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