Merge branch 'urgent' of git://git.kernel.org/pub/scm/linux/kernel/git/brodo/pcmcia-2.6
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / fs / ceph / dir.c
blobea8ee2e526aae94d5fd10c16cf4e78ca16090818
1 #include "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"
12 * Directory operations: readdir, lookup, create, link, unlink,
13 * rename, etc.
17 * Ceph MDS operations are specified in terms of a base ino and
18 * relative path. Thus, the client can specify an operation on a
19 * specific inode (e.g., a getattr due to fstat(2)), or as a path
20 * relative to, say, the root directory.
22 * Normally, we limit ourselves to strict inode ops (no path component)
23 * or dentry operations (a single path component relative to an ino). The
24 * exception to this is open_root_dentry(), which will open the mount
25 * point by name.
28 const struct inode_operations ceph_dir_iops;
29 const struct file_operations ceph_dir_fops;
30 struct dentry_operations ceph_dentry_ops;
33 * Initialize ceph dentry state.
35 int ceph_init_dentry(struct dentry *dentry)
37 struct ceph_dentry_info *di;
39 if (dentry->d_fsdata)
40 return 0;
42 if (ceph_snap(dentry->d_parent->d_inode) == CEPH_NOSNAP)
43 dentry->d_op = &ceph_dentry_ops;
44 else if (ceph_snap(dentry->d_parent->d_inode) == CEPH_SNAPDIR)
45 dentry->d_op = &ceph_snapdir_dentry_ops;
46 else
47 dentry->d_op = &ceph_snap_dentry_ops;
49 di = kmem_cache_alloc(ceph_dentry_cachep, GFP_NOFS);
50 if (!di)
51 return -ENOMEM; /* oh well */
53 spin_lock(&dentry->d_lock);
54 if (dentry->d_fsdata) /* lost a race */
55 goto out_unlock;
56 di->dentry = dentry;
57 di->lease_session = NULL;
58 dentry->d_fsdata = di;
59 dentry->d_time = jiffies;
60 ceph_dentry_lru_add(dentry);
61 out_unlock:
62 spin_unlock(&dentry->d_lock);
63 return 0;
69 * for readdir, we encode the directory frag and offset within that
70 * frag into f_pos.
72 static unsigned fpos_frag(loff_t p)
74 return p >> 32;
76 static unsigned fpos_off(loff_t p)
78 return p & 0xffffffff;
82 * When possible, we try to satisfy a readdir by peeking at the
83 * dcache. We make this work by carefully ordering dentries on
84 * d_u.d_child when we initially get results back from the MDS, and
85 * falling back to a "normal" sync readdir if any dentries in the dir
86 * are dropped.
88 * I_COMPLETE tells indicates we have all dentries in the dir. It is
89 * defined IFF we hold CEPH_CAP_FILE_SHARED (which will be revoked by
90 * the MDS if/when the directory is modified).
92 static int __dcache_readdir(struct file *filp,
93 void *dirent, filldir_t filldir)
95 struct inode *inode = filp->f_dentry->d_inode;
96 struct ceph_file_info *fi = filp->private_data;
97 struct dentry *parent = filp->f_dentry;
98 struct inode *dir = parent->d_inode;
99 struct list_head *p;
100 struct dentry *dentry, *last;
101 struct ceph_dentry_info *di;
102 int err = 0;
104 /* claim ref on last dentry we returned */
105 last = fi->dentry;
106 fi->dentry = NULL;
108 dout("__dcache_readdir %p at %llu (last %p)\n", dir, filp->f_pos,
109 last);
111 spin_lock(&dcache_lock);
113 /* start at beginning? */
114 if (filp->f_pos == 2 || (last &&
115 filp->f_pos < ceph_dentry(last)->offset)) {
116 if (list_empty(&parent->d_subdirs))
117 goto out_unlock;
118 p = parent->d_subdirs.prev;
119 dout(" initial p %p/%p\n", p->prev, p->next);
120 } else {
121 p = last->d_u.d_child.prev;
124 more:
125 dentry = list_entry(p, struct dentry, d_u.d_child);
126 di = ceph_dentry(dentry);
127 while (1) {
128 dout(" p %p/%p d_subdirs %p/%p\n", p->prev, p->next,
129 parent->d_subdirs.prev, parent->d_subdirs.next);
130 if (p == &parent->d_subdirs) {
131 fi->at_end = 1;
132 goto out_unlock;
134 if (!d_unhashed(dentry) && dentry->d_inode &&
135 ceph_snap(dentry->d_inode) != CEPH_SNAPDIR &&
136 ceph_ino(dentry->d_inode) != CEPH_INO_CEPH &&
137 filp->f_pos <= di->offset)
138 break;
139 dout(" skipping %p %.*s at %llu (%llu)%s%s\n", dentry,
140 dentry->d_name.len, dentry->d_name.name, di->offset,
141 filp->f_pos, d_unhashed(dentry) ? " unhashed" : "",
142 !dentry->d_inode ? " null" : "");
143 p = p->prev;
144 dentry = list_entry(p, struct dentry, d_u.d_child);
145 di = ceph_dentry(dentry);
148 atomic_inc(&dentry->d_count);
149 spin_unlock(&dcache_lock);
150 spin_unlock(&inode->i_lock);
152 dout(" %llu (%llu) dentry %p %.*s %p\n", di->offset, filp->f_pos,
153 dentry, dentry->d_name.len, dentry->d_name.name, dentry->d_inode);
154 filp->f_pos = di->offset;
155 err = filldir(dirent, dentry->d_name.name,
156 dentry->d_name.len, di->offset,
157 dentry->d_inode->i_ino,
158 dentry->d_inode->i_mode >> 12);
160 if (last) {
161 if (err < 0) {
162 /* remember our position */
163 fi->dentry = last;
164 fi->next_offset = di->offset;
165 } else {
166 dput(last);
168 last = NULL;
171 spin_lock(&inode->i_lock);
172 spin_lock(&dcache_lock);
174 last = dentry;
176 if (err < 0)
177 goto out_unlock;
179 p = p->prev;
180 filp->f_pos++;
182 /* make sure a dentry wasn't dropped while we didn't have dcache_lock */
183 if ((ceph_inode(dir)->i_ceph_flags & CEPH_I_COMPLETE))
184 goto more;
185 dout(" lost I_COMPLETE on %p; falling back to mds\n", dir);
186 err = -EAGAIN;
188 out_unlock:
189 spin_unlock(&dcache_lock);
191 if (last) {
192 spin_unlock(&inode->i_lock);
193 dput(last);
194 spin_lock(&inode->i_lock);
197 return err;
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
204 * server.
