]>
Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
3d14c5d2 | 2 | #include <linux/ceph/ceph_debug.h> |
355da1eb SW |
3 | |
4 | #include <linux/module.h> | |
5 | #include <linux/fs.h> | |
355da1eb SW |
6 | #include <linux/slab.h> |
7 | #include <linux/string.h> | |
8 | #include <linux/uaccess.h> | |
9 | #include <linux/kernel.h> | |
355da1eb SW |
10 | #include <linux/writeback.h> |
11 | #include <linux/vmalloc.h> | |
2cdeb1e4 | 12 | #include <linux/xattr.h> |
4db658ea | 13 | #include <linux/posix_acl.h> |
3e7fbe9c | 14 | #include <linux/random.h> |
a407846e | 15 | #include <linux/sort.h> |
355da1eb SW |
16 | |
17 | #include "super.h" | |
3d14c5d2 | 18 | #include "mds_client.h" |
99ccbd22 | 19 | #include "cache.h" |
3d14c5d2 | 20 | #include <linux/ceph/decode.h> |
355da1eb SW |
21 | |
22 | /* | |
23 | * Ceph inode operations | |
24 | * | |
25 | * Implement basic inode helpers (get, alloc) and inode ops (getattr, | |
26 | * setattr, etc.), xattr helpers, and helpers for assimilating | |
27 | * metadata returned by the MDS into our cache. | |
28 | * | |
29 | * Also define helpers for doing asynchronous writeback, invalidation, | |
30 | * and truncation for the benefit of those who can't afford to block | |
31 | * (typically because they are in the message handler path). | |
32 | */ | |
33 | ||
34 | static const struct inode_operations ceph_symlink_iops; | |
35 | ||
d91016e8 | 36 | static void ceph_inode_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); | |
d37b1d99 | 54 | if (!inode) |
355da1eb SW |
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, | |
355da1eb | 96 | .listxattr = ceph_listxattr, |
7221fe4c | 97 | .get_acl = ceph_get_acl, |
72466d0b | 98 | .set_acl = ceph_set_acl, |
355da1eb SW |
99 | }; |
100 | ||
101 | ||
102 | /* | |
103 | * We use a 'frag tree' to keep track of the MDS's directory fragments | |
104 | * for a given inode (usually there is just a single fragment). We | |
105 | * need to know when a child frag is delegated to a new MDS, or when | |
106 | * it is flagged as replicated, so we can direct our requests | |
107 | * accordingly. | |
108 | */ | |
109 | ||
110 | /* | |
111 | * find/create a frag in the tree | |
112 | */ | |
113 | static struct ceph_inode_frag *__get_or_create_frag(struct ceph_inode_info *ci, | |
114 | u32 f) | |
115 | { | |
116 | struct rb_node **p; | |
117 | struct rb_node *parent = NULL; | |
118 | struct ceph_inode_frag *frag; | |
119 | int c; | |
120 | ||
121 | p = &ci->i_fragtree.rb_node; | |
122 | while (*p) { | |
123 | parent = *p; | |
124 | frag = rb_entry(parent, struct ceph_inode_frag, node); | |
125 | c = ceph_frag_compare(f, frag->frag); | |
126 | if (c < 0) | |
127 | p = &(*p)->rb_left; | |
128 | else if (c > 0) | |
129 | p = &(*p)->rb_right; | |
130 | else | |
131 | return frag; | |
132 | } | |
133 | ||
134 | frag = kmalloc(sizeof(*frag), GFP_NOFS); | |
51308806 | 135 | if (!frag) |
355da1eb | 136 | return ERR_PTR(-ENOMEM); |
51308806 | 137 | |
355da1eb SW |
138 | frag->frag = f; |
139 | frag->split_by = 0; | |
140 | frag->mds = -1; | |
141 | frag->ndist = 0; | |
142 | ||
143 | rb_link_node(&frag->node, parent, p); | |
144 | rb_insert_color(&frag->node, &ci->i_fragtree); | |
145 | ||
146 | dout("get_or_create_frag added %llx.%llx frag %x\n", | |
147 | ceph_vinop(&ci->vfs_inode), f); | |
148 | return frag; | |
149 | } | |
150 | ||
151 | /* | |
152 | * find a specific frag @f | |
153 | */ | |
154 | struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci, u32 f) | |
155 | { | |
156 | struct rb_node *n = ci->i_fragtree.rb_node; | |
157 | ||
158 | while (n) { | |
159 | struct ceph_inode_frag *frag = | |
160 | rb_entry(n, struct ceph_inode_frag, node); | |
161 | int c = ceph_frag_compare(f, frag->frag); | |
162 | if (c < 0) | |
163 | n = n->rb_left; | |
164 | else if (c > 0) | |
165 | n = n->rb_right; | |
166 | else | |
167 | return frag; | |
168 | } | |
169 | return NULL; | |
170 | } | |
171 | ||
172 | /* | |
173 | * Choose frag containing the given value @v. If @pfrag is | |
174 | * specified, copy the frag delegation info to the caller if | |
175 | * it is present. | |
176 | */ | |
3e7fbe9c YZ |
177 | static u32 __ceph_choose_frag(struct ceph_inode_info *ci, u32 v, |
178 | struct ceph_inode_frag *pfrag, int *found) | |
355da1eb SW |
179 | { |
180 | u32 t = ceph_frag_make(0, 0); | |
181 | struct ceph_inode_frag *frag; | |
182 | unsigned nway, i; | |
183 | u32 n; | |
184 | ||
185 | if (found) | |
186 | *found = 0; | |
187 | ||
355da1eb SW |
188 | while (1) { |
189 | WARN_ON(!ceph_frag_contains_value(t, v)); | |
190 | frag = __ceph_find_frag(ci, t); | |
191 | if (!frag) | |
192 | break; /* t is a leaf */ | |
193 | if (frag->split_by == 0) { | |
194 | if (pfrag) | |
195 | memcpy(pfrag, frag, sizeof(*pfrag)); | |
196 | if (found) | |
197 | *found = 1; | |
198 | break; | |
199 | } | |
200 | ||
201 | /* choose child */ | |
202 | nway = 1 << frag->split_by; | |
203 | dout("choose_frag(%x) %x splits by %d (%d ways)\n", v, t, | |
204 | frag->split_by, nway); | |
205 | for (i = 0; i < nway; i++) { | |
206 | n = ceph_frag_make_child(t, frag->split_by, i); | |
207 | if (ceph_frag_contains_value(n, v)) { | |
208 | t = n; | |
209 | break; | |
210 | } | |
211 | } | |
212 | BUG_ON(i == nway); | |
213 | } | |
214 | dout("choose_frag(%x) = %x\n", v, t); | |
215 | ||
355da1eb SW |
216 | return t; |
217 | } | |
218 | ||
3e7fbe9c YZ |
219 | u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v, |
220 | struct ceph_inode_frag *pfrag, int *found) | |
221 | { | |
222 | u32 ret; | |
223 | mutex_lock(&ci->i_fragtree_mutex); | |
224 | ret = __ceph_choose_frag(ci, v, pfrag, found); | |
225 | mutex_unlock(&ci->i_fragtree_mutex); | |
226 | return ret; | |
227 | } | |
228 | ||
355da1eb SW |
229 | /* |
230 | * Process dirfrag (delegation) info from the mds. Include leaf | |
231 | * fragment in tree ONLY if ndist > 0. Otherwise, only | |
232 | * branches/splits are included in i_fragtree) | |
233 | */ | |
234 | static int ceph_fill_dirfrag(struct inode *inode, | |
235 | struct ceph_mds_reply_dirfrag *dirinfo) | |
236 | { | |
237 | struct ceph_inode_info *ci = ceph_inode(inode); | |
238 | struct ceph_inode_frag *frag; | |
239 | u32 id = le32_to_cpu(dirinfo->frag); | |
240 | int mds = le32_to_cpu(dirinfo->auth); | |
241 | int ndist = le32_to_cpu(dirinfo->ndist); | |
8d08503c | 242 | int diri_auth = -1; |
355da1eb SW |
243 | int i; |
244 | int err = 0; | |
245 | ||
8d08503c YZ |
246 | spin_lock(&ci->i_ceph_lock); |
247 | if (ci->i_auth_cap) | |
248 | diri_auth = ci->i_auth_cap->mds; | |
249 | spin_unlock(&ci->i_ceph_lock); | |
250 | ||
42172119 YZ |
251 | if (mds == -1) /* CDIR_AUTH_PARENT */ |
252 | mds = diri_auth; | |
253 | ||
355da1eb | 254 | mutex_lock(&ci->i_fragtree_mutex); |
8d08503c | 255 | if (ndist == 0 && mds == diri_auth) { |
355da1eb SW |
256 | /* no delegation info needed. */ |
257 | frag = __ceph_find_frag(ci, id); | |
258 | if (!frag) | |
259 | goto out; | |
260 | if (frag->split_by == 0) { | |
261 | /* tree leaf, remove */ | |
262 | dout("fill_dirfrag removed %llx.%llx frag %x" | |
263 | " (no ref)\n", ceph_vinop(inode), id); | |
264 | rb_erase(&frag->node, &ci->i_fragtree); | |
265 | kfree(frag); | |
266 | } else { | |
267 | /* tree branch, keep and clear */ | |
268 | dout("fill_dirfrag cleared %llx.%llx frag %x" | |
269 | " referral\n", ceph_vinop(inode), id); | |
270 | frag->mds = -1; | |
271 | frag->ndist = 0; | |
272 | } | |
273 | goto out; | |
274 | } | |
275 | ||
276 | ||
277 | /* find/add this frag to store mds delegation info */ | |
278 | frag = __get_or_create_frag(ci, id); | |
279 | if (IS_ERR(frag)) { | |
280 | /* this is not the end of the world; we can continue | |
281 | with bad/inaccurate delegation info */ | |
282 | pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n", | |
283 | ceph_vinop(inode), le32_to_cpu(dirinfo->frag)); | |
284 | err = -ENOMEM; | |
285 | goto out; | |
286 | } | |
287 | ||
288 | frag->mds = mds; | |
289 | frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP); | |
290 | for (i = 0; i < frag->ndist; i++) | |
291 | frag->dist[i] = le32_to_cpu(dirinfo->dist[i]); | |
292 | dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n", | |
293 | ceph_vinop(inode), frag->frag, frag->ndist); | |
294 | ||
295 | out: | |
296 | mutex_unlock(&ci->i_fragtree_mutex); | |
297 | return err; | |
298 | } | |
299 | ||
a407846e YZ |
300 | static int frag_tree_split_cmp(const void *l, const void *r) |
301 | { | |
302 | struct ceph_frag_tree_split *ls = (struct ceph_frag_tree_split*)l; | |
303 | struct ceph_frag_tree_split *rs = (struct ceph_frag_tree_split*)r; | |
fe2ed425 JL |
304 | return ceph_frag_compare(le32_to_cpu(ls->frag), |
305 | le32_to_cpu(rs->frag)); | |
a407846e YZ |
306 | } |
307 | ||
a4b7431f YZ |
308 | static bool is_frag_child(u32 f, struct ceph_inode_frag *frag) |
309 | { | |
310 | if (!frag) | |
311 | return f == ceph_frag_make(0, 0); | |
312 | if (ceph_frag_bits(f) != ceph_frag_bits(frag->frag) + frag->split_by) | |
313 | return false; | |
314 | return ceph_frag_contains_value(frag->frag, ceph_frag_value(f)); | |
315 | } | |
316 | ||
3e7fbe9c YZ |
317 | static int ceph_fill_fragtree(struct inode *inode, |
318 | struct ceph_frag_tree_head *fragtree, | |
319 | struct ceph_mds_reply_dirfrag *dirinfo) | |
320 | { | |
321 | struct ceph_inode_info *ci = ceph_inode(inode); | |
a4b7431f | 322 | struct ceph_inode_frag *frag, *prev_frag = NULL; |
3e7fbe9c | 323 | struct rb_node *rb_node; |
1b1bc16d YZ |
324 | unsigned i, split_by, nsplits; |
325 | u32 id; | |
3e7fbe9c YZ |
326 | bool update = false; |
327 | ||
328 | mutex_lock(&ci->i_fragtree_mutex); | |
329 | nsplits = le32_to_cpu(fragtree->nsplits); | |
1b1bc16d YZ |
330 | if (nsplits != ci->i_fragtree_nsplits) { |
331 | update = true; | |
332 | } else if (nsplits) { | |
3e7fbe9c YZ |
333 | i = prandom_u32() % nsplits; |
334 | id = le32_to_cpu(fragtree->splits[i].frag); | |
335 | if (!__ceph_find_frag(ci, id)) | |
336 | update = true; | |
337 | } else if (!RB_EMPTY_ROOT(&ci->i_fragtree)) { | |
338 | rb_node = rb_first(&ci->i_fragtree); | |
339 | frag = rb_entry(rb_node, struct ceph_inode_frag, node); | |
340 | if (frag->frag != ceph_frag_make(0, 0) || rb_next(rb_node)) | |
341 | update = true; | |
342 | } | |
343 | if (!update && dirinfo) { | |
344 | id = le32_to_cpu(dirinfo->frag); | |
345 | if (id != __ceph_choose_frag(ci, id, NULL, NULL)) | |
346 | update = true; | |
347 | } | |
348 | if (!update) | |
349 | goto out_unlock; | |
350 | ||
a407846e YZ |
351 | if (nsplits > 1) { |
352 | sort(fragtree->splits, nsplits, sizeof(fragtree->splits[0]), | |
353 | frag_tree_split_cmp, NULL); | |
354 | } | |
355 | ||
3e7fbe9c YZ |
356 | dout("fill_fragtree %llx.%llx\n", ceph_vinop(inode)); |
357 | rb_node = rb_first(&ci->i_fragtree); | |
358 | for (i = 0; i < nsplits; i++) { | |
359 | id = le32_to_cpu(fragtree->splits[i].frag); | |
