]>
Commit | Line | Data |
---|---|---|
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> | |
355da1eb SW |
9 | #include <linux/writeback.h> |
10 | #include <linux/vmalloc.h> | |
4db658ea | 11 | #include <linux/posix_acl.h> |
3e7fbe9c | 12 | #include <linux/random.h> |
a407846e | 13 | #include <linux/sort.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 | ||
32 | static const struct inode_operations ceph_symlink_iops; | |
33 | ||
3c6f6b79 SW |
34 | static void ceph_invalidate_work(struct work_struct *work); |
35 | static void ceph_writeback_work(struct work_struct *work); | |
36 | static 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 |
41 | static 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 |
48 | struct 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 | */ | |
70 | struct 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 | ||
92 | const 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 | */ | |
116 | static 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 | */ | |
160 | struct 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 |
183 | static 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 |
225 | u32 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 | */ | |
240 | static 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 | ||
42172119 YZ |
257 | if (mds == -1) /* CDIR_AUTH_PARENT */ |
258 | mds = diri_auth; | |
259 | ||
355da1eb | 260 | mutex_lock(&ci->i_fragtree_mutex); |
8d08503c | 261 | if (ndist == 0 && mds == diri_auth) { |
355da1eb SW |
262 | /* no delegation info needed. */ |
263 | frag = __ceph_find_frag(ci, id); | |
264 | if (!frag) | |
265 | goto out; | |
266 | if (frag->split_by == 0) { | |
267 | /* tree leaf, remove */ | |
268 | dout("fill_dirfrag removed %llx.%llx frag %x" | |
269 | " (no ref)\n", ceph_vinop(inode), id); | |
270 | rb_erase(&frag->node, &ci->i_fragtree); | |
271 | kfree(frag); | |
272 | } else { | |
273 | /* tree branch, keep and clear */ | |
274 | dout("fill_dirfrag cleared %llx.%llx frag %x" | |
275 | " referral\n", ceph_vinop(inode), id); | |
276 | frag->mds = -1; | |
277 | frag->ndist = 0; | |
278 | } | |
279 | goto out; | |
280 | } | |
281 | ||
282 | ||
283 | /* find/add this frag to store mds delegation info */ | |
284 | frag = __get_or_create_frag(ci, id); | |
285 | if (IS_ERR(frag)) { | |
286 | /* this is not the end of the world; we can continue | |
287 | with bad/inaccurate delegation info */ | |
288 | pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n", | |
289 | ceph_vinop(inode), le32_to_cpu(dirinfo->frag)); | |
290 | err = -ENOMEM; | |
291 | goto out; | |
292 | } | |
293 | ||
294 | frag->mds = mds; | |
295 | frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP); | |
296 | for (i = 0; i < frag->ndist; i++) | |
297 | frag->dist[i] = le32_to_cpu(dirinfo->dist[i]); | |
298 | dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n", | |
299 | ceph_vinop(inode), frag->frag, frag->ndist); | |
300 | ||
301 | out: | |
302 | mutex_unlock(&ci->i_fragtree_mutex); | |
303 | return err; | |
304 | } | |
305 | ||
a407846e YZ |
306 | static int frag_tree_split_cmp(const void *l, const void *r) |
307 | { | |
308 | struct ceph_frag_tree_split *ls = (struct ceph_frag_tree_split*)l; | |
309 | struct ceph_frag_tree_split *rs = (struct ceph_frag_tree_split*)r; | |
310 | return ceph_frag_compare(ls->frag, rs->frag); | |
311 | } | |
312 | ||
3e7fbe9c YZ |
313 | static int ceph_fill_fragtree(struct inode *inode, |
314 | struct ceph_frag_tree_head *fragtree, | |
315 | struct ceph_mds_reply_dirfrag *dirinfo) | |
316 | { | |
317 | struct ceph_inode_info *ci = ceph_inode(inode); | |
318 | struct ceph_inode_frag *frag; | |
319 | struct rb_node *rb_node; | |
320 | int i; | |
321 | u32 id, nsplits; | |
322 | bool update = false; | |
323 | ||
324 | mutex_lock(&ci->i_fragtree_mutex); | |
325 | nsplits = le32_to_cpu(fragtree->nsplits); | |
326 | if (nsplits) { | |
327 | i = prandom_u32() % nsplits; | |
328 | id = le32_to_cpu(fragtree->splits[i].frag); | |
329 | if (!__ceph_find_frag(ci, id)) | |
330 | update = true; | |
331 | } else if (!RB_EMPTY_ROOT(&ci->i_fragtree)) { | |
332 | rb_node = rb_first(&ci->i_fragtree); | |
333 | frag = rb_entry(rb_node, struct ceph_inode_frag, node); | |
334 | if (frag->frag != ceph_frag_make(0, 0) || rb_next(rb_node)) | |
335 | update = true; | |
336 | } | |
337 | if (!update && dirinfo) { | |
338 | id = le32_to_cpu(dirinfo->frag); | |
339 | if (id != __ceph_choose_frag(ci, id, NULL, NULL)) | |
340 | update = true; | |
341 | } | |
342 | if (!update) | |
343 | goto out_unlock; | |
344 | ||
a407846e YZ |
345 | if (nsplits > 1) { |
346 | sort(fragtree->splits, nsplits, sizeof(fragtree->splits[0]), | |
347 | frag_tree_split_cmp, NULL); | |
348 | } | |
349 | ||
3e7fbe9c YZ |
350 | dout("fill_fragtree %llx.%llx\n", ceph_vinop(inode)); |
351 | rb_node = rb_first(&ci->i_fragtree); | |
352 | for (i = 0; i < nsplits; i++) { | |
353 | id = le32_to_cpu(fragtree->splits[i].frag); | |
354 | frag = NULL; | |
355 | while (rb_node) { | |
356 | frag = rb_entry(rb_node, struct ceph_inode_frag, node); | |
357 | if (ceph_frag_compare(frag->frag, id) >= 0) { | |
358 | if (frag->frag != id) | |
359 | frag = NULL; | |
360 | else | |
361 | rb_node = rb_next(rb_node); | |
362 | break; | |
363 | } | |
364 | rb_node = rb_next(rb_node); | |
365 | rb_erase(&frag->node, &ci->i_fragtree); | |
366 | kfree(frag); | |
367 | frag = NULL; | |
368 | } | |
369 | if (!frag) { | |
370 | frag = __get_or_create_frag(ci, id); | |
371 | if (IS_ERR(frag)) | |
372 | continue; | |
373 | } | |
374 | frag->split_by = le32_to_cpu(fragtree->splits[i].by); | |
375 | dout(" frag %x split by %d\n", frag->frag, frag->split_by); | |
376 | } | |
377 | while (rb_node) { | |
378 | frag = rb_entry(rb_node, struct ceph_inode_frag, node); | |
379 | rb_node = rb_next(rb_node); | |
380 | rb_erase(&frag->node, &ci->i_fragtree); | |
381 | kfree(frag); | |
382 | } | |
383 | out_unlock: | |
384 | mutex_unlock(&ci->i_fragtree_mutex); | |
385 | return 0; | |
386 | } | |
355da1eb SW |
387 | |
388 | /* | |
389 | * initialize a newly allocated inode. | |
390 | */ | |
391 | struct inode *ceph_alloc_inode(struct super_block *sb) | |
392 | { | |
393 | struct ceph_inode_info *ci; | |
394 | int i; | |
395 | ||
396 | ci = kmem_cache_alloc(ceph_inode_cachep, GFP_NOFS); | |
397 | if (!ci) | |
398 | return NULL; | |
399 | ||
400 | dout("alloc_inode %p\n", &ci->vfs_inode); | |
401 | ||
be655596 SW |
402 | spin_lock_init(&ci->i_ceph_lock); |
403 | ||
355da1eb | 404 | ci->i_version = 0; |
31c542a1 | 405 | ci->i_inline_version = 0; |
355da1eb SW |
406 | ci->i_time_warp_seq = 0; |
407 | ci->i_ceph_flags = 0; | |
fdd4e158 YZ |
408 | atomic64_set(&ci->i_ordered_count, 1); |
409 | atomic64_set(&ci->i_release_count, 1); | |
410 | atomic64_set(&ci->i_complete_seq[0], 0); | |
411 | atomic64_set(&ci->i_complete_seq[1], 0); | |
355da1eb SW |
412 | ci->i_symlink = NULL; |
413 | ||
6c0f3af7 | 414 | memset(&ci->i_dir_layout, 0, sizeof(ci->i_dir_layout)); |
5ea5c5e0 | 415 | ci->i_pool_ns_len = 0; |
6c0f3af7 | 416 | |
355da1eb SW |
417 | ci->i_fragtree = RB_ROOT; |
418 | mutex_init(&ci->i_fragtree_mutex); | |
419 | ||
420 | ci->i_xattrs.blob = NULL; | |
421 | ci->i_xattrs.prealloc_blob = NULL; | |
422 | ci->i_xattrs.dirty = false; | |
423 | ci->i_xattrs.index = RB_ROOT; | |
424 | ci->i_xattrs.count = 0; | |
425 | ci->i_xattrs.names_size = 0; | |
426 | ci->i_xattrs.vals_size = 0; | |
427 | ci->i_xattrs.version = 0; | |
428 | ci->i_xattrs.index_version = 0; | |
429 | ||
430 | ci->i_caps = RB_ROOT; | |
431 | ci->i_auth_cap = NULL; | |
432 | ci->i_dirty_caps = 0; | |
433 | ci->i_flushing_caps = 0; | |
434 | INIT_LIST_HEAD(&ci->i_dirty_item); | |
435 | INIT_LIST_HEAD(&ci->i_flushing_item); | |
f66fd9f0 | 436 | ci->i_prealloc_cap_flush = NULL; |
553adfd9 | 437 | ci->i_cap_flush_tree = RB_ROOT; |
355da1eb SW |
438 | init_waitqueue_head(&ci->i_cap_wq); |
439 | ci->i_hold_caps_min = 0; | |
440 | ci->i_hold_caps_max = 0; | |
441 | INIT_LIST_HEAD(&ci->i_cap_delay_list); | |
355da1eb SW |
442 | INIT_LIST_HEAD(&ci->i_cap_snaps); |
443 | ci->i_head_snapc = NULL; | |
444 | ci->i_snap_caps = 0; | |
445 | ||
446 | for (i = 0; i < CEPH_FILE_MODE_NUM; i++) | |
447 | ci->i_nr_by_mode[i] = 0; | |
448 | ||
b0d7c223 | 449 | mutex_init(&ci->i_truncate_mutex); |
355da1eb SW |
450 | ci->i_truncate_seq = 0; |
451 | ci->i_truncate_size = 0; | |
452 | ci->i_truncate_pending = 0; | |
453 | ||
454 | ci->i_max_size = 0; | |
455 | ci->i_reported_size = 0; | |
456 | ci->i_wanted_max_size = 0; | |
457 | ci->i_requested_max_size = 0; | |
458 | ||
459 | ci->i_pin_ref = 0; | |
460 | ci->i_rd_ref = 0; | |
461 | ci->i_rdcache_ref = 0; | |
462 | ci->i_wr_ref = 0; | |
d3d0720d | 463 | ci->i_wb_ref = 0; |
355da1eb SW |
464 | ci->i_wrbuffer_ref = 0; |
465 | ci->i_wrbuffer_ref_head = 0; | |
466 | ci->i_shared_gen = 0; | |
467 | ci->i_rdcache_gen = 0; | |
468 | ci->i_rdcache_revoking = 0; | |
469 | ||
470 | INIT_LIST_HEAD(&ci->i_unsafe_writes); | |
471 | INIT_LIST_HEAD(&ci->i_unsafe_dirops); | |
68cd5b4b | 472 | INIT_LIST_HEAD(&ci->i_unsafe_iops); |
355da1eb SW |
473 | spin_lock_init(&ci->i_unsafe_lock); |
474 | ||
475 | ci->i_snap_realm = NULL; | |
476 | INIT_LIST_HEAD(&ci->i_snap_realm_item); | |
477 | INIT_LIST_HEAD(&ci->i_snap_flush_item); | |
478 | ||
3c6f6b79 SW |
479 | INIT_WORK(&ci->i_wb_work, ceph_writeback_work); |
480 | INIT_WORK(&ci->i_pg_inv_work, ceph_invalidate_work); | |
355da1eb SW |
481 | |
482 | INIT_WORK(&ci->i_vmtruncate_work, ceph_vmtruncate_work); | |
