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;
33 * Initialize ceph dentry state.
35 static int ceph_d_init(struct dentry *dentry)
37 struct ceph_dentry_info *di;
39 di = kmem_cache_zalloc(ceph_dentry_cachep, GFP_KERNEL);
41 return -ENOMEM; /* oh well */
44 di->lease_session = NULL;
46 dentry->d_fsdata = di;
47 ceph_dentry_lru_add(dentry);
52 * for f_pos for readdir:
54 * (0xff << 52) | ((24 bits hash) << 28) |
55 * (the nth entry has hash collision);
57 * ((frag value) << 28) | (the nth entry in frag);
59 #define OFFSET_BITS 28
60 #define OFFSET_MASK ((1 << OFFSET_BITS) - 1)
61 #define HASH_ORDER (0xffull << (OFFSET_BITS + 24))
62 loff_t ceph_make_fpos(unsigned high, unsigned off, bool hash_order)
64 loff_t fpos = ((loff_t)high << 28) | (loff_t)off;
70 static bool is_hash_order(loff_t p)
72 return (p & HASH_ORDER) == HASH_ORDER;
75 static unsigned fpos_frag(loff_t p)
77 return p >> OFFSET_BITS;
80 static unsigned fpos_hash(loff_t p)
82 return ceph_frag_value(fpos_frag(p));
85 static unsigned fpos_off(loff_t p)
87 return p & OFFSET_MASK;
90 static int fpos_cmp(loff_t l, loff_t r)
92 int v = ceph_frag_compare(fpos_frag(l), fpos_frag(r));
95 return (int)(fpos_off(l) - fpos_off(r));
99 * make note of the last dentry we read, so we can
100 * continue at the same lexicographical point,
101 * regardless of what dir changes take place on the
104 static int note_last_dentry(struct ceph_dir_file_info *dfi, const char *name,
105 int len, unsigned next_offset)
107 char *buf = kmalloc(len+1, GFP_KERNEL);
110 kfree(dfi->last_name);
111 dfi->last_name = buf;
112 memcpy(dfi->last_name, name, len);
113 dfi->last_name[len] = 0;
114 dfi->next_offset = next_offset;
115 dout("note_last_dentry '%s'\n", dfi->last_name);
120 static struct dentry *
121 __dcache_find_get_entry(struct dentry *parent, u64 idx,
122 struct ceph_readdir_cache_control *cache_ctl)
124 struct inode *dir = d_inode(parent);
125 struct dentry *dentry;
126 unsigned idx_mask = (PAGE_SIZE / sizeof(struct dentry *)) - 1;
127 loff_t ptr_pos = idx * sizeof(struct dentry *);
128 pgoff_t ptr_pgoff = ptr_pos >> PAGE_SHIFT;
130 if (ptr_pos >= i_size_read(dir))
133 if (!cache_ctl->page || ptr_pgoff != page_index(cache_ctl->page)) {
134 ceph_readdir_cache_release(cache_ctl);
135 cache_ctl->page = find_lock_page(&dir->i_data, ptr_pgoff);
136 if (!cache_ctl->page) {
137 dout(" page %lu not found\n", ptr_pgoff);
138 return ERR_PTR(-EAGAIN);
140 /* reading/filling the cache are serialized by
141 i_mutex, no need to use page lock */
142 unlock_page(cache_ctl->page);
143 cache_ctl->dentries = kmap(cache_ctl->page);
146 cache_ctl->index = idx & idx_mask;
149 spin_lock(&parent->d_lock);
150 /* check i_size again here, because empty directory can be
151 * marked as complete while not holding the i_mutex. */
152 if (ceph_dir_is_complete_ordered(dir) && ptr_pos < i_size_read(dir))
153 dentry = cache_ctl->dentries[cache_ctl->index];
156 spin_unlock(&parent->d_lock);
157 if (dentry && !lockref_get_not_dead(&dentry->d_lockref))
160 return dentry ? : ERR_PTR(-EAGAIN);
164 * When possible, we try to satisfy a readdir by peeking at the
165 * dcache. We make this work by carefully ordering dentries on
166 * d_child when we initially get results back from the MDS, and
167 * falling back to a "normal" sync readdir if any dentries in the dir
170 * Complete dir indicates that we have all dentries in the dir. It is
171 * defined IFF we hold CEPH_CAP_FILE_SHARED (which will be revoked by
172 * the MDS if/when the directory is modified).
