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 inode_operations ceph_dir_iops;
30 const struct file_operations ceph_dir_fops;
31 const struct dentry_operations ceph_dentry_ops;
34 * Initialize ceph dentry state.
36 int ceph_init_dentry(struct dentry *dentry)
38 struct ceph_dentry_info *di;
43 di = kmem_cache_alloc(ceph_dentry_cachep, GFP_NOFS | __GFP_ZERO);
45 return -ENOMEM; /* oh well */
47 spin_lock(&dentry->d_lock);
48 if (dentry->d_fsdata) {
50 kmem_cache_free(ceph_dentry_cachep, di);
54 if (dentry->d_parent == NULL || /* nfs fh_to_dentry */
55 ceph_snap(dentry->d_parent->d_inode) == CEPH_NOSNAP)
56 d_set_d_op(dentry, &ceph_dentry_ops);
57 else if (ceph_snap(dentry->d_parent->d_inode) == CEPH_SNAPDIR)
58 d_set_d_op(dentry, &ceph_snapdir_dentry_ops);
60 d_set_d_op(dentry, &ceph_snap_dentry_ops);
63 di->lease_session = NULL;
64 dentry->d_time = jiffies;
65 /* avoid reordering d_fsdata setup so that the check above is safe */
67 dentry->d_fsdata = di;
68 ceph_dentry_lru_add(dentry);
70 spin_unlock(&dentry->d_lock);
74 struct inode *ceph_get_dentry_parent_inode(struct dentry *dentry)
76 struct inode *inode = NULL;
81 spin_lock(&dentry->d_lock);
82 if (dentry->d_parent) {
83 inode = dentry->d_parent->d_inode;
86 spin_unlock(&dentry->d_lock);
92 * for readdir, we encode the directory frag and offset within that
95 static unsigned fpos_frag(loff_t p)
99 static unsigned fpos_off(loff_t p)
101 return p & 0xffffffff;
105 * When possible, we try to satisfy a readdir by peeking at the
106 * dcache. We make this work by carefully ordering dentries on
107 * d_u.d_child when we initially get results back from the MDS, and
108 * falling back to a "normal" sync readdir if any dentries in the dir
111 * I_COMPLETE tells indicates we have all dentries in the dir. It is
112 * defined IFF we hold CEPH_CAP_FILE_SHARED (which will be revoked by
113 * the MDS if/when the directory is modified).
115 static int __dcache_readdir(struct file *filp,
116 void *dirent, filldir_t filldir)
118 struct ceph_file_info *fi = filp->private_data;
119 struct dentry *parent = filp->f_dentry;
120 struct inode *dir = parent->d_inode;
122 struct dentry *dentry, *last;
123 struct ceph_dentry_info *di;
126 /* claim ref on last dentry we returned */
130 dout("__dcache_readdir %p at %llu (last %p)\n", dir, filp->f_pos,
133 spin_lock(&parent->d_lock);
135 /* start at beginning? */
136 if (filp->f_pos == 2 || last == NULL ||
137 filp->f_pos < ceph_dentry(last)->offset) {
138 if (list_empty(&parent->d_subdirs))
140 p = parent->d_subdirs.prev;
141 dout(" initial p %p/%p\n", p->prev, p->next);
143 p = last->d_u.d_child.prev;
147 dentry = list_entry(p, struct dentry, d_u.d_child);
148 di = ceph_dentry(dentry);
150 dout(" p %p/%p %s d_subdirs %p/%p\n", p->prev, p->next,
151 d_unhashed(dentry) ? "!hashed" : "hashed",
152 parent->d_subdirs.prev, parent->d_subdirs.next);
153 if (p == &parent->d_subdirs) {
154 fi->flags |= CEPH_F_ATEND;
157 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
158 if (!d_unhashed(dentry) && dentry->d_inode &&
159 ceph_snap(dentry->d_inode) != CEPH_SNAPDIR &&
160 ceph_ino(dentry->d_inode) != CEPH_INO_CEPH &&
161 filp->f_pos <= di->offset)
163 dout(" skipping %p %.*s at %llu (%llu)%s%s\n", dentry,
164 dentry->d_name.len, dentry->d_name.name, di->offset,
165 filp->f_pos, d_unhashed(dentry) ? " unhashed" : "",
166 !dentry->d_inode ? " null" : "");
