?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
gpio: rcar: Fix runtime PM imbalance on error
[linux/fpc-iii.git] / fs / ceph / dir.c
blobd594c26274305ac44d46c647545b503c787232bc
1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/ceph/ceph_debug.h>
3
4 #include <linux/spinlock.h>
5 #include <linux/namei.h>
6 #include <linux/slab.h>
7 #include <linux/sched.h>
8 #include <linux/xattr.h>
9
10 #include "super.h"
11 #include "mds_client.h"
12
13 /*
14 * Directory operations: readdir, lookup, create, link, unlink,
15 * rename, etc.
16 */
17
18 /*
19 * Ceph MDS operations are specified in terms of a base ino and
20 * relative path. Thus, the client can specify an operation on a
21 * specific inode (e.g., a getattr due to fstat(2)), or as a path
22 * relative to, say, the root directory.
23 *
24 * Normally, we limit ourselves to strict inode ops (no path component)
25 * or dentry operations (a single path component relative to an ino). The
26 * exception to this is open_root_dentry(), which will open the mount
27 * point by name.
28 */
29
30 const struct dentry_operations ceph_dentry_ops;
31
32 static bool __dentry_lease_is_valid(struct ceph_dentry_info *di);
33 static int __dir_lease_try_check(const struct dentry *dentry);
34
35 /*
36 * Initialize ceph dentry state.
37 */
38 static int ceph_d_init(struct dentry *dentry)
39 {
40 struct ceph_dentry_info *di;
41
42 di = kmem_cache_zalloc(ceph_dentry_cachep, GFP_KERNEL);
43 if (!di)
44 return -ENOMEM; /* oh well */
45
46 di->dentry = dentry;
47 di->lease_session = NULL;
48 di->time = jiffies;
49 dentry->d_fsdata = di;
50 INIT_LIST_HEAD(&di->lease_list);
51 return 0;
52 }
53
54 /*
55 * for f_pos for readdir:
56 * - hash order:
57 * (0xff << 52) | ((24 bits hash) << 28) |
58 * (the nth entry has hash collision);
59 * - frag+name order;
60 * ((frag value) << 28) | (the nth entry in frag);
61 */
62 #define OFFSET_BITS 28
63 #define OFFSET_MASK ((1 << OFFSET_BITS) - 1)
64 #define HASH_ORDER (0xffull << (OFFSET_BITS + 24))
65 loff_t ceph_make_fpos(unsigned high, unsigned off, bool hash_order)
66 {
67 loff_t fpos = ((loff_t)high << 28) | (loff_t)off;
68 if (hash_order)
69 fpos |= HASH_ORDER;
70 return fpos;
71 }
72
73 static bool is_hash_order(loff_t p)
74 {
75 return (p & HASH_ORDER) == HASH_ORDER;
76 }
77
78 static unsigned fpos_frag(loff_t p)
79 {
80 return p >> OFFSET_BITS;
81 }
82
83 static unsigned fpos_hash(loff_t p)
84 {
85 return ceph_frag_value(fpos_frag(p));
86 }
87
88 static unsigned fpos_off(loff_t p)
89 {
90 return p & OFFSET_MASK;
91 }
92
93 static int fpos_cmp(loff_t l, loff_t r)
94 {
95 int v = ceph_frag_compare(fpos_frag(l), fpos_frag(r));
96 if (v)
97 return v;
98 return (int)(fpos_off(l) - fpos_off(r));
99 }
100
101 /*
102 * make note of the last dentry we read, so we can
103 * continue at the same lexicographical point,
104 * regardless of what dir changes take place on the
105 * server.
106 */
107 static int note_last_dentry(struct ceph_dir_file_info *dfi, const char *name,
108 int len, unsigned next_offset)
109 {
110 char *buf = kmalloc(len+1, GFP_KERNEL);
111 if (!buf)
112 return -ENOMEM;
113 kfree(dfi->last_name);
114 dfi->last_name = buf;
115 memcpy(dfi->last_name, name, len);
116 dfi->last_name[len] = 0;
117 dfi->next_offset = next_offset;
118 dout("note_last_dentry '%s'\n", dfi->last_name);
119 return 0;
120 }
121
122
123 static struct dentry *
124 __dcache_find_get_entry(struct dentry *parent, u64 idx,
125 struct ceph_readdir_cache_control *cache_ctl)
126 {
127 struct inode *dir = d_inode(parent);
128 struct dentry *dentry;
129 unsigned idx_mask = (PAGE_SIZE / sizeof(struct dentry *)) - 1;
130 loff_t ptr_pos = idx * sizeof(struct dentry *);
131 pgoff_t ptr_pgoff = ptr_pos >> PAGE_SHIFT;
132
133 if (ptr_pos >= i_size_read(dir))
134 return NULL;
135
136 if (!cache_ctl->page || ptr_pgoff != page_index(cache_ctl->page)) {
137 ceph_readdir_cache_release(cache_ctl);
138 cache_ctl->page = find_lock_page(&dir->i_data, ptr_pgoff);
139 if (!cache_ctl->page) {
140 dout(" page %lu not found\n", ptr_pgoff);
141 return ERR_PTR(-EAGAIN);
142 }
143 /* reading/filling the cache are serialized by
144 i_mutex, no need to use page lock */
145 unlock_page(cache_ctl->page);
146 cache_ctl->dentries = kmap(cache_ctl->page);
147 }
148
149 cache_ctl->index = idx & idx_mask;
150
151 rcu_read_lock();
152 spin_lock(&parent->d_lock);
153 /* check i_size again here, because empty directory can be
154 * marked as complete while not holding the i_mutex. */
155 if (ceph_dir_is_complete_ordered(dir) && ptr_pos < i_size_read(dir))
156 dentry = cache_ctl->dentries[cache_ctl->index];
157 else
158 dentry = NULL;
159 spin_unlock(&parent->d_lock);
160 if (dentry && !lockref_get_not_dead(&dentry->d_lockref))
161 dentry = NULL;
162 rcu_read_unlock();
163 return dentry ? : ERR_PTR(-EAGAIN);
164 }
165
166 /*
167 * When possible, we try to satisfy a readdir by peeking at the
168 * dcache. We make this work by carefully ordering dentries on
169 * d_child when we initially get results back from the MDS, and
170 * falling back to a "normal" sync readdir if any dentries in the dir
171 * are dropped.
172 *
173 * Complete dir indicates that we have all dentries in the dir. It is
174 * defined IFF we hold CEPH_CAP_FILE_SHARED (which will be revoked by
175 * the MDS if/when the directory is modified).
176 */
177 static int __dcache_readdir(struct file *file, struct dir_context *ctx,
178 int shared_gen)
179 {
180 struct ceph_dir_file_info *dfi = file->private_data;
181 struct dentry *parent = file->f_path.dentry;
182 struct inode *dir = d_inode(parent);
183 struct dentry *dentry, *last = NULL;
184 struct ceph_dentry_info *di;
185 struct ceph_readdir_cache_control cache_ctl = {};
186 u64 idx = 0;
187 int err = 0;
188
189 dout("__dcache_readdir %p v%u at %llx\n", dir, (unsigned)shared_gen, ctx->pos);
190
191 /* search start position */
192 if (ctx->pos > 2) {
193 u64 count = div_u64(i_size_read(dir), sizeof(struct dentry *));
194 while (count > 0) {
195 u64 step = count >> 1;
196 dentry = __dcache_find_get_entry(parent, idx + step,
197 &cache_ctl);
198 if (!dentry) {
199 /* use linar search */
200 idx = 0;
201 break;
202 }
203 if (IS_ERR(dentry)) {
204 err = PTR_ERR(dentry);
205 goto out;
206 }
207 di = ceph_dentry(dentry);
208 spin_lock(&dentry->d_lock);
209 if (fpos_cmp(di->offset, ctx->pos) < 0) {
210 idx += step + 1;
211 count -= step + 1;
212 } else {
213 count = step;
214 }
215 spin_unlock(&dentry->d_lock);
216 dput(dentry);
217 }
218
219 dout("__dcache_readdir %p cache idx %llu\n", dir, idx);
220 }
221
222
223 for (;;) {
224 bool emit_dentry = false;
225 dentry = __dcache_find_get_entry(parent, idx++, &cache_ctl);
226 if (!dentry) {
227 dfi->file_info.flags |= CEPH_F_ATEND;
228 err = 0;
229 break;
230 }
231 if (IS_ERR(dentry)) {
232 err = PTR_ERR(dentry);
233 goto out;
234 }
235
236 spin_lock(&dentry->d_lock);
237 di = ceph_dentry(dentry);
238 if (d_unhashed(dentry) ||
239 d_really_is_negative(dentry) ||
240 di->lease_shared_gen != shared_gen) {
