1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/ceph/ceph_debug.h>
4 #include <linux/spinlock.h>
5 #include <linux/fs_struct.h>
6 #include <linux/namei.h>
7 #include <linux/slab.h>
8 #include <linux/sched.h>
9 #include <linux/xattr.h>
12 #include "mds_client.h"
15 * Directory operations: readdir, lookup, create, link, unlink,
20 * Ceph MDS operations are specified in terms of a base ino and
21 * relative path. Thus, the client can specify an operation on a
22 * specific inode (e.g., a getattr due to fstat(2)), or as a path
23 * relative to, say, the root directory.
25 * Normally, we limit ourselves to strict inode ops (no path component)
26 * or dentry operations (a single path component relative to an ino). The
27 * exception to this is open_root_dentry(), which will open the mount
31 const struct dentry_operations ceph_dentry_ops;
34 * Initialize ceph dentry state.
36 static int ceph_d_init(struct dentry *dentry)
38 struct ceph_dentry_info *di;
40 di = kmem_cache_zalloc(ceph_dentry_cachep, GFP_KERNEL);
42 return -ENOMEM; /* oh well */
45 di->lease_session = NULL;
47 dentry->d_fsdata = di;
48 ceph_dentry_lru_add(dentry);
53 * for f_pos for readdir:
55 * (0xff << 52) | ((24 bits hash) << 28) |
56 * (the nth entry has hash collision);
58 * ((frag value) << 28) | (the nth entry in frag);
60 #define OFFSET_BITS 28
61 #define OFFSET_MASK ((1 << OFFSET_BITS) - 1)
62 #define HASH_ORDER (0xffull << (OFFSET_BITS + 24))
63 loff_t ceph_make_fpos(unsigned high, unsigned off, bool hash_order)
65 loff_t fpos = ((loff_t)high << 28) | (loff_t)off;
71 static bool is_hash_order(loff_t p)
73 return (p & HASH_ORDER) == HASH_ORDER;
76 static unsigned fpos_frag(loff_t p)
78 return p >> OFFSET_BITS;
81 static unsigned fpos_hash(loff_t p)
83 return ceph_frag_value(fpos_frag(p));
86 static unsigned fpos_off(loff_t p)
88 return p & OFFSET_MASK;
91 static int fpos_cmp(loff_t l, loff_t r)
93 int v = ceph_frag_compare(fpos_frag(l), fpos_frag(r));
96 return (int)(fpos_off(l) - fpos_off(r));
100 * make note of the last dentry we read, so we can
101 * continue at the same lexicographical point,
102 * regardless of what dir changes take place on the
105 static int note_last_dentry(struct ceph_file_info *fi, const char *name,
106 int len, unsigned next_offset)
108 char *buf = kmalloc(len+1, GFP_KERNEL);
111 kfree(fi->last_name);
113 memcpy(fi->last_name, name, len);
114 fi->last_name[len] = 0;
115 fi->next_offset = next_offset;
116 dout("note_last_dentry '%s'\n", fi->last_name);
121 static struct dentry *
122 __dcache_find_get_entry(struct dentry *parent, u64 idx,
123 struct ceph_readdir_cache_control *cache_ctl)
125 struct inode *dir = d_inode(parent);
126 struct dentry *dentry;
127 unsigned idx_mask = (PAGE_SIZE / sizeof(struct dentry *)) - 1;
128 loff_t ptr_pos = idx * sizeof(struct dentry *);
129 pgoff_t ptr_pgoff = ptr_pos >> PAGE_SHIFT;
131 if (ptr_pos >= i_size_read(dir))
134 if (!cache_ctl->page || ptr_pgoff != page_index(cache_ctl->page)) {
135 ceph_readdir_cache_release(cache_ctl);
136 cache_ctl->page = find_lock_page(&dir->i_data, ptr_pgoff);
137 if (!cache_ctl->page) {
138 dout(" page %lu not found\n", ptr_pgoff);
139 return ERR_PTR(-EAGAIN);
141 /* reading/filling the cache are serialized by
142 i_mutex, no need to use page lock */
143 unlock_page(cache_ctl->page);
144 cache_ctl->dentries = kmap(cache_ctl->page);
147 cache_ctl->index = idx & idx_mask;
150 spin_lock(&parent->d_lock);
151 /* check i_size again here, because empty directory can be
152 * marked as complete while not holding the i_mutex. */
153 if (ceph_dir_is_complete_ordered(dir) && ptr_pos < i_size_read(dir))
154 dentry = cache_ctl->dentries[cache_ctl->index];
157 spin_unlock(&parent->d_lock);
158 if (dentry && !lockref_get_not_dead(&dentry->d_lockref))
161 return dentry ? : ERR_PTR(-EAGAIN);
165 * When possible, we try to satisfy a readdir by peeking at the
166 * dcache. We make this work by carefully ordering dentries on
167 * d_child when we initially get results back from the MDS, and
168 * falling back to a "normal" sync readdir if any dentries in the dir
171 * Complete dir indicates that we have all dentries in the dir. It is
172 * defined IFF we hold CEPH_CAP_FILE_SHARED (which will be revoked by
173 * the MDS if/when the directory is modified).
