Linux 3.12.5
[linux/fpc-iii.git] / fs / ceph / dir.c
blob868b61d56cac77f3a8328d5ba4851ec7947fe827
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 *file, struct dir_context *ctx)
116 struct ceph_file_info *fi = file->private_data;
117 struct dentry *parent = file->f_dentry;
118 struct inode *dir = parent->d_inode;
119 struct list_head *p;
120 struct dentry *dentry, *last;
121 struct ceph_dentry_info *di;
122 int err = 0;
124 /* claim ref on last dentry we returned */
125 last = fi->dentry;
126 fi->dentry = NULL;
128 dout("__dcache_readdir %p at %llu (last %p)\n", dir, ctx->pos,
129 last);
131 spin_lock(&parent->d_lock);
133 /* start at beginning? */
134 if (ctx->pos == 2 || last == NULL ||
135 ctx->pos < ceph_dentry(last)->offset) {
136 if (list_empty(&parent->d_subdirs))
137 goto out_unlock;
138 p = parent->d_subdirs.prev;
139 dout(" initial p %p/%p\n", p->prev, p->next);
140 } else {
141 p = last->d_u.d_child.prev;
144 more:
145 dentry = list_entry(p, struct dentry, d_u.d_child);
146 di = ceph_dentry(dentry);
147 while (1) {
148 dout(" p %p/%p %s d_subdirs %p/%p\n", p->prev, p->next,
149 d_unhashed(dentry) ? "!hashed" : "hashed",
150 parent->d_subdirs.prev, parent->d_subdirs.next);
151 if (p == &parent->d_subdirs) {
152 fi->flags |= CEPH_F_ATEND;
153 goto out_unlock;
155 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
156 if (!d_unhashed(dentry) && dentry->d_inode &&
157 ceph_snap(dentry->d_inode) != CEPH_SNAPDIR &&
158 ceph_ino(dentry->d_inode) != CEPH_INO_CEPH &&
159 ctx->pos <= di->offset)
160 break;
161 dout(" skipping %p %.*s at %llu (%llu)%s%s\n", dentry,
162 dentry->d_name.len, dentry->d_name.name, di->offset,
163 ctx->pos, d_unhashed(dentry) ? " unhashed" : "",
164 !dentry->d_inode ? " null" : "");
165 spin_unlock(&dentry->d_lock);
166 p = p->prev;
167 dentry = list_entry(p, struct dentry, d_u.d_child);
168 di = ceph_dentry(dentry);
171 dget_dlock(dentry);
172 spin_unlock(&dentry->d_lock);
173 spin_unlock(&parent->d_lock);
175 dout(" %llu (%llu) dentry %p %.*s %p\n", di->offset, ctx->pos,
176 dentry, dentry->d_name.len, dentry->d_name.name, dentry->d_inode);
177 ctx->pos = di->offset;
178 if (!dir_emit(ctx, dentry->d_name.name,
179 dentry->d_name.len,
180 ceph_translate_ino(dentry->d_sb, dentry->d_inode->i_ino),
181 dentry->d_inode->i_mode >> 12)) {
182 if (last) {
183 /* remember our position */
184 fi->dentry = last;
185 fi->next_offset = di->offset;
187 dput(dentry);
188 return 0;
191 if (last)
192 dput(last);
193 last = dentry;
195 ctx->pos++;
197 /* make sure a dentry wasn't dropped while we didn't have parent lock */
198 if (!ceph_dir_is_complete(dir)) {
199 dout(" lost dir complete on %p; falling back to mds\n", dir);
200 err = -EAGAIN;
201 goto out;
204 spin_lock(&parent->d_lock);
205 p = p->prev; /* advance to next dentry */
206 goto more;
208 out_unlock:
209 spin_unlock(&parent->d_lock);
210 out:
211 if (last)
212 dput(last);
213 return err;
217 * make note of the last dentry we read, so we can
218 * continue at the same lexicographical point,
219 * regardless of what dir changes take place on the
220 * server.
