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 if (dentry->d_parent == NULL || /* nfs fh_to_dentry */
44 ceph_snap(dentry->d_parent->d_inode) == CEPH_NOSNAP)
45 dentry->d_op = &ceph_dentry_ops;
46 else if (ceph_snap(dentry->d_parent->d_inode) == CEPH_SNAPDIR)
47 dentry->d_op = &ceph_snapdir_dentry_ops;
49 dentry->d_op = &ceph_snap_dentry_ops;
51 di = kmem_cache_alloc(ceph_dentry_cachep, GFP_NOFS | __GFP_ZERO);
53 return -ENOMEM; /* oh well */
55 spin_lock(&dentry->d_lock);
56 if (dentry->d_fsdata) {
58 kmem_cache_free(ceph_dentry_cachep, di);
62 di->lease_session = NULL;
63 di->parent_inode = igrab(dentry->d_parent->d_inode);
64 dentry->d_fsdata = di;
65 dentry->d_time = jiffies;
66 ceph_dentry_lru_add(dentry);
68 spin_unlock(&dentry->d_lock);
75 * for readdir, we encode the directory frag and offset within that
78 static unsigned fpos_frag(loff_t p)
82 static unsigned fpos_off(loff_t p)
84 return p & 0xffffffff;
88 * When possible, we try to satisfy a readdir by peeking at the
89 * dcache. We make this work by carefully ordering dentries on
90 * d_u.d_child when we initially get results back from the MDS, and
91 * falling back to a "normal" sync readdir if any dentries in the dir
94 * I_COMPLETE tells indicates we have all dentries in the dir. It is
95 * defined IFF we hold CEPH_CAP_FILE_SHARED (which will be revoked by
96 * the MDS if/when the directory is modified).
98 static int __dcache_readdir(struct file *filp,
99 void *dirent, filldir_t filldir)
101 struct ceph_file_info *fi = filp->private_data;
102 struct dentry *parent = filp->f_dentry;
103 struct inode *dir = parent->d_inode;
105 struct dentry *dentry, *last;
106 struct ceph_dentry_info *di;
109 /* claim ref on last dentry we returned */
113 dout("__dcache_readdir %p at %llu (last %p)\n", dir, filp->f_pos,
116 spin_lock(&dcache_lock);
118 /* start at beginning? */
119 if (filp->f_pos == 2 || last == NULL ||
120 filp->f_pos < ceph_dentry(last)->offset) {
121 if (list_empty(&parent->d_subdirs))
123 p = parent->d_subdirs.prev;
124 dout(" initial p %p/%p\n", p->prev, p->next);
126 p = last->d_u.d_child.prev;
130 dentry = list_entry(p, struct dentry, d_u.d_child);
131 di = ceph_dentry(dentry);
133 dout(" p %p/%p %s d_subdirs %p/%p\n", p->prev, p->next,
134 d_unhashed(dentry) ? "!hashed" : "hashed",
135 parent->d_subdirs.prev, parent->d_subdirs.next);
136 if (p == &parent->d_subdirs) {
140 if (!d_unhashed(dentry) && dentry->d_inode &&
141 ceph_snap(dentry->d_inode) != CEPH_SNAPDIR &&
142 ceph_ino(dentry->d_inode) != CEPH_INO_CEPH &&
143 filp->f_pos <= di->offset)
145 dout(" skipping %p %.*s at %llu (%llu)%s%s\n", dentry,
146 dentry->d_name.len, dentry->d_name.name, di->offset,
147 filp->f_pos, d_unhashed(dentry) ? " unhashed" : "",
148 !dentry->d_inode ? " null" : "");
150 dentry = list_entry(p, struct dentry, d_u.d_child);
151 di = ceph_dentry(dentry);
154 atomic_inc(&dentry->d_count);
155 spin_unlock(&dcache_lock);
157 dout(" %llu (%llu) dentry %p %.*s %p\n", di->offset, filp->f_pos,
158 dentry, dentry->d_name.len, dentry->d_name.name, dentry->d_inode);
159 filp->f_pos = di->offset;
160 err = filldir(dirent, dentry->d_name.name,
161 dentry->d_name.len, di->offset,
162 dentry->d_inode->i_ino,
163 dentry->d_inode->i_mode >> 12);
167 /* remember our position */
169 fi->next_offset = di->offset;
181 /* make sure a dentry wasn't dropped while we didn't have dcache_lock */
182 if (!ceph_i_test(dir, CEPH_I_COMPLETE)) {
183 dout(" lost I_COMPLETE on %p; falling back to mds\n", dir);
188 spin_lock(&dcache_lock);
189 p = p->prev; /* advance to next dentry */
193 spin_unlock(&dcache_lock);
201 * make note of the last dentry we read, so we can
202 * continue at the same lexicographical point,
203 * regardless of what dir changes take place on the
206 static int note_last_dentry(struct ceph_file_info *fi, const char *name,
209 kfree(fi->last_name);
210 fi->last_name = kmalloc(len+1, GFP_NOFS);
213 memcpy(fi->last_name, name, len);
214 fi->last_name[len] = 0;
215 dout("note_last_dentry '%s'\n", fi->last_name);