175 static int __dcache_readdir(struct file *file, struct dir_context *ctx,
178 struct ceph_file_info *fi = file->private_data;
179 struct dentry *parent = file->f_path.dentry;
180 struct inode *dir = d_inode(parent);
181 struct dentry *dentry, *last = NULL;
182 struct ceph_dentry_info *di;
183 struct ceph_readdir_cache_control cache_ctl = {};
187 dout("__dcache_readdir %p v%u at %llx\n", dir, (unsigned)shared_gen, ctx->pos);
189 /* search start position */
191 u64 count = div_u64(i_size_read(dir), sizeof(struct dentry *));
193 u64 step = count >> 1;
194 dentry = __dcache_find_get_entry(parent, idx + step,
197 /* use linar search */
201 if (IS_ERR(dentry)) {
202 err = PTR_ERR(dentry);
205 di = ceph_dentry(dentry);
206 spin_lock(&dentry->d_lock);
207 if (fpos_cmp(di->offset, ctx->pos) < 0) {
213 spin_unlock(&dentry->d_lock);
217 dout("__dcache_readdir %p cache idx %llu\n", dir, idx);
222 bool emit_dentry = false;
223 dentry = __dcache_find_get_entry(parent, idx++, &cache_ctl);
225 fi->flags |= CEPH_F_ATEND;
229 if (IS_ERR(dentry)) {
230 err = PTR_ERR(dentry);
234 spin_lock(&dentry->d_lock);
235 di = ceph_dentry(dentry);
236 if (d_unhashed(dentry) ||
237 d_really_is_negative(dentry) ||
238 di->lease_shared_gen != shared_gen) {
239 spin_unlock(&dentry->d_lock);
244 if (fpos_cmp(ctx->pos, di->offset) <= 0) {
247 spin_unlock(&dentry->d_lock);
250 dout(" %llx dentry %p %pd %p\n", di->offset,
251 dentry, dentry, d_inode(dentry));
252 ctx->pos = di->offset;
253 if (!dir_emit(ctx, dentry->d_name.name,
255 ceph_translate_ino(dentry->d_sb,
256 d_inode(dentry)->i_ino),
257 d_inode(dentry)->i_mode >> 12)) {
272 ceph_readdir_cache_release(&cache_ctl);
275 di = ceph_dentry(last);
276 ret = note_last_dentry(fi, last->d_name.name, last->d_name.len,
277 fpos_off(di->offset) + 1);
281 /* last_name no longer match cache index */
282 if (fi->readdir_cache_idx >= 0) {
283 fi->readdir_cache_idx = -1;
284 fi->dir_release_count = 0;
290 static bool need_send_readdir(struct ceph_file_info *fi, loff_t pos)
292 if (!fi->last_readdir)
294 if (is_hash_order(pos))
295 return !ceph_frag_contains_value(fi->frag, fpos_hash(pos));
297 return fi->frag != fpos_frag(pos);
300 static int ceph_readdir(struct file *file, struct dir_context *ctx)
302 struct ceph_file_info *fi = file->private_data;
303 struct inode *inode = file_inode(file);
304 struct ceph_inode_info *ci = ceph_inode(inode);
305 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
306 struct ceph_mds_client *mdsc = fsc->mdsc;
310 struct ceph_mds_reply_info_parsed *rinfo;
312 dout("readdir %p file %p pos %llx\n", inode, file, ctx->pos);
313 if (fi->flags & CEPH_F_ATEND)
316 /* always start with . and .. */
318 dout("readdir off 0 -> '.'\n");
319 if (!dir_emit(ctx, ".", 1,
320 ceph_translate_ino(inode->i_sb, inode->i_ino),
321 inode->i_mode >> 12))
326 ino_t ino = parent_ino(file->f_path.dentry);
327 dout("readdir off 1 -> '..'\n");
328 if (!dir_emit(ctx, "..", 2,
329 ceph_translate_ino(inode->i_sb, ino),
330 inode->i_mode >> 12))
335 /* can we use the dcache? */
336 spin_lock(&ci->i_ceph_lock);
337 if (ceph_test_mount_opt(fsc, DCACHE) &&
338 !ceph_test_mount_opt(fsc, NOASYNCREADDIR) &&
339 ceph_snap(inode) != CEPH_SNAPDIR &&
340 __ceph_dir_is_complete_ordered(ci) &&
341 __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1)) {
342 int shared_gen = atomic_read(&ci->i_shared_gen);
343 spin_unlock(&ci->i_ceph_lock);
344 err = __dcache_readdir(file, ctx, shared_gen);
348 spin_unlock(&ci->i_ceph_lock);
351 /* proceed with a normal readdir */
353 /* do we have the correct frag content buffered? */
354 if (need_send_readdir(fi, ctx->pos)) {
355 struct ceph_mds_request *req;
356 int op = ceph_snap(inode) == CEPH_SNAPDIR ?
