?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
GUI: Fix Tomato RAF theme for all builds. Compilation typo.
[tomato.git] / release / src-rt-6.x.4708 / linux / linux-2.6.36 / fs / ceph / inode.c
blob53f99554f73d6701a81f59ee417c1e78b9b8e45c
1 #include "ceph_debug.h"
2
3 #include <linux/module.h>
4 #include <linux/fs.h>
5 #include <linux/smp_lock.h>
6 #include <linux/slab.h>
7 #include <linux/string.h>
8 #include <linux/uaccess.h>
9 #include <linux/kernel.h>
10 #include <linux/namei.h>
11 #include <linux/writeback.h>
12 #include <linux/vmalloc.h>
13 #include <linux/pagevec.h>
14
15 #include "super.h"
16 #include "decode.h"
17
18 /*
19 * Ceph inode operations
20 *
21 * Implement basic inode helpers (get, alloc) and inode ops (getattr,
22 * setattr, etc.), xattr helpers, and helpers for assimilating
23 * metadata returned by the MDS into our cache.
24 *
25 * Also define helpers for doing asynchronous writeback, invalidation,
26 * and truncation for the benefit of those who can't afford to block
27 * (typically because they are in the message handler path).
28 */
29
30 static const struct inode_operations ceph_symlink_iops;
31
32 static void ceph_invalidate_work(struct work_struct *work);
33 static void ceph_writeback_work(struct work_struct *work);
34 static void ceph_vmtruncate_work(struct work_struct *work);
35
36 /*
37 * find or create an inode, given the ceph ino number
38 */
39 struct inode *ceph_get_inode(struct super_block *sb, struct ceph_vino vino)
40 {
41 struct inode *inode;
42 ino_t t = ceph_vino_to_ino(vino);
43
44 inode = iget5_locked(sb, t, ceph_ino_compare, ceph_set_ino_cb, &vino);
45 if (inode == NULL)
46 return ERR_PTR(-ENOMEM);
47 if (inode->i_state & I_NEW) {
48 dout("get_inode created new inode %p %llx.%llx ino %llx\n",
49 inode, ceph_vinop(inode), (u64)inode->i_ino);
50 unlock_new_inode(inode);
51 }
52
53 dout("get_inode on %lu=%llx.%llx got %p\n", inode->i_ino, vino.ino,
54 vino.snap, inode);
55 return inode;
56 }
57
58 /*
59 * get/constuct snapdir inode for a given directory
60 */
61 struct inode *ceph_get_snapdir(struct inode *parent)
62 {
63 struct ceph_vino vino = {
64 .ino = ceph_ino(parent),
65 .snap = CEPH_SNAPDIR,
66 };
67 struct inode *inode = ceph_get_inode(parent->i_sb, vino);
68 struct ceph_inode_info *ci = ceph_inode(inode);
69
70 BUG_ON(!S_ISDIR(parent->i_mode));
71 if (IS_ERR(inode))
72 return inode;
73 inode->i_mode = parent->i_mode;
74 inode->i_uid = parent->i_uid;
75 inode->i_gid = parent->i_gid;
76 inode->i_op = &ceph_dir_iops;
77 inode->i_fop = &ceph_dir_fops;
78 ci->i_snap_caps = CEPH_CAP_PIN; /* so we can open */
79 ci->i_rbytes = 0;
80 return inode;
81 }
82
83 const struct inode_operations ceph_file_iops = {
84 .permission = ceph_permission,
85 .setattr = ceph_setattr,
86 .getattr = ceph_getattr,
87 .setxattr = ceph_setxattr,
88 .getxattr = ceph_getxattr,
89 .listxattr = ceph_listxattr,
90 .removexattr = ceph_removexattr,
91 };
92
93
94 /*
95 * We use a 'frag tree' to keep track of the MDS's directory fragments
96 * for a given inode (usually there is just a single fragment). We
97 * need to know when a child frag is delegated to a new MDS, or when
98 * it is flagged as replicated, so we can direct our requests
99 * accordingly.
100 */
101
102 /*
103 * find/create a frag in the tree
104 */
105 static struct ceph_inode_frag *__get_or_create_frag(struct ceph_inode_info *ci,
106 u32 f)
107 {
108 struct rb_node **p;
109 struct rb_node *parent = NULL;
110 struct ceph_inode_frag *frag;
111 int c;
112
113 p = &ci->i_fragtree.rb_node;
114 while (*p) {
115 parent = *p;
116 frag = rb_entry(parent, struct ceph_inode_frag, node);
117 c = ceph_frag_compare(f, frag->frag);
118 if (c < 0)
119 p = &(*p)->rb_left;
120 else if (c > 0)
121 p = &(*p)->rb_right;
122 else
123 return frag;
124 }
125
126 frag = kmalloc(sizeof(*frag), GFP_NOFS);
127 if (!frag) {
128 pr_err("__get_or_create_frag ENOMEM on %p %llx.%llx "
129 "frag %x\n", &ci->vfs_inode,
130 ceph_vinop(&ci->vfs_inode), f);
131 return ERR_PTR(-ENOMEM);
132 }
133 frag->frag = f;
134 frag->split_by = 0;
135 frag->mds = -1;
136 frag->ndist = 0;
137
138 rb_link_node(&frag->node, parent, p);
139 rb_insert_color(&frag->node, &ci->i_fragtree);
140
141 dout("get_or_create_frag added %llx.%llx frag %x\n",
142 ceph_vinop(&ci->vfs_inode), f);
143 return frag;
144 }
145
146 /*
147 * find a specific frag @f
148 */
149 struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci, u32 f)
150 {
151 struct rb_node *n = ci->i_fragtree.rb_node;
152
153 while (n) {
154 struct ceph_inode_frag *frag =
155 rb_entry(n, struct ceph_inode_frag, node);
156 int c = ceph_frag_compare(f, frag->frag);
157 if (c < 0)
158 n = n->rb_left;
159 else if (c > 0)
160 n = n->rb_right;
161 else
162 return frag;
163 }
164 return NULL;
165 }
166
167 /*
168 * Choose frag containing the given value @v. If @pfrag is
169 * specified, copy the frag delegation info to the caller if
170 * it is present.
171 */
172 u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
173 struct ceph_inode_frag *pfrag,
174 int *found)
175 {
176 u32 t = ceph_frag_make(0, 0);
177 struct ceph_inode_frag *frag;
178 unsigned nway, i;
179 u32 n;
180
181 if (found)
182 *found = 0;
183
184 mutex_lock(&ci->i_fragtree_mutex);
185 while (1) {
186 WARN_ON(!ceph_frag_contains_value(t, v));
187 frag = __ceph_find_frag(ci, t);
188 if (!frag)
189 break; /* t is a leaf */
190 if (frag->split_by == 0) {
191 if (pfrag)
192 memcpy(pfrag, frag, sizeof(*pfrag));
193 if (found)
194 *found = 1;
195 break;
196 }
197
198 /* choose child */
199 nway = 1 << frag->split_by;
200 dout("choose_frag(%x) %x splits by %d (%d ways)\n", v, t,
201 frag->split_by, nway);
202 for (i = 0; i < nway; i++) {
203 n = ceph_frag_make_child(t, frag->split_by, i);
204 if (ceph_frag_contains_value(n, v)) {
205 t = n;
206 break;
207 }
208 }
209 BUG_ON(i == nway);
210 }
211 dout("choose_frag(%x) = %x\n", v, t);
212
213 mutex_unlock(&ci->i_fragtree_mutex);
214 return t;
215 }
216
217 /*
218 * Process dirfrag (delegation) info from the mds. Include leaf
219 * fragment in tree ONLY if ndist > 0. Otherwise, only
220 * branches/splits are included in i_fragtree)
221 */
222 static int ceph_fill_dirfrag(struct inode *inode,
223 struct ceph_mds_reply_dirfrag *dirinfo)
224 {
225 struct ceph_inode_info *ci = ceph_inode(inode);
226 struct ceph_inode_frag *frag;
227 u32 id = le32_to_cpu(dirinfo->frag);
228 int mds = le32_to_cpu(dirinfo->auth);
229 int ndist = le32_to_cpu(dirinfo->ndist);
230 int i;
231 int err = 0;
232
233 mutex_lock(&ci->i_fragtree_mutex);
234 if (ndist == 0) {
235 /* no delegation info needed. */
236 frag = __ceph_find_frag(ci, id);
237 if (!frag)
238 goto out;
239 if (frag->split_by == 0) {
240 /* tree leaf, remove */
241 dout("fill_dirfrag removed %llx.%llx frag %x"
242 " (no ref)\n", ceph_vinop(inode), id);
243 rb_erase(&frag->node, &ci->i_fragtree);
244 kfree(frag);
245 } else {
246 /* tree branch, keep and clear */
247 dout("fill_dirfrag cleared %llx.%llx frag %x"
248 " referral\n", ceph_vinop(inode), id);
249 frag->mds = -1;
250 frag->ndist = 0;
251 }
252 goto out;
253 }
254
255
256 /* find/add this frag to store mds delegation info */
257 frag = __get_or_create_frag(ci, id);
258 if (IS_ERR(frag)) {
259 /* this is not the end of the world; we can continue
260 with bad/inaccurate delegation info */
261 pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n",
262 ceph_vinop(inode), le32_to_cpu(dirinfo->frag));
263 err = -ENOMEM;
264 goto out;
265 }
266
267 frag->mds = mds;
268 frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP);
269 for (i = 0; i < frag->ndist; i++)
270 frag->dist[i] = le32_to_cpu(dirinfo->dist[i]);
271 dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n",
272 ceph_vinop(inode), frag->frag, frag->ndist);
273
274 out:
275 mutex_unlock(&ci->i_fragtree_mutex);
276 return err;
277 }
278
279
280 /*
281 * initialize a newly allocated inode.
