]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - fs/ceph/dir.c
projects / mirror_ubuntu-bionic-kernel.git / blob
? search:


60ee531eece400887fc52a52a2596821d5b1382d
[mirror_ubuntu-bionic-kernel.git] / fs / ceph / dir.c
1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/ceph/ceph_debug.h>
3
4 #include <linux/spinlock.h>
5 #include <linux/fs_struct.h>
6 #include <linux/namei.h>
7 #include <linux/slab.h>
8 #include <linux/sched.h>
9 #include <linux/xattr.h>
10
11 #include "super.h"
12 #include "mds_client.h"
13
14 /*
15 * Directory operations: readdir, lookup, create, link, unlink,
16 * rename, etc.
17 */
18
19 /*
20 * Ceph MDS operations are specified in terms of a base ino and
21 * relative path. Thus, the client can specify an operation on a
22 * specific inode (e.g., a getattr due to fstat(2)), or as a path
23 * relative to, say, the root directory.
24 *
25 * Normally, we limit ourselves to strict inode ops (no path component)
26 * or dentry operations (a single path component relative to an ino). The
27 * exception to this is open_root_dentry(), which will open the mount
28 * point by name.
29 */
30
31 const struct dentry_operations ceph_dentry_ops;
32
33 /*
34 * Initialize ceph dentry state.
35 */
36 static int ceph_d_init(struct dentry *dentry)
37 {
38 struct ceph_dentry_info *di;
39
40 di = kmem_cache_zalloc(ceph_dentry_cachep, GFP_KERNEL);
41 if (!di)
42 return -ENOMEM; /* oh well */
43
44 di->dentry = dentry;
45 di->lease_session = NULL;
46 di->time = jiffies;
47 dentry->d_fsdata = di;
48 ceph_dentry_lru_add(dentry);
49 return 0;
50 }
51
52 /*
53 * for f_pos for readdir:
54 * - hash order:
55 * (0xff << 52) | ((24 bits hash) << 28) |
56 * (the nth entry has hash collision);
57 * - frag+name order;
58 * ((frag value) << 28) | (the nth entry in frag);
59 */
60 #define OFFSET_BITS 28
61 #define OFFSET_MASK ((1 << OFFSET_BITS) - 1)
62 #define HASH_ORDER (0xffull << (OFFSET_BITS + 24))
63 loff_t ceph_make_fpos(unsigned high, unsigned off, bool hash_order)
64 {
65 loff_t fpos = ((loff_t)high << 28) | (loff_t)off;
66 if (hash_order)
67 fpos |= HASH_ORDER;
68 return fpos;
69 }
70
71 static bool is_hash_order(loff_t p)
72 {
73 return (p & HASH_ORDER) == HASH_ORDER;
74 }
75
76 static unsigned fpos_frag(loff_t p)
77 {
78 return p >> OFFSET_BITS;
79 }
80
81 static unsigned fpos_hash(loff_t p)
82 {
83 return ceph_frag_value(fpos_frag(p));
84 }
85
86 static unsigned fpos_off(loff_t p)
87 {
88 return p & OFFSET_MASK;
89 }
90
91 static int fpos_cmp(loff_t l, loff_t r)
92 {
93 int v = ceph_frag_compare(fpos_frag(l), fpos_frag(r));
94 if (v)
95 return v;
96 return (int)(fpos_off(l) - fpos_off(r));
97 }
98
99 /*
100 * make note of the last dentry we read, so we can
101 * continue at the same lexicographical point,
102 * regardless of what dir changes take place on the
103 * server.
104 */
105 static int note_last_dentry(struct ceph_file_info *fi, const char *name,
106 int len, unsigned next_offset)
107 {
108 char *buf = kmalloc(len+1, GFP_KERNEL);
109 if (!buf)
110 return -ENOMEM;
111 kfree(fi->last_name);
112 fi->last_name = buf;
113 memcpy(fi->last_name, name, len);
114 fi->last_name[len] = 0;
115 fi->next_offset = next_offset;
116 dout("note_last_dentry '%s'\n", fi->last_name);
117 return 0;
118 }
119
120
121 static struct dentry *
122 __dcache_find_get_entry(struct dentry *parent, u64 idx,
123 struct ceph_readdir_cache_control *cache_ctl)
124 {
125 struct inode *dir = d_inode(parent);
126 struct dentry *dentry;
127 unsigned idx_mask = (PAGE_SIZE / sizeof(struct dentry *)) - 1;
128 loff_t ptr_pos = idx * sizeof(struct dentry *);
129 pgoff_t ptr_pgoff = ptr_pos >> PAGE_SHIFT;
130
131 if (ptr_pos >= i_size_read(dir))
132 return NULL;
133
134 if (!cache_ctl->page || ptr_pgoff != page_index(cache_ctl->page)) {
135 ceph_readdir_cache_release(cache_ctl);
136 cache_ctl->page = find_lock_page(&dir->i_data, ptr_pgoff);
137 if (!cache_ctl->page) {
138 dout(" page %lu not found\n", ptr_pgoff);
139 return ERR_PTR(-EAGAIN);
140 }
141 /* reading/filling the cache are serialized by
142 i_mutex, no need to use page lock */
143 unlock_page(cache_ctl->page);
144 cache_ctl->dentries = kmap(cache_ctl->page);
145 }
146
147 cache_ctl->index = idx & idx_mask;
148
149 rcu_read_lock();
150 spin_lock(&parent->d_lock);
151 /* check i_size again here, because empty directory can be
152 * marked as complete while not holding the i_mutex. */
153 if (ceph_dir_is_complete_ordered(dir) && ptr_pos < i_size_read(dir))
154 dentry = cache_ctl->dentries[cache_ctl->index];
155 else
156 dentry = NULL;
157 spin_unlock(&parent->d_lock);
158 if (dentry && !lockref_get_not_dead(&dentry->d_lockref))
159 dentry = NULL;
160 rcu_read_unlock();
161 return dentry ? : ERR_PTR(-EAGAIN);
162 }
163
164 /*
165 * When possible, we try to satisfy a readdir by peeking at the
166 * dcache. We make this work by carefully ordering dentries on
167 * d_child when we initially get results back from the MDS, and
168 * falling back to a "normal" sync readdir if any dentries in the dir
169 * are dropped.
170 *
171 * Complete dir indicates that we have all dentries in the dir. It is
172 * defined IFF we hold CEPH_CAP_FILE_SHARED (which will be revoked by
173 * the MDS if/when the directory is modified).