174 static int __dcache_readdir(struct file *file, struct dir_context *ctx,
177 struct ceph_dir_file_info *dfi = file->private_data;
178 struct dentry *parent = file->f_path.dentry;
179 struct inode *dir = d_inode(parent);
180 struct dentry *dentry, *last = NULL;
181 struct ceph_dentry_info *di;
182 struct ceph_readdir_cache_control cache_ctl = {};
186 dout("__dcache_readdir %p v%u at %llx\n", dir, (unsigned)shared_gen, ctx->pos);
188 /* search start position */
190 u64 count = div_u64(i_size_read(dir), sizeof(struct dentry *));
192 u64 step = count >> 1;
193 dentry = __dcache_find_get_entry(parent, idx + step,
196 /* use linar search */
200 if (IS_ERR(dentry)) {
201 err = PTR_ERR(dentry);
204 di = ceph_dentry(dentry);
205 spin_lock(&dentry->d_lock);
206 if (fpos_cmp(di->offset, ctx->pos) < 0) {
212 spin_unlock(&dentry->d_lock);
216 dout("__dcache_readdir %p cache idx %llu\n", dir, idx);
221 bool emit_dentry = false;
222 dentry = __dcache_find_get_entry(parent, idx++, &cache_ctl);
224 dfi->file_info.flags |= CEPH_F_ATEND;
228 if (IS_ERR(dentry)) {
229 err = PTR_ERR(dentry);
233 spin_lock(&dentry->d_lock);
234 di = ceph_dentry(dentry);
235 if (d_unhashed(dentry) ||
236 d_really_is_negative(dentry) ||
237 di->lease_shared_gen != shared_gen) {
238 spin_unlock(&dentry->d_lock);
243 if (fpos_cmp(ctx->pos, di->offset) <= 0) {
246 spin_unlock(&dentry->d_lock);
249 dout(" %llx dentry %p %pd %p\n", di->offset,
250 dentry, dentry, d_inode(dentry));
251 ctx->pos = di->offset;
252 if (!dir_emit(ctx, dentry->d_name.name,
254 ceph_translate_ino(dentry->d_sb,
255 d_inode(dentry)->i_ino),
256 d_inode(dentry)->i_mode >> 12)) {
271 ceph_readdir_cache_release(&cache_ctl);
274 di = ceph_dentry(last);
275 ret = note_last_dentry(dfi, last->d_name.name, last->d_name.len,
276 fpos_off(di->offset) + 1);
280 /* last_name no longer match cache index */
281 if (dfi->readdir_cache_idx >= 0) {
282 dfi->readdir_cache_idx = -1;
283 dfi->dir_release_count = 0;
289 static bool need_send_readdir(struct ceph_dir_file_info *dfi, loff_t pos)
291 if (!dfi->last_readdir)
293 if (is_hash_order(pos))
294 return !ceph_frag_contains_value(dfi->frag, fpos_hash(pos));
296 return dfi->frag != fpos_frag(pos);
299 static int ceph_readdir(struct file *file, struct dir_context *ctx)
301 struct ceph_dir_file_info *dfi = file->private_data;
302 struct inode *inode = file_inode(file);
303 struct ceph_inode_info *ci = ceph_inode(inode);
304 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
305 struct ceph_mds_client *mdsc = fsc->mdsc;
309 struct ceph_mds_reply_info_parsed *rinfo;
311 dout("readdir %p file %p pos %llx\n", inode, file, ctx->pos);
312 if (dfi->file_info.flags & CEPH_F_ATEND)
315 /* always start with . and .. */
317 dout("readdir off 0 -> '.'\n");
318 if (!dir_emit(ctx, ".", 1,
319 ceph_translate_ino(inode->i_sb, inode->i_ino),
320 inode->i_mode >> 12))
325 ino_t ino = parent_ino(file->f_path.dentry);
326 dout("readdir off 1 -> '..'\n");
327 if (!dir_emit(ctx, "..", 2,
328 ceph_translate_ino(inode->i_sb, ino),
329 inode->i_mode >> 12))
334 /* can we use the dcache? */
335 spin_lock(&ci->i_ceph_lock);
336 if (ceph_test_mount_opt(fsc, DCACHE) &&
337 !ceph_test_mount_opt(fsc, NOASYNCREADDIR) &&
338 ceph_snap(inode) != CEPH_SNAPDIR &&
339 __ceph_dir_is_complete_ordered(ci) &&
340 __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1)) {
341 int shared_gen = atomic_read(&ci->i_shared_gen);
342 spin_unlock(&ci->i_ceph_lock);
343 err = __dcache_readdir(file, ctx, shared_gen);
347 spin_unlock(&ci->i_ceph_lock);
350 /* proceed with a normal readdir */
352 /* do we have the correct frag content buffered? */
353 if (need_send_readdir(dfi, ctx->pos)) {
354 struct ceph_mds_request *req;
355 int op = ceph_snap(inode) == CEPH_SNAPDIR ?
356 CEPH_MDS_OP_LSSNAP : CEPH_MDS_OP_READDIR;
358 /* discard old result, if any */
359 if (dfi->last_readdir) {
360 ceph_mdsc_put_request(dfi->last_readdir);
361 dfi->last_readdir = NULL;
364 if (is_hash_order(ctx->pos)) {
365 /* fragtree isn't always accurate. choose frag
366 * based on previous reply when possible. */
367 if (frag == (unsigned)-1)
368 frag = ceph_choose_frag(ci, fpos_hash(ctx->pos),
371 frag = fpos_frag(ctx->pos);
374 dout("readdir fetching %llx.%llx frag %x offset '%s'\n",
375 ceph_vinop(inode), frag, dfi->last_name);
376 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
379 err = ceph_alloc_readdir_reply_buffer(req, inode);
