blob: 8549a48115f71b23e1f35ef444caf3eb32dbced3 [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"
Milosz Tanski99ccbd22013-08-21 17:29:54 -040015#include "cache.h"
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -070016#include <linux/ceph/decode.h>
Sage Weil355da1e2009-10-06 11:31:08 -070017
18/*
19 * Ceph inode operations
20 *
21 * Implement basic inode helpers (get, alloc) and inode ops (getattr,
22 * setattr, etc.), xattr helpers, and helpers for assimilating
23 * metadata returned by the MDS into our cache.
24 *
25 * Also define helpers for doing asynchronous writeback, invalidation,
26 * and truncation for the benefit of those who can't afford to block
27 * (typically because they are in the message handler path).
28 */
29
30static const struct inode_operations ceph_symlink_iops;
31
Sage Weil3c6f6b72010-02-09 15:24:44 -080032static void ceph_invalidate_work(struct work_struct *work);
33static void ceph_writeback_work(struct work_struct *work);
34static void ceph_vmtruncate_work(struct work_struct *work);
Sage Weil355da1e2009-10-06 11:31:08 -070035
36/*
37 * find or create an inode, given the ceph ino number
38 */
Yehuda Sadehad1fee92011-01-21 16:44:03 -080039static int ceph_set_ino_cb(struct inode *inode, void *data)
40{
41 ceph_inode(inode)->i_vino = *(struct ceph_vino *)data;
42 inode->i_ino = ceph_vino_to_ino(*(struct ceph_vino *)data);
43 return 0;
44}
45
Sage Weil355da1e2009-10-06 11:31:08 -070046struct inode *ceph_get_inode(struct super_block *sb, struct ceph_vino vino)
47{
48 struct inode *inode;
49 ino_t t = ceph_vino_to_ino(vino);
50
51 inode = iget5_locked(sb, t, ceph_ino_compare, ceph_set_ino_cb, &vino);
52 if (inode == NULL)
53 return ERR_PTR(-ENOMEM);
54 if (inode->i_state & I_NEW) {
55 dout("get_inode created new inode %p %llx.%llx ino %llx\n",
56 inode, ceph_vinop(inode), (u64)inode->i_ino);
57 unlock_new_inode(inode);
58 }
59
60 dout("get_inode on %lu=%llx.%llx got %p\n", inode->i_ino, vino.ino,
61 vino.snap, inode);
62 return inode;
63}
64
65/*
66 * get/constuct snapdir inode for a given directory
67 */
68struct inode *ceph_get_snapdir(struct inode *parent)
69{
70 struct ceph_vino vino = {
71 .ino = ceph_ino(parent),
72 .snap = CEPH_SNAPDIR,
73 };
74 struct inode *inode = ceph_get_inode(parent->i_sb, vino);
Sage Weilb377ff12009-11-11 15:22:37 -080075 struct ceph_inode_info *ci = ceph_inode(inode);
Sage Weil355da1e2009-10-06 11:31:08 -070076
77 BUG_ON(!S_ISDIR(parent->i_mode));
78 if (IS_ERR(inode))
Julia Lawall7e34bc52010-05-22 12:01:14 +020079 return inode;
Sage Weil355da1e2009-10-06 11:31:08 -070080 inode->i_mode = parent->i_mode;
81 inode->i_uid = parent->i_uid;
82 inode->i_gid = parent->i_gid;
83 inode->i_op = &ceph_dir_iops;
84 inode->i_fop = &ceph_dir_fops;
Sage Weilb377ff12009-11-11 15:22:37 -080085 ci->i_snap_caps = CEPH_CAP_PIN; /* so we can open */
86 ci->i_rbytes = 0;
Sage Weil355da1e2009-10-06 11:31:08 -070087 return inode;
88}
89
90const struct inode_operations ceph_file_iops = {
91 .permission = ceph_permission,
92 .setattr = ceph_setattr,
93 .getattr = ceph_getattr,
94 .setxattr = ceph_setxattr,
95 .getxattr = ceph_getxattr,
96 .listxattr = ceph_listxattr,
97 .removexattr = ceph_removexattr,
98};
99
100
101/*
102 * We use a 'frag tree' to keep track of the MDS's directory fragments
103 * for a given inode (usually there is just a single fragment). We
104 * need to know when a child frag is delegated to a new MDS, or when
105 * it is flagged as replicated, so we can direct our requests
106 * accordingly.
107 */
108
109/*
110 * find/create a frag in the tree
111 */
112static struct ceph_inode_frag *__get_or_create_frag(struct ceph_inode_info *ci,
113 u32 f)
114{
115 struct rb_node **p;
116 struct rb_node *parent = NULL;
117 struct ceph_inode_frag *frag;
118 int c;
119
120 p = &ci->i_fragtree.rb_node;
121 while (*p) {
122 parent = *p;
123 frag = rb_entry(parent, struct ceph_inode_frag, node);
124 c = ceph_frag_compare(f, frag->frag);
125 if (c < 0)
126 p = &(*p)->rb_left;
127 else if (c > 0)
128 p = &(*p)->rb_right;
129 else
130 return frag;
131 }
132
133 frag = kmalloc(sizeof(*frag), GFP_NOFS);
134 if (!frag) {
135 pr_err("__get_or_create_frag ENOMEM on %p %llx.%llx "
136 "frag %x\n", &ci->vfs_inode,
137 ceph_vinop(&ci->vfs_inode), f);
138 return ERR_PTR(-ENOMEM);
139 }
140 frag->frag = f;
141 frag->split_by = 0;
142 frag->mds = -1;
143 frag->ndist = 0;
144
145 rb_link_node(&frag->node, parent, p);
146 rb_insert_color(&frag->node, &ci->i_fragtree);
147
148 dout("get_or_create_frag added %llx.%llx frag %x\n",
149 ceph_vinop(&ci->vfs_inode), f);
150 return frag;
151}
152
153/*
154 * find a specific frag @f
155 */
156struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci, u32 f)
157{
158 struct rb_node *n = ci->i_fragtree.rb_node;
159
160 while (n) {
161 struct ceph_inode_frag *frag =
162 rb_entry(n, struct ceph_inode_frag, node);
163 int c = ceph_frag_compare(f, frag->frag);
164 if (c < 0)
165 n = n->rb_left;
166 else if (c > 0)
167 n = n->rb_right;
168 else
169 return frag;
170 }
171 return NULL;
172}
173
174/*
175 * Choose frag containing the given value @v. If @pfrag is
176 * specified, copy the frag delegation info to the caller if
177 * it is present.
178 */
179u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
180 struct ceph_inode_frag *pfrag,
181 int *found)
182{
183 u32 t = ceph_frag_make(0, 0);
184 struct ceph_inode_frag *frag;
185 unsigned nway, i;
186 u32 n;
187
188 if (found)
189 *found = 0;
190
191 mutex_lock(&ci->i_fragtree_mutex);
192 while (1) {
193 WARN_ON(!ceph_frag_contains_value(t, v));
194 frag = __ceph_find_frag(ci, t);
195 if (!frag)
196 break; /* t is a leaf */
197 if (frag->split_by == 0) {
198 if (pfrag)
199 memcpy(pfrag, frag, sizeof(*pfrag));
200 if (found)
201 *found = 1;
202 break;
203 }
204
205 /* choose child */
206 nway = 1 << frag->split_by;
207 dout("choose_frag(%x) %x splits by %d (%d ways)\n", v, t,
208 frag->split_by, nway);
209 for (i = 0; i < nway; i++) {
210 n = ceph_frag_make_child(t, frag->split_by, i);
211 if (ceph_frag_contains_value(n, v)) {
212 t = n;
213 break;
214 }
215 }
216 BUG_ON(i == nway);
217 }
218 dout("choose_frag(%x) = %x\n", v, t);
219
220 mutex_unlock(&ci->i_fragtree_mutex);
221 return t;
222}
223
224/*
225 * Process dirfrag (delegation) info from the mds. Include leaf
226 * fragment in tree ONLY if ndist > 0. Otherwise, only
227 * branches/splits are included in i_fragtree)
228 */
229static int ceph_fill_dirfrag(struct inode *inode,
230 struct ceph_mds_reply_dirfrag *dirinfo)
231{
232 struct ceph_inode_info *ci = ceph_inode(inode);
233 struct ceph_inode_frag *frag;
234 u32 id = le32_to_cpu(dirinfo->frag);
235 int mds = le32_to_cpu(dirinfo->auth);
236 int ndist = le32_to_cpu(dirinfo->ndist);
237 int i;
238 int err = 0;
239
240 mutex_lock(&ci->i_fragtree_mutex);
241 if (ndist == 0) {
242 /* no delegation info needed. */
243 frag = __ceph_find_frag(ci, id);
244 if (!frag)
245 goto out;
246 if (frag->split_by == 0) {
247 /* tree leaf, remove */
248 dout("fill_dirfrag removed %llx.%llx frag %x"
249 " (no ref)\n", ceph_vinop(inode), id);
250 rb_erase(&frag->node, &ci->i_fragtree);
251 kfree(frag);
252 } else {
253 /* tree branch, keep and clear */
254 dout("fill_dirfrag cleared %llx.%llx frag %x"
255 " referral\n", ceph_vinop(inode), id);
256 frag->mds = -1;
257 frag->ndist = 0;
258 }
259 goto out;
260 }
261
262
263 /* find/add this frag to store mds delegation info */
264 frag = __get_or_create_frag(ci, id);
265 if (IS_ERR(frag)) {
266 /* this is not the end of the world; we can continue
267 with bad/inaccurate delegation info */
268 pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n",
269 ceph_vinop(inode), le32_to_cpu(dirinfo->frag));
270 err = -ENOMEM;
271 goto out;
272 }
273
274 frag->mds = mds;
275 frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP);
276 for (i = 0; i < frag->ndist; i++)
277 frag->dist[i] = le32_to_cpu(dirinfo->dist[i]);
278 dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n",
279 ceph_vinop(inode), frag->frag, frag->ndist);
280
281out:
282 mutex_unlock(&ci->i_fragtree_mutex);
283 return err;
284}
285
286
287/*
288 * initialize a newly allocated inode.
