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


4caeeae392e195086074cb03c1eba73b79d59ce6
[mirror_ubuntu-bionic-kernel.git] / fs / ceph / inode.c
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
17 #include "super.h"
18 #include "mds_client.h"
19 #include "cache.h"
20 #include <linux/ceph/decode.h>
21
22 /*
23 * Ceph inode operations
24 *
25 * Implement basic inode helpers (get, alloc) and inode ops (getattr,
26 * setattr, etc.), xattr helpers, and helpers for assimilating
27 * metadata returned by the MDS into our cache.
28 *
29 * Also define helpers for doing asynchronous writeback, invalidation,
30 * and truncation for the benefit of those who can't afford to block
31 * (typically because they are in the message handler path).
32 */
33
34 static const struct inode_operations ceph_symlink_iops;
35
36 static void ceph_inode_work(struct work_struct *work);
37
38 /*
39 * find or create an inode, given the ceph ino number
40 */
41 static int ceph_set_ino_cb(struct inode *inode, void *data)
42 {
43 ceph_inode(inode)->i_vino = *(struct ceph_vino *)data;
44 inode->i_ino = ceph_vino_to_ino(*(struct ceph_vino *)data);
45 return 0;
46 }
47
48 struct inode *ceph_get_inode(struct super_block *sb, struct ceph_vino vino)
49 {
50 struct inode *inode;
51 ino_t t = ceph_vino_to_ino(vino);
52
53 inode = iget5_locked(sb, t, ceph_ino_compare, ceph_set_ino_cb, &vino);
54 if (!inode)
55 return ERR_PTR(-ENOMEM);
56 if (inode->i_state & I_NEW) {
57 dout("get_inode created new inode %p %llx.%llx ino %llx\n",
58 inode, ceph_vinop(inode), (u64)inode->i_ino);
59 unlock_new_inode(inode);
60 }
61
62 dout("get_inode on %lu=%llx.%llx got %p\n", inode->i_ino, vino.ino,
63 vino.snap, inode);
64 return inode;
65 }
66
67 /*
68 * get/constuct snapdir inode for a given directory
69 */
70 struct inode *ceph_get_snapdir(struct inode *parent)
71 {
72 struct ceph_vino vino = {
73 .ino = ceph_ino(parent),
74 .snap = CEPH_SNAPDIR,
75 };
76 struct inode *inode = ceph_get_inode(parent->i_sb, vino);
77 struct ceph_inode_info *ci = ceph_inode(inode);
78
79 BUG_ON(!S_ISDIR(parent->i_mode));
80 if (IS_ERR(inode))
81 return inode;
82 inode->i_mode = parent->i_mode;
83 inode->i_uid = parent->i_uid;
84 inode->i_gid = parent->i_gid;
85 inode->i_op = &ceph_snapdir_iops;
86 inode->i_fop = &ceph_snapdir_fops;
87 ci->i_snap_caps = CEPH_CAP_PIN; /* so we can open */
88 ci->i_rbytes = 0;
89 return inode;
90 }
91
92 const struct inode_operations ceph_file_iops = {
93 .permission = ceph_permission,
94 .setattr = ceph_setattr,
95 .getattr = ceph_getattr,
96 .listxattr = ceph_listxattr,
97 .get_acl = ceph_get_acl,
98 .set_acl = ceph_set_acl,
99 };
100
101
102 /*
103 * We use a 'frag tree' to keep track of the MDS's directory fragments
104 * for a given inode (usually there is just a single fragment). We
105 * need to know when a child frag is delegated to a new MDS, or when
106 * it is flagged as replicated, so we can direct our requests
107 * accordingly.
108 */
109
110 /*
111 * find/create a frag in the tree
112 */
113 static struct ceph_inode_frag *__get_or_create_frag(struct ceph_inode_info *ci,
114 u32 f)
115 {
116 struct rb_node **p;
117 struct rb_node *parent = NULL;
118 struct ceph_inode_frag *frag;
119 int c;
120
121 p = &ci->i_fragtree.rb_node;
122 while (*p) {
123 parent = *p;
124 frag = rb_entry(parent, struct ceph_inode_frag, node);
125 c = ceph_frag_compare(f, frag->frag);
126 if (c < 0)
127 p = &(*p)->rb_left;
128 else if (c > 0)
129 p = &(*p)->rb_right;
130 else
131 return frag;
132 }
133
134 frag = kmalloc(sizeof(*frag), GFP_NOFS);
135 if (!frag)
136 return ERR_PTR(-ENOMEM);
137
138 frag->frag = f;
139 frag->split_by = 0;
140 frag->mds = -1;
141 frag->ndist = 0;
142
143 rb_link_node(&frag->node, parent, p);
144 rb_insert_color(&frag->node, &ci->i_fragtree);
145
146 dout("get_or_create_frag added %llx.%llx frag %x\n",
147 ceph_vinop(&ci->vfs_inode), f);
148 return frag;
149 }
150
151 /*
152 * find a specific frag @f
153 */
154 struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci, u32 f)
155 {
156 struct rb_node *n = ci->i_fragtree.rb_node;
157
158 while (n) {
159 struct ceph_inode_frag *frag =
160 rb_entry(n, struct ceph_inode_frag, node);
161 int c = ceph_frag_compare(f, frag->frag);
162 if (c < 0)
163 n = n->rb_left;
164 else if (c > 0)
165 n = n->rb_right;
166 else
167 return frag;
168 }
169 return NULL;
170 }
171
172 /*
173 * Choose frag containing the given value @v. If @pfrag is
174 * specified, copy the frag delegation info to the caller if
175 * it is present.
176 */
177 static u32 __ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
178 struct ceph_inode_frag *pfrag, int *found)
179 {
180 u32 t = ceph_frag_make(0, 0);
181 struct ceph_inode_frag *frag;
182 unsigned nway, i;
183 u32 n;
184
185 if (found)
186 *found = 0;
187
188 while (1) {
189 WARN_ON(!ceph_frag_contains_value(t, v));
190 frag = __ceph_find_frag(ci, t);
191 if (!frag)
192 break; /* t is a leaf */
193 if (frag->split_by == 0) {
194 if (pfrag)
195 memcpy(pfrag, frag, sizeof(*pfrag));
196 if (found)
197 *found = 1;
198 break;
199 }
200
201 /* choose child */
202 nway = 1 << frag->split_by;
203 dout("choose_frag(%x) %x splits by %d (%d ways)\n", v, t,
204 frag->split_by, nway);
205 for (i = 0; i < nway; i++) {
206 n = ceph_frag_make_child(t, frag->split_by, i);
207 if (ceph_frag_contains_value(n, v)) {
208 t = n;
209 break;
210 }
211 }
212 BUG_ON(i == nway);
213 }
214 dout("choose_frag(%x) = %x\n", v, t);
215
216 return t;
217 }
218
219 u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
220 struct ceph_inode_frag *pfrag, int *found)
221 {
222 u32 ret;
223 mutex_lock(&ci->i_fragtree_mutex);
224 ret = __ceph_choose_frag(ci, v, pfrag, found);
225 mutex_unlock(&ci->i_fragtree_mutex);
226 return ret;
227 }
228
229 /*
230 * Process dirfrag (delegation) info from the mds. Include leaf
231 * fragment in tree ONLY if ndist > 0. Otherwise, only
232 * branches/splits are included in i_fragtree)
233 */
234 static int ceph_fill_dirfrag(struct inode *inode,
235 struct ceph_mds_reply_dirfrag *dirinfo)
236 {
237 struct ceph_inode_info *ci = ceph_inode(inode);
238 struct ceph_inode_frag *frag;
239 u32 id = le32_to_cpu(dirinfo->frag);
240 int mds = le32_to_cpu(dirinfo->auth);
241 int ndist = le32_to_cpu(dirinfo->ndist);
242 int diri_auth = -1;
243 int i;
244 int err = 0;
245
246 spin_lock(&ci->i_ceph_lock);
247 if (ci->i_auth_cap)
248 diri_auth = ci->i_auth_cap->mds;
249 spin_unlock(&ci->i_ceph_lock);
250
251 if (mds == -1) /* CDIR_AUTH_PARENT */
252 mds = diri_auth;
253
254 mutex_lock(&ci->i_fragtree_mutex);
255 if (ndist == 0 && mds == diri_auth) {
256 /* no delegation info needed. */
257 frag = __ceph_find_frag(ci, id);
258 if (!frag)
259 goto out;
260 if (frag->split_by == 0) {
261 /* tree leaf, remove */
262 dout("fill_dirfrag removed %llx.%llx frag %x"
263 " (no ref)\n", ceph_vinop(inode), id);
264 rb_erase(&frag->node, &ci->i_fragtree);
265 kfree(frag);
266 } else {
267 /* tree branch, keep and clear */
268 dout("fill_dirfrag cleared %llx.%llx frag %x"
269 " referral\n", ceph_vinop(inode), id);
270 frag->mds = -1;
271 frag->ndist = 0;
272 }
273 goto out;
274 }
275
276
277 /* find/add this frag to store mds delegation info */
278 frag = __get_or_create_frag(ci, id);
279 if (IS_ERR(frag)) {
280 /* this is not the end of the world; we can continue
281 with bad/inaccurate delegation info */
282 pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n",
283 ceph_vinop(inode), le32_to_cpu(dirinfo->frag));
284 err = -ENOMEM;
285 goto out;
286 }
287
288 frag->mds = mds;
289 frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP);
290 for (i = 0; i < frag->ndist; i++)
291 frag->dist[i] = le32_to_cpu(dirinfo->dist[i]);
292 dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n",
293 ceph_vinop(inode), frag->frag, frag->ndist);
294
295 out:
296 mutex_unlock(&ci->i_fragtree_mutex);
297 return err;
298 }
299
300 static int frag_tree_split_cmp(const void *l, const void *r)
301 {
302 struct ceph_frag_tree_split *ls = (struct ceph_frag_tree_split*)l;