282 */
283 struct inode *ceph_alloc_inode(struct super_block *sb)
284 {
285 struct ceph_inode_info *ci;
286 int i;
287
288 ci = kmem_cache_alloc(ceph_inode_cachep, GFP_NOFS);
289 if (!ci)
290 return NULL;
291
292 dout("alloc_inode %p\n", &ci->vfs_inode);
293
294 ci->i_version = 0;
295 ci->i_time_warp_seq = 0;
296 ci->i_ceph_flags = 0;
297 ci->i_release_count = 0;
298 ci->i_symlink = NULL;
299
300 ci->i_fragtree = RB_ROOT;
301 mutex_init(&ci->i_fragtree_mutex);
302
303 ci->i_xattrs.blob = NULL;
304 ci->i_xattrs.prealloc_blob = NULL;
305 ci->i_xattrs.dirty = false;
306 ci->i_xattrs.index = RB_ROOT;
307 ci->i_xattrs.count = 0;
308 ci->i_xattrs.names_size = 0;
309 ci->i_xattrs.vals_size = 0;
310 ci->i_xattrs.version = 0;
311 ci->i_xattrs.index_version = 0;
312
313 ci->i_caps = RB_ROOT;
314 ci->i_auth_cap = NULL;
315 ci->i_dirty_caps = 0;
316 ci->i_flushing_caps = 0;
317 INIT_LIST_HEAD(&ci->i_dirty_item);
318 INIT_LIST_HEAD(&ci->i_flushing_item);
319 ci->i_cap_flush_seq = 0;
320 ci->i_cap_flush_last_tid = 0;
321 memset(&ci->i_cap_flush_tid, 0, sizeof(ci->i_cap_flush_tid));
322 init_waitqueue_head(&ci->i_cap_wq);
323 ci->i_hold_caps_min = 0;
324 ci->i_hold_caps_max = 0;
325 INIT_LIST_HEAD(&ci->i_cap_delay_list);
326 ci->i_cap_exporting_mds = 0;
327 ci->i_cap_exporting_mseq = 0;
328 ci->i_cap_exporting_issued = 0;
329 INIT_LIST_HEAD(&ci->i_cap_snaps);
330 ci->i_head_snapc = NULL;
331 ci->i_snap_caps = 0;
332
333 for (i = 0; i < CEPH_FILE_MODE_NUM; i++)
334 ci->i_nr_by_mode[i] = 0;
335
336 ci->i_truncate_seq = 0;
337 ci->i_truncate_size = 0;
338 ci->i_truncate_pending = 0;
339
340 ci->i_max_size = 0;
341 ci->i_reported_size = 0;
342 ci->i_wanted_max_size = 0;
343 ci->i_requested_max_size = 0;
344
345 ci->i_pin_ref = 0;
346 ci->i_rd_ref = 0;
347 ci->i_rdcache_ref = 0;
348 ci->i_wr_ref = 0;
349 ci->i_wrbuffer_ref = 0;
350 ci->i_wrbuffer_ref_head = 0;
351 ci->i_shared_gen = 0;
352 ci->i_rdcache_gen = 0;
353 ci->i_rdcache_revoking = 0;
354
355 INIT_LIST_HEAD(&ci->i_unsafe_writes);
356 INIT_LIST_HEAD(&ci->i_unsafe_dirops);
357 spin_lock_init(&ci->i_unsafe_lock);
358
359 ci->i_snap_realm = NULL;
360 INIT_LIST_HEAD(&ci->i_snap_realm_item);
361 INIT_LIST_HEAD(&ci->i_snap_flush_item);
362
363 INIT_WORK(&ci->i_wb_work, ceph_writeback_work);
364 INIT_WORK(&ci->i_pg_inv_work, ceph_invalidate_work);
365
366 INIT_WORK(&ci->i_vmtruncate_work, ceph_vmtruncate_work);
367
368 return &ci->vfs_inode;
369 }
370
371 void ceph_destroy_inode(struct inode *inode)
372 {
373 struct ceph_inode_info *ci = ceph_inode(inode);
374 struct ceph_inode_frag *frag;
375 struct rb_node *n;
376
377 dout("destroy_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode));
378
379 ceph_queue_caps_release(inode);
380
381 /*
382 * we may still have a snap_realm reference if there are stray
383 * caps in i_cap_exporting_issued or i_snap_caps.
384 */
385 if (ci->i_snap_realm) {
386 struct ceph_mds_client *mdsc =
387 &ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc;
388 struct ceph_snap_realm *realm = ci->i_snap_realm;
389
390 dout(" dropping residual ref to snap realm %p\n", realm);
391 spin_lock(&realm->inodes_with_caps_lock);
392 list_del_init(&ci->i_snap_realm_item);
393 spin_unlock(&realm->inodes_with_caps_lock);
394 ceph_put_snap_realm(mdsc, realm);
395 }
396
397 kfree(ci->i_symlink);
398 while ((n = rb_first(&ci->i_fragtree)) != NULL) {
399 frag = rb_entry(n, struct ceph_inode_frag, node);
400 rb_erase(n, &ci->i_fragtree);
401 kfree(frag);
402 }
403
404 __ceph_destroy_xattrs(ci);
405 if (ci->i_xattrs.blob)
406 ceph_buffer_put(ci->i_xattrs.blob);
407 if (ci->i_xattrs.prealloc_blob)
408 ceph_buffer_put(ci->i_xattrs.prealloc_blob);
409
410 kmem_cache_free(ceph_inode_cachep, ci);
411 }
412
413
414 /*
415 * Helpers to fill in size, ctime, mtime, and atime. We have to be
416 * careful because either the client or MDS may have more up to date
417 * info, depending on which capabilities are held, and whether
418 * time_warp_seq or truncate_seq have increased. (Ordinarily, mtime
419 * and size are monotonically increasing, except when utimes() or
420 * truncate() increments the corresponding _seq values.)
