cgroup: clean up cgroup_serial_nr_cursor
[linux-2.6/btrfs-unstable.git] / fs / ceph / dir.c
blobf02d82b7933e4399abd235a2527403df3364a2fa
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>
9 #include "super.h"
10 #include "mds_client.h"
13 * Directory operations: readdir, lookup, create, link, unlink,
14 * rename, etc.
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
26 * point by name.
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;
40 if (dentry->d_fsdata)
41 return 0;
43 di = kmem_cache_alloc(ceph_dentry_cachep, GFP_NOFS | __GFP_ZERO);
44 if (!di)
45 return -ENOMEM; /* oh well */
47 spin_lock(&dentry->d_lock);
48 if (dentry->d_fsdata) {
49 /* lost a race */
50 kmem_cache_free(ceph_dentry_cachep, di);
51 goto out_unlock;
54 if (ceph_snap(dentry->d_parent->d_inode) == CEPH_NOSNAP)
55 d_set_d_op(dentry, &ceph_dentry_ops);
56 else if (ceph_snap(dentry->d_parent->d_inode) == CEPH_SNAPDIR)
57 d_set_d_op(dentry, &ceph_snapdir_dentry_ops);
58 else
59 d_set_d_op(dentry, &ceph_snap_dentry_ops);
61 di->dentry = dentry;
62 di->lease_session = NULL;
63 dentry->d_time = jiffies;
64 /* avoid reordering d_fsdata setup so that the check above is safe */
65 smp_mb();
66 dentry->d_fsdata = di;
67 ceph_dentry_lru_add(dentry);
68 out_unlock:
69 spin_unlock(&dentry->d_lock);
70 return 0;
73 struct inode *ceph_get_dentry_parent_inode(struct dentry *dentry)
75 struct inode *inode = NULL;
77 if (!dentry)
78 return NULL;
80 spin_lock(&dentry->d_lock);
81 if (!IS_ROOT(dentry)) {
82 inode = dentry->d_parent->d_inode;
83 ihold(inode);
85 spin_unlock(&dentry->d_lock);
86 return inode;
91 * for readdir, we encode the directory frag and offset within that
92 * frag into f_pos.
94 static unsigned fpos_frag(loff_t p)
96 return p >> 32;
98 static unsigned fpos_off(loff_t p)
100 return p & 0xffffffff;
104 * When possible, we try to satisfy a readdir by peeking at the
105 * dcache. We make this work by carefully ordering dentries on
106 * d_u.d_child when we initially get results back from the MDS, and
107 * falling back to a "normal" sync readdir if any dentries in the dir
108 * are dropped.
110 * Complete dir indicates that we have all dentries in the dir. It is
111 * defined IFF we hold CEPH_CAP_FILE_SHARED (which will be revoked by
112 * the MDS if/when the directory is modified).
114 static int __dcache_readdir(struct file *filp,
115 void *dirent, filldir_t filldir)
117 struct ceph_file_info *fi = filp->private_data;
118 struct dentry *parent = filp->f_dentry;
119 struct inode *dir = parent->d_inode;
120 struct list_head *p;
121 struct dentry *dentry, *last;
122 struct ceph_dentry_info *di;
123 int err = 0;
125 /* claim ref on last dentry we returned */
126 last = fi->dentry;
127 fi->dentry = NULL;
129 dout("__dcache_readdir %p at %llu (last %p)\n", dir, filp->f_pos,
130 last);
132 spin_lock(&parent->d_lock);
134 /* start at beginning? */
135 if (filp->f_pos == 2 || last == NULL ||
136 filp->f_pos < ceph_dentry(last)->offset) {
137 if (list_empty(&parent->d_subdirs))
138 goto out_unlock;
139 p = parent->d_subdirs.prev;
140 dout(" initial p %p/%p\n", p->prev, p->next);
141 } else {
142 p = last->d_u.d_child.prev;
145 more:
146 dentry = list_entry(p, struct dentry, d_u.d_child);
147 di = ceph_dentry(dentry);
148 while (1) {
149 dout(" p %p/%p %s d_subdirs %p/%p\n", p->prev, p->next,
150 d_unhashed(dentry) ? "!hashed" : "hashed",
151 parent->d_subdirs.prev, parent->d_subdirs.next);
152 if (p == &parent->d_subdirs) {
153 fi->flags |= CEPH_F_ATEND;
154 goto out_unlock;
156 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
157 if (!d_unhashed(dentry) && dentry->d_inode &&
158 ceph_snap(dentry->d_inode) != CEPH_SNAPDIR &&
159 ceph_ino(dentry->d_inode) != CEPH_INO_CEPH &&
160 filp->f_pos <= di->offset)
161 break;
162 dout(" skipping %p %.*s at %llu (%llu)%s%s\n", dentry,
163 dentry->d_name.len, dentry->d_name.name, di->offset,
164 filp->f_pos, d_unhashed(dentry) ? " unhashed" : "",
165 !dentry->d_inode ? " null" : "");
166 spin_unlock(&dentry->d_lock);
167 p = p->prev;
168 dentry = list_entry(p, struct dentry, d_u.d_child);
169 di = ceph_dentry(dentry);
172 dget_dlock(dentry);
173 spin_unlock(&dentry->d_lock);
174 spin_unlock(&parent->d_lock);
176 dout(" %llu (%llu) dentry %p %.*s %p\n", di->offset, filp->f_pos,
177 dentry, dentry->d_name.len, dentry->d_name.name, dentry->d_inode);
178 filp->f_pos = di->offset;
179 err = filldir(dirent, dentry->d_name.name,
180 dentry->d_name.len, di->offset,
181 ceph_translate_ino(dentry->d_sb, dentry->d_inode->i_ino),
182 dentry->d_inode->i_mode >> 12);
184 if (last) {
185 if (err < 0) {
186 /* remember our position */
187 fi->dentry = last;
188 fi->next_offset = di->offset;
189 } else {
190 dput(last);
193 last = dentry;
195 if (err < 0)
196 goto out;
198 filp->f_pos++;
200 /* make sure a dentry wasn't dropped while we didn't have parent lock */
201 if (!ceph_dir_is_complete(dir)) {
202 dout(" lost dir complete on %p; falling back to mds\n", dir);
203 err = -EAGAIN;
204 goto out;
207 spin_lock(&parent->d_lock);
208 p = p->prev; /* advance to next dentry */
209 goto more;
211 out_unlock:
212 spin_unlock(&parent->d_lock);
213 out:
214 if (last)
215 dput(last);
216 return err;
220 * make note of the last dentry we read, so we can
221 * continue at the same lexicographical point,
222 * regardless of what dir changes take place on the
223 * server.
