ceph: introduce a new inode flag indicating if cached dentries are ordered
[firefly-linux-kernel-4.4.55.git] / fs / ceph / inode.c
1 #include <linux/ceph/ceph_debug.h>
2
3 #include <linux/module.h>
4 #include <linux/fs.h>
5 #include <linux/slab.h>
6 #include <linux/string.h>
7 #include <linux/uaccess.h>
8 #include <linux/kernel.h>
9 #include <linux/namei.h>
10 #include <linux/writeback.h>
11 #include <linux/vmalloc.h>
12 #include <linux/posix_acl.h>
13 #include <linux/random.h>
14
15 #include "super.h"
16 #include "mds_client.h"
17 #include "cache.h"
18 #include <linux/ceph/decode.h>
19
20 /*
21  * Ceph inode operations
22  *
23  * Implement basic inode helpers (get, alloc) and inode ops (getattr,
24  * setattr, etc.), xattr helpers, and helpers for assimilating
25  * metadata returned by the MDS into our cache.
26  *
27  * Also define helpers for doing asynchronous writeback, invalidation,
28  * and truncation for the benefit of those who can't afford to block
29  * (typically because they are in the message handler path).
30  */
31
32 static const struct inode_operations ceph_symlink_iops;
33
34 static void ceph_invalidate_work(struct work_struct *work);
35 static void ceph_writeback_work(struct work_struct *work);
36 static void ceph_vmtruncate_work(struct work_struct *work);
37
38 /*
39  * find or create an inode, given the ceph ino number
40  */
41 static int ceph_set_ino_cb(struct inode *inode, void *data)
42 {
43         ceph_inode(inode)->i_vino = *(struct ceph_vino *)data;
44         inode->i_ino = ceph_vino_to_ino(*(struct ceph_vino *)data);
45         return 0;
46 }
47
48 struct inode *ceph_get_inode(struct super_block *sb, struct ceph_vino vino)
49 {
50         struct inode *inode;
51         ino_t t = ceph_vino_to_ino(vino);
52
53         inode = iget5_locked(sb, t, ceph_ino_compare, ceph_set_ino_cb, &vino);
54         if (inode == NULL)
55                 return ERR_PTR(-ENOMEM);
56         if (inode->i_state & I_NEW) {
57                 dout("get_inode created new inode %p %llx.%llx ino %llx\n",
58                      inode, ceph_vinop(inode), (u64)inode->i_ino);
59                 unlock_new_inode(inode);
60         }
61
62         dout("get_inode on %lu=%llx.%llx got %p\n", inode->i_ino, vino.ino,
63              vino.snap, inode);
64         return inode;
65 }
66
67 /*
68  * get/constuct snapdir inode for a given directory
69  */
70 struct inode *ceph_get_snapdir(struct inode *parent)
71 {
72         struct ceph_vino vino = {
73                 .ino = ceph_ino(parent),
74                 .snap = CEPH_SNAPDIR,
75         };
76         struct inode *inode = ceph_get_inode(parent->i_sb, vino);
77         struct ceph_inode_info *ci = ceph_inode(inode);
78
79         BUG_ON(!S_ISDIR(parent->i_mode));
80         if (IS_ERR(inode))
81                 return inode;
82         inode->i_mode = parent->i_mode;
83         inode->i_uid = parent->i_uid;
84         inode->i_gid = parent->i_gid;
85         inode->i_op = &ceph_dir_iops;
86         inode->i_fop = &ceph_dir_fops;
87         ci->i_snap_caps = CEPH_CAP_PIN; /* so we can open */
88         ci->i_rbytes = 0;
89         return inode;
90 }
91
92 const struct inode_operations ceph_file_iops = {
93         .permission = ceph_permission,
94         .setattr = ceph_setattr,
95         .getattr = ceph_getattr,
96         .setxattr = ceph_setxattr,
97         .getxattr = ceph_getxattr,
98         .listxattr = ceph_listxattr,
99         .removexattr = ceph_removexattr,
100         .get_acl = ceph_get_acl,
101         .set_acl = ceph_set_acl,
102 };
103
104
105 /*
106  * We use a 'frag tree' to keep track of the MDS's directory fragments
107  * for a given inode (usually there is just a single fragment).  We
108  * need to know when a child frag is delegated to a new MDS, or when
109  * it is flagged as replicated, so we can direct our requests
110  * accordingly.
111  */
112
113 /*
114  * find/create a frag in the tree
115  */
116 static struct ceph_inode_frag *__get_or_create_frag(struct ceph_inode_info *ci,
117                                                     u32 f)
118 {
119         struct rb_node **p;
120         struct rb_node *parent = NULL;
121         struct ceph_inode_frag *frag;
122         int c;
123
124         p = &ci->i_fragtree.rb_node;
125         while (*p) {
126                 parent = *p;
127                 frag = rb_entry(parent, struct ceph_inode_frag, node);
128                 c = ceph_frag_compare(f, frag->frag);
129                 if (c < 0)
130                         p = &(*p)->rb_left;
131                 else if (c > 0)
132                         p = &(*p)->rb_right;
133                 else
134                         return frag;
135         }
136
137         frag = kmalloc(sizeof(*frag), GFP_NOFS);
138         if (!frag) {
139                 pr_err("__get_or_create_frag ENOMEM on %p %llx.%llx "
140                        "frag %x\n", &ci->vfs_inode,
141                        ceph_vinop(&ci->vfs_inode), f);
142                 return ERR_PTR(-ENOMEM);
143         }
144         frag->frag = f;
145         frag->split_by = 0;
146         frag->mds = -1;
147         frag->ndist = 0;
148
149         rb_link_node(&frag->node, parent, p);
150         rb_insert_color(&frag->node, &ci->i_fragtree);
151
152         dout("get_or_create_frag added %llx.%llx frag %x\n",
153              ceph_vinop(&ci->vfs_inode), f);
154         return frag;
155 }
156
157 /*
158  * find a specific frag @f
159  */
160 struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci, u32 f)
161 {
162         struct rb_node *n = ci->i_fragtree.rb_node;
163
164         while (n) {
165                 struct ceph_inode_frag *frag =
166                         rb_entry(n, struct ceph_inode_frag, node);
167                 int c = ceph_frag_compare(f, frag->frag);
168                 if (c < 0)
169                         n = n->rb_left;
170                 else if (c > 0)
171                         n = n->rb_right;
172                 else
173                         return frag;
174         }
175         return NULL;
176 }
177
178 /*
179  * Choose frag containing the given value @v.  If @pfrag is
180  * specified, copy the frag delegation info to the caller if
181  * it is present.
182  */
183 static u32 __ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
184                               struct ceph_inode_frag *pfrag, int *found)
185 {
186         u32 t = ceph_frag_make(0, 0);
187         struct ceph_inode_frag *frag;
188         unsigned nway, i;
189         u32 n;
190
191         if (found)
192                 *found = 0;
193
194         while (1) {
195                 WARN_ON(!ceph_frag_contains_value(t, v));
196                 frag = __ceph_find_frag(ci, t);
197                 if (!frag)
198                         break; /* t is a leaf */
199                 if (frag->split_by == 0) {
200                         if (pfrag)
201                                 memcpy(pfrag, frag, sizeof(*pfrag));
202                         if (found)
203                                 *found = 1;
204                         break;
205                 }
206
207                 /* choose child */
208                 nway = 1 << frag->split_by;
209                 dout("choose_frag(%x) %x splits by %d (%d ways)\n", v, t,
210                      frag->split_by, nway);
211                 for (i = 0; i < nway; i++) {
212                         n = ceph_frag_make_child(t, frag->split_by, i);
213                         if (ceph_frag_contains_value(n, v)) {
214                                 t = n;
215                                 break;
216                         }
217                 }
218                 BUG_ON(i == nway);
219         }
220         dout("choose_frag(%x) = %x\n", v, t);
221
222         return t;
223 }
224
225 u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
226                      struct ceph_inode_frag *pfrag, int *found)
227 {
228         u32 ret;
229         mutex_lock(&ci->i_fragtree_mutex);
230         ret = __ceph_choose_frag(ci, v, pfrag, found);
231         mutex_unlock(&ci->i_fragtree_mutex);
232         return ret;
233 }
234
235 /*
236  * Process dirfrag (delegation) info from the mds.  Include leaf
237  * fragment in tree ONLY if ndist > 0.  Otherwise, only
238  * branches/splits are included in i_fragtree)
239  */
240 static int ceph_fill_dirfrag(struct inode *inode,
241                              struct ceph_mds_reply_dirfrag *dirinfo)
242 {
243         struct ceph_inode_info *ci = ceph_inode(inode);
244         struct ceph_inode_frag *frag;
245         u32 id = le32_to_cpu(dirinfo->frag);
246         int mds = le32_to_cpu(dirinfo->auth);
247         int ndist = le32_to_cpu(dirinfo->ndist);
248         int diri_auth = -1;
249         int i;
250         int err = 0;
251
252         spin_lock(&ci->i_ceph_lock);
253         if (ci->i_auth_cap)
254                 diri_auth = ci->i_auth_cap->mds;
255         spin_unlock(&ci->i_ceph_lock);
256
257         mutex_lock(&ci->i_fragtree_mutex);
258         if (ndist == 0 && mds == diri_auth) {
259                 /* no delegation info needed. */
260                 frag = __ceph_find_frag(ci, id);
261                 if (!frag)
262                         goto out;
263                 if (frag->split_by == 0) {
264                         /* tree leaf, remove */
265                         dout("fill_dirfrag removed %llx.%llx frag %x"
266                              " (no ref)\n", ceph_vinop(inode), id);
267                         rb_erase(&frag->node, &ci->i_fragtree);
268                         kfree(frag);
269                 } else {
270                         /* tree branch, keep and clear */
271                         dout("fill_dirfrag cleared %llx.%llx frag %x"