421 */
422 int ceph_fill_file_size(struct inode *inode, int issued,
423 u32 truncate_seq, u64 truncate_size, u64 size)
424 {
425 struct ceph_inode_info *ci = ceph_inode(inode);
426 int queue_trunc = 0;
427
428 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 ||
429 (truncate_seq == ci->i_truncate_seq && size > inode->i_size)) {
430 dout("size %lld -> %llu\n", inode->i_size, size);
431 inode->i_size = size;
432 inode->i_blocks = (size + (1<<9) - 1) >> 9;
433 ci->i_reported_size = size;
434 if (truncate_seq != ci->i_truncate_seq) {
435 dout("truncate_seq %u -> %u\n",
436 ci->i_truncate_seq, truncate_seq);
437 ci->i_truncate_seq = truncate_seq;
438 /*
439 * If we hold relevant caps, or in the case where we're
440 * not the only client referencing this file and we
441 * don't hold those caps, then we need to check whether
442 * the file is either opened or mmaped
443 */
444 if ((issued & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_RD|
445 CEPH_CAP_FILE_WR|CEPH_CAP_FILE_BUFFER|
446 CEPH_CAP_FILE_EXCL|
447 CEPH_CAP_FILE_LAZYIO)) ||
448 mapping_mapped(inode->i_mapping) ||
449 __ceph_caps_file_wanted(ci)) {
450 ci->i_truncate_pending++;
451 queue_trunc = 1;
452 }
453 }
454 }
455 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 &&
456 ci->i_truncate_size != truncate_size) {
457 dout("truncate_size %lld -> %llu\n", ci->i_truncate_size,
458 truncate_size);
459 ci->i_truncate_size = truncate_size;
460 }
461 return queue_trunc;
462 }
463
464 void ceph_fill_file_time(struct inode *inode, int issued,
465 u64 time_warp_seq, struct timespec *ctime,
466 struct timespec *mtime, struct timespec *atime)
467 {
468 struct ceph_inode_info *ci = ceph_inode(inode);
469 int warn = 0;
470
471 if (issued & (CEPH_CAP_FILE_EXCL|
472 CEPH_CAP_FILE_WR|
473 CEPH_CAP_FILE_BUFFER)) {
474 if (timespec_compare(ctime, &inode->i_ctime) > 0) {
475 dout("ctime %ld.%09ld -> %ld.%09ld inc w/ cap\n",
476 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
477 ctime->tv_sec, ctime->tv_nsec);
478 inode->i_ctime = *ctime;
479 }
480 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) {
481 /* the MDS did a utimes() */
482 dout("mtime %ld.%09ld -> %ld.%09ld "
483 "tw %d -> %d\n",
484 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
485 mtime->tv_sec, mtime->tv_nsec,
486 ci->i_time_warp_seq, (int)time_warp_seq);
487
488 inode->i_mtime = *mtime;
489 inode->i_atime = *atime;
490 ci->i_time_warp_seq = time_warp_seq;
491 } else if (time_warp_seq == ci->i_time_warp_seq) {
492 /* nobody did utimes(); take the max */
493 if (timespec_compare(mtime, &inode->i_mtime) > 0) {
494 dout("mtime %ld.%09ld -> %ld.%09ld inc\n",
495 inode->i_mtime.tv_sec,
496 inode->i_mtime.tv_nsec,
497 mtime->tv_sec, mtime->tv_nsec);
498 inode->i_mtime = *mtime;
499 }
500 if (timespec_compare(atime, &inode->i_atime) > 0) {
501 dout("atime %ld.%09ld -> %ld.%09ld inc\n",
502 inode->i_atime.tv_sec,
503 inode->i_atime.tv_nsec,
504 atime->tv_sec, atime->tv_nsec);
505 inode->i_atime = *atime;
506 }
507 } else if (issued & CEPH_CAP_FILE_EXCL) {
508 /* we did a utimes(); ignore mds values */
509 } else {
510 warn = 1;
511 }
512 } else {
513 /* we have no write caps; whatever the MDS says is true */
514 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) {
515 inode->i_ctime = *ctime;
516 inode->i_mtime = *mtime;
517 inode->i_atime = *atime;
518 ci->i_time_warp_seq = time_warp_seq;
519 } else {
520 warn = 1;
521 }
522 }
523 if (warn) /* time_warp_seq shouldn't go backwards */
524 dout("%p mds time_warp_seq %llu < %u\n",
525 inode, time_warp_seq, ci->i_time_warp_seq);
526 }
527
528 /*
529 * Populate an inode based on info from mds. May be called on new or
530 * existing inodes.
531 */
532 static int fill_inode(struct inode *inode,
533 struct ceph_mds_reply_info_in *iinfo,
534 struct ceph_mds_reply_dirfrag *dirinfo,
535 struct ceph_mds_session *session,
536 unsigned long ttl_from, int cap_fmode,
537 struct ceph_cap_reservation *caps_reservation)
538 {
539 struct ceph_mds_reply_inode *info = iinfo->in;
540 struct ceph_inode_info *ci = ceph_inode(inode);
541 int i;
542 int issued, implemented;
543 struct timespec mtime, atime, ctime;
544 u32 nsplits;
545 struct ceph_buffer *xattr_blob = NULL;
546 int err = 0;
547 int queue_trunc = 0;
548
549 dout("fill_inode %p ino %llx.%llx v %llu had %llu\n",
550 inode, ceph_vinop(inode), le64_to_cpu(info->version),
551 ci->i_version);
552
553 /*
554 * prealloc xattr data, if it looks like we'll need it. only
555 * if len > 4 (meaning there are actually xattrs; the first 4
556 * bytes are the xattr count).
557 */
558 if (iinfo->xattr_len > 4) {
559 xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS);
560 if (!xattr_blob)
561 pr_err("fill_inode ENOMEM xattr blob %d bytes\n",
562 iinfo->xattr_len);
563 }
564
565 spin_lock(&inode->i_lock);
566
567 /*
568 * provided version will be odd if inode value is projected,
569 * even if stable. skip the update if we have a newer info
570 * (e.g., due to inode info racing form multiple MDSs), or if
571 * we are getting projected (unstable) inode info.
572 */
573 if (le64_to_cpu(info->version) > 0 &&
574 (ci->i_version & ~1) > le64_to_cpu(info->version))
575 goto no_change;
576
577 issued = __ceph_caps_issued(ci, &implemented);
578 issued |= implemented | __ceph_caps_dirty(ci);
579
580 /* update inode */
581 ci->i_version = le64_to_cpu(info->version);
582 inode->i_version++;
583 inode->i_rdev = le32_to_cpu(info->rdev);
584
585 if ((issued & CEPH_CAP_AUTH_EXCL) == 0) {
586 inode->i_mode = le32_to_cpu(info->mode);
587 inode->i_uid = le32_to_cpu(info->uid);
588 inode->i_gid = le32_to_cpu(info->gid);
589 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
590 inode->i_uid, inode->i_gid);
591 }
592
593 if ((issued & CEPH_CAP_LINK_EXCL) == 0)
594 inode->i_nlink = le32_to_cpu(info->nlink);
595
596 /* be careful with mtime, atime, size */
597 ceph_decode_timespec(&atime, &info->atime);
598 ceph_decode_timespec(&mtime, &info->mtime);
599 ceph_decode_timespec(&ctime, &info->ctime);
600 queue_trunc = ceph_fill_file_size(inode, issued,
601 le32_to_cpu(info->truncate_seq),
602 le64_to_cpu(info->truncate_size),
603 le64_to_cpu(info->size));
604 ceph_fill_file_time(inode, issued,
605 le32_to_cpu(info->time_warp_seq),
606 &ctime, &mtime, &atime);
607
608 ci->i_max_size = le64_to_cpu(info->max_size);
609 ci->i_layout = info->layout;
610 inode->i_blkbits = fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1;
611
612 /* xattrs */
613 /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
614 if ((issued & CEPH_CAP_XATTR_EXCL) == 0 &&
615 le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) {
616 if (ci->i_xattrs.blob)
617 ceph_buffer_put(ci->i_xattrs.blob);
618 ci->i_xattrs.blob = xattr_blob;
619 if (xattr_blob)
620 memcpy(ci->i_xattrs.blob->vec.iov_base,
621 iinfo->xattr_data, iinfo->xattr_len);
622 ci->i_xattrs.version = le64_to_cpu(info->xattr_version);
623 xattr_blob = NULL;
624 }
625
626 inode->i_mapping->a_ops = &ceph_aops;
627 inode->i_mapping->backing_dev_info =
628 &ceph_sb_to_client(inode->i_sb)->backing_dev_info;
629
630 switch (inode->i_mode & S_IFMT) {
631 case S_IFIFO:
632 case S_IFBLK:
633 case S_IFCHR:
634 case S_IFSOCK:
635 init_special_inode(inode, inode->i_mode, inode->i_rdev);
636 inode->i_op = &ceph_file_iops;
637 break;
638 case S_IFREG:
639 inode->i_op = &ceph_file_iops;
