]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - fs/ceph/dir.c
projects / mirror_ubuntu-bionic-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
ceph: properly release page upon error
[mirror_ubuntu-bionic-kernel.git] / fs / ceph / dir.c
1 #include <linux/ceph/ceph_debug.h>
2
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>
8
9 #include "super.h"
10 #include "mds_client.h"
11
12 /*
13 * Directory operations: readdir, lookup, create, link, unlink,
14 * rename, etc.
15 */
16
17 /*
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.
22 *
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.
27 */
28
29 const struct dentry_operations ceph_dentry_ops;
30
31 /*
32 * Initialize ceph dentry state.
33 */
34 int ceph_init_dentry(struct dentry *dentry)
35 {
36 struct ceph_dentry_info *di;
37
38 if (dentry->d_fsdata)
39 return 0;
40
41 di = kmem_cache_alloc(ceph_dentry_cachep, GFP_NOFS | __GFP_ZERO);
42 if (!di)
43 return -ENOMEM; /* oh well */
44
45 spin_lock(&dentry->d_lock);
46 if (dentry->d_fsdata) {
47 /* lost a race */
48 kmem_cache_free(ceph_dentry_cachep, di);
49 goto out_unlock;
50 }
51
52 if (ceph_snap(dentry->d_parent->d_inode) == CEPH_NOSNAP)
53 d_set_d_op(dentry, &ceph_dentry_ops);
54 else if (ceph_snap(dentry->d_parent->d_inode) == CEPH_SNAPDIR)
55 d_set_d_op(dentry, &ceph_snapdir_dentry_ops);
56 else
57 d_set_d_op(dentry, &ceph_snap_dentry_ops);
58
59 di->dentry = dentry;
60 di->lease_session = NULL;
61 dentry->d_time = jiffies;
62 /* avoid reordering d_fsdata setup so that the check above is safe */
63 smp_mb();
64 dentry->d_fsdata = di;
65 ceph_dentry_lru_add(dentry);
66 out_unlock:
67 spin_unlock(&dentry->d_lock);
68 return 0;
69 }
70
71 struct inode *ceph_get_dentry_parent_inode(struct dentry *dentry)
72 {
73 struct inode *inode = NULL;
74
75 if (!dentry)
76 return NULL;
77
78 spin_lock(&dentry->d_lock);
79 if (!IS_ROOT(dentry)) {
80 inode = dentry->d_parent->d_inode;
81 ihold(inode);
82 }
83 spin_unlock(&dentry->d_lock);
84 return inode;
85 }
86
87
88 /*
89 * for readdir, we encode the directory frag and offset within that
90 * frag into f_pos.
91 */
92 static unsigned fpos_frag(loff_t p)
93 {
94 return p >> 32;
95 }
96 static unsigned fpos_off(loff_t p)
97 {
98 return p & 0xffffffff;
99 }
100
101 static int fpos_cmp(loff_t l, loff_t r)
102 {
103 int v = ceph_frag_compare(fpos_frag(l), fpos_frag(r));
104 if (v)
105 return v;
106 return (int)(fpos_off(l) - fpos_off(r));
107 }
108
109 /*
110 * When possible, we try to satisfy a readdir by peeking at the
111 * dcache. We make this work by carefully ordering dentries on
112 * d_child when we initially get results back from the MDS, and
113 * falling back to a "normal" sync readdir if any dentries in the dir
114 * are dropped.
115 *
116 * Complete dir indicates that we have all dentries in the dir. It is
117 * defined IFF we hold CEPH_CAP_FILE_SHARED (which will be revoked by
118 * the MDS if/when the directory is modified).
119 */
120 static int __dcache_readdir(struct file *file, struct dir_context *ctx,
121 u32 shared_gen)
122 {
123 struct ceph_file_info *fi = file->private_data;
124 struct dentry *parent = file->f_path.dentry;
125 struct inode *dir = parent->d_inode;
126 struct list_head *p;
127 struct dentry *dentry, *last;
128 struct ceph_dentry_info *di;
129 int err = 0;
130
131 /* claim ref on last dentry we returned */
132 last = fi->dentry;
133 fi->dentry = NULL;
134
135 dout("__dcache_readdir %p v%u at %llu (last %p)\n",
136 dir, shared_gen, ctx->pos, last);
137
138 spin_lock(&parent->d_lock);
139
140 /* start at beginning? */
141 if (ctx->pos == 2 || last == NULL ||
142 fpos_cmp(ctx->pos, ceph_dentry(last)->offset) < 0) {
143 if (list_empty(&parent->d_subdirs))
144 goto out_unlock;
145 p = parent->d_subdirs.prev;
146 dout(" initial p %p/%p\n", p->prev, p->next);
147 } else {
148 p = last->d_child.prev;
149 }
150
151 more:
152 dentry = list_entry(p, struct dentry, d_child);
153 di = ceph_dentry(dentry);
154 while (1) {
155 dout(" p %p/%p %s d_subdirs %p/%p\n", p->prev, p->next,
156 d_unhashed(dentry) ? "!hashed" : "hashed",
157 parent->d_subdirs.prev, parent->d_subdirs.next);
158 if (p == &parent->d_subdirs) {
159 fi->flags |= CEPH_F_ATEND;
160 goto out_unlock;
161 }
162 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
163 if (di->lease_shared_gen == shared_gen &&
164 !d_unhashed(dentry) && dentry->d_inode &&
165 ceph_snap(dentry->d_inode) != CEPH_SNAPDIR &&
166 ceph_ino(dentry->d_inode) != CEPH_INO_CEPH &&
167 fpos_cmp(ctx->pos, di->offset) <= 0)
168 break;
169 dout(" skipping %p %pd at %llu (%llu)%s%s\n", dentry,
170 dentry, di->offset,
171 ctx->pos, d_unhashed(dentry) ? " unhashed" : "",
172 !dentry->d_inode ? " null" : "");
173 spin_unlock(&dentry->d_lock);
174 p = p->prev;
175 dentry = list_entry(p, struct dentry, d_child);
176 di = ceph_dentry(dentry);
177 }
178
179 dget_dlock(dentry);
180 spin_unlock(&dentry->d_lock);
181 spin_unlock(&parent->d_lock);
182
183 /* make sure a dentry wasn't dropped while we didn't have parent lock */
184 if (!ceph_dir_is_complete_ordered(dir)) {
185 dout(" lost dir complete on %p; falling back to mds\n", dir);
186 dput(dentry);
187 err = -EAGAIN;
188 goto out;
189 }
190
191 dout(" %llu (%llu) dentry %p %pd %p\n", di->offset, ctx->pos,
192 dentry, dentry, dentry->d_inode);
193 if (!dir_emit(ctx, dentry->d_name.name,
194 dentry->d_name.len,
195 ceph_translate_ino(dentry->d_sb, dentry->d_inode->i_ino),
196 dentry->d_inode->i_mode >> 12)) {
197 if (last) {
198 /* remember our position */
199 fi->dentry = last;
200 fi->next_offset = fpos_off(di->offset);
201 }
202 dput(dentry);
203 return 0;
204 }
205
206 ctx->pos = di->offset + 1;
207
208 if (last)
209 dput(last);
210 last = dentry;
211
212 spin_lock(&parent->d_lock);
213 p = p->prev; /* advance to next dentry */
214 goto more;
215
216 out_unlock:
217 spin_unlock(&parent->d_lock);
218 out:
219 if (last)
220 dput(last);
221 return err;
222 }
223
224 /*
225 * make note of the last dentry we read, so we can
226 * continue at the same lexicographical point,
227 * regardless of what dir changes take place on the
228 * server.
