495decfc4b34849bf3b401cbcc07036de5eaaa63
[linux-2.6-block.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/writeback.h>
10 #include <linux/vmalloc.h>
11 #include <linux/posix_acl.h>
12 #include <linux/random.h>
13
14 #include "super.h"
15 #include "mds_client.h"
16 #include "cache.h"
17 #include <linux/ceph/decode.h>
18
19 /*
20  * Ceph inode operations
21  *
22  * Implement basic inode helpers (get, alloc) and inode ops (getattr,
23  * setattr, etc.), xattr helpers, and helpers for assimilating
24  * metadata returned by the MDS into our cache.
25  *
26  * Also define helpers for doing asynchronous writeback, invalidation,
27  * and truncation for the benefit of those who can't afford to block
28  * (typically because they are in the message handler path).
29  */
30
31 static const struct inode_operations ceph_symlink_iops;
32
33 static void ceph_invalidate_work(struct work_struct *work);
34 static void ceph_writeback_work(struct work_struct *work);
35 static void ceph_vmtruncate_work(struct work_struct *work);
36
37 /*
38  * find or create an inode, given the ceph ino number
39  */
40 static int ceph_set_ino_cb(struct inode *inode, void *data)
41 {
42         ceph_inode(inode)->i_vino = *(struct ceph_vino *)data;
43         inode->i_ino = ceph_vino_to_ino(*(struct ceph_vino *)data);
44         return 0;
45 }
46
47 struct inode *ceph_get_inode(struct super_block *sb, struct ceph_vino vino)
48 {
49         struct inode *inode;
50         ino_t t = ceph_vino_to_ino(vino);
51
52         inode = iget5_locked(sb, t, ceph_ino_compare, ceph_set_ino_cb, &vino);
53         if (inode == NULL)
54                 return ERR_PTR(-ENOMEM);
55         if (inode->i_state & I_NEW) {
56                 dout("get_inode created new inode %p %llx.%llx ino %llx\n",
57                      inode, ceph_vinop(inode), (u64)inode->i_ino);
58                 unlock_new_inode(inode);
59         }
60
61         dout("get_inode on %lu=%llx.%llx got %p\n", inode->i_ino, vino.ino,
62              vino.snap, inode);
63         return inode;
64 }
65
66 /*
67  * get/constuct snapdir inode for a given directory
68  */
69 struct inode *ceph_get_snapdir(struct inode *parent)
70 {
71         struct ceph_vino vino = {
72                 .ino = ceph_ino(parent),
73                 .snap = CEPH_SNAPDIR,
74         };
75         struct inode *inode = ceph_get_inode(parent->i_sb, vino);
76         struct ceph_inode_info *ci = ceph_inode(inode);
77
78         BUG_ON(!S_ISDIR(parent->i_mode));
79         if (IS_ERR(inode))
80                 return inode;
81         inode->i_mode = parent->i_mode;
82         inode->i_uid = parent->i_uid;
83         inode->i_gid = parent->i_gid;
84         inode->i_op = &ceph_snapdir_iops;
85         inode->i_fop = &ceph_snapdir_fops;
86         ci->i_snap_caps = CEPH_CAP_PIN; /* so we can open */
87         ci->i_rbytes = 0;
88         return inode;
89 }
90
91 const struct inode_operations ceph_file_iops = {
92         .permission = ceph_permission,
93         .setattr = ceph_setattr,
94         .getattr = ceph_getattr,
95         .setxattr = ceph_setxattr,
96         .getxattr = ceph_getxattr,
97         .listxattr = ceph_listxattr,
98         .removexattr = ceph_removexattr,
99         .get_acl = ceph_get_acl,
100         .set_acl = ceph_set_acl,
101 };
102
103
104 /*
105  * We use a 'frag tree' to keep track of the MDS's directory fragments
106  * for a given inode (usually there is just a single fragment).  We
107  * need to know when a child frag is delegated to a new MDS, or when
108  * it is flagged as replicated, so we can direct our requests
109  * accordingly.
110  */
111
112 /*
113  * find/create a frag in the tree
114  */
115 static struct ceph_inode_frag *__get_or_create_frag(struct ceph_inode_info *ci,
116                                                     u32 f)
117 {
118         struct rb_node **p;
119         struct rb_node *parent = NULL;
120         struct ceph_inode_frag *frag;
121         int c;
122
123         p = &ci->i_fragtree.rb_node;
124         while (*p) {
125                 parent = *p;
126                 frag = rb_entry(parent, struct ceph_inode_frag, node);
127                 c = ceph_frag_compare(f, frag->frag);
128                 if (c < 0)
129                         p = &(*p)->rb_left;
130                 else if (c > 0)
131                         p = &(*p)->rb_right;
132                 else
133                         return frag;
134         }
135
136         frag = kmalloc(sizeof(*frag), GFP_NOFS);
137         if (!frag) {
138                 pr_err("__get_or_create_frag ENOMEM on %p %llx.%llx "
139                        "frag %x\n", &ci->vfs_inode,
140                        ceph_vinop(&ci->vfs_inode), f);
141                 return ERR_PTR(-ENOMEM);
142         }
143         frag->frag = f;
144         frag->split_by = 0;
145         frag->mds = -1;
146         frag->ndist = 0;
147
148         rb_link_node(&frag->node, parent, p);
149         rb_insert_color(&frag->node, &ci->i_fragtree);
150
151         dout("get_or_create_frag added %llx.%llx frag %x\n",
152              ceph_vinop(&ci->vfs_inode), f);
153         return frag;
154 }
155
156 /*
157  * find a specific frag @f
158  */
159 struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci, u32 f)
160 {
161         struct rb_node *n = ci->i_fragtree.rb_node;
162
163         while (n) {
164                 struct ceph_inode_frag *frag =
165                         rb_entry(n, struct ceph_inode_frag, node);
166                 int c = ceph_frag_compare(f, frag->frag);
167                 if (c < 0)
168                         n = n->rb_left;
169                 else if (c > 0)
170                         n = n->rb_right;
171                 else
172                         return frag;
173         }
174         return NULL;
175 }
176
177 /*
178  * Choose frag containing the given value @v.  If @pfrag is
179  * specified, copy the frag delegation info to the caller if
180  * it is present.
181  */
182 static u32 __ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
183                               struct ceph_inode_frag *pfrag, int *found)
184 {
185         u32 t = ceph_frag_make(0, 0);
186         struct ceph_inode_frag *frag;
187         unsigned nway, i;
188         u32 n;
189
190         if (found)
191                 *found = 0;
192
193         while (1) {
194                 WARN_ON(!ceph_frag_contains_value(t, v));
195                 frag = __ceph_find_frag(ci, t);
196                 if (!frag)
197                         break; /* t is a leaf */
198                 if (frag->split_by == 0) {
199                         if (pfrag)
200                                 memcpy(pfrag, frag, sizeof(*pfrag));
201                         if (found)
202                                 *found = 1;
203                         break;
204                 }
205
206                 /* choose child */
207                 nway = 1 << frag->split_by;
208                 dout("choose_frag(%x) %x splits by %d (%d ways)\n", v, t,
209                      frag->split_by, nway);
210                 for (i = 0; i < nway; i++) {
211                         n = ceph_frag_make_child(t, frag->split_by, i);
212                         if (ceph_frag_contains_value(n, v)) {
213                                 t = n;
214                                 break;
215                         }
216                 }
217                 BUG_ON(i == nway);
218         }
219         dout("choose_frag(%x) = %x\n", v, t);
220
221         return t;
222 }
223
224 u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
225                      struct ceph_inode_frag *pfrag, int *found)
226 {
227         u32 ret;
228         mutex_lock(&ci->i_fragtree_mutex);
229         ret = __ceph_choose_frag(ci, v, pfrag, found);
230         mutex_unlock(&ci->i_fragtree_mutex);
231         return ret;
232 }
233
234 /*
235  * Process dirfrag (delegation) info from the mds.  Include leaf
236  * fragment in tree ONLY if ndist > 0.  Otherwise, only
237  * branches/splits are included in i_fragtree)
238  */
239 static int ceph_fill_dirfrag(struct inode *inode,
240                              struct ceph_mds_reply_dirfrag *dirinfo)
241 {
242         struct ceph_inode_info *ci = ceph_inode(inode);
243         struct ceph_inode_frag *frag;
244         u32 id = le32_to_cpu(dirinfo->frag);
245         int mds = le32_to_cpu(dirinfo->auth);
246         int ndist = le32_to_cpu(dirinfo->ndist);
247         int diri_auth = -1;
248         int i;
249         int err = 0;
250
251         spin_lock(&ci->i_ceph_lock);
252         if (ci->i_auth_cap)
253                 diri_auth = ci->i_auth_cap->mds;
254         spin_unlock(&ci->i_ceph_lock);
255
256         mutex_lock(&ci->i_fragtree_mutex);
257         if (ndist == 0 && mds == diri_auth) {
258                 /* no delegation info needed. */
259                 frag = __ceph_find_frag(ci, id);
260                 if (!frag)
261                         goto out;
262                 if (frag->split_by == 0) {
263                         /* tree leaf, remove */
264                         dout("fill_dirfrag removed %llx.%llx frag %x"
265                              " (no ref)\n", ceph_vinop(inode), id);
266                         rb_erase(&frag->node, &ci->i_fragtree);
267                         kfree(frag);
268                 } else {
269                         /* tree branch, keep and clear */
270                         dout("fill_dirfrag cleared %llx.%llx frag %x"
271                              " referral\n", ceph_vinop(inode), id);
272                         frag->mds = -1;
273                         frag->ndist = 0;
274                 }
275                 goto out;
276         }
277
278
279         /* find/add this frag to store mds delegation info */
280         frag = __get_or_create_frag(ci, id);
281         if (IS_ERR(frag)) {
282                 /* this is not the end of the world; we can continue
283                    with bad/inaccurate delegation info */
284                 pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n",
285                        ceph_vinop(inode), le32_to_cpu(dirinfo->frag));
286                 err = -ENOMEM;
287                 goto out;
288         }
289
290         frag->mds = mds;
291         frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP);
292         for (i = 0; i < frag->ndist; i++)
293                 frag->dist[i] = le32_to_cpu(dirinfo->dist[i]);
294         dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n",
295              ceph_vinop(inode), frag->frag, frag->ndist);
296
297 out:
298         mutex_unlock(&ci->i_fragtree_mutex);
299         return err;
300 }
301
302 static int ceph_fill_fragtree(struct inode *inode,
303                               struct ceph_frag_tree_head *fragtree,
304                               struct ceph_mds_reply_dirfrag *dirinfo)
305 {
306         struct ceph_inode_info *ci = ceph_inode(inode);
307         struct ceph_inode_frag *frag;
308         struct rb_node *rb_node;
309         int i;
310         u32 id, nsplits;
311         bool update = false;
312
313         mutex_lock(&ci->i_fragtree_mutex);
314         nsplits = le32_to_cpu(fragtree->nsplits);
315         if (nsplits) {
316                 i = prandom_u32() % nsplits;
317                 id = le32_to_cpu(fragtree->splits[i].frag);
318                 if (!__ceph_find_frag(ci, id))
319                         update = true;
320         } else if (!RB_EMPTY_ROOT(&ci->i_fragtree)) {
321                 rb_node = rb_first(&ci->i_fragtree);
322                 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
323                 if (frag->frag != ceph_frag_make(0, 0) || rb_next(rb_node))
324                         update = true;
325         }
326         if (!update && dirinfo) {
327                 id = le32_to_cpu(dirinfo->frag);
328                 if (id != __ceph_choose_frag(ci, id, NULL, NULL))
329                         update = true;
330         }
331         if (!update)
332                 goto out_unlock;
333
334         dout("fill_fragtree %llx.%llx\n", ceph_vinop(inode));
335         rb_node = rb_first(&ci->i_fragtree);
336         for (i = 0; i < nsplits; i++) {
337                 id = le32_to_cpu(fragtree->splits[i].frag);
338                 frag = NULL;
339                 while (rb_node) {
340                         frag = rb_entry(rb_node, struct ceph_inode_frag, node);
341                         if (ceph_frag_compare(frag->frag, id) >= 0) {
342                                 if (frag->frag != id)
343                                         frag = NULL;
344                                 else
345                                         rb_node = rb_next(rb_node);
346                                 break;
347                         }
348                         rb_node = rb_next(rb_node);
349                         rb_erase(&frag->node, &ci->i_fragtree);
350                         kfree(frag);
351                         frag = NULL;
352                 }
353                 if (!frag) {
354                         frag = __get_or_create_frag(ci, id);
355                         if (IS_ERR(frag))
356                                 continue;
357                 }
358                 frag->split_by = le32_to_cpu(fragtree->splits[i].by);
359                 dout(" frag %x split by %d\n", frag->frag, frag->split_by);
360         }
361         while (rb_node) {
362                 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
363                 rb_node = rb_next(rb_node);
364                 rb_erase(&frag->node, &ci->i_fragtree);
365                 kfree(frag);
366         }
367 out_unlock:
368         mutex_unlock(&ci->i_fragtree_mutex);
369         return 0;
370 }
371
372 /*
373  * initialize a newly allocated inode.
