aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorLinus Torvalds <torvalds@linux-foundation.org>2013-02-21 12:03:54 -0500
committerLinus Torvalds <torvalds@linux-foundation.org>2013-02-21 12:03:54 -0500
commitc4bc705e45d91f900f7ec2e23a458280b084ae91 (patch)
tree54dc459057ddbcd33321d255b0a94f213dc43910
parent2608e3d0fa63b892f37a9f1921c2d2b37c7933c1 (diff)
parent634734b63ac39e137a1c623ba74f3e062b6577db (diff)
Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mszeredi/fuse
Pull fuse updates from Miklos Szeredi: "The biggest part of this pull request is a patch series from Maxim Patlasov to optimize scatter-gather direct IO. There's also the addition of a "readdirplus" API, poll events and various fixes and cleanups. There's a one line change outside of fuse to mm/filemap.c which makes the argument of iov_iter_single_seg_count() const, required by Maxim's patches." * 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mszeredi/fuse: (22 commits) fuse: allow control of adaptive readdirplus use Synchronize fuse header with one used in library fuse: send poll events fuse: don't WARN when nlink is zero fuse: avoid out-of-scope stack access fuse: bump version for READDIRPLUS FUSE: Adapt readdirplus to application usage patterns Do not use RCU for current process credentials fuse: cleanup fuse_direct_io() fuse: optimize __fuse_direct_io() fuse: optimize fuse_get_user_pages() fuse: pass iov[] to fuse_get_user_pages() mm: minor cleanup of iov_iter_single_seg_count() fuse: use req->page_descs[] for argpages cases fuse: add per-page descriptor <offset, length> to fuse_req fuse: rework fuse_do_ioctl() fuse: rework fuse_perform_write() fuse: rework fuse_readpages() fuse: rework fuse_retrieve() fuse: categorize fuse_get_req() ...
Diffstat
-rw-r--r--fs/fuse/cuse.c10
-rw-r--r--fs/fuse/dev.c128
-rw-r--r--fs/fuse/dir.c259
-rw-r--r--fs/fuse/file.c238
-rw-r--r--fs/fuse/fuse_i.h74
-rw-r--r--fs/fuse/inode.c16
-rw-r--r--include/linux/fs.h2
-rw-r--r--include/uapi/linux/fuse.h60
-rw-r--r--mm/filemap.c2
9 files changed, 609 insertions, 180 deletions
diff --git a/fs/fuse/cuse.c b/fs/fuse/cuse.c
index e397b675b029..6f96a8def147 100644
--- a/fs/fuse/cuse.c
+++ b/fs/fuse/cuse.c
@@ -91,19 +91,22 @@ static ssize_t cuse_read(struct file *file, char __user *buf, size_t count,
91 loff_t *ppos) 91 loff_t *ppos)
92{ 92{
93 loff_t pos = 0; 93 loff_t pos = 0;
94 struct iovec iov = { .iov_base = buf, .iov_len = count };
94 95
95 return fuse_direct_io(file, buf, count, &pos, 0); 96 return fuse_direct_io(file, &iov, 1, count, &pos, 0);
96} 97}
97 98
98static ssize_t cuse_write(struct file *file, const char __user *buf, 99static ssize_t cuse_write(struct file *file, const char __user *buf,
99 size_t count, loff_t *ppos) 100 size_t count, loff_t *ppos)
100{ 101{
101 loff_t pos = 0; 102 loff_t pos = 0;
103 struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = count };
104
102 /* 105 /*
103 * No locking or generic_write_checks(), the server is 106 * No locking or generic_write_checks(), the server is
104 * responsible for locking and sanity checks. 107 * responsible for locking and sanity checks.
105 */ 108 */
106 return fuse_direct_io(file, buf, count, &pos, 1); 109 return fuse_direct_io(file, &iov, 1, count, &pos, 1);
107} 110}
108 111
109static int cuse_open(struct inode *inode, struct file *file) 112static int cuse_open(struct inode *inode, struct file *file)
@@ -419,7 +422,7 @@ static int cuse_send_init(struct cuse_conn *cc)
419 422
420 BUILD_BUG_ON(CUSE_INIT_INFO_MAX > PAGE_SIZE); 423 BUILD_BUG_ON(CUSE_INIT_INFO_MAX > PAGE_SIZE);
421 424
422 req = fuse_get_req(fc); 425 req = fuse_get_req(fc, 1);
423 if (IS_ERR(req)) { 426 if (IS_ERR(req)) {
424 rc = PTR_ERR(req); 427 rc = PTR_ERR(req);
425 goto err; 428 goto err;
@@ -449,6 +452,7 @@ static int cuse_send_init(struct cuse_conn *cc)
449 req->out.argvar = 1; 452 req->out.argvar = 1;
450 req->out.argpages = 1; 453 req->out.argpages = 1;
451 req->pages[0] = page; 454 req->pages[0] = page;
455 req->page_descs[0].length = req->out.args[1].size;
452 req->num_pages = 1; 456 req->num_pages = 1;
453 req->end = cuse_process_init_reply; 457 req->end = cuse_process_init_reply;
454 fuse_request_send_background(fc, req); 458 fuse_request_send_background(fc, req);
diff --git a/fs/fuse/dev.c b/fs/fuse/dev.c
index e83351aa5bad..e9bdec0b16d9 100644
--- a/fs/fuse/dev.c
+++ b/fs/fuse/dev.c
@@ -34,34 +34,67 @@ static struct fuse_conn *fuse_get_conn(struct file *file)
34 return file->private_data; 34 return file->private_data;
35} 35}
36 36
37static void fuse_request_init(struct fuse_req *req) 37static void fuse_request_init(struct fuse_req *req, struct page **pages,
38 struct fuse_page_desc *page_descs,
39 unsigned npages)
38{ 40{
39 memset(req, 0, sizeof(*req)); 41 memset(req, 0, sizeof(*req));
42 memset(pages, 0, sizeof(*pages) * npages);
43 memset(page_descs, 0, sizeof(*page_descs) * npages);
40 INIT_LIST_HEAD(&req->list); 44 INIT_LIST_HEAD(&req->list);
41 INIT_LIST_HEAD(&req->intr_entry); 45 INIT_LIST_HEAD(&req->intr_entry);
42 init_waitqueue_head(&req->waitq); 46 init_waitqueue_head(&req->waitq);
43 atomic_set(&req->count, 1); 47 atomic_set(&req->count, 1);
48 req->pages = pages;
49 req->page_descs = page_descs;
50 req->max_pages = npages;
44} 51}
45 52
46struct fuse_req *fuse_request_alloc(void) 53static struct fuse_req *__fuse_request_alloc(unsigned npages, gfp_t flags)
47{ 54{
48 struct fuse_req *req = kmem_cache_alloc(fuse_req_cachep, GFP_KERNEL); 55 struct fuse_req *req = kmem_cache_alloc(fuse_req_cachep, flags);
49 if (req) 56 if (req) {
50 fuse_request_init(req); 57 struct page **pages;
58 struct fuse_page_desc *page_descs;
59
60 if (npages <= FUSE_REQ_INLINE_PAGES) {
61 pages = req->inline_pages;
62 page_descs = req->inline_page_descs;
63 } else {
64 pages = kmalloc(sizeof(struct page *) * npages, flags);
65 page_descs = kmalloc(sizeof(struct fuse_page_desc) *
66 npages, flags);
67 }
68
69 if (!pages || !page_descs) {
70 kfree(pages);
71 kfree(page_descs);
72 kmem_cache_free(fuse_req_cachep, req);
73 return NULL;
74 }
75
76 fuse_request_init(req, pages, page_descs, npages);
77 }
51 return req; 78 return req;
52} 79}
80
81struct fuse_req *fuse_request_alloc(unsigned npages)
82{
83 return __fuse_request_alloc(npages, GFP_KERNEL);
84}
53EXPORT_SYMBOL_GPL(fuse_request_alloc); 85EXPORT_SYMBOL_GPL(fuse_request_alloc);
54 86
55struct fuse_req *fuse_request_alloc_nofs(void) 87struct fuse_req *fuse_request_alloc_nofs(unsigned npages)
56{ 88{
57 struct fuse_req *req = kmem_cache_alloc(fuse_req_cachep, GFP_NOFS); 89 return __fuse_request_alloc(npages, GFP_NOFS);
58 if (req)
59 fuse_request_init(req);
60 return req;
61} 90}
62 91
63void fuse_request_free(struct fuse_req *req) 92void fuse_request_free(struct fuse_req *req)
64{ 93{
94 if (req->pages != req->inline_pages) {
95 kfree(req->pages);
96 kfree(req->page_descs);
97 }
65 kmem_cache_free(fuse_req_cachep, req); 98 kmem_cache_free(fuse_req_cachep, req);
66} 99}
67 100
@@ -97,7 +130,7 @@ static void fuse_req_init_context(struct fuse_req *req)
97 req->in.h.pid = current->pid; 130 req->in.h.pid = current->pid;
98} 131}
99 132
100struct fuse_req *fuse_get_req(struct fuse_conn *fc) 133struct fuse_req *fuse_get_req(struct fuse_conn *fc, unsigned npages)
101{ 134{
102 struct fuse_req *req; 135 struct fuse_req *req;
103 sigset_t oldset; 136 sigset_t oldset;
@@ -116,7 +149,7 @@ struct fuse_req *fuse_get_req(struct fuse_conn *fc)
116 if (!fc->connected) 149 if (!fc->connected)
117 goto out; 150 goto out;
118 151
119 req = fuse_request_alloc(); 152 req = fuse_request_alloc(npages);
120 err = -ENOMEM; 153 err = -ENOMEM;
121 if (!req) 154 if (!req)
122 goto out; 155 goto out;
@@ -165,7 +198,7 @@ static void put_reserved_req(struct fuse_conn *fc, struct fuse_req *req)
165 struct fuse_file *ff = file->private_data; 198 struct fuse_file *ff = file->private_data;
166 199
167 spin_lock(&fc->lock); 200 spin_lock(&fc->lock);
168 fuse_request_init(req); 201 fuse_request_init(req, req->pages, req->page_descs, req->max_pages);
169 BUG_ON(ff->reserved_req); 202 BUG_ON(ff->reserved_req);
170 ff->reserved_req = req; 203 ff->reserved_req = req;
171 wake_up_all(&fc->reserved_req_waitq); 204 wake_up_all(&fc->reserved_req_waitq);
@@ -186,13 +219,14 @@ static void put_reserved_req(struct fuse_conn *fc, struct fuse_req *req)
186 * filesystem should not have it's own file open. If deadlock is 219 * filesystem should not have it's own file open. If deadlock is
187 * intentional, it can still be broken by "aborting" the filesystem. 220 * intentional, it can still be broken by "aborting" the filesystem.
