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> |
a35ead31 | 16 | #include <linux/iversion.h> |
355da1eb SW |
17 | |
18 | #include "super.h" | |
3d14c5d2 | 19 | #include "mds_client.h" |
99ccbd22 | 20 | #include "cache.h" |
3d14c5d2 | 21 | #include <linux/ceph/decode.h> |
355da1eb SW |
22 | |
23 | /* | |
24 | * Ceph inode operations | |
25 | * | |
26 | * Implement basic inode helpers (get, alloc) and inode ops (getattr, | |
27 | * setattr, etc.), xattr helpers, and helpers for assimilating | |
28 | * metadata returned by the MDS into our cache. | |
29 | * | |
30 | * Also define helpers for doing asynchronous writeback, invalidation, | |
31 | * and truncation for the benefit of those who can't afford to block | |
32 | * (typically because they are in the message handler path). | |
33 | */ | |
34 | ||
35 | static const struct inode_operations ceph_symlink_iops; | |
36 | ||
1cf89a8d | 37 | static void ceph_inode_work(struct work_struct *work); |
355da1eb SW |
38 | |
39 | /* | |
40 | * find or create an inode, given the ceph ino number | |
41 | */ | |
ad1fee96 YS |
42 | static int ceph_set_ino_cb(struct inode *inode, void *data) |
43 | { | |
44 | ceph_inode(inode)->i_vino = *(struct ceph_vino *)data; | |
45 | inode->i_ino = ceph_vino_to_ino(*(struct ceph_vino *)data); | |
a35ead31 | 46 | inode_set_iversion_raw(inode, 0); |
ad1fee96 YS |
47 | return 0; |
48 | } | |
49 | ||
355da1eb SW |
50 | struct inode *ceph_get_inode(struct super_block *sb, struct ceph_vino vino) |
51 | { | |
52 | struct inode *inode; | |
53 | ino_t t = ceph_vino_to_ino(vino); | |
54 | ||
55 | inode = iget5_locked(sb, t, ceph_ino_compare, ceph_set_ino_cb, &vino); | |
d37b1d99 | 56 | if (!inode) |
355da1eb SW |
57 | return ERR_PTR(-ENOMEM); |
58 | if (inode->i_state & I_NEW) { | |
59 | dout("get_inode created new inode %p %llx.%llx ino %llx\n", | |
60 | inode, ceph_vinop(inode), (u64)inode->i_ino); | |
61 | unlock_new_inode(inode); | |
62 | } | |
63 | ||
64 | dout("get_inode on %lu=%llx.%llx got %p\n", inode->i_ino, vino.ino, | |
65 | vino.snap, inode); | |
66 | return inode; | |
67 | } | |
68 | ||
69 | /* | |
70 | * get/constuct snapdir inode for a given directory | |
71 | */ | |
72 | struct inode *ceph_get_snapdir(struct inode *parent) | |
73 | { | |
74 | struct ceph_vino vino = { | |
75 | .ino = ceph_ino(parent), | |
76 | .snap = CEPH_SNAPDIR, | |
77 | }; | |
78 | struct inode *inode = ceph_get_inode(parent->i_sb, vino); | |
b377ff13 | 79 | struct ceph_inode_info *ci = ceph_inode(inode); |
355da1eb SW |
80 | |
81 | BUG_ON(!S_ISDIR(parent->i_mode)); | |
82 | if (IS_ERR(inode)) | |
7e34bc52 | 83 | return inode; |
355da1eb SW |
84 | inode->i_mode = parent->i_mode; |
85 | inode->i_uid = parent->i_uid; | |
86 | inode->i_gid = parent->i_gid; | |
38c48b5f YZ |
87 | inode->i_op = &ceph_snapdir_iops; |
88 | inode->i_fop = &ceph_snapdir_fops; | |
b377ff13 SW |
89 | ci->i_snap_caps = CEPH_CAP_PIN; /* so we can open */ |
90 | ci->i_rbytes = 0; | |
355da1eb SW |
91 | return inode; |
92 | } | |
93 | ||
94 | const struct inode_operations ceph_file_iops = { | |
95 | .permission = ceph_permission, | |
96 | .setattr = ceph_setattr, | |
97 | .getattr = ceph_getattr, | |
355da1eb | 98 | .listxattr = ceph_listxattr, |
7221fe4c | 99 | .get_acl = ceph_get_acl, |
72466d0b | 100 | .set_acl = ceph_set_acl, |
355da1eb SW |
101 | }; |
102 | ||
103 | ||
104 | /* | |
105 | * We use a 'frag tree' to keep track of the MDS's directory fragments | |
106 | * for a given inode (usually there is just a single fragment). We | |
107 | * need to know when a child frag is delegated to a new MDS, or when | |
108 | * it is flagged as replicated, so we can direct our requests | |
109 | * accordingly. | |
110 | */ | |
111 | ||
112 | /* | |
113 | * find/create a frag in the tree | |
114 | */ | |
115 | static struct ceph_inode_frag *__get_or_create_frag(struct ceph_inode_info *ci, | |
116 | u32 f) | |
117 | { | |
118 | struct rb_node **p; | |
119 | struct rb_node *parent = NULL; | |
120 | struct ceph_inode_frag *frag; | |
121 | int c; | |
122 | ||
123 | p = &ci->i_fragtree.rb_node; | |
124 | while (*p) { | |
125 | parent = *p; | |
126 | frag = rb_entry(parent, struct ceph_inode_frag, node); | |
127 | c = ceph_frag_compare(f, frag->frag); | |
128 | if (c < 0) | |
129 | p = &(*p)->rb_left; | |
130 | else if (c > 0) | |
131 | p = &(*p)->rb_right; | |
132 | else | |
133 | return frag; | |
134 | } | |
135 | ||
136 | frag = kmalloc(sizeof(*frag), GFP_NOFS); | |
51308806 | 137 | if (!frag) |
355da1eb | 138 | return ERR_PTR(-ENOMEM); |
51308806 | 139 | |
355da1eb SW |
140 | frag->frag = f; |
141 | frag->split_by = 0; | |
142 | frag->mds = -1; | |
143 | frag->ndist = 0; | |
144 | ||
145 | rb_link_node(&frag->node, parent, p); | |
146 | rb_insert_color(&frag->node, &ci->i_fragtree); | |
147 | ||
148 | dout("get_or_create_frag added %llx.%llx frag %x\n", | |
149 | ceph_vinop(&ci->vfs_inode), f); | |
150 | return frag; | |
151 | } | |
152 | ||
153 | /* | |
154 | * find a specific frag @f | |
155 | */ | |
156 | struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci, u32 f) | |
157 | { | |
158 | struct rb_node *n = ci->i_fragtree.rb_node; | |
159 | ||
160 | while (n) { | |
161 | struct ceph_inode_frag *frag = | |
162 | rb_entry(n, struct ceph_inode_frag, node); | |
163 | int c = ceph_frag_compare(f, frag->frag); | |
164 | if (c < 0) | |
165 | n = n->rb_left; | |
166 | else if (c > 0) | |
167 | n = n->rb_right; | |
168 | else | |
169 | return frag; | |
170 | } | |
171 | return NULL; | |
172 | } | |
173 | ||
174 | /* | |
175 | * Choose frag containing the given value @v. If @pfrag is | |