206 static int note_last_dentry(struct ceph_file_info *fi, const char *name,
207 int len)
209 kfree(fi->last_name);
210 fi->last_name = kmalloc(len+1, GFP_NOFS);
211 if (!fi->last_name)
212 return -ENOMEM;
213 memcpy(fi->last_name, name, len);
214 fi->last_name[len] = 0;
215 dout("note_last_dentry '%s'\n", fi->last_name);
216 return 0;
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_client *client = ceph_inode_to_client(inode);
225 struct ceph_mds_client *mdsc = &client->mdsc;
226 unsigned frag = fpos_frag(filp->f_pos);
227 int off = fpos_off(filp->f_pos);
228 int err;
229 u32 ftype;
230 struct ceph_mds_reply_info_parsed *rinfo;
231 const int max_entries = client->mount_args->max_readdir;
233 dout("readdir %p filp %p frag %u off %u\n", inode, filp, frag, off);
234 if (fi->at_end)
235 return 0;
237 /* always start with . and .. */
238 if (filp->f_pos == 0) {
239 /* note dir version at start of readdir so we can tell
240 * if any dentries get dropped */
241 fi->dir_release_count = ci->i_release_count;
243 dout("readdir off 0 -> '.'\n");
244 if (filldir(dirent, ".", 1, ceph_make_fpos(0, 0),
245 inode->i_ino, inode->i_mode >> 12) < 0)
246 return 0;
247 filp->f_pos = 1;
248 off = 1;
250 if (filp->f_pos == 1) {
251 dout("readdir off 1 -> '..'\n");
252 if (filldir(dirent, "..", 2, ceph_make_fpos(0, 1),
253 filp->f_dentry->d_parent->d_inode->i_ino,
254 inode->i_mode >> 12) < 0)
255 return 0;
256 filp->f_pos = 2;
257 off = 2;
260 /* can we use the dcache? */
261 spin_lock(&inode->i_lock);
262 if ((filp->f_pos == 2 || fi->dentry) &&
263 !ceph_test_opt(client, NOASYNCREADDIR) &&
264 (ci->i_ceph_flags & CEPH_I_COMPLETE) &&
265 __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1)) {
266 err = __dcache_readdir(filp, dirent, filldir);
267 if (err != -EAGAIN) {
268 spin_unlock(&inode->i_lock);
269 return err;
272 spin_unlock(&inode->i_lock);
273 if (fi->dentry) {
274 err = note_last_dentry(fi, fi->dentry->d_name.name,
275 fi->dentry->d_name.len);
276 if (err)
277 return err;
278 dput(fi->dentry);
279 fi->dentry = NULL;
282 /* proceed with a normal readdir */
284 more:
285 /* do we have the correct frag content buffered? */
286 if (fi->frag != frag || fi->last_readdir == NULL) {
287 struct ceph_mds_request *req;
288 int op = ceph_snap(inode) == CEPH_SNAPDIR ?
