Loading...
1#include <linux/ceph/ceph_debug.h>
2
3#include <linux/module.h>
4#include <linux/fs.h>
5#include <linux/slab.h>
6#include <linux/string.h>
7#include <linux/uaccess.h>
8#include <linux/kernel.h>
9#include <linux/namei.h>
10#include <linux/writeback.h>
11#include <linux/vmalloc.h>
12
13#include "super.h"
14#include "mds_client.h"
15#include <linux/ceph/decode.h>
16
17/*
18 * Ceph inode operations
19 *
20 * Implement basic inode helpers (get, alloc) and inode ops (getattr,
21 * setattr, etc.), xattr helpers, and helpers for assimilating
22 * metadata returned by the MDS into our cache.
23 *
24 * Also define helpers for doing asynchronous writeback, invalidation,
25 * and truncation for the benefit of those who can't afford to block
26 * (typically because they are in the message handler path).
27 */
28
29static const struct inode_operations ceph_symlink_iops;
30
31static void ceph_invalidate_work(struct work_struct *work);
32static void ceph_writeback_work(struct work_struct *work);
33static void ceph_vmtruncate_work(struct work_struct *work);
34
35/*
36 * find or create an inode, given the ceph ino number
37 */
38static int ceph_set_ino_cb(struct inode *inode, void *data)
39{
40 ceph_inode(inode)->i_vino = *(struct ceph_vino *)data;
41 inode->i_ino = ceph_vino_to_ino(*(struct ceph_vino *)data);
42 return 0;
43}
44
45struct inode *ceph_get_inode(struct super_block *sb, struct ceph_vino vino)
46{
47 struct inode *inode;
48 ino_t t = ceph_vino_to_ino(vino);
49
50 inode = iget5_locked(sb, t, ceph_ino_compare, ceph_set_ino_cb, &vino);
51 if (inode == NULL)
52 return ERR_PTR(-ENOMEM);
53 if (inode->i_state & I_NEW) {
54 dout("get_inode created new inode %p %llx.%llx ino %llx\n",
55 inode, ceph_vinop(inode), (u64)inode->i_ino);
56 unlock_new_inode(inode);
57 }
58
59 dout("get_inode on %lu=%llx.%llx got %p\n", inode->i_ino, vino.ino,
60 vino.snap, inode);
61 return inode;
62}
63
64/*
65 * get/constuct snapdir inode for a given directory
66 */
67struct inode *ceph_get_snapdir(struct inode *parent)
68{
69 struct ceph_vino vino = {
70 .ino = ceph_ino(parent),
71 .snap = CEPH_SNAPDIR,
72 };
73 struct inode *inode = ceph_get_inode(parent->i_sb, vino);
74 struct ceph_inode_info *ci = ceph_inode(inode);
75
76 BUG_ON(!S_ISDIR(parent->i_mode));
77 if (IS_ERR(inode))
78 return inode;
79 inode->i_mode = parent->i_mode;
80 inode->i_uid = parent->i_uid;
81 inode->i_gid = parent->i_gid;
82 inode->i_op = &ceph_dir_iops;
83 inode->i_fop = &ceph_dir_fops;
84 ci->i_snap_caps = CEPH_CAP_PIN; /* so we can open */
85 ci->i_rbytes = 0;
86 return inode;
87}
88
89const struct inode_operations ceph_file_iops = {
90 .permission = ceph_permission,
91 .setattr = ceph_setattr,
92 .getattr = ceph_getattr,
93 .setxattr = ceph_setxattr,
94 .getxattr = ceph_getxattr,
95 .listxattr = ceph_listxattr,
96 .removexattr = ceph_removexattr,
97};
98
99
100/*
101 * We use a 'frag tree' to keep track of the MDS's directory fragments
102 * for a given inode (usually there is just a single fragment). We
103 * need to know when a child frag is delegated to a new MDS, or when
104 * it is flagged as replicated, so we can direct our requests
105 * accordingly.
106 */
107
108/*
109 * find/create a frag in the tree
110 */
111static struct ceph_inode_frag *__get_or_create_frag(struct ceph_inode_info *ci,
112 u32 f)
113{
114 struct rb_node **p;
115 struct rb_node *parent = NULL;
116 struct ceph_inode_frag *frag;
117 int c;
118
119 p = &ci->i_fragtree.rb_node;
120 while (*p) {
121 parent = *p;
122 frag = rb_entry(parent, struct ceph_inode_frag, node);
123 c = ceph_frag_compare(f, frag->frag);
124 if (c < 0)
125 p = &(*p)->rb_left;
126 else if (c > 0)
127 p = &(*p)->rb_right;
128 else
129 return frag;
130 }
131
132 frag = kmalloc(sizeof(*frag), GFP_NOFS);
133 if (!frag) {
134 pr_err("__get_or_create_frag ENOMEM on %p %llx.%llx "
135 "frag %x\n", &ci->vfs_inode,
136 ceph_vinop(&ci->vfs_inode), f);
137 return ERR_PTR(-ENOMEM);
138 }
139 frag->frag = f;
140 frag->split_by = 0;
141 frag->mds = -1;
142 frag->ndist = 0;
143
144 rb_link_node(&frag->node, parent, p);
145 rb_insert_color(&frag->node, &ci->i_fragtree);
146
147 dout("get_or_create_frag added %llx.%llx frag %x\n",
148 ceph_vinop(&ci->vfs_inode), f);
149 return frag;
150}
151
152/*
153 * find a specific frag @f
154 */
155struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci, u32 f)
156{
157 struct rb_node *n = ci->i_fragtree.rb_node;
158
159 while (n) {
160 struct ceph_inode_frag *frag =
161 rb_entry(n, struct ceph_inode_frag, node);
162 int c = ceph_frag_compare(f, frag->frag);
163 if (c < 0)
164 n = n->rb_left;
165 else if (c > 0)
166 n = n->rb_right;
167 else
168 return frag;
169 }
170 return NULL;
171}
172
173/*
174 * Choose frag containing the given value @v. If @pfrag is
175 * specified, copy the frag delegation info to the caller if
176 * it is present.
177 */
178u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
179 struct ceph_inode_frag *pfrag,
180 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 mutex_lock(&ci->i_fragtree_mutex);
191 while (1) {
192 WARN_ON(!ceph_frag_contains_value(t, v));
193 frag = __ceph_find_frag(ci, t);
194 if (!frag)
195 break; /* t is a leaf */
196 if (frag->split_by == 0) {
197 if (pfrag)
198 memcpy(pfrag, frag, sizeof(*pfrag));
199 if (found)
200 *found = 1;
201 break;
202 }
203
204 /* choose child */
205 nway = 1 << frag->split_by;
206 dout("choose_frag(%x) %x splits by %d (%d ways)\n", v, t,
207 frag->split_by, nway);
208 for (i = 0; i < nway; i++) {
209 n = ceph_frag_make_child(t, frag->split_by, i);
210 if (ceph_frag_contains_value(n, v)) {
211 t = n;
212 break;
213 }
214 }
215 BUG_ON(i == nway);
216 }
217 dout("choose_frag(%x) = %x\n", v, t);
218
219 mutex_unlock(&ci->i_fragtree_mutex);
220 return t;
221}
222
223/*
224 * Process dirfrag (delegation) info from the mds. Include leaf
225 * fragment in tree ONLY if ndist > 0. Otherwise, only
226 * branches/splits are included in i_fragtree)
227 */
228static int ceph_fill_dirfrag(struct inode *inode,
229 struct ceph_mds_reply_dirfrag *dirinfo)
230{
231 struct ceph_inode_info *ci = ceph_inode(inode);
232 struct ceph_inode_frag *frag;
233 u32 id = le32_to_cpu(dirinfo->frag);
234 int mds = le32_to_cpu(dirinfo->auth);
235 int ndist = le32_to_cpu(dirinfo->ndist);
236 int i;
237 int err = 0;
238
239 mutex_lock(&ci->i_fragtree_mutex);
240 if (ndist == 0) {
241 /* no delegation info needed. */
242 frag = __ceph_find_frag(ci, id);
243 if (!frag)
244 goto out;
245 if (frag->split_by == 0) {
246 /* tree leaf, remove */
247 dout("fill_dirfrag removed %llx.%llx frag %x"
248 " (no ref)\n", ceph_vinop(inode), id);
249 rb_erase(&frag->node, &ci->i_fragtree);
250 kfree(frag);
251 } else {
252 /* tree branch, keep and clear */
253 dout("fill_dirfrag cleared %llx.%llx frag %x"
254 " referral\n", ceph_vinop(inode), id);
255 frag->mds = -1;
256 frag->ndist = 0;
257 }
258 goto out;
259 }
260
261
262 /* find/add this frag to store mds delegation info */
263 frag = __get_or_create_frag(ci, id);
264 if (IS_ERR(frag)) {
265 /* this is not the end of the world; we can continue
266 with bad/inaccurate delegation info */
267 pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n",
268 ceph_vinop(inode), le32_to_cpu(dirinfo->frag));
269 err = -ENOMEM;
270 goto out;
271 }
272
273 frag->mds = mds;
274 frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP);
275 for (i = 0; i < frag->ndist; i++)
276 frag->dist[i] = le32_to_cpu(dirinfo->dist[i]);
277 dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n",
278 ceph_vinop(inode), frag->frag, frag->ndist);
279
280out:
281 mutex_unlock(&ci->i_fragtree_mutex);
282 return err;
283}
284
285
286/*
287 * initialize a newly allocated inode.
288 */
289struct inode *ceph_alloc_inode(struct super_block *sb)
290{
291 struct ceph_inode_info *ci;
292 int i;
293
294 ci = kmem_cache_alloc(ceph_inode_cachep, GFP_NOFS);
295 if (!ci)
296 return NULL;
297
298 dout("alloc_inode %p\n", &ci->vfs_inode);
299
300 spin_lock_init(&ci->i_ceph_lock);
301
302 ci->i_version = 0;
303 ci->i_time_warp_seq = 0;
304 ci->i_ceph_flags = 0;
305 ci->i_release_count = 0;
306 ci->i_symlink = NULL;
307
308 memset(&ci->i_dir_layout, 0, sizeof(ci->i_dir_layout));
309
310 ci->i_fragtree = RB_ROOT;
311 mutex_init(&ci->i_fragtree_mutex);
312
313 ci->i_xattrs.blob = NULL;
314 ci->i_xattrs.prealloc_blob = NULL;
315 ci->i_xattrs.dirty = false;
316 ci->i_xattrs.index = RB_ROOT;
317 ci->i_xattrs.count = 0;
318 ci->i_xattrs.names_size = 0;
319 ci->i_xattrs.vals_size = 0;
320 ci->i_xattrs.version = 0;
321 ci->i_xattrs.index_version = 0;
322
323 ci->i_caps = RB_ROOT;
324 ci->i_auth_cap = NULL;
325 ci->i_dirty_caps = 0;
326 ci->i_flushing_caps = 0;
327 INIT_LIST_HEAD(&ci->i_dirty_item);
328 INIT_LIST_HEAD(&ci->i_flushing_item);
329 ci->i_cap_flush_seq = 0;
330 ci->i_cap_flush_last_tid = 0;
331 memset(&ci->i_cap_flush_tid, 0, sizeof(ci->i_cap_flush_tid));
332 init_waitqueue_head(&ci->i_cap_wq);
333 ci->i_hold_caps_min = 0;
334 ci->i_hold_caps_max = 0;
335 INIT_LIST_HEAD(&ci->i_cap_delay_list);
336 ci->i_cap_exporting_mds = 0;
337 ci->i_cap_exporting_mseq = 0;
338 ci->i_cap_exporting_issued = 0;
339 INIT_LIST_HEAD(&ci->i_cap_snaps);
340 ci->i_head_snapc = NULL;
341 ci->i_snap_caps = 0;
342
343 for (i = 0; i < CEPH_FILE_MODE_NUM; i++)
344 ci->i_nr_by_mode[i] = 0;
345
346 ci->i_truncate_seq = 0;
347 ci->i_truncate_size = 0;
348 ci->i_truncate_pending = 0;
349
350 ci->i_max_size = 0;
351 ci->i_reported_size = 0;
352 ci->i_wanted_max_size = 0;
353 ci->i_requested_max_size = 0;
354
355 ci->i_pin_ref = 0;
356 ci->i_rd_ref = 0;
357 ci->i_rdcache_ref = 0;
358 ci->i_wr_ref = 0;
359 ci->i_wb_ref = 0;
360 ci->i_wrbuffer_ref = 0;
361 ci->i_wrbuffer_ref_head = 0;
362 ci->i_shared_gen = 0;
363 ci->i_rdcache_gen = 0;
364 ci->i_rdcache_revoking = 0;
365
366 INIT_LIST_HEAD(&ci->i_unsafe_writes);
367 INIT_LIST_HEAD(&ci->i_unsafe_dirops);
368 spin_lock_init(&ci->i_unsafe_lock);
369
370 ci->i_snap_realm = NULL;
371 INIT_LIST_HEAD(&ci->i_snap_realm_item);
372 INIT_LIST_HEAD(&ci->i_snap_flush_item);
373
374 INIT_WORK(&ci->i_wb_work, ceph_writeback_work);
375 INIT_WORK(&ci->i_pg_inv_work, ceph_invalidate_work);
376
377 INIT_WORK(&ci->i_vmtruncate_work, ceph_vmtruncate_work);
378
379 return &ci->vfs_inode;
380}
381
382static void ceph_i_callback(struct rcu_head *head)
383{
384 struct inode *inode = container_of(head, struct inode, i_rcu);
385 struct ceph_inode_info *ci = ceph_inode(inode);
386
387 kmem_cache_free(ceph_inode_cachep, ci);
388}
389
390void ceph_destroy_inode(struct inode *inode)
391{
392 struct ceph_inode_info *ci = ceph_inode(inode);
393 struct ceph_inode_frag *frag;
394 struct rb_node *n;
395
396 dout("destroy_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode));
397
398 ceph_queue_caps_release(inode);
399
400 /*
401 * we may still have a snap_realm reference if there are stray
402 * caps in i_cap_exporting_issued or i_snap_caps.
403 */
404 if (ci->i_snap_realm) {
405 struct ceph_mds_client *mdsc =
406 ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc;
407 struct ceph_snap_realm *realm = ci->i_snap_realm;
408
409 dout(" dropping residual ref to snap realm %p\n", realm);
410 spin_lock(&realm->inodes_with_caps_lock);
411 list_del_init(&ci->i_snap_realm_item);
412 spin_unlock(&realm->inodes_with_caps_lock);
413 ceph_put_snap_realm(mdsc, realm);
414 }
415
416 kfree(ci->i_symlink);
417 while ((n = rb_first(&ci->i_fragtree)) != NULL) {
418 frag = rb_entry(n, struct ceph_inode_frag, node);
419 rb_erase(n, &ci->i_fragtree);
420 kfree(frag);
421 }
422
423 __ceph_destroy_xattrs(ci);
424 if (ci->i_xattrs.blob)
425 ceph_buffer_put(ci->i_xattrs.blob);
426 if (ci->i_xattrs.prealloc_blob)
427 ceph_buffer_put(ci->i_xattrs.prealloc_blob);
428
429 call_rcu(&inode->i_rcu, ceph_i_callback);
430}
431
432
433/*
434 * Helpers to fill in size, ctime, mtime, and atime. We have to be
435 * careful because either the client or MDS may have more up to date
436 * info, depending on which capabilities are held, and whether
437 * time_warp_seq or truncate_seq have increased. (Ordinarily, mtime
438 * and size are monotonically increasing, except when utimes() or
439 * truncate() increments the corresponding _seq values.)
440 */
441int ceph_fill_file_size(struct inode *inode, int issued,
442 u32 truncate_seq, u64 truncate_size, u64 size)
443{
444 struct ceph_inode_info *ci = ceph_inode(inode);
445 int queue_trunc = 0;
446
447 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 ||
448 (truncate_seq == ci->i_truncate_seq && size > inode->i_size)) {
449 dout("size %lld -> %llu\n", inode->i_size, size);
450 inode->i_size = size;
451 inode->i_blocks = (size + (1<<9) - 1) >> 9;
452 ci->i_reported_size = size;
453 if (truncate_seq != ci->i_truncate_seq) {
454 dout("truncate_seq %u -> %u\n",
455 ci->i_truncate_seq, truncate_seq);
456 ci->i_truncate_seq = truncate_seq;
457 /*
458 * If we hold relevant caps, or in the case where we're
459 * not the only client referencing this file and we
460 * don't hold those caps, then we need to check whether
461 * the file is either opened or mmaped
462 */
463 if ((issued & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_RD|
464 CEPH_CAP_FILE_WR|CEPH_CAP_FILE_BUFFER|
465 CEPH_CAP_FILE_EXCL|
466 CEPH_CAP_FILE_LAZYIO)) ||
467 mapping_mapped(inode->i_mapping) ||
468 __ceph_caps_file_wanted(ci)) {
469 ci->i_truncate_pending++;
470 queue_trunc = 1;
471 }
472 }
473 }
474 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 &&
475 ci->i_truncate_size != truncate_size) {
476 dout("truncate_size %lld -> %llu\n", ci->i_truncate_size,
477 truncate_size);
478 ci->i_truncate_size = truncate_size;
479 }
480 return queue_trunc;
481}
482
483void ceph_fill_file_time(struct inode *inode, int issued,
484 u64 time_warp_seq, struct timespec *ctime,
485 struct timespec *mtime, struct timespec *atime)
486{
487 struct ceph_inode_info *ci = ceph_inode(inode);
488 int warn = 0;
489
490 if (issued & (CEPH_CAP_FILE_EXCL|
491 CEPH_CAP_FILE_WR|
492 CEPH_CAP_FILE_BUFFER|
493 CEPH_CAP_AUTH_EXCL|
494 CEPH_CAP_XATTR_EXCL)) {
495 if (timespec_compare(ctime, &inode->i_ctime) > 0) {
496 dout("ctime %ld.%09ld -> %ld.%09ld inc w/ cap\n",
497 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
498 ctime->tv_sec, ctime->tv_nsec);
499 inode->i_ctime = *ctime;
500 }
501 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) {
502 /* the MDS did a utimes() */
503 dout("mtime %ld.%09ld -> %ld.%09ld "
504 "tw %d -> %d\n",
505 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
506 mtime->tv_sec, mtime->tv_nsec,
507 ci->i_time_warp_seq, (int)time_warp_seq);
508
509 inode->i_mtime = *mtime;
510 inode->i_atime = *atime;
511 ci->i_time_warp_seq = time_warp_seq;
512 } else if (time_warp_seq == ci->i_time_warp_seq) {
513 /* nobody did utimes(); take the max */
514 if (timespec_compare(mtime, &inode->i_mtime) > 0) {
515 dout("mtime %ld.%09ld -> %ld.%09ld inc\n",
516 inode->i_mtime.tv_sec,
517 inode->i_mtime.tv_nsec,
518 mtime->tv_sec, mtime->tv_nsec);
519 inode->i_mtime = *mtime;
520 }
521 if (timespec_compare(atime, &inode->i_atime) > 0) {
522 dout("atime %ld.%09ld -> %ld.%09ld inc\n",
523 inode->i_atime.tv_sec,
524 inode->i_atime.tv_nsec,
525 atime->tv_sec, atime->tv_nsec);
526 inode->i_atime = *atime;
527 }
528 } else if (issued & CEPH_CAP_FILE_EXCL) {
529 /* we did a utimes(); ignore mds values */
530 } else {
531 warn = 1;
532 }
533 } else {
534 /* we have no write|excl caps; whatever the MDS says is true */
535 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) {
536 inode->i_ctime = *ctime;
537 inode->i_mtime = *mtime;
538 inode->i_atime = *atime;
539 ci->i_time_warp_seq = time_warp_seq;
540 } else {
541 warn = 1;
542 }
543 }
544 if (warn) /* time_warp_seq shouldn't go backwards */
545 dout("%p mds time_warp_seq %llu < %u\n",
546 inode, time_warp_seq, ci->i_time_warp_seq);
547}
548
549/*
550 * Populate an inode based on info from mds. May be called on new or
551 * existing inodes.
