projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge branch 'iommu/fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/joro...
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / fs / ceph / inode.c
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 ci->i_version = 0;
301 ci->i_time_warp_seq = 0;
302 ci->i_ceph_flags = 0;
303 ci->i_release_count = 0;
304 ci->i_symlink = NULL;
305
306 memset(&ci->i_dir_layout, 0, sizeof(ci->i_dir_layout));
307
308 ci->i_fragtree = RB_ROOT;
309 mutex_init(&ci->i_fragtree_mutex);
310
311 ci->i_xattrs.blob = NULL;
312 ci->i_xattrs.prealloc_blob = NULL;
313 ci->i_xattrs.dirty = false;
314 ci->i_xattrs.index = RB_ROOT;
315 ci->i_xattrs.count = 0;
316 ci->i_xattrs.names_size = 0;
317 ci->i_xattrs.vals_size = 0;
318 ci->i_xattrs.version = 0;
319 ci->i_xattrs.index_version = 0;
320
321 ci->i_caps = RB_ROOT;
322 ci->i_auth_cap = NULL;
323 ci->i_dirty_caps = 0;
324 ci->i_flushing_caps = 0;
325 INIT_LIST_HEAD(&ci->i_dirty_item);
326 INIT_LIST_HEAD(&ci->i_flushing_item);
327 ci->i_cap_flush_seq = 0;
328 ci->i_cap_flush_last_tid = 0;
329 memset(&ci->i_cap_flush_tid, 0, sizeof(ci->i_cap_flush_tid));
330 init_waitqueue_head(&ci->i_cap_wq);
331 ci->i_hold_caps_min = 0;
332 ci->i_hold_caps_max = 0;
333 INIT_LIST_HEAD(&ci->i_cap_delay_list);
334 ci->i_cap_exporting_mds = 0;
335 ci->i_cap_exporting_mseq = 0;
336 ci->i_cap_exporting_issued = 0;
337 INIT_LIST_HEAD(&ci->i_cap_snaps);
338 ci->i_head_snapc = NULL;
339 ci->i_snap_caps = 0;
340
341 for (i = 0; i < CEPH_FILE_MODE_NUM; i++)
342 ci->i_nr_by_mode[i] = 0;
343
344 ci->i_truncate_seq = 0;
345 ci->i_truncate_size = 0;
346 ci->i_truncate_pending = 0;
347
348 ci->i_max_size = 0;
349 ci->i_reported_size = 0;
350 ci->i_wanted_max_size = 0;
351 ci->i_requested_max_size = 0;
352
353 ci->i_pin_ref = 0;
354 ci->i_rd_ref = 0;
355 ci->i_rdcache_ref = 0;
356 ci->i_wr_ref = 0;
357 ci->i_wb_ref = 0;
358 ci->i_wrbuffer_ref = 0;
359 ci->i_wrbuffer_ref_head = 0;
360 ci->i_shared_gen = 0;
361 ci->i_rdcache_gen = 0;
362 ci->i_rdcache_revoking = 0;
363
364 INIT_LIST_HEAD(&ci->i_unsafe_writes);
365 INIT_LIST_HEAD(&ci->i_unsafe_dirops);
366 spin_lock_init(&ci->i_unsafe_lock);
367
368 ci->i_snap_realm = NULL;
369 INIT_LIST_HEAD(&ci->i_snap_realm_item);
370 INIT_LIST_HEAD(&ci->i_snap_flush_item);
371
372 INIT_WORK(&ci->i_wb_work, ceph_writeback_work);
373 INIT_WORK(&ci->i_pg_inv_work, ceph_invalidate_work);
374
375 INIT_WORK(&ci->i_vmtruncate_work, ceph_vmtruncate_work);
376
377 return &ci->vfs_inode;
378 }
379
380 static void ceph_i_callback(struct rcu_head *head)
381 {
382 struct inode *inode = container_of(head, struct inode, i_rcu);
383 struct ceph_inode_info *ci = ceph_inode(inode);
384
385 INIT_LIST_HEAD(&inode->i_dentry);
386 kmem_cache_free(ceph_inode_cachep, ci);
387 }
388
389 void ceph_destroy_inode(struct inode *inode)
390 {
391 struct ceph_inode_info *ci = ceph_inode(inode);
392 struct ceph_inode_frag *frag;
393 struct rb_node *n;
394
395 dout("destroy_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode));
396
397 ceph_queue_caps_release(inode);
398
399 /*
400 * we may still have a snap_realm reference if there are stray
401 * caps in i_cap_exporting_issued or i_snap_caps.
402 */
403 if (ci->i_snap_realm) {
404 struct ceph_mds_client *mdsc =
405 ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc;
406 struct ceph_snap_realm *realm = ci->i_snap_realm;
407
408 dout(" dropping residual ref to snap realm %p\n", realm);
409 spin_lock(&realm->inodes_with_caps_lock);
410 list_del_init(&ci->i_snap_realm_item);
411 spin_unlock(&realm->inodes_with_caps_lock);
412 ceph_put_snap_realm(mdsc, realm);
413 }
414
415 kfree(ci->i_symlink);
416 while ((n = rb_first(&ci->i_fragtree)) != NULL) {
417 frag = rb_entry(n, struct ceph_inode_frag, node);
418 rb_erase(n, &ci->i_fragtree);
419 kfree(frag);
420 }
421
422 __ceph_destroy_xattrs(ci);
423 if (ci->i_xattrs.blob)
424 ceph_buffer_put(ci->i_xattrs.blob);
425 if (ci->i_xattrs.prealloc_blob)
426 ceph_buffer_put(ci->i_xattrs.prealloc_blob);
427
428 call_rcu(&inode->i_rcu, ceph_i_callback);
429 }
430
431
432 /*
433 * Helpers to fill in size, ctime, mtime, and atime. We have to be
434 * careful because either the client or MDS may have more up to date
435 * info, depending on which capabilities are held, and whether
436 * time_warp_seq or truncate_seq have increased. (Ordinarily, mtime
437 * and size are monotonically increasing, except when utimes() or
438 * truncate() increments the corresponding _seq values.)
