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 * Complete dir indicates that 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 *file, struct dir_context *ctx)
116 struct ceph_file_info *fi = file->private_data;
117 struct dentry *parent = file->f_dentry;
118 struct inode *dir = parent->d_inode;
120 struct dentry *dentry, *last;
121 struct ceph_dentry_info *di;
124 /* claim ref on last dentry we returned */
128 dout("__dcache_readdir %p at %llu (last %p)\n", dir, ctx->pos,
131 spin_lock(&parent->d_lock);
133 /* start at beginning? */
134 if (ctx->pos == 2 || last == NULL ||
135 ctx->pos < ceph_dentry(last)->offset) {
136 if (list_empty(&parent->d_subdirs))
138 p = parent->d_subdirs.prev;
139 dout(" initial p %p/%p\n", p->prev, p->next);
141 p = last->d_u.d_child.prev;
145 dentry = list_entry(p, struct dentry, d_u.d_child);
146 di = ceph_dentry(dentry);
148 dout(" p %p/%p %s d_subdirs %p/%p\n", p->prev, p->next,
149 d_unhashed(dentry) ? "!hashed" : "hashed",
150 parent->d_subdirs.prev, parent->d_subdirs.next);
151 if (p == &parent->d_subdirs) {
152 fi->flags |= CEPH_F_ATEND;
155 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
156 if (!d_unhashed(dentry) && dentry->d_inode &&
157 ceph_snap(dentry->d_inode) != CEPH_SNAPDIR &&
158 ceph_ino(dentry->d_inode) != CEPH_INO_CEPH &&
159 ctx->pos <= di->offset)
161 dout(" skipping %p %.*s at %llu (%llu)%s%s\n", dentry,
162 dentry->d_name.len, dentry->d_name.name, di->offset,
163 ctx->pos, d_unhashed(dentry) ? " unhashed" : "",
164 !dentry->d_inode ? " null" : "");
165 spin_unlock(&dentry->d_lock);
167 dentry = list_entry(p, struct dentry, d_u.d_child);
168 di = ceph_dentry(dentry);
172 spin_unlock(&dentry->d_lock);
173 spin_unlock(&parent->d_lock);
175 dout(" %llu (%llu) dentry %p %.*s %p\n", di->offset, ctx->pos,
176 dentry, dentry->d_name.len, dentry->d_name.name, dentry->d_inode);
177 ctx->pos = di->offset;
178 if (!dir_emit(ctx, dentry->d_name.name,
180 ceph_translate_ino(dentry->d_sb, dentry->d_inode->i_ino),
181 dentry->d_inode->i_mode >> 12)) {
183 /* remember our position */
185 fi->next_offset = di->offset;
197 /* make sure a dentry wasn't dropped while we didn't have parent lock */
198 if (!ceph_dir_is_complete(dir)) {
199 dout(" lost dir complete on %p; falling back to mds\n", dir);
204 spin_lock(&parent->d_lock);
205 p = p->prev; /* advance to next dentry */
209 spin_unlock(&parent->d_lock);
217 * make note of the last dentry we read, so we can
218 * continue at the same lexicographical point,
219 * regardless of what dir changes take place on the
222 static int note_last_dentry(struct ceph_file_info *fi, const char *name,
225 kfree(fi->last_name);
226 fi->last_name = kmalloc(len+1, GFP_NOFS);
229 memcpy(fi->last_name, name, len);
230 fi->last_name[len] = 0;
231 dout("note_last_dentry '%s'\n", fi->last_name);
235 static int ceph_readdir(struct file *file, struct dir_context *ctx)
237 struct ceph_file_info *fi = file->private_data;
238 struct inode *inode = file_inode(file);
239 struct ceph_inode_info *ci = ceph_inode(inode);
240 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
241 struct ceph_mds_client *mdsc = fsc->mdsc;
242 unsigned frag = fpos_frag(ctx->pos);
243 int off = fpos_off(ctx->pos);
246 struct ceph_mds_reply_info_parsed *rinfo;
247 const int max_entries = fsc->mount_options->max_readdir;
248 const int max_bytes = fsc->mount_options->max_readdir_bytes;
250 dout("readdir %p file %p frag %u off %u\n", inode, file, frag, off);
