aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
Diffstat
-rw-r--r--fs/exec.c12
-rw-r--r--fs/lockd/host.c4
-rw-r--r--fs/locks.c48
-rw-r--r--fs/namei.c95
-rw-r--r--fs/nfs/delegation.c2
-rw-r--r--fs/nfs/delegation.h16
-rw-r--r--fs/nfs/dir.c49
-rw-r--r--fs/nfs/file.c27
-rw-r--r--fs/nfs/inode.c96
-rw-r--r--fs/nfs/nfs3proc.c4
-rw-r--r--fs/nfs/nfs4_fs.h53
-rw-r--r--fs/nfs/nfs4proc.c581
-rw-r--r--fs/nfs/nfs4state.c181
-rw-r--r--fs/nfs/nfs4xdr.c65
-rw-r--r--fs/nfs/proc.c2
-rw-r--r--fs/open.c79
-rw-r--r--include/linux/namei.h8
-rw-r--r--include/linux/nfs_fs.h4
-rw-r--r--include/linux/nfs_xdr.h36
-rw-r--r--include/linux/sunrpc/auth.h7
-rw-r--r--include/linux/sunrpc/debug.h3
-rw-r--r--include/linux/sunrpc/gss_api.h27
-rw-r--r--include/linux/sunrpc/gss_err.h10
-rw-r--r--include/linux/sunrpc/gss_krb5.h27
-rw-r--r--include/linux/sunrpc/gss_spkm3.h4
-rw-r--r--include/linux/sunrpc/msg_prot.h25
-rw-r--r--include/linux/sunrpc/xdr.h6
-rw-r--r--include/linux/sunrpc/xprt.h227
-rw-r--r--net/sunrpc/Makefile2
-rw-r--r--net/sunrpc/auth.c1
-rw-r--r--net/sunrpc/auth_gss/Makefile2
-rw-r--r--net/sunrpc/auth_gss/auth_gss.c187
-rw-r--r--net/sunrpc/auth_gss/gss_krb5_crypto.c260
-rw-r--r--net/sunrpc/auth_gss/gss_krb5_mech.c41
-rw-r--r--net/sunrpc/auth_gss/gss_krb5_seal.c44
-rw-r--r--net/sunrpc/auth_gss/gss_krb5_unseal.c39
-rw-r--r--net/sunrpc/auth_gss/gss_krb5_wrap.c363
-rw-r--r--net/sunrpc/auth_gss/gss_mech_switch.c29
-rw-r--r--net/sunrpc/auth_gss/gss_spkm3_mech.c21
-rw-r--r--net/sunrpc/auth_gss/gss_spkm3_seal.c4
-rw-r--r--net/sunrpc/auth_gss/gss_spkm3_unseal.c2
-rw-r--r--net/sunrpc/auth_gss/svcauth_gss.c9
-rw-r--r--net/sunrpc/auth_null.c2
-rw-r--r--net/sunrpc/auth_unix.c2
-rw-r--r--net/sunrpc/clnt.c147
-rw-r--r--net/sunrpc/pmap_clnt.c12
-rw-r--r--net/sunrpc/socklib.c175
-rw-r--r--net/sunrpc/sunrpc_syms.c1
-rw-r--r--net/sunrpc/svcsock.c3
-rw-r--r--net/sunrpc/sysctl.c32
-rw-r--r--net/sunrpc/xdr.c177
-rw-r--r--net/sunrpc/xprt.c1613
-rw-r--r--net/sunrpc/xprtsock.c1252
53 files changed, 3954 insertions, 2164 deletions
diff --git a/fs/exec.c b/fs/exec.c
index a04a575ad433..d2208f7c87db 100644
--- a/fs/exec.c
+++ b/fs/exec.c
@@ -126,8 +126,7 @@ asmlinkage long sys_uselib(const char __user * library)
126 struct nameidata nd; 126 struct nameidata nd;
127 int error; 127 int error;
128 128
129 nd.intent.open.flags = FMODE_READ; 129 error = __user_path_lookup_open(library, LOOKUP_FOLLOW, &nd, FMODE_READ);
130 error = __user_walk(library, LOOKUP_FOLLOW|LOOKUP_OPEN, &nd);
131 if (error) 130 if (error)
132 goto out; 131 goto out;
133 132
@@ -139,7 +138,7 @@ asmlinkage long sys_uselib(const char __user * library)
139 if (error) 138 if (error)
140 goto exit; 139 goto exit;
141 140
142 file = dentry_open(nd.dentry, nd.mnt, O_RDONLY); 141 file = nameidata_to_filp(&nd, O_RDONLY);
143 error = PTR_ERR(file); 142 error = PTR_ERR(file);
144 if (IS_ERR(file)) 143 if (IS_ERR(file))
145 goto out; 144 goto out;
@@ -167,6 +166,7 @@ asmlinkage long sys_uselib(const char __user * library)
167out: 166out:
168 return error; 167 return error;
169exit: 168exit:
169 release_open_intent(&nd);
170 path_release(&nd); 170 path_release(&nd);
171 goto out; 171 goto out;
172} 172}
@@ -490,8 +490,7 @@ struct file *open_exec(const char *name)
490 int err; 490 int err;
491 struct file *file; 491 struct file *file;
492 492
493 nd.intent.open.flags = FMODE_READ; 493 err = path_lookup_open(name, LOOKUP_FOLLOW, &nd, FMODE_READ);
494 err = path_lookup(name, LOOKUP_FOLLOW|LOOKUP_OPEN, &nd);
495 file = ERR_PTR(err); 494 file = ERR_PTR(err);
496 495
497 if (!err) { 496 if (!err) {
@@ -504,7 +503,7 @@ struct file *open_exec(const char *name)
504 err = -EACCES; 503 err = -EACCES;
505 file = ERR_PTR(err); 504 file = ERR_PTR(err);
506 if (!err) { 505 if (!err) {
507 file = dentry_open(nd.dentry, nd.mnt, O_RDONLY); 506 file = nameidata_to_filp(&nd, O_RDONLY);
508 if (!IS_ERR(file)) { 507 if (!IS_ERR(file)) {
509 err = deny_write_access(file); 508 err = deny_write_access(file);
510 if (err) { 509 if (err) {
@@ -516,6 +515,7 @@ out:
516 return file; 515 return file;
517 } 516 }
518 } 517 }
518 release_open_intent(&nd);
519 path_release(&nd); 519 path_release(&nd);
520 } 520 }
521 goto out; 521 goto out;
diff --git a/fs/lockd/host.c b/fs/lockd/host.c
index 82c77df81c5f..c4c8601096e0 100644
--- a/fs/lockd/host.c
+++ b/fs/lockd/host.c
@@ -173,11 +173,10 @@ nlm_bind_host(struct nlm_host *host)
173 173
174 /* If we've already created an RPC client, check whether 174 /* If we've already created an RPC client, check whether
175 * RPC rebind is required 175 * RPC rebind is required
176 * Note: why keep rebinding if we're on a tcp connection?
177 */ 176 */
178 if ((clnt = host->h_rpcclnt) != NULL) { 177 if ((clnt = host->h_rpcclnt) != NULL) {
179 xprt = clnt->cl_xprt; 178 xprt = clnt->cl_xprt;
180 if (!xprt->stream && time_after_eq(jiffies, host->h_nextrebind)) { 179 if (time_after_eq(jiffies, host->h_nextrebind)) {
181 clnt->cl_port = 0; 180 clnt->cl_port = 0;
182 host->h_nextrebind = jiffies + NLM_HOST_REBIND; 181 host->h_nextrebind = jiffies + NLM_HOST_REBIND;
183 dprintk("lockd: next rebind in %ld jiffies\n", 182 dprintk("lockd: next rebind in %ld jiffies\n",
@@ -189,7 +188,6 @@ nlm_bind_host(struct nlm_host *host)
189 goto forgetit; 188 goto forgetit;
190 189
191 xprt_set_timeout(&xprt->timeout, 5, nlmsvc_timeout); 190 xprt_set_timeout(&xprt->timeout, 5, nlmsvc_timeout);
192 xprt->nocong = 1; /* No congestion control for NLM */
193 xprt->resvport = 1; /* NLM requires a reserved port */ 191 xprt->resvport = 1; /* NLM requires a reserved port */
194 192
195 /* Existing NLM servers accept AUTH_UNIX only */ 193 /* Existing NLM servers accept AUTH_UNIX only */
diff --git a/fs/locks.c b/fs/locks.c
index f7daa5f48949..a1e8b2248014 100644
--- a/fs/locks.c
+++ b/fs/locks.c
@@ -316,21 +316,22 @@ static int flock_to_posix_lock(struct file *filp, struct file_lock *fl,
316 /* POSIX-1996 leaves the case l->l_len < 0 undefined; 316 /* POSIX-1996 leaves the case l->l_len < 0 undefined;
317 POSIX-2001 defines it. */ 317 POSIX-2001 defines it. */
318 start += l->l_start; 318 start += l->l_start;
319 end = start + l->l_len - 1; 319 if (start < 0)
320 if (l->l_len < 0) { 320 return -EINVAL;
321 fl->fl_end = OFFSET_MAX;
322 if (l->l_len > 0) {
323 end = start + l->l_len - 1;
324 fl->fl_end = end;
325 } else if (l->l_len < 0) {
321 end = start - 1; 326 end = start - 1;
327 fl->fl_end = end;
322 start += l->l_len; 328 start += l->l_len;
329 if (start < 0)
330 return -EINVAL;
323 } 331 }
324
325 if (start < 0)
326 return -EINVAL;
327 if (l->l_len > 0 && end < 0)
328 return -EOVERFLOW;
329
330 fl->fl_start = start; /* we record the absolute position */ 332 fl->fl_start = start; /* we record the absolute position */
331 fl->fl_end = end; 333 if (fl->fl_end < fl->fl_start)
332 if (l->l_len == 0) 334 return -EOVERFLOW;
333 fl->fl_end = OFFSET_MAX;
334 335
335 fl->fl_owner = current->files; 336 fl->fl_owner = current->files;
336 fl->fl_pid = current->tgid; 337 fl->fl_pid = current->tgid;
@@ -362,14 +363,21 @@ static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl,
362 return -EINVAL; 363 return -EINVAL;
363 } 364 }
364 365
365 if (((start += l->l_start) < 0) || (l->l_len < 0)) 366 start += l->l_start;
367 if (start < 0)
366 return -EINVAL; 368 return -EINVAL;
367 fl->fl_end = start + l->l_len - 1; 369 fl->fl_end = OFFSET_MAX;
368 if (l->l_len > 0 && fl->fl_end < 0) 370 if (l->l_len > 0) {
369 return -EOVERFLOW; 371 fl->fl_end = start + l->l_len - 1;
372 } else if (l->l_len < 0) {
373 fl->fl_end = start - 1;
374 start += l->l_len;
375 if (start < 0)
376 return -EINVAL;
377 }
370 fl->fl_start = start; /* we record the absolute position */ 378 fl->fl_start = start; /* we record the absolute position */
371 if (l->l_len == 0) 379 if (fl->fl_end < fl->fl_start)
372 fl->fl_end = OFFSET_MAX; 380 return -EOVERFLOW;
373 381
374 fl->fl_owner = current->files; 382 fl->fl_owner = current->files;
375 fl->fl_pid = current->tgid; 383 fl->fl_pid = current->tgid;
@@ -829,12 +837,16 @@ static int __posix_lock_file(struct inode *inode, struct file_lock *request)
829 /* Detect adjacent or overlapping regions (if same lock type) 837 /* Detect adjacent or overlapping regions (if same lock type)
830 */ 838 */
831 if (request->fl_type == fl->fl_type) { 839 if (request->fl_type == fl->fl_type) {
840 /* In all comparisons of start vs end, use
841 * "start - 1" rather than "end + 1". If end
842 * is OFFSET_MAX, end + 1 will become negative.
843 */
832 if (fl->fl_end < request->fl_start - 1) 844 if (fl->fl_end < request->fl_start - 1)
833 goto next_lock; 845 goto next_lock;
834 /* If the next lock in the list has entirely bigger 846 /* If the next lock in the list has entirely bigger
835 * addresses than the new one, insert the lock here. 847 * addresses than the new one, insert the lock here.
836 */ 848 */
837 if (fl->fl_start > request->fl_end + 1) 849 if (fl->fl_start - 1 > request->fl_end)
838 break; 850 break;
839 851
840 /* If we come here, the new and old lock are of the 852 /* If we come here, the new and old lock are of the
diff --git a/fs/namei.c b/fs/namei.c
index aa62dbda93ac..aaaa81036234 100644
--- a/fs/namei.c
+++ b/fs/namei.c
@@ -28,6 +28,7 @@
28#include <linux/syscalls.h> 28#include <linux/syscalls.h>
29#include <linux/mount.h> 29#include <linux/mount.h>
30#include <linux/audit.h> 30#include <linux/audit.h>
31#include <linux/file.h>
31#include <asm/namei.h> 32#include <asm/namei.h>
32#include <asm/uaccess.h> 33#include <asm/uaccess.h>
33 34
@@ -317,6 +318,18 @@ void path_release_on_umount(struct nameidata *nd)
317 mntput_no_expire(nd->mnt); 318 mntput_no_expire(nd->mnt);
318} 319}
319 320
321/**
322 * release_open_intent - free up open intent resources
323 * @nd: pointer to nameidata
324 */
325void release_open_intent(struct nameidata *nd)
326{
327 if (nd->intent.open.file->f_dentry == NULL)
328 put_filp(nd->intent.open.file);
329 else
330 fput(nd->intent.open.file);
331}
332
320/* 333/*
321 * Internal lookup() using the new generic dcache. 334 * Internal lookup() using the new generic dcache.
322 * SMP-safe 335 * SMP-safe
@@ -750,6 +763,7 @@ static fastcall int __link_path_walk(const char * name, struct nameidata *nd)
750 struct qstr this; 763 struct qstr this;
751 unsigned int c; 764 unsigned int c;
752 765
766 nd->flags |= LOOKUP_CONTINUE;
753 err = exec_permission_lite(inode, nd); 767 err = exec_permission_lite(inode, nd);
754 if (err == -EAGAIN) { 768 if (err == -EAGAIN) {
755 err = permission(inode, MAY_EXEC, nd); 769 err = permission(inode, MAY_EXEC, nd);
@@ -802,7 +816,6 @@ static fastcall int __link_path_walk(const char * name, struct nameidata *nd)
802 if (err < 0) 816 if (err < 0)
803 break; 817 break;
804 } 818 }
805 nd->flags |= LOOKUP_CONTINUE;
806 /* This does the actual lookups.. */ 819 /* This does the actual lookups.. */
807 err = do_lookup(nd, &this, &next); 820 err = do_lookup(nd, &this, &next);
808 if (err) 821 if (err)
@@ -1052,6 +1065,70 @@ out:
1052 return retval; 1065 return retval;
1053} 1066}
1054 1067
1068static int __path_lookup_intent_open(const char *name, unsigned int lookup_flags,
1069 struct nameidata *nd, int open_flags, int create_mode)
1070{
1071 struct file *filp = get_empty_filp();
1072 int err;
1073
1074 if (filp == NULL)
1075 return -ENFILE;
1076 nd->intent.open.file = filp;
1077 nd->intent.open.flags = open_flags;
1078 nd->intent.open.create_mode = create_mode;
1079 err = path_lookup(name, lookup_flags|LOOKUP_OPEN, nd);
1080 if (IS_ERR(nd->intent.open.file)) {
1081 if (err == 0) {
1082 err = PTR_ERR(nd->intent.open.file);
1083 path_release(nd);
1084 }
1085 } else if (err != 0)
1086 release_open_intent(nd);
1087 return err;
1088}
1089
1090/**
1091 * path_lookup_open - lookup a file path with open intent
1092 * @name: pointer to file name
1093 * @lookup_flags: lookup intent flags
1094 * @nd: pointer to nameidata
1095 * @open_flags: open intent flags
1096 */
1097int path_lookup_open(const char *name, unsigned int lookup_flags,
1098 struct nameidata *nd, int open_flags)
1099{
1100 return __path_lookup_intent_open(name, lookup_flags, nd,
1101 open_flags, 0);
1102}
1103
1104/**
1105 * path_lookup_create - lookup a file path with open + create intent
1106 * @name: pointer to file name
1107 * @lookup_flags: lookup intent flags
1108 * @nd: pointer to nameidata
1109 * @open_flags: open intent flags
1110 * @create_mode: create intent flags
1111 */
1112int path_lookup_create(const char *name, unsigned int lookup_flags,
1113 struct nameidata *nd, int open_flags, int create_mode)
1114{
1115 return __path_lookup_intent_open(name, lookup_flags|LOOKUP_CREATE, nd,
1116 open_flags, create_mode);
1117}
1118
1119int __user_path_lookup_open(const char __user *name, unsigned int lookup_flags,
1120 struct nameidata *nd, int open_flags)
1121{
1122 char *tmp = getname(name);
1123 int err = PTR_ERR(tmp);
1124
1125 if (!IS_ERR(tmp)) {
1126 err = __path_lookup_intent_open(tmp, lookup_flags, nd, open_flags, 0);
1127 putname(tmp);
1128 }
1129 return err;
1130}
1131
1055/* 1132/*
1056 * Restricted form of lookup. Doesn't follow links, single-component only, 1133 * Restricted form of lookup. Doesn't follow links, single-component only,
1057 * needs parent already locked. Doesn't follow mounts. 1134 * needs parent already locked. Doesn't follow mounts.
@@ -1416,27 +1493,27 @@ int may_open(struct nameidata *nd, int acc_mode, int flag)
1416 */ 1493 */
1417int open_namei(const char * pathname, int flag, int mode, struct nameidata *nd) 1494int open_namei(const char * pathname, int flag, int mode, struct nameidata *nd)
1418{ 1495{
1419 int acc_mode, error = 0; 1496 int acc_mode, error;
1420 struct path path; 1497 struct path path;
1421 struct dentry *dir; 1498 struct dentry *dir;
1422 int count = 0; 1499 int count = 0;
1423 1500
1424 acc_mode = ACC_MODE(flag); 1501 acc_mode = ACC_MODE(flag);
1425 1502
1503 /* O_TRUNC implies we need access checks for write permissions */
1504 if (flag & O_TRUNC)
1505 acc_mode |= MAY_WRITE;
1506
1426 /* Allow the LSM permission hook to distinguish append 1507 /* Allow the LSM permission hook to distinguish append
1427 access from general write access. */ 1508 access from general write access. */
1428 if (flag & O_APPEND) 1509 if (flag & O_APPEND)
1429 acc_mode |= MAY_APPEND; 1510 acc_mode |= MAY_APPEND;
1430 1511
1431 /* Fill in the open() intent data */
1432 nd->intent.open.flags = flag;
1433 nd->intent.open.create_mode = mode;
1434
1435 /* 1512 /*
1436 * The simplest case - just a plain lookup. 1513 * The simplest case - just a plain lookup.
1437 */ 1514 */
1438 if (!(flag & O_CREAT)) { 1515 if (!(flag & O_CREAT)) {
1439 error = path_lookup(pathname, lookup_flags(flag)|LOOKUP_OPEN, nd); 1516 error = path_lookup_open(pathname, lookup_flags(flag), nd, flag);
1440 if (error) 1517 if (error)
1441 return error; 1518 return error;
1442 goto ok; 1519 goto ok;
@@ -1445,7 +1522,7 @@ int open_namei(const char * pathname, int flag, int mode, struct nameidata *nd)
1445 /* 1522 /*
1446 * Create - we need to know the parent. 1523 * Create - we need to know the parent.
1447 */ 1524 */
1448 error = path_lookup(pathname, LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE, nd); 1525 error = path_lookup_create(pathname, LOOKUP_PARENT, nd, flag, mode);
1449 if (error) 1526 if (error)
1450 return error; 1527 return error;
1451 1528
@@ -1520,6 +1597,8 @@ ok:
1520exit_dput: 1597exit_dput:
1521 dput_path(&path, nd); 1598 dput_path(&path, nd);
1522exit: 1599exit:
1600 if (!IS_ERR(nd->intent.open.file))
1601 release_open_intent(nd);
1523 path_release(nd); 1602 path_release(nd);
1524 return error; 1603 return error;
1525 1604
diff --git a/fs/nfs/delegation.c b/fs/nfs/delegation.c
index 4a36839f0bbd..44135af9894c 100644
--- a/fs/nfs/delegation.c
+++ b/fs/nfs/delegation.c
@@ -142,7 +142,7 @@ static void nfs_msync_inode(struct inode *inode)
142/* 142/*
143 * Basic procedure for returning a delegation to the server 143 * Basic procedure for returning a delegation to the server
144 */ 144 */
145int nfs_inode_return_delegation(struct inode *inode) 145int __nfs_inode_return_delegation(struct inode *inode)
146{ 146{
147 struct nfs4_client *clp = NFS_SERVER(inode)->nfs4_state; 147 struct nfs4_client *clp = NFS_SERVER(inode)->nfs4_state;
148 struct nfs_inode *nfsi = NFS_I(inode); 148 struct nfs_inode *nfsi = NFS_I(inode);
diff --git a/fs/nfs/delegation.h b/fs/nfs/delegation.h
index 3f6c45a29d6a..8017846b561f 100644
--- a/fs/nfs/delegation.h
+++ b/fs/nfs/delegation.h
@@ -25,7 +25,7 @@ struct nfs_delegation {
25 25
26int nfs_inode_set_delegation(struct inode *inode, struct rpc_cred *cred, struct nfs_openres *res); 26int nfs_inode_set_delegation(struct inode *inode, struct rpc_cred *cred, struct nfs_openres *res);
27void nfs_inode_reclaim_delegation(struct inode *inode, struct rpc_cred *cred, struct nfs_openres *res); 27void nfs_inode_reclaim_delegation(struct inode *inode, struct rpc_cred *cred, struct nfs_openres *res);
28int nfs_inode_return_delegation(struct inode *inode); 28int __nfs_inode_return_delegation(struct inode *inode);
29int nfs_async_inode_return_delegation(struct inode *inode, const nfs4_stateid *stateid); 29int nfs_async_inode_return_delegation(struct inode *inode, const nfs4_stateid *stateid);
30 30
31struct inode *nfs_delegation_find_inode(struct nfs4_client *clp, const struct nfs_fh *fhandle); 31struct inode *nfs_delegation_find_inode(struct nfs4_client *clp, const struct nfs_fh *fhandle);
@@ -47,11 +47,25 @@ static inline int nfs_have_delegation(struct inode *inode, int flags)
47 return 1; 47 return 1;
48 return 0; 48 return 0;
49} 49}
50
51static inline int nfs_inode_return_delegation(struct inode *inode)
52{
53 int err = 0;
54
55 if (NFS_I(inode)->delegation != NULL)
56 err = __nfs_inode_return_delegation(inode);
57 return err;
58}
50#else 59#else
51static inline int nfs_have_delegation(struct inode *inode, int flags) 60static inline int nfs_have_delegation(struct inode *inode, int flags)
52{ 61{
53 return 0; 62 return 0;
54} 63}
64
65static inline int nfs_inode_return_delegation(struct inode *inode)
66{
67 return 0;
68}
55#endif 69#endif
56 70
57#endif 71#endif
diff --git a/fs/nfs/dir.c b/fs/nfs/dir.c
index 2df639f143e8..eb50c19fc253 100644
--- a/fs/nfs/dir.c
+++ b/fs/nfs/dir.c
@@ -565,8 +565,6 @@ static int nfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
565 } 565 }
566 } 566 }
567 unlock_kernel(); 567 unlock_kernel();
568 if (desc->error < 0)
569 return desc->error;
570 if (res < 0) 568 if (res < 0)
571 return res; 569 return res;
572 return 0; 570 return 0;
@@ -803,6 +801,7 @@ static int nfs_dentry_delete(struct dentry *dentry)
803 */ 801 */
804static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode) 802static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
805{ 803{
804 nfs_inode_return_delegation(inode);
806 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { 805 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
807 lock_kernel(); 806 lock_kernel();
808 inode->i_nlink--; 807 inode->i_nlink--;
@@ -916,7 +915,6 @@ static int is_atomic_open(struct inode *dir, struct nameidata *nd)
916static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd) 915static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
917{ 916{
918 struct dentry *res = NULL; 917 struct dentry *res = NULL;
919 struct inode *inode = NULL;
920 int error; 918 int error;
921 919
922 /* Check that we are indeed trying to open this file */ 920 /* Check that we are indeed trying to open this file */
@@ -930,8 +928,10 @@ static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry
930 dentry->d_op = NFS_PROTO(dir)->dentry_ops; 928 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
931 929
932 /* Let vfs_create() deal with O_EXCL */ 930 /* Let vfs_create() deal with O_EXCL */
933 if (nd->intent.open.flags & O_EXCL) 931 if (nd->intent.open.flags & O_EXCL) {
934 goto no_entry; 932 d_add(dentry, NULL);
933 goto out;
934 }
935 935
936 /* Open the file on the server */ 936 /* Open the file on the server */
937 lock_kernel(); 937 lock_kernel();
@@ -945,32 +945,30 @@ static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry
945 945
946 if (nd->intent.open.flags & O_CREAT) { 946 if (nd->intent.open.flags & O_CREAT) {
947 nfs_begin_data_update(dir); 947 nfs_begin_data_update(dir);
948 inode = nfs4_atomic_open(dir, dentry, nd); 948 res = nfs4_atomic_open(dir, dentry, nd);
949 nfs_end_data_update(dir); 949 nfs_end_data_update(dir);
950 } else 950 } else
951 inode = nfs4_atomic_open(dir, dentry, nd); 951 res = nfs4_atomic_open(dir, dentry, nd);
952 unlock_kernel(); 952 unlock_kernel();
953 if (IS_ERR(inode)) { 953 if (IS_ERR(res)) {
954 error = PTR_ERR(inode); 954 error = PTR_ERR(res);
955 switch (error) { 955 switch (error) {
956 /* Make a negative dentry */ 956 /* Make a negative dentry */
957 case -ENOENT: 957 case -ENOENT:
958 inode = NULL; 958 res = NULL;
959 break; 959 goto out;
960 /* This turned out not to be a regular file */ 960 /* This turned out not to be a regular file */
961 case -EISDIR:
962 case -ENOTDIR:
963 goto no_open;
961 case -ELOOP: 964 case -ELOOP:
962 if (!(nd->intent.open.flags & O_NOFOLLOW)) 965 if (!(nd->intent.open.flags & O_NOFOLLOW))
963 goto no_open; 966 goto no_open;
964 /* case -EISDIR: */
965 /* case -EINVAL: */ 967 /* case -EINVAL: */
966 default: 968 default:
967 res = ERR_PTR(error);
968 goto out; 969 goto out;
969 } 970 }
970 } 971 } else if (res != NULL)
971no_entry:
972 res = d_add_unique(dentry, inode);
973 if (res != NULL)
974 dentry = res; 972 dentry = res;
975 nfs_renew_times(dentry); 973 nfs_renew_times(dentry);
976 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 974 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
@@ -1014,7 +1012,7 @@ static int nfs_open_revalidate(struct dentry *dentry, struct nameidata *nd)
1014 */ 1012 */
1015 lock_kernel(); 1013 lock_kernel();
1016 verifier = nfs_save_change_attribute(dir); 1014 verifier = nfs_save_change_attribute(dir);
1017 ret = nfs4_open_revalidate(dir, dentry, openflags); 1015 ret = nfs4_open_revalidate(dir, dentry, openflags, nd);
1018 if (!ret) 1016 if (!ret)
1019 nfs_set_verifier(dentry, verifier); 1017 nfs_set_verifier(dentry, verifier);
1020 unlock_kernel(); 1018 unlock_kernel();
@@ -1137,7 +1135,7 @@ static int nfs_create(struct inode *dir, struct dentry *dentry, int mode,
1137 1135
1138 lock_kernel(); 1136 lock_kernel();
1139 nfs_begin_data_update(dir); 1137 nfs_begin_data_update(dir);
1140 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags); 1138 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, nd);
1141 nfs_end_data_update(dir); 1139 nfs_end_data_update(dir);
1142 if (error != 0) 1140 if (error != 0)
1143 goto out_err; 1141 goto out_err;
@@ -1332,6 +1330,7 @@ static int nfs_safe_remove(struct dentry *dentry)
1332 1330
1333 nfs_begin_data_update(dir); 1331 nfs_begin_data_update(dir);
1334 if (inode != NULL) { 1332 if (inode != NULL) {
1333 nfs_inode_return_delegation(inode);
1335 nfs_begin_data_update(inode); 1334 nfs_begin_data_update(inode);
1336 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name); 1335 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1337 /* The VFS may want to delete this inode */ 1336 /* The VFS may want to delete this inode */
@@ -1512,9 +1511,11 @@ static int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1512 */ 1511 */
1513 if (!new_inode) 1512 if (!new_inode)
1514 goto go_ahead; 1513 goto go_ahead;
1515 if (S_ISDIR(new_inode->i_mode)) 1514 if (S_ISDIR(new_inode->i_mode)) {
1516 goto out; 1515 error = -EISDIR;
1517 else if (atomic_read(&new_dentry->d_count) > 2) { 1516 if (!S_ISDIR(old_inode->i_mode))
1517 goto out;
1518 } else if (atomic_read(&new_dentry->d_count) > 2) {
1518 int err; 1519 int err;
1519 /* copy the target dentry's name */ 1520 /* copy the target dentry's name */
1520 dentry = d_alloc(new_dentry->d_parent, 1521 dentry = d_alloc(new_dentry->d_parent,
@@ -1539,7 +1540,8 @@ static int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1539#endif 1540#endif
1540 goto out; 1541 goto out;
1541 } 1542 }
1542 } 1543 } else
1544 new_inode->i_nlink--;
1543 1545
1544go_ahead: 1546go_ahead:
1545 /* 1547 /*
@@ -1549,6 +1551,7 @@ go_ahead:
1549 nfs_wb_all(old_inode); 1551 nfs_wb_all(old_inode);
1550 shrink_dcache_parent(old_dentry); 1552 shrink_dcache_parent(old_dentry);
1551 } 1553 }
1554 nfs_inode_return_delegation(old_inode);
1552 1555
1553 if (new_inode) 1556 if (new_inode)
1554 d_delete(new_dentry); 1557 d_delete(new_dentry);
diff --git a/fs/nfs/file.c b/fs/nfs/file.c
index 6bdcfa95de94..572d8593486f 100644
--- a/fs/nfs/file.c
+++ b/fs/nfs/file.c
@@ -376,22 +376,31 @@ out_swapfile:
376 376
377static int do_getlk(struct file *filp, int cmd, struct file_lock *fl) 377static int do_getlk(struct file *filp, int cmd, struct file_lock *fl)
378{ 378{
379 struct file_lock *cfl;
379 struct inode *inode = filp->f_mapping->host; 380 struct inode *inode = filp->f_mapping->host;
380 int status = 0; 381 int status = 0;
381 382
382 lock_kernel(); 383 lock_kernel();
383 /* Use local locking if mounted with "-onolock" */ 384 /* Try local locking first */
384 if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM)) 385 cfl = posix_test_lock(filp, fl);
385 status = NFS_PROTO(inode)->lock(filp, cmd, fl); 386 if (cfl != NULL) {
386 else { 387 locks_copy_lock(fl, cfl);
387 struct file_lock *cfl = posix_test_lock(filp, fl); 388 goto out;
388
389 fl->fl_type = F_UNLCK;
390 if (cfl != NULL)
391 memcpy(fl, cfl, sizeof(*fl));
392 } 389 }
390
391 if (nfs_have_delegation(inode, FMODE_READ))
392 goto out_noconflict;
393
394 if (NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM)
395 goto out_noconflict;
396
397 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
398out:
393 unlock_kernel(); 399 unlock_kernel();
394 return status; 400 return status;
401out_noconflict:
402 fl->fl_type = F_UNLCK;
403 goto out;
395} 404}
396 405
397static int do_vfs_lock(struct file *file, struct file_lock *fl) 406static int do_vfs_lock(struct file *file, struct file_lock *fl)
diff --git a/fs/nfs/inode.c b/fs/nfs/inode.c
index d4eadeea128e..65d5ab45ddc5 100644
--- a/fs/nfs/inode.c
+++ b/fs/nfs/inode.c
@@ -358,6 +358,35 @@ out_no_root:
358 return no_root_error; 358 return no_root_error;
359} 359}
360 360
361static void nfs_init_timeout_values(struct rpc_timeout *to, int proto, unsigned int timeo, unsigned int retrans)
362{
363 to->to_initval = timeo * HZ / 10;
364 to->to_retries = retrans;
365 if (!to->to_retries)
366 to->to_retries = 2;
367
368 switch (proto) {
369 case IPPROTO_TCP:
370 if (!to->to_initval)
371 to->to_initval = 60 * HZ;
372 if (to->to_initval > NFS_MAX_TCP_TIMEOUT)
373 to->to_initval = NFS_MAX_TCP_TIMEOUT;
374 to->to_increment = to->to_initval;
375 to->to_maxval = to->to_initval + (to->to_increment * to->to_retries);
376 to->to_exponential = 0;
377 break;
378 case IPPROTO_UDP:
379 default:
380 if (!to->to_initval)
381 to->to_initval = 11 * HZ / 10;
382 if (to->to_initval > NFS_MAX_UDP_TIMEOUT)
383 to->to_initval = NFS_MAX_UDP_TIMEOUT;
384 to->to_maxval = NFS_MAX_UDP_TIMEOUT;
385 to->to_exponential = 1;
386 break;
387 }
388}
389
361/* 390/*
362 * Create an RPC client handle. 391 * Create an RPC client handle.
363 */ 392 */
@@ -367,22 +396,12 @@ nfs_create_client(struct nfs_server *server, const struct nfs_mount_data *data)
367 struct rpc_timeout timeparms; 396 struct rpc_timeout timeparms;
368 struct rpc_xprt *xprt = NULL; 397 struct rpc_xprt *xprt = NULL;
369 struct rpc_clnt *clnt = NULL; 398 struct rpc_clnt *clnt = NULL;
370 int tcp = (data->flags & NFS_MOUNT_TCP); 399 int proto = (data->flags & NFS_MOUNT_TCP) ? IPPROTO_TCP : IPPROTO_UDP;
371 400
372 /* Initialize timeout values */ 401 nfs_init_timeout_values(&timeparms, proto, data->timeo, data->retrans);
373 timeparms.to_initval = data->timeo * HZ / 10;
374 timeparms.to_retries = data->retrans;
375 timeparms.to_maxval = tcp ? RPC_MAX_TCP_TIMEOUT : RPC_MAX_UDP_TIMEOUT;
376 timeparms.to_exponential = 1;
377
378 if (!timeparms.to_initval)
379 timeparms.to_initval = (tcp ? 600 : 11) * HZ / 10;
380 if (!timeparms.to_retries)
381 timeparms.to_retries = 5;
382 402
383 /* create transport and client */ 403 /* create transport and client */
384 xprt = xprt_create_proto(tcp ? IPPROTO_TCP : IPPROTO_UDP, 404 xprt = xprt_create_proto(proto, &server->addr, &timeparms);
385 &server->addr, &timeparms);
386 if (IS_ERR(xprt)) { 405 if (IS_ERR(xprt)) {
387 dprintk("%s: cannot create RPC transport. Error = %ld\n", 406 dprintk("%s: cannot create RPC transport. Error = %ld\n",
388 __FUNCTION__, PTR_ERR(xprt)); 407 __FUNCTION__, PTR_ERR(xprt));
@@ -576,7 +595,6 @@ static int nfs_show_options(struct seq_file *m, struct vfsmount *mnt)
576 { NFS_MOUNT_SOFT, ",soft", ",hard" }, 595 { NFS_MOUNT_SOFT, ",soft", ",hard" },
577 { NFS_MOUNT_INTR, ",intr", "" }, 596 { NFS_MOUNT_INTR, ",intr", "" },
578 { NFS_MOUNT_POSIX, ",posix", "" }, 597 { NFS_MOUNT_POSIX, ",posix", "" },
579 { NFS_MOUNT_TCP, ",tcp", ",udp" },
580 { NFS_MOUNT_NOCTO, ",nocto", "" }, 598 { NFS_MOUNT_NOCTO, ",nocto", "" },
581 { NFS_MOUNT_NOAC, ",noac", "" }, 599 { NFS_MOUNT_NOAC, ",noac", "" },
582 { NFS_MOUNT_NONLM, ",nolock", ",lock" }, 600 { NFS_MOUNT_NONLM, ",nolock", ",lock" },
@@ -585,6 +603,8 @@ static int nfs_show_options(struct seq_file *m, struct vfsmount *mnt)
585 }; 603 };
586 struct proc_nfs_info *nfs_infop; 604 struct proc_nfs_info *nfs_infop;
587 struct nfs_server *nfss = NFS_SB(mnt->mnt_sb); 605 struct nfs_server *nfss = NFS_SB(mnt->mnt_sb);
606 char buf[12];
607 char *proto;
588 608
589 seq_printf(m, ",v%d", nfss->rpc_ops->version); 609 seq_printf(m, ",v%d", nfss->rpc_ops->version);
590 seq_printf(m, ",rsize=%d", nfss->rsize); 610 seq_printf(m, ",rsize=%d", nfss->rsize);
@@ -603,6 +623,18 @@ static int nfs_show_options(struct seq_file *m, struct vfsmount *mnt)
603 else 623 else
604 seq_puts(m, nfs_infop->nostr); 624 seq_puts(m, nfs_infop->nostr);
605 } 625 }
626 switch (nfss->client->cl_xprt->prot) {
627 case IPPROTO_TCP:
628 proto = "tcp";
629 break;
630 case IPPROTO_UDP:
631 proto = "udp";
632 break;
633 default:
634 snprintf(buf, sizeof(buf), "%u", nfss->client->cl_xprt->prot);
635 proto = buf;
636 }
637 seq_printf(m, ",proto=%s", proto);
606 seq_puts(m, ",addr="); 638 seq_puts(m, ",addr=");
607 seq_escape(m, nfss->hostname, " \t\n\\"); 639 seq_escape(m, nfss->hostname, " \t\n\\");
608 return 0; 640 return 0;
@@ -821,6 +853,11 @@ nfs_setattr(struct dentry *dentry, struct iattr *attr)
821 filemap_fdatawait(inode->i_mapping); 853 filemap_fdatawait(inode->i_mapping);
822 nfs_wb_all(inode); 854 nfs_wb_all(inode);
823 } 855 }
856 /*
857 * Return any delegations if we're going to change ACLs
858 */
859 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0)
860 nfs_inode_return_delegation(inode);
824 error = NFS_PROTO(inode)->setattr(dentry, &fattr, attr); 861 error = NFS_PROTO(inode)->setattr(dentry, &fattr, attr);
825 if (error == 0) 862 if (error == 0)
826 nfs_refresh_inode(inode, &fattr); 863 nfs_refresh_inode(inode, &fattr);
@@ -1639,8 +1676,7 @@ static void nfs4_clear_inode(struct inode *inode)
1639 struct nfs_inode *nfsi = NFS_I(inode); 1676 struct nfs_inode *nfsi = NFS_I(inode);
1640 1677
1641 /* If we are holding a delegation, return it! */ 1678 /* If we are holding a delegation, return it! */
1642 if (nfsi->delegation != NULL) 1679 nfs_inode_return_delegation(inode);
1643 nfs_inode_return_delegation(inode);
1644 /* First call standard NFS clear_inode() code */ 1680 /* First call standard NFS clear_inode() code */
1645 nfs_clear_inode(inode); 1681 nfs_clear_inode(inode);
1646 /* Now clear out any remaining state */ 1682 /* Now clear out any remaining state */
@@ -1669,7 +1705,7 @@ static int nfs4_fill_super(struct super_block *sb, struct nfs4_mount_data *data,
1669 struct rpc_clnt *clnt = NULL; 1705 struct rpc_clnt *clnt = NULL;
1670 struct rpc_timeout timeparms; 1706 struct rpc_timeout timeparms;
1671 rpc_authflavor_t authflavour; 1707 rpc_authflavor_t authflavour;
1672 int proto, err = -EIO; 1708 int err = -EIO;
1673 1709
1674 sb->s_blocksize_bits = 0; 1710 sb->s_blocksize_bits = 0;
1675 sb->s_blocksize = 0; 1711 sb->s_blocksize = 0;
@@ -1687,30 +1723,8 @@ static int nfs4_fill_super(struct super_block *sb, struct nfs4_mount_data *data,
1687 server->acdirmax = data->acdirmax*HZ; 1723 server->acdirmax = data->acdirmax*HZ;
1688 1724
1689 server->rpc_ops = &nfs_v4_clientops; 1725 server->rpc_ops = &nfs_v4_clientops;
1690 /* Initialize timeout values */
1691
1692 timeparms.to_initval = data->timeo * HZ / 10;
1693 timeparms.to_retries = data->retrans;
1694 timeparms.to_exponential = 1;
1695 if (!timeparms.to_retries)
1696 timeparms.to_retries = 5;
1697 1726
1698 proto = data->proto; 1727 nfs_init_timeout_values(&timeparms, data->proto, data->timeo, data->retrans);
1699 /* Which IP protocol do we use? */
1700 switch (proto) {
1701 case IPPROTO_TCP:
1702 timeparms.to_maxval = RPC_MAX_TCP_TIMEOUT;
1703 if (!timeparms.to_initval)
1704 timeparms.to_initval = 600 * HZ / 10;
1705 break;
1706 case IPPROTO_UDP:
1707 timeparms.to_maxval = RPC_MAX_UDP_TIMEOUT;
1708 if (!timeparms.to_initval)
1709 timeparms.to_initval = 11 * HZ / 10;
1710 break;
1711 default:
1712 return -EINVAL;
1713 }
1714 1728
1715 clp = nfs4_get_client(&server->addr.sin_addr); 1729 clp = nfs4_get_client(&server->addr.sin_addr);
1716 if (!clp) { 1730 if (!clp) {
@@ -1735,7 +1749,7 @@ static int nfs4_fill_super(struct super_block *sb, struct nfs4_mount_data *data,
1735 1749
1736 down_write(&clp->cl_sem); 1750 down_write(&clp->cl_sem);
1737 if (IS_ERR(clp->cl_rpcclient)) { 1751 if (IS_ERR(clp->cl_rpcclient)) {
1738 xprt = xprt_create_proto(proto, &server->addr, &timeparms); 1752 xprt = xprt_create_proto(data->proto, &server->addr, &timeparms);
1739 if (IS_ERR(xprt)) { 1753 if (IS_ERR(xprt)) {
1740 up_write(&clp->cl_sem); 1754 up_write(&clp->cl_sem);
1741 err = PTR_ERR(xprt); 1755 err = PTR_ERR(xprt);
diff --git a/fs/nfs/nfs3proc.c b/fs/nfs/nfs3proc.c
index edc95514046d..e4a1cd48195e 100644
--- a/fs/nfs/nfs3proc.c
+++ b/fs/nfs/nfs3proc.c
@@ -299,7 +299,7 @@ static int nfs3_proc_commit(struct nfs_write_data *cdata)
299 */ 299 */
300static int 300static int
301nfs3_proc_create(struct inode *dir, struct dentry *dentry, struct iattr *sattr, 301nfs3_proc_create(struct inode *dir, struct dentry *dentry, struct iattr *sattr,
302 int flags) 302 int flags, struct nameidata *nd)
303{ 303{
304 struct nfs_fh fhandle; 304 struct nfs_fh fhandle;
305 struct nfs_fattr fattr; 305 struct nfs_fattr fattr;
@@ -735,7 +735,7 @@ extern u32 *nfs3_decode_dirent(u32 *, struct nfs_entry *, int);
735static void 735static void
736nfs3_read_done(struct rpc_task *task) 736nfs3_read_done(struct rpc_task *task)
737{ 737{
738 struct nfs_write_data *data = (struct nfs_write_data *) task->tk_calldata; 738 struct nfs_read_data *data = (struct nfs_read_data *) task->tk_calldata;
739 739
740 if (nfs3_async_handle_jukebox(task)) 740 if (nfs3_async_handle_jukebox(task))
741 return; 741 return;
diff --git a/fs/nfs/nfs4_fs.h b/fs/nfs/nfs4_fs.h
index ec1a22d7b876..78a53f5a9f18 100644
--- a/fs/nfs/nfs4_fs.h
+++ b/fs/nfs/nfs4_fs.h
@@ -93,25 +93,50 @@ struct nfs4_client {
93}; 93};
94 94
95/* 95/*
96 * struct rpc_sequence ensures that RPC calls are sent in the exact
97 * order that they appear on the list.
98 */
99struct rpc_sequence {
100 struct rpc_wait_queue wait; /* RPC call delay queue */
101 spinlock_t lock; /* Protects the list */
102 struct list_head list; /* Defines sequence of RPC calls */
103};
104
105#define NFS_SEQID_CONFIRMED 1
106struct nfs_seqid_counter {
107 struct rpc_sequence *sequence;
108 int flags;
109 u32 counter;
110};
111
112struct nfs_seqid {
113 struct nfs_seqid_counter *sequence;
114 struct list_head list;
115};
116
117static inline void nfs_confirm_seqid(struct nfs_seqid_counter *seqid, int status)
118{
119 if (seqid_mutating_err(-status))
120 seqid->flags |= NFS_SEQID_CONFIRMED;
121}
122
123/*
96 * NFS4 state_owners and lock_owners are simply labels for ordered 124 * NFS4 state_owners and lock_owners are simply labels for ordered
97 * sequences of RPC calls. Their sole purpose is to provide once-only 125 * sequences of RPC calls. Their sole purpose is to provide once-only
98 * semantics by allowing the server to identify replayed requests. 126 * semantics by allowing the server to identify replayed requests.
99 *
100 * The ->so_sema is held during all state_owner seqid-mutating operations:
101 * OPEN, OPEN_DOWNGRADE, and CLOSE. Its purpose is to properly serialize
102 * so_seqid.
103 */ 127 */
104struct nfs4_state_owner { 128struct nfs4_state_owner {
129 spinlock_t so_lock;
105 struct list_head so_list; /* per-clientid list of state_owners */ 130 struct list_head so_list; /* per-clientid list of state_owners */
106 struct nfs4_client *so_client; 131 struct nfs4_client *so_client;
107 u32 so_id; /* 32-bit identifier, unique */ 132 u32 so_id; /* 32-bit identifier, unique */
108 struct semaphore so_sema;
109 u32 so_seqid; /* protected by so_sema */
110 atomic_t so_count; 133 atomic_t so_count;
111 134
112 struct rpc_cred *so_cred; /* Associated cred */ 135 struct rpc_cred *so_cred; /* Associated cred */
113 struct list_head so_states; 136 struct list_head so_states;
114 struct list_head so_delegations; 137 struct list_head so_delegations;
138 struct nfs_seqid_counter so_seqid;
139 struct rpc_sequence so_sequence;
115}; 140};
116 141
117/* 142/*
@@ -132,7 +157,7 @@ struct nfs4_lock_state {
132 fl_owner_t ls_owner; /* POSIX lock owner */ 157 fl_owner_t ls_owner; /* POSIX lock owner */
133#define NFS_LOCK_INITIALIZED 1 158#define NFS_LOCK_INITIALIZED 1
134 int ls_flags; 159 int ls_flags;
135 u32 ls_seqid; 160 struct nfs_seqid_counter ls_seqid;
136 u32 ls_id; 161 u32 ls_id;
137 nfs4_stateid ls_stateid; 162 nfs4_stateid ls_stateid;
138 atomic_t ls_count; 163 atomic_t ls_count;
@@ -153,7 +178,6 @@ struct nfs4_state {
153 struct inode *inode; /* Pointer to the inode */ 178 struct inode *inode; /* Pointer to the inode */
154 179
155 unsigned long flags; /* Do we hold any locks? */ 180 unsigned long flags; /* Do we hold any locks? */
156 struct semaphore lock_sema; /* Serializes file locking operations */
157 spinlock_t state_lock; /* Protects the lock_states list */ 181 spinlock_t state_lock; /* Protects the lock_states list */
158 182
159 nfs4_stateid stateid; 183 nfs4_stateid stateid;
@@ -191,8 +215,8 @@ extern int nfs4_proc_setclientid_confirm(struct nfs4_client *);
191extern int nfs4_proc_async_renew(struct nfs4_client *); 215extern int nfs4_proc_async_renew(struct nfs4_client *);
192extern int nfs4_proc_renew(struct nfs4_client *); 216extern int nfs4_proc_renew(struct nfs4_client *);
193extern int nfs4_do_close(struct inode *inode, struct nfs4_state *state, mode_t mode); 217extern int nfs4_do_close(struct inode *inode, struct nfs4_state *state, mode_t mode);
194extern struct inode *nfs4_atomic_open(struct inode *, struct dentry *, struct nameidata *); 218extern struct dentry *nfs4_atomic_open(struct inode *, struct dentry *, struct nameidata *);
195extern int nfs4_open_revalidate(struct inode *, struct dentry *, int); 219extern int nfs4_open_revalidate(struct inode *, struct dentry *, int, struct nameidata *);
196 220
197extern struct nfs4_state_recovery_ops nfs4_reboot_recovery_ops; 221extern struct nfs4_state_recovery_ops nfs4_reboot_recovery_ops;
198extern struct nfs4_state_recovery_ops nfs4_network_partition_recovery_ops; 222extern struct nfs4_state_recovery_ops nfs4_network_partition_recovery_ops;
@@ -224,12 +248,17 @@ extern struct nfs4_state * nfs4_get_open_state(struct inode *, struct nfs4_state
224extern void nfs4_put_open_state(struct nfs4_state *); 248extern void nfs4_put_open_state(struct nfs4_state *);
225extern void nfs4_close_state(struct nfs4_state *, mode_t); 249extern void nfs4_close_state(struct nfs4_state *, mode_t);
226extern struct nfs4_state *nfs4_find_state(struct inode *, struct rpc_cred *, mode_t mode); 250extern struct nfs4_state *nfs4_find_state(struct inode *, struct rpc_cred *, mode_t mode);
227extern void nfs4_increment_seqid(int status, struct nfs4_state_owner *sp);
228extern void nfs4_schedule_state_recovery(struct nfs4_client *); 251extern void nfs4_schedule_state_recovery(struct nfs4_client *);
252extern void nfs4_put_lock_state(struct nfs4_lock_state *lsp);
229extern int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl); 253extern int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl);
230extern void nfs4_increment_lock_seqid(int status, struct nfs4_lock_state *ls);
231extern void nfs4_copy_stateid(nfs4_stateid *, struct nfs4_state *, fl_owner_t); 254extern void nfs4_copy_stateid(nfs4_stateid *, struct nfs4_state *, fl_owner_t);
232 255
256extern struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter);
257extern int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task);
258extern void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid);
259extern void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid);
260extern void nfs_free_seqid(struct nfs_seqid *seqid);
261
233extern const nfs4_stateid zero_stateid; 262extern const nfs4_stateid zero_stateid;
234 263
235/* nfs4xdr.c */ 264/* nfs4xdr.c */
diff --git a/fs/nfs/nfs4proc.c b/fs/nfs/nfs4proc.c
index 9701ca8c9428..9c1da34036aa 100644
--- a/fs/nfs/nfs4proc.c
+++ b/fs/nfs/nfs4proc.c
@@ -47,6 +47,7 @@
47#include <linux/nfs_page.h> 47#include <linux/nfs_page.h>
48#include <linux/smp_lock.h> 48#include <linux/smp_lock.h>
49#include <linux/namei.h> 49#include <linux/namei.h>
50#include <linux/mount.h>
50 51
51#include "nfs4_fs.h" 52#include "nfs4_fs.h"
52#include "delegation.h" 53#include "delegation.h"
@@ -56,10 +57,11 @@
56#define NFS4_POLL_RETRY_MIN (1*HZ) 57#define NFS4_POLL_RETRY_MIN (1*HZ)
57#define NFS4_POLL_RETRY_MAX (15*HZ) 58#define NFS4_POLL_RETRY_MAX (15*HZ)
58 59
60static int _nfs4_proc_open_confirm(struct rpc_clnt *clnt, const struct nfs_fh *fh, struct nfs4_state_owner *sp, nfs4_stateid *stateid, struct nfs_seqid *seqid);
59static int nfs4_do_fsinfo(struct nfs_server *, struct nfs_fh *, struct nfs_fsinfo *); 61static int nfs4_do_fsinfo(struct nfs_server *, struct nfs_fh *, struct nfs_fsinfo *);
60static int nfs4_async_handle_error(struct rpc_task *, struct nfs_server *); 62static int nfs4_async_handle_error(struct rpc_task *, const struct nfs_server *);
61static int _nfs4_proc_access(struct inode *inode, struct nfs_access_entry *entry); 63static int _nfs4_proc_access(struct inode *inode, struct nfs_access_entry *entry);
62static int nfs4_handle_exception(struct nfs_server *server, int errorcode, struct nfs4_exception *exception); 64static int nfs4_handle_exception(const struct nfs_server *server, int errorcode, struct nfs4_exception *exception);
63extern u32 *nfs4_decode_dirent(u32 *p, struct nfs_entry *entry, int plus); 65extern u32 *nfs4_decode_dirent(u32 *p, struct nfs_entry *entry, int plus);
64extern struct rpc_procinfo nfs4_procedures[]; 66extern struct rpc_procinfo nfs4_procedures[];
65 67
@@ -189,12 +191,28 @@ static void update_changeattr(struct inode *inode, struct nfs4_change_info *cinf
189 nfsi->change_attr = cinfo->after; 191 nfsi->change_attr = cinfo->after;
190} 192}
191 193
194/* Helper for asynchronous RPC calls */
195static int nfs4_call_async(struct rpc_clnt *clnt, rpc_action tk_begin,
196 rpc_action tk_exit, void *calldata)
197{
198 struct rpc_task *task;
199
200 if (!(task = rpc_new_task(clnt, tk_exit, RPC_TASK_ASYNC)))
201 return -ENOMEM;
202
203 task->tk_calldata = calldata;
204 task->tk_action = tk_begin;
205 rpc_execute(task);
206 return 0;
207}
208
192static void update_open_stateid(struct nfs4_state *state, nfs4_stateid *stateid, int open_flags) 209static void update_open_stateid(struct nfs4_state *state, nfs4_stateid *stateid, int open_flags)
193{ 210{
194 struct inode *inode = state->inode; 211 struct inode *inode = state->inode;
195 212
196 open_flags &= (FMODE_READ|FMODE_WRITE); 213 open_flags &= (FMODE_READ|FMODE_WRITE);
197 /* Protect against nfs4_find_state() */ 214 /* Protect against nfs4_find_state() */
215 spin_lock(&state->owner->so_lock);
198 spin_lock(&inode->i_lock); 216 spin_lock(&inode->i_lock);
199 state->state |= open_flags; 217 state->state |= open_flags;
200 /* NB! List reordering - see the reclaim code for why. */ 218 /* NB! List reordering - see the reclaim code for why. */
@@ -204,12 +222,12 @@ static void update_open_stateid(struct nfs4_state *state, nfs4_stateid *stateid,
204 state->nreaders++; 222 state->nreaders++;
205 memcpy(&state->stateid, stateid, sizeof(state->stateid)); 223 memcpy(&state->stateid, stateid, sizeof(state->stateid));
206 spin_unlock(&inode->i_lock); 224 spin_unlock(&inode->i_lock);
225 spin_unlock(&state->owner->so_lock);
207} 226}
208 227
209/* 228/*
210 * OPEN_RECLAIM: 229 * OPEN_RECLAIM:
211 * reclaim state on the server after a reboot. 230 * reclaim state on the server after a reboot.