439 */
440 int ceph_fill_file_size(struct inode *inode, int issued,
441 u32 truncate_seq, u64 truncate_size, u64 size)
442 {
443 struct ceph_inode_info *ci = ceph_inode(inode);
444 int queue_trunc = 0;
445
446 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 ||
447 (truncate_seq == ci->i_truncate_seq && size > inode->i_size)) {
448 dout("size %lld -> %llu\n", inode->i_size, size);
449 inode->i_size = size;
450 inode->i_blocks = (size + (1<<9) - 1) >> 9;
451 ci->i_reported_size = size;
452 if (truncate_seq != ci->i_truncate_seq) {
453 dout("truncate_seq %u -> %u\n",
454 ci->i_truncate_seq, truncate_seq);
455 ci->i_truncate_seq = truncate_seq;
456 /*
457 * If we hold relevant caps, or in the case where we're
458 * not the only client referencing this file and we
459 * don't hold those caps, then we need to check whether
460 * the file is either opened or mmaped
461 */
462 if ((issued & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_RD|
463 CEPH_CAP_FILE_WR|CEPH_CAP_FILE_BUFFER|
464 CEPH_CAP_FILE_EXCL|
465 CEPH_CAP_FILE_LAZYIO)) ||
466 mapping_mapped(inode->i_mapping) ||
467 __ceph_caps_file_wanted(ci)) {
468 ci->i_truncate_pending++;
469 queue_trunc = 1;
470 }
471 }
472 }
473 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 &&
474 ci->i_truncate_size != truncate_size) {
475 dout("truncate_size %lld -> %llu\n", ci->i_truncate_size,
476 truncate_size);
477 ci->i_truncate_size = truncate_size;
478 }
479 return queue_trunc;
480 }
481
482 void ceph_fill_file_time(struct inode *inode, int issued,
483 u64 time_warp_seq, struct timespec *ctime,
484 struct timespec *mtime, struct timespec *atime)
485 {
486 struct ceph_inode_info *ci = ceph_inode(inode);
487 int warn = 0;
488
489 if (issued & (CEPH_CAP_FILE_EXCL|
490 CEPH_CAP_FILE_WR|
491 CEPH_CAP_FILE_BUFFER|
492 CEPH_CAP_AUTH_EXCL|
493 CEPH_CAP_XATTR_EXCL)) {
494 if (timespec_compare(ctime, &inode->i_ctime) > 0) {
495 dout("ctime %ld.%09ld -> %ld.%09ld inc w/ cap\n",
496 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
497 ctime->tv_sec, ctime->tv_nsec);
498 inode->i_ctime = *ctime;
499 }
500 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) {
501 /* the MDS did a utimes() */
502 dout("mtime %ld.%09ld -> %ld.%09ld "
503 "tw %d -> %d\n",
504 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
505 mtime->tv_sec, mtime->tv_nsec,
506 ci->i_time_warp_seq, (int)time_warp_seq);
507
508 inode->i_mtime = *mtime;
509 inode->i_atime = *atime;
510 ci->i_time_warp_seq = time_warp_seq;
511 } else if (time_warp_seq == ci->i_time_warp_seq) {
512 /* nobody did utimes(); take the max */
513 if (timespec_compare(mtime, &inode->i_mtime) > 0) {
514 dout("mtime %ld.%09ld -> %ld.%09ld inc\n",
515 inode->i_mtime.tv_sec,
516 inode->i_mtime.tv_nsec,
517 mtime->tv_sec, mtime->tv_nsec);
518 inode->i_mtime = *mtime;
519 }
520 if (timespec_compare(atime, &inode->i_atime) > 0) {
521 dout("atime %ld.%09ld -> %ld.%09ld inc\n",
522 inode->i_atime.tv_sec,
523 inode->i_atime.tv_nsec,
524 atime->tv_sec, atime->tv_nsec);
525 inode->i_atime = *atime;
526 }
527 } else if (issued & CEPH_CAP_FILE_EXCL) {
528 /* we did a utimes(); ignore mds values */
529 } else {
530 warn = 1;
531 }
532 } else {
533 /* we have no write|excl caps; whatever the MDS says is true */
534 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) {
535 inode->i_ctime = *ctime;
536 inode->i_mtime = *mtime;
537 inode->i_atime = *atime;
538 ci->i_time_warp_seq = time_warp_seq;
539 } else {
540 warn = 1;
541 }
542 }
543 if (warn) /* time_warp_seq shouldn't go backwards */
544 dout("%p mds time_warp_seq %llu < %u\n",
545 inode, time_warp_seq, ci->i_time_warp_seq);
546 }
547
548 /*
549 * Populate an inode based on info from mds. May be called on new or
550 * existing inodes.
551 */
552 static int fill_inode(struct inode *inode,
553 struct ceph_mds_reply_info_in *iinfo,
554 struct ceph_mds_reply_dirfrag *dirinfo,
555 struct ceph_mds_session *session,
556 unsigned long ttl_from, int cap_fmode,
557 struct ceph_cap_reservation *caps_reservation)
558 {
559 struct ceph_mds_reply_inode *info = iinfo->in;
560 struct ceph_inode_info *ci = ceph_inode(inode);
561 int i;
562 int issued = 0, implemented;
563 int updating_inode = 0;
564 struct timespec mtime, atime, ctime;
565 u32 nsplits;
566 struct ceph_buffer *xattr_blob = NULL;
567 int err = 0;
568 int queue_trunc = 0;
569
570 dout("fill_inode %p ino %llx.%llx v %llu had %llu\n",
571 inode, ceph_vinop(inode), le64_to_cpu(info->version),
572 ci->i_version);
573
574 /*
575 * prealloc xattr data, if it looks like we'll need it. only
576 * if len > 4 (meaning there are actually xattrs; the first 4
577 * bytes are the xattr count).
578 */
579 if (iinfo->xattr_len > 4) {
580 xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS);
581 if (!xattr_blob)
582 pr_err("fill_inode ENOMEM xattr blob %d bytes\n",
583 iinfo->xattr_len);
584 }
585
586 spin_lock(&inode->i_lock);
587
588 /*
589 * provided version will be odd if inode value is projected,
590 * even if stable. skip the update if we have newer stable
591 * info (ours>=theirs, e.g. due to racing mds replies), unless
592 * we are getting projected (unstable) info (in which case the
593 * version is odd, and we want ours>theirs).
594 * us them
595 * 2 2 skip
596 * 3 2 skip
597 * 3 3 update
598 */
599 if (le64_to_cpu(info->version) > 0 &&
600 (ci->i_version & ~1) >= le64_to_cpu(info->version))
601 goto no_change;
602
603 updating_inode = 1;
604 issued = __ceph_caps_issued(ci, &implemented);
605 issued |= implemented | __ceph_caps_dirty(ci);
606
607 /* update inode */
608 ci->i_version = le64_to_cpu(info->version);
609 inode->i_version++;
610 inode->i_rdev = le32_to_cpu(info->rdev);
611
612 if ((issued & CEPH_CAP_AUTH_EXCL) == 0) {
613 inode->i_mode = le32_to_cpu(info->mode);
614 inode->i_uid = le32_to_cpu(info->uid);
615 inode->i_gid = le32_to_cpu(info->gid);
616 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
617 inode->i_uid, inode->i_gid);
618 }
619
620 if ((issued & CEPH_CAP_LINK_EXCL) == 0)
621 set_nlink(inode, le32_to_cpu(info->nlink));
622
623 /* be careful with mtime, atime, size */
624 ceph_decode_timespec(&atime, &info->atime);
625 ceph_decode_timespec(&mtime, &info->mtime);
626 ceph_decode_timespec(&ctime, &info->ctime);
627 queue_trunc = ceph_fill_file_size(inode, issued,
628 le32_to_cpu(info->truncate_seq),
629 le64_to_cpu(info->truncate_size),
630 le64_to_cpu(info->size));
631 ceph_fill_file_time(inode, issued,