176 | * specified, copy the frag delegation info to the caller if | |
177 | * it is present. | |
178 | */ | |
3e7fbe9c YZ |
179 | static u32 __ceph_choose_frag(struct ceph_inode_info *ci, u32 v, |
180 | struct ceph_inode_frag *pfrag, int *found) | |
355da1eb SW |
181 | { |
182 | u32 t = ceph_frag_make(0, 0); | |
183 | struct ceph_inode_frag *frag; | |
184 | unsigned nway, i; | |
185 | u32 n; | |
186 | ||
187 | if (found) | |
188 | *found = 0; | |
189 | ||
355da1eb SW |
190 | while (1) { |
191 | WARN_ON(!ceph_frag_contains_value(t, v)); | |
192 | frag = __ceph_find_frag(ci, t); | |
193 | if (!frag) | |
194 | break; /* t is a leaf */ | |
195 | if (frag->split_by == 0) { | |
196 | if (pfrag) | |
197 | memcpy(pfrag, frag, sizeof(*pfrag)); | |
198 | if (found) | |
199 | *found = 1; | |
200 | break; | |
201 | } | |
202 | ||
203 | /* choose child */ | |
204 | nway = 1 << frag->split_by; | |
205 | dout("choose_frag(%x) %x splits by %d (%d ways)\n", v, t, | |
206 | frag->split_by, nway); | |
207 | for (i = 0; i < nway; i++) { | |
208 | n = ceph_frag_make_child(t, frag->split_by, i); | |
209 | if (ceph_frag_contains_value(n, v)) { | |
210 | t = n; | |
211 | break; | |
212 | } | |
213 | } | |
214 | BUG_ON(i == nway); | |
215 | } | |
216 | dout("choose_frag(%x) = %x\n", v, t); | |
217 | ||
355da1eb SW |
218 | return t; |
219 | } | |
220 | ||
3e7fbe9c YZ |
221 | u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v, |
222 | struct ceph_inode_frag *pfrag, int *found) | |
223 | { | |
224 | u32 ret; | |
225 | mutex_lock(&ci->i_fragtree_mutex); | |
226 | ret = __ceph_choose_frag(ci, v, pfrag, found); | |
227 | mutex_unlock(&ci->i_fragtree_mutex); | |
228 | return ret; | |
229 | } | |
230 | ||
355da1eb SW |
231 | /* |
232 | * Process dirfrag (delegation) info from the mds. Include leaf | |
233 | * fragment in tree ONLY if ndist > 0. Otherwise, only | |
234 | * branches/splits are included in i_fragtree) | |
235 | */ | |
236 | static int ceph_fill_dirfrag(struct inode *inode, | |
237 | struct ceph_mds_reply_dirfrag *dirinfo) | |
238 | { | |
239 | struct ceph_inode_info *ci = ceph_inode(inode); | |
240 | struct ceph_inode_frag *frag; | |
241 | u32 id = le32_to_cpu(dirinfo->frag); | |
242 | int mds = le32_to_cpu(dirinfo->auth); | |
243 | int ndist = le32_to_cpu(dirinfo->ndist); | |
8d08503c | 244 | int diri_auth = -1; |
355da1eb SW |
245 | int i; |
246 | int err = 0; | |
247 | ||
8d08503c YZ |
248 | spin_lock(&ci->i_ceph_lock); |
249 | if (ci->i_auth_cap) | |
250 | diri_auth = ci->i_auth_cap->mds; | |
251 | spin_unlock(&ci->i_ceph_lock); | |
252 | ||
42172119 YZ |
253 | if (mds == -1) /* CDIR_AUTH_PARENT */ |
254 | mds = diri_auth; | |
255 | ||
355da1eb | 256 | mutex_lock(&ci->i_fragtree_mutex); |
8d08503c | 257 | if (ndist == 0 && mds == diri_auth) { |
355da1eb SW |
258 | /* no delegation info needed. */ |
259 | frag = __ceph_find_frag(ci, id); | |
260 | if (!frag) | |
261 | goto out; | |
262 | if (frag->split_by == 0) { | |
263 | /* tree leaf, remove */ | |
264 | dout("fill_dirfrag removed %llx.%llx frag %x" | |
265 | " (no ref)\n", ceph_vinop(inode), id); | |
266 | rb_erase(&frag->node, &ci->i_fragtree); | |
267 | kfree(frag); | |
268 | } else { | |
269 | /* tree branch, keep and clear */ | |
270 | dout("fill_dirfrag cleared %llx.%llx frag %x" | |
271 | " referral\n", ceph_vinop(inode), id); | |
272 | frag->mds = -1; | |
273 | frag->ndist = 0; | |
274 | } | |
275 | goto out; | |
276 | } | |
277 | ||
278 | ||
279 | /* find/add this frag to store mds delegation info */ | |
280 | frag = __get_or_create_frag(ci, id); | |
281 | if (IS_ERR(frag)) { | |
282 | /* this is not the end of the world; we can continue | |
283 | with bad/inaccurate delegation info */ | |
284 | pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n", | |
285 | ceph_vinop(inode), le32_to_cpu(dirinfo->frag)); | |
286 | err = -ENOMEM; | |
287 | goto out; | |
288 | } | |
289 | ||
290 | frag->mds = mds; | |
291 | frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP); | |
292 | for (i = 0; i < frag->ndist; i++) | |
293 | frag->dist[i] = le32_to_cpu(dirinfo->dist[i]); | |
294 | dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n", | |
295 | ceph_vinop(inode), frag->frag, frag->ndist); | |
296 | ||
297 | out: | |
298 | mutex_unlock(&ci->i_fragtree_mutex); | |
299 | return err; | |
300 | } | |
301 | ||
a407846e YZ |
302 | static int frag_tree_split_cmp(const void *l, const void *r) |
303 | { | |
304 | struct ceph_frag_tree_split *ls = (struct ceph_frag_tree_split*)l; | |
305 | struct ceph_frag_tree_split *rs = (struct ceph_frag_tree_split*)r; | |
fe2ed425 JL |
306 | return ceph_frag_compare(le32_to_cpu(ls->frag), |
307 | le32_to_cpu(rs->frag)); | |
a407846e YZ |
308 | } |
309 | ||
a4b7431f YZ |
310 | static bool is_frag_child(u32 f, struct ceph_inode_frag *frag) |
311 | { | |
312 | if (!frag) | |
313 | return f == ceph_frag_make(0, 0); | |
314 | if (ceph_frag_bits(f) != ceph_frag_bits(frag->frag) + frag->split_by) | |
315 | return false; | |
316 | return ceph_frag_contains_value(frag->frag, ceph_frag_value(f)); | |
317 | } | |
318 | ||
3e7fbe9c YZ |
319 | static int ceph_fill_fragtree(struct inode *inode, |
320 | struct ceph_frag_tree_head *fragtree, | |
321 | struct ceph_mds_reply_dirfrag *dirinfo) | |
322 | { | |
323 | struct ceph_inode_info *ci = ceph_inode(inode); | |
a4b7431f | 324 | struct ceph_inode_frag *frag, *prev_frag = NULL; |
3e7fbe9c | 325 | struct rb_node *rb_node; |
1b1bc16d YZ |
326 | unsigned i, split_by, nsplits; |
327 | u32 id; | |
3e7fbe9c YZ |
328 | bool update = false; |