303 struct ceph_frag_tree_split *rs = (struct ceph_frag_tree_split*)r;
304 return ceph_frag_compare(le32_to_cpu(ls->frag),
305 le32_to_cpu(rs->frag));
306 }
307
308 static bool is_frag_child(u32 f, struct ceph_inode_frag *frag)
309 {
310 if (!frag)
311 return f == ceph_frag_make(0, 0);
312 if (ceph_frag_bits(f) != ceph_frag_bits(frag->frag) + frag->split_by)
313 return false;
314 return ceph_frag_contains_value(frag->frag, ceph_frag_value(f));
315 }
316
317 static int ceph_fill_fragtree(struct inode *inode,
318 struct ceph_frag_tree_head *fragtree,
319 struct ceph_mds_reply_dirfrag *dirinfo)
320 {
321 struct ceph_inode_info *ci = ceph_inode(inode);
322 struct ceph_inode_frag *frag, *prev_frag = NULL;
323 struct rb_node *rb_node;
324 unsigned i, split_by, nsplits;
325 u32 id;
326 bool update = false;
327
328 mutex_lock(&ci->i_fragtree_mutex);
329 nsplits = le32_to_cpu(fragtree->nsplits);
330 if (nsplits != ci->i_fragtree_nsplits) {
331 update = true;
332 } else if (nsplits) {
333 i = prandom_u32() % nsplits;
334 id = le32_to_cpu(fragtree->splits[i].frag);
335 if (!__ceph_find_frag(ci, id))
336 update = true;
337 } else if (!RB_EMPTY_ROOT(&ci->i_fragtree)) {
338 rb_node = rb_first(&ci->i_fragtree);
339 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
340 if (frag->frag != ceph_frag_make(0, 0) || rb_next(rb_node))
341 update = true;
342 }
343 if (!update && dirinfo) {
344 id = le32_to_cpu(dirinfo->frag);
345 if (id != __ceph_choose_frag(ci, id, NULL, NULL))
346 update = true;
347 }
348 if (!update)
349 goto out_unlock;
350
351 if (nsplits > 1) {
352 sort(fragtree->splits, nsplits, sizeof(fragtree->splits[0]),
353 frag_tree_split_cmp, NULL);
354 }
355
356 dout("fill_fragtree %llx.%llx\n", ceph_vinop(inode));
357 rb_node = rb_first(&ci->i_fragtree);
358 for (i = 0; i < nsplits; i++) {
359 id = le32_to_cpu(fragtree->splits[i].frag);
360 split_by = le32_to_cpu(fragtree->splits[i].by);
361 if (split_by == 0 || ceph_frag_bits(id) + split_by > 24) {
362 pr_err("fill_fragtree %llx.%llx invalid split %d/%u, "
363 "frag %x split by %d\n", ceph_vinop(inode),
364 i, nsplits, id, split_by);
365 continue;
366 }
367 frag = NULL;
368 while (rb_node) {
369 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
370 if (ceph_frag_compare(frag->frag, id) >= 0) {
371 if (frag->frag != id)
372 frag = NULL;
373 else
374 rb_node = rb_next(rb_node);
375 break;
376 }
377 rb_node = rb_next(rb_node);
378 /* delete stale split/leaf node */
379 if (frag->split_by > 0 ||
380 !is_frag_child(frag->frag, prev_frag)) {
381 rb_erase(&frag->node, &ci->i_fragtree);
382 if (frag->split_by > 0)
383 ci->i_fragtree_nsplits--;
384 kfree(frag);
385 }
386 frag = NULL;
387 }
388 if (!frag) {
389 frag = __get_or_create_frag(ci, id);
390 if (IS_ERR(frag))
391 continue;
392 }
393 if (frag->split_by == 0)
394 ci->i_fragtree_nsplits++;
395 frag->split_by = split_by;
396 dout(" frag %x split by %d\n", frag->frag, frag->split_by);
397 prev_frag = frag;
398 }
399 while (rb_node) {
400 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
401 rb_node = rb_next(rb_node);
402 /* delete stale split/leaf node */
403 if (frag->split_by > 0 ||
404 !is_frag_child(frag->frag, prev_frag)) {
405 rb_erase(&frag->node, &ci->i_fragtree);
406 if (frag->split_by > 0)
407 ci->i_fragtree_nsplits--;
408 kfree(frag);
409 }
410 }
411 out_unlock:
412 mutex_unlock(&ci->i_fragtree_mutex);
413 return 0;
414 }
415
416 /*
417 * initialize a newly allocated inode.
418 */
419 struct inode *ceph_alloc_inode(struct super_block *sb)
420 {
421 struct ceph_inode_info *ci;
422 int i;
423
424 ci = kmem_cache_alloc(ceph_inode_cachep, GFP_NOFS);
425 if (!ci)
426 return NULL;
427
428 dout("alloc_inode %p\n", &ci->vfs_inode);
429
430 spin_lock_init(&ci->i_ceph_lock);
431
432 ci->i_version = 0;
433 ci->i_inline_version = 0;
434 ci->i_time_warp_seq = 0;
435 ci->i_ceph_flags = 0;
436 atomic64_set(&ci->i_ordered_count, 1);
437 atomic64_set(&ci->i_release_count, 1);
438 atomic64_set(&ci->i_complete_seq[0], 0);
439 atomic64_set(&ci->i_complete_seq[1], 0);
440 ci->i_symlink = NULL;
441
442 memset(&ci->i_dir_layout, 0, sizeof(ci->i_dir_layout));
443 RCU_INIT_POINTER(ci->i_layout.pool_ns, NULL);
444
445 ci->i_fragtree = RB_ROOT;
446 mutex_init(&ci->i_fragtree_mutex);
447
448 ci->i_xattrs.blob = NULL;
449 ci->i_xattrs.prealloc_blob = NULL;
450 ci->i_xattrs.dirty = false;
451 ci->i_xattrs.index = RB_ROOT;
452 ci->i_xattrs.count = 0;
453 ci->i_xattrs.names_size = 0;
454 ci->i_xattrs.vals_size = 0;
455 ci->i_xattrs.version = 0;
456 ci->i_xattrs.index_version = 0;
457
458 ci->i_caps = RB_ROOT;
459 ci->i_auth_cap = NULL;
460 ci->i_dirty_caps = 0;
461 ci->i_flushing_caps = 0;
462 INIT_LIST_HEAD(&ci->i_dirty_item);
463 INIT_LIST_HEAD(&ci->i_flushing_item);
464 ci->i_prealloc_cap_flush = NULL;
465 INIT_LIST_HEAD(&ci->i_cap_flush_list);
466 init_waitqueue_head(&ci->i_cap_wq);
467 ci->i_hold_caps_min = 0;
468 ci->i_hold_caps_max = 0;
469 INIT_LIST_HEAD(&ci->i_cap_delay_list);
470 INIT_LIST_HEAD(&ci->i_cap_snaps);
471 ci->i_head_snapc = NULL;
472 ci->i_snap_caps = 0;
473
474 for (i = 0; i < CEPH_FILE_MODE_BITS; i++)
475 ci->i_nr_by_mode[i] = 0;
476
477 mutex_init(&ci->i_truncate_mutex);
478 ci->i_truncate_seq = 0;
479 ci->i_truncate_size = 0;
480 ci->i_truncate_pending = 0;
481
482 ci->i_max_size = 0;
483 ci->i_reported_size = 0;
484 ci->i_wanted_max_size = 0;
485 ci->i_requested_max_size = 0;
486
487 ci->i_pin_ref = 0;
488 ci->i_rd_ref = 0;
489 ci->i_rdcache_ref = 0;
490 ci->i_wr_ref = 0;
491 ci->i_wb_ref = 0;
492 ci->i_wrbuffer_ref = 0;
493 ci->i_wrbuffer_ref_head = 0;
494 atomic_set(&ci->i_filelock_ref, 0);
495 atomic_set(&ci->i_shared_gen, 0);
496 ci->i_rdcache_gen = 0;
497 ci->i_rdcache_revoking = 0;
498
499 INIT_LIST_HEAD(&ci->i_unsafe_dirops);
500 INIT_LIST_HEAD(&ci->i_unsafe_iops);
501 spin_lock_init(&ci->i_unsafe_lock);
502
503 ci->i_snap_realm = NULL;
504 INIT_LIST_HEAD(&ci->i_snap_realm_item);
505 INIT_LIST_HEAD(&ci->i_snap_flush_item);
506
507 INIT_WORK(&ci->i_work, ceph_inode_work);
508 ci->i_work_mask = 0;
509
510 ceph_fscache_inode_init(ci);
511
512 return &ci->vfs_inode;
513 }
514
515 static void ceph_i_callback(struct rcu_head *head)
516 {
517 struct inode *inode = container_of(head, struct inode, i_rcu);
518 struct ceph_inode_info *ci = ceph_inode(inode);
519
520 kmem_cache_free(ceph_inode_cachep, ci);
521 }
522
523 void ceph_destroy_inode(struct inode *inode)
524 {
525 struct ceph_inode_info *ci = ceph_inode(inode);
526 struct ceph_inode_frag *frag;
527 struct rb_node *n;
528
529 dout("destroy_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode));
530
531 ceph_fscache_unregister_inode_cookie(ci);
532
533 __ceph_remove_caps(inode);
534
535 /*
536 * we may still have a snap_realm reference if there are stray
537 * caps in i_snap_caps.
538 */
539 if (ci->i_snap_realm) {
540 struct ceph_mds_client *mdsc =
541 ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc;
542 struct ceph_snap_realm *realm = ci->i_snap_realm;
543
544 dout(" dropping residual ref to snap realm %p\n", realm);
545 spin_lock(&realm->inodes_with_caps_lock);
546 list_del_init(&ci->i_snap_realm_item);
547 spin_unlock(&realm->inodes_with_caps_lock);
548 ceph_put_snap_realm(mdsc, realm);
549 }
550
551 kfree(ci->i_symlink);
552 while ((n = rb_first(&ci->i_fragtree)) != NULL) {
553 frag = rb_entry(n, struct ceph_inode_frag, node);
554 rb_erase(n, &ci->i_fragtree);
555 kfree(frag);
556 }
557 ci->i_fragtree_nsplits = 0;
558
559 __ceph_destroy_xattrs(ci);
560 if (ci->i_xattrs.blob)
561 ceph_buffer_put(ci->i_xattrs.blob);
562 if (ci->i_xattrs.prealloc_blob)
563 ceph_buffer_put(ci->i_xattrs.prealloc_blob);
564
565 ceph_put_string(rcu_dereference_raw(ci->i_layout.pool_ns));
566
567 call_rcu(&inode->i_rcu, ceph_i_callback);
568 }
569
570 int ceph_drop_inode(struct inode *inode)
571 {
572 /*
573 * Positve dentry and corresponding inode are always accompanied
574 * in MDS reply. So no need to keep inode in the cache after
575 * dropping all its aliases.
