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 di = kmem_cache_alloc(ceph_dentry_cachep, GFP_NOFS | __GFP_ZERO);
45 return -ENOMEM; /* oh well */
47 spin_lock(&dentry->d_lock);
48 if (dentry->d_fsdata) {
50 kmem_cache_free(ceph_dentry_cachep, di);
54 if (ceph_snap(dentry->d_parent->d_inode) == CEPH_NOSNAP)
55 d_set_d_op(dentry, &ceph_dentry_ops);
56 else if (ceph_snap(dentry->d_parent->d_inode) == CEPH_SNAPDIR)
57 d_set_d_op(dentry, &ceph_snapdir_dentry_ops);
59 d_set_d_op(dentry, &ceph_snap_dentry_ops);
62 di->lease_session = NULL;
63 dentry->d_time = jiffies;
64 /* avoid reordering d_fsdata setup so that the check above is safe */
66 dentry->d_fsdata = di;
67 ceph_dentry_lru_add(dentry);
69 spin_unlock(&dentry->d_lock);
73 struct inode *ceph_get_dentry_parent_inode(struct dentry *dentry)
75 struct inode *inode = NULL;
80 spin_lock(&dentry->d_lock);
81 if (!IS_ROOT(dentry)) {
82 inode = dentry->d_parent->d_inode;
85 spin_unlock(&dentry->d_lock);
91 * for readdir, we encode the directory frag and offset within that
94 static unsigned fpos_frag(loff_t p)
98 static unsigned fpos_off(loff_t p)
100 return p & 0xffffffff;
104 * When possible, we try to satisfy a readdir by peeking at the
105 * dcache. We make this work by carefully ordering dentries on
106 * d_u.d_child when we initially get results back from the MDS, and
107 * falling back to a "normal" sync readdir if any dentries in the dir
110 * D_COMPLETE tells indicates we have all dentries in the dir. It is
111 * defined IFF we hold CEPH_CAP_FILE_SHARED (which will be revoked by
112 * the MDS if/when the directory is modified).
114 static int __dcache_readdir(struct file *filp,
115 void *dirent, filldir_t filldir)
117 struct ceph_file_info *fi = filp->private_data;
118 struct dentry *parent = filp->f_dentry;
119 struct inode *dir = parent->d_inode;
121 struct dentry *dentry, *last;
122 struct ceph_dentry_info *di;
125 /* claim ref on last dentry we returned */
129 dout("__dcache_readdir %p at %llu (last %p)\n", dir, filp->f_pos,
132 spin_lock(&parent->d_lock);
134 /* start at beginning? */
135 if (filp->f_pos == 2 || last == NULL ||
136 filp->f_pos < ceph_dentry(last)->offset) {
137 if (list_empty(&parent->d_subdirs))
139 p = parent->d_subdirs.prev;
140 dout(" initial p %p/%p\n", p->prev, p->next);
142 p = last->d_u.d_child.prev;
146 dentry = list_entry(p, struct dentry, d_u.d_child);
147 di = ceph_dentry(dentry);
149 dout(" p %p/%p %s d_subdirs %p/%p\n", p->prev, p->next,
150 d_unhashed(dentry) ? "!hashed" : "hashed",
151 parent->d_subdirs.prev, parent->d_subdirs.next);
152 if (p == &parent->d_subdirs) {
153 fi->flags |= CEPH_F_ATEND;
156 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
157 if (!d_unhashed(dentry) && dentry->d_inode &&
158 ceph_snap(dentry->d_inode) != CEPH_SNAPDIR &&
159 ceph_ino(dentry->d_inode) != CEPH_INO_CEPH &&
160 filp->f_pos <= di->offset)
162 dout(" skipping %p %.*s at %llu (%llu)%s%s\n", dentry,
163 dentry->d_name.len, dentry->d_name.name, di->offset,
164 filp->f_pos, d_unhashed(dentry) ? " unhashed" : "",
165 !dentry->d_inode ? " null" : "");
166 spin_unlock(&dentry->d_lock);
168 dentry = list_entry(p, struct dentry, d_u.d_child);
169 di = ceph_dentry(dentry);
173 spin_unlock(&dentry->d_lock);
174 spin_unlock(&parent->d_lock);
176 dout(" %llu (%llu) dentry %p %.*s %p\n", di->offset, filp->f_pos,
177 dentry, dentry->d_name.len, dentry->d_name.name, dentry->d_inode);
178 filp->f_pos = di->offset;
179 err = filldir(dirent, dentry->d_name.name,
180 dentry->d_name.len, di->offset,
181 ceph_translate_ino(dentry->d_sb, dentry->d_inode->i_ino),
182 dentry->d_inode->i_mode >> 12);
186 /* remember our position */
188 fi->next_offset = di->offset;
200 /* make sure a dentry wasn't dropped while we didn't have parent lock */
201 if (!ceph_dir_test_complete(dir)) {
202 dout(" lost D_COMPLETE on %p; falling back to mds\n", dir);
207 spin_lock(&parent->d_lock);
208 p = p->prev; /* advance to next dentry */
212 spin_unlock(&parent->d_lock);
220 * make note of the last dentry we read, so we can