632 le32_to_cpu(info->time_warp_seq),
633 &ctime, &mtime, &atime);
634
635 /* only update max_size on auth cap */
636 if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
637 ci->i_max_size != le64_to_cpu(info->max_size)) {
638 dout("max_size %lld -> %llu\n", ci->i_max_size,
639 le64_to_cpu(info->max_size));
640 ci->i_max_size = le64_to_cpu(info->max_size);
641 }
642
643 ci->i_layout = info->layout;
644 inode->i_blkbits = fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1;
645
646 /* xattrs */
647 /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
648 if ((issued & CEPH_CAP_XATTR_EXCL) == 0 &&
649 le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) {
650 if (ci->i_xattrs.blob)
651 ceph_buffer_put(ci->i_xattrs.blob);
652 ci->i_xattrs.blob = xattr_blob;
653 if (xattr_blob)
654 memcpy(ci->i_xattrs.blob->vec.iov_base,
655 iinfo->xattr_data, iinfo->xattr_len);
656 ci->i_xattrs.version = le64_to_cpu(info->xattr_version);
657 xattr_blob = NULL;
658 }
659
660 inode->i_mapping->a_ops = &ceph_aops;
661 inode->i_mapping->backing_dev_info =
662 &ceph_sb_to_client(inode->i_sb)->backing_dev_info;
663
664 switch (inode->i_mode & S_IFMT) {
665 case S_IFIFO:
666 case S_IFBLK:
667 case S_IFCHR:
668 case S_IFSOCK:
669 init_special_inode(inode, inode->i_mode, inode->i_rdev);
670 inode->i_op = &ceph_file_iops;
671 break;
672 case S_IFREG:
673 inode->i_op = &ceph_file_iops;
674 inode->i_fop = &ceph_file_fops;
675 break;
676 case S_IFLNK:
677 inode->i_op = &ceph_symlink_iops;
678 if (!ci->i_symlink) {
679 int symlen = iinfo->symlink_len;
680 char *sym;
681
682 BUG_ON(symlen != inode->i_size);
683 spin_unlock(&inode->i_lock);
684
685 err = -ENOMEM;
686 sym = kmalloc(symlen+1, GFP_NOFS);
687 if (!sym)
688 goto out;
689 memcpy(sym, iinfo->symlink, symlen);
690 sym[symlen] = 0;
691
692 spin_lock(&inode->i_lock);
693 if (!ci->i_symlink)
694 ci->i_symlink = sym;
695 else
696 kfree(sym); /* lost a race */
697 }
698 break;
699 case S_IFDIR:
700 inode->i_op = &ceph_dir_iops;
701 inode->i_fop = &ceph_dir_fops;
702
703 ci->i_dir_layout = iinfo->dir_layout;
704
705 ci->i_files = le64_to_cpu(info->files);
706 ci->i_subdirs = le64_to_cpu(info->subdirs);
707 ci->i_rbytes = le64_to_cpu(info->rbytes);
708 ci->i_rfiles = le64_to_cpu(info->rfiles);
709 ci->i_rsubdirs = le64_to_cpu(info->rsubdirs);
710 ceph_decode_timespec(&ci->i_rctime, &info->rctime);
711 break;
712 default:
713 pr_err("fill_inode %llx.%llx BAD mode 0%o\n",
714 ceph_vinop(inode), inode->i_mode);
715 }
716
717 no_change:
718 spin_unlock(&inode->i_lock);
719
720 /* queue truncate if we saw i_size decrease */
721 if (queue_trunc)
722 ceph_queue_vmtruncate(inode);
723
724 /* populate frag tree */
725 /* FIXME: move me up, if/when version reflects fragtree changes */
726 nsplits = le32_to_cpu(info->fragtree.nsplits);
727 mutex_lock(&ci->i_fragtree_mutex);
728 for (i = 0; i < nsplits; i++) {
729 u32 id = le32_to_cpu(info->fragtree.splits[i].frag);
730 struct ceph_inode_frag *frag = __get_or_create_frag(ci, id);
731
732 if (IS_ERR(frag))
733 continue;
734 frag->split_by = le32_to_cpu(info->fragtree.splits[i].by);
735 dout(" frag %x split by %d\n", frag->frag, frag->split_by);
736 }
737 mutex_unlock(&ci->i_fragtree_mutex);
738
739 /* were we issued a capability? */
740 if (info->cap.caps) {
741 if (ceph_snap(inode) == CEPH_NOSNAP) {
742 ceph_add_cap(inode, session,
743 le64_to_cpu(info->cap.cap_id),
744 cap_fmode,
745 le32_to_cpu(info->cap.caps),
746 le32_to_cpu(info->cap.wanted),
747 le32_to_cpu(info->cap.seq),
748 le32_to_cpu(info->cap.mseq),
749 le64_to_cpu(info->cap.realm),
750 info->cap.flags,
751 caps_reservation);
752 } else {
753 spin_lock(&inode->i_lock);
754 dout(" %p got snap_caps %s\n", inode,
755 ceph_cap_string(le32_to_cpu(info->cap.caps)));
756 ci->i_snap_caps |= le32_to_cpu(info->cap.caps);
757 if (cap_fmode >= 0)
758 __ceph_get_fmode(ci, cap_fmode);
759 spin_unlock(&inode->i_lock);
760 }
761 } else if (cap_fmode >= 0) {
762 pr_warning("mds issued no caps on %llx.%llx\n",
763 ceph_vinop(inode));
764 __ceph_get_fmode(ci, cap_fmode);
765 }
766
767 /* set dir completion flag? */
768 if (S_ISDIR(inode->i_mode) &&
769 updating_inode && /* didn't jump to no_change */
770 ci->i_files == 0 && ci->i_subdirs == 0 &&
771 ceph_snap(inode) == CEPH_NOSNAP &&
772 (le32_to_cpu(info->cap.caps) & CEPH_CAP_FILE_SHARED) &&
773 (issued & CEPH_CAP_FILE_EXCL) == 0 &&
774 !ceph_dir_test_complete(inode)) {
775 dout(" marking %p complete (empty)\n", inode);
776 ceph_dir_set_complete(inode);
777 ci->i_max_offset = 2;
778 }
779
780 /* update delegation info? */
781 if (dirinfo)
782 ceph_fill_dirfrag(inode, dirinfo);
783
784 err = 0;
785
786 out:
787 if (xattr_blob)
788 ceph_buffer_put(xattr_blob);
789 return err;
790 }
791
792 /*
793 * caller should hold session s_mutex.
794 */
795 static void update_dentry_lease(struct dentry *dentry,
796 struct ceph_mds_reply_lease *lease,
797 struct ceph_mds_session *session,
798 unsigned long from_time)
799 {
800 struct ceph_dentry_info *di = ceph_dentry(dentry);
801 long unsigned duration = le32_to_cpu(lease->duration_ms);
802 long unsigned ttl = from_time + (duration * HZ) / 1000;
803 long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000;
804 struct inode *dir;
805
806 /* only track leases on regular dentries */
807 if (dentry->d_op != &ceph_dentry_ops)
808 return;
809
810 spin_lock(&dentry->d_lock);
811 dout("update_dentry_lease %p duration %lu ms ttl %lu\n",
812 dentry, duration, ttl);
813
814 /* make lease_rdcache_gen match directory */
815 dir = dentry->d_parent->d_inode;
816 di->lease_shared_gen = ceph_inode(dir)->i_shared_gen;
817
818 if (duration == 0)
819 goto out_unlock;
820
821 if (di->lease_gen == session->s_cap_gen &&
822 time_before(ttl, dentry->d_time))
823 goto out_unlock; /* we already have a newer lease. */
824
825 if (di->lease_session && di->lease_session != session)
826 goto out_unlock;
827
828 ceph_dentry_lru_touch(dentry);
829
830 if (!di->lease_session)
831 di->lease_session = ceph_get_mds_session(session);
832 di->lease_gen = session->s_cap_gen;
833 di->lease_seq = le32_to_cpu(lease->seq);
834 di->lease_renew_after = half_ttl;
835 di->lease_renew_from = 0;
836 dentry->d_time = ttl;
837 out_unlock:
838 spin_unlock(&dentry->d_lock);
839 return;
840 }
841
842 /*
843 * Set dentry's directory position based on the current dir's max, and
844 * order it in d_subdirs, so that dcache_readdir behaves.