251 if (fi->flags & CEPH_F_ATEND)
254 /* always start with . and .. */
256 /* note dir version at start of readdir so we can tell
257 * if any dentries get dropped */
258 fi->dir_release_count = atomic_read(&ci->i_release_count);
260 dout("readdir off 0 -> '.'\n");
261 if (!dir_emit(ctx, ".", 1,
262 ceph_translate_ino(inode->i_sb, inode->i_ino),
263 inode->i_mode >> 12))
269 ino_t ino = parent_ino(file->f_dentry);
270 dout("readdir off 1 -> '..'\n");
271 if (!dir_emit(ctx, "..", 2,
272 ceph_translate_ino(inode->i_sb, ino),
273 inode->i_mode >> 12))
279 /* can we use the dcache? */
280 spin_lock(&ci->i_ceph_lock);
281 if ((ctx->pos == 2 || fi->dentry) &&
282 !ceph_test_mount_opt(fsc, NOASYNCREADDIR) &&
283 ceph_snap(inode) != CEPH_SNAPDIR &&
284 __ceph_dir_is_complete(ci) &&
285 __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1)) {
286 spin_unlock(&ci->i_ceph_lock);
287 err = __dcache_readdir(file, ctx);
291 spin_unlock(&ci->i_ceph_lock);
294 err = note_last_dentry(fi, fi->dentry->d_name.name,
295 fi->dentry->d_name.len);
302 /* proceed with a normal readdir */
305 /* do we have the correct frag content buffered? */
306 if (fi->frag != frag || fi->last_readdir == NULL) {
307 struct ceph_mds_request *req;
308 int op = ceph_snap(inode) == CEPH_SNAPDIR ?
309 CEPH_MDS_OP_LSSNAP : CEPH_MDS_OP_READDIR;
311 /* discard old result, if any */
312 if (fi->last_readdir) {
313 ceph_mdsc_put_request(fi->last_readdir);
314 fi->last_readdir = NULL;
317 /* requery frag tree, as the frag topology may have changed */
318 frag = ceph_choose_frag(ceph_inode(inode), frag, NULL, NULL);
320 dout("readdir fetching %llx.%llx frag %x offset '%s'\n",
321 ceph_vinop(inode), frag, fi->last_name);
322 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
325 req->r_inode = inode;
327 req->r_dentry = dget(file->f_dentry);
328 /* hints to request -> mds selection code */
329 req->r_direct_mode = USE_AUTH_MDS;
330 req->r_direct_hash = ceph_frag_value(frag);
331 req->r_direct_is_hash = true;
332 req->r_path2 = kstrdup(fi->last_name, GFP_NOFS);
333 req->r_readdir_offset = fi->next_offset;
334 req->r_args.readdir.frag = cpu_to_le32(frag);
335 req->r_args.readdir.max_entries = cpu_to_le32(max_entries);
336 req->r_args.readdir.max_bytes = cpu_to_le32(max_bytes);
337 req->r_num_caps = max_entries + 1;
338 err = ceph_mdsc_do_request(mdsc, NULL, req);
340 ceph_mdsc_put_request(req);
343 dout("readdir got and parsed readdir result=%d"
344 " on frag %x, end=%d, complete=%d\n", err, frag,
345 (int)req->r_reply_info.dir_end,
346 (int)req->r_reply_info.dir_complete);
348 if (!req->r_did_prepopulate) {
349 dout("readdir !did_prepopulate");
350 /* preclude from marking dir complete */
351 fi->dir_release_count--;
354 /* note next offset and last dentry name */
355 rinfo = &req->r_reply_info;
356 if (le32_to_cpu(rinfo->dir_dir->frag) != frag) {
357 frag = le32_to_cpu(rinfo->dir_dir->frag);
358 if (ceph_frag_is_leftmost(frag))
362 off = fi->next_offset;
364 fi->offset = fi->next_offset;
365 fi->last_readdir = req;
368 if (req->r_reply_info.dir_end) {
369 kfree(fi->last_name);
370 fi->last_name = NULL;
371 if (ceph_frag_is_rightmost(frag))
376 err = note_last_dentry(fi,
377 rinfo->dir_dname[rinfo->dir_nr-1],
378 rinfo->dir_dname_len[rinfo->dir_nr-1]);
381 fi->next_offset += rinfo->dir_nr;
385 rinfo = &fi->last_readdir->r_reply_info;
386 dout("readdir frag %x num %d off %d chunkoff %d\n", frag,
