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;
103 static int fpos_cmp(loff_t l, loff_t r)
105 int v = ceph_frag_compare(fpos_frag(l), fpos_frag(r));
108 return (int)(fpos_off(l) - fpos_off(r));
112 * When possible, we try to satisfy a readdir by peeking at the
113 * dcache. We make this work by carefully ordering dentries on
114 * d_u.d_child when we initially get results back from the MDS, and
115 * falling back to a "normal" sync readdir if any dentries in the dir
118 * Complete dir indicates that we have all dentries in the dir. It is
119 * defined IFF we hold CEPH_CAP_FILE_SHARED (which will be revoked by
120 * the MDS if/when the directory is modified).
122 static int __dcache_readdir(struct file *file, struct dir_context *ctx,
125 struct ceph_file_info *fi = file->private_data;
126 struct dentry *parent = file->f_dentry;
127 struct inode *dir = parent->d_inode;
129 struct dentry *dentry, *last;
130 struct ceph_dentry_info *di;
133 /* claim ref on last dentry we returned */
137 dout("__dcache_readdir %p v%u at %llu (last %p)\n",
138 dir, shared_gen, ctx->pos, last);
140 spin_lock(&parent->d_lock);
142 /* start at beginning? */
143 if (ctx->pos == 2 || last == NULL ||
144 ctx->pos < ceph_dentry(last)->offset) {
145 if (list_empty(&parent->d_subdirs))
147 p = parent->d_subdirs.prev;
148 dout(" initial p %p/%p\n", p->prev, p->next);
150 p = last->d_u.d_child.prev;
154 dentry = list_entry(p, struct dentry, d_u.d_child);
155 di = ceph_dentry(dentry);
157 dout(" p %p/%p %s d_subdirs %p/%p\n", p->prev, p->next,
158 d_unhashed(dentry) ? "!hashed" : "hashed",
159 parent->d_subdirs.prev, parent->d_subdirs.next);
160 if (p == &parent->d_subdirs) {
161 fi->flags |= CEPH_F_ATEND;
164 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
165 if (di->lease_shared_gen == shared_gen &&
166 !d_unhashed(dentry) && dentry->d_inode &&
167 ceph_snap(dentry->d_inode) != CEPH_SNAPDIR &&
168 ceph_ino(dentry->d_inode) != CEPH_INO_CEPH &&
169 fpos_cmp(ctx->pos, di->offset) <= 0)
171 dout(" skipping %p %.*s at %llu (%llu)%s%s\n", dentry,
172 dentry->d_name.len, dentry->d_name.name, di->offset,
173 ctx->pos, d_unhashed(dentry) ? " unhashed" : "",
174 !dentry->d_inode ? " null" : "");
175 spin_unlock(&dentry->d_lock);
177 dentry = list_entry(p, struct dentry, d_u.d_child);
178 di = ceph_dentry(dentry);
182 spin_unlock(&dentry->d_lock);
183 spin_unlock(&parent->d_lock);
185 /* make sure a dentry wasn't dropped while we didn't have parent lock */
186 if (!ceph_dir_is_complete(dir)) {
187 dout(" lost dir complete on %p; falling back to mds\n", dir);
193 dout(" %llu (%llu) dentry %p %.*s %p\n", di->offset, ctx->pos,
194 dentry, dentry->d_name.len, dentry->d_name.name, dentry->d_inode);
195 if (!dir_emit(ctx, dentry->d_name.name,
197 ceph_translate_ino(dentry->d_sb, dentry->d_inode->i_ino),
198 dentry->d_inode->i_mode >> 12)) {
200 /* remember our position */
202 fi->next_offset = fpos_off(di->offset);
208 ctx->pos = di->offset + 1;
214 spin_lock(&parent->d_lock);
215 p = p->prev; /* advance to next dentry */
219 spin_unlock(&parent->d_lock);
227 * make note of the last dentry we read, so we can
228 * continue at the same lexicographical point,
229 * regardless of what dir changes take place on the
232 static int note_last_dentry(struct ceph_file_info *fi, const char *name,
235 kfree(fi->last_name);
236 fi->last_name = kmalloc(len+1, GFP_NOFS);
239 memcpy(fi->last_name, name, len);
240 fi->last_name[len] = 0;
241 dout("note_last_dentry '%s'\n", fi->last_name);
245 static int ceph_readdir(struct file *file, struct dir_context *ctx)
247 struct ceph_file_info *fi = file->private_data;
248 struct inode *inode = file_inode(file);
249 struct ceph_inode_info *ci = ceph_inode(inode);