221 * continue at the same lexicographical point,
222 * regardless of what dir changes take place on the
225 static int note_last_dentry(struct ceph_file_info *fi, const char *name,
228 kfree(fi->last_name);
229 fi->last_name = kmalloc(len+1, GFP_NOFS);
232 memcpy(fi->last_name, name, len);
233 fi->last_name[len] = 0;
234 dout("note_last_dentry '%s'\n", fi->last_name);
238 static int ceph_readdir(struct file *filp, void *dirent, filldir_t filldir)
240 struct ceph_file_info *fi = filp->private_data;
241 struct inode *inode = filp->f_dentry->d_inode;
242 struct ceph_inode_info *ci = ceph_inode(inode);
243 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
244 struct ceph_mds_client *mdsc = fsc->mdsc;
245 unsigned frag = fpos_frag(filp->f_pos);
246 int off = fpos_off(filp->f_pos);
249 struct ceph_mds_reply_info_parsed *rinfo;
250 const int max_entries = fsc->mount_options->max_readdir;
251 const int max_bytes = fsc->mount_options->max_readdir_bytes;
253 dout("readdir %p filp %p frag %u off %u\n", inode, filp, frag, off);
254 if (fi->flags & CEPH_F_ATEND)
257 /* always start with . and .. */
258 if (filp->f_pos == 0) {
259 /* note dir version at start of readdir so we can tell
260 * if any dentries get dropped */
261 fi->dir_release_count = ci->i_release_count;
263 dout("readdir off 0 -> '.'\n");
264 if (filldir(dirent, ".", 1, ceph_make_fpos(0, 0),
265 ceph_translate_ino(inode->i_sb, inode->i_ino),
266 inode->i_mode >> 12) < 0)
271 if (filp->f_pos == 1) {
272 ino_t ino = parent_ino(filp->f_dentry);
273 dout("readdir off 1 -> '..'\n");
274 if (filldir(dirent, "..", 2, ceph_make_fpos(0, 1),
275 ceph_translate_ino(inode->i_sb, ino),
276 inode->i_mode >> 12) < 0)
282 /* can we use the dcache? */
283 spin_lock(&ci->i_ceph_lock);
284 if ((filp->f_pos == 2 || fi->dentry) &&
285 !ceph_test_mount_opt(fsc, NOASYNCREADDIR) &&
286 ceph_snap(inode) != CEPH_SNAPDIR &&
287 ceph_dir_test_complete(inode) &&
288 __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1)) {
289 spin_unlock(&ci->i_ceph_lock);
290 err = __dcache_readdir(filp, dirent, filldir);
294 spin_unlock(&ci->i_ceph_lock);
297 err = note_last_dentry(fi, fi->dentry->d_name.name,
298 fi->dentry->d_name.len);
305 /* proceed with a normal readdir */
308 /* do we have the correct frag content buffered? */
309 if (fi->frag != frag || fi->last_readdir == NULL) {
310 struct ceph_mds_request *req;
311 int op = ceph_snap(inode) == CEPH_SNAPDIR ?
312 CEPH_MDS_OP_LSSNAP : CEPH_MDS_OP_READDIR;
314 /* discard old result, if any */
315 if (fi->last_readdir) {
316 ceph_mdsc_put_request(fi->last_readdir);
317 fi->last_readdir = NULL;
320 /* requery frag tree, as the frag topology may have changed */
321 frag = ceph_choose_frag(ceph_inode(inode), frag, NULL, NULL);
323 dout("readdir fetching %llx.%llx frag %x offset '%s'\n",
324 ceph_vinop(inode), frag, fi->last_name);
325 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
328 req->r_inode = inode;
330 req->r_dentry = dget(filp->f_dentry);
331 /* hints to request -> mds selection code */
332 req->r_direct_mode = USE_AUTH_MDS;
333 req->r_direct_hash = ceph_frag_value(frag);
334 req->r_direct_is_hash = true;
335 req->r_path2 = kstrdup(fi->last_name, GFP_NOFS);
336 req->r_readdir_offset = fi->next_offset;
337 req->r_args.readdir.frag = cpu_to_le32(frag);
338 req->r_args.readdir.max_entries = cpu_to_le32(max_entries);
339 req->r_args.readdir.max_bytes = cpu_to_le32(max_bytes);
340 req->r_num_caps = max_entries + 1;
341 err = ceph_mdsc_do_request(mdsc, NULL, req);
343 ceph_mdsc_put_request(req);
346 dout("readdir got and parsed readdir result=%d"
347 " on frag %x, end=%d, complete=%d\n", err, frag,
348 (int)req->r_reply_info.dir_end,
349 (int)req->r_reply_info.dir_complete);
351 if (!req->r_did_prepopulate) {
352 dout("readdir !did_prepopulate");
353 fi->dir_release_count--; /* preclude D_COMPLETE */
356 /* note next offset and last dentry name */
357 fi->offset = fi->next_offset;
358 fi->last_readdir = req;
360 if (req->r_reply_info.dir_end) {
361 kfree(fi->last_name);
362 fi->last_name = NULL;
