blob: f61a74115bebbe0c48f2b8d3edb7e85ff39d197f [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>
Linus Torvalds4db658e2014-01-28 18:06:18 -080012#include <linux/posix_acl.h>
Yan, Zheng3e7fbe92014-04-18 16:00:30 +080013#include <linux/random.h>
Sage Weil355da1e2009-10-06 11:31:08 -070014
15#include "super.h"
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -070016#include "mds_client.h"
Milosz Tanski99ccbd22013-08-21 17:29:54 -040017#include "cache.h"
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -070018#include <linux/ceph/decode.h>
Sage Weil355da1e2009-10-06 11:31:08 -070019
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
Sage Weil3c6f6b72010-02-09 15:24:44 -080034static 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);
Sage Weil355da1e2009-10-06 11:31:08 -070037
38/*
39 * find or create an inode, given the ceph ino number
40 */
Yehuda Sadehad1fee92011-01-21 16:44:03 -080041static 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
Sage Weil355da1e2009-10-06 11:31:08 -070048struct 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);
Sage Weilb377ff12009-11-11 15:22:37 -080077 struct ceph_inode_info *ci = ceph_inode(inode);
Sage Weil355da1e2009-10-06 11:31:08 -070078
79 BUG_ON(!S_ISDIR(parent->i_mode));
80 if (IS_ERR(inode))
Julia Lawall7e34bc52010-05-22 12:01:14 +020081 return inode;
Sage Weil355da1e2009-10-06 11:31:08 -070082 inode->i_mode = parent->i_mode;
83 inode->i_uid = parent->i_uid;
84 inode->i_gid = parent->i_gid;
85 inode->i_op = &ceph_dir_iops;
86 inode->i_fop = &ceph_dir_fops;
Sage Weilb377ff12009-11-11 15:22:37 -080087 ci->i_snap_caps = CEPH_CAP_PIN; /* so we can open */
88 ci->i_rbytes = 0;
Sage Weil355da1e2009-10-06 11:31:08 -070089 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,
Guangliang Zhao7221fe42013-11-11 15:18:03 +0800100 .get_acl = ceph_get_acl,
Sage Weil72466d02014-01-29 06:22:25 -0800101 .set_acl = ceph_set_acl,
Sage Weil355da1e2009-10-06 11:31:08 -0700102};
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 */
Yan, Zheng3e7fbe92014-04-18 16:00:30 +0800183static u32 __ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
184 struct ceph_inode_frag *pfrag, int *found)
Sage Weil355da1e2009-10-06 11:31:08 -0700185{
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
Sage Weil355da1e2009-10-06 11:31:08 -0700194 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
Sage Weil355da1e2009-10-06 11:31:08 -0700222 return t;
223}
224
Yan, Zheng3e7fbe92014-04-18 16:00:30 +0800225u32 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
Sage Weil355da1e2009-10-06 11:31:08 -0700235/*
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);
Yan, Zheng8d085032014-04-18 22:01:38 +0800248 int diri_auth = -1;
Sage Weil355da1e2009-10-06 11:31:08 -0700249 int i;
250 int err = 0;
251
Yan, Zheng8d085032014-04-18 22:01:38 +0800252 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
Sage Weil355da1e2009-10-06 11:31:08 -0700257 mutex_lock(&ci->i_fragtree_mutex);
Yan, Zheng8d085032014-04-18 22:01:38 +0800258 if (ndist == 0 && mds == diri_auth) {
Sage Weil355da1e2009-10-06 11:31:08 -0700259 /* 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
Yan, Zheng3e7fbe92014-04-18 16:00:30 +0800303static 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}
Sage Weil355da1e2009-10-06 11:31:08 -0700372
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
Sage Weilbe655592011-11-30 09:47:09 -0800387 spin_lock_init(&ci->i_ceph_lock);
388
Sage Weil355da1e2009-10-06 11:31:08 -0700389 ci->i_version = 0;
Yan, Zheng31c542a2014-11-14 21:41:55 +0800390 ci->i_inline_version = 0;
Sage Weil355da1e2009-10-06 11:31:08 -0700391 ci->i_time_warp_seq = 0;
392 ci->i_ceph_flags = 0;
Yan, Zheng70db4f32014-10-21 18:09:56 -0700393 ci->i_ordered_count = 0;
Yan, Zheng2f276c52013-03-13 19:44:32 +0800394 atomic_set(&ci->i_release_count, 1);
395 atomic_set(&ci->i_complete_count, 0);
Sage Weil355da1e2009-10-06 11:31:08 -0700396 ci->i_symlink = NULL;
397
Sage Weil6c0f3af2010-11-16 11:14:34 -0800398 memset(&ci->i_dir_layout, 0, sizeof(ci->i_dir_layout));
399
Sage Weil355da1e2009-10-06 11:31:08 -0700400 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);
Sage Weil355da1e2009-10-06 11:31:08 -0700426 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
Yan, Zhengb0d7c222013-08-12 21:42:15 -0700433 mutex_init(&ci->i_truncate_mutex);
Sage Weil355da1e2009-10-06 11:31:08 -0700434 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;
Henry C Changd3d07202011-05-11 10:29:54 +0000447 ci->i_wb_ref = 0;
Sage Weil355da1e2009-10-06 11:31:08 -0700448 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
Sage Weil3c6f6b72010-02-09 15:24:44 -0800462 INIT_WORK(&ci->i_wb_work, ceph_writeback_work);
463 INIT_WORK(&ci->i_pg_inv_work, ceph_invalidate_work);
Sage Weil355da1e2009-10-06 11:31:08 -0700464
465 INIT_WORK(&ci->i_vmtruncate_work, ceph_vmtruncate_work);
466
Milosz Tanski99ccbd22013-08-21 17:29:54 -0400467 ceph_fscache_inode_init(ci);
468
Sage Weil355da1e2009-10-06 11:31:08 -0700469 return &ci->vfs_inode;
470}
471
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100472static 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
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100477 kmem_cache_free(ceph_inode_cachep, ci);
478}
479
Sage Weil355da1e2009-10-06 11:31:08 -0700480void 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
Milosz Tanski99ccbd22013-08-21 17:29:54 -0400488 ceph_fscache_unregister_inode_cookie(ci);
489
Sage Weil355da1e2009-10-06 11:31:08 -0700490 ceph_queue_caps_release(inode);
491
Sage Weil8b218b82010-03-09 12:59:08 -0800492 /*
493 * we may still have a snap_realm reference if there are stray
Yan, Zhengd9df2782014-04-18 09:57:11 +0800494 * caps in i_snap_caps.
Sage Weil8b218b82010-03-09 12:59:08 -0800495 */
496 if (ci->i_snap_realm) {
497 struct ceph_mds_client *mdsc =
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700498 ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc;
Sage Weil8b218b82010-03-09 12:59:08 -0800499 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
Sage Weil355da1e2009-10-06 11:31:08 -0700508 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);
Sage Weilb6c1d5b2009-12-07 12:17:17 -0800516 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);
Sage Weil355da1e2009-10-06 11:31:08 -0700520
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100521 call_rcu(&inode->i_rcu, ceph_i_callback);
Sage Weil355da1e2009-10-06 11:31:08 -0700522}
523
Yan, Zheng9f12bd12013-09-20 19:55:31 +0800524int 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
Sage Weil355da1e2009-10-06 11:31:08 -0700534/*
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;
Yan, Zhengb0d7c222013-08-12 21:42:15 -0700558
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));
Yehuda Sadeh3d497d82010-02-09 11:08:40 -0800564 /*
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 */
Yan, Zhengb0d7c222013-08-12 21:42:15 -0700570 if ((issued & (CEPH_CAP_FILE_CACHE|
571 CEPH_CAP_FILE_BUFFER)) ||
Yehuda Sadeh3d497d82010-02-09 11:08:40 -0800572 mapping_mapped(inode->i_mapping) ||
573 __ceph_caps_file_wanted(ci)) {
Sage Weil355da1e2009-10-06 11:31:08 -0700574 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 }
Milosz Tanski99ccbd22013-08-21 17:29:54 -0400585
586 if (queue_trunc)
587 ceph_fscache_invalidate(inode);
588
Sage Weil355da1e2009-10-06 11:31:08 -0700589 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|
Sage Weild8672d62010-11-08 09:24:34 -0800601 CEPH_CAP_FILE_BUFFER|
602 CEPH_CAP_AUTH_EXCL|
603 CEPH_CAP_XATTR_EXCL)) {
Sage Weil355da1e2009-10-06 11:31:08 -0700604 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 {
Sage Weild8672d62010-11-08 09:24:34 -0800643 /* we have no write|excl caps; whatever the MDS says is true */
Sage Weil355da1e2009-10-06 11:31:08 -0700644 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 */
Yan, Zheng01deead2014-11-14 21:56:29 +0800662static int fill_inode(struct inode *inode, struct page *locked_page,
Sage Weil355da1e2009-10-06 11:31:08 -0700663 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{
Yan, Zhengd9df2782014-04-18 09:57:11 +0800669 struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
Sage Weil355da1e2009-10-06 11:31:08 -0700670 struct ceph_mds_reply_inode *info = iinfo->in;
671 struct ceph_inode_info *ci = ceph_inode(inode);
Yan, Zhengf98a1282014-04-17 08:55:50 +0800672 int issued = 0, implemented, new_issued;
Sage Weil355da1e2009-10-06 11:31:08 -0700673 struct timespec mtime, atime, ctime;
Sage Weil355da1e2009-10-06 11:31:08 -0700674 struct ceph_buffer *xattr_blob = NULL;
Yan, Zhengd9df2782014-04-18 09:57:11 +0800675 struct ceph_cap *new_cap = NULL;
Sage Weil355da1e2009-10-06 11:31:08 -0700676 int err = 0;
Yan, Zhengd9df2782014-04-18 09:57:11 +0800677 bool wake = false;
Yan, Zhengf98a1282014-04-17 08:55:50 +0800678 bool queue_trunc = false;
679 bool new_version = false;
Yan, Zheng31c542a2014-11-14 21:41:55 +0800680 bool fill_inline = false;
Sage Weil355da1e2009-10-06 11:31:08 -0700681
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
Yan, Zhengd9df2782014-04-18 09:57:11 +0800686 /* prealloc new cap struct */
687 if (info->cap.caps && ceph_snap(inode) == CEPH_NOSNAP)
688 new_cap = ceph_get_cap(mdsc, caps_reservation);
689
Sage Weil355da1e2009-10-06 11:31:08 -0700690 /*
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) {
Sage Weilb6c1d5b2009-12-07 12:17:17 -0800696 xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS);