167 spin_unlock(&dentry->d_lock);
169 dentry = list_entry(p, struct dentry, d_u.d_child);
170 di = ceph_dentry(dentry);
174 spin_unlock(&dentry->d_lock);
175 spin_unlock(&parent->d_lock);
177 dout(" %llu (%llu) dentry %p %.*s %p\n", di->offset, filp->f_pos,
178 dentry, dentry->d_name.len, dentry->d_name.name, dentry->d_inode);
179 filp->f_pos = di->offset;
180 err = filldir(dirent, dentry->d_name.name,
181 dentry->d_name.len, di->offset,
182 ceph_translate_ino(dentry->d_sb, dentry->d_inode->i_ino),
183 dentry->d_inode->i_mode >> 12);
187 /* remember our position */
189 fi->next_offset = di->offset;
201 /* make sure a dentry wasn't dropped while we didn't have parent lock */
202 if (!ceph_i_test(dir, CEPH_I_COMPLETE)) {
203 dout(" lost I_COMPLETE on %p; falling back to mds\n", dir);
208 spin_lock(&parent->d_lock);
209 p = p->prev; /* advance to next dentry */
213 spin_unlock(&parent->d_lock);
221 * make note of the last dentry we read, so we can
222 * continue at the same lexicographical point,
223 * regardless of what dir changes take place on the
226 static int note_last_dentry(struct ceph_file_info *fi, const char *name,
229 kfree(fi->last_name);
230 fi->last_name = kmalloc(len+1, GFP_NOFS);
233 memcpy(fi->last_name, name, len);
234 fi->last_name[len] = 0;
235 dout("note_last_dentry '%s'\n", fi->last_name);
239 static int ceph_readdir(struct file *filp, void *dirent, filldir_t filldir)
241 struct ceph_file_info *fi = filp->private_data;
242 struct inode *inode = filp->f_dentry->d_inode;
243 struct ceph_inode_info *ci = ceph_inode(inode);
244 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
245 struct ceph_mds_client *mdsc = fsc->mdsc;
246 unsigned frag = fpos_frag(filp->f_pos);
247 int off = fpos_off(filp->f_pos);
250 struct ceph_mds_reply_info_parsed *rinfo;
251 const int max_entries = fsc->mount_options->max_readdir;
252 const int max_bytes = fsc->mount_options->max_readdir_bytes;
254 dout("readdir %p filp %p frag %u off %u\n", inode, filp, frag, off);
255 if (fi->flags & CEPH_F_ATEND)
258 /* always start with . and .. */
259 if (filp->f_pos == 0) {
260 /* note dir version at start of readdir so we can tell
261 * if any dentries get dropped */
262 fi->dir_release_count = ci->i_release_count;
264 dout("readdir off 0 -> '.'\n");
265 if (filldir(dirent, ".", 1, ceph_make_fpos(0, 0),
266 ceph_translate_ino(inode->i_sb, inode->i_ino),
267 inode->i_mode >> 12) < 0)
272 if (filp->f_pos == 1) {
273 ino_t ino = filp->f_dentry->d_parent->d_inode->i_ino;
274 dout("readdir off 1 -> '..'\n");
275 if (filldir(dirent, "..", 2, ceph_make_fpos(0, 1),
276 ceph_translate_ino(inode->i_sb, ino),
277 inode->i_mode >> 12) < 0)
283 /* can we use the dcache? */
284 spin_lock(&inode->i_lock);
285 if ((filp->f_pos == 2 || fi->dentry) &&
286 !ceph_test_mount_opt(fsc, NOASYNCREADDIR) &&
287 ceph_snap(inode) != CEPH_SNAPDIR &&
288 (ci->i_ceph_flags & CEPH_I_COMPLETE) &&
289 __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1)) {
290 spin_unlock(&inode->i_lock);
291 err = __dcache_readdir(filp, dirent, filldir);
295 spin_unlock(&inode->i_lock);
298 err = note_last_dentry(fi, fi->dentry->d_name.name,
299 fi->dentry->d_name.len);
306 /* proceed with a normal readdir */
309 /* do we have the correct frag content buffered? */
310 if (fi->frag != frag || fi->last_readdir == NULL) {
311 struct ceph_mds_request *req;
312 int op = ceph_snap(inode) == CEPH_SNAPDIR ?
