blob: f3a2abf28a77df362faf5c38dc471a64dcbfdffc [file] [log] [blame]
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001#include <linux/ceph/ceph_debug.h>
Sage Weil355da1e2009-10-06 11:31:08 -07002
3#include <linux/module.h>
4#include <linux/fs.h>
Sage Weil355da1e2009-10-06 11:31:08 -07005#include <linux/slab.h>
6#include <linux/string.h>
7#include <linux/uaccess.h>
8#include <linux/kernel.h>
9#include <linux/namei.h>
10#include <linux/writeback.h>
11#include <linux/vmalloc.h>
12
13#include "super.h"
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -070014#include "mds_client.h"
15#include <linux/ceph/decode.h>
Sage Weil355da1e2009-10-06 11:31:08 -070016
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
29static const struct inode_operations ceph_symlink_iops;
30
Sage Weil3c6f6b72010-02-09 15:24:44 -080031static void ceph_invalidate_work(struct work_struct *work);
32static void ceph_writeback_work(struct work_struct *work);
33static void ceph_vmtruncate_work(struct work_struct *work);
Sage Weil355da1e2009-10-06 11:31:08 -070034
35/*
36 * find or create an inode, given the ceph ino number
37 */
Yehuda Sadehad1fee92011-01-21 16:44:03 -080038static int ceph_set_ino_cb(struct inode *inode, void *data)
39{
40 ceph_inode(inode)->i_vino = *(struct ceph_vino *)data;
41 inode->i_ino = ceph_vino_to_ino(*(struct ceph_vino *)data);
42 return 0;
43}
44
Sage Weil355da1e2009-10-06 11:31:08 -070045struct inode *ceph_get_inode(struct super_block *sb, struct ceph_vino vino)
46{
47 struct inode *inode;
48 ino_t t = ceph_vino_to_ino(vino);
49
50 inode = iget5_locked(sb, t, ceph_ino_compare, ceph_set_ino_cb, &vino);
51 if (inode == NULL)
52 return ERR_PTR(-ENOMEM);
53 if (inode->i_state & I_NEW) {
54 dout("get_inode created new inode %p %llx.%llx ino %llx\n",
55 inode, ceph_vinop(inode), (u64)inode->i_ino);
56 unlock_new_inode(inode);
57 }
58
59 dout("get_inode on %lu=%llx.%llx got %p\n", inode->i_ino, vino.ino,
60 vino.snap, inode);
61 return inode;
62}
63
64/*
65 * get/constuct snapdir inode for a given directory
66 */
67struct inode *ceph_get_snapdir(struct inode *parent)
68{
69 struct ceph_vino vino = {
70 .ino = ceph_ino(parent),
71 .snap = CEPH_SNAPDIR,
72 };
73 struct inode *inode = ceph_get_inode(parent->i_sb, vino);
Sage Weilb377ff12009-11-11 15:22:37 -080074 struct ceph_inode_info *ci = ceph_inode(inode);
Sage Weil355da1e2009-10-06 11:31:08 -070075
76 BUG_ON(!S_ISDIR(parent->i_mode));
77 if (IS_ERR(inode))
Julia Lawall7e34bc52010-05-22 12:01:14 +020078 return inode;
Sage Weil355da1e2009-10-06 11:31:08 -070079 inode->i_mode = parent->i_mode;
80 inode->i_uid = parent->i_uid;
81 inode->i_gid = parent->i_gid;
82 inode->i_op = &ceph_dir_iops;
83 inode->i_fop = &ceph_dir_fops;
Sage Weilb377ff12009-11-11 15:22:37 -080084 ci->i_snap_caps = CEPH_CAP_PIN; /* so we can open */
85 ci->i_rbytes = 0;
Sage Weil355da1e2009-10-06 11:31:08 -070086 return inode;
87}
88
89const struct inode_operations ceph_file_iops = {
90 .permission = ceph_permission,
91 .setattr = ceph_setattr,
92 .getattr = ceph_getattr,
93 .setxattr = ceph_setxattr,
94 .getxattr = ceph_getxattr,
95 .listxattr = ceph_listxattr,
96 .removexattr = ceph_removexattr,
97};
98
99
100/*
101 * We use a 'frag tree' to keep track of the MDS's directory fragments
102 * for a given inode (usually there is just a single fragment). We
103 * need to know when a child frag is delegated to a new MDS, or when
104 * it is flagged as replicated, so we can direct our requests
105 * accordingly.
106 */
107
108/*
109 * find/create a frag in the tree
110 */
111static struct ceph_inode_frag *__get_or_create_frag(struct ceph_inode_info *ci,
112 u32 f)
113{
114 struct rb_node **p;
115 struct rb_node *parent = NULL;
116 struct ceph_inode_frag *frag;
117 int c;
118
119 p = &ci->i_fragtree.rb_node;
120 while (*p) {
121 parent = *p;
122 frag = rb_entry(parent, struct ceph_inode_frag, node);
123 c = ceph_frag_compare(f, frag->frag);
124 if (c < 0)
125 p = &(*p)->rb_left;
126 else if (c > 0)
127 p = &(*p)->rb_right;
128 else
129 return frag;
130 }
131
132 frag = kmalloc(sizeof(*frag), GFP_NOFS);
133 if (!frag) {
134 pr_err("__get_or_create_frag ENOMEM on %p %llx.%llx "
135 "frag %x\n", &ci->vfs_inode,
136 ceph_vinop(&ci->vfs_inode), f);
137 return ERR_PTR(-ENOMEM);
138 }
139 frag->frag = f;
140 frag->split_by = 0;
141 frag->mds = -1;
142 frag->ndist = 0;
143
144 rb_link_node(&frag->node, parent, p);
145 rb_insert_color(&frag->node, &ci->i_fragtree);
146
147 dout("get_or_create_frag added %llx.%llx frag %x\n",
148 ceph_vinop(&ci->vfs_inode), f);
149 return frag;
150}
151
152/*
153 * find a specific frag @f
154 */
155struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci, u32 f)
156{
157 struct rb_node *n = ci->i_fragtree.rb_node;
158
159 while (n) {
160 struct ceph_inode_frag *frag =
161 rb_entry(n, struct ceph_inode_frag, node);
162 int c = ceph_frag_compare(f, frag->frag);
163 if (c < 0)
164 n = n->rb_left;
165 else if (c > 0)
166 n = n->rb_right;
167 else
168 return frag;
169 }
170 return NULL;
171}
172
173/*
174 * Choose frag containing the given value @v. If @pfrag is
175 * specified, copy the frag delegation info to the caller if
176 * it is present.
177 */
178u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
179 struct ceph_inode_frag *pfrag,
180 int *found)
181{
182 u32 t = ceph_frag_make(0, 0);
183 struct ceph_inode_frag *frag;
184 unsigned nway, i;
185 u32 n;
186
187 if (found)
188 *found = 0;
189
190 mutex_lock(&ci->i_fragtree_mutex);
191 while (1) {
192 WARN_ON(!ceph_frag_contains_value(t, v));
193 frag = __ceph_find_frag(ci, t);
194 if (!frag)
195 break; /* t is a leaf */
196 if (frag->split_by == 0) {
197 if (pfrag)
198 memcpy(pfrag, frag, sizeof(*pfrag));
199 if (found)
200 *found = 1;
201 break;
202 }
203
204 /* choose child */
205 nway = 1 << frag->split_by;
206 dout("choose_frag(%x) %x splits by %d (%d ways)\n", v, t,
207 frag->split_by, nway);
208 for (i = 0; i < nway; i++) {
209 n = ceph_frag_make_child(t, frag->split_by, i);
210 if (ceph_frag_contains_value(n, v)) {
211 t = n;
212 break;
213 }
214 }
215 BUG_ON(i == nway);
216 }
217 dout("choose_frag(%x) = %x\n", v, t);
218
219 mutex_unlock(&ci->i_fragtree_mutex);
220 return t;
221}
222
223/*
224 * Process dirfrag (delegation) info from the mds. Include leaf
225 * fragment in tree ONLY if ndist > 0. Otherwise, only
226 * branches/splits are included in i_fragtree)
227 */
228static int ceph_fill_dirfrag(struct inode *inode,
229 struct ceph_mds_reply_dirfrag *dirinfo)
230{
231 struct ceph_inode_info *ci = ceph_inode(inode);
232 struct ceph_inode_frag *frag;
233 u32 id = le32_to_cpu(dirinfo->frag);
234 int mds = le32_to_cpu(dirinfo->auth);
235 int ndist = le32_to_cpu(dirinfo->ndist);
236 int i;
237 int err = 0;
238
239 mutex_lock(&ci->i_fragtree_mutex);
240 if (ndist == 0) {
241 /* no delegation info needed. */
242 frag = __ceph_find_frag(ci, id);
243 if (!frag)
244 goto out;
245 if (frag->split_by == 0) {
246 /* tree leaf, remove */
247 dout("fill_dirfrag removed %llx.%llx frag %x"
248 " (no ref)\n", ceph_vinop(inode), id);
249 rb_erase(&frag->node, &ci->i_fragtree);
250 kfree(frag);
251 } else {
252 /* tree branch, keep and clear */
253 dout("fill_dirfrag cleared %llx.%llx frag %x"
254 " referral\n", ceph_vinop(inode), id);
255 frag->mds = -1;
256 frag->ndist = 0;
257 }
258 goto out;
259 }
260
261
262 /* find/add this frag to store mds delegation info */
263 frag = __get_or_create_frag(ci, id);
264 if (IS_ERR(frag)) {
265 /* this is not the end of the world; we can continue
266 with bad/inaccurate delegation info */
267 pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n",
268 ceph_vinop(inode), le32_to_cpu(dirinfo->frag));
269 err = -ENOMEM;
270 goto out;
271 }
272
273 frag->mds = mds;
274 frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP);
275 for (i = 0; i < frag->ndist; i++)
276 frag->dist[i] = le32_to_cpu(dirinfo->dist[i]);
277 dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n",
278 ceph_vinop(inode), frag->frag, frag->ndist);
279
280out:
281 mutex_unlock(&ci->i_fragtree_mutex);
282 return err;
283}
284
285
286/*
287 * initialize a newly allocated inode.