289 CEPH_MDS_OP_LSSNAP : CEPH_MDS_OP_READDIR;
291 /* discard old result, if any */
292 if (fi->last_readdir) {
293 ceph_mdsc_put_request(fi->last_readdir);
294 fi->last_readdir = NULL;
297 /* requery frag tree, as the frag topology may have changed */
298 frag = ceph_choose_frag(ceph_inode(inode), frag, NULL, NULL);
300 dout("readdir fetching %llx.%llx frag %x offset '%s'\n",
301 ceph_vinop(inode), frag, fi->last_name);
302 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
303 if (IS_ERR(req))
304 return PTR_ERR(req);
305 req->r_inode = igrab(inode);
306 req->r_dentry = dget(filp->f_dentry);
307 /* hints to request -> mds selection code */
308 req->r_direct_mode = USE_AUTH_MDS;
309 req->r_direct_hash = ceph_frag_value(frag);
310 req->r_direct_is_hash = true;
311 req->r_path2 = kstrdup(fi->last_name, GFP_NOFS);
312 req->r_readdir_offset = fi->next_offset;
313 req->r_args.readdir.frag = cpu_to_le32(frag);
314 req->r_args.readdir.max_entries = cpu_to_le32(max_entries);
315 req->r_num_caps = max_entries + 1;
316 err = ceph_mdsc_do_request(mdsc, NULL, req);
317 if (err < 0) {
318 ceph_mdsc_put_request(req);
319 return err;
321 dout("readdir got and parsed readdir result=%d"
322 " on frag %x, end=%d, complete=%d\n", err, frag,
323 (int)req->r_reply_info.dir_end,
324 (int)req->r_reply_info.dir_complete);
326 if (!req->r_did_prepopulate) {
327 dout("readdir !did_prepopulate");
328 fi->dir_release_count--; /* preclude I_COMPLETE */
331 /* note next offset and last dentry name */
332 fi->offset = fi->next_offset;
333 fi->last_readdir = req;
335 if (req->r_reply_info.dir_end) {
336 kfree(fi->last_name);
337 fi->last_name = NULL;
338 fi->next_offset = 0;
339 } else {
340 rinfo = &req->r_reply_info;
341 err = note_last_dentry(fi,
342 rinfo->dir_dname[rinfo->dir_nr-1],
343 rinfo->dir_dname_len[rinfo->dir_nr-1]);
344 if (err)
345 return err;
346 fi->next_offset += rinfo->dir_nr;
350 rinfo = &fi->last_readdir->r_reply_info;
351 dout("readdir frag %x num %d off %d chunkoff %d\n", frag,
352 rinfo->dir_nr, off, fi->offset);
353 while (off - fi->offset >= 0 && off - fi->offset < rinfo->dir_nr) {
354 u64 pos = ceph_make_fpos(frag, off);
355 struct ceph_mds_reply_inode *in =
356 rinfo->dir_in[off - fi->offset].in;
357 dout("readdir off %d (%d/%d) -> %lld '%.*s' %p\n",
358 off, off - fi->offset, rinfo->dir_nr, pos,
359 rinfo->dir_dname_len[off - fi->offset],
360 rinfo->dir_dname[off - fi->offset], in);
361 BUG_ON(!in);
362 ftype = le32_to_cpu(in->mode) >> 12;
363 if (filldir(dirent,
364 rinfo->dir_dname[off - fi->offset],
365 rinfo->dir_dname_len[off - fi->offset],
366 pos,
367 le64_to_cpu(in->ino),
368 ftype) < 0) {
369 dout("filldir stopping us...\n");
370 return 0;
372 off++;
373 filp->f_pos = pos + 1;
376 if (fi->last_name) {
377 ceph_mdsc_put_request(fi->last_readdir);
378 fi->last_readdir = NULL;
379 goto more;
382 /* more frags? */
383 if (!ceph_frag_is_rightmost(frag)) {
384 frag = ceph_frag_next(frag);
385 off = 0;
386 filp->f_pos = ceph_make_fpos(frag, off);
387 dout("readdir next frag is %x\n", frag);
388 goto more;
390 fi->at_end = 1;
393 * if dir_release_count still matches the dir, no dentries
394 * were released during the whole readdir, and we should have
395 * the complete dir contents in our cache.
397 spin_lock(&inode->i_lock);
398 if (ci->i_release_count == fi->dir_release_count) {
399 dout(" marking %p complete\n", inode);
400 ci->i_ceph_flags |= CEPH_I_COMPLETE;
401 ci->i_max_offset = filp->f_pos;
403 spin_unlock(&inode->i_lock);
405 dout("readdir %p filp %p done.\n", inode, filp);
406 return 0;
409 static void reset_readdir(struct ceph_file_info *fi)
411 if (fi->last_readdir) {
412 ceph_mdsc_put_request(fi->last_readdir);
413 fi->last_readdir = NULL;
415 kfree(fi->last_name);
416 fi->next_offset = 2; /* compensate for . and .. */
417 if (fi->dentry) {
418 dput(fi->dentry);
419 fi->dentry = NULL;
421 fi->at_end = 0;
424 static loff_t ceph_dir_llseek(struct file *file, loff_t offset, int origin)
426 struct ceph_file_info *fi = file->private_data;
427 struct inode *inode = file->f_mapping->host;
428 loff_t old_offset = offset;
429 loff_t retval;
431 mutex_lock(&inode->i_mutex);
432 switch (origin) {
433 case SEEK_END:
434 offset += inode->i_size + 2; /* FIXME */
435 break;
436 case SEEK_CUR:
437 offset += file->f_pos;
439 retval = -EINVAL;
440 if (offset >= 0 && offset <= inode->i_sb->s_maxbytes) {
441 if (offset != file->f_pos) {
442 file->f_pos = offset;
443 file->f_version = 0;
444 fi->at_end = 0;
446 retval = offset;
449 * discard buffered readdir content on seekdir(0), or
450 * seek to new frag, or seek prior to current chunk.
452 if (offset == 0 ||
453 fpos_frag(offset) != fpos_frag(old_offset) ||
454 fpos_off(offset) < fi->offset) {
455 dout("dir_llseek dropping %p content\n", file);
456 reset_readdir(fi);
459 /* bump dir_release_count if we did a forward seek */
460 if (offset > old_offset)
461 fi->dir_release_count--;
463 mutex_unlock(&inode->i_mutex);
464 return retval;
468 * Process result of a lookup/open request.
470 * Mainly, make sure we return the final req->r_dentry (if it already
471 * existed) in place of the original VFS-provided dentry when they
472 * differ.
474 * Gracefully handle the case where the MDS replies with -ENOENT and
475 * no trace (which it may do, at its discretion, e.g., if it doesn't
476 * care to issue a lease on the negative dentry).