552 */
553static int fill_inode(struct inode *inode,
554 struct ceph_mds_reply_info_in *iinfo,
555 struct ceph_mds_reply_dirfrag *dirinfo,
556 struct ceph_mds_session *session,
557 unsigned long ttl_from, int cap_fmode,
558 struct ceph_cap_reservation *caps_reservation)
559{
560 struct ceph_mds_reply_inode *info = iinfo->in;
561 struct ceph_inode_info *ci = ceph_inode(inode);
562 int i;
563 int issued = 0, implemented;
564 int updating_inode = 0;
565 struct timespec mtime, atime, ctime;
566 u32 nsplits;
567 struct ceph_buffer *xattr_blob = NULL;
568 int err = 0;
569 int queue_trunc = 0;
570
571 dout("fill_inode %p ino %llx.%llx v %llu had %llu\n",
572 inode, ceph_vinop(inode), le64_to_cpu(info->version),
573 ci->i_version);
574
575 /*
576 * prealloc xattr data, if it looks like we'll need it. only
577 * if len > 4 (meaning there are actually xattrs; the first 4
578 * bytes are the xattr count).
579 */
580 if (iinfo->xattr_len > 4) {
581 xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS);
582 if (!xattr_blob)
583 pr_err("fill_inode ENOMEM xattr blob %d bytes\n",
584 iinfo->xattr_len);
585 }
586
587 spin_lock(&ci->i_ceph_lock);
588
589 /*
590 * provided version will be odd if inode value is projected,
591 * even if stable. skip the update if we have newer stable
592 * info (ours>=theirs, e.g. due to racing mds replies), unless
593 * we are getting projected (unstable) info (in which case the
594 * version is odd, and we want ours>theirs).
595 * us them
596 * 2 2 skip
597 * 3 2 skip
598 * 3 3 update
599 */
600 if (le64_to_cpu(info->version) > 0 &&
601 (ci->i_version & ~1) >= le64_to_cpu(info->version))
602 goto no_change;
603
604 updating_inode = 1;
605 issued = __ceph_caps_issued(ci, &implemented);
606 issued |= implemented | __ceph_caps_dirty(ci);
607
608 /* update inode */
609 ci->i_version = le64_to_cpu(info->version);
610 inode->i_version++;
611 inode->i_rdev = le32_to_cpu(info->rdev);
612
613 if ((issued & CEPH_CAP_AUTH_EXCL) == 0) {
614 inode->i_mode = le32_to_cpu(info->mode);
615 inode->i_uid = le32_to_cpu(info->uid);
616 inode->i_gid = le32_to_cpu(info->gid);
617 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
618 inode->i_uid, inode->i_gid);
619 }
620
621 if ((issued & CEPH_CAP_LINK_EXCL) == 0)
622 set_nlink(inode, le32_to_cpu(info->nlink));
623
624 /* be careful with mtime, atime, size */
625 ceph_decode_timespec(&atime, &info->atime);
626 ceph_decode_timespec(&mtime, &info->mtime);
627 ceph_decode_timespec(&ctime, &info->ctime);
628 queue_trunc = ceph_fill_file_size(inode, issued,
629 le32_to_cpu(info->truncate_seq),
630 le64_to_cpu(info->truncate_size),
631 le64_to_cpu(info->size));
632 ceph_fill_file_time(inode, issued,
633 le32_to_cpu(info->time_warp_seq),
634 &ctime, &mtime, &atime);
635
636 /* only update max_size on auth cap */
637 if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
638 ci->i_max_size != le64_to_cpu(info->max_size)) {
639 dout("max_size %lld -> %llu\n", ci->i_max_size,
640 le64_to_cpu(info->max_size));
641 ci->i_max_size = le64_to_cpu(info->max_size);
642 }
643
644 ci->i_layout = info->layout;
645 inode->i_blkbits = fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1;
646
647 /* xattrs */
648 /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
649 if ((issued & CEPH_CAP_XATTR_EXCL) == 0 &&
650 le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) {
651 if (ci->i_xattrs.blob)
652 ceph_buffer_put(ci->i_xattrs.blob);
653 ci->i_xattrs.blob = xattr_blob;
654 if (xattr_blob)
655 memcpy(ci->i_xattrs.blob->vec.iov_base,
656 iinfo->xattr_data, iinfo->xattr_len);
657 ci->i_xattrs.version = le64_to_cpu(info->xattr_version);
658 xattr_blob = NULL;
659 }
660
661 inode->i_mapping->a_ops = &ceph_aops;
662 inode->i_mapping->backing_dev_info =
663 &ceph_sb_to_client(inode->i_sb)->backing_dev_info;
664
665 switch (inode->i_mode & S_IFMT) {
666 case S_IFIFO:
667 case S_IFBLK:
668 case S_IFCHR:
669 case S_IFSOCK:
670 init_special_inode(inode, inode->i_mode, inode->i_rdev);
671 inode->i_op = &ceph_file_iops;
672 break;
673 case S_IFREG:
674 inode->i_op = &ceph_file_iops;
675 inode->i_fop = &ceph_file_fops;
676 break;
677 case S_IFLNK:
678 inode->i_op = &ceph_symlink_iops;
679 if (!ci->i_symlink) {
680 u32 symlen = iinfo->symlink_len;
681 char *sym;
682
683 spin_unlock(&ci->i_ceph_lock);
684
685 err = -EINVAL;
686 if (WARN_ON(symlen != inode->i_size))
687 goto out;
688
689 err = -ENOMEM;
690 sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS);
691 if (!sym)
692 goto out;
693
694 spin_lock(&ci->i_ceph_lock);
695 if (!ci->i_symlink)
696 ci->i_symlink = sym;
697 else
698 kfree(sym); /* lost a race */
699 }
700 break;
701 case S_IFDIR:
702 inode->i_op = &ceph_dir_iops;
703 inode->i_fop = &ceph_dir_fops;
704
705 ci->i_dir_layout = iinfo->dir_layout;
706
707 ci->i_files = le64_to_cpu(info->files);
708 ci->i_subdirs = le64_to_cpu(info->subdirs);
709 ci->i_rbytes = le64_to_cpu(info->rbytes);
710 ci->i_rfiles = le64_to_cpu(info->rfiles);
711 ci->i_rsubdirs = le64_to_cpu(info->rsubdirs);
712 ceph_decode_timespec(&ci->i_rctime, &info->rctime);
713 break;
714 default:
715 pr_err("fill_inode %llx.%llx BAD mode 0%o\n",
716 ceph_vinop(inode), inode->i_mode);
717 }
718
719no_change:
720 spin_unlock(&ci->i_ceph_lock);
721
722 /* queue truncate if we saw i_size decrease */
723 if (queue_trunc)
724 ceph_queue_vmtruncate(inode);
725
726 /* populate frag tree */
727 /* FIXME: move me up, if/when version reflects fragtree changes */
728 nsplits = le32_to_cpu(info->fragtree.nsplits);
729 mutex_lock(&ci->i_fragtree_mutex);
730 for (i = 0; i < nsplits; i++) {
731 u32 id = le32_to_cpu(info->fragtree.splits[i].frag);
732 struct ceph_inode_frag *frag = __get_or_create_frag(ci, id);
733
734 if (IS_ERR(frag))
735 continue;
736 frag->split_by = le32_to_cpu(info->fragtree.splits[i].by);
737 dout(" frag %x split by %d\n", frag->frag, frag->split_by);
738 }
739 mutex_unlock(&ci->i_fragtree_mutex);
740
741 /* were we issued a capability? */
742 if (info->cap.caps) {
743 if (ceph_snap(inode) == CEPH_NOSNAP) {
744 ceph_add_cap(inode, session,
745 le64_to_cpu(info->cap.cap_id),
746 cap_fmode,
747 le32_to_cpu(info->cap.caps),
748 le32_to_cpu(info->cap.wanted),
749 le32_to_cpu(info->cap.seq),
750 le32_to_cpu(info->cap.mseq),
751 le64_to_cpu(info->cap.realm),
752 info->cap.flags,
753 caps_reservation);
754 } else {
755 spin_lock(&ci->i_ceph_lock);
756 dout(" %p got snap_caps %s\n", inode,
757 ceph_cap_string(le32_to_cpu(info->cap.caps)));
758 ci->i_snap_caps |= le32_to_cpu(info->cap.caps);
759 if (cap_fmode >= 0)
760 __ceph_get_fmode(ci, cap_fmode);
761 spin_unlock(&ci->i_ceph_lock);
762 }
763 } else if (cap_fmode >= 0) {
764 pr_warning("mds issued no caps on %llx.%llx\n",
765 ceph_vinop(inode));
766 __ceph_get_fmode(ci, cap_fmode);
767 }
768
769 /* set dir completion flag? */
770 if (S_ISDIR(inode->i_mode) &&
771 updating_inode && /* didn't jump to no_change */
772 ci->i_files == 0 && ci->i_subdirs == 0 &&
773 ceph_snap(inode) == CEPH_NOSNAP &&
774 (le32_to_cpu(info->cap.caps) & CEPH_CAP_FILE_SHARED) &&
775 (issued & CEPH_CAP_FILE_EXCL) == 0 &&
776 !ceph_dir_test_complete(inode)) {
777 dout(" marking %p complete (empty)\n", inode);
778 ceph_dir_set_complete(inode);
779 ci->i_max_offset = 2;
780 }
781
782 /* update delegation info? */
783 if (dirinfo)
784 ceph_fill_dirfrag(inode, dirinfo);
785
786 err = 0;
787
788out:
789 if (xattr_blob)
790 ceph_buffer_put(xattr_blob);
791 return err;
792}
793
794/*
795 * caller should hold session s_mutex.
796 */
797static void update_dentry_lease(struct dentry *dentry,
798 struct ceph_mds_reply_lease *lease,
799 struct ceph_mds_session *session,
800 unsigned long from_time)
801{
802 struct ceph_dentry_info *di = ceph_dentry(dentry);
803 long unsigned duration = le32_to_cpu(lease->duration_ms);
804 long unsigned ttl = from_time + (duration * HZ) / 1000;
805 long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000;
806 struct inode *dir;
807
808 /* only track leases on regular dentries */
809 if (dentry->d_op != &ceph_dentry_ops)
810 return;
811
812 spin_lock(&dentry->d_lock);
813 dout("update_dentry_lease %p duration %lu ms ttl %lu\n",
814 dentry, duration, ttl);
815
816 /* make lease_rdcache_gen match directory */
817 dir = dentry->d_parent->d_inode;
818 di->lease_shared_gen = ceph_inode(dir)->i_shared_gen;
819
820 if (duration == 0)
821 goto out_unlock;
822
823 if (di->lease_gen == session->s_cap_gen &&
824 time_before(ttl, dentry->d_time))
825 goto out_unlock; /* we already have a newer lease. */
826
827 if (di->lease_session && di->lease_session != session)
828 goto out_unlock;
829
830 ceph_dentry_lru_touch(dentry);
831
832 if (!di->lease_session)
833 di->lease_session = ceph_get_mds_session(session);
834 di->lease_gen = session->s_cap_gen;
835 di->lease_seq = le32_to_cpu(lease->seq);
836 di->lease_renew_after = half_ttl;
837 di->lease_renew_from = 0;
838 dentry->d_time = ttl;
839out_unlock:
840 spin_unlock(&dentry->d_lock);
841 return;
842}
843
844/*
845 * Set dentry's directory position based on the current dir's max, and
846 * order it in d_subdirs, so that dcache_readdir behaves.
847 *
848 * Always called under directory's i_mutex.
849 */
850static void ceph_set_dentry_offset(struct dentry *dn)
851{
852 struct dentry *dir = dn->d_parent;
853 struct inode *inode = dir->d_inode;
854 struct ceph_inode_info *ci;
855 struct ceph_dentry_info *di;
856
857 BUG_ON(!inode);
858
859 ci = ceph_inode(inode);
860 di = ceph_dentry(dn);
861
862 spin_lock(&ci->i_ceph_lock);
863 if (!ceph_dir_test_complete(inode)) {
864 spin_unlock(&ci->i_ceph_lock);
865 return;
866 }
867 di->offset = ceph_inode(inode)->i_max_offset++;
868 spin_unlock(&ci->i_ceph_lock);
869
870 spin_lock(&dir->d_lock);
871 spin_lock_nested(&dn->d_lock, DENTRY_D_LOCK_NESTED);
872 list_move(&dn->d_u.d_child, &dir->d_subdirs);
873 dout("set_dentry_offset %p %lld (%p %p)\n", dn, di->offset,
874 dn->d_u.d_child.prev, dn->d_u.d_child.next);
875 spin_unlock(&dn->d_lock);
876 spin_unlock(&dir->d_lock);
877}
878
879/*
880 * splice a dentry to an inode.
881 * caller must hold directory i_mutex for this to be safe.
882 *
883 * we will only rehash the resulting dentry if @prehash is
884 * true; @prehash will be set to false (for the benefit of
885 * the caller) if we fail.
886 */
887static struct dentry *splice_dentry(struct dentry *dn, struct inode *in,
888 bool *prehash, bool set_offset)
889{
890 struct dentry *realdn;
891
892 BUG_ON(dn->d_inode);
893
894 /* dn must be unhashed */
895 if (!d_unhashed(dn))
896 d_drop(dn);
897 realdn = d_materialise_unique(dn, in);
898 if (IS_ERR(realdn)) {
899 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
900 PTR_ERR(realdn), dn, in, ceph_vinop(in));
901 if (prehash)
902 *prehash = false; /* don't rehash on error */
903 dn = realdn; /* note realdn contains the error */
904 goto out;
905 } else if (realdn) {
906 dout("dn %p (%d) spliced with %p (%d) "
907 "inode %p ino %llx.%llx\n",
908 dn, dn->d_count,
909 realdn, realdn->d_count,
910 realdn->d_inode, ceph_vinop(realdn->d_inode));
911 dput(dn);
912 dn = realdn;
913 } else {
914 BUG_ON(!ceph_dentry(dn));
915 dout("dn %p attached to %p ino %llx.%llx\n",
916 dn, dn->d_inode, ceph_vinop(dn->d_inode));
917 }
918 if ((!prehash || *prehash) && d_unhashed(dn))
919 d_rehash(dn);
920 if (set_offset)
921 ceph_set_dentry_offset(dn);
922out:
923 return dn;
924}
925
926/*
927 * Incorporate results into the local cache. This is either just
928 * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
929 * after a lookup).
930 *
931 * A reply may contain
932 * a directory inode along with a dentry.
933 * and/or a target inode
934 *
935 * Called with snap_rwsem (read).
936 */
937int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req,
938 struct ceph_mds_session *session)
939{
940 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
941 struct inode *in = NULL;
942 struct ceph_mds_reply_inode *ininfo;
943 struct ceph_vino vino;
944 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
945 int i = 0;
946 int err = 0;
947
948 dout("fill_trace %p is_dentry %d is_target %d\n", req,
949 rinfo->head->is_dentry, rinfo->head->is_target);
950
951#if 0
952 /*
953 * Debugging hook:
954 *
955 * If we resend completed ops to a recovering mds, we get no
956 * trace. Since that is very rare, pretend this is the case
957 * to ensure the 'no trace' handlers in the callers behave.
958 *
959 * Fill in inodes unconditionally to avoid breaking cap
960 * invariants.
961 */
962 if (rinfo->head->op & CEPH_MDS_OP_WRITE) {
963 pr_info("fill_trace faking empty trace on %lld %s\n",
964 req->r_tid, ceph_mds_op_name(rinfo->head->op));
965 if (rinfo->head->is_dentry) {
966 rinfo->head->is_dentry = 0;
967 err = fill_inode(req->r_locked_dir,
968 &rinfo->diri, rinfo->dirfrag,
969 session, req->r_request_started, -1);
970 }
971 if (rinfo->head->is_target) {
972 rinfo->head->is_target = 0;
973 ininfo = rinfo->targeti.in;
974 vino.ino = le64_to_cpu(ininfo->ino);
975 vino.snap = le64_to_cpu(ininfo->snapid);
976 in = ceph_get_inode(sb, vino);
977 err = fill_inode(in, &rinfo->targeti, NULL,
978 session, req->r_request_started,
979 req->r_fmode);
980 iput(in);
981 }
982 }
983#endif
984
985 if (!rinfo->head->is_target && !rinfo->head->is_dentry) {
986 dout("fill_trace reply is empty!\n");
987 if (rinfo->head->result == 0 && req->r_locked_dir)
988 ceph_invalidate_dir_request(req);
989 return 0;
990 }
991
992 if (rinfo->head->is_dentry) {
993 struct inode *dir = req->r_locked_dir;
994
995 err = fill_inode(dir, &rinfo->diri, rinfo->dirfrag,
996 session, req->r_request_started, -1,
997 &req->r_caps_reservation);
998 if (err < 0)
999 return err;
1000 }
1001
1002 /*
1003 * ignore null lease/binding on snapdir ENOENT, or else we
1004 * will have trouble splicing in the virtual snapdir later
1005 */
1006 if (rinfo->head->is_dentry && !req->r_aborted &&
1007 (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name,
1008 fsc->mount_options->snapdir_name,
1009 req->r_dentry->d_name.len))) {
1010 /*
1011 * lookup link rename : null -> possibly existing inode
1012 * mknod symlink mkdir : null -> new inode
1013 * unlink : linked -> null
1014 */
1015 struct inode *dir = req->r_locked_dir;
1016 struct dentry *dn = req->r_dentry;
1017 bool have_dir_cap, have_lease;
1018
1019 BUG_ON(!dn);
1020 BUG_ON(!dir);
1021 BUG_ON(dn->d_parent->d_inode != dir);
1022 BUG_ON(ceph_ino(dir) !=
1023 le64_to_cpu(rinfo->diri.in->ino));
1024 BUG_ON(ceph_snap(dir) !=
1025 le64_to_cpu(rinfo->diri.in->snapid));
1026
1027 /* do we have a lease on the whole dir? */
1028 have_dir_cap =
1029 (le32_to_cpu(rinfo->diri.in->cap.caps) &
1030 CEPH_CAP_FILE_SHARED);
1031
1032 /* do we have a dn lease? */
1033 have_lease = have_dir_cap ||
1034 le32_to_cpu(rinfo->dlease->duration_ms);
1035 if (!have_lease)
1036 dout("fill_trace no dentry lease or dir cap\n");
1037
1038 /* rename? */
1039 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) {
1040 dout(" src %p '%.*s' dst %p '%.*s'\n",
1041 req->r_old_dentry,
1042 req->r_old_dentry->d_name.len,
1043 req->r_old_dentry->d_name.name,
1044 dn, dn->d_name.len, dn->d_name.name);
1045 dout("fill_trace doing d_move %p -> %p\n",
1046 req->r_old_dentry, dn);
1047
1048 d_move(req->r_old_dentry, dn);
1049 dout(" src %p '%.*s' dst %p '%.*s'\n",
1050 req->r_old_dentry,
1051 req->r_old_dentry->d_name.len,
1052 req->r_old_dentry->d_name.name,
1053 dn, dn->d_name.len, dn->d_name.name);
1054
1055 /* ensure target dentry is invalidated, despite
1056 rehashing bug in vfs_rename_dir */
1057 ceph_invalidate_dentry_lease(dn);
1058
1059 /*
1060 * d_move() puts the renamed dentry at the end of
1061 * d_subdirs. We need to assign it an appropriate
1062 * directory offset so we can behave when holding
1063 * D_COMPLETE.