288 */
289struct inode *ceph_alloc_inode(struct super_block *sb)
290{
291 struct ceph_inode_info *ci;
292 int i;
293
294 ci = kmem_cache_alloc(ceph_inode_cachep, GFP_NOFS);
295 if (!ci)
296 return NULL;
297
298 dout("alloc_inode %p\n", &ci->vfs_inode);
299
Sage Weilbe655592011-11-30 09:47:09 -0800300 spin_lock_init(&ci->i_ceph_lock);
301
Sage Weil355da1e2009-10-06 11:31:08 -0700302 ci->i_version = 0;
303 ci->i_time_warp_seq = 0;
304 ci->i_ceph_flags = 0;
Yan, Zheng2f276c52013-03-13 19:44:32 +0800305 atomic_set(&ci->i_release_count, 1);
306 atomic_set(&ci->i_complete_count, 0);
Sage Weil355da1e2009-10-06 11:31:08 -0700307 ci->i_symlink = NULL;
308
Sage Weil6c0f3af2010-11-16 11:14:34 -0800309 memset(&ci->i_dir_layout, 0, sizeof(ci->i_dir_layout));
310
Sage Weil355da1e2009-10-06 11:31:08 -0700311 ci->i_fragtree = RB_ROOT;
312 mutex_init(&ci->i_fragtree_mutex);
313
314 ci->i_xattrs.blob = NULL;
315 ci->i_xattrs.prealloc_blob = NULL;
316 ci->i_xattrs.dirty = false;
317 ci->i_xattrs.index = RB_ROOT;
318 ci->i_xattrs.count = 0;
319 ci->i_xattrs.names_size = 0;
320 ci->i_xattrs.vals_size = 0;
321 ci->i_xattrs.version = 0;
322 ci->i_xattrs.index_version = 0;
323
324 ci->i_caps = RB_ROOT;
325 ci->i_auth_cap = NULL;
326 ci->i_dirty_caps = 0;
327 ci->i_flushing_caps = 0;
328 INIT_LIST_HEAD(&ci->i_dirty_item);
329 INIT_LIST_HEAD(&ci->i_flushing_item);
330 ci->i_cap_flush_seq = 0;
331 ci->i_cap_flush_last_tid = 0;
332 memset(&ci->i_cap_flush_tid, 0, sizeof(ci->i_cap_flush_tid));
333 init_waitqueue_head(&ci->i_cap_wq);
334 ci->i_hold_caps_min = 0;
335 ci->i_hold_caps_max = 0;
336 INIT_LIST_HEAD(&ci->i_cap_delay_list);
337 ci->i_cap_exporting_mds = 0;
338 ci->i_cap_exporting_mseq = 0;
339 ci->i_cap_exporting_issued = 0;
340 INIT_LIST_HEAD(&ci->i_cap_snaps);
341 ci->i_head_snapc = NULL;
342 ci->i_snap_caps = 0;
343
344 for (i = 0; i < CEPH_FILE_MODE_NUM; i++)
345 ci->i_nr_by_mode[i] = 0;
346
347 ci->i_truncate_seq = 0;
348 ci->i_truncate_size = 0;
349 ci->i_truncate_pending = 0;
350
351 ci->i_max_size = 0;
352 ci->i_reported_size = 0;
353 ci->i_wanted_max_size = 0;
354 ci->i_requested_max_size = 0;
355
356 ci->i_pin_ref = 0;
357 ci->i_rd_ref = 0;
358 ci->i_rdcache_ref = 0;
359 ci->i_wr_ref = 0;
Henry C Changd3d07202011-05-11 10:29:54 +0000360 ci->i_wb_ref = 0;
Sage Weil355da1e2009-10-06 11:31:08 -0700361 ci->i_wrbuffer_ref = 0;
362 ci->i_wrbuffer_ref_head = 0;
363 ci->i_shared_gen = 0;
364 ci->i_rdcache_gen = 0;
365 ci->i_rdcache_revoking = 0;
366
367 INIT_LIST_HEAD(&ci->i_unsafe_writes);
368 INIT_LIST_HEAD(&ci->i_unsafe_dirops);
369 spin_lock_init(&ci->i_unsafe_lock);
370
371 ci->i_snap_realm = NULL;
372 INIT_LIST_HEAD(&ci->i_snap_realm_item);
373 INIT_LIST_HEAD(&ci->i_snap_flush_item);
374
Sage Weil3c6f6b72010-02-09 15:24:44 -0800375 INIT_WORK(&ci->i_wb_work, ceph_writeback_work);
376 INIT_WORK(&ci->i_pg_inv_work, ceph_invalidate_work);
Sage Weil355da1e2009-10-06 11:31:08 -0700377
378 INIT_WORK(&ci->i_vmtruncate_work, ceph_vmtruncate_work);
379
380 return &ci->vfs_inode;
381}
382
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100383static void ceph_i_callback(struct rcu_head *head)
384{
385 struct inode *inode = container_of(head, struct inode, i_rcu);
386 struct ceph_inode_info *ci = ceph_inode(inode);
387
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100388 kmem_cache_free(ceph_inode_cachep, ci);
389}
390
Sage Weil355da1e2009-10-06 11:31:08 -0700391void ceph_destroy_inode(struct inode *inode)
392{
393 struct ceph_inode_info *ci = ceph_inode(inode);
394 struct ceph_inode_frag *frag;
395 struct rb_node *n;
396
397 dout("destroy_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode));
398
399 ceph_queue_caps_release(inode);
400
Sage Weil8b218b82010-03-09 12:59:08 -0800401 /*
402 * we may still have a snap_realm reference if there are stray
403 * caps in i_cap_exporting_issued or i_snap_caps.
404 */
405 if (ci->i_snap_realm) {
406 struct ceph_mds_client *mdsc =
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700407 ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc;
Sage Weil8b218b82010-03-09 12:59:08 -0800408 struct ceph_snap_realm *realm = ci->i_snap_realm;
409
410 dout(" dropping residual ref to snap realm %p\n", realm);
411 spin_lock(&realm->inodes_with_caps_lock);
412 list_del_init(&ci->i_snap_realm_item);
413 spin_unlock(&realm->inodes_with_caps_lock);
414 ceph_put_snap_realm(mdsc, realm);
415 }
416
Sage Weil355da1e2009-10-06 11:31:08 -0700417 kfree(ci->i_symlink);
418 while ((n = rb_first(&ci->i_fragtree)) != NULL) {
419 frag = rb_entry(n, struct ceph_inode_frag, node);
420 rb_erase(n, &ci->i_fragtree);
421 kfree(frag);
422 }
423
424 __ceph_destroy_xattrs(ci);
Sage Weilb6c1d5b2009-12-07 12:17:17 -0800425 if (ci->i_xattrs.blob)
426 ceph_buffer_put(ci->i_xattrs.blob);
427 if (ci->i_xattrs.prealloc_blob)
428 ceph_buffer_put(ci->i_xattrs.prealloc_blob);
Sage Weil355da1e2009-10-06 11:31:08 -0700429
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100430 call_rcu(&inode->i_rcu, ceph_i_callback);
Sage Weil355da1e2009-10-06 11:31:08 -0700431}
432
433
434/*
435 * Helpers to fill in size, ctime, mtime, and atime. We have to be
436 * careful because either the client or MDS may have more up to date
437 * info, depending on which capabilities are held, and whether
438 * time_warp_seq or truncate_seq have increased. (Ordinarily, mtime
439 * and size are monotonically increasing, except when utimes() or
440 * truncate() increments the corresponding _seq values.)
441 */
442int ceph_fill_file_size(struct inode *inode, int issued,
443 u32 truncate_seq, u64 truncate_size, u64 size)
444{
445 struct ceph_inode_info *ci = ceph_inode(inode);
446 int queue_trunc = 0;
447
448 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 ||
449 (truncate_seq == ci->i_truncate_seq && size > inode->i_size)) {
450 dout("size %lld -> %llu\n", inode->i_size, size);
451 inode->i_size = size;
452 inode->i_blocks = (size + (1<<9) - 1) >> 9;
453 ci->i_reported_size = size;
454 if (truncate_seq != ci->i_truncate_seq) {
455 dout("truncate_seq %u -> %u\n",
456 ci->i_truncate_seq, truncate_seq);
457 ci->i_truncate_seq = truncate_seq;
Yehuda Sadeh3d497d82010-02-09 11:08:40 -0800458 /*
459 * If we hold relevant caps, or in the case where we're
460 * not the only client referencing this file and we
461 * don't hold those caps, then we need to check whether
462 * the file is either opened or mmaped
463 */
464 if ((issued & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_RD|
Sage Weil29625072010-05-27 10:40:43 -0700465 CEPH_CAP_FILE_WR|CEPH_CAP_FILE_BUFFER|
466 CEPH_CAP_FILE_EXCL|
467 CEPH_CAP_FILE_LAZYIO)) ||
Yehuda Sadeh3d497d82010-02-09 11:08:40 -0800468 mapping_mapped(inode->i_mapping) ||
469 __ceph_caps_file_wanted(ci)) {
Sage Weil355da1e2009-10-06 11:31:08 -0700470 ci->i_truncate_pending++;
471 queue_trunc = 1;
472 }
473 }
474 }
475 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 &&
476 ci->i_truncate_size != truncate_size) {
477 dout("truncate_size %lld -> %llu\n", ci->i_truncate_size,
478 truncate_size);
479 ci->i_truncate_size = truncate_size;
480 }
481 return queue_trunc;
482}
483
484void ceph_fill_file_time(struct inode *inode, int issued,
485 u64 time_warp_seq, struct timespec *ctime,
486 struct timespec *mtime, struct timespec *atime)
487{
488 struct ceph_inode_info *ci = ceph_inode(inode);
489 int warn = 0;
490
491 if (issued & (CEPH_CAP_FILE_EXCL|
492 CEPH_CAP_FILE_WR|
Sage Weild8672d62010-11-08 09:24:34 -0800493 CEPH_CAP_FILE_BUFFER|
494 CEPH_CAP_AUTH_EXCL|
495 CEPH_CAP_XATTR_EXCL)) {
Sage Weil355da1e2009-10-06 11:31:08 -0700496 if (timespec_compare(ctime, &inode->i_ctime) > 0) {
497 dout("ctime %ld.%09ld -> %ld.%09ld inc w/ cap\n",
498 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
499 ctime->tv_sec, ctime->tv_nsec);
500 inode->i_ctime = *ctime;
501 }
502 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) {
503 /* the MDS did a utimes() */
504 dout("mtime %ld.%09ld -> %ld.%09ld "
505 "tw %d -> %d\n",
506 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
507 mtime->tv_sec, mtime->tv_nsec,
508 ci->i_time_warp_seq, (int)time_warp_seq);
509
510 inode->i_mtime = *mtime;
511 inode->i_atime = *atime;
512 ci->i_time_warp_seq = time_warp_seq;
513 } else if (time_warp_seq == ci->i_time_warp_seq) {
514 /* nobody did utimes(); take the max */
515 if (timespec_compare(mtime, &inode->i_mtime) > 0) {
516 dout("mtime %ld.%09ld -> %ld.%09ld inc\n",
517 inode->i_mtime.tv_sec,
518 inode->i_mtime.tv_nsec,
519 mtime->tv_sec, mtime->tv_nsec);
520 inode->i_mtime = *mtime;
521 }
522 if (timespec_compare(atime, &inode->i_atime) > 0) {
523 dout("atime %ld.%09ld -> %ld.%09ld inc\n",
524 inode->i_atime.tv_sec,
525 inode->i_atime.tv_nsec,
526 atime->tv_sec, atime->tv_nsec);
527 inode->i_atime = *atime;
528 }
529 } else if (issued & CEPH_CAP_FILE_EXCL) {
530 /* we did a utimes(); ignore mds values */
531 } else {
532 warn = 1;
533 }
534 } else {
Sage Weild8672d62010-11-08 09:24:34 -0800535 /* we have no write|excl caps; whatever the MDS says is true */
Sage Weil355da1e2009-10-06 11:31:08 -0700536 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) {
537 inode->i_ctime = *ctime;
538 inode->i_mtime = *mtime;
539 inode->i_atime = *atime;
540 ci->i_time_warp_seq = time_warp_seq;
541 } else {
542 warn = 1;
543 }
544 }
545 if (warn) /* time_warp_seq shouldn't go backwards */
546 dout("%p mds time_warp_seq %llu < %u\n",
547 inode, time_warp_seq, ci->i_time_warp_seq);
548}
549
550/*
551 * Populate an inode based on info from mds. May be called on new or
552 * existing inodes.
