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>
10 #include "mds_client.h"
13 * Directory operations: readdir, lookup, create, link, unlink,
18 * Ceph MDS operations are specified in terms of a base ino and
19 * relative path. Thus, the client can specify an operation on a
20 * specific inode (e.g., a getattr due to fstat(2)), or as a path
21 * relative to, say, the root directory.
23 * Normally, we limit ourselves to strict inode ops (no path component)
24 * or dentry operations (a single path component relative to an ino). The
25 * exception to this is open_root_dentry(), which will open the mount
29 const struct dentry_operations ceph_dentry_ops;
32 * Initialize ceph dentry state.
34 int ceph_init_dentry(struct dentry *dentry)
36 struct ceph_dentry_info *di;
41 di = kmem_cache_alloc(ceph_dentry_cachep, GFP_NOFS | __GFP_ZERO);
43 return -ENOMEM; /* oh well */
45 spin_lock(&dentry->d_lock);
46 if (dentry->d_fsdata) {
48 kmem_cache_free(ceph_dentry_cachep, di);
52 if (ceph_snap(dentry->d_parent->d_inode) == CEPH_NOSNAP)
53 d_set_d_op(dentry, &ceph_dentry_ops);
54 else if (ceph_snap(dentry->d_parent->d_inode) == CEPH_SNAPDIR)
55 d_set_d_op(dentry, &ceph_snapdir_dentry_ops);
57 d_set_d_op(dentry, &ceph_snap_dentry_ops);
60 di->lease_session = NULL;
61 dentry->d_time = jiffies;
62 /* avoid reordering d_fsdata setup so that the check above is safe */
64 dentry->d_fsdata = di;
65 ceph_dentry_lru_add(dentry);
67 spin_unlock(&dentry->d_lock);
71 struct inode *ceph_get_dentry_parent_inode(struct dentry *dentry)
73 struct inode *inode = NULL;
78 spin_lock(&dentry->d_lock);
79 if (!IS_ROOT(dentry)) {
80 inode = dentry->d_parent->d_inode;
83 spin_unlock(&dentry->d_lock);
89 * for readdir, we encode the directory frag and offset within that
92 static unsigned fpos_frag(loff_t p)
96 static unsigned fpos_off(loff_t p)
98 return p & 0xffffffff;
101 static int fpos_cmp(loff_t l, loff_t r)
103 int v = ceph_frag_compare(fpos_frag(l), fpos_frag(r));
106 return (int)(fpos_off(l) - fpos_off(r));
110 * When possible, we try to satisfy a readdir by peeking at the
111 * dcache. We make this work by carefully ordering dentries on
112 * d_child when we initially get results back from the MDS, and
113 * falling back to a "normal" sync readdir if any dentries in the dir
116 * Complete dir indicates that we have all dentries in the dir. It is
117 * defined IFF we hold CEPH_CAP_FILE_SHARED (which will be revoked by
118 * the MDS if/when the directory is modified).
120 static int __dcache_readdir(struct file *file, struct dir_context *ctx,
123 struct ceph_file_info *fi = file->private_data;
124 struct dentry *parent = file->f_path.dentry;
125 struct inode *dir = parent->d_inode;
127 struct dentry *dentry, *last;
128 struct ceph_dentry_info *di;
131 /* claim ref on last dentry we returned */
135 dout("__dcache_readdir %p v%u at %llu (last %p)\n",
136 dir, shared_gen, ctx->pos, last);
138 spin_lock(&parent->d_lock);
140 /* start at beginning? */
141 if (ctx->pos == 2 || last == NULL ||
142 fpos_cmp(ctx->pos, ceph_dentry(last)->offset) < 0) {
143 if (list_empty(&parent->d_subdirs))
145 p = parent->d_subdirs.prev;
146 dout(" initial p %p/%p\n", p->prev, p->next);
148 p = last->d_child.prev;
152 dentry = list_entry(p, struct dentry, d_child);
153 di = ceph_dentry(dentry);
155 dout(" p %p/%p %s d_subdirs %p/%p\n", p->prev, p->next,
156 d_unhashed(dentry) ? "!hashed" : "hashed",
157 parent->d_subdirs.prev, parent->d_subdirs.next);
158 if (p == &parent->d_subdirs) {
159 fi->flags |= CEPH_F_ATEND;
162 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
