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 d_set_d_op(dentry, &ceph_dentry_ops);
46 else if (ceph_snap(dentry->d_parent->d_inode) == CEPH_SNAPDIR)
47 d_set_d_op(dentry, &ceph_snapdir_dentry_ops);
49 d_set_d_op(dentry, &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(&parent->d_lock);
117 /* start at beginning? */
118 if (filp->f_pos == 2 || last == NULL ||
119 filp->f_pos < ceph_dentry(last)->offset) {
120 if (list_empty(&parent->d_subdirs))
122 p = parent->d_subdirs.prev;
123 dout(" initial p %p/%p\n", p->prev, p->next);
125 p = last->d_u.d_child.prev;
129 dentry = list_entry(p, struct dentry, d_u.d_child);
130 di = ceph_dentry(dentry);
132 dout(" p %p/%p %s d_subdirs %p/%p\n", p->prev, p->next,
133 d_unhashed(dentry) ? "!hashed" : "hashed",
134 parent->d_subdirs.prev, parent->d_subdirs.next);
135 if (p == &parent->d_subdirs) {
139 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
140 if (!d_unhashed(dentry) && dentry->d_inode &&
141 ceph_snap(dentry->d_inode) != CEPH_SNAPDIR &&
142 ceph_ino(dentry->d_inode) != CEPH_INO_CEPH &&
143 filp->f_pos <= di->offset)
145 dout(" skipping %p %.*s at %llu (%llu)%s%s\n", dentry,
146 dentry->d_name.len, dentry->d_name.name, di->offset,
147 filp->f_pos, d_unhashed(dentry) ? " unhashed" : "",
148 !dentry->d_inode ? " null" : "");
149 spin_unlock(&dentry->d_lock);
151 dentry = list_entry(p, struct dentry, d_u.d_child);
152 di = ceph_dentry(dentry);
156 spin_unlock(&dentry->d_lock);
157 spin_unlock(&parent->d_lock);
159 dout(" %llu (%llu) dentry %p %.*s %p\n", di->offset, filp->f_pos,
160 dentry, dentry->d_name.len, dentry->d_name.name, dentry->d_inode);
161 filp->f_pos = di->offset;
162 err = filldir(dirent, dentry->d_name.name,
163 dentry->d_name.len, di->offset,
164 ceph_translate_ino(dentry->d_sb, dentry->d_inode->i_ino),
165 dentry->d_inode->i_mode >> 12);
169 /* remember our position */
171 fi->next_offset = di->offset;
183 /* make sure a dentry wasn't dropped while we didn't have parent lock */
184 if (!ceph_i_test(dir, CEPH_I_COMPLETE)) {
185 dout(" lost I_COMPLETE on %p; falling back to mds\n", dir);
190 spin_lock(&parent->d_lock);
191 p = p->prev; /* advance to next dentry */
195 spin_unlock(&parent->d_lock);
203 * make note of the last dentry we read, so we can
204 * continue at the same lexicographical point,
205 * regardless of what dir changes take place on the
208 static int note_last_dentry(struct ceph_file_info *fi, const char *name,
211 kfree(fi->last_name);
212 fi->last_name = kmalloc(len+1, GFP_NOFS);
215 memcpy(fi->last_name, name, len);
216 fi->last_name[len] = 0;
217 dout("note_last_dentry '%s'\n", fi->last_name);
221 static int ceph_readdir(struct file *filp, void *dirent, filldir_t filldir)
223 struct ceph_file_info *fi = filp->private_data;
224 struct inode *inode = filp->f_dentry->d_inode;
225 struct ceph_inode_info *ci = ceph_inode(inode);
226 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
227 struct ceph_mds_client *mdsc = fsc->mdsc;
228 unsigned frag = fpos_frag(filp->f_pos);
229 int off = fpos_off(filp->f_pos);
232 struct ceph_mds_reply_info_parsed *rinfo;
233 const int max_entries = fsc->mount_options->max_readdir;
234 const int max_bytes = fsc->mount_options->max_readdir_bytes;
236 dout("readdir %p filp %p frag %u off %u\n", inode, filp, frag, off);
240 /* always start with . and .. */
241 if (filp->f_pos == 0) {
242 /* note dir version at start of readdir so we can tell
