Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/sage/ceph...
[linux-2.6.git] / fs / ceph / inode.c
1 #include "ceph_debug.h"
2
3 #include <linux/module.h>
4 #include <linux/fs.h>
5 #include <linux/smp_lock.h>
6 #include <linux/slab.h>
7 #include <linux/string.h>
8 #include <linux/uaccess.h>
9 #include <linux/kernel.h>
10 #include <linux/namei.h>
11 #include <linux/writeback.h>
12 #include <linux/vmalloc.h>
13 #include <linux/pagevec.h>
14
15 #include "super.h"
16 #include "decode.h"
17
18 /*
19  * Ceph inode operations
20  *
21  * Implement basic inode helpers (get, alloc) and inode ops (getattr,
22  * setattr, etc.), xattr helpers, and helpers for assimilating
23  * metadata returned by the MDS into our cache.
24  *
25  * Also define helpers for doing asynchronous writeback, invalidation,
26  * and truncation for the benefit of those who can't afford to block
27  * (typically because they are in the message handler path).
28  */
29
30 static const struct inode_operations ceph_symlink_iops;
31
32 static void ceph_invalidate_work(struct work_struct *work);
33 static void ceph_writeback_work(struct work_struct *work);
34 static void ceph_vmtruncate_work(struct work_struct *work);
35
36 /*
37  * find or create an inode, given the ceph ino number
38  */
39 struct inode *ceph_get_inode(struct super_block *sb, struct ceph_vino vino)
40 {
41         struct inode *inode;
42         ino_t t = ceph_vino_to_ino(vino);
43
44         inode = iget5_locked(sb, t, ceph_ino_compare, ceph_set_ino_cb, &vino);
45         if (inode == NULL)
46                 return ERR_PTR(-ENOMEM);
47         if (inode->i_state & I_NEW) {
48                 dout("get_inode created new inode %p %llx.%llx ino %llx\n",
49                      inode, ceph_vinop(inode), (u64)inode->i_ino);
50                 unlock_new_inode(inode);
51         }
52
53         dout("get_inode on %lu=%llx.%llx got %p\n", inode->i_ino, vino.ino,
54              vino.snap, inode);
55         return inode;
56 }
57
58 /*
59  * get/constuct snapdir inode for a given directory
60  */
61 struct inode *ceph_get_snapdir(struct inode *parent)
62 {
63         struct ceph_vino vino = {
64                 .ino = ceph_ino(parent),
65                 .snap = CEPH_SNAPDIR,
66         };
67         struct inode *inode = ceph_get_inode(parent->i_sb, vino);
68         struct ceph_inode_info *ci = ceph_inode(inode);
69
70         BUG_ON(!S_ISDIR(parent->i_mode));
71         if (IS_ERR(inode))
72                 return ERR_PTR(PTR_ERR(inode));
73         inode->i_mode = parent->i_mode;
74         inode->i_uid = parent->i_uid;
75         inode->i_gid = parent->i_gid;
76         inode->i_op = &ceph_dir_iops;
77         inode->i_fop = &ceph_dir_fops;
78         ci->i_snap_caps = CEPH_CAP_PIN; /* so we can open */
79         ci->i_rbytes = 0;
80         return inode;
81 }
82
83 const struct inode_operations ceph_file_iops = {
84         .permission = ceph_permission,
85         .setattr = ceph_setattr,
86         .getattr = ceph_getattr,
87         .setxattr = ceph_setxattr,
88         .getxattr = ceph_getxattr,
89         .listxattr = ceph_listxattr,
90         .removexattr = ceph_removexattr,
91 };
92
93
94 /*
95  * We use a 'frag tree' to keep track of the MDS's directory fragments
96  * for a given inode (usually there is just a single fragment).  We
97  * need to know when a child frag is delegated to a new MDS, or when
98  * it is flagged as replicated, so we can direct our requests
99  * accordingly.
100  */
101
102 /*
103  * find/create a frag in the tree
104  */
105 static struct ceph_inode_frag *__get_or_create_frag(struct ceph_inode_info *ci,
106                                                     u32 f)
107 {
108         struct rb_node **p;
109         struct rb_node *parent = NULL;
110         struct ceph_inode_frag *frag;
111         int c;
112
113         p = &ci->i_fragtree.rb_node;
114         while (*p) {
115                 parent = *p;
116                 frag = rb_entry(parent, struct ceph_inode_frag, node);
117                 c = ceph_frag_compare(f, frag->frag);
118                 if (c < 0)
119                         p = &(*p)->rb_left;
120                 else if (c > 0)
121                         p = &(*p)->rb_right;
122                 else
123                         return frag;
124         }
125
126         frag = kmalloc(sizeof(*frag), GFP_NOFS);
127         if (!frag) {
128                 pr_err("__get_or_create_frag ENOMEM on %p %llx.%llx "
129                        "frag %x\n", &ci->vfs_inode,
130                        ceph_vinop(&ci->vfs_inode), f);
131                 return ERR_PTR(-ENOMEM);
132         }
133         frag->frag = f;
134         frag->split_by = 0;
135         frag->mds = -1;
136         frag->ndist = 0;
137
138         rb_link_node(&frag->node, parent, p);
139         rb_insert_color(&frag->node, &ci->i_fragtree);
140
141         dout("get_or_create_frag added %llx.%llx frag %x\n",
142              ceph_vinop(&ci->vfs_inode), f);
143         return frag;
144 }
145
146 /*
147  * find a specific frag @f
148  */
149 struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci, u32 f)
150 {
151         struct rb_node *n = ci->i_fragtree.rb_node;
152
153         while (n) {
154                 struct ceph_inode_frag *frag =
155                         rb_entry(n, struct ceph_inode_frag, node);
156                 int c = ceph_frag_compare(f, frag->frag);
157                 if (c < 0)
158                         n = n->rb_left;
159                 else if (c > 0)
160                         n = n->rb_right;
161                 else
162                         return frag;
163         }
164         return NULL;
165 }
166
167 /*
168  * Choose frag containing the given value @v.  If @pfrag is
169  * specified, copy the frag delegation info to the caller if
170  * it is present.
171  */
172 u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
173                      struct ceph_inode_frag *pfrag,
174                      int *found)
175 {
176         u32 t = ceph_frag_make(0, 0);
177         struct ceph_inode_frag *frag;
178         unsigned nway, i;
179         u32 n;
180
181         if (found)
182                 *found = 0;
183
184         mutex_lock(&ci->i_fragtree_mutex);
185         while (1) {
186                 WARN_ON(!ceph_frag_contains_value(t, v));
187                 frag = __ceph_find_frag(ci, t);
188                 if (!frag)
189                         break; /* t is a leaf */
190                 if (frag->split_by == 0) {
191                         if (pfrag)
192                                 memcpy(pfrag, frag, sizeof(*pfrag));
193                         if (found)
194                                 *found = 1;
195                         break;
196                 }
197
198                 /* choose child */
199                 nway = 1 << frag->split_by;
200                 dout("choose_frag(%x) %x splits by %d (%d ways)\n", v, t,
201                      frag->split_by, nway);
202                 for (i = 0; i < nway; i++) {
203                         n = ceph_frag_make_child(t, frag->split_by, i);
204                         if (ceph_frag_contains_value(n, v)) {
205                                 t = n;
206                                 break;
207                         }
208                 }
209                 BUG_ON(i == nway);
210         }
211         dout("choose_frag(%x) = %x\n", v, t);
212
213         mutex_unlock(&ci->i_fragtree_mutex);
214         return t;
215 }
216
217 /*
218  * Process dirfrag (delegation) info from the mds.  Include leaf
219  * fragment in tree ONLY if ndist > 0.  Otherwise, only
220  * branches/splits are included in i_fragtree)
221  */
222 static int ceph_fill_dirfrag(struct inode *inode,
223                              struct ceph_mds_reply_dirfrag *dirinfo)
224 {
225         struct ceph_inode_info *ci = ceph_inode(inode);
226         struct ceph_inode_frag *frag;
227         u32 id = le32_to_cpu(dirinfo->frag);
228         int mds = le32_to_cpu(dirinfo->auth);
229         int ndist = le32_to_cpu(dirinfo->ndist);
230         int i;
231         int err = 0;
232
233         mutex_lock(&ci->i_fragtree_mutex);
234         if (ndist == 0) {
235                 /* no delegation info needed. */
236                 frag = __ceph_find_frag(ci, id);
237                 if (!frag)
238                         goto out;
239                 if (frag->split_by == 0) {
240                         /* tree leaf, remove */
241                         dout("fill_dirfrag removed %llx.%llx frag %x"
242                              " (no ref)\n", ceph_vinop(inode), id);
243                         rb_erase(&frag->node, &ci->i_fragtree);
244                         kfree(frag);
245                 } else {
246                         /* tree branch, keep and clear */
247                         dout("fill_dirfrag cleared %llx.%llx frag %x"
248                              " referral\n", ceph_vinop(inode), id);
249                         frag->mds = -1;
250                         frag->ndist = 0;
251                 }
252                 goto out;
253         }
254
255
256         /* find/add this frag to store mds delegation info */
257         frag = __get_or_create_frag(ci, id);
258         if (IS_ERR(frag)) {
259                 /* this is not the end of the world; we can continue
260                    with bad/inaccurate delegation info */
261                 pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n",
262                        ceph_vinop(inode), le32_to_cpu(dirinfo->frag));
263                 err = -ENOMEM;
264                 goto out;
265         }
266
267         frag->mds = mds;
268         frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP);
269         for (i = 0; i < frag->ndist; i++)
270                 frag->dist[i] = le32_to_cpu(dirinfo->dist[i]);
271         dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n",
272              ceph_vinop(inode), frag->frag, frag->ndist);
273
274 out:
275         mutex_unlock(&ci->i_fragtree_mutex);
276         return err;
277 }
278
279
280 /*
281  * initialize a newly allocated inode.