374  */
375 struct inode *ceph_alloc_inode(struct super_block *sb)
376 {
377         struct ceph_inode_info *ci;
378         int i;
379
380         ci = kmem_cache_alloc(ceph_inode_cachep, GFP_NOFS);
381         if (!ci)
382                 return NULL;
383
384         dout("alloc_inode %p\n", &ci->vfs_inode);
385
386         spin_lock_init(&ci->i_ceph_lock);
387
388         ci->i_version = 0;
389         ci->i_inline_version = 0;
390         ci->i_time_warp_seq = 0;
391         ci->i_ceph_flags = 0;
392         atomic64_set(&ci->i_ordered_count, 1);
393         atomic64_set(&ci->i_release_count, 1);
394         atomic64_set(&ci->i_complete_seq[0], 0);
395         atomic64_set(&ci->i_complete_seq[1], 0);
396         ci->i_symlink = NULL;
397
398         memset(&ci->i_dir_layout, 0, sizeof(ci->i_dir_layout));
399         ci->i_pool_ns_len = 0;
400
401         ci->i_fragtree = RB_ROOT;
402         mutex_init(&ci->i_fragtree_mutex);
403
404         ci->i_xattrs.blob = NULL;
405         ci->i_xattrs.prealloc_blob = NULL;
406         ci->i_xattrs.dirty = false;
407         ci->i_xattrs.index = RB_ROOT;
408         ci->i_xattrs.count = 0;
409         ci->i_xattrs.names_size = 0;
410         ci->i_xattrs.vals_size = 0;
411         ci->i_xattrs.version = 0;
412         ci->i_xattrs.index_version = 0;
413
414         ci->i_caps = RB_ROOT;
415         ci->i_auth_cap = NULL;
416         ci->i_dirty_caps = 0;
417         ci->i_flushing_caps = 0;
418         INIT_LIST_HEAD(&ci->i_dirty_item);
419         INIT_LIST_HEAD(&ci->i_flushing_item);
420         ci->i_prealloc_cap_flush = NULL;
421         ci->i_cap_flush_tree = RB_ROOT;
422         init_waitqueue_head(&ci->i_cap_wq);
423         ci->i_hold_caps_min = 0;
424         ci->i_hold_caps_max = 0;
425         INIT_LIST_HEAD(&ci->i_cap_delay_list);
426         INIT_LIST_HEAD(&ci->i_cap_snaps);
427         ci->i_head_snapc = NULL;
428         ci->i_snap_caps = 0;
429
430         for (i = 0; i < CEPH_FILE_MODE_NUM; i++)
431                 ci->i_nr_by_mode[i] = 0;
432
433         mutex_init(&ci->i_truncate_mutex);
434         ci->i_truncate_seq = 0;
435         ci->i_truncate_size = 0;
436         ci->i_truncate_pending = 0;
437
438         ci->i_max_size = 0;
439         ci->i_reported_size = 0;
440         ci->i_wanted_max_size = 0;
441         ci->i_requested_max_size = 0;
442
443         ci->i_pin_ref = 0;
444         ci->i_rd_ref = 0;
445         ci->i_rdcache_ref = 0;
446         ci->i_wr_ref = 0;
447         ci->i_wb_ref = 0;
448         ci->i_wrbuffer_ref = 0;
449         ci->i_wrbuffer_ref_head = 0;
450         ci->i_shared_gen = 0;
451         ci->i_rdcache_gen = 0;
452         ci->i_rdcache_revoking = 0;
453
454         INIT_LIST_HEAD(&ci->i_unsafe_writes);
455         INIT_LIST_HEAD(&ci->i_unsafe_dirops);
456         INIT_LIST_HEAD(&ci->i_unsafe_iops);
457         spin_lock_init(&ci->i_unsafe_lock);
458
459         ci->i_snap_realm = NULL;
460         INIT_LIST_HEAD(&ci->i_snap_realm_item);
461         INIT_LIST_HEAD(&ci->i_snap_flush_item);
462
463         INIT_WORK(&ci->i_wb_work, ceph_writeback_work);
464         INIT_WORK(&ci->i_pg_inv_work, ceph_invalidate_work);
465
466         INIT_WORK(&ci->i_vmtruncate_work, ceph_vmtruncate_work);
467
468         ceph_fscache_inode_init(ci);
469
470         return &ci->vfs_inode;
471 }
472
473 static void ceph_i_callback(struct rcu_head *head)
474 {
475         struct inode *inode = container_of(head, struct inode, i_rcu);
476         struct ceph_inode_info *ci = ceph_inode(inode);
477
478         kmem_cache_free(ceph_inode_cachep, ci);
479 }
480
481 void ceph_destroy_inode(struct inode *inode)
482 {
483         struct ceph_inode_info *ci = ceph_inode(inode);
484         struct ceph_inode_frag *frag;
485         struct rb_node *n;
486
487         dout("destroy_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode));
488
489         ceph_fscache_unregister_inode_cookie(ci);
490
491         ceph_queue_caps_release(inode);
492
493         /*
494          * we may still have a snap_realm reference if there are stray
495          * caps in i_snap_caps.
496          */
497         if (ci->i_snap_realm) {
498                 struct ceph_mds_client *mdsc =
499                         ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc;
500                 struct ceph_snap_realm *realm = ci->i_snap_realm;
501
502                 dout(" dropping residual ref to snap realm %p\n", realm);
503                 spin_lock(&realm->inodes_with_caps_lock);
504                 list_del_init(&ci->i_snap_realm_item);
505                 spin_unlock(&realm->inodes_with_caps_lock);
506                 ceph_put_snap_realm(mdsc, realm);
507         }
508
509         kfree(ci->i_symlink);
510         while ((n = rb_first(&ci->i_fragtree)) != NULL) {
511                 frag = rb_entry(n, struct ceph_inode_frag, node);
512                 rb_erase(n, &ci->i_fragtree);
513                 kfree(frag);
514         }
515
516         __ceph_destroy_xattrs(ci);
517         if (ci->i_xattrs.blob)
518                 ceph_buffer_put(ci->i_xattrs.blob);
519         if (ci->i_xattrs.prealloc_blob)
520                 ceph_buffer_put(ci->i_xattrs.prealloc_blob);
521
522         call_rcu(&inode->i_rcu, ceph_i_callback);
523 }
524
525 int ceph_drop_inode(struct inode *inode)
526 {
527         /*
528          * Positve dentry and corresponding inode are always accompanied
529          * in MDS reply. So no need to keep inode in the cache after
530          * dropping all its aliases.
531          */
532         return 1;
533 }
534
535 /*
536  * Helpers to fill in size, ctime, mtime, and atime.  We have to be
537  * careful because either the client or MDS may have more up to date
538  * info, depending on which capabilities are held, and whether
539  * time_warp_seq or truncate_seq have increased.  (Ordinarily, mtime
540  * and size are monotonically increasing, except when utimes() or
541  * truncate() increments the corresponding _seq values.)