478 struct dentry *ceph_finish_lookup(struct ceph_mds_request *req,
479 struct dentry *dentry, int err)
481 struct ceph_client *client = ceph_client(dentry->d_sb);
482 struct inode *parent = dentry->d_parent->d_inode;
484 /* .snap dir? */
485 if (err == -ENOENT &&
486 ceph_vino(parent).ino != CEPH_INO_ROOT && /* no .snap in root dir */
487 strcmp(dentry->d_name.name,
488 client->mount_args->snapdir_name) == 0) {
489 struct inode *inode = ceph_get_snapdir(parent);
490 dout("ENOENT on snapdir %p '%.*s', linking to snapdir %p\n",
491 dentry, dentry->d_name.len, dentry->d_name.name, inode);
492 BUG_ON(!d_unhashed(dentry));
493 d_add(dentry, inode);
494 err = 0;
497 if (err == -ENOENT) {
498 /* no trace? */
499 err = 0;
500 if (!req->r_reply_info.head->is_dentry) {
501 dout("ENOENT and no trace, dentry %p inode %p\n",
502 dentry, dentry->d_inode);
503 if (dentry->d_inode) {
504 d_drop(dentry);
505 err = -ENOENT;
506 } else {
507 d_add(dentry, NULL);
511 if (err)
512 dentry = ERR_PTR(err);
513 else if (dentry != req->r_dentry)
514 dentry = dget(req->r_dentry); /* we got spliced */
515 else
516 dentry = NULL;
517 return dentry;
520 static int is_root_ceph_dentry(struct inode *inode, struct dentry *dentry)
522 return ceph_ino(inode) == CEPH_INO_ROOT &&
523 strncmp(dentry->d_name.name, ".ceph", 5) == 0;
527 * Look up a single dir entry. If there is a lookup intent, inform
528 * the MDS so that it gets our 'caps wanted' value in a single op.
530 static struct dentry *ceph_lookup(struct inode *dir, struct dentry *dentry,
531 struct nameidata *nd)
533 struct ceph_client *client = ceph_sb_to_client(dir->i_sb);
534 struct ceph_mds_client *mdsc = &client->mdsc;
535 struct ceph_mds_request *req;
536 int op;
537 int err;
539 dout("lookup %p dentry %p '%.*s'\n",
540 dir, dentry, dentry->d_name.len, dentry->d_name.name);
542 if (dentry->d_name.len > NAME_MAX)
543 return ERR_PTR(-ENAMETOOLONG);
545 err = ceph_init_dentry(dentry);
546 if (err < 0)
547 return ERR_PTR(err);
549 /* open (but not create!) intent? */
550 if (nd &&
551 (nd->flags & LOOKUP_OPEN) &&
552 (nd->flags & LOOKUP_CONTINUE) == 0 && /* only open last component */
553 !(nd->intent.open.flags & O_CREAT)) {
554 int mode = nd->intent.open.create_mode & ~current->fs->umask;
555 return ceph_lookup_open(dir, dentry, nd, mode, 1);
558 /* can we conclude ENOENT locally? */
559 if (dentry->d_inode == NULL) {
560 struct ceph_inode_info *ci = ceph_inode(dir);
561 struct ceph_dentry_info *di = ceph_dentry(dentry);
563 spin_lock(&dir->i_lock);
564 dout(" dir %p flags are %d\n", dir, ci->i_ceph_flags);
565 if (strncmp(dentry->d_name.name,
566 client->mount_args->snapdir_name,
567 dentry->d_name.len) &&
568 !is_root_ceph_dentry(dir, dentry) &&
569 (ci->i_ceph_flags & CEPH_I_COMPLETE) &&
570 (__ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1))) {
571 di->offset = ci->i_max_offset++;
572 spin_unlock(&dir->i_lock);
573 dout(" dir %p complete, -ENOENT\n", dir);
574 d_add(dentry, NULL);
575 di->lease_shared_gen = ci->i_shared_gen;
576 return NULL;
578 spin_unlock(&dir->i_lock);
581 op = ceph_snap(dir) == CEPH_SNAPDIR ?
582 CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP;
583 req = ceph_mdsc_create_request(mdsc, op, USE_ANY_MDS);
584 if (IS_ERR(req))
585 return ERR_PTR(PTR_ERR(req));
586 req->r_dentry = dget(dentry);
587 req->r_num_caps = 2;
588 /* we only need inode linkage */
589 req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
590 req->r_locked_dir = dir;
591 err = ceph_mdsc_do_request(mdsc, NULL, req);
592 dentry = ceph_finish_lookup(req, dentry, err);
593 ceph_mdsc_put_request(req); /* will dput(dentry) */
594 dout("lookup result=%p\n", dentry);
595 return dentry;
599 * If we do a create but get no trace back from the MDS, follow up with
600 * a lookup (the VFS expects us to link up the provided dentry).
602 int ceph_handle_notrace_create(struct inode *dir, struct dentry *dentry)
604 struct dentry *result = ceph_lookup(dir, dentry, NULL);
606 if (result && !IS_ERR(result)) {
608 * We created the item, then did a lookup, and found
609 * it was already linked to another inode we already
610 * had in our cache (and thus got spliced). Link our
611 * dentry to that inode, but don't hash it, just in
612 * case the VFS wants to dereference it.
