]> git.openfabrics.org - ~shefty/rdma-dev.git/blob - fs/ceph/inode.c
a7db56f1523b7634ce4a6da8dde1ae8a75874e6b
[~shefty/rdma-dev.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/pagevec.h>
13
14 #include "super.h"
15 #include "mds_client.h"
16 #include <linux/ceph/decode.h>
17
18 /*
19  * Ceph inode operations
20  *
21  * Implement basic inode helpers (get, alloc) and inode ops (getattr,
22  * setattr, etc.), xattr helpers, and helpers for assimilating
23  * metadata returned by the MDS into our cache.
24  *
25  * Also define helpers for doing asynchronous writeback, invalidation,
26  * and truncation for the benefit of those who can't afford to block
27  * (typically because they are in the message handler path).
28  */
29
30 static const struct inode_operations ceph_symlink_iops;
31
32 static void ceph_invalidate_work(struct work_struct *work);
33 static void ceph_writeback_work(struct work_struct *work);
34 static void ceph_vmtruncate_work(struct work_struct *work);
35
36 /*
37  * find or create an inode, given the ceph ino number
38  */
39 static int ceph_set_ino_cb(struct inode *inode, void *data)
40 {
41         ceph_inode(inode)->i_vino = *(struct ceph_vino *)data;
42         inode->i_ino = ceph_vino_to_ino(*(struct ceph_vino *)data);
43         return 0;
44 }
45
46 struct inode *ceph_get_inode(struct super_block *sb, struct ceph_vino vino)
47 {
48         struct inode *inode;
49         ino_t t = ceph_vino_to_ino(vino);
50
51         inode = iget5_locked(sb, t, ceph_ino_compare, ceph_set_ino_cb, &vino);
52         if (inode == NULL)
53                 return ERR_PTR(-ENOMEM);
54         if (inode->i_state & I_NEW) {
55                 dout("get_inode created new inode %p %llx.%llx ino %llx\n",
56                      inode, ceph_vinop(inode), (u64)inode->i_ino);
57                 unlock_new_inode(inode);
58         }
59
60         dout("get_inode on %lu=%llx.%llx got %p\n", inode->i_ino, vino.ino,
61              vino.snap, inode);
62         return inode;
63 }
64
65 /*
66  * get/constuct snapdir inode for a given directory
67  */
68 struct inode *ceph_get_snapdir(struct inode *parent)
69 {
70         struct ceph_vino vino = {
71                 .ino = ceph_ino(parent),
72                 .snap = CEPH_SNAPDIR,
73         };
74         struct inode *inode = ceph_get_inode(parent->i_sb, vino);
75         struct ceph_inode_info *ci = ceph_inode(inode);
76
77         BUG_ON(!S_ISDIR(parent->i_mode));
78         if (IS_ERR(inode))
79                 return inode;
80         inode->i_mode = parent->i_mode;
81         inode->i_uid = parent->i_uid;
82         inode->i_gid = parent->i_gid;
83         inode->i_op = &ceph_dir_iops;
84         inode->i_fop = &ceph_dir_fops;
85         ci->i_snap_caps = CEPH_CAP_PIN; /* so we can open */
86         ci->i_rbytes = 0;
87         return inode;
88 }
89
90 const struct inode_operations ceph_file_iops = {
91         .permission = ceph_permission,
92         .setattr = ceph_setattr,
93         .getattr = ceph_getattr,
94         .setxattr = ceph_setxattr,
95         .getxattr = ceph_getxattr,
96         .listxattr = ceph_listxattr,
97         .removexattr = ceph_removexattr,
98 };
99
100
101 /*
102  * We use a 'frag tree' to keep track of the MDS's directory fragments
103  * for a given inode (usually there is just a single fragment).  We
104  * need to know when a child frag is delegated to a new MDS, or when
105  * it is flagged as replicated, so we can direct our requests
106  * accordingly.
107  */
108
109 /*
110  * find/create a frag in the tree
111  */
112 static struct ceph_inode_frag *__get_or_create_frag(struct ceph_inode_info *ci,
113                                                     u32 f)
114 {
115         struct rb_node **p;
116         struct rb_node *parent = NULL;
117         struct ceph_inode_frag *frag;
118         int c;
119
120         p = &ci->i_fragtree.rb_node;
121         while (*p) {
122                 parent = *p;
123                 frag = rb_entry(parent, struct ceph_inode_frag, node);
124                 c = ceph_frag_compare(f, frag->frag);
125                 if (c < 0)
126                         p = &(*p)->rb_left;
127                 else if (c > 0)
128                         p = &(*p)->rb_right;
129                 else
130                         return frag;
131         }
132
133         frag = kmalloc(sizeof(*frag), GFP_NOFS);
134         if (!frag) {
135                 pr_err("__get_or_create_frag ENOMEM on %p %llx.%llx "
136                        "frag %x\n", &ci->vfs_inode,
137                        ceph_vinop(&ci->vfs_inode), f);
138                 return ERR_PTR(-ENOMEM);
139         }
140         frag->frag = f;
141         frag->split_by = 0;
142         frag->mds = -1;
143         frag->ndist = 0;
144
145         rb_link_node(&frag->node, parent, p);
146         rb_insert_color(&frag->node, &ci->i_fragtree);
147
148         dout("get_or_create_frag added %llx.%llx frag %x\n",
149              ceph_vinop(&ci->vfs_inode), f);
150         return frag;
151 }
152
153 /*
154  * find a specific frag @f
155  */
156 struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci, u32 f)
157 {
158         struct rb_node *n = ci->i_fragtree.rb_node;
159
160         while (n) {
161                 struct ceph_inode_frag *frag =
162                         rb_entry(n, struct ceph_inode_frag, node);
163                 int c = ceph_frag_compare(f, frag->frag);
164                 if (c < 0)
165                         n = n->rb_left;
166                 else if (c > 0)
167                         n = n->rb_right;
168                 else
169                         return frag;
170         }
171         return NULL;
172 }
173
174 /*
175  * Choose frag containing the given value @v.  If @pfrag is
176  * specified, copy the frag delegation info to the caller if
177  * it is present.
178  */
179 u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
180                      struct ceph_inode_frag *pfrag,
181                      int *found)
182 {
183         u32 t = ceph_frag_make(0, 0);
184         struct ceph_inode_frag *frag;
185         unsigned nway, i;
186         u32 n;
187
188         if (found)
189                 *found = 0;
190
191         mutex_lock(&ci->i_fragtree_mutex);
192         while (1) {
193                 WARN_ON(!ceph_frag_contains_value(t, v));
194                 frag = __ceph_find_frag(ci, t);
195                 if (!frag)
196                         break; /* t is a leaf */
197                 if (frag->split_by == 0) {
198                         if (pfrag)
199                                 memcpy(pfrag, frag, sizeof(*pfrag));
200                         if (found)
201                                 *found = 1;
202                         break;
203                 }
204
205                 /* choose child */
206                 nway = 1 << frag->split_by;
207                 dout("choose_frag(%x) %x splits by %d (%d ways)\n", v, t,
208                      frag->split_by, nway);
209                 for (i = 0; i < nway; i++) {
210                         n = ceph_frag_make_child(t, frag->split_by, i);
211                         if (ceph_frag_contains_value(n, v)) {
212                                 t = n;
213                                 break;
214                         }
215                 }
216                 BUG_ON(i == nway);
217         }
218         dout("choose_frag(%x) = %x\n", v, t);
219
220         mutex_unlock(&ci->i_fragtree_mutex);
221         return t;
222 }
223
224 /*
225  * Process dirfrag (delegation) info from the mds.  Include leaf
226  * fragment in tree ONLY if ndist > 0.  Otherwise, only
227  * branches/splits are included in i_fragtree)
228  */
229 static int ceph_fill_dirfrag(struct inode *inode,
230                              struct ceph_mds_reply_dirfrag *dirinfo)
231 {
232         struct ceph_inode_info *ci = ceph_inode(inode);
233         struct ceph_inode_frag *frag;
234         u32 id = le32_to_cpu(dirinfo->frag);
235         int mds = le32_to_cpu(dirinfo->auth);
236         int ndist = le32_to_cpu(dirinfo->ndist);
237         int i;
238         int err = 0;
239
240         mutex_lock(&ci->i_fragtree_mutex);
241         if (ndist == 0) {
242                 /* no delegation info needed. */
243                 frag = __ceph_find_frag(ci, id);
244                 if (!frag)
245                         goto out;
246                 if (frag->split_by == 0) {
247                         /* tree leaf, remove */
248                         dout("fill_dirfrag removed %llx.%llx frag %x"
249                              " (no ref)\n", ceph_vinop(inode), id);
250                         rb_erase(&frag->node, &ci->i_fragtree);
251                         kfree(frag);
252                 } else {
253                         /* tree branch, keep and clear */
254                         dout("fill_dirfrag cleared %llx.%llx frag %x"
255                              " referral\n", ceph_vinop(inode), id);
256                         frag->mds = -1;
257                         frag->ndist = 0;
258                 }
259                 goto out;
260         }
261
262
263         /* find/add this frag to store mds delegation info */
264         frag = __get_or_create_frag(ci, id);
265         if (IS_ERR(frag)) {
266                 /* this is not the end of the world; we can continue
267                    with bad/inaccurate delegation info */
268                 pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n",
269                        ceph_vinop(inode), le32_to_cpu(dirinfo->frag));
270                 err = -ENOMEM;
271                 goto out;
272         }
273
274         frag->mds = mds;
275         frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP);
276         for (i = 0; i < frag->ndist; i++)
277                 frag->dist[i] = le32_to_cpu(dirinfo->dist[i]);
278         dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n",
279              ceph_vinop(inode), frag->frag, frag->ndist);
280
281 out:
282         mutex_unlock(&ci->i_fragtree_mutex);
283         return err;
284 }
285
286
287 /*
288  * initialize a newly allocated inode.