483 | ||
99ccbd22 MT |
484 | ceph_fscache_inode_init(ci); |
485 | ||
355da1eb SW |
486 | return &ci->vfs_inode; |
487 | } | |
488 | ||
fa0d7e3d NP |
489 | static void ceph_i_callback(struct rcu_head *head) |
490 | { | |
491 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
492 | struct ceph_inode_info *ci = ceph_inode(inode); | |
493 | ||
fa0d7e3d NP |
494 | kmem_cache_free(ceph_inode_cachep, ci); |
495 | } | |
496 | ||
355da1eb SW |
497 | void ceph_destroy_inode(struct inode *inode) |
498 | { | |
499 | struct ceph_inode_info *ci = ceph_inode(inode); | |
500 | struct ceph_inode_frag *frag; | |
501 | struct rb_node *n; | |
502 | ||
503 | dout("destroy_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode)); | |
504 | ||
99ccbd22 MT |
505 | ceph_fscache_unregister_inode_cookie(ci); |
506 | ||
355da1eb SW |
507 | ceph_queue_caps_release(inode); |
508 | ||
8b218b8a SW |
509 | /* |
510 | * we may still have a snap_realm reference if there are stray | |
d9df2783 | 511 | * caps in i_snap_caps. |
8b218b8a SW |
512 | */ |
513 | if (ci->i_snap_realm) { | |
514 | struct ceph_mds_client *mdsc = | |
3d14c5d2 | 515 | ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc; |
8b218b8a SW |
516 | struct ceph_snap_realm *realm = ci->i_snap_realm; |
517 | ||
518 | dout(" dropping residual ref to snap realm %p\n", realm); | |
519 | spin_lock(&realm->inodes_with_caps_lock); | |
520 | list_del_init(&ci->i_snap_realm_item); | |
521 | spin_unlock(&realm->inodes_with_caps_lock); | |
522 | ceph_put_snap_realm(mdsc, realm); | |
523 | } | |
524 | ||
355da1eb SW |
525 | kfree(ci->i_symlink); |
526 | while ((n = rb_first(&ci->i_fragtree)) != NULL) { | |
527 | frag = rb_entry(n, struct ceph_inode_frag, node); | |
528 | rb_erase(n, &ci->i_fragtree); | |
529 | kfree(frag); | |
530 | } | |
531 | ||
532 | __ceph_destroy_xattrs(ci); | |
b6c1d5b8 SW |
533 | if (ci->i_xattrs.blob) |
534 | ceph_buffer_put(ci->i_xattrs.blob); | |
535 | if (ci->i_xattrs.prealloc_blob) | |
536 | ceph_buffer_put(ci->i_xattrs.prealloc_blob); | |
355da1eb | 537 | |
fa0d7e3d | 538 | call_rcu(&inode->i_rcu, ceph_i_callback); |
355da1eb SW |
539 | } |
540 | ||
9f12bd11 YZ |
541 | int ceph_drop_inode(struct inode *inode) |
542 | { | |
543 | /* | |
544 | * Positve dentry and corresponding inode are always accompanied | |
545 | * in MDS reply. So no need to keep inode in the cache after | |
546 | * dropping all its aliases. | |
547 | */ | |
548 | return 1; | |
549 | } | |
550 | ||
355da1eb SW |
551 | /* |
552 | * Helpers to fill in size, ctime, mtime, and atime. We have to be | |
553 | * careful because either the client or MDS may have more up to date | |
554 | * info, depending on which capabilities are held, and whether | |
555 | * time_warp_seq or truncate_seq have increased. (Ordinarily, mtime | |
556 | * and size are monotonically increasing, except when utimes() or | |
557 | * truncate() increments the corresponding _seq values.) | |
558 | */ | |
559 | int ceph_fill_file_size(struct inode *inode, int issued, | |
560 | u32 truncate_seq, u64 truncate_size, u64 size) | |
561 | { | |
562 | struct ceph_inode_info *ci = ceph_inode(inode); | |
563 | int queue_trunc = 0; | |
564 | ||
565 | if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 || | |
566 | (truncate_seq == ci->i_truncate_seq && size > inode->i_size)) { | |
567 | dout("size %lld -> %llu\n", inode->i_size, size); | |
a3d714c3 YZ |
568 | if (size > 0 && S_ISDIR(inode->i_mode)) { |
569 | pr_err("fill_file_size non-zero size for directory\n"); | |
570 | size = 0; | |
571 | } | |
99c88e69 | 572 | i_size_write(inode, size); |
355da1eb SW |
573 | inode->i_blocks = (size + (1<<9) - 1) >> 9; |
574 | ci->i_reported_size = size; | |
575 | if (truncate_seq != ci->i_truncate_seq) { | |
576 | dout("truncate_seq %u -> %u\n", | |
577 | ci->i_truncate_seq, truncate_seq); | |
578 | ci->i_truncate_seq = truncate_seq; | |
b0d7c223 YZ |
579 | |
580 | /* the MDS should have revoked these caps */ | |
581 | WARN_ON_ONCE(issued & (CEPH_CAP_FILE_EXCL | | |
582 | CEPH_CAP_FILE_RD | | |
583 | CEPH_CAP_FILE_WR | | |
584 | CEPH_CAP_FILE_LAZYIO)); | |
3d497d85 YS |
585 | /* |
586 | * If we hold relevant caps, or in the case where we're | |
587 | * not the only client referencing this file and we | |
588 | * don't hold those caps, then we need to check whether | |
589 | * the file is either opened or mmaped | |
590 | */ | |
b0d7c223 YZ |
591 | if ((issued & (CEPH_CAP_FILE_CACHE| |
592 | CEPH_CAP_FILE_BUFFER)) || | |
3d497d85 YS |
593 | mapping_mapped(inode->i_mapping) || |
594 | __ceph_caps_file_wanted(ci)) { | |
355da1eb SW |
595 | ci->i_truncate_pending++; |
596 | queue_trunc = 1; | |
597 | } | |
598 | } | |
599 | } | |
600 | if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 && | |
601 | ci->i_truncate_size != truncate_size) { | |
602 | dout("truncate_size %lld -> %llu\n", ci->i_truncate_size, | |
603 | truncate_size); | |
604 | ci->i_truncate_size = truncate_size; | |
605 | } | |
99ccbd22 MT |
606 | |
607 | if (queue_trunc) | |
608 | ceph_fscache_invalidate(inode); | |
609 | ||
355da1eb SW |
610 | return queue_trunc; |
611 | } | |
612 | ||
613 | void ceph_fill_file_time(struct inode *inode, int issued, | |
614 | u64 time_warp_seq, struct timespec *ctime, | |
615 | struct timespec *mtime, struct timespec *atime) | |
616 | { | |
617 | struct ceph_inode_info *ci = ceph_inode(inode); | |
618 | int warn = 0; | |
619 | ||
620 | if (issued & (CEPH_CAP_FILE_EXCL| | |
621 | CEPH_CAP_FILE_WR| | |
d8672d64 SW |
622 | CEPH_CAP_FILE_BUFFER| |
623 | CEPH_CAP_AUTH_EXCL| | |
624 | CEPH_CAP_XATTR_EXCL)) { | |
355da1eb SW |
625 | if (timespec_compare(ctime, &inode->i_ctime) > 0) { |
626 | dout("ctime %ld.%09ld -> %ld.%09ld inc w/ cap\n", | |
627 | inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec, | |
628 | ctime->tv_sec, ctime->tv_nsec); | |
629 | inode->i_ctime = *ctime; | |
630 | } | |
631 | if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) { | |
632 | /* the MDS did a utimes() */ | |
633 | dout("mtime %ld.%09ld -> %ld.%09ld " | |
634 | "tw %d -> %d\n", | |
635 | inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec, | |
636 | mtime->tv_sec, mtime->tv_nsec, | |
637 | ci->i_time_warp_seq, (int)time_warp_seq); | |
638 | ||
639 | inode->i_mtime = *mtime; | |
640 | inode->i_atime = *atime; | |
641 | ci->i_time_warp_seq = time_warp_seq; | |
642 | } else if (time_warp_seq == ci->i_time_warp_seq) { | |
643 | /* nobody did utimes(); take the max */ | |
644 | if (timespec_compare(mtime, &inode->i_mtime) > 0) { | |
645 | dout("mtime %ld.%09ld -> %ld.%09ld inc\n", | |
646 | inode->i_mtime.tv_sec, | |
647 | inode->i_mtime.tv_nsec, | |
648 | mtime->tv_sec, mtime->tv_nsec); | |
649 | inode->i_mtime = *mtime; | |
650 | } | |
651 | if (timespec_compare(atime, &inode->i_atime) > 0) { | |
652 | dout("atime %ld.%09ld -> %ld.%09ld inc\n", | |
653 | inode->i_atime.tv_sec, | |
654 | inode->i_atime.tv_nsec, | |
655 | atime->tv_sec, atime->tv_nsec); | |
656 | inode->i_atime = *atime; | |
657 | } | |
658 | } else if (issued & CEPH_CAP_FILE_EXCL) { | |
659 | /* we did a utimes(); ignore mds values */ | |
660 | } else { | |
661 | warn = 1; | |
662 | } | |
663 | } else { | |
d8672d64 | 664 | /* we have no write|excl caps; whatever the MDS says is true */ |
355da1eb SW |
665 | if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) { |
666 | inode->i_ctime = *ctime; | |
667 | inode->i_mtime = *mtime; | |
668 | inode->i_atime = *atime; | |
669 | ci->i_time_warp_seq = time_warp_seq; | |
670 | } else { | |
671 | warn = 1; | |
672 | } | |
673 | } | |
674 | if (warn) /* time_warp_seq shouldn't go backwards */ | |
675 | dout("%p mds time_warp_seq %llu < %u\n", | |
676 | inode, time_warp_seq, ci->i_time_warp_seq); | |
677 | } | |
678 | ||
679 | /* | |
680 | * Populate an inode based on info from mds. May be called on new or | |
681 | * existing inodes. | |
682 | */ | |
01deead0 | 683 | static int fill_inode(struct inode *inode, struct page *locked_page, |
355da1eb SW |
684 | struct ceph_mds_reply_info_in *iinfo, |
685 | struct ceph_mds_reply_dirfrag *dirinfo, | |
686 | struct ceph_mds_session *session, | |
687 | unsigned long ttl_from, int cap_fmode, | |
688 | struct ceph_cap_reservation *caps_reservation) | |
689 | { | |
d9df2783 | 690 | struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc; |
355da1eb SW |
691 | struct ceph_mds_reply_inode *info = iinfo->in; |
692 | struct ceph_inode_info *ci = ceph_inode(inode); | |
f98a128a | 693 | int issued = 0, implemented, new_issued; |
355da1eb | 694 | struct timespec mtime, atime, ctime; |
355da1eb | 695 | struct ceph_buffer *xattr_blob = NULL; |
d9df2783 | 696 | struct ceph_cap *new_cap = NULL; |
355da1eb | 697 | int err = 0; |
d9df2783 | 698 | bool wake = false; |
f98a128a YZ |
699 | bool queue_trunc = false; |
700 | bool new_version = false; | |
31c542a1 | 701 | bool fill_inline = false; |
355da1eb SW |
702 | |
703 | dout("fill_inode %p ino %llx.%llx v %llu had %llu\n", | |
704 | inode, ceph_vinop(inode), le64_to_cpu(info->version), | |
705 | ci->i_version); | |
706 | ||
d9df2783 YZ |
707 | /* prealloc new cap struct */ |
708 | if (info->cap.caps && ceph_snap(inode) == CEPH_NOSNAP) | |
709 | new_cap = ceph_get_cap(mdsc, caps_reservation); | |
710 | ||
355da1eb SW |
711 | /* |
712 | * prealloc xattr data, if it looks like we'll need it. only | |
713 | * if len > 4 (meaning there are actually xattrs; the first 4 | |
714 | * bytes are the xattr count). | |
715 | */ | |
716 | if (iinfo->xattr_len > 4) { | |