Sage Weil355da1e2009-10-06 11:31:08 -0700697 if (!xattr_blob)
698 pr_err("fill_inode ENOMEM xattr blob %d bytes\n",
699 iinfo->xattr_len);
700 }
701
Sage Weilbe655592011-11-30 09:47:09 -0800702 spin_lock(&ci->i_ceph_lock);
Sage Weil355da1e2009-10-06 11:31:08 -0700703
704 /*
705 * provided version will be odd if inode value is projected,
Sage Weil8bd59e02010-11-08 09:23:12 -0800706 * 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
Sage Weil355da1e2009-10-06 11:31:08 -0700714 */
Yan, Zhengf98a1282014-04-17 08:55:50 +0800715 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
Sage Weil355da1e2009-10-06 11:31:08 -0700720 issued = __ceph_caps_issued(ci, &implemented);
721 issued |= implemented | __ceph_caps_dirty(ci);
Yan, Zhengf98a1282014-04-17 08:55:50 +0800722 new_issued = ~issued & le32_to_cpu(info->cap.caps);
Sage Weil355da1e2009-10-06 11:31:08 -0700723
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);
Yan, Zhengf98a1282014-04-17 08:55:50 +0800728 inode->i_blkbits = fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1;
Sage Weil355da1e2009-10-06 11:31:08 -0700729
Yan, Zhengf98a1282014-04-17 08:55:50 +0800730 if ((new_version || (new_issued & CEPH_CAP_AUTH_SHARED)) &&
731 (issued & CEPH_CAP_AUTH_EXCL) == 0) {
Sage Weil355da1e2009-10-06 11:31:08 -0700732 inode->i_mode = le32_to_cpu(info->mode);
Eric W. Biedermanab871b92013-01-31 03:40:12 -0800733 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));
Sage Weil355da1e2009-10-06 11:31:08 -0700735 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
Eric W. Biedermanbd2bae62013-01-31 04:05:39 -0800736 from_kuid(&init_user_ns, inode->i_uid),
737 from_kgid(&init_user_ns, inode->i_gid));
Sage Weil355da1e2009-10-06 11:31:08 -0700738 }
739
Yan, Zhengf98a1282014-04-17 08:55:50 +0800740 if ((new_version || (new_issued & CEPH_CAP_LINK_SHARED)) &&
741 (issued & CEPH_CAP_LINK_EXCL) == 0)
Miklos Szeredibfe86842011-10-28 14:13:29 +0200742 set_nlink(inode, le32_to_cpu(info->nlink));
Sage Weil355da1e2009-10-06 11:31:08 -0700743
Yan, Zhengf98a1282014-04-17 08:55:50 +0800744 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 }
Sage Weil355da1e2009-10-06 11:31:08 -0700753
Yan, Zhengf98a1282014-04-17 08:55:50 +0800754 if (new_version ||
755 (new_issued & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR))) {
756 ci->i_layout = info->layout;
757 queue_trunc = ceph_fill_file_size(inode, issued,
758 le32_to_cpu(info->truncate_seq),
759 le64_to_cpu(info->truncate_size),
760 le64_to_cpu(info->size));
761 /* only update max_size on auth cap */
762 if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
763 ci->i_max_size != le64_to_cpu(info->max_size)) {
764 dout("max_size %lld -> %llu\n", ci->i_max_size,
765 le64_to_cpu(info->max_size));
766 ci->i_max_size = le64_to_cpu(info->max_size);
767 }
768 }
Sage Weil355da1e2009-10-06 11:31:08 -0700769
770 /* xattrs */
771 /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
Yan, Zheng508b32d2014-09-16 21:46:17 +0800772 if ((ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL)) &&
Sage Weil355da1e2009-10-06 11:31:08 -0700773 le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) {
774 if (ci->i_xattrs.blob)
775 ceph_buffer_put(ci->i_xattrs.blob);
776 ci->i_xattrs.blob = xattr_blob;
777 if (xattr_blob)
778 memcpy(ci->i_xattrs.blob->vec.iov_base,
779 iinfo->xattr_data, iinfo->xattr_len);
780 ci->i_xattrs.version = le64_to_cpu(info->xattr_version);
Guangliang Zhao7221fe42013-11-11 15:18:03 +0800781 ceph_forget_all_cached_acls(inode);
Sage Weila6424e42010-04-29 09:28:11 -0700782 xattr_blob = NULL;
Sage Weil355da1e2009-10-06 11:31:08 -0700783 }
784
785 inode->i_mapping->a_ops = &ceph_aops;
786 inode->i_mapping->backing_dev_info =
Cheng Renquan640ef792010-03-26 17:40:33 +0800787 &ceph_sb_to_client(inode->i_sb)->backing_dev_info;
Sage Weil355da1e2009-10-06 11:31:08 -0700788
789 switch (inode->i_mode & S_IFMT) {
790 case S_IFIFO:
791 case S_IFBLK:
792 case S_IFCHR:
793 case S_IFSOCK:
794 init_special_inode(inode, inode->i_mode, inode->i_rdev);
795 inode->i_op = &ceph_file_iops;
796 break;
797 case S_IFREG:
798 inode->i_op = &ceph_file_iops;
799 inode->i_fop = &ceph_file_fops;
800 break;
801 case S_IFLNK:
802 inode->i_op = &ceph_symlink_iops;
803 if (!ci->i_symlink) {
Xi Wang810339ec2012-02-03 09:55:36 -0500804 u32 symlen = iinfo->symlink_len;
Sage Weil355da1e2009-10-06 11:31:08 -0700805 char *sym;
806
Sage Weilbe655592011-11-30 09:47:09 -0800807 spin_unlock(&ci->i_ceph_lock);
Sage Weil355da1e2009-10-06 11:31:08 -0700808
Xi Wang810339ec2012-02-03 09:55:36 -0500809 err = -EINVAL;
810 if (WARN_ON(symlen != inode->i_size))
811 goto out;
812
Sage Weil355da1e2009-10-06 11:31:08 -0700813 err = -ENOMEM;
Xi Wang810339ec2012-02-03 09:55:36 -0500814 sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS);
Sage Weil355da1e2009-10-06 11:31:08 -0700815 if (!sym)
816 goto out;
Sage Weil355da1e2009-10-06 11:31:08 -0700817
Sage Weilbe655592011-11-30 09:47:09 -0800818 spin_lock(&ci->i_ceph_lock);
Sage Weil355da1e2009-10-06 11:31:08 -0700819 if (!ci->i_symlink)
820 ci->i_symlink = sym;
821 else
822 kfree(sym); /* lost a race */
823 }
824 break;
825 case S_IFDIR:
826 inode->i_op = &ceph_dir_iops;
827 inode->i_fop = &ceph_dir_fops;
828
Sage Weil14303d22010-12-14 17:37:52 -0800829 ci->i_dir_layout = iinfo->dir_layout;
830
Sage Weil355da1e2009-10-06 11:31:08 -0700831 ci->i_files = le64_to_cpu(info->files);
832 ci->i_subdirs = le64_to_cpu(info->subdirs);
833 ci->i_rbytes = le64_to_cpu(info->rbytes);
834 ci->i_rfiles = le64_to_cpu(info->rfiles);
835 ci->i_rsubdirs = le64_to_cpu(info->rsubdirs);
836 ceph_decode_timespec(&ci->i_rctime, &info->rctime);
Sage Weil355da1e2009-10-06 11:31:08 -0700837 break;
838 default:
839 pr_err("fill_inode %llx.%llx BAD mode 0%o\n",
840 ceph_vinop(inode), inode->i_mode);
841 }
842
Yan, Zhenga8673d62013-02-18 16:38:14 +0800843 /* set dir completion flag? */
844 if (S_ISDIR(inode->i_mode) &&
845 ci->i_files == 0 && ci->i_subdirs == 0 &&
846 ceph_snap(inode) == CEPH_NOSNAP &&
847 (le32_to_cpu(info->cap.caps) & CEPH_CAP_FILE_SHARED) &&
848 (issued & CEPH_CAP_FILE_EXCL) == 0 &&
Yan, Zheng2f276c52013-03-13 19:44:32 +0800849 !__ceph_dir_is_complete(ci)) {
Yan, Zhenga8673d62013-02-18 16:38:14 +0800850 dout(" marking %p complete (empty)\n", inode);
Yan, Zheng70db4f32014-10-21 18:09:56 -0700851 __ceph_dir_set_complete(ci, atomic_read(&ci->i_release_count),
852 ci->i_ordered_count);
Yan, Zhenga8673d62013-02-18 16:38:14 +0800853 }
Sage Weil355da1e2009-10-06 11:31:08 -0700854
855 /* were we issued a capability? */
856 if (info->cap.caps) {
857 if (ceph_snap(inode) == CEPH_NOSNAP) {
858 ceph_add_cap(inode, session,
859 le64_to_cpu(info->cap.cap_id),
860 cap_fmode,
861 le32_to_cpu(info->cap.caps),
862 le32_to_cpu(info->cap.wanted),
863 le32_to_cpu(info->cap.seq),
864 le32_to_cpu(info->cap.mseq),
865 le64_to_cpu(info->cap.realm),
Yan, Zhengd9df2782014-04-18 09:57:11 +0800866 info->cap.flags, &new_cap);
867 wake = true;
Sage Weil355da1e2009-10-06 11:31:08 -0700868 } else {
Sage Weil355da1e2009-10-06 11:31:08 -0700869 dout(" %p got snap_caps %s\n", inode,
870 ceph_cap_string(le32_to_cpu(info->cap.caps)));
871 ci->i_snap_caps |= le32_to_cpu(info->cap.caps);
872 if (cap_fmode >= 0)
873 __ceph_get_fmode(ci, cap_fmode);
Sage Weil355da1e2009-10-06 11:31:08 -0700874 }
Sage Weil04d000e2010-05-07 11:26:34 -0700875 } else if (cap_fmode >= 0) {
Fabian Frederickf3ae1b92014-06-06 14:35:37 -0700876 pr_warn("mds issued no caps on %llx.%llx\n",
Sage Weil04d000e2010-05-07 11:26:34 -0700877 ceph_vinop(inode));
878 __ceph_get_fmode(ci, cap_fmode);
Sage Weil355da1e2009-10-06 11:31:08 -0700879 }
Yan, Zheng31c542a2014-11-14 21:41:55 +0800880
881 if (iinfo->inline_version > 0 &&
882 iinfo->inline_version >= ci->i_inline_version) {
883 int cache_caps = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
884 ci->i_inline_version = iinfo->inline_version;
885 if (ci->i_inline_version != CEPH_INLINE_NONE &&
Yan, Zheng01deead2014-11-14 21:56:29 +0800886 (locked_page ||
887 (le32_to_cpu(info->cap.caps) & cache_caps)))
Yan, Zheng31c542a2014-11-14 21:41:55 +0800888 fill_inline = true;
889 }
890
Sage Weil355da1e2009-10-06 11:31:08 -0700891 spin_unlock(&ci->i_ceph_lock);
892
Yan, Zheng31c542a2014-11-14 21:41:55 +0800893 if (fill_inline)
Yan, Zheng01deead2014-11-14 21:56:29 +0800894 ceph_fill_inline_data(inode, locked_page,
Yan, Zheng31c542a2014-11-14 21:41:55 +0800895 iinfo->inline_data, iinfo->inline_len);
896
Yan, Zhengd9df2782014-04-18 09:57:11 +0800897 if (wake)
898 wake_up_all(&ci->i_cap_wq);
899
Sage Weil355da1e2009-10-06 11:31:08 -0700900 /* queue truncate if we saw i_size decrease */
901 if (queue_trunc)
902 ceph_queue_vmtruncate(inode);
903
904 /* populate frag tree */
Yan, Zheng3e7fbe92014-04-18 16:00:30 +0800905 if (S_ISDIR(inode->i_mode))
906 ceph_fill_fragtree(inode, &info->fragtree, dirinfo);
Sage Weil355da1e2009-10-06 11:31:08 -0700907
908 /* update delegation info? */
909 if (dirinfo)
910 ceph_fill_dirfrag(inode, dirinfo);
911
912 err = 0;
Sage Weil355da1e2009-10-06 11:31:08 -0700913out:
Yan, Zhengd9df2782014-04-18 09:57:11 +0800914 if (new_cap)
915 ceph_put_cap(mdsc, new_cap);
Sage Weilb6c1d5b2009-12-07 12:17:17 -0800916 if (xattr_blob)
917 ceph_buffer_put(xattr_blob);
Sage Weil355da1e2009-10-06 11:31:08 -0700918 return err;
919}
920
921/*
922 * caller should hold session s_mutex.