212 * Assumes caller is holding the sp->so_sem
213 */ 231 */
214static int _nfs4_open_reclaim(struct nfs4_state_owner *sp, struct nfs4_state *state) 232static int _nfs4_open_reclaim(struct nfs4_state_owner *sp, struct nfs4_state *state)
215{ 233{
@@ -218,7 +236,6 @@ static int _nfs4_open_reclaim(struct nfs4_state_owner *sp, struct nfs4_state *st
218 struct nfs_delegation *delegation = NFS_I(inode)->delegation; 236 struct nfs_delegation *delegation = NFS_I(inode)->delegation;
219 struct nfs_openargs o_arg = { 237 struct nfs_openargs o_arg = {
220 .fh = NFS_FH(inode), 238 .fh = NFS_FH(inode),
221 .seqid = sp->so_seqid,
222 .id = sp->so_id, 239 .id = sp->so_id,
223 .open_flags = state->state, 240 .open_flags = state->state,
224 .clientid = server->nfs4_state->cl_clientid, 241 .clientid = server->nfs4_state->cl_clientid,
@@ -245,8 +262,13 @@ static int _nfs4_open_reclaim(struct nfs4_state_owner *sp, struct nfs4_state *st
245 } 262 }
246 o_arg.u.delegation_type = delegation->type; 263 o_arg.u.delegation_type = delegation->type;
247 } 264 }
265 o_arg.seqid = nfs_alloc_seqid(&sp->so_seqid);
266 if (o_arg.seqid == NULL)
267 return -ENOMEM;
248 status = rpc_call_sync(server->client, &msg, RPC_TASK_NOINTR); 268 status = rpc_call_sync(server->client, &msg, RPC_TASK_NOINTR);
249 nfs4_increment_seqid(status, sp); 269 /* Confirm the sequence as being established */
270 nfs_confirm_seqid(&sp->so_seqid, status);
271 nfs_increment_open_seqid(status, o_arg.seqid);
250 if (status == 0) { 272 if (status == 0) {
251 memcpy(&state->stateid, &o_res.stateid, sizeof(state->stateid)); 273 memcpy(&state->stateid, &o_res.stateid, sizeof(state->stateid));
252 if (o_res.delegation_type != 0) { 274 if (o_res.delegation_type != 0) {
@@ -256,6 +278,7 @@ static int _nfs4_open_reclaim(struct nfs4_state_owner *sp, struct nfs4_state *st
256 nfs_async_inode_return_delegation(inode, &o_res.stateid); 278 nfs_async_inode_return_delegation(inode, &o_res.stateid);
257 } 279 }
258 } 280 }
281 nfs_free_seqid(o_arg.seqid);
259 clear_bit(NFS_DELEGATED_STATE, &state->flags); 282 clear_bit(NFS_DELEGATED_STATE, &state->flags);
260 /* Ensure we update the inode attributes */ 283 /* Ensure we update the inode attributes */
261 NFS_CACHEINV(inode); 284 NFS_CACHEINV(inode);
@@ -302,23 +325,35 @@ static int _nfs4_open_delegation_recall(struct dentry *dentry, struct nfs4_state
302 }; 325 };
303 int status = 0; 326 int status = 0;
304 327
305 down(&sp->so_sema);
306 if (!test_bit(NFS_DELEGATED_STATE, &state->flags)) 328 if (!test_bit(NFS_DELEGATED_STATE, &state->flags))
307 goto out; 329 goto out;
308 if (state->state == 0) 330 if (state->state == 0)
309 goto out; 331 goto out;
310 arg.seqid = sp->so_seqid; 332 arg.seqid = nfs_alloc_seqid(&sp->so_seqid);
333 status = -ENOMEM;
334 if (arg.seqid == NULL)
335 goto out;
311 arg.open_flags = state->state; 336 arg.open_flags = state->state;
312 memcpy(arg.u.delegation.data, state->stateid.data, sizeof(arg.u.delegation.data)); 337 memcpy(arg.u.delegation.data, state->stateid.data, sizeof(arg.u.delegation.data));
313 status = rpc_call_sync(server->client, &msg, RPC_TASK_NOINTR); 338 status = rpc_call_sync(server->client, &msg, RPC_TASK_NOINTR);
314 nfs4_increment_seqid(status, sp); 339 nfs_increment_open_seqid(status, arg.seqid);
340 if (status != 0)
341 goto out_free;
342 if(res.rflags & NFS4_OPEN_RESULT_CONFIRM) {
343 status = _nfs4_proc_open_confirm(server->client, NFS_FH(inode),
344 sp, &res.stateid, arg.seqid);
345 if (status != 0)
346 goto out_free;
347 }
348 nfs_confirm_seqid(&sp->so_seqid, 0);
315 if (status >= 0) { 349 if (status >= 0) {
316 memcpy(state->stateid.data, res.stateid.data, 350 memcpy(state->stateid.data, res.stateid.data,
317 sizeof(state->stateid.data)); 351 sizeof(state->stateid.data));
318 clear_bit(NFS_DELEGATED_STATE, &state->flags); 352 clear_bit(NFS_DELEGATED_STATE, &state->flags);
319 } 353 }
354out_free:
355 nfs_free_seqid(arg.seqid);
320out: 356out:
321 up(&sp->so_sema);
322 dput(parent); 357 dput(parent);
323 return status; 358 return status;
324} 359}
@@ -345,11 +380,11 @@ int nfs4_open_delegation_recall(struct dentry *dentry, struct nfs4_state *state)
345 return err; 380 return err;
346} 381}
347 382
348static inline int _nfs4_proc_open_confirm(struct rpc_clnt *clnt, const struct nfs_fh *fh, struct nfs4_state_owner *sp, nfs4_stateid *stateid) 383static int _nfs4_proc_open_confirm(struct rpc_clnt *clnt, const struct nfs_fh *fh, struct nfs4_state_owner *sp, nfs4_stateid *stateid, struct nfs_seqid *seqid)
349{ 384{
350 struct nfs_open_confirmargs arg = { 385 struct nfs_open_confirmargs arg = {
351 .fh = fh, 386 .fh = fh,
352 .seqid = sp->so_seqid, 387 .seqid = seqid,
353 .stateid = *stateid, 388 .stateid = *stateid,
354 }; 389 };
355 struct nfs_open_confirmres res; 390 struct nfs_open_confirmres res;
@@ -362,7 +397,9 @@ static inline int _nfs4_proc_open_confirm(struct rpc_clnt *clnt, const struct nf
362 int status; 397 int status;
363 398
364 status = rpc_call_sync(clnt, &msg, RPC_TASK_NOINTR); 399 status = rpc_call_sync(clnt, &msg, RPC_TASK_NOINTR);
365 nfs4_increment_seqid(status, sp); 400 /* Confirm the sequence as being established */
401 nfs_confirm_seqid(&sp->so_seqid, status);
402 nfs_increment_open_seqid(status, seqid);
366 if (status >= 0) 403 if (status >= 0)
367 memcpy(stateid, &res.stateid, sizeof(*stateid)); 404 memcpy(stateid, &res.stateid, sizeof(*stateid));
368 return status; 405 return status;
@@ -380,21 +417,37 @@ static int _nfs4_proc_open(struct inode *dir, struct nfs4_state_owner *sp, stru
380 int status; 417 int status;
381 418
382 /* Update sequence id. The caller must serialize! */ 419 /* Update sequence id. The caller must serialize! */
383 o_arg->seqid = sp->so_seqid;
384 o_arg->id = sp->so_id; 420 o_arg->id = sp->so_id;
385 o_arg->clientid = sp->so_client->cl_clientid; 421 o_arg->clientid = sp->so_client->cl_clientid;
386 422
387 status = rpc_call_sync(server->client, &msg, RPC_TASK_NOINTR); 423 status = rpc_call_sync(server->client, &msg, RPC_TASK_NOINTR);
388 nfs4_increment_seqid(status, sp); 424 if (status == 0) {
425 /* OPEN on anything except a regular file is disallowed in NFSv4 */
426 switch (o_res->f_attr->mode & S_IFMT) {
427 case S_IFREG:
428 break;
429 case S_IFLNK:
430 status = -ELOOP;
431 break;
432 case S_IFDIR:
433 status = -EISDIR;
434 break;
435 default:
436 status = -ENOTDIR;
437 }
438 }
439
440 nfs_increment_open_seqid(status, o_arg->seqid);
389 if (status != 0) 441 if (status != 0)
390 goto out; 442 goto out;
391 update_changeattr(dir, &o_res->cinfo); 443 update_changeattr(dir, &o_res->cinfo);
392 if(o_res->rflags & NFS4_OPEN_RESULT_CONFIRM) { 444 if(o_res->rflags & NFS4_OPEN_RESULT_CONFIRM) {
393 status = _nfs4_proc_open_confirm(server->client, &o_res->fh, 445 status = _nfs4_proc_open_confirm(server->client, &o_res->fh,
394 sp, &o_res->stateid); 446 sp, &o_res->stateid, o_arg->seqid);
395 if (status != 0) 447 if (status != 0)
396 goto out; 448 goto out;
397 } 449 }
450 nfs_confirm_seqid(&sp->so_seqid, 0);
398 if (!(o_res->f_attr->valid & NFS_ATTR_FATTR)) 451 if (!(o_res->f_attr->valid & NFS_ATTR_FATTR))
399 status = server->rpc_ops->getattr(server, &o_res->fh, o_res->f_attr); 452 status = server->rpc_ops->getattr(server, &o_res->fh, o_res->f_attr);
400out: 453out:
@@ -465,6 +518,10 @@ static int _nfs4_open_expired(struct nfs4_state_owner *sp, struct nfs4_state *st
465 set_bit(NFS_DELEGATED_STATE, &state->flags); 518 set_bit(NFS_DELEGATED_STATE, &state->flags);
466 goto out; 519 goto out;
467 } 520 }
521 o_arg.seqid = nfs_alloc_seqid(&sp->so_seqid);
522 status = -ENOMEM;
523 if (o_arg.seqid == NULL)
524 goto out;
468 status = _nfs4_proc_open(dir, sp, &o_arg, &o_res); 525 status = _nfs4_proc_open(dir, sp, &o_arg, &o_res);
469 if (status != 0) 526 if (status != 0)
470 goto out_nodeleg; 527 goto out_nodeleg;
@@ -490,6 +547,7 @@ static int _nfs4_open_expired(struct nfs4_state_owner *sp, struct nfs4_state *st
490 nfs_inode_reclaim_delegation(inode, sp->so_cred, &o_res); 547 nfs_inode_reclaim_delegation(inode, sp->so_cred, &o_res);
491 } 548 }
492out_nodeleg: 549out_nodeleg:
550 nfs_free_seqid(o_arg.seqid);
493 clear_bit(NFS_DELEGATED_STATE, &state->flags); 551 clear_bit(NFS_DELEGATED_STATE, &state->flags);
494out: 552out:
495 dput(parent); 553 dput(parent);
@@ -564,7 +622,6 @@ static int _nfs4_open_delegated(struct inode *inode, int flags, struct rpc_cred
564 dprintk("%s: nfs4_get_state_owner failed!\n", __FUNCTION__); 622 dprintk("%s: nfs4_get_state_owner failed!\n", __FUNCTION__);
565 goto out_err; 623 goto out_err;
566 } 624 }
567 down(&sp->so_sema);
568 state = nfs4_get_open_state(inode, sp); 625 state = nfs4_get_open_state(inode, sp);
569 if (state == NULL) 626 if (state == NULL)
570 goto out_err; 627 goto out_err;
@@ -589,7 +646,6 @@ static int _nfs4_open_delegated(struct inode *inode, int flags, struct rpc_cred
589 set_bit(NFS_DELEGATED_STATE, &state->flags); 646 set_bit(NFS_DELEGATED_STATE, &state->flags);
590 update_open_stateid(state, &delegation->stateid, open_flags); 647 update_open_stateid(state, &delegation->stateid, open_flags);
591out_ok: 648out_ok:
592 up(&sp->so_sema);
593 nfs4_put_state_owner(sp); 649 nfs4_put_state_owner(sp);
594 up_read(&nfsi->rwsem); 650 up_read(&nfsi->rwsem);
595 up_read(&clp->cl_sem); 651 up_read(&clp->cl_sem);
@@ -600,11 +656,12 @@ out_err:
600 if (sp != NULL) { 656 if (sp != NULL) {
601 if (state != NULL) 657 if (state != NULL)
602 nfs4_put_open_state(state); 658 nfs4_put_open_state(state);
603 up(&sp->so_sema);
604 nfs4_put_state_owner(sp); 659 nfs4_put_state_owner(sp);
605 } 660 }
606 up_read(&nfsi->rwsem); 661 up_read(&nfsi->rwsem);
607 up_read(&clp->cl_sem); 662 up_read(&clp->cl_sem);
663 if (err != -EACCES)
664 nfs_inode_return_delegation(inode);
608 return err; 665 return err;
609} 666}
610 667
@@ -665,8 +722,10 @@ static int _nfs4_do_open(struct inode *dir, struct dentry *dentry, int flags, st
665 } else 722 } else
666 o_arg.u.attrs = sattr; 723 o_arg.u.attrs = sattr;
667 /* Serialization for the sequence id */ 724 /* Serialization for the sequence id */
668 down(&sp->so_sema);
669 725
726 o_arg.seqid = nfs_alloc_seqid(&sp->so_seqid);
727 if (o_arg.seqid == NULL)
728 return -ENOMEM;
670 status = _nfs4_proc_open(dir, sp, &o_arg, &o_res); 729 status = _nfs4_proc_open(dir, sp, &o_arg, &o_res);
671 if (status != 0) 730 if (status != 0)
672 goto out_err; 731 goto out_err;
@@ -681,7 +740,7 @@ static int _nfs4_do_open(struct inode *dir, struct dentry *dentry, int flags, st
681 update_open_stateid(state, &o_res.stateid, flags); 740 update_open_stateid(state, &o_res.stateid, flags);
682 if (o_res.delegation_type != 0) 741 if (o_res.delegation_type != 0)
683 nfs_inode_set_delegation(inode, cred, &o_res); 742 nfs_inode_set_delegation(inode, cred, &o_res);
684 up(&sp->so_sema); 743 nfs_free_seqid(o_arg.seqid);
685 nfs4_put_state_owner(sp); 744 nfs4_put_state_owner(sp);
686 up_read(&clp->cl_sem); 745 up_read(&clp->cl_sem);
687 *res = state; 746 *res = state;
@@ -690,7 +749,7 @@ out_err:
690 if (sp != NULL) { 749 if (sp != NULL) {
691 if (state != NULL) 750 if (state != NULL)
692 nfs4_put_open_state(state); 751 nfs4_put_open_state(state);
693 up(&sp->so_sema); 752 nfs_free_seqid(o_arg.seqid);
694 nfs4_put_state_owner(sp); 753 nfs4_put_state_owner(sp);
695 } 754 }
696 /* Note: clp->cl_sem must be released before nfs4_put_open_state()! */ 755 /* Note: clp->cl_sem must be released before nfs4_put_open_state()! */
@@ -718,7 +777,7 @@ static struct nfs4_state *nfs4_do_open(struct inode *dir, struct dentry *dentry,
718 * It is actually a sign of a bug on the client or on the server. 777 * It is actually a sign of a bug on the client or on the server.
719 * 778 *
720 * If we receive a BAD_SEQID error in the particular case of 779 * If we receive a BAD_SEQID error in the particular case of
721 * doing an OPEN, we assume that nfs4_increment_seqid() will 780 * doing an OPEN, we assume that nfs_increment_open_seqid() will
722 * have unhashed the old state_owner for us, and that we can 781 * have unhashed the old state_owner for us, and that we can
723 * therefore safely retry using a new one. We should still warn 782 * therefore safely retry using a new one. We should still warn
724 * the user though... 783 * the user though...
@@ -728,6 +787,16 @@ static struct nfs4_state *nfs4_do_open(struct inode *dir, struct dentry *dentry,
728 exception.retry = 1; 787 exception.retry = 1;
729 continue; 788 continue;
730 } 789 }
790 /*
791 * BAD_STATEID on OPEN means that the server cancelled our
792 * state before it received the OPEN_CONFIRM.
793 * Recover by retrying the request as per the discussion
794 * on Page 181 of RFC3530.
795 */
796 if (status == -NFS4ERR_BAD_STATEID) {
797 exception.retry = 1;
798 continue;
799 }
731 res = ERR_PTR(nfs4_handle_exception(NFS_SERVER(dir), 800 res = ERR_PTR(nfs4_handle_exception(NFS_SERVER(dir),
732 status, &exception)); 801 status, &exception));
733 } while (exception.retry); 802 } while (exception.retry);
@@ -789,17 +858,27 @@ struct nfs4_closedata {
789 struct nfs_closeres res; 858 struct nfs_closeres res;
790}; 859};
791 860
861static void nfs4_free_closedata(struct nfs4_closedata *calldata)
862{
863 struct nfs4_state *state = calldata->state;
864 struct nfs4_state_owner *sp = state->owner;
865
866 nfs4_put_open_state(calldata->state);
867 nfs_free_seqid(calldata->arg.seqid);
868 nfs4_put_state_owner(sp);
869 kfree(calldata);
870}
871
792static void nfs4_close_done(struct rpc_task *task) 872static void nfs4_close_done(struct rpc_task *task)
793{ 873{
794 struct nfs4_closedata *calldata = (struct nfs4_closedata *)task->tk_calldata; 874 struct nfs4_closedata *calldata = (struct nfs4_closedata *)task->tk_calldata;
795 struct nfs4_state *state = calldata->state; 875 struct nfs4_state *state = calldata->state;
796 struct nfs4_state_owner *sp = state->owner;
797 struct nfs_server *server = NFS_SERVER(calldata->inode); 876 struct nfs_server *server = NFS_SERVER(calldata->inode);
798 877
799 /* hmm. we are done with the inode, and in the process of freeing 878 /* hmm. we are done with the inode, and in the process of freeing
800 * the state_owner. we keep this around to process errors 879 * the state_owner. we keep this around to process errors
801 */ 880 */
802 nfs4_increment_seqid(task->tk_status, sp); 881 nfs_increment_open_seqid(task->tk_status, calldata->arg.seqid);
803 switch (task->tk_status) { 882 switch (task->tk_status) {
804 case 0: 883 case 0:
805 memcpy(&state->stateid, &calldata->res.stateid, 884 memcpy(&state->stateid, &calldata->res.stateid,
@@ -817,24 +896,46 @@ static void nfs4_close_done(struct rpc_task *task)
817 } 896 }
818 } 897 }
819 state->state = calldata->arg.open_flags; 898 state->state = calldata->arg.open_flags;
820 nfs4_put_open_state(state); 899 nfs4_free_closedata(calldata);
821 up(&sp->so_sema);
822 nfs4_put_state_owner(sp);
823 up_read(&server->nfs4_state->cl_sem);
824 kfree(calldata);
825} 900}
826 901
827static inline int nfs4_close_call(struct rpc_clnt *clnt, struct nfs4_closedata *calldata) 902static void nfs4_close_begin(struct rpc_task *task)
828{ 903{
904 struct nfs4_closedata *calldata = (struct nfs4_closedata *)task->tk_calldata;
905 struct nfs4_state *state = calldata->state;
829 struct rpc_message msg = { 906 struct rpc_message msg = {
830 .rpc_proc = &nfs4_procedures[NFSPROC4_CLNT_CLOSE], 907 .rpc_proc = &nfs4_procedures[NFSPROC4_CLNT_CLOSE],
831 .rpc_argp = &calldata->arg, 908 .rpc_argp = &calldata->arg,
832 .rpc_resp = &calldata->res, 909 .rpc_resp = &calldata->res,
833 .rpc_cred = calldata->state->owner->so_cred, 910 .rpc_cred = state->owner->so_cred,
834 }; 911 };
835 if (calldata->arg.open_flags != 0) 912 int mode = 0;
913 int status;
914
915 status = nfs_wait_on_sequence(calldata->arg.seqid, task);
916 if (status != 0)
917 return;
918 /* Don't reorder reads */
919 smp_rmb();
920 /* Recalculate the new open mode in case someone reopened the file
921 * while we were waiting in line to be scheduled.
922 */
923 if (state->nreaders != 0)
924 mode |= FMODE_READ;
925 if (state->nwriters != 0)
926 mode |= FMODE_WRITE;
927 if (test_bit(NFS_DELEGATED_STATE, &state->flags))
928 state->state = mode;
929 if (mode == state->state) {
930 nfs4_free_closedata(calldata);
931 task->tk_exit = NULL;
932 rpc_exit(task, 0);
933 return;
934 }
935 if (mode != 0)
836 msg.rpc_proc = &nfs4_procedures[NFSPROC4_CLNT_OPEN_DOWNGRADE]; 936 msg.rpc_proc = &nfs4_procedures[NFSPROC4_CLNT_OPEN_DOWNGRADE];
837 return rpc_call_async(clnt, &msg, 0, nfs4_close_done, calldata); 937 calldata->arg.open_flags = mode;
938 rpc_call_setup(task, &msg, 0);
838} 939}
839 940
840/* 941/*
@@ -851,39 +952,52 @@ static inline int nfs4_close_call(struct rpc_clnt *clnt, struct nfs4_closedata *
851int nfs4_do_close(struct inode *inode, struct nfs4_state *state, mode_t mode) 952int nfs4_do_close(struct inode *inode, struct nfs4_state *state, mode_t mode)
852{ 953{
853 struct nfs4_closedata *calldata; 954 struct nfs4_closedata *calldata;
854 int status; 955 int status = -ENOMEM;
855 956
856 /* Tell caller we're done */ 957 calldata = kmalloc(sizeof(*calldata), GFP_KERNEL);
857 if (test_bit(NFS_DELEGATED_STATE, &state->flags)) {
858 state->state = mode;
859 return 0;
860 }
861 calldata = (struct nfs4_closedata *)kmalloc(sizeof(*calldata), GFP_KERNEL);
862 if (calldata == NULL) 958 if (calldata == NULL)
863 return -ENOMEM; 959 goto out;
864 calldata->inode = inode; 960 calldata->inode = inode;
865 calldata->state = state; 961 calldata->state = state;
866 calldata->arg.fh = NFS_FH(inode); 962 calldata->arg.fh = NFS_FH(inode);
963 calldata->arg.stateid = &state->stateid;
867 /* Serialization for the sequence id */ 964 /* Serialization for the sequence id */
868 calldata->arg.seqid = state->owner->so_seqid; 965 calldata->arg.seqid = nfs_alloc_seqid(&state->owner->so_seqid);
869 calldata->arg.open_flags = mode; 966 if (calldata->arg.seqid == NULL)
870 memcpy(&calldata->arg.stateid, &state->stateid, 967 goto out_free_calldata;
871 sizeof(calldata->arg.stateid)); 968
872 status = nfs4_close_call(NFS_SERVER(inode)->client, calldata); 969 status = nfs4_call_async(NFS_SERVER(inode)->client, nfs4_close_begin,
873 /* 970 nfs4_close_done, calldata);
874 * Return -EINPROGRESS on success in order to indicate to the 971 if (status == 0)
875 * caller that an asynchronous RPC call has been launched, and 972 goto out;
876 * that it will release the semaphores on completion. 973
877 */ 974 nfs_free_seqid(calldata->arg.seqid);
878 return (status == 0) ? -EINPROGRESS : status; 975out_free_calldata:
976 kfree(calldata);
977out:
978 return status;
979}
980
981static void nfs4_intent_set_file(struct nameidata *nd, struct dentry *dentry, struct nfs4_state *state)
982{
983 struct file *filp;
984
985 filp = lookup_instantiate_filp(nd, dentry, NULL);
986 if (!IS_ERR(filp)) {
987 struct nfs_open_context *ctx;
988 ctx = (struct nfs_open_context *)filp->private_data;
989 ctx->state = state;
990 } else
991 nfs4_close_state(state, nd->intent.open.flags);
879} 992}
880 993
881struct inode * 994struct dentry *
882nfs4_atomic_open(struct inode *dir, struct dentry *dentry, struct nameidata *nd) 995nfs4_atomic_open(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
883{ 996{
884 struct iattr attr; 997 struct iattr attr;
885 struct rpc_cred *cred; 998 struct rpc_cred *cred;
886 struct nfs4_state *state; 999 struct nfs4_state *state;
1000 struct dentry *res;
887 1001
888 if (nd->flags & LOOKUP_CREATE) { 1002 if (nd->flags & LOOKUP_CREATE) {
889 attr.ia_mode = nd->intent.open.create_mode; 1003 attr.ia_mode = nd->intent.open.create_mode;
@@ -897,16 +1011,23 @@ nfs4_atomic_open(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
897 1011
898 cred = rpcauth_lookupcred(NFS_SERVER(dir)->client->cl_auth, 0); 1012 cred = rpcauth_lookupcred(NFS_SERVER(dir)->client->cl_auth, 0);
899 if (IS_ERR(cred)) 1013 if (IS_ERR(cred))
900 return (struct inode *)cred; 1014 return (struct dentry *)cred;
901 state = nfs4_do_open(dir, dentry, nd->intent.open.flags, &attr, cred); 1015 state = nfs4_do_open(dir, dentry, nd->intent.open.flags, &attr, cred);
902 put_rpccred(cred); 1016 put_rpccred(cred);
903 if (IS_ERR(state)) 1017 if (IS_ERR(state)) {
904 return (struct inode *)state; 1018 if (PTR_ERR(state) == -ENOENT)
905 return state->inode; 1019 d_add(dentry, NULL);
1020 return (struct dentry *)state;
1021 }
1022 res = d_add_unique(dentry, state->inode);
1023 if (res != NULL)
1024 dentry = res;
1025 nfs4_intent_set_file(nd, dentry, state);
1026 return res;
906} 1027}
907 1028
908int 1029int
909nfs4_open_revalidate(struct inode *dir, struct dentry *dentry, int openflags) 1030nfs4_open_revalidate(struct inode *dir, struct dentry *dentry, int openflags, struct nameidata *nd)
910{ 1031{
911 struct rpc_cred *cred; 1032 struct rpc_cred *cred;
912 struct nfs4_state *state; 1033 struct nfs4_state *state;
@@ -919,18 +1040,30 @@ nfs4_open_revalidate(struct inode *dir, struct dentry *dentry, int openflags)
919 if (IS_ERR(state)) 1040 if (IS_ERR(state))
920 state = nfs4_do_open(dir, dentry, openflags, NULL, cred); 1041 state = nfs4_do_open(dir, dentry, openflags, NULL, cred);
921 put_rpccred(cred); 1042 put_rpccred(cred);
922 if (state == ERR_PTR(-ENOENT) && dentry->d_inode == 0) 1043 if (IS_ERR(state)) {
923 return 1; 1044 switch (PTR_ERR(state)) {
924 if (IS_ERR(state)) 1045 case -EPERM:
925 return 0; 1046 case -EACCES:
1047 case -EDQUOT:
1048 case -ENOSPC:
1049 case -EROFS:
1050 lookup_instantiate_filp(nd, (struct dentry *)state, NULL);
1051 return 1;
1052 case -ENOENT:
1053 if (dentry->d_inode == NULL)
1054 return 1;
1055 }
1056 goto out_drop;
1057 }
926 inode = state->inode; 1058 inode = state->inode;
1059 iput(inode);
927 if (inode == dentry->d_inode) { 1060 if (inode == dentry->d_inode) {
928 iput(inode); 1061 nfs4_intent_set_file(nd, dentry, state);
929 return 1; 1062 return 1;
930 } 1063 }
931 d_drop(dentry);
932 nfs4_close_state(state, openflags); 1064 nfs4_close_state(state, openflags);
933 iput(inode); 1065out_drop:
1066 d_drop(dentry);
934 return 0; 1067 return 0;
935} 1068}
936 1069
@@ -1431,7 +1564,7 @@ static int nfs4_proc_commit(struct nfs_write_data *cdata)
1431 1564
1432static int 1565static int
1433nfs4_proc_create(struct inode *dir, struct dentry *dentry, struct iattr *sattr, 1566nfs4_proc_create(struct inode *dir, struct dentry *dentry, struct iattr *sattr,
1434 int flags) 1567 int flags, struct nameidata *nd)
1435{ 1568{
1436 struct nfs4_state *state; 1569 struct nfs4_state *state;
1437 struct rpc_cred *cred; 1570 struct rpc_cred *cred;
@@ -1453,13 +1586,13 @@ nfs4_proc_create(struct inode *dir, struct dentry *dentry, struct iattr *sattr,
1453 struct nfs_fattr fattr; 1586 struct nfs_fattr fattr;
1454 status = nfs4_do_setattr(NFS_SERVER(dir), &fattr, 1587 status = nfs4_do_setattr(NFS_SERVER(dir), &fattr,
1455 NFS_FH(state->inode), sattr, state); 1588 NFS_FH(state->inode), sattr, state);
1456 if (status == 0) { 1589 if (status == 0)
1457 nfs_setattr_update_inode(state->inode, sattr); 1590 nfs_setattr_update_inode(state->inode, sattr);
1458 goto out; 1591 }
1459 } 1592 if (status == 0 && nd != NULL && (nd->flags & LOOKUP_OPEN))
1460 } else if (flags != 0) 1593 nfs4_intent_set_file(nd, dentry, state);
1461 goto out; 1594 else
1462 nfs4_close_state(state, flags); 1595 nfs4_close_state(state, flags);
1463out: 1596out:
1464 return status; 1597 return status;
1465} 1598}
@@ -2106,65 +2239,6 @@ nfs4_proc_renew(struct nfs4_client *clp)
2106 return 0; 2239 return 0;
2107} 2240}
2108 2241
2109/*
2110 * We will need to arrange for the VFS layer to provide an atomic open.
2111 * Until then, this open method is prone to inefficiency and race conditions
2112 * due to the lookup, potential create, and open VFS calls from sys_open()
2113 * placed on the wire.
2114 */
2115static int
2116nfs4_proc_file_open(struct inode *inode, struct file *filp)
2117{
2118 struct dentry *dentry = filp->f_dentry;
2119 struct nfs_open_context *ctx;
2120 struct nfs4_state *state = NULL;
2121 struct rpc_cred *cred;
2122 int status = -ENOMEM;
2123
2124 dprintk("nfs4_proc_file_open: starting on (%.*s/%.*s)\n",
2125 (int)dentry->d_parent->d_name.len,
2126 dentry->d_parent->d_name.name,
2127 (int)dentry->d_name.len, dentry->d_name.name);
2128
2129
2130 /* Find our open stateid */
2131 cred = rpcauth_lookupcred(NFS_SERVER(inode)->client->cl_auth, 0);
2132 if (IS_ERR(cred))
2133 return PTR_ERR(cred);
2134 ctx = alloc_nfs_open_context(dentry, cred);
2135 put_rpccred(cred);
2136 if (unlikely(ctx == NULL))
2137 return -ENOMEM;
2138 status = -EIO; /* ERACE actually */
2139 state = nfs4_find_state(inode, cred, filp->f_mode);
2140 if (unlikely(state == NULL))
2141 goto no_state;
2142 ctx->state = state;
2143 nfs4_close_state(state, filp->f_mode);
2144 ctx->mode = filp->f_mode;
2145 nfs_file_set_open_context(filp, ctx);
2146 put_nfs_open_context(ctx);
2147 if (filp->f_mode & FMODE_WRITE)
2148 nfs_begin_data_update(inode);
2149 return 0;
2150no_state:
2151 printk(KERN_WARNING "NFS: v4 raced in function %s\n", __FUNCTION__);
2152 put_nfs_open_context(ctx);
2153 return status;
2154}
2155
2156/*
2157 * Release our state
2158 */
2159static int
2160nfs4_proc_file_release(struct inode *inode, struct file *filp)
2161{
2162 if (filp->f_mode & FMODE_WRITE)
2163 nfs_end_data_update(inode);
2164 nfs_file_clear_open_context(filp);
2165 return 0;
2166}
2167
2168static inline int nfs4_server_supports_acls(struct nfs_server *server) 2242static inline int nfs4_server_supports_acls(struct nfs_server *server)
2169{ 2243{
2170 return (server->caps & NFS_CAP_ACLS) 2244 return (server->caps & NFS_CAP_ACLS)
@@ -2285,7 +2359,7 @@ static inline ssize_t nfs4_get_acl_uncached(struct inode *inode, void *buf, size
2285 return -ENOMEM; 2359 return -ENOMEM;
2286 args.acl_pages[0] = localpage; 2360 args.acl_pages[0] = localpage;
2287 args.acl_pgbase = 0; 2361 args.acl_pgbase = 0;
2288 args.acl_len = PAGE_SIZE; 2362 resp_len = args.acl_len = PAGE_SIZE;
2289 } else { 2363 } else {
2290 resp_buf = buf; 2364 resp_buf = buf;
2291 buf_to_pages(buf, buflen, args.acl_pages, &args.acl_pgbase); 2365 buf_to_pages(buf, buflen, args.acl_pages, &args.acl_pgbase);
@@ -2345,6 +2419,7 @@ static int nfs4_proc_set_acl(struct inode *inode, const void *buf, size_t buflen
2345 2419
2346 if (!nfs4_server_supports_acls(server)) 2420 if (!nfs4_server_supports_acls(server))
2347 return -EOPNOTSUPP; 2421 return -EOPNOTSUPP;
2422 nfs_inode_return_delegation(inode);
2348 buf_to_pages(buf, buflen, arg.acl_pages, &arg.acl_pgbase); 2423 buf_to_pages(buf, buflen, arg.acl_pages, &arg.acl_pgbase);
2349 ret = rpc_call_sync(NFS_SERVER(inode)->client, &msg, 0); 2424 ret = rpc_call_sync(NFS_SERVER(inode)->client, &msg, 0);
2350 if (ret == 0) 2425 if (ret == 0)
@@ -2353,7 +2428,7 @@ static int nfs4_proc_set_acl(struct inode *inode, const void *buf, size_t buflen
2353} 2428}
2354 2429
2355static int 2430static int
2356nfs4_async_handle_error(struct rpc_task *task, struct nfs_server *server) 2431nfs4_async_handle_error(struct rpc_task *task, const struct nfs_server *server)
2357{ 2432{
2358 struct nfs4_client *clp = server->nfs4_state; 2433 struct nfs4_client *clp = server->nfs4_state;
2359 2434
@@ -2431,7 +2506,7 @@ static int nfs4_delay(struct rpc_clnt *clnt, long *timeout)
2431/* This is the error handling routine for processes that are allowed 2506/* This is the error handling routine for processes that are allowed
2432 * to sleep. 2507 * to sleep.
2433 */ 2508 */
2434int nfs4_handle_exception(struct nfs_server *server, int errorcode, struct nfs4_exception *exception) 2509int nfs4_handle_exception(const struct nfs_server *server, int errorcode, struct nfs4_exception *exception)
2435{ 2510{
2436 struct nfs4_client *clp = server->nfs4_state; 2511 struct nfs4_client *clp = server->nfs4_state;
2437 int ret = errorcode; 2512 int ret = errorcode;
@@ -2632,7 +2707,6 @@ static int _nfs4_proc_getlk(struct nfs4_state *state, int cmd, struct file_lock
2632 2707
2633 down_read(&clp->cl_sem); 2708 down_read(&clp->cl_sem);
2634 nlo.clientid = clp->cl_clientid; 2709 nlo.clientid = clp->cl_clientid;
2635 down(&state->lock_sema);
2636 status = nfs4_set_lock_state(state, request); 2710 status = nfs4_set_lock_state(state, request);
2637 if (status != 0) 2711 if (status != 0)
2638 goto out; 2712 goto out;
@@ -2659,7 +2733,6 @@ static int _nfs4_proc_getlk(struct nfs4_state *state, int cmd, struct file_lock
2659 status = 0; 2733 status = 0;
2660 } 2734 }
2661out: 2735out:
2662 up(&state->lock_sema);
2663 up_read(&clp->cl_sem); 2736 up_read(&clp->cl_sem);
2664 return status; 2737 return status;
2665} 2738}
@@ -2696,79 +2769,149 @@ static int do_vfs_lock(struct file *file, struct file_lock *fl)
2696 return res; 2769 return res;
2697} 2770}
2698 2771
2699static int _nfs4_proc_unlck(struct nfs4_state *state, int cmd, struct file_lock *request) 2772struct nfs4_unlockdata {
2773 struct nfs_lockargs arg;
2774 struct nfs_locku_opargs luargs;
2775 struct nfs_lockres res;
2776 struct nfs4_lock_state *lsp;
2777 struct nfs_open_context *ctx;
2778 atomic_t refcount;
2779 struct completion completion;
2780};
2781
2782static void nfs4_locku_release_calldata(struct nfs4_unlockdata *calldata)
2700{ 2783{
2701 struct inode *inode = state->inode; 2784 if (atomic_dec_and_test(&calldata->refcount)) {
2702 struct nfs_server *server = NFS_SERVER(inode); 2785 nfs_free_seqid(calldata->luargs.seqid);
2703 struct nfs4_client *clp = server->nfs4_state; 2786 nfs4_put_lock_state(calldata->lsp);
2704 struct nfs_lockargs arg = { 2787 put_nfs_open_context(calldata->ctx);
2705 .fh = NFS_FH(inode), 2788 kfree(calldata);
2706 .type = nfs4_lck_type(cmd, request), 2789 }
2707 .offset = request->fl_start, 2790}
2708 .length = nfs4_lck_length(request), 2791
2709 }; 2792static void nfs4_locku_complete(struct nfs4_unlockdata *calldata)
2710 struct nfs_lockres res = { 2793{
2711 .server = server, 2794 complete(&calldata->completion);
2712 }; 2795 nfs4_locku_release_calldata(calldata);
2796}
2797
2798static void nfs4_locku_done(struct rpc_task *task)
2799{
2800 struct nfs4_unlockdata *calldata = (struct nfs4_unlockdata *)task->tk_calldata;
2801
2802 nfs_increment_lock_seqid(task->tk_status, calldata->luargs.seqid);
2803 switch (task->tk_status) {
2804 case 0:
2805 memcpy(calldata->lsp->ls_stateid.data,
2806 calldata->res.u.stateid.data,
2807 sizeof(calldata->lsp->ls_stateid.data));
2808 break;
2809 case -NFS4ERR_STALE_STATEID:
2810 case -NFS4ERR_EXPIRED:
2811 nfs4_schedule_state_recovery(calldata->res.server->nfs4_state);
2812 break;
2813 default:
2814 if (nfs4_async_handle_error(task, calldata->res.server) == -EAGAIN) {
2815 rpc_restart_call(task);
2816 return;
2817 }
2818 }
2819 nfs4_locku_complete(calldata);
2820}
2821
2822static void nfs4_locku_begin(struct rpc_task *task)
2823{
2824 struct nfs4_unlockdata *calldata = (struct nfs4_unlockdata *)task->tk_calldata;
2713 struct rpc_message msg = { 2825 struct rpc_message msg = {
2714 .rpc_proc = &nfs4_procedures[NFSPROC4_CLNT_LOCKU], 2826 .rpc_proc = &nfs4_procedures[NFSPROC4_CLNT_LOCKU],
2715 .rpc_argp = &arg, 2827 .rpc_argp = &calldata->arg,
2716 .rpc_resp = &res, 2828 .rpc_resp = &calldata->res,
2717 .rpc_cred = state->owner->so_cred, 2829 .rpc_cred = calldata->lsp->ls_state->owner->so_cred,
2718 }; 2830 };
2831 int status;
2832
2833 status = nfs_wait_on_sequence(calldata->luargs.seqid, task);
2834 if (status != 0)
2835 return;
2836 if ((calldata->lsp->ls_flags & NFS_LOCK_INITIALIZED) == 0) {
2837 nfs4_locku_complete(calldata);
2838 task->tk_exit = NULL;
2839 rpc_exit(task, 0);
2840 return;
2841 }
2842 rpc_call_setup(task, &msg, 0);
2843}
2844
2845static int nfs4_proc_unlck(struct nfs4_state *state, int cmd, struct file_lock *request)
2846{
2847 struct nfs4_unlockdata *calldata;
2848 struct inode *inode = state->inode;
2849 struct nfs_server *server = NFS_SERVER(inode);
2719 struct nfs4_lock_state *lsp; 2850 struct nfs4_lock_state *lsp;
2720 struct nfs_locku_opargs luargs;
2721 int status; 2851 int status;
2722 2852
2723 down_read(&clp->cl_sem);
2724 down(&state->lock_sema);
2725 status = nfs4_set_lock_state(state, request); 2853 status = nfs4_set_lock_state(state, request);
2726 if (status != 0) 2854 if (status != 0)
2727 goto out; 2855 return status;
2728 lsp = request->fl_u.nfs4_fl.owner; 2856 lsp = request->fl_u.nfs4_fl.owner;
2729 /* We might have lost the locks! */ 2857 /* We might have lost the locks! */
2730 if ((lsp->ls_flags & NFS_LOCK_INITIALIZED) == 0) 2858 if ((lsp->ls_flags & NFS_LOCK_INITIALIZED) == 0)
2731 goto out; 2859 return 0;
2732 luargs.seqid = lsp->ls_seqid; 2860 calldata = kmalloc(sizeof(*calldata), GFP_KERNEL);
2733 memcpy(&luargs.stateid, &lsp->ls_stateid, sizeof(luargs.stateid)); 2861 if (calldata == NULL)
2734 arg.u.locku = &luargs; 2862 return -ENOMEM;
2735 status = rpc_call_sync(server->client, &msg, RPC_TASK_NOINTR); 2863 calldata->luargs.seqid = nfs_alloc_seqid(&lsp->ls_seqid);
2736 nfs4_increment_lock_seqid(status, lsp); 2864 if (calldata->luargs.seqid == NULL) {
2737 2865 kfree(calldata);
2738 if (status == 0) 2866 return -ENOMEM;
2739 memcpy(&lsp->ls_stateid, &res.u.stateid, 2867 }
2740 sizeof(lsp->ls_stateid)); 2868 calldata->luargs.stateid = &lsp->ls_stateid;
2741out: 2869 calldata->arg.fh = NFS_FH(inode);
2742 up(&state->lock_sema); 2870 calldata->arg.type = nfs4_lck_type(cmd, request);
2871 calldata->arg.offset = request->fl_start;
2872 calldata->arg.length = nfs4_lck_length(request);
2873 calldata->arg.u.locku = &calldata->luargs;
2874 calldata->res.server = server;
2875 calldata->lsp = lsp;
2876 atomic_inc(&lsp->ls_count);
2877
2878 /* Ensure we don't close file until we're done freeing locks! */
2879 calldata->ctx = get_nfs_open_context((struct nfs_open_context*)request->fl_file->private_data);
2880
2881 atomic_set(&calldata->refcount, 2);
2882 init_completion(&calldata->completion);
2883
2884 status = nfs4_call_async(NFS_SERVER(inode)->client, nfs4_locku_begin,
2885 nfs4_locku_done, calldata);
2743 if (status == 0) 2886 if (status == 0)
2744 do_vfs_lock(request->fl_file, request); 2887 wait_for_completion_interruptible(&calldata->completion);
2745 up_read(&clp->cl_sem); 2888 do_vfs_lock(request->fl_file, request);
2889 nfs4_locku_release_calldata(calldata);
2746 return status; 2890 return status;
2747} 2891}
2748 2892
2749static int nfs4_proc_unlck(struct nfs4_state *state, int cmd, struct file_lock *request)
2750{
2751 struct nfs4_exception exception = { };
2752 int err;
2753
2754 do {
2755 err = nfs4_handle_exception(NFS_SERVER(state->inode),
2756 _nfs4_proc_unlck(state, cmd, request),
2757 &exception);
2758 } while (exception.retry);
2759 return err;
2760}
2761
2762static int _nfs4_do_setlk(struct nfs4_state *state, int cmd, struct file_lock *request, int reclaim) 2893static int _nfs4_do_setlk(struct nfs4_state *state, int cmd, struct file_lock *request, int reclaim)
2763{ 2894{
2764 struct inode *inode = state->inode; 2895 struct inode *inode = state->inode;
2765 struct nfs_server *server = NFS_SERVER(inode); 2896 struct nfs_server *server = NFS_SERVER(inode);
2766 struct nfs4_lock_state *lsp = request->fl_u.nfs4_fl.owner; 2897 struct nfs4_lock_state *lsp = request->fl_u.nfs4_fl.owner;
2898 struct nfs_lock_opargs largs = {
2899 .lock_stateid = &lsp->ls_stateid,
2900 .open_stateid = &state->stateid,
2901 .lock_owner = {
2902 .clientid = server->nfs4_state->cl_clientid,
2903 .id = lsp->ls_id,
2904 },
2905 .reclaim = reclaim,
2906 };
2767 struct nfs_lockargs arg = { 2907 struct nfs_lockargs arg = {
2768 .fh = NFS_FH(inode), 2908 .fh = NFS_FH(inode),
2769 .type = nfs4_lck_type(cmd, request), 2909 .type = nfs4_lck_type(cmd, request),
2770 .offset = request->fl_start, 2910 .offset = request->fl_start,
2771 .length = nfs4_lck_length(request), 2911 .length = nfs4_lck_length(request),
2912 .u = {
2913 .lock = &largs,
2914 },
2772 }; 2915 };
2773 struct nfs_lockres res = { 2916 struct nfs_lockres res = {
2774 .server = server, 2917 .server = server,
@@ -2779,53 +2922,39 @@ static int _nfs4_do_setlk(struct nfs4_state *state, int cmd, struct file_lock *r
2779 .rpc_resp = &res, 2922 .rpc_resp = &res,
2780 .rpc_cred = state->owner->so_cred, 2923 .rpc_cred = state->owner->so_cred,
2781 }; 2924 };
2782 struct nfs_lock_opargs largs = { 2925 int status = -ENOMEM;
2783 .reclaim = reclaim,
2784 .new_lock_owner = 0,
2785 };
2786 int status;
2787 2926
2788 if (!(lsp->ls_flags & NFS_LOCK_INITIALIZED)) { 2927 largs.lock_seqid = nfs_alloc_seqid(&lsp->ls_seqid);
2928 if (largs.lock_seqid == NULL)
2929 return -ENOMEM;
2930 if (!(lsp->ls_seqid.flags & NFS_SEQID_CONFIRMED)) {
2789 struct nfs4_state_owner *owner = state->owner; 2931 struct nfs4_state_owner *owner = state->owner;
2790 struct nfs_open_to_lock otl = { 2932
2791 .lock_owner = { 2933 largs.open_seqid = nfs_alloc_seqid(&owner->so_seqid);
2792 .clientid = server->nfs4_state->cl_clientid, 2934 if (largs.open_seqid == NULL)
2793 }, 2935 goto out;
2794 };
2795
2796 otl.lock_seqid = lsp->ls_seqid;
2797 otl.lock_owner.id = lsp->ls_id;
2798 memcpy(&otl.open_stateid, &state->stateid, sizeof(otl.open_stateid));
2799 largs.u.open_lock = &otl;
2800 largs.new_lock_owner = 1; 2936 largs.new_lock_owner = 1;
2801 arg.u.lock = &largs;
2802 down(&owner->so_sema);
2803 otl.open_seqid = owner->so_seqid;
2804 status = rpc_call_sync(server->client, &msg, RPC_TASK_NOINTR); 2937 status = rpc_call_sync(server->client, &msg, RPC_TASK_NOINTR);
2805 /* increment open_owner seqid on success, and 2938 /* increment open seqid on success, and seqid mutating errors */
2806 * seqid mutating errors */ 2939 if (largs.new_lock_owner != 0) {
2807 nfs4_increment_seqid(status, owner); 2940 nfs_increment_open_seqid(status, largs.open_seqid);
2808 up(&owner->so_sema); 2941 if (status == 0)
2809 if (status == 0) { 2942 nfs_confirm_seqid(&lsp->ls_seqid, 0);
2810 lsp->ls_flags |= NFS_LOCK_INITIALIZED;
2811 lsp->ls_seqid++;
2812 } 2943 }
2813 } else { 2944 nfs_free_seqid(largs.open_seqid);
2814 struct nfs_exist_lock el = { 2945 } else
2815 .seqid = lsp->ls_seqid,
2816 };
2817 memcpy(&el.stateid, &lsp->ls_stateid, sizeof(el.stateid));
2818 largs.u.exist_lock = &el;
2819 arg.u.lock = &largs;
2820 status = rpc_call_sync(server->client, &msg, RPC_TASK_NOINTR); 2946 status = rpc_call_sync(server->client, &msg, RPC_TASK_NOINTR);
2821 /* increment seqid on success, and * seqid mutating errors*/ 2947 /* increment lock seqid on success, and seqid mutating errors*/
2822 nfs4_increment_lock_seqid(status, lsp); 2948 nfs_increment_lock_seqid(status, largs.lock_seqid);
2823 }
2824 /* save the returned stateid. */ 2949 /* save the returned stateid. */
2825 if (status == 0) 2950 if (status == 0) {
2826 memcpy(&lsp->ls_stateid, &res.u.stateid, sizeof(nfs4_stateid)); 2951 memcpy(lsp->ls_stateid.data, res.u.stateid.data,
2827 else if (status == -NFS4ERR_DENIED) 2952 sizeof(lsp->ls_stateid.data));
2953 lsp->ls_flags |= NFS_LOCK_INITIALIZED;
2954 } else if (status == -NFS4ERR_DENIED)
2828 status = -EAGAIN; 2955 status = -EAGAIN;
2956out:
2957 nfs_free_seqid(largs.lock_seqid);
2829 return status; 2958 return status;
2830} 2959}
2831 2960
@@ -2865,11 +2994,9 @@ static int _nfs4_proc_setlk(struct nfs4_state *state, int cmd, struct file_lock
2865 int status; 2994 int status;
2866 2995
2867 down_read(&clp->cl_sem); 2996 down_read(&clp->cl_sem);
2868 down(&state->lock_sema);
2869 status = nfs4_set_lock_state(state, request); 2997 status = nfs4_set_lock_state(state, request);
2870 if (status == 0) 2998 if (status == 0)
2871 status = _nfs4_do_setlk(state, cmd, request, 0); 2999 status = _nfs4_do_setlk(state, cmd, request, 0);
2872 up(&state->lock_sema);
2873 if (status == 0) { 3000 if (status == 0) {
2874 /* Note: we always want to sleep here! */ 3001 /* Note: we always want to sleep here! */
2875 request->fl_flags |= FL_SLEEP; 3002 request->fl_flags |= FL_SLEEP;
@@ -3024,8 +3151,8 @@ struct nfs_rpc_ops nfs_v4_clientops = {
3024 .read_setup = nfs4_proc_read_setup, 3151 .read_setup = nfs4_proc_read_setup,
3025 .write_setup = nfs4_proc_write_setup, 3152 .write_setup = nfs4_proc_write_setup,
3026 .commit_setup = nfs4_proc_commit_setup, 3153 .commit_setup = nfs4_proc_commit_setup,
3027 .file_open = nfs4_proc_file_open, 3154 .file_open = nfs_open,
3028 .file_release = nfs4_proc_file_release, 3155 .file_release = nfs_release,
3029 .lock = nfs4_proc_lock, 3156 .lock = nfs4_proc_lock,
3030 .clear_acl_cache = nfs4_zap_acl_attr, 3157 .clear_acl_cache = nfs4_zap_acl_attr,
3031}; 3158};
diff --git a/fs/nfs/nfs4state.c b/fs/nfs/nfs4state.c
index afe587d82f1e..2d5a6a2b9dec 100644
--- a/fs/nfs/nfs4state.c
+++ b/fs/nfs/nfs4state.c
@@ -264,13 +264,16 @@ nfs4_alloc_state_owner(void)
264{ 264{
265 struct nfs4_state_owner *sp; 265 struct nfs4_state_owner *sp;
266 266
267 sp = kmalloc(sizeof(*sp),GFP_KERNEL); 267 sp = kzalloc(sizeof(*sp),GFP_KERNEL);
268 if (!sp) 268 if (!sp)
269 return NULL; 269 return NULL;
270 init_MUTEX(&sp->so_sema); 270 spin_lock_init(&sp->so_lock);
271 sp->so_seqid = 0; /* arbitrary */
272 INIT_LIST_HEAD(&sp->so_states); 271 INIT_LIST_HEAD(&sp->so_states);
273 INIT_LIST_HEAD(&sp->so_delegations); 272 INIT_LIST_HEAD(&sp->so_delegations);
273 rpc_init_wait_queue(&sp->so_sequence.wait, "Seqid_waitqueue");
274 sp->so_seqid.sequence = &sp->so_sequence;
275 spin_lock_init(&sp->so_sequence.lock);
276 INIT_LIST_HEAD(&sp->so_sequence.list);
274 atomic_set(&sp->so_count, 1); 277 atomic_set(&sp->so_count, 1);
275 return sp; 278 return sp;
276} 279}
@@ -359,7 +362,6 @@ nfs4_alloc_open_state(void)
359 memset(state->stateid.data, 0, sizeof(state->stateid.data)); 362 memset(state->stateid.data, 0, sizeof(state->stateid.data));
360 atomic_set(&state->count, 1); 363 atomic_set(&state->count, 1);
361 INIT_LIST_HEAD(&state->lock_states); 364 INIT_LIST_HEAD(&state->lock_states);
362 init_MUTEX(&state->lock_sema);
363 spin_lock_init(&state->state_lock); 365 spin_lock_init(&state->state_lock);
364 return state; 366 return state;
365} 367}
@@ -437,21 +439,23 @@ nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
437 if (state) 439 if (state)
438 goto out; 440 goto out;
439 new = nfs4_alloc_open_state(); 441 new = nfs4_alloc_open_state();
442 spin_lock(&owner->so_lock);
440 spin_lock(&inode->i_lock); 443 spin_lock(&inode->i_lock);
441 state = __nfs4_find_state_byowner(inode, owner); 444 state = __nfs4_find_state_byowner(inode, owner);
442 if (state == NULL && new != NULL) { 445 if (state == NULL && new != NULL) {
443 state = new; 446 state = new;
444 /* Caller *must* be holding owner->so_sem */
445 /* Note: The reclaim code dictates that we add stateless
446 * and read-only stateids to the end of the list */
447 list_add_tail(&state->open_states, &owner->so_states);
448 state->owner = owner; 447 state->owner = owner;
449 atomic_inc(&owner->so_count); 448 atomic_inc(&owner->so_count);
450 list_add(&state->inode_states, &nfsi->open_states); 449 list_add(&state->inode_states, &nfsi->open_states);
451 state->inode = igrab(inode); 450 state->inode = igrab(inode);
452 spin_unlock(&inode->i_lock); 451 spin_unlock(&inode->i_lock);
452 /* Note: The reclaim code dictates that we add stateless
453 * and read-only stateids to the end of the list */
454 list_add_tail(&state->open_states, &owner->so_states);
455 spin_unlock(&owner->so_lock);
453 } else { 456 } else {
454 spin_unlock(&inode->i_lock); 457 spin_unlock(&inode->i_lock);
458 spin_unlock(&owner->so_lock);
455 if (new) 459 if (new)
456 nfs4_free_open_state(new); 460 nfs4_free_open_state(new);
457 } 461 }
@@ -461,19 +465,21 @@ out:
461 465
462/* 466/*
463 * Beware! Caller must be holding exactly one 467 * Beware! Caller must be holding exactly one
464 * reference to clp->cl_sem and owner->so_sema! 468 * reference to clp->cl_sem!