243 * if any dentries get dropped */
244 fi->dir_release_count = ci->i_release_count;
246 dout("readdir off 0 -> '.'\n");
247 if (filldir(dirent, ".", 1, ceph_make_fpos(0, 0),
248 ceph_translate_ino(inode->i_sb, inode->i_ino),
249 inode->i_mode >> 12) < 0)
254 if (filp->f_pos == 1) {
255 ino_t ino = parent_ino(filp->f_dentry);
256 dout("readdir off 1 -> '..'\n");
257 if (filldir(dirent, "..", 2, ceph_make_fpos(0, 1),
258 ceph_translate_ino(inode->i_sb, ino),
259 inode->i_mode >> 12) < 0)
265 /* can we use the dcache? */
266 spin_lock(&inode->i_lock);
267 if ((filp->f_pos == 2 || fi->dentry) &&
268 !ceph_test_mount_opt(fsc, NOASYNCREADDIR) &&
269 ceph_snap(inode) != CEPH_SNAPDIR &&
270 (ci->i_ceph_flags & CEPH_I_COMPLETE) &&
271 __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1)) {
272 spin_unlock(&inode->i_lock);
273 err = __dcache_readdir(filp, dirent, filldir);
277 spin_unlock(&inode->i_lock);
280 err = note_last_dentry(fi, fi->dentry->d_name.name,
281 fi->dentry->d_name.len);
288 /* proceed with a normal readdir */
291 /* do we have the correct frag content buffered? */
292 if (fi->frag != frag || fi->last_readdir == NULL) {
293 struct ceph_mds_request *req;
294 int op = ceph_snap(inode) == CEPH_SNAPDIR ?
295 CEPH_MDS_OP_LSSNAP : CEPH_MDS_OP_READDIR;
297 /* discard old result, if any */
298 if (fi->last_readdir) {
299 ceph_mdsc_put_request(fi->last_readdir);
300 fi->last_readdir = NULL;
303 /* requery frag tree, as the frag topology may have changed */
304 frag = ceph_choose_frag(ceph_inode(inode), frag, NULL, NULL);
306 dout("readdir fetching %llx.%llx frag %x offset '%s'\n",
307 ceph_vinop(inode), frag, fi->last_name);
308 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
311 req->r_inode = inode;
313 req->r_dentry = dget(filp->f_dentry);
314 /* hints to request -> mds selection code */
315 req->r_direct_mode = USE_AUTH_MDS;
316 req->r_direct_hash = ceph_frag_value(frag);
317 req->r_direct_is_hash = true;
318 req->r_path2 = kstrdup(fi->last_name, GFP_NOFS);
319 req->r_readdir_offset = fi->next_offset;
320 req->r_args.readdir.frag = cpu_to_le32(frag);
321 req->r_args.readdir.max_entries = cpu_to_le32(max_entries);
322 req->r_args.readdir.max_bytes = cpu_to_le32(max_bytes);
323 req->r_num_caps = max_entries + 1;
324 err = ceph_mdsc_do_request(mdsc, NULL, req);
326 ceph_mdsc_put_request(req);
329 dout("readdir got and parsed readdir result=%d"
330 " on frag %x, end=%d, complete=%d\n", err, frag,
331 (int)req->r_reply_info.dir_end,
332 (int)req->r_reply_info.dir_complete);
334 if (!req->r_did_prepopulate) {
335 dout("readdir !did_prepopulate");
336 fi->dir_release_count--; /* preclude I_COMPLETE */
339 /* note next offset and last dentry name */
340 fi->offset = fi->next_offset;
341 fi->last_readdir = req;
343 if (req->r_reply_info.dir_end) {
344 kfree(fi->last_name);
345 fi->last_name = NULL;
346 if (ceph_frag_is_rightmost(frag))
351 rinfo = &req->r_reply_info;
352 err = note_last_dentry(fi,
353 rinfo->dir_dname[rinfo->dir_nr-1],
354 rinfo->dir_dname_len[rinfo->dir_nr-1]);
357 fi->next_offset += rinfo->dir_nr;
361 rinfo = &fi->last_readdir->r_reply_info;
362 dout("readdir frag %x num %d off %d chunkoff %d\n", frag,
363 rinfo->dir_nr, off, fi->offset);
364 while (off >= fi->offset && off - fi->offset < rinfo->dir_nr) {
365 u64 pos = ceph_make_fpos(frag, off);
366 struct ceph_mds_reply_inode *in =
367 rinfo->dir_in[off - fi->offset].in;
368 struct ceph_vino vino;