282  */
283 struct inode *ceph_alloc_inode(struct super_block *sb)
284 {
285         struct ceph_inode_info *ci;
286         int i;
287
288         ci = kmem_cache_alloc(ceph_inode_cachep, GFP_NOFS);
289         if (!ci)
290                 return NULL;
291
292         dout("alloc_inode %p\n", &ci->vfs_inode);
293
294         ci->i_version = 0;
295         ci->i_time_warp_seq = 0;
296         ci->i_ceph_flags = 0;
297         ci->i_release_count = 0;
298         ci->i_symlink = NULL;
299
300         ci->i_fragtree = RB_ROOT;
301         mutex_init(&ci->i_fragtree_mutex);
302
303         ci->i_xattrs.blob = NULL;
304         ci->i_xattrs.prealloc_blob = NULL;
305         ci->i_xattrs.dirty = false;
306         ci->i_xattrs.index = RB_ROOT;
307         ci->i_xattrs.count = 0;
308         ci->i_xattrs.names_size = 0;
309         ci->i_xattrs.vals_size = 0;
310         ci->i_xattrs.version = 0;
311         ci->i_xattrs.index_version = 0;
312
313         ci->i_caps = RB_ROOT;
314         ci->i_auth_cap = NULL;
315         ci->i_dirty_caps = 0;
316         ci->i_flushing_caps = 0;
317         INIT_LIST_HEAD(&ci->i_dirty_item);
318         INIT_LIST_HEAD(&ci->i_flushing_item);
319         ci->i_cap_flush_seq = 0;
320         ci->i_cap_flush_last_tid = 0;
321         memset(&ci->i_cap_flush_tid, 0, sizeof(ci->i_cap_flush_tid));
322         init_waitqueue_head(&ci->i_cap_wq);
323         ci->i_hold_caps_min = 0;
324         ci->i_hold_caps_max = 0;
325         INIT_LIST_HEAD(&ci->i_cap_delay_list);
326         ci->i_cap_exporting_mds = 0;
327         ci->i_cap_exporting_mseq = 0;
328         ci->i_cap_exporting_issued = 0;
329         INIT_LIST_HEAD(&ci->i_cap_snaps);
330         ci->i_head_snapc = NULL;
331         ci->i_snap_caps = 0;
332
333         for (i = 0; i < CEPH_FILE_MODE_NUM; i++)
334                 ci->i_nr_by_mode[i] = 0;
335
336         ci->i_truncate_seq = 0;
337         ci->i_truncate_size = 0;
338         ci->i_truncate_pending = 0;
339
340         ci->i_max_size = 0;
341         ci->i_reported_size = 0;
342         ci->i_wanted_max_size = 0;
343         ci->i_requested_max_size = 0;
344
345         ci->i_pin_ref = 0;
346         ci->i_rd_ref = 0;
347         ci->i_rdcache_ref = 0;
348         ci->i_wr_ref = 0;
349         ci->i_wrbuffer_ref = 0;
350         ci->i_wrbuffer_ref_head = 0;
351         ci->i_shared_gen = 0;
352         ci->i_rdcache_gen = 0;
353         ci->i_rdcache_revoking = 0;
354
355         INIT_LIST_HEAD(&ci->i_unsafe_writes);
356         INIT_LIST_HEAD(&ci->i_unsafe_dirops);
357         spin_lock_init(&ci->i_unsafe_lock);
358
359         ci->i_snap_realm = NULL;
360         INIT_LIST_HEAD(&ci->i_snap_realm_item);
361         INIT_LIST_HEAD(&ci->i_snap_flush_item);
362
363         INIT_WORK(&ci->i_wb_work, ceph_writeback_work);
364         INIT_WORK(&ci->i_pg_inv_work, ceph_invalidate_work);
365
366         INIT_WORK(&ci->i_vmtruncate_work, ceph_vmtruncate_work);
367
368         return &ci->vfs_inode;
369 }
370
371 void ceph_destroy_inode(struct inode *inode)
372 {
373         struct ceph_inode_info *ci = ceph_inode(inode);
374         struct ceph_inode_frag *frag;
375         struct rb_node *n;
376
377         dout("destroy_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode));
378
379         ceph_queue_caps_release(inode);
380
381         /*
382          * we may still have a snap_realm reference if there are stray
383          * caps in i_cap_exporting_issued or i_snap_caps.
384          */
385         if (ci->i_snap_realm) {
386                 struct ceph_mds_client *mdsc =
387                         &ceph_client(ci->vfs_inode.i_sb)->mdsc;
388                 struct ceph_snap_realm *realm = ci->i_snap_realm;
389
390                 dout(" dropping residual ref to snap realm %p\n", realm);
391                 spin_lock(&realm->inodes_with_caps_lock);
392                 list_del_init(&ci->i_snap_realm_item);
393                 spin_unlock(&realm->inodes_with_caps_lock);
394                 ceph_put_snap_realm(mdsc, realm);
395         }
396
397         kfree(ci->i_symlink);
398         while ((n = rb_first(&ci->i_fragtree)) != NULL) {
399                 frag = rb_entry(n, struct ceph_inode_frag, node);
400                 rb_erase(n, &ci->i_fragtree);
401                 kfree(frag);
402         }
403
404         __ceph_destroy_xattrs(ci);
405         if (ci->i_xattrs.blob)
406                 ceph_buffer_put(ci->i_xattrs.blob);
407         if (ci->i_xattrs.prealloc_blob)
408                 ceph_buffer_put(ci->i_xattrs.prealloc_blob);
409
410         kmem_cache_free(ceph_inode_cachep, ci);
411 }
412
413
414 /*
415  * Helpers to fill in size, ctime, mtime, and atime.  We have to be
416  * careful because either the client or MDS may have more up to date
417  * info, depending on which capabilities are held, and whether
418  * time_warp_seq or truncate_seq have increased.  (Ordinarily, mtime
419  * and size are monotonically increasing, except when utimes() or
420  * truncate() increments the corresponding _seq values.)
