projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
ceph: avoid sending unnessesary FLUSHSNAP message
[linux-2.6-block.git] / fs / ceph / inode.c
CommitLineData
3d14c5d2 1#include <linux/ceph/ceph_debug.h>
355da1eb
SW		 2
3#include <linux/module.h>
4#include <linux/fs.h>
355da1eb
SW		 5#include <linux/slab.h>
6#include <linux/string.h>
7#include <linux/uaccess.h>
8#include <linux/kernel.h>
9#include <linux/namei.h>
10#include <linux/writeback.h>
11#include <linux/vmalloc.h>
4db658ea 12#include <linux/posix_acl.h>
3e7fbe9c 13#include <linux/random.h>
355da1eb
SW		 14
15#include "super.h"
3d14c5d2 16#include "mds_client.h"
99ccbd22 17#include "cache.h"
3d14c5d2 18#include <linux/ceph/decode.h>
355da1eb
SW		 19
20/*
21 * Ceph inode operations
22 *
23 * Implement basic inode helpers (get, alloc) and inode ops (getattr,
24 * setattr, etc.), xattr helpers, and helpers for assimilating
25 * metadata returned by the MDS into our cache.
26 *
27 * Also define helpers for doing asynchronous writeback, invalidation,
28 * and truncation for the benefit of those who can't afford to block
29 * (typically because they are in the message handler path).
30 */
31
32static const struct inode_operations ceph_symlink_iops;
33
3c6f6b79
SW		 34static void ceph_invalidate_work(struct work_struct *work);
35static void ceph_writeback_work(struct work_struct *work);
36static void ceph_vmtruncate_work(struct work_struct *work);
355da1eb
SW		 37
38/*
39 * find or create an inode, given the ceph ino number
40 */
ad1fee96
YS		 41static int ceph_set_ino_cb(struct inode *inode, void *data)
42{
43 ceph_inode(inode)->i_vino = *(struct ceph_vino *)data;
44 inode->i_ino = ceph_vino_to_ino(*(struct ceph_vino *)data);
45 return 0;
46}
47
355da1eb
SW		 48struct inode *ceph_get_inode(struct super_block *sb, struct ceph_vino vino)
49{
50 struct inode *inode;
51 ino_t t = ceph_vino_to_ino(vino);
52
53 inode = iget5_locked(sb, t, ceph_ino_compare, ceph_set_ino_cb, &vino);
54 if (inode == NULL)
55 return ERR_PTR(-ENOMEM);
56 if (inode->i_state & I_NEW) {
57 dout("get_inode created new inode %p %llx.%llx ino %llx\n",
58 inode, ceph_vinop(inode), (u64)inode->i_ino);
59 unlock_new_inode(inode);
60 }
61
62 dout("get_inode on %lu=%llx.%llx got %p\n", inode->i_ino, vino.ino,
63 vino.snap, inode);
64 return inode;
65}
66
67/*
68 * get/constuct snapdir inode for a given directory
69 */
70struct inode *ceph_get_snapdir(struct inode *parent)
71{
72 struct ceph_vino vino = {
73 .ino = ceph_ino(parent),
74 .snap = CEPH_SNAPDIR,
75 };
76 struct inode *inode = ceph_get_inode(parent->i_sb, vino);
b377ff13 77 struct ceph_inode_info *ci = ceph_inode(inode);
355da1eb
SW		 78
79 BUG_ON(!S_ISDIR(parent->i_mode));
80 if (IS_ERR(inode))
7e34bc52 81 return inode;
355da1eb
SW		 82 inode->i_mode = parent->i_mode;
83 inode->i_uid = parent->i_uid;
84 inode->i_gid = parent->i_gid;
38c48b5f
YZ		 85 inode->i_op = &ceph_snapdir_iops;
86 inode->i_fop = &ceph_snapdir_fops;
b377ff13
SW		 87 ci->i_snap_caps = CEPH_CAP_PIN; /* so we can open */
88 ci->i_rbytes = 0;
355da1eb
SW		 89 return inode;
90}
91
92const struct inode_operations ceph_file_iops = {
93 .permission = ceph_permission,
94 .setattr = ceph_setattr,
95 .getattr = ceph_getattr,
96 .setxattr = ceph_setxattr,
97 .getxattr = ceph_getxattr,
98 .listxattr = ceph_listxattr,
99 .removexattr = ceph_removexattr,
7221fe4c 100 .get_acl = ceph_get_acl,
72466d0b 101 .set_acl = ceph_set_acl,
355da1eb
SW		 102};
103
104
105/*
106 * We use a 'frag tree' to keep track of the MDS's directory fragments
107 * for a given inode (usually there is just a single fragment). We
108 * need to know when a child frag is delegated to a new MDS, or when
109 * it is flagged as replicated, so we can direct our requests
110 * accordingly.
111 */
112
113/*
114 * find/create a frag in the tree
115 */
116static struct ceph_inode_frag *__get_or_create_frag(struct ceph_inode_info *ci,
117 u32 f)
118{
119 struct rb_node **p;
120 struct rb_node *parent = NULL;
121 struct ceph_inode_frag *frag;
122 int c;
123
124 p = &ci->i_fragtree.rb_node;
125 while (*p) {
126 parent = *p;
127 frag = rb_entry(parent, struct ceph_inode_frag, node);
128 c = ceph_frag_compare(f, frag->frag);
129 if (c < 0)
130 p = &(*p)->rb_left;
131 else if (c > 0)
132 p = &(*p)->rb_right;
133 else
134 return frag;
135 }
136
137 frag = kmalloc(sizeof(*frag), GFP_NOFS);
138 if (!frag) {
139 pr_err("__get_or_create_frag ENOMEM on %p %llx.%llx "
140 "frag %x\n", &ci->vfs_inode,
141 ceph_vinop(&ci->vfs_inode), f);
142 return ERR_PTR(-ENOMEM);
143 }
144 frag->frag = f;
145 frag->split_by = 0;
146 frag->mds = -1;
147 frag->ndist = 0;
148
149 rb_link_node(&frag->node, parent, p);
150 rb_insert_color(&frag->node, &ci->i_fragtree);
151
152 dout("get_or_create_frag added %llx.%llx frag %x\n",
153 ceph_vinop(&ci->vfs_inode), f);
154 return frag;
155}
156
157/*
158 * find a specific frag @f
159 */
160struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci, u32 f)
161{
162 struct rb_node *n = ci->i_fragtree.rb_node;
163
164 while (n) {
165 struct ceph_inode_frag *frag =
166 rb_entry(n, struct ceph_inode_frag, node);
167 int c = ceph_frag_compare(f, frag->frag);
168 if (c < 0)
169 n = n->rb_left;
170 else if (c > 0)
171 n = n->rb_right;
172 else
173 return frag;
174 }
175 return NULL;
176}
177
178/*
179 * Choose frag containing the given value @v. If @pfrag is
180 * specified, copy the frag delegation info to the caller if
181 * it is present.
182 */
3e7fbe9c
YZ		 183static u32 __ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
184 struct ceph_inode_frag *pfrag, int *found)
355da1eb
SW		 185{
186 u32 t = ceph_frag_make(0, 0);
187 struct ceph_inode_frag *frag;
188 unsigned nway, i;
189 u32 n;
190
191 if (found)
192 *found = 0;
193
355da1eb
SW		 194 while (1) {
195 WARN_ON(!ceph_frag_contains_value(t, v));
196 frag = __ceph_find_frag(ci, t);
197 if (!frag)
198 break; /* t is a leaf */
199 if (frag->split_by == 0) {
200 if (pfrag)
201 memcpy(pfrag, frag, sizeof(*pfrag));
202 if (found)
203 *found = 1;
204 break;
205 }
206
207 /* choose child */
208 nway = 1 << frag->split_by;
209 dout("choose_frag(%x) %x splits by %d (%d ways)\n", v, t,
210 frag->split_by, nway);
211 for (i = 0; i < nway; i++) {
212 n = ceph_frag_make_child(t, frag->split_by, i);
213 if (ceph_frag_contains_value(n, v)) {
214 t = n;
215 break;
216 }
217 }
218 BUG_ON(i == nway);
219 }
220 dout("choose_frag(%x) = %x\n", v, t);
221
355da1eb
SW		 222 return t;
223}
224
3e7fbe9c
YZ		 225u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
226 struct ceph_inode_frag *pfrag, int *found)
227{
228 u32 ret;
229 mutex_lock(&ci->i_fragtree_mutex);
230 ret = __ceph_choose_frag(ci, v, pfrag, found);
231 mutex_unlock(&ci->i_fragtree_mutex);
232 return ret;
233}
234
355da1eb
SW		 235/*
236 * Process dirfrag (delegation) info from the mds. Include leaf
237 * fragment in tree ONLY if ndist > 0. Otherwise, only
238 * branches/splits are included in i_fragtree)
239 */
240static int ceph_fill_dirfrag(struct inode *inode,
241 struct ceph_mds_reply_dirfrag *dirinfo)
242{
243 struct ceph_inode_info *ci = ceph_inode(inode);
244 struct ceph_inode_frag *frag;
245 u32 id = le32_to_cpu(dirinfo->frag);
246 int mds = le32_to_cpu(dirinfo->auth);
247 int ndist = le32_to_cpu(dirinfo->ndist);
8d08503c 248 int diri_auth = -1;
355da1eb
SW		 249 int i;
250 int err = 0;
251
8d08503c
YZ		 252 spin_lock(&ci->i_ceph_lock);
253 if (ci->i_auth_cap)
254 diri_auth = ci->i_auth_cap->mds;
255 spin_unlock(&ci->i_ceph_lock);
256
355da1eb 257 mutex_lock(&ci->i_fragtree_mutex);
8d08503c 258 if (ndist == 0 && mds == diri_auth) {
355da1eb
SW		 259 /* no delegation info needed. */
260 frag = __ceph_find_frag(ci, id);
261 if (!frag)
262 goto out;
263 if (frag->split_by == 0) {
264 /* tree leaf, remove */
265 dout("fill_dirfrag removed %llx.%llx frag %x"
266 " (no ref)\n", ceph_vinop(inode), id);
267 rb_erase(&frag->node, &ci->i_fragtree);
268 kfree(frag);
269 } else {
270 /* tree branch, keep and clear */
271 dout("fill_dirfrag cleared %llx.%llx frag %x"
272 " referral\n", ceph_vinop(inode), id);
273 frag->mds = -1;
274 frag->ndist = 0;
275 }
276 goto out;
277 }
278
279
280 /* find/add this frag to store mds delegation info */
281 frag = __get_or_create_frag(ci, id);
282 if (IS_ERR(frag)) {
283 /* this is not the end of the world; we can continue
284 with bad/inaccurate delegation info */
285 pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n",
286 ceph_vinop(inode), le32_to_cpu(dirinfo->frag));
287 err = -ENOMEM;
288 goto out;
289 }
290
291 frag->mds = mds;
292 frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP);
293 for (i = 0; i < frag->ndist; i++)
294 frag->dist[i] = le32_to_cpu(dirinfo->dist[i]);
295 dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n",
296 ceph_vinop(inode), frag->frag, frag->ndist);
297
298out:
299 mutex_unlock(&ci->i_fragtree_mutex);
300 return err;
301}
302
3e7fbe9c
YZ		 303static int ceph_fill_fragtree(struct inode *inode,
304 struct ceph_frag_tree_head *fragtree,
305 struct ceph_mds_reply_dirfrag *dirinfo)
306{
307 struct ceph_inode_info *ci = ceph_inode(inode);
308 struct ceph_inode_frag *frag;
309 struct rb_node *rb_node;
310 int i;
311 u32 id, nsplits;
312 bool update = false;
313
314 mutex_lock(&ci->i_fragtree_mutex);
315 nsplits = le32_to_cpu(fragtree->nsplits);
316 if (nsplits) {
317 i = prandom_u32() % nsplits;
318 id = le32_to_cpu(fragtree->splits[i].frag);
319 if (!__ceph_find_frag(ci, id))
320 update = true;
321 } else if (!RB_EMPTY_ROOT(&ci->i_fragtree)) {
322 rb_node = rb_first(&ci->i_fragtree);
323 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
324 if (frag->frag != ceph_frag_make(0, 0) || rb_next(rb_node))
325 update = true;
326 }
327 if (!update && dirinfo) {
328 id = le32_to_cpu(dirinfo->frag);
329 if (id != __ceph_choose_frag(ci, id, NULL, NULL))
330 update = true;
331 }
332 if (!update)
333 goto out_unlock;
334
335 dout("fill_fragtree %llx.%llx\n", ceph_vinop(inode));
336 rb_node = rb_first(&ci->i_fragtree);
337 for (i = 0; i < nsplits; i++) {
338 id = le32_to_cpu(fragtree->splits[i].frag);
339 frag = NULL;
340 while (rb_node) {
341 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
342 if (ceph_frag_compare(frag->frag, id) >= 0) {
343 if (frag->frag != id)
344 frag = NULL;
345 else
346 rb_node = rb_next(rb_node);
347 break;
348 }
349 rb_node = rb_next(rb_node);
350 rb_erase(&frag->node, &ci->i_fragtree);
351 kfree(frag);
352 frag = NULL;
353 }
354 if (!frag) {
355 frag = __get_or_create_frag(ci, id);
356 if (IS_ERR(frag))
357 continue;
358 }
359 frag->split_by = le32_to_cpu(fragtree->splits[i].by);
360 dout(" frag %x split by %d\n", frag->frag, frag->split_by);
361 }
362 while (rb_node) {
363 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
364 rb_node = rb_next(rb_node);
365 rb_erase(&frag->node, &ci->i_fragtree);
366 kfree(frag);
367 }
368out_unlock:
369 mutex_unlock(&ci->i_fragtree_mutex);
370 return 0;
371}
355da1eb
SW		 372
373/*
374 * initialize a newly allocated inode.