329 | ||
330 | mutex_lock(&ci->i_fragtree_mutex); | |
331 | nsplits = le32_to_cpu(fragtree->nsplits); | |
1b1bc16d YZ |
332 | if (nsplits != ci->i_fragtree_nsplits) { |
333 | update = true; | |
334 | } else if (nsplits) { | |
3e7fbe9c YZ |
335 | i = prandom_u32() % nsplits; |
336 | id = le32_to_cpu(fragtree->splits[i].frag); | |
337 | if (!__ceph_find_frag(ci, id)) | |
338 | update = true; | |
339 | } else if (!RB_EMPTY_ROOT(&ci->i_fragtree)) { | |
340 | rb_node = rb_first(&ci->i_fragtree); | |
341 | frag = rb_entry(rb_node, struct ceph_inode_frag, node); | |
342 | if (frag->frag != ceph_frag_make(0, 0) || rb_next(rb_node)) | |
343 | update = true; | |
344 | } | |
345 | if (!update && dirinfo) { | |
346 | id = le32_to_cpu(dirinfo->frag); | |
347 | if (id != __ceph_choose_frag(ci, id, NULL, NULL)) | |
348 | update = true; | |
349 | } | |
350 | if (!update) | |
351 | goto out_unlock; | |
352 | ||
a407846e YZ |
353 | if (nsplits > 1) { |
354 | sort(fragtree->splits, nsplits, sizeof(fragtree->splits[0]), | |
355 | frag_tree_split_cmp, NULL); | |
356 | } | |
357 | ||
3e7fbe9c YZ |
358 | dout("fill_fragtree %llx.%llx\n", ceph_vinop(inode)); |
359 | rb_node = rb_first(&ci->i_fragtree); | |
360 | for (i = 0; i < nsplits; i++) { | |
361 | id = le32_to_cpu(fragtree->splits[i].frag); | |
1b1bc16d YZ |
362 | split_by = le32_to_cpu(fragtree->splits[i].by); |
363 | if (split_by == 0 || ceph_frag_bits(id) + split_by > 24) { | |
364 | pr_err("fill_fragtree %llx.%llx invalid split %d/%u, " | |
365 | "frag %x split by %d\n", ceph_vinop(inode), | |
366 | i, nsplits, id, split_by); | |
367 | continue; | |
368 | } | |
3e7fbe9c YZ |
369 | frag = NULL; |
370 | while (rb_node) { | |
371 | frag = rb_entry(rb_node, struct ceph_inode_frag, node); | |
372 | if (ceph_frag_compare(frag->frag, id) >= 0) { | |
373 | if (frag->frag != id) | |
374 | frag = NULL; | |
375 | else | |
376 | rb_node = rb_next(rb_node); | |
377 | break; | |
378 | } | |
379 | rb_node = rb_next(rb_node); | |
a4b7431f YZ |
380 | /* delete stale split/leaf node */ |
381 | if (frag->split_by > 0 || | |
382 | !is_frag_child(frag->frag, prev_frag)) { | |
383 | rb_erase(&frag->node, &ci->i_fragtree); | |
1b1bc16d YZ |
384 | if (frag->split_by > 0) |
385 | ci->i_fragtree_nsplits--; | |
a4b7431f YZ |
386 | kfree(frag); |
387 | } | |
3e7fbe9c YZ |
388 | frag = NULL; |
389 | } | |
390 | if (!frag) { | |
391 | frag = __get_or_create_frag(ci, id); | |
392 | if (IS_ERR(frag)) | |
393 | continue; | |
394 | } | |
1b1bc16d YZ |
395 | if (frag->split_by == 0) |
396 | ci->i_fragtree_nsplits++; | |
397 | frag->split_by = split_by; | |
3e7fbe9c | 398 | dout(" frag %x split by %d\n", frag->frag, frag->split_by); |
a4b7431f | 399 | prev_frag = frag; |
3e7fbe9c YZ |
400 | } |
401 | while (rb_node) { | |
402 | frag = rb_entry(rb_node, struct ceph_inode_frag, node); | |
403 | rb_node = rb_next(rb_node); | |
a4b7431f YZ |
404 | /* delete stale split/leaf node */ |
405 | if (frag->split_by > 0 || | |
406 | !is_frag_child(frag->frag, prev_frag)) { | |
407 | rb_erase(&frag->node, &ci->i_fragtree); | |
1b1bc16d YZ |
408 | if (frag->split_by > 0) |
409 | ci->i_fragtree_nsplits--; | |
a4b7431f YZ |
410 | kfree(frag); |
411 | } | |
3e7fbe9c YZ |
412 | } |
413 | out_unlock: | |
414 | mutex_unlock(&ci->i_fragtree_mutex); | |
415 | return 0; | |
416 | } | |
355da1eb SW |
417 | |
418 | /* | |
419 | * initialize a newly allocated inode. | |
420 | */ | |
421 | struct inode *ceph_alloc_inode(struct super_block *sb) | |
422 | { | |
423 | struct ceph_inode_info *ci; | |
424 | int i; | |
425 | ||
426 | ci = kmem_cache_alloc(ceph_inode_cachep, GFP_NOFS); | |
427 | if (!ci) | |
428 | return NULL; | |
429 | ||
430 | dout("alloc_inode %p\n", &ci->vfs_inode); | |
431 | ||
be655596 SW |
432 | spin_lock_init(&ci->i_ceph_lock); |
433 | ||
355da1eb | 434 | ci->i_version = 0; |
31c542a1 | 435 | ci->i_inline_version = 0; |
355da1eb SW |
436 | ci->i_time_warp_seq = 0; |
437 | ci->i_ceph_flags = 0; | |
fdd4e158 YZ |
438 | atomic64_set(&ci->i_ordered_count, 1); |
439 | atomic64_set(&ci->i_release_count, 1); | |
440 | atomic64_set(&ci->i_complete_seq[0], 0); | |
441 | atomic64_set(&ci->i_complete_seq[1], 0); | |
355da1eb SW |
442 | ci->i_symlink = NULL; |
443 | ||
fb18a575 LH |
444 | ci->i_max_bytes = 0; |
445 | ci->i_max_files = 0; | |
446 | ||
6c0f3af7 | 447 | memset(&ci->i_dir_layout, 0, sizeof(ci->i_dir_layout)); |
30c156d9 | 448 | RCU_INIT_POINTER(ci->i_layout.pool_ns, NULL); |
6c0f3af7 | 449 | |
355da1eb SW |
450 | ci->i_fragtree = RB_ROOT; |
451 | mutex_init(&ci->i_fragtree_mutex); | |
452 | ||
453 | ci->i_xattrs.blob = NULL; | |
454 | ci->i_xattrs.prealloc_blob = NULL; | |
455 | ci->i_xattrs.dirty = false; | |
456 | ci->i_xattrs.index = RB_ROOT; | |
457 | ci->i_xattrs.count = 0; | |
458 | ci->i_xattrs.names_size = 0; | |
459 | ci->i_xattrs.vals_size = 0; | |
460 | ci->i_xattrs.version = 0; | |
461 | ci->i_xattrs.index_version = 0; | |
462 | ||
463 | ci->i_caps = RB_ROOT; | |
464 | ci->i_auth_cap = NULL; | |
465 | ci->i_dirty_caps = 0; | |
466 | ci->i_flushing_caps = 0; | |
467 | INIT_LIST_HEAD(&ci->i_dirty_item); | |
468 | INIT_LIST_HEAD(&ci->i_flushing_item); | |
f66fd9f0 | 469 | ci->i_prealloc_cap_flush = NULL; |
e4500b5e | 470 | INIT_LIST_HEAD(&ci->i_cap_flush_list); |
355da1eb SW |
471 | init_waitqueue_head(&ci->i_cap_wq); |
472 | ci->i_hold_caps_min = 0; | |
473 | ci->i_hold_caps_max = 0; | |
474 | INIT_LIST_HEAD(&ci->i_cap_delay_list); | |
355da1eb SW |
475 | INIT_LIST_HEAD(&ci->i_cap_snaps); |
476 | ci->i_head_snapc = NULL; | |
477 | ci->i_snap_caps = 0; | |
478 | ||