163 if (di->lease_shared_gen == shared_gen &&
164 !d_unhashed(dentry) && dentry->d_inode &&
165 ceph_snap(dentry->d_inode) != CEPH_SNAPDIR &&
166 ceph_ino(dentry->d_inode) != CEPH_INO_CEPH &&
167 fpos_cmp(ctx->pos, di->offset) <= 0)
169 dout(" skipping %p %pd at %llu (%llu)%s%s\n", dentry,
171 ctx->pos, d_unhashed(dentry) ? " unhashed" : "",
172 !dentry->d_inode ? " null" : "");
173 spin_unlock(&dentry->d_lock);
175 dentry = list_entry(p, struct dentry, d_child);
176 di = ceph_dentry(dentry);
180 spin_unlock(&dentry->d_lock);
181 spin_unlock(&parent->d_lock);
183 /* make sure a dentry wasn't dropped while we didn't have parent lock */
184 if (!ceph_dir_is_complete_ordered(dir)) {
185 dout(" lost dir complete on %p; falling back to mds\n", dir);
191 dout(" %llu (%llu) dentry %p %pd %p\n", di->offset, ctx->pos,
192 dentry, dentry, dentry->d_inode);
193 if (!dir_emit(ctx, dentry->d_name.name,
195 ceph_translate_ino(dentry->d_sb, dentry->d_inode->i_ino),
196 dentry->d_inode->i_mode >> 12)) {
198 /* remember our position */
200 fi->next_offset = fpos_off(di->offset);
206 ctx->pos = di->offset + 1;
212 spin_lock(&parent->d_lock);
213 p = p->prev; /* advance to next dentry */
217 spin_unlock(&parent->d_lock);
225 * make note of the last dentry we read, so we can
226 * continue at the same lexicographical point,
227 * regardless of what dir changes take place on the
230 static int note_last_dentry(struct ceph_file_info *fi, const char *name,
233 kfree(fi->last_name);
234 fi->last_name = kmalloc(len+1, GFP_NOFS);
237 memcpy(fi->last_name, name, len);
238 fi->last_name[len] = 0;
239 dout("note_last_dentry '%s'\n", fi->last_name);
243 static int ceph_readdir(struct file *file, struct dir_context *ctx)
245 struct ceph_file_info *fi = file->private_data;
246 struct inode *inode = file_inode(file);
247 struct ceph_inode_info *ci = ceph_inode(inode);
248 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
249 struct ceph_mds_client *mdsc = fsc->mdsc;
250 unsigned frag = fpos_frag(ctx->pos);
251 int off = fpos_off(ctx->pos);
254 struct ceph_mds_reply_info_parsed *rinfo;
256 dout("readdir %p file %p frag %u off %u\n", inode, file, frag, off);
257 if (fi->flags & CEPH_F_ATEND)
260 /* always start with . and .. */
262 dout("readdir off 0 -> '.'\n");
263 if (!dir_emit(ctx, ".", 1,
264 ceph_translate_ino(inode->i_sb, inode->i_ino),
265 inode->i_mode >> 12))
271 ino_t ino = parent_ino(file->f_path.dentry);
272 dout("readdir off 1 -> '..'\n");
273 if (!dir_emit(ctx, "..", 2,
274 ceph_translate_ino(inode->i_sb, ino),
275 inode->i_mode >> 12))
281 /* can we use the dcache? */
282 spin_lock(&ci->i_ceph_lock);
283 if ((ctx->pos == 2 || fi->dentry) &&
284 !ceph_test_mount_opt(fsc, NOASYNCREADDIR) &&
285 ceph_snap(inode) != CEPH_SNAPDIR &&
286 __ceph_dir_is_complete_ordered(ci) &&
287 __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1)) {
288 u32 shared_gen = ci->i_shared_gen;
289 spin_unlock(&ci->i_ceph_lock);
290 err = __dcache_readdir(file, ctx, shared_gen);
293 frag = fpos_frag(ctx->pos);
294 off = fpos_off(ctx->pos);
296 spin_unlock(&ci->i_ceph_lock);
299 err = note_last_dentry(fi, fi->dentry->d_name.name,
300 fi->dentry->d_name.len);
307 /* proceed with a normal readdir */
310 /* note dir version at start of readdir so we can tell
311 * if any dentries get dropped */
312 fi->dir_release_count = atomic_read(&ci->i_release_count);
313 fi->dir_ordered_count = ci->i_ordered_count;
317 /* do we have the correct frag content buffered? */
318 if (fi->frag != frag || fi->last_readdir == NULL) {
319 struct ceph_mds_request *req;
320 int op = ceph_snap(inode) == CEPH_SNAPDIR ?