375 */
376struct inode *ceph_alloc_inode(struct super_block *sb)
377{
378 struct ceph_inode_info *ci;
379 int i;
380
381 ci = kmem_cache_alloc(ceph_inode_cachep, GFP_NOFS);
382 if (!ci)
383 return NULL;
384
385 dout("alloc_inode %p\n", &ci->vfs_inode);
386
be655596
SW		 387 spin_lock_init(&ci->i_ceph_lock);
388
355da1eb 389 ci->i_version = 0;
31c542a1 390 ci->i_inline_version = 0;
355da1eb
SW		 391 ci->i_time_warp_seq = 0;
392 ci->i_ceph_flags = 0;
70db4f36 393 ci->i_ordered_count = 0;
2f276c51
YZ		 394 atomic_set(&ci->i_release_count, 1);
395 atomic_set(&ci->i_complete_count, 0);
355da1eb
SW		 396 ci->i_symlink = NULL;
397
6c0f3af7
SW		 398 memset(&ci->i_dir_layout, 0, sizeof(ci->i_dir_layout));
399
355da1eb
SW		 400 ci->i_fragtree = RB_ROOT;
401 mutex_init(&ci->i_fragtree_mutex);
402
403 ci->i_xattrs.blob = NULL;
404 ci->i_xattrs.prealloc_blob = NULL;
405 ci->i_xattrs.dirty = false;
406 ci->i_xattrs.index = RB_ROOT;
407 ci->i_xattrs.count = 0;
408 ci->i_xattrs.names_size = 0;
409 ci->i_xattrs.vals_size = 0;
410 ci->i_xattrs.version = 0;
411 ci->i_xattrs.index_version = 0;
412
413 ci->i_caps = RB_ROOT;
414 ci->i_auth_cap = NULL;
415 ci->i_dirty_caps = 0;
416 ci->i_flushing_caps = 0;
417 INIT_LIST_HEAD(&ci->i_dirty_item);
418 INIT_LIST_HEAD(&ci->i_flushing_item);
419 ci->i_cap_flush_seq = 0;
420 ci->i_cap_flush_last_tid = 0;
421 memset(&ci->i_cap_flush_tid, 0, sizeof(ci->i_cap_flush_tid));
422 init_waitqueue_head(&ci->i_cap_wq);
423 ci->i_hold_caps_min = 0;
424 ci->i_hold_caps_max = 0;
425 INIT_LIST_HEAD(&ci->i_cap_delay_list);
355da1eb
SW		 426 INIT_LIST_HEAD(&ci->i_cap_snaps);
427 ci->i_head_snapc = NULL;
428 ci->i_snap_caps = 0;
429
430 for (i = 0; i < CEPH_FILE_MODE_NUM; i++)
431 ci->i_nr_by_mode[i] = 0;
432
b0d7c223 433 mutex_init(&ci->i_truncate_mutex);
355da1eb
SW		 434 ci->i_truncate_seq = 0;
435 ci->i_truncate_size = 0;
436 ci->i_truncate_pending = 0;
437
438 ci->i_max_size = 0;
439 ci->i_reported_size = 0;
440 ci->i_wanted_max_size = 0;
441 ci->i_requested_max_size = 0;
442
443 ci->i_pin_ref = 0;
444 ci->i_rd_ref = 0;
445 ci->i_rdcache_ref = 0;
446 ci->i_wr_ref = 0;
d3d0720d 447 ci->i_wb_ref = 0;
355da1eb
SW		 448 ci->i_wrbuffer_ref = 0;
449 ci->i_wrbuffer_ref_head = 0;
450 ci->i_shared_gen = 0;
451 ci->i_rdcache_gen = 0;
452 ci->i_rdcache_revoking = 0;
453
454 INIT_LIST_HEAD(&ci->i_unsafe_writes);
455 INIT_LIST_HEAD(&ci->i_unsafe_dirops);
456 spin_lock_init(&ci->i_unsafe_lock);
457
458 ci->i_snap_realm = NULL;
459 INIT_LIST_HEAD(&ci->i_snap_realm_item);
460 INIT_LIST_HEAD(&ci->i_snap_flush_item);
461
3c6f6b79
SW		 462 INIT_WORK(&ci->i_wb_work, ceph_writeback_work);
463 INIT_WORK(&ci->i_pg_inv_work, ceph_invalidate_work);
355da1eb
SW		 464
465 INIT_WORK(&ci->i_vmtruncate_work, ceph_vmtruncate_work);
466
99ccbd22
MT		 467 ceph_fscache_inode_init(ci);
468
355da1eb
SW		 469 return &ci->vfs_inode;
470}
471
fa0d7e3d
NP		 472static void ceph_i_callback(struct rcu_head *head)
473{
474 struct inode *inode = container_of(head, struct inode, i_rcu);
475 struct ceph_inode_info *ci = ceph_inode(inode);
476
fa0d7e3d
NP		 477 kmem_cache_free(ceph_inode_cachep, ci);
478}
479
355da1eb
SW		 480void ceph_destroy_inode(struct inode *inode)
481{
482 struct ceph_inode_info *ci = ceph_inode(inode);
483 struct ceph_inode_frag *frag;
484 struct rb_node *n;
485
486 dout("destroy_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode));
487
99ccbd22
MT		 488 ceph_fscache_unregister_inode_cookie(ci);
489
355da1eb
SW		 490 ceph_queue_caps_release(inode);
491
8b218b8a
SW		 492 /*
493 * we may still have a snap_realm reference if there are stray
d9df2783 494 * caps in i_snap_caps.
8b218b8a
SW		 495 */
496 if (ci->i_snap_realm) {
497 struct ceph_mds_client *mdsc =
3d14c5d2 498 ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc;
8b218b8a
SW		 499 struct ceph_snap_realm *realm = ci->i_snap_realm;
500
501 dout(" dropping residual ref to snap realm %p\n", realm);
502 spin_lock(&realm->inodes_with_caps_lock);
503 list_del_init(&ci->i_snap_realm_item);
504 spin_unlock(&realm->inodes_with_caps_lock);
505 ceph_put_snap_realm(mdsc, realm);
506 }
507
355da1eb
SW		 508 kfree(ci->i_symlink);
509 while ((n = rb_first(&ci->i_fragtree)) != NULL) {
510 frag = rb_entry(n, struct ceph_inode_frag, node);
511 rb_erase(n, &ci->i_fragtree);
512 kfree(frag);
513 }
514
515 __ceph_destroy_xattrs(ci);
b6c1d5b8
SW		 516 if (ci->i_xattrs.blob)
517 ceph_buffer_put(ci->i_xattrs.blob);
518 if (ci->i_xattrs.prealloc_blob)
519 ceph_buffer_put(ci->i_xattrs.prealloc_blob);
355da1eb 520
fa0d7e3d 521 call_rcu(&inode->i_rcu, ceph_i_callback);
355da1eb
SW		 522}
523
9f12bd11
YZ		 524int ceph_drop_inode(struct inode *inode)
525{
526 /*
527 * Positve dentry and corresponding inode are always accompanied
528 * in MDS reply. So no need to keep inode in the cache after
529 * dropping all its aliases.
530 */
531 return 1;
532}
533
355da1eb
SW		 534/*
535 * Helpers to fill in size, ctime, mtime, and atime. We have to be
536 * careful because either the client or MDS may have more up to date
537 * info, depending on which capabilities are held, and whether
538 * time_warp_seq or truncate_seq have increased. (Ordinarily, mtime
539 * and size are monotonically increasing, except when utimes() or
540 * truncate() increments the corresponding _seq values.)
