1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/ceph/ceph_debug.h>
4 #include <linux/spinlock.h>
5 #include <linux/namei.h>
6 #include <linux/slab.h>
7 #include <linux/sched.h>
8 #include <linux/xattr.h>
11 #include "mds_client.h"
14 * Directory operations: readdir, lookup, create, link, unlink,
19 * Ceph MDS operations are specified in terms of a base ino and
20 * relative path. Thus, the client can specify an operation on a
21 * specific inode (e.g., a getattr due to fstat(2)), or as a path
22 * relative to, say, the root directory.
24 * Normally, we limit ourselves to strict inode ops (no path component)
25 * or dentry operations (a single path component relative to an ino). The
26 * exception to this is open_root_dentry(), which will open the mount
30 const struct dentry_operations ceph_dentry_ops;
32 static bool __dentry_lease_is_valid(struct ceph_dentry_info *di);
33 static int __dir_lease_try_check(const struct dentry *dentry);
36 * Initialize ceph dentry state.
38 static int ceph_d_init(struct dentry *dentry)
40 struct ceph_dentry_info *di;
41 struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(dentry->d_sb);
43 di = kmem_cache_zalloc(ceph_dentry_cachep, GFP_KERNEL);
45 return -ENOMEM; /* oh well */
48 di->lease_session = NULL;
50 dentry->d_fsdata = di;
51 INIT_LIST_HEAD(&di->lease_list);
53 atomic64_inc(&mdsc->metric.total_dentries);
59 * for f_pos for readdir:
61 * (0xff << 52) | ((24 bits hash) << 28) |
62 * (the nth entry has hash collision);
64 * ((frag value) << 28) | (the nth entry in frag);
66 #define OFFSET_BITS 28
67 #define OFFSET_MASK ((1 << OFFSET_BITS) - 1)
68 #define HASH_ORDER (0xffull << (OFFSET_BITS + 24))
69 loff_t ceph_make_fpos(unsigned high, unsigned off, bool hash_order)
71 loff_t fpos = ((loff_t)high << 28) | (loff_t)off;
77 static bool is_hash_order(loff_t p)
79 return (p & HASH_ORDER) == HASH_ORDER;
82 static unsigned fpos_frag(loff_t p)
84 return p >> OFFSET_BITS;
87 static unsigned fpos_hash(loff_t p)
89 return ceph_frag_value(fpos_frag(p));
92 static unsigned fpos_off(loff_t p)
94 return p & OFFSET_MASK;
97 static int fpos_cmp(loff_t l, loff_t r)
99 int v = ceph_frag_compare(fpos_frag(l), fpos_frag(r));
102 return (int)(fpos_off(l) - fpos_off(r));
106 * make note of the last dentry we read, so we can
107 * continue at the same lexicographical point,
108 * regardless of what dir changes take place on the
111 static int note_last_dentry(struct ceph_dir_file_info *dfi, const char *name,
112 int len, unsigned next_offset)
114 char *buf = kmalloc(len+1, GFP_KERNEL);
117 kfree(dfi->last_name);
118 dfi->last_name = buf;
119 memcpy(dfi->last_name, name, len);
120 dfi->last_name[len] = 0;
121 dfi->next_offset = next_offset;
122 dout("note_last_dentry '%s'\n", dfi->last_name);
127 static struct dentry *
128 __dcache_find_get_entry(struct dentry *parent, u64 idx,
129 struct ceph_readdir_cache_control *cache_ctl)
131 struct inode *dir = d_inode(parent);
132 struct dentry *dentry;
133 unsigned idx_mask = (PAGE_SIZE / sizeof(struct dentry *)) - 1;
134 loff_t ptr_pos = idx * sizeof(struct dentry *);
135 pgoff_t ptr_pgoff = ptr_pos >> PAGE_SHIFT;
137 if (ptr_pos >= i_size_read(dir))
140 if (!cache_ctl->page || ptr_pgoff != page_index(cache_ctl->page)) {
141 ceph_readdir_cache_release(cache_ctl);
142 cache_ctl->page = find_lock_page(&dir->i_data, ptr_pgoff);
143 if (!cache_ctl->page) {
144 dout(" page %lu not found\n", ptr_pgoff);
145 return ERR_PTR(-EAGAIN);
147 /* reading/filling the cache are serialized by
148 i_mutex, no need to use page lock */
149 unlock_page(cache_ctl->page);
150 cache_ctl->dentries = kmap(cache_ctl->page);
153 cache_ctl->index = idx & idx_mask;
156 spin_lock(&parent->d_lock);
157 /* check i_size again here, because empty directory can be
158 * marked as complete while not holding the i_mutex. */
159 if (ceph_dir_is_complete_ordered(dir) && ptr_pos < i_size_read(dir))
160 dentry = cache_ctl->dentries[cache_ctl->index];
163 spin_unlock(&parent->d_lock);
164 if (dentry && !lockref_get_not_dead(&dentry->d_lockref))
167 return dentry ? : ERR_PTR(-EAGAIN);
171 * When possible, we try to satisfy a readdir by peeking at the
172 * dcache. We make this work by carefully ordering dentries on
173 * d_child when we initially get results back from the MDS, and
174 * falling back to a "normal" sync readdir if any dentries in the dir
177 * Complete dir indicates that we have all dentries in the dir. It is
178 * defined IFF we hold CEPH_CAP_FILE_SHARED (which will be revoked by
179 * the MDS if/when the directory is modified).