357 CEPH_MDS_OP_LSSNAP : CEPH_MDS_OP_READDIR;
359 /* discard old result, if any */
360 if (fi->last_readdir) {
361 ceph_mdsc_put_request(fi->last_readdir);
362 fi->last_readdir = NULL;
365 if (is_hash_order(ctx->pos)) {
366 /* fragtree isn't always accurate. choose frag
367 * based on previous reply when possible. */
368 if (frag == (unsigned)-1)
369 frag = ceph_choose_frag(ci, fpos_hash(ctx->pos),
372 frag = fpos_frag(ctx->pos);
375 dout("readdir fetching %llx.%llx frag %x offset '%s'\n",
376 ceph_vinop(inode), frag, fi->last_name);
377 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
380 err = ceph_alloc_readdir_reply_buffer(req, inode);
382 ceph_mdsc_put_request(req);
385 /* hints to request -> mds selection code */
386 req->r_direct_mode = USE_AUTH_MDS;
387 if (op == CEPH_MDS_OP_READDIR) {
388 req->r_direct_hash = ceph_frag_value(frag);
389 __set_bit(CEPH_MDS_R_DIRECT_IS_HASH, &req->r_req_flags);
390 req->r_inode_drop = CEPH_CAP_FILE_EXCL;
393 req->r_path2 = kstrdup(fi->last_name, GFP_KERNEL);
395 ceph_mdsc_put_request(req);
398 } else if (is_hash_order(ctx->pos)) {
399 req->r_args.readdir.offset_hash =
400 cpu_to_le32(fpos_hash(ctx->pos));
403 req->r_dir_release_cnt = fi->dir_release_count;
404 req->r_dir_ordered_cnt = fi->dir_ordered_count;
405 req->r_readdir_cache_idx = fi->readdir_cache_idx;
406 req->r_readdir_offset = fi->next_offset;
407 req->r_args.readdir.frag = cpu_to_le32(frag);
408 req->r_args.readdir.flags =
409 cpu_to_le16(CEPH_READDIR_REPLY_BITFLAGS);
411 req->r_inode = inode;
413 req->r_dentry = dget(file->f_path.dentry);
414 err = ceph_mdsc_do_request(mdsc, NULL, req);
416 ceph_mdsc_put_request(req);
419 dout("readdir got and parsed readdir result=%d on "
420 "frag %x, end=%d, complete=%d, hash_order=%d\n",
422 (int)req->r_reply_info.dir_end,
423 (int)req->r_reply_info.dir_complete,
424 (int)req->r_reply_info.hash_order);
426 rinfo = &req->r_reply_info;
427 if (le32_to_cpu(rinfo->dir_dir->frag) != frag) {
428 frag = le32_to_cpu(rinfo->dir_dir->frag);
429 if (!rinfo->hash_order) {
430 fi->next_offset = req->r_readdir_offset;
431 /* adjust ctx->pos to beginning of frag */
432 ctx->pos = ceph_make_fpos(frag,
439 fi->last_readdir = req;
441 if (test_bit(CEPH_MDS_R_DID_PREPOPULATE, &req->r_req_flags)) {
442 fi->readdir_cache_idx = req->r_readdir_cache_idx;
443 if (fi->readdir_cache_idx < 0) {
444 /* preclude from marking dir ordered */
445 fi->dir_ordered_count = 0;
446 } else if (ceph_frag_is_leftmost(frag) &&
447 fi->next_offset == 2) {
448 /* note dir version at start of readdir so
449 * we can tell if any dentries get dropped */
450 fi->dir_release_count = req->r_dir_release_cnt;
451 fi->dir_ordered_count = req->r_dir_ordered_cnt;
454 dout("readdir !did_prepopulate");
455 /* disable readdir cache */
456 fi->readdir_cache_idx = -1;
457 /* preclude from marking dir complete */
458 fi->dir_release_count = 0;
461 /* note next offset and last dentry name */
462 if (rinfo->dir_nr > 0) {
463 struct ceph_mds_reply_dir_entry *rde =
464 rinfo->dir_entries + (rinfo->dir_nr-1);
465 unsigned next_offset = req->r_reply_info.dir_end ?
466 2 : (fpos_off(rde->offset) + 1);
467 err = note_last_dentry(fi, rde->name, rde->name_len,
471 } else if (req->r_reply_info.dir_end) {
477 rinfo = &fi->last_readdir->r_reply_info;
478 dout("readdir frag %x num %d pos %llx chunk first %llx\n",
479 fi->frag, rinfo->dir_nr, ctx->pos,
480 rinfo->dir_nr ? rinfo->dir_entries[0].offset : 0LL);
483 /* search start position */
484 if (rinfo->dir_nr > 0) {
485 int step, nr = rinfo->dir_nr;
488 if (rinfo->dir_entries[i + step].offset < ctx->pos) {
496 for (; i < rinfo->dir_nr; i++) {
497 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
498 struct ceph_vino vino;
502 BUG_ON(rde->offset < ctx->pos);
504 ctx->pos = rde->offset;
505 dout("readdir (%d/%d) -> %llx '%.*s' %p\n",
506 i, rinfo->dir_nr, ctx->pos,
507 rde->name_len, rde->name, &rde->inode.in);
509 BUG_ON(!rde->inode.in);
510 ftype = le32_to_cpu(rde->inode.in->mode) >> 12;
511 vino.ino = le64_to_cpu(rde->inode.in->ino);
512 vino.snap = le64_to_cpu(rde->inode.in->snapid);
513 ino = ceph_vino_to_ino(vino);
515 if (!dir_emit(ctx, rde->name, rde->name_len,
516 ceph_translate_ino(inode->i_sb, ino), ftype)) {
517 dout("filldir stopping us...\n");
523 ceph_mdsc_put_request(fi->last_readdir);
524 fi->last_readdir = NULL;
526 if (fi->next_offset > 2) {
532 if (!ceph_frag_is_rightmost(fi->frag)) {
533 frag = ceph_frag_next(fi->frag);
534 if (is_hash_order(ctx->pos)) {
535 loff_t new_pos = ceph_make_fpos(ceph_frag_value(frag),
536 fi->next_offset, true);
537 if (new_pos > ctx->pos)
541 ctx->pos = ceph_make_fpos(frag, fi->next_offset, false);
542 kfree(fi->last_name);
543 fi->last_name = NULL;
545 dout("readdir next frag is %x\n", frag);
548 fi->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) == fi->dir_release_count) {
556 spin_lock(&ci->i_ceph_lock);
557 if (fi->dir_ordered_count == atomic64_read(&ci->i_ordered_count)) {