174 */
175 static int __dcache_readdir(struct file *file, struct dir_context *ctx,
176 int shared_gen)
177 {
178 struct ceph_file_info *fi = file->private_data;
179 struct dentry *parent = file->f_path.dentry;
180 struct inode *dir = d_inode(parent);
181 struct dentry *dentry, *last = NULL;
182 struct ceph_dentry_info *di;
183 struct ceph_readdir_cache_control cache_ctl = {};
184 u64 idx = 0;
185 int err = 0;
186
187 dout("__dcache_readdir %p v%u at %llx\n", dir, (unsigned)shared_gen, ctx->pos);
188
189 /* search start position */
190 if (ctx->pos > 2) {
191 u64 count = div_u64(i_size_read(dir), sizeof(struct dentry *));
192 while (count > 0) {
193 u64 step = count >> 1;
194 dentry = __dcache_find_get_entry(parent, idx + step,
195 &cache_ctl);
196 if (!dentry) {
197 /* use linar search */
198 idx = 0;
199 break;
200 }
201 if (IS_ERR(dentry)) {
202 err = PTR_ERR(dentry);
203 goto out;
204 }
205 di = ceph_dentry(dentry);
206 spin_lock(&dentry->d_lock);
207 if (fpos_cmp(di->offset, ctx->pos) < 0) {
208 idx += step + 1;
209 count -= step + 1;
210 } else {
211 count = step;
212 }
213 spin_unlock(&dentry->d_lock);
214 dput(dentry);
215 }
216
217 dout("__dcache_readdir %p cache idx %llu\n", dir, idx);
218 }
219
220
221 for (;;) {
222 bool emit_dentry = false;
223 dentry = __dcache_find_get_entry(parent, idx++, &cache_ctl);
224 if (!dentry) {
225 fi->flags |= CEPH_F_ATEND;
226 err = 0;
227 break;
228 }
229 if (IS_ERR(dentry)) {
230 err = PTR_ERR(dentry);
231 goto out;
232 }
233
234 spin_lock(&dentry->d_lock);
235 di = ceph_dentry(dentry);
236 if (d_unhashed(dentry) ||
237 d_really_is_negative(dentry) ||
238 di->lease_shared_gen != shared_gen) {
239 spin_unlock(&dentry->d_lock);
240 dput(dentry);
241 err = -EAGAIN;
242 goto out;
243 }
244 if (fpos_cmp(ctx->pos, di->offset) <= 0) {
245 emit_dentry = true;
246 }
247 spin_unlock(&dentry->d_lock);
248
249 if (emit_dentry) {
250 dout(" %llx dentry %p %pd %p\n", di->offset,
251 dentry, dentry, d_inode(dentry));
252 ctx->pos = di->offset;
253 if (!dir_emit(ctx, dentry->d_name.name,
254 dentry->d_name.len,
255 ceph_translate_ino(dentry->d_sb,
256 d_inode(dentry)->i_ino),
257 d_inode(dentry)->i_mode >> 12)) {
258 dput(dentry);
259 err = 0;
260 break;
261 }
262 ctx->pos++;
263
264 if (last)
265 dput(last);
266 last = dentry;
267 } else {
268 dput(dentry);
269 }
270 }
271 out:
272 ceph_readdir_cache_release(&cache_ctl);
273 if (last) {
274 int ret;
275 di = ceph_dentry(last);
276 ret = note_last_dentry(fi, last->d_name.name, last->d_name.len,
277 fpos_off(di->offset) + 1);
278 if (ret < 0)
279 err = ret;
280 dput(last);
281 /* last_name no longer match cache index */
282 if (fi->readdir_cache_idx >= 0) {
283 fi->readdir_cache_idx = -1;
284 fi->dir_release_count = 0;
285 }
286 }
287 return err;
288 }
289
290 static bool need_send_readdir(struct ceph_file_info *fi, loff_t pos)
291 {
292 if (!fi->last_readdir)
293 return true;
294 if (is_hash_order(pos))
295 return !ceph_frag_contains_value(fi->frag, fpos_hash(pos));
296 else
297 return fi->frag != fpos_frag(pos);
298 }
299
300 static int ceph_readdir(struct file *file, struct dir_context *ctx)
301 {
302 struct ceph_file_info *fi = file->private_data;
303 struct inode *inode = file_inode(file);
304 struct ceph_inode_info *ci = ceph_inode(inode);
305 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
306 struct ceph_mds_client *mdsc = fsc->mdsc;
307 int i;
308 int err;
309 unsigned frag = -1;
310 struct ceph_mds_reply_info_parsed *rinfo;
311
312 dout("readdir %p file %p pos %llx\n", inode, file, ctx->pos);
313 if (fi->flags & CEPH_F_ATEND)
314 return 0;
315
316 /* always start with . and .. */
317 if (ctx->pos == 0) {
318 dout("readdir off 0 -> '.'\n");
319 if (!dir_emit(ctx, ".", 1,
320 ceph_translate_ino(inode->i_sb, inode->i_ino),
321 inode->i_mode >> 12))
322 return 0;
323 ctx->pos = 1;
324 }
325 if (ctx->pos == 1) {
326 ino_t ino = parent_ino(file->f_path.dentry);
327 dout("readdir off 1 -> '..'\n");
328 if (!dir_emit(ctx, "..", 2,
329 ceph_translate_ino(inode->i_sb, ino),
330 inode->i_mode >> 12))
331 return 0;
332 ctx->pos = 2;
333 }
334
335 /* can we use the dcache? */
336 spin_lock(&ci->i_ceph_lock);
337 if (ceph_test_mount_opt(fsc, DCACHE) &&
338 !ceph_test_mount_opt(fsc, NOASYNCREADDIR) &&
339 ceph_snap(inode) != CEPH_SNAPDIR &&
340 __ceph_dir_is_complete_ordered(ci) &&
341 __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1)) {
342 int shared_gen = atomic_read(&ci->i_shared_gen);
343 spin_unlock(&ci->i_ceph_lock);
344 err = __dcache_readdir(file, ctx, shared_gen);
345 if (err != -EAGAIN)
346 return err;
347 } else {
348 spin_unlock(&ci->i_ceph_lock);
349 }
350
351 /* proceed with a normal readdir */
352 more:
353 /* do we have the correct frag content buffered? */
354 if (need_send_readdir(fi, ctx->pos)) {
355 struct ceph_mds_request *req;
356 int op = ceph_snap(inode) == CEPH_SNAPDIR ?