923 */
924static void update_dentry_lease(struct dentry *dentry,
925 struct ceph_mds_reply_lease *lease,
926 struct ceph_mds_session *session,
927 unsigned long from_time)
928{
929 struct ceph_dentry_info *di = ceph_dentry(dentry);
930 long unsigned duration = le32_to_cpu(lease->duration_ms);
931 long unsigned ttl = from_time + (duration * HZ) / 1000;
932 long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000;
933 struct inode *dir;
934
935 /* only track leases on regular dentries */
936 if (dentry->d_op != &ceph_dentry_ops)
937 return;
938
939 spin_lock(&dentry->d_lock);
Sage Weil2f90b852011-07-26 11:28:25 -0700940 dout("update_dentry_lease %p duration %lu ms ttl %lu\n",
941 dentry, duration, ttl);
Sage Weil355da1e2009-10-06 11:31:08 -0700942
943 /* make lease_rdcache_gen match directory */
944 dir = dentry->d_parent->d_inode;
945 di->lease_shared_gen = ceph_inode(dir)->i_shared_gen;
946
Sage Weil2f90b852011-07-26 11:28:25 -0700947 if (duration == 0)
Sage Weil355da1e2009-10-06 11:31:08 -0700948 goto out_unlock;
949
950 if (di->lease_gen == session->s_cap_gen &&
951 time_before(ttl, dentry->d_time))
952 goto out_unlock; /* we already have a newer lease. */
953
954 if (di->lease_session && di->lease_session != session)
955 goto out_unlock;
956
957 ceph_dentry_lru_touch(dentry);
958
959 if (!di->lease_session)
960 di->lease_session = ceph_get_mds_session(session);
961 di->lease_gen = session->s_cap_gen;
962 di->lease_seq = le32_to_cpu(lease->seq);
963 di->lease_renew_after = half_ttl;
964 di->lease_renew_from = 0;
965 dentry->d_time = ttl;
966out_unlock:
967 spin_unlock(&dentry->d_lock);
968 return;
969}
970
971/*
Sage Weil1cd39352010-05-03 22:08:02 -0700972 * splice a dentry to an inode.
973 * caller must hold directory i_mutex for this to be safe.
974 *
975 * we will only rehash the resulting dentry if @prehash is
976 * true; @prehash will be set to false (for the benefit of
977 * the caller) if we fail.
978 */
979static struct dentry *splice_dentry(struct dentry *dn, struct inode *in,
Yan, Zheng0a8a70f2014-04-14 13:13:02 +0800980 bool *prehash)
Sage Weil1cd39352010-05-03 22:08:02 -0700981{
982 struct dentry *realdn;
983
984 BUG_ON(dn->d_inode);
985
986 /* dn must be unhashed */
987 if (!d_unhashed(dn))
988 d_drop(dn);
Al Viro41d28bc2014-10-12 22:24:21 -0400989 realdn = d_splice_alias(in, dn);
Sage Weil1cd39352010-05-03 22:08:02 -0700990 if (IS_ERR(realdn)) {
Sage Weild69ed052010-06-21 10:38:14 -0700991 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
992 PTR_ERR(realdn), dn, in, ceph_vinop(in));
Sage Weil1cd39352010-05-03 22:08:02 -0700993 if (prehash)
994 *prehash = false; /* don't rehash on error */
995 dn = realdn; /* note realdn contains the error */
996 goto out;
997 } else if (realdn) {
998 dout("dn %p (%d) spliced with %p (%d) "
999 "inode %p ino %llx.%llx\n",
Al Viro84d08fa2013-07-05 18:59:33 +04001000 dn, d_count(dn),
1001 realdn, d_count(realdn),
Sage Weil1cd39352010-05-03 22:08:02 -07001002 realdn->d_inode, ceph_vinop(realdn->d_inode));
1003 dput(dn);
1004 dn = realdn;
1005 } else {
1006 BUG_ON(!ceph_dentry(dn));
1007 dout("dn %p attached to %p ino %llx.%llx\n",
1008 dn, dn->d_inode, ceph_vinop(dn->d_inode));
1009 }
1010 if ((!prehash || *prehash) && d_unhashed(dn))
1011 d_rehash(dn);
Sage Weil1cd39352010-05-03 22:08:02 -07001012out:
1013 return dn;
1014}
1015
1016/*
Sage Weil355da1e2009-10-06 11:31:08 -07001017 * Incorporate results into the local cache. This is either just
1018 * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
1019 * after a lookup).
1020 *
1021 * A reply may contain
1022 * a directory inode along with a dentry.
1023 * and/or a target inode
1024 *
1025 * Called with snap_rwsem (read).
1026 */
1027int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req,
1028 struct ceph_mds_session *session)
1029{
1030 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1031 struct inode *in = NULL;
Sage Weil355da1e2009-10-06 11:31:08 -07001032 struct ceph_vino vino;
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001033 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
Sage Weil355da1e2009-10-06 11:31:08 -07001034 int err = 0;
1035
1036 dout("fill_trace %p is_dentry %d is_target %d\n", req,
1037 rinfo->head->is_dentry, rinfo->head->is_target);
1038
1039#if 0
1040 /*
1041 * Debugging hook:
1042 *
1043 * If we resend completed ops to a recovering mds, we get no
1044 * trace. Since that is very rare, pretend this is the case
1045 * to ensure the 'no trace' handlers in the callers behave.
1046 *
1047 * Fill in inodes unconditionally to avoid breaking cap
1048 * invariants.