1b1bc16d YZ |
360 | split_by = le32_to_cpu(fragtree->splits[i].by); |
361 | if (split_by == 0 || ceph_frag_bits(id) + split_by > 24) { | |
362 | pr_err("fill_fragtree %llx.%llx invalid split %d/%u, " | |
363 | "frag %x split by %d\n", ceph_vinop(inode), | |
364 | i, nsplits, id, split_by); | |
365 | continue; | |
366 | } | |
3e7fbe9c YZ |
367 | frag = NULL; |
368 | while (rb_node) { | |
369 | frag = rb_entry(rb_node, struct ceph_inode_frag, node); | |
370 | if (ceph_frag_compare(frag->frag, id) >= 0) { | |
371 | if (frag->frag != id) | |
372 | frag = NULL; | |
373 | else | |
374 | rb_node = rb_next(rb_node); | |
375 | break; | |
376 | } | |
377 | rb_node = rb_next(rb_node); | |
a4b7431f YZ |
378 | /* delete stale split/leaf node */ |
379 | if (frag->split_by > 0 || | |
380 | !is_frag_child(frag->frag, prev_frag)) { | |
381 | rb_erase(&frag->node, &ci->i_fragtree); | |
1b1bc16d YZ |
382 | if (frag->split_by > 0) |
383 | ci->i_fragtree_nsplits--; | |
a4b7431f YZ |
384 | kfree(frag); |
385 | } | |
3e7fbe9c YZ |
386 | frag = NULL; |
387 | } | |
388 | if (!frag) { | |
389 | frag = __get_or_create_frag(ci, id); | |
390 | if (IS_ERR(frag)) | |
391 | continue; | |
392 | } | |
1b1bc16d YZ |
393 | if (frag->split_by == 0) |
394 | ci->i_fragtree_nsplits++; | |
395 | frag->split_by = split_by; | |
3e7fbe9c | 396 | dout(" frag %x split by %d\n", frag->frag, frag->split_by); |
a4b7431f | 397 | prev_frag = frag; |
3e7fbe9c YZ |
398 | } |
399 | while (rb_node) { | |
400 | frag = rb_entry(rb_node, struct ceph_inode_frag, node); | |
401 | rb_node = rb_next(rb_node); | |
a4b7431f YZ |
402 | /* delete stale split/leaf node */ |
403 | if (frag->split_by > 0 || | |
404 | !is_frag_child(frag->frag, prev_frag)) { | |
405 | rb_erase(&frag->node, &ci->i_fragtree); | |
1b1bc16d YZ |
406 | if (frag->split_by > 0) |
407 | ci->i_fragtree_nsplits--; | |
a4b7431f YZ |
408 | kfree(frag); |
409 | } | |
3e7fbe9c YZ |
410 | } |
411 | out_unlock: | |
412 | mutex_unlock(&ci->i_fragtree_mutex); | |
413 | return 0; | |
414 | } | |
355da1eb SW |
415 | |
416 | /* | |
417 | * initialize a newly allocated inode. | |
418 | */ | |
419 | struct inode *ceph_alloc_inode(struct super_block *sb) | |
420 | { | |
421 | struct ceph_inode_info *ci; | |
422 | int i; | |
423 | ||
424 | ci = kmem_cache_alloc(ceph_inode_cachep, GFP_NOFS); | |
425 | if (!ci) | |
426 | return NULL; | |
427 | ||
428 | dout("alloc_inode %p\n", &ci->vfs_inode); | |
429 | ||
be655596 SW |
430 | spin_lock_init(&ci->i_ceph_lock); |
431 | ||
355da1eb | 432 | ci->i_version = 0; |
31c542a1 | 433 | ci->i_inline_version = 0; |
355da1eb SW |
434 | ci->i_time_warp_seq = 0; |
435 | ci->i_ceph_flags = 0; | |
fdd4e158 YZ |
436 | atomic64_set(&ci->i_ordered_count, 1); |
437 | atomic64_set(&ci->i_release_count, 1); | |
438 | atomic64_set(&ci->i_complete_seq[0], 0); | |
439 | atomic64_set(&ci->i_complete_seq[1], 0); | |
355da1eb SW |
440 | ci->i_symlink = NULL; |
441 | ||
6c0f3af7 | 442 | memset(&ci->i_dir_layout, 0, sizeof(ci->i_dir_layout)); |
30c156d9 | 443 | RCU_INIT_POINTER(ci->i_layout.pool_ns, NULL); |
6c0f3af7 | 444 | |
355da1eb SW |
445 | ci->i_fragtree = RB_ROOT; |
446 | mutex_init(&ci->i_fragtree_mutex); | |
447 | ||
448 | ci->i_xattrs.blob = NULL; | |
449 | ci->i_xattrs.prealloc_blob = NULL; | |
450 | ci->i_xattrs.dirty = false; | |
451 | ci->i_xattrs.index = RB_ROOT; | |
452 | ci->i_xattrs.count = 0; | |
453 | ci->i_xattrs.names_size = 0; | |
454 | ci->i_xattrs.vals_size = 0; | |
455 | ci->i_xattrs.version = 0; | |
456 | ci->i_xattrs.index_version = 0; | |
457 | ||
458 | ci->i_caps = RB_ROOT; | |
459 | ci->i_auth_cap = NULL; | |
460 | ci->i_dirty_caps = 0; | |
461 | ci->i_flushing_caps = 0; | |
462 | INIT_LIST_HEAD(&ci->i_dirty_item); | |
463 | INIT_LIST_HEAD(&ci->i_flushing_item); | |
f66fd9f0 | 464 | ci->i_prealloc_cap_flush = NULL; |
e4500b5e | 465 | INIT_LIST_HEAD(&ci->i_cap_flush_list); |
355da1eb SW |
466 | init_waitqueue_head(&ci->i_cap_wq); |
467 | ci->i_hold_caps_min = 0; | |
468 | ci->i_hold_caps_max = 0; | |
469 | INIT_LIST_HEAD(&ci->i_cap_delay_list); | |
355da1eb SW |
470 | INIT_LIST_HEAD(&ci->i_cap_snaps); |
471 | ci->i_head_snapc = NULL; | |
472 | ci->i_snap_caps = 0; | |
473 | ||
774a6a11 | 474 | for (i = 0; i < CEPH_FILE_MODE_BITS; i++) |
355da1eb SW |
475 | ci->i_nr_by_mode[i] = 0; |
476 | ||
b0d7c223 | 477 | mutex_init(&ci->i_truncate_mutex); |
355da1eb SW |
478 | ci->i_truncate_seq = 0; |
479 | ci->i_truncate_size = 0; | |
480 | ci->i_truncate_pending = 0; | |
481 | ||
482 | ci->i_max_size = 0; | |
483 | ci->i_reported_size = 0; | |
484 | ci->i_wanted_max_size = 0; | |
485 | ci->i_requested_max_size = 0; | |
486 | ||
487 | ci->i_pin_ref = 0; | |
488 | ci->i_rd_ref = 0; | |
489 | ci->i_rdcache_ref = 0; | |
490 | ci->i_wr_ref = 0; | |
d3d0720d | 491 | ci->i_wb_ref = 0; |
355da1eb SW |
492 | ci->i_wrbuffer_ref = 0; |
493 | ci->i_wrbuffer_ref_head = 0; | |
89aa5930 | 494 | atomic_set(&ci->i_filelock_ref, 0); |
7389f7b2 | 495 | atomic_set(&ci->i_shared_gen, 0); |
355da1eb SW |
496 | ci->i_rdcache_gen = 0; |
497 | ci->i_rdcache_revoking = 0; | |
498 | ||
355da1eb | 499 | INIT_LIST_HEAD(&ci->i_unsafe_dirops); |
68cd5b4b | 500 | INIT_LIST_HEAD(&ci->i_unsafe_iops); |
355da1eb SW |
501 | spin_lock_init(&ci->i_unsafe_lock); |
502 | ||
503 | ci->i_snap_realm = NULL; | |
504 | INIT_LIST_HEAD(&ci->i_snap_realm_item); | |
505 | INIT_LIST_HEAD(&ci->i_snap_flush_item); | |
506 | ||
d91016e8 YZ |
507 | INIT_WORK(&ci->i_work, ceph_inode_work); |
508 | ci->i_work_mask = 0; | |
355da1eb | 509 | |
99ccbd22 MT |
510 | ceph_fscache_inode_init(ci); |
511 | ||
355da1eb SW |
512 | return &ci->vfs_inode; |
513 | } | |
514 | ||
fa0d7e3d NP |
515 | static void ceph_i_callback(struct rcu_head *head) |
516 | { | |
517 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
518 | struct ceph_inode_info *ci = ceph_inode(inode); | |
519 | ||
fa0d7e3d NP |
520 | kmem_cache_free(ceph_inode_cachep, ci); |
521 | } | |
522 | ||
355da1eb SW |
523 | void ceph_destroy_inode(struct inode *inode) |
524 | { | |
525 | struct ceph_inode_info *ci = ceph_inode(inode); | |
526 | struct ceph_inode_frag *frag; | |
527 | struct rb_node *n; | |
528 | ||
529 | dout("destroy_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode)); | |
530 | ||
99ccbd22 MT |
531 | ceph_fscache_unregister_inode_cookie(ci); |
532 | ||
1de82206 | 533 | __ceph_remove_caps(inode); |
355da1eb | 534 | |
8b218b8a SW |
535 | /* |
536 | * we may still have a snap_realm reference if there are stray | |
d9df2783 | 537 | * caps in i_snap_caps. |
8b218b8a SW |
538 | */ |
539 | if (ci->i_snap_realm) { | |
540 | struct ceph_mds_client *mdsc = | |
3d14c5d2 | 541 | ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc; |
8b218b8a SW |
542 | struct ceph_snap_realm *realm = ci->i_snap_realm; |
543 | ||
544 | dout(" dropping residual ref to snap realm %p\n", realm); | |
545 | spin_lock(&realm->inodes_with_caps_lock); | |
546 | list_del_init(&ci->i_snap_realm_item); | |
547 | spin_unlock(&realm->inodes_with_caps_lock); | |
548 | ceph_put_snap_realm(mdsc, realm); | |
549 | } | |
550 | ||
355da1eb SW |
551 | kfree(ci->i_symlink); |
552 | while ((n = rb_first(&ci->i_fragtree)) != NULL) { | |
553 | frag = rb_entry(n, struct ceph_inode_frag, node); | |
554 | rb_erase(n, &ci->i_fragtree); | |
555 | kfree(frag); | |
556 | } | |
1b1bc16d | 557 | ci->i_fragtree_nsplits = 0; |
355da1eb SW |
558 | |
559 | __ceph_destroy_xattrs(ci); | |
b6c1d5b8 SW |
560 | if (ci->i_xattrs.blob) |
561 | ceph_buffer_put(ci->i_xattrs.blob); | |
562 | if (ci->i_xattrs.prealloc_blob) | |
563 | ceph_buffer_put(ci->i_xattrs.prealloc_blob); | |
355da1eb | 564 | |
779fe0fb | 565 | ceph_put_string(rcu_dereference_raw(ci->i_layout.pool_ns)); |
30c156d9 | 566 | |
fa0d7e3d | 567 | call_rcu(&inode->i_rcu, ceph_i_callback); |
355da1eb SW |
568 | } |
569 | ||
9f12bd11 YZ |
570 | int ceph_drop_inode(struct inode *inode) |
571 | { | |
572 | /* | |
573 | * Positve dentry and corresponding inode are always accompanied | |
574 | * in MDS reply. So no need to keep inode in the cache after | |
575 | * dropping all its aliases. | |
576 | */ | |
577 | return 1; | |
578 | } | |
579 | ||
224a7542 YZ |
580 | static inline blkcnt_t calc_inode_blocks(u64 size) |
581 | { | |
582 | return (size + (1<<9) - 1) >> 9; | |
583 | } | |
584 | ||
355da1eb SW |
585 | /* |
586 | * Helpers to fill in size, ctime, mtime, and atime. We have to be | |
587 | * careful because either the client or MDS may have more up to date | |
588 | * info, depending on which capabilities are held, and whether | |
589 | * time_warp_seq or truncate_seq have increased. (Ordinarily, mtime | |
590 | * and size are monotonically increasing, except when utimes() or | |
591 | * truncate() increments the corresponding _seq values.) | |
592 | */ | |
593 | int ceph_fill_file_size(struct inode *inode, int issued, | |
594 | u32 truncate_seq, u64 truncate_size, u64 size) | |
595 | { | |
596 | struct ceph_inode_info *ci = ceph_inode(inode); | |
597 | int queue_trunc = 0; | |
598 | ||
599 | if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 || | |
600 | (truncate_seq == ci->i_truncate_seq && size > inode->i_size)) { | |
601 | dout("size %lld -> %llu\n", inode->i_size, size); | |
a3d714c3 YZ |
602 | if (size > 0 && S_ISDIR(inode->i_mode)) { |
603 | pr_err("fill_file_size non-zero size for directory\n"); | |
604 | size = 0; | |
605 | } | |
99c88e69 | 606 | i_size_write(inode, size); |
224a7542 | 607 | inode->i_blocks = calc_inode_blocks(size); |
355da1eb SW |
608 | ci->i_reported_size = size; |
609 | if (truncate_seq != ci->i_truncate_seq) { | |
610 | dout("truncate_seq %u -> %u\n", | |
611 | ci->i_truncate_seq, truncate_seq); | |
612 | ci->i_truncate_seq = truncate_seq; | |
b0d7c223 YZ |
613 | |
614 | /* the MDS should have revoked these caps */ | |
615 | WARN_ON_ONCE(issued & (CEPH_CAP_FILE_EXCL | | |
616 | CEPH_CAP_FILE_RD | | |
617 | CEPH_CAP_FILE_WR | | |
618 | CEPH_CAP_FILE_LAZYIO)); | |
3d497d85 YS |
619 | /* |
620 | * If we hold relevant caps, or in the case where we're | |
621 | * not the only client referencing this file and we | |
622 | * don't hold those caps, then we need to check whether | |
623 | * the file is either opened or mmaped | |
624 | */ | |
b0d7c223 YZ |
625 | if ((issued & (CEPH_CAP_FILE_CACHE| |
626 | CEPH_CAP_FILE_BUFFER)) || | |
3d497d85 YS |
627 | mapping_mapped(inode->i_mapping) || |
628 | __ceph_caps_file_wanted(ci)) { | |
355da1eb SW |
629 | ci->i_truncate_pending++; |
630 | queue_trunc = 1; | |
631 | } | |
632 | } | |
633 | } | |
634 | if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 && | |
635 | ci->i_truncate_size != truncate_size) { | |
636 | dout("truncate_size %lld -> %llu\n", ci->i_truncate_size, | |
637 | truncate_size); | |
638 | ci->i_truncate_size = truncate_size; | |
639 | } | |
99ccbd22 MT |
640 | |
641 | if (queue_trunc) | |
642 | ceph_fscache_invalidate(inode); | |