387 rinfo->dir_nr, off, fi->offset);
389 ctx->pos = ceph_make_fpos(frag, off);
390 while (off >= fi->offset && off - fi->offset < rinfo->dir_nr) {
391 struct ceph_mds_reply_inode *in =
392 rinfo->dir_in[off - fi->offset].in;
393 struct ceph_vino vino;
396 dout("readdir off %d (%d/%d) -> %lld '%.*s' %p\n",
397 off, off - fi->offset, rinfo->dir_nr, ctx->pos,
398 rinfo->dir_dname_len[off - fi->offset],
399 rinfo->dir_dname[off - fi->offset], in);
401 ftype = le32_to_cpu(in->mode) >> 12;
402 vino.ino = le64_to_cpu(in->ino);
403 vino.snap = le64_to_cpu(in->snapid);
404 ino = ceph_vino_to_ino(vino);
406 rinfo->dir_dname[off - fi->offset],
407 rinfo->dir_dname_len[off - fi->offset],
408 ceph_translate_ino(inode->i_sb, ino), ftype)) {
409 dout("filldir stopping us...\n");
417 ceph_mdsc_put_request(fi->last_readdir);
418 fi->last_readdir = NULL;
423 if (!ceph_frag_is_rightmost(frag)) {
424 frag = ceph_frag_next(frag);
426 ctx->pos = ceph_make_fpos(frag, off);
427 dout("readdir next frag is %x\n", frag);
430 fi->flags |= CEPH_F_ATEND;
433 * if dir_release_count still matches the dir, no dentries
434 * were released during the whole readdir, and we should have
435 * the complete dir contents in our cache.
437 spin_lock(&ci->i_ceph_lock);
438 if (atomic_read(&ci->i_release_count) == fi->dir_release_count) {
439 dout(" marking %p complete\n", inode);
440 __ceph_dir_set_complete(ci, fi->dir_release_count);
441 ci->i_max_offset = ctx->pos;
443 spin_unlock(&ci->i_ceph_lock);
445 dout("readdir %p file %p done.\n", inode, file);
449 static void reset_readdir(struct ceph_file_info *fi)
451 if (fi->last_readdir) {
452 ceph_mdsc_put_request(fi->last_readdir);
453 fi->last_readdir = NULL;
455 kfree(fi->last_name);
456 fi->last_name = NULL;
457 fi->next_offset = 2; /* compensate for . and .. */
462 fi->flags &= ~CEPH_F_ATEND;
465 static loff_t ceph_dir_llseek(struct file *file, loff_t offset, int whence)
467 struct ceph_file_info *fi = file->private_data;
468 struct inode *inode = file->f_mapping->host;
469 loff_t old_offset = offset;
472 mutex_lock(&inode->i_mutex);
476 offset += inode->i_size + 2; /* FIXME */
479 offset += file->f_pos;
486 if (offset >= 0 && offset <= inode->i_sb->s_maxbytes) {
487 if (offset != file->f_pos) {
488 file->f_pos = offset;
490 fi->flags &= ~CEPH_F_ATEND;
495 * discard buffered readdir content on seekdir(0), or
496 * seek to new frag, or seek prior to current chunk.
499 fpos_frag(offset) != fpos_frag(old_offset) ||
500 fpos_off(offset) < fi->offset) {
501 dout("dir_llseek dropping %p content\n", file);
505 /* bump dir_release_count if we did a forward seek */
506 if (offset > old_offset)
507 fi->dir_release_count--;
510 mutex_unlock(&inode->i_mutex);
515 * Handle lookups for the hidden .snap directory.
517 int ceph_handle_snapdir(struct ceph_mds_request *req,
518 struct dentry *dentry, int err)
520 struct ceph_fs_client *fsc = ceph_sb_to_client(dentry->d_sb);
521 struct inode *parent = dentry->d_parent->d_inode; /* we hold i_mutex */
524 if (err == -ENOENT &&
525 ceph_snap(parent) == CEPH_NOSNAP &&
526 strcmp(dentry->d_name.name,
527 fsc->mount_options->snapdir_name) == 0) {
528 struct inode *inode = ceph_get_snapdir(parent);
529 dout("ENOENT on snapdir %p '%.*s', linking to snapdir %p\n",
530 dentry, dentry->d_name.len, dentry->d_name.name, inode);
531 BUG_ON(!d_unhashed(dentry));
532 d_add(dentry, inode);
539 * Figure out final result of a lookup/open request.
