?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
ext4: Optimize ext4 DIO overwrites
[linux/fpc-iii.git] / fs / ceph / inode.c
blobc07407586ce873f236a1cf61d0ea48efa27590c4
1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/ceph/ceph_debug.h>
3
4 #include <linux/module.h>
5 #include <linux/fs.h>
6 #include <linux/slab.h>
7 #include <linux/string.h>
8 #include <linux/uaccess.h>
9 #include <linux/kernel.h>
10 #include <linux/writeback.h>
11 #include <linux/vmalloc.h>
12 #include <linux/xattr.h>
13 #include <linux/posix_acl.h>
14 #include <linux/random.h>
15 #include <linux/sort.h>
16 #include <linux/iversion.h>
17
18 #include "super.h"
19 #include "mds_client.h"
20 #include "cache.h"
21 #include <linux/ceph/decode.h>
22
23 /*
24 * Ceph inode operations
25 *
26 * Implement basic inode helpers (get, alloc) and inode ops (getattr,
27 * setattr, etc.), xattr helpers, and helpers for assimilating
28 * metadata returned by the MDS into our cache.
29 *
30 * Also define helpers for doing asynchronous writeback, invalidation,
31 * and truncation for the benefit of those who can't afford to block
32 * (typically because they are in the message handler path).
33 */
34
35 static const struct inode_operations ceph_symlink_iops;
36
37 static void ceph_inode_work(struct work_struct *work);
38
39 /*
40 * find or create an inode, given the ceph ino number
41 */
42 static int ceph_set_ino_cb(struct inode *inode, void *data)
43 {
44 ceph_inode(inode)->i_vino = *(struct ceph_vino *)data;
45 inode->i_ino = ceph_vino_to_ino(*(struct ceph_vino *)data);
46 inode_set_iversion_raw(inode, 0);
47 return 0;
48 }
49
50 struct inode *ceph_get_inode(struct super_block *sb, struct ceph_vino vino)
51 {
52 struct inode *inode;
53 ino_t t = ceph_vino_to_ino(vino);
54
55 inode = iget5_locked(sb, t, ceph_ino_compare, ceph_set_ino_cb, &vino);
56 if (!inode)
57 return ERR_PTR(-ENOMEM);
58 if (inode->i_state & I_NEW) {
59 dout("get_inode created new inode %p %llx.%llx ino %llx\n",
60 inode, ceph_vinop(inode), (u64)inode->i_ino);
61 unlock_new_inode(inode);
62 }
63
64 dout("get_inode on %lu=%llx.%llx got %p\n", inode->i_ino, vino.ino,
65 vino.snap, inode);
66 return inode;
67 }
68
69 /*
70 * get/constuct snapdir inode for a given directory
71 */
72 struct inode *ceph_get_snapdir(struct inode *parent)
73 {
74 struct ceph_vino vino = {
75 .ino = ceph_ino(parent),
76 .snap = CEPH_SNAPDIR,
77 };
78 struct inode *inode = ceph_get_inode(parent->i_sb, vino);
79 struct ceph_inode_info *ci = ceph_inode(inode);
80
81 BUG_ON(!S_ISDIR(parent->i_mode));
82 if (IS_ERR(inode))
83 return inode;
84 inode->i_mode = parent->i_mode;
85 inode->i_uid = parent->i_uid;
86 inode->i_gid = parent->i_gid;
87 inode->i_op = &ceph_snapdir_iops;
88 inode->i_fop = &ceph_snapdir_fops;
89 ci->i_snap_caps = CEPH_CAP_PIN; /* so we can open */
90 ci->i_rbytes = 0;
91 return inode;
92 }
93
94 const struct inode_operations ceph_file_iops = {
95 .permission = ceph_permission,
96 .setattr = ceph_setattr,
97 .getattr = ceph_getattr,
98 .listxattr = ceph_listxattr,
99 .get_acl = ceph_get_acl,
100 .set_acl = ceph_set_acl,
101 };
102
103
104 /*
105 * We use a 'frag tree' to keep track of the MDS's directory fragments
106 * for a given inode (usually there is just a single fragment). We
107 * need to know when a child frag is delegated to a new MDS, or when
108 * it is flagged as replicated, so we can direct our requests
109 * accordingly.
110 */
111
112 /*
113 * find/create a frag in the tree
114 */
115 static struct ceph_inode_frag *__get_or_create_frag(struct ceph_inode_info *ci,
116 u32 f)
117 {
118 struct rb_node **p;
119 struct rb_node *parent = NULL;
120 struct ceph_inode_frag *frag;
121 int c;
122
123 p = &ci->i_fragtree.rb_node;
124 while (*p) {
125 parent = *p;
126 frag = rb_entry(parent, struct ceph_inode_frag, node);
127 c = ceph_frag_compare(f, frag->frag);
128 if (c < 0)
129 p = &(*p)->rb_left;
130 else if (c > 0)
131 p = &(*p)->rb_right;
132 else
133 return frag;
134 }
135
136 frag = kmalloc(sizeof(*frag), GFP_NOFS);
137 if (!frag)
138 return ERR_PTR(-ENOMEM);
139
140 frag->frag = f;
141 frag->split_by = 0;
142 frag->mds = -1;
143 frag->ndist = 0;
144
145 rb_link_node(&frag->node, parent, p);
146 rb_insert_color(&frag->node, &ci->i_fragtree);
147
148 dout("get_or_create_frag added %llx.%llx frag %x\n",
149 ceph_vinop(&ci->vfs_inode), f);
150 return frag;
151 }
152
153 /*
154 * find a specific frag @f
155 */
156 struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci, u32 f)
157 {
158 struct rb_node *n = ci->i_fragtree.rb_node;
159
160 while (n) {
161 struct ceph_inode_frag *frag =
162 rb_entry(n, struct ceph_inode_frag, node);
163 int c = ceph_frag_compare(f, frag->frag);
164 if (c < 0)
165 n = n->rb_left;
166 else if (c > 0)
167 n = n->rb_right;
168 else
169 return frag;
170 }
171 return NULL;
172 }
173
174 /*
175 * Choose frag containing the given value @v. If @pfrag is
176 * specified, copy the frag delegation info to the caller if
177 * it is present.
178 */
179 static u32 __ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
180 struct ceph_inode_frag *pfrag, int *found)
181 {
182 u32 t = ceph_frag_make(0, 0);
183 struct ceph_inode_frag *frag;
184 unsigned nway, i;
185 u32 n;
186
187 if (found)
188 *found = 0;
189
190 while (1) {
191 WARN_ON(!ceph_frag_contains_value(t, v));
192 frag = __ceph_find_frag(ci, t);
193 if (!frag)
194 break; /* t is a leaf */
195 if (frag->split_by == 0) {
196 if (pfrag)
197 memcpy(pfrag, frag, sizeof(*pfrag));
198 if (found)
199 *found = 1;
200 break;
201 }
202
203 /* choose child */
204 nway = 1 << frag->split_by;
205 dout("choose_frag(%x) %x splits by %d (%d ways)\n", v, t,
206 frag->split_by, nway);
207 for (i = 0; i < nway; i++) {
208 n = ceph_frag_make_child(t, frag->split_by, i);
209 if (ceph_frag_contains_value(n, v)) {
210 t = n;
211 break;
212 }
213 }
214 BUG_ON(i == nway);
215 }
216 dout("choose_frag(%x) = %x\n", v, t);
217
218 return t;
219 }
220
221 u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
222 struct ceph_inode_frag *pfrag, int *found)
223 {
224 u32 ret;
225 mutex_lock(&ci->i_fragtree_mutex);
226 ret = __ceph_choose_frag(ci, v, pfrag, found);
227 mutex_unlock(&ci->i_fragtree_mutex);
228 return ret;
229 }
230
231 /*
232 * Process dirfrag (delegation) info from the mds. Include leaf
233 * fragment in tree ONLY if ndist > 0. Otherwise, only
234 * branches/splits are included in i_fragtree)
235 */
236 static int ceph_fill_dirfrag(struct inode *inode,
237 struct ceph_mds_reply_dirfrag *dirinfo)
238 {
239 struct ceph_inode_info *ci = ceph_inode(inode);
240 struct ceph_inode_frag *frag;
241 u32 id = le32_to_cpu(dirinfo->frag);
242 int mds = le32_to_cpu(dirinfo->auth);
243 int ndist = le32_to_cpu(dirinfo->ndist);
244 int diri_auth = -1;
245 int i;
246 int err = 0;
247
248 spin_lock(&ci->i_ceph_lock);
249 if (ci->i_auth_cap)
250 diri_auth = ci->i_auth_cap->mds;
251 spin_unlock(&ci->i_ceph_lock);
252
253 if (mds == -1) /* CDIR_AUTH_PARENT */
254 mds = diri_auth;
255
256 mutex_lock(&ci->i_fragtree_mutex);
257 if (ndist == 0 && mds == diri_auth) {
258 /* no delegation info needed. */
259 frag = __ceph_find_frag(ci, id);
260 if (!frag)
261 goto out;
262 if (frag->split_by == 0) {
263 /* tree leaf, remove */
264 dout("fill_dirfrag removed %llx.%llx frag %x"
265 " (no ref)\n", ceph_vinop(inode), id);
266 rb_erase(&frag->node, &ci->i_fragtree);
267 kfree(frag);
268 } else {
269 /* tree branch, keep and clear */
270 dout("fill_dirfrag cleared %llx.%llx frag %x"
271 " referral\n", ceph_vinop(inode), id);
272 frag->mds = -1;
273 frag->ndist = 0;
274 }
275 goto out;
276 }
277
278
279 /* find/add this frag to store mds delegation info */
280 frag = __get_or_create_frag(ci, id);
281 if (IS_ERR(frag)) {
282 /* this is not the end of the world; we can continue
283 with bad/inaccurate delegation info */
284 pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n",
285 ceph_vinop(inode), le32_to_cpu(dirinfo->frag));
286 err = -ENOMEM;
287 goto out;
288 }
289
290 frag->mds = mds;
291 frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP);
292 for (i = 0; i < frag->ndist; i++)
293 frag->dist[i] = le32_to_cpu(dirinfo->dist[i]);
294 dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n",
295 ceph_vinop(inode), frag->frag, frag->ndist);
296
297 out:
298 mutex_unlock(&ci->i_fragtree_mutex);
299 return err;
300 }
301
302 static int frag_tree_split_cmp(const void *l, const void *r)
303 {
304 struct ceph_frag_tree_split *ls = (struct ceph_frag_tree_split*)l;
305 struct ceph_frag_tree_split *rs = (struct ceph_frag_tree_split*)r;