289 */
290struct inode *ceph_alloc_inode(struct super_block *sb)
291{
292 struct ceph_inode_info *ci;
293 int i;
294
295 ci = kmem_cache_alloc(ceph_inode_cachep, GFP_NOFS);
296 if (!ci)
297 return NULL;
298
299 dout("alloc_inode %p\n", &ci->vfs_inode);
300
Sage Weilbe655592011-11-30 09:47:09 -0800301 spin_lock_init(&ci->i_ceph_lock);
302
Sage Weil355da1e2009-10-06 11:31:08 -0700303 ci->i_version = 0;
304 ci->i_time_warp_seq = 0;
305 ci->i_ceph_flags = 0;
Yan, Zheng2f276c52013-03-13 19:44:32 +0800306 atomic_set(&ci->i_release_count, 1);
307 atomic_set(&ci->i_complete_count, 0);
Sage Weil355da1e2009-10-06 11:31:08 -0700308 ci->i_symlink = NULL;
309
Sage Weil6c0f3af2010-11-16 11:14:34 -0800310 memset(&ci->i_dir_layout, 0, sizeof(ci->i_dir_layout));
311
Sage Weil355da1e2009-10-06 11:31:08 -0700312 ci->i_fragtree = RB_ROOT;
313 mutex_init(&ci->i_fragtree_mutex);
314
315 ci->i_xattrs.blob = NULL;
316 ci->i_xattrs.prealloc_blob = NULL;
317 ci->i_xattrs.dirty = false;
318 ci->i_xattrs.index = RB_ROOT;
319 ci->i_xattrs.count = 0;
320 ci->i_xattrs.names_size = 0;
321 ci->i_xattrs.vals_size = 0;
322 ci->i_xattrs.version = 0;
323 ci->i_xattrs.index_version = 0;
324
325 ci->i_caps = RB_ROOT;
326 ci->i_auth_cap = NULL;
327 ci->i_dirty_caps = 0;
328 ci->i_flushing_caps = 0;
329 INIT_LIST_HEAD(&ci->i_dirty_item);
330 INIT_LIST_HEAD(&ci->i_flushing_item);
331 ci->i_cap_flush_seq = 0;
332 ci->i_cap_flush_last_tid = 0;
333 memset(&ci->i_cap_flush_tid, 0, sizeof(ci->i_cap_flush_tid));
334 init_waitqueue_head(&ci->i_cap_wq);
335 ci->i_hold_caps_min = 0;
336 ci->i_hold_caps_max = 0;
337 INIT_LIST_HEAD(&ci->i_cap_delay_list);
338 ci->i_cap_exporting_mds = 0;
339 ci->i_cap_exporting_mseq = 0;
340 ci->i_cap_exporting_issued = 0;
341 INIT_LIST_HEAD(&ci->i_cap_snaps);
342 ci->i_head_snapc = NULL;
343 ci->i_snap_caps = 0;
344
345 for (i = 0; i < CEPH_FILE_MODE_NUM; i++)
346 ci->i_nr_by_mode[i] = 0;
347
Yan, Zhengb0d7c222013-08-12 21:42:15 -0700348 mutex_init(&ci->i_truncate_mutex);
Sage Weil355da1e2009-10-06 11:31:08 -0700349 ci->i_truncate_seq = 0;
350 ci->i_truncate_size = 0;
351 ci->i_truncate_pending = 0;
352
353 ci->i_max_size = 0;
354 ci->i_reported_size = 0;
355 ci->i_wanted_max_size = 0;
356 ci->i_requested_max_size = 0;
357
358 ci->i_pin_ref = 0;
359 ci->i_rd_ref = 0;
360 ci->i_rdcache_ref = 0;
361 ci->i_wr_ref = 0;
Henry C Changd3d07202011-05-11 10:29:54 +0000362 ci->i_wb_ref = 0;
Sage Weil355da1e2009-10-06 11:31:08 -0700363 ci->i_wrbuffer_ref = 0;
364 ci->i_wrbuffer_ref_head = 0;
365 ci->i_shared_gen = 0;
366 ci->i_rdcache_gen = 0;
367 ci->i_rdcache_revoking = 0;
368
369 INIT_LIST_HEAD(&ci->i_unsafe_writes);
370 INIT_LIST_HEAD(&ci->i_unsafe_dirops);
371 spin_lock_init(&ci->i_unsafe_lock);
372
373 ci->i_snap_realm = NULL;
374 INIT_LIST_HEAD(&ci->i_snap_realm_item);
375 INIT_LIST_HEAD(&ci->i_snap_flush_item);
376
Sage Weil3c6f6b72010-02-09 15:24:44 -0800377 INIT_WORK(&ci->i_wb_work, ceph_writeback_work);
378 INIT_WORK(&ci->i_pg_inv_work, ceph_invalidate_work);
Sage Weil355da1e2009-10-06 11:31:08 -0700379
380 INIT_WORK(&ci->i_vmtruncate_work, ceph_vmtruncate_work);
381
Milosz Tanski99ccbd22013-08-21 17:29:54 -0400382 ceph_fscache_inode_init(ci);
383
Sage Weil355da1e2009-10-06 11:31:08 -0700384 return &ci->vfs_inode;
385}
386
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100387static void ceph_i_callback(struct rcu_head *head)
388{
389 struct inode *inode = container_of(head, struct inode, i_rcu);
390 struct ceph_inode_info *ci = ceph_inode(inode);
391
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100392 kmem_cache_free(ceph_inode_cachep, ci);
393}
394
Sage Weil355da1e2009-10-06 11:31:08 -0700395void ceph_destroy_inode(struct inode *inode)
396{
397 struct ceph_inode_info *ci = ceph_inode(inode);
398 struct ceph_inode_frag *frag;
399 struct rb_node *n;
400
401 dout("destroy_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode));
402
Milosz Tanski99ccbd22013-08-21 17:29:54 -0400403 ceph_fscache_unregister_inode_cookie(ci);
404
Sage Weil355da1e2009-10-06 11:31:08 -0700405 ceph_queue_caps_release(inode);
406
Sage Weil8b218b82010-03-09 12:59:08 -0800407 /*
408 * we may still have a snap_realm reference if there are stray
409 * caps in i_cap_exporting_issued or i_snap_caps.
410 */
411 if (ci->i_snap_realm) {
412 struct ceph_mds_client *mdsc =
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700413 ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc;
Sage Weil8b218b82010-03-09 12:59:08 -0800414 struct ceph_snap_realm *realm = ci->i_snap_realm;
415
416 dout(" dropping residual ref to snap realm %p\n", realm);
417 spin_lock(&realm->inodes_with_caps_lock);
418 list_del_init(&ci->i_snap_realm_item);
419 spin_unlock(&realm->inodes_with_caps_lock);
420 ceph_put_snap_realm(mdsc, realm);
421 }
422
Sage Weil355da1e2009-10-06 11:31:08 -0700423 kfree(ci->i_symlink);
424 while ((n = rb_first(&ci->i_fragtree)) != NULL) {
425 frag = rb_entry(n, struct ceph_inode_frag, node);
426 rb_erase(n, &ci->i_fragtree);
427 kfree(frag);
428 }
429
430 __ceph_destroy_xattrs(ci);
Sage Weilb6c1d5b2009-12-07 12:17:17 -0800431 if (ci->i_xattrs.blob)
432 ceph_buffer_put(ci->i_xattrs.blob);
433 if (ci->i_xattrs.prealloc_blob)
434 ceph_buffer_put(ci->i_xattrs.prealloc_blob);
Sage Weil355da1e2009-10-06 11:31:08 -0700435
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100436 call_rcu(&inode->i_rcu, ceph_i_callback);
Sage Weil355da1e2009-10-06 11:31:08 -0700437}
438
Sage Weil355da1e2009-10-06 11:31:08 -0700439/*
440 * Helpers to fill in size, ctime, mtime, and atime. We have to be
441 * careful because either the client or MDS may have more up to date
442 * info, depending on which capabilities are held, and whether
443 * time_warp_seq or truncate_seq have increased. (Ordinarily, mtime
444 * and size are monotonically increasing, except when utimes() or
445 * truncate() increments the corresponding _seq values.)
446 */
447int ceph_fill_file_size(struct inode *inode, int issued,
448 u32 truncate_seq, u64 truncate_size, u64 size)
449{
450 struct ceph_inode_info *ci = ceph_inode(inode);
451 int queue_trunc = 0;
452
453 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 ||
454 (truncate_seq == ci->i_truncate_seq && size > inode->i_size)) {
455 dout("size %lld -> %llu\n", inode->i_size, size);
456 inode->i_size = size;
457 inode->i_blocks = (size + (1<<9) - 1) >> 9;
458 ci->i_reported_size = size;
459 if (truncate_seq != ci->i_truncate_seq) {
460 dout("truncate_seq %u -> %u\n",
461 ci->i_truncate_seq, truncate_seq);
462 ci->i_truncate_seq = truncate_seq;
Yan, Zhengb0d7c222013-08-12 21:42:15 -0700463
464 /* the MDS should have revoked these caps */
465 WARN_ON_ONCE(issued & (CEPH_CAP_FILE_EXCL |
466 CEPH_CAP_FILE_RD |
467 CEPH_CAP_FILE_WR |
468 CEPH_CAP_FILE_LAZYIO));
Yehuda Sadeh3d497d82010-02-09 11:08:40 -0800469 /*
470 * If we hold relevant caps, or in the case where we're
471 * not the only client referencing this file and we
472 * don't hold those caps, then we need to check whether
473 * the file is either opened or mmaped
474 */
Yan, Zhengb0d7c222013-08-12 21:42:15 -0700475 if ((issued & (CEPH_CAP_FILE_CACHE|
476 CEPH_CAP_FILE_BUFFER)) ||
Yehuda Sadeh3d497d82010-02-09 11:08:40 -0800477 mapping_mapped(inode->i_mapping) ||
478 __ceph_caps_file_wanted(ci)) {
Sage Weil355da1e2009-10-06 11:31:08 -0700479 ci->i_truncate_pending++;
480 queue_trunc = 1;
481 }
482 }
483 }
484 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 &&
485 ci->i_truncate_size != truncate_size) {
486 dout("truncate_size %lld -> %llu\n", ci->i_truncate_size,
487 truncate_size);
488 ci->i_truncate_size = truncate_size;
489 }
Milosz Tanski99ccbd22013-08-21 17:29:54 -0400490
491 if (queue_trunc)
492 ceph_fscache_invalidate(inode);
493
Sage Weil355da1e2009-10-06 11:31:08 -0700494 return queue_trunc;
495}
496
497void ceph_fill_file_time(struct inode *inode, int issued,
498 u64 time_warp_seq, struct timespec *ctime,
499 struct timespec *mtime, struct timespec *atime)
500{
501 struct ceph_inode_info *ci = ceph_inode(inode);
502 int warn = 0;
503
504 if (issued & (CEPH_CAP_FILE_EXCL|
505 CEPH_CAP_FILE_WR|
Sage Weild8672d62010-11-08 09:24:34 -0800506 CEPH_CAP_FILE_BUFFER|
507 CEPH_CAP_AUTH_EXCL|
508 CEPH_CAP_XATTR_EXCL)) {
Sage Weil355da1e2009-10-06 11:31:08 -0700509 if (timespec_compare(ctime, &inode->i_ctime) > 0) {
510 dout("ctime %ld.%09ld -> %ld.%09ld inc w/ cap\n",
511 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
512 ctime->tv_sec, ctime->tv_nsec);
513 inode->i_ctime = *ctime;
514 }
515 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) {
516 /* the MDS did a utimes() */
517 dout("mtime %ld.%09ld -> %ld.%09ld "
518 "tw %d -> %d\n",
519 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
520 mtime->tv_sec, mtime->tv_nsec,
521 ci->i_time_warp_seq, (int)time_warp_seq);
522
523 inode->i_mtime = *mtime;
524 inode->i_atime = *atime;
525 ci->i_time_warp_seq = time_warp_seq;
526 } else if (time_warp_seq == ci->i_time_warp_seq) {
527 /* nobody did utimes(); take the max */
528 if (timespec_compare(mtime, &inode->i_mtime) > 0) {
529 dout("mtime %ld.%09ld -> %ld.%09ld inc\n",
530 inode->i_mtime.tv_sec,
531 inode->i_mtime.tv_nsec,
532 mtime->tv_sec, mtime->tv_nsec);
533 inode->i_mtime = *mtime;
534 }
535 if (timespec_compare(atime, &inode->i_atime) > 0) {
536 dout("atime %ld.%09ld -> %ld.%09ld inc\n",
537 inode->i_atime.tv_sec,
538 inode->i_atime.tv_nsec,
539 atime->tv_sec, atime->tv_nsec);
540 inode->i_atime = *atime;
541 }
542 } else if (issued & CEPH_CAP_FILE_EXCL) {
543 /* we did a utimes(); ignore mds values */
544 } else {
545 warn = 1;
546 }
547 } else {
Sage Weild8672d62010-11-08 09:24:34 -0800548 /* we have no write|excl caps; whatever the MDS says is true */
Sage Weil355da1e2009-10-06 11:31:08 -0700549 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) {
550 inode->i_ctime = *ctime;
551 inode->i_mtime = *mtime;
552 inode->i_atime = *atime;
553 ci->i_time_warp_seq = time_warp_seq;
554 } else {
555 warn = 1;
556 }
557 }
558 if (warn) /* time_warp_seq shouldn't go backwards */
559 dout("%p mds time_warp_seq %llu < %u\n",
560 inode, time_warp_seq, ci->i_time_warp_seq);
561}
562
563/*
564 * Populate an inode based on info from mds. May be called on new or
565 * existing inodes.