363 if (ceph_frag_is_rightmost(frag))
368 rinfo = &req->r_reply_info;
369 err = note_last_dentry(fi,
370 rinfo->dir_dname[rinfo->dir_nr-1],
371 rinfo->dir_dname_len[rinfo->dir_nr-1]);
374 fi->next_offset += rinfo->dir_nr;
378 rinfo = &fi->last_readdir->r_reply_info;
379 dout("readdir frag %x num %d off %d chunkoff %d\n", frag,
380 rinfo->dir_nr, off, fi->offset);
381 while (off >= fi->offset && off - fi->offset < rinfo->dir_nr) {
382 u64 pos = ceph_make_fpos(frag, off);
383 struct ceph_mds_reply_inode *in =
384 rinfo->dir_in[off - fi->offset].in;
385 struct ceph_vino vino;
388 dout("readdir off %d (%d/%d) -> %lld '%.*s' %p\n",
389 off, off - fi->offset, rinfo->dir_nr, pos,
390 rinfo->dir_dname_len[off - fi->offset],
391 rinfo->dir_dname[off - fi->offset], in);
393 ftype = le32_to_cpu(in->mode) >> 12;
394 vino.ino = le64_to_cpu(in->ino);
395 vino.snap = le64_to_cpu(in->snapid);
396 ino = ceph_vino_to_ino(vino);
398 rinfo->dir_dname[off - fi->offset],
399 rinfo->dir_dname_len[off - fi->offset],
401 ceph_translate_ino(inode->i_sb, ino), ftype) < 0) {
402 dout("filldir stopping us...\n");
406 filp->f_pos = pos + 1;
410 ceph_mdsc_put_request(fi->last_readdir);
411 fi->last_readdir = NULL;
416 if (!ceph_frag_is_rightmost(frag)) {
417 frag = ceph_frag_next(frag);
419 filp->f_pos = ceph_make_fpos(frag, off);
420 dout("readdir next frag is %x\n", frag);
423 fi->flags |= CEPH_F_ATEND;
426 * if dir_release_count still matches the dir, no dentries
427 * were released during the whole readdir, and we should have
428 * the complete dir contents in our cache.
430 spin_lock(&ci->i_ceph_lock);
431 if (ci->i_release_count == fi->dir_release_count) {
432 ceph_dir_set_complete(inode);
433 ci->i_max_offset = filp->f_pos;
435 spin_unlock(&ci->i_ceph_lock);
437 dout("readdir %p filp %p done.\n", inode, filp);
441 static void reset_readdir(struct ceph_file_info *fi)
443 if (fi->last_readdir) {
444 ceph_mdsc_put_request(fi->last_readdir);
445 fi->last_readdir = NULL;
447 kfree(fi->last_name);
448 fi->last_name = NULL;
449 fi->next_offset = 2; /* compensate for . and .. */
454 fi->flags &= ~CEPH_F_ATEND;
457 static loff_t ceph_dir_llseek(struct file *file, loff_t offset, int origin)
459 struct ceph_file_info *fi = file->private_data;
460 struct inode *inode = file->f_mapping->host;
461 loff_t old_offset = offset;
464 mutex_lock(&inode->i_mutex);
468 offset += inode->i_size + 2; /* FIXME */
471 offset += file->f_pos;
478 if (offset >= 0 && offset <= inode->i_sb->s_maxbytes) {
479 if (offset != file->f_pos) {
480 file->f_pos = offset;
482 fi->flags &= ~CEPH_F_ATEND;
487 * discard buffered readdir content on seekdir(0), or
488 * seek to new frag, or seek prior to current chunk.
491 fpos_frag(offset) != fpos_frag(old_offset) ||
492 fpos_off(offset) < fi->offset) {
493 dout("dir_llseek dropping %p content\n", file);
497 /* bump dir_release_count if we did a forward seek */
498 if (offset > old_offset)
499 fi->dir_release_count--;
502 mutex_unlock(&inode->i_mutex);
507 * Handle lookups for the hidden .snap directory.
509 int ceph_handle_snapdir(struct ceph_mds_request *req,
510 struct dentry *dentry, int err)
512 struct ceph_fs_client *fsc = ceph_sb_to_client(dentry->d_sb);
513 struct inode *parent = dentry->d_parent->d_inode; /* we hold i_mutex */
516 if (err == -ENOENT &&
517 ceph_snap(parent) == CEPH_NOSNAP &&
518 strcmp(dentry->d_name.name,
519 fsc->mount_options->snapdir_name) == 0) {
520 struct inode *inode = ceph_get_snapdir(parent);
521 dout("ENOENT on snapdir %p '%.*s', linking to snapdir %p\n",
522 dentry, dentry->d_name.len, dentry->d_name.name, inode);
523 BUG_ON(!d_unhashed(dentry));
524 d_add(dentry, inode);
531 * Figure out final result of a lookup/open request.
533 * Mainly, make sure we return the final req->r_dentry (if it already
534 * existed) in place of the original VFS-provided dentry when they
537 * Gracefully handle the case where the MDS replies with -ENOENT and
538 * no trace (which it may do, at its discretion, e.g., if it doesn't
539 * care to issue a lease on the negative dentry).