576 */
577 return 1;
578 }
579
580 static inline blkcnt_t calc_inode_blocks(u64 size)
581 {
582 return (size + (1<<9) - 1) >> 9;
583 }
584
585 /*
586 * Helpers to fill in size, ctime, mtime, and atime. We have to be
587 * careful because either the client or MDS may have more up to date
588 * info, depending on which capabilities are held, and whether
589 * time_warp_seq or truncate_seq have increased. (Ordinarily, mtime
590 * and size are monotonically increasing, except when utimes() or
591 * truncate() increments the corresponding _seq values.)
592 */
593 int ceph_fill_file_size(struct inode *inode, int issued,
594 u32 truncate_seq, u64 truncate_size, u64 size)
595 {
596 struct ceph_inode_info *ci = ceph_inode(inode);
597 int queue_trunc = 0;
598
599 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 ||
600 (truncate_seq == ci->i_truncate_seq && size > inode->i_size)) {
601 dout("size %lld -> %llu\n", inode->i_size, size);
602 if (size > 0 && S_ISDIR(inode->i_mode)) {
603 pr_err("fill_file_size non-zero size for directory\n");
604 size = 0;
605 }
606 i_size_write(inode, size);
607 inode->i_blocks = calc_inode_blocks(size);
608 ci->i_reported_size = size;
609 if (truncate_seq != ci->i_truncate_seq) {
610 dout("truncate_seq %u -> %u\n",
611 ci->i_truncate_seq, truncate_seq);
612 ci->i_truncate_seq = truncate_seq;
613
614 /* the MDS should have revoked these caps */
615 WARN_ON_ONCE(issued & (CEPH_CAP_FILE_EXCL |
616 CEPH_CAP_FILE_RD |
617 CEPH_CAP_FILE_WR |
618 CEPH_CAP_FILE_LAZYIO));
619 /*
620 * If we hold relevant caps, or in the case where we're
621 * not the only client referencing this file and we
622 * don't hold those caps, then we need to check whether
623 * the file is either opened or mmaped
624 */
625 if ((issued & (CEPH_CAP_FILE_CACHE|
626 CEPH_CAP_FILE_BUFFER)) ||
627 mapping_mapped(inode->i_mapping) ||
628 __ceph_caps_file_wanted(ci)) {
629 ci->i_truncate_pending++;
630 queue_trunc = 1;
631 }
632 }
633 }
634 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 &&
635 ci->i_truncate_size != truncate_size) {
636 dout("truncate_size %lld -> %llu\n", ci->i_truncate_size,
637 truncate_size);
638 ci->i_truncate_size = truncate_size;
639 }
640
641 if (queue_trunc)
642 ceph_fscache_invalidate(inode);
643
644 return queue_trunc;
645 }
646
647 void ceph_fill_file_time(struct inode *inode, int issued,
648 u64 time_warp_seq, struct timespec *ctime,
649 struct timespec *mtime, struct timespec *atime)
650 {
651 struct ceph_inode_info *ci = ceph_inode(inode);
652 int warn = 0;
653
654 if (issued & (CEPH_CAP_FILE_EXCL|
655 CEPH_CAP_FILE_WR|
656 CEPH_CAP_FILE_BUFFER|
657 CEPH_CAP_AUTH_EXCL|
658 CEPH_CAP_XATTR_EXCL)) {
659 if (ci->i_version == 0 ||
660 timespec_compare(ctime, &inode->i_ctime) > 0) {
661 dout("ctime %ld.%09ld -> %ld.%09ld inc w/ cap\n",
662 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
663 ctime->tv_sec, ctime->tv_nsec);
664 inode->i_ctime = *ctime;
665 }
666 if (ci->i_version == 0 ||
667 ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) {
668 /* the MDS did a utimes() */
669 dout("mtime %ld.%09ld -> %ld.%09ld "
670 "tw %d -> %d\n",
671 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
672 mtime->tv_sec, mtime->tv_nsec,
673 ci->i_time_warp_seq, (int)time_warp_seq);
674
675 inode->i_mtime = *mtime;
676 inode->i_atime = *atime;
677 ci->i_time_warp_seq = time_warp_seq;
678 } else if (time_warp_seq == ci->i_time_warp_seq) {
679 /* nobody did utimes(); take the max */
680 if (timespec_compare(mtime, &inode->i_mtime) > 0) {
681 dout("mtime %ld.%09ld -> %ld.%09ld inc\n",
682 inode->i_mtime.tv_sec,
683 inode->i_mtime.tv_nsec,
684 mtime->tv_sec, mtime->tv_nsec);
685 inode->i_mtime = *mtime;
686 }
687 if (timespec_compare(atime, &inode->i_atime) > 0) {
688 dout("atime %ld.%09ld -> %ld.%09ld inc\n",
689 inode->i_atime.tv_sec,
690 inode->i_atime.tv_nsec,
691 atime->tv_sec, atime->tv_nsec);
692 inode->i_atime = *atime;
693 }
694 } else if (issued & CEPH_CAP_FILE_EXCL) {
695 /* we did a utimes(); ignore mds values */
696 } else {
697 warn = 1;
698 }
699 } else {
700 /* we have no write|excl caps; whatever the MDS says is true */
701 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) {
702 inode->i_ctime = *ctime;
703 inode->i_mtime = *mtime;
704 inode->i_atime = *atime;
705 ci->i_time_warp_seq = time_warp_seq;
706 } else {
707 warn = 1;
708 }
709 }
710 if (warn) /* time_warp_seq shouldn't go backwards */
711 dout("%p mds time_warp_seq %llu < %u\n",
712 inode, time_warp_seq, ci->i_time_warp_seq);
713 }
714
715 /*
716 * Populate an inode based on info from mds. May be called on new or
717 * existing inodes.
718 */
719 static int fill_inode(struct inode *inode, struct page *locked_page,
720 struct ceph_mds_reply_info_in *iinfo,
721 struct ceph_mds_reply_dirfrag *dirinfo,
722 struct ceph_mds_session *session,
723 unsigned long ttl_from, int cap_fmode,
724 struct ceph_cap_reservation *caps_reservation)
725 {
726 struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
727 struct ceph_mds_reply_inode *info = iinfo->in;
728 struct ceph_inode_info *ci = ceph_inode(inode);
729 int issued = 0, implemented, new_issued;
730 struct timespec mtime, atime, ctime;
731 struct ceph_buffer *xattr_blob = NULL;
732 struct ceph_string *pool_ns = NULL;
733 struct ceph_cap *new_cap = NULL;
734 int err = 0;
735 bool wake = false;
736 bool queue_trunc = false;
737 bool new_version = false;
738 bool fill_inline = false;
739
740 dout("fill_inode %p ino %llx.%llx v %llu had %llu\n",
741 inode, ceph_vinop(inode), le64_to_cpu(info->version),
742 ci->i_version);
743
744 /* prealloc new cap struct */
745 if (info->cap.caps && ceph_snap(inode) == CEPH_NOSNAP)
746 new_cap = ceph_get_cap(mdsc, caps_reservation);
747
748 /*
749 * prealloc xattr data, if it looks like we'll need it. only
750 * if len > 4 (meaning there are actually xattrs; the first 4
751 * bytes are the xattr count).
752 */
753 if (iinfo->xattr_len > 4) {
754 xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS);
755 if (!xattr_blob)
756 pr_err("fill_inode ENOMEM xattr blob %d bytes\n",
757 iinfo->xattr_len);
758 }
759
760 if (iinfo->pool_ns_len > 0)
761 pool_ns = ceph_find_or_create_string(iinfo->pool_ns_data,
762 iinfo->pool_ns_len);
763
764 spin_lock(&ci->i_ceph_lock);
765
766 /*
767 * provided version will be odd if inode value is projected,
768 * even if stable. skip the update if we have newer stable
769 * info (ours>=theirs, e.g. due to racing mds replies), unless
770 * we are getting projected (unstable) info (in which case the
771 * version is odd, and we want ours>theirs).