289  */
290 struct inode *ceph_alloc_inode(struct super_block *sb)
291 {
292         struct ceph_inode_info *ci;
293         int i;
294
295         ci = kmem_cache_alloc(ceph_inode_cachep, GFP_NOFS);
296         if (!ci)
297                 return NULL;
298
299         dout("alloc_inode %p\n", &ci->vfs_inode);
300
301         ci->i_version = 0;
302         ci->i_time_warp_seq = 0;
303         ci->i_ceph_flags = 0;
304         ci->i_release_count = 0;
305         ci->i_symlink = NULL;
306
307         memset(&ci->i_dir_layout, 0, sizeof(ci->i_dir_layout));
308
309         ci->i_fragtree = RB_ROOT;
310         mutex_init(&ci->i_fragtree_mutex);
311
312         ci->i_xattrs.blob = NULL;
313         ci->i_xattrs.prealloc_blob = NULL;
314         ci->i_xattrs.dirty = false;
315         ci->i_xattrs.index = RB_ROOT;
316         ci->i_xattrs.count = 0;
317         ci->i_xattrs.names_size = 0;
318         ci->i_xattrs.vals_size = 0;
319         ci->i_xattrs.version = 0;
320         ci->i_xattrs.index_version = 0;
321
322         ci->i_caps = RB_ROOT;
323         ci->i_auth_cap = NULL;
324         ci->i_dirty_caps = 0;
325         ci->i_flushing_caps = 0;
326         INIT_LIST_HEAD(&ci->i_dirty_item);
327         INIT_LIST_HEAD(&ci->i_flushing_item);
328         ci->i_cap_flush_seq = 0;
329         ci->i_cap_flush_last_tid = 0;
330         memset(&ci->i_cap_flush_tid, 0, sizeof(ci->i_cap_flush_tid));
331         init_waitqueue_head(&ci->i_cap_wq);
332         ci->i_hold_caps_min = 0;
333         ci->i_hold_caps_max = 0;
334         INIT_LIST_HEAD(&ci->i_cap_delay_list);
335         ci->i_cap_exporting_mds = 0;
336         ci->i_cap_exporting_mseq = 0;
337         ci->i_cap_exporting_issued = 0;
338         INIT_LIST_HEAD(&ci->i_cap_snaps);
339         ci->i_head_snapc = NULL;
340         ci->i_snap_caps = 0;
341
342         for (i = 0; i < CEPH_FILE_MODE_NUM; i++)
343                 ci->i_nr_by_mode[i] = 0;
344
345         ci->i_truncate_seq = 0;
346         ci->i_truncate_size = 0;
347         ci->i_truncate_pending = 0;
348
349         ci->i_max_size = 0;
350         ci->i_reported_size = 0;
351         ci->i_wanted_max_size = 0;
352         ci->i_requested_max_size = 0;
353
354         ci->i_pin_ref = 0;
355         ci->i_rd_ref = 0;
356         ci->i_rdcache_ref = 0;
357         ci->i_wr_ref = 0;
358         ci->i_wb_ref = 0;
359         ci->i_wrbuffer_ref = 0;
360         ci->i_wrbuffer_ref_head = 0;
361         ci->i_shared_gen = 0;
362         ci->i_rdcache_gen = 0;
363         ci->i_rdcache_revoking = 0;
364
365         INIT_LIST_HEAD(&ci->i_unsafe_writes);
366         INIT_LIST_HEAD(&ci->i_unsafe_dirops);
367         spin_lock_init(&ci->i_unsafe_lock);
368
369         ci->i_snap_realm = NULL;
370         INIT_LIST_HEAD(&ci->i_snap_realm_item);
371         INIT_LIST_HEAD(&ci->i_snap_flush_item);
372
373         INIT_WORK(&ci->i_wb_work, ceph_writeback_work);
374         INIT_WORK(&ci->i_pg_inv_work, ceph_invalidate_work);
375
376         INIT_WORK(&ci->i_vmtruncate_work, ceph_vmtruncate_work);
377
378         return &ci->vfs_inode;
379 }
380
381 static void ceph_i_callback(struct rcu_head *head)
382 {
383         struct inode *inode = container_of(head, struct inode, i_rcu);
384         struct ceph_inode_info *ci = ceph_inode(inode);
385
386         INIT_LIST_HEAD(&inode->i_dentry);
387         kmem_cache_free(ceph_inode_cachep, ci);
388 }
389
390 void ceph_destroy_inode(struct inode *inode)
391 {
392         struct ceph_inode_info *ci = ceph_inode(inode);
393         struct ceph_inode_frag *frag;
394         struct rb_node *n;
395
396         dout("destroy_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode));
397
398         ceph_queue_caps_release(inode);
399
400         /*
401          * we may still have a snap_realm reference if there are stray
402          * caps in i_cap_exporting_issued or i_snap_caps.
403          */
404         if (ci->i_snap_realm) {
405                 struct ceph_mds_client *mdsc =
406                         ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc;
407                 struct ceph_snap_realm *realm = ci->i_snap_realm;
408
409                 dout(" dropping residual ref to snap realm %p\n", realm);
410                 spin_lock(&realm->inodes_with_caps_lock);
411                 list_del_init(&ci->i_snap_realm_item);
412                 spin_unlock(&realm->inodes_with_caps_lock);
413                 ceph_put_snap_realm(mdsc, realm);
414         }
415
416         kfree(ci->i_symlink);
417         while ((n = rb_first(&ci->i_fragtree)) != NULL) {
418                 frag = rb_entry(n, struct ceph_inode_frag, node);
419                 rb_erase(n, &ci->i_fragtree);
420                 kfree(frag);
421         }
422
423         __ceph_destroy_xattrs(ci);
424         if (ci->i_xattrs.blob)
425                 ceph_buffer_put(ci->i_xattrs.blob);
426         if (ci->i_xattrs.prealloc_blob)
427                 ceph_buffer_put(ci->i_xattrs.prealloc_blob);
428
429         call_rcu(&inode->i_rcu, ceph_i_callback);
430 }
431
432
433 /*
434  * Helpers to fill in size, ctime, mtime, and atime.  We have to be
435  * careful because either the client or MDS may have more up to date
436  * info, depending on which capabilities are held, and whether
437  * time_warp_seq or truncate_seq have increased.  (Ordinarily, mtime
438  * and size are monotonically increasing, except when utimes() or
439  * truncate() increments the corresponding _seq values.)
