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);
382 req->r_dir_release_cnt = fi->dir_release_count;
383 req->r_dir_ordered_cnt = fi->dir_ordered_count;
384 req->r_readdir_cache_idx = fi->readdir_cache_idx;
385 req->r_readdir_offset = fi->next_offset;
386 req->r_args.readdir.frag = cpu_to_le32(frag);
387 req->r_args.readdir.flags =
388 cpu_to_le16(CEPH_READDIR_REPLY_BITFLAGS);
390 req->r_inode = inode;
392 req->r_dentry = dget(file->f_path.dentry);
393 err = ceph_mdsc_do_request(mdsc, NULL, req);
395 ceph_mdsc_put_request(req);
398 dout("readdir got and parsed readdir result=%d on "
399 "frag %x, end=%d, complete=%d, hash_order=%d\n",
401 (int)req->r_reply_info.dir_end,
402 (int)req->r_reply_info.dir_complete,
403 (int)req->r_reply_info.hash_order);
405 rinfo = &req->r_reply_info;
406 if (le32_to_cpu(rinfo->dir_dir->frag) != frag) {
407 frag = le32_to_cpu(rinfo->dir_dir->frag);
408 if (!rinfo->hash_order) {
409 fi->next_offset = req->r_readdir_offset;
410 /* adjust ctx->pos to beginning of frag */
411 ctx->pos = ceph_make_fpos(frag,
418 fi->last_readdir = req;
420 if (test_bit(CEPH_MDS_R_DID_PREPOPULATE, &req->r_req_flags)) {
421 fi->readdir_cache_idx = req->r_readdir_cache_idx;
422 if (fi->readdir_cache_idx < 0) {
423 /* preclude from marking dir ordered */
424 fi->dir_ordered_count = 0;
425 } else if (ceph_frag_is_leftmost(frag) &&
426 fi->next_offset == 2) {
427 /* note dir version at start of readdir so
428 * we can tell if any dentries get dropped */
429 fi->dir_release_count = req->r_dir_release_cnt;
430 fi->dir_ordered_count = req->r_dir_ordered_cnt;
433 dout("readdir !did_prepopulate");
434 /* disable readdir cache */
435 fi->readdir_cache_idx = -1;
436 /* preclude from marking dir complete */
437 fi->dir_release_count = 0;
440 /* note next offset and last dentry name */
441 if (rinfo->dir_nr > 0) {
442 struct ceph_mds_reply_dir_entry *rde =
443 rinfo->dir_entries + (rinfo->dir_nr-1);
444 unsigned next_offset = req->r_reply_info.dir_end ?
445 2 : (fpos_off(rde->offset) + 1);
446 err = note_last_dentry(fi, rde->name, rde->name_len,
450 } else if (req->r_reply_info.dir_end) {
456 rinfo = &fi->last_readdir->r_reply_info;
457 dout("readdir frag %x num %d pos %llx chunk first %llx\n",
458 fi->frag, rinfo->dir_nr, ctx->pos,
459 rinfo->dir_nr ? rinfo->dir_entries[0].offset : 0LL);
462 /* search start position */
463 if (rinfo->dir_nr > 0) {
464 int step, nr = rinfo->dir_nr;
467 if (rinfo->dir_entries[i + step].offset < ctx->pos) {
475 for (; i < rinfo->dir_nr; i++) {
476 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
477 struct ceph_vino vino;
480 BUG_ON(rde->offset < ctx->pos);
482 ctx->pos = rde->offset;
483 dout("readdir (%d/%d) -> %llx '%.*s' %p\n",
484 i, rinfo->dir_nr, ctx->pos,
485 rde->name_len, rde->name, &rde->inode.in);
487 BUG_ON(!rde->inode.in);
488 ftype = le32_to_cpu(rde->inode.in->mode) >> 12;
489 vino.ino = le64_to_cpu(rde->inode.in->ino);
490 vino.snap = le64_to_cpu(rde->inode.in->snapid);
491 ino = ceph_vino_to_ino(vino);
493 if (!dir_emit(ctx, rde->name, rde->name_len,
494 ceph_translate_ino(inode->i_sb, ino), ftype)) {
495 dout("filldir stopping us...\n");
501 if (fi->next_offset > 2) {
502 ceph_mdsc_put_request(fi->last_readdir);
503 fi->last_readdir = NULL;
508 if (!ceph_frag_is_rightmost(fi->frag)) {
509 unsigned frag = ceph_frag_next(fi->frag);
510 if (is_hash_order(ctx->pos)) {
511 loff_t new_pos = ceph_make_fpos(ceph_frag_value(frag),
512 fi->next_offset, true);
513 if (new_pos > ctx->pos)
517 ctx->pos = ceph_make_fpos(frag, fi->next_offset, false);
518 kfree(fi->last_name);
519 fi->last_name = NULL;
521 dout("readdir next frag is %x\n", frag);
524 fi->flags |= CEPH_F_ATEND;
527 * if dir_release_count still matches the dir, no dentries
528 * were released during the whole readdir, and we should have
529 * the complete dir contents in our cache.