381 ceph_mdsc_put_request(req);
384 /* hints to request -> mds selection code */
385 req->r_direct_mode = USE_AUTH_MDS;
386 if (op == CEPH_MDS_OP_READDIR) {
387 req->r_direct_hash = ceph_frag_value(frag);
388 __set_bit(CEPH_MDS_R_DIRECT_IS_HASH, &req->r_req_flags);
389 req->r_inode_drop = CEPH_CAP_FILE_EXCL;
391 if (dfi->last_name) {
392 req->r_path2 = kstrdup(dfi->last_name, GFP_KERNEL);
394 ceph_mdsc_put_request(req);
397 } else if (is_hash_order(ctx->pos)) {
398 req->r_args.readdir.offset_hash =
399 cpu_to_le32(fpos_hash(ctx->pos));
402 req->r_dir_release_cnt = dfi->dir_release_count;
403 req->r_dir_ordered_cnt = dfi->dir_ordered_count;
404 req->r_readdir_cache_idx = dfi->readdir_cache_idx;
405 req->r_readdir_offset = dfi->next_offset;
406 req->r_args.readdir.frag = cpu_to_le32(frag);
407 req->r_args.readdir.flags =
408 cpu_to_le16(CEPH_READDIR_REPLY_BITFLAGS);
410 req->r_inode = inode;
412 req->r_dentry = dget(file->f_path.dentry);
413 err = ceph_mdsc_do_request(mdsc, NULL, req);
415 ceph_mdsc_put_request(req);
418 dout("readdir got and parsed readdir result=%d on "
419 "frag %x, end=%d, complete=%d, hash_order=%d\n",
421 (int)req->r_reply_info.dir_end,
422 (int)req->r_reply_info.dir_complete,
423 (int)req->r_reply_info.hash_order);
425 rinfo = &req->r_reply_info;
426 if (le32_to_cpu(rinfo->dir_dir->frag) != frag) {
427 frag = le32_to_cpu(rinfo->dir_dir->frag);
428 if (!rinfo->hash_order) {
429 dfi->next_offset = req->r_readdir_offset;
430 /* adjust ctx->pos to beginning of frag */
431 ctx->pos = ceph_make_fpos(frag,
438 dfi->last_readdir = req;
440 if (test_bit(CEPH_MDS_R_DID_PREPOPULATE, &req->r_req_flags)) {
441 dfi->readdir_cache_idx = req->r_readdir_cache_idx;
442 if (dfi->readdir_cache_idx < 0) {
443 /* preclude from marking dir ordered */
444 dfi->dir_ordered_count = 0;
445 } else if (ceph_frag_is_leftmost(frag) &&
446 dfi->next_offset == 2) {
447 /* note dir version at start of readdir so
448 * we can tell if any dentries get dropped */
449 dfi->dir_release_count = req->r_dir_release_cnt;
450 dfi->dir_ordered_count = req->r_dir_ordered_cnt;
453 dout("readdir !did_prepopulate\n");
454 /* disable readdir cache */
455 dfi->readdir_cache_idx = -1;
456 /* preclude from marking dir complete */
457 dfi->dir_release_count = 0;
460 /* note next offset and last dentry name */
461 if (rinfo->dir_nr > 0) {
462 struct ceph_mds_reply_dir_entry *rde =
463 rinfo->dir_entries + (rinfo->dir_nr-1);
464 unsigned next_offset = req->r_reply_info.dir_end ?
465 2 : (fpos_off(rde->offset) + 1);
466 err = note_last_dentry(dfi, rde->name, rde->name_len,
470 } else if (req->r_reply_info.dir_end) {
471 dfi->next_offset = 2;
476 rinfo = &dfi->last_readdir->r_reply_info;
477 dout("readdir frag %x num %d pos %llx chunk first %llx\n",
478 dfi->frag, rinfo->dir_nr, ctx->pos,
479 rinfo->dir_nr ? rinfo->dir_entries[0].offset : 0LL);
482 /* search start position */
483 if (rinfo->dir_nr > 0) {
484 int step, nr = rinfo->dir_nr;
487 if (rinfo->dir_entries[i + step].offset < ctx->pos) {
495 for (; i < rinfo->dir_nr; i++) {
496 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
497 struct ceph_vino vino;
501 BUG_ON(rde->offset < ctx->pos);
503 ctx->pos = rde->offset;
504 dout("readdir (%d/%d) -> %llx '%.*s' %p\n",
505 i, rinfo->dir_nr, ctx->pos,
506 rde->name_len, rde->name, &rde->inode.in);
508 BUG_ON(!rde->inode.in);
509 ftype = le32_to_cpu(rde->inode.in->mode) >> 12;
510 vino.ino = le64_to_cpu(rde->inode.in->ino);
511 vino.snap = le64_to_cpu(rde->inode.in->snapid);
512 ino = ceph_vino_to_ino(vino);
514 if (!dir_emit(ctx, rde->name, rde->name_len,
515 ceph_translate_ino(inode->i_sb, ino), ftype)) {
516 dout("filldir stopping us...\n");
522 ceph_mdsc_put_request(dfi->last_readdir);
523 dfi->last_readdir = NULL;
525 if (dfi->next_offset > 2) {
531 if (!ceph_frag_is_rightmost(dfi->frag)) {
532 frag = ceph_frag_next(dfi->frag);
533 if (is_hash_order(ctx->pos)) {
534 loff_t new_pos = ceph_make_fpos(ceph_frag_value(frag),
535 dfi->next_offset, true);
536 if (new_pos > ctx->pos)
540 ctx->pos = ceph_make_fpos(frag, dfi->next_offset,
542 kfree(dfi->last_name);
543 dfi->last_name = NULL;
545 dout("readdir next frag is %x\n", frag);
548 dfi->file_info.flags |= CEPH_F_ATEND;
551 * if dir_release_count still matches the dir, no dentries
552 * were released during the whole readdir, and we should have
553 * the complete dir contents in our cache.