229 */
230 static int note_last_dentry(struct ceph_file_info *fi, const char *name,
231 int len)
232 {
233 kfree(fi->last_name);
234 fi->last_name = kmalloc(len+1, GFP_NOFS);
235 if (!fi->last_name)
236 return -ENOMEM;
237 memcpy(fi->last_name, name, len);
238 fi->last_name[len] = 0;
239 dout("note_last_dentry '%s'\n", fi->last_name);
240 return 0;
241 }
242
243 static int ceph_readdir(struct file *file, struct dir_context *ctx)
244 {
245 struct ceph_file_info *fi = file->private_data;
246 struct inode *inode = file_inode(file);
247 struct ceph_inode_info *ci = ceph_inode(inode);
248 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
249 struct ceph_mds_client *mdsc = fsc->mdsc;
250 unsigned frag = fpos_frag(ctx->pos);
251 int off = fpos_off(ctx->pos);
252 int err;
253 u32 ftype;
254 struct ceph_mds_reply_info_parsed *rinfo;
255
256 dout("readdir %p file %p frag %u off %u\n", inode, file, frag, off);
257 if (fi->flags & CEPH_F_ATEND)
258 return 0;
259
260 /* always start with . and .. */
261 if (ctx->pos == 0) {
262 dout("readdir off 0 -> '.'\n");
263 if (!dir_emit(ctx, ".", 1,
264 ceph_translate_ino(inode->i_sb, inode->i_ino),
265 inode->i_mode >> 12))
266 return 0;
267 ctx->pos = 1;
268 off = 1;
269 }
270 if (ctx->pos == 1) {
271 ino_t ino = parent_ino(file->f_path.dentry);
272 dout("readdir off 1 -> '..'\n");
273 if (!dir_emit(ctx, "..", 2,
274 ceph_translate_ino(inode->i_sb, ino),
275 inode->i_mode >> 12))
276 return 0;
277 ctx->pos = 2;
278 off = 2;
279 }
280
281 /* can we use the dcache? */
282 spin_lock(&ci->i_ceph_lock);
283 if ((ctx->pos == 2 || fi->dentry) &&
284 ceph_test_mount_opt(fsc, DCACHE) &&
285 !ceph_test_mount_opt(fsc, NOASYNCREADDIR) &&
286 ceph_snap(inode) != CEPH_SNAPDIR &&
287 __ceph_dir_is_complete_ordered(ci) &&
288 __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1)) {
289 u32 shared_gen = ci->i_shared_gen;
290 spin_unlock(&ci->i_ceph_lock);
291 err = __dcache_readdir(file, ctx, shared_gen);
292 if (err != -EAGAIN)
293 return err;
294 frag = fpos_frag(ctx->pos);
295 off = fpos_off(ctx->pos);
296 } else {
297 spin_unlock(&ci->i_ceph_lock);
298 }
299 if (fi->dentry) {
300 err = note_last_dentry(fi, fi->dentry->d_name.name,
301 fi->dentry->d_name.len);
302 if (err)
303 return err;
304 dput(fi->dentry);
305 fi->dentry = NULL;
306 }
307
308 /* proceed with a normal readdir */
309
310 if (ctx->pos == 2) {
311 /* note dir version at start of readdir so we can tell
312 * if any dentries get dropped */
313 fi->dir_release_count = atomic_read(&ci->i_release_count);
314 fi->dir_ordered_count = ci->i_ordered_count;
315 }
316
317 more:
318 /* do we have the correct frag content buffered? */
319 if (fi->frag != frag || fi->last_readdir == NULL) {
320 struct ceph_mds_request *req;
321 int op = ceph_snap(inode) == CEPH_SNAPDIR ?