440  */
441 int ceph_fill_file_size(struct inode *inode, int issued,
442                         u32 truncate_seq, u64 truncate_size, u64 size)
443 {
444         struct ceph_inode_info *ci = ceph_inode(inode);
445         int queue_trunc = 0;
446
447         if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 ||
448             (truncate_seq == ci->i_truncate_seq && size > inode->i_size)) {
449                 dout("size %lld -> %llu\n", inode->i_size, size);
450                 inode->i_size = size;
451                 inode->i_blocks = (size + (1<<9) - 1) >> 9;
452                 ci->i_reported_size = size;
453                 if (truncate_seq != ci->i_truncate_seq) {
454                         dout("truncate_seq %u -> %u\n",
455                              ci->i_truncate_seq, truncate_seq);
456                         ci->i_truncate_seq = truncate_seq;
457                         /*
458                          * If we hold relevant caps, or in the case where we're
459                          * not the only client referencing this file and we
460                          * don't hold those caps, then we need to check whether
461                          * the file is either opened or mmaped
462                          */
463                         if ((issued & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_RD|
464                                        CEPH_CAP_FILE_WR|CEPH_CAP_FILE_BUFFER|
465                                        CEPH_CAP_FILE_EXCL|
466                                        CEPH_CAP_FILE_LAZYIO)) ||
467                             mapping_mapped(inode->i_mapping) ||
468                             __ceph_caps_file_wanted(ci)) {
469                                 ci->i_truncate_pending++;
470                                 queue_trunc = 1;
471                         }
472                 }
473         }
474         if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 &&
475             ci->i_truncate_size != truncate_size) {
476                 dout("truncate_size %lld -> %llu\n", ci->i_truncate_size,
477                      truncate_size);
478                 ci->i_truncate_size = truncate_size;
479         }
480         return queue_trunc;
481 }
482
483 void ceph_fill_file_time(struct inode *inode, int issued,
484                          u64 time_warp_seq, struct timespec *ctime,
485                          struct timespec *mtime, struct timespec *atime)
486 {
487         struct ceph_inode_info *ci = ceph_inode(inode);
488         int warn = 0;
489
490         if (issued & (CEPH_CAP_FILE_EXCL|
491                       CEPH_CAP_FILE_WR|
492                       CEPH_CAP_FILE_BUFFER|
493                       CEPH_CAP_AUTH_EXCL|
494                       CEPH_CAP_XATTR_EXCL)) {
495                 if (timespec_compare(ctime, &inode->i_ctime) > 0) {
496                         dout("ctime %ld.%09ld -> %ld.%09ld inc w/ cap\n",
497                              inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
498                              ctime->tv_sec, ctime->tv_nsec);
499                         inode->i_ctime = *ctime;
500                 }
501                 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) {
502                         /* the MDS did a utimes() */
503                         dout("mtime %ld.%09ld -> %ld.%09ld "
504                              "tw %d -> %d\n",
505                              inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
506                              mtime->tv_sec, mtime->tv_nsec,
507                              ci->i_time_warp_seq, (int)time_warp_seq);
508
509                         inode->i_mtime = *mtime;
510                         inode->i_atime = *atime;
511                         ci->i_time_warp_seq = time_warp_seq;
512                 } else if (time_warp_seq == ci->i_time_warp_seq) {
513                         /* nobody did utimes(); take the max */
514                         if (timespec_compare(mtime, &inode->i_mtime) > 0) {
515                                 dout("mtime %ld.%09ld -> %ld.%09ld inc\n",
516                                      inode->i_mtime.tv_sec,
517                                      inode->i_mtime.tv_nsec,
518                                      mtime->tv_sec, mtime->tv_nsec);
519                                 inode->i_mtime = *mtime;
520                         }
521                         if (timespec_compare(atime, &inode->i_atime) > 0) {
522                                 dout("atime %ld.%09ld -> %ld.%09ld inc\n",
523                                      inode->i_atime.tv_sec,
524                                      inode->i_atime.tv_nsec,
525                                      atime->tv_sec, atime->tv_nsec);
526                                 inode->i_atime = *atime;
527                         }
528                 } else if (issued & CEPH_CAP_FILE_EXCL) {
529                         /* we did a utimes(); ignore mds values */
530                 } else {
531                         warn = 1;
532                 }
533         } else {
534                 /* we have no write|excl caps; whatever the MDS says is true */
535                 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) {
536                         inode->i_ctime = *ctime;
537                         inode->i_mtime = *mtime;
538                         inode->i_atime = *atime;
539                         ci->i_time_warp_seq = time_warp_seq;
540                 } else {
541                         warn = 1;
542                 }
543         }
544         if (warn) /* time_warp_seq shouldn't go backwards */
545                 dout("%p mds time_warp_seq %llu < %u\n",
546                      inode, time_warp_seq, ci->i_time_warp_seq);
547 }
548
549 /*
550  * Populate an inode based on info from mds.  May be called on new or
551  * existing inodes.
552  */
553 static int fill_inode(struct inode *inode,
554                       struct ceph_mds_reply_info_in *iinfo,
555                       struct ceph_mds_reply_dirfrag *dirinfo,
556                       struct ceph_mds_session *session,
557                       unsigned long ttl_from, int cap_fmode,
558                       struct ceph_cap_reservation *caps_reservation)
559 {
560         struct ceph_mds_reply_inode *info = iinfo->in;
561         struct ceph_inode_info *ci = ceph_inode(inode);
562         int i;
563         int issued = 0, implemented;
564         int updating_inode = 0;
565         struct timespec mtime, atime, ctime;
566         u32 nsplits;
567         struct ceph_buffer *xattr_blob = NULL;
568         int err = 0;
569         int queue_trunc = 0;
570
571         dout("fill_inode %p ino %llx.%llx v %llu had %llu\n",
572              inode, ceph_vinop(inode), le64_to_cpu(info->version),
573              ci->i_version);
574
575         /*
576          * prealloc xattr data, if it looks like we'll need it.  only
577          * if len > 4 (meaning there are actually xattrs; the first 4
578          * bytes are the xattr count).
579          */
580         if (iinfo->xattr_len > 4) {
581                 xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS);
582                 if (!xattr_blob)
583                         pr_err("fill_inode ENOMEM xattr blob %d bytes\n",
584                                iinfo->xattr_len);
585         }
586
587         spin_lock(&inode->i_lock);
588
589         /*
590          * provided version will be odd if inode value is projected,
591          * even if stable.  skip the update if we have newer stable
592          * info (ours>=theirs, e.g. due to racing mds replies), unless
593          * we are getting projected (unstable) info (in which case the
594          * version is odd, and we want ours>theirs).
595          *   us   them
596          *   2    2     skip
597          *   3    2     skip
598          *   3    3     update
599          */
600         if (le64_to_cpu(info->version) > 0 &&
601             (ci->i_version & ~1) >= le64_to_cpu(info->version))
602                 goto no_change;
603         
604         updating_inode = 1;
605         issued = __ceph_caps_issued(ci, &implemented);
606         issued |= implemented | __ceph_caps_dirty(ci);
607
608         /* update inode */
609         ci->i_version = le64_to_cpu(info->version);
610         inode->i_version++;
611         inode->i_rdev = le32_to_cpu(info->rdev);
612
613         if ((issued & CEPH_CAP_AUTH_EXCL) == 0) {
614                 inode->i_mode = le32_to_cpu(info->mode);
615                 inode->i_uid = le32_to_cpu(info->uid);
616                 inode->i_gid = le32_to_cpu(info->gid);
617                 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
618                      inode->i_uid, inode->i_gid);
619         }
620
621         if ((issued & CEPH_CAP_LINK_EXCL) == 0)
622                 inode->i_nlink = le32_to_cpu(info->nlink);
623
624         /* be careful with mtime, atime, size */
625         ceph_decode_timespec(&atime, &info->atime);
626         ceph_decode_timespec(&mtime, &info->mtime);
627         ceph_decode_timespec(&ctime, &info->ctime);
628         queue_trunc = ceph_fill_file_size(inode, issued,
629                                           le32_to_cpu(info->truncate_seq),
630                                           le64_to_cpu(info->truncate_size),
631                                           le64_to_cpu(info->size));
632         ceph_fill_file_time(inode, issued,
633                             le32_to_cpu(info->time_warp_seq),
634                             &ctime, &mtime, &atime);
635
636         /* only update max_size on auth cap */
637         if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
638             ci->i_max_size != le64_to_cpu(info->max_size)) {
639                 dout("max_size %lld -> %llu\n", ci->i_max_size,
640                      le64_to_cpu(info->max_size));
641                 ci->i_max_size = le64_to_cpu(info->max_size);
642         }
643
644         ci->i_layout = info->layout;
645         inode->i_blkbits = fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1;
646
647         /* xattrs */
648         /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
649         if ((issued & CEPH_CAP_XATTR_EXCL) == 0 &&
650             le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) {
651                 if (ci->i_xattrs.blob)
652                         ceph_buffer_put(ci->i_xattrs.blob);
653                 ci->i_xattrs.blob = xattr_blob;
654                 if (xattr_blob)
655                         memcpy(ci->i_xattrs.blob->vec.iov_base,
656                                iinfo->xattr_data, iinfo->xattr_len);
657                 ci->i_xattrs.version = le64_to_cpu(info->xattr_version);
658                 xattr_blob = NULL;
