Loading...
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
29const struct inode_operations ceph_dir_iops;
30const struct file_operations ceph_dir_fops;
31const struct dentry_operations ceph_dentry_ops;
32
33/*
34 * Initialize ceph dentry state.
35 */
36int ceph_init_dentry(struct dentry *dentry)
37{
38 struct ceph_dentry_info *di;
39
40 if (dentry->d_fsdata)
41 return 0;
42
43 di = kmem_cache_alloc(ceph_dentry_cachep, GFP_NOFS | __GFP_ZERO);
44 if (!di)
45 return -ENOMEM; /* oh well */
46
47 spin_lock(&dentry->d_lock);
48 if (dentry->d_fsdata) {
49 /* lost a race */
50 kmem_cache_free(ceph_dentry_cachep, di);
51 goto out_unlock;
52 }
53
54 if (ceph_snap(dentry->d_parent->d_inode) == CEPH_NOSNAP)
55 d_set_d_op(dentry, &ceph_dentry_ops);
56 else if (ceph_snap(dentry->d_parent->d_inode) == CEPH_SNAPDIR)
57 d_set_d_op(dentry, &ceph_snapdir_dentry_ops);
58 else
59 d_set_d_op(dentry, &ceph_snap_dentry_ops);
60
61 di->dentry = dentry;
62 di->lease_session = NULL;
63 dentry->d_time = jiffies;
64 /* avoid reordering d_fsdata setup so that the check above is safe */
65 smp_mb();
66 dentry->d_fsdata = di;
67 ceph_dentry_lru_add(dentry);
68out_unlock:
69 spin_unlock(&dentry->d_lock);
70 return 0;
71}
72
73struct inode *ceph_get_dentry_parent_inode(struct dentry *dentry)
74{
75 struct inode *inode = NULL;
76
77 if (!dentry)
78 return NULL;
79
80 spin_lock(&dentry->d_lock);
81 if (!IS_ROOT(dentry)) {
82 inode = dentry->d_parent->d_inode;
83 ihold(inode);
84 }
85 spin_unlock(&dentry->d_lock);
86 return inode;
87}
88
89
90/*
91 * for readdir, we encode the directory frag and offset within that
92 * frag into f_pos.
93 */
94static unsigned fpos_frag(loff_t p)
95{
96 return p >> 32;
97}
98static unsigned fpos_off(loff_t p)
99{
100 return p & 0xffffffff;
101}
102
103static int fpos_cmp(loff_t l, loff_t r)
104{
105 int v = ceph_frag_compare(fpos_frag(l), fpos_frag(r));
106 if (v)
107 return v;
108 return (int)(fpos_off(l) - fpos_off(r));
109}
110
111/*
112 * When possible, we try to satisfy a readdir by peeking at the
113 * dcache. We make this work by carefully ordering dentries on
114 * d_u.d_child when we initially get results back from the MDS, and
115 * falling back to a "normal" sync readdir if any dentries in the dir
116 * are dropped.
117 *
118 * Complete dir indicates that we have all dentries in the dir. It is
119 * defined IFF we hold CEPH_CAP_FILE_SHARED (which will be revoked by
120 * the MDS if/when the directory is modified).
121 */
122static int __dcache_readdir(struct file *file, struct dir_context *ctx,
123 u32 shared_gen)
124{
125 struct ceph_file_info *fi = file->private_data;
126 struct dentry *parent = file->f_dentry;
127 struct inode *dir = parent->d_inode;
128 struct list_head *p;
129 struct dentry *dentry, *last;
130 struct ceph_dentry_info *di;
131 int err = 0;
132
133 /* claim ref on last dentry we returned */
134 last = fi->dentry;
135 fi->dentry = NULL;
136
137 dout("__dcache_readdir %p v%u at %llu (last %p)\n",
138 dir, shared_gen, ctx->pos, last);
139
140 spin_lock(&parent->d_lock);
141
142 /* start at beginning? */
143 if (ctx->pos == 2 || last == NULL ||
144 fpos_cmp(ctx->pos, ceph_dentry(last)->offset) < 0) {
145 if (list_empty(&parent->d_subdirs))
146 goto out_unlock;
147 p = parent->d_subdirs.prev;
148 dout(" initial p %p/%p\n", p->prev, p->next);
149 } else {
150 p = last->d_u.d_child.prev;
151 }
152
153more:
154 dentry = list_entry(p, struct dentry, d_u.d_child);
155 di = ceph_dentry(dentry);
156 while (1) {
157 dout(" p %p/%p %s d_subdirs %p/%p\n", p->prev, p->next,
158 d_unhashed(dentry) ? "!hashed" : "hashed",
159 parent->d_subdirs.prev, parent->d_subdirs.next);
160 if (p == &parent->d_subdirs) {
161 fi->flags |= CEPH_F_ATEND;
162 goto out_unlock;
163 }
164 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
165 if (di->lease_shared_gen == shared_gen &&
166 !d_unhashed(dentry) && dentry->d_inode &&
167 ceph_snap(dentry->d_inode) != CEPH_SNAPDIR &&
168 ceph_ino(dentry->d_inode) != CEPH_INO_CEPH &&
169 fpos_cmp(ctx->pos, di->offset) <= 0)
170 break;
171 dout(" skipping %p %.*s at %llu (%llu)%s%s\n", dentry,
172 dentry->d_name.len, dentry->d_name.name, di->offset,
173 ctx->pos, d_unhashed(dentry) ? " unhashed" : "",
174 !dentry->d_inode ? " null" : "");
175 spin_unlock(&dentry->d_lock);
176 p = p->prev;
177 dentry = list_entry(p, struct dentry, d_u.d_child);
178 di = ceph_dentry(dentry);
179 }
180
181 dget_dlock(dentry);
182 spin_unlock(&dentry->d_lock);
183 spin_unlock(&parent->d_lock);
184
185 /* make sure a dentry wasn't dropped while we didn't have parent lock */
186 if (!ceph_dir_is_complete(dir)) {
187 dout(" lost dir complete on %p; falling back to mds\n", dir);
188 dput(dentry);
189 err = -EAGAIN;
190 goto out;
191 }
192
193 dout(" %llu (%llu) dentry %p %.*s %p\n", di->offset, ctx->pos,
194 dentry, dentry->d_name.len, dentry->d_name.name, dentry->d_inode);
195 if (!dir_emit(ctx, dentry->d_name.name,
196 dentry->d_name.len,
197 ceph_translate_ino(dentry->d_sb, dentry->d_inode->i_ino),
198 dentry->d_inode->i_mode >> 12)) {
199 if (last) {
200 /* remember our position */
201 fi->dentry = last;
202 fi->next_offset = fpos_off(di->offset);
203 }
204 dput(dentry);
205 return 0;
206 }
207
208 ctx->pos = di->offset + 1;
209
210 if (last)
211 dput(last);
212 last = dentry;
213
214 spin_lock(&parent->d_lock);
215 p = p->prev; /* advance to next dentry */
216 goto more;
217
218out_unlock:
219 spin_unlock(&parent->d_lock);
220out:
221 if (last)
222 dput(last);
223 return err;
224}
225
226/*
227 * make note of the last dentry we read, so we can
228 * continue at the same lexicographical point,
229 * regardless of what dir changes take place on the
230 * server.
231 */
232static int note_last_dentry(struct ceph_file_info *fi, const char *name,
233 int len)
234{
235 kfree(fi->last_name);
236 fi->last_name = kmalloc(len+1, GFP_NOFS);
237 if (!fi->last_name)
238 return -ENOMEM;
239 memcpy(fi->last_name, name, len);
240 fi->last_name[len] = 0;
241 dout("note_last_dentry '%s'\n", fi->last_name);
242 return 0;
243}
244
245static int ceph_readdir(struct file *file, struct dir_context *ctx)
246{
247 struct ceph_file_info *fi = file->private_data;
248 struct inode *inode = file_inode(file);
249 struct ceph_inode_info *ci = ceph_inode(inode);
250 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
251 struct ceph_mds_client *mdsc = fsc->mdsc;
252 unsigned frag = fpos_frag(ctx->pos);
253 int off = fpos_off(ctx->pos);
254 int err;
255 u32 ftype;
256 struct ceph_mds_reply_info_parsed *rinfo;
257
258 dout("readdir %p file %p frag %u off %u\n", inode, file, frag, off);
259 if (fi->flags & CEPH_F_ATEND)
260 return 0;
261
262 /* always start with . and .. */
263 if (ctx->pos == 0) {
264 /* note dir version at start of readdir so we can tell
265 * if any dentries get dropped */
266 fi->dir_release_count = atomic_read(&ci->i_release_count);
267
268 dout("readdir off 0 -> '.'\n");
269 if (!dir_emit(ctx, ".", 1,
270 ceph_translate_ino(inode->i_sb, inode->i_ino),
271 inode->i_mode >> 12))
272 return 0;
273 ctx->pos = 1;
274 off = 1;
275 }
276 if (ctx->pos == 1) {
277 ino_t ino = parent_ino(file->f_dentry);
278 dout("readdir off 1 -> '..'\n");
279 if (!dir_emit(ctx, "..", 2,
280 ceph_translate_ino(inode->i_sb, ino),
281 inode->i_mode >> 12))
282 return 0;
283 ctx->pos = 2;
284 off = 2;
285 }
286
287 /* can we use the dcache? */
288 spin_lock(&ci->i_ceph_lock);
289 if ((ctx->pos == 2 || fi->dentry) &&
290 !ceph_test_mount_opt(fsc, NOASYNCREADDIR) &&
291 ceph_snap(inode) != CEPH_SNAPDIR &&
292 __ceph_dir_is_complete(ci) &&
293 __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1)) {
294 u32 shared_gen = ci->i_shared_gen;
295 spin_unlock(&ci->i_ceph_lock);
296 err = __dcache_readdir(file, ctx, shared_gen);
297 if (err != -EAGAIN)
298 return err;
299 frag = fpos_frag(ctx->pos);
300 off = fpos_off(ctx->pos);
301 } else {
302 spin_unlock(&ci->i_ceph_lock);
303 }
304 if (fi->dentry) {
305 err = note_last_dentry(fi, fi->dentry->d_name.name,
306 fi->dentry->d_name.len);
307 if (err)
308 return err;
309 dput(fi->dentry);
310 fi->dentry = NULL;
311 }
312
313 /* proceed with a normal readdir */
314
315more:
316 /* do we have the correct frag content buffered? */
317 if (fi->frag != frag || fi->last_readdir == NULL) {
318 struct ceph_mds_request *req;
319 int op = ceph_snap(inode) == CEPH_SNAPDIR ?
320 CEPH_MDS_OP_LSSNAP : CEPH_MDS_OP_READDIR;
321
322 /* discard old result, if any */
323 if (fi->last_readdir) {
324 ceph_mdsc_put_request(fi->last_readdir);
325 fi->last_readdir = NULL;
326 }
327
328 dout("readdir fetching %llx.%llx frag %x offset '%s'\n",
329 ceph_vinop(inode), frag, fi->last_name);
330 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
331 if (IS_ERR(req))
332 return PTR_ERR(req);
333 err = ceph_alloc_readdir_reply_buffer(req, inode);
334 if (err) {
335 ceph_mdsc_put_request(req);
336 return err;
337 }
338 req->r_inode = inode;
339 ihold(inode);
340 req->r_dentry = dget(file->f_dentry);
341 /* hints to request -> mds selection code */
342 req->r_direct_mode = USE_AUTH_MDS;
343 req->r_direct_hash = ceph_frag_value(frag);
344 req->r_direct_is_hash = true;
345 req->r_path2 = kstrdup(fi->last_name, GFP_NOFS);
346 req->r_readdir_offset = fi->next_offset;
347 req->r_args.readdir.frag = cpu_to_le32(frag);
348 err = ceph_mdsc_do_request(mdsc, NULL, req);
349 if (err < 0) {
350 ceph_mdsc_put_request(req);
351 return err;
352 }
353 dout("readdir got and parsed readdir result=%d"
354 " on frag %x, end=%d, complete=%d\n", err, frag,
355 (int)req->r_reply_info.dir_end,
356 (int)req->r_reply_info.dir_complete);
357
358 if (!req->r_did_prepopulate) {
359 dout("readdir !did_prepopulate");
360 /* preclude from marking dir complete */
361 fi->dir_release_count--;
362 }
363
364 /* note next offset and last dentry name */
365 rinfo = &req->r_reply_info;
366 if (le32_to_cpu(rinfo->dir_dir->frag) != frag) {
367 frag = le32_to_cpu(rinfo->dir_dir->frag);
368 if (ceph_frag_is_leftmost(frag))
369 fi->next_offset = 2;
370 else
371 fi->next_offset = 0;
372 off = fi->next_offset;
373 }
374 fi->frag = frag;
375 fi->offset = fi->next_offset;
376 fi->last_readdir = req;
377
378 if (req->r_reply_info.dir_end) {
379 kfree(fi->last_name);
380 fi->last_name = NULL;
381 if (ceph_frag_is_rightmost(frag))
382 fi->next_offset = 2;
383 else
384 fi->next_offset = 0;
385 } else {
386 err = note_last_dentry(fi,
387 rinfo->dir_dname[rinfo->dir_nr-1],
388 rinfo->dir_dname_len[rinfo->dir_nr-1]);
389 if (err)
390 return err;
391 fi->next_offset += rinfo->dir_nr;
392 }
393 }
394
395 rinfo = &fi->last_readdir->r_reply_info;
396 dout("readdir frag %x num %d off %d chunkoff %d\n", frag,
397 rinfo->dir_nr, off, fi->offset);
398
399 ctx->pos = ceph_make_fpos(frag, off);
400 while (off >= fi->offset && off - fi->offset < rinfo->dir_nr) {
401 struct ceph_mds_reply_inode *in =
402 rinfo->dir_in[off - fi->offset].in;
403 struct ceph_vino vino;
404 ino_t ino;
405
406 dout("readdir off %d (%d/%d) -> %lld '%.*s' %p\n",
407 off, off - fi->offset, rinfo->dir_nr, ctx->pos,
408 rinfo->dir_dname_len[off - fi->offset],
409 rinfo->dir_dname[off - fi->offset], in);
410 BUG_ON(!in);
411 ftype = le32_to_cpu(in->mode) >> 12;
412 vino.ino = le64_to_cpu(in->ino);
413 vino.snap = le64_to_cpu(in->snapid);
414 ino = ceph_vino_to_ino(vino);
415 if (!dir_emit(ctx,
416 rinfo->dir_dname[off - fi->offset],
417 rinfo->dir_dname_len[off - fi->offset],
418 ceph_translate_ino(inode->i_sb, ino), ftype)) {
419 dout("filldir stopping us...\n");
420 return 0;
421 }
422 off++;
423 ctx->pos++;
424 }
425
426 if (fi->last_name) {
427 ceph_mdsc_put_request(fi->last_readdir);
428 fi->last_readdir = NULL;
429 goto more;
430 }
431
432 /* more frags? */
433 if (!ceph_frag_is_rightmost(frag)) {
434 frag = ceph_frag_next(frag);
435 off = 0;
436 ctx->pos = ceph_make_fpos(frag, off);
437 dout("readdir next frag is %x\n", frag);
438 goto more;
439 }
440 fi->flags |= CEPH_F_ATEND;
441
442 /*
443 * if dir_release_count still matches the dir, no dentries
444 * were released during the whole readdir, and we should have
445 * the complete dir contents in our cache.
446 */
447 spin_lock(&ci->i_ceph_lock);
448 if (atomic_read(&ci->i_release_count) == fi->dir_release_count) {
449 dout(" marking %p complete\n", inode);
450 __ceph_dir_set_complete(ci, fi->dir_release_count);
451 }
452 spin_unlock(&ci->i_ceph_lock);
453
454 dout("readdir %p file %p done.\n", inode, file);
455 return 0;
456}
457
458static void reset_readdir(struct ceph_file_info *fi, unsigned frag)
459{
460 if (fi->last_readdir) {
461 ceph_mdsc_put_request(fi->last_readdir);
462 fi->last_readdir = NULL;
463 }
464 kfree(fi->last_name);
465 fi->last_name = NULL;
466 if (ceph_frag_is_leftmost(frag))
467 fi->next_offset = 2; /* compensate for . and .. */
468 else
469 fi->next_offset = 0;
470 if (fi->dentry) {
471 dput(fi->dentry);
472 fi->dentry = NULL;
473 }
474 fi->flags &= ~CEPH_F_ATEND;
475}
476
477static loff_t ceph_dir_llseek(struct file *file, loff_t offset, int whence)
478{
479 struct ceph_file_info *fi = file->private_data;
480 struct inode *inode = file->f_mapping->host;
481 loff_t old_offset = ceph_make_fpos(fi->frag, fi->next_offset);
482 loff_t retval;
483
484 mutex_lock(&inode->i_mutex);
485 retval = -EINVAL;
486 switch (whence) {
487 case SEEK_END:
488 offset += inode->i_size + 2; /* FIXME */
489 break;
490 case SEEK_CUR:
491 offset += file->f_pos;
492 case SEEK_SET:
493 break;
494 default:
495 goto out;
496 }
497
498 if (offset >= 0) {
499 if (offset != file->f_pos) {
500 file->f_pos = offset;
501 file->f_version = 0;
502 fi->flags &= ~CEPH_F_ATEND;
503 }
504 retval = offset;
505
506 /*
507 * discard buffered readdir content on seekdir(0), or
508 * seek to new frag, or seek prior to current chunk.