322 CEPH_MDS_OP_LSSNAP : CEPH_MDS_OP_READDIR;
323
324 /* discard old result, if any */
325 if (fi->last_readdir) {
326 ceph_mdsc_put_request(fi->last_readdir);
327 fi->last_readdir = NULL;
328 }
329
330 dout("readdir fetching %llx.%llx frag %x offset '%s'\n",
331 ceph_vinop(inode), frag, fi->last_name);
332 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
333 if (IS_ERR(req))
334 return PTR_ERR(req);
335 err = ceph_alloc_readdir_reply_buffer(req, inode);
336 if (err) {
337 ceph_mdsc_put_request(req);
338 return err;
339 }
340 req->r_inode = inode;
341 ihold(inode);
342 req->r_dentry = dget(file->f_path.dentry);
343 /* hints to request -> mds selection code */
344 req->r_direct_mode = USE_AUTH_MDS;
345 req->r_direct_hash = ceph_frag_value(frag);
346 req->r_direct_is_hash = true;
347 req->r_path2 = kstrdup(fi->last_name, GFP_NOFS);
348 req->r_readdir_offset = fi->next_offset;
349 req->r_args.readdir.frag = cpu_to_le32(frag);
350 err = ceph_mdsc_do_request(mdsc, NULL, req);
351 if (err < 0) {
352 ceph_mdsc_put_request(req);
353 return err;
354 }
355 dout("readdir got and parsed readdir result=%d"
356 " on frag %x, end=%d, complete=%d\n", err, frag,
357 (int)req->r_reply_info.dir_end,
358 (int)req->r_reply_info.dir_complete);
359
360 if (!req->r_did_prepopulate) {
361 dout("readdir !did_prepopulate");
362 /* preclude from marking dir complete */
363 fi->dir_release_count--;
364 }
365
366 /* note next offset and last dentry name */
367 rinfo = &req->r_reply_info;
368 if (le32_to_cpu(rinfo->dir_dir->frag) != frag) {
369 frag = le32_to_cpu(rinfo->dir_dir->frag);
370 if (ceph_frag_is_leftmost(frag))
371 fi->next_offset = 2;
372 else
373 fi->next_offset = 0;
374 off = fi->next_offset;
375 }
376 fi->frag = frag;
377 fi->offset = fi->next_offset;
378 fi->last_readdir = req;
379
380 if (req->r_reply_info.dir_end) {
381 kfree(fi->last_name);
382 fi->last_name = NULL;
383 if (ceph_frag_is_rightmost(frag))
384 fi->next_offset = 2;
385 else
386 fi->next_offset = 0;
387 } else {
388 err = note_last_dentry(fi,
389 rinfo->dir_dname[rinfo->dir_nr-1],
390 rinfo->dir_dname_len[rinfo->dir_nr-1]);
391 if (err)
392 return err;
393 fi->next_offset += rinfo->dir_nr;
394 }
395 }
396
397 rinfo = &fi->last_readdir->r_reply_info;
398 dout("readdir frag %x num %d off %d chunkoff %d\n", frag,
399 rinfo->dir_nr, off, fi->offset);
400
401 ctx->pos = ceph_make_fpos(frag, off);
402 while (off >= fi->offset && off - fi->offset < rinfo->dir_nr) {
403 struct ceph_mds_reply_inode *in =
404 rinfo->dir_in[off - fi->offset].in;
405 struct ceph_vino vino;
406 ino_t ino;
407
408 dout("readdir off %d (%d/%d) -> %lld '%.*s' %p\n",
409 off, off - fi->offset, rinfo->dir_nr, ctx->pos,
410 rinfo->dir_dname_len[off - fi->offset],
411 rinfo->dir_dname[off - fi->offset], in);
412 BUG_ON(!in);
413 ftype = le32_to_cpu(in->mode) >> 12;
414 vino.ino = le64_to_cpu(in->ino);
415 vino.snap = le64_to_cpu(in->snapid);
416 ino = ceph_vino_to_ino(vino);
417 if (!dir_emit(ctx,
418 rinfo->dir_dname[off - fi->offset],
419 rinfo->dir_dname_len[off - fi->offset],
420 ceph_translate_ino(inode->i_sb, ino), ftype)) {
421 dout("filldir stopping us...\n");
422 return 0;
423 }
424 off++;
425 ctx->pos++;
426 }
427
428 if (fi->last_name) {
429 ceph_mdsc_put_request(fi->last_readdir);
430 fi->last_readdir = NULL;
431 goto more;
432 }
433
434 /* more frags? */
435 if (!ceph_frag_is_rightmost(frag)) {
436 frag = ceph_frag_next(frag);
437 off = 0;
438 ctx->pos = ceph_make_fpos(frag, off);
439 dout("readdir next frag is %x\n", frag);
440 goto more;
441 }
442 fi->flags |= CEPH_F_ATEND;
443
444 /*
445 * if dir_release_count still matches the dir, no dentries
446 * were released during the whole readdir, and we should have
447 * the complete dir contents in our cache.
448 */
449 spin_lock(&ci->i_ceph_lock);
450 if (atomic_read(&ci->i_release_count) == fi->dir_release_count) {
451 if (ci->i_ordered_count == fi->dir_ordered_count)
452 dout(" marking %p complete and ordered\n", inode);
453 else
454 dout(" marking %p complete\n", inode);
455 __ceph_dir_set_complete(ci, fi->dir_release_count,
456 fi->dir_ordered_count);
457 }
458 spin_unlock(&ci->i_ceph_lock);
459
460 dout("readdir %p file %p done.\n", inode, file);
461 return 0;
462 }
463
464 static void reset_readdir(struct ceph_file_info *fi, unsigned frag)
465 {
466 if (fi->last_readdir) {
467 ceph_mdsc_put_request(fi->last_readdir);
468 fi->last_readdir = NULL;
469 }
470 kfree(fi->last_name);
471 fi->last_name = NULL;
472 if (ceph_frag_is_leftmost(frag))
473 fi->next_offset = 2; /* compensate for . and .. */
474 else
475 fi->next_offset = 0;
476 if (fi->dentry) {
477 dput(fi->dentry);
478 fi->dentry = NULL;
479 }
480 fi->flags &= ~CEPH_F_ATEND;
481 }
482
483 static loff_t ceph_dir_llseek(struct file *file, loff_t offset, int whence)
484 {
485 struct ceph_file_info *fi = file->private_data;
486 struct inode *inode = file->f_mapping->host;
487 loff_t old_offset = ceph_make_fpos(fi->frag, fi->next_offset);
488 loff_t retval;
489
490 mutex_lock(&inode->i_mutex);
491 retval = -EINVAL;
492 switch (whence) {
493 case SEEK_END:
494 offset += inode->i_size + 2; /* FIXME */
495 break;
496 case SEEK_CUR:
497 offset += file->f_pos;
498 case SEEK_SET:
499 break;
500 default:
501 goto out;
502 }
503
504 if (offset >= 0) {
505 if (offset != file->f_pos) {
506 file->f_pos = offset;
507 file->f_version = 0;
508 fi->flags &= ~CEPH_F_ATEND;
509 }
510 retval = offset;
511
512 /*
513 * discard buffered readdir content on seekdir(0), or
514 * seek to new frag, or seek prior to current chunk.
515 */
516 if (offset == 0 ||
517 fpos_frag(offset) != fi->frag ||
518 fpos_off(offset) < fi->offset) {
519 dout("dir_llseek dropping %p content\n", file);
520 reset_readdir(fi, fpos_frag(offset));
521 }
522
523 /* bump dir_release_count if we did a forward seek */
524 if (fpos_cmp(offset, old_offset) > 0)
525 fi->dir_release_count--;
526 }
527 out:
528 mutex_unlock(&inode->i_mutex);
529 return retval;
530 }
531
532 /*
533 * Handle lookups for the hidden .snap directory.
