aboutsummaryrefslogtreecommitdiffstats
path: root/fs/ceph
diff options
context:
space:
mode:
authorSage Weil <sage@newdream.net>2011-11-03 12:23:36 -0400
committerSage Weil <sage@newdream.net>2011-11-06 00:10:10 -0400
commitc6ffe10015f4e6fba8a915318b319c43aed1836f (patch)
treedd39730aeb6d00713ca27a1bf34b116d31479151 /fs/ceph
parentb58dc4100b9190f2cb437f1f67ffcb9f9acc4923 (diff)
ceph: use new D_COMPLETE dentry flag
We used to use a flag on the directory inode to track whether the dcache contents for a directory were a complete cached copy. Switch to a dentry flag CEPH_D_COMPLETE that is safely updated by ->d_prune(). Signed-off-by: Sage Weil <sage@newdream.net>
Diffstat (limited to 'fs/ceph')
-rw-r--r--fs/ceph/caps.c8
-rw-r--r--fs/ceph/dir.c59
-rw-r--r--fs/ceph/inode.c8
-rw-r--r--fs/ceph/mds_client.c6
-rw-r--r--fs/ceph/super.h10
5 files changed, 68 insertions, 23 deletions
diff --git a/fs/ceph/caps.c b/fs/ceph/caps.c
index 15b21e35078a..0f327c6c9679 100644
--- a/fs/ceph/caps.c
+++ b/fs/ceph/caps.c
@@ -487,17 +487,15 @@ static void __check_cap_issue(struct ceph_inode_info *ci, struct ceph_cap *cap,
487 ci->i_rdcache_gen++; 487 ci->i_rdcache_gen++;
488 488
489 /* 489 /*
490 * if we are newly issued FILE_SHARED, clear I_COMPLETE; we 490 * if we are newly issued FILE_SHARED, clear D_COMPLETE; we
491 * don't know what happened to this directory while we didn't 491 * don't know what happened to this directory while we didn't
492 * have the cap. 492 * have the cap.
493 */ 493 */
494 if ((issued & CEPH_CAP_FILE_SHARED) && 494 if ((issued & CEPH_CAP_FILE_SHARED) &&
495 (had & CEPH_CAP_FILE_SHARED) == 0) { 495 (had & CEPH_CAP_FILE_SHARED) == 0) {
496 ci->i_shared_gen++; 496 ci->i_shared_gen++;
497 if (S_ISDIR(ci->vfs_inode.i_mode)) { 497 if (S_ISDIR(ci->vfs_inode.i_mode))
498 dout(" marking %p NOT complete\n", &ci->vfs_inode); 498 ceph_dir_clear_complete(&ci->vfs_inode);
499 ci->i_ceph_flags &= ~CEPH_I_COMPLETE;
500 }
501 } 499 }
502} 500}
503 501
diff --git a/fs/ceph/dir.c b/fs/ceph/dir.c
index 97fb155b3242..2abd0dfad7f8 100644
--- a/fs/ceph/dir.c
+++ b/fs/ceph/dir.c
@@ -108,7 +108,7 @@ static unsigned fpos_off(loff_t p)
108 * falling back to a "normal" sync readdir if any dentries in the dir 108 * falling back to a "normal" sync readdir if any dentries in the dir
109 * are dropped. 109 * are dropped.
110 * 110 *
111 * I_COMPLETE tells indicates we have all dentries in the dir. It is 111 * D_COMPLETE tells indicates we have all dentries in the dir. It is
112 * defined IFF we hold CEPH_CAP_FILE_SHARED (which will be revoked by 112 * defined IFF we hold CEPH_CAP_FILE_SHARED (which will be revoked by
113 * the MDS if/when the directory is modified). 113 * the MDS if/when the directory is modified).