371 dout("readdir off %d (%d/%d) -> %lld '%.*s' %p\n",
372 off, off - fi->offset, rinfo->dir_nr, pos,
373 rinfo->dir_dname_len[off - fi->offset],
374 rinfo->dir_dname[off - fi->offset], in);
376 ftype = le32_to_cpu(in->mode) >> 12;
377 vino.ino = le64_to_cpu(in->ino);
378 vino.snap = le64_to_cpu(in->snapid);
379 ino = ceph_vino_to_ino(vino);
381 rinfo->dir_dname[off - fi->offset],
382 rinfo->dir_dname_len[off - fi->offset],
384 ceph_translate_ino(inode->i_sb, 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 ceph_snap(parent) == CEPH_NOSNAP &&
504 strcmp(dentry->d_name.name,
505 fsc->mount_options->snapdir_name) == 0) {
506 struct inode *inode = ceph_get_snapdir(parent);
507 dout("ENOENT on snapdir %p '%.*s', linking to snapdir %p\n",
508 dentry, dentry->d_name.len, dentry->d_name.name, inode);
509 BUG_ON(!d_unhashed(dentry));
510 d_add(dentry, inode);
514 if (err == -ENOENT) {
517 if (!req->r_reply_info.head->is_dentry) {
518 dout("ENOENT and no trace, dentry %p inode %p\n",
519 dentry, dentry->d_inode);
520 if (dentry->d_inode) {
529 dentry = ERR_PTR(err);
530 else if (dentry != req->r_dentry)
531 dentry = dget(req->r_dentry); /* we got spliced */
537 static int is_root_ceph_dentry(struct inode *inode, struct dentry *dentry)
539 return ceph_ino(inode) == CEPH_INO_ROOT &&
540 strncmp(dentry->d_name.name, ".ceph", 5) == 0;
544 * Look up a single dir entry. If there is a lookup intent, inform
545 * the MDS so that it gets our 'caps wanted' value in a single op.
547 static struct dentry *ceph_lookup(struct inode *dir, struct dentry *dentry,
548 struct nameidata *nd)
550 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
551 struct ceph_mds_client *mdsc = fsc->mdsc;
552 struct ceph_mds_request *req;
556 dout("lookup %p dentry %p '%.*s'\n",
557 dir, dentry, dentry->d_name.len, dentry->d_name.name);
559 if (dentry->d_name.len > NAME_MAX)
560 return ERR_PTR(-ENAMETOOLONG);
562 err = ceph_init_dentry(dentry);
566 /* open (but not create!) intent? */
568 (nd->flags & LOOKUP_OPEN) &&
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 ihold(old_dentry->d_inode);
794 d_instantiate(dentry, old_dentry->d_inode);
796 ceph_mdsc_put_request(req);
801 * For a soon-to-be unlinked file, drop the AUTH_RDCACHE caps. If it
802 * looks like the link count will hit 0, drop any other caps (other
803 * than PIN) we don't specifically want (due to the file still being
806 static int drop_caps_for_unlink(struct inode *inode)
808 struct ceph_inode_info *ci = ceph_inode(inode);
809 int drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL;
811 spin_lock(&inode->i_lock);
812 if (inode->i_nlink == 1) {
813 drop |= ~(__ceph_caps_wanted(ci) | CEPH_CAP_PIN);
814 ci->i_ceph_flags |= CEPH_I_NODELAY;
816 spin_unlock(&inode->i_lock);
821 * rmdir and unlink are differ only by the metadata op code
823 static int ceph_unlink(struct inode *dir, struct dentry *dentry)
825 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
826 struct ceph_mds_client *mdsc = fsc->mdsc;
827 struct inode *inode = dentry->d_inode;
828 struct ceph_mds_request *req;
832 if (ceph_snap(dir) == CEPH_SNAPDIR) {
833 /* rmdir .snap/foo is RMSNAP */
834 dout("rmsnap dir %p '%.*s' dn %p\n", dir, dentry->d_name.len,
835 dentry->d_name.name, dentry);
836 op = CEPH_MDS_OP_RMSNAP;
837 } else if (ceph_snap(dir) == CEPH_NOSNAP) {
838 dout("unlink/rmdir dir %p dn %p inode %p\n",
840 op = ((dentry->d_inode->i_mode & S_IFMT) == S_IFDIR) ?