614 BUG_ON(!result->d_inode);
615 d_instantiate(dentry, result->d_inode);
616 return 0;
618 return PTR_ERR(result);
621 static int ceph_mknod(struct inode *dir, struct dentry *dentry,
622 int mode, dev_t rdev)
624 struct ceph_client *client = ceph_sb_to_client(dir->i_sb);
625 struct ceph_mds_client *mdsc = &client->mdsc;
626 struct ceph_mds_request *req;
627 int err;
629 if (ceph_snap(dir) != CEPH_NOSNAP)
630 return -EROFS;
632 dout("mknod in dir %p dentry %p mode 0%o rdev %d\n",
633 dir, dentry, mode, rdev);
634 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_MKNOD, USE_AUTH_MDS);
635 if (IS_ERR(req)) {
636 d_drop(dentry);
637 return PTR_ERR(req);
639 req->r_dentry = dget(dentry);
640 req->r_num_caps = 2;
641 req->r_locked_dir = dir;
642 req->r_args.mknod.mode = cpu_to_le32(mode);
643 req->r_args.mknod.rdev = cpu_to_le32(rdev);
644 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
645 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
646 err = ceph_mdsc_do_request(mdsc, dir, req);
647 if (!err && !req->r_reply_info.head->is_dentry)
648 err = ceph_handle_notrace_create(dir, dentry);
649 ceph_mdsc_put_request(req);
650 if (err)
651 d_drop(dentry);
652 return err;
655 static int ceph_create(struct inode *dir, struct dentry *dentry, int mode,
656 struct nameidata *nd)
658 dout("create in dir %p dentry %p name '%.*s'\n",
659 dir, dentry, dentry->d_name.len, dentry->d_name.name);
661 if (ceph_snap(dir) != CEPH_NOSNAP)
662 return -EROFS;
664 if (nd) {
665 BUG_ON((nd->flags & LOOKUP_OPEN) == 0);
666 dentry = ceph_lookup_open(dir, dentry, nd, mode, 0);
667 /* hrm, what should i do here if we get aliased? */
668 if (IS_ERR(dentry))
669 return PTR_ERR(dentry);
670 return 0;
673 /* fall back to mknod */
674 return ceph_mknod(dir, dentry, (mode & ~S_IFMT) | S_IFREG, 0);
677 static int ceph_symlink(struct inode *dir, struct dentry *dentry,
678 const char *dest)
680 struct ceph_client *client = ceph_sb_to_client(dir->i_sb);
681 struct ceph_mds_client *mdsc = &client->mdsc;
682 struct ceph_mds_request *req;
683 int err;
685 if (ceph_snap(dir) != CEPH_NOSNAP)
686 return -EROFS;
688 dout("symlink in dir %p dentry %p to '%s'\n", dir, dentry, dest);
689 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SYMLINK, USE_AUTH_MDS);
690 if (IS_ERR(req)) {
691 d_drop(dentry);
692 return PTR_ERR(req);
694 req->r_dentry = dget(dentry);
695 req->r_num_caps = 2;
696 req->r_path2 = kstrdup(dest, GFP_NOFS);
697 req->r_locked_dir = dir;
698 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
699 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
700 err = ceph_mdsc_do_request(mdsc, dir, req);
701 if (!err && !req->r_reply_info.head->is_dentry)
702 err = ceph_handle_notrace_create(dir, dentry);
703 ceph_mdsc_put_request(req);
704 if (err)
705 d_drop(dentry);
706 return err;
709 static int ceph_mkdir(struct inode *dir, struct dentry *dentry, int mode)
711 struct ceph_client *client = ceph_sb_to_client(dir->i_sb);
712 struct ceph_mds_client *mdsc = &client->mdsc;
713 struct ceph_mds_request *req;
714 int err = -EROFS;
715 int op;
717 if (ceph_snap(dir) == CEPH_SNAPDIR) {
718 /* mkdir .snap/foo is a MKSNAP */
719 op = CEPH_MDS_OP_MKSNAP;
720 dout("mksnap dir %p snap '%.*s' dn %p\n", dir,
721 dentry->d_name.len, dentry->d_name.name, dentry);
722 } else if (ceph_snap(dir) == CEPH_NOSNAP) {
723 dout("mkdir dir %p dn %p mode 0%o\n", dir, dentry, mode);
724 op = CEPH_MDS_OP_MKDIR;
725 } else {
726 goto out;
728 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
729 if (IS_ERR(req)) {
730 err = PTR_ERR(req);
731 goto out;
734 req->r_dentry = dget(dentry);
735 req->r_num_caps = 2;
736 req->r_locked_dir = dir;
737 req->r_args.mkdir.mode = cpu_to_le32(mode);
738 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
739 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
740 err = ceph_mdsc_do_request(mdsc, dir, req);
741 if (!err && !req->r_reply_info.head->is_dentry)
742 err = ceph_handle_notrace_create(dir, dentry);
743 ceph_mdsc_put_request(req);
744 out:
745 if (err < 0)
746 d_drop(dentry);
747 return err;
750 static int ceph_link(struct dentry *old_dentry, struct inode *dir,
751 struct dentry *dentry)
753 struct ceph_client *client = ceph_sb_to_client(dir->i_sb);
754 struct ceph_mds_client *mdsc = &client->mdsc;
755 struct ceph_mds_request *req;
756 int err;
758 if (ceph_snap(dir) != CEPH_NOSNAP)
759 return -EROFS;
761 dout("link in dir %p old_dentry %p dentry %p\n", dir,
762 old_dentry, dentry);
763 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LINK, USE_AUTH_MDS);
764 if (IS_ERR(req)) {
765 d_drop(dentry);
766 return PTR_ERR(req);
768 req->r_dentry = dget(dentry);
769 req->r_num_caps = 2;
770 req->r_old_dentry = dget(old_dentry); /* or inode? hrm. */
771 req->r_locked_dir = dir;
772 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
773 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
774 err = ceph_mdsc_do_request(mdsc, dir, req);
775 if (err)
776 d_drop(dentry);
777 else if (!req->r_reply_info.head->is_dentry)
778 d_instantiate(dentry, igrab(old_dentry->d_inode));
779 ceph_mdsc_put_request(req);
780 return err;
784 * For a soon-to-be unlinked file, drop the AUTH_RDCACHE caps. If it
785 * looks like the link count will hit 0, drop any other caps (other
786 * than PIN) we don't specifically want (due to the file still being
787 * open).