b6c1d5b8 | 717 | xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS); |
355da1eb SW |
718 | if (!xattr_blob) |
719 | pr_err("fill_inode ENOMEM xattr blob %d bytes\n", | |
720 | iinfo->xattr_len); | |
721 | } | |
722 | ||
be655596 | 723 | spin_lock(&ci->i_ceph_lock); |
355da1eb SW |
724 | |
725 | /* | |
726 | * provided version will be odd if inode value is projected, | |
8bd59e01 SW |
727 | * even if stable. skip the update if we have newer stable |
728 | * info (ours>=theirs, e.g. due to racing mds replies), unless | |
729 | * we are getting projected (unstable) info (in which case the | |
730 | * version is odd, and we want ours>theirs). | |
731 | * us them | |
732 | * 2 2 skip | |
733 | * 3 2 skip | |
734 | * 3 3 update | |
355da1eb | 735 | */ |
f98a128a YZ |
736 | if (ci->i_version == 0 || |
737 | ((info->cap.flags & CEPH_CAP_FLAG_AUTH) && | |
738 | le64_to_cpu(info->version) > (ci->i_version & ~1))) | |
739 | new_version = true; | |
740 | ||
355da1eb SW |
741 | issued = __ceph_caps_issued(ci, &implemented); |
742 | issued |= implemented | __ceph_caps_dirty(ci); | |
f98a128a | 743 | new_issued = ~issued & le32_to_cpu(info->cap.caps); |
355da1eb SW |
744 | |
745 | /* update inode */ | |
746 | ci->i_version = le64_to_cpu(info->version); | |
747 | inode->i_version++; | |
748 | inode->i_rdev = le32_to_cpu(info->rdev); | |
f98a128a | 749 | inode->i_blkbits = fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1; |
355da1eb | 750 | |
f98a128a YZ |
751 | if ((new_version || (new_issued & CEPH_CAP_AUTH_SHARED)) && |
752 | (issued & CEPH_CAP_AUTH_EXCL) == 0) { | |
355da1eb | 753 | inode->i_mode = le32_to_cpu(info->mode); |
ab871b90 EB |
754 | inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(info->uid)); |
755 | inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(info->gid)); | |
355da1eb | 756 | dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode, |
bd2bae6a EB |
757 | from_kuid(&init_user_ns, inode->i_uid), |
758 | from_kgid(&init_user_ns, inode->i_gid)); | |
355da1eb SW |
759 | } |
760 | ||
f98a128a YZ |
761 | if ((new_version || (new_issued & CEPH_CAP_LINK_SHARED)) && |
762 | (issued & CEPH_CAP_LINK_EXCL) == 0) | |
bfe86848 | 763 | set_nlink(inode, le32_to_cpu(info->nlink)); |
355da1eb | 764 | |
f98a128a YZ |
765 | if (new_version || (new_issued & CEPH_CAP_ANY_RD)) { |
766 | /* be careful with mtime, atime, size */ | |
767 | ceph_decode_timespec(&atime, &info->atime); | |
768 | ceph_decode_timespec(&mtime, &info->mtime); | |
769 | ceph_decode_timespec(&ctime, &info->ctime); | |
770 | ceph_fill_file_time(inode, issued, | |
771 | le32_to_cpu(info->time_warp_seq), | |
772 | &ctime, &mtime, &atime); | |
773 | } | |
774 | ||
775 | if (new_version || | |
776 | (new_issued & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR))) { | |
10183a69 YZ |
777 | if (ci->i_layout.fl_pg_pool != info->layout.fl_pg_pool) |
778 | ci->i_ceph_flags &= ~CEPH_I_POOL_PERM; | |
f98a128a | 779 | ci->i_layout = info->layout; |
5ea5c5e0 | 780 | ci->i_pool_ns_len = iinfo->pool_ns_len; |
10183a69 | 781 | |
f98a128a YZ |
782 | queue_trunc = ceph_fill_file_size(inode, issued, |
783 | le32_to_cpu(info->truncate_seq), | |
784 | le64_to_cpu(info->truncate_size), | |
785 | le64_to_cpu(info->size)); | |
786 | /* only update max_size on auth cap */ | |
787 | if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) && | |
788 | ci->i_max_size != le64_to_cpu(info->max_size)) { | |
789 | dout("max_size %lld -> %llu\n", ci->i_max_size, | |
790 | le64_to_cpu(info->max_size)); | |
791 | ci->i_max_size = le64_to_cpu(info->max_size); | |
792 | } | |
793 | } | |
355da1eb SW |
794 | |
795 | /* xattrs */ | |
796 | /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */ | |
508b32d8 | 797 | if ((ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL)) && |
355da1eb SW |
798 | le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) { |
799 | if (ci->i_xattrs.blob) | |
800 | ceph_buffer_put(ci->i_xattrs.blob); | |
801 | ci->i_xattrs.blob = xattr_blob; | |
802 | if (xattr_blob) | |
803 | memcpy(ci->i_xattrs.blob->vec.iov_base, | |
804 | iinfo->xattr_data, iinfo->xattr_len); | |
805 | ci->i_xattrs.version = le64_to_cpu(info->xattr_version); | |
7221fe4c | 806 | ceph_forget_all_cached_acls(inode); |
a6424e48 | 807 | xattr_blob = NULL; |
355da1eb SW |
808 | } |
809 | ||
810 | inode->i_mapping->a_ops = &ceph_aops; | |
355da1eb SW |
811 | |
812 | switch (inode->i_mode & S_IFMT) { | |
813 | case S_IFIFO: | |
814 | case S_IFBLK: | |
815 | case S_IFCHR: | |
816 | case S_IFSOCK: | |
817 | init_special_inode(inode, inode->i_mode, inode->i_rdev); | |
818 | inode->i_op = &ceph_file_iops; | |
819 | break; | |
820 | case S_IFREG: | |
821 | inode->i_op = &ceph_file_iops; | |
822 | inode->i_fop = &ceph_file_fops; | |
823 | break; | |
824 | case S_IFLNK: | |
825 | inode->i_op = &ceph_symlink_iops; | |
826 | if (!ci->i_symlink) { | |
810339ec | 827 | u32 symlen = iinfo->symlink_len; |
355da1eb SW |
828 | char *sym; |
829 | ||
be655596 | 830 | spin_unlock(&ci->i_ceph_lock); |
355da1eb | 831 | |
810339ec | 832 | err = -EINVAL; |
99c88e69 | 833 | if (WARN_ON(symlen != i_size_read(inode))) |
810339ec XW |
834 | goto out; |
835 | ||
355da1eb | 836 | err = -ENOMEM; |
810339ec | 837 | sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS); |
355da1eb SW |
838 | if (!sym) |
839 | goto out; | |
355da1eb | 840 | |
be655596 | 841 | spin_lock(&ci->i_ceph_lock); |
355da1eb SW |
842 | if (!ci->i_symlink) |
843 | ci->i_symlink = sym; | |
844 | else | |
845 | kfree(sym); /* lost a race */ | |
846 | } | |
ac194dcc | 847 | inode->i_link = ci->i_symlink; |
355da1eb SW |
848 | break; |
849 | case S_IFDIR: | |
850 | inode->i_op = &ceph_dir_iops; | |
851 | inode->i_fop = &ceph_dir_fops; | |
852 | ||
14303d20 SW |
853 | ci->i_dir_layout = iinfo->dir_layout; |
854 | ||
355da1eb SW |
855 | ci->i_files = le64_to_cpu(info->files); |
856 | ci->i_subdirs = le64_to_cpu(info->subdirs); | |
857 | ci->i_rbytes = le64_to_cpu(info->rbytes); | |
858 | ci->i_rfiles = le64_to_cpu(info->rfiles); | |
859 | ci->i_rsubdirs = le64_to_cpu(info->rsubdirs); | |
860 | ceph_decode_timespec(&ci->i_rctime, &info->rctime); | |
355da1eb SW |
861 | break; |
862 | default: | |
863 | pr_err("fill_inode %llx.%llx BAD mode 0%o\n", | |
864 | ceph_vinop(inode), inode->i_mode); | |
865 | } | |
866 | ||
355da1eb SW |
867 | /* were we issued a capability? */ |
868 | if (info->cap.caps) { | |
869 | if (ceph_snap(inode) == CEPH_NOSNAP) { | |
2f92b3d0 | 870 | unsigned caps = le32_to_cpu(info->cap.caps); |
355da1eb SW |
871 | ceph_add_cap(inode, session, |
872 | le64_to_cpu(info->cap.cap_id), | |
2f92b3d0 | 873 | cap_fmode, caps, |
355da1eb SW |
874 | le32_to_cpu(info->cap.wanted), |
875 | le32_to_cpu(info->cap.seq), | |
876 | le32_to_cpu(info->cap.mseq), | |
877 | le64_to_cpu(info->cap.realm), | |
d9df2783 | 878 | info->cap.flags, &new_cap); |
2f92b3d0 YZ |
879 | |
880 | /* set dir completion flag? */ | |
881 | if (S_ISDIR(inode->i_mode) && | |
882 | ci->i_files == 0 && ci->i_subdirs == 0 && | |
883 | (caps & CEPH_CAP_FILE_SHARED) && | |
884 | (issued & CEPH_CAP_FILE_EXCL) == 0 && | |
885 | !__ceph_dir_is_complete(ci)) { | |
886 | dout(" marking %p complete (empty)\n", inode); | |
fdd4e158 | 887 | i_size_write(inode, 0); |
2f92b3d0 | 888 | __ceph_dir_set_complete(ci, |
fdd4e158 YZ |
889 | atomic64_read(&ci->i_release_count), |
890 | atomic64_read(&ci->i_ordered_count)); | |
2f92b3d0 YZ |
891 | } |
892 | ||
d9df2783 | 893 | wake = true; |
355da1eb | 894 | } else { |
355da1eb SW |
895 | dout(" %p got snap_caps %s\n", inode, |
896 | ceph_cap_string(le32_to_cpu(info->cap.caps))); | |
897 | ci->i_snap_caps |= le32_to_cpu(info->cap.caps); | |
898 | if (cap_fmode >= 0) | |
899 | __ceph_get_fmode(ci, cap_fmode); | |
355da1eb | 900 | } |
04d000eb | 901 | } else if (cap_fmode >= 0) { |
f3ae1b97 | 902 | pr_warn("mds issued no caps on %llx.%llx\n", |
04d000eb SW |
903 | ceph_vinop(inode)); |
904 | __ceph_get_fmode(ci, cap_fmode); | |
355da1eb | 905 | } |
31c542a1 YZ |
906 | |
907 | if (iinfo->inline_version > 0 && | |
908 | iinfo->inline_version >= ci->i_inline_version) { | |
909 | int cache_caps = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO; | |
910 | ci->i_inline_version = iinfo->inline_version; | |
911 | if (ci->i_inline_version != CEPH_INLINE_NONE && | |
01deead0 YZ |
912 | (locked_page || |
913 | (le32_to_cpu(info->cap.caps) & cache_caps))) | |
31c542a1 YZ |
914 | fill_inline = true; |
915 | } | |
916 | ||
be655596 | 917 | spin_unlock(&ci->i_ceph_lock); |
355da1eb | 918 | |
31c542a1 | 919 | if (fill_inline) |
01deead0 | 920 | ceph_fill_inline_data(inode, locked_page, |
31c542a1 YZ |
921 | iinfo->inline_data, iinfo->inline_len); |
922 | ||
d9df2783 YZ |
923 | if (wake) |
924 | wake_up_all(&ci->i_cap_wq); | |
925 | ||
355da1eb SW |
926 | /* queue truncate if we saw i_size decrease */ |
927 | if (queue_trunc) | |
3c6f6b79 | 928 | ceph_queue_vmtruncate(inode); |
355da1eb SW |
929 | |
930 | /* populate frag tree */ | |
3e7fbe9c YZ |
931 | if (S_ISDIR(inode->i_mode)) |
932 | ceph_fill_fragtree(inode, &info->fragtree, dirinfo); | |
355da1eb SW |
933 | |
934 | /* update delegation info? */ | |
935 | if (dirinfo) | |
936 | ceph_fill_dirfrag(inode, dirinfo); | |
937 | ||
938 | err = 0; | |
355da1eb | 939 | out: |
d9df2783 YZ |
940 | if (new_cap) |
941 | ceph_put_cap(mdsc, new_cap); | |
b6c1d5b8 SW |
942 | if (xattr_blob) |
943 | ceph_buffer_put(xattr_blob); | |
355da1eb SW |
944 | return err; |
945 | } | |
946 | ||
947 | /* | |
948 | * caller should hold session s_mutex. | |
