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). Link our
674 * dentry to that inode, but don't hash it, just in
675 * case the VFS wants to dereference it.
677 BUG_ON(!result->d_inode);
678 d_instantiate(dentry, result->d_inode);
681 return PTR_ERR(result);
684 static int ceph_mknod(struct inode *dir, struct dentry *dentry,
685 umode_t mode, dev_t rdev)
687 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
688 struct ceph_mds_client *mdsc = fsc->mdsc;
689 struct ceph_mds_request *req;
690 struct ceph_acls_info acls = {};
693 if (ceph_snap(dir) != CEPH_NOSNAP)
696 err = ceph_pre_init_acls(dir, &mode, &acls);
700 dout("mknod in dir %p dentry %p mode 0%ho rdev %d\n",
701 dir, dentry, mode, rdev);
702 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_MKNOD, USE_AUTH_MDS);
707 req->r_dentry = dget(dentry);
709 req->r_locked_dir = dir;
710 req->r_args.mknod.mode = cpu_to_le32(mode);
711 req->r_args.mknod.rdev = cpu_to_le32(rdev);
712 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
713 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
715 req->r_pagelist = acls.pagelist;
716 acls.pagelist = NULL;
718 err = ceph_mdsc_do_request(mdsc, dir, req);
719 if (!err && !req->r_reply_info.head->is_dentry)
720 err = ceph_handle_notrace_create(dir, dentry);
721 ceph_mdsc_put_request(req);
724 ceph_init_inode_acls(dentry->d_inode, &acls);
727 ceph_release_acls_info(&acls);
731 static int ceph_create(struct inode *dir, struct dentry *dentry, umode_t mode,
734 return ceph_mknod(dir, dentry, mode, 0);
737 static int ceph_symlink(struct inode *dir, struct dentry *dentry,
740 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
741 struct ceph_mds_client *mdsc = fsc->mdsc;
742 struct ceph_mds_request *req;
745 if (ceph_snap(dir) != CEPH_NOSNAP)
748 dout("symlink in dir %p dentry %p to '%s'\n", dir, dentry, dest);
749 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SYMLINK, USE_AUTH_MDS);
754 req->r_dentry = dget(dentry);
756 req->r_path2 = kstrdup(dest, GFP_NOFS);
757 req->r_locked_dir = dir;
758 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
759 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
760 err = ceph_mdsc_do_request(mdsc, dir, req);
761 if (!err && !req->r_reply_info.head->is_dentry)
762 err = ceph_handle_notrace_create(dir, dentry);
763 ceph_mdsc_put_request(req);
770 static int ceph_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
772 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
773 struct ceph_mds_client *mdsc = fsc->mdsc;
774 struct ceph_mds_request *req;
775 struct ceph_acls_info acls = {};
779 if (ceph_snap(dir) == CEPH_SNAPDIR) {
780 /* mkdir .snap/foo is a MKSNAP */
781 op = CEPH_MDS_OP_MKSNAP;
782 dout("mksnap dir %p snap '%pd' dn %p\n", dir,
784 } else if (ceph_snap(dir) == CEPH_NOSNAP) {
785 dout("mkdir dir %p dn %p mode 0%ho\n", dir, dentry, mode);
786 op = CEPH_MDS_OP_MKDIR;
792 err = ceph_pre_init_acls(dir, &mode, &acls);
796 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
802 req->r_dentry = dget(dentry);
804 req->r_locked_dir = dir;
805 req->r_args.mkdir.mode = cpu_to_le32(mode);
806 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
807 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
809 req->r_pagelist = acls.pagelist;
810 acls.pagelist = NULL;
812 err = ceph_mdsc_do_request(mdsc, dir, req);
814 !req->r_reply_info.head->is_target &&
815 !req->r_reply_info.head->is_dentry)
816 err = ceph_handle_notrace_create(dir, dentry);
817 ceph_mdsc_put_request(req);
820 ceph_init_inode_acls(dentry->d_inode, &acls);
823 ceph_release_acls_info(&acls);
827 static int ceph_link(struct dentry *old_dentry, struct inode *dir,
828 struct dentry *dentry)
830 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
831 struct ceph_mds_client *mdsc = fsc->mdsc;
832 struct ceph_mds_request *req;
835 if (ceph_snap(dir) != CEPH_NOSNAP)
838 dout("link in dir %p old_dentry %p dentry %p\n", dir,
840 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LINK, USE_AUTH_MDS);