558 dout(" marking %p complete and ordered\n", inode);
559 /* use i_size to track number of entries in
561 BUG_ON(fi->readdir_cache_idx < 0);
562 i_size_write(inode, fi->readdir_cache_idx *
563 sizeof(struct dentry*));
565 dout(" marking %p complete\n", inode);
567 __ceph_dir_set_complete(ci, fi->dir_release_count,
568 fi->dir_ordered_count);
569 spin_unlock(&ci->i_ceph_lock);
572 dout("readdir %p file %p done.\n", inode, file);
576 static void reset_readdir(struct ceph_file_info *fi)
578 if (fi->last_readdir) {
579 ceph_mdsc_put_request(fi->last_readdir);
580 fi->last_readdir = NULL;
582 kfree(fi->last_name);
583 fi->last_name = NULL;
584 fi->dir_release_count = 0;
585 fi->readdir_cache_idx = -1;
586 fi->next_offset = 2; /* compensate for . and .. */
587 fi->flags &= ~CEPH_F_ATEND;
591 * discard buffered readdir content on seekdir(0), or seek to new frag,
592 * or seek prior to current chunk
594 static bool need_reset_readdir(struct ceph_file_info *fi, loff_t new_pos)
596 struct ceph_mds_reply_info_parsed *rinfo;
600 if (is_hash_order(new_pos)) {
601 /* no need to reset last_name for a forward seek when
602 * dentries are sotred in hash order */
603 } else if (fi->frag != fpos_frag(new_pos)) {
606 rinfo = fi->last_readdir ? &fi->last_readdir->r_reply_info : NULL;
607 if (!rinfo || !rinfo->dir_nr)
609 chunk_offset = rinfo->dir_entries[0].offset;
610 return new_pos < chunk_offset ||
611 is_hash_order(new_pos) != is_hash_order(chunk_offset);
614 static loff_t ceph_dir_llseek(struct file *file, loff_t offset, int whence)
616 struct ceph_file_info *fi = file->private_data;
617 struct inode *inode = file->f_mapping->host;
624 offset += file->f_pos;
628 retval = -EOPNOTSUPP;
634 if (need_reset_readdir(fi, offset)) {
635 dout("dir_llseek dropping %p content\n", file);
637 } else if (is_hash_order(offset) && offset > file->f_pos) {
638 /* for hash offset, we don't know if a forward seek
639 * is within same frag */
640 fi->dir_release_count = 0;
641 fi->readdir_cache_idx = -1;
644 if (offset != file->f_pos) {
645 file->f_pos = offset;
647 fi->flags &= ~CEPH_F_ATEND;
657 * Handle lookups for the hidden .snap directory.
659 int ceph_handle_snapdir(struct ceph_mds_request *req,
660 struct dentry *dentry, int err)
662 struct ceph_fs_client *fsc = ceph_sb_to_client(dentry->d_sb);
663 struct inode *parent = d_inode(dentry->d_parent); /* we hold i_mutex */
666 if (err == -ENOENT &&
667 ceph_snap(parent) == CEPH_NOSNAP &&
668 strcmp(dentry->d_name.name,
669 fsc->mount_options->snapdir_name) == 0) {
670 struct inode *inode = ceph_get_snapdir(parent);
671 dout("ENOENT on snapdir %p '%pd', linking to snapdir %p\n",
672 dentry, dentry, inode);
673 BUG_ON(!d_unhashed(dentry));
674 d_add(dentry, inode);
681 * Figure out final result of a lookup/open request.
683 * Mainly, make sure we return the final req->r_dentry (if it already
684 * existed) in place of the original VFS-provided dentry when they
687 * Gracefully handle the case where the MDS replies with -ENOENT and
688 * no trace (which it may do, at its discretion, e.g., if it doesn't
689 * care to issue a lease on the negative dentry).
691 struct dentry *ceph_finish_lookup(struct ceph_mds_request *req,
692 struct dentry *dentry, int err)
694 if (err == -ENOENT) {
697 if (!req->r_reply_info.head->is_dentry) {
698 dout("ENOENT and no trace, dentry %p inode %p\n",
699 dentry, d_inode(dentry));
700 if (d_really_is_positive(dentry)) {
709 dentry = ERR_PTR(err);
710 else if (dentry != req->r_dentry)
711 dentry = dget(req->r_dentry); /* we got spliced */
717 static bool is_root_ceph_dentry(struct inode *inode, struct dentry *dentry)
719 return ceph_ino(inode) == CEPH_INO_ROOT &&
720 strncmp(dentry->d_name.name, ".ceph", 5) == 0;
724 * Look up a single dir entry. If there is a lookup intent, inform
725 * the MDS so that it gets our 'caps wanted' value in a single op.
727 static struct dentry *ceph_lookup(struct inode *dir, struct dentry *dentry,
730 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
731 struct ceph_mds_client *mdsc = fsc->mdsc;
732 struct ceph_mds_request *req;
737 dout("lookup %p dentry %p '%pd'\n",
738 dir, dentry, dentry);
740 if (dentry->d_name.len > NAME_MAX)
741 return ERR_PTR(-ENAMETOOLONG);
743 /* can we conclude ENOENT locally? */
744 if (d_really_is_negative(dentry)) {
745 struct ceph_inode_info *ci = ceph_inode(dir);
746 struct ceph_dentry_info *di = ceph_dentry(dentry);
748 spin_lock(&ci->i_ceph_lock);
749 dout(" dir %p flags are %d\n", dir, ci->i_ceph_flags);
750 if (strncmp(dentry->d_name.name,
751 fsc->mount_options->snapdir_name,
752 dentry->d_name.len) &&
753 !is_root_ceph_dentry(dir, dentry) &&
754 ceph_test_mount_opt(fsc, DCACHE) &&
755 __ceph_dir_is_complete(ci) &&
756 (__ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1))) {
757 spin_unlock(&ci->i_ceph_lock);
758 dout(" dir %p complete, -ENOENT\n", dir);
760 di->lease_shared_gen = atomic_read(&ci->i_shared_gen);
763 spin_unlock(&ci->i_ceph_lock);
766 op = ceph_snap(dir) == CEPH_SNAPDIR ?