541 struct dentry *ceph_finish_lookup(struct ceph_mds_request *req,
542 struct dentry *dentry, int err)
544 if (err == -ENOENT) {
547 if (!req->r_reply_info.head->is_dentry) {
548 dout("ENOENT and no trace, dentry %p inode %p\n",
549 dentry, dentry->d_inode);
550 if (dentry->d_inode) {
559 dentry = ERR_PTR(err);
560 else if (dentry != req->r_dentry)
561 dentry = dget(req->r_dentry); /* we got spliced */
567 static int is_root_ceph_dentry(struct inode *inode, struct dentry *dentry)
569 return ceph_ino(inode) == CEPH_INO_ROOT &&
570 strncmp(dentry->d_name.name, ".ceph", 5) == 0;
574 * Look up a single dir entry. If there is a lookup intent, inform
575 * the MDS so that it gets our 'caps wanted' value in a single op.
577 static struct dentry *ceph_lookup(struct inode *dir, struct dentry *dentry,
578 struct nameidata *nd)
580 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
581 struct ceph_mds_client *mdsc = fsc->mdsc;
582 struct ceph_mds_request *req;
586 dout("lookup %p dentry %p '%.*s'\n",
587 dir, dentry, dentry->d_name.len, dentry->d_name.name);
589 if (dentry->d_name.len > NAME_MAX)
590 return ERR_PTR(-ENAMETOOLONG);
592 err = ceph_init_dentry(dentry);
596 /* open (but not create!) intent? */
598 (nd->flags & LOOKUP_OPEN) &&
599 !(nd->intent.open.flags & O_CREAT)) {
600 int mode = nd->intent.open.create_mode & ~current->fs->umask;
601 return ceph_lookup_open(dir, dentry, nd, mode, 1);
604 /* can we conclude ENOENT locally? */
605 if (dentry->d_inode == NULL) {
606 struct ceph_inode_info *ci = ceph_inode(dir);
607 struct ceph_dentry_info *di = ceph_dentry(dentry);
609 spin_lock(&ci->i_ceph_lock);
610 dout(" dir %p flags are %d\n", dir, ci->i_ceph_flags);
611 if (strncmp(dentry->d_name.name,
612 fsc->mount_options->snapdir_name,
613 dentry->d_name.len) &&
614 !is_root_ceph_dentry(dir, dentry) &&
615 ceph_dir_test_complete(dir) &&
616 (__ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1))) {
617 spin_unlock(&ci->i_ceph_lock);
618 dout(" dir %p complete, -ENOENT\n", dir);
620 di->lease_shared_gen = ci->i_shared_gen;
623 spin_unlock(&ci->i_ceph_lock);
626 op = ceph_snap(dir) == CEPH_SNAPDIR ?
627 CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP;
628 req = ceph_mdsc_create_request(mdsc, op, USE_ANY_MDS);
630 return ERR_CAST(req);
631 req->r_dentry = dget(dentry);
633 /* we only need inode linkage */
634 req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
635 req->r_locked_dir = dir;
636 err = ceph_mdsc_do_request(mdsc, NULL, req);
637 err = ceph_handle_snapdir(req, dentry, err);
638 dentry = ceph_finish_lookup(req, dentry, err);
639 ceph_mdsc_put_request(req); /* will dput(dentry) */
640 dout("lookup result=%p\n", dentry);
645 * If we do a create but get no trace back from the MDS, follow up with
646 * a lookup (the VFS expects us to link up the provided dentry).
648 int ceph_handle_notrace_create(struct inode *dir, struct dentry *dentry)
650 struct dentry *result = ceph_lookup(dir, dentry, NULL);
652 if (result && !IS_ERR(result)) {
654 * We created the item, then did a lookup, and found
655 * it was already linked to another inode we already
656 * had in our cache (and thus got spliced). Link our
657 * dentry to that inode, but don't hash it, just in
658 * case the VFS wants to dereference it.