643 | ||
355da1eb SW |
644 | return queue_trunc; |
645 | } | |
646 | ||
647 | void ceph_fill_file_time(struct inode *inode, int issued, | |
648 | u64 time_warp_seq, struct timespec *ctime, | |
649 | struct timespec *mtime, struct timespec *atime) | |
650 | { | |
651 | struct ceph_inode_info *ci = ceph_inode(inode); | |
652 | int warn = 0; | |
653 | ||
654 | if (issued & (CEPH_CAP_FILE_EXCL| | |
655 | CEPH_CAP_FILE_WR| | |
d8672d64 SW |
656 | CEPH_CAP_FILE_BUFFER| |
657 | CEPH_CAP_AUTH_EXCL| | |
658 | CEPH_CAP_XATTR_EXCL)) { | |
fb64b047 YZ |
659 | if (ci->i_version == 0 || |
660 | timespec_compare(ctime, &inode->i_ctime) > 0) { | |
355da1eb SW |
661 | dout("ctime %ld.%09ld -> %ld.%09ld inc w/ cap\n", |
662 | inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec, | |
663 | ctime->tv_sec, ctime->tv_nsec); | |
664 | inode->i_ctime = *ctime; | |
665 | } | |
fb64b047 YZ |
666 | if (ci->i_version == 0 || |
667 | ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) { | |
355da1eb SW |
668 | /* the MDS did a utimes() */ |
669 | dout("mtime %ld.%09ld -> %ld.%09ld " | |
670 | "tw %d -> %d\n", | |
671 | inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec, | |
672 | mtime->tv_sec, mtime->tv_nsec, | |
673 | ci->i_time_warp_seq, (int)time_warp_seq); | |
674 | ||
675 | inode->i_mtime = *mtime; | |
676 | inode->i_atime = *atime; | |
677 | ci->i_time_warp_seq = time_warp_seq; | |
678 | } else if (time_warp_seq == ci->i_time_warp_seq) { | |
679 | /* nobody did utimes(); take the max */ | |
680 | if (timespec_compare(mtime, &inode->i_mtime) > 0) { | |
681 | dout("mtime %ld.%09ld -> %ld.%09ld inc\n", | |
682 | inode->i_mtime.tv_sec, | |
683 | inode->i_mtime.tv_nsec, | |
684 | mtime->tv_sec, mtime->tv_nsec); | |
685 | inode->i_mtime = *mtime; | |
686 | } | |
687 | if (timespec_compare(atime, &inode->i_atime) > 0) { | |
688 | dout("atime %ld.%09ld -> %ld.%09ld inc\n", | |
689 | inode->i_atime.tv_sec, | |
690 | inode->i_atime.tv_nsec, | |
691 | atime->tv_sec, atime->tv_nsec); | |
692 | inode->i_atime = *atime; | |
693 | } | |
694 | } else if (issued & CEPH_CAP_FILE_EXCL) { | |
695 | /* we did a utimes(); ignore mds values */ | |
696 | } else { | |
697 | warn = 1; | |
698 | } | |
699 | } else { | |
d8672d64 | 700 | /* we have no write|excl caps; whatever the MDS says is true */ |
355da1eb SW |
701 | if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) { |
702 | inode->i_ctime = *ctime; | |
703 | inode->i_mtime = *mtime; | |
704 | inode->i_atime = *atime; | |
705 | ci->i_time_warp_seq = time_warp_seq; | |
706 | } else { | |
707 | warn = 1; | |
708 | } | |
709 | } | |
710 | if (warn) /* time_warp_seq shouldn't go backwards */ | |
711 | dout("%p mds time_warp_seq %llu < %u\n", | |
712 | inode, time_warp_seq, ci->i_time_warp_seq); | |
713 | } | |
714 | ||
715 | /* | |
716 | * Populate an inode based on info from mds. May be called on new or | |
717 | * existing inodes. | |
718 | */ | |
01deead0 | 719 | static int fill_inode(struct inode *inode, struct page *locked_page, |
355da1eb SW |
720 | struct ceph_mds_reply_info_in *iinfo, |
721 | struct ceph_mds_reply_dirfrag *dirinfo, | |
722 | struct ceph_mds_session *session, | |
723 | unsigned long ttl_from, int cap_fmode, | |
724 | struct ceph_cap_reservation *caps_reservation) | |
725 | { | |
d9df2783 | 726 | struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc; |
355da1eb SW |
727 | struct ceph_mds_reply_inode *info = iinfo->in; |
728 | struct ceph_inode_info *ci = ceph_inode(inode); | |
f98a128a | 729 | int issued = 0, implemented, new_issued; |
355da1eb | 730 | struct timespec mtime, atime, ctime; |
355da1eb | 731 | struct ceph_buffer *xattr_blob = NULL; |
779fe0fb | 732 | struct ceph_string *pool_ns = NULL; |
d9df2783 | 733 | struct ceph_cap *new_cap = NULL; |
355da1eb | 734 | int err = 0; |
d9df2783 | 735 | bool wake = false; |
f98a128a YZ |
736 | bool queue_trunc = false; |
737 | bool new_version = false; | |
31c542a1 | 738 | bool fill_inline = false; |
355da1eb SW |
739 | |
740 | dout("fill_inode %p ino %llx.%llx v %llu had %llu\n", | |
741 | inode, ceph_vinop(inode), le64_to_cpu(info->version), | |
742 | ci->i_version); | |
743 | ||
d9df2783 YZ |
744 | /* prealloc new cap struct */ |
745 | if (info->cap.caps && ceph_snap(inode) == CEPH_NOSNAP) | |
746 | new_cap = ceph_get_cap(mdsc, caps_reservation); | |
747 | ||
355da1eb SW |
748 | /* |
749 | * prealloc xattr data, if it looks like we'll need it. only | |
750 | * if len > 4 (meaning there are actually xattrs; the first 4 | |
751 | * bytes are the xattr count). | |
752 | */ | |
753 | if (iinfo->xattr_len > 4) { | |
b6c1d5b8 | 754 | xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS); |
355da1eb SW |
755 | if (!xattr_blob) |
756 | pr_err("fill_inode ENOMEM xattr blob %d bytes\n", | |
757 | iinfo->xattr_len); | |
758 | } | |
759 | ||
779fe0fb YZ |
760 | if (iinfo->pool_ns_len > 0) |
761 | pool_ns = ceph_find_or_create_string(iinfo->pool_ns_data, | |
762 | iinfo->pool_ns_len); | |
763 | ||
be655596 | 764 | spin_lock(&ci->i_ceph_lock); |
355da1eb SW |
765 | |
766 | /* | |
767 | * provided version will be odd if inode value is projected, | |
8bd59e01 SW |
768 | * even if stable. skip the update if we have newer stable |
769 | * info (ours>=theirs, e.g. due to racing mds replies), unless | |
770 | * we are getting projected (unstable) info (in which case the | |
771 | * version is odd, and we want ours>theirs). | |
772 | * us them | |
773 | * 2 2 skip | |
774 | * 3 2 skip | |
775 | * 3 3 update | |
355da1eb | 776 | */ |
f98a128a YZ |
777 | if (ci->i_version == 0 || |
778 | ((info->cap.flags & CEPH_CAP_FLAG_AUTH) && | |
779 | le64_to_cpu(info->version) > (ci->i_version & ~1))) | |
780 | new_version = true; | |
781 | ||
355da1eb SW |
782 | issued = __ceph_caps_issued(ci, &implemented); |
783 | issued |= implemented | __ceph_caps_dirty(ci); | |
f98a128a | 784 | new_issued = ~issued & le32_to_cpu(info->cap.caps); |
355da1eb SW |
785 | |
786 | /* update inode */ | |
355da1eb | 787 | inode->i_rdev = le32_to_cpu(info->rdev); |
f98a128a | 788 | inode->i_blkbits = fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1; |
355da1eb | 789 | |
f98a128a YZ |
790 | if ((new_version || (new_issued & CEPH_CAP_AUTH_SHARED)) && |
791 | (issued & CEPH_CAP_AUTH_EXCL) == 0) { | |
355da1eb | 792 | inode->i_mode = le32_to_cpu(info->mode); |
ab871b90 EB |
793 | inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(info->uid)); |
794 | inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(info->gid)); | |
355da1eb | 795 | dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode, |
bd2bae6a EB |
796 | from_kuid(&init_user_ns, inode->i_uid), |
797 | from_kgid(&init_user_ns, inode->i_gid)); | |
355da1eb SW |
798 | } |
799 | ||
f98a128a YZ |
800 | if ((new_version || (new_issued & CEPH_CAP_LINK_SHARED)) && |
801 | (issued & CEPH_CAP_LINK_EXCL) == 0) | |
bfe86848 | 802 | set_nlink(inode, le32_to_cpu(info->nlink)); |
355da1eb | 803 | |
f98a128a YZ |
804 | if (new_version || (new_issued & CEPH_CAP_ANY_RD)) { |
805 | /* be careful with mtime, atime, size */ | |
806 | ceph_decode_timespec(&atime, &info->atime); | |
807 | ceph_decode_timespec(&mtime, &info->mtime); | |
808 | ceph_decode_timespec(&ctime, &info->ctime); | |
809 | ceph_fill_file_time(inode, issued, | |
810 | le32_to_cpu(info->time_warp_seq), | |
811 | &ctime, &mtime, &atime); | |
812 | } | |
813 | ||
814 | if (new_version || | |
815 | (new_issued & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR))) { | |
7627151e | 816 | s64 old_pool = ci->i_layout.pool_id; |
779fe0fb YZ |
817 | struct ceph_string *old_ns; |
818 | ||
7627151e | 819 | ceph_file_layout_from_legacy(&ci->i_layout, &info->layout); |
779fe0fb YZ |
820 | old_ns = rcu_dereference_protected(ci->i_layout.pool_ns, |
821 | lockdep_is_held(&ci->i_ceph_lock)); | |
822 | rcu_assign_pointer(ci->i_layout.pool_ns, pool_ns); | |
823 | ||
824 | if (ci->i_layout.pool_id != old_pool || pool_ns != old_ns) | |
10183a69 | 825 | ci->i_ceph_flags &= ~CEPH_I_POOL_PERM; |
10183a69 | 826 | |
779fe0fb | 827 | pool_ns = old_ns; |
10183a69 | 828 | |
f98a128a YZ |
829 | queue_trunc = ceph_fill_file_size(inode, issued, |
830 | le32_to_cpu(info->truncate_seq), | |
831 | le64_to_cpu(info->truncate_size), | |
832 | le64_to_cpu(info->size)); | |
833 | /* only update max_size on auth cap */ | |
834 | if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) && | |
835 | ci->i_max_size != le64_to_cpu(info->max_size)) { | |
836 | dout("max_size %lld -> %llu\n", ci->i_max_size, | |
837 | le64_to_cpu(info->max_size)); | |
838 | ci->i_max_size = le64_to_cpu(info->max_size); | |
839 | } | |
840 | } | |
355da1eb SW |
841 | |
842 | /* xattrs */ | |
843 | /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */ | |
508b32d8 | 844 | if ((ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL)) && |
355da1eb SW |
845 | le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) { |
846 | if (ci->i_xattrs.blob) | |
847 | ceph_buffer_put(ci->i_xattrs.blob); | |
848 | ci->i_xattrs.blob = xattr_blob; | |
849 | if (xattr_blob) | |
850 | memcpy(ci->i_xattrs.blob->vec.iov_base, | |
851 | iinfo->xattr_data, iinfo->xattr_len); | |
852 | ci->i_xattrs.version = le64_to_cpu(info->xattr_version); | |
7221fe4c | 853 | ceph_forget_all_cached_acls(inode); |
a6424e48 | 854 | xattr_blob = NULL; |
355da1eb SW |
855 | } |
856 | ||
fb64b047 YZ |
857 | /* finally update i_version */ |
858 | ci->i_version = le64_to_cpu(info->version); | |
859 | ||
355da1eb | 860 | inode->i_mapping->a_ops = &ceph_aops; |
355da1eb SW |
861 | |
862 | switch (inode->i_mode & S_IFMT) { | |
863 | case S_IFIFO: | |
864 | case S_IFBLK: | |
865 | case S_IFCHR: | |
866 | case S_IFSOCK: | |
867 | init_special_inode(inode, inode->i_mode, inode->i_rdev); | |
868 | inode->i_op = &ceph_file_iops; | |
869 | break; | |
870 | case S_IFREG: | |
871 | inode->i_op = &ceph_file_iops; | |
872 | inode->i_fop = &ceph_file_fops; | |
873 | break; | |
874 | case S_IFLNK: | |
875 | inode->i_op = &ceph_symlink_iops; | |
876 | if (!ci->i_symlink) { | |
810339ec | 877 | u32 symlen = iinfo->symlink_len; |
355da1eb SW |
878 | char *sym; |
879 | ||
be655596 | 880 | spin_unlock(&ci->i_ceph_lock); |
355da1eb | 881 | |
224a7542 YZ |
882 | if (symlen != i_size_read(inode)) { |
883 | pr_err("fill_inode %llx.%llx BAD symlink " | |
884 | "size %lld\n", ceph_vinop(inode), | |
885 | i_size_read(inode)); | |
886 | i_size_write(inode, symlen); | |
887 | inode->i_blocks = calc_inode_blocks(symlen); | |
888 | } | |
810339ec | 889 | |
355da1eb | 890 | err = -ENOMEM; |
810339ec | 891 | sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS); |
355da1eb SW |
892 | if (!sym) |
893 | goto out; | |
355da1eb | 894 | |
be655596 | 895 | spin_lock(&ci->i_ceph_lock); |
355da1eb SW |
896 | if (!ci->i_symlink) |
897 | ci->i_symlink = sym; | |
898 | else | |
899 | kfree(sym); /* lost a race */ | |
900 | } | |