181 static int __dcache_readdir(struct file *file, struct dir_context *ctx,
184 struct ceph_dir_file_info *dfi = file->private_data;
185 struct dentry *parent = file->f_path.dentry;
186 struct inode *dir = d_inode(parent);
187 struct dentry *dentry, *last = NULL;
188 struct ceph_dentry_info *di;
189 struct ceph_readdir_cache_control cache_ctl = {};
193 dout("__dcache_readdir %p v%u at %llx\n", dir, (unsigned)shared_gen, ctx->pos);
195 /* search start position */
197 u64 count = div_u64(i_size_read(dir), sizeof(struct dentry *));
199 u64 step = count >> 1;
200 dentry = __dcache_find_get_entry(parent, idx + step,
203 /* use linar search */
207 if (IS_ERR(dentry)) {
208 err = PTR_ERR(dentry);
211 di = ceph_dentry(dentry);
212 spin_lock(&dentry->d_lock);
213 if (fpos_cmp(di->offset, ctx->pos) < 0) {
219 spin_unlock(&dentry->d_lock);
223 dout("__dcache_readdir %p cache idx %llu\n", dir, idx);
228 bool emit_dentry = false;
229 dentry = __dcache_find_get_entry(parent, idx++, &cache_ctl);
231 dfi->file_info.flags |= CEPH_F_ATEND;
235 if (IS_ERR(dentry)) {
236 err = PTR_ERR(dentry);
240 spin_lock(&dentry->d_lock);
241 di = ceph_dentry(dentry);
242 if (d_unhashed(dentry) ||
243 d_really_is_negative(dentry) ||
244 di->lease_shared_gen != shared_gen) {
245 spin_unlock(&dentry->d_lock);
250 if (fpos_cmp(ctx->pos, di->offset) <= 0) {
251 __ceph_dentry_dir_lease_touch(di);
254 spin_unlock(&dentry->d_lock);
257 dout(" %llx dentry %p %pd %p\n", di->offset,
258 dentry, dentry, d_inode(dentry));
259 ctx->pos = di->offset;
260 if (!dir_emit(ctx, dentry->d_name.name,
261 dentry->d_name.len, ceph_present_inode(d_inode(dentry)),
262 d_inode(dentry)->i_mode >> 12)) {
277 ceph_readdir_cache_release(&cache_ctl);
280 di = ceph_dentry(last);
281 ret = note_last_dentry(dfi, last->d_name.name, last->d_name.len,
282 fpos_off(di->offset) + 1);
286 /* last_name no longer match cache index */
287 if (dfi->readdir_cache_idx >= 0) {
288 dfi->readdir_cache_idx = -1;
289 dfi->dir_release_count = 0;
295 static bool need_send_readdir(struct ceph_dir_file_info *dfi, loff_t pos)
297 if (!dfi->last_readdir)
299 if (is_hash_order(pos))
300 return !ceph_frag_contains_value(dfi->frag, fpos_hash(pos));
302 return dfi->frag != fpos_frag(pos);
305 static int ceph_readdir(struct file *file, struct dir_context *ctx)
307 struct ceph_dir_file_info *dfi = file->private_data;
308 struct inode *inode = file_inode(file);
309 struct ceph_inode_info *ci = ceph_inode(inode);
310 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
311 struct ceph_mds_client *mdsc = fsc->mdsc;
315 struct ceph_mds_reply_info_parsed *rinfo;
317 dout("readdir %p file %p pos %llx\n", inode, file, ctx->pos);
318 if (dfi->file_info.flags & CEPH_F_ATEND)
321 /* always start with . and .. */
323 dout("readdir off 0 -> '.'\n");
324 if (!dir_emit(ctx, ".", 1, ceph_present_inode(inode),
325 inode->i_mode >> 12))
331 struct dentry *dentry = file->f_path.dentry;
333 spin_lock(&dentry->d_lock);
334 ino = ceph_present_inode(dentry->d_parent->d_inode);
335 spin_unlock(&dentry->d_lock);
337 dout("readdir off 1 -> '..'\n");
338 if (!dir_emit(ctx, "..", 2, ino, inode->i_mode >> 12))
343 spin_lock(&ci->i_ceph_lock);
344 /* request Fx cap. if have Fx, we don't need to release Fs cap
345 * for later create/unlink. */
346 __ceph_touch_fmode(ci, mdsc, CEPH_FILE_MODE_WR);
347 /* can we use the dcache? */
348 if (ceph_test_mount_opt(fsc, DCACHE) &&
349 !ceph_test_mount_opt(fsc, NOASYNCREADDIR) &&
350 ceph_snap(inode) != CEPH_SNAPDIR &&
351 __ceph_dir_is_complete_ordered(ci) &&
352 __ceph_caps_issued_mask_metric(ci, CEPH_CAP_FILE_SHARED, 1)) {
353 int shared_gen = atomic_read(&ci->i_shared_gen);
355 spin_unlock(&ci->i_ceph_lock);
356 err = __dcache_readdir(file, ctx, shared_gen);
360 spin_unlock(&ci->i_ceph_lock);
363 /* proceed with a normal readdir */
365 /* do we have the correct frag content buffered? */
366 if (need_send_readdir(dfi, ctx->pos)) {
367 struct ceph_mds_request *req;
368 int op = ceph_snap(inode) == CEPH_SNAPDIR ?
369 CEPH_MDS_OP_LSSNAP : CEPH_MDS_OP_READDIR;
371 /* discard old result, if any */
372 if (dfi->last_readdir) {
373 ceph_mdsc_put_request(dfi->last_readdir);
374 dfi->last_readdir = NULL;
377 if (is_hash_order(ctx->pos)) {
378 /* fragtree isn't always accurate. choose frag
379 * based on previous reply when possible. */
380 if (frag == (unsigned)-1)
381 frag = ceph_choose_frag(ci, fpos_hash(ctx->pos),
384 frag = fpos_frag(ctx->pos);
387 dout("readdir fetching %llx.%llx frag %x offset '%s'\n",
388 ceph_vinop(inode), frag, dfi->last_name);
389 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
392 err = ceph_alloc_readdir_reply_buffer(req, inode);
394 ceph_mdsc_put_request(req);
397 /* hints to request -> mds selection code */
398 req->r_direct_mode = USE_AUTH_MDS;
399 if (op == CEPH_MDS_OP_READDIR) {
400 req->r_direct_hash = ceph_frag_value(frag);
401 __set_bit(CEPH_MDS_R_DIRECT_IS_HASH, &req->r_req_flags);
402 req->r_inode_drop = CEPH_CAP_FILE_EXCL;
404 if (dfi->last_name) {
405 req->r_path2 = kstrdup(dfi->last_name, GFP_KERNEL);
407 ceph_mdsc_put_request(req);
410 } else if (is_hash_order(ctx->pos)) {
411 req->r_args.readdir.offset_hash =
412 cpu_to_le32(fpos_hash(ctx->pos));
415 req->r_dir_release_cnt = dfi->dir_release_count;
416 req->r_dir_ordered_cnt = dfi->dir_ordered_count;
417 req->r_readdir_cache_idx = dfi->readdir_cache_idx;
418 req->r_readdir_offset = dfi->next_offset;
419 req->r_args.readdir.frag = cpu_to_le32(frag);
420 req->r_args.readdir.flags =
421 cpu_to_le16(CEPH_READDIR_REPLY_BITFLAGS);
423 req->r_inode = inode;
425 req->r_dentry = dget(file->f_path.dentry);
426 err = ceph_mdsc_do_request(mdsc, NULL, req);
428 ceph_mdsc_put_request(req);
431 dout("readdir got and parsed readdir result=%d on "
432 "frag %x, end=%d, complete=%d, hash_order=%d\n",
434 (int)req->r_reply_info.dir_end,
435 (int)req->r_reply_info.dir_complete,
436 (int)req->r_reply_info.hash_order);
438 rinfo = &req->r_reply_info;
439 if (le32_to_cpu(rinfo->dir_dir->frag) != frag) {
440 frag = le32_to_cpu(rinfo->dir_dir->frag);
441 if (!rinfo->hash_order) {
442 dfi->next_offset = req->r_readdir_offset;
443 /* adjust ctx->pos to beginning of frag */
444 ctx->pos = ceph_make_fpos(frag,
451 dfi->last_readdir = req;
453 if (test_bit(CEPH_MDS_R_DID_PREPOPULATE, &req->r_req_flags)) {
454 dfi->readdir_cache_idx = req->r_readdir_cache_idx;
455 if (dfi->readdir_cache_idx < 0) {
456 /* preclude from marking dir ordered */
457 dfi->dir_ordered_count = 0;
458 } else if (ceph_frag_is_leftmost(frag) &&
459 dfi->next_offset == 2) {
460 /* note dir version at start of readdir so
461 * we can tell if any dentries get dropped */
462 dfi->dir_release_count = req->r_dir_release_cnt;
463 dfi->dir_ordered_count = req->r_dir_ordered_cnt;
466 dout("readdir !did_prepopulate\n");
467 /* disable readdir cache */
468 dfi->readdir_cache_idx = -1;
469 /* preclude from marking dir complete */
470 dfi->dir_release_count = 0;
473 /* note next offset and last dentry name */
474 if (rinfo->dir_nr > 0) {
475 struct ceph_mds_reply_dir_entry *rde =
476 rinfo->dir_entries + (rinfo->dir_nr-1);
477 unsigned next_offset = req->r_reply_info.dir_end ?