250 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
251 struct ceph_mds_client *mdsc = fsc->mdsc;
252 unsigned frag = fpos_frag(ctx->pos);
253 int off = fpos_off(ctx->pos);
256 struct ceph_mds_reply_info_parsed *rinfo;
258 dout("readdir %p file %p frag %u off %u\n", inode, file, frag, off);
259 if (fi->flags & CEPH_F_ATEND)
262 /* always start with . and .. */
264 /* note dir version at start of readdir so we can tell
265 * if any dentries get dropped */
266 fi->dir_release_count = atomic_read(&ci->i_release_count);
268 dout("readdir off 0 -> '.'\n");
269 if (!dir_emit(ctx, ".", 1,
270 ceph_translate_ino(inode->i_sb, inode->i_ino),
271 inode->i_mode >> 12))
277 ino_t ino = parent_ino(file->f_dentry);
278 dout("readdir off 1 -> '..'\n");
279 if (!dir_emit(ctx, "..", 2,
280 ceph_translate_ino(inode->i_sb, ino),
281 inode->i_mode >> 12))
287 /* can we use the dcache? */
288 spin_lock(&ci->i_ceph_lock);
289 if ((ctx->pos == 2 || fi->dentry) &&
290 !ceph_test_mount_opt(fsc, NOASYNCREADDIR) &&
291 ceph_snap(inode) != CEPH_SNAPDIR &&
292 __ceph_dir_is_complete(ci) &&
293 __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1)) {
294 u32 shared_gen = ci->i_shared_gen;
295 spin_unlock(&ci->i_ceph_lock);
296 err = __dcache_readdir(file, ctx, shared_gen);
299 frag = fpos_frag(ctx->pos);
300 off = fpos_off(ctx->pos);
302 spin_unlock(&ci->i_ceph_lock);
305 err = note_last_dentry(fi, fi->dentry->d_name.name,
306 fi->dentry->d_name.len);
313 /* proceed with a normal readdir */
316 /* do we have the correct frag content buffered? */
317 if (fi->frag != frag || fi->last_readdir == NULL) {
318 struct ceph_mds_request *req;
319 int op = ceph_snap(inode) == CEPH_SNAPDIR ?
320 CEPH_MDS_OP_LSSNAP : CEPH_MDS_OP_READDIR;
322 /* discard old result, if any */
323 if (fi->last_readdir) {
324 ceph_mdsc_put_request(fi->last_readdir);
325 fi->last_readdir = NULL;
328 dout("readdir fetching %llx.%llx frag %x offset '%s'\n",
329 ceph_vinop(inode), frag, fi->last_name);
330 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
333 err = ceph_alloc_readdir_reply_buffer(req, inode);
335 ceph_mdsc_put_request(req);
338 req->r_inode = inode;
340 req->r_dentry = dget(file->f_dentry);
341 /* hints to request -> mds selection code */
342 req->r_direct_mode = USE_AUTH_MDS;
343 req->r_direct_hash = ceph_frag_value(frag);
344 req->r_direct_is_hash = true;
345 req->r_path2 = kstrdup(fi->last_name, GFP_NOFS);
346 req->r_readdir_offset = fi->next_offset;
347 req->r_args.readdir.frag = cpu_to_le32(frag);
348 err = ceph_mdsc_do_request(mdsc, NULL, req);
350 ceph_mdsc_put_request(req);
353 dout("readdir got and parsed readdir result=%d"
354 " on frag %x, end=%d, complete=%d\n", err, frag,
355 (int)req->r_reply_info.dir_end,
356 (int)req->r_reply_info.dir_complete);
358 if (!req->r_did_prepopulate) {
359 dout("readdir !did_prepopulate");
360 /* preclude from marking dir complete */
361 fi->dir_release_count--;
364 /* note next offset and last dentry name */
365 rinfo = &req->r_reply_info;
366 if (le32_to_cpu(rinfo->dir_dir->frag) != frag) {
367 frag = le32_to_cpu(rinfo->dir_dir->frag);
368 if (ceph_frag_is_leftmost(frag))
372 off = fi->next_offset;
375 fi->offset = fi->next_offset;
376 fi->last_readdir = req;
378 if (req->r_reply_info.dir_end) {
379 kfree(fi->last_name);
380 fi->last_name = NULL;
381 if (ceph_frag_is_rightmost(frag))
386 err = note_last_dentry(fi,
387 rinfo->dir_dname[rinfo->dir_nr-1],
388 rinfo->dir_dname_len[rinfo->dir_nr-1]);
391 fi->next_offset += rinfo->dir_nr;
395 rinfo = &fi->last_readdir->r_reply_info;
396 dout("readdir frag %x num %d off %d chunkoff %d\n", frag,
397 rinfo->dir_nr, off, fi->offset);
399 ctx->pos = ceph_make_fpos(frag, off);
400 while (off >= fi->offset && off - fi->offset < rinfo->dir_nr) {