531 if (atomic64_read(&ci->i_release_count) == fi->dir_release_count) {
532 spin_lock(&ci->i_ceph_lock);
533 if (fi->dir_ordered_count == atomic64_read(&ci->i_ordered_count)) {
534 dout(" marking %p complete and ordered\n", inode);
535 /* use i_size to track number of entries in
537 BUG_ON(fi->readdir_cache_idx < 0);
538 i_size_write(inode, fi->readdir_cache_idx *
539 sizeof(struct dentry*));
541 dout(" marking %p complete\n", inode);
543 __ceph_dir_set_complete(ci, fi->dir_release_count,
544 fi->dir_ordered_count);
545 spin_unlock(&ci->i_ceph_lock);
548 dout("readdir %p file %p done.\n", inode, file);
552 static void reset_readdir(struct ceph_file_info *fi)
554 if (fi->last_readdir) {
555 ceph_mdsc_put_request(fi->last_readdir);
556 fi->last_readdir = NULL;
558 kfree(fi->last_name);
559 fi->last_name = NULL;
560 fi->dir_release_count = 0;
561 fi->readdir_cache_idx = -1;
562 fi->next_offset = 2; /* compensate for . and .. */
563 fi->flags &= ~CEPH_F_ATEND;
567 * discard buffered readdir content on seekdir(0), or seek to new frag,
568 * or seek prior to current chunk
570 static bool need_reset_readdir(struct ceph_file_info *fi, loff_t new_pos)
572 struct ceph_mds_reply_info_parsed *rinfo;
576 if (is_hash_order(new_pos)) {
577 /* no need to reset last_name for a forward seek when
578 * dentries are sotred in hash order */
579 } else if (fi->frag != fpos_frag(new_pos)) {
582 rinfo = fi->last_readdir ? &fi->last_readdir->r_reply_info : NULL;
583 if (!rinfo || !rinfo->dir_nr)
585 chunk_offset = rinfo->dir_entries[0].offset;
586 return new_pos < chunk_offset ||
587 is_hash_order(new_pos) != is_hash_order(chunk_offset);
590 static loff_t ceph_dir_llseek(struct file *file, loff_t offset, int whence)
592 struct ceph_file_info *fi = file->private_data;
593 struct inode *inode = file->f_mapping->host;
600 offset += file->f_pos;
604 retval = -EOPNOTSUPP;
610 if (need_reset_readdir(fi, offset)) {
611 dout("dir_llseek dropping %p content\n", file);
613 } else if (is_hash_order(offset) && offset > file->f_pos) {
614 /* for hash offset, we don't know if a forward seek
615 * is within same frag */
616 fi->dir_release_count = 0;
617 fi->readdir_cache_idx = -1;
620 if (offset != file->f_pos) {
621 file->f_pos = offset;
623 fi->flags &= ~CEPH_F_ATEND;
633 * Handle lookups for the hidden .snap directory.
635 int ceph_handle_snapdir(struct ceph_mds_request *req,
636 struct dentry *dentry, int err)
638 struct ceph_fs_client *fsc = ceph_sb_to_client(dentry->d_sb);
639 struct inode *parent = d_inode(dentry->d_parent); /* we hold i_mutex */
642 if (err == -ENOENT &&
643 ceph_snap(parent) == CEPH_NOSNAP &&
644 strcmp(dentry->d_name.name,
645 fsc->mount_options->snapdir_name) == 0) {
646 struct inode *inode = ceph_get_snapdir(parent);
647 dout("ENOENT on snapdir %p '%pd', linking to snapdir %p\n",
648 dentry, dentry, inode);
649 BUG_ON(!d_unhashed(dentry));
650 d_add(dentry, inode);
657 * Figure out final result of a lookup/open request.