306 return ceph_frag_compare(le32_to_cpu(ls->frag),
307 le32_to_cpu(rs->frag));
308 }
309
310 static bool is_frag_child(u32 f, struct ceph_inode_frag *frag)
311 {
312 if (!frag)
313 return f == ceph_frag_make(0, 0);
314 if (ceph_frag_bits(f) != ceph_frag_bits(frag->frag) + frag->split_by)
315 return false;
316 return ceph_frag_contains_value(frag->frag, ceph_frag_value(f));
317 }
318
319 static int ceph_fill_fragtree(struct inode *inode,
320 struct ceph_frag_tree_head *fragtree,
321 struct ceph_mds_reply_dirfrag *dirinfo)
322 {
323 struct ceph_inode_info *ci = ceph_inode(inode);
324 struct ceph_inode_frag *frag, *prev_frag = NULL;
325 struct rb_node *rb_node;
326 unsigned i, split_by, nsplits;
327 u32 id;
328 bool update = false;
329
330 mutex_lock(&ci->i_fragtree_mutex);
331 nsplits = le32_to_cpu(fragtree->nsplits);
332 if (nsplits != ci->i_fragtree_nsplits) {
333 update = true;
334 } else if (nsplits) {
335 i = prandom_u32() % nsplits;
336 id = le32_to_cpu(fragtree->splits[i].frag);
337 if (!__ceph_find_frag(ci, id))
338 update = true;
339 } else if (!RB_EMPTY_ROOT(&ci->i_fragtree)) {
340 rb_node = rb_first(&ci->i_fragtree);
341 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
342 if (frag->frag != ceph_frag_make(0, 0) || rb_next(rb_node))
343 update = true;
344 }
345 if (!update && dirinfo) {
346 id = le32_to_cpu(dirinfo->frag);
347 if (id != __ceph_choose_frag(ci, id, NULL, NULL))
348 update = true;
349 }
350 if (!update)
351 goto out_unlock;
352
353 if (nsplits > 1) {
354 sort(fragtree->splits, nsplits, sizeof(fragtree->splits[0]),
355 frag_tree_split_cmp, NULL);
356 }
357
358 dout("fill_fragtree %llx.%llx\n", ceph_vinop(inode));
359 rb_node = rb_first(&ci->i_fragtree);
360 for (i = 0; i < nsplits; i++) {
361 id = le32_to_cpu(fragtree->splits[i].frag);
362 split_by = le32_to_cpu(fragtree->splits[i].by);
363 if (split_by == 0 || ceph_frag_bits(id) + split_by > 24) {
364 pr_err("fill_fragtree %llx.%llx invalid split %d/%u, "
365 "frag %x split by %d\n", ceph_vinop(inode),
366 i, nsplits, id, split_by);
367 continue;
368 }
369 frag = NULL;
370 while (rb_node) {
371 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
372 if (ceph_frag_compare(frag->frag, id) >= 0) {
373 if (frag->frag != id)
374 frag = NULL;
375 else
376 rb_node = rb_next(rb_node);
377 break;
378 }
379 rb_node = rb_next(rb_node);
380 /* delete stale split/leaf node */
381 if (frag->split_by > 0 ||
382 !is_frag_child(frag->frag, prev_frag)) {
383 rb_erase(&frag->node, &ci->i_fragtree);
384 if (frag->split_by > 0)
385 ci->i_fragtree_nsplits--;
386 kfree(frag);
387 }
388 frag = NULL;
389 }
390 if (!frag) {
391 frag = __get_or_create_frag(ci, id);
392 if (IS_ERR(frag))
393 continue;
394 }
395 if (frag->split_by == 0)
396 ci->i_fragtree_nsplits++;
397 frag->split_by = split_by;
398 dout(" frag %x split by %d\n", frag->frag, frag->split_by);
399 prev_frag = frag;
400 }
401 while (rb_node) {
402 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
403 rb_node = rb_next(rb_node);
404 /* delete stale split/leaf node */
405 if (frag->split_by > 0 ||
406 !is_frag_child(frag->frag, prev_frag)) {
407 rb_erase(&frag->node, &ci->i_fragtree);
408 if (frag->split_by > 0)
409 ci->i_fragtree_nsplits--;
410 kfree(frag);
411 }
412 }
413 out_unlock:
414 mutex_unlock(&ci->i_fragtree_mutex);
415 return 0;
416 }
417
418 /*
419 * initialize a newly allocated inode.
420 */
421 struct inode *ceph_alloc_inode(struct super_block *sb)
422 {
423 struct ceph_inode_info *ci;
424 int i;
425
426 ci = kmem_cache_alloc(ceph_inode_cachep, GFP_NOFS);
427 if (!ci)
428 return NULL;
429
430 dout("alloc_inode %p\n", &ci->vfs_inode);
431
432 spin_lock_init(&ci->i_ceph_lock);
433
434 ci->i_version = 0;
435 ci->i_inline_version = 0;
436 ci->i_time_warp_seq = 0;
437 ci->i_ceph_flags = 0;
438 atomic64_set(&ci->i_ordered_count, 1);
439 atomic64_set(&ci->i_release_count, 1);
440 atomic64_set(&ci->i_complete_seq[0], 0);
441 atomic64_set(&ci->i_complete_seq[1], 0);
442 ci->i_symlink = NULL;
443
444 ci->i_max_bytes = 0;
445 ci->i_max_files = 0;
446
447 memset(&ci->i_dir_layout, 0, sizeof(ci->i_dir_layout));
448 RCU_INIT_POINTER(ci->i_layout.pool_ns, NULL);
449
450 ci->i_fragtree = RB_ROOT;
451 mutex_init(&ci->i_fragtree_mutex);
452
453 ci->i_xattrs.blob = NULL;
454 ci->i_xattrs.prealloc_blob = NULL;
455 ci->i_xattrs.dirty = false;
456 ci->i_xattrs.index = RB_ROOT;
457 ci->i_xattrs.count = 0;
458 ci->i_xattrs.names_size = 0;
459 ci->i_xattrs.vals_size = 0;
460 ci->i_xattrs.version = 0;
461 ci->i_xattrs.index_version = 0;
462
463 ci->i_caps = RB_ROOT;
464 ci->i_auth_cap = NULL;
465 ci->i_dirty_caps = 0;
466 ci->i_flushing_caps = 0;
467 INIT_LIST_HEAD(&ci->i_dirty_item);
468 INIT_LIST_HEAD(&ci->i_flushing_item);
469 ci->i_prealloc_cap_flush = NULL;
470 INIT_LIST_HEAD(&ci->i_cap_flush_list);
471 init_waitqueue_head(&ci->i_cap_wq);
472 ci->i_hold_caps_min = 0;
473 ci->i_hold_caps_max = 0;
474 INIT_LIST_HEAD(&ci->i_cap_delay_list);
475 INIT_LIST_HEAD(&ci->i_cap_snaps);
476 ci->i_head_snapc = NULL;
477 ci->i_snap_caps = 0;
478
479 for (i = 0; i < CEPH_FILE_MODE_BITS; i++)
480 ci->i_nr_by_mode[i] = 0;
481
482 mutex_init(&ci->i_truncate_mutex);
483 ci->i_truncate_seq = 0;
484 ci->i_truncate_size = 0;
485 ci->i_truncate_pending = 0;
486
487 ci->i_max_size = 0;
488 ci->i_reported_size = 0;
489 ci->i_wanted_max_size = 0;
490 ci->i_requested_max_size = 0;
491
492 ci->i_pin_ref = 0;
493 ci->i_rd_ref = 0;
494 ci->i_rdcache_ref = 0;
495 ci->i_wr_ref = 0;
496 ci->i_wb_ref = 0;
497 ci->i_wrbuffer_ref = 0;
498 ci->i_wrbuffer_ref_head = 0;
499 atomic_set(&ci->i_filelock_ref, 0);
500 atomic_set(&ci->i_shared_gen, 1);
501 ci->i_rdcache_gen = 0;
502 ci->i_rdcache_revoking = 0;
503
504 INIT_LIST_HEAD(&ci->i_unsafe_dirops);
505 INIT_LIST_HEAD(&ci->i_unsafe_iops);
506 spin_lock_init(&ci->i_unsafe_lock);
507
508 ci->i_snap_realm = NULL;
509 INIT_LIST_HEAD(&ci->i_snap_realm_item);
510 INIT_LIST_HEAD(&ci->i_snap_flush_item);
511
512 INIT_WORK(&ci->i_work, ceph_inode_work);
513 ci->i_work_mask = 0;
514 memset(&ci->i_btime, '\0', sizeof(ci->i_btime));
515
516 ceph_fscache_inode_init(ci);
517
518 ci->i_meta_err = 0;
519
520 return &ci->vfs_inode;
521 }
522
523 void ceph_free_inode(struct inode *inode)
524 {
525 struct ceph_inode_info *ci = ceph_inode(inode);
526
527 kfree(ci->i_symlink);
528 kmem_cache_free(ceph_inode_cachep, ci);
529 }
530
531 void ceph_evict_inode(struct inode *inode)
532 {
533 struct ceph_inode_info *ci = ceph_inode(inode);
534 struct ceph_inode_frag *frag;
535 struct rb_node *n;
536
537 dout("evict_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode));
538
539 truncate_inode_pages_final(&inode->i_data);
540 clear_inode(inode);
541
542 ceph_fscache_unregister_inode_cookie(ci);
543
544 __ceph_remove_caps(ci);
545
546 if (__ceph_has_any_quota(ci))
547 ceph_adjust_quota_realms_count(inode, false);
548
549 /*
550 * we may still have a snap_realm reference if there are stray
551 * caps in i_snap_caps.
552 */
553 if (ci->i_snap_realm) {
554 struct ceph_mds_client *mdsc =
555 ceph_inode_to_client(inode)->mdsc;
556 if (ceph_snap(inode) == CEPH_NOSNAP) {
557 struct ceph_snap_realm *realm = ci->i_snap_realm;
558 dout(" dropping residual ref to snap realm %p\n",
559 realm);
560 spin_lock(&realm->inodes_with_caps_lock);
561 list_del_init(&ci->i_snap_realm_item);
562 ci->i_snap_realm = NULL;
563 if (realm->ino == ci->i_vino.ino)
564 realm->inode = NULL;
565 spin_unlock(&realm->inodes_with_caps_lock);
566 ceph_put_snap_realm(mdsc, realm);
567 } else {
568 ceph_put_snapid_map(mdsc, ci->i_snapid_map);
569 ci->i_snap_realm = NULL;
570 }
571 }
572
573 while ((n = rb_first(&ci->i_fragtree)) != NULL) {
574 frag = rb_entry(n, struct ceph_inode_frag, node);
575 rb_erase(n, &ci->i_fragtree);
576 kfree(frag);
577 }
578 ci->i_fragtree_nsplits = 0;
579
580 __ceph_destroy_xattrs(ci);
581 if (ci->i_xattrs.blob)
582 ceph_buffer_put(ci->i_xattrs.blob);
583 if (ci->i_xattrs.prealloc_blob)
584 ceph_buffer_put(ci->i_xattrs.prealloc_blob);
585
586 ceph_put_string(rcu_dereference_raw(ci->i_layout.pool_ns));
587 }
588
589 static inline blkcnt_t calc_inode_blocks(u64 size)
590 {
591 return (size + (1<<9) - 1) >> 9;
592 }
593
594 /*
595 * Helpers to fill in size, ctime, mtime, and atime. We have to be
596 * careful because either the client or MDS may have more up to date
597 * info, depending on which capabilities are held, and whether
598 * time_warp_seq or truncate_seq have increased. (Ordinarily, mtime
599 * and size are monotonically increasing, except when utimes() or
600 * truncate() increments the corresponding _seq values.)