767 CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP;
768 req = ceph_mdsc_create_request(mdsc, op, USE_ANY_MDS);
770 return ERR_CAST(req);
771 req->r_dentry = dget(dentry);
774 mask = CEPH_STAT_CAP_INODE | CEPH_CAP_AUTH_SHARED;
775 if (ceph_security_xattr_wanted(dir))
776 mask |= CEPH_CAP_XATTR_SHARED;
777 req->r_args.getattr.mask = cpu_to_le32(mask);
780 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
781 err = ceph_mdsc_do_request(mdsc, NULL, req);
782 err = ceph_handle_snapdir(req, dentry, err);
783 dentry = ceph_finish_lookup(req, dentry, err);
784 ceph_mdsc_put_request(req); /* will dput(dentry) */
785 dout("lookup result=%p\n", dentry);
790 * If we do a create but get no trace back from the MDS, follow up with
791 * a lookup (the VFS expects us to link up the provided dentry).
793 int ceph_handle_notrace_create(struct inode *dir, struct dentry *dentry)
795 struct dentry *result = ceph_lookup(dir, dentry, 0);
797 if (result && !IS_ERR(result)) {
799 * We created the item, then did a lookup, and found
800 * it was already linked to another inode we already
801 * had in our cache (and thus got spliced). To not
802 * confuse VFS (especially when inode is a directory),
803 * we don't link our dentry to that inode, return an
806 * This event should be rare and it happens only when
807 * we talk to old MDS. Recent MDS does not send traceless
808 * reply for request that creates new inode.
813 return PTR_ERR(result);
816 static int ceph_mknod(struct inode *dir, struct dentry *dentry,
817 umode_t mode, dev_t rdev)
819 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
820 struct ceph_mds_client *mdsc = fsc->mdsc;
821 struct ceph_mds_request *req;
822 struct ceph_acls_info acls = {};
825 if (ceph_snap(dir) != CEPH_NOSNAP)
828 err = ceph_pre_init_acls(dir, &mode, &acls);
832 dout("mknod in dir %p dentry %p mode 0%ho rdev %d\n",
833 dir, dentry, mode, rdev);
834 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_MKNOD, USE_AUTH_MDS);
839 req->r_dentry = dget(dentry);
842 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
843 req->r_args.mknod.mode = cpu_to_le32(mode);
844 req->r_args.mknod.rdev = cpu_to_le32(rdev);
845 req->r_dentry_drop = CEPH_CAP_FILE_SHARED | CEPH_CAP_AUTH_EXCL;
846 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
848 req->r_pagelist = acls.pagelist;
849 acls.pagelist = NULL;
851 err = ceph_mdsc_do_request(mdsc, dir, req);
852 if (!err && !req->r_reply_info.head->is_dentry)
853 err = ceph_handle_notrace_create(dir, dentry);
854 ceph_mdsc_put_request(req);
857 ceph_init_inode_acls(d_inode(dentry), &acls);
860 ceph_release_acls_info(&acls);
864 static int ceph_create(struct inode *dir, struct dentry *dentry, umode_t mode,
867 return ceph_mknod(dir, dentry, mode, 0);
870 static int ceph_symlink(struct inode *dir, struct dentry *dentry,
873 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
874 struct ceph_mds_client *mdsc = fsc->mdsc;
875 struct ceph_mds_request *req;
878 if (ceph_snap(dir) != CEPH_NOSNAP)
881 dout("symlink in dir %p dentry %p to '%s'\n", dir, dentry, dest);
882 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SYMLINK, USE_AUTH_MDS);
887 req->r_path2 = kstrdup(dest, GFP_KERNEL);
890 ceph_mdsc_put_request(req);
894 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
895 req->r_dentry = dget(dentry);
897 req->r_dentry_drop = CEPH_CAP_FILE_SHARED | CEPH_CAP_AUTH_EXCL;
898 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
899 err = ceph_mdsc_do_request(mdsc, dir, req);
900 if (!err && !req->r_reply_info.head->is_dentry)
901 err = ceph_handle_notrace_create(dir, dentry);
902 ceph_mdsc_put_request(req);
909 static int ceph_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
911 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
912 struct ceph_mds_client *mdsc = fsc->mdsc;
913 struct ceph_mds_request *req;
914 struct ceph_acls_info acls = {};
918 if (ceph_snap(dir) == CEPH_SNAPDIR) {
919 /* mkdir .snap/foo is a MKSNAP */
920 op = CEPH_MDS_OP_MKSNAP;
921 dout("mksnap dir %p snap '%pd' dn %p\n", dir,
923 } else if (ceph_snap(dir) == CEPH_NOSNAP) {
924 dout("mkdir dir %p dn %p mode 0%ho\n", dir, dentry, mode);
925 op = CEPH_MDS_OP_MKDIR;
931 err = ceph_pre_init_acls(dir, &mode, &acls);
935 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
941 req->r_dentry = dget(dentry);
944 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
945 req->r_args.mkdir.mode = cpu_to_le32(mode);
946 req->r_dentry_drop = CEPH_CAP_FILE_SHARED | CEPH_CAP_AUTH_EXCL;
947 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
949 req->r_pagelist = acls.pagelist;
950 acls.pagelist = NULL;
952 err = ceph_mdsc_do_request(mdsc, dir, req);
954 !req->r_reply_info.head->is_target &&
955 !req->r_reply_info.head->is_dentry)
956 err = ceph_handle_notrace_create(dir, dentry);
957 ceph_mdsc_put_request(req);
960 ceph_init_inode_acls(d_inode(dentry), &acls);
963 ceph_release_acls_info(&acls);
967 static int ceph_link(struct dentry *old_dentry, struct inode *dir,
968 struct dentry *dentry)
970 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
971 struct ceph_mds_client *mdsc = fsc->mdsc;
972 struct ceph_mds_request *req;
975 if (ceph_snap(dir) != CEPH_NOSNAP)