659 * Mainly, make sure we return the final req->r_dentry (if it already
660 * existed) in place of the original VFS-provided dentry when they
663 * Gracefully handle the case where the MDS replies with -ENOENT and
664 * no trace (which it may do, at its discretion, e.g., if it doesn't
665 * care to issue a lease on the negative dentry).
667 struct dentry *ceph_finish_lookup(struct ceph_mds_request *req,
668 struct dentry *dentry, int err)
670 if (err == -ENOENT) {
673 if (!req->r_reply_info.head->is_dentry) {
674 dout("ENOENT and no trace, dentry %p inode %p\n",
675 dentry, d_inode(dentry));
676 if (d_really_is_positive(dentry)) {
685 dentry = ERR_PTR(err);
686 else if (dentry != req->r_dentry)
687 dentry = dget(req->r_dentry); /* we got spliced */
693 static bool is_root_ceph_dentry(struct inode *inode, struct dentry *dentry)
695 return ceph_ino(inode) == CEPH_INO_ROOT &&
696 strncmp(dentry->d_name.name, ".ceph", 5) == 0;
700 * Look up a single dir entry. If there is a lookup intent, inform
701 * the MDS so that it gets our 'caps wanted' value in a single op.
703 static struct dentry *ceph_lookup(struct inode *dir, struct dentry *dentry,
706 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
707 struct ceph_mds_client *mdsc = fsc->mdsc;
708 struct ceph_mds_request *req;
713 dout("lookup %p dentry %p '%pd'\n",
714 dir, dentry, dentry);
716 if (dentry->d_name.len > NAME_MAX)
717 return ERR_PTR(-ENAMETOOLONG);
719 /* can we conclude ENOENT locally? */
720 if (d_really_is_negative(dentry)) {
721 struct ceph_inode_info *ci = ceph_inode(dir);
722 struct ceph_dentry_info *di = ceph_dentry(dentry);
724 spin_lock(&ci->i_ceph_lock);
725 dout(" dir %p flags are %d\n", dir, ci->i_ceph_flags);
726 if (strncmp(dentry->d_name.name,
727 fsc->mount_options->snapdir_name,
728 dentry->d_name.len) &&
729 !is_root_ceph_dentry(dir, dentry) &&
730 ceph_test_mount_opt(fsc, DCACHE) &&
731 __ceph_dir_is_complete(ci) &&
732 (__ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1))) {
733 spin_unlock(&ci->i_ceph_lock);
734 dout(" dir %p complete, -ENOENT\n", dir);
736 di->lease_shared_gen = ci->i_shared_gen;
739 spin_unlock(&ci->i_ceph_lock);
742 op = ceph_snap(dir) == CEPH_SNAPDIR ?
743 CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP;
744 req = ceph_mdsc_create_request(mdsc, op, USE_ANY_MDS);
746 return ERR_CAST(req);
747 req->r_dentry = dget(dentry);
750 mask = CEPH_STAT_CAP_INODE | CEPH_CAP_AUTH_SHARED;
751 if (ceph_security_xattr_wanted(dir))
752 mask |= CEPH_CAP_XATTR_SHARED;
753 req->r_args.getattr.mask = cpu_to_le32(mask);
756 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
757 err = ceph_mdsc_do_request(mdsc, NULL, req);
758 err = ceph_handle_snapdir(req, dentry, err);
759 dentry = ceph_finish_lookup(req, dentry, err);
760 ceph_mdsc_put_request(req); /* will dput(dentry) */
761 dout("lookup result=%p\n", dentry);
766 * If we do a create but get no trace back from the MDS, follow up with
767 * a lookup (the VFS expects us to link up the provided dentry).
769 int ceph_handle_notrace_create(struct inode *dir, struct dentry *dentry)
771 struct dentry *result = ceph_lookup(dir, dentry, 0);
773 if (result && !IS_ERR(result)) {
775 * We created the item, then did a lookup, and found
776 * it was already linked to another inode we already
777 * had in our cache (and thus got spliced). To not
778 * confuse VFS (especially when inode is a directory),
779 * we don't link our dentry to that inode, return an
782 * This event should be rare and it happens only when
783 * we talk to old MDS. Recent MDS does not send traceless
784 * reply for request that creates new inode.