401 struct ceph_mds_reply_inode *in =
402 rinfo->dir_in[off - fi->offset].in;
403 struct ceph_vino vino;
406 dout("readdir off %d (%d/%d) -> %lld '%.*s' %p\n",
407 off, off - fi->offset, rinfo->dir_nr, ctx->pos,
408 rinfo->dir_dname_len[off - fi->offset],
409 rinfo->dir_dname[off - fi->offset], in);
411 ftype = le32_to_cpu(in->mode) >> 12;
412 vino.ino = le64_to_cpu(in->ino);
413 vino.snap = le64_to_cpu(in->snapid);
414 ino = ceph_vino_to_ino(vino);
416 rinfo->dir_dname[off - fi->offset],
417 rinfo->dir_dname_len[off - fi->offset],
418 ceph_translate_ino(inode->i_sb, ino), ftype)) {
419 dout("filldir stopping us...\n");
427 ceph_mdsc_put_request(fi->last_readdir);
428 fi->last_readdir = NULL;
433 if (!ceph_frag_is_rightmost(frag)) {
434 frag = ceph_frag_next(frag);
436 ctx->pos = ceph_make_fpos(frag, off);
437 dout("readdir next frag is %x\n", frag);
440 fi->flags |= CEPH_F_ATEND;
443 * if dir_release_count still matches the dir, no dentries
444 * were released during the whole readdir, and we should have
445 * the complete dir contents in our cache.
447 spin_lock(&ci->i_ceph_lock);
448 if (atomic_read(&ci->i_release_count) == fi->dir_release_count) {
449 dout(" marking %p complete\n", inode);
450 __ceph_dir_set_complete(ci, fi->dir_release_count);
451 ci->i_max_offset = ctx->pos;
453 spin_unlock(&ci->i_ceph_lock);
455 dout("readdir %p file %p done.\n", inode, file);
459 static void reset_readdir(struct ceph_file_info *fi, unsigned frag)
461 if (fi->last_readdir) {
462 ceph_mdsc_put_request(fi->last_readdir);
463 fi->last_readdir = NULL;
465 kfree(fi->last_name);
466 fi->last_name = NULL;
467 if (ceph_frag_is_leftmost(frag))
468 fi->next_offset = 2; /* compensate for . and .. */
475 fi->flags &= ~CEPH_F_ATEND;
478 static loff_t ceph_dir_llseek(struct file *file, loff_t offset, int whence)
480 struct ceph_file_info *fi = file->private_data;
481 struct inode *inode = file->f_mapping->host;
482 loff_t old_offset = ceph_make_fpos(fi->frag, fi->next_offset);
485 mutex_lock(&inode->i_mutex);
489 offset += inode->i_size + 2; /* FIXME */
492 offset += file->f_pos;
500 if (offset != file->f_pos) {
501 file->f_pos = offset;
503 fi->flags &= ~CEPH_F_ATEND;
508 * discard buffered readdir content on seekdir(0), or
509 * seek to new frag, or seek prior to current chunk.
512 fpos_frag(offset) != fi->frag ||
513 fpos_off(offset) < fi->offset) {
514 dout("dir_llseek dropping %p content\n", file);
515 reset_readdir(fi, fpos_frag(offset));
518 /* bump dir_release_count if we did a forward seek */
519 if (fpos_cmp(offset, old_offset) > 0)
520 fi->dir_release_count--;
523 mutex_unlock(&inode->i_mutex);
528 * Handle lookups for the hidden .snap directory.
530 int ceph_handle_snapdir(struct ceph_mds_request *req,
531 struct dentry *dentry, int err)
533 struct ceph_fs_client *fsc = ceph_sb_to_client(dentry->d_sb);
534 struct inode *parent = dentry->d_parent->d_inode; /* we hold i_mutex */
537 if (err == -ENOENT &&
538 ceph_snap(parent) == CEPH_NOSNAP &&
539 strcmp(dentry->d_name.name,
540 fsc->mount_options->snapdir_name) == 0) {
541 struct inode *inode = ceph_get_snapdir(parent);
542 dout("ENOENT on snapdir %p '%.*s', linking to snapdir %p\n",
543 dentry, dentry->d_name.len, dentry->d_name.name, inode);
544 BUG_ON(!d_unhashed(dentry));
545 d_add(dentry, inode);
552 * Figure out final result of a lookup/open request.
554 * Mainly, make sure we return the final req->r_dentry (if it already
555 * existed) in place of the original VFS-provided dentry when they
558 * Gracefully handle the case where the MDS replies with -ENOENT and
559 * no trace (which it may do, at its discretion, e.g., if it doesn't
560 * care to issue a lease on the negative dentry).