272                              " referral\n", ceph_vinop(inode), id);
273                         frag->mds = -1;
274                         frag->ndist = 0;
275                 }
276                 goto out;
277         }
278
279
280         /* find/add this frag to store mds delegation info */
281         frag = __get_or_create_frag(ci, id);
282         if (IS_ERR(frag)) {
283                 /* this is not the end of the world; we can continue
284                    with bad/inaccurate delegation info */
285                 pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n",
286                        ceph_vinop(inode), le32_to_cpu(dirinfo->frag));
287                 err = -ENOMEM;
288                 goto out;
289         }
290
291         frag->mds = mds;
292         frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP);
293         for (i = 0; i < frag->ndist; i++)
294                 frag->dist[i] = le32_to_cpu(dirinfo->dist[i]);
295         dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n",
296              ceph_vinop(inode), frag->frag, frag->ndist);
297
298 out:
299         mutex_unlock(&ci->i_fragtree_mutex);
300         return err;
301 }
302
303 static int ceph_fill_fragtree(struct inode *inode,
304                               struct ceph_frag_tree_head *fragtree,
305                               struct ceph_mds_reply_dirfrag *dirinfo)
306 {
307         struct ceph_inode_info *ci = ceph_inode(inode);
308         struct ceph_inode_frag *frag;
309         struct rb_node *rb_node;
310         int i;
311         u32 id, nsplits;
312         bool update = false;
313
314         mutex_lock(&ci->i_fragtree_mutex);
315         nsplits = le32_to_cpu(fragtree->nsplits);
316         if (nsplits) {
317                 i = prandom_u32() % nsplits;
318                 id = le32_to_cpu(fragtree->splits[i].frag);
319                 if (!__ceph_find_frag(ci, id))
320                         update = true;
321         } else if (!RB_EMPTY_ROOT(&ci->i_fragtree)) {
322                 rb_node = rb_first(&ci->i_fragtree);
323                 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
324                 if (frag->frag != ceph_frag_make(0, 0) || rb_next(rb_node))
325                         update = true;
326         }
327         if (!update && dirinfo) {
328                 id = le32_to_cpu(dirinfo->frag);
329                 if (id != __ceph_choose_frag(ci, id, NULL, NULL))
330                         update = true;
331         }
332         if (!update)
333                 goto out_unlock;
334
335         dout("fill_fragtree %llx.%llx\n", ceph_vinop(inode));
336         rb_node = rb_first(&ci->i_fragtree);
337         for (i = 0; i < nsplits; i++) {
338                 id = le32_to_cpu(fragtree->splits[i].frag);
339                 frag = NULL;
340                 while (rb_node) {
341                         frag = rb_entry(rb_node, struct ceph_inode_frag, node);
342                         if (ceph_frag_compare(frag->frag, id) >= 0) {
343                                 if (frag->frag != id)
344                                         frag = NULL;
345                                 else
346                                         rb_node = rb_next(rb_node);
347                                 break;
348                         }
349                         rb_node = rb_next(rb_node);
350                         rb_erase(&frag->node, &ci->i_fragtree);
351                         kfree(frag);
352                         frag = NULL;
353                 }
354                 if (!frag) {
355                         frag = __get_or_create_frag(ci, id);
356                         if (IS_ERR(frag))
357                                 continue;
358                 }
359                 frag->split_by = le32_to_cpu(fragtree->splits[i].by);
360                 dout(" frag %x split by %d\n", frag->frag, frag->split_by);
361         }
362         while (rb_node) {
363                 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
364                 rb_node = rb_next(rb_node);
365                 rb_erase(&frag->node, &ci->i_fragtree);
366                 kfree(frag);
367         }
368 out_unlock:
369         mutex_unlock(&ci->i_fragtree_mutex);
370         return 0;
371 }
372
373 /*
374  * initialize a newly allocated inode.
375  */
376 struct inode *ceph_alloc_inode(struct super_block *sb)
377 {
378         struct ceph_inode_info *ci;
379         int i;
380
381         ci = kmem_cache_alloc(ceph_inode_cachep, GFP_NOFS);
382         if (!ci)
383                 return NULL;
384
385         dout("alloc_inode %p\n", &ci->vfs_inode);
386
387         spin_lock_init(&ci->i_ceph_lock);
388
389         ci->i_version = 0;
390         ci->i_time_warp_seq = 0;
391         ci->i_ceph_flags = 0;
392         ci->i_ordered_count = 0;
393         atomic_set(&ci->i_release_count, 1);
394         atomic_set(&ci->i_complete_count, 0);
395         ci->i_symlink = NULL;
396
397         memset(&ci->i_dir_layout, 0, sizeof(ci->i_dir_layout));
398
399         ci->i_fragtree = RB_ROOT;
400         mutex_init(&ci->i_fragtree_mutex);
401
402         ci->i_xattrs.blob = NULL;
403         ci->i_xattrs.prealloc_blob = NULL;
404         ci->i_xattrs.dirty = false;
405         ci->i_xattrs.index = RB_ROOT;
406         ci->i_xattrs.count = 0;
407         ci->i_xattrs.names_size = 0;
408         ci->i_xattrs.vals_size = 0;
409         ci->i_xattrs.version = 0;
410         ci->i_xattrs.index_version = 0;
411
412         ci->i_caps = RB_ROOT;
413         ci->i_auth_cap = NULL;
414         ci->i_dirty_caps = 0;
415         ci->i_flushing_caps = 0;
416         INIT_LIST_HEAD(&ci->i_dirty_item);
417         INIT_LIST_HEAD(&ci->i_flushing_item);
418         ci->i_cap_flush_seq = 0;
419         ci->i_cap_flush_last_tid = 0;
420         memset(&ci->i_cap_flush_tid, 0, sizeof(ci->i_cap_flush_tid));
421         init_waitqueue_head(&ci->i_cap_wq);
422         ci->i_hold_caps_min = 0;
423         ci->i_hold_caps_max = 0;
424         INIT_LIST_HEAD(&ci->i_cap_delay_list);
425         INIT_LIST_HEAD(&ci->i_cap_snaps);
426         ci->i_head_snapc = NULL;
427         ci->i_snap_caps = 0;
428
429         for (i = 0; i < CEPH_FILE_MODE_NUM; i++)
430                 ci->i_nr_by_mode[i] = 0;
431
432         mutex_init(&ci->i_truncate_mutex);
433         ci->i_truncate_seq = 0;
434         ci->i_truncate_size = 0;
435         ci->i_truncate_pending = 0;
436
437         ci->i_max_size = 0;
438         ci->i_reported_size = 0;
439         ci->i_wanted_max_size = 0;
440         ci->i_requested_max_size = 0;
441
442         ci->i_pin_ref = 0;
443         ci->i_rd_ref = 0;
444         ci->i_rdcache_ref = 0;
445         ci->i_wr_ref = 0;
446         ci->i_wb_ref = 0;
447         ci->i_wrbuffer_ref = 0;
448         ci->i_wrbuffer_ref_head = 0;
449         ci->i_shared_gen = 0;
450         ci->i_rdcache_gen = 0;
451         ci->i_rdcache_revoking = 0;
452
453         INIT_LIST_HEAD(&ci->i_unsafe_writes);
454         INIT_LIST_HEAD(&ci->i_unsafe_dirops);
455         spin_lock_init(&ci->i_unsafe_lock);
456
457         ci->i_snap_realm = NULL;
458         INIT_LIST_HEAD(&ci->i_snap_realm_item);
459         INIT_LIST_HEAD(&ci->i_snap_flush_item);
460
461         INIT_WORK(&ci->i_wb_work, ceph_writeback_work);
462         INIT_WORK(&ci->i_pg_inv_work, ceph_invalidate_work);
463
464         INIT_WORK(&ci->i_vmtruncate_work, ceph_vmtruncate_work);
465
466         ceph_fscache_inode_init(ci);
467
468         return &ci->vfs_inode;
469 }
470
471 static void ceph_i_callback(struct rcu_head *head)
472 {
473         struct inode *inode = container_of(head, struct inode, i_rcu);
474         struct ceph_inode_info *ci = ceph_inode(inode);
475
476         kmem_cache_free(ceph_inode_cachep, ci);
477 }
478
479 void ceph_destroy_inode(struct inode *inode)
480 {
481         struct ceph_inode_info *ci = ceph_inode(inode);
482         struct ceph_inode_frag *frag;
483         struct rb_node *n;
484
485         dout("destroy_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode));
486
487         ceph_fscache_unregister_inode_cookie(ci);
488
489         ceph_queue_caps_release(inode);
490
491         /*
492          * we may still have a snap_realm reference if there are stray
493          * caps in i_snap_caps.
494          */
495         if (ci->i_snap_realm) {
496                 struct ceph_mds_client *mdsc =
497                         ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc;
498                 struct ceph_snap_realm *realm = ci->i_snap_realm;
499
500                 dout(" dropping residual ref to snap realm %p\n", realm);
501                 spin_lock(&realm->inodes_with_caps_lock);
502                 list_del_init(&ci->i_snap_realm_item);
503                 spin_unlock(&realm->inodes_with_caps_lock);
504                 ceph_put_snap_realm(mdsc, realm);
505         }
506
507         kfree(ci->i_symlink);
508         while ((n = rb_first(&ci->i_fragtree)) != NULL) {
509                 frag = rb_entry(n, struct ceph_inode_frag, node);
510                 rb_erase(n, &ci->i_fragtree);
511                 kfree(frag);
512         }
513
514         __ceph_destroy_xattrs(ci);
515         if (ci->i_xattrs.blob)
516                 ceph_buffer_put(ci->i_xattrs.blob);
517         if (ci->i_xattrs.prealloc_blob)
518                 ceph_buffer_put(ci->i_xattrs.prealloc_blob);
519
520         call_rcu(&inode->i_rcu, ceph_i_callback);
521 }
522
523 int ceph_drop_inode(struct inode *inode)
524 {
525         /*
526          * Positve dentry and corresponding inode are always accompanied
527          * in MDS reply. So no need to keep inode in the cache after
528          * dropping all its aliases.