845 req->r_dentry = dget(dentry);
847 req->r_old_dentry = dget(old_dentry);
848 req->r_locked_dir = dir;
849 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
850 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
851 /* release LINK_SHARED on source inode (mds will lock it) */
852 req->r_old_inode_drop = CEPH_CAP_LINK_SHARED;
853 err = ceph_mdsc_do_request(mdsc, dir, req);
856 } else if (!req->r_reply_info.head->is_dentry) {
857 ihold(old_dentry->d_inode);
858 d_instantiate(dentry, old_dentry->d_inode);
860 ceph_mdsc_put_request(req);
865 * For a soon-to-be unlinked file, drop the AUTH_RDCACHE caps. If it
866 * looks like the link count will hit 0, drop any other caps (other
867 * than PIN) we don't specifically want (due to the file still being
870 static int drop_caps_for_unlink(struct inode *inode)
872 struct ceph_inode_info *ci = ceph_inode(inode);
873 int drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL;
875 spin_lock(&ci->i_ceph_lock);
876 if (inode->i_nlink == 1) {
877 drop |= ~(__ceph_caps_wanted(ci) | CEPH_CAP_PIN);
878 ci->i_ceph_flags |= CEPH_I_NODELAY;
880 spin_unlock(&ci->i_ceph_lock);
885 * rmdir and unlink are differ only by the metadata op code
887 static int ceph_unlink(struct inode *dir, struct dentry *dentry)
889 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
890 struct ceph_mds_client *mdsc = fsc->mdsc;
891 struct inode *inode = dentry->d_inode;
892 struct ceph_mds_request *req;
896 if (ceph_snap(dir) == CEPH_SNAPDIR) {
897 /* rmdir .snap/foo is RMSNAP */
898 dout("rmsnap dir %p '%pd' dn %p\n", dir, dentry, dentry);
899 op = CEPH_MDS_OP_RMSNAP;
900 } else if (ceph_snap(dir) == CEPH_NOSNAP) {
901 dout("unlink/rmdir dir %p dn %p inode %p\n",
903 op = S_ISDIR(dentry->d_inode->i_mode) ?
904 CEPH_MDS_OP_RMDIR : CEPH_MDS_OP_UNLINK;
907 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
912 req->r_dentry = dget(dentry);
914 req->r_locked_dir = dir;
915 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
916 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
917 req->r_inode_drop = drop_caps_for_unlink(inode);
918 err = ceph_mdsc_do_request(mdsc, dir, req);
919 if (!err && !req->r_reply_info.head->is_dentry)
921 ceph_mdsc_put_request(req);
926 static int ceph_rename(struct inode *old_dir, struct dentry *old_dentry,
927 struct inode *new_dir, struct dentry *new_dentry)
929 struct ceph_fs_client *fsc = ceph_sb_to_client(old_dir->i_sb);
930 struct ceph_mds_client *mdsc = fsc->mdsc;
931 struct ceph_mds_request *req;
934 if (ceph_snap(old_dir) != ceph_snap(new_dir))
936 if (ceph_snap(old_dir) != CEPH_NOSNAP ||
937 ceph_snap(new_dir) != CEPH_NOSNAP)
939 dout("rename dir %p dentry %p to dir %p dentry %p\n",
940 old_dir, old_dentry, new_dir, new_dentry);
941 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_RENAME, USE_AUTH_MDS);
945 req->r_dentry = dget(new_dentry);
947 req->r_old_dentry = dget(old_dentry);
948 req->r_old_dentry_dir = old_dir;
949 req->r_locked_dir = new_dir;
950 req->r_old_dentry_drop = CEPH_CAP_FILE_SHARED;
951 req->r_old_dentry_unless = CEPH_CAP_FILE_EXCL;
952 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
953 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
954 /* release LINK_RDCACHE on source inode (mds will lock it) */
955 req->r_old_inode_drop = CEPH_CAP_LINK_SHARED;
956 if (new_dentry->d_inode)
957 req->r_inode_drop = drop_caps_for_unlink(new_dentry->d_inode);
958 err = ceph_mdsc_do_request(mdsc, old_dir, req);
959 if (!err && !req->r_reply_info.head->is_dentry) {
961 * Normally d_move() is done by fill_trace (called by
962 * do_request, above). If there is no trace, we need
966 d_move(old_dentry, new_dentry);
968 /* ensure target dentry is invalidated, despite
969 rehashing bug in vfs_rename_dir */
970 ceph_invalidate_dentry_lease(new_dentry);
972 /* d_move screws up sibling dentries' offsets */
973 ceph_dir_clear_complete(old_dir);
974 ceph_dir_clear_complete(new_dir);
977 ceph_mdsc_put_request(req);
982 * Ensure a dentry lease will no longer revalidate.