555 if (atomic64_read(&ci->i_release_count) ==
556 dfi->dir_release_count) {
557 spin_lock(&ci->i_ceph_lock);
558 if (dfi->dir_ordered_count ==
559 atomic64_read(&ci->i_ordered_count)) {
560 dout(" marking %p complete and ordered\n", inode);
561 /* use i_size to track number of entries in
563 BUG_ON(dfi->readdir_cache_idx < 0);
564 i_size_write(inode, dfi->readdir_cache_idx *
565 sizeof(struct dentry*));
567 dout(" marking %p complete\n", inode);
569 __ceph_dir_set_complete(ci, dfi->dir_release_count,
570 dfi->dir_ordered_count);
571 spin_unlock(&ci->i_ceph_lock);
574 dout("readdir %p file %p done.\n", inode, file);
578 static void reset_readdir(struct ceph_dir_file_info *dfi)
580 if (dfi->last_readdir) {
581 ceph_mdsc_put_request(dfi->last_readdir);
582 dfi->last_readdir = NULL;
584 kfree(dfi->last_name);
585 dfi->last_name = NULL;
586 dfi->dir_release_count = 0;
587 dfi->readdir_cache_idx = -1;
588 dfi->next_offset = 2; /* compensate for . and .. */
589 dfi->file_info.flags &= ~CEPH_F_ATEND;
593 * discard buffered readdir content on seekdir(0), or seek to new frag,
594 * or seek prior to current chunk
596 static bool need_reset_readdir(struct ceph_dir_file_info *dfi, loff_t new_pos)
598 struct ceph_mds_reply_info_parsed *rinfo;
602 if (is_hash_order(new_pos)) {
603 /* no need to reset last_name for a forward seek when
604 * dentries are sotred in hash order */
605 } else if (dfi->frag != fpos_frag(new_pos)) {
608 rinfo = dfi->last_readdir ? &dfi->last_readdir->r_reply_info : NULL;
609 if (!rinfo || !rinfo->dir_nr)
611 chunk_offset = rinfo->dir_entries[0].offset;
612 return new_pos < chunk_offset ||
613 is_hash_order(new_pos) != is_hash_order(chunk_offset);
616 static loff_t ceph_dir_llseek(struct file *file, loff_t offset, int whence)
618 struct ceph_dir_file_info *dfi = file->private_data;
619 struct inode *inode = file->f_mapping->host;
626 offset += file->f_pos;
630 retval = -EOPNOTSUPP;
636 if (need_reset_readdir(dfi, offset)) {
637 dout("dir_llseek dropping %p content\n", file);
639 } else if (is_hash_order(offset) && offset > file->f_pos) {
640 /* for hash offset, we don't know if a forward seek
641 * is within same frag */
642 dfi->dir_release_count = 0;
643 dfi->readdir_cache_idx = -1;
646 if (offset != file->f_pos) {
647 file->f_pos = offset;
649 dfi->file_info.flags &= ~CEPH_F_ATEND;
659 * Handle lookups for the hidden .snap directory.
661 int ceph_handle_snapdir(struct ceph_mds_request *req,
662 struct dentry *dentry, int err)
664 struct ceph_fs_client *fsc = ceph_sb_to_client(dentry->d_sb);
665 struct inode *parent = d_inode(dentry->d_parent); /* we hold i_mutex */
668 if (err == -ENOENT &&
669 ceph_snap(parent) == CEPH_NOSNAP &&
670 strcmp(dentry->d_name.name,
671 fsc->mount_options->snapdir_name) == 0) {
672 struct inode *inode = ceph_get_snapdir(parent);
673 dout("ENOENT on snapdir %p '%pd', linking to snapdir %p\n",
674 dentry, dentry, inode);
675 BUG_ON(!d_unhashed(dentry));
676 d_add(dentry, inode);
683 * Figure out final result of a lookup/open request.
685 * Mainly, make sure we return the final req->r_dentry (if it already
686 * existed) in place of the original VFS-provided dentry when they
689 * Gracefully handle the case where the MDS replies with -ENOENT and
690 * no trace (which it may do, at its discretion, e.g., if it doesn't
691 * care to issue a lease on the negative dentry).
693 struct dentry *ceph_finish_lookup(struct ceph_mds_request *req,
694 struct dentry *dentry, int err)
696 if (err == -ENOENT) {
699 if (!req->r_reply_info.head->is_dentry) {
700 dout("ENOENT and no trace, dentry %p inode %p\n",
701 dentry, d_inode(dentry));
702 if (d_really_is_positive(dentry)) {
711 dentry = ERR_PTR(err);
712 else if (dentry != req->r_dentry)
713 dentry = dget(req->r_dentry); /* we got spliced */
719 static bool is_root_ceph_dentry(struct inode *inode, struct dentry *dentry)
721 return ceph_ino(inode) == CEPH_INO_ROOT &&
722 strncmp(dentry->d_name.name, ".ceph", 5) == 0;
726 * Look up a single dir entry. If there is a lookup intent, inform
727 * the MDS so that it gets our 'caps wanted' value in a single op.
729 static struct dentry *ceph_lookup(struct inode *dir, struct dentry *dentry,
732 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
733 struct ceph_mds_client *mdsc = fsc->mdsc;
734 struct ceph_mds_request *req;
739 dout("lookup %p dentry %p '%pd'\n",
740 dir, dentry, dentry);
742 if (dentry->d_name.len > NAME_MAX)
743 return ERR_PTR(-ENAMETOOLONG);
745 /* can we conclude ENOENT locally? */
746 if (d_really_is_negative(dentry)) {
747 struct ceph_inode_info *ci = ceph_inode(dir);
748 struct ceph_dentry_info *di = ceph_dentry(dentry);
750 spin_lock(&ci->i_ceph_lock);
751 dout(" dir %p flags are %d\n", dir, ci->i_ceph_flags);
752 if (strncmp(dentry->d_name.name,
753 fsc->mount_options->snapdir_name,
754 dentry->d_name.len) &&
755 !is_root_ceph_dentry(dir, dentry) &&
756 ceph_test_mount_opt(fsc, DCACHE) &&
757 __ceph_dir_is_complete(ci) &&
758 (__ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1))) {
759 spin_unlock(&ci->i_ceph_lock);
760 dout(" dir %p complete, -ENOENT\n", dir);
762 di->lease_shared_gen = atomic_read(&ci->i_shared_gen);
765 spin_unlock(&ci->i_ceph_lock);
768 op = ceph_snap(dir) == CEPH_SNAPDIR ?
