1 #include <linux/ceph/ceph_debug.h>
3 #include <linux/spinlock.h>
4 #include <linux/fs_struct.h>
5 #include <linux/namei.h>
6 #include <linux/slab.h>
7 #include <linux/sched.h>
10 #include "mds_client.h"
13 * Directory operations: readdir, lookup, create, link, unlink,
18 * Ceph MDS operations are specified in terms of a base ino and
19 * relative path. Thus, the client can specify an operation on a
20 * specific inode (e.g., a getattr due to fstat(2)), or as a path
21 * relative to, say, the root directory.
23 * Normally, we limit ourselves to strict inode ops (no path component)
24 * or dentry operations (a single path component relative to an ino). The
25 * exception to this is open_root_dentry(), which will open the mount
29 const struct inode_operations ceph_dir_iops;
30 const struct file_operations ceph_dir_fops;
31 const struct dentry_operations ceph_dentry_ops;
34 * Initialize ceph dentry state.
36 int ceph_init_dentry(struct dentry *dentry)
38 struct ceph_dentry_info *di;
43 if (dentry->d_parent == NULL || /* nfs fh_to_dentry */
44 ceph_snap(dentry->d_parent->d_inode) == CEPH_NOSNAP)
45 dentry->d_op = &ceph_dentry_ops;
46 else if (ceph_snap(dentry->d_parent->d_inode) == CEPH_SNAPDIR)
47 dentry->d_op = &ceph_snapdir_dentry_ops;
49 dentry->d_op = &ceph_snap_dentry_ops;
51 di = kmem_cache_alloc(ceph_dentry_cachep, GFP_NOFS | __GFP_ZERO);
53 return -ENOMEM; /* oh well */
55 spin_lock(&dentry->d_lock);
56 if (dentry->d_fsdata) {
58 kmem_cache_free(ceph_dentry_cachep, di);
62 di->lease_session = NULL;
63 dentry->d_fsdata = di;
64 dentry->d_time = jiffies;
65 ceph_dentry_lru_add(dentry);
67 spin_unlock(&dentry->d_lock);
74 * for readdir, we encode the directory frag and offset within that
77 static unsigned fpos_frag(loff_t p)
81 static unsigned fpos_off(loff_t p)
83 return p & 0xffffffff;
87 * When possible, we try to satisfy a readdir by peeking at the
88 * dcache. We make this work by carefully ordering dentries on
89 * d_u.d_child when we initially get results back from the MDS, and
90 * falling back to a "normal" sync readdir if any dentries in the dir
93 * I_COMPLETE tells indicates we have all dentries in the dir. It is
94 * defined IFF we hold CEPH_CAP_FILE_SHARED (which will be revoked by
95 * the MDS if/when the directory is modified).
97 static int __dcache_readdir(struct file *filp,
98 void *dirent, filldir_t filldir)
100 struct ceph_file_info *fi = filp->private_data;
101 struct dentry *parent = filp->f_dentry;
102 struct inode *dir = parent->d_inode;
104 struct dentry *dentry, *last;
105 struct ceph_dentry_info *di;
108 /* claim ref on last dentry we returned */
112 dout("__dcache_readdir %p at %llu (last %p)\n", dir, filp->f_pos,
115 spin_lock(&dcache_lock);
116 spin_lock(&parent->d_lock);
118 /* start at beginning? */
119 if (filp->f_pos == 2 || last == NULL ||
120 filp->f_pos < ceph_dentry(last)->offset) {
121 if (list_empty(&parent->d_subdirs))
123 p = parent->d_subdirs.prev;
124 dout(" initial p %p/%p\n", p->prev, p->next);
126 p = last->d_u.d_child.prev;
130 dentry = list_entry(p, struct dentry, d_u.d_child);
131 di = ceph_dentry(dentry);
133 dout(" p %p/%p %s d_subdirs %p/%p\n", p->prev, p->next,
134 d_unhashed(dentry) ? "!hashed" : "hashed",
135 parent->d_subdirs.prev, parent->d_subdirs.next);
136 if (p == &parent->d_subdirs) {
140 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
141 if (!d_unhashed(dentry) && dentry->d_inode &&
142 ceph_snap(dentry->d_inode) != CEPH_SNAPDIR &&
143 ceph_ino(dentry->d_inode) != CEPH_INO_CEPH &&
144 filp->f_pos <= di->offset)