321 CEPH_MDS_OP_LSSNAP : CEPH_MDS_OP_READDIR;
323 /* discard old result, if any */
324 if (fi->last_readdir) {
325 ceph_mdsc_put_request(fi->last_readdir);
326 fi->last_readdir = NULL;
329 dout("readdir fetching %llx.%llx frag %x offset '%s'\n",
330 ceph_vinop(inode), frag, fi->last_name);
331 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
334 err = ceph_alloc_readdir_reply_buffer(req, inode);
336 ceph_mdsc_put_request(req);
339 req->r_inode = inode;
341 req->r_dentry = dget(file->f_path.dentry);
342 /* hints to request -> mds selection code */
343 req->r_direct_mode = USE_AUTH_MDS;
344 req->r_direct_hash = ceph_frag_value(frag);
345 req->r_direct_is_hash = true;
346 req->r_path2 = kstrdup(fi->last_name, GFP_NOFS);
347 req->r_readdir_offset = fi->next_offset;
348 req->r_args.readdir.frag = cpu_to_le32(frag);
349 err = ceph_mdsc_do_request(mdsc, NULL, req);
351 ceph_mdsc_put_request(req);
354 dout("readdir got and parsed readdir result=%d"
355 " on frag %x, end=%d, complete=%d\n", err, frag,
356 (int)req->r_reply_info.dir_end,
357 (int)req->r_reply_info.dir_complete);
359 if (!req->r_did_prepopulate) {
360 dout("readdir !did_prepopulate");
361 /* preclude from marking dir complete */
362 fi->dir_release_count--;
365 /* note next offset and last dentry name */
366 rinfo = &req->r_reply_info;
367 if (le32_to_cpu(rinfo->dir_dir->frag) != frag) {
368 frag = le32_to_cpu(rinfo->dir_dir->frag);
369 if (ceph_frag_is_leftmost(frag))
373 off = fi->next_offset;
376 fi->offset = fi->next_offset;
377 fi->last_readdir = req;
379 if (req->r_reply_info.dir_end) {
380 kfree(fi->last_name);
381 fi->last_name = NULL;
382 if (ceph_frag_is_rightmost(frag))
387 err = note_last_dentry(fi,
388 rinfo->dir_dname[rinfo->dir_nr-1],
389 rinfo->dir_dname_len[rinfo->dir_nr-1]);
392 fi->next_offset += rinfo->dir_nr;
396 rinfo = &fi->last_readdir->r_reply_info;
397 dout("readdir frag %x num %d off %d chunkoff %d\n", frag,
398 rinfo->dir_nr, off, fi->offset);
400 ctx->pos = ceph_make_fpos(frag, off);
401 while (off >= fi->offset && off - fi->offset < rinfo->dir_nr) {
402 struct ceph_mds_reply_inode *in =
403 rinfo->dir_in[off - fi->offset].in;
404 struct ceph_vino vino;
407 dout("readdir off %d (%d/%d) -> %lld '%.*s' %p\n",
408 off, off - fi->offset, rinfo->dir_nr, ctx->pos,
409 rinfo->dir_dname_len[off - fi->offset],
410 rinfo->dir_dname[off - fi->offset], in);
412 ftype = le32_to_cpu(in->mode) >> 12;
413 vino.ino = le64_to_cpu(in->ino);
414 vino.snap = le64_to_cpu(in->snapid);
415 ino = ceph_vino_to_ino(vino);
417 rinfo->dir_dname[off - fi->offset],
418 rinfo->dir_dname_len[off - fi->offset],
419 ceph_translate_ino(inode->i_sb, ino), ftype)) {
420 dout("filldir stopping us...\n");
428 ceph_mdsc_put_request(fi->last_readdir);
429 fi->last_readdir = NULL;
434 if (!ceph_frag_is_rightmost(frag)) {
435 frag = ceph_frag_next(frag);
437 ctx->pos = ceph_make_fpos(frag, off);
438 dout("readdir next frag is %x\n", frag);
441 fi->flags |= CEPH_F_ATEND;
444 * if dir_release_count still matches the dir, no dentries
445 * were released during the whole readdir, and we should have
446 * the complete dir contents in our cache.
448 spin_lock(&ci->i_ceph_lock);
449 if (atomic_read(&ci->i_release_count) == fi->dir_release_count) {
450 if (ci->i_ordered_count == fi->dir_ordered_count)
451 dout(" marking %p complete and ordered\n", inode);
453 dout(" marking %p complete\n", inode);
454 __ceph_dir_set_complete(ci, fi->dir_release_count,
455 fi->dir_ordered_count);
457 spin_unlock(&ci->i_ceph_lock);
459 dout("readdir %p file %p done.\n", inode, file);
463 static void reset_readdir(struct ceph_file_info *fi, unsigned frag)
465 if (fi->last_readdir) {
466 ceph_mdsc_put_request(fi->last_readdir);
467 fi->last_readdir = NULL;
469 kfree(fi->last_name);
470 fi->last_name = NULL;
471 if (ceph_frag_is_leftmost(frag))
472 fi->next_offset = 2; /* compensate for . and .. */
479 fi->flags &= ~CEPH_F_ATEND;
482 static loff_t ceph_dir_llseek(struct file *file, loff_t offset, int whence)
484 struct ceph_file_info *fi = file->private_data;
485 struct inode *inode = file->f_mapping->host;
486 loff_t old_offset = ceph_make_fpos(fi->frag, fi->next_offset);
489 mutex_lock(&inode->i_mutex);
493 offset += inode->i_size + 2; /* FIXME */
496 offset += file->f_pos;
504 if (offset != file->f_pos) {
505 file->f_pos = offset;
507 fi->flags &= ~CEPH_F_ATEND;
512 * discard buffered readdir content on seekdir(0), or
513 * seek to new frag, or seek prior to current chunk.
516 fpos_frag(offset) != fi->frag ||
517 fpos_off(offset) < fi->offset) {
518 dout("dir_llseek dropping %p content\n", file);
519 reset_readdir(fi, fpos_frag(offset));
522 /* bump dir_release_count if we did a forward seek */
523 if (fpos_cmp(offset, old_offset) > 0)
524 fi->dir_release_count--;
527 mutex_unlock(&inode->i_mutex);
532 * Handle lookups for the hidden .snap directory.