1049 */
1050 if (rinfo->head->op & CEPH_MDS_OP_WRITE) {
1051 pr_info("fill_trace faking empty trace on %lld %s\n",
1052 req->r_tid, ceph_mds_op_name(rinfo->head->op));
1053 if (rinfo->head->is_dentry) {
1054 rinfo->head->is_dentry = 0;
1055 err = fill_inode(req->r_locked_dir,
1056 &rinfo->diri, rinfo->dirfrag,
1057 session, req->r_request_started, -1);
1058 }
1059 if (rinfo->head->is_target) {
1060 rinfo->head->is_target = 0;
1061 ininfo = rinfo->targeti.in;
1062 vino.ino = le64_to_cpu(ininfo->ino);
1063 vino.snap = le64_to_cpu(ininfo->snapid);
1064 in = ceph_get_inode(sb, vino);
1065 err = fill_inode(in, &rinfo->targeti, NULL,
1066 session, req->r_request_started,
1067 req->r_fmode);
1068 iput(in);
1069 }
1070 }
1071#endif
1072
1073 if (!rinfo->head->is_target && !rinfo->head->is_dentry) {
1074 dout("fill_trace reply is empty!\n");
Sage Weil167c9e32010-05-14 10:02:57 -07001075 if (rinfo->head->result == 0 && req->r_locked_dir)
1076 ceph_invalidate_dir_request(req);
Sage Weil355da1e2009-10-06 11:31:08 -07001077 return 0;
1078 }
1079
1080 if (rinfo->head->is_dentry) {
Sage Weil5b1daec2010-01-25 11:33:08 -08001081 struct inode *dir = req->r_locked_dir;
1082
Sage Weil6c5e50f2012-08-21 15:55:25 -07001083 if (dir) {
Yan, Zheng01deead2014-11-14 21:56:29 +08001084 err = fill_inode(dir, NULL,
1085 &rinfo->diri, rinfo->dirfrag,
Sage Weil6c5e50f2012-08-21 15:55:25 -07001086 session, req->r_request_started, -1,
1087 &req->r_caps_reservation);
1088 if (err < 0)
Yan, Zheng19913b42014-03-06 16:40:32 +08001089 goto done;
Sage Weil6c5e50f2012-08-21 15:55:25 -07001090 } else {
1091 WARN_ON_ONCE(1);
1092 }
Yan, Zheng19913b42014-03-06 16:40:32 +08001093
1094 if (dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME) {
1095 struct qstr dname;
1096 struct dentry *dn, *parent;
1097
1098 BUG_ON(!rinfo->head->is_target);
1099 BUG_ON(req->r_dentry);
1100
1101 parent = d_find_any_alias(dir);
1102 BUG_ON(!parent);
1103
1104 dname.name = rinfo->dname;
1105 dname.len = rinfo->dname_len;
1106 dname.hash = full_name_hash(dname.name, dname.len);
1107 vino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1108 vino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1109retry_lookup:
1110 dn = d_lookup(parent, &dname);
1111 dout("d_lookup on parent=%p name=%.*s got %p\n",
1112 parent, dname.len, dname.name, dn);
1113
1114 if (!dn) {
1115 dn = d_alloc(parent, &dname);
1116 dout("d_alloc %p '%.*s' = %p\n", parent,
1117 dname.len, dname.name, dn);
1118 if (dn == NULL) {
1119 dput(parent);
1120 err = -ENOMEM;
1121 goto done;
1122 }
1123 err = ceph_init_dentry(dn);
1124 if (err < 0) {
1125 dput(dn);
1126 dput(parent);
1127 goto done;
1128 }
1129 } else if (dn->d_inode &&
1130 (ceph_ino(dn->d_inode) != vino.ino ||
1131 ceph_snap(dn->d_inode) != vino.snap)) {
1132 dout(" dn %p points to wrong inode %p\n",
1133 dn, dn->d_inode);
1134 d_delete(dn);
1135 dput(dn);
1136 goto retry_lookup;
1137 }
1138
1139 req->r_dentry = dn;
1140 dput(parent);
1141 }
Sage Weil5b1daec2010-01-25 11:33:08 -08001142 }
1143
Yan, Zheng86b58d12013-12-05 12:38:59 +08001144 if (rinfo->head->is_target) {
1145 vino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1146 vino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1147
1148 in = ceph_get_inode(sb, vino);
1149 if (IS_ERR(in)) {
1150 err = PTR_ERR(in);
1151 goto done;
1152 }
1153 req->r_target_inode = in;
1154
Yan, Zheng01deead2014-11-14 21:56:29 +08001155 err = fill_inode(in, req->r_locked_page, &rinfo->targeti, NULL,
Yan, Zheng86b58d12013-12-05 12:38:59 +08001156 session, req->r_request_started,
Yan, Zheng48193012014-04-01 20:34:18 +08001157 (!req->r_aborted && rinfo->head->result == 0) ?
Yan, Zheng86b58d12013-12-05 12:38:59 +08001158 req->r_fmode : -1,
1159 &req->r_caps_reservation);
1160 if (err < 0) {
1161 pr_err("fill_inode badness %p %llx.%llx\n",
1162 in, ceph_vinop(in));
1163 goto done;
1164 }
1165 }
1166
Sage Weil9358c6d2010-03-30 13:54:41 -07001167 /*
1168 * ignore null lease/binding on snapdir ENOENT, or else we
1169 * will have trouble splicing in the virtual snapdir later
1170 */
1171 if (rinfo->head->is_dentry && !req->r_aborted &&
Sage Weil6c5e50f2012-08-21 15:55:25 -07001172 req->r_locked_dir &&
Sage Weil9358c6d2010-03-30 13:54:41 -07001173 (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name,
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001174 fsc->mount_options->snapdir_name,
Sage Weil9358c6d2010-03-30 13:54:41 -07001175 req->r_dentry->d_name.len))) {
Sage Weil355da1e2009-10-06 11:31:08 -07001176 /*
1177 * lookup link rename : null -> possibly existing inode
1178 * mknod symlink mkdir : null -> new inode
1179 * unlink : linked -> null
1180 */
1181 struct inode *dir = req->r_locked_dir;
1182 struct dentry *dn = req->r_dentry;
1183 bool have_dir_cap, have_lease;
1184
1185 BUG_ON(!dn);
1186 BUG_ON(!dir);
1187 BUG_ON(dn->d_parent->d_inode != dir);
1188 BUG_ON(ceph_ino(dir) !=
1189 le64_to_cpu(rinfo->diri.in->ino));
1190 BUG_ON(ceph_snap(dir) !=
1191 le64_to_cpu(rinfo->diri.in->snapid));
1192
Sage Weil355da1e2009-10-06 11:31:08 -07001193 /* do we have a lease on the whole dir? */
1194 have_dir_cap =
1195 (le32_to_cpu(rinfo->diri.in->cap.caps) &
1196 CEPH_CAP_FILE_SHARED);
1197
1198 /* do we have a dn lease? */
1199 have_lease = have_dir_cap ||
Sage Weil2f90b852011-07-26 11:28:25 -07001200 le32_to_cpu(rinfo->dlease->duration_ms);
Sage Weil355da1e2009-10-06 11:31:08 -07001201 if (!have_lease)
1202 dout("fill_trace no dentry lease or dir cap\n");
1203
1204 /* rename? */
1205 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) {
Yan, Zheng0a8a70f2014-04-14 13:13:02 +08001206 struct inode *olddir = req->r_old_dentry_dir;
1207 BUG_ON(!olddir);
1208
Al Viroa4555892014-10-21 20:11:25 -04001209 dout(" src %p '%pd' dst %p '%pd'\n",
Sage Weil355da1e2009-10-06 11:31:08 -07001210 req->r_old_dentry,
Al Viroa4555892014-10-21 20:11:25 -04001211 req->r_old_dentry,
1212 dn, dn);
Sage Weil355da1e2009-10-06 11:31:08 -07001213 dout("fill_trace doing d_move %p -> %p\n",
1214 req->r_old_dentry, dn);
Sage Weilc10f5e12010-04-16 12:56:11 -07001215
Sage Weil355da1e2009-10-06 11:31:08 -07001216 d_move(req->r_old_dentry, dn);
Al Viroa4555892014-10-21 20:11:25 -04001217 dout(" src %p '%pd' dst %p '%pd'\n",
Sage Weil355da1e2009-10-06 11:31:08 -07001218 req->r_old_dentry,
Al Viroa4555892014-10-21 20:11:25 -04001219 req->r_old_dentry,
1220 dn, dn);
Sage Weil81a6cf22010-05-14 09:35:38 -07001221
Sage Weilc4a29f22009-12-21 11:42:18 -08001222 /* ensure target dentry is invalidated, despite
1223 rehashing bug in vfs_rename_dir */
Sage Weil81a6cf22010-05-14 09:35:38 -07001224 ceph_invalidate_dentry_lease(dn);
1225
Yan, Zheng0a8a70f2014-04-14 13:13:02 +08001226 /* d_move screws up sibling dentries' offsets */
Yan, Zheng70db4f32014-10-21 18:09:56 -07001227 ceph_dir_clear_ordered(dir);
1228 ceph_dir_clear_ordered(olddir);
Yan, Zheng0a8a70f2014-04-14 13:13:02 +08001229
Milosz Tanski99ccbd22013-08-21 17:29:54 -04001230 dout("dn %p gets new offset %lld\n", req->r_old_dentry,
Sage Weil1cd39352010-05-03 22:08:02 -07001231 ceph_dentry(req->r_old_dentry)->offset);
Sage Weil81a6cf22010-05-14 09:35:38 -07001232
Sage Weil355da1e2009-10-06 11:31:08 -07001233 dn = req->r_old_dentry; /* use old_dentry */
Sage Weil355da1e2009-10-06 11:31:08 -07001234 }
1235
1236 /* null dentry? */
1237 if (!rinfo->head->is_target) {
1238 dout("fill_trace null dentry\n");
1239 if (dn->d_inode) {
Yan, Zheng70db4f32014-10-21 18:09:56 -07001240 ceph_dir_clear_ordered(dir);
Sage Weil355da1e2009-10-06 11:31:08 -07001241 dout("d_delete %p\n", dn);
1242 d_delete(dn);
1243 } else {
1244 dout("d_instantiate %p NULL\n", dn);
1245 d_instantiate(dn, NULL);
1246 if (have_lease && d_unhashed(dn))
1247 d_rehash(dn);
1248 update_dentry_lease(dn, rinfo->dlease,
1249 session,
1250 req->r_request_started);
1251 }
1252 goto done;
1253 }
1254
1255 /* attach proper inode */
Yan, Zheng86b58d12013-12-05 12:38:59 +08001256 if (!dn->d_inode) {
Yan, Zheng70db4f32014-10-21 18:09:56 -07001257 ceph_dir_clear_ordered(dir);
Yan, Zheng86b58d12013-12-05 12:38:59 +08001258 ihold(in);
Yan, Zheng0a8a70f2014-04-14 13:13:02 +08001259 dn = splice_dentry(dn, in, &have_lease);
Sage Weil355da1e2009-10-06 11:31:08 -07001260 if (IS_ERR(dn)) {
1261 err = PTR_ERR(dn);
1262 goto done;
1263 }
1264 req->r_dentry = dn; /* may have spliced */
Yan, Zheng86b58d12013-12-05 12:38:59 +08001265 } else if (dn->d_inode && dn->d_inode != in) {
Sage Weil355da1e2009-10-06 11:31:08 -07001266 dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
Yan, Zheng86b58d12013-12-05 12:38:59 +08001267 dn, dn->d_inode, ceph_vinop(dn->d_inode),
1268 ceph_vinop(in));
Sage Weil355da1e2009-10-06 11:31:08 -07001269 have_lease = false;
Sage Weil355da1e2009-10-06 11:31:08 -07001270 }
1271
1272 if (have_lease)
1273 update_dentry_lease(dn, rinfo->dlease, session,
1274 req->r_request_started);
1275 dout(" final dn %p\n", dn);
Yan, Zheng86b58d12013-12-05 12:38:59 +08001276 } else if (!req->r_aborted &&
1277 (req->r_op == CEPH_MDS_OP_LOOKUPSNAP ||
1278 req->r_op == CEPH_MDS_OP_MKSNAP)) {
Sage Weil355da1e2009-10-06 11:31:08 -07001279 struct dentry *dn = req->r_dentry;
Yan, Zheng0a8a70f2014-04-14 13:13:02 +08001280 struct inode *dir = req->r_locked_dir;
Sage Weil355da1e2009-10-06 11:31:08 -07001281
1282 /* fill out a snapdir LOOKUPSNAP dentry */
1283 BUG_ON(!dn);
Yan, Zheng0a8a70f2014-04-14 13:13:02 +08001284 BUG_ON(!dir);
1285 BUG_ON(ceph_snap(dir) != CEPH_SNAPDIR);
Sage Weil355da1e2009-10-06 11:31:08 -07001286 dout(" linking snapped dir %p to dn %p\n", in, dn);
Yan, Zheng70db4f32014-10-21 18:09:56 -07001287 ceph_dir_clear_ordered(dir);
Yan, Zheng86b58d12013-12-05 12:38:59 +08001288 ihold(in);
Yan, Zheng0a8a70f2014-04-14 13:13:02 +08001289 dn = splice_dentry(dn, in, NULL);
Sage Weil355da1e2009-10-06 11:31:08 -07001290 if (IS_ERR(dn)) {
1291 err = PTR_ERR(dn);
1292 goto done;
1293 }
1294 req->r_dentry = dn; /* may have spliced */
Sage Weil355da1e2009-10-06 11:31:08 -07001295 }
Sage Weil355da1e2009-10-06 11:31:08 -07001296done:
1297 dout("fill_trace done err=%d\n", err);
1298 return err;
1299}
1300
1301/*
1302 * Prepopulate our cache with readdir results, leases, etc.