659         }
660
661         inode->i_mapping->a_ops = &ceph_aops;
662         inode->i_mapping->backing_dev_info =
663                 &ceph_sb_to_client(inode->i_sb)->backing_dev_info;
664
665         switch (inode->i_mode & S_IFMT) {
666         case S_IFIFO:
667         case S_IFBLK:
668         case S_IFCHR:
669         case S_IFSOCK:
670                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
671                 inode->i_op = &ceph_file_iops;
672                 break;
673         case S_IFREG:
674                 inode->i_op = &ceph_file_iops;
675                 inode->i_fop = &ceph_file_fops;
676                 break;
677         case S_IFLNK:
678                 inode->i_op = &ceph_symlink_iops;
679                 if (!ci->i_symlink) {
680                         int symlen = iinfo->symlink_len;
681                         char *sym;
682
683                         BUG_ON(symlen != inode->i_size);
684                         spin_unlock(&inode->i_lock);
685
686                         err = -ENOMEM;
687                         sym = kmalloc(symlen+1, GFP_NOFS);
688                         if (!sym)
689                                 goto out;
690                         memcpy(sym, iinfo->symlink, symlen);
691                         sym[symlen] = 0;
692
693                         spin_lock(&inode->i_lock);
694                         if (!ci->i_symlink)
695                                 ci->i_symlink = sym;
696                         else
697                                 kfree(sym); /* lost a race */
698                 }
699                 break;
700         case S_IFDIR:
701                 inode->i_op = &ceph_dir_iops;
702                 inode->i_fop = &ceph_dir_fops;
703
704                 ci->i_dir_layout = iinfo->dir_layout;
705
706                 ci->i_files = le64_to_cpu(info->files);
707                 ci->i_subdirs = le64_to_cpu(info->subdirs);
708                 ci->i_rbytes = le64_to_cpu(info->rbytes);
709                 ci->i_rfiles = le64_to_cpu(info->rfiles);
710                 ci->i_rsubdirs = le64_to_cpu(info->rsubdirs);
711                 ceph_decode_timespec(&ci->i_rctime, &info->rctime);
712                 break;
713         default:
714                 pr_err("fill_inode %llx.%llx BAD mode 0%o\n",
715                        ceph_vinop(inode), inode->i_mode);
716         }
717
718 no_change:
719         spin_unlock(&inode->i_lock);
720
721         /* queue truncate if we saw i_size decrease */
722         if (queue_trunc)
723                 ceph_queue_vmtruncate(inode);
724
725         /* populate frag tree */
726         /* FIXME: move me up, if/when version reflects fragtree changes */
727         nsplits = le32_to_cpu(info->fragtree.nsplits);
728         mutex_lock(&ci->i_fragtree_mutex);
729         for (i = 0; i < nsplits; i++) {
730                 u32 id = le32_to_cpu(info->fragtree.splits[i].frag);
731                 struct ceph_inode_frag *frag = __get_or_create_frag(ci, id);
732
733                 if (IS_ERR(frag))
734                         continue;
735                 frag->split_by = le32_to_cpu(info->fragtree.splits[i].by);
736                 dout(" frag %x split by %d\n", frag->frag, frag->split_by);
737         }
738         mutex_unlock(&ci->i_fragtree_mutex);
739
740         /* were we issued a capability? */
741         if (info->cap.caps) {
742                 if (ceph_snap(inode) == CEPH_NOSNAP) {
743                         ceph_add_cap(inode, session,
744                                      le64_to_cpu(info->cap.cap_id),
745                                      cap_fmode,
746                                      le32_to_cpu(info->cap.caps),
747                                      le32_to_cpu(info->cap.wanted),
748                                      le32_to_cpu(info->cap.seq),
749                                      le32_to_cpu(info->cap.mseq),
750                                      le64_to_cpu(info->cap.realm),
751                                      info->cap.flags,
752                                      caps_reservation);
753                 } else {
754                         spin_lock(&inode->i_lock);
755                         dout(" %p got snap_caps %s\n", inode,
756                              ceph_cap_string(le32_to_cpu(info->cap.caps)));
757                         ci->i_snap_caps |= le32_to_cpu(info->cap.caps);
758                         if (cap_fmode >= 0)
759                                 __ceph_get_fmode(ci, cap_fmode);
760                         spin_unlock(&inode->i_lock);
761                 }
762         } else if (cap_fmode >= 0) {
763                 pr_warning("mds issued no caps on %llx.%llx\n",
764                            ceph_vinop(inode));
765                 __ceph_get_fmode(ci, cap_fmode);
766         }
767
768         /* set dir completion flag? */
769         if (S_ISDIR(inode->i_mode) &&
770             updating_inode &&                 /* didn't jump to no_change */
771             ci->i_files == 0 && ci->i_subdirs == 0 &&
772             ceph_snap(inode) == CEPH_NOSNAP &&
773             (le32_to_cpu(info->cap.caps) & CEPH_CAP_FILE_SHARED) &&
774             (issued & CEPH_CAP_FILE_EXCL) == 0 &&
775             (ci->i_ceph_flags & CEPH_I_COMPLETE) == 0) {
776                 dout(" marking %p complete (empty)\n", inode);
777                 /* ci->i_ceph_flags |= CEPH_I_COMPLETE; */
778                 ci->i_max_offset = 2;
779         }
780
781         /* update delegation info? */
782         if (dirinfo)
783                 ceph_fill_dirfrag(inode, dirinfo);
784
785         err = 0;
786
787 out:
788         if (xattr_blob)
789                 ceph_buffer_put(xattr_blob);
790         return err;
791 }
792
793 /*
794  * caller should hold session s_mutex.
795  */
796 static void update_dentry_lease(struct dentry *dentry,
797                                 struct ceph_mds_reply_lease *lease,
798                                 struct ceph_mds_session *session,
799                                 unsigned long from_time)
800 {
801         struct ceph_dentry_info *di = ceph_dentry(dentry);
802         long unsigned duration = le32_to_cpu(lease->duration_ms);
803         long unsigned ttl = from_time + (duration * HZ) / 1000;
804         long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000;
805         struct inode *dir;
806
807         /* only track leases on regular dentries */
808         if (dentry->d_op != &ceph_dentry_ops)
809                 return;
810
811         spin_lock(&dentry->d_lock);
812         dout("update_dentry_lease %p duration %lu ms ttl %lu\n",
813              dentry, duration, ttl);
814
815         /* make lease_rdcache_gen match directory */
816         dir = dentry->d_parent->d_inode;
817         di->lease_shared_gen = ceph_inode(dir)->i_shared_gen;
818
819         if (duration == 0)
820                 goto out_unlock;
821
822         if (di->lease_gen == session->s_cap_gen &&
823             time_before(ttl, dentry->d_time))
824                 goto out_unlock;  /* we already have a newer lease. */
825
826         if (di->lease_session && di->lease_session != session)
827                 goto out_unlock;
828
829         ceph_dentry_lru_touch(dentry);
830
831         if (!di->lease_session)
832                 di->lease_session = ceph_get_mds_session(session);
833         di->lease_gen = session->s_cap_gen;
834         di->lease_seq = le32_to_cpu(lease->seq);
835         di->lease_renew_after = half_ttl;
836         di->lease_renew_from = 0;
837         dentry->d_time = ttl;
838 out_unlock:
839         spin_unlock(&dentry->d_lock);
840         return;
841 }
842
843 /*
844  * Set dentry's directory position based on the current dir's max, and
845  * order it in d_subdirs, so that dcache_readdir behaves.
846  */
847 static void ceph_set_dentry_offset(struct dentry *dn)
848 {
849         struct dentry *dir = dn->d_parent;
850         struct inode *inode = dn->d_parent->d_inode;
851         struct ceph_dentry_info *di;
852
853         BUG_ON(!inode);
854
855         di = ceph_dentry(dn);
856
857         spin_lock(&inode->i_lock);
858         if ((ceph_inode(inode)->i_ceph_flags & CEPH_I_COMPLETE) == 0) {
859                 spin_unlock(&inode->i_lock);
860                 return;
861         }
862         di->offset = ceph_inode(inode)->i_max_offset++;
863         spin_unlock(&inode->i_lock);
864
865         spin_lock(&dir->d_lock);
866         spin_lock_nested(&dn->d_lock, DENTRY_D_LOCK_NESTED);
867         list_move(&dn->d_u.d_child, &dir->d_subdirs);
868         dout("set_dentry_offset %p %lld (%p %p)\n", dn, di->offset,
869              dn->d_u.d_child.prev, dn->d_u.d_child.next);
870         spin_unlock(&dn->d_lock);
871         spin_unlock(&dir->d_lock);
872 }
873
874 /*
875  * splice a dentry to an inode.
876  * caller must hold directory i_mutex for this to be safe.
877  *
878  * we will only rehash the resulting dentry if @prehash is
879  * true; @prehash will be set to false (for the benefit of
880  * the caller) if we fail.
881  */
882 static struct dentry *splice_dentry(struct dentry *dn, struct inode *in,
883                                     bool *prehash, bool set_offset)
884 {
885         struct dentry *realdn;
886
887         BUG_ON(dn->d_inode);
888
889         /* dn must be unhashed */
890         if (!d_unhashed(dn))
891                 d_drop(dn);
892         realdn = d_materialise_unique(dn, in);
893         if (IS_ERR(realdn)) {
894                 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
895                        PTR_ERR(realdn), dn, in, ceph_vinop(in));
896                 if (prehash)
897                         *prehash = false; /* don't rehash on error */
898                 dn = realdn; /* note realdn contains the error */
899                 goto out;
900         } else if (realdn) {
901                 dout("dn %p (%d) spliced with %p (%d) "
902                      "inode %p ino %llx.%llx\n",
903                      dn, dn->d_count,
904                      realdn, realdn->d_count,
905                      realdn->d_inode, ceph_vinop(realdn->d_inode));
906                 dput(dn);
907                 dn = realdn;
908         } else {
909                 BUG_ON(!ceph_dentry(dn));
910                 dout("dn %p attached to %p ino %llx.%llx\n",
911                      dn, dn->d_inode, ceph_vinop(dn->d_inode));
912         }
913         if ((!prehash || *prehash) && d_unhashed(dn))
914                 d_rehash(dn);
915         if (set_offset)
916                 ceph_set_dentry_offset(dn);
917 out:
918         return dn;
919 }
920
921 /*
922  * Incorporate results into the local cache.  This is either just
923  * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
924  * after a lookup).