845 *
846 * Always called under directory's i_mutex.
847 */
848 static void ceph_set_dentry_offset(struct dentry *dn)
849 {
850 struct dentry *dir = dn->d_parent;
851 struct inode *inode = dir->d_inode;
852 struct ceph_dentry_info *di;
853
854 BUG_ON(!inode);
855
856 di = ceph_dentry(dn);
857
858 spin_lock(&inode->i_lock);
859 if (!ceph_dir_test_complete(inode)) {
860 spin_unlock(&inode->i_lock);
861 return;
862 }
863 di->offset = ceph_inode(inode)->i_max_offset++;
864 spin_unlock(&inode->i_lock);
865
866 spin_lock(&dir->d_lock);
867 spin_lock_nested(&dn->d_lock, DENTRY_D_LOCK_NESTED);
868 list_move(&dn->d_u.d_child, &dir->d_subdirs);
869 dout("set_dentry_offset %p %lld (%p %p)\n", dn, di->offset,
870 dn->d_u.d_child.prev, dn->d_u.d_child.next);
871 spin_unlock(&dn->d_lock);
872 spin_unlock(&dir->d_lock);
873 }
874
875 /*
876 * splice a dentry to an inode.
877 * caller must hold directory i_mutex for this to be safe.
878 *
879 * we will only rehash the resulting dentry if @prehash is
880 * true; @prehash will be set to false (for the benefit of
881 * the caller) if we fail.
882 */
883 static struct dentry *splice_dentry(struct dentry *dn, struct inode *in,
884 bool *prehash, bool set_offset)
885 {
886 struct dentry *realdn;
887
888 BUG_ON(dn->d_inode);
889
890 /* dn must be unhashed */
891 if (!d_unhashed(dn))
892 d_drop(dn);
893 realdn = d_materialise_unique(dn, in);
894 if (IS_ERR(realdn)) {
895 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
896 PTR_ERR(realdn), dn, in, ceph_vinop(in));
897 if (prehash)
898 *prehash = false; /* don't rehash on error */
899 dn = realdn; /* note realdn contains the error */
900 goto out;
901 } else if (realdn) {
902 dout("dn %p (%d) spliced with %p (%d) "
903 "inode %p ino %llx.%llx\n",
904 dn, dn->d_count,
905 realdn, realdn->d_count,
906 realdn->d_inode, ceph_vinop(realdn->d_inode));
907 dput(dn);
908 dn = realdn;
909 } else {
910 BUG_ON(!ceph_dentry(dn));
911 dout("dn %p attached to %p ino %llx.%llx\n",
912 dn, dn->d_inode, ceph_vinop(dn->d_inode));
913 }
914 if ((!prehash || *prehash) && d_unhashed(dn))
915 d_rehash(dn);
916 if (set_offset)
917 ceph_set_dentry_offset(dn);
918 out:
919 return dn;
920 }
921
922 /*
923 * Incorporate results into the local cache. This is either just
924 * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
925 * after a lookup).
926 *
927 * A reply may contain
928 * a directory inode along with a dentry.
929 * and/or a target inode
930 *
931 * Called with snap_rwsem (read).
932 */
933 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req,
934 struct ceph_mds_session *session)
935 {
936 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
937 struct inode *in = NULL;
938 struct ceph_mds_reply_inode *ininfo;
939 struct ceph_vino vino;
940 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
941 int i = 0;
942 int err = 0;
943
944 dout("fill_trace %p is_dentry %d is_target %d\n", req,
945 rinfo->head->is_dentry, rinfo->head->is_target);
946
947 #if 0
948 /*
949 * Debugging hook:
950 *
951 * If we resend completed ops to a recovering mds, we get no
952 * trace. Since that is very rare, pretend this is the case
953 * to ensure the 'no trace' handlers in the callers behave.
954 *
955 * Fill in inodes unconditionally to avoid breaking cap
956 * invariants.
957 */
958 if (rinfo->head->op & CEPH_MDS_OP_WRITE) {
959 pr_info("fill_trace faking empty trace on %lld %s\n",
960 req->r_tid, ceph_mds_op_name(rinfo->head->op));
961 if (rinfo->head->is_dentry) {
962 rinfo->head->is_dentry = 0;
963 err = fill_inode(req->r_locked_dir,
964 &rinfo->diri, rinfo->dirfrag,
965 session, req->r_request_started, -1);
966 }
967 if (rinfo->head->is_target) {
968 rinfo->head->is_target = 0;
969 ininfo = rinfo->targeti.in;
970 vino.ino = le64_to_cpu(ininfo->ino);
971 vino.snap = le64_to_cpu(ininfo->snapid);
972 in = ceph_get_inode(sb, vino);
973 err = fill_inode(in, &rinfo->targeti, NULL,
974 session, req->r_request_started,
975 req->r_fmode);
976 iput(in);
977 }
978 }
979 #endif
980
981 if (!rinfo->head->is_target && !rinfo->head->is_dentry) {
982 dout("fill_trace reply is empty!\n");
983 if (rinfo->head->result == 0 && req->r_locked_dir)
984 ceph_invalidate_dir_request(req);
985 return 0;
986 }
987
988 if (rinfo->head->is_dentry) {
989 struct inode *dir = req->r_locked_dir;
990
991 err = fill_inode(dir, &rinfo->diri, rinfo->dirfrag,
992 session, req->r_request_started, -1,
993 &req->r_caps_reservation);
994 if (err < 0)
995 return err;
996 }
997
998 /*
999 * ignore null lease/binding on snapdir ENOENT, or else we
1000 * will have trouble splicing in the virtual snapdir later
1001 */
1002 if (rinfo->head->is_dentry && !req->r_aborted &&
1003 (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name,
1004 fsc->mount_options->snapdir_name,
1005 req->r_dentry->d_name.len))) {
1006 /*
1007 * lookup link rename : null -> possibly existing inode
1008 * mknod symlink mkdir : null -> new inode
1009 * unlink : linked -> null
1010 */
1011 struct inode *dir = req->r_locked_dir;
1012 struct dentry *dn = req->r_dentry;
1013 bool have_dir_cap, have_lease;
1014
1015 BUG_ON(!dn);
1016 BUG_ON(!dir);
1017 BUG_ON(dn->d_parent->d_inode != dir);
1018 BUG_ON(ceph_ino(dir) !=
1019 le64_to_cpu(rinfo->diri.in->ino));
1020 BUG_ON(ceph_snap(dir) !=
1021 le64_to_cpu(rinfo->diri.in->snapid));
1022
1023 /* do we have a lease on the whole dir? */
1024 have_dir_cap =
1025 (le32_to_cpu(rinfo->diri.in->cap.caps) &
1026 CEPH_CAP_FILE_SHARED);
1027
1028 /* do we have a dn lease? */
1029 have_lease = have_dir_cap ||
1030 le32_to_cpu(rinfo->dlease->duration_ms);
1031 if (!have_lease)
1032 dout("fill_trace no dentry lease or dir cap\n");
1033
1034 /* rename? */
1035 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) {
1036 dout(" src %p '%.*s' dst %p '%.*s'\n",
1037 req->r_old_dentry,
1038 req->r_old_dentry->d_name.len,
1039 req->r_old_dentry->d_name.name,
1040 dn, dn->d_name.len, dn->d_name.name);
1041 dout("fill_trace doing d_move %p -> %p\n",
1042 req->r_old_dentry, dn);
1043
1044 d_move(req->r_old_dentry, dn);
1045 dout(" src %p '%.*s' dst %p '%.*s'\n",
1046 req->r_old_dentry,
1047 req->r_old_dentry->d_name.len,
1048 req->r_old_dentry->d_name.name,
1049 dn, dn->d_name.len, dn->d_name.name);
1050
1051 /* ensure target dentry is invalidated, despite
1052 rehashing bug in vfs_rename_dir */
1053 ceph_invalidate_dentry_lease(dn);
1054
1055 /*
1056 * d_move() puts the renamed dentry at the end of
1057 * d_subdirs. We need to assign it an appropriate
1058 * directory offset so we can behave when holding
1059 * D_COMPLETE.