562 struct dentry *ceph_finish_lookup(struct ceph_mds_request *req,
563 struct dentry *dentry, int err)
565 if (err == -ENOENT) {
568 if (!req->r_reply_info.head->is_dentry) {
569 dout("ENOENT and no trace, dentry %p inode %p\n",
570 dentry, dentry->d_inode);
571 if (dentry->d_inode) {
580 dentry = ERR_PTR(err);
581 else if (dentry != req->r_dentry)
582 dentry = dget(req->r_dentry); /* we got spliced */
588 static int is_root_ceph_dentry(struct inode *inode, struct dentry *dentry)
590 return ceph_ino(inode) == CEPH_INO_ROOT &&
591 strncmp(dentry->d_name.name, ".ceph", 5) == 0;
595 * Look up a single dir entry. If there is a lookup intent, inform
596 * the MDS so that it gets our 'caps wanted' value in a single op.
598 static struct dentry *ceph_lookup(struct inode *dir, struct dentry *dentry,
601 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
602 struct ceph_mds_client *mdsc = fsc->mdsc;
603 struct ceph_mds_request *req;
607 dout("lookup %p dentry %p '%.*s'\n",
608 dir, dentry, dentry->d_name.len, dentry->d_name.name);
610 if (dentry->d_name.len > NAME_MAX)
611 return ERR_PTR(-ENAMETOOLONG);
613 err = ceph_init_dentry(dentry);
617 /* can we conclude ENOENT locally? */
618 if (dentry->d_inode == NULL) {
619 struct ceph_inode_info *ci = ceph_inode(dir);
620 struct ceph_dentry_info *di = ceph_dentry(dentry);
622 spin_lock(&ci->i_ceph_lock);
623 dout(" dir %p flags are %d\n", dir, ci->i_ceph_flags);
624 if (strncmp(dentry->d_name.name,
625 fsc->mount_options->snapdir_name,
626 dentry->d_name.len) &&
627 !is_root_ceph_dentry(dir, dentry) &&
628 __ceph_dir_is_complete(ci) &&
629 (__ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1))) {
630 spin_unlock(&ci->i_ceph_lock);
631 dout(" dir %p complete, -ENOENT\n", dir);
633 di->lease_shared_gen = ci->i_shared_gen;
636 spin_unlock(&ci->i_ceph_lock);
639 op = ceph_snap(dir) == CEPH_SNAPDIR ?
640 CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP;
641 req = ceph_mdsc_create_request(mdsc, op, USE_ANY_MDS);
643 return ERR_CAST(req);
644 req->r_dentry = dget(dentry);
646 /* we only need inode linkage */
647 req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
648 req->r_locked_dir = dir;
649 err = ceph_mdsc_do_request(mdsc, NULL, req);
650 err = ceph_handle_snapdir(req, dentry, err);
651 dentry = ceph_finish_lookup(req, dentry, err);
652 ceph_mdsc_put_request(req); /* will dput(dentry) */
653 dout("lookup result=%p\n", dentry);
658 * If we do a create but get no trace back from the MDS, follow up with
659 * a lookup (the VFS expects us to link up the provided dentry).
661 int ceph_handle_notrace_create(struct inode *dir, struct dentry *dentry)
663 struct dentry *result = ceph_lookup(dir, dentry, 0);
665 if (result && !IS_ERR(result)) {
667 * We created the item, then did a lookup, and found
668 * it was already linked to another inode we already
669 * had in our cache (and thus got spliced). Link our
670 * dentry to that inode, but don't hash it, just in
671 * case the VFS wants to dereference it.