949 | */ | |
950 | static void update_dentry_lease(struct dentry *dentry, | |
951 | struct ceph_mds_reply_lease *lease, | |
952 | struct ceph_mds_session *session, | |
953 | unsigned long from_time) | |
954 | { | |
955 | struct ceph_dentry_info *di = ceph_dentry(dentry); | |
956 | long unsigned duration = le32_to_cpu(lease->duration_ms); | |
957 | long unsigned ttl = from_time + (duration * HZ) / 1000; | |
958 | long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000; | |
959 | struct inode *dir; | |
960 | ||
961 | /* only track leases on regular dentries */ | |
962 | if (dentry->d_op != &ceph_dentry_ops) | |
963 | return; | |
964 | ||
965 | spin_lock(&dentry->d_lock); | |
2f90b852 SW |
966 | dout("update_dentry_lease %p duration %lu ms ttl %lu\n", |
967 | dentry, duration, ttl); | |
355da1eb SW |
968 | |
969 | /* make lease_rdcache_gen match directory */ | |
2b0143b5 | 970 | dir = d_inode(dentry->d_parent); |
355da1eb SW |
971 | di->lease_shared_gen = ceph_inode(dir)->i_shared_gen; |
972 | ||
2f90b852 | 973 | if (duration == 0) |
355da1eb SW |
974 | goto out_unlock; |
975 | ||
976 | if (di->lease_gen == session->s_cap_gen && | |
977 | time_before(ttl, dentry->d_time)) | |
978 | goto out_unlock; /* we already have a newer lease. */ | |
979 | ||
980 | if (di->lease_session && di->lease_session != session) | |
981 | goto out_unlock; | |
982 | ||
983 | ceph_dentry_lru_touch(dentry); | |
984 | ||
985 | if (!di->lease_session) | |
986 | di->lease_session = ceph_get_mds_session(session); | |
987 | di->lease_gen = session->s_cap_gen; | |
988 | di->lease_seq = le32_to_cpu(lease->seq); | |
989 | di->lease_renew_after = half_ttl; | |
990 | di->lease_renew_from = 0; | |
991 | dentry->d_time = ttl; | |
992 | out_unlock: | |
993 | spin_unlock(&dentry->d_lock); | |
994 | return; | |
995 | } | |
996 | ||
997 | /* | |
998 | * splice a dentry to an inode. | |
999 | * caller must hold directory i_mutex for this to be safe. | |
355da1eb | 1000 | */ |
f7380af0 | 1001 | static struct dentry *splice_dentry(struct dentry *dn, struct inode *in) |
355da1eb SW |
1002 | { |
1003 | struct dentry *realdn; | |
1004 | ||
2b0143b5 | 1005 | BUG_ON(d_inode(dn)); |
1cd3935b | 1006 | |
355da1eb SW |
1007 | /* dn must be unhashed */ |
1008 | if (!d_unhashed(dn)) | |
1009 | d_drop(dn); | |
41d28bca | 1010 | realdn = d_splice_alias(in, dn); |
355da1eb | 1011 | if (IS_ERR(realdn)) { |
d69ed05a SW |
1012 | pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n", |
1013 | PTR_ERR(realdn), dn, in, ceph_vinop(in)); | |
355da1eb SW |
1014 | dn = realdn; /* note realdn contains the error */ |
1015 | goto out; | |
1016 | } else if (realdn) { | |
1017 | dout("dn %p (%d) spliced with %p (%d) " | |
1018 | "inode %p ino %llx.%llx\n", | |
84d08fa8 AV |
1019 | dn, d_count(dn), |
1020 | realdn, d_count(realdn), | |
2b0143b5 | 1021 | d_inode(realdn), ceph_vinop(d_inode(realdn))); |
355da1eb SW |
1022 | dput(dn); |
1023 | dn = realdn; | |
1024 | } else { | |
1025 | BUG_ON(!ceph_dentry(dn)); | |
355da1eb | 1026 | dout("dn %p attached to %p ino %llx.%llx\n", |
2b0143b5 | 1027 | dn, d_inode(dn), ceph_vinop(d_inode(dn))); |
355da1eb | 1028 | } |
355da1eb SW |
1029 | out: |
1030 | return dn; | |
1031 | } | |
1032 | ||
1033 | /* | |
1034 | * Incorporate results into the local cache. This is either just | |
1035 | * one inode, or a directory, dentry, and possibly linked-to inode (e.g., | |
1036 | * after a lookup). | |
1037 | * | |
1038 | * A reply may contain | |
1039 | * a directory inode along with a dentry. | |
1040 | * and/or a target inode | |
1041 | * | |
1042 | * Called with snap_rwsem (read). | |
1043 | */ | |
1044 | int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req, | |
1045 | struct ceph_mds_session *session) | |
1046 | { | |
1047 | struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info; | |
1048 | struct inode *in = NULL; | |
355da1eb | 1049 | struct ceph_vino vino; |
3d14c5d2 | 1050 | struct ceph_fs_client *fsc = ceph_sb_to_client(sb); |
355da1eb SW |
1051 | int err = 0; |
1052 | ||
1053 | dout("fill_trace %p is_dentry %d is_target %d\n", req, | |
1054 | rinfo->head->is_dentry, rinfo->head->is_target); | |
1055 | ||
1056 | #if 0 | |
1057 | /* | |
1058 | * Debugging hook: | |
1059 | * | |
1060 | * If we resend completed ops to a recovering mds, we get no | |
1061 | * trace. Since that is very rare, pretend this is the case | |
1062 | * to ensure the 'no trace' handlers in the callers behave. | |
1063 | * | |
1064 | * Fill in inodes unconditionally to avoid breaking cap | |
1065 | * invariants. | |
1066 | */ | |
1067 | if (rinfo->head->op & CEPH_MDS_OP_WRITE) { | |
1068 | pr_info("fill_trace faking empty trace on %lld %s\n", | |
1069 | req->r_tid, ceph_mds_op_name(rinfo->head->op)); | |
1070 | if (rinfo->head->is_dentry) { | |
1071 | rinfo->head->is_dentry = 0; | |
1072 | err = fill_inode(req->r_locked_dir, | |
1073 | &rinfo->diri, rinfo->dirfrag, | |
1074 | session, req->r_request_started, -1); | |
1075 | } | |
1076 | if (rinfo->head->is_target) { | |
1077 | rinfo->head->is_target = 0; | |
1078 | ininfo = rinfo->targeti.in; | |
1079 | vino.ino = le64_to_cpu(ininfo->ino); | |
1080 | vino.snap = le64_to_cpu(ininfo->snapid); | |
1081 | in = ceph_get_inode(sb, vino); | |
1082 | err = fill_inode(in, &rinfo->targeti, NULL, | |
1083 | session, req->r_request_started, | |
1084 | req->r_fmode); | |
1085 | iput(in); | |
1086 | } | |
1087 | } | |
1088 | #endif | |
1089 | ||
1090 | if (!rinfo->head->is_target && !rinfo->head->is_dentry) { | |
1091 | dout("fill_trace reply is empty!\n"); | |
167c9e35 SW |
1092 | if (rinfo->head->result == 0 && req->r_locked_dir) |
1093 | ceph_invalidate_dir_request(req); | |
355da1eb SW |
1094 | return 0; |
1095 | } | |
1096 | ||
1097 | if (rinfo->head->is_dentry) { | |
5b1daecd SW |
1098 | struct inode *dir = req->r_locked_dir; |
1099 | ||
6c5e50fa | 1100 | if (dir) { |
01deead0 YZ |
1101 | err = fill_inode(dir, NULL, |
1102 | &rinfo->diri, rinfo->dirfrag, | |
6c5e50fa SW |
1103 | session, req->r_request_started, -1, |
1104 | &req->r_caps_reservation); | |
1105 | if (err < 0) | |
19913b4e | 1106 | goto done; |
6c5e50fa SW |
1107 | } else { |
1108 | WARN_ON_ONCE(1); | |
1109 | } | |
19913b4e YZ |
1110 | |
1111 | if (dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME) { | |
1112 | struct qstr dname; | |
1113 | struct dentry *dn, *parent; | |
1114 | ||
1115 | BUG_ON(!rinfo->head->is_target); | |
1116 | BUG_ON(req->r_dentry); | |
1117 | ||
1118 | parent = d_find_any_alias(dir); | |
1119 | BUG_ON(!parent); | |
1120 | ||
1121 | dname.name = rinfo->dname; | |
1122 | dname.len = rinfo->dname_len; | |
1123 | dname.hash = full_name_hash(dname.name, dname.len); | |
1124 | vino.ino = le64_to_cpu(rinfo->targeti.in->ino); | |
1125 | vino.snap = le64_to_cpu(rinfo->targeti.in->snapid); | |
1126 | retry_lookup: | |
1127 | dn = d_lookup(parent, &dname); | |
1128 | dout("d_lookup on parent=%p name=%.*s got %p\n", | |
1129 | parent, dname.len, dname.name, dn); | |
1130 | ||
1131 | if (!dn) { | |
1132 | dn = d_alloc(parent, &dname); | |
1133 | dout("d_alloc %p '%.*s' = %p\n", parent, | |
1134 | dname.len, dname.name, dn); | |
1135 | if (dn == NULL) { | |
1136 | dput(parent); | |
1137 | err = -ENOMEM; | |
1138 | goto done; | |
1139 | } | |
1140 | err = ceph_init_dentry(dn); | |
1141 | if (err < 0) { | |
1142 | dput(dn); | |
1143 | dput(parent); | |
1144 | goto done; | |
1145 | } | |
2b0143b5 DH |
1146 | } else if (d_really_is_positive(dn) && |
1147 | (ceph_ino(d_inode(dn)) != vino.ino || | |
1148 | ceph_snap(d_inode(dn)) != vino.snap)) { | |
19913b4e | 1149 | dout(" dn %p points to wrong inode %p\n", |
2b0143b5 | 1150 | dn, d_inode(dn)); |
19913b4e YZ |
1151 | d_delete(dn); |
1152 | dput(dn); | |
1153 | goto retry_lookup; | |
1154 | } | |
1155 | ||
1156 | req->r_dentry = dn; | |
1157 | dput(parent); | |
1158 | } | |
5b1daecd SW |
1159 | } |
1160 | ||
86b58d13 YZ |
1161 | if (rinfo->head->is_target) { |
1162 | vino.ino = le64_to_cpu(rinfo->targeti.in->ino); | |
1163 | vino.snap = le64_to_cpu(rinfo->targeti.in->snapid); | |
1164 | ||
1165 | in = ceph_get_inode(sb, vino); | |
1166 | if (IS_ERR(in)) { | |
1167 | err = PTR_ERR(in); | |
1168 | goto done; | |
1169 | } | |
1170 | req->r_target_inode = in; | |
1171 | ||
01deead0 | 1172 | err = fill_inode(in, req->r_locked_page, &rinfo->targeti, NULL, |
86b58d13 | 1173 | session, req->r_request_started, |
48193012 | 1174 | (!req->r_aborted && rinfo->head->result == 0) ? |
86b58d13 YZ |
1175 | req->r_fmode : -1, |
1176 | &req->r_caps_reservation); | |
1177 | if (err < 0) { | |
1178 | pr_err("fill_inode badness %p %llx.%llx\n", | |
1179 | in, ceph_vinop(in)); | |
1180 | goto done; | |
1181 | } | |
1182 | } | |
1183 | ||
9358c6d4 SW |
1184 | /* |
1185 | * ignore null lease/binding on snapdir ENOENT, or else we | |
1186 | * will have trouble splicing in the virtual snapdir later | |
1187 | */ | |
1188 | if (rinfo->head->is_dentry && !req->r_aborted && | |
6c5e50fa | 1189 | req->r_locked_dir && |
9358c6d4 | 1190 | (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name, |
3d14c5d2 | 1191 | fsc->mount_options->snapdir_name, |
9358c6d4 | 1192 | req->r_dentry->d_name.len))) { |
355da1eb SW |
1193 | /* |
1194 | * lookup link rename : null -> possibly existing inode | |
1195 | * mknod symlink mkdir : null -> new inode | |
1196 | * unlink : linked -> null | |
1197 | */ | |
1198 | struct inode *dir = req->r_locked_dir; | |
1199 | struct dentry *dn = req->r_dentry; | |
1200 | bool have_dir_cap, have_lease; | |