357 CEPH_MDS_OP_LSSNAP : CEPH_MDS_OP_READDIR;
358
359 /* discard old result, if any */
360 if (fi->last_readdir) {
361 ceph_mdsc_put_request(fi->last_readdir);
362 fi->last_readdir = NULL;
363 }
364
365 if (is_hash_order(ctx->pos)) {
366 /* fragtree isn't always accurate. choose frag
367 * based on previous reply when possible. */
368 if (frag == (unsigned)-1)
369 frag = ceph_choose_frag(ci, fpos_hash(ctx->pos),
370 NULL, NULL);
371 } else {
372 frag = fpos_frag(ctx->pos);
373 }
374
375 dout("readdir fetching %llx.%llx frag %x offset '%s'\n",
376 ceph_vinop(inode), frag, fi->last_name);
377 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
378 if (IS_ERR(req))
379 return PTR_ERR(req);
380 err = ceph_alloc_readdir_reply_buffer(req, inode);
381 if (err) {
382 ceph_mdsc_put_request(req);
383 return err;
384 }
385 /* hints to request -> mds selection code */
386 req->r_direct_mode = USE_AUTH_MDS;
387 if (op == CEPH_MDS_OP_READDIR) {
388 req->r_direct_hash = ceph_frag_value(frag);
389 __set_bit(CEPH_MDS_R_DIRECT_IS_HASH, &req->r_req_flags);
390 }
391 if (fi->last_name) {
392 req->r_path2 = kstrdup(fi->last_name, GFP_KERNEL);
393 if (!req->r_path2) {
394 ceph_mdsc_put_request(req);
395 return -ENOMEM;
396 }
397 } else if (is_hash_order(ctx->pos)) {
398 req->r_args.readdir.offset_hash =
399 cpu_to_le32(fpos_hash(ctx->pos));
400 }
401
402 req->r_dir_release_cnt = fi->dir_release_count;
403 req->r_dir_ordered_cnt = fi->dir_ordered_count;
404 req->r_readdir_cache_idx = fi->readdir_cache_idx;
405 req->r_readdir_offset = fi->next_offset;
406 req->r_args.readdir.frag = cpu_to_le32(frag);
407 req->r_args.readdir.flags =
408 cpu_to_le16(CEPH_READDIR_REPLY_BITFLAGS);
409
410 req->r_inode = inode;
411 ihold(inode);
412 req->r_dentry = dget(file->f_path.dentry);
413 err = ceph_mdsc_do_request(mdsc, NULL, req);
414 if (err < 0) {
415 ceph_mdsc_put_request(req);
416 return err;
417 }
418 dout("readdir got and parsed readdir result=%d on "
419 "frag %x, end=%d, complete=%d, hash_order=%d\n",
420 err, frag,
421 (int)req->r_reply_info.dir_end,
422 (int)req->r_reply_info.dir_complete,
423 (int)req->r_reply_info.hash_order);
424
425 rinfo = &req->r_reply_info;
426 if (le32_to_cpu(rinfo->dir_dir->frag) != frag) {
427 frag = le32_to_cpu(rinfo->dir_dir->frag);
428 if (!rinfo->hash_order) {
429 fi->next_offset = req->r_readdir_offset;
430 /* adjust ctx->pos to beginning of frag */
431 ctx->pos = ceph_make_fpos(frag,
432 fi->next_offset,
433 false);
434 }
435 }
436
437 fi->frag = frag;
438 fi->last_readdir = req;
439
440 if (test_bit(CEPH_MDS_R_DID_PREPOPULATE, &req->r_req_flags)) {
441 fi->readdir_cache_idx = req->r_readdir_cache_idx;
442 if (fi->readdir_cache_idx < 0) {
443 /* preclude from marking dir ordered */
444 fi->dir_ordered_count = 0;
445 } else if (ceph_frag_is_leftmost(frag) &&
446 fi->next_offset == 2) {
447 /* note dir version at start of readdir so
448 * we can tell if any dentries get dropped */
449 fi->dir_release_count = req->r_dir_release_cnt;
450 fi->dir_ordered_count = req->r_dir_ordered_cnt;
451 }
452 } else {
453 dout("readdir !did_prepopulate");
454 /* disable readdir cache */
455 fi->readdir_cache_idx = -1;
456 /* preclude from marking dir complete */
457 fi->dir_release_count = 0;
458 }
459
460 /* note next offset and last dentry name */
461 if (rinfo->dir_nr > 0) {
462 struct ceph_mds_reply_dir_entry *rde =
463 rinfo->dir_entries + (rinfo->dir_nr-1);
464 unsigned next_offset = req->r_reply_info.dir_end ?
465 2 : (fpos_off(rde->offset) + 1);
466 err = note_last_dentry(fi, rde->name, rde->name_len,
467 next_offset);
468 if (err)
469 return err;
470 } else if (req->r_reply_info.dir_end) {
471 fi->next_offset = 2;
472 /* keep last name */
473 }
474 }
475
476 rinfo = &fi->last_readdir->r_reply_info;
477 dout("readdir frag %x num %d pos %llx chunk first %llx\n",
478 fi->frag, rinfo->dir_nr, ctx->pos,
479 rinfo->dir_nr ? rinfo->dir_entries[0].offset : 0LL);
480
481 i = 0;
482 /* search start position */
483 if (rinfo->dir_nr > 0) {
484 int step, nr = rinfo->dir_nr;
485 while (nr > 0) {
486 step = nr >> 1;
487 if (rinfo->dir_entries[i + step].offset < ctx->pos) {
488 i += step + 1;
489 nr -= step + 1;
490 } else {
491 nr = step;
492 }
493 }
494 }
495 for (; i < rinfo->dir_nr; i++) {
496 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
497 struct ceph_vino vino;
498 ino_t ino;
499 u32 ftype;
500
501 BUG_ON(rde->offset < ctx->pos);
502
503 ctx->pos = rde->offset;
504 dout("readdir (%d/%d) -> %llx '%.*s' %p\n",
505 i, rinfo->dir_nr, ctx->pos,
506 rde->name_len, rde->name, &rde->inode.in);
507
508 BUG_ON(!rde->inode.in);
509 ftype = le32_to_cpu(rde->inode.in->mode) >> 12;
510 vino.ino = le64_to_cpu(rde->inode.in->ino);
511 vino.snap = le64_to_cpu(rde->inode.in->snapid);
512 ino = ceph_vino_to_ino(vino);
513
514 if (!dir_emit(ctx, rde->name, rde->name_len,
515 ceph_translate_ino(inode->i_sb, ino), ftype)) {
516 dout("filldir stopping us...\n");
517 return 0;
518 }
519 ctx->pos++;
520 }
521
522 ceph_mdsc_put_request(fi->last_readdir);
523 fi->last_readdir = NULL;
524
525 if (fi->next_offset > 2) {
526 frag = fi->frag;
527 goto more;
528 }
529
530 /* more frags? */
531 if (!ceph_frag_is_rightmost(fi->frag)) {
532 frag = ceph_frag_next(fi->frag);
533 if (is_hash_order(ctx->pos)) {
534 loff_t new_pos = ceph_make_fpos(ceph_frag_value(frag),
535 fi->next_offset, true);
536 if (new_pos > ctx->pos)
537 ctx->pos = new_pos;
538 /* keep last_name */
539 } else {
540 ctx->pos = ceph_make_fpos(frag, fi->next_offset, false);
541 kfree(fi->last_name);
542 fi->last_name = NULL;
543 }
544 dout("readdir next frag is %x\n", frag);
545 goto more;
546 }
547 fi->flags |= CEPH_F_ATEND;
548
549 /*
550 * if dir_release_count still matches the dir, no dentries
551 * were released during the whole readdir, and we should have
552 * the complete dir contents in our cache.