925  *
926  * A reply may contain
927  *         a directory inode along with a dentry.
928  *  and/or a target inode
929  *
930  * Called with snap_rwsem (read).
931  */
932 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req,
933                     struct ceph_mds_session *session)
934 {
935         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
936         struct inode *in = NULL;
937         struct ceph_mds_reply_inode *ininfo;
938         struct ceph_vino vino;
939         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
940         int i = 0;
941         int err = 0;
942
943         dout("fill_trace %p is_dentry %d is_target %d\n", req,
944              rinfo->head->is_dentry, rinfo->head->is_target);
945
946 #if 0
947         /*
948          * Debugging hook:
949          *
950          * If we resend completed ops to a recovering mds, we get no
951          * trace.  Since that is very rare, pretend this is the case
952          * to ensure the 'no trace' handlers in the callers behave.
953          *
954          * Fill in inodes unconditionally to avoid breaking cap
955          * invariants.
956          */
957         if (rinfo->head->op & CEPH_MDS_OP_WRITE) {
958                 pr_info("fill_trace faking empty trace on %lld %s\n",
959                         req->r_tid, ceph_mds_op_name(rinfo->head->op));
960                 if (rinfo->head->is_dentry) {
961                         rinfo->head->is_dentry = 0;
962                         err = fill_inode(req->r_locked_dir,
963                                          &rinfo->diri, rinfo->dirfrag,
964                                          session, req->r_request_started, -1);
965                 }
966                 if (rinfo->head->is_target) {
967                         rinfo->head->is_target = 0;
968                         ininfo = rinfo->targeti.in;
969                         vino.ino = le64_to_cpu(ininfo->ino);
970                         vino.snap = le64_to_cpu(ininfo->snapid);
971                         in = ceph_get_inode(sb, vino);
972                         err = fill_inode(in, &rinfo->targeti, NULL,
973                                          session, req->r_request_started,
974                                          req->r_fmode);
975                         iput(in);
976                 }
977         }
978 #endif
979
980         if (!rinfo->head->is_target && !rinfo->head->is_dentry) {
981                 dout("fill_trace reply is empty!\n");
982                 if (rinfo->head->result == 0 && req->r_locked_dir)
983                         ceph_invalidate_dir_request(req);
984                 return 0;
985         }
986
987         if (rinfo->head->is_dentry) {
988                 struct inode *dir = req->r_locked_dir;
989
990                 err = fill_inode(dir, &rinfo->diri, rinfo->dirfrag,
991                                  session, req->r_request_started, -1,
992                                  &req->r_caps_reservation);
993                 if (err < 0)
994                         return err;
995         }
996
997         /*
998          * ignore null lease/binding on snapdir ENOENT, or else we
999          * will have trouble splicing in the virtual snapdir later
1000          */
1001         if (rinfo->head->is_dentry && !req->r_aborted &&
1002             (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name,
1003                                                fsc->mount_options->snapdir_name,
1004                                                req->r_dentry->d_name.len))) {
1005                 /*
1006                  * lookup link rename   : null -> possibly existing inode
1007                  * mknod symlink mkdir  : null -> new inode
1008                  * unlink               : linked -> null
1009                  */
1010                 struct inode *dir = req->r_locked_dir;
1011                 struct dentry *dn = req->r_dentry;
1012                 bool have_dir_cap, have_lease;
1013
1014                 BUG_ON(!dn);
1015                 BUG_ON(!dir);
1016                 BUG_ON(dn->d_parent->d_inode != dir);
1017                 BUG_ON(ceph_ino(dir) !=
1018                        le64_to_cpu(rinfo->diri.in->ino));
1019                 BUG_ON(ceph_snap(dir) !=
1020                        le64_to_cpu(rinfo->diri.in->snapid));
1021
1022                 /* do we have a lease on the whole dir? */
1023                 have_dir_cap =
1024                         (le32_to_cpu(rinfo->diri.in->cap.caps) &
1025                          CEPH_CAP_FILE_SHARED);
1026
1027                 /* do we have a dn lease? */
1028                 have_lease = have_dir_cap ||
1029                         le32_to_cpu(rinfo->dlease->duration_ms);
1030                 if (!have_lease)
1031                         dout("fill_trace  no dentry lease or dir cap\n");
1032
1033                 /* rename? */
1034                 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) {
1035                         dout(" src %p '%.*s' dst %p '%.*s'\n",
1036                              req->r_old_dentry,
1037                              req->r_old_dentry->d_name.len,
1038                              req->r_old_dentry->d_name.name,
1039                              dn, dn->d_name.len, dn->d_name.name);
1040                         dout("fill_trace doing d_move %p -> %p\n",
1041                              req->r_old_dentry, dn);
1042
1043                         d_move(req->r_old_dentry, dn);
1044                         dout(" src %p '%.*s' dst %p '%.*s'\n",
1045                              req->r_old_dentry,
1046                              req->r_old_dentry->d_name.len,
1047                              req->r_old_dentry->d_name.name,
1048                              dn, dn->d_name.len, dn->d_name.name);
1049
1050                         /* ensure target dentry is invalidated, despite
1051                            rehashing bug in vfs_rename_dir */
1052                         ceph_invalidate_dentry_lease(dn);
1053
1054                         /*
1055                          * d_move() puts the renamed dentry at the end of
1056                          * d_subdirs.  We need to assign it an appropriate
1057                          * directory offset so we can behave when holding
1058                          * I_COMPLETE.
1059                          */
1060                         ceph_set_dentry_offset(req->r_old_dentry);
1061                         dout("dn %p gets new offset %lld\n", req->r_old_dentry, 
1062                              ceph_dentry(req->r_old_dentry)->offset);
1063
1064                         dn = req->r_old_dentry;  /* use old_dentry */
1065                         in = dn->d_inode;
1066                 }
1067
1068                 /* null dentry? */
1069                 if (!rinfo->head->is_target) {
1070                         dout("fill_trace null dentry\n");
1071                         if (dn->d_inode) {
1072                                 dout("d_delete %p\n", dn);
1073                                 d_delete(dn);
1074                         } else {
1075                                 dout("d_instantiate %p NULL\n", dn);
1076                                 d_instantiate(dn, NULL);
1077                                 if (have_lease && d_unhashed(dn))
1078                                         d_rehash(dn);
1079                                 update_dentry_lease(dn, rinfo->dlease,
1080                                                     session,
1081                                                     req->r_request_started);
1082                         }
1083                         goto done;
1084                 }
1085
1086                 /* attach proper inode */
1087                 ininfo = rinfo->targeti.in;
1088                 vino.ino = le64_to_cpu(ininfo->ino);
1089                 vino.snap = le64_to_cpu(ininfo->snapid);
1090                 in = dn->d_inode;
1091                 if (!in) {
1092                         in = ceph_get_inode(sb, vino);
1093                         if (IS_ERR(in)) {
1094                                 pr_err("fill_trace bad get_inode "
1095                                        "%llx.%llx\n", vino.ino, vino.snap);
1096                                 err = PTR_ERR(in);
1097                                 d_delete(dn);
1098                                 goto done;
1099                         }
1100                         dn = splice_dentry(dn, in, &have_lease, true);
1101                         if (IS_ERR(dn)) {
1102                                 err = PTR_ERR(dn);
1103                                 goto done;
1104                         }
1105                         req->r_dentry = dn;  /* may have spliced */
1106                         ihold(in);
1107                 } else if (ceph_ino(in) == vino.ino &&
1108                            ceph_snap(in) == vino.snap) {
1109                         ihold(in);
1110                 } else {
1111                         dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1112                              dn, in, ceph_ino(in), ceph_snap(in),
1113                              vino.ino, vino.snap);
1114                         have_lease = false;
1115                         in = NULL;
1116                 }
1117
1118                 if (have_lease)
1119                         update_dentry_lease(dn, rinfo->dlease, session,
1120                                             req->r_request_started);
1121                 dout(" final dn %p\n", dn);
1122                 i++;
1123         } else if (req->r_op == CEPH_MDS_OP_LOOKUPSNAP ||
1124                    req->r_op == CEPH_MDS_OP_MKSNAP) {
1125                 struct dentry *dn = req->r_dentry;
1126
1127                 /* fill out a snapdir LOOKUPSNAP dentry */
1128                 BUG_ON(!dn);
1129                 BUG_ON(!req->r_locked_dir);
1130                 BUG_ON(ceph_snap(req->r_locked_dir) != CEPH_SNAPDIR);
1131                 ininfo = rinfo->targeti.in;
1132                 vino.ino = le64_to_cpu(ininfo->ino);
1133                 vino.snap = le64_to_cpu(ininfo->snapid);
1134                 in = ceph_get_inode(sb, vino);
1135                 if (IS_ERR(in)) {
1136                         pr_err("fill_inode get_inode badness %llx.%llx\n",
1137                                vino.ino, vino.snap);
1138                         err = PTR_ERR(in);
1139                         d_delete(dn);
1140                         goto done;
1141                 }
1142                 dout(" linking snapped dir %p to dn %p\n", in, dn);
1143                 dn = splice_dentry(dn, in, NULL, true);
1144                 if (IS_ERR(dn)) {
1145                         err = PTR_ERR(dn);
1146                         goto done;
1147                 }
1148                 req->r_dentry = dn;  /* may have spliced */
1149                 ihold(in);
1150                 rinfo->head->is_dentry = 1;  /* fool notrace handlers */
1151         }
1152
1153         if (rinfo->head->is_target) {
1154                 vino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1155                 vino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1156
1157                 if (in == NULL || ceph_ino(in) != vino.ino ||
1158                     ceph_snap(in) != vino.snap) {
1159                         in = ceph_get_inode(sb, vino);
1160                         if (IS_ERR(in)) {
1161                                 err = PTR_ERR(in);
1162                                 goto done;
1163                         }
1164                 }
1165                 req->r_target_inode = in;
1166
1167                 err = fill_inode(in,
1168                                  &rinfo->targeti, NULL,
1169                                  session, req->r_request_started,
1170                                  (le32_to_cpu(rinfo->head->result) == 0) ?
