1 #include <linux/ceph/ceph_debug.h>
3 #include <linux/spinlock.h>
4 #include <linux/fs_struct.h>
5 #include <linux/namei.h>
6 #include <linux/slab.h>
7 #include <linux/sched.h>
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_file_info *fi, const char *name,
105 int len, unsigned next_offset)
107 char *buf = kmalloc(len+1, GFP_KERNEL);
110 kfree(fi->last_name);
112 memcpy(fi->last_name, name, len);
113 fi->last_name[len] = 0;
114 fi->next_offset = next_offset;
115 dout("note_last_dentry '%s'\n", fi->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_file_info *fi = 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, 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 fi->flags |= CEPH_F_ATEND;
228 if (IS_ERR(dentry)) {
229 err = PTR_ERR(dentry);
233 di = ceph_dentry(dentry);
234 spin_lock(&dentry->d_lock);
235 if (di->lease_shared_gen == shared_gen &&
236 d_really_is_positive(dentry) &&
237 fpos_cmp(ctx->pos, di->offset) <= 0) {
240 spin_unlock(&dentry->d_lock);
243 dout(" %llx dentry %p %pd %p\n", di->offset,
244 dentry, dentry, d_inode(dentry));
245 ctx->pos = di->offset;
246 if (!dir_emit(ctx, dentry->d_name.name,
248 ceph_translate_ino(dentry->d_sb,
249 d_inode(dentry)->i_ino),
250 d_inode(dentry)->i_mode >> 12)) {
265 ceph_readdir_cache_release(&cache_ctl);
268 di = ceph_dentry(last);
269 ret = note_last_dentry(fi, last->d_name.name, last->d_name.len,
270 fpos_off(di->offset) + 1);
278 static bool need_send_readdir(struct ceph_file_info *fi, loff_t pos)
280 if (!fi->last_readdir)
282 if (is_hash_order(pos))
283 return !ceph_frag_contains_value(fi->frag, fpos_hash(pos));
285 return fi->frag != fpos_frag(pos);
288 static int ceph_readdir(struct file *file, struct dir_context *ctx)
290 struct ceph_file_info *fi = file->private_data;
291 struct inode *inode = file_inode(file);
292 struct ceph_inode_info *ci = ceph_inode(inode);
293 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
294 struct ceph_mds_client *mdsc = fsc->mdsc;
298 struct ceph_mds_reply_info_parsed *rinfo;
300 dout("readdir %p file %p pos %llx\n", inode, file, ctx->pos);
301 if (fi->flags & CEPH_F_ATEND)
304 /* always start with . and .. */
306 dout("readdir off 0 -> '.'\n");
307 if (!dir_emit(ctx, ".", 1,
308 ceph_translate_ino(inode->i_sb, inode->i_ino),
309 inode->i_mode >> 12))
314 ino_t ino = parent_ino(file->f_path.dentry);
315 dout("readdir off 1 -> '..'\n");
316 if (!dir_emit(ctx, "..", 2,
317 ceph_translate_ino(inode->i_sb, ino),
318 inode->i_mode >> 12))
323 /* can we use the dcache? */
324 spin_lock(&ci->i_ceph_lock);
325 if (ceph_test_mount_opt(fsc, DCACHE) &&
326 !ceph_test_mount_opt(fsc, NOASYNCREADDIR) &&
327 ceph_snap(inode) != CEPH_SNAPDIR &&
328 __ceph_dir_is_complete_ordered(ci) &&
329 __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1)) {
330 u32 shared_gen = ci->i_shared_gen;
331 spin_unlock(&ci->i_ceph_lock);
332 err = __dcache_readdir(file, ctx, shared_gen);
336 spin_unlock(&ci->i_ceph_lock);
339 /* proceed with a normal readdir */
341 /* do we have the correct frag content buffered? */
342 if (need_send_readdir(fi, ctx->pos)) {
343 struct ceph_mds_request *req;
345 int op = ceph_snap(inode) == CEPH_SNAPDIR ?