146 dout(" skipping %p %.*s at %llu (%llu)%s%s\n", dentry,
147 dentry->d_name.len, dentry->d_name.name, di->offset,
148 filp->f_pos, d_unhashed(dentry) ? " unhashed" : "",
149 !dentry->d_inode ? " null" : "");
150 spin_unlock(&dentry->d_lock);
152 dentry = list_entry(p, struct dentry, d_u.d_child);
153 di = ceph_dentry(dentry);
157 spin_unlock(&dentry->d_lock);
158 spin_unlock(&parent->d_lock);
159 spin_unlock(&dcache_lock);
161 dout(" %llu (%llu) dentry %p %.*s %p\n", di->offset, filp->f_pos,
162 dentry, dentry->d_name.len, dentry->d_name.name, dentry->d_inode);
163 filp->f_pos = di->offset;
164 err = filldir(dirent, dentry->d_name.name,
165 dentry->d_name.len, di->offset,
166 dentry->d_inode->i_ino,
167 dentry->d_inode->i_mode >> 12);
171 /* remember our position */
173 fi->next_offset = di->offset;
185 /* make sure a dentry wasn't dropped while we didn't have dcache_lock */
186 if (!ceph_i_test(dir, CEPH_I_COMPLETE)) {
187 dout(" lost I_COMPLETE on %p; falling back to mds\n", dir);
192 spin_lock(&dcache_lock);
193 spin_lock(&parent->d_lock);
194 p = p->prev; /* advance to next dentry */
198 spin_unlock(&parent->d_lock);
199 spin_unlock(&dcache_lock);
207 * make note of the last dentry we read, so we can
208 * continue at the same lexicographical point,
209 * regardless of what dir changes take place on the
212 static int note_last_dentry(struct ceph_file_info *fi, const char *name,
215 kfree(fi->last_name);
216 fi->last_name = kmalloc(len+1, GFP_NOFS);
219 memcpy(fi->last_name, name, len);
220 fi->last_name[len] = 0;
221 dout("note_last_dentry '%s'\n", fi->last_name);
225 static int ceph_readdir(struct file *filp, void *dirent, filldir_t filldir)
227 struct ceph_file_info *fi = filp->private_data;
228 struct inode *inode = filp->f_dentry->d_inode;
229 struct ceph_inode_info *ci = ceph_inode(inode);
230 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
231 struct ceph_mds_client *mdsc = fsc->mdsc;
232 unsigned frag = fpos_frag(filp->f_pos);
233 int off = fpos_off(filp->f_pos);
236 struct ceph_mds_reply_info_parsed *rinfo;
237 const int max_entries = fsc->mount_options->max_readdir;
238 const int max_bytes = fsc->mount_options->max_readdir_bytes;
240 dout("readdir %p filp %p frag %u off %u\n", inode, filp, frag, off);
244 /* always start with . and .. */
245 if (filp->f_pos == 0) {
246 /* note dir version at start of readdir so we can tell
247 * if any dentries get dropped */
248 fi->dir_release_count = ci->i_release_count;
250 dout("readdir off 0 -> '.'\n");
251 if (filldir(dirent, ".", 1, ceph_make_fpos(0, 0),
252 inode->i_ino, inode->i_mode >> 12) < 0)
257 if (filp->f_pos == 1) {
258 dout("readdir off 1 -> '..'\n");
259 if (filldir(dirent, "..", 2, ceph_make_fpos(0, 1),
260 filp->f_dentry->d_parent->d_inode->i_ino,
261 inode->i_mode >> 12) < 0)
267 /* can we use the dcache? */
268 spin_lock(&inode->i_lock);
269 if ((filp->f_pos == 2 || fi->dentry) &&
270 !ceph_test_mount_opt(fsc, NOASYNCREADDIR) &&
271 ceph_snap(inode) != CEPH_SNAPDIR &&
272 (ci->i_ceph_flags & CEPH_I_COMPLETE) &&
273 __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1)) {
274 spin_unlock(&inode->i_lock);
275 err = __dcache_readdir(filp, dirent, filldir);
279 spin_unlock(&inode->i_lock);
282 err = note_last_dentry(fi, fi->dentry->d_name.name,
283 fi->dentry->d_name.len);
290 /* proceed with a normal readdir */
293 /* do we have the correct frag content buffered? */
294 if (fi->frag != frag || fi->last_readdir == NULL) {
295 struct ceph_mds_request *req;
296 int op = ceph_snap(inode) == CEPH_SNAPDIR ?