566 */
567static int fill_inode(struct inode *inode,
568 struct ceph_mds_reply_info_in *iinfo,
569 struct ceph_mds_reply_dirfrag *dirinfo,
570 struct ceph_mds_session *session,
571 unsigned long ttl_from, int cap_fmode,
572 struct ceph_cap_reservation *caps_reservation)
573{
574 struct ceph_mds_reply_inode *info = iinfo->in;
575 struct ceph_inode_info *ci = ceph_inode(inode);
576 int i;
Sage Weildfabbed62011-07-26 11:30:02 -0700577 int issued = 0, implemented;
Sage Weil355da1e2009-10-06 11:31:08 -0700578 struct timespec mtime, atime, ctime;
579 u32 nsplits;
580 struct ceph_buffer *xattr_blob = NULL;
581 int err = 0;
582 int queue_trunc = 0;
583
584 dout("fill_inode %p ino %llx.%llx v %llu had %llu\n",
585 inode, ceph_vinop(inode), le64_to_cpu(info->version),
586 ci->i_version);
587
588 /*
589 * prealloc xattr data, if it looks like we'll need it. only
590 * if len > 4 (meaning there are actually xattrs; the first 4
591 * bytes are the xattr count).
592 */
593 if (iinfo->xattr_len > 4) {
Sage Weilb6c1d5b2009-12-07 12:17:17 -0800594 xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS);
Sage Weil355da1e2009-10-06 11:31:08 -0700595 if (!xattr_blob)
596 pr_err("fill_inode ENOMEM xattr blob %d bytes\n",
597 iinfo->xattr_len);
598 }
599
Sage Weilbe655592011-11-30 09:47:09 -0800600 spin_lock(&ci->i_ceph_lock);
Sage Weil355da1e2009-10-06 11:31:08 -0700601
602 /*
603 * provided version will be odd if inode value is projected,
Sage Weil8bd59e02010-11-08 09:23:12 -0800604 * even if stable. skip the update if we have newer stable
605 * info (ours>=theirs, e.g. due to racing mds replies), unless
606 * we are getting projected (unstable) info (in which case the
607 * version is odd, and we want ours>theirs).
608 * us them
609 * 2 2 skip
610 * 3 2 skip
611 * 3 3 update
Sage Weil355da1e2009-10-06 11:31:08 -0700612 */
613 if (le64_to_cpu(info->version) > 0 &&
Sage Weil8bd59e02010-11-08 09:23:12 -0800614 (ci->i_version & ~1) >= le64_to_cpu(info->version))
Sage Weil355da1e2009-10-06 11:31:08 -0700615 goto no_change;
Sage Weildfabbed62011-07-26 11:30:02 -0700616
Sage Weil355da1e2009-10-06 11:31:08 -0700617 issued = __ceph_caps_issued(ci, &implemented);
618 issued |= implemented | __ceph_caps_dirty(ci);
619
620 /* update inode */
621 ci->i_version = le64_to_cpu(info->version);
622 inode->i_version++;
623 inode->i_rdev = le32_to_cpu(info->rdev);
624
625 if ((issued & CEPH_CAP_AUTH_EXCL) == 0) {
626 inode->i_mode = le32_to_cpu(info->mode);
Eric W. Biedermanab871b92013-01-31 03:40:12 -0800627 inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(info->uid));
628 inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(info->gid));
Sage Weil355da1e2009-10-06 11:31:08 -0700629 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
Eric W. Biedermanbd2bae62013-01-31 04:05:39 -0800630 from_kuid(&init_user_ns, inode->i_uid),
631 from_kgid(&init_user_ns, inode->i_gid));
Sage Weil355da1e2009-10-06 11:31:08 -0700632 }
633
634 if ((issued & CEPH_CAP_LINK_EXCL) == 0)
Miklos Szeredibfe86842011-10-28 14:13:29 +0200635 set_nlink(inode, le32_to_cpu(info->nlink));
Sage Weil355da1e2009-10-06 11:31:08 -0700636
637 /* be careful with mtime, atime, size */
638 ceph_decode_timespec(&atime, &info->atime);
639 ceph_decode_timespec(&mtime, &info->mtime);
640 ceph_decode_timespec(&ctime, &info->ctime);
641 queue_trunc = ceph_fill_file_size(inode, issued,
642 le32_to_cpu(info->truncate_seq),
643 le64_to_cpu(info->truncate_size),
Sage Weil355da1e2009-10-06 11:31:08 -0700644 le64_to_cpu(info->size));
645 ceph_fill_file_time(inode, issued,
646 le32_to_cpu(info->time_warp_seq),
647 &ctime, &mtime, &atime);
648
Sage Weil912a9b02010-11-07 09:37:25 -0800649 /* only update max_size on auth cap */
650 if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
651 ci->i_max_size != le64_to_cpu(info->max_size)) {
652 dout("max_size %lld -> %llu\n", ci->i_max_size,
653 le64_to_cpu(info->max_size));
654 ci->i_max_size = le64_to_cpu(info->max_size);
655 }
656
Sage Weil355da1e2009-10-06 11:31:08 -0700657 ci->i_layout = info->layout;
658 inode->i_blkbits = fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1;
659
660 /* xattrs */
661 /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
662 if ((issued & CEPH_CAP_XATTR_EXCL) == 0 &&
663 le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) {
664 if (ci->i_xattrs.blob)
665 ceph_buffer_put(ci->i_xattrs.blob);
666 ci->i_xattrs.blob = xattr_blob;
667 if (xattr_blob)
668 memcpy(ci->i_xattrs.blob->vec.iov_base,
669 iinfo->xattr_data, iinfo->xattr_len);
670 ci->i_xattrs.version = le64_to_cpu(info->xattr_version);
Sage Weila6424e42010-04-29 09:28:11 -0700671 xattr_blob = NULL;
Sage Weil355da1e2009-10-06 11:31:08 -0700672 }
673
674 inode->i_mapping->a_ops = &ceph_aops;
675 inode->i_mapping->backing_dev_info =
Cheng Renquan640ef792010-03-26 17:40:33 +0800676 &ceph_sb_to_client(inode->i_sb)->backing_dev_info;
Sage Weil355da1e2009-10-06 11:31:08 -0700677
678 switch (inode->i_mode & S_IFMT) {
679 case S_IFIFO:
680 case S_IFBLK:
681 case S_IFCHR:
682 case S_IFSOCK:
683 init_special_inode(inode, inode->i_mode, inode->i_rdev);
684 inode->i_op = &ceph_file_iops;
685 break;
686 case S_IFREG:
687 inode->i_op = &ceph_file_iops;
688 inode->i_fop = &ceph_file_fops;
689 break;
690 case S_IFLNK:
691 inode->i_op = &ceph_symlink_iops;
692 if (!ci->i_symlink) {
Xi Wang810339ec2012-02-03 09:55:36 -0500693 u32 symlen = iinfo->symlink_len;
Sage Weil355da1e2009-10-06 11:31:08 -0700694 char *sym;
695
Sage Weilbe655592011-11-30 09:47:09 -0800696 spin_unlock(&ci->i_ceph_lock);
Sage Weil355da1e2009-10-06 11:31:08 -0700697
Xi Wang810339ec2012-02-03 09:55:36 -0500698 err = -EINVAL;
699 if (WARN_ON(symlen != inode->i_size))
700 goto out;
701
Sage Weil355da1e2009-10-06 11:31:08 -0700702 err = -ENOMEM;
Xi Wang810339ec2012-02-03 09:55:36 -0500703 sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS);
Sage Weil355da1e2009-10-06 11:31:08 -0700704 if (!sym)
705 goto out;
Sage Weil355da1e2009-10-06 11:31:08 -0700706
Sage Weilbe655592011-11-30 09:47:09 -0800707 spin_lock(&ci->i_ceph_lock);
Sage Weil355da1e2009-10-06 11:31:08 -0700708 if (!ci->i_symlink)
709 ci->i_symlink = sym;
710 else
711 kfree(sym); /* lost a race */
712 }
713 break;
714 case S_IFDIR:
715 inode->i_op = &ceph_dir_iops;
716 inode->i_fop = &ceph_dir_fops;
717
Sage Weil14303d22010-12-14 17:37:52 -0800718 ci->i_dir_layout = iinfo->dir_layout;
719
Sage Weil355da1e2009-10-06 11:31:08 -0700720 ci->i_files = le64_to_cpu(info->files);
721 ci->i_subdirs = le64_to_cpu(info->subdirs);
722 ci->i_rbytes = le64_to_cpu(info->rbytes);
723 ci->i_rfiles = le64_to_cpu(info->rfiles);
724 ci->i_rsubdirs = le64_to_cpu(info->rsubdirs);
725 ceph_decode_timespec(&ci->i_rctime, &info->rctime);
Sage Weil355da1e2009-10-06 11:31:08 -0700726 break;
727 default:
728 pr_err("fill_inode %llx.%llx BAD mode 0%o\n",
729 ceph_vinop(inode), inode->i_mode);
730 }
731
Yan, Zhenga8673d62013-02-18 16:38:14 +0800732 /* set dir completion flag? */
733 if (S_ISDIR(inode->i_mode) &&
734 ci->i_files == 0 && ci->i_subdirs == 0 &&
735 ceph_snap(inode) == CEPH_NOSNAP &&
736 (le32_to_cpu(info->cap.caps) & CEPH_CAP_FILE_SHARED) &&
737 (issued & CEPH_CAP_FILE_EXCL) == 0 &&
Yan, Zheng2f276c52013-03-13 19:44:32 +0800738 !__ceph_dir_is_complete(ci)) {
Yan, Zhenga8673d62013-02-18 16:38:14 +0800739 dout(" marking %p complete (empty)\n", inode);
Yan, Zheng2f276c52013-03-13 19:44:32 +0800740 __ceph_dir_set_complete(ci, atomic_read(&ci->i_release_count));
Yan, Zhenga8673d62013-02-18 16:38:14 +0800741 ci->i_max_offset = 2;
742 }
Sage Weil355da1e2009-10-06 11:31:08 -0700743no_change:
Sage Weilbe655592011-11-30 09:47:09 -0800744 spin_unlock(&ci->i_ceph_lock);
Sage Weil355da1e2009-10-06 11:31:08 -0700745
746 /* queue truncate if we saw i_size decrease */
747 if (queue_trunc)
Sage Weil3c6f6b72010-02-09 15:24:44 -0800748 ceph_queue_vmtruncate(inode);
Sage Weil355da1e2009-10-06 11:31:08 -0700749
750 /* populate frag tree */
751 /* FIXME: move me up, if/when version reflects fragtree changes */
752 nsplits = le32_to_cpu(info->fragtree.nsplits);
753 mutex_lock(&ci->i_fragtree_mutex);
754 for (i = 0; i < nsplits; i++) {
755 u32 id = le32_to_cpu(info->fragtree.splits[i].frag);
756 struct ceph_inode_frag *frag = __get_or_create_frag(ci, id);
757
758 if (IS_ERR(frag))
759 continue;
760 frag->split_by = le32_to_cpu(info->fragtree.splits[i].by);
761 dout(" frag %x split by %d\n", frag->frag, frag->split_by);
762 }
763 mutex_unlock(&ci->i_fragtree_mutex);
764
765 /* were we issued a capability? */
766 if (info->cap.caps) {
767 if (ceph_snap(inode) == CEPH_NOSNAP) {
768 ceph_add_cap(inode, session,
769 le64_to_cpu(info->cap.cap_id),
770 cap_fmode,
771 le32_to_cpu(info->cap.caps),
772 le32_to_cpu(info->cap.wanted),
773 le32_to_cpu(info->cap.seq),
774 le32_to_cpu(info->cap.mseq),
775 le64_to_cpu(info->cap.realm),
776 info->cap.flags,
777 caps_reservation);
778 } else {
Sage Weilbe655592011-11-30 09:47:09 -0800779 spin_lock(&ci->i_ceph_lock);
Sage Weil355da1e2009-10-06 11:31:08 -0700780 dout(" %p got snap_caps %s\n", inode,
781 ceph_cap_string(le32_to_cpu(info->cap.caps)));
782 ci->i_snap_caps |= le32_to_cpu(info->cap.caps);
783 if (cap_fmode >= 0)
784 __ceph_get_fmode(ci, cap_fmode);
Sage Weilbe655592011-11-30 09:47:09 -0800785 spin_unlock(&ci->i_ceph_lock);
Sage Weil355da1e2009-10-06 11:31:08 -0700786 }
Sage Weil04d000e2010-05-07 11:26:34 -0700787 } else if (cap_fmode >= 0) {
788 pr_warning("mds issued no caps on %llx.%llx\n",
789 ceph_vinop(inode));
790 __ceph_get_fmode(ci, cap_fmode);
Sage Weil355da1e2009-10-06 11:31:08 -0700791 }
792
793 /* update delegation info? */
794 if (dirinfo)
795 ceph_fill_dirfrag(inode, dirinfo);
796
797 err = 0;
798
799out:
Sage Weilb6c1d5b2009-12-07 12:17:17 -0800800 if (xattr_blob)
801 ceph_buffer_put(xattr_blob);
Sage Weil355da1e2009-10-06 11:31:08 -0700802 return err;
803}
804
805/*
806 * caller should hold session s_mutex.