978 dout("link in dir %p old_dentry %p dentry %p\n", dir,
980 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LINK, USE_AUTH_MDS);
985 req->r_dentry = dget(dentry);
987 req->r_old_dentry = dget(old_dentry);
989 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
990 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
991 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
992 /* release LINK_SHARED on source inode (mds will lock it) */
993 req->r_old_inode_drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL;
994 err = ceph_mdsc_do_request(mdsc, dir, req);
997 } else if (!req->r_reply_info.head->is_dentry) {
998 ihold(d_inode(old_dentry));
999 d_instantiate(dentry, d_inode(old_dentry));
1001 ceph_mdsc_put_request(req);
1006 * rmdir and unlink are differ only by the metadata op code
1008 static int ceph_unlink(struct inode *dir, struct dentry *dentry)
1010 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
1011 struct ceph_mds_client *mdsc = fsc->mdsc;
1012 struct inode *inode = d_inode(dentry);
1013 struct ceph_mds_request *req;
1017 if (ceph_snap(dir) == CEPH_SNAPDIR) {
1018 /* rmdir .snap/foo is RMSNAP */
1019 dout("rmsnap dir %p '%pd' dn %p\n", dir, dentry, dentry);
1020 op = CEPH_MDS_OP_RMSNAP;
1021 } else if (ceph_snap(dir) == CEPH_NOSNAP) {
1022 dout("unlink/rmdir dir %p dn %p inode %p\n",
1023 dir, dentry, inode);
1024 op = d_is_dir(dentry) ?
1025 CEPH_MDS_OP_RMDIR : CEPH_MDS_OP_UNLINK;
1028 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
1033 req->r_dentry = dget(dentry);
1034 req->r_num_caps = 2;
1035 req->r_parent = dir;
1036 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
1037 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
1038 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
1039 req->r_inode_drop = ceph_drop_caps_for_unlink(inode);
1040 err = ceph_mdsc_do_request(mdsc, dir, req);
1041 if (!err && !req->r_reply_info.head->is_dentry)
1043 ceph_mdsc_put_request(req);
1048 static int ceph_rename(struct inode *old_dir, struct dentry *old_dentry,
1049 struct inode *new_dir, struct dentry *new_dentry,
1052 struct ceph_fs_client *fsc = ceph_sb_to_client(old_dir->i_sb);
1053 struct ceph_mds_client *mdsc = fsc->mdsc;
1054 struct ceph_mds_request *req;
1055 int op = CEPH_MDS_OP_RENAME;
1061 if (ceph_snap(old_dir) != ceph_snap(new_dir))
1063 if (ceph_snap(old_dir) != CEPH_NOSNAP) {
1064 if (old_dir == new_dir && ceph_snap(old_dir) == CEPH_SNAPDIR)
1065 op = CEPH_MDS_OP_RENAMESNAP;
1069 dout("rename dir %p dentry %p to dir %p dentry %p\n",
1070 old_dir, old_dentry, new_dir, new_dentry);
1071 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
1073 return PTR_ERR(req);
1075 req->r_dentry = dget(new_dentry);
1076 req->r_num_caps = 2;
1077 req->r_old_dentry = dget(old_dentry);
1078 req->r_old_dentry_dir = old_dir;
1079 req->r_parent = new_dir;
1080 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
1081 req->r_old_dentry_drop = CEPH_CAP_FILE_SHARED;
1082 req->r_old_dentry_unless = CEPH_CAP_FILE_EXCL;
1083 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
1084 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
1085 /* release LINK_RDCACHE on source inode (mds will lock it) */
1086 req->r_old_inode_drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL;
1087 if (d_really_is_positive(new_dentry)) {
1089 ceph_drop_caps_for_unlink(d_inode(new_dentry));
1091 err = ceph_mdsc_do_request(mdsc, old_dir, req);
1092 if (!err && !req->r_reply_info.head->is_dentry) {
1094 * Normally d_move() is done by fill_trace (called by
1095 * do_request, above). If there is no trace, we need
1098 d_move(old_dentry, new_dentry);
1100 ceph_mdsc_put_request(req);
1105 * Ensure a dentry lease will no longer revalidate.
1107 void ceph_invalidate_dentry_lease(struct dentry *dentry)
1109 spin_lock(&dentry->d_lock);
1110 ceph_dentry(dentry)->time = jiffies;
1111 ceph_dentry(dentry)->lease_shared_gen = 0;
1112 spin_unlock(&dentry->d_lock);
1116 * Check if dentry lease is valid. If not, delete the lease. Try to
1117 * renew if the least is more than half up.
1119 static int dentry_lease_is_valid(struct dentry *dentry, unsigned int flags,
1122 struct ceph_dentry_info *di;
1123 struct ceph_mds_session *s;
1127 struct ceph_mds_session *session = NULL;
1130 spin_lock(&dentry->d_lock);
1131 di = ceph_dentry(dentry);
1132 if (di && di->lease_session) {
1133 s = di->lease_session;
1134 spin_lock(&s->s_gen_ttl_lock);
1137 spin_unlock(&s->s_gen_ttl_lock);
1139 if (di->lease_gen == gen &&
1140 time_before(jiffies, di->time) &&
1141 time_before(jiffies, ttl)) {
1143 if (di->lease_renew_after &&
1144 time_after(jiffies, di->lease_renew_after)) {
1146 * We should renew. If we're in RCU walk mode
1147 * though, we can't do that so just return
1150 if (flags & LOOKUP_RCU) {
1153 session = ceph_get_mds_session(s);
1154 seq = di->lease_seq;
1155 di->lease_renew_after = 0;
1156 di->lease_renew_from = jiffies;
1161 spin_unlock(&dentry->d_lock);
1164 ceph_mdsc_lease_send_msg(session, dir, dentry,
1165 CEPH_MDS_LEASE_RENEW, seq);
1166 ceph_put_mds_session(session);
1168 dout("dentry_lease_is_valid - dentry %p = %d\n", dentry, valid);
1173 * Check if directory-wide content lease/cap is valid.