534 int ceph_handle_snapdir(struct ceph_mds_request *req,
535 struct dentry *dentry, int err)
537 struct ceph_fs_client *fsc = ceph_sb_to_client(dentry->d_sb);
538 struct inode *parent = dentry->d_parent->d_inode; /* we hold i_mutex */
541 if (err == -ENOENT &&
542 ceph_snap(parent) == CEPH_NOSNAP &&
543 strcmp(dentry->d_name.name,
544 fsc->mount_options->snapdir_name) == 0) {
545 struct inode *inode = ceph_get_snapdir(parent);
546 dout("ENOENT on snapdir %p '%pd', linking to snapdir %p\n",
547 dentry, dentry, inode);
548 BUG_ON(!d_unhashed(dentry));
549 d_add(dentry, inode);
556 * Figure out final result of a lookup/open request.
558 * Mainly, make sure we return the final req->r_dentry (if it already
559 * existed) in place of the original VFS-provided dentry when they
562 * Gracefully handle the case where the MDS replies with -ENOENT and
563 * no trace (which it may do, at its discretion, e.g., if it doesn't
564 * care to issue a lease on the negative dentry).
566 struct dentry *ceph_finish_lookup(struct ceph_mds_request *req,
567 struct dentry *dentry, int err)
569 if (err == -ENOENT) {
572 if (!req->r_reply_info.head->is_dentry) {
573 dout("ENOENT and no trace, dentry %p inode %p\n",
574 dentry, dentry->d_inode);
575 if (dentry->d_inode) {
584 dentry = ERR_PTR(err);
585 else if (dentry != req->r_dentry)
586 dentry = dget(req->r_dentry); /* we got spliced */
592 static int is_root_ceph_dentry(struct inode *inode, struct dentry *dentry)
594 return ceph_ino(inode) == CEPH_INO_ROOT &&
595 strncmp(dentry->d_name.name, ".ceph", 5) == 0;
599 * Look up a single dir entry. If there is a lookup intent, inform
600 * the MDS so that it gets our 'caps wanted' value in a single op.
602 static struct dentry *ceph_lookup(struct inode *dir, struct dentry *dentry,
605 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
606 struct ceph_mds_client *mdsc = fsc->mdsc;
607 struct ceph_mds_request *req;
611 dout("lookup %p dentry %p '%pd'\n",
612 dir, dentry, dentry);
614 if (dentry->d_name.len > NAME_MAX)
615 return ERR_PTR(-ENAMETOOLONG);
617 err = ceph_init_dentry(dentry);
621 /* can we conclude ENOENT locally? */
622 if (dentry->d_inode == NULL) {
623 struct ceph_inode_info *ci = ceph_inode(dir);
624 struct ceph_dentry_info *di = ceph_dentry(dentry);
626 spin_lock(&ci->i_ceph_lock);
627 dout(" dir %p flags are %d\n", dir, ci->i_ceph_flags);
628 if (strncmp(dentry->d_name.name,
629 fsc->mount_options->snapdir_name,
630 dentry->d_name.len) &&
631 !is_root_ceph_dentry(dir, dentry) &&
632 __ceph_dir_is_complete(ci) &&
633 (__ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1))) {
634 spin_unlock(&ci->i_ceph_lock);
635 dout(" dir %p complete, -ENOENT\n", dir);
637 di->lease_shared_gen = ci->i_shared_gen;
640 spin_unlock(&ci->i_ceph_lock);
643 op = ceph_snap(dir) == CEPH_SNAPDIR ?
644 CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP;
645 req = ceph_mdsc_create_request(mdsc, op, USE_ANY_MDS);
647 return ERR_CAST(req);
648 req->r_dentry = dget(dentry);
650 /* we only need inode linkage */
651 req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
652 req->r_locked_dir = dir;
653 err = ceph_mdsc_do_request(mdsc, NULL, req);
654 err = ceph_handle_snapdir(req, dentry, err);
655 dentry = ceph_finish_lookup(req, dentry, err);
656 ceph_mdsc_put_request(req); /* will dput(dentry) */
657 dout("lookup result=%p\n", dentry);
662 * If we do a create but get no trace back from the MDS, follow up with
663 * a lookup (the VFS expects us to link up the provided dentry).
665 int ceph_handle_notrace_create(struct inode *dir, struct dentry *dentry)
667 struct dentry *result = ceph_lookup(dir, dentry, 0);
669 if (result && !IS_ERR(result)) {
671 * We created the item, then did a lookup, and found
672 * it was already linked to another inode we already
673 * had in our cache (and thus got spliced). To not
674 * confuse VFS (especially when inode is a directory),
675 * we don't link our dentry to that inode, return an
678 * This event should be rare and it happens only when
679 * we talk to old MDS. Recent MDS does not send traceless
680 * reply for request that creates new inode.