188 */ 221 */
189struct fuse_req *fuse_get_req_nofail(struct fuse_conn *fc, struct file *file) 222struct fuse_req *fuse_get_req_nofail_nopages(struct fuse_conn *fc,
223 struct file *file)
190{ 224{
191 struct fuse_req *req; 225 struct fuse_req *req;
192 226
193 atomic_inc(&fc->num_waiting); 227 atomic_inc(&fc->num_waiting);
194 wait_event(fc->blocked_waitq, !fc->blocked); 228 wait_event(fc->blocked_waitq, !fc->blocked);
195 req = fuse_request_alloc(); 229 req = fuse_request_alloc(0);
196 if (!req) 230 if (!req)
197 req = get_reserved_req(fc, file); 231 req = get_reserved_req(fc, file);
198 232
@@ -406,9 +440,8 @@ __acquires(fc->lock)
406 } 440 }
407} 441}
408 442
409void fuse_request_send(struct fuse_conn *fc, struct fuse_req *req) 443static void __fuse_request_send(struct fuse_conn *fc, struct fuse_req *req)
410{ 444{
411 req->isreply = 1;
412 spin_lock(&fc->lock); 445 spin_lock(&fc->lock);
413 if (!fc->connected) 446 if (!fc->connected)
414 req->out.h.error = -ENOTCONN; 447 req->out.h.error = -ENOTCONN;
@@ -425,6 +458,12 @@ void fuse_request_send(struct fuse_conn *fc, struct fuse_req *req)
425 } 458 }
426 spin_unlock(&fc->lock); 459 spin_unlock(&fc->lock);
427} 460}
461
462void fuse_request_send(struct fuse_conn *fc, struct fuse_req *req)
463{
464 req->isreply = 1;
465 __fuse_request_send(fc, req);
466}
428EXPORT_SYMBOL_GPL(fuse_request_send); 467EXPORT_SYMBOL_GPL(fuse_request_send);
429 468
430static void fuse_request_send_nowait_locked(struct fuse_conn *fc, 469static void fuse_request_send_nowait_locked(struct fuse_conn *fc,
@@ -491,6 +530,27 @@ void fuse_request_send_background_locked(struct fuse_conn *fc,
491 fuse_request_send_nowait_locked(fc, req); 530 fuse_request_send_nowait_locked(fc, req);
492} 531}
493 532
533void fuse_force_forget(struct file *file, u64 nodeid)
534{
535 struct inode *inode = file->f_path.dentry->d_inode;
536 struct fuse_conn *fc = get_fuse_conn(inode);
537 struct fuse_req *req;
538 struct fuse_forget_in inarg;
539
540 memset(&inarg, 0, sizeof(inarg));
541 inarg.nlookup = 1;
542 req = fuse_get_req_nofail_nopages(fc, file);
543 req->in.h.opcode = FUSE_FORGET;
544 req->in.h.nodeid = nodeid;
545 req->in.numargs = 1;
546 req->in.args[0].size = sizeof(inarg);
547 req->in.args[0].value = &inarg;
548 req->isreply = 0;
549 __fuse_request_send(fc, req);
550 /* ignore errors */
551 fuse_put_request(fc, req);
552}
553
494/* 554/*
495 * Lock the request. Up to the next unlock_request() there mustn't be 555 * Lock the request. Up to the next unlock_request() there mustn't be
496 * anything that could cause a page-fault. If the request was already 556 * anything that could cause a page-fault. If the request was already
@@ -850,11 +910,11 @@ static int fuse_copy_pages(struct fuse_copy_state *cs, unsigned nbytes,
850{ 910{
851 unsigned i; 911 unsigned i;
852 struct fuse_req *req = cs->req; 912 struct fuse_req *req = cs->req;
853 unsigned offset = req->page_offset;
854 unsigned count = min(nbytes, (unsigned) PAGE_SIZE - offset);
855 913
856 for (i = 0; i < req->num_pages && (nbytes || zeroing); i++) { 914 for (i = 0; i < req->num_pages && (nbytes || zeroing); i++) {
857 int err; 915 int err;
916 unsigned offset = req->page_descs[i].offset;
917 unsigned count = min(nbytes, req->page_descs[i].length);
858 918
859 err = fuse_copy_page(cs, &req->pages[i], offset, count, 919 err = fuse_copy_page(cs, &req->pages[i], offset, count,
860 zeroing); 920 zeroing);
@@ -862,8 +922,6 @@ static int fuse_copy_pages(struct fuse_copy_state *cs, unsigned nbytes,
862 return err; 922 return err;
863 923
864 nbytes -= count; 924 nbytes -= count;
865 count = min(nbytes, (unsigned) PAGE_SIZE);
866 offset = 0;
867 } 925 }
868 return 0; 926 return 0;
869} 927}
@@ -1536,29 +1594,34 @@ static int fuse_retrieve(struct fuse_conn *fc, struct inode *inode,
1536 unsigned int num; 1594 unsigned int num;
1537 unsigned int offset; 1595 unsigned int offset;
1538 size_t total_len = 0; 1596 size_t total_len = 0;
1597 int num_pages;
1598
1599 offset = outarg->offset & ~PAGE_CACHE_MASK;
1600 file_size = i_size_read(inode);
1601
1602 num = outarg->size;
1603 if (outarg->offset > file_size)
1604 num = 0;
1605 else if (outarg->offset + num > file_size)
1606 num = file_size - outarg->offset;
1607
1608 num_pages = (num + offset + PAGE_SIZE - 1) >> PAGE_SHIFT;
1609 num_pages = min(num_pages, FUSE_MAX_PAGES_PER_REQ);
1539 1610
1540 req = fuse_get_req(fc); 1611 req = fuse_get_req(fc, num_pages);
1541 if (IS_ERR(req)) 1612 if (IS_ERR(req))
1542 return PTR_ERR(req); 1613 return PTR_ERR(req);
1543 1614
1544 offset = outarg->offset & ~PAGE_CACHE_MASK;
1545
1546 req->in.h.opcode = FUSE_NOTIFY_REPLY; 1615 req->in.h.opcode = FUSE_NOTIFY_REPLY;
1547 req->in.h.nodeid = outarg->nodeid; 1616 req->in.h.nodeid = outarg->nodeid;
1548 req->in.numargs = 2; 1617 req->in.numargs = 2;
1549 req->in.argpages = 1; 1618 req->in.argpages = 1;
1550 req->page_offset = offset; 1619 req->page_descs[0].offset = offset;
1551 req->end = fuse_retrieve_end; 1620 req->end = fuse_retrieve_end;
1552 1621
1553 index = outarg->offset >> PAGE_CACHE_SHIFT; 1622 index = outarg->offset >> PAGE_CACHE_SHIFT;
1554 file_size = i_size_read(inode);
1555 num = outarg->size;
1556 if (outarg->offset > file_size)
1557 num = 0;
1558 else if (outarg->offset + num > file_size)
1559 num = file_size - outarg->offset;
1560 1623
1561 while (num && req->num_pages < FUSE_MAX_PAGES_PER_REQ) { 1624 while (num && req->num_pages < num_pages) {
1562 struct page *page; 1625 struct page *page;
1563 unsigned int this_num; 1626 unsigned int this_num;
1564 1627
@@ -1568,6 +1631,7 @@ static int fuse_retrieve(struct fuse_conn *fc, struct inode *inode,
1568 1631
1569 this_num = min_t(unsigned, num, PAGE_CACHE_SIZE - offset); 1632 this_num = min_t(unsigned, num, PAGE_CACHE_SIZE - offset);
1570 req->pages[req->num_pages] = page; 1633 req->pages[req->num_pages] = page;
1634 req->page_descs[req->num_pages].length = this_num;
1571 req->num_pages++; 1635 req->num_pages++;
1572 1636
1573 offset = 0; 1637 offset = 0;
diff --git a/fs/fuse/dir.c b/fs/fuse/dir.c
index b7c09f9eb40c..85065221a58a 100644
--- a/fs/fuse/dir.c
+++ b/fs/fuse/dir.c
@@ -14,6 +14,29 @@
14#include <linux/namei.h> 14#include <linux/namei.h>
15#include <linux/slab.h> 15#include <linux/slab.h>
16 16
17static bool fuse_use_readdirplus(struct inode *dir, struct file *filp)
18{
19 struct fuse_conn *fc = get_fuse_conn(dir);
20 struct fuse_inode *fi = get_fuse_inode(dir);
21
22 if (!fc->do_readdirplus)
23 return false;
24 if (!fc->readdirplus_auto)
25 return true;
26 if (test_and_clear_bit(FUSE_I_ADVISE_RDPLUS, &fi->state))
27 return true;
28 if (filp->f_pos == 0)
29 return true;
30 return false;
31}
32
33static void fuse_advise_use_readdirplus(struct inode *dir)
34{
35 struct fuse_inode *fi = get_fuse_inode(dir);
36
37 set_bit(FUSE_I_ADVISE_RDPLUS, &fi->state);
38}
39
17#if BITS_PER_LONG >= 64 40#if BITS_PER_LONG >= 64
18static inline void fuse_dentry_settime(struct dentry *entry, u64 time) 41static inline void fuse_dentry_settime(struct dentry *entry, u64 time)
19{ 42{
@@ -178,7 +201,7 @@ static int fuse_dentry_revalidate(struct dentry *entry, unsigned int flags)
178 return -ECHILD; 201 return -ECHILD;
179 202
180 fc = get_fuse_conn(inode); 203 fc = get_fuse_conn(inode);
181 req = fuse_get_req(fc); 204 req = fuse_get_req_nopages(fc);
182 if (IS_ERR(req)) 205 if (IS_ERR(req))
183 return 0; 206 return 0;
184 207
@@ -219,6 +242,7 @@ static int fuse_dentry_revalidate(struct dentry *entry, unsigned int flags)
219 attr_version); 242 attr_version);
220 fuse_change_entry_timeout(entry, &outarg); 243 fuse_change_entry_timeout(entry, &outarg);
221 } 244 }
245 fuse_advise_use_readdirplus(inode);
222 return 1; 246 return 1;
223} 247}
224 248
@@ -271,7 +295,7 @@ int fuse_lookup_name(struct super_block *sb, u64 nodeid, struct qstr *name,
271 if (name->len > FUSE_NAME_MAX) 295 if (name->len > FUSE_NAME_MAX)
272 goto out; 296 goto out;
273 297
274 req = fuse_get_req(fc); 298 req = fuse_get_req_nopages(fc);
275 err = PTR_ERR(req); 299 err = PTR_ERR(req);
276 if (IS_ERR(req)) 300 if (IS_ERR(req))
277 goto out; 301 goto out;
@@ -355,6 +379,7 @@ static struct dentry *fuse_lookup(struct inode *dir, struct dentry *entry,
355 else 379 else
356 fuse_invalidate_entry_cache(entry); 380 fuse_invalidate_entry_cache(entry);
357 381
382 fuse_advise_use_readdirplus(dir);
358 return newent; 383 return newent;
359 384
360 out_iput: 385 out_iput:
@@ -391,7 +416,7 @@ static int fuse_create_open(struct inode *dir, struct dentry *entry,
391 if (!forget) 416 if (!forget)
392 goto out_err; 417 goto out_err;
393 418
394 req = fuse_get_req(fc); 419 req = fuse_get_req_nopages(fc);
395 err = PTR_ERR(req); 420 err = PTR_ERR(req);
396 if (IS_ERR(req)) 421 if (IS_ERR(req))
397 goto out_put_forget_req; 422 goto out_put_forget_req;
@@ -592,7 +617,7 @@ static int fuse_mknod(struct inode *dir, struct dentry *entry, umode_t mode,
592{ 617{
593 struct fuse_mknod_in inarg; 618 struct fuse_mknod_in inarg;
594 struct fuse_conn *fc = get_fuse_conn(dir); 619 struct fuse_conn *fc = get_fuse_conn(dir);
595 struct fuse_req *req = fuse_get_req(fc); 620 struct fuse_req *req = fuse_get_req_nopages(fc);
596 if (IS_ERR(req)) 621 if (IS_ERR(req))
597 return PTR_ERR(req); 622 return PTR_ERR(req);
598 623
@@ -623,7 +648,7 @@ static int fuse_mkdir(struct inode *dir, struct dentry *entry, umode_t mode)
623{ 648{
624 struct fuse_mkdir_in inarg; 649 struct fuse_mkdir_in inarg;
625 struct fuse_conn *fc = get_fuse_conn(dir); 650 struct fuse_conn *fc = get_fuse_conn(dir);
626 struct fuse_req *req = fuse_get_req(fc); 651 struct fuse_req *req = fuse_get_req_nopages(fc);
627 if (IS_ERR(req)) 652 if (IS_ERR(req))
628 return PTR_ERR(req); 653 return PTR_ERR(req);
629 654
@@ -647,7 +672,7 @@ static int fuse_symlink(struct inode *dir, struct dentry *entry,