789 static int drop_caps_for_unlink(struct inode *inode)
791 struct ceph_inode_info *ci = ceph_inode(inode);
792 int drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL;
794 spin_lock(&inode->i_lock);
795 if (inode->i_nlink == 1) {
796 drop |= ~(__ceph_caps_wanted(ci) | CEPH_CAP_PIN);
797 ci->i_ceph_flags |= CEPH_I_NODELAY;
799 spin_unlock(&inode->i_lock);
800 return drop;
804 * rmdir and unlink are differ only by the metadata op code
806 static int ceph_unlink(struct inode *dir, struct dentry *dentry)
808 struct ceph_client *client = ceph_sb_to_client(dir->i_sb);
809 struct ceph_mds_client *mdsc = &client->mdsc;
810 struct inode *inode = dentry->d_inode;
811 struct ceph_mds_request *req;
812 int err = -EROFS;
813 int op;
815 if (ceph_snap(dir) == CEPH_SNAPDIR) {
816 /* rmdir .snap/foo is RMSNAP */
817 dout("rmsnap dir %p '%.*s' dn %p\n", dir, dentry->d_name.len,
818 dentry->d_name.name, dentry);
819 op = CEPH_MDS_OP_RMSNAP;
820 } else if (ceph_snap(dir) == CEPH_NOSNAP) {
821 dout("unlink/rmdir dir %p dn %p inode %p\n",
822 dir, dentry, inode);
823 op = ((dentry->d_inode->i_mode & S_IFMT) == S_IFDIR) ?
824 CEPH_MDS_OP_RMDIR : CEPH_MDS_OP_UNLINK;
825 } else
826 goto out;
827 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
828 if (IS_ERR(req)) {
829 err = PTR_ERR(req);
830 goto out;
832 req->r_dentry = dget(dentry);
833 req->r_num_caps = 2;
834 req->r_locked_dir = dir;
835 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
836 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
837 req->r_inode_drop = drop_caps_for_unlink(inode);
838 err = ceph_mdsc_do_request(mdsc, dir, req);
839 if (!err && !req->r_reply_info.head->is_dentry)
840 d_delete(dentry);
841 ceph_mdsc_put_request(req);
842 out:
843 return err;
846 static int ceph_rename(struct inode *old_dir, struct dentry *old_dentry,
847 struct inode *new_dir, struct dentry *new_dentry)
849 struct ceph_client *client = ceph_sb_to_client(old_dir->i_sb);
850 struct ceph_mds_client *mdsc = &client->mdsc;
851 struct ceph_mds_request *req;
852 int err;
854 if (ceph_snap(old_dir) != ceph_snap(new_dir))
855 return -EXDEV;
856 if (ceph_snap(old_dir) != CEPH_NOSNAP ||
857 ceph_snap(new_dir) != CEPH_NOSNAP)
858 return -EROFS;
859 dout("rename dir %p dentry %p to dir %p dentry %p\n",
860 old_dir, old_dentry, new_dir, new_dentry);
861 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_RENAME, USE_AUTH_MDS);
862 if (IS_ERR(req))
863 return PTR_ERR(req);
864 req->r_dentry = dget(new_dentry);
865 req->r_num_caps = 2;
866 req->r_old_dentry = dget(old_dentry);
867 req->r_locked_dir = new_dir;
868 req->r_old_dentry_drop = CEPH_CAP_FILE_SHARED;
869 req->r_old_dentry_unless = CEPH_CAP_FILE_EXCL;
870 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
871 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
872 /* release LINK_RDCACHE on source inode (mds will lock it) */
873 req->r_old_inode_drop = CEPH_CAP_LINK_SHARED;
874 if (new_dentry->d_inode)
875 req->r_inode_drop = drop_caps_for_unlink(new_dentry->d_inode);
876 err = ceph_mdsc_do_request(mdsc, old_dir, req);
877 if (!err && !req->r_reply_info.head->is_dentry) {
879 * Normally d_move() is done by fill_trace (called by
880 * do_request, above). If there is no trace, we need
881 * to do it here.
883 d_move(old_dentry, new_dentry);
885 ceph_mdsc_put_request(req);
886 return err;
891 * Check if dentry lease is valid. If not, delete the lease. Try to
892 * renew if the least is more than half up.
894 static int dentry_lease_is_valid(struct dentry *dentry)
896 struct ceph_dentry_info *di;
897 struct ceph_mds_session *s;
898 int valid = 0;
899 u32 gen;
900 unsigned long ttl;
901 struct ceph_mds_session *session = NULL;
902 struct inode *dir = NULL;
903 u32 seq = 0;
905 spin_lock(&dentry->d_lock);
906 di = ceph_dentry(dentry);
907 if (di && di->lease_session) {
908 s = di->lease_session;
909 spin_lock(&s->s_cap_lock);
910 gen = s->s_cap_gen;
911 ttl = s->s_cap_ttl;
912 spin_unlock(&s->s_cap_lock);
914 if (di->lease_gen == gen &&
915 time_before(jiffies, dentry->d_time) &&
916 time_before(jiffies, ttl)) {
917 valid = 1;
918 if (di->lease_renew_after &&
919 time_after(jiffies, di->lease_renew_after)) {
920 /* we should renew */
921 dir = dentry->d_parent->d_inode;
922 session = ceph_get_mds_session(s);
923 seq = di->lease_seq;
924 di->lease_renew_after = 0;
925 di->lease_renew_from = jiffies;
929 spin_unlock(&dentry->d_lock);
931 if (session) {
932 ceph_mdsc_lease_send_msg(session, dir, dentry,
933 CEPH_MDS_LEASE_RENEW, seq);
934 ceph_put_mds_session(session);
936 dout("dentry_lease_is_valid - dentry %p = %d\n", dentry, valid);
937 return valid;
941 * Check if directory-wide content lease/cap is valid.