529          */
530         return 1;
531 }
532
533 /*
534  * Helpers to fill in size, ctime, mtime, and atime.  We have to be
535  * careful because either the client or MDS may have more up to date
536  * info, depending on which capabilities are held, and whether
537  * time_warp_seq or truncate_seq have increased.  (Ordinarily, mtime
538  * and size are monotonically increasing, except when utimes() or
539  * truncate() increments the corresponding _seq values.)
540  */
541 int ceph_fill_file_size(struct inode *inode, int issued,
542                         u32 truncate_seq, u64 truncate_size, u64 size)
543 {
544         struct ceph_inode_info *ci = ceph_inode(inode);
545         int queue_trunc = 0;
546
547         if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 ||
548             (truncate_seq == ci->i_truncate_seq && size > inode->i_size)) {
549                 dout("size %lld -> %llu\n", inode->i_size, size);
550                 inode->i_size = size;
551                 inode->i_blocks = (size + (1<<9) - 1) >> 9;
552                 ci->i_reported_size = size;
553                 if (truncate_seq != ci->i_truncate_seq) {
554                         dout("truncate_seq %u -> %u\n",
555                              ci->i_truncate_seq, truncate_seq);
556                         ci->i_truncate_seq = truncate_seq;
557
558                         /* the MDS should have revoked these caps */
559                         WARN_ON_ONCE(issued & (CEPH_CAP_FILE_EXCL |
560                                                CEPH_CAP_FILE_RD |
561                                                CEPH_CAP_FILE_WR |
562                                                CEPH_CAP_FILE_LAZYIO));
563                         /*
564                          * If we hold relevant caps, or in the case where we're
565                          * not the only client referencing this file and we
566                          * don't hold those caps, then we need to check whether
567                          * the file is either opened or mmaped
568                          */
569                         if ((issued & (CEPH_CAP_FILE_CACHE|
570                                        CEPH_CAP_FILE_BUFFER)) ||
571                             mapping_mapped(inode->i_mapping) ||
572                             __ceph_caps_file_wanted(ci)) {
573                                 ci->i_truncate_pending++;
574                                 queue_trunc = 1;
575                         }
576                 }
577         }
578         if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 &&
579             ci->i_truncate_size != truncate_size) {
580                 dout("truncate_size %lld -> %llu\n", ci->i_truncate_size,
581                      truncate_size);
582                 ci->i_truncate_size = truncate_size;
583         }
584
585         if (queue_trunc)
586                 ceph_fscache_invalidate(inode);
587
588         return queue_trunc;
589 }
590
591 void ceph_fill_file_time(struct inode *inode, int issued,
592                          u64 time_warp_seq, struct timespec *ctime,
593                          struct timespec *mtime, struct timespec *atime)
594 {
595         struct ceph_inode_info *ci = ceph_inode(inode);
596         int warn = 0;
597
598         if (issued & (CEPH_CAP_FILE_EXCL|
599                       CEPH_CAP_FILE_WR|
600                       CEPH_CAP_FILE_BUFFER|
601                       CEPH_CAP_AUTH_EXCL|
602                       CEPH_CAP_XATTR_EXCL)) {
603                 if (timespec_compare(ctime, &inode->i_ctime) > 0) {
604                         dout("ctime %ld.%09ld -> %ld.%09ld inc w/ cap\n",
605                              inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
606                              ctime->tv_sec, ctime->tv_nsec);
607                         inode->i_ctime = *ctime;
608                 }
609                 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) {
610                         /* the MDS did a utimes() */
611                         dout("mtime %ld.%09ld -> %ld.%09ld "
612                              "tw %d -> %d\n",
613                              inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
614                              mtime->tv_sec, mtime->tv_nsec,
615                              ci->i_time_warp_seq, (int)time_warp_seq);
616
617                         inode->i_mtime = *mtime;
618                         inode->i_atime = *atime;
619                         ci->i_time_warp_seq = time_warp_seq;
620                 } else if (time_warp_seq == ci->i_time_warp_seq) {
621                         /* nobody did utimes(); take the max */
622                         if (timespec_compare(mtime, &inode->i_mtime) > 0) {
623                                 dout("mtime %ld.%09ld -> %ld.%09ld inc\n",
624                                      inode->i_mtime.tv_sec,
625                                      inode->i_mtime.tv_nsec,
626                                      mtime->tv_sec, mtime->tv_nsec);
627                                 inode->i_mtime = *mtime;
628                         }
629                         if (timespec_compare(atime, &inode->i_atime) > 0) {
630                                 dout("atime %ld.%09ld -> %ld.%09ld inc\n",
631                                      inode->i_atime.tv_sec,
632                                      inode->i_atime.tv_nsec,
633                                      atime->tv_sec, atime->tv_nsec);
634                                 inode->i_atime = *atime;
635                         }
636                 } else if (issued & CEPH_CAP_FILE_EXCL) {
637                         /* we did a utimes(); ignore mds values */
638                 } else {
639                         warn = 1;
640                 }
641         } else {
642                 /* we have no write|excl caps; whatever the MDS says is true */
643                 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) {
644                         inode->i_ctime = *ctime;
645                         inode->i_mtime = *mtime;
646                         inode->i_atime = *atime;
647                         ci->i_time_warp_seq = time_warp_seq;
648                 } else {
649                         warn = 1;
650                 }
651         }
652         if (warn) /* time_warp_seq shouldn't go backwards */
653                 dout("%p mds time_warp_seq %llu < %u\n",
654                      inode, time_warp_seq, ci->i_time_warp_seq);
655 }
656
657 /*
658  * Populate an inode based on info from mds.  May be called on new or
659  * existing inodes.
660  */
661 static int fill_inode(struct inode *inode,
662                       struct ceph_mds_reply_info_in *iinfo,
663                       struct ceph_mds_reply_dirfrag *dirinfo,
664                       struct ceph_mds_session *session,
665                       unsigned long ttl_from, int cap_fmode,
666                       struct ceph_cap_reservation *caps_reservation)
667 {
668         struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
669         struct ceph_mds_reply_inode *info = iinfo->in;
670         struct ceph_inode_info *ci = ceph_inode(inode);
671         int issued = 0, implemented, new_issued;
672         struct timespec mtime, atime, ctime;
673         struct ceph_buffer *xattr_blob = NULL;
674         struct ceph_cap *new_cap = NULL;
675         int err = 0;
676         bool wake = false;
677         bool queue_trunc = false;
678         bool new_version = false;
679
680         dout("fill_inode %p ino %llx.%llx v %llu had %llu\n",
681              inode, ceph_vinop(inode), le64_to_cpu(info->version),
682              ci->i_version);
683
684         /* prealloc new cap struct */
685         if (info->cap.caps && ceph_snap(inode) == CEPH_NOSNAP)
686                 new_cap = ceph_get_cap(mdsc, caps_reservation);
687
688         /*
689          * prealloc xattr data, if it looks like we'll need it.  only
690          * if len > 4 (meaning there are actually xattrs; the first 4
691          * bytes are the xattr count).
692          */
693         if (iinfo->xattr_len > 4) {
694                 xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS);
695                 if (!xattr_blob)
696                         pr_err("fill_inode ENOMEM xattr blob %d bytes\n",
697                                iinfo->xattr_len);
698         }
699
700         spin_lock(&ci->i_ceph_lock);
701
702         /*
703          * provided version will be odd if inode value is projected,
704          * even if stable.  skip the update if we have newer stable
705          * info (ours>=theirs, e.g. due to racing mds replies), unless
706          * we are getting projected (unstable) info (in which case the
707          * version is odd, and we want ours>theirs).