465 */ 469 */
466void nfs4_put_open_state(struct nfs4_state *state) 470void nfs4_put_open_state(struct nfs4_state *state)
467{ 471{
468 struct inode *inode = state->inode; 472 struct inode *inode = state->inode;
469 struct nfs4_state_owner *owner = state->owner; 473 struct nfs4_state_owner *owner = state->owner;
470 474
471 if (!atomic_dec_and_lock(&state->count, &inode->i_lock)) 475 if (!atomic_dec_and_lock(&state->count, &owner->so_lock))
472 return; 476 return;
477 spin_lock(&inode->i_lock);
473 if (!list_empty(&state->inode_states)) 478 if (!list_empty(&state->inode_states))
474 list_del(&state->inode_states); 479 list_del(&state->inode_states);
475 spin_unlock(&inode->i_lock);
476 list_del(&state->open_states); 480 list_del(&state->open_states);
481 spin_unlock(&inode->i_lock);
482 spin_unlock(&owner->so_lock);
477 iput(inode); 483 iput(inode);
478 BUG_ON (state->state != 0); 484 BUG_ON (state->state != 0);
479 nfs4_free_open_state(state); 485 nfs4_free_open_state(state);
@@ -481,20 +487,17 @@ void nfs4_put_open_state(struct nfs4_state *state)
481} 487}
482 488
483/* 489/*
484 * Beware! Caller must be holding no references to clp->cl_sem! 490 * Close the current file.
485 * of owner->so_sema!
486 */ 491 */
487void nfs4_close_state(struct nfs4_state *state, mode_t mode) 492void nfs4_close_state(struct nfs4_state *state, mode_t mode)
488{ 493{
489 struct inode *inode = state->inode; 494 struct inode *inode = state->inode;
490 struct nfs4_state_owner *owner = state->owner; 495 struct nfs4_state_owner *owner = state->owner;
491 struct nfs4_client *clp = owner->so_client;
492 int newstate; 496 int newstate;
493 497
494 atomic_inc(&owner->so_count); 498 atomic_inc(&owner->so_count);
495 down_read(&clp->cl_sem);
496 down(&owner->so_sema);
497 /* Protect against nfs4_find_state() */ 499 /* Protect against nfs4_find_state() */
500 spin_lock(&owner->so_lock);
498 spin_lock(&inode->i_lock); 501 spin_lock(&inode->i_lock);
499 if (mode & FMODE_READ) 502 if (mode & FMODE_READ)
500 state->nreaders--; 503 state->nreaders--;
@@ -507,6 +510,7 @@ void nfs4_close_state(struct nfs4_state *state, mode_t mode)
507 list_move_tail(&state->open_states, &owner->so_states); 510 list_move_tail(&state->open_states, &owner->so_states);
508 } 511 }
509 spin_unlock(&inode->i_lock); 512 spin_unlock(&inode->i_lock);
513 spin_unlock(&owner->so_lock);
510 newstate = 0; 514 newstate = 0;
511 if (state->state != 0) { 515 if (state->state != 0) {
512 if (state->nreaders) 516 if (state->nreaders)
@@ -515,14 +519,16 @@ void nfs4_close_state(struct nfs4_state *state, mode_t mode)
515 newstate |= FMODE_WRITE; 519 newstate |= FMODE_WRITE;
516 if (state->state == newstate) 520 if (state->state == newstate)
517 goto out; 521 goto out;
518 if (nfs4_do_close(inode, state, newstate) == -EINPROGRESS) 522 if (test_bit(NFS_DELEGATED_STATE, &state->flags)) {
523 state->state = newstate;
524 goto out;
525 }
526 if (nfs4_do_close(inode, state, newstate) == 0)
519 return; 527 return;
520 } 528 }
521out: 529out:
522 nfs4_put_open_state(state); 530 nfs4_put_open_state(state);
523 up(&owner->so_sema);
524 nfs4_put_state_owner(owner); 531 nfs4_put_state_owner(owner);
525 up_read(&clp->cl_sem);
526} 532}
527 533
528/* 534/*
@@ -546,19 +552,16 @@ __nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
546 * Return a compatible lock_state. If no initialized lock_state structure 552 * Return a compatible lock_state. If no initialized lock_state structure
547 * exists, return an uninitialized one. 553 * exists, return an uninitialized one.
548 * 554 *
549 * The caller must be holding state->lock_sema
550 */ 555 */
551static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, fl_owner_t fl_owner) 556static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
552{ 557{
553 struct nfs4_lock_state *lsp; 558 struct nfs4_lock_state *lsp;
554 struct nfs4_client *clp = state->owner->so_client; 559 struct nfs4_client *clp = state->owner->so_client;
555 560
556 lsp = kmalloc(sizeof(*lsp), GFP_KERNEL); 561 lsp = kzalloc(sizeof(*lsp), GFP_KERNEL);
557 if (lsp == NULL) 562 if (lsp == NULL)
558 return NULL; 563 return NULL;
559 lsp->ls_flags = 0; 564 lsp->ls_seqid.sequence = &state->owner->so_sequence;
560 lsp->ls_seqid = 0; /* arbitrary */
561 memset(lsp->ls_stateid.data, 0, sizeof(lsp->ls_stateid.data));
562 atomic_set(&lsp->ls_count, 1); 565 atomic_set(&lsp->ls_count, 1);
563 lsp->ls_owner = fl_owner; 566 lsp->ls_owner = fl_owner;
564 spin_lock(&clp->cl_lock); 567 spin_lock(&clp->cl_lock);
@@ -572,7 +575,7 @@ static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, f
572 * Return a compatible lock_state. If no initialized lock_state structure 575 * Return a compatible lock_state. If no initialized lock_state structure
573 * exists, return an uninitialized one. 576 * exists, return an uninitialized one.
574 * 577 *
575 * The caller must be holding state->lock_sema and clp->cl_sem 578 * The caller must be holding clp->cl_sem
576 */ 579 */
577static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner) 580static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner)
578{ 581{
@@ -605,7 +608,7 @@ static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_
605 * Release reference to lock_state, and free it if we see that 608 * Release reference to lock_state, and free it if we see that
606 * it is no longer in use 609 * it is no longer in use
607 */ 610 */
608static void nfs4_put_lock_state(struct nfs4_lock_state *lsp) 611void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
609{ 612{
610 struct nfs4_state *state; 613 struct nfs4_state *state;
611 614
@@ -673,29 +676,94 @@ void nfs4_copy_stateid(nfs4_stateid *dst, struct nfs4_state *state, fl_owner_t f
673 nfs4_put_lock_state(lsp); 676 nfs4_put_lock_state(lsp);
674} 677}
675 678
676/* 679struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter)
677* Called with state->lock_sema and clp->cl_sem held.
678*/
679void nfs4_increment_lock_seqid(int status, struct nfs4_lock_state *lsp)
680{ 680{
681 if (status == NFS_OK || seqid_mutating_err(-status)) 681 struct nfs_seqid *new;
682 lsp->ls_seqid++; 682
683 new = kmalloc(sizeof(*new), GFP_KERNEL);
684 if (new != NULL) {
685 new->sequence = counter;
686 INIT_LIST_HEAD(&new->list);
687 }
688 return new;
689}
690
691void nfs_free_seqid(struct nfs_seqid *seqid)
692{
693 struct rpc_sequence *sequence = seqid->sequence->sequence;
694
695 if (!list_empty(&seqid->list)) {
696 spin_lock(&sequence->lock);
697 list_del(&seqid->list);
698 spin_unlock(&sequence->lock);
699 }
700 rpc_wake_up_next(&sequence->wait);
701 kfree(seqid);
683} 702}
684 703
685/* 704/*
686* Called with sp->so_sema and clp->cl_sem held. 705 * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
687* 706 * failed with a seqid incrementing error -
688* Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or 707 * see comments nfs_fs.h:seqid_mutating_error()
689* failed with a seqid incrementing error - 708 */
690* see comments nfs_fs.h:seqid_mutating_error() 709static inline void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
691*/ 710{
692void nfs4_increment_seqid(int status, struct nfs4_state_owner *sp) 711 switch (status) {
693{ 712 case 0:
694 if (status == NFS_OK || seqid_mutating_err(-status)) 713 break;
695 sp->so_seqid++; 714 case -NFS4ERR_BAD_SEQID:
696 /* If the server returns BAD_SEQID, unhash state_owner here */ 715 case -NFS4ERR_STALE_CLIENTID:
697 if (status == -NFS4ERR_BAD_SEQID) 716 case -NFS4ERR_STALE_STATEID:
717 case -NFS4ERR_BAD_STATEID:
718 case -NFS4ERR_BADXDR:
719 case -NFS4ERR_RESOURCE:
720 case -NFS4ERR_NOFILEHANDLE:
721 /* Non-seqid mutating errors */
722 return;
723 };
724 /*
725 * Note: no locking needed as we are guaranteed to be first
726 * on the sequence list
727 */
728 seqid->sequence->counter++;
729}
730
731void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
732{
733 if (status == -NFS4ERR_BAD_SEQID) {
734 struct nfs4_state_owner *sp = container_of(seqid->sequence,
735 struct nfs4_state_owner, so_seqid);
698 nfs4_drop_state_owner(sp); 736 nfs4_drop_state_owner(sp);
737 }
738 return nfs_increment_seqid(status, seqid);
739}
740
741/*
742 * Increment the seqid if the LOCK/LOCKU succeeded, or
743 * failed with a seqid incrementing error -
744 * see comments nfs_fs.h:seqid_mutating_error()
745 */
746void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
747{
748 return nfs_increment_seqid(status, seqid);
749}
750
751int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
752{
753 struct rpc_sequence *sequence = seqid->sequence->sequence;
754 int status = 0;
755
756 if (sequence->list.next == &seqid->list)
757 goto out;
758 spin_lock(&sequence->lock);
759 if (!list_empty(&sequence->list)) {
760 rpc_sleep_on(&sequence->wait, task, NULL, NULL);
761 status = -EAGAIN;
762 } else
763 list_add(&seqid->list, &sequence->list);
764 spin_unlock(&sequence->lock);
765out:
766 return status;
699} 767}
700 768
701static int reclaimer(void *); 769static int reclaimer(void *);
@@ -791,8 +859,6 @@ static int nfs4_reclaim_open_state(struct nfs4_state_recovery_ops *ops, struct n
791 if (state->state == 0) 859 if (state->state == 0)
792 continue; 860 continue;
793 status = ops->recover_open(sp, state); 861 status = ops->recover_open(sp, state);
794 list_for_each_entry(lock, &state->lock_states, ls_locks)
795 lock->ls_flags &= ~NFS_LOCK_INITIALIZED;
796 if (status >= 0) { 862 if (status >= 0) {
797 status = nfs4_reclaim_locks(ops, state); 863 status = nfs4_reclaim_locks(ops, state);
798 if (status < 0) 864 if (status < 0)
@@ -831,6 +897,28 @@ out_err:
831 return status; 897 return status;
832} 898}
833 899
900static void nfs4_state_mark_reclaim(struct nfs4_client *clp)
901{
902 struct nfs4_state_owner *sp;
903 struct nfs4_state *state;
904 struct nfs4_lock_state *lock;
905
906 /* Reset all sequence ids to zero */
907 list_for_each_entry(sp, &clp->cl_state_owners, so_list) {
908 sp->so_seqid.counter = 0;
909 sp->so_seqid.flags = 0;
910 spin_lock(&sp->so_lock);
911 list_for_each_entry(state, &sp->so_states, open_states) {
912 list_for_each_entry(lock, &state->lock_states, ls_locks) {
913 lock->ls_seqid.counter = 0;
914 lock->ls_seqid.flags = 0;
915 lock->ls_flags &= ~NFS_LOCK_INITIALIZED;
916 }
917 }
918 spin_unlock(&sp->so_lock);
919 }
920}
921
834static int reclaimer(void *ptr) 922static int reclaimer(void *ptr)
835{ 923{
836 struct reclaimer_args *args = (struct reclaimer_args *)ptr; 924 struct reclaimer_args *args = (struct reclaimer_args *)ptr;
@@ -864,6 +952,7 @@ restart_loop:
864 default: 952 default:
865 ops = &nfs4_network_partition_recovery_ops; 953 ops = &nfs4_network_partition_recovery_ops;
866 }; 954 };
955 nfs4_state_mark_reclaim(clp);
867 status = __nfs4_init_client(clp); 956 status = __nfs4_init_client(clp);
868 if (status) 957 if (status)
869 goto out_error; 958 goto out_error;
diff --git a/fs/nfs/nfs4xdr.c b/fs/nfs/nfs4xdr.c
index 6c564ef9489e..cd762648fa9a 100644
--- a/fs/nfs/nfs4xdr.c
+++ b/fs/nfs/nfs4xdr.c
@@ -602,10 +602,10 @@ static int encode_close(struct xdr_stream *xdr, const struct nfs_closeargs *arg)
602{ 602{
603 uint32_t *p; 603 uint32_t *p;
604 604
605 RESERVE_SPACE(8+sizeof(arg->stateid.data)); 605 RESERVE_SPACE(8+sizeof(arg->stateid->data));
606 WRITE32(OP_CLOSE); 606 WRITE32(OP_CLOSE);
607 WRITE32(arg->seqid); 607 WRITE32(arg->seqid->sequence->counter);
608 WRITEMEM(arg->stateid.data, sizeof(arg->stateid.data)); 608 WRITEMEM(arg->stateid->data, sizeof(arg->stateid->data));
609 609
610 return 0; 610 return 0;
611} 611}
@@ -729,22 +729,18 @@ static int encode_lock(struct xdr_stream *xdr, const struct nfs_lockargs *arg)
729 WRITE64(arg->length); 729 WRITE64(arg->length);
730 WRITE32(opargs->new_lock_owner); 730 WRITE32(opargs->new_lock_owner);
731 if (opargs->new_lock_owner){ 731 if (opargs->new_lock_owner){
732 struct nfs_open_to_lock *ol = opargs->u.open_lock;
733
734 RESERVE_SPACE(40); 732 RESERVE_SPACE(40);
735 WRITE32(ol->open_seqid); 733 WRITE32(opargs->open_seqid->sequence->counter);
736 WRITEMEM(&ol->open_stateid, sizeof(ol->open_stateid)); 734 WRITEMEM(opargs->open_stateid->data, sizeof(opargs->open_stateid->data));
737 WRITE32(ol->lock_seqid); 735 WRITE32(opargs->lock_seqid->sequence->counter);
738 WRITE64(ol->lock_owner.clientid); 736 WRITE64(opargs->lock_owner.clientid);
739 WRITE32(4); 737 WRITE32(4);
740 WRITE32(ol->lock_owner.id); 738 WRITE32(opargs->lock_owner.id);
741 } 739 }
742 else { 740 else {
743 struct nfs_exist_lock *el = opargs->u.exist_lock;
744
745 RESERVE_SPACE(20); 741 RESERVE_SPACE(20);
746 WRITEMEM(&el->stateid, sizeof(el->stateid)); 742 WRITEMEM(opargs->lock_stateid->data, sizeof(opargs->lock_stateid->data));
747 WRITE32(el->seqid); 743 WRITE32(opargs->lock_seqid->sequence->counter);
748 } 744 }
749 745
750 return 0; 746 return 0;
@@ -775,8 +771,8 @@ static int encode_locku(struct xdr_stream *xdr, const struct nfs_lockargs *arg)
775 RESERVE_SPACE(44); 771 RESERVE_SPACE(44);
776 WRITE32(OP_LOCKU); 772 WRITE32(OP_LOCKU);
777 WRITE32(arg->type); 773 WRITE32(arg->type);
778 WRITE32(opargs->seqid); 774 WRITE32(opargs->seqid->sequence->counter);
779 WRITEMEM(&opargs->stateid, sizeof(opargs->stateid)); 775 WRITEMEM(opargs->stateid->data, sizeof(opargs->stateid->data));
780 WRITE64(arg->offset); 776 WRITE64(arg->offset);
781 WRITE64(arg->length); 777 WRITE64(arg->length);
782 778
@@ -826,7 +822,7 @@ static inline void encode_openhdr(struct xdr_stream *xdr, const struct nfs_opena
826 */ 822 */
827 RESERVE_SPACE(8); 823 RESERVE_SPACE(8);
828 WRITE32(OP_OPEN); 824 WRITE32(OP_OPEN);
829 WRITE32(arg->seqid); 825 WRITE32(arg->seqid->sequence->counter);
830 encode_share_access(xdr, arg->open_flags); 826 encode_share_access(xdr, arg->open_flags);
831 RESERVE_SPACE(16); 827 RESERVE_SPACE(16);
832 WRITE64(arg->clientid); 828 WRITE64(arg->clientid);
@@ -941,7 +937,7 @@ static int encode_open_confirm(struct xdr_stream *xdr, const struct nfs_open_con
941 RESERVE_SPACE(8+sizeof(arg->stateid.data)); 937 RESERVE_SPACE(8+sizeof(arg->stateid.data));
942 WRITE32(OP_OPEN_CONFIRM); 938 WRITE32(OP_OPEN_CONFIRM);
943 WRITEMEM(arg->stateid.data, sizeof(arg->stateid.data)); 939 WRITEMEM(arg->stateid.data, sizeof(arg->stateid.data));
944 WRITE32(arg->seqid); 940 WRITE32(arg->seqid->sequence->counter);
945 941
946 return 0; 942 return 0;
947} 943}
@@ -950,10 +946,10 @@ static int encode_open_downgrade(struct xdr_stream *xdr, const struct nfs_closea
950{ 946{
951 uint32_t *p; 947 uint32_t *p;
952 948
953 RESERVE_SPACE(8+sizeof(arg->stateid.data)); 949 RESERVE_SPACE(8+sizeof(arg->stateid->data));
954 WRITE32(OP_OPEN_DOWNGRADE); 950 WRITE32(OP_OPEN_DOWNGRADE);
955 WRITEMEM(arg->stateid.data, sizeof(arg->stateid.data)); 951 WRITEMEM(arg->stateid->data, sizeof(arg->stateid->data));
956 WRITE32(arg->seqid); 952 WRITE32(arg->seqid->sequence->counter);
957 encode_share_access(xdr, arg->open_flags); 953 encode_share_access(xdr, arg->open_flags);
958 return 0; 954 return 0;
959} 955}
@@ -1437,6 +1433,9 @@ static int nfs4_xdr_enc_open(struct rpc_rqst *req, uint32_t *p, struct nfs_opena
1437 }; 1433 };
1438 int status; 1434 int status;
1439 1435
1436 status = nfs_wait_on_sequence(args->seqid, req->rq_task);
1437 if (status != 0)
1438 goto out;
1440 xdr_init_encode(&xdr, &req->rq_snd_buf, p); 1439 xdr_init_encode(&xdr, &req->rq_snd_buf, p);
1441 encode_compound_hdr(&xdr, &hdr); 1440 encode_compound_hdr(&xdr, &hdr);
1442 status = encode_putfh(&xdr, args->fh); 1441 status = encode_putfh(&xdr, args->fh);
@@ -1464,6 +1463,9 @@ static int nfs4_xdr_enc_open_confirm(struct rpc_rqst *req, uint32_t *p, struct n
1464 }; 1463 };
1465 int status; 1464 int status;
1466 1465
1466 status = nfs_wait_on_sequence(args->seqid, req->rq_task);
1467 if (status != 0)
1468 goto out;
1467 xdr_init_encode(&xdr, &req->rq_snd_buf, p); 1469 xdr_init_encode(&xdr, &req->rq_snd_buf, p);
1468 encode_compound_hdr(&xdr, &hdr); 1470 encode_compound_hdr(&xdr, &hdr);
1469 status = encode_putfh(&xdr, args->fh); 1471 status = encode_putfh(&xdr, args->fh);
@@ -1485,6 +1487,9 @@ static int nfs4_xdr_enc_open_noattr(struct rpc_rqst *req, uint32_t *p, struct nf
1485 }; 1487 };
1486 int status; 1488 int status;
1487 1489
1490 status = nfs_wait_on_sequence(args->seqid, req->rq_task);
1491 if (status != 0)
1492 goto out;
1488 xdr_init_encode(&xdr, &req->rq_snd_buf, p); 1493 xdr_init_encode(&xdr, &req->rq_snd_buf, p);
1489 encode_compound_hdr(&xdr, &hdr); 1494 encode_compound_hdr(&xdr, &hdr);
1490 status = encode_putfh(&xdr, args->fh); 1495 status = encode_putfh(&xdr, args->fh);
@@ -1525,8 +1530,15 @@ static int nfs4_xdr_enc_lock(struct rpc_rqst *req, uint32_t *p, struct nfs_locka
1525 struct compound_hdr hdr = { 1530 struct compound_hdr hdr = {
1526 .nops = 2, 1531 .nops = 2,
1527 }; 1532 };
1533 struct nfs_lock_opargs *opargs = args->u.lock;
1528 int status; 1534 int status;
1529 1535
1536 status = nfs_wait_on_sequence(opargs->lock_seqid, req->rq_task);
1537 if (status != 0)
1538 goto out;
1539 /* Do we need to do an open_to_lock_owner? */
1540 if (opargs->lock_seqid->sequence->flags & NFS_SEQID_CONFIRMED)
1541 opargs->new_lock_owner = 0;
1530 xdr_init_encode(&xdr, &req->rq_snd_buf, p); 1542 xdr_init_encode(&xdr, &req->rq_snd_buf, p);
1531 encode_compound_hdr(&xdr, &hdr); 1543 encode_compound_hdr(&xdr, &hdr);
1532 status = encode_putfh(&xdr, args->fh); 1544 status = encode_putfh(&xdr, args->fh);
@@ -2890,8 +2902,8 @@ static int decode_lock(struct xdr_stream *xdr, struct nfs_lockres *res)
2890 2902
2891 status = decode_op_hdr(xdr, OP_LOCK); 2903 status = decode_op_hdr(xdr, OP_LOCK);
2892 if (status == 0) { 2904 if (status == 0) {
2893 READ_BUF(sizeof(nfs4_stateid)); 2905 READ_BUF(sizeof(res->u.stateid.data));
2894 COPYMEM(&res->u.stateid, sizeof(res->u.stateid)); 2906 COPYMEM(res->u.stateid.data, sizeof(res->u.stateid.data));
2895 } else if (status == -NFS4ERR_DENIED) 2907 } else if (status == -NFS4ERR_DENIED)
2896 return decode_lock_denied(xdr, &res->u.denied); 2908 return decode_lock_denied(xdr, &res->u.denied);
2897 return status; 2909 return status;
@@ -2913,8 +2925,8 @@ static int decode_locku(struct xdr_stream *xdr, struct nfs_lockres *res)
2913 2925
2914 status = decode_op_hdr(xdr, OP_LOCKU); 2926 status = decode_op_hdr(xdr, OP_LOCKU);
2915 if (status == 0) { 2927 if (status == 0) {
2916 READ_BUF(sizeof(nfs4_stateid)); 2928 READ_BUF(sizeof(res->u.stateid.data));
2917 COPYMEM(&res->u.stateid, sizeof(res->u.stateid)); 2929 COPYMEM(res->u.stateid.data, sizeof(res->u.stateid.data));
2918 } 2930 }
2919 return status; 2931 return status;
2920} 2932}
@@ -3243,7 +3255,8 @@ static int decode_getacl(struct xdr_stream *xdr, struct rpc_rqst *req,
3243 if (attrlen <= *acl_len) 3255 if (attrlen <= *acl_len)
3244 xdr_read_pages(xdr, attrlen); 3256 xdr_read_pages(xdr, attrlen);
3245 *acl_len = attrlen; 3257 *acl_len = attrlen;
3246 } 3258 } else
3259 status = -EOPNOTSUPP;
3247 3260
3248out: 3261out:
3249 return status; 3262 return status;
diff --git a/fs/nfs/proc.c b/fs/nfs/proc.c
index be23c3fb9260..8fef86523d7f 100644
--- a/fs/nfs/proc.c
+++ b/fs/nfs/proc.c
@@ -216,7 +216,7 @@ static int nfs_proc_write(struct nfs_write_data *wdata)
216 216
217static int 217static int
218nfs_proc_create(struct inode *dir, struct dentry *dentry, struct iattr *sattr, 218nfs_proc_create(struct inode *dir, struct dentry *dentry, struct iattr *sattr,
219 int flags) 219 int flags, struct nameidata *nd)
220{ 220{
221 struct nfs_fh fhandle; 221 struct nfs_fh fhandle;
222 struct nfs_fattr fattr; 222 struct nfs_fattr fattr;
diff --git a/fs/open.c b/fs/open.c
index f0d90cf0495c..8d06ec911fd9 100644
--- a/fs/open.c
+++ b/fs/open.c
@@ -739,7 +739,8 @@ asmlinkage long sys_fchown(unsigned int fd, uid_t user, gid_t group)
739} 739}
740 740
741static struct file *__dentry_open(struct dentry *dentry, struct vfsmount *mnt, 741static struct file *__dentry_open(struct dentry *dentry, struct vfsmount *mnt,
742 int flags, struct file *f) 742 int flags, struct file *f,
743 int (*open)(struct inode *, struct file *))
743{ 744{
744 struct inode *inode; 745 struct inode *inode;
745 int error; 746 int error;
@@ -761,11 +762,14 @@ static struct file *__dentry_open(struct dentry *dentry, struct vfsmount *mnt,
761 f->f_op = fops_get(inode->i_fop); 762 f->f_op = fops_get(inode->i_fop);
762 file_move(f, &inode->i_sb->s_files); 763 file_move(f, &inode->i_sb->s_files);
763 764
764 if (f->f_op && f->f_op->open) { 765 if (!open && f->f_op)
765 error = f->f_op->open(inode,f); 766 open = f->f_op->open;
767 if (open) {
768 error = open(inode, f);
766 if (error) 769 if (error)
767 goto cleanup_all; 770 goto cleanup_all;
768 } 771 }
772
769 f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC); 773 f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
770 774
771 file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping); 775 file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
@@ -814,28 +818,75 @@ struct file *filp_open(const char * filename, int flags, int mode)
814{ 818{
815 int namei_flags, error; 819 int namei_flags, error;
816 struct nameidata nd; 820 struct nameidata nd;
817 struct file *f;
818 821
819 namei_flags = flags; 822 namei_flags = flags;
820 if ((namei_flags+1) & O_ACCMODE) 823 if ((namei_flags+1) & O_ACCMODE)
821 namei_flags++; 824 namei_flags++;
822 if (namei_flags & O_TRUNC)
823 namei_flags |= 2;
824
825 error = -ENFILE;
826 f = get_empty_filp();
827 if (f == NULL)
828 return ERR_PTR(error);
829 825
830 error = open_namei(filename, namei_flags, mode, &nd); 826 error = open_namei(filename, namei_flags, mode, &nd);
831 if (!error) 827 if (!error)
832 return __dentry_open(nd.dentry, nd.mnt, flags, f); 828 return nameidata_to_filp(&nd, flags);
833 829
834 put_filp(f);
835 return ERR_PTR(error); 830 return ERR_PTR(error);
836} 831}
837EXPORT_SYMBOL(filp_open); 832EXPORT_SYMBOL(filp_open);
838 833
834/**
835 * lookup_instantiate_filp - instantiates the open intent filp
836 * @nd: pointer to nameidata
837 * @dentry: pointer to dentry
838 * @open: open callback
839 *
840 * Helper for filesystems that want to use lookup open intents and pass back
841 * a fully instantiated struct file to the caller.
842 * This function is meant to be called from within a filesystem's
843 * lookup method.
844 * Note that in case of error, nd->intent.open.file is destroyed, but the
845 * path information remains valid.
846 * If the open callback is set to NULL, then the standard f_op->open()
847 * filesystem callback is substituted.
848 */
849struct file *lookup_instantiate_filp(struct nameidata *nd, struct dentry *dentry,
850 int (*open)(struct inode *, struct file *))
851{
852 if (IS_ERR(nd->intent.open.file))
853 goto out;
854 if (IS_ERR(dentry))
855 goto out_err;
856 nd->intent.open.file = __dentry_open(dget(dentry), mntget(nd->mnt),
857 nd->intent.open.flags - 1,
858 nd->intent.open.file,
859 open);
860out:
861 return nd->intent.open.file;
862out_err:
863 release_open_intent(nd);
864 nd->intent.open.file = (struct file *)dentry;
865 goto out;
866}
867EXPORT_SYMBOL_GPL(lookup_instantiate_filp);
868
869/**
870 * nameidata_to_filp - convert a nameidata to an open filp.
871 * @nd: pointer to nameidata
872 * @flags: open flags
873 *
874 * Note that this function destroys the original nameidata
875 */
876struct file *nameidata_to_filp(struct nameidata *nd, int flags)
877{
878 struct file *filp;
879
880 /* Pick up the filp from the open intent */
881 filp = nd->intent.open.file;
882 /* Has the filesystem initialised the file for us? */
883 if (filp->f_dentry == NULL)
884 filp = __dentry_open(nd->dentry, nd->mnt, flags, filp, NULL);
885 else
886 path_release(nd);
887 return filp;
888}
889
839struct file *dentry_open(struct dentry *dentry, struct vfsmount *mnt, int flags) 890struct file *dentry_open(struct dentry *dentry, struct vfsmount *mnt, int flags)
840{ 891{
841 int error; 892 int error;
@@ -846,7 +897,7 @@ struct file *dentry_open(struct dentry *dentry, struct vfsmount *mnt, int flags)
846 if (f == NULL) 897 if (f == NULL)
847 return ERR_PTR(error); 898 return ERR_PTR(error);
848 899
849 return __dentry_open(dentry, mnt, flags, f); 900 return __dentry_open(dentry, mnt, flags, f, NULL);
850} 901}
851EXPORT_SYMBOL(dentry_open); 902EXPORT_SYMBOL(dentry_open);
852 903
diff --git a/include/linux/namei.h b/include/linux/namei.h
index 7db67b008cac..1c975d0d9e94 100644
--- a/include/linux/namei.h
+++ b/include/linux/namei.h
@@ -8,6 +8,7 @@ struct vfsmount;
8struct open_intent { 8struct open_intent {
9 int flags; 9 int flags;
10 int create_mode; 10 int create_mode;
11 struct file *file;
11}; 12};
12 13
13enum { MAX_NESTED_LINKS = 5 }; 14enum { MAX_NESTED_LINKS = 5 };
@@ -65,6 +66,13 @@ extern int FASTCALL(link_path_walk(const char *, struct nameidata *));
65extern void path_release(struct nameidata *); 66extern void path_release(struct nameidata *);
66extern void path_release_on_umount(struct nameidata *); 67extern void path_release_on_umount(struct nameidata *);
67 68
69extern int __user_path_lookup_open(const char __user *, unsigned lookup_flags, struct nameidata *nd, int open_flags);
70extern int path_lookup_open(const char *, unsigned lookup_flags, struct nameidata *, int open_flags);
71extern struct file *lookup_instantiate_filp(struct nameidata *nd, struct dentry *dentry,
72 int (*open)(struct inode *, struct file *));
73extern struct file *nameidata_to_filp(struct nameidata *nd, int flags);
74extern void release_open_intent(struct nameidata *);
75
68extern struct dentry * lookup_one_len(const char *, struct dentry *, int); 76extern struct dentry * lookup_one_len(const char *, struct dentry *, int);
69extern struct dentry * lookup_hash(struct qstr *, struct dentry *); 77extern struct dentry * lookup_hash(struct qstr *, struct dentry *);
70 78
diff --git a/include/linux/nfs_fs.h b/include/linux/nfs_fs.h
index 9a6047ff1b25..7bac2785c6e4 100644
--- a/include/linux/nfs_fs.h
+++ b/include/linux/nfs_fs.h
@@ -41,6 +41,10 @@
41#define NFS_MAX_FILE_IO_BUFFER_SIZE 32768 41#define NFS_MAX_FILE_IO_BUFFER_SIZE 32768
42#define NFS_DEF_FILE_IO_BUFFER_SIZE 4096 42#define NFS_DEF_FILE_IO_BUFFER_SIZE 4096
43 43
44/* Default timeout values */
45#define NFS_MAX_UDP_TIMEOUT (60*HZ)
46#define NFS_MAX_TCP_TIMEOUT (600*HZ)
47
44/* 48/*
45 * superblock magic number for NFS 49 * superblock magic number for NFS
46 */ 50 */
diff --git a/include/linux/nfs_xdr.h b/include/linux/nfs_xdr.h
index a2bf6914ff1b..60086dac11d5 100644
--- a/include/linux/nfs_xdr.h
+++ b/include/linux/nfs_xdr.h
@@ -96,12 +96,13 @@ struct nfs4_change_info {
96 u64 after; 96 u64 after;
97}; 97};
98 98
99struct nfs_seqid;
99/* 100/*
100 * Arguments to the open call. 101 * Arguments to the open call.
101 */ 102 */
102struct nfs_openargs { 103struct nfs_openargs {
103 const struct nfs_fh * fh; 104 const struct nfs_fh * fh;
104 __u32 seqid; 105 struct nfs_seqid * seqid;
105 int open_flags; 106 int open_flags;
106 __u64 clientid; 107 __u64 clientid;
107 __u32 id; 108 __u32 id;
@@ -136,7 +137,7 @@ struct nfs_openres {
136struct nfs_open_confirmargs { 137struct nfs_open_confirmargs {
137 const struct nfs_fh * fh; 138 const struct nfs_fh * fh;
138 nfs4_stateid stateid; 139 nfs4_stateid stateid;
139 __u32 seqid; 140 struct nfs_seqid * seqid;
140}; 141};
141 142
142struct nfs_open_confirmres { 143struct nfs_open_confirmres {
@@ -148,8 +149,8 @@ struct nfs_open_confirmres {
148 */ 149 */
149struct nfs_closeargs { 150struct nfs_closeargs {
150 struct nfs_fh * fh; 151 struct nfs_fh * fh;
151 nfs4_stateid stateid; 152 nfs4_stateid * stateid;
152 __u32 seqid; 153 struct nfs_seqid * seqid;
153 int open_flags; 154 int open_flags;
154}; 155};
155 156
@@ -164,30 +165,19 @@ struct nfs_lowner {
164 u32 id; 165 u32 id;
165}; 166};
166 167
167struct nfs_open_to_lock {
168 __u32 open_seqid;
169 nfs4_stateid open_stateid;
170 __u32 lock_seqid;
171 struct nfs_lowner lock_owner;
172};
173
174struct nfs_exist_lock {
175 nfs4_stateid stateid;
176 __u32 seqid;
177};
178
179struct nfs_lock_opargs { 168struct nfs_lock_opargs {
169 struct nfs_seqid * lock_seqid;
170 nfs4_stateid * lock_stateid;
171 struct nfs_seqid * open_seqid;
172 nfs4_stateid * open_stateid;
173 struct nfs_lowner lock_owner;
180 __u32 reclaim; 174 __u32 reclaim;
181 __u32 new_lock_owner; 175 __u32 new_lock_owner;
182 union {
183 struct nfs_open_to_lock *open_lock;
184 struct nfs_exist_lock *exist_lock;
185 } u;
186}; 176};
187 177
188struct nfs_locku_opargs { 178struct nfs_locku_opargs {
189 __u32 seqid; 179 struct nfs_seqid * seqid;
190 nfs4_stateid stateid; 180 nfs4_stateid * stateid;
191}; 181};
192 182
193struct nfs_lockargs { 183struct nfs_lockargs {
@@ -722,7 +712,7 @@ struct nfs_rpc_ops {
722 int (*write) (struct nfs_write_data *); 712 int (*write) (struct nfs_write_data *);
723 int (*commit) (struct nfs_write_data *); 713 int (*commit) (struct nfs_write_data *);
724 int (*create) (struct inode *, struct dentry *, 714 int (*create) (struct inode *, struct dentry *,
725 struct iattr *, int); 715 struct iattr *, int, struct nameidata *);
726 int (*remove) (struct inode *, struct qstr *); 716 int (*remove) (struct inode *, struct qstr *);
727 int (*unlink_setup) (struct rpc_message *, 717 int (*unlink_setup) (struct rpc_message *,
728 struct dentry *, struct qstr *); 718 struct dentry *, struct qstr *);
diff --git a/include/linux/sunrpc/auth.h b/include/linux/sunrpc/auth.h
index 04ebc24db348..b68c11a2d6dd 100644
--- a/include/linux/sunrpc/auth.h
+++ b/include/linux/sunrpc/auth.h
@@ -66,7 +66,12 @@ struct rpc_cred_cache {
66 66
67struct rpc_auth { 67struct rpc_auth {
68 unsigned int au_cslack; /* call cred size estimate */ 68 unsigned int au_cslack; /* call cred size estimate */
69 unsigned int au_rslack; /* reply verf size guess */ 69 /* guess at number of u32's auth adds before
70 * reply data; normally the verifier size: */
71 unsigned int au_rslack;
72 /* for gss, used to calculate au_rslack: */
73 unsigned int au_verfsize;
74
70 unsigned int au_flags; /* various flags */ 75 unsigned int au_flags; /* various flags */
71 struct rpc_authops * au_ops; /* operations */ 76 struct rpc_authops * au_ops; /* operations */
72 rpc_authflavor_t au_flavor; /* pseudoflavor (note may 77 rpc_authflavor_t au_flavor; /* pseudoflavor (note may
diff --git a/include/linux/sunrpc/debug.h b/include/linux/sunrpc/debug.h
index eadb31e3c198..1a42d902bc11 100644
--- a/include/linux/sunrpc/debug.h
+++ b/include/linux/sunrpc/debug.h
@@ -32,6 +32,7 @@
32#define RPCDBG_AUTH 0x0010 32#define RPCDBG_AUTH 0x0010
33#define RPCDBG_PMAP 0x0020 33#define RPCDBG_PMAP 0x0020
34#define RPCDBG_SCHED 0x0040 34#define RPCDBG_SCHED 0x0040
35#define RPCDBG_TRANS 0x0080
35#define RPCDBG_SVCSOCK 0x0100 36#define RPCDBG_SVCSOCK 0x0100
36#define RPCDBG_SVCDSP 0x0200 37#define RPCDBG_SVCDSP 0x0200
37#define RPCDBG_MISC 0x0400 38#define RPCDBG_MISC 0x0400
@@ -94,6 +95,8 @@ enum {
94 CTL_NLMDEBUG, 95 CTL_NLMDEBUG,
95 CTL_SLOTTABLE_UDP, 96 CTL_SLOTTABLE_UDP,
96 CTL_SLOTTABLE_TCP, 97 CTL_SLOTTABLE_TCP,
98 CTL_MIN_RESVPORT,
99 CTL_MAX_RESVPORT,
97}; 100};
98 101
99#endif /* _LINUX_SUNRPC_DEBUG_H_ */ 102#endif /* _LINUX_SUNRPC_DEBUG_H_ */
diff --git a/include/linux/sunrpc/gss_api.h b/include/linux/sunrpc/gss_api.h
index 689262f63059..9b8bcf125c18 100644
--- a/include/linux/sunrpc/gss_api.h
+++ b/include/linux/sunrpc/gss_api.h
@@ -40,14 +40,21 @@ int gss_import_sec_context(
40 struct gss_ctx **ctx_id); 40 struct gss_ctx **ctx_id);
41u32 gss_get_mic( 41u32 gss_get_mic(
42 struct gss_ctx *ctx_id, 42 struct gss_ctx *ctx_id,
43 u32 qop,
44 struct xdr_buf *message, 43 struct xdr_buf *message,
45 struct xdr_netobj *mic_token); 44 struct xdr_netobj *mic_token);
46u32 gss_verify_mic( 45u32 gss_verify_mic(
47 struct gss_ctx *ctx_id, 46 struct gss_ctx *ctx_id,
48 struct xdr_buf *message, 47 struct xdr_buf *message,
49 struct xdr_netobj *mic_token, 48 struct xdr_netobj *mic_token);
50 u32 *qstate); 49u32 gss_wrap(
50 struct gss_ctx *ctx_id,
51 int offset,
52 struct xdr_buf *outbuf,
53 struct page **inpages);
54u32 gss_unwrap(
55 struct gss_ctx *ctx_id,
56 int offset,
57 struct xdr_buf *inbuf);
51u32 gss_delete_sec_context( 58u32 gss_delete_sec_context(
52 struct gss_ctx **ctx_id); 59 struct gss_ctx **ctx_id);
53 60
@@ -56,7 +63,6 @@ char *gss_service_to_auth_domain_name(struct gss_api_mech *, u32 service);
56 63
57struct pf_desc { 64struct pf_desc {
58 u32 pseudoflavor; 65 u32 pseudoflavor;
59 u32 qop;
60 u32 service; 66 u32 service;
61 char *name; 67 char *name;
62 char *auth_domain_name; 68 char *auth_domain_name;
@@ -85,14 +91,21 @@ struct gss_api_ops {
85 struct gss_ctx *ctx_id); 91 struct gss_ctx *ctx_id);
86 u32 (*gss_get_mic)( 92 u32 (*gss_get_mic)(
87 struct gss_ctx *ctx_id, 93 struct gss_ctx *ctx_id,
88 u32 qop,
89 struct xdr_buf *message, 94 struct xdr_buf *message,
90 struct xdr_netobj *mic_token); 95 struct xdr_netobj *mic_token);
91 u32 (*gss_verify_mic)( 96 u32 (*gss_verify_mic)(
92 struct gss_ctx *ctx_id, 97 struct gss_ctx *ctx_id,
93 struct xdr_buf *message, 98 struct xdr_buf *message,
94 struct xdr_netobj *mic_token, 99 struct xdr_netobj *mic_token);
95 u32 *qstate); 100 u32 (*gss_wrap)(
101 struct gss_ctx *ctx_id,
102 int offset,
103 struct xdr_buf *outbuf,
104 struct page **inpages);
105 u32 (*gss_unwrap)(
106 struct gss_ctx *ctx_id,
107 int offset,
108 struct xdr_buf *buf);
96 void (*gss_delete_sec_context)( 109 void (*gss_delete_sec_context)(
97 void *internal_ctx_id); 110 void *internal_ctx_id);
98}; 111};
diff --git a/include/linux/sunrpc/gss_err.h b/include/linux/sunrpc/gss_err.h
index 92608a2e574c..a6807867bd21 100644
--- a/include/linux/sunrpc/gss_err.h
+++ b/include/linux/sunrpc/gss_err.h
@@ -66,16 +66,6 @@ typedef unsigned int OM_uint32;
66 66
67 67
68/* 68/*
69 * Define the default Quality of Protection for per-message services. Note
70 * that an implementation that offers multiple levels of QOP may either reserve
71 * a value (for example zero, as assumed here) to mean "default protection", or
72 * alternatively may simply equate GSS_C_QOP_DEFAULT to a specific explicit
73 * QOP value. However a value of 0 should always be interpreted by a GSSAPI
74 * implementation as a request for the default protection level.