219 static int ceph_readdir(struct file *filp, void *dirent, filldir_t filldir)
221 struct ceph_file_info *fi = filp->private_data;
222 struct inode *inode = filp->f_dentry->d_inode;
223 struct ceph_inode_info *ci = ceph_inode(inode);
224 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
225 struct ceph_mds_client *mdsc = fsc->mdsc;
226 unsigned frag = fpos_frag(filp->f_pos);
227 int off = fpos_off(filp->f_pos);
230 struct ceph_mds_reply_info_parsed *rinfo;
231 const int max_entries = fsc->mount_options->max_readdir;
232 const int max_bytes = fsc->mount_options->max_readdir_bytes;
234 dout("readdir %p filp %p frag %u off %u\n", inode, filp, frag, off);
238 /* always start with . and .. */
239 if (filp->f_pos == 0) {
240 /* note dir version at start of readdir so we can tell
241 * if any dentries get dropped */
242 fi->dir_release_count = ci->i_release_count;
244 dout("readdir off 0 -> '.'\n");
245 if (filldir(dirent, ".", 1, ceph_make_fpos(0, 0),
246 inode->i_ino, inode->i_mode >> 12) < 0)
251 if (filp->f_pos == 1) {
252 dout("readdir off 1 -> '..'\n");
253 if (filldir(dirent, "..", 2, ceph_make_fpos(0, 1),
254 filp->f_dentry->d_parent->d_inode->i_ino,
255 inode->i_mode >> 12) < 0)
261 /* can we use the dcache? */
262 spin_lock(&inode->i_lock);
263 if ((filp->f_pos == 2 || fi->dentry) &&
264 !ceph_test_mount_opt(fsc, NOASYNCREADDIR) &&
265 ceph_snap(inode) != CEPH_SNAPDIR &&
266 (ci->i_ceph_flags & CEPH_I_COMPLETE) &&
267 __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1)) {
268 spin_unlock(&inode->i_lock);
269 err = __dcache_readdir(filp, dirent, filldir);
273 spin_unlock(&inode->i_lock);
276 err = note_last_dentry(fi, fi->dentry->d_name.name,
277 fi->dentry->d_name.len);
284 /* proceed with a normal readdir */
287 /* do we have the correct frag content buffered? */
288 if (fi->frag != frag || fi->last_readdir == NULL) {
289 struct ceph_mds_request *req;
290 int op = ceph_snap(inode) == CEPH_SNAPDIR ?
291 CEPH_MDS_OP_LSSNAP : CEPH_MDS_OP_READDIR;
293 /* discard old result, if any */
294 if (fi->last_readdir) {
295 ceph_mdsc_put_request(fi->last_readdir);
296 fi->last_readdir = NULL;
299 /* requery frag tree, as the frag topology may have changed */
300 frag = ceph_choose_frag(ceph_inode(inode), frag, NULL, NULL);
302 dout("readdir fetching %llx.%llx frag %x offset '%s'\n",
303 ceph_vinop(inode), frag, fi->last_name);
304 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
307 req->r_inode = igrab(inode);
308 req->r_dentry = dget(filp->f_dentry);
309 /* hints to request -> mds selection code */
310 req->r_direct_mode = USE_AUTH_MDS;
311 req->r_direct_hash = ceph_frag_value(frag);
312 req->r_direct_is_hash = true;
313 req->r_path2 = kstrdup(fi->last_name, GFP_NOFS);
314 req->r_readdir_offset = fi->next_offset;
315 req->r_args.readdir.frag = cpu_to_le32(frag);
316 req->r_args.readdir.max_entries = cpu_to_le32(max_entries);
317 req->r_args.readdir.max_bytes = cpu_to_le32(max_bytes);
318 req->r_num_caps = max_entries + 1;
319 err = ceph_mdsc_do_request(mdsc, NULL, req);
321 ceph_mdsc_put_request(req);
324 dout("readdir got and parsed readdir result=%d"
325 " on frag %x, end=%d, complete=%d\n", err, frag,
326 (int)req->r_reply_info.dir_end,
327 (int)req->r_reply_info.dir_complete);
329 if (!req->r_did_prepopulate) {
330 dout("readdir !did_prepopulate");
331 fi->dir_release_count--; /* preclude I_COMPLETE */
334 /* note next offset and last dentry name */
335 fi->offset = fi->next_offset;
336 fi->last_readdir = req;
338 if (req->r_reply_info.dir_end) {
339 kfree(fi->last_name);
340 fi->last_name = NULL;
341 if (ceph_frag_is_rightmost(frag))
346 rinfo = &req->r_reply_info;
347 err = note_last_dentry(fi,
348 rinfo->dir_dname[rinfo->dir_nr-1],
349 rinfo->dir_dname_len[rinfo->dir_nr-1]);
352 fi->next_offset += rinfo->dir_nr;
356 rinfo = &fi->last_readdir->r_reply_info;
357 dout("readdir frag %x num %d off %d chunkoff %d\n", frag,
358 rinfo->dir_nr, off, fi->offset);
359 while (off - fi->offset >= 0 && off - fi->offset < rinfo->dir_nr) {