509 */
510 if (offset == 0 ||
511 fpos_frag(offset) != fi->frag ||
512 fpos_off(offset) < fi->offset) {
513 dout("dir_llseek dropping %p content\n", file);
514 reset_readdir(fi, fpos_frag(offset));
515 }
516
517 /* bump dir_release_count if we did a forward seek */
518 if (fpos_cmp(offset, old_offset) > 0)
519 fi->dir_release_count--;
520 }
521out:
522 mutex_unlock(&inode->i_mutex);
523 return retval;
524}
525
526/*
527 * Handle lookups for the hidden .snap directory.
528 */
529int ceph_handle_snapdir(struct ceph_mds_request *req,
530 struct dentry *dentry, int err)
531{
532 struct ceph_fs_client *fsc = ceph_sb_to_client(dentry->d_sb);
533 struct inode *parent = dentry->d_parent->d_inode; /* we hold i_mutex */
534
535 /* .snap dir? */
536 if (err == -ENOENT &&
537 ceph_snap(parent) == CEPH_NOSNAP &&
538 strcmp(dentry->d_name.name,
539 fsc->mount_options->snapdir_name) == 0) {
540 struct inode *inode = ceph_get_snapdir(parent);
541 dout("ENOENT on snapdir %p '%.*s', linking to snapdir %p\n",
542 dentry, dentry->d_name.len, dentry->d_name.name, inode);
543 BUG_ON(!d_unhashed(dentry));
544 d_add(dentry, inode);
545 err = 0;
546 }
547 return err;
548}
549
550/*
551 * Figure out final result of a lookup/open request.
552 *
553 * Mainly, make sure we return the final req->r_dentry (if it already
554 * existed) in place of the original VFS-provided dentry when they
555 * differ.
556 *
557 * Gracefully handle the case where the MDS replies with -ENOENT and
558 * no trace (which it may do, at its discretion, e.g., if it doesn't
559 * care to issue a lease on the negative dentry).
560 */
561struct dentry *ceph_finish_lookup(struct ceph_mds_request *req,
562 struct dentry *dentry, int err)
563{
564 if (err == -ENOENT) {
565 /* no trace? */
566 err = 0;
567 if (!req->r_reply_info.head->is_dentry) {
568 dout("ENOENT and no trace, dentry %p inode %p\n",
569 dentry, dentry->d_inode);
570 if (dentry->d_inode) {
571 d_drop(dentry);
572 err = -ENOENT;
573 } else {
574 d_add(dentry, NULL);
575 }
576 }
577 }
578 if (err)
579 dentry = ERR_PTR(err);
580 else if (dentry != req->r_dentry)
581 dentry = dget(req->r_dentry); /* we got spliced */
582 else
583 dentry = NULL;
584 return dentry;
585}
586
587static int is_root_ceph_dentry(struct inode *inode, struct dentry *dentry)
588{
589 return ceph_ino(inode) == CEPH_INO_ROOT &&
590 strncmp(dentry->d_name.name, ".ceph", 5) == 0;
591}
592
593/*
594 * Look up a single dir entry. If there is a lookup intent, inform
595 * the MDS so that it gets our 'caps wanted' value in a single op.
596 */
597static struct dentry *ceph_lookup(struct inode *dir, struct dentry *dentry,
598 unsigned int flags)
599{
600 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
601 struct ceph_mds_client *mdsc = fsc->mdsc;
602 struct ceph_mds_request *req;
603 int op;
604 int err;
605
606 dout("lookup %p dentry %p '%.*s'\n",
607 dir, dentry, dentry->d_name.len, dentry->d_name.name);
608
609 if (dentry->d_name.len > NAME_MAX)
610 return ERR_PTR(-ENAMETOOLONG);
611
612 err = ceph_init_dentry(dentry);
613 if (err < 0)
614 return ERR_PTR(err);
615
616 /* can we conclude ENOENT locally? */
617 if (dentry->d_inode == NULL) {
618 struct ceph_inode_info *ci = ceph_inode(dir);
619 struct ceph_dentry_info *di = ceph_dentry(dentry);
620
621 spin_lock(&ci->i_ceph_lock);
622 dout(" dir %p flags are %d\n", dir, ci->i_ceph_flags);
623 if (strncmp(dentry->d_name.name,
624 fsc->mount_options->snapdir_name,
625 dentry->d_name.len) &&
626 !is_root_ceph_dentry(dir, dentry) &&
627 __ceph_dir_is_complete(ci) &&
628 (__ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1))) {
629 spin_unlock(&ci->i_ceph_lock);
630 dout(" dir %p complete, -ENOENT\n", dir);
631 d_add(dentry, NULL);
632 di->lease_shared_gen = ci->i_shared_gen;
633 return NULL;
634 }
635 spin_unlock(&ci->i_ceph_lock);
636 }
637
638 op = ceph_snap(dir) == CEPH_SNAPDIR ?
639 CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP;
640 req = ceph_mdsc_create_request(mdsc, op, USE_ANY_MDS);
641 if (IS_ERR(req))
642 return ERR_CAST(req);
643 req->r_dentry = dget(dentry);
644 req->r_num_caps = 2;
645 /* we only need inode linkage */
646 req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
647 req->r_locked_dir = dir;
648 err = ceph_mdsc_do_request(mdsc, NULL, req);
649 err = ceph_handle_snapdir(req, dentry, err);
650 dentry = ceph_finish_lookup(req, dentry, err);
651 ceph_mdsc_put_request(req); /* will dput(dentry) */
652 dout("lookup result=%p\n", dentry);
653 return dentry;
654}
655
656/*
657 * If we do a create but get no trace back from the MDS, follow up with
658 * a lookup (the VFS expects us to link up the provided dentry).
659 */
660int ceph_handle_notrace_create(struct inode *dir, struct dentry *dentry)
661{
662 struct dentry *result = ceph_lookup(dir, dentry, 0);
663
664 if (result && !IS_ERR(result)) {
665 /*
666 * We created the item, then did a lookup, and found
667 * it was already linked to another inode we already
668 * had in our cache (and thus got spliced). Link our
669 * dentry to that inode, but don't hash it, just in
670 * case the VFS wants to dereference it.
671 */
672 BUG_ON(!result->d_inode);
673 d_instantiate(dentry, result->d_inode);
674 return 0;
675 }
676 return PTR_ERR(result);
677}
678
679static int ceph_mknod(struct inode *dir, struct dentry *dentry,
680 umode_t mode, dev_t rdev)
681{
682 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
683 struct ceph_mds_client *mdsc = fsc->mdsc;
684 struct ceph_mds_request *req;
685 int err;
686
687 if (ceph_snap(dir) != CEPH_NOSNAP)
688 return -EROFS;
689
690 dout("mknod in dir %p dentry %p mode 0%ho rdev %d\n",
691 dir, dentry, mode, rdev);
692 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_MKNOD, USE_AUTH_MDS);
693 if (IS_ERR(req)) {
694 d_drop(dentry);
695 return PTR_ERR(req);
696 }
697 req->r_dentry = dget(dentry);
698 req->r_num_caps = 2;
699 req->r_locked_dir = dir;
700 req->r_args.mknod.mode = cpu_to_le32(mode);
701 req->r_args.mknod.rdev = cpu_to_le32(rdev);
702 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
703 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
704 err = ceph_mdsc_do_request(mdsc, dir, req);
705 if (!err && !req->r_reply_info.head->is_dentry)
706 err = ceph_handle_notrace_create(dir, dentry);
707 ceph_mdsc_put_request(req);
708
709 if (!err)
710 ceph_init_acl(dentry, dentry->d_inode, dir);
711 else
712 d_drop(dentry);
713 return err;
714}
715
716static int ceph_create(struct inode *dir, struct dentry *dentry, umode_t mode,
717 bool excl)
718{
719 return ceph_mknod(dir, dentry, mode, 0);
720}
721
722static int ceph_symlink(struct inode *dir, struct dentry *dentry,
723 const char *dest)
724{
725 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
726 struct ceph_mds_client *mdsc = fsc->mdsc;
727 struct ceph_mds_request *req;
728 int err;
729
730 if (ceph_snap(dir) != CEPH_NOSNAP)
731 return -EROFS;
732
733 dout("symlink in dir %p dentry %p to '%s'\n", dir, dentry, dest);
734 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SYMLINK, USE_AUTH_MDS);
735 if (IS_ERR(req)) {
736 d_drop(dentry);
737 return PTR_ERR(req);
738 }
739 req->r_dentry = dget(dentry);
740 req->r_num_caps = 2;
741 req->r_path2 = kstrdup(dest, GFP_NOFS);
742 req->r_locked_dir = dir;
743 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
744 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
745 err = ceph_mdsc_do_request(mdsc, dir, req);
746 if (!err && !req->r_reply_info.head->is_dentry)
747 err = ceph_handle_notrace_create(dir, dentry);
748 ceph_mdsc_put_request(req);
749 if (!err)
750 ceph_init_acl(dentry, dentry->d_inode, dir);
751 else
752 d_drop(dentry);
753 return err;
754}
755
756static int ceph_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
757{
758 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
759 struct ceph_mds_client *mdsc = fsc->mdsc;
760 struct ceph_mds_request *req;
761 int err = -EROFS;
762 int op;
763
764 if (ceph_snap(dir) == CEPH_SNAPDIR) {
765 /* mkdir .snap/foo is a MKSNAP */
766 op = CEPH_MDS_OP_MKSNAP;
767 dout("mksnap dir %p snap '%.*s' dn %p\n", dir,
768 dentry->d_name.len, dentry->d_name.name, dentry);
769 } else if (ceph_snap(dir) == CEPH_NOSNAP) {
770 dout("mkdir dir %p dn %p mode 0%ho\n", dir, dentry, mode);
771 op = CEPH_MDS_OP_MKDIR;
772 } else {
773 goto out;
774 }
775 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
776 if (IS_ERR(req)) {
777 err = PTR_ERR(req);
778 goto out;
779 }
780
781 req->r_dentry = dget(dentry);
782 req->r_num_caps = 2;
783 req->r_locked_dir = dir;
784 req->r_args.mkdir.mode = cpu_to_le32(mode);
785 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
786 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
787 err = ceph_mdsc_do_request(mdsc, dir, req);
788 if (!err && !req->r_reply_info.head->is_dentry)
789 err = ceph_handle_notrace_create(dir, dentry);
790 ceph_mdsc_put_request(req);
791out:
792 if (!err)
793 ceph_init_acl(dentry, dentry->d_inode, dir);
794 else
795 d_drop(dentry);
796 return err;
797}
798
799static int ceph_link(struct dentry *old_dentry, struct inode *dir,
800 struct dentry *dentry)
801{
802 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
803 struct ceph_mds_client *mdsc = fsc->mdsc;
804 struct ceph_mds_request *req;
805 int err;
806
807 if (ceph_snap(dir) != CEPH_NOSNAP)
808 return -EROFS;
809
810 dout("link in dir %p old_dentry %p dentry %p\n", dir,
811 old_dentry, dentry);
812 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LINK, USE_AUTH_MDS);
813 if (IS_ERR(req)) {
814 d_drop(dentry);
815 return PTR_ERR(req);
816 }
817 req->r_dentry = dget(dentry);
818 req->r_num_caps = 2;
819 req->r_old_dentry = dget(old_dentry);
820 req->r_locked_dir = dir;
821 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
822 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
823 /* release LINK_SHARED on source inode (mds will lock it) */
824 req->r_old_inode_drop = CEPH_CAP_LINK_SHARED;
825 err = ceph_mdsc_do_request(mdsc, dir, req);
826 if (err) {
827 d_drop(dentry);
828 } else if (!req->r_reply_info.head->is_dentry) {
829 ihold(old_dentry->d_inode);
830 d_instantiate(dentry, old_dentry->d_inode);
831 }
832 ceph_mdsc_put_request(req);
833 return err;
834}
835
836/*
837 * For a soon-to-be unlinked file, drop the AUTH_RDCACHE caps. If it
838 * looks like the link count will hit 0, drop any other caps (other
839 * than PIN) we don't specifically want (due to the file still being
840 * open).
841 */
842static int drop_caps_for_unlink(struct inode *inode)
843{
844 struct ceph_inode_info *ci = ceph_inode(inode);
845 int drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL;
846
847 spin_lock(&ci->i_ceph_lock);
848 if (inode->i_nlink == 1) {
849 drop |= ~(__ceph_caps_wanted(ci) | CEPH_CAP_PIN);
850 ci->i_ceph_flags |= CEPH_I_NODELAY;
851 }
852 spin_unlock(&ci->i_ceph_lock);
853 return drop;
854}
855
856/*
857 * rmdir and unlink are differ only by the metadata op code
858 */
859static int ceph_unlink(struct inode *dir, struct dentry *dentry)
860{
861 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
862 struct ceph_mds_client *mdsc = fsc->mdsc;
863 struct inode *inode = dentry->d_inode;
864 struct ceph_mds_request *req;
865 int err = -EROFS;
866 int op;
867
868 if (ceph_snap(dir) == CEPH_SNAPDIR) {
869 /* rmdir .snap/foo is RMSNAP */
870 dout("rmsnap dir %p '%.*s' dn %p\n", dir, dentry->d_name.len,
871 dentry->d_name.name, dentry);
872 op = CEPH_MDS_OP_RMSNAP;
873 } else if (ceph_snap(dir) == CEPH_NOSNAP) {
874 dout("unlink/rmdir dir %p dn %p inode %p\n",
875 dir, dentry, inode);
876 op = S_ISDIR(dentry->d_inode->i_mode) ?
877 CEPH_MDS_OP_RMDIR : CEPH_MDS_OP_UNLINK;
878 } else
879 goto out;
880 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
881 if (IS_ERR(req)) {
882 err = PTR_ERR(req);
883 goto out;
884 }
885 req->r_dentry = dget(dentry);
886 req->r_num_caps = 2;
887 req->r_locked_dir = dir;
888 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
889 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
890 req->r_inode_drop = drop_caps_for_unlink(inode);
891 err = ceph_mdsc_do_request(mdsc, dir, req);
892 if (!err && !req->r_reply_info.head->is_dentry)
893 d_delete(dentry);
894 ceph_mdsc_put_request(req);
895out:
896 return err;
897}
898
899static int ceph_rename(struct inode *old_dir, struct dentry *old_dentry,
900 struct inode *new_dir, struct dentry *new_dentry)
901{
902 struct ceph_fs_client *fsc = ceph_sb_to_client(old_dir->i_sb);
903 struct ceph_mds_client *mdsc = fsc->mdsc;
904 struct ceph_mds_request *req;
905 int err;
906
907 if (ceph_snap(old_dir) != ceph_snap(new_dir))
908 return -EXDEV;
909 if (ceph_snap(old_dir) != CEPH_NOSNAP ||
910 ceph_snap(new_dir) != CEPH_NOSNAP)
911 return -EROFS;
912 dout("rename dir %p dentry %p to dir %p dentry %p\n",
913 old_dir, old_dentry, new_dir, new_dentry);
914 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_RENAME, USE_AUTH_MDS);
915 if (IS_ERR(req))
916 return PTR_ERR(req);
917 ihold(old_dir);
918 req->r_dentry = dget(new_dentry);
919 req->r_num_caps = 2;
920 req->r_old_dentry = dget(old_dentry);
921 req->r_old_dentry_dir = old_dir;
922 req->r_locked_dir = new_dir;
923 req->r_old_dentry_drop = CEPH_CAP_FILE_SHARED;
924 req->r_old_dentry_unless = CEPH_CAP_FILE_EXCL;
925 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
926 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
927 /* release LINK_RDCACHE on source inode (mds will lock it) */
928 req->r_old_inode_drop = CEPH_CAP_LINK_SHARED;
929 if (new_dentry->d_inode)
930 req->r_inode_drop = drop_caps_for_unlink(new_dentry->d_inode);
931 err = ceph_mdsc_do_request(mdsc, old_dir, req);
932 if (!err && !req->r_reply_info.head->is_dentry) {
933 /*
934 * Normally d_move() is done by fill_trace (called by
935 * do_request, above). If there is no trace, we need
936 * to do it here.