421  */
422 int ceph_fill_file_size(struct inode *inode, int issued,
423                         u32 truncate_seq, u64 truncate_size, u64 size)
424 {
425         struct ceph_inode_info *ci = ceph_inode(inode);
426         int queue_trunc = 0;
427
428         if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 ||
429             (truncate_seq == ci->i_truncate_seq && size > inode->i_size)) {
430                 dout("size %lld -> %llu\n", inode->i_size, size);
431                 inode->i_size = size;
432                 inode->i_blocks = (size + (1<<9) - 1) >> 9;
433                 ci->i_reported_size = size;
434                 if (truncate_seq != ci->i_truncate_seq) {
435                         dout("truncate_seq %u -> %u\n",
436                              ci->i_truncate_seq, truncate_seq);
437                         ci->i_truncate_seq = truncate_seq;
438                         /*
439                          * If we hold relevant caps, or in the case where we're
440                          * not the only client referencing this file and we
441                          * don't hold those caps, then we need to check whether
442                          * the file is either opened or mmaped
443                          */
444                         if ((issued & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_RD|
445                                       CEPH_CAP_FILE_WR|CEPH_CAP_FILE_BUFFER|
446                                       CEPH_CAP_FILE_EXCL)) ||
447                             mapping_mapped(inode->i_mapping) ||
448                             __ceph_caps_file_wanted(ci)) {
449                                 ci->i_truncate_pending++;
450                                 queue_trunc = 1;
451                         }
452                 }
453         }
454         if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 &&
455             ci->i_truncate_size != truncate_size) {
456                 dout("truncate_size %lld -> %llu\n", ci->i_truncate_size,
457                      truncate_size);
458                 ci->i_truncate_size = truncate_size;
459         }
460         return queue_trunc;
461 }
462
463 void ceph_fill_file_time(struct inode *inode, int issued,
464                          u64 time_warp_seq, struct timespec *ctime,
465                          struct timespec *mtime, struct timespec *atime)
466 {
467         struct ceph_inode_info *ci = ceph_inode(inode);
468         int warn = 0;
469
470         if (issued & (CEPH_CAP_FILE_EXCL|
471                       CEPH_CAP_FILE_WR|
472                       CEPH_CAP_FILE_BUFFER)) {
473                 if (timespec_compare(ctime, &inode->i_ctime) > 0) {
474                         dout("ctime %ld.%09ld -> %ld.%09ld inc w/ cap\n",
475                              inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
476                              ctime->tv_sec, ctime->tv_nsec);
477                         inode->i_ctime = *ctime;
478                 }
479                 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) {
480                         /* the MDS did a utimes() */
481                         dout("mtime %ld.%09ld -> %ld.%09ld "
482                              "tw %d -> %d\n",
483                              inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
484                              mtime->tv_sec, mtime->tv_nsec,
485                              ci->i_time_warp_seq, (int)time_warp_seq);
486
487                         inode->i_mtime = *mtime;
488                         inode->i_atime = *atime;
489                         ci->i_time_warp_seq = time_warp_seq;
490                 } else if (time_warp_seq == ci->i_time_warp_seq) {
491                         /* nobody did utimes(); take the max */
492                         if (timespec_compare(mtime, &inode->i_mtime) > 0) {
493                                 dout("mtime %ld.%09ld -> %ld.%09ld inc\n",
494                                      inode->i_mtime.tv_sec,
495                                      inode->i_mtime.tv_nsec,
496                                      mtime->tv_sec, mtime->tv_nsec);
497                                 inode->i_mtime = *mtime;
498                         }
499                         if (timespec_compare(atime, &inode->i_atime) > 0) {
500                                 dout("atime %ld.%09ld -> %ld.%09ld inc\n",
501                                      inode->i_atime.tv_sec,
502                                      inode->i_atime.tv_nsec,
503                                      atime->tv_sec, atime->tv_nsec);
504                                 inode->i_atime = *atime;
505                         }
506                 } else if (issued & CEPH_CAP_FILE_EXCL) {
507                         /* we did a utimes(); ignore mds values */
508                 } else {
509                         warn = 1;
510                 }
511         } else {
512                 /* we have no write caps; whatever the MDS says is true */
513                 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) {
514                         inode->i_ctime = *ctime;
515                         inode->i_mtime = *mtime;
516                         inode->i_atime = *atime;
517                         ci->i_time_warp_seq = time_warp_seq;
518                 } else {
519                         warn = 1;
520                 }
521         }
522         if (warn) /* time_warp_seq shouldn't go backwards */
523                 dout("%p mds time_warp_seq %llu < %u\n",
524                      inode, time_warp_seq, ci->i_time_warp_seq);
525 }
526
527 /*
528  * Populate an inode based on info from mds.  May be called on new or
529  * existing inodes.
530  */
531 static int fill_inode(struct inode *inode,
532                       struct ceph_mds_reply_info_in *iinfo,
533                       struct ceph_mds_reply_dirfrag *dirinfo,
534                       struct ceph_mds_session *session,
535                       unsigned long ttl_from, int cap_fmode,
536                       struct ceph_cap_reservation *caps_reservation)
537 {
538         struct ceph_mds_reply_inode *info = iinfo->in;
539         struct ceph_inode_info *ci = ceph_inode(inode);
540         int i;
541         int issued, implemented;
542         struct timespec mtime, atime, ctime;
543         u32 nsplits;
544         struct ceph_buffer *xattr_blob = NULL;
545         int err = 0;
546         int queue_trunc = 0;
547
548         dout("fill_inode %p ino %llx.%llx v %llu had %llu\n",
549              inode, ceph_vinop(inode), le64_to_cpu(info->version),
550              ci->i_version);
551
552         /*
553          * prealloc xattr data, if it looks like we'll need it.  only
554          * if len > 4 (meaning there are actually xattrs; the first 4
555          * bytes are the xattr count).
556          */
557         if (iinfo->xattr_len > 4) {
558                 xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS);
559                 if (!xattr_blob)
560                         pr_err("fill_inode ENOMEM xattr blob %d bytes\n",
561                                iinfo->xattr_len);
562         }
563
564         spin_lock(&inode->i_lock);
565
566         /*
567          * provided version will be odd if inode value is projected,
568          * even if stable.  skip the update if we have a newer info
569          * (e.g., due to inode info racing form multiple MDSs), or if
570          * we are getting projected (unstable) inode info.
571          */
572         if (le64_to_cpu(info->version) > 0 &&
573             (ci->i_version & ~1) > le64_to_cpu(info->version))
574                 goto no_change;
575
576         issued = __ceph_caps_issued(ci, &implemented);
577         issued |= implemented | __ceph_caps_dirty(ci);
578
579         /* update inode */
580         ci->i_version = le64_to_cpu(info->version);
581         inode->i_version++;
582         inode->i_rdev = le32_to_cpu(info->rdev);
583
584         if ((issued & CEPH_CAP_AUTH_EXCL) == 0) {
585                 inode->i_mode = le32_to_cpu(info->mode);
586                 inode->i_uid = le32_to_cpu(info->uid);
587                 inode->i_gid = le32_to_cpu(info->gid);
588                 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
589                      inode->i_uid, inode->i_gid);
590         }
591
592         if ((issued & CEPH_CAP_LINK_EXCL) == 0)
593                 inode->i_nlink = le32_to_cpu(info->nlink);
594
595         /* be careful with mtime, atime, size */
596         ceph_decode_timespec(&atime, &info->atime);
597         ceph_decode_timespec(&mtime, &info->mtime);
598         ceph_decode_timespec(&ctime, &info->ctime);
599         queue_trunc = ceph_fill_file_size(inode, issued,
600                                           le32_to_cpu(info->truncate_seq),
601                                           le64_to_cpu(info->truncate_size),
602                                           le64_to_cpu(info->size));
603         ceph_fill_file_time(inode, issued,
604                             le32_to_cpu(info->time_warp_seq),
605                             &ctime, &mtime, &atime);
606
607         ci->i_max_size = le64_to_cpu(info->max_size);
608         ci->i_layout = info->layout;
609         inode->i_blkbits = fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1;
610
611         /* xattrs */
612         /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
613         if ((issued & CEPH_CAP_XATTR_EXCL) == 0 &&
614             le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) {
615                 if (ci->i_xattrs.blob)
616                         ceph_buffer_put(ci->i_xattrs.blob);
617                 ci->i_xattrs.blob = xattr_blob;
618                 if (xattr_blob)
619                         memcpy(ci->i_xattrs.blob->vec.iov_base,
620                                iinfo->xattr_data, iinfo->xattr_len);
621                 ci->i_xattrs.version = le64_to_cpu(info->xattr_version);
622         }
623
624         inode->i_mapping->a_ops = &ceph_aops;
625         inode->i_mapping->backing_dev_info =
626                 &ceph_client(inode->i_sb)->backing_dev_info;
627
628         switch (inode->i_mode & S_IFMT) {