541 */
542int ceph_fill_file_size(struct inode *inode, int issued,
543 u32 truncate_seq, u64 truncate_size, u64 size)
544{
545 struct ceph_inode_info *ci = ceph_inode(inode);
546 int queue_trunc = 0;
547
548 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 ||
549 (truncate_seq == ci->i_truncate_seq && size > inode->i_size)) {
550 dout("size %lld -> %llu\n", inode->i_size, size);
551 inode->i_size = size;
552 inode->i_blocks = (size + (1<<9) - 1) >> 9;
553 ci->i_reported_size = size;
554 if (truncate_seq != ci->i_truncate_seq) {
555 dout("truncate_seq %u -> %u\n",
556 ci->i_truncate_seq, truncate_seq);
557 ci->i_truncate_seq = truncate_seq;
b0d7c223
YZ		 558
559 /* the MDS should have revoked these caps */
560 WARN_ON_ONCE(issued & (CEPH_CAP_FILE_EXCL |
561 CEPH_CAP_FILE_RD |
562 CEPH_CAP_FILE_WR |
563 CEPH_CAP_FILE_LAZYIO));
3d497d85
YS		 564 /*
565 * If we hold relevant caps, or in the case where we're
566 * not the only client referencing this file and we
567 * don't hold those caps, then we need to check whether
568 * the file is either opened or mmaped
569 */
b0d7c223
YZ		 570 if ((issued & (CEPH_CAP_FILE_CACHE|
571 CEPH_CAP_FILE_BUFFER)) ||
3d497d85
YS		 572 mapping_mapped(inode->i_mapping) ||
573 __ceph_caps_file_wanted(ci)) {
355da1eb
SW		 574 ci->i_truncate_pending++;
575 queue_trunc = 1;
576 }
577 }
578 }
579 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 &&
580 ci->i_truncate_size != truncate_size) {
581 dout("truncate_size %lld -> %llu\n", ci->i_truncate_size,
582 truncate_size);
583 ci->i_truncate_size = truncate_size;
584 }
99ccbd22
MT		 585
586 if (queue_trunc)
587 ceph_fscache_invalidate(inode);
588
355da1eb
SW		 589 return queue_trunc;
590}
591
592void ceph_fill_file_time(struct inode *inode, int issued,
593 u64 time_warp_seq, struct timespec *ctime,
594 struct timespec *mtime, struct timespec *atime)
595{
596 struct ceph_inode_info *ci = ceph_inode(inode);
597 int warn = 0;
598
599 if (issued & (CEPH_CAP_FILE_EXCL|
600 CEPH_CAP_FILE_WR|
d8672d64
SW		 601 CEPH_CAP_FILE_BUFFER|
602 CEPH_CAP_AUTH_EXCL|
603 CEPH_CAP_XATTR_EXCL)) {
355da1eb
SW		 604 if (timespec_compare(ctime, &inode->i_ctime) > 0) {
605 dout("ctime %ld.%09ld -> %ld.%09ld inc w/ cap\n",
606 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
607 ctime->tv_sec, ctime->tv_nsec);
608 inode->i_ctime = *ctime;
609 }
610 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) {
611 /* the MDS did a utimes() */
612 dout("mtime %ld.%09ld -> %ld.%09ld "
613 "tw %d -> %d\n",
614 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
615 mtime->tv_sec, mtime->tv_nsec,
616 ci->i_time_warp_seq, (int)time_warp_seq);
617
618 inode->i_mtime = *mtime;
619 inode->i_atime = *atime;
620 ci->i_time_warp_seq = time_warp_seq;
621 } else if (time_warp_seq == ci->i_time_warp_seq) {
622 /* nobody did utimes(); take the max */
623 if (timespec_compare(mtime, &inode->i_mtime) > 0) {
624 dout("mtime %ld.%09ld -> %ld.%09ld inc\n",
625 inode->i_mtime.tv_sec,
626 inode->i_mtime.tv_nsec,
627 mtime->tv_sec, mtime->tv_nsec);
628 inode->i_mtime = *mtime;
629 }
630 if (timespec_compare(atime, &inode->i_atime) > 0) {
631 dout("atime %ld.%09ld -> %ld.%09ld inc\n",
632 inode->i_atime.tv_sec,
633 inode->i_atime.tv_nsec,
634 atime->tv_sec, atime->tv_nsec);
635 inode->i_atime = *atime;
636 }
637 } else if (issued & CEPH_CAP_FILE_EXCL) {
638 /* we did a utimes(); ignore mds values */
639 } else {
640 warn = 1;
641 }
642 } else {
d8672d64 643 /* we have no write|excl caps; whatever the MDS says is true */
355da1eb
SW		 644 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) {
645 inode->i_ctime = *ctime;
646 inode->i_mtime = *mtime;
647 inode->i_atime = *atime;
648 ci->i_time_warp_seq = time_warp_seq;
649 } else {
650 warn = 1;
651 }
652 }
653 if (warn) /* time_warp_seq shouldn't go backwards */
654 dout("%p mds time_warp_seq %llu < %u\n",
655 inode, time_warp_seq, ci->i_time_warp_seq);
656}
657
658/*
659 * Populate an inode based on info from mds. May be called on new or
660 * existing inodes.
661 */
01deead0 662static int fill_inode(struct inode *inode, struct page *locked_page,
355da1eb
SW		 663 struct ceph_mds_reply_info_in *iinfo,
664 struct ceph_mds_reply_dirfrag *dirinfo,
665 struct ceph_mds_session *session,
666 unsigned long ttl_from, int cap_fmode,
667 struct ceph_cap_reservation *caps_reservation)
668{
d9df2783 669 struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
355da1eb
SW		 670 struct ceph_mds_reply_inode *info = iinfo->in;
671 struct ceph_inode_info *ci = ceph_inode(inode);
f98a128a 672 int issued = 0, implemented, new_issued;
355da1eb 673 struct timespec mtime, atime, ctime;
355da1eb 674 struct ceph_buffer *xattr_blob = NULL;
d9df2783 675 struct ceph_cap *new_cap = NULL;
355da1eb 676 int err = 0;
d9df2783 677 bool wake = false;
f98a128a
YZ		 678 bool queue_trunc = false;
679 bool new_version = false;
31c542a1 680 bool fill_inline = false;
355da1eb
SW		 681
682 dout("fill_inode %p ino %llx.%llx v %llu had %llu\n",
683 inode, ceph_vinop(inode), le64_to_cpu(info->version),
684 ci->i_version);
685
d9df2783
YZ		 686 /* prealloc new cap struct */
687 if (info->cap.caps && ceph_snap(inode) == CEPH_NOSNAP)
688 new_cap = ceph_get_cap(mdsc, caps_reservation);
689
355da1eb
SW		 690 /*
691 * prealloc xattr data, if it looks like we'll need it. only
692 * if len > 4 (meaning there are actually xattrs; the first 4
693 * bytes are the xattr count).
694 */
695 if (iinfo->xattr_len > 4) {
b6c1d5b8 696 xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS);
355da1eb
SW		 697 if (!xattr_blob)
698 pr_err("fill_inode ENOMEM xattr blob %d bytes\n",
699 iinfo->xattr_len);
700 }
701
be655596 702 spin_lock(&ci->i_ceph_lock);
355da1eb
SW		 703
704 /*
705 * provided version will be odd if inode value is projected,
8bd59e01
SW		 706 * even if stable. skip the update if we have newer stable
707 * info (ours>=theirs, e.g. due to racing mds replies), unless
708 * we are getting projected (unstable) info (in which case the
709 * version is odd, and we want ours>theirs).
710 * us them
711 * 2 2 skip
712 * 3 2 skip
713 * 3 3 update
355da1eb 714 */
f98a128a
YZ		 715 if (ci->i_version == 0 ||
716 ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
717 le64_to_cpu(info->version) > (ci->i_version & ~1)))
718 new_version = true;
719
355da1eb
SW		 720 issued = __ceph_caps_issued(ci, &implemented);
721 issued |= implemented | __ceph_caps_dirty(ci);
f98a128a 722 new_issued = ~issued & le32_to_cpu(info->cap.caps);
355da1eb
SW		 723
724 /* update inode */
725 ci->i_version = le64_to_cpu(info->version);
726 inode->i_version++;
727 inode->i_rdev = le32_to_cpu(info->rdev);
f98a128a 728 inode->i_blkbits = fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1;
355da1eb 729
f98a128a
YZ		 730 if ((new_version || (new_issued & CEPH_CAP_AUTH_SHARED)) &&
731 (issued & CEPH_CAP_AUTH_EXCL) == 0) {
355da1eb 732 inode->i_mode = le32_to_cpu(info->mode);
ab871b90
EB		 733 inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(info->uid));
734 inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(info->gid));
355da1eb 735 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
bd2bae6a
EB		 736 from_kuid(&init_user_ns, inode->i_uid),
737 from_kgid(&init_user_ns, inode->i_gid));
355da1eb
SW		 738 }
739
f98a128a
YZ		 740 if ((new_version || (new_issued & CEPH_CAP_LINK_SHARED)) &&
741 (issued & CEPH_CAP_LINK_EXCL) == 0)
bfe86848 742 set_nlink(inode, le32_to_cpu(info->nlink));
355da1eb 743
f98a128a
YZ		 744 if (new_version || (new_issued & CEPH_CAP_ANY_RD)) {
745 /* be careful with mtime, atime, size */
746 ceph_decode_timespec(&atime, &info->atime);
747 ceph_decode_timespec(&mtime, &info->mtime);
748 ceph_decode_timespec(&ctime, &info->ctime);
749 ceph_fill_file_time(inode, issued,
750 le32_to_cpu(info->time_warp_seq),
751 &ctime, &mtime, &atime);
752 }
753
754 if (new_version ||
755 (new_issued & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR))) {
10183a69
YZ		 756 if (ci->i_layout.fl_pg_pool != info->layout.fl_pg_pool)
757 ci->i_ceph_flags &= ~CEPH_I_POOL_PERM;
f98a128a 758 ci->i_layout = info->layout;
10183a69 759
f98a128a
YZ		 760 queue_trunc = ceph_fill_file_size(inode, issued,
761 le32_to_cpu(info->truncate_seq),
762 le64_to_cpu(info->truncate_size),
763 le64_to_cpu(info->size));
764 /* only update max_size on auth cap */
765 if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
766 ci->i_max_size != le64_to_cpu(info->max_size)) {