807 */
808static void update_dentry_lease(struct dentry *dentry,
809 struct ceph_mds_reply_lease *lease,
810 struct ceph_mds_session *session,
811 unsigned long from_time)
812{
813 struct ceph_dentry_info *di = ceph_dentry(dentry);
814 long unsigned duration = le32_to_cpu(lease->duration_ms);
815 long unsigned ttl = from_time + (duration * HZ) / 1000;
816 long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000;
817 struct inode *dir;
818
819 /* only track leases on regular dentries */
820 if (dentry->d_op != &ceph_dentry_ops)
821 return;
822
823 spin_lock(&dentry->d_lock);
Sage Weil2f90b852011-07-26 11:28:25 -0700824 dout("update_dentry_lease %p duration %lu ms ttl %lu\n",
825 dentry, duration, ttl);
Sage Weil355da1e2009-10-06 11:31:08 -0700826
827 /* make lease_rdcache_gen match directory */
828 dir = dentry->d_parent->d_inode;
829 di->lease_shared_gen = ceph_inode(dir)->i_shared_gen;
830
Sage Weil2f90b852011-07-26 11:28:25 -0700831 if (duration == 0)
Sage Weil355da1e2009-10-06 11:31:08 -0700832 goto out_unlock;
833
834 if (di->lease_gen == session->s_cap_gen &&
835 time_before(ttl, dentry->d_time))
836 goto out_unlock; /* we already have a newer lease. */
837
838 if (di->lease_session && di->lease_session != session)
839 goto out_unlock;
840
841 ceph_dentry_lru_touch(dentry);
842
843 if (!di->lease_session)
844 di->lease_session = ceph_get_mds_session(session);
845 di->lease_gen = session->s_cap_gen;
846 di->lease_seq = le32_to_cpu(lease->seq);
847 di->lease_renew_after = half_ttl;
848 di->lease_renew_from = 0;
849 dentry->d_time = ttl;
850out_unlock:
851 spin_unlock(&dentry->d_lock);
852 return;
853}
854
855/*
Yehuda Sadeh4baa75e2010-01-07 15:36:32 -0800856 * Set dentry's directory position based on the current dir's max, and
857 * order it in d_subdirs, so that dcache_readdir behaves.
Sage Weil4f177262011-07-26 11:31:08 -0700858 *
859 * Always called under directory's i_mutex.
Yehuda Sadeh4baa75e2010-01-07 15:36:32 -0800860 */
861static void ceph_set_dentry_offset(struct dentry *dn)
862{
863 struct dentry *dir = dn->d_parent;
Sage Weil4f177262011-07-26 11:31:08 -0700864 struct inode *inode = dir->d_inode;
Yehuda Sadehb8cd9522011-12-13 09:56:30 -0800865 struct ceph_inode_info *ci;
Yehuda Sadeh4baa75e2010-01-07 15:36:32 -0800866 struct ceph_dentry_info *di;
867
868 BUG_ON(!inode);
869
Yehuda Sadehb8cd9522011-12-13 09:56:30 -0800870 ci = ceph_inode(inode);
Yehuda Sadeh4baa75e2010-01-07 15:36:32 -0800871 di = ceph_dentry(dn);
872
Sage Weilbe655592011-11-30 09:47:09 -0800873 spin_lock(&ci->i_ceph_lock);
Yan, Zheng2f276c52013-03-13 19:44:32 +0800874 if (!__ceph_dir_is_complete(ci)) {
Sage Weilbe655592011-11-30 09:47:09 -0800875 spin_unlock(&ci->i_ceph_lock);
Sage Weile8a74982010-04-15 14:08:49 -0700876 return;
877 }
Yehuda Sadeh4baa75e2010-01-07 15:36:32 -0800878 di->offset = ceph_inode(inode)->i_max_offset++;
Sage Weilbe655592011-11-30 09:47:09 -0800879 spin_unlock(&ci->i_ceph_lock);
Yehuda Sadeh4baa75e2010-01-07 15:36:32 -0800880
Nick Piggin2fd6b7f2011-01-07 17:49:34 +1100881 spin_lock(&dir->d_lock);
882 spin_lock_nested(&dn->d_lock, DENTRY_D_LOCK_NESTED);
Henry C Chang13a42142010-06-01 11:31:08 -0700883 list_move(&dn->d_u.d_child, &dir->d_subdirs);
Yehuda Sadeh4baa75e2010-01-07 15:36:32 -0800884 dout("set_dentry_offset %p %lld (%p %p)\n", dn, di->offset,
885 dn->d_u.d_child.prev, dn->d_u.d_child.next);
886 spin_unlock(&dn->d_lock);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +1100887 spin_unlock(&dir->d_lock);
Yehuda Sadeh4baa75e2010-01-07 15:36:32 -0800888}
889
890/*
Sage Weil1cd39352010-05-03 22:08:02 -0700891 * splice a dentry to an inode.
892 * caller must hold directory i_mutex for this to be safe.
893 *
894 * we will only rehash the resulting dentry if @prehash is
895 * true; @prehash will be set to false (for the benefit of
896 * the caller) if we fail.
897 */
898static struct dentry *splice_dentry(struct dentry *dn, struct inode *in,
Sage Weil467c5252010-09-13 11:39:20 -0700899 bool *prehash, bool set_offset)
Sage Weil1cd39352010-05-03 22:08:02 -0700900{
901 struct dentry *realdn;
902
903 BUG_ON(dn->d_inode);
904
905 /* dn must be unhashed */
906 if (!d_unhashed(dn))
907 d_drop(dn);
908 realdn = d_materialise_unique(dn, in);
909 if (IS_ERR(realdn)) {
Sage Weild69ed052010-06-21 10:38:14 -0700910 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
911 PTR_ERR(realdn), dn, in, ceph_vinop(in));
Sage Weil1cd39352010-05-03 22:08:02 -0700912 if (prehash)
913 *prehash = false; /* don't rehash on error */
914 dn = realdn; /* note realdn contains the error */
915 goto out;
916 } else if (realdn) {
917 dout("dn %p (%d) spliced with %p (%d) "
918 "inode %p ino %llx.%llx\n",
Al Viro84d08fa2013-07-05 18:59:33 +0400919 dn, d_count(dn),
920 realdn, d_count(realdn),
Sage Weil1cd39352010-05-03 22:08:02 -0700921 realdn->d_inode, ceph_vinop(realdn->d_inode));
922 dput(dn);
923 dn = realdn;
924 } else {
925 BUG_ON(!ceph_dentry(dn));
926 dout("dn %p attached to %p ino %llx.%llx\n",
927 dn, dn->d_inode, ceph_vinop(dn->d_inode));
928 }
929 if ((!prehash || *prehash) && d_unhashed(dn))
930 d_rehash(dn);
Sage Weil467c5252010-09-13 11:39:20 -0700931 if (set_offset)
932 ceph_set_dentry_offset(dn);
Sage Weil1cd39352010-05-03 22:08:02 -0700933out:
934 return dn;
935}
936
937/*
Sage Weil355da1e2009-10-06 11:31:08 -0700938 * Incorporate results into the local cache. This is either just
939 * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
940 * after a lookup).
941 *
942 * A reply may contain
943 * a directory inode along with a dentry.
944 * and/or a target inode
945 *
946 * Called with snap_rwsem (read).
947 */
948int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req,
949 struct ceph_mds_session *session)
950{
951 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
952 struct inode *in = NULL;
953 struct ceph_mds_reply_inode *ininfo;
954 struct ceph_vino vino;
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700955 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
Sage Weil355da1e2009-10-06 11:31:08 -0700956 int i = 0;
957 int err = 0;
958
959 dout("fill_trace %p is_dentry %d is_target %d\n", req,
960 rinfo->head->is_dentry, rinfo->head->is_target);
961
962#if 0
963 /*
964 * Debugging hook:
965 *
966 * If we resend completed ops to a recovering mds, we get no
967 * trace. Since that is very rare, pretend this is the case
968 * to ensure the 'no trace' handlers in the callers behave.
969 *
970 * Fill in inodes unconditionally to avoid breaking cap
971 * invariants.