313 CEPH_MDS_OP_LSSNAP : CEPH_MDS_OP_READDIR;
315 /* discard old result, if any */
316 if (fi->last_readdir) {
317 ceph_mdsc_put_request(fi->last_readdir);
318 fi->last_readdir = NULL;
321 /* requery frag tree, as the frag topology may have changed */
322 frag = ceph_choose_frag(ceph_inode(inode), frag, NULL, NULL);
324 dout("readdir fetching %llx.%llx frag %x offset '%s'\n",
325 ceph_vinop(inode), frag, fi->last_name);
326 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
329 req->r_inode = inode;
331 req->r_dentry = dget(filp->f_dentry);
332 /* hints to request -> mds selection code */
333 req->r_direct_mode = USE_AUTH_MDS;
334 req->r_direct_hash = ceph_frag_value(frag);
335 req->r_direct_is_hash = true;
336 req->r_path2 = kstrdup(fi->last_name, GFP_NOFS);
337 req->r_readdir_offset = fi->next_offset;
338 req->r_args.readdir.frag = cpu_to_le32(frag);
339 req->r_args.readdir.max_entries = cpu_to_le32(max_entries);
340 req->r_args.readdir.max_bytes = cpu_to_le32(max_bytes);
341 req->r_num_caps = max_entries + 1;
342 err = ceph_mdsc_do_request(mdsc, NULL, req);
344 ceph_mdsc_put_request(req);
347 dout("readdir got and parsed readdir result=%d"
348 " on frag %x, end=%d, complete=%d\n", err, frag,
349 (int)req->r_reply_info.dir_end,
350 (int)req->r_reply_info.dir_complete);
352 if (!req->r_did_prepopulate) {
353 dout("readdir !did_prepopulate");
354 fi->dir_release_count--; /* preclude I_COMPLETE */
357 /* note next offset and last dentry name */
358 fi->offset = fi->next_offset;
359 fi->last_readdir = req;
361 if (req->r_reply_info.dir_end) {
362 kfree(fi->last_name);
363 fi->last_name = NULL;
364 if (ceph_frag_is_rightmost(frag))
369 rinfo = &req->r_reply_info;
370 err = note_last_dentry(fi,
371 rinfo->dir_dname[rinfo->dir_nr-1],
372 rinfo->dir_dname_len[rinfo->dir_nr-1]);
375 fi->next_offset += rinfo->dir_nr;
379 rinfo = &fi->last_readdir->r_reply_info;
380 dout("readdir frag %x num %d off %d chunkoff %d\n", frag,
381 rinfo->dir_nr, off, fi->offset);
382 while (off >= fi->offset && off - fi->offset < rinfo->dir_nr) {
383 u64 pos = ceph_make_fpos(frag, off);
384 struct ceph_mds_reply_inode *in =
385 rinfo->dir_in[off - fi->offset].in;
386 struct ceph_vino vino;
389 dout("readdir off %d (%d/%d) -> %lld '%.*s' %p\n",
390 off, off - fi->offset, rinfo->dir_nr, pos,
391 rinfo->dir_dname_len[off - fi->offset],
392 rinfo->dir_dname[off - fi->offset], in);
394 ftype = le32_to_cpu(in->mode) >> 12;
395 vino.ino = le64_to_cpu(in->ino);
396 vino.snap = le64_to_cpu(in->snapid);
397 ino = ceph_vino_to_ino(vino);
399 rinfo->dir_dname[off - fi->offset],
400 rinfo->dir_dname_len[off - fi->offset],
402 ceph_translate_ino(inode->i_sb, ino), ftype) < 0) {
403 dout("filldir stopping us...\n");
407 filp->f_pos = pos + 1;
411 ceph_mdsc_put_request(fi->last_readdir);
412 fi->last_readdir = NULL;
417 if (!ceph_frag_is_rightmost(frag)) {
418 frag = ceph_frag_next(frag);
420 filp->f_pos = ceph_make_fpos(frag, off);
421 dout("readdir next frag is %x\n", frag);
424 fi->flags |= CEPH_F_ATEND;
427 * if dir_release_count still matches the dir, no dentries
428 * were released during the whole readdir, and we should have
429 * the complete dir contents in our cache.
431 spin_lock(&inode->i_lock);
432 if (ci->i_release_count == fi->dir_release_count) {
433 dout(" marking %p complete\n", inode);
434 /* ci->i_ceph_flags |= CEPH_I_COMPLETE; */
435 ci->i_max_offset = filp->f_pos;
437 spin_unlock(&inode->i_lock);
439 dout("readdir %p filp %p done.\n", inode, filp);
443 static void reset_readdir(struct ceph_file_info *fi)
445 if (fi->last_readdir) {
446 ceph_mdsc_put_request(fi->last_readdir);
447 fi->last_readdir = NULL;
449 kfree(fi->last_name);
450 fi->last_name = NULL;
451 fi->next_offset = 2; /* compensate for . and .. */
456 fi->flags &= ~CEPH_F_ATEND;
459 static loff_t ceph_dir_llseek(struct file *file, loff_t offset, int origin)
461 struct ceph_file_info *fi = file->private_data;
462 struct inode *inode = file->f_mapping->host;
463 loff_t old_offset = offset;
466 mutex_lock(&inode->i_mutex);
469 offset += inode->i_size + 2; /* FIXME */
472 offset += file->f_pos;
475 if (offset >= 0 && offset <= inode->i_sb->s_maxbytes) {
476 if (offset != file->f_pos) {
477 file->f_pos = offset;
479 fi->flags &= ~CEPH_F_ATEND;
484 * discard buffered readdir content on seekdir(0), or
485 * seek to new frag, or seek prior to current chunk.
488 fpos_frag(offset) != fpos_frag(old_offset) ||
489 fpos_off(offset) < fi->offset) {
490 dout("dir_llseek dropping %p content\n", file);
494 /* bump dir_release_count if we did a forward seek */
495 if (offset > old_offset)
496 fi->dir_release_count--;
498 mutex_unlock(&inode->i_mutex);
503 * Handle lookups for the hidden .snap directory.