1175 static int dir_lease_is_valid(struct inode *dir, struct dentry *dentry)
1177 struct ceph_inode_info *ci = ceph_inode(dir);
1178 struct ceph_dentry_info *di = ceph_dentry(dentry);
1181 spin_lock(&ci->i_ceph_lock);
1182 if (atomic_read(&ci->i_shared_gen) == di->lease_shared_gen)
1183 valid = __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1);
1184 spin_unlock(&ci->i_ceph_lock);
1185 dout("dir_lease_is_valid dir %p v%u dentry %p v%u = %d\n",
1186 dir, (unsigned)atomic_read(&ci->i_shared_gen),
1187 dentry, (unsigned)di->lease_shared_gen, valid);
1192 * Check if cached dentry can be trusted.
1194 static int ceph_d_revalidate(struct dentry *dentry, unsigned int flags)
1197 struct dentry *parent;
1200 if (flags & LOOKUP_RCU) {
1201 parent = READ_ONCE(dentry->d_parent);
1202 dir = d_inode_rcu(parent);
1206 parent = dget_parent(dentry);
1207 dir = d_inode(parent);
1210 dout("d_revalidate %p '%pd' inode %p offset %lld\n", dentry,
1211 dentry, d_inode(dentry), ceph_dentry(dentry)->offset);
1213 /* always trust cached snapped dentries, snapdir dentry */
1214 if (ceph_snap(dir) != CEPH_NOSNAP) {
1215 dout("d_revalidate %p '%pd' inode %p is SNAPPED\n", dentry,
1216 dentry, d_inode(dentry));
1218 } else if (d_really_is_positive(dentry) &&
1219 ceph_snap(d_inode(dentry)) == CEPH_SNAPDIR) {
1222 valid = dentry_lease_is_valid(dentry, flags, dir);
1223 if (valid == -ECHILD)
1225 if (valid || dir_lease_is_valid(dir, dentry)) {
1226 if (d_really_is_positive(dentry))
1227 valid = ceph_is_any_caps(d_inode(dentry));
1234 struct ceph_mds_client *mdsc =
1235 ceph_sb_to_client(dir->i_sb)->mdsc;
1236 struct ceph_mds_request *req;
1240 if (flags & LOOKUP_RCU)
1243 op = ceph_snap(dir) == CEPH_SNAPDIR ?
1244 CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP;
1245 req = ceph_mdsc_create_request(mdsc, op, USE_ANY_MDS);
1247 req->r_dentry = dget(dentry);
1248 req->r_num_caps = 2;
1249 req->r_parent = dir;
1251 mask = CEPH_STAT_CAP_INODE | CEPH_CAP_AUTH_SHARED;
1252 if (ceph_security_xattr_wanted(dir))
1253 mask |= CEPH_CAP_XATTR_SHARED;
1254 req->r_args.getattr.mask = cpu_to_le32(mask);
1256 err = ceph_mdsc_do_request(mdsc, NULL, req);
1259 if (d_really_is_positive(dentry) &&
1260 d_inode(dentry) == req->r_target_inode)
1264 if (d_really_is_negative(dentry))
1270 ceph_mdsc_put_request(req);
1271 dout("d_revalidate %p lookup result=%d\n",
1276 dout("d_revalidate %p %s\n", dentry, valid ? "valid" : "invalid");
1278 ceph_dentry_lru_touch(dentry);
1280 ceph_dir_clear_complete(dir);
1283 if (!(flags & LOOKUP_RCU))
1289 * Release our ceph_dentry_info.
1291 static void ceph_d_release(struct dentry *dentry)
1293 struct ceph_dentry_info *di = ceph_dentry(dentry);
1295 dout("d_release %p\n", dentry);
1296 ceph_dentry_lru_del(dentry);
1298 spin_lock(&dentry->d_lock);
1299 dentry->d_fsdata = NULL;
1300 spin_unlock(&dentry->d_lock);
1302 if (di->lease_session)
1303 ceph_put_mds_session(di->lease_session);
1304 kmem_cache_free(ceph_dentry_cachep, di);
1308 * When the VFS prunes a dentry from the cache, we need to clear the
1309 * complete flag on the parent directory.
1311 * Called under dentry->d_lock.
1313 static void ceph_d_prune(struct dentry *dentry)
1315 struct ceph_inode_info *dir_ci;
1316 struct ceph_dentry_info *di;
1318 dout("ceph_d_prune %pd %p\n", dentry, dentry);
1320 /* do we have a valid parent? */
1321 if (IS_ROOT(dentry))
1324 /* we hold d_lock, so d_parent is stable */
1325 dir_ci = ceph_inode(d_inode(dentry->d_parent));
1326 if (dir_ci->i_vino.snap == CEPH_SNAPDIR)
1329 /* who calls d_delete() should also disable dcache readdir */
1330 if (d_really_is_negative(dentry))
1333 /* d_fsdata does not get cleared until d_release */
1334 if (!d_unhashed(dentry)) {
1335 __ceph_dir_clear_complete(dir_ci);
1339 /* Disable dcache readdir just in case that someone called d_drop()
1340 * or d_invalidate(), but MDS didn't revoke CEPH_CAP_FILE_SHARED
1341 * properly (dcache readdir is still enabled) */
1342 di = ceph_dentry(dentry);
1343 if (di->offset > 0 &&
1344 di->lease_shared_gen == atomic_read(&dir_ci->i_shared_gen))
1345 __ceph_dir_clear_ordered(dir_ci);
1349 * read() on a dir. This weird interface hack only works if mounted
1350 * with '-o dirstat'.