225 static int note_last_dentry(struct ceph_file_info *fi, const char *name,
226 int len)
228 kfree(fi->last_name);
229 fi->last_name = kmalloc(len+1, GFP_NOFS);
230 if (!fi->last_name)
231 return -ENOMEM;
232 memcpy(fi->last_name, name, len);
233 fi->last_name[len] = 0;
234 dout("note_last_dentry '%s'\n", fi->last_name);
235 return 0;
238 static int ceph_readdir(struct file *filp, void *dirent, filldir_t filldir)
240 struct ceph_file_info *fi = filp->private_data;
241 struct inode *inode = file_inode(filp);
242 struct ceph_inode_info *ci = ceph_inode(inode);
243 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
244 struct ceph_mds_client *mdsc = fsc->mdsc;
245 unsigned frag = fpos_frag(filp->f_pos);
246 int off = fpos_off(filp->f_pos);
247 int err;
248 u32 ftype;
249 struct ceph_mds_reply_info_parsed *rinfo;
250 const int max_entries = fsc->mount_options->max_readdir;
251 const int max_bytes = fsc->mount_options->max_readdir_bytes;
253 dout("readdir %p filp %p frag %u off %u\n", inode, filp, frag, off);
254 if (fi->flags & CEPH_F_ATEND)
255 return 0;
257 /* always start with . and .. */
258 if (filp->f_pos == 0) {
259 /* note dir version at start of readdir so we can tell
260 * if any dentries get dropped */
261 fi->dir_release_count = atomic_read(&ci->i_release_count);
263 dout("readdir off 0 -> '.'\n");
264 if (filldir(dirent, ".", 1, ceph_make_fpos(0, 0),
265 ceph_translate_ino(inode->i_sb, inode->i_ino),
266 inode->i_mode >> 12) < 0)
267 return 0;
268 filp->f_pos = 1;
269 off = 1;
271 if (filp->f_pos == 1) {
272 ino_t ino = parent_ino(filp->f_dentry);
273 dout("readdir off 1 -> '..'\n");
274 if (filldir(dirent, "..", 2, ceph_make_fpos(0, 1),
275 ceph_translate_ino(inode->i_sb, ino),
276 inode->i_mode >> 12) < 0)
277 return 0;
278 filp->f_pos = 2;
279 off = 2;
282 /* can we use the dcache? */
283 spin_lock(&ci->i_ceph_lock);
284 if ((filp->f_pos == 2 || fi->dentry) &&
285 !ceph_test_mount_opt(fsc, NOASYNCREADDIR) &&
286 ceph_snap(inode) != CEPH_SNAPDIR &&
287 __ceph_dir_is_complete(ci) &&
288 __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1)) {
289 spin_unlock(&ci->i_ceph_lock);
290 err = __dcache_readdir(filp, dirent, filldir);
291 if (err != -EAGAIN)
292 return err;
293 } else {
294 spin_unlock(&ci->i_ceph_lock);
296 if (fi->dentry) {
297 err = note_last_dentry(fi, fi->dentry->d_name.name,
298 fi->dentry->d_name.len);
299 if (err)
300 return err;
301 dput(fi->dentry);
302 fi->dentry = NULL;
305 /* proceed with a normal readdir */
307 more:
308 /* do we have the correct frag content buffered? */
309 if (fi->frag != frag || fi->last_readdir == NULL) {
310 struct ceph_mds_request *req;
311 int op = ceph_snap(inode) == CEPH_SNAPDIR ?