534 */
535 int ceph_handle_snapdir(struct ceph_mds_request *req,
536 struct dentry *dentry, int err)
537 {
538 struct ceph_fs_client *fsc = ceph_sb_to_client(dentry->d_sb);
539 struct inode *parent = dentry->d_parent->d_inode; /* we hold i_mutex */
540
541 /* .snap dir? */
542 if (err == -ENOENT &&
543 ceph_snap(parent) == CEPH_NOSNAP &&
544 strcmp(dentry->d_name.name,
545 fsc->mount_options->snapdir_name) == 0) {
546 struct inode *inode = ceph_get_snapdir(parent);
547 dout("ENOENT on snapdir %p '%pd', linking to snapdir %p\n",
548 dentry, dentry, inode);
549 BUG_ON(!d_unhashed(dentry));
550 d_add(dentry, inode);
551 err = 0;
552 }
553 return err;
554 }
555
556 /*
557 * Figure out final result of a lookup/open request.
558 *
559 * Mainly, make sure we return the final req->r_dentry (if it already
560 * existed) in place of the original VFS-provided dentry when they
561 * differ.
562 *
563 * Gracefully handle the case where the MDS replies with -ENOENT and
564 * no trace (which it may do, at its discretion, e.g., if it doesn't
565 * care to issue a lease on the negative dentry).
566 */
567 struct dentry *ceph_finish_lookup(struct ceph_mds_request *req,
568 struct dentry *dentry, int err)
569 {
570 if (err == -ENOENT) {
571 /* no trace? */
572 err = 0;
573 if (!req->r_reply_info.head->is_dentry) {
574 dout("ENOENT and no trace, dentry %p inode %p\n",
575 dentry, dentry->d_inode);
576 if (dentry->d_inode) {
577 d_drop(dentry);
578 err = -ENOENT;
579 } else {
580 d_add(dentry, NULL);
581 }
582 }
583 }
584 if (err)
585 dentry = ERR_PTR(err);
586 else if (dentry != req->r_dentry)
587 dentry = dget(req->r_dentry); /* we got spliced */
588 else
589 dentry = NULL;
590 return dentry;
591 }
592
593 static int is_root_ceph_dentry(struct inode *inode, struct dentry *dentry)
594 {
595 return ceph_ino(inode) == CEPH_INO_ROOT &&
596 strncmp(dentry->d_name.name, ".ceph", 5) == 0;
597 }
598
599 /*
600 * Look up a single dir entry. If there is a lookup intent, inform
601 * the MDS so that it gets our 'caps wanted' value in a single op.
602 */
603 static struct dentry *ceph_lookup(struct inode *dir, struct dentry *dentry,
604 unsigned int flags)
605 {
606 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
607 struct ceph_mds_client *mdsc = fsc->mdsc;
608 struct ceph_mds_request *req;
609 int op;
610 int err;
611
612 dout("lookup %p dentry %p '%pd'\n",
613 dir, dentry, dentry);
614
615 if (dentry->d_name.len > NAME_MAX)
616 return ERR_PTR(-ENAMETOOLONG);
617
618 err = ceph_init_dentry(dentry);
619 if (err < 0)
620 return ERR_PTR(err);
621
622 /* can we conclude ENOENT locally? */
623 if (dentry->d_inode == NULL) {
624 struct ceph_inode_info *ci = ceph_inode(dir);
625 struct ceph_dentry_info *di = ceph_dentry(dentry);
626
627 spin_lock(&ci->i_ceph_lock);
628 dout(" dir %p flags are %d\n", dir, ci->i_ceph_flags);
629 if (strncmp(dentry->d_name.name,
630 fsc->mount_options->snapdir_name,
631 dentry->d_name.len) &&
632 !is_root_ceph_dentry(dir, dentry) &&
633 ceph_test_mount_opt(fsc, DCACHE) &&
634 __ceph_dir_is_complete(ci) &&
635 (__ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1))) {
636 spin_unlock(&ci->i_ceph_lock);
637 dout(" dir %p complete, -ENOENT\n", dir);
638 d_add(dentry, NULL);
639 di->lease_shared_gen = ci->i_shared_gen;
640 return NULL;
641 }
642 spin_unlock(&ci->i_ceph_lock);
643 }
644
645 op = ceph_snap(dir) == CEPH_SNAPDIR ?
646 CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP;
647 req = ceph_mdsc_create_request(mdsc, op, USE_ANY_MDS);
648 if (IS_ERR(req))
649 return ERR_CAST(req);
650 req->r_dentry = dget(dentry);
651 req->r_num_caps = 2;
652 /* we only need inode linkage */
653 req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
654 req->r_locked_dir = dir;
655 err = ceph_mdsc_do_request(mdsc, NULL, req);
656 err = ceph_handle_snapdir(req, dentry, err);
657 dentry = ceph_finish_lookup(req, dentry, err);
658 ceph_mdsc_put_request(req); /* will dput(dentry) */
659 dout("lookup result=%p\n", dentry);
660 return dentry;
661 }
662
663 /*
664 * If we do a create but get no trace back from the MDS, follow up with
665 * a lookup (the VFS expects us to link up the provided dentry).
666 */
667 int ceph_handle_notrace_create(struct inode *dir, struct dentry *dentry)
668 {
669 struct dentry *result = ceph_lookup(dir, dentry, 0);
670
671 if (result && !IS_ERR(result)) {
672 /*
673 * We created the item, then did a lookup, and found
674 * it was already linked to another inode we already
675 * had in our cache (and thus got spliced). To not
676 * confuse VFS (especially when inode is a directory),
677 * we don't link our dentry to that inode, return an
678 * error instead.
679 *
680 * This event should be rare and it happens only when
681 * we talk to old MDS. Recent MDS does not send traceless
682 * reply for request that creates new inode.