685 return PTR_ERR(result);
688 static int ceph_mknod(struct inode *dir, struct dentry *dentry,
689 umode_t mode, dev_t rdev)
691 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
692 struct ceph_mds_client *mdsc = fsc->mdsc;
693 struct ceph_mds_request *req;
694 struct ceph_acls_info acls = {};
697 if (ceph_snap(dir) != CEPH_NOSNAP)
700 err = ceph_pre_init_acls(dir, &mode, &acls);
704 dout("mknod in dir %p dentry %p mode 0%ho rdev %d\n",
705 dir, dentry, mode, rdev);
706 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_MKNOD, USE_AUTH_MDS);
711 req->r_dentry = dget(dentry);
713 req->r_locked_dir = dir;
714 req->r_args.mknod.mode = cpu_to_le32(mode);
715 req->r_args.mknod.rdev = cpu_to_le32(rdev);
716 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
717 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
719 req->r_pagelist = acls.pagelist;
720 acls.pagelist = NULL;
722 err = ceph_mdsc_do_request(mdsc, dir, req);
723 if (!err && !req->r_reply_info.head->is_dentry)
724 err = ceph_handle_notrace_create(dir, dentry);
725 ceph_mdsc_put_request(req);
728 ceph_init_inode_acls(dentry->d_inode, &acls);
731 ceph_release_acls_info(&acls);
735 static int ceph_create(struct inode *dir, struct dentry *dentry, umode_t mode,
738 return ceph_mknod(dir, dentry, mode, 0);
741 static int ceph_symlink(struct inode *dir, struct dentry *dentry,
744 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
745 struct ceph_mds_client *mdsc = fsc->mdsc;
746 struct ceph_mds_request *req;
749 if (ceph_snap(dir) != CEPH_NOSNAP)
752 dout("symlink in dir %p dentry %p to '%s'\n", dir, dentry, dest);
753 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SYMLINK, USE_AUTH_MDS);
758 req->r_dentry = dget(dentry);
760 req->r_path2 = kstrdup(dest, GFP_NOFS);
761 req->r_locked_dir = dir;
762 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
763 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
764 err = ceph_mdsc_do_request(mdsc, dir, req);
765 if (!err && !req->r_reply_info.head->is_dentry)
766 err = ceph_handle_notrace_create(dir, dentry);
767 ceph_mdsc_put_request(req);
774 static int ceph_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
776 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
777 struct ceph_mds_client *mdsc = fsc->mdsc;
778 struct ceph_mds_request *req;
779 struct ceph_acls_info acls = {};
783 if (ceph_snap(dir) == CEPH_SNAPDIR) {
784 /* mkdir .snap/foo is a MKSNAP */
785 op = CEPH_MDS_OP_MKSNAP;
786 dout("mksnap dir %p snap '%pd' dn %p\n", dir,
788 } else if (ceph_snap(dir) == CEPH_NOSNAP) {
789 dout("mkdir dir %p dn %p mode 0%ho\n", dir, dentry, mode);
790 op = CEPH_MDS_OP_MKDIR;
796 err = ceph_pre_init_acls(dir, &mode, &acls);
800 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
806 req->r_dentry = dget(dentry);
808 req->r_locked_dir = dir;
809 req->r_args.mkdir.mode = cpu_to_le32(mode);
810 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
811 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
813 req->r_pagelist = acls.pagelist;
814 acls.pagelist = NULL;
816 err = ceph_mdsc_do_request(mdsc, dir, req);
818 !req->r_reply_info.head->is_target &&
819 !req->r_reply_info.head->is_dentry)
820 err = ceph_handle_notrace_create(dir, dentry);
821 ceph_mdsc_put_request(req);
824 ceph_init_inode_acls(dentry->d_inode, &acls);
827 ceph_release_acls_info(&acls);
831 static int ceph_link(struct dentry *old_dentry, struct inode *dir,
832 struct dentry *dentry)
834 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
835 struct ceph_mds_client *mdsc = fsc->mdsc;
836 struct ceph_mds_request *req;
839 if (ceph_snap(dir) != CEPH_NOSNAP)
842 dout("link in dir %p old_dentry %p dentry %p\n", dir,
844 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LINK, USE_AUTH_MDS);
849 req->r_dentry = dget(dentry);
851 req->r_old_dentry = dget(old_dentry);
852 req->r_locked_dir = dir;
853 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
854 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
855 /* release LINK_SHARED on source inode (mds will lock it) */
856 req->r_old_inode_drop = CEPH_CAP_LINK_SHARED;
857 err = ceph_mdsc_do_request(mdsc, dir, req);
860 } else if (!req->r_reply_info.head->is_dentry) {
861 ihold(old_dentry->d_inode);
862 d_instantiate(dentry, old_dentry->d_inode);
864 ceph_mdsc_put_request(req);
869 * For a soon-to-be unlinked file, drop the AUTH_RDCACHE caps. If it
870 * looks like the link count will hit 0, drop any other caps (other
871 * than PIN) we don't specifically want (due to the file still being
874 static int drop_caps_for_unlink(struct inode *inode)
876 struct ceph_inode_info *ci = ceph_inode(inode);
877 int drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL;
879 spin_lock(&ci->i_ceph_lock);
880 if (inode->i_nlink == 1) {
881 drop |= ~(__ceph_caps_wanted(ci) | CEPH_CAP_PIN);
882 ci->i_ceph_flags |= CEPH_I_NODELAY;
884 spin_unlock(&ci->i_ceph_lock);
889 * rmdir and unlink are differ only by the metadata op code
891 static int ceph_unlink(struct inode *dir, struct dentry *dentry)
893 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
894 struct ceph_mds_client *mdsc = fsc->mdsc;
895 struct inode *inode = dentry->d_inode;
896 struct ceph_mds_request *req;
900 if (ceph_snap(dir) == CEPH_SNAPDIR) {
901 /* rmdir .snap/foo is RMSNAP */
902 dout("rmsnap dir %p '%pd' dn %p\n", dir, dentry, dentry);
903 op = CEPH_MDS_OP_RMSNAP;
904 } else if (ceph_snap(dir) == CEPH_NOSNAP) {
905 dout("unlink/rmdir dir %p dn %p inode %p\n",
907 op = d_is_dir(dentry) ?