769 CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP;
770 req = ceph_mdsc_create_request(mdsc, op, USE_ANY_MDS);
772 return ERR_CAST(req);
773 req->r_dentry = dget(dentry);
776 mask = CEPH_STAT_CAP_INODE | CEPH_CAP_AUTH_SHARED;
777 if (ceph_security_xattr_wanted(dir))
778 mask |= CEPH_CAP_XATTR_SHARED;
779 req->r_args.getattr.mask = cpu_to_le32(mask);
782 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
783 err = ceph_mdsc_do_request(mdsc, NULL, req);
784 err = ceph_handle_snapdir(req, dentry, err);
785 dentry = ceph_finish_lookup(req, dentry, err);
786 ceph_mdsc_put_request(req); /* will dput(dentry) */
787 dout("lookup result=%p\n", dentry);
792 * If we do a create but get no trace back from the MDS, follow up with
793 * a lookup (the VFS expects us to link up the provided dentry).
795 int ceph_handle_notrace_create(struct inode *dir, struct dentry *dentry)
797 struct dentry *result = ceph_lookup(dir, dentry, 0);
799 if (result && !IS_ERR(result)) {
801 * We created the item, then did a lookup, and found
802 * it was already linked to another inode we already
803 * had in our cache (and thus got spliced). To not
804 * confuse VFS (especially when inode is a directory),
805 * we don't link our dentry to that inode, return an
808 * This event should be rare and it happens only when
809 * we talk to old MDS. Recent MDS does not send traceless
810 * reply for request that creates new inode.
815 return PTR_ERR(result);
818 static int ceph_mknod(struct inode *dir, struct dentry *dentry,
819 umode_t mode, dev_t rdev)
821 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
822 struct ceph_mds_client *mdsc = fsc->mdsc;
823 struct ceph_mds_request *req;
824 struct ceph_acls_info acls = {};
827 if (ceph_snap(dir) != CEPH_NOSNAP)
830 if (ceph_quota_is_max_files_exceeded(dir))
833 err = ceph_pre_init_acls(dir, &mode, &acls);
837 dout("mknod in dir %p dentry %p mode 0%ho rdev %d\n",
838 dir, dentry, mode, rdev);
839 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_MKNOD, USE_AUTH_MDS);
844 req->r_dentry = dget(dentry);
847 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
848 req->r_args.mknod.mode = cpu_to_le32(mode);
849 req->r_args.mknod.rdev = cpu_to_le32(rdev);
850 req->r_dentry_drop = CEPH_CAP_FILE_SHARED | CEPH_CAP_AUTH_EXCL;
851 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
853 req->r_pagelist = acls.pagelist;
854 acls.pagelist = NULL;
856 err = ceph_mdsc_do_request(mdsc, dir, req);
857 if (!err && !req->r_reply_info.head->is_dentry)
858 err = ceph_handle_notrace_create(dir, dentry);
859 ceph_mdsc_put_request(req);
862 ceph_init_inode_acls(d_inode(dentry), &acls);
865 ceph_release_acls_info(&acls);
869 static int ceph_create(struct inode *dir, struct dentry *dentry, umode_t mode,
872 return ceph_mknod(dir, dentry, mode, 0);
875 static int ceph_symlink(struct inode *dir, struct dentry *dentry,
878 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
879 struct ceph_mds_client *mdsc = fsc->mdsc;
880 struct ceph_mds_request *req;
883 if (ceph_snap(dir) != CEPH_NOSNAP)
886 if (ceph_quota_is_max_files_exceeded(dir))
889 dout("symlink in dir %p dentry %p to '%s'\n", dir, dentry, dest);
890 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SYMLINK, USE_AUTH_MDS);
895 req->r_path2 = kstrdup(dest, GFP_KERNEL);
898 ceph_mdsc_put_request(req);
902 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
903 req->r_dentry = dget(dentry);
905 req->r_dentry_drop = CEPH_CAP_FILE_SHARED | CEPH_CAP_AUTH_EXCL;
906 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
907 err = ceph_mdsc_do_request(mdsc, dir, req);
908 if (!err && !req->r_reply_info.head->is_dentry)
909 err = ceph_handle_notrace_create(dir, dentry);
910 ceph_mdsc_put_request(req);
917 static int ceph_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
919 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
920 struct ceph_mds_client *mdsc = fsc->mdsc;
921 struct ceph_mds_request *req;
922 struct ceph_acls_info acls = {};
926 if (ceph_snap(dir) == CEPH_SNAPDIR) {
927 /* mkdir .snap/foo is a MKSNAP */
928 op = CEPH_MDS_OP_MKSNAP;
929 dout("mksnap dir %p snap '%pd' dn %p\n", dir,
931 } else if (ceph_snap(dir) == CEPH_NOSNAP) {
932 dout("mkdir dir %p dn %p mode 0%ho\n", dir, dentry, mode);
933 op = CEPH_MDS_OP_MKDIR;
938 if (op == CEPH_MDS_OP_MKDIR &&
939 ceph_quota_is_max_files_exceeded(dir)) {
945 err = ceph_pre_init_acls(dir, &mode, &acls);
949 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
955 req->r_dentry = dget(dentry);
958 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
959 req->r_args.mkdir.mode = cpu_to_le32(mode);
960 req->r_dentry_drop = CEPH_CAP_FILE_SHARED | CEPH_CAP_AUTH_EXCL;
961 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
963 req->r_pagelist = acls.pagelist;
964 acls.pagelist = NULL;
966 err = ceph_mdsc_do_request(mdsc, dir, req);
968 !req->r_reply_info.head->is_target &&
969 !req->r_reply_info.head->is_dentry)
970 err = ceph_handle_notrace_create(dir, dentry);
971 ceph_mdsc_put_request(req);
974 ceph_init_inode_acls(d_inode(dentry), &acls);
977 ceph_release_acls_info(&acls);
981 static int ceph_link(struct dentry *old_dentry, struct inode *dir,
982 struct dentry *dentry)
984 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
985 struct ceph_mds_client *mdsc = fsc->mdsc;
986 struct ceph_mds_request *req;
989 if (ceph_snap(dir) != CEPH_NOSNAP)