789 return PTR_ERR(result);
792 static int ceph_mknod(struct inode *dir, struct dentry *dentry,
793 umode_t mode, dev_t rdev)
795 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
796 struct ceph_mds_client *mdsc = fsc->mdsc;
797 struct ceph_mds_request *req;
798 struct ceph_acls_info acls = {};
801 if (ceph_snap(dir) != CEPH_NOSNAP)
804 err = ceph_pre_init_acls(dir, &mode, &acls);
808 dout("mknod in dir %p dentry %p mode 0%ho rdev %d\n",
809 dir, dentry, mode, rdev);
810 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_MKNOD, USE_AUTH_MDS);
815 req->r_dentry = dget(dentry);
818 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
819 req->r_args.mknod.mode = cpu_to_le32(mode);
820 req->r_args.mknod.rdev = cpu_to_le32(rdev);
821 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
822 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
824 req->r_pagelist = acls.pagelist;
825 acls.pagelist = NULL;
827 err = ceph_mdsc_do_request(mdsc, dir, req);
828 if (!err && !req->r_reply_info.head->is_dentry)
829 err = ceph_handle_notrace_create(dir, dentry);
830 ceph_mdsc_put_request(req);
833 ceph_init_inode_acls(d_inode(dentry), &acls);
836 ceph_release_acls_info(&acls);
840 static int ceph_create(struct inode *dir, struct dentry *dentry, umode_t mode,
843 return ceph_mknod(dir, dentry, mode, 0);
846 static int ceph_symlink(struct inode *dir, struct dentry *dentry,
849 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
850 struct ceph_mds_client *mdsc = fsc->mdsc;
851 struct ceph_mds_request *req;
854 if (ceph_snap(dir) != CEPH_NOSNAP)
857 dout("symlink in dir %p dentry %p to '%s'\n", dir, dentry, dest);
858 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SYMLINK, USE_AUTH_MDS);
863 req->r_path2 = kstrdup(dest, GFP_KERNEL);
866 ceph_mdsc_put_request(req);
870 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
871 req->r_dentry = dget(dentry);
873 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
874 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
875 err = ceph_mdsc_do_request(mdsc, dir, req);
876 if (!err && !req->r_reply_info.head->is_dentry)
877 err = ceph_handle_notrace_create(dir, dentry);
878 ceph_mdsc_put_request(req);
885 static int ceph_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
887 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
888 struct ceph_mds_client *mdsc = fsc->mdsc;
889 struct ceph_mds_request *req;
890 struct ceph_acls_info acls = {};
894 if (ceph_snap(dir) == CEPH_SNAPDIR) {
895 /* mkdir .snap/foo is a MKSNAP */
896 op = CEPH_MDS_OP_MKSNAP;
897 dout("mksnap dir %p snap '%pd' dn %p\n", dir,
899 } else if (ceph_snap(dir) == CEPH_NOSNAP) {
900 dout("mkdir dir %p dn %p mode 0%ho\n", dir, dentry, mode);
901 op = CEPH_MDS_OP_MKDIR;
907 err = ceph_pre_init_acls(dir, &mode, &acls);
911 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
917 req->r_dentry = dget(dentry);
920 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
921 req->r_args.mkdir.mode = cpu_to_le32(mode);
922 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
923 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
925 req->r_pagelist = acls.pagelist;
926 acls.pagelist = NULL;
928 err = ceph_mdsc_do_request(mdsc, dir, req);
930 !req->r_reply_info.head->is_target &&
931 !req->r_reply_info.head->is_dentry)
932 err = ceph_handle_notrace_create(dir, dentry);
933 ceph_mdsc_put_request(req);
936 ceph_init_inode_acls(d_inode(dentry), &acls);
939 ceph_release_acls_info(&acls);
943 static int ceph_link(struct dentry *old_dentry, struct inode *dir,
944 struct dentry *dentry)
946 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
947 struct ceph_mds_client *mdsc = fsc->mdsc;
948 struct ceph_mds_request *req;
951 if (ceph_snap(dir) != CEPH_NOSNAP)
954 dout("link in dir %p old_dentry %p dentry %p\n", dir,
956 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LINK, USE_AUTH_MDS);
961 req->r_dentry = dget(dentry);
963 req->r_old_dentry = dget(old_dentry);
965 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
966 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
967 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