542  */
543 int ceph_fill_file_size(struct inode *inode, int issued,
544                         u32 truncate_seq, u64 truncate_size, u64 size)
545 {
546         struct ceph_inode_info *ci = ceph_inode(inode);
547         int queue_trunc = 0;
548
549         if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 ||
550             (truncate_seq == ci->i_truncate_seq && size > inode->i_size)) {
551                 dout("size %lld -> %llu\n", inode->i_size, size);
552                 i_size_write(inode, size);
553                 inode->i_blocks = (size + (1<<9) - 1) >> 9;
554                 ci->i_reported_size = size;
555                 if (truncate_seq != ci->i_truncate_seq) {
556                         dout("truncate_seq %u -> %u\n",
557                              ci->i_truncate_seq, truncate_seq);
558                         ci->i_truncate_seq = truncate_seq;
559
560                         /* the MDS should have revoked these caps */
561                         WARN_ON_ONCE(issued & (CEPH_CAP_FILE_EXCL |
562                                                CEPH_CAP_FILE_RD |
563                                                CEPH_CAP_FILE_WR |
564                                                CEPH_CAP_FILE_LAZYIO));
565                         /*
566                          * If we hold relevant caps, or in the case where we're
567                          * not the only client referencing this file and we
568                          * don't hold those caps, then we need to check whether
569                          * the file is either opened or mmaped
570                          */
571                         if ((issued & (CEPH_CAP_FILE_CACHE|
572                                        CEPH_CAP_FILE_BUFFER)) ||
573                             mapping_mapped(inode->i_mapping) ||
574                             __ceph_caps_file_wanted(ci)) {
575                                 ci->i_truncate_pending++;
576                                 queue_trunc = 1;
577                         }
578                 }
579         }
580         if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 &&
581             ci->i_truncate_size != truncate_size) {
582                 dout("truncate_size %lld -> %llu\n", ci->i_truncate_size,
583                      truncate_size);
584                 ci->i_truncate_size = truncate_size;
585         }
586
587         if (queue_trunc)
588                 ceph_fscache_invalidate(inode);
589
590         return queue_trunc;
591 }
592
593 void ceph_fill_file_time(struct inode *inode, int issued,
594                          u64 time_warp_seq, struct timespec *ctime,
595                          struct timespec *mtime, struct timespec *atime)
596 {
597         struct ceph_inode_info *ci = ceph_inode(inode);
598         int warn = 0;
599
600         if (issued & (CEPH_CAP_FILE_EXCL|
601                       CEPH_CAP_FILE_WR|
602                       CEPH_CAP_FILE_BUFFER|
603                       CEPH_CAP_AUTH_EXCL|
604                       CEPH_CAP_XATTR_EXCL)) {
605                 if (timespec_compare(ctime, &inode->i_ctime) > 0) {
606                         dout("ctime %ld.%09ld -> %ld.%09ld inc w/ cap\n",
607                              inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
608                              ctime->tv_sec, ctime->tv_nsec);
609                         inode->i_ctime = *ctime;
610                 }
611                 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) {
612                         /* the MDS did a utimes() */
613                         dout("mtime %ld.%09ld -> %ld.%09ld "
614                              "tw %d -> %d\n",
615                              inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
616                              mtime->tv_sec, mtime->tv_nsec,
617                              ci->i_time_warp_seq, (int)time_warp_seq);
618
619                         inode->i_mtime = *mtime;
620                         inode->i_atime = *atime;
621                         ci->i_time_warp_seq = time_warp_seq;
622                 } else if (time_warp_seq == ci->i_time_warp_seq) {
623                         /* nobody did utimes(); take the max */
624                         if (timespec_compare(mtime, &inode->i_mtime) > 0) {
625                                 dout("mtime %ld.%09ld -> %ld.%09ld inc\n",
626                                      inode->i_mtime.tv_sec,
627                                      inode->i_mtime.tv_nsec,
628                                      mtime->tv_sec, mtime->tv_nsec);
629                                 inode->i_mtime = *mtime;
630                         }
631                         if (timespec_compare(atime, &inode->i_atime) > 0) {
632                                 dout("atime %ld.%09ld -> %ld.%09ld inc\n",
633                                      inode->i_atime.tv_sec,
634                                      inode->i_atime.tv_nsec,
635                                      atime->tv_sec, atime->tv_nsec);
636                                 inode->i_atime = *atime;
637                         }
638                 } else if (issued & CEPH_CAP_FILE_EXCL) {
639                         /* we did a utimes(); ignore mds values */
640                 } else {
641                         warn = 1;
642                 }
643         } else {
644                 /* we have no write|excl caps; whatever the MDS says is true */
645                 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) {
646                         inode->i_ctime = *ctime;
647                         inode->i_mtime = *mtime;
648                         inode->i_atime = *atime;
649                         ci->i_time_warp_seq = time_warp_seq;
650                 } else {
651                         warn = 1;
652                 }
653         }
654         if (warn) /* time_warp_seq shouldn't go backwards */
655                 dout("%p mds time_warp_seq %llu < %u\n",
656                      inode, time_warp_seq, ci->i_time_warp_seq);
657 }
658
659 /*
660  * Populate an inode based on info from mds.  May be called on new or
661  * existing inodes.
662  */
663 static int fill_inode(struct inode *inode, struct page *locked_page,
664                       struct ceph_mds_reply_info_in *iinfo,
665                       struct ceph_mds_reply_dirfrag *dirinfo,
666                       struct ceph_mds_session *session,
667                       unsigned long ttl_from, int cap_fmode,
668                       struct ceph_cap_reservation *caps_reservation)
669 {
670         struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
671         struct ceph_mds_reply_inode *info = iinfo->in;
672         struct ceph_inode_info *ci = ceph_inode(inode);
673         int issued = 0, implemented, new_issued;
674         struct timespec mtime, atime, ctime;
675         struct ceph_buffer *xattr_blob = NULL;
676         struct ceph_cap *new_cap = NULL;
677         int err = 0;
678         bool wake = false;
679         bool queue_trunc = false;
680         bool new_version = false;
681         bool fill_inline = false;
682
683         dout("fill_inode %p ino %llx.%llx v %llu had %llu\n",
684              inode, ceph_vinop(inode), le64_to_cpu(info->version),
685              ci->i_version);
686
687         /* prealloc new cap struct */
688         if (info->cap.caps && ceph_snap(inode) == CEPH_NOSNAP)
689                 new_cap = ceph_get_cap(mdsc, caps_reservation);
690
691         /*
692          * prealloc xattr data, if it looks like we'll need it.  only
693          * if len > 4 (meaning there are actually xattrs; the first 4
694          * bytes are the xattr count).
695          */
696         if (iinfo->xattr_len > 4) {
697                 xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS);
698                 if (!xattr_blob)
699                         pr_err("fill_inode ENOMEM xattr blob %d bytes\n",
700                                iinfo->xattr_len);
701         }
702
703         spin_lock(&ci->i_ceph_lock);
704
705         /*
706          * provided version will be odd if inode value is projected,
707          * even if stable.  skip the update if we have newer stable
708          * info (ours>=theirs, e.g. due to racing mds replies), unless
709          * we are getting projected (unstable) info (in which case the
710          * version is odd, and we want ours>theirs).
711          *   us   them
712          *   2    2     skip
713          *   3    2     skip
714          *   3    3     update
715          */
716         if (ci->i_version == 0 ||
717             ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
718              le64_to_cpu(info->version) > (ci->i_version & ~1)))
719                 new_version = true;
720
721         issued = __ceph_caps_issued(ci, &implemented);
722         issued |= implemented | __ceph_caps_dirty(ci);
723         new_issued = ~issued & le32_to_cpu(info->cap.caps);
724
725         /* update inode */
726         ci->i_version = le64_to_cpu(info->version);
727         inode->i_version++;
728         inode->i_rdev = le32_to_cpu(info->rdev);
729         inode->i_blkbits = fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1;
730
731         if ((new_version || (new_issued & CEPH_CAP_AUTH_SHARED)) &&
732             (issued & CEPH_CAP_AUTH_EXCL) == 0) {
733                 inode->i_mode = le32_to_cpu(info->mode);
734                 inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(info->uid));
735                 inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(info->gid));
736                 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
737                      from_kuid(&init_user_ns, inode->i_uid),
738                      from_kgid(&init_user_ns, inode->i_gid));
739         }
740
741         if ((new_version || (new_issued & CEPH_CAP_LINK_SHARED)) &&
742             (issued & CEPH_CAP_LINK_EXCL) == 0)
743                 set_nlink(inode, le32_to_cpu(info->nlink));
744
745         if (new_version || (new_issued & CEPH_CAP_ANY_RD)) {
746                 /* be careful with mtime, atime, size */
747                 ceph_decode_timespec(&atime, &info->atime);
748                 ceph_decode_timespec(&mtime, &info->mtime);
749                 ceph_decode_timespec(&ctime, &info->ctime);
750                 ceph_fill_file_time(inode, issued,
751                                 le32_to_cpu(info->time_warp_seq),
752                                 &ctime, &mtime, &atime);
753         }
754
755         if (new_version ||
756             (new_issued & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR))) {
757                 if (ci->i_layout.fl_pg_pool != info->layout.fl_pg_pool)
758                         ci->i_ceph_flags &= ~CEPH_I_POOL_PERM;
759                 ci->i_layout = info->layout;
760                 ci->i_pool_ns_len = iinfo->pool_ns_len;
761
762                 queue_trunc = ceph_fill_file_size(inode, issued,
763                                         le32_to_cpu(info->truncate_seq),
764                                         le64_to_cpu(info->truncate_size),
765                                         le64_to_cpu(info->size));
766                 /* only update max_size on auth cap */
767                 if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
768                     ci->i_max_size != le64_to_cpu(info->max_size)) {
769                         dout("max_size %lld -> %llu\n", ci->i_max_size,