241 spin_unlock(&dentry->d_lock);
242 dput(dentry);
243 err = -EAGAIN;
244 goto out;
245 }
246 if (fpos_cmp(ctx->pos, di->offset) <= 0) {
247 __ceph_dentry_dir_lease_touch(di);
248 emit_dentry = true;
249 }
250 spin_unlock(&dentry->d_lock);
251
252 if (emit_dentry) {
253 dout(" %llx dentry %p %pd %p\n", di->offset,
254 dentry, dentry, d_inode(dentry));
255 ctx->pos = di->offset;
256 if (!dir_emit(ctx, dentry->d_name.name,
257 dentry->d_name.len,
258 ceph_translate_ino(dentry->d_sb,
259 d_inode(dentry)->i_ino),
260 d_inode(dentry)->i_mode >> 12)) {
261 dput(dentry);
262 err = 0;
263 break;
264 }
265 ctx->pos++;
266
267 if (last)
268 dput(last);
269 last = dentry;
270 } else {
271 dput(dentry);
272 }
273 }
274 out:
275 ceph_readdir_cache_release(&cache_ctl);
276 if (last) {
277 int ret;
278 di = ceph_dentry(last);
279 ret = note_last_dentry(dfi, last->d_name.name, last->d_name.len,
280 fpos_off(di->offset) + 1);
281 if (ret < 0)
282 err = ret;
283 dput(last);
284 /* last_name no longer match cache index */
285 if (dfi->readdir_cache_idx >= 0) {
286 dfi->readdir_cache_idx = -1;
287 dfi->dir_release_count = 0;
288 }
289 }
290 return err;
291 }
292
293 static bool need_send_readdir(struct ceph_dir_file_info *dfi, loff_t pos)
294 {
295 if (!dfi->last_readdir)
296 return true;
297 if (is_hash_order(pos))
298 return !ceph_frag_contains_value(dfi->frag, fpos_hash(pos));
299 else
300 return dfi->frag != fpos_frag(pos);
301 }
302
303 static int ceph_readdir(struct file *file, struct dir_context *ctx)
304 {
305 struct ceph_dir_file_info *dfi = file->private_data;
306 struct inode *inode = file_inode(file);
307 struct ceph_inode_info *ci = ceph_inode(inode);
308 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
309 struct ceph_mds_client *mdsc = fsc->mdsc;
310 int i;
311 int err;
312 unsigned frag = -1;
313 struct ceph_mds_reply_info_parsed *rinfo;
314
315 dout("readdir %p file %p pos %llx\n", inode, file, ctx->pos);
316 if (dfi->file_info.flags & CEPH_F_ATEND)
317 return 0;
318
319 /* always start with . and .. */
320 if (ctx->pos == 0) {
321 dout("readdir off 0 -> '.'\n");
322 if (!dir_emit(ctx, ".", 1,
323 ceph_translate_ino(inode->i_sb, inode->i_ino),
324 inode->i_mode >> 12))
325 return 0;
326 ctx->pos = 1;
327 }
328 if (ctx->pos == 1) {
329 ino_t ino = parent_ino(file->f_path.dentry);
330 dout("readdir off 1 -> '..'\n");
331 if (!dir_emit(ctx, "..", 2,
332 ceph_translate_ino(inode->i_sb, ino),
333 inode->i_mode >> 12))
334 return 0;
335 ctx->pos = 2;
336 }
337
338 spin_lock(&ci->i_ceph_lock);
339 /* request Fx cap. if have Fx, we don't need to release Fs cap
340 * for later create/unlink. */
341 __ceph_touch_fmode(ci, mdsc, CEPH_FILE_MODE_WR);
342 /* can we use the dcache? */
343 if (ceph_test_mount_opt(fsc, DCACHE) &&
344 !ceph_test_mount_opt(fsc, NOASYNCREADDIR) &&
345 ceph_snap(inode) != CEPH_SNAPDIR &&
346 __ceph_dir_is_complete_ordered(ci) &&
347 __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1)) {
348 int shared_gen = atomic_read(&ci->i_shared_gen);
349 spin_unlock(&ci->i_ceph_lock);
350 err = __dcache_readdir(file, ctx, shared_gen);
351 if (err != -EAGAIN)
352 return err;
353 } else {
354 spin_unlock(&ci->i_ceph_lock);
355 }
356
357 /* proceed with a normal readdir */
358 more:
359 /* do we have the correct frag content buffered? */
360 if (need_send_readdir(dfi, ctx->pos)) {
361 struct ceph_mds_request *req;
362 int op = ceph_snap(inode) == CEPH_SNAPDIR ?
363 CEPH_MDS_OP_LSSNAP : CEPH_MDS_OP_READDIR;
364
365 /* discard old result, if any */
366 if (dfi->last_readdir) {
367 ceph_mdsc_put_request(dfi->last_readdir);
368 dfi->last_readdir = NULL;
369 }
370
371 if (is_hash_order(ctx->pos)) {
372 /* fragtree isn't always accurate. choose frag
373 * based on previous reply when possible. */
374 if (frag == (unsigned)-1)
375 frag = ceph_choose_frag(ci, fpos_hash(ctx->pos),
376 NULL, NULL);
377 } else {
378 frag = fpos_frag(ctx->pos);
379 }
380
381 dout("readdir fetching %llx.%llx frag %x offset '%s'\n",
382 ceph_vinop(inode), frag, dfi->last_name);
383 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
384 if (IS_ERR(req))
385 return PTR_ERR(req);
386 err = ceph_alloc_readdir_reply_buffer(req, inode);
387 if (err) {
388 ceph_mdsc_put_request(req);
389 return err;
390 }
391 /* hints to request -> mds selection code */
392 req->r_direct_mode = USE_AUTH_MDS;
393 if (op == CEPH_MDS_OP_READDIR) {
394 req->r_direct_hash = ceph_frag_value(frag);
395 __set_bit(CEPH_MDS_R_DIRECT_IS_HASH, &req->r_req_flags);
396 req->r_inode_drop = CEPH_CAP_FILE_EXCL;
397 }
398 if (dfi->last_name) {
399 req->r_path2 = kstrdup(dfi->last_name, GFP_KERNEL);
400 if (!req->r_path2) {
401 ceph_mdsc_put_request(req);
402 return -ENOMEM;
403 }
404 } else if (is_hash_order(ctx->pos)) {
405 req->r_args.readdir.offset_hash =
406 cpu_to_le32(fpos_hash(ctx->pos));
407 }
408
409 req->r_dir_release_cnt = dfi->dir_release_count;
410 req->r_dir_ordered_cnt = dfi->dir_ordered_count;
411 req->r_readdir_cache_idx = dfi->readdir_cache_idx;
412 req->r_readdir_offset = dfi->next_offset;
413 req->r_args.readdir.frag = cpu_to_le32(frag);
414 req->r_args.readdir.flags =
415 cpu_to_le16(CEPH_READDIR_REPLY_BITFLAGS);
416
417 req->r_inode = inode;
418 ihold(inode);
419 req->r_dentry = dget(file->f_path.dentry);
420 err = ceph_mdsc_do_request(mdsc, NULL, req);
421 if (err < 0) {
422 ceph_mdsc_put_request(req);
423 return err;
424 }
425 dout("readdir got and parsed readdir result=%d on "
426 "frag %x, end=%d, complete=%d, hash_order=%d\n",
427 err, frag,
428 (int)req->r_reply_info.dir_end,
429 (int)req->r_reply_info.dir_complete,
430 (int)req->r_reply_info.hash_order);
431
432 rinfo = &req->r_reply_info;
433 if (le32_to_cpu(rinfo->dir_dir->frag) != frag) {
434 frag = le32_to_cpu(rinfo->dir_dir->frag);
435 if (!rinfo->hash_order) {
436 dfi->next_offset = req->r_readdir_offset;
437 /* adjust ctx->pos to beginning of frag */
438 ctx->pos = ceph_make_fpos(frag,
439 dfi->next_offset,
440 false);
441 }
442 }
443
444 dfi->frag = frag;
445 dfi->last_readdir = req;
446
447 if (test_bit(CEPH_MDS_R_DID_PREPOPULATE, &req->r_req_flags)) {
448 dfi->readdir_cache_idx = req->r_readdir_cache_idx;
449 if (dfi->readdir_cache_idx < 0) {
450 /* preclude from marking dir ordered */
451 dfi->dir_ordered_count = 0;
452 } else if (ceph_frag_is_leftmost(frag) &&
453 dfi->next_offset == 2) {
454 /* note dir version at start of readdir so
455 * we can tell if any dentries get dropped */
456 dfi->dir_release_count = req->r_dir_release_cnt;
457 dfi->dir_ordered_count = req->r_dir_ordered_cnt;
458 }
459 } else {
460 dout("readdir !did_prepopulate\n");
461 /* disable readdir cache */
462 dfi->readdir_cache_idx = -1;
463 /* preclude from marking dir complete */
464 dfi->dir_release_count = 0;
465 }
466
467 /* note next offset and last dentry name */
468 if (rinfo->dir_nr > 0) {
469 struct ceph_mds_reply_dir_entry *rde =
470 rinfo->dir_entries + (rinfo->dir_nr-1);
471 unsigned next_offset = req->r_reply_info.dir_end ?