478 2 : (fpos_off(rde->offset) + 1);
479 err = note_last_dentry(dfi, rde->name, rde->name_len,
483 } else if (req->r_reply_info.dir_end) {
484 dfi->next_offset = 2;
489 rinfo = &dfi->last_readdir->r_reply_info;
490 dout("readdir frag %x num %d pos %llx chunk first %llx\n",
491 dfi->frag, rinfo->dir_nr, ctx->pos,
492 rinfo->dir_nr ? rinfo->dir_entries[0].offset : 0LL);
495 /* search start position */
496 if (rinfo->dir_nr > 0) {
497 int step, nr = rinfo->dir_nr;
500 if (rinfo->dir_entries[i + step].offset < ctx->pos) {
508 for (; i < rinfo->dir_nr; i++) {
509 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
511 BUG_ON(rde->offset < ctx->pos);
513 ctx->pos = rde->offset;
514 dout("readdir (%d/%d) -> %llx '%.*s' %p\n",
515 i, rinfo->dir_nr, ctx->pos,
516 rde->name_len, rde->name, &rde->inode.in);
518 BUG_ON(!rde->inode.in);
520 if (!dir_emit(ctx, rde->name, rde->name_len,
521 ceph_present_ino(inode->i_sb, le64_to_cpu(rde->inode.in->ino)),
522 le32_to_cpu(rde->inode.in->mode) >> 12)) {
523 dout("filldir stopping us...\n");
529 ceph_mdsc_put_request(dfi->last_readdir);
530 dfi->last_readdir = NULL;
532 if (dfi->next_offset > 2) {
538 if (!ceph_frag_is_rightmost(dfi->frag)) {
539 frag = ceph_frag_next(dfi->frag);
540 if (is_hash_order(ctx->pos)) {
541 loff_t new_pos = ceph_make_fpos(ceph_frag_value(frag),
542 dfi->next_offset, true);
543 if (new_pos > ctx->pos)
547 ctx->pos = ceph_make_fpos(frag, dfi->next_offset,
549 kfree(dfi->last_name);
550 dfi->last_name = NULL;
552 dout("readdir next frag is %x\n", frag);
555 dfi->file_info.flags |= CEPH_F_ATEND;
558 * if dir_release_count still matches the dir, no dentries
559 * were released during the whole readdir, and we should have
560 * the complete dir contents in our cache.
562 if (atomic64_read(&ci->i_release_count) ==
563 dfi->dir_release_count) {
564 spin_lock(&ci->i_ceph_lock);
565 if (dfi->dir_ordered_count ==
566 atomic64_read(&ci->i_ordered_count)) {
567 dout(" marking %p complete and ordered\n", inode);
568 /* use i_size to track number of entries in
570 BUG_ON(dfi->readdir_cache_idx < 0);
571 i_size_write(inode, dfi->readdir_cache_idx *
572 sizeof(struct dentry*));
574 dout(" marking %p complete\n", inode);
576 __ceph_dir_set_complete(ci, dfi->dir_release_count,
577 dfi->dir_ordered_count);
578 spin_unlock(&ci->i_ceph_lock);
581 dout("readdir %p file %p done.\n", inode, file);
585 static void reset_readdir(struct ceph_dir_file_info *dfi)
587 if (dfi->last_readdir) {
588 ceph_mdsc_put_request(dfi->last_readdir);
589 dfi->last_readdir = NULL;
591 kfree(dfi->last_name);
592 dfi->last_name = NULL;
593 dfi->dir_release_count = 0;
594 dfi->readdir_cache_idx = -1;
595 dfi->next_offset = 2; /* compensate for . and .. */
596 dfi->file_info.flags &= ~CEPH_F_ATEND;
600 * discard buffered readdir content on seekdir(0), or seek to new frag,
601 * or seek prior to current chunk
603 static bool need_reset_readdir(struct ceph_dir_file_info *dfi, loff_t new_pos)
605 struct ceph_mds_reply_info_parsed *rinfo;
609 if (is_hash_order(new_pos)) {
610 /* no need to reset last_name for a forward seek when
611 * dentries are sotred in hash order */
612 } else if (dfi->frag != fpos_frag(new_pos)) {
615 rinfo = dfi->last_readdir ? &dfi->last_readdir->r_reply_info : NULL;
616 if (!rinfo || !rinfo->dir_nr)
618 chunk_offset = rinfo->dir_entries[0].offset;
619 return new_pos < chunk_offset ||
620 is_hash_order(new_pos) != is_hash_order(chunk_offset);
623 static loff_t ceph_dir_llseek(struct file *file, loff_t offset, int whence)
625 struct ceph_dir_file_info *dfi = file->private_data;
626 struct inode *inode = file->f_mapping->host;
633 offset += file->f_pos;
638 retval = -EOPNOTSUPP;
645 if (need_reset_readdir(dfi, offset)) {
646 dout("dir_llseek dropping %p content\n", file);
648 } else if (is_hash_order(offset) && offset > file->f_pos) {
649 /* for hash offset, we don't know if a forward seek
650 * is within same frag */
651 dfi->dir_release_count = 0;
652 dfi->readdir_cache_idx = -1;
655 if (offset != file->f_pos) {
656 file->f_pos = offset;
658 dfi->file_info.flags &= ~CEPH_F_ATEND;
668 * Handle lookups for the hidden .snap directory.
670 struct dentry *ceph_handle_snapdir(struct ceph_mds_request *req,
671 struct dentry *dentry)
673 struct ceph_fs_client *fsc = ceph_sb_to_client(dentry->d_sb);
674 struct inode *parent = d_inode(dentry->d_parent); /* we hold i_mutex */
677 if (ceph_snap(parent) == CEPH_NOSNAP &&
678 strcmp(dentry->d_name.name, fsc->mount_options->snapdir_name) == 0) {
680 struct inode *inode = ceph_get_snapdir(parent);
682 res = d_splice_alias(inode, dentry);
683 dout("ENOENT on snapdir %p '%pd', linking to snapdir %p. Spliced dentry %p\n",
684 dentry, dentry, inode, res);
692 * Figure out final result of a lookup/open request.
694 * Mainly, make sure we return the final req->r_dentry (if it already
695 * existed) in place of the original VFS-provided dentry when they
698 * Gracefully handle the case where the MDS replies with -ENOENT and
699 * no trace (which it may do, at its discretion, e.g., if it doesn't
700 * care to issue a lease on the negative dentry).
702 struct dentry *ceph_finish_lookup(struct ceph_mds_request *req,
703 struct dentry *dentry, int err)
705 if (err == -ENOENT) {
708 if (!req->r_reply_info.head->is_dentry) {
709 dout("ENOENT and no trace, dentry %p inode %p\n",
710 dentry, d_inode(dentry));
711 if (d_really_is_positive(dentry)) {
720 dentry = ERR_PTR(err);
721 else if (dentry != req->r_dentry)
722 dentry = dget(req->r_dentry); /* we got spliced */
728 static bool is_root_ceph_dentry(struct inode *inode, struct dentry *dentry)
730 return ceph_ino(inode) == CEPH_INO_ROOT &&
731 strncmp(dentry->d_name.name, ".ceph", 5) == 0;
735 * Look up a single dir entry. If there is a lookup intent, inform
736 * the MDS so that it gets our 'caps wanted' value in a single op.