541 * Mainly, make sure we return the final req->r_dentry (if it already
542 * existed) in place of the original VFS-provided dentry when they
545 * Gracefully handle the case where the MDS replies with -ENOENT and
546 * no trace (which it may do, at its discretion, e.g., if it doesn't
547 * care to issue a lease on the negative dentry).
549 struct dentry *ceph_finish_lookup(struct ceph_mds_request *req,
550 struct dentry *dentry, int err)
552 if (err == -ENOENT) {
555 if (!req->r_reply_info.head->is_dentry) {
556 dout("ENOENT and no trace, dentry %p inode %p\n",
557 dentry, dentry->d_inode);
558 if (dentry->d_inode) {
567 dentry = ERR_PTR(err);
568 else if (dentry != req->r_dentry)
569 dentry = dget(req->r_dentry); /* we got spliced */
575 static int is_root_ceph_dentry(struct inode *inode, struct dentry *dentry)
577 return ceph_ino(inode) == CEPH_INO_ROOT &&
578 strncmp(dentry->d_name.name, ".ceph", 5) == 0;
582 * Look up a single dir entry. If there is a lookup intent, inform
583 * the MDS so that it gets our 'caps wanted' value in a single op.
585 static struct dentry *ceph_lookup(struct inode *dir, struct dentry *dentry,
588 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
589 struct ceph_mds_client *mdsc = fsc->mdsc;
590 struct ceph_mds_request *req;
594 dout("lookup %p dentry %p '%.*s'\n",
595 dir, dentry, dentry->d_name.len, dentry->d_name.name);
597 if (dentry->d_name.len > NAME_MAX)
598 return ERR_PTR(-ENAMETOOLONG);
600 err = ceph_init_dentry(dentry);
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_is_complete(ci) &&
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, 0);
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,
704 return ceph_mknod(dir, dentry, mode, 0);
707 static int ceph_symlink(struct inode *dir, struct dentry *dentry,
710 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
711 struct ceph_mds_client *mdsc = fsc->mdsc;
712 struct ceph_mds_request *req;
715 if (ceph_snap(dir) != CEPH_NOSNAP)
718 dout("symlink in dir %p dentry %p to '%s'\n", dir, dentry, dest);
719 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SYMLINK, USE_AUTH_MDS);
724 req->r_dentry = dget(dentry);
726 req->r_path2 = kstrdup(dest, GFP_NOFS);
727 req->r_locked_dir = dir;
728 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
729 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
730 err = ceph_mdsc_do_request(mdsc, dir, req);
731 if (!err && !req->r_reply_info.head->is_dentry)
732 err = ceph_handle_notrace_create(dir, dentry);
733 ceph_mdsc_put_request(req);
739 static int ceph_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
741 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
742 struct ceph_mds_client *mdsc = fsc->mdsc;
743 struct ceph_mds_request *req;
747 if (ceph_snap(dir) == CEPH_SNAPDIR) {
748 /* mkdir .snap/foo is a MKSNAP */
749 op = CEPH_MDS_OP_MKSNAP;
750 dout("mksnap dir %p snap '%.*s' dn %p\n", dir,
751 dentry->d_name.len, dentry->d_name.name, dentry);
752 } else if (ceph_snap(dir) == CEPH_NOSNAP) {
753 dout("mkdir dir %p dn %p mode 0%ho\n", dir, dentry, mode);
754 op = CEPH_MDS_OP_MKDIR;
758 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
764 req->r_dentry = dget(dentry);
766 req->r_locked_dir = dir;
767 req->r_args.mkdir.mode = cpu_to_le32(mode);
768 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
769 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
770 err = ceph_mdsc_do_request(mdsc, dir, req);
771 if (!err && !req->r_reply_info.head->is_dentry)
772 err = ceph_handle_notrace_create(dir, dentry);
773 ceph_mdsc_put_request(req);
780 static int ceph_link(struct dentry *old_dentry, struct inode *dir,
781 struct dentry *dentry)
783 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
784 struct ceph_mds_client *mdsc = fsc->mdsc;
785 struct ceph_mds_request *req;
788 if (ceph_snap(dir) != CEPH_NOSNAP)
791 dout("link in dir %p old_dentry %p dentry %p\n", dir,
793 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LINK, USE_AUTH_MDS);
798 req->r_dentry = dget(dentry);
800 req->r_old_dentry = dget(old_dentry); /* or inode? hrm. */