346 CEPH_MDS_OP_LSSNAP : CEPH_MDS_OP_READDIR;
348 /* discard old result, if any */
349 if (fi->last_readdir) {
350 ceph_mdsc_put_request(fi->last_readdir);
351 fi->last_readdir = NULL;
354 if (is_hash_order(ctx->pos)) {
355 frag = ceph_choose_frag(ci, fpos_hash(ctx->pos),
358 frag = fpos_frag(ctx->pos);
361 dout("readdir fetching %llx.%llx frag %x offset '%s'\n",
362 ceph_vinop(inode), frag, fi->last_name);
363 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
366 err = ceph_alloc_readdir_reply_buffer(req, inode);
368 ceph_mdsc_put_request(req);
371 /* hints to request -> mds selection code */
372 req->r_direct_mode = USE_AUTH_MDS;
373 req->r_direct_hash = ceph_frag_value(frag);
374 __set_bit(CEPH_MDS_R_DIRECT_IS_HASH, &req->r_req_flags);
376 req->r_path2 = kstrdup(fi->last_name, GFP_KERNEL);
378 ceph_mdsc_put_request(req);
381 } else if (is_hash_order(ctx->pos)) {
382 req->r_args.readdir.offset_hash =
383 cpu_to_le32(fpos_hash(ctx->pos));
386 req->r_dir_release_cnt = fi->dir_release_count;
387 req->r_dir_ordered_cnt = fi->dir_ordered_count;
388 req->r_readdir_cache_idx = fi->readdir_cache_idx;
389 req->r_readdir_offset = fi->next_offset;
390 req->r_args.readdir.frag = cpu_to_le32(frag);
391 req->r_args.readdir.flags =
392 cpu_to_le16(CEPH_READDIR_REPLY_BITFLAGS);
394 req->r_inode = inode;
396 req->r_dentry = dget(file->f_path.dentry);
397 err = ceph_mdsc_do_request(mdsc, NULL, req);
399 ceph_mdsc_put_request(req);
402 dout("readdir got and parsed readdir result=%d on "
403 "frag %x, end=%d, complete=%d, hash_order=%d\n",
405 (int)req->r_reply_info.dir_end,
406 (int)req->r_reply_info.dir_complete,
407 (int)req->r_reply_info.hash_order);
409 rinfo = &req->r_reply_info;
410 if (le32_to_cpu(rinfo->dir_dir->frag) != frag) {
411 frag = le32_to_cpu(rinfo->dir_dir->frag);
412 if (!rinfo->hash_order) {
413 fi->next_offset = req->r_readdir_offset;
414 /* adjust ctx->pos to beginning of frag */
415 ctx->pos = ceph_make_fpos(frag,
422 fi->last_readdir = req;
424 if (test_bit(CEPH_MDS_R_DID_PREPOPULATE, &req->r_req_flags)) {
425 fi->readdir_cache_idx = req->r_readdir_cache_idx;
426 if (fi->readdir_cache_idx < 0) {
427 /* preclude from marking dir ordered */
428 fi->dir_ordered_count = 0;
429 } else if (ceph_frag_is_leftmost(frag) &&
430 fi->next_offset == 2) {
431 /* note dir version at start of readdir so
432 * we can tell if any dentries get dropped */
433 fi->dir_release_count = req->r_dir_release_cnt;
434 fi->dir_ordered_count = req->r_dir_ordered_cnt;
437 dout("readdir !did_prepopulate");
438 /* disable readdir cache */
439 fi->readdir_cache_idx = -1;
440 /* preclude from marking dir complete */
441 fi->dir_release_count = 0;
444 /* note next offset and last dentry name */
445 if (rinfo->dir_nr > 0) {
446 struct ceph_mds_reply_dir_entry *rde =
447 rinfo->dir_entries + (rinfo->dir_nr-1);
448 unsigned next_offset = req->r_reply_info.dir_end ?
449 2 : (fpos_off(rde->offset) + 1);
450 err = note_last_dentry(fi, rde->name, rde->name_len,
454 } else if (req->r_reply_info.dir_end) {
460 rinfo = &fi->last_readdir->r_reply_info;
461 dout("readdir frag %x num %d pos %llx chunk first %llx\n",
462 fi->frag, rinfo->dir_nr, ctx->pos,
463 rinfo->dir_nr ? rinfo->dir_entries[0].offset : 0LL);
466 /* search start position */
467 if (rinfo->dir_nr > 0) {
468 int step, nr = rinfo->dir_nr;
471 if (rinfo->dir_entries[i + step].offset < ctx->pos) {
479 for (; i < rinfo->dir_nr; i++) {
480 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
481 struct ceph_vino vino;
484 BUG_ON(rde->offset < ctx->pos);
486 ctx->pos = rde->offset;
487 dout("readdir (%d/%d) -> %llx '%.*s' %p\n",
488 i, rinfo->dir_nr, ctx->pos,
489 rde->name_len, rde->name, &rde->inode.in);
491 BUG_ON(!rde->inode.in);
492 ftype = le32_to_cpu(rde->inode.in->mode) >> 12;
493 vino.ino = le64_to_cpu(rde->inode.in->ino);
494 vino.snap = le64_to_cpu(rde->inode.in->snapid);
495 ino = ceph_vino_to_ino(vino);
497 if (!dir_emit(ctx, rde->name, rde->name_len,
498 ceph_translate_ino(inode->i_sb, ino), ftype)) {
499 dout("filldir stopping us...\n");
505 if (fi->next_offset > 2) {
506 ceph_mdsc_put_request(fi->last_readdir);
507 fi->last_readdir = NULL;
512 if (!ceph_frag_is_rightmost(fi->frag)) {
513 unsigned frag = ceph_frag_next(fi->frag);
514 if (is_hash_order(ctx->pos)) {
515 loff_t new_pos = ceph_make_fpos(ceph_frag_value(frag),
516 fi->next_offset, true);
517 if (new_pos > ctx->pos)
521 ctx->pos = ceph_make_fpos(frag, fi->next_offset, false);
522 kfree(fi->last_name);
523 fi->last_name = NULL;
525 dout("readdir next frag is %x\n", frag);
528 fi->flags |= CEPH_F_ATEND;
531 * if dir_release_count still matches the dir, no dentries
532 * were released during the whole readdir, and we should have
533 * the complete dir contents in our cache.