767 dout("max_size %lld -> %llu\n", ci->i_max_size,
768 le64_to_cpu(info->max_size));
769 ci->i_max_size = le64_to_cpu(info->max_size);
770 }
771 }
355da1eb
SW		 772
773 /* xattrs */
774 /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
508b32d8 775 if ((ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL)) &&
355da1eb
SW		 776 le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) {
777 if (ci->i_xattrs.blob)
778 ceph_buffer_put(ci->i_xattrs.blob);
779 ci->i_xattrs.blob = xattr_blob;
780 if (xattr_blob)
781 memcpy(ci->i_xattrs.blob->vec.iov_base,
782 iinfo->xattr_data, iinfo->xattr_len);
783 ci->i_xattrs.version = le64_to_cpu(info->xattr_version);
7221fe4c 784 ceph_forget_all_cached_acls(inode);
a6424e48 785 xattr_blob = NULL;
355da1eb
SW		 786 }
787
788 inode->i_mapping->a_ops = &ceph_aops;
355da1eb
SW		 789
790 switch (inode->i_mode & S_IFMT) {
791 case S_IFIFO:
792 case S_IFBLK:
793 case S_IFCHR:
794 case S_IFSOCK:
795 init_special_inode(inode, inode->i_mode, inode->i_rdev);
796 inode->i_op = &ceph_file_iops;
797 break;
798 case S_IFREG:
799 inode->i_op = &ceph_file_iops;
800 inode->i_fop = &ceph_file_fops;
801 break;
802 case S_IFLNK:
803 inode->i_op = &ceph_symlink_iops;
804 if (!ci->i_symlink) {
810339ec 805 u32 symlen = iinfo->symlink_len;
355da1eb
SW		 806 char *sym;
807
be655596 808 spin_unlock(&ci->i_ceph_lock);
355da1eb 809
810339ec
XW		 810 err = -EINVAL;
811 if (WARN_ON(symlen != inode->i_size))
812 goto out;
813
355da1eb 814 err = -ENOMEM;
810339ec 815 sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS);
355da1eb
SW		 816 if (!sym)
817 goto out;
355da1eb 818
be655596 819 spin_lock(&ci->i_ceph_lock);
355da1eb
SW		 820 if (!ci->i_symlink)
821 ci->i_symlink = sym;
822 else
823 kfree(sym); /* lost a race */
824 }
825 break;
826 case S_IFDIR:
827 inode->i_op = &ceph_dir_iops;
828 inode->i_fop = &ceph_dir_fops;
829
14303d20
SW		 830 ci->i_dir_layout = iinfo->dir_layout;
831
355da1eb
SW		 832 ci->i_files = le64_to_cpu(info->files);
833 ci->i_subdirs = le64_to_cpu(info->subdirs);
834 ci->i_rbytes = le64_to_cpu(info->rbytes);
835 ci->i_rfiles = le64_to_cpu(info->rfiles);
836 ci->i_rsubdirs = le64_to_cpu(info->rsubdirs);
837 ceph_decode_timespec(&ci->i_rctime, &info->rctime);
355da1eb
SW		 838 break;
839 default:
840 pr_err("fill_inode %llx.%llx BAD mode 0%o\n",
841 ceph_vinop(inode), inode->i_mode);
842 }
843
355da1eb
SW		 844 /* were we issued a capability? */
845 if (info->cap.caps) {
846 if (ceph_snap(inode) == CEPH_NOSNAP) {
2f92b3d0 847 unsigned caps = le32_to_cpu(info->cap.caps);
355da1eb
SW		 848 ceph_add_cap(inode, session,
849 le64_to_cpu(info->cap.cap_id),
2f92b3d0 850 cap_fmode, caps,
355da1eb
SW		 851 le32_to_cpu(info->cap.wanted),
852 le32_to_cpu(info->cap.seq),
853 le32_to_cpu(info->cap.mseq),
854 le64_to_cpu(info->cap.realm),
d9df2783 855 info->cap.flags, &new_cap);
2f92b3d0
YZ		 856
857 /* set dir completion flag? */
858 if (S_ISDIR(inode->i_mode) &&
859 ci->i_files == 0 && ci->i_subdirs == 0 &&
860 (caps & CEPH_CAP_FILE_SHARED) &&
861 (issued & CEPH_CAP_FILE_EXCL) == 0 &&
862 !__ceph_dir_is_complete(ci)) {
863 dout(" marking %p complete (empty)\n", inode);
864 __ceph_dir_set_complete(ci,
865 atomic_read(&ci->i_release_count),
866 ci->i_ordered_count);
867 }
868
d9df2783 869 wake = true;
355da1eb 870 } else {
355da1eb
SW		 871 dout(" %p got snap_caps %s\n", inode,
872 ceph_cap_string(le32_to_cpu(info->cap.caps)));
873 ci->i_snap_caps |= le32_to_cpu(info->cap.caps);
874 if (cap_fmode >= 0)
875 __ceph_get_fmode(ci, cap_fmode);
355da1eb 876 }
04d000eb 877 } else if (cap_fmode >= 0) {
f3ae1b97 878 pr_warn("mds issued no caps on %llx.%llx\n",
04d000eb
SW		 879 ceph_vinop(inode));
880 __ceph_get_fmode(ci, cap_fmode);
355da1eb 881 }
31c542a1
YZ		 882
883 if (iinfo->inline_version > 0 &&
884 iinfo->inline_version >= ci->i_inline_version) {
885 int cache_caps = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
886 ci->i_inline_version = iinfo->inline_version;
887 if (ci->i_inline_version != CEPH_INLINE_NONE &&
01deead0
YZ		 888 (locked_page ||
889 (le32_to_cpu(info->cap.caps) & cache_caps)))
31c542a1
YZ		 890 fill_inline = true;
891 }
892
be655596 893 spin_unlock(&ci->i_ceph_lock);
355da1eb 894
31c542a1 895 if (fill_inline)
01deead0 896 ceph_fill_inline_data(inode, locked_page,
31c542a1
YZ		 897 iinfo->inline_data, iinfo->inline_len);
898
d9df2783
YZ		 899 if (wake)
900 wake_up_all(&ci->i_cap_wq);
901
355da1eb
SW		 902 /* queue truncate if we saw i_size decrease */
903 if (queue_trunc)
3c6f6b79 904 ceph_queue_vmtruncate(inode);
355da1eb
SW		 905
906 /* populate frag tree */
3e7fbe9c
YZ		 907 if (S_ISDIR(inode->i_mode))
908 ceph_fill_fragtree(inode, &info->fragtree, dirinfo);
355da1eb
SW		 909
910 /* update delegation info? */
911 if (dirinfo)
912 ceph_fill_dirfrag(inode, dirinfo);
913
914 err = 0;
355da1eb 915out:
d9df2783
YZ		 916 if (new_cap)
917 ceph_put_cap(mdsc, new_cap);
b6c1d5b8
SW		 918 if (xattr_blob)
919 ceph_buffer_put(xattr_blob);
355da1eb
SW		 920 return err;
921}
922
923/*
924 * caller should hold session s_mutex.
925 */
926static void update_dentry_lease(struct dentry *dentry,
927 struct ceph_mds_reply_lease *lease,
928 struct ceph_mds_session *session,
929 unsigned long from_time)
930{
931 struct ceph_dentry_info *di = ceph_dentry(dentry);
932 long unsigned duration = le32_to_cpu(lease->duration_ms);
933 long unsigned ttl = from_time + (duration * HZ) / 1000;
934 long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000;
935 struct inode *dir;
936
937 /* only track leases on regular dentries */
938 if (dentry->d_op != &ceph_dentry_ops)
939 return;
940
941 spin_lock(&dentry->d_lock);
2f90b852
SW		 942 dout("update_dentry_lease %p duration %lu ms ttl %lu\n",
943 dentry, duration, ttl);
355da1eb
SW		 944
945 /* make lease_rdcache_gen match directory */
2b0143b5 946 dir = d_inode(dentry->d_parent);
355da1eb
SW		 947 di->lease_shared_gen = ceph_inode(dir)->i_shared_gen;
948
2f90b852 949 if (duration == 0)
355da1eb
SW		 950 goto out_unlock;
951
952 if (di->lease_gen == session->s_cap_gen &&
953 time_before(ttl, dentry->d_time))
954 goto out_unlock; /* we already have a newer lease. */
955
956 if (di->lease_session && di->lease_session != session)
957 goto out_unlock;
958
959 ceph_dentry_lru_touch(dentry);
960
961 if (!di->lease_session)
962 di->lease_session = ceph_get_mds_session(session);
963 di->lease_gen = session->s_cap_gen;
964 di->lease_seq = le32_to_cpu(lease->seq);
965 di->lease_renew_after = half_ttl;
966 di->lease_renew_from = 0;
967 dentry->d_time = ttl;
968out_unlock:
969 spin_unlock(&dentry->d_lock);
970 return;
971}
972
973/*
974 * splice a dentry to an inode.
975 * caller must hold directory i_mutex for this to be safe.
976 *
977 * we will only rehash the resulting dentry if @prehash is
978 * true; @prehash will be set to false (for the benefit of
979 * the caller) if we fail.
980 */
981static struct dentry *splice_dentry(struct dentry *dn, struct inode *in,
0a8a70f9 982 bool *prehash)
355da1eb
SW		 983{
984 struct dentry *realdn;
985
2b0143b5 986 BUG_ON(d_inode(dn));
1cd3935b 987
355da1eb
SW		 988 /* dn must be unhashed */
989 if (!d_unhashed(dn))
990 d_drop(dn);
41d28bca 991 realdn = d_splice_alias(in, dn);
355da1eb 992 if (IS_ERR(realdn)) {
d69ed05a
SW		 993 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
994 PTR_ERR(realdn), dn, in, ceph_vinop(in));
355da1eb
SW		 995 if (prehash)
996 *prehash = false; /* don't rehash on error */
997 dn = realdn; /* note realdn contains the error */
998 goto out;
999 } else if (realdn) {
1000 dout("dn %p (%d) spliced with %p (%d) "
1001 "inode %p ino %llx.%llx\n",
84d08fa8
AV		 1002 dn, d_count(dn),
1003 realdn, d_count(realdn),
2b0143b5 1004 d_inode(realdn), ceph_vinop(d_inode(realdn)));
355da1eb
SW		 1005 dput(dn);
1006 dn = realdn;
1007 } else {
1008 BUG_ON(!ceph_dentry(dn));
355da1eb 1009 dout("dn %p attached to %p ino %llx.%llx\n",
2b0143b5 1010 dn, d_inode(dn), ceph_vinop(d_inode(dn)));
355da1eb
SW		 1011 }
1012 if ((!prehash || *prehash) && d_unhashed(dn))
1013 d_rehash(dn);
1014out:
1015 return dn;
1016}
1017
1018/*
1019 * Incorporate results into the local cache. This is either just
1020 * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
1021 * after a lookup).
1022 *
1023 * A reply may contain
1024 * a directory inode along with a dentry.
1025 * and/or a target inode
1026 *
1027 * Called with snap_rwsem (read).