505 int ceph_handle_snapdir(struct ceph_mds_request *req,
506 struct dentry *dentry, int err)
508 struct ceph_fs_client *fsc = ceph_sb_to_client(dentry->d_sb);
509 struct inode *parent = dentry->d_parent->d_inode;
512 if (err == -ENOENT &&
513 ceph_snap(parent) == CEPH_NOSNAP &&
514 strcmp(dentry->d_name.name,
515 fsc->mount_options->snapdir_name) == 0) {
516 struct inode *inode = ceph_get_snapdir(parent);
517 dout("ENOENT on snapdir %p '%.*s', linking to snapdir %p\n",
518 dentry, dentry->d_name.len, dentry->d_name.name, inode);
519 BUG_ON(!d_unhashed(dentry));
520 d_add(dentry, inode);
527 * Figure out final result of a lookup/open request.
529 * Mainly, make sure we return the final req->r_dentry (if it already
530 * existed) in place of the original VFS-provided dentry when they
533 * Gracefully handle the case where the MDS replies with -ENOENT and
534 * no trace (which it may do, at its discretion, e.g., if it doesn't
535 * care to issue a lease on the negative dentry).
537 struct dentry *ceph_finish_lookup(struct ceph_mds_request *req,
538 struct dentry *dentry, int err)
540 if (err == -ENOENT) {
543 if (!req->r_reply_info.head->is_dentry) {
544 dout("ENOENT and no trace, dentry %p inode %p\n",
545 dentry, dentry->d_inode);
546 if (dentry->d_inode) {
555 dentry = ERR_PTR(err);
556 else if (dentry != req->r_dentry)
557 dentry = dget(req->r_dentry); /* we got spliced */
563 static int is_root_ceph_dentry(struct inode *inode, struct dentry *dentry)
565 return ceph_ino(inode) == CEPH_INO_ROOT &&
566 strncmp(dentry->d_name.name, ".ceph", 5) == 0;
570 * Look up a single dir entry. If there is a lookup intent, inform
571 * the MDS so that it gets our 'caps wanted' value in a single op.
573 static struct dentry *ceph_lookup(struct inode *dir, struct dentry *dentry,
574 struct nameidata *nd)
576 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
577 struct ceph_mds_client *mdsc = fsc->mdsc;
578 struct ceph_mds_request *req;
582 dout("lookup %p dentry %p '%.*s'\n",
583 dir, dentry, dentry->d_name.len, dentry->d_name.name);
585 if (dentry->d_name.len > NAME_MAX)
586 return ERR_PTR(-ENAMETOOLONG);
588 err = ceph_init_dentry(dentry);
592 /* open (but not create!) intent? */
594 (nd->flags & LOOKUP_OPEN) &&
595 (nd->flags & LOOKUP_CONTINUE) == 0 && /* only open last component */
596 !(nd->intent.open.flags & O_CREAT)) {
597 int mode = nd->intent.open.create_mode & ~current->fs->umask;
598 return ceph_lookup_open(dir, dentry, nd, mode, 1);
601 /* can we conclude ENOENT locally? */
602 if (dentry->d_inode == NULL) {
603 struct ceph_inode_info *ci = ceph_inode(dir);
604 struct ceph_dentry_info *di = ceph_dentry(dentry);
606 spin_lock(&dir->i_lock);
607 dout(" dir %p flags are %d\n", dir, ci->i_ceph_flags);
608 if (strncmp(dentry->d_name.name,
609 fsc->mount_options->snapdir_name,
610 dentry->d_name.len) &&
611 !is_root_ceph_dentry(dir, dentry) &&
612 (ci->i_ceph_flags & CEPH_I_COMPLETE) &&
613 (__ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1))) {
614 spin_unlock(&dir->i_lock);
615 dout(" dir %p complete, -ENOENT\n", dir);
617 di->lease_shared_gen = ci->i_shared_gen;
620 spin_unlock(&dir->i_lock);
623 op = ceph_snap(dir) == CEPH_SNAPDIR ?
624 CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP;
625 req = ceph_mdsc_create_request(mdsc, op, USE_ANY_MDS);
627 return ERR_CAST(req);
628 req->r_dentry = dget(dentry);
630 /* we only need inode linkage */
631 req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
632 req->r_locked_dir = dir;
633 err = ceph_mdsc_do_request(mdsc, NULL, req);
634 err = ceph_handle_snapdir(req, dentry, err);
635 dentry = ceph_finish_lookup(req, dentry, err);
636 ceph_mdsc_put_request(req); /* will dput(dentry) */
637 dout("lookup result=%p\n", dentry);
642 * If we do a create but get no trace back from the MDS, follow up with
643 * a lookup (the VFS expects us to link up the provided dentry).
645 int ceph_handle_notrace_create(struct inode *dir, struct dentry *dentry)
647 struct dentry *result = ceph_lookup(dir, dentry, NULL);
649 if (result && !IS_ERR(result)) {
651 * We created the item, then did a lookup, and found
652 * it was already linked to another inode we already
653 * had in our cache (and thus got spliced). Link our
654 * dentry to that inode, but don't hash it, just in
655 * case the VFS wants to dereference it.