472 2 : (fpos_off(rde->offset) + 1);
473 err = note_last_dentry(dfi, rde->name, rde->name_len,
474 next_offset);
475 if (err)
476 return err;
477 } else if (req->r_reply_info.dir_end) {
478 dfi->next_offset = 2;
479 /* keep last name */
480 }
481 }
482
483 rinfo = &dfi->last_readdir->r_reply_info;
484 dout("readdir frag %x num %d pos %llx chunk first %llx\n",
485 dfi->frag, rinfo->dir_nr, ctx->pos,
486 rinfo->dir_nr ? rinfo->dir_entries[0].offset : 0LL);
487
488 i = 0;
489 /* search start position */
490 if (rinfo->dir_nr > 0) {
491 int step, nr = rinfo->dir_nr;
492 while (nr > 0) {
493 step = nr >> 1;
494 if (rinfo->dir_entries[i + step].offset < ctx->pos) {
495 i += step + 1;
496 nr -= step + 1;
497 } else {
498 nr = step;
499 }
500 }
501 }
502 for (; i < rinfo->dir_nr; i++) {
503 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
504 struct ceph_vino vino;
505 ino_t ino;
506 u32 ftype;
507
508 BUG_ON(rde->offset < ctx->pos);
509
510 ctx->pos = rde->offset;
511 dout("readdir (%d/%d) -> %llx '%.*s' %p\n",
512 i, rinfo->dir_nr, ctx->pos,
513 rde->name_len, rde->name, &rde->inode.in);
514
515 BUG_ON(!rde->inode.in);
516 ftype = le32_to_cpu(rde->inode.in->mode) >> 12;
517 vino.ino = le64_to_cpu(rde->inode.in->ino);
518 vino.snap = le64_to_cpu(rde->inode.in->snapid);
519 ino = ceph_vino_to_ino(vino);
520
521 if (!dir_emit(ctx, rde->name, rde->name_len,
522 ceph_translate_ino(inode->i_sb, ino), ftype)) {
523 dout("filldir stopping us...\n");
524 return 0;
525 }
526 ctx->pos++;
527 }
528
529 ceph_mdsc_put_request(dfi->last_readdir);
530 dfi->last_readdir = NULL;
531
532 if (dfi->next_offset > 2) {
533 frag = dfi->frag;
534 goto more;
535 }
536
537 /* more frags? */
538 if (!ceph_frag_is_rightmost(dfi->frag)) {
539 frag = ceph_frag_next(dfi->frag);
540 if (is_hash_order(ctx->pos)) {
541 loff_t new_pos = ceph_make_fpos(ceph_frag_value(frag),
542 dfi->next_offset, true);
543 if (new_pos > ctx->pos)
544 ctx->pos = new_pos;
545 /* keep last_name */
546 } else {
547 ctx->pos = ceph_make_fpos(frag, dfi->next_offset,
548 false);
549 kfree(dfi->last_name);
550 dfi->last_name = NULL;
551 }
552 dout("readdir next frag is %x\n", frag);
553 goto more;
554 }
555 dfi->file_info.flags |= CEPH_F_ATEND;
556
557 /*
558 * if dir_release_count still matches the dir, no dentries
559 * were released during the whole readdir, and we should have
560 * the complete dir contents in our cache.
561 */
562 if (atomic64_read(&ci->i_release_count) ==
563 dfi->dir_release_count) {
564 spin_lock(&ci->i_ceph_lock);
565 if (dfi->dir_ordered_count ==
566 atomic64_read(&ci->i_ordered_count)) {
567 dout(" marking %p complete and ordered\n", inode);
568 /* use i_size to track number of entries in
569 * readdir cache */
570 BUG_ON(dfi->readdir_cache_idx < 0);
571 i_size_write(inode, dfi->readdir_cache_idx *
572 sizeof(struct dentry*));
573 } else {
574 dout(" marking %p complete\n", inode);
575 }
576 __ceph_dir_set_complete(ci, dfi->dir_release_count,
577 dfi->dir_ordered_count);
578 spin_unlock(&ci->i_ceph_lock);
579 }
580
581 dout("readdir %p file %p done.\n", inode, file);
582 return 0;
583 }
584
585 static void reset_readdir(struct ceph_dir_file_info *dfi)
586 {
587 if (dfi->last_readdir) {
588 ceph_mdsc_put_request(dfi->last_readdir);
589 dfi->last_readdir = NULL;
590 }
591 kfree(dfi->last_name);
592 dfi->last_name = NULL;
593 dfi->dir_release_count = 0;
594 dfi->readdir_cache_idx = -1;
595 dfi->next_offset = 2; /* compensate for . and .. */
596 dfi->file_info.flags &= ~CEPH_F_ATEND;
597 }
598
599 /*
600 * discard buffered readdir content on seekdir(0), or seek to new frag,
601 * or seek prior to current chunk
602 */
603 static bool need_reset_readdir(struct ceph_dir_file_info *dfi, loff_t new_pos)
604 {
605 struct ceph_mds_reply_info_parsed *rinfo;
606 loff_t chunk_offset;
607 if (new_pos == 0)
608 return true;
609 if (is_hash_order(new_pos)) {
610 /* no need to reset last_name for a forward seek when
611 * dentries are sotred in hash order */
612 } else if (dfi->frag != fpos_frag(new_pos)) {
613 return true;
614 }
615 rinfo = dfi->last_readdir ? &dfi->last_readdir->r_reply_info : NULL;
616 if (!rinfo || !rinfo->dir_nr)
617 return true;
618 chunk_offset = rinfo->dir_entries[0].offset;
619 return new_pos < chunk_offset ||
620 is_hash_order(new_pos) != is_hash_order(chunk_offset);
621 }
622
623 static loff_t ceph_dir_llseek(struct file *file, loff_t offset, int whence)
624 {
625 struct ceph_dir_file_info *dfi = file->private_data;
626 struct inode *inode = file->f_mapping->host;
627 loff_t retval;
628
629 inode_lock(inode);
630 retval = -EINVAL;
631 switch (whence) {
632 case SEEK_CUR:
633 offset += file->f_pos;
634 case SEEK_SET:
635 break;
636 case SEEK_END:
637 retval = -EOPNOTSUPP;
638 default:
639 goto out;
640 }
641
642 if (offset >= 0) {
643 if (need_reset_readdir(dfi, offset)) {
644 dout("dir_llseek dropping %p content\n", file);
645 reset_readdir(dfi);
646 } else if (is_hash_order(offset) && offset > file->f_pos) {
647 /* for hash offset, we don't know if a forward seek
648 * is within same frag */
649 dfi->dir_release_count = 0;
650 dfi->readdir_cache_idx = -1;
651 }
652
653 if (offset != file->f_pos) {
654 file->f_pos = offset;
655 file->f_version = 0;
656 dfi->file_info.flags &= ~CEPH_F_ATEND;
657 }
658 retval = offset;
659 }
660 out:
661 inode_unlock(inode);
662 return retval;
663 }
664
665 /*
666 * Handle lookups for the hidden .snap directory.
667 */
668 int ceph_handle_snapdir(struct ceph_mds_request *req,
669 struct dentry *dentry, int err)
670 {
671 struct ceph_fs_client *fsc = ceph_sb_to_client(dentry->d_sb);
672 struct inode *parent = d_inode(dentry->d_parent); /* we hold i_mutex */
673
674 /* .snap dir? */
675 if (err == -ENOENT &&
676 ceph_snap(parent) == CEPH_NOSNAP &&
677 strcmp(dentry->d_name.name,
678 fsc->mount_options->snapdir_name) == 0) {
679 struct inode *inode = ceph_get_snapdir(parent);
680 dout("ENOENT on snapdir %p '%pd', linking to snapdir %p\n",
681 dentry, dentry, inode);
682 BUG_ON(!d_unhashed(dentry));
683 d_add(dentry, inode);
684 err = 0;
685 }
686 return err;
687 }
688
689 /*
690 * Figure out final result of a lookup/open request.
691 *
692 * Mainly, make sure we return the final req->r_dentry (if it already
693 * existed) in place of the original VFS-provided dentry when they
694 * differ.
695 *
696 * Gracefully handle the case where the MDS replies with -ENOENT and
697 * no trace (which it may do, at its discretion, e.g., if it doesn't
698 * care to issue a lease on the negative dentry).
699 */
700 struct dentry *ceph_finish_lookup(struct ceph_mds_request *req,
701 struct dentry *dentry, int err)
702 {
703 if (err == -ENOENT) {
704 /* no trace? */
705 err = 0;
706 if (!req->r_reply_info.head->is_dentry) {
707 dout("ENOENT and no trace, dentry %p inode %p\n",
708 dentry, d_inode(dentry));
709 if (d_really_is_positive(dentry)) {
710 d_drop(dentry);
711 err = -ENOENT;
712 } else {
713 d_add(dentry, NULL);
714 }
715 }
716 }
717 if (err)
718 dentry = ERR_PTR(err);
719 else if (dentry != req->r_dentry)
720 dentry = dget(req->r_dentry); /* we got spliced */
721 else
722 dentry = NULL;
723 return dentry;
724 }
725
726 static bool is_root_ceph_dentry(struct inode *inode, struct dentry *dentry)
727 {
728 return ceph_ino(inode) == CEPH_INO_ROOT &&
729 strncmp(dentry->d_name.name, ".ceph", 5) == 0;
730 }
731
732 /*
733 * Look up a single dir entry. If there is a lookup intent, inform
734 * the MDS so that it gets our 'caps wanted' value in a single op.