972 */
973 if (rinfo->head->op & CEPH_MDS_OP_WRITE) {
974 pr_info("fill_trace faking empty trace on %lld %s\n",
975 req->r_tid, ceph_mds_op_name(rinfo->head->op));
976 if (rinfo->head->is_dentry) {
977 rinfo->head->is_dentry = 0;
978 err = fill_inode(req->r_locked_dir,
979 &rinfo->diri, rinfo->dirfrag,
980 session, req->r_request_started, -1);
981 }
982 if (rinfo->head->is_target) {
983 rinfo->head->is_target = 0;
984 ininfo = rinfo->targeti.in;
985 vino.ino = le64_to_cpu(ininfo->ino);
986 vino.snap = le64_to_cpu(ininfo->snapid);
987 in = ceph_get_inode(sb, vino);
988 err = fill_inode(in, &rinfo->targeti, NULL,
989 session, req->r_request_started,
990 req->r_fmode);
991 iput(in);
992 }
993 }
994#endif
995
996 if (!rinfo->head->is_target && !rinfo->head->is_dentry) {
997 dout("fill_trace reply is empty!\n");
Sage Weil167c9e32010-05-14 10:02:57 -0700998 if (rinfo->head->result == 0 && req->r_locked_dir)
999 ceph_invalidate_dir_request(req);
Sage Weil355da1e2009-10-06 11:31:08 -07001000 return 0;
1001 }
1002
1003 if (rinfo->head->is_dentry) {
Sage Weil5b1daec2010-01-25 11:33:08 -08001004 struct inode *dir = req->r_locked_dir;
1005
Sage Weil6c5e50f2012-08-21 15:55:25 -07001006 if (dir) {
1007 err = fill_inode(dir, &rinfo->diri, rinfo->dirfrag,
1008 session, req->r_request_started, -1,
1009 &req->r_caps_reservation);
1010 if (err < 0)
1011 return err;
1012 } else {
1013 WARN_ON_ONCE(1);
1014 }
Sage Weil5b1daec2010-01-25 11:33:08 -08001015 }
1016
Sage Weil9358c6d2010-03-30 13:54:41 -07001017 /*
1018 * ignore null lease/binding on snapdir ENOENT, or else we
1019 * will have trouble splicing in the virtual snapdir later
1020 */
1021 if (rinfo->head->is_dentry && !req->r_aborted &&
Sage Weil6c5e50f2012-08-21 15:55:25 -07001022 req->r_locked_dir &&
Sage Weil9358c6d2010-03-30 13:54:41 -07001023 (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name,
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001024 fsc->mount_options->snapdir_name,
Sage Weil9358c6d2010-03-30 13:54:41 -07001025 req->r_dentry->d_name.len))) {
Sage Weil355da1e2009-10-06 11:31:08 -07001026 /*
1027 * lookup link rename : null -> possibly existing inode
1028 * mknod symlink mkdir : null -> new inode
1029 * unlink : linked -> null
1030 */
1031 struct inode *dir = req->r_locked_dir;
1032 struct dentry *dn = req->r_dentry;
1033 bool have_dir_cap, have_lease;
1034
1035 BUG_ON(!dn);
1036 BUG_ON(!dir);
1037 BUG_ON(dn->d_parent->d_inode != dir);
1038 BUG_ON(ceph_ino(dir) !=
1039 le64_to_cpu(rinfo->diri.in->ino));
1040 BUG_ON(ceph_snap(dir) !=
1041 le64_to_cpu(rinfo->diri.in->snapid));
1042
Sage Weil355da1e2009-10-06 11:31:08 -07001043 /* do we have a lease on the whole dir? */
1044 have_dir_cap =
1045 (le32_to_cpu(rinfo->diri.in->cap.caps) &
1046 CEPH_CAP_FILE_SHARED);
1047
1048 /* do we have a dn lease? */
1049 have_lease = have_dir_cap ||
Sage Weil2f90b852011-07-26 11:28:25 -07001050 le32_to_cpu(rinfo->dlease->duration_ms);
Sage Weil355da1e2009-10-06 11:31:08 -07001051 if (!have_lease)
1052 dout("fill_trace no dentry lease or dir cap\n");
1053
1054 /* rename? */
1055 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) {
1056 dout(" src %p '%.*s' dst %p '%.*s'\n",
1057 req->r_old_dentry,
1058 req->r_old_dentry->d_name.len,
1059 req->r_old_dentry->d_name.name,
1060 dn, dn->d_name.len, dn->d_name.name);
1061 dout("fill_trace doing d_move %p -> %p\n",
1062 req->r_old_dentry, dn);
Sage Weilc10f5e12010-04-16 12:56:11 -07001063
Sage Weil355da1e2009-10-06 11:31:08 -07001064 d_move(req->r_old_dentry, dn);
1065 dout(" src %p '%.*s' dst %p '%.*s'\n",
1066 req->r_old_dentry,
1067 req->r_old_dentry->d_name.len,
1068 req->r_old_dentry->d_name.name,
1069 dn, dn->d_name.len, dn->d_name.name);
Sage Weil81a6cf22010-05-14 09:35:38 -07001070
Sage Weilc4a29f22009-12-21 11:42:18 -08001071 /* ensure target dentry is invalidated, despite
1072 rehashing bug in vfs_rename_dir */
Sage Weil81a6cf22010-05-14 09:35:38 -07001073 ceph_invalidate_dentry_lease(dn);
1074
Sage Weil09adc802011-02-04 21:38:47 -08001075 /*
1076 * d_move() puts the renamed dentry at the end of
1077 * d_subdirs. We need to assign it an appropriate
Yan, Zheng2f276c52013-03-13 19:44:32 +08001078 * directory offset so we can behave when dir is
1079 * complete.
Sage Weil09adc802011-02-04 21:38:47 -08001080 */
1081 ceph_set_dentry_offset(req->r_old_dentry);
Milosz Tanski99ccbd22013-08-21 17:29:54 -04001082 dout("dn %p gets new offset %lld\n", req->r_old_dentry,
Sage Weil1cd39352010-05-03 22:08:02 -07001083 ceph_dentry(req->r_old_dentry)->offset);
Sage Weil81a6cf22010-05-14 09:35:38 -07001084
Sage Weil355da1e2009-10-06 11:31:08 -07001085 dn = req->r_old_dentry; /* use old_dentry */
1086 in = dn->d_inode;
1087 }
1088
1089 /* null dentry? */
1090 if (!rinfo->head->is_target) {
1091 dout("fill_trace null dentry\n");
1092 if (dn->d_inode) {
1093 dout("d_delete %p\n", dn);
1094 d_delete(dn);
1095 } else {
1096 dout("d_instantiate %p NULL\n", dn);
1097 d_instantiate(dn, NULL);
1098 if (have_lease && d_unhashed(dn))
1099 d_rehash(dn);
1100 update_dentry_lease(dn, rinfo->dlease,
1101 session,
1102 req->r_request_started);
1103 }
1104 goto done;
1105 }
1106
1107 /* attach proper inode */
1108 ininfo = rinfo->targeti.in;
1109 vino.ino = le64_to_cpu(ininfo->ino);
1110 vino.snap = le64_to_cpu(ininfo->snapid);
Sage Weild8b16b32010-11-06 12:41:16 -07001111 in = dn->d_inode;
1112 if (!in) {
Sage Weil355da1e2009-10-06 11:31:08 -07001113 in = ceph_get_inode(sb, vino);
1114 if (IS_ERR(in)) {
1115 pr_err("fill_trace bad get_inode "
1116 "%llx.%llx\n", vino.ino, vino.snap);
1117 err = PTR_ERR(in);
Al Viro2744c172012-09-26 21:41:05 -04001118 d_drop(dn);
Sage Weil355da1e2009-10-06 11:31:08 -07001119 goto done;
1120 }
Sage Weil467c5252010-09-13 11:39:20 -07001121 dn = splice_dentry(dn, in, &have_lease, true);
Sage Weil355da1e2009-10-06 11:31:08 -07001122 if (IS_ERR(dn)) {
1123 err = PTR_ERR(dn);
1124 goto done;
1125 }
1126 req->r_dentry = dn; /* may have spliced */
Sage Weil70b666c2011-05-27 09:24:26 -07001127 ihold(in);
Sage Weil355da1e2009-10-06 11:31:08 -07001128 } else if (ceph_ino(in) == vino.ino &&
1129 ceph_snap(in) == vino.snap) {
Sage Weil70b666c2011-05-27 09:24:26 -07001130 ihold(in);
Sage Weil355da1e2009-10-06 11:31:08 -07001131 } else {
1132 dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1133 dn, in, ceph_ino(in), ceph_snap(in),
1134 vino.ino, vino.snap);
1135 have_lease = false;
1136 in = NULL;
1137 }
1138
1139 if (have_lease)
1140 update_dentry_lease(dn, rinfo->dlease, session,
1141 req->r_request_started);
1142 dout(" final dn %p\n", dn);
1143 i++;
Sage Weil79f9f992013-01-29 02:55:31 -05001144 } else if ((req->r_op == CEPH_MDS_OP_LOOKUPSNAP ||
1145 req->r_op == CEPH_MDS_OP_MKSNAP) && !req->r_aborted) {
Sage Weil355da1e2009-10-06 11:31:08 -07001146 struct dentry *dn = req->r_dentry;
1147
1148 /* fill out a snapdir LOOKUPSNAP dentry */
1149 BUG_ON(!dn);
1150 BUG_ON(!req->r_locked_dir);
1151 BUG_ON(ceph_snap(req->r_locked_dir) != CEPH_SNAPDIR);
1152 ininfo = rinfo->targeti.in;
1153 vino.ino = le64_to_cpu(ininfo->ino);
1154 vino.snap = le64_to_cpu(ininfo->snapid);
1155 in = ceph_get_inode(sb, vino);
1156 if (IS_ERR(in)) {
1157 pr_err("fill_inode get_inode badness %llx.%llx\n",
1158 vino.ino, vino.snap);
1159 err = PTR_ERR(in);
1160 d_delete(dn);
1161 goto done;
1162 }
1163 dout(" linking snapped dir %p to dn %p\n", in, dn);
Sage Weil467c5252010-09-13 11:39:20 -07001164 dn = splice_dentry(dn, in, NULL, true);
Sage Weil355da1e2009-10-06 11:31:08 -07001165 if (IS_ERR(dn)) {
1166 err = PTR_ERR(dn);
1167 goto done;
1168 }
1169 req->r_dentry = dn; /* may have spliced */
Sage Weil70b666c2011-05-27 09:24:26 -07001170 ihold(in);
Sage Weil355da1e2009-10-06 11:31:08 -07001171 rinfo->head->is_dentry = 1; /* fool notrace handlers */
1172 }
1173
1174 if (rinfo->head->is_target) {
1175 vino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1176 vino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1177
1178 if (in == NULL || ceph_ino(in) != vino.ino ||
1179 ceph_snap(in) != vino.snap) {
1180 in = ceph_get_inode(sb, vino);
1181 if (IS_ERR(in)) {
1182 err = PTR_ERR(in);
1183 goto done;
1184 }
1185 }
1186 req->r_target_inode = in;
1187
1188 err = fill_inode(in,
1189 &rinfo->targeti, NULL,
1190 session, req->r_request_started,
1191 (le32_to_cpu(rinfo->head->result) == 0) ?
1192 req->r_fmode : -1,
1193 &req->r_caps_reservation);
1194 if (err < 0) {
1195 pr_err("fill_inode badness %p %llx.%llx\n",
1196 in, ceph_vinop(in));
1197 goto done;
1198 }
1199 }
1200
1201done:
1202 dout("fill_trace done err=%d\n", err);
1203 return err;
1204}
1205
1206/*
1207 * Prepopulate our cache with readdir results, leases, etc.