629         case S_IFIFO:
630         case S_IFBLK:
631         case S_IFCHR:
632         case S_IFSOCK:
633                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
634                 inode->i_op = &ceph_file_iops;
635                 break;
636         case S_IFREG:
637                 inode->i_op = &ceph_file_iops;
638                 inode->i_fop = &ceph_file_fops;
639                 break;
640         case S_IFLNK:
641                 inode->i_op = &ceph_symlink_iops;
642                 if (!ci->i_symlink) {
643                         int symlen = iinfo->symlink_len;
644                         char *sym;
645
646                         BUG_ON(symlen != inode->i_size);
647                         spin_unlock(&inode->i_lock);
648
649                         err = -ENOMEM;
650                         sym = kmalloc(symlen+1, GFP_NOFS);
651                         if (!sym)
652                                 goto out;
653                         memcpy(sym, iinfo->symlink, symlen);
654                         sym[symlen] = 0;
655
656                         spin_lock(&inode->i_lock);
657                         if (!ci->i_symlink)
658                                 ci->i_symlink = sym;
659                         else
660                                 kfree(sym); /* lost a race */
661                 }
662                 break;
663         case S_IFDIR:
664                 inode->i_op = &ceph_dir_iops;
665                 inode->i_fop = &ceph_dir_fops;
666
667                 ci->i_files = le64_to_cpu(info->files);
668                 ci->i_subdirs = le64_to_cpu(info->subdirs);
669                 ci->i_rbytes = le64_to_cpu(info->rbytes);
670                 ci->i_rfiles = le64_to_cpu(info->rfiles);
671                 ci->i_rsubdirs = le64_to_cpu(info->rsubdirs);
672                 ceph_decode_timespec(&ci->i_rctime, &info->rctime);
673
674                 /* set dir completion flag? */
675                 if (ci->i_files == 0 && ci->i_subdirs == 0 &&
676                     ceph_snap(inode) == CEPH_NOSNAP &&
677                     (le32_to_cpu(info->cap.caps) & CEPH_CAP_FILE_SHARED)) {
678                         dout(" marking %p complete (empty)\n", inode);
679                         ci->i_ceph_flags |= CEPH_I_COMPLETE;
680                         ci->i_max_offset = 2;
681                 }
682
683                 /* it may be better to set st_size in getattr instead? */
684                 if (ceph_test_opt(ceph_client(inode->i_sb), RBYTES))
685                         inode->i_size = ci->i_rbytes;
686                 break;
687         default:
688                 pr_err("fill_inode %llx.%llx BAD mode 0%o\n",
689                        ceph_vinop(inode), inode->i_mode);
690         }
691
692 no_change:
693         spin_unlock(&inode->i_lock);
694
695         /* queue truncate if we saw i_size decrease */
696         if (queue_trunc)
697                 ceph_queue_vmtruncate(inode);
698
699         /* populate frag tree */
700         /* FIXME: move me up, if/when version reflects fragtree changes */
701         nsplits = le32_to_cpu(info->fragtree.nsplits);
702         mutex_lock(&ci->i_fragtree_mutex);
703         for (i = 0; i < nsplits; i++) {
704                 u32 id = le32_to_cpu(info->fragtree.splits[i].frag);
705                 struct ceph_inode_frag *frag = __get_or_create_frag(ci, id);
706
707                 if (IS_ERR(frag))
708                         continue;
709                 frag->split_by = le32_to_cpu(info->fragtree.splits[i].by);
710                 dout(" frag %x split by %d\n", frag->frag, frag->split_by);
711         }
712         mutex_unlock(&ci->i_fragtree_mutex);
713
714         /* were we issued a capability? */
715         if (info->cap.caps) {
716                 if (ceph_snap(inode) == CEPH_NOSNAP) {
717                         ceph_add_cap(inode, session,
718                                      le64_to_cpu(info->cap.cap_id),
719                                      cap_fmode,
720                                      le32_to_cpu(info->cap.caps),
721                                      le32_to_cpu(info->cap.wanted),
722                                      le32_to_cpu(info->cap.seq),
723                                      le32_to_cpu(info->cap.mseq),
724                                      le64_to_cpu(info->cap.realm),
725                                      info->cap.flags,
726                                      caps_reservation);
727                 } else {
728                         spin_lock(&inode->i_lock);
729                         dout(" %p got snap_caps %s\n", inode,
730                              ceph_cap_string(le32_to_cpu(info->cap.caps)));
731                         ci->i_snap_caps |= le32_to_cpu(info->cap.caps);
732                         if (cap_fmode >= 0)
733                                 __ceph_get_fmode(ci, cap_fmode);
734                         spin_unlock(&inode->i_lock);
735                 }
736         }
737
738         /* update delegation info? */
739         if (dirinfo)
740                 ceph_fill_dirfrag(inode, dirinfo);
741
742         err = 0;
743
744 out:
745         if (xattr_blob)
746                 ceph_buffer_put(xattr_blob);
747         return err;
748 }
749
750 /*
751  * caller should hold session s_mutex.
752  */
753 static void update_dentry_lease(struct dentry *dentry,
754                                 struct ceph_mds_reply_lease *lease,
755                                 struct ceph_mds_session *session,
756                                 unsigned long from_time)
757 {
758         struct ceph_dentry_info *di = ceph_dentry(dentry);
759         long unsigned duration = le32_to_cpu(lease->duration_ms);
760         long unsigned ttl = from_time + (duration * HZ) / 1000;
761         long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000;
762         struct inode *dir;
763
764         /* only track leases on regular dentries */
765         if (dentry->d_op != &ceph_dentry_ops)
766                 return;
767
768         spin_lock(&dentry->d_lock);
769         dout("update_dentry_lease %p mask %d duration %lu ms ttl %lu\n",
770              dentry, le16_to_cpu(lease->mask), duration, ttl);
771
772         /* make lease_rdcache_gen match directory */
773         dir = dentry->d_parent->d_inode;
774         di->lease_shared_gen = ceph_inode(dir)->i_shared_gen;
775
776         if (lease->mask == 0)
777                 goto out_unlock;
778
779         if (di->lease_gen == session->s_cap_gen &&
780             time_before(ttl, dentry->d_time))
781                 goto out_unlock;  /* we already have a newer lease. */
782
783         if (di->lease_session && di->lease_session != session)
784                 goto out_unlock;
785
786         ceph_dentry_lru_touch(dentry);
787
788         if (!di->lease_session)
789                 di->lease_session = ceph_get_mds_session(session);
790         di->lease_gen = session->s_cap_gen;
791         di->lease_seq = le32_to_cpu(lease->seq);
792         di->lease_renew_after = half_ttl;
793         di->lease_renew_from = 0;
794         dentry->d_time = ttl;
795 out_unlock:
796         spin_unlock(&dentry->d_lock);
797         return;
798 }
799
800 /*
801  * splice a dentry to an inode.
802  * caller must hold directory i_mutex for this to be safe.
803  *
804  * we will only rehash the resulting dentry if @prehash is
805  * true; @prehash will be set to false (for the benefit of
806  * the caller) if we fail.
807  */
808 static struct dentry *splice_dentry(struct dentry *dn, struct inode *in,
809                                     bool *prehash)
810 {
811         struct dentry *realdn;
812
813         /* dn must be unhashed */
814         if (!d_unhashed(dn))
815                 d_drop(dn);
816         realdn = d_materialise_unique(dn, in);
817         if (IS_ERR(realdn)) {
818                 pr_err("splice_dentry error %p inode %p ino %llx.%llx\n",
819                        dn, in, ceph_vinop(in));
820                 if (prehash)
821                         *prehash = false; /* don't rehash on error */
822                 dn = realdn; /* note realdn contains the error */
823                 goto out;
824         } else if (realdn) {
825                 dout("dn %p (%d) spliced with %p (%d) "
826                      "inode %p ino %llx.%llx\n",
827                      dn, atomic_read(&dn->d_count),
828                      realdn, atomic_read(&realdn->d_count),
829                      realdn->d_inode, ceph_vinop(realdn->d_inode));
830                 dput(dn);
831                 dn = realdn;
832         } else {
833                 BUG_ON(!ceph_dentry(dn));
834
835                 dout("dn %p attached to %p ino %llx.%llx\n",
836                      dn, dn->d_inode, ceph_vinop(dn->d_inode));
837         }
838         if ((!prehash || *prehash) && d_unhashed(dn))
839                 d_rehash(dn);
840 out:
841         return dn;
842 }
843
844 /*
845  * Set dentry's directory position based on the current dir's max, and
846  * order it in d_subdirs, so that dcache_readdir behaves.
847  */
848 static void ceph_set_dentry_offset(struct dentry *dn)
849 {
850         struct dentry *dir = dn->d_parent;
851         struct inode *inode = dn->d_parent->d_inode;
852         struct ceph_dentry_info *di;
853
854         BUG_ON(!inode);
855
856         di = ceph_dentry(dn);
857
858         spin_lock(&inode->i_lock);
859         di->offset = ceph_inode(inode)->i_max_offset++;
860         spin_unlock(&inode->i_lock);
861
862         spin_lock(&dcache_lock);
863         spin_lock(&dn->d_lock);
864         list_move_tail(&dir->d_subdirs, &dn->d_u.d_child);
865         dout("set_dentry_offset %p %lld (%p %p)\n", dn, di->offset,
866              dn->d_u.d_child.prev, dn->d_u.d_child.next);
867         spin_unlock(&dn->d_lock);
868         spin_unlock(&dcache_lock);
869 }
870
871 /*
872  * Incorporate results into the local cache.  This is either just
873  * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
874  * after a lookup).
875  *
876  * A reply may contain
877  *         a directory inode along with a dentry.
878  *  and/or a target inode
879  *
880  * Called with snap_rwsem (read).