738 static struct dentry *ceph_lookup(struct inode *dir, struct dentry *dentry,
741 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
742 struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(dir->i_sb);
743 struct ceph_mds_request *req;
748 dout("lookup %p dentry %p '%pd'\n",
749 dir, dentry, dentry);
751 if (dentry->d_name.len > NAME_MAX)
752 return ERR_PTR(-ENAMETOOLONG);
754 /* can we conclude ENOENT locally? */
755 if (d_really_is_negative(dentry)) {
756 struct ceph_inode_info *ci = ceph_inode(dir);
757 struct ceph_dentry_info *di = ceph_dentry(dentry);
759 spin_lock(&ci->i_ceph_lock);
760 dout(" dir %p flags are 0x%lx\n", dir, ci->i_ceph_flags);
761 if (strncmp(dentry->d_name.name,
762 fsc->mount_options->snapdir_name,
763 dentry->d_name.len) &&
764 !is_root_ceph_dentry(dir, dentry) &&
765 ceph_test_mount_opt(fsc, DCACHE) &&
766 __ceph_dir_is_complete(ci) &&
767 __ceph_caps_issued_mask_metric(ci, CEPH_CAP_FILE_SHARED, 1)) {
768 __ceph_touch_fmode(ci, mdsc, CEPH_FILE_MODE_RD);
769 spin_unlock(&ci->i_ceph_lock);
770 dout(" dir %p complete, -ENOENT\n", dir);
772 di->lease_shared_gen = atomic_read(&ci->i_shared_gen);
775 spin_unlock(&ci->i_ceph_lock);
778 op = ceph_snap(dir) == CEPH_SNAPDIR ?
779 CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP;
780 req = ceph_mdsc_create_request(mdsc, op, USE_ANY_MDS);
782 return ERR_CAST(req);
783 req->r_dentry = dget(dentry);
786 mask = CEPH_STAT_CAP_INODE | CEPH_CAP_AUTH_SHARED;
787 if (ceph_security_xattr_wanted(dir))
788 mask |= CEPH_CAP_XATTR_SHARED;
789 req->r_args.getattr.mask = cpu_to_le32(mask);
792 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
793 err = ceph_mdsc_do_request(mdsc, NULL, req);
794 if (err == -ENOENT) {
797 res = ceph_handle_snapdir(req, dentry);
805 dentry = ceph_finish_lookup(req, dentry, err);
806 ceph_mdsc_put_request(req); /* will dput(dentry) */
807 dout("lookup result=%p\n", dentry);
812 * If we do a create but get no trace back from the MDS, follow up with
813 * a lookup (the VFS expects us to link up the provided dentry).
815 int ceph_handle_notrace_create(struct inode *dir, struct dentry *dentry)
817 struct dentry *result = ceph_lookup(dir, dentry, 0);
819 if (result && !IS_ERR(result)) {
821 * We created the item, then did a lookup, and found
822 * it was already linked to another inode we already
823 * had in our cache (and thus got spliced). To not
824 * confuse VFS (especially when inode is a directory),
825 * we don't link our dentry to that inode, return an
828 * This event should be rare and it happens only when
829 * we talk to old MDS. Recent MDS does not send traceless
830 * reply for request that creates new inode.
835 return PTR_ERR(result);
838 static int ceph_mknod(struct user_namespace *mnt_userns, struct inode *dir,
839 struct dentry *dentry, umode_t mode, dev_t rdev)
841 struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(dir->i_sb);
842 struct ceph_mds_request *req;
843 struct ceph_acl_sec_ctx as_ctx = {};
846 if (ceph_snap(dir) != CEPH_NOSNAP)
849 if (ceph_quota_is_max_files_exceeded(dir)) {
854 err = ceph_pre_init_acls(dir, &mode, &as_ctx);
857 err = ceph_security_init_secctx(dentry, mode, &as_ctx);
861 dout("mknod in dir %p dentry %p mode 0%ho rdev %d\n",
862 dir, dentry, mode, rdev);
863 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_MKNOD, USE_AUTH_MDS);
868 req->r_dentry = dget(dentry);
871 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
872 req->r_args.mknod.mode = cpu_to_le32(mode);
873 req->r_args.mknod.rdev = cpu_to_le32(rdev);
874 req->r_dentry_drop = CEPH_CAP_FILE_SHARED | CEPH_CAP_AUTH_EXCL;
875 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
876 if (as_ctx.pagelist) {
877 req->r_pagelist = as_ctx.pagelist;
878 as_ctx.pagelist = NULL;
880 err = ceph_mdsc_do_request(mdsc, dir, req);
881 if (!err && !req->r_reply_info.head->is_dentry)
882 err = ceph_handle_notrace_create(dir, dentry);
883 ceph_mdsc_put_request(req);
886 ceph_init_inode_acls(d_inode(dentry), &as_ctx);
889 ceph_release_acl_sec_ctx(&as_ctx);
893 static int ceph_create(struct user_namespace *mnt_userns, struct inode *dir,
894 struct dentry *dentry, umode_t mode, bool excl)
896 return ceph_mknod(mnt_userns, dir, dentry, mode, 0);
899 static int ceph_symlink(struct user_namespace *mnt_userns, struct inode *dir,
900 struct dentry *dentry, const char *dest)
902 struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(dir->i_sb);
903 struct ceph_mds_request *req;
904 struct ceph_acl_sec_ctx as_ctx = {};
907 if (ceph_snap(dir) != CEPH_NOSNAP)
910 if (ceph_quota_is_max_files_exceeded(dir)) {
915 err = ceph_security_init_secctx(dentry, S_IFLNK | 0777, &as_ctx);
919 dout("symlink in dir %p dentry %p to '%s'\n", dir, dentry, dest);
920 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SYMLINK, USE_AUTH_MDS);
925 req->r_path2 = kstrdup(dest, GFP_KERNEL);
928 ceph_mdsc_put_request(req);
932 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
933 req->r_dentry = dget(dentry);
935 req->r_dentry_drop = CEPH_CAP_FILE_SHARED | CEPH_CAP_AUTH_EXCL;
936 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
937 if (as_ctx.pagelist) {
938 req->r_pagelist = as_ctx.pagelist;
939 as_ctx.pagelist = NULL;
941 err = ceph_mdsc_do_request(mdsc, dir, req);
942 if (!err && !req->r_reply_info.head->is_dentry)
943 err = ceph_handle_notrace_create(dir, dentry);
944 ceph_mdsc_put_request(req);
948 ceph_release_acl_sec_ctx(&as_ctx);
952 static int ceph_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
953 struct dentry *dentry, umode_t mode)
955 struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(dir->i_sb);
956 struct ceph_mds_request *req;
957 struct ceph_acl_sec_ctx as_ctx = {};
961 if (ceph_snap(dir) == CEPH_SNAPDIR) {
962 /* mkdir .snap/foo is a MKSNAP */
963 op = CEPH_MDS_OP_MKSNAP;
964 dout("mksnap dir %p snap '%pd' dn %p\n", dir,
966 } else if (ceph_snap(dir) == CEPH_NOSNAP) {
967 dout("mkdir dir %p dn %p mode 0%ho\n", dir, dentry, mode);
968 op = CEPH_MDS_OP_MKDIR;
973 if (op == CEPH_MDS_OP_MKDIR &&
974 ceph_quota_is_max_files_exceeded(dir)) {
980 err = ceph_pre_init_acls(dir, &mode, &as_ctx);
983 err = ceph_security_init_secctx(dentry, mode, &as_ctx);
987 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
993 req->r_dentry = dget(dentry);
996 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
997 req->r_args.mkdir.mode = cpu_to_le32(mode);
998 req->r_dentry_drop = CEPH_CAP_FILE_SHARED | CEPH_CAP_AUTH_EXCL;
999 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
1000 if (as_ctx.pagelist) {
1001 req->r_pagelist = as_ctx.pagelist;
1002 as_ctx.pagelist = NULL;
1004 err = ceph_mdsc_do_request(mdsc, dir, req);
1006 !req->r_reply_info.head->is_target &&
1007 !req->r_reply_info.head->is_dentry)
1008 err = ceph_handle_notrace_create(dir, dentry);
1009 ceph_mdsc_put_request(req);
1012 ceph_init_inode_acls(d_inode(dentry), &as_ctx);