735 */
736 static struct dentry *ceph_lookup(struct inode *dir, struct dentry *dentry,
737 unsigned int flags)
738 {
739 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
740 struct ceph_mds_client *mdsc = fsc->mdsc;
741 struct ceph_mds_request *req;
742 int op;
743 int mask;
744 int err;
745
746 dout("lookup %p dentry %p '%pd'\n",
747 dir, dentry, dentry);
748
749 if (dentry->d_name.len > NAME_MAX)
750 return ERR_PTR(-ENAMETOOLONG);
751
752 /* can we conclude ENOENT locally? */
753 if (d_really_is_negative(dentry)) {
754 struct ceph_inode_info *ci = ceph_inode(dir);
755 struct ceph_dentry_info *di = ceph_dentry(dentry);
756
757 spin_lock(&ci->i_ceph_lock);
758 dout(" dir %p flags are 0x%lx\n", dir, ci->i_ceph_flags);
759 if (strncmp(dentry->d_name.name,
760 fsc->mount_options->snapdir_name,
761 dentry->d_name.len) &&
762 !is_root_ceph_dentry(dir, dentry) &&
763 ceph_test_mount_opt(fsc, DCACHE) &&
764 __ceph_dir_is_complete(ci) &&
765 (__ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1))) {
766 __ceph_touch_fmode(ci, mdsc, CEPH_FILE_MODE_RD);
767 spin_unlock(&ci->i_ceph_lock);
768 dout(" dir %p complete, -ENOENT\n", dir);
769 d_add(dentry, NULL);
770 di->lease_shared_gen = atomic_read(&ci->i_shared_gen);
771 return NULL;
772 }
773 spin_unlock(&ci->i_ceph_lock);
774 }
775
776 op = ceph_snap(dir) == CEPH_SNAPDIR ?
777 CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP;
778 req = ceph_mdsc_create_request(mdsc, op, USE_ANY_MDS);
779 if (IS_ERR(req))
780 return ERR_CAST(req);
781 req->r_dentry = dget(dentry);
782 req->r_num_caps = 2;
783
784 mask = CEPH_STAT_CAP_INODE | CEPH_CAP_AUTH_SHARED;
785 if (ceph_security_xattr_wanted(dir))
786 mask |= CEPH_CAP_XATTR_SHARED;
787 req->r_args.getattr.mask = cpu_to_le32(mask);
788
789 req->r_parent = dir;
790 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
791 err = ceph_mdsc_do_request(mdsc, NULL, req);
792 err = ceph_handle_snapdir(req, dentry, err);
793 dentry = ceph_finish_lookup(req, dentry, err);
794 ceph_mdsc_put_request(req); /* will dput(dentry) */
795 dout("lookup result=%p\n", dentry);
796 return dentry;
797 }
798
799 /*
800 * If we do a create but get no trace back from the MDS, follow up with
801 * a lookup (the VFS expects us to link up the provided dentry).
802 */
803 int ceph_handle_notrace_create(struct inode *dir, struct dentry *dentry)
804 {
805 struct dentry *result = ceph_lookup(dir, dentry, 0);
806
807 if (result && !IS_ERR(result)) {
808 /*
809 * We created the item, then did a lookup, and found
810 * it was already linked to another inode we already
811 * had in our cache (and thus got spliced). To not
812 * confuse VFS (especially when inode is a directory),
813 * we don't link our dentry to that inode, return an
814 * error instead.
815 *
816 * This event should be rare and it happens only when
817 * we talk to old MDS. Recent MDS does not send traceless
818 * reply for request that creates new inode.
819 */
820 d_drop(result);
821 return -ESTALE;
822 }
823 return PTR_ERR(result);
824 }
825
826 static int ceph_mknod(struct inode *dir, struct dentry *dentry,
827 umode_t mode, dev_t rdev)
828 {
829 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
830 struct ceph_mds_client *mdsc = fsc->mdsc;
831 struct ceph_mds_request *req;
832 struct ceph_acl_sec_ctx as_ctx = {};
833 int err;
834
835 if (ceph_snap(dir) != CEPH_NOSNAP)
836 return -EROFS;
837
838 if (ceph_quota_is_max_files_exceeded(dir)) {
839 err = -EDQUOT;
840 goto out;
841 }
842
843 err = ceph_pre_init_acls(dir, &mode, &as_ctx);
844 if (err < 0)
845 goto out;
846 err = ceph_security_init_secctx(dentry, mode, &as_ctx);
847 if (err < 0)
848 goto out;
849
850 dout("mknod in dir %p dentry %p mode 0%ho rdev %d\n",
851 dir, dentry, mode, rdev);
852 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_MKNOD, USE_AUTH_MDS);
853 if (IS_ERR(req)) {
854 err = PTR_ERR(req);
855 goto out;
856 }
857 req->r_dentry = dget(dentry);
858 req->r_num_caps = 2;
859 req->r_parent = dir;
860 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
861 req->r_args.mknod.mode = cpu_to_le32(mode);
862 req->r_args.mknod.rdev = cpu_to_le32(rdev);
863 req->r_dentry_drop = CEPH_CAP_FILE_SHARED | CEPH_CAP_AUTH_EXCL;
864 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
865 if (as_ctx.pagelist) {
866 req->r_pagelist = as_ctx.pagelist;
867 as_ctx.pagelist = NULL;
868 }
869 err = ceph_mdsc_do_request(mdsc, dir, req);
870 if (!err && !req->r_reply_info.head->is_dentry)
871 err = ceph_handle_notrace_create(dir, dentry);
872 ceph_mdsc_put_request(req);
873 out:
874 if (!err)
875 ceph_init_inode_acls(d_inode(dentry), &as_ctx);
876 else
877 d_drop(dentry);
878 ceph_release_acl_sec_ctx(&as_ctx);
879 return err;
880 }
881
882 static int ceph_create(struct inode *dir, struct dentry *dentry, umode_t mode,
883 bool excl)
884 {
885 return ceph_mknod(dir, dentry, mode, 0);
886 }
887
888 static int ceph_symlink(struct inode *dir, struct dentry *dentry,
889 const char *dest)
890 {
891 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
892 struct ceph_mds_client *mdsc = fsc->mdsc;
893 struct ceph_mds_request *req;
894 struct ceph_acl_sec_ctx as_ctx = {};
895 int err;
896
897 if (ceph_snap(dir) != CEPH_NOSNAP)
898 return -EROFS;
899
900 if (ceph_quota_is_max_files_exceeded(dir)) {
901 err = -EDQUOT;
902 goto out;
903 }
904
905 err = ceph_security_init_secctx(dentry, S_IFLNK | 0777, &as_ctx);
906 if (err < 0)
907 goto out;
908
909 dout("symlink in dir %p dentry %p to '%s'\n", dir, dentry, dest);
910 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SYMLINK, USE_AUTH_MDS);
911 if (IS_ERR(req)) {
912 err = PTR_ERR(req);
913 goto out;
914 }
915 req->r_path2 = kstrdup(dest, GFP_KERNEL);
916 if (!req->r_path2) {
917 err = -ENOMEM;
918 ceph_mdsc_put_request(req);
919 goto out;
920 }
921 req->r_parent = dir;
922 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
923 req->r_dentry = dget(dentry);
924 req->r_num_caps = 2;
925 req->r_dentry_drop = CEPH_CAP_FILE_SHARED | CEPH_CAP_AUTH_EXCL;
926 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
927 err = ceph_mdsc_do_request(mdsc, dir, req);
928 if (!err && !req->r_reply_info.head->is_dentry)
929 err = ceph_handle_notrace_create(dir, dentry);
930 ceph_mdsc_put_request(req);
931 out:
932 if (err)
933 d_drop(dentry);
934 ceph_release_acl_sec_ctx(&as_ctx);
935 return err;
936 }
937
938 static int ceph_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
939 {
940 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
941 struct ceph_mds_client *mdsc = fsc->mdsc;
942 struct ceph_mds_request *req;
943 struct ceph_acl_sec_ctx as_ctx = {};
944 int err = -EROFS;
945 int op;
946
947 if (ceph_snap(dir) == CEPH_SNAPDIR) {
948 /* mkdir .snap/foo is a MKSNAP */
949 op = CEPH_MDS_OP_MKSNAP;
950 dout("mksnap dir %p snap '%pd' dn %p\n", dir,
951 dentry, dentry);
952 } else if (ceph_snap(dir) == CEPH_NOSNAP) {
953 dout("mkdir dir %p dn %p mode 0%ho\n", dir, dentry, mode);
954 op = CEPH_MDS_OP_MKDIR;
955 } else {
956 goto out;
957 }
958
959 if (op == CEPH_MDS_OP_MKDIR &&
960 ceph_quota_is_max_files_exceeded(dir)) {
961 err = -EDQUOT;
962 goto out;
963 }
964
965 mode |= S_IFDIR;
966 err = ceph_pre_init_acls(dir, &mode, &as_ctx);
967 if (err < 0)
968 goto out;
969 err = ceph_security_init_secctx(dentry, mode, &as_ctx);
970 if (err < 0)
971 goto out;
972
973 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
974 if (IS_ERR(req)) {
975 err = PTR_ERR(req);
976 goto out;
977 }
978
979 req->r_dentry = dget(dentry);
980 req->r_num_caps = 2;
981 req->r_parent = dir;
982 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
983 req->r_args.mkdir.mode = cpu_to_le32(mode);
984 req->r_dentry_drop = CEPH_CAP_FILE_SHARED | CEPH_CAP_AUTH_EXCL;
985 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
986 if (as_ctx.pagelist) {
987 req->r_pagelist = as_ctx.pagelist;
988 as_ctx.pagelist = NULL;
989 }
990 err = ceph_mdsc_do_request(mdsc, dir, req);
991 if (!err &&
992 !req->r_reply_info.head->is_target &&
993 !req->r_reply_info.head->is_dentry)
994 err = ceph_handle_notrace_create(dir, dentry);