640 inode->i_fop = &ceph_file_fops;
641 break;
642 case S_IFLNK:
643 inode->i_op = &ceph_symlink_iops;
644 if (!ci->i_symlink) {
645 int symlen = iinfo->symlink_len;
646 char *sym;
647
648 BUG_ON(symlen != inode->i_size);
649 spin_unlock(&inode->i_lock);
650
651 err = -ENOMEM;
652 sym = kmalloc(symlen+1, GFP_NOFS);
653 if (!sym)
654 goto out;
655 memcpy(sym, iinfo->symlink, symlen);
656 sym[symlen] = 0;
657
658 spin_lock(&inode->i_lock);
659 if (!ci->i_symlink)
660 ci->i_symlink = sym;
661 else
662 kfree(sym); /* lost a race */
663 }
664 break;
665 case S_IFDIR:
666 inode->i_op = &ceph_dir_iops;
667 inode->i_fop = &ceph_dir_fops;
668
669 ci->i_files = le64_to_cpu(info->files);
670 ci->i_subdirs = le64_to_cpu(info->subdirs);
671 ci->i_rbytes = le64_to_cpu(info->rbytes);
672 ci->i_rfiles = le64_to_cpu(info->rfiles);
673 ci->i_rsubdirs = le64_to_cpu(info->rsubdirs);
674 ceph_decode_timespec(&ci->i_rctime, &info->rctime);
675
676 /* set dir completion flag? */
677 if (ci->i_files == 0 && ci->i_subdirs == 0 &&
678 ceph_snap(inode) == CEPH_NOSNAP &&
679 (le32_to_cpu(info->cap.caps) & CEPH_CAP_FILE_SHARED) &&
680 (issued & CEPH_CAP_FILE_EXCL) == 0 &&
681 (ci->i_ceph_flags & CEPH_I_COMPLETE) == 0) {
682 dout(" marking %p complete (empty)\n", inode);
683 ci->i_ceph_flags |= CEPH_I_COMPLETE;
684 ci->i_max_offset = 2;
685 }
686
687 /* it may be better to set st_size in getattr instead? */
688 if (ceph_test_opt(ceph_sb_to_client(inode->i_sb), RBYTES))
689 inode->i_size = ci->i_rbytes;
690 break;
691 default:
692 pr_err("fill_inode %llx.%llx BAD mode 0%o\n",
693 ceph_vinop(inode), inode->i_mode);
694 }
695
696 no_change:
697 spin_unlock(&inode->i_lock);
698
699 /* queue truncate if we saw i_size decrease */
700 if (queue_trunc)
701 ceph_queue_vmtruncate(inode);
702
703 /* populate frag tree */
704 nsplits = le32_to_cpu(info->fragtree.nsplits);
705 mutex_lock(&ci->i_fragtree_mutex);
706 for (i = 0; i < nsplits; i++) {
707 u32 id = le32_to_cpu(info->fragtree.splits[i].frag);
708 struct ceph_inode_frag *frag = __get_or_create_frag(ci, id);
709
710 if (IS_ERR(frag))
711 continue;
712 frag->split_by = le32_to_cpu(info->fragtree.splits[i].by);
713 dout(" frag %x split by %d\n", frag->frag, frag->split_by);
714 }
715 mutex_unlock(&ci->i_fragtree_mutex);
716
717 /* were we issued a capability? */
718 if (info->cap.caps) {
719 if (ceph_snap(inode) == CEPH_NOSNAP) {
720 ceph_add_cap(inode, session,
721 le64_to_cpu(info->cap.cap_id),
722 cap_fmode,
723 le32_to_cpu(info->cap.caps),
724 le32_to_cpu(info->cap.wanted),
725 le32_to_cpu(info->cap.seq),
726 le32_to_cpu(info->cap.mseq),
727 le64_to_cpu(info->cap.realm),
728 info->cap.flags,
729 caps_reservation);
730 } else {
731 spin_lock(&inode->i_lock);
732 dout(" %p got snap_caps %s\n", inode,
733 ceph_cap_string(le32_to_cpu(info->cap.caps)));
734 ci->i_snap_caps |= le32_to_cpu(info->cap.caps);
735 if (cap_fmode >= 0)
736 __ceph_get_fmode(ci, cap_fmode);
737 spin_unlock(&inode->i_lock);
738 }
739 } else if (cap_fmode >= 0) {
740 pr_warning("mds issued no caps on %llx.%llx\n",
741 ceph_vinop(inode));
742 __ceph_get_fmode(ci, cap_fmode);
743 }
744
745 /* update delegation info? */
746 if (dirinfo)
747 ceph_fill_dirfrag(inode, dirinfo);
748
749 err = 0;
750
751 out:
752 if (xattr_blob)
753 ceph_buffer_put(xattr_blob);
754 return err;
755 }
756
757 /*
758 * caller should hold session s_mutex.
759 */
760 static void update_dentry_lease(struct dentry *dentry,
761 struct ceph_mds_reply_lease *lease,
762 struct ceph_mds_session *session,
763 unsigned long from_time)
764 {
765 struct ceph_dentry_info *di = ceph_dentry(dentry);
766 long unsigned duration = le32_to_cpu(lease->duration_ms);
767 long unsigned ttl = from_time + (duration * HZ) / 1000;
768 long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000;
769 struct inode *dir;
770
771 /* only track leases on regular dentries */
772 if (dentry->d_op != &ceph_dentry_ops)
773 return;
774
775 spin_lock(&dentry->d_lock);
776 dout("update_dentry_lease %p mask %d duration %lu ms ttl %lu\n",
777 dentry, le16_to_cpu(lease->mask), duration, ttl);
778
779 /* make lease_rdcache_gen match directory */
780 dir = dentry->d_parent->d_inode;
781 di->lease_shared_gen = ceph_inode(dir)->i_shared_gen;
782
783 if (lease->mask == 0)
784 goto out_unlock;
785
786 if (di->lease_gen == session->s_cap_gen &&
787 time_before(ttl, dentry->d_time))
788 goto out_unlock; /* we already have a newer lease. */
789
790 if (di->lease_session && di->lease_session != session)
791 goto out_unlock;
792
793 ceph_dentry_lru_touch(dentry);
794
795 if (!di->lease_session)
796 di->lease_session = ceph_get_mds_session(session);
797 di->lease_gen = session->s_cap_gen;
798 di->lease_seq = le32_to_cpu(lease->seq);
799 di->lease_renew_after = half_ttl;
800 di->lease_renew_from = 0;
801 dentry->d_time = ttl;
802 out_unlock:
803 spin_unlock(&dentry->d_lock);
804 return;
805 }
806
807 /*
808 * Set dentry's directory position based on the current dir's max, and
809 * order it in d_subdirs, so that dcache_readdir behaves.
810 */
811 static void ceph_set_dentry_offset(struct dentry *dn)
812 {
813 struct dentry *dir = dn->d_parent;
814 struct inode *inode = dn->d_parent->d_inode;
815 struct ceph_dentry_info *di;
816
817 BUG_ON(!inode);
818
819 di = ceph_dentry(dn);
820
821 spin_lock(&inode->i_lock);
822 if ((ceph_inode(inode)->i_ceph_flags & CEPH_I_COMPLETE) == 0) {
823 spin_unlock(&inode->i_lock);
824 return;
825 }
826 di->offset = ceph_inode(inode)->i_max_offset++;
827 spin_unlock(&inode->i_lock);
828
829 spin_lock(&dcache_lock);
830 spin_lock(&dn->d_lock);
831 list_move(&dn->d_u.d_child, &dir->d_subdirs);
832 dout("set_dentry_offset %p %lld (%p %p)\n", dn, di->offset,
833 dn->d_u.d_child.prev, dn->d_u.d_child.next);
834 spin_unlock(&dn->d_lock);
835 spin_unlock(&dcache_lock);
836 }
837
838 /*
839 * splice a dentry to an inode.
840 * caller must hold directory i_mutex for this to be safe.
841 *
842 * we will only rehash the resulting dentry if @prehash is
843 * true; @prehash will be set to false (for the benefit of
844 * the caller) if we fail.
845 */
846 static struct dentry *splice_dentry(struct dentry *dn, struct inode *in,
847 bool *prehash, bool set_offset)
848 {
849 struct dentry *realdn;
850
851 BUG_ON(dn->d_inode);
852
853 /* dn must be unhashed */
854 if (!d_unhashed(dn))
855 d_drop(dn);
856 realdn = d_materialise_unique(dn, in);
857 if (IS_ERR(realdn)) {
858 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
859 PTR_ERR(realdn), dn, in, ceph_vinop(in));
860 if (prehash)
861 *prehash = false; /* don't rehash on error */
862 dn = realdn; /* note realdn contains the error */
863 goto out;
864 } else if (realdn) {
865 dout("dn %p (%d) spliced with %p (%d) "
866 "inode %p ino %llx.%llx\n",
867 dn, atomic_read(&dn->d_count),
868 realdn, atomic_read(&realdn->d_count),
869 realdn->d_inode, ceph_vinop(realdn->d_inode));
870 dput(dn);
871 dn = realdn;
872 } else {
873 BUG_ON(!ceph_dentry(dn));
874 dout("dn %p attached to %p ino %llx.%llx\n",
875 dn, dn->d_inode, ceph_vinop(dn->d_inode));
876 }
877 if ((!prehash || *prehash) && d_unhashed(dn))
878 d_rehash(dn);
879 if (set_offset)
880 ceph_set_dentry_offset(dn);
881 out:
882 return dn;
883 }
884
885 /*
886 * Incorporate results into the local cache. This is either just
887 * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
888 * after a lookup).