774a6a11 | 479 | for (i = 0; i < CEPH_FILE_MODE_BITS; i++) |
355da1eb SW |
480 | ci->i_nr_by_mode[i] = 0; |
481 | ||
b0d7c223 | 482 | mutex_init(&ci->i_truncate_mutex); |
355da1eb SW |
483 | ci->i_truncate_seq = 0; |
484 | ci->i_truncate_size = 0; | |
485 | ci->i_truncate_pending = 0; | |
486 | ||
487 | ci->i_max_size = 0; | |
488 | ci->i_reported_size = 0; | |
489 | ci->i_wanted_max_size = 0; | |
490 | ci->i_requested_max_size = 0; | |
491 | ||
492 | ci->i_pin_ref = 0; | |
493 | ci->i_rd_ref = 0; | |
494 | ci->i_rdcache_ref = 0; | |
495 | ci->i_wr_ref = 0; | |
d3d0720d | 496 | ci->i_wb_ref = 0; |
355da1eb SW |
497 | ci->i_wrbuffer_ref = 0; |
498 | ci->i_wrbuffer_ref_head = 0; | |
89aa5930 | 499 | atomic_set(&ci->i_filelock_ref, 0); |
1e9c2eb6 | 500 | atomic_set(&ci->i_shared_gen, 1); |
355da1eb SW |
501 | ci->i_rdcache_gen = 0; |
502 | ci->i_rdcache_revoking = 0; | |
503 | ||
355da1eb | 504 | INIT_LIST_HEAD(&ci->i_unsafe_dirops); |
68cd5b4b | 505 | INIT_LIST_HEAD(&ci->i_unsafe_iops); |
355da1eb SW |
506 | spin_lock_init(&ci->i_unsafe_lock); |
507 | ||
508 | ci->i_snap_realm = NULL; | |
509 | INIT_LIST_HEAD(&ci->i_snap_realm_item); | |
510 | INIT_LIST_HEAD(&ci->i_snap_flush_item); | |
511 | ||
1cf89a8d YZ |
512 | INIT_WORK(&ci->i_work, ceph_inode_work); |
513 | ci->i_work_mask = 0; | |
245ce991 | 514 | memset(&ci->i_btime, '\0', sizeof(ci->i_btime)); |
355da1eb | 515 | |
99ccbd22 MT |
516 | ceph_fscache_inode_init(ci); |
517 | ||
355da1eb SW |
518 | return &ci->vfs_inode; |
519 | } | |
520 | ||
cfa6d412 | 521 | void ceph_free_inode(struct inode *inode) |
fa0d7e3d | 522 | { |
fa0d7e3d NP |
523 | struct ceph_inode_info *ci = ceph_inode(inode); |
524 | ||
daf5cc27 | 525 | kfree(ci->i_symlink); |
fa0d7e3d NP |
526 | kmem_cache_free(ceph_inode_cachep, ci); |
527 | } | |
528 | ||
87bc5b89 | 529 | void ceph_evict_inode(struct inode *inode) |
355da1eb SW |
530 | { |
531 | struct ceph_inode_info *ci = ceph_inode(inode); | |
532 | struct ceph_inode_frag *frag; | |
533 | struct rb_node *n; | |
534 | ||
87bc5b89 YZ |
535 | dout("evict_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode)); |
536 | ||
537 | truncate_inode_pages_final(&inode->i_data); | |
538 | clear_inode(inode); | |
355da1eb | 539 | |
99ccbd22 MT |
540 | ceph_fscache_unregister_inode_cookie(ci); |
541 | ||
d6e47819 | 542 | __ceph_remove_caps(ci); |
355da1eb | 543 | |
d557c48d LH |
544 | if (__ceph_has_any_quota(ci)) |
545 | ceph_adjust_quota_realms_count(inode, false); | |
546 | ||
8b218b8a SW |
547 | /* |
548 | * we may still have a snap_realm reference if there are stray | |
d9df2783 | 549 | * caps in i_snap_caps. |
8b218b8a SW |
550 | */ |
551 | if (ci->i_snap_realm) { | |
552 | struct ceph_mds_client *mdsc = | |
75c9627e YZ |
553 | ceph_inode_to_client(inode)->mdsc; |
554 | if (ceph_snap(inode) == CEPH_NOSNAP) { | |
555 | struct ceph_snap_realm *realm = ci->i_snap_realm; | |
556 | dout(" dropping residual ref to snap realm %p\n", | |
557 | realm); | |
558 | spin_lock(&realm->inodes_with_caps_lock); | |
559 | list_del_init(&ci->i_snap_realm_item); | |
560 | ci->i_snap_realm = NULL; | |
561 | if (realm->ino == ci->i_vino.ino) | |
562 | realm->inode = NULL; | |
563 | spin_unlock(&realm->inodes_with_caps_lock); | |
564 | ceph_put_snap_realm(mdsc, realm); | |
565 | } else { | |
566 | ceph_put_snapid_map(mdsc, ci->i_snapid_map); | |
567 | ci->i_snap_realm = NULL; | |
568 | } | |
8b218b8a SW |
569 | } |
570 | ||
355da1eb SW |
571 | while ((n = rb_first(&ci->i_fragtree)) != NULL) { |
572 | frag = rb_entry(n, struct ceph_inode_frag, node); | |
573 | rb_erase(n, &ci->i_fragtree); | |
574 | kfree(frag); | |
575 | } | |
1b1bc16d | 576 | ci->i_fragtree_nsplits = 0; |
355da1eb SW |
577 | |
578 | __ceph_destroy_xattrs(ci); | |
b6c1d5b8 SW |
579 | if (ci->i_xattrs.blob) |
580 | ceph_buffer_put(ci->i_xattrs.blob); | |
581 | if (ci->i_xattrs.prealloc_blob) | |
582 | ceph_buffer_put(ci->i_xattrs.prealloc_blob); | |
355da1eb | 583 | |
779fe0fb | 584 | ceph_put_string(rcu_dereference_raw(ci->i_layout.pool_ns)); |
355da1eb SW |
585 | } |
586 | ||
9f12bd11 YZ |
587 | int ceph_drop_inode(struct inode *inode) |
588 | { | |
589 | /* | |
590 | * Positve dentry and corresponding inode are always accompanied | |
591 | * in MDS reply. So no need to keep inode in the cache after | |
592 | * dropping all its aliases. | |
593 | */ | |
594 | return 1; | |
595 | } | |
596 | ||
224a7542 YZ |
597 | static inline blkcnt_t calc_inode_blocks(u64 size) |
598 | { | |
599 | return (size + (1<<9) - 1) >> 9; | |
600 | } | |
601 | ||
355da1eb SW |
602 | /* |
603 | * Helpers to fill in size, ctime, mtime, and atime. We have to be | |
604 | * careful because either the client or MDS may have more up to date | |
605 | * info, depending on which capabilities are held, and whether | |
606 | * time_warp_seq or truncate_seq have increased. (Ordinarily, mtime | |
607 | * and size are monotonically increasing, except when utimes() or | |
608 | * truncate() increments the corresponding _seq values.) | |
609 | */ | |
610 | int ceph_fill_file_size(struct inode *inode, int issued, | |
611 | u32 truncate_seq, u64 truncate_size, u64 size) | |
612 | { | |
613 | struct ceph_inode_info *ci = ceph_inode(inode); | |
614 | int queue_trunc = 0; | |
615 | ||
616 | if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 || | |
617 | (truncate_seq == ci->i_truncate_seq && size > inode->i_size)) { | |
618 | dout("size %lld -> %llu\n", inode->i_size, size); | |
a3d714c3 YZ |
619 | if (size > 0 && S_ISDIR(inode->i_mode)) { |
620 | pr_err("fill_file_size non-zero size for directory\n"); | |