772 * us them
773 * 2 2 skip
774 * 3 2 skip
775 * 3 3 update
776 */
777 if (ci->i_version == 0 ||
778 ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
779 le64_to_cpu(info->version) > (ci->i_version & ~1)))
780 new_version = true;
781
782 issued = __ceph_caps_issued(ci, &implemented);
783 issued |= implemented | __ceph_caps_dirty(ci);
784 new_issued = ~issued & le32_to_cpu(info->cap.caps);
785
786 /* update inode */
787 inode->i_rdev = le32_to_cpu(info->rdev);
788 inode->i_blkbits = fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1;
789
790 if ((new_version || (new_issued & CEPH_CAP_AUTH_SHARED)) &&
791 (issued & CEPH_CAP_AUTH_EXCL) == 0) {
792 inode->i_mode = le32_to_cpu(info->mode);
793 inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(info->uid));
794 inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(info->gid));
795 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
796 from_kuid(&init_user_ns, inode->i_uid),
797 from_kgid(&init_user_ns, inode->i_gid));
798 }
799
800 if ((new_version || (new_issued & CEPH_CAP_LINK_SHARED)) &&
801 (issued & CEPH_CAP_LINK_EXCL) == 0)
802 set_nlink(inode, le32_to_cpu(info->nlink));
803
804 if (new_version || (new_issued & CEPH_CAP_ANY_RD)) {
805 /* be careful with mtime, atime, size */
806 ceph_decode_timespec(&atime, &info->atime);
807 ceph_decode_timespec(&mtime, &info->mtime);
808 ceph_decode_timespec(&ctime, &info->ctime);
809 ceph_fill_file_time(inode, issued,
810 le32_to_cpu(info->time_warp_seq),
811 &ctime, &mtime, &atime);
812 }
813
814 if (new_version ||
815 (new_issued & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR))) {
816 s64 old_pool = ci->i_layout.pool_id;
817 struct ceph_string *old_ns;
818
819 ceph_file_layout_from_legacy(&ci->i_layout, &info->layout);
820 old_ns = rcu_dereference_protected(ci->i_layout.pool_ns,
821 lockdep_is_held(&ci->i_ceph_lock));
822 rcu_assign_pointer(ci->i_layout.pool_ns, pool_ns);
823
824 if (ci->i_layout.pool_id != old_pool || pool_ns != old_ns)
825 ci->i_ceph_flags &= ~CEPH_I_POOL_PERM;
826
827 pool_ns = old_ns;
828
829 queue_trunc = ceph_fill_file_size(inode, issued,
830 le32_to_cpu(info->truncate_seq),
831 le64_to_cpu(info->truncate_size),
832 le64_to_cpu(info->size));
833 /* only update max_size on auth cap */
834 if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
835 ci->i_max_size != le64_to_cpu(info->max_size)) {
836 dout("max_size %lld -> %llu\n", ci->i_max_size,
837 le64_to_cpu(info->max_size));
838 ci->i_max_size = le64_to_cpu(info->max_size);
839 }
840 }
841
842 /* xattrs */
843 /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
844 if ((ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL)) &&
845 le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) {
846 if (ci->i_xattrs.blob)
847 ceph_buffer_put(ci->i_xattrs.blob);
848 ci->i_xattrs.blob = xattr_blob;
849 if (xattr_blob)
850 memcpy(ci->i_xattrs.blob->vec.iov_base,
851 iinfo->xattr_data, iinfo->xattr_len);
852 ci->i_xattrs.version = le64_to_cpu(info->xattr_version);
853 ceph_forget_all_cached_acls(inode);
854 xattr_blob = NULL;
855 }
856
857 /* finally update i_version */
858 ci->i_version = le64_to_cpu(info->version);
859
860 inode->i_mapping->a_ops = &ceph_aops;
861
862 switch (inode->i_mode & S_IFMT) {
863 case S_IFIFO:
864 case S_IFBLK:
865 case S_IFCHR:
866 case S_IFSOCK:
867 init_special_inode(inode, inode->i_mode, inode->i_rdev);
868 inode->i_op = &ceph_file_iops;
869 break;
870 case S_IFREG:
871 inode->i_op = &ceph_file_iops;
872 inode->i_fop = &ceph_file_fops;
873 break;
874 case S_IFLNK:
875 inode->i_op = &ceph_symlink_iops;
876 if (!ci->i_symlink) {
877 u32 symlen = iinfo->symlink_len;
878 char *sym;
879
880 spin_unlock(&ci->i_ceph_lock);
881
882 if (symlen != i_size_read(inode)) {
883 pr_err("fill_inode %llx.%llx BAD symlink "
884 "size %lld\n", ceph_vinop(inode),
885 i_size_read(inode));
886 i_size_write(inode, symlen);
887 inode->i_blocks = calc_inode_blocks(symlen);
888 }
889
890 err = -ENOMEM;
891 sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS);
892 if (!sym)
893 goto out;
894
895 spin_lock(&ci->i_ceph_lock);
896 if (!ci->i_symlink)
897 ci->i_symlink = sym;
898 else
899 kfree(sym); /* lost a race */
900 }
901 inode->i_link = ci->i_symlink;
902 break;
903 case S_IFDIR:
904 inode->i_op = &ceph_dir_iops;
905 inode->i_fop = &ceph_dir_fops;
906
907 ci->i_dir_layout = iinfo->dir_layout;
908
909 ci->i_files = le64_to_cpu(info->files);
910 ci->i_subdirs = le64_to_cpu(info->subdirs);
911 ci->i_rbytes = le64_to_cpu(info->rbytes);
912 ci->i_rfiles = le64_to_cpu(info->rfiles);
913 ci->i_rsubdirs = le64_to_cpu(info->rsubdirs);
914 ceph_decode_timespec(&ci->i_rctime, &info->rctime);
915 break;
916 default:
917 pr_err("fill_inode %llx.%llx BAD mode 0%o\n",
918 ceph_vinop(inode), inode->i_mode);
919 }
920
921 /* were we issued a capability? */
922 if (info->cap.caps) {
923 if (ceph_snap(inode) == CEPH_NOSNAP) {
924 unsigned caps = le32_to_cpu(info->cap.caps);
925 ceph_add_cap(inode, session,
926 le64_to_cpu(info->cap.cap_id),
927 cap_fmode, caps,
928 le32_to_cpu(info->cap.wanted),
929 le32_to_cpu(info->cap.seq),
930 le32_to_cpu(info->cap.mseq),
931 le64_to_cpu(info->cap.realm),
932 info->cap.flags, &new_cap);
933
934 /* set dir completion flag? */
935 if (S_ISDIR(inode->i_mode) &&
936 ci->i_files == 0 && ci->i_subdirs == 0 &&
937 (caps & CEPH_CAP_FILE_SHARED) &&
938 (issued & CEPH_CAP_FILE_EXCL) == 0 &&
939 !__ceph_dir_is_complete(ci)) {
940 dout(" marking %p complete (empty)\n", inode);
941 i_size_write(inode, 0);
942 __ceph_dir_set_complete(ci,
943 atomic64_read(&ci->i_release_count),
944 atomic64_read(&ci->i_ordered_count));
945 }
946
947 wake = true;
948 } else {
949 dout(" %p got snap_caps %s\n", inode,
950 ceph_cap_string(le32_to_cpu(info->cap.caps)));
951 ci->i_snap_caps |= le32_to_cpu(info->cap.caps);
952 if (cap_fmode >= 0)
953 __ceph_get_fmode(ci, cap_fmode);
954 }
955 } else if (cap_fmode >= 0) {
956 pr_warn("mds issued no caps on %llx.%llx\n",
957 ceph_vinop(inode));
958 __ceph_get_fmode(ci, cap_fmode);
959 }
960
961 if (iinfo->inline_version > 0 &&
962 iinfo->inline_version >= ci->i_inline_version) {
963 int cache_caps = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
964 ci->i_inline_version = iinfo->inline_version;
965 if (ci->i_inline_version != CEPH_INLINE_NONE &&
966 (locked_page ||
967 (le32_to_cpu(info->cap.caps) & cache_caps)))
968 fill_inline = true;
969 }
970
971 spin_unlock(&ci->i_ceph_lock);
972
973 if (fill_inline)
974 ceph_fill_inline_data(inode, locked_page,
975 iinfo->inline_data, iinfo->inline_len);
976
977 if (wake)
978 wake_up_all(&ci->i_cap_wq);
979
980 /* queue truncate if we saw i_size decrease */
981 if (queue_trunc)
982 ceph_queue_vmtruncate(inode);
983
984 /* populate frag tree */
985 if (S_ISDIR(inode->i_mode))
986 ceph_fill_fragtree(inode, &info->fragtree, dirinfo);
987
988 /* update delegation info? */
989 if (dirinfo)
990 ceph_fill_dirfrag(inode, dirinfo);
991
992 err = 0;
993 out:
994 if (new_cap)
995 ceph_put_cap(mdsc, new_cap);
996 if (xattr_blob)
997 ceph_buffer_put(xattr_blob);
998 ceph_put_string(pool_ns);
999 return err;
1000 }
1001
1002 /*
1003 * caller should hold session s_mutex.
1004 */
1005 static void update_dentry_lease(struct dentry *dentry,
1006 struct ceph_mds_reply_lease *lease,
1007 struct ceph_mds_session *session,
1008 unsigned long from_time,
1009 struct ceph_vino *tgt_vino,
1010 struct ceph_vino *dir_vino)
1011 {
1012 struct ceph_dentry_info *di = ceph_dentry(dentry);
1013 long unsigned duration = le32_to_cpu(lease->duration_ms);
1014 long unsigned ttl = from_time + (duration * HZ) / 1000;
1015 long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000;
1016 struct inode *dir;
1017 struct ceph_mds_session *old_lease_session = NULL;
1018
1019 /*
1020 * Make sure dentry's inode matches tgt_vino. NULL tgt_vino means that
1021 * we expect a negative dentry.
1022 */
1023 if (!tgt_vino && d_really_is_positive(dentry))
1024 return;
1025
1026 if (tgt_vino && (d_really_is_negative(dentry) ||
1027 !ceph_ino_compare(d_inode(dentry), tgt_vino)))
1028 return;
1029
1030 spin_lock(&dentry->d_lock);
1031 dout("update_dentry_lease %p duration %lu ms ttl %lu\n",
1032 dentry, duration, ttl);
1033
1034 dir = d_inode(dentry->d_parent);
1035
1036 /* make sure parent matches dir_vino */
1037 if (!ceph_ino_compare(dir, dir_vino))
1038 goto out_unlock;
1039
1040 /* only track leases on regular dentries */
1041 if (ceph_snap(dir) != CEPH_NOSNAP)
1042 goto out_unlock;
1043
1044 di->lease_shared_gen = atomic_read(&ceph_inode(dir)->i_shared_gen);
1045
1046 if (duration == 0)
1047 goto out_unlock;
1048
1049 if (di->lease_gen == session->s_cap_gen &&
1050 time_before(ttl, di->time))
1051 goto out_unlock; /* we already have a newer lease. */
1052
1053 if (di->lease_session && di->lease_session != session) {
1054 old_lease_session = di->lease_session;
1055 di->lease_session = NULL;
1056 }
1057
1058 ceph_dentry_lru_touch(dentry);
1059
1060 if (!di->lease_session)
1061 di->lease_session = ceph_get_mds_session(session);
1062 di->lease_gen = session->s_cap_gen;
1063 di->lease_seq = le32_to_cpu(lease->seq);
1064 di->lease_renew_after = half_ttl;
1065 di->lease_renew_from = 0;
1066 di->time = ttl;
1067 out_unlock:
1068 spin_unlock(&dentry->d_lock);
1069 if (old_lease_session)
1070 ceph_put_mds_session(old_lease_session);
1071 }
1072
1073 /*
1074 * splice a dentry to an inode.
1075 * caller must hold directory i_mutex for this to be safe.
1076 */
1077 static struct dentry *splice_dentry(struct dentry *dn, struct inode *in)
1078 {
1079 struct dentry *realdn;
1080
1081 BUG_ON(d_inode(dn));
1082
1083 /* dn must be unhashed */
1084 if (!d_unhashed(dn))
1085 d_drop(dn);
1086 realdn = d_splice_alias(in, dn);
1087 if (IS_ERR(realdn)) {
1088 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
1089 PTR_ERR(realdn), dn, in, ceph_vinop(in));
1090 dn = realdn; /* note realdn contains the error */
1091 goto out;
1092 } else if (realdn) {
1093 dout("dn %p (%d) spliced with %p (%d) "
1094 "inode %p ino %llx.%llx\n",
1095 dn, d_count(dn),
1096 realdn, d_count(realdn),
1097 d_inode(realdn), ceph_vinop(d_inode(realdn)));
1098 dput(dn);
1099 dn = realdn;
1100 } else {
1101 BUG_ON(!ceph_dentry(dn));
1102 dout("dn %p attached to %p ino %llx.%llx\n",
1103 dn, d_inode(dn), ceph_vinop(d_inode(dn)));
1104 }
1105 out:
1106 return dn;
1107 }
1108
1109 /*
1110 * Incorporate results into the local cache. This is either just
1111 * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
1112 * after a lookup).
1113 *
1114 * A reply may contain
1115 * a directory inode along with a dentry.