1303 */
Sage Weil79f9f992013-01-29 02:55:31 -05001304static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req,
1305 struct ceph_mds_session *session)
1306{
1307 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1308 int i, err = 0;
1309
1310 for (i = 0; i < rinfo->dir_nr; i++) {
1311 struct ceph_vino vino;
1312 struct inode *in;
1313 int rc;
1314
1315 vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino);
1316 vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid);
1317
1318 in = ceph_get_inode(req->r_dentry->d_sb, vino);
1319 if (IS_ERR(in)) {
1320 err = PTR_ERR(in);
1321 dout("new_inode badness got %d\n", err);
1322 continue;
1323 }
Yan, Zheng01deead2014-11-14 21:56:29 +08001324 rc = fill_inode(in, NULL, &rinfo->dir_in[i], NULL, session,
Sage Weil79f9f992013-01-29 02:55:31 -05001325 req->r_request_started, -1,
1326 &req->r_caps_reservation);
1327 if (rc < 0) {
1328 pr_err("fill_inode badness on %p got %d\n", in, rc);
1329 err = rc;
1330 continue;
1331 }
1332 }
1333
1334 return err;
1335}
1336
Sage Weil355da1e2009-10-06 11:31:08 -07001337int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1338 struct ceph_mds_session *session)
1339{
1340 struct dentry *parent = req->r_dentry;
1341 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1342 struct qstr dname;
1343 struct dentry *dn;
1344 struct inode *in;
Yan, Zheng86b58d12013-12-05 12:38:59 +08001345 int err = 0, ret, i;
Sage Weil355da1e2009-10-06 11:31:08 -07001346 struct inode *snapdir = NULL;
1347 struct ceph_mds_request_head *rhead = req->r_request->front.iov_base;
Sage Weil355da1e2009-10-06 11:31:08 -07001348 struct ceph_dentry_info *di;
Yan, Zheng81c6aea2013-09-18 09:44:13 +08001349 u64 r_readdir_offset = req->r_readdir_offset;
1350 u32 frag = le32_to_cpu(rhead->args.readdir.frag);
1351
1352 if (rinfo->dir_dir &&
1353 le32_to_cpu(rinfo->dir_dir->frag) != frag) {
1354 dout("readdir_prepopulate got new frag %x -> %x\n",
1355 frag, le32_to_cpu(rinfo->dir_dir->frag));
1356 frag = le32_to_cpu(rinfo->dir_dir->frag);
1357 if (ceph_frag_is_leftmost(frag))
1358 r_readdir_offset = 2;
1359 else
1360 r_readdir_offset = 0;
1361 }
Sage Weil355da1e2009-10-06 11:31:08 -07001362
Sage Weil79f9f992013-01-29 02:55:31 -05001363 if (req->r_aborted)
1364 return readdir_prepopulate_inodes_only(req, session);
1365
Sage Weil355da1e2009-10-06 11:31:08 -07001366 if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
1367 snapdir = ceph_get_snapdir(parent->d_inode);
1368 parent = d_find_alias(snapdir);
1369 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1370 rinfo->dir_nr, parent);
1371 } else {
1372 dout("readdir_prepopulate %d items under dn %p\n",
1373 rinfo->dir_nr, parent);
1374 if (rinfo->dir_dir)
1375 ceph_fill_dirfrag(parent->d_inode, rinfo->dir_dir);
1376 }
1377
Yan, Zheng86b58d12013-12-05 12:38:59 +08001378 /* FIXME: release caps/leases if error occurs */
Sage Weil355da1e2009-10-06 11:31:08 -07001379 for (i = 0; i < rinfo->dir_nr; i++) {
1380 struct ceph_vino vino;
1381
1382 dname.name = rinfo->dir_dname[i];
1383 dname.len = rinfo->dir_dname_len[i];
1384 dname.hash = full_name_hash(dname.name, dname.len);
1385
1386 vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino);
1387 vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid);
1388
1389retry_lookup:
1390 dn = d_lookup(parent, &dname);
1391 dout("d_lookup on parent=%p name=%.*s got %p\n",
1392 parent, dname.len, dname.name, dn);
1393
1394 if (!dn) {
1395 dn = d_alloc(parent, &dname);
1396 dout("d_alloc %p '%.*s' = %p\n", parent,
1397 dname.len, dname.name, dn);
1398 if (dn == NULL) {
1399 dout("d_alloc badness\n");
1400 err = -ENOMEM;
1401 goto out;
1402 }
Yan, Zheng86b58d12013-12-05 12:38:59 +08001403 ret = ceph_init_dentry(dn);
1404 if (ret < 0) {
Sage Weil8c696732010-07-22 14:11:56 -07001405 dput(dn);
Yan, Zheng86b58d12013-12-05 12:38:59 +08001406 err = ret;
Sage Weil355da1e2009-10-06 11:31:08 -07001407 goto out;
Sage Weil8c696732010-07-22 14:11:56 -07001408 }
Sage Weil355da1e2009-10-06 11:31:08 -07001409 } else if (dn->d_inode &&
1410 (ceph_ino(dn->d_inode) != vino.ino ||
1411 ceph_snap(dn->d_inode) != vino.snap)) {
1412 dout(" dn %p points to wrong inode %p\n",
1413 dn, dn->d_inode);
1414 d_delete(dn);
1415 dput(dn);
1416 goto retry_lookup;
1417 } else {
1418 /* reorder parent's d_subdirs */
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001419 spin_lock(&parent->d_lock);
1420 spin_lock_nested(&dn->d_lock, DENTRY_D_LOCK_NESTED);
Al Viro946e51f2014-10-26 19:19:16 -04001421 list_move(&dn->d_child, &parent->d_subdirs);
Sage Weil355da1e2009-10-06 11:31:08 -07001422 spin_unlock(&dn->d_lock);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001423 spin_unlock(&parent->d_lock);
Sage Weil355da1e2009-10-06 11:31:08 -07001424 }
1425
Sage Weil355da1e2009-10-06 11:31:08 -07001426 /* inode */
1427 if (dn->d_inode) {
1428 in = dn->d_inode;
1429 } else {
1430 in = ceph_get_inode(parent->d_sb, vino);
Dan Carpenterac1f12e2010-08-25 09:11:35 +02001431 if (IS_ERR(in)) {
Sage Weil355da1e2009-10-06 11:31:08 -07001432 dout("new_inode badness\n");
Al Viro2744c172012-09-26 21:41:05 -04001433 d_drop(dn);
Sage Weil355da1e2009-10-06 11:31:08 -07001434 dput(dn);
Dan Carpenterac1f12e2010-08-25 09:11:35 +02001435 err = PTR_ERR(in);
Sage Weil355da1e2009-10-06 11:31:08 -07001436 goto out;
1437 }
Sage Weil355da1e2009-10-06 11:31:08 -07001438 }
1439
Yan, Zheng01deead2014-11-14 21:56:29 +08001440 if (fill_inode(in, NULL, &rinfo->dir_in[i], NULL, session,
Sage Weil355da1e2009-10-06 11:31:08 -07001441 req->r_request_started, -1,
1442 &req->r_caps_reservation) < 0) {
1443 pr_err("fill_inode badness on %p\n", in);
Yan, Zheng86b58d12013-12-05 12:38:59 +08001444 if (!dn->d_inode)
1445 iput(in);
1446 d_drop(dn);
Sage Weild69ed052010-06-21 10:38:14 -07001447 goto next_item;
Sage Weil355da1e2009-10-06 11:31:08 -07001448 }
Yan, Zheng86b58d12013-12-05 12:38:59 +08001449
1450 if (!dn->d_inode) {
Yan, Zheng0a8a70f2014-04-14 13:13:02 +08001451 dn = splice_dentry(dn, in, NULL);
Yan, Zheng86b58d12013-12-05 12:38:59 +08001452 if (IS_ERR(dn)) {
1453 err = PTR_ERR(dn);
1454 dn = NULL;
1455 goto next_item;
1456 }
1457 }
1458
1459 di = dn->d_fsdata;
1460 di->offset = ceph_make_fpos(frag, i + r_readdir_offset);
1461
1462 update_dentry_lease(dn, rinfo->dir_dlease[i],
1463 req->r_session,
1464 req->r_request_started);
Sage Weild69ed052010-06-21 10:38:14 -07001465next_item:
1466 if (dn)
1467 dput(dn);
Sage Weil355da1e2009-10-06 11:31:08 -07001468 }
Yan, Zheng86b58d12013-12-05 12:38:59 +08001469 if (err == 0)
1470 req->r_did_prepopulate = true;
Sage Weil355da1e2009-10-06 11:31:08 -07001471
1472out:
1473 if (snapdir) {
1474 iput(snapdir);
1475 dput(parent);
1476 }
1477 dout("readdir_prepopulate done\n");
1478 return err;
1479}
1480
1481int ceph_inode_set_size(struct inode *inode, loff_t size)
1482{
1483 struct ceph_inode_info *ci = ceph_inode(inode);
1484 int ret = 0;
1485
Sage Weilbe655592011-11-30 09:47:09 -08001486 spin_lock(&ci->i_ceph_lock);
Sage Weil355da1e2009-10-06 11:31:08 -07001487 dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size);
1488 inode->i_size = size;
1489 inode->i_blocks = (size + (1 << 9) - 1) >> 9;
1490
1491 /* tell the MDS if we are approaching max_size */
1492 if ((size << 1) >= ci->i_max_size &&
1493 (ci->i_reported_size << 1) < ci->i_max_size)
1494 ret = 1;
1495
Sage Weilbe655592011-11-30 09:47:09 -08001496 spin_unlock(&ci->i_ceph_lock);
Sage Weil355da1e2009-10-06 11:31:08 -07001497 return ret;
1498}
1499
1500/*
1501 * Write back inode data in a worker thread. (This can't be done
1502 * in the message handler context.)