1201 | ||
1202 | BUG_ON(!dn); | |
1203 | BUG_ON(!dir); | |
2b0143b5 | 1204 | BUG_ON(d_inode(dn->d_parent) != dir); |
355da1eb SW |
1205 | BUG_ON(ceph_ino(dir) != |
1206 | le64_to_cpu(rinfo->diri.in->ino)); | |
1207 | BUG_ON(ceph_snap(dir) != | |
1208 | le64_to_cpu(rinfo->diri.in->snapid)); | |
1209 | ||
355da1eb SW |
1210 | /* do we have a lease on the whole dir? */ |
1211 | have_dir_cap = | |
1212 | (le32_to_cpu(rinfo->diri.in->cap.caps) & | |
1213 | CEPH_CAP_FILE_SHARED); | |
1214 | ||
1215 | /* do we have a dn lease? */ | |
1216 | have_lease = have_dir_cap || | |
2f90b852 | 1217 | le32_to_cpu(rinfo->dlease->duration_ms); |
355da1eb SW |
1218 | if (!have_lease) |
1219 | dout("fill_trace no dentry lease or dir cap\n"); | |
1220 | ||
1221 | /* rename? */ | |
1222 | if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) { | |
0a8a70f9 YZ |
1223 | struct inode *olddir = req->r_old_dentry_dir; |
1224 | BUG_ON(!olddir); | |
1225 | ||
a455589f | 1226 | dout(" src %p '%pd' dst %p '%pd'\n", |
355da1eb | 1227 | req->r_old_dentry, |
a455589f AV |
1228 | req->r_old_dentry, |
1229 | dn, dn); | |
355da1eb SW |
1230 | dout("fill_trace doing d_move %p -> %p\n", |
1231 | req->r_old_dentry, dn); | |
c10f5e12 | 1232 | |
fdd4e158 YZ |
1233 | /* d_move screws up sibling dentries' offsets */ |
1234 | ceph_dir_clear_ordered(dir); | |
1235 | ceph_dir_clear_ordered(olddir); | |
1236 | ||
355da1eb | 1237 | d_move(req->r_old_dentry, dn); |
a455589f AV |
1238 | dout(" src %p '%pd' dst %p '%pd'\n", |
1239 | req->r_old_dentry, | |
355da1eb | 1240 | req->r_old_dentry, |
a455589f | 1241 | dn, dn); |
81a6cf2d | 1242 | |
c4a29f26 SW |
1243 | /* ensure target dentry is invalidated, despite |
1244 | rehashing bug in vfs_rename_dir */ | |
81a6cf2d SW |
1245 | ceph_invalidate_dentry_lease(dn); |
1246 | ||
99ccbd22 | 1247 | dout("dn %p gets new offset %lld\n", req->r_old_dentry, |
1cd3935b | 1248 | ceph_dentry(req->r_old_dentry)->offset); |
81a6cf2d | 1249 | |
355da1eb | 1250 | dn = req->r_old_dentry; /* use old_dentry */ |
355da1eb SW |
1251 | } |
1252 | ||
1253 | /* null dentry? */ | |
1254 | if (!rinfo->head->is_target) { | |
1255 | dout("fill_trace null dentry\n"); | |
2b0143b5 | 1256 | if (d_really_is_positive(dn)) { |
70db4f36 | 1257 | ceph_dir_clear_ordered(dir); |
355da1eb SW |
1258 | dout("d_delete %p\n", dn); |
1259 | d_delete(dn); | |
1260 | } else { | |
355da1eb | 1261 | if (have_lease && d_unhashed(dn)) |
f8b31710 | 1262 | d_add(dn, NULL); |
355da1eb SW |
1263 | update_dentry_lease(dn, rinfo->dlease, |
1264 | session, | |
1265 | req->r_request_started); | |
1266 | } | |
1267 | goto done; | |
1268 | } | |
1269 | ||
1270 | /* attach proper inode */ | |
2b0143b5 | 1271 | if (d_really_is_negative(dn)) { |
70db4f36 | 1272 | ceph_dir_clear_ordered(dir); |
86b58d13 | 1273 | ihold(in); |
f7380af0 | 1274 | dn = splice_dentry(dn, in); |
355da1eb SW |
1275 | if (IS_ERR(dn)) { |
1276 | err = PTR_ERR(dn); | |
1277 | goto done; | |
1278 | } | |
1279 | req->r_dentry = dn; /* may have spliced */ | |
2b0143b5 | 1280 | } else if (d_really_is_positive(dn) && d_inode(dn) != in) { |
355da1eb | 1281 | dout(" %p links to %p %llx.%llx, not %llx.%llx\n", |
2b0143b5 | 1282 | dn, d_inode(dn), ceph_vinop(d_inode(dn)), |
86b58d13 | 1283 | ceph_vinop(in)); |
200fd27c | 1284 | d_invalidate(dn); |
355da1eb | 1285 | have_lease = false; |
355da1eb SW |
1286 | } |
1287 | ||
1288 | if (have_lease) | |
1289 | update_dentry_lease(dn, rinfo->dlease, session, | |
1290 | req->r_request_started); | |
1291 | dout(" final dn %p\n", dn); | |
86b58d13 YZ |
1292 | } else if (!req->r_aborted && |
1293 | (req->r_op == CEPH_MDS_OP_LOOKUPSNAP || | |
1294 | req->r_op == CEPH_MDS_OP_MKSNAP)) { | |
355da1eb | 1295 | struct dentry *dn = req->r_dentry; |
0a8a70f9 | 1296 | struct inode *dir = req->r_locked_dir; |
355da1eb SW |
1297 | |
1298 | /* fill out a snapdir LOOKUPSNAP dentry */ | |
1299 | BUG_ON(!dn); | |
0a8a70f9 YZ |
1300 | BUG_ON(!dir); |
1301 | BUG_ON(ceph_snap(dir) != CEPH_SNAPDIR); | |
355da1eb | 1302 | dout(" linking snapped dir %p to dn %p\n", in, dn); |
70db4f36 | 1303 | ceph_dir_clear_ordered(dir); |
86b58d13 | 1304 | ihold(in); |
f7380af0 | 1305 | dn = splice_dentry(dn, in); |
355da1eb SW |
1306 | if (IS_ERR(dn)) { |
1307 | err = PTR_ERR(dn); | |
1308 | goto done; | |
1309 | } | |
1310 | req->r_dentry = dn; /* may have spliced */ | |
355da1eb | 1311 | } |
355da1eb SW |
1312 | done: |
1313 | dout("fill_trace done err=%d\n", err); | |
1314 | return err; | |
1315 | } | |
1316 | ||
1317 | /* | |
1318 | * Prepopulate our cache with readdir results, leases, etc. | |
1319 | */ | |
79f9f99a SW |
1320 | static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req, |
1321 | struct ceph_mds_session *session) | |
1322 | { | |
1323 | struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info; | |
1324 | int i, err = 0; | |
1325 | ||
1326 | for (i = 0; i < rinfo->dir_nr; i++) { | |
2a5beea3 | 1327 | struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i; |
79f9f99a SW |
1328 | struct ceph_vino vino; |
1329 | struct inode *in; | |
1330 | int rc; | |
1331 | ||
2a5beea3 YZ |
1332 | vino.ino = le64_to_cpu(rde->inode.in->ino); |
1333 | vino.snap = le64_to_cpu(rde->inode.in->snapid); | |
79f9f99a SW |
1334 | |
1335 | in = ceph_get_inode(req->r_dentry->d_sb, vino); | |
1336 | if (IS_ERR(in)) { | |
1337 | err = PTR_ERR(in); | |
1338 | dout("new_inode badness got %d\n", err); | |
1339 | continue; | |
1340 | } | |
2a5beea3 | 1341 | rc = fill_inode(in, NULL, &rde->inode, NULL, session, |
79f9f99a SW |
1342 | req->r_request_started, -1, |
1343 | &req->r_caps_reservation); | |
1344 | if (rc < 0) { | |
1345 | pr_err("fill_inode badness on %p got %d\n", in, rc); | |
1346 | err = rc; | |
79f9f99a | 1347 | } |
209ae762 | 1348 | iput(in); |
79f9f99a SW |
1349 | } |
1350 | ||
1351 | return err; | |
1352 | } | |
1353 | ||
fdd4e158 YZ |
1354 | void ceph_readdir_cache_release(struct ceph_readdir_cache_control *ctl) |
1355 | { | |
1356 | if (ctl->page) { | |
1357 | kunmap(ctl->page); | |
09cbfeaf | 1358 | put_page(ctl->page); |
fdd4e158 YZ |
1359 | ctl->page = NULL; |
1360 | } | |
1361 | } | |
1362 | ||
1363 | static int fill_readdir_cache(struct inode *dir, struct dentry *dn, | |
1364 | struct ceph_readdir_cache_control *ctl, | |
1365 | struct ceph_mds_request *req) | |
1366 | { | |
1367 | struct ceph_inode_info *ci = ceph_inode(dir); | |
09cbfeaf | 1368 | unsigned nsize = PAGE_SIZE / sizeof(struct dentry*); |
fdd4e158 YZ |
1369 | unsigned idx = ctl->index % nsize; |
1370 | pgoff_t pgoff = ctl->index / nsize; | |
1371 | ||
1372 | if (!ctl->page || pgoff != page_index(ctl->page)) { | |
1373 | ceph_readdir_cache_release(ctl); | |
af5e5eb5 YZ |
1374 | if (idx == 0) |
1375 | ctl->page = grab_cache_page(&dir->i_data, pgoff); | |
1376 | else | |
1377 | ctl->page = find_lock_page(&dir->i_data, pgoff); | |
fdd4e158 YZ |
1378 | if (!ctl->page) { |
1379 | ctl->index = -1; | |
af5e5eb5 | 1380 | return idx == 0 ? -ENOMEM : 0; |
fdd4e158 YZ |
1381 | } |
1382 | /* reading/filling the cache are serialized by | |
1383 | * i_mutex, no need to use page lock */ | |
1384 | unlock_page(ctl->page); | |
1385 | ctl->dentries = kmap(ctl->page); | |
af5e5eb5 | 1386 | if (idx == 0) |
09cbfeaf | 1387 | memset(ctl->dentries, 0, PAGE_SIZE); |
fdd4e158 YZ |
1388 | } |
1389 | ||
1390 | if (req->r_dir_release_cnt == atomic64_read(&ci->i_release_count) && | |
1391 | req->r_dir_ordered_cnt == atomic64_read(&ci->i_ordered_count)) { | |
1392 | dout("readdir cache dn %p idx %d\n", dn, ctl->index); | |
1393 | ctl->dentries[idx] = dn; | |
1394 | ctl->index++; | |
1395 | } else { | |
1396 | dout("disable readdir cache\n"); | |
1397 | ctl->index = -1; | |
1398 | } | |
1399 | return 0; | |
1400 | } | |
1401 | ||
355da1eb SW |
1402 | int ceph_readdir_prepopulate(struct ceph_mds_request *req, |
1403 | struct ceph_mds_session *session) | |
1404 | { | |
1405 | struct dentry *parent = req->r_dentry; | |
f3c4ebe6 | 1406 | struct ceph_inode_info *ci = ceph_inode(d_inode(parent)); |
355da1eb SW |
1407 | struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info; |
1408 | struct qstr dname; | |
1409 | struct dentry *dn; | |
1410 | struct inode *in; | |
315f2408 | 1411 | int err = 0, skipped = 0, ret, i; |
355da1eb SW |
1412 | struct inode *snapdir = NULL; |
1413 | struct ceph_mds_request_head *rhead = req->r_request->front.iov_base; | |
81c6aea5 | 1414 | u32 frag = le32_to_cpu(rhead->args.readdir.frag); |
f3c4ebe6 YZ |
1415 | u32 last_hash = 0; |
1416 | u32 fpos_offset; | |
fdd4e158 YZ |
1417 | struct ceph_readdir_cache_control cache_ctl = {}; |
1418 | ||
1419 | if (req->r_aborted) | |
1420 | return readdir_prepopulate_inodes_only(req, session); | |
81c6aea5 | 1421 | |
f3c4ebe6 YZ |
1422 | if (rinfo->hash_order && req->r_path2) { |
1423 | last_hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash, | |
1424 | req->r_path2, strlen(req->r_path2)); | |
1425 | last_hash = ceph_frag_value(last_hash); | |
1426 | } | |
1427 | ||
81c6aea5 YZ |
1428 | if (rinfo->dir_dir && |
1429 | le32_to_cpu(rinfo->dir_dir->frag) != frag) { | |
1430 | dout("readdir_prepopulate got new frag %x -> %x\n", | |
1431 | frag, le32_to_cpu(rinfo->dir_dir->frag)); | |
1432 | frag = le32_to_cpu(rinfo->dir_dir->frag); | |
f3c4ebe6 YZ |
1433 | if (!rinfo->hash_order) |
1434 | req->r_readdir_offset = 2; | |
81c6aea5 | 1435 | } |
355da1eb SW |
1436 | |
1437 | if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) { | |
2b0143b5 | 1438 | snapdir = ceph_get_snapdir(d_inode(parent)); |