601 */
602 int ceph_fill_file_size(struct inode *inode, int issued,
603 u32 truncate_seq, u64 truncate_size, u64 size)
604 {
605 struct ceph_inode_info *ci = ceph_inode(inode);
606 int queue_trunc = 0;
607
608 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 ||
609 (truncate_seq == ci->i_truncate_seq && size > inode->i_size)) {
610 dout("size %lld -> %llu\n", inode->i_size, size);
611 if (size > 0 && S_ISDIR(inode->i_mode)) {
612 pr_err("fill_file_size non-zero size for directory\n");
613 size = 0;
614 }
615 i_size_write(inode, size);
616 inode->i_blocks = calc_inode_blocks(size);
617 ci->i_reported_size = size;
618 if (truncate_seq != ci->i_truncate_seq) {
619 dout("truncate_seq %u -> %u\n",
620 ci->i_truncate_seq, truncate_seq);
621 ci->i_truncate_seq = truncate_seq;
622
623 /* the MDS should have revoked these caps */
624 WARN_ON_ONCE(issued & (CEPH_CAP_FILE_EXCL |
625 CEPH_CAP_FILE_RD |
626 CEPH_CAP_FILE_WR |
627 CEPH_CAP_FILE_LAZYIO));
628 /*
629 * If we hold relevant caps, or in the case where we're
630 * not the only client referencing this file and we
631 * don't hold those caps, then we need to check whether
632 * the file is either opened or mmaped
633 */
634 if ((issued & (CEPH_CAP_FILE_CACHE|
635 CEPH_CAP_FILE_BUFFER)) ||
636 mapping_mapped(inode->i_mapping) ||
637 __ceph_caps_file_wanted(ci)) {
638 ci->i_truncate_pending++;
639 queue_trunc = 1;
640 }
641 }
642 }
643 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 &&
644 ci->i_truncate_size != truncate_size) {
645 dout("truncate_size %lld -> %llu\n", ci->i_truncate_size,
646 truncate_size);
647 ci->i_truncate_size = truncate_size;
648 }
649
650 if (queue_trunc)
651 ceph_fscache_invalidate(inode);
652
653 return queue_trunc;
654 }
655
656 void ceph_fill_file_time(struct inode *inode, int issued,
657 u64 time_warp_seq, struct timespec64 *ctime,
658 struct timespec64 *mtime, struct timespec64 *atime)
659 {
660 struct ceph_inode_info *ci = ceph_inode(inode);
661 int warn = 0;
662
663 if (issued & (CEPH_CAP_FILE_EXCL|
664 CEPH_CAP_FILE_WR|
665 CEPH_CAP_FILE_BUFFER|
666 CEPH_CAP_AUTH_EXCL|
667 CEPH_CAP_XATTR_EXCL)) {
668 if (ci->i_version == 0 ||
669 timespec64_compare(ctime, &inode->i_ctime) > 0) {
670 dout("ctime %lld.%09ld -> %lld.%09ld inc w/ cap\n",
671 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
672 ctime->tv_sec, ctime->tv_nsec);
673 inode->i_ctime = *ctime;
674 }
675 if (ci->i_version == 0 ||
676 ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) {
677 /* the MDS did a utimes() */
678 dout("mtime %lld.%09ld -> %lld.%09ld "
679 "tw %d -> %d\n",
680 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
681 mtime->tv_sec, mtime->tv_nsec,
682 ci->i_time_warp_seq, (int)time_warp_seq);
683
684 inode->i_mtime = *mtime;
685 inode->i_atime = *atime;
686 ci->i_time_warp_seq = time_warp_seq;
687 } else if (time_warp_seq == ci->i_time_warp_seq) {
688 /* nobody did utimes(); take the max */
689 if (timespec64_compare(mtime, &inode->i_mtime) > 0) {
690 dout("mtime %lld.%09ld -> %lld.%09ld inc\n",
691 inode->i_mtime.tv_sec,
692 inode->i_mtime.tv_nsec,
693 mtime->tv_sec, mtime->tv_nsec);
694 inode->i_mtime = *mtime;
695 }
696 if (timespec64_compare(atime, &inode->i_atime) > 0) {
697 dout("atime %lld.%09ld -> %lld.%09ld inc\n",
698 inode->i_atime.tv_sec,
699 inode->i_atime.tv_nsec,
700 atime->tv_sec, atime->tv_nsec);
701 inode->i_atime = *atime;
702 }
703 } else if (issued & CEPH_CAP_FILE_EXCL) {
704 /* we did a utimes(); ignore mds values */
705 } else {
706 warn = 1;
707 }
708 } else {
709 /* we have no write|excl caps; whatever the MDS says is true */
710 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) {
711 inode->i_ctime = *ctime;
712 inode->i_mtime = *mtime;
713 inode->i_atime = *atime;
714 ci->i_time_warp_seq = time_warp_seq;
715 } else {
716 warn = 1;
717 }
718 }
719 if (warn) /* time_warp_seq shouldn't go backwards */
720 dout("%p mds time_warp_seq %llu < %u\n",
721 inode, time_warp_seq, ci->i_time_warp_seq);
722 }
723
724 /*
725 * Populate an inode based on info from mds. May be called on new or
726 * existing inodes.
727 */
728 static int fill_inode(struct inode *inode, struct page *locked_page,
729 struct ceph_mds_reply_info_in *iinfo,
730 struct ceph_mds_reply_dirfrag *dirinfo,
731 struct ceph_mds_session *session,
732 unsigned long ttl_from, int cap_fmode,
733 struct ceph_cap_reservation *caps_reservation)
734 {
735 struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
736 struct ceph_mds_reply_inode *info = iinfo->in;
737 struct ceph_inode_info *ci = ceph_inode(inode);
738 int issued, new_issued, info_caps;
739 struct timespec64 mtime, atime, ctime;
740 struct ceph_buffer *xattr_blob = NULL;
741 struct ceph_buffer *old_blob = NULL;
742 struct ceph_string *pool_ns = NULL;
743 struct ceph_cap *new_cap = NULL;
744 int err = 0;
745 bool wake = false;
746 bool queue_trunc = false;
747 bool new_version = false;
748 bool fill_inline = false;
749
750 dout("fill_inode %p ino %llx.%llx v %llu had %llu\n",
751 inode, ceph_vinop(inode), le64_to_cpu(info->version),
752 ci->i_version);
753
754 info_caps = le32_to_cpu(info->cap.caps);
755
756 /* prealloc new cap struct */
757 if (info_caps && ceph_snap(inode) == CEPH_NOSNAP)
758 new_cap = ceph_get_cap(mdsc, caps_reservation);
759
760 /*
761 * prealloc xattr data, if it looks like we'll need it. only
762 * if len > 4 (meaning there are actually xattrs; the first 4
763 * bytes are the xattr count).
764 */
765 if (iinfo->xattr_len > 4) {
766 xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS);
767 if (!xattr_blob)
768 pr_err("fill_inode ENOMEM xattr blob %d bytes\n",
769 iinfo->xattr_len);
770 }
771
772 if (iinfo->pool_ns_len > 0)
773 pool_ns = ceph_find_or_create_string(iinfo->pool_ns_data,
774 iinfo->pool_ns_len);
775
776 if (ceph_snap(inode) != CEPH_NOSNAP && !ci->i_snapid_map)
777 ci->i_snapid_map = ceph_get_snapid_map(mdsc, ceph_snap(inode));
778
779 spin_lock(&ci->i_ceph_lock);
780
781 /*
782 * provided version will be odd if inode value is projected,
783 * even if stable. skip the update if we have newer stable
784 * info (ours>=theirs, e.g. due to racing mds replies), unless
785 * we are getting projected (unstable) info (in which case the
786 * version is odd, and we want ours>theirs).
787 * us them
788 * 2 2 skip
789 * 3 2 skip
790 * 3 3 update
791 */
792 if (ci->i_version == 0 ||
793 ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
794 le64_to_cpu(info->version) > (ci->i_version & ~1)))
795 new_version = true;
796
797 /* Update change_attribute */
798 inode_set_max_iversion_raw(inode, iinfo->change_attr);
799
800 __ceph_caps_issued(ci, &issued);
801 issued |= __ceph_caps_dirty(ci);
802 new_issued = ~issued & info_caps;
803
804 /* update inode */
805 inode->i_rdev = le32_to_cpu(info->rdev);
806 /* directories have fl_stripe_unit set to zero */
807 if (le32_to_cpu(info->layout.fl_stripe_unit))
808 inode->i_blkbits =
809 fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1;
810 else
811 inode->i_blkbits = CEPH_BLOCK_SHIFT;
812
813 __ceph_update_quota(ci, iinfo->max_bytes, iinfo->max_files);
814
815 if ((new_version || (new_issued & CEPH_CAP_AUTH_SHARED)) &&
816 (issued & CEPH_CAP_AUTH_EXCL) == 0) {
817 inode->i_mode = le32_to_cpu(info->mode);
818 inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(info->uid));
819 inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(info->gid));
820 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
821 from_kuid(&init_user_ns, inode->i_uid),
822 from_kgid(&init_user_ns, inode->i_gid));
823 ceph_decode_timespec64(&ci->i_btime, &iinfo->btime);
824 ceph_decode_timespec64(&ci->i_snap_btime, &iinfo->snap_btime);
825 }
826
827 if ((new_version || (new_issued & CEPH_CAP_LINK_SHARED)) &&
828 (issued & CEPH_CAP_LINK_EXCL) == 0)
829 set_nlink(inode, le32_to_cpu(info->nlink));
830
831 if (new_version || (new_issued & CEPH_CAP_ANY_RD)) {
832 /* be careful with mtime, atime, size */
833 ceph_decode_timespec64(&atime, &info->atime);
834 ceph_decode_timespec64(&mtime, &info->mtime);
835 ceph_decode_timespec64(&ctime, &info->ctime);
836 ceph_fill_file_time(inode, issued,
837 le32_to_cpu(info->time_warp_seq),
838 &ctime, &mtime, &atime);
839 }
840
841 if (new_version || (info_caps & CEPH_CAP_FILE_SHARED)) {
842 ci->i_files = le64_to_cpu(info->files);
843 ci->i_subdirs = le64_to_cpu(info->subdirs);
844 }
845
846 if (new_version ||
847 (new_issued & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR))) {
848 s64 old_pool = ci->i_layout.pool_id;
849 struct ceph_string *old_ns;
850
851 ceph_file_layout_from_legacy(&ci->i_layout, &info->layout);
852 old_ns = rcu_dereference_protected(ci->i_layout.pool_ns,
853 lockdep_is_held(&ci->i_ceph_lock));
854 rcu_assign_pointer(ci->i_layout.pool_ns, pool_ns);
855
856 if (ci->i_layout.pool_id != old_pool || pool_ns != old_ns)
857 ci->i_ceph_flags &= ~CEPH_I_POOL_PERM;
858
859 pool_ns = old_ns;
860
861 queue_trunc = ceph_fill_file_size(inode, issued,
862 le32_to_cpu(info->truncate_seq),
863 le64_to_cpu(info->truncate_size),
864 le64_to_cpu(info->size));
865 /* only update max_size on auth cap */
866 if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
867 ci->i_max_size != le64_to_cpu(info->max_size)) {
868 dout("max_size %lld -> %llu\n", ci->i_max_size,
869 le64_to_cpu(info->max_size));
870 ci->i_max_size = le64_to_cpu(info->max_size);
871 }
872 }
873
874 /* layout and rstat are not tracked by capability, update them if
875 * the inode info is from auth mds */
876 if (new_version || (info->cap.flags & CEPH_CAP_FLAG_AUTH)) {
877 if (S_ISDIR(inode->i_mode)) {
878 ci->i_dir_layout = iinfo->dir_layout;
879 ci->i_rbytes = le64_to_cpu(info->rbytes);