660 BUG_ON(!result->d_inode);
661 d_instantiate(dentry, result->d_inode);
664 return PTR_ERR(result);
667 static int ceph_mknod(struct inode *dir, struct dentry *dentry,
668 umode_t mode, dev_t rdev)
670 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
671 struct ceph_mds_client *mdsc = fsc->mdsc;
672 struct ceph_mds_request *req;
675 if (ceph_snap(dir) != CEPH_NOSNAP)
678 dout("mknod in dir %p dentry %p mode 0%ho rdev %d\n",
679 dir, dentry, mode, rdev);
680 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_MKNOD, USE_AUTH_MDS);
685 req->r_dentry = dget(dentry);
687 req->r_locked_dir = dir;
688 req->r_args.mknod.mode = cpu_to_le32(mode);
689 req->r_args.mknod.rdev = cpu_to_le32(rdev);
690 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
691 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
692 err = ceph_mdsc_do_request(mdsc, dir, req);
693 if (!err && !req->r_reply_info.head->is_dentry)
694 err = ceph_handle_notrace_create(dir, dentry);
695 ceph_mdsc_put_request(req);
701 static int ceph_create(struct inode *dir, struct dentry *dentry, umode_t mode,
702 struct nameidata *nd)
704 dout("create in dir %p dentry %p name '%.*s'\n",
705 dir, dentry, dentry->d_name.len, dentry->d_name.name);
707 if (ceph_snap(dir) != CEPH_NOSNAP)
711 BUG_ON((nd->flags & LOOKUP_OPEN) == 0);
712 dentry = ceph_lookup_open(dir, dentry, nd, mode, 0);
713 /* hrm, what should i do here if we get aliased? */
715 return PTR_ERR(dentry);
719 /* fall back to mknod */
720 return ceph_mknod(dir, dentry, (mode & ~S_IFMT) | S_IFREG, 0);
723 static int ceph_symlink(struct inode *dir, struct dentry *dentry,
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;
731 if (ceph_snap(dir) != CEPH_NOSNAP)
734 dout("symlink in dir %p dentry %p to '%s'\n", dir, dentry, dest);
735 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SYMLINK, USE_AUTH_MDS);
740 req->r_dentry = dget(dentry);
742 req->r_path2 = kstrdup(dest, GFP_NOFS);
743 req->r_locked_dir = dir;
744 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
745 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
746 err = ceph_mdsc_do_request(mdsc, dir, req);
747 if (!err && !req->r_reply_info.head->is_dentry)
748 err = ceph_handle_notrace_create(dir, dentry);
749 ceph_mdsc_put_request(req);
755 static int ceph_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
757 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
758 struct ceph_mds_client *mdsc = fsc->mdsc;
759 struct ceph_mds_request *req;
763 if (ceph_snap(dir) == CEPH_SNAPDIR) {
764 /* mkdir .snap/foo is a MKSNAP */
765 op = CEPH_MDS_OP_MKSNAP;
766 dout("mksnap dir %p snap '%.*s' dn %p\n", dir,
767 dentry->d_name.len, dentry->d_name.name, dentry);
768 } else if (ceph_snap(dir) == CEPH_NOSNAP) {
769 dout("mkdir dir %p dn %p mode 0%ho\n", dir, dentry, mode);
770 op = CEPH_MDS_OP_MKDIR;
774 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
780 req->r_dentry = dget(dentry);
782 req->r_locked_dir = dir;
783 req->r_args.mkdir.mode = cpu_to_le32(mode);
784 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
785 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
786 err = ceph_mdsc_do_request(mdsc, dir, req);
787 if (!err && !req->r_reply_info.head->is_dentry)
788 err = ceph_handle_notrace_create(dir, dentry);
789 ceph_mdsc_put_request(req);
796 static int ceph_link(struct dentry *old_dentry, struct inode *dir,
797 struct dentry *dentry)
799 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
800 struct ceph_mds_client *mdsc = fsc->mdsc;
801 struct ceph_mds_request *req;
804 if (ceph_snap(dir) != CEPH_NOSNAP)
807 dout("link in dir %p old_dentry %p dentry %p\n", dir,
809 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LINK, USE_AUTH_MDS);
814 req->r_dentry = dget(dentry);
816 req->r_old_dentry = dget(old_dentry); /* or inode? hrm. */
817 req->r_old_dentry_dir = ceph_get_dentry_parent_inode(old_dentry);
818 req->r_locked_dir = dir;
819 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
820 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
821 err = ceph_mdsc_do_request(mdsc, dir, req);
824 } else if (!req->r_reply_info.head->is_dentry) {
825 ihold(old_dentry->d_inode);
826 d_instantiate(dentry, old_dentry->d_inode);
828 ceph_mdsc_put_request(req);
833 * For a soon-to-be unlinked file, drop the AUTH_RDCACHE caps. If it
834 * looks like the link count will hit 0, drop any other caps (other
835 * than PIN) we don't specifically want (due to the file still being
838 static int drop_caps_for_unlink(struct inode *inode)
840 struct ceph_inode_info *ci = ceph_inode(inode);
841 int drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL;
843 spin_lock(&ci->i_ceph_lock);
844 if (inode->i_nlink == 1) {
845 drop |= ~(__ceph_caps_wanted(ci) | CEPH_CAP_PIN);
846 ci->i_ceph_flags |= CEPH_I_NODELAY;
848 spin_unlock(&ci->i_ceph_lock);
853 * rmdir and unlink are differ only by the metadata op code
855 static int ceph_unlink(struct inode *dir, struct dentry *dentry)
857 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
858 struct ceph_mds_client *mdsc = fsc->mdsc;
859 struct inode *inode = dentry->d_inode;
860 struct ceph_mds_request *req;
864 if (ceph_snap(dir) == CEPH_SNAPDIR) {
865 /* rmdir .snap/foo is RMSNAP */
866 dout("rmsnap dir %p '%.*s' dn %p\n", dir, dentry->d_name.len,
867 dentry->d_name.name, dentry);
868 op = CEPH_MDS_OP_RMSNAP;
869 } else if (ceph_snap(dir) == CEPH_NOSNAP) {
870 dout("unlink/rmdir dir %p dn %p inode %p\n",
872 op = S_ISDIR(dentry->d_inode->i_mode) ?