657 BUG_ON(!result->d_inode);
658 d_instantiate(dentry, result->d_inode);
661 return PTR_ERR(result);
664 static int ceph_mknod(struct inode *dir, struct dentry *dentry,
665 int mode, dev_t rdev)
667 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
668 struct ceph_mds_client *mdsc = fsc->mdsc;
669 struct ceph_mds_request *req;
672 if (ceph_snap(dir) != CEPH_NOSNAP)
675 dout("mknod in dir %p dentry %p mode 0%o rdev %d\n",
676 dir, dentry, mode, rdev);
677 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_MKNOD, USE_AUTH_MDS);
682 req->r_dentry = dget(dentry);
684 req->r_locked_dir = dir;
685 req->r_args.mknod.mode = cpu_to_le32(mode);
686 req->r_args.mknod.rdev = cpu_to_le32(rdev);
687 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
688 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
689 err = ceph_mdsc_do_request(mdsc, dir, req);
690 if (!err && !req->r_reply_info.head->is_dentry)
691 err = ceph_handle_notrace_create(dir, dentry);
692 ceph_mdsc_put_request(req);
698 static int ceph_create(struct inode *dir, struct dentry *dentry, int mode,
699 struct nameidata *nd)
701 dout("create in dir %p dentry %p name '%.*s'\n",
702 dir, dentry, dentry->d_name.len, dentry->d_name.name);
704 if (ceph_snap(dir) != CEPH_NOSNAP)
708 BUG_ON((nd->flags & LOOKUP_OPEN) == 0);
709 dentry = ceph_lookup_open(dir, dentry, nd, mode, 0);
710 /* hrm, what should i do here if we get aliased? */
712 return PTR_ERR(dentry);
716 /* fall back to mknod */
717 return ceph_mknod(dir, dentry, (mode & ~S_IFMT) | S_IFREG, 0);
720 static int ceph_symlink(struct inode *dir, struct dentry *dentry,
723 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
724 struct ceph_mds_client *mdsc = fsc->mdsc;
725 struct ceph_mds_request *req;
728 if (ceph_snap(dir) != CEPH_NOSNAP)
731 dout("symlink in dir %p dentry %p to '%s'\n", dir, dentry, dest);
732 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SYMLINK, USE_AUTH_MDS);
737 req->r_dentry = dget(dentry);
739 req->r_path2 = kstrdup(dest, GFP_NOFS);
740 req->r_locked_dir = dir;
741 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
742 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
743 err = ceph_mdsc_do_request(mdsc, dir, req);
744 if (!err && !req->r_reply_info.head->is_dentry)
745 err = ceph_handle_notrace_create(dir, dentry);
746 ceph_mdsc_put_request(req);
752 static int ceph_mkdir(struct inode *dir, struct dentry *dentry, int mode)
754 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
755 struct ceph_mds_client *mdsc = fsc->mdsc;
756 struct ceph_mds_request *req;
760 if (ceph_snap(dir) == CEPH_SNAPDIR) {
761 /* mkdir .snap/foo is a MKSNAP */
762 op = CEPH_MDS_OP_MKSNAP;
763 dout("mksnap dir %p snap '%.*s' dn %p\n", dir,
764 dentry->d_name.len, dentry->d_name.name, dentry);
765 } else if (ceph_snap(dir) == CEPH_NOSNAP) {
766 dout("mkdir dir %p dn %p mode 0%o\n", dir, dentry, mode);
767 op = CEPH_MDS_OP_MKDIR;
771 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
777 req->r_dentry = dget(dentry);
779 req->r_locked_dir = dir;
780 req->r_args.mkdir.mode = cpu_to_le32(mode);
781 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
782 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
783 err = ceph_mdsc_do_request(mdsc, dir, req);
784 if (!err && !req->r_reply_info.head->is_dentry)
785 err = ceph_handle_notrace_create(dir, dentry);
786 ceph_mdsc_put_request(req);
793 static int ceph_link(struct dentry *old_dentry, struct inode *dir,
794 struct dentry *dentry)
796 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
797 struct ceph_mds_client *mdsc = fsc->mdsc;
798 struct ceph_mds_request *req;
801 if (ceph_snap(dir) != CEPH_NOSNAP)
804 dout("link in dir %p old_dentry %p dentry %p\n", dir,
806 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LINK, USE_AUTH_MDS);
811 req->r_dentry = dget(dentry);
813 req->r_old_dentry = dget(old_dentry); /* or inode? hrm. */
814 req->r_old_dentry_dir = ceph_get_dentry_parent_inode(old_dentry);
815 req->r_locked_dir = dir;
816 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
817 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
818 err = ceph_mdsc_do_request(mdsc, dir, req);
821 } else if (!req->r_reply_info.head->is_dentry) {
822 ihold(old_dentry->d_inode);
823 d_instantiate(dentry, old_dentry->d_inode);
825 ceph_mdsc_put_request(req);
830 * For a soon-to-be unlinked file, drop the AUTH_RDCACHE caps. If it
831 * looks like the link count will hit 0, drop any other caps (other
832 * than PIN) we don't specifically want (due to the file still being
835 static int drop_caps_for_unlink(struct inode *inode)
837 struct ceph_inode_info *ci = ceph_inode(inode);
838 int drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL;
840 spin_lock(&inode->i_lock);
841 if (inode->i_nlink == 1) {
842 drop |= ~(__ceph_caps_wanted(ci) | CEPH_CAP_PIN);
843 ci->i_ceph_flags |= CEPH_I_NODELAY;
845 spin_unlock(&inode->i_lock);
850 * rmdir and unlink are differ only by the metadata op code
852 static int ceph_unlink(struct inode *dir, struct dentry *dentry)
854 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
855 struct ceph_mds_client *mdsc = fsc->mdsc;
856 struct inode *inode = dentry->d_inode;
857 struct ceph_mds_request *req;
861 if (ceph_snap(dir) == CEPH_SNAPDIR) {
862 /* rmdir .snap/foo is RMSNAP */
863 dout("rmsnap dir %p '%.*s' dn %p\n", dir, dentry->d_name.len,
864 dentry->d_name.name, dentry);
865 op = CEPH_MDS_OP_RMSNAP;
866 } else if (ceph_snap(dir) == CEPH_NOSNAP) {
867 dout("unlink/rmdir dir %p dn %p inode %p\n",
869 op = ((dentry->d_inode->i_mode & S_IFMT) == S_IFDIR) ?