770                                         le64_to_cpu(info->max_size));
771                         ci->i_max_size = le64_to_cpu(info->max_size);
772                 }
773         }
774
775         /* xattrs */
776         /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
777         if ((ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL))  &&
778             le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) {
779                 if (ci->i_xattrs.blob)
780                         ceph_buffer_put(ci->i_xattrs.blob);
781                 ci->i_xattrs.blob = xattr_blob;
782                 if (xattr_blob)
783                         memcpy(ci->i_xattrs.blob->vec.iov_base,
784                                iinfo->xattr_data, iinfo->xattr_len);
785                 ci->i_xattrs.version = le64_to_cpu(info->xattr_version);
786                 ceph_forget_all_cached_acls(inode);
787                 xattr_blob = NULL;
788         }
789
790         inode->i_mapping->a_ops = &ceph_aops;
791
792         switch (inode->i_mode & S_IFMT) {
793         case S_IFIFO:
794         case S_IFBLK:
795         case S_IFCHR:
796         case S_IFSOCK:
797                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
798                 inode->i_op = &ceph_file_iops;
799                 break;
800         case S_IFREG:
801                 inode->i_op = &ceph_file_iops;
802                 inode->i_fop = &ceph_file_fops;
803                 break;
804         case S_IFLNK:
805                 inode->i_op = &ceph_symlink_iops;
806                 if (!ci->i_symlink) {
807                         u32 symlen = iinfo->symlink_len;
808                         char *sym;
809
810                         spin_unlock(&ci->i_ceph_lock);
811
812                         err = -EINVAL;
813                         if (WARN_ON(symlen != i_size_read(inode)))
814                                 goto out;
815
816                         err = -ENOMEM;
817                         sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS);
818                         if (!sym)
819                                 goto out;
820
821                         spin_lock(&ci->i_ceph_lock);
822                         if (!ci->i_symlink)
823                                 ci->i_symlink = sym;
824                         else
825                                 kfree(sym); /* lost a race */
826                 }
827                 inode->i_link = ci->i_symlink;
828                 break;
829         case S_IFDIR:
830                 inode->i_op = &ceph_dir_iops;
831                 inode->i_fop = &ceph_dir_fops;
832
833                 ci->i_dir_layout = iinfo->dir_layout;
834
835                 ci->i_files = le64_to_cpu(info->files);
836                 ci->i_subdirs = le64_to_cpu(info->subdirs);
837                 ci->i_rbytes = le64_to_cpu(info->rbytes);
838                 ci->i_rfiles = le64_to_cpu(info->rfiles);
839                 ci->i_rsubdirs = le64_to_cpu(info->rsubdirs);
840                 ceph_decode_timespec(&ci->i_rctime, &info->rctime);
841                 break;
842         default:
843                 pr_err("fill_inode %llx.%llx BAD mode 0%o\n",
844                        ceph_vinop(inode), inode->i_mode);
845         }
846
847         /* were we issued a capability? */
848         if (info->cap.caps) {
849                 if (ceph_snap(inode) == CEPH_NOSNAP) {
850                         unsigned caps = le32_to_cpu(info->cap.caps);
851                         ceph_add_cap(inode, session,
852                                      le64_to_cpu(info->cap.cap_id),
853                                      cap_fmode, caps,
854                                      le32_to_cpu(info->cap.wanted),
855                                      le32_to_cpu(info->cap.seq),
856                                      le32_to_cpu(info->cap.mseq),
857                                      le64_to_cpu(info->cap.realm),
858                                      info->cap.flags, &new_cap);
859
860                         /* set dir completion flag? */
861                         if (S_ISDIR(inode->i_mode) &&
862                             ci->i_files == 0 && ci->i_subdirs == 0 &&
863                             (caps & CEPH_CAP_FILE_SHARED) &&
864                             (issued & CEPH_CAP_FILE_EXCL) == 0 &&
865                             !__ceph_dir_is_complete(ci)) {
866                                 dout(" marking %p complete (empty)\n", inode);
867                                 i_size_write(inode, 0);
868                                 __ceph_dir_set_complete(ci,
869                                         atomic64_read(&ci->i_release_count),
870                                         atomic64_read(&ci->i_ordered_count));
871                         }
872
873                         wake = true;
874                 } else {
875                         dout(" %p got snap_caps %s\n", inode,
876                              ceph_cap_string(le32_to_cpu(info->cap.caps)));
877                         ci->i_snap_caps |= le32_to_cpu(info->cap.caps);
878                         if (cap_fmode >= 0)
879                                 __ceph_get_fmode(ci, cap_fmode);
880                 }
881         } else if (cap_fmode >= 0) {
882                 pr_warn("mds issued no caps on %llx.%llx\n",
883                            ceph_vinop(inode));
884                 __ceph_get_fmode(ci, cap_fmode);
885         }
886
887         if (iinfo->inline_version > 0 &&
888             iinfo->inline_version >= ci->i_inline_version) {
889                 int cache_caps = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
890                 ci->i_inline_version = iinfo->inline_version;
891                 if (ci->i_inline_version != CEPH_INLINE_NONE &&
892                     (locked_page ||
893                      (le32_to_cpu(info->cap.caps) & cache_caps)))
894                         fill_inline = true;
895         }
896
897         spin_unlock(&ci->i_ceph_lock);
898
899         if (fill_inline)
900                 ceph_fill_inline_data(inode, locked_page,
901                                       iinfo->inline_data, iinfo->inline_len);
902
903         if (wake)
904                 wake_up_all(&ci->i_cap_wq);
905
906         /* queue truncate if we saw i_size decrease */
907         if (queue_trunc)
908                 ceph_queue_vmtruncate(inode);
909
910         /* populate frag tree */
911         if (S_ISDIR(inode->i_mode))
912                 ceph_fill_fragtree(inode, &info->fragtree, dirinfo);
913
914         /* update delegation info? */
915         if (dirinfo)
916                 ceph_fill_dirfrag(inode, dirinfo);
917
918         err = 0;
919 out:
920         if (new_cap)
921                 ceph_put_cap(mdsc, new_cap);
922         if (xattr_blob)
923                 ceph_buffer_put(xattr_blob);
924         return err;
925 }
926
927 /*
928  * caller should hold session s_mutex.
929  */
930 static void update_dentry_lease(struct dentry *dentry,
931                                 struct ceph_mds_reply_lease *lease,
932                                 struct ceph_mds_session *session,
933                                 unsigned long from_time)
934 {
935         struct ceph_dentry_info *di = ceph_dentry(dentry);
936         long unsigned duration = le32_to_cpu(lease->duration_ms);
937         long unsigned ttl = from_time + (duration * HZ) / 1000;
938         long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000;
939         struct inode *dir;
940
941         /* only track leases on regular dentries */
942         if (dentry->d_op != &ceph_dentry_ops)
943                 return;
944
945         spin_lock(&dentry->d_lock);
946         dout("update_dentry_lease %p duration %lu ms ttl %lu\n",
947              dentry, duration, ttl);
948
949         /* make lease_rdcache_gen match directory */
950         dir = d_inode(dentry->d_parent);
951         di->lease_shared_gen = ceph_inode(dir)->i_shared_gen;
952
953         if (duration == 0)
954                 goto out_unlock;
955
956         if (di->lease_gen == session->s_cap_gen &&
957             time_before(ttl, dentry->d_time))
958                 goto out_unlock;  /* we already have a newer lease. */
959
960         if (di->lease_session && di->lease_session != session)
961                 goto out_unlock;
962
963         ceph_dentry_lru_touch(dentry);
964
965         if (!di->lease_session)
966                 di->lease_session = ceph_get_mds_session(session);
967         di->lease_gen = session->s_cap_gen;
968         di->lease_seq = le32_to_cpu(lease->seq);
969         di->lease_renew_after = half_ttl;
970         di->lease_renew_from = 0;
971         dentry->d_time = ttl;
972 out_unlock:
973         spin_unlock(&dentry->d_lock);
974         return;
975 }
976
977 /*
978  * splice a dentry to an inode.
979  * caller must hold directory i_mutex for this to be safe.
980  */
981 static struct dentry *splice_dentry(struct dentry *dn, struct inode *in)
982 {
983         struct dentry *realdn;
984
985         BUG_ON(d_inode(dn));
986
987         /* dn must be unhashed */
988         if (!d_unhashed(dn))
989                 d_drop(dn);
990         realdn = d_splice_alias(in, dn);
991         if (IS_ERR(realdn)) {
992                 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
993                        PTR_ERR(realdn), dn, in, ceph_vinop(in));
994                 dn = realdn; /* note realdn contains the error */
995                 goto out;
996         } else if (realdn) {
997                 dout("dn %p (%d) spliced with %p (%d) "
998                      "inode %p ino %llx.%llx\n",
999                      dn, d_count(dn),
1000                      realdn, d_count(realdn),
1001                      d_inode(realdn), ceph_vinop(d_inode(realdn)));
1002                 dput(dn);
1003                 dn = realdn;
1004         } else {
1005                 BUG_ON(!ceph_dentry(dn));
1006                 dout("dn %p attached to %p ino %llx.%llx\n",
1007                      dn, d_inode(dn), ceph_vinop(d_inode(dn)));
1008         }
1009 out:
1010         return dn;
1011 }
1012
1013 /*
1014  * Incorporate results into the local cache.  This is either just
1015  * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
1016  * after a lookup).
1017  *
1018  * A reply may contain
1019  *         a directory inode along with a dentry.
1020  *  and/or a target inode
1021  *
1022  * Called with snap_rwsem (read).
1023  */
1024 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req,
1025                     struct ceph_mds_session *session)
1026 {
1027         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1028         struct inode *in = NULL;
1029         struct ceph_vino vino;
1030         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
1031         int err = 0;
1032
1033         dout("fill_trace %p is_dentry %d is_target %d\n", req,
1034              rinfo->head->is_dentry, rinfo->head->is_target);
1035
1036 #if 0
1037         /*
1038          * Debugging hook:
1039          *
1040          * If we resend completed ops to a recovering mds, we get no
1041          * trace.  Since that is very rare, pretend this is the case
1042          * to ensure the 'no trace' handlers in the callers behave.
1043          *
1044          * Fill in inodes unconditionally to avoid breaking cap
1045          * invariants.