535 if (atomic64_read(&ci->i_release_count) == fi->dir_release_count) {
536 spin_lock(&ci->i_ceph_lock);
537 if (fi->dir_ordered_count == atomic64_read(&ci->i_ordered_count)) {
538 dout(" marking %p complete and ordered\n", inode);
539 /* use i_size to track number of entries in
541 BUG_ON(fi->readdir_cache_idx < 0);
542 i_size_write(inode, fi->readdir_cache_idx *
543 sizeof(struct dentry*));
545 dout(" marking %p complete\n", inode);
547 __ceph_dir_set_complete(ci, fi->dir_release_count,
548 fi->dir_ordered_count);
549 spin_unlock(&ci->i_ceph_lock);
552 dout("readdir %p file %p done.\n", inode, file);
556 static void reset_readdir(struct ceph_file_info *fi)
558 if (fi->last_readdir) {
559 ceph_mdsc_put_request(fi->last_readdir);
560 fi->last_readdir = NULL;
562 kfree(fi->last_name);
563 fi->last_name = NULL;
564 fi->dir_release_count = 0;
565 fi->readdir_cache_idx = -1;
566 fi->next_offset = 2; /* compensate for . and .. */
567 fi->flags &= ~CEPH_F_ATEND;
571 * discard buffered readdir content on seekdir(0), or seek to new frag,
572 * or seek prior to current chunk
574 static bool need_reset_readdir(struct ceph_file_info *fi, loff_t new_pos)
576 struct ceph_mds_reply_info_parsed *rinfo;
580 if (is_hash_order(new_pos)) {
581 /* no need to reset last_name for a forward seek when
582 * dentries are sotred in hash order */
583 } else if (fi->frag != fpos_frag(new_pos)) {
586 rinfo = fi->last_readdir ? &fi->last_readdir->r_reply_info : NULL;
587 if (!rinfo || !rinfo->dir_nr)
589 chunk_offset = rinfo->dir_entries[0].offset;
590 return new_pos < chunk_offset ||
591 is_hash_order(new_pos) != is_hash_order(chunk_offset);
594 static loff_t ceph_dir_llseek(struct file *file, loff_t offset, int whence)
596 struct ceph_file_info *fi = file->private_data;
597 struct inode *inode = file->f_mapping->host;
604 offset += file->f_pos;
608 retval = -EOPNOTSUPP;
614 if (need_reset_readdir(fi, offset)) {
615 dout("dir_llseek dropping %p content\n", file);
617 } else if (is_hash_order(offset) && offset > file->f_pos) {
618 /* for hash offset, we don't know if a forward seek
619 * is within same frag */
620 fi->dir_release_count = 0;
621 fi->readdir_cache_idx = -1;
624 if (offset != file->f_pos) {
625 file->f_pos = offset;
627 fi->flags &= ~CEPH_F_ATEND;
637 * Handle lookups for the hidden .snap directory.
639 int ceph_handle_snapdir(struct ceph_mds_request *req,
640 struct dentry *dentry, int err)
642 struct ceph_fs_client *fsc = ceph_sb_to_client(dentry->d_sb);
643 struct inode *parent = d_inode(dentry->d_parent); /* we hold i_mutex */
646 if (err == -ENOENT &&
647 ceph_snap(parent) == CEPH_NOSNAP &&
648 strcmp(dentry->d_name.name,
649 fsc->mount_options->snapdir_name) == 0) {
650 struct inode *inode = ceph_get_snapdir(parent);
651 dout("ENOENT on snapdir %p '%pd', linking to snapdir %p\n",
652 dentry, dentry, inode);
653 BUG_ON(!d_unhashed(dentry));
654 d_add(dentry, inode);
661 * Figure out final result of a lookup/open request.