995 ceph_mdsc_put_request(req);
996 out:
997 if (!err)
998 ceph_init_inode_acls(d_inode(dentry), &as_ctx);
999 else
1000 d_drop(dentry);
1001 ceph_release_acl_sec_ctx(&as_ctx);
1002 return err;
1003 }
1004
1005 static int ceph_link(struct dentry *old_dentry, struct inode *dir,
1006 struct dentry *dentry)
1007 {
1008 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
1009 struct ceph_mds_client *mdsc = fsc->mdsc;
1010 struct ceph_mds_request *req;
1011 int err;
1012
1013 if (ceph_snap(dir) != CEPH_NOSNAP)
1014 return -EROFS;
1015
1016 dout("link in dir %p old_dentry %p dentry %p\n", dir,
1017 old_dentry, dentry);
1018 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LINK, USE_AUTH_MDS);
1019 if (IS_ERR(req)) {
1020 d_drop(dentry);
1021 return PTR_ERR(req);
1022 }
1023 req->r_dentry = dget(dentry);
1024 req->r_num_caps = 2;
1025 req->r_old_dentry = dget(old_dentry);
1026 req->r_parent = dir;
1027 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
1028 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
1029 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
1030 /* release LINK_SHARED on source inode (mds will lock it) */
1031 req->r_old_inode_drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL;
1032 err = ceph_mdsc_do_request(mdsc, dir, req);
1033 if (err) {
1034 d_drop(dentry);
1035 } else if (!req->r_reply_info.head->is_dentry) {
1036 ihold(d_inode(old_dentry));
1037 d_instantiate(dentry, d_inode(old_dentry));
1038 }
1039 ceph_mdsc_put_request(req);
1040 return err;
1041 }
1042
1043 static void ceph_async_unlink_cb(struct ceph_mds_client *mdsc,
1044 struct ceph_mds_request *req)
1045 {
1046 int result = req->r_err ? req->r_err :
1047 le32_to_cpu(req->r_reply_info.head->result);
1048
1049 if (result == -EJUKEBOX)
1050 goto out;
1051
1052 /* If op failed, mark everyone involved for errors */
1053 if (result) {
1054 int pathlen;
1055 u64 base;
1056 char *path = ceph_mdsc_build_path(req->r_dentry, &pathlen,
1057 &base, 0);
1058
1059 /* mark error on parent + clear complete */
1060 mapping_set_error(req->r_parent->i_mapping, result);
1061 ceph_dir_clear_complete(req->r_parent);
1062
1063 /* drop the dentry -- we don't know its status */
1064 if (!d_unhashed(req->r_dentry))
1065 d_drop(req->r_dentry);
1066
1067 /* mark inode itself for an error (since metadata is bogus) */
1068 mapping_set_error(req->r_old_inode->i_mapping, result);
1069
1070 pr_warn("ceph: async unlink failure path=(%llx)%s result=%d!\n",
1071 base, IS_ERR(path) ? "<<bad>>" : path, result);
1072 ceph_mdsc_free_path(path, pathlen);
1073 }
1074 out:
1075 iput(req->r_old_inode);
1076 ceph_mdsc_release_dir_caps(req);
1077 }
1078
1079 static int get_caps_for_async_unlink(struct inode *dir, struct dentry *dentry)
1080 {
1081 struct ceph_inode_info *ci = ceph_inode(dir);
1082 struct ceph_dentry_info *di;
1083 int got = 0, want = CEPH_CAP_FILE_EXCL | CEPH_CAP_DIR_UNLINK;
1084
1085 spin_lock(&ci->i_ceph_lock);
1086 if ((__ceph_caps_issued(ci, NULL) & want) == want) {
1087 ceph_take_cap_refs(ci, want, false);
1088 got = want;
1089 }
1090 spin_unlock(&ci->i_ceph_lock);
1091
1092 /* If we didn't get anything, return 0 */
1093 if (!got)
1094 return 0;
1095
1096 spin_lock(&dentry->d_lock);
1097 di = ceph_dentry(dentry);
1098 /*
1099 * - We are holding Fx, which implies Fs caps.
1100 * - Only support async unlink for primary linkage
1101 */
1102 if (atomic_read(&ci->i_shared_gen) != di->lease_shared_gen ||
1103 !(di->flags & CEPH_DENTRY_PRIMARY_LINK))
1104 want = 0;
1105 spin_unlock(&dentry->d_lock);
1106
1107 /* Do we still want what we've got? */
1108 if (want == got)
1109 return got;
1110
1111 ceph_put_cap_refs(ci, got);
1112 return 0;
1113 }
1114
1115 /*
1116 * rmdir and unlink are differ only by the metadata op code
1117 */
1118 static int ceph_unlink(struct inode *dir, struct dentry *dentry)
1119 {
1120 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
1121 struct ceph_mds_client *mdsc = fsc->mdsc;
1122 struct inode *inode = d_inode(dentry);
1123 struct ceph_mds_request *req;
1124 bool try_async = ceph_test_mount_opt(fsc, ASYNC_DIROPS);
1125 int err = -EROFS;
1126 int op;
1127
1128 if (ceph_snap(dir) == CEPH_SNAPDIR) {
1129 /* rmdir .snap/foo is RMSNAP */
1130 dout("rmsnap dir %p '%pd' dn %p\n", dir, dentry, dentry);
1131 op = CEPH_MDS_OP_RMSNAP;
1132 } else if (ceph_snap(dir) == CEPH_NOSNAP) {
1133 dout("unlink/rmdir dir %p dn %p inode %p\n",
1134 dir, dentry, inode);
1135 op = d_is_dir(dentry) ?
1136 CEPH_MDS_OP_RMDIR : CEPH_MDS_OP_UNLINK;
1137 } else
1138 goto out;
1139 retry:
1140 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
1141 if (IS_ERR(req)) {
1142 err = PTR_ERR(req);
1143 goto out;
1144 }
1145 req->r_dentry = dget(dentry);
1146 req->r_num_caps = 2;
1147 req->r_parent = dir;
1148 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
1149 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
1150 req->r_inode_drop = ceph_drop_caps_for_unlink(inode);
1151
1152 if (try_async && op == CEPH_MDS_OP_UNLINK &&
1153 (req->r_dir_caps = get_caps_for_async_unlink(dir, dentry))) {
1154 dout("async unlink on %lu/%.*s caps=%s", dir->i_ino,
1155 dentry->d_name.len, dentry->d_name.name,
1156 ceph_cap_string(req->r_dir_caps));
1157 set_bit(CEPH_MDS_R_ASYNC, &req->r_req_flags);
1158 req->r_callback = ceph_async_unlink_cb;
1159 req->r_old_inode = d_inode(dentry);
1160 ihold(req->r_old_inode);
1161 err = ceph_mdsc_submit_request(mdsc, dir, req);
1162 if (!err) {
1163 /*
1164 * We have enough caps, so we assume that the unlink
1165 * will succeed. Fix up the target inode and dcache.
1166 */
1167 drop_nlink(inode);
1168 d_delete(dentry);
1169 } else if (err == -EJUKEBOX) {
1170 try_async = false;
1171 ceph_mdsc_put_request(req);
1172 goto retry;
1173 }
1174 } else {
1175 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
1176 err = ceph_mdsc_do_request(mdsc, dir, req);
1177 if (!err && !req->r_reply_info.head->is_dentry)
1178 d_delete(dentry);
1179 }
1180
1181 ceph_mdsc_put_request(req);
1182 out:
1183 return err;
1184 }
1185
1186 static int ceph_rename(struct inode *old_dir, struct dentry *old_dentry,
1187 struct inode *new_dir, struct dentry *new_dentry,
1188 unsigned int flags)
1189 {
1190 struct ceph_fs_client *fsc = ceph_sb_to_client(old_dir->i_sb);
1191 struct ceph_mds_client *mdsc = fsc->mdsc;
1192 struct ceph_mds_request *req;
1193 int op = CEPH_MDS_OP_RENAME;
1194 int err;
1195
1196 if (flags)
1197 return -EINVAL;
1198
1199 if (ceph_snap(old_dir) != ceph_snap(new_dir))
1200 return -EXDEV;
1201 if (ceph_snap(old_dir) != CEPH_NOSNAP) {
1202 if (old_dir == new_dir && ceph_snap(old_dir) == CEPH_SNAPDIR)
1203 op = CEPH_MDS_OP_RENAMESNAP;
1204 else
1205 return -EROFS;
1206 }
1207 /* don't allow cross-quota renames */
1208 if ((old_dir != new_dir) &&
1209 (!ceph_quota_is_same_realm(old_dir, new_dir)))
1210 return -EXDEV;
1211
1212 dout("rename dir %p dentry %p to dir %p dentry %p\n",
1213 old_dir, old_dentry, new_dir, new_dentry);
1214 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
1215 if (IS_ERR(req))
1216 return PTR_ERR(req);
1217 ihold(old_dir);
1218 req->r_dentry = dget(new_dentry);
1219 req->r_num_caps = 2;
1220 req->r_old_dentry = dget(old_dentry);
1221 req->r_old_dentry_dir = old_dir;
1222 req->r_parent = new_dir;
1223 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
1224 req->r_old_dentry_drop = CEPH_CAP_FILE_SHARED;
1225 req->r_old_dentry_unless = CEPH_CAP_FILE_EXCL;
1226 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
1227 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
1228 /* release LINK_RDCACHE on source inode (mds will lock it) */
1229 req->r_old_inode_drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL;
1230 if (d_really_is_positive(new_dentry)) {
1231 req->r_inode_drop =
1232 ceph_drop_caps_for_unlink(d_inode(new_dentry));
1233 }
1234 err = ceph_mdsc_do_request(mdsc, old_dir, req);
1235 if (!err && !req->r_reply_info.head->is_dentry) {
1236 /*
1237 * Normally d_move() is done by fill_trace (called by
1238 * do_request, above). If there is no trace, we need
1239 * to do it here.