881  */
882 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req,
883                     struct ceph_mds_session *session)
884 {
885         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
886         struct inode *in = NULL;
887         struct ceph_mds_reply_inode *ininfo;
888         struct ceph_vino vino;
889         int i = 0;
890         int err = 0;
891
892         dout("fill_trace %p is_dentry %d is_target %d\n", req,
893              rinfo->head->is_dentry, rinfo->head->is_target);
894
895 #if 0
896         /*
897          * Debugging hook:
898          *
899          * If we resend completed ops to a recovering mds, we get no
900          * trace.  Since that is very rare, pretend this is the case
901          * to ensure the 'no trace' handlers in the callers behave.
902          *
903          * Fill in inodes unconditionally to avoid breaking cap
904          * invariants.
905          */
906         if (rinfo->head->op & CEPH_MDS_OP_WRITE) {
907                 pr_info("fill_trace faking empty trace on %lld %s\n",
908                         req->r_tid, ceph_mds_op_name(rinfo->head->op));
909                 if (rinfo->head->is_dentry) {
910                         rinfo->head->is_dentry = 0;
911                         err = fill_inode(req->r_locked_dir,
912                                          &rinfo->diri, rinfo->dirfrag,
913                                          session, req->r_request_started, -1);
914                 }
915                 if (rinfo->head->is_target) {
916                         rinfo->head->is_target = 0;
917                         ininfo = rinfo->targeti.in;
918                         vino.ino = le64_to_cpu(ininfo->ino);
919                         vino.snap = le64_to_cpu(ininfo->snapid);
920                         in = ceph_get_inode(sb, vino);
921                         err = fill_inode(in, &rinfo->targeti, NULL,
922                                          session, req->r_request_started,
923                                          req->r_fmode);
924                         iput(in);
925                 }
926         }
927 #endif
928
929         if (!rinfo->head->is_target && !rinfo->head->is_dentry) {
930                 dout("fill_trace reply is empty!\n");
931                 if (rinfo->head->result == 0 && req->r_locked_dir) {
932                         struct ceph_inode_info *ci =
933                                 ceph_inode(req->r_locked_dir);
934                         dout(" clearing %p complete (empty trace)\n",
935                              req->r_locked_dir);
936                         ci->i_ceph_flags &= ~CEPH_I_COMPLETE;
937                         ci->i_release_count++;
938                 }
939                 return 0;
940         }
941
942         if (rinfo->head->is_dentry) {
943                 struct inode *dir = req->r_locked_dir;
944
945                 err = fill_inode(dir, &rinfo->diri, rinfo->dirfrag,
946                                  session, req->r_request_started, -1,
947                                  &req->r_caps_reservation);
948                 if (err < 0)
949                         return err;
950         }
951
952         if (rinfo->head->is_dentry && !req->r_aborted) {
953                 /*
954                  * lookup link rename   : null -> possibly existing inode
955                  * mknod symlink mkdir  : null -> new inode
956                  * unlink               : linked -> null
957                  */
958                 struct inode *dir = req->r_locked_dir;
959                 struct dentry *dn = req->r_dentry;
960                 bool have_dir_cap, have_lease;
961
962                 BUG_ON(!dn);
963                 BUG_ON(!dir);
964                 BUG_ON(dn->d_parent->d_inode != dir);
965                 BUG_ON(ceph_ino(dir) !=
966                        le64_to_cpu(rinfo->diri.in->ino));
967                 BUG_ON(ceph_snap(dir) !=
968                        le64_to_cpu(rinfo->diri.in->snapid));
969
970                 /* do we have a lease on the whole dir? */
971                 have_dir_cap =
972                         (le32_to_cpu(rinfo->diri.in->cap.caps) &
973                          CEPH_CAP_FILE_SHARED);
974
975                 /* do we have a dn lease? */
976                 have_lease = have_dir_cap ||
977                         (le16_to_cpu(rinfo->dlease->mask) &
978                          CEPH_LOCK_DN);
979
980                 if (!have_lease)
981                         dout("fill_trace  no dentry lease or dir cap\n");
982
983                 /* rename? */
984                 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) {
985                         dout(" src %p '%.*s' dst %p '%.*s'\n",
986                              req->r_old_dentry,
987                              req->r_old_dentry->d_name.len,
988                              req->r_old_dentry->d_name.name,
989                              dn, dn->d_name.len, dn->d_name.name);
990                         dout("fill_trace doing d_move %p -> %p\n",
991                              req->r_old_dentry, dn);
992                         d_move(req->r_old_dentry, dn);
993                         dout(" src %p '%.*s' dst %p '%.*s'\n",
994                              req->r_old_dentry,
995                              req->r_old_dentry->d_name.len,
996                              req->r_old_dentry->d_name.name,
997                              dn, dn->d_name.len, dn->d_name.name);
998                         /* ensure target dentry is invalidated, despite
999                            rehashing bug in vfs_rename_dir */
1000                         dn->d_time = jiffies;
1001                         ceph_dentry(dn)->lease_shared_gen = 0;
1002                         /* take overwritten dentry's readdir offset */
1003                         ceph_dentry(req->r_old_dentry)->offset =
1004                                 ceph_dentry(dn)->offset;
1005                         dn = req->r_old_dentry;  /* use old_dentry */
1006                         in = dn->d_inode;
1007                 }
1008
1009                 /* null dentry? */
1010                 if (!rinfo->head->is_target) {
1011                         dout("fill_trace null dentry\n");
1012                         if (dn->d_inode) {
1013                                 dout("d_delete %p\n", dn);
1014                                 d_delete(dn);
1015                         } else {
1016                                 dout("d_instantiate %p NULL\n", dn);
1017                                 d_instantiate(dn, NULL);
1018                                 if (have_lease && d_unhashed(dn))
1019                                         d_rehash(dn);
1020                                 update_dentry_lease(dn, rinfo->dlease,
1021                                                     session,
1022                                                     req->r_request_started);
1023                         }
1024                         goto done;
1025                 }
1026
1027                 /* attach proper inode */
1028                 ininfo = rinfo->targeti.in;
1029                 vino.ino = le64_to_cpu(ininfo->ino);
1030                 vino.snap = le64_to_cpu(ininfo->snapid);
1031                 if (!dn->d_inode) {
1032                         in = ceph_get_inode(sb, vino);
1033                         if (IS_ERR(in)) {
1034                                 pr_err("fill_trace bad get_inode "
1035                                        "%llx.%llx\n", vino.ino, vino.snap);
1036                                 err = PTR_ERR(in);
1037                                 d_delete(dn);
1038                                 goto done;
1039                         }
1040                         dn = splice_dentry(dn, in, &have_lease);
1041                         if (IS_ERR(dn)) {
1042                                 err = PTR_ERR(dn);
1043                                 goto done;
1044                         }
1045                         req->r_dentry = dn;  /* may have spliced */
1046                         ceph_set_dentry_offset(dn);
1047                         igrab(in);
1048                 } else if (ceph_ino(in) == vino.ino &&
1049                            ceph_snap(in) == vino.snap) {
1050                         igrab(in);
1051                 } else {
1052                         dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1053                              dn, in, ceph_ino(in), ceph_snap(in),
1054                              vino.ino, vino.snap);
1055                         have_lease = false;
1056                         in = NULL;
1057                 }
1058
1059                 if (have_lease)
1060                         update_dentry_lease(dn, rinfo->dlease, session,
1061                                             req->r_request_started);
1062                 dout(" final dn %p\n", dn);
1063                 i++;
1064         } else if (req->r_op == CEPH_MDS_OP_LOOKUPSNAP ||
1065                    req->r_op == CEPH_MDS_OP_MKSNAP) {
1066                 struct dentry *dn = req->r_dentry;
1067
1068                 /* fill out a snapdir LOOKUPSNAP dentry */
1069                 BUG_ON(!dn);
1070                 BUG_ON(!req->r_locked_dir);
1071                 BUG_ON(ceph_snap(req->r_locked_dir) != CEPH_SNAPDIR);
1072                 ininfo = rinfo->targeti.in;
1073                 vino.ino = le64_to_cpu(ininfo->ino);
1074                 vino.snap = le64_to_cpu(ininfo->snapid);
1075                 in = ceph_get_inode(sb, vino);
1076                 if (IS_ERR(in)) {
1077                         pr_err("fill_inode get_inode badness %llx.%llx\n",
1078                                vino.ino, vino.snap);
1079                         err = PTR_ERR(in);
1080                         d_delete(dn);
1081                         goto done;
1082                 }
1083                 dout(" linking snapped dir %p to dn %p\n", in, dn);
1084                 dn = splice_dentry(dn, in, NULL);
1085                 if (IS_ERR(dn)) {
1086                         err = PTR_ERR(dn);
1087                         goto done;
1088                 }
1089                 ceph_set_dentry_offset(dn);
1090                 req->r_dentry = dn;  /* may have spliced */
1091                 igrab(in);
1092                 rinfo->head->is_dentry = 1;  /* fool notrace handlers */
1093         }
1094
1095         if (rinfo->head->is_target) {
1096                 vino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1097                 vino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1098
1099                 if (in == NULL || ceph_ino(in) != vino.ino ||
1100                     ceph_snap(in) != vino.snap) {
1101                         in = ceph_get_inode(sb, vino);
1102                         if (IS_ERR(in)) {
1103                                 err = PTR_ERR(in);
1104                                 goto done;
1105                         }
1106                 }
1107                 req->r_target_inode = in;
1108
1109                 err = fill_inode(in,
1110                                  &rinfo->targeti, NULL,
1111                                  session, req->r_request_started,
1112                                  (le32_to_cpu(rinfo->head->result) == 0) ?