873 CEPH_MDS_OP_RMDIR : CEPH_MDS_OP_UNLINK;
876 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
881 req->r_dentry = dget(dentry);
883 req->r_locked_dir = dir;
884 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
885 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
886 req->r_inode_drop = drop_caps_for_unlink(inode);
887 err = ceph_mdsc_do_request(mdsc, dir, req);
888 if (!err && !req->r_reply_info.head->is_dentry)
890 ceph_mdsc_put_request(req);
895 static int ceph_rename(struct inode *old_dir, struct dentry *old_dentry,
896 struct inode *new_dir, struct dentry *new_dentry)
898 struct ceph_fs_client *fsc = ceph_sb_to_client(old_dir->i_sb);
899 struct ceph_mds_client *mdsc = fsc->mdsc;
900 struct ceph_mds_request *req;
903 if (ceph_snap(old_dir) != ceph_snap(new_dir))
905 if (ceph_snap(old_dir) != CEPH_NOSNAP ||
906 ceph_snap(new_dir) != CEPH_NOSNAP)
908 dout("rename dir %p dentry %p to dir %p dentry %p\n",
909 old_dir, old_dentry, new_dir, new_dentry);
910 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_RENAME, USE_AUTH_MDS);
913 req->r_dentry = dget(new_dentry);
915 req->r_old_dentry = dget(old_dentry);
916 req->r_old_dentry_dir = ceph_get_dentry_parent_inode(old_dentry);
917 req->r_locked_dir = new_dir;
918 req->r_old_dentry_drop = CEPH_CAP_FILE_SHARED;
919 req->r_old_dentry_unless = CEPH_CAP_FILE_EXCL;
920 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
921 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
922 /* release LINK_RDCACHE on source inode (mds will lock it) */
923 req->r_old_inode_drop = CEPH_CAP_LINK_SHARED;
924 if (new_dentry->d_inode)
925 req->r_inode_drop = drop_caps_for_unlink(new_dentry->d_inode);
926 err = ceph_mdsc_do_request(mdsc, old_dir, req);
927 if (!err && !req->r_reply_info.head->is_dentry) {
929 * Normally d_move() is done by fill_trace (called by
930 * do_request, above). If there is no trace, we need
934 /* d_move screws up d_subdirs order */
935 ceph_dir_clear_complete(new_dir);
937 d_move(old_dentry, new_dentry);
939 /* ensure target dentry is invalidated, despite
940 rehashing bug in vfs_rename_dir */
941 ceph_invalidate_dentry_lease(new_dentry);
943 ceph_mdsc_put_request(req);
948 * Ensure a dentry lease will no longer revalidate.
950 void ceph_invalidate_dentry_lease(struct dentry *dentry)
952 spin_lock(&dentry->d_lock);
953 dentry->d_time = jiffies;
954 ceph_dentry(dentry)->lease_shared_gen = 0;
955 spin_unlock(&dentry->d_lock);
959 * Check if dentry lease is valid. If not, delete the lease. Try to
960 * renew if the least is more than half up.
962 static int dentry_lease_is_valid(struct dentry *dentry)
964 struct ceph_dentry_info *di;
965 struct ceph_mds_session *s;
969 struct ceph_mds_session *session = NULL;
970 struct inode *dir = NULL;
973 spin_lock(&dentry->d_lock);
974 di = ceph_dentry(dentry);
975 if (di->lease_session) {
976 s = di->lease_session;
977 spin_lock(&s->s_gen_ttl_lock);
980 spin_unlock(&s->s_gen_ttl_lock);
982 if (di->lease_gen == gen &&
983 time_before(jiffies, dentry->d_time) &&
984 time_before(jiffies, ttl)) {
986 if (di->lease_renew_after &&
987 time_after(jiffies, di->lease_renew_after)) {
988 /* we should renew */
989 dir = dentry->d_parent->d_inode;
990 session = ceph_get_mds_session(s);
992 di->lease_renew_after = 0;
993 di->lease_renew_from = jiffies;
997 spin_unlock(&dentry->d_lock);
1000 ceph_mdsc_lease_send_msg(session, dir, dentry,
1001 CEPH_MDS_LEASE_RENEW, seq);
1002 ceph_put_mds_session(session);
1004 dout("dentry_lease_is_valid - dentry %p = %d\n", dentry, valid);
1009 * Check if directory-wide content lease/cap is valid.
1011 static int dir_lease_is_valid(struct inode *dir, struct dentry *dentry)
1013 struct ceph_inode_info *ci = ceph_inode(dir);
1014 struct ceph_dentry_info *di = ceph_dentry(dentry);
1017 spin_lock(&ci->i_ceph_lock);
1018 if (ci->i_shared_gen == di->lease_shared_gen)
1019 valid = __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1);
1020 spin_unlock(&ci->i_ceph_lock);
1021 dout("dir_lease_is_valid dir %p v%u dentry %p v%u = %d\n",
1022 dir, (unsigned)ci->i_shared_gen, dentry,
1023 (unsigned)di->lease_shared_gen, valid);
1028 * Check if cached dentry can be trusted.