992 dout("link in dir %p old_dentry %p dentry %p\n", dir,
994 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LINK, USE_AUTH_MDS);
999 req->r_dentry = dget(dentry);
1000 req->r_num_caps = 2;
1001 req->r_old_dentry = dget(old_dentry);
1002 req->r_parent = dir;
1003 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
1004 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
1005 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
1006 /* release LINK_SHARED on source inode (mds will lock it) */
1007 req->r_old_inode_drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL;
1008 err = ceph_mdsc_do_request(mdsc, dir, req);
1011 } else if (!req->r_reply_info.head->is_dentry) {
1012 ihold(d_inode(old_dentry));
1013 d_instantiate(dentry, d_inode(old_dentry));
1015 ceph_mdsc_put_request(req);
1020 * rmdir and unlink are differ only by the metadata op code
1022 static int ceph_unlink(struct inode *dir, struct dentry *dentry)
1024 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
1025 struct ceph_mds_client *mdsc = fsc->mdsc;
1026 struct inode *inode = d_inode(dentry);
1027 struct ceph_mds_request *req;
1031 if (ceph_snap(dir) == CEPH_SNAPDIR) {
1032 /* rmdir .snap/foo is RMSNAP */
1033 dout("rmsnap dir %p '%pd' dn %p\n", dir, dentry, dentry);
1034 op = CEPH_MDS_OP_RMSNAP;
1035 } else if (ceph_snap(dir) == CEPH_NOSNAP) {
1036 dout("unlink/rmdir dir %p dn %p inode %p\n",
1037 dir, dentry, inode);
1038 op = d_is_dir(dentry) ?
1039 CEPH_MDS_OP_RMDIR : CEPH_MDS_OP_UNLINK;
1042 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
1047 req->r_dentry = dget(dentry);
1048 req->r_num_caps = 2;
1049 req->r_parent = dir;
1050 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
1051 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
1052 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
1053 req->r_inode_drop = ceph_drop_caps_for_unlink(inode);
1054 err = ceph_mdsc_do_request(mdsc, dir, req);
1055 if (!err && !req->r_reply_info.head->is_dentry)
1057 ceph_mdsc_put_request(req);
1062 static int ceph_rename(struct inode *old_dir, struct dentry *old_dentry,
1063 struct inode *new_dir, struct dentry *new_dentry,
1066 struct ceph_fs_client *fsc = ceph_sb_to_client(old_dir->i_sb);
1067 struct ceph_mds_client *mdsc = fsc->mdsc;
1068 struct ceph_mds_request *req;
1069 int op = CEPH_MDS_OP_RENAME;
1075 if (ceph_snap(old_dir) != ceph_snap(new_dir))
1077 if (ceph_snap(old_dir) != CEPH_NOSNAP) {
1078 if (old_dir == new_dir && ceph_snap(old_dir) == CEPH_SNAPDIR)
1079 op = CEPH_MDS_OP_RENAMESNAP;
1083 /* don't allow cross-quota renames */
1084 if ((old_dir != new_dir) &&
1085 (!ceph_quota_is_same_realm(old_dir, new_dir)))
1088 dout("rename dir %p dentry %p to dir %p dentry %p\n",
1089 old_dir, old_dentry, new_dir, new_dentry);
1090 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
1092 return PTR_ERR(req);
1094 req->r_dentry = dget(new_dentry);
1095 req->r_num_caps = 2;
1096 req->r_old_dentry = dget(old_dentry);
1097 req->r_old_dentry_dir = old_dir;
1098 req->r_parent = new_dir;
1099 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
1100 req->r_old_dentry_drop = CEPH_CAP_FILE_SHARED;
1101 req->r_old_dentry_unless = CEPH_CAP_FILE_EXCL;
1102 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
1103 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
1104 /* release LINK_RDCACHE on source inode (mds will lock it) */
1105 req->r_old_inode_drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL;
1106 if (d_really_is_positive(new_dentry)) {
1108 ceph_drop_caps_for_unlink(d_inode(new_dentry));
1110 err = ceph_mdsc_do_request(mdsc, old_dir, req);
1111 if (!err && !req->r_reply_info.head->is_dentry) {
1113 * Normally d_move() is done by fill_trace (called by
1114 * do_request, above). If there is no trace, we need
1117 d_move(old_dentry, new_dentry);
1119 ceph_mdsc_put_request(req);
1124 * Ensure a dentry lease will no longer revalidate.
1126 void ceph_invalidate_dentry_lease(struct dentry *dentry)
1128 spin_lock(&dentry->d_lock);
1129 ceph_dentry(dentry)->time = jiffies;
1130 ceph_dentry(dentry)->lease_shared_gen = 0;
1131 spin_unlock(&dentry->d_lock);
1135 * Check if dentry lease is valid. If not, delete the lease. Try to
1136 * renew if the least is more than half up.
1138 static int dentry_lease_is_valid(struct dentry *dentry, unsigned int flags,
1141 struct ceph_dentry_info *di;
1142 struct ceph_mds_session *s;
1146 struct ceph_mds_session *session = NULL;
1149 spin_lock(&dentry->d_lock);
1150 di = ceph_dentry(dentry);
1151 if (di && di->lease_session) {
1152 s = di->lease_session;
1153 spin_lock(&s->s_gen_ttl_lock);
1156 spin_unlock(&s->s_gen_ttl_lock);
1158 if (di->lease_gen == gen &&
1159 time_before(jiffies, di->time) &&
1160 time_before(jiffies, ttl)) {
1162 if (di->lease_renew_after &&
1163 time_after(jiffies, di->lease_renew_after)) {
1165 * We should renew. If we're in RCU walk mode
1166 * though, we can't do that so just return
1169 if (flags & LOOKUP_RCU) {
1172 session = ceph_get_mds_session(s);
1173 seq = di->lease_seq;
1174 di->lease_renew_after = 0;
1175 di->lease_renew_from = jiffies;
1180 spin_unlock(&dentry->d_lock);
1183 ceph_mdsc_lease_send_msg(session, dir, dentry,
1184 CEPH_MDS_LEASE_RENEW, seq);
1185 ceph_put_mds_session(session);
1187 dout("dentry_lease_is_valid - dentry %p = %d\n", dentry, valid);
1192 * Check if directory-wide content lease/cap is valid.