663 * Mainly, make sure we return the final req->r_dentry (if it already
664 * existed) in place of the original VFS-provided dentry when they
667 * Gracefully handle the case where the MDS replies with -ENOENT and
668 * no trace (which it may do, at its discretion, e.g., if it doesn't
669 * care to issue a lease on the negative dentry).
671 struct dentry *ceph_finish_lookup(struct ceph_mds_request *req,
672 struct dentry *dentry, int err)
674 if (err == -ENOENT) {
677 if (!req->r_reply_info.head->is_dentry) {
678 dout("ENOENT and no trace, dentry %p inode %p\n",
679 dentry, d_inode(dentry));
680 if (d_really_is_positive(dentry)) {
689 dentry = ERR_PTR(err);
690 else if (dentry != req->r_dentry)
691 dentry = dget(req->r_dentry); /* we got spliced */
697 static bool is_root_ceph_dentry(struct inode *inode, struct dentry *dentry)
699 return ceph_ino(inode) == CEPH_INO_ROOT &&
700 strncmp(dentry->d_name.name, ".ceph", 5) == 0;
704 * Look up a single dir entry. If there is a lookup intent, inform
705 * the MDS so that it gets our 'caps wanted' value in a single op.
707 static struct dentry *ceph_lookup(struct inode *dir, struct dentry *dentry,
710 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
711 struct ceph_mds_client *mdsc = fsc->mdsc;
712 struct ceph_mds_request *req;
717 dout("lookup %p dentry %p '%pd'\n",
718 dir, dentry, dentry);
720 if (dentry->d_name.len > NAME_MAX)
721 return ERR_PTR(-ENAMETOOLONG);
723 /* can we conclude ENOENT locally? */
724 if (d_really_is_negative(dentry)) {
725 struct ceph_inode_info *ci = ceph_inode(dir);
726 struct ceph_dentry_info *di = ceph_dentry(dentry);
728 spin_lock(&ci->i_ceph_lock);
729 dout(" dir %p flags are %d\n", dir, ci->i_ceph_flags);
730 if (strncmp(dentry->d_name.name,
731 fsc->mount_options->snapdir_name,
732 dentry->d_name.len) &&
733 !is_root_ceph_dentry(dir, dentry) &&
734 ceph_test_mount_opt(fsc, DCACHE) &&
735 __ceph_dir_is_complete(ci) &&
736 (__ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1))) {
737 spin_unlock(&ci->i_ceph_lock);
738 dout(" dir %p complete, -ENOENT\n", dir);
740 di->lease_shared_gen = ci->i_shared_gen;
743 spin_unlock(&ci->i_ceph_lock);
746 op = ceph_snap(dir) == CEPH_SNAPDIR ?
747 CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP;
748 req = ceph_mdsc_create_request(mdsc, op, USE_ANY_MDS);
750 return ERR_CAST(req);
751 req->r_dentry = dget(dentry);
754 mask = CEPH_STAT_CAP_INODE | CEPH_CAP_AUTH_SHARED;
755 if (ceph_security_xattr_wanted(dir))
756 mask |= CEPH_CAP_XATTR_SHARED;
757 req->r_args.getattr.mask = cpu_to_le32(mask);
760 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
761 err = ceph_mdsc_do_request(mdsc, NULL, req);
762 err = ceph_handle_snapdir(req, dentry, err);
763 dentry = ceph_finish_lookup(req, dentry, err);
764 ceph_mdsc_put_request(req); /* will dput(dentry) */
765 dout("lookup result=%p\n", dentry);
770 * If we do a create but get no trace back from the MDS, follow up with
771 * a lookup (the VFS expects us to link up the provided dentry).
773 int ceph_handle_notrace_create(struct inode *dir, struct dentry *dentry)
775 struct dentry *result = ceph_lookup(dir, dentry, 0);
777 if (result && !IS_ERR(result)) {
779 * We created the item, then did a lookup, and found
780 * it was already linked to another inode we already
781 * had in our cache (and thus got spliced). To not
782 * confuse VFS (especially when inode is a directory),
783 * we don't link our dentry to that inode, return an
786 * This event should be rare and it happens only when
787 * we talk to old MDS. Recent MDS does not send traceless
788 * reply for request that creates new inode.