870 CEPH_MDS_OP_RMDIR : CEPH_MDS_OP_UNLINK;
873 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
878 req->r_dentry = dget(dentry);
880 req->r_locked_dir = dir;
881 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
882 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
883 req->r_inode_drop = drop_caps_for_unlink(inode);
884 err = ceph_mdsc_do_request(mdsc, dir, req);
885 if (!err && !req->r_reply_info.head->is_dentry)
887 ceph_mdsc_put_request(req);
892 static int ceph_rename(struct inode *old_dir, struct dentry *old_dentry,
893 struct inode *new_dir, struct dentry *new_dentry)
895 struct ceph_fs_client *fsc = ceph_sb_to_client(old_dir->i_sb);
896 struct ceph_mds_client *mdsc = fsc->mdsc;
897 struct ceph_mds_request *req;
900 if (ceph_snap(old_dir) != ceph_snap(new_dir))
902 if (ceph_snap(old_dir) != CEPH_NOSNAP ||
903 ceph_snap(new_dir) != CEPH_NOSNAP)
905 dout("rename dir %p dentry %p to dir %p dentry %p\n",
906 old_dir, old_dentry, new_dir, new_dentry);
907 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_RENAME, USE_AUTH_MDS);
910 req->r_dentry = dget(new_dentry);
912 req->r_old_dentry = dget(old_dentry);
913 req->r_old_dentry_dir = ceph_get_dentry_parent_inode(old_dentry);
914 req->r_locked_dir = new_dir;
915 req->r_old_dentry_drop = CEPH_CAP_FILE_SHARED;
916 req->r_old_dentry_unless = CEPH_CAP_FILE_EXCL;
917 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
918 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
919 /* release LINK_RDCACHE on source inode (mds will lock it) */
920 req->r_old_inode_drop = CEPH_CAP_LINK_SHARED;
921 if (new_dentry->d_inode)
922 req->r_inode_drop = drop_caps_for_unlink(new_dentry->d_inode);
923 err = ceph_mdsc_do_request(mdsc, old_dir, req);
924 if (!err && !req->r_reply_info.head->is_dentry) {
926 * Normally d_move() is done by fill_trace (called by
927 * do_request, above). If there is no trace, we need
931 /* d_move screws up d_subdirs order */
932 ceph_i_clear(new_dir, CEPH_I_COMPLETE);
934 d_move(old_dentry, new_dentry);
936 /* ensure target dentry is invalidated, despite
937 rehashing bug in vfs_rename_dir */
938 ceph_invalidate_dentry_lease(new_dentry);
940 ceph_mdsc_put_request(req);
945 * Ensure a dentry lease will no longer revalidate.
947 void ceph_invalidate_dentry_lease(struct dentry *dentry)
949 spin_lock(&dentry->d_lock);
950 dentry->d_time = jiffies;
951 ceph_dentry(dentry)->lease_shared_gen = 0;
952 spin_unlock(&dentry->d_lock);
956 * Check if dentry lease is valid. If not, delete the lease. Try to
957 * renew if the least is more than half up.