841 CEPH_MDS_OP_RMDIR : CEPH_MDS_OP_UNLINK;
844 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
849 req->r_dentry = dget(dentry);
851 req->r_locked_dir = dir;
852 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
853 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
854 req->r_inode_drop = drop_caps_for_unlink(inode);
855 err = ceph_mdsc_do_request(mdsc, dir, req);
856 if (!err && !req->r_reply_info.head->is_dentry)
858 ceph_mdsc_put_request(req);
863 static int ceph_rename(struct inode *old_dir, struct dentry *old_dentry,
864 struct inode *new_dir, struct dentry *new_dentry)
866 struct ceph_fs_client *fsc = ceph_sb_to_client(old_dir->i_sb);
867 struct ceph_mds_client *mdsc = fsc->mdsc;
868 struct ceph_mds_request *req;
871 if (ceph_snap(old_dir) != ceph_snap(new_dir))
873 if (ceph_snap(old_dir) != CEPH_NOSNAP ||
874 ceph_snap(new_dir) != CEPH_NOSNAP)
876 dout("rename dir %p dentry %p to dir %p dentry %p\n",
877 old_dir, old_dentry, new_dir, new_dentry);
878 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_RENAME, USE_AUTH_MDS);
881 req->r_dentry = dget(new_dentry);
883 req->r_old_dentry = dget(old_dentry);
884 req->r_locked_dir = new_dir;
885 req->r_old_dentry_drop = CEPH_CAP_FILE_SHARED;
886 req->r_old_dentry_unless = CEPH_CAP_FILE_EXCL;
887 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
888 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
889 /* release LINK_RDCACHE on source inode (mds will lock it) */
890 req->r_old_inode_drop = CEPH_CAP_LINK_SHARED;
891 if (new_dentry->d_inode)
892 req->r_inode_drop = drop_caps_for_unlink(new_dentry->d_inode);
893 err = ceph_mdsc_do_request(mdsc, old_dir, req);
894 if (!err && !req->r_reply_info.head->is_dentry) {
896 * Normally d_move() is done by fill_trace (called by
897 * do_request, above). If there is no trace, we need
901 /* d_move screws up d_subdirs order */
902 ceph_i_clear(new_dir, CEPH_I_COMPLETE);
904 d_move(old_dentry, new_dentry);
906 /* ensure target dentry is invalidated, despite
907 rehashing bug in vfs_rename_dir */
908 ceph_invalidate_dentry_lease(new_dentry);
910 ceph_mdsc_put_request(req);
915 * Ensure a dentry lease will no longer revalidate.
917 void ceph_invalidate_dentry_lease(struct dentry *dentry)
919 spin_lock(&dentry->d_lock);
920 dentry->d_time = jiffies;
921 ceph_dentry(dentry)->lease_shared_gen = 0;
922 spin_unlock(&dentry->d_lock);
926 * Check if dentry lease is valid. If not, delete the lease. Try to
927 * renew if the least is more than half up.
929 static int dentry_lease_is_valid(struct dentry *dentry)
931 struct ceph_dentry_info *di;
932 struct ceph_mds_session *s;
936 struct ceph_mds_session *session = NULL;
937 struct inode *dir = NULL;
940 spin_lock(&dentry->d_lock);
941 di = ceph_dentry(dentry);
942 if (di && di->lease_session) {
943 s = di->lease_session;
944 spin_lock(&s->s_cap_lock);
947 spin_unlock(&s->s_cap_lock);
949 if (di->lease_gen == gen &&
950 time_before(jiffies, dentry->d_time) &&
951 time_before(jiffies, ttl)) {
953 if (di->lease_renew_after &&
954 time_after(jiffies, di->lease_renew_after)) {
955 /* we should renew */
956 dir = dentry->d_parent->d_inode;
957 session = ceph_get_mds_session(s);
959 di->lease_renew_after = 0;
960 di->lease_renew_from = jiffies;
964 spin_unlock(&dentry->d_lock);
967 ceph_mdsc_lease_send_msg(session, dir, dentry,
968 CEPH_MDS_LEASE_RENEW, seq);
969 ceph_put_mds_session(session);
971 dout("dentry_lease_is_valid - dentry %p = %d\n", dentry, valid);
976 * Check if directory-wide content lease/cap is valid.
978 static int dir_lease_is_valid(struct inode *dir, struct dentry *dentry)
980 struct ceph_inode_info *ci = ceph_inode(dir);
981 struct ceph_dentry_info *di = ceph_dentry(dentry);
984 spin_lock(&dir->i_lock);
985 if (ci->i_shared_gen == di->lease_shared_gen)
986 valid = __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1);
987 spin_unlock(&dir->i_lock);
988 dout("dir_lease_is_valid dir %p v%u dentry %p v%u = %d\n",
989 dir, (unsigned)ci->i_shared_gen, dentry,
990 (unsigned)di->lease_shared_gen, valid);
995 * Check if cached dentry can be trusted.