880 ci->i_rfiles = le64_to_cpu(info->rfiles);
881 ci->i_rsubdirs = le64_to_cpu(info->rsubdirs);
882 ci->i_dir_pin = iinfo->dir_pin;
883 ceph_decode_timespec64(&ci->i_rctime, &info->rctime);
884 }
885 }
886
887 /* xattrs */
888 /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
889 if ((ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL)) &&
890 le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) {
891 if (ci->i_xattrs.blob)
892 old_blob = ci->i_xattrs.blob;
893 ci->i_xattrs.blob = xattr_blob;
894 if (xattr_blob)
895 memcpy(ci->i_xattrs.blob->vec.iov_base,
896 iinfo->xattr_data, iinfo->xattr_len);
897 ci->i_xattrs.version = le64_to_cpu(info->xattr_version);
898 ceph_forget_all_cached_acls(inode);
899 ceph_security_invalidate_secctx(inode);
900 xattr_blob = NULL;
901 }
902
903 /* finally update i_version */
904 if (le64_to_cpu(info->version) > ci->i_version)
905 ci->i_version = le64_to_cpu(info->version);
906
907 inode->i_mapping->a_ops = &ceph_aops;
908
909 switch (inode->i_mode & S_IFMT) {
910 case S_IFIFO:
911 case S_IFBLK:
912 case S_IFCHR:
913 case S_IFSOCK:
914 inode->i_blkbits = PAGE_SHIFT;
915 init_special_inode(inode, inode->i_mode, inode->i_rdev);
916 inode->i_op = &ceph_file_iops;
917 break;
918 case S_IFREG:
919 inode->i_op = &ceph_file_iops;
920 inode->i_fop = &ceph_file_fops;
921 break;
922 case S_IFLNK:
923 inode->i_op = &ceph_symlink_iops;
924 if (!ci->i_symlink) {
925 u32 symlen = iinfo->symlink_len;
926 char *sym;
927
928 spin_unlock(&ci->i_ceph_lock);
929
930 if (symlen != i_size_read(inode)) {
931 pr_err("fill_inode %llx.%llx BAD symlink "
932 "size %lld\n", ceph_vinop(inode),
933 i_size_read(inode));
934 i_size_write(inode, symlen);
935 inode->i_blocks = calc_inode_blocks(symlen);
936 }
937
938 err = -ENOMEM;
939 sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS);
940 if (!sym)
941 goto out;
942
943 spin_lock(&ci->i_ceph_lock);
944 if (!ci->i_symlink)
945 ci->i_symlink = sym;
946 else
947 kfree(sym); /* lost a race */
948 }
949 inode->i_link = ci->i_symlink;
950 break;
951 case S_IFDIR:
952 inode->i_op = &ceph_dir_iops;
953 inode->i_fop = &ceph_dir_fops;
954 break;
955 default:
956 pr_err("fill_inode %llx.%llx BAD mode 0%o\n",
957 ceph_vinop(inode), inode->i_mode);
958 }
959
960 /* were we issued a capability? */
961 if (info_caps) {
962 if (ceph_snap(inode) == CEPH_NOSNAP) {
963 ceph_add_cap(inode, session,
964 le64_to_cpu(info->cap.cap_id),
965 cap_fmode, info_caps,
966 le32_to_cpu(info->cap.wanted),
967 le32_to_cpu(info->cap.seq),
968 le32_to_cpu(info->cap.mseq),
969 le64_to_cpu(info->cap.realm),
970 info->cap.flags, &new_cap);
971
972 /* set dir completion flag? */
973 if (S_ISDIR(inode->i_mode) &&
974 ci->i_files == 0 && ci->i_subdirs == 0 &&
975 (info_caps & CEPH_CAP_FILE_SHARED) &&
976 (issued & CEPH_CAP_FILE_EXCL) == 0 &&
977 !__ceph_dir_is_complete(ci)) {
978 dout(" marking %p complete (empty)\n", inode);
979 i_size_write(inode, 0);
980 __ceph_dir_set_complete(ci,
981 atomic64_read(&ci->i_release_count),
982 atomic64_read(&ci->i_ordered_count));
983 }
984
985 wake = true;
986 } else {
987 dout(" %p got snap_caps %s\n", inode,
988 ceph_cap_string(info_caps));
989 ci->i_snap_caps |= info_caps;
990 if (cap_fmode >= 0)
991 __ceph_get_fmode(ci, cap_fmode);
992 }
993 } else if (cap_fmode >= 0) {
994 pr_warn("mds issued no caps on %llx.%llx\n",
995 ceph_vinop(inode));
996 __ceph_get_fmode(ci, cap_fmode);
997 }
998
999 if (iinfo->inline_version > 0 &&
1000 iinfo->inline_version >= ci->i_inline_version) {
1001 int cache_caps = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
1002 ci->i_inline_version = iinfo->inline_version;
1003 if (ci->i_inline_version != CEPH_INLINE_NONE &&
1004 (locked_page || (info_caps & cache_caps)))
1005 fill_inline = true;
1006 }
1007
1008 spin_unlock(&ci->i_ceph_lock);
1009
1010 if (fill_inline)
1011 ceph_fill_inline_data(inode, locked_page,
1012 iinfo->inline_data, iinfo->inline_len);
1013
1014 if (wake)
1015 wake_up_all(&ci->i_cap_wq);
1016
1017 /* queue truncate if we saw i_size decrease */
1018 if (queue_trunc)
1019 ceph_queue_vmtruncate(inode);
1020
1021 /* populate frag tree */
1022 if (S_ISDIR(inode->i_mode))
1023 ceph_fill_fragtree(inode, &info->fragtree, dirinfo);
1024
1025 /* update delegation info? */
1026 if (dirinfo)
1027 ceph_fill_dirfrag(inode, dirinfo);
1028
1029 err = 0;
1030 out:
1031 if (new_cap)
1032 ceph_put_cap(mdsc, new_cap);
1033 ceph_buffer_put(old_blob);
1034 ceph_buffer_put(xattr_blob);
1035 ceph_put_string(pool_ns);
1036 return err;
1037 }
1038
1039 /*
1040 * caller should hold session s_mutex and dentry->d_lock.
1041 */
1042 static void __update_dentry_lease(struct inode *dir, struct dentry *dentry,
1043 struct ceph_mds_reply_lease *lease,
1044 struct ceph_mds_session *session,
1045 unsigned long from_time,
1046 struct ceph_mds_session **old_lease_session)
1047 {
1048 struct ceph_dentry_info *di = ceph_dentry(dentry);
1049 long unsigned duration = le32_to_cpu(lease->duration_ms);
1050 long unsigned ttl = from_time + (duration * HZ) / 1000;
1051 long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000;
1052
1053 dout("update_dentry_lease %p duration %lu ms ttl %lu\n",
1054 dentry, duration, ttl);
1055
1056 /* only track leases on regular dentries */
1057 if (ceph_snap(dir) != CEPH_NOSNAP)
1058 return;
1059
1060 di->lease_shared_gen = atomic_read(&ceph_inode(dir)->i_shared_gen);
1061 if (duration == 0) {
1062 __ceph_dentry_dir_lease_touch(di);
1063 return;
1064 }
1065
1066 if (di->lease_gen == session->s_cap_gen &&
1067 time_before(ttl, di->time))
1068 return; /* we already have a newer lease. */
1069
1070 if (di->lease_session && di->lease_session != session) {
1071 *old_lease_session = di->lease_session;
1072 di->lease_session = NULL;
1073 }
1074
1075 if (!di->lease_session)
1076 di->lease_session = ceph_get_mds_session(session);
1077 di->lease_gen = session->s_cap_gen;
1078 di->lease_seq = le32_to_cpu(lease->seq);
1079 di->lease_renew_after = half_ttl;
1080 di->lease_renew_from = 0;
1081 di->time = ttl;
1082
1083 __ceph_dentry_lease_touch(di);
1084 }
1085
1086 static inline void update_dentry_lease(struct inode *dir, struct dentry *dentry,
1087 struct ceph_mds_reply_lease *lease,
1088 struct ceph_mds_session *session,
1089 unsigned long from_time)
1090 {
1091 struct ceph_mds_session *old_lease_session = NULL;
1092 spin_lock(&dentry->d_lock);
1093 __update_dentry_lease(dir, dentry, lease, session, from_time,
1094 &old_lease_session);
1095 spin_unlock(&dentry->d_lock);
1096 if (old_lease_session)
1097 ceph_put_mds_session(old_lease_session);
1098 }
1099
1100 /*
1101 * update dentry lease without having parent inode locked
1102 */
1103 static void update_dentry_lease_careful(struct dentry *dentry,
1104 struct ceph_mds_reply_lease *lease,
1105 struct ceph_mds_session *session,
1106 unsigned long from_time,
1107 char *dname, u32 dname_len,
1108 struct ceph_vino *pdvino,
1109 struct ceph_vino *ptvino)
1110
1111 {
1112 struct inode *dir;
1113 struct ceph_mds_session *old_lease_session = NULL;
1114
1115 spin_lock(&dentry->d_lock);
1116 /* make sure dentry's name matches target */
1117 if (dentry->d_name.len != dname_len ||
1118 memcmp(dentry->d_name.name, dname, dname_len))
1119 goto out_unlock;
1120
1121 dir = d_inode(dentry->d_parent);
1122 /* make sure parent matches dvino */
1123 if (!ceph_ino_compare(dir, pdvino))
1124 goto out_unlock;
1125
1126 /* make sure dentry's inode matches target. NULL ptvino means that
1127 * we expect a negative dentry */
1128 if (ptvino) {
1129 if (d_really_is_negative(dentry))
1130 goto out_unlock;
1131 if (!ceph_ino_compare(d_inode(dentry), ptvino))
1132 goto out_unlock;
1133 } else {
1134 if (d_really_is_positive(dentry))
1135 goto out_unlock;
1136 }
1137
1138 __update_dentry_lease(dir, dentry, lease, session,
1139 from_time, &old_lease_session);
1140 out_unlock:
1141 spin_unlock(&dentry->d_lock);
1142 if (old_lease_session)
1143 ceph_put_mds_session(old_lease_session);
1144 }
1145
1146 /*
1147 * splice a dentry to an inode.
1148 * caller must hold directory i_mutex for this to be safe.
1149 */
1150 static int splice_dentry(struct dentry **pdn, struct inode *in)
1151 {
1152 struct dentry *dn = *pdn;
1153 struct dentry *realdn;
1154
1155 BUG_ON(d_inode(dn));
1156
1157 if (S_ISDIR(in->i_mode)) {
1158 /* If inode is directory, d_splice_alias() below will remove
1159 * 'realdn' from its origin parent. We need to ensure that
1160 * origin parent's readdir cache will not reference 'realdn'
1161 */
1162 realdn = d_find_any_alias(in);
1163 if (realdn) {
1164 struct ceph_dentry_info *di = ceph_dentry(realdn);
1165 spin_lock(&realdn->d_lock);
1166
1167 realdn->d_op->d_prune(realdn);
1168
1169 di->time = jiffies;
1170 di->lease_shared_gen = 0;
1171 di->offset = 0;
1172
1173 spin_unlock(&realdn->d_lock);
1174 dput(realdn);
1175 }
1176 }
1177
1178 /* dn must be unhashed */
1179 if (!d_unhashed(dn))
1180 d_drop(dn);
1181 realdn = d_splice_alias(in, dn);
1182 if (IS_ERR(realdn)) {
1183 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
1184 PTR_ERR(realdn), dn, in, ceph_vinop(in));
1185 return PTR_ERR(realdn);
1186 }
1187
1188 if (realdn) {
1189 dout("dn %p (%d) spliced with %p (%d) "
1190 "inode %p ino %llx.%llx\n",
1191 dn, d_count(dn),
1192 realdn, d_count(realdn),
1193 d_inode(realdn), ceph_vinop(d_inode(realdn)));
1194 dput(dn);
1195 *pdn = realdn;
1196 } else {
1197 BUG_ON(!ceph_dentry(dn));
1198 dout("dn %p attached to %p ino %llx.%llx\n",
1199 dn, d_inode(dn), ceph_vinop(d_inode(dn)));
1200 }
1201 return 0;
1202 }
1203
1204 /*
1205 * Incorporate results into the local cache. This is either just
1206 * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
1207 * after a lookup).