673 BUG_ON(!result->d_inode);
674 d_instantiate(dentry, result->d_inode);
677 return PTR_ERR(result);
680 static int ceph_mknod(struct inode *dir, struct dentry *dentry,
681 umode_t mode, dev_t rdev)
683 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
684 struct ceph_mds_client *mdsc = fsc->mdsc;
685 struct ceph_mds_request *req;
688 if (ceph_snap(dir) != CEPH_NOSNAP)
691 dout("mknod in dir %p dentry %p mode 0%ho rdev %d\n",
692 dir, dentry, mode, rdev);
693 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_MKNOD, USE_AUTH_MDS);
698 req->r_dentry = dget(dentry);
700 req->r_locked_dir = dir;
701 req->r_args.mknod.mode = cpu_to_le32(mode);
702 req->r_args.mknod.rdev = cpu_to_le32(rdev);
703 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
704 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
705 err = ceph_mdsc_do_request(mdsc, dir, req);
706 if (!err && !req->r_reply_info.head->is_dentry)
707 err = ceph_handle_notrace_create(dir, dentry);
708 ceph_mdsc_put_request(req);
711 ceph_init_acl(dentry, dentry->d_inode, dir);
717 static int ceph_create(struct inode *dir, struct dentry *dentry, umode_t mode,
720 return ceph_mknod(dir, dentry, mode, 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);
751 ceph_init_acl(dentry, dentry->d_inode, dir);
757 static int ceph_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
759 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
760 struct ceph_mds_client *mdsc = fsc->mdsc;
761 struct ceph_mds_request *req;
765 if (ceph_snap(dir) == CEPH_SNAPDIR) {
766 /* mkdir .snap/foo is a MKSNAP */
767 op = CEPH_MDS_OP_MKSNAP;
768 dout("mksnap dir %p snap '%.*s' dn %p\n", dir,
769 dentry->d_name.len, dentry->d_name.name, dentry);
770 } else if (ceph_snap(dir) == CEPH_NOSNAP) {
771 dout("mkdir dir %p dn %p mode 0%ho\n", dir, dentry, mode);
772 op = CEPH_MDS_OP_MKDIR;
776 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
782 req->r_dentry = dget(dentry);
784 req->r_locked_dir = dir;
785 req->r_args.mkdir.mode = cpu_to_le32(mode);
786 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
787 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
788 err = ceph_mdsc_do_request(mdsc, dir, req);
789 if (!err && !req->r_reply_info.head->is_dentry)
790 err = ceph_handle_notrace_create(dir, dentry);
791 ceph_mdsc_put_request(req);
794 ceph_init_acl(dentry, dentry->d_inode, dir);
800 static int ceph_link(struct dentry *old_dentry, struct inode *dir,
801 struct dentry *dentry)
803 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
804 struct ceph_mds_client *mdsc = fsc->mdsc;
805 struct ceph_mds_request *req;
808 if (ceph_snap(dir) != CEPH_NOSNAP)
811 dout("link in dir %p old_dentry %p dentry %p\n", dir,
813 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LINK, USE_AUTH_MDS);
818 req->r_dentry = dget(dentry);
820 req->r_old_dentry = dget(old_dentry);
821 req->r_locked_dir = dir;
822 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
823 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
824 /* release LINK_SHARED on source inode (mds will lock it) */
825 req->r_old_inode_drop = CEPH_CAP_LINK_SHARED;
826 err = ceph_mdsc_do_request(mdsc, dir, req);
829 } else if (!req->r_reply_info.head->is_dentry) {
830 ihold(old_dentry->d_inode);
831 d_instantiate(dentry, old_dentry->d_inode);
833 ceph_mdsc_put_request(req);
838 * For a soon-to-be unlinked file, drop the AUTH_RDCACHE caps. If it
839 * looks like the link count will hit 0, drop any other caps (other
840 * than PIN) we don't specifically want (due to the file still being
843 static int drop_caps_for_unlink(struct inode *inode)
845 struct ceph_inode_info *ci = ceph_inode(inode);
846 int drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL;
848 spin_lock(&ci->i_ceph_lock);
849 if (inode->i_nlink == 1) {
850 drop |= ~(__ceph_caps_wanted(ci) | CEPH_CAP_PIN);
851 ci->i_ceph_flags |= CEPH_I_NODELAY;
853 spin_unlock(&ci->i_ceph_lock);
858 * rmdir and unlink are differ only by the metadata op code
860 static int ceph_unlink(struct inode *dir, struct dentry *dentry)
862 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
863 struct ceph_mds_client *mdsc = fsc->mdsc;
864 struct inode *inode = dentry->d_inode;
865 struct ceph_mds_request *req;
869 if (ceph_snap(dir) == CEPH_SNAPDIR) {
870 /* rmdir .snap/foo is RMSNAP */
871 dout("rmsnap dir %p '%.*s' dn %p\n", dir, dentry->d_name.len,
872 dentry->d_name.name, dentry);
873 op = CEPH_MDS_OP_RMSNAP;
874 } else if (ceph_snap(dir) == CEPH_NOSNAP) {
875 dout("unlink/rmdir dir %p dn %p inode %p\n",
877 op = S_ISDIR(dentry->d_inode->i_mode) ?