1208 */
Sage Weil79f9f992013-01-29 02:55:31 -05001209static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req,
1210 struct ceph_mds_session *session)
1211{
1212 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1213 int i, err = 0;
1214
1215 for (i = 0; i < rinfo->dir_nr; i++) {
1216 struct ceph_vino vino;
1217 struct inode *in;
1218 int rc;
1219
1220 vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino);
1221 vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid);
1222
1223 in = ceph_get_inode(req->r_dentry->d_sb, vino);
1224 if (IS_ERR(in)) {
1225 err = PTR_ERR(in);
1226 dout("new_inode badness got %d\n", err);
1227 continue;
1228 }
1229 rc = fill_inode(in, &rinfo->dir_in[i], NULL, session,
1230 req->r_request_started, -1,
1231 &req->r_caps_reservation);
1232 if (rc < 0) {
1233 pr_err("fill_inode badness on %p got %d\n", in, rc);
1234 err = rc;
1235 continue;
1236 }
1237 }
1238
1239 return err;
1240}
1241
Sage Weil355da1e2009-10-06 11:31:08 -07001242int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1243 struct ceph_mds_session *session)
1244{
1245 struct dentry *parent = req->r_dentry;
1246 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1247 struct qstr dname;
1248 struct dentry *dn;
1249 struct inode *in;
1250 int err = 0, i;
1251 struct inode *snapdir = NULL;
1252 struct ceph_mds_request_head *rhead = req->r_request->front.iov_base;
1253 u64 frag = le32_to_cpu(rhead->args.readdir.frag);
1254 struct ceph_dentry_info *di;
1255
Sage Weil79f9f992013-01-29 02:55:31 -05001256 if (req->r_aborted)
1257 return readdir_prepopulate_inodes_only(req, session);
1258
Sage Weil355da1e2009-10-06 11:31:08 -07001259 if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
1260 snapdir = ceph_get_snapdir(parent->d_inode);
1261 parent = d_find_alias(snapdir);
1262 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1263 rinfo->dir_nr, parent);
1264 } else {
1265 dout("readdir_prepopulate %d items under dn %p\n",
1266 rinfo->dir_nr, parent);
1267 if (rinfo->dir_dir)
1268 ceph_fill_dirfrag(parent->d_inode, rinfo->dir_dir);
1269 }
1270
1271 for (i = 0; i < rinfo->dir_nr; i++) {
1272 struct ceph_vino vino;
1273
1274 dname.name = rinfo->dir_dname[i];
1275 dname.len = rinfo->dir_dname_len[i];
1276 dname.hash = full_name_hash(dname.name, dname.len);
1277
1278 vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino);
1279 vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid);
1280
1281retry_lookup:
1282 dn = d_lookup(parent, &dname);
1283 dout("d_lookup on parent=%p name=%.*s got %p\n",
1284 parent, dname.len, dname.name, dn);
1285
1286 if (!dn) {
1287 dn = d_alloc(parent, &dname);
1288 dout("d_alloc %p '%.*s' = %p\n", parent,
1289 dname.len, dname.name, dn);
1290 if (dn == NULL) {
1291 dout("d_alloc badness\n");
1292 err = -ENOMEM;
1293 goto out;
1294 }
1295 err = ceph_init_dentry(dn);
Sage Weil8c696732010-07-22 14:11:56 -07001296 if (err < 0) {
1297 dput(dn);
Sage Weil355da1e2009-10-06 11:31:08 -07001298 goto out;
Sage Weil8c696732010-07-22 14:11:56 -07001299 }
Sage Weil355da1e2009-10-06 11:31:08 -07001300 } else if (dn->d_inode &&
1301 (ceph_ino(dn->d_inode) != vino.ino ||
1302 ceph_snap(dn->d_inode) != vino.snap)) {
1303 dout(" dn %p points to wrong inode %p\n",
1304 dn, dn->d_inode);
1305 d_delete(dn);
1306 dput(dn);
1307 goto retry_lookup;
1308 } else {
1309 /* reorder parent's d_subdirs */
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001310 spin_lock(&parent->d_lock);
1311 spin_lock_nested(&dn->d_lock, DENTRY_D_LOCK_NESTED);
Sage Weil355da1e2009-10-06 11:31:08 -07001312 list_move(&dn->d_u.d_child, &parent->d_subdirs);
1313 spin_unlock(&dn->d_lock);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001314 spin_unlock(&parent->d_lock);
Sage Weil355da1e2009-10-06 11:31:08 -07001315 }
1316
1317 di = dn->d_fsdata;
1318 di->offset = ceph_make_fpos(frag, i + req->r_readdir_offset);
1319
1320 /* inode */
1321 if (dn->d_inode) {
1322 in = dn->d_inode;
1323 } else {
1324 in = ceph_get_inode(parent->d_sb, vino);
Dan Carpenterac1f12e2010-08-25 09:11:35 +02001325 if (IS_ERR(in)) {
Sage Weil355da1e2009-10-06 11:31:08 -07001326 dout("new_inode badness\n");
Al Viro2744c172012-09-26 21:41:05 -04001327 d_drop(dn);
Sage Weil355da1e2009-10-06 11:31:08 -07001328 dput(dn);
Dan Carpenterac1f12e2010-08-25 09:11:35 +02001329 err = PTR_ERR(in);
Sage Weil355da1e2009-10-06 11:31:08 -07001330 goto out;
1331 }
Sage Weil467c5252010-09-13 11:39:20 -07001332 dn = splice_dentry(dn, in, NULL, false);
Sage Weild69ed052010-06-21 10:38:14 -07001333 if (IS_ERR(dn))
1334 dn = NULL;
Sage Weil355da1e2009-10-06 11:31:08 -07001335 }
1336
1337 if (fill_inode(in, &rinfo->dir_in[i], NULL, session,
1338 req->r_request_started, -1,
1339 &req->r_caps_reservation) < 0) {
1340 pr_err("fill_inode badness on %p\n", in);
Sage Weild69ed052010-06-21 10:38:14 -07001341 goto next_item;
Sage Weil355da1e2009-10-06 11:31:08 -07001342 }
Sage Weild69ed052010-06-21 10:38:14 -07001343 if (dn)
1344 update_dentry_lease(dn, rinfo->dir_dlease[i],
1345 req->r_session,
1346 req->r_request_started);
1347next_item:
1348 if (dn)
1349 dput(dn);
Sage Weil355da1e2009-10-06 11:31:08 -07001350 }
1351 req->r_did_prepopulate = true;
1352
1353out:
1354 if (snapdir) {
1355 iput(snapdir);
1356 dput(parent);
1357 }
1358 dout("readdir_prepopulate done\n");
1359 return err;
1360}
1361
1362int ceph_inode_set_size(struct inode *inode, loff_t size)
1363{
1364 struct ceph_inode_info *ci = ceph_inode(inode);
1365 int ret = 0;
1366
Sage Weilbe655592011-11-30 09:47:09 -08001367 spin_lock(&ci->i_ceph_lock);
Sage Weil355da1e2009-10-06 11:31:08 -07001368 dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size);
1369 inode->i_size = size;
1370 inode->i_blocks = (size + (1 << 9) - 1) >> 9;
1371
1372 /* tell the MDS if we are approaching max_size */
1373 if ((size << 1) >= ci->i_max_size &&
1374 (ci->i_reported_size << 1) < ci->i_max_size)
1375 ret = 1;
1376
Sage Weilbe655592011-11-30 09:47:09 -08001377 spin_unlock(&ci->i_ceph_lock);
Sage Weil355da1e2009-10-06 11:31:08 -07001378 return ret;
1379}
1380
1381/*
1382 * Write back inode data in a worker thread. (This can't be done
1383 * in the message handler context.)
1384 */
Sage Weil3c6f6b72010-02-09 15:24:44 -08001385void ceph_queue_writeback(struct inode *inode)
1386{
Sage Weil15a20152011-11-05 22:06:31 -07001387 ihold(inode);
Sage Weil3c6f6b72010-02-09 15:24:44 -08001388 if (queue_work(ceph_inode_to_client(inode)->wb_wq,
1389 &ceph_inode(inode)->i_wb_work)) {
Sage Weil2c27c9a2010-02-17 15:45:51 -08001390 dout("ceph_queue_writeback %p\n", inode);
Sage Weil3c6f6b72010-02-09 15:24:44 -08001391 } else {
Sage Weil2c27c9a2010-02-17 15:45:51 -08001392 dout("ceph_queue_writeback %p failed\n", inode);
Sage Weil15a20152011-11-05 22:06:31 -07001393 iput(inode);
Sage Weil3c6f6b72010-02-09 15:24:44 -08001394 }
1395}
1396
1397static void ceph_writeback_work(struct work_struct *work)
Sage Weil355da1e2009-10-06 11:31:08 -07001398{
1399 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1400 i_wb_work);
1401 struct inode *inode = &ci->vfs_inode;
1402
1403 dout("writeback %p\n", inode);
1404 filemap_fdatawrite(&inode->i_data);
1405 iput(inode);
1406}
1407
1408/*
Sage Weil3c6f6b72010-02-09 15:24:44 -08001409 * queue an async invalidation
1410 */
1411void ceph_queue_invalidate(struct inode *inode)
1412{
Sage Weil15a20152011-11-05 22:06:31 -07001413 ihold(inode);
Sage Weil3c6f6b72010-02-09 15:24:44 -08001414 if (queue_work(ceph_inode_to_client(inode)->pg_inv_wq,
1415 &ceph_inode(inode)->i_pg_inv_work)) {
1416 dout("ceph_queue_invalidate %p\n", inode);
Sage Weil3c6f6b72010-02-09 15:24:44 -08001417 } else {
1418 dout("ceph_queue_invalidate %p failed\n", inode);
Sage Weil15a20152011-11-05 22:06:31 -07001419 iput(inode);
Sage Weil3c6f6b72010-02-09 15:24:44 -08001420 }
1421}
1422
1423/*
Sage Weil355da1e2009-10-06 11:31:08 -07001424 * Invalidate inode pages in a worker thread. (This can't be done
1425 * in the message handler context.)
1426 */
Sage Weil3c6f6b72010-02-09 15:24:44 -08001427static void ceph_invalidate_work(struct work_struct *work)
Sage Weil355da1e2009-10-06 11:31:08 -07001428{
1429 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1430 i_pg_inv_work);
1431 struct inode *inode = &ci->vfs_inode;
1432 u32 orig_gen;
1433 int check = 0;
1434
Yan, Zhengb0d7c222013-08-12 21:42:15 -07001435 mutex_lock(&ci->i_truncate_mutex);
Sage Weilbe655592011-11-30 09:47:09 -08001436 spin_lock(&ci->i_ceph_lock);
Sage Weil355da1e2009-10-06 11:31:08 -07001437 dout("invalidate_pages %p gen %d revoking %d\n", inode,
1438 ci->i_rdcache_gen, ci->i_rdcache_revoking);
Sage Weilcd045cb2010-11-04 11:05:05 -07001439 if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
Sage Weil355da1e2009-10-06 11:31:08 -07001440 /* nevermind! */
Sage Weilbe655592011-11-30 09:47:09 -08001441 spin_unlock(&ci->i_ceph_lock);
Yan, Zhengb0d7c222013-08-12 21:42:15 -07001442 mutex_unlock(&ci->i_truncate_mutex);
Sage Weil355da1e2009-10-06 11:31:08 -07001443 goto out;
1444 }
1445 orig_gen = ci->i_rdcache_gen;
Sage Weilbe655592011-11-30 09:47:09 -08001446 spin_unlock(&ci->i_ceph_lock);
Sage Weil355da1e2009-10-06 11:31:08 -07001447
Yan, Zhengb0d7c222013-08-12 21:42:15 -07001448 truncate_inode_pages(inode->i_mapping, 0);
Sage Weil355da1e2009-10-06 11:31:08 -07001449
Sage Weilbe655592011-11-30 09:47:09 -08001450 spin_lock(&ci->i_ceph_lock);
Sage Weilcd045cb2010-11-04 11:05:05 -07001451 if (orig_gen == ci->i_rdcache_gen &&
1452 orig_gen == ci->i_rdcache_revoking) {
Sage Weil355da1e2009-10-06 11:31:08 -07001453 dout("invalidate_pages %p gen %d successful\n", inode,
1454 ci->i_rdcache_gen);
Sage Weilcd045cb2010-11-04 11:05:05 -07001455 ci->i_rdcache_revoking--;
Sage Weil355da1e2009-10-06 11:31:08 -07001456 check = 1;
1457 } else {
Sage Weilcd045cb2010-11-04 11:05:05 -07001458 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1459 inode, orig_gen, ci->i_rdcache_gen,
1460 ci->i_rdcache_revoking);
Sage Weil355da1e2009-10-06 11:31:08 -07001461 }
Sage Weilbe655592011-11-30 09:47:09 -08001462 spin_unlock(&ci->i_ceph_lock);
Yan, Zhengb0d7c222013-08-12 21:42:15 -07001463 mutex_unlock(&ci->i_truncate_mutex);
Sage Weil355da1e2009-10-06 11:31:08 -07001464
1465 if (check)
1466 ceph_check_caps(ci, 0, NULL);
1467out:
1468 iput(inode);
1469}
1470
1471
1472/*
Yan, Zheng3f999692013-03-01 10:57:54 +08001473 * called by trunc_wq;
Sage Weil355da1e2009-10-06 11:31:08 -07001474 *
1475 * We also truncate in a separate thread as well.