1064 */
1065 ceph_set_dentry_offset(req->r_old_dentry);
1066 dout("dn %p gets new offset %lld\n", req->r_old_dentry,
1067 ceph_dentry(req->r_old_dentry)->offset);
1068
1069 dn = req->r_old_dentry; /* use old_dentry */
1070 in = dn->d_inode;
1071 }
1072
1073 /* null dentry? */
1074 if (!rinfo->head->is_target) {
1075 dout("fill_trace null dentry\n");
1076 if (dn->d_inode) {
1077 dout("d_delete %p\n", dn);
1078 d_delete(dn);
1079 } else {
1080 dout("d_instantiate %p NULL\n", dn);
1081 d_instantiate(dn, NULL);
1082 if (have_lease && d_unhashed(dn))
1083 d_rehash(dn);
1084 update_dentry_lease(dn, rinfo->dlease,
1085 session,
1086 req->r_request_started);
1087 }
1088 goto done;
1089 }
1090
1091 /* attach proper inode */
1092 ininfo = rinfo->targeti.in;
1093 vino.ino = le64_to_cpu(ininfo->ino);
1094 vino.snap = le64_to_cpu(ininfo->snapid);
1095 in = dn->d_inode;
1096 if (!in) {
1097 in = ceph_get_inode(sb, vino);
1098 if (IS_ERR(in)) {
1099 pr_err("fill_trace bad get_inode "
1100 "%llx.%llx\n", vino.ino, vino.snap);
1101 err = PTR_ERR(in);
1102 d_delete(dn);
1103 goto done;
1104 }
1105 dn = splice_dentry(dn, in, &have_lease, true);
1106 if (IS_ERR(dn)) {
1107 err = PTR_ERR(dn);
1108 goto done;
1109 }
1110 req->r_dentry = dn; /* may have spliced */
1111 ihold(in);
1112 } else if (ceph_ino(in) == vino.ino &&
1113 ceph_snap(in) == vino.snap) {
1114 ihold(in);
1115 } else {
1116 dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1117 dn, in, ceph_ino(in), ceph_snap(in),
1118 vino.ino, vino.snap);
1119 have_lease = false;
1120 in = NULL;
1121 }
1122
1123 if (have_lease)
1124 update_dentry_lease(dn, rinfo->dlease, session,
1125 req->r_request_started);
1126 dout(" final dn %p\n", dn);
1127 i++;
1128 } else if (req->r_op == CEPH_MDS_OP_LOOKUPSNAP ||
1129 req->r_op == CEPH_MDS_OP_MKSNAP) {
1130 struct dentry *dn = req->r_dentry;
1131
1132 /* fill out a snapdir LOOKUPSNAP dentry */
1133 BUG_ON(!dn);
1134 BUG_ON(!req->r_locked_dir);
1135 BUG_ON(ceph_snap(req->r_locked_dir) != CEPH_SNAPDIR);
1136 ininfo = rinfo->targeti.in;
1137 vino.ino = le64_to_cpu(ininfo->ino);
1138 vino.snap = le64_to_cpu(ininfo->snapid);
1139 in = ceph_get_inode(sb, vino);
1140 if (IS_ERR(in)) {
1141 pr_err("fill_inode get_inode badness %llx.%llx\n",
1142 vino.ino, vino.snap);
1143 err = PTR_ERR(in);
1144 d_delete(dn);
1145 goto done;
1146 }
1147 dout(" linking snapped dir %p to dn %p\n", in, dn);
1148 dn = splice_dentry(dn, in, NULL, true);
1149 if (IS_ERR(dn)) {
1150 err = PTR_ERR(dn);
1151 goto done;
1152 }
1153 req->r_dentry = dn; /* may have spliced */
1154 ihold(in);
1155 rinfo->head->is_dentry = 1; /* fool notrace handlers */
1156 }
1157
1158 if (rinfo->head->is_target) {
1159 vino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1160 vino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1161
1162 if (in == NULL || ceph_ino(in) != vino.ino ||
1163 ceph_snap(in) != vino.snap) {
1164 in = ceph_get_inode(sb, vino);
1165 if (IS_ERR(in)) {
1166 err = PTR_ERR(in);
1167 goto done;
1168 }
1169 }
1170 req->r_target_inode = in;
1171
1172 err = fill_inode(in,
1173 &rinfo->targeti, NULL,
1174 session, req->r_request_started,
1175 (le32_to_cpu(rinfo->head->result) == 0) ?
1176 req->r_fmode : -1,
1177 &req->r_caps_reservation);
1178 if (err < 0) {
1179 pr_err("fill_inode badness %p %llx.%llx\n",
1180 in, ceph_vinop(in));
1181 goto done;
1182 }
1183 }
1184
1185done:
1186 dout("fill_trace done err=%d\n", err);
1187 return err;
1188}
1189
1190/*
1191 * Prepopulate our cache with readdir results, leases, etc.
1192 */
1193int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1194 struct ceph_mds_session *session)
1195{
1196 struct dentry *parent = req->r_dentry;
1197 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1198 struct qstr dname;
1199 struct dentry *dn;
1200 struct inode *in;
1201 int err = 0, i;
1202 struct inode *snapdir = NULL;
1203 struct ceph_mds_request_head *rhead = req->r_request->front.iov_base;
1204 u64 frag = le32_to_cpu(rhead->args.readdir.frag);
1205 struct ceph_dentry_info *di;
1206
1207 if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
1208 snapdir = ceph_get_snapdir(parent->d_inode);
1209 parent = d_find_alias(snapdir);
1210 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1211 rinfo->dir_nr, parent);
1212 } else {
1213 dout("readdir_prepopulate %d items under dn %p\n",
1214 rinfo->dir_nr, parent);
1215 if (rinfo->dir_dir)
1216 ceph_fill_dirfrag(parent->d_inode, rinfo->dir_dir);
1217 }
1218
1219 for (i = 0; i < rinfo->dir_nr; i++) {
1220 struct ceph_vino vino;
1221
1222 dname.name = rinfo->dir_dname[i];
1223 dname.len = rinfo->dir_dname_len[i];
1224 dname.hash = full_name_hash(dname.name, dname.len);
1225
1226 vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino);
1227 vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid);
1228
1229retry_lookup:
1230 dn = d_lookup(parent, &dname);
1231 dout("d_lookup on parent=%p name=%.*s got %p\n",
1232 parent, dname.len, dname.name, dn);
1233
1234 if (!dn) {
1235 dn = d_alloc(parent, &dname);
1236 dout("d_alloc %p '%.*s' = %p\n", parent,
1237 dname.len, dname.name, dn);
1238 if (dn == NULL) {
1239 dout("d_alloc badness\n");
1240 err = -ENOMEM;
1241 goto out;
1242 }
1243 err = ceph_init_dentry(dn);
1244 if (err < 0) {
1245 dput(dn);
1246 goto out;
1247 }
1248 } else if (dn->d_inode &&
1249 (ceph_ino(dn->d_inode) != vino.ino ||
1250 ceph_snap(dn->d_inode) != vino.snap)) {
1251 dout(" dn %p points to wrong inode %p\n",
1252 dn, dn->d_inode);
1253 d_delete(dn);
1254 dput(dn);
1255 goto retry_lookup;
1256 } else {
1257 /* reorder parent's d_subdirs */
1258 spin_lock(&parent->d_lock);
1259 spin_lock_nested(&dn->d_lock, DENTRY_D_LOCK_NESTED);
1260 list_move(&dn->d_u.d_child, &parent->d_subdirs);
1261 spin_unlock(&dn->d_lock);
1262 spin_unlock(&parent->d_lock);
1263 }
1264
1265 di = dn->d_fsdata;
1266 di->offset = ceph_make_fpos(frag, i + req->r_readdir_offset);
1267
1268 /* inode */
1269 if (dn->d_inode) {
1270 in = dn->d_inode;
1271 } else {
1272 in = ceph_get_inode(parent->d_sb, vino);
1273 if (IS_ERR(in)) {
1274 dout("new_inode badness\n");
1275 d_delete(dn);
1276 dput(dn);
1277 err = PTR_ERR(in);
1278 goto out;
1279 }
1280 dn = splice_dentry(dn, in, NULL, false);
1281 if (IS_ERR(dn))
1282 dn = NULL;
1283 }
1284
1285 if (fill_inode(in, &rinfo->dir_in[i], NULL, session,
1286 req->r_request_started, -1,
1287 &req->r_caps_reservation) < 0) {
1288 pr_err("fill_inode badness on %p\n", in);
1289 goto next_item;
1290 }
1291 if (dn)
1292 update_dentry_lease(dn, rinfo->dir_dlease[i],
1293 req->r_session,
1294 req->r_request_started);
1295next_item:
1296 if (dn)
1297 dput(dn);
1298 }
1299 req->r_did_prepopulate = true;
1300
1301out:
1302 if (snapdir) {
1303 iput(snapdir);
1304 dput(parent);
1305 }
1306 dout("readdir_prepopulate done\n");
1307 return err;
1308}
1309
1310int ceph_inode_set_size(struct inode *inode, loff_t size)
1311{
1312 struct ceph_inode_info *ci = ceph_inode(inode);
1313 int ret = 0;
1314
1315 spin_lock(&ci->i_ceph_lock);
1316 dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size);
1317 inode->i_size = size;
1318 inode->i_blocks = (size + (1 << 9) - 1) >> 9;
1319
1320 /* tell the MDS if we are approaching max_size */
1321 if ((size << 1) >= ci->i_max_size &&
1322 (ci->i_reported_size << 1) < ci->i_max_size)
1323 ret = 1;
1324
1325 spin_unlock(&ci->i_ceph_lock);
1326 return ret;
1327}
1328
1329/*
1330 * Write back inode data in a worker thread. (This can't be done
1331 * in the message handler context.)
1332 */
1333void ceph_queue_writeback(struct inode *inode)
1334{
1335 ihold(inode);
1336 if (queue_work(ceph_inode_to_client(inode)->wb_wq,
1337 &ceph_inode(inode)->i_wb_work)) {
1338 dout("ceph_queue_writeback %p\n", inode);
1339 } else {
1340 dout("ceph_queue_writeback %p failed\n", inode);
1341 iput(inode);
1342 }
1343}
1344
1345static void ceph_writeback_work(struct work_struct *work)
1346{
1347 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1348 i_wb_work);
1349 struct inode *inode = &ci->vfs_inode;
1350
1351 dout("writeback %p\n", inode);
1352 filemap_fdatawrite(&inode->i_data);
1353 iput(inode);
1354}
1355
1356/*
1357 * queue an async invalidation
1358 */
1359void ceph_queue_invalidate(struct inode *inode)
1360{
1361 ihold(inode);
1362 if (queue_work(ceph_inode_to_client(inode)->pg_inv_wq,
1363 &ceph_inode(inode)->i_pg_inv_work)) {
1364 dout("ceph_queue_invalidate %p\n", inode);
1365 } else {
1366 dout("ceph_queue_invalidate %p failed\n", inode);
1367 iput(inode);
1368 }
1369}
1370
1371/*
1372 * Invalidate inode pages in a worker thread. (This can't be done
1373 * in the message handler context.)
1374 */
1375static void ceph_invalidate_work(struct work_struct *work)
1376{
1377 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1378 i_pg_inv_work);
1379 struct inode *inode = &ci->vfs_inode;
1380 u32 orig_gen;
1381 int check = 0;
1382
1383 spin_lock(&ci->i_ceph_lock);
1384 dout("invalidate_pages %p gen %d revoking %d\n", inode,
1385 ci->i_rdcache_gen, ci->i_rdcache_revoking);
1386 if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
1387 /* nevermind! */
1388 spin_unlock(&ci->i_ceph_lock);
1389 goto out;
1390 }
1391 orig_gen = ci->i_rdcache_gen;
1392 spin_unlock(&ci->i_ceph_lock);
1393
1394 truncate_inode_pages(&inode->i_data, 0);
1395
1396 spin_lock(&ci->i_ceph_lock);
1397 if (orig_gen == ci->i_rdcache_gen &&
1398 orig_gen == ci->i_rdcache_revoking) {
1399 dout("invalidate_pages %p gen %d successful\n", inode,
1400 ci->i_rdcache_gen);
1401 ci->i_rdcache_revoking--;
1402 check = 1;
1403 } else {
1404 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1405 inode, orig_gen, ci->i_rdcache_gen,
1406 ci->i_rdcache_revoking);
1407 }
1408 spin_unlock(&ci->i_ceph_lock);
1409
1410 if (check)
1411 ceph_check_caps(ci, 0, NULL);
1412out:
1413 iput(inode);
1414}
1415
1416
1417/*
1418 * called by trunc_wq; take i_mutex ourselves
1419 *
1420 * We also truncate in a separate thread as well.
1421 */
1422static void ceph_vmtruncate_work(struct work_struct *work)
1423{
1424 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1425 i_vmtruncate_work);
1426 struct inode *inode = &ci->vfs_inode;
1427
1428 dout("vmtruncate_work %p\n", inode);
1429 mutex_lock(&inode->i_mutex);
1430 __ceph_do_pending_vmtruncate(inode);
1431 mutex_unlock(&inode->i_mutex);
1432 iput(inode);
1433}
1434
1435/*
1436 * Queue an async vmtruncate. If we fail to queue work, we will handle
1437 * the truncation the next time we call __ceph_do_pending_vmtruncate.
1438 */
1439void ceph_queue_vmtruncate(struct inode *inode)
1440{
1441 struct ceph_inode_info *ci = ceph_inode(inode);
1442
1443 ihold(inode);
1444 if (queue_work(ceph_sb_to_client(inode->i_sb)->trunc_wq,
1445 &ci->i_vmtruncate_work)) {
1446 dout("ceph_queue_vmtruncate %p\n", inode);
1447 } else {
1448 dout("ceph_queue_vmtruncate %p failed, pending=%d\n",
1449 inode, ci->i_truncate_pending);
1450 iput(inode);
1451 }
1452}
1453
1454/*
1455 * called with i_mutex held.
1456 *
1457 * Make sure any pending truncation is applied before doing anything
1458 * that may depend on it.
1459 */
1460void __ceph_do_pending_vmtruncate(struct inode *inode)
1461{
1462 struct ceph_inode_info *ci = ceph_inode(inode);
1463 u64 to;
1464 int wrbuffer_refs, wake = 0;
1465
1466retry:
1467 spin_lock(&ci->i_ceph_lock);
1468 if (ci->i_truncate_pending == 0) {
1469 dout("__do_pending_vmtruncate %p none pending\n", inode);
1470 spin_unlock(&ci->i_ceph_lock);
1471 return;
1472 }
1473
1474 /*
1475 * make sure any dirty snapped pages are flushed before we
1476 * possibly truncate them.. so write AND block!