621 | size = 0; | |
622 | } | |
99c88e69 | 623 | i_size_write(inode, size); |
224a7542 | 624 | inode->i_blocks = calc_inode_blocks(size); |
355da1eb SW |
625 | ci->i_reported_size = size; |
626 | if (truncate_seq != ci->i_truncate_seq) { | |
627 | dout("truncate_seq %u -> %u\n", | |
628 | ci->i_truncate_seq, truncate_seq); | |
629 | ci->i_truncate_seq = truncate_seq; | |
b0d7c223 YZ |
630 | |
631 | /* the MDS should have revoked these caps */ | |
632 | WARN_ON_ONCE(issued & (CEPH_CAP_FILE_EXCL | | |
633 | CEPH_CAP_FILE_RD | | |
634 | CEPH_CAP_FILE_WR | | |
635 | CEPH_CAP_FILE_LAZYIO)); | |
3d497d85 YS |
636 | /* |
637 | * If we hold relevant caps, or in the case where we're | |
638 | * not the only client referencing this file and we | |
639 | * don't hold those caps, then we need to check whether | |
640 | * the file is either opened or mmaped | |
641 | */ | |
b0d7c223 YZ |
642 | if ((issued & (CEPH_CAP_FILE_CACHE| |
643 | CEPH_CAP_FILE_BUFFER)) || | |
3d497d85 YS |
644 | mapping_mapped(inode->i_mapping) || |
645 | __ceph_caps_file_wanted(ci)) { | |
355da1eb SW |
646 | ci->i_truncate_pending++; |
647 | queue_trunc = 1; | |
648 | } | |
649 | } | |
650 | } | |
651 | if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 && | |
652 | ci->i_truncate_size != truncate_size) { | |
653 | dout("truncate_size %lld -> %llu\n", ci->i_truncate_size, | |
654 | truncate_size); | |
655 | ci->i_truncate_size = truncate_size; | |
656 | } | |
99ccbd22 MT |
657 | |
658 | if (queue_trunc) | |
659 | ceph_fscache_invalidate(inode); | |
660 | ||
355da1eb SW |
661 | return queue_trunc; |
662 | } | |
663 | ||
664 | void ceph_fill_file_time(struct inode *inode, int issued, | |
9bbeab41 AB |
665 | u64 time_warp_seq, struct timespec64 *ctime, |
666 | struct timespec64 *mtime, struct timespec64 *atime) | |
355da1eb SW |
667 | { |
668 | struct ceph_inode_info *ci = ceph_inode(inode); | |
669 | int warn = 0; | |
670 | ||
671 | if (issued & (CEPH_CAP_FILE_EXCL| | |
672 | CEPH_CAP_FILE_WR| | |
d8672d64 SW |
673 | CEPH_CAP_FILE_BUFFER| |
674 | CEPH_CAP_AUTH_EXCL| | |
675 | CEPH_CAP_XATTR_EXCL)) { | |
ffdeec7a | 676 | if (ci->i_version == 0 || |
9bbeab41 | 677 | timespec64_compare(ctime, &inode->i_ctime) > 0) { |
13442b03 | 678 | dout("ctime %lld.%09ld -> %lld.%09ld inc w/ cap\n", |
9bbeab41 AB |
679 | inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec, |
680 | ctime->tv_sec, ctime->tv_nsec); | |
681 | inode->i_ctime = *ctime; | |
355da1eb | 682 | } |
ffdeec7a YZ |
683 | if (ci->i_version == 0 || |
684 | ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) { | |
355da1eb | 685 | /* the MDS did a utimes() */ |
13442b03 | 686 | dout("mtime %lld.%09ld -> %lld.%09ld " |
355da1eb | 687 | "tw %d -> %d\n", |
9bbeab41 AB |
688 | inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec, |
689 | mtime->tv_sec, mtime->tv_nsec, | |
355da1eb SW |
690 | ci->i_time_warp_seq, (int)time_warp_seq); |
691 | ||
9bbeab41 AB |
692 | inode->i_mtime = *mtime; |
693 | inode->i_atime = *atime; | |
355da1eb SW |
694 | ci->i_time_warp_seq = time_warp_seq; |
695 | } else if (time_warp_seq == ci->i_time_warp_seq) { | |
696 | /* nobody did utimes(); take the max */ | |
9bbeab41 | 697 | if (timespec64_compare(mtime, &inode->i_mtime) > 0) { |
13442b03 | 698 | dout("mtime %lld.%09ld -> %lld.%09ld inc\n", |
9bbeab41 | 699 | inode->i_mtime.tv_sec, |
355da1eb | 700 | inode->i_mtime.tv_nsec, |
9bbeab41 AB |
701 | mtime->tv_sec, mtime->tv_nsec); |
702 | inode->i_mtime = *mtime; | |
355da1eb | 703 | } |
9bbeab41 | 704 | if (timespec64_compare(atime, &inode->i_atime) > 0) { |
13442b03 | 705 | dout("atime %lld.%09ld -> %lld.%09ld inc\n", |
9bbeab41 | 706 | inode->i_atime.tv_sec, |
355da1eb | 707 | inode->i_atime.tv_nsec, |
9bbeab41 AB |
708 | atime->tv_sec, atime->tv_nsec); |
709 | inode->i_atime = *atime; | |
355da1eb SW |
710 | } |
711 | } else if (issued & CEPH_CAP_FILE_EXCL) { | |
712 | /* we did a utimes(); ignore mds values */ | |
713 | } else { | |
714 | warn = 1; | |
715 | } | |
716 | } else { | |
d8672d64 | 717 | /* we have no write|excl caps; whatever the MDS says is true */ |
355da1eb | 718 | if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) { |
9bbeab41 AB |
719 | inode->i_ctime = *ctime; |
720 | inode->i_mtime = *mtime; | |
721 | inode->i_atime = *atime; | |
355da1eb SW |
722 | ci->i_time_warp_seq = time_warp_seq; |
723 | } else { | |
724 | warn = 1; | |
725 | } | |
726 | } | |
727 | if (warn) /* time_warp_seq shouldn't go backwards */ | |
728 | dout("%p mds time_warp_seq %llu < %u\n", | |
729 | inode, time_warp_seq, ci->i_time_warp_seq); | |
730 | } | |
731 | ||
732 | /* | |
733 | * Populate an inode based on info from mds. May be called on new or | |
734 | * existing inodes. | |
735 | */ | |
01deead0 | 736 | static int fill_inode(struct inode *inode, struct page *locked_page, |
355da1eb SW |
737 | struct ceph_mds_reply_info_in *iinfo, |
738 | struct ceph_mds_reply_dirfrag *dirinfo, | |
739 | struct ceph_mds_session *session, | |
740 | unsigned long ttl_from, int cap_fmode, | |
741 | struct ceph_cap_reservation *caps_reservation) | |
742 | { | |
d9df2783 | 743 | struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc; |
355da1eb SW |
744 | struct ceph_mds_reply_inode *info = iinfo->in; |
745 | struct ceph_inode_info *ci = ceph_inode(inode); | |
2af54a72 | 746 | int issued, new_issued, info_caps; |
9bbeab41 | 747 | struct timespec64 mtime, atime, ctime; |
355da1eb | 748 | struct ceph_buffer *xattr_blob = NULL; |
779fe0fb | 749 | struct ceph_string *pool_ns = NULL; |
d9df2783 | 750 | struct ceph_cap *new_cap = NULL; |