889 *
890 * A reply may contain
891 * a directory inode along with a dentry.
892 * and/or a target inode
893 *
894 * Called with snap_rwsem (read).
895 */
896 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req,
897 struct ceph_mds_session *session)
898 {
899 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
900 struct inode *in = NULL;
901 struct ceph_mds_reply_inode *ininfo;
902 struct ceph_vino vino;
903 struct ceph_client *client = ceph_sb_to_client(sb);
904 int i = 0;
905 int err = 0;
906
907 dout("fill_trace %p is_dentry %d is_target %d\n", req,
908 rinfo->head->is_dentry, rinfo->head->is_target);
909
910
911 if (!rinfo->head->is_target && !rinfo->head->is_dentry) {
912 dout("fill_trace reply is empty!\n");
913 if (rinfo->head->result == 0 && req->r_locked_dir)
914 ceph_invalidate_dir_request(req);
915 return 0;
916 }
917
918 if (rinfo->head->is_dentry) {
919 struct inode *dir = req->r_locked_dir;
920
921 err = fill_inode(dir, &rinfo->diri, rinfo->dirfrag,
922 session, req->r_request_started, -1,
923 &req->r_caps_reservation);
924 if (err < 0)
925 return err;
926 }
927
928 /*
929 * ignore null lease/binding on snapdir ENOENT, or else we
930 * will have trouble splicing in the virtual snapdir later
931 */
932 if (rinfo->head->is_dentry && !req->r_aborted &&
933 (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name,
934 client->mount_args->snapdir_name,
935 req->r_dentry->d_name.len))) {
936 /*
937 * lookup link rename : null -> possibly existing inode
938 * mknod symlink mkdir : null -> new inode
939 * unlink : linked -> null
940 */
941 struct inode *dir = req->r_locked_dir;
942 struct dentry *dn = req->r_dentry;
943 bool have_dir_cap, have_lease;
944
945 BUG_ON(!dn);
946 BUG_ON(!dir);
947 BUG_ON(dn->d_parent->d_inode != dir);
948 BUG_ON(ceph_ino(dir) !=
949 le64_to_cpu(rinfo->diri.in->ino));
950 BUG_ON(ceph_snap(dir) !=
951 le64_to_cpu(rinfo->diri.in->snapid));
952
953 /* do we have a lease on the whole dir? */
954 have_dir_cap =
955 (le32_to_cpu(rinfo->diri.in->cap.caps) &
956 CEPH_CAP_FILE_SHARED);
957
958 /* do we have a dn lease? */
959 have_lease = have_dir_cap ||
960 (le16_to_cpu(rinfo->dlease->mask) &
961 CEPH_LOCK_DN);
962
963 if (!have_lease)
964 dout("fill_trace no dentry lease or dir cap\n");
965
966 /* rename? */
967 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) {
968 dout(" src %p '%.*s' dst %p '%.*s'\n",
969 req->r_old_dentry,
970 req->r_old_dentry->d_name.len,
971 req->r_old_dentry->d_name.name,
972 dn, dn->d_name.len, dn->d_name.name);
973 dout("fill_trace doing d_move %p -> %p\n",
974 req->r_old_dentry, dn);
975
976 /* d_move screws up d_subdirs order */
977 ceph_i_clear(dir, CEPH_I_COMPLETE);
978
979 d_move(req->r_old_dentry, dn);
980 dout(" src %p '%.*s' dst %p '%.*s'\n",
981 req->r_old_dentry,
982 req->r_old_dentry->d_name.len,
983 req->r_old_dentry->d_name.name,
984 dn, dn->d_name.len, dn->d_name.name);
985
986 /* ensure target dentry is invalidated, despite
987 rehashing bug in vfs_rename_dir */
988 ceph_invalidate_dentry_lease(dn);
989
990 /* take overwritten dentry's readdir offset */
991 dout("dn %p gets %p offset %lld (old offset %lld)\n",
992 req->r_old_dentry, dn, ceph_dentry(dn)->offset,
993 ceph_dentry(req->r_old_dentry)->offset);
994 ceph_dentry(req->r_old_dentry)->offset =
995 ceph_dentry(dn)->offset;
996
997 dn = req->r_old_dentry; /* use old_dentry */
998 in = dn->d_inode;
999 }
1000
1001 /* null dentry? */
1002 if (!rinfo->head->is_target) {
1003 dout("fill_trace null dentry\n");
1004 if (dn->d_inode) {
1005 dout("d_delete %p\n", dn);
1006 d_delete(dn);
1007 } else {
1008 dout("d_instantiate %p NULL\n", dn);
1009 d_instantiate(dn, NULL);
1010 if (have_lease && d_unhashed(dn))
1011 d_rehash(dn);
1012 update_dentry_lease(dn, rinfo->dlease,
1013 session,
1014 req->r_request_started);
1015 }
1016 goto done;
1017 }
1018
1019 /* attach proper inode */
1020 ininfo = rinfo->targeti.in;
1021 vino.ino = le64_to_cpu(ininfo->ino);
1022 vino.snap = le64_to_cpu(ininfo->snapid);
1023 if (!dn->d_inode) {
1024 in = ceph_get_inode(sb, vino);
1025 if (IS_ERR(in)) {
1026 pr_err("fill_trace bad get_inode "
1027 "%llx.%llx\n", vino.ino, vino.snap);
1028 err = PTR_ERR(in);
1029 d_delete(dn);
1030 goto done;
1031 }
1032 dn = splice_dentry(dn, in, &have_lease, true);
1033 if (IS_ERR(dn)) {
1034 err = PTR_ERR(dn);
1035 goto done;
1036 }
1037 req->r_dentry = dn; /* may have spliced */
1038 igrab(in);
1039 } else if (ceph_ino(in) == vino.ino &&
1040 ceph_snap(in) == vino.snap) {
1041 igrab(in);
1042 } else {
1043 dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1044 dn, in, ceph_ino(in), ceph_snap(in),
1045 vino.ino, vino.snap);
1046 have_lease = false;
1047 in = NULL;
1048 }
1049
1050 if (have_lease)
1051 update_dentry_lease(dn, rinfo->dlease, session,
1052 req->r_request_started);
1053 dout(" final dn %p\n", dn);
1054 i++;
1055 } else if (req->r_op == CEPH_MDS_OP_LOOKUPSNAP ||
1056 req->r_op == CEPH_MDS_OP_MKSNAP) {
1057 struct dentry *dn = req->r_dentry;
1058
1059 /* fill out a snapdir LOOKUPSNAP dentry */
1060 BUG_ON(!dn);
1061 BUG_ON(!req->r_locked_dir);
1062 BUG_ON(ceph_snap(req->r_locked_dir) != CEPH_SNAPDIR);
1063 ininfo = rinfo->targeti.in;
1064 vino.ino = le64_to_cpu(ininfo->ino);
1065 vino.snap = le64_to_cpu(ininfo->snapid);
1066 in = ceph_get_inode(sb, vino);
1067 if (IS_ERR(in)) {
1068 pr_err("fill_inode get_inode badness %llx.%llx\n",
1069 vino.ino, vino.snap);
1070 err = PTR_ERR(in);
1071 d_delete(dn);
1072 goto done;
1073 }
1074 dout(" linking snapped dir %p to dn %p\n", in, dn);
1075 dn = splice_dentry(dn, in, NULL, true);
1076 if (IS_ERR(dn)) {
1077 err = PTR_ERR(dn);
1078 goto done;
1079 }
1080 req->r_dentry = dn; /* may have spliced */
1081 igrab(in);
1082 rinfo->head->is_dentry = 1; /* fool notrace handlers */
1083 }
1084
1085 if (rinfo->head->is_target) {
1086 vino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1087 vino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1088
1089 if (in == NULL || ceph_ino(in) != vino.ino ||
1090 ceph_snap(in) != vino.snap) {
1091 in = ceph_get_inode(sb, vino);
1092 if (IS_ERR(in)) {
1093 err = PTR_ERR(in);
1094 goto done;
1095 }
1096 }
1097 req->r_target_inode = in;
1098
1099 err = fill_inode(in,
1100 &rinfo->targeti, NULL,
1101 session, req->r_request_started,
1102 (le32_to_cpu(rinfo->head->result) == 0) ?