297 CEPH_MDS_OP_LSSNAP : CEPH_MDS_OP_READDIR;
299 /* discard old result, if any */
300 if (fi->last_readdir) {
301 ceph_mdsc_put_request(fi->last_readdir);
302 fi->last_readdir = NULL;
305 /* requery frag tree, as the frag topology may have changed */
306 frag = ceph_choose_frag(ceph_inode(inode), frag, NULL, NULL);
308 dout("readdir fetching %llx.%llx frag %x offset '%s'\n",
309 ceph_vinop(inode), frag, fi->last_name);
310 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
313 req->r_inode = igrab(inode);
314 req->r_dentry = dget(filp->f_dentry);
315 /* hints to request -> mds selection code */
316 req->r_direct_mode = USE_AUTH_MDS;
317 req->r_direct_hash = ceph_frag_value(frag);
318 req->r_direct_is_hash = true;
319 req->r_path2 = kstrdup(fi->last_name, GFP_NOFS);
320 req->r_readdir_offset = fi->next_offset;
321 req->r_args.readdir.frag = cpu_to_le32(frag);
322 req->r_args.readdir.max_entries = cpu_to_le32(max_entries);
323 req->r_args.readdir.max_bytes = cpu_to_le32(max_bytes);
324 req->r_num_caps = max_entries + 1;
325 err = ceph_mdsc_do_request(mdsc, NULL, req);
327 ceph_mdsc_put_request(req);
330 dout("readdir got and parsed readdir result=%d"
331 " on frag %x, end=%d, complete=%d\n", err, frag,
332 (int)req->r_reply_info.dir_end,
333 (int)req->r_reply_info.dir_complete);
335 if (!req->r_did_prepopulate) {
336 dout("readdir !did_prepopulate");
337 fi->dir_release_count--; /* preclude I_COMPLETE */
340 /* note next offset and last dentry name */
341 fi->offset = fi->next_offset;
342 fi->last_readdir = req;
344 if (req->r_reply_info.dir_end) {
345 kfree(fi->last_name);
346 fi->last_name = NULL;
347 if (ceph_frag_is_rightmost(frag))
352 rinfo = &req->r_reply_info;
353 err = note_last_dentry(fi,
354 rinfo->dir_dname[rinfo->dir_nr-1],
355 rinfo->dir_dname_len[rinfo->dir_nr-1]);
358 fi->next_offset += rinfo->dir_nr;
362 rinfo = &fi->last_readdir->r_reply_info;
363 dout("readdir frag %x num %d off %d chunkoff %d\n", frag,
364 rinfo->dir_nr, off, fi->offset);
365 while (off - fi->offset >= 0 && off - fi->offset < rinfo->dir_nr) {
366 u64 pos = ceph_make_fpos(frag, off);
367 struct ceph_mds_reply_inode *in =
368 rinfo->dir_in[off - fi->offset].in;
369 struct ceph_vino vino;
372 dout("readdir off %d (%d/%d) -> %lld '%.*s' %p\n",
373 off, off - fi->offset, rinfo->dir_nr, pos,
374 rinfo->dir_dname_len[off - fi->offset],
375 rinfo->dir_dname[off - fi->offset], in);
377 ftype = le32_to_cpu(in->mode) >> 12;
378 vino.ino = le64_to_cpu(in->ino);
379 vino.snap = le64_to_cpu(in->snapid);
380 ino = ceph_vino_to_ino(vino);
382 rinfo->dir_dname[off - fi->offset],
383 rinfo->dir_dname_len[off - fi->offset],
384 pos, ino, ftype) < 0) {
385 dout("filldir stopping us...\n");
389 filp->f_pos = pos + 1;
393 ceph_mdsc_put_request(fi->last_readdir);
394 fi->last_readdir = NULL;
399 if (!ceph_frag_is_rightmost(frag)) {
400 frag = ceph_frag_next(frag);
402 filp->f_pos = ceph_make_fpos(frag, off);
403 dout("readdir next frag is %x\n", frag);
409 * if dir_release_count still matches the dir, no dentries
410 * were released during the whole readdir, and we should have
411 * the complete dir contents in our cache.
413 spin_lock(&inode->i_lock);
414 if (ci->i_release_count == fi->dir_release_count) {
415 dout(" marking %p complete\n", inode);
416 ci->i_ceph_flags |= CEPH_I_COMPLETE;
417 ci->i_max_offset = filp->f_pos;
419 spin_unlock(&inode->i_lock);
421 dout("readdir %p filp %p done.\n", inode, filp);
425 static void reset_readdir(struct ceph_file_info *fi)
427 if (fi->last_readdir) {
428 ceph_mdsc_put_request(fi->last_readdir);
429 fi->last_readdir = NULL;
431 kfree(fi->last_name);
432 fi->last_name = NULL;
433 fi->next_offset = 2; /* compensate for . and .. */
441 static loff_t ceph_dir_llseek(struct file *file, loff_t offset, int origin)
443 struct ceph_file_info *fi = file->private_data;
444 struct inode *inode = file->f_mapping->host;
445 loff_t old_offset = offset;
448 mutex_lock(&inode->i_mutex);
451 offset += inode->i_size + 2; /* FIXME */
454 offset += file->f_pos;
457 if (offset >= 0 && offset <= inode->i_sb->s_maxbytes) {
458 if (offset != file->f_pos) {
459 file->f_pos = offset;
466 * discard buffered readdir content on seekdir(0), or
467 * seek to new frag, or seek prior to current chunk.
470 fpos_frag(offset) != fpos_frag(old_offset) ||
471 fpos_off(offset) < fi->offset) {
472 dout("dir_llseek dropping %p content\n", file);
476 /* bump dir_release_count if we did a forward seek */
477 if (offset > old_offset)
478 fi->dir_release_count--;
480 mutex_unlock(&inode->i_mutex);
485 * Process result of a lookup/open request.