801 req->r_old_dentry_dir = ceph_get_dentry_parent_inode(old_dentry);
802 req->r_locked_dir = dir;
803 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
804 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
805 /* release LINK_SHARED on source inode (mds will lock it) */
806 req->r_old_inode_drop = CEPH_CAP_LINK_SHARED;
807 err = ceph_mdsc_do_request(mdsc, dir, req);
810 } else if (!req->r_reply_info.head->is_dentry) {
811 ihold(old_dentry->d_inode);
812 d_instantiate(dentry, old_dentry->d_inode);
814 ceph_mdsc_put_request(req);
819 * For a soon-to-be unlinked file, drop the AUTH_RDCACHE caps. If it
820 * looks like the link count will hit 0, drop any other caps (other
821 * than PIN) we don't specifically want (due to the file still being
824 static int drop_caps_for_unlink(struct inode *inode)
826 struct ceph_inode_info *ci = ceph_inode(inode);
827 int drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL;
829 spin_lock(&ci->i_ceph_lock);
830 if (inode->i_nlink == 1) {
831 drop |= ~(__ceph_caps_wanted(ci) | CEPH_CAP_PIN);
832 ci->i_ceph_flags |= CEPH_I_NODELAY;
834 spin_unlock(&ci->i_ceph_lock);
839 * rmdir and unlink are differ only by the metadata op code
841 static int ceph_unlink(struct inode *dir, struct dentry *dentry)
843 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
844 struct ceph_mds_client *mdsc = fsc->mdsc;
845 struct inode *inode = dentry->d_inode;
846 struct ceph_mds_request *req;
850 if (ceph_snap(dir) == CEPH_SNAPDIR) {
851 /* rmdir .snap/foo is RMSNAP */
852 dout("rmsnap dir %p '%.*s' dn %p\n", dir, dentry->d_name.len,
853 dentry->d_name.name, dentry);
854 op = CEPH_MDS_OP_RMSNAP;
855 } else if (ceph_snap(dir) == CEPH_NOSNAP) {
856 dout("unlink/rmdir dir %p dn %p inode %p\n",
858 op = S_ISDIR(dentry->d_inode->i_mode) ?
859 CEPH_MDS_OP_RMDIR : CEPH_MDS_OP_UNLINK;
862 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
867 req->r_dentry = dget(dentry);
869 req->r_locked_dir = dir;
870 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
871 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
872 req->r_inode_drop = drop_caps_for_unlink(inode);
873 err = ceph_mdsc_do_request(mdsc, dir, req);
874 if (!err && !req->r_reply_info.head->is_dentry)
876 ceph_mdsc_put_request(req);
881 static int ceph_rename(struct inode *old_dir, struct dentry *old_dentry,
882 struct inode *new_dir, struct dentry *new_dentry)
884 struct ceph_fs_client *fsc = ceph_sb_to_client(old_dir->i_sb);
885 struct ceph_mds_client *mdsc = fsc->mdsc;
886 struct ceph_mds_request *req;
889 if (ceph_snap(old_dir) != ceph_snap(new_dir))
891 if (ceph_snap(old_dir) != CEPH_NOSNAP ||
892 ceph_snap(new_dir) != CEPH_NOSNAP)
894 dout("rename dir %p dentry %p to dir %p dentry %p\n",
895 old_dir, old_dentry, new_dir, new_dentry);
896 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_RENAME, USE_AUTH_MDS);
899 req->r_dentry = dget(new_dentry);
901 req->r_old_dentry = dget(old_dentry);
902 req->r_old_dentry_dir = ceph_get_dentry_parent_inode(old_dentry);
903 req->r_locked_dir = new_dir;
904 req->r_old_dentry_drop = CEPH_CAP_FILE_SHARED;
905 req->r_old_dentry_unless = CEPH_CAP_FILE_EXCL;
906 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
907 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
908 /* release LINK_RDCACHE on source inode (mds will lock it) */
909 req->r_old_inode_drop = CEPH_CAP_LINK_SHARED;
910 if (new_dentry->d_inode)
911 req->r_inode_drop = drop_caps_for_unlink(new_dentry->d_inode);
912 err = ceph_mdsc_do_request(mdsc, old_dir, req);
913 if (!err && !req->r_reply_info.head->is_dentry) {
915 * Normally d_move() is done by fill_trace (called by
916 * do_request, above). If there is no trace, we need
920 /* d_move screws up d_subdirs order */
921 ceph_dir_clear_complete(new_dir);
923 d_move(old_dentry, new_dentry);
925 /* ensure target dentry is invalidated, despite
926 rehashing bug in vfs_rename_dir */
927 ceph_invalidate_dentry_lease(new_dentry);
929 ceph_mdsc_put_request(req);
934 * Ensure a dentry lease will no longer revalidate.