1103 req->r_fmode : -1,
1104 &req->r_caps_reservation);
1105 if (err < 0) {
1106 pr_err("fill_inode badness %p %llx.%llx\n",
1107 in, ceph_vinop(in));
1108 goto done;
1109 }
1110 }
1111
1112 done:
1113 dout("fill_trace done err=%d\n", err);
1114 return err;
1115 }
1116
1117 /*
1118 * Prepopulate our cache with readdir results, leases, etc.
1119 */
1120 int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1121 struct ceph_mds_session *session)
1122 {
1123 struct dentry *parent = req->r_dentry;
1124 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1125 struct qstr dname;
1126 struct dentry *dn;
1127 struct inode *in;
1128 int err = 0, i;
1129 struct inode *snapdir = NULL;
1130 struct ceph_mds_request_head *rhead = req->r_request->front.iov_base;
1131 u64 frag = le32_to_cpu(rhead->args.readdir.frag);
1132 struct ceph_dentry_info *di;
1133
1134 if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
1135 snapdir = ceph_get_snapdir(parent->d_inode);
1136 parent = d_find_alias(snapdir);
1137 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1138 rinfo->dir_nr, parent);
1139 } else {
1140 dout("readdir_prepopulate %d items under dn %p\n",
1141 rinfo->dir_nr, parent);
1142 if (rinfo->dir_dir)
1143 ceph_fill_dirfrag(parent->d_inode, rinfo->dir_dir);
1144 }
1145
1146 for (i = 0; i < rinfo->dir_nr; i++) {
1147 struct ceph_vino vino;
1148
1149 dname.name = rinfo->dir_dname[i];
1150 dname.len = rinfo->dir_dname_len[i];
1151 dname.hash = full_name_hash(dname.name, dname.len);
1152
1153 vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino);
1154 vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid);
1155
1156 retry_lookup:
1157 dn = d_lookup(parent, &dname);
1158 dout("d_lookup on parent=%p name=%.*s got %p\n",
1159 parent, dname.len, dname.name, dn);
1160
1161 if (!dn) {
1162 dn = d_alloc(parent, &dname);
1163 dout("d_alloc %p '%.*s' = %p\n", parent,
1164 dname.len, dname.name, dn);
1165 if (dn == NULL) {
1166 dout("d_alloc badness\n");
1167 err = -ENOMEM;
1168 goto out;
1169 }
1170 err = ceph_init_dentry(dn);
1171 if (err < 0) {
1172 dput(dn);
1173 goto out;
1174 }
1175 } else if (dn->d_inode &&
1176 (ceph_ino(dn->d_inode) != vino.ino ||
1177 ceph_snap(dn->d_inode) != vino.snap)) {
1178 dout(" dn %p points to wrong inode %p\n",
1179 dn, dn->d_inode);
1180 d_delete(dn);
1181 dput(dn);
1182 goto retry_lookup;
1183 } else {
1184 /* reorder parent's d_subdirs */
1185 spin_lock(&dcache_lock);
1186 spin_lock(&dn->d_lock);
1187 list_move(&dn->d_u.d_child, &parent->d_subdirs);
1188 spin_unlock(&dn->d_lock);
1189 spin_unlock(&dcache_lock);
1190 }
1191
1192 di = dn->d_fsdata;
1193 di->offset = ceph_make_fpos(frag, i + req->r_readdir_offset);
1194
1195 /* inode */
1196 if (dn->d_inode) {
1197 in = dn->d_inode;
1198 } else {
1199 in = ceph_get_inode(parent->d_sb, vino);
1200 if (IS_ERR(in)) {
1201 dout("new_inode badness\n");
1202 d_delete(dn);
1203 dput(dn);
1204 err = PTR_ERR(in);
1205 goto out;
1206 }
1207 dn = splice_dentry(dn, in, NULL, false);
1208 if (IS_ERR(dn))
1209 dn = NULL;
1210 }
1211
1212 if (fill_inode(in, &rinfo->dir_in[i], NULL, session,
1213 req->r_request_started, -1,
1214 &req->r_caps_reservation) < 0) {
1215 pr_err("fill_inode badness on %p\n", in);
1216 goto next_item;
1217 }
1218 if (dn)
1219 update_dentry_lease(dn, rinfo->dir_dlease[i],
1220 req->r_session,
1221 req->r_request_started);
1222 next_item:
1223 if (dn)
1224 dput(dn);
1225 }
1226 req->r_did_prepopulate = true;
1227
1228 out:
1229 if (snapdir) {
1230 iput(snapdir);
1231 dput(parent);
1232 }
1233 dout("readdir_prepopulate done\n");
1234 return err;
1235 }
1236
1237 int ceph_inode_set_size(struct inode *inode, loff_t size)
1238 {
1239 struct ceph_inode_info *ci = ceph_inode(inode);
1240 int ret = 0;
1241
1242 spin_lock(&inode->i_lock);
1243 dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size);
1244 inode->i_size = size;
1245 inode->i_blocks = (size + (1 << 9) - 1) >> 9;
1246
1247 /* tell the MDS if we are approaching max_size */
1248 if ((size << 1) >= ci->i_max_size &&
1249 (ci->i_reported_size << 1) < ci->i_max_size)
1250 ret = 1;
1251
1252 spin_unlock(&inode->i_lock);
1253 return ret;
1254 }
1255
1256 /*
1257 * Write back inode data in a worker thread. (This can't be done
1258 * in the message handler context.)
1259 */
1260 void ceph_queue_writeback(struct inode *inode)
1261 {
1262 if (queue_work(ceph_inode_to_client(inode)->wb_wq,
1263 &ceph_inode(inode)->i_wb_work)) {
1264 dout("ceph_queue_writeback %p\n", inode);
1265 igrab(inode);
1266 } else {
1267 dout("ceph_queue_writeback %p failed\n", inode);
1268 }
1269 }
1270
1271 static void ceph_writeback_work(struct work_struct *work)
1272 {
1273 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1274 i_wb_work);
1275 struct inode *inode = &ci->vfs_inode;
1276
1277 dout("writeback %p\n", inode);
1278 filemap_fdatawrite(&inode->i_data);
1279 iput(inode);
1280 }
1281
1282 /*
1283 * queue an async invalidation
1284 */
1285 void ceph_queue_invalidate(struct inode *inode)
1286 {
1287 if (queue_work(ceph_inode_to_client(inode)->pg_inv_wq,
1288 &ceph_inode(inode)->i_pg_inv_work)) {
1289 dout("ceph_queue_invalidate %p\n", inode);
1290 igrab(inode);
1291 } else {
1292 dout("ceph_queue_invalidate %p failed\n", inode);
1293 }
1294 }
1295
1296 /*
1297 * invalidate any pages that are not dirty or under writeback. this
1298 * includes pages that are clean and mapped.
1299 */
1300 static void ceph_invalidate_nondirty_pages(struct address_space *mapping)
1301 {
1302 struct pagevec pvec;
1303 pgoff_t next = 0;
1304 int i;
1305
1306 pagevec_init(&pvec, 0);
1307 while (pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
1308 for (i = 0; i < pagevec_count(&pvec); i++) {
1309 struct page *page = pvec.pages[i];
1310 pgoff_t index;
1311 int skip_page =
1312 (PageDirty(page) || PageWriteback(page));
1313
1314 if (!skip_page)
1315 skip_page = !trylock_page(page);
1316
1317 /*
1318 * We really shouldn't be looking at the ->index of an
1319 * unlocked page. But we're not allowed to lock these
1320 * pages. So we rely upon nobody altering the ->index
1321 * of this (pinned-by-us) page.
1322 */
1323 index = page->index;
1324 if (index > next)
1325 next = index;
1326 next++;
1327
1328 if (skip_page)
1329 continue;
1330
1331 generic_error_remove_page(mapping, page);
1332 unlock_page(page);
1333 }
1334 pagevec_release(&pvec);
1335 cond_resched();
1336 }
1337 }
1338
1339 /*
1340 * Invalidate inode pages in a worker thread. (This can't be done
1341 * in the message handler context.)