708          *   us   them
709          *   2    2     skip
710          *   3    2     skip
711          *   3    3     update
712          */
713         if (ci->i_version == 0 ||
714             ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
715              le64_to_cpu(info->version) > (ci->i_version & ~1)))
716                 new_version = true;
717
718         issued = __ceph_caps_issued(ci, &implemented);
719         issued |= implemented | __ceph_caps_dirty(ci);
720         new_issued = ~issued & le32_to_cpu(info->cap.caps);
721
722         /* update inode */
723         ci->i_version = le64_to_cpu(info->version);
724         inode->i_version++;
725         inode->i_rdev = le32_to_cpu(info->rdev);
726         inode->i_blkbits = fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1;
727
728         if ((new_version || (new_issued & CEPH_CAP_AUTH_SHARED)) &&
729             (issued & CEPH_CAP_AUTH_EXCL) == 0) {
730                 inode->i_mode = le32_to_cpu(info->mode);
731                 inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(info->uid));
732                 inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(info->gid));
733                 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
734                      from_kuid(&init_user_ns, inode->i_uid),
735                      from_kgid(&init_user_ns, inode->i_gid));
736         }
737
738         if ((new_version || (new_issued & CEPH_CAP_LINK_SHARED)) &&
739             (issued & CEPH_CAP_LINK_EXCL) == 0)
740                 set_nlink(inode, le32_to_cpu(info->nlink));
741
742         if (new_version || (new_issued & CEPH_CAP_ANY_RD)) {
743                 /* be careful with mtime, atime, size */
744                 ceph_decode_timespec(&atime, &info->atime);
745                 ceph_decode_timespec(&mtime, &info->mtime);
746                 ceph_decode_timespec(&ctime, &info->ctime);
747                 ceph_fill_file_time(inode, issued,
748                                 le32_to_cpu(info->time_warp_seq),
749                                 &ctime, &mtime, &atime);
750         }
751
752         if (new_version ||
753             (new_issued & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR))) {
754                 ci->i_layout = info->layout;
755                 queue_trunc = ceph_fill_file_size(inode, issued,
756                                         le32_to_cpu(info->truncate_seq),
757                                         le64_to_cpu(info->truncate_size),
758                                         le64_to_cpu(info->size));
759                 /* only update max_size on auth cap */
760                 if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
761                     ci->i_max_size != le64_to_cpu(info->max_size)) {
762                         dout("max_size %lld -> %llu\n", ci->i_max_size,
763                                         le64_to_cpu(info->max_size));
764                         ci->i_max_size = le64_to_cpu(info->max_size);
765                 }
766         }
767
768         /* xattrs */
769         /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
770         if ((ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL))  &&
771             le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) {
772                 if (ci->i_xattrs.blob)
773                         ceph_buffer_put(ci->i_xattrs.blob);
774                 ci->i_xattrs.blob = xattr_blob;
775                 if (xattr_blob)
776                         memcpy(ci->i_xattrs.blob->vec.iov_base,
777                                iinfo->xattr_data, iinfo->xattr_len);
778                 ci->i_xattrs.version = le64_to_cpu(info->xattr_version);
779                 ceph_forget_all_cached_acls(inode);
780                 xattr_blob = NULL;
781         }
782
783         inode->i_mapping->a_ops = &ceph_aops;
784         inode->i_mapping->backing_dev_info =
785                 &ceph_sb_to_client(inode->i_sb)->backing_dev_info;
786
787         switch (inode->i_mode & S_IFMT) {
788         case S_IFIFO:
789         case S_IFBLK:
790         case S_IFCHR:
791         case S_IFSOCK:
792                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
793                 inode->i_op = &ceph_file_iops;
794                 break;
795         case S_IFREG:
796                 inode->i_op = &ceph_file_iops;
797                 inode->i_fop = &ceph_file_fops;
798                 break;
799         case S_IFLNK:
800                 inode->i_op = &ceph_symlink_iops;
801                 if (!ci->i_symlink) {
802                         u32 symlen = iinfo->symlink_len;
803                         char *sym;
804
805                         spin_unlock(&ci->i_ceph_lock);
806
807                         err = -EINVAL;
808                         if (WARN_ON(symlen != inode->i_size))
809                                 goto out;
810
811                         err = -ENOMEM;
812                         sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS);
813                         if (!sym)
814                                 goto out;
815
816                         spin_lock(&ci->i_ceph_lock);
817                         if (!ci->i_symlink)
818                                 ci->i_symlink = sym;
819                         else
820                                 kfree(sym); /* lost a race */
821                 }
822                 break;
823         case S_IFDIR:
824                 inode->i_op = &ceph_dir_iops;
825                 inode->i_fop = &ceph_dir_fops;
826
827                 ci->i_dir_layout = iinfo->dir_layout;
828
829                 ci->i_files = le64_to_cpu(info->files);
830                 ci->i_subdirs = le64_to_cpu(info->subdirs);
831                 ci->i_rbytes = le64_to_cpu(info->rbytes);
832                 ci->i_rfiles = le64_to_cpu(info->rfiles);
833                 ci->i_rsubdirs = le64_to_cpu(info->rsubdirs);
834                 ceph_decode_timespec(&ci->i_rctime, &info->rctime);
835                 break;
836         default:
837                 pr_err("fill_inode %llx.%llx BAD mode 0%o\n",
838                        ceph_vinop(inode), inode->i_mode);
839         }
840
841         /* set dir completion flag? */
842         if (S_ISDIR(inode->i_mode) &&
843             ci->i_files == 0 && ci->i_subdirs == 0 &&
844             ceph_snap(inode) == CEPH_NOSNAP &&
845             (le32_to_cpu(info->cap.caps) & CEPH_CAP_FILE_SHARED) &&
846             (issued & CEPH_CAP_FILE_EXCL) == 0 &&
847             !__ceph_dir_is_complete(ci)) {
848                 dout(" marking %p complete (empty)\n", inode);
849                 __ceph_dir_set_complete(ci, atomic_read(&ci->i_release_count),
850                                         ci->i_ordered_count);
851         }
852
853         /* were we issued a capability? */
854         if (info->cap.caps) {
855                 if (ceph_snap(inode) == CEPH_NOSNAP) {
856                         ceph_add_cap(inode, session,
857                                      le64_to_cpu(info->cap.cap_id),
858                                      cap_fmode,
859                                      le32_to_cpu(info->cap.caps),
860                                      le32_to_cpu(info->cap.wanted),
861                                      le32_to_cpu(info->cap.seq),
862                                      le32_to_cpu(info->cap.mseq),
863                                      le64_to_cpu(info->cap.realm),
864                                      info->cap.flags, &new_cap);
865                         wake = true;
866                 } else {
867                         dout(" %p got snap_caps %s\n", inode,
868                              ceph_cap_string(le32_to_cpu(info->cap.caps)));
869                         ci->i_snap_caps |= le32_to_cpu(info->cap.caps);
870                         if (cap_fmode >= 0)
871                                 __ceph_get_fmode(ci, cap_fmode);
872                 }
873         } else if (cap_fmode >= 0) {
874                 pr_warn("mds issued no caps on %llx.%llx\n",
875                            ceph_vinop(inode));
876                 __ceph_get_fmode(ci, cap_fmode);
877         }
878         spin_unlock(&ci->i_ceph_lock);
879
880         if (wake)
881                 wake_up_all(&ci->i_cap_wq);
882
883         /* queue truncate if we saw i_size decrease */
884         if (queue_trunc)
885                 ceph_queue_vmtruncate(inode);
886
887         /* populate frag tree */
888         if (S_ISDIR(inode->i_mode))
889                 ceph_fill_fragtree(inode, &info->fragtree, dirinfo);
890
891         /* update delegation info? */
892         if (dirinfo)
893                 ceph_fill_dirfrag(inode, dirinfo);
894
895         err = 0;
896 out:
897         if (new_cap)
898                 ceph_put_cap(mdsc, new_cap);
899         if (xattr_blob)
900                 ceph_buffer_put(xattr_blob);
901         return err;
902 }
903
904 /*
905  * caller should hold session s_mutex.
906  */
907 static void update_dentry_lease(struct dentry *dentry,
908                                 struct ceph_mds_reply_lease *lease,
909                                 struct ceph_mds_session *session,
910                                 unsigned long from_time)
911 {
912         struct ceph_dentry_info *di = ceph_dentry(dentry);
913         long unsigned duration = le32_to_cpu(lease->duration_ms);
914         long unsigned ttl = from_time + (duration * HZ) / 1000;
915         long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000;
916         struct inode *dir;
917
918         /* only track leases on regular dentries */
919         if (dentry->d_op != &ceph_dentry_ops)
920                 return;
921
922         spin_lock(&dentry->d_lock);
923         dout("update_dentry_lease %p duration %lu ms ttl %lu\n",
924              dentry, duration, ttl);
925
926         /* make lease_rdcache_gen match directory */
927         dir = dentry->d_parent->d_inode;
928         di->lease_shared_gen = ceph_inode(dir)->i_shared_gen;
929
930         if (duration == 0)
931                 goto out_unlock;
932
933         if (di->lease_gen == session->s_cap_gen &&
934             time_before(ttl, dentry->d_time))
935                 goto out_unlock;  /* we already have a newer lease. */
936
937         if (di->lease_session && di->lease_session != session)
938                 goto out_unlock;
939
940         ceph_dentry_lru_touch(dentry);
941
942         if (!di->lease_session)
943                 di->lease_session = ceph_get_mds_session(session);
944         di->lease_gen = session->s_cap_gen;
945         di->lease_seq = le32_to_cpu(lease->seq);
946         di->lease_renew_after = half_ttl;
947         di->lease_renew_from = 0;
948         dentry->d_time = ttl;
949 out_unlock:
950         spin_unlock(&dentry->d_lock);
951         return;
952 }
953
954 /*
955  * splice a dentry to an inode.
956  * caller must hold directory i_mutex for this to be safe.
957  *
958  * we will only rehash the resulting dentry if @prehash is
959  * true; @prehash will be set to false (for the benefit of
960  * the caller) if we fail.
961  */
962 static struct dentry *splice_dentry(struct dentry *dn, struct inode *in,
963                                     bool *prehash)
964 {
965         struct dentry *realdn;
966
967         BUG_ON(dn->d_inode);
968
969         /* dn must be unhashed */
970         if (!d_unhashed(dn))
971                 d_drop(dn);
972         realdn = d_materialise_unique(dn, in);
973         if (IS_ERR(realdn)) {
974                 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
975                        PTR_ERR(realdn), dn, in, ceph_vinop(in));
976                 if (prehash)
977                         *prehash = false; /* don't rehash on error */
978                 dn = realdn; /* note realdn contains the error */
979                 goto out;
980         } else if (realdn) {
981                 dout("dn %p (%d) spliced with %p (%d) "
982                      "inode %p ino %llx.%llx\n",
983                      dn, d_count(dn),
984                      realdn, d_count(realdn),
985                      realdn->d_inode, ceph_vinop(realdn->d_inode));
986                 dput(dn);
987                 dn = realdn;
988         } else {
989                 BUG_ON(!ceph_dentry(dn));
990                 dout("dn %p attached to %p ino %llx.%llx\n",
991                      dn, dn->d_inode, ceph_vinop(dn->d_inode));
992         }
993         if ((!prehash || *prehash) && d_unhashed(dn))
994                 d_rehash(dn);
995 out:
996         return dn;
997 }
998
999 /*
1000  * Incorporate results into the local cache.  This is either just
1001  * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
1002  * after a lookup).