936 void ceph_invalidate_dentry_lease(struct dentry *dentry)
938 spin_lock(&dentry->d_lock);
939 dentry->d_time = jiffies;
940 ceph_dentry(dentry)->lease_shared_gen = 0;
941 spin_unlock(&dentry->d_lock);
945 * Check if dentry lease is valid. If not, delete the lease. Try to
946 * renew if the least is more than half up.
948 static int dentry_lease_is_valid(struct dentry *dentry)
950 struct ceph_dentry_info *di;
951 struct ceph_mds_session *s;
955 struct ceph_mds_session *session = NULL;
956 struct inode *dir = NULL;
959 spin_lock(&dentry->d_lock);
960 di = ceph_dentry(dentry);
961 if (di->lease_session) {
962 s = di->lease_session;
963 spin_lock(&s->s_gen_ttl_lock);
966 spin_unlock(&s->s_gen_ttl_lock);
968 if (di->lease_gen == gen &&
969 time_before(jiffies, dentry->d_time) &&
970 time_before(jiffies, ttl)) {
972 if (di->lease_renew_after &&
973 time_after(jiffies, di->lease_renew_after)) {
974 /* we should renew */
975 dir = dentry->d_parent->d_inode;
976 session = ceph_get_mds_session(s);
978 di->lease_renew_after = 0;
979 di->lease_renew_from = jiffies;
983 spin_unlock(&dentry->d_lock);
986 ceph_mdsc_lease_send_msg(session, dir, dentry,
987 CEPH_MDS_LEASE_RENEW, seq);
988 ceph_put_mds_session(session);
990 dout("dentry_lease_is_valid - dentry %p = %d\n", dentry, valid);
995 * Check if directory-wide content lease/cap is valid.
997 static int dir_lease_is_valid(struct inode *dir, struct dentry *dentry)
999 struct ceph_inode_info *ci = ceph_inode(dir);
1000 struct ceph_dentry_info *di = ceph_dentry(dentry);
1003 spin_lock(&ci->i_ceph_lock);
1004 if (ci->i_shared_gen == di->lease_shared_gen)
1005 valid = __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1);
1006 spin_unlock(&ci->i_ceph_lock);
1007 dout("dir_lease_is_valid dir %p v%u dentry %p v%u = %d\n",
1008 dir, (unsigned)ci->i_shared_gen, dentry,
1009 (unsigned)di->lease_shared_gen, valid);
1014 * Check if cached dentry can be trusted.
1016 static int ceph_d_revalidate(struct dentry *dentry, unsigned int flags)
1021 if (flags & LOOKUP_RCU)
1024 dout("d_revalidate %p '%.*s' inode %p offset %lld\n", dentry,
1025 dentry->d_name.len, dentry->d_name.name, dentry->d_inode,
1026 ceph_dentry(dentry)->offset);
1028 dir = ceph_get_dentry_parent_inode(dentry);
1030 /* always trust cached snapped dentries, snapdir dentry */
1031 if (ceph_snap(dir) != CEPH_NOSNAP) {
1032 dout("d_revalidate %p '%.*s' inode %p is SNAPPED\n", dentry,
1033 dentry->d_name.len, dentry->d_name.name, dentry->d_inode);
1035 } else if (dentry->d_inode &&
1036 ceph_snap(dentry->d_inode) == CEPH_SNAPDIR) {
1038 } else if (dentry_lease_is_valid(dentry) ||
1039 dir_lease_is_valid(dir, dentry)) {
1043 dout("d_revalidate %p %s\n", dentry, valid ? "valid" : "invalid");
1045 ceph_dentry_lru_touch(dentry);
1053 * Release our ceph_dentry_info.