75 */
76#define GSS_C_QOP_DEFAULT 0
77
78/*
79 * Expiration time of 2^32-1 seconds means infinite lifetime for a 69 * Expiration time of 2^32-1 seconds means infinite lifetime for a
80 * credential or security context 70 * credential or security context
81 */ 71 */
diff --git a/include/linux/sunrpc/gss_krb5.h b/include/linux/sunrpc/gss_krb5.h
index ffe31d2eb9ec..2c3601d31045 100644
--- a/include/linux/sunrpc/gss_krb5.h
+++ b/include/linux/sunrpc/gss_krb5.h
@@ -116,18 +116,22 @@ enum seal_alg {
116 116
117s32 117s32
118make_checksum(s32 cksumtype, char *header, int hdrlen, struct xdr_buf *body, 118make_checksum(s32 cksumtype, char *header, int hdrlen, struct xdr_buf *body,
119 struct xdr_netobj *cksum); 119 int body_offset, struct xdr_netobj *cksum);
120
121u32 gss_get_mic_kerberos(struct gss_ctx *, struct xdr_buf *,
122 struct xdr_netobj *);
123
124u32 gss_verify_mic_kerberos(struct gss_ctx *, struct xdr_buf *,
125 struct xdr_netobj *);
120 126
121u32 127u32
122krb5_make_token(struct krb5_ctx *context_handle, int qop_req, 128gss_wrap_kerberos(struct gss_ctx *ctx_id, int offset,
123 struct xdr_buf *input_message_buffer, 129 struct xdr_buf *outbuf, struct page **pages);
124 struct xdr_netobj *output_message_buffer, int toktype);
125 130
126u32 131u32
127krb5_read_token(struct krb5_ctx *context_handle, 132gss_unwrap_kerberos(struct gss_ctx *ctx_id, int offset,
128 struct xdr_netobj *input_token_buffer, 133 struct xdr_buf *buf);
129 struct xdr_buf *message_buffer, 134
130 int *qop_state, int toktype);
131 135
132u32 136u32
133krb5_encrypt(struct crypto_tfm * key, 137krb5_encrypt(struct crypto_tfm * key,
@@ -137,6 +141,13 @@ u32
137krb5_decrypt(struct crypto_tfm * key, 141krb5_decrypt(struct crypto_tfm * key,
138 void *iv, void *in, void *out, int length); 142 void *iv, void *in, void *out, int length);
139 143
144int
145gss_encrypt_xdr_buf(struct crypto_tfm *tfm, struct xdr_buf *outbuf, int offset,
146 struct page **pages);
147
148int
149gss_decrypt_xdr_buf(struct crypto_tfm *tfm, struct xdr_buf *inbuf, int offset);
150
140s32 151s32
141krb5_make_seq_num(struct crypto_tfm * key, 152krb5_make_seq_num(struct crypto_tfm * key,
142 int direction, 153 int direction,
diff --git a/include/linux/sunrpc/gss_spkm3.h b/include/linux/sunrpc/gss_spkm3.h
index b5c9968c3c17..0beb2cf00a84 100644
--- a/include/linux/sunrpc/gss_spkm3.h
+++ b/include/linux/sunrpc/gss_spkm3.h
@@ -41,9 +41,9 @@ struct spkm3_ctx {
41#define SPKM_WRAP_TOK 5 41#define SPKM_WRAP_TOK 5
42#define SPKM_DEL_TOK 6 42#define SPKM_DEL_TOK 6
43 43
44u32 spkm3_make_token(struct spkm3_ctx *ctx, int qop_req, struct xdr_buf * text, struct xdr_netobj * token, int toktype); 44u32 spkm3_make_token(struct spkm3_ctx *ctx, struct xdr_buf * text, struct xdr_netobj * token, int toktype);
45 45
46u32 spkm3_read_token(struct spkm3_ctx *ctx, struct xdr_netobj *read_token, struct xdr_buf *message_buffer, int *qop_state, int toktype); 46u32 spkm3_read_token(struct spkm3_ctx *ctx, struct xdr_netobj *read_token, struct xdr_buf *message_buffer, int toktype);
47 47
48#define CKSUMTYPE_RSA_MD5 0x0007 48#define CKSUMTYPE_RSA_MD5 0x0007
49 49
diff --git a/include/linux/sunrpc/msg_prot.h b/include/linux/sunrpc/msg_prot.h
index 15f115332389..f43f237360ae 100644
--- a/include/linux/sunrpc/msg_prot.h
+++ b/include/linux/sunrpc/msg_prot.h
@@ -76,5 +76,30 @@ enum rpc_auth_stat {
76 76
77#define RPC_MAXNETNAMELEN 256 77#define RPC_MAXNETNAMELEN 256
78 78
79/*
80 * From RFC 1831:
81 *
82 * "A record is composed of one or more record fragments. A record
83 * fragment is a four-byte header followed by 0 to (2**31) - 1 bytes of
84 * fragment data. The bytes encode an unsigned binary number; as with
85 * XDR integers, the byte order is from highest to lowest. The number
86 * encodes two values -- a boolean which indicates whether the fragment
87 * is the last fragment of the record (bit value 1 implies the fragment
88 * is the last fragment) and a 31-bit unsigned binary value which is the
89 * length in bytes of the fragment's data. The boolean value is the
90 * highest-order bit of the header; the length is the 31 low-order bits.
91 * (Note that this record specification is NOT in XDR standard form!)"
92 *
93 * The Linux RPC client always sends its requests in a single record
94 * fragment, limiting the maximum payload size for stream transports to
95 * 2GB.
96 */
97
98typedef u32 rpc_fraghdr;
99
100#define RPC_LAST_STREAM_FRAGMENT (1U << 31)
101#define RPC_FRAGMENT_SIZE_MASK (~RPC_LAST_STREAM_FRAGMENT)
102#define RPC_MAX_FRAGMENT_SIZE ((1U << 31) - 1)
103
79#endif /* __KERNEL__ */ 104#endif /* __KERNEL__ */
80#endif /* _LINUX_SUNRPC_MSGPROT_H_ */ 105#endif /* _LINUX_SUNRPC_MSGPROT_H_ */
diff --git a/include/linux/sunrpc/xdr.h b/include/linux/sunrpc/xdr.h
index 23448d0fb5bc..5da968729cf8 100644
--- a/include/linux/sunrpc/xdr.h
+++ b/include/linux/sunrpc/xdr.h
@@ -161,14 +161,10 @@ typedef struct {
161 161
162typedef size_t (*skb_read_actor_t)(skb_reader_t *desc, void *to, size_t len); 162typedef size_t (*skb_read_actor_t)(skb_reader_t *desc, void *to, size_t len);
163 163
164extern int csum_partial_copy_to_xdr(struct xdr_buf *, struct sk_buff *);
164extern ssize_t xdr_partial_copy_from_skb(struct xdr_buf *, unsigned int, 165extern ssize_t xdr_partial_copy_from_skb(struct xdr_buf *, unsigned int,
165 skb_reader_t *, skb_read_actor_t); 166 skb_reader_t *, skb_read_actor_t);
166 167
167struct socket;
168struct sockaddr;
169extern int xdr_sendpages(struct socket *, struct sockaddr *, int,
170 struct xdr_buf *, unsigned int, int);
171
172extern int xdr_encode_word(struct xdr_buf *, int, u32); 168extern int xdr_encode_word(struct xdr_buf *, int, u32);
173extern int xdr_decode_word(struct xdr_buf *, int, u32 *); 169extern int xdr_decode_word(struct xdr_buf *, int, u32 *);
174 170
diff --git a/include/linux/sunrpc/xprt.h b/include/linux/sunrpc/xprt.h
index e618c1649814..3b8b6e823c70 100644
--- a/include/linux/sunrpc/xprt.h
+++ b/include/linux/sunrpc/xprt.h
@@ -1,5 +1,5 @@
1/* 1/*
2 * linux/include/linux/sunrpc/clnt_xprt.h 2 * linux/include/linux/sunrpc/xprt.h
3 * 3 *
4 * Declarations for the RPC transport interface. 4 * Declarations for the RPC transport interface.
5 * 5 *
@@ -15,20 +15,6 @@
15#include <linux/sunrpc/sched.h> 15#include <linux/sunrpc/sched.h>
16#include <linux/sunrpc/xdr.h> 16#include <linux/sunrpc/xdr.h>
17 17
18/*
19 * The transport code maintains an estimate on the maximum number of out-
20 * standing RPC requests, using a smoothed version of the congestion
21 * avoidance implemented in 44BSD. This is basically the Van Jacobson
22 * congestion algorithm: If a retransmit occurs, the congestion window is
23 * halved; otherwise, it is incremented by 1/cwnd when
24 *
25 * - a reply is received and
26 * - a full number of requests are outstanding and
27 * - the congestion window hasn't been updated recently.
28 *
29 * Upper procedures may check whether a request would block waiting for
30 * a free RPC slot by using the RPC_CONGESTED() macro.
31 */
32extern unsigned int xprt_udp_slot_table_entries; 18extern unsigned int xprt_udp_slot_table_entries;
33extern unsigned int xprt_tcp_slot_table_entries; 19extern unsigned int xprt_tcp_slot_table_entries;
34 20
@@ -36,34 +22,23 @@ extern unsigned int xprt_tcp_slot_table_entries;
36#define RPC_DEF_SLOT_TABLE (16U) 22#define RPC_DEF_SLOT_TABLE (16U)
37#define RPC_MAX_SLOT_TABLE (128U) 23#define RPC_MAX_SLOT_TABLE (128U)
38 24
39#define RPC_CWNDSHIFT (8U)
40#define RPC_CWNDSCALE (1U << RPC_CWNDSHIFT)
41#define RPC_INITCWND RPC_CWNDSCALE
42#define RPC_MAXCWND(xprt) ((xprt)->max_reqs << RPC_CWNDSHIFT)
43#define RPCXPRT_CONGESTED(xprt) ((xprt)->cong >= (xprt)->cwnd)
44
45/* Default timeout values */
46#define RPC_MAX_UDP_TIMEOUT (60*HZ)
47#define RPC_MAX_TCP_TIMEOUT (600*HZ)
48
49/* 25/*
50 * Wait duration for an RPC TCP connection to be established. Solaris 26 * RPC call and reply header size as number of 32bit words (verifier
51 * NFS over TCP uses 60 seconds, for example, which is in line with how 27 * size computed separately)
52 * long a server takes to reboot.
53 */ 28 */
54#define RPC_CONNECT_TIMEOUT (60*HZ) 29#define RPC_CALLHDRSIZE 6
30#define RPC_REPHDRSIZE 4
55 31
56/* 32/*
57 * Delay an arbitrary number of seconds before attempting to reconnect 33 * Parameters for choosing a free port
58 * after an error.
59 */ 34 */
60#define RPC_REESTABLISH_TIMEOUT (15*HZ) 35extern unsigned int xprt_min_resvport;
36extern unsigned int xprt_max_resvport;
61 37
62/* RPC call and reply header size as number of 32bit words (verifier 38#define RPC_MIN_RESVPORT (1U)
63 * size computed separately) 39#define RPC_MAX_RESVPORT (65535U)
64 */ 40#define RPC_DEF_MIN_RESVPORT (650U)
65#define RPC_CALLHDRSIZE 6 41#define RPC_DEF_MAX_RESVPORT (1023U)
66#define RPC_REPHDRSIZE 4
67 42
68/* 43/*
69 * This describes a timeout strategy 44 * This describes a timeout strategy
@@ -76,6 +51,9 @@ struct rpc_timeout {
76 unsigned char to_exponential; 51 unsigned char to_exponential;
77}; 52};
78 53
54struct rpc_task;
55struct rpc_xprt;
56
79/* 57/*
80 * This describes a complete RPC request 58 * This describes a complete RPC request
81 */ 59 */
@@ -95,7 +73,10 @@ struct rpc_rqst {
95 int rq_cong; /* has incremented xprt->cong */ 73 int rq_cong; /* has incremented xprt->cong */
96 int rq_received; /* receive completed */ 74 int rq_received; /* receive completed */
97 u32 rq_seqno; /* gss seq no. used on req. */ 75 u32 rq_seqno; /* gss seq no. used on req. */
98 76 int rq_enc_pages_num;
77 struct page **rq_enc_pages; /* scratch pages for use by
78 gss privacy code */
79 void (*rq_release_snd_buf)(struct rpc_rqst *); /* release rq_enc_pages */
99 struct list_head rq_list; 80 struct list_head rq_list;
100 81
101 struct xdr_buf rq_private_buf; /* The receive buffer 82 struct xdr_buf rq_private_buf; /* The receive buffer
@@ -121,12 +102,21 @@ struct rpc_rqst {
121#define rq_svec rq_snd_buf.head 102#define rq_svec rq_snd_buf.head
122#define rq_slen rq_snd_buf.len 103#define rq_slen rq_snd_buf.len
123 104
124#define XPRT_LAST_FRAG (1 << 0) 105struct rpc_xprt_ops {
125#define XPRT_COPY_RECM (1 << 1) 106 void (*set_buffer_size)(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize);
126#define XPRT_COPY_XID (1 << 2) 107 int (*reserve_xprt)(struct rpc_task *task);
127#define XPRT_COPY_DATA (1 << 3) 108 void (*release_xprt)(struct rpc_xprt *xprt, struct rpc_task *task);
109 void (*connect)(struct rpc_task *task);
110 int (*send_request)(struct rpc_task *task);
111 void (*set_retrans_timeout)(struct rpc_task *task);
112 void (*timer)(struct rpc_task *task);
113 void (*release_request)(struct rpc_task *task);
114 void (*close)(struct rpc_xprt *xprt);
115 void (*destroy)(struct rpc_xprt *xprt);
116};
128 117
129struct rpc_xprt { 118struct rpc_xprt {
119 struct rpc_xprt_ops * ops; /* transport methods */
130 struct socket * sock; /* BSD socket layer */ 120 struct socket * sock; /* BSD socket layer */
131 struct sock * inet; /* INET layer */ 121 struct sock * inet; /* INET layer */
132 122
@@ -137,11 +127,13 @@ struct rpc_xprt {
137 unsigned long cong; /* current congestion */ 127 unsigned long cong; /* current congestion */
138 unsigned long cwnd; /* congestion window */ 128 unsigned long cwnd; /* congestion window */
139 129
140 unsigned int rcvsize, /* socket receive buffer size */ 130 size_t rcvsize, /* transport rcv buffer size */
141 sndsize; /* socket send buffer size */ 131 sndsize; /* transport send buffer size */
142 132
143 size_t max_payload; /* largest RPC payload size, 133 size_t max_payload; /* largest RPC payload size,
144 in bytes */ 134 in bytes */
135 unsigned int tsh_size; /* size of transport specific
136 header */
145 137
146 struct rpc_wait_queue sending; /* requests waiting to send */ 138 struct rpc_wait_queue sending; /* requests waiting to send */
147 struct rpc_wait_queue resend; /* requests waiting to resend */ 139 struct rpc_wait_queue resend; /* requests waiting to resend */
@@ -150,11 +142,9 @@ struct rpc_xprt {
150 struct list_head free; /* free slots */ 142 struct list_head free; /* free slots */
151 struct rpc_rqst * slot; /* slot table storage */ 143 struct rpc_rqst * slot; /* slot table storage */
152 unsigned int max_reqs; /* total slots */ 144 unsigned int max_reqs; /* total slots */
153 unsigned long sockstate; /* Socket state */ 145 unsigned long state; /* transport state */
154 unsigned char shutdown : 1, /* being shut down */ 146 unsigned char shutdown : 1, /* being shut down */
155 nocong : 1, /* no congestion control */ 147 resvport : 1; /* use a reserved port */
156 resvport : 1, /* use a reserved port */
157 stream : 1; /* TCP */
158 148
159 /* 149 /*
160 * XID 150 * XID
@@ -171,22 +161,27 @@ struct rpc_xprt {
171 unsigned long tcp_copied, /* copied to request */ 161 unsigned long tcp_copied, /* copied to request */
172 tcp_flags; 162 tcp_flags;
173 /* 163 /*
174 * Connection of sockets 164 * Connection of transports
175 */ 165 */
176 struct work_struct sock_connect; 166 unsigned long connect_timeout,
167 bind_timeout,
168 reestablish_timeout;
169 struct work_struct connect_worker;
177 unsigned short port; 170 unsigned short port;
171
178 /* 172 /*
179 * Disconnection of idle sockets 173 * Disconnection of idle transports
180 */ 174 */
181 struct work_struct task_cleanup; 175 struct work_struct task_cleanup;
182 struct timer_list timer; 176 struct timer_list timer;
183 unsigned long last_used; 177 unsigned long last_used,
178 idle_timeout;
184 179
185 /* 180 /*
186 * Send stuff 181 * Send stuff
187 */ 182 */
188 spinlock_t sock_lock; /* lock socket info */ 183 spinlock_t transport_lock; /* lock transport info */
189 spinlock_t xprt_lock; /* lock xprt info */ 184 spinlock_t reserve_lock; /* lock slot table */
190 struct rpc_task * snd_task; /* Task blocked in send */ 185 struct rpc_task * snd_task; /* Task blocked in send */
191 186
192 struct list_head recv; 187 struct list_head recv;
@@ -195,37 +190,111 @@ struct rpc_xprt {
195 void (*old_data_ready)(struct sock *, int); 190 void (*old_data_ready)(struct sock *, int);
196 void (*old_state_change)(struct sock *); 191 void (*old_state_change)(struct sock *);
197 void (*old_write_space)(struct sock *); 192 void (*old_write_space)(struct sock *);
198
199 wait_queue_head_t cong_wait;
200}; 193};
201 194
195#define XPRT_LAST_FRAG (1 << 0)
196#define XPRT_COPY_RECM (1 << 1)
197#define XPRT_COPY_XID (1 << 2)
198#define XPRT_COPY_DATA (1 << 3)
199
202#ifdef __KERNEL__ 200#ifdef __KERNEL__
203 201
204struct rpc_xprt * xprt_create_proto(int proto, struct sockaddr_in *addr, 202/*
205 struct rpc_timeout *toparms); 203 * Transport operations used by ULPs
206int xprt_destroy(struct rpc_xprt *); 204 */
207void xprt_set_timeout(struct rpc_timeout *, unsigned int, 205struct rpc_xprt * xprt_create_proto(int proto, struct sockaddr_in *addr, struct rpc_timeout *to);
208 unsigned long); 206void xprt_set_timeout(struct rpc_timeout *to, unsigned int retr, unsigned long incr);
209 207
210void xprt_reserve(struct rpc_task *); 208/*
211int xprt_prepare_transmit(struct rpc_task *); 209 * Generic internal transport functions
212void xprt_transmit(struct rpc_task *); 210 */
213void xprt_receive(struct rpc_task *); 211void xprt_connect(struct rpc_task *task);
212void xprt_reserve(struct rpc_task *task);
213int xprt_reserve_xprt(struct rpc_task *task);
214int xprt_reserve_xprt_cong(struct rpc_task *task);
215int xprt_prepare_transmit(struct rpc_task *task);
216void xprt_transmit(struct rpc_task *task);
217void xprt_abort_transmit(struct rpc_task *task);
214int xprt_adjust_timeout(struct rpc_rqst *req); 218int xprt_adjust_timeout(struct rpc_rqst *req);
215void xprt_release(struct rpc_task *); 219void xprt_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task);
216void xprt_connect(struct rpc_task *); 220void xprt_release_xprt_cong(struct rpc_xprt *xprt, struct rpc_task *task);
217void xprt_sock_setbufsize(struct rpc_xprt *); 221void xprt_release(struct rpc_task *task);
218 222int xprt_destroy(struct rpc_xprt *xprt);
219#define XPRT_LOCKED 0 223
220#define XPRT_CONNECT 1 224static inline u32 *xprt_skip_transport_header(struct rpc_xprt *xprt, u32 *p)
221#define XPRT_CONNECTING 2 225{
222 226 return p + xprt->tsh_size;
223#define xprt_connected(xp) (test_bit(XPRT_CONNECT, &(xp)->sockstate)) 227}
224#define xprt_set_connected(xp) (set_bit(XPRT_CONNECT, &(xp)->sockstate)) 228
225#define xprt_test_and_set_connected(xp) (test_and_set_bit(XPRT_CONNECT, &(xp)->sockstate)) 229/*
226#define xprt_test_and_clear_connected(xp) \ 230 * Transport switch helper functions
227 (test_and_clear_bit(XPRT_CONNECT, &(xp)->sockstate)) 231 */
228#define xprt_clear_connected(xp) (clear_bit(XPRT_CONNECT, &(xp)->sockstate)) 232void xprt_set_retrans_timeout_def(struct rpc_task *task);
233void xprt_set_retrans_timeout_rtt(struct rpc_task *task);
234void xprt_wake_pending_tasks(struct rpc_xprt *xprt, int status);
235void xprt_wait_for_buffer_space(struct rpc_task *task);
236void xprt_write_space(struct rpc_xprt *xprt);
237void xprt_update_rtt(struct rpc_task *task);
238void xprt_adjust_cwnd(struct rpc_task *task, int result);
239struct rpc_rqst * xprt_lookup_rqst(struct rpc_xprt *xprt, u32 xid);
240void xprt_complete_rqst(struct rpc_task *task, int copied);
241void xprt_release_rqst_cong(struct rpc_task *task);
242void xprt_disconnect(struct rpc_xprt *xprt);
243
244/*
245 * Socket transport setup operations
246 */
247int xs_setup_udp(struct rpc_xprt *xprt, struct rpc_timeout *to);
248int xs_setup_tcp(struct rpc_xprt *xprt, struct rpc_timeout *to);
249
250/*
251 * Reserved bit positions in xprt->state
252 */
253#define XPRT_LOCKED (0)
254#define XPRT_CONNECTED (1)
255#define XPRT_CONNECTING (2)
256
257static inline void xprt_set_connected(struct rpc_xprt *xprt)
258{
259 set_bit(XPRT_CONNECTED, &xprt->state);
260}
261
262static inline void xprt_clear_connected(struct rpc_xprt *xprt)
263{
264 clear_bit(XPRT_CONNECTED, &xprt->state);
265}
266
267static inline int xprt_connected(struct rpc_xprt *xprt)
268{
269 return test_bit(XPRT_CONNECTED, &xprt->state);
270}
271
272static inline int xprt_test_and_set_connected(struct rpc_xprt *xprt)
273{
274 return test_and_set_bit(XPRT_CONNECTED, &xprt->state);
275}
276
277static inline int xprt_test_and_clear_connected(struct rpc_xprt *xprt)
278{
279 return test_and_clear_bit(XPRT_CONNECTED, &xprt->state);
280}
281
282static inline void xprt_clear_connecting(struct rpc_xprt *xprt)
283{
284 smp_mb__before_clear_bit();
285 clear_bit(XPRT_CONNECTING, &xprt->state);
286 smp_mb__after_clear_bit();
287}
288
289static inline int xprt_connecting(struct rpc_xprt *xprt)
290{
291 return test_bit(XPRT_CONNECTING, &xprt->state);
292}
293
294static inline int xprt_test_and_set_connecting(struct rpc_xprt *xprt)
295{
296 return test_and_set_bit(XPRT_CONNECTING, &xprt->state);
297}
229 298
230#endif /* __KERNEL__*/ 299#endif /* __KERNEL__*/
231 300
diff --git a/net/sunrpc/Makefile b/net/sunrpc/Makefile
index 46a2ce00a29b..cdcab9ca4c60 100644
--- a/net/sunrpc/Makefile
+++ b/net/sunrpc/Makefile
@@ -6,7 +6,7 @@
6obj-$(CONFIG_SUNRPC) += sunrpc.o 6obj-$(CONFIG_SUNRPC) += sunrpc.o
7obj-$(CONFIG_SUNRPC_GSS) += auth_gss/ 7obj-$(CONFIG_SUNRPC_GSS) += auth_gss/
8 8
9sunrpc-y := clnt.o xprt.o sched.o \ 9sunrpc-y := clnt.o xprt.o socklib.o xprtsock.o sched.o \
10 auth.o auth_null.o auth_unix.o \ 10 auth.o auth_null.o auth_unix.o \
11 svc.o svcsock.o svcauth.o svcauth_unix.o \ 11 svc.o svcsock.o svcauth.o svcauth_unix.o \
12 pmap_clnt.o timer.o xdr.o \ 12 pmap_clnt.o timer.o xdr.o \
diff --git a/net/sunrpc/auth.c b/net/sunrpc/auth.c
index 505e2d4b3d62..a415d99c394d 100644
--- a/net/sunrpc/auth.c
+++ b/net/sunrpc/auth.c
@@ -11,7 +11,6 @@
11#include <linux/module.h> 11#include <linux/module.h>
12#include <linux/slab.h> 12#include <linux/slab.h>
13#include <linux/errno.h> 13#include <linux/errno.h>
14#include <linux/socket.h>
15#include <linux/sunrpc/clnt.h> 14#include <linux/sunrpc/clnt.h>
16#include <linux/spinlock.h> 15#include <linux/spinlock.h>
17 16
diff --git a/net/sunrpc/auth_gss/Makefile b/net/sunrpc/auth_gss/Makefile
index fe1b874084bc..f3431a7e33da 100644
--- a/net/sunrpc/auth_gss/Makefile
+++ b/net/sunrpc/auth_gss/Makefile
@@ -10,7 +10,7 @@ auth_rpcgss-objs := auth_gss.o gss_generic_token.o \
10obj-$(CONFIG_RPCSEC_GSS_KRB5) += rpcsec_gss_krb5.o 10obj-$(CONFIG_RPCSEC_GSS_KRB5) += rpcsec_gss_krb5.o
11 11
12rpcsec_gss_krb5-objs := gss_krb5_mech.o gss_krb5_seal.o gss_krb5_unseal.o \ 12rpcsec_gss_krb5-objs := gss_krb5_mech.o gss_krb5_seal.o gss_krb5_unseal.o \
13 gss_krb5_seqnum.o 13 gss_krb5_seqnum.o gss_krb5_wrap.o
14 14
15obj-$(CONFIG_RPCSEC_GSS_SPKM3) += rpcsec_gss_spkm3.o 15obj-$(CONFIG_RPCSEC_GSS_SPKM3) += rpcsec_gss_spkm3.o
16 16
diff --git a/net/sunrpc/auth_gss/auth_gss.c b/net/sunrpc/auth_gss/auth_gss.c
index 2f7b867161d2..f44f46f1d8e0 100644
--- a/net/sunrpc/auth_gss/auth_gss.c
+++ b/net/sunrpc/auth_gss/auth_gss.c
@@ -42,9 +42,8 @@
42#include <linux/init.h> 42#include <linux/init.h>
43#include <linux/types.h> 43#include <linux/types.h>
44#include <linux/slab.h> 44#include <linux/slab.h>
45#include <linux/socket.h>
46#include <linux/in.h>
47#include <linux/sched.h> 45#include <linux/sched.h>
46#include <linux/pagemap.h>
48#include <linux/sunrpc/clnt.h> 47#include <linux/sunrpc/clnt.h>
49#include <linux/sunrpc/auth.h> 48#include <linux/sunrpc/auth.h>
50#include <linux/sunrpc/auth_gss.h> 49#include <linux/sunrpc/auth_gss.h>
@@ -846,10 +845,8 @@ gss_marshal(struct rpc_task *task, u32 *p)
846 845
847 /* We compute the checksum for the verifier over the xdr-encoded bytes 846 /* We compute the checksum for the verifier over the xdr-encoded bytes
848 * starting with the xid and ending at the end of the credential: */ 847 * starting with the xid and ending at the end of the credential: */
849 iov.iov_base = req->rq_snd_buf.head[0].iov_base; 848 iov.iov_base = xprt_skip_transport_header(task->tk_xprt,
850 if (task->tk_client->cl_xprt->stream) 849 req->rq_snd_buf.head[0].iov_base);
851 /* See clnt.c:call_header() */
852 iov.iov_base += 4;
853 iov.iov_len = (u8 *)p - (u8 *)iov.iov_base; 850 iov.iov_len = (u8 *)p - (u8 *)iov.iov_base;
854 xdr_buf_from_iov(&iov, &verf_buf); 851 xdr_buf_from_iov(&iov, &verf_buf);
855 852
@@ -857,9 +854,7 @@ gss_marshal(struct rpc_task *task, u32 *p)
857 *p++ = htonl(RPC_AUTH_GSS); 854 *p++ = htonl(RPC_AUTH_GSS);
858 855
859 mic.data = (u8 *)(p + 1); 856 mic.data = (u8 *)(p + 1);
860 maj_stat = gss_get_mic(ctx->gc_gss_ctx, 857 maj_stat = gss_get_mic(ctx->gc_gss_ctx, &verf_buf, &mic);
861 GSS_C_QOP_DEFAULT,
862 &verf_buf, &mic);
863 if (maj_stat == GSS_S_CONTEXT_EXPIRED) { 858 if (maj_stat == GSS_S_CONTEXT_EXPIRED) {
864 cred->cr_flags &= ~RPCAUTH_CRED_UPTODATE; 859 cred->cr_flags &= ~RPCAUTH_CRED_UPTODATE;
865 } else if (maj_stat != 0) { 860 } else if (maj_stat != 0) {
@@ -890,10 +885,8 @@ static u32 *
890gss_validate(struct rpc_task *task, u32 *p) 885gss_validate(struct rpc_task *task, u32 *p)
891{ 886{
892 struct rpc_cred *cred = task->tk_msg.rpc_cred; 887 struct rpc_cred *cred = task->tk_msg.rpc_cred;
893 struct gss_cred *gss_cred = container_of(cred, struct gss_cred,
894 gc_base);
895 struct gss_cl_ctx *ctx = gss_cred_get_ctx(cred); 888 struct gss_cl_ctx *ctx = gss_cred_get_ctx(cred);
896 u32 seq, qop_state; 889 u32 seq;
897 struct kvec iov; 890 struct kvec iov;
898 struct xdr_buf verf_buf; 891 struct xdr_buf verf_buf;
899 struct xdr_netobj mic; 892 struct xdr_netobj mic;
@@ -914,23 +907,14 @@ gss_validate(struct rpc_task *task, u32 *p)
914 mic.data = (u8 *)p; 907 mic.data = (u8 *)p;
915 mic.len = len; 908 mic.len = len;
916 909
917 maj_stat = gss_verify_mic(ctx->gc_gss_ctx, &verf_buf, &mic, &qop_state); 910 maj_stat = gss_verify_mic(ctx->gc_gss_ctx, &verf_buf, &mic);
918 if (maj_stat == GSS_S_CONTEXT_EXPIRED) 911 if (maj_stat == GSS_S_CONTEXT_EXPIRED)
919 cred->cr_flags &= ~RPCAUTH_CRED_UPTODATE; 912 cred->cr_flags &= ~RPCAUTH_CRED_UPTODATE;
920 if (maj_stat) 913 if (maj_stat)
921 goto out_bad; 914 goto out_bad;
922 switch (gss_cred->gc_service) { 915 /* We leave it to unwrap to calculate au_rslack. For now we just
923 case RPC_GSS_SVC_NONE: 916 * calculate the length of the verifier: */
924 /* verifier data, flavor, length: */ 917 task->tk_auth->au_verfsize = XDR_QUADLEN(len) + 2;
925 task->tk_auth->au_rslack = XDR_QUADLEN(len) + 2;
926 break;
927 case RPC_GSS_SVC_INTEGRITY:
928 /* verifier data, flavor, length, length, sequence number: */
929 task->tk_auth->au_rslack = XDR_QUADLEN(len) + 4;
930 break;
931 case RPC_GSS_SVC_PRIVACY:
932 goto out_bad;
933 }
934 gss_put_ctx(ctx); 918 gss_put_ctx(ctx);
935 dprintk("RPC: %4u GSS gss_validate: gss_verify_mic succeeded.\n", 919 dprintk("RPC: %4u GSS gss_validate: gss_verify_mic succeeded.\n",
936 task->tk_pid); 920 task->tk_pid);
@@ -975,8 +959,7 @@ gss_wrap_req_integ(struct rpc_cred *cred, struct gss_cl_ctx *ctx,
975 p = iov->iov_base + iov->iov_len; 959 p = iov->iov_base + iov->iov_len;
976 mic.data = (u8 *)(p + 1); 960 mic.data = (u8 *)(p + 1);
977 961
978 maj_stat = gss_get_mic(ctx->gc_gss_ctx, 962 maj_stat = gss_get_mic(ctx->gc_gss_ctx, &integ_buf, &mic);
979 GSS_C_QOP_DEFAULT, &integ_buf, &mic);
980 status = -EIO; /* XXX? */ 963 status = -EIO; /* XXX? */
981 if (maj_stat == GSS_S_CONTEXT_EXPIRED) 964 if (maj_stat == GSS_S_CONTEXT_EXPIRED)
982 cred->cr_flags &= ~RPCAUTH_CRED_UPTODATE; 965 cred->cr_flags &= ~RPCAUTH_CRED_UPTODATE;
@@ -990,6 +973,113 @@ gss_wrap_req_integ(struct rpc_cred *cred, struct gss_cl_ctx *ctx,
990 return 0; 973 return 0;
991} 974}
992 975
976static void
977priv_release_snd_buf(struct rpc_rqst *rqstp)
978{
979 int i;
980
981 for (i=0; i < rqstp->rq_enc_pages_num; i++)
982 __free_page(rqstp->rq_enc_pages[i]);
983 kfree(rqstp->rq_enc_pages);
984}
985
986static int
987alloc_enc_pages(struct rpc_rqst *rqstp)
988{
989 struct xdr_buf *snd_buf = &rqstp->rq_snd_buf;
990 int first, last, i;
991
992 if (snd_buf->page_len == 0) {
993 rqstp->rq_enc_pages_num = 0;
994 return 0;
995 }
996
997 first = snd_buf->page_base >> PAGE_CACHE_SHIFT;
998 last = (snd_buf->page_base + snd_buf->page_len - 1) >> PAGE_CACHE_SHIFT;
999 rqstp->rq_enc_pages_num = last - first + 1 + 1;
1000 rqstp->rq_enc_pages
1001 = kmalloc(rqstp->rq_enc_pages_num * sizeof(struct page *),
1002 GFP_NOFS);
1003 if (!rqstp->rq_enc_pages)
1004 goto out;
1005 for (i=0; i < rqstp->rq_enc_pages_num; i++) {
1006 rqstp->rq_enc_pages[i] = alloc_page(GFP_NOFS);
1007 if (rqstp->rq_enc_pages[i] == NULL)
1008 goto out_free;
1009 }
1010 rqstp->rq_release_snd_buf = priv_release_snd_buf;
1011 return 0;
1012out_free:
1013 for (i--; i >= 0; i--) {
1014 __free_page(rqstp->rq_enc_pages[i]);
1015 }
1016out:
1017 return -EAGAIN;
1018}
1019
1020static inline int
1021gss_wrap_req_priv(struct rpc_cred *cred, struct gss_cl_ctx *ctx,
1022 kxdrproc_t encode, struct rpc_rqst *rqstp, u32 *p, void *obj)
1023{
1024 struct xdr_buf *snd_buf = &rqstp->rq_snd_buf;
1025 u32 offset;
1026 u32 maj_stat;
1027 int status;
1028 u32 *opaque_len;
1029 struct page **inpages;
1030 int first;
1031 int pad;
1032 struct kvec *iov;
1033 char *tmp;
1034
1035 opaque_len = p++;
1036 offset = (u8 *)p - (u8 *)snd_buf->head[0].iov_base;
1037 *p++ = htonl(rqstp->rq_seqno);
1038
1039 status = encode(rqstp, p, obj);
1040 if (status)
1041 return status;
1042
1043 status = alloc_enc_pages(rqstp);
1044 if (status)
1045 return status;
1046 first = snd_buf->page_base >> PAGE_CACHE_SHIFT;
1047 inpages = snd_buf->pages + first;
1048 snd_buf->pages = rqstp->rq_enc_pages;
1049 snd_buf->page_base -= first << PAGE_CACHE_SHIFT;
1050 /* Give the tail its own page, in case we need extra space in the
1051 * head when wrapping: */
1052 if (snd_buf->page_len || snd_buf->tail[0].iov_len) {
1053 tmp = page_address(rqstp->rq_enc_pages[rqstp->rq_enc_pages_num - 1]);
1054 memcpy(tmp, snd_buf->tail[0].iov_base, snd_buf->tail[0].iov_len);
1055 snd_buf->tail[0].iov_base = tmp;
1056 }
1057 maj_stat = gss_wrap(ctx->gc_gss_ctx, offset, snd_buf, inpages);
1058 /* RPC_SLACK_SPACE should prevent this ever happening: */
1059 BUG_ON(snd_buf->len > snd_buf->buflen);
1060 status = -EIO;
1061 /* We're assuming that when GSS_S_CONTEXT_EXPIRED, the encryption was
1062 * done anyway, so it's safe to put the request on the wire: */
1063 if (maj_stat == GSS_S_CONTEXT_EXPIRED)
1064 cred->cr_flags &= ~RPCAUTH_CRED_UPTODATE;
1065 else if (maj_stat)
1066 return status;
1067
1068 *opaque_len = htonl(snd_buf->len - offset);
1069 /* guess whether we're in the head or the tail: */
1070 if (snd_buf->page_len || snd_buf->tail[0].iov_len)
1071 iov = snd_buf->tail;
1072 else
1073 iov = snd_buf->head;
1074 p = iov->iov_base + iov->iov_len;
1075 pad = 3 - ((snd_buf->len - offset - 1) & 3);
1076 memset(p, 0, pad);
1077 iov->iov_len += pad;
1078 snd_buf->len += pad;
1079
1080 return 0;
1081}
1082
993static int 1083static int
994gss_wrap_req(struct rpc_task *task, 1084gss_wrap_req(struct rpc_task *task,
995 kxdrproc_t encode, void *rqstp, u32 *p, void *obj) 1085 kxdrproc_t encode, void *rqstp, u32 *p, void *obj)
@@ -1017,6 +1107,8 @@ gss_wrap_req(struct rpc_task *task,
1017 rqstp, p, obj); 1107 rqstp, p, obj);
1018 break; 1108 break;
1019 case RPC_GSS_SVC_PRIVACY: 1109 case RPC_GSS_SVC_PRIVACY:
1110 status = gss_wrap_req_priv(cred, ctx, encode,
1111 rqstp, p, obj);
1020 break; 1112 break;
1021 } 1113 }
1022out: 1114out:
@@ -1054,8 +1146,7 @@ gss_unwrap_resp_integ(struct rpc_cred *cred, struct gss_cl_ctx *ctx,
1054 if (xdr_buf_read_netobj(rcv_buf, &mic, mic_offset)) 1146 if (xdr_buf_read_netobj(rcv_buf, &mic, mic_offset))
1055 return status; 1147 return status;
1056 1148
1057 maj_stat = gss_verify_mic(ctx->gc_gss_ctx, &integ_buf, 1149 maj_stat = gss_verify_mic(ctx->gc_gss_ctx, &integ_buf, &mic);
1058 &mic, NULL);
1059 if (maj_stat == GSS_S_CONTEXT_EXPIRED) 1150 if (maj_stat == GSS_S_CONTEXT_EXPIRED)
1060 cred->cr_flags &= ~RPCAUTH_CRED_UPTODATE; 1151 cred->cr_flags &= ~RPCAUTH_CRED_UPTODATE;
1061 if (maj_stat != GSS_S_COMPLETE) 1152 if (maj_stat != GSS_S_COMPLETE)
@@ -1063,6 +1154,35 @@ gss_unwrap_resp_integ(struct rpc_cred *cred, struct gss_cl_ctx *ctx,
1063 return 0; 1154 return 0;
1064} 1155}
1065 1156
1157static inline int
1158gss_unwrap_resp_priv(struct rpc_cred *cred, struct gss_cl_ctx *ctx,
1159 struct rpc_rqst *rqstp, u32 **p)
1160{
1161 struct xdr_buf *rcv_buf = &rqstp->rq_rcv_buf;
1162 u32 offset;
1163 u32 opaque_len;
1164 u32 maj_stat;
1165 int status = -EIO;
1166
1167 opaque_len = ntohl(*(*p)++);
1168 offset = (u8 *)(*p) - (u8 *)rcv_buf->head[0].iov_base;
1169 if (offset + opaque_len > rcv_buf->len)
1170 return status;
1171 /* remove padding: */
1172 rcv_buf->len = offset + opaque_len;
1173
1174 maj_stat = gss_unwrap(ctx->gc_gss_ctx, offset, rcv_buf);
1175 if (maj_stat == GSS_S_CONTEXT_EXPIRED)
1176 cred->cr_flags &= ~RPCAUTH_CRED_UPTODATE;
1177 if (maj_stat != GSS_S_COMPLETE)
1178 return status;
1179 if (ntohl(*(*p)++) != rqstp->rq_seqno)
1180 return status;
1181
1182 return 0;
1183}
1184
1185
1066static int 1186static int
1067gss_unwrap_resp(struct rpc_task *task, 1187gss_unwrap_resp(struct rpc_task *task,
1068 kxdrproc_t decode, void *rqstp, u32 *p, void *obj) 1188 kxdrproc_t decode, void *rqstp, u32 *p, void *obj)
@@ -1071,6 +1191,9 @@ gss_unwrap_resp(struct rpc_task *task,
1071 struct gss_cred *gss_cred = container_of(cred, struct gss_cred, 1191 struct gss_cred *gss_cred = container_of(cred, struct gss_cred,
1072 gc_base); 1192 gc_base);
1073 struct gss_cl_ctx *ctx = gss_cred_get_ctx(cred); 1193 struct gss_cl_ctx *ctx = gss_cred_get_ctx(cred);
1194 u32 *savedp = p;
1195 struct kvec *head = ((struct rpc_rqst *)rqstp)->rq_rcv_buf.head;
1196 int savedlen = head->iov_len;
1074 int status = -EIO; 1197 int status = -EIO;
1075 1198
1076 if (ctx->gc_proc != RPC_GSS_PROC_DATA) 1199 if (ctx->gc_proc != RPC_GSS_PROC_DATA)
@@ -1084,8 +1207,14 @@ gss_unwrap_resp(struct rpc_task *task,
1084 goto out; 1207 goto out;
1085 break; 1208 break;
1086 case RPC_GSS_SVC_PRIVACY: 1209 case RPC_GSS_SVC_PRIVACY:
1210 status = gss_unwrap_resp_priv(cred, ctx, rqstp, &p);
1211 if (status)
1212 goto out;
1087 break; 1213 break;
1088 } 1214 }
1215 /* take into account extra slack for integrity and privacy cases: */
1216 task->tk_auth->au_rslack = task->tk_auth->au_verfsize + (p - savedp)
1217 + (savedlen - head->iov_len);
1089out_decode: 1218out_decode:
1090 status = decode(rqstp, p, obj); 1219 status = decode(rqstp, p, obj);
1091out: 1220out:
diff --git a/net/sunrpc/auth_gss/gss_krb5_crypto.c b/net/sunrpc/auth_gss/gss_krb5_crypto.c
index ee6ae74cd1b2..3f3d5437f02d 100644
--- a/net/sunrpc/auth_gss/gss_krb5_crypto.c
+++ b/net/sunrpc/auth_gss/gss_krb5_crypto.c
@@ -139,17 +139,91 @@ buf_to_sg(struct scatterlist *sg, char *ptr, int len) {
139 sg->length = len; 139 sg->length = len;
140} 140}
141 141
142static int
143process_xdr_buf(struct xdr_buf *buf, int offset, int len,
144 int (*actor)(struct scatterlist *, void *), void *data)
145{
146 int i, page_len, thislen, page_offset, ret = 0;
147 struct scatterlist sg[1];
148
149 if (offset >= buf->head[0].iov_len) {
150 offset -= buf->head[0].iov_len;
151 } else {
152 thislen = buf->head[0].iov_len - offset;
153 if (thislen > len)
154 thislen = len;
155 buf_to_sg(sg, buf->head[0].iov_base + offset, thislen);
156 ret = actor(sg, data);
157 if (ret)
158 goto out;
159 offset = 0;
160 len -= thislen;
161 }
162 if (len == 0)
163 goto out;
164
165 if (offset >= buf->page_len) {
166 offset -= buf->page_len;
167 } else {
168 page_len = buf->page_len - offset;
169 if (page_len > len)
170 page_len = len;
171 len -= page_len;
172 page_offset = (offset + buf->page_base) & (PAGE_CACHE_SIZE - 1);
173 i = (offset + buf->page_base) >> PAGE_CACHE_SHIFT;
174 thislen = PAGE_CACHE_SIZE - page_offset;
175 do {
176 if (thislen > page_len)
177 thislen = page_len;
178 sg->page = buf->pages[i];
179 sg->offset = page_offset;
180 sg->length = thislen;
181 ret = actor(sg, data);
182 if (ret)
183 goto out;
184 page_len -= thislen;
185 i++;
186 page_offset = 0;
187 thislen = PAGE_CACHE_SIZE;
188 } while (page_len != 0);
189 offset = 0;
190 }
191 if (len == 0)
192 goto out;
193
194 if (offset < buf->tail[0].iov_len) {
195 thislen = buf->tail[0].iov_len - offset;
196 if (thislen > len)
197 thislen = len;
198 buf_to_sg(sg, buf->tail[0].iov_base + offset, thislen);
199 ret = actor(sg, data);
200 len -= thislen;
201 }
202 if (len != 0)
203 ret = -EINVAL;
204out:
205 return ret;
206}
207
208static int
209checksummer(struct scatterlist *sg, void *data)
210{
211 struct crypto_tfm *tfm = (struct crypto_tfm *)data;
212
213 crypto_digest_update(tfm, sg, 1);
214
215 return 0;
216}
217
142/* checksum the plaintext data and hdrlen bytes of the token header */ 218/* checksum the plaintext data and hdrlen bytes of the token header */
143s32 219s32
144make_checksum(s32 cksumtype, char *header, int hdrlen, struct xdr_buf *body, 220make_checksum(s32 cksumtype, char *header, int hdrlen, struct xdr_buf *body,
145 struct xdr_netobj *cksum) 221 int body_offset, struct xdr_netobj *cksum)
146{ 222{
147 char *cksumname; 223 char *cksumname;
148 struct crypto_tfm *tfm = NULL; /* XXX add to ctx? */ 224 struct crypto_tfm *tfm = NULL; /* XXX add to ctx? */
149 struct scatterlist sg[1]; 225 struct scatterlist sg[1];
150 u32 code = GSS_S_FAILURE; 226 u32 code = GSS_S_FAILURE;
151 int len, thislen, offset;
152 int i;
153 227
154 switch (cksumtype) { 228 switch (cksumtype) {
155 case CKSUMTYPE_RSA_MD5: 229 case CKSUMTYPE_RSA_MD5:
@@ -169,33 +243,8 @@ make_checksum(s32 cksumtype, char *header, int hdrlen, struct xdr_buf *body,
169 crypto_digest_init(tfm); 243 crypto_digest_init(tfm);
170 buf_to_sg(sg, header, hdrlen); 244 buf_to_sg(sg, header, hdrlen);
171 crypto_digest_update(tfm, sg, 1); 245 crypto_digest_update(tfm, sg, 1);
172 if (body->head[0].iov_len) { 246 process_xdr_buf(body, body_offset, body->len - body_offset,
173 buf_to_sg(sg, body->head[0].iov_base, body->head[0].iov_len); 247 checksummer, tfm);
174 crypto_digest_update(tfm, sg, 1);
175 }
176
177 len = body->page_len;
178 if (len != 0) {
179 offset = body->page_base & (PAGE_CACHE_SIZE - 1);
180 i = body->page_base >> PAGE_CACHE_SHIFT;
181 thislen = PAGE_CACHE_SIZE - offset;
182 do {
183 if (thislen > len)
184 thislen = len;
185 sg->page = body->pages[i];
186 sg->offset = offset;
187 sg->length = thislen;
188 crypto_digest_update(tfm, sg, 1);
189 len -= thislen;
190 i++;
191 offset = 0;
192 thislen = PAGE_CACHE_SIZE;
193 } while(len != 0);
194 }
195 if (body->tail[0].iov_len) {
196 buf_to_sg(sg, body->tail[0].iov_base, body->tail[0].iov_len);
197 crypto_digest_update(tfm, sg, 1);
198 }
199 crypto_digest_final(tfm, cksum->data); 248 crypto_digest_final(tfm, cksum->data);
200 code = 0; 249 code = 0;
201out: 250out:
@@ -204,3 +253,154 @@ out:
204} 253}
205 254
206EXPORT_SYMBOL(make_checksum); 255EXPORT_SYMBOL(make_checksum);
256
257struct encryptor_desc {
258 u8 iv[8]; /* XXX hard-coded blocksize */
259 struct crypto_tfm *tfm;
260 int pos;
261 struct xdr_buf *outbuf;
262 struct page **pages;
263 struct scatterlist infrags[4];
264 struct scatterlist outfrags[4];
265 int fragno;
266 int fraglen;
267};
268
269static int
270encryptor(struct scatterlist *sg, void *data)
271{
272 struct encryptor_desc *desc = data;
273 struct xdr_buf *outbuf = desc->outbuf;
274 struct page *in_page;
275 int thislen = desc->fraglen + sg->length;
276 int fraglen, ret;
277 int page_pos;
278
279 /* Worst case is 4 fragments: head, end of page 1, start
280 * of page 2, tail. Anything more is a bug. */
281 BUG_ON(desc->fragno > 3);
282 desc->infrags[desc->fragno] = *sg;
283 desc->outfrags[desc->fragno] = *sg;
284
285 page_pos = desc->pos - outbuf->head[0].iov_len;
286 if (page_pos >= 0 && page_pos < outbuf->page_len) {
287 /* pages are not in place: */
288 int i = (page_pos + outbuf->page_base) >> PAGE_CACHE_SHIFT;
289 in_page = desc->pages[i];
290 } else {
291 in_page = sg->page;
292 }
293 desc->infrags[desc->fragno].page = in_page;
294 desc->fragno++;
295 desc->fraglen += sg->length;
296 desc->pos += sg->length;
297
298 fraglen = thislen & 7; /* XXX hardcoded blocksize */
299 thislen -= fraglen;
300
301 if (thislen == 0)
302 return 0;
303
304 ret = crypto_cipher_encrypt_iv(desc->tfm, desc->outfrags, desc->infrags,
305 thislen, desc->iv);
306 if (ret)
307 return ret;
308 if (fraglen) {
309 desc->outfrags[0].page = sg->page;
310 desc->outfrags[0].offset = sg->offset + sg->length - fraglen;
311 desc->outfrags[0].length = fraglen;
312 desc->infrags[0] = desc->outfrags[0];
313 desc->infrags[0].page = in_page;
314 desc->fragno = 1;
315 desc->fraglen = fraglen;
316 } else {
317 desc->fragno = 0;
318 desc->fraglen = 0;
319 }
320 return 0;
321}
322
323int
324gss_encrypt_xdr_buf(struct crypto_tfm *tfm, struct xdr_buf *buf, int offset,
325 struct page **pages)
326{
327 int ret;
328 struct encryptor_desc desc;
329
330 BUG_ON((buf->len - offset) % crypto_tfm_alg_blocksize(tfm) != 0);
331
332 memset(desc.iv, 0, sizeof(desc.iv));
333 desc.tfm = tfm;
334 desc.pos = offset;
335 desc.outbuf = buf;
336 desc.pages = pages;
337 desc.fragno = 0;
338 desc.fraglen = 0;
339
340 ret = process_xdr_buf(buf, offset, buf->len - offset, encryptor, &desc);
341 return ret;
342}
343
344EXPORT_SYMBOL(gss_encrypt_xdr_buf);
345
346struct decryptor_desc {
347 u8 iv[8]; /* XXX hard-coded blocksize */
348 struct crypto_tfm *tfm;
349 struct scatterlist frags[4];
350 int fragno;
351 int fraglen;
352};
353
354static int
355decryptor(struct scatterlist *sg, void *data)
356{
357 struct decryptor_desc *desc = data;
358 int thislen = desc->fraglen + sg->length;
359 int fraglen, ret;
360
361 /* Worst case is 4 fragments: head, end of page 1, start
362 * of page 2, tail. Anything more is a bug. */
363 BUG_ON(desc->fragno > 3);
364 desc->frags[desc->fragno] = *sg;
365 desc->fragno++;
366 desc->fraglen += sg->length;
367
368 fraglen = thislen & 7; /* XXX hardcoded blocksize */
369 thislen -= fraglen;
370
371 if (thislen == 0)
372 return 0;
373
374 ret = crypto_cipher_decrypt_iv(desc->tfm, desc->frags, desc->frags,
375 thislen, desc->iv);
376 if (ret)
377 return ret;
378 if (fraglen) {
379 desc->frags[0].page = sg->page;
380 desc->frags[0].offset = sg->offset + sg->length - fraglen;
381 desc->frags[0].length = fraglen;
382 desc->fragno = 1;
383 desc->fraglen = fraglen;
384 } else {
385 desc->fragno = 0;
386 desc->fraglen = 0;
387 }
388 return 0;
389}
390
391int
392gss_decrypt_xdr_buf(struct crypto_tfm *tfm, struct xdr_buf *buf, int offset)
393{
394 struct decryptor_desc desc;
395
396 /* XXXJBF: */
397 BUG_ON((buf->len - offset) % crypto_tfm_alg_blocksize(tfm) != 0);
398
399 memset(desc.iv, 0, sizeof(desc.iv));
400 desc.tfm = tfm;
401 desc.fragno = 0;
402 desc.fraglen = 0;
403 return process_xdr_buf(buf, offset, buf->len - offset, decryptor, &desc);
404}
405
406EXPORT_SYMBOL(gss_decrypt_xdr_buf);
diff --git a/net/sunrpc/auth_gss/gss_krb5_mech.c b/net/sunrpc/auth_gss/gss_krb5_mech.c
index 606a8a82cafb..5f1f806a0b11 100644
--- a/net/sunrpc/auth_gss/gss_krb5_mech.c
+++ b/net/sunrpc/auth_gss/gss_krb5_mech.c
@@ -39,7 +39,6 @@
39#include <linux/types.h> 39#include <linux/types.h>
40#include <linux/slab.h> 40#include <linux/slab.h>
41#include <linux/sunrpc/auth.h> 41#include <linux/sunrpc/auth.h>
42#include <linux/in.h>
43#include <linux/sunrpc/gss_krb5.h> 42#include <linux/sunrpc/gss_krb5.h>
44#include <linux/sunrpc/xdr.h> 43#include <linux/sunrpc/xdr.h>
45#include <linux/crypto.h> 44#include <linux/crypto.h>
@@ -191,43 +190,12 @@ gss_delete_sec_context_kerberos(void *internal_ctx) {
191 kfree(kctx); 190 kfree(kctx);
192} 191}
193 192
194static u32
195gss_verify_mic_kerberos(struct gss_ctx *ctx,
196 struct xdr_buf *message,
197 struct xdr_netobj *mic_token,
198 u32 *qstate) {
199 u32 maj_stat = 0;
200 int qop_state;
201 struct krb5_ctx *kctx = ctx->internal_ctx_id;
202
203 maj_stat = krb5_read_token(kctx, mic_token, message, &qop_state,
204 KG_TOK_MIC_MSG);
205 if (!maj_stat && qop_state)
206 *qstate = qop_state;
207
208 dprintk("RPC: gss_verify_mic_kerberos returning %d\n", maj_stat);
209 return maj_stat;
210}
211
212static u32
213gss_get_mic_kerberos(struct gss_ctx *ctx,
214 u32 qop,
215 struct xdr_buf *message,
216 struct xdr_netobj *mic_token) {
217 u32 err = 0;
218 struct krb5_ctx *kctx = ctx->internal_ctx_id;
219
220 err = krb5_make_token(kctx, qop, message, mic_token, KG_TOK_MIC_MSG);
221
222 dprintk("RPC: gss_get_mic_kerberos returning %d\n",err);
223
224 return err;
225}
226
227static struct gss_api_ops gss_kerberos_ops = { 193static struct gss_api_ops gss_kerberos_ops = {
228 .gss_import_sec_context = gss_import_sec_context_kerberos, 194 .gss_import_sec_context = gss_import_sec_context_kerberos,
229 .gss_get_mic = gss_get_mic_kerberos, 195 .gss_get_mic = gss_get_mic_kerberos,
230 .gss_verify_mic = gss_verify_mic_kerberos, 196 .gss_verify_mic = gss_verify_mic_kerberos,
197 .gss_wrap = gss_wrap_kerberos,
198 .gss_unwrap = gss_unwrap_kerberos,
231 .gss_delete_sec_context = gss_delete_sec_context_kerberos, 199 .gss_delete_sec_context = gss_delete_sec_context_kerberos,
232}; 200};
233 201
@@ -242,6 +210,11 @@ static struct pf_desc gss_kerberos_pfs[] = {
242 .service = RPC_GSS_SVC_INTEGRITY, 210 .service = RPC_GSS_SVC_INTEGRITY,
243 .name = "krb5i", 211 .name = "krb5i",
244 }, 212 },
213 [2] = {
214 .pseudoflavor = RPC_AUTH_GSS_KRB5P,
215 .service = RPC_GSS_SVC_PRIVACY,
216 .name = "krb5p",
217 },
245}; 218};
246 219
247static struct gss_api_mech gss_kerberos_mech = { 220static struct gss_api_mech gss_kerberos_mech = {
diff --git a/net/sunrpc/auth_gss/gss_krb5_seal.c b/net/sunrpc/auth_gss/gss_krb5_seal.c
index afeeb8715a77..13f8ae979454 100644
--- a/net/sunrpc/auth_gss/gss_krb5_seal.c
+++ b/net/sunrpc/auth_gss/gss_krb5_seal.c
@@ -70,22 +70,13 @@
70# define RPCDBG_FACILITY RPCDBG_AUTH 70# define RPCDBG_FACILITY RPCDBG_AUTH
71#endif 71#endif
72 72
73static inline int
74gss_krb5_padding(int blocksize, int length) {
75 /* Most of the code is block-size independent but in practice we
76 * use only 8: */
77 BUG_ON(blocksize != 8);
78 return 8 - (length & 7);
79}
80
81u32 73u32
82krb5_make_token(struct krb5_ctx *ctx, int qop_req, 74gss_get_mic_kerberos(struct gss_ctx *gss_ctx, struct xdr_buf *text,
83 struct xdr_buf *text, struct xdr_netobj *token, 75 struct xdr_netobj *token)
84 int toktype)
85{ 76{
77 struct krb5_ctx *ctx = gss_ctx->internal_ctx_id;
86 s32 checksum_type; 78 s32 checksum_type;
87 struct xdr_netobj md5cksum = {.len = 0, .data = NULL}; 79 struct xdr_netobj md5cksum = {.len = 0, .data = NULL};
88 int blocksize = 0, tmsglen;
89 unsigned char *ptr, *krb5_hdr, *msg_start; 80 unsigned char *ptr, *krb5_hdr, *msg_start;
90 s32 now; 81 s32 now;
91 82
@@ -93,9 +84,6 @@ krb5_make_token(struct krb5_ctx *ctx, int qop_req,
93 84
94 now = get_seconds(); 85 now = get_seconds();
95 86
96 if (qop_req != 0)
97 goto out_err;
98
99 switch (ctx->signalg) { 87 switch (ctx->signalg) {
100 case SGN_ALG_DES_MAC_MD5: 88 case SGN_ALG_DES_MAC_MD5:
101 checksum_type = CKSUMTYPE_RSA_MD5; 89 checksum_type = CKSUMTYPE_RSA_MD5;
@@ -111,21 +99,13 @@ krb5_make_token(struct krb5_ctx *ctx, int qop_req,
111 goto out_err; 99 goto out_err;
112 } 100 }
113 101
114 if (toktype == KG_TOK_WRAP_MSG) { 102 token->len = g_token_size(&ctx->mech_used, 22);
115 blocksize = crypto_tfm_alg_blocksize(ctx->enc);
116 tmsglen = blocksize + text->len
117 + gss_krb5_padding(blocksize, blocksize + text->len);
118 } else {
119 tmsglen = 0;
120 }
121
122 token->len = g_token_size(&ctx->mech_used, 22 + tmsglen);
123 103
124 ptr = token->data; 104 ptr = token->data;
125 g_make_token_header(&ctx->mech_used, 22 + tmsglen, &ptr); 105 g_make_token_header(&ctx->mech_used, 22, &ptr);
126 106
127 *ptr++ = (unsigned char) ((toktype>>8)&0xff); 107 *ptr++ = (unsigned char) ((KG_TOK_MIC_MSG>>8)&0xff);
128 *ptr++ = (unsigned char) (toktype&0xff); 108 *ptr++ = (unsigned char) (KG_TOK_MIC_MSG&0xff);
129 109
130 /* ptr now at byte 2 of header described in rfc 1964, section 1.2.1: */ 110 /* ptr now at byte 2 of header described in rfc 1964, section 1.2.1: */
131 krb5_hdr = ptr - 2; 111 krb5_hdr = ptr - 2;
@@ -133,17 +113,9 @@ krb5_make_token(struct krb5_ctx *ctx, int qop_req,
133 113
134 *(u16 *)(krb5_hdr + 2) = htons(ctx->signalg); 114 *(u16 *)(krb5_hdr + 2) = htons(ctx->signalg);
135 memset(krb5_hdr + 4, 0xff, 4); 115 memset(krb5_hdr + 4, 0xff, 4);
136 if (toktype == KG_TOK_WRAP_MSG)
137 *(u16 *)(krb5_hdr + 4) = htons(ctx->sealalg);
138 116
139 if (toktype == KG_TOK_WRAP_MSG) { 117 if (make_checksum(checksum_type, krb5_hdr, 8, text, 0, &md5cksum))
140 /* XXX removing support for now */
141 goto out_err;
142 } else { /* Sign only. */
143 if (make_checksum(checksum_type, krb5_hdr, 8, text,
144 &md5cksum))
145 goto out_err; 118 goto out_err;
146 }
147 119
148 switch (ctx->signalg) { 120 switch (ctx->signalg) {
149 case SGN_ALG_DES_MAC_MD5: 121 case SGN_ALG_DES_MAC_MD5:
diff --git a/net/sunrpc/auth_gss/gss_krb5_unseal.c b/net/sunrpc/auth_gss/gss_krb5_unseal.c
index 8767fc53183d..2030475d98ed 100644
--- a/net/sunrpc/auth_gss/gss_krb5_unseal.c
+++ b/net/sunrpc/auth_gss/gss_krb5_unseal.c
@@ -68,21 +68,14 @@
68#endif 68#endif
69 69
70 70
71/* message_buffer is an input if toktype is MIC and an output if it is WRAP: 71/* read_token is a mic token, and message_buffer is the data that the mic was
72 * If toktype is MIC: read_token is a mic token, and message_buffer is the 72 * supposedly taken over. */
73 * data that the mic was supposedly taken over.