1171                                  req->r_fmode : -1,
1172                                  &req->r_caps_reservation);
1173                 if (err < 0) {
1174                         pr_err("fill_inode badness %p %llx.%llx\n",
1175                                in, ceph_vinop(in));
1176                         goto done;
1177                 }
1178         }
1179
1180 done:
1181         dout("fill_trace done err=%d\n", err);
1182         return err;
1183 }
1184
1185 /*
1186  * Prepopulate our cache with readdir results, leases, etc.
1187  */
1188 int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1189                              struct ceph_mds_session *session)
1190 {
1191         struct dentry *parent = req->r_dentry;
1192         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1193         struct qstr dname;
1194         struct dentry *dn;
1195         struct inode *in;
1196         int err = 0, i;
1197         struct inode *snapdir = NULL;
1198         struct ceph_mds_request_head *rhead = req->r_request->front.iov_base;
1199         u64 frag = le32_to_cpu(rhead->args.readdir.frag);
1200         struct ceph_dentry_info *di;
1201
1202         if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
1203                 snapdir = ceph_get_snapdir(parent->d_inode);
1204                 parent = d_find_alias(snapdir);
1205                 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1206                      rinfo->dir_nr, parent);
1207         } else {
1208                 dout("readdir_prepopulate %d items under dn %p\n",
1209                      rinfo->dir_nr, parent);
1210                 if (rinfo->dir_dir)
1211                         ceph_fill_dirfrag(parent->d_inode, rinfo->dir_dir);
1212         }
1213
1214         for (i = 0; i < rinfo->dir_nr; i++) {
1215                 struct ceph_vino vino;
1216
1217                 dname.name = rinfo->dir_dname[i];
1218                 dname.len = rinfo->dir_dname_len[i];
1219                 dname.hash = full_name_hash(dname.name, dname.len);
1220
1221                 vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino);
1222                 vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid);
1223
1224 retry_lookup:
1225                 dn = d_lookup(parent, &dname);
1226                 dout("d_lookup on parent=%p name=%.*s got %p\n",
1227                      parent, dname.len, dname.name, dn);
1228
1229                 if (!dn) {
1230                         dn = d_alloc(parent, &dname);
1231                         dout("d_alloc %p '%.*s' = %p\n", parent,
1232                              dname.len, dname.name, dn);
1233                         if (dn == NULL) {
1234                                 dout("d_alloc badness\n");
1235                                 err = -ENOMEM;
1236                                 goto out;
1237                         }
1238                         err = ceph_init_dentry(dn);
1239                         if (err < 0) {
1240                                 dput(dn);
1241                                 goto out;
1242                         }
1243                 } else if (dn->d_inode &&
1244                            (ceph_ino(dn->d_inode) != vino.ino ||
1245                             ceph_snap(dn->d_inode) != vino.snap)) {
1246                         dout(" dn %p points to wrong inode %p\n",
1247                              dn, dn->d_inode);
1248                         d_delete(dn);
1249                         dput(dn);
1250                         goto retry_lookup;
1251                 } else {
1252                         /* reorder parent's d_subdirs */
1253                         spin_lock(&parent->d_lock);
1254                         spin_lock_nested(&dn->d_lock, DENTRY_D_LOCK_NESTED);
1255                         list_move(&dn->d_u.d_child, &parent->d_subdirs);
1256                         spin_unlock(&dn->d_lock);
1257                         spin_unlock(&parent->d_lock);
1258                 }
1259
1260                 di = dn->d_fsdata;
1261                 di->offset = ceph_make_fpos(frag, i + req->r_readdir_offset);
1262
1263                 /* inode */
1264                 if (dn->d_inode) {
1265                         in = dn->d_inode;
1266                 } else {
1267                         in = ceph_get_inode(parent->d_sb, vino);
1268                         if (IS_ERR(in)) {
1269                                 dout("new_inode badness\n");
1270                                 d_delete(dn);
1271                                 dput(dn);
1272                                 err = PTR_ERR(in);
1273                                 goto out;
1274                         }
1275                         dn = splice_dentry(dn, in, NULL, false);
1276                         if (IS_ERR(dn))
1277                                 dn = NULL;
1278                 }
1279
1280                 if (fill_inode(in, &rinfo->dir_in[i], NULL, session,
1281                                req->r_request_started, -1,
1282                                &req->r_caps_reservation) < 0) {
1283                         pr_err("fill_inode badness on %p\n", in);
1284                         goto next_item;
1285                 }
1286                 if (dn)
1287                         update_dentry_lease(dn, rinfo->dir_dlease[i],
1288                                             req->r_session,
1289                                             req->r_request_started);
1290 next_item:
1291                 if (dn)
1292                         dput(dn);
1293         }
1294         req->r_did_prepopulate = true;
1295
1296 out:
1297         if (snapdir) {
1298                 iput(snapdir);
1299                 dput(parent);
1300         }
1301         dout("readdir_prepopulate done\n");
1302         return err;
1303 }
1304
1305 int ceph_inode_set_size(struct inode *inode, loff_t size)
1306 {
1307         struct ceph_inode_info *ci = ceph_inode(inode);
1308         int ret = 0;
1309
1310         spin_lock(&inode->i_lock);
1311         dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size);
1312         inode->i_size = size;
1313         inode->i_blocks = (size + (1 << 9) - 1) >> 9;
1314
1315         /* tell the MDS if we are approaching max_size */
1316         if ((size << 1) >= ci->i_max_size &&
1317             (ci->i_reported_size << 1) < ci->i_max_size)
1318                 ret = 1;
1319
1320         spin_unlock(&inode->i_lock);
1321         return ret;
1322 }
1323
1324 /*
1325  * Write back inode data in a worker thread.  (This can't be done
1326  * in the message handler context.)
1327  */
1328 void ceph_queue_writeback(struct inode *inode)
1329 {
1330         if (queue_work(ceph_inode_to_client(inode)->wb_wq,
1331                        &ceph_inode(inode)->i_wb_work)) {
1332                 dout("ceph_queue_writeback %p\n", inode);
1333                 ihold(inode);
1334         } else {
1335                 dout("ceph_queue_writeback %p failed\n", inode);
1336         }
1337 }
1338
1339 static void ceph_writeback_work(struct work_struct *work)
1340 {
1341         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1342                                                   i_wb_work);
1343         struct inode *inode = &ci->vfs_inode;
1344
1345         dout("writeback %p\n", inode);
1346         filemap_fdatawrite(&inode->i_data);
1347         iput(inode);
1348 }
1349
1350 /*
1351  * queue an async invalidation
1352  */
1353 void ceph_queue_invalidate(struct inode *inode)
1354 {
1355         if (queue_work(ceph_inode_to_client(inode)->pg_inv_wq,
1356                        &ceph_inode(inode)->i_pg_inv_work)) {
1357                 dout("ceph_queue_invalidate %p\n", inode);
1358                 ihold(inode);
1359         } else {
1360                 dout("ceph_queue_invalidate %p failed\n", inode);
1361         }
1362 }
1363
1364 /*
1365  * invalidate any pages that are not dirty or under writeback.  this
1366  * includes pages that are clean and mapped.
1367  */
1368 static void ceph_invalidate_nondirty_pages(struct address_space *mapping)
1369 {
1370         struct pagevec pvec;
1371         pgoff_t next = 0;
1372         int i;
1373
1374         pagevec_init(&pvec, 0);
1375         while (pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
1376                 for (i = 0; i < pagevec_count(&pvec); i++) {
1377                         struct page *page = pvec.pages[i];
1378                         pgoff_t index;
1379                         int skip_page =
1380                                 (PageDirty(page) || PageWriteback(page));
1381
1382                         if (!skip_page)
1383                                 skip_page = !trylock_page(page);
1384
1385                         /*
1386                          * We really shouldn't be looking at the ->index of an
1387                          * unlocked page.  But we're not allowed to lock these
1388                          * pages.  So we rely upon nobody altering the ->index
1389                          * of this (pinned-by-us) page.