1116 * and/or a target inode
1117 *
1118 * Called with snap_rwsem (read).
1119 */
1120 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req)
1121 {
1122 struct ceph_mds_session *session = req->r_session;
1123 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1124 struct inode *in = NULL;
1125 struct ceph_vino tvino, dvino;
1126 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
1127 int err = 0;
1128
1129 dout("fill_trace %p is_dentry %d is_target %d\n", req,
1130 rinfo->head->is_dentry, rinfo->head->is_target);
1131
1132 if (!rinfo->head->is_target && !rinfo->head->is_dentry) {
1133 dout("fill_trace reply is empty!\n");
1134 if (rinfo->head->result == 0 && req->r_parent)
1135 ceph_invalidate_dir_request(req);
1136 return 0;
1137 }
1138
1139 if (rinfo->head->is_dentry) {
1140 struct inode *dir = req->r_parent;
1141
1142 if (dir) {
1143 err = fill_inode(dir, NULL,
1144 &rinfo->diri, rinfo->dirfrag,
1145 session, req->r_request_started, -1,
1146 &req->r_caps_reservation);
1147 if (err < 0)
1148 goto done;
1149 } else {
1150 WARN_ON_ONCE(1);
1151 }
1152
1153 if (dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME) {
1154 struct qstr dname;
1155 struct dentry *dn, *parent;
1156
1157 BUG_ON(!rinfo->head->is_target);
1158 BUG_ON(req->r_dentry);
1159
1160 parent = d_find_any_alias(dir);
1161 BUG_ON(!parent);
1162
1163 dname.name = rinfo->dname;
1164 dname.len = rinfo->dname_len;
1165 dname.hash = full_name_hash(parent, dname.name, dname.len);
1166 tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1167 tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1168 retry_lookup:
1169 dn = d_lookup(parent, &dname);
1170 dout("d_lookup on parent=%p name=%.*s got %p\n",
1171 parent, dname.len, dname.name, dn);
1172
1173 if (!dn) {
1174 dn = d_alloc(parent, &dname);
1175 dout("d_alloc %p '%.*s' = %p\n", parent,
1176 dname.len, dname.name, dn);
1177 if (!dn) {
1178 dput(parent);
1179 err = -ENOMEM;
1180 goto done;
1181 }
1182 err = 0;
1183 } else if (d_really_is_positive(dn) &&
1184 (ceph_ino(d_inode(dn)) != tvino.ino ||
1185 ceph_snap(d_inode(dn)) != tvino.snap)) {
1186 dout(" dn %p points to wrong inode %p\n",
1187 dn, d_inode(dn));
1188 ceph_dir_clear_ordered(dir);
1189 d_delete(dn);
1190 dput(dn);
1191 goto retry_lookup;
1192 }
1193
1194 req->r_dentry = dn;
1195 dput(parent);
1196 }
1197 }
1198
1199 if (rinfo->head->is_target) {
1200 tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1201 tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1202
1203 in = ceph_get_inode(sb, tvino);
1204 if (IS_ERR(in)) {
1205 err = PTR_ERR(in);
1206 goto done;
1207 }
1208 req->r_target_inode = in;
1209
1210 err = fill_inode(in, req->r_locked_page, &rinfo->targeti, NULL,
1211 session, req->r_request_started,
1212 (!test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) &&
1213 rinfo->head->result == 0) ? req->r_fmode : -1,
1214 &req->r_caps_reservation);
1215 if (err < 0) {
1216 pr_err("fill_inode badness %p %llx.%llx\n",
1217 in, ceph_vinop(in));
1218 goto done;
1219 }
1220 }
1221
1222 /*
1223 * ignore null lease/binding on snapdir ENOENT, or else we
1224 * will have trouble splicing in the virtual snapdir later
1225 */
1226 if (rinfo->head->is_dentry &&
1227 !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) &&
1228 test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1229 (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name,
1230 fsc->mount_options->snapdir_name,
1231 req->r_dentry->d_name.len))) {
1232 /*
1233 * lookup link rename : null -> possibly existing inode
1234 * mknod symlink mkdir : null -> new inode
1235 * unlink : linked -> null
1236 */
1237 struct inode *dir = req->r_parent;
1238 struct dentry *dn = req->r_dentry;
1239 bool have_dir_cap, have_lease;
1240
1241 BUG_ON(!dn);
1242 BUG_ON(!dir);
1243 BUG_ON(d_inode(dn->d_parent) != dir);
1244
1245 dvino.ino = le64_to_cpu(rinfo->diri.in->ino);
1246 dvino.snap = le64_to_cpu(rinfo->diri.in->snapid);
1247
1248 BUG_ON(ceph_ino(dir) != dvino.ino);
1249 BUG_ON(ceph_snap(dir) != dvino.snap);
1250
1251 /* do we have a lease on the whole dir? */
1252 have_dir_cap =
1253 (le32_to_cpu(rinfo->diri.in->cap.caps) &
1254 CEPH_CAP_FILE_SHARED);
1255
1256 /* do we have a dn lease? */
1257 have_lease = have_dir_cap ||
1258 le32_to_cpu(rinfo->dlease->duration_ms);
1259 if (!have_lease)
1260 dout("fill_trace no dentry lease or dir cap\n");
1261
1262 /* rename? */
1263 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) {
1264 struct inode *olddir = req->r_old_dentry_dir;
1265 BUG_ON(!olddir);
1266
1267 dout(" src %p '%pd' dst %p '%pd'\n",
1268 req->r_old_dentry,
1269 req->r_old_dentry,
1270 dn, dn);
1271 dout("fill_trace doing d_move %p -> %p\n",
1272 req->r_old_dentry, dn);
1273
1274 /* d_move screws up sibling dentries' offsets */
1275 ceph_dir_clear_ordered(dir);
1276 ceph_dir_clear_ordered(olddir);
1277
1278 d_move(req->r_old_dentry, dn);
1279 dout(" src %p '%pd' dst %p '%pd'\n",
1280 req->r_old_dentry,
1281 req->r_old_dentry,
1282 dn, dn);
1283
1284 /* ensure target dentry is invalidated, despite
1285 rehashing bug in vfs_rename_dir */
1286 ceph_invalidate_dentry_lease(dn);
1287
1288 dout("dn %p gets new offset %lld\n", req->r_old_dentry,
1289 ceph_dentry(req->r_old_dentry)->offset);
1290
1291 dn = req->r_old_dentry; /* use old_dentry */
1292 }
1293
1294 /* null dentry? */
1295 if (!rinfo->head->is_target) {
1296 dout("fill_trace null dentry\n");
1297 if (d_really_is_positive(dn)) {
1298 ceph_dir_clear_ordered(dir);
1299 dout("d_delete %p\n", dn);
1300 d_delete(dn);
1301 } else if (have_lease) {
1302 if (d_unhashed(dn))
1303 d_add(dn, NULL);
1304 update_dentry_lease(dn, rinfo->dlease,
1305 session,
1306 req->r_request_started,
1307 NULL, &dvino);
1308 }
1309 goto done;
1310 }
1311
1312 /* attach proper inode */
1313 if (d_really_is_negative(dn)) {
1314 ceph_dir_clear_ordered(dir);
1315 ihold(in);
1316 dn = splice_dentry(dn, in);
1317 if (IS_ERR(dn)) {
1318 err = PTR_ERR(dn);
1319 goto done;
1320 }
1321 req->r_dentry = dn; /* may have spliced */
1322 } else if (d_really_is_positive(dn) && d_inode(dn) != in) {
1323 dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1324 dn, d_inode(dn), ceph_vinop(d_inode(dn)),
1325 ceph_vinop(in));
1326 ceph_dir_clear_ordered(dir);
1327 d_invalidate(dn);
1328 have_lease = false;
1329 }
1330
1331 if (have_lease) {
1332 tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1333 tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1334 update_dentry_lease(dn, rinfo->dlease, session,
1335 req->r_request_started,
1336 &tvino, &dvino);
1337 }
1338 dout(" final dn %p\n", dn);
1339 } else if ((req->r_op == CEPH_MDS_OP_LOOKUPSNAP ||
1340 req->r_op == CEPH_MDS_OP_MKSNAP) &&
1341 !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
1342 struct dentry *dn = req->r_dentry;
1343 struct inode *dir = req->r_parent;
1344
1345 /* fill out a snapdir LOOKUPSNAP dentry */
1346 BUG_ON(!dn);
1347 BUG_ON(!dir);
1348 BUG_ON(ceph_snap(dir) != CEPH_SNAPDIR);
1349 dout(" linking snapped dir %p to dn %p\n", in, dn);
1350 ceph_dir_clear_ordered(dir);
1351 ihold(in);
1352 dn = splice_dentry(dn, in);
1353 if (IS_ERR(dn)) {
1354 err = PTR_ERR(dn);
1355 goto done;
1356 }
1357 req->r_dentry = dn; /* may have spliced */
1358 } else if (rinfo->head->is_dentry) {
1359 struct ceph_vino *ptvino = NULL;
1360
1361 if ((le32_to_cpu(rinfo->diri.in->cap.caps) & CEPH_CAP_FILE_SHARED) ||
1362 le32_to_cpu(rinfo->dlease->duration_ms)) {
1363 dvino.ino = le64_to_cpu(rinfo->diri.in->ino);
1364 dvino.snap = le64_to_cpu(rinfo->diri.in->snapid);
1365
1366 if (rinfo->head->is_target) {
1367 tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1368 tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1369 ptvino = &tvino;
1370 }
1371
1372 update_dentry_lease(req->r_dentry, rinfo->dlease,
1373 session, req->r_request_started, ptvino,
1374 &dvino);
1375 } else {
1376 dout("%s: no dentry lease or dir cap\n", __func__);
1377 }
1378 }
1379 done:
1380 dout("fill_trace done err=%d\n", err);
1381 return err;
1382 }
1383
1384 /*
1385 * Prepopulate our cache with readdir results, leases, etc.