312 CEPH_MDS_OP_LSSNAP : CEPH_MDS_OP_READDIR;
314 /* discard old result, if any */
315 if (fi->last_readdir) {
316 ceph_mdsc_put_request(fi->last_readdir);
317 fi->last_readdir = NULL;
320 /* requery frag tree, as the frag topology may have changed */
321 frag = ceph_choose_frag(ceph_inode(inode), frag, NULL, NULL);
323 dout("readdir fetching %llx.%llx frag %x offset '%s'\n",
324 ceph_vinop(inode), frag, fi->last_name);
325 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
326 if (IS_ERR(req))
327 return PTR_ERR(req);
328 req->r_inode = inode;
329 ihold(inode);
330 req->r_dentry = dget(filp->f_dentry);
331 /* hints to request -> mds selection code */
332 req->r_direct_mode = USE_AUTH_MDS;
333 req->r_direct_hash = ceph_frag_value(frag);
334 req->r_direct_is_hash = true;
335 req->r_path2 = kstrdup(fi->last_name, GFP_NOFS);
336 req->r_readdir_offset = fi->next_offset;
337 req->r_args.readdir.frag = cpu_to_le32(frag);
338 req->r_args.readdir.max_entries = cpu_to_le32(max_entries);
339 req->r_args.readdir.max_bytes = cpu_to_le32(max_bytes);
340 req->r_num_caps = max_entries + 1;
341 err = ceph_mdsc_do_request(mdsc, NULL, req);
342 if (err < 0) {
343 ceph_mdsc_put_request(req);
344 return err;
346 dout("readdir got and parsed readdir result=%d"
347 " on frag %x, end=%d, complete=%d\n", err, frag,
348 (int)req->r_reply_info.dir_end,
349 (int)req->r_reply_info.dir_complete);
351 if (!req->r_did_prepopulate) {
352 dout("readdir !did_prepopulate");
353 /* preclude from marking dir complete */
354 fi->dir_release_count--;
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))
365 fi->next_offset = 2;
366 else
367 fi->next_offset = 0;
368 } else {
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]);
373 if (err)
374 return err;
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;
387 ino_t ino;
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);
393 BUG_ON(!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);
398 if (filldir(dirent,
399 rinfo->dir_dname[off - fi->offset],
400 rinfo->dir_dname_len[off - fi->offset],
401 pos,
402 ceph_translate_ino(inode->i_sb, ino), ftype) < 0) {
403 dout("filldir stopping us...\n");
404 return 0;
406 off++;
407 filp->f_pos = pos + 1;
410 if (fi->last_name) {
411 ceph_mdsc_put_request(fi->last_readdir);
412 fi->last_readdir = NULL;
413 goto more;
416 /* more frags? */
417 if (!ceph_frag_is_rightmost(frag)) {
418 frag = ceph_frag_next(frag);
419 off = 0;
420 filp->f_pos = ceph_make_fpos(frag, off);
421 dout("readdir next frag is %x\n", frag);
422 goto more;
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(&ci->i_ceph_lock);
432 if (atomic_read(&ci->i_release_count) == fi->dir_release_count) {
433 dout(" marking %p complete\n", inode);
434 __ceph_dir_set_complete(ci, fi->dir_release_count);
435 ci->i_max_offset = filp->f_pos;
437 spin_unlock(&ci->i_ceph_lock);
439 dout("readdir %p filp %p done.\n", inode, filp);
440 return 0;
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 .. */
452 if (fi->dentry) {
453 dput(fi->dentry);
454 fi->dentry = NULL;
456 fi->flags &= ~CEPH_F_ATEND;
459 static loff_t ceph_dir_llseek(struct file *file, loff_t offset, int whence)
461 struct ceph_file_info *fi = file->private_data;
462 struct inode *inode = file->f_mapping->host;
463 loff_t old_offset = offset;
464 loff_t retval;
466 mutex_lock(&inode->i_mutex);
467 retval = -EINVAL;
468 switch (whence) {
469 case SEEK_END:
470 offset += inode->i_size + 2; /* FIXME */
471 break;
472 case SEEK_CUR:
473 offset += file->f_pos;
474 case SEEK_SET:
475 break;
476 default:
477 goto out;
480 if (offset >= 0 && offset <= inode->i_sb->s_maxbytes) {
481 if (offset != file->f_pos) {
482 file->f_pos = offset;
483 file->f_version = 0;
484 fi->flags &= ~CEPH_F_ATEND;
486 retval = offset;
489 * discard buffered readdir content on seekdir(0), or
490 * seek to new frag, or seek prior to current chunk.
492 if (offset == 0 ||
493 fpos_frag(offset) != fpos_frag(old_offset) ||
494 fpos_off(offset) < fi->offset) {
495 dout("dir_llseek dropping %p content\n", file);
496 reset_readdir(fi);
499 /* bump dir_release_count if we did a forward seek */
500 if (offset > old_offset)
501 fi->dir_release_count--;
503 out:
504 mutex_unlock(&inode->i_mutex);
505 return retval;
509 * Handle lookups for the hidden .snap directory.
511 int ceph_handle_snapdir(struct ceph_mds_request *req,
512 struct dentry *dentry, int err)
514 struct ceph_fs_client *fsc = ceph_sb_to_client(dentry->d_sb);
515 struct inode *parent = dentry->d_parent->d_inode; /* we hold i_mutex */
517 /* .snap dir? */
518 if (err == -ENOENT &&
519 ceph_snap(parent) == CEPH_NOSNAP &&
520 strcmp(dentry->d_name.name,
521 fsc->mount_options->snapdir_name) == 0) {
522 struct inode *inode = ceph_get_snapdir(parent);
523 dout("ENOENT on snapdir %p '%.*s', linking to snapdir %p\n",
524 dentry, dentry->d_name.len, dentry->d_name.name, inode);
525 BUG_ON(!d_unhashed(dentry));
526 d_add(dentry, inode);
527 err = 0;
529 return err;
533 * Figure out final result of a lookup/open request.
535 * Mainly, make sure we return the final req->r_dentry (if it already
536 * existed) in place of the original VFS-provided dentry when they
537 * differ.
539 * Gracefully handle the case where the MDS replies with -ENOENT and
540 * no trace (which it may do, at its discretion, e.g., if it doesn't
541 * care to issue a lease on the negative dentry).