553 */
554static int fill_inode(struct inode *inode,
555 struct ceph_mds_reply_info_in *iinfo,
556 struct ceph_mds_reply_dirfrag *dirinfo,
557 struct ceph_mds_session *session,
558 unsigned long ttl_from, int cap_fmode,
559 struct ceph_cap_reservation *caps_reservation)
560{
561 struct ceph_mds_reply_inode *info = iinfo->in;
562 struct ceph_inode_info *ci = ceph_inode(inode);
563 int i;
Sage Weildfabbed62011-07-26 11:30:02 -0700564 int issued = 0, implemented;
Sage Weil355da1e2009-10-06 11:31:08 -0700565 struct timespec mtime, atime, ctime;
566 u32 nsplits;
567 struct ceph_buffer *xattr_blob = NULL;
568 int err = 0;
569 int queue_trunc = 0;
570
571 dout("fill_inode %p ino %llx.%llx v %llu had %llu\n",
572 inode, ceph_vinop(inode), le64_to_cpu(info->version),
573 ci->i_version);
574
575 /*
576 * prealloc xattr data, if it looks like we'll need it. only
577 * if len > 4 (meaning there are actually xattrs; the first 4
578 * bytes are the xattr count).
579 */
580 if (iinfo->xattr_len > 4) {
Sage Weilb6c1d5b2009-12-07 12:17:17 -0800581 xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS);
Sage Weil355da1e2009-10-06 11:31:08 -0700582 if (!xattr_blob)
583 pr_err("fill_inode ENOMEM xattr blob %d bytes\n",
584 iinfo->xattr_len);
585 }
586
Sage Weilbe655592011-11-30 09:47:09 -0800587 spin_lock(&ci->i_ceph_lock);
Sage Weil355da1e2009-10-06 11:31:08 -0700588
589 /*
590 * provided version will be odd if inode value is projected,
Sage Weil8bd59e02010-11-08 09:23:12 -0800591 * even if stable. skip the update if we have newer stable
592 * info (ours>=theirs, e.g. due to racing mds replies), unless
593 * we are getting projected (unstable) info (in which case the
594 * version is odd, and we want ours>theirs).
595 * us them
596 * 2 2 skip
597 * 3 2 skip
598 * 3 3 update
Sage Weil355da1e2009-10-06 11:31:08 -0700599 */
600 if (le64_to_cpu(info->version) > 0 &&
Sage Weil8bd59e02010-11-08 09:23:12 -0800601 (ci->i_version & ~1) >= le64_to_cpu(info->version))
Sage Weil355da1e2009-10-06 11:31:08 -0700602 goto no_change;
Sage Weildfabbed62011-07-26 11:30:02 -0700603
Sage Weil355da1e2009-10-06 11:31:08 -0700604 issued = __ceph_caps_issued(ci, &implemented);
605 issued |= implemented | __ceph_caps_dirty(ci);
606
607 /* update inode */
608 ci->i_version = le64_to_cpu(info->version);
609 inode->i_version++;
610 inode->i_rdev = le32_to_cpu(info->rdev);
611
612 if ((issued & CEPH_CAP_AUTH_EXCL) == 0) {
613 inode->i_mode = le32_to_cpu(info->mode);
Eric W. Biedermanab871b92013-01-31 03:40:12 -0800614 inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(info->uid));
615 inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(info->gid));
Sage Weil355da1e2009-10-06 11:31:08 -0700616 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
Eric W. Biedermanbd2bae62013-01-31 04:05:39 -0800617 from_kuid(&init_user_ns, inode->i_uid),
618 from_kgid(&init_user_ns, inode->i_gid));
Sage Weil355da1e2009-10-06 11:31:08 -0700619 }
620
621 if ((issued & CEPH_CAP_LINK_EXCL) == 0)
Miklos Szeredibfe86842011-10-28 14:13:29 +0200622 set_nlink(inode, le32_to_cpu(info->nlink));
Sage Weil355da1e2009-10-06 11:31:08 -0700623
624 /* be careful with mtime, atime, size */
625 ceph_decode_timespec(&atime, &info->atime);
626 ceph_decode_timespec(&mtime, &info->mtime);
627 ceph_decode_timespec(&ctime, &info->ctime);
628 queue_trunc = ceph_fill_file_size(inode, issued,
629 le32_to_cpu(info->truncate_seq),
630 le64_to_cpu(info->truncate_size),
Sage Weil355da1e2009-10-06 11:31:08 -0700631 le64_to_cpu(info->size));
632 ceph_fill_file_time(inode, issued,
633 le32_to_cpu(info->time_warp_seq),
634 &ctime, &mtime, &atime);
635
Sage Weil912a9b02010-11-07 09:37:25 -0800636 /* only update max_size on auth cap */
637 if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
638 ci->i_max_size != le64_to_cpu(info->max_size)) {
639 dout("max_size %lld -> %llu\n", ci->i_max_size,
640 le64_to_cpu(info->max_size));
641 ci->i_max_size = le64_to_cpu(info->max_size);
642 }
643
Sage Weil355da1e2009-10-06 11:31:08 -0700644 ci->i_layout = info->layout;
645 inode->i_blkbits = fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1;
646
647 /* xattrs */
648 /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
649 if ((issued & CEPH_CAP_XATTR_EXCL) == 0 &&
650 le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) {
651 if (ci->i_xattrs.blob)
652 ceph_buffer_put(ci->i_xattrs.blob);
653 ci->i_xattrs.blob = xattr_blob;
654 if (xattr_blob)
655 memcpy(ci->i_xattrs.blob->vec.iov_base,
656 iinfo->xattr_data, iinfo->xattr_len);
657 ci->i_xattrs.version = le64_to_cpu(info->xattr_version);
Sage Weila6424e42010-04-29 09:28:11 -0700658 xattr_blob = NULL;
Sage Weil355da1e2009-10-06 11:31:08 -0700659 }
660
661 inode->i_mapping->a_ops = &ceph_aops;
662 inode->i_mapping->backing_dev_info =
Cheng Renquan640ef792010-03-26 17:40:33 +0800663 &ceph_sb_to_client(inode->i_sb)->backing_dev_info;
Sage Weil355da1e2009-10-06 11:31:08 -0700664
665 switch (inode->i_mode & S_IFMT) {
666 case S_IFIFO:
667 case S_IFBLK:
668 case S_IFCHR:
669 case S_IFSOCK:
670 init_special_inode(inode, inode->i_mode, inode->i_rdev);
671 inode->i_op = &ceph_file_iops;
672 break;
673 case S_IFREG:
674 inode->i_op = &ceph_file_iops;
675 inode->i_fop = &ceph_file_fops;
676 break;
677 case S_IFLNK:
678 inode->i_op = &ceph_symlink_iops;
679 if (!ci->i_symlink) {
Xi Wang810339ec2012-02-03 09:55:36 -0500680 u32 symlen = iinfo->symlink_len;
Sage Weil355da1e2009-10-06 11:31:08 -0700681 char *sym;
682
Sage Weilbe655592011-11-30 09:47:09 -0800683 spin_unlock(&ci->i_ceph_lock);
Sage Weil355da1e2009-10-06 11:31:08 -0700684
Xi Wang810339ec2012-02-03 09:55:36 -0500685 err = -EINVAL;
686 if (WARN_ON(symlen != inode->i_size))
687 goto out;
688
Sage Weil355da1e2009-10-06 11:31:08 -0700689 err = -ENOMEM;
Xi Wang810339ec2012-02-03 09:55:36 -0500690 sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS);
Sage Weil355da1e2009-10-06 11:31:08 -0700691 if (!sym)
692 goto out;
Sage Weil355da1e2009-10-06 11:31:08 -0700693
Sage Weilbe655592011-11-30 09:47:09 -0800694 spin_lock(&ci->i_ceph_lock);
Sage Weil355da1e2009-10-06 11:31:08 -0700695 if (!ci->i_symlink)
696 ci->i_symlink = sym;
697 else
698 kfree(sym); /* lost a race */
699 }
700 break;
701 case S_IFDIR:
702 inode->i_op = &ceph_dir_iops;
703 inode->i_fop = &ceph_dir_fops;
704
Sage Weil14303d22010-12-14 17:37:52 -0800705 ci->i_dir_layout = iinfo->dir_layout;
706
Sage Weil355da1e2009-10-06 11:31:08 -0700707 ci->i_files = le64_to_cpu(info->files);
708 ci->i_subdirs = le64_to_cpu(info->subdirs);
709 ci->i_rbytes = le64_to_cpu(info->rbytes);
710 ci->i_rfiles = le64_to_cpu(info->rfiles);
711 ci->i_rsubdirs = le64_to_cpu(info->rsubdirs);
712 ceph_decode_timespec(&ci->i_rctime, &info->rctime);
Sage Weil355da1e2009-10-06 11:31:08 -0700713 break;
714 default:
715 pr_err("fill_inode %llx.%llx BAD mode 0%o\n",
716 ceph_vinop(inode), inode->i_mode);
717 }
718
Yan, Zhenga8673d62013-02-18 16:38:14 +0800719 /* set dir completion flag? */
720 if (S_ISDIR(inode->i_mode) &&
721 ci->i_files == 0 && ci->i_subdirs == 0 &&
722 ceph_snap(inode) == CEPH_NOSNAP &&
723 (le32_to_cpu(info->cap.caps) & CEPH_CAP_FILE_SHARED) &&
724 (issued & CEPH_CAP_FILE_EXCL) == 0 &&
Yan, Zheng2f276c52013-03-13 19:44:32 +0800725 !__ceph_dir_is_complete(ci)) {
Yan, Zhenga8673d62013-02-18 16:38:14 +0800726 dout(" marking %p complete (empty)\n", inode);
Yan, Zheng2f276c52013-03-13 19:44:32 +0800727 __ceph_dir_set_complete(ci, atomic_read(&ci->i_release_count));
Yan, Zhenga8673d62013-02-18 16:38:14 +0800728 ci->i_max_offset = 2;
729 }
Sage Weil355da1e2009-10-06 11:31:08 -0700730no_change:
Sage Weilbe655592011-11-30 09:47:09 -0800731 spin_unlock(&ci->i_ceph_lock);
Sage Weil355da1e2009-10-06 11:31:08 -0700732
733 /* queue truncate if we saw i_size decrease */
734 if (queue_trunc)
Sage Weil3c6f6b72010-02-09 15:24:44 -0800735 ceph_queue_vmtruncate(inode);
Sage Weil355da1e2009-10-06 11:31:08 -0700736
737 /* populate frag tree */
738 /* FIXME: move me up, if/when version reflects fragtree changes */
739 nsplits = le32_to_cpu(info->fragtree.nsplits);
740 mutex_lock(&ci->i_fragtree_mutex);
741 for (i = 0; i < nsplits; i++) {
742 u32 id = le32_to_cpu(info->fragtree.splits[i].frag);
743 struct ceph_inode_frag *frag = __get_or_create_frag(ci, id);
744
745 if (IS_ERR(frag))
746 continue;
747 frag->split_by = le32_to_cpu(info->fragtree.splits[i].by);
748 dout(" frag %x split by %d\n", frag->frag, frag->split_by);
749 }
750 mutex_unlock(&ci->i_fragtree_mutex);
751
752 /* were we issued a capability? */
753 if (info->cap.caps) {
754 if (ceph_snap(inode) == CEPH_NOSNAP) {
755 ceph_add_cap(inode, session,
756 le64_to_cpu(info->cap.cap_id),
757 cap_fmode,
758 le32_to_cpu(info->cap.caps),
759 le32_to_cpu(info->cap.wanted),
760 le32_to_cpu(info->cap.seq),
761 le32_to_cpu(info->cap.mseq),
762 le64_to_cpu(info->cap.realm),
763 info->cap.flags,
764 caps_reservation);
765 } else {
Sage Weilbe655592011-11-30 09:47:09 -0800766 spin_lock(&ci->i_ceph_lock);
Sage Weil355da1e2009-10-06 11:31:08 -0700767 dout(" %p got snap_caps %s\n", inode,
768 ceph_cap_string(le32_to_cpu(info->cap.caps)));
769 ci->i_snap_caps |= le32_to_cpu(info->cap.caps);
770 if (cap_fmode >= 0)
771 __ceph_get_fmode(ci, cap_fmode);
Sage Weilbe655592011-11-30 09:47:09 -0800772 spin_unlock(&ci->i_ceph_lock);
Sage Weil355da1e2009-10-06 11:31:08 -0700773 }
Sage Weil04d000e2010-05-07 11:26:34 -0700774 } else if (cap_fmode >= 0) {
775 pr_warning("mds issued no caps on %llx.%llx\n",
776 ceph_vinop(inode));
777 __ceph_get_fmode(ci, cap_fmode);
Sage Weil355da1e2009-10-06 11:31:08 -0700778 }
779
780 /* update delegation info? */
781 if (dirinfo)
782 ceph_fill_dirfrag(inode, dirinfo);
783
784 err = 0;
785
786out:
Sage Weilb6c1d5b2009-12-07 12:17:17 -0800787 if (xattr_blob)
788 ceph_buffer_put(xattr_blob);
Sage Weil355da1e2009-10-06 11:31:08 -0700789 return err;
790}
791
792/*
793 * caller should hold session s_mutex.