1386 */
1387 static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req,
1388 struct ceph_mds_session *session)
1389 {
1390 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1391 int i, err = 0;
1392
1393 for (i = 0; i < rinfo->dir_nr; i++) {
1394 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1395 struct ceph_vino vino;
1396 struct inode *in;
1397 int rc;
1398
1399 vino.ino = le64_to_cpu(rde->inode.in->ino);
1400 vino.snap = le64_to_cpu(rde->inode.in->snapid);
1401
1402 in = ceph_get_inode(req->r_dentry->d_sb, vino);
1403 if (IS_ERR(in)) {
1404 err = PTR_ERR(in);
1405 dout("new_inode badness got %d\n", err);
1406 continue;
1407 }
1408 rc = fill_inode(in, NULL, &rde->inode, NULL, session,
1409 req->r_request_started, -1,
1410 &req->r_caps_reservation);
1411 if (rc < 0) {
1412 pr_err("fill_inode badness on %p got %d\n", in, rc);
1413 err = rc;
1414 }
1415 iput(in);
1416 }
1417
1418 return err;
1419 }
1420
1421 void ceph_readdir_cache_release(struct ceph_readdir_cache_control *ctl)
1422 {
1423 if (ctl->page) {
1424 kunmap(ctl->page);
1425 put_page(ctl->page);
1426 ctl->page = NULL;
1427 }
1428 }
1429
1430 static int fill_readdir_cache(struct inode *dir, struct dentry *dn,
1431 struct ceph_readdir_cache_control *ctl,
1432 struct ceph_mds_request *req)
1433 {
1434 struct ceph_inode_info *ci = ceph_inode(dir);
1435 unsigned nsize = PAGE_SIZE / sizeof(struct dentry*);
1436 unsigned idx = ctl->index % nsize;
1437 pgoff_t pgoff = ctl->index / nsize;
1438
1439 if (!ctl->page || pgoff != page_index(ctl->page)) {
1440 ceph_readdir_cache_release(ctl);
1441 if (idx == 0)
1442 ctl->page = grab_cache_page(&dir->i_data, pgoff);
1443 else
1444 ctl->page = find_lock_page(&dir->i_data, pgoff);
1445 if (!ctl->page) {
1446 ctl->index = -1;
1447 return idx == 0 ? -ENOMEM : 0;
1448 }
1449 /* reading/filling the cache are serialized by
1450 * i_mutex, no need to use page lock */
1451 unlock_page(ctl->page);
1452 ctl->dentries = kmap(ctl->page);
1453 if (idx == 0)
1454 memset(ctl->dentries, 0, PAGE_SIZE);
1455 }
1456
1457 if (req->r_dir_release_cnt == atomic64_read(&ci->i_release_count) &&
1458 req->r_dir_ordered_cnt == atomic64_read(&ci->i_ordered_count)) {
1459 dout("readdir cache dn %p idx %d\n", dn, ctl->index);
1460 ctl->dentries[idx] = dn;
1461 ctl->index++;
1462 } else {
1463 dout("disable readdir cache\n");
1464 ctl->index = -1;
1465 }
1466 return 0;
1467 }
1468
1469 int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1470 struct ceph_mds_session *session)
1471 {
1472 struct dentry *parent = req->r_dentry;
1473 struct ceph_inode_info *ci = ceph_inode(d_inode(parent));
1474 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1475 struct qstr dname;
1476 struct dentry *dn;
1477 struct inode *in;
1478 int err = 0, skipped = 0, ret, i;
1479 struct ceph_mds_request_head *rhead = req->r_request->front.iov_base;
1480 u32 frag = le32_to_cpu(rhead->args.readdir.frag);
1481 u32 last_hash = 0;
1482 u32 fpos_offset;
1483 struct ceph_readdir_cache_control cache_ctl = {};
1484
1485 if (test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags))
1486 return readdir_prepopulate_inodes_only(req, session);
1487
1488 if (rinfo->hash_order) {
1489 if (req->r_path2) {
1490 last_hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1491 req->r_path2,
1492 strlen(req->r_path2));
1493 last_hash = ceph_frag_value(last_hash);
1494 } else if (rinfo->offset_hash) {
1495 /* mds understands offset_hash */
1496 WARN_ON_ONCE(req->r_readdir_offset != 2);
1497 last_hash = le32_to_cpu(rhead->args.readdir.offset_hash);
1498 }
1499 }
1500
1501 if (rinfo->dir_dir &&
1502 le32_to_cpu(rinfo->dir_dir->frag) != frag) {
1503 dout("readdir_prepopulate got new frag %x -> %x\n",
1504 frag, le32_to_cpu(rinfo->dir_dir->frag));
1505 frag = le32_to_cpu(rinfo->dir_dir->frag);
1506 if (!rinfo->hash_order)
1507 req->r_readdir_offset = 2;
1508 }
1509
1510 if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
1511 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1512 rinfo->dir_nr, parent);
1513 } else {
1514 dout("readdir_prepopulate %d items under dn %p\n",
1515 rinfo->dir_nr, parent);
1516 if (rinfo->dir_dir)
1517 ceph_fill_dirfrag(d_inode(parent), rinfo->dir_dir);
1518
1519 if (ceph_frag_is_leftmost(frag) &&
1520 req->r_readdir_offset == 2 &&
1521 !(rinfo->hash_order && last_hash)) {
1522 /* note dir version at start of readdir so we can
1523 * tell if any dentries get dropped */
1524 req->r_dir_release_cnt =
1525 atomic64_read(&ci->i_release_count);
1526 req->r_dir_ordered_cnt =
1527 atomic64_read(&ci->i_ordered_count);
1528 req->r_readdir_cache_idx = 0;
1529 }
1530 }
1531
1532 cache_ctl.index = req->r_readdir_cache_idx;
1533 fpos_offset = req->r_readdir_offset;
1534
1535 /* FIXME: release caps/leases if error occurs */
1536 for (i = 0; i < rinfo->dir_nr; i++) {
1537 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1538 struct ceph_vino tvino, dvino;
1539
1540 dname.name = rde->name;
1541 dname.len = rde->name_len;
1542 dname.hash = full_name_hash(parent, dname.name, dname.len);
1543
1544 tvino.ino = le64_to_cpu(rde->inode.in->ino);
1545 tvino.snap = le64_to_cpu(rde->inode.in->snapid);
1546
1547 if (rinfo->hash_order) {
1548 u32 hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1549 rde->name, rde->name_len);
1550 hash = ceph_frag_value(hash);
1551 if (hash != last_hash)
1552 fpos_offset = 2;
1553 last_hash = hash;
1554 rde->offset = ceph_make_fpos(hash, fpos_offset++, true);
1555 } else {
1556 rde->offset = ceph_make_fpos(frag, fpos_offset++, false);
1557 }
1558
1559 retry_lookup:
1560 dn = d_lookup(parent, &dname);
1561 dout("d_lookup on parent=%p name=%.*s got %p\n",
1562 parent, dname.len, dname.name, dn);
1563
1564 if (!dn) {
1565 dn = d_alloc(parent, &dname);
1566 dout("d_alloc %p '%.*s' = %p\n", parent,
1567 dname.len, dname.name, dn);
1568 if (!dn) {
1569 dout("d_alloc badness\n");
1570 err = -ENOMEM;
1571 goto out;
1572 }
1573 } else if (d_really_is_positive(dn) &&
1574 (ceph_ino(d_inode(dn)) != tvino.ino ||
1575 ceph_snap(d_inode(dn)) != tvino.snap)) {
1576 dout(" dn %p points to wrong inode %p\n",
1577 dn, d_inode(dn));
1578 __ceph_dir_clear_ordered(ci);
1579 d_delete(dn);
1580 dput(dn);
1581 goto retry_lookup;
1582 }
1583
1584 /* inode */
1585 if (d_really_is_positive(dn)) {
1586 in = d_inode(dn);
1587 } else {
1588 in = ceph_get_inode(parent->d_sb, tvino);
1589 if (IS_ERR(in)) {
1590 dout("new_inode badness\n");
1591 d_drop(dn);
1592 dput(dn);
1593 err = PTR_ERR(in);
1594 goto out;
1595 }
1596 }
1597
1598 ret = fill_inode(in, NULL, &rde->inode, NULL, session,
1599 req->r_request_started, -1,
1600 &req->r_caps_reservation);
1601 if (ret < 0) {
1602 pr_err("fill_inode badness on %p\n", in);
1603 if (d_really_is_positive(dn))
1604 __ceph_dir_clear_ordered(ci);
1605 else
1606 iput(in);
1607 d_drop(dn);
1608 err = ret;
1609 goto next_item;
1610 }
1611
1612 if (d_really_is_negative(dn)) {
1613 struct dentry *realdn;
1614
1615 if (ceph_security_xattr_deadlock(in)) {
1616 dout(" skip splicing dn %p to inode %p"
1617 " (security xattr deadlock)\n", dn, in);
1618 iput(in);
1619 skipped++;
1620 goto next_item;
1621 }
1622
1623 realdn = splice_dentry(dn, in);
1624 if (IS_ERR(realdn)) {
1625 err = PTR_ERR(realdn);
1626 d_drop(dn);
1627 dn = NULL;
1628 goto next_item;
1629 }
1630 dn = realdn;
1631 }
1632
1633 ceph_dentry(dn)->offset = rde->offset;
1634
1635 dvino = ceph_vino(d_inode(parent));
1636 update_dentry_lease(dn, rde->lease, req->r_session,
1637 req->r_request_started, &tvino, &dvino);
1638
1639 if (err == 0 && skipped == 0 && cache_ctl.index >= 0) {
1640 ret = fill_readdir_cache(d_inode(parent), dn,
1641 &cache_ctl, req);
1642 if (ret < 0)
1643 err = ret;
1644 }
1645 next_item:
1646 if (dn)
1647 dput(dn);
1648 }
1649 out:
1650 if (err == 0 && skipped == 0) {
1651 set_bit(CEPH_MDS_R_DID_PREPOPULATE, &req->r_req_flags);
1652 req->r_readdir_cache_idx = cache_ctl.index;
1653 }
1654 ceph_readdir_cache_release(&cache_ctl);
1655 dout("readdir_prepopulate done\n");
1656 return err;
1657 }
1658
1659 bool ceph_inode_set_size(struct inode *inode, loff_t size)
1660 {
1661 struct ceph_inode_info *ci = ceph_inode(inode);
1662 bool ret;
1663
1664 spin_lock(&ci->i_ceph_lock);
1665 dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size);
1666 i_size_write(inode, size);
1667 inode->i_blocks = calc_inode_blocks(size);
1668
1669 ret = __ceph_should_report_size(ci);
1670
1671 spin_unlock(&ci->i_ceph_lock);
1672 return ret;
1673 }
1674
1675 /*
1676 * Write back inode data in a worker thread. (This can't be done
1677 * in the message handler context.)