543 struct dentry *ceph_finish_lookup(struct ceph_mds_request *req,
544 struct dentry *dentry, int err)
546 if (err == -ENOENT) {
547 /* no trace? */
548 err = 0;
549 if (!req->r_reply_info.head->is_dentry) {
550 dout("ENOENT and no trace, dentry %p inode %p\n",
551 dentry, dentry->d_inode);
552 if (dentry->d_inode) {
553 d_drop(dentry);
554 err = -ENOENT;
555 } else {
556 d_add(dentry, NULL);
560 if (err)
561 dentry = ERR_PTR(err);
562 else if (dentry != req->r_dentry)
563 dentry = dget(req->r_dentry); /* we got spliced */
564 else
565 dentry = NULL;
566 return dentry;
569 static int is_root_ceph_dentry(struct inode *inode, struct dentry *dentry)
571 return ceph_ino(inode) == CEPH_INO_ROOT &&
572 strncmp(dentry->d_name.name, ".ceph", 5) == 0;
576 * Look up a single dir entry. If there is a lookup intent, inform
577 * the MDS so that it gets our 'caps wanted' value in a single op.
579 static struct dentry *ceph_lookup(struct inode *dir, struct dentry *dentry,
580 unsigned int flags)
582 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
583 struct ceph_mds_client *mdsc = fsc->mdsc;
584 struct ceph_mds_request *req;
585 int op;
586 int err;
588 dout("lookup %p dentry %p '%.*s'\n",
589 dir, dentry, dentry->d_name.len, dentry->d_name.name);
591 if (dentry->d_name.len > NAME_MAX)
592 return ERR_PTR(-ENAMETOOLONG);
594 err = ceph_init_dentry(dentry);
595 if (err < 0)
596 return ERR_PTR(err);
598 /* can we conclude ENOENT locally? */
599 if (dentry->d_inode == NULL) {
600 struct ceph_inode_info *ci = ceph_inode(dir);
601 struct ceph_dentry_info *di = ceph_dentry(dentry);
603 spin_lock(&ci->i_ceph_lock);
604 dout(" dir %p flags are %d\n", dir, ci->i_ceph_flags);
605 if (strncmp(dentry->d_name.name,
606 fsc->mount_options->snapdir_name,
607 dentry->d_name.len) &&
608 !is_root_ceph_dentry(dir, dentry) &&
609 __ceph_dir_is_complete(ci) &&
610 (__ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1))) {
611 spin_unlock(&ci->i_ceph_lock);
612 dout(" dir %p complete, -ENOENT\n", dir);
613 d_add(dentry, NULL);
614 di->lease_shared_gen = ci->i_shared_gen;
615 return NULL;
617 spin_unlock(&ci->i_ceph_lock);
620 op = ceph_snap(dir) == CEPH_SNAPDIR ?
621 CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP;
622 req = ceph_mdsc_create_request(mdsc, op, USE_ANY_MDS);
623 if (IS_ERR(req))
624 return ERR_CAST(req);
625 req->r_dentry = dget(dentry);
626 req->r_num_caps = 2;
627 /* we only need inode linkage */
628 req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
629 req->r_locked_dir = dir;
630 err = ceph_mdsc_do_request(mdsc, NULL, req);
631 err = ceph_handle_snapdir(req, dentry, err);
632 dentry = ceph_finish_lookup(req, dentry, err);
633 ceph_mdsc_put_request(req); /* will dput(dentry) */
634 dout("lookup result=%p\n", dentry);
635 return dentry;
639 * If we do a create but get no trace back from the MDS, follow up with
640 * a lookup (the VFS expects us to link up the provided dentry).
642 int ceph_handle_notrace_create(struct inode *dir, struct dentry *dentry)
644 struct dentry *result = ceph_lookup(dir, dentry, 0);
646 if (result && !IS_ERR(result)) {
648 * We created the item, then did a lookup, and found
649 * it was already linked to another inode we already
650 * had in our cache (and thus got spliced). Link our
651 * dentry to that inode, but don't hash it, just in
652 * case the VFS wants to dereference it.