360 u64 pos = ceph_make_fpos(frag, off);
361 struct ceph_mds_reply_inode *in =
362 rinfo->dir_in[off - fi->offset].in;
363 struct ceph_vino vino;
366 dout("readdir off %d (%d/%d) -> %lld '%.*s' %p\n",
367 off, off - fi->offset, rinfo->dir_nr, pos,
368 rinfo->dir_dname_len[off - fi->offset],
369 rinfo->dir_dname[off - fi->offset], in);
371 ftype = le32_to_cpu(in->mode) >> 12;
372 vino.ino = le64_to_cpu(in->ino);
373 vino.snap = le64_to_cpu(in->snapid);
374 ino = ceph_vino_to_ino(vino);
376 rinfo->dir_dname[off - fi->offset],
377 rinfo->dir_dname_len[off - fi->offset],
378 pos, ino, ftype) < 0) {
379 dout("filldir stopping us...\n");
383 filp->f_pos = pos + 1;
387 ceph_mdsc_put_request(fi->last_readdir);
388 fi->last_readdir = NULL;
393 if (!ceph_frag_is_rightmost(frag)) {
394 frag = ceph_frag_next(frag);
396 filp->f_pos = ceph_make_fpos(frag, off);
397 dout("readdir next frag is %x\n", frag);
403 * if dir_release_count still matches the dir, no dentries
404 * were released during the whole readdir, and we should have
405 * the complete dir contents in our cache.
407 spin_lock(&inode->i_lock);
408 if (ci->i_release_count == fi->dir_release_count) {
409 dout(" marking %p complete\n", inode);
410 ci->i_ceph_flags |= CEPH_I_COMPLETE;
411 ci->i_max_offset = filp->f_pos;
413 spin_unlock(&inode->i_lock);
415 dout("readdir %p filp %p done.\n", inode, filp);
419 static void reset_readdir(struct ceph_file_info *fi)
421 if (fi->last_readdir) {
422 ceph_mdsc_put_request(fi->last_readdir);
423 fi->last_readdir = NULL;
425 kfree(fi->last_name);
426 fi->last_name = NULL;
427 fi->next_offset = 2; /* compensate for . and .. */
435 static loff_t ceph_dir_llseek(struct file *file, loff_t offset, int origin)
437 struct ceph_file_info *fi = file->private_data;
438 struct inode *inode = file->f_mapping->host;
439 loff_t old_offset = offset;
442 mutex_lock(&inode->i_mutex);
445 offset += inode->i_size + 2; /* FIXME */
448 offset += file->f_pos;
451 if (offset >= 0 && offset <= inode->i_sb->s_maxbytes) {
452 if (offset != file->f_pos) {
453 file->f_pos = offset;
460 * discard buffered readdir content on seekdir(0), or
461 * seek to new frag, or seek prior to current chunk.
464 fpos_frag(offset) != fpos_frag(old_offset) ||
465 fpos_off(offset) < fi->offset) {
466 dout("dir_llseek dropping %p content\n", file);
470 /* bump dir_release_count if we did a forward seek */
471 if (offset > old_offset)
472 fi->dir_release_count--;
474 mutex_unlock(&inode->i_mutex);
479 * Process result of a lookup/open request.
481 * Mainly, make sure we return the final req->r_dentry (if it already
482 * existed) in place of the original VFS-provided dentry when they
485 * Gracefully handle the case where the MDS replies with -ENOENT and
486 * no trace (which it may do, at its discretion, e.g., if it doesn't
487 * care to issue a lease on the negative dentry).
489 struct dentry *ceph_finish_lookup(struct ceph_mds_request *req,
490 struct dentry *dentry, int err)
492 struct ceph_fs_client *fsc = ceph_sb_to_client(dentry->d_sb);
493 struct inode *parent = dentry->d_parent->d_inode;
496 if (err == -ENOENT &&
497 strcmp(dentry->d_name.name,
498 fsc->mount_options->snapdir_name) == 0) {
499 struct inode *inode = ceph_get_snapdir(parent);
500 dout("ENOENT on snapdir %p '%.*s', linking to snapdir %p\n",
501 dentry, dentry->d_name.len, dentry->d_name.name, inode);
502 BUG_ON(!d_unhashed(dentry));
503 d_add(dentry, inode);
507 if (err == -ENOENT) {
510 if (!req->r_reply_info.head->is_dentry) {
511 dout("ENOENT and no trace, dentry %p inode %p\n",
512 dentry, dentry->d_inode);
513 if (dentry->d_inode) {
522 dentry = ERR_PTR(err);
523 else if (dentry != req->r_dentry)
524 dentry = dget(req->r_dentry); /* we got spliced */
530 static int is_root_ceph_dentry(struct inode *inode, struct dentry *dentry)
532 return ceph_ino(inode) == CEPH_INO_ROOT &&
533 strncmp(dentry->d_name.name, ".ceph", 5) == 0;
537 * Look up a single dir entry. If there is a lookup intent, inform
538 * the MDS so that it gets our 'caps wanted' value in a single op.