968 /* release LINK_SHARED on source inode (mds will lock it) */
969 req->r_old_inode_drop = CEPH_CAP_LINK_SHARED;
970 err = ceph_mdsc_do_request(mdsc, dir, req);
973 } else if (!req->r_reply_info.head->is_dentry) {
974 ihold(d_inode(old_dentry));
975 d_instantiate(dentry, d_inode(old_dentry));
977 ceph_mdsc_put_request(req);
982 * For a soon-to-be unlinked file, drop the AUTH_RDCACHE caps. If it
983 * looks like the link count will hit 0, drop any other caps (other
984 * than PIN) we don't specifically want (due to the file still being
987 static int drop_caps_for_unlink(struct inode *inode)
989 struct ceph_inode_info *ci = ceph_inode(inode);
990 int drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL;
992 spin_lock(&ci->i_ceph_lock);
993 if (inode->i_nlink == 1) {
994 drop |= ~(__ceph_caps_wanted(ci) | CEPH_CAP_PIN);
995 ci->i_ceph_flags |= CEPH_I_NODELAY;
997 spin_unlock(&ci->i_ceph_lock);
1002 * rmdir and unlink are differ only by the metadata op code
1004 static int ceph_unlink(struct inode *dir, struct dentry *dentry)
1006 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
1007 struct ceph_mds_client *mdsc = fsc->mdsc;
1008 struct inode *inode = d_inode(dentry);
1009 struct ceph_mds_request *req;
1013 if (ceph_snap(dir) == CEPH_SNAPDIR) {
1014 /* rmdir .snap/foo is RMSNAP */
1015 dout("rmsnap dir %p '%pd' dn %p\n", dir, dentry, dentry);
1016 op = CEPH_MDS_OP_RMSNAP;
1017 } else if (ceph_snap(dir) == CEPH_NOSNAP) {
1018 dout("unlink/rmdir dir %p dn %p inode %p\n",
1019 dir, dentry, inode);
1020 op = d_is_dir(dentry) ?
1021 CEPH_MDS_OP_RMDIR : CEPH_MDS_OP_UNLINK;
1024 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
1029 req->r_dentry = dget(dentry);
1030 req->r_num_caps = 2;
1031 req->r_parent = dir;
1032 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
1033 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
1034 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
1035 req->r_inode_drop = drop_caps_for_unlink(inode);
1036 err = ceph_mdsc_do_request(mdsc, dir, req);
1037 if (!err && !req->r_reply_info.head->is_dentry)
1039 ceph_mdsc_put_request(req);
1044 static int ceph_rename(struct inode *old_dir, struct dentry *old_dentry,
1045 struct inode *new_dir, struct dentry *new_dentry,
1048 struct ceph_fs_client *fsc = ceph_sb_to_client(old_dir->i_sb);
1049 struct ceph_mds_client *mdsc = fsc->mdsc;
1050 struct ceph_mds_request *req;
1051 int op = CEPH_MDS_OP_RENAME;
1057 if (ceph_snap(old_dir) != ceph_snap(new_dir))
1059 if (ceph_snap(old_dir) != CEPH_NOSNAP) {
1060 if (old_dir == new_dir && ceph_snap(old_dir) == CEPH_SNAPDIR)
1061 op = CEPH_MDS_OP_RENAMESNAP;
1065 dout("rename dir %p dentry %p to dir %p dentry %p\n",
1066 old_dir, old_dentry, new_dir, new_dentry);
1067 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
1069 return PTR_ERR(req);
1071 req->r_dentry = dget(new_dentry);
1072 req->r_num_caps = 2;
1073 req->r_old_dentry = dget(old_dentry);
1074 req->r_old_dentry_dir = old_dir;
1075 req->r_parent = new_dir;
1076 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
1077 req->r_old_dentry_drop = CEPH_CAP_FILE_SHARED;
1078 req->r_old_dentry_unless = CEPH_CAP_FILE_EXCL;
1079 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
1080 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
1081 /* release LINK_RDCACHE on source inode (mds will lock it) */
1082 req->r_old_inode_drop = CEPH_CAP_LINK_SHARED;
1083 if (d_really_is_positive(new_dentry))
1084 req->r_inode_drop = drop_caps_for_unlink(d_inode(new_dentry));
1085 err = ceph_mdsc_do_request(mdsc, old_dir, req);
1086 if (!err && !req->r_reply_info.head->is_dentry) {
1088 * Normally d_move() is done by fill_trace (called by
1089 * do_request, above). If there is no trace, we need
1093 /* d_move screws up sibling dentries' offsets */
1094 ceph_dir_clear_complete(old_dir);
1095 ceph_dir_clear_complete(new_dir);
1097 d_move(old_dentry, new_dentry);
1099 /* ensure target dentry is invalidated, despite
1100 rehashing bug in vfs_rename_dir */
1101 ceph_invalidate_dentry_lease(new_dentry);
1103 ceph_mdsc_put_request(req);
1108 * Ensure a dentry lease will no longer revalidate.