654 BUG_ON(!result->d_inode);
655 d_instantiate(dentry, result->d_inode);
656 return 0;
658 return PTR_ERR(result);
661 static int ceph_mknod(struct inode *dir, struct dentry *dentry,
662 umode_t mode, dev_t rdev)
664 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
665 struct ceph_mds_client *mdsc = fsc->mdsc;
666 struct ceph_mds_request *req;
667 int err;
669 if (ceph_snap(dir) != CEPH_NOSNAP)
670 return -EROFS;
672 dout("mknod in dir %p dentry %p mode 0%ho rdev %d\n",
673 dir, dentry, mode, rdev);
674 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_MKNOD, USE_AUTH_MDS);
675 if (IS_ERR(req)) {
676 d_drop(dentry);
677 return PTR_ERR(req);
679 req->r_dentry = dget(dentry);
680 req->r_num_caps = 2;
681 req->r_locked_dir = dir;
682 req->r_args.mknod.mode = cpu_to_le32(mode);
683 req->r_args.mknod.rdev = cpu_to_le32(rdev);
684 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
685 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
686 err = ceph_mdsc_do_request(mdsc, dir, req);
687 if (!err && !req->r_reply_info.head->is_dentry)
688 err = ceph_handle_notrace_create(dir, dentry);
689 ceph_mdsc_put_request(req);
690 if (err)
691 d_drop(dentry);
692 return err;
695 static int ceph_create(struct inode *dir, struct dentry *dentry, umode_t mode,
696 bool excl)
698 return ceph_mknod(dir, dentry, mode, 0);
701 static int ceph_symlink(struct inode *dir, struct dentry *dentry,
702 const char *dest)
704 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
705 struct ceph_mds_client *mdsc = fsc->mdsc;
706 struct ceph_mds_request *req;
707 int err;
709 if (ceph_snap(dir) != CEPH_NOSNAP)
710 return -EROFS;
712 dout("symlink in dir %p dentry %p to '%s'\n", dir, dentry, dest);
713 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SYMLINK, USE_AUTH_MDS);
714 if (IS_ERR(req)) {
715 d_drop(dentry);
716 return PTR_ERR(req);
718 req->r_dentry = dget(dentry);
719 req->r_num_caps = 2;
720 req->r_path2 = kstrdup(dest, GFP_NOFS);
721 req->r_locked_dir = dir;
722 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
723 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
724 err = ceph_mdsc_do_request(mdsc, dir, req);
725 if (!err && !req->r_reply_info.head->is_dentry)
726 err = ceph_handle_notrace_create(dir, dentry);
727 ceph_mdsc_put_request(req);
728 if (err)
729 d_drop(dentry);
730 return err;
733 static int ceph_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
735 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
736 struct ceph_mds_client *mdsc = fsc->mdsc;
737 struct ceph_mds_request *req;
738 int err = -EROFS;
739 int op;
741 if (ceph_snap(dir) == CEPH_SNAPDIR) {
742 /* mkdir .snap/foo is a MKSNAP */
743 op = CEPH_MDS_OP_MKSNAP;
744 dout("mksnap dir %p snap '%.*s' dn %p\n", dir,
745 dentry->d_name.len, dentry->d_name.name, dentry);
746 } else if (ceph_snap(dir) == CEPH_NOSNAP) {
747 dout("mkdir dir %p dn %p mode 0%ho\n", dir, dentry, mode);
748 op = CEPH_MDS_OP_MKDIR;
749 } else {
750 goto out;
752 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
753 if (IS_ERR(req)) {
754 err = PTR_ERR(req);
755 goto out;
758 req->r_dentry = dget(dentry);
759 req->r_num_caps = 2;
760 req->r_locked_dir = dir;
761 req->r_args.mkdir.mode = cpu_to_le32(mode);
762 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
763 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
764 err = ceph_mdsc_do_request(mdsc, dir, req);
765 if (!err && !req->r_reply_info.head->is_dentry)
766 err = ceph_handle_notrace_create(dir, dentry);
767 ceph_mdsc_put_request(req);
768 out:
769 if (err < 0)
770 d_drop(dentry);
771 return err;
774 static int ceph_link(struct dentry *old_dentry, struct inode *dir,
775 struct dentry *dentry)
777 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
778 struct ceph_mds_client *mdsc = fsc->mdsc;
779 struct ceph_mds_request *req;
780 int err;
782 if (ceph_snap(dir) != CEPH_NOSNAP)
783 return -EROFS;
785 dout("link in dir %p old_dentry %p dentry %p\n", dir,
786 old_dentry, dentry);
787 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LINK, USE_AUTH_MDS);
788 if (IS_ERR(req)) {
789 d_drop(dentry);
790 return PTR_ERR(req);
792 req->r_dentry = dget(dentry);
793 req->r_num_caps = 2;
794 req->r_old_dentry = dget(old_dentry); /* or inode? hrm. */
795 req->r_old_dentry_dir = ceph_get_dentry_parent_inode(old_dentry);
796 req->r_locked_dir = dir;
797 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
798 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
799 err = ceph_mdsc_do_request(mdsc, dir, req);
800 if (err) {
801 d_drop(dentry);
802 } else if (!req->r_reply_info.head->is_dentry) {
803 ihold(old_dentry->d_inode);
804 d_instantiate(dentry, old_dentry->d_inode);
806 ceph_mdsc_put_request(req);
807 return err;
811 * For a soon-to-be unlinked file, drop the AUTH_RDCACHE caps. If it
812 * looks like the link count will hit 0, drop any other caps (other
813 * than PIN) we don't specifically want (due to the file still being
814 * open).
816 static int drop_caps_for_unlink(struct inode *inode)
818 struct ceph_inode_info *ci = ceph_inode(inode);
819 int drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL;
821 spin_lock(&ci->i_ceph_lock);
822 if (inode->i_nlink == 1) {
823 drop |= ~(__ceph_caps_wanted(ci) | CEPH_CAP_PIN);
824 ci->i_ceph_flags |= CEPH_I_NODELAY;
826 spin_unlock(&ci->i_ceph_lock);
827 return drop;
831 * rmdir and unlink are differ only by the metadata op code
833 static int ceph_unlink(struct inode *dir, struct dentry *dentry)
835 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
836 struct ceph_mds_client *mdsc = fsc->mdsc;
837 struct inode *inode = dentry->d_inode;
838 struct ceph_mds_request *req;
839 int err = -EROFS;
840 int op;
842 if (ceph_snap(dir) == CEPH_SNAPDIR) {
843 /* rmdir .snap/foo is RMSNAP */
844 dout("rmsnap dir %p '%.*s' dn %p\n", dir, dentry->d_name.len,
845 dentry->d_name.name, dentry);
846 op = CEPH_MDS_OP_RMSNAP;
847 } else if (ceph_snap(dir) == CEPH_NOSNAP) {
848 dout("unlink/rmdir dir %p dn %p inode %p\n",
849 dir, dentry, inode);
850 op = S_ISDIR(dentry->d_inode->i_mode) ?