793 return PTR_ERR(result);
796 static int ceph_mknod(struct inode *dir, struct dentry *dentry,
797 umode_t mode, dev_t rdev)
799 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
800 struct ceph_mds_client *mdsc = fsc->mdsc;
801 struct ceph_mds_request *req;
802 struct ceph_acls_info acls = {};
805 if (ceph_snap(dir) != CEPH_NOSNAP)
808 err = ceph_pre_init_acls(dir, &mode, &acls);
812 dout("mknod in dir %p dentry %p mode 0%ho rdev %d\n",
813 dir, dentry, mode, rdev);
814 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_MKNOD, USE_AUTH_MDS);
819 req->r_dentry = dget(dentry);
822 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
823 req->r_args.mknod.mode = cpu_to_le32(mode);
824 req->r_args.mknod.rdev = cpu_to_le32(rdev);
825 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
826 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
828 req->r_pagelist = acls.pagelist;
829 acls.pagelist = NULL;
831 err = ceph_mdsc_do_request(mdsc, dir, req);
832 if (!err && !req->r_reply_info.head->is_dentry)
833 err = ceph_handle_notrace_create(dir, dentry);
834 ceph_mdsc_put_request(req);
837 ceph_init_inode_acls(d_inode(dentry), &acls);
840 ceph_release_acls_info(&acls);
844 static int ceph_create(struct inode *dir, struct dentry *dentry, umode_t mode,
847 return ceph_mknod(dir, dentry, mode, 0);
850 static int ceph_symlink(struct inode *dir, struct dentry *dentry,
853 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
854 struct ceph_mds_client *mdsc = fsc->mdsc;
855 struct ceph_mds_request *req;
858 if (ceph_snap(dir) != CEPH_NOSNAP)
861 dout("symlink in dir %p dentry %p to '%s'\n", dir, dentry, dest);
862 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SYMLINK, USE_AUTH_MDS);
867 req->r_path2 = kstrdup(dest, GFP_KERNEL);
870 ceph_mdsc_put_request(req);
874 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
875 req->r_dentry = dget(dentry);
877 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
878 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
879 err = ceph_mdsc_do_request(mdsc, dir, req);
880 if (!err && !req->r_reply_info.head->is_dentry)
881 err = ceph_handle_notrace_create(dir, dentry);
882 ceph_mdsc_put_request(req);
889 static int ceph_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
891 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
892 struct ceph_mds_client *mdsc = fsc->mdsc;
893 struct ceph_mds_request *req;
894 struct ceph_acls_info acls = {};
898 if (ceph_snap(dir) == CEPH_SNAPDIR) {
899 /* mkdir .snap/foo is a MKSNAP */
900 op = CEPH_MDS_OP_MKSNAP;
901 dout("mksnap dir %p snap '%pd' dn %p\n", dir,
903 } else if (ceph_snap(dir) == CEPH_NOSNAP) {
904 dout("mkdir dir %p dn %p mode 0%ho\n", dir, dentry, mode);
905 op = CEPH_MDS_OP_MKDIR;
911 err = ceph_pre_init_acls(dir, &mode, &acls);
915 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
921 req->r_dentry = dget(dentry);
924 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
925 req->r_args.mkdir.mode = cpu_to_le32(mode);
926 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
927 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
929 req->r_pagelist = acls.pagelist;
930 acls.pagelist = NULL;
932 err = ceph_mdsc_do_request(mdsc, dir, req);
934 !req->r_reply_info.head->is_target &&
935 !req->r_reply_info.head->is_dentry)
936 err = ceph_handle_notrace_create(dir, dentry);
937 ceph_mdsc_put_request(req);
940 ceph_init_inode_acls(d_inode(dentry), &acls);
943 ceph_release_acls_info(&acls);
947 static int ceph_link(struct dentry *old_dentry, struct inode *dir,
948 struct dentry *dentry)
950 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
951 struct ceph_mds_client *mdsc = fsc->mdsc;
952 struct ceph_mds_request *req;
955 if (ceph_snap(dir) != CEPH_NOSNAP)
958 dout("link in dir %p old_dentry %p dentry %p\n", dir,
960 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LINK, USE_AUTH_MDS);
965 req->r_dentry = dget(dentry);
967 req->r_old_dentry = dget(old_dentry);