1678 */
1679 void ceph_queue_writeback(struct inode *inode)
1680 {
1681 struct ceph_inode_info *ci = ceph_inode(inode);
1682 set_bit(CEPH_I_WORK_WRITEBACK, &ci->i_work_mask);
1683
1684 ihold(inode);
1685 if (queue_work(ceph_inode_to_client(inode)->inode_wq,
1686 &ci->i_work)) {
1687 dout("ceph_queue_writeback %p\n", inode);
1688 } else {
1689 dout("ceph_queue_writeback %p already queued, mask=%lx\n",
1690 inode, ci->i_work_mask);
1691 iput(inode);
1692 }
1693 }
1694
1695 /*
1696 * queue an async invalidation
1697 */
1698 void ceph_queue_invalidate(struct inode *inode)
1699 {
1700 struct ceph_inode_info *ci = ceph_inode(inode);
1701 set_bit(CEPH_I_WORK_INVALIDATE_PAGES, &ci->i_work_mask);
1702
1703 ihold(inode);
1704 if (queue_work(ceph_inode_to_client(inode)->inode_wq,
1705 &ceph_inode(inode)->i_work)) {
1706 dout("ceph_queue_invalidate %p\n", inode);
1707 } else {
1708 dout("ceph_queue_invalidate %p already queued, mask=%lx\n",
1709 inode, ci->i_work_mask);
1710 iput(inode);
1711 }
1712 }
1713
1714 /*
1715 * Queue an async vmtruncate. If we fail to queue work, we will handle
1716 * the truncation the next time we call __ceph_do_pending_vmtruncate.
1717 */
1718 void ceph_queue_vmtruncate(struct inode *inode)
1719 {
1720 struct ceph_inode_info *ci = ceph_inode(inode);
1721 set_bit(CEPH_I_WORK_VMTRUNCATE, &ci->i_work_mask);
1722
1723 ihold(inode);
1724 if (queue_work(ceph_inode_to_client(inode)->inode_wq,
1725 &ci->i_work)) {
1726 dout("ceph_queue_vmtruncate %p\n", inode);
1727 } else {
1728 dout("ceph_queue_vmtruncate %p already queued, mask=%lx\n",
1729 inode, ci->i_work_mask);
1730 iput(inode);
1731 }
1732 }
1733
1734 static void ceph_do_invalidate_pages(struct inode *inode)
1735 {
1736 struct ceph_inode_info *ci = ceph_inode(inode);
1737 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1738 u32 orig_gen;
1739 int check = 0;
1740
1741 mutex_lock(&ci->i_truncate_mutex);
1742
1743 if (READ_ONCE(fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) {
1744 pr_warn_ratelimited("invalidate_pages %p %lld forced umount\n",
1745 inode, ceph_ino(inode));
1746 mapping_set_error(inode->i_mapping, -EIO);
1747 truncate_pagecache(inode, 0);
1748 mutex_unlock(&ci->i_truncate_mutex);
1749 goto out;
1750 }
1751
1752 spin_lock(&ci->i_ceph_lock);
1753 dout("invalidate_pages %p gen %d revoking %d\n", inode,
1754 ci->i_rdcache_gen, ci->i_rdcache_revoking);
1755 if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
1756 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1757 check = 1;
1758 spin_unlock(&ci->i_ceph_lock);
1759 mutex_unlock(&ci->i_truncate_mutex);
1760 goto out;
1761 }
1762 orig_gen = ci->i_rdcache_gen;
1763 spin_unlock(&ci->i_ceph_lock);
1764
1765 if (invalidate_inode_pages2(inode->i_mapping) < 0) {
1766 pr_err("invalidate_pages %p fails\n", inode);
1767 }
1768
1769 spin_lock(&ci->i_ceph_lock);
1770 if (orig_gen == ci->i_rdcache_gen &&
1771 orig_gen == ci->i_rdcache_revoking) {
1772 dout("invalidate_pages %p gen %d successful\n", inode,
1773 ci->i_rdcache_gen);
1774 ci->i_rdcache_revoking--;
1775 check = 1;
1776 } else {
1777 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1778 inode, orig_gen, ci->i_rdcache_gen,
1779 ci->i_rdcache_revoking);
1780 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1781 check = 1;
1782 }
1783 spin_unlock(&ci->i_ceph_lock);
1784 mutex_unlock(&ci->i_truncate_mutex);
1785 out:
1786 if (check)
1787 ceph_check_caps(ci, 0, NULL);
1788 }
1789
1790 /*
1791 * Make sure any pending truncation is applied before doing anything
1792 * that may depend on it.
1793 */
1794 void __ceph_do_pending_vmtruncate(struct inode *inode)
1795 {
1796 struct ceph_inode_info *ci = ceph_inode(inode);
1797 u64 to;
1798 int wrbuffer_refs, finish = 0;
1799
1800 mutex_lock(&ci->i_truncate_mutex);
1801 retry:
1802 spin_lock(&ci->i_ceph_lock);
1803 if (ci->i_truncate_pending == 0) {
1804 dout("__do_pending_vmtruncate %p none pending\n", inode);
1805 spin_unlock(&ci->i_ceph_lock);
1806 mutex_unlock(&ci->i_truncate_mutex);
1807 return;
1808 }
1809
1810 /*
1811 * make sure any dirty snapped pages are flushed before we
1812 * possibly truncate them.. so write AND block!
1813 */
1814 if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) {
1815 struct ceph_cap_snap *capsnap;
1816 to = ci->i_truncate_size;
1817 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
1818 // MDS should have revoked Frw caps
1819 WARN_ON_ONCE(capsnap->writing);
1820 if (capsnap->dirty_pages && capsnap->size > to)
1821 to = capsnap->size;
1822 }
1823 spin_unlock(&ci->i_ceph_lock);
1824 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1825 inode);
1826
1827 truncate_pagecache(inode, to);
1828
1829 filemap_write_and_wait_range(&inode->i_data, 0,
1830 inode->i_sb->s_maxbytes);
1831 goto retry;
1832 }
1833
1834 /* there should be no reader or writer */
1835 WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref);
1836
1837 to = ci->i_truncate_size;
1838 wrbuffer_refs = ci->i_wrbuffer_ref;
1839 dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode,
1840 ci->i_truncate_pending, to);
1841 spin_unlock(&ci->i_ceph_lock);
1842
1843 truncate_pagecache(inode, to);
1844
1845 spin_lock(&ci->i_ceph_lock);
1846 if (to == ci->i_truncate_size) {
1847 ci->i_truncate_pending = 0;
1848 finish = 1;
1849 }
1850 spin_unlock(&ci->i_ceph_lock);
1851 if (!finish)
1852 goto retry;
1853
1854 mutex_unlock(&ci->i_truncate_mutex);
1855
1856 if (wrbuffer_refs == 0)
1857 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
1858
1859 wake_up_all(&ci->i_cap_wq);
1860 }
1861
1862 static void ceph_inode_work(struct work_struct *work)
1863 {
1864 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1865 i_work);
1866 struct inode *inode = &ci->vfs_inode;
1867
1868 if (test_and_clear_bit(CEPH_I_WORK_WRITEBACK, &ci->i_work_mask)) {
1869 dout("writeback %p\n", inode);
1870 filemap_fdatawrite(&inode->i_data);
1871 }
1872 if (test_and_clear_bit(CEPH_I_WORK_INVALIDATE_PAGES, &ci->i_work_mask))
1873 ceph_do_invalidate_pages(inode);
1874
1875 if (test_and_clear_bit(CEPH_I_WORK_VMTRUNCATE, &ci->i_work_mask))
1876 __ceph_do_pending_vmtruncate(inode);
1877
1878 iput(inode);
1879 }
1880
1881 /*
1882 * symlinks
1883 */
1884 static const struct inode_operations ceph_symlink_iops = {
1885 .get_link = simple_get_link,
1886 .setattr = ceph_setattr,
1887 .getattr = ceph_getattr,
1888 .listxattr = ceph_listxattr,
1889 };
1890
1891 int __ceph_setattr(struct inode *inode, struct iattr *attr)
1892 {
1893 struct ceph_inode_info *ci = ceph_inode(inode);
1894 const unsigned int ia_valid = attr->ia_valid;
1895 struct ceph_mds_request *req;
1896 struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
1897 struct ceph_cap_flush *prealloc_cf;
1898 int issued;
1899 int release = 0, dirtied = 0;
1900 int mask = 0;
1901 int err = 0;
1902 int inode_dirty_flags = 0;
1903 bool lock_snap_rwsem = false;
1904
1905 prealloc_cf = ceph_alloc_cap_flush();
1906 if (!prealloc_cf)
1907 return -ENOMEM;
1908
1909 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR,
1910 USE_AUTH_MDS);
1911 if (IS_ERR(req)) {
1912 ceph_free_cap_flush(prealloc_cf);
1913 return PTR_ERR(req);
1914 }
1915
1916 spin_lock(&ci->i_ceph_lock);
1917 issued = __ceph_caps_issued(ci, NULL);
1918
1919 if (!ci->i_head_snapc &&
1920 (issued & (CEPH_CAP_ANY_EXCL | CEPH_CAP_FILE_WR))) {
1921 lock_snap_rwsem = true;
1922 if (!down_read_trylock(&mdsc->snap_rwsem)) {
1923 spin_unlock(&ci->i_ceph_lock);
1924 down_read(&mdsc->snap_rwsem);
1925 spin_lock(&ci->i_ceph_lock);
1926 issued = __ceph_caps_issued(ci, NULL);
1927 }
1928 }
1929
1930 dout("setattr %p issued %s\n", inode, ceph_cap_string(issued));
1931
1932 if (ia_valid & ATTR_UID) {
1933 dout("setattr %p uid %d -> %d\n", inode,
1934 from_kuid(&init_user_ns, inode->i_uid),
1935 from_kuid(&init_user_ns, attr->ia_uid));
1936 if (issued & CEPH_CAP_AUTH_EXCL) {
1937 inode->i_uid = attr->ia_uid;
1938 dirtied |= CEPH_CAP_AUTH_EXCL;
1939 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1940 !uid_eq(attr->ia_uid, inode->i_uid)) {
1941 req->r_args.setattr.uid = cpu_to_le32(
1942 from_kuid(&init_user_ns, attr->ia_uid));
1943 mask |= CEPH_SETATTR_UID;
1944 release |= CEPH_CAP_AUTH_SHARED;
1945 }
1946 }
1947 if (ia_valid & ATTR_GID) {
1948 dout("setattr %p gid %d -> %d\n", inode,
1949 from_kgid(&init_user_ns, inode->i_gid),
1950 from_kgid(&init_user_ns, attr->ia_gid));
1951 if (issued & CEPH_CAP_AUTH_EXCL) {
1952 inode->i_gid = attr->ia_gid;
1953 dirtied |= CEPH_CAP_AUTH_EXCL;
1954 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1955 !gid_eq(attr->ia_gid, inode->i_gid)) {