1342 */
1343 static void ceph_invalidate_work(struct work_struct *work)
1344 {
1345 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1346 i_pg_inv_work);
1347 struct inode *inode = &ci->vfs_inode;
1348 u32 orig_gen;
1349 int check = 0;
1350
1351 spin_lock(&inode->i_lock);
1352 dout("invalidate_pages %p gen %d revoking %d\n", inode,
1353 ci->i_rdcache_gen, ci->i_rdcache_revoking);
1354 if (ci->i_rdcache_gen == 0 ||
1355 ci->i_rdcache_revoking != ci->i_rdcache_gen) {
1356 BUG_ON(ci->i_rdcache_revoking > ci->i_rdcache_gen);
1357 /* nevermind! */
1358 ci->i_rdcache_revoking = 0;
1359 spin_unlock(&inode->i_lock);
1360 goto out;
1361 }
1362 orig_gen = ci->i_rdcache_gen;
1363 spin_unlock(&inode->i_lock);
1364
1365 ceph_invalidate_nondirty_pages(inode->i_mapping);
1366
1367 spin_lock(&inode->i_lock);
1368 if (orig_gen == ci->i_rdcache_gen) {
1369 dout("invalidate_pages %p gen %d successful\n", inode,
1370 ci->i_rdcache_gen);
1371 ci->i_rdcache_gen = 0;
1372 ci->i_rdcache_revoking = 0;
1373 check = 1;
1374 } else {
1375 dout("invalidate_pages %p gen %d raced, gen now %d\n",
1376 inode, orig_gen, ci->i_rdcache_gen);
1377 }
1378 spin_unlock(&inode->i_lock);
1379
1380 if (check)
1381 ceph_check_caps(ci, 0, NULL);
1382 out:
1383 iput(inode);
1384 }
1385
1386
1387 /*
1388 * called by trunc_wq; take i_mutex ourselves
1389 *
1390 * We also truncate in a separate thread as well.
1391 */
1392 static void ceph_vmtruncate_work(struct work_struct *work)
1393 {
1394 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1395 i_vmtruncate_work);
1396 struct inode *inode = &ci->vfs_inode;
1397
1398 dout("vmtruncate_work %p\n", inode);
1399 mutex_lock(&inode->i_mutex);
1400 __ceph_do_pending_vmtruncate(inode);
1401 mutex_unlock(&inode->i_mutex);
1402 iput(inode);
1403 }
1404
1405 /*
1406 * Queue an async vmtruncate. If we fail to queue work, we will handle
1407 * the truncation the next time we call __ceph_do_pending_vmtruncate.
1408 */
1409 void ceph_queue_vmtruncate(struct inode *inode)
1410 {
1411 struct ceph_inode_info *ci = ceph_inode(inode);
1412
1413 if (queue_work(ceph_sb_to_client(inode->i_sb)->trunc_wq,
1414 &ci->i_vmtruncate_work)) {
1415 dout("ceph_queue_vmtruncate %p\n", inode);
1416 igrab(inode);
1417 } else {
1418 dout("ceph_queue_vmtruncate %p failed, pending=%d\n",
1419 inode, ci->i_truncate_pending);
1420 }
1421 }
1422
1423 /*
1424 * called with i_mutex held.
1425 *
1426 * Make sure any pending truncation is applied before doing anything
1427 * that may depend on it.
1428 */
1429 void __ceph_do_pending_vmtruncate(struct inode *inode)
1430 {
1431 struct ceph_inode_info *ci = ceph_inode(inode);
1432 u64 to;
1433 int wrbuffer_refs, wake = 0;
1434
1435 retry:
1436 spin_lock(&inode->i_lock);
1437 if (ci->i_truncate_pending == 0) {
1438 dout("__do_pending_vmtruncate %p none pending\n", inode);
1439 spin_unlock(&inode->i_lock);
1440 return;
1441 }
1442
1443 /*
1444 * make sure any dirty snapped pages are flushed before we
1445 * possibly truncate them.. so write AND block!
1446 */
1447 if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) {
1448 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1449 inode);
1450 spin_unlock(&inode->i_lock);
1451 filemap_write_and_wait_range(&inode->i_data, 0,
1452 inode->i_sb->s_maxbytes);
1453 goto retry;
1454 }
1455
1456 to = ci->i_truncate_size;
1457 wrbuffer_refs = ci->i_wrbuffer_ref;
1458 dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode,
1459 ci->i_truncate_pending, to);
1460 spin_unlock(&inode->i_lock);
1461
1462 truncate_inode_pages(inode->i_mapping, to);
1463
1464 spin_lock(&inode->i_lock);
1465 ci->i_truncate_pending--;
1466 if (ci->i_truncate_pending == 0)
1467 wake = 1;
1468 spin_unlock(&inode->i_lock);
1469
1470 if (wrbuffer_refs == 0)
1471 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
1472 if (wake)
1473 wake_up_all(&ci->i_cap_wq);
1474 }
1475
1476
1477 /*
1478 * symlinks
1479 */
1480 static void *ceph_sym_follow_link(struct dentry *dentry, struct nameidata *nd)
1481 {
1482 struct ceph_inode_info *ci = ceph_inode(dentry->d_inode);
1483 nd_set_link(nd, ci->i_symlink);
1484 return NULL;
1485 }
1486
1487 static const struct inode_operations ceph_symlink_iops = {
1488 .readlink = generic_readlink,
1489 .follow_link = ceph_sym_follow_link,
1490 };
1491
1492 /*
1493 * setattr
1494 */
1495 int ceph_setattr(struct dentry *dentry, struct iattr *attr)
1496 {
1497 struct inode *inode = dentry->d_inode;
1498 struct ceph_inode_info *ci = ceph_inode(inode);
1499 struct inode *parent_inode = dentry->d_parent->d_inode;
1500 const unsigned int ia_valid = attr->ia_valid;
1501 struct ceph_mds_request *req;
1502 struct ceph_mds_client *mdsc = &ceph_sb_to_client(dentry->d_sb)->mdsc;
1503 int issued;
1504 int release = 0, dirtied = 0;
1505 int mask = 0;
1506 int err = 0;
1507
1508 if (ceph_snap(inode) != CEPH_NOSNAP)
1509 return -EROFS;
1510
1511 __ceph_do_pending_vmtruncate(inode);
1512
1513 err = inode_change_ok(inode, attr);
1514 if (err != 0)
1515 return err;
1516
1517 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR,
1518 USE_AUTH_MDS);
1519 if (IS_ERR(req))
1520 return PTR_ERR(req);
1521
1522 spin_lock(&inode->i_lock);
1523 issued = __ceph_caps_issued(ci, NULL);
1524 dout("setattr %p issued %s\n", inode, ceph_cap_string(issued));
1525
1526 if (ia_valid & ATTR_UID) {
1527 dout("setattr %p uid %d -> %d\n", inode,
1528 inode->i_uid, attr->ia_uid);
1529 if (issued & CEPH_CAP_AUTH_EXCL) {
1530 inode->i_uid = attr->ia_uid;
1531 dirtied |= CEPH_CAP_AUTH_EXCL;
1532 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1533 attr->ia_uid != inode->i_uid) {
1534 req->r_args.setattr.uid = cpu_to_le32(attr->ia_uid);
1535 mask |= CEPH_SETATTR_UID;
1536 release |= CEPH_CAP_AUTH_SHARED;
1537 }
1538 }
1539 if (ia_valid & ATTR_GID) {
1540 dout("setattr %p gid %d -> %d\n", inode,
1541 inode->i_gid, attr->ia_gid);
1542 if (issued & CEPH_CAP_AUTH_EXCL) {
1543 inode->i_gid = attr->ia_gid;
1544 dirtied |= CEPH_CAP_AUTH_EXCL;
1545 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1546 attr->ia_gid != inode->i_gid) {
1547 req->r_args.setattr.gid = cpu_to_le32(attr->ia_gid);
1548 mask |= CEPH_SETATTR_GID;
1549 release |= CEPH_CAP_AUTH_SHARED;
1550 }
1551 }
1552 if (ia_valid & ATTR_MODE) {
1553 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode,
1554 attr->ia_mode);
1555 if (issued & CEPH_CAP_AUTH_EXCL) {
1556 inode->i_mode = attr->ia_mode;
1557 dirtied |= CEPH_CAP_AUTH_EXCL;
1558 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1559 attr->ia_mode != inode->i_mode) {
1560 req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode);