878 CEPH_MDS_OP_RMDIR : CEPH_MDS_OP_UNLINK;
881 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
886 req->r_dentry = dget(dentry);
888 req->r_locked_dir = dir;
889 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
890 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
891 req->r_inode_drop = drop_caps_for_unlink(inode);
892 err = ceph_mdsc_do_request(mdsc, dir, req);
893 if (!err && !req->r_reply_info.head->is_dentry)
895 ceph_mdsc_put_request(req);
900 static int ceph_rename(struct inode *old_dir, struct dentry *old_dentry,
901 struct inode *new_dir, struct dentry *new_dentry)
903 struct ceph_fs_client *fsc = ceph_sb_to_client(old_dir->i_sb);
904 struct ceph_mds_client *mdsc = fsc->mdsc;
905 struct ceph_mds_request *req;
908 if (ceph_snap(old_dir) != ceph_snap(new_dir))
910 if (ceph_snap(old_dir) != CEPH_NOSNAP ||
911 ceph_snap(new_dir) != CEPH_NOSNAP)
913 dout("rename dir %p dentry %p to dir %p dentry %p\n",
914 old_dir, old_dentry, new_dir, new_dentry);
915 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_RENAME, USE_AUTH_MDS);
919 req->r_dentry = dget(new_dentry);
921 req->r_old_dentry = dget(old_dentry);
922 req->r_old_dentry_dir = old_dir;
923 req->r_locked_dir = new_dir;
924 req->r_old_dentry_drop = CEPH_CAP_FILE_SHARED;
925 req->r_old_dentry_unless = CEPH_CAP_FILE_EXCL;
926 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
927 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
928 /* release LINK_RDCACHE on source inode (mds will lock it) */
929 req->r_old_inode_drop = CEPH_CAP_LINK_SHARED;
930 if (new_dentry->d_inode)
931 req->r_inode_drop = drop_caps_for_unlink(new_dentry->d_inode);
932 err = ceph_mdsc_do_request(mdsc, old_dir, req);
933 if (!err && !req->r_reply_info.head->is_dentry) {
935 * Normally d_move() is done by fill_trace (called by
936 * do_request, above). If there is no trace, we need
940 /* d_move screws up d_subdirs order */
941 ceph_dir_clear_complete(new_dir);
943 d_move(old_dentry, new_dentry);
945 /* ensure target dentry is invalidated, despite
946 rehashing bug in vfs_rename_dir */
947 ceph_invalidate_dentry_lease(new_dentry);
949 ceph_mdsc_put_request(req);
954 * Ensure a dentry lease will no longer revalidate.
956 void ceph_invalidate_dentry_lease(struct dentry *dentry)
958 spin_lock(&dentry->d_lock);
959 dentry->d_time = jiffies;
960 ceph_dentry(dentry)->lease_shared_gen = 0;
961 spin_unlock(&dentry->d_lock);
965 * Check if dentry lease is valid. If not, delete the lease. Try to
966 * renew if the least is more than half up.
968 static int dentry_lease_is_valid(struct dentry *dentry)
970 struct ceph_dentry_info *di;
971 struct ceph_mds_session *s;
975 struct ceph_mds_session *session = NULL;
976 struct inode *dir = NULL;
979 spin_lock(&dentry->d_lock);
980 di = ceph_dentry(dentry);
981 if (di->lease_session) {
982 s = di->lease_session;
983 spin_lock(&s->s_gen_ttl_lock);
986 spin_unlock(&s->s_gen_ttl_lock);
988 if (di->lease_gen == gen &&
989 time_before(jiffies, dentry->d_time) &&
990 time_before(jiffies, ttl)) {
992 if (di->lease_renew_after &&
993 time_after(jiffies, di->lease_renew_after)) {
994 /* we should renew */
995 dir = dentry->d_parent->d_inode;
996 session = ceph_get_mds_session(s);
998 di->lease_renew_after = 0;
999 di->lease_renew_from = jiffies;
1003 spin_unlock(&dentry->d_lock);
1006 ceph_mdsc_lease_send_msg(session, dir, dentry,
1007 CEPH_MDS_LEASE_RENEW, seq);
1008 ceph_put_mds_session(session);
1010 dout("dentry_lease_is_valid - dentry %p = %d\n", dentry, valid);
1015 * Check if directory-wide content lease/cap is valid.
1017 static int dir_lease_is_valid(struct inode *dir, struct dentry *dentry)
1019 struct ceph_inode_info *ci = ceph_inode(dir);
1020 struct ceph_dentry_info *di = ceph_dentry(dentry);
1023 spin_lock(&ci->i_ceph_lock);
1024 if (ci->i_shared_gen == di->lease_shared_gen)
1025 valid = __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1);
1026 spin_unlock(&ci->i_ceph_lock);
1027 dout("dir_lease_is_valid dir %p v%u dentry %p v%u = %d\n",
1028 dir, (unsigned)ci->i_shared_gen, dentry,
1029 (unsigned)di->lease_shared_gen, valid);
1034 * Check if cached dentry can be trusted.