114 */ 114 */
@@ -199,8 +199,8 @@ more:
199 filp->f_pos++; 199 filp->f_pos++;
200 200
201 /* make sure a dentry wasn't dropped while we didn't have parent lock */ 201 /* make sure a dentry wasn't dropped while we didn't have parent lock */
202 if (!ceph_i_test(dir, CEPH_I_COMPLETE)) { 202 if (!ceph_dir_test_complete(dir)) {
203 dout(" lost I_COMPLETE on %p; falling back to mds\n", dir); 203 dout(" lost D_COMPLETE on %p; falling back to mds\n", dir);
204 err = -EAGAIN; 204 err = -EAGAIN;
205 goto out; 205 goto out;
206 } 206 }
@@ -285,7 +285,7 @@ static int ceph_readdir(struct file *filp, void *dirent, filldir_t filldir)
285 if ((filp->f_pos == 2 || fi->dentry) && 285 if ((filp->f_pos == 2 || fi->dentry) &&
286 !ceph_test_mount_opt(fsc, NOASYNCREADDIR) && 286 !ceph_test_mount_opt(fsc, NOASYNCREADDIR) &&
287 ceph_snap(inode) != CEPH_SNAPDIR && 287 ceph_snap(inode) != CEPH_SNAPDIR &&
288 (ci->i_ceph_flags & CEPH_I_COMPLETE) && 288 ceph_dir_test_complete(inode) &&
289 __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1)) { 289 __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1)) {
290 spin_unlock(&inode->i_lock); 290 spin_unlock(&inode->i_lock);
291 err = __dcache_readdir(filp, dirent, filldir); 291 err = __dcache_readdir(filp, dirent, filldir);
@@ -351,7 +351,7 @@ more:
351 351
352 if (!req->r_did_prepopulate) { 352 if (!req->r_did_prepopulate) {
353 dout("readdir !did_prepopulate"); 353 dout("readdir !did_prepopulate");
354 fi->dir_release_count--; /* preclude I_COMPLETE */ 354 fi->dir_release_count--; /* preclude D_COMPLETE */
355 } 355 }
356 356
357 /* note next offset and last dentry name */ 357 /* note next offset and last dentry name */
@@ -430,8 +430,7 @@ more:
430 */ 430 */
431 spin_lock(&inode->i_lock); 431 spin_lock(&inode->i_lock);
432 if (ci->i_release_count == fi->dir_release_count) { 432 if (ci->i_release_count == fi->dir_release_count) {
433 dout(" marking %p complete\n", inode); 433 ceph_dir_set_complete(inode);
434 /* ci->i_ceph_flags |= CEPH_I_COMPLETE; */
435 ci->i_max_offset = filp->f_pos; 434 ci->i_max_offset = filp->f_pos;
436 } 435 }
437 spin_unlock(&inode->i_lock); 436 spin_unlock(&inode->i_lock);
@@ -614,7 +613,7 @@ static struct dentry *ceph_lookup(struct inode *dir, struct dentry *dentry,
614 fsc->mount_options->snapdir_name, 613 fsc->mount_options->snapdir_name,
615 dentry->d_name.len) && 614 dentry->d_name.len) &&
616 !is_root_ceph_dentry(dir, dentry) && 615 !is_root_ceph_dentry(dir, dentry) &&
617 (ci->i_ceph_flags & CEPH_I_COMPLETE) && 616 ceph_dir_test_complete(dir) &&
618 (__ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1))) { 617 (__ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1))) {
619 spin_unlock(&dir->i_lock); 618 spin_unlock(&dir->i_lock);
620 dout(" dir %p complete, -ENOENT\n", dir); 619 dout(" dir %p complete, -ENOENT\n", dir);
@@ -934,7 +933,7 @@ static int ceph_rename(struct inode *old_dir, struct dentry *old_dentry,
934 */ 933 */
935 934
936 /* d_move screws up d_subdirs order */ 935 /* d_move screws up d_subdirs order */
937 ceph_i_clear(new_dir, CEPH_I_COMPLETE); 936 ceph_dir_clear_complete(new_dir);
938 937
939 d_move(old_dentry, new_dentry); 938 d_move(old_dentry, new_dentry);
940 939
@@ -1093,6 +1092,48 @@ static int ceph_snapdir_d_revalidate(struct dentry *dentry,
1093} 1092}
1094 1093
1095/* 1094/*
1095 * Set/clear/test dir complete flag on the dir's dentry.