851 CEPH_MDS_OP_RMDIR : CEPH_MDS_OP_UNLINK;
852 } else
853 goto out;
854 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
855 if (IS_ERR(req)) {
856 err = PTR_ERR(req);
857 goto out;
859 req->r_dentry = dget(dentry);
860 req->r_num_caps = 2;
861 req->r_locked_dir = dir;
862 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
863 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
864 req->r_inode_drop = drop_caps_for_unlink(inode);
865 err = ceph_mdsc_do_request(mdsc, dir, req);
866 if (!err && !req->r_reply_info.head->is_dentry)
867 d_delete(dentry);
868 ceph_mdsc_put_request(req);
869 out:
870 return err;
873 static int ceph_rename(struct inode *old_dir, struct dentry *old_dentry,
874 struct inode *new_dir, struct dentry *new_dentry)
876 struct ceph_fs_client *fsc = ceph_sb_to_client(old_dir->i_sb);
877 struct ceph_mds_client *mdsc = fsc->mdsc;
878 struct ceph_mds_request *req;
879 int err;
881 if (ceph_snap(old_dir) != ceph_snap(new_dir))
882 return -EXDEV;
883 if (ceph_snap(old_dir) != CEPH_NOSNAP ||
884 ceph_snap(new_dir) != CEPH_NOSNAP)
885 return -EROFS;
886 dout("rename dir %p dentry %p to dir %p dentry %p\n",
887 old_dir, old_dentry, new_dir, new_dentry);
888 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_RENAME, USE_AUTH_MDS);
889 if (IS_ERR(req))
890 return PTR_ERR(req);
891 req->r_dentry = dget(new_dentry);
892 req->r_num_caps = 2;
893 req->r_old_dentry = dget(old_dentry);
894 req->r_old_dentry_dir = ceph_get_dentry_parent_inode(old_dentry);
895 req->r_locked_dir = new_dir;
896 req->r_old_dentry_drop = CEPH_CAP_FILE_SHARED;
897 req->r_old_dentry_unless = CEPH_CAP_FILE_EXCL;
898 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
899 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
900 /* release LINK_RDCACHE on source inode (mds will lock it) */
901 req->r_old_inode_drop = CEPH_CAP_LINK_SHARED;
902 if (new_dentry->d_inode)
903 req->r_inode_drop = drop_caps_for_unlink(new_dentry->d_inode);
904 err = ceph_mdsc_do_request(mdsc, old_dir, req);
905 if (!err && !req->r_reply_info.head->is_dentry) {
907 * Normally d_move() is done by fill_trace (called by
908 * do_request, above). If there is no trace, we need
909 * to do it here.
912 /* d_move screws up d_subdirs order */
913 ceph_dir_clear_complete(new_dir);
915 d_move(old_dentry, new_dentry);
917 /* ensure target dentry is invalidated, despite
918 rehashing bug in vfs_rename_dir */
919 ceph_invalidate_dentry_lease(new_dentry);
921 ceph_mdsc_put_request(req);
922 return err;
926 * Ensure a dentry lease will no longer revalidate.
928 void ceph_invalidate_dentry_lease(struct dentry *dentry)
930 spin_lock(&dentry->d_lock);
931 dentry->d_time = jiffies;
932 ceph_dentry(dentry)->lease_shared_gen = 0;
933 spin_unlock(&dentry->d_lock);
937 * Check if dentry lease is valid. If not, delete the lease. Try to
938 * renew if the least is more than half up.
940 static int dentry_lease_is_valid(struct dentry *dentry)
942 struct ceph_dentry_info *di;
943 struct ceph_mds_session *s;
944 int valid = 0;
945 u32 gen;
946 unsigned long ttl;
947 struct ceph_mds_session *session = NULL;
948 struct inode *dir = NULL;
949 u32 seq = 0;
951 spin_lock(&dentry->d_lock);
952 di = ceph_dentry(dentry);
953 if (di->lease_session) {
954 s = di->lease_session;
955 spin_lock(&s->s_gen_ttl_lock);
956 gen = s->s_cap_gen;
957 ttl = s->s_cap_ttl;
958 spin_unlock(&s->s_gen_ttl_lock);
960 if (di->lease_gen == gen &&
961 time_before(jiffies, dentry->d_time) &&
962 time_before(jiffies, ttl)) {
963 valid = 1;
964 if (di->lease_renew_after &&
965 time_after(jiffies, di->lease_renew_after)) {
966 /* we should renew */
967 dir = dentry->d_parent->d_inode;
968 session = ceph_get_mds_session(s);
969 seq = di->lease_seq;
970 di->lease_renew_after = 0;
971 di->lease_renew_from = jiffies;
975 spin_unlock(&dentry->d_lock);
977 if (session) {
978 ceph_mdsc_lease_send_msg(session, dir, dentry,
979 CEPH_MDS_LEASE_RENEW, seq);
980 ceph_put_mds_session(session);
982 dout("dentry_lease_is_valid - dentry %p = %d\n", dentry, valid);
983 return valid;
987 * Check if directory-wide content lease/cap is valid.
989 static int dir_lease_is_valid(struct inode *dir, struct dentry *dentry)
991 struct ceph_inode_info *ci = ceph_inode(dir);
992 struct ceph_dentry_info *di = ceph_dentry(dentry);
993 int valid = 0;
995 spin_lock(&ci->i_ceph_lock);
996 if (ci->i_shared_gen == di->lease_shared_gen)
997 valid = __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1);
998 spin_unlock(&ci->i_ceph_lock);
999 dout("dir_lease_is_valid dir %p v%u dentry %p v%u = %d\n",
1000 dir, (unsigned)ci->i_shared_gen, dentry,
1001 (unsigned)di->lease_shared_gen, valid);
1002 return valid;
1006 * Check if cached dentry can be trusted.