540 static struct dentry *ceph_lookup(struct inode *dir, struct dentry *dentry,
541 struct nameidata *nd)
543 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
544 struct ceph_mds_client *mdsc = fsc->mdsc;
545 struct ceph_mds_request *req;
549 dout("lookup %p dentry %p '%.*s'\n",
550 dir, dentry, dentry->d_name.len, dentry->d_name.name);
552 if (dentry->d_name.len > NAME_MAX)
553 return ERR_PTR(-ENAMETOOLONG);
555 err = ceph_init_dentry(dentry);
559 /* open (but not create!) intent? */
561 (nd->flags & LOOKUP_OPEN) &&
562 (nd->flags & LOOKUP_CONTINUE) == 0 && /* only open last component */
563 !(nd->intent.open.flags & O_CREAT)) {
564 int mode = nd->intent.open.create_mode & ~current->fs->umask;
565 return ceph_lookup_open(dir, dentry, nd, mode, 1);
568 /* can we conclude ENOENT locally? */
569 if (dentry->d_inode == NULL) {
570 struct ceph_inode_info *ci = ceph_inode(dir);
571 struct ceph_dentry_info *di = ceph_dentry(dentry);
573 spin_lock(&dir->i_lock);
574 dout(" dir %p flags are %d\n", dir, ci->i_ceph_flags);
575 if (strncmp(dentry->d_name.name,
576 fsc->mount_options->snapdir_name,
577 dentry->d_name.len) &&
578 !is_root_ceph_dentry(dir, dentry) &&
579 (ci->i_ceph_flags & CEPH_I_COMPLETE) &&
580 (__ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1))) {
581 spin_unlock(&dir->i_lock);
582 dout(" dir %p complete, -ENOENT\n", dir);
584 di->lease_shared_gen = ci->i_shared_gen;
587 spin_unlock(&dir->i_lock);
590 op = ceph_snap(dir) == CEPH_SNAPDIR ?
591 CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP;
592 req = ceph_mdsc_create_request(mdsc, op, USE_ANY_MDS);
594 return ERR_CAST(req);
595 req->r_dentry = dget(dentry);
597 /* we only need inode linkage */
598 req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
599 req->r_locked_dir = dir;
600 err = ceph_mdsc_do_request(mdsc, NULL, req);
601 dentry = ceph_finish_lookup(req, dentry, err);
602 ceph_mdsc_put_request(req); /* will dput(dentry) */
603 dout("lookup result=%p\n", dentry);
608 * If we do a create but get no trace back from the MDS, follow up with
609 * a lookup (the VFS expects us to link up the provided dentry).
611 int ceph_handle_notrace_create(struct inode *dir, struct dentry *dentry)
613 struct dentry *result = ceph_lookup(dir, dentry, NULL);
615 if (result && !IS_ERR(result)) {
617 * We created the item, then did a lookup, and found
618 * it was already linked to another inode we already
619 * had in our cache (and thus got spliced). Link our
620 * dentry to that inode, but don't hash it, just in
621 * case the VFS wants to dereference it.