908 CEPH_MDS_OP_RMDIR : CEPH_MDS_OP_UNLINK;
911 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
916 req->r_dentry = dget(dentry);
918 req->r_locked_dir = dir;
919 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
920 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
921 req->r_inode_drop = drop_caps_for_unlink(inode);
922 err = ceph_mdsc_do_request(mdsc, dir, req);
923 if (!err && !req->r_reply_info.head->is_dentry)
925 ceph_mdsc_put_request(req);
930 static int ceph_rename(struct inode *old_dir, struct dentry *old_dentry,
931 struct inode *new_dir, struct dentry *new_dentry)
933 struct ceph_fs_client *fsc = ceph_sb_to_client(old_dir->i_sb);
934 struct ceph_mds_client *mdsc = fsc->mdsc;
935 struct ceph_mds_request *req;
938 if (ceph_snap(old_dir) != ceph_snap(new_dir))
940 if (ceph_snap(old_dir) != CEPH_NOSNAP ||
941 ceph_snap(new_dir) != CEPH_NOSNAP)
943 dout("rename dir %p dentry %p to dir %p dentry %p\n",
944 old_dir, old_dentry, new_dir, new_dentry);
945 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_RENAME, USE_AUTH_MDS);
949 req->r_dentry = dget(new_dentry);
951 req->r_old_dentry = dget(old_dentry);
952 req->r_old_dentry_dir = old_dir;
953 req->r_locked_dir = new_dir;
954 req->r_old_dentry_drop = CEPH_CAP_FILE_SHARED;
955 req->r_old_dentry_unless = CEPH_CAP_FILE_EXCL;
956 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
957 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
958 /* release LINK_RDCACHE on source inode (mds will lock it) */
959 req->r_old_inode_drop = CEPH_CAP_LINK_SHARED;
960 if (new_dentry->d_inode)
961 req->r_inode_drop = drop_caps_for_unlink(new_dentry->d_inode);
962 err = ceph_mdsc_do_request(mdsc, old_dir, req);
963 if (!err && !req->r_reply_info.head->is_dentry) {
965 * Normally d_move() is done by fill_trace (called by
966 * do_request, above). If there is no trace, we need
970 d_move(old_dentry, new_dentry);
972 /* ensure target dentry is invalidated, despite
973 rehashing bug in vfs_rename_dir */
974 ceph_invalidate_dentry_lease(new_dentry);
976 /* d_move screws up sibling dentries' offsets */
977 ceph_dir_clear_complete(old_dir);
978 ceph_dir_clear_complete(new_dir);
981 ceph_mdsc_put_request(req);
986 * Ensure a dentry lease will no longer revalidate.
988 void ceph_invalidate_dentry_lease(struct dentry *dentry)
990 spin_lock(&dentry->d_lock);
991 dentry->d_time = jiffies;
992 ceph_dentry(dentry)->lease_shared_gen = 0;
993 spin_unlock(&dentry->d_lock);
997 * Check if dentry lease is valid. If not, delete the lease. Try to
998 * renew if the least is more than half up.
1000 static int dentry_lease_is_valid(struct dentry *dentry)
1002 struct ceph_dentry_info *di;
1003 struct ceph_mds_session *s;
1007 struct ceph_mds_session *session = NULL;
1008 struct inode *dir = NULL;
1011 spin_lock(&dentry->d_lock);
1012 di = ceph_dentry(dentry);
1013 if (di->lease_session) {
1014 s = di->lease_session;
1015 spin_lock(&s->s_gen_ttl_lock);
1018 spin_unlock(&s->s_gen_ttl_lock);
1020 if (di->lease_gen == gen &&
1021 time_before(jiffies, dentry->d_time) &&
1022 time_before(jiffies, ttl)) {
1024 if (di->lease_renew_after &&
1025 time_after(jiffies, di->lease_renew_after)) {
1026 /* we should renew */
1027 dir = dentry->d_parent->d_inode;
1028 session = ceph_get_mds_session(s);
1029 seq = di->lease_seq;
1030 di->lease_renew_after = 0;
1031 di->lease_renew_from = jiffies;
1035 spin_unlock(&dentry->d_lock);
1038 ceph_mdsc_lease_send_msg(session, dir, dentry,
1039 CEPH_MDS_LEASE_RENEW, seq);
1040 ceph_put_mds_session(session);
1042 dout("dentry_lease_is_valid - dentry %p = %d\n", dentry, valid);
1047 * Check if directory-wide content lease/cap is valid.