1036 static int ceph_d_revalidate(struct dentry *dentry, unsigned int flags)
1041 if (flags & LOOKUP_RCU)
1044 dout("d_revalidate %p '%.*s' inode %p offset %lld\n", dentry,
1045 dentry->d_name.len, dentry->d_name.name, dentry->d_inode,
1046 ceph_dentry(dentry)->offset);
1048 dir = ceph_get_dentry_parent_inode(dentry);
1050 /* always trust cached snapped dentries, snapdir dentry */
1051 if (ceph_snap(dir) != CEPH_NOSNAP) {
1052 dout("d_revalidate %p '%.*s' inode %p is SNAPPED\n", dentry,
1053 dentry->d_name.len, dentry->d_name.name, dentry->d_inode);
1055 } else if (dentry->d_inode &&
1056 ceph_snap(dentry->d_inode) == CEPH_SNAPDIR) {
1058 } else if (dentry_lease_is_valid(dentry) ||
1059 dir_lease_is_valid(dir, dentry)) {
1060 if (dentry->d_inode)
1061 valid = ceph_is_any_caps(dentry->d_inode);
1066 dout("d_revalidate %p %s\n", dentry, valid ? "valid" : "invalid");
1068 ceph_dentry_lru_touch(dentry);
1070 ceph_dir_clear_complete(dir);
1078 * Release our ceph_dentry_info.
1080 static void ceph_d_release(struct dentry *dentry)
1082 struct ceph_dentry_info *di = ceph_dentry(dentry);
1084 dout("d_release %p\n", dentry);
1085 ceph_dentry_lru_del(dentry);
1086 if (di->lease_session)
1087 ceph_put_mds_session(di->lease_session);
1088 kmem_cache_free(ceph_dentry_cachep, di);
1089 dentry->d_fsdata = NULL;
1092 static int ceph_snapdir_d_revalidate(struct dentry *dentry,
1096 * Eventually, we'll want to revalidate snapped metadata
1097 * too... probably...
1103 * When the VFS prunes a dentry from the cache, we need to clear the
1104 * complete flag on the parent directory.
1106 * Called under dentry->d_lock.
1108 static void ceph_d_prune(struct dentry *dentry)
1110 dout("ceph_d_prune %p\n", dentry);
1112 /* do we have a valid parent? */
1113 if (IS_ROOT(dentry))
1116 /* if we are not hashed, we don't affect dir's completeness */
1117 if (d_unhashed(dentry))
1121 * we hold d_lock, so d_parent is stable, and d_fsdata is never
1122 * cleared until d_release
1124 ceph_dir_clear_complete(dentry->d_parent->d_inode);
1128 * read() on a dir. This weird interface hack only works if mounted
1129 * with '-o dirstat'.
1131 static ssize_t ceph_read_dir(struct file *file, char __user *buf, size_t size,
1134 struct ceph_file_info *cf = file->private_data;
1135 struct inode *inode = file_inode(file);
1136 struct ceph_inode_info *ci = ceph_inode(inode);
1138 const int bufsize = 1024;
1140 if (!ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb), DIRSTAT))
1143 if (!cf->dir_info) {
1144 cf->dir_info = kmalloc(bufsize, GFP_NOFS);
1148 snprintf(cf->dir_info, bufsize,
1151 " subdirs: %20lld\n"
1152 "rentries: %20lld\n"
1154 " rsubdirs: %20lld\n"
1156 "rctime: %10ld.%09ld\n",
1157 ci->i_files + ci->i_subdirs,
1160 ci->i_rfiles + ci->i_rsubdirs,
1164 (long)ci->i_rctime.tv_sec,
1165 (long)ci->i_rctime.tv_nsec);
1168 if (*ppos >= cf->dir_info_len)
1170 size = min_t(unsigned, size, cf->dir_info_len-*ppos);
1171 left = copy_to_user(buf, cf->dir_info + *ppos, size);
1174 *ppos += (size - left);
1179 * an fsync() on a dir will wait for any uncommitted directory
1180 * operations to commit.
1182 static int ceph_dir_fsync(struct file *file, loff_t start, loff_t end,
1185 struct inode *inode = file_inode(file);
1186 struct ceph_inode_info *ci = ceph_inode(inode);
1187 struct list_head *head = &ci->i_unsafe_dirops;
1188 struct ceph_mds_request *req;
1192 dout("dir_fsync %p\n", inode);
1193 ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
1196 mutex_lock(&inode->i_mutex);
1198 spin_lock(&ci->i_unsafe_lock);
1199 if (list_empty(head))
1202 req = list_entry(head->prev,
1203 struct ceph_mds_request, r_unsafe_dir_item);
1204 last_tid = req->r_tid;
1207 ceph_mdsc_get_request(req);
1208 spin_unlock(&ci->i_unsafe_lock);
1210 dout("dir_fsync %p wait on tid %llu (until %llu)\n",
1211 inode, req->r_tid, last_tid);
1212 if (req->r_timeout) {
1213 ret = wait_for_completion_timeout(
1214 &req->r_safe_completion, req->r_timeout);
1218 ret = -EIO; /* timed out */
1220 wait_for_completion(&req->r_safe_completion);
1222 ceph_mdsc_put_request(req);
1224 spin_lock(&ci->i_unsafe_lock);
1225 if (ret || list_empty(head))
1227 req = list_entry(head->next,
1228 struct ceph_mds_request, r_unsafe_dir_item);
1229 } while (req->r_tid < last_tid);
1231 spin_unlock(&ci->i_unsafe_lock);
1232 mutex_unlock(&inode->i_mutex);
1238 * We maintain a private dentry LRU.