1208 *
1209 * A reply may contain
1210 * a directory inode along with a dentry.
1211 * and/or a target inode
1212 *
1213 * Called with snap_rwsem (read).
1214 */
1215 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req)
1216 {
1217 struct ceph_mds_session *session = req->r_session;
1218 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1219 struct inode *in = NULL;
1220 struct ceph_vino tvino, dvino;
1221 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
1222 int err = 0;
1223
1224 dout("fill_trace %p is_dentry %d is_target %d\n", req,
1225 rinfo->head->is_dentry, rinfo->head->is_target);
1226
1227 if (!rinfo->head->is_target && !rinfo->head->is_dentry) {
1228 dout("fill_trace reply is empty!\n");
1229 if (rinfo->head->result == 0 && req->r_parent)
1230 ceph_invalidate_dir_request(req);
1231 return 0;
1232 }
1233
1234 if (rinfo->head->is_dentry) {
1235 struct inode *dir = req->r_parent;
1236
1237 if (dir) {
1238 err = fill_inode(dir, NULL,
1239 &rinfo->diri, rinfo->dirfrag,
1240 session, req->r_request_started, -1,
1241 &req->r_caps_reservation);
1242 if (err < 0)
1243 goto done;
1244 } else {
1245 WARN_ON_ONCE(1);
1246 }
1247
1248 if (dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME &&
1249 test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1250 !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
1251 struct qstr dname;
1252 struct dentry *dn, *parent;
1253
1254 BUG_ON(!rinfo->head->is_target);
1255 BUG_ON(req->r_dentry);
1256
1257 parent = d_find_any_alias(dir);
1258 BUG_ON(!parent);
1259
1260 dname.name = rinfo->dname;
1261 dname.len = rinfo->dname_len;
1262 dname.hash = full_name_hash(parent, dname.name, dname.len);
1263 tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1264 tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1265 retry_lookup:
1266 dn = d_lookup(parent, &dname);
1267 dout("d_lookup on parent=%p name=%.*s got %p\n",
1268 parent, dname.len, dname.name, dn);
1269
1270 if (!dn) {
1271 dn = d_alloc(parent, &dname);
1272 dout("d_alloc %p '%.*s' = %p\n", parent,
1273 dname.len, dname.name, dn);
1274 if (!dn) {
1275 dput(parent);
1276 err = -ENOMEM;
1277 goto done;
1278 }
1279 err = 0;
1280 } else if (d_really_is_positive(dn) &&
1281 (ceph_ino(d_inode(dn)) != tvino.ino ||
1282 ceph_snap(d_inode(dn)) != tvino.snap)) {
1283 dout(" dn %p points to wrong inode %p\n",
1284 dn, d_inode(dn));
1285 ceph_dir_clear_ordered(dir);
1286 d_delete(dn);
1287 dput(dn);
1288 goto retry_lookup;
1289 }
1290
1291 req->r_dentry = dn;
1292 dput(parent);
1293 }
1294 }
1295
1296 if (rinfo->head->is_target) {
1297 tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1298 tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1299
1300 in = ceph_get_inode(sb, tvino);
1301 if (IS_ERR(in)) {
1302 err = PTR_ERR(in);
1303 goto done;
1304 }
1305 req->r_target_inode = in;
1306
1307 err = fill_inode(in, req->r_locked_page, &rinfo->targeti, NULL,
1308 session, req->r_request_started,
1309 (!test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) &&
1310 rinfo->head->result == 0) ? req->r_fmode : -1,
1311 &req->r_caps_reservation);
1312 if (err < 0) {
1313 pr_err("fill_inode badness %p %llx.%llx\n",
1314 in, ceph_vinop(in));
1315 goto done;
1316 }
1317 }
1318
1319 /*
1320 * ignore null lease/binding on snapdir ENOENT, or else we
1321 * will have trouble splicing in the virtual snapdir later
1322 */
1323 if (rinfo->head->is_dentry &&
1324 !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) &&
1325 test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1326 (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name,
1327 fsc->mount_options->snapdir_name,
1328 req->r_dentry->d_name.len))) {
1329 /*
1330 * lookup link rename : null -> possibly existing inode
1331 * mknod symlink mkdir : null -> new inode
1332 * unlink : linked -> null
1333 */
1334 struct inode *dir = req->r_parent;
1335 struct dentry *dn = req->r_dentry;
1336 bool have_dir_cap, have_lease;
1337
1338 BUG_ON(!dn);
1339 BUG_ON(!dir);
1340 BUG_ON(d_inode(dn->d_parent) != dir);
1341
1342 dvino.ino = le64_to_cpu(rinfo->diri.in->ino);
1343 dvino.snap = le64_to_cpu(rinfo->diri.in->snapid);
1344
1345 BUG_ON(ceph_ino(dir) != dvino.ino);
1346 BUG_ON(ceph_snap(dir) != dvino.snap);
1347
1348 /* do we have a lease on the whole dir? */
1349 have_dir_cap =
1350 (le32_to_cpu(rinfo->diri.in->cap.caps) &
1351 CEPH_CAP_FILE_SHARED);
1352
1353 /* do we have a dn lease? */
1354 have_lease = have_dir_cap ||
1355 le32_to_cpu(rinfo->dlease->duration_ms);
1356 if (!have_lease)
1357 dout("fill_trace no dentry lease or dir cap\n");
1358
1359 /* rename? */
1360 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) {
1361 struct inode *olddir = req->r_old_dentry_dir;
1362 BUG_ON(!olddir);
1363
1364 dout(" src %p '%pd' dst %p '%pd'\n",
1365 req->r_old_dentry,
1366 req->r_old_dentry,
1367 dn, dn);
1368 dout("fill_trace doing d_move %p -> %p\n",
1369 req->r_old_dentry, dn);
1370
1371 /* d_move screws up sibling dentries' offsets */
1372 ceph_dir_clear_ordered(dir);
1373 ceph_dir_clear_ordered(olddir);
1374
1375 d_move(req->r_old_dentry, dn);
1376 dout(" src %p '%pd' dst %p '%pd'\n",
1377 req->r_old_dentry,
1378 req->r_old_dentry,
1379 dn, dn);
1380
1381 /* ensure target dentry is invalidated, despite
1382 rehashing bug in vfs_rename_dir */
1383 ceph_invalidate_dentry_lease(dn);
1384
1385 dout("dn %p gets new offset %lld\n", req->r_old_dentry,
1386 ceph_dentry(req->r_old_dentry)->offset);
1387
1388 /* swap r_dentry and r_old_dentry in case that
1389 * splice_dentry() gets called later. This is safe
1390 * because no other place will use them */
1391 req->r_dentry = req->r_old_dentry;
1392 req->r_old_dentry = dn;
1393 dn = req->r_dentry;
1394 }
1395
1396 /* null dentry? */
1397 if (!rinfo->head->is_target) {
1398 dout("fill_trace null dentry\n");
1399 if (d_really_is_positive(dn)) {
1400 dout("d_delete %p\n", dn);
1401 ceph_dir_clear_ordered(dir);
1402 d_delete(dn);
1403 } else if (have_lease) {
1404 if (d_unhashed(dn))
1405 d_add(dn, NULL);
1406 update_dentry_lease(dir, dn,
1407 rinfo->dlease, session,
1408 req->r_request_started);
1409 }
1410 goto done;
1411 }
1412
1413 /* attach proper inode */
1414 if (d_really_is_negative(dn)) {
1415 ceph_dir_clear_ordered(dir);
1416 ihold(in);
1417 err = splice_dentry(&req->r_dentry, in);
1418 if (err < 0)
1419 goto done;
1420 dn = req->r_dentry; /* may have spliced */
1421 } else if (d_really_is_positive(dn) && d_inode(dn) != in) {
1422 dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1423 dn, d_inode(dn), ceph_vinop(d_inode(dn)),
1424 ceph_vinop(in));
1425 d_invalidate(dn);
1426 have_lease = false;
1427 }
1428
1429 if (have_lease) {
1430 update_dentry_lease(dir, dn,
1431 rinfo->dlease, session,
1432 req->r_request_started);
1433 }
1434 dout(" final dn %p\n", dn);
1435 } else if ((req->r_op == CEPH_MDS_OP_LOOKUPSNAP ||
1436 req->r_op == CEPH_MDS_OP_MKSNAP) &&
1437 test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1438 !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
1439 struct inode *dir = req->r_parent;
1440
1441 /* fill out a snapdir LOOKUPSNAP dentry */
1442 BUG_ON(!dir);
1443 BUG_ON(ceph_snap(dir) != CEPH_SNAPDIR);
1444 BUG_ON(!req->r_dentry);
1445 dout(" linking snapped dir %p to dn %p\n", in, req->r_dentry);
1446 ceph_dir_clear_ordered(dir);
1447 ihold(in);
1448 err = splice_dentry(&req->r_dentry, in);
1449 if (err < 0)
1450 goto done;
1451 } else if (rinfo->head->is_dentry && req->r_dentry) {
1452 /* parent inode is not locked, be carefull */
1453 struct ceph_vino *ptvino = NULL;
1454 dvino.ino = le64_to_cpu(rinfo->diri.in->ino);
1455 dvino.snap = le64_to_cpu(rinfo->diri.in->snapid);
1456 if (rinfo->head->is_target) {
1457 tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1458 tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1459 ptvino = &tvino;
1460 }
1461 update_dentry_lease_careful(req->r_dentry, rinfo->dlease,
1462 session, req->r_request_started,
1463 rinfo->dname, rinfo->dname_len,
1464 &dvino, ptvino);
1465 }
1466 done:
1467 dout("fill_trace done err=%d\n", err);
1468 return err;
1469 }
1470
1471 /*
1472 * Prepopulate our cache with readdir results, leases, etc.