487 * Mainly, make sure we return the final req->r_dentry (if it already
488 * existed) in place of the original VFS-provided dentry when they
491 * Gracefully handle the case where the MDS replies with -ENOENT and
492 * no trace (which it may do, at its discretion, e.g., if it doesn't
493 * care to issue a lease on the negative dentry).
495 struct dentry *ceph_finish_lookup(struct ceph_mds_request *req,
496 struct dentry *dentry, int err)
498 struct ceph_fs_client *fsc = ceph_sb_to_client(dentry->d_sb);
499 struct inode *parent = dentry->d_parent->d_inode;
502 if (err == -ENOENT &&
503 strcmp(dentry->d_name.name,
504 fsc->mount_options->snapdir_name) == 0) {
505 struct inode *inode = ceph_get_snapdir(parent);
506 dout("ENOENT on snapdir %p '%.*s', linking to snapdir %p\n",
507 dentry, dentry->d_name.len, dentry->d_name.name, inode);
508 BUG_ON(!d_unhashed(dentry));
509 d_add(dentry, inode);
513 if (err == -ENOENT) {
516 if (!req->r_reply_info.head->is_dentry) {
517 dout("ENOENT and no trace, dentry %p inode %p\n",
518 dentry, dentry->d_inode);
519 if (dentry->d_inode) {
528 dentry = ERR_PTR(err);
529 else if (dentry != req->r_dentry)
530 dentry = dget(req->r_dentry); /* we got spliced */
536 static int is_root_ceph_dentry(struct inode *inode, struct dentry *dentry)
538 return ceph_ino(inode) == CEPH_INO_ROOT &&
539 strncmp(dentry->d_name.name, ".ceph", 5) == 0;
543 * Look up a single dir entry. If there is a lookup intent, inform
544 * the MDS so that it gets our 'caps wanted' value in a single op.
546 static struct dentry *ceph_lookup(struct inode *dir, struct dentry *dentry,
547 struct nameidata *nd)
549 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
550 struct ceph_mds_client *mdsc = fsc->mdsc;
551 struct ceph_mds_request *req;
555 dout("lookup %p dentry %p '%.*s'\n",
556 dir, dentry, dentry->d_name.len, dentry->d_name.name);
558 if (dentry->d_name.len > NAME_MAX)
559 return ERR_PTR(-ENAMETOOLONG);
561 err = ceph_init_dentry(dentry);
565 /* open (but not create!) intent? */
567 (nd->flags & LOOKUP_OPEN) &&
568 (nd->flags & LOOKUP_CONTINUE) == 0 && /* only open last component */
569 !(nd->intent.open.flags & O_CREAT)) {
570 int mode = nd->intent.open.create_mode & ~current->fs->umask;
571 return ceph_lookup_open(dir, dentry, nd, mode, 1);
574 /* can we conclude ENOENT locally? */
575 if (dentry->d_inode == NULL) {
576 struct ceph_inode_info *ci = ceph_inode(dir);
577 struct ceph_dentry_info *di = ceph_dentry(dentry);
579 spin_lock(&dir->i_lock);
580 dout(" dir %p flags are %d\n", dir, ci->i_ceph_flags);
581 if (strncmp(dentry->d_name.name,
582 fsc->mount_options->snapdir_name,
583 dentry->d_name.len) &&
584 !is_root_ceph_dentry(dir, dentry) &&
585 (ci->i_ceph_flags & CEPH_I_COMPLETE) &&
586 (__ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1))) {
587 spin_unlock(&dir->i_lock);
588 dout(" dir %p complete, -ENOENT\n", dir);
590 di->lease_shared_gen = ci->i_shared_gen;
593 spin_unlock(&dir->i_lock);
596 op = ceph_snap(dir) == CEPH_SNAPDIR ?
597 CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP;
598 req = ceph_mdsc_create_request(mdsc, op, USE_ANY_MDS);
600 return ERR_CAST(req);
601 req->r_dentry = dget(dentry);
603 /* we only need inode linkage */
604 req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
605 req->r_locked_dir = dir;
606 err = ceph_mdsc_do_request(mdsc, NULL, req);
607 dentry = ceph_finish_lookup(req, dentry, err);
608 ceph_mdsc_put_request(req); /* will dput(dentry) */
609 dout("lookup result=%p\n", dentry);
614 * If we do a create but get no trace back from the MDS, follow up with
615 * a lookup (the VFS expects us to link up the provided dentry).
617 int ceph_handle_notrace_create(struct inode *dir, struct dentry *dentry)
619 struct dentry *result = ceph_lookup(dir, dentry, NULL);
621 if (result && !IS_ERR(result)) {
623 * We created the item, then did a lookup, and found
624 * it was already linked to another inode we already
625 * had in our cache (and thus got spliced). Link our
626 * dentry to that inode, but don't hash it, just in
627 * case the VFS wants to dereference it.