1352 static ssize_t ceph_read_dir(struct file *file, char __user *buf, size_t size,
1355 struct ceph_file_info *cf = file->private_data;
1356 struct inode *inode = file_inode(file);
1357 struct ceph_inode_info *ci = ceph_inode(inode);
1359 const int bufsize = 1024;
1361 if (!ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb), DIRSTAT))
1364 if (!cf->dir_info) {
1365 cf->dir_info = kmalloc(bufsize, GFP_KERNEL);
1369 snprintf(cf->dir_info, bufsize,
1372 " subdirs: %20lld\n"
1373 "rentries: %20lld\n"
1375 " rsubdirs: %20lld\n"
1377 "rctime: %10ld.%09ld\n",
1378 ci->i_files + ci->i_subdirs,
1381 ci->i_rfiles + ci->i_rsubdirs,
1385 (long)ci->i_rctime.tv_sec,
1386 (long)ci->i_rctime.tv_nsec);
1389 if (*ppos >= cf->dir_info_len)
1391 size = min_t(unsigned, size, cf->dir_info_len-*ppos);
1392 left = copy_to_user(buf, cf->dir_info + *ppos, size);
1395 *ppos += (size - left);
1400 * We maintain a private dentry LRU.
1402 * FIXME: this needs to be changed to a per-mds lru to be useful.
1404 void ceph_dentry_lru_add(struct dentry *dn)
1406 struct ceph_dentry_info *di = ceph_dentry(dn);
1407 struct ceph_mds_client *mdsc;
1409 dout("dentry_lru_add %p %p '%pd'\n", di, dn, dn);
1410 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1411 spin_lock(&mdsc->dentry_lru_lock);
1412 list_add_tail(&di->lru, &mdsc->dentry_lru);
1414 spin_unlock(&mdsc->dentry_lru_lock);
1417 void ceph_dentry_lru_touch(struct dentry *dn)
1419 struct ceph_dentry_info *di = ceph_dentry(dn);
1420 struct ceph_mds_client *mdsc;
1422 dout("dentry_lru_touch %p %p '%pd' (offset %lld)\n", di, dn, dn,
1424 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1425 spin_lock(&mdsc->dentry_lru_lock);
1426 list_move_tail(&di->lru, &mdsc->dentry_lru);
1427 spin_unlock(&mdsc->dentry_lru_lock);
1430 void ceph_dentry_lru_del(struct dentry *dn)
1432 struct ceph_dentry_info *di = ceph_dentry(dn);
1433 struct ceph_mds_client *mdsc;
1435 dout("dentry_lru_del %p %p '%pd'\n", di, dn, dn);
1436 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1437 spin_lock(&mdsc->dentry_lru_lock);
1438 list_del_init(&di->lru);
1440 spin_unlock(&mdsc->dentry_lru_lock);
1444 * Return name hash for a given dentry. This is dependent on
1445 * the parent directory's hash function.
1447 unsigned ceph_dentry_hash(struct inode *dir, struct dentry *dn)
1449 struct ceph_inode_info *dci = ceph_inode(dir);
1451 switch (dci->i_dir_layout.dl_dir_hash) {
1452 case 0: /* for backward compat */
1453 case CEPH_STR_HASH_LINUX:
1454 return dn->d_name.hash;
1457 return ceph_str_hash(dci->i_dir_layout.dl_dir_hash,
1458 dn->d_name.name, dn->d_name.len);
1462 const struct file_operations ceph_dir_fops = {
1463 .read = ceph_read_dir,
1464 .iterate = ceph_readdir,
1465 .llseek = ceph_dir_llseek,
1467 .release = ceph_release,
1468 .unlocked_ioctl = ceph_ioctl,
1469 .fsync = ceph_fsync,
1472 const struct file_operations ceph_snapdir_fops = {
1473 .iterate = ceph_readdir,
1474 .llseek = ceph_dir_llseek,
1476 .release = ceph_release,
1479 const struct inode_operations ceph_dir_iops = {
1480 .lookup = ceph_lookup,
1481 .permission = ceph_permission,
1482 .getattr = ceph_getattr,
1483 .setattr = ceph_setattr,
1484 .listxattr = ceph_listxattr,
1485 .get_acl = ceph_get_acl,
1486 .set_acl = ceph_set_acl,
1487 .mknod = ceph_mknod,
1488 .symlink = ceph_symlink,
1489 .mkdir = ceph_mkdir,
1491 .unlink = ceph_unlink,
1492 .rmdir = ceph_unlink,
1493 .rename = ceph_rename,
1494 .create = ceph_create,
1495 .atomic_open = ceph_atomic_open,
1498 const struct inode_operations ceph_snapdir_iops = {
1499 .lookup = ceph_lookup,
1500 .permission = ceph_permission,
1501 .getattr = ceph_getattr,
1502 .mkdir = ceph_mkdir,
1503 .rmdir = ceph_unlink,
1504 .rename = ceph_rename,
1507 const struct dentry_operations ceph_dentry_ops = {
1508 .d_revalidate = ceph_d_revalidate,
1509 .d_release = ceph_d_release,
1510 .d_prune = ceph_d_prune,
1511 .d_init = ceph_d_init,