1008 static int ceph_d_revalidate(struct dentry *dentry, unsigned int flags)
1010 int valid = 0;
1011 struct inode *dir;
1013 if (flags & LOOKUP_RCU)
1014 return -ECHILD;
1016 dout("d_revalidate %p '%.*s' inode %p offset %lld\n", dentry,
1017 dentry->d_name.len, dentry->d_name.name, dentry->d_inode,
1018 ceph_dentry(dentry)->offset);
1020 dir = ceph_get_dentry_parent_inode(dentry);
1022 /* always trust cached snapped dentries, snapdir dentry */
1023 if (ceph_snap(dir) != CEPH_NOSNAP) {
1024 dout("d_revalidate %p '%.*s' inode %p is SNAPPED\n", dentry,
1025 dentry->d_name.len, dentry->d_name.name, dentry->d_inode);
1026 valid = 1;
1027 } else if (dentry->d_inode &&
1028 ceph_snap(dentry->d_inode) == CEPH_SNAPDIR) {
1029 valid = 1;
1030 } else if (dentry_lease_is_valid(dentry) ||
1031 dir_lease_is_valid(dir, dentry)) {
1032 valid = 1;
1035 dout("d_revalidate %p %s\n", dentry, valid ? "valid" : "invalid");
1036 if (valid)
1037 ceph_dentry_lru_touch(dentry);
1038 else
1039 d_drop(dentry);
1040 iput(dir);
1041 return valid;
1045 * Release our ceph_dentry_info.
1047 static void ceph_d_release(struct dentry *dentry)
1049 struct ceph_dentry_info *di = ceph_dentry(dentry);
1051 dout("d_release %p\n", dentry);
1052 ceph_dentry_lru_del(dentry);
1053 if (di->lease_session)
1054 ceph_put_mds_session(di->lease_session);
1055 kmem_cache_free(ceph_dentry_cachep, di);
1056 dentry->d_fsdata = NULL;
1059 static int ceph_snapdir_d_revalidate(struct dentry *dentry,
1060 unsigned int flags)
1063 * Eventually, we'll want to revalidate snapped metadata
1064 * too... probably...
1066 return 1;
1070 * When the VFS prunes a dentry from the cache, we need to clear the
1071 * complete flag on the parent directory.
1073 * Called under dentry->d_lock.
1075 static void ceph_d_prune(struct dentry *dentry)
1077 dout("ceph_d_prune %p\n", dentry);
1079 /* do we have a valid parent? */
1080 if (IS_ROOT(dentry))
1081 return;
1083 /* if we are not hashed, we don't affect dir's completeness */
1084 if (d_unhashed(dentry))
1085 return;
1088 * we hold d_lock, so d_parent is stable, and d_fsdata is never
1089 * cleared until d_release
1091 ceph_dir_clear_complete(dentry->d_parent->d_inode);
1095 * read() on a dir. This weird interface hack only works if mounted
1096 * with '-o dirstat'.
1098 static ssize_t ceph_read_dir(struct file *file, char __user *buf, size_t size,
1099 loff_t *ppos)
1101 struct ceph_file_info *cf = file->private_data;
1102 struct inode *inode = file_inode(file);
1103 struct ceph_inode_info *ci = ceph_inode(inode);
1104 int left;
1105 const int bufsize = 1024;
1107 if (!ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb), DIRSTAT))
1108 return -EISDIR;
1110 if (!cf->dir_info) {
1111 cf->dir_info = kmalloc(bufsize, GFP_NOFS);
1112 if (!cf->dir_info)
1113 return -ENOMEM;
1114 cf->dir_info_len =
1115 snprintf(cf->dir_info, bufsize,
1116 "entries: %20lld\n"
1117 " files: %20lld\n"
1118 " subdirs: %20lld\n"
1119 "rentries: %20lld\n"
1120 " rfiles: %20lld\n"
1121 " rsubdirs: %20lld\n"
1122 "rbytes: %20lld\n"
1123 "rctime: %10ld.%09ld\n",
1124 ci->i_files + ci->i_subdirs,
1125 ci->i_files,
1126 ci->i_subdirs,
1127 ci->i_rfiles + ci->i_rsubdirs,
1128 ci->i_rfiles,
1129 ci->i_rsubdirs,
1130 ci->i_rbytes,
1131 (long)ci->i_rctime.tv_sec,
1132 (long)ci->i_rctime.tv_nsec);
1135 if (*ppos >= cf->dir_info_len)
1136 return 0;
1137 size = min_t(unsigned, size, cf->dir_info_len-*ppos);
1138 left = copy_to_user(buf, cf->dir_info + *ppos, size);
1139 if (left == size)
1140 return -EFAULT;
1141 *ppos += (size - left);
1142 return size - left;
1146 * an fsync() on a dir will wait for any uncommitted directory
1147 * operations to commit.