647{ 672{
648 struct fuse_conn *fc = get_fuse_conn(dir); 673 struct fuse_conn *fc = get_fuse_conn(dir);
649 unsigned len = strlen(link) + 1; 674 unsigned len = strlen(link) + 1;
650 struct fuse_req *req = fuse_get_req(fc); 675 struct fuse_req *req = fuse_get_req_nopages(fc);
651 if (IS_ERR(req)) 676 if (IS_ERR(req))
652 return PTR_ERR(req); 677 return PTR_ERR(req);
653 678
@@ -664,7 +689,7 @@ static int fuse_unlink(struct inode *dir, struct dentry *entry)
664{ 689{
665 int err; 690 int err;
666 struct fuse_conn *fc = get_fuse_conn(dir); 691 struct fuse_conn *fc = get_fuse_conn(dir);
667 struct fuse_req *req = fuse_get_req(fc); 692 struct fuse_req *req = fuse_get_req_nopages(fc);
668 if (IS_ERR(req)) 693 if (IS_ERR(req))
669 return PTR_ERR(req); 694 return PTR_ERR(req);
670 695
@@ -682,7 +707,14 @@ static int fuse_unlink(struct inode *dir, struct dentry *entry)
682 707
683 spin_lock(&fc->lock); 708 spin_lock(&fc->lock);
684 fi->attr_version = ++fc->attr_version; 709 fi->attr_version = ++fc->attr_version;
685 drop_nlink(inode); 710 /*
711 * If i_nlink == 0 then unlink doesn't make sense, yet this can
712 * happen if userspace filesystem is careless. It would be
713 * difficult to enforce correct nlink usage so just ignore this
714 * condition here
715 */
716 if (inode->i_nlink > 0)
717 drop_nlink(inode);
686 spin_unlock(&fc->lock); 718 spin_unlock(&fc->lock);
687 fuse_invalidate_attr(inode); 719 fuse_invalidate_attr(inode);
688 fuse_invalidate_attr(dir); 720 fuse_invalidate_attr(dir);
@@ -696,7 +728,7 @@ static int fuse_rmdir(struct inode *dir, struct dentry *entry)
696{ 728{
697 int err; 729 int err;
698 struct fuse_conn *fc = get_fuse_conn(dir); 730 struct fuse_conn *fc = get_fuse_conn(dir);
699 struct fuse_req *req = fuse_get_req(fc); 731 struct fuse_req *req = fuse_get_req_nopages(fc);
700 if (IS_ERR(req)) 732 if (IS_ERR(req))
701 return PTR_ERR(req); 733 return PTR_ERR(req);
702 734
@@ -723,7 +755,7 @@ static int fuse_rename(struct inode *olddir, struct dentry *oldent,
723 int err; 755 int err;
724 struct fuse_rename_in inarg; 756 struct fuse_rename_in inarg;
725 struct fuse_conn *fc = get_fuse_conn(olddir); 757 struct fuse_conn *fc = get_fuse_conn(olddir);
726 struct fuse_req *req = fuse_get_req(fc); 758 struct fuse_req *req = fuse_get_req_nopages(fc);
727 759
728 if (IS_ERR(req)) 760 if (IS_ERR(req))
729 return PTR_ERR(req); 761 return PTR_ERR(req);
@@ -776,7 +808,7 @@ static int fuse_link(struct dentry *entry, struct inode *newdir,
776 struct fuse_link_in inarg; 808 struct fuse_link_in inarg;
777 struct inode *inode = entry->d_inode; 809 struct inode *inode = entry->d_inode;
778 struct fuse_conn *fc = get_fuse_conn(inode); 810 struct fuse_conn *fc = get_fuse_conn(inode);
779 struct fuse_req *req = fuse_get_req(fc); 811 struct fuse_req *req = fuse_get_req_nopages(fc);
780 if (IS_ERR(req)) 812 if (IS_ERR(req))
781 return PTR_ERR(req); 813 return PTR_ERR(req);
782 814
@@ -848,7 +880,7 @@ static int fuse_do_getattr(struct inode *inode, struct kstat *stat,
848 struct fuse_req *req; 880 struct fuse_req *req;
849 u64 attr_version; 881 u64 attr_version;
850 882
851 req = fuse_get_req(fc); 883 req = fuse_get_req_nopages(fc);
852 if (IS_ERR(req)) 884 if (IS_ERR(req))
853 return PTR_ERR(req); 885 return PTR_ERR(req);
854 886
@@ -985,7 +1017,7 @@ int fuse_reverse_inval_entry(struct super_block *sb, u64 parent_nodeid,
985 1017
986/* 1018/*
987 * Calling into a user-controlled filesystem gives the filesystem 1019 * Calling into a user-controlled filesystem gives the filesystem
988 * daemon ptrace-like capabilities over the requester process. This 1020 * daemon ptrace-like capabilities over the current process. This
989 * means, that the filesystem daemon is able to record the exact 1021 * means, that the filesystem daemon is able to record the exact
990 * filesystem operations performed, and can also control the behavior 1022 * filesystem operations performed, and can also control the behavior
991 * of the requester process in otherwise impossible ways. For example 1023 * of the requester process in otherwise impossible ways. For example
@@ -996,27 +1028,23 @@ int fuse_reverse_inval_entry(struct super_block *sb, u64 parent_nodeid,
996 * for which the owner of the mount has ptrace privilege. This 1028 * for which the owner of the mount has ptrace privilege. This
997 * excludes processes started by other users, suid or sgid processes. 1029 * excludes processes started by other users, suid or sgid processes.
998 */ 1030 */
999int fuse_allow_task(struct fuse_conn *fc, struct task_struct *task) 1031int fuse_allow_current_process(struct fuse_conn *fc)
1000{ 1032{
1001 const struct cred *cred; 1033 const struct cred *cred;
1002 int ret;
1003 1034
1004 if (fc->flags & FUSE_ALLOW_OTHER) 1035 if (fc->flags & FUSE_ALLOW_OTHER)
1005 return 1; 1036 return 1;
1006 1037
1007 rcu_read_lock(); 1038 cred = current_cred();
1008 ret = 0;
1009 cred = __task_cred(task);
1010 if (uid_eq(cred->euid, fc->user_id) && 1039 if (uid_eq(cred->euid, fc->user_id) &&
1011 uid_eq(cred->suid, fc->user_id) && 1040 uid_eq(cred->suid, fc->user_id) &&
1012 uid_eq(cred->uid, fc->user_id) && 1041 uid_eq(cred->uid, fc->user_id) &&
1013 gid_eq(cred->egid, fc->group_id) && 1042 gid_eq(cred->egid, fc->group_id) &&
1014 gid_eq(cred->sgid, fc->group_id) && 1043 gid_eq(cred->sgid, fc->group_id) &&
1015 gid_eq(cred->gid, fc->group_id)) 1044 gid_eq(cred->gid, fc->group_id))
1016 ret = 1; 1045 return 1;
1017 rcu_read_unlock();
1018 1046
1019 return ret; 1047 return 0;
1020} 1048}
1021 1049
1022static int fuse_access(struct inode *inode, int mask) 1050static int fuse_access(struct inode *inode, int mask)
@@ -1029,7 +1057,7 @@ static int fuse_access(struct inode *inode, int mask)
1029 if (fc->no_access) 1057 if (fc->no_access)
1030 return 0; 1058 return 0;
1031 1059
1032 req = fuse_get_req(fc); 1060 req = fuse_get_req_nopages(fc);
1033 if (IS_ERR(req)) 1061 if (IS_ERR(req))
1034 return PTR_ERR(req); 1062 return PTR_ERR(req);
1035 1063
@@ -1077,7 +1105,7 @@ static int fuse_permission(struct inode *inode, int mask)
1077 bool refreshed = false; 1105 bool refreshed = false;
1078 int err = 0; 1106 int err = 0;
1079 1107
1080 if (!fuse_allow_task(fc, current)) 1108 if (!fuse_allow_current_process(fc))
1081 return -EACCES; 1109 return -EACCES;
1082 1110
1083 /* 1111 /*
@@ -1155,19 +1183,157 @@ static int parse_dirfile(char *buf, size_t nbytes, struct file *file,
1155 return 0; 1183 return 0;
1156} 1184}
1157 1185
1158static int fuse_readdir(struct file *file, void *dstbuf, filldir_t filldir) 1186static int fuse_direntplus_link(struct file *file,
1187 struct fuse_direntplus *direntplus,
1188 u64 attr_version)
1159{ 1189{
1160 int err; 1190 int err;
1191 struct fuse_entry_out *o = &direntplus->entry_out;
1192 struct fuse_dirent *dirent = &direntplus->dirent;
1193 struct dentry *parent = file->f_path.dentry;
1194 struct qstr name = QSTR_INIT(dirent->name, dirent->namelen);
1195 struct dentry *dentry;
1196 struct dentry *alias;
1197 struct inode *dir = parent->d_inode;
1198 struct fuse_conn *fc;
1199 struct inode *inode;
1200
1201 if (!o->nodeid) {
1202 /*
1203 * Unlike in the case of fuse_lookup, zero nodeid does not mean
1204 * ENOENT. Instead, it only means the userspace filesystem did
1205 * not want to return attributes/handle for this entry.
1206 *
1207 * So do nothing.
1208 */
1209 return 0;
1210 }
1211
1212 if (name.name[0] == '.') {
1213 /*
1214 * We could potentially refresh the attributes of the directory
1215 * and its parent?
1216 */
1217 if (name.len == 1)
1218 return 0;
1219 if (name.name[1] == '.' && name.len == 2)
1220 return 0;
1221 }
1222 fc = get_fuse_conn(dir);
1223
1224 name.hash = full_name_hash(name.name, name.len);
1225 dentry = d_lookup(parent, &name);
1226 if (dentry && dentry->d_inode) {
1227 inode = dentry->d_inode;
1228 if (get_node_id(inode) == o->nodeid) {
1229 struct fuse_inode *fi;
1230 fi = get_fuse_inode(inode);
1231 spin_lock(&fc->lock);
1232 fi->nlookup++;
1233 spin_unlock(&fc->lock);
1234
1235 /*
1236 * The other branch to 'found' comes via fuse_iget()
1237 * which bumps nlookup inside
1238 */
1239 goto found;
1240 }
1241 err = d_invalidate(dentry);
1242 if (err)
1243 goto out;
1244 dput(dentry);
1245 dentry = NULL;
1246 }
1247
1248 dentry = d_alloc(parent, &name);
1249 err = -ENOMEM;
1250 if (!dentry)
1251 goto out;
1252
1253 inode = fuse_iget(dir->i_sb, o->nodeid, o->generation,
1254 &o->attr, entry_attr_timeout(o), attr_version);
1255 if (!inode)
1256 goto out;
1257
1258 alias = d_materialise_unique(dentry, inode);
1259 err = PTR_ERR(alias);
1260 if (IS_ERR(alias))
1261 goto out;
1262 if (alias) {
1263 dput(dentry);
1264 dentry = alias;
1265 }
1266
1267found:
1268 fuse_change_attributes(inode, &o->attr, entry_attr_timeout(o),
1269 attr_version);
1270
1271 fuse_change_entry_timeout(dentry, o);
1272
1273 err = 0;
1274out:
1275 if (dentry)
1276 dput(dentry);
1277 return err;
1278}
1279
1280static int parse_dirplusfile(char *buf, size_t nbytes, struct file *file,
1281 void *dstbuf, filldir_t filldir, u64 attr_version)
1282{
1283 struct fuse_direntplus *direntplus;
1284 struct fuse_dirent *dirent;
1285 size_t reclen;
1286 int over = 0;
1287 int ret;
1288
1289 while (nbytes >= FUSE_NAME_OFFSET_DIRENTPLUS) {
1290 direntplus = (struct fuse_direntplus *) buf;
1291 dirent = &direntplus->dirent;
1292 reclen = FUSE_DIRENTPLUS_SIZE(direntplus);
1293
1294 if (!dirent->namelen || dirent->namelen > FUSE_NAME_MAX)
1295 return -EIO;
1296 if (reclen > nbytes)
1297 break;
1298
1299 if (!over) {
1300 /* We fill entries into dstbuf only as much as
1301 it can hold. But we still continue iterating
1302 over remaining entries to link them. If not,
1303 we need to send a FORGET for each of those
1304 which we did not link.