74 * If toktype is WRAP: read_token is a wrap token, and message_buffer is used
75 * to return the decrypted data.
76 */
77 73
78/* XXX will need to change prototype and/or just split into a separate function
79 * when we add privacy (because read_token will be in pages too). */
80u32 74u32
81krb5_read_token(struct krb5_ctx *ctx, 75gss_verify_mic_kerberos(struct gss_ctx *gss_ctx,
82 struct xdr_netobj *read_token, 76 struct xdr_buf *message_buffer, struct xdr_netobj *read_token)
83 struct xdr_buf *message_buffer,
84 int *qop_state, int toktype)
85{ 77{
78 struct krb5_ctx *ctx = gss_ctx->internal_ctx_id;
86 int signalg; 79 int signalg;
87 int sealalg; 80 int sealalg;
88 s32 checksum_type; 81 s32 checksum_type;
@@ -100,16 +93,12 @@ krb5_read_token(struct krb5_ctx *ctx,
100 read_token->len)) 93 read_token->len))
101 goto out; 94 goto out;
102 95
103 if ((*ptr++ != ((toktype>>8)&0xff)) || (*ptr++ != (toktype&0xff))) 96 if ((*ptr++ != ((KG_TOK_MIC_MSG>>8)&0xff)) ||
97 (*ptr++ != ( KG_TOK_MIC_MSG &0xff)) )
104 goto out; 98 goto out;
105 99
106 /* XXX sanity-check bodysize?? */ 100 /* XXX sanity-check bodysize?? */
107 101
108 if (toktype == KG_TOK_WRAP_MSG) {
109 /* XXX gone */
110 goto out;
111 }
112
113 /* get the sign and seal algorithms */ 102 /* get the sign and seal algorithms */
114 103
115 signalg = ptr[0] + (ptr[1] << 8); 104 signalg = ptr[0] + (ptr[1] << 8);
@@ -120,14 +109,7 @@ krb5_read_token(struct krb5_ctx *ctx,
120 if ((ptr[4] != 0xff) || (ptr[5] != 0xff)) 109 if ((ptr[4] != 0xff) || (ptr[5] != 0xff))
121 goto out; 110 goto out;
122 111
123 if (((toktype != KG_TOK_WRAP_MSG) && (sealalg != 0xffff)) || 112 if (sealalg != 0xffff)
124 ((toktype == KG_TOK_WRAP_MSG) && (sealalg == 0xffff)))
125 goto out;
126
127 /* in the current spec, there is only one valid seal algorithm per
128 key type, so a simple comparison is ok */
129
130 if ((toktype == KG_TOK_WRAP_MSG) && !(sealalg == ctx->sealalg))
131 goto out; 113 goto out;
132 114
133 /* there are several mappings of seal algorithms to sign algorithms, 115 /* there are several mappings of seal algorithms to sign algorithms,
@@ -154,7 +136,7 @@ krb5_read_token(struct krb5_ctx *ctx,
154 switch (signalg) { 136 switch (signalg) {
155 case SGN_ALG_DES_MAC_MD5: 137 case SGN_ALG_DES_MAC_MD5:
156 ret = make_checksum(checksum_type, ptr - 2, 8, 138 ret = make_checksum(checksum_type, ptr - 2, 8,
157 message_buffer, &md5cksum); 139 message_buffer, 0, &md5cksum);
158 if (ret) 140 if (ret)
159 goto out; 141 goto out;
160 142
@@ -175,9 +157,6 @@ krb5_read_token(struct krb5_ctx *ctx,
175 157
176 /* it got through unscathed. Make sure the context is unexpired */ 158 /* it got through unscathed. Make sure the context is unexpired */
177 159
178 if (qop_state)
179 *qop_state = GSS_C_QOP_DEFAULT;
180
181 now = get_seconds(); 160 now = get_seconds();
182 161
183 ret = GSS_S_CONTEXT_EXPIRED; 162 ret = GSS_S_CONTEXT_EXPIRED;
diff --git a/net/sunrpc/auth_gss/gss_krb5_wrap.c b/net/sunrpc/auth_gss/gss_krb5_wrap.c
new file mode 100644
index 000000000000..af777cf9f251
--- /dev/null
+++ b/net/sunrpc/auth_gss/gss_krb5_wrap.c
@@ -0,0 +1,363 @@
1#include <linux/types.h>
2#include <linux/slab.h>
3#include <linux/jiffies.h>
4#include <linux/sunrpc/gss_krb5.h>
5#include <linux/random.h>
6#include <linux/pagemap.h>
7#include <asm/scatterlist.h>
8#include <linux/crypto.h>
9
10#ifdef RPC_DEBUG
11# define RPCDBG_FACILITY RPCDBG_AUTH
12#endif
13
14static inline int
15gss_krb5_padding(int blocksize, int length)
16{
17 /* Most of the code is block-size independent but currently we
18 * use only 8: */
19 BUG_ON(blocksize != 8);
20 return 8 - (length & 7);
21}
22
23static inline void
24gss_krb5_add_padding(struct xdr_buf *buf, int offset, int blocksize)
25{
26 int padding = gss_krb5_padding(blocksize, buf->len - offset);
27 char *p;
28 struct kvec *iov;
29
30 if (buf->page_len || buf->tail[0].iov_len)
31 iov = &buf->tail[0];
32 else
33 iov = &buf->head[0];
34 p = iov->iov_base + iov->iov_len;
35 iov->iov_len += padding;
36 buf->len += padding;
37 memset(p, padding, padding);
38}
39
40static inline int
41gss_krb5_remove_padding(struct xdr_buf *buf, int blocksize)
42{
43 u8 *ptr;
44 u8 pad;
45 int len = buf->len;
46
47 if (len <= buf->head[0].iov_len) {
48 pad = *(u8 *)(buf->head[0].iov_base + len - 1);
49 if (pad > buf->head[0].iov_len)
50 return -EINVAL;
51 buf->head[0].iov_len -= pad;
52 goto out;
53 } else
54 len -= buf->head[0].iov_len;
55 if (len <= buf->page_len) {
56 int last = (buf->page_base + len - 1)
57 >>PAGE_CACHE_SHIFT;
58 int offset = (buf->page_base + len - 1)
59 & (PAGE_CACHE_SIZE - 1);
60 ptr = kmap_atomic(buf->pages[last], KM_SKB_SUNRPC_DATA);
61 pad = *(ptr + offset);
62 kunmap_atomic(ptr, KM_SKB_SUNRPC_DATA);
63 goto out;
64 } else
65 len -= buf->page_len;
66 BUG_ON(len > buf->tail[0].iov_len);
67 pad = *(u8 *)(buf->tail[0].iov_base + len - 1);
68out:
69 /* XXX: NOTE: we do not adjust the page lengths--they represent
70 * a range of data in the real filesystem page cache, and we need
71 * to know that range so the xdr code can properly place read data.
72 * However adjusting the head length, as we do above, is harmless.
73 * In the case of a request that fits into a single page, the server
74 * also uses length and head length together to determine the original
75 * start of the request to copy the request for deferal; so it's
76 * easier on the server if we adjust head and tail length in tandem.
77 * It's not really a problem that we don't fool with the page and
78 * tail lengths, though--at worst badly formed xdr might lead the
79 * server to attempt to parse the padding.
80 * XXX: Document all these weird requirements for gss mechanism
81 * wrap/unwrap functions. */
82 if (pad > blocksize)
83 return -EINVAL;
84 if (buf->len > pad)
85 buf->len -= pad;
86 else
87 return -EINVAL;
88 return 0;
89}
90
91static inline void
92make_confounder(char *p, int blocksize)
93{
94 static u64 i = 0;
95 u64 *q = (u64 *)p;
96
97 /* rfc1964 claims this should be "random". But all that's really
98 * necessary is that it be unique. And not even that is necessary in
99 * our case since our "gssapi" implementation exists only to support
100 * rpcsec_gss, so we know that the only buffers we will ever encrypt
101 * already begin with a unique sequence number. Just to hedge my bets
102 * I'll make a half-hearted attempt at something unique, but ensuring
103 * uniqueness would mean worrying about atomicity and rollover, and I
104 * don't care enough. */
105
106 BUG_ON(blocksize != 8);
107 *q = i++;
108}
109
110/* Assumptions: the head and tail of inbuf are ours to play with.
111 * The pages, however, may be real pages in the page cache and we replace
112 * them with scratch pages from **pages before writing to them. */
113/* XXX: obviously the above should be documentation of wrap interface,
114 * and shouldn't be in this kerberos-specific file. */
115
116/* XXX factor out common code with seal/unseal. */
117
118u32
119gss_wrap_kerberos(struct gss_ctx *ctx, int offset,
120 struct xdr_buf *buf, struct page **pages)
121{
122 struct krb5_ctx *kctx = ctx->internal_ctx_id;
123 s32 checksum_type;
124 struct xdr_netobj md5cksum = {.len = 0, .data = NULL};
125 int blocksize = 0, plainlen;
126 unsigned char *ptr, *krb5_hdr, *msg_start;
127 s32 now;
128 int headlen;
129 struct page **tmp_pages;
130
131 dprintk("RPC: gss_wrap_kerberos\n");
132
133 now = get_seconds();
134
135 switch (kctx->signalg) {
136 case SGN_ALG_DES_MAC_MD5:
137 checksum_type = CKSUMTYPE_RSA_MD5;
138 break;
139 default:
140 dprintk("RPC: gss_krb5_seal: kctx->signalg %d not"
141 " supported\n", kctx->signalg);
142 goto out_err;
143 }
144 if (kctx->sealalg != SEAL_ALG_NONE && kctx->sealalg != SEAL_ALG_DES) {
145 dprintk("RPC: gss_krb5_seal: kctx->sealalg %d not supported\n",
146 kctx->sealalg);
147 goto out_err;
148 }
149
150 blocksize = crypto_tfm_alg_blocksize(kctx->enc);
151 gss_krb5_add_padding(buf, offset, blocksize);
152 BUG_ON((buf->len - offset) % blocksize);
153 plainlen = blocksize + buf->len - offset;
154
155 headlen = g_token_size(&kctx->mech_used, 22 + plainlen) -
156 (buf->len - offset);
157
158 ptr = buf->head[0].iov_base + offset;
159 /* shift data to make room for header. */
160 /* XXX Would be cleverer to encrypt while copying. */
161 /* XXX bounds checking, slack, etc. */
162 memmove(ptr + headlen, ptr, buf->head[0].iov_len - offset);
163 buf->head[0].iov_len += headlen;
164 buf->len += headlen;
165 BUG_ON((buf->len - offset - headlen) % blocksize);
166
167 g_make_token_header(&kctx->mech_used, 22 + plainlen, &ptr);
168
169
170 *ptr++ = (unsigned char) ((KG_TOK_WRAP_MSG>>8)&0xff);
171 *ptr++ = (unsigned char) (KG_TOK_WRAP_MSG&0xff);
172
173 /* ptr now at byte 2 of header described in rfc 1964, section 1.2.1: */
174 krb5_hdr = ptr - 2;
175 msg_start = krb5_hdr + 24;
176 /* XXXJBF: */ BUG_ON(buf->head[0].iov_base + offset + headlen != msg_start + blocksize);
177
178 *(u16 *)(krb5_hdr + 2) = htons(kctx->signalg);
179 memset(krb5_hdr + 4, 0xff, 4);
180 *(u16 *)(krb5_hdr + 4) = htons(kctx->sealalg);
181
182 make_confounder(msg_start, blocksize);
183
184 /* XXXJBF: UGH!: */
185 tmp_pages = buf->pages;
186 buf->pages = pages;
187 if (make_checksum(checksum_type, krb5_hdr, 8, buf,
188 offset + headlen - blocksize, &md5cksum))
189 goto out_err;
190 buf->pages = tmp_pages;
191
192 switch (kctx->signalg) {
193 case SGN_ALG_DES_MAC_MD5:
194 if (krb5_encrypt(kctx->seq, NULL, md5cksum.data,
195 md5cksum.data, md5cksum.len))
196 goto out_err;
197 memcpy(krb5_hdr + 16,
198 md5cksum.data + md5cksum.len - KRB5_CKSUM_LENGTH,
199 KRB5_CKSUM_LENGTH);
200
201 dprintk("RPC: make_seal_token: cksum data: \n");
202 print_hexl((u32 *) (krb5_hdr + 16), KRB5_CKSUM_LENGTH, 0);
203 break;
204 default:
205 BUG();
206 }
207
208 kfree(md5cksum.data);
209
210 /* XXX would probably be more efficient to compute checksum
211 * and encrypt at the same time: */
212 if ((krb5_make_seq_num(kctx->seq, kctx->initiate ? 0 : 0xff,
213 kctx->seq_send, krb5_hdr + 16, krb5_hdr + 8)))
214 goto out_err;
215
216 if (gss_encrypt_xdr_buf(kctx->enc, buf, offset + headlen - blocksize,
217 pages))
218 goto out_err;
219
220 kctx->seq_send++;
221
222 return ((kctx->endtime < now) ? GSS_S_CONTEXT_EXPIRED : GSS_S_COMPLETE);
223out_err:
224 if (md5cksum.data) kfree(md5cksum.data);
225 return GSS_S_FAILURE;
226}
227
228u32
229gss_unwrap_kerberos(struct gss_ctx *ctx, int offset, struct xdr_buf *buf)
230{
231 struct krb5_ctx *kctx = ctx->internal_ctx_id;
232 int signalg;
233 int sealalg;
234 s32 checksum_type;
235 struct xdr_netobj md5cksum = {.len = 0, .data = NULL};
236 s32 now;
237 int direction;
238 s32 seqnum;
239 unsigned char *ptr;
240 int bodysize;
241 u32 ret = GSS_S_DEFECTIVE_TOKEN;
242 void *data_start, *orig_start;
243 int data_len;
244 int blocksize;
245
246 dprintk("RPC: gss_unwrap_kerberos\n");
247
248 ptr = (u8 *)buf->head[0].iov_base + offset;
249 if (g_verify_token_header(&kctx->mech_used, &bodysize, &ptr,
250 buf->len - offset))
251 goto out;
252
253 if ((*ptr++ != ((KG_TOK_WRAP_MSG>>8)&0xff)) ||
254 (*ptr++ != (KG_TOK_WRAP_MSG &0xff)) )
255 goto out;
256
257 /* XXX sanity-check bodysize?? */
258
259 /* get the sign and seal algorithms */
260
261 signalg = ptr[0] + (ptr[1] << 8);
262 sealalg = ptr[2] + (ptr[3] << 8);
263
264 /* Sanity checks */
265
266 if ((ptr[4] != 0xff) || (ptr[5] != 0xff))
267 goto out;
268
269 if (sealalg == 0xffff)
270 goto out;
271
272 /* in the current spec, there is only one valid seal algorithm per
273 key type, so a simple comparison is ok */
274
275 if (sealalg != kctx->sealalg)
276 goto out;
277
278 /* there are several mappings of seal algorithms to sign algorithms,
279 but few enough that we can try them all. */
280
281 if ((kctx->sealalg == SEAL_ALG_NONE && signalg > 1) ||
282 (kctx->sealalg == SEAL_ALG_1 && signalg != SGN_ALG_3) ||
283 (kctx->sealalg == SEAL_ALG_DES3KD &&
284 signalg != SGN_ALG_HMAC_SHA1_DES3_KD))
285 goto out;
286
287 if (gss_decrypt_xdr_buf(kctx->enc, buf,
288 ptr + 22 - (unsigned char *)buf->head[0].iov_base))
289 goto out;
290
291 /* compute the checksum of the message */
292
293 /* initialize the the cksum */
294 switch (signalg) {
295 case SGN_ALG_DES_MAC_MD5:
296 checksum_type = CKSUMTYPE_RSA_MD5;
297 break;
298 default:
299 ret = GSS_S_DEFECTIVE_TOKEN;
300 goto out;
301 }
302
303 switch (signalg) {
304 case SGN_ALG_DES_MAC_MD5:
305 ret = make_checksum(checksum_type, ptr - 2, 8, buf,
306 ptr + 22 - (unsigned char *)buf->head[0].iov_base, &md5cksum);
307 if (ret)
308 goto out;
309
310 ret = krb5_encrypt(kctx->seq, NULL, md5cksum.data,
311 md5cksum.data, md5cksum.len);
312 if (ret)
313 goto out;
314
315 if (memcmp(md5cksum.data + 8, ptr + 14, 8)) {
316 ret = GSS_S_BAD_SIG;
317 goto out;
318 }
319 break;
320 default:
321 ret = GSS_S_DEFECTIVE_TOKEN;
322 goto out;
323 }
324
325 /* it got through unscathed. Make sure the context is unexpired */
326
327 now = get_seconds();
328
329 ret = GSS_S_CONTEXT_EXPIRED;
330 if (now > kctx->endtime)
331 goto out;
332
333 /* do sequencing checks */
334
335 ret = GSS_S_BAD_SIG;
336 if ((ret = krb5_get_seq_num(kctx->seq, ptr + 14, ptr + 6, &direction,
337 &seqnum)))
338 goto out;
339
340 if ((kctx->initiate && direction != 0xff) ||
341 (!kctx->initiate && direction != 0))
342 goto out;
343
344 /* Copy the data back to the right position. XXX: Would probably be
345 * better to copy and encrypt at the same time. */
346
347 blocksize = crypto_tfm_alg_blocksize(kctx->enc);
348 data_start = ptr + 22 + blocksize;
349 orig_start = buf->head[0].iov_base + offset;
350 data_len = (buf->head[0].iov_base + buf->head[0].iov_len) - data_start;
351 memmove(orig_start, data_start, data_len);
352 buf->head[0].iov_len -= (data_start - orig_start);
353 buf->len -= (data_start - orig_start);
354
355 ret = GSS_S_DEFECTIVE_TOKEN;
356 if (gss_krb5_remove_padding(buf, blocksize))
357 goto out;
358
359 ret = GSS_S_COMPLETE;
360out:
361 if (md5cksum.data) kfree(md5cksum.data);
362 return ret;
363}
diff --git a/net/sunrpc/auth_gss/gss_mech_switch.c b/net/sunrpc/auth_gss/gss_mech_switch.c
index 9dfb68377d69..b048bf672da2 100644
--- a/net/sunrpc/auth_gss/gss_mech_switch.c
+++ b/net/sunrpc/auth_gss/gss_mech_switch.c
@@ -35,7 +35,6 @@
35 35
36#include <linux/types.h> 36#include <linux/types.h>
37#include <linux/slab.h> 37#include <linux/slab.h>
38#include <linux/socket.h>
39#include <linux/module.h> 38#include <linux/module.h>
40#include <linux/sunrpc/msg_prot.h> 39#include <linux/sunrpc/msg_prot.h>
41#include <linux/sunrpc/gss_asn1.h> 40#include <linux/sunrpc/gss_asn1.h>
@@ -251,13 +250,11 @@ gss_import_sec_context(const void *input_token, size_t bufsize,
251 250
252u32 251u32
253gss_get_mic(struct gss_ctx *context_handle, 252gss_get_mic(struct gss_ctx *context_handle,
254 u32 qop,
255 struct xdr_buf *message, 253 struct xdr_buf *message,
256 struct xdr_netobj *mic_token) 254 struct xdr_netobj *mic_token)
257{ 255{
258 return context_handle->mech_type->gm_ops 256 return context_handle->mech_type->gm_ops
259 ->gss_get_mic(context_handle, 257 ->gss_get_mic(context_handle,
260 qop,
261 message, 258 message,
262 mic_token); 259 mic_token);
263} 260}
@@ -267,16 +264,34 @@ gss_get_mic(struct gss_ctx *context_handle,
267u32 264u32
268gss_verify_mic(struct gss_ctx *context_handle, 265gss_verify_mic(struct gss_ctx *context_handle,
269 struct xdr_buf *message, 266 struct xdr_buf *message,
270 struct xdr_netobj *mic_token, 267 struct xdr_netobj *mic_token)
271 u32 *qstate)
272{ 268{
273 return context_handle->mech_type->gm_ops 269 return context_handle->mech_type->gm_ops
274 ->gss_verify_mic(context_handle, 270 ->gss_verify_mic(context_handle,
275 message, 271 message,
276 mic_token, 272 mic_token);
277 qstate);
278} 273}
279 274
275u32
276gss_wrap(struct gss_ctx *ctx_id,
277 int offset,
278 struct xdr_buf *buf,
279 struct page **inpages)
280{
281 return ctx_id->mech_type->gm_ops
282 ->gss_wrap(ctx_id, offset, buf, inpages);
283}
284
285u32
286gss_unwrap(struct gss_ctx *ctx_id,
287 int offset,
288 struct xdr_buf *buf)
289{
290 return ctx_id->mech_type->gm_ops
291 ->gss_unwrap(ctx_id, offset, buf);
292}
293
294
280/* gss_delete_sec_context: free all resources associated with context_handle. 295/* gss_delete_sec_context: free all resources associated with context_handle.
281 * Note this differs from the RFC 2744-specified prototype in that we don't 296 * Note this differs from the RFC 2744-specified prototype in that we don't
282 * bother returning an output token, since it would never be used anyway. */ 297 * bother returning an output token, since it would never be used anyway. */
diff --git a/net/sunrpc/auth_gss/gss_spkm3_mech.c b/net/sunrpc/auth_gss/gss_spkm3_mech.c
index 6c97d61baa9b..39b3edc14694 100644
--- a/net/sunrpc/auth_gss/gss_spkm3_mech.c
+++ b/net/sunrpc/auth_gss/gss_spkm3_mech.c
@@ -224,18 +224,13 @@ gss_delete_sec_context_spkm3(void *internal_ctx) {
224static u32 224static u32
225gss_verify_mic_spkm3(struct gss_ctx *ctx, 225gss_verify_mic_spkm3(struct gss_ctx *ctx,
226 struct xdr_buf *signbuf, 226 struct xdr_buf *signbuf,
227 struct xdr_netobj *checksum, 227 struct xdr_netobj *checksum)
228 u32 *qstate) { 228{
229 u32 maj_stat = 0; 229 u32 maj_stat = 0;
230 int qop_state = 0;
231 struct spkm3_ctx *sctx = ctx->internal_ctx_id; 230 struct spkm3_ctx *sctx = ctx->internal_ctx_id;
232 231
233 dprintk("RPC: gss_verify_mic_spkm3 calling spkm3_read_token\n"); 232 dprintk("RPC: gss_verify_mic_spkm3 calling spkm3_read_token\n");
234 maj_stat = spkm3_read_token(sctx, checksum, signbuf, &qop_state, 233 maj_stat = spkm3_read_token(sctx, checksum, signbuf, SPKM_MIC_TOK);
235 SPKM_MIC_TOK);
236
237 if (!maj_stat && qop_state)
238 *qstate = qop_state;
239 234
240 dprintk("RPC: gss_verify_mic_spkm3 returning %d\n", maj_stat); 235 dprintk("RPC: gss_verify_mic_spkm3 returning %d\n", maj_stat);
241 return maj_stat; 236 return maj_stat;
@@ -243,15 +238,15 @@ gss_verify_mic_spkm3(struct gss_ctx *ctx,
243 238
244static u32 239static u32
245gss_get_mic_spkm3(struct gss_ctx *ctx, 240gss_get_mic_spkm3(struct gss_ctx *ctx,
246 u32 qop,
247 struct xdr_buf *message_buffer, 241 struct xdr_buf *message_buffer,
248 struct xdr_netobj *message_token) { 242 struct xdr_netobj *message_token)
243{
249 u32 err = 0; 244 u32 err = 0;
250 struct spkm3_ctx *sctx = ctx->internal_ctx_id; 245 struct spkm3_ctx *sctx = ctx->internal_ctx_id;
251 246
252 dprintk("RPC: gss_get_mic_spkm3\n"); 247 dprintk("RPC: gss_get_mic_spkm3\n");
253 248
254 err = spkm3_make_token(sctx, qop, message_buffer, 249 err = spkm3_make_token(sctx, message_buffer,
255 message_token, SPKM_MIC_TOK); 250 message_token, SPKM_MIC_TOK);
256 return err; 251 return err;
257} 252}
@@ -264,8 +259,8 @@ static struct gss_api_ops gss_spkm3_ops = {
264}; 259};
265 260
266static struct pf_desc gss_spkm3_pfs[] = { 261static struct pf_desc gss_spkm3_pfs[] = {
267 {RPC_AUTH_GSS_SPKM, 0, RPC_GSS_SVC_NONE, "spkm3"}, 262 {RPC_AUTH_GSS_SPKM, RPC_GSS_SVC_NONE, "spkm3"},
268 {RPC_AUTH_GSS_SPKMI, 0, RPC_GSS_SVC_INTEGRITY, "spkm3i"}, 263 {RPC_AUTH_GSS_SPKMI, RPC_GSS_SVC_INTEGRITY, "spkm3i"},
269}; 264};
270 265
271static struct gss_api_mech gss_spkm3_mech = { 266static struct gss_api_mech gss_spkm3_mech = {
diff --git a/net/sunrpc/auth_gss/gss_spkm3_seal.c b/net/sunrpc/auth_gss/gss_spkm3_seal.c
index 25339868d462..148201e929d0 100644
--- a/net/sunrpc/auth_gss/gss_spkm3_seal.c
+++ b/net/sunrpc/auth_gss/gss_spkm3_seal.c
@@ -51,7 +51,7 @@
51 */ 51 */
52 52
53u32 53u32
54spkm3_make_token(struct spkm3_ctx *ctx, int qop_req, 54spkm3_make_token(struct spkm3_ctx *ctx,
55 struct xdr_buf * text, struct xdr_netobj * token, 55 struct xdr_buf * text, struct xdr_netobj * token,
56 int toktype) 56 int toktype)
57{ 57{
@@ -68,8 +68,6 @@ spkm3_make_token(struct spkm3_ctx *ctx, int qop_req,
68 dprintk("RPC: spkm3_make_token\n"); 68 dprintk("RPC: spkm3_make_token\n");
69 69
70 now = jiffies; 70 now = jiffies;
71 if (qop_req != 0)
72 goto out_err;
73 71
74 if (ctx->ctx_id.len != 16) { 72 if (ctx->ctx_id.len != 16) {
75 dprintk("RPC: spkm3_make_token BAD ctx_id.len %d\n", 73 dprintk("RPC: spkm3_make_token BAD ctx_id.len %d\n",
diff --git a/net/sunrpc/auth_gss/gss_spkm3_unseal.c b/net/sunrpc/auth_gss/gss_spkm3_unseal.c
index 65ce81bf0bc4..c3c0d9586103 100644
--- a/net/sunrpc/auth_gss/gss_spkm3_unseal.c
+++ b/net/sunrpc/auth_gss/gss_spkm3_unseal.c
@@ -52,7 +52,7 @@ u32
52spkm3_read_token(struct spkm3_ctx *ctx, 52spkm3_read_token(struct spkm3_ctx *ctx,
53 struct xdr_netobj *read_token, /* checksum */ 53 struct xdr_netobj *read_token, /* checksum */
54 struct xdr_buf *message_buffer, /* signbuf */ 54 struct xdr_buf *message_buffer, /* signbuf */
55 int *qop_state, int toktype) 55 int toktype)
56{ 56{
57 s32 code; 57 s32 code;
58 struct xdr_netobj wire_cksum = {.len =0, .data = NULL}; 58 struct xdr_netobj wire_cksum = {.len =0, .data = NULL};
diff --git a/net/sunrpc/auth_gss/svcauth_gss.c b/net/sunrpc/auth_gss/svcauth_gss.c
index e3308195374e..e4ada15ed856 100644
--- a/net/sunrpc/auth_gss/svcauth_gss.c
+++ b/net/sunrpc/auth_gss/svcauth_gss.c
@@ -566,8 +566,7 @@ gss_verify_header(struct svc_rqst *rqstp, struct rsc *rsci,
566 566
567 if (rqstp->rq_deferred) /* skip verification of revisited request */ 567 if (rqstp->rq_deferred) /* skip verification of revisited request */
568 return SVC_OK; 568 return SVC_OK;
569 if (gss_verify_mic(ctx_id, &rpchdr, &checksum, NULL) 569 if (gss_verify_mic(ctx_id, &rpchdr, &checksum) != GSS_S_COMPLETE) {
570 != GSS_S_COMPLETE) {
571 *authp = rpcsec_gsserr_credproblem; 570 *authp = rpcsec_gsserr_credproblem;
572 return SVC_DENIED; 571 return SVC_DENIED;
573 } 572 }
@@ -604,7 +603,7 @@ gss_write_verf(struct svc_rqst *rqstp, struct gss_ctx *ctx_id, u32 seq)
604 xdr_buf_from_iov(&iov, &verf_data); 603 xdr_buf_from_iov(&iov, &verf_data);
605 p = rqstp->rq_res.head->iov_base + rqstp->rq_res.head->iov_len; 604 p = rqstp->rq_res.head->iov_base + rqstp->rq_res.head->iov_len;
606 mic.data = (u8 *)(p + 1); 605 mic.data = (u8 *)(p + 1);
607 maj_stat = gss_get_mic(ctx_id, 0, &verf_data, &mic); 606 maj_stat = gss_get_mic(ctx_id, &verf_data, &mic);
608 if (maj_stat != GSS_S_COMPLETE) 607 if (maj_stat != GSS_S_COMPLETE)
609 return -1; 608 return -1;
610 *p++ = htonl(mic.len); 609 *p++ = htonl(mic.len);
@@ -710,7 +709,7 @@ unwrap_integ_data(struct xdr_buf *buf, u32 seq, struct gss_ctx *ctx)
710 goto out; 709 goto out;
711 if (read_bytes_from_xdr_buf(buf, integ_len + 4, mic.data, mic.len)) 710 if (read_bytes_from_xdr_buf(buf, integ_len + 4, mic.data, mic.len))
712 goto out; 711 goto out;
713 maj_stat = gss_verify_mic(ctx, &integ_buf, &mic, NULL); 712 maj_stat = gss_verify_mic(ctx, &integ_buf, &mic);
714 if (maj_stat != GSS_S_COMPLETE) 713 if (maj_stat != GSS_S_COMPLETE)
715 goto out; 714 goto out;
716 if (ntohl(svc_getu32(&buf->head[0])) != seq) 715 if (ntohl(svc_getu32(&buf->head[0])) != seq)
@@ -1012,7 +1011,7 @@ svcauth_gss_release(struct svc_rqst *rqstp)
1012 resv = &resbuf->tail[0]; 1011 resv = &resbuf->tail[0];
1013 } 1012 }
1014 mic.data = (u8 *)resv->iov_base + resv->iov_len + 4; 1013 mic.data = (u8 *)resv->iov_base + resv->iov_len + 4;
1015 if (gss_get_mic(gsd->rsci->mechctx, 0, &integ_buf, &mic)) 1014 if (gss_get_mic(gsd->rsci->mechctx, &integ_buf, &mic))
1016 goto out_err; 1015 goto out_err;
1017 svc_putu32(resv, htonl(mic.len)); 1016 svc_putu32(resv, htonl(mic.len));
1018 memset(mic.data + mic.len, 0, 1017 memset(mic.data + mic.len, 0,
diff --git a/net/sunrpc/auth_null.c b/net/sunrpc/auth_null.c
index 9b72d3abf823..f56767aaa927 100644
--- a/net/sunrpc/auth_null.c
+++ b/net/sunrpc/auth_null.c
@@ -7,9 +7,7 @@
7 */ 7 */
8 8
9#include <linux/types.h> 9#include <linux/types.h>
10#include <linux/socket.h>
11#include <linux/module.h> 10#include <linux/module.h>
12#include <linux/in.h>
13#include <linux/utsname.h> 11#include <linux/utsname.h>
14#include <linux/sunrpc/clnt.h> 12#include <linux/sunrpc/clnt.h>
15#include <linux/sched.h> 13#include <linux/sched.h>
diff --git a/net/sunrpc/auth_unix.c b/net/sunrpc/auth_unix.c
index 4ff297a9b15b..890fb5ea0dcb 100644
--- a/net/sunrpc/auth_unix.c
+++ b/net/sunrpc/auth_unix.c
@@ -9,8 +9,6 @@
9#include <linux/types.h> 9#include <linux/types.h>
10#include <linux/sched.h> 10#include <linux/sched.h>
11#include <linux/module.h> 11#include <linux/module.h>
12#include <linux/socket.h>
13#include <linux/in.h>
14#include <linux/sunrpc/clnt.h> 12#include <linux/sunrpc/clnt.h>
15#include <linux/sunrpc/auth.h> 13#include <linux/sunrpc/auth.h>
16 14
diff --git a/net/sunrpc/clnt.c b/net/sunrpc/clnt.c
index f17e6153b688..702ede309b06 100644
--- a/net/sunrpc/clnt.c
+++ b/net/sunrpc/clnt.c
@@ -1,5 +1,5 @@
1/* 1/*
2 * linux/net/sunrpc/rpcclnt.c 2 * linux/net/sunrpc/clnt.c
3 * 3 *
4 * This file contains the high-level RPC interface. 4 * This file contains the high-level RPC interface.