1113                                  req->r_fmode : -1,
1114                                  &req->r_caps_reservation);
1115                 if (err < 0) {
1116                         pr_err("fill_inode badness %p %llx.%llx\n",
1117                                in, ceph_vinop(in));
1118                         goto done;
1119                 }
1120         }
1121
1122 done:
1123         dout("fill_trace done err=%d\n", err);
1124         return err;
1125 }
1126
1127 /*
1128  * Prepopulate our cache with readdir results, leases, etc.
1129  */
1130 int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1131                              struct ceph_mds_session *session)
1132 {
1133         struct dentry *parent = req->r_dentry;
1134         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1135         struct qstr dname;
1136         struct dentry *dn;
1137         struct inode *in;
1138         int err = 0, i;
1139         struct inode *snapdir = NULL;
1140         struct ceph_mds_request_head *rhead = req->r_request->front.iov_base;
1141         u64 frag = le32_to_cpu(rhead->args.readdir.frag);
1142         struct ceph_dentry_info *di;
1143
1144         if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
1145                 snapdir = ceph_get_snapdir(parent->d_inode);
1146                 parent = d_find_alias(snapdir);
1147                 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1148                      rinfo->dir_nr, parent);
1149         } else {
1150                 dout("readdir_prepopulate %d items under dn %p\n",
1151                      rinfo->dir_nr, parent);
1152                 if (rinfo->dir_dir)
1153                         ceph_fill_dirfrag(parent->d_inode, rinfo->dir_dir);
1154         }
1155
1156         for (i = 0; i < rinfo->dir_nr; i++) {
1157                 struct ceph_vino vino;
1158
1159                 dname.name = rinfo->dir_dname[i];
1160                 dname.len = rinfo->dir_dname_len[i];
1161                 dname.hash = full_name_hash(dname.name, dname.len);
1162
1163                 vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino);
1164                 vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid);
1165
1166 retry_lookup:
1167                 dn = d_lookup(parent, &dname);
1168                 dout("d_lookup on parent=%p name=%.*s got %p\n",
1169                      parent, dname.len, dname.name, dn);
1170
1171                 if (!dn) {
1172                         dn = d_alloc(parent, &dname);
1173                         dout("d_alloc %p '%.*s' = %p\n", parent,
1174                              dname.len, dname.name, dn);
1175                         if (dn == NULL) {
1176                                 dout("d_alloc badness\n");
1177                                 err = -ENOMEM;
1178                                 goto out;
1179                         }
1180                         err = ceph_init_dentry(dn);
1181                         if (err < 0)
1182                                 goto out;
1183                 } else if (dn->d_inode &&
1184                            (ceph_ino(dn->d_inode) != vino.ino ||
1185                             ceph_snap(dn->d_inode) != vino.snap)) {
1186                         dout(" dn %p points to wrong inode %p\n",
1187                              dn, dn->d_inode);
1188                         d_delete(dn);
1189                         dput(dn);
1190                         goto retry_lookup;
1191                 } else {
1192                         /* reorder parent's d_subdirs */
1193                         spin_lock(&dcache_lock);
1194                         spin_lock(&dn->d_lock);
1195                         list_move(&dn->d_u.d_child, &parent->d_subdirs);
1196                         spin_unlock(&dn->d_lock);
1197                         spin_unlock(&dcache_lock);
1198                 }
1199
1200                 di = dn->d_fsdata;
1201                 di->offset = ceph_make_fpos(frag, i + req->r_readdir_offset);
1202
1203                 /* inode */
1204                 if (dn->d_inode) {
1205                         in = dn->d_inode;
1206                 } else {
1207                         in = ceph_get_inode(parent->d_sb, vino);
1208                         if (in == NULL) {
1209                                 dout("new_inode badness\n");
1210                                 d_delete(dn);
1211                                 dput(dn);
1212                                 err = -ENOMEM;
1213                                 goto out;
1214                         }
1215                         dn = splice_dentry(dn, in, NULL);
1216                 }
1217
1218                 if (fill_inode(in, &rinfo->dir_in[i], NULL, session,
1219                                req->r_request_started, -1,
1220                                &req->r_caps_reservation) < 0) {
1221                         pr_err("fill_inode badness on %p\n", in);
1222                         dput(dn);
1223                         continue;
1224                 }
1225                 update_dentry_lease(dn, rinfo->dir_dlease[i],
1226                                     req->r_session, req->r_request_started);
1227                 dput(dn);
1228         }
1229         req->r_did_prepopulate = true;
1230
1231 out:
1232         if (snapdir) {
1233                 iput(snapdir);
1234                 dput(parent);
1235         }
1236         dout("readdir_prepopulate done\n");
1237         return err;
1238 }
1239
1240 int ceph_inode_set_size(struct inode *inode, loff_t size)
1241 {
1242         struct ceph_inode_info *ci = ceph_inode(inode);
1243         int ret = 0;
1244
1245         spin_lock(&inode->i_lock);
1246         dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size);
1247         inode->i_size = size;
1248         inode->i_blocks = (size + (1 << 9) - 1) >> 9;
1249
1250         /* tell the MDS if we are approaching max_size */
1251         if ((size << 1) >= ci->i_max_size &&
1252             (ci->i_reported_size << 1) < ci->i_max_size)
1253                 ret = 1;
1254
1255         spin_unlock(&inode->i_lock);
1256         return ret;
1257 }
1258
1259 /*
1260  * Write back inode data in a worker thread.  (This can't be done
1261  * in the message handler context.)
1262  */
1263 void ceph_queue_writeback(struct inode *inode)
1264 {
1265         if (queue_work(ceph_inode_to_client(inode)->wb_wq,
1266                        &ceph_inode(inode)->i_wb_work)) {
1267                 dout("ceph_queue_writeback %p\n", inode);
1268                 igrab(inode);
1269         } else {
1270                 dout("ceph_queue_writeback %p failed\n", inode);
1271         }
1272 }
1273
1274 static void ceph_writeback_work(struct work_struct *work)
1275 {
1276         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1277                                                   i_wb_work);
1278         struct inode *inode = &ci->vfs_inode;
1279
1280         dout("writeback %p\n", inode);
1281         filemap_fdatawrite(&inode->i_data);
1282         iput(inode);
1283 }
1284
1285 /*
1286  * queue an async invalidation
1287  */
1288 void ceph_queue_invalidate(struct inode *inode)
1289 {
1290         if (queue_work(ceph_inode_to_client(inode)->pg_inv_wq,
1291                        &ceph_inode(inode)->i_pg_inv_work)) {
1292                 dout("ceph_queue_invalidate %p\n", inode);
1293                 igrab(inode);
1294         } else {
1295                 dout("ceph_queue_invalidate %p failed\n", inode);
1296         }
1297 }
1298
1299 /*
1300  * invalidate any pages that are not dirty or under writeback.  this
1301  * includes pages that are clean and mapped.
1302  */
1303 static void ceph_invalidate_nondirty_pages(struct address_space *mapping)
1304 {
1305         struct pagevec pvec;
1306         pgoff_t next = 0;
1307         int i;
1308
1309         pagevec_init(&pvec, 0);
1310         while (pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
1311                 for (i = 0; i < pagevec_count(&pvec); i++) {
1312                         struct page *page = pvec.pages[i];
1313                         pgoff_t index;
1314                         int skip_page =
1315                                 (PageDirty(page) || PageWriteback(page));
1316
1317                         if (!skip_page)
1318                                 skip_page = !trylock_page(page);
1319
1320                         /*
1321                          * We really shouldn't be looking at the ->index of an
1322                          * unlocked page.  But we're not allowed to lock these
1323                          * pages.  So we rely upon nobody altering the ->index
1324                          * of this (pinned-by-us) page.