1476 */
Sage Weil3c6f6b72010-02-09 15:24:44 -08001477static void ceph_vmtruncate_work(struct work_struct *work)
Sage Weil355da1e2009-10-06 11:31:08 -07001478{
1479 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1480 i_vmtruncate_work);
1481 struct inode *inode = &ci->vfs_inode;
1482
1483 dout("vmtruncate_work %p\n", inode);
Yan, Zhengb415bf42013-07-02 12:40:19 +08001484 __ceph_do_pending_vmtruncate(inode);
Sage Weil355da1e2009-10-06 11:31:08 -07001485 iput(inode);
1486}
1487
1488/*
Sage Weil3c6f6b72010-02-09 15:24:44 -08001489 * Queue an async vmtruncate. If we fail to queue work, we will handle
1490 * the truncation the next time we call __ceph_do_pending_vmtruncate.
1491 */
1492void ceph_queue_vmtruncate(struct inode *inode)
1493{
1494 struct ceph_inode_info *ci = ceph_inode(inode);
1495
Sage Weil15a20152011-11-05 22:06:31 -07001496 ihold(inode);
Milosz Tanski99ccbd22013-08-21 17:29:54 -04001497
Cheng Renquan640ef792010-03-26 17:40:33 +08001498 if (queue_work(ceph_sb_to_client(inode->i_sb)->trunc_wq,
Sage Weil3c6f6b72010-02-09 15:24:44 -08001499 &ci->i_vmtruncate_work)) {
1500 dout("ceph_queue_vmtruncate %p\n", inode);
Sage Weil3c6f6b72010-02-09 15:24:44 -08001501 } else {
1502 dout("ceph_queue_vmtruncate %p failed, pending=%d\n",
1503 inode, ci->i_truncate_pending);
Sage Weil15a20152011-11-05 22:06:31 -07001504 iput(inode);
Sage Weil3c6f6b72010-02-09 15:24:44 -08001505 }
1506}
1507
1508/*
Sage Weil355da1e2009-10-06 11:31:08 -07001509 * Make sure any pending truncation is applied before doing anything
1510 * that may depend on it.
1511 */
Yan, Zhengb415bf42013-07-02 12:40:19 +08001512void __ceph_do_pending_vmtruncate(struct inode *inode)
Sage Weil355da1e2009-10-06 11:31:08 -07001513{
1514 struct ceph_inode_info *ci = ceph_inode(inode);
1515 u64 to;
Yan, Zhenga85f50b2012-11-19 10:49:08 +08001516 int wrbuffer_refs, finish = 0;
Sage Weil355da1e2009-10-06 11:31:08 -07001517
Yan, Zhengb0d7c222013-08-12 21:42:15 -07001518 mutex_lock(&ci->i_truncate_mutex);
Sage Weil355da1e2009-10-06 11:31:08 -07001519retry:
Sage Weilbe655592011-11-30 09:47:09 -08001520 spin_lock(&ci->i_ceph_lock);
Sage Weil355da1e2009-10-06 11:31:08 -07001521 if (ci->i_truncate_pending == 0) {
1522 dout("__do_pending_vmtruncate %p none pending\n", inode);
Sage Weilbe655592011-11-30 09:47:09 -08001523 spin_unlock(&ci->i_ceph_lock);
Yan, Zhengb0d7c222013-08-12 21:42:15 -07001524 mutex_unlock(&ci->i_truncate_mutex);
Sage Weil355da1e2009-10-06 11:31:08 -07001525 return;
1526 }
1527
1528 /*
1529 * make sure any dirty snapped pages are flushed before we
1530 * possibly truncate them.. so write AND block!
1531 */
1532 if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) {
1533 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1534 inode);
Sage Weilbe655592011-11-30 09:47:09 -08001535 spin_unlock(&ci->i_ceph_lock);
Sage Weil355da1e2009-10-06 11:31:08 -07001536 filemap_write_and_wait_range(&inode->i_data, 0,
1537 inode->i_sb->s_maxbytes);
1538 goto retry;
1539 }
1540
Yan, Zhengb0d7c222013-08-12 21:42:15 -07001541 /* there should be no reader or writer */
1542 WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref);
1543
Sage Weil355da1e2009-10-06 11:31:08 -07001544 to = ci->i_truncate_size;
1545 wrbuffer_refs = ci->i_wrbuffer_ref;
1546 dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode,
1547 ci->i_truncate_pending, to);
Sage Weilbe655592011-11-30 09:47:09 -08001548 spin_unlock(&ci->i_ceph_lock);
Sage Weil355da1e2009-10-06 11:31:08 -07001549
1550 truncate_inode_pages(inode->i_mapping, to);
1551
Sage Weilbe655592011-11-30 09:47:09 -08001552 spin_lock(&ci->i_ceph_lock);
Yan, Zhenga85f50b2012-11-19 10:49:08 +08001553 if (to == ci->i_truncate_size) {
1554 ci->i_truncate_pending = 0;
1555 finish = 1;
1556 }
Sage Weilbe655592011-11-30 09:47:09 -08001557 spin_unlock(&ci->i_ceph_lock);
Yan, Zhenga85f50b2012-11-19 10:49:08 +08001558 if (!finish)
1559 goto retry;
Sage Weil355da1e2009-10-06 11:31:08 -07001560
Yan, Zhengb0d7c222013-08-12 21:42:15 -07001561 mutex_unlock(&ci->i_truncate_mutex);
1562
Sage Weil355da1e2009-10-06 11:31:08 -07001563 if (wrbuffer_refs == 0)
1564 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
Yan, Zhenga85f50b2012-11-19 10:49:08 +08001565
1566 wake_up_all(&ci->i_cap_wq);
Sage Weil355da1e2009-10-06 11:31:08 -07001567}
1568
Sage Weil355da1e2009-10-06 11:31:08 -07001569/*
1570 * symlinks
1571 */
1572static void *ceph_sym_follow_link(struct dentry *dentry, struct nameidata *nd)
1573{
1574 struct ceph_inode_info *ci = ceph_inode(dentry->d_inode);
1575 nd_set_link(nd, ci->i_symlink);
1576 return NULL;
1577}
1578
1579static const struct inode_operations ceph_symlink_iops = {
1580 .readlink = generic_readlink,
1581 .follow_link = ceph_sym_follow_link,
Yan, Zheng0b932672013-04-07 16:28:49 +08001582 .setattr = ceph_setattr,
1583 .getattr = ceph_getattr,
1584 .setxattr = ceph_setxattr,
1585 .getxattr = ceph_getxattr,
1586 .listxattr = ceph_listxattr,
1587 .removexattr = ceph_removexattr,
Sage Weil355da1e2009-10-06 11:31:08 -07001588};
1589
1590/*
1591 * setattr
1592 */
1593int ceph_setattr(struct dentry *dentry, struct iattr *attr)
1594{
1595 struct inode *inode = dentry->d_inode;
1596 struct ceph_inode_info *ci = ceph_inode(inode);
Sage Weil5f21c962011-07-26 11:30:29 -07001597 struct inode *parent_inode;
Sage Weil355da1e2009-10-06 11:31:08 -07001598 const unsigned int ia_valid = attr->ia_valid;
1599 struct ceph_mds_request *req;
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001600 struct ceph_mds_client *mdsc = ceph_sb_to_client(dentry->d_sb)->mdsc;
Sage Weil355da1e2009-10-06 11:31:08 -07001601 int issued;
1602 int release = 0, dirtied = 0;
1603 int mask = 0;
1604 int err = 0;
Sage Weilfca65b42011-05-04 11:33:47 -07001605 int inode_dirty_flags = 0;
Sage Weil355da1e2009-10-06 11:31:08 -07001606
1607 if (ceph_snap(inode) != CEPH_NOSNAP)
1608 return -EROFS;
1609
Sage Weil355da1e2009-10-06 11:31:08 -07001610 err = inode_change_ok(inode, attr);
1611 if (err != 0)
1612 return err;
1613
1614 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR,
1615 USE_AUTH_MDS);
1616 if (IS_ERR(req))
1617 return PTR_ERR(req);
1618
Sage Weilbe655592011-11-30 09:47:09 -08001619 spin_lock(&ci->i_ceph_lock);
Sage Weil355da1e2009-10-06 11:31:08 -07001620 issued = __ceph_caps_issued(ci, NULL);
1621 dout("setattr %p issued %s\n", inode, ceph_cap_string(issued));
1622
1623 if (ia_valid & ATTR_UID) {
1624 dout("setattr %p uid %d -> %d\n", inode,
Eric W. Biedermanbd2bae62013-01-31 04:05:39 -08001625 from_kuid(&init_user_ns, inode->i_uid),
1626 from_kuid(&init_user_ns, attr->ia_uid));
Sage Weil355da1e2009-10-06 11:31:08 -07001627 if (issued & CEPH_CAP_AUTH_EXCL) {
1628 inode->i_uid = attr->ia_uid;
1629 dirtied |= CEPH_CAP_AUTH_EXCL;
1630 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
Eric W. Biedermanab871b92013-01-31 03:40:12 -08001631 !uid_eq(attr->ia_uid, inode->i_uid)) {
1632 req->r_args.setattr.uid = cpu_to_le32(
1633 from_kuid(&init_user_ns, attr->ia_uid));
Sage Weil355da1e2009-10-06 11:31:08 -07001634 mask |= CEPH_SETATTR_UID;
1635 release |= CEPH_CAP_AUTH_SHARED;
1636 }
1637 }
1638 if (ia_valid & ATTR_GID) {
1639 dout("setattr %p gid %d -> %d\n", inode,
Eric W. Biedermanbd2bae62013-01-31 04:05:39 -08001640 from_kgid(&init_user_ns, inode->i_gid),
1641 from_kgid(&init_user_ns, attr->ia_gid));
Sage Weil355da1e2009-10-06 11:31:08 -07001642 if (issued & CEPH_CAP_AUTH_EXCL) {
1643 inode->i_gid = attr->ia_gid;
1644 dirtied |= CEPH_CAP_AUTH_EXCL;
1645 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
Eric W. Biedermanab871b92013-01-31 03:40:12 -08001646 !gid_eq(attr->ia_gid, inode->i_gid)) {
1647 req->r_args.setattr.gid = cpu_to_le32(
1648 from_kgid(&init_user_ns, attr->ia_gid));
Sage Weil355da1e2009-10-06 11:31:08 -07001649 mask |= CEPH_SETATTR_GID;
1650 release |= CEPH_CAP_AUTH_SHARED;
1651 }
1652 }
1653 if (ia_valid & ATTR_MODE) {
1654 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode,
1655 attr->ia_mode);
1656 if (issued & CEPH_CAP_AUTH_EXCL) {
1657 inode->i_mode = attr->ia_mode;
1658 dirtied |= CEPH_CAP_AUTH_EXCL;
1659 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1660 attr->ia_mode != inode->i_mode) {
1661 req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode);
1662 mask |= CEPH_SETATTR_MODE;
1663 release |= CEPH_CAP_AUTH_SHARED;
1664 }
1665 }
1666
1667 if (ia_valid & ATTR_ATIME) {
1668 dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode,
1669 inode->i_atime.tv_sec, inode->i_atime.tv_nsec,