969 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
970 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
971 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
972 /* release LINK_SHARED on source inode (mds will lock it) */
973 req->r_old_inode_drop = CEPH_CAP_LINK_SHARED;
974 err = ceph_mdsc_do_request(mdsc, dir, req);
977 } else if (!req->r_reply_info.head->is_dentry) {
978 ihold(d_inode(old_dentry));
979 d_instantiate(dentry, d_inode(old_dentry));
981 ceph_mdsc_put_request(req);
986 * For a soon-to-be unlinked file, drop the AUTH_RDCACHE caps. If it
987 * looks like the link count will hit 0, drop any other caps (other
988 * than PIN) we don't specifically want (due to the file still being
991 static int drop_caps_for_unlink(struct inode *inode)
993 struct ceph_inode_info *ci = ceph_inode(inode);
994 int drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL;
996 spin_lock(&ci->i_ceph_lock);
997 if (inode->i_nlink == 1) {
998 drop |= ~(__ceph_caps_wanted(ci) | CEPH_CAP_PIN);
999 ci->i_ceph_flags |= CEPH_I_NODELAY;
1001 spin_unlock(&ci->i_ceph_lock);
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 = 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;
1087 if (d_really_is_positive(new_dentry))
1088 req->r_inode_drop = drop_caps_for_unlink(d_inode(new_dentry));
1089 err = ceph_mdsc_do_request(mdsc, old_dir, req);
1090 if (!err && !req->r_reply_info.head->is_dentry) {
1092 * Normally d_move() is done by fill_trace (called by
1093 * do_request, above). If there is no trace, we need
1097 /* d_move screws up sibling dentries' offsets */
1098 ceph_dir_clear_complete(old_dir);
1099 ceph_dir_clear_complete(new_dir);
1101 d_move(old_dentry, new_dentry);
1103 /* ensure target dentry is invalidated, despite
1104 rehashing bug in vfs_rename_dir */
1105 ceph_invalidate_dentry_lease(new_dentry);
1107 ceph_mdsc_put_request(req);
1112 * Ensure a dentry lease will no longer revalidate.
1114 void ceph_invalidate_dentry_lease(struct dentry *dentry)
1116 spin_lock(&dentry->d_lock);
1117 ceph_dentry(dentry)->time = jiffies;
1118 ceph_dentry(dentry)->lease_shared_gen = 0;
1119 spin_unlock(&dentry->d_lock);
1123 * Check if dentry lease is valid. If not, delete the lease. Try to
1124 * renew if the least is more than half up.
1126 static int dentry_lease_is_valid(struct dentry *dentry, unsigned int flags,
1129 struct ceph_dentry_info *di;
1130 struct ceph_mds_session *s;
1134 struct ceph_mds_session *session = NULL;
1137 spin_lock(&dentry->d_lock);
1138 di = ceph_dentry(dentry);
1139 if (di && di->lease_session) {
1140 s = di->lease_session;
1141 spin_lock(&s->s_gen_ttl_lock);
1144 spin_unlock(&s->s_gen_ttl_lock);
1146 if (di->lease_gen == gen &&
1147 time_before(jiffies, di->time) &&
1148 time_before(jiffies, ttl)) {
1150 if (di->lease_renew_after &&
1151 time_after(jiffies, di->lease_renew_after)) {
1153 * We should renew. If we're in RCU walk mode
1154 * though, we can't do that so just return
1157 if (flags & LOOKUP_RCU) {
1160 session = ceph_get_mds_session(s);
1161 seq = di->lease_seq;
1162 di->lease_renew_after = 0;
1163 di->lease_renew_from = jiffies;
1168 spin_unlock(&dentry->d_lock);
1171 ceph_mdsc_lease_send_msg(session, dir, dentry,
1172 CEPH_MDS_LEASE_RENEW, seq);
1173 ceph_put_mds_session(session);
1175 dout("dentry_lease_is_valid - dentry %p = %d\n", dentry, valid);
1180 * Check if directory-wide content lease/cap is valid.
1182 static int dir_lease_is_valid(struct inode *dir, struct dentry *dentry)
1184 struct ceph_inode_info *ci = ceph_inode(dir);
1185 struct ceph_dentry_info *di = ceph_dentry(dentry);
1188 spin_lock(&ci->i_ceph_lock);
1189 if (ci->i_shared_gen == di->lease_shared_gen)
1190 valid = __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1);
1191 spin_unlock(&ci->i_ceph_lock);
1192 dout("dir_lease_is_valid dir %p v%u dentry %p v%u = %d\n",
1193 dir, (unsigned)ci->i_shared_gen, dentry,
1194 (unsigned)di->lease_shared_gen, valid);
1199 * Check if cached dentry can be trusted.