1046          */
1047         if (rinfo->head->op & CEPH_MDS_OP_WRITE) {
1048                 pr_info("fill_trace faking empty trace on %lld %s\n",
1049                         req->r_tid, ceph_mds_op_name(rinfo->head->op));
1050                 if (rinfo->head->is_dentry) {
1051                         rinfo->head->is_dentry = 0;
1052                         err = fill_inode(req->r_locked_dir,
1053                                          &rinfo->diri, rinfo->dirfrag,
1054                                          session, req->r_request_started, -1);
1055                 }
1056                 if (rinfo->head->is_target) {
1057                         rinfo->head->is_target = 0;
1058                         ininfo = rinfo->targeti.in;
1059                         vino.ino = le64_to_cpu(ininfo->ino);
1060                         vino.snap = le64_to_cpu(ininfo->snapid);
1061                         in = ceph_get_inode(sb, vino);
1062                         err = fill_inode(in, &rinfo->targeti, NULL,
1063                                          session, req->r_request_started,
1064                                          req->r_fmode);
1065                         iput(in);
1066                 }
1067         }
1068 #endif
1069
1070         if (!rinfo->head->is_target && !rinfo->head->is_dentry) {
1071                 dout("fill_trace reply is empty!\n");
1072                 if (rinfo->head->result == 0 && req->r_locked_dir)
1073                         ceph_invalidate_dir_request(req);
1074                 return 0;
1075         }
1076
1077         if (rinfo->head->is_dentry) {
1078                 struct inode *dir = req->r_locked_dir;
1079
1080                 if (dir) {
1081                         err = fill_inode(dir, NULL,
1082                                          &rinfo->diri, rinfo->dirfrag,
1083                                          session, req->r_request_started, -1,
1084                                          &req->r_caps_reservation);
1085                         if (err < 0)
1086                                 goto done;
1087                 } else {
1088                         WARN_ON_ONCE(1);
1089                 }
1090
1091                 if (dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME) {
1092                         struct qstr dname;
1093                         struct dentry *dn, *parent;
1094
1095                         BUG_ON(!rinfo->head->is_target);
1096                         BUG_ON(req->r_dentry);
1097
1098                         parent = d_find_any_alias(dir);
1099                         BUG_ON(!parent);
1100
1101                         dname.name = rinfo->dname;
1102                         dname.len = rinfo->dname_len;
1103                         dname.hash = full_name_hash(dname.name, dname.len);
1104                         vino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1105                         vino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1106 retry_lookup:
1107                         dn = d_lookup(parent, &dname);
1108                         dout("d_lookup on parent=%p name=%.*s got %p\n",
1109                              parent, dname.len, dname.name, dn);
1110
1111                         if (!dn) {
1112                                 dn = d_alloc(parent, &dname);
1113                                 dout("d_alloc %p '%.*s' = %p\n", parent,
1114                                      dname.len, dname.name, dn);
1115                                 if (dn == NULL) {
1116                                         dput(parent);
1117                                         err = -ENOMEM;
1118                                         goto done;
1119                                 }
1120                                 err = ceph_init_dentry(dn);
1121                                 if (err < 0) {
1122                                         dput(dn);
1123                                         dput(parent);
1124                                         goto done;
1125                                 }
1126                         } else if (d_really_is_positive(dn) &&
1127                                    (ceph_ino(d_inode(dn)) != vino.ino ||
1128                                     ceph_snap(d_inode(dn)) != vino.snap)) {
1129                                 dout(" dn %p points to wrong inode %p\n",
1130                                      dn, d_inode(dn));
1131                                 d_delete(dn);
1132                                 dput(dn);
1133                                 goto retry_lookup;
1134                         }
1135
1136                         req->r_dentry = dn;
1137                         dput(parent);
1138                 }
1139         }
1140
1141         if (rinfo->head->is_target) {
1142                 vino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1143                 vino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1144
1145                 in = ceph_get_inode(sb, vino);
1146                 if (IS_ERR(in)) {
1147                         err = PTR_ERR(in);
1148                         goto done;
1149                 }
1150                 req->r_target_inode = in;
1151
1152                 err = fill_inode(in, req->r_locked_page, &rinfo->targeti, NULL,
1153                                 session, req->r_request_started,
1154                                 (!req->r_aborted && rinfo->head->result == 0) ?
1155                                 req->r_fmode : -1,
1156                                 &req->r_caps_reservation);
1157                 if (err < 0) {
1158                         pr_err("fill_inode badness %p %llx.%llx\n",
1159                                 in, ceph_vinop(in));
1160                         goto done;
1161                 }
1162         }
1163
1164         /*
1165          * ignore null lease/binding on snapdir ENOENT, or else we
1166          * will have trouble splicing in the virtual snapdir later
1167          */
1168         if (rinfo->head->is_dentry && !req->r_aborted &&
1169             req->r_locked_dir &&
1170             (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name,
1171                                                fsc->mount_options->snapdir_name,
1172                                                req->r_dentry->d_name.len))) {
1173                 /*
1174                  * lookup link rename   : null -> possibly existing inode
1175                  * mknod symlink mkdir  : null -> new inode
1176                  * unlink               : linked -> null
1177                  */
1178                 struct inode *dir = req->r_locked_dir;
1179                 struct dentry *dn = req->r_dentry;
1180                 bool have_dir_cap, have_lease;
1181
1182                 BUG_ON(!dn);
1183                 BUG_ON(!dir);
1184                 BUG_ON(d_inode(dn->d_parent) != dir);
1185                 BUG_ON(ceph_ino(dir) !=
1186                        le64_to_cpu(rinfo->diri.in->ino));
1187                 BUG_ON(ceph_snap(dir) !=
1188                        le64_to_cpu(rinfo->diri.in->snapid));
1189
1190                 /* do we have a lease on the whole dir? */
1191                 have_dir_cap =
1192                         (le32_to_cpu(rinfo->diri.in->cap.caps) &
1193                          CEPH_CAP_FILE_SHARED);
1194
1195                 /* do we have a dn lease? */
1196                 have_lease = have_dir_cap ||
1197                         le32_to_cpu(rinfo->dlease->duration_ms);
1198                 if (!have_lease)
1199                         dout("fill_trace  no dentry lease or dir cap\n");
1200
1201                 /* rename? */
1202                 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) {
1203                         struct inode *olddir = req->r_old_dentry_dir;
1204                         BUG_ON(!olddir);
1205
1206                         dout(" src %p '%pd' dst %p '%pd'\n",
1207                              req->r_old_dentry,
1208                              req->r_old_dentry,
1209                              dn, dn);
1210                         dout("fill_trace doing d_move %p -> %p\n",
1211                              req->r_old_dentry, dn);
1212
1213                         /* d_move screws up sibling dentries' offsets */
1214                         ceph_dir_clear_ordered(dir);
1215                         ceph_dir_clear_ordered(olddir);
1216
1217                         d_move(req->r_old_dentry, dn);
1218                         dout(" src %p '%pd' dst %p '%pd'\n",
1219                              req->r_old_dentry,
1220                              req->r_old_dentry,
1221                              dn, dn);
1222
1223                         /* ensure target dentry is invalidated, despite
1224                            rehashing bug in vfs_rename_dir */
1225                         ceph_invalidate_dentry_lease(dn);
1226
1227                         dout("dn %p gets new offset %lld\n", req->r_old_dentry,
1228                              ceph_dentry(req->r_old_dentry)->offset);
1229
1230                         dn = req->r_old_dentry;  /* use old_dentry */
1231                 }
1232
1233                 /* null dentry? */
1234                 if (!rinfo->head->is_target) {
1235                         dout("fill_trace null dentry\n");
1236                         if (d_really_is_positive(dn)) {
1237                                 ceph_dir_clear_ordered(dir);
1238                                 dout("d_delete %p\n", dn);
1239                                 d_delete(dn);
1240                         } else {
1241                                 if (have_lease && d_unhashed(dn))
1242                                         d_add(dn, NULL);
1243                                 update_dentry_lease(dn, rinfo->dlease,
1244                                                     session,
1245                                                     req->r_request_started);
1246                         }
1247                         goto done;
1248                 }
1249
1250                 /* attach proper inode */
1251                 if (d_really_is_negative(dn)) {
1252                         ceph_dir_clear_ordered(dir);
1253                         ihold(in);
1254                         dn = splice_dentry(dn, in);
1255                         if (IS_ERR(dn)) {
1256                                 err = PTR_ERR(dn);
1257                                 goto done;
1258                         }
1259                         req->r_dentry = dn;  /* may have spliced */
1260                 } else if (d_really_is_positive(dn) && d_inode(dn) != in) {
1261                         dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1262                              dn, d_inode(dn), ceph_vinop(d_inode(dn)),
1263                              ceph_vinop(in));
1264                         have_lease = false;
1265                 }
1266
1267                 if (have_lease)
1268                         update_dentry_lease(dn, rinfo->dlease, session,
1269                                             req->r_request_started);
1270                 dout(" final dn %p\n", dn);
1271         } else if (!req->r_aborted &&
1272                    (req->r_op == CEPH_MDS_OP_LOOKUPSNAP ||
1273                     req->r_op == CEPH_MDS_OP_MKSNAP)) {
1274                 struct dentry *dn = req->r_dentry;
1275                 struct inode *dir = req->r_locked_dir;
1276
1277                 /* fill out a snapdir LOOKUPSNAP dentry */
1278                 BUG_ON(!dn);
1279                 BUG_ON(!dir);
1280                 BUG_ON(ceph_snap(dir) != CEPH_SNAPDIR);
1281                 dout(" linking snapped dir %p to dn %p\n", in, dn);
1282                 ceph_dir_clear_ordered(dir);
1283                 ihold(in);
1284                 dn = splice_dentry(dn, in);
1285                 if (IS_ERR(dn)) {
1286                         err = PTR_ERR(dn);
1287                         goto done;
1288                 }
1289                 req->r_dentry = dn;  /* may have spliced */
1290         }
1291 done:
1292         dout("fill_trace done err=%d\n", err);
1293         return err;
1294 }
1295
1296 /*
1297  * Prepopulate our cache with readdir results, leases, etc.