1015 ceph_release_acl_sec_ctx(&as_ctx);
1019 static int ceph_link(struct dentry *old_dentry, struct inode *dir,
1020 struct dentry *dentry)
1022 struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(dir->i_sb);
1023 struct ceph_mds_request *req;
1026 if (ceph_snap(dir) != CEPH_NOSNAP)
1029 dout("link in dir %p old_dentry %p dentry %p\n", dir,
1030 old_dentry, dentry);
1031 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LINK, USE_AUTH_MDS);
1034 return PTR_ERR(req);
1036 req->r_dentry = dget(dentry);
1037 req->r_num_caps = 2;
1038 req->r_old_dentry = dget(old_dentry);
1039 req->r_parent = dir;
1040 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
1041 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
1042 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
1043 /* release LINK_SHARED on source inode (mds will lock it) */
1044 req->r_old_inode_drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL;
1045 err = ceph_mdsc_do_request(mdsc, dir, req);
1048 } else if (!req->r_reply_info.head->is_dentry) {
1049 ihold(d_inode(old_dentry));
1050 d_instantiate(dentry, d_inode(old_dentry));
1052 ceph_mdsc_put_request(req);
1056 static void ceph_async_unlink_cb(struct ceph_mds_client *mdsc,
1057 struct ceph_mds_request *req)
1059 int result = req->r_err ? req->r_err :
1060 le32_to_cpu(req->r_reply_info.head->result);
1062 if (result == -EJUKEBOX)
1065 /* If op failed, mark everyone involved for errors */
1069 char *path = ceph_mdsc_build_path(req->r_dentry, &pathlen,
1072 /* mark error on parent + clear complete */
1073 mapping_set_error(req->r_parent->i_mapping, result);
1074 ceph_dir_clear_complete(req->r_parent);
1076 /* drop the dentry -- we don't know its status */
1077 if (!d_unhashed(req->r_dentry))
1078 d_drop(req->r_dentry);
1080 /* mark inode itself for an error (since metadata is bogus) */
1081 mapping_set_error(req->r_old_inode->i_mapping, result);
1083 pr_warn("ceph: async unlink failure path=(%llx)%s result=%d!\n",
1084 base, IS_ERR(path) ? "<<bad>>" : path, result);
1085 ceph_mdsc_free_path(path, pathlen);
1088 iput(req->r_old_inode);
1089 ceph_mdsc_release_dir_caps(req);
1092 static int get_caps_for_async_unlink(struct inode *dir, struct dentry *dentry)
1094 struct ceph_inode_info *ci = ceph_inode(dir);
1095 struct ceph_dentry_info *di;
1096 int got = 0, want = CEPH_CAP_FILE_EXCL | CEPH_CAP_DIR_UNLINK;
1098 spin_lock(&ci->i_ceph_lock);
1099 if ((__ceph_caps_issued(ci, NULL) & want) == want) {
1100 ceph_take_cap_refs(ci, want, false);
1103 spin_unlock(&ci->i_ceph_lock);
1105 /* If we didn't get anything, return 0 */
1109 spin_lock(&dentry->d_lock);
1110 di = ceph_dentry(dentry);
1112 * - We are holding Fx, which implies Fs caps.
1113 * - Only support async unlink for primary linkage
1115 if (atomic_read(&ci->i_shared_gen) != di->lease_shared_gen ||
1116 !(di->flags & CEPH_DENTRY_PRIMARY_LINK))
1118 spin_unlock(&dentry->d_lock);
1120 /* Do we still want what we've got? */
1124 ceph_put_cap_refs(ci, got);
1129 * rmdir and unlink are differ only by the metadata op code
1131 static int ceph_unlink(struct inode *dir, struct dentry *dentry)
1133 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
1134 struct ceph_mds_client *mdsc = fsc->mdsc;
1135 struct inode *inode = d_inode(dentry);
1136 struct ceph_mds_request *req;
1137 bool try_async = ceph_test_mount_opt(fsc, ASYNC_DIROPS);
1141 if (ceph_snap(dir) == CEPH_SNAPDIR) {
1142 /* rmdir .snap/foo is RMSNAP */
1143 dout("rmsnap dir %p '%pd' dn %p\n", dir, dentry, dentry);
1144 op = CEPH_MDS_OP_RMSNAP;
1145 } else if (ceph_snap(dir) == CEPH_NOSNAP) {
1146 dout("unlink/rmdir dir %p dn %p inode %p\n",
1147 dir, dentry, inode);
1148 op = d_is_dir(dentry) ?
1149 CEPH_MDS_OP_RMDIR : CEPH_MDS_OP_UNLINK;
1153 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
1158 req->r_dentry = dget(dentry);
1159 req->r_num_caps = 2;
1160 req->r_parent = dir;
1161 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
1162 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
1163 req->r_inode_drop = ceph_drop_caps_for_unlink(inode);
1165 if (try_async && op == CEPH_MDS_OP_UNLINK &&
1166 (req->r_dir_caps = get_caps_for_async_unlink(dir, dentry))) {
1167 dout("async unlink on %llu/%.*s caps=%s", ceph_ino(dir),
1168 dentry->d_name.len, dentry->d_name.name,
1169 ceph_cap_string(req->r_dir_caps));
1170 set_bit(CEPH_MDS_R_ASYNC, &req->r_req_flags);
1171 req->r_callback = ceph_async_unlink_cb;
1172 req->r_old_inode = d_inode(dentry);
1173 ihold(req->r_old_inode);
1174 err = ceph_mdsc_submit_request(mdsc, dir, req);
1177 * We have enough caps, so we assume that the unlink
1178 * will succeed. Fix up the target inode and dcache.
1182 } else if (err == -EJUKEBOX) {
1184 ceph_mdsc_put_request(req);
1188 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
1189 err = ceph_mdsc_do_request(mdsc, dir, req);
1190 if (!err && !req->r_reply_info.head->is_dentry)
1194 ceph_mdsc_put_request(req);
1199 static int ceph_rename(struct user_namespace *mnt_userns, struct inode *old_dir,
1200 struct dentry *old_dentry, struct inode *new_dir,
1201 struct dentry *new_dentry, unsigned int flags)
1203 struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(old_dir->i_sb);
1204 struct ceph_mds_request *req;
1205 int op = CEPH_MDS_OP_RENAME;
1211 if (ceph_snap(old_dir) != ceph_snap(new_dir))
1213 if (ceph_snap(old_dir) != CEPH_NOSNAP) {
1214 if (old_dir == new_dir && ceph_snap(old_dir) == CEPH_SNAPDIR)
1215 op = CEPH_MDS_OP_RENAMESNAP;
1219 /* don't allow cross-quota renames */
1220 if ((old_dir != new_dir) &&
1221 (!ceph_quota_is_same_realm(old_dir, new_dir)))
1224 dout("rename dir %p dentry %p to dir %p dentry %p\n",
1225 old_dir, old_dentry, new_dir, new_dentry);
1226 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
1228 return PTR_ERR(req);
1230 req->r_dentry = dget(new_dentry);
1231 req->r_num_caps = 2;
1232 req->r_old_dentry = dget(old_dentry);
1233 req->r_old_dentry_dir = old_dir;
1234 req->r_parent = new_dir;
1235 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
1236 req->r_old_dentry_drop = CEPH_CAP_FILE_SHARED;
1237 req->r_old_dentry_unless = CEPH_CAP_FILE_EXCL;
1238 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
1239 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
1240 /* release LINK_RDCACHE on source inode (mds will lock it) */
1241 req->r_old_inode_drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL;
1242 if (d_really_is_positive(new_dentry)) {
1244 ceph_drop_caps_for_unlink(d_inode(new_dentry));
1246 err = ceph_mdsc_do_request(mdsc, old_dir, req);
1247 if (!err && !req->r_reply_info.head->is_dentry) {
1249 * Normally d_move() is done by fill_trace (called by
1250 * do_request, above). If there is no trace, we need
1253 d_move(old_dentry, new_dentry);
1255 ceph_mdsc_put_request(req);
1260 * Move dentry to tail of mdsc->dentry_leases list when lease is updated.