629 BUG_ON(!result->d_inode);
630 d_instantiate(dentry, result->d_inode);
633 return PTR_ERR(result);
636 static int ceph_mknod(struct inode *dir, struct dentry *dentry,
637 int mode, dev_t rdev)
639 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
640 struct ceph_mds_client *mdsc = fsc->mdsc;
641 struct ceph_mds_request *req;
644 if (ceph_snap(dir) != CEPH_NOSNAP)
647 dout("mknod in dir %p dentry %p mode 0%o rdev %d\n",
648 dir, dentry, mode, rdev);
649 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_MKNOD, USE_AUTH_MDS);
654 req->r_dentry = dget(dentry);
656 req->r_locked_dir = dir;
657 req->r_args.mknod.mode = cpu_to_le32(mode);
658 req->r_args.mknod.rdev = cpu_to_le32(rdev);
659 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
660 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
661 err = ceph_mdsc_do_request(mdsc, dir, req);
662 if (!err && !req->r_reply_info.head->is_dentry)
663 err = ceph_handle_notrace_create(dir, dentry);
664 ceph_mdsc_put_request(req);
670 static int ceph_create(struct inode *dir, struct dentry *dentry, int mode,
671 struct nameidata *nd)
673 dout("create in dir %p dentry %p name '%.*s'\n",
674 dir, dentry, dentry->d_name.len, dentry->d_name.name);
676 if (ceph_snap(dir) != CEPH_NOSNAP)
680 BUG_ON((nd->flags & LOOKUP_OPEN) == 0);
681 dentry = ceph_lookup_open(dir, dentry, nd, mode, 0);
682 /* hrm, what should i do here if we get aliased? */
684 return PTR_ERR(dentry);
688 /* fall back to mknod */
689 return ceph_mknod(dir, dentry, (mode & ~S_IFMT) | S_IFREG, 0);
692 static int ceph_symlink(struct inode *dir, struct dentry *dentry,
695 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
696 struct ceph_mds_client *mdsc = fsc->mdsc;
697 struct ceph_mds_request *req;
700 if (ceph_snap(dir) != CEPH_NOSNAP)
703 dout("symlink in dir %p dentry %p to '%s'\n", dir, dentry, dest);
704 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SYMLINK, USE_AUTH_MDS);
709 req->r_dentry = dget(dentry);
711 req->r_path2 = kstrdup(dest, GFP_NOFS);
712 req->r_locked_dir = dir;
713 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
714 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
715 err = ceph_mdsc_do_request(mdsc, dir, req);
716 if (!err && !req->r_reply_info.head->is_dentry)
717 err = ceph_handle_notrace_create(dir, dentry);
718 ceph_mdsc_put_request(req);
724 static int ceph_mkdir(struct inode *dir, struct dentry *dentry, int mode)
726 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
727 struct ceph_mds_client *mdsc = fsc->mdsc;
728 struct ceph_mds_request *req;
732 if (ceph_snap(dir) == CEPH_SNAPDIR) {
733 /* mkdir .snap/foo is a MKSNAP */
734 op = CEPH_MDS_OP_MKSNAP;
735 dout("mksnap dir %p snap '%.*s' dn %p\n", dir,
736 dentry->d_name.len, dentry->d_name.name, dentry);
737 } else if (ceph_snap(dir) == CEPH_NOSNAP) {
738 dout("mkdir dir %p dn %p mode 0%o\n", dir, dentry, mode);
739 op = CEPH_MDS_OP_MKDIR;
743 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
749 req->r_dentry = dget(dentry);
751 req->r_locked_dir = dir;
752 req->r_args.mkdir.mode = cpu_to_le32(mode);
753 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
754 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
755 err = ceph_mdsc_do_request(mdsc, dir, req);
756 if (!err && !req->r_reply_info.head->is_dentry)
757 err = ceph_handle_notrace_create(dir, dentry);
758 ceph_mdsc_put_request(req);
765 static int ceph_link(struct dentry *old_dentry, struct inode *dir,
766 struct dentry *dentry)
768 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
769 struct ceph_mds_client *mdsc = fsc->mdsc;
770 struct ceph_mds_request *req;
773 if (ceph_snap(dir) != CEPH_NOSNAP)
776 dout("link in dir %p old_dentry %p dentry %p\n", dir,
778 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LINK, USE_AUTH_MDS);
783 req->r_dentry = dget(dentry);
785 req->r_old_dentry = dget(old_dentry); /* or inode? hrm. */
786 req->r_locked_dir = dir;
787 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
788 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
789 err = ceph_mdsc_do_request(mdsc, dir, req);
792 else if (!req->r_reply_info.head->is_dentry)
793 d_instantiate(dentry, igrab(old_dentry->d_inode));
794 ceph_mdsc_put_request(req);
799 * For a soon-to-be unlinked file, drop the AUTH_RDCACHE caps. If it
800 * looks like the link count will hit 0, drop any other caps (other
801 * than PIN) we don't specifically want (due to the file still being
804 static int drop_caps_for_unlink(struct inode *inode)
806 struct ceph_inode_info *ci = ceph_inode(inode);
807 int drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL;
809 spin_lock(&inode->i_lock);
810 if (inode->i_nlink == 1) {
811 drop |= ~(__ceph_caps_wanted(ci) | CEPH_CAP_PIN);
812 ci->i_ceph_flags |= CEPH_I_NODELAY;
814 spin_unlock(&inode->i_lock);
819 * rmdir and unlink are differ only by the metadata op code
821 static int ceph_unlink(struct inode *dir, struct dentry *dentry)
823 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
824 struct ceph_mds_client *mdsc = fsc->mdsc;
825 struct inode *inode = dentry->d_inode;
826 struct ceph_mds_request *req;
830 if (ceph_snap(dir) == CEPH_SNAPDIR) {
831 /* rmdir .snap/foo is RMSNAP */
832 dout("rmsnap dir %p '%.*s' dn %p\n", dir, dentry->d_name.len,
833 dentry->d_name.name, dentry);
834 op = CEPH_MDS_OP_RMSNAP;
835 } else if (ceph_snap(dir) == CEPH_NOSNAP) {
836 dout("unlink/rmdir dir %p dn %p inode %p\n",
838 op = ((dentry->d_inode->i_mode & S_IFMT) == S_IFDIR) ?