623 BUG_ON(!result->d_inode);
624 d_instantiate(dentry, result->d_inode);
627 return PTR_ERR(result);
630 static int ceph_mknod(struct inode *dir, struct dentry *dentry,
631 int mode, dev_t rdev)
633 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
634 struct ceph_mds_client *mdsc = fsc->mdsc;
635 struct ceph_mds_request *req;
638 if (ceph_snap(dir) != CEPH_NOSNAP)
641 dout("mknod in dir %p dentry %p mode 0%o rdev %d\n",
642 dir, dentry, mode, rdev);
643 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_MKNOD, USE_AUTH_MDS);
648 req->r_dentry = dget(dentry);
650 req->r_locked_dir = dir;
651 req->r_args.mknod.mode = cpu_to_le32(mode);
652 req->r_args.mknod.rdev = cpu_to_le32(rdev);
653 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
654 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
655 err = ceph_mdsc_do_request(mdsc, dir, req);
656 if (!err && !req->r_reply_info.head->is_dentry)
657 err = ceph_handle_notrace_create(dir, dentry);
658 ceph_mdsc_put_request(req);
664 static int ceph_create(struct inode *dir, struct dentry *dentry, int mode,
665 struct nameidata *nd)
667 dout("create in dir %p dentry %p name '%.*s'\n",
668 dir, dentry, dentry->d_name.len, dentry->d_name.name);
670 if (ceph_snap(dir) != CEPH_NOSNAP)
674 BUG_ON((nd->flags & LOOKUP_OPEN) == 0);
675 dentry = ceph_lookup_open(dir, dentry, nd, mode, 0);
676 /* hrm, what should i do here if we get aliased? */
678 return PTR_ERR(dentry);
682 /* fall back to mknod */
683 return ceph_mknod(dir, dentry, (mode & ~S_IFMT) | S_IFREG, 0);
686 static int ceph_symlink(struct inode *dir, struct dentry *dentry,
689 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
690 struct ceph_mds_client *mdsc = fsc->mdsc;
691 struct ceph_mds_request *req;
694 if (ceph_snap(dir) != CEPH_NOSNAP)
697 dout("symlink in dir %p dentry %p to '%s'\n", dir, dentry, dest);
698 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SYMLINK, USE_AUTH_MDS);
703 req->r_dentry = dget(dentry);
705 req->r_path2 = kstrdup(dest, GFP_NOFS);
706 req->r_locked_dir = dir;
707 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
708 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
709 err = ceph_mdsc_do_request(mdsc, dir, req);
710 if (!err && !req->r_reply_info.head->is_dentry)
711 err = ceph_handle_notrace_create(dir, dentry);
712 ceph_mdsc_put_request(req);
718 static int ceph_mkdir(struct inode *dir, struct dentry *dentry, int mode)
720 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
721 struct ceph_mds_client *mdsc = fsc->mdsc;
722 struct ceph_mds_request *req;
726 if (ceph_snap(dir) == CEPH_SNAPDIR) {
727 /* mkdir .snap/foo is a MKSNAP */
728 op = CEPH_MDS_OP_MKSNAP;
729 dout("mksnap dir %p snap '%.*s' dn %p\n", dir,
730 dentry->d_name.len, dentry->d_name.name, dentry);
731 } else if (ceph_snap(dir) == CEPH_NOSNAP) {
732 dout("mkdir dir %p dn %p mode 0%o\n", dir, dentry, mode);
733 op = CEPH_MDS_OP_MKDIR;
737 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
743 req->r_dentry = dget(dentry);
745 req->r_locked_dir = dir;
746 req->r_args.mkdir.mode = cpu_to_le32(mode);
747 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
748 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
749 err = ceph_mdsc_do_request(mdsc, dir, req);
750 if (!err && !req->r_reply_info.head->is_dentry)
751 err = ceph_handle_notrace_create(dir, dentry);
752 ceph_mdsc_put_request(req);
759 static int ceph_link(struct dentry *old_dentry, struct inode *dir,
760 struct dentry *dentry)
762 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
763 struct ceph_mds_client *mdsc = fsc->mdsc;
764 struct ceph_mds_request *req;
767 if (ceph_snap(dir) != CEPH_NOSNAP)
770 dout("link in dir %p old_dentry %p dentry %p\n", dir,
772 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LINK, USE_AUTH_MDS);
777 req->r_dentry = dget(dentry);
779 req->r_old_dentry = dget(old_dentry); /* or inode? hrm. */
780 req->r_locked_dir = dir;
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);
786 else if (!req->r_reply_info.head->is_dentry)
787 d_instantiate(dentry, igrab(old_dentry->d_inode));
788 ceph_mdsc_put_request(req);
793 * For a soon-to-be unlinked file, drop the AUTH_RDCACHE caps. If it
794 * looks like the link count will hit 0, drop any other caps (other
795 * than PIN) we don't specifically want (due to the file still being
798 static int drop_caps_for_unlink(struct inode *inode)
800 struct ceph_inode_info *ci = ceph_inode(inode);
801 int drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL;
803 spin_lock(&inode->i_lock);
804 if (inode->i_nlink == 1) {
805 drop |= ~(__ceph_caps_wanted(ci) | CEPH_CAP_PIN);
806 ci->i_ceph_flags |= CEPH_I_NODELAY;
808 spin_unlock(&inode->i_lock);
813 * rmdir and unlink are differ only by the metadata op code
815 static int ceph_unlink(struct inode *dir, struct dentry *dentry)
817 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
818 struct ceph_mds_client *mdsc = fsc->mdsc;
819 struct inode *inode = dentry->d_inode;
820 struct ceph_mds_request *req;
824 if (ceph_snap(dir) == CEPH_SNAPDIR) {
825 /* rmdir .snap/foo is RMSNAP */
826 dout("rmsnap dir %p '%.*s' dn %p\n", dir, dentry->d_name.len,
827 dentry->d_name.name, dentry);
828 op = CEPH_MDS_OP_RMSNAP;
829 } else if (ceph_snap(dir) == CEPH_NOSNAP) {
830 dout("unlink/rmdir dir %p dn %p inode %p\n",
832 op = ((dentry->d_inode->i_mode & S_IFMT) == S_IFDIR) ?