1305 */
1306 over = filldir(dstbuf, dirent->name, dirent->namelen,
1307 file->f_pos, dirent->ino,
1308 dirent->type);
1309 file->f_pos = dirent->off;
1310 }
1311
1312 buf += reclen;
1313 nbytes -= reclen;
1314
1315 ret = fuse_direntplus_link(file, direntplus, attr_version);
1316 if (ret)
1317 fuse_force_forget(file, direntplus->entry_out.nodeid);
1318 }
1319
1320 return 0;
1321}
1322
1323static int fuse_readdir(struct file *file, void *dstbuf, filldir_t filldir)
1324{
1325 int plus, err;
1161 size_t nbytes; 1326 size_t nbytes;
1162 struct page *page; 1327 struct page *page;
1163 struct inode *inode = file->f_path.dentry->d_inode; 1328 struct inode *inode = file->f_path.dentry->d_inode;
1164 struct fuse_conn *fc = get_fuse_conn(inode); 1329 struct fuse_conn *fc = get_fuse_conn(inode);
1165 struct fuse_req *req; 1330 struct fuse_req *req;
1331 u64 attr_version = 0;
1166 1332
1167 if (is_bad_inode(inode)) 1333 if (is_bad_inode(inode))
1168 return -EIO; 1334 return -EIO;
1169 1335
1170 req = fuse_get_req(fc); 1336 req = fuse_get_req(fc, 1);
1171 if (IS_ERR(req)) 1337 if (IS_ERR(req))
1172 return PTR_ERR(req); 1338 return PTR_ERR(req);
1173 1339
@@ -1176,17 +1342,34 @@ static int fuse_readdir(struct file *file, void *dstbuf, filldir_t filldir)
1176 fuse_put_request(fc, req); 1342 fuse_put_request(fc, req);
1177 return -ENOMEM; 1343 return -ENOMEM;
1178 } 1344 }
1345
1346 plus = fuse_use_readdirplus(inode, file);
1179 req->out.argpages = 1; 1347 req->out.argpages = 1;
1180 req->num_pages = 1; 1348 req->num_pages = 1;
1181 req->pages[0] = page; 1349 req->pages[0] = page;
1182 fuse_read_fill(req, file, file->f_pos, PAGE_SIZE, FUSE_READDIR); 1350 req->page_descs[0].length = PAGE_SIZE;
1351 if (plus) {
1352 attr_version = fuse_get_attr_version(fc);
1353 fuse_read_fill(req, file, file->f_pos, PAGE_SIZE,
1354 FUSE_READDIRPLUS);
1355 } else {
1356 fuse_read_fill(req, file, file->f_pos, PAGE_SIZE,
1357 FUSE_READDIR);
1358 }
1183 fuse_request_send(fc, req); 1359 fuse_request_send(fc, req);
1184 nbytes = req->out.args[0].size; 1360 nbytes = req->out.args[0].size;
1185 err = req->out.h.error; 1361 err = req->out.h.error;
1186 fuse_put_request(fc, req); 1362 fuse_put_request(fc, req);
1187 if (!err) 1363 if (!err) {
1188 err = parse_dirfile(page_address(page), nbytes, file, dstbuf, 1364 if (plus) {
1189 filldir); 1365 err = parse_dirplusfile(page_address(page), nbytes,
1366 file, dstbuf, filldir,
1367 attr_version);
1368 } else {
1369 err = parse_dirfile(page_address(page), nbytes, file,
1370 dstbuf, filldir);
1371 }
1372 }
1190 1373
1191 __free_page(page); 1374 __free_page(page);
1192 fuse_invalidate_attr(inode); /* atime changed */ 1375 fuse_invalidate_attr(inode); /* atime changed */
@@ -1197,7 +1380,7 @@ static char *read_link(struct dentry *dentry)
1197{ 1380{
1198 struct inode *inode = dentry->d_inode; 1381 struct inode *inode = dentry->d_inode;
1199 struct fuse_conn *fc = get_fuse_conn(inode); 1382 struct fuse_conn *fc = get_fuse_conn(inode);
1200 struct fuse_req *req = fuse_get_req(fc); 1383 struct fuse_req *req = fuse_get_req_nopages(fc);
1201 char *link; 1384 char *link;
1202 1385
1203 if (IS_ERR(req)) 1386 if (IS_ERR(req))
@@ -1391,7 +1574,7 @@ static int fuse_do_setattr(struct dentry *entry, struct iattr *attr,
1391 loff_t oldsize; 1574 loff_t oldsize;
1392 int err; 1575 int err;
1393 1576
1394 if (!fuse_allow_task(fc, current)) 1577 if (!fuse_allow_current_process(fc))
1395 return -EACCES; 1578 return -EACCES;
1396 1579
1397 if (!(fc->flags & FUSE_DEFAULT_PERMISSIONS)) 1580 if (!(fc->flags & FUSE_DEFAULT_PERMISSIONS))
@@ -1410,7 +1593,7 @@ static int fuse_do_setattr(struct dentry *entry, struct iattr *attr,
1410 if (attr->ia_valid & ATTR_SIZE) 1593 if (attr->ia_valid & ATTR_SIZE)
1411 is_truncate = true; 1594 is_truncate = true;
1412 1595
1413 req = fuse_get_req(fc); 1596 req = fuse_get_req_nopages(fc);
1414 if (IS_ERR(req)) 1597 if (IS_ERR(req))
1415 return PTR_ERR(req); 1598 return PTR_ERR(req);
1416 1599
@@ -1500,7 +1683,7 @@ static int fuse_getattr(struct vfsmount *mnt, struct dentry *entry,
1500 struct inode *inode = entry->d_inode; 1683 struct inode *inode = entry->d_inode;
1501 struct fuse_conn *fc = get_fuse_conn(inode); 1684 struct fuse_conn *fc = get_fuse_conn(inode);
1502 1685
1503 if (!fuse_allow_task(fc, current)) 1686 if (!fuse_allow_current_process(fc))
1504 return -EACCES; 1687 return -EACCES;
1505 1688
1506 return fuse_update_attributes(inode, stat, NULL, NULL); 1689 return fuse_update_attributes(inode, stat, NULL, NULL);
@@ -1518,7 +1701,7 @@ static int fuse_setxattr(struct dentry *entry, const char *name,
1518 if (fc->no_setxattr) 1701 if (fc->no_setxattr)
1519 return -EOPNOTSUPP; 1702 return -EOPNOTSUPP;
1520 1703
1521 req = fuse_get_req(fc); 1704 req = fuse_get_req_nopages(fc);
1522 if (IS_ERR(req)) 1705 if (IS_ERR(req))
1523 return PTR_ERR(req); 1706 return PTR_ERR(req);
1524 1707
@@ -1557,7 +1740,7 @@ static ssize_t fuse_getxattr(struct dentry *entry, const char *name,
1557 if (fc->no_getxattr) 1740 if (fc->no_getxattr)
1558 return -EOPNOTSUPP; 1741 return -EOPNOTSUPP;
1559 1742
1560 req = fuse_get_req(fc); 1743 req = fuse_get_req_nopages(fc);
1561 if (IS_ERR(req)) 1744 if (IS_ERR(req))
1562 return PTR_ERR(req); 1745 return PTR_ERR(req);
1563 1746
@@ -1603,13 +1786,13 @@ static ssize_t fuse_listxattr(struct dentry *entry, char *list, size_t size)
1603 struct fuse_getxattr_out outarg; 1786 struct fuse_getxattr_out outarg;
1604 ssize_t ret; 1787 ssize_t ret;
1605 1788
1606 if (!fuse_allow_task(fc, current)) 1789 if (!fuse_allow_current_process(fc))
1607 return -EACCES; 1790 return -EACCES;
1608 1791
1609 if (fc->no_listxattr) 1792 if (fc->no_listxattr)
1610 return -EOPNOTSUPP; 1793 return -EOPNOTSUPP;
1611 1794
1612 req = fuse_get_req(fc); 1795 req = fuse_get_req_nopages(fc);
1613 if (IS_ERR(req)) 1796 if (IS_ERR(req))
1614 return PTR_ERR(req); 1797 return PTR_ERR(req);
1615 1798
@@ -1654,7 +1837,7 @@ static int fuse_removexattr(struct dentry *entry, const char *name)
1654 if (fc->no_removexattr) 1837 if (fc->no_removexattr)
1655 return -EOPNOTSUPP; 1838 return -EOPNOTSUPP;
1656 1839
1657 req = fuse_get_req(fc); 1840 req = fuse_get_req_nopages(fc);
1658 if (IS_ERR(req)) 1841 if (IS_ERR(req))
1659 return PTR_ERR(req); 1842 return PTR_ERR(req);
1660 1843
diff --git a/fs/fuse/file.c b/fs/fuse/file.c
index f3ab824fa302..c8071768b950 100644
--- a/fs/fuse/file.c
+++ b/fs/fuse/file.c
@@ -25,7 +25,7 @@ static int fuse_send_open(struct fuse_conn *fc, u64 nodeid, struct file *file,
25 struct fuse_req *req; 25 struct fuse_req *req;
26 int err; 26 int err;
27 27
28 req = fuse_get_req(fc); 28 req = fuse_get_req_nopages(fc);
29 if (IS_ERR(req)) 29 if (IS_ERR(req))
30 return PTR_ERR(req); 30 return PTR_ERR(req);
31 31
@@ -57,7 +57,7 @@ struct fuse_file *fuse_file_alloc(struct fuse_conn *fc)
57 return NULL; 57 return NULL;
58 58
59 ff->fc = fc; 59 ff->fc = fc;
60 ff->reserved_req = fuse_request_alloc(); 60 ff->reserved_req = fuse_request_alloc(0);
61 if (unlikely(!ff->reserved_req)) { 61 if (unlikely(!ff->reserved_req)) {
62 kfree(ff); 62 kfree(ff);
63 return NULL; 63 return NULL;
@@ -368,7 +368,7 @@ static int fuse_flush(struct file *file, fl_owner_t id)
368 if (fc->no_flush) 368 if (fc->no_flush)
369 return 0; 369 return 0;
370 370
371 req = fuse_get_req_nofail(fc, file); 371 req = fuse_get_req_nofail_nopages(fc, file);
372 memset(&inarg, 0, sizeof(inarg)); 372 memset(&inarg, 0, sizeof(inarg));
373 inarg.fh = ff->fh; 373 inarg.fh = ff->fh;
374 inarg.lock_owner = fuse_lock_owner_id(fc, id); 374 inarg.lock_owner = fuse_lock_owner_id(fc, id);
@@ -436,7 +436,7 @@ int fuse_fsync_common(struct file *file, loff_t start, loff_t end,
436 436
437 fuse_sync_writes(inode); 437 fuse_sync_writes(inode);
438 438
439 req = fuse_get_req(fc); 439 req = fuse_get_req_nopages(fc);
440 if (IS_ERR(req)) { 440 if (IS_ERR(req)) {
441 err = PTR_ERR(req); 441 err = PTR_ERR(req);
442 goto out; 442 goto out;
@@ -544,7 +544,7 @@ static int fuse_readpage(struct file *file, struct page *page)
544 */ 544 */
545 fuse_wait_on_page_writeback(inode, page->index); 545 fuse_wait_on_page_writeback(inode, page->index);
546 546
547 req = fuse_get_req(fc); 547 req = fuse_get_req(fc, 1);
548 err = PTR_ERR(req); 548 err = PTR_ERR(req);
549 if (IS_ERR(req)) 549 if (IS_ERR(req))
550 goto out; 550 goto out;
@@ -555,6 +555,7 @@ static int fuse_readpage(struct file *file, struct page *page)
555 req->out.argpages = 1; 555 req->out.argpages = 1;
556 req->num_pages = 1; 556 req->num_pages = 1;
557 req->pages[0] = page; 557 req->pages[0] = page;
558 req->page_descs[0].length = count;
558 num_read = fuse_send_read(req, file, pos, count, NULL); 559 num_read = fuse_send_read(req, file, pos, count, NULL);
559 err = req->out.h.error; 560 err = req->out.h.error;
560 fuse_put_request(fc, req); 561 fuse_put_request(fc, req);
@@ -641,6 +642,7 @@ struct fuse_fill_data {
641 struct fuse_req *req; 642 struct fuse_req *req;
642 struct file *file; 643 struct file *file;
643 struct inode *inode; 644 struct inode *inode;
645 unsigned nr_pages;