1110 void ceph_invalidate_dentry_lease(struct dentry *dentry)
1112 spin_lock(&dentry->d_lock);
1113 ceph_dentry(dentry)->time = jiffies;
1114 ceph_dentry(dentry)->lease_shared_gen = 0;
1115 spin_unlock(&dentry->d_lock);
1119 * Check if dentry lease is valid. If not, delete the lease. Try to
1120 * renew if the least is more than half up.
1122 static int dentry_lease_is_valid(struct dentry *dentry, unsigned int flags,
1125 struct ceph_dentry_info *di;
1126 struct ceph_mds_session *s;
1130 struct ceph_mds_session *session = NULL;
1133 spin_lock(&dentry->d_lock);
1134 di = ceph_dentry(dentry);
1135 if (di && di->lease_session) {
1136 s = di->lease_session;
1137 spin_lock(&s->s_gen_ttl_lock);
1140 spin_unlock(&s->s_gen_ttl_lock);
1142 if (di->lease_gen == gen &&
1143 time_before(jiffies, di->time) &&
1144 time_before(jiffies, ttl)) {
1146 if (di->lease_renew_after &&
1147 time_after(jiffies, di->lease_renew_after)) {
1149 * We should renew. If we're in RCU walk mode
1150 * though, we can't do that so just return
1153 if (flags & LOOKUP_RCU) {
1156 session = ceph_get_mds_session(s);
1157 seq = di->lease_seq;
1158 di->lease_renew_after = 0;
1159 di->lease_renew_from = jiffies;
1164 spin_unlock(&dentry->d_lock);
1167 ceph_mdsc_lease_send_msg(session, dir, dentry,
1168 CEPH_MDS_LEASE_RENEW, seq);
1169 ceph_put_mds_session(session);
1171 dout("dentry_lease_is_valid - dentry %p = %d\n", dentry, valid);
1176 * Check if directory-wide content lease/cap is valid.
1178 static int dir_lease_is_valid(struct inode *dir, struct dentry *dentry)
1180 struct ceph_inode_info *ci = ceph_inode(dir);
1181 struct ceph_dentry_info *di = ceph_dentry(dentry);
1184 spin_lock(&ci->i_ceph_lock);
1185 if (ci->i_shared_gen == di->lease_shared_gen)
1186 valid = __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1);
1187 spin_unlock(&ci->i_ceph_lock);
1188 dout("dir_lease_is_valid dir %p v%u dentry %p v%u = %d\n",
1189 dir, (unsigned)ci->i_shared_gen, dentry,
1190 (unsigned)di->lease_shared_gen, valid);
1195 * Check if cached dentry can be trusted.