ac194dcc | 901 | inode->i_link = ci->i_symlink; |
355da1eb SW |
902 | break; |
903 | case S_IFDIR: | |
904 | inode->i_op = &ceph_dir_iops; | |
905 | inode->i_fop = &ceph_dir_fops; | |
906 | ||
14303d20 SW |
907 | ci->i_dir_layout = iinfo->dir_layout; |
908 | ||
355da1eb SW |
909 | ci->i_files = le64_to_cpu(info->files); |
910 | ci->i_subdirs = le64_to_cpu(info->subdirs); | |
911 | ci->i_rbytes = le64_to_cpu(info->rbytes); | |
912 | ci->i_rfiles = le64_to_cpu(info->rfiles); | |
913 | ci->i_rsubdirs = le64_to_cpu(info->rsubdirs); | |
914 | ceph_decode_timespec(&ci->i_rctime, &info->rctime); | |
355da1eb SW |
915 | break; |
916 | default: | |
917 | pr_err("fill_inode %llx.%llx BAD mode 0%o\n", | |
918 | ceph_vinop(inode), inode->i_mode); | |
919 | } | |
920 | ||
355da1eb SW |
921 | /* were we issued a capability? */ |
922 | if (info->cap.caps) { | |
923 | if (ceph_snap(inode) == CEPH_NOSNAP) { | |
2f92b3d0 | 924 | unsigned caps = le32_to_cpu(info->cap.caps); |
355da1eb SW |
925 | ceph_add_cap(inode, session, |
926 | le64_to_cpu(info->cap.cap_id), | |
2f92b3d0 | 927 | cap_fmode, caps, |
355da1eb SW |
928 | le32_to_cpu(info->cap.wanted), |
929 | le32_to_cpu(info->cap.seq), | |
930 | le32_to_cpu(info->cap.mseq), | |
931 | le64_to_cpu(info->cap.realm), | |
d9df2783 | 932 | info->cap.flags, &new_cap); |
2f92b3d0 YZ |
933 | |
934 | /* set dir completion flag? */ | |
935 | if (S_ISDIR(inode->i_mode) && | |
936 | ci->i_files == 0 && ci->i_subdirs == 0 && | |
937 | (caps & CEPH_CAP_FILE_SHARED) && | |
938 | (issued & CEPH_CAP_FILE_EXCL) == 0 && | |
939 | !__ceph_dir_is_complete(ci)) { | |
940 | dout(" marking %p complete (empty)\n", inode); | |
fdd4e158 | 941 | i_size_write(inode, 0); |
2f92b3d0 | 942 | __ceph_dir_set_complete(ci, |
fdd4e158 YZ |
943 | atomic64_read(&ci->i_release_count), |
944 | atomic64_read(&ci->i_ordered_count)); | |
2f92b3d0 YZ |
945 | } |
946 | ||
d9df2783 | 947 | wake = true; |
355da1eb | 948 | } else { |
355da1eb SW |
949 | dout(" %p got snap_caps %s\n", inode, |
950 | ceph_cap_string(le32_to_cpu(info->cap.caps))); | |
951 | ci->i_snap_caps |= le32_to_cpu(info->cap.caps); | |
952 | if (cap_fmode >= 0) | |
953 | __ceph_get_fmode(ci, cap_fmode); | |
355da1eb | 954 | } |
04d000eb | 955 | } else if (cap_fmode >= 0) { |
f3ae1b97 | 956 | pr_warn("mds issued no caps on %llx.%llx\n", |
04d000eb SW |
957 | ceph_vinop(inode)); |
958 | __ceph_get_fmode(ci, cap_fmode); | |
355da1eb | 959 | } |
31c542a1 YZ |
960 | |
961 | if (iinfo->inline_version > 0 && | |
962 | iinfo->inline_version >= ci->i_inline_version) { | |
963 | int cache_caps = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO; | |
964 | ci->i_inline_version = iinfo->inline_version; | |
965 | if (ci->i_inline_version != CEPH_INLINE_NONE && | |
01deead0 YZ |
966 | (locked_page || |
967 | (le32_to_cpu(info->cap.caps) & cache_caps))) | |
31c542a1 YZ |
968 | fill_inline = true; |
969 | } | |
970 | ||
be655596 | 971 | spin_unlock(&ci->i_ceph_lock); |
355da1eb | 972 | |
31c542a1 | 973 | if (fill_inline) |
01deead0 | 974 | ceph_fill_inline_data(inode, locked_page, |
31c542a1 YZ |
975 | iinfo->inline_data, iinfo->inline_len); |
976 | ||
d9df2783 YZ |
977 | if (wake) |
978 | wake_up_all(&ci->i_cap_wq); | |
979 | ||
355da1eb SW |
980 | /* queue truncate if we saw i_size decrease */ |
981 | if (queue_trunc) | |
3c6f6b79 | 982 | ceph_queue_vmtruncate(inode); |
355da1eb SW |
983 | |
984 | /* populate frag tree */ | |
3e7fbe9c YZ |
985 | if (S_ISDIR(inode->i_mode)) |
986 | ceph_fill_fragtree(inode, &info->fragtree, dirinfo); | |
355da1eb SW |
987 | |
988 | /* update delegation info? */ | |
989 | if (dirinfo) | |
990 | ceph_fill_dirfrag(inode, dirinfo); | |
991 | ||
992 | err = 0; | |
355da1eb | 993 | out: |
d9df2783 YZ |
994 | if (new_cap) |
995 | ceph_put_cap(mdsc, new_cap); | |
b6c1d5b8 SW |
996 | if (xattr_blob) |
997 | ceph_buffer_put(xattr_blob); | |
779fe0fb | 998 | ceph_put_string(pool_ns); |
355da1eb SW |
999 | return err; |
1000 | } | |
1001 | ||
1002 | /* | |
1003 | * caller should hold session s_mutex. | |
1004 | */ | |
1005 | static void update_dentry_lease(struct dentry *dentry, | |
1006 | struct ceph_mds_reply_lease *lease, | |
1007 | struct ceph_mds_session *session, | |
f5d55f03 JL |
1008 | unsigned long from_time, |
1009 | struct ceph_vino *tgt_vino, | |
1010 | struct ceph_vino *dir_vino) | |
355da1eb SW |
1011 | { |
1012 | struct ceph_dentry_info *di = ceph_dentry(dentry); | |
1013 | long unsigned duration = le32_to_cpu(lease->duration_ms); | |
1014 | long unsigned ttl = from_time + (duration * HZ) / 1000; | |
1015 | long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000; | |
1016 | struct inode *dir; | |
481f001f | 1017 | struct ceph_mds_session *old_lease_session = NULL; |
355da1eb | 1018 | |
f5d55f03 JL |
1019 | /* |
1020 | * Make sure dentry's inode matches tgt_vino. NULL tgt_vino means that | |
1021 | * we expect a negative dentry. | |
1022 | */ | |
1023 | if (!tgt_vino && d_really_is_positive(dentry)) | |
1024 | return; | |
1025 | ||
1026 | if (tgt_vino && (d_really_is_negative(dentry) || | |
1027 | !ceph_ino_compare(d_inode(dentry), tgt_vino))) | |
1028 | return; | |
1029 | ||
355da1eb | 1030 | spin_lock(&dentry->d_lock); |
2f90b852 SW |
1031 | dout("update_dentry_lease %p duration %lu ms ttl %lu\n", |
1032 | dentry, duration, ttl); | |
355da1eb | 1033 | |
2b0143b5 | 1034 | dir = d_inode(dentry->d_parent); |
18fc8abd | 1035 | |
f5d55f03 JL |
1036 | /* make sure parent matches dir_vino */ |
1037 | if (!ceph_ino_compare(dir, dir_vino)) | |
1038 | goto out_unlock; | |
1039 | ||
18fc8abd AV |
1040 | /* only track leases on regular dentries */ |
1041 | if (ceph_snap(dir) != CEPH_NOSNAP) | |
1042 | goto out_unlock; | |
1043 | ||
7389f7b2 | 1044 | di->lease_shared_gen = atomic_read(&ceph_inode(dir)->i_shared_gen); |
355da1eb | 1045 | |
2f90b852 | 1046 | if (duration == 0) |
355da1eb SW |
1047 | goto out_unlock; |
1048 | ||
1049 | if (di->lease_gen == session->s_cap_gen && | |
9b16f03c | 1050 | time_before(ttl, di->time)) |
355da1eb SW |
1051 | goto out_unlock; /* we already have a newer lease. */ |
1052 | ||
481f001f YZ |
1053 | if (di->lease_session && di->lease_session != session) { |
1054 | old_lease_session = di->lease_session; | |
1055 | di->lease_session = NULL; | |
1056 | } | |
355da1eb SW |
1057 | |
1058 | ceph_dentry_lru_touch(dentry); | |
1059 | ||
1060 | if (!di->lease_session) | |
1061 | di->lease_session = ceph_get_mds_session(session); | |
1062 | di->lease_gen = session->s_cap_gen; | |
1063 | di->lease_seq = le32_to_cpu(lease->seq); | |
1064 | di->lease_renew_after = half_ttl; | |
1065 | di->lease_renew_from = 0; | |
9b16f03c | 1066 | di->time = ttl; |
355da1eb SW |
1067 | out_unlock: |
1068 | spin_unlock(&dentry->d_lock); | |
481f001f YZ |
1069 | if (old_lease_session) |
1070 | ceph_put_mds_session(old_lease_session); | |
355da1eb SW |
1071 | } |
1072 | ||
1073 | /* | |
1074 | * splice a dentry to an inode. | |
1075 | * caller must hold directory i_mutex for this to be safe. | |
355da1eb | 1076 | */ |
f7380af0 | 1077 | static struct dentry *splice_dentry(struct dentry *dn, struct inode *in) |
355da1eb SW |
1078 | { |
1079 | struct dentry *realdn; | |
1080 | ||
2b0143b5 | 1081 | BUG_ON(d_inode(dn)); |
1cd3935b | 1082 | |
48697e8a YZ |
1083 | if (S_ISDIR(in->i_mode)) { |
1084 | /* If inode is directory, d_splice_alias() below will remove | |
1085 | * 'realdn' from its origin parent. We need to ensure that | |
1086 | * origin parent's readdir cache will not reference 'realdn' | |
1087 | */ | |
1088 | realdn = d_find_any_alias(in); | |
1089 | if (realdn) { | |
1090 | struct ceph_dentry_info *di = ceph_dentry(realdn); | |
1091 | spin_lock(&realdn->d_lock); | |
1092 | ||
1093 | realdn->d_op->d_prune(realdn); | |
1094 | ||
1095 | di->time = jiffies; | |
1096 | di->lease_shared_gen = 0; | |
1097 | di->offset = 0; | |
1098 | ||
1099 | spin_unlock(&realdn->d_lock); | |
1100 | dput(realdn); | |
1101 | } | |
1102 | } | |
1103 | ||
355da1eb SW |
1104 | /* dn must be unhashed */ |
1105 | if (!d_unhashed(dn)) | |
1106 | d_drop(dn); | |
41d28bca | 1107 | realdn = d_splice_alias(in, dn); |
355da1eb | 1108 | if (IS_ERR(realdn)) { |
d69ed05a SW |
1109 | pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n", |
1110 | PTR_ERR(realdn), dn, in, ceph_vinop(in)); | |
355da1eb SW |
1111 | dn = realdn; /* note realdn contains the error */ |
1112 | goto out; | |
1113 | } else if (realdn) { | |
1114 | dout("dn %p (%d) spliced with %p (%d) " | |
1115 | "inode %p ino %llx.%llx\n", | |
84d08fa8 AV |
1116 | dn, d_count(dn), |
1117 | realdn, d_count(realdn), | |
2b0143b5 | 1118 | d_inode(realdn), ceph_vinop(d_inode(realdn))); |
355da1eb SW |
1119 | dput(dn); |
1120 | dn = realdn; | |
1121 | } else { | |
1122 | BUG_ON(!ceph_dentry(dn)); | |
355da1eb | 1123 | dout("dn %p attached to %p ino %llx.%llx\n", |
2b0143b5 | 1124 | dn, d_inode(dn), ceph_vinop(d_inode(dn))); |
355da1eb | 1125 | } |
355da1eb SW |
1126 | out: |
1127 | return dn; | |
1128 | } | |
1129 | ||
1130 | /* | |
1131 | * Incorporate results into the local cache. This is either just | |
1132 | * one inode, or a directory, dentry, and possibly linked-to inode (e.g., | |
1133 | * after a lookup). | |
1134 | * | |
1135 | * A reply may contain | |
1136 | * a directory inode along with a dentry. | |
1137 | * and/or a target inode | |
1138 | * | |
1139 | * Called with snap_rwsem (read). | |
1140 | */ | |
f5a03b08 | 1141 | int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req) |
355da1eb | 1142 | { |
f5a03b08 | 1143 | struct ceph_mds_session *session = req->r_session; |
355da1eb SW |
1144 | struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info; |
1145 | struct inode *in = NULL; | |
f5d55f03 | 1146 | struct ceph_vino tvino, dvino; |
3d14c5d2 | 1147 | struct ceph_fs_client *fsc = ceph_sb_to_client(sb); |
355da1eb SW |
1148 | int err = 0; |
1149 | ||
1150 | dout("fill_trace %p is_dentry %d is_target %d\n", req, | |
1151 | rinfo->head->is_dentry, rinfo->head->is_target); | |
1152 | ||
355da1eb SW |
1153 | if (!rinfo->head->is_target && !rinfo->head->is_dentry) { |
1154 | dout("fill_trace reply is empty!\n"); | |
3dd69aab | 1155 | if (rinfo->head->result == 0 && req->r_parent) |
167c9e35 | 1156 | ceph_invalidate_dir_request(req); |
355da1eb SW |
1157 | return 0; |
1158 | } | |
1159 | ||
1160 | if (rinfo->head->is_dentry) { | |
3dd69aab | 1161 | struct inode *dir = req->r_parent; |
5b1daecd | 1162 | |
6c5e50fa | 1163 | if (dir) { |
01deead0 YZ |
1164 | err = fill_inode(dir, NULL, |
1165 | &rinfo->diri, rinfo->dirfrag, | |
6c5e50fa SW |
1166 | session, req->r_request_started, -1, |
1167 | &req->r_caps_reservation); | |
1168 | if (err < 0) | |
19913b4e | 1169 | goto done; |
6c5e50fa SW |
1170 | } else { |
1171 | WARN_ON_ONCE(1); | |
1172 | } | |
19913b4e YZ |
1173 | |
1174 | if (dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME) { | |