222 static int note_last_dentry(struct ceph_file_info *fi, const char *name,
223 int len)
225 kfree(fi->last_name);
226 fi->last_name = kmalloc(len+1, GFP_NOFS);
227 if (!fi->last_name)
228 return -ENOMEM;
229 memcpy(fi->last_name, name, len);
230 fi->last_name[len] = 0;
231 dout("note_last_dentry '%s'\n", fi->last_name);
232 return 0;
235 static int ceph_readdir(struct file *file, struct dir_context *ctx)
237 struct ceph_file_info *fi = file->private_data;
238 struct inode *inode = file_inode(file);
239 struct ceph_inode_info *ci = ceph_inode(inode);
240 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
241 struct ceph_mds_client *mdsc = fsc->mdsc;
242 unsigned frag = fpos_frag(ctx->pos);
243 int off = fpos_off(ctx->pos);
244 int err;
245 u32 ftype;
246 struct ceph_mds_reply_info_parsed *rinfo;
247 const int max_entries = fsc->mount_options->max_readdir;
248 const int max_bytes = fsc->mount_options->max_readdir_bytes;
250 dout("readdir %p file %p frag %u off %u\n", inode, file, frag, off);
251 if (fi->flags & CEPH_F_ATEND)
252 return 0;
254 /* always start with . and .. */
255 if (ctx->pos == 0) {
256 /* note dir version at start of readdir so we can tell
257 * if any dentries get dropped */
258 fi->dir_release_count = atomic_read(&ci->i_release_count);
260 dout("readdir off 0 -> '.'\n");
261 if (!dir_emit(ctx, ".", 1,
262 ceph_translate_ino(inode->i_sb, inode->i_ino),
263 inode->i_mode >> 12))
264 return 0;
265 ctx->pos = 1;
266 off = 1;
268 if (ctx->pos == 1) {
269 ino_t ino = parent_ino(file->f_dentry);
270 dout("readdir off 1 -> '..'\n");
271 if (!dir_emit(ctx, "..", 2,
272 ceph_translate_ino(inode->i_sb, ino),
273 inode->i_mode >> 12))
274 return 0;
275 ctx->pos = 2;
276 off = 2;
279 /* can we use the dcache? */
280 spin_lock(&ci->i_ceph_lock);
281 if ((ctx->pos == 2 || fi->dentry) &&
282 !ceph_test_mount_opt(fsc, NOASYNCREADDIR) &&
283 ceph_snap(inode) != CEPH_SNAPDIR &&
284 __ceph_dir_is_complete(ci) &&
285 __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1)) {
286 spin_unlock(&ci->i_ceph_lock);
287 err = __dcache_readdir(file, ctx);
288 if (err != -EAGAIN)
289 return err;
290 } else {
291 spin_unlock(&ci->i_ceph_lock);
293 if (fi->dentry) {
294 err = note_last_dentry(fi, fi->dentry->d_name.name,
295 fi->dentry->d_name.len);
296 if (err)
297 return err;
298 dput(fi->dentry);
299 fi->dentry = NULL;
302 /* proceed with a normal readdir */
304 more:
305 /* do we have the correct frag content buffered? */
306 if (fi->frag != frag || fi->last_readdir == NULL) {
307 struct ceph_mds_request *req;
308 int op = ceph_snap(inode) == CEPH_SNAPDIR ?
309 CEPH_MDS_OP_LSSNAP : CEPH_MDS_OP_READDIR;
311 /* discard old result, if any */
312 if (fi->last_readdir) {
313 ceph_mdsc_put_request(fi->last_readdir);
314 fi->last_readdir = NULL;
317 /* requery frag tree, as the frag topology may have changed */
318 frag = ceph_choose_frag(ceph_inode(inode), frag, NULL, NULL);
320 dout("readdir fetching %llx.%llx frag %x offset '%s'\n",
321 ceph_vinop(inode), frag, fi->last_name);
322 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
323 if (IS_ERR(req))
324 return PTR_ERR(req);
325 req->r_inode = inode;
326 ihold(inode);
327 req->r_dentry = dget(file->f_dentry);
328 /* hints to request -> mds selection code */
329 req->r_direct_mode = USE_AUTH_MDS;
330 req->r_direct_hash = ceph_frag_value(frag);
331 req->r_direct_is_hash = true;
332 req->r_path2 = kstrdup(fi->last_name, GFP_NOFS);
333 req->r_readdir_offset = fi->next_offset;
334 req->r_args.readdir.frag = cpu_to_le32(frag);
335 req->r_args.readdir.max_entries = cpu_to_le32(max_entries);
336 req->r_args.readdir.max_bytes = cpu_to_le32(max_bytes);
337 req->r_num_caps = max_entries + 1;
338 err = ceph_mdsc_do_request(mdsc, NULL, req);
339 if (err < 0) {
340 ceph_mdsc_put_request(req);
341 return err;
343 dout("readdir got and parsed readdir result=%d"
344 " on frag %x, end=%d, complete=%d\n", err, frag,
345 (int)req->r_reply_info.dir_end,
346 (int)req->r_reply_info.dir_complete);
348 if (!req->r_did_prepopulate) {
349 dout("readdir !did_prepopulate");
350 /* preclude from marking dir complete */
351 fi->dir_release_count--;
354 /* note next offset and last dentry name */
355 fi->offset = fi->next_offset;
356 fi->last_readdir = req;
358 if (req->r_reply_info.dir_end) {
359 kfree(fi->last_name);
360 fi->last_name = NULL;
361 if (ceph_frag_is_rightmost(frag))
362 fi->next_offset = 2;
363 else
364 fi->next_offset = 0;
365 } else {
366 rinfo = &req->r_reply_info;
367 err = note_last_dentry(fi,
368 rinfo->dir_dname[rinfo->dir_nr-1],
369 rinfo->dir_dname_len[rinfo->dir_nr-1]);
370 if (err)
371 return err;
372 fi->next_offset += rinfo->dir_nr;
376 rinfo = &fi->last_readdir->r_reply_info;
377 dout("readdir frag %x num %d off %d chunkoff %d\n", frag,
378 rinfo->dir_nr, off, fi->offset);
380 ctx->pos = ceph_make_fpos(frag, off);
381 while (off >= fi->offset && off - fi->offset < rinfo->dir_nr) {
382 struct ceph_mds_reply_inode *in =
383 rinfo->dir_in[off - fi->offset].in;
384 struct ceph_vino vino;
385 ino_t ino;
387 dout("readdir off %d (%d/%d) -> %lld '%.*s' %p\n",
388 off, off - fi->offset, rinfo->dir_nr, ctx->pos,
389 rinfo->dir_dname_len[off - fi->offset],
390 rinfo->dir_dname[off - fi->offset], in);
391 BUG_ON(!in);
392 ftype = le32_to_cpu(in->mode) >> 12;
393 vino.ino = le64_to_cpu(in->ino);
394 vino.snap = le64_to_cpu(in->snapid);
395 ino = ceph_vino_to_ino(vino);
396 if (!dir_emit(ctx,
397 rinfo->dir_dname[off - fi->offset],
398 rinfo->dir_dname_len[off - fi->offset],
399 ceph_translate_ino(inode->i_sb, ino), ftype)) {
400 dout("filldir stopping us...\n");
401 return 0;
403 off++;
404 ctx->pos++;
407 if (fi->last_name) {
408 ceph_mdsc_put_request(fi->last_readdir);
409 fi->last_readdir = NULL;
410 goto more;
413 /* more frags? */
414 if (!ceph_frag_is_rightmost(frag)) {
415 frag = ceph_frag_next(frag);
416 off = 0;
417 ctx->pos = ceph_make_fpos(frag, off);
418 dout("readdir next frag is %x\n", frag);
419 goto more;
421 fi->flags |= CEPH_F_ATEND;
424 * if dir_release_count still matches the dir, no dentries
425 * were released during the whole readdir, and we should have
426 * the complete dir contents in our cache.