1149 static int ceph_dir_fsync(struct file *file, loff_t start, loff_t end,
1150 int datasync)
1152 struct inode *inode = file_inode(file);
1153 struct ceph_inode_info *ci = ceph_inode(inode);
1154 struct list_head *head = &ci->i_unsafe_dirops;
1155 struct ceph_mds_request *req;
1156 u64 last_tid;
1157 int ret = 0;
1159 dout("dir_fsync %p\n", inode);
1160 ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
1161 if (ret)
1162 return ret;
1163 mutex_lock(&inode->i_mutex);
1165 spin_lock(&ci->i_unsafe_lock);
1166 if (list_empty(head))
1167 goto out;
1169 req = list_entry(head->prev,
1170 struct ceph_mds_request, r_unsafe_dir_item);
1171 last_tid = req->r_tid;
1173 do {
1174 ceph_mdsc_get_request(req);
1175 spin_unlock(&ci->i_unsafe_lock);
1177 dout("dir_fsync %p wait on tid %llu (until %llu)\n",
1178 inode, req->r_tid, last_tid);
1179 if (req->r_timeout) {
1180 ret = wait_for_completion_timeout(
1181 &req->r_safe_completion, req->r_timeout);
1182 if (ret > 0)
1183 ret = 0;
1184 else if (ret == 0)
1185 ret = -EIO; /* timed out */
1186 } else {
1187 wait_for_completion(&req->r_safe_completion);
1189 ceph_mdsc_put_request(req);
1191 spin_lock(&ci->i_unsafe_lock);
1192 if (ret || list_empty(head))
1193 break;
1194 req = list_entry(head->next,
1195 struct ceph_mds_request, r_unsafe_dir_item);
1196 } while (req->r_tid < last_tid);
1197 out:
1198 spin_unlock(&ci->i_unsafe_lock);
1199 mutex_unlock(&inode->i_mutex);
1201 return ret;
1205 * We maintain a private dentry LRU.
1207 * FIXME: this needs to be changed to a per-mds lru to be useful.
1209 void ceph_dentry_lru_add(struct dentry *dn)
1211 struct ceph_dentry_info *di = ceph_dentry(dn);
1212 struct ceph_mds_client *mdsc;
1214 dout("dentry_lru_add %p %p '%.*s'\n", di, dn,
1215 dn->d_name.len, dn->d_name.name);
1216 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1217 spin_lock(&mdsc->dentry_lru_lock);
1218 list_add_tail(&di->lru, &mdsc->dentry_lru);
1219 mdsc->num_dentry++;
1220 spin_unlock(&mdsc->dentry_lru_lock);
1223 void ceph_dentry_lru_touch(struct dentry *dn)
1225 struct ceph_dentry_info *di = ceph_dentry(dn);
1226 struct ceph_mds_client *mdsc;
1228 dout("dentry_lru_touch %p %p '%.*s' (offset %lld)\n", di, dn,
1229 dn->d_name.len, dn->d_name.name, di->offset);
1230 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1231 spin_lock(&mdsc->dentry_lru_lock);
1232 list_move_tail(&di->lru, &mdsc->dentry_lru);
1233 spin_unlock(&mdsc->dentry_lru_lock);
1236 void ceph_dentry_lru_del(struct dentry *dn)
1238 struct ceph_dentry_info *di = ceph_dentry(dn);
1239 struct ceph_mds_client *mdsc;
1241 dout("dentry_lru_del %p %p '%.*s'\n", di, dn,
1242 dn->d_name.len, dn->d_name.name);
1243 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1244 spin_lock(&mdsc->dentry_lru_lock);
1245 list_del_init(&di->lru);
1246 mdsc->num_dentry--;
1247 spin_unlock(&mdsc->dentry_lru_lock);
1251 * Return name hash for a given dentry. This is dependent on
1252 * the parent directory's hash function.
1254 unsigned ceph_dentry_hash(struct inode *dir, struct dentry *dn)
1256 struct ceph_inode_info *dci = ceph_inode(dir);
1258 switch (dci->i_dir_layout.dl_dir_hash) {
1259 case 0: /* for backward compat */
1260 case CEPH_STR_HASH_LINUX:
1261 return dn->d_name.hash;
1263 default:
1264 return ceph_str_hash(dci->i_dir_layout.dl_dir_hash,
1265 dn->d_name.name, dn->d_name.len);
1269 const struct file_operations ceph_dir_fops = {
1270 .read = ceph_read_dir,
1271 .readdir = ceph_readdir,
1272 .llseek = ceph_dir_llseek,
1273 .open = ceph_open,
1274 .release = ceph_release,
1275 .unlocked_ioctl = ceph_ioctl,
1276 .fsync = ceph_dir_fsync,
1279 const struct inode_operations ceph_dir_iops = {
1280 .lookup = ceph_lookup,
1281 .permission = ceph_permission,
1282 .getattr = ceph_getattr,
1283 .setattr = ceph_setattr,
1284 .setxattr = ceph_setxattr,
1285 .getxattr = ceph_getxattr,
1286 .listxattr = ceph_listxattr,
1287 .removexattr = ceph_removexattr,
1288 .mknod = ceph_mknod,
1289 .symlink = ceph_symlink,
1290 .mkdir = ceph_mkdir,
1291 .link = ceph_link,
1292 .unlink = ceph_unlink,
1293 .rmdir = ceph_unlink,
1294 .rename = ceph_rename,
1295 .create = ceph_create,
1296 .atomic_open = ceph_atomic_open,
1299 const struct dentry_operations ceph_dentry_ops = {
1300 .d_revalidate = ceph_d_revalidate,
1301 .d_release = ceph_d_release,
1302 .d_prune = ceph_d_prune,
1305 const struct dentry_operations ceph_snapdir_dentry_ops = {
1306 .d_revalidate = ceph_snapdir_d_revalidate,
1307 .d_release = ceph_d_release,
1310 const struct dentry_operations ceph_snap_dentry_ops = {
1311 .d_release = ceph_d_release,
1312 .d_prune = ceph_d_prune,