937 */
938
939 d_move(old_dentry, new_dentry);
940
941 /* ensure target dentry is invalidated, despite
942 rehashing bug in vfs_rename_dir */
943 ceph_invalidate_dentry_lease(new_dentry);
944
945 /* d_move screws up sibling dentries' offsets */
946 ceph_dir_clear_complete(old_dir);
947 ceph_dir_clear_complete(new_dir);
948
949 }
950 ceph_mdsc_put_request(req);
951 return err;
952}
953
954/*
955 * Ensure a dentry lease will no longer revalidate.
956 */
957void ceph_invalidate_dentry_lease(struct dentry *dentry)
958{
959 spin_lock(&dentry->d_lock);
960 dentry->d_time = jiffies;
961 ceph_dentry(dentry)->lease_shared_gen = 0;
962 spin_unlock(&dentry->d_lock);
963}
964
965/*
966 * Check if dentry lease is valid. If not, delete the lease. Try to
967 * renew if the least is more than half up.
968 */
969static int dentry_lease_is_valid(struct dentry *dentry)
970{
971 struct ceph_dentry_info *di;
972 struct ceph_mds_session *s;
973 int valid = 0;
974 u32 gen;
975 unsigned long ttl;
976 struct ceph_mds_session *session = NULL;
977 struct inode *dir = NULL;
978 u32 seq = 0;
979
980 spin_lock(&dentry->d_lock);
981 di = ceph_dentry(dentry);
982 if (di->lease_session) {
983 s = di->lease_session;
984 spin_lock(&s->s_gen_ttl_lock);
985 gen = s->s_cap_gen;
986 ttl = s->s_cap_ttl;
987 spin_unlock(&s->s_gen_ttl_lock);
988
989 if (di->lease_gen == gen &&
990 time_before(jiffies, dentry->d_time) &&
991 time_before(jiffies, ttl)) {
992 valid = 1;
993 if (di->lease_renew_after &&
994 time_after(jiffies, di->lease_renew_after)) {
995 /* we should renew */
996 dir = dentry->d_parent->d_inode;
997 session = ceph_get_mds_session(s);
998 seq = di->lease_seq;
999 di->lease_renew_after = 0;
1000 di->lease_renew_from = jiffies;
1001 }
1002 }
1003 }
1004 spin_unlock(&dentry->d_lock);
1005
1006 if (session) {
1007 ceph_mdsc_lease_send_msg(session, dir, dentry,
1008 CEPH_MDS_LEASE_RENEW, seq);
1009 ceph_put_mds_session(session);
1010 }
1011 dout("dentry_lease_is_valid - dentry %p = %d\n", dentry, valid);
1012 return valid;
1013}
1014
1015/*
1016 * Check if directory-wide content lease/cap is valid.
1017 */
1018static int dir_lease_is_valid(struct inode *dir, struct dentry *dentry)
1019{
1020 struct ceph_inode_info *ci = ceph_inode(dir);
1021 struct ceph_dentry_info *di = ceph_dentry(dentry);
1022 int valid = 0;
1023
1024 spin_lock(&ci->i_ceph_lock);
1025 if (ci->i_shared_gen == di->lease_shared_gen)
1026 valid = __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1);
1027 spin_unlock(&ci->i_ceph_lock);
1028 dout("dir_lease_is_valid dir %p v%u dentry %p v%u = %d\n",
1029 dir, (unsigned)ci->i_shared_gen, dentry,
1030 (unsigned)di->lease_shared_gen, valid);
1031 return valid;
1032}
1033
1034/*
1035 * Check if cached dentry can be trusted.
1036 */
1037static int ceph_d_revalidate(struct dentry *dentry, unsigned int flags)
1038{
1039 int valid = 0;
1040 struct inode *dir;
1041
1042 if (flags & LOOKUP_RCU)
1043 return -ECHILD;
1044
1045 dout("d_revalidate %p '%.*s' inode %p offset %lld\n", dentry,
1046 dentry->d_name.len, dentry->d_name.name, dentry->d_inode,
1047 ceph_dentry(dentry)->offset);
1048
1049 dir = ceph_get_dentry_parent_inode(dentry);
1050
1051 /* always trust cached snapped dentries, snapdir dentry */
1052 if (ceph_snap(dir) != CEPH_NOSNAP) {
1053 dout("d_revalidate %p '%.*s' inode %p is SNAPPED\n", dentry,
1054 dentry->d_name.len, dentry->d_name.name, dentry->d_inode);
1055 valid = 1;
1056 } else if (dentry->d_inode &&
1057 ceph_snap(dentry->d_inode) == CEPH_SNAPDIR) {
1058 valid = 1;
1059 } else if (dentry_lease_is_valid(dentry) ||
1060 dir_lease_is_valid(dir, dentry)) {
1061 if (dentry->d_inode)
1062 valid = ceph_is_any_caps(dentry->d_inode);
1063 else
1064 valid = 1;
1065 }
1066
1067 dout("d_revalidate %p %s\n", dentry, valid ? "valid" : "invalid");
1068 if (valid) {
1069 ceph_dentry_lru_touch(dentry);
1070 } else {
1071 ceph_dir_clear_complete(dir);
1072 d_drop(dentry);
1073 }
1074 iput(dir);
1075 return valid;
1076}
1077
1078/*
1079 * Release our ceph_dentry_info.
1080 */
1081static void ceph_d_release(struct dentry *dentry)
1082{
1083 struct ceph_dentry_info *di = ceph_dentry(dentry);
1084
1085 dout("d_release %p\n", dentry);
1086 ceph_dentry_lru_del(dentry);
1087 if (di->lease_session)
1088 ceph_put_mds_session(di->lease_session);
1089 kmem_cache_free(ceph_dentry_cachep, di);
1090 dentry->d_fsdata = NULL;
1091}
1092
1093static int ceph_snapdir_d_revalidate(struct dentry *dentry,
1094 unsigned int flags)
1095{
1096 /*
1097 * Eventually, we'll want to revalidate snapped metadata
1098 * too... probably...
1099 */
1100 return 1;
1101}
1102
1103/*
1104 * When the VFS prunes a dentry from the cache, we need to clear the
1105 * complete flag on the parent directory.
1106 *
1107 * Called under dentry->d_lock.
1108 */
1109static void ceph_d_prune(struct dentry *dentry)
1110{
1111 dout("ceph_d_prune %p\n", dentry);
1112
1113 /* do we have a valid parent? */
1114 if (IS_ROOT(dentry))
1115 return;
1116
1117 /* if we are not hashed, we don't affect dir's completeness */
1118 if (d_unhashed(dentry))
1119 return;
1120
1121 /*
1122 * we hold d_lock, so d_parent is stable, and d_fsdata is never
1123 * cleared until d_release
1124 */
1125 ceph_dir_clear_complete(dentry->d_parent->d_inode);
1126}
1127
1128/*
1129 * read() on a dir. This weird interface hack only works if mounted
1130 * with '-o dirstat'.
1131 */
1132static ssize_t ceph_read_dir(struct file *file, char __user *buf, size_t size,
1133 loff_t *ppos)
1134{
1135 struct ceph_file_info *cf = file->private_data;
1136 struct inode *inode = file_inode(file);
1137 struct ceph_inode_info *ci = ceph_inode(inode);
1138 int left;
1139 const int bufsize = 1024;
1140
1141 if (!ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb), DIRSTAT))
1142 return -EISDIR;
1143
1144 if (!cf->dir_info) {
1145 cf->dir_info = kmalloc(bufsize, GFP_NOFS);
1146 if (!cf->dir_info)
1147 return -ENOMEM;
1148 cf->dir_info_len =
1149 snprintf(cf->dir_info, bufsize,
1150 "entries: %20lld\n"
1151 " files: %20lld\n"
1152 " subdirs: %20lld\n"
1153 "rentries: %20lld\n"
1154 " rfiles: %20lld\n"
1155 " rsubdirs: %20lld\n"
1156 "rbytes: %20lld\n"
1157 "rctime: %10ld.%09ld\n",
1158 ci->i_files + ci->i_subdirs,
1159 ci->i_files,
1160 ci->i_subdirs,
1161 ci->i_rfiles + ci->i_rsubdirs,
1162 ci->i_rfiles,
1163 ci->i_rsubdirs,
1164 ci->i_rbytes,
1165 (long)ci->i_rctime.tv_sec,
1166 (long)ci->i_rctime.tv_nsec);
1167 }
1168
1169 if (*ppos >= cf->dir_info_len)
1170 return 0;
1171 size = min_t(unsigned, size, cf->dir_info_len-*ppos);
1172 left = copy_to_user(buf, cf->dir_info + *ppos, size);
1173 if (left == size)
1174 return -EFAULT;
1175 *ppos += (size - left);
1176 return size - left;
1177}
1178
1179/*
1180 * an fsync() on a dir will wait for any uncommitted directory
1181 * operations to commit.
1182 */
1183static int ceph_dir_fsync(struct file *file, loff_t start, loff_t end,
1184 int datasync)
1185{
1186 struct inode *inode = file_inode(file);
1187 struct ceph_inode_info *ci = ceph_inode(inode);
1188 struct list_head *head = &ci->i_unsafe_dirops;
1189 struct ceph_mds_request *req;
1190 u64 last_tid;
1191 int ret = 0;
1192
1193 dout("dir_fsync %p\n", inode);
1194 ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
1195 if (ret)
1196 return ret;
1197 mutex_lock(&inode->i_mutex);
1198
1199 spin_lock(&ci->i_unsafe_lock);
1200 if (list_empty(head))
1201 goto out;
1202
1203 req = list_entry(head->prev,
1204 struct ceph_mds_request, r_unsafe_dir_item);
1205 last_tid = req->r_tid;
1206
1207 do {
1208 ceph_mdsc_get_request(req);
1209 spin_unlock(&ci->i_unsafe_lock);
1210
1211 dout("dir_fsync %p wait on tid %llu (until %llu)\n",
1212 inode, req->r_tid, last_tid);
1213 if (req->r_timeout) {
1214 ret = wait_for_completion_timeout(
1215 &req->r_safe_completion, req->r_timeout);
1216 if (ret > 0)
1217 ret = 0;
1218 else if (ret == 0)
1219 ret = -EIO; /* timed out */
1220 } else {
1221 wait_for_completion(&req->r_safe_completion);
1222 }
1223 ceph_mdsc_put_request(req);
1224
1225 spin_lock(&ci->i_unsafe_lock);
1226 if (ret || list_empty(head))
1227 break;
1228 req = list_entry(head->next,
1229 struct ceph_mds_request, r_unsafe_dir_item);
1230 } while (req->r_tid < last_tid);
1231out:
1232 spin_unlock(&ci->i_unsafe_lock);
1233 mutex_unlock(&inode->i_mutex);
1234
1235 return ret;
1236}
1237
1238/*
1239 * We maintain a private dentry LRU.
1240 *
1241 * FIXME: this needs to be changed to a per-mds lru to be useful.
1242 */
1243void ceph_dentry_lru_add(struct dentry *dn)
1244{
1245 struct ceph_dentry_info *di = ceph_dentry(dn);
1246 struct ceph_mds_client *mdsc;
1247
1248 dout("dentry_lru_add %p %p '%.*s'\n", di, dn,
1249 dn->d_name.len, dn->d_name.name);
1250 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1251 spin_lock(&mdsc->dentry_lru_lock);
1252 list_add_tail(&di->lru, &mdsc->dentry_lru);
1253 mdsc->num_dentry++;
1254 spin_unlock(&mdsc->dentry_lru_lock);
1255}
1256
1257void ceph_dentry_lru_touch(struct dentry *dn)
1258{
1259 struct ceph_dentry_info *di = ceph_dentry(dn);
1260 struct ceph_mds_client *mdsc;
1261
1262 dout("dentry_lru_touch %p %p '%.*s' (offset %lld)\n", di, dn,
1263 dn->d_name.len, dn->d_name.name, di->offset);
1264 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1265 spin_lock(&mdsc->dentry_lru_lock);
1266 list_move_tail(&di->lru, &mdsc->dentry_lru);
1267 spin_unlock(&mdsc->dentry_lru_lock);
1268}
1269
1270void ceph_dentry_lru_del(struct dentry *dn)
1271{
1272 struct ceph_dentry_info *di = ceph_dentry(dn);
1273 struct ceph_mds_client *mdsc;
1274
1275 dout("dentry_lru_del %p %p '%.*s'\n", di, dn,
1276 dn->d_name.len, dn->d_name.name);
1277 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1278 spin_lock(&mdsc->dentry_lru_lock);
1279 list_del_init(&di->lru);
1280 mdsc->num_dentry--;
1281 spin_unlock(&mdsc->dentry_lru_lock);
1282}
1283
1284/*
1285 * Return name hash for a given dentry. This is dependent on
1286 * the parent directory's hash function.
1287 */
1288unsigned ceph_dentry_hash(struct inode *dir, struct dentry *dn)
1289{
1290 struct ceph_inode_info *dci = ceph_inode(dir);
1291
1292 switch (dci->i_dir_layout.dl_dir_hash) {
1293 case 0: /* for backward compat */
1294 case CEPH_STR_HASH_LINUX:
1295 return dn->d_name.hash;
1296
1297 default:
1298 return ceph_str_hash(dci->i_dir_layout.dl_dir_hash,
1299 dn->d_name.name, dn->d_name.len);
1300 }
1301}
1302
1303const struct file_operations ceph_dir_fops = {
1304 .read = ceph_read_dir,
1305 .iterate = ceph_readdir,
1306 .llseek = ceph_dir_llseek,
1307 .open = ceph_open,
1308 .release = ceph_release,
1309 .unlocked_ioctl = ceph_ioctl,
1310 .fsync = ceph_dir_fsync,
1311};
1312
1313const struct inode_operations ceph_dir_iops = {
1314 .lookup = ceph_lookup,
1315 .permission = ceph_permission,
1316 .getattr = ceph_getattr,
1317 .setattr = ceph_setattr,
1318 .setxattr = ceph_setxattr,
1319 .getxattr = ceph_getxattr,
1320 .listxattr = ceph_listxattr,
1321 .removexattr = ceph_removexattr,
1322 .get_acl = ceph_get_acl,
1323 .set_acl = ceph_set_acl,
1324 .mknod = ceph_mknod,
1325 .symlink = ceph_symlink,
1326 .mkdir = ceph_mkdir,
1327 .link = ceph_link,
1328 .unlink = ceph_unlink,
1329 .rmdir = ceph_unlink,
1330 .rename = ceph_rename,
1331 .create = ceph_create,
1332 .atomic_open = ceph_atomic_open,
1333};
1334
1335const struct dentry_operations ceph_dentry_ops = {
1336 .d_revalidate = ceph_d_revalidate,
1337 .d_release = ceph_d_release,
1338 .d_prune = ceph_d_prune,
1339};
1340
1341const struct dentry_operations ceph_snapdir_dentry_ops = {
1342 .d_revalidate = ceph_snapdir_d_revalidate,
1343 .d_release = ceph_d_release,
1344};
1345
1346const struct dentry_operations ceph_snap_dentry_ops = {
1347 .d_release = ceph_d_release,
1348 .d_prune = ceph_d_prune,
1349};
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#include "crypto.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
31const struct dentry_operations ceph_dentry_ops;
32
33static bool __dentry_lease_is_valid(struct ceph_dentry_info *di);
34static int __dir_lease_try_check(const struct dentry *dentry);
35
36/*
37 * Initialize ceph dentry state.
38 */
39static int ceph_d_init(struct dentry *dentry)
40{
41 struct ceph_dentry_info *di;
42 struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(dentry->d_sb);
43
44 di = kmem_cache_zalloc(ceph_dentry_cachep, GFP_KERNEL);
45 if (!di)
46 return -ENOMEM; /* oh well */
47
48 di->dentry = dentry;
49 di->lease_session = NULL;
50 di->time = jiffies;
51 dentry->d_fsdata = di;
52 INIT_LIST_HEAD(&di->lease_list);
53
54 atomic64_inc(&mdsc->metric.total_dentries);
55
56 return 0;
57}
58
59/*
60 * for f_pos for readdir:
61 * - hash order:
62 * (0xff << 52) | ((24 bits hash) << 28) |
63 * (the nth entry has hash collision);
64 * - frag+name order;
65 * ((frag value) << 28) | (the nth entry in frag);
66 */
67#define OFFSET_BITS 28
68#define OFFSET_MASK ((1 << OFFSET_BITS) - 1)
69#define HASH_ORDER (0xffull << (OFFSET_BITS + 24))
70loff_t ceph_make_fpos(unsigned high, unsigned off, bool hash_order)
71{
72 loff_t fpos = ((loff_t)high << 28) | (loff_t)off;
73 if (hash_order)
74 fpos |= HASH_ORDER;
75 return fpos;
76}
77
78static bool is_hash_order(loff_t p)
79{
80 return (p & HASH_ORDER) == HASH_ORDER;
81}
82
83static unsigned fpos_frag(loff_t p)
84{
85 return p >> OFFSET_BITS;
86}
87
88static unsigned fpos_hash(loff_t p)
89{
90 return ceph_frag_value(fpos_frag(p));
91}
92
93static unsigned fpos_off(loff_t p)
94{
95 return p & OFFSET_MASK;
96}
97
98static int fpos_cmp(loff_t l, loff_t r)
99{
100 int v = ceph_frag_compare(fpos_frag(l), fpos_frag(r));
101 if (v)
102 return v;
103 return (int)(fpos_off(l) - fpos_off(r));
104}
105
106/*
107 * make note of the last dentry we read, so we can
108 * continue at the same lexicographical point,
109 * regardless of what dir changes take place on the
110 * server.