1049 static int dir_lease_is_valid(struct inode *dir, struct dentry *dentry)
1051 struct ceph_inode_info *ci = ceph_inode(dir);
1052 struct ceph_dentry_info *di = ceph_dentry(dentry);
1055 spin_lock(&ci->i_ceph_lock);
1056 if (ci->i_shared_gen == di->lease_shared_gen)
1057 valid = __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1);
1058 spin_unlock(&ci->i_ceph_lock);
1059 dout("dir_lease_is_valid dir %p v%u dentry %p v%u = %d\n",
1060 dir, (unsigned)ci->i_shared_gen, dentry,
1061 (unsigned)di->lease_shared_gen, valid);
1066 * Check if cached dentry can be trusted.
1068 static int ceph_d_revalidate(struct dentry *dentry, unsigned int flags)
1073 if (flags & LOOKUP_RCU)
1076 dout("d_revalidate %p '%pd' inode %p offset %lld\n", dentry,
1077 dentry, dentry->d_inode, ceph_dentry(dentry)->offset);
1079 dir = ceph_get_dentry_parent_inode(dentry);
1081 /* always trust cached snapped dentries, snapdir dentry */
1082 if (ceph_snap(dir) != CEPH_NOSNAP) {
1083 dout("d_revalidate %p '%pd' inode %p is SNAPPED\n", dentry,
1084 dentry, dentry->d_inode);
1086 } else if (dentry->d_inode &&
1087 ceph_snap(dentry->d_inode) == CEPH_SNAPDIR) {
1089 } else if (dentry_lease_is_valid(dentry) ||
1090 dir_lease_is_valid(dir, dentry)) {
1091 if (dentry->d_inode)
1092 valid = ceph_is_any_caps(dentry->d_inode);
1097 dout("d_revalidate %p %s\n", dentry, valid ? "valid" : "invalid");
1099 ceph_dentry_lru_touch(dentry);
1101 ceph_dir_clear_complete(dir);
1108 * Release our ceph_dentry_info.
1110 static void ceph_d_release(struct dentry *dentry)
1112 struct ceph_dentry_info *di = ceph_dentry(dentry);
1114 dout("d_release %p\n", dentry);
1115 ceph_dentry_lru_del(dentry);
1116 if (di->lease_session)
1117 ceph_put_mds_session(di->lease_session);
1118 kmem_cache_free(ceph_dentry_cachep, di);
1119 dentry->d_fsdata = NULL;
1122 static int ceph_snapdir_d_revalidate(struct dentry *dentry,
1126 * Eventually, we'll want to revalidate snapped metadata
1127 * too... probably...
1133 * When the VFS prunes a dentry from the cache, we need to clear the
1134 * complete flag on the parent directory.
1136 * Called under dentry->d_lock.
1138 static void ceph_d_prune(struct dentry *dentry)
1140 dout("ceph_d_prune %p\n", dentry);
1142 /* do we have a valid parent? */
1143 if (IS_ROOT(dentry))
1146 /* if we are not hashed, we don't affect dir's completeness */
1147 if (d_unhashed(dentry))
1151 * we hold d_lock, so d_parent is stable, and d_fsdata is never
1152 * cleared until d_release
1154 ceph_dir_clear_complete(dentry->d_parent->d_inode);
1158 * read() on a dir. This weird interface hack only works if mounted
1159 * with '-o dirstat'.
1161 static ssize_t ceph_read_dir(struct file *file, char __user *buf, size_t size,
1164 struct ceph_file_info *cf = file->private_data;
1165 struct inode *inode = file_inode(file);
1166 struct ceph_inode_info *ci = ceph_inode(inode);
1168 const int bufsize = 1024;
1170 if (!ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb), DIRSTAT))
1173 if (!cf->dir_info) {
1174 cf->dir_info = kmalloc(bufsize, GFP_NOFS);
1178 snprintf(cf->dir_info, bufsize,
1181 " subdirs: %20lld\n"
1182 "rentries: %20lld\n"
1184 " rsubdirs: %20lld\n"
1186 "rctime: %10ld.%09ld\n",
1187 ci->i_files + ci->i_subdirs,
1190 ci->i_rfiles + ci->i_rsubdirs,
1194 (long)ci->i_rctime.tv_sec,
1195 (long)ci->i_rctime.tv_nsec);
1198 if (*ppos >= cf->dir_info_len)
1200 size = min_t(unsigned, size, cf->dir_info_len-*ppos);
1201 left = copy_to_user(buf, cf->dir_info + *ppos, size);
1204 *ppos += (size - left);
1209 * an fsync() on a dir will wait for any uncommitted directory
1210 * operations to commit.