553 */
554 if (atomic64_read(&ci->i_release_count) == fi->dir_release_count) {
555 spin_lock(&ci->i_ceph_lock);
556 if (fi->dir_ordered_count == atomic64_read(&ci->i_ordered_count)) {
557 dout(" marking %p complete and ordered\n", inode);
558 /* use i_size to track number of entries in
559 * readdir cache */
560 BUG_ON(fi->readdir_cache_idx < 0);
561 i_size_write(inode, fi->readdir_cache_idx *
562 sizeof(struct dentry*));
563 } else {
564 dout(" marking %p complete\n", inode);
565 }
566 __ceph_dir_set_complete(ci, fi->dir_release_count,
567 fi->dir_ordered_count);
568 spin_unlock(&ci->i_ceph_lock);
569 }
570
571 dout("readdir %p file %p done.\n", inode, file);
572 return 0;
573 }
574
575 static void reset_readdir(struct ceph_file_info *fi)
576 {
577 if (fi->last_readdir) {
578 ceph_mdsc_put_request(fi->last_readdir);
579 fi->last_readdir = NULL;
580 }
581 kfree(fi->last_name);
582 fi->last_name = NULL;
583 fi->dir_release_count = 0;
584 fi->readdir_cache_idx = -1;
585 fi->next_offset = 2; /* compensate for . and .. */
586 fi->flags &= ~CEPH_F_ATEND;
587 }
588
589 /*
590 * discard buffered readdir content on seekdir(0), or seek to new frag,
591 * or seek prior to current chunk
592 */
593 static bool need_reset_readdir(struct ceph_file_info *fi, loff_t new_pos)
594 {
595 struct ceph_mds_reply_info_parsed *rinfo;
596 loff_t chunk_offset;
597 if (new_pos == 0)
598 return true;
599 if (is_hash_order(new_pos)) {
600 /* no need to reset last_name for a forward seek when
601 * dentries are sotred in hash order */
602 } else if (fi->frag != fpos_frag(new_pos)) {
603 return true;
604 }
605 rinfo = fi->last_readdir ? &fi->last_readdir->r_reply_info : NULL;
606 if (!rinfo || !rinfo->dir_nr)
607 return true;
608 chunk_offset = rinfo->dir_entries[0].offset;
609 return new_pos < chunk_offset ||
610 is_hash_order(new_pos) != is_hash_order(chunk_offset);
611 }
612
613 static loff_t ceph_dir_llseek(struct file *file, loff_t offset, int whence)
614 {
615 struct ceph_file_info *fi = file->private_data;
616 struct inode *inode = file->f_mapping->host;
617 loff_t retval;
618
619 inode_lock(inode);
620 retval = -EINVAL;
621 switch (whence) {
622 case SEEK_CUR:
623 offset += file->f_pos;
624 case SEEK_SET:
625 break;
626 case SEEK_END:
627 retval = -EOPNOTSUPP;
628 default:
629 goto out;
630 }
631
632 if (offset >= 0) {
633 if (need_reset_readdir(fi, offset)) {
634 dout("dir_llseek dropping %p content\n", file);
635 reset_readdir(fi);
636 } else if (is_hash_order(offset) && offset > file->f_pos) {
637 /* for hash offset, we don't know if a forward seek
638 * is within same frag */
639 fi->dir_release_count = 0;
640 fi->readdir_cache_idx = -1;
641 }
642
643 if (offset != file->f_pos) {
644 file->f_pos = offset;
645 file->f_version = 0;
646 fi->flags &= ~CEPH_F_ATEND;
647 }
648 retval = offset;
649 }
650 out:
651 inode_unlock(inode);
652 return retval;
653 }
654
655 /*
656 * Handle lookups for the hidden .snap directory.
657 */
658 int ceph_handle_snapdir(struct ceph_mds_request *req,
659 struct dentry *dentry, int err)
660 {
661 struct ceph_fs_client *fsc = ceph_sb_to_client(dentry->d_sb);
662 struct inode *parent = d_inode(dentry->d_parent); /* we hold i_mutex */
663
664 /* .snap dir? */
665 if (err == -ENOENT &&
666 ceph_snap(parent) == CEPH_NOSNAP &&
667 strcmp(dentry->d_name.name,
668 fsc->mount_options->snapdir_name) == 0) {
669 struct inode *inode = ceph_get_snapdir(parent);
670 dout("ENOENT on snapdir %p '%pd', linking to snapdir %p\n",
671 dentry, dentry, inode);
672 BUG_ON(!d_unhashed(dentry));
673 d_add(dentry, inode);
674 err = 0;
675 }
676 return err;
677 }
678
679 /*
680 * Figure out final result of a lookup/open request.
681 *
682 * Mainly, make sure we return the final req->r_dentry (if it already
683 * existed) in place of the original VFS-provided dentry when they
684 * differ.
685 *
686 * Gracefully handle the case where the MDS replies with -ENOENT and
687 * no trace (which it may do, at its discretion, e.g., if it doesn't
688 * care to issue a lease on the negative dentry).
689 */
690 struct dentry *ceph_finish_lookup(struct ceph_mds_request *req,
691 struct dentry *dentry, int err)
692 {
693 if (err == -ENOENT) {
694 /* no trace? */
695 err = 0;
696 if (!req->r_reply_info.head->is_dentry) {
697 dout("ENOENT and no trace, dentry %p inode %p\n",
698 dentry, d_inode(dentry));
699 if (d_really_is_positive(dentry)) {
700 d_drop(dentry);
701 err = -ENOENT;
702 } else {
703 d_add(dentry, NULL);
704 }
705 }
706 }
707 if (err)
708 dentry = ERR_PTR(err);
709 else if (dentry != req->r_dentry)
710 dentry = dget(req->r_dentry); /* we got spliced */
711 else
712 dentry = NULL;
713 return dentry;
714 }
715
716 static bool is_root_ceph_dentry(struct inode *inode, struct dentry *dentry)
717 {
718 return ceph_ino(inode) == CEPH_INO_ROOT &&
719 strncmp(dentry->d_name.name, ".ceph", 5) == 0;
720 }
721
722 /*
723 * Look up a single dir entry. If there is a lookup intent, inform
724 * the MDS so that it gets our 'caps wanted' value in a single op.
725 */
726 static struct dentry *ceph_lookup(struct inode *dir, struct dentry *dentry,
727 unsigned int flags)
728 {
729 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
730 struct ceph_mds_client *mdsc = fsc->mdsc;
731 struct ceph_mds_request *req;
732 int op;
733 int mask;
734 int err;
735
736 dout("lookup %p dentry %p '%pd'\n",
737 dir, dentry, dentry);
738
739 if (dentry->d_name.len > NAME_MAX)
740 return ERR_PTR(-ENAMETOOLONG);
741
742 /* can we conclude ENOENT locally? */
743 if (d_really_is_negative(dentry)) {
744 struct ceph_inode_info *ci = ceph_inode(dir);
745 struct ceph_dentry_info *di = ceph_dentry(dentry);
746
747 spin_lock(&ci->i_ceph_lock);
748 dout(" dir %p flags are %d\n", dir, ci->i_ceph_flags);
749 if (strncmp(dentry->d_name.name,
750 fsc->mount_options->snapdir_name,
751 dentry->d_name.len) &&
752 !is_root_ceph_dentry(dir, dentry) &&
753 ceph_test_mount_opt(fsc, DCACHE) &&
754 __ceph_dir_is_complete(ci) &&
755 (__ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1))) {
756 spin_unlock(&ci->i_ceph_lock);
757 dout(" dir %p complete, -ENOENT\n", dir);
758 d_add(dentry, NULL);
759 di->lease_shared_gen = atomic_read(&ci->i_shared_gen);
760 return NULL;
761 }
762 spin_unlock(&ci->i_ceph_lock);
763 }
764
765 op = ceph_snap(dir) == CEPH_SNAPDIR ?