1473 */
1474 static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req,
1475 struct ceph_mds_session *session)
1476 {
1477 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1478 int i, err = 0;
1479
1480 for (i = 0; i < rinfo->dir_nr; i++) {
1481 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1482 struct ceph_vino vino;
1483 struct inode *in;
1484 int rc;
1485
1486 vino.ino = le64_to_cpu(rde->inode.in->ino);
1487 vino.snap = le64_to_cpu(rde->inode.in->snapid);
1488
1489 in = ceph_get_inode(req->r_dentry->d_sb, vino);
1490 if (IS_ERR(in)) {
1491 err = PTR_ERR(in);
1492 dout("new_inode badness got %d\n", err);
1493 continue;
1494 }
1495 rc = fill_inode(in, NULL, &rde->inode, NULL, session,
1496 req->r_request_started, -1,
1497 &req->r_caps_reservation);
1498 if (rc < 0) {
1499 pr_err("fill_inode badness on %p got %d\n", in, rc);
1500 err = rc;
1501 }
1502 /* avoid calling iput_final() in mds dispatch threads */
1503 ceph_async_iput(in);
1504 }
1505
1506 return err;
1507 }
1508
1509 void ceph_readdir_cache_release(struct ceph_readdir_cache_control *ctl)
1510 {
1511 if (ctl->page) {
1512 kunmap(ctl->page);
1513 put_page(ctl->page);
1514 ctl->page = NULL;
1515 }
1516 }
1517
1518 static int fill_readdir_cache(struct inode *dir, struct dentry *dn,
1519 struct ceph_readdir_cache_control *ctl,
1520 struct ceph_mds_request *req)
1521 {
1522 struct ceph_inode_info *ci = ceph_inode(dir);
1523 unsigned nsize = PAGE_SIZE / sizeof(struct dentry*);
1524 unsigned idx = ctl->index % nsize;
1525 pgoff_t pgoff = ctl->index / nsize;
1526
1527 if (!ctl->page || pgoff != page_index(ctl->page)) {
1528 ceph_readdir_cache_release(ctl);
1529 if (idx == 0)
1530 ctl->page = grab_cache_page(&dir->i_data, pgoff);
1531 else
1532 ctl->page = find_lock_page(&dir->i_data, pgoff);
1533 if (!ctl->page) {
1534 ctl->index = -1;
1535 return idx == 0 ? -ENOMEM : 0;
1536 }
1537 /* reading/filling the cache are serialized by
1538 * i_mutex, no need to use page lock */
1539 unlock_page(ctl->page);
1540 ctl->dentries = kmap(ctl->page);
1541 if (idx == 0)
1542 memset(ctl->dentries, 0, PAGE_SIZE);
1543 }
1544
1545 if (req->r_dir_release_cnt == atomic64_read(&ci->i_release_count) &&
1546 req->r_dir_ordered_cnt == atomic64_read(&ci->i_ordered_count)) {
1547 dout("readdir cache dn %p idx %d\n", dn, ctl->index);
1548 ctl->dentries[idx] = dn;
1549 ctl->index++;
1550 } else {
1551 dout("disable readdir cache\n");
1552 ctl->index = -1;
1553 }
1554 return 0;
1555 }
1556
1557 int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1558 struct ceph_mds_session *session)
1559 {
1560 struct dentry *parent = req->r_dentry;
1561 struct ceph_inode_info *ci = ceph_inode(d_inode(parent));
1562 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1563 struct qstr dname;
1564 struct dentry *dn;
1565 struct inode *in;
1566 int err = 0, skipped = 0, ret, i;
1567 struct ceph_mds_request_head *rhead = req->r_request->front.iov_base;
1568 u32 frag = le32_to_cpu(rhead->args.readdir.frag);
1569 u32 last_hash = 0;
1570 u32 fpos_offset;
1571 struct ceph_readdir_cache_control cache_ctl = {};
1572
1573 if (test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags))
1574 return readdir_prepopulate_inodes_only(req, session);
1575
1576 if (rinfo->hash_order) {
1577 if (req->r_path2) {
1578 last_hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1579 req->r_path2,
1580 strlen(req->r_path2));
1581 last_hash = ceph_frag_value(last_hash);
1582 } else if (rinfo->offset_hash) {
1583 /* mds understands offset_hash */
1584 WARN_ON_ONCE(req->r_readdir_offset != 2);
1585 last_hash = le32_to_cpu(rhead->args.readdir.offset_hash);
1586 }
1587 }
1588
1589 if (rinfo->dir_dir &&
1590 le32_to_cpu(rinfo->dir_dir->frag) != frag) {
1591 dout("readdir_prepopulate got new frag %x -> %x\n",
1592 frag, le32_to_cpu(rinfo->dir_dir->frag));
1593 frag = le32_to_cpu(rinfo->dir_dir->frag);
1594 if (!rinfo->hash_order)
1595 req->r_readdir_offset = 2;
1596 }
1597
1598 if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
1599 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1600 rinfo->dir_nr, parent);
1601 } else {
1602 dout("readdir_prepopulate %d items under dn %p\n",
1603 rinfo->dir_nr, parent);
1604 if (rinfo->dir_dir)
1605 ceph_fill_dirfrag(d_inode(parent), rinfo->dir_dir);
1606
1607 if (ceph_frag_is_leftmost(frag) &&
1608 req->r_readdir_offset == 2 &&
1609 !(rinfo->hash_order && last_hash)) {
1610 /* note dir version at start of readdir so we can
1611 * tell if any dentries get dropped */
1612 req->r_dir_release_cnt =
1613 atomic64_read(&ci->i_release_count);
1614 req->r_dir_ordered_cnt =
1615 atomic64_read(&ci->i_ordered_count);
1616 req->r_readdir_cache_idx = 0;
1617 }
1618 }
1619
1620 cache_ctl.index = req->r_readdir_cache_idx;
1621 fpos_offset = req->r_readdir_offset;
1622
1623 /* FIXME: release caps/leases if error occurs */
1624 for (i = 0; i < rinfo->dir_nr; i++) {
1625 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1626 struct ceph_vino tvino;
1627
1628 dname.name = rde->name;
1629 dname.len = rde->name_len;
1630 dname.hash = full_name_hash(parent, dname.name, dname.len);
1631
1632 tvino.ino = le64_to_cpu(rde->inode.in->ino);
1633 tvino.snap = le64_to_cpu(rde->inode.in->snapid);
1634
1635 if (rinfo->hash_order) {
1636 u32 hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1637 rde->name, rde->name_len);
1638 hash = ceph_frag_value(hash);
1639 if (hash != last_hash)
1640 fpos_offset = 2;
1641 last_hash = hash;
1642 rde->offset = ceph_make_fpos(hash, fpos_offset++, true);
1643 } else {
1644 rde->offset = ceph_make_fpos(frag, fpos_offset++, false);
1645 }
1646
1647 retry_lookup:
1648 dn = d_lookup(parent, &dname);
1649 dout("d_lookup on parent=%p name=%.*s got %p\n",
1650 parent, dname.len, dname.name, dn);
1651
1652 if (!dn) {
1653 dn = d_alloc(parent, &dname);
1654 dout("d_alloc %p '%.*s' = %p\n", parent,
1655 dname.len, dname.name, dn);
1656 if (!dn) {
1657 dout("d_alloc badness\n");
1658 err = -ENOMEM;
1659 goto out;
1660 }
1661 } else if (d_really_is_positive(dn) &&
1662 (ceph_ino(d_inode(dn)) != tvino.ino ||
1663 ceph_snap(d_inode(dn)) != tvino.snap)) {
1664 struct ceph_dentry_info *di = ceph_dentry(dn);
1665 dout(" dn %p points to wrong inode %p\n",
1666 dn, d_inode(dn));
1667
1668 spin_lock(&dn->d_lock);
1669 if (di->offset > 0 &&
1670 di->lease_shared_gen ==
1671 atomic_read(&ci->i_shared_gen)) {
1672 __ceph_dir_clear_ordered(ci);
1673 di->offset = 0;
1674 }
1675 spin_unlock(&dn->d_lock);
1676
1677 d_delete(dn);
1678 dput(dn);
1679 goto retry_lookup;
1680 }
1681
1682 /* inode */
1683 if (d_really_is_positive(dn)) {
1684 in = d_inode(dn);
1685 } else {
1686 in = ceph_get_inode(parent->d_sb, tvino);
1687 if (IS_ERR(in)) {
1688 dout("new_inode badness\n");
1689 d_drop(dn);
1690 dput(dn);
1691 err = PTR_ERR(in);
1692 goto out;
1693 }
1694 }
1695
1696 ret = fill_inode(in, NULL, &rde->inode, NULL, session,
1697 req->r_request_started, -1,
1698 &req->r_caps_reservation);
1699 if (ret < 0) {
1700 pr_err("fill_inode badness on %p\n", in);
1701 if (d_really_is_negative(dn)) {
1702 /* avoid calling iput_final() in mds
1703 * dispatch threads */
1704 ceph_async_iput(in);
1705 }
1706 d_drop(dn);
1707 err = ret;
1708 goto next_item;
1709 }
1710
1711 if (d_really_is_negative(dn)) {
1712 if (ceph_security_xattr_deadlock(in)) {
1713 dout(" skip splicing dn %p to inode %p"
1714 " (security xattr deadlock)\n", dn, in);
1715 ceph_async_iput(in);
1716 skipped++;
1717 goto next_item;
1718 }
1719
1720 err = splice_dentry(&dn, in);
1721 if (err < 0)
1722 goto next_item;
1723 }
1724
1725 ceph_dentry(dn)->offset = rde->offset;
1726
1727 update_dentry_lease(d_inode(parent), dn,
1728 rde->lease, req->r_session,
1729 req->r_request_started);
1730
1731 if (err == 0 && skipped == 0 && cache_ctl.index >= 0) {
1732 ret = fill_readdir_cache(d_inode(parent), dn,
1733 &cache_ctl, req);
1734 if (ret < 0)
1735 err = ret;
1736 }
1737 next_item:
1738 dput(dn);
1739 }
1740 out:
1741 if (err == 0 && skipped == 0) {
1742 set_bit(CEPH_MDS_R_DID_PREPOPULATE, &req->r_req_flags);
1743 req->r_readdir_cache_idx = cache_ctl.index;
1744 }
1745 ceph_readdir_cache_release(&cache_ctl);
1746 dout("readdir_prepopulate done\n");
1747 return err;
1748 }
1749
1750 bool ceph_inode_set_size(struct inode *inode, loff_t size)
1751 {
1752 struct ceph_inode_info *ci = ceph_inode(inode);
1753 bool ret;
1754
1755 spin_lock(&ci->i_ceph_lock);
1756 dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size);
1757 i_size_write(inode, size);
1758 inode->i_blocks = calc_inode_blocks(size);
1759
1760 ret = __ceph_should_report_size(ci);
1761
1762 spin_unlock(&ci->i_ceph_lock);
1763 return ret;
1764 }
1765
1766 /*
1767 * Put reference to inode, but avoid calling iput_final() in current thread.
1768 * iput_final() may wait for reahahead pages. The wait can cause deadlock in
1769 * some contexts.
1770 */
1771 void ceph_async_iput(struct inode *inode)
1772 {
1773 if (!inode)
1774 return;
1775 for (;;) {
1776 if (atomic_add_unless(&inode->i_count, -1, 1))
1777 break;
1778 if (queue_work(ceph_inode_to_client(inode)->inode_wq,
1779 &ceph_inode(inode)->i_work))
1780 break;
1781 /* queue work failed, i_count must be at least 2 */
1782 }
1783 }
1784
1785 /*
1786 * Write back inode data in a worker thread. (This can't be done
1787 * in the message handler context.)