1060 */
1061 ceph_set_dentry_offset(req->r_old_dentry);
1062 dout("dn %p gets new offset %lld\n", req->r_old_dentry,
1063 ceph_dentry(req->r_old_dentry)->offset);
1064
1065 dn = req->r_old_dentry; /* use old_dentry */
1066 in = dn->d_inode;
1067 }
1068
1069 /* null dentry? */
1070 if (!rinfo->head->is_target) {
1071 dout("fill_trace null dentry\n");
1072 if (dn->d_inode) {
1073 dout("d_delete %p\n", dn);
1074 d_delete(dn);
1075 } else {
1076 dout("d_instantiate %p NULL\n", dn);
1077 d_instantiate(dn, NULL);
1078 if (have_lease && d_unhashed(dn))
1079 d_rehash(dn);
1080 update_dentry_lease(dn, rinfo->dlease,
1081 session,
1082 req->r_request_started);
1083 }
1084 goto done;
1085 }
1086
1087 /* attach proper inode */
1088 ininfo = rinfo->targeti.in;
1089 vino.ino = le64_to_cpu(ininfo->ino);
1090 vino.snap = le64_to_cpu(ininfo->snapid);
1091 in = dn->d_inode;
1092 if (!in) {
1093 in = ceph_get_inode(sb, vino);
1094 if (IS_ERR(in)) {
1095 pr_err("fill_trace bad get_inode "
1096 "%llx.%llx\n", vino.ino, vino.snap);
1097 err = PTR_ERR(in);
1098 d_delete(dn);
1099 goto done;
1100 }
1101 dn = splice_dentry(dn, in, &have_lease, true);
1102 if (IS_ERR(dn)) {
1103 err = PTR_ERR(dn);
1104 goto done;
1105 }
1106 req->r_dentry = dn; /* may have spliced */
1107 ihold(in);
1108 } else if (ceph_ino(in) == vino.ino &&
1109 ceph_snap(in) == vino.snap) {
1110 ihold(in);
1111 } else {
1112 dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1113 dn, in, ceph_ino(in), ceph_snap(in),
1114 vino.ino, vino.snap);
1115 have_lease = false;
1116 in = NULL;
1117 }
1118
1119 if (have_lease)
1120 update_dentry_lease(dn, rinfo->dlease, session,
1121 req->r_request_started);
1122 dout(" final dn %p\n", dn);
1123 i++;
1124 } else if (req->r_op == CEPH_MDS_OP_LOOKUPSNAP ||
1125 req->r_op == CEPH_MDS_OP_MKSNAP) {
1126 struct dentry *dn = req->r_dentry;
1127
1128 /* fill out a snapdir LOOKUPSNAP dentry */
1129 BUG_ON(!dn);
1130 BUG_ON(!req->r_locked_dir);
1131 BUG_ON(ceph_snap(req->r_locked_dir) != CEPH_SNAPDIR);
1132 ininfo = rinfo->targeti.in;
1133 vino.ino = le64_to_cpu(ininfo->ino);
1134 vino.snap = le64_to_cpu(ininfo->snapid);
1135 in = ceph_get_inode(sb, vino);
1136 if (IS_ERR(in)) {
1137 pr_err("fill_inode get_inode badness %llx.%llx\n",
1138 vino.ino, vino.snap);
1139 err = PTR_ERR(in);
1140 d_delete(dn);
1141 goto done;
1142 }
1143 dout(" linking snapped dir %p to dn %p\n", in, dn);
1144 dn = splice_dentry(dn, in, NULL, true);
1145 if (IS_ERR(dn)) {
1146 err = PTR_ERR(dn);
1147 goto done;
1148 }
1149 req->r_dentry = dn; /* may have spliced */
1150 ihold(in);
1151 rinfo->head->is_dentry = 1; /* fool notrace handlers */
1152 }
1153
1154 if (rinfo->head->is_target) {
1155 vino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1156 vino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1157
1158 if (in == NULL || ceph_ino(in) != vino.ino ||
1159 ceph_snap(in) != vino.snap) {
1160 in = ceph_get_inode(sb, vino);
1161 if (IS_ERR(in)) {
1162 err = PTR_ERR(in);
1163 goto done;
1164 }
1165 }
1166 req->r_target_inode = in;
1167
1168 err = fill_inode(in,
1169 &rinfo->targeti, NULL,
1170 session, req->r_request_started,
1171 (le32_to_cpu(rinfo->head->result) == 0) ?
1172 req->r_fmode : -1,
1173 &req->r_caps_reservation);
1174 if (err < 0) {
1175 pr_err("fill_inode badness %p %llx.%llx\n",
1176 in, ceph_vinop(in));
1177 goto done;
1178 }
1179 }
1180
1181 done:
1182 dout("fill_trace done err=%d\n", err);
1183 return err;
1184 }
1185
1186 /*
1187 * Prepopulate our cache with readdir results, leases, etc.