111 */
112static int note_last_dentry(struct ceph_fs_client *fsc,
113 struct ceph_dir_file_info *dfi,
114 const char *name,
115 int len, unsigned next_offset)
116{
117 char *buf = kmalloc(len+1, GFP_KERNEL);
118 if (!buf)
119 return -ENOMEM;
120 kfree(dfi->last_name);
121 dfi->last_name = buf;
122 memcpy(dfi->last_name, name, len);
123 dfi->last_name[len] = 0;
124 dfi->next_offset = next_offset;
125 doutc(fsc->client, "'%s'\n", dfi->last_name);
126 return 0;
127}
128
129
130static struct dentry *
131__dcache_find_get_entry(struct dentry *parent, u64 idx,
132 struct ceph_readdir_cache_control *cache_ctl)
133{
134 struct inode *dir = d_inode(parent);
135 struct ceph_client *cl = ceph_inode_to_client(dir);
136 struct dentry *dentry;
137 unsigned idx_mask = (PAGE_SIZE / sizeof(struct dentry *)) - 1;
138 loff_t ptr_pos = idx * sizeof(struct dentry *);
139 pgoff_t ptr_pgoff = ptr_pos >> PAGE_SHIFT;
140
141 if (ptr_pos >= i_size_read(dir))
142 return NULL;
143
144 if (!cache_ctl->page || ptr_pgoff != page_index(cache_ctl->page)) {
145 ceph_readdir_cache_release(cache_ctl);
146 cache_ctl->page = find_lock_page(&dir->i_data, ptr_pgoff);
147 if (!cache_ctl->page) {
148 doutc(cl, " page %lu not found\n", ptr_pgoff);
149 return ERR_PTR(-EAGAIN);
150 }
151 /* reading/filling the cache are serialized by
152 i_rwsem, no need to use page lock */
153 unlock_page(cache_ctl->page);
154 cache_ctl->dentries = kmap(cache_ctl->page);
155 }
156
157 cache_ctl->index = idx & idx_mask;
158
159 rcu_read_lock();
160 spin_lock(&parent->d_lock);
161 /* check i_size again here, because empty directory can be
162 * marked as complete while not holding the i_rwsem. */
163 if (ceph_dir_is_complete_ordered(dir) && ptr_pos < i_size_read(dir))
164 dentry = cache_ctl->dentries[cache_ctl->index];
165 else
166 dentry = NULL;
167 spin_unlock(&parent->d_lock);
168 if (dentry && !lockref_get_not_dead(&dentry->d_lockref))
169 dentry = NULL;
170 rcu_read_unlock();
171 return dentry ? : ERR_PTR(-EAGAIN);
172}
173
174/*
175 * When possible, we try to satisfy a readdir by peeking at the
176 * dcache. We make this work by carefully ordering dentries on
177 * d_children when we initially get results back from the MDS, and
178 * falling back to a "normal" sync readdir if any dentries in the dir
179 * are dropped.
180 *
181 * Complete dir indicates that we have all dentries in the dir. It is
182 * defined IFF we hold CEPH_CAP_FILE_SHARED (which will be revoked by
183 * the MDS if/when the directory is modified).
184 */
185static int __dcache_readdir(struct file *file, struct dir_context *ctx,
186 int shared_gen)
187{
188 struct ceph_dir_file_info *dfi = file->private_data;
189 struct dentry *parent = file->f_path.dentry;
190 struct inode *dir = d_inode(parent);
191 struct ceph_fs_client *fsc = ceph_inode_to_fs_client(dir);
192 struct ceph_client *cl = ceph_inode_to_client(dir);
193 struct dentry *dentry, *last = NULL;
194 struct ceph_dentry_info *di;
195 struct ceph_readdir_cache_control cache_ctl = {};
196 u64 idx = 0;
197 int err = 0;
198
199 doutc(cl, "%p %llx.%llx v%u at %llx\n", dir, ceph_vinop(dir),
200 (unsigned)shared_gen, ctx->pos);
201
202 /* search start position */
203 if (ctx->pos > 2) {
204 u64 count = div_u64(i_size_read(dir), sizeof(struct dentry *));
205 while (count > 0) {
206 u64 step = count >> 1;
207 dentry = __dcache_find_get_entry(parent, idx + step,
208 &cache_ctl);
209 if (!dentry) {
210 /* use linar search */
211 idx = 0;
212 break;
213 }
214 if (IS_ERR(dentry)) {
215 err = PTR_ERR(dentry);
216 goto out;
217 }
218 di = ceph_dentry(dentry);
219 spin_lock(&dentry->d_lock);
220 if (fpos_cmp(di->offset, ctx->pos) < 0) {
221 idx += step + 1;
222 count -= step + 1;
223 } else {
224 count = step;
225 }
226 spin_unlock(&dentry->d_lock);
227 dput(dentry);
228 }
229
230 doutc(cl, "%p %llx.%llx cache idx %llu\n", dir,
231 ceph_vinop(dir), idx);
232 }
233
234
235 for (;;) {
236 bool emit_dentry = false;
237 dentry = __dcache_find_get_entry(parent, idx++, &cache_ctl);
238 if (!dentry) {
239 dfi->file_info.flags |= CEPH_F_ATEND;
240 err = 0;
241 break;
242 }
243 if (IS_ERR(dentry)) {
244 err = PTR_ERR(dentry);
245 goto out;
246 }
247
248 spin_lock(&dentry->d_lock);
249 di = ceph_dentry(dentry);
250 if (d_unhashed(dentry) ||
251 d_really_is_negative(dentry) ||
252 di->lease_shared_gen != shared_gen ||
253 ((dentry->d_flags & DCACHE_NOKEY_NAME) &&
254 fscrypt_has_encryption_key(dir))) {
255 spin_unlock(&dentry->d_lock);
256 dput(dentry);
257 err = -EAGAIN;
258 goto out;
259 }
260 if (fpos_cmp(ctx->pos, di->offset) <= 0) {
261 __ceph_dentry_dir_lease_touch(di);
262 emit_dentry = true;
263 }
264 spin_unlock(&dentry->d_lock);
265
266 if (emit_dentry) {
267 doutc(cl, " %llx dentry %p %pd %p\n", di->offset,
268 dentry, dentry, d_inode(dentry));
269 ctx->pos = di->offset;
270 if (!dir_emit(ctx, dentry->d_name.name,
271 dentry->d_name.len, ceph_present_inode(d_inode(dentry)),
272 d_inode(dentry)->i_mode >> 12)) {
273 dput(dentry);
274 err = 0;
275 break;
276 }
277 ctx->pos++;
278
279 if (last)
280 dput(last);
281 last = dentry;
282 } else {
283 dput(dentry);
284 }
285 }
286out:
287 ceph_readdir_cache_release(&cache_ctl);
288 if (last) {
289 int ret;
290 di = ceph_dentry(last);
291 ret = note_last_dentry(fsc, dfi, last->d_name.name,
292 last->d_name.len,
293 fpos_off(di->offset) + 1);
294 if (ret < 0)
295 err = ret;
296 dput(last);
297 /* last_name no longer match cache index */
298 if (dfi->readdir_cache_idx >= 0) {
299 dfi->readdir_cache_idx = -1;
300 dfi->dir_release_count = 0;
301 }
302 }
303 return err;
304}
305
306static bool need_send_readdir(struct ceph_dir_file_info *dfi, loff_t pos)
307{
308 if (!dfi->last_readdir)
309 return true;
310 if (is_hash_order(pos))
311 return !ceph_frag_contains_value(dfi->frag, fpos_hash(pos));
312 else
313 return dfi->frag != fpos_frag(pos);
314}
315
316static int ceph_readdir(struct file *file, struct dir_context *ctx)
317{
318 struct ceph_dir_file_info *dfi = file->private_data;
319 struct inode *inode = file_inode(file);
320 struct ceph_inode_info *ci = ceph_inode(inode);
321 struct ceph_fs_client *fsc = ceph_inode_to_fs_client(inode);
322 struct ceph_mds_client *mdsc = fsc->mdsc;
323 struct ceph_client *cl = fsc->client;
324 int i;
325 int err;
326 unsigned frag = -1;
327 struct ceph_mds_reply_info_parsed *rinfo;
328
329 doutc(cl, "%p %llx.%llx file %p pos %llx\n", inode,
330 ceph_vinop(inode), file, ctx->pos);
331 if (dfi->file_info.flags & CEPH_F_ATEND)
332 return 0;
333
334 /* always start with . and .. */
335 if (ctx->pos == 0) {
336 doutc(cl, "%p %llx.%llx off 0 -> '.'\n", inode,
337 ceph_vinop(inode));
338 if (!dir_emit(ctx, ".", 1, ceph_present_inode(inode),
339 inode->i_mode >> 12))
340 return 0;
341 ctx->pos = 1;
342 }
343 if (ctx->pos == 1) {
344 u64 ino;
345 struct dentry *dentry = file->f_path.dentry;
346
347 spin_lock(&dentry->d_lock);
348 ino = ceph_present_inode(dentry->d_parent->d_inode);
349 spin_unlock(&dentry->d_lock);
350
351 doutc(cl, "%p %llx.%llx off 1 -> '..'\n", inode,
352 ceph_vinop(inode));
353 if (!dir_emit(ctx, "..", 2, ino, inode->i_mode >> 12))
354 return 0;
355 ctx->pos = 2;
356 }
357
358 err = ceph_fscrypt_prepare_readdir(inode);
359 if (err < 0)
360 return err;
361
362 spin_lock(&ci->i_ceph_lock);
363 /* request Fx cap. if have Fx, we don't need to release Fs cap
364 * for later create/unlink. */
365 __ceph_touch_fmode(ci, mdsc, CEPH_FILE_MODE_WR);
366 /* can we use the dcache? */
367 if (ceph_test_mount_opt(fsc, DCACHE) &&
368 !ceph_test_mount_opt(fsc, NOASYNCREADDIR) &&
369 ceph_snap(inode) != CEPH_SNAPDIR &&
370 __ceph_dir_is_complete_ordered(ci) &&
371 __ceph_caps_issued_mask_metric(ci, CEPH_CAP_FILE_SHARED, 1)) {
372 int shared_gen = atomic_read(&ci->i_shared_gen);
373
374 spin_unlock(&ci->i_ceph_lock);
375 err = __dcache_readdir(file, ctx, shared_gen);
376 if (err != -EAGAIN)
377 return err;
378 } else {
379 spin_unlock(&ci->i_ceph_lock);
380 }
381
382 /* proceed with a normal readdir */
383more:
384 /* do we have the correct frag content buffered? */
385 if (need_send_readdir(dfi, ctx->pos)) {
386 struct ceph_mds_request *req;
387 int op = ceph_snap(inode) == CEPH_SNAPDIR ?
388 CEPH_MDS_OP_LSSNAP : CEPH_MDS_OP_READDIR;
389
390 /* discard old result, if any */
391 if (dfi->last_readdir) {
392 ceph_mdsc_put_request(dfi->last_readdir);
393 dfi->last_readdir = NULL;
394 }
395
396 if (is_hash_order(ctx->pos)) {
397 /* fragtree isn't always accurate. choose frag
398 * based on previous reply when possible. */
399 if (frag == (unsigned)-1)
400 frag = ceph_choose_frag(ci, fpos_hash(ctx->pos),
401 NULL, NULL);
402 } else {
403 frag = fpos_frag(ctx->pos);
404 }
405
406 doutc(cl, "fetching %p %llx.%llx frag %x offset '%s'\n",
407 inode, ceph_vinop(inode), frag, dfi->last_name);
408 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
409 if (IS_ERR(req))
410 return PTR_ERR(req);
411
412 err = ceph_alloc_readdir_reply_buffer(req, inode);
413 if (err) {
414 ceph_mdsc_put_request(req);
415 return err;
416 }
417 /* hints to request -> mds selection code */
418 req->r_direct_mode = USE_AUTH_MDS;
419 if (op == CEPH_MDS_OP_READDIR) {
420 req->r_direct_hash = ceph_frag_value(frag);
421 __set_bit(CEPH_MDS_R_DIRECT_IS_HASH, &req->r_req_flags);
422 req->r_inode_drop = CEPH_CAP_FILE_EXCL;
423 }
424 if (dfi->last_name) {
425 struct qstr d_name = { .name = dfi->last_name,
426 .len = strlen(dfi->last_name) };
427
428 req->r_path2 = kzalloc(NAME_MAX + 1, GFP_KERNEL);
429 if (!req->r_path2) {
430 ceph_mdsc_put_request(req);
431 return -ENOMEM;
432 }
433
434 err = ceph_encode_encrypted_dname(inode, &d_name,
435 req->r_path2);
436 if (err < 0) {
437 ceph_mdsc_put_request(req);
438 return err;
439 }
440 } else if (is_hash_order(ctx->pos)) {
441 req->r_args.readdir.offset_hash =
442 cpu_to_le32(fpos_hash(ctx->pos));
443 }
444
445 req->r_dir_release_cnt = dfi->dir_release_count;
446 req->r_dir_ordered_cnt = dfi->dir_ordered_count;
447 req->r_readdir_cache_idx = dfi->readdir_cache_idx;
448 req->r_readdir_offset = dfi->next_offset;
449 req->r_args.readdir.frag = cpu_to_le32(frag);
450 req->r_args.readdir.flags =
451 cpu_to_le16(CEPH_READDIR_REPLY_BITFLAGS);
452
453 req->r_inode = inode;
454 ihold(inode);
455 req->r_dentry = dget(file->f_path.dentry);
456 err = ceph_mdsc_do_request(mdsc, NULL, req);
457 if (err < 0) {
458 ceph_mdsc_put_request(req);
459 return err;
460 }
461 doutc(cl, "%p %llx.%llx got and parsed readdir result=%d"
462 "on frag %x, end=%d, complete=%d, hash_order=%d\n",
463 inode, ceph_vinop(inode), err, frag,
464 (int)req->r_reply_info.dir_end,
465 (int)req->r_reply_info.dir_complete,
466 (int)req->r_reply_info.hash_order);
467
468 rinfo = &req->r_reply_info;
469 if (le32_to_cpu(rinfo->dir_dir->frag) != frag) {
470 frag = le32_to_cpu(rinfo->dir_dir->frag);
471 if (!rinfo->hash_order) {
472 dfi->next_offset = req->r_readdir_offset;
473 /* adjust ctx->pos to beginning of frag */
474 ctx->pos = ceph_make_fpos(frag,
475 dfi->next_offset,
476 false);
477 }
478 }
479
480 dfi->frag = frag;
481 dfi->last_readdir = req;
482
483 if (test_bit(CEPH_MDS_R_DID_PREPOPULATE, &req->r_req_flags)) {
484 dfi->readdir_cache_idx = req->r_readdir_cache_idx;
485 if (dfi->readdir_cache_idx < 0) {
486 /* preclude from marking dir ordered */
487 dfi->dir_ordered_count = 0;
488 } else if (ceph_frag_is_leftmost(frag) &&
489 dfi->next_offset == 2) {
490 /* note dir version at start of readdir so
491 * we can tell if any dentries get dropped */
492 dfi->dir_release_count = req->r_dir_release_cnt;
493 dfi->dir_ordered_count = req->r_dir_ordered_cnt;
494 }
495 } else {
496 doutc(cl, "%p %llx.%llx !did_prepopulate\n", inode,
497 ceph_vinop(inode));
498 /* disable readdir cache */
499 dfi->readdir_cache_idx = -1;
500 /* preclude from marking dir complete */
501 dfi->dir_release_count = 0;
502 }
503
504 /* note next offset and last dentry name */
505 if (rinfo->dir_nr > 0) {
506 struct ceph_mds_reply_dir_entry *rde =
507 rinfo->dir_entries + (rinfo->dir_nr-1);
508 unsigned next_offset = req->r_reply_info.dir_end ?