833 CEPH_MDS_OP_RMDIR : CEPH_MDS_OP_UNLINK;
836 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
841 req->r_dentry = dget(dentry);
843 req->r_locked_dir = dir;
844 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
845 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
846 req->r_inode_drop = drop_caps_for_unlink(inode);
847 err = ceph_mdsc_do_request(mdsc, dir, req);
848 if (!err && !req->r_reply_info.head->is_dentry)
850 ceph_mdsc_put_request(req);
855 static int ceph_rename(struct inode *old_dir, struct dentry *old_dentry,
856 struct inode *new_dir, struct dentry *new_dentry)
858 struct ceph_fs_client *fsc = ceph_sb_to_client(old_dir->i_sb);
859 struct ceph_mds_client *mdsc = fsc->mdsc;
860 struct ceph_mds_request *req;
863 if (ceph_snap(old_dir) != ceph_snap(new_dir))
865 if (ceph_snap(old_dir) != CEPH_NOSNAP ||
866 ceph_snap(new_dir) != CEPH_NOSNAP)
868 dout("rename dir %p dentry %p to dir %p dentry %p\n",
869 old_dir, old_dentry, new_dir, new_dentry);
870 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_RENAME, USE_AUTH_MDS);
873 req->r_dentry = dget(new_dentry);
875 req->r_old_dentry = dget(old_dentry);
876 req->r_locked_dir = new_dir;
877 req->r_old_dentry_drop = CEPH_CAP_FILE_SHARED;
878 req->r_old_dentry_unless = CEPH_CAP_FILE_EXCL;
879 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
880 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
881 /* release LINK_RDCACHE on source inode (mds will lock it) */
882 req->r_old_inode_drop = CEPH_CAP_LINK_SHARED;
883 if (new_dentry->d_inode)
884 req->r_inode_drop = drop_caps_for_unlink(new_dentry->d_inode);
885 err = ceph_mdsc_do_request(mdsc, old_dir, req);
886 if (!err && !req->r_reply_info.head->is_dentry) {
888 * Normally d_move() is done by fill_trace (called by
889 * do_request, above). If there is no trace, we need
893 /* d_move screws up d_subdirs order */
894 ceph_i_clear(new_dir, CEPH_I_COMPLETE);
896 d_move(old_dentry, new_dentry);
898 /* ensure target dentry is invalidated, despite
899 rehashing bug in vfs_rename_dir */
900 ceph_invalidate_dentry_lease(new_dentry);
902 ceph_mdsc_put_request(req);
907 * Ensure a dentry lease will no longer revalidate.
909 void ceph_invalidate_dentry_lease(struct dentry *dentry)
911 spin_lock(&dentry->d_lock);
912 dentry->d_time = jiffies;
913 ceph_dentry(dentry)->lease_shared_gen = 0;
914 spin_unlock(&dentry->d_lock);
918 * Check if dentry lease is valid. If not, delete the lease. Try to
919 * renew if the least is more than half up.
921 static int dentry_lease_is_valid(struct dentry *dentry)
923 struct ceph_dentry_info *di;
924 struct ceph_mds_session *s;
928 struct ceph_mds_session *session = NULL;
929 struct inode *dir = NULL;
932 spin_lock(&dentry->d_lock);
933 di = ceph_dentry(dentry);
934 if (di && di->lease_session) {
935 s = di->lease_session;
936 spin_lock(&s->s_cap_lock);
939 spin_unlock(&s->s_cap_lock);
941 if (di->lease_gen == gen &&
942 time_before(jiffies, dentry->d_time) &&
943 time_before(jiffies, ttl)) {
945 if (di->lease_renew_after &&
946 time_after(jiffies, di->lease_renew_after)) {
947 /* we should renew */
948 dir = dentry->d_parent->d_inode;
949 session = ceph_get_mds_session(s);
951 di->lease_renew_after = 0;
952 di->lease_renew_from = jiffies;
956 spin_unlock(&dentry->d_lock);
959 ceph_mdsc_lease_send_msg(session, dir, dentry,
960 CEPH_MDS_LEASE_RENEW, seq);
961 ceph_put_mds_session(session);
963 dout("dentry_lease_is_valid - dentry %p = %d\n", dentry, valid);
968 * Check if directory-wide content lease/cap is valid.
970 static int dir_lease_is_valid(struct inode *dir, struct dentry *dentry)
972 struct ceph_inode_info *ci = ceph_inode(dir);
973 struct ceph_dentry_info *di = ceph_dentry(dentry);
976 spin_lock(&dir->i_lock);
977 if (ci->i_shared_gen == di->lease_shared_gen)
978 valid = __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1);
979 spin_unlock(&dir->i_lock);
980 dout("dir_lease_is_valid dir %p v%u dentry %p v%u = %d\n",
981 dir, (unsigned)ci->i_shared_gen, dentry,
982 (unsigned)di->lease_shared_gen, valid);
987 * Check if cached dentry can be trusted.