1503 */
Sage Weil3c6f6b72010-02-09 15:24:44 -08001504void ceph_queue_writeback(struct inode *inode)
1505{
Sage Weil15a20152011-11-05 22:06:31 -07001506 ihold(inode);
Sage Weil3c6f6b72010-02-09 15:24:44 -08001507 if (queue_work(ceph_inode_to_client(inode)->wb_wq,
1508 &ceph_inode(inode)->i_wb_work)) {
Sage Weil2c27c9a2010-02-17 15:45:51 -08001509 dout("ceph_queue_writeback %p\n", inode);
Sage Weil3c6f6b72010-02-09 15:24:44 -08001510 } else {
Sage Weil2c27c9a2010-02-17 15:45:51 -08001511 dout("ceph_queue_writeback %p failed\n", inode);
Sage Weil15a20152011-11-05 22:06:31 -07001512 iput(inode);
Sage Weil3c6f6b72010-02-09 15:24:44 -08001513 }
1514}
1515
1516static void ceph_writeback_work(struct work_struct *work)
Sage Weil355da1e2009-10-06 11:31:08 -07001517{
1518 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1519 i_wb_work);
1520 struct inode *inode = &ci->vfs_inode;
1521
1522 dout("writeback %p\n", inode);
1523 filemap_fdatawrite(&inode->i_data);
1524 iput(inode);
1525}
1526
1527/*
Sage Weil3c6f6b72010-02-09 15:24:44 -08001528 * queue an async invalidation
1529 */
1530void ceph_queue_invalidate(struct inode *inode)
1531{
Sage Weil15a20152011-11-05 22:06:31 -07001532 ihold(inode);
Sage Weil3c6f6b72010-02-09 15:24:44 -08001533 if (queue_work(ceph_inode_to_client(inode)->pg_inv_wq,
1534 &ceph_inode(inode)->i_pg_inv_work)) {
1535 dout("ceph_queue_invalidate %p\n", inode);
Sage Weil3c6f6b72010-02-09 15:24:44 -08001536 } else {
1537 dout("ceph_queue_invalidate %p failed\n", inode);
Sage Weil15a20152011-11-05 22:06:31 -07001538 iput(inode);
Sage Weil3c6f6b72010-02-09 15:24:44 -08001539 }
1540}
1541
1542/*
Sage Weil355da1e2009-10-06 11:31:08 -07001543 * Invalidate inode pages in a worker thread. (This can't be done
1544 * in the message handler context.)
1545 */
Sage Weil3c6f6b72010-02-09 15:24:44 -08001546static void ceph_invalidate_work(struct work_struct *work)
Sage Weil355da1e2009-10-06 11:31:08 -07001547{
1548 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1549 i_pg_inv_work);
1550 struct inode *inode = &ci->vfs_inode;
1551 u32 orig_gen;
1552 int check = 0;
1553
Yan, Zhengb0d7c222013-08-12 21:42:15 -07001554 mutex_lock(&ci->i_truncate_mutex);
Sage Weilbe655592011-11-30 09:47:09 -08001555 spin_lock(&ci->i_ceph_lock);
Sage Weil355da1e2009-10-06 11:31:08 -07001556 dout("invalidate_pages %p gen %d revoking %d\n", inode,
1557 ci->i_rdcache_gen, ci->i_rdcache_revoking);
Sage Weilcd045cb2010-11-04 11:05:05 -07001558 if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
Yan, Zheng9563f882013-11-22 13:50:45 +08001559 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1560 check = 1;
Sage Weilbe655592011-11-30 09:47:09 -08001561 spin_unlock(&ci->i_ceph_lock);
Yan, Zhengb0d7c222013-08-12 21:42:15 -07001562 mutex_unlock(&ci->i_truncate_mutex);
Sage Weil355da1e2009-10-06 11:31:08 -07001563 goto out;
1564 }
1565 orig_gen = ci->i_rdcache_gen;
Sage Weilbe655592011-11-30 09:47:09 -08001566 spin_unlock(&ci->i_ceph_lock);
Sage Weil355da1e2009-10-06 11:31:08 -07001567
Yan, Zheng4e217b52014-06-08 05:08:56 +08001568 truncate_pagecache(inode, 0);
Sage Weil355da1e2009-10-06 11:31:08 -07001569
Sage Weilbe655592011-11-30 09:47:09 -08001570 spin_lock(&ci->i_ceph_lock);
Sage Weilcd045cb2010-11-04 11:05:05 -07001571 if (orig_gen == ci->i_rdcache_gen &&
1572 orig_gen == ci->i_rdcache_revoking) {
Sage Weil355da1e2009-10-06 11:31:08 -07001573 dout("invalidate_pages %p gen %d successful\n", inode,
1574 ci->i_rdcache_gen);
Sage Weilcd045cb2010-11-04 11:05:05 -07001575 ci->i_rdcache_revoking--;
Sage Weil355da1e2009-10-06 11:31:08 -07001576 check = 1;
1577 } else {
Sage Weilcd045cb2010-11-04 11:05:05 -07001578 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1579 inode, orig_gen, ci->i_rdcache_gen,
1580 ci->i_rdcache_revoking);
Yan, Zheng9563f882013-11-22 13:50:45 +08001581 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1582 check = 1;
Sage Weil355da1e2009-10-06 11:31:08 -07001583 }
Sage Weilbe655592011-11-30 09:47:09 -08001584 spin_unlock(&ci->i_ceph_lock);
Yan, Zhengb0d7c222013-08-12 21:42:15 -07001585 mutex_unlock(&ci->i_truncate_mutex);
Yan, Zheng9563f882013-11-22 13:50:45 +08001586out:
Sage Weil355da1e2009-10-06 11:31:08 -07001587 if (check)
1588 ceph_check_caps(ci, 0, NULL);
Sage Weil355da1e2009-10-06 11:31:08 -07001589 iput(inode);
1590}
1591
1592
1593/*
Yan, Zheng3f999692013-03-01 10:57:54 +08001594 * called by trunc_wq;
Sage Weil355da1e2009-10-06 11:31:08 -07001595 *
1596 * We also truncate in a separate thread as well.
1597 */
Sage Weil3c6f6b72010-02-09 15:24:44 -08001598static void ceph_vmtruncate_work(struct work_struct *work)
Sage Weil355da1e2009-10-06 11:31:08 -07001599{
1600 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1601 i_vmtruncate_work);
1602 struct inode *inode = &ci->vfs_inode;
1603
1604 dout("vmtruncate_work %p\n", inode);
Yan, Zhengb415bf42013-07-02 12:40:19 +08001605 __ceph_do_pending_vmtruncate(inode);
Sage Weil355da1e2009-10-06 11:31:08 -07001606 iput(inode);
1607}
1608
1609/*
Sage Weil3c6f6b72010-02-09 15:24:44 -08001610 * Queue an async vmtruncate. If we fail to queue work, we will handle
1611 * the truncation the next time we call __ceph_do_pending_vmtruncate.
1612 */
1613void ceph_queue_vmtruncate(struct inode *inode)
1614{
1615 struct ceph_inode_info *ci = ceph_inode(inode);
1616
Sage Weil15a20152011-11-05 22:06:31 -07001617 ihold(inode);
Milosz Tanski99ccbd22013-08-21 17:29:54 -04001618
Cheng Renquan640ef792010-03-26 17:40:33 +08001619 if (queue_work(ceph_sb_to_client(inode->i_sb)->trunc_wq,
Sage Weil3c6f6b72010-02-09 15:24:44 -08001620 &ci->i_vmtruncate_work)) {
1621 dout("ceph_queue_vmtruncate %p\n", inode);
Sage Weil3c6f6b72010-02-09 15:24:44 -08001622 } else {
1623 dout("ceph_queue_vmtruncate %p failed, pending=%d\n",
1624 inode, ci->i_truncate_pending);
Sage Weil15a20152011-11-05 22:06:31 -07001625 iput(inode);
Sage Weil3c6f6b72010-02-09 15:24:44 -08001626 }
1627}
1628
1629/*
Sage Weil355da1e2009-10-06 11:31:08 -07001630 * Make sure any pending truncation is applied before doing anything
1631 * that may depend on it.
1632 */
Yan, Zhengb415bf42013-07-02 12:40:19 +08001633void __ceph_do_pending_vmtruncate(struct inode *inode)
Sage Weil355da1e2009-10-06 11:31:08 -07001634{
1635 struct ceph_inode_info *ci = ceph_inode(inode);
1636 u64 to;
Yan, Zhenga85f50b2012-11-19 10:49:08 +08001637 int wrbuffer_refs, finish = 0;
Sage Weil355da1e2009-10-06 11:31:08 -07001638
Yan, Zhengb0d7c222013-08-12 21:42:15 -07001639 mutex_lock(&ci->i_truncate_mutex);
Sage Weil355da1e2009-10-06 11:31:08 -07001640retry:
Sage Weilbe655592011-11-30 09:47:09 -08001641 spin_lock(&ci->i_ceph_lock);
Sage Weil355da1e2009-10-06 11:31:08 -07001642 if (ci->i_truncate_pending == 0) {
1643 dout("__do_pending_vmtruncate %p none pending\n", inode);
Sage Weilbe655592011-11-30 09:47:09 -08001644 spin_unlock(&ci->i_ceph_lock);
Yan, Zhengb0d7c222013-08-12 21:42:15 -07001645 mutex_unlock(&ci->i_truncate_mutex);
Sage Weil355da1e2009-10-06 11:31:08 -07001646 return;
1647 }
1648
1649 /*
1650 * make sure any dirty snapped pages are flushed before we
1651 * possibly truncate them.. so write AND block!