1197 static int ceph_d_revalidate(struct dentry *dentry, unsigned int flags)
1200 struct dentry *parent;
1203 if (flags & LOOKUP_RCU) {
1204 parent = READ_ONCE(dentry->d_parent);
1205 dir = d_inode_rcu(parent);
1209 parent = dget_parent(dentry);
1210 dir = d_inode(parent);
1213 dout("d_revalidate %p '%pd' inode %p offset %lld\n", dentry,
1214 dentry, d_inode(dentry), ceph_dentry(dentry)->offset);
1216 /* always trust cached snapped dentries, snapdir dentry */
1217 if (ceph_snap(dir) != CEPH_NOSNAP) {
1218 dout("d_revalidate %p '%pd' inode %p is SNAPPED\n", dentry,
1219 dentry, d_inode(dentry));
1221 } else if (d_really_is_positive(dentry) &&
1222 ceph_snap(d_inode(dentry)) == CEPH_SNAPDIR) {
1225 valid = dentry_lease_is_valid(dentry, flags, dir);
1226 if (valid == -ECHILD)
1228 if (valid || dir_lease_is_valid(dir, dentry)) {
1229 if (d_really_is_positive(dentry))
1230 valid = ceph_is_any_caps(d_inode(dentry));
1237 struct ceph_mds_client *mdsc =
1238 ceph_sb_to_client(dir->i_sb)->mdsc;
1239 struct ceph_mds_request *req;
1243 if (flags & LOOKUP_RCU)
1246 op = ceph_snap(dir) == CEPH_SNAPDIR ?
1247 CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP;
1248 req = ceph_mdsc_create_request(mdsc, op, USE_ANY_MDS);
1250 req->r_dentry = dget(dentry);
1251 req->r_num_caps = 2;
1252 req->r_parent = dir;
1254 mask = CEPH_STAT_CAP_INODE | CEPH_CAP_AUTH_SHARED;
1255 if (ceph_security_xattr_wanted(dir))
1256 mask |= CEPH_CAP_XATTR_SHARED;
1257 req->r_args.getattr.mask = cpu_to_le32(mask);
1259 err = ceph_mdsc_do_request(mdsc, NULL, req);
1262 if (d_really_is_positive(dentry) &&
1263 d_inode(dentry) == req->r_target_inode)
1267 if (d_really_is_negative(dentry))
1273 ceph_mdsc_put_request(req);
1274 dout("d_revalidate %p lookup result=%d\n",
1279 dout("d_revalidate %p %s\n", dentry, valid ? "valid" : "invalid");
1281 ceph_dentry_lru_touch(dentry);
1283 ceph_dir_clear_complete(dir);
1286 if (!(flags & LOOKUP_RCU))
1292 * Release our ceph_dentry_info.
1294 static void ceph_d_release(struct dentry *dentry)
1296 struct ceph_dentry_info *di = ceph_dentry(dentry);
1298 dout("d_release %p\n", dentry);
1299 ceph_dentry_lru_del(dentry);
1301 spin_lock(&dentry->d_lock);
1302 dentry->d_fsdata = NULL;
1303 spin_unlock(&dentry->d_lock);
1305 if (di->lease_session)
1306 ceph_put_mds_session(di->lease_session);
1307 kmem_cache_free(ceph_dentry_cachep, di);
1311 * When the VFS prunes a dentry from the cache, we need to clear the
1312 * complete flag on the parent directory.
1314 * Called under dentry->d_lock.
1316 static void ceph_d_prune(struct dentry *dentry)
1318 dout("ceph_d_prune %p\n", dentry);
1320 /* do we have a valid parent? */
1321 if (IS_ROOT(dentry))
1324 /* if we are not hashed, we don't affect dir's completeness */
1325 if (d_unhashed(dentry))
1328 if (ceph_snap(d_inode(dentry->d_parent)) == CEPH_SNAPDIR)
1332 * we hold d_lock, so d_parent is stable, and d_fsdata is never
1333 * cleared until d_release
1335 ceph_dir_clear_complete(d_inode(dentry->d_parent));
1339 * read() on a dir. This weird interface hack only works if mounted
1340 * with '-o dirstat'.
1342 static ssize_t ceph_read_dir(struct file *file, char __user *buf, size_t size,
1345 struct ceph_file_info *cf = file->private_data;
1346 struct inode *inode = file_inode(file);
1347 struct ceph_inode_info *ci = ceph_inode(inode);
1349 const int bufsize = 1024;
1351 if (!ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb), DIRSTAT))
1354 if (!cf->dir_info) {
1355 cf->dir_info = kmalloc(bufsize, GFP_KERNEL);
1359 snprintf(cf->dir_info, bufsize,
1362 " subdirs: %20lld\n"
1363 "rentries: %20lld\n"
1365 " rsubdirs: %20lld\n"
1367 "rctime: %10ld.%09ld\n",
1368 ci->i_files + ci->i_subdirs,
1371 ci->i_rfiles + ci->i_rsubdirs,
1375 (long)ci->i_rctime.tv_sec,
1376 (long)ci->i_rctime.tv_nsec);
1379 if (*ppos >= cf->dir_info_len)
1381 size = min_t(unsigned, size, cf->dir_info_len-*ppos);
1382 left = copy_to_user(buf, cf->dir_info + *ppos, size);
1385 *ppos += (size - left);
1390 * We maintain a private dentry LRU.