1261 * Leases at front of the list will expire first. (Assume all leases have
1264 * Called under dentry->d_lock.
1266 void __ceph_dentry_lease_touch(struct ceph_dentry_info *di)
1268 struct dentry *dn = di->dentry;
1269 struct ceph_mds_client *mdsc;
1271 dout("dentry_lease_touch %p %p '%pd'\n", di, dn, dn);
1273 di->flags |= CEPH_DENTRY_LEASE_LIST;
1274 if (di->flags & CEPH_DENTRY_SHRINK_LIST) {
1275 di->flags |= CEPH_DENTRY_REFERENCED;
1279 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1280 spin_lock(&mdsc->dentry_list_lock);
1281 list_move_tail(&di->lease_list, &mdsc->dentry_leases);
1282 spin_unlock(&mdsc->dentry_list_lock);
1285 static void __dentry_dir_lease_touch(struct ceph_mds_client* mdsc,
1286 struct ceph_dentry_info *di)
1288 di->flags &= ~(CEPH_DENTRY_LEASE_LIST | CEPH_DENTRY_REFERENCED);
1291 list_move_tail(&di->lease_list, &mdsc->dentry_dir_leases);
1295 * When dir lease is used, add dentry to tail of mdsc->dentry_dir_leases
1296 * list if it's not in the list, otherwise set 'referenced' flag.
1298 * Called under dentry->d_lock.
1300 void __ceph_dentry_dir_lease_touch(struct ceph_dentry_info *di)
1302 struct dentry *dn = di->dentry;
1303 struct ceph_mds_client *mdsc;
1305 dout("dentry_dir_lease_touch %p %p '%pd' (offset 0x%llx)\n",
1306 di, dn, dn, di->offset);
1308 if (!list_empty(&di->lease_list)) {
1309 if (di->flags & CEPH_DENTRY_LEASE_LIST) {
1310 /* don't remove dentry from dentry lease list
1311 * if its lease is valid */
1312 if (__dentry_lease_is_valid(di))
1315 di->flags |= CEPH_DENTRY_REFERENCED;
1320 if (di->flags & CEPH_DENTRY_SHRINK_LIST) {
1321 di->flags |= CEPH_DENTRY_REFERENCED;
1322 di->flags &= ~CEPH_DENTRY_LEASE_LIST;
1326 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1327 spin_lock(&mdsc->dentry_list_lock);
1328 __dentry_dir_lease_touch(mdsc, di),
1329 spin_unlock(&mdsc->dentry_list_lock);
1332 static void __dentry_lease_unlist(struct ceph_dentry_info *di)
1334 struct ceph_mds_client *mdsc;
1335 if (di->flags & CEPH_DENTRY_SHRINK_LIST)
1337 if (list_empty(&di->lease_list))
1340 mdsc = ceph_sb_to_client(di->dentry->d_sb)->mdsc;
1341 spin_lock(&mdsc->dentry_list_lock);
1342 list_del_init(&di->lease_list);
1343 spin_unlock(&mdsc->dentry_list_lock);
1353 struct ceph_lease_walk_control {
1355 bool expire_dir_lease;
1356 unsigned long nr_to_scan;
1357 unsigned long dir_lease_ttl;
1360 static unsigned long
1361 __dentry_leases_walk(struct ceph_mds_client *mdsc,
1362 struct ceph_lease_walk_control *lwc,
1363 int (*check)(struct dentry*, void*))
1365 struct ceph_dentry_info *di, *tmp;
1366 struct dentry *dentry, *last = NULL;
1367 struct list_head* list;
1369 unsigned long freed = 0;
1372 list = lwc->dir_lease ? &mdsc->dentry_dir_leases : &mdsc->dentry_leases;
1373 spin_lock(&mdsc->dentry_list_lock);
1374 list_for_each_entry_safe(di, tmp, list, lease_list) {
1375 if (!lwc->nr_to_scan)
1379 dentry = di->dentry;
1383 if (!spin_trylock(&dentry->d_lock))
1386 if (__lockref_is_dead(&dentry->d_lockref)) {
1387 list_del_init(&di->lease_list);
1391 ret = check(dentry, lwc);
1393 /* move it into tail of dir lease list */
1394 __dentry_dir_lease_touch(mdsc, di);
1400 di->flags &= ~CEPH_DENTRY_REFERENCED;
1401 if (dentry->d_lockref.count > 0) {
1402 /* update_dentry_lease() will re-add
1403 * it to lease list, or
1404 * ceph_d_delete() will return 1 when
1405 * last reference is dropped */
1406 list_del_init(&di->lease_list);
1408 di->flags |= CEPH_DENTRY_SHRINK_LIST;
1409 list_move_tail(&di->lease_list, &dispose);
1414 spin_unlock(&dentry->d_lock);
1418 spin_unlock(&mdsc->dentry_list_lock);
1420 while (!list_empty(&dispose)) {
1421 di = list_first_entry(&dispose, struct ceph_dentry_info,
1423 dentry = di->dentry;
1424 spin_lock(&dentry->d_lock);
1426 list_del_init(&di->lease_list);
1427 di->flags &= ~CEPH_DENTRY_SHRINK_LIST;
1428 if (di->flags & CEPH_DENTRY_REFERENCED) {
1429 spin_lock(&mdsc->dentry_list_lock);
1430 if (di->flags & CEPH_DENTRY_LEASE_LIST) {
1431 list_add_tail(&di->lease_list,
1432 &mdsc->dentry_leases);
1434 __dentry_dir_lease_touch(mdsc, di);
1436 spin_unlock(&mdsc->dentry_list_lock);
1441 spin_unlock(&dentry->d_lock);
1442 /* ceph_d_delete() does the trick */
1448 static int __dentry_lease_check(struct dentry *dentry, void *arg)
1450 struct ceph_dentry_info *di = ceph_dentry(dentry);
1453 if (__dentry_lease_is_valid(di))
1455 ret = __dir_lease_try_check(dentry);
1463 static int __dir_lease_check(struct dentry *dentry, void *arg)
1465 struct ceph_lease_walk_control *lwc = arg;
1466 struct ceph_dentry_info *di = ceph_dentry(dentry);
1468 int ret = __dir_lease_try_check(dentry);
1472 if (time_before(jiffies, di->time + lwc->dir_lease_ttl))
1474 /* Move dentry to tail of dir lease list if we don't want
1475 * to delete it. So dentries in the list are checked in a
1476 * round robin manner */
1477 if (!lwc->expire_dir_lease)
1479 if (dentry->d_lockref.count > 0 ||
1480 (di->flags & CEPH_DENTRY_REFERENCED))
1482 /* invalidate dir lease */
1483 di->lease_shared_gen = 0;
1488 int ceph_trim_dentries(struct ceph_mds_client *mdsc)
1490 struct ceph_lease_walk_control lwc;
1491 unsigned long count;
1492 unsigned long freed;
1494 spin_lock(&mdsc->caps_list_lock);
1495 if (mdsc->caps_use_max > 0 &&
1496 mdsc->caps_use_count > mdsc->caps_use_max)
1497 count = mdsc->caps_use_count - mdsc->caps_use_max;
1500 spin_unlock(&mdsc->caps_list_lock);
1502 lwc.dir_lease = false;
1503 lwc.nr_to_scan = CEPH_CAPS_PER_RELEASE * 2;
1504 freed = __dentry_leases_walk(mdsc, &lwc, __dentry_lease_check);
1505 if (!lwc.nr_to_scan) /* more invalid leases */
1508 if (lwc.nr_to_scan < CEPH_CAPS_PER_RELEASE)
1509 lwc.nr_to_scan = CEPH_CAPS_PER_RELEASE;
1511 lwc.dir_lease = true;
1512 lwc.expire_dir_lease = freed < count;
1513 lwc.dir_lease_ttl = mdsc->fsc->mount_options->caps_wanted_delay_max * HZ;
1514 freed +=__dentry_leases_walk(mdsc, &lwc, __dir_lease_check);
1515 if (!lwc.nr_to_scan) /* more to check */
1518 return freed > 0 ? 1 : 0;
1522 * Ensure a dentry lease will no longer revalidate.