1298  */
1299 static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req,
1300                                            struct ceph_mds_session *session)
1301 {
1302         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1303         int i, err = 0;
1304
1305         for (i = 0; i < rinfo->dir_nr; i++) {
1306                 struct ceph_vino vino;
1307                 struct inode *in;
1308                 int rc;
1309
1310                 vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino);
1311                 vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid);
1312
1313                 in = ceph_get_inode(req->r_dentry->d_sb, vino);
1314                 if (IS_ERR(in)) {
1315                         err = PTR_ERR(in);
1316                         dout("new_inode badness got %d\n", err);
1317                         continue;
1318                 }
1319                 rc = fill_inode(in, NULL, &rinfo->dir_in[i], NULL, session,
1320                                 req->r_request_started, -1,
1321                                 &req->r_caps_reservation);
1322                 if (rc < 0) {
1323                         pr_err("fill_inode badness on %p got %d\n", in, rc);
1324                         err = rc;
1325                         continue;
1326                 }
1327         }
1328
1329         return err;
1330 }
1331
1332 void ceph_readdir_cache_release(struct ceph_readdir_cache_control *ctl)
1333 {
1334         if (ctl->page) {
1335                 kunmap(ctl->page);
1336                 page_cache_release(ctl->page);
1337                 ctl->page = NULL;
1338         }
1339 }
1340
1341 static int fill_readdir_cache(struct inode *dir, struct dentry *dn,
1342                               struct ceph_readdir_cache_control *ctl,
1343                               struct ceph_mds_request *req)
1344 {
1345         struct ceph_inode_info *ci = ceph_inode(dir);
1346         unsigned nsize = PAGE_CACHE_SIZE / sizeof(struct dentry*);
1347         unsigned idx = ctl->index % nsize;
1348         pgoff_t pgoff = ctl->index / nsize;
1349
1350         if (!ctl->page || pgoff != page_index(ctl->page)) {
1351                 ceph_readdir_cache_release(ctl);
1352                 if (idx == 0)
1353                         ctl->page = grab_cache_page(&dir->i_data, pgoff);
1354                 else
1355                         ctl->page = find_lock_page(&dir->i_data, pgoff);
1356                 if (!ctl->page) {
1357                         ctl->index = -1;
1358                         return idx == 0 ? -ENOMEM : 0;
1359                 }
1360                 /* reading/filling the cache are serialized by
1361                  * i_mutex, no need to use page lock */
1362                 unlock_page(ctl->page);
1363                 ctl->dentries = kmap(ctl->page);
1364                 if (idx == 0)
1365                         memset(ctl->dentries, 0, PAGE_CACHE_SIZE);
1366         }
1367
1368         if (req->r_dir_release_cnt == atomic64_read(&ci->i_release_count) &&
1369             req->r_dir_ordered_cnt == atomic64_read(&ci->i_ordered_count)) {
1370                 dout("readdir cache dn %p idx %d\n", dn, ctl->index);
1371                 ctl->dentries[idx] = dn;
1372                 ctl->index++;
1373         } else {
1374                 dout("disable readdir cache\n");
1375                 ctl->index = -1;
1376         }
1377         return 0;
1378 }
1379
1380 int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1381                              struct ceph_mds_session *session)
1382 {
1383         struct dentry *parent = req->r_dentry;
1384         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1385         struct qstr dname;
1386         struct dentry *dn;
1387         struct inode *in;
1388         int err = 0, ret, i;
1389         struct inode *snapdir = NULL;
1390         struct ceph_mds_request_head *rhead = req->r_request->front.iov_base;
1391         struct ceph_dentry_info *di;
1392         u32 frag = le32_to_cpu(rhead->args.readdir.frag);
1393         struct ceph_readdir_cache_control cache_ctl = {};
1394
1395         if (req->r_aborted)
1396                 return readdir_prepopulate_inodes_only(req, session);
1397
1398         if (rinfo->dir_dir &&
1399             le32_to_cpu(rinfo->dir_dir->frag) != frag) {
1400                 dout("readdir_prepopulate got new frag %x -> %x\n",
1401                      frag, le32_to_cpu(rinfo->dir_dir->frag));
1402                 frag = le32_to_cpu(rinfo->dir_dir->frag);
1403                 if (ceph_frag_is_leftmost(frag))
1404                         req->r_readdir_offset = 2;
1405                 else
1406                         req->r_readdir_offset = 0;
1407         }
1408
1409         if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
1410                 snapdir = ceph_get_snapdir(d_inode(parent));
1411                 parent = d_find_alias(snapdir);
1412                 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1413                      rinfo->dir_nr, parent);
1414         } else {
1415                 dout("readdir_prepopulate %d items under dn %p\n",
1416                      rinfo->dir_nr, parent);
1417                 if (rinfo->dir_dir)
1418                         ceph_fill_dirfrag(d_inode(parent), rinfo->dir_dir);
1419         }
1420
1421         if (ceph_frag_is_leftmost(frag) && req->r_readdir_offset == 2) {
1422                 /* note dir version at start of readdir so we can tell
1423                  * if any dentries get dropped */
1424                 struct ceph_inode_info *ci = ceph_inode(d_inode(parent));
1425                 req->r_dir_release_cnt = atomic64_read(&ci->i_release_count);
1426                 req->r_dir_ordered_cnt = atomic64_read(&ci->i_ordered_count);
1427                 req->r_readdir_cache_idx = 0;
1428         }
1429
1430         cache_ctl.index = req->r_readdir_cache_idx;
1431
1432         /* FIXME: release caps/leases if error occurs */
1433         for (i = 0; i < rinfo->dir_nr; i++) {
1434                 struct ceph_vino vino;
1435
1436                 dname.name = rinfo->dir_dname[i];
1437                 dname.len = rinfo->dir_dname_len[i];
1438                 dname.hash = full_name_hash(dname.name, dname.len);
1439
1440                 vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino);
1441                 vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid);
1442
1443 retry_lookup:
1444                 dn = d_lookup(parent, &dname);
1445                 dout("d_lookup on parent=%p name=%.*s got %p\n",
1446                      parent, dname.len, dname.name, dn);
1447
1448                 if (!dn) {
1449                         dn = d_alloc(parent, &dname);
1450                         dout("d_alloc %p '%.*s' = %p\n", parent,
1451                              dname.len, dname.name, dn);
1452                         if (dn == NULL) {
1453                                 dout("d_alloc badness\n");
1454                                 err = -ENOMEM;
1455                                 goto out;
1456                         }
1457                         ret = ceph_init_dentry(dn);
1458                         if (ret < 0) {
1459                                 dput(dn);
1460                                 err = ret;
1461                                 goto out;
1462                         }
1463                 } else if (d_really_is_positive(dn) &&
1464                            (ceph_ino(d_inode(dn)) != vino.ino ||
1465                             ceph_snap(d_inode(dn)) != vino.snap)) {
1466                         dout(" dn %p points to wrong inode %p\n",
1467                              dn, d_inode(dn));
1468                         d_delete(dn);
1469                         dput(dn);
1470                         goto retry_lookup;
1471                 }
1472
1473                 /* inode */
1474                 if (d_really_is_positive(dn)) {
1475                         in = d_inode(dn);
1476                 } else {
1477                         in = ceph_get_inode(parent->d_sb, vino);
1478                         if (IS_ERR(in)) {
1479                                 dout("new_inode badness\n");
1480                                 d_drop(dn);
1481                                 dput(dn);
1482                                 err = PTR_ERR(in);
1483                                 goto out;
1484                         }
1485                 }
1486
1487                 ret = fill_inode(in, NULL, &rinfo->dir_in[i], NULL, session,
1488                                  req->r_request_started, -1,
1489                                  &req->r_caps_reservation);
1490                 if (ret < 0) {
1491                         pr_err("fill_inode badness on %p\n", in);
1492                         if (d_really_is_negative(dn))
1493                                 iput(in);
1494                         d_drop(dn);
1495                         err = ret;
1496                         goto next_item;
1497                 }
1498
1499                 if (d_really_is_negative(dn)) {
1500                         struct dentry *realdn = splice_dentry(dn, in);
1501                         if (IS_ERR(realdn)) {
1502                                 err = PTR_ERR(realdn);
1503                                 d_drop(dn);
1504                                 dn = NULL;
1505                                 goto next_item;
1506                         }
1507                         dn = realdn;
1508                 }
1509
1510                 di = dn->d_fsdata;
1511                 di->offset = ceph_make_fpos(frag, i + req->r_readdir_offset);
1512
1513                 update_dentry_lease(dn, rinfo->dir_dlease[i],
1514                                     req->r_session,
1515                                     req->r_request_started);
1516
1517                 if (err == 0 && cache_ctl.index >= 0) {
1518                         ret = fill_readdir_cache(d_inode(parent), dn,
1519                                                  &cache_ctl, req);
1520                         if (ret < 0)
1521                                 err = ret;
1522                 }
1523 next_item:
1524                 if (dn)
1525                         dput(dn);
1526         }
1527 out:
1528         if (err == 0) {
1529                 req->r_did_prepopulate = true;
1530                 req->r_readdir_cache_idx = cache_ctl.index;
1531         }
1532         ceph_readdir_cache_release(&cache_ctl);
1533         if (snapdir) {
1534                 iput(snapdir);
1535                 dput(parent);
1536         }
1537         dout("readdir_prepopulate done\n");
1538         return err;
1539 }
1540
1541 int ceph_inode_set_size(struct inode *inode, loff_t size)
1542 {
1543         struct ceph_inode_info *ci = ceph_inode(inode);
1544         int ret = 0;
1545
1546         spin_lock(&ci->i_ceph_lock);
1547         dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size);
1548         i_size_write(inode, size);
1549         inode->i_blocks = (size + (1 << 9) - 1) >> 9;
1550
1551         /* tell the MDS if we are approaching max_size */
1552         if ((size << 1) >= ci->i_max_size &&
1553             (ci->i_reported_size << 1) < ci->i_max_size)
1554                 ret = 1;
1555
1556         spin_unlock(&ci->i_ceph_lock);
1557         return ret;
1558 }
1559
1560 /*
1561  * Write back inode data in a worker thread.  (This can't be done
1562  * in the message handler context.)