1188 */
1189 int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1190 struct ceph_mds_session *session)
1191 {
1192 struct dentry *parent = req->r_dentry;
1193 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1194 struct qstr dname;
1195 struct dentry *dn;
1196 struct inode *in;
1197 int err = 0, i;
1198 struct inode *snapdir = NULL;
1199 struct ceph_mds_request_head *rhead = req->r_request->front.iov_base;
1200 u64 frag = le32_to_cpu(rhead->args.readdir.frag);
1201 struct ceph_dentry_info *di;
1202
1203 if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
1204 snapdir = ceph_get_snapdir(parent->d_inode);
1205 parent = d_find_alias(snapdir);
1206 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1207 rinfo->dir_nr, parent);
1208 } else {
1209 dout("readdir_prepopulate %d items under dn %p\n",
1210 rinfo->dir_nr, parent);
1211 if (rinfo->dir_dir)
1212 ceph_fill_dirfrag(parent->d_inode, rinfo->dir_dir);
1213 }
1214
1215 for (i = 0; i < rinfo->dir_nr; i++) {
1216 struct ceph_vino vino;
1217
1218 dname.name = rinfo->dir_dname[i];
1219 dname.len = rinfo->dir_dname_len[i];
1220 dname.hash = full_name_hash(dname.name, dname.len);
1221
1222 vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino);
1223 vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid);
1224
1225 retry_lookup:
1226 dn = d_lookup(parent, &dname);
1227 dout("d_lookup on parent=%p name=%.*s got %p\n",
1228 parent, dname.len, dname.name, dn);
1229
1230 if (!dn) {
1231 dn = d_alloc(parent, &dname);
1232 dout("d_alloc %p '%.*s' = %p\n", parent,
1233 dname.len, dname.name, dn);
1234 if (dn == NULL) {
1235 dout("d_alloc badness\n");
1236 err = -ENOMEM;
1237 goto out;
1238 }
1239 err = ceph_init_dentry(dn);
1240 if (err < 0) {
1241 dput(dn);
1242 goto out;
1243 }
1244 } else if (dn->d_inode &&
1245 (ceph_ino(dn->d_inode) != vino.ino ||
1246 ceph_snap(dn->d_inode) != vino.snap)) {
1247 dout(" dn %p points to wrong inode %p\n",
1248 dn, dn->d_inode);
1249 d_delete(dn);
1250 dput(dn);
1251 goto retry_lookup;
1252 } else {
1253 /* reorder parent's d_subdirs */
1254 spin_lock(&parent->d_lock);
1255 spin_lock_nested(&dn->d_lock, DENTRY_D_LOCK_NESTED);
1256 list_move(&dn->d_u.d_child, &parent->d_subdirs);
1257 spin_unlock(&dn->d_lock);
1258 spin_unlock(&parent->d_lock);
1259 }
1260
1261 di = dn->d_fsdata;
1262 di->offset = ceph_make_fpos(frag, i + req->r_readdir_offset);
1263
1264 /* inode */
1265 if (dn->d_inode) {
1266 in = dn->d_inode;
1267 } else {
1268 in = ceph_get_inode(parent->d_sb, vino);
1269 if (IS_ERR(in)) {
1270 dout("new_inode badness\n");
1271 d_delete(dn);
1272 dput(dn);
1273 err = PTR_ERR(in);
1274 goto out;
1275 }
1276 dn = splice_dentry(dn, in, NULL, false);
1277 if (IS_ERR(dn))
1278 dn = NULL;
1279 }
1280
1281 if (fill_inode(in, &rinfo->dir_in[i], NULL, session,
1282 req->r_request_started, -1,
1283 &req->r_caps_reservation) < 0) {
1284 pr_err("fill_inode badness on %p\n", in);
1285 goto next_item;
1286 }
1287 if (dn)
1288 update_dentry_lease(dn, rinfo->dir_dlease[i],
1289 req->r_session,
1290 req->r_request_started);
1291 next_item:
1292 if (dn)
1293 dput(dn);
1294 }
1295 req->r_did_prepopulate = true;
1296
1297 out:
1298 if (snapdir) {
1299 iput(snapdir);
1300 dput(parent);
1301 }
1302 dout("readdir_prepopulate done\n");
1303 return err;
1304 }
1305
1306 int ceph_inode_set_size(struct inode *inode, loff_t size)
1307 {
1308 struct ceph_inode_info *ci = ceph_inode(inode);
1309 int ret = 0;
1310
1311 spin_lock(&inode->i_lock);
1312 dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size);
1313 inode->i_size = size;
1314 inode->i_blocks = (size + (1 << 9) - 1) >> 9;
1315
1316 /* tell the MDS if we are approaching max_size */
1317 if ((size << 1) >= ci->i_max_size &&
1318 (ci->i_reported_size << 1) < ci->i_max_size)
1319 ret = 1;
1320
1321 spin_unlock(&inode->i_lock);
1322 return ret;
1323 }
1324
1325 /*
1326 * Write back inode data in a worker thread. (This can't be done
1327 * in the message handler context.)
1328 */
1329 void ceph_queue_writeback(struct inode *inode)
1330 {
1331 ihold(inode);
1332 if (queue_work(ceph_inode_to_client(inode)->wb_wq,
1333 &ceph_inode(inode)->i_wb_work)) {
1334 dout("ceph_queue_writeback %p\n", inode);
1335 } else {
1336 dout("ceph_queue_writeback %p failed\n", inode);
1337 iput(inode);
1338 }
1339 }
1340
1341 static void ceph_writeback_work(struct work_struct *work)
1342 {
1343 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1344 i_wb_work);
1345 struct inode *inode = &ci->vfs_inode;
1346
1347 dout("writeback %p\n", inode);
1348 filemap_fdatawrite(&inode->i_data);
1349 iput(inode);
1350 }
1351
1352 /*
1353 * queue an async invalidation
1354 */
1355 void ceph_queue_invalidate(struct inode *inode)
1356 {
1357 ihold(inode);
1358 if (queue_work(ceph_inode_to_client(inode)->pg_inv_wq,
1359 &ceph_inode(inode)->i_pg_inv_work)) {
1360 dout("ceph_queue_invalidate %p\n", inode);
1361 } else {
1362 dout("ceph_queue_invalidate %p failed\n", inode);
1363 iput(inode);
1364 }
1365 }
1366
1367 /*
1368 * Invalidate inode pages in a worker thread. (This can't be done
1369 * in the message handler context.)
1370 */
1371 static void ceph_invalidate_work(struct work_struct *work)
1372 {
1373 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1374 i_pg_inv_work);
1375 struct inode *inode = &ci->vfs_inode;
1376 u32 orig_gen;
1377 int check = 0;
1378
1379 spin_lock(&inode->i_lock);
1380 dout("invalidate_pages %p gen %d revoking %d\n", inode,
1381 ci->i_rdcache_gen, ci->i_rdcache_revoking);
1382 if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
1383 /* nevermind! */
1384 spin_unlock(&inode->i_lock);
1385 goto out;
1386 }
1387 orig_gen = ci->i_rdcache_gen;
1388 spin_unlock(&inode->i_lock);
1389
1390 truncate_inode_pages(&inode->i_data, 0);
1391
1392 spin_lock(&inode->i_lock);
1393 if (orig_gen == ci->i_rdcache_gen &&
1394 orig_gen == ci->i_rdcache_revoking) {
1395 dout("invalidate_pages %p gen %d successful\n", inode,
1396 ci->i_rdcache_gen);
1397 ci->i_rdcache_revoking--;
1398 check = 1;
1399 } else {
1400 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1401 inode, orig_gen, ci->i_rdcache_gen,
1402 ci->i_rdcache_revoking);
1403 }
1404 spin_unlock(&inode->i_lock);
1405
1406 if (check)
1407 ceph_check_caps(ci, 0, NULL);
1408 out:
1409 iput(inode);
1410 }
1411
1412
1413 /*
1414 * called by trunc_wq; take i_mutex ourselves
1415 *
1416 * We also truncate in a separate thread as well.
1417 */
1418 static void ceph_vmtruncate_work(struct work_struct *work)
1419 {
1420 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1421 i_vmtruncate_work);
1422 struct inode *inode = &ci->vfs_inode;
1423
1424 dout("vmtruncate_work %p\n", inode);
1425 mutex_lock(&inode->i_mutex);
1426 __ceph_do_pending_vmtruncate(inode);
1427 mutex_unlock(&inode->i_mutex);
1428 iput(inode);
1429 }
1430
1431 /*
1432 * Queue an async vmtruncate. If we fail to queue work, we will handle
1433 * the truncation the next time we call __ceph_do_pending_vmtruncate.