989 static int ceph_d_revalidate(struct dentry *dentry, struct nameidata *nd)
991 struct inode *dir = dentry->d_parent->d_inode;
993 dout("d_revalidate %p '%.*s' inode %p offset %lld\n", dentry,
994 dentry->d_name.len, dentry->d_name.name, dentry->d_inode,
995 ceph_dentry(dentry)->offset);
997 /* always trust cached snapped dentries, snapdir dentry */
998 if (ceph_snap(dir) != CEPH_NOSNAP) {
999 dout("d_revalidate %p '%.*s' inode %p is SNAPPED\n", dentry,
1000 dentry->d_name.len, dentry->d_name.name, dentry->d_inode);
1003 if (dentry->d_inode && ceph_snap(dentry->d_inode) == CEPH_SNAPDIR)
1006 if (dentry_lease_is_valid(dentry) ||
1007 dir_lease_is_valid(dir, dentry))
1010 dout("d_revalidate %p invalid\n", dentry);
1014 ceph_dentry_lru_touch(dentry);
1019 * When a dentry is released, clear the dir I_COMPLETE if it was part
1020 * of the current dir gen or if this is in the snapshot namespace.
1022 static void ceph_dentry_release(struct dentry *dentry)
1024 struct ceph_dentry_info *di = ceph_dentry(dentry);
1025 struct inode *parent_inode = NULL;
1026 u64 snapid = CEPH_NOSNAP;
1028 if (!IS_ROOT(dentry)) {
1029 parent_inode = di->parent_inode;
1031 snapid = ceph_snap(parent_inode);
1033 dout("dentry_release %p parent %p\n", dentry, parent_inode);
1034 if (parent_inode && snapid != CEPH_SNAPDIR) {
1035 struct ceph_inode_info *ci = ceph_inode(parent_inode);
1037 spin_lock(&parent_inode->i_lock);
1038 if (ci->i_shared_gen == di->lease_shared_gen ||
1039 snapid <= CEPH_MAXSNAP) {
1040 dout(" clearing %p complete (d_release)\n",
1042 ci->i_ceph_flags &= ~CEPH_I_COMPLETE;
1043 ci->i_release_count++;
1045 spin_unlock(&parent_inode->i_lock);
1048 ceph_dentry_lru_del(dentry);
1049 if (di->lease_session)
1050 ceph_put_mds_session(di->lease_session);
1051 kmem_cache_free(ceph_dentry_cachep, di);
1052 dentry->d_fsdata = NULL;
1058 static int ceph_snapdir_d_revalidate(struct dentry *dentry,
1059 struct nameidata *nd)
1062 * Eventually, we'll want to revalidate snapped metadata
1063 * too... probably...
1071 * read() on a dir. This weird interface hack only works if mounted
1072 * with '-o dirstat'.
1074 static ssize_t ceph_read_dir(struct file *file, char __user *buf, size_t size,
1077 struct ceph_file_info *cf = file->private_data;
1078 struct inode *inode = file->f_dentry->d_inode;
1079 struct ceph_inode_info *ci = ceph_inode(inode);
1082 if (!ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb), DIRSTAT))
1085 if (!cf->dir_info) {
1086 cf->dir_info = kmalloc(1024, GFP_NOFS);
1090 sprintf(cf->dir_info,
1093 " subdirs: %20lld\n"
1094 "rentries: %20lld\n"
1096 " rsubdirs: %20lld\n"
1098 "rctime: %10ld.%09ld\n",
1099 ci->i_files + ci->i_subdirs,
1102 ci->i_rfiles + ci->i_rsubdirs,
1106 (long)ci->i_rctime.tv_sec,
1107 (long)ci->i_rctime.tv_nsec);
1110 if (*ppos >= cf->dir_info_len)
1112 size = min_t(unsigned, size, cf->dir_info_len-*ppos);
1113 left = copy_to_user(buf, cf->dir_info + *ppos, size);
1116 *ppos += (size - left);
1121 * an fsync() on a dir will wait for any uncommitted directory
1122 * operations to commit.