644}; 646};
645 647
646static int fuse_readpages_fill(void *_data, struct page *page) 648static int fuse_readpages_fill(void *_data, struct page *page)
@@ -656,16 +658,26 @@ static int fuse_readpages_fill(void *_data, struct page *page)
656 (req->num_pages == FUSE_MAX_PAGES_PER_REQ || 658 (req->num_pages == FUSE_MAX_PAGES_PER_REQ ||
657 (req->num_pages + 1) * PAGE_CACHE_SIZE > fc->max_read || 659 (req->num_pages + 1) * PAGE_CACHE_SIZE > fc->max_read ||
658 req->pages[req->num_pages - 1]->index + 1 != page->index)) { 660 req->pages[req->num_pages - 1]->index + 1 != page->index)) {
661 int nr_alloc = min_t(unsigned, data->nr_pages,
662 FUSE_MAX_PAGES_PER_REQ);
659 fuse_send_readpages(req, data->file); 663 fuse_send_readpages(req, data->file);
660 data->req = req = fuse_get_req(fc); 664 data->req = req = fuse_get_req(fc, nr_alloc);
661 if (IS_ERR(req)) { 665 if (IS_ERR(req)) {
662 unlock_page(page); 666 unlock_page(page);
663 return PTR_ERR(req); 667 return PTR_ERR(req);
664 } 668 }
665 } 669 }
670
671 if (WARN_ON(req->num_pages >= req->max_pages)) {
672 fuse_put_request(fc, req);
673 return -EIO;
674 }
675
666 page_cache_get(page); 676 page_cache_get(page);
667 req->pages[req->num_pages] = page; 677 req->pages[req->num_pages] = page;
678 req->page_descs[req->num_pages].length = PAGE_SIZE;
668 req->num_pages++; 679 req->num_pages++;
680 data->nr_pages--;
669 return 0; 681 return 0;
670} 682}
671 683
@@ -676,6 +688,7 @@ static int fuse_readpages(struct file *file, struct address_space *mapping,
676 struct fuse_conn *fc = get_fuse_conn(inode); 688 struct fuse_conn *fc = get_fuse_conn(inode);
677 struct fuse_fill_data data; 689 struct fuse_fill_data data;
678 int err; 690 int err;
691 int nr_alloc = min_t(unsigned, nr_pages, FUSE_MAX_PAGES_PER_REQ);
679 692
680 err = -EIO; 693 err = -EIO;
681 if (is_bad_inode(inode)) 694 if (is_bad_inode(inode))
@@ -683,7 +696,8 @@ static int fuse_readpages(struct file *file, struct address_space *mapping,
683 696
684 data.file = file; 697 data.file = file;
685 data.inode = inode; 698 data.inode = inode;
686 data.req = fuse_get_req(fc); 699 data.req = fuse_get_req(fc, nr_alloc);
700 data.nr_pages = nr_pages;
687 err = PTR_ERR(data.req); 701 err = PTR_ERR(data.req);
688 if (IS_ERR(data.req)) 702 if (IS_ERR(data.req))
689 goto out; 703 goto out;
@@ -786,7 +800,7 @@ static size_t fuse_send_write_pages(struct fuse_req *req, struct file *file,
786 800
787 res = fuse_send_write(req, file, pos, count, NULL); 801 res = fuse_send_write(req, file, pos, count, NULL);
788 802
789 offset = req->page_offset; 803 offset = req->page_descs[0].offset;
790 count = res; 804 count = res;
791 for (i = 0; i < req->num_pages; i++) { 805 for (i = 0; i < req->num_pages; i++) {
792 struct page *page = req->pages[i]; 806 struct page *page = req->pages[i];
@@ -817,7 +831,7 @@ static ssize_t fuse_fill_write_pages(struct fuse_req *req,
817 int err; 831 int err;
818 832
819 req->in.argpages = 1; 833 req->in.argpages = 1;
820 req->page_offset = offset; 834 req->page_descs[0].offset = offset;
821 835
822 do { 836 do {
823 size_t tmp; 837 size_t tmp;
@@ -857,6 +871,7 @@ static ssize_t fuse_fill_write_pages(struct fuse_req *req,
857 871
858 err = 0; 872 err = 0;
859 req->pages[req->num_pages] = page; 873 req->pages[req->num_pages] = page;
874 req->page_descs[req->num_pages].length = tmp;
860 req->num_pages++; 875 req->num_pages++;
861 876
862 iov_iter_advance(ii, tmp); 877 iov_iter_advance(ii, tmp);
@@ -869,11 +884,19 @@ static ssize_t fuse_fill_write_pages(struct fuse_req *req,
869 if (!fc->big_writes) 884 if (!fc->big_writes)
870 break; 885 break;
871 } while (iov_iter_count(ii) && count < fc->max_write && 886 } while (iov_iter_count(ii) && count < fc->max_write &&
872 req->num_pages < FUSE_MAX_PAGES_PER_REQ && offset == 0); 887 req->num_pages < req->max_pages && offset == 0);
873 888
874 return count > 0 ? count : err; 889 return count > 0 ? count : err;
875} 890}
876 891
892static inline unsigned fuse_wr_pages(loff_t pos, size_t len)
893{
894 return min_t(unsigned,
895 ((pos + len - 1) >> PAGE_CACHE_SHIFT) -
896 (pos >> PAGE_CACHE_SHIFT) + 1,
897 FUSE_MAX_PAGES_PER_REQ);
898}
899
877static ssize_t fuse_perform_write(struct file *file, 900static ssize_t fuse_perform_write(struct file *file,
878 struct address_space *mapping, 901 struct address_space *mapping,
879 struct iov_iter *ii, loff_t pos) 902 struct iov_iter *ii, loff_t pos)
@@ -889,8 +912,9 @@ static ssize_t fuse_perform_write(struct file *file,
889 do { 912 do {
890 struct fuse_req *req; 913 struct fuse_req *req;
891 ssize_t count; 914 ssize_t count;
915 unsigned nr_pages = fuse_wr_pages(pos, iov_iter_count(ii));
892 916
893 req = fuse_get_req(fc); 917 req = fuse_get_req(fc, nr_pages);
894 if (IS_ERR(req)) { 918 if (IS_ERR(req)) {
895 err = PTR_ERR(req); 919 err = PTR_ERR(req);
896 break; 920 break;
@@ -1023,47 +1047,110 @@ static void fuse_release_user_pages(struct fuse_req *req, int write)
1023 } 1047 }
1024} 1048}
1025 1049
1026static int fuse_get_user_pages(struct fuse_req *req, const char __user *buf, 1050static inline void fuse_page_descs_length_init(struct fuse_req *req,
1051 unsigned index, unsigned nr_pages)
1052{
1053 int i;
1054
1055 for (i = index; i < index + nr_pages; i++)
1056 req->page_descs[i].length = PAGE_SIZE -
1057 req->page_descs[i].offset;
1058}
1059
1060static inline unsigned long fuse_get_user_addr(const struct iov_iter *ii)
1061{
1062 return (unsigned long)ii->iov->iov_base + ii->iov_offset;
1063}
1064
1065static inline size_t fuse_get_frag_size(const struct iov_iter *ii,
1066 size_t max_size)
1067{
1068 return min(iov_iter_single_seg_count(ii), max_size);
1069}
1070
1071static int fuse_get_user_pages(struct fuse_req *req, struct iov_iter *ii,
1027 size_t *nbytesp, int write) 1072 size_t *nbytesp, int write)
1028{ 1073{
1029 size_t nbytes = *nbytesp; 1074 size_t nbytes = 0; /* # bytes already packed in req */
1030 unsigned long user_addr = (unsigned long) buf;
1031 unsigned offset = user_addr & ~PAGE_MASK;
1032 int npages;
1033 1075
1034 /* Special case for kernel I/O: can copy directly into the buffer */ 1076 /* Special case for kernel I/O: can copy directly into the buffer */
1035 if (segment_eq(get_fs(), KERNEL_DS)) { 1077 if (segment_eq(get_fs(), KERNEL_DS)) {
1078 unsigned long user_addr = fuse_get_user_addr(ii);
1079 size_t frag_size = fuse_get_frag_size(ii, *nbytesp);
1080
1036 if (write) 1081 if (write)
1037 req->in.args[1].value = (void *) user_addr; 1082 req->in.args[1].value = (void *) user_addr;
1038 else 1083 else
1039 req->out.args[0].value = (void *) user_addr; 1084 req->out.args[0].value = (void *) user_addr;
1040 1085
1086 iov_iter_advance(ii, frag_size);
1087 *nbytesp = frag_size;
1041 return 0; 1088 return 0;
1042 } 1089 }
1043 1090
1044 nbytes = min_t(size_t, nbytes, FUSE_MAX_PAGES_PER_REQ << PAGE_SHIFT); 1091 while (nbytes < *nbytesp && req->num_pages < req->max_pages) {
1045 npages = (nbytes + offset + PAGE_SIZE - 1) >> PAGE_SHIFT; 1092 unsigned npages;
1046 npages = clamp(npages, 1, FUSE_MAX_PAGES_PER_REQ); 1093 unsigned long user_addr = fuse_get_user_addr(ii);
1047 npages = get_user_pages_fast(user_addr, npages, !write, req->pages); 1094 unsigned offset = user_addr & ~PAGE_MASK;
1048 if (npages < 0) 1095 size_t frag_size = fuse_get_frag_size(ii, *nbytesp - nbytes);
1049 return npages; 1096 int ret;
1097
1098 unsigned n = req->max_pages - req->num_pages;
1099 frag_size = min_t(size_t, frag_size, n << PAGE_SHIFT);
1100
1101 npages = (frag_size + offset + PAGE_SIZE - 1) >> PAGE_SHIFT;
1102 npages = clamp(npages, 1U, n);
1103
1104 ret = get_user_pages_fast(user_addr, npages, !write,
1105 &req->pages[req->num_pages]);
1106 if (ret < 0)
1107 return ret;
1050 1108
1051 req->num_pages = npages; 1109 npages = ret;
1052 req->page_offset = offset; 1110 frag_size = min_t(size_t, frag_size,
1111 (npages << PAGE_SHIFT) - offset);
1112 iov_iter_advance(ii, frag_size);
1113
1114 req->page_descs[req->num_pages].offset = offset;
1115 fuse_page_descs_length_init(req, req->num_pages, npages);
1116
1117 req->num_pages += npages;
1118 req->page_descs[req->num_pages - 1].length -=
1119 (npages << PAGE_SHIFT) - offset - frag_size;
1120
1121 nbytes += frag_size;
1122 }
1053 1123
1054 if (write) 1124 if (write)
1055 req->in.argpages = 1; 1125 req->in.argpages = 1;
1056 else 1126 else
1057 req->out.argpages = 1; 1127 req->out.argpages = 1;
1058 1128
1059 nbytes = (req->num_pages << PAGE_SHIFT) - req->page_offset; 1129 *nbytesp = nbytes;
1060 *nbytesp = min(*nbytesp, nbytes);
1061 1130
1062 return 0; 1131 return 0;
1063} 1132}
1064 1133
1065ssize_t fuse_direct_io(struct file *file, const char __user *buf, 1134static inline int fuse_iter_npages(const struct iov_iter *ii_p)
1066 size_t count, loff_t *ppos, int write) 1135{
1136 struct iov_iter ii = *ii_p;
1137 int npages = 0;
1138
1139 while (iov_iter_count(&ii) && npages < FUSE_MAX_PAGES_PER_REQ) {
1140 unsigned long user_addr = fuse_get_user_addr(&ii);
1141 unsigned offset = user_addr & ~PAGE_MASK;
1142 size_t frag_size = iov_iter_single_seg_count(&ii);