1434 */
1435 void ceph_queue_vmtruncate(struct inode *inode)
1436 {
1437 struct ceph_inode_info *ci = ceph_inode(inode);
1438
1439 ihold(inode);
1440 if (queue_work(ceph_sb_to_client(inode->i_sb)->trunc_wq,
1441 &ci->i_vmtruncate_work)) {
1442 dout("ceph_queue_vmtruncate %p\n", inode);
1443 } else {
1444 dout("ceph_queue_vmtruncate %p failed, pending=%d\n",
1445 inode, ci->i_truncate_pending);
1446 iput(inode);
1447 }
1448 }
1449
1450 /*
1451 * called with i_mutex held.
1452 *
1453 * Make sure any pending truncation is applied before doing anything
1454 * that may depend on it.
1455 */
1456 void __ceph_do_pending_vmtruncate(struct inode *inode)
1457 {
1458 struct ceph_inode_info *ci = ceph_inode(inode);
1459 u64 to;
1460 int wrbuffer_refs, wake = 0;
1461
1462 retry:
1463 spin_lock(&inode->i_lock);
1464 if (ci->i_truncate_pending == 0) {
1465 dout("__do_pending_vmtruncate %p none pending\n", inode);
1466 spin_unlock(&inode->i_lock);
1467 return;
1468 }
1469
1470 /*
1471 * make sure any dirty snapped pages are flushed before we
1472 * possibly truncate them.. so write AND block!
1473 */
1474 if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) {
1475 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1476 inode);
1477 spin_unlock(&inode->i_lock);
1478 filemap_write_and_wait_range(&inode->i_data, 0,
1479 inode->i_sb->s_maxbytes);
1480 goto retry;
1481 }
1482
1483 to = ci->i_truncate_size;
1484 wrbuffer_refs = ci->i_wrbuffer_ref;
1485 dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode,
1486 ci->i_truncate_pending, to);
1487 spin_unlock(&inode->i_lock);
1488
1489 truncate_inode_pages(inode->i_mapping, to);
1490
1491 spin_lock(&inode->i_lock);
1492 ci->i_truncate_pending--;
1493 if (ci->i_truncate_pending == 0)
1494 wake = 1;
1495 spin_unlock(&inode->i_lock);
1496
1497 if (wrbuffer_refs == 0)
1498 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
1499 if (wake)
1500 wake_up_all(&ci->i_cap_wq);
1501 }
1502
1503
1504 /*
1505 * symlinks
1506 */
1507 static void *ceph_sym_follow_link(struct dentry *dentry, struct nameidata *nd)
1508 {
1509 struct ceph_inode_info *ci = ceph_inode(dentry->d_inode);
1510 nd_set_link(nd, ci->i_symlink);
1511 return NULL;
1512 }
1513
1514 static const struct inode_operations ceph_symlink_iops = {
1515 .readlink = generic_readlink,
1516 .follow_link = ceph_sym_follow_link,
1517 };
1518
1519 /*
1520 * setattr
1521 */
1522 int ceph_setattr(struct dentry *dentry, struct iattr *attr)
1523 {
1524 struct inode *inode = dentry->d_inode;
1525 struct ceph_inode_info *ci = ceph_inode(inode);
1526 struct inode *parent_inode;
1527 const unsigned int ia_valid = attr->ia_valid;
1528 struct ceph_mds_request *req;
1529 struct ceph_mds_client *mdsc = ceph_sb_to_client(dentry->d_sb)->mdsc;
1530 int issued;
1531 int release = 0, dirtied = 0;
1532 int mask = 0;
1533 int err = 0;
1534 int inode_dirty_flags = 0;
1535
1536 if (ceph_snap(inode) != CEPH_NOSNAP)
1537 return -EROFS;
1538
1539 __ceph_do_pending_vmtruncate(inode);
1540
1541 err = inode_change_ok(inode, attr);
1542 if (err != 0)
1543 return err;
1544
1545 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR,
1546 USE_AUTH_MDS);
1547 if (IS_ERR(req))
1548 return PTR_ERR(req);
1549
1550 spin_lock(&inode->i_lock);
1551 issued = __ceph_caps_issued(ci, NULL);
1552 dout("setattr %p issued %s\n", inode, ceph_cap_string(issued));
1553
1554 if (ia_valid & ATTR_UID) {
1555 dout("setattr %p uid %d -> %d\n", inode,
1556 inode->i_uid, attr->ia_uid);
1557 if (issued & CEPH_CAP_AUTH_EXCL) {
1558 inode->i_uid = attr->ia_uid;
1559 dirtied |= CEPH_CAP_AUTH_EXCL;
1560 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1561 attr->ia_uid != inode->i_uid) {
1562 req->r_args.setattr.uid = cpu_to_le32(attr->ia_uid);
1563 mask |= CEPH_SETATTR_UID;
1564 release |= CEPH_CAP_AUTH_SHARED;
1565 }
1566 }
1567 if (ia_valid & ATTR_GID) {
1568 dout("setattr %p gid %d -> %d\n", inode,
1569 inode->i_gid, attr->ia_gid);
1570 if (issued & CEPH_CAP_AUTH_EXCL) {
1571 inode->i_gid = attr->ia_gid;
1572 dirtied |= CEPH_CAP_AUTH_EXCL;
1573 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1574 attr->ia_gid != inode->i_gid) {
1575 req->r_args.setattr.gid = cpu_to_le32(attr->ia_gid);
1576 mask |= CEPH_SETATTR_GID;
1577 release |= CEPH_CAP_AUTH_SHARED;
1578 }
1579 }
1580 if (ia_valid & ATTR_MODE) {
1581 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode,
1582 attr->ia_mode);
1583 if (issued & CEPH_CAP_AUTH_EXCL) {
1584 inode->i_mode = attr->ia_mode;
1585 dirtied |= CEPH_CAP_AUTH_EXCL;
1586 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1587 attr->ia_mode != inode->i_mode) {
1588 req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode);
1589 mask |= CEPH_SETATTR_MODE;
1590 release |= CEPH_CAP_AUTH_SHARED;
1591 }
1592 }
1593
1594 if (ia_valid & ATTR_ATIME) {
1595 dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode,
1596 inode->i_atime.tv_sec, inode->i_atime.tv_nsec,
1597 attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec);
1598 if (issued & CEPH_CAP_FILE_EXCL) {
1599 ci->i_time_warp_seq++;
1600 inode->i_atime = attr->ia_atime;
1601 dirtied |= CEPH_CAP_FILE_EXCL;
1602 } else if ((issued & CEPH_CAP_FILE_WR) &&
1603 timespec_compare(&inode->i_atime,
1604 &attr->ia_atime) < 0) {
1605 inode->i_atime = attr->ia_atime;
1606 dirtied |= CEPH_CAP_FILE_WR;
1607 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1608 !timespec_equal(&inode->i_atime, &attr->ia_atime)) {
1609 ceph_encode_timespec(&req->r_args.setattr.atime,
1610 &attr->ia_atime);
1611 mask |= CEPH_SETATTR_ATIME;
1612 release |= CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_RD |
1613 CEPH_CAP_FILE_WR;
1614 }
1615 }
1616 if (ia_valid & ATTR_MTIME) {
1617 dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode,
1618 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
1619 attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec);