1670 attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec);
1671 if (issued & CEPH_CAP_FILE_EXCL) {
1672 ci->i_time_warp_seq++;
1673 inode->i_atime = attr->ia_atime;
1674 dirtied |= CEPH_CAP_FILE_EXCL;
1675 } else if ((issued & CEPH_CAP_FILE_WR) &&
1676 timespec_compare(&inode->i_atime,
1677 &attr->ia_atime) < 0) {
1678 inode->i_atime = attr->ia_atime;
1679 dirtied |= CEPH_CAP_FILE_WR;
1680 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1681 !timespec_equal(&inode->i_atime, &attr->ia_atime)) {
1682 ceph_encode_timespec(&req->r_args.setattr.atime,
1683 &attr->ia_atime);
1684 mask |= CEPH_SETATTR_ATIME;
1685 release |= CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_RD |
1686 CEPH_CAP_FILE_WR;
1687 }
1688 }
1689 if (ia_valid & ATTR_MTIME) {
1690 dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode,
1691 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
1692 attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec);
1693 if (issued & CEPH_CAP_FILE_EXCL) {
1694 ci->i_time_warp_seq++;
1695 inode->i_mtime = attr->ia_mtime;
1696 dirtied |= CEPH_CAP_FILE_EXCL;
1697 } else if ((issued & CEPH_CAP_FILE_WR) &&
1698 timespec_compare(&inode->i_mtime,
1699 &attr->ia_mtime) < 0) {
1700 inode->i_mtime = attr->ia_mtime;
1701 dirtied |= CEPH_CAP_FILE_WR;
1702 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1703 !timespec_equal(&inode->i_mtime, &attr->ia_mtime)) {
1704 ceph_encode_timespec(&req->r_args.setattr.mtime,
1705 &attr->ia_mtime);
1706 mask |= CEPH_SETATTR_MTIME;
1707 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
1708 CEPH_CAP_FILE_WR;
1709 }
1710 }
1711 if (ia_valid & ATTR_SIZE) {
1712 dout("setattr %p size %lld -> %lld\n", inode,
1713 inode->i_size, attr->ia_size);
1714 if (attr->ia_size > inode->i_sb->s_maxbytes) {
1715 err = -EINVAL;
1716 goto out;
1717 }
1718 if ((issued & CEPH_CAP_FILE_EXCL) &&
1719 attr->ia_size > inode->i_size) {
1720 inode->i_size = attr->ia_size;
Sage Weil355da1e2009-10-06 11:31:08 -07001721 inode->i_blocks =
1722 (attr->ia_size + (1 << 9) - 1) >> 9;
1723 inode->i_ctime = attr->ia_ctime;
1724 ci->i_reported_size = attr->ia_size;
1725 dirtied |= CEPH_CAP_FILE_EXCL;
1726 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1727 attr->ia_size != inode->i_size) {
1728 req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
1729 req->r_args.setattr.old_size =
1730 cpu_to_le64(inode->i_size);
1731 mask |= CEPH_SETATTR_SIZE;
1732 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
1733 CEPH_CAP_FILE_WR;
1734 }
1735 }
1736
1737 /* these do nothing */
1738 if (ia_valid & ATTR_CTIME) {
1739 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME|
1740 ATTR_MODE|ATTR_UID|ATTR_GID)) == 0;
1741 dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode,
1742 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
1743 attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec,
1744 only ? "ctime only" : "ignored");
1745 inode->i_ctime = attr->ia_ctime;
1746 if (only) {
1747 /*
1748 * if kernel wants to dirty ctime but nothing else,
1749 * we need to choose a cap to dirty under, or do
1750 * a almost-no-op setattr
1751 */
1752 if (issued & CEPH_CAP_AUTH_EXCL)
1753 dirtied |= CEPH_CAP_AUTH_EXCL;
1754 else if (issued & CEPH_CAP_FILE_EXCL)
1755 dirtied |= CEPH_CAP_FILE_EXCL;
1756 else if (issued & CEPH_CAP_XATTR_EXCL)
1757 dirtied |= CEPH_CAP_XATTR_EXCL;
1758 else
1759 mask |= CEPH_SETATTR_CTIME;
1760 }
1761 }
1762 if (ia_valid & ATTR_FILE)
1763 dout("setattr %p ATTR_FILE ... hrm!\n", inode);
1764
1765 if (dirtied) {
Sage Weilfca65b42011-05-04 11:33:47 -07001766 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied);
Sage Weil355da1e2009-10-06 11:31:08 -07001767 inode->i_ctime = CURRENT_TIME;
1768 }
1769
1770 release &= issued;
Sage Weilbe655592011-11-30 09:47:09 -08001771 spin_unlock(&ci->i_ceph_lock);
Sage Weil355da1e2009-10-06 11:31:08 -07001772
Sage Weilfca65b42011-05-04 11:33:47 -07001773 if (inode_dirty_flags)
1774 __mark_inode_dirty(inode, inode_dirty_flags);
1775
Sage Weil355da1e2009-10-06 11:31:08 -07001776 if (mask) {
Sage Weil70b666c2011-05-27 09:24:26 -07001777 req->r_inode = inode;
1778 ihold(inode);
Sage Weil355da1e2009-10-06 11:31:08 -07001779 req->r_inode_drop = release;
1780 req->r_args.setattr.mask = cpu_to_le32(mask);
1781 req->r_num_caps = 1;
Sage Weil5f21c962011-07-26 11:30:29 -07001782 parent_inode = ceph_get_dentry_parent_inode(dentry);
Sage Weil355da1e2009-10-06 11:31:08 -07001783 err = ceph_mdsc_do_request(mdsc, parent_inode, req);
Sage Weil5f21c962011-07-26 11:30:29 -07001784 iput(parent_inode);
Sage Weil355da1e2009-10-06 11:31:08 -07001785 }
1786 dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err,
1787 ceph_cap_string(dirtied), mask);
1788
1789 ceph_mdsc_put_request(req);
Yan, Zhengb0d7c222013-08-12 21:42:15 -07001790 if (mask & CEPH_SETATTR_SIZE)
1791 __ceph_do_pending_vmtruncate(inode);
Sage Weil355da1e2009-10-06 11:31:08 -07001792 return err;
1793out:
Sage Weilbe655592011-11-30 09:47:09 -08001794 spin_unlock(&ci->i_ceph_lock);
Sage Weil355da1e2009-10-06 11:31:08 -07001795 ceph_mdsc_put_request(req);
1796 return err;
1797}
1798
1799/*
1800 * Verify that we have a lease on the given mask. If not,
1801 * do a getattr against an mds.
1802 */
1803int ceph_do_getattr(struct inode *inode, int mask)
1804{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001805 struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
1806 struct ceph_mds_client *mdsc = fsc->mdsc;
Sage Weil355da1e2009-10-06 11:31:08 -07001807 struct ceph_mds_request *req;
1808 int err;
1809
1810 if (ceph_snap(inode) == CEPH_SNAPDIR) {
1811 dout("do_getattr inode %p SNAPDIR\n", inode);
1812 return 0;
1813 }
1814
Sage Weilb7495fc2010-11-09 12:43:12 -08001815 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 -07001816 if (ceph_caps_issued_mask(ceph_inode(inode), mask, 1))
1817 return 0;
1818
1819 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
1820 if (IS_ERR(req))
1821 return PTR_ERR(req);
Sage Weil70b666c2011-05-27 09:24:26 -07001822 req->r_inode = inode;
1823 ihold(inode);
Sage Weil355da1e2009-10-06 11:31:08 -07001824 req->r_num_caps = 1;
1825 req->r_args.getattr.mask = cpu_to_le32(mask);
1826 err = ceph_mdsc_do_request(mdsc, NULL, req);
1827 ceph_mdsc_put_request(req);
1828 dout("do_getattr result=%d\n", err);
1829 return err;
1830}
1831
1832
1833/*
1834 * Check inode permissions. We verify we have a valid value for
1835 * the AUTH cap, then call the generic handler.
1836 */
Al Viro10556cb2011-06-20 19:28:19 -04001837int ceph_permission(struct inode *inode, int mask)
Sage Weil355da1e2009-10-06 11:31:08 -07001838{
Nick Pigginb74c79e2011-01-07 17:49:58 +11001839 int err;
1840
Al Viro10556cb2011-06-20 19:28:19 -04001841 if (mask & MAY_NOT_BLOCK)
Nick Pigginb74c79e2011-01-07 17:49:58 +11001842 return -ECHILD;
1843
1844 err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED);
Sage Weil355da1e2009-10-06 11:31:08 -07001845
1846 if (!err)
Al Viro2830ba72011-06-20 19:16:29 -04001847 err = generic_permission(inode, mask);
Sage Weil355da1e2009-10-06 11:31:08 -07001848 return err;
1849}
1850
1851/*
1852 * Get all attributes. Hopefully somedata we'll have a statlite()
1853 * and can limit the fields we require to be accurate.
1854 */
1855int ceph_getattr(struct vfsmount *mnt, struct dentry *dentry,
1856 struct kstat *stat)
1857{
1858 struct inode *inode = dentry->d_inode;
Sage Weil232d4b02009-10-21 11:21:49 -07001859 struct ceph_inode_info *ci = ceph_inode(inode);
Sage Weil355da1e2009-10-06 11:31:08 -07001860 int err;
1861
1862 err = ceph_do_getattr(inode, CEPH_STAT_CAP_INODE_ALL);
1863 if (!err) {
1864 generic_fillattr(inode, stat);
Yehuda Sadehad1fee92011-01-21 16:44:03 -08001865 stat->ino = ceph_translate_ino(inode->i_sb, inode->i_ino);
Sage Weil355da1e2009-10-06 11:31:08 -07001866 if (ceph_snap(inode) != CEPH_NOSNAP)
1867 stat->dev = ceph_snap(inode);
1868 else
1869 stat->dev = 0;
Sage Weil232d4b02009-10-21 11:21:49 -07001870 if (S_ISDIR(inode->i_mode)) {
Yehuda Sadeh1c1266b2011-01-12 16:53:27 -08001871 if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb),
1872 RBYTES))
1873 stat->size = ci->i_rbytes;
1874 else
1875 stat->size = ci->i_files + ci->i_subdirs;
Sage Weil232d4b02009-10-21 11:21:49 -07001876 stat->blocks = 0;
Sage Weil355da1e2009-10-06 11:31:08 -07001877 stat->blksize = 65536;
Sage Weil232d4b02009-10-21 11:21:49 -07001878 }
Sage Weil355da1e2009-10-06 11:31:08 -07001879 }
1880 return err;
1881}