428 spin_lock(&ci->i_ceph_lock);
429 if (atomic_read(&ci->i_release_count) == fi->dir_release_count) {
430 dout(" marking %p complete\n", inode);
431 __ceph_dir_set_complete(ci, fi->dir_release_count);
432 ci->i_max_offset = ctx->pos;
434 spin_unlock(&ci->i_ceph_lock);
436 dout("readdir %p file %p done.\n", inode, file);
437 return 0;
440 static void reset_readdir(struct ceph_file_info *fi)
442 if (fi->last_readdir) {
443 ceph_mdsc_put_request(fi->last_readdir);
444 fi->last_readdir = NULL;
446 kfree(fi->last_name);
447 fi->last_name = NULL;
448 fi->next_offset = 2; /* compensate for . and .. */
449 if (fi->dentry) {
450 dput(fi->dentry);
451 fi->dentry = NULL;
453 fi->flags &= ~CEPH_F_ATEND;
456 static loff_t ceph_dir_llseek(struct file *file, loff_t offset, int whence)
458 struct ceph_file_info *fi = file->private_data;
459 struct inode *inode = file->f_mapping->host;
460 loff_t old_offset = offset;
461 loff_t retval;
463 mutex_lock(&inode->i_mutex);
464 retval = -EINVAL;
465 switch (whence) {
466 case SEEK_END:
467 offset += inode->i_size + 2; /* FIXME */
468 break;
469 case SEEK_CUR:
470 offset += file->f_pos;
471 case SEEK_SET:
472 break;
473 default:
474 goto out;
477 if (offset >= 0 && offset <= inode->i_sb->s_maxbytes) {
478 if (offset != file->f_pos) {
479 file->f_pos = offset;
480 file->f_version = 0;
481 fi->flags &= ~CEPH_F_ATEND;
483 retval = offset;
486 * discard buffered readdir content on seekdir(0), or
487 * seek to new frag, or seek prior to current chunk.
489 if (offset == 0 ||
490 fpos_frag(offset) != fpos_frag(old_offset) ||
491 fpos_off(offset) < fi->offset) {
492 dout("dir_llseek dropping %p content\n", file);
493 reset_readdir(fi);
496 /* bump dir_release_count if we did a forward seek */
497 if (offset > old_offset)
498 fi->dir_release_count--;
500 out:
501 mutex_unlock(&inode->i_mutex);
502 return retval;
506 * Handle lookups for the hidden .snap directory.
508 int ceph_handle_snapdir(struct ceph_mds_request *req,
509 struct dentry *dentry, int err)
511 struct ceph_fs_client *fsc = ceph_sb_to_client(dentry->d_sb);
512 struct inode *parent = dentry->d_parent->d_inode; /* we hold i_mutex */
514 /* .snap dir? */
515 if (err == -ENOENT &&
516 ceph_snap(parent) == CEPH_NOSNAP &&
517 strcmp(dentry->d_name.name,
518 fsc->mount_options->snapdir_name) == 0) {
519 struct inode *inode = ceph_get_snapdir(parent);
520 dout("ENOENT on snapdir %p '%.*s', linking to snapdir %p\n",
521 dentry, dentry->d_name.len, dentry->d_name.name, inode);
522 BUG_ON(!d_unhashed(dentry));
523 d_add(dentry, inode);
524 err = 0;
526 return err;
530 * Figure out final result of a lookup/open request.
532 * Mainly, make sure we return the final req->r_dentry (if it already
533 * existed) in place of the original VFS-provided dentry when they
534 * differ.
536 * Gracefully handle the case where the MDS replies with -ENOENT and
537 * no trace (which it may do, at its discretion, e.g., if it doesn't
538 * care to issue a lease on the negative dentry).
540 struct dentry *ceph_finish_lookup(struct ceph_mds_request *req,
541 struct dentry *dentry, int err)
543 if (err == -ENOENT) {
544 /* no trace? */
545 err = 0;
546 if (!req->r_reply_info.head->is_dentry) {
547 dout("ENOENT and no trace, dentry %p inode %p\n",
548 dentry, dentry->d_inode);
549 if (dentry->d_inode) {
550 d_drop(dentry);
551 err = -ENOENT;
552 } else {
553 d_add(dentry, NULL);
557 if (err)
558 dentry = ERR_PTR(err);
559 else if (dentry != req->r_dentry)
560 dentry = dget(req->r_dentry); /* we got spliced */
561 else
562 dentry = NULL;
563 return dentry;
566 static int is_root_ceph_dentry(struct inode *inode, struct dentry *dentry)
568 return ceph_ino(inode) == CEPH_INO_ROOT &&
569 strncmp(dentry->d_name.name, ".ceph", 5) == 0;
573 * Look up a single dir entry. If there is a lookup intent, inform
574 * the MDS so that it gets our 'caps wanted' value in a single op.