683 */
684 d_drop(result);
685 return -ESTALE;
686 }
687 return PTR_ERR(result);
688 }
689
690 static int ceph_mknod(struct inode *dir, struct dentry *dentry,
691 umode_t mode, dev_t rdev)
692 {
693 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
694 struct ceph_mds_client *mdsc = fsc->mdsc;
695 struct ceph_mds_request *req;
696 struct ceph_acls_info acls = {};
697 int err;
698
699 if (ceph_snap(dir) != CEPH_NOSNAP)
700 return -EROFS;
701
702 err = ceph_pre_init_acls(dir, &mode, &acls);
703 if (err < 0)
704 return err;
705
706 dout("mknod in dir %p dentry %p mode 0%ho rdev %d\n",
707 dir, dentry, mode, rdev);
708 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_MKNOD, USE_AUTH_MDS);
709 if (IS_ERR(req)) {
710 err = PTR_ERR(req);
711 goto out;
712 }
713 req->r_dentry = dget(dentry);
714 req->r_num_caps = 2;
715 req->r_locked_dir = dir;
716 req->r_args.mknod.mode = cpu_to_le32(mode);
717 req->r_args.mknod.rdev = cpu_to_le32(rdev);
718 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
719 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
720 if (acls.pagelist) {
721 req->r_pagelist = acls.pagelist;
722 acls.pagelist = NULL;
723 }
724 err = ceph_mdsc_do_request(mdsc, dir, req);
725 if (!err && !req->r_reply_info.head->is_dentry)
726 err = ceph_handle_notrace_create(dir, dentry);
727 ceph_mdsc_put_request(req);
728 out:
729 if (!err)
730 ceph_init_inode_acls(dentry->d_inode, &acls);
731 else
732 d_drop(dentry);
733 ceph_release_acls_info(&acls);
734 return err;
735 }
736
737 static int ceph_create(struct inode *dir, struct dentry *dentry, umode_t mode,
738 bool excl)
739 {
740 return ceph_mknod(dir, dentry, mode, 0);
741 }
742
743 static int ceph_symlink(struct inode *dir, struct dentry *dentry,
744 const char *dest)
745 {
746 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
747 struct ceph_mds_client *mdsc = fsc->mdsc;
748 struct ceph_mds_request *req;
749 int err;
750
751 if (ceph_snap(dir) != CEPH_NOSNAP)
752 return -EROFS;
753
754 dout("symlink in dir %p dentry %p to '%s'\n", dir, dentry, dest);
755 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SYMLINK, USE_AUTH_MDS);
756 if (IS_ERR(req)) {
757 err = PTR_ERR(req);
758 goto out;
759 }
760 req->r_dentry = dget(dentry);
761 req->r_num_caps = 2;
762 req->r_path2 = kstrdup(dest, GFP_NOFS);
763 req->r_locked_dir = dir;
764 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
765 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
766 err = ceph_mdsc_do_request(mdsc, dir, req);
767 if (!err && !req->r_reply_info.head->is_dentry)
768 err = ceph_handle_notrace_create(dir, dentry);
769 ceph_mdsc_put_request(req);
770 out:
771 if (err)
772 d_drop(dentry);
773 return err;
774 }
775
776 static int ceph_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
777 {
778 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
779 struct ceph_mds_client *mdsc = fsc->mdsc;
780 struct ceph_mds_request *req;
781 struct ceph_acls_info acls = {};
782 int err = -EROFS;
783 int op;
784
785 if (ceph_snap(dir) == CEPH_SNAPDIR) {
786 /* mkdir .snap/foo is a MKSNAP */
787 op = CEPH_MDS_OP_MKSNAP;
788 dout("mksnap dir %p snap '%pd' dn %p\n", dir,
789 dentry, dentry);
790 } else if (ceph_snap(dir) == CEPH_NOSNAP) {
791 dout("mkdir dir %p dn %p mode 0%ho\n", dir, dentry, mode);
792 op = CEPH_MDS_OP_MKDIR;
793 } else {
794 goto out;
795 }
796
797 mode |= S_IFDIR;
798 err = ceph_pre_init_acls(dir, &mode, &acls);
799 if (err < 0)
800 goto out;
801
802 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
803 if (IS_ERR(req)) {
804 err = PTR_ERR(req);
805 goto out;
806 }
807
808 req->r_dentry = dget(dentry);
809 req->r_num_caps = 2;
810 req->r_locked_dir = dir;
811 req->r_args.mkdir.mode = cpu_to_le32(mode);
812 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
813 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
814 if (acls.pagelist) {
815 req->r_pagelist = acls.pagelist;
816 acls.pagelist = NULL;
817 }
818 err = ceph_mdsc_do_request(mdsc, dir, req);
819 if (!err &&
820 !req->r_reply_info.head->is_target &&
821 !req->r_reply_info.head->is_dentry)
822 err = ceph_handle_notrace_create(dir, dentry);
823 ceph_mdsc_put_request(req);
824 out:
825 if (!err)
826 ceph_init_inode_acls(dentry->d_inode, &acls);
827 else
828 d_drop(dentry);
829 ceph_release_acls_info(&acls);
830 return err;
831 }
832
833 static int ceph_link(struct dentry *old_dentry, struct inode *dir,
834 struct dentry *dentry)
835 {
836 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
837 struct ceph_mds_client *mdsc = fsc->mdsc;
838 struct ceph_mds_request *req;
839 int err;
840
841 if (ceph_snap(dir) != CEPH_NOSNAP)
842 return -EROFS;
843
844 dout("link in dir %p old_dentry %p dentry %p\n", dir,
845 old_dentry, dentry);
846 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LINK, USE_AUTH_MDS);
847 if (IS_ERR(req)) {
848 d_drop(dentry);
849 return PTR_ERR(req);
850 }
851 req->r_dentry = dget(dentry);
852 req->r_num_caps = 2;
853 req->r_old_dentry = dget(old_dentry);
854 req->r_locked_dir = dir;
855 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
856 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
857 /* release LINK_SHARED on source inode (mds will lock it) */
858 req->r_old_inode_drop = CEPH_CAP_LINK_SHARED;
859 err = ceph_mdsc_do_request(mdsc, dir, req);
860 if (err) {
861 d_drop(dentry);
862 } else if (!req->r_reply_info.head->is_dentry) {
863 ihold(old_dentry->d_inode);
864 d_instantiate(dentry, old_dentry->d_inode);
865 }
866 ceph_mdsc_put_request(req);
867 return err;
868 }
869
870 /*
871 * For a soon-to-be unlinked file, drop the AUTH_RDCACHE caps. If it
872 * looks like the link count will hit 0, drop any other caps (other
873 * than PIN) we don't specifically want (due to the file still being
874 * open).