1477 */
1478 if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) {
1479 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1480 inode);
1481 spin_unlock(&ci->i_ceph_lock);
1482 filemap_write_and_wait_range(&inode->i_data, 0,
1483 inode->i_sb->s_maxbytes);
1484 goto retry;
1485 }
1486
1487 to = ci->i_truncate_size;
1488 wrbuffer_refs = ci->i_wrbuffer_ref;
1489 dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode,
1490 ci->i_truncate_pending, to);
1491 spin_unlock(&ci->i_ceph_lock);
1492
1493 truncate_inode_pages(inode->i_mapping, to);
1494
1495 spin_lock(&ci->i_ceph_lock);
1496 ci->i_truncate_pending--;
1497 if (ci->i_truncate_pending == 0)
1498 wake = 1;
1499 spin_unlock(&ci->i_ceph_lock);
1500
1501 if (wrbuffer_refs == 0)
1502 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
1503 if (wake)
1504 wake_up_all(&ci->i_cap_wq);
1505}
1506
1507
1508/*
1509 * symlinks
1510 */
1511static void *ceph_sym_follow_link(struct dentry *dentry, struct nameidata *nd)
1512{
1513 struct ceph_inode_info *ci = ceph_inode(dentry->d_inode);
1514 nd_set_link(nd, ci->i_symlink);
1515 return NULL;
1516}
1517
1518static const struct inode_operations ceph_symlink_iops = {
1519 .readlink = generic_readlink,
1520 .follow_link = ceph_sym_follow_link,
1521};
1522
1523/*
1524 * setattr
1525 */
1526int ceph_setattr(struct dentry *dentry, struct iattr *attr)
1527{
1528 struct inode *inode = dentry->d_inode;
1529 struct ceph_inode_info *ci = ceph_inode(inode);
1530 struct inode *parent_inode;
1531 const unsigned int ia_valid = attr->ia_valid;
1532 struct ceph_mds_request *req;
1533 struct ceph_mds_client *mdsc = ceph_sb_to_client(dentry->d_sb)->mdsc;
1534 int issued;
1535 int release = 0, dirtied = 0;
1536 int mask = 0;
1537 int err = 0;
1538 int inode_dirty_flags = 0;
1539
1540 if (ceph_snap(inode) != CEPH_NOSNAP)
1541 return -EROFS;
1542
1543 __ceph_do_pending_vmtruncate(inode);
1544
1545 err = inode_change_ok(inode, attr);
1546 if (err != 0)
1547 return err;
1548
1549 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR,
1550 USE_AUTH_MDS);
1551 if (IS_ERR(req))
1552 return PTR_ERR(req);
1553
1554 spin_lock(&ci->i_ceph_lock);
1555 issued = __ceph_caps_issued(ci, NULL);
1556 dout("setattr %p issued %s\n", inode, ceph_cap_string(issued));
1557
1558 if (ia_valid & ATTR_UID) {
1559 dout("setattr %p uid %d -> %d\n", inode,
1560 inode->i_uid, attr->ia_uid);
1561 if (issued & CEPH_CAP_AUTH_EXCL) {
1562 inode->i_uid = attr->ia_uid;
1563 dirtied |= CEPH_CAP_AUTH_EXCL;
1564 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1565 attr->ia_uid != inode->i_uid) {
1566 req->r_args.setattr.uid = cpu_to_le32(attr->ia_uid);
1567 mask |= CEPH_SETATTR_UID;
1568 release |= CEPH_CAP_AUTH_SHARED;
1569 }
1570 }
1571 if (ia_valid & ATTR_GID) {
1572 dout("setattr %p gid %d -> %d\n", inode,
1573 inode->i_gid, attr->ia_gid);
1574 if (issued & CEPH_CAP_AUTH_EXCL) {
1575 inode->i_gid = attr->ia_gid;
1576 dirtied |= CEPH_CAP_AUTH_EXCL;
1577 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1578 attr->ia_gid != inode->i_gid) {
1579 req->r_args.setattr.gid = cpu_to_le32(attr->ia_gid);
1580 mask |= CEPH_SETATTR_GID;
1581 release |= CEPH_CAP_AUTH_SHARED;
1582 }
1583 }
1584 if (ia_valid & ATTR_MODE) {
1585 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode,
1586 attr->ia_mode);
1587 if (issued & CEPH_CAP_AUTH_EXCL) {
1588 inode->i_mode = attr->ia_mode;
1589 dirtied |= CEPH_CAP_AUTH_EXCL;
1590 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1591 attr->ia_mode != inode->i_mode) {
1592 req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode);
1593 mask |= CEPH_SETATTR_MODE;
1594 release |= CEPH_CAP_AUTH_SHARED;
1595 }
1596 }
1597
1598 if (ia_valid & ATTR_ATIME) {
1599 dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode,
1600 inode->i_atime.tv_sec, inode->i_atime.tv_nsec,
1601 attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec);
1602 if (issued & CEPH_CAP_FILE_EXCL) {
1603 ci->i_time_warp_seq++;
1604 inode->i_atime = attr->ia_atime;
1605 dirtied |= CEPH_CAP_FILE_EXCL;
1606 } else if ((issued & CEPH_CAP_FILE_WR) &&
1607 timespec_compare(&inode->i_atime,
1608 &attr->ia_atime) < 0) {
1609 inode->i_atime = attr->ia_atime;
1610 dirtied |= CEPH_CAP_FILE_WR;
1611 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1612 !timespec_equal(&inode->i_atime, &attr->ia_atime)) {
1613 ceph_encode_timespec(&req->r_args.setattr.atime,
1614 &attr->ia_atime);
1615 mask |= CEPH_SETATTR_ATIME;
1616 release |= CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_RD |
1617 CEPH_CAP_FILE_WR;
1618 }
1619 }
1620 if (ia_valid & ATTR_MTIME) {
1621 dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode,
1622 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
1623 attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec);
1624 if (issued & CEPH_CAP_FILE_EXCL) {
1625 ci->i_time_warp_seq++;
1626 inode->i_mtime = attr->ia_mtime;
1627 dirtied |= CEPH_CAP_FILE_EXCL;
1628 } else if ((issued & CEPH_CAP_FILE_WR) &&
1629 timespec_compare(&inode->i_mtime,
1630 &attr->ia_mtime) < 0) {
1631 inode->i_mtime = attr->ia_mtime;
1632 dirtied |= CEPH_CAP_FILE_WR;
1633 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1634 !timespec_equal(&inode->i_mtime, &attr->ia_mtime)) {
1635 ceph_encode_timespec(&req->r_args.setattr.mtime,
1636 &attr->ia_mtime);
1637 mask |= CEPH_SETATTR_MTIME;
1638 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
1639 CEPH_CAP_FILE_WR;
1640 }
1641 }
1642 if (ia_valid & ATTR_SIZE) {
1643 dout("setattr %p size %lld -> %lld\n", inode,
1644 inode->i_size, attr->ia_size);
1645 if (attr->ia_size > inode->i_sb->s_maxbytes) {
1646 err = -EINVAL;
1647 goto out;
1648 }
1649 if ((issued & CEPH_CAP_FILE_EXCL) &&
1650 attr->ia_size > inode->i_size) {
1651 inode->i_size = attr->ia_size;
1652 inode->i_blocks =
1653 (attr->ia_size + (1 << 9) - 1) >> 9;
1654 inode->i_ctime = attr->ia_ctime;
1655 ci->i_reported_size = attr->ia_size;
1656 dirtied |= CEPH_CAP_FILE_EXCL;
1657 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1658 attr->ia_size != inode->i_size) {
1659 req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
1660 req->r_args.setattr.old_size =
1661 cpu_to_le64(inode->i_size);
1662 mask |= CEPH_SETATTR_SIZE;
1663 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
1664 CEPH_CAP_FILE_WR;
1665 }
1666 }
1667
1668 /* these do nothing */
1669 if (ia_valid & ATTR_CTIME) {
1670 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME|
1671 ATTR_MODE|ATTR_UID|ATTR_GID)) == 0;
1672 dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode,
1673 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
1674 attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec,
1675 only ? "ctime only" : "ignored");
1676 inode->i_ctime = attr->ia_ctime;
1677 if (only) {
1678 /*
1679 * if kernel wants to dirty ctime but nothing else,
1680 * we need to choose a cap to dirty under, or do
1681 * a almost-no-op setattr
1682 */
1683 if (issued & CEPH_CAP_AUTH_EXCL)
1684 dirtied |= CEPH_CAP_AUTH_EXCL;
1685 else if (issued & CEPH_CAP_FILE_EXCL)
1686 dirtied |= CEPH_CAP_FILE_EXCL;
1687 else if (issued & CEPH_CAP_XATTR_EXCL)
1688 dirtied |= CEPH_CAP_XATTR_EXCL;
1689 else
1690 mask |= CEPH_SETATTR_CTIME;
1691 }
1692 }
1693 if (ia_valid & ATTR_FILE)
1694 dout("setattr %p ATTR_FILE ... hrm!\n", inode);
1695
1696 if (dirtied) {
1697 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied);
1698 inode->i_ctime = CURRENT_TIME;
1699 }
1700
1701 release &= issued;
1702 spin_unlock(&ci->i_ceph_lock);
1703
1704 if (inode_dirty_flags)
1705 __mark_inode_dirty(inode, inode_dirty_flags);
1706
1707 if (mask) {
1708 req->r_inode = inode;
1709 ihold(inode);
1710 req->r_inode_drop = release;
1711 req->r_args.setattr.mask = cpu_to_le32(mask);
1712 req->r_num_caps = 1;
1713 parent_inode = ceph_get_dentry_parent_inode(dentry);
1714 err = ceph_mdsc_do_request(mdsc, parent_inode, req);
1715 iput(parent_inode);
1716 }
1717 dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err,
1718 ceph_cap_string(dirtied), mask);
1719
1720 ceph_mdsc_put_request(req);
1721 __ceph_do_pending_vmtruncate(inode);
1722 return err;
1723out:
1724 spin_unlock(&ci->i_ceph_lock);
1725 ceph_mdsc_put_request(req);
1726 return err;
1727}
1728
1729/*
1730 * Verify that we have a lease on the given mask. If not,
1731 * do a getattr against an mds.
1732 */
1733int ceph_do_getattr(struct inode *inode, int mask)
1734{
1735 struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
1736 struct ceph_mds_client *mdsc = fsc->mdsc;
1737 struct ceph_mds_request *req;
1738 int err;
1739
1740 if (ceph_snap(inode) == CEPH_SNAPDIR) {
1741 dout("do_getattr inode %p SNAPDIR\n", inode);
1742 return 0;
1743 }
1744
1745 dout("do_getattr inode %p mask %s mode 0%o\n", inode, ceph_cap_string(mask), inode->i_mode);
1746 if (ceph_caps_issued_mask(ceph_inode(inode), mask, 1))
1747 return 0;
1748
1749 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
1750 if (IS_ERR(req))
1751 return PTR_ERR(req);
1752 req->r_inode = inode;
1753 ihold(inode);
1754 req->r_num_caps = 1;
1755 req->r_args.getattr.mask = cpu_to_le32(mask);
1756 err = ceph_mdsc_do_request(mdsc, NULL, req);
1757 ceph_mdsc_put_request(req);
1758 dout("do_getattr result=%d\n", err);
1759 return err;
1760}
1761
1762
1763/*
1764 * Check inode permissions. We verify we have a valid value for
1765 * the AUTH cap, then call the generic handler.
1766 */
1767int ceph_permission(struct inode *inode, int mask)
1768{
1769 int err;
1770
1771 if (mask & MAY_NOT_BLOCK)
1772 return -ECHILD;
1773
1774 err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED);
1775
1776 if (!err)
1777 err = generic_permission(inode, mask);
1778 return err;
1779}
1780
1781/*
1782 * Get all attributes. Hopefully somedata we'll have a statlite()
1783 * and can limit the fields we require to be accurate.
1784 */
1785int ceph_getattr(struct vfsmount *mnt, struct dentry *dentry,
1786 struct kstat *stat)
1787{
1788 struct inode *inode = dentry->d_inode;
1789 struct ceph_inode_info *ci = ceph_inode(inode);
1790 int err;
1791
1792 err = ceph_do_getattr(inode, CEPH_STAT_CAP_INODE_ALL);
1793 if (!err) {
1794 generic_fillattr(inode, stat);
1795 stat->ino = ceph_translate_ino(inode->i_sb, inode->i_ino);
1796 if (ceph_snap(inode) != CEPH_NOSNAP)
1797 stat->dev = ceph_snap(inode);
1798 else
1799 stat->dev = 0;
1800 if (S_ISDIR(inode->i_mode)) {
1801 if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb),
1802 RBYTES))
1803 stat->size = ci->i_rbytes;
1804 else
1805 stat->size = ci->i_files + ci->i_subdirs;
1806 stat->blocks = 0;
1807 stat->blksize = 65536;
1808 }
1809 }
1810 return err;
1811}
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
35static const struct inode_operations ceph_symlink_iops;
36
37static void ceph_inode_work(struct work_struct *work);
38
39/*
40 * find or create an inode, given the ceph ino number
41 */
42static int ceph_set_ino_cb(struct inode *inode, void *data)
43{
44 struct ceph_inode_info *ci = ceph_inode(inode);
45 struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb);
46
47 ci->i_vino = *(struct ceph_vino *)data;
48 inode->i_ino = ceph_vino_to_ino_t(ci->i_vino);
49 inode_set_iversion_raw(inode, 0);
50 percpu_counter_inc(&mdsc->metric.total_inodes);
51
52 return 0;
53}
54
55struct inode *ceph_get_inode(struct super_block *sb, struct ceph_vino vino)
56{
57 struct inode *inode;
58
59 if (ceph_vino_is_reserved(vino))
60 return ERR_PTR(-EREMOTEIO);
61
62 inode = iget5_locked(sb, (unsigned long)vino.ino, ceph_ino_compare,
63 ceph_set_ino_cb, &vino);
64 if (!inode)
65 return ERR_PTR(-ENOMEM);
66
67 dout("get_inode on %llu=%llx.%llx got %p new %d\n", ceph_present_inode(inode),
68 ceph_vinop(inode), inode, !!(inode->i_state & I_NEW));
69 return inode;
70}
71
72/*
73 * get/constuct snapdir inode for a given directory
74 */
75struct inode *ceph_get_snapdir(struct inode *parent)
76{
77 struct ceph_vino vino = {
78 .ino = ceph_ino(parent),
79 .snap = CEPH_SNAPDIR,
80 };
81 struct inode *inode = ceph_get_inode(parent->i_sb, vino);
82 struct ceph_inode_info *ci = ceph_inode(inode);
83
84 if (IS_ERR(inode))
85 return inode;
86
87 if (!S_ISDIR(parent->i_mode)) {
88 pr_warn_once("bad snapdir parent type (mode=0%o)\n",
89 parent->i_mode);
90 return ERR_PTR(-ENOTDIR);
91 }
92
93 if (!(inode->i_state & I_NEW) && !S_ISDIR(inode->i_mode)) {
94 pr_warn_once("bad snapdir inode type (mode=0%o)\n",
95 inode->i_mode);
96 return ERR_PTR(-ENOTDIR);
97 }
98
99 inode->i_mode = parent->i_mode;
100 inode->i_uid = parent->i_uid;
101 inode->i_gid = parent->i_gid;
102 inode->i_mtime = parent->i_mtime;
103 inode->i_ctime = parent->i_ctime;
104 inode->i_atime = parent->i_atime;
105 ci->i_rbytes = 0;
106 ci->i_btime = ceph_inode(parent)->i_btime;
107
108 if (inode->i_state & I_NEW) {
109 inode->i_op = &ceph_snapdir_iops;
110 inode->i_fop = &ceph_snapdir_fops;
111 ci->i_snap_caps = CEPH_CAP_PIN; /* so we can open */
112 unlock_new_inode(inode);
113 }
114
115 return inode;
116}
117
118const struct inode_operations ceph_file_iops = {
119 .permission = ceph_permission,
120 .setattr = ceph_setattr,
121 .getattr = ceph_getattr,
122 .listxattr = ceph_listxattr,
123 .get_acl = ceph_get_acl,
124 .set_acl = ceph_set_acl,
125};
126
127
128/*
129 * We use a 'frag tree' to keep track of the MDS's directory fragments
130 * for a given inode (usually there is just a single fragment). We
131 * need to know when a child frag is delegated to a new MDS, or when
132 * it is flagged as replicated, so we can direct our requests
133 * accordingly.
134 */
135
136/*
137 * find/create a frag in the tree
138 */
139static struct ceph_inode_frag *__get_or_create_frag(struct ceph_inode_info *ci,
140 u32 f)
141{
142 struct rb_node **p;
143 struct rb_node *parent = NULL;
144 struct ceph_inode_frag *frag;
145 int c;
146
147 p = &ci->i_fragtree.rb_node;
148 while (*p) {
149 parent = *p;
150 frag = rb_entry(parent, struct ceph_inode_frag, node);
151 c = ceph_frag_compare(f, frag->frag);
152 if (c < 0)
153 p = &(*p)->rb_left;
154 else if (c > 0)
155 p = &(*p)->rb_right;
156 else
157 return frag;
158 }
159
160 frag = kmalloc(sizeof(*frag), GFP_NOFS);
161 if (!frag)
162 return ERR_PTR(-ENOMEM);
163
164 frag->frag = f;
165 frag->split_by = 0;
166 frag->mds = -1;
167 frag->ndist = 0;
168
169 rb_link_node(&frag->node, parent, p);
170 rb_insert_color(&frag->node, &ci->i_fragtree);
171
172 dout("get_or_create_frag added %llx.%llx frag %x\n",
173 ceph_vinop(&ci->vfs_inode), f);
174 return frag;
175}
176
177/*
178 * find a specific frag @f
179 */
180struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci, u32 f)
181{
182 struct rb_node *n = ci->i_fragtree.rb_node;
183
184 while (n) {
185 struct ceph_inode_frag *frag =
186 rb_entry(n, struct ceph_inode_frag, node);
187 int c = ceph_frag_compare(f, frag->frag);
188 if (c < 0)
189 n = n->rb_left;
190 else if (c > 0)
191 n = n->rb_right;
192 else
193 return frag;
194 }
195 return NULL;
196}
197
198/*
199 * Choose frag containing the given value @v. If @pfrag is
200 * specified, copy the frag delegation info to the caller if
201 * it is present.
202 */
203static u32 __ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
204 struct ceph_inode_frag *pfrag, int *found)
205{
206 u32 t = ceph_frag_make(0, 0);
207 struct ceph_inode_frag *frag;
208 unsigned nway, i;
209 u32 n;
210
211 if (found)
212 *found = 0;
213
214 while (1) {
215 WARN_ON(!ceph_frag_contains_value(t, v));
216 frag = __ceph_find_frag(ci, t);
217 if (!frag)
218 break; /* t is a leaf */
219 if (frag->split_by == 0) {
220 if (pfrag)
221 memcpy(pfrag, frag, sizeof(*pfrag));
222 if (found)
223 *found = 1;
224 break;
225 }
226
227 /* choose child */
228 nway = 1 << frag->split_by;
229 dout("choose_frag(%x) %x splits by %d (%d ways)\n", v, t,
230 frag->split_by, nway);
231 for (i = 0; i < nway; i++) {
232 n = ceph_frag_make_child(t, frag->split_by, i);
233 if (ceph_frag_contains_value(n, v)) {
234 t = n;
235 break;
236 }
237 }
238 BUG_ON(i == nway);
239 }
240 dout("choose_frag(%x) = %x\n", v, t);
241
242 return t;
243}
244
245u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
246 struct ceph_inode_frag *pfrag, int *found)
247{
248 u32 ret;
249 mutex_lock(&ci->i_fragtree_mutex);
250 ret = __ceph_choose_frag(ci, v, pfrag, found);
251 mutex_unlock(&ci->i_fragtree_mutex);
252 return ret;
253}
254
255/*
256 * Process dirfrag (delegation) info from the mds. Include leaf
257 * fragment in tree ONLY if ndist > 0. Otherwise, only
258 * branches/splits are included in i_fragtree)
259 */
260static int ceph_fill_dirfrag(struct inode *inode,
261 struct ceph_mds_reply_dirfrag *dirinfo)
262{
263 struct ceph_inode_info *ci = ceph_inode(inode);
264 struct ceph_inode_frag *frag;
265 u32 id = le32_to_cpu(dirinfo->frag);
266 int mds = le32_to_cpu(dirinfo->auth);
267 int ndist = le32_to_cpu(dirinfo->ndist);
268 int diri_auth = -1;
269 int i;
270 int err = 0;
271
272 spin_lock(&ci->i_ceph_lock);
273 if (ci->i_auth_cap)
274 diri_auth = ci->i_auth_cap->mds;
275 spin_unlock(&ci->i_ceph_lock);
276
277 if (mds == -1) /* CDIR_AUTH_PARENT */
278 mds = diri_auth;
279
280 mutex_lock(&ci->i_fragtree_mutex);
281 if (ndist == 0 && mds == diri_auth) {
282 /* no delegation info needed. */
283 frag = __ceph_find_frag(ci, id);
284 if (!frag)
285 goto out;
286 if (frag->split_by == 0) {
287 /* tree leaf, remove */
288 dout("fill_dirfrag removed %llx.%llx frag %x"
289 " (no ref)\n", ceph_vinop(inode), id);
290 rb_erase(&frag->node, &ci->i_fragtree);
291 kfree(frag);
292 } else {
293 /* tree branch, keep and clear */
294 dout("fill_dirfrag cleared %llx.%llx frag %x"
295 " referral\n", ceph_vinop(inode), id);
296 frag->mds = -1;
297 frag->ndist = 0;
298 }
299 goto out;
300 }
301
302
303 /* find/add this frag to store mds delegation info */
304 frag = __get_or_create_frag(ci, id);
305 if (IS_ERR(frag)) {
306 /* this is not the end of the world; we can continue
307 with bad/inaccurate delegation info */
308 pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n",
309 ceph_vinop(inode), le32_to_cpu(dirinfo->frag));
310 err = -ENOMEM;
311 goto out;
312 }
313
314 frag->mds = mds;
315 frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP);
316 for (i = 0; i < frag->ndist; i++)
317 frag->dist[i] = le32_to_cpu(dirinfo->dist[i]);
318 dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n",
319 ceph_vinop(inode), frag->frag, frag->ndist);
320
321out:
322 mutex_unlock(&ci->i_fragtree_mutex);
323 return err;
324}
325
326static int frag_tree_split_cmp(const void *l, const void *r)
327{
328 struct ceph_frag_tree_split *ls = (struct ceph_frag_tree_split*)l;
329 struct ceph_frag_tree_split *rs = (struct ceph_frag_tree_split*)r;
330 return ceph_frag_compare(le32_to_cpu(ls->frag),
331 le32_to_cpu(rs->frag));
332}
333
334static bool is_frag_child(u32 f, struct ceph_inode_frag *frag)
335{
336 if (!frag)
337 return f == ceph_frag_make(0, 0);
338 if (ceph_frag_bits(f) != ceph_frag_bits(frag->frag) + frag->split_by)
339 return false;
340 return ceph_frag_contains_value(frag->frag, ceph_frag_value(f));
341}
342
343static int ceph_fill_fragtree(struct inode *inode,
344 struct ceph_frag_tree_head *fragtree,
345 struct ceph_mds_reply_dirfrag *dirinfo)
346{
347 struct ceph_inode_info *ci = ceph_inode(inode);
348 struct ceph_inode_frag *frag, *prev_frag = NULL;
349 struct rb_node *rb_node;
350 unsigned i, split_by, nsplits;
351 u32 id;
352 bool update = false;
353
354 mutex_lock(&ci->i_fragtree_mutex);
355 nsplits = le32_to_cpu(fragtree->nsplits);
356 if (nsplits != ci->i_fragtree_nsplits) {
357 update = true;
358 } else if (nsplits) {
359 i = prandom_u32() % nsplits;
360 id = le32_to_cpu(fragtree->splits[i].frag);
361 if (!__ceph_find_frag(ci, id))
362 update = true;
363 } else if (!RB_EMPTY_ROOT(&ci->i_fragtree)) {
364 rb_node = rb_first(&ci->i_fragtree);
365 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
366 if (frag->frag != ceph_frag_make(0, 0) || rb_next(rb_node))
367 update = true;
368 }
369 if (!update && dirinfo) {
370 id = le32_to_cpu(dirinfo->frag);
371 if (id != __ceph_choose_frag(ci, id, NULL, NULL))
372 update = true;
373 }
374 if (!update)
375 goto out_unlock;
376
377 if (nsplits > 1) {
378 sort(fragtree->splits, nsplits, sizeof(fragtree->splits[0]),
379 frag_tree_split_cmp, NULL);
380 }
381
382 dout("fill_fragtree %llx.%llx\n", ceph_vinop(inode));
383 rb_node = rb_first(&ci->i_fragtree);
384 for (i = 0; i < nsplits; i++) {
385 id = le32_to_cpu(fragtree->splits[i].frag);
386 split_by = le32_to_cpu(fragtree->splits[i].by);
387 if (split_by == 0 || ceph_frag_bits(id) + split_by > 24) {
388 pr_err("fill_fragtree %llx.%llx invalid split %d/%u, "
389 "frag %x split by %d\n", ceph_vinop(inode),
390 i, nsplits, id, split_by);
391 continue;
392 }
393 frag = NULL;
394 while (rb_node) {
395 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
396 if (ceph_frag_compare(frag->frag, id) >= 0) {
397 if (frag->frag != id)
398 frag = NULL;
399 else
400 rb_node = rb_next(rb_node);
401 break;
402 }
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 frag = NULL;
413 }
414 if (!frag) {
415 frag = __get_or_create_frag(ci, id);
416 if (IS_ERR(frag))
417 continue;
418 }
419 if (frag->split_by == 0)
420 ci->i_fragtree_nsplits++;
421 frag->split_by = split_by;
422 dout(" frag %x split by %d\n", frag->frag, frag->split_by);
423 prev_frag = frag;
424 }
425 while (rb_node) {
426 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
427 rb_node = rb_next(rb_node);
428 /* delete stale split/leaf node */
429 if (frag->split_by > 0 ||
430 !is_frag_child(frag->frag, prev_frag)) {
431 rb_erase(&frag->node, &ci->i_fragtree);
432 if (frag->split_by > 0)
433 ci->i_fragtree_nsplits--;
434 kfree(frag);
435 }
436 }
437out_unlock:
438 mutex_unlock(&ci->i_fragtree_mutex);
439 return 0;
440}
441
442/*
443 * initialize a newly allocated inode.