794 */
795static void update_dentry_lease(struct dentry *dentry,
796 struct ceph_mds_reply_lease *lease,
797 struct ceph_mds_session *session,
798 unsigned long from_time)
799{
800 struct ceph_dentry_info *di = ceph_dentry(dentry);
801 long unsigned duration = le32_to_cpu(lease->duration_ms);
802 long unsigned ttl = from_time + (duration * HZ) / 1000;
803 long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000;
804 struct inode *dir;
805
806 /* only track leases on regular dentries */
807 if (dentry->d_op != &ceph_dentry_ops)
808 return;
809
810 spin_lock(&dentry->d_lock);
Sage Weil2f90b852011-07-26 11:28:25 -0700811 dout("update_dentry_lease %p duration %lu ms ttl %lu\n",
812 dentry, duration, ttl);
Sage Weil355da1e2009-10-06 11:31:08 -0700813
814 /* make lease_rdcache_gen match directory */
815 dir = dentry->d_parent->d_inode;
816 di->lease_shared_gen = ceph_inode(dir)->i_shared_gen;
817
Sage Weil2f90b852011-07-26 11:28:25 -0700818 if (duration == 0)
Sage Weil355da1e2009-10-06 11:31:08 -0700819 goto out_unlock;
820
821 if (di->lease_gen == session->s_cap_gen &&
822 time_before(ttl, dentry->d_time))
823 goto out_unlock; /* we already have a newer lease. */
824
825 if (di->lease_session && di->lease_session != session)
826 goto out_unlock;
827
828 ceph_dentry_lru_touch(dentry);
829
830 if (!di->lease_session)
831 di->lease_session = ceph_get_mds_session(session);
832 di->lease_gen = session->s_cap_gen;
833 di->lease_seq = le32_to_cpu(lease->seq);
834 di->lease_renew_after = half_ttl;
835 di->lease_renew_from = 0;
836 dentry->d_time = ttl;
837out_unlock:
838 spin_unlock(&dentry->d_lock);
839 return;
840}
841
842/*
Yehuda Sadeh4baa75e2010-01-07 15:36:32 -0800843 * Set dentry's directory position based on the current dir's max, and
844 * order it in d_subdirs, so that dcache_readdir behaves.
Sage Weil4f177262011-07-26 11:31:08 -0700845 *
846 * Always called under directory's i_mutex.
Yehuda Sadeh4baa75e2010-01-07 15:36:32 -0800847 */
848static void ceph_set_dentry_offset(struct dentry *dn)
849{
850 struct dentry *dir = dn->d_parent;
Sage Weil4f177262011-07-26 11:31:08 -0700851 struct inode *inode = dir->d_inode;
Yehuda Sadehb8cd9522011-12-13 09:56:30 -0800852 struct ceph_inode_info *ci;
Yehuda Sadeh4baa75e2010-01-07 15:36:32 -0800853 struct ceph_dentry_info *di;
854
855 BUG_ON(!inode);
856
Yehuda Sadehb8cd9522011-12-13 09:56:30 -0800857 ci = ceph_inode(inode);
Yehuda Sadeh4baa75e2010-01-07 15:36:32 -0800858 di = ceph_dentry(dn);
859
Sage Weilbe655592011-11-30 09:47:09 -0800860 spin_lock(&ci->i_ceph_lock);
Yan, Zheng2f276c52013-03-13 19:44:32 +0800861 if (!__ceph_dir_is_complete(ci)) {
Sage Weilbe655592011-11-30 09:47:09 -0800862 spin_unlock(&ci->i_ceph_lock);
Sage Weile8a74982010-04-15 14:08:49 -0700863 return;
864 }
Yehuda Sadeh4baa75e2010-01-07 15:36:32 -0800865 di->offset = ceph_inode(inode)->i_max_offset++;
Sage Weilbe655592011-11-30 09:47:09 -0800866 spin_unlock(&ci->i_ceph_lock);
Yehuda Sadeh4baa75e2010-01-07 15:36:32 -0800867
Nick Piggin2fd6b7f2011-01-07 17:49:34 +1100868 spin_lock(&dir->d_lock);
869 spin_lock_nested(&dn->d_lock, DENTRY_D_LOCK_NESTED);
Henry C Chang13a42142010-06-01 11:31:08 -0700870 list_move(&dn->d_u.d_child, &dir->d_subdirs);
Yehuda Sadeh4baa75e2010-01-07 15:36:32 -0800871 dout("set_dentry_offset %p %lld (%p %p)\n", dn, di->offset,
872 dn->d_u.d_child.prev, dn->d_u.d_child.next);
873 spin_unlock(&dn->d_lock);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +1100874 spin_unlock(&dir->d_lock);
Yehuda Sadeh4baa75e2010-01-07 15:36:32 -0800875}
876
877/*
Sage Weil1cd39352010-05-03 22:08:02 -0700878 * splice a dentry to an inode.
879 * caller must hold directory i_mutex for this to be safe.
880 *
881 * we will only rehash the resulting dentry if @prehash is
882 * true; @prehash will be set to false (for the benefit of
883 * the caller) if we fail.
884 */
885static struct dentry *splice_dentry(struct dentry *dn, struct inode *in,
Sage Weil467c5252010-09-13 11:39:20 -0700886 bool *prehash, bool set_offset)
Sage Weil1cd39352010-05-03 22:08:02 -0700887{
888 struct dentry *realdn;
889
890 BUG_ON(dn->d_inode);
891
892 /* dn must be unhashed */
893 if (!d_unhashed(dn))
894 d_drop(dn);
895 realdn = d_materialise_unique(dn, in);
896 if (IS_ERR(realdn)) {
Sage Weild69ed052010-06-21 10:38:14 -0700897 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
898 PTR_ERR(realdn), dn, in, ceph_vinop(in));
Sage Weil1cd39352010-05-03 22:08:02 -0700899 if (prehash)
900 *prehash = false; /* don't rehash on error */
901 dn = realdn; /* note realdn contains the error */
902 goto out;
903 } else if (realdn) {
904 dout("dn %p (%d) spliced with %p (%d) "
905 "inode %p ino %llx.%llx\n",
Al Viro84d08fa2013-07-05 18:59:33 +0400906 dn, d_count(dn),
907 realdn, d_count(realdn),
Sage Weil1cd39352010-05-03 22:08:02 -0700908 realdn->d_inode, ceph_vinop(realdn->d_inode));
909 dput(dn);
910 dn = realdn;
911 } else {
912 BUG_ON(!ceph_dentry(dn));
913 dout("dn %p attached to %p ino %llx.%llx\n",
914 dn, dn->d_inode, ceph_vinop(dn->d_inode));
915 }
916 if ((!prehash || *prehash) && d_unhashed(dn))
917 d_rehash(dn);
Sage Weil467c5252010-09-13 11:39:20 -0700918 if (set_offset)
919 ceph_set_dentry_offset(dn);
Sage Weil1cd39352010-05-03 22:08:02 -0700920out:
921 return dn;
922}
923
924/*
Sage Weil355da1e2009-10-06 11:31:08 -0700925 * Incorporate results into the local cache. This is either just
926 * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
927 * after a lookup).
928 *
929 * A reply may contain
930 * a directory inode along with a dentry.
931 * and/or a target inode
932 *
933 * Called with snap_rwsem (read).
934 */
935int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req,
936 struct ceph_mds_session *session)
937{
938 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
939 struct inode *in = NULL;
940 struct ceph_mds_reply_inode *ininfo;
941 struct ceph_vino vino;
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700942 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
Sage Weil355da1e2009-10-06 11:31:08 -0700943 int i = 0;
944 int err = 0;
945
946 dout("fill_trace %p is_dentry %d is_target %d\n", req,
947 rinfo->head->is_dentry, rinfo->head->is_target);
948
949#if 0
950 /*
951 * Debugging hook:
952 *
953 * If we resend completed ops to a recovering mds, we get no
954 * trace. Since that is very rare, pretend this is the case
955 * to ensure the 'no trace' handlers in the callers behave.
956 *
957 * Fill in inodes unconditionally to avoid breaking cap
958 * invariants.