1003  *
1004  * A reply may contain
1005  *         a directory inode along with a dentry.
1006  *  and/or a target inode
1007  *
1008  * Called with snap_rwsem (read).
1009  */
1010 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req,
1011                     struct ceph_mds_session *session)
1012 {
1013         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1014         struct inode *in = NULL;
1015         struct ceph_vino vino;
1016         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
1017         int err = 0;
1018
1019         dout("fill_trace %p is_dentry %d is_target %d\n", req,
1020              rinfo->head->is_dentry, rinfo->head->is_target);
1021
1022 #if 0
1023         /*
1024          * Debugging hook:
1025          *
1026          * If we resend completed ops to a recovering mds, we get no
1027          * trace.  Since that is very rare, pretend this is the case
1028          * to ensure the 'no trace' handlers in the callers behave.
1029          *
1030          * Fill in inodes unconditionally to avoid breaking cap
1031          * invariants.
1032          */
1033         if (rinfo->head->op & CEPH_MDS_OP_WRITE) {
1034                 pr_info("fill_trace faking empty trace on %lld %s\n",
1035                         req->r_tid, ceph_mds_op_name(rinfo->head->op));
1036                 if (rinfo->head->is_dentry) {
1037                         rinfo->head->is_dentry = 0;
1038                         err = fill_inode(req->r_locked_dir,
1039                                          &rinfo->diri, rinfo->dirfrag,
1040                                          session, req->r_request_started, -1);
1041                 }
1042                 if (rinfo->head->is_target) {
1043                         rinfo->head->is_target = 0;
1044                         ininfo = rinfo->targeti.in;
1045                         vino.ino = le64_to_cpu(ininfo->ino);
1046                         vino.snap = le64_to_cpu(ininfo->snapid);
1047                         in = ceph_get_inode(sb, vino);
1048                         err = fill_inode(in, &rinfo->targeti, NULL,
1049                                          session, req->r_request_started,
1050                                          req->r_fmode);
1051                         iput(in);
1052                 }
1053         }
1054 #endif
1055
1056         if (!rinfo->head->is_target && !rinfo->head->is_dentry) {
1057                 dout("fill_trace reply is empty!\n");
1058                 if (rinfo->head->result == 0 && req->r_locked_dir)
1059                         ceph_invalidate_dir_request(req);
1060                 return 0;
1061         }
1062
1063         if (rinfo->head->is_dentry) {
1064                 struct inode *dir = req->r_locked_dir;
1065
1066                 if (dir) {
1067                         err = fill_inode(dir, &rinfo->diri, rinfo->dirfrag,
1068                                          session, req->r_request_started, -1,
1069                                          &req->r_caps_reservation);
1070                         if (err < 0)
1071                                 goto done;
1072                 } else {
1073                         WARN_ON_ONCE(1);
1074                 }
1075
1076                 if (dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME) {
1077                         struct qstr dname;
1078                         struct dentry *dn, *parent;
1079
1080                         BUG_ON(!rinfo->head->is_target);
1081                         BUG_ON(req->r_dentry);
1082
1083                         parent = d_find_any_alias(dir);
1084                         BUG_ON(!parent);
1085
1086                         dname.name = rinfo->dname;
1087                         dname.len = rinfo->dname_len;
1088                         dname.hash = full_name_hash(dname.name, dname.len);
1089                         vino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1090                         vino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1091 retry_lookup:
1092                         dn = d_lookup(parent, &dname);
1093                         dout("d_lookup on parent=%p name=%.*s got %p\n",
1094                              parent, dname.len, dname.name, dn);
1095
1096                         if (!dn) {
1097                                 dn = d_alloc(parent, &dname);
1098                                 dout("d_alloc %p '%.*s' = %p\n", parent,
1099                                      dname.len, dname.name, dn);
1100                                 if (dn == NULL) {
1101                                         dput(parent);
1102                                         err = -ENOMEM;
1103                                         goto done;
1104                                 }
1105                                 err = ceph_init_dentry(dn);
1106                                 if (err < 0) {
1107                                         dput(dn);
1108                                         dput(parent);
1109                                         goto done;
1110                                 }
1111                         } else if (dn->d_inode &&
1112                                    (ceph_ino(dn->d_inode) != vino.ino ||
1113                                     ceph_snap(dn->d_inode) != vino.snap)) {
1114                                 dout(" dn %p points to wrong inode %p\n",
1115                                      dn, dn->d_inode);
1116                                 d_delete(dn);
1117                                 dput(dn);
1118                                 goto retry_lookup;
1119                         }
1120
1121                         req->r_dentry = dn;
1122                         dput(parent);
1123                 }
1124         }
1125
1126         if (rinfo->head->is_target) {
1127                 vino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1128                 vino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1129
1130                 in = ceph_get_inode(sb, vino);
1131                 if (IS_ERR(in)) {
1132                         err = PTR_ERR(in);
1133                         goto done;
1134                 }
1135                 req->r_target_inode = in;
1136
1137                 err = fill_inode(in, &rinfo->targeti, NULL,
1138                                 session, req->r_request_started,
1139                                 (!req->r_aborted && rinfo->head->result == 0) ?
1140                                 req->r_fmode : -1,
1141                                 &req->r_caps_reservation);
1142                 if (err < 0) {
1143                         pr_err("fill_inode badness %p %llx.%llx\n",
1144                                 in, ceph_vinop(in));
1145                         goto done;
1146                 }
1147         }
1148
1149         /*
1150          * ignore null lease/binding on snapdir ENOENT, or else we
1151          * will have trouble splicing in the virtual snapdir later
1152          */
1153         if (rinfo->head->is_dentry && !req->r_aborted &&
1154             req->r_locked_dir &&
1155             (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name,
1156                                                fsc->mount_options->snapdir_name,
1157                                                req->r_dentry->d_name.len))) {
1158                 /*
1159                  * lookup link rename   : null -> possibly existing inode
1160                  * mknod symlink mkdir  : null -> new inode
1161                  * unlink               : linked -> null
1162                  */
1163                 struct inode *dir = req->r_locked_dir;
1164                 struct dentry *dn = req->r_dentry;
1165                 bool have_dir_cap, have_lease;
1166
1167                 BUG_ON(!dn);
1168                 BUG_ON(!dir);
1169                 BUG_ON(dn->d_parent->d_inode != dir);
1170                 BUG_ON(ceph_ino(dir) !=
1171                        le64_to_cpu(rinfo->diri.in->ino));
1172                 BUG_ON(ceph_snap(dir) !=
1173                        le64_to_cpu(rinfo->diri.in->snapid));
1174
1175                 /* do we have a lease on the whole dir? */
1176                 have_dir_cap =
1177                         (le32_to_cpu(rinfo->diri.in->cap.caps) &
1178                          CEPH_CAP_FILE_SHARED);
1179
1180                 /* do we have a dn lease? */
1181                 have_lease = have_dir_cap ||
1182                         le32_to_cpu(rinfo->dlease->duration_ms);
1183                 if (!have_lease)
1184                         dout("fill_trace  no dentry lease or dir cap\n");
1185
1186                 /* rename? */
1187                 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) {
1188                         struct inode *olddir = req->r_old_dentry_dir;
1189                         BUG_ON(!olddir);
1190
1191                         dout(" src %p '%.*s' dst %p '%.*s'\n",
1192                              req->r_old_dentry,
1193                              req->r_old_dentry->d_name.len,
1194                              req->r_old_dentry->d_name.name,
1195                              dn, dn->d_name.len, dn->d_name.name);
1196                         dout("fill_trace doing d_move %p -> %p\n",
1197                              req->r_old_dentry, dn);
1198
1199                         d_move(req->r_old_dentry, dn);
1200                         dout(" src %p '%.*s' dst %p '%.*s'\n",
1201                              req->r_old_dentry,
1202                              req->r_old_dentry->d_name.len,
1203                              req->r_old_dentry->d_name.name,
1204                              dn, dn->d_name.len, dn->d_name.name);
1205
1206                         /* ensure target dentry is invalidated, despite
1207                            rehashing bug in vfs_rename_dir */
1208                         ceph_invalidate_dentry_lease(dn);
1209
1210                         /* d_move screws up sibling dentries' offsets */
1211                         ceph_dir_clear_ordered(dir);
1212                         ceph_dir_clear_ordered(olddir);
1213
1214                         dout("dn %p gets new offset %lld\n", req->r_old_dentry,
1215                              ceph_dentry(req->r_old_dentry)->offset);
1216
1217                         dn = req->r_old_dentry;  /* use old_dentry */
1218                 }
1219
1220                 /* null dentry? */
1221                 if (!rinfo->head->is_target) {
1222                         dout("fill_trace null dentry\n");
1223                         if (dn->d_inode) {
1224                                 ceph_dir_clear_ordered(dir);
1225                                 dout("d_delete %p\n", dn);
1226                                 d_delete(dn);
1227                         } else {
1228                                 dout("d_instantiate %p NULL\n", dn);
1229                                 d_instantiate(dn, NULL);
1230                                 if (have_lease && d_unhashed(dn))
1231                                         d_rehash(dn);
1232                                 update_dentry_lease(dn, rinfo->dlease,
1233                                                     session,
1234                                                     req->r_request_started);
1235                         }
1236                         goto done;
1237                 }
1238
1239                 /* attach proper inode */
1240                 if (!dn->d_inode) {
1241                         ceph_dir_clear_ordered(dir);
1242                         ihold(in);
1243                         dn = splice_dentry(dn, in, &have_lease);
1244                         if (IS_ERR(dn)) {
1245                                 err = PTR_ERR(dn);
1246                                 goto done;
1247                         }
1248                         req->r_dentry = dn;  /* may have spliced */
1249                 } else if (dn->d_inode && dn->d_inode != in) {
1250                         dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1251                              dn, dn->d_inode, ceph_vinop(dn->d_inode),
1252                              ceph_vinop(in));
1253                         have_lease = false;
1254                 }
1255
1256                 if (have_lease)
1257                         update_dentry_lease(dn, rinfo->dlease, session,
1258                                             req->r_request_started);
1259                 dout(" final dn %p\n", dn);
1260         } else if (!req->r_aborted &&
1261                    (req->r_op == CEPH_MDS_OP_LOOKUPSNAP ||
1262                     req->r_op == CEPH_MDS_OP_MKSNAP)) {
1263                 struct dentry *dn = req->r_dentry;
1264                 struct inode *dir = req->r_locked_dir;
1265
1266                 /* fill out a snapdir LOOKUPSNAP dentry */
1267                 BUG_ON(!dn);
1268                 BUG_ON(!dir);
1269                 BUG_ON(ceph_snap(dir) != CEPH_SNAPDIR);
1270                 dout(" linking snapped dir %p to dn %p\n", in, dn);
1271                 ceph_dir_clear_ordered(dir);
1272                 ihold(in);
1273                 dn = splice_dentry(dn, in, NULL);
1274                 if (IS_ERR(dn)) {
1275                         err = PTR_ERR(dn);
1276                         goto done;
1277                 }
1278                 req->r_dentry = dn;  /* may have spliced */
1279         }
1280 done:
1281         dout("fill_trace done err=%d\n", err);
1282         return err;
1283 }
1284
1285 /*
1286  * Prepopulate our cache with readdir results, leases, etc.