1096 */
1097static struct dentry * __d_find_any_alias(struct inode *inode)
1098{
1099 struct dentry *alias;
1100
1101 if (list_empty(&inode->i_dentry))
1102 return NULL;
1103 alias = list_first_entry(&inode->i_dentry, struct dentry, d_alias);
1104 return alias;
1105}
1106
1107void ceph_dir_set_complete(struct inode *inode)
1108{
1109 struct dentry *dentry = __d_find_any_alias(inode);
1110
1111 if (dentry && ceph_dentry(dentry)) {
1112 dout(" marking %p (%p) complete\n", inode, dentry);
1113 set_bit(CEPH_D_COMPLETE, &ceph_dentry(dentry)->flags);
1114 }
1115}
1116
1117void ceph_dir_clear_complete(struct inode *inode)
1118{
1119 struct dentry *dentry = __d_find_any_alias(inode);
1120
1121 if (dentry && ceph_dentry(dentry)) {
1122 dout(" marking %p (%p) NOT complete\n", inode, dentry);
1123 clear_bit(CEPH_D_COMPLETE, &ceph_dentry(dentry)->flags);
1124 }
1125}
1126
1127bool ceph_dir_test_complete(struct inode *inode)
1128{
1129 struct dentry *dentry = __d_find_any_alias(inode);
1130
1131 if (dentry && ceph_dentry(dentry))
1132 return test_bit(CEPH_D_COMPLETE, &ceph_dentry(dentry)->flags);
1133 return false;
1134}
1135
1136/*
1096 * When the VFS prunes a dentry from the cache, we need to clear the 1137 * When the VFS prunes a dentry from the cache, we need to clear the
1097 * complete flag on the parent directory. 1138 * complete flag on the parent directory.
1098 * 1139 *
diff --git a/fs/ceph/inode.c b/fs/ceph/inode.c
index 1616a0d37cbd..e392bfce84a3 100644
--- a/fs/ceph/inode.c
+++ b/fs/ceph/inode.c
@@ -771,9 +771,9 @@ no_change:
771 ceph_snap(inode) == CEPH_NOSNAP && 771 ceph_snap(inode) == CEPH_NOSNAP &&
772 (le32_to_cpu(info->cap.caps) & CEPH_CAP_FILE_SHARED) && 772 (le32_to_cpu(info->cap.caps) & CEPH_CAP_FILE_SHARED) &&
773 (issued & CEPH_CAP_FILE_EXCL) == 0 && 773 (issued & CEPH_CAP_FILE_EXCL) == 0 &&
774 (ci->i_ceph_flags & CEPH_I_COMPLETE) == 0) { 774 !ceph_dir_test_complete(inode)) {
775 dout(" marking %p complete (empty)\n", inode); 775 dout(" marking %p complete (empty)\n", inode);
776 /* ci->i_ceph_flags |= CEPH_I_COMPLETE; */ 776 ceph_dir_set_complete(inode);
777 ci->i_max_offset = 2; 777 ci->i_max_offset = 2;
778 } 778 }
779 779
@@ -856,7 +856,7 @@ static void ceph_set_dentry_offset(struct dentry *dn)
856 di = ceph_dentry(dn); 856 di = ceph_dentry(dn);
857 857
858 spin_lock(&inode->i_lock); 858 spin_lock(&inode->i_lock);
859 if ((ceph_inode(inode)->i_ceph_flags & CEPH_I_COMPLETE) == 0) { 859 if (!ceph_dir_test_complete(inode)) {
860 spin_unlock(&inode->i_lock); 860 spin_unlock(&inode->i_lock);
861 return; 861 return;
862 } 862 }
@@ -1056,7 +1056,7 @@ int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req,
1056 * d_move() puts the renamed dentry at the end of 1056 * d_move() puts the renamed dentry at the end of
1057 * d_subdirs. We need to assign it an appropriate 1057 * d_subdirs. We need to assign it an appropriate
1058 * directory offset so we can behave when holding 1058 * directory offset so we can behave when holding
1059 * I_COMPLETE. 1059 * D_COMPLETE.