1143
1144 npages += (frag_size + offset + PAGE_SIZE - 1) >> PAGE_SHIFT;
1145 iov_iter_advance(&ii, frag_size);
1146 }
1147
1148 return min(npages, FUSE_MAX_PAGES_PER_REQ);
1149}
1150
1151ssize_t fuse_direct_io(struct file *file, const struct iovec *iov,
1152 unsigned long nr_segs, size_t count, loff_t *ppos,
1153 int write)
1067{ 1154{
1068 struct fuse_file *ff = file->private_data; 1155 struct fuse_file *ff = file->private_data;
1069 struct fuse_conn *fc = ff->fc; 1156 struct fuse_conn *fc = ff->fc;
@@ -1071,8 +1158,11 @@ ssize_t fuse_direct_io(struct file *file, const char __user *buf,
1071 loff_t pos = *ppos; 1158 loff_t pos = *ppos;
1072 ssize_t res = 0; 1159 ssize_t res = 0;
1073 struct fuse_req *req; 1160 struct fuse_req *req;
1161 struct iov_iter ii;
1162
1163 iov_iter_init(&ii, iov, nr_segs, count, 0);
1074 1164
1075 req = fuse_get_req(fc); 1165 req = fuse_get_req(fc, fuse_iter_npages(&ii));
1076 if (IS_ERR(req)) 1166 if (IS_ERR(req))
1077 return PTR_ERR(req); 1167 return PTR_ERR(req);
1078 1168
@@ -1080,7 +1170,7 @@ ssize_t fuse_direct_io(struct file *file, const char __user *buf,
1080 size_t nres; 1170 size_t nres;
1081 fl_owner_t owner = current->files; 1171 fl_owner_t owner = current->files;
1082 size_t nbytes = min(count, nmax); 1172 size_t nbytes = min(count, nmax);
1083 int err = fuse_get_user_pages(req, buf, &nbytes, write); 1173 int err = fuse_get_user_pages(req, &ii, &nbytes, write);
1084 if (err) { 1174 if (err) {
1085 res = err; 1175 res = err;
1086 break; 1176 break;
@@ -1103,12 +1193,11 @@ ssize_t fuse_direct_io(struct file *file, const char __user *buf,
1103 count -= nres; 1193 count -= nres;
1104 res += nres; 1194 res += nres;
1105 pos += nres; 1195 pos += nres;
1106 buf += nres;
1107 if (nres != nbytes) 1196 if (nres != nbytes)
1108 break; 1197 break;
1109 if (count) { 1198 if (count) {
1110 fuse_put_request(fc, req); 1199 fuse_put_request(fc, req);
1111 req = fuse_get_req(fc); 1200 req = fuse_get_req(fc, fuse_iter_npages(&ii));
1112 if (IS_ERR(req)) 1201 if (IS_ERR(req))
1113 break; 1202 break;
1114 } 1203 }
@@ -1122,8 +1211,8 @@ ssize_t fuse_direct_io(struct file *file, const char __user *buf,
1122} 1211}
1123EXPORT_SYMBOL_GPL(fuse_direct_io); 1212EXPORT_SYMBOL_GPL(fuse_direct_io);
1124 1213
1125static ssize_t fuse_direct_read(struct file *file, char __user *buf, 1214static ssize_t __fuse_direct_read(struct file *file, const struct iovec *iov,
1126 size_t count, loff_t *ppos) 1215 unsigned long nr_segs, loff_t *ppos)
1127{ 1216{
1128 ssize_t res; 1217 ssize_t res;
1129 struct inode *inode = file->f_path.dentry->d_inode; 1218 struct inode *inode = file->f_path.dentry->d_inode;
@@ -1131,22 +1220,31 @@ static ssize_t fuse_direct_read(struct file *file, char __user *buf,
1131 if (is_bad_inode(inode)) 1220 if (is_bad_inode(inode))
1132 return -EIO; 1221 return -EIO;
1133 1222
1134 res = fuse_direct_io(file, buf, count, ppos, 0); 1223 res = fuse_direct_io(file, iov, nr_segs, iov_length(iov, nr_segs),
1224 ppos, 0);
1135 1225
1136 fuse_invalidate_attr(inode); 1226 fuse_invalidate_attr(inode);
1137 1227
1138 return res; 1228 return res;
1139} 1229}
1140 1230
1141static ssize_t __fuse_direct_write(struct file *file, const char __user *buf, 1231static ssize_t fuse_direct_read(struct file *file, char __user *buf,
1142 size_t count, loff_t *ppos) 1232 size_t count, loff_t *ppos)
1233{
1234 struct iovec iov = { .iov_base = buf, .iov_len = count };
1235 return __fuse_direct_read(file, &iov, 1, ppos);
1236}
1237
1238static ssize_t __fuse_direct_write(struct file *file, const struct iovec *iov,
1239 unsigned long nr_segs, loff_t *ppos)
1143{ 1240{
1144 struct inode *inode = file->f_path.dentry->d_inode; 1241 struct inode *inode = file->f_path.dentry->d_inode;
1242 size_t count = iov_length(iov, nr_segs);
1145 ssize_t res; 1243 ssize_t res;
1146 1244
1147 res = generic_write_checks(file, ppos, &count, 0); 1245 res = generic_write_checks(file, ppos, &count, 0);
1148 if (!res) { 1246 if (!res) {
1149 res = fuse_direct_io(file, buf, count, ppos, 1); 1247 res = fuse_direct_io(file, iov, nr_segs, count, ppos, 1);
1150 if (res > 0) 1248 if (res > 0)
1151 fuse_write_update_size(inode, *ppos); 1249 fuse_write_update_size(inode, *ppos);
1152 } 1250 }
@@ -1159,6 +1257,7 @@ static ssize_t __fuse_direct_write(struct file *file, const char __user *buf,
1159static ssize_t fuse_direct_write(struct file *file, const char __user *buf, 1257static ssize_t fuse_direct_write(struct file *file, const char __user *buf,
1160 size_t count, loff_t *ppos) 1258 size_t count, loff_t *ppos)
1161{ 1259{
1260 struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = count };
1162 struct inode *inode = file->f_path.dentry->d_inode; 1261 struct inode *inode = file->f_path.dentry->d_inode;
1163 ssize_t res; 1262 ssize_t res;
1164 1263
@@ -1167,7 +1266,7 @@ static ssize_t fuse_direct_write(struct file *file, const char __user *buf,
1167 1266
1168 /* Don't allow parallel writes to the same file */ 1267 /* Don't allow parallel writes to the same file */
1169 mutex_lock(&inode->i_mutex); 1268 mutex_lock(&inode->i_mutex);
1170 res = __fuse_direct_write(file, buf, count, ppos); 1269 res = __fuse_direct_write(file, &iov, 1, ppos);
1171 mutex_unlock(&inode->i_mutex); 1270 mutex_unlock(&inode->i_mutex);
1172 1271
1173 return res; 1272 return res;
@@ -1272,7 +1371,7 @@ static int fuse_writepage_locked(struct page *page)
1272 1371
1273 set_page_writeback(page); 1372 set_page_writeback(page);
1274 1373
1275 req = fuse_request_alloc_nofs(); 1374 req = fuse_request_alloc_nofs(1);
1276 if (!req) 1375 if (!req)
1277 goto err; 1376 goto err;
1278 1377
@@ -1293,7 +1392,8 @@ static int fuse_writepage_locked(struct page *page)
1293 req->in.argpages = 1; 1392 req->in.argpages = 1;
1294 req->num_pages = 1; 1393 req->num_pages = 1;
1295 req->pages[0] = tmp_page; 1394 req->pages[0] = tmp_page;
1296 req->page_offset = 0; 1395 req->page_descs[0].offset = 0;
1396 req->page_descs[0].length = PAGE_SIZE;
1297 req->end = fuse_writepage_end; 1397 req->end = fuse_writepage_end;
1298 req->inode = inode; 1398 req->inode = inode;
1299 1399
@@ -1471,7 +1571,7 @@ static int fuse_getlk(struct file *file, struct file_lock *fl)
1471 struct fuse_lk_out outarg; 1571 struct fuse_lk_out outarg;
1472 int err; 1572 int err;
1473 1573
1474 req = fuse_get_req(fc); 1574 req = fuse_get_req_nopages(fc);
1475 if (IS_ERR(req)) 1575 if (IS_ERR(req))
1476 return PTR_ERR(req); 1576 return PTR_ERR(req);
1477 1577
@@ -1506,7 +1606,7 @@ static int fuse_setlk(struct file *file, struct file_lock *fl, int flock)
1506 if (fl->fl_flags & FL_CLOSE) 1606 if (fl->fl_flags & FL_CLOSE)
1507 return 0; 1607 return 0;
1508 1608
1509 req = fuse_get_req(fc); 1609 req = fuse_get_req_nopages(fc);
1510 if (IS_ERR(req)) 1610 if (IS_ERR(req))
1511 return PTR_ERR(req); 1611 return PTR_ERR(req);
1512 1612
@@ -1575,7 +1675,7 @@ static sector_t fuse_bmap(struct address_space *mapping, sector_t block)
1575 if (!inode->i_sb->s_bdev || fc->no_bmap) 1675 if (!inode->i_sb->s_bdev || fc->no_bmap)
1576 return 0; 1676 return 0;
1577 1677
1578 req = fuse_get_req(fc); 1678 req = fuse_get_req_nopages(fc);
1579 if (IS_ERR(req)) 1679 if (IS_ERR(req))
1580 return 0; 1680 return 0;
1581 1681
@@ -1873,7 +1973,7 @@ long fuse_do_ioctl(struct file *file, unsigned int cmd, unsigned long arg,
1873 num_pages++; 1973 num_pages++;
1874 } 1974 }
1875 1975
1876 req = fuse_get_req(fc); 1976 req = fuse_get_req(fc, num_pages);
1877 if (IS_ERR(req)) { 1977 if (IS_ERR(req)) {
1878 err = PTR_ERR(req); 1978 err = PTR_ERR(req);
1879 req = NULL; 1979 req = NULL;
@@ -1881,6 +1981,7 @@ long fuse_do_ioctl(struct file *file, unsigned int cmd, unsigned long arg,
1881 } 1981 }
1882 memcpy(req->pages, pages, sizeof(req->pages[0]) * num_pages); 1982 memcpy(req->pages, pages, sizeof(req->pages[0]) * num_pages);
1883 req->num_pages = num_pages; 1983 req->num_pages = num_pages;
1984 fuse_page_descs_length_init(req, 0, req->num_pages);
1884 1985
1885 /* okay, let's send it to the client */ 1986 /* okay, let's send it to the client */
1886 req->in.h.opcode = FUSE_IOCTL; 1987 req->in.h.opcode = FUSE_IOCTL;
@@ -1981,7 +2082,7 @@ long fuse_ioctl_common(struct file *file, unsigned int cmd,
1981 struct inode *inode = file->f_dentry->d_inode; 2082 struct inode *inode = file->f_dentry->d_inode;
1982 struct fuse_conn *fc = get_fuse_conn(inode); 2083 struct fuse_conn *fc = get_fuse_conn(inode);
1983 2084
1984 if (!fuse_allow_task(fc, current)) 2085 if (!fuse_allow_current_process(fc))
1985 return -EACCES; 2086 return -EACCES;
1986 2087
1987 if (is_bad_inode(inode)) 2088 if (is_bad_inode(inode))
@@ -2066,6 +2167,7 @@ unsigned fuse_file_poll(struct file *file, poll_table *wait)
2066 return DEFAULT_POLLMASK; 2167 return DEFAULT_POLLMASK;
2067 2168
2068 poll_wait(file, &ff->poll_wait, wait); 2169 poll_wait(file, &ff->poll_wait, wait);
2170 inarg.events = (__u32)poll_requested_events(wait);
2069 2171
2070 /* 2172 /*
2071 * Ask for notification iff there's someone waiting for it. 2173 * Ask for notification iff there's someone waiting for it.