766 CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP;
767 req = ceph_mdsc_create_request(mdsc, op, USE_ANY_MDS);
768 if (IS_ERR(req))
769 return ERR_CAST(req);
770 req->r_dentry = dget(dentry);
771 req->r_num_caps = 2;
772
773 mask = CEPH_STAT_CAP_INODE | CEPH_CAP_AUTH_SHARED;
774 if (ceph_security_xattr_wanted(dir))
775 mask |= CEPH_CAP_XATTR_SHARED;
776 req->r_args.getattr.mask = cpu_to_le32(mask);
777
778 req->r_parent = dir;
779 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
780 err = ceph_mdsc_do_request(mdsc, NULL, req);
781 err = ceph_handle_snapdir(req, dentry, err);
782 dentry = ceph_finish_lookup(req, dentry, err);
783 ceph_mdsc_put_request(req); /* will dput(dentry) */
784 dout("lookup result=%p\n", dentry);
785 return dentry;
786 }
787
788 /*
789 * If we do a create but get no trace back from the MDS, follow up with
790 * a lookup (the VFS expects us to link up the provided dentry).
791 */
792 int ceph_handle_notrace_create(struct inode *dir, struct dentry *dentry)
793 {
794 struct dentry *result = ceph_lookup(dir, dentry, 0);
795
796 if (result && !IS_ERR(result)) {
797 /*
798 * We created the item, then did a lookup, and found
799 * it was already linked to another inode we already
800 * had in our cache (and thus got spliced). To not
801 * confuse VFS (especially when inode is a directory),
802 * we don't link our dentry to that inode, return an
803 * error instead.
804 *
805 * This event should be rare and it happens only when
806 * we talk to old MDS. Recent MDS does not send traceless
807 * reply for request that creates new inode.
808 */
809 d_drop(result);
810 return -ESTALE;
811 }
812 return PTR_ERR(result);
813 }
814
815 static int ceph_mknod(struct inode *dir, struct dentry *dentry,
816 umode_t mode, dev_t rdev)
817 {
818 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
819 struct ceph_mds_client *mdsc = fsc->mdsc;
820 struct ceph_mds_request *req;
821 struct ceph_acls_info acls = {};
822 int err;
823
824 if (ceph_snap(dir) != CEPH_NOSNAP)
825 return -EROFS;
826
827 err = ceph_pre_init_acls(dir, &mode, &acls);
828 if (err < 0)
829 return err;
830
831 dout("mknod in dir %p dentry %p mode 0%ho rdev %d\n",
832 dir, dentry, mode, rdev);
833 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_MKNOD, USE_AUTH_MDS);
834 if (IS_ERR(req)) {
835 err = PTR_ERR(req);
836 goto out;
837 }
838 req->r_dentry = dget(dentry);
839 req->r_num_caps = 2;
840 req->r_parent = dir;
841 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
842 req->r_args.mknod.mode = cpu_to_le32(mode);
843 req->r_args.mknod.rdev = cpu_to_le32(rdev);
844 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
845 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
846 if (acls.pagelist) {
847 req->r_pagelist = acls.pagelist;
848 acls.pagelist = NULL;
849 }
850 err = ceph_mdsc_do_request(mdsc, dir, req);
851 if (!err && !req->r_reply_info.head->is_dentry)
852 err = ceph_handle_notrace_create(dir, dentry);
853 ceph_mdsc_put_request(req);
854 out:
855 if (!err)
856 ceph_init_inode_acls(d_inode(dentry), &acls);
857 else
858 d_drop(dentry);
859 ceph_release_acls_info(&acls);
860 return err;
861 }
862
863 static int ceph_create(struct inode *dir, struct dentry *dentry, umode_t mode,
864 bool excl)
865 {
866 return ceph_mknod(dir, dentry, mode, 0);
867 }
868
869 static int ceph_symlink(struct inode *dir, struct dentry *dentry,
870 const char *dest)
871 {
872 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
873 struct ceph_mds_client *mdsc = fsc->mdsc;
874 struct ceph_mds_request *req;
875 int err;
876
877 if (ceph_snap(dir) != CEPH_NOSNAP)
878 return -EROFS;
879
880 dout("symlink in dir %p dentry %p to '%s'\n", dir, dentry, dest);
881 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SYMLINK, USE_AUTH_MDS);
882 if (IS_ERR(req)) {
883 err = PTR_ERR(req);
884 goto out;
885 }
886 req->r_path2 = kstrdup(dest, GFP_KERNEL);
887 if (!req->r_path2) {
888 err = -ENOMEM;
889 ceph_mdsc_put_request(req);
890 goto out;
891 }
892 req->r_parent = dir;
893 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
894 req->r_dentry = dget(dentry);
895 req->r_num_caps = 2;
896 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
897 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
898 err = ceph_mdsc_do_request(mdsc, dir, req);
899 if (!err && !req->r_reply_info.head->is_dentry)
900 err = ceph_handle_notrace_create(dir, dentry);
901 ceph_mdsc_put_request(req);
902 out:
903 if (err)
904 d_drop(dentry);
905 return err;
906 }
907
908 static int ceph_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
909 {
910 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
911 struct ceph_mds_client *mdsc = fsc->mdsc;
912 struct ceph_mds_request *req;
913 struct ceph_acls_info acls = {};
914 int err = -EROFS;
915 int op;
916
917 if (ceph_snap(dir) == CEPH_SNAPDIR) {
918 /* mkdir .snap/foo is a MKSNAP */
919 op = CEPH_MDS_OP_MKSNAP;
920 dout("mksnap dir %p snap '%pd' dn %p\n", dir,
921 dentry, dentry);
922 } else if (ceph_snap(dir) == CEPH_NOSNAP) {
923 dout("mkdir dir %p dn %p mode 0%ho\n", dir, dentry, mode);
924 op = CEPH_MDS_OP_MKDIR;
925 } else {
926 goto out;
927 }
928
929 mode |= S_IFDIR;
930 err = ceph_pre_init_acls(dir, &mode, &acls);
931 if (err < 0)
932 goto out;
933
934 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
935 if (IS_ERR(req)) {
936 err = PTR_ERR(req);
937 goto out;
938 }
939
940 req->r_dentry = dget(dentry);
941 req->r_num_caps = 2;
942 req->r_parent = dir;
943 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
944 req->r_args.mkdir.mode = cpu_to_le32(mode);
945 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
946 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
947 if (acls.pagelist) {
948 req->r_pagelist = acls.pagelist;
949 acls.pagelist = NULL;
950 }
951 err = ceph_mdsc_do_request(mdsc, dir, req);
952 if (!err &&
953 !req->r_reply_info.head->is_target &&
954 !req->r_reply_info.head->is_dentry)
955 err = ceph_handle_notrace_create(dir, dentry);
956 ceph_mdsc_put_request(req);
957 out:
958 if (!err)
959 ceph_init_inode_acls(d_inode(dentry), &acls);
960 else
961 d_drop(dentry);
962 ceph_release_acls_info(&acls);
963 return err;
964 }
965
966 static int ceph_link(struct dentry *old_dentry, struct inode *dir,
967 struct dentry *dentry)
968 {
969 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
970 struct ceph_mds_client *mdsc = fsc->mdsc;
971 struct ceph_mds_request *req;
972 int err;
973
974 if (ceph_snap(dir) != CEPH_NOSNAP)
975 return -EROFS;
976
977 dout("link in dir %p old_dentry %p dentry %p\n", dir,
978 old_dentry, dentry);
979 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LINK, USE_AUTH_MDS);
980 if (IS_ERR(req)) {
981 d_drop(dentry);
982 return PTR_ERR(req);
983 }
984 req->r_dentry = dget(dentry);
985 req->r_num_caps = 2;
986 req->r_old_dentry = dget(old_dentry);
987 req->r_parent = dir;
988 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
989 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
990 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
991 /* release LINK_SHARED on source inode (mds will lock it) */
992 req->r_old_inode_drop = CEPH_CAP_LINK_SHARED;
993 err = ceph_mdsc_do_request(mdsc, dir, req);
994 if (err) {
995 d_drop(dentry);
996 } else if (!req->r_reply_info.head->is_dentry) {
997 ihold(d_inode(old_dentry));
998 d_instantiate(dentry, d_inode(old_dentry));
999 }
1000 ceph_mdsc_put_request(req);
1001 return err;
1002 }
1003
1004 /*
1005 * For a soon-to-be unlinked file, drop the AUTH_RDCACHE caps. If it
1006 * looks like the link count will hit 0, drop any other caps (other
1007 * than PIN) we don't specifically want (due to the file still being
1008 * open).