1055 static void ceph_d_release(struct dentry *dentry)
1057 struct ceph_dentry_info *di = ceph_dentry(dentry);
1059 dout("d_release %p\n", dentry);
1060 ceph_dentry_lru_del(dentry);
1061 if (di->lease_session)
1062 ceph_put_mds_session(di->lease_session);
1063 kmem_cache_free(ceph_dentry_cachep, di);
1064 dentry->d_fsdata = NULL;
1067 static int ceph_snapdir_d_revalidate(struct dentry *dentry,
1071 * Eventually, we'll want to revalidate snapped metadata
1072 * too... probably...
1078 * When the VFS prunes a dentry from the cache, we need to clear the
1079 * complete flag on the parent directory.
1081 * Called under dentry->d_lock.
1083 static void ceph_d_prune(struct dentry *dentry)
1085 dout("ceph_d_prune %p\n", dentry);
1087 /* do we have a valid parent? */
1088 if (IS_ROOT(dentry))
1091 /* if we are not hashed, we don't affect dir's completeness */
1092 if (d_unhashed(dentry))
1096 * we hold d_lock, so d_parent is stable, and d_fsdata is never
1097 * cleared until d_release
1099 ceph_dir_clear_complete(dentry->d_parent->d_inode);
1103 * read() on a dir. This weird interface hack only works if mounted
1104 * with '-o dirstat'.
1106 static ssize_t ceph_read_dir(struct file *file, char __user *buf, size_t size,
1109 struct ceph_file_info *cf = file->private_data;
1110 struct inode *inode = file_inode(file);
1111 struct ceph_inode_info *ci = ceph_inode(inode);
1113 const int bufsize = 1024;
1115 if (!ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb), DIRSTAT))
1118 if (!cf->dir_info) {
1119 cf->dir_info = kmalloc(bufsize, GFP_NOFS);
1123 snprintf(cf->dir_info, bufsize,
1126 " subdirs: %20lld\n"
1127 "rentries: %20lld\n"
1129 " rsubdirs: %20lld\n"
1131 "rctime: %10ld.%09ld\n",
1132 ci->i_files + ci->i_subdirs,
1135 ci->i_rfiles + ci->i_rsubdirs,
1139 (long)ci->i_rctime.tv_sec,
1140 (long)ci->i_rctime.tv_nsec);
1143 if (*ppos >= cf->dir_info_len)
1145 size = min_t(unsigned, size, cf->dir_info_len-*ppos);
1146 left = copy_to_user(buf, cf->dir_info + *ppos, size);
1149 *ppos += (size - left);
1154 * an fsync() on a dir will wait for any uncommitted directory
1155 * operations to commit.
1157 static int ceph_dir_fsync(struct file *file, loff_t start, loff_t end,
1160 struct inode *inode = file_inode(file);
1161 struct ceph_inode_info *ci = ceph_inode(inode);
1162 struct list_head *head = &ci->i_unsafe_dirops;
1163 struct ceph_mds_request *req;
1167 dout("dir_fsync %p\n", inode);
1168 ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
1171 mutex_lock(&inode->i_mutex);
1173 spin_lock(&ci->i_unsafe_lock);
1174 if (list_empty(head))
1177 req = list_entry(head->prev,
1178 struct ceph_mds_request, r_unsafe_dir_item);
1179 last_tid = req->r_tid;
1182 ceph_mdsc_get_request(req);
1183 spin_unlock(&ci->i_unsafe_lock);
1185 dout("dir_fsync %p wait on tid %llu (until %llu)\n",
1186 inode, req->r_tid, last_tid);
1187 if (req->r_timeout) {
1188 ret = wait_for_completion_timeout(
1189 &req->r_safe_completion, req->r_timeout);
1193 ret = -EIO; /* timed out */
1195 wait_for_completion(&req->r_safe_completion);
1197 ceph_mdsc_put_request(req);
1199 spin_lock(&ci->i_unsafe_lock);
1200 if (ret || list_empty(head))
1202 req = list_entry(head->next,
1203 struct ceph_mds_request, r_unsafe_dir_item);
1204 } while (req->r_tid < last_tid);
1206 spin_unlock(&ci->i_unsafe_lock);
1207 mutex_unlock(&inode->i_mutex);
1213 * We maintain a private dentry LRU.