1175 | struct qstr dname; | |
1176 | struct dentry *dn, *parent; | |
1177 | ||
1178 | BUG_ON(!rinfo->head->is_target); | |
1179 | BUG_ON(req->r_dentry); | |
1180 | ||
1181 | parent = d_find_any_alias(dir); | |
1182 | BUG_ON(!parent); | |
1183 | ||
1184 | dname.name = rinfo->dname; | |
1185 | dname.len = rinfo->dname_len; | |
8387ff25 | 1186 | dname.hash = full_name_hash(parent, dname.name, dname.len); |
f5d55f03 JL |
1187 | tvino.ino = le64_to_cpu(rinfo->targeti.in->ino); |
1188 | tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid); | |
19913b4e YZ |
1189 | retry_lookup: |
1190 | dn = d_lookup(parent, &dname); | |
1191 | dout("d_lookup on parent=%p name=%.*s got %p\n", | |
1192 | parent, dname.len, dname.name, dn); | |
1193 | ||
1194 | if (!dn) { | |
1195 | dn = d_alloc(parent, &dname); | |
1196 | dout("d_alloc %p '%.*s' = %p\n", parent, | |
1197 | dname.len, dname.name, dn); | |
d37b1d99 | 1198 | if (!dn) { |
19913b4e YZ |
1199 | dput(parent); |
1200 | err = -ENOMEM; | |
1201 | goto done; | |
1202 | } | |
ad5cb123 | 1203 | err = 0; |
2b0143b5 | 1204 | } else if (d_really_is_positive(dn) && |
f5d55f03 JL |
1205 | (ceph_ino(d_inode(dn)) != tvino.ino || |
1206 | ceph_snap(d_inode(dn)) != tvino.snap)) { | |
19913b4e | 1207 | dout(" dn %p points to wrong inode %p\n", |
2b0143b5 | 1208 | dn, d_inode(dn)); |
933ad2c9 | 1209 | ceph_dir_clear_ordered(dir); |
19913b4e YZ |
1210 | d_delete(dn); |
1211 | dput(dn); | |
1212 | goto retry_lookup; | |
1213 | } | |
1214 | ||
1215 | req->r_dentry = dn; | |
1216 | dput(parent); | |
1217 | } | |
5b1daecd SW |
1218 | } |
1219 | ||
86b58d13 | 1220 | if (rinfo->head->is_target) { |
f5d55f03 JL |
1221 | tvino.ino = le64_to_cpu(rinfo->targeti.in->ino); |
1222 | tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid); | |
86b58d13 | 1223 | |
f5d55f03 | 1224 | in = ceph_get_inode(sb, tvino); |
86b58d13 YZ |
1225 | if (IS_ERR(in)) { |
1226 | err = PTR_ERR(in); | |
1227 | goto done; | |
1228 | } | |
1229 | req->r_target_inode = in; | |
1230 | ||
01deead0 | 1231 | err = fill_inode(in, req->r_locked_page, &rinfo->targeti, NULL, |
86b58d13 | 1232 | session, req->r_request_started, |
bc2de10d JL |
1233 | (!test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) && |
1234 | rinfo->head->result == 0) ? req->r_fmode : -1, | |
86b58d13 YZ |
1235 | &req->r_caps_reservation); |
1236 | if (err < 0) { | |
1237 | pr_err("fill_inode badness %p %llx.%llx\n", | |
1238 | in, ceph_vinop(in)); | |
1239 | goto done; | |
1240 | } | |
1241 | } | |
1242 | ||
9358c6d4 SW |
1243 | /* |
1244 | * ignore null lease/binding on snapdir ENOENT, or else we | |
1245 | * will have trouble splicing in the virtual snapdir later | |
1246 | */ | |
3dd69aab JL |
1247 | if (rinfo->head->is_dentry && |
1248 | !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) && | |
1249 | test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) && | |
9358c6d4 | 1250 | (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name, |
3d14c5d2 | 1251 | fsc->mount_options->snapdir_name, |
9358c6d4 | 1252 | req->r_dentry->d_name.len))) { |
355da1eb SW |
1253 | /* |
1254 | * lookup link rename : null -> possibly existing inode | |
1255 | * mknod symlink mkdir : null -> new inode | |
1256 | * unlink : linked -> null | |
1257 | */ | |
3dd69aab | 1258 | struct inode *dir = req->r_parent; |
355da1eb SW |
1259 | struct dentry *dn = req->r_dentry; |
1260 | bool have_dir_cap, have_lease; | |
1261 | ||
1262 | BUG_ON(!dn); | |
1263 | BUG_ON(!dir); | |
2b0143b5 | 1264 | BUG_ON(d_inode(dn->d_parent) != dir); |
f5d55f03 JL |
1265 | |
1266 | dvino.ino = le64_to_cpu(rinfo->diri.in->ino); | |
1267 | dvino.snap = le64_to_cpu(rinfo->diri.in->snapid); | |
1268 | ||
1269 | BUG_ON(ceph_ino(dir) != dvino.ino); | |
1270 | BUG_ON(ceph_snap(dir) != dvino.snap); | |
355da1eb | 1271 | |
355da1eb SW |
1272 | /* do we have a lease on the whole dir? */ |
1273 | have_dir_cap = | |
1274 | (le32_to_cpu(rinfo->diri.in->cap.caps) & | |
1275 | CEPH_CAP_FILE_SHARED); | |
1276 | ||
1277 | /* do we have a dn lease? */ | |
1278 | have_lease = have_dir_cap || | |
2f90b852 | 1279 | le32_to_cpu(rinfo->dlease->duration_ms); |
355da1eb SW |
1280 | if (!have_lease) |
1281 | dout("fill_trace no dentry lease or dir cap\n"); | |
1282 | ||
1283 | /* rename? */ | |
1284 | if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) { | |
0a8a70f9 YZ |
1285 | struct inode *olddir = req->r_old_dentry_dir; |
1286 | BUG_ON(!olddir); | |
1287 | ||
a455589f | 1288 | dout(" src %p '%pd' dst %p '%pd'\n", |
355da1eb | 1289 | req->r_old_dentry, |
a455589f AV |
1290 | req->r_old_dentry, |
1291 | dn, dn); | |
355da1eb SW |
1292 | dout("fill_trace doing d_move %p -> %p\n", |
1293 | req->r_old_dentry, dn); | |
c10f5e12 | 1294 | |
fdd4e158 YZ |
1295 | /* d_move screws up sibling dentries' offsets */ |
1296 | ceph_dir_clear_ordered(dir); | |
1297 | ceph_dir_clear_ordered(olddir); | |
1298 | ||
355da1eb | 1299 | d_move(req->r_old_dentry, dn); |
a455589f AV |
1300 | dout(" src %p '%pd' dst %p '%pd'\n", |
1301 | req->r_old_dentry, | |
355da1eb | 1302 | req->r_old_dentry, |
a455589f | 1303 | dn, dn); |
81a6cf2d | 1304 | |
c4a29f26 SW |
1305 | /* ensure target dentry is invalidated, despite |
1306 | rehashing bug in vfs_rename_dir */ | |
81a6cf2d SW |
1307 | ceph_invalidate_dentry_lease(dn); |
1308 | ||
99ccbd22 | 1309 | dout("dn %p gets new offset %lld\n", req->r_old_dentry, |
1cd3935b | 1310 | ceph_dentry(req->r_old_dentry)->offset); |
81a6cf2d | 1311 | |
355da1eb | 1312 | dn = req->r_old_dentry; /* use old_dentry */ |
355da1eb SW |
1313 | } |
1314 | ||
1315 | /* null dentry? */ | |
1316 | if (!rinfo->head->is_target) { | |
1317 | dout("fill_trace null dentry\n"); | |
2b0143b5 | 1318 | if (d_really_is_positive(dn)) { |
355da1eb | 1319 | dout("d_delete %p\n", dn); |
48697e8a | 1320 | ceph_dir_clear_ordered(dir); |
355da1eb | 1321 | d_delete(dn); |
80d025ff JL |
1322 | } else if (have_lease) { |
1323 | if (d_unhashed(dn)) | |
f8b31710 | 1324 | d_add(dn, NULL); |
355da1eb SW |
1325 | update_dentry_lease(dn, rinfo->dlease, |
1326 | session, | |
f5d55f03 JL |
1327 | req->r_request_started, |
1328 | NULL, &dvino); | |
355da1eb SW |
1329 | } |
1330 | goto done; | |
1331 | } | |
1332 | ||
1333 | /* attach proper inode */ | |
2b0143b5 | 1334 | if (d_really_is_negative(dn)) { |
70db4f36 | 1335 | ceph_dir_clear_ordered(dir); |
86b58d13 | 1336 | ihold(in); |
f7380af0 | 1337 | dn = splice_dentry(dn, in); |
355da1eb SW |
1338 | if (IS_ERR(dn)) { |
1339 | err = PTR_ERR(dn); | |
1340 | goto done; | |
1341 | } | |
1342 | req->r_dentry = dn; /* may have spliced */ | |
2b0143b5 | 1343 | } else if (d_really_is_positive(dn) && d_inode(dn) != in) { |
355da1eb | 1344 | dout(" %p links to %p %llx.%llx, not %llx.%llx\n", |
2b0143b5 | 1345 | dn, d_inode(dn), ceph_vinop(d_inode(dn)), |
86b58d13 | 1346 | ceph_vinop(in)); |
200fd27c | 1347 | d_invalidate(dn); |
355da1eb | 1348 | have_lease = false; |
355da1eb SW |
1349 | } |
1350 | ||
f5d55f03 JL |
1351 | if (have_lease) { |
1352 | tvino.ino = le64_to_cpu(rinfo->targeti.in->ino); | |
1353 | tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid); | |
355da1eb | 1354 | update_dentry_lease(dn, rinfo->dlease, session, |
f5d55f03 JL |
1355 | req->r_request_started, |
1356 | &tvino, &dvino); | |
1357 | } | |
355da1eb | 1358 | dout(" final dn %p\n", dn); |
bc2de10d JL |
1359 | } else if ((req->r_op == CEPH_MDS_OP_LOOKUPSNAP || |
1360 | req->r_op == CEPH_MDS_OP_MKSNAP) && | |
1361 | !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) { | |
355da1eb | 1362 | struct dentry *dn = req->r_dentry; |
3dd69aab | 1363 | struct inode *dir = req->r_parent; |
355da1eb SW |
1364 | |
1365 | /* fill out a snapdir LOOKUPSNAP dentry */ | |
1366 | BUG_ON(!dn); | |
0a8a70f9 YZ |
1367 | BUG_ON(!dir); |
1368 | BUG_ON(ceph_snap(dir) != CEPH_SNAPDIR); | |
355da1eb | 1369 | dout(" linking snapped dir %p to dn %p\n", in, dn); |
70db4f36 | 1370 | ceph_dir_clear_ordered(dir); |
86b58d13 | 1371 | ihold(in); |
f7380af0 | 1372 | dn = splice_dentry(dn, in); |
355da1eb SW |
1373 | if (IS_ERR(dn)) { |
1374 | err = PTR_ERR(dn); | |
1375 | goto done; | |
1376 | } | |
1377 | req->r_dentry = dn; /* may have spliced */ | |
cdde7c43 JL |
1378 | } else if (rinfo->head->is_dentry) { |
1379 | struct ceph_vino *ptvino = NULL; | |
1380 | ||
1381 | if ((le32_to_cpu(rinfo->diri.in->cap.caps) & CEPH_CAP_FILE_SHARED) || | |
1382 | le32_to_cpu(rinfo->dlease->duration_ms)) { | |
1383 | dvino.ino = le64_to_cpu(rinfo->diri.in->ino); | |
1384 | dvino.snap = le64_to_cpu(rinfo->diri.in->snapid); | |
1385 | ||
1386 | if (rinfo->head->is_target) { | |
1387 | tvino.ino = le64_to_cpu(rinfo->targeti.in->ino); | |
1388 | tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid); | |
1389 | ptvino = &tvino; | |
1390 | } | |
1391 | ||
1392 | update_dentry_lease(req->r_dentry, rinfo->dlease, | |
1393 | session, req->r_request_started, ptvino, | |
1394 | &dvino); | |
1395 | } else { | |
1396 | dout("%s: no dentry lease or dir cap\n", __func__); | |
1397 | } | |
355da1eb | 1398 | } |
355da1eb SW |
1399 | done: |
1400 | dout("fill_trace done err=%d\n", err); | |
1401 | return err; | |
1402 | } | |
1403 | ||
1404 | /* | |
1405 | * Prepopulate our cache with readdir results, leases, etc. | |
1406 | */ | |
79f9f99a SW |
1407 | static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req, |
1408 | struct ceph_mds_session *session) | |
1409 | { | |
1410 | struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info; | |
1411 | int i, err = 0; | |
1412 | ||
1413 | for (i = 0; i < rinfo->dir_nr; i++) { | |
2a5beea3 | 1414 | struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i; |
79f9f99a SW |
1415 | struct ceph_vino vino; |
1416 | struct inode *in; | |
1417 | int rc; | |
1418 | ||
2a5beea3 YZ |
1419 | vino.ino = le64_to_cpu(rde->inode.in->ino); |
1420 | vino.snap = le64_to_cpu(rde->inode.in->snapid); | |
79f9f99a SW |
1421 | |
1422 | in = ceph_get_inode(req->r_dentry->d_sb, vino); | |
1423 | if (IS_ERR(in)) { | |
1424 | err = PTR_ERR(in); | |
1425 | dout("new_inode badness got %d\n", err); | |
1426 | continue; | |
1427 | } | |
2a5beea3 | 1428 | rc = fill_inode(in, NULL, &rde->inode, NULL, session, |
79f9f99a SW |
1429 | req->r_request_started, -1, |
1430 | &req->r_caps_reservation); | |
1431 | if (rc < 0) { | |
1432 | pr_err("fill_inode badness on %p got %d\n", in, rc); | |
1433 | err = rc; | |
79f9f99a | 1434 | } |
209ae762 | 1435 | iput(in); |
79f9f99a SW |
1436 | } |
1437 | ||
1438 | return err; | |
1439 | } | |
1440 | ||
fdd4e158 YZ |
1441 | void ceph_readdir_cache_release(struct ceph_readdir_cache_control *ctl) |
1442 | { | |
1443 | if (ctl->page) { | |
1444 | kunmap(ctl->page); | |
09cbfeaf | 1445 | put_page(ctl->page); |
fdd4e158 YZ |
1446 | ctl->page = NULL; |
1447 | } | |
1448 | } | |
1449 | ||
1450 | static int fill_readdir_cache(struct inode *dir, struct dentry *dn, | |