959 */
960 if (rinfo->head->op & CEPH_MDS_OP_WRITE) {
961 pr_info("fill_trace faking empty trace on %lld %s\n",
962 req->r_tid, ceph_mds_op_name(rinfo->head->op));
963 if (rinfo->head->is_dentry) {
964 rinfo->head->is_dentry = 0;
965 err = fill_inode(req->r_locked_dir,
966 &rinfo->diri, rinfo->dirfrag,
967 session, req->r_request_started, -1);
968 }
969 if (rinfo->head->is_target) {
970 rinfo->head->is_target = 0;
971 ininfo = rinfo->targeti.in;
972 vino.ino = le64_to_cpu(ininfo->ino);
973 vino.snap = le64_to_cpu(ininfo->snapid);
974 in = ceph_get_inode(sb, vino);
975 err = fill_inode(in, &rinfo->targeti, NULL,
976 session, req->r_request_started,
977 req->r_fmode);
978 iput(in);
979 }
980 }
981#endif
982
983 if (!rinfo->head->is_target && !rinfo->head->is_dentry) {
984 dout("fill_trace reply is empty!\n");
Sage Weil167c9e32010-05-14 10:02:57 -0700985 if (rinfo->head->result == 0 && req->r_locked_dir)
986 ceph_invalidate_dir_request(req);
Sage Weil355da1e2009-10-06 11:31:08 -0700987 return 0;
988 }
989
990 if (rinfo->head->is_dentry) {
Sage Weil5b1daec2010-01-25 11:33:08 -0800991 struct inode *dir = req->r_locked_dir;
992
Sage Weil6c5e50f2012-08-21 15:55:25 -0700993 if (dir) {
994 err = fill_inode(dir, &rinfo->diri, rinfo->dirfrag,
995 session, req->r_request_started, -1,
996 &req->r_caps_reservation);
997 if (err < 0)
998 return err;
999 } else {
1000 WARN_ON_ONCE(1);
1001 }
Sage Weil5b1daec2010-01-25 11:33:08 -08001002 }
1003
Sage Weil9358c6d2010-03-30 13:54:41 -07001004 /*
1005 * ignore null lease/binding on snapdir ENOENT, or else we
1006 * will have trouble splicing in the virtual snapdir later
1007 */
1008 if (rinfo->head->is_dentry && !req->r_aborted &&
Sage Weil6c5e50f2012-08-21 15:55:25 -07001009 req->r_locked_dir &&
Sage Weil9358c6d2010-03-30 13:54:41 -07001010 (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name,
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001011 fsc->mount_options->snapdir_name,
Sage Weil9358c6d2010-03-30 13:54:41 -07001012 req->r_dentry->d_name.len))) {
Sage Weil355da1e2009-10-06 11:31:08 -07001013 /*
1014 * lookup link rename : null -> possibly existing inode
1015 * mknod symlink mkdir : null -> new inode
1016 * unlink : linked -> null
1017 */
1018 struct inode *dir = req->r_locked_dir;
1019 struct dentry *dn = req->r_dentry;
1020 bool have_dir_cap, have_lease;
1021
1022 BUG_ON(!dn);
1023 BUG_ON(!dir);
1024 BUG_ON(dn->d_parent->d_inode != dir);
1025 BUG_ON(ceph_ino(dir) !=
1026 le64_to_cpu(rinfo->diri.in->ino));
1027 BUG_ON(ceph_snap(dir) !=
1028 le64_to_cpu(rinfo->diri.in->snapid));
1029
Sage Weil355da1e2009-10-06 11:31:08 -07001030 /* do we have a lease on the whole dir? */
1031 have_dir_cap =
1032 (le32_to_cpu(rinfo->diri.in->cap.caps) &
1033 CEPH_CAP_FILE_SHARED);
1034
1035 /* do we have a dn lease? */
1036 have_lease = have_dir_cap ||
Sage Weil2f90b852011-07-26 11:28:25 -07001037 le32_to_cpu(rinfo->dlease->duration_ms);
Sage Weil355da1e2009-10-06 11:31:08 -07001038 if (!have_lease)
1039 dout("fill_trace no dentry lease or dir cap\n");
1040
1041 /* rename? */
1042 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) {
1043 dout(" src %p '%.*s' dst %p '%.*s'\n",
1044 req->r_old_dentry,
1045 req->r_old_dentry->d_name.len,
1046 req->r_old_dentry->d_name.name,
1047 dn, dn->d_name.len, dn->d_name.name);
1048 dout("fill_trace doing d_move %p -> %p\n",
1049 req->r_old_dentry, dn);
Sage Weilc10f5e12010-04-16 12:56:11 -07001050
Sage Weil355da1e2009-10-06 11:31:08 -07001051 d_move(req->r_old_dentry, dn);
1052 dout(" src %p '%.*s' dst %p '%.*s'\n",
1053 req->r_old_dentry,
1054 req->r_old_dentry->d_name.len,
1055 req->r_old_dentry->d_name.name,
1056 dn, dn->d_name.len, dn->d_name.name);
Sage Weil81a6cf22010-05-14 09:35:38 -07001057
Sage Weilc4a29f22009-12-21 11:42:18 -08001058 /* ensure target dentry is invalidated, despite
1059 rehashing bug in vfs_rename_dir */
Sage Weil81a6cf22010-05-14 09:35:38 -07001060 ceph_invalidate_dentry_lease(dn);
1061
Sage Weil09adc802011-02-04 21:38:47 -08001062 /*
1063 * d_move() puts the renamed dentry at the end of
1064 * d_subdirs. We need to assign it an appropriate
Yan, Zheng2f276c52013-03-13 19:44:32 +08001065 * directory offset so we can behave when dir is
1066 * complete.
Sage Weil09adc802011-02-04 21:38:47 -08001067 */
1068 ceph_set_dentry_offset(req->r_old_dentry);
1069 dout("dn %p gets new offset %lld\n", req->r_old_dentry,
Sage Weil1cd39352010-05-03 22:08:02 -07001070 ceph_dentry(req->r_old_dentry)->offset);
Sage Weil81a6cf22010-05-14 09:35:38 -07001071
Sage Weil355da1e2009-10-06 11:31:08 -07001072 dn = req->r_old_dentry; /* use old_dentry */
1073 in = dn->d_inode;
1074 }
1075
1076 /* null dentry? */
1077 if (!rinfo->head->is_target) {
1078 dout("fill_trace null dentry\n");
1079 if (dn->d_inode) {
1080 dout("d_delete %p\n", dn);
1081 d_delete(dn);
1082 } else {
1083 dout("d_instantiate %p NULL\n", dn);
1084 d_instantiate(dn, NULL);
1085 if (have_lease && d_unhashed(dn))
1086 d_rehash(dn);
1087 update_dentry_lease(dn, rinfo->dlease,
1088 session,
1089 req->r_request_started);
1090 }
1091 goto done;
1092 }
1093
1094 /* attach proper inode */
1095 ininfo = rinfo->targeti.in;
1096 vino.ino = le64_to_cpu(ininfo->ino);
1097 vino.snap = le64_to_cpu(ininfo->snapid);
Sage Weild8b16b32010-11-06 12:41:16 -07001098 in = dn->d_inode;
1099 if (!in) {
Sage Weil355da1e2009-10-06 11:31:08 -07001100 in = ceph_get_inode(sb, vino);
1101 if (IS_ERR(in)) {
1102 pr_err("fill_trace bad get_inode "
1103 "%llx.%llx\n", vino.ino, vino.snap);
1104 err = PTR_ERR(in);
Al Viro2744c172012-09-26 21:41:05 -04001105 d_drop(dn);
Sage Weil355da1e2009-10-06 11:31:08 -07001106 goto done;
1107 }
Sage Weil467c5252010-09-13 11:39:20 -07001108 dn = splice_dentry(dn, in, &have_lease, true);
Sage Weil355da1e2009-10-06 11:31:08 -07001109 if (IS_ERR(dn)) {
1110 err = PTR_ERR(dn);
1111 goto done;
1112 }
1113 req->r_dentry = dn; /* may have spliced */
Sage Weil70b666c2011-05-27 09:24:26 -07001114 ihold(in);
Sage Weil355da1e2009-10-06 11:31:08 -07001115 } else if (ceph_ino(in) == vino.ino &&
1116 ceph_snap(in) == vino.snap) {
Sage Weil70b666c2011-05-27 09:24:26 -07001117 ihold(in);
Sage Weil355da1e2009-10-06 11:31:08 -07001118 } else {
1119 dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1120 dn, in, ceph_ino(in), ceph_snap(in),
1121 vino.ino, vino.snap);
1122 have_lease = false;
1123 in = NULL;
1124 }
1125
1126 if (have_lease)
1127 update_dentry_lease(dn, rinfo->dlease, session,
1128 req->r_request_started);
1129 dout(" final dn %p\n", dn);
1130 i++;
Sage Weil79f9f992013-01-29 02:55:31 -05001131 } else if ((req->r_op == CEPH_MDS_OP_LOOKUPSNAP ||
1132 req->r_op == CEPH_MDS_OP_MKSNAP) && !req->r_aborted) {
Sage Weil355da1e2009-10-06 11:31:08 -07001133 struct dentry *dn = req->r_dentry;
1134
1135 /* fill out a snapdir LOOKUPSNAP dentry */
1136 BUG_ON(!dn);
1137 BUG_ON(!req->r_locked_dir);
1138 BUG_ON(ceph_snap(req->r_locked_dir) != CEPH_SNAPDIR);
1139 ininfo = rinfo->targeti.in;
1140 vino.ino = le64_to_cpu(ininfo->ino);
1141 vino.snap = le64_to_cpu(ininfo->snapid);
1142 in = ceph_get_inode(sb, vino);
1143 if (IS_ERR(in)) {
1144 pr_err("fill_inode get_inode badness %llx.%llx\n",
1145 vino.ino, vino.snap);
1146 err = PTR_ERR(in);
1147 d_delete(dn);
1148 goto done;
1149 }
1150 dout(" linking snapped dir %p to dn %p\n", in, dn);
Sage Weil467c5252010-09-13 11:39:20 -07001151 dn = splice_dentry(dn, in, NULL, true);
Sage Weil355da1e2009-10-06 11:31:08 -07001152 if (IS_ERR(dn)) {
1153 err = PTR_ERR(dn);
1154 goto done;
1155 }
1156 req->r_dentry = dn; /* may have spliced */
Sage Weil70b666c2011-05-27 09:24:26 -07001157 ihold(in);
Sage Weil355da1e2009-10-06 11:31:08 -07001158 rinfo->head->is_dentry = 1; /* fool notrace handlers */
1159 }
1160
1161 if (rinfo->head->is_target) {
1162 vino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1163 vino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1164
1165 if (in == NULL || ceph_ino(in) != vino.ino ||
1166 ceph_snap(in) != vino.snap) {
1167 in = ceph_get_inode(sb, vino);
1168 if (IS_ERR(in)) {
1169 err = PTR_ERR(in);
1170 goto done;
1171 }
1172 }
1173 req->r_target_inode = in;
1174
1175 err = fill_inode(in,
1176 &rinfo->targeti, NULL,
1177 session, req->r_request_started,
1178 (le32_to_cpu(rinfo->head->result) == 0) ?
1179 req->r_fmode : -1,
1180 &req->r_caps_reservation);
1181 if (err < 0) {
1182 pr_err("fill_inode badness %p %llx.%llx\n",
1183 in, ceph_vinop(in));
1184 goto done;
1185 }
1186 }
1187
1188done:
1189 dout("fill_trace done err=%d\n", err);
1190 return err;
1191}
1192
1193/*
1194 * Prepopulate our cache with readdir results, leases, etc.