1240 */
1241 d_move(old_dentry, new_dentry);
1242 }
1243 ceph_mdsc_put_request(req);
1244 return err;
1245 }
1246
1247 /*
1248 * Move dentry to tail of mdsc->dentry_leases list when lease is updated.
1249 * Leases at front of the list will expire first. (Assume all leases have
1250 * similar duration)
1251 *
1252 * Called under dentry->d_lock.
1253 */
1254 void __ceph_dentry_lease_touch(struct ceph_dentry_info *di)
1255 {
1256 struct dentry *dn = di->dentry;
1257 struct ceph_mds_client *mdsc;
1258
1259 dout("dentry_lease_touch %p %p '%pd'\n", di, dn, dn);
1260
1261 di->flags |= CEPH_DENTRY_LEASE_LIST;
1262 if (di->flags & CEPH_DENTRY_SHRINK_LIST) {
1263 di->flags |= CEPH_DENTRY_REFERENCED;
1264 return;
1265 }
1266
1267 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1268 spin_lock(&mdsc->dentry_list_lock);
1269 list_move_tail(&di->lease_list, &mdsc->dentry_leases);
1270 spin_unlock(&mdsc->dentry_list_lock);
1271 }
1272
1273 static void __dentry_dir_lease_touch(struct ceph_mds_client* mdsc,
1274 struct ceph_dentry_info *di)
1275 {
1276 di->flags &= ~(CEPH_DENTRY_LEASE_LIST | CEPH_DENTRY_REFERENCED);
1277 di->lease_gen = 0;
1278 di->time = jiffies;
1279 list_move_tail(&di->lease_list, &mdsc->dentry_dir_leases);
1280 }
1281
1282 /*
1283 * When dir lease is used, add dentry to tail of mdsc->dentry_dir_leases
1284 * list if it's not in the list, otherwise set 'referenced' flag.
1285 *
1286 * Called under dentry->d_lock.
1287 */
1288 void __ceph_dentry_dir_lease_touch(struct ceph_dentry_info *di)
1289 {
1290 struct dentry *dn = di->dentry;
1291 struct ceph_mds_client *mdsc;
1292
1293 dout("dentry_dir_lease_touch %p %p '%pd' (offset 0x%llx)\n",
1294 di, dn, dn, di->offset);
1295
1296 if (!list_empty(&di->lease_list)) {
1297 if (di->flags & CEPH_DENTRY_LEASE_LIST) {
1298 /* don't remove dentry from dentry lease list
1299 * if its lease is valid */
1300 if (__dentry_lease_is_valid(di))
1301 return;
1302 } else {
1303 di->flags |= CEPH_DENTRY_REFERENCED;
1304 return;
1305 }
1306 }
1307
1308 if (di->flags & CEPH_DENTRY_SHRINK_LIST) {
1309 di->flags |= CEPH_DENTRY_REFERENCED;
1310 di->flags &= ~CEPH_DENTRY_LEASE_LIST;
1311 return;
1312 }
1313
1314 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1315 spin_lock(&mdsc->dentry_list_lock);
1316 __dentry_dir_lease_touch(mdsc, di),
1317 spin_unlock(&mdsc->dentry_list_lock);
1318 }
1319
1320 static void __dentry_lease_unlist(struct ceph_dentry_info *di)
1321 {
1322 struct ceph_mds_client *mdsc;
1323 if (di->flags & CEPH_DENTRY_SHRINK_LIST)
1324 return;
1325 if (list_empty(&di->lease_list))
1326 return;
1327
1328 mdsc = ceph_sb_to_client(di->dentry->d_sb)->mdsc;
1329 spin_lock(&mdsc->dentry_list_lock);
1330 list_del_init(&di->lease_list);
1331 spin_unlock(&mdsc->dentry_list_lock);
1332 }
1333
1334 enum {
1335 KEEP = 0,
1336 DELETE = 1,
1337 TOUCH = 2,
1338 STOP = 4,
1339 };
1340
1341 struct ceph_lease_walk_control {
1342 bool dir_lease;
1343 bool expire_dir_lease;
1344 unsigned long nr_to_scan;
1345 unsigned long dir_lease_ttl;
1346 };
1347
1348 static unsigned long
1349 __dentry_leases_walk(struct ceph_mds_client *mdsc,
1350 struct ceph_lease_walk_control *lwc,
1351 int (*check)(struct dentry*, void*))
1352 {
1353 struct ceph_dentry_info *di, *tmp;
1354 struct dentry *dentry, *last = NULL;
1355 struct list_head* list;
1356 LIST_HEAD(dispose);
1357 unsigned long freed = 0;
1358 int ret = 0;
1359
1360 list = lwc->dir_lease ? &mdsc->dentry_dir_leases : &mdsc->dentry_leases;
1361 spin_lock(&mdsc->dentry_list_lock);
1362 list_for_each_entry_safe(di, tmp, list, lease_list) {
1363 if (!lwc->nr_to_scan)
1364 break;
1365 --lwc->nr_to_scan;
1366
1367 dentry = di->dentry;
1368 if (last == dentry)
1369 break;
1370
1371 if (!spin_trylock(&dentry->d_lock))
1372 continue;
1373
1374 if (__lockref_is_dead(&dentry->d_lockref)) {
1375 list_del_init(&di->lease_list);
1376 goto next;
1377 }
1378
1379 ret = check(dentry, lwc);
1380 if (ret & TOUCH) {
1381 /* move it into tail of dir lease list */
1382 __dentry_dir_lease_touch(mdsc, di);
1383 if (!last)
1384 last = dentry;
1385 }
1386 if (ret & DELETE) {
1387 /* stale lease */
1388 di->flags &= ~CEPH_DENTRY_REFERENCED;
1389 if (dentry->d_lockref.count > 0) {
1390 /* update_dentry_lease() will re-add
1391 * it to lease list, or
1392 * ceph_d_delete() will return 1 when
1393 * last reference is dropped */
1394 list_del_init(&di->lease_list);
1395 } else {
1396 di->flags |= CEPH_DENTRY_SHRINK_LIST;
1397 list_move_tail(&di->lease_list, &dispose);
1398 dget_dlock(dentry);
1399 }
1400 }
1401 next:
1402 spin_unlock(&dentry->d_lock);
1403 if (ret & STOP)
1404 break;
1405 }
1406 spin_unlock(&mdsc->dentry_list_lock);
1407
1408 while (!list_empty(&dispose)) {
1409 di = list_first_entry(&dispose, struct ceph_dentry_info,
1410 lease_list);
1411 dentry = di->dentry;
1412 spin_lock(&dentry->d_lock);
1413
1414 list_del_init(&di->lease_list);
1415 di->flags &= ~CEPH_DENTRY_SHRINK_LIST;
1416 if (di->flags & CEPH_DENTRY_REFERENCED) {
1417 spin_lock(&mdsc->dentry_list_lock);
1418 if (di->flags & CEPH_DENTRY_LEASE_LIST) {
1419 list_add_tail(&di->lease_list,
1420 &mdsc->dentry_leases);
1421 } else {
1422 __dentry_dir_lease_touch(mdsc, di);
1423 }
1424 spin_unlock(&mdsc->dentry_list_lock);
1425 } else {
1426 freed++;
1427 }
1428
1429 spin_unlock(&dentry->d_lock);
1430 /* ceph_d_delete() does the trick */
1431 dput(dentry);
1432 }
1433 return freed;
1434 }
1435
1436 static int __dentry_lease_check(struct dentry *dentry, void *arg)
1437 {
1438 struct ceph_dentry_info *di = ceph_dentry(dentry);
1439 int ret;
1440
1441 if (__dentry_lease_is_valid(di))
1442 return STOP;
1443 ret = __dir_lease_try_check(dentry);
1444 if (ret == -EBUSY)
1445 return KEEP;
1446 if (ret > 0)
1447 return TOUCH;
1448 return DELETE;
1449 }
1450
1451 static int __dir_lease_check(struct dentry *dentry, void *arg)
1452 {
1453 struct ceph_lease_walk_control *lwc = arg;
1454 struct ceph_dentry_info *di = ceph_dentry(dentry);
1455
1456 int ret = __dir_lease_try_check(dentry);
1457 if (ret == -EBUSY)
1458 return KEEP;
1459 if (ret > 0) {
1460 if (time_before(jiffies, di->time + lwc->dir_lease_ttl))
1461 return STOP;
1462 /* Move dentry to tail of dir lease list if we don't want
1463 * to delete it. So dentries in the list are checked in a
1464 * round robin manner */
1465 if (!lwc->expire_dir_lease)
1466 return TOUCH;
1467 if (dentry->d_lockref.count > 0 ||
1468 (di->flags & CEPH_DENTRY_REFERENCED))
1469 return TOUCH;
1470 /* invalidate dir lease */
1471 di->lease_shared_gen = 0;
1472 }
1473 return DELETE;
1474 }
1475
1476 int ceph_trim_dentries(struct ceph_mds_client *mdsc)
1477 {
1478 struct ceph_lease_walk_control lwc;
1479 unsigned long count;
1480 unsigned long freed;
1481
1482 spin_lock(&mdsc->caps_list_lock);
1483 if (mdsc->caps_use_max > 0 &&
1484 mdsc->caps_use_count > mdsc->caps_use_max)
1485 count = mdsc->caps_use_count - mdsc->caps_use_max;
1486 else
1487 count = 0;
1488 spin_unlock(&mdsc->caps_list_lock);
1489
1490 lwc.dir_lease = false;
1491 lwc.nr_to_scan = CEPH_CAPS_PER_RELEASE * 2;
1492 freed = __dentry_leases_walk(mdsc, &lwc, __dentry_lease_check);
1493 if (!lwc.nr_to_scan) /* more invalid leases */
1494 return -EAGAIN;
1495
1496 if (lwc.nr_to_scan < CEPH_CAPS_PER_RELEASE)
1497 lwc.nr_to_scan = CEPH_CAPS_PER_RELEASE;
1498
1499 lwc.dir_lease = true;
1500 lwc.expire_dir_lease = freed < count;
1501 lwc.dir_lease_ttl = mdsc->fsc->mount_options->caps_wanted_delay_max * HZ;
1502 freed +=__dentry_leases_walk(mdsc, &lwc, __dir_lease_check);
1503 if (!lwc.nr_to_scan) /* more to check */
1504 return -EAGAIN;
1505
1506 return freed > 0 ? 1 : 0;
1507 }
1508
1509 /*
1510 * Ensure a dentry lease will no longer revalidate.