1194 static int dir_lease_is_valid(struct inode *dir, struct dentry *dentry)
1196 struct ceph_inode_info *ci = ceph_inode(dir);
1197 struct ceph_dentry_info *di = ceph_dentry(dentry);
1200 spin_lock(&ci->i_ceph_lock);
1201 if (atomic_read(&ci->i_shared_gen) == di->lease_shared_gen)
1202 valid = __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1);
1203 spin_unlock(&ci->i_ceph_lock);
1204 dout("dir_lease_is_valid dir %p v%u dentry %p v%u = %d\n",
1205 dir, (unsigned)atomic_read(&ci->i_shared_gen),
1206 dentry, (unsigned)di->lease_shared_gen, valid);
1211 * Check if cached dentry can be trusted.
1213 static int ceph_d_revalidate(struct dentry *dentry, unsigned int flags)
1216 struct dentry *parent;
1219 if (flags & LOOKUP_RCU) {
1220 parent = READ_ONCE(dentry->d_parent);
1221 dir = d_inode_rcu(parent);
1225 parent = dget_parent(dentry);
1226 dir = d_inode(parent);
1229 dout("d_revalidate %p '%pd' inode %p offset %lld\n", dentry,
1230 dentry, d_inode(dentry), ceph_dentry(dentry)->offset);
1232 /* always trust cached snapped dentries, snapdir dentry */
1233 if (ceph_snap(dir) != CEPH_NOSNAP) {
1234 dout("d_revalidate %p '%pd' inode %p is SNAPPED\n", dentry,
1235 dentry, d_inode(dentry));
1237 } else if (d_really_is_positive(dentry) &&
1238 ceph_snap(d_inode(dentry)) == CEPH_SNAPDIR) {
1241 valid = dentry_lease_is_valid(dentry, flags, dir);
1242 if (valid == -ECHILD)
1244 if (valid || dir_lease_is_valid(dir, dentry)) {
1245 if (d_really_is_positive(dentry))
1246 valid = ceph_is_any_caps(d_inode(dentry));
1253 struct ceph_mds_client *mdsc =
1254 ceph_sb_to_client(dir->i_sb)->mdsc;
1255 struct ceph_mds_request *req;
1259 if (flags & LOOKUP_RCU)
1262 op = ceph_snap(dir) == CEPH_SNAPDIR ?
1263 CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP;
1264 req = ceph_mdsc_create_request(mdsc, op, USE_ANY_MDS);
1266 req->r_dentry = dget(dentry);
1267 req->r_num_caps = 2;
1268 req->r_parent = dir;
1270 mask = CEPH_STAT_CAP_INODE | CEPH_CAP_AUTH_SHARED;
1271 if (ceph_security_xattr_wanted(dir))
1272 mask |= CEPH_CAP_XATTR_SHARED;
1273 req->r_args.getattr.mask = cpu_to_le32(mask);
1275 err = ceph_mdsc_do_request(mdsc, NULL, req);
1278 if (d_really_is_positive(dentry) &&
1279 d_inode(dentry) == req->r_target_inode)
1283 if (d_really_is_negative(dentry))
1289 ceph_mdsc_put_request(req);
1290 dout("d_revalidate %p lookup result=%d\n",
1295 dout("d_revalidate %p %s\n", dentry, valid ? "valid" : "invalid");
1297 ceph_dentry_lru_touch(dentry);
1299 ceph_dir_clear_complete(dir);
1302 if (!(flags & LOOKUP_RCU))
1308 * Release our ceph_dentry_info.
1310 static void ceph_d_release(struct dentry *dentry)
1312 struct ceph_dentry_info *di = ceph_dentry(dentry);
1314 dout("d_release %p\n", dentry);
1315 ceph_dentry_lru_del(dentry);
1317 spin_lock(&dentry->d_lock);
1318 dentry->d_fsdata = NULL;
1319 spin_unlock(&dentry->d_lock);
1321 if (di->lease_session)
1322 ceph_put_mds_session(di->lease_session);
1323 kmem_cache_free(ceph_dentry_cachep, di);
1327 * When the VFS prunes a dentry from the cache, we need to clear the
1328 * complete flag on the parent directory.
1330 * Called under dentry->d_lock.
1332 static void ceph_d_prune(struct dentry *dentry)
1334 struct ceph_inode_info *dir_ci;
1335 struct ceph_dentry_info *di;
1337 dout("ceph_d_prune %pd %p\n", dentry, dentry);
1339 /* do we have a valid parent? */
1340 if (IS_ROOT(dentry))
1343 /* we hold d_lock, so d_parent is stable */
1344 dir_ci = ceph_inode(d_inode(dentry->d_parent));
1345 if (dir_ci->i_vino.snap == CEPH_SNAPDIR)
1348 /* who calls d_delete() should also disable dcache readdir */
1349 if (d_really_is_negative(dentry))
1352 /* d_fsdata does not get cleared until d_release */
1353 if (!d_unhashed(dentry)) {
1354 __ceph_dir_clear_complete(dir_ci);
1358 /* Disable dcache readdir just in case that someone called d_drop()
1359 * or d_invalidate(), but MDS didn't revoke CEPH_CAP_FILE_SHARED
1360 * properly (dcache readdir is still enabled) */
1361 di = ceph_dentry(dentry);
1362 if (di->offset > 0 &&
1363 di->lease_shared_gen == atomic_read(&dir_ci->i_shared_gen))
1364 __ceph_dir_clear_ordered(dir_ci);
1368 * read() on a dir. This weird interface hack only works if mounted
1369 * with '-o dirstat'.