1325                          */
1326                         index = page->index;
1327                         if (index > next)
1328                                 next = index;
1329                         next++;
1330
1331                         if (skip_page)
1332                                 continue;
1333
1334                         generic_error_remove_page(mapping, page);
1335                         unlock_page(page);
1336                 }
1337                 pagevec_release(&pvec);
1338                 cond_resched();
1339         }
1340 }
1341
1342 /*
1343  * Invalidate inode pages in a worker thread.  (This can't be done
1344  * in the message handler context.)
1345  */
1346 static void ceph_invalidate_work(struct work_struct *work)
1347 {
1348         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1349                                                   i_pg_inv_work);
1350         struct inode *inode = &ci->vfs_inode;
1351         u32 orig_gen;
1352         int check = 0;
1353
1354         spin_lock(&inode->i_lock);
1355         dout("invalidate_pages %p gen %d revoking %d\n", inode,
1356              ci->i_rdcache_gen, ci->i_rdcache_revoking);
1357         if (ci->i_rdcache_gen == 0 ||
1358             ci->i_rdcache_revoking != ci->i_rdcache_gen) {
1359                 BUG_ON(ci->i_rdcache_revoking > ci->i_rdcache_gen);
1360                 /* nevermind! */
1361                 ci->i_rdcache_revoking = 0;
1362                 spin_unlock(&inode->i_lock);
1363                 goto out;
1364         }
1365         orig_gen = ci->i_rdcache_gen;
1366         spin_unlock(&inode->i_lock);
1367
1368         ceph_invalidate_nondirty_pages(inode->i_mapping);
1369
1370         spin_lock(&inode->i_lock);
1371         if (orig_gen == ci->i_rdcache_gen) {
1372                 dout("invalidate_pages %p gen %d successful\n", inode,
1373                      ci->i_rdcache_gen);
1374                 ci->i_rdcache_gen = 0;
1375                 ci->i_rdcache_revoking = 0;
1376                 check = 1;
1377         } else {
1378                 dout("invalidate_pages %p gen %d raced, gen now %d\n",
1379                      inode, orig_gen, ci->i_rdcache_gen);
1380         }
1381         spin_unlock(&inode->i_lock);
1382
1383         if (check)
1384                 ceph_check_caps(ci, 0, NULL);
1385 out:
1386         iput(inode);
1387 }
1388
1389
1390 /*
1391  * called by trunc_wq; take i_mutex ourselves
1392  *
1393  * We also truncate in a separate thread as well.
1394  */
1395 static void ceph_vmtruncate_work(struct work_struct *work)
1396 {
1397         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1398                                                   i_vmtruncate_work);
1399         struct inode *inode = &ci->vfs_inode;
1400
1401         dout("vmtruncate_work %p\n", inode);
1402         mutex_lock(&inode->i_mutex);
1403         __ceph_do_pending_vmtruncate(inode);
1404         mutex_unlock(&inode->i_mutex);
1405         iput(inode);
1406 }
1407
1408 /*
1409  * Queue an async vmtruncate.  If we fail to queue work, we will handle
1410  * the truncation the next time we call __ceph_do_pending_vmtruncate.
1411  */
1412 void ceph_queue_vmtruncate(struct inode *inode)
1413 {
1414         struct ceph_inode_info *ci = ceph_inode(inode);
1415
1416         if (queue_work(ceph_client(inode->i_sb)->trunc_wq,
1417                        &ci->i_vmtruncate_work)) {
1418                 dout("ceph_queue_vmtruncate %p\n", inode);
1419                 igrab(inode);
1420         } else {
1421                 dout("ceph_queue_vmtruncate %p failed, pending=%d\n",
1422                      inode, ci->i_truncate_pending);
1423         }
1424 }
1425
1426 /*
1427  * called with i_mutex held.
1428  *
1429  * Make sure any pending truncation is applied before doing anything
1430  * that may depend on it.
1431  */
1432 void __ceph_do_pending_vmtruncate(struct inode *inode)
1433 {
1434         struct ceph_inode_info *ci = ceph_inode(inode);
1435         u64 to;
1436         int wrbuffer_refs, wake = 0;
1437
1438 retry:
1439         spin_lock(&inode->i_lock);
1440         if (ci->i_truncate_pending == 0) {
1441                 dout("__do_pending_vmtruncate %p none pending\n", inode);
1442                 spin_unlock(&inode->i_lock);
1443                 return;
1444         }
1445
1446         /*
1447          * make sure any dirty snapped pages are flushed before we
1448          * possibly truncate them.. so write AND block!
1449          */
1450         if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) {
1451                 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1452                      inode);
1453                 spin_unlock(&inode->i_lock);
1454                 filemap_write_and_wait_range(&inode->i_data, 0,
1455                                              inode->i_sb->s_maxbytes);
1456                 goto retry;
1457         }
1458
1459         to = ci->i_truncate_size;
1460         wrbuffer_refs = ci->i_wrbuffer_ref;
1461         dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode,
1462              ci->i_truncate_pending, to);
1463         spin_unlock(&inode->i_lock);
1464
1465         truncate_inode_pages(inode->i_mapping, to);
1466
1467         spin_lock(&inode->i_lock);
1468         ci->i_truncate_pending--;
1469         if (ci->i_truncate_pending == 0)
1470                 wake = 1;
1471         spin_unlock(&inode->i_lock);
1472
1473         if (wrbuffer_refs == 0)
1474                 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
1475         if (wake)
1476                 wake_up(&ci->i_cap_wq);
1477 }
1478
1479
1480 /*
1481  * symlinks
1482  */
1483 static void *ceph_sym_follow_link(struct dentry *dentry, struct nameidata *nd)
1484 {
1485         struct ceph_inode_info *ci = ceph_inode(dentry->d_inode);
1486         nd_set_link(nd, ci->i_symlink);
1487         return NULL;
1488 }
1489
1490 static const struct inode_operations ceph_symlink_iops = {
1491         .readlink = generic_readlink,
1492         .follow_link = ceph_sym_follow_link,
1493 };
1494
1495 /*
1496  * setattr
1497  */
1498 int ceph_setattr(struct dentry *dentry, struct iattr *attr)
1499 {
1500         struct inode *inode = dentry->d_inode;
1501         struct ceph_inode_info *ci = ceph_inode(inode);
1502         struct inode *parent_inode = dentry->d_parent->d_inode;
1503         const unsigned int ia_valid = attr->ia_valid;
1504         struct ceph_mds_request *req;
1505         struct ceph_mds_client *mdsc = &ceph_client(dentry->d_sb)->mdsc;
1506         int issued;
1507         int release = 0, dirtied = 0;
1508         int mask = 0;
1509         int err = 0;
1510
1511         if (ceph_snap(inode) != CEPH_NOSNAP)
1512                 return -EROFS;
1513
1514         __ceph_do_pending_vmtruncate(inode);
1515
1516         err = inode_change_ok(inode, attr);
1517         if (err != 0)
1518                 return err;
1519
1520         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR,
1521                                        USE_AUTH_MDS);
1522         if (IS_ERR(req))
1523                 return PTR_ERR(req);
1524
1525         spin_lock(&inode->i_lock);
1526         issued = __ceph_caps_issued(ci, NULL);
1527         dout("setattr %p issued %s\n", inode, ceph_cap_string(issued));
1528
1529         if (ia_valid & ATTR_UID) {
1530                 dout("setattr %p uid %d -> %d\n", inode,
1531                      inode->i_uid, attr->ia_uid);
1532                 if (issued & CEPH_CAP_AUTH_EXCL) {
1533                         inode->i_uid = attr->ia_uid;
1534                         dirtied |= CEPH_CAP_AUTH_EXCL;
1535                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1536                            attr->ia_uid != inode->i_uid) {
1537                         req->r_args.setattr.uid = cpu_to_le32(attr->ia_uid);
1538                         mask |= CEPH_SETATTR_UID;
1539                         release |= CEPH_CAP_AUTH_SHARED;
1540                 }
1541         }
1542         if (ia_valid & ATTR_GID) {
1543                 dout("setattr %p gid %d -> %d\n", inode,
1544                      inode->i_gid, attr->ia_gid);
1545                 if (issued & CEPH_CAP_AUTH_EXCL) {
1546                         inode->i_gid = attr->ia_gid;
1547                         dirtied |= CEPH_CAP_AUTH_EXCL;
1548                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1549                            attr->ia_gid != inode->i_gid) {
1550                         req->r_args.setattr.gid = cpu_to_le32(attr->ia_gid);
1551                         mask |= CEPH_SETATTR_GID;
1552                         release |= CEPH_CAP_AUTH_SHARED;
1553                 }
1554         }
1555         if (ia_valid & ATTR_MODE) {
1556                 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode,
1557                      attr->ia_mode);
1558                 if (issued & CEPH_CAP_AUTH_EXCL) {
1559                         inode->i_mode = attr->ia_mode;
1560                         dirtied |= CEPH_CAP_AUTH_EXCL;
1561                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1562                            attr->ia_mode != inode->i_mode) {
1563                         req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode);
1564                         mask |= CEPH_SETATTR_MODE;
1565                         release |= CEPH_CAP_AUTH_SHARED;