984 void ceph_invalidate_dentry_lease(struct dentry *dentry)
986 spin_lock(&dentry->d_lock);
987 dentry->d_time = jiffies;
988 ceph_dentry(dentry)->lease_shared_gen = 0;
989 spin_unlock(&dentry->d_lock);
993 * Check if dentry lease is valid. If not, delete the lease. Try to
994 * renew if the least is more than half up.
996 static int dentry_lease_is_valid(struct dentry *dentry)
998 struct ceph_dentry_info *di;
999 struct ceph_mds_session *s;
1003 struct ceph_mds_session *session = NULL;
1004 struct inode *dir = NULL;
1007 spin_lock(&dentry->d_lock);
1008 di = ceph_dentry(dentry);
1009 if (di->lease_session) {
1010 s = di->lease_session;
1011 spin_lock(&s->s_gen_ttl_lock);
1014 spin_unlock(&s->s_gen_ttl_lock);
1016 if (di->lease_gen == gen &&
1017 time_before(jiffies, dentry->d_time) &&
1018 time_before(jiffies, ttl)) {
1020 if (di->lease_renew_after &&
1021 time_after(jiffies, di->lease_renew_after)) {
1022 /* we should renew */
1023 dir = dentry->d_parent->d_inode;
1024 session = ceph_get_mds_session(s);
1025 seq = di->lease_seq;
1026 di->lease_renew_after = 0;
1027 di->lease_renew_from = jiffies;
1031 spin_unlock(&dentry->d_lock);
1034 ceph_mdsc_lease_send_msg(session, dir, dentry,
1035 CEPH_MDS_LEASE_RENEW, seq);
1036 ceph_put_mds_session(session);
1038 dout("dentry_lease_is_valid - dentry %p = %d\n", dentry, valid);
1043 * Check if directory-wide content lease/cap is valid.
1045 static int dir_lease_is_valid(struct inode *dir, struct dentry *dentry)
1047 struct ceph_inode_info *ci = ceph_inode(dir);
1048 struct ceph_dentry_info *di = ceph_dentry(dentry);
1051 spin_lock(&ci->i_ceph_lock);
1052 if (ci->i_shared_gen == di->lease_shared_gen)
1053 valid = __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1);
1054 spin_unlock(&ci->i_ceph_lock);
1055 dout("dir_lease_is_valid dir %p v%u dentry %p v%u = %d\n",
1056 dir, (unsigned)ci->i_shared_gen, dentry,
1057 (unsigned)di->lease_shared_gen, valid);
1062 * Check if cached dentry can be trusted.
1064 static int ceph_d_revalidate(struct dentry *dentry, unsigned int flags)
1069 if (flags & LOOKUP_RCU)
1072 dout("d_revalidate %p '%pd' inode %p offset %lld\n", dentry,
1073 dentry, dentry->d_inode, ceph_dentry(dentry)->offset);
1075 dir = ceph_get_dentry_parent_inode(dentry);
1077 /* always trust cached snapped dentries, snapdir dentry */
1078 if (ceph_snap(dir) != CEPH_NOSNAP) {
1079 dout("d_revalidate %p '%pd' inode %p is SNAPPED\n", dentry,
1080 dentry, dentry->d_inode);
1082 } else if (dentry->d_inode &&
1083 ceph_snap(dentry->d_inode) == CEPH_SNAPDIR) {
1085 } else if (dentry_lease_is_valid(dentry) ||
1086 dir_lease_is_valid(dir, dentry)) {
1087 if (dentry->d_inode)
1088 valid = ceph_is_any_caps(dentry->d_inode);
1093 dout("d_revalidate %p %s\n", dentry, valid ? "valid" : "invalid");
1095 ceph_dentry_lru_touch(dentry);
1097 ceph_dir_clear_complete(dir);
1104 * Release our ceph_dentry_info.