576 static struct dentry *ceph_lookup(struct inode *dir, struct dentry *dentry,
577 unsigned int flags)
579 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
580 struct ceph_mds_client *mdsc = fsc->mdsc;
581 struct ceph_mds_request *req;
582 int op;
583 int err;
585 dout("lookup %p dentry %p '%.*s'\n",
586 dir, dentry, dentry->d_name.len, dentry->d_name.name);
588 if (dentry->d_name.len > NAME_MAX)
589 return ERR_PTR(-ENAMETOOLONG);
591 err = ceph_init_dentry(dentry);
592 if (err < 0)
593 return ERR_PTR(err);
595 /* can we conclude ENOENT locally? */
596 if (dentry->d_inode == NULL) {
597 struct ceph_inode_info *ci = ceph_inode(dir);
598 struct ceph_dentry_info *di = ceph_dentry(dentry);
600 spin_lock(&ci->i_ceph_lock);
601 dout(" dir %p flags are %d\n", dir, ci->i_ceph_flags);
602 if (strncmp(dentry->d_name.name,
603 fsc->mount_options->snapdir_name,
604 dentry->d_name.len) &&
605 !is_root_ceph_dentry(dir, dentry) &&
606 __ceph_dir_is_complete(ci) &&
607 (__ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1))) {
608 spin_unlock(&ci->i_ceph_lock);
609 dout(" dir %p complete, -ENOENT\n", dir);
610 d_add(dentry, NULL);
611 di->lease_shared_gen = ci->i_shared_gen;
612 return NULL;
614 spin_unlock(&ci->i_ceph_lock);
617 op = ceph_snap(dir) == CEPH_SNAPDIR ?
618 CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP;
619 req = ceph_mdsc_create_request(mdsc, op, USE_ANY_MDS);
620 if (IS_ERR(req))
621 return ERR_CAST(req);
622 req->r_dentry = dget(dentry);
623 req->r_num_caps = 2;
624 /* we only need inode linkage */
625 req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
626 req->r_locked_dir = dir;
627 err = ceph_mdsc_do_request(mdsc, NULL, req);
628 err = ceph_handle_snapdir(req, dentry, err);
629 dentry = ceph_finish_lookup(req, dentry, err);
630 ceph_mdsc_put_request(req); /* will dput(dentry) */
631 dout("lookup result=%p\n", dentry);
632 return dentry;
636 * If we do a create but get no trace back from the MDS, follow up with
637 * a lookup (the VFS expects us to link up the provided dentry).
639 int ceph_handle_notrace_create(struct inode *dir, struct dentry *dentry)
641 struct dentry *result = ceph_lookup(dir, dentry, 0);
643 if (result && !IS_ERR(result)) {
645 * We created the item, then did a lookup, and found
646 * it was already linked to another inode we already
647 * had in our cache (and thus got spliced). Link our
648 * dentry to that inode, but don't hash it, just in
649 * case the VFS wants to dereference it.
651 BUG_ON(!result->d_inode);
652 d_instantiate(dentry, result->d_inode);
653 return 0;
655 return PTR_ERR(result);
658 static int ceph_mknod(struct inode *dir, struct dentry *dentry,
659 umode_t mode, dev_t rdev)
661 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
662 struct ceph_mds_client *mdsc = fsc->mdsc;
663 struct ceph_mds_request *req;
664 int err;
666 if (ceph_snap(dir) != CEPH_NOSNAP)
667 return -EROFS;
669 dout("mknod in dir %p dentry %p mode 0%ho rdev %d\n",
670 dir, dentry, mode, rdev);
671 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_MKNOD, USE_AUTH_MDS);
672 if (IS_ERR(req)) {
673 d_drop(dentry);
674 return PTR_ERR(req);
676 req->r_dentry = dget(dentry);
677 req->r_num_caps = 2;
678 req->r_locked_dir = dir;
679 req->r_args.mknod.mode = cpu_to_le32(mode);
680 req->r_args.mknod.rdev = cpu_to_le32(rdev);
681 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
682 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
683 err = ceph_mdsc_do_request(mdsc, dir, req);
684 if (!err && !req->r_reply_info.head->is_dentry)
685 err = ceph_handle_notrace_create(dir, dentry);
686 ceph_mdsc_put_request(req);
687 if (err)
688 d_drop(dentry);
689 return err;
692 static int ceph_create(struct inode *dir, struct dentry *dentry, umode_t mode,
693 bool excl)
695 return ceph_mknod(dir, dentry, mode, 0);
698 static int ceph_symlink(struct inode *dir, struct dentry *dentry,
699 const char *dest)
701 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
702 struct ceph_mds_client *mdsc = fsc->mdsc;
703 struct ceph_mds_request *req;
704 int err;
706 if (ceph_snap(dir) != CEPH_NOSNAP)
707 return -EROFS;
709 dout("symlink in dir %p dentry %p to '%s'\n", dir, dentry, dest);
710 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SYMLINK, USE_AUTH_MDS);
711 if (IS_ERR(req)) {
712 d_drop(dentry);