1124 static int ceph_dir_fsync(struct file *file, int datasync)
1126 struct inode *inode = file->f_path.dentry->d_inode;
1127 struct ceph_inode_info *ci = ceph_inode(inode);
1128 struct list_head *head = &ci->i_unsafe_dirops;
1129 struct ceph_mds_request *req;
1133 dout("dir_fsync %p\n", inode);
1134 spin_lock(&ci->i_unsafe_lock);
1135 if (list_empty(head))
1138 req = list_entry(head->prev,
1139 struct ceph_mds_request, r_unsafe_dir_item);
1140 last_tid = req->r_tid;
1143 ceph_mdsc_get_request(req);
1144 spin_unlock(&ci->i_unsafe_lock);
1145 dout("dir_fsync %p wait on tid %llu (until %llu)\n",
1146 inode, req->r_tid, last_tid);
1147 if (req->r_timeout) {
1148 ret = wait_for_completion_timeout(
1149 &req->r_safe_completion, req->r_timeout);
1153 ret = -EIO; /* timed out */
1155 wait_for_completion(&req->r_safe_completion);
1157 spin_lock(&ci->i_unsafe_lock);
1158 ceph_mdsc_put_request(req);
1160 if (ret || list_empty(head))
1162 req = list_entry(head->next,
1163 struct ceph_mds_request, r_unsafe_dir_item);
1164 } while (req->r_tid < last_tid);
1166 spin_unlock(&ci->i_unsafe_lock);
1171 * We maintain a private dentry LRU.
1173 * FIXME: this needs to be changed to a per-mds lru to be useful.
1175 void ceph_dentry_lru_add(struct dentry *dn)
1177 struct ceph_dentry_info *di = ceph_dentry(dn);
1178 struct ceph_mds_client *mdsc;
1180 dout("dentry_lru_add %p %p '%.*s'\n", di, dn,
1181 dn->d_name.len, dn->d_name.name);
1183 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1184 spin_lock(&mdsc->dentry_lru_lock);
1185 list_add_tail(&di->lru, &mdsc->dentry_lru);
1187 spin_unlock(&mdsc->dentry_lru_lock);
1191 void ceph_dentry_lru_touch(struct dentry *dn)
1193 struct ceph_dentry_info *di = ceph_dentry(dn);
1194 struct ceph_mds_client *mdsc;
1196 dout("dentry_lru_touch %p %p '%.*s' (offset %lld)\n", di, dn,
1197 dn->d_name.len, dn->d_name.name, di->offset);
1199 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1200 spin_lock(&mdsc->dentry_lru_lock);
1201 list_move_tail(&di->lru, &mdsc->dentry_lru);
1202 spin_unlock(&mdsc->dentry_lru_lock);
1206 void ceph_dentry_lru_del(struct dentry *dn)
1208 struct ceph_dentry_info *di = ceph_dentry(dn);
1209 struct ceph_mds_client *mdsc;
1211 dout("dentry_lru_del %p %p '%.*s'\n", di, dn,
1212 dn->d_name.len, dn->d_name.name);
1214 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1215 spin_lock(&mdsc->dentry_lru_lock);
1216 list_del_init(&di->lru);
1218 spin_unlock(&mdsc->dentry_lru_lock);
1223 * Return name hash for a given dentry. This is dependent on
1224 * the parent directory's hash function.
1226 unsigned ceph_dentry_hash(struct dentry *dn)
1228 struct inode *dir = dn->d_parent->d_inode;
1229 struct ceph_inode_info *dci = ceph_inode(dir);
1231 switch (dci->i_dir_layout.dl_dir_hash) {
1232 case 0: /* for backward compat */
1233 case CEPH_STR_HASH_LINUX:
1234 return dn->d_name.hash;
1237 return ceph_str_hash(dci->i_dir_layout.dl_dir_hash,
1238 dn->d_name.name, dn->d_name.len);
1242 const struct file_operations ceph_dir_fops = {
1243 .read = ceph_read_dir,
1244 .readdir = ceph_readdir,
1245 .llseek = ceph_dir_llseek,
1247 .release = ceph_release,
1248 .unlocked_ioctl = ceph_ioctl,
1249 .fsync = ceph_dir_fsync,
1252 const struct inode_operations ceph_dir_iops = {
1253 .lookup = ceph_lookup,
1254 .permission = ceph_permission,
1255 .getattr = ceph_getattr,
1256 .setattr = ceph_setattr,
1257 .setxattr = ceph_setxattr,
1258 .getxattr = ceph_getxattr,
1259 .listxattr = ceph_listxattr,
1260 .removexattr = ceph_removexattr,
1261 .mknod = ceph_mknod,
1262 .symlink = ceph_symlink,
1263 .mkdir = ceph_mkdir,
1265 .unlink = ceph_unlink,
1266 .rmdir = ceph_unlink,
1267 .rename = ceph_rename,
1268 .create = ceph_create,
1271 const struct dentry_operations ceph_dentry_ops = {
1272 .d_revalidate = ceph_d_revalidate,
1273 .d_release = ceph_dentry_release,
1276 const struct dentry_operations ceph_snapdir_dentry_ops = {
1277 .d_revalidate = ceph_snapdir_d_revalidate,
1278 .d_release = ceph_dentry_release,
1281 const struct dentry_operations ceph_snap_dentry_ops = {
1282 .d_release = ceph_dentry_release,