1287  */
1288 static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req,
1289                                            struct ceph_mds_session *session)
1290 {
1291         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1292         int i, err = 0;
1293
1294         for (i = 0; i < rinfo->dir_nr; i++) {
1295                 struct ceph_vino vino;
1296                 struct inode *in;
1297                 int rc;
1298
1299                 vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino);
1300                 vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid);
1301
1302                 in = ceph_get_inode(req->r_dentry->d_sb, vino);
1303                 if (IS_ERR(in)) {
1304                         err = PTR_ERR(in);
1305                         dout("new_inode badness got %d\n", err);
1306                         continue;
1307                 }
1308                 rc = fill_inode(in, &rinfo->dir_in[i], NULL, session,
1309                                 req->r_request_started, -1,
1310                                 &req->r_caps_reservation);
1311                 if (rc < 0) {
1312                         pr_err("fill_inode badness on %p got %d\n", in, rc);
1313                         err = rc;
1314                         continue;
1315                 }
1316         }
1317
1318         return err;
1319 }
1320
1321 int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1322                              struct ceph_mds_session *session)
1323 {
1324         struct dentry *parent = req->r_dentry;
1325         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1326         struct qstr dname;
1327         struct dentry *dn;
1328         struct inode *in;
1329         int err = 0, ret, i;
1330         struct inode *snapdir = NULL;
1331         struct ceph_mds_request_head *rhead = req->r_request->front.iov_base;
1332         struct ceph_dentry_info *di;
1333         u64 r_readdir_offset = req->r_readdir_offset;
1334         u32 frag = le32_to_cpu(rhead->args.readdir.frag);
1335
1336         if (rinfo->dir_dir &&
1337             le32_to_cpu(rinfo->dir_dir->frag) != frag) {
1338                 dout("readdir_prepopulate got new frag %x -> %x\n",
1339                      frag, le32_to_cpu(rinfo->dir_dir->frag));
1340                 frag = le32_to_cpu(rinfo->dir_dir->frag);
1341                 if (ceph_frag_is_leftmost(frag))
1342                         r_readdir_offset = 2;
1343                 else
1344                         r_readdir_offset = 0;
1345         }
1346
1347         if (req->r_aborted)
1348                 return readdir_prepopulate_inodes_only(req, session);
1349
1350         if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
1351                 snapdir = ceph_get_snapdir(parent->d_inode);
1352                 parent = d_find_alias(snapdir);
1353                 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1354                      rinfo->dir_nr, parent);
1355         } else {
1356                 dout("readdir_prepopulate %d items under dn %p\n",
1357                      rinfo->dir_nr, parent);
1358                 if (rinfo->dir_dir)
1359                         ceph_fill_dirfrag(parent->d_inode, rinfo->dir_dir);
1360         }
1361
1362         /* FIXME: release caps/leases if error occurs */
1363         for (i = 0; i < rinfo->dir_nr; i++) {
1364                 struct ceph_vino vino;
1365
1366                 dname.name = rinfo->dir_dname[i];
1367                 dname.len = rinfo->dir_dname_len[i];
1368                 dname.hash = full_name_hash(dname.name, dname.len);
1369
1370                 vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino);
1371                 vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid);
1372
1373 retry_lookup:
1374                 dn = d_lookup(parent, &dname);
1375                 dout("d_lookup on parent=%p name=%.*s got %p\n",
1376                      parent, dname.len, dname.name, dn);
1377
1378                 if (!dn) {
1379                         dn = d_alloc(parent, &dname);
1380                         dout("d_alloc %p '%.*s' = %p\n", parent,
1381                              dname.len, dname.name, dn);
1382                         if (dn == NULL) {
1383                                 dout("d_alloc badness\n");
1384                                 err = -ENOMEM;
1385                                 goto out;
1386                         }
1387                         ret = ceph_init_dentry(dn);
1388                         if (ret < 0) {
1389                                 dput(dn);
1390                                 err = ret;
1391                                 goto out;
1392                         }
1393                 } else if (dn->d_inode &&
1394                            (ceph_ino(dn->d_inode) != vino.ino ||
1395                             ceph_snap(dn->d_inode) != vino.snap)) {
1396                         dout(" dn %p points to wrong inode %p\n",
1397                              dn, dn->d_inode);
1398                         d_delete(dn);
1399                         dput(dn);
1400                         goto retry_lookup;
1401                 } else {
1402                         /* reorder parent's d_subdirs */
1403                         spin_lock(&parent->d_lock);
1404                         spin_lock_nested(&dn->d_lock, DENTRY_D_LOCK_NESTED);
1405                         list_move(&dn->d_u.d_child, &parent->d_subdirs);
1406                         spin_unlock(&dn->d_lock);
1407                         spin_unlock(&parent->d_lock);
1408                 }
1409
1410                 /* inode */
1411                 if (dn->d_inode) {
1412                         in = dn->d_inode;
1413                 } else {
1414                         in = ceph_get_inode(parent->d_sb, vino);
1415                         if (IS_ERR(in)) {
1416                                 dout("new_inode badness\n");
1417                                 d_drop(dn);
1418                                 dput(dn);
1419                                 err = PTR_ERR(in);
1420                                 goto out;
1421                         }
1422                 }
1423
1424                 if (fill_inode(in, &rinfo->dir_in[i], NULL, session,
1425                                req->r_request_started, -1,
1426                                &req->r_caps_reservation) < 0) {
1427                         pr_err("fill_inode badness on %p\n", in);
1428                         if (!dn->d_inode)
1429                                 iput(in);
1430                         d_drop(dn);
1431                         goto next_item;
1432                 }
1433
1434                 if (!dn->d_inode) {
1435                         dn = splice_dentry(dn, in, NULL);
1436                         if (IS_ERR(dn)) {
1437                                 err = PTR_ERR(dn);
1438                                 dn = NULL;
1439                                 goto next_item;
1440                         }
1441                 }
1442
1443                 di = dn->d_fsdata;
1444                 di->offset = ceph_make_fpos(frag, i + r_readdir_offset);
1445
1446                 update_dentry_lease(dn, rinfo->dir_dlease[i],
1447                                     req->r_session,
1448                                     req->r_request_started);
1449 next_item:
1450                 if (dn)
1451                         dput(dn);
1452         }
1453         if (err == 0)
1454                 req->r_did_prepopulate = true;
1455
1456 out:
1457         if (snapdir) {
1458                 iput(snapdir);
1459                 dput(parent);
1460         }
1461         dout("readdir_prepopulate done\n");
1462         return err;
1463 }
1464
1465 int ceph_inode_set_size(struct inode *inode, loff_t size)
1466 {
1467         struct ceph_inode_info *ci = ceph_inode(inode);
1468         int ret = 0;
1469
1470         spin_lock(&ci->i_ceph_lock);
1471         dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size);
1472         inode->i_size = size;
1473         inode->i_blocks = (size + (1 << 9) - 1) >> 9;
1474
1475         /* tell the MDS if we are approaching max_size */
1476         if ((size << 1) >= ci->i_max_size &&
1477             (ci->i_reported_size << 1) < ci->i_max_size)
1478                 ret = 1;
1479
1480         spin_unlock(&ci->i_ceph_lock);
1481         return ret;
1482 }
1483
1484 /*
1485  * Write back inode data in a worker thread.  (This can't be done
1486  * in the message handler context.)