1201 static int ceph_d_revalidate(struct dentry *dentry, unsigned int flags)
1204 struct dentry *parent;
1207 if (flags & LOOKUP_RCU) {
1208 parent = READ_ONCE(dentry->d_parent);
1209 dir = d_inode_rcu(parent);
1213 parent = dget_parent(dentry);
1214 dir = d_inode(parent);
1217 dout("d_revalidate %p '%pd' inode %p offset %lld\n", dentry,
1218 dentry, d_inode(dentry), ceph_dentry(dentry)->offset);
1220 /* always trust cached snapped dentries, snapdir dentry */
1221 if (ceph_snap(dir) != CEPH_NOSNAP) {
1222 dout("d_revalidate %p '%pd' inode %p is SNAPPED\n", dentry,
1223 dentry, d_inode(dentry));
1225 } else if (d_really_is_positive(dentry) &&
1226 ceph_snap(d_inode(dentry)) == CEPH_SNAPDIR) {
1229 valid = dentry_lease_is_valid(dentry, flags, dir);
1230 if (valid == -ECHILD)
1232 if (valid || dir_lease_is_valid(dir, dentry)) {
1233 if (d_really_is_positive(dentry))
1234 valid = ceph_is_any_caps(d_inode(dentry));
1241 struct ceph_mds_client *mdsc =
1242 ceph_sb_to_client(dir->i_sb)->mdsc;
1243 struct ceph_mds_request *req;
1247 if (flags & LOOKUP_RCU)
1250 op = ceph_snap(dir) == CEPH_SNAPDIR ?
1251 CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP;
1252 req = ceph_mdsc_create_request(mdsc, op, USE_ANY_MDS);
1254 req->r_dentry = dget(dentry);
1255 req->r_num_caps = 2;
1256 req->r_parent = dir;
1258 mask = CEPH_STAT_CAP_INODE | CEPH_CAP_AUTH_SHARED;
1259 if (ceph_security_xattr_wanted(dir))
1260 mask |= CEPH_CAP_XATTR_SHARED;
1261 req->r_args.getattr.mask = cpu_to_le32(mask);
1263 err = ceph_mdsc_do_request(mdsc, NULL, req);
1266 if (d_really_is_positive(dentry) &&
1267 d_inode(dentry) == req->r_target_inode)
1271 if (d_really_is_negative(dentry))
1277 ceph_mdsc_put_request(req);
1278 dout("d_revalidate %p lookup result=%d\n",
1283 dout("d_revalidate %p %s\n", dentry, valid ? "valid" : "invalid");
1285 ceph_dentry_lru_touch(dentry);
1287 ceph_dir_clear_complete(dir);
1290 if (!(flags & LOOKUP_RCU))
1296 * Release our ceph_dentry_info.
1298 static void ceph_d_release(struct dentry *dentry)
1300 struct ceph_dentry_info *di = ceph_dentry(dentry);
1302 dout("d_release %p\n", dentry);
1303 ceph_dentry_lru_del(dentry);
1305 spin_lock(&dentry->d_lock);
1306 dentry->d_fsdata = NULL;
1307 spin_unlock(&dentry->d_lock);
1309 if (di->lease_session)
1310 ceph_put_mds_session(di->lease_session);
1311 kmem_cache_free(ceph_dentry_cachep, di);
1315 * When the VFS prunes a dentry from the cache, we need to clear the
1316 * complete flag on the parent directory.
1318 * Called under dentry->d_lock.
1320 static void ceph_d_prune(struct dentry *dentry)
1322 dout("ceph_d_prune %p\n", dentry);
1324 /* do we have a valid parent? */
1325 if (IS_ROOT(dentry))
1328 /* if we are not hashed, we don't affect dir's completeness */
1329 if (d_unhashed(dentry))
1332 if (ceph_snap(d_inode(dentry->d_parent)) == CEPH_SNAPDIR)
1336 * we hold d_lock, so d_parent is stable, and d_fsdata is never
1337 * cleared until d_release
1339 ceph_dir_clear_complete(d_inode(dentry->d_parent));
1343 * read() on a dir. This weird interface hack only works if mounted
1344 * with '-o dirstat'.