1195 */
Sage Weil79f9f992013-01-29 02:55:31 -05001196static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req,
1197 struct ceph_mds_session *session)
1198{
1199 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1200 int i, err = 0;
1201
1202 for (i = 0; i < rinfo->dir_nr; i++) {
1203 struct ceph_vino vino;
1204 struct inode *in;
1205 int rc;
1206
1207 vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino);
1208 vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid);
1209
1210 in = ceph_get_inode(req->r_dentry->d_sb, vino);
1211 if (IS_ERR(in)) {
1212 err = PTR_ERR(in);
1213 dout("new_inode badness got %d\n", err);
1214 continue;
1215 }
1216 rc = fill_inode(in, &rinfo->dir_in[i], NULL, session,
1217 req->r_request_started, -1,
1218 &req->r_caps_reservation);
1219 if (rc < 0) {
1220 pr_err("fill_inode badness on %p got %d\n", in, rc);
1221 err = rc;
1222 continue;
1223 }
1224 }
1225
1226 return err;
1227}
1228
Sage Weil355da1e2009-10-06 11:31:08 -07001229int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1230 struct ceph_mds_session *session)
1231{
1232 struct dentry *parent = req->r_dentry;
1233 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1234 struct qstr dname;
1235 struct dentry *dn;
1236 struct inode *in;
1237 int err = 0, i;
1238 struct inode *snapdir = NULL;
1239 struct ceph_mds_request_head *rhead = req->r_request->front.iov_base;
1240 u64 frag = le32_to_cpu(rhead->args.readdir.frag);
1241 struct ceph_dentry_info *di;
1242
Sage Weil79f9f992013-01-29 02:55:31 -05001243 if (req->r_aborted)
1244 return readdir_prepopulate_inodes_only(req, session);
1245
Sage Weil355da1e2009-10-06 11:31:08 -07001246 if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
1247 snapdir = ceph_get_snapdir(parent->d_inode);
1248 parent = d_find_alias(snapdir);
1249 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1250 rinfo->dir_nr, parent);
1251 } else {
1252 dout("readdir_prepopulate %d items under dn %p\n",
1253 rinfo->dir_nr, parent);
1254 if (rinfo->dir_dir)
1255 ceph_fill_dirfrag(parent->d_inode, rinfo->dir_dir);
1256 }
1257
1258 for (i = 0; i < rinfo->dir_nr; i++) {
1259 struct ceph_vino vino;
1260
1261 dname.name = rinfo->dir_dname[i];
1262 dname.len = rinfo->dir_dname_len[i];
1263 dname.hash = full_name_hash(dname.name, dname.len);
1264
1265 vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino);
1266 vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid);
1267
1268retry_lookup:
1269 dn = d_lookup(parent, &dname);
1270 dout("d_lookup on parent=%p name=%.*s got %p\n",
1271 parent, dname.len, dname.name, dn);
1272
1273 if (!dn) {
1274 dn = d_alloc(parent, &dname);
1275 dout("d_alloc %p '%.*s' = %p\n", parent,
1276 dname.len, dname.name, dn);
1277 if (dn == NULL) {
1278 dout("d_alloc badness\n");
1279 err = -ENOMEM;
1280 goto out;
1281 }
1282 err = ceph_init_dentry(dn);
Sage Weil8c696732010-07-22 14:11:56 -07001283 if (err < 0) {
1284 dput(dn);
Sage Weil355da1e2009-10-06 11:31:08 -07001285 goto out;
Sage Weil8c696732010-07-22 14:11:56 -07001286 }
Sage Weil355da1e2009-10-06 11:31:08 -07001287 } else if (dn->d_inode &&
1288 (ceph_ino(dn->d_inode) != vino.ino ||
1289 ceph_snap(dn->d_inode) != vino.snap)) {
1290 dout(" dn %p points to wrong inode %p\n",
1291 dn, dn->d_inode);
1292 d_delete(dn);
1293 dput(dn);
1294 goto retry_lookup;
1295 } else {
1296 /* reorder parent's d_subdirs */
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001297 spin_lock(&parent->d_lock);
1298 spin_lock_nested(&dn->d_lock, DENTRY_D_LOCK_NESTED);
Sage Weil355da1e2009-10-06 11:31:08 -07001299 list_move(&dn->d_u.d_child, &parent->d_subdirs);
1300 spin_unlock(&dn->d_lock);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001301 spin_unlock(&parent->d_lock);
Sage Weil355da1e2009-10-06 11:31:08 -07001302 }
1303
1304 di = dn->d_fsdata;
1305 di->offset = ceph_make_fpos(frag, i + req->r_readdir_offset);
1306
1307 /* inode */
1308 if (dn->d_inode) {
1309 in = dn->d_inode;
1310 } else {
1311 in = ceph_get_inode(parent->d_sb, vino);
Dan Carpenterac1f12e2010-08-25 09:11:35 +02001312 if (IS_ERR(in)) {
Sage Weil355da1e2009-10-06 11:31:08 -07001313 dout("new_inode badness\n");
Al Viro2744c172012-09-26 21:41:05 -04001314 d_drop(dn);
Sage Weil355da1e2009-10-06 11:31:08 -07001315 dput(dn);
Dan Carpenterac1f12e2010-08-25 09:11:35 +02001316 err = PTR_ERR(in);
Sage Weil355da1e2009-10-06 11:31:08 -07001317 goto out;
1318 }
Sage Weil467c5252010-09-13 11:39:20 -07001319 dn = splice_dentry(dn, in, NULL, false);
Sage Weild69ed052010-06-21 10:38:14 -07001320 if (IS_ERR(dn))
1321 dn = NULL;
Sage Weil355da1e2009-10-06 11:31:08 -07001322 }
1323
1324 if (fill_inode(in, &rinfo->dir_in[i], NULL, session,
1325 req->r_request_started, -1,
1326 &req->r_caps_reservation) < 0) {
1327 pr_err("fill_inode badness on %p\n", in);
Sage Weild69ed052010-06-21 10:38:14 -07001328 goto next_item;
Sage Weil355da1e2009-10-06 11:31:08 -07001329 }
Sage Weild69ed052010-06-21 10:38:14 -07001330 if (dn)
1331 update_dentry_lease(dn, rinfo->dir_dlease[i],
1332 req->r_session,
1333 req->r_request_started);
1334next_item:
1335 if (dn)
1336 dput(dn);
Sage Weil355da1e2009-10-06 11:31:08 -07001337 }
1338 req->r_did_prepopulate = true;
1339
1340out:
1341 if (snapdir) {
1342 iput(snapdir);
1343 dput(parent);
1344 }
1345 dout("readdir_prepopulate done\n");
1346 return err;
1347}
1348
1349int ceph_inode_set_size(struct inode *inode, loff_t size)
1350{
1351 struct ceph_inode_info *ci = ceph_inode(inode);
1352 int ret = 0;
1353
Sage Weilbe655592011-11-30 09:47:09 -08001354 spin_lock(&ci->i_ceph_lock);
Sage Weil355da1e2009-10-06 11:31:08 -07001355 dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size);
1356 inode->i_size = size;
1357 inode->i_blocks = (size + (1 << 9) - 1) >> 9;
1358
1359 /* tell the MDS if we are approaching max_size */
1360 if ((size << 1) >= ci->i_max_size &&
1361 (ci->i_reported_size << 1) < ci->i_max_size)
1362 ret = 1;
1363
Sage Weilbe655592011-11-30 09:47:09 -08001364 spin_unlock(&ci->i_ceph_lock);
Sage Weil355da1e2009-10-06 11:31:08 -07001365 return ret;
1366}
1367
1368/*
1369 * Write back inode data in a worker thread. (This can't be done
1370 * in the message handler context.)
1371 */
Sage Weil3c6f6b72010-02-09 15:24:44 -08001372void ceph_queue_writeback(struct inode *inode)
1373{
Sage Weil15a20152011-11-05 22:06:31 -07001374 ihold(inode);
Sage Weil3c6f6b72010-02-09 15:24:44 -08001375 if (queue_work(ceph_inode_to_client(inode)->wb_wq,
1376 &ceph_inode(inode)->i_wb_work)) {
Sage Weil2c27c9a2010-02-17 15:45:51 -08001377 dout("ceph_queue_writeback %p\n", inode);
Sage Weil3c6f6b72010-02-09 15:24:44 -08001378 } else {
Sage Weil2c27c9a2010-02-17 15:45:51 -08001379 dout("ceph_queue_writeback %p failed\n", inode);
Sage Weil15a20152011-11-05 22:06:31 -07001380 iput(inode);
Sage Weil3c6f6b72010-02-09 15:24:44 -08001381 }
1382}
1383
1384static void ceph_writeback_work(struct work_struct *work)
Sage Weil355da1e2009-10-06 11:31:08 -07001385{
1386 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1387 i_wb_work);
1388 struct inode *inode = &ci->vfs_inode;
1389
1390 dout("writeback %p\n", inode);
1391 filemap_fdatawrite(&inode->i_data);
1392 iput(inode);
1393}
1394
1395/*
Sage Weil3c6f6b72010-02-09 15:24:44 -08001396 * queue an async invalidation
1397 */
1398void ceph_queue_invalidate(struct inode *inode)
1399{
Sage Weil15a20152011-11-05 22:06:31 -07001400 ihold(inode);
Sage Weil3c6f6b72010-02-09 15:24:44 -08001401 if (queue_work(ceph_inode_to_client(inode)->pg_inv_wq,
1402 &ceph_inode(inode)->i_pg_inv_work)) {
1403 dout("ceph_queue_invalidate %p\n", inode);
Sage Weil3c6f6b72010-02-09 15:24:44 -08001404 } else {
1405 dout("ceph_queue_invalidate %p failed\n", inode);
Sage Weil15a20152011-11-05 22:06:31 -07001406 iput(inode);
Sage Weil3c6f6b72010-02-09 15:24:44 -08001407 }
1408}
1409
1410/*
Sage Weil355da1e2009-10-06 11:31:08 -07001411 * Invalidate inode pages in a worker thread. (This can't be done
1412 * in the message handler context.)
1413 */
Sage Weil3c6f6b72010-02-09 15:24:44 -08001414static void ceph_invalidate_work(struct work_struct *work)
Sage Weil355da1e2009-10-06 11:31:08 -07001415{
1416 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1417 i_pg_inv_work);
1418 struct inode *inode = &ci->vfs_inode;
1419 u32 orig_gen;
1420 int check = 0;
1421
Sage Weilbe655592011-11-30 09:47:09 -08001422 spin_lock(&ci->i_ceph_lock);
Sage Weil355da1e2009-10-06 11:31:08 -07001423 dout("invalidate_pages %p gen %d revoking %d\n", inode,
1424 ci->i_rdcache_gen, ci->i_rdcache_revoking);
Sage Weilcd045cb2010-11-04 11:05:05 -07001425 if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
Sage Weil355da1e2009-10-06 11:31:08 -07001426 /* nevermind! */
Sage Weilbe655592011-11-30 09:47:09 -08001427 spin_unlock(&ci->i_ceph_lock);
Sage Weil355da1e2009-10-06 11:31:08 -07001428 goto out;
1429 }
1430 orig_gen = ci->i_rdcache_gen;
Sage Weilbe655592011-11-30 09:47:09 -08001431 spin_unlock(&ci->i_ceph_lock);
Sage Weil355da1e2009-10-06 11:31:08 -07001432
Sage Weil83eaea22011-08-24 14:07:01 -07001433 truncate_inode_pages(&inode->i_data, 0);
Sage Weil355da1e2009-10-06 11:31:08 -07001434
Sage Weilbe655592011-11-30 09:47:09 -08001435 spin_lock(&ci->i_ceph_lock);
Sage Weilcd045cb2010-11-04 11:05:05 -07001436 if (orig_gen == ci->i_rdcache_gen &&
1437 orig_gen == ci->i_rdcache_revoking) {
Sage Weil355da1e2009-10-06 11:31:08 -07001438 dout("invalidate_pages %p gen %d successful\n", inode,
1439 ci->i_rdcache_gen);
Sage Weilcd045cb2010-11-04 11:05:05 -07001440 ci->i_rdcache_revoking--;
Sage Weil355da1e2009-10-06 11:31:08 -07001441 check = 1;
1442 } else {
Sage Weilcd045cb2010-11-04 11:05:05 -07001443 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1444 inode, orig_gen, ci->i_rdcache_gen,
1445 ci->i_rdcache_revoking);
Sage Weil355da1e2009-10-06 11:31:08 -07001446 }
Sage Weilbe655592011-11-30 09:47:09 -08001447 spin_unlock(&ci->i_ceph_lock);
Sage Weil355da1e2009-10-06 11:31:08 -07001448
1449 if (check)
1450 ceph_check_caps(ci, 0, NULL);
1451out:
1452 iput(inode);
1453}
1454
1455
1456/*
Yan, Zheng3f999692013-03-01 10:57:54 +08001457 * called by trunc_wq;
Sage Weil355da1e2009-10-06 11:31:08 -07001458 *
1459 * We also truncate in a separate thread as well.