1511 */
1512 void ceph_invalidate_dentry_lease(struct dentry *dentry)
1513 {
1514 struct ceph_dentry_info *di = ceph_dentry(dentry);
1515 spin_lock(&dentry->d_lock);
1516 di->time = jiffies;
1517 di->lease_shared_gen = 0;
1518 di->flags &= ~CEPH_DENTRY_PRIMARY_LINK;
1519 __dentry_lease_unlist(di);
1520 spin_unlock(&dentry->d_lock);
1521 }
1522
1523 /*
1524 * Check if dentry lease is valid. If not, delete the lease. Try to
1525 * renew if the least is more than half up.
1526 */
1527 static bool __dentry_lease_is_valid(struct ceph_dentry_info *di)
1528 {
1529 struct ceph_mds_session *session;
1530
1531 if (!di->lease_gen)
1532 return false;
1533
1534 session = di->lease_session;
1535 if (session) {
1536 u32 gen;
1537 unsigned long ttl;
1538
1539 spin_lock(&session->s_gen_ttl_lock);
1540 gen = session->s_cap_gen;
1541 ttl = session->s_cap_ttl;
1542 spin_unlock(&session->s_gen_ttl_lock);
1543
1544 if (di->lease_gen == gen &&
1545 time_before(jiffies, ttl) &&
1546 time_before(jiffies, di->time))
1547 return true;
1548 }
1549 di->lease_gen = 0;
1550 return false;
1551 }
1552
1553 static int dentry_lease_is_valid(struct dentry *dentry, unsigned int flags)
1554 {
1555 struct ceph_dentry_info *di;
1556 struct ceph_mds_session *session = NULL;
1557 u32 seq = 0;
1558 int valid = 0;
1559
1560 spin_lock(&dentry->d_lock);
1561 di = ceph_dentry(dentry);
1562 if (di && __dentry_lease_is_valid(di)) {
1563 valid = 1;
1564
1565 if (di->lease_renew_after &&
1566 time_after(jiffies, di->lease_renew_after)) {
1567 /*
1568 * We should renew. If we're in RCU walk mode
1569 * though, we can't do that so just return
1570 * -ECHILD.
1571 */
1572 if (flags & LOOKUP_RCU) {
1573 valid = -ECHILD;
1574 } else {
1575 session = ceph_get_mds_session(di->lease_session);
1576 seq = di->lease_seq;
1577 di->lease_renew_after = 0;
1578 di->lease_renew_from = jiffies;
1579 }
1580 }
1581 }
1582 spin_unlock(&dentry->d_lock);
1583
1584 if (session) {
1585 ceph_mdsc_lease_send_msg(session, dentry,
1586 CEPH_MDS_LEASE_RENEW, seq);
1587 ceph_put_mds_session(session);
1588 }
1589 dout("dentry_lease_is_valid - dentry %p = %d\n", dentry, valid);
1590 return valid;
1591 }
1592
1593 /*
1594 * Called under dentry->d_lock.
1595 */
1596 static int __dir_lease_try_check(const struct dentry *dentry)
1597 {
1598 struct ceph_dentry_info *di = ceph_dentry(dentry);
1599 struct inode *dir;
1600 struct ceph_inode_info *ci;
1601 int valid = 0;
1602
1603 if (!di->lease_shared_gen)
1604 return 0;
1605 if (IS_ROOT(dentry))
1606 return 0;
1607
1608 dir = d_inode(dentry->d_parent);
1609 ci = ceph_inode(dir);
1610
1611 if (spin_trylock(&ci->i_ceph_lock)) {
1612 if (atomic_read(&ci->i_shared_gen) == di->lease_shared_gen &&
1613 __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 0))
1614 valid = 1;
1615 spin_unlock(&ci->i_ceph_lock);
1616 } else {
1617 valid = -EBUSY;
1618 }
1619
1620 if (!valid)
1621 di->lease_shared_gen = 0;
1622 return valid;
1623 }
1624
1625 /*
1626 * Check if directory-wide content lease/cap is valid.
1627 */
1628 static int dir_lease_is_valid(struct inode *dir, struct dentry *dentry,
1629 struct ceph_mds_client *mdsc)
1630 {
1631 struct ceph_inode_info *ci = ceph_inode(dir);
1632 int valid;
1633 int shared_gen;
1634
1635 spin_lock(&ci->i_ceph_lock);
1636 valid = __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1);
1637 if (valid) {
1638 __ceph_touch_fmode(ci, mdsc, CEPH_FILE_MODE_RD);
1639 shared_gen = atomic_read(&ci->i_shared_gen);
1640 }
1641 spin_unlock(&ci->i_ceph_lock);
1642 if (valid) {
1643 struct ceph_dentry_info *di;
1644 spin_lock(&dentry->d_lock);
1645 di = ceph_dentry(dentry);
1646 if (dir == d_inode(dentry->d_parent) &&
1647 di && di->lease_shared_gen == shared_gen)
1648 __ceph_dentry_dir_lease_touch(di);
1649 else
1650 valid = 0;
1651 spin_unlock(&dentry->d_lock);
1652 }
1653 dout("dir_lease_is_valid dir %p v%u dentry %p = %d\n",
1654 dir, (unsigned)atomic_read(&ci->i_shared_gen), dentry, valid);
1655 return valid;
1656 }
1657
1658 /*
1659 * Check if cached dentry can be trusted.
1660 */
1661 static int ceph_d_revalidate(struct dentry *dentry, unsigned int flags)
1662 {
1663 int valid = 0;
1664 struct dentry *parent;
1665 struct inode *dir, *inode;
1666 struct ceph_mds_client *mdsc;
1667
1668 if (flags & LOOKUP_RCU) {
1669 parent = READ_ONCE(dentry->d_parent);
1670 dir = d_inode_rcu(parent);
1671 if (!dir)
1672 return -ECHILD;
1673 inode = d_inode_rcu(dentry);
1674 } else {
1675 parent = dget_parent(dentry);
1676 dir = d_inode(parent);
1677 inode = d_inode(dentry);
1678 }
1679
1680 dout("d_revalidate %p '%pd' inode %p offset 0x%llx\n", dentry,
1681 dentry, inode, ceph_dentry(dentry)->offset);
1682
1683 mdsc = ceph_sb_to_client(dir->i_sb)->mdsc;
1684
1685 /* always trust cached snapped dentries, snapdir dentry */
1686 if (ceph_snap(dir) != CEPH_NOSNAP) {
1687 dout("d_revalidate %p '%pd' inode %p is SNAPPED\n", dentry,
1688 dentry, inode);
1689 valid = 1;
1690 } else if (inode && ceph_snap(inode) == CEPH_SNAPDIR) {
1691 valid = 1;
1692 } else {
1693 valid = dentry_lease_is_valid(dentry, flags);
1694 if (valid == -ECHILD)
1695 return valid;
1696 if (valid || dir_lease_is_valid(dir, dentry, mdsc)) {
1697 if (inode)
1698 valid = ceph_is_any_caps(inode);
1699 else
1700 valid = 1;
1701 }
1702 }
1703
1704 if (!valid) {
1705 struct ceph_mds_request *req;
1706 int op, err;
1707 u32 mask;
1708
1709 if (flags & LOOKUP_RCU)
1710 return -ECHILD;
1711
1712 op = ceph_snap(dir) == CEPH_SNAPDIR ?