1563  */
1564 void ceph_queue_writeback(struct inode *inode)
1565 {
1566         ihold(inode);
1567         if (queue_work(ceph_inode_to_client(inode)->wb_wq,
1568                        &ceph_inode(inode)->i_wb_work)) {
1569                 dout("ceph_queue_writeback %p\n", inode);
1570         } else {
1571                 dout("ceph_queue_writeback %p failed\n", inode);
1572                 iput(inode);
1573         }
1574 }
1575
1576 static void ceph_writeback_work(struct work_struct *work)
1577 {
1578         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1579                                                   i_wb_work);
1580         struct inode *inode = &ci->vfs_inode;
1581
1582         dout("writeback %p\n", inode);
1583         filemap_fdatawrite(&inode->i_data);
1584         iput(inode);
1585 }
1586
1587 /*
1588  * queue an async invalidation
1589  */
1590 void ceph_queue_invalidate(struct inode *inode)
1591 {
1592         ihold(inode);
1593         if (queue_work(ceph_inode_to_client(inode)->pg_inv_wq,
1594                        &ceph_inode(inode)->i_pg_inv_work)) {
1595                 dout("ceph_queue_invalidate %p\n", inode);
1596         } else {
1597                 dout("ceph_queue_invalidate %p failed\n", inode);
1598                 iput(inode);
1599         }
1600 }
1601
1602 /*
1603  * Invalidate inode pages in a worker thread.  (This can't be done
1604  * in the message handler context.)
1605  */
1606 static void ceph_invalidate_work(struct work_struct *work)
1607 {
1608         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1609                                                   i_pg_inv_work);
1610         struct inode *inode = &ci->vfs_inode;
1611         u32 orig_gen;
1612         int check = 0;
1613
1614         mutex_lock(&ci->i_truncate_mutex);
1615         spin_lock(&ci->i_ceph_lock);
1616         dout("invalidate_pages %p gen %d revoking %d\n", inode,
1617              ci->i_rdcache_gen, ci->i_rdcache_revoking);
1618         if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
1619                 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1620                         check = 1;
1621                 spin_unlock(&ci->i_ceph_lock);
1622                 mutex_unlock(&ci->i_truncate_mutex);
1623                 goto out;
1624         }
1625         orig_gen = ci->i_rdcache_gen;
1626         spin_unlock(&ci->i_ceph_lock);
1627
1628         truncate_pagecache(inode, 0);
1629
1630         spin_lock(&ci->i_ceph_lock);
1631         if (orig_gen == ci->i_rdcache_gen &&
1632             orig_gen == ci->i_rdcache_revoking) {
1633                 dout("invalidate_pages %p gen %d successful\n", inode,
1634                      ci->i_rdcache_gen);
1635                 ci->i_rdcache_revoking--;
1636                 check = 1;
1637         } else {
1638                 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1639                      inode, orig_gen, ci->i_rdcache_gen,
1640                      ci->i_rdcache_revoking);
1641                 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1642                         check = 1;
1643         }
1644         spin_unlock(&ci->i_ceph_lock);
1645         mutex_unlock(&ci->i_truncate_mutex);
1646 out:
1647         if (check)
1648                 ceph_check_caps(ci, 0, NULL);
1649         iput(inode);
1650 }
1651
1652
1653 /*
1654  * called by trunc_wq;
1655  *
1656  * We also truncate in a separate thread as well.
1657  */
1658 static void ceph_vmtruncate_work(struct work_struct *work)
1659 {
1660         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1661                                                   i_vmtruncate_work);
1662         struct inode *inode = &ci->vfs_inode;
1663
1664         dout("vmtruncate_work %p\n", inode);
1665         __ceph_do_pending_vmtruncate(inode);
1666         iput(inode);
1667 }
1668
1669 /*
1670  * Queue an async vmtruncate.  If we fail to queue work, we will handle
1671  * the truncation the next time we call __ceph_do_pending_vmtruncate.
1672  */
1673 void ceph_queue_vmtruncate(struct inode *inode)
1674 {
1675         struct ceph_inode_info *ci = ceph_inode(inode);
1676
1677         ihold(inode);
1678
1679         if (queue_work(ceph_sb_to_client(inode->i_sb)->trunc_wq,
1680                        &ci->i_vmtruncate_work)) {
1681                 dout("ceph_queue_vmtruncate %p\n", inode);
1682         } else {
1683                 dout("ceph_queue_vmtruncate %p failed, pending=%d\n",
1684                      inode, ci->i_truncate_pending);
1685                 iput(inode);
1686         }
1687 }
1688
1689 /*
1690  * Make sure any pending truncation is applied before doing anything
1691  * that may depend on it.
1692  */
1693 void __ceph_do_pending_vmtruncate(struct inode *inode)
1694 {
1695         struct ceph_inode_info *ci = ceph_inode(inode);
1696         u64 to;
1697         int wrbuffer_refs, finish = 0;
1698
1699         mutex_lock(&ci->i_truncate_mutex);
1700 retry:
1701         spin_lock(&ci->i_ceph_lock);
1702         if (ci->i_truncate_pending == 0) {
1703                 dout("__do_pending_vmtruncate %p none pending\n", inode);
1704                 spin_unlock(&ci->i_ceph_lock);
1705                 mutex_unlock(&ci->i_truncate_mutex);
1706                 return;
1707         }
1708
1709         /*
1710          * make sure any dirty snapped pages are flushed before we
1711          * possibly truncate them.. so write AND block!
1712          */
1713         if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) {
1714                 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1715                      inode);
1716                 spin_unlock(&ci->i_ceph_lock);
1717                 filemap_write_and_wait_range(&inode->i_data, 0,
1718                                              inode->i_sb->s_maxbytes);
1719                 goto retry;
1720         }
1721
1722         /* there should be no reader or writer */
1723         WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref);
1724
1725         to = ci->i_truncate_size;
1726         wrbuffer_refs = ci->i_wrbuffer_ref;
1727         dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode,
1728              ci->i_truncate_pending, to);
1729         spin_unlock(&ci->i_ceph_lock);
1730
1731         truncate_pagecache(inode, to);
1732
1733         spin_lock(&ci->i_ceph_lock);
1734         if (to == ci->i_truncate_size) {
1735                 ci->i_truncate_pending = 0;
1736                 finish = 1;
1737         }
1738         spin_unlock(&ci->i_ceph_lock);
1739         if (!finish)
1740                 goto retry;
1741
1742         mutex_unlock(&ci->i_truncate_mutex);
1743
1744         if (wrbuffer_refs == 0)
1745                 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
1746
1747         wake_up_all(&ci->i_cap_wq);
1748 }
1749
1750 /*
1751  * symlinks
1752  */
1753 static const struct inode_operations ceph_symlink_iops = {
1754         .readlink = generic_readlink,
1755         .get_link = simple_get_link,
1756         .setattr = ceph_setattr,
1757         .getattr = ceph_getattr,
1758         .setxattr = ceph_setxattr,
1759         .getxattr = ceph_getxattr,
1760         .listxattr = ceph_listxattr,
1761         .removexattr = ceph_removexattr,
1762 };
1763
1764 /*
1765  * setattr
1766  */
1767 int ceph_setattr(struct dentry *dentry, struct iattr *attr)
1768 {
1769         struct inode *inode = d_inode(dentry);
1770         struct ceph_inode_info *ci = ceph_inode(inode);
1771         const unsigned int ia_valid = attr->ia_valid;
1772         struct ceph_mds_request *req;
1773         struct ceph_mds_client *mdsc = ceph_sb_to_client(dentry->d_sb)->mdsc;
1774         struct ceph_cap_flush *prealloc_cf;
1775         int issued;
1776         int release = 0, dirtied = 0;
1777         int mask = 0;
1778         int err = 0;
1779         int inode_dirty_flags = 0;
1780         bool lock_snap_rwsem = false;
1781
1782         if (ceph_snap(inode) != CEPH_NOSNAP)
1783                 return -EROFS;
1784
1785         err = inode_change_ok(inode, attr);
1786         if (err != 0)
1787                 return err;
1788
1789         prealloc_cf = ceph_alloc_cap_flush();
1790         if (!prealloc_cf)
1791                 return -ENOMEM;
1792
1793         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR,
1794                                        USE_AUTH_MDS);
1795         if (IS_ERR(req)) {
1796                 ceph_free_cap_flush(prealloc_cf);
1797                 return PTR_ERR(req);
1798         }
1799
1800         spin_lock(&ci->i_ceph_lock);
1801         issued = __ceph_caps_issued(ci, NULL);
1802
1803         if (!ci->i_head_snapc &&
1804             (issued & (CEPH_CAP_ANY_EXCL | CEPH_CAP_FILE_WR))) {
1805                 lock_snap_rwsem = true;
1806                 if (!down_read_trylock(&mdsc->snap_rwsem)) {
1807                         spin_unlock(&ci->i_ceph_lock);
1808                         down_read(&mdsc->snap_rwsem);
1809                         spin_lock(&ci->i_ceph_lock);
1810                         issued = __ceph_caps_issued(ci, NULL);
1811                 }
1812         }
1813
1814         dout("setattr %p issued %s\n", inode, ceph_cap_string(issued));
1815
1816         if (ia_valid & ATTR_UID) {
1817                 dout("setattr %p uid %d -> %d\n", inode,
1818                      from_kuid(&init_user_ns, inode->i_uid),
1819                      from_kuid(&init_user_ns, attr->ia_uid));
1820                 if (issued & CEPH_CAP_AUTH_EXCL) {
1821                         inode->i_uid = attr->ia_uid;
1822                         dirtied |= CEPH_CAP_AUTH_EXCL;
1823                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1824                            !uid_eq(attr->ia_uid, inode->i_uid)) {
1825                         req->r_args.setattr.uid = cpu_to_le32(
1826                                 from_kuid(&init_user_ns, attr->ia_uid));
1827                         mask |= CEPH_SETATTR_UID;
1828                         release |= CEPH_CAP_AUTH_SHARED;
1829                 }
1830         }
1831         if (ia_valid & ATTR_GID) {
1832                 dout("setattr %p gid %d -> %d\n", inode,
1833                      from_kgid(&init_user_ns, inode->i_gid),
1834                      from_kgid(&init_user_ns, attr->ia_gid));
1835                 if (issued & CEPH_CAP_AUTH_EXCL) {
1836                         inode->i_gid = attr->ia_gid;
1837                         dirtied |= CEPH_CAP_AUTH_EXCL;
1838                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1839                            !gid_eq(attr->ia_gid, inode->i_gid)) {
1840                         req->r_args.setattr.gid = cpu_to_le32(
1841                                 from_kgid(&init_user_ns, attr->ia_gid));
1842                         mask |= CEPH_SETATTR_GID;
1843                         release |= CEPH_CAP_AUTH_SHARED;
1844                 }
1845         }
1846         if (ia_valid & ATTR_MODE) {
1847                 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode,
1848                      attr->ia_mode);
1849                 if (issued & CEPH_CAP_AUTH_EXCL) {
1850                         inode->i_mode = attr->ia_mode;
1851                         dirtied |= CEPH_CAP_AUTH_EXCL;