444 */
445struct inode *ceph_alloc_inode(struct super_block *sb)
446{
447 struct ceph_inode_info *ci;
448 int i;
449
450 ci = kmem_cache_alloc(ceph_inode_cachep, GFP_NOFS);
451 if (!ci)
452 return NULL;
453
454 dout("alloc_inode %p\n", &ci->vfs_inode);
455
456 spin_lock_init(&ci->i_ceph_lock);
457
458 ci->i_version = 0;
459 ci->i_inline_version = 0;
460 ci->i_time_warp_seq = 0;
461 ci->i_ceph_flags = 0;
462 atomic64_set(&ci->i_ordered_count, 1);
463 atomic64_set(&ci->i_release_count, 1);
464 atomic64_set(&ci->i_complete_seq[0], 0);
465 atomic64_set(&ci->i_complete_seq[1], 0);
466 ci->i_symlink = NULL;
467
468 ci->i_max_bytes = 0;
469 ci->i_max_files = 0;
470
471 memset(&ci->i_dir_layout, 0, sizeof(ci->i_dir_layout));
472 memset(&ci->i_cached_layout, 0, sizeof(ci->i_cached_layout));
473 RCU_INIT_POINTER(ci->i_layout.pool_ns, NULL);
474
475 ci->i_fragtree = RB_ROOT;
476 mutex_init(&ci->i_fragtree_mutex);
477
478 ci->i_xattrs.blob = NULL;
479 ci->i_xattrs.prealloc_blob = NULL;
480 ci->i_xattrs.dirty = false;
481 ci->i_xattrs.index = RB_ROOT;
482 ci->i_xattrs.count = 0;
483 ci->i_xattrs.names_size = 0;
484 ci->i_xattrs.vals_size = 0;
485 ci->i_xattrs.version = 0;
486 ci->i_xattrs.index_version = 0;
487
488 ci->i_caps = RB_ROOT;
489 ci->i_auth_cap = NULL;
490 ci->i_dirty_caps = 0;
491 ci->i_flushing_caps = 0;
492 INIT_LIST_HEAD(&ci->i_dirty_item);
493 INIT_LIST_HEAD(&ci->i_flushing_item);
494 ci->i_prealloc_cap_flush = NULL;
495 INIT_LIST_HEAD(&ci->i_cap_flush_list);
496 init_waitqueue_head(&ci->i_cap_wq);
497 ci->i_hold_caps_max = 0;
498 INIT_LIST_HEAD(&ci->i_cap_delay_list);
499 INIT_LIST_HEAD(&ci->i_cap_snaps);
500 ci->i_head_snapc = NULL;
501 ci->i_snap_caps = 0;
502
503 ci->i_last_rd = ci->i_last_wr = jiffies - 3600 * HZ;
504 for (i = 0; i < CEPH_FILE_MODE_BITS; i++)
505 ci->i_nr_by_mode[i] = 0;
506
507 mutex_init(&ci->i_truncate_mutex);
508 ci->i_truncate_seq = 0;
509 ci->i_truncate_size = 0;
510 ci->i_truncate_pending = 0;
511
512 ci->i_max_size = 0;
513 ci->i_reported_size = 0;
514 ci->i_wanted_max_size = 0;
515 ci->i_requested_max_size = 0;
516
517 ci->i_pin_ref = 0;
518 ci->i_rd_ref = 0;
519 ci->i_rdcache_ref = 0;
520 ci->i_wr_ref = 0;
521 ci->i_wb_ref = 0;
522 ci->i_fx_ref = 0;
523 ci->i_wrbuffer_ref = 0;
524 ci->i_wrbuffer_ref_head = 0;
525 atomic_set(&ci->i_filelock_ref, 0);
526 atomic_set(&ci->i_shared_gen, 1);
527 ci->i_rdcache_gen = 0;
528 ci->i_rdcache_revoking = 0;
529
530 INIT_LIST_HEAD(&ci->i_unsafe_dirops);
531 INIT_LIST_HEAD(&ci->i_unsafe_iops);
532 spin_lock_init(&ci->i_unsafe_lock);
533
534 ci->i_snap_realm = NULL;
535 INIT_LIST_HEAD(&ci->i_snap_realm_item);
536 INIT_LIST_HEAD(&ci->i_snap_flush_item);
537
538 INIT_WORK(&ci->i_work, ceph_inode_work);
539 ci->i_work_mask = 0;
540 memset(&ci->i_btime, '\0', sizeof(ci->i_btime));
541
542 ceph_fscache_inode_init(ci);
543
544 return &ci->vfs_inode;
545}
546
547void ceph_free_inode(struct inode *inode)
548{
549 struct ceph_inode_info *ci = ceph_inode(inode);
550
551 kfree(ci->i_symlink);
552 kmem_cache_free(ceph_inode_cachep, ci);
553}
554
555void ceph_evict_inode(struct inode *inode)
556{
557 struct ceph_inode_info *ci = ceph_inode(inode);
558 struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb);
559 struct ceph_inode_frag *frag;
560 struct rb_node *n;
561
562 dout("evict_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode));
563
564 percpu_counter_dec(&mdsc->metric.total_inodes);
565
566 truncate_inode_pages_final(&inode->i_data);
567 clear_inode(inode);
568
569 ceph_fscache_unregister_inode_cookie(ci);
570
571 __ceph_remove_caps(ci);
572
573 if (__ceph_has_any_quota(ci))
574 ceph_adjust_quota_realms_count(inode, false);
575
576 /*
577 * we may still have a snap_realm reference if there are stray
578 * caps in i_snap_caps.
579 */
580 if (ci->i_snap_realm) {
581 if (ceph_snap(inode) == CEPH_NOSNAP) {
582 struct ceph_snap_realm *realm = ci->i_snap_realm;
583 dout(" dropping residual ref to snap realm %p\n",
584 realm);
585 spin_lock(&realm->inodes_with_caps_lock);
586 list_del_init(&ci->i_snap_realm_item);
587 ci->i_snap_realm = NULL;
588 if (realm->ino == ci->i_vino.ino)
589 realm->inode = NULL;
590 spin_unlock(&realm->inodes_with_caps_lock);
591 ceph_put_snap_realm(mdsc, realm);
592 } else {
593 ceph_put_snapid_map(mdsc, ci->i_snapid_map);
594 ci->i_snap_realm = NULL;
595 }
596 }
597
598 while ((n = rb_first(&ci->i_fragtree)) != NULL) {
599 frag = rb_entry(n, struct ceph_inode_frag, node);
600 rb_erase(n, &ci->i_fragtree);
601 kfree(frag);
602 }
603 ci->i_fragtree_nsplits = 0;
604
605 __ceph_destroy_xattrs(ci);
606 if (ci->i_xattrs.blob)
607 ceph_buffer_put(ci->i_xattrs.blob);
608 if (ci->i_xattrs.prealloc_blob)
609 ceph_buffer_put(ci->i_xattrs.prealloc_blob);
610
611 ceph_put_string(rcu_dereference_raw(ci->i_layout.pool_ns));
612 ceph_put_string(rcu_dereference_raw(ci->i_cached_layout.pool_ns));
613}
614
615static inline blkcnt_t calc_inode_blocks(u64 size)
616{
617 return (size + (1<<9) - 1) >> 9;
618}
619
620/*
621 * Helpers to fill in size, ctime, mtime, and atime. We have to be
622 * careful because either the client or MDS may have more up to date
623 * info, depending on which capabilities are held, and whether
624 * time_warp_seq or truncate_seq have increased. (Ordinarily, mtime
625 * and size are monotonically increasing, except when utimes() or
626 * truncate() increments the corresponding _seq values.)
627 */
628int ceph_fill_file_size(struct inode *inode, int issued,
629 u32 truncate_seq, u64 truncate_size, u64 size)
630{
631 struct ceph_inode_info *ci = ceph_inode(inode);
632 int queue_trunc = 0;
633 loff_t isize = i_size_read(inode);
634
635 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 ||
636 (truncate_seq == ci->i_truncate_seq && size > isize)) {
637 dout("size %lld -> %llu\n", isize, size);
638 if (size > 0 && S_ISDIR(inode->i_mode)) {
639 pr_err("fill_file_size non-zero size for directory\n");
640 size = 0;
641 }
642 i_size_write(inode, size);
643 inode->i_blocks = calc_inode_blocks(size);
644 ci->i_reported_size = size;
645 if (truncate_seq != ci->i_truncate_seq) {
646 dout("truncate_seq %u -> %u\n",
647 ci->i_truncate_seq, truncate_seq);
648 ci->i_truncate_seq = truncate_seq;
649
650 /* the MDS should have revoked these caps */
651 WARN_ON_ONCE(issued & (CEPH_CAP_FILE_EXCL |
652 CEPH_CAP_FILE_RD |
653 CEPH_CAP_FILE_WR |
654 CEPH_CAP_FILE_LAZYIO));
655 /*
656 * If we hold relevant caps, or in the case where we're
657 * not the only client referencing this file and we
658 * don't hold those caps, then we need to check whether
659 * the file is either opened or mmaped
660 */
661 if ((issued & (CEPH_CAP_FILE_CACHE|
662 CEPH_CAP_FILE_BUFFER)) ||
663 mapping_mapped(inode->i_mapping) ||
664 __ceph_is_file_opened(ci)) {
665 ci->i_truncate_pending++;
666 queue_trunc = 1;
667 }
668 }
669 }
670 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 &&
671 ci->i_truncate_size != truncate_size) {
672 dout("truncate_size %lld -> %llu\n", ci->i_truncate_size,
673 truncate_size);
674 ci->i_truncate_size = truncate_size;
675 }
676
677 if (queue_trunc)
678 ceph_fscache_invalidate(inode);
679
680 return queue_trunc;
681}
682
683void ceph_fill_file_time(struct inode *inode, int issued,
684 u64 time_warp_seq, struct timespec64 *ctime,
685 struct timespec64 *mtime, struct timespec64 *atime)
686{
687 struct ceph_inode_info *ci = ceph_inode(inode);
688 int warn = 0;
689
690 if (issued & (CEPH_CAP_FILE_EXCL|
691 CEPH_CAP_FILE_WR|
692 CEPH_CAP_FILE_BUFFER|
693 CEPH_CAP_AUTH_EXCL|
694 CEPH_CAP_XATTR_EXCL)) {
695 if (ci->i_version == 0 ||
696 timespec64_compare(ctime, &inode->i_ctime) > 0) {
697 dout("ctime %lld.%09ld -> %lld.%09ld inc w/ cap\n",
698 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
699 ctime->tv_sec, ctime->tv_nsec);
700 inode->i_ctime = *ctime;
701 }
702 if (ci->i_version == 0 ||
703 ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) {
704 /* the MDS did a utimes() */
705 dout("mtime %lld.%09ld -> %lld.%09ld "
706 "tw %d -> %d\n",
707 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
708 mtime->tv_sec, mtime->tv_nsec,
709 ci->i_time_warp_seq, (int)time_warp_seq);
710
711 inode->i_mtime = *mtime;
712 inode->i_atime = *atime;
713 ci->i_time_warp_seq = time_warp_seq;
714 } else if (time_warp_seq == ci->i_time_warp_seq) {
715 /* nobody did utimes(); take the max */
716 if (timespec64_compare(mtime, &inode->i_mtime) > 0) {
717 dout("mtime %lld.%09ld -> %lld.%09ld inc\n",
718 inode->i_mtime.tv_sec,
719 inode->i_mtime.tv_nsec,
720 mtime->tv_sec, mtime->tv_nsec);
721 inode->i_mtime = *mtime;
722 }
723 if (timespec64_compare(atime, &inode->i_atime) > 0) {
724 dout("atime %lld.%09ld -> %lld.%09ld inc\n",
725 inode->i_atime.tv_sec,
726 inode->i_atime.tv_nsec,
727 atime->tv_sec, atime->tv_nsec);
728 inode->i_atime = *atime;
729 }
730 } else if (issued & CEPH_CAP_FILE_EXCL) {
731 /* we did a utimes(); ignore mds values */
732 } else {
733 warn = 1;
734 }
735 } else {
736 /* we have no write|excl caps; whatever the MDS says is true */
737 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) {
738 inode->i_ctime = *ctime;
739 inode->i_mtime = *mtime;
740 inode->i_atime = *atime;
741 ci->i_time_warp_seq = time_warp_seq;
742 } else {
743 warn = 1;
744 }
745 }
746 if (warn) /* time_warp_seq shouldn't go backwards */
747 dout("%p mds time_warp_seq %llu < %u\n",
748 inode, time_warp_seq, ci->i_time_warp_seq);
749}
750
751/*
752 * Populate an inode based on info from mds. May be called on new or
753 * existing inodes.
754 */
755int ceph_fill_inode(struct inode *inode, struct page *locked_page,
756 struct ceph_mds_reply_info_in *iinfo,
757 struct ceph_mds_reply_dirfrag *dirinfo,
758 struct ceph_mds_session *session, int cap_fmode,
759 struct ceph_cap_reservation *caps_reservation)
760{
761 struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb);
762 struct ceph_mds_reply_inode *info = iinfo->in;
763 struct ceph_inode_info *ci = ceph_inode(inode);
764 int issued, new_issued, info_caps;
765 struct timespec64 mtime, atime, ctime;
766 struct ceph_buffer *xattr_blob = NULL;
767 struct ceph_buffer *old_blob = NULL;
768 struct ceph_string *pool_ns = NULL;
769 struct ceph_cap *new_cap = NULL;
770 int err = 0;
771 bool wake = false;
772 bool queue_trunc = false;
773 bool new_version = false;
774 bool fill_inline = false;
775 umode_t mode = le32_to_cpu(info->mode);
776 dev_t rdev = le32_to_cpu(info->rdev);
777
778 lockdep_assert_held(&mdsc->snap_rwsem);
779
780 dout("%s %p ino %llx.%llx v %llu had %llu\n", __func__,
781 inode, ceph_vinop(inode), le64_to_cpu(info->version),
782 ci->i_version);
783
784 /* Once I_NEW is cleared, we can't change type or dev numbers */
785 if (inode->i_state & I_NEW) {
786 inode->i_mode = mode;
787 } else {
788 if (inode_wrong_type(inode, mode)) {
789 pr_warn_once("inode type changed! (ino %llx.%llx is 0%o, mds says 0%o)\n",
790 ceph_vinop(inode), inode->i_mode, mode);
791 return -ESTALE;
792 }
793
794 if ((S_ISCHR(mode) || S_ISBLK(mode)) && inode->i_rdev != rdev) {
795 pr_warn_once("dev inode rdev changed! (ino %llx.%llx is %u:%u, mds says %u:%u)\n",
796 ceph_vinop(inode), MAJOR(inode->i_rdev),
797 MINOR(inode->i_rdev), MAJOR(rdev),
798 MINOR(rdev));
799 return -ESTALE;
800 }
801 }
802
803 info_caps = le32_to_cpu(info->cap.caps);
804
805 /* prealloc new cap struct */
806 if (info_caps && ceph_snap(inode) == CEPH_NOSNAP) {
807 new_cap = ceph_get_cap(mdsc, caps_reservation);
808 if (!new_cap)
809 return -ENOMEM;
810 }
811
812 /*
813 * prealloc xattr data, if it looks like we'll need it. only
814 * if len > 4 (meaning there are actually xattrs; the first 4
815 * bytes are the xattr count).
816 */
817 if (iinfo->xattr_len > 4) {
818 xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS);
819 if (!xattr_blob)
820 pr_err("%s ENOMEM xattr blob %d bytes\n", __func__,
821 iinfo->xattr_len);
822 }
823
824 if (iinfo->pool_ns_len > 0)
825 pool_ns = ceph_find_or_create_string(iinfo->pool_ns_data,
826 iinfo->pool_ns_len);
827
828 if (ceph_snap(inode) != CEPH_NOSNAP && !ci->i_snapid_map)
829 ci->i_snapid_map = ceph_get_snapid_map(mdsc, ceph_snap(inode));
830
831 spin_lock(&ci->i_ceph_lock);
832
833 /*
834 * provided version will be odd if inode value is projected,
835 * even if stable. skip the update if we have newer stable
836 * info (ours>=theirs, e.g. due to racing mds replies), unless
837 * we are getting projected (unstable) info (in which case the
838 * version is odd, and we want ours>theirs).