1460 */
Sage Weil3c6f6b72010-02-09 15:24:44 -08001461static void ceph_vmtruncate_work(struct work_struct *work)
Sage Weil355da1e2009-10-06 11:31:08 -07001462{
1463 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1464 i_vmtruncate_work);
1465 struct inode *inode = &ci->vfs_inode;
1466
1467 dout("vmtruncate_work %p\n", inode);
Yan, Zhengb415bf42013-07-02 12:40:19 +08001468 mutex_lock(&inode->i_mutex);
1469 __ceph_do_pending_vmtruncate(inode);
1470 mutex_unlock(&inode->i_mutex);
Sage Weil355da1e2009-10-06 11:31:08 -07001471 iput(inode);
1472}
1473
1474/*
Sage Weil3c6f6b72010-02-09 15:24:44 -08001475 * Queue an async vmtruncate. If we fail to queue work, we will handle
1476 * the truncation the next time we call __ceph_do_pending_vmtruncate.
1477 */
1478void ceph_queue_vmtruncate(struct inode *inode)
1479{
1480 struct ceph_inode_info *ci = ceph_inode(inode);
1481
Sage Weil15a20152011-11-05 22:06:31 -07001482 ihold(inode);
Cheng Renquan640ef792010-03-26 17:40:33 +08001483 if (queue_work(ceph_sb_to_client(inode->i_sb)->trunc_wq,
Sage Weil3c6f6b72010-02-09 15:24:44 -08001484 &ci->i_vmtruncate_work)) {
1485 dout("ceph_queue_vmtruncate %p\n", inode);
Sage Weil3c6f6b72010-02-09 15:24:44 -08001486 } else {
1487 dout("ceph_queue_vmtruncate %p failed, pending=%d\n",
1488 inode, ci->i_truncate_pending);
Sage Weil15a20152011-11-05 22:06:31 -07001489 iput(inode);
Sage Weil3c6f6b72010-02-09 15:24:44 -08001490 }
1491}
1492
1493/*
Sage Weil355da1e2009-10-06 11:31:08 -07001494 * Make sure any pending truncation is applied before doing anything
1495 * that may depend on it.
1496 */
Yan, Zhengb415bf42013-07-02 12:40:19 +08001497void __ceph_do_pending_vmtruncate(struct inode *inode)
Sage Weil355da1e2009-10-06 11:31:08 -07001498{
1499 struct ceph_inode_info *ci = ceph_inode(inode);
1500 u64 to;
Yan, Zhenga85f50b2012-11-19 10:49:08 +08001501 int wrbuffer_refs, finish = 0;
Sage Weil355da1e2009-10-06 11:31:08 -07001502
1503retry:
Sage Weilbe655592011-11-30 09:47:09 -08001504 spin_lock(&ci->i_ceph_lock);
Sage Weil355da1e2009-10-06 11:31:08 -07001505 if (ci->i_truncate_pending == 0) {
1506 dout("__do_pending_vmtruncate %p none pending\n", inode);
Sage Weilbe655592011-11-30 09:47:09 -08001507 spin_unlock(&ci->i_ceph_lock);
Sage Weil355da1e2009-10-06 11:31:08 -07001508 return;
1509 }
1510
1511 /*
1512 * make sure any dirty snapped pages are flushed before we
1513 * possibly truncate them.. so write AND block!
1514 */
1515 if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) {
1516 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1517 inode);
Sage Weilbe655592011-11-30 09:47:09 -08001518 spin_unlock(&ci->i_ceph_lock);
Sage Weil355da1e2009-10-06 11:31:08 -07001519 filemap_write_and_wait_range(&inode->i_data, 0,
1520 inode->i_sb->s_maxbytes);
1521 goto retry;
1522 }
1523
1524 to = ci->i_truncate_size;
1525 wrbuffer_refs = ci->i_wrbuffer_ref;
1526 dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode,
1527 ci->i_truncate_pending, to);
Sage Weilbe655592011-11-30 09:47:09 -08001528 spin_unlock(&ci->i_ceph_lock);
Sage Weil355da1e2009-10-06 11:31:08 -07001529
1530 truncate_inode_pages(inode->i_mapping, to);
1531
Sage Weilbe655592011-11-30 09:47:09 -08001532 spin_lock(&ci->i_ceph_lock);
Yan, Zhenga85f50b2012-11-19 10:49:08 +08001533 if (to == ci->i_truncate_size) {
1534 ci->i_truncate_pending = 0;
1535 finish = 1;
1536 }
Sage Weilbe655592011-11-30 09:47:09 -08001537 spin_unlock(&ci->i_ceph_lock);
Yan, Zhenga85f50b2012-11-19 10:49:08 +08001538 if (!finish)
1539 goto retry;
Sage Weil355da1e2009-10-06 11:31:08 -07001540
1541 if (wrbuffer_refs == 0)
1542 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
Yan, Zhenga85f50b2012-11-19 10:49:08 +08001543
1544 wake_up_all(&ci->i_cap_wq);
Sage Weil355da1e2009-10-06 11:31:08 -07001545}
1546
1547
1548/*
1549 * symlinks
1550 */
1551static void *ceph_sym_follow_link(struct dentry *dentry, struct nameidata *nd)
1552{
1553 struct ceph_inode_info *ci = ceph_inode(dentry->d_inode);
1554 nd_set_link(nd, ci->i_symlink);
1555 return NULL;
1556}
1557
1558static const struct inode_operations ceph_symlink_iops = {
1559 .readlink = generic_readlink,
1560 .follow_link = ceph_sym_follow_link,
Yan, Zheng0b932672013-04-07 16:28:49 +08001561 .setattr = ceph_setattr,
1562 .getattr = ceph_getattr,
1563 .setxattr = ceph_setxattr,
1564 .getxattr = ceph_getxattr,
1565 .listxattr = ceph_listxattr,
1566 .removexattr = ceph_removexattr,
Sage Weil355da1e2009-10-06 11:31:08 -07001567};
1568
1569/*
1570 * setattr
1571 */
1572int ceph_setattr(struct dentry *dentry, struct iattr *attr)
1573{
1574 struct inode *inode = dentry->d_inode;
1575 struct ceph_inode_info *ci = ceph_inode(inode);
Sage Weil5f21c962011-07-26 11:30:29 -07001576 struct inode *parent_inode;
Sage Weil355da1e2009-10-06 11:31:08 -07001577 const unsigned int ia_valid = attr->ia_valid;
1578 struct ceph_mds_request *req;
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001579 struct ceph_mds_client *mdsc = ceph_sb_to_client(dentry->d_sb)->mdsc;
Sage Weil355da1e2009-10-06 11:31:08 -07001580 int issued;
1581 int release = 0, dirtied = 0;
1582 int mask = 0;
1583 int err = 0;
Sage Weilfca65b42011-05-04 11:33:47 -07001584 int inode_dirty_flags = 0;
Sage Weil355da1e2009-10-06 11:31:08 -07001585
1586 if (ceph_snap(inode) != CEPH_NOSNAP)
1587 return -EROFS;
1588
Yan, Zhengb415bf42013-07-02 12:40:19 +08001589 __ceph_do_pending_vmtruncate(inode);
Sage Weil355da1e2009-10-06 11:31:08 -07001590
1591 err = inode_change_ok(inode, attr);
1592 if (err != 0)
1593 return err;
1594
1595 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR,
1596 USE_AUTH_MDS);
1597 if (IS_ERR(req))
1598 return PTR_ERR(req);
1599
Sage Weilbe655592011-11-30 09:47:09 -08001600 spin_lock(&ci->i_ceph_lock);
Sage Weil355da1e2009-10-06 11:31:08 -07001601 issued = __ceph_caps_issued(ci, NULL);
1602 dout("setattr %p issued %s\n", inode, ceph_cap_string(issued));
1603
1604 if (ia_valid & ATTR_UID) {
1605 dout("setattr %p uid %d -> %d\n", inode,
Eric W. Biedermanbd2bae62013-01-31 04:05:39 -08001606 from_kuid(&init_user_ns, inode->i_uid),
1607 from_kuid(&init_user_ns, attr->ia_uid));
Sage Weil355da1e2009-10-06 11:31:08 -07001608 if (issued & CEPH_CAP_AUTH_EXCL) {
1609 inode->i_uid = attr->ia_uid;
1610 dirtied |= CEPH_CAP_AUTH_EXCL;
1611 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
Eric W. Biedermanab871b92013-01-31 03:40:12 -08001612 !uid_eq(attr->ia_uid, inode->i_uid)) {
1613 req->r_args.setattr.uid = cpu_to_le32(
1614 from_kuid(&init_user_ns, attr->ia_uid));
Sage Weil355da1e2009-10-06 11:31:08 -07001615 mask |= CEPH_SETATTR_UID;
1616 release |= CEPH_CAP_AUTH_SHARED;
1617 }
1618 }
1619 if (ia_valid & ATTR_GID) {
1620 dout("setattr %p gid %d -> %d\n", inode,
Eric W. Biedermanbd2bae62013-01-31 04:05:39 -08001621 from_kgid(&init_user_ns, inode->i_gid),
1622 from_kgid(&init_user_ns, attr->ia_gid));
Sage Weil355da1e2009-10-06 11:31:08 -07001623 if (issued & CEPH_CAP_AUTH_EXCL) {
1624 inode->i_gid = attr->ia_gid;
1625 dirtied |= CEPH_CAP_AUTH_EXCL;
1626 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
Eric W. Biedermanab871b92013-01-31 03:40:12 -08001627 !gid_eq(attr->ia_gid, inode->i_gid)) {
1628 req->r_args.setattr.gid = cpu_to_le32(
1629 from_kgid(&init_user_ns, attr->ia_gid));
Sage Weil355da1e2009-10-06 11:31:08 -07001630 mask |= CEPH_SETATTR_GID;
1631 release |= CEPH_CAP_AUTH_SHARED;
1632 }
1633 }
1634 if (ia_valid & ATTR_MODE) {
1635 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode,
1636 attr->ia_mode);
1637 if (issued & CEPH_CAP_AUTH_EXCL) {
1638 inode->i_mode = attr->ia_mode;
1639 dirtied |= CEPH_CAP_AUTH_EXCL;
1640 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1641 attr->ia_mode != inode->i_mode) {
1642 req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode);
1643 mask |= CEPH_SETATTR_MODE;
1644 release |= CEPH_CAP_AUTH_SHARED;
1645 }
1646 }
1647
1648 if (ia_valid & ATTR_ATIME) {
1649 dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode,
1650 inode->i_atime.tv_sec, inode->i_atime.tv_nsec,
1651 attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec);
1652 if (issued & CEPH_CAP_FILE_EXCL) {
1653 ci->i_time_warp_seq++;
1654 inode->i_atime = attr->ia_atime;
1655 dirtied |= CEPH_CAP_FILE_EXCL;
1656 } else if ((issued & CEPH_CAP_FILE_WR) &&