1392 * FIXME: this needs to be changed to a per-mds lru to be useful.
1394 void ceph_dentry_lru_add(struct dentry *dn)
1396 struct ceph_dentry_info *di = ceph_dentry(dn);
1397 struct ceph_mds_client *mdsc;
1399 dout("dentry_lru_add %p %p '%pd'\n", di, dn, dn);
1400 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1401 spin_lock(&mdsc->dentry_lru_lock);
1402 list_add_tail(&di->lru, &mdsc->dentry_lru);
1404 spin_unlock(&mdsc->dentry_lru_lock);
1407 void ceph_dentry_lru_touch(struct dentry *dn)
1409 struct ceph_dentry_info *di = ceph_dentry(dn);
1410 struct ceph_mds_client *mdsc;
1412 dout("dentry_lru_touch %p %p '%pd' (offset %lld)\n", di, dn, dn,
1414 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1415 spin_lock(&mdsc->dentry_lru_lock);
1416 list_move_tail(&di->lru, &mdsc->dentry_lru);
1417 spin_unlock(&mdsc->dentry_lru_lock);
1420 void ceph_dentry_lru_del(struct dentry *dn)
1422 struct ceph_dentry_info *di = ceph_dentry(dn);
1423 struct ceph_mds_client *mdsc;
1425 dout("dentry_lru_del %p %p '%pd'\n", di, dn, dn);
1426 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1427 spin_lock(&mdsc->dentry_lru_lock);
1428 list_del_init(&di->lru);
1430 spin_unlock(&mdsc->dentry_lru_lock);
1434 * Return name hash for a given dentry. This is dependent on
1435 * the parent directory's hash function.
1437 unsigned ceph_dentry_hash(struct inode *dir, struct dentry *dn)
1439 struct ceph_inode_info *dci = ceph_inode(dir);
1441 switch (dci->i_dir_layout.dl_dir_hash) {
1442 case 0: /* for backward compat */
1443 case CEPH_STR_HASH_LINUX:
1444 return dn->d_name.hash;
1447 return ceph_str_hash(dci->i_dir_layout.dl_dir_hash,
1448 dn->d_name.name, dn->d_name.len);
1452 const struct file_operations ceph_dir_fops = {
1453 .read = ceph_read_dir,
1454 .iterate = ceph_readdir,
1455 .llseek = ceph_dir_llseek,
1457 .release = ceph_release,
1458 .unlocked_ioctl = ceph_ioctl,
1459 .fsync = ceph_fsync,
1462 const struct file_operations ceph_snapdir_fops = {
1463 .iterate = ceph_readdir,
1464 .llseek = ceph_dir_llseek,
1466 .release = ceph_release,
1469 const struct inode_operations ceph_dir_iops = {
1470 .lookup = ceph_lookup,
1471 .permission = ceph_permission,
1472 .getattr = ceph_getattr,
1473 .setattr = ceph_setattr,
1474 .listxattr = ceph_listxattr,
1475 .get_acl = ceph_get_acl,
1476 .set_acl = ceph_set_acl,
1477 .mknod = ceph_mknod,
1478 .symlink = ceph_symlink,
1479 .mkdir = ceph_mkdir,
1481 .unlink = ceph_unlink,
1482 .rmdir = ceph_unlink,
1483 .rename = ceph_rename,
1484 .create = ceph_create,
1485 .atomic_open = ceph_atomic_open,
1488 const struct inode_operations ceph_snapdir_iops = {
1489 .lookup = ceph_lookup,
1490 .permission = ceph_permission,
1491 .getattr = ceph_getattr,
1492 .mkdir = ceph_mkdir,
1493 .rmdir = ceph_unlink,
1494 .rename = ceph_rename,
1497 const struct dentry_operations ceph_dentry_ops = {
1498 .d_revalidate = ceph_d_revalidate,
1499 .d_release = ceph_d_release,
1500 .d_prune = ceph_d_prune,
1501 .d_init = ceph_d_init,