355da1eb | 751 | int err = 0; |
d9df2783 | 752 | bool wake = false; |
f98a128a YZ |
753 | bool queue_trunc = false; |
754 | bool new_version = false; | |
31c542a1 | 755 | bool fill_inline = false; |
355da1eb SW |
756 | |
757 | dout("fill_inode %p ino %llx.%llx v %llu had %llu\n", | |
758 | inode, ceph_vinop(inode), le64_to_cpu(info->version), | |
759 | ci->i_version); | |
760 | ||
2af54a72 YZ |
761 | info_caps = le32_to_cpu(info->cap.caps); |
762 | ||
d9df2783 | 763 | /* prealloc new cap struct */ |
2af54a72 | 764 | if (info_caps && ceph_snap(inode) == CEPH_NOSNAP) |
d9df2783 YZ |
765 | new_cap = ceph_get_cap(mdsc, caps_reservation); |
766 | ||
355da1eb SW |
767 | /* |
768 | * prealloc xattr data, if it looks like we'll need it. only | |
769 | * if len > 4 (meaning there are actually xattrs; the first 4 | |
770 | * bytes are the xattr count). | |
771 | */ | |
772 | if (iinfo->xattr_len > 4) { | |
b6c1d5b8 | 773 | xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS); |
355da1eb SW |
774 | if (!xattr_blob) |
775 | pr_err("fill_inode ENOMEM xattr blob %d bytes\n", | |
776 | iinfo->xattr_len); | |
777 | } | |
778 | ||
779fe0fb YZ |
779 | if (iinfo->pool_ns_len > 0) |
780 | pool_ns = ceph_find_or_create_string(iinfo->pool_ns_data, | |
781 | iinfo->pool_ns_len); | |
782 | ||
75c9627e YZ |
783 | if (ceph_snap(inode) != CEPH_NOSNAP && !ci->i_snapid_map) |
784 | ci->i_snapid_map = ceph_get_snapid_map(mdsc, ceph_snap(inode)); | |
785 | ||
be655596 | 786 | spin_lock(&ci->i_ceph_lock); |
355da1eb SW |
787 | |
788 | /* | |
789 | * provided version will be odd if inode value is projected, | |
8bd59e01 SW |
790 | * even if stable. skip the update if we have newer stable |
791 | * info (ours>=theirs, e.g. due to racing mds replies), unless | |
792 | * we are getting projected (unstable) info (in which case the | |
793 | * version is odd, and we want ours>theirs). | |
794 | * us them | |
795 | * 2 2 skip | |
796 | * 3 2 skip | |
797 | * 3 3 update | |
355da1eb | 798 | */ |
f98a128a YZ |
799 | if (ci->i_version == 0 || |
800 | ((info->cap.flags & CEPH_CAP_FLAG_AUTH) && | |
801 | le64_to_cpu(info->version) > (ci->i_version & ~1))) | |
802 | new_version = true; | |
803 | ||
a35ead31 JL |
804 | /* Update change_attribute */ |
805 | inode_set_max_iversion_raw(inode, iinfo->change_attr); | |
806 | ||
2af54a72 YZ |
807 | __ceph_caps_issued(ci, &issued); |
808 | issued |= __ceph_caps_dirty(ci); | |
809 | new_issued = ~issued & info_caps; | |
355da1eb SW |
810 | |
811 | /* update inode */ | |
355da1eb | 812 | inode->i_rdev = le32_to_cpu(info->rdev); |
f98a128a | 813 | inode->i_blkbits = fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1; |
355da1eb | 814 | |
d557c48d | 815 | __ceph_update_quota(ci, iinfo->max_bytes, iinfo->max_files); |
fb18a575 | 816 | |
f98a128a YZ |
817 | if ((new_version || (new_issued & CEPH_CAP_AUTH_SHARED)) && |
818 | (issued & CEPH_CAP_AUTH_EXCL) == 0) { | |
355da1eb | 819 | inode->i_mode = le32_to_cpu(info->mode); |
ab871b90 EB |
820 | inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(info->uid)); |
821 | inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(info->gid)); | |
355da1eb | 822 | dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode, |
bd2bae6a EB |
823 | from_kuid(&init_user_ns, inode->i_uid), |
824 | from_kgid(&init_user_ns, inode->i_gid)); | |
245ce991 | 825 | ceph_decode_timespec64(&ci->i_btime, &iinfo->btime); |
193e7b37 | 826 | ceph_decode_timespec64(&ci->i_snap_btime, &iinfo->snap_btime); |
355da1eb SW |
827 | } |
828 | ||
f98a128a YZ |
829 | if ((new_version || (new_issued & CEPH_CAP_LINK_SHARED)) && |
830 | (issued & CEPH_CAP_LINK_EXCL) == 0) | |
bfe86848 | 831 | set_nlink(inode, le32_to_cpu(info->nlink)); |
355da1eb | 832 | |
f98a128a YZ |
833 | if (new_version || (new_issued & CEPH_CAP_ANY_RD)) { |
834 | /* be careful with mtime, atime, size */ | |
9bbeab41 AB |
835 | ceph_decode_timespec64(&atime, &info->atime); |
836 | ceph_decode_timespec64(&mtime, &info->mtime); | |
837 | ceph_decode_timespec64(&ctime, &info->ctime); | |
f98a128a YZ |
838 | ceph_fill_file_time(inode, issued, |
839 | le32_to_cpu(info->time_warp_seq), | |
840 | &ctime, &mtime, &atime); | |
841 | } | |
842 | ||
2af54a72 YZ |
843 | if (new_version || (info_caps & CEPH_CAP_FILE_SHARED)) { |
844 | ci->i_files = le64_to_cpu(info->files); | |
845 | ci->i_subdirs = le64_to_cpu(info->subdirs); | |
846 | } | |
847 | ||
f98a128a YZ |
848 | if (new_version || |
849 | (new_issued & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR))) { | |
7627151e | 850 | s64 old_pool = ci->i_layout.pool_id; |
779fe0fb YZ |
851 | struct ceph_string *old_ns; |
852 | ||
7627151e | 853 | ceph_file_layout_from_legacy(&ci->i_layout, &info->layout); |
779fe0fb YZ |
854 | old_ns = rcu_dereference_protected(ci->i_layout.pool_ns, |
855 | lockdep_is_held(&ci->i_ceph_lock)); | |
856 | rcu_assign_pointer(ci->i_layout.pool_ns, pool_ns); | |
857 | ||
858 | if (ci->i_layout.pool_id != old_pool || pool_ns != old_ns) | |
10183a69 | 859 | ci->i_ceph_flags &= ~CEPH_I_POOL_PERM; |
10183a69 | 860 | |
779fe0fb | 861 | pool_ns = old_ns; |
10183a69 | 862 | |
f98a128a YZ |
863 | queue_trunc = ceph_fill_file_size(inode, issued, |
864 | le32_to_cpu(info->truncate_seq), | |
865 | le64_to_cpu(info->truncate_size), | |
866 | le64_to_cpu(info->size)); | |
867 | /* only update max_size on auth cap */ | |
868 | if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) && | |
869 | ci->i_max_size != le64_to_cpu(info->max_size)) { | |
870 | dout("max_size %lld -> %llu\n", ci->i_max_size, | |
871 | le64_to_cpu(info->max_size)); | |