@@ -2076,7 +2178,7 @@ unsigned fuse_file_poll(struct file *file, poll_table *wait)
2076 fuse_register_polled_file(fc, ff); 2178 fuse_register_polled_file(fc, ff);
2077 } 2179 }
2078 2180
2079 req = fuse_get_req(fc); 2181 req = fuse_get_req_nopages(fc);
2080 if (IS_ERR(req)) 2182 if (IS_ERR(req))
2081 return POLLERR; 2183 return POLLERR;
2082 2184
@@ -2126,41 +2228,6 @@ int fuse_notify_poll_wakeup(struct fuse_conn *fc,
2126 return 0; 2228 return 0;
2127} 2229}
2128 2230
2129static ssize_t fuse_loop_dio(struct file *filp, const struct iovec *iov,
2130 unsigned long nr_segs, loff_t *ppos, int rw)
2131{
2132 const struct iovec *vector = iov;
2133 ssize_t ret = 0;
2134
2135 while (nr_segs > 0) {
2136 void __user *base;
2137 size_t len;
2138 ssize_t nr;
2139
2140 base = vector->iov_base;
2141 len = vector->iov_len;
2142 vector++;
2143 nr_segs--;
2144
2145 if (rw == WRITE)
2146 nr = __fuse_direct_write(filp, base, len, ppos);
2147 else
2148 nr = fuse_direct_read(filp, base, len, ppos);
2149
2150 if (nr < 0) {
2151 if (!ret)
2152 ret = nr;
2153 break;
2154 }
2155 ret += nr;
2156 if (nr != len)
2157 break;
2158 }
2159
2160 return ret;
2161}
2162
2163
2164static ssize_t 2231static ssize_t
2165fuse_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov, 2232fuse_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
2166 loff_t offset, unsigned long nr_segs) 2233 loff_t offset, unsigned long nr_segs)
@@ -2172,7 +2239,10 @@ fuse_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
2172 file = iocb->ki_filp; 2239 file = iocb->ki_filp;
2173 pos = offset; 2240 pos = offset;
2174 2241
2175 ret = fuse_loop_dio(file, iov, nr_segs, &pos, rw); 2242 if (rw == WRITE)
2243 ret = __fuse_direct_write(file, iov, nr_segs, &pos);
2244 else
2245 ret = __fuse_direct_read(file, iov, nr_segs, &pos);
2176 2246
2177 return ret; 2247 return ret;
2178} 2248}
@@ -2194,7 +2264,7 @@ static long fuse_file_fallocate(struct file *file, int mode, loff_t offset,
2194 if (fc->no_fallocate) 2264 if (fc->no_fallocate)
2195 return -EOPNOTSUPP; 2265 return -EOPNOTSUPP;
2196 2266
2197 req = fuse_get_req(fc); 2267 req = fuse_get_req_nopages(fc);
2198 if (IS_ERR(req)) 2268 if (IS_ERR(req))
2199 return PTR_ERR(req); 2269 return PTR_ERR(req);
2200 2270
diff --git a/fs/fuse/fuse_i.h b/fs/fuse/fuse_i.h
index e105a53fc72d..6aeba864f070 100644
--- a/fs/fuse/fuse_i.h
+++ b/fs/fuse/fuse_i.h
@@ -44,6 +44,9 @@
44 doing the mount will be allowed to access the filesystem */ 44 doing the mount will be allowed to access the filesystem */
45#define FUSE_ALLOW_OTHER (1 << 1) 45#define FUSE_ALLOW_OTHER (1 << 1)
46 46
47/** Number of page pointers embedded in fuse_req */
48#define FUSE_REQ_INLINE_PAGES 1
49
47/** List of active connections */ 50/** List of active connections */
48extern struct list_head fuse_conn_list; 51extern struct list_head fuse_conn_list;
49 52
@@ -103,6 +106,15 @@ struct fuse_inode {
103 106
104 /** List of writepage requestst (pending or sent) */ 107 /** List of writepage requestst (pending or sent) */
105 struct list_head writepages; 108 struct list_head writepages;
109
110 /** Miscellaneous bits describing inode state */
111 unsigned long state;
112};
113
114/** FUSE inode state bits */
115enum {
116 /** Advise readdirplus */
117 FUSE_I_ADVISE_RDPLUS,
106}; 118};
107 119
108struct fuse_conn; 120struct fuse_conn;
@@ -200,6 +212,12 @@ struct fuse_out {
200 struct fuse_arg args[3]; 212 struct fuse_arg args[3];
201}; 213};
202 214
215/** FUSE page descriptor */
216struct fuse_page_desc {
217 unsigned int length;
218 unsigned int offset;
219};
220
203/** The request state */ 221/** The request state */
204enum fuse_req_state { 222enum fuse_req_state {
205 FUSE_REQ_INIT = 0, 223 FUSE_REQ_INIT = 0,
@@ -291,14 +309,23 @@ struct fuse_req {
291 } misc; 309 } misc;
292 310
293 /** page vector */ 311 /** page vector */
294 struct page *pages[FUSE_MAX_PAGES_PER_REQ]; 312 struct page **pages;
313
314 /** page-descriptor vector */
315 struct fuse_page_desc *page_descs;
316
317 /** size of the 'pages' array */
318 unsigned max_pages;
319
320 /** inline page vector */
321 struct page *inline_pages[FUSE_REQ_INLINE_PAGES];
322
323 /** inline page-descriptor vector */
324 struct fuse_page_desc inline_page_descs[FUSE_REQ_INLINE_PAGES];
295 325
296 /** number of pages in vector */ 326 /** number of pages in vector */
297 unsigned num_pages; 327 unsigned num_pages;
298 328
299 /** offset of data on first page */
300 unsigned page_offset;
301
302 /** File used in the request (or NULL) */ 329 /** File used in the request (or NULL) */
303 struct fuse_file *ff; 330 struct fuse_file *ff;
304 331
@@ -487,6 +514,12 @@ struct fuse_conn {
487 /** Use enhanced/automatic page cache invalidation. */ 514 /** Use enhanced/automatic page cache invalidation. */
488 unsigned auto_inval_data:1; 515 unsigned auto_inval_data:1;
489 516
517 /** Does the filesystem support readdirplus? */
518 unsigned do_readdirplus:1;
519
520 /** Does the filesystem want adaptive readdirplus? */
521 unsigned readdirplus_auto:1;
522
490 /** The number of requests waiting for completion */ 523 /** The number of requests waiting for completion */
491 atomic_t num_waiting; 524 atomic_t num_waiting;
492 525
@@ -578,6 +611,9 @@ void fuse_queue_forget(struct fuse_conn *fc, struct fuse_forget_link *forget,
578 611
579struct fuse_forget_link *fuse_alloc_forget(void); 612struct fuse_forget_link *fuse_alloc_forget(void);
580 613
614/* Used by READDIRPLUS */
615void fuse_force_forget(struct file *file, u64 nodeid);
616
581/** 617/**
582 * Initialize READ or READDIR request 618 * Initialize READ or READDIR request
583 */ 619 */
@@ -658,9 +694,9 @@ void fuse_ctl_cleanup(void);
658/** 694/**
659 * Allocate a request 695 * Allocate a request
660 */ 696 */
661struct fuse_req *fuse_request_alloc(void); 697struct fuse_req *fuse_request_alloc(unsigned npages);
662 698
663struct fuse_req *fuse_request_alloc_nofs(void); 699struct fuse_req *fuse_request_alloc_nofs(unsigned npages);
664 700
665/** 701/**
666 * Free a request 702 * Free a request
@@ -668,14 +704,25 @@ struct fuse_req *fuse_request_alloc_nofs(void);
668void fuse_request_free(struct fuse_req *req); 704void fuse_request_free(struct fuse_req *req);
669 705
670/** 706/**
671 * Get a request, may fail with -ENOMEM 707 * Get a request, may fail with -ENOMEM,
708 * caller should specify # elements in req->pages[] explicitly
672 */ 709 */
673struct fuse_req *fuse_get_req(struct fuse_conn *fc); 710struct fuse_req *fuse_get_req(struct fuse_conn *fc, unsigned npages);
711
712/**
713 * Get a request, may fail with -ENOMEM,
714 * useful for callers who doesn't use req->pages[]
715 */
716static inline struct fuse_req *fuse_get_req_nopages(struct fuse_conn *fc)
717{
718 return fuse_get_req(fc, 0);
719}
674 720
675/** 721/**
676 * Gets a requests for a file operation, always succeeds 722 * Gets a requests for a file operation, always succeeds
677 */ 723 */
678struct fuse_req *fuse_get_req_nofail(struct fuse_conn *fc, struct file *file); 724struct fuse_req *fuse_get_req_nofail_nopages(struct fuse_conn *fc,
725 struct file *file);
679 726
680/** 727/**
681 * Decrement reference count of a request. If count goes to zero free 728 * Decrement reference count of a request. If count goes to zero free
@@ -739,9 +786,9 @@ void fuse_ctl_remove_conn(struct fuse_conn *fc);
739int fuse_valid_type(int m); 786int fuse_valid_type(int m);
740 787
741/** 788/**
742 * Is task allowed to perform filesystem operation? 789 * Is current process allowed to perform filesystem operation?