839 CEPH_MDS_OP_RMDIR : CEPH_MDS_OP_UNLINK;
842 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
847 req->r_dentry = dget(dentry);
849 req->r_locked_dir = dir;
850 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
851 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
852 req->r_inode_drop = drop_caps_for_unlink(inode);
853 err = ceph_mdsc_do_request(mdsc, dir, req);
854 if (!err && !req->r_reply_info.head->is_dentry)
856 ceph_mdsc_put_request(req);
861 static int ceph_rename(struct inode *old_dir, struct dentry *old_dentry,
862 struct inode *new_dir, struct dentry *new_dentry)
864 struct ceph_fs_client *fsc = ceph_sb_to_client(old_dir->i_sb);
865 struct ceph_mds_client *mdsc = fsc->mdsc;
866 struct ceph_mds_request *req;
869 if (ceph_snap(old_dir) != ceph_snap(new_dir))
871 if (ceph_snap(old_dir) != CEPH_NOSNAP ||
872 ceph_snap(new_dir) != CEPH_NOSNAP)
874 dout("rename dir %p dentry %p to dir %p dentry %p\n",
875 old_dir, old_dentry, new_dir, new_dentry);
876 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_RENAME, USE_AUTH_MDS);
879 req->r_dentry = dget(new_dentry);
881 req->r_old_dentry = dget(old_dentry);
882 req->r_locked_dir = new_dir;
883 req->r_old_dentry_drop = CEPH_CAP_FILE_SHARED;
884 req->r_old_dentry_unless = CEPH_CAP_FILE_EXCL;
885 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
886 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
887 /* release LINK_RDCACHE on source inode (mds will lock it) */
888 req->r_old_inode_drop = CEPH_CAP_LINK_SHARED;
889 if (new_dentry->d_inode)
890 req->r_inode_drop = drop_caps_for_unlink(new_dentry->d_inode);
891 err = ceph_mdsc_do_request(mdsc, old_dir, req);
892 if (!err && !req->r_reply_info.head->is_dentry) {
894 * Normally d_move() is done by fill_trace (called by
895 * do_request, above). If there is no trace, we need
899 /* d_move screws up d_subdirs order */
900 ceph_i_clear(new_dir, CEPH_I_COMPLETE);
902 d_move(old_dentry, new_dentry);
904 /* ensure target dentry is invalidated, despite
905 rehashing bug in vfs_rename_dir */
906 ceph_invalidate_dentry_lease(new_dentry);
908 ceph_mdsc_put_request(req);
913 * Ensure a dentry lease will no longer revalidate.
915 void ceph_invalidate_dentry_lease(struct dentry *dentry)
917 spin_lock(&dentry->d_lock);
918 dentry->d_time = jiffies;
919 ceph_dentry(dentry)->lease_shared_gen = 0;
920 spin_unlock(&dentry->d_lock);
924 * Check if dentry lease is valid. If not, delete the lease. Try to
925 * renew if the least is more than half up.
927 static int dentry_lease_is_valid(struct dentry *dentry)
929 struct ceph_dentry_info *di;
930 struct ceph_mds_session *s;
934 struct ceph_mds_session *session = NULL;
935 struct inode *dir = NULL;
938 spin_lock(&dentry->d_lock);
939 di = ceph_dentry(dentry);
940 if (di && di->lease_session) {
941 s = di->lease_session;
942 spin_lock(&s->s_cap_lock);
945 spin_unlock(&s->s_cap_lock);
947 if (di->lease_gen == gen &&
948 time_before(jiffies, dentry->d_time) &&
949 time_before(jiffies, ttl)) {
951 if (di->lease_renew_after &&
952 time_after(jiffies, di->lease_renew_after)) {
953 /* we should renew */
954 dir = dentry->d_parent->d_inode;
955 session = ceph_get_mds_session(s);
957 di->lease_renew_after = 0;
958 di->lease_renew_from = jiffies;
962 spin_unlock(&dentry->d_lock);
965 ceph_mdsc_lease_send_msg(session, dir, dentry,
966 CEPH_MDS_LEASE_RENEW, seq);
967 ceph_put_mds_session(session);
969 dout("dentry_lease_is_valid - dentry %p = %d\n", dentry, valid);
974 * Check if directory-wide content lease/cap is valid.