839 * us them
840 * 2 2 skip
841 * 3 2 skip
842 * 3 3 update
843 */
844 if (ci->i_version == 0 ||
845 ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
846 le64_to_cpu(info->version) > (ci->i_version & ~1)))
847 new_version = true;
848
849 /* Update change_attribute */
850 inode_set_max_iversion_raw(inode, iinfo->change_attr);
851
852 __ceph_caps_issued(ci, &issued);
853 issued |= __ceph_caps_dirty(ci);
854 new_issued = ~issued & info_caps;
855
856 /* directories have fl_stripe_unit set to zero */
857 if (le32_to_cpu(info->layout.fl_stripe_unit))
858 inode->i_blkbits =
859 fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1;
860 else
861 inode->i_blkbits = CEPH_BLOCK_SHIFT;
862
863 __ceph_update_quota(ci, iinfo->max_bytes, iinfo->max_files);
864
865 if ((new_version || (new_issued & CEPH_CAP_AUTH_SHARED)) &&
866 (issued & CEPH_CAP_AUTH_EXCL) == 0) {
867 inode->i_mode = mode;
868 inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(info->uid));
869 inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(info->gid));
870 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
871 from_kuid(&init_user_ns, inode->i_uid),
872 from_kgid(&init_user_ns, inode->i_gid));
873 ceph_decode_timespec64(&ci->i_btime, &iinfo->btime);
874 ceph_decode_timespec64(&ci->i_snap_btime, &iinfo->snap_btime);
875 }
876
877 if ((new_version || (new_issued & CEPH_CAP_LINK_SHARED)) &&
878 (issued & CEPH_CAP_LINK_EXCL) == 0)
879 set_nlink(inode, le32_to_cpu(info->nlink));
880
881 if (new_version || (new_issued & CEPH_CAP_ANY_RD)) {
882 /* be careful with mtime, atime, size */
883 ceph_decode_timespec64(&atime, &info->atime);
884 ceph_decode_timespec64(&mtime, &info->mtime);
885 ceph_decode_timespec64(&ctime, &info->ctime);
886 ceph_fill_file_time(inode, issued,
887 le32_to_cpu(info->time_warp_seq),
888 &ctime, &mtime, &atime);
889 }
890
891 if (new_version || (info_caps & CEPH_CAP_FILE_SHARED)) {
892 ci->i_files = le64_to_cpu(info->files);
893 ci->i_subdirs = le64_to_cpu(info->subdirs);
894 }
895
896 if (new_version ||
897 (new_issued & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR))) {
898 s64 old_pool = ci->i_layout.pool_id;
899 struct ceph_string *old_ns;
900
901 ceph_file_layout_from_legacy(&ci->i_layout, &info->layout);
902 old_ns = rcu_dereference_protected(ci->i_layout.pool_ns,
903 lockdep_is_held(&ci->i_ceph_lock));
904 rcu_assign_pointer(ci->i_layout.pool_ns, pool_ns);
905
906 if (ci->i_layout.pool_id != old_pool || pool_ns != old_ns)
907 ci->i_ceph_flags &= ~CEPH_I_POOL_PERM;
908
909 pool_ns = old_ns;
910
911 queue_trunc = ceph_fill_file_size(inode, issued,
912 le32_to_cpu(info->truncate_seq),
913 le64_to_cpu(info->truncate_size),
914 le64_to_cpu(info->size));
915 /* only update max_size on auth cap */
916 if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
917 ci->i_max_size != le64_to_cpu(info->max_size)) {
918 dout("max_size %lld -> %llu\n", ci->i_max_size,
919 le64_to_cpu(info->max_size));
920 ci->i_max_size = le64_to_cpu(info->max_size);
921 }
922 }
923
924 /* layout and rstat are not tracked by capability, update them if
925 * the inode info is from auth mds */
926 if (new_version || (info->cap.flags & CEPH_CAP_FLAG_AUTH)) {
927 if (S_ISDIR(inode->i_mode)) {
928 ci->i_dir_layout = iinfo->dir_layout;
929 ci->i_rbytes = le64_to_cpu(info->rbytes);
930 ci->i_rfiles = le64_to_cpu(info->rfiles);
931 ci->i_rsubdirs = le64_to_cpu(info->rsubdirs);
932 ci->i_dir_pin = iinfo->dir_pin;
933 ci->i_rsnaps = iinfo->rsnaps;
934 ceph_decode_timespec64(&ci->i_rctime, &info->rctime);
935 }
936 }
937
938 /* xattrs */
939 /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
940 if ((ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL)) &&
941 le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) {
942 if (ci->i_xattrs.blob)
943 old_blob = ci->i_xattrs.blob;
944 ci->i_xattrs.blob = xattr_blob;
945 if (xattr_blob)
946 memcpy(ci->i_xattrs.blob->vec.iov_base,
947 iinfo->xattr_data, iinfo->xattr_len);
948 ci->i_xattrs.version = le64_to_cpu(info->xattr_version);
949 ceph_forget_all_cached_acls(inode);
950 ceph_security_invalidate_secctx(inode);
951 xattr_blob = NULL;
952 }
953
954 /* finally update i_version */
955 if (le64_to_cpu(info->version) > ci->i_version)
956 ci->i_version = le64_to_cpu(info->version);
957
958 inode->i_mapping->a_ops = &ceph_aops;
959
960 switch (inode->i_mode & S_IFMT) {
961 case S_IFIFO:
962 case S_IFBLK:
963 case S_IFCHR:
964 case S_IFSOCK:
965 inode->i_blkbits = PAGE_SHIFT;
966 init_special_inode(inode, inode->i_mode, rdev);
967 inode->i_op = &ceph_file_iops;
968 break;
969 case S_IFREG:
970 inode->i_op = &ceph_file_iops;
971 inode->i_fop = &ceph_file_fops;
972 break;
973 case S_IFLNK:
974 inode->i_op = &ceph_symlink_iops;
975 if (!ci->i_symlink) {
976 u32 symlen = iinfo->symlink_len;
977 char *sym;
978
979 spin_unlock(&ci->i_ceph_lock);
980
981 if (symlen != i_size_read(inode)) {
982 pr_err("%s %llx.%llx BAD symlink "
983 "size %lld\n", __func__,
984 ceph_vinop(inode),
985 i_size_read(inode));
986 i_size_write(inode, symlen);
987 inode->i_blocks = calc_inode_blocks(symlen);
988 }
989
990 err = -ENOMEM;
991 sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS);
992 if (!sym)
993 goto out;
994
995 spin_lock(&ci->i_ceph_lock);
996 if (!ci->i_symlink)
997 ci->i_symlink = sym;
998 else
999 kfree(sym); /* lost a race */
1000 }
1001 inode->i_link = ci->i_symlink;
1002 break;
1003 case S_IFDIR:
1004 inode->i_op = &ceph_dir_iops;
1005 inode->i_fop = &ceph_dir_fops;
1006 break;
1007 default:
1008 pr_err("%s %llx.%llx BAD mode 0%o\n", __func__,
1009 ceph_vinop(inode), inode->i_mode);
1010 }
1011
1012 /* were we issued a capability? */
1013 if (info_caps) {
1014 if (ceph_snap(inode) == CEPH_NOSNAP) {
1015 ceph_add_cap(inode, session,
1016 le64_to_cpu(info->cap.cap_id),
1017 info_caps,
1018 le32_to_cpu(info->cap.wanted),
1019 le32_to_cpu(info->cap.seq),
1020 le32_to_cpu(info->cap.mseq),
1021 le64_to_cpu(info->cap.realm),
1022 info->cap.flags, &new_cap);
1023
1024 /* set dir completion flag? */
1025 if (S_ISDIR(inode->i_mode) &&
1026 ci->i_files == 0 && ci->i_subdirs == 0 &&
1027 (info_caps & CEPH_CAP_FILE_SHARED) &&
1028 (issued & CEPH_CAP_FILE_EXCL) == 0 &&
1029 !__ceph_dir_is_complete(ci)) {
1030 dout(" marking %p complete (empty)\n", inode);
1031 i_size_write(inode, 0);
1032 __ceph_dir_set_complete(ci,
1033 atomic64_read(&ci->i_release_count),
1034 atomic64_read(&ci->i_ordered_count));
1035 }
1036
1037 wake = true;
1038 } else {
1039 dout(" %p got snap_caps %s\n", inode,
1040 ceph_cap_string(info_caps));
1041 ci->i_snap_caps |= info_caps;
1042 }
1043 }
1044
1045 if (iinfo->inline_version > 0 &&
1046 iinfo->inline_version >= ci->i_inline_version) {
1047 int cache_caps = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
1048 ci->i_inline_version = iinfo->inline_version;
1049 if (ci->i_inline_version != CEPH_INLINE_NONE &&
1050 (locked_page || (info_caps & cache_caps)))
1051 fill_inline = true;
1052 }
1053
1054 if (cap_fmode >= 0) {
1055 if (!info_caps)
1056 pr_warn("mds issued no caps on %llx.%llx\n",
1057 ceph_vinop(inode));
1058 __ceph_touch_fmode(ci, mdsc, cap_fmode);
1059 }
1060
1061 spin_unlock(&ci->i_ceph_lock);
1062
1063 if (fill_inline)
1064 ceph_fill_inline_data(inode, locked_page,
1065 iinfo->inline_data, iinfo->inline_len);
1066
1067 if (wake)
1068 wake_up_all(&ci->i_cap_wq);
1069
1070 /* queue truncate if we saw i_size decrease */
1071 if (queue_trunc)
1072 ceph_queue_vmtruncate(inode);
1073
1074 /* populate frag tree */
1075 if (S_ISDIR(inode->i_mode))
1076 ceph_fill_fragtree(inode, &info->fragtree, dirinfo);
1077
1078 /* update delegation info? */
1079 if (dirinfo)
1080 ceph_fill_dirfrag(inode, dirinfo);
1081
1082 err = 0;
1083out:
1084 if (new_cap)
1085 ceph_put_cap(mdsc, new_cap);
1086 ceph_buffer_put(old_blob);
1087 ceph_buffer_put(xattr_blob);
1088 ceph_put_string(pool_ns);
1089 return err;
1090}
1091
1092/*
1093 * caller should hold session s_mutex and dentry->d_lock.
1094 */
1095static void __update_dentry_lease(struct inode *dir, struct dentry *dentry,
1096 struct ceph_mds_reply_lease *lease,
1097 struct ceph_mds_session *session,
1098 unsigned long from_time,
1099 struct ceph_mds_session **old_lease_session)
1100{
1101 struct ceph_dentry_info *di = ceph_dentry(dentry);
1102 unsigned mask = le16_to_cpu(lease->mask);
1103 long unsigned duration = le32_to_cpu(lease->duration_ms);
1104 long unsigned ttl = from_time + (duration * HZ) / 1000;
1105 long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000;
1106
1107 dout("update_dentry_lease %p duration %lu ms ttl %lu\n",
1108 dentry, duration, ttl);
1109
1110 /* only track leases on regular dentries */
1111 if (ceph_snap(dir) != CEPH_NOSNAP)
1112 return;
1113
1114 if (mask & CEPH_LEASE_PRIMARY_LINK)
1115 di->flags |= CEPH_DENTRY_PRIMARY_LINK;
1116 else
1117 di->flags &= ~CEPH_DENTRY_PRIMARY_LINK;
1118
1119 di->lease_shared_gen = atomic_read(&ceph_inode(dir)->i_shared_gen);
1120 if (!(mask & CEPH_LEASE_VALID)) {
1121 __ceph_dentry_dir_lease_touch(di);
1122 return;
1123 }
1124
1125 if (di->lease_gen == atomic_read(&session->s_cap_gen) &&
1126 time_before(ttl, di->time))
1127 return; /* we already have a newer lease. */
1128
1129 if (di->lease_session && di->lease_session != session) {
1130 *old_lease_session = di->lease_session;
1131 di->lease_session = NULL;
1132 }
1133
1134 if (!di->lease_session)
1135 di->lease_session = ceph_get_mds_session(session);
1136 di->lease_gen = atomic_read(&session->s_cap_gen);
1137 di->lease_seq = le32_to_cpu(lease->seq);
1138 di->lease_renew_after = half_ttl;
1139 di->lease_renew_from = 0;
1140 di->time = ttl;
1141
1142 __ceph_dentry_lease_touch(di);
1143}
1144
1145static inline void update_dentry_lease(struct inode *dir, struct dentry *dentry,
1146 struct ceph_mds_reply_lease *lease,
1147 struct ceph_mds_session *session,
1148 unsigned long from_time)
1149{
1150 struct ceph_mds_session *old_lease_session = NULL;
1151 spin_lock(&dentry->d_lock);
1152 __update_dentry_lease(dir, dentry, lease, session, from_time,
1153 &old_lease_session);
1154 spin_unlock(&dentry->d_lock);
1155 ceph_put_mds_session(old_lease_session);
1156}
1157
1158/*
1159 * update dentry lease without having parent inode locked
1160 */
1161static void update_dentry_lease_careful(struct dentry *dentry,
1162 struct ceph_mds_reply_lease *lease,
1163 struct ceph_mds_session *session,
1164 unsigned long from_time,
1165 char *dname, u32 dname_len,
1166 struct ceph_vino *pdvino,
1167 struct ceph_vino *ptvino)
1168
1169{
1170 struct inode *dir;
1171 struct ceph_mds_session *old_lease_session = NULL;
1172
1173 spin_lock(&dentry->d_lock);
1174 /* make sure dentry's name matches target */
1175 if (dentry->d_name.len != dname_len ||
1176 memcmp(dentry->d_name.name, dname, dname_len))
1177 goto out_unlock;
1178
1179 dir = d_inode(dentry->d_parent);
1180 /* make sure parent matches dvino */
1181 if (!ceph_ino_compare(dir, pdvino))
1182 goto out_unlock;
1183
1184 /* make sure dentry's inode matches target. NULL ptvino means that
1185 * we expect a negative dentry */
1186 if (ptvino) {
1187 if (d_really_is_negative(dentry))
1188 goto out_unlock;
1189 if (!ceph_ino_compare(d_inode(dentry), ptvino))
1190 goto out_unlock;
1191 } else {
1192 if (d_really_is_positive(dentry))
1193 goto out_unlock;
1194 }
1195
1196 __update_dentry_lease(dir, dentry, lease, session,
1197 from_time, &old_lease_session);
1198out_unlock:
1199 spin_unlock(&dentry->d_lock);
1200 ceph_put_mds_session(old_lease_session);
1201}
1202
1203/*
1204 * splice a dentry to an inode.
1205 * caller must hold directory i_mutex for this to be safe.
1206 */
1207static int splice_dentry(struct dentry **pdn, struct inode *in)
1208{
1209 struct dentry *dn = *pdn;
1210 struct dentry *realdn;
1211
1212 BUG_ON(d_inode(dn));
1213
1214 if (S_ISDIR(in->i_mode)) {
1215 /* If inode is directory, d_splice_alias() below will remove
1216 * 'realdn' from its origin parent. We need to ensure that
1217 * origin parent's readdir cache will not reference 'realdn'
1218 */
1219 realdn = d_find_any_alias(in);
1220 if (realdn) {
1221 struct ceph_dentry_info *di = ceph_dentry(realdn);
1222 spin_lock(&realdn->d_lock);
1223
1224 realdn->d_op->d_prune(realdn);
1225
1226 di->time = jiffies;
1227 di->lease_shared_gen = 0;
1228 di->offset = 0;
1229
1230 spin_unlock(&realdn->d_lock);
1231 dput(realdn);
1232 }
1233 }
1234
1235 /* dn must be unhashed */
1236 if (!d_unhashed(dn))
1237 d_drop(dn);
1238 realdn = d_splice_alias(in, dn);
1239 if (IS_ERR(realdn)) {
1240 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
1241 PTR_ERR(realdn), dn, in, ceph_vinop(in));
1242 return PTR_ERR(realdn);
1243 }
1244
1245 if (realdn) {
1246 dout("dn %p (%d) spliced with %p (%d) "
1247 "inode %p ino %llx.%llx\n",
1248 dn, d_count(dn),
1249 realdn, d_count(realdn),
1250 d_inode(realdn), ceph_vinop(d_inode(realdn)));
1251 dput(dn);
1252 *pdn = realdn;
1253 } else {
1254 BUG_ON(!ceph_dentry(dn));
1255 dout("dn %p attached to %p ino %llx.%llx\n",
1256 dn, d_inode(dn), ceph_vinop(d_inode(dn)));
1257 }
1258 return 0;
1259}
1260
1261/*
1262 * Incorporate results into the local cache. This is either just
1263 * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
1264 * after a lookup).
1265 *
1266 * A reply may contain
1267 * a directory inode along with a dentry.
1268 * and/or a target inode
1269 *
1270 * Called with snap_rwsem (read).