875 */
876 static int drop_caps_for_unlink(struct inode *inode)
877 {
878 struct ceph_inode_info *ci = ceph_inode(inode);
879 int drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL;
880
881 spin_lock(&ci->i_ceph_lock);
882 if (inode->i_nlink == 1) {
883 drop |= ~(__ceph_caps_wanted(ci) | CEPH_CAP_PIN);
884 ci->i_ceph_flags |= CEPH_I_NODELAY;
885 }
886 spin_unlock(&ci->i_ceph_lock);
887 return drop;
888 }
889
890 /*
891 * rmdir and unlink are differ only by the metadata op code
892 */
893 static int ceph_unlink(struct inode *dir, struct dentry *dentry)
894 {
895 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
896 struct ceph_mds_client *mdsc = fsc->mdsc;
897 struct inode *inode = dentry->d_inode;
898 struct ceph_mds_request *req;
899 int err = -EROFS;
900 int op;
901
902 if (ceph_snap(dir) == CEPH_SNAPDIR) {
903 /* rmdir .snap/foo is RMSNAP */
904 dout("rmsnap dir %p '%pd' dn %p\n", dir, dentry, dentry);
905 op = CEPH_MDS_OP_RMSNAP;
906 } else if (ceph_snap(dir) == CEPH_NOSNAP) {
907 dout("unlink/rmdir dir %p dn %p inode %p\n",
908 dir, dentry, inode);
909 op = d_is_dir(dentry) ?
910 CEPH_MDS_OP_RMDIR : CEPH_MDS_OP_UNLINK;
911 } else
912 goto out;
913 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
914 if (IS_ERR(req)) {
915 err = PTR_ERR(req);
916 goto out;
917 }
918 req->r_dentry = dget(dentry);
919 req->r_num_caps = 2;
920 req->r_locked_dir = dir;
921 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
922 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
923 req->r_inode_drop = drop_caps_for_unlink(inode);
924 err = ceph_mdsc_do_request(mdsc, dir, req);
925 if (!err && !req->r_reply_info.head->is_dentry)
926 d_delete(dentry);
927 ceph_mdsc_put_request(req);
928 out:
929 return err;
930 }
931
932 static int ceph_rename(struct inode *old_dir, struct dentry *old_dentry,
933 struct inode *new_dir, struct dentry *new_dentry)
934 {
935 struct ceph_fs_client *fsc = ceph_sb_to_client(old_dir->i_sb);
936 struct ceph_mds_client *mdsc = fsc->mdsc;
937 struct ceph_mds_request *req;
938 int err;
939
940 if (ceph_snap(old_dir) != ceph_snap(new_dir))
941 return -EXDEV;
942 if (ceph_snap(old_dir) != CEPH_NOSNAP ||
943 ceph_snap(new_dir) != CEPH_NOSNAP)
944 return -EROFS;
945 dout("rename dir %p dentry %p to dir %p dentry %p\n",
946 old_dir, old_dentry, new_dir, new_dentry);
947 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_RENAME, USE_AUTH_MDS);
948 if (IS_ERR(req))
949 return PTR_ERR(req);
950 ihold(old_dir);
951 req->r_dentry = dget(new_dentry);
952 req->r_num_caps = 2;
953 req->r_old_dentry = dget(old_dentry);
954 req->r_old_dentry_dir = old_dir;
955 req->r_locked_dir = new_dir;
956 req->r_old_dentry_drop = CEPH_CAP_FILE_SHARED;
957 req->r_old_dentry_unless = CEPH_CAP_FILE_EXCL;
958 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
959 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
960 /* release LINK_RDCACHE on source inode (mds will lock it) */
961 req->r_old_inode_drop = CEPH_CAP_LINK_SHARED;
962 if (new_dentry->d_inode)
963 req->r_inode_drop = drop_caps_for_unlink(new_dentry->d_inode);
964 err = ceph_mdsc_do_request(mdsc, old_dir, req);
965 if (!err && !req->r_reply_info.head->is_dentry) {
966 /*
967 * Normally d_move() is done by fill_trace (called by
968 * do_request, above). If there is no trace, we need
969 * to do it here.
970 */
971
972 d_move(old_dentry, new_dentry);
973
974 /* ensure target dentry is invalidated, despite
975 rehashing bug in vfs_rename_dir */
976 ceph_invalidate_dentry_lease(new_dentry);
977
978 /* d_move screws up sibling dentries' offsets */
979 ceph_dir_clear_complete(old_dir);
980 ceph_dir_clear_complete(new_dir);
981
982 }
983 ceph_mdsc_put_request(req);
984 return err;
985 }
986
987 /*
988 * Ensure a dentry lease will no longer revalidate.
989 */
990 void ceph_invalidate_dentry_lease(struct dentry *dentry)
991 {
992 spin_lock(&dentry->d_lock);
993 dentry->d_time = jiffies;
994 ceph_dentry(dentry)->lease_shared_gen = 0;
995 spin_unlock(&dentry->d_lock);
996 }
997
998 /*
999 * Check if dentry lease is valid. If not, delete the lease. Try to
1000 * renew if the least is more than half up.
1001 */
1002 static int dentry_lease_is_valid(struct dentry *dentry)
1003 {
1004 struct ceph_dentry_info *di;
1005 struct ceph_mds_session *s;
1006 int valid = 0;
1007 u32 gen;
1008 unsigned long ttl;
1009 struct ceph_mds_session *session = NULL;
1010 struct inode *dir = NULL;
1011 u32 seq = 0;
1012
1013 spin_lock(&dentry->d_lock);
1014 di = ceph_dentry(dentry);
1015 if (di->lease_session) {
1016 s = di->lease_session;
1017 spin_lock(&s->s_gen_ttl_lock);
1018 gen = s->s_cap_gen;
1019 ttl = s->s_cap_ttl;
1020 spin_unlock(&s->s_gen_ttl_lock);
1021
1022 if (di->lease_gen == gen &&
1023 time_before(jiffies, dentry->d_time) &&
1024 time_before(jiffies, ttl)) {
1025 valid = 1;
1026 if (di->lease_renew_after &&
1027 time_after(jiffies, di->lease_renew_after)) {
1028 /* we should renew */
1029 dir = dentry->d_parent->d_inode;
1030 session = ceph_get_mds_session(s);
1031 seq = di->lease_seq;
1032 di->lease_renew_after = 0;
1033 di->lease_renew_from = jiffies;
1034 }
1035 }
1036 }
1037 spin_unlock(&dentry->d_lock);
1038
1039 if (session) {
1040 ceph_mdsc_lease_send_msg(session, dir, dentry,
1041 CEPH_MDS_LEASE_RENEW, seq);
1042 ceph_put_mds_session(session);
1043 }
1044 dout("dentry_lease_is_valid - dentry %p = %d\n", dentry, valid);
1045 return valid;
1046 }
1047
1048 /*
1049 * Check if directory-wide content lease/cap is valid.