5 * It is modeled as a finite state machine to support both synchronous 5 * It is modeled as a finite state machine to support both synchronous
@@ -27,7 +27,6 @@
27#include <linux/types.h> 27#include <linux/types.h>
28#include <linux/mm.h> 28#include <linux/mm.h>
29#include <linux/slab.h> 29#include <linux/slab.h>
30#include <linux/in.h>
31#include <linux/utsname.h> 30#include <linux/utsname.h>
32 31
33#include <linux/sunrpc/clnt.h> 32#include <linux/sunrpc/clnt.h>
@@ -53,6 +52,7 @@ static void call_allocate(struct rpc_task *task);
53static void call_encode(struct rpc_task *task); 52static void call_encode(struct rpc_task *task);
54static void call_decode(struct rpc_task *task); 53static void call_decode(struct rpc_task *task);
55static void call_bind(struct rpc_task *task); 54static void call_bind(struct rpc_task *task);
55static void call_bind_status(struct rpc_task *task);
56static void call_transmit(struct rpc_task *task); 56static void call_transmit(struct rpc_task *task);
57static void call_status(struct rpc_task *task); 57static void call_status(struct rpc_task *task);
58static void call_refresh(struct rpc_task *task); 58static void call_refresh(struct rpc_task *task);
@@ -517,15 +517,8 @@ void
517rpc_setbufsize(struct rpc_clnt *clnt, unsigned int sndsize, unsigned int rcvsize) 517rpc_setbufsize(struct rpc_clnt *clnt, unsigned int sndsize, unsigned int rcvsize)
518{ 518{
519 struct rpc_xprt *xprt = clnt->cl_xprt; 519 struct rpc_xprt *xprt = clnt->cl_xprt;
520 520 if (xprt->ops->set_buffer_size)
521 xprt->sndsize = 0; 521 xprt->ops->set_buffer_size(xprt, sndsize, rcvsize);
522 if (sndsize)
523 xprt->sndsize = sndsize + RPC_SLACK_SPACE;
524 xprt->rcvsize = 0;
525 if (rcvsize)
526 xprt->rcvsize = rcvsize + RPC_SLACK_SPACE;
527 if (xprt_connected(xprt))
528 xprt_sock_setbufsize(xprt);
529} 522}
530 523
531/* 524/*
@@ -685,13 +678,11 @@ call_allocate(struct rpc_task *task)
685static void 678static void
686call_encode(struct rpc_task *task) 679call_encode(struct rpc_task *task)
687{ 680{
688 struct rpc_clnt *clnt = task->tk_client;
689 struct rpc_rqst *req = task->tk_rqstp; 681 struct rpc_rqst *req = task->tk_rqstp;
690 struct xdr_buf *sndbuf = &req->rq_snd_buf; 682 struct xdr_buf *sndbuf = &req->rq_snd_buf;
691 struct xdr_buf *rcvbuf = &req->rq_rcv_buf; 683 struct xdr_buf *rcvbuf = &req->rq_rcv_buf;
692 unsigned int bufsiz; 684 unsigned int bufsiz;
693 kxdrproc_t encode; 685 kxdrproc_t encode;
694 int status;
695 u32 *p; 686 u32 *p;
696 687
697 dprintk("RPC: %4d call_encode (status %d)\n", 688 dprintk("RPC: %4d call_encode (status %d)\n",
@@ -719,11 +710,15 @@ call_encode(struct rpc_task *task)
719 rpc_exit(task, -EIO); 710 rpc_exit(task, -EIO);
720 return; 711 return;
721 } 712 }
722 if (encode && (status = rpcauth_wrap_req(task, encode, req, p, 713 if (encode == NULL)
723 task->tk_msg.rpc_argp)) < 0) { 714 return;
724 printk(KERN_WARNING "%s: can't encode arguments: %d\n", 715
725 clnt->cl_protname, -status); 716 task->tk_status = rpcauth_wrap_req(task, encode, req, p,
726 rpc_exit(task, status); 717 task->tk_msg.rpc_argp);
718 if (task->tk_status == -ENOMEM) {
719 /* XXX: Is this sane? */
720 rpc_delay(task, 3*HZ);
721 task->tk_status = -EAGAIN;
727 } 722 }
728} 723}
729 724
@@ -734,43 +729,95 @@ static void
734call_bind(struct rpc_task *task) 729call_bind(struct rpc_task *task)
735{ 730{
736 struct rpc_clnt *clnt = task->tk_client; 731 struct rpc_clnt *clnt = task->tk_client;
737 struct rpc_xprt *xprt = clnt->cl_xprt;
738
739 dprintk("RPC: %4d call_bind xprt %p %s connected\n", task->tk_pid,
740 xprt, (xprt_connected(xprt) ? "is" : "is not"));
741 732
742 task->tk_action = (xprt_connected(xprt)) ? call_transmit : call_connect; 733 dprintk("RPC: %4d call_bind (status %d)\n",
734 task->tk_pid, task->tk_status);
743 735
736 task->tk_action = call_connect;
744 if (!clnt->cl_port) { 737 if (!clnt->cl_port) {
745 task->tk_action = call_connect; 738 task->tk_action = call_bind_status;
746 task->tk_timeout = RPC_CONNECT_TIMEOUT; 739 task->tk_timeout = task->tk_xprt->bind_timeout;
747 rpc_getport(task, clnt); 740 rpc_getport(task, clnt);
748 } 741 }
749} 742}
750 743
751/* 744/*
752 * 4a. Connect to the RPC server (TCP case) 745 * 4a. Sort out bind result
746 */
747static void
748call_bind_status(struct rpc_task *task)
749{
750 int status = -EACCES;
751
752 if (task->tk_status >= 0) {
753 dprintk("RPC: %4d call_bind_status (status %d)\n",
754 task->tk_pid, task->tk_status);
755 task->tk_status = 0;
756 task->tk_action = call_connect;
757 return;
758 }
759
760 switch (task->tk_status) {
761 case -EACCES:
762 dprintk("RPC: %4d remote rpcbind: RPC program/version unavailable\n",
763 task->tk_pid);
764 rpc_delay(task, 3*HZ);
765 goto retry_bind;
766 case -ETIMEDOUT:
767 dprintk("RPC: %4d rpcbind request timed out\n",
768 task->tk_pid);
769 if (RPC_IS_SOFT(task)) {
770 status = -EIO;
771 break;
772 }
773 goto retry_bind;
774 case -EPFNOSUPPORT:
775 dprintk("RPC: %4d remote rpcbind service unavailable\n",
776 task->tk_pid);
777 break;
778 case -EPROTONOSUPPORT:
779 dprintk("RPC: %4d remote rpcbind version 2 unavailable\n",
780 task->tk_pid);
781 break;
782 default:
783 dprintk("RPC: %4d unrecognized rpcbind error (%d)\n",
784 task->tk_pid, -task->tk_status);
785 status = -EIO;
786 break;
787 }
788
789 rpc_exit(task, status);
790 return;
791
792retry_bind:
793 task->tk_status = 0;
794 task->tk_action = call_bind;
795 return;
796}
797
798/*
799 * 4b. Connect to the RPC server
753 */ 800 */
754static void 801static void
755call_connect(struct rpc_task *task) 802call_connect(struct rpc_task *task)
756{ 803{
757 struct rpc_clnt *clnt = task->tk_client; 804 struct rpc_xprt *xprt = task->tk_xprt;
758 805
759 dprintk("RPC: %4d call_connect status %d\n", 806 dprintk("RPC: %4d call_connect xprt %p %s connected\n",
760 task->tk_pid, task->tk_status); 807 task->tk_pid, xprt,
808 (xprt_connected(xprt) ? "is" : "is not"));
761 809
762 if (xprt_connected(clnt->cl_xprt)) { 810 task->tk_action = call_transmit;
763 task->tk_action = call_transmit; 811 if (!xprt_connected(xprt)) {
764 return; 812 task->tk_action = call_connect_status;
813 if (task->tk_status < 0)
814 return;
815 xprt_connect(task);
765 } 816 }
766 task->tk_action = call_connect_status;
767 if (task->tk_status < 0)
768 return;
769 xprt_connect(task);
770} 817}
771 818
772/* 819/*
773 * 4b. Sort out connect result 820 * 4c. Sort out connect result
774 */ 821 */
775static void 822static void
776call_connect_status(struct rpc_task *task) 823call_connect_status(struct rpc_task *task)
@@ -778,6 +825,9 @@ call_connect_status(struct rpc_task *task)
778 struct rpc_clnt *clnt = task->tk_client; 825 struct rpc_clnt *clnt = task->tk_client;
779 int status = task->tk_status; 826 int status = task->tk_status;
780 827
828 dprintk("RPC: %5u call_connect_status (status %d)\n",
829 task->tk_pid, task->tk_status);
830
781 task->tk_status = 0; 831 task->tk_status = 0;
782 if (status >= 0) { 832 if (status >= 0) {
783 clnt->cl_stats->netreconn++; 833 clnt->cl_stats->netreconn++;
@@ -785,17 +835,19 @@ call_connect_status(struct rpc_task *task)
785 return; 835 return;
786 } 836 }
787 837
788 /* Something failed: we may have to rebind */ 838 /* Something failed: remote service port may have changed */
789 if (clnt->cl_autobind) 839 if (clnt->cl_autobind)
790 clnt->cl_port = 0; 840 clnt->cl_port = 0;
841
791 switch (status) { 842 switch (status) {
792 case -ENOTCONN: 843 case -ENOTCONN:
793 case -ETIMEDOUT: 844 case -ETIMEDOUT:
794 case -EAGAIN: 845 case -EAGAIN:
795 task->tk_action = (clnt->cl_port == 0) ? call_bind : call_connect; 846 task->tk_action = call_bind;
796 break; 847 break;
797 default: 848 default:
798 rpc_exit(task, -EIO); 849 rpc_exit(task, -EIO);
850 break;
799 } 851 }
800} 852}
801 853
@@ -815,10 +867,12 @@ call_transmit(struct rpc_task *task)
815 if (task->tk_status != 0) 867 if (task->tk_status != 0)
816 return; 868 return;
817 /* Encode here so that rpcsec_gss can use correct sequence number. */ 869 /* Encode here so that rpcsec_gss can use correct sequence number. */
818 if (!task->tk_rqstp->rq_bytes_sent) 870 if (task->tk_rqstp->rq_bytes_sent == 0) {
819 call_encode(task); 871 call_encode(task);
820 if (task->tk_status < 0) 872 /* Did the encode result in an error condition? */
821 return; 873 if (task->tk_status != 0)
874 goto out_nosend;
875 }
822 xprt_transmit(task); 876 xprt_transmit(task);
823 if (task->tk_status < 0) 877 if (task->tk_status < 0)
824 return; 878 return;
@@ -826,6 +880,10 @@ call_transmit(struct rpc_task *task)
826 task->tk_action = NULL; 880 task->tk_action = NULL;
827 rpc_wake_up_task(task); 881 rpc_wake_up_task(task);
828 } 882 }
883 return;
884out_nosend:
885 /* release socket write lock before attempting to handle error */
886 xprt_abort_transmit(task);
829} 887}
830 888
831/* 889/*
@@ -1020,13 +1078,12 @@ static u32 *
1020call_header(struct rpc_task *task) 1078call_header(struct rpc_task *task)
1021{ 1079{
1022 struct rpc_clnt *clnt = task->tk_client; 1080 struct rpc_clnt *clnt = task->tk_client;
1023 struct rpc_xprt *xprt = clnt->cl_xprt;
1024 struct rpc_rqst *req = task->tk_rqstp; 1081 struct rpc_rqst *req = task->tk_rqstp;
1025 u32 *p = req->rq_svec[0].iov_base; 1082 u32 *p = req->rq_svec[0].iov_base;
1026 1083
1027 /* FIXME: check buffer size? */ 1084 /* FIXME: check buffer size? */
1028 if (xprt->stream) 1085
1029 *p++ = 0; /* fill in later */ 1086 p = xprt_skip_transport_header(task->tk_xprt, p);
1030 *p++ = req->rq_xid; /* XID */ 1087 *p++ = req->rq_xid; /* XID */
1031 *p++ = htonl(RPC_CALL); /* CALL */ 1088 *p++ = htonl(RPC_CALL); /* CALL */
1032 *p++ = htonl(RPC_VERSION); /* RPC version */ 1089 *p++ = htonl(RPC_VERSION); /* RPC version */
diff --git a/net/sunrpc/pmap_clnt.c b/net/sunrpc/pmap_clnt.c
index 4e81f2766923..a398575f94b8 100644
--- a/net/sunrpc/pmap_clnt.c
+++ b/net/sunrpc/pmap_clnt.c
@@ -26,7 +26,7 @@
26#define PMAP_GETPORT 3 26#define PMAP_GETPORT 3
27 27
28static struct rpc_procinfo pmap_procedures[]; 28static struct rpc_procinfo pmap_procedures[];
29static struct rpc_clnt * pmap_create(char *, struct sockaddr_in *, int); 29static struct rpc_clnt * pmap_create(char *, struct sockaddr_in *, int, int);
30static void pmap_getport_done(struct rpc_task *); 30static void pmap_getport_done(struct rpc_task *);
31static struct rpc_program pmap_program; 31static struct rpc_program pmap_program;
32static DEFINE_SPINLOCK(pmap_lock); 32static DEFINE_SPINLOCK(pmap_lock);
@@ -65,7 +65,7 @@ rpc_getport(struct rpc_task *task, struct rpc_clnt *clnt)
65 map->pm_binding = 1; 65 map->pm_binding = 1;
66 spin_unlock(&pmap_lock); 66 spin_unlock(&pmap_lock);
67 67
68 pmap_clnt = pmap_create(clnt->cl_server, sap, map->pm_prot); 68 pmap_clnt = pmap_create(clnt->cl_server, sap, map->pm_prot, 0);
69 if (IS_ERR(pmap_clnt)) { 69 if (IS_ERR(pmap_clnt)) {
70 task->tk_status = PTR_ERR(pmap_clnt); 70 task->tk_status = PTR_ERR(pmap_clnt);
71 goto bailout; 71 goto bailout;
@@ -112,7 +112,7 @@ rpc_getport_external(struct sockaddr_in *sin, __u32 prog, __u32 vers, int prot)
112 NIPQUAD(sin->sin_addr.s_addr), prog, vers, prot); 112 NIPQUAD(sin->sin_addr.s_addr), prog, vers, prot);
113 113
114 sprintf(hostname, "%u.%u.%u.%u", NIPQUAD(sin->sin_addr.s_addr)); 114 sprintf(hostname, "%u.%u.%u.%u", NIPQUAD(sin->sin_addr.s_addr));
115 pmap_clnt = pmap_create(hostname, sin, prot); 115 pmap_clnt = pmap_create(hostname, sin, prot, 0);
116 if (IS_ERR(pmap_clnt)) 116 if (IS_ERR(pmap_clnt))
117 return PTR_ERR(pmap_clnt); 117 return PTR_ERR(pmap_clnt);
118 118
@@ -171,7 +171,7 @@ rpc_register(u32 prog, u32 vers, int prot, unsigned short port, int *okay)
171 171
172 sin.sin_family = AF_INET; 172 sin.sin_family = AF_INET;
173 sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK); 173 sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
174 pmap_clnt = pmap_create("localhost", &sin, IPPROTO_UDP); 174 pmap_clnt = pmap_create("localhost", &sin, IPPROTO_UDP, 1);
175 if (IS_ERR(pmap_clnt)) { 175 if (IS_ERR(pmap_clnt)) {
176 error = PTR_ERR(pmap_clnt); 176 error = PTR_ERR(pmap_clnt);
177 dprintk("RPC: couldn't create pmap client. Error = %d\n", error); 177 dprintk("RPC: couldn't create pmap client. Error = %d\n", error);
@@ -198,7 +198,7 @@ rpc_register(u32 prog, u32 vers, int prot, unsigned short port, int *okay)
198} 198}
199 199
200static struct rpc_clnt * 200static struct rpc_clnt *
201pmap_create(char *hostname, struct sockaddr_in *srvaddr, int proto) 201pmap_create(char *hostname, struct sockaddr_in *srvaddr, int proto, int privileged)
202{ 202{
203 struct rpc_xprt *xprt; 203 struct rpc_xprt *xprt;
204 struct rpc_clnt *clnt; 204 struct rpc_clnt *clnt;
@@ -208,6 +208,8 @@ pmap_create(char *hostname, struct sockaddr_in *srvaddr, int proto)
208 if (IS_ERR(xprt)) 208 if (IS_ERR(xprt))
209 return (struct rpc_clnt *)xprt; 209 return (struct rpc_clnt *)xprt;
210 xprt->addr.sin_port = htons(RPC_PMAP_PORT); 210 xprt->addr.sin_port = htons(RPC_PMAP_PORT);
211 if (!privileged)
212 xprt->resvport = 0;
211 213
212 /* printk("pmap: create clnt\n"); */ 214 /* printk("pmap: create clnt\n"); */
213 clnt = rpc_new_client(xprt, hostname, 215 clnt = rpc_new_client(xprt, hostname,
diff --git a/net/sunrpc/socklib.c b/net/sunrpc/socklib.c
new file mode 100644
index 000000000000..8f97e90f36c8
--- /dev/null
+++ b/net/sunrpc/socklib.c
@@ -0,0 +1,175 @@
1/*
2 * linux/net/sunrpc/socklib.c
3 *
4 * Common socket helper routines for RPC client and server
5 *
6 * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de>
7 */
8
9#include <linux/types.h>
10#include <linux/pagemap.h>
11#include <linux/udp.h>
12#include <linux/sunrpc/xdr.h>
13
14
15/**
16 * skb_read_bits - copy some data bits from skb to internal buffer
17 * @desc: sk_buff copy helper
18 * @to: copy destination
19 * @len: number of bytes to copy
20 *
21 * Possibly called several times to iterate over an sk_buff and copy
22 * data out of it.
23 */
24static size_t skb_read_bits(skb_reader_t *desc, void *to, size_t len)
25{
26 if (len > desc->count)
27 len = desc->count;
28 if (skb_copy_bits(desc->skb, desc->offset, to, len))
29 return 0;
30 desc->count -= len;
31 desc->offset += len;
32 return len;
33}
34
35/**
36 * skb_read_and_csum_bits - copy and checksum from skb to buffer
37 * @desc: sk_buff copy helper
38 * @to: copy destination
39 * @len: number of bytes to copy
40 *
41 * Same as skb_read_bits, but calculate a checksum at the same time.
42 */
43static size_t skb_read_and_csum_bits(skb_reader_t *desc, void *to, size_t len)
44{
45 unsigned int csum2, pos;
46
47 if (len > desc->count)
48 len = desc->count;
49 pos = desc->offset;
50 csum2 = skb_copy_and_csum_bits(desc->skb, pos, to, len, 0);
51 desc->csum = csum_block_add(desc->csum, csum2, pos);
52 desc->count -= len;
53 desc->offset += len;
54 return len;
55}
56
57/**
58 * xdr_partial_copy_from_skb - copy data out of an skb
59 * @xdr: target XDR buffer
60 * @base: starting offset
61 * @desc: sk_buff copy helper
62 * @copy_actor: virtual method for copying data
63 *
64 */
65ssize_t xdr_partial_copy_from_skb(struct xdr_buf *xdr, unsigned int base, skb_reader_t *desc, skb_read_actor_t copy_actor)
66{
67 struct page **ppage = xdr->pages;
68 unsigned int len, pglen = xdr->page_len;
69 ssize_t copied = 0;
70 int ret;
71
72 len = xdr->head[0].iov_len;
73 if (base < len) {
74 len -= base;
75 ret = copy_actor(desc, (char *)xdr->head[0].iov_base + base, len);
76 copied += ret;
77 if (ret != len || !desc->count)
78 goto out;
79 base = 0;
80 } else
81 base -= len;
82
83 if (unlikely(pglen == 0))
84 goto copy_tail;
85 if (unlikely(base >= pglen)) {
86 base -= pglen;
87 goto copy_tail;
88 }
89 if (base || xdr->page_base) {
90 pglen -= base;
91 base += xdr->page_base;
92 ppage += base >> PAGE_CACHE_SHIFT;
93 base &= ~PAGE_CACHE_MASK;
94 }
95 do {
96 char *kaddr;
97
98 /* ACL likes to be lazy in allocating pages - ACLs
99 * are small by default but can get huge. */
100 if (unlikely(*ppage == NULL)) {
101 *ppage = alloc_page(GFP_ATOMIC);
102 if (unlikely(*ppage == NULL)) {
103 if (copied == 0)
104 copied = -ENOMEM;
105 goto out;
106 }
107 }
108
109 len = PAGE_CACHE_SIZE;
110 kaddr = kmap_atomic(*ppage, KM_SKB_SUNRPC_DATA);
111 if (base) {
112 len -= base;
113 if (pglen < len)
114 len = pglen;
115 ret = copy_actor(desc, kaddr + base, len);
116 base = 0;
117 } else {
118 if (pglen < len)
119 len = pglen;
120 ret = copy_actor(desc, kaddr, len);
121 }
122 flush_dcache_page(*ppage);
123 kunmap_atomic(kaddr, KM_SKB_SUNRPC_DATA);
124 copied += ret;
125 if (ret != len || !desc->count)
126 goto out;
127 ppage++;
128 } while ((pglen -= len) != 0);
129copy_tail:
130 len = xdr->tail[0].iov_len;
131 if (base < len)
132 copied += copy_actor(desc, (char *)xdr->tail[0].iov_base + base, len - base);
133out:
134 return copied;
135}
136
137/**
138 * csum_partial_copy_to_xdr - checksum and copy data
139 * @xdr: target XDR buffer
140 * @skb: source skb
141 *
142 * We have set things up such that we perform the checksum of the UDP
143 * packet in parallel with the copies into the RPC client iovec. -DaveM
144 */
145int csum_partial_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb)
146{
147 skb_reader_t desc;
148
149 desc.skb = skb;
150 desc.offset = sizeof(struct udphdr);
151 desc.count = skb->len - desc.offset;
152
153 if (skb->ip_summed == CHECKSUM_UNNECESSARY)
154 goto no_checksum;
155
156 desc.csum = csum_partial(skb->data, desc.offset, skb->csum);
157 if (xdr_partial_copy_from_skb(xdr, 0, &desc, skb_read_and_csum_bits) < 0)
158 return -1;
159 if (desc.offset != skb->len) {
160 unsigned int csum2;
161 csum2 = skb_checksum(skb, desc.offset, skb->len - desc.offset, 0);
162 desc.csum = csum_block_add(desc.csum, csum2, desc.offset);
163 }
164 if (desc.count)
165 return -1;
166 if ((unsigned short)csum_fold(desc.csum))
167 return -1;
168 return 0;
169no_checksum:
170 if (xdr_partial_copy_from_skb(xdr, 0, &desc, skb_read_bits) < 0)
171 return -1;
172 if (desc.count)
173 return -1;
174 return 0;
175}
diff --git a/net/sunrpc/sunrpc_syms.c b/net/sunrpc/sunrpc_syms.c
index ed48ff022d35..2387e7b823ff 100644
--- a/net/sunrpc/sunrpc_syms.c
+++ b/net/sunrpc/sunrpc_syms.c
@@ -10,7 +10,6 @@
10#include <linux/module.h> 10#include <linux/module.h>
11 11
12#include <linux/types.h> 12#include <linux/types.h>
13#include <linux/socket.h>
14#include <linux/sched.h> 13#include <linux/sched.h>
15#include <linux/uio.h> 14#include <linux/uio.h>
16#include <linux/unistd.h> 15#include <linux/unistd.h>
diff --git a/net/sunrpc/svcsock.c b/net/sunrpc/svcsock.c
index 691dea4a58e7..f16e7cdd6150 100644
--- a/net/sunrpc/svcsock.c
+++ b/net/sunrpc/svcsock.c
@@ -548,9 +548,6 @@ svc_write_space(struct sock *sk)
548/* 548/*
549 * Receive a datagram from a UDP socket. 549 * Receive a datagram from a UDP socket.
550 */ 550 */
551extern int
552csum_partial_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb);
553
554static int 551static int
555svc_udp_recvfrom(struct svc_rqst *rqstp) 552svc_udp_recvfrom(struct svc_rqst *rqstp)
556{ 553{
diff --git a/net/sunrpc/sysctl.c b/net/sunrpc/sysctl.c
index 1b9616a12e24..d0c9f460e411 100644
--- a/net/sunrpc/sysctl.c
+++ b/net/sunrpc/sysctl.c
@@ -119,8 +119,18 @@ done:
119 return 0; 119 return 0;
120} 120}
121 121
122unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
123unsigned int xprt_tcp_slot_table_entries = RPC_DEF_SLOT_TABLE;
124unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
125EXPORT_SYMBOL(xprt_min_resvport);
126unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
127EXPORT_SYMBOL(xprt_max_resvport);
128
129
122static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE; 130static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
123static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE; 131static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
132static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
133static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
124 134
125static ctl_table debug_table[] = { 135static ctl_table debug_table[] = {
126 { 136 {
@@ -177,6 +187,28 @@ static ctl_table debug_table[] = {
177 .extra1 = &min_slot_table_size, 187 .extra1 = &min_slot_table_size,
178 .extra2 = &max_slot_table_size 188 .extra2 = &max_slot_table_size
179 }, 189 },
190 {
191 .ctl_name = CTL_MIN_RESVPORT,
192 .procname = "min_resvport",
193 .data = &xprt_min_resvport,
194 .maxlen = sizeof(unsigned int),
195 .mode = 0644,
196 .proc_handler = &proc_dointvec_minmax,
197 .strategy = &sysctl_intvec,
198 .extra1 = &xprt_min_resvport_limit,
199 .extra2 = &xprt_max_resvport_limit
200 },
201 {
202 .ctl_name = CTL_MAX_RESVPORT,
203 .procname = "max_resvport",
204 .data = &xprt_max_resvport,
205 .maxlen = sizeof(unsigned int),
206 .mode = 0644,
207 .proc_handler = &proc_dointvec_minmax,
208 .strategy = &sysctl_intvec,
209 .extra1 = &xprt_min_resvport_limit,
210 .extra2 = &xprt_max_resvport_limit
211 },
180 { .ctl_name = 0 } 212 { .ctl_name = 0 }
181}; 213};
182 214
diff --git a/net/sunrpc/xdr.c b/net/sunrpc/xdr.c
index fde16f40a581..32df43372ee9 100644
--- a/net/sunrpc/xdr.c
+++ b/net/sunrpc/xdr.c
@@ -6,15 +6,12 @@
6 * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de> 6 * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de>
7 */ 7 */
8 8
9#include <linux/module.h>
9#include <linux/types.h> 10#include <linux/types.h>
10#include <linux/socket.h>
11#include <linux/string.h> 11#include <linux/string.h>
12#include <linux/kernel.h> 12#include <linux/kernel.h>
13#include <linux/pagemap.h> 13#include <linux/pagemap.h>
14#include <linux/errno.h> 14#include <linux/errno.h>
15#include <linux/in.h>
16#include <linux/net.h>
17#include <net/sock.h>
18#include <linux/sunrpc/xdr.h> 15#include <linux/sunrpc/xdr.h>
19#include <linux/sunrpc/msg_prot.h> 16#include <linux/sunrpc/msg_prot.h>
20 17
@@ -176,178 +173,6 @@ xdr_inline_pages(struct xdr_buf *xdr, unsigned int offset,
176 xdr->buflen += len; 173 xdr->buflen += len;
177} 174}
178 175
179ssize_t
180xdr_partial_copy_from_skb(struct xdr_buf *xdr, unsigned int base,
181 skb_reader_t *desc,
182 skb_read_actor_t copy_actor)
183{
184 struct page **ppage = xdr->pages;
185 unsigned int len, pglen = xdr->page_len;
186 ssize_t copied = 0;
187 int ret;
188
189 len = xdr->head[0].iov_len;
190 if (base < len) {
191 len -= base;
192 ret = copy_actor(desc, (char *)xdr->head[0].iov_base + base, len);
193 copied += ret;
194 if (ret != len || !desc->count)
195 goto out;
196 base = 0;
197 } else
198 base -= len;
199
200 if (pglen == 0)
201 goto copy_tail;
202 if (base >= pglen) {
203 base -= pglen;
204 goto copy_tail;
205 }
206 if (base || xdr->page_base) {
207 pglen -= base;
208 base += xdr->page_base;
209 ppage += base >> PAGE_CACHE_SHIFT;
210 base &= ~PAGE_CACHE_MASK;
211 }
212 do {
213 char *kaddr;
214
215 /* ACL likes to be lazy in allocating pages - ACLs
216 * are small by default but can get huge. */
217 if (unlikely(*ppage == NULL)) {
218 *ppage = alloc_page(GFP_ATOMIC);
219 if (unlikely(*ppage == NULL)) {
220 if (copied == 0)
221 copied = -ENOMEM;
222 goto out;
223 }
224 }
225
226 len = PAGE_CACHE_SIZE;
227 kaddr = kmap_atomic(*ppage, KM_SKB_SUNRPC_DATA);
228 if (base) {
229 len -= base;
230 if (pglen < len)
231 len = pglen;
232 ret = copy_actor(desc, kaddr + base, len);
233 base = 0;
234 } else {
235 if (pglen < len)
236 len = pglen;
237 ret = copy_actor(desc, kaddr, len);
238 }
239 flush_dcache_page(*ppage);
240 kunmap_atomic(kaddr, KM_SKB_SUNRPC_DATA);
241 copied += ret;
242 if (ret != len || !desc->count)
243 goto out;
244 ppage++;
245 } while ((pglen -= len) != 0);
246copy_tail:
247 len = xdr->tail[0].iov_len;
248 if (base < len)
249 copied += copy_actor(desc, (char *)xdr->tail[0].iov_base + base, len - base);
250out:
251 return copied;
252}
253
254
255int
256xdr_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen,
257 struct xdr_buf *xdr, unsigned int base, int msgflags)
258{
259 struct page **ppage = xdr->pages;
260 unsigned int len, pglen = xdr->page_len;
261 int err, ret = 0;
262 ssize_t (*sendpage)(struct socket *, struct page *, int, size_t, int);
263
264 len = xdr->head[0].iov_len;
265 if (base < len || (addr != NULL && base == 0)) {
266 struct kvec iov = {
267 .iov_base = xdr->head[0].iov_base + base,
268 .iov_len = len - base,
269 };
270 struct msghdr msg = {
271 .msg_name = addr,
272 .msg_namelen = addrlen,
273 .msg_flags = msgflags,
274 };
275 if (xdr->len > len)
276 msg.msg_flags |= MSG_MORE;
277
278 if (iov.iov_len != 0)
279 err = kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
280 else
281 err = kernel_sendmsg(sock, &msg, NULL, 0, 0);
282 if (ret == 0)
283 ret = err;
284 else if (err > 0)
285 ret += err;
286 if (err != iov.iov_len)
287 goto out;
288 base = 0;
289 } else
290 base -= len;
291
292 if (pglen == 0)
293 goto copy_tail;
294 if (base >= pglen) {
295 base -= pglen;
296 goto copy_tail;
297 }
298 if (base || xdr->page_base) {
299 pglen -= base;
300 base += xdr->page_base;
301 ppage += base >> PAGE_CACHE_SHIFT;
302 base &= ~PAGE_CACHE_MASK;
303 }
304
305 sendpage = sock->ops->sendpage ? : sock_no_sendpage;
306 do {
307 int flags = msgflags;
308
309 len = PAGE_CACHE_SIZE;
310 if (base)
311 len -= base;
312 if (pglen < len)
313 len = pglen;
314
315 if (pglen != len || xdr->tail[0].iov_len != 0)
316 flags |= MSG_MORE;
317
318 /* Hmm... We might be dealing with highmem pages */
319 if (PageHighMem(*ppage))
320 sendpage = sock_no_sendpage;
321 err = sendpage(sock, *ppage, base, len, flags);
322 if (ret == 0)
323 ret = err;
324 else if (err > 0)
325 ret += err;
326 if (err != len)
327 goto out;
328 base = 0;
329 ppage++;
330 } while ((pglen -= len) != 0);
331copy_tail:
332 len = xdr->tail[0].iov_len;
333 if (base < len) {
334 struct kvec iov = {
335 .iov_base = xdr->tail[0].iov_base + base,
336 .iov_len = len - base,
337 };
338 struct msghdr msg = {
339 .msg_flags = msgflags,
340 };
341 err = kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
342 if (ret == 0)
343 ret = err;
344 else if (err > 0)
345 ret += err;
346 }
347out:
348 return ret;
349}
350
351 176
352/* 177/*
353 * Helper routines for doing 'memmove' like operations on a struct xdr_buf 178 * Helper routines for doing 'memmove' like operations on a struct xdr_buf
diff --git a/net/sunrpc/xprt.c b/net/sunrpc/xprt.c
index 3c654e06b084..6dda3860351f 100644
--- a/net/sunrpc/xprt.c
+++ b/net/sunrpc/xprt.c
@@ -10,12 +10,12 @@
10 * one is available. Otherwise, it sleeps on the backlog queue 10 * one is available. Otherwise, it sleeps on the backlog queue
11 * (xprt_reserve). 11 * (xprt_reserve).
12 * - Next, the caller puts together the RPC message, stuffs it into 12 * - Next, the caller puts together the RPC message, stuffs it into
13 * the request struct, and calls xprt_call(). 13 * the request struct, and calls xprt_transmit().
14 * - xprt_call transmits the message and installs the caller on the 14 * - xprt_transmit sends the message and installs the caller on the
15 * socket's wait list. At the same time, it installs a timer that 15 * transport's wait list. At the same time, it installs a timer that
16 * is run after the packet's timeout has expired. 16 * is run after the packet's timeout has expired.
17 * - When a packet arrives, the data_ready handler walks the list of 17 * - When a packet arrives, the data_ready handler walks the list of
18 * pending requests for that socket. If a matching XID is found, the 18 * pending requests for that transport. If a matching XID is found, the
19 * caller is woken up, and the timer removed. 19 * caller is woken up, and the timer removed.
20 * - When no reply arrives within the timeout interval, the timer is 20 * - When no reply arrives within the timeout interval, the timer is
21 * fired by the kernel and runs xprt_timer(). It either adjusts the 21 * fired by the kernel and runs xprt_timer(). It either adjusts the
@@ -33,36 +33,17 @@
33 * 33 *
34 * Copyright (C) 1995-1997, Olaf Kirch <okir@monad.swb.de> 34 * Copyright (C) 1995-1997, Olaf Kirch <okir@monad.swb.de>
35 * 35 *
36 * TCP callback races fixes (C) 1998 Red Hat Software <alan@redhat.com> 36 * Transport switch API copyright (C) 2005, Chuck Lever <cel@netapp.com>
37 * TCP send fixes (C) 1998 Red Hat Software <alan@redhat.com>
38 * TCP NFS related read + write fixes
39 * (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
40 *
41 * Rewrite of larges part of the code in order to stabilize TCP stuff.
42 * Fix behaviour when socket buffer is full.
43 * (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
44 */ 37 */
45 38
39#include <linux/module.h>
40
46#include <linux/types.h> 41#include <linux/types.h>
47#include <linux/slab.h> 42#include <linux/interrupt.h>
48#include <linux/capability.h>
49#include <linux/sched.h>
50#include <linux/errno.h>
51#include <linux/socket.h>
52#include <linux/in.h>
53#include <linux/net.h>
54#include <linux/mm.h>
55#include <linux/udp.h>
56#include <linux/tcp.h>
57#include <linux/sunrpc/clnt.h>
58#include <linux/file.h>
59#include <linux/workqueue.h> 43#include <linux/workqueue.h>
60#include <linux/random.h> 44#include <linux/random.h>
61 45
62#include <net/sock.h> 46#include <linux/sunrpc/clnt.h>
63#include <net/checksum.h>
64#include <net/udp.h>
65#include <net/tcp.h>
66 47
67/* 48/*
68 * Local variables 49 * Local variables
@@ -73,81 +54,90 @@
73# define RPCDBG_FACILITY RPCDBG_XPRT 54# define RPCDBG_FACILITY RPCDBG_XPRT
74#endif 55#endif
75 56
76#define XPRT_MAX_BACKOFF (8)
77#define XPRT_IDLE_TIMEOUT (5*60*HZ)
78#define XPRT_MAX_RESVPORT (800)
79
80/* 57/*
81 * Local functions 58 * Local functions
82 */ 59 */
83static void xprt_request_init(struct rpc_task *, struct rpc_xprt *); 60static void xprt_request_init(struct rpc_task *, struct rpc_xprt *);
84static inline void do_xprt_reserve(struct rpc_task *); 61static inline void do_xprt_reserve(struct rpc_task *);
85static void xprt_disconnect(struct rpc_xprt *);
86static void xprt_connect_status(struct rpc_task *task); 62static void xprt_connect_status(struct rpc_task *task);
87static struct rpc_xprt * xprt_setup(int proto, struct sockaddr_in *ap,
88 struct rpc_timeout *to);
89static struct socket *xprt_create_socket(struct rpc_xprt *, int, int);
90static void xprt_bind_socket(struct rpc_xprt *, struct socket *);
91static int __xprt_get_cong(struct rpc_xprt *, struct rpc_task *); 63static int __xprt_get_cong(struct rpc_xprt *, struct rpc_task *);
92 64
93static int xprt_clear_backlog(struct rpc_xprt *xprt);
94
95#ifdef RPC_DEBUG_DATA
96/* 65/*
97 * Print the buffer contents (first 128 bytes only--just enough for 66 * The transport code maintains an estimate on the maximum number of out-
98 * diropres return). 67 * standing RPC requests, using a smoothed version of the congestion
68 * avoidance implemented in 44BSD. This is basically the Van Jacobson
69 * congestion algorithm: If a retransmit occurs, the congestion window is
70 * halved; otherwise, it is incremented by 1/cwnd when
71 *
72 * - a reply is received and
73 * - a full number of requests are outstanding and
74 * - the congestion window hasn't been updated recently.
99 */ 75 */
100static void 76#define RPC_CWNDSHIFT (8U)
101xprt_pktdump(char *msg, u32 *packet, unsigned int count) 77#define RPC_CWNDSCALE (1U << RPC_CWNDSHIFT)
102{ 78#define RPC_INITCWND RPC_CWNDSCALE
103 u8 *buf = (u8 *) packet; 79#define RPC_MAXCWND(xprt) ((xprt)->max_reqs << RPC_CWNDSHIFT)
104 int j;
105
106 dprintk("RPC: %s\n", msg);
107 for (j = 0; j < count && j < 128; j += 4) {
108 if (!(j & 31)) {
109 if (j)
110 dprintk("\n");
111 dprintk("0x%04x ", j);
112 }
113 dprintk("%02x%02x%02x%02x ",
114 buf[j], buf[j+1], buf[j+2], buf[j+3]);
115 }
116 dprintk("\n");
117}
118#else
119static inline void
120xprt_pktdump(char *msg, u32 *packet, unsigned int count)
121{
122 /* NOP */
123}
124#endif
125 80
126/* 81#define RPCXPRT_CONGESTED(xprt) ((xprt)->cong >= (xprt)->cwnd)
127 * Look up RPC transport given an INET socket 82
83/**
84 * xprt_reserve_xprt - serialize write access to transports
85 * @task: task that is requesting access to the transport
86 *
87 * This prevents mixing the payload of separate requests, and prevents
88 * transport connects from colliding with writes. No congestion control
89 * is provided.
128 */ 90 */
129static inline struct rpc_xprt * 91int xprt_reserve_xprt(struct rpc_task *task)
130xprt_from_sock(struct sock *sk)
131{ 92{
132 return (struct rpc_xprt *) sk->sk_user_data; 93 struct rpc_xprt *xprt = task->tk_xprt;
94 struct rpc_rqst *req = task->tk_rqstp;
95
96 if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) {
97 if (task == xprt->snd_task)
98 return 1;
99 if (task == NULL)
100 return 0;
101 goto out_sleep;
102 }
103 xprt->snd_task = task;
104 if (req) {
105 req->rq_bytes_sent = 0;
106 req->rq_ntrans++;
107 }
108 return 1;
109
110out_sleep:
111 dprintk("RPC: %4d failed to lock transport %p\n",
112 task->tk_pid, xprt);
113 task->tk_timeout = 0;
114 task->tk_status = -EAGAIN;
115 if (req && req->rq_ntrans)
116 rpc_sleep_on(&xprt->resend, task, NULL, NULL);
117 else
118 rpc_sleep_on(&xprt->sending, task, NULL, NULL);
119 return 0;
133} 120}
134 121
135/* 122/*
136 * Serialize write access to sockets, in order to prevent different 123 * xprt_reserve_xprt_cong - serialize write access to transports
137 * requests from interfering with each other. 124 * @task: task that is requesting access to the transport
138 * Also prevents TCP socket connects from colliding with writes. 125 *
126 * Same as xprt_reserve_xprt, but Van Jacobson congestion control is
127 * integrated into the decision of whether a request is allowed to be
128 * woken up and given access to the transport.
139 */ 129 */
140static int 130int xprt_reserve_xprt_cong(struct rpc_task *task)
141__xprt_lock_write(struct rpc_xprt *xprt, struct rpc_task *task)
142{ 131{
132 struct rpc_xprt *xprt = task->tk_xprt;
143 struct rpc_rqst *req = task->tk_rqstp; 133 struct rpc_rqst *req = task->tk_rqstp;
144 134
145 if (test_and_set_bit(XPRT_LOCKED, &xprt->sockstate)) { 135 if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) {
146 if (task == xprt->snd_task) 136 if (task == xprt->snd_task)
147 return 1; 137 return 1;
148 goto out_sleep; 138 goto out_sleep;
149 } 139 }
150 if (xprt->nocong || __xprt_get_cong(xprt, task)) { 140 if (__xprt_get_cong(xprt, task)) {
151 xprt->snd_task = task; 141 xprt->snd_task = task;
152 if (req) { 142 if (req) {
153 req->rq_bytes_sent = 0; 143 req->rq_bytes_sent = 0;
@@ -156,10 +146,10 @@ __xprt_lock_write(struct rpc_xprt *xprt, struct rpc_task *task)
156 return 1; 146 return 1;
157 } 147 }
158 smp_mb__before_clear_bit(); 148 smp_mb__before_clear_bit();
159 clear_bit(XPRT_LOCKED, &xprt->sockstate); 149 clear_bit(XPRT_LOCKED, &xprt->state);
160 smp_mb__after_clear_bit(); 150 smp_mb__after_clear_bit();
161out_sleep: 151out_sleep:
162 dprintk("RPC: %4d failed to lock socket %p\n", task->tk_pid, xprt); 152 dprintk("RPC: %4d failed to lock transport %p\n", task->tk_pid, xprt);
163 task->tk_timeout = 0; 153 task->tk_timeout = 0;
164 task->tk_status = -EAGAIN; 154 task->tk_status = -EAGAIN;
165 if (req && req->rq_ntrans) 155 if (req && req->rq_ntrans)
@@ -169,26 +159,52 @@ out_sleep:
169 return 0; 159 return 0;
170} 160}
171 161
172static inline int 162static inline int xprt_lock_write(struct rpc_xprt *xprt, struct rpc_task *task)
173xprt_lock_write(struct rpc_xprt *xprt, struct rpc_task *task)
174{ 163{
175 int retval; 164 int retval;
176 165
177 spin_lock_bh(&xprt->sock_lock); 166 spin_lock_bh(&xprt->transport_lock);
178 retval = __xprt_lock_write(xprt, task); 167 retval = xprt->ops->reserve_xprt(task);
179 spin_unlock_bh(&xprt->sock_lock); 168 spin_unlock_bh(&xprt->transport_lock);
180 return retval; 169 return retval;
181} 170}
182 171
172static void __xprt_lock_write_next(struct rpc_xprt *xprt)
173{
174 struct rpc_task *task;
175 struct rpc_rqst *req;
183 176
184static void 177 if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
185__xprt_lock_write_next(struct rpc_xprt *xprt) 178 return;
179
180 task = rpc_wake_up_next(&xprt->resend);
181 if (!task) {
182 task = rpc_wake_up_next(&xprt->sending);
183 if (!task)
184 goto out_unlock;
185 }
186
187 req = task->tk_rqstp;
188 xprt->snd_task = task;
189 if (req) {
190 req->rq_bytes_sent = 0;
191 req->rq_ntrans++;
192 }
193 return;
194
195out_unlock:
196 smp_mb__before_clear_bit();
197 clear_bit(XPRT_LOCKED, &xprt->state);
198 smp_mb__after_clear_bit();
199}
200
201static void __xprt_lock_write_next_cong(struct rpc_xprt *xprt)
186{ 202{
187 struct rpc_task *task; 203 struct rpc_task *task;
188 204
189 if (test_and_set_bit(XPRT_LOCKED, &xprt->sockstate)) 205 if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
190 return; 206 return;
191 if (!xprt->nocong && RPCXPRT_CONGESTED(xprt)) 207 if (RPCXPRT_CONGESTED(xprt))
192 goto out_unlock; 208 goto out_unlock;
193 task = rpc_wake_up_next(&xprt->resend); 209 task = rpc_wake_up_next(&xprt->resend);
194 if (!task) { 210 if (!task) {
@@ -196,7 +212,7 @@ __xprt_lock_write_next(struct rpc_xprt *xprt)
196 if (!task) 212 if (!task)
197 goto out_unlock; 213 goto out_unlock;
198 } 214 }
199 if (xprt->nocong || __xprt_get_cong(xprt, task)) { 215 if (__xprt_get_cong(xprt, task)) {
200 struct rpc_rqst *req = task->tk_rqstp; 216 struct rpc_rqst *req = task->tk_rqstp;
201 xprt->snd_task = task; 217 xprt->snd_task = task;
202 if (req) { 218 if (req) {
@@ -207,87 +223,52 @@ __xprt_lock_write_next(struct rpc_xprt *xprt)
207 } 223 }
208out_unlock: 224out_unlock:
209 smp_mb__before_clear_bit(); 225 smp_mb__before_clear_bit();
210 clear_bit(XPRT_LOCKED, &xprt->sockstate); 226 clear_bit(XPRT_LOCKED, &xprt->state);
211 smp_mb__after_clear_bit(); 227 smp_mb__after_clear_bit();
212} 228}
213 229
214/* 230/**
215 * Releases the socket for use by other requests. 231 * xprt_release_xprt - allow other requests to use a transport
232 * @xprt: transport with other tasks potentially waiting
233 * @task: task that is releasing access to the transport
234 *
235 * Note that "task" can be NULL. No congestion control is provided.
216 */ 236 */
217static void 237void xprt_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
218__xprt_release_write(struct rpc_xprt *xprt, struct rpc_task *task)
219{ 238{
220 if (xprt->snd_task == task) { 239 if (xprt->snd_task == task) {
221 xprt->snd_task = NULL; 240 xprt->snd_task = NULL;
222 smp_mb__before_clear_bit(); 241 smp_mb__before_clear_bit();
223 clear_bit(XPRT_LOCKED, &xprt->sockstate); 242 clear_bit(XPRT_LOCKED, &xprt->state);
224 smp_mb__after_clear_bit(); 243 smp_mb__after_clear_bit();
225 __xprt_lock_write_next(xprt); 244 __xprt_lock_write_next(xprt);
226 } 245 }
227} 246}
228 247
229static inline void 248/**
230xprt_release_write(struct rpc_xprt *xprt, struct rpc_task *task) 249 * xprt_release_xprt_cong - allow other requests to use a transport
231{ 250 * @xprt: transport with other tasks potentially waiting
232 spin_lock_bh(&xprt->sock_lock); 251 * @task: task that is releasing access to the transport
233 __xprt_release_write(xprt, task); 252 *
234 spin_unlock_bh(&xprt->sock_lock); 253 * Note that "task" can be NULL. Another task is awoken to use the
235} 254 * transport if the transport's congestion window allows it.
236
237/*
238 * Write data to socket.
239 */ 255 */
240static inline int 256void xprt_release_xprt_cong(struct rpc_xprt *xprt, struct rpc_task *task)
241xprt_sendmsg(struct rpc_xprt *xprt, struct rpc_rqst *req)
242{ 257{
243 struct socket *sock = xprt->sock; 258 if (xprt->snd_task == task) {
244 struct xdr_buf *xdr = &req->rq_snd_buf; 259 xprt->snd_task = NULL;
245 struct sockaddr *addr = NULL; 260 smp_mb__before_clear_bit();
246 int addrlen = 0; 261 clear_bit(XPRT_LOCKED, &xprt->state);
247 unsigned int skip; 262 smp_mb__after_clear_bit();
248 int result; 263 __xprt_lock_write_next_cong(xprt);
249
250 if (!sock)
251 return -ENOTCONN;
252
253 xprt_pktdump("packet data:",
254 req->rq_svec->iov_base,
255 req->rq_svec->iov_len);
256
257 /* For UDP, we need to provide an address */
258 if (!xprt->stream) {
259 addr = (struct sockaddr *) &xprt->addr;
260 addrlen = sizeof(xprt->addr);
261 } 264 }
262 /* Dont repeat bytes */ 265}
263 skip = req->rq_bytes_sent;
264
265 clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
266 result = xdr_sendpages(sock, addr, addrlen, xdr, skip, MSG_DONTWAIT);
267
268 dprintk("RPC: xprt_sendmsg(%d) = %d\n", xdr->len - skip, result);
269
270 if (result >= 0)
271 return result;
272 266
273 switch (result) { 267static inline void xprt_release_write(struct rpc_xprt *xprt, struct rpc_task *task)
274 case -ECONNREFUSED: 268{
275 /* When the server has died, an ICMP port unreachable message 269 spin_lock_bh(&xprt->transport_lock);
276 * prompts ECONNREFUSED. 270 xprt->ops->release_xprt(xprt, task);
277 */ 271 spin_unlock_bh(&xprt->transport_lock);
278 case -EAGAIN:
279 break;
280 case -ECONNRESET:
281 case -ENOTCONN:
282 case -EPIPE:
283 /* connection broken */
284 if (xprt->stream)
285 result = -ENOTCONN;
286 break;
287 default:
288 printk(KERN_NOTICE "RPC: sendmsg returned error %d\n", -result);
289 }
290 return result;
291} 272}
292 273
293/* 274/*
@@ -321,26 +302,40 @@ __xprt_put_cong(struct rpc_xprt *xprt, struct rpc_rqst *req)
321 return; 302 return;
322 req->rq_cong = 0; 303 req->rq_cong = 0;
323 xprt->cong -= RPC_CWNDSCALE; 304 xprt->cong -= RPC_CWNDSCALE;
324 __xprt_lock_write_next(xprt); 305 __xprt_lock_write_next_cong(xprt);
325} 306}
326 307
327/* 308/**
328 * Adjust RPC congestion window 309 * xprt_release_rqst_cong - housekeeping when request is complete
310 * @task: RPC request that recently completed
311 *
312 * Useful for transports that require congestion control.
313 */
314void xprt_release_rqst_cong(struct rpc_task *task)
315{
316 __xprt_put_cong(task->tk_xprt, task->tk_rqstp);
317}
318
319/**
320 * xprt_adjust_cwnd - adjust transport congestion window
321 * @task: recently completed RPC request used to adjust window
322 * @result: result code of completed RPC request
323 *
329 * We use a time-smoothed congestion estimator to avoid heavy oscillation. 324 * We use a time-smoothed congestion estimator to avoid heavy oscillation.
330 */ 325 */
331static void 326void xprt_adjust_cwnd(struct rpc_task *task, int result)
332xprt_adjust_cwnd(struct rpc_xprt *xprt, int result)
333{ 327{
334 unsigned long cwnd; 328 struct rpc_rqst *req = task->tk_rqstp;
329 struct rpc_xprt *xprt = task->tk_xprt;
330 unsigned long cwnd = xprt->cwnd;
335 331
336 cwnd = xprt->cwnd;
337 if (result >= 0 && cwnd <= xprt->cong) { 332 if (result >= 0 && cwnd <= xprt->cong) {
338 /* The (cwnd >> 1) term makes sure 333 /* The (cwnd >> 1) term makes sure
339 * the result gets rounded properly. */ 334 * the result gets rounded properly. */
340 cwnd += (RPC_CWNDSCALE * RPC_CWNDSCALE + (cwnd >> 1)) / cwnd; 335 cwnd += (RPC_CWNDSCALE * RPC_CWNDSCALE + (cwnd >> 1)) / cwnd;
341 if (cwnd > RPC_MAXCWND(xprt)) 336 if (cwnd > RPC_MAXCWND(xprt))
342 cwnd = RPC_MAXCWND(xprt); 337 cwnd = RPC_MAXCWND(xprt);
343 __xprt_lock_write_next(xprt); 338 __xprt_lock_write_next_cong(xprt);
344 } else if (result == -ETIMEDOUT) { 339 } else if (result == -ETIMEDOUT) {
345 cwnd >>= 1; 340 cwnd >>= 1;
346 if (cwnd < RPC_CWNDSCALE) 341 if (cwnd < RPC_CWNDSCALE)
@@ -349,11 +344,89 @@ xprt_adjust_cwnd(struct rpc_xprt *xprt, int result)
349 dprintk("RPC: cong %ld, cwnd was %ld, now %ld\n", 344 dprintk("RPC: cong %ld, cwnd was %ld, now %ld\n",
350 xprt->cong, xprt->cwnd, cwnd); 345 xprt->cong, xprt->cwnd, cwnd);
351 xprt->cwnd = cwnd; 346 xprt->cwnd = cwnd;
347 __xprt_put_cong(xprt, req);
348}
349
350/**
351 * xprt_wake_pending_tasks - wake all tasks on a transport's pending queue
352 * @xprt: transport with waiting tasks
353 * @status: result code to plant in each task before waking it
354 *
355 */
356void xprt_wake_pending_tasks(struct rpc_xprt *xprt, int status)
357{
358 if (status < 0)
359 rpc_wake_up_status(&xprt->pending, status);
360 else
361 rpc_wake_up(&xprt->pending);
362}
363
364/**
365 * xprt_wait_for_buffer_space - wait for transport output buffer to clear
366 * @task: task to be put to sleep
367 *
368 */
369void xprt_wait_for_buffer_space(struct rpc_task *task)
370{
371 struct rpc_rqst *req = task->tk_rqstp;
372 struct rpc_xprt *xprt = req->rq_xprt;
373
374 task->tk_timeout = req->rq_timeout;
375 rpc_sleep_on(&xprt->pending, task, NULL, NULL);
376}
377
378/**
379 * xprt_write_space - wake the task waiting for transport output buffer space
380 * @xprt: transport with waiting tasks
381 *
382 * Can be called in a soft IRQ context, so xprt_write_space never sleeps.
383 */
384void xprt_write_space(struct rpc_xprt *xprt)
385{
386 if (unlikely(xprt->shutdown))
387 return;
388
389 spin_lock_bh(&xprt->transport_lock);
390 if (xprt->snd_task) {
391 dprintk("RPC: write space: waking waiting task on xprt %p\n",
392 xprt);
393 rpc_wake_up_task(xprt->snd_task);
394 }
395 spin_unlock_bh(&xprt->transport_lock);
396}
397
398/**
399 * xprt_set_retrans_timeout_def - set a request's retransmit timeout
400 * @task: task whose timeout is to be set
401 *
402 * Set a request's retransmit timeout based on the transport's
403 * default timeout parameters. Used by transports that don't adjust
404 * the retransmit timeout based on round-trip time estimation.
405 */
406void xprt_set_retrans_timeout_def(struct rpc_task *task)
407{
408 task->tk_timeout = task->tk_rqstp->rq_timeout;
352} 409}
353 410
354/* 411/*
355 * Reset the major timeout value 412 * xprt_set_retrans_timeout_rtt - set a request's retransmit timeout
413 * @task: task whose timeout is to be set
414 *
415 * Set a request's retransmit timeout using the RTT estimator.