1652 */
1653 if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) {
1654 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1655 inode);
Sage Weilbe655592011-11-30 09:47:09 -08001656 spin_unlock(&ci->i_ceph_lock);
Sage Weil355da1e2009-10-06 11:31:08 -07001657 filemap_write_and_wait_range(&inode->i_data, 0,
1658 inode->i_sb->s_maxbytes);
1659 goto retry;
1660 }
1661
Yan, Zhengb0d7c222013-08-12 21:42:15 -07001662 /* there should be no reader or writer */
1663 WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref);
1664
Sage Weil355da1e2009-10-06 11:31:08 -07001665 to = ci->i_truncate_size;
1666 wrbuffer_refs = ci->i_wrbuffer_ref;
1667 dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode,
1668 ci->i_truncate_pending, to);
Sage Weilbe655592011-11-30 09:47:09 -08001669 spin_unlock(&ci->i_ceph_lock);
Sage Weil355da1e2009-10-06 11:31:08 -07001670
Yan, Zheng4e217b52014-06-08 05:08:56 +08001671 truncate_pagecache(inode, to);
Sage Weil355da1e2009-10-06 11:31:08 -07001672
Sage Weilbe655592011-11-30 09:47:09 -08001673 spin_lock(&ci->i_ceph_lock);
Yan, Zhenga85f50b2012-11-19 10:49:08 +08001674 if (to == ci->i_truncate_size) {
1675 ci->i_truncate_pending = 0;
1676 finish = 1;
1677 }
Sage Weilbe655592011-11-30 09:47:09 -08001678 spin_unlock(&ci->i_ceph_lock);
Yan, Zhenga85f50b2012-11-19 10:49:08 +08001679 if (!finish)
1680 goto retry;
Sage Weil355da1e2009-10-06 11:31:08 -07001681
Yan, Zhengb0d7c222013-08-12 21:42:15 -07001682 mutex_unlock(&ci->i_truncate_mutex);
1683
Sage Weil355da1e2009-10-06 11:31:08 -07001684 if (wrbuffer_refs == 0)
1685 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
Yan, Zhenga85f50b2012-11-19 10:49:08 +08001686
1687 wake_up_all(&ci->i_cap_wq);
Sage Weil355da1e2009-10-06 11:31:08 -07001688}
1689
Sage Weil355da1e2009-10-06 11:31:08 -07001690/*
1691 * symlinks
1692 */
1693static void *ceph_sym_follow_link(struct dentry *dentry, struct nameidata *nd)
1694{
1695 struct ceph_inode_info *ci = ceph_inode(dentry->d_inode);
1696 nd_set_link(nd, ci->i_symlink);
1697 return NULL;
1698}
1699
1700static const struct inode_operations ceph_symlink_iops = {
1701 .readlink = generic_readlink,
1702 .follow_link = ceph_sym_follow_link,
Yan, Zheng0b932672013-04-07 16:28:49 +08001703 .setattr = ceph_setattr,
1704 .getattr = ceph_getattr,
1705 .setxattr = ceph_setxattr,
1706 .getxattr = ceph_getxattr,
1707 .listxattr = ceph_listxattr,
1708 .removexattr = ceph_removexattr,
Sage Weil355da1e2009-10-06 11:31:08 -07001709};
1710
1711/*
1712 * setattr
1713 */
1714int ceph_setattr(struct dentry *dentry, struct iattr *attr)
1715{
1716 struct inode *inode = dentry->d_inode;
1717 struct ceph_inode_info *ci = ceph_inode(inode);
Sage Weil355da1e2009-10-06 11:31:08 -07001718 const unsigned int ia_valid = attr->ia_valid;
1719 struct ceph_mds_request *req;
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001720 struct ceph_mds_client *mdsc = ceph_sb_to_client(dentry->d_sb)->mdsc;
Sage Weil355da1e2009-10-06 11:31:08 -07001721 int issued;
1722 int release = 0, dirtied = 0;
1723 int mask = 0;
1724 int err = 0;
Sage Weilfca65b42011-05-04 11:33:47 -07001725 int inode_dirty_flags = 0;
Sage Weil355da1e2009-10-06 11:31:08 -07001726
1727 if (ceph_snap(inode) != CEPH_NOSNAP)
1728 return -EROFS;
1729
Sage Weil355da1e2009-10-06 11:31:08 -07001730 err = inode_change_ok(inode, attr);
1731 if (err != 0)
1732 return err;
1733
1734 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR,
1735 USE_AUTH_MDS);
1736 if (IS_ERR(req))
1737 return PTR_ERR(req);
1738
Sage Weilbe655592011-11-30 09:47:09 -08001739 spin_lock(&ci->i_ceph_lock);
Sage Weil355da1e2009-10-06 11:31:08 -07001740 issued = __ceph_caps_issued(ci, NULL);
1741 dout("setattr %p issued %s\n", inode, ceph_cap_string(issued));
1742
1743 if (ia_valid & ATTR_UID) {
1744 dout("setattr %p uid %d -> %d\n", inode,
Eric W. Biedermanbd2bae62013-01-31 04:05:39 -08001745 from_kuid(&init_user_ns, inode->i_uid),
1746 from_kuid(&init_user_ns, attr->ia_uid));
Sage Weil355da1e2009-10-06 11:31:08 -07001747 if (issued & CEPH_CAP_AUTH_EXCL) {
1748 inode->i_uid = attr->ia_uid;
1749 dirtied |= CEPH_CAP_AUTH_EXCL;
1750 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
Eric W. Biedermanab871b92013-01-31 03:40:12 -08001751 !uid_eq(attr->ia_uid, inode->i_uid)) {
1752 req->r_args.setattr.uid = cpu_to_le32(
1753 from_kuid(&init_user_ns, attr->ia_uid));
Sage Weil355da1e2009-10-06 11:31:08 -07001754 mask |= CEPH_SETATTR_UID;
1755 release |= CEPH_CAP_AUTH_SHARED;
1756 }
1757 }
1758 if (ia_valid & ATTR_GID) {
1759 dout("setattr %p gid %d -> %d\n", inode,
Eric W. Biedermanbd2bae62013-01-31 04:05:39 -08001760 from_kgid(&init_user_ns, inode->i_gid),
1761 from_kgid(&init_user_ns, attr->ia_gid));
Sage Weil355da1e2009-10-06 11:31:08 -07001762 if (issued & CEPH_CAP_AUTH_EXCL) {
1763 inode->i_gid = attr->ia_gid;
1764 dirtied |= CEPH_CAP_AUTH_EXCL;
1765 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
Eric W. Biedermanab871b92013-01-31 03:40:12 -08001766 !gid_eq(attr->ia_gid, inode->i_gid)) {
1767 req->r_args.setattr.gid = cpu_to_le32(
1768 from_kgid(&init_user_ns, attr->ia_gid));
Sage Weil355da1e2009-10-06 11:31:08 -07001769 mask |= CEPH_SETATTR_GID;
1770 release |= CEPH_CAP_AUTH_SHARED;
1771 }
1772 }
1773 if (ia_valid & ATTR_MODE) {
1774 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode,
1775 attr->ia_mode);
1776 if (issued & CEPH_CAP_AUTH_EXCL) {
1777 inode->i_mode = attr->ia_mode;
1778 dirtied |= CEPH_CAP_AUTH_EXCL;
1779 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1780 attr->ia_mode != inode->i_mode) {
Guangliang Zhao7221fe42013-11-11 15:18:03 +08001781 inode->i_mode = attr->ia_mode;
Sage Weil355da1e2009-10-06 11:31:08 -07001782 req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode);
1783 mask |= CEPH_SETATTR_MODE;
1784 release |= CEPH_CAP_AUTH_SHARED;
1785 }
1786 }
1787
1788 if (ia_valid & ATTR_ATIME) {
1789 dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode,
1790 inode->i_atime.tv_sec, inode->i_atime.tv_nsec,
1791 attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec);
1792 if (issued & CEPH_CAP_FILE_EXCL) {
1793 ci->i_time_warp_seq++;
1794 inode->i_atime = attr->ia_atime;
1795 dirtied |= CEPH_CAP_FILE_EXCL;
1796 } else if ((issued & CEPH_CAP_FILE_WR) &&
1797 timespec_compare(&inode->i_atime,
1798 &attr->ia_atime) < 0) {
1799 inode->i_atime = attr->ia_atime;
1800 dirtied |= CEPH_CAP_FILE_WR;
1801 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1802 !timespec_equal(&inode->i_atime, &attr->ia_atime)) {
1803 ceph_encode_timespec(&req->r_args.setattr.atime,
1804 &attr->ia_atime);
1805 mask |= CEPH_SETATTR_ATIME;
1806 release |= CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_RD |
1807 CEPH_CAP_FILE_WR;
1808 }
1809 }
1810 if (ia_valid & ATTR_MTIME) {
1811 dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode,
1812 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
1813 attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec);
1814 if (issued & CEPH_CAP_FILE_EXCL) {
1815 ci->i_time_warp_seq++;
1816 inode->i_mtime = attr->ia_mtime;
1817 dirtied |= CEPH_CAP_FILE_EXCL;
1818 } else if ((issued & CEPH_CAP_FILE_WR) &&
1819 timespec_compare(&inode->i_mtime,
1820 &attr->ia_mtime) < 0) {
1821 inode->i_mtime = attr->ia_mtime;
1822 dirtied |= CEPH_CAP_FILE_WR;
1823 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1824 !timespec_equal(&inode->i_mtime, &attr->ia_mtime)) {
1825 ceph_encode_timespec(&req->r_args.setattr.mtime,
1826 &attr->ia_mtime);
1827 mask |= CEPH_SETATTR_MTIME;
1828 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
1829 CEPH_CAP_FILE_WR;
1830 }
1831 }
1832 if (ia_valid & ATTR_SIZE) {
1833 dout("setattr %p size %lld -> %lld\n", inode,
1834 inode->i_size, attr->ia_size);