1030 static int ceph_d_revalidate(struct dentry *dentry, struct nameidata *nd)
1035 if (nd && nd->flags & LOOKUP_RCU)
1038 dout("d_revalidate %p '%.*s' inode %p offset %lld\n", dentry,
1039 dentry->d_name.len, dentry->d_name.name, dentry->d_inode,
1040 ceph_dentry(dentry)->offset);
1042 dir = ceph_get_dentry_parent_inode(dentry);
1044 /* always trust cached snapped dentries, snapdir dentry */
1045 if (ceph_snap(dir) != CEPH_NOSNAP) {
1046 dout("d_revalidate %p '%.*s' inode %p is SNAPPED\n", dentry,
1047 dentry->d_name.len, dentry->d_name.name, dentry->d_inode);
1049 } else if (dentry->d_inode &&
1050 ceph_snap(dentry->d_inode) == CEPH_SNAPDIR) {
1052 } else if (dentry_lease_is_valid(dentry) ||
1053 dir_lease_is_valid(dir, dentry)) {
1057 dout("d_revalidate %p %s\n", dentry, valid ? "valid" : "invalid");
1059 ceph_dentry_lru_touch(dentry);
1067 * Release our ceph_dentry_info.
1069 static void ceph_d_release(struct dentry *dentry)
1071 struct ceph_dentry_info *di = ceph_dentry(dentry);
1073 dout("d_release %p\n", dentry);
1074 ceph_dentry_lru_del(dentry);
1075 if (di->lease_session)
1076 ceph_put_mds_session(di->lease_session);
1077 kmem_cache_free(ceph_dentry_cachep, di);
1078 dentry->d_fsdata = NULL;
1081 static int ceph_snapdir_d_revalidate(struct dentry *dentry,
1082 struct nameidata *nd)
1085 * Eventually, we'll want to revalidate snapped metadata
1086 * too... probably...
1092 * Set/clear/test dir complete flag on the dir's dentry.
1094 void ceph_dir_set_complete(struct inode *inode)
1096 struct dentry *dentry = d_find_any_alias(inode);
1098 if (dentry && ceph_dentry(dentry) &&
1099 ceph_test_mount_opt(ceph_sb_to_client(dentry->d_sb), DCACHE)) {
1100 dout(" marking %p (%p) complete\n", inode, dentry);
1101 set_bit(CEPH_D_COMPLETE, &ceph_dentry(dentry)->flags);
1106 void ceph_dir_clear_complete(struct inode *inode)
1108 struct dentry *dentry = d_find_any_alias(inode);
1110 if (dentry && ceph_dentry(dentry)) {
1111 dout(" marking %p (%p) complete\n", inode, dentry);
1112 set_bit(CEPH_D_COMPLETE, &ceph_dentry(dentry)->flags);
1117 bool ceph_dir_test_complete(struct inode *inode)
1119 struct dentry *dentry = d_find_any_alias(inode);
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);
1130 * When the VFS prunes a dentry from the cache, we need to clear the
1131 * complete flag on the parent directory.
1133 * Called under dentry->d_lock.
1135 static void ceph_d_prune(struct dentry *dentry)
1137 struct ceph_dentry_info *di;
1139 dout("ceph_d_prune %p\n", dentry);
1141 /* do we have a valid parent? */
1142 if (IS_ROOT(dentry))
1145 /* if we are not hashed, we don't affect D_COMPLETE */
1146 if (d_unhashed(dentry))
1150 * we hold d_lock, so d_parent is stable, and d_fsdata is never
1151 * cleared until d_release
1153 di = ceph_dentry(dentry->d_parent);
1154 clear_bit(CEPH_D_COMPLETE, &di->flags);
1158 * read() on a dir. This weird interface hack only works if mounted
1159 * with '-o dirstat'.
1161 static ssize_t ceph_read_dir(struct file *file, char __user *buf, size_t size,
1164 struct ceph_file_info *cf = file->private_data;
1165 struct inode *inode = file->f_dentry->d_inode;
1166 struct ceph_inode_info *ci = ceph_inode(inode);
1168 const int bufsize = 1024;
1170 if (!ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb), DIRSTAT))
1173 if (!cf->dir_info) {
1174 cf->dir_info = kmalloc(bufsize, GFP_NOFS);
1178 snprintf(cf->dir_info, bufsize,
1181 " subdirs: %20lld\n"
1182 "rentries: %20lld\n"
1184 " rsubdirs: %20lld\n"
1186 "rctime: %10ld.%09ld\n",
1187 ci->i_files + ci->i_subdirs,
1190 ci->i_rfiles + ci->i_rsubdirs,
1194 (long)ci->i_rctime.tv_sec,
1195 (long)ci->i_rctime.tv_nsec);
1198 if (*ppos >= cf->dir_info_len)
1200 size = min_t(unsigned, size, cf->dir_info_len-*ppos);
1201 left = copy_to_user(buf, cf->dir_info + *ppos, size);
1204 *ppos += (size - left);
1209 * an fsync() on a dir will wait for any uncommitted directory
1210 * operations to commit.