943 static int dir_lease_is_valid(struct inode *dir, struct dentry *dentry)
945 struct ceph_inode_info *ci = ceph_inode(dir);
946 struct ceph_dentry_info *di = ceph_dentry(dentry);
947 int valid = 0;
949 spin_lock(&dir->i_lock);
950 if (ci->i_shared_gen == di->lease_shared_gen)
951 valid = __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1);
952 spin_unlock(&dir->i_lock);
953 dout("dir_lease_is_valid dir %p v%u dentry %p v%u = %d\n",
954 dir, (unsigned)ci->i_shared_gen, dentry,
955 (unsigned)di->lease_shared_gen, valid);
956 return valid;
960 * Check if cached dentry can be trusted.
962 static int ceph_d_revalidate(struct dentry *dentry, struct nameidata *nd)
964 struct inode *dir = dentry->d_parent->d_inode;
966 dout("d_revalidate %p '%.*s' inode %p\n", dentry,
967 dentry->d_name.len, dentry->d_name.name, dentry->d_inode);
969 /* always trust cached snapped dentries, snapdir dentry */
970 if (ceph_snap(dir) != CEPH_NOSNAP) {
971 dout("d_revalidate %p '%.*s' inode %p is SNAPPED\n", dentry,
972 dentry->d_name.len, dentry->d_name.name, dentry->d_inode);
973 goto out_touch;
975 if (dentry->d_inode && ceph_snap(dentry->d_inode) == CEPH_SNAPDIR)
976 goto out_touch;
978 if (dentry_lease_is_valid(dentry) ||
979 dir_lease_is_valid(dir, dentry))
980 goto out_touch;
982 dout("d_revalidate %p invalid\n", dentry);
983 d_drop(dentry);
984 return 0;
985 out_touch:
986 ceph_dentry_lru_touch(dentry);
987 return 1;
991 * When a dentry is released, clear the dir I_COMPLETE if it was part
992 * of the current dir gen.
994 static void ceph_dentry_release(struct dentry *dentry)
996 struct ceph_dentry_info *di = ceph_dentry(dentry);
997 struct inode *parent_inode = dentry->d_parent->d_inode;
999 if (parent_inode) {
1000 struct ceph_inode_info *ci = ceph_inode(parent_inode);
1002 spin_lock(&parent_inode->i_lock);
1003 if (ci->i_shared_gen == di->lease_shared_gen) {
1004 dout(" clearing %p complete (d_release)\n",
1005 parent_inode);
1006 ci->i_ceph_flags &= ~CEPH_I_COMPLETE;
1007 ci->i_release_count++;
1009 spin_unlock(&parent_inode->i_lock);
1011 if (di) {
1012 ceph_dentry_lru_del(dentry);
1013 if (di->lease_session)
1014 ceph_put_mds_session(di->lease_session);
1015 kmem_cache_free(ceph_dentry_cachep, di);
1016 dentry->d_fsdata = NULL;
1020 static int ceph_snapdir_d_revalidate(struct dentry *dentry,
1021 struct nameidata *nd)
1024 * Eventually, we'll want to revalidate snapped metadata
1025 * too... probably...
1027 return 1;
1033 * read() on a dir. This weird interface hack only works if mounted
1034 * with '-o dirstat'.
1036 static ssize_t ceph_read_dir(struct file *file, char __user *buf, size_t size,
1037 loff_t *ppos)
1039 struct ceph_file_info *cf = file->private_data;
1040 struct inode *inode = file->f_dentry->d_inode;
1041 struct ceph_inode_info *ci = ceph_inode(inode);
1042 int left;
1044 if (!ceph_test_opt(ceph_client(inode->i_sb), DIRSTAT))
1045 return -EISDIR;
1047 if (!cf->dir_info) {
1048 cf->dir_info = kmalloc(1024, GFP_NOFS);
1049 if (!cf->dir_info)
1050 return -ENOMEM;
1051 cf->dir_info_len =
1052 sprintf(cf->dir_info,
1053 "entries: %20lld\n"
1054 " files: %20lld\n"
1055 " subdirs: %20lld\n"
1056 "rentries: %20lld\n"
1057 " rfiles: %20lld\n"
1058 " rsubdirs: %20lld\n"
1059 "rbytes: %20lld\n"
1060 "rctime: %10ld.%09ld\n",
1061 ci->i_files + ci->i_subdirs,
1062 ci->i_files,
1063 ci->i_subdirs,
1064 ci->i_rfiles + ci->i_rsubdirs,
1065 ci->i_rfiles,
1066 ci->i_rsubdirs,
1067 ci->i_rbytes,
1068 (long)ci->i_rctime.tv_sec,
1069 (long)ci->i_rctime.tv_nsec);
1072 if (*ppos >= cf->dir_info_len)
1073 return 0;
1074 size = min_t(unsigned, size, cf->dir_info_len-*ppos);
1075 left = copy_to_user(buf, cf->dir_info + *ppos, size);
1076 if (left == size)
1077 return -EFAULT;
1078 *ppos += (size - left);
1079 return size - left;
1083 * an fsync() on a dir will wait for any uncommitted directory
1084 * operations to commit.