1524 void ceph_invalidate_dentry_lease(struct dentry *dentry)
1526 struct ceph_dentry_info *di = ceph_dentry(dentry);
1527 spin_lock(&dentry->d_lock);
1529 di->lease_shared_gen = 0;
1530 di->flags &= ~CEPH_DENTRY_PRIMARY_LINK;
1531 __dentry_lease_unlist(di);
1532 spin_unlock(&dentry->d_lock);
1536 * Check if dentry lease is valid. If not, delete the lease. Try to
1537 * renew if the least is more than half up.
1539 static bool __dentry_lease_is_valid(struct ceph_dentry_info *di)
1541 struct ceph_mds_session *session;
1546 session = di->lease_session;
1551 spin_lock(&session->s_gen_ttl_lock);
1552 gen = session->s_cap_gen;
1553 ttl = session->s_cap_ttl;
1554 spin_unlock(&session->s_gen_ttl_lock);
1556 if (di->lease_gen == gen &&
1557 time_before(jiffies, ttl) &&
1558 time_before(jiffies, di->time))
1565 static int dentry_lease_is_valid(struct dentry *dentry, unsigned int flags)
1567 struct ceph_dentry_info *di;
1568 struct ceph_mds_session *session = NULL;
1572 spin_lock(&dentry->d_lock);
1573 di = ceph_dentry(dentry);
1574 if (di && __dentry_lease_is_valid(di)) {
1577 if (di->lease_renew_after &&
1578 time_after(jiffies, di->lease_renew_after)) {
1580 * We should renew. If we're in RCU walk mode
1581 * though, we can't do that so just return
1584 if (flags & LOOKUP_RCU) {
1587 session = ceph_get_mds_session(di->lease_session);
1588 seq = di->lease_seq;
1589 di->lease_renew_after = 0;
1590 di->lease_renew_from = jiffies;
1594 spin_unlock(&dentry->d_lock);
1597 ceph_mdsc_lease_send_msg(session, dentry,
1598 CEPH_MDS_LEASE_RENEW, seq);
1599 ceph_put_mds_session(session);
1601 dout("dentry_lease_is_valid - dentry %p = %d\n", dentry, valid);
1606 * Called under dentry->d_lock.
1608 static int __dir_lease_try_check(const struct dentry *dentry)
1610 struct ceph_dentry_info *di = ceph_dentry(dentry);
1612 struct ceph_inode_info *ci;
1615 if (!di->lease_shared_gen)
1617 if (IS_ROOT(dentry))
1620 dir = d_inode(dentry->d_parent);
1621 ci = ceph_inode(dir);
1623 if (spin_trylock(&ci->i_ceph_lock)) {
1624 if (atomic_read(&ci->i_shared_gen) == di->lease_shared_gen &&
1625 __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 0))
1627 spin_unlock(&ci->i_ceph_lock);
1633 di->lease_shared_gen = 0;
1638 * Check if directory-wide content lease/cap is valid.
1640 static int dir_lease_is_valid(struct inode *dir, struct dentry *dentry,
1641 struct ceph_mds_client *mdsc)
1643 struct ceph_inode_info *ci = ceph_inode(dir);
1647 spin_lock(&ci->i_ceph_lock);
1648 valid = __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1);
1650 __ceph_touch_fmode(ci, mdsc, CEPH_FILE_MODE_RD);
1651 shared_gen = atomic_read(&ci->i_shared_gen);
1653 spin_unlock(&ci->i_ceph_lock);
1655 struct ceph_dentry_info *di;
1656 spin_lock(&dentry->d_lock);
1657 di = ceph_dentry(dentry);
1658 if (dir == d_inode(dentry->d_parent) &&
1659 di && di->lease_shared_gen == shared_gen)
1660 __ceph_dentry_dir_lease_touch(di);
1663 spin_unlock(&dentry->d_lock);
1665 dout("dir_lease_is_valid dir %p v%u dentry %p = %d\n",
1666 dir, (unsigned)atomic_read(&ci->i_shared_gen), dentry, valid);
1671 * Check if cached dentry can be trusted.
1673 static int ceph_d_revalidate(struct dentry *dentry, unsigned int flags)
1676 struct dentry *parent;
1677 struct inode *dir, *inode;
1678 struct ceph_mds_client *mdsc;
1680 if (flags & LOOKUP_RCU) {
1681 parent = READ_ONCE(dentry->d_parent);
1682 dir = d_inode_rcu(parent);
1685 inode = d_inode_rcu(dentry);
1687 parent = dget_parent(dentry);
1688 dir = d_inode(parent);
1689 inode = d_inode(dentry);
1692 dout("d_revalidate %p '%pd' inode %p offset 0x%llx\n", dentry,
1693 dentry, inode, ceph_dentry(dentry)->offset);
1695 mdsc = ceph_sb_to_client(dir->i_sb)->mdsc;
1697 /* always trust cached snapped dentries, snapdir dentry */
1698 if (ceph_snap(dir) != CEPH_NOSNAP) {
1699 dout("d_revalidate %p '%pd' inode %p is SNAPPED\n", dentry,
1702 } else if (inode && ceph_snap(inode) == CEPH_SNAPDIR) {
1705 valid = dentry_lease_is_valid(dentry, flags);
1706 if (valid == -ECHILD)
1708 if (valid || dir_lease_is_valid(dir, dentry, mdsc)) {
1710 valid = ceph_is_any_caps(inode);
1717 struct ceph_mds_request *req;
1721 if (flags & LOOKUP_RCU)
1724 percpu_counter_inc(&mdsc->metric.d_lease_mis);
1726 op = ceph_snap(dir) == CEPH_SNAPDIR ?