743 */ 790 */
744int fuse_allow_task(struct fuse_conn *fc, struct task_struct *task); 791int fuse_allow_current_process(struct fuse_conn *fc);
745 792
746u64 fuse_lock_owner_id(struct fuse_conn *fc, fl_owner_t id); 793u64 fuse_lock_owner_id(struct fuse_conn *fc, fl_owner_t id);
747 794
@@ -776,8 +823,9 @@ int fuse_reverse_inval_entry(struct super_block *sb, u64 parent_nodeid,
776 823
777int fuse_do_open(struct fuse_conn *fc, u64 nodeid, struct file *file, 824int fuse_do_open(struct fuse_conn *fc, u64 nodeid, struct file *file,
778 bool isdir); 825 bool isdir);
779ssize_t fuse_direct_io(struct file *file, const char __user *buf, 826ssize_t fuse_direct_io(struct file *file, const struct iovec *iov,
780 size_t count, loff_t *ppos, int write); 827 unsigned long nr_segs, size_t count, loff_t *ppos,
828 int write);
781long fuse_do_ioctl(struct file *file, unsigned int cmd, unsigned long arg, 829long fuse_do_ioctl(struct file *file, unsigned int cmd, unsigned long arg,
782 unsigned int flags); 830 unsigned int flags);
783long fuse_ioctl_common(struct file *file, unsigned int cmd, 831long fuse_ioctl_common(struct file *file, unsigned int cmd,
diff --git a/fs/fuse/inode.c b/fs/fuse/inode.c
index 73ca6b72beaf..01353ed75750 100644
--- a/fs/fuse/inode.c
+++ b/fs/fuse/inode.c
@@ -92,6 +92,7 @@ static struct inode *fuse_alloc_inode(struct super_block *sb)
92 fi->attr_version = 0; 92 fi->attr_version = 0;
93 fi->writectr = 0; 93 fi->writectr = 0;
94 fi->orig_ino = 0; 94 fi->orig_ino = 0;
95 fi->state = 0;
95 INIT_LIST_HEAD(&fi->write_files); 96 INIT_LIST_HEAD(&fi->write_files);
96 INIT_LIST_HEAD(&fi->queued_writes); 97 INIT_LIST_HEAD(&fi->queued_writes);
97 INIT_LIST_HEAD(&fi->writepages); 98 INIT_LIST_HEAD(&fi->writepages);
@@ -408,12 +409,12 @@ static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
408 struct fuse_statfs_out outarg; 409 struct fuse_statfs_out outarg;
409 int err; 410 int err;
410 411
411 if (!fuse_allow_task(fc, current)) { 412 if (!fuse_allow_current_process(fc)) {
412 buf->f_type = FUSE_SUPER_MAGIC; 413 buf->f_type = FUSE_SUPER_MAGIC;
413 return 0; 414 return 0;
414 } 415 }
415 416
416 req = fuse_get_req(fc); 417 req = fuse_get_req_nopages(fc);
417 if (IS_ERR(req)) 418 if (IS_ERR(req))
418 return PTR_ERR(req); 419 return PTR_ERR(req);
419 420
@@ -863,6 +864,10 @@ static void process_init_reply(struct fuse_conn *fc, struct fuse_req *req)
863 fc->dont_mask = 1; 864 fc->dont_mask = 1;
864 if (arg->flags & FUSE_AUTO_INVAL_DATA) 865 if (arg->flags & FUSE_AUTO_INVAL_DATA)
865 fc->auto_inval_data = 1; 866 fc->auto_inval_data = 1;
867 if (arg->flags & FUSE_DO_READDIRPLUS)
868 fc->do_readdirplus = 1;
869 if (arg->flags & FUSE_READDIRPLUS_AUTO)
870 fc->readdirplus_auto = 1;
866 } else { 871 } else {
867 ra_pages = fc->max_read / PAGE_CACHE_SIZE; 872 ra_pages = fc->max_read / PAGE_CACHE_SIZE;
868 fc->no_lock = 1; 873 fc->no_lock = 1;
@@ -889,7 +894,8 @@ static void fuse_send_init(struct fuse_conn *fc, struct fuse_req *req)
889 arg->flags |= FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC | 894 arg->flags |= FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
890 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK | 895 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
891 FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ | 896 FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ |
892 FUSE_FLOCK_LOCKS | FUSE_IOCTL_DIR | FUSE_AUTO_INVAL_DATA; 897 FUSE_FLOCK_LOCKS | FUSE_IOCTL_DIR | FUSE_AUTO_INVAL_DATA |
898 FUSE_DO_READDIRPLUS | FUSE_READDIRPLUS_AUTO;
893 req->in.h.opcode = FUSE_INIT; 899 req->in.h.opcode = FUSE_INIT;
894 req->in.numargs = 1; 900 req->in.numargs = 1;
895 req->in.args[0].size = sizeof(*arg); 901 req->in.args[0].size = sizeof(*arg);
@@ -1034,12 +1040,12 @@ static int fuse_fill_super(struct super_block *sb, void *data, int silent)
1034 /* only now - we want root dentry with NULL ->d_op */ 1040 /* only now - we want root dentry with NULL ->d_op */
1035 sb->s_d_op = &fuse_dentry_operations; 1041 sb->s_d_op = &fuse_dentry_operations;
1036 1042
1037 init_req = fuse_request_alloc(); 1043 init_req = fuse_request_alloc(0);
1038 if (!init_req) 1044 if (!init_req)
1039 goto err_put_root; 1045 goto err_put_root;
1040 1046
1041 if (is_bdev) { 1047 if (is_bdev) {
1042 fc->destroy_req = fuse_request_alloc(); 1048 fc->destroy_req = fuse_request_alloc(0);
1043 if (!fc->destroy_req) 1049 if (!fc->destroy_req)
1044 goto err_free_init_req; 1050 goto err_free_init_req;
1045 } 1051 }
diff --git a/include/linux/fs.h b/include/linux/fs.h
index 7617ee04f066..7d2e893ec3d1 100644
--- a/include/linux/fs.h
+++ b/include/linux/fs.h
@@ -301,7 +301,7 @@ size_t iov_iter_copy_from_user(struct page *page,
301 struct iov_iter *i, unsigned long offset, size_t bytes); 301 struct iov_iter *i, unsigned long offset, size_t bytes);
302void iov_iter_advance(struct iov_iter *i, size_t bytes); 302void iov_iter_advance(struct iov_iter *i, size_t bytes);
303int iov_iter_fault_in_readable(struct iov_iter *i, size_t bytes); 303int iov_iter_fault_in_readable(struct iov_iter *i, size_t bytes);
304size_t iov_iter_single_seg_count(struct iov_iter *i); 304size_t iov_iter_single_seg_count(const struct iov_iter *i);
305 305
306static inline void iov_iter_init(struct iov_iter *i, 306static inline void iov_iter_init(struct iov_iter *i,
307 const struct iovec *iov, unsigned long nr_segs, 307 const struct iovec *iov, unsigned long nr_segs,
diff --git a/include/uapi/linux/fuse.h b/include/uapi/linux/fuse.h
index d8c713e148e3..4c43b4448792 100644
--- a/include/uapi/linux/fuse.h
+++ b/include/uapi/linux/fuse.h
@@ -1,9 +1,35 @@
1/* 1/*
2 FUSE: Filesystem in Userspace 2 This file defines the kernel interface of FUSE
3 Copyright (C) 2001-2008 Miklos Szeredi <miklos@szeredi.hu> 3 Copyright (C) 2001-2008 Miklos Szeredi <miklos@szeredi.hu>
4 4
5 This program can be distributed under the terms of the GNU GPL. 5 This program can be distributed under the terms of the GNU GPL.
6 See the file COPYING. 6 See the file COPYING.
7
8 This -- and only this -- header file may also be distributed under
9 the terms of the BSD Licence as follows:
10
11 Copyright (C) 2001-2007 Miklos Szeredi. All rights reserved.
12
13 Redistribution and use in source and binary forms, with or without
14 modification, are permitted provided that the following conditions
15 are met:
16 1. Redistributions of source code must retain the above copyright
17 notice, this list of conditions and the following disclaimer.
18 2. Redistributions in binary form must reproduce the above copyright
19 notice, this list of conditions and the following disclaimer in the
20 documentation and/or other materials provided with the distribution.
21
22 THIS SOFTWARE IS PROVIDED BY AUTHOR AND CONTRIBUTORS ``AS IS'' AND
23 ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 ARE DISCLAIMED. IN NO EVENT SHALL AUTHOR OR CONTRIBUTORS BE LIABLE
26 FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 SUCH DAMAGE.
7*/ 33*/
8 34
9/* 35/*
@@ -60,12 +86,25 @@
60 * 86 *
61 * 7.20 87 * 7.20
62 * - add FUSE_AUTO_INVAL_DATA 88 * - add FUSE_AUTO_INVAL_DATA
89 *
90 * 7.21
91 * - add FUSE_READDIRPLUS
92 * - send the requested events in POLL request
63 */ 93 */
64 94
65#ifndef _LINUX_FUSE_H 95#ifndef _LINUX_FUSE_H
66#define _LINUX_FUSE_H 96#define _LINUX_FUSE_H
67 97
98#ifdef __linux__
68#include <linux/types.h> 99#include <linux/types.h>
100#else
101#include <stdint.h>
102#define __u64 uint64_t
103#define __s64 int64_t
104#define __u32 uint32_t
105#define __s32 int32_t
106#define __u16 uint16_t
107#endif
69 108
70/* 109/*
71 * Version negotiation: 110 * Version negotiation:
@@ -91,7 +130,7 @@
91#define FUSE_KERNEL_VERSION 7 130#define FUSE_KERNEL_VERSION 7
92 131
93/** Minor version number of this interface */ 132/** Minor version number of this interface */
94#define FUSE_KERNEL_MINOR_VERSION 20 133#define FUSE_KERNEL_MINOR_VERSION 21
95 134
96/** The node ID of the root inode */ 135/** The node ID of the root inode */
97#define FUSE_ROOT_ID 1 136#define FUSE_ROOT_ID 1
@@ -179,6 +218,8 @@ struct fuse_file_lock {
179 * FUSE_FLOCK_LOCKS: remote locking for BSD style file locks 218 * FUSE_FLOCK_LOCKS: remote locking for BSD style file locks
180 * FUSE_HAS_IOCTL_DIR: kernel supports ioctl on directories 219 * FUSE_HAS_IOCTL_DIR: kernel supports ioctl on directories
181 * FUSE_AUTO_INVAL_DATA: automatically invalidate cached pages 220 * FUSE_AUTO_INVAL_DATA: automatically invalidate cached pages
221 * FUSE_DO_READDIRPLUS: do READDIRPLUS (READDIR+LOOKUP in one)
222 * FUSE_READDIRPLUS_AUTO: adaptive readdirplus
182 */ 223 */
183#define FUSE_ASYNC_READ (1 << 0) 224#define FUSE_ASYNC_READ (1 << 0)
184#define FUSE_POSIX_LOCKS (1 << 1) 225#define FUSE_POSIX_LOCKS (1 << 1)
@@ -193,6 +234,8 @@ struct fuse_file_lock {
193#define FUSE_FLOCK_LOCKS (1 << 10) 234#define FUSE_FLOCK_LOCKS (1 << 10)
194#define FUSE_HAS_IOCTL_DIR (1 << 11) 235#define FUSE_HAS_IOCTL_DIR (1 << 11)
195#define FUSE_AUTO_INVAL_DATA (1 << 12) 236#define FUSE_AUTO_INVAL_DATA (1 << 12)
237#define FUSE_DO_READDIRPLUS (1 << 13)
238#define FUSE_READDIRPLUS_AUTO (1 << 14)
196 239
197/** 240/**
198 * CUSE INIT request/reply flags 241 * CUSE INIT request/reply flags
@@ -299,6 +342,7 @@ enum fuse_opcode {
299 FUSE_NOTIFY_REPLY = 41, 342 FUSE_NOTIFY_REPLY = 41,
300 FUSE_BATCH_FORGET = 42, 343 FUSE_BATCH_FORGET = 42,
301 FUSE_FALLOCATE = 43, 344 FUSE_FALLOCATE = 43,
345 FUSE_READDIRPLUS = 44,
302 346
303 /* CUSE specific operations */ 347 /* CUSE specific operations */
304 CUSE_INIT = 4096, 348 CUSE_INIT = 4096,
@@ -580,7 +624,7 @@ struct fuse_poll_in {
580 __u64 fh; 624 __u64 fh;
581 __u64 kh; 625 __u64 kh;
582 __u32 flags; 626 __u32 flags;
583 __u32 padding; 627 __u32 events;
584}; 628};
585 629
586struct fuse_poll_out { 630struct fuse_poll_out {
@@ -630,6 +674,16 @@ struct fuse_dirent {
630#define FUSE_DIRENT_SIZE(d) \ 674#define FUSE_DIRENT_SIZE(d) \
631 FUSE_DIRENT_ALIGN(FUSE_NAME_OFFSET + (d)->namelen) 675 FUSE_DIRENT_ALIGN(FUSE_NAME_OFFSET + (d)->namelen)
632 676
677struct fuse_direntplus {
678 struct fuse_entry_out entry_out;
679 struct fuse_dirent dirent;
680};
681
682#define FUSE_NAME_OFFSET_DIRENTPLUS \
683 offsetof(struct fuse_direntplus, dirent.name)
684#define FUSE_DIRENTPLUS_SIZE(d) \
685 FUSE_DIRENT_ALIGN(FUSE_NAME_OFFSET_DIRENTPLUS + (d)->dirent.namelen)
686
633struct fuse_notify_inval_inode_out { 687struct fuse_notify_inval_inode_out {
634 __u64 ino; 688 __u64 ino;
635 __s64 off; 689 __s64 off;
diff --git a/mm/filemap.c b/mm/filemap.c
index 83efee76a5c0..24a7ea583f0c 100644
--- a/mm/filemap.c
+++ b/mm/filemap.c
@@ -2056,7 +2056,7 @@ EXPORT_SYMBOL(iov_iter_fault_in_readable);
2056/* 2056/*
2057 * Return the count of just the current iov_iter segment. 2057 * Return the count of just the current iov_iter segment.
2058 */ 2058 */
2059size_t iov_iter_single_seg_count(struct iov_iter *i) 2059size_t iov_iter_single_seg_count(const struct iov_iter *i)
2060{ 2060{
2061 const struct iovec *iov = i->iov; 2061 const struct iovec *iov = i->iov;
2062 if (i->nr_segs == 1) 2062 if (i->nr_segs == 1)
generated by cgit v1.2.2 (git 2.25.0) at 2025-07-16 08:22:51 -0400