872 | ci->i_max_size = le64_to_cpu(info->max_size); | |
873 | } | |
874 | } | |
355da1eb | 875 | |
49a9f4f6 YZ |
876 | /* layout and rstat are not tracked by capability, update them if |
877 | * the inode info is from auth mds */ | |
878 | if (new_version || (info->cap.flags & CEPH_CAP_FLAG_AUTH)) { | |
879 | if (S_ISDIR(inode->i_mode)) { | |
880 | ci->i_dir_layout = iinfo->dir_layout; | |
881 | ci->i_rbytes = le64_to_cpu(info->rbytes); | |
882 | ci->i_rfiles = le64_to_cpu(info->rfiles); | |
883 | ci->i_rsubdirs = le64_to_cpu(info->rsubdirs); | |
08796873 | 884 | ci->i_dir_pin = iinfo->dir_pin; |
9bbeab41 | 885 | ceph_decode_timespec64(&ci->i_rctime, &info->rctime); |
49a9f4f6 YZ |
886 | } |
887 | } | |
888 | ||
355da1eb SW |
889 | /* xattrs */ |
890 | /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */ | |
508b32d8 | 891 | if ((ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL)) && |
355da1eb SW |
892 | le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) { |
893 | if (ci->i_xattrs.blob) | |
894 | ceph_buffer_put(ci->i_xattrs.blob); | |
895 | ci->i_xattrs.blob = xattr_blob; | |
896 | if (xattr_blob) | |
897 | memcpy(ci->i_xattrs.blob->vec.iov_base, | |
898 | iinfo->xattr_data, iinfo->xattr_len); | |
899 | ci->i_xattrs.version = le64_to_cpu(info->xattr_version); | |
7221fe4c | 900 | ceph_forget_all_cached_acls(inode); |
ac6713cc | 901 | ceph_security_invalidate_secctx(inode); |
a6424e48 | 902 | xattr_blob = NULL; |
355da1eb SW |
903 | } |
904 | ||
ffdeec7a | 905 | /* finally update i_version */ |
aae1a442 YZ |
906 | if (le64_to_cpu(info->version) > ci->i_version) |
907 | ci->i_version = le64_to_cpu(info->version); | |
ffdeec7a | 908 | |
355da1eb | 909 | inode->i_mapping->a_ops = &ceph_aops; |
355da1eb SW |
910 | |
911 | switch (inode->i_mode & S_IFMT) { | |
912 | case S_IFIFO: | |
913 | case S_IFBLK: | |
914 | case S_IFCHR: | |
915 | case S_IFSOCK: | |
5ba72e60 | 916 | inode->i_blkbits = PAGE_SHIFT; |
355da1eb SW |
917 | init_special_inode(inode, inode->i_mode, inode->i_rdev); |
918 | inode->i_op = &ceph_file_iops; | |
919 | break; | |
920 | case S_IFREG: | |
921 | inode->i_op = &ceph_file_iops; | |
922 | inode->i_fop = &ceph_file_fops; | |
923 | break; | |
924 | case S_IFLNK: | |
925 | inode->i_op = &ceph_symlink_iops; | |
926 | if (!ci->i_symlink) { | |
810339ec | 927 | u32 symlen = iinfo->symlink_len; |
355da1eb SW |
928 | char *sym; |
929 | ||
be655596 | 930 | spin_unlock(&ci->i_ceph_lock); |
355da1eb | 931 | |
224a7542 YZ |
932 | if (symlen != i_size_read(inode)) { |
933 | pr_err("fill_inode %llx.%llx BAD symlink " | |
934 | "size %lld\n", ceph_vinop(inode), | |
935 | i_size_read(inode)); | |
936 | i_size_write(inode, symlen); | |
937 | inode->i_blocks = calc_inode_blocks(symlen); | |
938 | } | |
810339ec | 939 | |
355da1eb | 940 | err = -ENOMEM; |
810339ec | 941 | sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS); |
355da1eb SW |
942 | if (!sym) |
943 | goto out; | |
355da1eb | 944 | |
be655596 | 945 | spin_lock(&ci->i_ceph_lock); |
355da1eb SW |
946 | if (!ci->i_symlink) |
947 | ci->i_symlink = sym; | |
948 | else | |
949 | kfree(sym); /* lost a race */ | |
950 | } | |
ac194dcc | 951 | inode->i_link = ci->i_symlink; |
355da1eb SW |
952 | break; |
953 | case S_IFDIR: | |
954 | inode->i_op = &ceph_dir_iops; | |
955 | inode->i_fop = &ceph_dir_fops; | |
355da1eb SW |
956 | break; |
957 | default: | |
958 | pr_err("fill_inode %llx.%llx BAD mode 0%o\n", | |
959 | ceph_vinop(inode), inode->i_mode); | |
960 | } | |
961 | ||
355da1eb | 962 | /* were we issued a capability? */ |
2af54a72 | 963 | if (info_caps) { |
355da1eb SW |
964 | if (ceph_snap(inode) == CEPH_NOSNAP) { |
965 | ceph_add_cap(inode, session, | |
966 | le64_to_cpu(info->cap.cap_id), | |
2af54a72 | 967 | cap_fmode, info_caps, |
355da1eb SW |
968 | le32_to_cpu(info->cap.wanted), |
969 | le32_to_cpu(info->cap.seq), | |
970 | le32_to_cpu(info->cap.mseq), | |
971 | le64_to_cpu(info->cap.realm), | |
d9df2783 | 972 | info->cap.flags, &new_cap); |
2f92b3d0 YZ |
973 | |
974 | /* set dir completion flag? */ | |
975 | if (S_ISDIR(inode->i_mode) && | |
976 | ci->i_files == 0 && ci->i_subdirs == 0 && | |
2af54a72 | 977 | (info_caps & CEPH_CAP_FILE_SHARED) && |
2f92b3d0 YZ |
978 | (issued & CEPH_CAP_FILE_EXCL) == 0 && |
979 | !__ceph_dir_is_complete(ci)) { | |
980 | dout(" marking %p complete (empty)\n", inode); | |
fdd4e158 | 981 | i_size_write(inode, 0); |
2f92b3d0 | 982 | __ceph_dir_set_complete(ci, |
fdd4e158 YZ |
983 | atomic64_read(&ci->i_release_count), |
984 | atomic64_read(&ci->i_ordered_count)); | |
2f92b3d0 YZ |
985 | } |
986 | ||
d9df2783 | 987 | wake = true; |
355da1eb | 988 | } else { |
355da1eb | 989 | dout(" %p got snap_caps %s\n", inode, |
2af54a72 YZ |
990 | ceph_cap_string(info_caps)); |
991 | ci->i_snap_caps |= info_caps; | |
355da1eb SW |
992 | if (cap_fmode >= 0) |
993 | __ceph_get_fmode(ci, cap_fmode); | |
355da1eb | 994 | } |
04d000eb | 995 | } else if (cap_fmode >= 0) { |
f3ae1b97 | 996 | pr_warn("mds issued no caps on %llx.%llx\n", |
04d000eb SW |
997 | ceph_vinop(inode)); |
998 | __ceph_get_fmode(ci, cap_fmode); | |
355da1eb | 999 | } |
31c542a1 YZ |
1000 | |
1001 | if (iinfo->inline_version > 0 && | |
1002 | iinfo->inline_version >= ci->i_inline_version) { | |
1003 | int cache_caps = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO; | |
1004 | ci->i_inline_version = iinfo->inline_version; | |
1005 | if (ci->i_inline_version != CEPH_INLINE_NONE && | |
2af54a72 | 1006 | (locked_page || (info_caps & cache_caps))) |
31c542a1 YZ |
1007 | fill_inline = true; |
1008 | } | |