1106 static void ceph_d_release(struct dentry *dentry)
1108 struct ceph_dentry_info *di = ceph_dentry(dentry);
1110 dout("d_release %p\n", dentry);
1111 ceph_dentry_lru_del(dentry);
1112 if (di->lease_session)
1113 ceph_put_mds_session(di->lease_session);
1114 kmem_cache_free(ceph_dentry_cachep, di);
1115 dentry->d_fsdata = NULL;
1118 static int ceph_snapdir_d_revalidate(struct dentry *dentry,
1122 * Eventually, we'll want to revalidate snapped metadata
1123 * too... probably...
1129 * When the VFS prunes a dentry from the cache, we need to clear the
1130 * complete flag on the parent directory.
1132 * Called under dentry->d_lock.
1134 static void ceph_d_prune(struct dentry *dentry)
1136 dout("ceph_d_prune %p\n", dentry);
1138 /* do we have a valid parent? */
1139 if (IS_ROOT(dentry))
1142 /* if we are not hashed, we don't affect dir's completeness */
1143 if (d_unhashed(dentry))
1147 * we hold d_lock, so d_parent is stable, and d_fsdata is never
1148 * cleared until d_release
1150 ceph_dir_clear_complete(dentry->d_parent->d_inode);
1154 * read() on a dir. This weird interface hack only works if mounted
1155 * with '-o dirstat'.
1157 static ssize_t ceph_read_dir(struct file *file, char __user *buf, size_t size,
1160 struct ceph_file_info *cf = file->private_data;
1161 struct inode *inode = file_inode(file);
1162 struct ceph_inode_info *ci = ceph_inode(inode);
1164 const int bufsize = 1024;
1166 if (!ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb), DIRSTAT))
1169 if (!cf->dir_info) {
1170 cf->dir_info = kmalloc(bufsize, GFP_NOFS);
1174 snprintf(cf->dir_info, bufsize,
1177 " subdirs: %20lld\n"
1178 "rentries: %20lld\n"
1180 " rsubdirs: %20lld\n"
1182 "rctime: %10ld.%09ld\n",
1183 ci->i_files + ci->i_subdirs,
1186 ci->i_rfiles + ci->i_rsubdirs,
1190 (long)ci->i_rctime.tv_sec,
1191 (long)ci->i_rctime.tv_nsec);
1194 if (*ppos >= cf->dir_info_len)
1196 size = min_t(unsigned, size, cf->dir_info_len-*ppos);
1197 left = copy_to_user(buf, cf->dir_info + *ppos, size);
1200 *ppos += (size - left);
1205 * an fsync() on a dir will wait for any uncommitted directory
1206 * operations to commit.
1208 static int ceph_dir_fsync(struct file *file, loff_t start, loff_t end,
1211 struct inode *inode = file_inode(file);
1212 struct ceph_inode_info *ci = ceph_inode(inode);
1213 struct list_head *head = &ci->i_unsafe_dirops;
1214 struct ceph_mds_request *req;
1218 dout("dir_fsync %p\n", inode);
1219 ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
1222 mutex_lock(&inode->i_mutex);
1224 spin_lock(&ci->i_unsafe_lock);
1225 if (list_empty(head))
1228 req = list_entry(head->prev,
1229 struct ceph_mds_request, r_unsafe_dir_item);
1230 last_tid = req->r_tid;
1233 ceph_mdsc_get_request(req);
1234 spin_unlock(&ci->i_unsafe_lock);
1236 dout("dir_fsync %p wait on tid %llu (until %llu)\n",
1237 inode, req->r_tid, last_tid);
1238 if (req->r_timeout) {
1239 ret = wait_for_completion_timeout(
1240 &req->r_safe_completion, req->r_timeout);
1244 ret = -EIO; /* timed out */
1246 wait_for_completion(&req->r_safe_completion);
1248 ceph_mdsc_put_request(req);
1250 spin_lock(&ci->i_unsafe_lock);
1251 if (ret || list_empty(head))
1253 req = list_entry(head->next,
1254 struct ceph_mds_request, r_unsafe_dir_item);
1255 } while (req->r_tid < last_tid);
1257 spin_unlock(&ci->i_unsafe_lock);
1258 mutex_unlock(&inode->i_mutex);
1264 * We maintain a private dentry LRU.
1266 * FIXME: this needs to be changed to a per-mds lru to be useful.