1212 static int ceph_dir_fsync(struct file *file, loff_t start, loff_t end,
1215 struct inode *inode = file_inode(file);
1216 struct ceph_inode_info *ci = ceph_inode(inode);
1217 struct list_head *head = &ci->i_unsafe_dirops;
1218 struct ceph_mds_request *req;
1222 dout("dir_fsync %p\n", inode);
1223 ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
1226 mutex_lock(&inode->i_mutex);
1228 spin_lock(&ci->i_unsafe_lock);
1229 if (list_empty(head))
1232 req = list_entry(head->prev,
1233 struct ceph_mds_request, r_unsafe_dir_item);
1234 last_tid = req->r_tid;
1237 ceph_mdsc_get_request(req);
1238 spin_unlock(&ci->i_unsafe_lock);
1240 dout("dir_fsync %p wait on tid %llu (until %llu)\n",
1241 inode, req->r_tid, last_tid);
1242 if (req->r_timeout) {
1243 ret = wait_for_completion_timeout(
1244 &req->r_safe_completion, req->r_timeout);
1248 ret = -EIO; /* timed out */
1250 wait_for_completion(&req->r_safe_completion);
1252 ceph_mdsc_put_request(req);
1254 spin_lock(&ci->i_unsafe_lock);
1255 if (ret || list_empty(head))
1257 req = list_entry(head->next,
1258 struct ceph_mds_request, r_unsafe_dir_item);
1259 } while (req->r_tid < last_tid);
1261 spin_unlock(&ci->i_unsafe_lock);
1262 mutex_unlock(&inode->i_mutex);
1268 * We maintain a private dentry LRU.
1270 * FIXME: this needs to be changed to a per-mds lru to be useful.
1272 void ceph_dentry_lru_add(struct dentry *dn)
1274 struct ceph_dentry_info *di = ceph_dentry(dn);
1275 struct ceph_mds_client *mdsc;
1277 dout("dentry_lru_add %p %p '%pd'\n", di, dn, dn);
1278 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1279 spin_lock(&mdsc->dentry_lru_lock);
1280 list_add_tail(&di->lru, &mdsc->dentry_lru);
1282 spin_unlock(&mdsc->dentry_lru_lock);
1285 void ceph_dentry_lru_touch(struct dentry *dn)
1287 struct ceph_dentry_info *di = ceph_dentry(dn);
1288 struct ceph_mds_client *mdsc;
1290 dout("dentry_lru_touch %p %p '%pd' (offset %lld)\n", di, dn, dn,
1292 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1293 spin_lock(&mdsc->dentry_lru_lock);
1294 list_move_tail(&di->lru, &mdsc->dentry_lru);
1295 spin_unlock(&mdsc->dentry_lru_lock);
1298 void ceph_dentry_lru_del(struct dentry *dn)
1300 struct ceph_dentry_info *di = ceph_dentry(dn);
1301 struct ceph_mds_client *mdsc;
1303 dout("dentry_lru_del %p %p '%pd'\n", di, dn, dn);
1304 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1305 spin_lock(&mdsc->dentry_lru_lock);
1306 list_del_init(&di->lru);
1308 spin_unlock(&mdsc->dentry_lru_lock);
1312 * Return name hash for a given dentry. This is dependent on
1313 * the parent directory's hash function.
1315 unsigned ceph_dentry_hash(struct inode *dir, struct dentry *dn)
1317 struct ceph_inode_info *dci = ceph_inode(dir);
1319 switch (dci->i_dir_layout.dl_dir_hash) {
1320 case 0: /* for backward compat */
1321 case CEPH_STR_HASH_LINUX:
1322 return dn->d_name.hash;
1325 return ceph_str_hash(dci->i_dir_layout.dl_dir_hash,
1326 dn->d_name.name, dn->d_name.len);
1330 const struct file_operations ceph_dir_fops = {
1331 .read = ceph_read_dir,
1332 .iterate = ceph_readdir,
1333 .llseek = ceph_dir_llseek,
1335 .release = ceph_release,
1336 .unlocked_ioctl = ceph_ioctl,
1337 .fsync = ceph_dir_fsync,
1340 const struct file_operations ceph_snapdir_fops = {
1341 .iterate = ceph_readdir,
1342 .llseek = ceph_dir_llseek,
1344 .release = ceph_release,
1347 const struct inode_operations ceph_dir_iops = {
1348 .lookup = ceph_lookup,
1349 .permission = ceph_permission,
1350 .getattr = ceph_getattr,
1351 .setattr = ceph_setattr,
1352 .setxattr = ceph_setxattr,
1353 .getxattr = ceph_getxattr,
1354 .listxattr = ceph_listxattr,
1355 .removexattr = ceph_removexattr,
1356 .get_acl = ceph_get_acl,
1357 .set_acl = ceph_set_acl,
1358 .mknod = ceph_mknod,
1359 .symlink = ceph_symlink,
1360 .mkdir = ceph_mkdir,
1362 .unlink = ceph_unlink,
1363 .rmdir = ceph_unlink,
1364 .rename = ceph_rename,
1365 .create = ceph_create,
1366 .atomic_open = ceph_atomic_open,
1369 const struct inode_operations ceph_snapdir_iops = {
1370 .lookup = ceph_lookup,
1371 .permission = ceph_permission,
1372 .getattr = ceph_getattr,
1373 .mkdir = ceph_mkdir,
1374 .rmdir = ceph_unlink,
1377 const struct dentry_operations ceph_dentry_ops = {
1378 .d_revalidate = ceph_d_revalidate,
1379 .d_release = ceph_d_release,
1380 .d_prune = ceph_d_prune,
1383 const struct dentry_operations ceph_snapdir_dentry_ops = {
1384 .d_revalidate = ceph_snapdir_d_revalidate,
1385 .d_release = ceph_d_release,
1388 const struct dentry_operations ceph_snap_dentry_ops = {
1389 .d_release = ceph_d_release,
1390 .d_prune = ceph_d_prune,