959 static int dentry_lease_is_valid(struct dentry *dentry)
961 struct ceph_dentry_info *di;
962 struct ceph_mds_session *s;
966 struct ceph_mds_session *session = NULL;
967 struct inode *dir = NULL;
970 spin_lock(&dentry->d_lock);
971 di = ceph_dentry(dentry);
972 if (di && di->lease_session) {
973 s = di->lease_session;
974 spin_lock(&s->s_cap_lock);
977 spin_unlock(&s->s_cap_lock);
979 if (di->lease_gen == gen &&
980 time_before(jiffies, dentry->d_time) &&
981 time_before(jiffies, ttl)) {
983 if (di->lease_renew_after &&
984 time_after(jiffies, di->lease_renew_after)) {
985 /* we should renew */
986 dir = dentry->d_parent->d_inode;
987 session = ceph_get_mds_session(s);
989 di->lease_renew_after = 0;
990 di->lease_renew_from = jiffies;
994 spin_unlock(&dentry->d_lock);
997 ceph_mdsc_lease_send_msg(session, dir, dentry,
998 CEPH_MDS_LEASE_RENEW, seq);
999 ceph_put_mds_session(session);
1001 dout("dentry_lease_is_valid - dentry %p = %d\n", dentry, valid);
1006 * Check if directory-wide content lease/cap is valid.
1008 static int dir_lease_is_valid(struct inode *dir, struct dentry *dentry)
1010 struct ceph_inode_info *ci = ceph_inode(dir);
1011 struct ceph_dentry_info *di = ceph_dentry(dentry);
1014 spin_lock(&dir->i_lock);
1015 if (ci->i_shared_gen == di->lease_shared_gen)
1016 valid = __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1);
1017 spin_unlock(&dir->i_lock);
1018 dout("dir_lease_is_valid dir %p v%u dentry %p v%u = %d\n",
1019 dir, (unsigned)ci->i_shared_gen, dentry,
1020 (unsigned)di->lease_shared_gen, valid);
1025 * Check if cached dentry can be trusted.
1027 static int ceph_d_revalidate(struct dentry *dentry, struct nameidata *nd)
1032 if (nd && nd->flags & LOOKUP_RCU)
1035 dout("d_revalidate %p '%.*s' inode %p offset %lld\n", dentry,
1036 dentry->d_name.len, dentry->d_name.name, dentry->d_inode,
1037 ceph_dentry(dentry)->offset);
1039 dir = ceph_get_dentry_parent_inode(dentry);
1041 /* always trust cached snapped dentries, snapdir dentry */
1042 if (ceph_snap(dir) != CEPH_NOSNAP) {
1043 dout("d_revalidate %p '%.*s' inode %p is SNAPPED\n", dentry,
1044 dentry->d_name.len, dentry->d_name.name, dentry->d_inode);
1046 } else if (dentry->d_inode &&
1047 ceph_snap(dentry->d_inode) == CEPH_SNAPDIR) {
1049 } else if (dentry_lease_is_valid(dentry) ||
1050 dir_lease_is_valid(dir, dentry)) {
1054 dout("d_revalidate %p %s\n", dentry, valid ? "valid" : "invalid");
1056 ceph_dentry_lru_touch(dentry);
1064 * Release our ceph_dentry_info.
1066 static void ceph_d_release(struct dentry *dentry)
1068 struct ceph_dentry_info *di = ceph_dentry(dentry);
1070 dout("d_release %p\n", dentry);
1072 ceph_dentry_lru_del(dentry);
1073 if (di->lease_session)
1074 ceph_put_mds_session(di->lease_session);
1075 kmem_cache_free(ceph_dentry_cachep, di);
1076 dentry->d_fsdata = NULL;
1080 static int ceph_snapdir_d_revalidate(struct dentry *dentry,
1081 struct nameidata *nd)
1084 * Eventually, we'll want to revalidate snapped metadata
1085 * too... probably...
1093 * read() on a dir. This weird interface hack only works if mounted
1094 * with '-o dirstat'.
1096 static ssize_t ceph_read_dir(struct file *file, char __user *buf, size_t size,
1099 struct ceph_file_info *cf = file->private_data;
1100 struct inode *inode = file->f_dentry->d_inode;
1101 struct ceph_inode_info *ci = ceph_inode(inode);
1103 const int bufsize = 1024;
1105 if (!ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb), DIRSTAT))
1108 if (!cf->dir_info) {
1109 cf->dir_info = kmalloc(bufsize, GFP_NOFS);
1113 snprintf(cf->dir_info, bufsize,
1116 " subdirs: %20lld\n"
1117 "rentries: %20lld\n"
1119 " rsubdirs: %20lld\n"
1121 "rctime: %10ld.%09ld\n",
1122 ci->i_files + ci->i_subdirs,
1125 ci->i_rfiles + ci->i_rsubdirs,
1129 (long)ci->i_rctime.tv_sec,
1130 (long)ci->i_rctime.tv_nsec);
1133 if (*ppos >= cf->dir_info_len)
1135 size = min_t(unsigned, size, cf->dir_info_len-*ppos);
1136 left = copy_to_user(buf, cf->dir_info + *ppos, size);
1139 *ppos += (size - left);
1144 * an fsync() on a dir will wait for any uncommitted directory
1145 * operations to commit.