1852                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1853                            attr->ia_mode != inode->i_mode) {
1854                         inode->i_mode = attr->ia_mode;
1855                         req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode);
1856                         mask |= CEPH_SETATTR_MODE;
1857                         release |= CEPH_CAP_AUTH_SHARED;
1858                 }
1859         }
1860
1861         if (ia_valid & ATTR_ATIME) {
1862                 dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode,
1863                      inode->i_atime.tv_sec, inode->i_atime.tv_nsec,
1864                      attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec);
1865                 if (issued & CEPH_CAP_FILE_EXCL) {
1866                         ci->i_time_warp_seq++;
1867                         inode->i_atime = attr->ia_atime;
1868                         dirtied |= CEPH_CAP_FILE_EXCL;
1869                 } else if ((issued & CEPH_CAP_FILE_WR) &&
1870                            timespec_compare(&inode->i_atime,
1871                                             &attr->ia_atime) < 0) {
1872                         inode->i_atime = attr->ia_atime;
1873                         dirtied |= CEPH_CAP_FILE_WR;
1874                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1875                            !timespec_equal(&inode->i_atime, &attr->ia_atime)) {
1876                         ceph_encode_timespec(&req->r_args.setattr.atime,
1877                                              &attr->ia_atime);
1878                         mask |= CEPH_SETATTR_ATIME;
1879                         release |= CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_RD |
1880                                 CEPH_CAP_FILE_WR;
1881                 }
1882         }
1883         if (ia_valid & ATTR_MTIME) {
1884                 dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode,
1885                      inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
1886                      attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec);
1887                 if (issued & CEPH_CAP_FILE_EXCL) {
1888                         ci->i_time_warp_seq++;
1889                         inode->i_mtime = attr->ia_mtime;
1890                         dirtied |= CEPH_CAP_FILE_EXCL;
1891                 } else if ((issued & CEPH_CAP_FILE_WR) &&
1892                            timespec_compare(&inode->i_mtime,
1893                                             &attr->ia_mtime) < 0) {
1894                         inode->i_mtime = attr->ia_mtime;
1895                         dirtied |= CEPH_CAP_FILE_WR;
1896                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1897                            !timespec_equal(&inode->i_mtime, &attr->ia_mtime)) {
1898                         ceph_encode_timespec(&req->r_args.setattr.mtime,
1899                                              &attr->ia_mtime);
1900                         mask |= CEPH_SETATTR_MTIME;
1901                         release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
1902                                 CEPH_CAP_FILE_WR;
1903                 }
1904         }
1905         if (ia_valid & ATTR_SIZE) {
1906                 dout("setattr %p size %lld -> %lld\n", inode,
1907                      inode->i_size, attr->ia_size);
1908                 if ((issued & CEPH_CAP_FILE_EXCL) &&
1909                     attr->ia_size > inode->i_size) {
1910                         i_size_write(inode, attr->ia_size);
1911                         inode->i_blocks =
1912                                 (attr->ia_size + (1 << 9) - 1) >> 9;
1913                         inode->i_ctime = attr->ia_ctime;
1914                         ci->i_reported_size = attr->ia_size;
1915                         dirtied |= CEPH_CAP_FILE_EXCL;
1916                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1917                            attr->ia_size != inode->i_size) {
1918                         req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
1919                         req->r_args.setattr.old_size =
1920                                 cpu_to_le64(inode->i_size);
1921                         mask |= CEPH_SETATTR_SIZE;
1922                         release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
1923                                 CEPH_CAP_FILE_WR;
1924                 }
1925         }
1926
1927         /* these do nothing */
1928         if (ia_valid & ATTR_CTIME) {
1929                 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME|
1930                                          ATTR_MODE|ATTR_UID|ATTR_GID)) == 0;
1931                 dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode,
1932                      inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
1933                      attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec,
1934                      only ? "ctime only" : "ignored");
1935                 inode->i_ctime = attr->ia_ctime;
1936                 if (only) {
1937                         /*
1938                          * if kernel wants to dirty ctime but nothing else,
1939                          * we need to choose a cap to dirty under, or do
1940                          * a almost-no-op setattr
1941                          */
1942                         if (issued & CEPH_CAP_AUTH_EXCL)
1943                                 dirtied |= CEPH_CAP_AUTH_EXCL;
1944                         else if (issued & CEPH_CAP_FILE_EXCL)
1945                                 dirtied |= CEPH_CAP_FILE_EXCL;
1946                         else if (issued & CEPH_CAP_XATTR_EXCL)
1947                                 dirtied |= CEPH_CAP_XATTR_EXCL;
1948                         else
1949                                 mask |= CEPH_SETATTR_CTIME;
1950                 }
1951         }
1952         if (ia_valid & ATTR_FILE)
1953                 dout("setattr %p ATTR_FILE ... hrm!\n", inode);
1954
1955         if (dirtied) {
1956                 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied,
1957                                                            &prealloc_cf);
1958                 inode->i_ctime = current_fs_time(inode->i_sb);
1959         }
1960
1961         release &= issued;
1962         spin_unlock(&ci->i_ceph_lock);
1963         if (lock_snap_rwsem)
1964                 up_read(&mdsc->snap_rwsem);
1965
1966         if (inode_dirty_flags)
1967                 __mark_inode_dirty(inode, inode_dirty_flags);
1968
1969         if (ia_valid & ATTR_MODE) {
1970                 err = posix_acl_chmod(inode, attr->ia_mode);
1971                 if (err)
1972                         goto out_put;
1973         }
1974
1975         if (mask) {
1976                 req->r_inode = inode;
1977                 ihold(inode);
1978                 req->r_inode_drop = release;
1979                 req->r_args.setattr.mask = cpu_to_le32(mask);
1980                 req->r_num_caps = 1;
1981                 err = ceph_mdsc_do_request(mdsc, NULL, req);
1982         }
1983         dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err,
1984              ceph_cap_string(dirtied), mask);
1985
1986         ceph_mdsc_put_request(req);
1987         if (mask & CEPH_SETATTR_SIZE)
1988                 __ceph_do_pending_vmtruncate(inode);
1989         ceph_free_cap_flush(prealloc_cf);
1990         return err;
1991 out_put:
1992         ceph_mdsc_put_request(req);
1993         ceph_free_cap_flush(prealloc_cf);
1994         return err;
1995 }
1996
1997 /*
1998  * Verify that we have a lease on the given mask.  If not,
1999  * do a getattr against an mds.
2000  */
2001 int __ceph_do_getattr(struct inode *inode, struct page *locked_page,
2002                       int mask, bool force)
2003 {
2004         struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
2005         struct ceph_mds_client *mdsc = fsc->mdsc;
2006         struct ceph_mds_request *req;
2007         int err;
2008
2009         if (ceph_snap(inode) == CEPH_SNAPDIR) {
2010                 dout("do_getattr inode %p SNAPDIR\n", inode);
2011                 return 0;
2012         }
2013
2014         dout("do_getattr inode %p mask %s mode 0%o\n",
2015              inode, ceph_cap_string(mask), inode->i_mode);
2016         if (!force && ceph_caps_issued_mask(ceph_inode(inode), mask, 1))
2017                 return 0;
2018
2019         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
2020         if (IS_ERR(req))
2021                 return PTR_ERR(req);
2022         req->r_inode = inode;
2023         ihold(inode);
2024         req->r_num_caps = 1;
2025         req->r_args.getattr.mask = cpu_to_le32(mask);
2026         req->r_locked_page = locked_page;
2027         err = ceph_mdsc_do_request(mdsc, NULL, req);
2028         if (locked_page && err == 0) {
2029                 u64 inline_version = req->r_reply_info.targeti.inline_version;
2030                 if (inline_version == 0) {
2031                         /* the reply is supposed to contain inline data */
2032                         err = -EINVAL;
2033                 } else if (inline_version == CEPH_INLINE_NONE) {
2034                         err = -ENODATA;
2035                 } else {
2036                         err = req->r_reply_info.targeti.inline_len;
2037                 }
2038         }
2039         ceph_mdsc_put_request(req);
2040         dout("do_getattr result=%d\n", err);
2041         return err;
2042 }
2043
2044
2045 /*
2046  * Check inode permissions.  We verify we have a valid value for
2047  * the AUTH cap, then call the generic handler.
2048  */
2049 int ceph_permission(struct inode *inode, int mask)
2050 {
2051         int err;
2052
2053         if (mask & MAY_NOT_BLOCK)
2054                 return -ECHILD;
2055
2056         err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED, false);
2057
2058         if (!err)
2059                 err = generic_permission(inode, mask);
2060         return err;
2061 }
2062
2063 /*
2064  * Get all attributes.  Hopefully somedata we'll have a statlite()
2065  * and can limit the fields we require to be accurate.
2066  */
2067 int ceph_getattr(struct vfsmount *mnt, struct dentry *dentry,
2068                  struct kstat *stat)
2069 {
2070         struct inode *inode = d_inode(dentry);
2071         struct ceph_inode_info *ci = ceph_inode(inode);
2072         int err;
2073
2074         err = ceph_do_getattr(inode, CEPH_STAT_CAP_INODE_ALL, false);
2075         if (!err) {
2076                 generic_fillattr(inode, stat);
2077                 stat->ino = ceph_translate_ino(inode->i_sb, inode->i_ino);
2078                 if (ceph_snap(inode) != CEPH_NOSNAP)
2079                         stat->dev = ceph_snap(inode);
2080                 else
2081                         stat->dev = 0;
2082                 if (S_ISDIR(inode->i_mode)) {
2083                         if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb),
2084                                                 RBYTES))
2085                                 stat->size = ci->i_rbytes;
2086                         else
2087                                 stat->size = ci->i_files + ci->i_subdirs;
2088                         stat->blocks = 0;
2089                         stat->blksize = 65536;
2090                 }
2091         }
2092         return err;
2093 }