509 2 : (fpos_off(rde->offset) + 1);
510 err = note_last_dentry(fsc, dfi, rde->name,
511 rde->name_len, next_offset);
512 if (err) {
513 ceph_mdsc_put_request(dfi->last_readdir);
514 dfi->last_readdir = NULL;
515 return err;
516 }
517 } else if (req->r_reply_info.dir_end) {
518 dfi->next_offset = 2;
519 /* keep last name */
520 }
521 }
522
523 rinfo = &dfi->last_readdir->r_reply_info;
524 doutc(cl, "%p %llx.%llx frag %x num %d pos %llx chunk first %llx\n",
525 inode, ceph_vinop(inode), dfi->frag, rinfo->dir_nr, ctx->pos,
526 rinfo->dir_nr ? rinfo->dir_entries[0].offset : 0LL);
527
528 i = 0;
529 /* search start position */
530 if (rinfo->dir_nr > 0) {
531 int step, nr = rinfo->dir_nr;
532 while (nr > 0) {
533 step = nr >> 1;
534 if (rinfo->dir_entries[i + step].offset < ctx->pos) {
535 i += step + 1;
536 nr -= step + 1;
537 } else {
538 nr = step;
539 }
540 }
541 }
542 for (; i < rinfo->dir_nr; i++) {
543 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
544
545 if (rde->offset < ctx->pos) {
546 pr_warn_client(cl,
547 "%p %llx.%llx rde->offset 0x%llx ctx->pos 0x%llx\n",
548 inode, ceph_vinop(inode), rde->offset, ctx->pos);
549 return -EIO;
550 }
551
552 if (WARN_ON_ONCE(!rde->inode.in))
553 return -EIO;
554
555 ctx->pos = rde->offset;
556 doutc(cl, "%p %llx.%llx (%d/%d) -> %llx '%.*s' %p\n", inode,
557 ceph_vinop(inode), i, rinfo->dir_nr, ctx->pos,
558 rde->name_len, rde->name, &rde->inode.in);
559
560 if (!dir_emit(ctx, rde->name, rde->name_len,
561 ceph_present_ino(inode->i_sb, le64_to_cpu(rde->inode.in->ino)),
562 le32_to_cpu(rde->inode.in->mode) >> 12)) {
563 /*
564 * NOTE: Here no need to put the 'dfi->last_readdir',
565 * because when dir_emit stops us it's most likely
566 * doesn't have enough memory, etc. So for next readdir
567 * it will continue.
568 */
569 doutc(cl, "filldir stopping us...\n");
570 return 0;
571 }
572
573 /* Reset the lengths to their original allocated vals */
574 ctx->pos++;
575 }
576
577 ceph_mdsc_put_request(dfi->last_readdir);
578 dfi->last_readdir = NULL;
579
580 if (dfi->next_offset > 2) {
581 frag = dfi->frag;
582 goto more;
583 }
584
585 /* more frags? */
586 if (!ceph_frag_is_rightmost(dfi->frag)) {
587 frag = ceph_frag_next(dfi->frag);
588 if (is_hash_order(ctx->pos)) {
589 loff_t new_pos = ceph_make_fpos(ceph_frag_value(frag),
590 dfi->next_offset, true);
591 if (new_pos > ctx->pos)
592 ctx->pos = new_pos;
593 /* keep last_name */
594 } else {
595 ctx->pos = ceph_make_fpos(frag, dfi->next_offset,
596 false);
597 kfree(dfi->last_name);
598 dfi->last_name = NULL;
599 }
600 doutc(cl, "%p %llx.%llx next frag is %x\n", inode,
601 ceph_vinop(inode), frag);
602 goto more;
603 }
604 dfi->file_info.flags |= CEPH_F_ATEND;
605
606 /*
607 * if dir_release_count still matches the dir, no dentries
608 * were released during the whole readdir, and we should have
609 * the complete dir contents in our cache.
610 */
611 if (atomic64_read(&ci->i_release_count) ==
612 dfi->dir_release_count) {
613 spin_lock(&ci->i_ceph_lock);
614 if (dfi->dir_ordered_count ==
615 atomic64_read(&ci->i_ordered_count)) {
616 doutc(cl, " marking %p %llx.%llx complete and ordered\n",
617 inode, ceph_vinop(inode));
618 /* use i_size to track number of entries in
619 * readdir cache */
620 BUG_ON(dfi->readdir_cache_idx < 0);
621 i_size_write(inode, dfi->readdir_cache_idx *
622 sizeof(struct dentry*));
623 } else {
624 doutc(cl, " marking %llx.%llx complete\n",
625 ceph_vinop(inode));
626 }
627 __ceph_dir_set_complete(ci, dfi->dir_release_count,
628 dfi->dir_ordered_count);
629 spin_unlock(&ci->i_ceph_lock);
630 }
631 doutc(cl, "%p %llx.%llx file %p done.\n", inode, ceph_vinop(inode),
632 file);
633 return 0;
634}
635
636static void reset_readdir(struct ceph_dir_file_info *dfi)
637{
638 if (dfi->last_readdir) {
639 ceph_mdsc_put_request(dfi->last_readdir);
640 dfi->last_readdir = NULL;
641 }
642 kfree(dfi->last_name);
643 dfi->last_name = NULL;
644 dfi->dir_release_count = 0;
645 dfi->readdir_cache_idx = -1;
646 dfi->next_offset = 2; /* compensate for . and .. */
647 dfi->file_info.flags &= ~CEPH_F_ATEND;
648}
649
650/*
651 * discard buffered readdir content on seekdir(0), or seek to new frag,
652 * or seek prior to current chunk
653 */
654static bool need_reset_readdir(struct ceph_dir_file_info *dfi, loff_t new_pos)
655{
656 struct ceph_mds_reply_info_parsed *rinfo;
657 loff_t chunk_offset;
658 if (new_pos == 0)
659 return true;
660 if (is_hash_order(new_pos)) {
661 /* no need to reset last_name for a forward seek when
662 * dentries are sotred in hash order */
663 } else if (dfi->frag != fpos_frag(new_pos)) {
664 return true;
665 }
666 rinfo = dfi->last_readdir ? &dfi->last_readdir->r_reply_info : NULL;
667 if (!rinfo || !rinfo->dir_nr)
668 return true;
669 chunk_offset = rinfo->dir_entries[0].offset;
670 return new_pos < chunk_offset ||
671 is_hash_order(new_pos) != is_hash_order(chunk_offset);
672}
673
674static loff_t ceph_dir_llseek(struct file *file, loff_t offset, int whence)
675{
676 struct ceph_dir_file_info *dfi = file->private_data;
677 struct inode *inode = file->f_mapping->host;
678 struct ceph_client *cl = ceph_inode_to_client(inode);
679 loff_t retval;
680
681 inode_lock(inode);
682 retval = -EINVAL;
683 switch (whence) {
684 case SEEK_CUR:
685 offset += file->f_pos;
686 break;
687 case SEEK_SET:
688 break;
689 case SEEK_END:
690 retval = -EOPNOTSUPP;
691 goto out;
692 default:
693 goto out;
694 }
695
696 if (offset >= 0) {
697 if (need_reset_readdir(dfi, offset)) {
698 doutc(cl, "%p %llx.%llx dropping %p content\n",
699 inode, ceph_vinop(inode), file);
700 reset_readdir(dfi);
701 } else if (is_hash_order(offset) && offset > file->f_pos) {
702 /* for hash offset, we don't know if a forward seek
703 * is within same frag */
704 dfi->dir_release_count = 0;
705 dfi->readdir_cache_idx = -1;
706 }
707
708 if (offset != file->f_pos) {
709 file->f_pos = offset;
710 file->f_version = 0;
711 dfi->file_info.flags &= ~CEPH_F_ATEND;
712 }
713 retval = offset;
714 }
715out:
716 inode_unlock(inode);
717 return retval;
718}
719
720/*
721 * Handle lookups for the hidden .snap directory.
722 */
723struct dentry *ceph_handle_snapdir(struct ceph_mds_request *req,
724 struct dentry *dentry)
725{
726 struct ceph_fs_client *fsc = ceph_sb_to_fs_client(dentry->d_sb);
727 struct inode *parent = d_inode(dentry->d_parent); /* we hold i_rwsem */
728 struct ceph_client *cl = ceph_inode_to_client(parent);
729
730 /* .snap dir? */
731 if (ceph_snap(parent) == CEPH_NOSNAP &&
732 strcmp(dentry->d_name.name, fsc->mount_options->snapdir_name) == 0) {
733 struct dentry *res;
734 struct inode *inode = ceph_get_snapdir(parent);
735
736 res = d_splice_alias(inode, dentry);
737 doutc(cl, "ENOENT on snapdir %p '%pd', linking to "
738 "snapdir %p %llx.%llx. Spliced dentry %p\n",
739 dentry, dentry, inode, ceph_vinop(inode), res);
740 if (res)
741 dentry = res;
742 }
743 return dentry;
744}
745
746/*
747 * Figure out final result of a lookup/open request.
748 *
749 * Mainly, make sure we return the final req->r_dentry (if it already
750 * existed) in place of the original VFS-provided dentry when they
751 * differ.
752 *
753 * Gracefully handle the case where the MDS replies with -ENOENT and
754 * no trace (which it may do, at its discretion, e.g., if it doesn't
755 * care to issue a lease on the negative dentry).
756 */
757struct dentry *ceph_finish_lookup(struct ceph_mds_request *req,
758 struct dentry *dentry, int err)
759{
760 struct ceph_client *cl = req->r_mdsc->fsc->client;
761
762 if (err == -ENOENT) {
763 /* no trace? */
764 err = 0;
765 if (!req->r_reply_info.head->is_dentry) {
766 doutc(cl,
767 "ENOENT and no trace, dentry %p inode %llx.%llx\n",
768 dentry, ceph_vinop(d_inode(dentry)));
769 if (d_really_is_positive(dentry)) {
770 d_drop(dentry);
771 err = -ENOENT;
772 } else {
773 d_add(dentry, NULL);
774 }
775 }
776 }
777 if (err)
778 dentry = ERR_PTR(err);
779 else if (dentry != req->r_dentry)
780 dentry = dget(req->r_dentry); /* we got spliced */
781 else
782 dentry = NULL;
783 return dentry;
784}
785
786static bool is_root_ceph_dentry(struct inode *inode, struct dentry *dentry)
787{
788 return ceph_ino(inode) == CEPH_INO_ROOT &&
789 strncmp(dentry->d_name.name, ".ceph", 5) == 0;
790}
791
792/*
793 * Look up a single dir entry. If there is a lookup intent, inform
794 * the MDS so that it gets our 'caps wanted' value in a single op.
795 */
796static struct dentry *ceph_lookup(struct inode *dir, struct dentry *dentry,
797 unsigned int flags)
798{
799 struct ceph_fs_client *fsc = ceph_sb_to_fs_client(dir->i_sb);
800 struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(dir->i_sb);
801 struct ceph_client *cl = fsc->client;
802 struct ceph_mds_request *req;
803 int op;
804 int mask;
805 int err;
806
807 doutc(cl, "%p %llx.%llx/'%pd' dentry %p\n", dir, ceph_vinop(dir),
808 dentry, dentry);
809
810 if (dentry->d_name.len > NAME_MAX)
811 return ERR_PTR(-ENAMETOOLONG);
812
813 if (IS_ENCRYPTED(dir)) {
814 bool had_key = fscrypt_has_encryption_key(dir);
815
816 err = fscrypt_prepare_lookup_partial(dir, dentry);
817 if (err < 0)
818 return ERR_PTR(err);
819
820 /* mark directory as incomplete if it has been unlocked */
821 if (!had_key && fscrypt_has_encryption_key(dir))
822 ceph_dir_clear_complete(dir);
823 }
824
825 /* can we conclude ENOENT locally? */
826 if (d_really_is_negative(dentry)) {
827 struct ceph_inode_info *ci = ceph_inode(dir);
828 struct ceph_dentry_info *di = ceph_dentry(dentry);
829
830 spin_lock(&ci->i_ceph_lock);
831 doutc(cl, " dir %llx.%llx flags are 0x%lx\n",
832 ceph_vinop(dir), ci->i_ceph_flags);
833 if (strncmp(dentry->d_name.name,
834 fsc->mount_options->snapdir_name,
835 dentry->d_name.len) &&
836 !is_root_ceph_dentry(dir, dentry) &&
837 ceph_test_mount_opt(fsc, DCACHE) &&
838 __ceph_dir_is_complete(ci) &&
839 __ceph_caps_issued_mask_metric(ci, CEPH_CAP_FILE_SHARED, 1)) {
840 __ceph_touch_fmode(ci, mdsc, CEPH_FILE_MODE_RD);
841 spin_unlock(&ci->i_ceph_lock);
842 doutc(cl, " dir %llx.%llx complete, -ENOENT\n",
843 ceph_vinop(dir));
844 d_add(dentry, NULL);
845 di->lease_shared_gen = atomic_read(&ci->i_shared_gen);
846 return NULL;
847 }
848 spin_unlock(&ci->i_ceph_lock);
849 }
850
851 op = ceph_snap(dir) == CEPH_SNAPDIR ?
852 CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP;
853 req = ceph_mdsc_create_request(mdsc, op, USE_ANY_MDS);
854 if (IS_ERR(req))
855 return ERR_CAST(req);
856 req->r_dentry = dget(dentry);
857 req->r_num_caps = 2;
858
859 mask = CEPH_STAT_CAP_INODE | CEPH_CAP_AUTH_SHARED;
860 if (ceph_security_xattr_wanted(dir))
861 mask |= CEPH_CAP_XATTR_SHARED;
862 req->r_args.getattr.mask = cpu_to_le32(mask);
863
864 ihold(dir);
865 req->r_parent = dir;
866 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
867 err = ceph_mdsc_do_request(mdsc, NULL, req);
868 if (err == -ENOENT) {
869 struct dentry *res;
870
871 res = ceph_handle_snapdir(req, dentry);
872 if (IS_ERR(res)) {
873 err = PTR_ERR(res);
874 } else {
875 dentry = res;
876 err = 0;
877 }
878 }
879 dentry = ceph_finish_lookup(req, dentry, err);
880 ceph_mdsc_put_request(req); /* will dput(dentry) */
881 doutc(cl, "result=%p\n", dentry);
882 return dentry;
883}
884
885/*
886 * If we do a create but get no trace back from the MDS, follow up with
887 * a lookup (the VFS expects us to link up the provided dentry).
888 */
889int ceph_handle_notrace_create(struct inode *dir, struct dentry *dentry)
890{
891 struct dentry *result = ceph_lookup(dir, dentry, 0);
892
893 if (result && !IS_ERR(result)) {
894 /*
895 * We created the item, then did a lookup, and found
896 * it was already linked to another inode we already
897 * had in our cache (and thus got spliced). To not
898 * confuse VFS (especially when inode is a directory),
899 * we don't link our dentry to that inode, return an
900 * error instead.
901 *
902 * This event should be rare and it happens only when
903 * we talk to old MDS. Recent MDS does not send traceless
904 * reply for request that creates new inode.