1956 req->r_args.setattr.gid = cpu_to_le32(
1957 from_kgid(&init_user_ns, attr->ia_gid));
1958 mask |= CEPH_SETATTR_GID;
1959 release |= CEPH_CAP_AUTH_SHARED;
1960 }
1961 }
1962 if (ia_valid & ATTR_MODE) {
1963 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode,
1964 attr->ia_mode);
1965 if (issued & CEPH_CAP_AUTH_EXCL) {
1966 inode->i_mode = attr->ia_mode;
1967 dirtied |= CEPH_CAP_AUTH_EXCL;
1968 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1969 attr->ia_mode != inode->i_mode) {
1970 inode->i_mode = attr->ia_mode;
1971 req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode);
1972 mask |= CEPH_SETATTR_MODE;
1973 release |= CEPH_CAP_AUTH_SHARED;
1974 }
1975 }
1976
1977 if (ia_valid & ATTR_ATIME) {
1978 dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode,
1979 inode->i_atime.tv_sec, inode->i_atime.tv_nsec,
1980 attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec);
1981 if (issued & CEPH_CAP_FILE_EXCL) {
1982 ci->i_time_warp_seq++;
1983 inode->i_atime = attr->ia_atime;
1984 dirtied |= CEPH_CAP_FILE_EXCL;
1985 } else if ((issued & CEPH_CAP_FILE_WR) &&
1986 timespec_compare(&inode->i_atime,
1987 &attr->ia_atime) < 0) {
1988 inode->i_atime = attr->ia_atime;
1989 dirtied |= CEPH_CAP_FILE_WR;
1990 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1991 !timespec_equal(&inode->i_atime, &attr->ia_atime)) {
1992 ceph_encode_timespec(&req->r_args.setattr.atime,
1993 &attr->ia_atime);
1994 mask |= CEPH_SETATTR_ATIME;
1995 release |= CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_RD |
1996 CEPH_CAP_FILE_WR;
1997 }
1998 }
1999 if (ia_valid & ATTR_MTIME) {
2000 dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode,
2001 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
2002 attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec);
2003 if (issued & CEPH_CAP_FILE_EXCL) {
2004 ci->i_time_warp_seq++;
2005 inode->i_mtime = attr->ia_mtime;
2006 dirtied |= CEPH_CAP_FILE_EXCL;
2007 } else if ((issued & CEPH_CAP_FILE_WR) &&
2008 timespec_compare(&inode->i_mtime,
2009 &attr->ia_mtime) < 0) {
2010 inode->i_mtime = attr->ia_mtime;
2011 dirtied |= CEPH_CAP_FILE_WR;
2012 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2013 !timespec_equal(&inode->i_mtime, &attr->ia_mtime)) {
2014 ceph_encode_timespec(&req->r_args.setattr.mtime,
2015 &attr->ia_mtime);
2016 mask |= CEPH_SETATTR_MTIME;
2017 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
2018 CEPH_CAP_FILE_WR;
2019 }
2020 }
2021 if (ia_valid & ATTR_SIZE) {
2022 dout("setattr %p size %lld -> %lld\n", inode,
2023 inode->i_size, attr->ia_size);
2024 if ((issued & CEPH_CAP_FILE_EXCL) &&
2025 attr->ia_size > inode->i_size) {
2026 i_size_write(inode, attr->ia_size);
2027 inode->i_blocks = calc_inode_blocks(attr->ia_size);
2028 ci->i_reported_size = attr->ia_size;
2029 dirtied |= CEPH_CAP_FILE_EXCL;
2030 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2031 attr->ia_size != inode->i_size) {
2032 req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
2033 req->r_args.setattr.old_size =
2034 cpu_to_le64(inode->i_size);
2035 mask |= CEPH_SETATTR_SIZE;
2036 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
2037 CEPH_CAP_FILE_WR;
2038 }
2039 }
2040
2041 /* these do nothing */
2042 if (ia_valid & ATTR_CTIME) {
2043 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME|
2044 ATTR_MODE|ATTR_UID|ATTR_GID)) == 0;
2045 dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode,
2046 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
2047 attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec,
2048 only ? "ctime only" : "ignored");
2049 if (only) {
2050 /*
2051 * if kernel wants to dirty ctime but nothing else,
2052 * we need to choose a cap to dirty under, or do
2053 * a almost-no-op setattr
2054 */
2055 if (issued & CEPH_CAP_AUTH_EXCL)
2056 dirtied |= CEPH_CAP_AUTH_EXCL;
2057 else if (issued & CEPH_CAP_FILE_EXCL)
2058 dirtied |= CEPH_CAP_FILE_EXCL;
2059 else if (issued & CEPH_CAP_XATTR_EXCL)
2060 dirtied |= CEPH_CAP_XATTR_EXCL;
2061 else
2062 mask |= CEPH_SETATTR_CTIME;
2063 }
2064 }
2065 if (ia_valid & ATTR_FILE)
2066 dout("setattr %p ATTR_FILE ... hrm!\n", inode);
2067
2068 if (dirtied) {
2069 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied,
2070 &prealloc_cf);
2071 inode->i_ctime = attr->ia_ctime;
2072 }
2073
2074 release &= issued;
2075 spin_unlock(&ci->i_ceph_lock);
2076 if (lock_snap_rwsem)
2077 up_read(&mdsc->snap_rwsem);
2078
2079 if (inode_dirty_flags)
2080 __mark_inode_dirty(inode, inode_dirty_flags);
2081
2082
2083 if (mask) {
2084 req->r_inode = inode;
2085 ihold(inode);
2086 req->r_inode_drop = release;
2087 req->r_args.setattr.mask = cpu_to_le32(mask);
2088 req->r_num_caps = 1;
2089 req->r_stamp = attr->ia_ctime;
2090 err = ceph_mdsc_do_request(mdsc, NULL, req);
2091 }
2092 dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err,
2093 ceph_cap_string(dirtied), mask);
2094
2095 ceph_mdsc_put_request(req);
2096 ceph_free_cap_flush(prealloc_cf);
2097
2098 if (err >= 0 && (mask & CEPH_SETATTR_SIZE))
2099 __ceph_do_pending_vmtruncate(inode);
2100
2101 return err;
2102 }
2103
2104 /*
2105 * setattr
2106 */
2107 int ceph_setattr(struct dentry *dentry, struct iattr *attr)
2108 {
2109 struct inode *inode = d_inode(dentry);
2110 int err;
2111
2112 if (ceph_snap(inode) != CEPH_NOSNAP)
2113 return -EROFS;
2114
2115 err = setattr_prepare(dentry, attr);
2116 if (err != 0)
2117 return err;
2118
2119 err = __ceph_setattr(inode, attr);
2120
2121 if (err >= 0 && (attr->ia_valid & ATTR_MODE))
2122 err = posix_acl_chmod(inode, attr->ia_mode);
2123
2124 return err;
2125 }
2126
2127 /*
2128 * Verify that we have a lease on the given mask. If not,
2129 * do a getattr against an mds.
2130 */
2131 int __ceph_do_getattr(struct inode *inode, struct page *locked_page,
2132 int mask, bool force)
2133 {
2134 struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
2135 struct ceph_mds_client *mdsc = fsc->mdsc;
2136 struct ceph_mds_request *req;
2137 int err;
2138
2139 if (ceph_snap(inode) == CEPH_SNAPDIR) {
2140 dout("do_getattr inode %p SNAPDIR\n", inode);
2141 return 0;
2142 }
2143
2144 dout("do_getattr inode %p mask %s mode 0%o\n",
2145 inode, ceph_cap_string(mask), inode->i_mode);
2146 if (!force && ceph_caps_issued_mask(ceph_inode(inode), mask, 1))
2147 return 0;
2148
2149 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
2150 if (IS_ERR(req))
2151 return PTR_ERR(req);
2152 req->r_inode = inode;
2153 ihold(inode);
2154 req->r_num_caps = 1;
2155 req->r_args.getattr.mask = cpu_to_le32(mask);
2156 req->r_locked_page = locked_page;
2157 err = ceph_mdsc_do_request(mdsc, NULL, req);
2158 if (locked_page && err == 0) {
2159 u64 inline_version = req->r_reply_info.targeti.inline_version;
2160 if (inline_version == 0) {
2161 /* the reply is supposed to contain inline data */
2162 err = -EINVAL;
2163 } else if (inline_version == CEPH_INLINE_NONE) {
2164 err = -ENODATA;
2165 } else {
2166 err = req->r_reply_info.targeti.inline_len;
2167 }
2168 }
2169 ceph_mdsc_put_request(req);
2170 dout("do_getattr result=%d\n", err);
2171 return err;
2172 }
2173
2174
2175 /*
2176 * Check inode permissions. We verify we have a valid value for
2177 * the AUTH cap, then call the generic handler.
2178 */
2179 int ceph_permission(struct inode *inode, int mask)
2180 {
2181 int err;
2182
2183 if (mask & MAY_NOT_BLOCK)
2184 return -ECHILD;
2185
2186 err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED, false);
2187
2188 if (!err)
2189 err = generic_permission(inode, mask);
2190 return err;
2191 }
2192
2193 /*
2194 * Get all attributes. Hopefully somedata we'll have a statlite()
2195 * and can limit the fields we require to be accurate.
2196 */
2197 int ceph_getattr(const struct path *path, struct kstat *stat,
2198 u32 request_mask, unsigned int flags)
2199 {
2200 struct inode *inode = d_inode(path->dentry);
2201 struct ceph_inode_info *ci = ceph_inode(inode);
2202 int err;
2203
2204 err = ceph_do_getattr(inode, CEPH_STAT_CAP_INODE_ALL, false);
2205 if (!err) {
2206 generic_fillattr(inode, stat);
2207 stat->ino = ceph_translate_ino(inode->i_sb, inode->i_ino);
2208 if (ceph_snap(inode) != CEPH_NOSNAP)
2209 stat->dev = ceph_snap(inode);
2210 else
2211 stat->dev = 0;
2212 if (S_ISDIR(inode->i_mode)) {
2213 if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb),
2214 RBYTES))
2215 stat->size = ci->i_rbytes;
2216 else
2217 stat->size = ci->i_files + ci->i_subdirs;
2218 stat->blocks = 0;
2219 stat->blksize = 65536;
2220 }
2221 }
2222 return err;
2223 }
ubuntu-bionic-kernel mirror