1487  */
1488 void ceph_queue_writeback(struct inode *inode)
1489 {
1490         ihold(inode);
1491         if (queue_work(ceph_inode_to_client(inode)->wb_wq,
1492                        &ceph_inode(inode)->i_wb_work)) {
1493                 dout("ceph_queue_writeback %p\n", inode);
1494         } else {
1495                 dout("ceph_queue_writeback %p failed\n", inode);
1496                 iput(inode);
1497         }
1498 }
1499
1500 static void ceph_writeback_work(struct work_struct *work)
1501 {
1502         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1503                                                   i_wb_work);
1504         struct inode *inode = &ci->vfs_inode;
1505
1506         dout("writeback %p\n", inode);
1507         filemap_fdatawrite(&inode->i_data);
1508         iput(inode);
1509 }
1510
1511 /*
1512  * queue an async invalidation
1513  */
1514 void ceph_queue_invalidate(struct inode *inode)
1515 {
1516         ihold(inode);
1517         if (queue_work(ceph_inode_to_client(inode)->pg_inv_wq,
1518                        &ceph_inode(inode)->i_pg_inv_work)) {
1519                 dout("ceph_queue_invalidate %p\n", inode);
1520         } else {
1521                 dout("ceph_queue_invalidate %p failed\n", inode);
1522                 iput(inode);
1523         }
1524 }
1525
1526 /*
1527  * Invalidate inode pages in a worker thread.  (This can't be done
1528  * in the message handler context.)
1529  */
1530 static void ceph_invalidate_work(struct work_struct *work)
1531 {
1532         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1533                                                   i_pg_inv_work);
1534         struct inode *inode = &ci->vfs_inode;
1535         u32 orig_gen;
1536         int check = 0;
1537
1538         mutex_lock(&ci->i_truncate_mutex);
1539         spin_lock(&ci->i_ceph_lock);
1540         dout("invalidate_pages %p gen %d revoking %d\n", inode,
1541              ci->i_rdcache_gen, ci->i_rdcache_revoking);
1542         if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
1543                 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1544                         check = 1;
1545                 spin_unlock(&ci->i_ceph_lock);
1546                 mutex_unlock(&ci->i_truncate_mutex);
1547                 goto out;
1548         }
1549         orig_gen = ci->i_rdcache_gen;
1550         spin_unlock(&ci->i_ceph_lock);
1551
1552         truncate_pagecache(inode, 0);
1553
1554         spin_lock(&ci->i_ceph_lock);
1555         if (orig_gen == ci->i_rdcache_gen &&
1556             orig_gen == ci->i_rdcache_revoking) {
1557                 dout("invalidate_pages %p gen %d successful\n", inode,
1558                      ci->i_rdcache_gen);
1559                 ci->i_rdcache_revoking--;
1560                 check = 1;
1561         } else {
1562                 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1563                      inode, orig_gen, ci->i_rdcache_gen,
1564                      ci->i_rdcache_revoking);
1565                 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1566                         check = 1;
1567         }
1568         spin_unlock(&ci->i_ceph_lock);
1569         mutex_unlock(&ci->i_truncate_mutex);
1570 out:
1571         if (check)
1572                 ceph_check_caps(ci, 0, NULL);
1573         iput(inode);
1574 }
1575
1576
1577 /*
1578  * called by trunc_wq;
1579  *
1580  * We also truncate in a separate thread as well.
1581  */
1582 static void ceph_vmtruncate_work(struct work_struct *work)
1583 {
1584         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1585                                                   i_vmtruncate_work);
1586         struct inode *inode = &ci->vfs_inode;
1587
1588         dout("vmtruncate_work %p\n", inode);
1589         __ceph_do_pending_vmtruncate(inode);
1590         iput(inode);
1591 }
1592
1593 /*
1594  * Queue an async vmtruncate.  If we fail to queue work, we will handle
1595  * the truncation the next time we call __ceph_do_pending_vmtruncate.
1596  */
1597 void ceph_queue_vmtruncate(struct inode *inode)
1598 {
1599         struct ceph_inode_info *ci = ceph_inode(inode);
1600
1601         ihold(inode);
1602
1603         if (queue_work(ceph_sb_to_client(inode->i_sb)->trunc_wq,
1604                        &ci->i_vmtruncate_work)) {
1605                 dout("ceph_queue_vmtruncate %p\n", inode);
1606         } else {
1607                 dout("ceph_queue_vmtruncate %p failed, pending=%d\n",
1608                      inode, ci->i_truncate_pending);
1609                 iput(inode);
1610         }
1611 }
1612
1613 /*
1614  * Make sure any pending truncation is applied before doing anything
1615  * that may depend on it.
1616  */
1617 void __ceph_do_pending_vmtruncate(struct inode *inode)
1618 {
1619         struct ceph_inode_info *ci = ceph_inode(inode);
1620         u64 to;
1621         int wrbuffer_refs, finish = 0;
1622
1623         mutex_lock(&ci->i_truncate_mutex);
1624 retry:
1625         spin_lock(&ci->i_ceph_lock);
1626         if (ci->i_truncate_pending == 0) {
1627                 dout("__do_pending_vmtruncate %p none pending\n", inode);
1628                 spin_unlock(&ci->i_ceph_lock);
1629                 mutex_unlock(&ci->i_truncate_mutex);
1630                 return;
1631         }
1632
1633         /*
1634          * make sure any dirty snapped pages are flushed before we
1635          * possibly truncate them.. so write AND block!
1636          */
1637         if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) {
1638                 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1639                      inode);
1640                 spin_unlock(&ci->i_ceph_lock);
1641                 filemap_write_and_wait_range(&inode->i_data, 0,
1642                                              inode->i_sb->s_maxbytes);
1643                 goto retry;
1644         }
1645
1646         /* there should be no reader or writer */
1647         WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref);
1648
1649         to = ci->i_truncate_size;
1650         wrbuffer_refs = ci->i_wrbuffer_ref;
1651         dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode,
1652              ci->i_truncate_pending, to);
1653         spin_unlock(&ci->i_ceph_lock);
1654
1655         truncate_pagecache(inode, to);
1656
1657         spin_lock(&ci->i_ceph_lock);
1658         if (to == ci->i_truncate_size) {
1659                 ci->i_truncate_pending = 0;
1660                 finish = 1;
1661         }
1662         spin_unlock(&ci->i_ceph_lock);
1663         if (!finish)
1664                 goto retry;
1665
1666         mutex_unlock(&ci->i_truncate_mutex);
1667
1668         if (wrbuffer_refs == 0)
1669                 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
1670
1671         wake_up_all(&ci->i_cap_wq);
1672 }
1673
1674 /*
1675  * symlinks
1676  */
1677 static void *ceph_sym_follow_link(struct dentry *dentry, struct nameidata *nd)
1678 {
1679         struct ceph_inode_info *ci = ceph_inode(dentry->d_inode);
1680         nd_set_link(nd, ci->i_symlink);
1681         return NULL;
1682 }
1683
1684 static const struct inode_operations ceph_symlink_iops = {
1685         .readlink = generic_readlink,
1686         .follow_link = ceph_sym_follow_link,
1687         .setattr = ceph_setattr,
1688         .getattr = ceph_getattr,
1689         .setxattr = ceph_setxattr,
1690         .getxattr = ceph_getxattr,
1691         .listxattr = ceph_listxattr,
1692         .removexattr = ceph_removexattr,
1693 };
1694
1695 /*
1696  * setattr
1697  */
1698 int ceph_setattr(struct dentry *dentry, struct iattr *attr)
1699 {
1700         struct inode *inode = dentry->d_inode;
1701         struct ceph_inode_info *ci = ceph_inode(inode);
1702         const unsigned int ia_valid = attr->ia_valid;
1703         struct ceph_mds_request *req;
1704         struct ceph_mds_client *mdsc = ceph_sb_to_client(dentry->d_sb)->mdsc;
1705         int issued;
1706         int release = 0, dirtied = 0;
1707         int mask = 0;
1708         int err = 0;
1709         int inode_dirty_flags = 0;
1710
1711         if (ceph_snap(inode) != CEPH_NOSNAP)
1712                 return -EROFS;
1713
1714         err = inode_change_ok(inode, attr);
1715         if (err != 0)
1716                 return err;
1717
1718         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR,
1719                                        USE_AUTH_MDS);
1720         if (IS_ERR(req))
1721                 return PTR_ERR(req);
1722
1723         spin_lock(&ci->i_ceph_lock);
1724         issued = __ceph_caps_issued(ci, NULL);
1725         dout("setattr %p issued %s\n", inode, ceph_cap_string(issued));
1726
1727         if (ia_valid & ATTR_UID) {
1728                 dout("setattr %p uid %d -> %d\n", inode,
1729                      from_kuid(&init_user_ns, inode->i_uid),
1730                      from_kuid(&init_user_ns, attr->ia_uid));
1731                 if (issued & CEPH_CAP_AUTH_EXCL) {
1732                         inode->i_uid = attr->ia_uid;
1733                         dirtied |= CEPH_CAP_AUTH_EXCL;
1734                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1735                            !uid_eq(attr->ia_uid, inode->i_uid)) {
1736                         req->r_args.setattr.uid = cpu_to_le32(
1737                                 from_kuid(&init_user_ns, attr->ia_uid));
1738                         mask |= CEPH_SETATTR_UID;
1739                         release |= CEPH_CAP_AUTH_SHARED;
1740                 }
1741         }
1742         if (ia_valid & ATTR_GID) {
1743                 dout("setattr %p gid %d -> %d\n", inode,
1744                      from_kgid(&init_user_ns, inode->i_gid),
1745                      from_kgid(&init_user_ns, attr->ia_gid));
1746                 if (issued & CEPH_CAP_AUTH_EXCL) {
1747                         inode->i_gid = attr->ia_gid;
1748                         dirtied |= CEPH_CAP_AUTH_EXCL;
1749                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1750                            !gid_eq(attr->ia_gid, inode->i_gid)) {
1751                         req->r_args.setattr.gid = cpu_to_le32(
1752                                 from_kgid(&init_user_ns, attr->ia_gid));
1753                         mask |= CEPH_SETATTR_GID;
1754                         release |= CEPH_CAP_AUTH_SHARED;
1755                 }
1756         }
1757         if (ia_valid & ATTR_MODE) {
1758                 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode,
1759                      attr->ia_mode);
1760                 if (issued & CEPH_CAP_AUTH_EXCL) {