1451 | struct ceph_readdir_cache_control *ctl, | |
1452 | struct ceph_mds_request *req) | |
1453 | { | |
1454 | struct ceph_inode_info *ci = ceph_inode(dir); | |
09cbfeaf | 1455 | unsigned nsize = PAGE_SIZE / sizeof(struct dentry*); |
fdd4e158 YZ |
1456 | unsigned idx = ctl->index % nsize; |
1457 | pgoff_t pgoff = ctl->index / nsize; | |
1458 | ||
1459 | if (!ctl->page || pgoff != page_index(ctl->page)) { | |
1460 | ceph_readdir_cache_release(ctl); | |
af5e5eb5 YZ |
1461 | if (idx == 0) |
1462 | ctl->page = grab_cache_page(&dir->i_data, pgoff); | |
1463 | else | |
1464 | ctl->page = find_lock_page(&dir->i_data, pgoff); | |
fdd4e158 YZ |
1465 | if (!ctl->page) { |
1466 | ctl->index = -1; | |
af5e5eb5 | 1467 | return idx == 0 ? -ENOMEM : 0; |
fdd4e158 YZ |
1468 | } |
1469 | /* reading/filling the cache are serialized by | |
1470 | * i_mutex, no need to use page lock */ | |
1471 | unlock_page(ctl->page); | |
1472 | ctl->dentries = kmap(ctl->page); | |
af5e5eb5 | 1473 | if (idx == 0) |
09cbfeaf | 1474 | memset(ctl->dentries, 0, PAGE_SIZE); |
fdd4e158 YZ |
1475 | } |
1476 | ||
1477 | if (req->r_dir_release_cnt == atomic64_read(&ci->i_release_count) && | |
1478 | req->r_dir_ordered_cnt == atomic64_read(&ci->i_ordered_count)) { | |
1479 | dout("readdir cache dn %p idx %d\n", dn, ctl->index); | |
1480 | ctl->dentries[idx] = dn; | |
1481 | ctl->index++; | |
1482 | } else { | |
1483 | dout("disable readdir cache\n"); | |
1484 | ctl->index = -1; | |
1485 | } | |
1486 | return 0; | |
1487 | } | |
1488 | ||
355da1eb SW |
1489 | int ceph_readdir_prepopulate(struct ceph_mds_request *req, |
1490 | struct ceph_mds_session *session) | |
1491 | { | |
1492 | struct dentry *parent = req->r_dentry; | |
f3c4ebe6 | 1493 | struct ceph_inode_info *ci = ceph_inode(d_inode(parent)); |
355da1eb SW |
1494 | struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info; |
1495 | struct qstr dname; | |
1496 | struct dentry *dn; | |
1497 | struct inode *in; | |
315f2408 | 1498 | int err = 0, skipped = 0, ret, i; |
355da1eb | 1499 | struct ceph_mds_request_head *rhead = req->r_request->front.iov_base; |
81c6aea5 | 1500 | u32 frag = le32_to_cpu(rhead->args.readdir.frag); |
f3c4ebe6 YZ |
1501 | u32 last_hash = 0; |
1502 | u32 fpos_offset; | |
fdd4e158 YZ |
1503 | struct ceph_readdir_cache_control cache_ctl = {}; |
1504 | ||
bc2de10d | 1505 | if (test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) |
fdd4e158 | 1506 | return readdir_prepopulate_inodes_only(req, session); |
81c6aea5 | 1507 | |
79162547 YZ |
1508 | if (rinfo->hash_order) { |
1509 | if (req->r_path2) { | |
1510 | last_hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash, | |
1511 | req->r_path2, | |
1512 | strlen(req->r_path2)); | |
1513 | last_hash = ceph_frag_value(last_hash); | |
1514 | } else if (rinfo->offset_hash) { | |
1515 | /* mds understands offset_hash */ | |
1516 | WARN_ON_ONCE(req->r_readdir_offset != 2); | |
1517 | last_hash = le32_to_cpu(rhead->args.readdir.offset_hash); | |
1518 | } | |
f3c4ebe6 YZ |
1519 | } |
1520 | ||
81c6aea5 YZ |
1521 | if (rinfo->dir_dir && |
1522 | le32_to_cpu(rinfo->dir_dir->frag) != frag) { | |
1523 | dout("readdir_prepopulate got new frag %x -> %x\n", | |
1524 | frag, le32_to_cpu(rinfo->dir_dir->frag)); | |
1525 | frag = le32_to_cpu(rinfo->dir_dir->frag); | |
f3c4ebe6 | 1526 | if (!rinfo->hash_order) |
fdd4e158 | 1527 | req->r_readdir_offset = 2; |
81c6aea5 | 1528 | } |
355da1eb SW |
1529 | |
1530 | if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) { | |
355da1eb SW |
1531 | dout("readdir_prepopulate %d items under SNAPDIR dn %p\n", |
1532 | rinfo->dir_nr, parent); | |
1533 | } else { | |
1534 | dout("readdir_prepopulate %d items under dn %p\n", | |
1535 | rinfo->dir_nr, parent); | |
1536 | if (rinfo->dir_dir) | |
2b0143b5 | 1537 | ceph_fill_dirfrag(d_inode(parent), rinfo->dir_dir); |
355da1eb | 1538 | |
8d45b911 YZ |
1539 | if (ceph_frag_is_leftmost(frag) && |
1540 | req->r_readdir_offset == 2 && | |
1541 | !(rinfo->hash_order && last_hash)) { | |
1542 | /* note dir version at start of readdir so we can | |
1543 | * tell if any dentries get dropped */ | |
1544 | req->r_dir_release_cnt = | |
1545 | atomic64_read(&ci->i_release_count); | |
1546 | req->r_dir_ordered_cnt = | |
1547 | atomic64_read(&ci->i_ordered_count); | |
1548 | req->r_readdir_cache_idx = 0; | |
1549 | } | |
fdd4e158 YZ |
1550 | } |
1551 | ||
1552 | cache_ctl.index = req->r_readdir_cache_idx; | |
f3c4ebe6 | 1553 | fpos_offset = req->r_readdir_offset; |
fdd4e158 | 1554 | |
86b58d13 | 1555 | /* FIXME: release caps/leases if error occurs */ |
355da1eb | 1556 | for (i = 0; i < rinfo->dir_nr; i++) { |
2a5beea3 | 1557 | struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i; |
f5d55f03 | 1558 | struct ceph_vino tvino, dvino; |
355da1eb | 1559 | |
2a5beea3 YZ |
1560 | dname.name = rde->name; |
1561 | dname.len = rde->name_len; | |
8387ff25 | 1562 | dname.hash = full_name_hash(parent, dname.name, dname.len); |
355da1eb | 1563 | |
f5d55f03 JL |
1564 | tvino.ino = le64_to_cpu(rde->inode.in->ino); |
1565 | tvino.snap = le64_to_cpu(rde->inode.in->snapid); | |
355da1eb | 1566 | |
f3c4ebe6 YZ |
1567 | if (rinfo->hash_order) { |
1568 | u32 hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash, | |
1569 | rde->name, rde->name_len); | |
1570 | hash = ceph_frag_value(hash); | |
1571 | if (hash != last_hash) | |
1572 | fpos_offset = 2; | |
1573 | last_hash = hash; | |
1574 | rde->offset = ceph_make_fpos(hash, fpos_offset++, true); | |
1575 | } else { | |
1576 | rde->offset = ceph_make_fpos(frag, fpos_offset++, false); | |
1577 | } | |
355da1eb SW |
1578 | |
1579 | retry_lookup: | |
1580 | dn = d_lookup(parent, &dname); | |
1581 | dout("d_lookup on parent=%p name=%.*s got %p\n", | |
1582 | parent, dname.len, dname.name, dn); | |
1583 | ||
1584 | if (!dn) { | |
1585 | dn = d_alloc(parent, &dname); | |
1586 | dout("d_alloc %p '%.*s' = %p\n", parent, | |
1587 | dname.len, dname.name, dn); | |
d37b1d99 | 1588 | if (!dn) { |
355da1eb SW |
1589 | dout("d_alloc badness\n"); |
1590 | err = -ENOMEM; | |
1591 | goto out; | |
1592 | } | |
2b0143b5 | 1593 | } else if (d_really_is_positive(dn) && |
f5d55f03 JL |
1594 | (ceph_ino(d_inode(dn)) != tvino.ino || |
1595 | ceph_snap(d_inode(dn)) != tvino.snap)) { | |
48697e8a | 1596 | struct ceph_dentry_info *di = ceph_dentry(dn); |
355da1eb | 1597 | dout(" dn %p points to wrong inode %p\n", |
2b0143b5 | 1598 | dn, d_inode(dn)); |
48697e8a YZ |
1599 | |
1600 | spin_lock(&dn->d_lock); | |
1601 | if (di->offset > 0 && | |
1602 | di->lease_shared_gen == | |
1603 | atomic_read(&ci->i_shared_gen)) { | |
1604 | __ceph_dir_clear_ordered(ci); | |
1605 | di->offset = 0; | |
1606 | } | |
1607 | spin_unlock(&dn->d_lock); | |
1608 | ||
355da1eb SW |
1609 | d_delete(dn); |
1610 | dput(dn); | |
1611 | goto retry_lookup; | |
355da1eb SW |
1612 | } |
1613 | ||
355da1eb | 1614 | /* inode */ |
2b0143b5 DH |
1615 | if (d_really_is_positive(dn)) { |
1616 | in = d_inode(dn); | |
355da1eb | 1617 | } else { |
f5d55f03 | 1618 | in = ceph_get_inode(parent->d_sb, tvino); |
ac1f12ef | 1619 | if (IS_ERR(in)) { |
355da1eb | 1620 | dout("new_inode badness\n"); |
2744c171 | 1621 | d_drop(dn); |
355da1eb | 1622 | dput(dn); |
ac1f12ef | 1623 | err = PTR_ERR(in); |
355da1eb SW |
1624 | goto out; |
1625 | } | |
355da1eb SW |
1626 | } |
1627 | ||
2a5beea3 | 1628 | ret = fill_inode(in, NULL, &rde->inode, NULL, session, |
fdd4e158 YZ |
1629 | req->r_request_started, -1, |
1630 | &req->r_caps_reservation); | |
1631 | if (ret < 0) { | |
355da1eb | 1632 | pr_err("fill_inode badness on %p\n", in); |
48697e8a | 1633 | if (d_really_is_negative(dn)) |
86b58d13 YZ |
1634 | iput(in); |
1635 | d_drop(dn); | |
fdd4e158 | 1636 | err = ret; |
d69ed05a | 1637 | goto next_item; |
355da1eb | 1638 | } |
86b58d13 | 1639 | |
2b0143b5 | 1640 | if (d_really_is_negative(dn)) { |
315f2408 YZ |
1641 | struct dentry *realdn; |
1642 | ||
1643 | if (ceph_security_xattr_deadlock(in)) { | |
1644 | dout(" skip splicing dn %p to inode %p" | |
1645 | " (security xattr deadlock)\n", dn, in); | |
1646 | iput(in); | |
1647 | skipped++; | |
1648 | goto next_item; | |
1649 | } | |
1650 | ||
1651 | realdn = splice_dentry(dn, in); | |
5cba372c YZ |
1652 | if (IS_ERR(realdn)) { |
1653 | err = PTR_ERR(realdn); | |
1654 | d_drop(dn); | |
86b58d13 YZ |
1655 | dn = NULL; |
1656 | goto next_item; | |
1657 | } | |
5cba372c | 1658 | dn = realdn; |
86b58d13 YZ |
1659 | } |
1660 | ||
f3c4ebe6 | 1661 | ceph_dentry(dn)->offset = rde->offset; |
86b58d13 | 1662 | |
f5d55f03 | 1663 | dvino = ceph_vino(d_inode(parent)); |
2a5beea3 | 1664 | update_dentry_lease(dn, rde->lease, req->r_session, |
f5d55f03 | 1665 | req->r_request_started, &tvino, &dvino); |
fdd4e158 | 1666 | |
315f2408 | 1667 | if (err == 0 && skipped == 0 && cache_ctl.index >= 0) { |
fdd4e158 YZ |
1668 | ret = fill_readdir_cache(d_inode(parent), dn, |
1669 | &cache_ctl, req); | |
1670 | if (ret < 0) | |
1671 | err = ret; | |
1672 | } | |
d69ed05a SW |
1673 | next_item: |
1674 | if (dn) | |
1675 | dput(dn); | |
355da1eb | 1676 | } |
355da1eb | 1677 | out: |
315f2408 | 1678 | if (err == 0 && skipped == 0) { |
bc2de10d | 1679 | set_bit(CEPH_MDS_R_DID_PREPOPULATE, &req->r_req_flags); |
fdd4e158 YZ |
1680 | req->r_readdir_cache_idx = cache_ctl.index; |
1681 | } | |
1682 | ceph_readdir_cache_release(&cache_ctl); | |
355da1eb SW |
1683 | dout("readdir_prepopulate done\n"); |
1684 | return err; | |
1685 | } | |
1686 | ||
efb0ca76 | 1687 | bool ceph_inode_set_size(struct inode *inode, loff_t size) |
355da1eb SW |
1688 | { |
1689 | struct ceph_inode_info *ci = ceph_inode(inode); | |
efb0ca76 | 1690 | bool ret; |
355da1eb | 1691 | |
be655596 | 1692 | spin_lock(&ci->i_ceph_lock); |
355da1eb | 1693 | dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size); |
99c88e69 | 1694 | i_size_write(inode, size); |
224a7542 | 1695 | inode->i_blocks = calc_inode_blocks(size); |
355da1eb | 1696 | |
efb0ca76 | 1697 | ret = __ceph_should_report_size(ci); |
355da1eb | 1698 | |
be655596 | 1699 | spin_unlock(&ci->i_ceph_lock); |
355da1eb SW |
1700 | return ret; |
1701 | } | |
1702 | ||
1703 | /* | |
1704 | * Write back inode data in a worker thread. (This can't be done | |
1705 | * in the message handler context.) | |
1706 | */ | |
3c6f6b79 SW |
1707 | void ceph_queue_writeback(struct inode *inode) |
1708 | { | |
d91016e8 YZ |
1709 | struct ceph_inode_info *ci = ceph_inode(inode); |
1710 | set_bit(CEPH_I_WORK_WRITEBACK, &ci->i_work_mask); | |
1711 | ||
15a2015f | 1712 | ihold(inode); |
d91016e8 YZ |
1713 | if (queue_work(ceph_inode_to_client(inode)->inode_wq, |
1714 | &ci->i_work)) { | |
2c27c9a5 | 1715 | dout("ceph_queue_writeback %p\n", inode); |
3c6f6b79 | 1716 | } else { |
d91016e8 YZ |
1717 | dout("ceph_queue_writeback %p already queued, mask=%lx\n", |
1718 | inode, ci->i_work_mask); | |
15a2015f | 1719 | iput(inode); |
3c6f6b79 SW |
1720 | } |