1050 */
1051 static int dir_lease_is_valid(struct inode *dir, struct dentry *dentry)
1052 {
1053 struct ceph_inode_info *ci = ceph_inode(dir);
1054 struct ceph_dentry_info *di = ceph_dentry(dentry);
1055 int valid = 0;
1056
1057 spin_lock(&ci->i_ceph_lock);
1058 if (ci->i_shared_gen == di->lease_shared_gen)
1059 valid = __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1);
1060 spin_unlock(&ci->i_ceph_lock);
1061 dout("dir_lease_is_valid dir %p v%u dentry %p v%u = %d\n",
1062 dir, (unsigned)ci->i_shared_gen, dentry,
1063 (unsigned)di->lease_shared_gen, valid);
1064 return valid;
1065 }
1066
1067 /*
1068 * Check if cached dentry can be trusted.
1069 */
1070 static int ceph_d_revalidate(struct dentry *dentry, unsigned int flags)
1071 {
1072 int valid = 0;
1073 struct inode *dir;
1074
1075 if (flags & LOOKUP_RCU)
1076 return -ECHILD;
1077
1078 dout("d_revalidate %p '%pd' inode %p offset %lld\n", dentry,
1079 dentry, dentry->d_inode, ceph_dentry(dentry)->offset);
1080
1081 dir = ceph_get_dentry_parent_inode(dentry);
1082
1083 /* always trust cached snapped dentries, snapdir dentry */
1084 if (ceph_snap(dir) != CEPH_NOSNAP) {
1085 dout("d_revalidate %p '%pd' inode %p is SNAPPED\n", dentry,
1086 dentry, dentry->d_inode);
1087 valid = 1;
1088 } else if (dentry->d_inode &&
1089 ceph_snap(dentry->d_inode) == CEPH_SNAPDIR) {
1090 valid = 1;
1091 } else if (dentry_lease_is_valid(dentry) ||
1092 dir_lease_is_valid(dir, dentry)) {
1093 if (dentry->d_inode)
1094 valid = ceph_is_any_caps(dentry->d_inode);
1095 else
1096 valid = 1;
1097 }
1098
1099 dout("d_revalidate %p %s\n", dentry, valid ? "valid" : "invalid");
1100 if (valid) {
1101 ceph_dentry_lru_touch(dentry);
1102 } else {
1103 ceph_dir_clear_complete(dir);
1104 }
1105 iput(dir);
1106 return valid;
1107 }
1108
1109 /*
1110 * Release our ceph_dentry_info.
1111 */
1112 static void ceph_d_release(struct dentry *dentry)
1113 {
1114 struct ceph_dentry_info *di = ceph_dentry(dentry);
1115
1116 dout("d_release %p\n", dentry);
1117 ceph_dentry_lru_del(dentry);
1118 if (di->lease_session)
1119 ceph_put_mds_session(di->lease_session);
1120 kmem_cache_free(ceph_dentry_cachep, di);
1121 dentry->d_fsdata = NULL;
1122 }
1123
1124 static int ceph_snapdir_d_revalidate(struct dentry *dentry,
1125 unsigned int flags)
1126 {
1127 /*
1128 * Eventually, we'll want to revalidate snapped metadata
1129 * too... probably...
1130 */
1131 return 1;
1132 }
1133
1134 /*
1135 * When the VFS prunes a dentry from the cache, we need to clear the
1136 * complete flag on the parent directory.
1137 *
1138 * Called under dentry->d_lock.
1139 */
1140 static void ceph_d_prune(struct dentry *dentry)
1141 {
1142 dout("ceph_d_prune %p\n", dentry);
1143
1144 /* do we have a valid parent? */
1145 if (IS_ROOT(dentry))
1146 return;
1147
1148 /* if we are not hashed, we don't affect dir's completeness */
1149 if (d_unhashed(dentry))
1150 return;
1151
1152 /*
1153 * we hold d_lock, so d_parent is stable, and d_fsdata is never
1154 * cleared until d_release
1155 */
1156 ceph_dir_clear_complete(dentry->d_parent->d_inode);
1157 }
1158
1159 /*
1160 * read() on a dir. This weird interface hack only works if mounted
1161 * with '-o dirstat'.
1162 */
1163 static ssize_t ceph_read_dir(struct file *file, char __user *buf, size_t size,
1164 loff_t *ppos)
1165 {
1166 struct ceph_file_info *cf = file->private_data;
1167 struct inode *inode = file_inode(file);
1168 struct ceph_inode_info *ci = ceph_inode(inode);
1169 int left;
1170 const int bufsize = 1024;
1171
1172 if (!ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb), DIRSTAT))
1173 return -EISDIR;
1174
1175 if (!cf->dir_info) {
1176 cf->dir_info = kmalloc(bufsize, GFP_NOFS);
1177 if (!cf->dir_info)
1178 return -ENOMEM;
1179 cf->dir_info_len =
1180 snprintf(cf->dir_info, bufsize,
1181 "entries: %20lld\n"
1182 " files: %20lld\n"
1183 " subdirs: %20lld\n"
1184 "rentries: %20lld\n"
1185 " rfiles: %20lld\n"
1186 " rsubdirs: %20lld\n"
1187 "rbytes: %20lld\n"
1188 "rctime: %10ld.%09ld\n",
1189 ci->i_files + ci->i_subdirs,
1190 ci->i_files,
1191 ci->i_subdirs,
1192 ci->i_rfiles + ci->i_rsubdirs,
1193 ci->i_rfiles,
1194 ci->i_rsubdirs,
1195 ci->i_rbytes,
1196 (long)ci->i_rctime.tv_sec,
1197 (long)ci->i_rctime.tv_nsec);
1198 }
1199
1200 if (*ppos >= cf->dir_info_len)
1201 return 0;
1202 size = min_t(unsigned, size, cf->dir_info_len-*ppos);
1203 left = copy_to_user(buf, cf->dir_info + *ppos, size);
1204 if (left == size)
1205 return -EFAULT;
1206 *ppos += (size - left);
1207 return size - left;
1208 }
1209
1210 /*
1211 * an fsync() on a dir will wait for any uncommitted directory
1212 * operations to commit.