713 return PTR_ERR(req);
715 req->r_dentry = dget(dentry);
716 req->r_num_caps = 2;
717 req->r_path2 = kstrdup(dest, GFP_NOFS);
718 req->r_locked_dir = dir;
719 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
720 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
721 err = ceph_mdsc_do_request(mdsc, dir, req);
722 if (!err && !req->r_reply_info.head->is_dentry)
723 err = ceph_handle_notrace_create(dir, dentry);
724 ceph_mdsc_put_request(req);
725 if (err)
726 d_drop(dentry);
727 return err;
730 static int ceph_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
732 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
733 struct ceph_mds_client *mdsc = fsc->mdsc;
734 struct ceph_mds_request *req;
735 int err = -EROFS;
736 int op;
738 if (ceph_snap(dir) == CEPH_SNAPDIR) {
739 /* mkdir .snap/foo is a MKSNAP */
740 op = CEPH_MDS_OP_MKSNAP;
741 dout("mksnap dir %p snap '%.*s' dn %p\n", dir,
742 dentry->d_name.len, dentry->d_name.name, dentry);
743 } else if (ceph_snap(dir) == CEPH_NOSNAP) {
744 dout("mkdir dir %p dn %p mode 0%ho\n", dir, dentry, mode);
745 op = CEPH_MDS_OP_MKDIR;
746 } else {
747 goto out;
749 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
750 if (IS_ERR(req)) {
751 err = PTR_ERR(req);
752 goto out;
755 req->r_dentry = dget(dentry);
756 req->r_num_caps = 2;
757 req->r_locked_dir = dir;
758 req->r_args.mkdir.mode = cpu_to_le32(mode);
759 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
760 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
761 err = ceph_mdsc_do_request(mdsc, dir, req);
762 if (!err && !req->r_reply_info.head->is_dentry)
763 err = ceph_handle_notrace_create(dir, dentry);
764 ceph_mdsc_put_request(req);
765 out:
766 if (err < 0)
767 d_drop(dentry);
768 return err;
771 static int ceph_link(struct dentry *old_dentry, struct inode *dir,
772 struct dentry *dentry)
774 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
775 struct ceph_mds_client *mdsc = fsc->mdsc;
776 struct ceph_mds_request *req;
777 int err;
779 if (ceph_snap(dir) != CEPH_NOSNAP)
780 return -EROFS;
782 dout("link in dir %p old_dentry %p dentry %p\n", dir,
783 old_dentry, dentry);
784 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LINK, USE_AUTH_MDS);
785 if (IS_ERR(req)) {
786 d_drop(dentry);
787 return PTR_ERR(req);
789 req->r_dentry = dget(dentry);
790 req->r_num_caps = 2;
791 req->r_old_dentry = dget(old_dentry); /* or inode? hrm. */
792 req->r_old_dentry_dir = ceph_get_dentry_parent_inode(old_dentry);
793 req->r_locked_dir = dir;
794 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
795 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
796 /* release LINK_SHARED on source inode (mds will lock it) */
797 req->r_old_inode_drop = CEPH_CAP_LINK_SHARED;
798 err = ceph_mdsc_do_request(mdsc, dir, req);
799 if (err) {
800 d_drop(dentry);
801 } else if (!req->r_reply_info.head->is_dentry) {
802 ihold(old_dentry->d_inode);
803 d_instantiate(dentry, old_dentry->d_inode);
805 ceph_mdsc_put_request(req);
806 return err;
810 * For a soon-to-be unlinked file, drop the AUTH_RDCACHE caps. If it
811 * looks like the link count will hit 0, drop any other caps (other
812 * than PIN) we don't specifically want (due to the file still being
813 * open).
815 static int drop_caps_for_unlink(struct inode *inode)
817 struct ceph_inode_info *ci = ceph_inode(inode);
818 int drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL;
820 spin_lock(&ci->i_ceph_lock);
821 if (inode->i_nlink == 1) {
822 drop |= ~(__ceph_caps_wanted(ci) | CEPH_CAP_PIN);
823 ci->i_ceph_flags |= CEPH_I_NODELAY;
825 spin_unlock(&ci->i_ceph_lock);
826 return drop;
830 * rmdir and unlink are differ only by the metadata op code
832 static int ceph_unlink(struct inode *dir, struct dentry *dentry)
834 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
835 struct ceph_mds_client *mdsc = fsc->mdsc;
836 struct inode *inode = dentry->d_inode;
837 struct ceph_mds_request *req;
838 int err = -EROFS;
839 int op;
841 if (ceph_snap(dir) == CEPH_SNAPDIR) {
842 /* rmdir .snap/foo is RMSNAP */
843 dout("rmsnap dir %p '%.*s' dn %p\n", dir, dentry->d_name.len,
844 dentry->d_name.name, dentry);
845 op = CEPH_MDS_OP_RMSNAP;
846 } else if (ceph_snap(dir) == CEPH_NOSNAP) {
847 dout("unlink/rmdir dir %p dn %p inode %p\n",
848 dir, dentry, inode);
849 op = S_ISDIR(dentry->d_inode->i_mode) ?