905 */
906 d_drop(result);
907 return -ESTALE;
908 }
909 return PTR_ERR(result);
910}
911
912static int ceph_mknod(struct mnt_idmap *idmap, struct inode *dir,
913 struct dentry *dentry, umode_t mode, dev_t rdev)
914{
915 struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(dir->i_sb);
916 struct ceph_client *cl = mdsc->fsc->client;
917 struct ceph_mds_request *req;
918 struct ceph_acl_sec_ctx as_ctx = {};
919 int err;
920
921 if (ceph_snap(dir) != CEPH_NOSNAP)
922 return -EROFS;
923
924 err = ceph_wait_on_conflict_unlink(dentry);
925 if (err)
926 return err;
927
928 if (ceph_quota_is_max_files_exceeded(dir)) {
929 err = -EDQUOT;
930 goto out;
931 }
932
933 doutc(cl, "%p %llx.%llx/'%pd' dentry %p mode 0%ho rdev %d\n",
934 dir, ceph_vinop(dir), dentry, dentry, mode, rdev);
935 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_MKNOD, USE_AUTH_MDS);
936 if (IS_ERR(req)) {
937 err = PTR_ERR(req);
938 goto out;
939 }
940
941 req->r_new_inode = ceph_new_inode(dir, dentry, &mode, &as_ctx);
942 if (IS_ERR(req->r_new_inode)) {
943 err = PTR_ERR(req->r_new_inode);
944 req->r_new_inode = NULL;
945 goto out_req;
946 }
947
948 if (S_ISREG(mode) && IS_ENCRYPTED(dir))
949 set_bit(CEPH_MDS_R_FSCRYPT_FILE, &req->r_req_flags);
950
951 req->r_dentry = dget(dentry);
952 req->r_num_caps = 2;
953 req->r_parent = dir;
954 ihold(dir);
955 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
956 req->r_mnt_idmap = mnt_idmap_get(idmap);
957 req->r_args.mknod.mode = cpu_to_le32(mode);
958 req->r_args.mknod.rdev = cpu_to_le32(rdev);
959 req->r_dentry_drop = CEPH_CAP_FILE_SHARED | CEPH_CAP_AUTH_EXCL |
960 CEPH_CAP_XATTR_EXCL;
961 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
962
963 ceph_as_ctx_to_req(req, &as_ctx);
964
965 err = ceph_mdsc_do_request(mdsc, dir, req);
966 if (!err && !req->r_reply_info.head->is_dentry)
967 err = ceph_handle_notrace_create(dir, dentry);
968out_req:
969 ceph_mdsc_put_request(req);
970out:
971 if (!err)
972 ceph_init_inode_acls(d_inode(dentry), &as_ctx);
973 else
974 d_drop(dentry);
975 ceph_release_acl_sec_ctx(&as_ctx);
976 return err;
977}
978
979static int ceph_create(struct mnt_idmap *idmap, struct inode *dir,
980 struct dentry *dentry, umode_t mode, bool excl)
981{
982 return ceph_mknod(idmap, dir, dentry, mode, 0);
983}
984
985#if IS_ENABLED(CONFIG_FS_ENCRYPTION)
986static int prep_encrypted_symlink_target(struct ceph_mds_request *req,
987 const char *dest)
988{
989 int err;
990 int len = strlen(dest);
991 struct fscrypt_str osd_link = FSTR_INIT(NULL, 0);
992
993 err = fscrypt_prepare_symlink(req->r_parent, dest, len, PATH_MAX,
994 &osd_link);
995 if (err)
996 goto out;
997
998 err = fscrypt_encrypt_symlink(req->r_new_inode, dest, len, &osd_link);
999 if (err)
1000 goto out;
1001
1002 req->r_path2 = kmalloc(CEPH_BASE64_CHARS(osd_link.len) + 1, GFP_KERNEL);
1003 if (!req->r_path2) {
1004 err = -ENOMEM;
1005 goto out;
1006 }
1007
1008 len = ceph_base64_encode(osd_link.name, osd_link.len, req->r_path2);
1009 req->r_path2[len] = '\0';
1010out:
1011 fscrypt_fname_free_buffer(&osd_link);
1012 return err;
1013}
1014#else
1015static int prep_encrypted_symlink_target(struct ceph_mds_request *req,
1016 const char *dest)
1017{
1018 return -EOPNOTSUPP;
1019}
1020#endif
1021
1022static int ceph_symlink(struct mnt_idmap *idmap, struct inode *dir,
1023 struct dentry *dentry, const char *dest)
1024{
1025 struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(dir->i_sb);
1026 struct ceph_client *cl = mdsc->fsc->client;
1027 struct ceph_mds_request *req;
1028 struct ceph_acl_sec_ctx as_ctx = {};
1029 umode_t mode = S_IFLNK | 0777;
1030 int err;
1031
1032 if (ceph_snap(dir) != CEPH_NOSNAP)
1033 return -EROFS;
1034
1035 err = ceph_wait_on_conflict_unlink(dentry);
1036 if (err)
1037 return err;
1038
1039 if (ceph_quota_is_max_files_exceeded(dir)) {
1040 err = -EDQUOT;
1041 goto out;
1042 }
1043
1044 doutc(cl, "%p %llx.%llx/'%pd' to '%s'\n", dir, ceph_vinop(dir), dentry,
1045 dest);
1046 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SYMLINK, USE_AUTH_MDS);
1047 if (IS_ERR(req)) {
1048 err = PTR_ERR(req);
1049 goto out;
1050 }
1051
1052 req->r_new_inode = ceph_new_inode(dir, dentry, &mode, &as_ctx);
1053 if (IS_ERR(req->r_new_inode)) {
1054 err = PTR_ERR(req->r_new_inode);
1055 req->r_new_inode = NULL;
1056 goto out_req;
1057 }
1058
1059 req->r_parent = dir;
1060 ihold(dir);
1061
1062 if (IS_ENCRYPTED(req->r_new_inode)) {
1063 err = prep_encrypted_symlink_target(req, dest);
1064 if (err)
1065 goto out_req;
1066 } else {
1067 req->r_path2 = kstrdup(dest, GFP_KERNEL);
1068 if (!req->r_path2) {
1069 err = -ENOMEM;
1070 goto out_req;
1071 }
1072 }
1073
1074 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
1075 req->r_mnt_idmap = mnt_idmap_get(idmap);
1076 req->r_dentry = dget(dentry);
1077 req->r_num_caps = 2;
1078 req->r_dentry_drop = CEPH_CAP_FILE_SHARED | CEPH_CAP_AUTH_EXCL |
1079 CEPH_CAP_XATTR_EXCL;
1080 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
1081
1082 ceph_as_ctx_to_req(req, &as_ctx);
1083
1084 err = ceph_mdsc_do_request(mdsc, dir, req);
1085 if (!err && !req->r_reply_info.head->is_dentry)
1086 err = ceph_handle_notrace_create(dir, dentry);
1087out_req:
1088 ceph_mdsc_put_request(req);
1089out:
1090 if (err)
1091 d_drop(dentry);
1092 ceph_release_acl_sec_ctx(&as_ctx);
1093 return err;
1094}
1095
1096static int ceph_mkdir(struct mnt_idmap *idmap, struct inode *dir,
1097 struct dentry *dentry, umode_t mode)
1098{
1099 struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(dir->i_sb);
1100 struct ceph_client *cl = mdsc->fsc->client;
1101 struct ceph_mds_request *req;
1102 struct ceph_acl_sec_ctx as_ctx = {};
1103 int err;
1104 int op;
1105
1106 err = ceph_wait_on_conflict_unlink(dentry);
1107 if (err)
1108 return err;
1109
1110 if (ceph_snap(dir) == CEPH_SNAPDIR) {
1111 /* mkdir .snap/foo is a MKSNAP */
1112 op = CEPH_MDS_OP_MKSNAP;
1113 doutc(cl, "mksnap %llx.%llx/'%pd' dentry %p\n",
1114 ceph_vinop(dir), dentry, dentry);
1115 } else if (ceph_snap(dir) == CEPH_NOSNAP) {
1116 doutc(cl, "mkdir %llx.%llx/'%pd' dentry %p mode 0%ho\n",
1117 ceph_vinop(dir), dentry, dentry, mode);
1118 op = CEPH_MDS_OP_MKDIR;
1119 } else {
1120 err = -EROFS;
1121 goto out;
1122 }
1123
1124 if (op == CEPH_MDS_OP_MKDIR &&
1125 ceph_quota_is_max_files_exceeded(dir)) {
1126 err = -EDQUOT;
1127 goto out;
1128 }
1129 if ((op == CEPH_MDS_OP_MKSNAP) && IS_ENCRYPTED(dir) &&
1130 !fscrypt_has_encryption_key(dir)) {
1131 err = -ENOKEY;
1132 goto out;
1133 }
1134
1135
1136 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
1137 if (IS_ERR(req)) {
1138 err = PTR_ERR(req);
1139 goto out;
1140 }
1141
1142 mode |= S_IFDIR;
1143 req->r_new_inode = ceph_new_inode(dir, dentry, &mode, &as_ctx);
1144 if (IS_ERR(req->r_new_inode)) {
1145 err = PTR_ERR(req->r_new_inode);
1146 req->r_new_inode = NULL;
1147 goto out_req;
1148 }
1149
1150 req->r_dentry = dget(dentry);
1151 req->r_num_caps = 2;
1152 req->r_parent = dir;
1153 ihold(dir);
1154 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
1155 if (op == CEPH_MDS_OP_MKDIR)
1156 req->r_mnt_idmap = mnt_idmap_get(idmap);
1157 req->r_args.mkdir.mode = cpu_to_le32(mode);
1158 req->r_dentry_drop = CEPH_CAP_FILE_SHARED | CEPH_CAP_AUTH_EXCL |
1159 CEPH_CAP_XATTR_EXCL;
1160 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
1161
1162 ceph_as_ctx_to_req(req, &as_ctx);
1163
1164 err = ceph_mdsc_do_request(mdsc, dir, req);
1165 if (!err &&
1166 !req->r_reply_info.head->is_target &&
1167 !req->r_reply_info.head->is_dentry)
1168 err = ceph_handle_notrace_create(dir, dentry);
1169out_req:
1170 ceph_mdsc_put_request(req);
1171out:
1172 if (!err)
1173 ceph_init_inode_acls(d_inode(dentry), &as_ctx);
1174 else
1175 d_drop(dentry);
1176 ceph_release_acl_sec_ctx(&as_ctx);
1177 return err;
1178}
1179
1180static int ceph_link(struct dentry *old_dentry, struct inode *dir,
1181 struct dentry *dentry)
1182{
1183 struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(dir->i_sb);
1184 struct ceph_client *cl = mdsc->fsc->client;
1185 struct ceph_mds_request *req;
1186 int err;
1187
1188 if (dentry->d_flags & DCACHE_DISCONNECTED)
1189 return -EINVAL;
1190
1191 err = ceph_wait_on_conflict_unlink(dentry);
1192 if (err)
1193 return err;
1194
1195 if (ceph_snap(dir) != CEPH_NOSNAP)
1196 return -EROFS;
1197
1198 err = fscrypt_prepare_link(old_dentry, dir, dentry);
1199 if (err)
1200 return err;
1201
1202 doutc(cl, "%p %llx.%llx/'%pd' to '%pd'\n", dir, ceph_vinop(dir),
1203 old_dentry, dentry);
1204 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LINK, USE_AUTH_MDS);
1205 if (IS_ERR(req)) {
1206 d_drop(dentry);
1207 return PTR_ERR(req);
1208 }
1209 req->r_dentry = dget(dentry);
1210 req->r_num_caps = 2;
1211 req->r_old_dentry = dget(old_dentry);
1212 /*
1213 * The old_dentry maybe a DCACHE_DISCONNECTED dentry, then we
1214 * will just pass the ino# to MDSs.
1215 */
1216 if (old_dentry->d_flags & DCACHE_DISCONNECTED)
1217 req->r_ino2 = ceph_vino(d_inode(old_dentry));
1218 req->r_parent = dir;
1219 ihold(dir);
1220 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
1221 req->r_dentry_drop = CEPH_CAP_FILE_SHARED | CEPH_CAP_XATTR_EXCL;
1222 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
1223 /* release LINK_SHARED on source inode (mds will lock it) */
1224 req->r_old_inode_drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL;
1225 err = ceph_mdsc_do_request(mdsc, dir, req);
1226 if (err) {
1227 d_drop(dentry);
1228 } else if (!req->r_reply_info.head->is_dentry) {
1229 ihold(d_inode(old_dentry));
1230 d_instantiate(dentry, d_inode(old_dentry));
1231 }
1232 ceph_mdsc_put_request(req);
1233 return err;
1234}
1235
1236static void ceph_async_unlink_cb(struct ceph_mds_client *mdsc,
1237 struct ceph_mds_request *req)
1238{
1239 struct dentry *dentry = req->r_dentry;
1240 struct ceph_fs_client *fsc = ceph_sb_to_fs_client(dentry->d_sb);
1241 struct ceph_client *cl = fsc->client;
1242 struct ceph_dentry_info *di = ceph_dentry(dentry);
1243 int result = req->r_err ? req->r_err :
1244 le32_to_cpu(req->r_reply_info.head->result);
1245
1246 if (!test_bit(CEPH_DENTRY_ASYNC_UNLINK_BIT, &di->flags))
1247 pr_warn_client(cl,
1248 "dentry %p:%pd async unlink bit is not set\n",
1249 dentry, dentry);
1250
1251 spin_lock(&fsc->async_unlink_conflict_lock);
1252 hash_del_rcu(&di->hnode);
1253 spin_unlock(&fsc->async_unlink_conflict_lock);
1254
1255 spin_lock(&dentry->d_lock);
1256 di->flags &= ~CEPH_DENTRY_ASYNC_UNLINK;
1257 wake_up_bit(&di->flags, CEPH_DENTRY_ASYNC_UNLINK_BIT);
1258 spin_unlock(&dentry->d_lock);
1259
1260 synchronize_rcu();
1261
1262 if (result == -EJUKEBOX)
1263 goto out;
1264
1265 /* If op failed, mark everyone involved for errors */
1266 if (result) {
1267 int pathlen = 0;
1268 u64 base = 0;
1269 char *path = ceph_mdsc_build_path(mdsc, dentry, &pathlen,
1270 &base, 0);
1271
1272 /* mark error on parent + clear complete */
1273 mapping_set_error(req->r_parent->i_mapping, result);
1274 ceph_dir_clear_complete(req->r_parent);
1275
1276 /* drop the dentry -- we don't know its status */
1277 if (!d_unhashed(dentry))
1278 d_drop(dentry);
1279
1280 /* mark inode itself for an error (since metadata is bogus) */
1281 mapping_set_error(req->r_old_inode->i_mapping, result);
1282
1283 pr_warn_client(cl, "failure path=(%llx)%s result=%d!\n",
1284 base, IS_ERR(path) ? "<<bad>>" : path, result);
1285 ceph_mdsc_free_path(path, pathlen);
1286 }
1287out:
1288 iput(req->r_old_inode);
1289 ceph_mdsc_release_dir_caps(req);
1290}
1291
1292static int get_caps_for_async_unlink(struct inode *dir, struct dentry *dentry)
1293{
1294 struct ceph_inode_info *ci = ceph_inode(dir);
1295 struct ceph_dentry_info *di;
1296 int got = 0, want = CEPH_CAP_FILE_EXCL | CEPH_CAP_DIR_UNLINK;
1297
1298 spin_lock(&ci->i_ceph_lock);
1299 if ((__ceph_caps_issued(ci, NULL) & want) == want) {
1300 ceph_take_cap_refs(ci, want, false);
1301 got = want;
1302 }
1303 spin_unlock(&ci->i_ceph_lock);
1304
1305 /* If we didn't get anything, return 0 */
1306 if (!got)
1307 return 0;
1308
1309 spin_lock(&dentry->d_lock);
1310 di = ceph_dentry(dentry);
1311 /*
1312 * - We are holding Fx, which implies Fs caps.
1313 * - Only support async unlink for primary linkage
1314 */
1315 if (atomic_read(&ci->i_shared_gen) != di->lease_shared_gen ||
1316 !(di->flags & CEPH_DENTRY_PRIMARY_LINK))
1317 want = 0;
1318 spin_unlock(&dentry->d_lock);
1319
1320 /* Do we still want what we've got? */
1321 if (want == got)
1322 return got;
1323
1324 ceph_put_cap_refs(ci, got);
1325 return 0;
1326}
1327
1328/*
1329 * rmdir and unlink are differ only by the metadata op code
1330 */
1331static int ceph_unlink(struct inode *dir, struct dentry *dentry)
1332{
1333 struct ceph_fs_client *fsc = ceph_sb_to_fs_client(dir->i_sb);
1334 struct ceph_client *cl = fsc->client;
1335 struct ceph_mds_client *mdsc = fsc->mdsc;
1336 struct inode *inode = d_inode(dentry);
1337 struct ceph_mds_request *req;
1338 bool try_async = ceph_test_mount_opt(fsc, ASYNC_DIROPS);
1339 int err = -EROFS;
1340 int op;
1341
1342 if (ceph_snap(dir) == CEPH_SNAPDIR) {
1343 /* rmdir .snap/foo is RMSNAP */
1344 doutc(cl, "rmsnap %llx.%llx/'%pd' dn\n", ceph_vinop(dir),
1345 dentry);
1346 op = CEPH_MDS_OP_RMSNAP;
1347 } else if (ceph_snap(dir) == CEPH_NOSNAP) {
1348 doutc(cl, "unlink/rmdir %llx.%llx/'%pd' inode %llx.%llx\n",
1349 ceph_vinop(dir), dentry, ceph_vinop(inode));
1350 op = d_is_dir(dentry) ?
