?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
mm: consider all swapped back pages in used-once logic
[linux-2.6.git] / fs / ceph / inode.c
blob9fff9f3b17e4a5206a9073cec2be714fba44bda0
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
29 static const struct inode_operations ceph_symlink_iops;
30
31 static void ceph_invalidate_work(struct work_struct *work);
32 static void ceph_writeback_work(struct work_struct *work);
33 static void ceph_vmtruncate_work(struct work_struct *work);
34
35 /*
36 * find or create an inode, given the ceph ino number
37 */
38 static 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
45 struct 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 */
67 struct 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
89 const 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 */
111 static 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 */
155 struct 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 */
178 u32 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 */
228 static 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
280 out:
281 mutex_unlock(&ci->i_fragtree_mutex);
282 return err;
283 }
284
285
286 /*
287 * initialize a newly allocated inode.
288 */
289 struct 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
382 static 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
390 void 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 */
441 int 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
483 void 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 */
553 static 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
719 no_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
788 out:
789 if (xattr_blob)
790 ceph_buffer_put(xattr_blob);
791 return err;
792 }
793
794 /*
795 * caller should hold session s_mutex.
796 */
797 static 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;
839 out_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 */
850 static 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 */
887 static 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);
922 out:
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 */
937 int 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
1185 done:
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 */
1193 int 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
1229 retry_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);
1295 next_item:
1296 if (dn)
1297 dput(dn);
1298 }
1299 req->r_did_prepopulate = true;
1300
1301 out:
1302 if (snapdir) {
1303 iput(snapdir);
1304 dput(parent);
1305 }
1306 dout("readdir_prepopulate done\n");
1307 return err;
1308 }
1309
1310 int 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 */
1333 void 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
1345 static 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 */
1359 void 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 */
1375 static 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);
1412 out:
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 */
1422 static 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 */
1439 void 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 */
1460 void __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
1466 retry:
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 */
1511 static 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
1518 static const struct inode_operations ceph_symlink_iops = {
1519 .readlink = generic_readlink,
1520 .follow_link = ceph_sym_follow_link,
1521 };
1522
1523 /*
1524 * setattr
1525 */
1526 int 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;
1723 out:
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 */
1733 int 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 */
1767 int 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 */
1785 int 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 }
Linus' kernel tree