Sage Weil355da1e2009-10-06 11:31:08 -07001835 if ((issued & CEPH_CAP_FILE_EXCL) &&
1836 attr->ia_size > inode->i_size) {
1837 inode->i_size = attr->ia_size;
Sage Weil355da1e2009-10-06 11:31:08 -07001838 inode->i_blocks =
1839 (attr->ia_size + (1 << 9) - 1) >> 9;
1840 inode->i_ctime = attr->ia_ctime;
1841 ci->i_reported_size = attr->ia_size;
1842 dirtied |= CEPH_CAP_FILE_EXCL;
1843 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1844 attr->ia_size != inode->i_size) {
1845 req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
1846 req->r_args.setattr.old_size =
1847 cpu_to_le64(inode->i_size);
1848 mask |= CEPH_SETATTR_SIZE;
1849 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
1850 CEPH_CAP_FILE_WR;
1851 }
1852 }
1853
1854 /* these do nothing */
1855 if (ia_valid & ATTR_CTIME) {
1856 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME|
1857 ATTR_MODE|ATTR_UID|ATTR_GID)) == 0;
1858 dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode,
1859 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
1860 attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec,
1861 only ? "ctime only" : "ignored");
1862 inode->i_ctime = attr->ia_ctime;
1863 if (only) {
1864 /*
1865 * if kernel wants to dirty ctime but nothing else,
1866 * we need to choose a cap to dirty under, or do
1867 * a almost-no-op setattr
1868 */
1869 if (issued & CEPH_CAP_AUTH_EXCL)
1870 dirtied |= CEPH_CAP_AUTH_EXCL;
1871 else if (issued & CEPH_CAP_FILE_EXCL)
1872 dirtied |= CEPH_CAP_FILE_EXCL;
1873 else if (issued & CEPH_CAP_XATTR_EXCL)
1874 dirtied |= CEPH_CAP_XATTR_EXCL;
1875 else
1876 mask |= CEPH_SETATTR_CTIME;
1877 }
1878 }
1879 if (ia_valid & ATTR_FILE)
1880 dout("setattr %p ATTR_FILE ... hrm!\n", inode);
1881
1882 if (dirtied) {
Sage Weilfca65b42011-05-04 11:33:47 -07001883 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied);
Sage Weil355da1e2009-10-06 11:31:08 -07001884 inode->i_ctime = CURRENT_TIME;
1885 }
1886
1887 release &= issued;
Sage Weilbe655592011-11-30 09:47:09 -08001888 spin_unlock(&ci->i_ceph_lock);
Sage Weil355da1e2009-10-06 11:31:08 -07001889
Sage Weilfca65b42011-05-04 11:33:47 -07001890 if (inode_dirty_flags)
1891 __mark_inode_dirty(inode, inode_dirty_flags);
1892
Guangliang Zhao7221fe42013-11-11 15:18:03 +08001893 if (ia_valid & ATTR_MODE) {
Linus Torvalds4db658e2014-01-28 18:06:18 -08001894 err = posix_acl_chmod(inode, attr->ia_mode);
Guangliang Zhao7221fe42013-11-11 15:18:03 +08001895 if (err)
1896 goto out_put;
1897 }
1898
Sage Weil355da1e2009-10-06 11:31:08 -07001899 if (mask) {
Sage Weil70b666c2011-05-27 09:24:26 -07001900 req->r_inode = inode;
1901 ihold(inode);
Sage Weil355da1e2009-10-06 11:31:08 -07001902 req->r_inode_drop = release;
1903 req->r_args.setattr.mask = cpu_to_le32(mask);
1904 req->r_num_caps = 1;
Sage Weil752c8bd2013-02-05 13:52:29 -08001905 err = ceph_mdsc_do_request(mdsc, NULL, req);
Sage Weil355da1e2009-10-06 11:31:08 -07001906 }
1907 dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err,
1908 ceph_cap_string(dirtied), mask);
1909
1910 ceph_mdsc_put_request(req);
Yan, Zhengb0d7c222013-08-12 21:42:15 -07001911 if (mask & CEPH_SETATTR_SIZE)
1912 __ceph_do_pending_vmtruncate(inode);
Sage Weil355da1e2009-10-06 11:31:08 -07001913 return err;
Guangliang Zhao7221fe42013-11-11 15:18:03 +08001914out_put:
Sage Weil355da1e2009-10-06 11:31:08 -07001915 ceph_mdsc_put_request(req);
1916 return err;
1917}
1918
1919/*
1920 * Verify that we have a lease on the given mask. If not,
1921 * do a getattr against an mds.
1922 */
Yan, Zheng01deead2014-11-14 21:56:29 +08001923int __ceph_do_getattr(struct inode *inode, struct page *locked_page,
1924 int mask, bool force)
Sage Weil355da1e2009-10-06 11:31:08 -07001925{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001926 struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
1927 struct ceph_mds_client *mdsc = fsc->mdsc;
Sage Weil355da1e2009-10-06 11:31:08 -07001928 struct ceph_mds_request *req;
1929 int err;
1930
1931 if (ceph_snap(inode) == CEPH_SNAPDIR) {
1932 dout("do_getattr inode %p SNAPDIR\n", inode);
1933 return 0;
1934 }
1935
Yan, Zheng01deead2014-11-14 21:56:29 +08001936 dout("do_getattr inode %p mask %s mode 0%o\n",
1937 inode, ceph_cap_string(mask), inode->i_mode);
Yan, Zheng508b32d2014-09-16 21:46:17 +08001938 if (!force && ceph_caps_issued_mask(ceph_inode(inode), mask, 1))
Sage Weil355da1e2009-10-06 11:31:08 -07001939 return 0;
1940
1941 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
1942 if (IS_ERR(req))
1943 return PTR_ERR(req);
Sage Weil70b666c2011-05-27 09:24:26 -07001944 req->r_inode = inode;
1945 ihold(inode);
Sage Weil355da1e2009-10-06 11:31:08 -07001946 req->r_num_caps = 1;
1947 req->r_args.getattr.mask = cpu_to_le32(mask);
Yan, Zheng01deead2014-11-14 21:56:29 +08001948 req->r_locked_page = locked_page;
Sage Weil355da1e2009-10-06 11:31:08 -07001949 err = ceph_mdsc_do_request(mdsc, NULL, req);
Yan, Zheng01deead2014-11-14 21:56:29 +08001950 if (locked_page && err == 0) {
1951 u64 inline_version = req->r_reply_info.targeti.inline_version;
1952 if (inline_version == 0) {
1953 /* the reply is supposed to contain inline data */
1954 err = -EINVAL;
1955 } else if (inline_version == CEPH_INLINE_NONE) {
1956 err = -ENODATA;
1957 } else {
1958 err = req->r_reply_info.targeti.inline_len;
1959 }
1960 }
Sage Weil355da1e2009-10-06 11:31:08 -07001961 ceph_mdsc_put_request(req);
1962 dout("do_getattr result=%d\n", err);
1963 return err;
1964}
1965
1966
1967/*
1968 * Check inode permissions. We verify we have a valid value for
1969 * the AUTH cap, then call the generic handler.
1970 */
Al Viro10556cb2011-06-20 19:28:19 -04001971int ceph_permission(struct inode *inode, int mask)
Sage Weil355da1e2009-10-06 11:31:08 -07001972{
Nick Pigginb74c79e2011-01-07 17:49:58 +11001973 int err;
1974
Al Viro10556cb2011-06-20 19:28:19 -04001975 if (mask & MAY_NOT_BLOCK)
Nick Pigginb74c79e2011-01-07 17:49:58 +11001976 return -ECHILD;
1977
Yan, Zheng508b32d2014-09-16 21:46:17 +08001978 err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED, false);
Sage Weil355da1e2009-10-06 11:31:08 -07001979
1980 if (!err)
Al Viro2830ba72011-06-20 19:16:29 -04001981 err = generic_permission(inode, mask);
Sage Weil355da1e2009-10-06 11:31:08 -07001982 return err;
1983}
1984
1985/*
1986 * Get all attributes. Hopefully somedata we'll have a statlite()
1987 * and can limit the fields we require to be accurate.
1988 */
1989int ceph_getattr(struct vfsmount *mnt, struct dentry *dentry,
1990 struct kstat *stat)
1991{
1992 struct inode *inode = dentry->d_inode;
Sage Weil232d4b02009-10-21 11:21:49 -07001993 struct ceph_inode_info *ci = ceph_inode(inode);
Sage Weil355da1e2009-10-06 11:31:08 -07001994 int err;
1995
Yan, Zheng508b32d2014-09-16 21:46:17 +08001996 err = ceph_do_getattr(inode, CEPH_STAT_CAP_INODE_ALL, false);
Sage Weil355da1e2009-10-06 11:31:08 -07001997 if (!err) {
1998 generic_fillattr(inode, stat);
Yehuda Sadehad1fee92011-01-21 16:44:03 -08001999 stat->ino = ceph_translate_ino(inode->i_sb, inode->i_ino);
Sage Weil355da1e2009-10-06 11:31:08 -07002000 if (ceph_snap(inode) != CEPH_NOSNAP)
2001 stat->dev = ceph_snap(inode);
2002 else
2003 stat->dev = 0;
Sage Weil232d4b02009-10-21 11:21:49 -07002004 if (S_ISDIR(inode->i_mode)) {
Yehuda Sadeh1c1266b2011-01-12 16:53:27 -08002005 if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb),
2006 RBYTES))
2007 stat->size = ci->i_rbytes;
2008 else
2009 stat->size = ci->i_files + ci->i_subdirs;
Sage Weil232d4b02009-10-21 11:21:49 -07002010 stat->blocks = 0;
Sage Weil355da1e2009-10-06 11:31:08 -07002011 stat->blksize = 65536;
Sage Weil232d4b02009-10-21 11:21:49 -07002012 }
Sage Weil355da1e2009-10-06 11:31:08 -07002013 }
2014 return err;
2015}