1028 */
1029int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req,
1030 struct ceph_mds_session *session)
1031{
1032 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1033 struct inode *in = NULL;
355da1eb 1034 struct ceph_vino vino;
3d14c5d2 1035 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
355da1eb
SW		 1036 int err = 0;
1037
1038 dout("fill_trace %p is_dentry %d is_target %d\n", req,
1039 rinfo->head->is_dentry, rinfo->head->is_target);
1040
1041#if 0
1042 /*
1043 * Debugging hook:
1044 *
1045 * If we resend completed ops to a recovering mds, we get no
1046 * trace. Since that is very rare, pretend this is the case
1047 * to ensure the 'no trace' handlers in the callers behave.
1048 *
1049 * Fill in inodes unconditionally to avoid breaking cap
1050 * invariants.
1051 */
1052 if (rinfo->head->op & CEPH_MDS_OP_WRITE) {
1053 pr_info("fill_trace faking empty trace on %lld %s\n",
1054 req->r_tid, ceph_mds_op_name(rinfo->head->op));
1055 if (rinfo->head->is_dentry) {
1056 rinfo->head->is_dentry = 0;
1057 err = fill_inode(req->r_locked_dir,
1058 &rinfo->diri, rinfo->dirfrag,
1059 session, req->r_request_started, -1);
1060 }
1061 if (rinfo->head->is_target) {
1062 rinfo->head->is_target = 0;
1063 ininfo = rinfo->targeti.in;
1064 vino.ino = le64_to_cpu(ininfo->ino);
1065 vino.snap = le64_to_cpu(ininfo->snapid);
1066 in = ceph_get_inode(sb, vino);
1067 err = fill_inode(in, &rinfo->targeti, NULL,
1068 session, req->r_request_started,
1069 req->r_fmode);
1070 iput(in);
1071 }
1072 }
1073#endif
1074
1075 if (!rinfo->head->is_target && !rinfo->head->is_dentry) {
1076 dout("fill_trace reply is empty!\n");
167c9e35
SW		 1077 if (rinfo->head->result == 0 && req->r_locked_dir)
1078 ceph_invalidate_dir_request(req);
355da1eb
SW		 1079 return 0;
1080 }
1081
1082 if (rinfo->head->is_dentry) {
5b1daecd
SW		 1083 struct inode *dir = req->r_locked_dir;
1084
6c5e50fa 1085 if (dir) {
01deead0
YZ		 1086 err = fill_inode(dir, NULL,
1087 &rinfo->diri, rinfo->dirfrag,
6c5e50fa
SW		 1088 session, req->r_request_started, -1,
1089 &req->r_caps_reservation);
1090 if (err < 0)
19913b4e 1091 goto done;
6c5e50fa
SW		 1092 } else {
1093 WARN_ON_ONCE(1);
1094 }
19913b4e
YZ		 1095
1096 if (dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME) {
1097 struct qstr dname;
1098 struct dentry *dn, *parent;
1099
1100 BUG_ON(!rinfo->head->is_target);
1101 BUG_ON(req->r_dentry);
1102
1103 parent = d_find_any_alias(dir);
1104 BUG_ON(!parent);
1105
1106 dname.name = rinfo->dname;
1107 dname.len = rinfo->dname_len;
1108 dname.hash = full_name_hash(dname.name, dname.len);
1109 vino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1110 vino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1111retry_lookup:
1112 dn = d_lookup(parent, &dname);
1113 dout("d_lookup on parent=%p name=%.*s got %p\n",
1114 parent, dname.len, dname.name, dn);
1115
1116 if (!dn) {
1117 dn = d_alloc(parent, &dname);
1118 dout("d_alloc %p '%.*s' = %p\n", parent,
1119 dname.len, dname.name, dn);
1120 if (dn == NULL) {
1121 dput(parent);
1122 err = -ENOMEM;
1123 goto done;
1124 }
1125 err = ceph_init_dentry(dn);
1126 if (err < 0) {
1127 dput(dn);
1128 dput(parent);
1129 goto done;
1130 }
2b0143b5
DH		 1131 } else if (d_really_is_positive(dn) &&
1132 (ceph_ino(d_inode(dn)) != vino.ino ||
1133 ceph_snap(d_inode(dn)) != vino.snap)) {
19913b4e 1134 dout(" dn %p points to wrong inode %p\n",
2b0143b5 1135 dn, d_inode(dn));
19913b4e
YZ		 1136 d_delete(dn);
1137 dput(dn);
1138 goto retry_lookup;
1139 }
1140
1141 req->r_dentry = dn;
1142 dput(parent);
1143 }
5b1daecd
SW		 1144 }
1145
86b58d13
YZ		 1146 if (rinfo->head->is_target) {
1147 vino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1148 vino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1149
1150 in = ceph_get_inode(sb, vino);
1151 if (IS_ERR(in)) {
1152 err = PTR_ERR(in);
1153 goto done;
1154 }
1155 req->r_target_inode = in;
1156
01deead0 1157 err = fill_inode(in, req->r_locked_page, &rinfo->targeti, NULL,
86b58d13 1158 session, req->r_request_started,
48193012 1159 (!req->r_aborted && rinfo->head->result == 0) ?
86b58d13
YZ		 1160 req->r_fmode : -1,
1161 &req->r_caps_reservation);
1162 if (err < 0) {
1163 pr_err("fill_inode badness %p %llx.%llx\n",
1164 in, ceph_vinop(in));
1165 goto done;
1166 }
1167 }
1168
9358c6d4
SW		 1169 /*
1170 * ignore null lease/binding on snapdir ENOENT, or else we
1171 * will have trouble splicing in the virtual snapdir later
1172 */
1173 if (rinfo->head->is_dentry && !req->r_aborted &&
6c5e50fa 1174 req->r_locked_dir &&
9358c6d4 1175 (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name,
3d14c5d2 1176 fsc->mount_options->snapdir_name,
9358c6d4 1177 req->r_dentry->d_name.len))) {
355da1eb
SW		 1178 /*
1179 * lookup link rename : null -> possibly existing inode
1180 * mknod symlink mkdir : null -> new inode
1181 * unlink : linked -> null
1182 */
1183 struct inode *dir = req->r_locked_dir;
1184 struct dentry *dn = req->r_dentry;
1185 bool have_dir_cap, have_lease;
1186
1187 BUG_ON(!dn);
1188 BUG_ON(!dir);
2b0143b5 1189 BUG_ON(d_inode(dn->d_parent) != dir);
355da1eb
SW		 1190 BUG_ON(ceph_ino(dir) !=
1191 le64_to_cpu(rinfo->diri.in->ino));
1192 BUG_ON(ceph_snap(dir) !=
1193 le64_to_cpu(rinfo->diri.in->snapid));
1194
355da1eb
SW		 1195 /* do we have a lease on the whole dir? */
1196 have_dir_cap =
1197 (le32_to_cpu(rinfo->diri.in->cap.caps) &
1198 CEPH_CAP_FILE_SHARED);
1199
1200 /* do we have a dn lease? */
1201 have_lease = have_dir_cap ||
2f90b852 1202 le32_to_cpu(rinfo->dlease->duration_ms);
355da1eb
SW		 1203 if (!have_lease)
1204 dout("fill_trace no dentry lease or dir cap\n");
1205
1206 /* rename? */
1207 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) {
0a8a70f9
YZ		 1208 struct inode *olddir = req->r_old_dentry_dir;
1209 BUG_ON(!olddir);
1210
a455589f 1211 dout(" src %p '%pd' dst %p '%pd'\n",
355da1eb 1212 req->r_old_dentry,
a455589f
AV		 1213 req->r_old_dentry,
1214 dn, dn);
355da1eb
SW		 1215 dout("fill_trace doing d_move %p -> %p\n",
1216 req->r_old_dentry, dn);
c10f5e12 1217
355da1eb 1218 d_move(req->r_old_dentry, dn);
a455589f
AV		 1219 dout(" src %p '%pd' dst %p '%pd'\n",
1220 req->r_old_dentry,
355da1eb 1221 req->r_old_dentry,
a455589f 1222 dn, dn);
81a6cf2d 1223
c4a29f26
SW		 1224 /* ensure target dentry is invalidated, despite
1225 rehashing bug in vfs_rename_dir */
81a6cf2d
SW		 1226 ceph_invalidate_dentry_lease(dn);
1227
0a8a70f9 1228 /* d_move screws up sibling dentries' offsets */
70db4f36
YZ		 1229 ceph_dir_clear_ordered(dir);
1230 ceph_dir_clear_ordered(olddir);
0a8a70f9 1231
99ccbd22 1232 dout("dn %p gets new offset %lld\n", req->r_old_dentry,
1cd3935b 1233 ceph_dentry(req->r_old_dentry)->offset);
81a6cf2d 1234
355da1eb 1235 dn = req->r_old_dentry; /* use old_dentry */
355da1eb
SW		 1236 }
1237
1238 /* null dentry? */
1239 if (!rinfo->head->is_target) {
1240 dout("fill_trace null dentry\n");
2b0143b5 1241 if (d_really_is_positive(dn)) {
70db4f36 1242 ceph_dir_clear_ordered(dir);
355da1eb
SW		 1243 dout("d_delete %p\n", dn);
1244 d_delete(dn);
1245 } else {
1246 dout("d_instantiate %p NULL\n", dn);
1247 d_instantiate(dn, NULL);
1248 if (have_lease && d_unhashed(dn))
1249 d_rehash(dn);
1250 update_dentry_lease(dn, rinfo->dlease,
1251 session,
1252 req->r_request_started);
1253 }
1254 goto done;
1255 }
1256
1257 /* attach proper inode */
2b0143b5 1258 if (d_really_is_negative(dn)) {
70db4f36 1259 ceph_dir_clear_ordered(dir);
86b58d13 1260 ihold(in);
0a8a70f9 1261 dn = splice_dentry(dn, in, &have_lease);
355da1eb
SW		 1262 if (IS_ERR(dn)) {
1263 err = PTR_ERR(dn);
1264 goto done;
1265 }
1266 req->r_dentry = dn; /* may have spliced */
2b0143b5 1267 } else if (d_really_is_positive(dn) && d_inode(dn) != in) {
355da1eb 1268 dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
2b0143b5 1269 dn, d_inode(dn), ceph_vinop(d_inode(dn)),
86b58d13 1270 ceph_vinop(in));
355da1eb 1271 have_lease = false;
355da1eb
SW		 1272 }
1273
1274 if (have_lease)
1275 update_dentry_lease(dn, rinfo->dlease, session,
1276 req->r_request_started);
1277 dout(" final dn %p\n", dn);
86b58d13
YZ		 1278 } else if (!req->r_aborted &&
1279 (req->r_op == CEPH_MDS_OP_LOOKUPSNAP ||
1280 req->r_op == CEPH_MDS_OP_MKSNAP)) {
355da1eb 1281 struct dentry *dn = req->r_dentry;
0a8a70f9 1282 struct inode *dir = req->r_locked_dir;
355da1eb
SW		 1283
1284 /* fill out a snapdir LOOKUPSNAP dentry */
1285 BUG_ON(!dn);
0a8a70f9
YZ		 1286 BUG_ON(!dir);
1287 BUG_ON(ceph_snap(dir) != CEPH_SNAPDIR);
355da1eb 1288 dout(" linking snapped dir %p to dn %p\n", in, dn);
70db4f36 1289 ceph_dir_clear_ordered(dir);
86b58d13 1290 ihold(in);
0a8a70f9 1291 dn = splice_dentry(dn, in, NULL);
355da1eb
SW		 1292 if (IS_ERR(dn)) {
1293 err = PTR_ERR(dn);
1294 goto done;
1295 }
1296 req->r_dentry = dn; /* may have spliced */
355da1eb 1297 }
355da1eb
SW		 1298done:
1299 dout("fill_trace done err=%d\n", err);
1300 return err;
1301}
1302
1303/*
1304 * Prepopulate our cache with readdir results, leases, etc.