355da1eb SW |
1439 | parent = d_find_alias(snapdir); |
1440 | dout("readdir_prepopulate %d items under SNAPDIR dn %p\n", | |
1441 | rinfo->dir_nr, parent); | |
1442 | } else { | |
1443 | dout("readdir_prepopulate %d items under dn %p\n", | |
1444 | rinfo->dir_nr, parent); | |
1445 | if (rinfo->dir_dir) | |
2b0143b5 | 1446 | ceph_fill_dirfrag(d_inode(parent), rinfo->dir_dir); |
355da1eb SW |
1447 | } |
1448 | ||
fdd4e158 YZ |
1449 | if (ceph_frag_is_leftmost(frag) && req->r_readdir_offset == 2) { |
1450 | /* note dir version at start of readdir so we can tell | |
1451 | * if any dentries get dropped */ | |
fdd4e158 YZ |
1452 | req->r_dir_release_cnt = atomic64_read(&ci->i_release_count); |
1453 | req->r_dir_ordered_cnt = atomic64_read(&ci->i_ordered_count); | |
1454 | req->r_readdir_cache_idx = 0; | |
1455 | } | |
1456 | ||
1457 | cache_ctl.index = req->r_readdir_cache_idx; | |
f3c4ebe6 | 1458 | fpos_offset = req->r_readdir_offset; |
fdd4e158 | 1459 | |
86b58d13 | 1460 | /* FIXME: release caps/leases if error occurs */ |
355da1eb | 1461 | for (i = 0; i < rinfo->dir_nr; i++) { |
2a5beea3 | 1462 | struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i; |
355da1eb SW |
1463 | struct ceph_vino vino; |
1464 | ||
2a5beea3 YZ |
1465 | dname.name = rde->name; |
1466 | dname.len = rde->name_len; | |
355da1eb SW |
1467 | dname.hash = full_name_hash(dname.name, dname.len); |
1468 | ||
2a5beea3 YZ |
1469 | vino.ino = le64_to_cpu(rde->inode.in->ino); |
1470 | vino.snap = le64_to_cpu(rde->inode.in->snapid); | |
355da1eb | 1471 | |
f3c4ebe6 YZ |
1472 | if (rinfo->hash_order) { |
1473 | u32 hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash, | |
1474 | rde->name, rde->name_len); | |
1475 | hash = ceph_frag_value(hash); | |
1476 | if (hash != last_hash) | |
1477 | fpos_offset = 2; | |
1478 | last_hash = hash; | |
1479 | rde->offset = ceph_make_fpos(hash, fpos_offset++, true); | |
1480 | } else { | |
1481 | rde->offset = ceph_make_fpos(frag, fpos_offset++, false); | |
1482 | } | |
1483 | ||
355da1eb SW |
1484 | retry_lookup: |
1485 | dn = d_lookup(parent, &dname); | |
1486 | dout("d_lookup on parent=%p name=%.*s got %p\n", | |
1487 | parent, dname.len, dname.name, dn); | |
1488 | ||
1489 | if (!dn) { | |
1490 | dn = d_alloc(parent, &dname); | |
1491 | dout("d_alloc %p '%.*s' = %p\n", parent, | |
1492 | dname.len, dname.name, dn); | |
1493 | if (dn == NULL) { | |
1494 | dout("d_alloc badness\n"); | |
1495 | err = -ENOMEM; | |
1496 | goto out; | |
1497 | } | |
86b58d13 YZ |
1498 | ret = ceph_init_dentry(dn); |
1499 | if (ret < 0) { | |
8c696737 | 1500 | dput(dn); |
86b58d13 | 1501 | err = ret; |
355da1eb | 1502 | goto out; |
8c696737 | 1503 | } |
2b0143b5 DH |
1504 | } else if (d_really_is_positive(dn) && |
1505 | (ceph_ino(d_inode(dn)) != vino.ino || | |
1506 | ceph_snap(d_inode(dn)) != vino.snap)) { | |
355da1eb | 1507 | dout(" dn %p points to wrong inode %p\n", |
2b0143b5 | 1508 | dn, d_inode(dn)); |
355da1eb SW |
1509 | d_delete(dn); |
1510 | dput(dn); | |
1511 | goto retry_lookup; | |
355da1eb SW |
1512 | } |
1513 | ||
355da1eb | 1514 | /* inode */ |
2b0143b5 DH |
1515 | if (d_really_is_positive(dn)) { |
1516 | in = d_inode(dn); | |
355da1eb SW |
1517 | } else { |
1518 | in = ceph_get_inode(parent->d_sb, vino); | |
ac1f12ef | 1519 | if (IS_ERR(in)) { |
355da1eb | 1520 | dout("new_inode badness\n"); |
2744c171 | 1521 | d_drop(dn); |
355da1eb | 1522 | dput(dn); |
ac1f12ef | 1523 | err = PTR_ERR(in); |
355da1eb SW |
1524 | goto out; |
1525 | } | |
355da1eb SW |
1526 | } |
1527 | ||
2a5beea3 | 1528 | ret = fill_inode(in, NULL, &rde->inode, NULL, session, |
fdd4e158 YZ |
1529 | req->r_request_started, -1, |
1530 | &req->r_caps_reservation); | |
1531 | if (ret < 0) { | |
355da1eb | 1532 | pr_err("fill_inode badness on %p\n", in); |
2b0143b5 | 1533 | if (d_really_is_negative(dn)) |
86b58d13 YZ |
1534 | iput(in); |
1535 | d_drop(dn); | |
fdd4e158 | 1536 | err = ret; |
d69ed05a | 1537 | goto next_item; |
355da1eb | 1538 | } |
86b58d13 | 1539 | |
2b0143b5 | 1540 | if (d_really_is_negative(dn)) { |
315f2408 YZ |
1541 | struct dentry *realdn; |
1542 | ||
1543 | if (ceph_security_xattr_deadlock(in)) { | |
1544 | dout(" skip splicing dn %p to inode %p" | |
1545 | " (security xattr deadlock)\n", dn, in); | |
1546 | iput(in); | |
1547 | skipped++; | |
1548 | goto next_item; | |
1549 | } | |
1550 | ||
1551 | realdn = splice_dentry(dn, in); | |
5cba372c YZ |
1552 | if (IS_ERR(realdn)) { |
1553 | err = PTR_ERR(realdn); | |
1554 | d_drop(dn); | |
86b58d13 YZ |
1555 | dn = NULL; |
1556 | goto next_item; | |
1557 | } | |
5cba372c | 1558 | dn = realdn; |
86b58d13 YZ |
1559 | } |
1560 | ||
f3c4ebe6 | 1561 | ceph_dentry(dn)->offset = rde->offset; |
86b58d13 | 1562 | |
2a5beea3 | 1563 | update_dentry_lease(dn, rde->lease, req->r_session, |
86b58d13 | 1564 | req->r_request_started); |
fdd4e158 | 1565 | |
315f2408 | 1566 | if (err == 0 && skipped == 0 && cache_ctl.index >= 0) { |
fdd4e158 YZ |
1567 | ret = fill_readdir_cache(d_inode(parent), dn, |
1568 | &cache_ctl, req); | |
1569 | if (ret < 0) | |
1570 | err = ret; | |
1571 | } | |
d69ed05a SW |
1572 | next_item: |
1573 | if (dn) | |
1574 | dput(dn); | |
355da1eb | 1575 | } |
355da1eb | 1576 | out: |
315f2408 | 1577 | if (err == 0 && skipped == 0) { |
fdd4e158 YZ |
1578 | req->r_did_prepopulate = true; |
1579 | req->r_readdir_cache_idx = cache_ctl.index; | |
1580 | } | |
1581 | ceph_readdir_cache_release(&cache_ctl); | |
355da1eb SW |
1582 | if (snapdir) { |
1583 | iput(snapdir); | |
1584 | dput(parent); | |
1585 | } | |
1586 | dout("readdir_prepopulate done\n"); | |
1587 | return err; | |
1588 | } | |
1589 | ||
1590 | int ceph_inode_set_size(struct inode *inode, loff_t size) | |
1591 | { | |
1592 | struct ceph_inode_info *ci = ceph_inode(inode); | |
1593 | int ret = 0; | |
1594 | ||
be655596 | 1595 | spin_lock(&ci->i_ceph_lock); |
355da1eb | 1596 | dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size); |
99c88e69 | 1597 | i_size_write(inode, size); |
355da1eb SW |
1598 | inode->i_blocks = (size + (1 << 9) - 1) >> 9; |
1599 | ||
1600 | /* tell the MDS if we are approaching max_size */ | |
1601 | if ((size << 1) >= ci->i_max_size && | |
1602 | (ci->i_reported_size << 1) < ci->i_max_size) | |
1603 | ret = 1; | |
1604 | ||
be655596 | 1605 | spin_unlock(&ci->i_ceph_lock); |
355da1eb SW |
1606 | return ret; |
1607 | } | |
1608 | ||
1609 | /* | |
1610 | * Write back inode data in a worker thread. (This can't be done | |
1611 | * in the message handler context.) | |
1612 | */ | |
3c6f6b79 SW |
1613 | void ceph_queue_writeback(struct inode *inode) |
1614 | { | |
15a2015f | 1615 | ihold(inode); |
3c6f6b79 SW |
1616 | if (queue_work(ceph_inode_to_client(inode)->wb_wq, |
1617 | &ceph_inode(inode)->i_wb_work)) { | |
2c27c9a5 | 1618 | dout("ceph_queue_writeback %p\n", inode); |
3c6f6b79 | 1619 | } else { |
2c27c9a5 | 1620 | dout("ceph_queue_writeback %p failed\n", inode); |
15a2015f | 1621 | iput(inode); |
3c6f6b79 SW |
1622 | } |
1623 | } | |
1624 | ||
1625 | static void ceph_writeback_work(struct work_struct *work) | |
355da1eb SW |
1626 | { |
1627 | struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info, | |
1628 | i_wb_work); | |
1629 | struct inode *inode = &ci->vfs_inode; | |
1630 | ||
1631 | dout("writeback %p\n", inode); | |
1632 | filemap_fdatawrite(&inode->i_data); | |
1633 | iput(inode); | |
1634 | } | |
1635 | ||
3c6f6b79 SW |
1636 | /* |
1637 | * queue an async invalidation | |
1638 | */ | |
1639 | void ceph_queue_invalidate(struct inode *inode) | |
1640 | { | |
15a2015f | 1641 | ihold(inode); |
3c6f6b79 SW |
1642 | if (queue_work(ceph_inode_to_client(inode)->pg_inv_wq, |
1643 | &ceph_inode(inode)->i_pg_inv_work)) { | |
1644 | dout("ceph_queue_invalidate %p\n", inode); | |
3c6f6b79 SW |
1645 | } else { |
1646 | dout("ceph_queue_invalidate %p failed\n", inode); | |
15a2015f | 1647 | iput(inode); |
3c6f6b79 SW |
1648 | } |
1649 | } | |
1650 | ||
355da1eb SW |
1651 | /* |
1652 | * Invalidate inode pages in a worker thread. (This can't be done | |
1653 | * in the message handler context.) | |
1654 | */ | |
3c6f6b79 | 1655 | static void ceph_invalidate_work(struct work_struct *work) |
355da1eb SW |
1656 | { |
1657 | struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info, | |
1658 | i_pg_inv_work); | |
1659 | struct inode *inode = &ci->vfs_inode; | |
6c93df5d | 1660 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); |
355da1eb SW |
1661 | u32 orig_gen; |
1662 | int check = 0; | |
1663 | ||
b0d7c223 | 1664 | mutex_lock(&ci->i_truncate_mutex); |
6c93df5d YZ |
1665 | |
1666 | if (ACCESS_ONCE(fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) { | |
1667 | pr_warn_ratelimited("invalidate_pages %p %lld forced umount\n", | |
1668 | inode, ceph_ino(inode)); | |
1669 | mapping_set_error(inode->i_mapping, -EIO); | |
1670 | truncate_pagecache(inode, 0); | |
1671 | mutex_unlock(&ci->i_truncate_mutex); | |
1672 | goto out; | |
1673 | } | |
1674 | ||
be655596 | 1675 | spin_lock(&ci->i_ceph_lock); |
355da1eb SW |
1676 | dout("invalidate_pages %p gen %d revoking %d\n", inode, |
1677 | ci->i_rdcache_gen, ci->i_rdcache_revoking); | |
cd045cb4 | 1678 | if (ci->i_rdcache_revoking != ci->i_rdcache_gen) { |
9563f88c YZ |
1679 | if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE)) |
1680 | check = 1; | |
be655596 | 1681 | spin_unlock(&ci->i_ceph_lock); |
b0d7c223 | 1682 | mutex_unlock(&ci->i_truncate_mutex); |
355da1eb SW |
1683 | goto out; |
1684 | } | |
1685 | orig_gen = ci->i_rdcache_gen; | |