850 CEPH_MDS_OP_RMDIR : CEPH_MDS_OP_UNLINK;
851 } else
852 goto out;
853 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
854 if (IS_ERR(req)) {
855 err = PTR_ERR(req);
856 goto out;
858 req->r_dentry = dget(dentry);
859 req->r_num_caps = 2;
860 req->r_locked_dir = dir;
861 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
862 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
863 req->r_inode_drop = drop_caps_for_unlink(inode);
864 err = ceph_mdsc_do_request(mdsc, dir, req);
865 if (!err && !req->r_reply_info.head->is_dentry)
866 d_delete(dentry);
867 ceph_mdsc_put_request(req);
868 out:
869 return err;
872 static int ceph_rename(struct inode *old_dir, struct dentry *old_dentry,
873 struct inode *new_dir, struct dentry *new_dentry)
875 struct ceph_fs_client *fsc = ceph_sb_to_client(old_dir->i_sb);
876 struct ceph_mds_client *mdsc = fsc->mdsc;
877 struct ceph_mds_request *req;
878 int err;
880 if (ceph_snap(old_dir) != ceph_snap(new_dir))
881 return -EXDEV;
882 if (ceph_snap(old_dir) != CEPH_NOSNAP ||
883 ceph_snap(new_dir) != CEPH_NOSNAP)
884 return -EROFS;
885 dout("rename dir %p dentry %p to dir %p dentry %p\n",
886 old_dir, old_dentry, new_dir, new_dentry);
887 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_RENAME, USE_AUTH_MDS);
888 if (IS_ERR(req))
889 return PTR_ERR(req);
890 req->r_dentry = dget(new_dentry);
891 req->r_num_caps = 2;
892 req->r_old_dentry = dget(old_dentry);
893 req->r_old_dentry_dir = ceph_get_dentry_parent_inode(old_dentry);
894 req->r_locked_dir = new_dir;
895 req->r_old_dentry_drop = CEPH_CAP_FILE_SHARED;
896 req->r_old_dentry_unless = CEPH_CAP_FILE_EXCL;
897 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
898 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
899 /* release LINK_RDCACHE on source inode (mds will lock it) */
900 req->r_old_inode_drop = CEPH_CAP_LINK_SHARED;
901 if (new_dentry->d_inode)
902 req->r_inode_drop = drop_caps_for_unlink(new_dentry->d_inode);
903 err = ceph_mdsc_do_request(mdsc, old_dir, req);
904 if (!err && !req->r_reply_info.head->is_dentry) {
906 * Normally d_move() is done by fill_trace (called by
907 * do_request, above). If there is no trace, we need
908 * to do it here.
911 /* d_move screws up d_subdirs order */
912 ceph_dir_clear_complete(new_dir);
914 d_move(old_dentry, new_dentry);
916 /* ensure target dentry is invalidated, despite
917 rehashing bug in vfs_rename_dir */
918 ceph_invalidate_dentry_lease(new_dentry);
920 ceph_mdsc_put_request(req);
921 return err;
925 * Ensure a dentry lease will no longer revalidate.
927 void ceph_invalidate_dentry_lease(struct dentry *dentry)
929 spin_lock(&dentry->d_lock);
930 dentry->d_time = jiffies;
931 ceph_dentry(dentry)->lease_shared_gen = 0;
932 spin_unlock(&dentry->d_lock);
936 * Check if dentry lease is valid. If not, delete the lease. Try to
937 * renew if the least is more than half up.
939 static int dentry_lease_is_valid(struct dentry *dentry)
941 struct ceph_dentry_info *di;
942 struct ceph_mds_session *s;
943 int valid = 0;
944 u32 gen;
945 unsigned long ttl;
946 struct ceph_mds_session *session = NULL;
947 struct inode *dir = NULL;
948 u32 seq = 0;
950 spin_lock(&dentry->d_lock);
951 di = ceph_dentry(dentry);
952 if (di->lease_session) {
953 s = di->lease_session;
954 spin_lock(&s->s_gen_ttl_lock);
955 gen = s->s_cap_gen;
956 ttl = s->s_cap_ttl;
957 spin_unlock(&s->s_gen_ttl_lock);
959 if (di->lease_gen == gen &&
960 time_before(jiffies, dentry->d_time) &&
961 time_before(jiffies, ttl)) {
962 valid = 1;
963 if (di->lease_renew_after &&
964 time_after(jiffies, di->lease_renew_after)) {
965 /* we should renew */
966 dir = dentry->d_parent->d_inode;
967 session = ceph_get_mds_session(s);
968 seq = di->lease_seq;
969 di->lease_renew_after = 0;
970 di->lease_renew_from = jiffies;
974 spin_unlock(&dentry->d_lock);
976 if (session) {
977 ceph_mdsc_lease_send_msg(session, dir, dentry,
978 CEPH_MDS_LEASE_RENEW, seq);
979 ceph_put_mds_session(session);
981 dout("dentry_lease_is_valid - dentry %p = %d\n", dentry, valid);
982 return valid;
986 * Check if directory-wide content lease/cap is valid.
988 static int dir_lease_is_valid(struct inode *dir, struct dentry *dentry)
990 struct ceph_inode_info *ci = ceph_inode(dir);
991 struct ceph_dentry_info *di = ceph_dentry(dentry);
992 int valid = 0;
994 spin_lock(&ci->i_ceph_lock);
995 if (ci->i_shared_gen == di->lease_shared_gen)
996 valid = __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1);
997 spin_unlock(&ci->i_ceph_lock);
998 dout("dir_lease_is_valid dir %p v%u dentry %p v%u = %d\n",
999 dir, (unsigned)ci->i_shared_gen, dentry,
1000 (unsigned)di->lease_shared_gen, valid);
1001 return valid;
1005 * Check if cached dentry can be trusted.