1561 mask |= CEPH_SETATTR_MODE;
1562 release |= CEPH_CAP_AUTH_SHARED;
1563 }
1564 }
1565
1566 if (ia_valid & ATTR_ATIME) {
1567 dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode,
1568 inode->i_atime.tv_sec, inode->i_atime.tv_nsec,
1569 attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec);
1570 if (issued & CEPH_CAP_FILE_EXCL) {
1571 ci->i_time_warp_seq++;
1572 inode->i_atime = attr->ia_atime;
1573 dirtied |= CEPH_CAP_FILE_EXCL;
1574 } else if ((issued & CEPH_CAP_FILE_WR) &&
1575 timespec_compare(&inode->i_atime,
1576 &attr->ia_atime) < 0) {
1577 inode->i_atime = attr->ia_atime;
1578 dirtied |= CEPH_CAP_FILE_WR;
1579 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1580 !timespec_equal(&inode->i_atime, &attr->ia_atime)) {
1581 ceph_encode_timespec(&req->r_args.setattr.atime,
1582 &attr->ia_atime);
1583 mask |= CEPH_SETATTR_ATIME;
1584 release |= CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_RD |
1585 CEPH_CAP_FILE_WR;
1586 }
1587 }
1588 if (ia_valid & ATTR_MTIME) {
1589 dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode,
1590 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
1591 attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec);
1592 if (issued & CEPH_CAP_FILE_EXCL) {
1593 ci->i_time_warp_seq++;
1594 inode->i_mtime = attr->ia_mtime;
1595 dirtied |= CEPH_CAP_FILE_EXCL;
1596 } else if ((issued & CEPH_CAP_FILE_WR) &&
1597 timespec_compare(&inode->i_mtime,
1598 &attr->ia_mtime) < 0) {
1599 inode->i_mtime = attr->ia_mtime;
1600 dirtied |= CEPH_CAP_FILE_WR;
1601 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1602 !timespec_equal(&inode->i_mtime, &attr->ia_mtime)) {
1603 ceph_encode_timespec(&req->r_args.setattr.mtime,
1604 &attr->ia_mtime);
1605 mask |= CEPH_SETATTR_MTIME;
1606 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
1607 CEPH_CAP_FILE_WR;
1608 }
1609 }
1610 if (ia_valid & ATTR_SIZE) {
1611 dout("setattr %p size %lld -> %lld\n", inode,
1612 inode->i_size, attr->ia_size);
1613 if (attr->ia_size > inode->i_sb->s_maxbytes) {
1614 err = -EINVAL;
1615 goto out;
1616 }
1617 if ((issued & CEPH_CAP_FILE_EXCL) &&
1618 attr->ia_size > inode->i_size) {
1619 inode->i_size = attr->ia_size;
1620 inode->i_blocks =
1621 (attr->ia_size + (1 << 9) - 1) >> 9;
1622 inode->i_ctime = attr->ia_ctime;
1623 ci->i_reported_size = attr->ia_size;
1624 dirtied |= CEPH_CAP_FILE_EXCL;
1625 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1626 attr->ia_size != inode->i_size) {
1627 req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
1628 req->r_args.setattr.old_size =
1629 cpu_to_le64(inode->i_size);
1630 mask |= CEPH_SETATTR_SIZE;
1631 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
1632 CEPH_CAP_FILE_WR;
1633 }
1634 }
1635
1636 /* these do nothing */
1637 if (ia_valid & ATTR_CTIME) {
1638 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME|
1639 ATTR_MODE|ATTR_UID|ATTR_GID)) == 0;
1640 dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode,
1641 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
1642 attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec,
1643 only ? "ctime only" : "ignored");
1644 inode->i_ctime = attr->ia_ctime;
1645 if (only) {
1646 /*
1647 * if kernel wants to dirty ctime but nothing else,
1648 * we need to choose a cap to dirty under, or do
1649 * a almost-no-op setattr
1650 */
1651 if (issued & CEPH_CAP_AUTH_EXCL)
1652 dirtied |= CEPH_CAP_AUTH_EXCL;
1653 else if (issued & CEPH_CAP_FILE_EXCL)
1654 dirtied |= CEPH_CAP_FILE_EXCL;
1655 else if (issued & CEPH_CAP_XATTR_EXCL)
1656 dirtied |= CEPH_CAP_XATTR_EXCL;
1657 else
1658 mask |= CEPH_SETATTR_CTIME;
1659 }
1660 }
1661 if (ia_valid & ATTR_FILE)
1662 dout("setattr %p ATTR_FILE ... hrm!\n", inode);
1663
1664 if (dirtied) {
1665 __ceph_mark_dirty_caps(ci, dirtied);
1666 inode->i_ctime = CURRENT_TIME;
1667 }
1668
1669 release &= issued;
1670 spin_unlock(&inode->i_lock);
1671
1672 if (mask) {
1673 req->r_inode = igrab(inode);
1674 req->r_inode_drop = release;
1675 req->r_args.setattr.mask = cpu_to_le32(mask);
1676 req->r_num_caps = 1;
1677 err = ceph_mdsc_do_request(mdsc, parent_inode, req);
1678 }
1679 dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err,
1680 ceph_cap_string(dirtied), mask);
1681
1682 ceph_mdsc_put_request(req);
1683 __ceph_do_pending_vmtruncate(inode);
1684 return err;
1685 out:
1686 spin_unlock(&inode->i_lock);
1687 ceph_mdsc_put_request(req);
1688 return err;
1689 }
1690
1691 /*
1692 * Verify that we have a lease on the given mask. If not,
1693 * do a getattr against an mds.
1694 */
1695 int ceph_do_getattr(struct inode *inode, int mask)
1696 {
1697 struct ceph_client *client = ceph_sb_to_client(inode->i_sb);
1698 struct ceph_mds_client *mdsc = &client->mdsc;
1699 struct ceph_mds_request *req;
1700 int err;
1701
1702 if (ceph_snap(inode) == CEPH_SNAPDIR) {
1703 dout("do_getattr inode %p SNAPDIR\n", inode);
1704 return 0;
1705 }
1706
1707 dout("do_getattr inode %p mask %s\n", inode, ceph_cap_string(mask));
1708 if (ceph_caps_issued_mask(ceph_inode(inode), mask, 1))
1709 return 0;
1710
1711 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
1712 if (IS_ERR(req))
1713 return PTR_ERR(req);
1714 req->r_inode = igrab(inode);
1715 req->r_num_caps = 1;
1716 req->r_args.getattr.mask = cpu_to_le32(mask);
1717 err = ceph_mdsc_do_request(mdsc, NULL, req);
1718 ceph_mdsc_put_request(req);
1719 dout("do_getattr result=%d\n", err);
1720 return err;
1721 }
1722
1723
1724 /*
1725 * Check inode permissions. We verify we have a valid value for
1726 * the AUTH cap, then call the generic handler.
1727 */
1728 int ceph_permission(struct inode *inode, int mask)
1729 {
1730 int err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED);
1731
1732 if (!err)
1733 err = generic_permission(inode, mask, NULL);
1734 return err;
1735 }
1736
1737 /*
1738 * Get all attributes. Hopefully somedata we'll have a statlite()
1739 * and can limit the fields we require to be accurate.
1740 */
1741 int ceph_getattr(struct vfsmount *mnt, struct dentry *dentry,
1742 struct kstat *stat)
1743 {
1744 struct inode *inode = dentry->d_inode;
1745 struct ceph_inode_info *ci = ceph_inode(inode);
1746 int err;
1747
1748 err = ceph_do_getattr(inode, CEPH_STAT_CAP_INODE_ALL);
1749 if (!err) {
1750 generic_fillattr(inode, stat);
1751 stat->ino = inode->i_ino;
1752 if (ceph_snap(inode) != CEPH_NOSNAP)
1753 stat->dev = ceph_snap(inode);
1754 else
1755 stat->dev = 0;
1756 if (S_ISDIR(inode->i_mode)) {
1757 stat->size = ci->i_rbytes;
1758 stat->blocks = 0;
1759 stat->blksize = 65536;
1760 }
1761 }
1762 return err;
1763 }
Tomato firmware and modifications.