1060 */ 1060 */
1061 ceph_set_dentry_offset(req->r_old_dentry); 1061 ceph_set_dentry_offset(req->r_old_dentry);
1062 dout("dn %p gets new offset %lld\n", req->r_old_dentry, 1062 dout("dn %p gets new offset %lld\n", req->r_old_dentry,
diff --git a/fs/ceph/mds_client.c b/fs/ceph/mds_client.c
index 1d72f15fe9f4..b4c3efa163c2 100644
--- a/fs/ceph/mds_client.c
+++ b/fs/ceph/mds_client.c
@@ -2002,7 +2002,7 @@ out:
2002} 2002}
2003 2003
2004/* 2004/*
2005 * Invalidate dir I_COMPLETE, dentry lease state on an aborted MDS 2005 * Invalidate dir D_COMPLETE, dentry lease state on an aborted MDS
2006 * namespace request. 2006 * namespace request.
2007 */ 2007 */
2008void ceph_invalidate_dir_request(struct ceph_mds_request *req) 2008void ceph_invalidate_dir_request(struct ceph_mds_request *req)
@@ -2010,9 +2010,9 @@ void ceph_invalidate_dir_request(struct ceph_mds_request *req)
2010 struct inode *inode = req->r_locked_dir; 2010 struct inode *inode = req->r_locked_dir;
2011 struct ceph_inode_info *ci = ceph_inode(inode); 2011 struct ceph_inode_info *ci = ceph_inode(inode);
2012 2012
2013 dout("invalidate_dir_request %p (I_COMPLETE, lease(s))\n", inode); 2013 dout("invalidate_dir_request %p (D_COMPLETE, lease(s))\n", inode);
2014 spin_lock(&inode->i_lock); 2014 spin_lock(&inode->i_lock);
2015 ci->i_ceph_flags &= ~CEPH_I_COMPLETE; 2015 ceph_dir_clear_complete(inode);
2016 ci->i_release_count++; 2016 ci->i_release_count++;
2017 spin_unlock(&inode->i_lock); 2017 spin_unlock(&inode->i_lock);
2018 2018
diff --git a/fs/ceph/super.h b/fs/ceph/super.h
index 98e60693c9a1..01bf189e08a9 100644
--- a/fs/ceph/super.h
+++ b/fs/ceph/super.h
@@ -264,7 +264,7 @@ struct ceph_inode_info {
264 struct timespec i_rctime; 264 struct timespec i_rctime;
265 u64 i_rbytes, i_rfiles, i_rsubdirs; 265 u64 i_rbytes, i_rfiles, i_rsubdirs;
266 u64 i_files, i_subdirs; 266 u64 i_files, i_subdirs;
267 u64 i_max_offset; /* largest readdir offset, set with I_COMPLETE */ 267 u64 i_max_offset; /* largest readdir offset, set with D_COMPLETE */
268 268
269 struct rb_root i_fragtree; 269 struct rb_root i_fragtree;
270 struct mutex i_fragtree_mutex; 270 struct mutex i_fragtree_mutex;
@@ -429,7 +429,6 @@ static inline struct inode *ceph_find_inode(struct super_block *sb,
429/* 429/*
430 * Ceph inode. 430 * Ceph inode.
431 */ 431 */
432#define CEPH_I_COMPLETE 1 /* we have complete directory cached */
433#define CEPH_I_NODELAY 4 /* do not delay cap release */ 432#define CEPH_I_NODELAY 4 /* do not delay cap release */
434#define CEPH_I_FLUSH 8 /* do not delay flush of dirty metadata */ 433#define CEPH_I_FLUSH 8 /* do not delay flush of dirty metadata */
435#define CEPH_I_NOFLUSH 16 /* do not flush dirty caps */ 434#define CEPH_I_NOFLUSH 16 /* do not flush dirty caps */
@@ -487,6 +486,13 @@ static inline loff_t ceph_make_fpos(unsigned frag, unsigned off)
487} 486}
488 487
489/* 488/*
489 * set/clear directory D_COMPLETE flag
490 */
491void ceph_dir_set_complete(struct inode *inode);
492void ceph_dir_clear_complete(struct inode *inode);
493bool ceph_dir_test_complete(struct inode *inode);
494
495/*
490 * caps helpers 496 * caps helpers
491 */ 497 */
492static inline bool __ceph_is_any_real_caps(struct ceph_inode_info *ci) 498static inline bool __ceph_is_any_real_caps(struct ceph_inode_info *ci)