997 static int ceph_d_revalidate(struct dentry *dentry, struct nameidata *nd)
1001 if (nd && nd->flags & LOOKUP_RCU)
1004 dir = dentry->d_parent->d_inode;
1006 dout("d_revalidate %p '%.*s' inode %p offset %lld\n", dentry,
1007 dentry->d_name.len, dentry->d_name.name, dentry->d_inode,
1008 ceph_dentry(dentry)->offset);
1010 /* always trust cached snapped dentries, snapdir dentry */
1011 if (ceph_snap(dir) != CEPH_NOSNAP) {
1012 dout("d_revalidate %p '%.*s' inode %p is SNAPPED\n", dentry,
1013 dentry->d_name.len, dentry->d_name.name, dentry->d_inode);
1016 if (dentry->d_inode && ceph_snap(dentry->d_inode) == CEPH_SNAPDIR)
1019 if (dentry_lease_is_valid(dentry) ||
1020 dir_lease_is_valid(dir, dentry))
1023 dout("d_revalidate %p invalid\n", dentry);
1027 ceph_dentry_lru_touch(dentry);
1032 * Release our ceph_dentry_info.
1034 static void ceph_d_release(struct dentry *dentry)
1036 struct ceph_dentry_info *di = ceph_dentry(dentry);
1038 dout("d_release %p\n", dentry);
1040 ceph_dentry_lru_del(dentry);
1041 if (di->lease_session)
1042 ceph_put_mds_session(di->lease_session);
1043 kmem_cache_free(ceph_dentry_cachep, di);
1044 dentry->d_fsdata = NULL;
1048 static int ceph_snapdir_d_revalidate(struct dentry *dentry,
1049 struct nameidata *nd)
1052 * Eventually, we'll want to revalidate snapped metadata
1053 * too... probably...
1061 * read() on a dir. This weird interface hack only works if mounted
1062 * with '-o dirstat'.
1064 static ssize_t ceph_read_dir(struct file *file, char __user *buf, size_t size,
1067 struct ceph_file_info *cf = file->private_data;
1068 struct inode *inode = file->f_dentry->d_inode;
1069 struct ceph_inode_info *ci = ceph_inode(inode);
1071 const int bufsize = 1024;
1073 if (!ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb), DIRSTAT))
1076 if (!cf->dir_info) {
1077 cf->dir_info = kmalloc(bufsize, GFP_NOFS);
1081 snprintf(cf->dir_info, bufsize,
1084 " subdirs: %20lld\n"
1085 "rentries: %20lld\n"
1087 " rsubdirs: %20lld\n"
1089 "rctime: %10ld.%09ld\n",
1090 ci->i_files + ci->i_subdirs,
1093 ci->i_rfiles + ci->i_rsubdirs,
1097 (long)ci->i_rctime.tv_sec,
1098 (long)ci->i_rctime.tv_nsec);
1101 if (*ppos >= cf->dir_info_len)
1103 size = min_t(unsigned, size, cf->dir_info_len-*ppos);
1104 left = copy_to_user(buf, cf->dir_info + *ppos, size);
1107 *ppos += (size - left);
1112 * an fsync() on a dir will wait for any uncommitted directory
1113 * operations to commit.
1115 static int ceph_dir_fsync(struct file *file, int datasync)
1117 struct inode *inode = file->f_path.dentry->d_inode;
1118 struct ceph_inode_info *ci = ceph_inode(inode);
1119 struct list_head *head = &ci->i_unsafe_dirops;
1120 struct ceph_mds_request *req;
1124 dout("dir_fsync %p\n", inode);
1125 spin_lock(&ci->i_unsafe_lock);
1126 if (list_empty(head))
1129 req = list_entry(head->prev,
1130 struct ceph_mds_request, r_unsafe_dir_item);
1131 last_tid = req->r_tid;
1134 ceph_mdsc_get_request(req);
1135 spin_unlock(&ci->i_unsafe_lock);
1136 dout("dir_fsync %p wait on tid %llu (until %llu)\n",
1137 inode, req->r_tid, last_tid);
1138 if (req->r_timeout) {
1139 ret = wait_for_completion_timeout(
1140 &req->r_safe_completion, req->r_timeout);
1144 ret = -EIO; /* timed out */
1146 wait_for_completion(&req->r_safe_completion);
1148 spin_lock(&ci->i_unsafe_lock);
1149 ceph_mdsc_put_request(req);
1151 if (ret || list_empty(head))
1153 req = list_entry(head->next,
1154 struct ceph_mds_request, r_unsafe_dir_item);
1155 } while (req->r_tid < last_tid);
1157 spin_unlock(&ci->i_unsafe_lock);
1162 * We maintain a private dentry LRU.