1009 */
1010 static int drop_caps_for_unlink(struct inode *inode)
1011 {
1012 struct ceph_inode_info *ci = ceph_inode(inode);
1013 int drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL;
1014
1015 spin_lock(&ci->i_ceph_lock);
1016 if (inode->i_nlink == 1) {
1017 drop |= ~(__ceph_caps_wanted(ci) | CEPH_CAP_PIN);
1018 ci->i_ceph_flags |= CEPH_I_NODELAY;
1019 }
1020 spin_unlock(&ci->i_ceph_lock);
1021 return drop;
1022 }
1023
1024 /*
1025 * rmdir and unlink are differ only by the metadata op code
1026 */
1027 static int ceph_unlink(struct inode *dir, struct dentry *dentry)
1028 {
1029 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
1030 struct ceph_mds_client *mdsc = fsc->mdsc;
1031 struct inode *inode = d_inode(dentry);
1032 struct ceph_mds_request *req;
1033 int err = -EROFS;
1034 int op;
1035
1036 if (ceph_snap(dir) == CEPH_SNAPDIR) {
1037 /* rmdir .snap/foo is RMSNAP */
1038 dout("rmsnap dir %p '%pd' dn %p\n", dir, dentry, dentry);
1039 op = CEPH_MDS_OP_RMSNAP;
1040 } else if (ceph_snap(dir) == CEPH_NOSNAP) {
1041 dout("unlink/rmdir dir %p dn %p inode %p\n",
1042 dir, dentry, inode);
1043 op = d_is_dir(dentry) ?
1044 CEPH_MDS_OP_RMDIR : CEPH_MDS_OP_UNLINK;
1045 } else
1046 goto out;
1047 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
1048 if (IS_ERR(req)) {
1049 err = PTR_ERR(req);
1050 goto out;
1051 }
1052 req->r_dentry = dget(dentry);
1053 req->r_num_caps = 2;
1054 req->r_parent = dir;
1055 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
1056 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
1057 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
1058 req->r_inode_drop = drop_caps_for_unlink(inode);
1059 err = ceph_mdsc_do_request(mdsc, dir, req);
1060 if (!err && !req->r_reply_info.head->is_dentry)
1061 d_delete(dentry);
1062 ceph_mdsc_put_request(req);
1063 out:
1064 return err;
1065 }
1066
1067 static int ceph_rename(struct inode *old_dir, struct dentry *old_dentry,
1068 struct inode *new_dir, struct dentry *new_dentry,
1069 unsigned int flags)
1070 {
1071 struct ceph_fs_client *fsc = ceph_sb_to_client(old_dir->i_sb);
1072 struct ceph_mds_client *mdsc = fsc->mdsc;
1073 struct ceph_mds_request *req;
1074 int op = CEPH_MDS_OP_RENAME;
1075 int err;
1076
1077 if (flags)
1078 return -EINVAL;
1079
1080 if (ceph_snap(old_dir) != ceph_snap(new_dir))
1081 return -EXDEV;
1082 if (ceph_snap(old_dir) != CEPH_NOSNAP) {
1083 if (old_dir == new_dir && ceph_snap(old_dir) == CEPH_SNAPDIR)
1084 op = CEPH_MDS_OP_RENAMESNAP;
1085 else
1086 return -EROFS;
1087 }
1088 dout("rename dir %p dentry %p to dir %p dentry %p\n",
1089 old_dir, old_dentry, new_dir, new_dentry);
1090 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
1091 if (IS_ERR(req))
1092 return PTR_ERR(req);
1093 ihold(old_dir);
1094 req->r_dentry = dget(new_dentry);
1095 req->r_num_caps = 2;
1096 req->r_old_dentry = dget(old_dentry);
1097 req->r_old_dentry_dir = old_dir;
1098 req->r_parent = new_dir;
1099 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
1100 req->r_old_dentry_drop = CEPH_CAP_FILE_SHARED;
1101 req->r_old_dentry_unless = CEPH_CAP_FILE_EXCL;
1102 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
1103 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
1104 /* release LINK_RDCACHE on source inode (mds will lock it) */
1105 req->r_old_inode_drop = CEPH_CAP_LINK_SHARED;
1106 if (d_really_is_positive(new_dentry))
1107 req->r_inode_drop = drop_caps_for_unlink(d_inode(new_dentry));
1108 err = ceph_mdsc_do_request(mdsc, old_dir, req);
1109 if (!err && !req->r_reply_info.head->is_dentry) {
1110 /*
1111 * Normally d_move() is done by fill_trace (called by
1112 * do_request, above). If there is no trace, we need
1113 * to do it here.
1114 */
1115
1116 /* d_move screws up sibling dentries' offsets */
1117 ceph_dir_clear_complete(old_dir);
1118 ceph_dir_clear_complete(new_dir);
1119
1120 d_move(old_dentry, new_dentry);
1121
1122 /* ensure target dentry is invalidated, despite
1123 rehashing bug in vfs_rename_dir */
1124 ceph_invalidate_dentry_lease(new_dentry);
1125 }
1126 ceph_mdsc_put_request(req);
1127 return err;
1128 }
1129
1130 /*
1131 * Ensure a dentry lease will no longer revalidate.
1132 */
1133 void ceph_invalidate_dentry_lease(struct dentry *dentry)
1134 {
1135 spin_lock(&dentry->d_lock);
1136 ceph_dentry(dentry)->time = jiffies;
1137 ceph_dentry(dentry)->lease_shared_gen = 0;
1138 spin_unlock(&dentry->d_lock);
1139 }
1140
1141 /*
1142 * Check if dentry lease is valid. If not, delete the lease. Try to
1143 * renew if the least is more than half up.
1144 */
1145 static int dentry_lease_is_valid(struct dentry *dentry, unsigned int flags,
1146 struct inode *dir)
1147 {
1148 struct ceph_dentry_info *di;
1149 struct ceph_mds_session *s;
1150 int valid = 0;
1151 u32 gen;
1152 unsigned long ttl;
1153 struct ceph_mds_session *session = NULL;
1154 u32 seq = 0;
1155
1156 spin_lock(&dentry->d_lock);
1157 di = ceph_dentry(dentry);
1158 if (di && di->lease_session) {
1159 s = di->lease_session;
1160 spin_lock(&s->s_gen_ttl_lock);
1161 gen = s->s_cap_gen;
1162 ttl = s->s_cap_ttl;
1163 spin_unlock(&s->s_gen_ttl_lock);
1164
1165 if (di->lease_gen == gen &&
1166 time_before(jiffies, di->time) &&
1167 time_before(jiffies, ttl)) {
1168 valid = 1;
1169 if (di->lease_renew_after &&
1170 time_after(jiffies, di->lease_renew_after)) {
1171 /*
1172 * We should renew. If we're in RCU walk mode
1173 * though, we can't do that so just return
1174 * -ECHILD.