356 */ 416 */
417void xprt_set_retrans_timeout_rtt(struct rpc_task *task)
418{
419 int timer = task->tk_msg.rpc_proc->p_timer;
420 struct rpc_rtt *rtt = task->tk_client->cl_rtt;
421 struct rpc_rqst *req = task->tk_rqstp;
422 unsigned long max_timeout = req->rq_xprt->timeout.to_maxval;
423
424 task->tk_timeout = rpc_calc_rto(rtt, timer);
425 task->tk_timeout <<= rpc_ntimeo(rtt, timer) + req->rq_retries;
426 if (task->tk_timeout > max_timeout || task->tk_timeout == 0)
427 task->tk_timeout = max_timeout;
428}
429
357static void xprt_reset_majortimeo(struct rpc_rqst *req) 430static void xprt_reset_majortimeo(struct rpc_rqst *req)
358{ 431{
359 struct rpc_timeout *to = &req->rq_xprt->timeout; 432 struct rpc_timeout *to = &req->rq_xprt->timeout;
@@ -368,8 +441,10 @@ static void xprt_reset_majortimeo(struct rpc_rqst *req)
368 req->rq_majortimeo += jiffies; 441 req->rq_majortimeo += jiffies;
369} 442}
370 443
371/* 444/**
372 * Adjust timeout values etc for next retransmit 445 * xprt_adjust_timeout - adjust timeout values for next retransmit
446 * @req: RPC request containing parameters to use for the adjustment
447 *
373 */ 448 */
374int xprt_adjust_timeout(struct rpc_rqst *req) 449int xprt_adjust_timeout(struct rpc_rqst *req)
375{ 450{
@@ -391,9 +466,9 @@ int xprt_adjust_timeout(struct rpc_rqst *req)
391 req->rq_retries = 0; 466 req->rq_retries = 0;
392 xprt_reset_majortimeo(req); 467 xprt_reset_majortimeo(req);
393 /* Reset the RTT counters == "slow start" */ 468 /* Reset the RTT counters == "slow start" */
394 spin_lock_bh(&xprt->sock_lock); 469 spin_lock_bh(&xprt->transport_lock);
395 rpc_init_rtt(req->rq_task->tk_client->cl_rtt, to->to_initval); 470 rpc_init_rtt(req->rq_task->tk_client->cl_rtt, to->to_initval);
396 spin_unlock_bh(&xprt->sock_lock); 471 spin_unlock_bh(&xprt->transport_lock);
397 pprintk("RPC: %lu timeout\n", jiffies); 472 pprintk("RPC: %lu timeout\n", jiffies);
398 status = -ETIMEDOUT; 473 status = -ETIMEDOUT;
399 } 474 }
@@ -405,133 +480,52 @@ int xprt_adjust_timeout(struct rpc_rqst *req)
405 return status; 480 return status;
406} 481}
407 482
408/* 483static void xprt_autoclose(void *args)
409 * Close down a transport socket
410 */
411static void
412xprt_close(struct rpc_xprt *xprt)
413{
414 struct socket *sock = xprt->sock;
415 struct sock *sk = xprt->inet;
416
417 if (!sk)
418 return;
419
420 write_lock_bh(&sk->sk_callback_lock);
421 xprt->inet = NULL;
422 xprt->sock = NULL;
423
424 sk->sk_user_data = NULL;
425 sk->sk_data_ready = xprt->old_data_ready;
426 sk->sk_state_change = xprt->old_state_change;
427 sk->sk_write_space = xprt->old_write_space;
428 write_unlock_bh(&sk->sk_callback_lock);
429
430 sk->sk_no_check = 0;
431
432 sock_release(sock);
433}
434
435static void
436xprt_socket_autoclose(void *args)
437{ 484{
438 struct rpc_xprt *xprt = (struct rpc_xprt *)args; 485 struct rpc_xprt *xprt = (struct rpc_xprt *)args;
439 486
440 xprt_disconnect(xprt); 487 xprt_disconnect(xprt);
441 xprt_close(xprt); 488 xprt->ops->close(xprt);
442 xprt_release_write(xprt, NULL); 489 xprt_release_write(xprt, NULL);
443} 490}
444 491
445/* 492/**
446 * Mark a transport as disconnected 493 * xprt_disconnect - mark a transport as disconnected
494 * @xprt: transport to flag for disconnect
495 *
447 */ 496 */
448static void 497void xprt_disconnect(struct rpc_xprt *xprt)
449xprt_disconnect(struct rpc_xprt *xprt)
450{ 498{
451 dprintk("RPC: disconnected transport %p\n", xprt); 499 dprintk("RPC: disconnected transport %p\n", xprt);
452 spin_lock_bh(&xprt->sock_lock); 500 spin_lock_bh(&xprt->transport_lock);
453 xprt_clear_connected(xprt); 501 xprt_clear_connected(xprt);
454 rpc_wake_up_status(&xprt->pending, -ENOTCONN); 502 xprt_wake_pending_tasks(xprt, -ENOTCONN);
455 spin_unlock_bh(&xprt->sock_lock); 503 spin_unlock_bh(&xprt->transport_lock);
456} 504}
457 505
458/*
459 * Used to allow disconnection when we've been idle
460 */
461static void 506static void
462xprt_init_autodisconnect(unsigned long data) 507xprt_init_autodisconnect(unsigned long data)
463{ 508{
464 struct rpc_xprt *xprt = (struct rpc_xprt *)data; 509 struct rpc_xprt *xprt = (struct rpc_xprt *)data;
465 510
466 spin_lock(&xprt->sock_lock); 511 spin_lock(&xprt->transport_lock);
467 if (!list_empty(&xprt->recv) || xprt->shutdown) 512 if (!list_empty(&xprt->recv) || xprt->shutdown)
468 goto out_abort; 513 goto out_abort;
469 if (test_and_set_bit(XPRT_LOCKED, &xprt->sockstate)) 514 if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
470 goto out_abort; 515 goto out_abort;
471 spin_unlock(&xprt->sock_lock); 516 spin_unlock(&xprt->transport_lock);
472 /* Let keventd close the socket */ 517 if (xprt_connecting(xprt))
473 if (test_bit(XPRT_CONNECTING, &xprt->sockstate) != 0)
474 xprt_release_write(xprt, NULL); 518 xprt_release_write(xprt, NULL);
475 else 519 else
476 schedule_work(&xprt->task_cleanup); 520 schedule_work(&xprt->task_cleanup);
477 return; 521 return;
478out_abort: 522out_abort:
479 spin_unlock(&xprt->sock_lock); 523 spin_unlock(&xprt->transport_lock);
480}
481
482static void xprt_socket_connect(void *args)
483{
484 struct rpc_xprt *xprt = (struct rpc_xprt *)args;
485 struct socket *sock = xprt->sock;
486 int status = -EIO;
487
488 if (xprt->shutdown || xprt->addr.sin_port == 0)
489 goto out;
490
491 /*
492 * Start by resetting any existing state
493 */
494 xprt_close(xprt);
495 sock = xprt_create_socket(xprt, xprt->prot, xprt->resvport);
496 if (sock == NULL) {
497 /* couldn't create socket or bind to reserved port;
498 * this is likely a permanent error, so cause an abort */
499 goto out;
500 }
501 xprt_bind_socket(xprt, sock);
502 xprt_sock_setbufsize(xprt);
503
504 status = 0;
505 if (!xprt->stream)
506 goto out;
507
508 /*
509 * Tell the socket layer to start connecting...
510 */
511 status = sock->ops->connect(sock, (struct sockaddr *) &xprt->addr,
512 sizeof(xprt->addr), O_NONBLOCK);
513 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
514 xprt, -status, xprt_connected(xprt), sock->sk->sk_state);
515 if (status < 0) {
516 switch (status) {
517 case -EINPROGRESS:
518 case -EALREADY:
519 goto out_clear;
520 }
521 }
522out:
523 if (status < 0)
524 rpc_wake_up_status(&xprt->pending, status);
525 else
526 rpc_wake_up(&xprt->pending);
527out_clear:
528 smp_mb__before_clear_bit();
529 clear_bit(XPRT_CONNECTING, &xprt->sockstate);
530 smp_mb__after_clear_bit();
531} 524}
532 525
533/* 526/**
534 * Attempt to connect a TCP socket. 527 * xprt_connect - schedule a transport connect operation
528 * @task: RPC task that is requesting the connect
535 * 529 *
536 */ 530 */
537void xprt_connect(struct rpc_task *task) 531void xprt_connect(struct rpc_task *task)
@@ -552,37 +546,19 @@ void xprt_connect(struct rpc_task *task)
552 if (!xprt_lock_write(xprt, task)) 546 if (!xprt_lock_write(xprt, task))
553 return; 547 return;
554 if (xprt_connected(xprt)) 548 if (xprt_connected(xprt))
555 goto out_write; 549 xprt_release_write(xprt, task);
550 else {
551 if (task->tk_rqstp)
552 task->tk_rqstp->rq_bytes_sent = 0;
556 553
557 if (task->tk_rqstp) 554 task->tk_timeout = xprt->connect_timeout;
558 task->tk_rqstp->rq_bytes_sent = 0; 555 rpc_sleep_on(&xprt->pending, task, xprt_connect_status, NULL);
559 556 xprt->ops->connect(task);
560 task->tk_timeout = RPC_CONNECT_TIMEOUT;
561 rpc_sleep_on(&xprt->pending, task, xprt_connect_status, NULL);
562 if (!test_and_set_bit(XPRT_CONNECTING, &xprt->sockstate)) {
563 /* Note: if we are here due to a dropped connection
564 * we delay reconnecting by RPC_REESTABLISH_TIMEOUT/HZ
565 * seconds
566 */
567 if (xprt->sock != NULL)
568 schedule_delayed_work(&xprt->sock_connect,
569 RPC_REESTABLISH_TIMEOUT);
570 else {
571 schedule_work(&xprt->sock_connect);
572 if (!RPC_IS_ASYNC(task))
573 flush_scheduled_work();
574 }
575 } 557 }
576 return; 558 return;
577 out_write:
578 xprt_release_write(xprt, task);
579} 559}
580 560
581/* 561static void xprt_connect_status(struct rpc_task *task)
582 * We arrive here when awoken from waiting on connection establishment.
583 */
584static void
585xprt_connect_status(struct rpc_task *task)
586{ 562{
587 struct rpc_xprt *xprt = task->tk_xprt; 563 struct rpc_xprt *xprt = task->tk_xprt;
588 564
@@ -592,31 +568,42 @@ xprt_connect_status(struct rpc_task *task)
592 return; 568 return;
593 } 569 }
594 570
595 /* if soft mounted, just cause this RPC to fail */
596 if (RPC_IS_SOFT(task))
597 task->tk_status = -EIO;
598
599 switch (task->tk_status) { 571 switch (task->tk_status) {
600 case -ECONNREFUSED: 572 case -ECONNREFUSED:
601 case -ECONNRESET: 573 case -ECONNRESET:
574 dprintk("RPC: %4d xprt_connect_status: server %s refused connection\n",
575 task->tk_pid, task->tk_client->cl_server);
576 break;
602 case -ENOTCONN: 577 case -ENOTCONN:
603 return; 578 dprintk("RPC: %4d xprt_connect_status: connection broken\n",
579 task->tk_pid);
580 break;
604 case -ETIMEDOUT: 581 case -ETIMEDOUT:
605 dprintk("RPC: %4d xprt_connect_status: timed out\n", 582 dprintk("RPC: %4d xprt_connect_status: connect attempt timed out\n",
606 task->tk_pid); 583 task->tk_pid);
607 break; 584 break;
608 default: 585 default:
609 printk(KERN_ERR "RPC: error %d connecting to server %s\n", 586 dprintk("RPC: %4d xprt_connect_status: error %d connecting to server %s\n",
610 -task->tk_status, task->tk_client->cl_server); 587 task->tk_pid, -task->tk_status, task->tk_client->cl_server);
588 xprt_release_write(xprt, task);
589 task->tk_status = -EIO;
590 return;
591 }
592
593 /* if soft mounted, just cause this RPC to fail */
594 if (RPC_IS_SOFT(task)) {
595 xprt_release_write(xprt, task);
596 task->tk_status = -EIO;
611 } 597 }
612 xprt_release_write(xprt, task);
613} 598}
614 599
615/* 600/**
616 * Look up the RPC request corresponding to a reply, and then lock it. 601 * xprt_lookup_rqst - find an RPC request corresponding to an XID
602 * @xprt: transport on which the original request was transmitted
603 * @xid: RPC XID of incoming reply
604 *
617 */ 605 */
618static inline struct rpc_rqst * 606struct rpc_rqst *xprt_lookup_rqst(struct rpc_xprt *xprt, u32 xid)
619xprt_lookup_rqst(struct rpc_xprt *xprt, u32 xid)
620{ 607{
621 struct list_head *pos; 608 struct list_head *pos;
622 struct rpc_rqst *req = NULL; 609 struct rpc_rqst *req = NULL;
@@ -631,556 +618,68 @@ xprt_lookup_rqst(struct rpc_xprt *xprt, u32 xid)
631 return req; 618 return req;
632} 619}
633 620
634/* 621/**
635 * Complete reply received. 622 * xprt_update_rtt - update an RPC client's RTT state after receiving a reply
636 * The TCP code relies on us to remove the request from xprt->pending. 623 * @task: RPC request that recently completed
637 */ 624 *
638static void
639xprt_complete_rqst(struct rpc_xprt *xprt, struct rpc_rqst *req, int copied)
640{
641 struct rpc_task *task = req->rq_task;
642 struct rpc_clnt *clnt = task->tk_client;
643
644 /* Adjust congestion window */
645 if (!xprt->nocong) {
646 unsigned timer = task->tk_msg.rpc_proc->p_timer;
647 xprt_adjust_cwnd(xprt, copied);
648 __xprt_put_cong(xprt, req);
649 if (timer) {
650 if (req->rq_ntrans == 1)
651 rpc_update_rtt(clnt->cl_rtt, timer,
652 (long)jiffies - req->rq_xtime);
653 rpc_set_timeo(clnt->cl_rtt, timer, req->rq_ntrans - 1);
654 }
655 }
656
657#ifdef RPC_PROFILE
658 /* Profile only reads for now */
659 if (copied > 1024) {
660 static unsigned long nextstat;
661 static unsigned long pkt_rtt, pkt_len, pkt_cnt;
662
663 pkt_cnt++;
664 pkt_len += req->rq_slen + copied;
665 pkt_rtt += jiffies - req->rq_xtime;
666 if (time_before(nextstat, jiffies)) {
667 printk("RPC: %lu %ld cwnd\n", jiffies, xprt->cwnd);
668 printk("RPC: %ld %ld %ld %ld stat\n",
669 jiffies, pkt_cnt, pkt_len, pkt_rtt);
670 pkt_rtt = pkt_len = pkt_cnt = 0;
671 nextstat = jiffies + 5 * HZ;
672 }
673 }
674#endif
675
676 dprintk("RPC: %4d has input (%d bytes)\n", task->tk_pid, copied);
677 list_del_init(&req->rq_list);
678 req->rq_received = req->rq_private_buf.len = copied;
679
680 /* ... and wake up the process. */
681 rpc_wake_up_task(task);
682 return;
683}
684
685static size_t
686skb_read_bits(skb_reader_t *desc, void *to, size_t len)
687{
688 if (len > desc->count)
689 len = desc->count;
690 if (skb_copy_bits(desc->skb, desc->offset, to, len))
691 return 0;
692 desc->count -= len;
693 desc->offset += len;
694 return len;
695}
696
697static size_t
698skb_read_and_csum_bits(skb_reader_t *desc, void *to, size_t len)
699{
700 unsigned int csum2, pos;
701
702 if (len > desc->count)
703 len = desc->count;
704 pos = desc->offset;
705 csum2 = skb_copy_and_csum_bits(desc->skb, pos, to, len, 0);
706 desc->csum = csum_block_add(desc->csum, csum2, pos);
707 desc->count -= len;
708 desc->offset += len;
709 return len;
710}
711
712/*
713 * We have set things up such that we perform the checksum of the UDP
714 * packet in parallel with the copies into the RPC client iovec. -DaveM
715 */
716int
717csum_partial_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb)
718{
719 skb_reader_t desc;
720
721 desc.skb = skb;
722 desc.offset = sizeof(struct udphdr);
723 desc.count = skb->len - desc.offset;
724
725 if (skb->ip_summed == CHECKSUM_UNNECESSARY)
726 goto no_checksum;
727
728 desc.csum = csum_partial(skb->data, desc.offset, skb->csum);
729 if (xdr_partial_copy_from_skb(xdr, 0, &desc, skb_read_and_csum_bits) < 0)
730 return -1;
731 if (desc.offset != skb->len) {
732 unsigned int csum2;
733 csum2 = skb_checksum(skb, desc.offset, skb->len - desc.offset, 0);
734 desc.csum = csum_block_add(desc.csum, csum2, desc.offset);
735 }
736 if (desc.count)
737 return -1;
738 if ((unsigned short)csum_fold(desc.csum))
739 return -1;
740 return 0;
741no_checksum:
742 if (xdr_partial_copy_from_skb(xdr, 0, &desc, skb_read_bits) < 0)
743 return -1;
744 if (desc.count)
745 return -1;
746 return 0;
747}
748
749/*
750 * Input handler for RPC replies. Called from a bottom half and hence
751 * atomic.
752 */
753static void
754udp_data_ready(struct sock *sk, int len)
755{
756 struct rpc_task *task;
757 struct rpc_xprt *xprt;
758 struct rpc_rqst *rovr;
759 struct sk_buff *skb;
760 int err, repsize, copied;
761 u32 _xid, *xp;
762
763 read_lock(&sk->sk_callback_lock);
764 dprintk("RPC: udp_data_ready...\n");
765 if (!(xprt = xprt_from_sock(sk))) {
766 printk("RPC: udp_data_ready request not found!\n");
767 goto out;
768 }
769
770 dprintk("RPC: udp_data_ready client %p\n", xprt);
771
772 if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
773 goto out;
774
775 if (xprt->shutdown)
776 goto dropit;
777
778 repsize = skb->len - sizeof(struct udphdr);
779 if (repsize < 4) {
780 printk("RPC: impossible RPC reply size %d!\n", repsize);
781 goto dropit;
782 }
783
784 /* Copy the XID from the skb... */
785 xp = skb_header_pointer(skb, sizeof(struct udphdr),
786 sizeof(_xid), &_xid);
787 if (xp == NULL)
788 goto dropit;
789
790 /* Look up and lock the request corresponding to the given XID */
791 spin_lock(&xprt->sock_lock);
792 rovr = xprt_lookup_rqst(xprt, *xp);
793 if (!rovr)
794 goto out_unlock;
795 task = rovr->rq_task;
796
797 dprintk("RPC: %4d received reply\n", task->tk_pid);
798
799 if ((copied = rovr->rq_private_buf.buflen) > repsize)
800 copied = repsize;
801
802 /* Suck it into the iovec, verify checksum if not done by hw. */
803 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb))
804 goto out_unlock;
805
806 /* Something worked... */
807 dst_confirm(skb->dst);
808
809 xprt_complete_rqst(xprt, rovr, copied);
810
811 out_unlock:
812 spin_unlock(&xprt->sock_lock);
813 dropit:
814 skb_free_datagram(sk, skb);
815 out:
816 read_unlock(&sk->sk_callback_lock);
817}
818
819/*
820 * Copy from an skb into memory and shrink the skb.
821 */
822static inline size_t
823tcp_copy_data(skb_reader_t *desc, void *p, size_t len)
824{
825 if (len > desc->count)
826 len = desc->count;
827 if (skb_copy_bits(desc->skb, desc->offset, p, len)) {
828 dprintk("RPC: failed to copy %zu bytes from skb. %zu bytes remain\n",
829 len, desc->count);
830 return 0;
831 }
832 desc->offset += len;
833 desc->count -= len;
834 dprintk("RPC: copied %zu bytes from skb. %zu bytes remain\n",
835 len, desc->count);
836 return len;
837}
838
839/*
840 * TCP read fragment marker
841 */
842static inline void
843tcp_read_fraghdr(struct rpc_xprt *xprt, skb_reader_t *desc)
844{
845 size_t len, used;
846 char *p;
847
848 p = ((char *) &xprt->tcp_recm) + xprt->tcp_offset;
849 len = sizeof(xprt->tcp_recm) - xprt->tcp_offset;
850 used = tcp_copy_data(desc, p, len);
851 xprt->tcp_offset += used;
852 if (used != len)
853 return;
854 xprt->tcp_reclen = ntohl(xprt->tcp_recm);
855 if (xprt->tcp_reclen & 0x80000000)
856 xprt->tcp_flags |= XPRT_LAST_FRAG;
857 else
858 xprt->tcp_flags &= ~XPRT_LAST_FRAG;
859 xprt->tcp_reclen &= 0x7fffffff;
860 xprt->tcp_flags &= ~XPRT_COPY_RECM;
861 xprt->tcp_offset = 0;
862 /* Sanity check of the record length */
863 if (xprt->tcp_reclen < 4) {
864 printk(KERN_ERR "RPC: Invalid TCP record fragment length\n");
865 xprt_disconnect(xprt);
866 }
867 dprintk("RPC: reading TCP record fragment of length %d\n",
868 xprt->tcp_reclen);
869}
870
871static void
872tcp_check_recm(struct rpc_xprt *xprt)
873{
874 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u, tcp_flags = %lx\n",
875 xprt, xprt->tcp_copied, xprt->tcp_offset, xprt->tcp_reclen, xprt->tcp_flags);
876 if (xprt->tcp_offset == xprt->tcp_reclen) {
877 xprt->tcp_flags |= XPRT_COPY_RECM;
878 xprt->tcp_offset = 0;
879 if (xprt->tcp_flags & XPRT_LAST_FRAG) {
880 xprt->tcp_flags &= ~XPRT_COPY_DATA;
881 xprt->tcp_flags |= XPRT_COPY_XID;
882 xprt->tcp_copied = 0;
883 }
884 }
885}
886
887/*
888 * TCP read xid
889 */
890static inline void
891tcp_read_xid(struct rpc_xprt *xprt, skb_reader_t *desc)
892{
893 size_t len, used;
894 char *p;
895
896 len = sizeof(xprt->tcp_xid) - xprt->tcp_offset;
897 dprintk("RPC: reading XID (%Zu bytes)\n", len);
898 p = ((char *) &xprt->tcp_xid) + xprt->tcp_offset;
899 used = tcp_copy_data(desc, p, len);
900 xprt->tcp_offset += used;
901 if (used != len)
902 return;
903 xprt->tcp_flags &= ~XPRT_COPY_XID;
904 xprt->tcp_flags |= XPRT_COPY_DATA;
905 xprt->tcp_copied = 4;
906 dprintk("RPC: reading reply for XID %08x\n",
907 ntohl(xprt->tcp_xid));
908 tcp_check_recm(xprt);
909}
910
911/*
912 * TCP read and complete request
913 */
914static inline void
915tcp_read_request(struct rpc_xprt *xprt, skb_reader_t *desc)
916{
917 struct rpc_rqst *req;
918 struct xdr_buf *rcvbuf;
919 size_t len;
920 ssize_t r;
921
922 /* Find and lock the request corresponding to this xid */
923 spin_lock(&xprt->sock_lock);
924 req = xprt_lookup_rqst(xprt, xprt->tcp_xid);
925 if (!req) {
926 xprt->tcp_flags &= ~XPRT_COPY_DATA;
927 dprintk("RPC: XID %08x request not found!\n",
928 ntohl(xprt->tcp_xid));
929 spin_unlock(&xprt->sock_lock);
930 return;
931 }
932
933 rcvbuf = &req->rq_private_buf;
934 len = desc->count;
935 if (len > xprt->tcp_reclen - xprt->tcp_offset) {
936 skb_reader_t my_desc;
937
938 len = xprt->tcp_reclen - xprt->tcp_offset;
939 memcpy(&my_desc, desc, sizeof(my_desc));
940 my_desc.count = len;
941 r = xdr_partial_copy_from_skb(rcvbuf, xprt->tcp_copied,
942 &my_desc, tcp_copy_data);
943 desc->count -= r;
944 desc->offset += r;
945 } else
946 r = xdr_partial_copy_from_skb(rcvbuf, xprt->tcp_copied,
947 desc, tcp_copy_data);
948
949 if (r > 0) {
950 xprt->tcp_copied += r;
951 xprt->tcp_offset += r;
952 }
953 if (r != len) {
954 /* Error when copying to the receive buffer,
955 * usually because we weren't able to allocate
956 * additional buffer pages. All we can do now
957 * is turn off XPRT_COPY_DATA, so the request
958 * will not receive any additional updates,
959 * and time out.
960 * Any remaining data from this record will
961 * be discarded.
962 */
963 xprt->tcp_flags &= ~XPRT_COPY_DATA;
964 dprintk("RPC: XID %08x truncated request\n",
965 ntohl(xprt->tcp_xid));
966 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u\n",
967 xprt, xprt->tcp_copied, xprt->tcp_offset, xprt->tcp_reclen);
968 goto out;
969 }
970
971 dprintk("RPC: XID %08x read %Zd bytes\n",
972 ntohl(xprt->tcp_xid), r);
973 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u\n",
974 xprt, xprt->tcp_copied, xprt->tcp_offset, xprt->tcp_reclen);
975
976 if (xprt->tcp_copied == req->rq_private_buf.buflen)
977 xprt->tcp_flags &= ~XPRT_COPY_DATA;
978 else if (xprt->tcp_offset == xprt->tcp_reclen) {
979 if (xprt->tcp_flags & XPRT_LAST_FRAG)
980 xprt->tcp_flags &= ~XPRT_COPY_DATA;
981 }
982
983out:
984 if (!(xprt->tcp_flags & XPRT_COPY_DATA)) {
985 dprintk("RPC: %4d received reply complete\n",
986 req->rq_task->tk_pid);
987 xprt_complete_rqst(xprt, req, xprt->tcp_copied);
988 }
989 spin_unlock(&xprt->sock_lock);
990 tcp_check_recm(xprt);
991}
992
993/*
994 * TCP discard extra bytes from a short read
995 */
996static inline void
997tcp_read_discard(struct rpc_xprt *xprt, skb_reader_t *desc)
998{
999 size_t len;
1000
1001 len = xprt->tcp_reclen - xprt->tcp_offset;
1002 if (len > desc->count)
1003 len = desc->count;
1004 desc->count -= len;
1005 desc->offset += len;
1006 xprt->tcp_offset += len;
1007 dprintk("RPC: discarded %Zu bytes\n", len);
1008 tcp_check_recm(xprt);
1009}
1010
1011/*
1012 * TCP record receive routine
1013 * We first have to grab the record marker, then the XID, then the data.
1014 */ 625 */
1015static int 626void xprt_update_rtt(struct rpc_task *task)
1016tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb,
1017 unsigned int offset, size_t len)
1018{
1019 struct rpc_xprt *xprt = rd_desc->arg.data;
1020 skb_reader_t desc = {
1021 .skb = skb,
1022 .offset = offset,
1023 .count = len,
1024 .csum = 0
1025 };
1026
1027 dprintk("RPC: tcp_data_recv\n");
1028 do {
1029 /* Read in a new fragment marker if necessary */
1030 /* Can we ever really expect to get completely empty fragments? */
1031 if (xprt->tcp_flags & XPRT_COPY_RECM) {
1032 tcp_read_fraghdr(xprt, &desc);
1033 continue;
1034 }
1035 /* Read in the xid if necessary */
1036 if (xprt->tcp_flags & XPRT_COPY_XID) {
1037 tcp_read_xid(xprt, &desc);
1038 continue;
1039 }
1040 /* Read in the request data */
1041 if (xprt->tcp_flags & XPRT_COPY_DATA) {
1042 tcp_read_request(xprt, &desc);
1043 continue;
1044 }
1045 /* Skip over any trailing bytes on short reads */
1046 tcp_read_discard(xprt, &desc);
1047 } while (desc.count);
1048 dprintk("RPC: tcp_data_recv done\n");
1049 return len - desc.count;
1050}
1051
1052static void tcp_data_ready(struct sock *sk, int bytes)
1053{ 627{
1054 struct rpc_xprt *xprt; 628 struct rpc_rqst *req = task->tk_rqstp;
1055 read_descriptor_t rd_desc; 629 struct rpc_rtt *rtt = task->tk_client->cl_rtt;
1056 630 unsigned timer = task->tk_msg.rpc_proc->p_timer;
1057 read_lock(&sk->sk_callback_lock);
1058 dprintk("RPC: tcp_data_ready...\n");
1059 if (!(xprt = xprt_from_sock(sk))) {
1060 printk("RPC: tcp_data_ready socket info not found!\n");
1061 goto out;
1062 }
1063 if (xprt->shutdown)
1064 goto out;
1065
1066 /* We use rd_desc to pass struct xprt to tcp_data_recv */
1067 rd_desc.arg.data = xprt;
1068 rd_desc.count = 65536;
1069 tcp_read_sock(sk, &rd_desc, tcp_data_recv);
1070out:
1071 read_unlock(&sk->sk_callback_lock);
1072}
1073
1074static void
1075tcp_state_change(struct sock *sk)
1076{
1077 struct rpc_xprt *xprt;
1078 631
1079 read_lock(&sk->sk_callback_lock); 632 if (timer) {
1080 if (!(xprt = xprt_from_sock(sk))) 633 if (req->rq_ntrans == 1)
1081 goto out; 634 rpc_update_rtt(rtt, timer,
1082 dprintk("RPC: tcp_state_change client %p...\n", xprt); 635 (long)jiffies - req->rq_xtime);
1083 dprintk("RPC: state %x conn %d dead %d zapped %d\n", 636 rpc_set_timeo(rtt, timer, req->rq_ntrans - 1);
1084 sk->sk_state, xprt_connected(xprt),
1085 sock_flag(sk, SOCK_DEAD),
1086 sock_flag(sk, SOCK_ZAPPED));
1087
1088 switch (sk->sk_state) {
1089 case TCP_ESTABLISHED:
1090 spin_lock_bh(&xprt->sock_lock);
1091 if (!xprt_test_and_set_connected(xprt)) {
1092 /* Reset TCP record info */
1093 xprt->tcp_offset = 0;
1094 xprt->tcp_reclen = 0;
1095 xprt->tcp_copied = 0;
1096 xprt->tcp_flags = XPRT_COPY_RECM | XPRT_COPY_XID;
1097 rpc_wake_up(&xprt->pending);
1098 }
1099 spin_unlock_bh(&xprt->sock_lock);
1100 break;
1101 case TCP_SYN_SENT:
1102 case TCP_SYN_RECV:
1103 break;
1104 default:
1105 xprt_disconnect(xprt);
1106 break;
1107 } 637 }
1108 out:
1109 read_unlock(&sk->sk_callback_lock);
1110} 638}
1111 639
1112/* 640/**
1113 * Called when more output buffer space is available for this socket. 641 * xprt_complete_rqst - called when reply processing is complete
1114 * We try not to wake our writers until they can make "significant" 642 * @task: RPC request that recently completed
1115 * progress, otherwise we'll waste resources thrashing sock_sendmsg 643 * @copied: actual number of bytes received from the transport
1116 * with a bunch of small requests. 644 *
645 * Caller holds transport lock.
1117 */ 646 */
1118static void 647void xprt_complete_rqst(struct rpc_task *task, int copied)
1119xprt_write_space(struct sock *sk)
1120{ 648{
1121 struct rpc_xprt *xprt; 649 struct rpc_rqst *req = task->tk_rqstp;
1122 struct socket *sock;
1123
1124 read_lock(&sk->sk_callback_lock);
1125 if (!(xprt = xprt_from_sock(sk)) || !(sock = sk->sk_socket))
1126 goto out;
1127 if (xprt->shutdown)
1128 goto out;
1129
1130 /* Wait until we have enough socket memory */
1131 if (xprt->stream) {
1132 /* from net/core/stream.c:sk_stream_write_space */
1133 if (sk_stream_wspace(sk) < sk_stream_min_wspace(sk))
1134 goto out;
1135 } else {
1136 /* from net/core/sock.c:sock_def_write_space */
1137 if (!sock_writeable(sk))
1138 goto out;
1139 }
1140 650
1141 if (!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)) 651 dprintk("RPC: %5u xid %08x complete (%d bytes received)\n",
1142 goto out; 652 task->tk_pid, ntohl(req->rq_xid), copied);
1143 653
1144 spin_lock_bh(&xprt->sock_lock); 654 list_del_init(&req->rq_list);
1145 if (xprt->snd_task) 655 req->rq_received = req->rq_private_buf.len = copied;
1146 rpc_wake_up_task(xprt->snd_task); 656 rpc_wake_up_task(task);
1147 spin_unlock_bh(&xprt->sock_lock);
1148out:
1149 read_unlock(&sk->sk_callback_lock);
1150} 657}
1151 658
1152/* 659static void xprt_timer(struct rpc_task *task)
1153 * RPC receive timeout handler.
1154 */
1155static void
1156xprt_timer(struct rpc_task *task)
1157{ 660{
1158 struct rpc_rqst *req = task->tk_rqstp; 661 struct rpc_rqst *req = task->tk_rqstp;
1159 struct rpc_xprt *xprt = req->rq_xprt; 662 struct rpc_xprt *xprt = req->rq_xprt;
1160 663
1161 spin_lock(&xprt->sock_lock); 664 dprintk("RPC: %4d xprt_timer\n", task->tk_pid);
1162 if (req->rq_received)
1163 goto out;
1164
1165 xprt_adjust_cwnd(req->rq_xprt, -ETIMEDOUT);
1166 __xprt_put_cong(xprt, req);
1167 665
1168 dprintk("RPC: %4d xprt_timer (%s request)\n", 666 spin_lock(&xprt->transport_lock);
1169 task->tk_pid, req ? "pending" : "backlogged"); 667 if (!req->rq_received) {
1170 668 if (xprt->ops->timer)
1171 task->tk_status = -ETIMEDOUT; 669 xprt->ops->timer(task);
1172out: 670 task->tk_status = -ETIMEDOUT;
671 }
1173 task->tk_timeout = 0; 672 task->tk_timeout = 0;
1174 rpc_wake_up_task(task); 673 rpc_wake_up_task(task);
1175 spin_unlock(&xprt->sock_lock); 674 spin_unlock(&xprt->transport_lock);
1176} 675}
1177 676
1178/* 677/**
1179 * Place the actual RPC call. 678 * xprt_prepare_transmit - reserve the transport before sending a request
1180 * We have to copy the iovec because sendmsg fiddles with its contents. 679 * @task: RPC task about to send a request
680 *
1181 */ 681 */
1182int 682int xprt_prepare_transmit(struct rpc_task *task)
1183xprt_prepare_transmit(struct rpc_task *task)
1184{ 683{
1185 struct rpc_rqst *req = task->tk_rqstp; 684 struct rpc_rqst *req = task->tk_rqstp;
1186 struct rpc_xprt *xprt = req->rq_xprt; 685 struct rpc_xprt *xprt = req->rq_xprt;
@@ -1191,12 +690,12 @@ xprt_prepare_transmit(struct rpc_task *task)
1191 if (xprt->shutdown) 690 if (xprt->shutdown)
1192 return -EIO; 691 return -EIO;
1193 692
1194 spin_lock_bh(&xprt->sock_lock); 693 spin_lock_bh(&xprt->transport_lock);
1195 if (req->rq_received && !req->rq_bytes_sent) { 694 if (req->rq_received && !req->rq_bytes_sent) {
1196 err = req->rq_received; 695 err = req->rq_received;
1197 goto out_unlock; 696 goto out_unlock;
1198 } 697 }
1199 if (!__xprt_lock_write(xprt, task)) { 698 if (!xprt->ops->reserve_xprt(task)) {
1200 err = -EAGAIN; 699 err = -EAGAIN;
1201 goto out_unlock; 700 goto out_unlock;
1202 } 701 }
@@ -1206,39 +705,42 @@ xprt_prepare_transmit(struct rpc_task *task)
1206 goto out_unlock; 705 goto out_unlock;
1207 } 706 }
1208out_unlock: 707out_unlock:
1209 spin_unlock_bh(&xprt->sock_lock); 708 spin_unlock_bh(&xprt->transport_lock);
1210 return err; 709 return err;
1211} 710}
1212 711
1213void 712void
1214xprt_transmit(struct rpc_task *task) 713xprt_abort_transmit(struct rpc_task *task)
714{
715 struct rpc_xprt *xprt = task->tk_xprt;
716
717 xprt_release_write(xprt, task);
718}
719
720/**
721 * xprt_transmit - send an RPC request on a transport
722 * @task: controlling RPC task
723 *
724 * We have to copy the iovec because sendmsg fiddles with its contents.
725 */
726void xprt_transmit(struct rpc_task *task)
1215{ 727{
1216 struct rpc_clnt *clnt = task->tk_client;
1217 struct rpc_rqst *req = task->tk_rqstp; 728 struct rpc_rqst *req = task->tk_rqstp;
1218 struct rpc_xprt *xprt = req->rq_xprt; 729 struct rpc_xprt *xprt = req->rq_xprt;
1219 int status, retry = 0; 730 int status;
1220
1221 731
1222 dprintk("RPC: %4d xprt_transmit(%u)\n", task->tk_pid, req->rq_slen); 732 dprintk("RPC: %4d xprt_transmit(%u)\n", task->tk_pid, req->rq_slen);
1223 733
1224 /* set up everything as needed. */
1225 /* Write the record marker */
1226 if (xprt->stream) {
1227 u32 *marker = req->rq_svec[0].iov_base;
1228
1229 *marker = htonl(0x80000000|(req->rq_slen-sizeof(*marker)));
1230 }
1231
1232 smp_rmb(); 734 smp_rmb();
1233 if (!req->rq_received) { 735 if (!req->rq_received) {
1234 if (list_empty(&req->rq_list)) { 736 if (list_empty(&req->rq_list)) {
1235 spin_lock_bh(&xprt->sock_lock); 737 spin_lock_bh(&xprt->transport_lock);
1236 /* Update the softirq receive buffer */ 738 /* Update the softirq receive buffer */
1237 memcpy(&req->rq_private_buf, &req->rq_rcv_buf, 739 memcpy(&req->rq_private_buf, &req->rq_rcv_buf,
1238 sizeof(req->rq_private_buf)); 740 sizeof(req->rq_private_buf));
1239 /* Add request to the receive list */ 741 /* Add request to the receive list */
1240 list_add_tail(&req->rq_list, &xprt->recv); 742 list_add_tail(&req->rq_list, &xprt->recv);
1241 spin_unlock_bh(&xprt->sock_lock); 743 spin_unlock_bh(&xprt->transport_lock);
1242 xprt_reset_majortimeo(req); 744 xprt_reset_majortimeo(req);
1243 /* Turn off autodisconnect */ 745 /* Turn off autodisconnect */
1244 del_singleshot_timer_sync(&xprt->timer); 746 del_singleshot_timer_sync(&xprt->timer);
@@ -1246,40 +748,19 @@ xprt_transmit(struct rpc_task *task)
1246 } else if (!req->rq_bytes_sent) 748 } else if (!req->rq_bytes_sent)
1247 return; 749 return;
1248 750
1249 /* Continue transmitting the packet/record. We must be careful 751 status = xprt->ops->send_request(task);
1250 * to cope with writespace callbacks arriving _after_ we have 752 if (status == 0) {
1251 * called xprt_sendmsg(). 753 dprintk("RPC: %4d xmit complete\n", task->tk_pid);
1252 */ 754 spin_lock_bh(&xprt->transport_lock);
1253 while (1) { 755 xprt->ops->set_retrans_timeout(task);
1254 req->rq_xtime = jiffies; 756 /* Don't race with disconnect */
1255 status = xprt_sendmsg(xprt, req); 757 if (!xprt_connected(xprt))
1256 758 task->tk_status = -ENOTCONN;
1257 if (status < 0) 759 else if (!req->rq_received)
1258 break; 760 rpc_sleep_on(&xprt->pending, task, NULL, xprt_timer);
1259 761 xprt->ops->release_xprt(xprt, task);
1260 if (xprt->stream) { 762 spin_unlock_bh(&xprt->transport_lock);
1261 req->rq_bytes_sent += status; 763 return;
1262
1263 /* If we've sent the entire packet, immediately
1264 * reset the count of bytes sent. */
1265 if (req->rq_bytes_sent >= req->rq_slen) {
1266 req->rq_bytes_sent = 0;
1267 goto out_receive;
1268 }
1269 } else {
1270 if (status >= req->rq_slen)
1271 goto out_receive;
1272 status = -EAGAIN;
1273 break;
1274 }
1275
1276 dprintk("RPC: %4d xmit incomplete (%d left of %d)\n",
1277 task->tk_pid, req->rq_slen - req->rq_bytes_sent,
1278 req->rq_slen);
1279
1280 status = -EAGAIN;
1281 if (retry++ > 50)
1282 break;
1283 } 764 }
1284 765
1285 /* Note: at this point, task->tk_sleeping has not yet been set, 766 /* Note: at this point, task->tk_sleeping has not yet been set,
@@ -1289,60 +770,19 @@ xprt_transmit(struct rpc_task *task)
1289 task->tk_status = status; 770 task->tk_status = status;
1290 771
1291 switch (status) { 772 switch (status) {
1292 case -EAGAIN:
1293 if (test_bit(SOCK_ASYNC_NOSPACE, &xprt->sock->flags)) {
1294 /* Protect against races with xprt_write_space */
1295 spin_lock_bh(&xprt->sock_lock);
1296 /* Don't race with disconnect */
1297 if (!xprt_connected(xprt))
1298 task->tk_status = -ENOTCONN;
1299 else if (test_bit(SOCK_NOSPACE, &xprt->sock->flags)) {
1300 task->tk_timeout = req->rq_timeout;
1301 rpc_sleep_on(&xprt->pending, task, NULL, NULL);
1302 }
1303 spin_unlock_bh(&xprt->sock_lock);
1304 return;
1305 }
1306 /* Keep holding the socket if it is blocked */
1307 rpc_delay(task, HZ>>4);
1308 return;
1309 case -ECONNREFUSED: 773 case -ECONNREFUSED:
1310 task->tk_timeout = RPC_REESTABLISH_TIMEOUT;
1311 rpc_sleep_on(&xprt->sending, task, NULL, NULL); 774 rpc_sleep_on(&xprt->sending, task, NULL, NULL);
775 case -EAGAIN:
1312 case -ENOTCONN: 776 case -ENOTCONN:
1313 return; 777 return;
1314 default: 778 default:
1315 if (xprt->stream) 779 break;
1316 xprt_disconnect(xprt);
1317 } 780 }
1318 xprt_release_write(xprt, task); 781 xprt_release_write(xprt, task);
1319 return; 782 return;
1320 out_receive:
1321 dprintk("RPC: %4d xmit complete\n", task->tk_pid);
1322 /* Set the task's receive timeout value */
1323 spin_lock_bh(&xprt->sock_lock);
1324 if (!xprt->nocong) {
1325 int timer = task->tk_msg.rpc_proc->p_timer;
1326 task->tk_timeout = rpc_calc_rto(clnt->cl_rtt, timer);
1327 task->tk_timeout <<= rpc_ntimeo(clnt->cl_rtt, timer) + req->rq_retries;
1328 if (task->tk_timeout > xprt->timeout.to_maxval || task->tk_timeout == 0)
1329 task->tk_timeout = xprt->timeout.to_maxval;
1330 } else
1331 task->tk_timeout = req->rq_timeout;
1332 /* Don't race with disconnect */
1333 if (!xprt_connected(xprt))
1334 task->tk_status = -ENOTCONN;
1335 else if (!req->rq_received)
1336 rpc_sleep_on(&xprt->pending, task, NULL, xprt_timer);
1337 __xprt_release_write(xprt, task);
1338 spin_unlock_bh(&xprt->sock_lock);
1339} 783}
1340 784
1341/* 785static inline void do_xprt_reserve(struct rpc_task *task)
1342 * Reserve an RPC call slot.
1343 */
1344static inline void
1345do_xprt_reserve(struct rpc_task *task)
1346{ 786{
1347 struct rpc_xprt *xprt = task->tk_xprt; 787 struct rpc_xprt *xprt = task->tk_xprt;
1348 788
@@ -1362,22 +802,25 @@ do_xprt_reserve(struct rpc_task *task)
1362 rpc_sleep_on(&xprt->backlog, task, NULL, NULL); 802 rpc_sleep_on(&xprt->backlog, task, NULL, NULL);
1363} 803}
1364 804
1365void 805/**
1366xprt_reserve(struct rpc_task *task) 806 * xprt_reserve - allocate an RPC request slot
807 * @task: RPC task requesting a slot allocation
808 *
809 * If no more slots are available, place the task on the transport's
810 * backlog queue.
811 */
812void xprt_reserve(struct rpc_task *task)
1367{ 813{
1368 struct rpc_xprt *xprt = task->tk_xprt; 814 struct rpc_xprt *xprt = task->tk_xprt;
1369 815
1370 task->tk_status = -EIO; 816 task->tk_status = -EIO;
1371 if (!xprt->shutdown) { 817 if (!xprt->shutdown) {
1372 spin_lock(&xprt->xprt_lock); 818 spin_lock(&xprt->reserve_lock);
1373 do_xprt_reserve(task); 819 do_xprt_reserve(task);
1374 spin_unlock(&xprt->xprt_lock); 820 spin_unlock(&xprt->reserve_lock);
1375 } 821 }
1376} 822}
1377 823
1378/*
1379 * Allocate a 'unique' XID
1380 */
1381static inline u32 xprt_alloc_xid(struct rpc_xprt *xprt) 824static inline u32 xprt_alloc_xid(struct rpc_xprt *xprt)
1382{ 825{
1383 return xprt->xid++; 826 return xprt->xid++;
@@ -1388,11 +831,7 @@ static inline void xprt_init_xid(struct rpc_xprt *xprt)
1388 get_random_bytes(&xprt->xid, sizeof(xprt->xid)); 831 get_random_bytes(&xprt->xid, sizeof(xprt->xid));
1389} 832}
1390 833
1391/* 834static void xprt_request_init(struct rpc_task *task, struct rpc_xprt *xprt)
1392 * Initialize RPC request
1393 */
1394static void
1395xprt_request_init(struct rpc_task *task, struct rpc_xprt *xprt)
1396{ 835{
1397 struct rpc_rqst *req = task->tk_rqstp; 836 struct rpc_rqst *req = task->tk_rqstp;
1398 837
@@ -1400,128 +839,104 @@ xprt_request_init(struct rpc_task *task, struct rpc_xprt *xprt)
1400 req->rq_task = task; 839 req->rq_task = task;
1401 req->rq_xprt = xprt; 840 req->rq_xprt = xprt;
1402 req->rq_xid = xprt_alloc_xid(xprt); 841 req->rq_xid = xprt_alloc_xid(xprt);
842 req->rq_release_snd_buf = NULL;
1403 dprintk("RPC: %4d reserved req %p xid %08x\n", task->tk_pid, 843 dprintk("RPC: %4d reserved req %p xid %08x\n", task->tk_pid,
1404 req, ntohl(req->rq_xid)); 844 req, ntohl(req->rq_xid));
1405} 845}
1406 846
1407/* 847/**
1408 * Release an RPC call slot 848 * xprt_release - release an RPC request slot
849 * @task: task which is finished with the slot
850 *
1409 */ 851 */
1410void 852void xprt_release(struct rpc_task *task)
1411xprt_release(struct rpc_task *task)
1412{ 853{
1413 struct rpc_xprt *xprt = task->tk_xprt; 854 struct rpc_xprt *xprt = task->tk_xprt;
1414 struct rpc_rqst *req; 855 struct rpc_rqst *req;
1415 856
1416 if (!(req = task->tk_rqstp)) 857 if (!(req = task->tk_rqstp))
1417 return; 858 return;
1418 spin_lock_bh(&xprt->sock_lock); 859 spin_lock_bh(&xprt->transport_lock);
1419 __xprt_release_write(xprt, task); 860 xprt->ops->release_xprt(xprt, task);
1420 __xprt_put_cong(xprt, req); 861 if (xprt->ops->release_request)
862 xprt->ops->release_request(task);
1421 if (!list_empty(&req->rq_list)) 863 if (!list_empty(&req->rq_list))
1422 list_del(&req->rq_list); 864 list_del(&req->rq_list);
1423 xprt->last_used = jiffies; 865 xprt->last_used = jiffies;
1424 if (list_empty(&xprt->recv) && !xprt->shutdown) 866 if (list_empty(&xprt->recv) && !xprt->shutdown)
1425 mod_timer(&xprt->timer, xprt->last_used + XPRT_IDLE_TIMEOUT); 867 mod_timer(&xprt->timer,
1426 spin_unlock_bh(&xprt->sock_lock); 868 xprt->last_used + xprt->idle_timeout);
869 spin_unlock_bh(&xprt->transport_lock);
1427 task->tk_rqstp = NULL; 870 task->tk_rqstp = NULL;
871 if (req->rq_release_snd_buf)
872 req->rq_release_snd_buf(req);
1428 memset(req, 0, sizeof(*req)); /* mark unused */ 873 memset(req, 0, sizeof(*req)); /* mark unused */
1429 874
1430 dprintk("RPC: %4d release request %p\n", task->tk_pid, req); 875 dprintk("RPC: %4d release request %p\n", task->tk_pid, req);
1431 876
1432 spin_lock(&xprt->xprt_lock); 877 spin_lock(&xprt->reserve_lock);
1433 list_add(&req->rq_list, &xprt->free); 878 list_add(&req->rq_list, &xprt->free);
1434 xprt_clear_backlog(xprt); 879 rpc_wake_up_next(&xprt->backlog);
1435 spin_unlock(&xprt->xprt_lock); 880 spin_unlock(&xprt->reserve_lock);
1436}
1437
1438/*
1439 * Set default timeout parameters
1440 */
1441static void
1442xprt_default_timeout(struct rpc_timeout *to, int proto)
1443{
1444 if (proto == IPPROTO_UDP)
1445 xprt_set_timeout(to, 5, 5 * HZ);
1446 else
1447 xprt_set_timeout(to, 5, 60 * HZ);
1448} 881}
1449 882
1450/* 883/**
1451 * Set constant timeout 884 * xprt_set_timeout - set constant RPC timeout
885 * @to: RPC timeout parameters to set up
886 * @retr: number of retries
887 * @incr: amount of increase after each retry
888 *
1452 */ 889 */
1453void 890void xprt_set_timeout(struct rpc_timeout *to, unsigned int retr, unsigned long incr)
1454xprt_set_timeout(struct rpc_timeout *to, unsigned int retr, unsigned long incr)
1455{ 891{
1456 to->to_initval = 892 to->to_initval =
1457 to->to_increment = incr; 893 to->to_increment = incr;
1458 to->to_maxval = incr * retr; 894 to->to_maxval = to->to_initval + (incr * retr);
1459 to->to_retries = retr; 895 to->to_retries = retr;
1460 to->to_exponential = 0; 896 to->to_exponential = 0;
1461} 897}
1462 898
1463unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE; 899static struct rpc_xprt *xprt_setup(int proto, struct sockaddr_in *ap, struct rpc_timeout *to)
1464unsigned int xprt_tcp_slot_table_entries = RPC_DEF_SLOT_TABLE;
1465
1466/*
1467 * Initialize an RPC client
1468 */
1469static struct rpc_xprt *
1470xprt_setup(int proto, struct sockaddr_in *ap, struct rpc_timeout *to)
1471{ 900{
901 int result;
1472 struct rpc_xprt *xprt; 902 struct rpc_xprt *xprt;
1473 unsigned int entries;
1474 size_t slot_table_size;
1475 struct rpc_rqst *req; 903 struct rpc_rqst *req;
1476 904
1477 dprintk("RPC: setting up %s transport...\n",
1478 proto == IPPROTO_UDP? "UDP" : "TCP");
1479
1480 entries = (proto == IPPROTO_TCP)?