1788 */
1789 void ceph_queue_writeback(struct inode *inode)
1790 {
1791 struct ceph_inode_info *ci = ceph_inode(inode);
1792 set_bit(CEPH_I_WORK_WRITEBACK, &ci->i_work_mask);
1793
1794 ihold(inode);
1795 if (queue_work(ceph_inode_to_client(inode)->inode_wq,
1796 &ci->i_work)) {
1797 dout("ceph_queue_writeback %p\n", inode);
1798 } else {
1799 dout("ceph_queue_writeback %p already queued, mask=%lx\n",
1800 inode, ci->i_work_mask);
1801 iput(inode);
1802 }
1803 }
1804
1805 /*
1806 * queue an async invalidation
1807 */
1808 void ceph_queue_invalidate(struct inode *inode)
1809 {
1810 struct ceph_inode_info *ci = ceph_inode(inode);
1811 set_bit(CEPH_I_WORK_INVALIDATE_PAGES, &ci->i_work_mask);
1812
1813 ihold(inode);
1814 if (queue_work(ceph_inode_to_client(inode)->inode_wq,
1815 &ceph_inode(inode)->i_work)) {
1816 dout("ceph_queue_invalidate %p\n", inode);
1817 } else {
1818 dout("ceph_queue_invalidate %p already queued, mask=%lx\n",
1819 inode, ci->i_work_mask);
1820 iput(inode);
1821 }
1822 }
1823
1824 /*
1825 * Queue an async vmtruncate. If we fail to queue work, we will handle
1826 * the truncation the next time we call __ceph_do_pending_vmtruncate.
1827 */
1828 void ceph_queue_vmtruncate(struct inode *inode)
1829 {
1830 struct ceph_inode_info *ci = ceph_inode(inode);
1831 set_bit(CEPH_I_WORK_VMTRUNCATE, &ci->i_work_mask);
1832
1833 ihold(inode);
1834 if (queue_work(ceph_inode_to_client(inode)->inode_wq,
1835 &ci->i_work)) {
1836 dout("ceph_queue_vmtruncate %p\n", inode);
1837 } else {
1838 dout("ceph_queue_vmtruncate %p already queued, mask=%lx\n",
1839 inode, ci->i_work_mask);
1840 iput(inode);
1841 }
1842 }
1843
1844 static void ceph_do_invalidate_pages(struct inode *inode)
1845 {
1846 struct ceph_inode_info *ci = ceph_inode(inode);
1847 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1848 u32 orig_gen;
1849 int check = 0;
1850
1851 mutex_lock(&ci->i_truncate_mutex);
1852
1853 if (READ_ONCE(fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) {
1854 pr_warn_ratelimited("invalidate_pages %p %lld forced umount\n",
1855 inode, ceph_ino(inode));
1856 mapping_set_error(inode->i_mapping, -EIO);
1857 truncate_pagecache(inode, 0);
1858 mutex_unlock(&ci->i_truncate_mutex);
1859 goto out;
1860 }
1861
1862 spin_lock(&ci->i_ceph_lock);
1863 dout("invalidate_pages %p gen %d revoking %d\n", inode,
1864 ci->i_rdcache_gen, ci->i_rdcache_revoking);
1865 if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
1866 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1867 check = 1;
1868 spin_unlock(&ci->i_ceph_lock);
1869 mutex_unlock(&ci->i_truncate_mutex);
1870 goto out;
1871 }
1872 orig_gen = ci->i_rdcache_gen;
1873 spin_unlock(&ci->i_ceph_lock);
1874
1875 if (invalidate_inode_pages2(inode->i_mapping) < 0) {
1876 pr_err("invalidate_pages %p fails\n", inode);
1877 }
1878
1879 spin_lock(&ci->i_ceph_lock);
1880 if (orig_gen == ci->i_rdcache_gen &&
1881 orig_gen == ci->i_rdcache_revoking) {
1882 dout("invalidate_pages %p gen %d successful\n", inode,
1883 ci->i_rdcache_gen);
1884 ci->i_rdcache_revoking--;
1885 check = 1;
1886 } else {
1887 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1888 inode, orig_gen, ci->i_rdcache_gen,
1889 ci->i_rdcache_revoking);
1890 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1891 check = 1;
1892 }
1893 spin_unlock(&ci->i_ceph_lock);
1894 mutex_unlock(&ci->i_truncate_mutex);
1895 out:
1896 if (check)
1897 ceph_check_caps(ci, 0, NULL);
1898 }
1899
1900 /*
1901 * Make sure any pending truncation is applied before doing anything
1902 * that may depend on it.
1903 */
1904 void __ceph_do_pending_vmtruncate(struct inode *inode)
1905 {
1906 struct ceph_inode_info *ci = ceph_inode(inode);
1907 u64 to;
1908 int wrbuffer_refs, finish = 0;
1909
1910 mutex_lock(&ci->i_truncate_mutex);
1911 retry:
1912 spin_lock(&ci->i_ceph_lock);
1913 if (ci->i_truncate_pending == 0) {
1914 dout("__do_pending_vmtruncate %p none pending\n", inode);
1915 spin_unlock(&ci->i_ceph_lock);
1916 mutex_unlock(&ci->i_truncate_mutex);
1917 return;
1918 }
1919
1920 /*
1921 * make sure any dirty snapped pages are flushed before we
1922 * possibly truncate them.. so write AND block!
1923 */
1924 if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) {
1925 spin_unlock(&ci->i_ceph_lock);
1926 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1927 inode);
1928 filemap_write_and_wait_range(&inode->i_data, 0,
1929 inode->i_sb->s_maxbytes);
1930 goto retry;
1931 }
1932
1933 /* there should be no reader or writer */
1934 WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref);
1935
1936 to = ci->i_truncate_size;
1937 wrbuffer_refs = ci->i_wrbuffer_ref;
1938 dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode,
1939 ci->i_truncate_pending, to);
1940 spin_unlock(&ci->i_ceph_lock);
1941
1942 truncate_pagecache(inode, to);
1943
1944 spin_lock(&ci->i_ceph_lock);
1945 if (to == ci->i_truncate_size) {
1946 ci->i_truncate_pending = 0;
1947 finish = 1;
1948 }
1949 spin_unlock(&ci->i_ceph_lock);
1950 if (!finish)
1951 goto retry;
1952
1953 mutex_unlock(&ci->i_truncate_mutex);
1954
1955 if (wrbuffer_refs == 0)
1956 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
1957
1958 wake_up_all(&ci->i_cap_wq);
1959 }
1960
1961 static void ceph_inode_work(struct work_struct *work)
1962 {
1963 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1964 i_work);
1965 struct inode *inode = &ci->vfs_inode;
1966
1967 if (test_and_clear_bit(CEPH_I_WORK_WRITEBACK, &ci->i_work_mask)) {
1968 dout("writeback %p\n", inode);
1969 filemap_fdatawrite(&inode->i_data);
1970 }
1971 if (test_and_clear_bit(CEPH_I_WORK_INVALIDATE_PAGES, &ci->i_work_mask))
1972 ceph_do_invalidate_pages(inode);
1973
1974 if (test_and_clear_bit(CEPH_I_WORK_VMTRUNCATE, &ci->i_work_mask))
1975 __ceph_do_pending_vmtruncate(inode);
1976
1977 iput(inode);
1978 }
1979
1980 /*
1981 * symlinks
1982 */
1983 static const struct inode_operations ceph_symlink_iops = {
1984 .get_link = simple_get_link,
1985 .setattr = ceph_setattr,
1986 .getattr = ceph_getattr,
1987 .listxattr = ceph_listxattr,
1988 };
1989
1990 int __ceph_setattr(struct inode *inode, struct iattr *attr)
1991 {
1992 struct ceph_inode_info *ci = ceph_inode(inode);
1993 unsigned int ia_valid = attr->ia_valid;
1994 struct ceph_mds_request *req;
1995 struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
1996 struct ceph_cap_flush *prealloc_cf;
1997 int issued;
1998 int release = 0, dirtied = 0;
1999 int mask = 0;
2000 int err = 0;
2001 int inode_dirty_flags = 0;
2002 bool lock_snap_rwsem = false;
2003
2004 prealloc_cf = ceph_alloc_cap_flush();
2005 if (!prealloc_cf)
2006 return -ENOMEM;
2007
2008 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR,
2009 USE_AUTH_MDS);
2010 if (IS_ERR(req)) {
2011 ceph_free_cap_flush(prealloc_cf);
2012 return PTR_ERR(req);
2013 }
2014
2015 spin_lock(&ci->i_ceph_lock);
2016 issued = __ceph_caps_issued(ci, NULL);
2017
2018 if (!ci->i_head_snapc &&
2019 (issued & (CEPH_CAP_ANY_EXCL | CEPH_CAP_FILE_WR))) {
2020 lock_snap_rwsem = true;
2021 if (!down_read_trylock(&mdsc->snap_rwsem)) {
2022 spin_unlock(&ci->i_ceph_lock);
2023 down_read(&mdsc->snap_rwsem);
2024 spin_lock(&ci->i_ceph_lock);
2025 issued = __ceph_caps_issued(ci, NULL);
2026 }
2027 }
2028
2029 dout("setattr %p issued %s\n", inode, ceph_cap_string(issued));
2030
2031 if (ia_valid & ATTR_UID) {
2032 dout("setattr %p uid %d -> %d\n", inode,
2033 from_kuid(&init_user_ns, inode->i_uid),
2034 from_kuid(&init_user_ns, attr->ia_uid));
2035 if (issued & CEPH_CAP_AUTH_EXCL) {
2036 inode->i_uid = attr->ia_uid;
2037 dirtied |= CEPH_CAP_AUTH_EXCL;
2038 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2039 !uid_eq(attr->ia_uid, inode->i_uid)) {
2040 req->r_args.setattr.uid = cpu_to_le32(
2041 from_kuid(&init_user_ns, attr->ia_uid));
2042 mask |= CEPH_SETATTR_UID;
2043 release |= CEPH_CAP_AUTH_SHARED;
2044 }
2045 }
2046 if (ia_valid & ATTR_GID) {
2047 dout("setattr %p gid %d -> %d\n", inode,
2048 from_kgid(&init_user_ns, inode->i_gid),
2049 from_kgid(&init_user_ns, attr->ia_gid));
2050 if (issued & CEPH_CAP_AUTH_EXCL) {
2051 inode->i_gid = attr->ia_gid;
2052 dirtied |= CEPH_CAP_AUTH_EXCL;
2053 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2054 !gid_eq(attr->ia_gid, inode->i_gid)) {
2055 req->r_args.setattr.gid = cpu_to_le32(
2056 from_kgid(&init_user_ns, attr->ia_gid));
2057 mask |= CEPH_SETATTR_GID;
2058 release |= CEPH_CAP_AUTH_SHARED;
2059 }
2060 }
2061 if (ia_valid & ATTR_MODE) {
2062 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode,
2063 attr->ia_mode);
2064 if (issued & CEPH_CAP_AUTH_EXCL) {
2065 inode->i_mode = attr->ia_mode;
2066 dirtied |= CEPH_CAP_AUTH_EXCL;
2067 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2068 attr->ia_mode != inode->i_mode) {
2069 inode->i_mode = attr->ia_mode;
2070 req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode);
2071 mask |= CEPH_SETATTR_MODE;
2072 release |= CEPH_CAP_AUTH_SHARED;
2073 }
2074 }
2075
2076 if (ia_valid & ATTR_ATIME) {
2077 dout("setattr %p atime %lld.%ld -> %lld.%ld\n", inode,
2078 inode->i_atime.tv_sec, inode->i_atime.tv_nsec,
2079 attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec);
2080 if (issued & CEPH_CAP_FILE_EXCL) {
2081 ci->i_time_warp_seq++;
2082 inode->i_atime = attr->ia_atime;
2083 dirtied |= CEPH_CAP_FILE_EXCL;
2084 } else if ((issued & CEPH_CAP_FILE_WR) &&