1240 * FIXME: this needs to be changed to a per-mds lru to be useful.
1242 void ceph_dentry_lru_add(struct dentry *dn)
1244 struct ceph_dentry_info *di = ceph_dentry(dn);
1245 struct ceph_mds_client *mdsc;
1247 dout("dentry_lru_add %p %p '%.*s'\n", di, dn,
1248 dn->d_name.len, dn->d_name.name);
1249 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1250 spin_lock(&mdsc->dentry_lru_lock);
1251 list_add_tail(&di->lru, &mdsc->dentry_lru);
1253 spin_unlock(&mdsc->dentry_lru_lock);
1256 void ceph_dentry_lru_touch(struct dentry *dn)
1258 struct ceph_dentry_info *di = ceph_dentry(dn);
1259 struct ceph_mds_client *mdsc;
1261 dout("dentry_lru_touch %p %p '%.*s' (offset %lld)\n", di, dn,
1262 dn->d_name.len, dn->d_name.name, di->offset);
1263 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1264 spin_lock(&mdsc->dentry_lru_lock);
1265 list_move_tail(&di->lru, &mdsc->dentry_lru);
1266 spin_unlock(&mdsc->dentry_lru_lock);
1269 void ceph_dentry_lru_del(struct dentry *dn)
1271 struct ceph_dentry_info *di = ceph_dentry(dn);
1272 struct ceph_mds_client *mdsc;
1274 dout("dentry_lru_del %p %p '%.*s'\n", di, dn,
1275 dn->d_name.len, dn->d_name.name);
1276 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1277 spin_lock(&mdsc->dentry_lru_lock);
1278 list_del_init(&di->lru);
1280 spin_unlock(&mdsc->dentry_lru_lock);
1284 * Return name hash for a given dentry. This is dependent on
1285 * the parent directory's hash function.
1287 unsigned ceph_dentry_hash(struct inode *dir, struct dentry *dn)
1289 struct ceph_inode_info *dci = ceph_inode(dir);
1291 switch (dci->i_dir_layout.dl_dir_hash) {
1292 case 0: /* for backward compat */
1293 case CEPH_STR_HASH_LINUX:
1294 return dn->d_name.hash;
1297 return ceph_str_hash(dci->i_dir_layout.dl_dir_hash,
1298 dn->d_name.name, dn->d_name.len);
1302 const struct file_operations ceph_dir_fops = {
1303 .read = ceph_read_dir,
1304 .iterate = ceph_readdir,
1305 .llseek = ceph_dir_llseek,
1307 .release = ceph_release,
1308 .unlocked_ioctl = ceph_ioctl,
1309 .fsync = ceph_dir_fsync,
1312 const struct inode_operations ceph_dir_iops = {
1313 .lookup = ceph_lookup,
1314 .permission = ceph_permission,
1315 .getattr = ceph_getattr,
1316 .setattr = ceph_setattr,
1317 .setxattr = ceph_setxattr,
1318 .getxattr = ceph_getxattr,
1319 .listxattr = ceph_listxattr,
1320 .removexattr = ceph_removexattr,
1321 .get_acl = ceph_get_acl,
1322 .set_acl = ceph_set_acl,
1323 .mknod = ceph_mknod,
1324 .symlink = ceph_symlink,
1325 .mkdir = ceph_mkdir,
1327 .unlink = ceph_unlink,
1328 .rmdir = ceph_unlink,
1329 .rename = ceph_rename,
1330 .create = ceph_create,
1331 .atomic_open = ceph_atomic_open,
1334 const struct dentry_operations ceph_dentry_ops = {
1335 .d_revalidate = ceph_d_revalidate,
1336 .d_release = ceph_d_release,
1337 .d_prune = ceph_d_prune,
1340 const struct dentry_operations ceph_snapdir_dentry_ops = {
1341 .d_revalidate = ceph_snapdir_d_revalidate,
1342 .d_release = ceph_d_release,
1345 const struct dentry_operations ceph_snap_dentry_ops = {
1346 .d_release = ceph_d_release,
1347 .d_prune = ceph_d_prune,