1271 */
1272int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req)
1273{
1274 struct ceph_mds_session *session = req->r_session;
1275 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1276 struct inode *in = NULL;
1277 struct ceph_vino tvino, dvino;
1278 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
1279 int err = 0;
1280
1281 dout("fill_trace %p is_dentry %d is_target %d\n", req,
1282 rinfo->head->is_dentry, rinfo->head->is_target);
1283
1284 if (!rinfo->head->is_target && !rinfo->head->is_dentry) {
1285 dout("fill_trace reply is empty!\n");
1286 if (rinfo->head->result == 0 && req->r_parent)
1287 ceph_invalidate_dir_request(req);
1288 return 0;
1289 }
1290
1291 if (rinfo->head->is_dentry) {
1292 struct inode *dir = req->r_parent;
1293
1294 if (dir) {
1295 err = ceph_fill_inode(dir, NULL, &rinfo->diri,
1296 rinfo->dirfrag, session, -1,
1297 &req->r_caps_reservation);
1298 if (err < 0)
1299 goto done;
1300 } else {
1301 WARN_ON_ONCE(1);
1302 }
1303
1304 if (dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME &&
1305 test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1306 !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
1307 struct qstr dname;
1308 struct dentry *dn, *parent;
1309
1310 BUG_ON(!rinfo->head->is_target);
1311 BUG_ON(req->r_dentry);
1312
1313 parent = d_find_any_alias(dir);
1314 BUG_ON(!parent);
1315
1316 dname.name = rinfo->dname;
1317 dname.len = rinfo->dname_len;
1318 dname.hash = full_name_hash(parent, dname.name, dname.len);
1319 tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1320 tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1321retry_lookup:
1322 dn = d_lookup(parent, &dname);
1323 dout("d_lookup on parent=%p name=%.*s got %p\n",
1324 parent, dname.len, dname.name, dn);
1325
1326 if (!dn) {
1327 dn = d_alloc(parent, &dname);
1328 dout("d_alloc %p '%.*s' = %p\n", parent,
1329 dname.len, dname.name, dn);
1330 if (!dn) {
1331 dput(parent);
1332 err = -ENOMEM;
1333 goto done;
1334 }
1335 err = 0;
1336 } else if (d_really_is_positive(dn) &&
1337 (ceph_ino(d_inode(dn)) != tvino.ino ||
1338 ceph_snap(d_inode(dn)) != tvino.snap)) {
1339 dout(" dn %p points to wrong inode %p\n",
1340 dn, d_inode(dn));
1341 ceph_dir_clear_ordered(dir);
1342 d_delete(dn);
1343 dput(dn);
1344 goto retry_lookup;
1345 }
1346
1347 req->r_dentry = dn;
1348 dput(parent);
1349 }
1350 }
1351
1352 if (rinfo->head->is_target) {
1353 /* Should be filled in by handle_reply */
1354 BUG_ON(!req->r_target_inode);
1355
1356 in = req->r_target_inode;
1357 err = ceph_fill_inode(in, req->r_locked_page, &rinfo->targeti,
1358 NULL, session,
1359 (!test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) &&
1360 !test_bit(CEPH_MDS_R_ASYNC, &req->r_req_flags) &&
1361 rinfo->head->result == 0) ? req->r_fmode : -1,
1362 &req->r_caps_reservation);
1363 if (err < 0) {
1364 pr_err("ceph_fill_inode badness %p %llx.%llx\n",
1365 in, ceph_vinop(in));
1366 req->r_target_inode = NULL;
1367 if (in->i_state & I_NEW)
1368 discard_new_inode(in);
1369 else
1370 iput(in);
1371 goto done;
1372 }
1373 if (in->i_state & I_NEW)
1374 unlock_new_inode(in);
1375 }
1376
1377 /*
1378 * ignore null lease/binding on snapdir ENOENT, or else we
1379 * will have trouble splicing in the virtual snapdir later
1380 */
1381 if (rinfo->head->is_dentry &&
1382 !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) &&
1383 test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1384 (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name,
1385 fsc->mount_options->snapdir_name,
1386 req->r_dentry->d_name.len))) {
1387 /*
1388 * lookup link rename : null -> possibly existing inode
1389 * mknod symlink mkdir : null -> new inode
1390 * unlink : linked -> null
1391 */
1392 struct inode *dir = req->r_parent;
1393 struct dentry *dn = req->r_dentry;
1394 bool have_dir_cap, have_lease;
1395
1396 BUG_ON(!dn);
1397 BUG_ON(!dir);
1398 BUG_ON(d_inode(dn->d_parent) != dir);
1399
1400 dvino.ino = le64_to_cpu(rinfo->diri.in->ino);
1401 dvino.snap = le64_to_cpu(rinfo->diri.in->snapid);
1402
1403 BUG_ON(ceph_ino(dir) != dvino.ino);
1404 BUG_ON(ceph_snap(dir) != dvino.snap);
1405
1406 /* do we have a lease on the whole dir? */
1407 have_dir_cap =
1408 (le32_to_cpu(rinfo->diri.in->cap.caps) &
1409 CEPH_CAP_FILE_SHARED);
1410
1411 /* do we have a dn lease? */
1412 have_lease = have_dir_cap ||
1413 le32_to_cpu(rinfo->dlease->duration_ms);
1414 if (!have_lease)
1415 dout("fill_trace no dentry lease or dir cap\n");
1416
1417 /* rename? */
1418 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) {
1419 struct inode *olddir = req->r_old_dentry_dir;
1420 BUG_ON(!olddir);
1421
1422 dout(" src %p '%pd' dst %p '%pd'\n",
1423 req->r_old_dentry,
1424 req->r_old_dentry,
1425 dn, dn);
1426 dout("fill_trace doing d_move %p -> %p\n",
1427 req->r_old_dentry, dn);
1428
1429 /* d_move screws up sibling dentries' offsets */
1430 ceph_dir_clear_ordered(dir);
1431 ceph_dir_clear_ordered(olddir);
1432
1433 d_move(req->r_old_dentry, dn);
1434 dout(" src %p '%pd' dst %p '%pd'\n",
1435 req->r_old_dentry,
1436 req->r_old_dentry,
1437 dn, dn);
1438
1439 /* ensure target dentry is invalidated, despite
1440 rehashing bug in vfs_rename_dir */
1441 ceph_invalidate_dentry_lease(dn);
1442
1443 dout("dn %p gets new offset %lld\n", req->r_old_dentry,
1444 ceph_dentry(req->r_old_dentry)->offset);
1445
1446 /* swap r_dentry and r_old_dentry in case that
1447 * splice_dentry() gets called later. This is safe
1448 * because no other place will use them */
1449 req->r_dentry = req->r_old_dentry;
1450 req->r_old_dentry = dn;
1451 dn = req->r_dentry;
1452 }
1453
1454 /* null dentry? */
1455 if (!rinfo->head->is_target) {
1456 dout("fill_trace null dentry\n");
1457 if (d_really_is_positive(dn)) {
1458 dout("d_delete %p\n", dn);
1459 ceph_dir_clear_ordered(dir);
1460 d_delete(dn);
1461 } else if (have_lease) {
1462 if (d_unhashed(dn))
1463 d_add(dn, NULL);
1464 update_dentry_lease(dir, dn,
1465 rinfo->dlease, session,
1466 req->r_request_started);
1467 }
1468 goto done;
1469 }
1470
1471 /* attach proper inode */
1472 if (d_really_is_negative(dn)) {
1473 ceph_dir_clear_ordered(dir);
1474 ihold(in);
1475 err = splice_dentry(&req->r_dentry, in);
1476 if (err < 0)
1477 goto done;
1478 dn = req->r_dentry; /* may have spliced */
1479 } else if (d_really_is_positive(dn) && d_inode(dn) != in) {
1480 dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1481 dn, d_inode(dn), ceph_vinop(d_inode(dn)),
1482 ceph_vinop(in));
1483 d_invalidate(dn);
1484 have_lease = false;
1485 }
1486
1487 if (have_lease) {
1488 update_dentry_lease(dir, dn,
1489 rinfo->dlease, session,
1490 req->r_request_started);
1491 }
1492 dout(" final dn %p\n", dn);
1493 } else if ((req->r_op == CEPH_MDS_OP_LOOKUPSNAP ||
1494 req->r_op == CEPH_MDS_OP_MKSNAP) &&
1495 test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1496 !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
1497 struct inode *dir = req->r_parent;
1498
1499 /* fill out a snapdir LOOKUPSNAP dentry */
1500 BUG_ON(!dir);
1501 BUG_ON(ceph_snap(dir) != CEPH_SNAPDIR);
1502 BUG_ON(!req->r_dentry);
1503 dout(" linking snapped dir %p to dn %p\n", in, req->r_dentry);
1504 ceph_dir_clear_ordered(dir);
1505 ihold(in);
1506 err = splice_dentry(&req->r_dentry, in);
1507 if (err < 0)
1508 goto done;
1509 } else if (rinfo->head->is_dentry && req->r_dentry) {
1510 /* parent inode is not locked, be carefull */
1511 struct ceph_vino *ptvino = NULL;
1512 dvino.ino = le64_to_cpu(rinfo->diri.in->ino);
1513 dvino.snap = le64_to_cpu(rinfo->diri.in->snapid);
1514 if (rinfo->head->is_target) {
1515 tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1516 tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1517 ptvino = &tvino;
1518 }
1519 update_dentry_lease_careful(req->r_dentry, rinfo->dlease,
1520 session, req->r_request_started,
1521 rinfo->dname, rinfo->dname_len,
1522 &dvino, ptvino);
1523 }
1524done:
1525 dout("fill_trace done err=%d\n", err);
1526 return err;
1527}
1528
1529/*
1530 * Prepopulate our cache with readdir results, leases, etc.
1531 */
1532static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req,
1533 struct ceph_mds_session *session)
1534{
1535 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1536 int i, err = 0;
1537
1538 for (i = 0; i < rinfo->dir_nr; i++) {
1539 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1540 struct ceph_vino vino;
1541 struct inode *in;
1542 int rc;
1543
1544 vino.ino = le64_to_cpu(rde->inode.in->ino);
1545 vino.snap = le64_to_cpu(rde->inode.in->snapid);
1546
1547 in = ceph_get_inode(req->r_dentry->d_sb, vino);
1548 if (IS_ERR(in)) {
1549 err = PTR_ERR(in);
1550 dout("new_inode badness got %d\n", err);
1551 continue;
1552 }
1553 rc = ceph_fill_inode(in, NULL, &rde->inode, NULL, session,
1554 -1, &req->r_caps_reservation);
1555 if (rc < 0) {
1556 pr_err("ceph_fill_inode badness on %p got %d\n",
1557 in, rc);
1558 err = rc;
1559 if (in->i_state & I_NEW) {
1560 ihold(in);
1561 discard_new_inode(in);
1562 }
1563 } else if (in->i_state & I_NEW) {
1564 unlock_new_inode(in);
1565 }
1566
1567 iput(in);
1568 }
1569
1570 return err;
1571}
1572
1573void ceph_readdir_cache_release(struct ceph_readdir_cache_control *ctl)
1574{
1575 if (ctl->page) {
1576 kunmap(ctl->page);
1577 put_page(ctl->page);
1578 ctl->page = NULL;
1579 }
1580}
1581
1582static int fill_readdir_cache(struct inode *dir, struct dentry *dn,
1583 struct ceph_readdir_cache_control *ctl,
1584 struct ceph_mds_request *req)
1585{
1586 struct ceph_inode_info *ci = ceph_inode(dir);
1587 unsigned nsize = PAGE_SIZE / sizeof(struct dentry*);
1588 unsigned idx = ctl->index % nsize;
1589 pgoff_t pgoff = ctl->index / nsize;
1590
1591 if (!ctl->page || pgoff != page_index(ctl->page)) {
1592 ceph_readdir_cache_release(ctl);
1593 if (idx == 0)
1594 ctl->page = grab_cache_page(&dir->i_data, pgoff);
1595 else
1596 ctl->page = find_lock_page(&dir->i_data, pgoff);
1597 if (!ctl->page) {
1598 ctl->index = -1;
1599 return idx == 0 ? -ENOMEM : 0;
1600 }
1601 /* reading/filling the cache are serialized by
1602 * i_mutex, no need to use page lock */
1603 unlock_page(ctl->page);
1604 ctl->dentries = kmap(ctl->page);
1605 if (idx == 0)
1606 memset(ctl->dentries, 0, PAGE_SIZE);
1607 }
1608
1609 if (req->r_dir_release_cnt == atomic64_read(&ci->i_release_count) &&
1610 req->r_dir_ordered_cnt == atomic64_read(&ci->i_ordered_count)) {
1611 dout("readdir cache dn %p idx %d\n", dn, ctl->index);
1612 ctl->dentries[idx] = dn;
1613 ctl->index++;
1614 } else {
1615 dout("disable readdir cache\n");
1616 ctl->index = -1;
1617 }
1618 return 0;
1619}
1620
1621int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1622 struct ceph_mds_session *session)
1623{
1624 struct dentry *parent = req->r_dentry;
1625 struct ceph_inode_info *ci = ceph_inode(d_inode(parent));
1626 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1627 struct qstr dname;
1628 struct dentry *dn;
1629 struct inode *in;
1630 int err = 0, skipped = 0, ret, i;
1631 u32 frag = le32_to_cpu(req->r_args.readdir.frag);
1632 u32 last_hash = 0;
1633 u32 fpos_offset;
1634 struct ceph_readdir_cache_control cache_ctl = {};
1635
1636 if (test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags))
1637 return readdir_prepopulate_inodes_only(req, session);
1638
1639 if (rinfo->hash_order) {
1640 if (req->r_path2) {
1641 last_hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1642 req->r_path2,
1643 strlen(req->r_path2));
1644 last_hash = ceph_frag_value(last_hash);
1645 } else if (rinfo->offset_hash) {
1646 /* mds understands offset_hash */
1647 WARN_ON_ONCE(req->r_readdir_offset != 2);
1648 last_hash = le32_to_cpu(req->r_args.readdir.offset_hash);
1649 }
1650 }
1651
1652 if (rinfo->dir_dir &&
1653 le32_to_cpu(rinfo->dir_dir->frag) != frag) {
1654 dout("readdir_prepopulate got new frag %x -> %x\n",
1655 frag, le32_to_cpu(rinfo->dir_dir->frag));
1656 frag = le32_to_cpu(rinfo->dir_dir->frag);
1657 if (!rinfo->hash_order)
1658 req->r_readdir_offset = 2;
1659 }
1660
1661 if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
1662 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1663 rinfo->dir_nr, parent);
1664 } else {
1665 dout("readdir_prepopulate %d items under dn %p\n",
1666 rinfo->dir_nr, parent);
1667 if (rinfo->dir_dir)
1668 ceph_fill_dirfrag(d_inode(parent), rinfo->dir_dir);
1669
1670 if (ceph_frag_is_leftmost(frag) &&
1671 req->r_readdir_offset == 2 &&
1672 !(rinfo->hash_order && last_hash)) {
1673 /* note dir version at start of readdir so we can
1674 * tell if any dentries get dropped */
1675 req->r_dir_release_cnt =
1676 atomic64_read(&ci->i_release_count);
1677 req->r_dir_ordered_cnt =
1678 atomic64_read(&ci->i_ordered_count);
1679 req->r_readdir_cache_idx = 0;
1680 }
1681 }
1682
1683 cache_ctl.index = req->r_readdir_cache_idx;
1684 fpos_offset = req->r_readdir_offset;
1685
1686 /* FIXME: release caps/leases if error occurs */
1687 for (i = 0; i < rinfo->dir_nr; i++) {
1688 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1689 struct ceph_vino tvino;
1690
1691 dname.name = rde->name;
1692 dname.len = rde->name_len;
1693 dname.hash = full_name_hash(parent, dname.name, dname.len);
1694
1695 tvino.ino = le64_to_cpu(rde->inode.in->ino);
1696 tvino.snap = le64_to_cpu(rde->inode.in->snapid);
1697
1698 if (rinfo->hash_order) {
1699 u32 hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1700 rde->name, rde->name_len);
1701 hash = ceph_frag_value(hash);
1702 if (hash != last_hash)
1703 fpos_offset = 2;
1704 last_hash = hash;
1705 rde->offset = ceph_make_fpos(hash, fpos_offset++, true);
1706 } else {
1707 rde->offset = ceph_make_fpos(frag, fpos_offset++, false);
1708 }
1709
1710retry_lookup:
1711 dn = d_lookup(parent, &dname);
1712 dout("d_lookup on parent=%p name=%.*s got %p\n",
1713 parent, dname.len, dname.name, dn);
1714
1715 if (!dn) {
1716 dn = d_alloc(parent, &dname);
1717 dout("d_alloc %p '%.*s' = %p\n", parent,
1718 dname.len, dname.name, dn);
1719 if (!dn) {
1720 dout("d_alloc badness\n");
1721 err = -ENOMEM;
1722 goto out;
1723 }
1724 } else if (d_really_is_positive(dn) &&
1725 (ceph_ino(d_inode(dn)) != tvino.ino ||
1726 ceph_snap(d_inode(dn)) != tvino.snap)) {
1727 struct ceph_dentry_info *di = ceph_dentry(dn);
1728 dout(" dn %p points to wrong inode %p\n",
1729 dn, d_inode(dn));
1730
1731 spin_lock(&dn->d_lock);
1732 if (di->offset > 0 &&
1733 di->lease_shared_gen ==
1734 atomic_read(&ci->i_shared_gen)) {
1735 __ceph_dir_clear_ordered(ci);
1736 di->offset = 0;
1737 }
1738 spin_unlock(&dn->d_lock);
1739
1740 d_delete(dn);
1741 dput(dn);
1742 goto retry_lookup;
1743 }
1744
1745 /* inode */
1746 if (d_really_is_positive(dn)) {
1747 in = d_inode(dn);
1748 } else {
1749 in = ceph_get_inode(parent->d_sb, tvino);
1750 if (IS_ERR(in)) {
1751 dout("new_inode badness\n");
1752 d_drop(dn);
1753 dput(dn);
1754 err = PTR_ERR(in);
1755 goto out;
1756 }
1757 }
1758
1759 ret = ceph_fill_inode(in, NULL, &rde->inode, NULL, session,
1760 -1, &req->r_caps_reservation);
1761 if (ret < 0) {
1762 pr_err("ceph_fill_inode badness on %p\n", in);
1763 if (d_really_is_negative(dn)) {
1764 if (in->i_state & I_NEW) {
1765 ihold(in);
1766 discard_new_inode(in);
1767 }
1768 iput(in);
1769 }
1770 d_drop(dn);
1771 err = ret;
1772 goto next_item;
1773 }
1774 if (in->i_state & I_NEW)
1775 unlock_new_inode(in);
1776
1777 if (d_really_is_negative(dn)) {
1778 if (ceph_security_xattr_deadlock(in)) {
1779 dout(" skip splicing dn %p to inode %p"
1780 " (security xattr deadlock)\n", dn, in);
1781 iput(in);
1782 skipped++;
1783 goto next_item;
1784 }
1785
1786 err = splice_dentry(&dn, in);
1787 if (err < 0)
1788 goto next_item;
1789 }
1790
1791 ceph_dentry(dn)->offset = rde->offset;
1792
1793 update_dentry_lease(d_inode(parent), dn,
1794 rde->lease, req->r_session,
1795 req->r_request_started);
1796
1797 if (err == 0 && skipped == 0 && cache_ctl.index >= 0) {
1798 ret = fill_readdir_cache(d_inode(parent), dn,
1799 &cache_ctl, req);
1800 if (ret < 0)
1801 err = ret;
1802 }
1803next_item:
1804 dput(dn);
1805 }
1806out:
1807 if (err == 0 && skipped == 0) {
1808 set_bit(CEPH_MDS_R_DID_PREPOPULATE, &req->r_req_flags);
1809 req->r_readdir_cache_idx = cache_ctl.index;
1810 }
1811 ceph_readdir_cache_release(&cache_ctl);
1812 dout("readdir_prepopulate done\n");
1813 return err;
1814}
1815
1816bool ceph_inode_set_size(struct inode *inode, loff_t size)
1817{
1818 struct ceph_inode_info *ci = ceph_inode(inode);
1819 bool ret;
1820
1821 spin_lock(&ci->i_ceph_lock);
1822 dout("set_size %p %llu -> %llu\n", inode, i_size_read(inode), size);
1823 i_size_write(inode, size);
1824 inode->i_blocks = calc_inode_blocks(size);
1825
1826 ret = __ceph_should_report_size(ci);
1827
1828 spin_unlock(&ci->i_ceph_lock);
1829 return ret;
1830}
1831
1832void ceph_queue_inode_work(struct inode *inode, int work_bit)
1833{
1834 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1835 struct ceph_inode_info *ci = ceph_inode(inode);
1836 set_bit(work_bit, &ci->i_work_mask);
1837
1838 ihold(inode);
1839 if (queue_work(fsc->inode_wq, &ci->i_work)) {
1840 dout("queue_inode_work %p, mask=%lx\n", inode, ci->i_work_mask);
1841 } else {
1842 dout("queue_inode_work %p already queued, mask=%lx\n",
1843 inode, ci->i_work_mask);
1844 iput(inode);
1845 }
1846}
1847
1848static void ceph_do_invalidate_pages(struct inode *inode)
1849{
1850 struct ceph_inode_info *ci = ceph_inode(inode);
1851 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1852 u32 orig_gen;
1853 int check = 0;
1854
1855 mutex_lock(&ci->i_truncate_mutex);
1856
1857 if (READ_ONCE(fsc->mount_state) >= CEPH_MOUNT_SHUTDOWN) {
1858 pr_warn_ratelimited("invalidate_pages %p %lld forced umount\n",
1859 inode, ceph_ino(inode));
1860 mapping_set_error(inode->i_mapping, -EIO);
1861 truncate_pagecache(inode, 0);
1862 mutex_unlock(&ci->i_truncate_mutex);
1863 goto out;
1864 }
1865
1866 spin_lock(&ci->i_ceph_lock);
1867 dout("invalidate_pages %p gen %d revoking %d\n", inode,
1868 ci->i_rdcache_gen, ci->i_rdcache_revoking);
1869 if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
1870 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1871 check = 1;
1872 spin_unlock(&ci->i_ceph_lock);
1873 mutex_unlock(&ci->i_truncate_mutex);
1874 goto out;
1875 }
1876 orig_gen = ci->i_rdcache_gen;
1877 spin_unlock(&ci->i_ceph_lock);
1878
1879 ceph_fscache_invalidate(inode);
1880 if (invalidate_inode_pages2(inode->i_mapping) < 0) {
1881 pr_err("invalidate_pages %p fails\n", inode);
1882 }
1883
1884 spin_lock(&ci->i_ceph_lock);
1885 if (orig_gen == ci->i_rdcache_gen &&
1886 orig_gen == ci->i_rdcache_revoking) {
1887 dout("invalidate_pages %p gen %d successful\n", inode,
1888 ci->i_rdcache_gen);
1889 ci->i_rdcache_revoking--;
1890 check = 1;
1891 } else {
1892 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1893 inode, orig_gen, ci->i_rdcache_gen,
1894 ci->i_rdcache_revoking);
1895 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1896 check = 1;
1897 }
1898 spin_unlock(&ci->i_ceph_lock);
1899 mutex_unlock(&ci->i_truncate_mutex);
1900out:
1901 if (check)
1902 ceph_check_caps(ci, 0, NULL);
1903}
1904
1905/*
1906 * Make sure any pending truncation is applied before doing anything
1907 * that may depend on it.