1213 */
1214 static int ceph_dir_fsync(struct file *file, loff_t start, loff_t end,
1215 int datasync)
1216 {
1217 struct inode *inode = file_inode(file);
1218 struct ceph_inode_info *ci = ceph_inode(inode);
1219 struct list_head *head = &ci->i_unsafe_dirops;
1220 struct ceph_mds_request *req;
1221 u64 last_tid;
1222 int ret = 0;
1223
1224 dout("dir_fsync %p\n", inode);
1225 ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
1226 if (ret)
1227 return ret;
1228 mutex_lock(&inode->i_mutex);
1229
1230 spin_lock(&ci->i_unsafe_lock);
1231 if (list_empty(head))
1232 goto out;
1233
1234 req = list_entry(head->prev,
1235 struct ceph_mds_request, r_unsafe_dir_item);
1236 last_tid = req->r_tid;
1237
1238 do {
1239 ceph_mdsc_get_request(req);
1240 spin_unlock(&ci->i_unsafe_lock);
1241
1242 dout("dir_fsync %p wait on tid %llu (until %llu)\n",
1243 inode, req->r_tid, last_tid);
1244 if (req->r_timeout) {
1245 unsigned long time_left = wait_for_completion_timeout(
1246 &req->r_safe_completion,
1247 req->r_timeout);
1248 if (time_left > 0)
1249 ret = 0;
1250 else
1251 ret = -EIO; /* timed out */
1252 } else {
1253 wait_for_completion(&req->r_safe_completion);
1254 }
1255 ceph_mdsc_put_request(req);
1256
1257 spin_lock(&ci->i_unsafe_lock);
1258 if (ret || list_empty(head))
1259 break;
1260 req = list_entry(head->next,
1261 struct ceph_mds_request, r_unsafe_dir_item);
1262 } while (req->r_tid < last_tid);
1263 out:
1264 spin_unlock(&ci->i_unsafe_lock);
1265 mutex_unlock(&inode->i_mutex);
1266
1267 return ret;
1268 }
1269
1270 /*
1271 * We maintain a private dentry LRU.
1272 *
1273 * FIXME: this needs to be changed to a per-mds lru to be useful.
1274 */
1275 void ceph_dentry_lru_add(struct dentry *dn)
1276 {
1277 struct ceph_dentry_info *di = ceph_dentry(dn);
1278 struct ceph_mds_client *mdsc;
1279
1280 dout("dentry_lru_add %p %p '%pd'\n", di, dn, dn);
1281 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1282 spin_lock(&mdsc->dentry_lru_lock);
1283 list_add_tail(&di->lru, &mdsc->dentry_lru);
1284 mdsc->num_dentry++;
1285 spin_unlock(&mdsc->dentry_lru_lock);
1286 }
1287
1288 void ceph_dentry_lru_touch(struct dentry *dn)
1289 {
1290 struct ceph_dentry_info *di = ceph_dentry(dn);
1291 struct ceph_mds_client *mdsc;
1292
1293 dout("dentry_lru_touch %p %p '%pd' (offset %lld)\n", di, dn, dn,
1294 di->offset);
1295 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1296 spin_lock(&mdsc->dentry_lru_lock);
1297 list_move_tail(&di->lru, &mdsc->dentry_lru);
1298 spin_unlock(&mdsc->dentry_lru_lock);
1299 }
1300
1301 void ceph_dentry_lru_del(struct dentry *dn)
1302 {
1303 struct ceph_dentry_info *di = ceph_dentry(dn);
1304 struct ceph_mds_client *mdsc;
1305
1306 dout("dentry_lru_del %p %p '%pd'\n", di, dn, dn);
1307 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1308 spin_lock(&mdsc->dentry_lru_lock);
1309 list_del_init(&di->lru);
1310 mdsc->num_dentry--;
1311 spin_unlock(&mdsc->dentry_lru_lock);
1312 }
1313
1314 /*
1315 * Return name hash for a given dentry. This is dependent on
1316 * the parent directory's hash function.
1317 */
1318 unsigned ceph_dentry_hash(struct inode *dir, struct dentry *dn)
1319 {
1320 struct ceph_inode_info *dci = ceph_inode(dir);
1321
1322 switch (dci->i_dir_layout.dl_dir_hash) {
1323 case 0: /* for backward compat */
1324 case CEPH_STR_HASH_LINUX:
1325 return dn->d_name.hash;
1326
1327 default:
1328 return ceph_str_hash(dci->i_dir_layout.dl_dir_hash,
1329 dn->d_name.name, dn->d_name.len);
1330 }
1331 }
1332
1333 const struct file_operations ceph_dir_fops = {
1334 .read = ceph_read_dir,
1335 .iterate = ceph_readdir,
1336 .llseek = ceph_dir_llseek,
1337 .open = ceph_open,
1338 .release = ceph_release,
1339 .unlocked_ioctl = ceph_ioctl,
1340 .fsync = ceph_dir_fsync,
1341 };
1342
1343 const struct file_operations ceph_snapdir_fops = {
1344 .iterate = ceph_readdir,
1345 .llseek = ceph_dir_llseek,
1346 .open = ceph_open,
1347 .release = ceph_release,
1348 };
1349
1350 const struct inode_operations ceph_dir_iops = {
1351 .lookup = ceph_lookup,
1352 .permission = ceph_permission,
1353 .getattr = ceph_getattr,
1354 .setattr = ceph_setattr,
1355 .setxattr = ceph_setxattr,
1356 .getxattr = ceph_getxattr,
1357 .listxattr = ceph_listxattr,
1358 .removexattr = ceph_removexattr,
1359 .get_acl = ceph_get_acl,
1360 .set_acl = ceph_set_acl,
1361 .mknod = ceph_mknod,
1362 .symlink = ceph_symlink,
1363 .mkdir = ceph_mkdir,
1364 .link = ceph_link,
1365 .unlink = ceph_unlink,
1366 .rmdir = ceph_unlink,
1367 .rename = ceph_rename,
1368 .create = ceph_create,
1369 .atomic_open = ceph_atomic_open,
1370 };
1371
1372 const struct inode_operations ceph_snapdir_iops = {
1373 .lookup = ceph_lookup,
1374 .permission = ceph_permission,
1375 .getattr = ceph_getattr,
1376 .mkdir = ceph_mkdir,
1377 .rmdir = ceph_unlink,
1378 };
1379
1380 const struct dentry_operations ceph_dentry_ops = {
1381 .d_revalidate = ceph_d_revalidate,
1382 .d_release = ceph_d_release,
1383 .d_prune = ceph_d_prune,
1384 };
1385
1386 const struct dentry_operations ceph_snapdir_dentry_ops = {
1387 .d_revalidate = ceph_snapdir_d_revalidate,
1388 .d_release = ceph_d_release,
1389 };
1390
1391 const struct dentry_operations ceph_snap_dentry_ops = {
1392 .d_release = ceph_d_release,
1393 .d_prune = ceph_d_prune,
1394 };
ubuntu-bionic-kernel mirror