976 static int dir_lease_is_valid(struct inode *dir, struct dentry *dentry)
978 struct ceph_inode_info *ci = ceph_inode(dir);
979 struct ceph_dentry_info *di = ceph_dentry(dentry);
982 spin_lock(&dir->i_lock);
983 if (ci->i_shared_gen == di->lease_shared_gen)
984 valid = __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1);
985 spin_unlock(&dir->i_lock);
986 dout("dir_lease_is_valid dir %p v%u dentry %p v%u = %d\n",
987 dir, (unsigned)ci->i_shared_gen, dentry,
988 (unsigned)di->lease_shared_gen, valid);
993 * Check if cached dentry can be trusted.
995 static int ceph_d_revalidate(struct dentry *dentry, struct nameidata *nd)
997 struct inode *dir = dentry->d_parent->d_inode;
999 dout("d_revalidate %p '%.*s' inode %p offset %lld\n", dentry,
1000 dentry->d_name.len, dentry->d_name.name, dentry->d_inode,
1001 ceph_dentry(dentry)->offset);
1003 /* always trust cached snapped dentries, snapdir dentry */
1004 if (ceph_snap(dir) != CEPH_NOSNAP) {
1005 dout("d_revalidate %p '%.*s' inode %p is SNAPPED\n", dentry,
1006 dentry->d_name.len, dentry->d_name.name, dentry->d_inode);
1009 if (dentry->d_inode && ceph_snap(dentry->d_inode) == CEPH_SNAPDIR)
1012 if (dentry_lease_is_valid(dentry) ||
1013 dir_lease_is_valid(dir, dentry))
1016 dout("d_revalidate %p invalid\n", dentry);
1020 ceph_dentry_lru_touch(dentry);
1025 * When a dentry is released, clear the dir I_COMPLETE if it was part
1026 * of the current dir gen or if this is in the snapshot namespace.
1028 static void ceph_dentry_release(struct dentry *dentry)
1030 struct ceph_dentry_info *di = ceph_dentry(dentry);
1031 struct inode *parent_inode = NULL;
1032 u64 snapid = CEPH_NOSNAP;
1034 if (!IS_ROOT(dentry)) {
1035 parent_inode = dentry->d_parent->d_inode;
1037 snapid = ceph_snap(parent_inode);
1039 dout("dentry_release %p parent %p\n", dentry, parent_inode);
1040 if (parent_inode && snapid != CEPH_SNAPDIR) {
1041 struct ceph_inode_info *ci = ceph_inode(parent_inode);
1043 spin_lock(&parent_inode->i_lock);
1044 if (ci->i_shared_gen == di->lease_shared_gen ||
1045 snapid <= CEPH_MAXSNAP) {
1046 dout(" clearing %p complete (d_release)\n",
1048 ci->i_ceph_flags &= ~CEPH_I_COMPLETE;
1049 ci->i_release_count++;
1051 spin_unlock(&parent_inode->i_lock);
1054 ceph_dentry_lru_del(dentry);
1055 if (di->lease_session)
1056 ceph_put_mds_session(di->lease_session);
1057 kmem_cache_free(ceph_dentry_cachep, di);
1058 dentry->d_fsdata = NULL;
1062 static int ceph_snapdir_d_revalidate(struct dentry *dentry,
1063 struct nameidata *nd)
1066 * Eventually, we'll want to revalidate snapped metadata
1067 * too... probably...
1075 * read() on a dir. This weird interface hack only works if mounted
1076 * with '-o dirstat'.
1078 static ssize_t ceph_read_dir(struct file *file, char __user *buf, size_t size,
1081 struct ceph_file_info *cf = file->private_data;
1082 struct inode *inode = file->f_dentry->d_inode;
1083 struct ceph_inode_info *ci = ceph_inode(inode);
1086 if (!ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb), DIRSTAT))
1089 if (!cf->dir_info) {
1090 cf->dir_info = kmalloc(1024, GFP_NOFS);
1094 sprintf(cf->dir_info,
1097 " subdirs: %20lld\n"
1098 "rentries: %20lld\n"
1100 " rsubdirs: %20lld\n"
1102 "rctime: %10ld.%09ld\n",
1103 ci->i_files + ci->i_subdirs,
1106 ci->i_rfiles + ci->i_rsubdirs,
1110 (long)ci->i_rctime.tv_sec,
1111 (long)ci->i_rctime.tv_nsec);
1114 if (*ppos >= cf->dir_info_len)
1116 size = min_t(unsigned, size, cf->dir_info_len-*ppos);
1117 left = copy_to_user(buf, cf->dir_info + *ppos, size);
1120 *ppos += (size - left);
1125 * an fsync() on a dir will wait for any uncommitted directory
1126 * operations to commit.