1007 static int ceph_d_revalidate(struct dentry *dentry, unsigned int flags)
1009 int valid = 0;
1010 struct inode *dir;
1012 if (flags & LOOKUP_RCU)
1013 return -ECHILD;
1015 dout("d_revalidate %p '%.*s' inode %p offset %lld\n", dentry,
1016 dentry->d_name.len, dentry->d_name.name, dentry->d_inode,
1017 ceph_dentry(dentry)->offset);
1019 dir = ceph_get_dentry_parent_inode(dentry);
1021 /* always trust cached snapped dentries, snapdir dentry */
1022 if (ceph_snap(dir) != CEPH_NOSNAP) {
1023 dout("d_revalidate %p '%.*s' inode %p is SNAPPED\n", dentry,
1024 dentry->d_name.len, dentry->d_name.name, dentry->d_inode);
1025 valid = 1;
1026 } else if (dentry->d_inode &&
1027 ceph_snap(dentry->d_inode) == CEPH_SNAPDIR) {
1028 valid = 1;
1029 } else if (dentry_lease_is_valid(dentry) ||
1030 dir_lease_is_valid(dir, dentry)) {
1031 valid = 1;
1034 dout("d_revalidate %p %s\n", dentry, valid ? "valid" : "invalid");
1035 if (valid)
1036 ceph_dentry_lru_touch(dentry);
1037 else
1038 d_drop(dentry);
1039 iput(dir);
1040 return valid;
1044 * Release our ceph_dentry_info.
1046 static void ceph_d_release(struct dentry *dentry)
1048 struct ceph_dentry_info *di = ceph_dentry(dentry);
1050 dout("d_release %p\n", dentry);
1051 ceph_dentry_lru_del(dentry);
1052 if (di->lease_session)
1053 ceph_put_mds_session(di->lease_session);
1054 kmem_cache_free(ceph_dentry_cachep, di);
1055 dentry->d_fsdata = NULL;
1058 static int ceph_snapdir_d_revalidate(struct dentry *dentry,
1059 unsigned int flags)
1062 * Eventually, we'll want to revalidate snapped metadata
1063 * too... probably...
1065 return 1;
1069 * When the VFS prunes a dentry from the cache, we need to clear the
1070 * complete flag on the parent directory.
1072 * Called under dentry->d_lock.
1074 static void ceph_d_prune(struct dentry *dentry)
1076 dout("ceph_d_prune %p\n", dentry);
1078 /* do we have a valid parent? */
1079 if (IS_ROOT(dentry))
1080 return;
1082 /* if we are not hashed, we don't affect dir's completeness */
1083 if (d_unhashed(dentry))
1084 return;
1087 * we hold d_lock, so d_parent is stable, and d_fsdata is never
1088 * cleared until d_release
1090 ceph_dir_clear_complete(dentry->d_parent->d_inode);
1094 * read() on a dir. This weird interface hack only works if mounted
1095 * with '-o dirstat'.
1097 static ssize_t ceph_read_dir(struct file *file, char __user *buf, size_t size,
1098 loff_t *ppos)
1100 struct ceph_file_info *cf = file->private_data;
1101 struct inode *inode = file_inode(file);
1102 struct ceph_inode_info *ci = ceph_inode(inode);
1103 int left;
1104 const int bufsize = 1024;
1106 if (!ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb), DIRSTAT))
1107 return -EISDIR;
1109 if (!cf->dir_info) {
1110 cf->dir_info = kmalloc(bufsize, GFP_NOFS);
1111 if (!cf->dir_info)
1112 return -ENOMEM;
1113 cf->dir_info_len =
1114 snprintf(cf->dir_info, bufsize,
1115 "entries: %20lld\n"
1116 " files: %20lld\n"
1117 " subdirs: %20lld\n"
1118 "rentries: %20lld\n"
1119 " rfiles: %20lld\n"
1120 " rsubdirs: %20lld\n"
1121 "rbytes: %20lld\n"
1122 "rctime: %10ld.%09ld\n",
1123 ci->i_files + ci->i_subdirs,
1124 ci->i_files,
1125 ci->i_subdirs,
1126 ci->i_rfiles + ci->i_rsubdirs,
1127 ci->i_rfiles,
1128 ci->i_rsubdirs,
1129 ci->i_rbytes,
1130 (long)ci->i_rctime.tv_sec,
1131 (long)ci->i_rctime.tv_nsec);
1134 if (*ppos >= cf->dir_info_len)
1135 return 0;
1136 size = min_t(unsigned, size, cf->dir_info_len-*ppos);
1137 left = copy_to_user(buf, cf->dir_info + *ppos, size);
1138 if (left == size)
1139 return -EFAULT;
1140 *ppos += (size - left);
1141 return size - left;
1145 * an fsync() on a dir will wait for any uncommitted directory
1146 * operations to commit.