1721 | } | |
1722 | ||
3c6f6b79 SW |
1723 | /* |
1724 | * queue an async invalidation | |
1725 | */ | |
1726 | void ceph_queue_invalidate(struct inode *inode) | |
1727 | { | |
d91016e8 YZ |
1728 | struct ceph_inode_info *ci = ceph_inode(inode); |
1729 | set_bit(CEPH_I_WORK_INVALIDATE_PAGES, &ci->i_work_mask); | |
1730 | ||
15a2015f | 1731 | ihold(inode); |
d91016e8 YZ |
1732 | if (queue_work(ceph_inode_to_client(inode)->inode_wq, |
1733 | &ceph_inode(inode)->i_work)) { | |
3c6f6b79 | 1734 | dout("ceph_queue_invalidate %p\n", inode); |
3c6f6b79 | 1735 | } else { |
d91016e8 YZ |
1736 | dout("ceph_queue_invalidate %p already queued, mask=%lx\n", |
1737 | inode, ci->i_work_mask); | |
15a2015f | 1738 | iput(inode); |
3c6f6b79 SW |
1739 | } |
1740 | } | |
1741 | ||
355da1eb | 1742 | /* |
d91016e8 YZ |
1743 | * Queue an async vmtruncate. If we fail to queue work, we will handle |
1744 | * the truncation the next time we call __ceph_do_pending_vmtruncate. | |
355da1eb | 1745 | */ |
d91016e8 | 1746 | void ceph_queue_vmtruncate(struct inode *inode) |
355da1eb | 1747 | { |
d91016e8 YZ |
1748 | struct ceph_inode_info *ci = ceph_inode(inode); |
1749 | set_bit(CEPH_I_WORK_VMTRUNCATE, &ci->i_work_mask); | |
1750 | ||
1751 | ihold(inode); | |
1752 | if (queue_work(ceph_inode_to_client(inode)->inode_wq, | |
1753 | &ci->i_work)) { | |
1754 | dout("ceph_queue_vmtruncate %p\n", inode); | |
1755 | } else { | |
1756 | dout("ceph_queue_vmtruncate %p already queued, mask=%lx\n", | |
1757 | inode, ci->i_work_mask); | |
1758 | iput(inode); | |
1759 | } | |
1760 | } | |
1761 | ||
1762 | static void ceph_do_invalidate_pages(struct inode *inode) | |
1763 | { | |
1764 | struct ceph_inode_info *ci = ceph_inode(inode); | |
6c93df5d | 1765 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); |
355da1eb SW |
1766 | u32 orig_gen; |
1767 | int check = 0; | |
1768 | ||
b0d7c223 | 1769 | mutex_lock(&ci->i_truncate_mutex); |
6c93df5d | 1770 | |
52953d55 | 1771 | if (READ_ONCE(fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) { |
6c93df5d YZ |
1772 | pr_warn_ratelimited("invalidate_pages %p %lld forced umount\n", |
1773 | inode, ceph_ino(inode)); | |
1774 | mapping_set_error(inode->i_mapping, -EIO); | |
1775 | truncate_pagecache(inode, 0); | |
1776 | mutex_unlock(&ci->i_truncate_mutex); | |
1777 | goto out; | |
1778 | } | |
1779 | ||
be655596 | 1780 | spin_lock(&ci->i_ceph_lock); |
355da1eb SW |
1781 | dout("invalidate_pages %p gen %d revoking %d\n", inode, |
1782 | ci->i_rdcache_gen, ci->i_rdcache_revoking); | |
cd045cb4 | 1783 | if (ci->i_rdcache_revoking != ci->i_rdcache_gen) { |
9563f88c YZ |
1784 | if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE)) |
1785 | check = 1; | |
be655596 | 1786 | spin_unlock(&ci->i_ceph_lock); |
b0d7c223 | 1787 | mutex_unlock(&ci->i_truncate_mutex); |
355da1eb SW |
1788 | goto out; |
1789 | } | |
1790 | orig_gen = ci->i_rdcache_gen; | |
be655596 | 1791 | spin_unlock(&ci->i_ceph_lock); |
355da1eb | 1792 | |
9abd4db7 YZ |
1793 | if (invalidate_inode_pages2(inode->i_mapping) < 0) { |
1794 | pr_err("invalidate_pages %p fails\n", inode); | |
1795 | } | |
355da1eb | 1796 | |
be655596 | 1797 | spin_lock(&ci->i_ceph_lock); |
cd045cb4 SW |
1798 | if (orig_gen == ci->i_rdcache_gen && |
1799 | orig_gen == ci->i_rdcache_revoking) { | |
355da1eb SW |
1800 | dout("invalidate_pages %p gen %d successful\n", inode, |
1801 | ci->i_rdcache_gen); | |
cd045cb4 | 1802 | ci->i_rdcache_revoking--; |
355da1eb SW |
1803 | check = 1; |
1804 | } else { | |
cd045cb4 SW |
1805 | dout("invalidate_pages %p gen %d raced, now %d revoking %d\n", |
1806 | inode, orig_gen, ci->i_rdcache_gen, | |
1807 | ci->i_rdcache_revoking); | |
9563f88c YZ |
1808 | if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE)) |
1809 | check = 1; | |
355da1eb | 1810 | } |
be655596 | 1811 | spin_unlock(&ci->i_ceph_lock); |
b0d7c223 | 1812 | mutex_unlock(&ci->i_truncate_mutex); |
9563f88c | 1813 | out: |
355da1eb SW |
1814 | if (check) |
1815 | ceph_check_caps(ci, 0, NULL); | |
3c6f6b79 SW |
1816 | } |
1817 | ||
355da1eb | 1818 | /* |
355da1eb SW |
1819 | * Make sure any pending truncation is applied before doing anything |
1820 | * that may depend on it. | |
1821 | */ | |
b415bf4f | 1822 | void __ceph_do_pending_vmtruncate(struct inode *inode) |
355da1eb SW |
1823 | { |
1824 | struct ceph_inode_info *ci = ceph_inode(inode); | |
1825 | u64 to; | |
a85f50b6 | 1826 | int wrbuffer_refs, finish = 0; |
355da1eb | 1827 | |
b0d7c223 | 1828 | mutex_lock(&ci->i_truncate_mutex); |
355da1eb | 1829 | retry: |
be655596 | 1830 | spin_lock(&ci->i_ceph_lock); |
355da1eb SW |
1831 | if (ci->i_truncate_pending == 0) { |
1832 | dout("__do_pending_vmtruncate %p none pending\n", inode); | |
be655596 | 1833 | spin_unlock(&ci->i_ceph_lock); |
b0d7c223 | 1834 | mutex_unlock(&ci->i_truncate_mutex); |
355da1eb SW |
1835 | return; |
1836 | } | |
1837 | ||
1838 | /* | |
1839 | * make sure any dirty snapped pages are flushed before we | |
1840 | * possibly truncate them.. so write AND block! | |
1841 | */ | |
1842 | if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) { | |
c8fd0d37 YZ |
1843 | struct ceph_cap_snap *capsnap; |
1844 | to = ci->i_truncate_size; | |
1845 | list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) { | |
1846 | // MDS should have revoked Frw caps | |
1847 | WARN_ON_ONCE(capsnap->writing); | |
1848 | if (capsnap->dirty_pages && capsnap->size > to) | |
1849 | to = capsnap->size; | |
1850 | } | |
1851 | spin_unlock(&ci->i_ceph_lock); | |
355da1eb SW |
1852 | dout("__do_pending_vmtruncate %p flushing snaps first\n", |
1853 | inode); | |
c8fd0d37 YZ |
1854 | |
1855 | truncate_pagecache(inode, to); | |
1856 | ||
355da1eb SW |
1857 | filemap_write_and_wait_range(&inode->i_data, 0, |
1858 | inode->i_sb->s_maxbytes); | |
1859 | goto retry; | |
1860 | } | |
1861 | ||
b0d7c223 YZ |
1862 | /* there should be no reader or writer */ |
1863 | WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref); | |
1864 | ||
355da1eb SW |
1865 | to = ci->i_truncate_size; |
1866 | wrbuffer_refs = ci->i_wrbuffer_ref; | |
1867 | dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode, | |
1868 | ci->i_truncate_pending, to); | |
be655596 | 1869 | spin_unlock(&ci->i_ceph_lock); |
355da1eb | 1870 | |
4e217b5d | 1871 | truncate_pagecache(inode, to); |
355da1eb | 1872 | |
be655596 | 1873 | spin_lock(&ci->i_ceph_lock); |
a85f50b6 YZ |
1874 | if (to == ci->i_truncate_size) { |
1875 | ci->i_truncate_pending = 0; | |
1876 | finish = 1; | |
1877 | } | |
be655596 | 1878 | spin_unlock(&ci->i_ceph_lock); |
a85f50b6 YZ |
1879 | if (!finish) |
1880 | goto retry; | |
355da1eb | 1881 | |
b0d7c223 YZ |
1882 | mutex_unlock(&ci->i_truncate_mutex); |
1883 | ||
355da1eb SW |
1884 | if (wrbuffer_refs == 0) |
1885 | ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL); | |
a85f50b6 YZ |
1886 | |
1887 | wake_up_all(&ci->i_cap_wq); | |
355da1eb SW |
1888 | } |
1889 | ||
d91016e8 YZ |
1890 | static void ceph_inode_work(struct work_struct *work) |
1891 | { | |
1892 | struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info, | |
1893 | i_work); | |
1894 | struct inode *inode = &ci->vfs_inode; | |
1895 | ||
1896 | if (test_and_clear_bit(CEPH_I_WORK_WRITEBACK, &ci->i_work_mask)) { | |
1897 | dout("writeback %p\n", inode); | |
1898 | filemap_fdatawrite(&inode->i_data); | |
1899 | } | |
1900 | if (test_and_clear_bit(CEPH_I_WORK_INVALIDATE_PAGES, &ci->i_work_mask)) | |
1901 | ceph_do_invalidate_pages(inode); | |
1902 | ||
1903 | if (test_and_clear_bit(CEPH_I_WORK_VMTRUNCATE, &ci->i_work_mask)) | |
1904 | __ceph_do_pending_vmtruncate(inode); | |
1905 | ||
1906 | iput(inode); | |
1907 | } | |
1908 | ||
355da1eb SW |
1909 | /* |
1910 | * symlinks | |
1911 | */ | |
355da1eb | 1912 | static const struct inode_operations ceph_symlink_iops = { |
6b255391 | 1913 | .get_link = simple_get_link, |
0b932672 YZ |
1914 | .setattr = ceph_setattr, |
1915 | .getattr = ceph_getattr, | |
0b932672 | 1916 | .listxattr = ceph_listxattr, |
355da1eb SW |
1917 | }; |
1918 | ||
a26fecca | 1919 | int __ceph_setattr(struct inode *inode, struct iattr *attr) |
355da1eb | 1920 | { |
355da1eb | 1921 | struct ceph_inode_info *ci = ceph_inode(inode); |
355da1eb SW |
1922 | const unsigned int ia_valid = attr->ia_valid; |
1923 | struct ceph_mds_request *req; | |
a26fecca | 1924 | struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc; |
f66fd9f0 | 1925 | struct ceph_cap_flush *prealloc_cf; |
355da1eb SW |
1926 | int issued; |
1927 | int release = 0, dirtied = 0; | |
1928 | int mask = 0; | |
1929 | int err = 0; | |
fca65b4a | 1930 | int inode_dirty_flags = 0; |
604d1b02 | 1931 | bool lock_snap_rwsem = false; |
355da1eb | 1932 | |
f66fd9f0 YZ |
1933 | prealloc_cf = ceph_alloc_cap_flush(); |
1934 | if (!prealloc_cf) | |
1935 | return -ENOMEM; | |
1936 | ||
355da1eb SW |
1937 | req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR, |
1938 | USE_AUTH_MDS); | |
f66fd9f0 YZ |
1939 | if (IS_ERR(req)) { |
1940 | ceph_free_cap_flush(prealloc_cf); | |
355da1eb | 1941 | return PTR_ERR(req); |
f66fd9f0 | 1942 | } |
355da1eb | 1943 | |
be655596 | 1944 | spin_lock(&ci->i_ceph_lock); |
355da1eb | 1945 | issued = __ceph_caps_issued(ci, NULL); |
604d1b02 YZ |
1946 | |
1947 | if (!ci->i_head_snapc && | |
1948 | (issued & (CEPH_CAP_ANY_EXCL | CEPH_CAP_FILE_WR))) { | |
1949 | lock_snap_rwsem = true; | |
1950 | if (!down_read_trylock(&mdsc->snap_rwsem)) { | |
1951 | spin_unlock(&ci->i_ceph_lock); | |
1952 | down_read(&mdsc->snap_rwsem); | |
1953 | spin_lock(&ci->i_ceph_lock); | |
1954 | issued = __ceph_caps_issued(ci, NULL); | |
1955 | } | |
1956 | } | |
1957 | ||
355da1eb SW |
1958 | dout("setattr %p issued %s\n", inode, ceph_cap_string(issued)); |
1959 | ||
1960 | if (ia_valid & ATTR_UID) { | |
1961 | dout("setattr %p uid %d -> %d\n", inode, | |
bd2bae6a EB |
1962 | from_kuid(&init_user_ns, inode->i_uid), |
1963 | from_kuid(&init_user_ns, attr->ia_uid)); | |
355da1eb SW |
1964 | if (issued & CEPH_CAP_AUTH_EXCL) { |
1965 | inode->i_uid = attr->ia_uid; | |
1966 | dirtied |= CEPH_CAP_AUTH_EXCL; | |
1967 | } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || | |
ab871b90 EB |
1968 | !uid_eq(attr->ia_uid, inode->i_uid)) { |
1969 | req->r_args.setattr.uid = cpu_to_le32( | |
1970 | from_kuid(&init_user_ns, attr->ia_uid)); | |
355da1eb SW |
1971 | mask |= CEPH_SETATTR_UID; |
1972 | release |= CEPH_CAP_AUTH_SHARED; | |
1973 | } | |
1974 | } | |
1975 | if (ia_valid & ATTR_GID) { | |
1976 | dout("setattr %p gid %d -> %d\n", inode, | |
bd2bae6a EB |
1977 | from_kgid(&init_user_ns, inode->i_gid), |
1978 | from_kgid(&init_user_ns, attr->ia_gid)); | |
355da1eb SW |
1979 | if (issued & CEPH_CAP_AUTH_EXCL) { |
1980 | inode->i_gid = attr->ia_gid; | |
1981 | dirtied |= CEPH_CAP_AUTH_EXCL; | |
1982 | } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || | |
ab871b90 EB |
1983 | !gid_eq(attr->ia_gid, inode->i_gid)) { |