1481 xprt_tcp_slot_table_entries : xprt_udp_slot_table_entries;
1482
1483 if ((xprt = kmalloc(sizeof(struct rpc_xprt), GFP_KERNEL)) == NULL) 905 if ((xprt = kmalloc(sizeof(struct rpc_xprt), GFP_KERNEL)) == NULL)
1484 return ERR_PTR(-ENOMEM); 906 return ERR_PTR(-ENOMEM);
1485 memset(xprt, 0, sizeof(*xprt)); /* Nnnngh! */ 907 memset(xprt, 0, sizeof(*xprt)); /* Nnnngh! */
1486 xprt->max_reqs = entries;
1487 slot_table_size = entries * sizeof(xprt->slot[0]);
1488 xprt->slot = kmalloc(slot_table_size, GFP_KERNEL);
1489 if (xprt->slot == NULL) {
1490 kfree(xprt);
1491 return ERR_PTR(-ENOMEM);
1492 }
1493 memset(xprt->slot, 0, slot_table_size);
1494 908
1495 xprt->addr = *ap; 909 xprt->addr = *ap;
1496 xprt->prot = proto; 910
1497 xprt->stream = (proto == IPPROTO_TCP)? 1 : 0; 911 switch (proto) {
1498 if (xprt->stream) { 912 case IPPROTO_UDP:
1499 xprt->cwnd = RPC_MAXCWND(xprt); 913 result = xs_setup_udp(xprt, to);
1500 xprt->nocong = 1; 914 break;
1501 xprt->max_payload = (1U << 31) - 1; 915 case IPPROTO_TCP:
1502 } else { 916 result = xs_setup_tcp(xprt, to);
1503 xprt->cwnd = RPC_INITCWND; 917 break;
1504 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3); 918 default:
919 printk(KERN_ERR "RPC: unrecognized transport protocol: %d\n",
920 proto);
921 result = -EIO;
922 break;
923 }
924 if (result) {
925 kfree(xprt);
926 return ERR_PTR(result);
1505 } 927 }
1506 spin_lock_init(&xprt->sock_lock); 928
1507 spin_lock_init(&xprt->xprt_lock); 929 spin_lock_init(&xprt->transport_lock);
1508 init_waitqueue_head(&xprt->cong_wait); 930 spin_lock_init(&xprt->reserve_lock);
1509 931
1510 INIT_LIST_HEAD(&xprt->free); 932 INIT_LIST_HEAD(&xprt->free);
1511 INIT_LIST_HEAD(&xprt->recv); 933 INIT_LIST_HEAD(&xprt->recv);
1512 INIT_WORK(&xprt->sock_connect, xprt_socket_connect, xprt); 934 INIT_WORK(&xprt->task_cleanup, xprt_autoclose, xprt);
1513 INIT_WORK(&xprt->task_cleanup, xprt_socket_autoclose, xprt);
1514 init_timer(&xprt->timer); 935 init_timer(&xprt->timer);
1515 xprt->timer.function = xprt_init_autodisconnect; 936 xprt->timer.function = xprt_init_autodisconnect;
1516 xprt->timer.data = (unsigned long) xprt; 937 xprt->timer.data = (unsigned long) xprt;
1517 xprt->last_used = jiffies; 938 xprt->last_used = jiffies;
1518 xprt->port = XPRT_MAX_RESVPORT; 939 xprt->cwnd = RPC_INITCWND;
1519
1520 /* Set timeout parameters */
1521 if (to) {
1522 xprt->timeout = *to;
1523 } else
1524 xprt_default_timeout(&xprt->timeout, xprt->prot);
1525 940
1526 rpc_init_wait_queue(&xprt->pending, "xprt_pending"); 941 rpc_init_wait_queue(&xprt->pending, "xprt_pending");
1527 rpc_init_wait_queue(&xprt->sending, "xprt_sending"); 942 rpc_init_wait_queue(&xprt->sending, "xprt_sending");
@@ -1529,139 +944,25 @@ xprt_setup(int proto, struct sockaddr_in *ap, struct rpc_timeout *to)
1529 rpc_init_priority_wait_queue(&xprt->backlog, "xprt_backlog"); 944 rpc_init_priority_wait_queue(&xprt->backlog, "xprt_backlog");
1530 945
1531 /* initialize free list */ 946 /* initialize free list */
1532 for (req = &xprt->slot[entries-1]; req >= &xprt->slot[0]; req--) 947 for (req = &xprt->slot[xprt->max_reqs-1]; req >= &xprt->slot[0]; req--)
1533 list_add(&req->rq_list, &xprt->free); 948 list_add(&req->rq_list, &xprt->free);
1534 949
1535 xprt_init_xid(xprt); 950 xprt_init_xid(xprt);
1536 951
1537 /* Check whether we want to use a reserved port */
1538 xprt->resvport = capable(CAP_NET_BIND_SERVICE) ? 1 : 0;
1539
1540 dprintk("RPC: created transport %p with %u slots\n", xprt, 952 dprintk("RPC: created transport %p with %u slots\n", xprt,
1541 xprt->max_reqs); 953 xprt->max_reqs);
1542 954
1543 return xprt; 955 return xprt;
1544} 956}
1545 957
1546/* 958/**
1547 * Bind to a reserved port 959 * xprt_create_proto - create an RPC client transport
1548 */ 960 * @proto: requested transport protocol
1549static inline int xprt_bindresvport(struct rpc_xprt *xprt, struct socket *sock) 961 * @sap: remote peer's address
1550{ 962 * @to: timeout parameters for new transport
1551 struct sockaddr_in myaddr = { 963 *
1552 .sin_family = AF_INET,
1553 };
1554 int err, port;
1555
1556 /* Were we already bound to a given port? Try to reuse it */
1557 port = xprt->port;
1558 do {
1559 myaddr.sin_port = htons(port);
1560 err = sock->ops->bind(sock, (struct sockaddr *) &myaddr,
1561 sizeof(myaddr));
1562 if (err == 0) {
1563 xprt->port = port;
1564 return 0;
1565 }
1566 if (--port == 0)
1567 port = XPRT_MAX_RESVPORT;
1568 } while (err == -EADDRINUSE && port != xprt->port);
1569
1570 printk("RPC: Can't bind to reserved port (%d).\n", -err);
1571 return err;
1572}
1573
1574static void
1575xprt_bind_socket(struct rpc_xprt *xprt, struct socket *sock)
1576{
1577 struct sock *sk = sock->sk;
1578
1579 if (xprt->inet)
1580 return;
1581
1582 write_lock_bh(&sk->sk_callback_lock);
1583 sk->sk_user_data = xprt;
1584 xprt->old_data_ready = sk->sk_data_ready;
1585 xprt->old_state_change = sk->sk_state_change;
1586 xprt->old_write_space = sk->sk_write_space;
1587 if (xprt->prot == IPPROTO_UDP) {
1588 sk->sk_data_ready = udp_data_ready;
1589 sk->sk_no_check = UDP_CSUM_NORCV;
1590 xprt_set_connected(xprt);
1591 } else {
1592 tcp_sk(sk)->nonagle = 1; /* disable Nagle's algorithm */
1593 sk->sk_data_ready = tcp_data_ready;
1594 sk->sk_state_change = tcp_state_change;
1595 xprt_clear_connected(xprt);
1596 }
1597 sk->sk_write_space = xprt_write_space;
1598
1599 /* Reset to new socket */
1600 xprt->sock = sock;
1601 xprt->inet = sk;
1602 write_unlock_bh(&sk->sk_callback_lock);
1603
1604 return;
1605}
1606
1607/*
1608 * Set socket buffer length
1609 */
1610void
1611xprt_sock_setbufsize(struct rpc_xprt *xprt)
1612{
1613 struct sock *sk = xprt->inet;
1614
1615 if (xprt->stream)
1616 return;
1617 if (xprt->rcvsize) {
1618 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1619 sk->sk_rcvbuf = xprt->rcvsize * xprt->max_reqs * 2;
1620 }
1621 if (xprt->sndsize) {
1622 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1623 sk->sk_sndbuf = xprt->sndsize * xprt->max_reqs * 2;
1624 sk->sk_write_space(sk);
1625 }
1626}
1627
1628/*
1629 * Datastream sockets are created here, but xprt_connect will create
1630 * and connect stream sockets.
1631 */
1632static struct socket * xprt_create_socket(struct rpc_xprt *xprt, int proto, int resvport)
1633{
1634 struct socket *sock;
1635 int type, err;
1636
1637 dprintk("RPC: xprt_create_socket(%s %d)\n",
1638 (proto == IPPROTO_UDP)? "udp" : "tcp", proto);
1639
1640 type = (proto == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
1641
1642 if ((err = sock_create_kern(PF_INET, type, proto, &sock)) < 0) {
1643 printk("RPC: can't create socket (%d).\n", -err);
1644 return NULL;
1645 }
1646
1647 /* If the caller has the capability, bind to a reserved port */
1648 if (resvport && xprt_bindresvport(xprt, sock) < 0) {
1649 printk("RPC: can't bind to reserved port.\n");
1650 goto failed;
1651 }
1652
1653 return sock;
1654
1655failed:
1656 sock_release(sock);
1657 return NULL;
1658}
1659
1660/*
1661 * Create an RPC client transport given the protocol and peer address.
1662 */ 964 */
1663struct rpc_xprt * 965struct rpc_xprt *xprt_create_proto(int proto, struct sockaddr_in *sap, struct rpc_timeout *to)
1664xprt_create_proto(int proto, struct sockaddr_in *sap, struct rpc_timeout *to)
1665{ 966{
1666 struct rpc_xprt *xprt; 967 struct rpc_xprt *xprt;
1667 968
@@ -1673,46 +974,26 @@ xprt_create_proto(int proto, struct sockaddr_in *sap, struct rpc_timeout *to)
1673 return xprt; 974 return xprt;
1674} 975}
1675 976
1676/* 977static void xprt_shutdown(struct rpc_xprt *xprt)
1677 * Prepare for transport shutdown.
1678 */
1679static void
1680xprt_shutdown(struct rpc_xprt *xprt)
1681{ 978{
1682 xprt->shutdown = 1; 979 xprt->shutdown = 1;
1683 rpc_wake_up(&xprt->sending); 980 rpc_wake_up(&xprt->sending);
1684 rpc_wake_up(&xprt->resend); 981 rpc_wake_up(&xprt->resend);
1685 rpc_wake_up(&xprt->pending); 982 xprt_wake_pending_tasks(xprt, -EIO);
1686 rpc_wake_up(&xprt->backlog); 983 rpc_wake_up(&xprt->backlog);
1687 wake_up(&xprt->cong_wait);
1688 del_timer_sync(&xprt->timer); 984 del_timer_sync(&xprt->timer);
1689
1690 /* synchronously wait for connect worker to finish */
1691 cancel_delayed_work(&xprt->sock_connect);
1692 flush_scheduled_work();
1693} 985}
1694 986
1695/* 987/**
1696 * Clear the xprt backlog queue 988 * xprt_destroy - destroy an RPC transport, killing off all requests.
1697 */ 989 * @xprt: transport to destroy
1698static int 990 *
1699xprt_clear_backlog(struct rpc_xprt *xprt) {
1700 rpc_wake_up_next(&xprt->backlog);
1701 wake_up(&xprt->cong_wait);
1702 return 1;
1703}
1704
1705/*
1706 * Destroy an RPC transport, killing off all requests.
1707 */ 991 */
1708int 992int xprt_destroy(struct rpc_xprt *xprt)
1709xprt_destroy(struct rpc_xprt *xprt)
1710{ 993{
1711 dprintk("RPC: destroying transport %p\n", xprt); 994 dprintk("RPC: destroying transport %p\n", xprt);
1712 xprt_shutdown(xprt); 995 xprt_shutdown(xprt);
1713 xprt_disconnect(xprt); 996 xprt->ops->destroy(xprt);
1714 xprt_close(xprt);
1715 kfree(xprt->slot);
1716 kfree(xprt); 997 kfree(xprt);
1717 998
1718 return 0; 999 return 0;
diff --git a/net/sunrpc/xprtsock.c b/net/sunrpc/xprtsock.c
new file mode 100644
index 000000000000..2e1529217e65
--- /dev/null
+++ b/net/sunrpc/xprtsock.c
@@ -0,0 +1,1252 @@
1/*
2 * linux/net/sunrpc/xprtsock.c
3 *
4 * Client-side transport implementation for sockets.
5 *
6 * TCP callback races fixes (C) 1998 Red Hat Software <alan@redhat.com>
7 * TCP send fixes (C) 1998 Red Hat Software <alan@redhat.com>
8 * TCP NFS related read + write fixes
9 * (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
10 *
11 * Rewrite of larges part of the code in order to stabilize TCP stuff.
12 * Fix behaviour when socket buffer is full.
13 * (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
14 *
15 * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
16 */
17
18#include <linux/types.h>
19#include <linux/slab.h>
20#include <linux/capability.h>
21#include <linux/sched.h>
22#include <linux/pagemap.h>
23#include <linux/errno.h>
24#include <linux/socket.h>
25#include <linux/in.h>
26#include <linux/net.h>
27#include <linux/mm.h>
28#include <linux/udp.h>
29#include <linux/tcp.h>
30#include <linux/sunrpc/clnt.h>
31#include <linux/file.h>
32
33#include <net/sock.h>
34#include <net/checksum.h>
35#include <net/udp.h>
36#include <net/tcp.h>
37
38/*
39 * How many times to try sending a request on a socket before waiting
40 * for the socket buffer to clear.
41 */
42#define XS_SENDMSG_RETRY (10U)
43
44/*
45 * Time out for an RPC UDP socket connect. UDP socket connects are
46 * synchronous, but we set a timeout anyway in case of resource
47 * exhaustion on the local host.
48 */
49#define XS_UDP_CONN_TO (5U * HZ)
50
51/*
52 * Wait duration for an RPC TCP connection to be established. Solaris
53 * NFS over TCP uses 60 seconds, for example, which is in line with how
54 * long a server takes to reboot.
55 */
56#define XS_TCP_CONN_TO (60U * HZ)
57
58/*
59 * Wait duration for a reply from the RPC portmapper.
60 */
61#define XS_BIND_TO (60U * HZ)
62
63/*
64 * Delay if a UDP socket connect error occurs. This is most likely some
65 * kind of resource problem on the local host.
66 */
67#define XS_UDP_REEST_TO (2U * HZ)
68
69/*
70 * The reestablish timeout allows clients to delay for a bit before attempting
71 * to reconnect to a server that just dropped our connection.
72 *
73 * We implement an exponential backoff when trying to reestablish a TCP
74 * transport connection with the server. Some servers like to drop a TCP
75 * connection when they are overworked, so we start with a short timeout and
76 * increase over time if the server is down or not responding.
77 */
78#define XS_TCP_INIT_REEST_TO (3U * HZ)
79#define XS_TCP_MAX_REEST_TO (5U * 60 * HZ)
80
81/*
82 * TCP idle timeout; client drops the transport socket if it is idle
83 * for this long. Note that we also timeout UDP sockets to prevent
84 * holding port numbers when there is no RPC traffic.
85 */
86#define XS_IDLE_DISC_TO (5U * 60 * HZ)
87
88#ifdef RPC_DEBUG
89# undef RPC_DEBUG_DATA
90# define RPCDBG_FACILITY RPCDBG_TRANS
91#endif
92
93#ifdef RPC_DEBUG_DATA
94static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
95{
96 u8 *buf = (u8 *) packet;
97 int j;
98
99 dprintk("RPC: %s\n", msg);
100 for (j = 0; j < count && j < 128; j += 4) {
101 if (!(j & 31)) {
102 if (j)
103 dprintk("\n");
104 dprintk("0x%04x ", j);
105 }
106 dprintk("%02x%02x%02x%02x ",
107 buf[j], buf[j+1], buf[j+2], buf[j+3]);
108 }
109 dprintk("\n");
110}
111#else
112static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
113{
114 /* NOP */
115}
116#endif
117
118#define XS_SENDMSG_FLAGS (MSG_DONTWAIT | MSG_NOSIGNAL)
119
120static inline int xs_send_head(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base, unsigned int len)
121{
122 struct kvec iov = {
123 .iov_base = xdr->head[0].iov_base + base,
124 .iov_len = len - base,
125 };
126 struct msghdr msg = {
127 .msg_name = addr,
128 .msg_namelen = addrlen,
129 .msg_flags = XS_SENDMSG_FLAGS,
130 };
131
132 if (xdr->len > len)
133 msg.msg_flags |= MSG_MORE;
134
135 if (likely(iov.iov_len))
136 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
137 return kernel_sendmsg(sock, &msg, NULL, 0, 0);
138}
139
140static int xs_send_tail(struct socket *sock, struct xdr_buf *xdr, unsigned int base, unsigned int len)
141{
142 struct kvec iov = {
143 .iov_base = xdr->tail[0].iov_base + base,
144 .iov_len = len - base,
145 };
146 struct msghdr msg = {
147 .msg_flags = XS_SENDMSG_FLAGS,
148 };
149
150 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
151}
152
153/**
154 * xs_sendpages - write pages directly to a socket
155 * @sock: socket to send on
156 * @addr: UDP only -- address of destination
157 * @addrlen: UDP only -- length of destination address
158 * @xdr: buffer containing this request
159 * @base: starting position in the buffer
160 *
161 */
162static inline int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base)
163{
164 struct page **ppage = xdr->pages;
165 unsigned int len, pglen = xdr->page_len;
166 int err, ret = 0;
167 ssize_t (*sendpage)(struct socket *, struct page *, int, size_t, int);
168
169 if (unlikely(!sock))
170 return -ENOTCONN;
171
172 clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
173
174 len = xdr->head[0].iov_len;
175 if (base < len || (addr != NULL && base == 0)) {
176 err = xs_send_head(sock, addr, addrlen, xdr, base, len);
177 if (ret == 0)
178 ret = err;
179 else if (err > 0)
180 ret += err;
181 if (err != (len - base))
182 goto out;
183 base = 0;
184 } else
185 base -= len;
186
187 if (unlikely(pglen == 0))
188 goto copy_tail;
189 if (unlikely(base >= pglen)) {
190 base -= pglen;
191 goto copy_tail;
192 }
193 if (base || xdr->page_base) {
194 pglen -= base;
195 base += xdr->page_base;
196 ppage += base >> PAGE_CACHE_SHIFT;
197 base &= ~PAGE_CACHE_MASK;
198 }
199
200 sendpage = sock->ops->sendpage ? : sock_no_sendpage;
201 do {
202 int flags = XS_SENDMSG_FLAGS;
203
204 len = PAGE_CACHE_SIZE;
205 if (base)
206 len -= base;
207 if (pglen < len)
208 len = pglen;
209
210 if (pglen != len || xdr->tail[0].iov_len != 0)
211 flags |= MSG_MORE;
212
213 /* Hmm... We might be dealing with highmem pages */
214 if (PageHighMem(*ppage))
215 sendpage = sock_no_sendpage;
216 err = sendpage(sock, *ppage, base, len, flags);
217 if (ret == 0)
218 ret = err;
219 else if (err > 0)
220 ret += err;
221 if (err != len)
222 goto out;
223 base = 0;
224 ppage++;
225 } while ((pglen -= len) != 0);
226copy_tail:
227 len = xdr->tail[0].iov_len;
228 if (base < len) {
229 err = xs_send_tail(sock, xdr, base, len);
230 if (ret == 0)
231 ret = err;
232 else if (err > 0)
233 ret += err;
234 }
235out:
236 return ret;
237}
238
239/**
240 * xs_nospace - place task on wait queue if transmit was incomplete
241 * @task: task to put to sleep
242 *
243 */
244static void xs_nospace(struct rpc_task *task)
245{
246 struct rpc_rqst *req = task->tk_rqstp;
247 struct rpc_xprt *xprt = req->rq_xprt;
248
249 dprintk("RPC: %4d xmit incomplete (%u left of %u)\n",
250 task->tk_pid, req->rq_slen - req->rq_bytes_sent,
251 req->rq_slen);
252
253 if (test_bit(SOCK_ASYNC_NOSPACE, &xprt->sock->flags)) {
254 /* Protect against races with write_space */
255 spin_lock_bh(&xprt->transport_lock);
256
257 /* Don't race with disconnect */
258 if (!xprt_connected(xprt))
259 task->tk_status = -ENOTCONN;
260 else if (test_bit(SOCK_NOSPACE, &xprt->sock->flags))
261 xprt_wait_for_buffer_space(task);
262
263 spin_unlock_bh(&xprt->transport_lock);
264 } else
265 /* Keep holding the socket if it is blocked */
266 rpc_delay(task, HZ>>4);
267}
268
269/**
270 * xs_udp_send_request - write an RPC request to a UDP socket
271 * @task: address of RPC task that manages the state of an RPC request
272 *
273 * Return values:
274 * 0: The request has been sent
275 * EAGAIN: The socket was blocked, please call again later to
276 * complete the request
277 * ENOTCONN: Caller needs to invoke connect logic then call again
278 * other: Some other error occured, the request was not sent
279 */
280static int xs_udp_send_request(struct rpc_task *task)
281{
282 struct rpc_rqst *req = task->tk_rqstp;
283 struct rpc_xprt *xprt = req->rq_xprt;
284 struct xdr_buf *xdr = &req->rq_snd_buf;
285 int status;
286
287 xs_pktdump("packet data:",
288 req->rq_svec->iov_base,
289 req->rq_svec->iov_len);
290
291 req->rq_xtime = jiffies;
292 status = xs_sendpages(xprt->sock, (struct sockaddr *) &xprt->addr,
293 sizeof(xprt->addr), xdr, req->rq_bytes_sent);
294
295 dprintk("RPC: xs_udp_send_request(%u) = %d\n",
296 xdr->len - req->rq_bytes_sent, status);
297
298 if (likely(status >= (int) req->rq_slen))
299 return 0;
300
301 /* Still some bytes left; set up for a retry later. */
302 if (status > 0)
303 status = -EAGAIN;
304
305 switch (status) {
306 case -ENETUNREACH:
307 case -EPIPE:
308 case -ECONNREFUSED:
309 /* When the server has died, an ICMP port unreachable message
310 * prompts ECONNREFUSED. */
311 break;
312 case -EAGAIN:
313 xs_nospace(task);
314 break;
315 default:
316 dprintk("RPC: sendmsg returned unrecognized error %d\n",
317 -status);
318 break;
319 }
320
321 return status;
322}
323
324static inline void xs_encode_tcp_record_marker(struct xdr_buf *buf)
325{
326 u32 reclen = buf->len - sizeof(rpc_fraghdr);
327 rpc_fraghdr *base = buf->head[0].iov_base;
328 *base = htonl(RPC_LAST_STREAM_FRAGMENT | reclen);
329}
330
331/**
332 * xs_tcp_send_request - write an RPC request to a TCP socket
333 * @task: address of RPC task that manages the state of an RPC request
334 *
335 * Return values:
336 * 0: The request has been sent
337 * EAGAIN: The socket was blocked, please call again later to
338 * complete the request
339 * ENOTCONN: Caller needs to invoke connect logic then call again
340 * other: Some other error occured, the request was not sent
341 *
342 * XXX: In the case of soft timeouts, should we eventually give up
343 * if sendmsg is not able to make progress?
344 */
345static int xs_tcp_send_request(struct rpc_task *task)
346{
347 struct rpc_rqst *req = task->tk_rqstp;
348 struct rpc_xprt *xprt = req->rq_xprt;
349 struct xdr_buf *xdr = &req->rq_snd_buf;
350 int status, retry = 0;
351
352 xs_encode_tcp_record_marker(&req->rq_snd_buf);
353
354 xs_pktdump("packet data:",
355 req->rq_svec->iov_base,
356 req->rq_svec->iov_len);
357
358 /* Continue transmitting the packet/record. We must be careful
359 * to cope with writespace callbacks arriving _after_ we have
360 * called sendmsg(). */
361 while (1) {
362 req->rq_xtime = jiffies;
363 status = xs_sendpages(xprt->sock, NULL, 0, xdr,
364 req->rq_bytes_sent);
365
366 dprintk("RPC: xs_tcp_send_request(%u) = %d\n",
367 xdr->len - req->rq_bytes_sent, status);
368
369 if (unlikely(status < 0))
370 break;
371
372 /* If we've sent the entire packet, immediately
373 * reset the count of bytes sent. */
374 req->rq_bytes_sent += status;
375 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
376 req->rq_bytes_sent = 0;
377 return 0;
378 }
379
380 status = -EAGAIN;
381 if (retry++ > XS_SENDMSG_RETRY)
382 break;
383 }
384
385 switch (status) {
386 case -EAGAIN:
387 xs_nospace(task);
388 break;
389 case -ECONNREFUSED:
390 case -ECONNRESET:
391 case -ENOTCONN:
392 case -EPIPE:
393 status = -ENOTCONN;
394 break;
395 default:
396 dprintk("RPC: sendmsg returned unrecognized error %d\n",
397 -status);
398 xprt_disconnect(xprt);
399 break;
400 }
401
402 return status;
403}
404
405/**
406 * xs_close - close a socket
407 * @xprt: transport
408 *
409 * This is used when all requests are complete; ie, no DRC state remains
410 * on the server we want to save.
411 */
412static void xs_close(struct rpc_xprt *xprt)
413{
414 struct socket *sock = xprt->sock;
415 struct sock *sk = xprt->inet;
416
417 if (!sk)
418 return;
419
420 dprintk("RPC: xs_close xprt %p\n", xprt);
421
422 write_lock_bh(&sk->sk_callback_lock);
423 xprt->inet = NULL;
424 xprt->sock = NULL;
425
426 sk->sk_user_data = NULL;
427 sk->sk_data_ready = xprt->old_data_ready;
428 sk->sk_state_change = xprt->old_state_change;
429 sk->sk_write_space = xprt->old_write_space;
430 write_unlock_bh(&sk->sk_callback_lock);
431
432 sk->sk_no_check = 0;
433
434 sock_release(sock);
435}
436
437/**
438 * xs_destroy - prepare to shutdown a transport
439 * @xprt: doomed transport
440 *
441 */
442static void xs_destroy(struct rpc_xprt *xprt)
443{
444 dprintk("RPC: xs_destroy xprt %p\n", xprt);
445
446 cancel_delayed_work(&xprt->connect_worker);
447 flush_scheduled_work();
448
449 xprt_disconnect(xprt);
450 xs_close(xprt);
451 kfree(xprt->slot);
452}
453
454static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
455{
456 return (struct rpc_xprt *) sk->sk_user_data;
457}
458
459/**
460 * xs_udp_data_ready - "data ready" callback for UDP sockets
461 * @sk: socket with data to read
462 * @len: how much data to read
463 *
464 */
465static void xs_udp_data_ready(struct sock *sk, int len)
466{
467 struct rpc_task *task;
468 struct rpc_xprt *xprt;
469 struct rpc_rqst *rovr;
470 struct sk_buff *skb;
471 int err, repsize, copied;
472 u32 _xid, *xp;
473
474 read_lock(&sk->sk_callback_lock);
475 dprintk("RPC: xs_udp_data_ready...\n");
476 if (!(xprt = xprt_from_sock(sk)))
477 goto out;
478
479 if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
480 goto out;
481
482 if (xprt->shutdown)
483 goto dropit;
484
485 repsize = skb->len - sizeof(struct udphdr);
486 if (repsize < 4) {
487 dprintk("RPC: impossible RPC reply size %d!\n", repsize);
488 goto dropit;
489 }
490
491 /* Copy the XID from the skb... */
492 xp = skb_header_pointer(skb, sizeof(struct udphdr),
493 sizeof(_xid), &_xid);
494 if (xp == NULL)
495 goto dropit;
496
497 /* Look up and lock the request corresponding to the given XID */
498 spin_lock(&xprt->transport_lock);
499 rovr = xprt_lookup_rqst(xprt, *xp);
500 if (!rovr)
501 goto out_unlock;
502 task = rovr->rq_task;
503
504 if ((copied = rovr->rq_private_buf.buflen) > repsize)
505 copied = repsize;
506
507 /* Suck it into the iovec, verify checksum if not done by hw. */
508 if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb))
509 goto out_unlock;
510
511 /* Something worked... */
512 dst_confirm(skb->dst);
513
514 xprt_adjust_cwnd(task, copied);
515 xprt_update_rtt(task);
516 xprt_complete_rqst(task, copied);
517
518 out_unlock:
519 spin_unlock(&xprt->transport_lock);
520 dropit:
521 skb_free_datagram(sk, skb);
522 out:
523 read_unlock(&sk->sk_callback_lock);
524}
525
526static inline size_t xs_tcp_copy_data(skb_reader_t *desc, void *p, size_t len)
527{
528 if (len > desc->count)
529 len = desc->count;
530 if (skb_copy_bits(desc->skb, desc->offset, p, len)) {
531 dprintk("RPC: failed to copy %zu bytes from skb. %zu bytes remain\n",
532 len, desc->count);
533 return 0;
534 }
535 desc->offset += len;
536 desc->count -= len;
537 dprintk("RPC: copied %zu bytes from skb. %zu bytes remain\n",
538 len, desc->count);
539 return len;
540}
541
542static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, skb_reader_t *desc)
543{
544 size_t len, used;
545 char *p;
546
547 p = ((char *) &xprt->tcp_recm) + xprt->tcp_offset;
548 len = sizeof(xprt->tcp_recm) - xprt->tcp_offset;
549 used = xs_tcp_copy_data(desc, p, len);
550 xprt->tcp_offset += used;
551 if (used != len)
552 return;
553
554 xprt->tcp_reclen = ntohl(xprt->tcp_recm);
555 if (xprt->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
556 xprt->tcp_flags |= XPRT_LAST_FRAG;
557 else
558 xprt->tcp_flags &= ~XPRT_LAST_FRAG;
559 xprt->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
560
561 xprt->tcp_flags &= ~XPRT_COPY_RECM;
562 xprt->tcp_offset = 0;
563
564 /* Sanity check of the record length */
565 if (unlikely(xprt->tcp_reclen < 4)) {
566 dprintk("RPC: invalid TCP record fragment length\n");
567 xprt_disconnect(xprt);
568 return;
569 }
570 dprintk("RPC: reading TCP record fragment of length %d\n",
571 xprt->tcp_reclen);
572}
573
574static void xs_tcp_check_recm(struct rpc_xprt *xprt)
575{
576 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u, tcp_flags = %lx\n",
577 xprt, xprt->tcp_copied, xprt->tcp_offset, xprt->tcp_reclen, xprt->tcp_flags);
578 if (xprt->tcp_offset == xprt->tcp_reclen) {
579 xprt->tcp_flags |= XPRT_COPY_RECM;
580 xprt->tcp_offset = 0;
581 if (xprt->tcp_flags & XPRT_LAST_FRAG) {
582 xprt->tcp_flags &= ~XPRT_COPY_DATA;
583 xprt->tcp_flags |= XPRT_COPY_XID;
584 xprt->tcp_copied = 0;
585 }
586 }
587}
588
589static inline void xs_tcp_read_xid(struct rpc_xprt *xprt, skb_reader_t *desc)
590{
591 size_t len, used;
592 char *p;
593
594 len = sizeof(xprt->tcp_xid) - xprt->tcp_offset;
595 dprintk("RPC: reading XID (%Zu bytes)\n", len);
596 p = ((char *) &xprt->tcp_xid) + xprt->tcp_offset;
597 used = xs_tcp_copy_data(desc, p, len);
598 xprt->tcp_offset += used;
599 if (used != len)
600 return;
601 xprt->tcp_flags &= ~XPRT_COPY_XID;
602 xprt->tcp_flags |= XPRT_COPY_DATA;
603 xprt->tcp_copied = 4;
604 dprintk("RPC: reading reply for XID %08x\n",
605 ntohl(xprt->tcp_xid));
606 xs_tcp_check_recm(xprt);
607}
608
609static inline void xs_tcp_read_request(struct rpc_xprt *xprt, skb_reader_t *desc)
610{
611 struct rpc_rqst *req;
612 struct xdr_buf *rcvbuf;
613 size_t len;
614 ssize_t r;
615
616 /* Find and lock the request corresponding to this xid */
617 spin_lock(&xprt->transport_lock);
618 req = xprt_lookup_rqst(xprt, xprt->tcp_xid);
619 if (!req) {
620 xprt->tcp_flags &= ~XPRT_COPY_DATA;
621 dprintk("RPC: XID %08x request not found!\n",
622 ntohl(xprt->tcp_xid));
623 spin_unlock(&xprt->transport_lock);
624 return;
625 }
626
627 rcvbuf = &req->rq_private_buf;
628 len = desc->count;
629 if (len > xprt->tcp_reclen - xprt->tcp_offset) {
630 skb_reader_t my_desc;
631
632 len = xprt->tcp_reclen - xprt->tcp_offset;
633 memcpy(&my_desc, desc, sizeof(my_desc));
634 my_desc.count = len;
635 r = xdr_partial_copy_from_skb(rcvbuf, xprt->tcp_copied,
636 &my_desc, xs_tcp_copy_data);
637 desc->count -= r;
638 desc->offset += r;
639 } else
640 r = xdr_partial_copy_from_skb(rcvbuf, xprt->tcp_copied,
641 desc, xs_tcp_copy_data);
642
643 if (r > 0) {
644 xprt->tcp_copied += r;
645 xprt->tcp_offset += r;
646 }
647 if (r != len) {
648 /* Error when copying to the receive buffer,
649 * usually because we weren't able to allocate
650 * additional buffer pages. All we can do now
651 * is turn off XPRT_COPY_DATA, so the request
652 * will not receive any additional updates,
653 * and time out.
654 * Any remaining data from this record will
655 * be discarded.
656 */
657 xprt->tcp_flags &= ~XPRT_COPY_DATA;
658 dprintk("RPC: XID %08x truncated request\n",
659 ntohl(xprt->tcp_xid));
660 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u\n",
661 xprt, xprt->tcp_copied, xprt->tcp_offset, xprt->tcp_reclen);
662 goto out;
663 }
664
665 dprintk("RPC: XID %08x read %Zd bytes\n",
666 ntohl(xprt->tcp_xid), r);
667 dprintk("RPC: xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u\n",
668 xprt, xprt->tcp_copied, xprt->tcp_offset, xprt->tcp_reclen);
669
670 if (xprt->tcp_copied == req->rq_private_buf.buflen)
671 xprt->tcp_flags &= ~XPRT_COPY_DATA;
672 else if (xprt->tcp_offset == xprt->tcp_reclen) {
673 if (xprt->tcp_flags & XPRT_LAST_FRAG)
674 xprt->tcp_flags &= ~XPRT_COPY_DATA;
675 }
676
677out:
678 if (!(xprt->tcp_flags & XPRT_COPY_DATA))
679 xprt_complete_rqst(req->rq_task, xprt->tcp_copied);
680 spin_unlock(&xprt->transport_lock);
681 xs_tcp_check_recm(xprt);
682}
683
684static inline void xs_tcp_read_discard(struct rpc_xprt *xprt, skb_reader_t *desc)
685{
686 size_t len;
687
688 len = xprt->tcp_reclen - xprt->tcp_offset;
689 if (len > desc->count)
690 len = desc->count;
691 desc->count -= len;
692 desc->offset += len;
693 xprt->tcp_offset += len;
694 dprintk("RPC: discarded %Zu bytes\n", len);
695 xs_tcp_check_recm(xprt);
696}
697
698static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
699{
700 struct rpc_xprt *xprt = rd_desc->arg.data;
701 skb_reader_t desc = {
702 .skb = skb,
703 .offset = offset,
704 .count = len,
705 .csum = 0
706 };
707
708 dprintk("RPC: xs_tcp_data_recv started\n");
709 do {
710 /* Read in a new fragment marker if necessary */
711 /* Can we ever really expect to get completely empty fragments? */
712 if (xprt->tcp_flags & XPRT_COPY_RECM) {
713 xs_tcp_read_fraghdr(xprt, &desc);
714 continue;
715 }
716 /* Read in the xid if necessary */
717 if (xprt->tcp_flags & XPRT_COPY_XID) {
718 xs_tcp_read_xid(xprt, &desc);
719 continue;
720 }
721 /* Read in the request data */
722 if (xprt->tcp_flags & XPRT_COPY_DATA) {
723 xs_tcp_read_request(xprt, &desc);
724 continue;
725 }
726 /* Skip over any trailing bytes on short reads */
727 xs_tcp_read_discard(xprt, &desc);
728 } while (desc.count);
729 dprintk("RPC: xs_tcp_data_recv done\n");
730 return len - desc.count;
731}
732
733/**
734 * xs_tcp_data_ready - "data ready" callback for TCP sockets
735 * @sk: socket with data to read
736 * @bytes: how much data to read
737 *
738 */
739static void xs_tcp_data_ready(struct sock *sk, int bytes)
740{
741 struct rpc_xprt *xprt;
742 read_descriptor_t rd_desc;
743
744 read_lock(&sk->sk_callback_lock);
745 dprintk("RPC: xs_tcp_data_ready...\n");
746 if (!(xprt = xprt_from_sock(sk)))
747 goto out;
748 if (xprt->shutdown)
749 goto out;
750
751 /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
752 rd_desc.arg.data = xprt;
753 rd_desc.count = 65536;
754 tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
755out:
756 read_unlock(&sk->sk_callback_lock);
757}
758
759/**
760 * xs_tcp_state_change - callback to handle TCP socket state changes
761 * @sk: socket whose state has changed
762 *
763 */
764static void xs_tcp_state_change(struct sock *sk)
765{
766 struct rpc_xprt *xprt;
767
768 read_lock(&sk->sk_callback_lock);
769 if (!(xprt = xprt_from_sock(sk)))
770 goto out;
771 dprintk("RPC: xs_tcp_state_change client %p...\n", xprt);
772 dprintk("RPC: state %x conn %d dead %d zapped %d\n",
773 sk->sk_state, xprt_connected(xprt),
774 sock_flag(sk, SOCK_DEAD),
775 sock_flag(sk, SOCK_ZAPPED));
776
777 switch (sk->sk_state) {
778 case TCP_ESTABLISHED:
779 spin_lock_bh(&xprt->transport_lock);
780 if (!xprt_test_and_set_connected(xprt)) {
781 /* Reset TCP record info */
782 xprt->tcp_offset = 0;
783 xprt->tcp_reclen = 0;
784 xprt->tcp_copied = 0;
785 xprt->tcp_flags = XPRT_COPY_RECM | XPRT_COPY_XID;
786 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
787 xprt_wake_pending_tasks(xprt, 0);
788 }
789 spin_unlock_bh(&xprt->transport_lock);
790 break;
791 case TCP_SYN_SENT:
792 case TCP_SYN_RECV:
793 break;
794 default:
795 xprt_disconnect(xprt);
796 break;
797 }
798 out:
799 read_unlock(&sk->sk_callback_lock);
800}
801
802/**
803 * xs_udp_write_space - callback invoked when socket buffer space
804 * becomes available
805 * @sk: socket whose state has changed
806 *
807 * Called when more output buffer space is available for this socket.
808 * We try not to wake our writers until they can make "significant"
809 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
810 * with a bunch of small requests.
811 */
812static void xs_udp_write_space(struct sock *sk)
813{
814 read_lock(&sk->sk_callback_lock);
815
816 /* from net/core/sock.c:sock_def_write_space */
817 if (sock_writeable(sk)) {
818 struct socket *sock;
819 struct rpc_xprt *xprt;
820
821 if (unlikely(!(sock = sk->sk_socket)))
822 goto out;
823 if (unlikely(!(xprt = xprt_from_sock(sk))))
824 goto out;
825 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)))
826 goto out;
827
828 xprt_write_space(xprt);
829 }
830
831 out:
832 read_unlock(&sk->sk_callback_lock);
833}
834
835/**
836 * xs_tcp_write_space - callback invoked when socket buffer space
837 * becomes available
838 * @sk: socket whose state has changed
839 *
840 * Called when more output buffer space is available for this socket.
841 * We try not to wake our writers until they can make "significant"
842 * progress, otherwise we'll waste resources thrashing kernel_sendmsg
843 * with a bunch of small requests.
844 */
845static void xs_tcp_write_space(struct sock *sk)
846{
847 read_lock(&sk->sk_callback_lock);
848
849 /* from net/core/stream.c:sk_stream_write_space */
850 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
851 struct socket *sock;
852 struct rpc_xprt *xprt;
853
854 if (unlikely(!(sock = sk->sk_socket)))
855 goto out;
856 if (unlikely(!(xprt = xprt_from_sock(sk))))
857 goto out;
858 if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)))
859 goto out;
860
861 xprt_write_space(xprt);
862 }
863
864 out:
865 read_unlock(&sk->sk_callback_lock);
866}
867
868static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
869{
870 struct sock *sk = xprt->inet;
871
872 if (xprt->rcvsize) {
873 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
874 sk->sk_rcvbuf = xprt->rcvsize * xprt->max_reqs * 2;
875 }
876 if (xprt->sndsize) {
877 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
878 sk->sk_sndbuf = xprt->sndsize * xprt->max_reqs * 2;
879 sk->sk_write_space(sk);
880 }
881}
882
883/**
884 * xs_udp_set_buffer_size - set send and receive limits
885 * @xprt: generic transport
886 * @sndsize: requested size of send buffer, in bytes
887 * @rcvsize: requested size of receive buffer, in bytes
888 *
889 * Set socket send and receive buffer size limits.
890 */
891static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
892{
893 xprt->sndsize = 0;
894 if (sndsize)
895 xprt->sndsize = sndsize + 1024;
896 xprt->rcvsize = 0;
897 if (rcvsize)
898 xprt->rcvsize = rcvsize + 1024;
899
900 xs_udp_do_set_buffer_size(xprt);
901}
902
903/**
904 * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
905 * @task: task that timed out
906 *
907 * Adjust the congestion window after a retransmit timeout has occurred.
908 */
909static void xs_udp_timer(struct rpc_task *task)
910{
911 xprt_adjust_cwnd(task, -ETIMEDOUT);
912}
913
914static int xs_bindresvport(struct rpc_xprt *xprt, struct socket *sock)
915{
916 struct sockaddr_in myaddr = {
917 .sin_family = AF_INET,
918 };
919 int err;
920 unsigned short port = xprt->port;
921
922 do {
923 myaddr.sin_port = htons(port);
924 err = sock->ops->bind(sock, (struct sockaddr *) &myaddr,
925 sizeof(myaddr));
926 if (err == 0) {
927 xprt->port = port;
928 dprintk("RPC: xs_bindresvport bound to port %u\n",
929 port);
930 return 0;
931 }
932 if (port <= xprt_min_resvport)
933 port = xprt_max_resvport;
934 else
935 port--;
936 } while (err == -EADDRINUSE && port != xprt->port);
937
938 dprintk("RPC: can't bind to reserved port (%d).\n", -err);
939 return err;
940}
941
942/**
943 * xs_udp_connect_worker - set up a UDP socket
944 * @args: RPC transport to connect
945 *
946 * Invoked by a work queue tasklet.
947 */
948static void xs_udp_connect_worker(void *args)
949{
950 struct rpc_xprt *xprt = (struct rpc_xprt *) args;
951 struct socket *sock = xprt->sock;
952 int err, status = -EIO;
953
954 if (xprt->shutdown || xprt->addr.sin_port == 0)
955 goto out;
956
957 dprintk("RPC: xs_udp_connect_worker for xprt %p\n", xprt);
958
959 /* Start by resetting any existing state */
960 xs_close(xprt);
961
962 if ((err = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock)) < 0) {
963 dprintk("RPC: can't create UDP transport socket (%d).\n", -err);
964 goto out;
965 }
966
967 if (xprt->resvport && xs_bindresvport(xprt, sock) < 0) {
968 sock_release(sock);
969 goto out;
970 }
971
972 if (!xprt->inet) {
973 struct sock *sk = sock->sk;
974
975 write_lock_bh(&sk->sk_callback_lock);
976
977 sk->sk_user_data = xprt;
978 xprt->old_data_ready = sk->sk_data_ready;
979 xprt->old_state_change = sk->sk_state_change;
980 xprt->old_write_space = sk->sk_write_space;
981 sk->sk_data_ready = xs_udp_data_ready;
982 sk->sk_write_space = xs_udp_write_space;
983 sk->sk_no_check = UDP_CSUM_NORCV;
984
985 xprt_set_connected(xprt);
986
987 /* Reset to new socket */
988 xprt->sock = sock;
989 xprt->inet = sk;
990
991 write_unlock_bh(&sk->sk_callback_lock);
992 }
993 xs_udp_do_set_buffer_size(xprt);
994 status = 0;
995out:
996 xprt_wake_pending_tasks(xprt, status);
997 xprt_clear_connecting(xprt);
998}
999
1000/*
1001 * We need to preserve the port number so the reply cache on the server can
1002 * find our cached RPC replies when we get around to reconnecting.
1003 */
1004static void xs_tcp_reuse_connection(struct rpc_xprt *xprt)
1005{
1006 int result;
1007 struct socket *sock = xprt->sock;
1008 struct sockaddr any;
1009
1010 dprintk("RPC: disconnecting xprt %p to reuse port\n", xprt);
1011
1012 /*
1013 * Disconnect the transport socket by doing a connect operation
1014 * with AF_UNSPEC. This should return immediately...
1015 */
1016 memset(&any, 0, sizeof(any));
1017 any.sa_family = AF_UNSPEC;
1018 result = sock->ops->connect(sock, &any, sizeof(any), 0);
1019 if (result)
1020 dprintk("RPC: AF_UNSPEC connect return code %d\n",
1021 result);
1022}
1023
1024/**
1025 * xs_tcp_connect_worker - connect a TCP socket to a remote endpoint
1026 * @args: RPC transport to connect
1027 *
1028 * Invoked by a work queue tasklet.
1029 */
1030static void xs_tcp_connect_worker(void *args)
1031{
1032 struct rpc_xprt *xprt = (struct rpc_xprt *)args;
1033 struct socket *sock = xprt->sock;
1034 int err, status = -EIO;
1035
1036 if (xprt->shutdown || xprt->addr.sin_port == 0)
1037 goto out;
1038
1039 dprintk("RPC: xs_tcp_connect_worker for xprt %p\n", xprt);
1040
1041 if (!xprt->sock) {
1042 /* start from scratch */
1043 if ((err = sock_create_kern(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock)) < 0) {
1044 dprintk("RPC: can't create TCP transport socket (%d).\n", -err);
1045 goto out;
1046 }
1047
1048 if (xprt->resvport && xs_bindresvport(xprt, sock) < 0) {
1049 sock_release(sock);
1050 goto out;
1051 }
1052 } else
1053 /* "close" the socket, preserving the local port */
1054 xs_tcp_reuse_connection(xprt);
1055
1056 if (!xprt->inet) {
1057 struct sock *sk = sock->sk;
1058
1059 write_lock_bh(&sk->sk_callback_lock);
1060
1061 sk->sk_user_data = xprt;
1062 xprt->old_data_ready = sk->sk_data_ready;
1063 xprt->old_state_change = sk->sk_state_change;
1064 xprt->old_write_space = sk->sk_write_space;
1065 sk->sk_data_ready = xs_tcp_data_ready;
1066 sk->sk_state_change = xs_tcp_state_change;
1067 sk->sk_write_space = xs_tcp_write_space;
1068
1069 /* socket options */
1070 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
1071 sock_reset_flag(sk, SOCK_LINGER);
1072 tcp_sk(sk)->linger2 = 0;
1073 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1074
1075 xprt_clear_connected(xprt);
1076
1077 /* Reset to new socket */
1078 xprt->sock = sock;
1079 xprt->inet = sk;
1080
1081 write_unlock_bh(&sk->sk_callback_lock);
1082 }
1083
1084 /* Tell the socket layer to start connecting... */
1085 status = sock->ops->connect(sock, (struct sockaddr *) &xprt->addr,
1086 sizeof(xprt->addr), O_NONBLOCK);
1087 dprintk("RPC: %p connect status %d connected %d sock state %d\n",
1088 xprt, -status, xprt_connected(xprt), sock->sk->sk_state);
1089 if (status < 0) {
1090 switch (status) {
1091 case -EINPROGRESS:
1092 case -EALREADY:
1093 goto out_clear;
1094 case -ECONNREFUSED:
1095 case -ECONNRESET:
1096 /* retry with existing socket, after a delay */
1097 break;
1098 default:
1099 /* get rid of existing socket, and retry */
1100 xs_close(xprt);
1101 break;
1102 }
1103 }
1104out:
1105 xprt_wake_pending_tasks(xprt, status);
1106out_clear:
1107 xprt_clear_connecting(xprt);
1108}
1109
1110/**
1111 * xs_connect - connect a socket to a remote endpoint
1112 * @task: address of RPC task that manages state of connect request
1113 *
1114 * TCP: If the remote end dropped the connection, delay reconnecting.
1115 *
1116 * UDP socket connects are synchronous, but we use a work queue anyway
1117 * to guarantee that even unprivileged user processes can set up a
1118 * socket on a privileged port.
1119 *
1120 * If a UDP socket connect fails, the delay behavior here prevents
1121 * retry floods (hard mounts).
1122 */
1123static void xs_connect(struct rpc_task *task)
1124{
1125 struct rpc_xprt *xprt = task->tk_xprt;
1126
1127 if (xprt_test_and_set_connecting(xprt))
1128 return;
1129
1130 if (xprt->sock != NULL) {
1131 dprintk("RPC: xs_connect delayed xprt %p for %lu seconds\n",
1132 xprt, xprt->reestablish_timeout / HZ);
1133 schedule_delayed_work(&xprt->connect_worker,
1134 xprt->reestablish_timeout);
1135 xprt->reestablish_timeout <<= 1;
1136 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
1137 xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
1138 } else {
1139 dprintk("RPC: xs_connect scheduled xprt %p\n", xprt);
1140 schedule_work(&xprt->connect_worker);
1141
1142 /* flush_scheduled_work can sleep... */
1143 if (!RPC_IS_ASYNC(task))
1144 flush_scheduled_work();
1145 }
1146}
1147
1148static struct rpc_xprt_ops xs_udp_ops = {
1149 .set_buffer_size = xs_udp_set_buffer_size,
1150 .reserve_xprt = xprt_reserve_xprt_cong,
1151 .release_xprt = xprt_release_xprt_cong,
1152 .connect = xs_connect,
1153 .send_request = xs_udp_send_request,
1154 .set_retrans_timeout = xprt_set_retrans_timeout_rtt,
1155 .timer = xs_udp_timer,
1156 .release_request = xprt_release_rqst_cong,
1157 .close = xs_close,
1158 .destroy = xs_destroy,
1159};
1160
1161static struct rpc_xprt_ops xs_tcp_ops = {
1162 .reserve_xprt = xprt_reserve_xprt,
1163 .release_xprt = xprt_release_xprt,
1164 .connect = xs_connect,
1165 .send_request = xs_tcp_send_request,
1166 .set_retrans_timeout = xprt_set_retrans_timeout_def,
1167 .close = xs_close,
1168 .destroy = xs_destroy,
1169};
1170
1171/**
1172 * xs_setup_udp - Set up transport to use a UDP socket
1173 * @xprt: transport to set up
1174 * @to: timeout parameters
1175 *
1176 */
1177int xs_setup_udp(struct rpc_xprt *xprt, struct rpc_timeout *to)
1178{
1179 size_t slot_table_size;
1180
1181 dprintk("RPC: setting up udp-ipv4 transport...\n");
1182
1183 xprt->max_reqs = xprt_udp_slot_table_entries;
1184 slot_table_size = xprt->max_reqs * sizeof(xprt->slot[0]);
1185 xprt->slot = kmalloc(slot_table_size, GFP_KERNEL);
1186 if (xprt->slot == NULL)
1187 return -ENOMEM;
1188 memset(xprt->slot, 0, slot_table_size);
1189
1190 xprt->prot = IPPROTO_UDP;
1191 xprt->port = xprt_max_resvport;
1192 xprt->tsh_size = 0;
1193 xprt->resvport = capable(CAP_NET_BIND_SERVICE) ? 1 : 0;
1194 /* XXX: header size can vary due to auth type, IPv6, etc. */
1195 xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
1196
1197 INIT_WORK(&xprt->connect_worker, xs_udp_connect_worker, xprt);
1198 xprt->bind_timeout = XS_BIND_TO;
1199 xprt->connect_timeout = XS_UDP_CONN_TO;
1200 xprt->reestablish_timeout = XS_UDP_REEST_TO;
1201 xprt->idle_timeout = XS_IDLE_DISC_TO;
1202
1203 xprt->ops = &xs_udp_ops;
1204
1205 if (to)
1206 xprt->timeout = *to;
1207 else
1208 xprt_set_timeout(&xprt->timeout, 5, 5 * HZ);
1209
1210 return 0;
1211}
1212
1213/**
1214 * xs_setup_tcp - Set up transport to use a TCP socket
1215 * @xprt: transport to set up
1216 * @to: timeout parameters
1217 *
1218 */
1219int xs_setup_tcp(struct rpc_xprt *xprt, struct rpc_timeout *to)
1220{
1221 size_t slot_table_size;
1222
1223 dprintk("RPC: setting up tcp-ipv4 transport...\n");
1224
1225 xprt->max_reqs = xprt_tcp_slot_table_entries;
1226 slot_table_size = xprt->max_reqs * sizeof(xprt->slot[0]);
1227 xprt->slot = kmalloc(slot_table_size, GFP_KERNEL);
1228 if (xprt->slot == NULL)
1229 return -ENOMEM;
1230 memset(xprt->slot, 0, slot_table_size);
1231
1232 xprt->prot = IPPROTO_TCP;
1233 xprt->port = xprt_max_resvport;
1234 xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
1235 xprt->resvport = capable(CAP_NET_BIND_SERVICE) ? 1 : 0;
1236 xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
1237
1238 INIT_WORK(&xprt->connect_worker, xs_tcp_connect_worker, xprt);
1239 xprt->bind_timeout = XS_BIND_TO;
1240 xprt->connect_timeout = XS_TCP_CONN_TO;
1241 xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1242 xprt->idle_timeout = XS_IDLE_DISC_TO;
1243
1244 xprt->ops = &xs_tcp_ops;
1245
1246 if (to)
1247 xprt->timeout = *to;
1248 else
1249 xprt_set_timeout(&xprt->timeout, 2, 60 * HZ);
1250
1251 return 0;
1252}
generated by cgit v1.2.2 (git 2.25.0) at 2025-01-14 15:44:56 -0500