be655596 | 1686 | spin_unlock(&ci->i_ceph_lock); |
355da1eb | 1687 | |
4e217b5d | 1688 | truncate_pagecache(inode, 0); |
355da1eb | 1689 | |
be655596 | 1690 | spin_lock(&ci->i_ceph_lock); |
cd045cb4 SW |
1691 | if (orig_gen == ci->i_rdcache_gen && |
1692 | orig_gen == ci->i_rdcache_revoking) { | |
355da1eb SW |
1693 | dout("invalidate_pages %p gen %d successful\n", inode, |
1694 | ci->i_rdcache_gen); | |
cd045cb4 | 1695 | ci->i_rdcache_revoking--; |
355da1eb SW |
1696 | check = 1; |
1697 | } else { | |
cd045cb4 SW |
1698 | dout("invalidate_pages %p gen %d raced, now %d revoking %d\n", |
1699 | inode, orig_gen, ci->i_rdcache_gen, | |
1700 | ci->i_rdcache_revoking); | |
9563f88c YZ |
1701 | if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE)) |
1702 | check = 1; | |
355da1eb | 1703 | } |
be655596 | 1704 | spin_unlock(&ci->i_ceph_lock); |
b0d7c223 | 1705 | mutex_unlock(&ci->i_truncate_mutex); |
9563f88c | 1706 | out: |
355da1eb SW |
1707 | if (check) |
1708 | ceph_check_caps(ci, 0, NULL); | |
355da1eb SW |
1709 | iput(inode); |
1710 | } | |
1711 | ||
1712 | ||
1713 | /* | |
3f99969f | 1714 | * called by trunc_wq; |
355da1eb SW |
1715 | * |
1716 | * We also truncate in a separate thread as well. | |
1717 | */ | |
3c6f6b79 | 1718 | static void ceph_vmtruncate_work(struct work_struct *work) |
355da1eb SW |
1719 | { |
1720 | struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info, | |
1721 | i_vmtruncate_work); | |
1722 | struct inode *inode = &ci->vfs_inode; | |
1723 | ||
1724 | dout("vmtruncate_work %p\n", inode); | |
b415bf4f | 1725 | __ceph_do_pending_vmtruncate(inode); |
355da1eb SW |
1726 | iput(inode); |
1727 | } | |
1728 | ||
3c6f6b79 SW |
1729 | /* |
1730 | * Queue an async vmtruncate. If we fail to queue work, we will handle | |
1731 | * the truncation the next time we call __ceph_do_pending_vmtruncate. | |
1732 | */ | |
1733 | void ceph_queue_vmtruncate(struct inode *inode) | |
1734 | { | |
1735 | struct ceph_inode_info *ci = ceph_inode(inode); | |
1736 | ||
15a2015f | 1737 | ihold(inode); |
99ccbd22 | 1738 | |
640ef79d | 1739 | if (queue_work(ceph_sb_to_client(inode->i_sb)->trunc_wq, |
3c6f6b79 SW |
1740 | &ci->i_vmtruncate_work)) { |
1741 | dout("ceph_queue_vmtruncate %p\n", inode); | |
3c6f6b79 SW |
1742 | } else { |
1743 | dout("ceph_queue_vmtruncate %p failed, pending=%d\n", | |
1744 | inode, ci->i_truncate_pending); | |
15a2015f | 1745 | iput(inode); |
3c6f6b79 SW |
1746 | } |
1747 | } | |
1748 | ||
355da1eb | 1749 | /* |
355da1eb SW |
1750 | * Make sure any pending truncation is applied before doing anything |
1751 | * that may depend on it. | |
1752 | */ | |
b415bf4f | 1753 | void __ceph_do_pending_vmtruncate(struct inode *inode) |
355da1eb SW |
1754 | { |
1755 | struct ceph_inode_info *ci = ceph_inode(inode); | |
1756 | u64 to; | |
a85f50b6 | 1757 | int wrbuffer_refs, finish = 0; |
355da1eb | 1758 | |
b0d7c223 | 1759 | mutex_lock(&ci->i_truncate_mutex); |
355da1eb | 1760 | retry: |
be655596 | 1761 | spin_lock(&ci->i_ceph_lock); |
355da1eb SW |
1762 | if (ci->i_truncate_pending == 0) { |
1763 | dout("__do_pending_vmtruncate %p none pending\n", inode); | |
be655596 | 1764 | spin_unlock(&ci->i_ceph_lock); |
b0d7c223 | 1765 | mutex_unlock(&ci->i_truncate_mutex); |
355da1eb SW |
1766 | return; |
1767 | } | |
1768 | ||
1769 | /* | |
1770 | * make sure any dirty snapped pages are flushed before we | |
1771 | * possibly truncate them.. so write AND block! | |
1772 | */ | |
1773 | if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) { | |
1774 | dout("__do_pending_vmtruncate %p flushing snaps first\n", | |
1775 | inode); | |
be655596 | 1776 | spin_unlock(&ci->i_ceph_lock); |
355da1eb SW |
1777 | filemap_write_and_wait_range(&inode->i_data, 0, |
1778 | inode->i_sb->s_maxbytes); | |
1779 | goto retry; | |
1780 | } | |
1781 | ||
b0d7c223 YZ |
1782 | /* there should be no reader or writer */ |
1783 | WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref); | |
1784 | ||
355da1eb SW |
1785 | to = ci->i_truncate_size; |
1786 | wrbuffer_refs = ci->i_wrbuffer_ref; | |
1787 | dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode, | |
1788 | ci->i_truncate_pending, to); | |
be655596 | 1789 | spin_unlock(&ci->i_ceph_lock); |
355da1eb | 1790 | |
4e217b5d | 1791 | truncate_pagecache(inode, to); |
355da1eb | 1792 | |
be655596 | 1793 | spin_lock(&ci->i_ceph_lock); |
a85f50b6 YZ |
1794 | if (to == ci->i_truncate_size) { |
1795 | ci->i_truncate_pending = 0; | |
1796 | finish = 1; | |
1797 | } | |
be655596 | 1798 | spin_unlock(&ci->i_ceph_lock); |
a85f50b6 YZ |
1799 | if (!finish) |
1800 | goto retry; | |
355da1eb | 1801 | |
b0d7c223 YZ |
1802 | mutex_unlock(&ci->i_truncate_mutex); |
1803 | ||
355da1eb SW |
1804 | if (wrbuffer_refs == 0) |
1805 | ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL); | |
a85f50b6 YZ |
1806 | |
1807 | wake_up_all(&ci->i_cap_wq); | |
355da1eb SW |
1808 | } |
1809 | ||
355da1eb SW |
1810 | /* |
1811 | * symlinks | |
1812 | */ | |
355da1eb SW |
1813 | static const struct inode_operations ceph_symlink_iops = { |
1814 | .readlink = generic_readlink, | |
6b255391 | 1815 | .get_link = simple_get_link, |
0b932672 YZ |
1816 | .setattr = ceph_setattr, |
1817 | .getattr = ceph_getattr, | |
1818 | .setxattr = ceph_setxattr, | |
1819 | .getxattr = ceph_getxattr, | |
1820 | .listxattr = ceph_listxattr, | |
1821 | .removexattr = ceph_removexattr, | |
355da1eb SW |
1822 | }; |
1823 | ||
1824 | /* | |
1825 | * setattr | |
1826 | */ | |
1827 | int ceph_setattr(struct dentry *dentry, struct iattr *attr) | |
1828 | { | |
2b0143b5 | 1829 | struct inode *inode = d_inode(dentry); |
355da1eb | 1830 | struct ceph_inode_info *ci = ceph_inode(inode); |
355da1eb SW |
1831 | const unsigned int ia_valid = attr->ia_valid; |
1832 | struct ceph_mds_request *req; | |
3d14c5d2 | 1833 | struct ceph_mds_client *mdsc = ceph_sb_to_client(dentry->d_sb)->mdsc; |
f66fd9f0 | 1834 | struct ceph_cap_flush *prealloc_cf; |
355da1eb SW |
1835 | int issued; |
1836 | int release = 0, dirtied = 0; | |
1837 | int mask = 0; | |
1838 | int err = 0; | |
fca65b4a | 1839 | int inode_dirty_flags = 0; |
604d1b02 | 1840 | bool lock_snap_rwsem = false; |
355da1eb SW |
1841 | |
1842 | if (ceph_snap(inode) != CEPH_NOSNAP) | |
1843 | return -EROFS; | |
1844 | ||
355da1eb SW |
1845 | err = inode_change_ok(inode, attr); |
1846 | if (err != 0) | |
1847 | return err; | |
1848 | ||
f66fd9f0 YZ |
1849 | prealloc_cf = ceph_alloc_cap_flush(); |
1850 | if (!prealloc_cf) | |
1851 | return -ENOMEM; | |
1852 | ||
355da1eb SW |
1853 | req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR, |
1854 | USE_AUTH_MDS); | |
f66fd9f0 YZ |
1855 | if (IS_ERR(req)) { |
1856 | ceph_free_cap_flush(prealloc_cf); | |
355da1eb | 1857 | return PTR_ERR(req); |
f66fd9f0 | 1858 | } |
355da1eb | 1859 | |
be655596 | 1860 | spin_lock(&ci->i_ceph_lock); |
355da1eb | 1861 | issued = __ceph_caps_issued(ci, NULL); |
604d1b02 YZ |
1862 | |
1863 | if (!ci->i_head_snapc && | |
1864 | (issued & (CEPH_CAP_ANY_EXCL | CEPH_CAP_FILE_WR))) { | |
1865 | lock_snap_rwsem = true; | |
1866 | if (!down_read_trylock(&mdsc->snap_rwsem)) { | |
1867 | spin_unlock(&ci->i_ceph_lock); | |
1868 | down_read(&mdsc->snap_rwsem); | |
1869 | spin_lock(&ci->i_ceph_lock); | |
1870 | issued = __ceph_caps_issued(ci, NULL); | |
1871 | } | |
1872 | } | |
1873 | ||
355da1eb SW |
1874 | dout("setattr %p issued %s\n", inode, ceph_cap_string(issued)); |
1875 | ||
1876 | if (ia_valid & ATTR_UID) { | |
1877 | dout("setattr %p uid %d -> %d\n", inode, | |
bd2bae6a EB |
1878 | from_kuid(&init_user_ns, inode->i_uid), |
1879 | from_kuid(&init_user_ns, attr->ia_uid)); | |
355da1eb SW |
1880 | if (issued & CEPH_CAP_AUTH_EXCL) { |
1881 | inode->i_uid = attr->ia_uid; | |
1882 | dirtied |= CEPH_CAP_AUTH_EXCL; | |
1883 | } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || | |
ab871b90 EB |
1884 | !uid_eq(attr->ia_uid, inode->i_uid)) { |
1885 | req->r_args.setattr.uid = cpu_to_le32( | |
1886 | from_kuid(&init_user_ns, attr->ia_uid)); | |
355da1eb SW |
1887 | mask |= CEPH_SETATTR_UID; |
1888 | release |= CEPH_CAP_AUTH_SHARED; | |
1889 | } | |
1890 | } | |
1891 | if (ia_valid & ATTR_GID) { | |
1892 | dout("setattr %p gid %d -> %d\n", inode, | |
bd2bae6a EB |
1893 | from_kgid(&init_user_ns, inode->i_gid), |
1894 | from_kgid(&init_user_ns, attr->ia_gid)); | |
355da1eb SW |
1895 | if (issued & CEPH_CAP_AUTH_EXCL) { |
1896 | inode->i_gid = attr->ia_gid; | |
1897 | dirtied |= CEPH_CAP_AUTH_EXCL; | |
1898 | } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || | |
ab871b90 EB |
1899 | !gid_eq(attr->ia_gid, inode->i_gid)) { |
1900 | req->r_args.setattr.gid = cpu_to_le32( | |
1901 | from_kgid(&init_user_ns, attr->ia_gid)); | |
355da1eb SW |
1902 | mask |= CEPH_SETATTR_GID; |
1903 | release |= CEPH_CAP_AUTH_SHARED; | |
1904 | } | |
1905 | } | |
1906 | if (ia_valid & ATTR_MODE) { | |
1907 | dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode, | |
1908 | attr->ia_mode); | |
1909 | if (issued & CEPH_CAP_AUTH_EXCL) { | |
1910 | inode->i_mode = attr->ia_mode; | |
1911 | dirtied |= CEPH_CAP_AUTH_EXCL; | |
1912 | } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || | |
1913 | attr->ia_mode != inode->i_mode) { | |
7221fe4c | 1914 | inode->i_mode = attr->ia_mode; |
355da1eb SW |
1915 | req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode); |
1916 | mask |= CEPH_SETATTR_MODE; | |
1917 | release |= CEPH_CAP_AUTH_SHARED; | |
1918 | } | |
1919 | } | |
1920 | ||
1921 | if (ia_valid & ATTR_ATIME) { | |
1922 | dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode, | |