1984 | req->r_args.setattr.gid = cpu_to_le32( | |
1985 | from_kgid(&init_user_ns, attr->ia_gid)); | |
355da1eb SW |
1986 | mask |= CEPH_SETATTR_GID; |
1987 | release |= CEPH_CAP_AUTH_SHARED; | |
1988 | } | |
1989 | } | |
1990 | if (ia_valid & ATTR_MODE) { | |
1991 | dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode, | |
1992 | attr->ia_mode); | |
1993 | if (issued & CEPH_CAP_AUTH_EXCL) { | |
1994 | inode->i_mode = attr->ia_mode; | |
1995 | dirtied |= CEPH_CAP_AUTH_EXCL; | |
1996 | } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || | |
1997 | attr->ia_mode != inode->i_mode) { | |
7221fe4c | 1998 | inode->i_mode = attr->ia_mode; |
355da1eb SW |
1999 | req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode); |
2000 | mask |= CEPH_SETATTR_MODE; | |
2001 | release |= CEPH_CAP_AUTH_SHARED; | |
2002 | } | |
2003 | } | |
2004 | ||
2005 | if (ia_valid & ATTR_ATIME) { | |
2006 | dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode, | |
2007 | inode->i_atime.tv_sec, inode->i_atime.tv_nsec, | |
2008 | attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec); | |
2009 | if (issued & CEPH_CAP_FILE_EXCL) { | |
2010 | ci->i_time_warp_seq++; | |
2011 | inode->i_atime = attr->ia_atime; | |
2012 | dirtied |= CEPH_CAP_FILE_EXCL; | |
2013 | } else if ((issued & CEPH_CAP_FILE_WR) && | |
2014 | timespec_compare(&inode->i_atime, | |
2015 | &attr->ia_atime) < 0) { | |
2016 | inode->i_atime = attr->ia_atime; | |
2017 | dirtied |= CEPH_CAP_FILE_WR; | |
2018 | } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || | |
2019 | !timespec_equal(&inode->i_atime, &attr->ia_atime)) { | |
2020 | ceph_encode_timespec(&req->r_args.setattr.atime, | |
2021 | &attr->ia_atime); | |
2022 | mask |= CEPH_SETATTR_ATIME; | |
2023 | release |= CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_RD | | |
2024 | CEPH_CAP_FILE_WR; | |
2025 | } | |
2026 | } | |
2027 | if (ia_valid & ATTR_MTIME) { | |
2028 | dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode, | |
2029 | inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec, | |
2030 | attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec); | |
2031 | if (issued & CEPH_CAP_FILE_EXCL) { | |
2032 | ci->i_time_warp_seq++; | |
2033 | inode->i_mtime = attr->ia_mtime; | |
2034 | dirtied |= CEPH_CAP_FILE_EXCL; | |
2035 | } else if ((issued & CEPH_CAP_FILE_WR) && | |
2036 | timespec_compare(&inode->i_mtime, | |
2037 | &attr->ia_mtime) < 0) { | |
2038 | inode->i_mtime = attr->ia_mtime; | |
2039 | dirtied |= CEPH_CAP_FILE_WR; | |
2040 | } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || | |
2041 | !timespec_equal(&inode->i_mtime, &attr->ia_mtime)) { | |
2042 | ceph_encode_timespec(&req->r_args.setattr.mtime, | |
2043 | &attr->ia_mtime); | |
2044 | mask |= CEPH_SETATTR_MTIME; | |
2045 | release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD | | |
2046 | CEPH_CAP_FILE_WR; | |
2047 | } | |
2048 | } | |
2049 | if (ia_valid & ATTR_SIZE) { | |
2050 | dout("setattr %p size %lld -> %lld\n", inode, | |
2051 | inode->i_size, attr->ia_size); | |
355da1eb SW |
2052 | if ((issued & CEPH_CAP_FILE_EXCL) && |
2053 | attr->ia_size > inode->i_size) { | |
99c88e69 | 2054 | i_size_write(inode, attr->ia_size); |
224a7542 | 2055 | inode->i_blocks = calc_inode_blocks(attr->ia_size); |
355da1eb SW |
2056 | ci->i_reported_size = attr->ia_size; |
2057 | dirtied |= CEPH_CAP_FILE_EXCL; | |
2058 | } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || | |
2059 | attr->ia_size != inode->i_size) { | |
2060 | req->r_args.setattr.size = cpu_to_le64(attr->ia_size); | |
2061 | req->r_args.setattr.old_size = | |
2062 | cpu_to_le64(inode->i_size); | |
2063 | mask |= CEPH_SETATTR_SIZE; | |
2064 | release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD | | |
2065 | CEPH_CAP_FILE_WR; | |
2066 | } | |
2067 | } | |
2068 | ||
2069 | /* these do nothing */ | |
2070 | if (ia_valid & ATTR_CTIME) { | |
2071 | bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME| | |
2072 | ATTR_MODE|ATTR_UID|ATTR_GID)) == 0; | |
2073 | dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode, | |
2074 | inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec, | |
2075 | attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec, | |
2076 | only ? "ctime only" : "ignored"); | |
355da1eb SW |
2077 | if (only) { |
2078 | /* | |
2079 | * if kernel wants to dirty ctime but nothing else, | |
2080 | * we need to choose a cap to dirty under, or do | |
2081 | * a almost-no-op setattr | |
2082 | */ | |
2083 | if (issued & CEPH_CAP_AUTH_EXCL) | |
2084 | dirtied |= CEPH_CAP_AUTH_EXCL; | |
2085 | else if (issued & CEPH_CAP_FILE_EXCL) | |
2086 | dirtied |= CEPH_CAP_FILE_EXCL; | |
2087 | else if (issued & CEPH_CAP_XATTR_EXCL) | |
2088 | dirtied |= CEPH_CAP_XATTR_EXCL; | |
2089 | else | |
2090 | mask |= CEPH_SETATTR_CTIME; | |
2091 | } | |
2092 | } | |
2093 | if (ia_valid & ATTR_FILE) | |
2094 | dout("setattr %p ATTR_FILE ... hrm!\n", inode); | |
2095 | ||
2096 | if (dirtied) { | |
f66fd9f0 YZ |
2097 | inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied, |
2098 | &prealloc_cf); | |
4ca2fea6 | 2099 | inode->i_ctime = attr->ia_ctime; |
355da1eb SW |
2100 | } |
2101 | ||
2102 | release &= issued; | |
be655596 | 2103 | spin_unlock(&ci->i_ceph_lock); |
604d1b02 YZ |
2104 | if (lock_snap_rwsem) |
2105 | up_read(&mdsc->snap_rwsem); | |
355da1eb | 2106 | |
fca65b4a SW |
2107 | if (inode_dirty_flags) |
2108 | __mark_inode_dirty(inode, inode_dirty_flags); | |
2109 | ||
7221fe4c | 2110 | |
355da1eb | 2111 | if (mask) { |
70b666c3 SW |
2112 | req->r_inode = inode; |
2113 | ihold(inode); | |
355da1eb SW |
2114 | req->r_inode_drop = release; |
2115 | req->r_args.setattr.mask = cpu_to_le32(mask); | |
2116 | req->r_num_caps = 1; | |
4ca2fea6 | 2117 | req->r_stamp = attr->ia_ctime; |
752c8bdc | 2118 | err = ceph_mdsc_do_request(mdsc, NULL, req); |
355da1eb SW |
2119 | } |
2120 | dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err, | |
2121 | ceph_cap_string(dirtied), mask); | |
2122 | ||
355da1eb | 2123 | ceph_mdsc_put_request(req); |
f66fd9f0 | 2124 | ceph_free_cap_flush(prealloc_cf); |
8179a101 YZ |
2125 | |
2126 | if (err >= 0 && (mask & CEPH_SETATTR_SIZE)) | |
2127 | __ceph_do_pending_vmtruncate(inode); | |
2128 | ||
355da1eb SW |
2129 | return err; |
2130 | } | |
2131 | ||
a26fecca AG |
2132 | /* |
2133 | * setattr | |
2134 | */ | |
2135 | int ceph_setattr(struct dentry *dentry, struct iattr *attr) | |
2136 | { | |
fd5472ed JK |
2137 | struct inode *inode = d_inode(dentry); |
2138 | int err; | |
2139 | ||
2140 | if (ceph_snap(inode) != CEPH_NOSNAP) | |
2141 | return -EROFS; | |
2142 | ||
31051c85 | 2143 | err = setattr_prepare(dentry, attr); |
fd5472ed JK |
2144 | if (err != 0) |
2145 | return err; | |
2146 | ||
8179a101 YZ |
2147 | err = __ceph_setattr(inode, attr); |
2148 | ||
2149 | if (err >= 0 && (attr->ia_valid & ATTR_MODE)) | |
2150 | err = posix_acl_chmod(inode, attr->ia_mode); | |
2151 | ||
2152 | return err; | |
a26fecca AG |
2153 | } |
2154 | ||
355da1eb SW |
2155 | /* |
2156 | * Verify that we have a lease on the given mask. If not, | |
2157 | * do a getattr against an mds. | |
2158 | */ | |
01deead0 YZ |
2159 | int __ceph_do_getattr(struct inode *inode, struct page *locked_page, |
2160 | int mask, bool force) | |
355da1eb | 2161 | { |
3d14c5d2 YS |
2162 | struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb); |
2163 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
355da1eb SW |
2164 | struct ceph_mds_request *req; |
2165 | int err; | |
2166 | ||
2167 | if (ceph_snap(inode) == CEPH_SNAPDIR) { | |
2168 | dout("do_getattr inode %p SNAPDIR\n", inode); | |
2169 | return 0; | |
2170 | } | |
2171 | ||
01deead0 YZ |
2172 | dout("do_getattr inode %p mask %s mode 0%o\n", |
2173 | inode, ceph_cap_string(mask), inode->i_mode); | |
508b32d8 | 2174 | if (!force && ceph_caps_issued_mask(ceph_inode(inode), mask, 1)) |
355da1eb SW |
2175 | return 0; |
2176 | ||
2177 | req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS); | |
2178 | if (IS_ERR(req)) | |
2179 | return PTR_ERR(req); | |
70b666c3 SW |
2180 | req->r_inode = inode; |
2181 | ihold(inode); | |
355da1eb SW |
2182 | req->r_num_caps = 1; |
2183 | req->r_args.getattr.mask = cpu_to_le32(mask); | |
01deead0 | 2184 | req->r_locked_page = locked_page; |
355da1eb | 2185 | err = ceph_mdsc_do_request(mdsc, NULL, req); |
01deead0 YZ |
2186 | if (locked_page && err == 0) { |
2187 | u64 inline_version = req->r_reply_info.targeti.inline_version; | |
2188 | if (inline_version == 0) { | |
2189 | /* the reply is supposed to contain inline data */ | |
2190 | err = -EINVAL; | |
2191 | } else if (inline_version == CEPH_INLINE_NONE) { | |
2192 | err = -ENODATA; | |
2193 | } else { | |
2194 | err = req->r_reply_info.targeti.inline_len; | |
2195 | } | |
2196 | } | |
355da1eb SW |
2197 | ceph_mdsc_put_request(req); |
2198 | dout("do_getattr result=%d\n", err); | |
2199 | return err; | |
2200 | } | |
2201 | ||
2202 | ||
2203 | /* | |
2204 | * Check inode permissions. We verify we have a valid value for | |
2205 | * the AUTH cap, then call the generic handler. | |
2206 | */ | |
10556cb2 | 2207 | int ceph_permission(struct inode *inode, int mask) |
355da1eb | 2208 | { |
b74c79e9 NP |
2209 | int err; |
2210 | ||
10556cb2 | 2211 | if (mask & MAY_NOT_BLOCK) |
b74c79e9 NP |
2212 | return -ECHILD; |
2213 | ||
508b32d8 | 2214 | err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED, false); |
355da1eb SW |
2215 | |
2216 | if (!err) | |
2830ba7f | 2217 | err = generic_permission(inode, mask); |
355da1eb SW |
2218 | return err; |
2219 | } | |
2220 | ||
2221 | /* | |
2222 | * Get all attributes. Hopefully somedata we'll have a statlite() | |
2223 | * and can limit the fields we require to be accurate. | |
2224 | */ | |
a528d35e DH |
2225 | int ceph_getattr(const struct path *path, struct kstat *stat, |
2226 | u32 request_mask, unsigned int flags) | |
355da1eb | 2227 | { |
a528d35e | 2228 | struct inode *inode = d_inode(path->dentry); |
232d4b01 | 2229 | struct ceph_inode_info *ci = ceph_inode(inode); |
355da1eb SW |
2230 | int err; |
2231 | ||
508b32d8 | 2232 | err = ceph_do_getattr(inode, CEPH_STAT_CAP_INODE_ALL, false); |
355da1eb SW |
2233 | if (!err) { |
2234 | generic_fillattr(inode, stat); | |
ad1fee96 | 2235 | stat->ino = ceph_translate_ino(inode->i_sb, inode->i_ino); |
355da1eb SW |
2236 | if (ceph_snap(inode) != CEPH_NOSNAP) |
2237 | stat->dev = ceph_snap(inode); | |
2238 | else | |
2239 | stat->dev = 0; | |
232d4b01 | 2240 | if (S_ISDIR(inode->i_mode)) { |
1c1266bb YS |
2241 | if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb), |
2242 | RBYTES)) | |
2243 | stat->size = ci->i_rbytes; | |
2244 | else | |
2245 | stat->size = ci->i_files + ci->i_subdirs; | |
232d4b01 | 2246 | stat->blocks = 0; |
355da1eb | 2247 | stat->blksize = 65536; |
232d4b01 | 2248 | } |
355da1eb SW |
2249 | } |
2250 | return err; | |
2251 | } |