1908 */
1909void __ceph_do_pending_vmtruncate(struct inode *inode)
1910{
1911 struct ceph_inode_info *ci = ceph_inode(inode);
1912 u64 to;
1913 int wrbuffer_refs, finish = 0;
1914
1915 mutex_lock(&ci->i_truncate_mutex);
1916retry:
1917 spin_lock(&ci->i_ceph_lock);
1918 if (ci->i_truncate_pending == 0) {
1919 dout("__do_pending_vmtruncate %p none pending\n", inode);
1920 spin_unlock(&ci->i_ceph_lock);
1921 mutex_unlock(&ci->i_truncate_mutex);
1922 return;
1923 }
1924
1925 /*
1926 * make sure any dirty snapped pages are flushed before we
1927 * possibly truncate them.. so write AND block!
1928 */
1929 if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) {
1930 spin_unlock(&ci->i_ceph_lock);
1931 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1932 inode);
1933 filemap_write_and_wait_range(&inode->i_data, 0,
1934 inode->i_sb->s_maxbytes);
1935 goto retry;
1936 }
1937
1938 /* there should be no reader or writer */
1939 WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref);
1940
1941 to = ci->i_truncate_size;
1942 wrbuffer_refs = ci->i_wrbuffer_ref;
1943 dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode,
1944 ci->i_truncate_pending, to);
1945 spin_unlock(&ci->i_ceph_lock);
1946
1947 truncate_pagecache(inode, to);
1948
1949 spin_lock(&ci->i_ceph_lock);
1950 if (to == ci->i_truncate_size) {
1951 ci->i_truncate_pending = 0;
1952 finish = 1;
1953 }
1954 spin_unlock(&ci->i_ceph_lock);
1955 if (!finish)
1956 goto retry;
1957
1958 mutex_unlock(&ci->i_truncate_mutex);
1959
1960 if (wrbuffer_refs == 0)
1961 ceph_check_caps(ci, 0, NULL);
1962
1963 wake_up_all(&ci->i_cap_wq);
1964}
1965
1966static void ceph_inode_work(struct work_struct *work)
1967{
1968 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1969 i_work);
1970 struct inode *inode = &ci->vfs_inode;
1971
1972 if (test_and_clear_bit(CEPH_I_WORK_WRITEBACK, &ci->i_work_mask)) {
1973 dout("writeback %p\n", inode);
1974 filemap_fdatawrite(&inode->i_data);
1975 }
1976 if (test_and_clear_bit(CEPH_I_WORK_INVALIDATE_PAGES, &ci->i_work_mask))
1977 ceph_do_invalidate_pages(inode);
1978
1979 if (test_and_clear_bit(CEPH_I_WORK_VMTRUNCATE, &ci->i_work_mask))
1980 __ceph_do_pending_vmtruncate(inode);
1981
1982 if (test_and_clear_bit(CEPH_I_WORK_CHECK_CAPS, &ci->i_work_mask))
1983 ceph_check_caps(ci, 0, NULL);
1984
1985 if (test_and_clear_bit(CEPH_I_WORK_FLUSH_SNAPS, &ci->i_work_mask))
1986 ceph_flush_snaps(ci, NULL);
1987
1988 iput(inode);
1989}
1990
1991/*
1992 * symlinks
1993 */
1994static const struct inode_operations ceph_symlink_iops = {
1995 .get_link = simple_get_link,
1996 .setattr = ceph_setattr,
1997 .getattr = ceph_getattr,
1998 .listxattr = ceph_listxattr,
1999};
2000
2001int __ceph_setattr(struct inode *inode, struct iattr *attr)
2002{
2003 struct ceph_inode_info *ci = ceph_inode(inode);
2004 unsigned int ia_valid = attr->ia_valid;
2005 struct ceph_mds_request *req;
2006 struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
2007 struct ceph_cap_flush *prealloc_cf;
2008 int issued;
2009 int release = 0, dirtied = 0;
2010 int mask = 0;
2011 int err = 0;
2012 int inode_dirty_flags = 0;
2013 bool lock_snap_rwsem = false;
2014
2015 prealloc_cf = ceph_alloc_cap_flush();
2016 if (!prealloc_cf)
2017 return -ENOMEM;
2018
2019 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR,
2020 USE_AUTH_MDS);
2021 if (IS_ERR(req)) {
2022 ceph_free_cap_flush(prealloc_cf);
2023 return PTR_ERR(req);
2024 }
2025
2026 spin_lock(&ci->i_ceph_lock);
2027 issued = __ceph_caps_issued(ci, NULL);
2028
2029 if (!ci->i_head_snapc &&
2030 (issued & (CEPH_CAP_ANY_EXCL | CEPH_CAP_FILE_WR))) {
2031 lock_snap_rwsem = true;
2032 if (!down_read_trylock(&mdsc->snap_rwsem)) {
2033 spin_unlock(&ci->i_ceph_lock);
2034 down_read(&mdsc->snap_rwsem);
2035 spin_lock(&ci->i_ceph_lock);
2036 issued = __ceph_caps_issued(ci, NULL);
2037 }
2038 }
2039
2040 dout("setattr %p issued %s\n", inode, ceph_cap_string(issued));
2041
2042 if (ia_valid & ATTR_UID) {
2043 dout("setattr %p uid %d -> %d\n", inode,
2044 from_kuid(&init_user_ns, inode->i_uid),
2045 from_kuid(&init_user_ns, attr->ia_uid));
2046 if (issued & CEPH_CAP_AUTH_EXCL) {
2047 inode->i_uid = attr->ia_uid;
2048 dirtied |= CEPH_CAP_AUTH_EXCL;
2049 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2050 !uid_eq(attr->ia_uid, inode->i_uid)) {
2051 req->r_args.setattr.uid = cpu_to_le32(
2052 from_kuid(&init_user_ns, attr->ia_uid));
2053 mask |= CEPH_SETATTR_UID;
2054 release |= CEPH_CAP_AUTH_SHARED;
2055 }
2056 }
2057 if (ia_valid & ATTR_GID) {
2058 dout("setattr %p gid %d -> %d\n", inode,
2059 from_kgid(&init_user_ns, inode->i_gid),
2060 from_kgid(&init_user_ns, attr->ia_gid));
2061 if (issued & CEPH_CAP_AUTH_EXCL) {
2062 inode->i_gid = attr->ia_gid;
2063 dirtied |= CEPH_CAP_AUTH_EXCL;
2064 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2065 !gid_eq(attr->ia_gid, inode->i_gid)) {
2066 req->r_args.setattr.gid = cpu_to_le32(
2067 from_kgid(&init_user_ns, attr->ia_gid));
2068 mask |= CEPH_SETATTR_GID;
2069 release |= CEPH_CAP_AUTH_SHARED;
2070 }
2071 }
2072 if (ia_valid & ATTR_MODE) {
2073 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode,
2074 attr->ia_mode);
2075 if (issued & CEPH_CAP_AUTH_EXCL) {
2076 inode->i_mode = attr->ia_mode;
2077 dirtied |= CEPH_CAP_AUTH_EXCL;
2078 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2079 attr->ia_mode != inode->i_mode) {
2080 inode->i_mode = attr->ia_mode;
2081 req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode);
2082 mask |= CEPH_SETATTR_MODE;
2083 release |= CEPH_CAP_AUTH_SHARED;
2084 }
2085 }
2086
2087 if (ia_valid & ATTR_ATIME) {
2088 dout("setattr %p atime %lld.%ld -> %lld.%ld\n", inode,
2089 inode->i_atime.tv_sec, inode->i_atime.tv_nsec,
2090 attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec);
2091 if (issued & CEPH_CAP_FILE_EXCL) {
2092 ci->i_time_warp_seq++;
2093 inode->i_atime = attr->ia_atime;
2094 dirtied |= CEPH_CAP_FILE_EXCL;
2095 } else if ((issued & CEPH_CAP_FILE_WR) &&
2096 timespec64_compare(&inode->i_atime,
2097 &attr->ia_atime) < 0) {
2098 inode->i_atime = attr->ia_atime;
2099 dirtied |= CEPH_CAP_FILE_WR;
2100 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2101 !timespec64_equal(&inode->i_atime, &attr->ia_atime)) {
2102 ceph_encode_timespec64(&req->r_args.setattr.atime,
2103 &attr->ia_atime);
2104 mask |= CEPH_SETATTR_ATIME;
2105 release |= CEPH_CAP_FILE_SHARED |
2106 CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2107 }
2108 }
2109 if (ia_valid & ATTR_SIZE) {
2110 loff_t isize = i_size_read(inode);
2111
2112 dout("setattr %p size %lld -> %lld\n", inode, isize, attr->ia_size);
2113 if ((issued & CEPH_CAP_FILE_EXCL) && attr->ia_size > isize) {
2114 i_size_write(inode, attr->ia_size);
2115 inode->i_blocks = calc_inode_blocks(attr->ia_size);
2116 ci->i_reported_size = attr->ia_size;
2117 dirtied |= CEPH_CAP_FILE_EXCL;
2118 ia_valid |= ATTR_MTIME;
2119 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2120 attr->ia_size != isize) {
2121 req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
2122 req->r_args.setattr.old_size = cpu_to_le64(isize);
2123 mask |= CEPH_SETATTR_SIZE;
2124 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_EXCL |
2125 CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2126 }
2127 }
2128 if (ia_valid & ATTR_MTIME) {
2129 dout("setattr %p mtime %lld.%ld -> %lld.%ld\n", inode,
2130 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
2131 attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec);
2132 if (issued & CEPH_CAP_FILE_EXCL) {
2133 ci->i_time_warp_seq++;
2134 inode->i_mtime = attr->ia_mtime;
2135 dirtied |= CEPH_CAP_FILE_EXCL;
2136 } else if ((issued & CEPH_CAP_FILE_WR) &&
2137 timespec64_compare(&inode->i_mtime,
2138 &attr->ia_mtime) < 0) {
2139 inode->i_mtime = attr->ia_mtime;
2140 dirtied |= CEPH_CAP_FILE_WR;
2141 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2142 !timespec64_equal(&inode->i_mtime, &attr->ia_mtime)) {
2143 ceph_encode_timespec64(&req->r_args.setattr.mtime,
2144 &attr->ia_mtime);
2145 mask |= CEPH_SETATTR_MTIME;
2146 release |= CEPH_CAP_FILE_SHARED |
2147 CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2148 }
2149 }
2150
2151 /* these do nothing */
2152 if (ia_valid & ATTR_CTIME) {
2153 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME|
2154 ATTR_MODE|ATTR_UID|ATTR_GID)) == 0;
2155 dout("setattr %p ctime %lld.%ld -> %lld.%ld (%s)\n", inode,
2156 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
2157 attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec,
2158 only ? "ctime only" : "ignored");
2159 if (only) {
2160 /*
2161 * if kernel wants to dirty ctime but nothing else,
2162 * we need to choose a cap to dirty under, or do
2163 * a almost-no-op setattr
2164 */
2165 if (issued & CEPH_CAP_AUTH_EXCL)
2166 dirtied |= CEPH_CAP_AUTH_EXCL;
2167 else if (issued & CEPH_CAP_FILE_EXCL)
2168 dirtied |= CEPH_CAP_FILE_EXCL;
2169 else if (issued & CEPH_CAP_XATTR_EXCL)
2170 dirtied |= CEPH_CAP_XATTR_EXCL;
2171 else
2172 mask |= CEPH_SETATTR_CTIME;
2173 }
2174 }
2175 if (ia_valid & ATTR_FILE)
2176 dout("setattr %p ATTR_FILE ... hrm!\n", inode);
2177
2178 if (dirtied) {
2179 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied,
2180 &prealloc_cf);
2181 inode->i_ctime = attr->ia_ctime;
2182 }
2183
2184 release &= issued;
2185 spin_unlock(&ci->i_ceph_lock);
2186 if (lock_snap_rwsem)
2187 up_read(&mdsc->snap_rwsem);
2188
2189 if (inode_dirty_flags)
2190 __mark_inode_dirty(inode, inode_dirty_flags);
2191
2192
2193 if (mask) {
2194 req->r_inode = inode;
2195 ihold(inode);
2196 req->r_inode_drop = release;
2197 req->r_args.setattr.mask = cpu_to_le32(mask);
2198 req->r_num_caps = 1;
2199 req->r_stamp = attr->ia_ctime;
2200 err = ceph_mdsc_do_request(mdsc, NULL, req);
2201 }
2202 dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err,
2203 ceph_cap_string(dirtied), mask);
2204
2205 ceph_mdsc_put_request(req);
2206 ceph_free_cap_flush(prealloc_cf);
2207
2208 if (err >= 0 && (mask & CEPH_SETATTR_SIZE))
2209 __ceph_do_pending_vmtruncate(inode);
2210
2211 return err;
2212}
2213
2214/*
2215 * setattr
2216 */
2217int ceph_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
2218 struct iattr *attr)
2219{
2220 struct inode *inode = d_inode(dentry);
2221 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
2222 int err;
2223
2224 if (ceph_snap(inode) != CEPH_NOSNAP)
2225 return -EROFS;
2226
2227 err = setattr_prepare(&init_user_ns, dentry, attr);
2228 if (err != 0)
2229 return err;
2230
2231 if ((attr->ia_valid & ATTR_SIZE) &&
2232 attr->ia_size > max(i_size_read(inode), fsc->max_file_size))
2233 return -EFBIG;
2234
2235 if ((attr->ia_valid & ATTR_SIZE) &&
2236 ceph_quota_is_max_bytes_exceeded(inode, attr->ia_size))
2237 return -EDQUOT;
2238
2239 err = __ceph_setattr(inode, attr);
2240
2241 if (err >= 0 && (attr->ia_valid & ATTR_MODE))
2242 err = posix_acl_chmod(&init_user_ns, inode, attr->ia_mode);
2243
2244 return err;
2245}
2246
2247/*
2248 * Verify that we have a lease on the given mask. If not,
2249 * do a getattr against an mds.
2250 */
2251int __ceph_do_getattr(struct inode *inode, struct page *locked_page,
2252 int mask, bool force)
2253{
2254 struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
2255 struct ceph_mds_client *mdsc = fsc->mdsc;
2256 struct ceph_mds_request *req;
2257 int mode;
2258 int err;
2259
2260 if (ceph_snap(inode) == CEPH_SNAPDIR) {
2261 dout("do_getattr inode %p SNAPDIR\n", inode);
2262 return 0;
2263 }
2264
2265 dout("do_getattr inode %p mask %s mode 0%o\n",
2266 inode, ceph_cap_string(mask), inode->i_mode);
2267 if (!force && ceph_caps_issued_mask_metric(ceph_inode(inode), mask, 1))
2268 return 0;
2269
2270 mode = (mask & CEPH_STAT_RSTAT) ? USE_AUTH_MDS : USE_ANY_MDS;
2271 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, mode);
2272 if (IS_ERR(req))
2273 return PTR_ERR(req);
2274 req->r_inode = inode;
2275 ihold(inode);
2276 req->r_num_caps = 1;
2277 req->r_args.getattr.mask = cpu_to_le32(mask);
2278 req->r_locked_page = locked_page;
2279 err = ceph_mdsc_do_request(mdsc, NULL, req);
2280 if (locked_page && err == 0) {
2281 u64 inline_version = req->r_reply_info.targeti.inline_version;
2282 if (inline_version == 0) {
2283 /* the reply is supposed to contain inline data */
2284 err = -EINVAL;
2285 } else if (inline_version == CEPH_INLINE_NONE) {
2286 err = -ENODATA;
2287 } else {
2288 err = req->r_reply_info.targeti.inline_len;
2289 }
2290 }
2291 ceph_mdsc_put_request(req);
2292 dout("do_getattr result=%d\n", err);
2293 return err;
2294}
2295
2296
2297/*
2298 * Check inode permissions. We verify we have a valid value for
2299 * the AUTH cap, then call the generic handler.
2300 */
2301int ceph_permission(struct user_namespace *mnt_userns, struct inode *inode,
2302 int mask)
2303{
2304 int err;
2305
2306 if (mask & MAY_NOT_BLOCK)
2307 return -ECHILD;
2308
2309 err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED, false);
2310
2311 if (!err)
2312 err = generic_permission(&init_user_ns, inode, mask);
2313 return err;
2314}
2315
2316/* Craft a mask of needed caps given a set of requested statx attrs. */
2317static int statx_to_caps(u32 want, umode_t mode)
2318{
2319 int mask = 0;
2320
2321 if (want & (STATX_MODE|STATX_UID|STATX_GID|STATX_CTIME|STATX_BTIME))
2322 mask |= CEPH_CAP_AUTH_SHARED;
2323
2324 if (want & (STATX_NLINK|STATX_CTIME)) {
2325 /*
2326 * The link count for directories depends on inode->i_subdirs,
2327 * and that is only updated when Fs caps are held.
2328 */
2329 if (S_ISDIR(mode))
2330 mask |= CEPH_CAP_FILE_SHARED;
2331 else
2332 mask |= CEPH_CAP_LINK_SHARED;
2333 }
2334
2335 if (want & (STATX_ATIME|STATX_MTIME|STATX_CTIME|STATX_SIZE|
2336 STATX_BLOCKS))
2337 mask |= CEPH_CAP_FILE_SHARED;
2338
2339 if (want & (STATX_CTIME))
2340 mask |= CEPH_CAP_XATTR_SHARED;
2341
2342 return mask;
2343}
2344
2345/*
2346 * Get all the attributes. If we have sufficient caps for the requested attrs,
2347 * then we can avoid talking to the MDS at all.
2348 */
2349int ceph_getattr(struct user_namespace *mnt_userns, const struct path *path,
2350 struct kstat *stat, u32 request_mask, unsigned int flags)
2351{
2352 struct inode *inode = d_inode(path->dentry);
2353 struct ceph_inode_info *ci = ceph_inode(inode);
2354 u32 valid_mask = STATX_BASIC_STATS;
2355 int err = 0;
2356
2357 /* Skip the getattr altogether if we're asked not to sync */
2358 if (!(flags & AT_STATX_DONT_SYNC)) {
2359 err = ceph_do_getattr(inode,
2360 statx_to_caps(request_mask, inode->i_mode),
2361 flags & AT_STATX_FORCE_SYNC);
2362 if (err)
2363 return err;
2364 }
2365
2366 generic_fillattr(&init_user_ns, inode, stat);
2367 stat->ino = ceph_present_inode(inode);
2368
2369 /*
2370 * btime on newly-allocated inodes is 0, so if this is still set to
2371 * that, then assume that it's not valid.
2372 */
2373 if (ci->i_btime.tv_sec || ci->i_btime.tv_nsec) {
2374 stat->btime = ci->i_btime;
2375 valid_mask |= STATX_BTIME;
2376 }
2377
2378 if (ceph_snap(inode) == CEPH_NOSNAP)
2379 stat->dev = inode->i_sb->s_dev;
2380 else
2381 stat->dev = ci->i_snapid_map ? ci->i_snapid_map->dev : 0;
2382
2383 if (S_ISDIR(inode->i_mode)) {
2384 if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb),
2385 RBYTES))
2386 stat->size = ci->i_rbytes;
2387 else
2388 stat->size = ci->i_files + ci->i_subdirs;
2389 stat->blocks = 0;
2390 stat->blksize = 65536;
2391 /*
2392 * Some applications rely on the number of st_nlink
2393 * value on directories to be either 0 (if unlinked)
2394 * or 2 + number of subdirectories.
2395 */
2396 if (stat->nlink == 1)
2397 /* '.' + '..' + subdirs */
2398 stat->nlink = 1 + 1 + ci->i_subdirs;
2399 }
2400
2401 stat->result_mask = request_mask & valid_mask;
2402 return err;
2403}