1086 static int ceph_dir_fsync(struct file *file, struct dentry *dentry,
1087 int datasync)
1089 struct inode *inode = dentry->d_inode;
1090 struct ceph_inode_info *ci = ceph_inode(inode);
1091 struct list_head *head = &ci->i_unsafe_dirops;
1092 struct ceph_mds_request *req;
1093 u64 last_tid;
1094 int ret = 0;
1096 dout("dir_fsync %p\n", inode);
1097 spin_lock(&ci->i_unsafe_lock);
1098 if (list_empty(head))
1099 goto out;
1101 req = list_entry(head->prev,
1102 struct ceph_mds_request, r_unsafe_dir_item);
1103 last_tid = req->r_tid;
1105 do {
1106 ceph_mdsc_get_request(req);
1107 spin_unlock(&ci->i_unsafe_lock);
1108 dout("dir_fsync %p wait on tid %llu (until %llu)\n",
1109 inode, req->r_tid, last_tid);
1110 if (req->r_timeout) {
1111 ret = wait_for_completion_timeout(
1112 &req->r_safe_completion, req->r_timeout);
1113 if (ret > 0)
1114 ret = 0;
1115 else if (ret == 0)
1116 ret = -EIO; /* timed out */
1117 } else {
1118 wait_for_completion(&req->r_safe_completion);
1120 spin_lock(&ci->i_unsafe_lock);
1121 ceph_mdsc_put_request(req);
1123 if (ret || list_empty(head))
1124 break;
1125 req = list_entry(head->next,
1126 struct ceph_mds_request, r_unsafe_dir_item);
1127 } while (req->r_tid < last_tid);
1128 out:
1129 spin_unlock(&ci->i_unsafe_lock);
1130 return ret;
1134 * We maintain a private dentry LRU.
1136 * FIXME: this needs to be changed to a per-mds lru to be useful.
1138 void ceph_dentry_lru_add(struct dentry *dn)
1140 struct ceph_dentry_info *di = ceph_dentry(dn);
1141 struct ceph_mds_client *mdsc;
1143 dout("dentry_lru_add %p %p '%.*s'\n", di, dn,
1144 dn->d_name.len, dn->d_name.name);
1145 if (di) {
1146 mdsc = &ceph_client(dn->d_sb)->mdsc;
1147 spin_lock(&mdsc->dentry_lru_lock);
1148 list_add_tail(&di->lru, &mdsc->dentry_lru);
1149 mdsc->num_dentry++;
1150 spin_unlock(&mdsc->dentry_lru_lock);
1154 void ceph_dentry_lru_touch(struct dentry *dn)
1156 struct ceph_dentry_info *di = ceph_dentry(dn);
1157 struct ceph_mds_client *mdsc;
1159 dout("dentry_lru_touch %p %p '%.*s'\n", di, dn,
1160 dn->d_name.len, dn->d_name.name);
1161 if (di) {
1162 mdsc = &ceph_client(dn->d_sb)->mdsc;
1163 spin_lock(&mdsc->dentry_lru_lock);
1164 list_move_tail(&di->lru, &mdsc->dentry_lru);
1165 spin_unlock(&mdsc->dentry_lru_lock);
1169 void ceph_dentry_lru_del(struct dentry *dn)
1171 struct ceph_dentry_info *di = ceph_dentry(dn);
1172 struct ceph_mds_client *mdsc;
1174 dout("dentry_lru_del %p %p '%.*s'\n", di, dn,
1175 dn->d_name.len, dn->d_name.name);
1176 if (di) {
1177 mdsc = &ceph_client(dn->d_sb)->mdsc;
1178 spin_lock(&mdsc->dentry_lru_lock);
1179 list_del_init(&di->lru);
1180 mdsc->num_dentry--;
1181 spin_unlock(&mdsc->dentry_lru_lock);
1185 const struct file_operations ceph_dir_fops = {
1186 .read = ceph_read_dir,
1187 .readdir = ceph_readdir,
1188 .llseek = ceph_dir_llseek,
1189 .open = ceph_open,
1190 .release = ceph_release,
1191 .unlocked_ioctl = ceph_ioctl,
1192 .fsync = ceph_dir_fsync,
1195 const struct inode_operations ceph_dir_iops = {
1196 .lookup = ceph_lookup,
1197 .permission = ceph_permission,
1198 .getattr = ceph_getattr,
1199 .setattr = ceph_setattr,
1200 .setxattr = ceph_setxattr,
1201 .getxattr = ceph_getxattr,
1202 .listxattr = ceph_listxattr,
1203 .removexattr = ceph_removexattr,
1204 .mknod = ceph_mknod,
1205 .symlink = ceph_symlink,
1206 .mkdir = ceph_mkdir,
1207 .link = ceph_link,
1208 .unlink = ceph_unlink,
1209 .rmdir = ceph_unlink,
1210 .rename = ceph_rename,
1211 .create = ceph_create,
1214 struct dentry_operations ceph_dentry_ops = {
1215 .d_revalidate = ceph_d_revalidate,
1216 .d_release = ceph_dentry_release,
1219 struct dentry_operations ceph_snapdir_dentry_ops = {
1220 .d_revalidate = ceph_snapdir_d_revalidate,
1223 struct dentry_operations ceph_snap_dentry_ops = {