1147 static int ceph_dir_fsync(struct file *file, int datasync)
1149 struct inode *inode = file->f_path.dentry->d_inode;
1150 struct ceph_inode_info *ci = ceph_inode(inode);
1151 struct list_head *head = &ci->i_unsafe_dirops;
1152 struct ceph_mds_request *req;
1156 dout("dir_fsync %p\n", inode);
1157 spin_lock(&ci->i_unsafe_lock);
1158 if (list_empty(head))
1161 req = list_entry(head->prev,
1162 struct ceph_mds_request, r_unsafe_dir_item);
1163 last_tid = req->r_tid;
1166 ceph_mdsc_get_request(req);
1167 spin_unlock(&ci->i_unsafe_lock);
1168 dout("dir_fsync %p wait on tid %llu (until %llu)\n",
1169 inode, req->r_tid, last_tid);
1170 if (req->r_timeout) {
1171 ret = wait_for_completion_timeout(
1172 &req->r_safe_completion, req->r_timeout);
1176 ret = -EIO; /* timed out */
1178 wait_for_completion(&req->r_safe_completion);
1180 spin_lock(&ci->i_unsafe_lock);
1181 ceph_mdsc_put_request(req);
1183 if (ret || list_empty(head))
1185 req = list_entry(head->next,
1186 struct ceph_mds_request, r_unsafe_dir_item);
1187 } while (req->r_tid < last_tid);
1189 spin_unlock(&ci->i_unsafe_lock);
1194 * We maintain a private dentry LRU.
1196 * FIXME: this needs to be changed to a per-mds lru to be useful.
1198 void ceph_dentry_lru_add(struct dentry *dn)
1200 struct ceph_dentry_info *di = ceph_dentry(dn);
1201 struct ceph_mds_client *mdsc;
1203 dout("dentry_lru_add %p %p '%.*s'\n", di, dn,
1204 dn->d_name.len, dn->d_name.name);
1206 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1207 spin_lock(&mdsc->dentry_lru_lock);
1208 list_add_tail(&di->lru, &mdsc->dentry_lru);
1210 spin_unlock(&mdsc->dentry_lru_lock);
1214 void ceph_dentry_lru_touch(struct dentry *dn)
1216 struct ceph_dentry_info *di = ceph_dentry(dn);
1217 struct ceph_mds_client *mdsc;
1219 dout("dentry_lru_touch %p %p '%.*s' (offset %lld)\n", di, dn,
1220 dn->d_name.len, dn->d_name.name, di->offset);
1222 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1223 spin_lock(&mdsc->dentry_lru_lock);
1224 list_move_tail(&di->lru, &mdsc->dentry_lru);
1225 spin_unlock(&mdsc->dentry_lru_lock);
1229 void ceph_dentry_lru_del(struct dentry *dn)
1231 struct ceph_dentry_info *di = ceph_dentry(dn);
1232 struct ceph_mds_client *mdsc;
1234 dout("dentry_lru_del %p %p '%.*s'\n", di, dn,
1235 dn->d_name.len, dn->d_name.name);
1237 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1238 spin_lock(&mdsc->dentry_lru_lock);
1239 list_del_init(&di->lru);
1241 spin_unlock(&mdsc->dentry_lru_lock);
1246 * Return name hash for a given dentry. This is dependent on
1247 * the parent directory's hash function.
1249 unsigned ceph_dentry_hash(struct inode *dir, struct dentry *dn)
1251 struct ceph_inode_info *dci = ceph_inode(dir);
1253 switch (dci->i_dir_layout.dl_dir_hash) {
1254 case 0: /* for backward compat */
1255 case CEPH_STR_HASH_LINUX:
1256 return dn->d_name.hash;
1259 return ceph_str_hash(dci->i_dir_layout.dl_dir_hash,
1260 dn->d_name.name, dn->d_name.len);
1264 const struct file_operations ceph_dir_fops = {
1265 .read = ceph_read_dir,
1266 .readdir = ceph_readdir,
1267 .llseek = ceph_dir_llseek,
1269 .release = ceph_release,
1270 .unlocked_ioctl = ceph_ioctl,
1271 .fsync = ceph_dir_fsync,
1274 const struct inode_operations ceph_dir_iops = {
1275 .lookup = ceph_lookup,
1276 .permission = ceph_permission,
1277 .getattr = ceph_getattr,
1278 .setattr = ceph_setattr,
1279 .setxattr = ceph_setxattr,
1280 .getxattr = ceph_getxattr,
1281 .listxattr = ceph_listxattr,
1282 .removexattr = ceph_removexattr,
1283 .mknod = ceph_mknod,
1284 .symlink = ceph_symlink,
1285 .mkdir = ceph_mkdir,
1287 .unlink = ceph_unlink,
1288 .rmdir = ceph_unlink,
1289 .rename = ceph_rename,
1290 .create = ceph_create,
1293 const struct dentry_operations ceph_dentry_ops = {
1294 .d_revalidate = ceph_d_revalidate,
1295 .d_release = ceph_d_release,
1298 const struct dentry_operations ceph_snapdir_dentry_ops = {
1299 .d_revalidate = ceph_snapdir_d_revalidate,
1300 .d_release = ceph_d_release,
1303 const struct dentry_operations ceph_snap_dentry_ops = {
1304 .d_release = ceph_d_release,