1268 void ceph_dentry_lru_add(struct dentry *dn)
1270 struct ceph_dentry_info *di = ceph_dentry(dn);
1271 struct ceph_mds_client *mdsc;
1273 dout("dentry_lru_add %p %p '%pd'\n", di, dn, dn);
1274 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1275 spin_lock(&mdsc->dentry_lru_lock);
1276 list_add_tail(&di->lru, &mdsc->dentry_lru);
1278 spin_unlock(&mdsc->dentry_lru_lock);
1281 void ceph_dentry_lru_touch(struct dentry *dn)
1283 struct ceph_dentry_info *di = ceph_dentry(dn);
1284 struct ceph_mds_client *mdsc;
1286 dout("dentry_lru_touch %p %p '%pd' (offset %lld)\n", di, dn, dn,
1288 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1289 spin_lock(&mdsc->dentry_lru_lock);
1290 list_move_tail(&di->lru, &mdsc->dentry_lru);
1291 spin_unlock(&mdsc->dentry_lru_lock);
1294 void ceph_dentry_lru_del(struct dentry *dn)
1296 struct ceph_dentry_info *di = ceph_dentry(dn);
1297 struct ceph_mds_client *mdsc;
1299 dout("dentry_lru_del %p %p '%pd'\n", di, dn, dn);
1300 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1301 spin_lock(&mdsc->dentry_lru_lock);
1302 list_del_init(&di->lru);
1304 spin_unlock(&mdsc->dentry_lru_lock);
1308 * Return name hash for a given dentry. This is dependent on
1309 * the parent directory's hash function.
1311 unsigned ceph_dentry_hash(struct inode *dir, struct dentry *dn)
1313 struct ceph_inode_info *dci = ceph_inode(dir);
1315 switch (dci->i_dir_layout.dl_dir_hash) {
1316 case 0: /* for backward compat */
1317 case CEPH_STR_HASH_LINUX:
1318 return dn->d_name.hash;
1321 return ceph_str_hash(dci->i_dir_layout.dl_dir_hash,
1322 dn->d_name.name, dn->d_name.len);
1326 const struct file_operations ceph_dir_fops = {
1327 .read = ceph_read_dir,
1328 .iterate = ceph_readdir,
1329 .llseek = ceph_dir_llseek,
1331 .release = ceph_release,
1332 .unlocked_ioctl = ceph_ioctl,
1333 .fsync = ceph_dir_fsync,
1336 const struct file_operations ceph_snapdir_fops = {
1337 .iterate = ceph_readdir,
1338 .llseek = ceph_dir_llseek,
1340 .release = ceph_release,
1343 const struct inode_operations ceph_dir_iops = {
1344 .lookup = ceph_lookup,
1345 .permission = ceph_permission,
1346 .getattr = ceph_getattr,
1347 .setattr = ceph_setattr,
1348 .setxattr = ceph_setxattr,
1349 .getxattr = ceph_getxattr,
1350 .listxattr = ceph_listxattr,
1351 .removexattr = ceph_removexattr,
1352 .get_acl = ceph_get_acl,
1353 .set_acl = ceph_set_acl,
1354 .mknod = ceph_mknod,
1355 .symlink = ceph_symlink,
1356 .mkdir = ceph_mkdir,
1358 .unlink = ceph_unlink,
1359 .rmdir = ceph_unlink,
1360 .rename = ceph_rename,
1361 .create = ceph_create,
1362 .atomic_open = ceph_atomic_open,
1365 const struct inode_operations ceph_snapdir_iops = {
1366 .lookup = ceph_lookup,
1367 .permission = ceph_permission,
1368 .getattr = ceph_getattr,
1369 .mkdir = ceph_mkdir,
1370 .rmdir = ceph_unlink,
1373 const struct dentry_operations ceph_dentry_ops = {
1374 .d_revalidate = ceph_d_revalidate,
1375 .d_release = ceph_d_release,
1376 .d_prune = ceph_d_prune,
1379 const struct dentry_operations ceph_snapdir_dentry_ops = {
1380 .d_revalidate = ceph_snapdir_d_revalidate,
1381 .d_release = ceph_d_release,
1384 const struct dentry_operations ceph_snap_dentry_ops = {
1385 .d_release = ceph_d_release,
1386 .d_prune = ceph_d_prune,