1175 */
1176 if (flags & LOOKUP_RCU) {
1177 valid = -ECHILD;
1178 } else {
1179 session = ceph_get_mds_session(s);
1180 seq = di->lease_seq;
1181 di->lease_renew_after = 0;
1182 di->lease_renew_from = jiffies;
1183 }
1184 }
1185 }
1186 }
1187 spin_unlock(&dentry->d_lock);
1188
1189 if (session) {
1190 ceph_mdsc_lease_send_msg(session, dir, dentry,
1191 CEPH_MDS_LEASE_RENEW, seq);
1192 ceph_put_mds_session(session);
1193 }
1194 dout("dentry_lease_is_valid - dentry %p = %d\n", dentry, valid);
1195 return valid;
1196 }
1197
1198 /*
1199 * Check if directory-wide content lease/cap is valid.
1200 */
1201 static int dir_lease_is_valid(struct inode *dir, struct dentry *dentry)
1202 {
1203 struct ceph_inode_info *ci = ceph_inode(dir);
1204 struct ceph_dentry_info *di = ceph_dentry(dentry);
1205 int valid = 0;
1206
1207 spin_lock(&ci->i_ceph_lock);
1208 if (atomic_read(&ci->i_shared_gen) == di->lease_shared_gen)
1209 valid = __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1);
1210 spin_unlock(&ci->i_ceph_lock);
1211 dout("dir_lease_is_valid dir %p v%u dentry %p v%u = %d\n",
1212 dir, (unsigned)atomic_read(&ci->i_shared_gen),
1213 dentry, (unsigned)di->lease_shared_gen, valid);
1214 return valid;
1215 }
1216
1217 /*
1218 * Check if cached dentry can be trusted.
1219 */
1220 static int ceph_d_revalidate(struct dentry *dentry, unsigned int flags)
1221 {
1222 int valid = 0;
1223 struct dentry *parent;
1224 struct inode *dir;
1225
1226 if (flags & LOOKUP_RCU) {
1227 parent = READ_ONCE(dentry->d_parent);
1228 dir = d_inode_rcu(parent);
1229 if (!dir)
1230 return -ECHILD;
1231 } else {
1232 parent = dget_parent(dentry);
1233 dir = d_inode(parent);
1234 }
1235
1236 dout("d_revalidate %p '%pd' inode %p offset %lld\n", dentry,
1237 dentry, d_inode(dentry), ceph_dentry(dentry)->offset);
1238
1239 /* always trust cached snapped dentries, snapdir dentry */
1240 if (ceph_snap(dir) != CEPH_NOSNAP) {
1241 dout("d_revalidate %p '%pd' inode %p is SNAPPED\n", dentry,
1242 dentry, d_inode(dentry));
1243 valid = 1;
1244 } else if (d_really_is_positive(dentry) &&
1245 ceph_snap(d_inode(dentry)) == CEPH_SNAPDIR) {
1246 valid = 1;
1247 } else {
1248 valid = dentry_lease_is_valid(dentry, flags, dir);
1249 if (valid == -ECHILD)
1250 return valid;
1251 if (valid || dir_lease_is_valid(dir, dentry)) {
1252 if (d_really_is_positive(dentry))
1253 valid = ceph_is_any_caps(d_inode(dentry));
1254 else
1255 valid = 1;
1256 }
1257 }
1258
1259 if (!valid) {
1260 struct ceph_mds_client *mdsc =
1261 ceph_sb_to_client(dir->i_sb)->mdsc;
1262 struct ceph_mds_request *req;
1263 int op, err;
1264 u32 mask;
1265
1266 if (flags & LOOKUP_RCU)
1267 return -ECHILD;
1268
1269 op = ceph_snap(dir) == CEPH_SNAPDIR ?
1270 CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP;
1271 req = ceph_mdsc_create_request(mdsc, op, USE_ANY_MDS);
1272 if (!IS_ERR(req)) {
1273 req->r_dentry = dget(dentry);
1274 req->r_num_caps = 2;
1275 req->r_parent = dir;
1276
1277 mask = CEPH_STAT_CAP_INODE | CEPH_CAP_AUTH_SHARED;
1278 if (ceph_security_xattr_wanted(dir))
1279 mask |= CEPH_CAP_XATTR_SHARED;
1280 req->r_args.getattr.mask = cpu_to_le32(mask);
1281
1282 err = ceph_mdsc_do_request(mdsc, NULL, req);
1283 switch (err) {
1284 case 0:
1285 if (d_really_is_positive(dentry) &&
1286 d_inode(dentry) == req->r_target_inode)
1287 valid = 1;
1288 break;
1289 case -ENOENT:
1290 if (d_really_is_negative(dentry))
1291 valid = 1;
1292 /* Fallthrough */
1293 default:
1294 break;
1295 }
1296 ceph_mdsc_put_request(req);
1297 dout("d_revalidate %p lookup result=%d\n",
1298 dentry, err);
1299 }
1300 }
1301
1302 dout("d_revalidate %p %s\n", dentry, valid ? "valid" : "invalid");
1303 if (valid) {
1304 ceph_dentry_lru_touch(dentry);
1305 } else {
1306 ceph_dir_clear_complete(dir);
1307 }
1308
1309 if (!(flags & LOOKUP_RCU))
1310 dput(parent);
1311 return valid;
1312 }
1313
1314 /*
1315 * Release our ceph_dentry_info.
1316 */
1317 static void ceph_d_release(struct dentry *dentry)
1318 {
1319 struct ceph_dentry_info *di = ceph_dentry(dentry);
1320
1321 dout("d_release %p\n", dentry);
1322 ceph_dentry_lru_del(dentry);
1323
1324 spin_lock(&dentry->d_lock);
1325 dentry->d_fsdata = NULL;
1326 spin_unlock(&dentry->d_lock);
1327
1328 if (di->lease_session)
1329 ceph_put_mds_session(di->lease_session);
1330 kmem_cache_free(ceph_dentry_cachep, di);
1331 }
1332
1333 /*
1334 * When the VFS prunes a dentry from the cache, we need to clear the
1335 * complete flag on the parent directory.
1336 *
1337 * Called under dentry->d_lock.