1727 CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP;
1728 req = ceph_mdsc_create_request(mdsc, op, USE_ANY_MDS);
1730 req->r_dentry = dget(dentry);
1731 req->r_num_caps = 2;
1732 req->r_parent = dir;
1734 mask = CEPH_STAT_CAP_INODE | CEPH_CAP_AUTH_SHARED;
1735 if (ceph_security_xattr_wanted(dir))
1736 mask |= CEPH_CAP_XATTR_SHARED;
1737 req->r_args.getattr.mask = cpu_to_le32(mask);
1739 err = ceph_mdsc_do_request(mdsc, NULL, req);
1742 if (d_really_is_positive(dentry) &&
1743 d_inode(dentry) == req->r_target_inode)
1747 if (d_really_is_negative(dentry))
1753 ceph_mdsc_put_request(req);
1754 dout("d_revalidate %p lookup result=%d\n",
1758 percpu_counter_inc(&mdsc->metric.d_lease_hit);
1761 dout("d_revalidate %p %s\n", dentry, valid ? "valid" : "invalid");
1763 ceph_dir_clear_complete(dir);
1765 if (!(flags & LOOKUP_RCU))
1771 * Delete unused dentry that doesn't have valid lease
1773 * Called under dentry->d_lock.
1775 static int ceph_d_delete(const struct dentry *dentry)
1777 struct ceph_dentry_info *di;
1779 /* won't release caps */
1780 if (d_really_is_negative(dentry))
1782 if (ceph_snap(d_inode(dentry)) != CEPH_NOSNAP)
1785 di = ceph_dentry(dentry);
1787 if (__dentry_lease_is_valid(di))
1789 if (__dir_lease_try_check(dentry))
1796 * Release our ceph_dentry_info.
1798 static void ceph_d_release(struct dentry *dentry)
1800 struct ceph_dentry_info *di = ceph_dentry(dentry);
1801 struct ceph_fs_client *fsc = ceph_sb_to_client(dentry->d_sb);
1803 dout("d_release %p\n", dentry);
1805 atomic64_dec(&fsc->mdsc->metric.total_dentries);
1807 spin_lock(&dentry->d_lock);
1808 __dentry_lease_unlist(di);
1809 dentry->d_fsdata = NULL;
1810 spin_unlock(&dentry->d_lock);
1812 if (di->lease_session)
1813 ceph_put_mds_session(di->lease_session);
1814 kmem_cache_free(ceph_dentry_cachep, di);
1818 * When the VFS prunes a dentry from the cache, we need to clear the
1819 * complete flag on the parent directory.
1821 * Called under dentry->d_lock.
1823 static void ceph_d_prune(struct dentry *dentry)
1825 struct ceph_inode_info *dir_ci;
1826 struct ceph_dentry_info *di;
1828 dout("ceph_d_prune %pd %p\n", dentry, dentry);
1830 /* do we have a valid parent? */
1831 if (IS_ROOT(dentry))
1834 /* we hold d_lock, so d_parent is stable */
1835 dir_ci = ceph_inode(d_inode(dentry->d_parent));
1836 if (dir_ci->i_vino.snap == CEPH_SNAPDIR)
1839 /* who calls d_delete() should also disable dcache readdir */
1840 if (d_really_is_negative(dentry))
1843 /* d_fsdata does not get cleared until d_release */
1844 if (!d_unhashed(dentry)) {
1845 __ceph_dir_clear_complete(dir_ci);
1849 /* Disable dcache readdir just in case that someone called d_drop()
1850 * or d_invalidate(), but MDS didn't revoke CEPH_CAP_FILE_SHARED
1851 * properly (dcache readdir is still enabled) */
1852 di = ceph_dentry(dentry);
1853 if (di->offset > 0 &&
1854 di->lease_shared_gen == atomic_read(&dir_ci->i_shared_gen))
1855 __ceph_dir_clear_ordered(dir_ci);
1859 * read() on a dir. This weird interface hack only works if mounted
1860 * with '-o dirstat'.
1862 static ssize_t ceph_read_dir(struct file *file, char __user *buf, size_t size,
1865 struct ceph_dir_file_info *dfi = file->private_data;
1866 struct inode *inode = file_inode(file);
1867 struct ceph_inode_info *ci = ceph_inode(inode);
1869 const int bufsize = 1024;
1871 if (!ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb), DIRSTAT))
1874 if (!dfi->dir_info) {
1875 dfi->dir_info = kmalloc(bufsize, GFP_KERNEL);
1879 snprintf(dfi->dir_info, bufsize,
1882 " subdirs: %20lld\n"
1883 "rentries: %20lld\n"
1885 " rsubdirs: %20lld\n"
1887 "rctime: %10lld.%09ld\n",
1888 ci->i_files + ci->i_subdirs,
1891 ci->i_rfiles + ci->i_rsubdirs,
1895 ci->i_rctime.tv_sec,
1896 ci->i_rctime.tv_nsec);
1899 if (*ppos >= dfi->dir_info_len)
1901 size = min_t(unsigned, size, dfi->dir_info_len-*ppos);
1902 left = copy_to_user(buf, dfi->dir_info + *ppos, size);
1905 *ppos += (size - left);
1912 * Return name hash for a given dentry. This is dependent on
1913 * the parent directory's hash function.
1915 unsigned ceph_dentry_hash(struct inode *dir, struct dentry *dn)
1917 struct ceph_inode_info *dci = ceph_inode(dir);
1920 switch (dci->i_dir_layout.dl_dir_hash) {
1921 case 0: /* for backward compat */
1922 case CEPH_STR_HASH_LINUX:
1923 return dn->d_name.hash;
1926 spin_lock(&dn->d_lock);
1927 hash = ceph_str_hash(dci->i_dir_layout.dl_dir_hash,
1928 dn->d_name.name, dn->d_name.len);
1929 spin_unlock(&dn->d_lock);
1934 const struct file_operations ceph_dir_fops = {
1935 .read = ceph_read_dir,
1936 .iterate = ceph_readdir,
1937 .llseek = ceph_dir_llseek,
1939 .release = ceph_release,
1940 .unlocked_ioctl = ceph_ioctl,
1941 .compat_ioctl = compat_ptr_ioctl,
1942 .fsync = ceph_fsync,
1944 .flock = ceph_flock,
1947 const struct file_operations ceph_snapdir_fops = {
1948 .iterate = ceph_readdir,
1949 .llseek = ceph_dir_llseek,
1951 .release = ceph_release,
1954 const struct inode_operations ceph_dir_iops = {
1955 .lookup = ceph_lookup,
1956 .permission = ceph_permission,
1957 .getattr = ceph_getattr,
1958 .setattr = ceph_setattr,
1959 .listxattr = ceph_listxattr,
1960 .get_acl = ceph_get_acl,
1961 .set_acl = ceph_set_acl,
1962 .mknod = ceph_mknod,
1963 .symlink = ceph_symlink,
1964 .mkdir = ceph_mkdir,
1966 .unlink = ceph_unlink,
1967 .rmdir = ceph_unlink,
1968 .rename = ceph_rename,
1969 .create = ceph_create,
1970 .atomic_open = ceph_atomic_open,
1973 const struct inode_operations ceph_snapdir_iops = {
1974 .lookup = ceph_lookup,
1975 .permission = ceph_permission,
1976 .getattr = ceph_getattr,
1977 .mkdir = ceph_mkdir,
1978 .rmdir = ceph_unlink,
1979 .rename = ceph_rename,
1982 const struct dentry_operations ceph_dentry_ops = {
1983 .d_revalidate = ceph_d_revalidate,
1984 .d_delete = ceph_d_delete,
1985 .d_release = ceph_d_release,
1986 .d_prune = ceph_d_prune,
1987 .d_init = ceph_d_init,