1371 static ssize_t ceph_read_dir(struct file *file, char __user *buf, size_t size,
1374 struct ceph_dir_file_info *dfi = file->private_data;
1375 struct inode *inode = file_inode(file);
1376 struct ceph_inode_info *ci = ceph_inode(inode);
1378 const int bufsize = 1024;
1380 if (!ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb), DIRSTAT))
1383 if (!dfi->dir_info) {
1384 dfi->dir_info = kmalloc(bufsize, GFP_KERNEL);
1388 snprintf(dfi->dir_info, bufsize,
1391 " subdirs: %20lld\n"
1392 "rentries: %20lld\n"
1394 " rsubdirs: %20lld\n"
1396 "rctime: %10ld.%09ld\n",
1397 ci->i_files + ci->i_subdirs,
1400 ci->i_rfiles + ci->i_rsubdirs,
1404 (long)ci->i_rctime.tv_sec,
1405 (long)ci->i_rctime.tv_nsec);
1408 if (*ppos >= dfi->dir_info_len)
1410 size = min_t(unsigned, size, dfi->dir_info_len-*ppos);
1411 left = copy_to_user(buf, dfi->dir_info + *ppos, size);
1414 *ppos += (size - left);
1419 * We maintain a private dentry LRU.
1421 * FIXME: this needs to be changed to a per-mds lru to be useful.
1423 void ceph_dentry_lru_add(struct dentry *dn)
1425 struct ceph_dentry_info *di = ceph_dentry(dn);
1426 struct ceph_mds_client *mdsc;
1428 dout("dentry_lru_add %p %p '%pd'\n", di, dn, dn);
1429 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1430 spin_lock(&mdsc->dentry_lru_lock);
1431 list_add_tail(&di->lru, &mdsc->dentry_lru);
1433 spin_unlock(&mdsc->dentry_lru_lock);
1436 void ceph_dentry_lru_touch(struct dentry *dn)
1438 struct ceph_dentry_info *di = ceph_dentry(dn);
1439 struct ceph_mds_client *mdsc;
1441 dout("dentry_lru_touch %p %p '%pd' (offset %lld)\n", di, dn, dn,
1443 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1444 spin_lock(&mdsc->dentry_lru_lock);
1445 list_move_tail(&di->lru, &mdsc->dentry_lru);
1446 spin_unlock(&mdsc->dentry_lru_lock);
1449 void ceph_dentry_lru_del(struct dentry *dn)
1451 struct ceph_dentry_info *di = ceph_dentry(dn);
1452 struct ceph_mds_client *mdsc;
1454 dout("dentry_lru_del %p %p '%pd'\n", di, dn, dn);
1455 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1456 spin_lock(&mdsc->dentry_lru_lock);
1457 list_del_init(&di->lru);
1459 spin_unlock(&mdsc->dentry_lru_lock);
1463 * Return name hash for a given dentry. This is dependent on
1464 * the parent directory's hash function.
1466 unsigned ceph_dentry_hash(struct inode *dir, struct dentry *dn)
1468 struct ceph_inode_info *dci = ceph_inode(dir);
1470 switch (dci->i_dir_layout.dl_dir_hash) {
1471 case 0: /* for backward compat */
1472 case CEPH_STR_HASH_LINUX:
1473 return dn->d_name.hash;
1476 return ceph_str_hash(dci->i_dir_layout.dl_dir_hash,
1477 dn->d_name.name, dn->d_name.len);
1481 const struct file_operations ceph_dir_fops = {
1482 .read = ceph_read_dir,
1483 .iterate = ceph_readdir,
1484 .llseek = ceph_dir_llseek,
1486 .release = ceph_release,
1487 .unlocked_ioctl = ceph_ioctl,
1488 .fsync = ceph_fsync,
1490 .flock = ceph_flock,
1493 const struct file_operations ceph_snapdir_fops = {
1494 .iterate = ceph_readdir,
1495 .llseek = ceph_dir_llseek,
1497 .release = ceph_release,
1500 const struct inode_operations ceph_dir_iops = {
1501 .lookup = ceph_lookup,
1502 .permission = ceph_permission,
1503 .getattr = ceph_getattr,
1504 .setattr = ceph_setattr,
1505 .listxattr = ceph_listxattr,
1506 .get_acl = ceph_get_acl,
1507 .set_acl = ceph_set_acl,
1508 .mknod = ceph_mknod,
1509 .symlink = ceph_symlink,
1510 .mkdir = ceph_mkdir,
1512 .unlink = ceph_unlink,
1513 .rmdir = ceph_unlink,
1514 .rename = ceph_rename,
1515 .create = ceph_create,
1516 .atomic_open = ceph_atomic_open,
1519 const struct inode_operations ceph_snapdir_iops = {
1520 .lookup = ceph_lookup,
1521 .permission = ceph_permission,
1522 .getattr = ceph_getattr,
1523 .mkdir = ceph_mkdir,
1524 .rmdir = ceph_unlink,
1525 .rename = ceph_rename,
1528 const struct dentry_operations ceph_dentry_ops = {
1529 .d_revalidate = ceph_d_revalidate,
1530 .d_release = ceph_d_release,
1531 .d_prune = ceph_d_prune,
1532 .d_init = ceph_d_init,