1338 */
1339 static void ceph_d_prune(struct dentry *dentry)
1340 {
1341 struct ceph_inode_info *dir_ci;
1342 struct ceph_dentry_info *di;
1343
1344 dout("ceph_d_prune %pd %p\n", dentry, dentry);
1345
1346 /* do we have a valid parent? */
1347 if (IS_ROOT(dentry))
1348 return;
1349
1350 /* we hold d_lock, so d_parent is stable */
1351 dir_ci = ceph_inode(d_inode(dentry->d_parent));
1352 if (dir_ci->i_vino.snap == CEPH_SNAPDIR)
1353 return;
1354
1355 /* who calls d_delete() should also disable dcache readdir */
1356 if (d_really_is_negative(dentry))
1357 return;
1358
1359 /* d_fsdata does not get cleared until d_release */
1360 if (!d_unhashed(dentry)) {
1361 __ceph_dir_clear_complete(dir_ci);
1362 return;
1363 }
1364
1365 /* Disable dcache readdir just in case that someone called d_drop()
1366 * or d_invalidate(), but MDS didn't revoke CEPH_CAP_FILE_SHARED
1367 * properly (dcache readdir is still enabled) */
1368 di = ceph_dentry(dentry);
1369 if (di->offset > 0 &&
1370 di->lease_shared_gen == atomic_read(&dir_ci->i_shared_gen))
1371 __ceph_dir_clear_ordered(dir_ci);
1372 }
1373
1374 /*
1375 * read() on a dir. This weird interface hack only works if mounted
1376 * with '-o dirstat'.
1377 */
1378 static ssize_t ceph_read_dir(struct file *file, char __user *buf, size_t size,
1379 loff_t *ppos)
1380 {
1381 struct ceph_file_info *cf = file->private_data;
1382 struct inode *inode = file_inode(file);
1383 struct ceph_inode_info *ci = ceph_inode(inode);
1384 int left;
1385 const int bufsize = 1024;
1386
1387 if (!ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb), DIRSTAT))
1388 return -EISDIR;
1389
1390 if (!cf->dir_info) {
1391 cf->dir_info = kmalloc(bufsize, GFP_KERNEL);
1392 if (!cf->dir_info)
1393 return -ENOMEM;
1394 cf->dir_info_len =
1395 snprintf(cf->dir_info, bufsize,
1396 "entries: %20lld\n"
1397 " files: %20lld\n"
1398 " subdirs: %20lld\n"
1399 "rentries: %20lld\n"
1400 " rfiles: %20lld\n"
1401 " rsubdirs: %20lld\n"
1402 "rbytes: %20lld\n"
1403 "rctime: %10ld.%09ld\n",
1404 ci->i_files + ci->i_subdirs,
1405 ci->i_files,
1406 ci->i_subdirs,
1407 ci->i_rfiles + ci->i_rsubdirs,
1408 ci->i_rfiles,
1409 ci->i_rsubdirs,
1410 ci->i_rbytes,
1411 (long)ci->i_rctime.tv_sec,
1412 (long)ci->i_rctime.tv_nsec);
1413 }
1414
1415 if (*ppos >= cf->dir_info_len)
1416 return 0;
1417 size = min_t(unsigned, size, cf->dir_info_len-*ppos);
1418 left = copy_to_user(buf, cf->dir_info + *ppos, size);
1419 if (left == size)
1420 return -EFAULT;
1421 *ppos += (size - left);
1422 return size - left;
1423 }
1424
1425 /*
1426 * We maintain a private dentry LRU.
1427 *
1428 * FIXME: this needs to be changed to a per-mds lru to be useful.
1429 */
1430 void ceph_dentry_lru_add(struct dentry *dn)
1431 {
1432 struct ceph_dentry_info *di = ceph_dentry(dn);
1433 struct ceph_mds_client *mdsc;
1434
1435 dout("dentry_lru_add %p %p '%pd'\n", di, dn, dn);
1436 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1437 spin_lock(&mdsc->dentry_lru_lock);
1438 list_add_tail(&di->lru, &mdsc->dentry_lru);
1439 mdsc->num_dentry++;
1440 spin_unlock(&mdsc->dentry_lru_lock);
1441 }
1442
1443 void ceph_dentry_lru_touch(struct dentry *dn)
1444 {
1445 struct ceph_dentry_info *di = ceph_dentry(dn);
1446 struct ceph_mds_client *mdsc;
1447
1448 dout("dentry_lru_touch %p %p '%pd' (offset %lld)\n", di, dn, dn,
1449 di->offset);
1450 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1451 spin_lock(&mdsc->dentry_lru_lock);
1452 list_move_tail(&di->lru, &mdsc->dentry_lru);
1453 spin_unlock(&mdsc->dentry_lru_lock);
1454 }
1455
1456 void ceph_dentry_lru_del(struct dentry *dn)
1457 {
1458 struct ceph_dentry_info *di = ceph_dentry(dn);
1459 struct ceph_mds_client *mdsc;
1460
1461 dout("dentry_lru_del %p %p '%pd'\n", di, dn, dn);
1462 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1463 spin_lock(&mdsc->dentry_lru_lock);
1464 list_del_init(&di->lru);
1465 mdsc->num_dentry--;
1466 spin_unlock(&mdsc->dentry_lru_lock);
1467 }
1468
1469 /*
1470 * Return name hash for a given dentry. This is dependent on
1471 * the parent directory's hash function.
1472 */
1473 unsigned ceph_dentry_hash(struct inode *dir, struct dentry *dn)
1474 {
1475 struct ceph_inode_info *dci = ceph_inode(dir);
1476
1477 switch (dci->i_dir_layout.dl_dir_hash) {
1478 case 0: /* for backward compat */
1479 case CEPH_STR_HASH_LINUX:
1480 return dn->d_name.hash;
1481
1482 default:
1483 return ceph_str_hash(dci->i_dir_layout.dl_dir_hash,
1484 dn->d_name.name, dn->d_name.len);
1485 }
1486 }
1487
1488 const struct file_operations ceph_dir_fops = {
1489 .read = ceph_read_dir,
1490 .iterate = ceph_readdir,
1491 .llseek = ceph_dir_llseek,
1492 .open = ceph_open,
1493 .release = ceph_release,
1494 .unlocked_ioctl = ceph_ioctl,
1495 .fsync = ceph_fsync,
1496 };
1497
1498 const struct file_operations ceph_snapdir_fops = {
1499 .iterate = ceph_readdir,
1500 .llseek = ceph_dir_llseek,
1501 .open = ceph_open,
1502 .release = ceph_release,
1503 };
1504
1505 const struct inode_operations ceph_dir_iops = {
1506 .lookup = ceph_lookup,
1507 .permission = ceph_permission,
1508 .getattr = ceph_getattr,
1509 .setattr = ceph_setattr,
1510 .listxattr = ceph_listxattr,
1511 .get_acl = ceph_get_acl,
1512 .set_acl = ceph_set_acl,
1513 .mknod = ceph_mknod,
1514 .symlink = ceph_symlink,
1515 .mkdir = ceph_mkdir,
1516 .link = ceph_link,
1517 .unlink = ceph_unlink,
1518 .rmdir = ceph_unlink,
1519 .rename = ceph_rename,
1520 .create = ceph_create,
1521 .atomic_open = ceph_atomic_open,
1522 };
1523
1524 const struct inode_operations ceph_snapdir_iops = {
1525 .lookup = ceph_lookup,
1526 .permission = ceph_permission,
1527 .getattr = ceph_getattr,
1528 .mkdir = ceph_mkdir,
1529 .rmdir = ceph_unlink,
1530 .rename = ceph_rename,
1531 };
1532
1533 const struct dentry_operations ceph_dentry_ops = {
1534 .d_revalidate = ceph_d_revalidate,
1535 .d_release = ceph_d_release,
1536 .d_prune = ceph_d_prune,
1537 .d_init = ceph_d_init,
1538 };
ubuntu-bionic-kernel mirror