1215 * FIXME: this needs to be changed to a per-mds lru to be useful.
1217 void ceph_dentry_lru_add(struct dentry *dn)
1219 struct ceph_dentry_info *di = ceph_dentry(dn);
1220 struct ceph_mds_client *mdsc;
1222 dout("dentry_lru_add %p %p '%.*s'\n", di, dn,
1223 dn->d_name.len, dn->d_name.name);
1224 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1225 spin_lock(&mdsc->dentry_lru_lock);
1226 list_add_tail(&di->lru, &mdsc->dentry_lru);
1228 spin_unlock(&mdsc->dentry_lru_lock);
1231 void ceph_dentry_lru_touch(struct dentry *dn)
1233 struct ceph_dentry_info *di = ceph_dentry(dn);
1234 struct ceph_mds_client *mdsc;
1236 dout("dentry_lru_touch %p %p '%.*s' (offset %lld)\n", di, dn,
1237 dn->d_name.len, dn->d_name.name, di->offset);
1238 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1239 spin_lock(&mdsc->dentry_lru_lock);
1240 list_move_tail(&di->lru, &mdsc->dentry_lru);
1241 spin_unlock(&mdsc->dentry_lru_lock);
1244 void ceph_dentry_lru_del(struct dentry *dn)
1246 struct ceph_dentry_info *di = ceph_dentry(dn);
1247 struct ceph_mds_client *mdsc;
1249 dout("dentry_lru_del %p %p '%.*s'\n", di, dn,
1250 dn->d_name.len, dn->d_name.name);
1251 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1252 spin_lock(&mdsc->dentry_lru_lock);
1253 list_del_init(&di->lru);
1255 spin_unlock(&mdsc->dentry_lru_lock);
1259 * Return name hash for a given dentry. This is dependent on
1260 * the parent directory's hash function.
1262 unsigned ceph_dentry_hash(struct inode *dir, struct dentry *dn)
1264 struct ceph_inode_info *dci = ceph_inode(dir);
1266 switch (dci->i_dir_layout.dl_dir_hash) {
1267 case 0: /* for backward compat */
1268 case CEPH_STR_HASH_LINUX:
1269 return dn->d_name.hash;
1272 return ceph_str_hash(dci->i_dir_layout.dl_dir_hash,
1273 dn->d_name.name, dn->d_name.len);
1277 const struct file_operations ceph_dir_fops = {
1278 .read = ceph_read_dir,
1279 .iterate = ceph_readdir,
1280 .llseek = ceph_dir_llseek,
1282 .release = ceph_release,
1283 .unlocked_ioctl = ceph_ioctl,
1284 .fsync = ceph_dir_fsync,
1287 const struct inode_operations ceph_dir_iops = {
1288 .lookup = ceph_lookup,
1289 .permission = ceph_permission,
1290 .getattr = ceph_getattr,
1291 .setattr = ceph_setattr,
1292 .setxattr = ceph_setxattr,
1293 .getxattr = ceph_getxattr,
1294 .listxattr = ceph_listxattr,
1295 .removexattr = ceph_removexattr,
1296 .mknod = ceph_mknod,
1297 .symlink = ceph_symlink,
1298 .mkdir = ceph_mkdir,
1300 .unlink = ceph_unlink,
1301 .rmdir = ceph_unlink,
1302 .rename = ceph_rename,
1303 .create = ceph_create,
1304 .atomic_open = ceph_atomic_open,
1307 const struct dentry_operations ceph_dentry_ops = {
1308 .d_revalidate = ceph_d_revalidate,
1309 .d_release = ceph_d_release,
1310 .d_prune = ceph_d_prune,
1313 const struct dentry_operations ceph_snapdir_dentry_ops = {
1314 .d_revalidate = ceph_snapdir_d_revalidate,
1315 .d_release = ceph_d_release,
1318 const struct dentry_operations ceph_snap_dentry_ops = {
1319 .d_release = ceph_d_release,
1320 .d_prune = ceph_d_prune,