1346 static ssize_t ceph_read_dir(struct file *file, char __user *buf, size_t size,
1349 struct ceph_file_info *cf = file->private_data;
1350 struct inode *inode = file_inode(file);
1351 struct ceph_inode_info *ci = ceph_inode(inode);
1353 const int bufsize = 1024;
1355 if (!ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb), DIRSTAT))
1358 if (!cf->dir_info) {
1359 cf->dir_info = kmalloc(bufsize, GFP_KERNEL);
1363 snprintf(cf->dir_info, bufsize,
1366 " subdirs: %20lld\n"
1367 "rentries: %20lld\n"
1369 " rsubdirs: %20lld\n"
1371 "rctime: %10ld.%09ld\n",
1372 ci->i_files + ci->i_subdirs,
1375 ci->i_rfiles + ci->i_rsubdirs,
1379 (long)ci->i_rctime.tv_sec,
1380 (long)ci->i_rctime.tv_nsec);
1383 if (*ppos >= cf->dir_info_len)
1385 size = min_t(unsigned, size, cf->dir_info_len-*ppos);
1386 left = copy_to_user(buf, cf->dir_info + *ppos, size);
1389 *ppos += (size - left);
1394 * We maintain a private dentry LRU.
1396 * FIXME: this needs to be changed to a per-mds lru to be useful.
1398 void ceph_dentry_lru_add(struct dentry *dn)
1400 struct ceph_dentry_info *di = ceph_dentry(dn);
1401 struct ceph_mds_client *mdsc;
1403 dout("dentry_lru_add %p %p '%pd'\n", di, dn, dn);
1404 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1405 spin_lock(&mdsc->dentry_lru_lock);
1406 list_add_tail(&di->lru, &mdsc->dentry_lru);
1408 spin_unlock(&mdsc->dentry_lru_lock);
1411 void ceph_dentry_lru_touch(struct dentry *dn)
1413 struct ceph_dentry_info *di = ceph_dentry(dn);
1414 struct ceph_mds_client *mdsc;
1416 dout("dentry_lru_touch %p %p '%pd' (offset %lld)\n", di, dn, dn,
1418 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1419 spin_lock(&mdsc->dentry_lru_lock);
1420 list_move_tail(&di->lru, &mdsc->dentry_lru);
1421 spin_unlock(&mdsc->dentry_lru_lock);
1424 void ceph_dentry_lru_del(struct dentry *dn)
1426 struct ceph_dentry_info *di = ceph_dentry(dn);
1427 struct ceph_mds_client *mdsc;
1429 dout("dentry_lru_del %p %p '%pd'\n", di, dn, dn);
1430 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1431 spin_lock(&mdsc->dentry_lru_lock);
1432 list_del_init(&di->lru);
1434 spin_unlock(&mdsc->dentry_lru_lock);
1438 * Return name hash for a given dentry. This is dependent on
1439 * the parent directory's hash function.
1441 unsigned ceph_dentry_hash(struct inode *dir, struct dentry *dn)
1443 struct ceph_inode_info *dci = ceph_inode(dir);
1445 switch (dci->i_dir_layout.dl_dir_hash) {
1446 case 0: /* for backward compat */
1447 case CEPH_STR_HASH_LINUX:
1448 return dn->d_name.hash;
1451 return ceph_str_hash(dci->i_dir_layout.dl_dir_hash,
1452 dn->d_name.name, dn->d_name.len);
1456 const struct file_operations ceph_dir_fops = {
1457 .read = ceph_read_dir,
1458 .iterate = ceph_readdir,
1459 .llseek = ceph_dir_llseek,
1461 .release = ceph_release,
1462 .unlocked_ioctl = ceph_ioctl,
1463 .fsync = ceph_fsync,
1466 const struct file_operations ceph_snapdir_fops = {
1467 .iterate = ceph_readdir,
1468 .llseek = ceph_dir_llseek,
1470 .release = ceph_release,
1473 const struct inode_operations ceph_dir_iops = {
1474 .lookup = ceph_lookup,
1475 .permission = ceph_permission,
1476 .getattr = ceph_getattr,
1477 .setattr = ceph_setattr,
1478 .listxattr = ceph_listxattr,
1479 .get_acl = ceph_get_acl,
1480 .set_acl = ceph_set_acl,
1481 .mknod = ceph_mknod,
1482 .symlink = ceph_symlink,
1483 .mkdir = ceph_mkdir,
1485 .unlink = ceph_unlink,
1486 .rmdir = ceph_unlink,
1487 .rename = ceph_rename,
1488 .create = ceph_create,
1489 .atomic_open = ceph_atomic_open,
1492 const struct inode_operations ceph_snapdir_iops = {
1493 .lookup = ceph_lookup,
1494 .permission = ceph_permission,
1495 .getattr = ceph_getattr,
1496 .mkdir = ceph_mkdir,
1497 .rmdir = ceph_unlink,
1498 .rename = ceph_rename,
1501 const struct dentry_operations ceph_dentry_ops = {
1502 .d_revalidate = ceph_d_revalidate,
1503 .d_release = ceph_d_release,
1504 .d_prune = ceph_d_prune,
1505 .d_init = ceph_d_init,