1351 CEPH_MDS_OP_RMDIR : CEPH_MDS_OP_UNLINK;
1352 } else
1353 goto out;
1354retry:
1355 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
1356 if (IS_ERR(req)) {
1357 err = PTR_ERR(req);
1358 goto out;
1359 }
1360 req->r_dentry = dget(dentry);
1361 req->r_num_caps = 2;
1362 req->r_parent = dir;
1363 ihold(dir);
1364 req->r_dentry_drop = CEPH_CAP_FILE_SHARED | CEPH_CAP_XATTR_EXCL;
1365 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
1366 req->r_inode_drop = ceph_drop_caps_for_unlink(inode);
1367
1368 if (try_async && op == CEPH_MDS_OP_UNLINK &&
1369 (req->r_dir_caps = get_caps_for_async_unlink(dir, dentry))) {
1370 struct ceph_dentry_info *di = ceph_dentry(dentry);
1371
1372 doutc(cl, "async unlink on %llx.%llx/'%pd' caps=%s",
1373 ceph_vinop(dir), dentry,
1374 ceph_cap_string(req->r_dir_caps));
1375 set_bit(CEPH_MDS_R_ASYNC, &req->r_req_flags);
1376 req->r_callback = ceph_async_unlink_cb;
1377 req->r_old_inode = d_inode(dentry);
1378 ihold(req->r_old_inode);
1379
1380 spin_lock(&dentry->d_lock);
1381 di->flags |= CEPH_DENTRY_ASYNC_UNLINK;
1382 spin_unlock(&dentry->d_lock);
1383
1384 spin_lock(&fsc->async_unlink_conflict_lock);
1385 hash_add_rcu(fsc->async_unlink_conflict, &di->hnode,
1386 dentry->d_name.hash);
1387 spin_unlock(&fsc->async_unlink_conflict_lock);
1388
1389 err = ceph_mdsc_submit_request(mdsc, dir, req);
1390 if (!err) {
1391 /*
1392 * We have enough caps, so we assume that the unlink
1393 * will succeed. Fix up the target inode and dcache.
1394 */
1395 drop_nlink(inode);
1396 d_delete(dentry);
1397 } else {
1398 spin_lock(&fsc->async_unlink_conflict_lock);
1399 hash_del_rcu(&di->hnode);
1400 spin_unlock(&fsc->async_unlink_conflict_lock);
1401
1402 spin_lock(&dentry->d_lock);
1403 di->flags &= ~CEPH_DENTRY_ASYNC_UNLINK;
1404 spin_unlock(&dentry->d_lock);
1405
1406 if (err == -EJUKEBOX) {
1407 try_async = false;
1408 ceph_mdsc_put_request(req);
1409 goto retry;
1410 }
1411 }
1412 } else {
1413 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
1414 err = ceph_mdsc_do_request(mdsc, dir, req);
1415 if (!err && !req->r_reply_info.head->is_dentry)
1416 d_delete(dentry);
1417 }
1418
1419 ceph_mdsc_put_request(req);
1420out:
1421 return err;
1422}
1423
1424static int ceph_rename(struct mnt_idmap *idmap, struct inode *old_dir,
1425 struct dentry *old_dentry, struct inode *new_dir,
1426 struct dentry *new_dentry, unsigned int flags)
1427{
1428 struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(old_dir->i_sb);
1429 struct ceph_client *cl = mdsc->fsc->client;
1430 struct ceph_mds_request *req;
1431 int op = CEPH_MDS_OP_RENAME;
1432 int err;
1433
1434 if (flags)
1435 return -EINVAL;
1436
1437 if (ceph_snap(old_dir) != ceph_snap(new_dir))
1438 return -EXDEV;
1439 if (ceph_snap(old_dir) != CEPH_NOSNAP) {
1440 if (old_dir == new_dir && ceph_snap(old_dir) == CEPH_SNAPDIR)
1441 op = CEPH_MDS_OP_RENAMESNAP;
1442 else
1443 return -EROFS;
1444 }
1445 /* don't allow cross-quota renames */
1446 if ((old_dir != new_dir) &&
1447 (!ceph_quota_is_same_realm(old_dir, new_dir)))
1448 return -EXDEV;
1449
1450 err = ceph_wait_on_conflict_unlink(new_dentry);
1451 if (err)
1452 return err;
1453
1454 err = fscrypt_prepare_rename(old_dir, old_dentry, new_dir, new_dentry,
1455 flags);
1456 if (err)
1457 return err;
1458
1459 doutc(cl, "%llx.%llx/'%pd' to %llx.%llx/'%pd'\n",
1460 ceph_vinop(old_dir), old_dentry, ceph_vinop(new_dir),
1461 new_dentry);
1462 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
1463 if (IS_ERR(req))
1464 return PTR_ERR(req);
1465 ihold(old_dir);
1466 req->r_dentry = dget(new_dentry);
1467 req->r_num_caps = 2;
1468 req->r_old_dentry = dget(old_dentry);
1469 req->r_old_dentry_dir = old_dir;
1470 req->r_parent = new_dir;
1471 ihold(new_dir);
1472 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
1473 req->r_old_dentry_drop = CEPH_CAP_FILE_SHARED | CEPH_CAP_XATTR_EXCL;
1474 req->r_old_dentry_unless = CEPH_CAP_FILE_EXCL;
1475 req->r_dentry_drop = CEPH_CAP_FILE_SHARED | CEPH_CAP_XATTR_EXCL;
1476 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
1477 /* release LINK_RDCACHE on source inode (mds will lock it) */
1478 req->r_old_inode_drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL;
1479 if (d_really_is_positive(new_dentry)) {
1480 req->r_inode_drop =
1481 ceph_drop_caps_for_unlink(d_inode(new_dentry));
1482 }
1483 err = ceph_mdsc_do_request(mdsc, old_dir, req);
1484 if (!err && !req->r_reply_info.head->is_dentry) {
1485 /*
1486 * Normally d_move() is done by fill_trace (called by
1487 * do_request, above). If there is no trace, we need
1488 * to do it here.
1489 */
1490 d_move(old_dentry, new_dentry);
1491 }
1492 ceph_mdsc_put_request(req);
1493 return err;
1494}
1495
1496/*
1497 * Move dentry to tail of mdsc->dentry_leases list when lease is updated.
1498 * Leases at front of the list will expire first. (Assume all leases have
1499 * similar duration)
1500 *
1501 * Called under dentry->d_lock.
1502 */
1503void __ceph_dentry_lease_touch(struct ceph_dentry_info *di)
1504{
1505 struct dentry *dn = di->dentry;
1506 struct ceph_mds_client *mdsc = ceph_sb_to_fs_client(dn->d_sb)->mdsc;
1507 struct ceph_client *cl = mdsc->fsc->client;
1508
1509 doutc(cl, "%p %p '%pd'\n", di, dn, dn);
1510
1511 di->flags |= CEPH_DENTRY_LEASE_LIST;
1512 if (di->flags & CEPH_DENTRY_SHRINK_LIST) {
1513 di->flags |= CEPH_DENTRY_REFERENCED;
1514 return;
1515 }
1516
1517 spin_lock(&mdsc->dentry_list_lock);
1518 list_move_tail(&di->lease_list, &mdsc->dentry_leases);
1519 spin_unlock(&mdsc->dentry_list_lock);
1520}
1521
1522static void __dentry_dir_lease_touch(struct ceph_mds_client* mdsc,
1523 struct ceph_dentry_info *di)
1524{
1525 di->flags &= ~(CEPH_DENTRY_LEASE_LIST | CEPH_DENTRY_REFERENCED);
1526 di->lease_gen = 0;
1527 di->time = jiffies;
1528 list_move_tail(&di->lease_list, &mdsc->dentry_dir_leases);
1529}
1530
1531/*
1532 * When dir lease is used, add dentry to tail of mdsc->dentry_dir_leases
1533 * list if it's not in the list, otherwise set 'referenced' flag.
1534 *
1535 * Called under dentry->d_lock.
1536 */
1537void __ceph_dentry_dir_lease_touch(struct ceph_dentry_info *di)
1538{
1539 struct dentry *dn = di->dentry;
1540 struct ceph_mds_client *mdsc = ceph_sb_to_fs_client(dn->d_sb)->mdsc;
1541 struct ceph_client *cl = mdsc->fsc->client;
1542
1543 doutc(cl, "%p %p '%pd' (offset 0x%llx)\n", di, dn, dn, di->offset);
1544
1545 if (!list_empty(&di->lease_list)) {
1546 if (di->flags & CEPH_DENTRY_LEASE_LIST) {
1547 /* don't remove dentry from dentry lease list
1548 * if its lease is valid */
1549 if (__dentry_lease_is_valid(di))
1550 return;
1551 } else {
1552 di->flags |= CEPH_DENTRY_REFERENCED;
1553 return;
1554 }
1555 }
1556
1557 if (di->flags & CEPH_DENTRY_SHRINK_LIST) {
1558 di->flags |= CEPH_DENTRY_REFERENCED;
1559 di->flags &= ~CEPH_DENTRY_LEASE_LIST;
1560 return;
1561 }
1562
1563 spin_lock(&mdsc->dentry_list_lock);
1564 __dentry_dir_lease_touch(mdsc, di),
1565 spin_unlock(&mdsc->dentry_list_lock);
1566}
1567
1568static void __dentry_lease_unlist(struct ceph_dentry_info *di)
1569{
1570 struct ceph_mds_client *mdsc;
1571 if (di->flags & CEPH_DENTRY_SHRINK_LIST)
1572 return;
1573 if (list_empty(&di->lease_list))
1574 return;
1575
1576 mdsc = ceph_sb_to_fs_client(di->dentry->d_sb)->mdsc;
1577 spin_lock(&mdsc->dentry_list_lock);
1578 list_del_init(&di->lease_list);
1579 spin_unlock(&mdsc->dentry_list_lock);
1580}
1581
1582enum {
1583 KEEP = 0,
1584 DELETE = 1,
1585 TOUCH = 2,
1586 STOP = 4,
1587};
1588
1589struct ceph_lease_walk_control {
1590 bool dir_lease;
1591 bool expire_dir_lease;
1592 unsigned long nr_to_scan;
1593 unsigned long dir_lease_ttl;
1594};
1595
1596static int __dir_lease_check(const struct dentry *, struct ceph_lease_walk_control *);
1597static int __dentry_lease_check(const struct dentry *);
1598
1599static unsigned long
1600__dentry_leases_walk(struct ceph_mds_client *mdsc,
1601 struct ceph_lease_walk_control *lwc)
1602{
1603 struct ceph_dentry_info *di, *tmp;
1604 struct dentry *dentry, *last = NULL;
1605 struct list_head* list;
1606 LIST_HEAD(dispose);
1607 unsigned long freed = 0;
1608 int ret = 0;
1609
1610 list = lwc->dir_lease ? &mdsc->dentry_dir_leases : &mdsc->dentry_leases;
1611 spin_lock(&mdsc->dentry_list_lock);
1612 list_for_each_entry_safe(di, tmp, list, lease_list) {
1613 if (!lwc->nr_to_scan)
1614 break;
1615 --lwc->nr_to_scan;
1616
1617 dentry = di->dentry;
1618 if (last == dentry)
1619 break;
1620
1621 if (!spin_trylock(&dentry->d_lock))
1622 continue;
1623
1624 if (__lockref_is_dead(&dentry->d_lockref)) {
1625 list_del_init(&di->lease_list);
1626 goto next;
1627 }
1628
1629 if (lwc->dir_lease)
1630 ret = __dir_lease_check(dentry, lwc);
1631 else
1632 ret = __dentry_lease_check(dentry);
1633 if (ret & TOUCH) {
1634 /* move it into tail of dir lease list */
1635 __dentry_dir_lease_touch(mdsc, di);
1636 if (!last)
1637 last = dentry;
1638 }
1639 if (ret & DELETE) {
1640 /* stale lease */
1641 di->flags &= ~CEPH_DENTRY_REFERENCED;
1642 if (dentry->d_lockref.count > 0) {
1643 /* update_dentry_lease() will re-add
1644 * it to lease list, or
1645 * ceph_d_delete() will return 1 when
1646 * last reference is dropped */
1647 list_del_init(&di->lease_list);
1648 } else {
1649 di->flags |= CEPH_DENTRY_SHRINK_LIST;
1650 list_move_tail(&di->lease_list, &dispose);
1651 dget_dlock(dentry);
1652 }
1653 }
1654next:
1655 spin_unlock(&dentry->d_lock);
1656 if (ret & STOP)
1657 break;
1658 }
1659 spin_unlock(&mdsc->dentry_list_lock);
1660
1661 while (!list_empty(&dispose)) {
1662 di = list_first_entry(&dispose, struct ceph_dentry_info,
1663 lease_list);
1664 dentry = di->dentry;
1665 spin_lock(&dentry->d_lock);
1666
1667 list_del_init(&di->lease_list);
1668 di->flags &= ~CEPH_DENTRY_SHRINK_LIST;
1669 if (di->flags & CEPH_DENTRY_REFERENCED) {
1670 spin_lock(&mdsc->dentry_list_lock);
1671 if (di->flags & CEPH_DENTRY_LEASE_LIST) {
1672 list_add_tail(&di->lease_list,
1673 &mdsc->dentry_leases);
1674 } else {
1675 __dentry_dir_lease_touch(mdsc, di);
1676 }
1677 spin_unlock(&mdsc->dentry_list_lock);
1678 } else {
1679 freed++;
1680 }
1681
1682 spin_unlock(&dentry->d_lock);
1683 /* ceph_d_delete() does the trick */
1684 dput(dentry);
1685 }
1686 return freed;
1687}
1688
1689static int __dentry_lease_check(const struct dentry *dentry)
1690{
1691 struct ceph_dentry_info *di = ceph_dentry(dentry);
1692 int ret;
1693
1694 if (__dentry_lease_is_valid(di))
1695 return STOP;
1696 ret = __dir_lease_try_check(dentry);
1697 if (ret == -EBUSY)
1698 return KEEP;
1699 if (ret > 0)
1700 return TOUCH;
1701 return DELETE;
1702}
1703
1704static int __dir_lease_check(const struct dentry *dentry,
1705 struct ceph_lease_walk_control *lwc)
1706{
1707 struct ceph_dentry_info *di = ceph_dentry(dentry);
1708
1709 int ret = __dir_lease_try_check(dentry);
1710 if (ret == -EBUSY)
1711 return KEEP;
1712 if (ret > 0) {
1713 if (time_before(jiffies, di->time + lwc->dir_lease_ttl))
1714 return STOP;
1715 /* Move dentry to tail of dir lease list if we don't want
1716 * to delete it. So dentries in the list are checked in a
1717 * round robin manner */
1718 if (!lwc->expire_dir_lease)
1719 return TOUCH;
1720 if (dentry->d_lockref.count > 0 ||
1721 (di->flags & CEPH_DENTRY_REFERENCED))
1722 return TOUCH;
1723 /* invalidate dir lease */
1724 di->lease_shared_gen = 0;
1725 }
1726 return DELETE;
1727}
1728
1729int ceph_trim_dentries(struct ceph_mds_client *mdsc)
1730{
1731 struct ceph_lease_walk_control lwc;
1732 unsigned long count;
1733 unsigned long freed;
1734
1735 spin_lock(&mdsc->caps_list_lock);
1736 if (mdsc->caps_use_max > 0 &&
1737 mdsc->caps_use_count > mdsc->caps_use_max)
1738 count = mdsc->caps_use_count - mdsc->caps_use_max;
1739 else
1740 count = 0;
1741 spin_unlock(&mdsc->caps_list_lock);
1742
1743 lwc.dir_lease = false;
1744 lwc.nr_to_scan = CEPH_CAPS_PER_RELEASE * 2;
1745 freed = __dentry_leases_walk(mdsc, &lwc);
1746 if (!lwc.nr_to_scan) /* more invalid leases */
1747 return -EAGAIN;
1748
1749 if (lwc.nr_to_scan < CEPH_CAPS_PER_RELEASE)
1750 lwc.nr_to_scan = CEPH_CAPS_PER_RELEASE;
1751
1752 lwc.dir_lease = true;
1753 lwc.expire_dir_lease = freed < count;
1754 lwc.dir_lease_ttl = mdsc->fsc->mount_options->caps_wanted_delay_max * HZ;
1755 freed +=__dentry_leases_walk(mdsc, &lwc);
1756 if (!lwc.nr_to_scan) /* more to check */
1757 return -EAGAIN;
1758
1759 return freed > 0 ? 1 : 0;
1760}
1761
1762/*
1763 * Ensure a dentry lease will no longer revalidate.
1764 */
1765void ceph_invalidate_dentry_lease(struct dentry *dentry)
1766{
1767 struct ceph_dentry_info *di = ceph_dentry(dentry);
1768 spin_lock(&dentry->d_lock);
1769 di->time = jiffies;
1770 di->lease_shared_gen = 0;
1771 di->flags &= ~CEPH_DENTRY_PRIMARY_LINK;
1772 __dentry_lease_unlist(di);
1773 spin_unlock(&dentry->d_lock);
1774}
1775
1776/*
1777 * Check if dentry lease is valid. If not, delete the lease. Try to
1778 * renew if the least is more than half up.