1128 static int ceph_dir_fsync(struct file *file, int datasync)
1130 struct inode *inode = file->f_path.dentry->d_inode;
1131 struct ceph_inode_info *ci = ceph_inode(inode);
1132 struct list_head *head = &ci->i_unsafe_dirops;
1133 struct ceph_mds_request *req;
1137 dout("dir_fsync %p\n", inode);
1138 spin_lock(&ci->i_unsafe_lock);
1139 if (list_empty(head))
1142 req = list_entry(head->prev,
1143 struct ceph_mds_request, r_unsafe_dir_item);
1144 last_tid = req->r_tid;
1147 ceph_mdsc_get_request(req);
1148 spin_unlock(&ci->i_unsafe_lock);
1149 dout("dir_fsync %p wait on tid %llu (until %llu)\n",
1150 inode, req->r_tid, last_tid);
1151 if (req->r_timeout) {
1152 ret = wait_for_completion_timeout(
1153 &req->r_safe_completion, req->r_timeout);
1157 ret = -EIO; /* timed out */
1159 wait_for_completion(&req->r_safe_completion);
1161 spin_lock(&ci->i_unsafe_lock);
1162 ceph_mdsc_put_request(req);
1164 if (ret || list_empty(head))
1166 req = list_entry(head->next,
1167 struct ceph_mds_request, r_unsafe_dir_item);
1168 } while (req->r_tid < last_tid);
1170 spin_unlock(&ci->i_unsafe_lock);
1175 * We maintain a private dentry LRU.
1177 * FIXME: this needs to be changed to a per-mds lru to be useful.
1179 void ceph_dentry_lru_add(struct dentry *dn)
1181 struct ceph_dentry_info *di = ceph_dentry(dn);
1182 struct ceph_mds_client *mdsc;
1184 dout("dentry_lru_add %p %p '%.*s'\n", di, dn,
1185 dn->d_name.len, dn->d_name.name);
1187 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1188 spin_lock(&mdsc->dentry_lru_lock);
1189 list_add_tail(&di->lru, &mdsc->dentry_lru);
1191 spin_unlock(&mdsc->dentry_lru_lock);
1195 void ceph_dentry_lru_touch(struct dentry *dn)
1197 struct ceph_dentry_info *di = ceph_dentry(dn);
1198 struct ceph_mds_client *mdsc;
1200 dout("dentry_lru_touch %p %p '%.*s' (offset %lld)\n", di, dn,
1201 dn->d_name.len, dn->d_name.name, di->offset);
1203 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1204 spin_lock(&mdsc->dentry_lru_lock);
1205 list_move_tail(&di->lru, &mdsc->dentry_lru);
1206 spin_unlock(&mdsc->dentry_lru_lock);
1210 void ceph_dentry_lru_del(struct dentry *dn)
1212 struct ceph_dentry_info *di = ceph_dentry(dn);
1213 struct ceph_mds_client *mdsc;
1215 dout("dentry_lru_del %p %p '%.*s'\n", di, dn,
1216 dn->d_name.len, dn->d_name.name);
1218 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1219 spin_lock(&mdsc->dentry_lru_lock);
1220 list_del_init(&di->lru);
1222 spin_unlock(&mdsc->dentry_lru_lock);
1226 const struct file_operations ceph_dir_fops = {
1227 .read = ceph_read_dir,
1228 .readdir = ceph_readdir,
1229 .llseek = ceph_dir_llseek,
1231 .release = ceph_release,
1232 .unlocked_ioctl = ceph_ioctl,
1233 .fsync = ceph_dir_fsync,
1236 const struct inode_operations ceph_dir_iops = {
1237 .lookup = ceph_lookup,
1238 .permission = ceph_permission,
1239 .getattr = ceph_getattr,
1240 .setattr = ceph_setattr,
1241 .setxattr = ceph_setxattr,
1242 .getxattr = ceph_getxattr,
1243 .listxattr = ceph_listxattr,
1244 .removexattr = ceph_removexattr,
1245 .mknod = ceph_mknod,
1246 .symlink = ceph_symlink,
1247 .mkdir = ceph_mkdir,
1249 .unlink = ceph_unlink,
1250 .rmdir = ceph_unlink,
1251 .rename = ceph_rename,
1252 .create = ceph_create,
1255 const struct dentry_operations ceph_dentry_ops = {
1256 .d_revalidate = ceph_d_revalidate,
1257 .d_release = ceph_dentry_release,
1260 const struct dentry_operations ceph_snapdir_dentry_ops = {
1261 .d_revalidate = ceph_snapdir_d_revalidate,
1262 .d_release = ceph_dentry_release,
1265 const struct dentry_operations ceph_snap_dentry_ops = {
1266 .d_release = ceph_dentry_release,