1713 CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP;
1714 req = ceph_mdsc_create_request(mdsc, op, USE_ANY_MDS);
1715 if (!IS_ERR(req)) {
1716 req->r_dentry = dget(dentry);
1717 req->r_num_caps = 2;
1718 req->r_parent = dir;
1719
1720 mask = CEPH_STAT_CAP_INODE | CEPH_CAP_AUTH_SHARED;
1721 if (ceph_security_xattr_wanted(dir))
1722 mask |= CEPH_CAP_XATTR_SHARED;
1723 req->r_args.getattr.mask = cpu_to_le32(mask);
1724
1725 err = ceph_mdsc_do_request(mdsc, NULL, req);
1726 switch (err) {
1727 case 0:
1728 if (d_really_is_positive(dentry) &&
1729 d_inode(dentry) == req->r_target_inode)
1730 valid = 1;
1731 break;
1732 case -ENOENT:
1733 if (d_really_is_negative(dentry))
1734 valid = 1;
1735 /* Fallthrough */
1736 default:
1737 break;
1738 }
1739 ceph_mdsc_put_request(req);
1740 dout("d_revalidate %p lookup result=%d\n",
1741 dentry, err);
1742 }
1743 }
1744
1745 dout("d_revalidate %p %s\n", dentry, valid ? "valid" : "invalid");
1746 if (!valid)
1747 ceph_dir_clear_complete(dir);
1748
1749 if (!(flags & LOOKUP_RCU))
1750 dput(parent);
1751 return valid;
1752 }
1753
1754 /*
1755 * Delete unused dentry that doesn't have valid lease
1756 *
1757 * Called under dentry->d_lock.
1758 */
1759 static int ceph_d_delete(const struct dentry *dentry)
1760 {
1761 struct ceph_dentry_info *di;
1762
1763 /* won't release caps */
1764 if (d_really_is_negative(dentry))
1765 return 0;
1766 if (ceph_snap(d_inode(dentry)) != CEPH_NOSNAP)
1767 return 0;
1768 /* vaild lease? */
1769 di = ceph_dentry(dentry);
1770 if (di) {
1771 if (__dentry_lease_is_valid(di))
1772 return 0;
1773 if (__dir_lease_try_check(dentry))
1774 return 0;
1775 }
1776 return 1;
1777 }
1778
1779 /*
1780 * Release our ceph_dentry_info.
1781 */
1782 static void ceph_d_release(struct dentry *dentry)
1783 {
1784 struct ceph_dentry_info *di = ceph_dentry(dentry);
1785
1786 dout("d_release %p\n", dentry);
1787
1788 spin_lock(&dentry->d_lock);
1789 __dentry_lease_unlist(di);
1790 dentry->d_fsdata = NULL;
1791 spin_unlock(&dentry->d_lock);
1792
1793 if (di->lease_session)
1794 ceph_put_mds_session(di->lease_session);
1795 kmem_cache_free(ceph_dentry_cachep, di);
1796 }
1797
1798 /*
1799 * When the VFS prunes a dentry from the cache, we need to clear the
1800 * complete flag on the parent directory.
1801 *
1802 * Called under dentry->d_lock.
1803 */
1804 static void ceph_d_prune(struct dentry *dentry)
1805 {
1806 struct ceph_inode_info *dir_ci;
1807 struct ceph_dentry_info *di;
1808
1809 dout("ceph_d_prune %pd %p\n", dentry, dentry);
1810
1811 /* do we have a valid parent? */
1812 if (IS_ROOT(dentry))
1813 return;
1814
1815 /* we hold d_lock, so d_parent is stable */
1816 dir_ci = ceph_inode(d_inode(dentry->d_parent));
1817 if (dir_ci->i_vino.snap == CEPH_SNAPDIR)
1818 return;
1819
1820 /* who calls d_delete() should also disable dcache readdir */
1821 if (d_really_is_negative(dentry))
1822 return;
1823
1824 /* d_fsdata does not get cleared until d_release */
1825 if (!d_unhashed(dentry)) {
1826 __ceph_dir_clear_complete(dir_ci);
1827 return;
1828 }
1829
1830 /* Disable dcache readdir just in case that someone called d_drop()
1831 * or d_invalidate(), but MDS didn't revoke CEPH_CAP_FILE_SHARED
1832 * properly (dcache readdir is still enabled) */
1833 di = ceph_dentry(dentry);
1834 if (di->offset > 0 &&
1835 di->lease_shared_gen == atomic_read(&dir_ci->i_shared_gen))
1836 __ceph_dir_clear_ordered(dir_ci);
1837 }
1838
1839 /*
1840 * read() on a dir. This weird interface hack only works if mounted
1841 * with '-o dirstat'.
1842 */
1843 static ssize_t ceph_read_dir(struct file *file, char __user *buf, size_t size,
1844 loff_t *ppos)
1845 {
1846 struct ceph_dir_file_info *dfi = file->private_data;
1847 struct inode *inode = file_inode(file);
1848 struct ceph_inode_info *ci = ceph_inode(inode);
1849 int left;
1850 const int bufsize = 1024;
1851
1852 if (!ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb), DIRSTAT))
1853 return -EISDIR;
1854
1855 if (!dfi->dir_info) {
1856 dfi->dir_info = kmalloc(bufsize, GFP_KERNEL);
1857 if (!dfi->dir_info)
1858 return -ENOMEM;
1859 dfi->dir_info_len =
1860 snprintf(dfi->dir_info, bufsize,
1861 "entries: %20lld\n"
1862 " files: %20lld\n"
1863 " subdirs: %20lld\n"
1864 "rentries: %20lld\n"
1865 " rfiles: %20lld\n"
1866 " rsubdirs: %20lld\n"
1867 "rbytes: %20lld\n"
1868 "rctime: %10lld.%09ld\n",
1869 ci->i_files + ci->i_subdirs,
1870 ci->i_files,
1871 ci->i_subdirs,
1872 ci->i_rfiles + ci->i_rsubdirs,
1873 ci->i_rfiles,
1874 ci->i_rsubdirs,
1875 ci->i_rbytes,
1876 ci->i_rctime.tv_sec,
1877 ci->i_rctime.tv_nsec);
1878 }
1879
1880 if (*ppos >= dfi->dir_info_len)
1881 return 0;
1882 size = min_t(unsigned, size, dfi->dir_info_len-*ppos);
1883 left = copy_to_user(buf, dfi->dir_info + *ppos, size);
1884 if (left == size)
1885 return -EFAULT;
1886 *ppos += (size - left);
1887 return size - left;
1888 }
1889
1890
1891
1892 /*
1893 * Return name hash for a given dentry. This is dependent on
1894 * the parent directory's hash function.
1895 */
1896 unsigned ceph_dentry_hash(struct inode *dir, struct dentry *dn)
1897 {
1898 struct ceph_inode_info *dci = ceph_inode(dir);
1899 unsigned hash;
1900
1901 switch (dci->i_dir_layout.dl_dir_hash) {
1902 case 0: /* for backward compat */
1903 case CEPH_STR_HASH_LINUX:
1904 return dn->d_name.hash;
1905
1906 default:
1907 spin_lock(&dn->d_lock);
1908 hash = ceph_str_hash(dci->i_dir_layout.dl_dir_hash,
1909 dn->d_name.name, dn->d_name.len);
1910 spin_unlock(&dn->d_lock);
1911 return hash;
1912 }
1913 }
1914
1915 const struct file_operations ceph_dir_fops = {
1916 .read = ceph_read_dir,
1917 .iterate = ceph_readdir,
1918 .llseek = ceph_dir_llseek,
1919 .open = ceph_open,
1920 .release = ceph_release,
1921 .unlocked_ioctl = ceph_ioctl,
1922 .compat_ioctl = compat_ptr_ioctl,
1923 .fsync = ceph_fsync,
1924 .lock = ceph_lock,
1925 .flock = ceph_flock,
1926 };
1927
1928 const struct file_operations ceph_snapdir_fops = {
1929 .iterate = ceph_readdir,
1930 .llseek = ceph_dir_llseek,
1931 .open = ceph_open,
1932 .release = ceph_release,
1933 };
1934
1935 const struct inode_operations ceph_dir_iops = {
1936 .lookup = ceph_lookup,
1937 .permission = ceph_permission,
1938 .getattr = ceph_getattr,
1939 .setattr = ceph_setattr,
1940 .listxattr = ceph_listxattr,
1941 .get_acl = ceph_get_acl,
1942 .set_acl = ceph_set_acl,
1943 .mknod = ceph_mknod,
1944 .symlink = ceph_symlink,
1945 .mkdir = ceph_mkdir,
1946 .link = ceph_link,
1947 .unlink = ceph_unlink,
1948 .rmdir = ceph_unlink,
1949 .rename = ceph_rename,
1950 .create = ceph_create,
1951 .atomic_open = ceph_atomic_open,
1952 };
1953
1954 const struct inode_operations ceph_snapdir_iops = {
1955 .lookup = ceph_lookup,
1956 .permission = ceph_permission,
1957 .getattr = ceph_getattr,
1958 .mkdir = ceph_mkdir,
1959 .rmdir = ceph_unlink,
1960 .rename = ceph_rename,
1961 };
1962
1963 const struct dentry_operations ceph_dentry_ops = {
1964 .d_revalidate = ceph_d_revalidate,
1965 .d_delete = ceph_d_delete,
1966 .d_release = ceph_d_release,
1967 .d_prune = ceph_d_prune,
1968 .d_init = ceph_d_init,
1969 };
Linux with added FPC-III support