1148 static int ceph_dir_fsync(struct file *file, loff_t start, loff_t end,
1149 int datasync)
1151 struct inode *inode = file_inode(file);
1152 struct ceph_inode_info *ci = ceph_inode(inode);
1153 struct list_head *head = &ci->i_unsafe_dirops;
1154 struct ceph_mds_request *req;
1155 u64 last_tid;
1156 int ret = 0;
1158 dout("dir_fsync %p\n", inode);
1159 ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
1160 if (ret)
1161 return ret;
1162 mutex_lock(&inode->i_mutex);
1164 spin_lock(&ci->i_unsafe_lock);
1165 if (list_empty(head))
1166 goto out;
1168 req = list_entry(head->prev,
1169 struct ceph_mds_request, r_unsafe_dir_item);
1170 last_tid = req->r_tid;
1172 do {
1173 ceph_mdsc_get_request(req);
1174 spin_unlock(&ci->i_unsafe_lock);
1176 dout("dir_fsync %p wait on tid %llu (until %llu)\n",
1177 inode, req->r_tid, last_tid);
1178 if (req->r_timeout) {
1179 ret = wait_for_completion_timeout(
1180 &req->r_safe_completion, req->r_timeout);
1181 if (ret > 0)
1182 ret = 0;
1183 else if (ret == 0)
1184 ret = -EIO; /* timed out */
1185 } else {
1186 wait_for_completion(&req->r_safe_completion);
1188 ceph_mdsc_put_request(req);
1190 spin_lock(&ci->i_unsafe_lock);
1191 if (ret || list_empty(head))
1192 break;
1193 req = list_entry(head->next,
1194 struct ceph_mds_request, r_unsafe_dir_item);
1195 } while (req->r_tid < last_tid);
1196 out:
1197 spin_unlock(&ci->i_unsafe_lock);
1198 mutex_unlock(&inode->i_mutex);
1200 return ret;
1204 * We maintain a private dentry LRU.
1206 * FIXME: this needs to be changed to a per-mds lru to be useful.
1208 void ceph_dentry_lru_add(struct dentry *dn)
1210 struct ceph_dentry_info *di = ceph_dentry(dn);
1211 struct ceph_mds_client *mdsc;
1213 dout("dentry_lru_add %p %p '%.*s'\n", di, dn,
1214 dn->d_name.len, dn->d_name.name);
1215 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1216 spin_lock(&mdsc->dentry_lru_lock);
1217 list_add_tail(&di->lru, &mdsc->dentry_lru);
1218 mdsc->num_dentry++;
1219 spin_unlock(&mdsc->dentry_lru_lock);
1222 void ceph_dentry_lru_touch(struct dentry *dn)
1224 struct ceph_dentry_info *di = ceph_dentry(dn);
1225 struct ceph_mds_client *mdsc;
1227 dout("dentry_lru_touch %p %p '%.*s' (offset %lld)\n", di, dn,
1228 dn->d_name.len, dn->d_name.name, di->offset);
1229 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1230 spin_lock(&mdsc->dentry_lru_lock);
1231 list_move_tail(&di->lru, &mdsc->dentry_lru);
1232 spin_unlock(&mdsc->dentry_lru_lock);
1235 void ceph_dentry_lru_del(struct dentry *dn)
1237 struct ceph_dentry_info *di = ceph_dentry(dn);
1238 struct ceph_mds_client *mdsc;
1240 dout("dentry_lru_del %p %p '%.*s'\n", di, dn,
1241 dn->d_name.len, dn->d_name.name);
1242 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1243 spin_lock(&mdsc->dentry_lru_lock);
1244 list_del_init(&di->lru);
1245 mdsc->num_dentry--;
1246 spin_unlock(&mdsc->dentry_lru_lock);
1250 * Return name hash for a given dentry. This is dependent on
1251 * the parent directory's hash function.
1253 unsigned ceph_dentry_hash(struct inode *dir, struct dentry *dn)
1255 struct ceph_inode_info *dci = ceph_inode(dir);
1257 switch (dci->i_dir_layout.dl_dir_hash) {
1258 case 0: /* for backward compat */
1259 case CEPH_STR_HASH_LINUX:
1260 return dn->d_name.hash;
1262 default:
1263 return ceph_str_hash(dci->i_dir_layout.dl_dir_hash,
1264 dn->d_name.name, dn->d_name.len);
1268 const struct file_operations ceph_dir_fops = {
1269 .read = ceph_read_dir,
1270 .iterate = ceph_readdir,
1271 .llseek = ceph_dir_llseek,
1272 .open = ceph_open,
1273 .release = ceph_release,
1274 .unlocked_ioctl = ceph_ioctl,
1275 .fsync = ceph_dir_fsync,
1278 const struct inode_operations ceph_dir_iops = {
1279 .lookup = ceph_lookup,
1280 .permission = ceph_permission,
1281 .getattr = ceph_getattr,
1282 .setattr = ceph_setattr,
1283 .setxattr = ceph_setxattr,
1284 .getxattr = ceph_getxattr,
1285 .listxattr = ceph_listxattr,
1286 .removexattr = ceph_removexattr,
1287 .mknod = ceph_mknod,
1288 .symlink = ceph_symlink,
1289 .mkdir = ceph_mkdir,
1290 .link = ceph_link,
1291 .unlink = ceph_unlink,
1292 .rmdir = ceph_unlink,
1293 .rename = ceph_rename,
1294 .create = ceph_create,
1295 .atomic_open = ceph_atomic_open,
1298 const struct dentry_operations ceph_dentry_ops = {
1299 .d_revalidate = ceph_d_revalidate,
1300 .d_release = ceph_d_release,
1301 .d_prune = ceph_d_prune,
1304 const struct dentry_operations ceph_snapdir_dentry_ops = {
1305 .d_revalidate = ceph_snapdir_d_revalidate,
1306 .d_release = ceph_d_release,
1309 const struct dentry_operations ceph_snap_dentry_ops = {
1310 .d_release = ceph_d_release,
1311 .d_prune = ceph_d_prune,