1305 */
79f9f99a
SW		 1306static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req,
1307 struct ceph_mds_session *session)
1308{
1309 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1310 int i, err = 0;
1311
1312 for (i = 0; i < rinfo->dir_nr; i++) {
1313 struct ceph_vino vino;
1314 struct inode *in;
1315 int rc;
1316
1317 vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino);
1318 vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid);
1319
1320 in = ceph_get_inode(req->r_dentry->d_sb, vino);
1321 if (IS_ERR(in)) {
1322 err = PTR_ERR(in);
1323 dout("new_inode badness got %d\n", err);
1324 continue;
1325 }
01deead0 1326 rc = fill_inode(in, NULL, &rinfo->dir_in[i], NULL, session,
79f9f99a
SW		 1327 req->r_request_started, -1,
1328 &req->r_caps_reservation);
1329 if (rc < 0) {
1330 pr_err("fill_inode badness on %p got %d\n", in, rc);
1331 err = rc;
1332 continue;
1333 }
1334 }
1335
1336 return err;
1337}
1338
355da1eb
SW		 1339int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1340 struct ceph_mds_session *session)
1341{
1342 struct dentry *parent = req->r_dentry;
1343 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1344 struct qstr dname;
1345 struct dentry *dn;
1346 struct inode *in;
86b58d13 1347 int err = 0, ret, i;
355da1eb
SW		 1348 struct inode *snapdir = NULL;
1349 struct ceph_mds_request_head *rhead = req->r_request->front.iov_base;
355da1eb 1350 struct ceph_dentry_info *di;
81c6aea5
YZ		 1351 u64 r_readdir_offset = req->r_readdir_offset;
1352 u32 frag = le32_to_cpu(rhead->args.readdir.frag);
1353
1354 if (rinfo->dir_dir &&
1355 le32_to_cpu(rinfo->dir_dir->frag) != frag) {
1356 dout("readdir_prepopulate got new frag %x -> %x\n",
1357 frag, le32_to_cpu(rinfo->dir_dir->frag));
1358 frag = le32_to_cpu(rinfo->dir_dir->frag);
1359 if (ceph_frag_is_leftmost(frag))
1360 r_readdir_offset = 2;
1361 else
1362 r_readdir_offset = 0;
1363 }
355da1eb 1364
79f9f99a
SW		 1365 if (req->r_aborted)
1366 return readdir_prepopulate_inodes_only(req, session);
355da1eb
SW		 1367
1368 if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
2b0143b5 1369 snapdir = ceph_get_snapdir(d_inode(parent));
355da1eb
SW		 1370 parent = d_find_alias(snapdir);
1371 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1372 rinfo->dir_nr, parent);
1373 } else {
1374 dout("readdir_prepopulate %d items under dn %p\n",
1375 rinfo->dir_nr, parent);
1376 if (rinfo->dir_dir)
2b0143b5 1377 ceph_fill_dirfrag(d_inode(parent), rinfo->dir_dir);
355da1eb
SW		 1378 }
1379
86b58d13 1380 /* FIXME: release caps/leases if error occurs */
355da1eb
SW		 1381 for (i = 0; i < rinfo->dir_nr; i++) {
1382 struct ceph_vino vino;
1383
1384 dname.name = rinfo->dir_dname[i];
1385 dname.len = rinfo->dir_dname_len[i];
1386 dname.hash = full_name_hash(dname.name, dname.len);
1387
1388 vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino);
1389 vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid);
1390
1391retry_lookup:
1392 dn = d_lookup(parent, &dname);
1393 dout("d_lookup on parent=%p name=%.*s got %p\n",
1394 parent, dname.len, dname.name, dn);
1395
1396 if (!dn) {
1397 dn = d_alloc(parent, &dname);
1398 dout("d_alloc %p '%.*s' = %p\n", parent,
1399 dname.len, dname.name, dn);
1400 if (dn == NULL) {
1401 dout("d_alloc badness\n");
1402 err = -ENOMEM;
1403 goto out;
1404 }
86b58d13
YZ		 1405 ret = ceph_init_dentry(dn);
1406 if (ret < 0) {
8c696737 1407 dput(dn);
86b58d13 1408 err = ret;
355da1eb 1409 goto out;
8c696737 1410 }
2b0143b5
DH		 1411 } else if (d_really_is_positive(dn) &&
1412 (ceph_ino(d_inode(dn)) != vino.ino ||
1413 ceph_snap(d_inode(dn)) != vino.snap)) {
355da1eb 1414 dout(" dn %p points to wrong inode %p\n",
2b0143b5 1415 dn, d_inode(dn));
355da1eb
SW		 1416 d_delete(dn);
1417 dput(dn);
1418 goto retry_lookup;
1419 } else {
1420 /* reorder parent's d_subdirs */
2fd6b7f5
NP		 1421 spin_lock(&parent->d_lock);
1422 spin_lock_nested(&dn->d_lock, DENTRY_D_LOCK_NESTED);
946e51f2 1423 list_move(&dn->d_child, &parent->d_subdirs);
355da1eb 1424 spin_unlock(&dn->d_lock);
2fd6b7f5 1425 spin_unlock(&parent->d_lock);
355da1eb
SW		 1426 }
1427
355da1eb 1428 /* inode */
2b0143b5
DH		 1429 if (d_really_is_positive(dn)) {
1430 in = d_inode(dn);
355da1eb
SW		 1431 } else {
1432 in = ceph_get_inode(parent->d_sb, vino);
ac1f12ef 1433 if (IS_ERR(in)) {
355da1eb 1434 dout("new_inode badness\n");
2744c171 1435 d_drop(dn);
355da1eb 1436 dput(dn);
ac1f12ef 1437 err = PTR_ERR(in);
355da1eb
SW		 1438 goto out;
1439 }
355da1eb
SW		 1440 }
1441
01deead0 1442 if (fill_inode(in, NULL, &rinfo->dir_in[i], NULL, session,
355da1eb
SW		 1443 req->r_request_started, -1,
1444 &req->r_caps_reservation) < 0) {
1445 pr_err("fill_inode badness on %p\n", in);
2b0143b5 1446 if (d_really_is_negative(dn))
86b58d13
YZ		 1447 iput(in);
1448 d_drop(dn);
d69ed05a 1449 goto next_item;
355da1eb 1450 }
86b58d13 1451
2b0143b5 1452 if (d_really_is_negative(dn)) {
5cba372c
YZ		 1453 struct dentry *realdn = splice_dentry(dn, in, NULL);
1454 if (IS_ERR(realdn)) {
1455 err = PTR_ERR(realdn);
1456 d_drop(dn);
86b58d13
YZ		 1457 dn = NULL;
1458 goto next_item;
1459 }
5cba372c 1460 dn = realdn;
86b58d13
YZ		 1461 }
1462
1463 di = dn->d_fsdata;
1464 di->offset = ceph_make_fpos(frag, i + r_readdir_offset);
1465
1466 update_dentry_lease(dn, rinfo->dir_dlease[i],
1467 req->r_session,
1468 req->r_request_started);
d69ed05a
SW		 1469next_item:
1470 if (dn)
1471 dput(dn);
355da1eb 1472 }
86b58d13
YZ		 1473 if (err == 0)
1474 req->r_did_prepopulate = true;
355da1eb
SW		 1475
1476out:
1477 if (snapdir) {
1478 iput(snapdir);
1479 dput(parent);
1480 }
1481 dout("readdir_prepopulate done\n");
1482 return err;
1483}
1484
1485int ceph_inode_set_size(struct inode *inode, loff_t size)
1486{
1487 struct ceph_inode_info *ci = ceph_inode(inode);
1488 int ret = 0;
1489
be655596 1490 spin_lock(&ci->i_ceph_lock);
355da1eb
SW		 1491 dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size);
1492 inode->i_size = size;
1493 inode->i_blocks = (size + (1 << 9) - 1) >> 9;
1494
1495 /* tell the MDS if we are approaching max_size */
1496 if ((size << 1) >= ci->i_max_size &&
1497 (ci->i_reported_size << 1) < ci->i_max_size)
1498 ret = 1;
1499
be655596 1500 spin_unlock(&ci->i_ceph_lock);
355da1eb
SW		 1501 return ret;
1502}
1503
1504/*
1505 * Write back inode data in a worker thread. (This can't be done
1506 * in the message handler context.)
1507 */
3c6f6b79
SW		 1508void ceph_queue_writeback(struct inode *inode)
1509{
15a2015f 1510 ihold(inode);
3c6f6b79
SW		 1511 if (queue_work(ceph_inode_to_client(inode)->wb_wq,
1512 &ceph_inode(inode)->i_wb_work)) {
2c27c9a5 1513 dout("ceph_queue_writeback %p\n", inode);
3c6f6b79 1514 } else {
2c27c9a5 1515 dout("ceph_queue_writeback %p failed\n", inode);
15a2015f 1516 iput(inode);
3c6f6b79
SW		 1517 }
1518}
1519
1520static void ceph_writeback_work(struct work_struct *work)
355da1eb
SW		 1521{
1522 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1523 i_wb_work);
1524 struct inode *inode = &ci->vfs_inode;
1525
1526 dout("writeback %p\n", inode);
1527 filemap_fdatawrite(&inode->i_data);
1528 iput(inode);
1529}
1530
3c6f6b79
SW		 1531/*
1532 * queue an async invalidation
1533 */
1534void ceph_queue_invalidate(struct inode *inode)
1535{
15a2015f 1536 ihold(inode);
3c6f6b79
SW		 1537 if (queue_work(ceph_inode_to_client(inode)->pg_inv_wq,
1538 &ceph_inode(inode)->i_pg_inv_work)) {
1539 dout("ceph_queue_invalidate %p\n", inode);
3c6f6b79
SW		 1540 } else {
1541 dout("ceph_queue_invalidate %p failed\n", inode);
15a2015f 1542 iput(inode);
3c6f6b79
SW		 1543 }
1544}
1545
355da1eb
SW		 1546/*
1547 * Invalidate inode pages in a worker thread. (This can't be done
1548 * in the message handler context.)