1390                          */
1391                         index = page->index;
1392                         if (index > next)
1393                                 next = index;
1394                         next++;
1395
1396                         if (skip_page)
1397                                 continue;
1398
1399                         generic_error_remove_page(mapping, page);
1400                         unlock_page(page);
1401                 }
1402                 pagevec_release(&pvec);
1403                 cond_resched();
1404         }
1405 }
1406
1407 /*
1408  * Invalidate inode pages in a worker thread.  (This can't be done
1409  * in the message handler context.)
1410  */
1411 static void ceph_invalidate_work(struct work_struct *work)
1412 {
1413         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1414                                                   i_pg_inv_work);
1415         struct inode *inode = &ci->vfs_inode;
1416         u32 orig_gen;
1417         int check = 0;
1418
1419         spin_lock(&inode->i_lock);
1420         dout("invalidate_pages %p gen %d revoking %d\n", inode,
1421              ci->i_rdcache_gen, ci->i_rdcache_revoking);
1422         if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
1423                 /* nevermind! */
1424                 spin_unlock(&inode->i_lock);
1425                 goto out;
1426         }
1427         orig_gen = ci->i_rdcache_gen;
1428         spin_unlock(&inode->i_lock);
1429
1430         ceph_invalidate_nondirty_pages(inode->i_mapping);
1431
1432         spin_lock(&inode->i_lock);
1433         if (orig_gen == ci->i_rdcache_gen &&
1434             orig_gen == ci->i_rdcache_revoking) {
1435                 dout("invalidate_pages %p gen %d successful\n", inode,
1436                      ci->i_rdcache_gen);
1437                 ci->i_rdcache_revoking--;
1438                 check = 1;
1439         } else {
1440                 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1441                      inode, orig_gen, ci->i_rdcache_gen,
1442                      ci->i_rdcache_revoking);
1443         }
1444         spin_unlock(&inode->i_lock);
1445
1446         if (check)
1447                 ceph_check_caps(ci, 0, NULL);
1448 out:
1449         iput(inode);
1450 }
1451
1452
1453 /*
1454  * called by trunc_wq; take i_mutex ourselves
1455  *
1456  * We also truncate in a separate thread as well.
1457  */
1458 static void ceph_vmtruncate_work(struct work_struct *work)
1459 {
1460         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1461                                                   i_vmtruncate_work);
1462         struct inode *inode = &ci->vfs_inode;
1463
1464         dout("vmtruncate_work %p\n", inode);
1465         mutex_lock(&inode->i_mutex);
1466         __ceph_do_pending_vmtruncate(inode);
1467         mutex_unlock(&inode->i_mutex);
1468         iput(inode);
1469 }
1470
1471 /*
1472  * Queue an async vmtruncate.  If we fail to queue work, we will handle
1473  * the truncation the next time we call __ceph_do_pending_vmtruncate.
1474  */
1475 void ceph_queue_vmtruncate(struct inode *inode)
1476 {
1477         struct ceph_inode_info *ci = ceph_inode(inode);
1478
1479         if (queue_work(ceph_sb_to_client(inode->i_sb)->trunc_wq,
1480                        &ci->i_vmtruncate_work)) {
1481                 dout("ceph_queue_vmtruncate %p\n", inode);
1482                 ihold(inode);
1483         } else {
1484                 dout("ceph_queue_vmtruncate %p failed, pending=%d\n",
1485                      inode, ci->i_truncate_pending);
1486         }
1487 }
1488
1489 /*
1490  * called with i_mutex held.
1491  *
1492  * Make sure any pending truncation is applied before doing anything
1493  * that may depend on it.
1494  */
1495 void __ceph_do_pending_vmtruncate(struct inode *inode)
1496 {
1497         struct ceph_inode_info *ci = ceph_inode(inode);
1498         u64 to;
1499         int wrbuffer_refs, wake = 0;
1500
1501 retry:
1502         spin_lock(&inode->i_lock);
1503         if (ci->i_truncate_pending == 0) {
1504                 dout("__do_pending_vmtruncate %p none pending\n", inode);
1505                 spin_unlock(&inode->i_lock);
1506                 return;
1507         }
1508
1509         /*
1510          * make sure any dirty snapped pages are flushed before we
1511          * possibly truncate them.. so write AND block!
1512          */
1513         if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) {
1514                 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1515                      inode);
1516                 spin_unlock(&inode->i_lock);
1517                 filemap_write_and_wait_range(&inode->i_data, 0,
1518                                              inode->i_sb->s_maxbytes);
1519                 goto retry;
1520         }
1521
1522         to = ci->i_truncate_size;
1523         wrbuffer_refs = ci->i_wrbuffer_ref;
1524         dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode,
1525              ci->i_truncate_pending, to);
1526         spin_unlock(&inode->i_lock);
1527
1528         truncate_inode_pages(inode->i_mapping, to);
1529
1530         spin_lock(&inode->i_lock);
1531         ci->i_truncate_pending--;
1532         if (ci->i_truncate_pending == 0)
1533                 wake = 1;
1534         spin_unlock(&inode->i_lock);
1535
1536         if (wrbuffer_refs == 0)
1537                 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
1538         if (wake)
1539                 wake_up_all(&ci->i_cap_wq);
1540 }
1541
1542
1543 /*
1544  * symlinks
1545  */
1546 static void *ceph_sym_follow_link(struct dentry *dentry, struct nameidata *nd)
1547 {
1548         struct ceph_inode_info *ci = ceph_inode(dentry->d_inode);
1549         nd_set_link(nd, ci->i_symlink);
1550         return NULL;
1551 }
1552
1553 static const struct inode_operations ceph_symlink_iops = {
1554         .readlink = generic_readlink,
1555         .follow_link = ceph_sym_follow_link,
1556 };
1557
1558 /*
1559  * setattr
1560  */
1561 int ceph_setattr(struct dentry *dentry, struct iattr *attr)
1562 {
1563         struct inode *inode = dentry->d_inode;
1564         struct ceph_inode_info *ci = ceph_inode(inode);
1565         struct inode *parent_inode;
1566         const unsigned int ia_valid = attr->ia_valid;
1567         struct ceph_mds_request *req;
1568         struct ceph_mds_client *mdsc = ceph_sb_to_client(dentry->d_sb)->mdsc;
1569         int issued;
1570         int release = 0, dirtied = 0;
1571         int mask = 0;
1572         int err = 0;
1573         int inode_dirty_flags = 0;
1574
1575         if (ceph_snap(inode) != CEPH_NOSNAP)
1576                 return -EROFS;
1577
1578         __ceph_do_pending_vmtruncate(inode);
1579
1580         err = inode_change_ok(inode, attr);
1581         if (err != 0)
1582                 return err;
1583
1584         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR,
1585                                        USE_AUTH_MDS);
1586         if (IS_ERR(req))
1587                 return PTR_ERR(req);
1588
1589         spin_lock(&inode->i_lock);
1590         issued = __ceph_caps_issued(ci, NULL);
1591         dout("setattr %p issued %s\n", inode, ceph_cap_string(issued));
1592
1593         if (ia_valid & ATTR_UID) {
1594                 dout("setattr %p uid %d -> %d\n", inode,
1595                      inode->i_uid, attr->ia_uid);
1596                 if (issued & CEPH_CAP_AUTH_EXCL) {
1597                         inode->i_uid = attr->ia_uid;
1598                         dirtied |= CEPH_CAP_AUTH_EXCL;
1599                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1600                            attr->ia_uid != inode->i_uid) {
1601                         req->r_args.setattr.uid = cpu_to_le32(attr->ia_uid);
1602                         mask |= CEPH_SETATTR_UID;
1603                         release |= CEPH_CAP_AUTH_SHARED;
1604                 }
1605         }
1606         if (ia_valid & ATTR_GID) {
1607                 dout("setattr %p gid %d -> %d\n", inode,
1608                      inode->i_gid, attr->ia_gid);
1609                 if (issued & CEPH_CAP_AUTH_EXCL) {
1610                         inode->i_gid = attr->ia_gid;
1611                         dirtied |= CEPH_CAP_AUTH_EXCL;
1612                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1613                            attr->ia_gid != inode->i_gid) {
1614                         req->r_args.setattr.gid = cpu_to_le32(attr->ia_gid);
1615                         mask |= CEPH_SETATTR_GID;
1616                         release |= CEPH_CAP_AUTH_SHARED;
1617                 }
1618         }
1619         if (ia_valid & ATTR_MODE) {
1620                 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode,
1621                      attr->ia_mode);
1622                 if (issued & CEPH_CAP_AUTH_EXCL) {
1623                         inode->i_mode = attr->ia_mode;
1624                         dirtied |= CEPH_CAP_AUTH_EXCL;
1625                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1626                            attr->ia_mode != inode->i_mode) {
1627                         req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode);
1628                         mask |= CEPH_SETATTR_MODE;
1629                         release |= CEPH_CAP_AUTH_SHARED;
1630                 }
1631         }
1632
1633         if (ia_valid & ATTR_ATIME) {
1634                 dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode,
1635                      inode->i_atime.tv_sec, inode->i_atime.tv_nsec,
1636                      attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec);
1637                 if (issued & CEPH_CAP_FILE_EXCL) {