1164 * FIXME: this needs to be changed to a per-mds lru to be useful.
1166 void ceph_dentry_lru_add(struct dentry *dn)
1168 struct ceph_dentry_info *di = ceph_dentry(dn);
1169 struct ceph_mds_client *mdsc;
1171 dout("dentry_lru_add %p %p '%.*s'\n", di, dn,
1172 dn->d_name.len, dn->d_name.name);
1174 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1175 spin_lock(&mdsc->dentry_lru_lock);
1176 list_add_tail(&di->lru, &mdsc->dentry_lru);
1178 spin_unlock(&mdsc->dentry_lru_lock);
1182 void ceph_dentry_lru_touch(struct dentry *dn)
1184 struct ceph_dentry_info *di = ceph_dentry(dn);
1185 struct ceph_mds_client *mdsc;
1187 dout("dentry_lru_touch %p %p '%.*s' (offset %lld)\n", di, dn,
1188 dn->d_name.len, dn->d_name.name, di->offset);
1190 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1191 spin_lock(&mdsc->dentry_lru_lock);
1192 list_move_tail(&di->lru, &mdsc->dentry_lru);
1193 spin_unlock(&mdsc->dentry_lru_lock);
1197 void ceph_dentry_lru_del(struct dentry *dn)
1199 struct ceph_dentry_info *di = ceph_dentry(dn);
1200 struct ceph_mds_client *mdsc;
1202 dout("dentry_lru_del %p %p '%.*s'\n", di, dn,
1203 dn->d_name.len, dn->d_name.name);
1205 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1206 spin_lock(&mdsc->dentry_lru_lock);
1207 list_del_init(&di->lru);
1209 spin_unlock(&mdsc->dentry_lru_lock);
1214 * Return name hash for a given dentry. This is dependent on
1215 * the parent directory's hash function.
1217 unsigned ceph_dentry_hash(struct dentry *dn)
1219 struct inode *dir = dn->d_parent->d_inode;
1220 struct ceph_inode_info *dci = ceph_inode(dir);
1222 switch (dci->i_dir_layout.dl_dir_hash) {
1223 case 0: /* for backward compat */
1224 case CEPH_STR_HASH_LINUX:
1225 return dn->d_name.hash;
1228 return ceph_str_hash(dci->i_dir_layout.dl_dir_hash,
1229 dn->d_name.name, dn->d_name.len);
1233 const struct file_operations ceph_dir_fops = {
1234 .read = ceph_read_dir,
1235 .readdir = ceph_readdir,
1236 .llseek = ceph_dir_llseek,
1238 .release = ceph_release,
1239 .unlocked_ioctl = ceph_ioctl,
1240 .fsync = ceph_dir_fsync,
1243 const struct inode_operations ceph_dir_iops = {
1244 .lookup = ceph_lookup,
1245 .permission = ceph_permission,
1246 .getattr = ceph_getattr,
1247 .setattr = ceph_setattr,
1248 .setxattr = ceph_setxattr,
1249 .getxattr = ceph_getxattr,
1250 .listxattr = ceph_listxattr,
1251 .removexattr = ceph_removexattr,
1252 .mknod = ceph_mknod,
1253 .symlink = ceph_symlink,
1254 .mkdir = ceph_mkdir,
1256 .unlink = ceph_unlink,
1257 .rmdir = ceph_unlink,
1258 .rename = ceph_rename,
1259 .create = ceph_create,
1262 const struct dentry_operations ceph_dentry_ops = {
1263 .d_revalidate = ceph_d_revalidate,
1264 .d_release = ceph_d_release,
1267 const struct dentry_operations ceph_snapdir_dentry_ops = {
1268 .d_revalidate = ceph_snapdir_d_revalidate,
1269 .d_release = ceph_d_release,
1272 const struct dentry_operations ceph_snap_dentry_ops = {
1273 .d_release = ceph_d_release,