1549 */
3c6f6b79 1550static void ceph_invalidate_work(struct work_struct *work)
355da1eb
SW		 1551{
1552 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1553 i_pg_inv_work);
1554 struct inode *inode = &ci->vfs_inode;
1555 u32 orig_gen;
1556 int check = 0;
1557
b0d7c223 1558 mutex_lock(&ci->i_truncate_mutex);
be655596 1559 spin_lock(&ci->i_ceph_lock);
355da1eb
SW		 1560 dout("invalidate_pages %p gen %d revoking %d\n", inode,
1561 ci->i_rdcache_gen, ci->i_rdcache_revoking);
cd045cb4 1562 if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
9563f88c
YZ		 1563 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1564 check = 1;
be655596 1565 spin_unlock(&ci->i_ceph_lock);
b0d7c223 1566 mutex_unlock(&ci->i_truncate_mutex);
355da1eb
SW		 1567 goto out;
1568 }
1569 orig_gen = ci->i_rdcache_gen;
be655596 1570 spin_unlock(&ci->i_ceph_lock);
355da1eb 1571
4e217b5d 1572 truncate_pagecache(inode, 0);
355da1eb 1573
be655596 1574 spin_lock(&ci->i_ceph_lock);
cd045cb4
SW		 1575 if (orig_gen == ci->i_rdcache_gen &&
1576 orig_gen == ci->i_rdcache_revoking) {
355da1eb
SW		 1577 dout("invalidate_pages %p gen %d successful\n", inode,
1578 ci->i_rdcache_gen);
cd045cb4 1579 ci->i_rdcache_revoking--;
355da1eb
SW		 1580 check = 1;
1581 } else {
cd045cb4
SW		 1582 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1583 inode, orig_gen, ci->i_rdcache_gen,
1584 ci->i_rdcache_revoking);
9563f88c
YZ		 1585 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1586 check = 1;
355da1eb 1587 }
be655596 1588 spin_unlock(&ci->i_ceph_lock);
b0d7c223 1589 mutex_unlock(&ci->i_truncate_mutex);
9563f88c 1590out:
355da1eb
SW		 1591 if (check)
1592 ceph_check_caps(ci, 0, NULL);
355da1eb
SW		 1593 iput(inode);
1594}
1595
1596
1597/*
3f99969f 1598 * called by trunc_wq;
355da1eb
SW		 1599 *
1600 * We also truncate in a separate thread as well.
1601 */
3c6f6b79 1602static void ceph_vmtruncate_work(struct work_struct *work)
355da1eb
SW		 1603{
1604 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1605 i_vmtruncate_work);
1606 struct inode *inode = &ci->vfs_inode;
1607
1608 dout("vmtruncate_work %p\n", inode);
b415bf4f 1609 __ceph_do_pending_vmtruncate(inode);
355da1eb
SW		 1610 iput(inode);
1611}
1612
3c6f6b79
SW		 1613/*
1614 * Queue an async vmtruncate. If we fail to queue work, we will handle
1615 * the truncation the next time we call __ceph_do_pending_vmtruncate.
1616 */
1617void ceph_queue_vmtruncate(struct inode *inode)
1618{
1619 struct ceph_inode_info *ci = ceph_inode(inode);
1620
15a2015f 1621 ihold(inode);
99ccbd22 1622
640ef79d 1623 if (queue_work(ceph_sb_to_client(inode->i_sb)->trunc_wq,
3c6f6b79
SW		 1624 &ci->i_vmtruncate_work)) {
1625 dout("ceph_queue_vmtruncate %p\n", inode);
3c6f6b79
SW		 1626 } else {
1627 dout("ceph_queue_vmtruncate %p failed, pending=%d\n",
1628 inode, ci->i_truncate_pending);
15a2015f 1629 iput(inode);
3c6f6b79
SW		 1630 }
1631}
1632
355da1eb 1633/*
355da1eb
SW		 1634 * Make sure any pending truncation is applied before doing anything
1635 * that may depend on it.
1636 */
b415bf4f 1637void __ceph_do_pending_vmtruncate(struct inode *inode)
355da1eb
SW		 1638{
1639 struct ceph_inode_info *ci = ceph_inode(inode);
1640 u64 to;
a85f50b6 1641 int wrbuffer_refs, finish = 0;
355da1eb 1642
b0d7c223 1643 mutex_lock(&ci->i_truncate_mutex);
355da1eb 1644retry:
be655596 1645 spin_lock(&ci->i_ceph_lock);
355da1eb
SW		 1646 if (ci->i_truncate_pending == 0) {
1647 dout("__do_pending_vmtruncate %p none pending\n", inode);
be655596 1648 spin_unlock(&ci->i_ceph_lock);
b0d7c223 1649 mutex_unlock(&ci->i_truncate_mutex);
355da1eb
SW		 1650 return;
1651 }
1652
1653 /*
1654 * make sure any dirty snapped pages are flushed before we
1655 * possibly truncate them.. so write AND block!
1656 */
1657 if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) {
1658 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1659 inode);
be655596 1660 spin_unlock(&ci->i_ceph_lock);
355da1eb
SW		 1661 filemap_write_and_wait_range(&inode->i_data, 0,
1662 inode->i_sb->s_maxbytes);
1663 goto retry;
1664 }
1665
b0d7c223
YZ		 1666 /* there should be no reader or writer */
1667 WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref);
1668
355da1eb
SW		 1669 to = ci->i_truncate_size;
1670 wrbuffer_refs = ci->i_wrbuffer_ref;
1671 dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode,
1672 ci->i_truncate_pending, to);
be655596 1673 spin_unlock(&ci->i_ceph_lock);
355da1eb 1674
4e217b5d 1675 truncate_pagecache(inode, to);
355da1eb 1676
be655596 1677 spin_lock(&ci->i_ceph_lock);
a85f50b6
YZ		 1678 if (to == ci->i_truncate_size) {
1679 ci->i_truncate_pending = 0;
1680 finish = 1;
1681 }
be655596 1682 spin_unlock(&ci->i_ceph_lock);
a85f50b6
YZ		 1683 if (!finish)
1684 goto retry;
355da1eb 1685
b0d7c223
YZ		 1686 mutex_unlock(&ci->i_truncate_mutex);
1687
355da1eb
SW		 1688 if (wrbuffer_refs == 0)
1689 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
a85f50b6
YZ		 1690
1691 wake_up_all(&ci->i_cap_wq);
355da1eb
SW		 1692}
1693
355da1eb
SW		 1694/*
1695 * symlinks
1696 */
1697static void *ceph_sym_follow_link(struct dentry *dentry, struct nameidata *nd)
1698{
2b0143b5 1699 struct ceph_inode_info *ci = ceph_inode(d_inode(dentry));
355da1eb
SW		 1700 nd_set_link(nd, ci->i_symlink);
1701 return NULL;
1702}
1703
1704static const struct inode_operations ceph_symlink_iops = {
1705 .readlink = generic_readlink,
1706 .follow_link = ceph_sym_follow_link,
0b932672
YZ		 1707 .setattr = ceph_setattr,
1708 .getattr = ceph_getattr,
1709 .setxattr = ceph_setxattr,
1710 .getxattr = ceph_getxattr,
1711 .listxattr = ceph_listxattr,
1712 .removexattr = ceph_removexattr,
355da1eb
SW		 1713};
1714
1715/*
1716 * setattr
1717 */
1718int ceph_setattr(struct dentry *dentry, struct iattr *attr)
1719{
2b0143b5 1720 struct inode *inode = d_inode(dentry);
355da1eb 1721 struct ceph_inode_info *ci = ceph_inode(inode);
355da1eb
SW		 1722 const unsigned int ia_valid = attr->ia_valid;
1723 struct ceph_mds_request *req;
3d14c5d2 1724 struct ceph_mds_client *mdsc = ceph_sb_to_client(dentry->d_sb)->mdsc;
355da1eb
SW		 1725 int issued;
1726 int release = 0, dirtied = 0;
1727 int mask = 0;
1728 int err = 0;
fca65b4a 1729 int inode_dirty_flags = 0;
355da1eb
SW		 1730
1731 if (ceph_snap(inode) != CEPH_NOSNAP)
1732 return -EROFS;
1733
355da1eb
SW		 1734 err = inode_change_ok(inode, attr);
1735 if (err != 0)
1736 return err;
1737
1738 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR,
1739 USE_AUTH_MDS);
1740 if (IS_ERR(req))
1741 return PTR_ERR(req);
1742
be655596 1743 spin_lock(&ci->i_ceph_lock);
355da1eb
SW		 1744 issued = __ceph_caps_issued(ci, NULL);
1745 dout("setattr %p issued %s\n", inode, ceph_cap_string(issued));
1746
1747 if (ia_valid & ATTR_UID) {
1748 dout("setattr %p uid %d -> %d\n", inode,
bd2bae6a
EB		 1749 from_kuid(&init_user_ns, inode->i_uid),
1750 from_kuid(&init_user_ns, attr->ia_uid));
355da1eb
SW		 1751 if (issued & CEPH_CAP_AUTH_EXCL) {
1752 inode->i_uid = attr->ia_uid;
1753 dirtied |= CEPH_CAP_AUTH_EXCL;
1754 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
ab871b90
EB		 1755 !uid_eq(attr->ia_uid, inode->i_uid)) {
1756 req->r_args.setattr.uid = cpu_to_le32(
1757 from_kuid(&init_user_ns, attr->ia_uid));
355da1eb
SW		 1758 mask |= CEPH_SETATTR_UID;
1759 release |= CEPH_CAP_AUTH_SHARED;
1760 }
1761 }
1762 if (ia_valid & ATTR_GID) {
1763 dout("setattr %p gid %d -> %d\n", inode,
bd2bae6a
EB		 1764 from_kgid(&init_user_ns, inode->i_gid),
1765 from_kgid(&init_user_ns, attr->ia_gid));
355da1eb
SW		 1766 if (issued & CEPH_CAP_AUTH_EXCL) {
1767 inode->i_gid = attr->ia_gid;
1768 dirtied |= CEPH_CAP_AUTH_EXCL;
1769 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
ab871b90
EB		 1770 !gid_eq(attr->ia_gid, inode->i_gid)) {
1771 req->r_args.setattr.gid = cpu_to_le32(
1772 from_kgid(&init_user_ns, attr->ia_gid));
355da1eb
SW		 1773 mask |= CEPH_SETATTR_GID;
1774 release |= CEPH_CAP_AUTH_SHARED;
1775 }
1776 }
1777 if (ia_valid & ATTR_MODE) {
1778 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode,
1779 attr->ia_mode);
1780 if (issued & CEPH_CAP_AUTH_EXCL) {
1781 inode->i_mode = attr->ia_mode;
1782 dirtied |= CEPH_CAP_AUTH_EXCL;
1783 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1784 attr->ia_mode != inode->i_mode) {