1761                         inode->i_mode = attr->ia_mode;
1762                         dirtied |= CEPH_CAP_AUTH_EXCL;
1763                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1764                            attr->ia_mode != inode->i_mode) {
1765                         inode->i_mode = attr->ia_mode;
1766                         req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode);
1767                         mask |= CEPH_SETATTR_MODE;
1768                         release |= CEPH_CAP_AUTH_SHARED;
1769                 }
1770         }
1771
1772         if (ia_valid & ATTR_ATIME) {
1773                 dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode,
1774                      inode->i_atime.tv_sec, inode->i_atime.tv_nsec,
1775                      attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec);
1776                 if (issued & CEPH_CAP_FILE_EXCL) {
1777                         ci->i_time_warp_seq++;
1778                         inode->i_atime = attr->ia_atime;
1779                         dirtied |= CEPH_CAP_FILE_EXCL;
1780                 } else if ((issued & CEPH_CAP_FILE_WR) &&
1781                            timespec_compare(&inode->i_atime,
1782                                             &attr->ia_atime) < 0) {
1783                         inode->i_atime = attr->ia_atime;
1784                         dirtied |= CEPH_CAP_FILE_WR;
1785                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1786                            !timespec_equal(&inode->i_atime, &attr->ia_atime)) {
1787                         ceph_encode_timespec(&req->r_args.setattr.atime,
1788                                              &attr->ia_atime);
1789                         mask |= CEPH_SETATTR_ATIME;
1790                         release |= CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_RD |
1791                                 CEPH_CAP_FILE_WR;
1792                 }
1793         }
1794         if (ia_valid & ATTR_MTIME) {
1795                 dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode,
1796                      inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
1797                      attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec);
1798                 if (issued & CEPH_CAP_FILE_EXCL) {
1799                         ci->i_time_warp_seq++;
1800                         inode->i_mtime = attr->ia_mtime;
1801                         dirtied |= CEPH_CAP_FILE_EXCL;
1802                 } else if ((issued & CEPH_CAP_FILE_WR) &&
1803                            timespec_compare(&inode->i_mtime,
1804                                             &attr->ia_mtime) < 0) {
1805                         inode->i_mtime = attr->ia_mtime;
1806                         dirtied |= CEPH_CAP_FILE_WR;
1807                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1808                            !timespec_equal(&inode->i_mtime, &attr->ia_mtime)) {
1809                         ceph_encode_timespec(&req->r_args.setattr.mtime,
1810                                              &attr->ia_mtime);
1811                         mask |= CEPH_SETATTR_MTIME;
1812                         release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
1813                                 CEPH_CAP_FILE_WR;
1814                 }
1815         }
1816         if (ia_valid & ATTR_SIZE) {
1817                 dout("setattr %p size %lld -> %lld\n", inode,
1818                      inode->i_size, attr->ia_size);
1819                 if ((issued & CEPH_CAP_FILE_EXCL) &&
1820                     attr->ia_size > inode->i_size) {
1821                         inode->i_size = attr->ia_size;
1822                         inode->i_blocks =
1823                                 (attr->ia_size + (1 << 9) - 1) >> 9;
1824                         inode->i_ctime = attr->ia_ctime;
1825                         ci->i_reported_size = attr->ia_size;
1826                         dirtied |= CEPH_CAP_FILE_EXCL;
1827                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1828                            attr->ia_size != inode->i_size) {
1829                         req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
1830                         req->r_args.setattr.old_size =
1831                                 cpu_to_le64(inode->i_size);
1832                         mask |= CEPH_SETATTR_SIZE;
1833                         release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
1834                                 CEPH_CAP_FILE_WR;
1835                 }
1836         }
1837
1838         /* these do nothing */
1839         if (ia_valid & ATTR_CTIME) {
1840                 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME|
1841                                          ATTR_MODE|ATTR_UID|ATTR_GID)) == 0;
1842                 dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode,
1843                      inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
1844                      attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec,
1845                      only ? "ctime only" : "ignored");
1846                 inode->i_ctime = attr->ia_ctime;
1847                 if (only) {
1848                         /*
1849                          * if kernel wants to dirty ctime but nothing else,
1850                          * we need to choose a cap to dirty under, or do
1851                          * a almost-no-op setattr
1852                          */
1853                         if (issued & CEPH_CAP_AUTH_EXCL)
1854                                 dirtied |= CEPH_CAP_AUTH_EXCL;
1855                         else if (issued & CEPH_CAP_FILE_EXCL)
1856                                 dirtied |= CEPH_CAP_FILE_EXCL;
1857                         else if (issued & CEPH_CAP_XATTR_EXCL)
1858                                 dirtied |= CEPH_CAP_XATTR_EXCL;
1859                         else
1860                                 mask |= CEPH_SETATTR_CTIME;
1861                 }
1862         }
1863         if (ia_valid & ATTR_FILE)
1864                 dout("setattr %p ATTR_FILE ... hrm!\n", inode);
1865
1866         if (dirtied) {
1867                 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied);
1868                 inode->i_ctime = CURRENT_TIME;
1869         }
1870
1871         release &= issued;
1872         spin_unlock(&ci->i_ceph_lock);
1873
1874         if (inode_dirty_flags)
1875                 __mark_inode_dirty(inode, inode_dirty_flags);
1876
1877         if (ia_valid & ATTR_MODE) {
1878                 err = posix_acl_chmod(inode, attr->ia_mode);
1879                 if (err)
1880                         goto out_put;
1881         }
1882
1883         if (mask) {
1884                 req->r_inode = inode;
1885                 ihold(inode);
1886                 req->r_inode_drop = release;
1887                 req->r_args.setattr.mask = cpu_to_le32(mask);
1888                 req->r_num_caps = 1;
1889                 err = ceph_mdsc_do_request(mdsc, NULL, req);
1890         }
1891         dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err,
1892              ceph_cap_string(dirtied), mask);
1893
1894         ceph_mdsc_put_request(req);
1895         if (mask & CEPH_SETATTR_SIZE)
1896                 __ceph_do_pending_vmtruncate(inode);
1897         return err;
1898 out_put:
1899         ceph_mdsc_put_request(req);
1900         return err;
1901 }
1902
1903 /*
1904  * Verify that we have a lease on the given mask.  If not,
1905  * do a getattr against an mds.
1906  */
1907 int ceph_do_getattr(struct inode *inode, int mask, bool force)
1908 {
1909         struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
1910         struct ceph_mds_client *mdsc = fsc->mdsc;
1911         struct ceph_mds_request *req;
1912         int err;
1913
1914         if (ceph_snap(inode) == CEPH_SNAPDIR) {
1915                 dout("do_getattr inode %p SNAPDIR\n", inode);
1916                 return 0;
1917         }
1918
1919         dout("do_getattr inode %p mask %s mode 0%o\n", inode, ceph_cap_string(mask), inode->i_mode);
1920         if (!force && ceph_caps_issued_mask(ceph_inode(inode), mask, 1))
1921                 return 0;
1922
1923         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
1924         if (IS_ERR(req))
1925                 return PTR_ERR(req);
1926         req->r_inode = inode;
1927         ihold(inode);
1928         req->r_num_caps = 1;
1929         req->r_args.getattr.mask = cpu_to_le32(mask);
1930         err = ceph_mdsc_do_request(mdsc, NULL, req);
1931         ceph_mdsc_put_request(req);
1932         dout("do_getattr result=%d\n", err);
1933         return err;
1934 }
1935
1936
1937 /*
1938  * Check inode permissions.  We verify we have a valid value for
1939  * the AUTH cap, then call the generic handler.
1940  */
1941 int ceph_permission(struct inode *inode, int mask)
1942 {
1943         int err;
1944
1945         if (mask & MAY_NOT_BLOCK)
1946                 return -ECHILD;
1947
1948         err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED, false);
1949
1950         if (!err)
1951                 err = generic_permission(inode, mask);
1952         return err;
1953 }
1954
1955 /*
1956  * Get all attributes.  Hopefully somedata we'll have a statlite()
1957  * and can limit the fields we require to be accurate.
1958  */
1959 int ceph_getattr(struct vfsmount *mnt, struct dentry *dentry,
1960                  struct kstat *stat)
1961 {
1962         struct inode *inode = dentry->d_inode;
1963         struct ceph_inode_info *ci = ceph_inode(inode);
1964         int err;
1965
1966         err = ceph_do_getattr(inode, CEPH_STAT_CAP_INODE_ALL, false);
1967         if (!err) {
1968                 generic_fillattr(inode, stat);
1969                 stat->ino = ceph_translate_ino(inode->i_sb, inode->i_ino);
1970                 if (ceph_snap(inode) != CEPH_NOSNAP)
1971                         stat->dev = ceph_snap(inode);
1972                 else
1973                         stat->dev = 0;
1974                 if (S_ISDIR(inode->i_mode)) {
1975                         if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb),
1976                                                 RBYTES))
1977                                 stat->size = ci->i_rbytes;
1978                         else
1979                                 stat->size = ci->i_files + ci->i_subdirs;
1980                         stat->blocks = 0;
1981                         stat->blksize = 65536;
1982                 }
1983         }
1984         return err;
1985 }