2085 timespec64_compare(&inode->i_atime,
2086 &attr->ia_atime) < 0) {
2087 inode->i_atime = attr->ia_atime;
2088 dirtied |= CEPH_CAP_FILE_WR;
2089 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2090 !timespec64_equal(&inode->i_atime, &attr->ia_atime)) {
2091 ceph_encode_timespec64(&req->r_args.setattr.atime,
2092 &attr->ia_atime);
2093 mask |= CEPH_SETATTR_ATIME;
2094 release |= CEPH_CAP_FILE_SHARED |
2095 CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2096 }
2097 }
2098 if (ia_valid & ATTR_SIZE) {
2099 dout("setattr %p size %lld -> %lld\n", inode,
2100 inode->i_size, attr->ia_size);
2101 if ((issued & CEPH_CAP_FILE_EXCL) &&
2102 attr->ia_size > inode->i_size) {
2103 i_size_write(inode, attr->ia_size);
2104 inode->i_blocks = calc_inode_blocks(attr->ia_size);
2105 ci->i_reported_size = attr->ia_size;
2106 dirtied |= CEPH_CAP_FILE_EXCL;
2107 ia_valid |= ATTR_MTIME;
2108 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2109 attr->ia_size != inode->i_size) {
2110 req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
2111 req->r_args.setattr.old_size =
2112 cpu_to_le64(inode->i_size);
2113 mask |= CEPH_SETATTR_SIZE;
2114 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_EXCL |
2115 CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2116 }
2117 }
2118 if (ia_valid & ATTR_MTIME) {
2119 dout("setattr %p mtime %lld.%ld -> %lld.%ld\n", inode,
2120 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
2121 attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec);
2122 if (issued & CEPH_CAP_FILE_EXCL) {
2123 ci->i_time_warp_seq++;
2124 inode->i_mtime = attr->ia_mtime;
2125 dirtied |= CEPH_CAP_FILE_EXCL;
2126 } else if ((issued & CEPH_CAP_FILE_WR) &&
2127 timespec64_compare(&inode->i_mtime,
2128 &attr->ia_mtime) < 0) {
2129 inode->i_mtime = attr->ia_mtime;
2130 dirtied |= CEPH_CAP_FILE_WR;
2131 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2132 !timespec64_equal(&inode->i_mtime, &attr->ia_mtime)) {
2133 ceph_encode_timespec64(&req->r_args.setattr.mtime,
2134 &attr->ia_mtime);
2135 mask |= CEPH_SETATTR_MTIME;
2136 release |= CEPH_CAP_FILE_SHARED |
2137 CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2138 }
2139 }
2140
2141 /* these do nothing */
2142 if (ia_valid & ATTR_CTIME) {
2143 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME|
2144 ATTR_MODE|ATTR_UID|ATTR_GID)) == 0;
2145 dout("setattr %p ctime %lld.%ld -> %lld.%ld (%s)\n", inode,
2146 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
2147 attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec,
2148 only ? "ctime only" : "ignored");
2149 if (only) {
2150 /*
2151 * if kernel wants to dirty ctime but nothing else,
2152 * we need to choose a cap to dirty under, or do
2153 * a almost-no-op setattr
2154 */
2155 if (issued & CEPH_CAP_AUTH_EXCL)
2156 dirtied |= CEPH_CAP_AUTH_EXCL;
2157 else if (issued & CEPH_CAP_FILE_EXCL)
2158 dirtied |= CEPH_CAP_FILE_EXCL;
2159 else if (issued & CEPH_CAP_XATTR_EXCL)
2160 dirtied |= CEPH_CAP_XATTR_EXCL;
2161 else
2162 mask |= CEPH_SETATTR_CTIME;
2163 }
2164 }
2165 if (ia_valid & ATTR_FILE)
2166 dout("setattr %p ATTR_FILE ... hrm!\n", inode);
2167
2168 if (dirtied) {
2169 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied,
2170 &prealloc_cf);
2171 inode->i_ctime = attr->ia_ctime;
2172 }
2173
2174 release &= issued;
2175 spin_unlock(&ci->i_ceph_lock);
2176 if (lock_snap_rwsem)
2177 up_read(&mdsc->snap_rwsem);
2178
2179 if (inode_dirty_flags)
2180 __mark_inode_dirty(inode, inode_dirty_flags);
2181
2182
2183 if (mask) {
2184 req->r_inode = inode;
2185 ihold(inode);
2186 req->r_inode_drop = release;
2187 req->r_args.setattr.mask = cpu_to_le32(mask);
2188 req->r_num_caps = 1;
2189 req->r_stamp = attr->ia_ctime;
2190 err = ceph_mdsc_do_request(mdsc, NULL, req);
2191 }
2192 dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err,
2193 ceph_cap_string(dirtied), mask);
2194
2195 ceph_mdsc_put_request(req);
2196 ceph_free_cap_flush(prealloc_cf);
2197
2198 if (err >= 0 && (mask & CEPH_SETATTR_SIZE))
2199 __ceph_do_pending_vmtruncate(inode);
2200
2201 return err;
2202 }
2203
2204 /*
2205 * setattr
2206 */
2207 int ceph_setattr(struct dentry *dentry, struct iattr *attr)
2208 {
2209 struct inode *inode = d_inode(dentry);
2210 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
2211 int err;
2212
2213 if (ceph_snap(inode) != CEPH_NOSNAP)
2214 return -EROFS;
2215
2216 err = setattr_prepare(dentry, attr);
2217 if (err != 0)
2218 return err;
2219
2220 if ((attr->ia_valid & ATTR_SIZE) &&
2221 attr->ia_size > max(inode->i_size, fsc->max_file_size))
2222 return -EFBIG;
2223
2224 if ((attr->ia_valid & ATTR_SIZE) &&
2225 ceph_quota_is_max_bytes_exceeded(inode, attr->ia_size))
2226 return -EDQUOT;
2227
2228 err = __ceph_setattr(inode, attr);
2229
2230 if (err >= 0 && (attr->ia_valid & ATTR_MODE))
2231 err = posix_acl_chmod(inode, attr->ia_mode);
2232
2233 return err;
2234 }
2235
2236 /*
2237 * Verify that we have a lease on the given mask. If not,
2238 * do a getattr against an mds.
2239 */
2240 int __ceph_do_getattr(struct inode *inode, struct page *locked_page,
2241 int mask, bool force)
2242 {
2243 struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
2244 struct ceph_mds_client *mdsc = fsc->mdsc;
2245 struct ceph_mds_request *req;
2246 int mode;
2247 int err;
2248
2249 if (ceph_snap(inode) == CEPH_SNAPDIR) {
2250 dout("do_getattr inode %p SNAPDIR\n", inode);
2251 return 0;
2252 }
2253
2254 dout("do_getattr inode %p mask %s mode 0%o\n",
2255 inode, ceph_cap_string(mask), inode->i_mode);
2256 if (!force && ceph_caps_issued_mask(ceph_inode(inode), mask, 1))
2257 return 0;
2258
2259 mode = (mask & CEPH_STAT_RSTAT) ? USE_AUTH_MDS : USE_ANY_MDS;
2260 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, mode);
2261 if (IS_ERR(req))
2262 return PTR_ERR(req);
2263 req->r_inode = inode;
2264 ihold(inode);
2265 req->r_num_caps = 1;
2266 req->r_args.getattr.mask = cpu_to_le32(mask);
2267 req->r_locked_page = locked_page;
2268 err = ceph_mdsc_do_request(mdsc, NULL, req);
2269 if (locked_page && err == 0) {
2270 u64 inline_version = req->r_reply_info.targeti.inline_version;
2271 if (inline_version == 0) {
2272 /* the reply is supposed to contain inline data */
2273 err = -EINVAL;
2274 } else if (inline_version == CEPH_INLINE_NONE) {
2275 err = -ENODATA;
2276 } else {
2277 err = req->r_reply_info.targeti.inline_len;
2278 }
2279 }
2280 ceph_mdsc_put_request(req);
2281 dout("do_getattr result=%d\n", err);
2282 return err;
2283 }
2284
2285
2286 /*
2287 * Check inode permissions. We verify we have a valid value for
2288 * the AUTH cap, then call the generic handler.
2289 */
2290 int ceph_permission(struct inode *inode, int mask)
2291 {
2292 int err;
2293
2294 if (mask & MAY_NOT_BLOCK)
2295 return -ECHILD;
2296
2297 err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED, false);
2298
2299 if (!err)
2300 err = generic_permission(inode, mask);
2301 return err;
2302 }
2303
2304 /* Craft a mask of needed caps given a set of requested statx attrs. */
2305 static int statx_to_caps(u32 want)
2306 {
2307 int mask = 0;
2308
2309 if (want & (STATX_MODE|STATX_UID|STATX_GID|STATX_CTIME|STATX_BTIME))
2310 mask |= CEPH_CAP_AUTH_SHARED;
2311
2312 if (want & (STATX_NLINK|STATX_CTIME))
2313 mask |= CEPH_CAP_LINK_SHARED;
2314
2315 if (want & (STATX_ATIME|STATX_MTIME|STATX_CTIME|STATX_SIZE|
2316 STATX_BLOCKS))
2317 mask |= CEPH_CAP_FILE_SHARED;
2318
2319 if (want & (STATX_CTIME))
2320 mask |= CEPH_CAP_XATTR_SHARED;
2321
2322 return mask;
2323 }
2324
2325 /*
2326 * Get all the attributes. If we have sufficient caps for the requested attrs,
2327 * then we can avoid talking to the MDS at all.
2328 */
2329 int ceph_getattr(const struct path *path, struct kstat *stat,
2330 u32 request_mask, unsigned int flags)
2331 {
2332 struct inode *inode = d_inode(path->dentry);
2333 struct ceph_inode_info *ci = ceph_inode(inode);
2334 u32 valid_mask = STATX_BASIC_STATS;
2335 int err = 0;
2336
2337 /* Skip the getattr altogether if we're asked not to sync */
2338 if (!(flags & AT_STATX_DONT_SYNC)) {
2339 err = ceph_do_getattr(inode, statx_to_caps(request_mask),
2340 flags & AT_STATX_FORCE_SYNC);
2341 if (err)
2342 return err;
2343 }
2344
2345 generic_fillattr(inode, stat);
2346 stat->ino = ceph_translate_ino(inode->i_sb, inode->i_ino);
2347
2348 /*
2349 * btime on newly-allocated inodes is 0, so if this is still set to
2350 * that, then assume that it's not valid.
2351 */
2352 if (ci->i_btime.tv_sec || ci->i_btime.tv_nsec) {
2353 stat->btime = ci->i_btime;
2354 valid_mask |= STATX_BTIME;
2355 }
2356
2357 if (ceph_snap(inode) == CEPH_NOSNAP)
2358 stat->dev = inode->i_sb->s_dev;
2359 else
2360 stat->dev = ci->i_snapid_map ? ci->i_snapid_map->dev : 0;
2361
2362 if (S_ISDIR(inode->i_mode)) {
2363 if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb),
2364 RBYTES))
2365 stat->size = ci->i_rbytes;
2366 else
2367 stat->size = ci->i_files + ci->i_subdirs;
2368 stat->blocks = 0;
2369 stat->blksize = 65536;
2370 /*
2371 * Some applications rely on the number of st_nlink
2372 * value on directories to be either 0 (if unlinked)
2373 * or 2 + number of subdirectories.
2374 */
2375 if (stat->nlink == 1)
2376 /* '.' + '..' + subdirs */
2377 stat->nlink = 1 + 1 + ci->i_subdirs;
2378 }
2379
2380 stat->result_mask = request_mask & valid_mask;
2381 return err;
2382 }
Linux with added FPC-III support