1657 timespec_compare(&inode->i_atime,
1658 &attr->ia_atime) < 0) {
1659 inode->i_atime = attr->ia_atime;
1660 dirtied |= CEPH_CAP_FILE_WR;
1661 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1662 !timespec_equal(&inode->i_atime, &attr->ia_atime)) {
1663 ceph_encode_timespec(&req->r_args.setattr.atime,
1664 &attr->ia_atime);
1665 mask |= CEPH_SETATTR_ATIME;
1666 release |= CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_RD |
1667 CEPH_CAP_FILE_WR;
1668 }
1669 }
1670 if (ia_valid & ATTR_MTIME) {
1671 dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode,
1672 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
1673 attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec);
1674 if (issued & CEPH_CAP_FILE_EXCL) {
1675 ci->i_time_warp_seq++;
1676 inode->i_mtime = attr->ia_mtime;
1677 dirtied |= CEPH_CAP_FILE_EXCL;
1678 } else if ((issued & CEPH_CAP_FILE_WR) &&
1679 timespec_compare(&inode->i_mtime,
1680 &attr->ia_mtime) < 0) {
1681 inode->i_mtime = attr->ia_mtime;
1682 dirtied |= CEPH_CAP_FILE_WR;
1683 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1684 !timespec_equal(&inode->i_mtime, &attr->ia_mtime)) {
1685 ceph_encode_timespec(&req->r_args.setattr.mtime,
1686 &attr->ia_mtime);
1687 mask |= CEPH_SETATTR_MTIME;
1688 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
1689 CEPH_CAP_FILE_WR;
1690 }
1691 }
1692 if (ia_valid & ATTR_SIZE) {
1693 dout("setattr %p size %lld -> %lld\n", inode,
1694 inode->i_size, attr->ia_size);
1695 if (attr->ia_size > inode->i_sb->s_maxbytes) {
1696 err = -EINVAL;
1697 goto out;
1698 }
1699 if ((issued & CEPH_CAP_FILE_EXCL) &&
1700 attr->ia_size > inode->i_size) {
1701 inode->i_size = attr->ia_size;
Sage Weil355da1e2009-10-06 11:31:08 -07001702 inode->i_blocks =
1703 (attr->ia_size + (1 << 9) - 1) >> 9;
1704 inode->i_ctime = attr->ia_ctime;
1705 ci->i_reported_size = attr->ia_size;
1706 dirtied |= CEPH_CAP_FILE_EXCL;
1707 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1708 attr->ia_size != inode->i_size) {
1709 req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
1710 req->r_args.setattr.old_size =
1711 cpu_to_le64(inode->i_size);
1712 mask |= CEPH_SETATTR_SIZE;
1713 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
1714 CEPH_CAP_FILE_WR;
1715 }
1716 }
1717
1718 /* these do nothing */
1719 if (ia_valid & ATTR_CTIME) {
1720 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME|
1721 ATTR_MODE|ATTR_UID|ATTR_GID)) == 0;
1722 dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode,
1723 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
1724 attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec,
1725 only ? "ctime only" : "ignored");
1726 inode->i_ctime = attr->ia_ctime;
1727 if (only) {
1728 /*
1729 * if kernel wants to dirty ctime but nothing else,
1730 * we need to choose a cap to dirty under, or do
1731 * a almost-no-op setattr
1732 */
1733 if (issued & CEPH_CAP_AUTH_EXCL)
1734 dirtied |= CEPH_CAP_AUTH_EXCL;
1735 else if (issued & CEPH_CAP_FILE_EXCL)
1736 dirtied |= CEPH_CAP_FILE_EXCL;
1737 else if (issued & CEPH_CAP_XATTR_EXCL)
1738 dirtied |= CEPH_CAP_XATTR_EXCL;
1739 else
1740 mask |= CEPH_SETATTR_CTIME;
1741 }
1742 }
1743 if (ia_valid & ATTR_FILE)
1744 dout("setattr %p ATTR_FILE ... hrm!\n", inode);
1745
1746 if (dirtied) {
Sage Weilfca65b42011-05-04 11:33:47 -07001747 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied);
Sage Weil355da1e2009-10-06 11:31:08 -07001748 inode->i_ctime = CURRENT_TIME;
1749 }
1750
1751 release &= issued;
Sage Weilbe655592011-11-30 09:47:09 -08001752 spin_unlock(&ci->i_ceph_lock);
Sage Weil355da1e2009-10-06 11:31:08 -07001753
Sage Weilfca65b42011-05-04 11:33:47 -07001754 if (inode_dirty_flags)
1755 __mark_inode_dirty(inode, inode_dirty_flags);
1756
Sage Weil355da1e2009-10-06 11:31:08 -07001757 if (mask) {
Sage Weil70b666c2011-05-27 09:24:26 -07001758 req->r_inode = inode;
1759 ihold(inode);
Sage Weil355da1e2009-10-06 11:31:08 -07001760 req->r_inode_drop = release;
1761 req->r_args.setattr.mask = cpu_to_le32(mask);
1762 req->r_num_caps = 1;
Sage Weil5f21c962011-07-26 11:30:29 -07001763 parent_inode = ceph_get_dentry_parent_inode(dentry);
Sage Weil355da1e2009-10-06 11:31:08 -07001764 err = ceph_mdsc_do_request(mdsc, parent_inode, req);
Sage Weil5f21c962011-07-26 11:30:29 -07001765 iput(parent_inode);
Sage Weil355da1e2009-10-06 11:31:08 -07001766 }
1767 dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err,
1768 ceph_cap_string(dirtied), mask);
1769
1770 ceph_mdsc_put_request(req);
Yan, Zhengb415bf42013-07-02 12:40:19 +08001771 __ceph_do_pending_vmtruncate(inode);
Sage Weil355da1e2009-10-06 11:31:08 -07001772 return err;
1773out:
Sage Weilbe655592011-11-30 09:47:09 -08001774 spin_unlock(&ci->i_ceph_lock);
Sage Weil355da1e2009-10-06 11:31:08 -07001775 ceph_mdsc_put_request(req);
1776 return err;
1777}
1778
1779/*
1780 * Verify that we have a lease on the given mask. If not,
1781 * do a getattr against an mds.
1782 */
1783int ceph_do_getattr(struct inode *inode, int mask)
1784{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001785 struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
1786 struct ceph_mds_client *mdsc = fsc->mdsc;
Sage Weil355da1e2009-10-06 11:31:08 -07001787 struct ceph_mds_request *req;
1788 int err;
1789
1790 if (ceph_snap(inode) == CEPH_SNAPDIR) {
1791 dout("do_getattr inode %p SNAPDIR\n", inode);
1792 return 0;
1793 }
1794
Sage Weilb7495fc2010-11-09 12:43:12 -08001795 dout("do_getattr inode %p mask %s mode 0%o\n", inode, ceph_cap_string(mask), inode->i_mode);
Sage Weil355da1e2009-10-06 11:31:08 -07001796 if (ceph_caps_issued_mask(ceph_inode(inode), mask, 1))
1797 return 0;
1798
1799 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
1800 if (IS_ERR(req))
1801 return PTR_ERR(req);
Sage Weil70b666c2011-05-27 09:24:26 -07001802 req->r_inode = inode;
1803 ihold(inode);
Sage Weil355da1e2009-10-06 11:31:08 -07001804 req->r_num_caps = 1;
1805 req->r_args.getattr.mask = cpu_to_le32(mask);
1806 err = ceph_mdsc_do_request(mdsc, NULL, req);
1807 ceph_mdsc_put_request(req);
1808 dout("do_getattr result=%d\n", err);
1809 return err;
1810}
1811
1812
1813/*
1814 * Check inode permissions. We verify we have a valid value for
1815 * the AUTH cap, then call the generic handler.
1816 */
Al Viro10556cb2011-06-20 19:28:19 -04001817int ceph_permission(struct inode *inode, int mask)
Sage Weil355da1e2009-10-06 11:31:08 -07001818{
Nick Pigginb74c79e2011-01-07 17:49:58 +11001819 int err;
1820
Al Viro10556cb2011-06-20 19:28:19 -04001821 if (mask & MAY_NOT_BLOCK)
Nick Pigginb74c79e2011-01-07 17:49:58 +11001822 return -ECHILD;
1823
1824 err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED);
Sage Weil355da1e2009-10-06 11:31:08 -07001825
1826 if (!err)
Al Viro2830ba72011-06-20 19:16:29 -04001827 err = generic_permission(inode, mask);
Sage Weil355da1e2009-10-06 11:31:08 -07001828 return err;
1829}
1830
1831/*
1832 * Get all attributes. Hopefully somedata we'll have a statlite()
1833 * and can limit the fields we require to be accurate.
1834 */
1835int ceph_getattr(struct vfsmount *mnt, struct dentry *dentry,
1836 struct kstat *stat)
1837{
1838 struct inode *inode = dentry->d_inode;
Sage Weil232d4b02009-10-21 11:21:49 -07001839 struct ceph_inode_info *ci = ceph_inode(inode);
Sage Weil355da1e2009-10-06 11:31:08 -07001840 int err;
1841
1842 err = ceph_do_getattr(inode, CEPH_STAT_CAP_INODE_ALL);
1843 if (!err) {
1844 generic_fillattr(inode, stat);
Yehuda Sadehad1fee92011-01-21 16:44:03 -08001845 stat->ino = ceph_translate_ino(inode->i_sb, inode->i_ino);
Sage Weil355da1e2009-10-06 11:31:08 -07001846 if (ceph_snap(inode) != CEPH_NOSNAP)
1847 stat->dev = ceph_snap(inode);
1848 else
1849 stat->dev = 0;
Sage Weil232d4b02009-10-21 11:21:49 -07001850 if (S_ISDIR(inode->i_mode)) {
Yehuda Sadeh1c1266b2011-01-12 16:53:27 -08001851 if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb),
1852 RBYTES))
1853 stat->size = ci->i_rbytes;
1854 else
1855 stat->size = ci->i_files + ci->i_subdirs;
Sage Weil232d4b02009-10-21 11:21:49 -07001856 stat->blocks = 0;
Sage Weil355da1e2009-10-06 11:31:08 -07001857 stat->blksize = 65536;
Sage Weil232d4b02009-10-21 11:21:49 -07001858 }
Sage Weil355da1e2009-10-06 11:31:08 -07001859 }
1860 return err;
1861}