1923 | inode->i_atime.tv_sec, inode->i_atime.tv_nsec, | |
1924 | attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec); | |
1925 | if (issued & CEPH_CAP_FILE_EXCL) { | |
1926 | ci->i_time_warp_seq++; | |
1927 | inode->i_atime = attr->ia_atime; | |
1928 | dirtied |= CEPH_CAP_FILE_EXCL; | |
1929 | } else if ((issued & CEPH_CAP_FILE_WR) && | |
1930 | timespec_compare(&inode->i_atime, | |
1931 | &attr->ia_atime) < 0) { | |
1932 | inode->i_atime = attr->ia_atime; | |
1933 | dirtied |= CEPH_CAP_FILE_WR; | |
1934 | } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || | |
1935 | !timespec_equal(&inode->i_atime, &attr->ia_atime)) { | |
1936 | ceph_encode_timespec(&req->r_args.setattr.atime, | |
1937 | &attr->ia_atime); | |
1938 | mask |= CEPH_SETATTR_ATIME; | |
1939 | release |= CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_RD | | |
1940 | CEPH_CAP_FILE_WR; | |
1941 | } | |
1942 | } | |
1943 | if (ia_valid & ATTR_MTIME) { | |
1944 | dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode, | |
1945 | inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec, | |
1946 | attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec); | |
1947 | if (issued & CEPH_CAP_FILE_EXCL) { | |
1948 | ci->i_time_warp_seq++; | |
1949 | inode->i_mtime = attr->ia_mtime; | |
1950 | dirtied |= CEPH_CAP_FILE_EXCL; | |
1951 | } else if ((issued & CEPH_CAP_FILE_WR) && | |
1952 | timespec_compare(&inode->i_mtime, | |
1953 | &attr->ia_mtime) < 0) { | |
1954 | inode->i_mtime = attr->ia_mtime; | |
1955 | dirtied |= CEPH_CAP_FILE_WR; | |
1956 | } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || | |
1957 | !timespec_equal(&inode->i_mtime, &attr->ia_mtime)) { | |
1958 | ceph_encode_timespec(&req->r_args.setattr.mtime, | |
1959 | &attr->ia_mtime); | |
1960 | mask |= CEPH_SETATTR_MTIME; | |
1961 | release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD | | |
1962 | CEPH_CAP_FILE_WR; | |
1963 | } | |
1964 | } | |
1965 | if (ia_valid & ATTR_SIZE) { | |
1966 | dout("setattr %p size %lld -> %lld\n", inode, | |
1967 | inode->i_size, attr->ia_size); | |
355da1eb SW |
1968 | if ((issued & CEPH_CAP_FILE_EXCL) && |
1969 | attr->ia_size > inode->i_size) { | |
99c88e69 | 1970 | i_size_write(inode, attr->ia_size); |
355da1eb SW |
1971 | inode->i_blocks = |
1972 | (attr->ia_size + (1 << 9) - 1) >> 9; | |
1973 | inode->i_ctime = attr->ia_ctime; | |
1974 | ci->i_reported_size = attr->ia_size; | |
1975 | dirtied |= CEPH_CAP_FILE_EXCL; | |
1976 | } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || | |
1977 | attr->ia_size != inode->i_size) { | |
1978 | req->r_args.setattr.size = cpu_to_le64(attr->ia_size); | |
1979 | req->r_args.setattr.old_size = | |
1980 | cpu_to_le64(inode->i_size); | |
1981 | mask |= CEPH_SETATTR_SIZE; | |
1982 | release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD | | |
1983 | CEPH_CAP_FILE_WR; | |
1984 | } | |
1985 | } | |
1986 | ||
1987 | /* these do nothing */ | |
1988 | if (ia_valid & ATTR_CTIME) { | |
1989 | bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME| | |
1990 | ATTR_MODE|ATTR_UID|ATTR_GID)) == 0; | |
1991 | dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode, | |
1992 | inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec, | |
1993 | attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec, | |
1994 | only ? "ctime only" : "ignored"); | |
1995 | inode->i_ctime = attr->ia_ctime; | |
1996 | if (only) { | |
1997 | /* | |
1998 | * if kernel wants to dirty ctime but nothing else, | |
1999 | * we need to choose a cap to dirty under, or do | |
2000 | * a almost-no-op setattr | |
2001 | */ | |
2002 | if (issued & CEPH_CAP_AUTH_EXCL) | |
2003 | dirtied |= CEPH_CAP_AUTH_EXCL; | |
2004 | else if (issued & CEPH_CAP_FILE_EXCL) | |
2005 | dirtied |= CEPH_CAP_FILE_EXCL; | |
2006 | else if (issued & CEPH_CAP_XATTR_EXCL) | |
2007 | dirtied |= CEPH_CAP_XATTR_EXCL; | |
2008 | else | |
2009 | mask |= CEPH_SETATTR_CTIME; | |
2010 | } | |
2011 | } | |
2012 | if (ia_valid & ATTR_FILE) | |
2013 | dout("setattr %p ATTR_FILE ... hrm!\n", inode); | |
2014 | ||
2015 | if (dirtied) { | |
f66fd9f0 YZ |
2016 | inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied, |
2017 | &prealloc_cf); | |
8bbd4714 | 2018 | inode->i_ctime = current_fs_time(inode->i_sb); |
355da1eb SW |
2019 | } |
2020 | ||
2021 | release &= issued; | |
be655596 | 2022 | spin_unlock(&ci->i_ceph_lock); |
604d1b02 YZ |
2023 | if (lock_snap_rwsem) |
2024 | up_read(&mdsc->snap_rwsem); | |
355da1eb | 2025 | |
fca65b4a SW |
2026 | if (inode_dirty_flags) |
2027 | __mark_inode_dirty(inode, inode_dirty_flags); | |
2028 | ||
7221fe4c | 2029 | if (ia_valid & ATTR_MODE) { |
4db658ea | 2030 | err = posix_acl_chmod(inode, attr->ia_mode); |
7221fe4c GZ |
2031 | if (err) |
2032 | goto out_put; | |
2033 | } | |
2034 | ||
355da1eb | 2035 | if (mask) { |
70b666c3 SW |
2036 | req->r_inode = inode; |
2037 | ihold(inode); | |
355da1eb SW |
2038 | req->r_inode_drop = release; |
2039 | req->r_args.setattr.mask = cpu_to_le32(mask); | |
2040 | req->r_num_caps = 1; | |
752c8bdc | 2041 | err = ceph_mdsc_do_request(mdsc, NULL, req); |
355da1eb SW |
2042 | } |
2043 | dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err, | |
2044 | ceph_cap_string(dirtied), mask); | |
2045 | ||
2046 | ceph_mdsc_put_request(req); | |
b0d7c223 YZ |
2047 | if (mask & CEPH_SETATTR_SIZE) |
2048 | __ceph_do_pending_vmtruncate(inode); | |
f66fd9f0 | 2049 | ceph_free_cap_flush(prealloc_cf); |
355da1eb | 2050 | return err; |
7221fe4c | 2051 | out_put: |
355da1eb | 2052 | ceph_mdsc_put_request(req); |
f66fd9f0 | 2053 | ceph_free_cap_flush(prealloc_cf); |
355da1eb SW |
2054 | return err; |
2055 | } | |
2056 | ||
2057 | /* | |
2058 | * Verify that we have a lease on the given mask. If not, | |
2059 | * do a getattr against an mds. | |
2060 | */ | |
01deead0 YZ |
2061 | int __ceph_do_getattr(struct inode *inode, struct page *locked_page, |
2062 | int mask, bool force) | |
355da1eb | 2063 | { |
3d14c5d2 YS |
2064 | struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb); |
2065 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
355da1eb SW |
2066 | struct ceph_mds_request *req; |
2067 | int err; | |
2068 | ||
2069 | if (ceph_snap(inode) == CEPH_SNAPDIR) { | |
2070 | dout("do_getattr inode %p SNAPDIR\n", inode); | |
2071 | return 0; | |
2072 | } | |
2073 | ||
01deead0 YZ |
2074 | dout("do_getattr inode %p mask %s mode 0%o\n", |
2075 | inode, ceph_cap_string(mask), inode->i_mode); | |
508b32d8 | 2076 | if (!force && ceph_caps_issued_mask(ceph_inode(inode), mask, 1)) |
355da1eb SW |
2077 | return 0; |
2078 | ||
2079 | req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS); | |
2080 | if (IS_ERR(req)) | |
2081 | return PTR_ERR(req); | |
70b666c3 SW |
2082 | req->r_inode = inode; |
2083 | ihold(inode); | |
355da1eb SW |
2084 | req->r_num_caps = 1; |
2085 | req->r_args.getattr.mask = cpu_to_le32(mask); | |
01deead0 | 2086 | req->r_locked_page = locked_page; |
355da1eb | 2087 | err = ceph_mdsc_do_request(mdsc, NULL, req); |
01deead0 YZ |
2088 | if (locked_page && err == 0) { |
2089 | u64 inline_version = req->r_reply_info.targeti.inline_version; | |
2090 | if (inline_version == 0) { | |
2091 | /* the reply is supposed to contain inline data */ | |
2092 | err = -EINVAL; | |
2093 | } else if (inline_version == CEPH_INLINE_NONE) { | |
2094 | err = -ENODATA; | |
2095 | } else { | |
2096 | err = req->r_reply_info.targeti.inline_len; | |
2097 | } | |
2098 | } | |
355da1eb SW |
2099 | ceph_mdsc_put_request(req); |
2100 | dout("do_getattr result=%d\n", err); | |
2101 | return err; | |
2102 | } | |
2103 | ||
2104 | ||
2105 | /* | |
2106 | * Check inode permissions. We verify we have a valid value for | |
2107 | * the AUTH cap, then call the generic handler. | |
2108 | */ | |
10556cb2 | 2109 | int ceph_permission(struct inode *inode, int mask) |
355da1eb | 2110 | { |
b74c79e9 NP |
2111 | int err; |
2112 | ||
10556cb2 | 2113 | if (mask & MAY_NOT_BLOCK) |
b74c79e9 NP |
2114 | return -ECHILD; |
2115 | ||
508b32d8 | 2116 | err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED, false); |
355da1eb SW |
2117 | |
2118 | if (!err) | |
2830ba7f | 2119 | err = generic_permission(inode, mask); |
355da1eb SW |
2120 | return err; |
2121 | } | |
2122 | ||
2123 | /* | |
2124 | * Get all attributes. Hopefully somedata we'll have a statlite() | |
2125 | * and can limit the fields we require to be accurate. | |
2126 | */ | |
2127 | int ceph_getattr(struct vfsmount *mnt, struct dentry *dentry, | |
2128 | struct kstat *stat) | |
2129 | { | |
2b0143b5 | 2130 | struct inode *inode = d_inode(dentry); |
232d4b01 | 2131 | struct ceph_inode_info *ci = ceph_inode(inode); |
355da1eb SW |
2132 | int err; |
2133 | ||
508b32d8 | 2134 | err = ceph_do_getattr(inode, CEPH_STAT_CAP_INODE_ALL, false); |
355da1eb SW |
2135 | if (!err) { |
2136 | generic_fillattr(inode, stat); | |
ad1fee96 | 2137 | stat->ino = ceph_translate_ino(inode->i_sb, inode->i_ino); |
355da1eb SW |
2138 | if (ceph_snap(inode) != CEPH_NOSNAP) |
2139 | stat->dev = ceph_snap(inode); | |
2140 | else | |
2141 | stat->dev = 0; | |
232d4b01 | 2142 | if (S_ISDIR(inode->i_mode)) { |
1c1266bb YS |
2143 | if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb), |
2144 | RBYTES)) | |
2145 | stat->size = ci->i_rbytes; | |
2146 | else | |
2147 | stat->size = ci->i_files + ci->i_subdirs; | |
232d4b01 | 2148 | stat->blocks = 0; |
355da1eb | 2149 | stat->blksize = 65536; |
232d4b01 | 2150 | } |
355da1eb SW |
2151 | } |
2152 | return err; | |
2153 | } |