1638                         ci->i_time_warp_seq++;
1639                         inode->i_atime = attr->ia_atime;
1640                         dirtied |= CEPH_CAP_FILE_EXCL;
1641                 } else if ((issued & CEPH_CAP_FILE_WR) &&
1642                            timespec_compare(&inode->i_atime,
1643                                             &attr->ia_atime) < 0) {
1644                         inode->i_atime = attr->ia_atime;
1645                         dirtied |= CEPH_CAP_FILE_WR;
1646                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1647                            !timespec_equal(&inode->i_atime, &attr->ia_atime)) {
1648                         ceph_encode_timespec(&req->r_args.setattr.atime,
1649                                              &attr->ia_atime);
1650                         mask |= CEPH_SETATTR_ATIME;
1651                         release |= CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_RD |
1652                                 CEPH_CAP_FILE_WR;
1653                 }
1654         }
1655         if (ia_valid & ATTR_MTIME) {
1656                 dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode,
1657                      inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
1658                      attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec);
1659                 if (issued & CEPH_CAP_FILE_EXCL) {
1660                         ci->i_time_warp_seq++;
1661                         inode->i_mtime = attr->ia_mtime;
1662                         dirtied |= CEPH_CAP_FILE_EXCL;
1663                 } else if ((issued & CEPH_CAP_FILE_WR) &&
1664                            timespec_compare(&inode->i_mtime,
1665                                             &attr->ia_mtime) < 0) {
1666                         inode->i_mtime = attr->ia_mtime;
1667                         dirtied |= CEPH_CAP_FILE_WR;
1668                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1669                            !timespec_equal(&inode->i_mtime, &attr->ia_mtime)) {
1670                         ceph_encode_timespec(&req->r_args.setattr.mtime,
1671                                              &attr->ia_mtime);
1672                         mask |= CEPH_SETATTR_MTIME;
1673                         release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
1674                                 CEPH_CAP_FILE_WR;
1675                 }
1676         }
1677         if (ia_valid & ATTR_SIZE) {
1678                 dout("setattr %p size %lld -> %lld\n", inode,
1679                      inode->i_size, attr->ia_size);
1680                 if (attr->ia_size > inode->i_sb->s_maxbytes) {
1681                         err = -EINVAL;
1682                         goto out;
1683                 }
1684                 if ((issued & CEPH_CAP_FILE_EXCL) &&
1685                     attr->ia_size > inode->i_size) {
1686                         inode->i_size = attr->ia_size;
1687                         inode->i_blocks =
1688                                 (attr->ia_size + (1 << 9) - 1) >> 9;
1689                         inode->i_ctime = attr->ia_ctime;
1690                         ci->i_reported_size = attr->ia_size;
1691                         dirtied |= CEPH_CAP_FILE_EXCL;
1692                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1693                            attr->ia_size != inode->i_size) {
1694                         req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
1695                         req->r_args.setattr.old_size =
1696                                 cpu_to_le64(inode->i_size);
1697                         mask |= CEPH_SETATTR_SIZE;
1698                         release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
1699                                 CEPH_CAP_FILE_WR;
1700                 }
1701         }
1702
1703         /* these do nothing */
1704         if (ia_valid & ATTR_CTIME) {
1705                 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME|
1706                                          ATTR_MODE|ATTR_UID|ATTR_GID)) == 0;
1707                 dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode,
1708                      inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
1709                      attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec,
1710                      only ? "ctime only" : "ignored");
1711                 inode->i_ctime = attr->ia_ctime;
1712                 if (only) {
1713                         /*
1714                          * if kernel wants to dirty ctime but nothing else,
1715                          * we need to choose a cap to dirty under, or do
1716                          * a almost-no-op setattr
1717                          */
1718                         if (issued & CEPH_CAP_AUTH_EXCL)
1719                                 dirtied |= CEPH_CAP_AUTH_EXCL;
1720                         else if (issued & CEPH_CAP_FILE_EXCL)
1721                                 dirtied |= CEPH_CAP_FILE_EXCL;
1722                         else if (issued & CEPH_CAP_XATTR_EXCL)
1723                                 dirtied |= CEPH_CAP_XATTR_EXCL;
1724                         else
1725                                 mask |= CEPH_SETATTR_CTIME;
1726                 }
1727         }
1728         if (ia_valid & ATTR_FILE)
1729                 dout("setattr %p ATTR_FILE ... hrm!\n", inode);
1730
1731         if (dirtied) {
1732                 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied);
1733                 inode->i_ctime = CURRENT_TIME;
1734         }
1735
1736         release &= issued;
1737         spin_unlock(&inode->i_lock);
1738
1739         if (inode_dirty_flags)
1740                 __mark_inode_dirty(inode, inode_dirty_flags);
1741
1742         if (mask) {
1743                 req->r_inode = inode;
1744                 ihold(inode);
1745                 req->r_inode_drop = release;
1746                 req->r_args.setattr.mask = cpu_to_le32(mask);
1747                 req->r_num_caps = 1;
1748                 parent_inode = ceph_get_dentry_parent_inode(dentry);
1749                 err = ceph_mdsc_do_request(mdsc, parent_inode, req);
1750                 iput(parent_inode);
1751         }
1752         dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err,
1753              ceph_cap_string(dirtied), mask);
1754
1755         ceph_mdsc_put_request(req);
1756         __ceph_do_pending_vmtruncate(inode);
1757         return err;
1758 out:
1759         spin_unlock(&inode->i_lock);
1760         ceph_mdsc_put_request(req);
1761         return err;
1762 }
1763
1764 /*
1765  * Verify that we have a lease on the given mask.  If not,
1766  * do a getattr against an mds.
1767  */
1768 int ceph_do_getattr(struct inode *inode, int mask)
1769 {
1770         struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
1771         struct ceph_mds_client *mdsc = fsc->mdsc;
1772         struct ceph_mds_request *req;
1773         int err;
1774
1775         if (ceph_snap(inode) == CEPH_SNAPDIR) {
1776                 dout("do_getattr inode %p SNAPDIR\n", inode);
1777                 return 0;
1778         }
1779
1780         dout("do_getattr inode %p mask %s mode 0%o\n", inode, ceph_cap_string(mask), inode->i_mode);
1781         if (ceph_caps_issued_mask(ceph_inode(inode), mask, 1))
1782                 return 0;
1783
1784         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
1785         if (IS_ERR(req))
1786                 return PTR_ERR(req);
1787         req->r_inode = inode;
1788         ihold(inode);
1789         req->r_num_caps = 1;
1790         req->r_args.getattr.mask = cpu_to_le32(mask);
1791         err = ceph_mdsc_do_request(mdsc, NULL, req);
1792         ceph_mdsc_put_request(req);
1793         dout("do_getattr result=%d\n", err);
1794         return err;
1795 }
1796
1797
1798 /*
1799  * Check inode permissions.  We verify we have a valid value for
1800  * the AUTH cap, then call the generic handler.
1801  */
1802 int ceph_permission(struct inode *inode, int mask, unsigned int flags)
1803 {
1804         int err;
1805
1806         if (flags & IPERM_FLAG_RCU)
1807                 return -ECHILD;
1808
1809         err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED);
1810
1811         if (!err)
1812                 err = generic_permission(inode, mask, flags, NULL);
1813         return err;
1814 }
1815
1816 /*
1817  * Get all attributes.  Hopefully somedata we'll have a statlite()
1818  * and can limit the fields we require to be accurate.
1819  */
1820 int ceph_getattr(struct vfsmount *mnt, struct dentry *dentry,
1821                  struct kstat *stat)
1822 {
1823         struct inode *inode = dentry->d_inode;
1824         struct ceph_inode_info *ci = ceph_inode(inode);
1825         int err;
1826
1827         err = ceph_do_getattr(inode, CEPH_STAT_CAP_INODE_ALL);
1828         if (!err) {
1829                 generic_fillattr(inode, stat);
1830                 stat->ino = ceph_translate_ino(inode->i_sb, inode->i_ino);
1831                 if (ceph_snap(inode) != CEPH_NOSNAP)
1832                         stat->dev = ceph_snap(inode);
1833                 else
1834                         stat->dev = 0;
1835                 if (S_ISDIR(inode->i_mode)) {
1836                         if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb),
1837                                                 RBYTES))
1838                                 stat->size = ci->i_rbytes;
1839                         else
1840                                 stat->size = ci->i_files + ci->i_subdirs;
1841                         stat->blocks = 0;
1842                         stat->blksize = 65536;
1843                 }
1844         }
1845         return err;
1846 }