1212 static int ceph_dir_fsync(struct file *file, loff_t start, loff_t end,
1215 struct inode *inode = file->f_path.dentry->d_inode;
1216 struct ceph_inode_info *ci = ceph_inode(inode);
1217 struct list_head *head = &ci->i_unsafe_dirops;
1218 struct ceph_mds_request *req;
1222 dout("dir_fsync %p\n", inode);
1223 ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
1226 mutex_lock(&inode->i_mutex);
1228 spin_lock(&ci->i_unsafe_lock);
1229 if (list_empty(head))
1232 req = list_entry(head->prev,
1233 struct ceph_mds_request, r_unsafe_dir_item);
1234 last_tid = req->r_tid;
1237 ceph_mdsc_get_request(req);
1238 spin_unlock(&ci->i_unsafe_lock);
1240 dout("dir_fsync %p wait on tid %llu (until %llu)\n",
1241 inode, req->r_tid, last_tid);
1242 if (req->r_timeout) {
1243 ret = wait_for_completion_timeout(
1244 &req->r_safe_completion, req->r_timeout);
1248 ret = -EIO; /* timed out */
1250 wait_for_completion(&req->r_safe_completion);
1252 ceph_mdsc_put_request(req);
1254 spin_lock(&ci->i_unsafe_lock);
1255 if (ret || list_empty(head))
1257 req = list_entry(head->next,
1258 struct ceph_mds_request, r_unsafe_dir_item);
1259 } while (req->r_tid < last_tid);
1261 spin_unlock(&ci->i_unsafe_lock);
1262 mutex_unlock(&inode->i_mutex);
1268 * We maintain a private dentry LRU.
1270 * FIXME: this needs to be changed to a per-mds lru to be useful.
1272 void ceph_dentry_lru_add(struct dentry *dn)
1274 struct ceph_dentry_info *di = ceph_dentry(dn);
1275 struct ceph_mds_client *mdsc;
1277 dout("dentry_lru_add %p %p '%.*s'\n", di, dn,
1278 dn->d_name.len, dn->d_name.name);
1279 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1280 spin_lock(&mdsc->dentry_lru_lock);
1281 list_add_tail(&di->lru, &mdsc->dentry_lru);
1283 spin_unlock(&mdsc->dentry_lru_lock);
1286 void ceph_dentry_lru_touch(struct dentry *dn)
1288 struct ceph_dentry_info *di = ceph_dentry(dn);
1289 struct ceph_mds_client *mdsc;
1291 dout("dentry_lru_touch %p %p '%.*s' (offset %lld)\n", di, dn,
1292 dn->d_name.len, dn->d_name.name, di->offset);
1293 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1294 spin_lock(&mdsc->dentry_lru_lock);
1295 list_move_tail(&di->lru, &mdsc->dentry_lru);
1296 spin_unlock(&mdsc->dentry_lru_lock);
1299 void ceph_dentry_lru_del(struct dentry *dn)
1301 struct ceph_dentry_info *di = ceph_dentry(dn);
1302 struct ceph_mds_client *mdsc;
1304 dout("dentry_lru_del %p %p '%.*s'\n", di, dn,
1305 dn->d_name.len, dn->d_name.name);
1306 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1307 spin_lock(&mdsc->dentry_lru_lock);
1308 list_del_init(&di->lru);
1310 spin_unlock(&mdsc->dentry_lru_lock);
1314 * Return name hash for a given dentry. This is dependent on
1315 * the parent directory's hash function.
1317 unsigned ceph_dentry_hash(struct inode *dir, struct dentry *dn)
1319 struct ceph_inode_info *dci = ceph_inode(dir);
1321 switch (dci->i_dir_layout.dl_dir_hash) {
1322 case 0: /* for backward compat */
1323 case CEPH_STR_HASH_LINUX:
1324 return dn->d_name.hash;
1327 return ceph_str_hash(dci->i_dir_layout.dl_dir_hash,
1328 dn->d_name.name, dn->d_name.len);
1332 const struct file_operations ceph_dir_fops = {
1333 .read = ceph_read_dir,
1334 .readdir = ceph_readdir,
1335 .llseek = ceph_dir_llseek,
1337 .release = ceph_release,
1338 .unlocked_ioctl = ceph_ioctl,
1339 .fsync = ceph_dir_fsync,
1342 const struct inode_operations ceph_dir_iops = {
1343 .lookup = ceph_lookup,
1344 .permission = ceph_permission,
1345 .getattr = ceph_getattr,
1346 .setattr = ceph_setattr,
1347 .setxattr = ceph_setxattr,
1348 .getxattr = ceph_getxattr,
1349 .listxattr = ceph_listxattr,
1350 .removexattr = ceph_removexattr,
1351 .mknod = ceph_mknod,
1352 .symlink = ceph_symlink,
1353 .mkdir = ceph_mkdir,
1355 .unlink = ceph_unlink,
1356 .rmdir = ceph_unlink,
1357 .rename = ceph_rename,
1358 .create = ceph_create,
1361 const struct dentry_operations ceph_dentry_ops = {
1362 .d_revalidate = ceph_d_revalidate,
1363 .d_release = ceph_d_release,
1364 .d_prune = ceph_d_prune,
1367 const struct dentry_operations ceph_snapdir_dentry_ops = {
1368 .d_revalidate = ceph_snapdir_d_revalidate,
1369 .d_release = ceph_d_release,
1372 const struct dentry_operations ceph_snap_dentry_ops = {
1373 .d_release = ceph_d_release,
1374 .d_prune = ceph_d_prune,