1566                 }
1567         }
1568
1569         if (ia_valid & ATTR_ATIME) {
1570                 dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode,
1571                      inode->i_atime.tv_sec, inode->i_atime.tv_nsec,
1572                      attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec);
1573                 if (issued & CEPH_CAP_FILE_EXCL) {
1574                         ci->i_time_warp_seq++;
1575                         inode->i_atime = attr->ia_atime;
1576                         dirtied |= CEPH_CAP_FILE_EXCL;
1577                 } else if ((issued & CEPH_CAP_FILE_WR) &&
1578                            timespec_compare(&inode->i_atime,
1579                                             &attr->ia_atime) < 0) {
1580                         inode->i_atime = attr->ia_atime;
1581                         dirtied |= CEPH_CAP_FILE_WR;
1582                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1583                            !timespec_equal(&inode->i_atime, &attr->ia_atime)) {
1584                         ceph_encode_timespec(&req->r_args.setattr.atime,
1585                                              &attr->ia_atime);
1586                         mask |= CEPH_SETATTR_ATIME;
1587                         release |= CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_RD |
1588                                 CEPH_CAP_FILE_WR;
1589                 }
1590         }
1591         if (ia_valid & ATTR_MTIME) {
1592                 dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode,
1593                      inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
1594                      attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec);
1595                 if (issued & CEPH_CAP_FILE_EXCL) {
1596                         ci->i_time_warp_seq++;
1597                         inode->i_mtime = attr->ia_mtime;
1598                         dirtied |= CEPH_CAP_FILE_EXCL;
1599                 } else if ((issued & CEPH_CAP_FILE_WR) &&
1600                            timespec_compare(&inode->i_mtime,
1601                                             &attr->ia_mtime) < 0) {
1602                         inode->i_mtime = attr->ia_mtime;
1603                         dirtied |= CEPH_CAP_FILE_WR;
1604                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1605                            !timespec_equal(&inode->i_mtime, &attr->ia_mtime)) {
1606                         ceph_encode_timespec(&req->r_args.setattr.mtime,
1607                                              &attr->ia_mtime);
1608                         mask |= CEPH_SETATTR_MTIME;
1609                         release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
1610                                 CEPH_CAP_FILE_WR;
1611                 }
1612         }
1613         if (ia_valid & ATTR_SIZE) {
1614                 dout("setattr %p size %lld -> %lld\n", inode,
1615                      inode->i_size, attr->ia_size);
1616                 if (attr->ia_size > inode->i_sb->s_maxbytes) {
1617                         err = -EINVAL;
1618                         goto out;
1619                 }
1620                 if ((issued & CEPH_CAP_FILE_EXCL) &&
1621                     attr->ia_size > inode->i_size) {
1622                         inode->i_size = attr->ia_size;
1623                         inode->i_blocks =
1624                                 (attr->ia_size + (1 << 9) - 1) >> 9;
1625                         inode->i_ctime = attr->ia_ctime;
1626                         ci->i_reported_size = attr->ia_size;
1627                         dirtied |= CEPH_CAP_FILE_EXCL;
1628                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1629                            attr->ia_size != inode->i_size) {
1630                         req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
1631                         req->r_args.setattr.old_size =
1632                                 cpu_to_le64(inode->i_size);
1633                         mask |= CEPH_SETATTR_SIZE;
1634                         release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
1635                                 CEPH_CAP_FILE_WR;
1636                 }
1637         }
1638
1639         /* these do nothing */
1640         if (ia_valid & ATTR_CTIME) {
1641                 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME|
1642                                          ATTR_MODE|ATTR_UID|ATTR_GID)) == 0;
1643                 dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode,
1644                      inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
1645                      attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec,
1646                      only ? "ctime only" : "ignored");
1647                 inode->i_ctime = attr->ia_ctime;
1648                 if (only) {
1649                         /*
1650                          * if kernel wants to dirty ctime but nothing else,
1651                          * we need to choose a cap to dirty under, or do
1652                          * a almost-no-op setattr
1653                          */
1654                         if (issued & CEPH_CAP_AUTH_EXCL)
1655                                 dirtied |= CEPH_CAP_AUTH_EXCL;
1656                         else if (issued & CEPH_CAP_FILE_EXCL)
1657                                 dirtied |= CEPH_CAP_FILE_EXCL;
1658                         else if (issued & CEPH_CAP_XATTR_EXCL)
1659                                 dirtied |= CEPH_CAP_XATTR_EXCL;
1660                         else
1661                                 mask |= CEPH_SETATTR_CTIME;
1662                 }
1663         }
1664         if (ia_valid & ATTR_FILE)
1665                 dout("setattr %p ATTR_FILE ... hrm!\n", inode);
1666
1667         if (dirtied) {
1668                 __ceph_mark_dirty_caps(ci, dirtied);
1669                 inode->i_ctime = CURRENT_TIME;
1670         }
1671
1672         release &= issued;
1673         spin_unlock(&inode->i_lock);
1674
1675         if (mask) {
1676                 req->r_inode = igrab(inode);
1677                 req->r_inode_drop = release;
1678                 req->r_args.setattr.mask = cpu_to_le32(mask);
1679                 req->r_num_caps = 1;
1680                 err = ceph_mdsc_do_request(mdsc, parent_inode, req);
1681         }
1682         dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err,
1683              ceph_cap_string(dirtied), mask);
1684
1685         ceph_mdsc_put_request(req);
1686         __ceph_do_pending_vmtruncate(inode);
1687         return err;
1688 out:
1689         spin_unlock(&inode->i_lock);
1690         ceph_mdsc_put_request(req);
1691         return err;
1692 }
1693
1694 /*
1695  * Verify that we have a lease on the given mask.  If not,
1696  * do a getattr against an mds.
1697  */
1698 int ceph_do_getattr(struct inode *inode, int mask)
1699 {
1700         struct ceph_client *client = ceph_sb_to_client(inode->i_sb);
1701         struct ceph_mds_client *mdsc = &client->mdsc;
1702         struct ceph_mds_request *req;
1703         int err;
1704
1705         if (ceph_snap(inode) == CEPH_SNAPDIR) {
1706                 dout("do_getattr inode %p SNAPDIR\n", inode);
1707                 return 0;
1708         }
1709
1710         dout("do_getattr inode %p mask %s\n", inode, ceph_cap_string(mask));
1711         if (ceph_caps_issued_mask(ceph_inode(inode), mask, 1))
1712                 return 0;
1713
1714         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
1715         if (IS_ERR(req))
1716                 return PTR_ERR(req);
1717         req->r_inode = igrab(inode);
1718         req->r_num_caps = 1;
1719         req->r_args.getattr.mask = cpu_to_le32(mask);
1720         err = ceph_mdsc_do_request(mdsc, NULL, req);
1721         ceph_mdsc_put_request(req);
1722         dout("do_getattr result=%d\n", err);
1723         return err;
1724 }
1725
1726
1727 /*
1728  * Check inode permissions.  We verify we have a valid value for
1729  * the AUTH cap, then call the generic handler.
1730  */
1731 int ceph_permission(struct inode *inode, int mask)
1732 {
1733         int err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED);
1734
1735         if (!err)
1736                 err = generic_permission(inode, mask, NULL);
1737         return err;
1738 }
1739
1740 /*
1741  * Get all attributes.  Hopefully somedata we'll have a statlite()
1742  * and can limit the fields we require to be accurate.
1743  */
1744 int ceph_getattr(struct vfsmount *mnt, struct dentry *dentry,
1745                  struct kstat *stat)
1746 {
1747         struct inode *inode = dentry->d_inode;
1748         struct ceph_inode_info *ci = ceph_inode(inode);
1749         int err;
1750
1751         err = ceph_do_getattr(inode, CEPH_STAT_CAP_INODE_ALL);
1752         if (!err) {
1753                 generic_fillattr(inode, stat);
1754                 stat->ino = inode->i_ino;
1755                 if (ceph_snap(inode) != CEPH_NOSNAP)
1756                         stat->dev = ceph_snap(inode);
1757                 else
1758                         stat->dev = 0;
1759                 if (S_ISDIR(inode->i_mode)) {
1760                         stat->size = ci->i_rbytes;
1761                         stat->blocks = 0;
1762                         stat->blksize = 65536;
1763                 }
1764         }
1765         return err;
1766 }