7221fe4c 1785 inode->i_mode = attr->ia_mode;
355da1eb
SW		 1786 req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode);
1787 mask |= CEPH_SETATTR_MODE;
1788 release |= CEPH_CAP_AUTH_SHARED;
1789 }
1790 }
1791
1792 if (ia_valid & ATTR_ATIME) {
1793 dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode,
1794 inode->i_atime.tv_sec, inode->i_atime.tv_nsec,
1795 attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec);
1796 if (issued & CEPH_CAP_FILE_EXCL) {
1797 ci->i_time_warp_seq++;
1798 inode->i_atime = attr->ia_atime;
1799 dirtied |= CEPH_CAP_FILE_EXCL;
1800 } else if ((issued & CEPH_CAP_FILE_WR) &&
1801 timespec_compare(&inode->i_atime,
1802 &attr->ia_atime) < 0) {
1803 inode->i_atime = attr->ia_atime;
1804 dirtied |= CEPH_CAP_FILE_WR;
1805 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1806 !timespec_equal(&inode->i_atime, &attr->ia_atime)) {
1807 ceph_encode_timespec(&req->r_args.setattr.atime,
1808 &attr->ia_atime);
1809 mask |= CEPH_SETATTR_ATIME;
1810 release |= CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_RD |
1811 CEPH_CAP_FILE_WR;
1812 }
1813 }
1814 if (ia_valid & ATTR_MTIME) {
1815 dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode,
1816 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
1817 attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec);
1818 if (issued & CEPH_CAP_FILE_EXCL) {
1819 ci->i_time_warp_seq++;
1820 inode->i_mtime = attr->ia_mtime;
1821 dirtied |= CEPH_CAP_FILE_EXCL;
1822 } else if ((issued & CEPH_CAP_FILE_WR) &&
1823 timespec_compare(&inode->i_mtime,
1824 &attr->ia_mtime) < 0) {
1825 inode->i_mtime = attr->ia_mtime;
1826 dirtied |= CEPH_CAP_FILE_WR;
1827 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1828 !timespec_equal(&inode->i_mtime, &attr->ia_mtime)) {
1829 ceph_encode_timespec(&req->r_args.setattr.mtime,
1830 &attr->ia_mtime);
1831 mask |= CEPH_SETATTR_MTIME;
1832 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
1833 CEPH_CAP_FILE_WR;
1834 }
1835 }
1836 if (ia_valid & ATTR_SIZE) {
1837 dout("setattr %p size %lld -> %lld\n", inode,
1838 inode->i_size, attr->ia_size);
355da1eb
SW		 1839 if ((issued & CEPH_CAP_FILE_EXCL) &&
1840 attr->ia_size > inode->i_size) {
1841 inode->i_size = attr->ia_size;
355da1eb
SW		 1842 inode->i_blocks =
1843 (attr->ia_size + (1 << 9) - 1) >> 9;
1844 inode->i_ctime = attr->ia_ctime;
1845 ci->i_reported_size = attr->ia_size;
1846 dirtied |= CEPH_CAP_FILE_EXCL;
1847 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1848 attr->ia_size != inode->i_size) {
1849 req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
1850 req->r_args.setattr.old_size =
1851 cpu_to_le64(inode->i_size);
1852 mask |= CEPH_SETATTR_SIZE;
1853 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
1854 CEPH_CAP_FILE_WR;
1855 }
1856 }
1857
1858 /* these do nothing */
1859 if (ia_valid & ATTR_CTIME) {
1860 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME|
1861 ATTR_MODE|ATTR_UID|ATTR_GID)) == 0;
1862 dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode,
1863 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
1864 attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec,
1865 only ? "ctime only" : "ignored");
1866 inode->i_ctime = attr->ia_ctime;
1867 if (only) {
1868 /*
1869 * if kernel wants to dirty ctime but nothing else,
1870 * we need to choose a cap to dirty under, or do
1871 * a almost-no-op setattr
1872 */
1873 if (issued & CEPH_CAP_AUTH_EXCL)
1874 dirtied |= CEPH_CAP_AUTH_EXCL;
1875 else if (issued & CEPH_CAP_FILE_EXCL)
1876 dirtied |= CEPH_CAP_FILE_EXCL;
1877 else if (issued & CEPH_CAP_XATTR_EXCL)
1878 dirtied |= CEPH_CAP_XATTR_EXCL;
1879 else
1880 mask |= CEPH_SETATTR_CTIME;
1881 }
1882 }
1883 if (ia_valid & ATTR_FILE)
1884 dout("setattr %p ATTR_FILE ... hrm!\n", inode);
1885
1886 if (dirtied) {
fca65b4a 1887 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied);
355da1eb
SW		 1888 inode->i_ctime = CURRENT_TIME;
1889 }
1890
1891 release &= issued;
be655596 1892 spin_unlock(&ci->i_ceph_lock);
355da1eb 1893
fca65b4a
SW		 1894 if (inode_dirty_flags)
1895 __mark_inode_dirty(inode, inode_dirty_flags);
1896
7221fe4c 1897 if (ia_valid & ATTR_MODE) {
4db658ea 1898 err = posix_acl_chmod(inode, attr->ia_mode);
7221fe4c
GZ		 1899 if (err)
1900 goto out_put;
1901 }
1902
355da1eb 1903 if (mask) {
70b666c3
SW		 1904 req->r_inode = inode;
1905 ihold(inode);
355da1eb
SW		 1906 req->r_inode_drop = release;
1907 req->r_args.setattr.mask = cpu_to_le32(mask);
1908 req->r_num_caps = 1;
752c8bdc 1909 err = ceph_mdsc_do_request(mdsc, NULL, req);
355da1eb
SW		 1910 }
1911 dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err,
1912 ceph_cap_string(dirtied), mask);
1913
1914 ceph_mdsc_put_request(req);
b0d7c223
YZ		 1915 if (mask & CEPH_SETATTR_SIZE)
1916 __ceph_do_pending_vmtruncate(inode);
355da1eb 1917 return err;
7221fe4c 1918out_put:
355da1eb
SW		 1919 ceph_mdsc_put_request(req);
1920 return err;
1921}
1922
1923/*
1924 * Verify that we have a lease on the given mask. If not,
1925 * do a getattr against an mds.
1926 */
01deead0
YZ		 1927int __ceph_do_getattr(struct inode *inode, struct page *locked_page,
1928 int mask, bool force)
355da1eb 1929{
3d14c5d2
YS		 1930 struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
1931 struct ceph_mds_client *mdsc = fsc->mdsc;
355da1eb
SW		 1932 struct ceph_mds_request *req;
1933 int err;
1934
1935 if (ceph_snap(inode) == CEPH_SNAPDIR) {
1936 dout("do_getattr inode %p SNAPDIR\n", inode);
1937 return 0;
1938 }
1939
01deead0
YZ		 1940 dout("do_getattr inode %p mask %s mode 0%o\n",
1941 inode, ceph_cap_string(mask), inode->i_mode);
508b32d8 1942 if (!force && ceph_caps_issued_mask(ceph_inode(inode), mask, 1))
355da1eb
SW		 1943 return 0;
1944
1945 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
1946 if (IS_ERR(req))
1947 return PTR_ERR(req);
70b666c3
SW		 1948 req->r_inode = inode;
1949 ihold(inode);
355da1eb
SW		 1950 req->r_num_caps = 1;
1951 req->r_args.getattr.mask = cpu_to_le32(mask);
01deead0 1952 req->r_locked_page = locked_page;
355da1eb 1953 err = ceph_mdsc_do_request(mdsc, NULL, req);
01deead0
YZ		 1954 if (locked_page && err == 0) {
1955 u64 inline_version = req->r_reply_info.targeti.inline_version;
1956 if (inline_version == 0) {
1957 /* the reply is supposed to contain inline data */
1958 err = -EINVAL;
1959 } else if (inline_version == CEPH_INLINE_NONE) {
1960 err = -ENODATA;
1961 } else {
1962 err = req->r_reply_info.targeti.inline_len;
1963 }
1964 }
355da1eb
SW		 1965 ceph_mdsc_put_request(req);
1966 dout("do_getattr result=%d\n", err);
1967 return err;
1968}
1969
1970
1971/*
1972 * Check inode permissions. We verify we have a valid value for
1973 * the AUTH cap, then call the generic handler.
1974 */
10556cb2 1975int ceph_permission(struct inode *inode, int mask)
355da1eb 1976{
b74c79e9
NP		 1977 int err;
1978
10556cb2 1979 if (mask & MAY_NOT_BLOCK)
b74c79e9
NP		 1980 return -ECHILD;
1981
508b32d8 1982 err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED, false);
355da1eb
SW		 1983
1984 if (!err)
2830ba7f 1985 err = generic_permission(inode, mask);
355da1eb
SW		 1986 return err;
1987}
1988
1989/*
1990 * Get all attributes. Hopefully somedata we'll have a statlite()
1991 * and can limit the fields we require to be accurate.
1992 */
1993int ceph_getattr(struct vfsmount *mnt, struct dentry *dentry,
1994 struct kstat *stat)
1995{
2b0143b5 1996 struct inode *inode = d_inode(dentry);
232d4b01 1997 struct ceph_inode_info *ci = ceph_inode(inode);
355da1eb
SW		 1998 int err;
1999
508b32d8 2000 err = ceph_do_getattr(inode, CEPH_STAT_CAP_INODE_ALL, false);
355da1eb
SW		 2001 if (!err) {
2002 generic_fillattr(inode, stat);
ad1fee96 2003 stat->ino = ceph_translate_ino(inode->i_sb, inode->i_ino);
355da1eb
SW		 2004 if (ceph_snap(inode) != CEPH_NOSNAP)
2005 stat->dev = ceph_snap(inode);
2006 else
2007 stat->dev = 0;
232d4b01 2008 if (S_ISDIR(inode->i_mode)) {
1c1266bb
YS		 2009 if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb),
2010 RBYTES))
2011 stat->size = ci->i_rbytes;
2012 else
2013 stat->size = ci->i_files + ci->i_subdirs;
232d4b01 2014 stat->blocks = 0;
355da1eb 2015 stat->blksize = 65536;
232d4b01 2016 }
355da1eb
SW		 2017 }
2018 return err;
2019}
Linux 6.x block layer and io_uring tree(s)