1779 */
1780static bool __dentry_lease_is_valid(struct ceph_dentry_info *di)
1781{
1782 struct ceph_mds_session *session;
1783
1784 if (!di->lease_gen)
1785 return false;
1786
1787 session = di->lease_session;
1788 if (session) {
1789 u32 gen;
1790 unsigned long ttl;
1791
1792 gen = atomic_read(&session->s_cap_gen);
1793 ttl = session->s_cap_ttl;
1794
1795 if (di->lease_gen == gen &&
1796 time_before(jiffies, ttl) &&
1797 time_before(jiffies, di->time))
1798 return true;
1799 }
1800 di->lease_gen = 0;
1801 return false;
1802}
1803
1804static int dentry_lease_is_valid(struct dentry *dentry, unsigned int flags)
1805{
1806 struct ceph_dentry_info *di;
1807 struct ceph_mds_session *session = NULL;
1808 struct ceph_mds_client *mdsc = ceph_sb_to_fs_client(dentry->d_sb)->mdsc;
1809 struct ceph_client *cl = mdsc->fsc->client;
1810 u32 seq = 0;
1811 int valid = 0;
1812
1813 spin_lock(&dentry->d_lock);
1814 di = ceph_dentry(dentry);
1815 if (di && __dentry_lease_is_valid(di)) {
1816 valid = 1;
1817
1818 if (di->lease_renew_after &&
1819 time_after(jiffies, di->lease_renew_after)) {
1820 /*
1821 * We should renew. If we're in RCU walk mode
1822 * though, we can't do that so just return
1823 * -ECHILD.
1824 */
1825 if (flags & LOOKUP_RCU) {
1826 valid = -ECHILD;
1827 } else {
1828 session = ceph_get_mds_session(di->lease_session);
1829 seq = di->lease_seq;
1830 di->lease_renew_after = 0;
1831 di->lease_renew_from = jiffies;
1832 }
1833 }
1834 }
1835 spin_unlock(&dentry->d_lock);
1836
1837 if (session) {
1838 ceph_mdsc_lease_send_msg(session, dentry,
1839 CEPH_MDS_LEASE_RENEW, seq);
1840 ceph_put_mds_session(session);
1841 }
1842 doutc(cl, "dentry %p = %d\n", dentry, valid);
1843 return valid;
1844}
1845
1846/*
1847 * Called under dentry->d_lock.
1848 */
1849static int __dir_lease_try_check(const struct dentry *dentry)
1850{
1851 struct ceph_dentry_info *di = ceph_dentry(dentry);
1852 struct inode *dir;
1853 struct ceph_inode_info *ci;
1854 int valid = 0;
1855
1856 if (!di->lease_shared_gen)
1857 return 0;
1858 if (IS_ROOT(dentry))
1859 return 0;
1860
1861 dir = d_inode(dentry->d_parent);
1862 ci = ceph_inode(dir);
1863
1864 if (spin_trylock(&ci->i_ceph_lock)) {
1865 if (atomic_read(&ci->i_shared_gen) == di->lease_shared_gen &&
1866 __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 0))
1867 valid = 1;
1868 spin_unlock(&ci->i_ceph_lock);
1869 } else {
1870 valid = -EBUSY;
1871 }
1872
1873 if (!valid)
1874 di->lease_shared_gen = 0;
1875 return valid;
1876}
1877
1878/*
1879 * Check if directory-wide content lease/cap is valid.
1880 */
1881static int dir_lease_is_valid(struct inode *dir, struct dentry *dentry,
1882 struct ceph_mds_client *mdsc)
1883{
1884 struct ceph_inode_info *ci = ceph_inode(dir);
1885 struct ceph_client *cl = mdsc->fsc->client;
1886 int valid;
1887 int shared_gen;
1888
1889 spin_lock(&ci->i_ceph_lock);
1890 valid = __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1);
1891 if (valid) {
1892 __ceph_touch_fmode(ci, mdsc, CEPH_FILE_MODE_RD);
1893 shared_gen = atomic_read(&ci->i_shared_gen);
1894 }
1895 spin_unlock(&ci->i_ceph_lock);
1896 if (valid) {
1897 struct ceph_dentry_info *di;
1898 spin_lock(&dentry->d_lock);
1899 di = ceph_dentry(dentry);
1900 if (dir == d_inode(dentry->d_parent) &&
1901 di && di->lease_shared_gen == shared_gen)
1902 __ceph_dentry_dir_lease_touch(di);
1903 else
1904 valid = 0;
1905 spin_unlock(&dentry->d_lock);
1906 }
1907 doutc(cl, "dir %p %llx.%llx v%u dentry %p '%pd' = %d\n", dir,
1908 ceph_vinop(dir), (unsigned)atomic_read(&ci->i_shared_gen),
1909 dentry, dentry, valid);
1910 return valid;
1911}
1912
1913/*
1914 * Check if cached dentry can be trusted.
1915 */
1916static int ceph_d_revalidate(struct dentry *dentry, unsigned int flags)
1917{
1918 struct ceph_mds_client *mdsc = ceph_sb_to_fs_client(dentry->d_sb)->mdsc;
1919 struct ceph_client *cl = mdsc->fsc->client;
1920 int valid = 0;
1921 struct dentry *parent;
1922 struct inode *dir, *inode;
1923
1924 valid = fscrypt_d_revalidate(dentry, flags);
1925 if (valid <= 0)
1926 return valid;
1927
1928 if (flags & LOOKUP_RCU) {
1929 parent = READ_ONCE(dentry->d_parent);
1930 dir = d_inode_rcu(parent);
1931 if (!dir)
1932 return -ECHILD;
1933 inode = d_inode_rcu(dentry);
1934 } else {
1935 parent = dget_parent(dentry);
1936 dir = d_inode(parent);
1937 inode = d_inode(dentry);
1938 }
1939
1940 doutc(cl, "%p '%pd' inode %p offset 0x%llx nokey %d\n",
1941 dentry, dentry, inode, ceph_dentry(dentry)->offset,
1942 !!(dentry->d_flags & DCACHE_NOKEY_NAME));
1943
1944 mdsc = ceph_sb_to_fs_client(dir->i_sb)->mdsc;
1945
1946 /* always trust cached snapped dentries, snapdir dentry */
1947 if (ceph_snap(dir) != CEPH_NOSNAP) {
1948 doutc(cl, "%p '%pd' inode %p is SNAPPED\n", dentry,
1949 dentry, inode);
1950 valid = 1;
1951 } else if (inode && ceph_snap(inode) == CEPH_SNAPDIR) {
1952 valid = 1;
1953 } else {
1954 valid = dentry_lease_is_valid(dentry, flags);
1955 if (valid == -ECHILD)
1956 return valid;
1957 if (valid || dir_lease_is_valid(dir, dentry, mdsc)) {
1958 if (inode)
1959 valid = ceph_is_any_caps(inode);
1960 else
1961 valid = 1;
1962 }
1963 }
1964
1965 if (!valid) {
1966 struct ceph_mds_request *req;
1967 int op, err;
1968 u32 mask;
1969
1970 if (flags & LOOKUP_RCU)
1971 return -ECHILD;
1972
1973 percpu_counter_inc(&mdsc->metric.d_lease_mis);
1974
1975 op = ceph_snap(dir) == CEPH_SNAPDIR ?
1976 CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP;
1977 req = ceph_mdsc_create_request(mdsc, op, USE_ANY_MDS);
1978 if (!IS_ERR(req)) {
1979 req->r_dentry = dget(dentry);
1980 req->r_num_caps = 2;
1981 req->r_parent = dir;
1982 ihold(dir);
1983
1984 mask = CEPH_STAT_CAP_INODE | CEPH_CAP_AUTH_SHARED;
1985 if (ceph_security_xattr_wanted(dir))
1986 mask |= CEPH_CAP_XATTR_SHARED;
1987 req->r_args.getattr.mask = cpu_to_le32(mask);
1988
1989 err = ceph_mdsc_do_request(mdsc, NULL, req);
1990 switch (err) {
1991 case 0:
1992 if (d_really_is_positive(dentry) &&
1993 d_inode(dentry) == req->r_target_inode)
1994 valid = 1;
1995 break;
1996 case -ENOENT:
1997 if (d_really_is_negative(dentry))
1998 valid = 1;
1999 fallthrough;
2000 default:
2001 break;
2002 }
2003 ceph_mdsc_put_request(req);
2004 doutc(cl, "%p '%pd', lookup result=%d\n", dentry,
2005 dentry, err);
2006 }
2007 } else {
2008 percpu_counter_inc(&mdsc->metric.d_lease_hit);
2009 }
2010
2011 doutc(cl, "%p '%pd' %s\n", dentry, dentry, valid ? "valid" : "invalid");
2012 if (!valid)
2013 ceph_dir_clear_complete(dir);
2014
2015 if (!(flags & LOOKUP_RCU))
2016 dput(parent);
2017 return valid;
2018}
2019
2020/*
2021 * Delete unused dentry that doesn't have valid lease
2022 *
2023 * Called under dentry->d_lock.
2024 */
2025static int ceph_d_delete(const struct dentry *dentry)
2026{
2027 struct ceph_dentry_info *di;
2028
2029 /* won't release caps */
2030 if (d_really_is_negative(dentry))
2031 return 0;
2032 if (ceph_snap(d_inode(dentry)) != CEPH_NOSNAP)
2033 return 0;
2034 /* vaild lease? */
2035 di = ceph_dentry(dentry);
2036 if (di) {
2037 if (__dentry_lease_is_valid(di))
2038 return 0;
2039 if (__dir_lease_try_check(dentry))
2040 return 0;
2041 }
2042 return 1;
2043}
2044
2045/*
2046 * Release our ceph_dentry_info.
2047 */
2048static void ceph_d_release(struct dentry *dentry)
2049{
2050 struct ceph_dentry_info *di = ceph_dentry(dentry);
2051 struct ceph_fs_client *fsc = ceph_sb_to_fs_client(dentry->d_sb);
2052
2053 doutc(fsc->client, "dentry %p '%pd'\n", dentry, dentry);
2054
2055 atomic64_dec(&fsc->mdsc->metric.total_dentries);
2056
2057 spin_lock(&dentry->d_lock);
2058 __dentry_lease_unlist(di);
2059 dentry->d_fsdata = NULL;
2060 spin_unlock(&dentry->d_lock);
2061
2062 ceph_put_mds_session(di->lease_session);
2063 kmem_cache_free(ceph_dentry_cachep, di);
2064}
2065
2066/*
2067 * When the VFS prunes a dentry from the cache, we need to clear the
2068 * complete flag on the parent directory.
2069 *
2070 * Called under dentry->d_lock.
2071 */
2072static void ceph_d_prune(struct dentry *dentry)
2073{
2074 struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(dentry->d_sb);
2075 struct ceph_client *cl = mdsc->fsc->client;
2076 struct ceph_inode_info *dir_ci;
2077 struct ceph_dentry_info *di;
2078
2079 doutc(cl, "dentry %p '%pd'\n", dentry, dentry);
2080
2081 /* do we have a valid parent? */
2082 if (IS_ROOT(dentry))
2083 return;
2084
2085 /* we hold d_lock, so d_parent is stable */
2086 dir_ci = ceph_inode(d_inode(dentry->d_parent));
2087 if (dir_ci->i_vino.snap == CEPH_SNAPDIR)
2088 return;
2089
2090 /* who calls d_delete() should also disable dcache readdir */
2091 if (d_really_is_negative(dentry))
2092 return;
2093
2094 /* d_fsdata does not get cleared until d_release */
2095 if (!d_unhashed(dentry)) {
2096 __ceph_dir_clear_complete(dir_ci);
2097 return;
2098 }
2099
2100 /* Disable dcache readdir just in case that someone called d_drop()
2101 * or d_invalidate(), but MDS didn't revoke CEPH_CAP_FILE_SHARED
2102 * properly (dcache readdir is still enabled) */
2103 di = ceph_dentry(dentry);
2104 if (di->offset > 0 &&
2105 di->lease_shared_gen == atomic_read(&dir_ci->i_shared_gen))
2106 __ceph_dir_clear_ordered(dir_ci);
2107}
2108
2109/*
2110 * read() on a dir. This weird interface hack only works if mounted
2111 * with '-o dirstat'.
2112 */
2113static ssize_t ceph_read_dir(struct file *file, char __user *buf, size_t size,
2114 loff_t *ppos)
2115{
2116 struct ceph_dir_file_info *dfi = file->private_data;
2117 struct inode *inode = file_inode(file);
2118 struct ceph_inode_info *ci = ceph_inode(inode);
2119 int left;
2120 const int bufsize = 1024;
2121
2122 if (!ceph_test_mount_opt(ceph_sb_to_fs_client(inode->i_sb), DIRSTAT))
2123 return -EISDIR;
2124
2125 if (!dfi->dir_info) {
2126 dfi->dir_info = kmalloc(bufsize, GFP_KERNEL);
2127 if (!dfi->dir_info)
2128 return -ENOMEM;
2129 dfi->dir_info_len =
2130 snprintf(dfi->dir_info, bufsize,
2131 "entries: %20lld\n"
2132 " files: %20lld\n"
2133 " subdirs: %20lld\n"
2134 "rentries: %20lld\n"
2135 " rfiles: %20lld\n"
2136 " rsubdirs: %20lld\n"
2137 "rbytes: %20lld\n"
2138 "rctime: %10lld.%09ld\n",
2139 ci->i_files + ci->i_subdirs,
2140 ci->i_files,
2141 ci->i_subdirs,
2142 ci->i_rfiles + ci->i_rsubdirs,
2143 ci->i_rfiles,
2144 ci->i_rsubdirs,
2145 ci->i_rbytes,
2146 ci->i_rctime.tv_sec,
2147 ci->i_rctime.tv_nsec);
2148 }
2149
2150 if (*ppos >= dfi->dir_info_len)
2151 return 0;
2152 size = min_t(unsigned, size, dfi->dir_info_len-*ppos);
2153 left = copy_to_user(buf, dfi->dir_info + *ppos, size);
2154 if (left == size)
2155 return -EFAULT;
2156 *ppos += (size - left);
2157 return size - left;
2158}
2159
2160
2161
2162/*
2163 * Return name hash for a given dentry. This is dependent on
2164 * the parent directory's hash function.
2165 */
2166unsigned ceph_dentry_hash(struct inode *dir, struct dentry *dn)
2167{
2168 struct ceph_inode_info *dci = ceph_inode(dir);
2169 unsigned hash;
2170
2171 switch (dci->i_dir_layout.dl_dir_hash) {
2172 case 0: /* for backward compat */
2173 case CEPH_STR_HASH_LINUX:
2174 return dn->d_name.hash;
2175
2176 default:
2177 spin_lock(&dn->d_lock);
2178 hash = ceph_str_hash(dci->i_dir_layout.dl_dir_hash,
2179 dn->d_name.name, dn->d_name.len);
2180 spin_unlock(&dn->d_lock);
2181 return hash;
2182 }
2183}
2184
2185WRAP_DIR_ITER(ceph_readdir) // FIXME!
2186const struct file_operations ceph_dir_fops = {
2187 .read = ceph_read_dir,
2188 .iterate_shared = shared_ceph_readdir,
2189 .llseek = ceph_dir_llseek,
2190 .open = ceph_open,
2191 .release = ceph_release,
2192 .unlocked_ioctl = ceph_ioctl,
2193 .compat_ioctl = compat_ptr_ioctl,
2194 .fsync = ceph_fsync,
2195 .lock = ceph_lock,
2196 .flock = ceph_flock,
2197};
2198
2199const struct file_operations ceph_snapdir_fops = {
2200 .iterate_shared = shared_ceph_readdir,
2201 .llseek = ceph_dir_llseek,
2202 .open = ceph_open,
2203 .release = ceph_release,
2204};
2205
2206const struct inode_operations ceph_dir_iops = {
2207 .lookup = ceph_lookup,
2208 .permission = ceph_permission,
2209 .getattr = ceph_getattr,
2210 .setattr = ceph_setattr,
2211 .listxattr = ceph_listxattr,
2212 .get_inode_acl = ceph_get_acl,
2213 .set_acl = ceph_set_acl,
2214 .mknod = ceph_mknod,
2215 .symlink = ceph_symlink,
2216 .mkdir = ceph_mkdir,
2217 .link = ceph_link,
2218 .unlink = ceph_unlink,
2219 .rmdir = ceph_unlink,
2220 .rename = ceph_rename,
2221 .create = ceph_create,
2222 .atomic_open = ceph_atomic_open,
2223};
2224
2225const struct inode_operations ceph_snapdir_iops = {
2226 .lookup = ceph_lookup,
2227 .permission = ceph_permission,
2228 .getattr = ceph_getattr,
2229 .mkdir = ceph_mkdir,
2230 .rmdir = ceph_unlink,
2231 .rename = ceph_rename,
2232};
2233
2234const struct dentry_operations ceph_dentry_ops = {
2235 .d_revalidate = ceph_d_revalidate,
2236 .d_delete = ceph_d_delete,
2237 .d_release = ceph_d_release,
2238 .d_prune = ceph_d_prune,
2239 .d_init = ceph_d_init,
2240};