1620 if (issued & CEPH_CAP_FILE_EXCL) {
1621 ci->i_time_warp_seq++;
1622 inode->i_mtime = attr->ia_mtime;
1623 dirtied |= CEPH_CAP_FILE_EXCL;
1624 } else if ((issued & CEPH_CAP_FILE_WR) &&
1625 timespec_compare(&inode->i_mtime,
1626 &attr->ia_mtime) < 0) {
1627 inode->i_mtime = attr->ia_mtime;
1628 dirtied |= CEPH_CAP_FILE_WR;
1629 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1630 !timespec_equal(&inode->i_mtime, &attr->ia_mtime)) {
1631 ceph_encode_timespec(&req->r_args.setattr.mtime,
1632 &attr->ia_mtime);
1633 mask |= CEPH_SETATTR_MTIME;
1634 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
1635 CEPH_CAP_FILE_WR;
1636 }
1637 }
1638 if (ia_valid & ATTR_SIZE) {
1639 dout("setattr %p size %lld -> %lld\n", inode,
1640 inode->i_size, attr->ia_size);
1641 if (attr->ia_size > inode->i_sb->s_maxbytes) {
1642 err = -EINVAL;
1643 goto out;
1644 }
1645 if ((issued & CEPH_CAP_FILE_EXCL) &&
1646 attr->ia_size > inode->i_size) {
1647 inode->i_size = attr->ia_size;
1648 inode->i_blocks =
1649 (attr->ia_size + (1 << 9) - 1) >> 9;
1650 inode->i_ctime = attr->ia_ctime;
1651 ci->i_reported_size = attr->ia_size;
1652 dirtied |= CEPH_CAP_FILE_EXCL;
1653 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1654 attr->ia_size != inode->i_size) {
1655 req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
1656 req->r_args.setattr.old_size =
1657 cpu_to_le64(inode->i_size);
1658 mask |= CEPH_SETATTR_SIZE;
1659 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
1660 CEPH_CAP_FILE_WR;
1661 }
1662 }
1663
1664 /* these do nothing */
1665 if (ia_valid & ATTR_CTIME) {
1666 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME|
1667 ATTR_MODE|ATTR_UID|ATTR_GID)) == 0;
1668 dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode,
1669 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
1670 attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec,
1671 only ? "ctime only" : "ignored");
1672 inode->i_ctime = attr->ia_ctime;
1673 if (only) {
1674 /*
1675 * if kernel wants to dirty ctime but nothing else,
1676 * we need to choose a cap to dirty under, or do
1677 * a almost-no-op setattr
1678 */
1679 if (issued & CEPH_CAP_AUTH_EXCL)
1680 dirtied |= CEPH_CAP_AUTH_EXCL;
1681 else if (issued & CEPH_CAP_FILE_EXCL)
1682 dirtied |= CEPH_CAP_FILE_EXCL;
1683 else if (issued & CEPH_CAP_XATTR_EXCL)
1684 dirtied |= CEPH_CAP_XATTR_EXCL;
1685 else
1686 mask |= CEPH_SETATTR_CTIME;
1687 }
1688 }
1689 if (ia_valid & ATTR_FILE)
1690 dout("setattr %p ATTR_FILE ... hrm!\n", inode);
1691
1692 if (dirtied) {
1693 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied);
1694 inode->i_ctime = CURRENT_TIME;
1695 }
1696
1697 release &= issued;
1698 spin_unlock(&inode->i_lock);
1699
1700 if (inode_dirty_flags)
1701 __mark_inode_dirty(inode, inode_dirty_flags);
1702
1703 if (mask) {
1704 req->r_inode = inode;
1705 ihold(inode);
1706 req->r_inode_drop = release;
1707 req->r_args.setattr.mask = cpu_to_le32(mask);
1708 req->r_num_caps = 1;
1709 parent_inode = ceph_get_dentry_parent_inode(dentry);
1710 err = ceph_mdsc_do_request(mdsc, parent_inode, req);
1711 iput(parent_inode);
1712 }
1713 dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err,
1714 ceph_cap_string(dirtied), mask);
1715
1716 ceph_mdsc_put_request(req);
1717 __ceph_do_pending_vmtruncate(inode);
1718 return err;
1719 out:
1720 spin_unlock(&inode->i_lock);
1721 ceph_mdsc_put_request(req);
1722 return err;
1723 }
1724
1725 /*
1726 * Verify that we have a lease on the given mask. If not,
1727 * do a getattr against an mds.
1728 */
1729 int ceph_do_getattr(struct inode *inode, int mask)
1730 {
1731 struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
1732 struct ceph_mds_client *mdsc = fsc->mdsc;
1733 struct ceph_mds_request *req;
1734 int err;
1735
1736 if (ceph_snap(inode) == CEPH_SNAPDIR) {
1737 dout("do_getattr inode %p SNAPDIR\n", inode);
1738 return 0;
1739 }
1740
1741 dout("do_getattr inode %p mask %s mode 0%o\n", inode, ceph_cap_string(mask), inode->i_mode);
1742 if (ceph_caps_issued_mask(ceph_inode(inode), mask, 1))
1743 return 0;
1744
1745 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
1746 if (IS_ERR(req))
1747 return PTR_ERR(req);
1748 req->r_inode = inode;
1749 ihold(inode);
1750 req->r_num_caps = 1;
1751 req->r_args.getattr.mask = cpu_to_le32(mask);
1752 err = ceph_mdsc_do_request(mdsc, NULL, req);
1753 ceph_mdsc_put_request(req);
1754 dout("do_getattr result=%d\n", err);
1755 return err;
1756 }
1757
1758
1759 /*
1760 * Check inode permissions. We verify we have a valid value for
1761 * the AUTH cap, then call the generic handler.
1762 */
1763 int ceph_permission(struct inode *inode, int mask)
1764 {
1765 int err;
1766
1767 if (mask & MAY_NOT_BLOCK)
1768 return -ECHILD;
1769
1770 err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED);
1771
1772 if (!err)
1773 err = generic_permission(inode, mask);
1774 return err;
1775 }
1776
1777 /*
1778 * Get all attributes. Hopefully somedata we'll have a statlite()
1779 * and can limit the fields we require to be accurate.
1780 */
1781 int ceph_getattr(struct vfsmount *mnt, struct dentry *dentry,
1782 struct kstat *stat)
1783 {
1784 struct inode *inode = dentry->d_inode;
1785 struct ceph_inode_info *ci = ceph_inode(inode);
1786 int err;
1787
1788 err = ceph_do_getattr(inode, CEPH_STAT_CAP_INODE_ALL);
1789 if (!err) {
1790 generic_fillattr(inode, stat);
1791 stat->ino = ceph_translate_ino(inode->i_sb, inode->i_ino);
1792 if (ceph_snap(inode) != CEPH_NOSNAP)
1793 stat->dev = ceph_snap(inode);
1794 else
1795 stat->dev = 0;
1796 if (S_ISDIR(inode->i_mode)) {
1797 if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb),
1798 RBYTES))
1799 stat->size = ci->i_rbytes;
1800 else
1801 stat->size = ci->i_files + ci->i_subdirs;
1802 stat->blocks = 0;
1803 stat->blksize = 65536;
1804 }
1805 }
1806 return err;
1807 }
kernel source for telekom puls (alcatel/mediatek)