1009 | ||
be655596 | 1010 | spin_unlock(&ci->i_ceph_lock); |
355da1eb | 1011 | |
31c542a1 | 1012 | if (fill_inline) |
01deead0 | 1013 | ceph_fill_inline_data(inode, locked_page, |
31c542a1 YZ |
1014 | iinfo->inline_data, iinfo->inline_len); |
1015 | ||
d9df2783 YZ |
1016 | if (wake) |
1017 | wake_up_all(&ci->i_cap_wq); | |
1018 | ||
355da1eb SW |
1019 | /* queue truncate if we saw i_size decrease */ |
1020 | if (queue_trunc) | |
3c6f6b79 | 1021 | ceph_queue_vmtruncate(inode); |
355da1eb SW |
1022 | |
1023 | /* populate frag tree */ | |
3e7fbe9c YZ |
1024 | if (S_ISDIR(inode->i_mode)) |
1025 | ceph_fill_fragtree(inode, &info->fragtree, dirinfo); | |
355da1eb SW |
1026 | |
1027 | /* update delegation info? */ | |
1028 | if (dirinfo) | |
1029 | ceph_fill_dirfrag(inode, dirinfo); | |
1030 | ||
1031 | err = 0; | |
355da1eb | 1032 | out: |
d9df2783 YZ |
1033 | if (new_cap) |
1034 | ceph_put_cap(mdsc, new_cap); | |
b6c1d5b8 SW |
1035 | if (xattr_blob) |
1036 | ceph_buffer_put(xattr_blob); | |
779fe0fb | 1037 | ceph_put_string(pool_ns); |
355da1eb SW |
1038 | return err; |
1039 | } | |
1040 | ||
1041 | /* | |
543212b3 | 1042 | * caller should hold session s_mutex and dentry->d_lock. |
355da1eb | 1043 | */ |
543212b3 YZ |
1044 | static void __update_dentry_lease(struct inode *dir, struct dentry *dentry, |
1045 | struct ceph_mds_reply_lease *lease, | |
1046 | struct ceph_mds_session *session, | |
1047 | unsigned long from_time, | |
1048 | struct ceph_mds_session **old_lease_session) | |
355da1eb SW |
1049 | { |
1050 | struct ceph_dentry_info *di = ceph_dentry(dentry); | |
1051 | long unsigned duration = le32_to_cpu(lease->duration_ms); | |
1052 | long unsigned ttl = from_time + (duration * HZ) / 1000; | |
1053 | long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000; | |
f5d55f03 | 1054 | |
2f90b852 SW |
1055 | dout("update_dentry_lease %p duration %lu ms ttl %lu\n", |
1056 | dentry, duration, ttl); | |
355da1eb | 1057 | |
18fc8abd AV |
1058 | /* only track leases on regular dentries */ |
1059 | if (ceph_snap(dir) != CEPH_NOSNAP) | |
543212b3 | 1060 | return; |
18fc8abd | 1061 | |
97aeb6bf | 1062 | di->lease_shared_gen = atomic_read(&ceph_inode(dir)->i_shared_gen); |
37c4efc1 YZ |
1063 | if (duration == 0) { |
1064 | __ceph_dentry_dir_lease_touch(di); | |
543212b3 | 1065 | return; |
37c4efc1 | 1066 | } |
355da1eb SW |
1067 | |
1068 | if (di->lease_gen == session->s_cap_gen && | |
9b16f03c | 1069 | time_before(ttl, di->time)) |
543212b3 | 1070 | return; /* we already have a newer lease. */ |
355da1eb | 1071 | |
481f001f | 1072 | if (di->lease_session && di->lease_session != session) { |
543212b3 | 1073 | *old_lease_session = di->lease_session; |
481f001f YZ |
1074 | di->lease_session = NULL; |
1075 | } | |
355da1eb | 1076 | |
355da1eb SW |
1077 | if (!di->lease_session) |
1078 | di->lease_session = ceph_get_mds_session(session); | |
1079 | di->lease_gen = session->s_cap_gen; | |
1080 | di->lease_seq = le32_to_cpu(lease->seq); | |
1081 | di->lease_renew_after = half_ttl; | |
1082 | di->lease_renew_from = 0; | |
9b16f03c | 1083 | di->time = ttl; |
37c4efc1 YZ |
1084 | |
1085 | __ceph_dentry_lease_touch(di); | |
543212b3 YZ |
1086 | } |
1087 | ||
1088 | static inline void update_dentry_lease(struct inode *dir, struct dentry *dentry, | |
1089 | struct ceph_mds_reply_lease *lease, | |
1090 | struct ceph_mds_session *session, | |
1091 | unsigned long from_time) | |
1092 | { | |
1093 | struct ceph_mds_session *old_lease_session = NULL; | |
1094 | spin_lock(&dentry->d_lock); | |
1095 | __update_dentry_lease(dir, dentry, lease, session, from_time, | |
1096 | &old_lease_session); | |
1097 | spin_unlock(&dentry->d_lock); | |
1098 | if (old_lease_session) | |
1099 | ceph_put_mds_session(old_lease_session); | |
1100 | } | |
1101 | ||
1102 | /* | |
1103 | * update dentry lease without having parent inode locked | |
1104 | */ | |
1105 | static void update_dentry_lease_careful(struct dentry *dentry, | |
1106 | struct ceph_mds_reply_lease *lease, | |
1107 | struct ceph_mds_session *session, | |
1108 | unsigned long from_time, | |
1109 | char *dname, u32 dname_len, | |
1110 | struct ceph_vino *pdvino, | |
1111 | struct ceph_vino *ptvino) | |
1112 | ||
1113 | { | |
1114 | struct inode *dir; | |
1115 | struct ceph_mds_session *old_lease_session = NULL; | |
1116 | ||
1117 | spin_lock(&dentry->d_lock); | |
1118 | /* make sure dentry's name matches target */ | |
1119 | if (dentry->d_name.len != dname_len || | |
1120 | memcmp(dentry->d_name.name, dname, dname_len)) | |
1121 | goto out_unlock; | |
1122 | ||
1123 | dir = d_inode(dentry->d_parent); | |
1124 | /* make sure parent matches dvino */ | |
1125 | if (!ceph_ino_compare(dir, pdvino)) | |
1126 | goto out_unlock; | |
1127 | ||
1128 | /* make sure dentry's inode matches target. NULL ptvino means that | |
1129 | * we expect a negative dentry */ | |
1130 | if (ptvino) { | |
1131 | if (d_really_is_negative(dentry)) | |
1132 | goto out_unlock; | |
1133 | if (!ceph_ino_compare(d_inode(dentry), ptvino)) | |
1134 | goto out_unlock; | |
1135 | } else { | |
1136 | if (d_really_is_positive(dentry)) | |
1137 | goto out_unlock; | |
1138 | } | |
1139 | ||
1140 | __update_dentry_lease(dir, dentry, lease, session, | |
1141 | from_time, &old_lease_session); | |
355da1eb SW |
1142 | out_unlock: |
1143 | spin_unlock(&dentry->d_lock); | |
481f001f YZ |
1144 | if (old_lease_session) |
1145 | ceph_put_mds_session(old_lease_session); | |
355da1eb SW |
1146 | } |
1147 | ||
1148 | /* | |
1149 | * splice a dentry to an inode. | |
1150 | * caller must hold directory i_mutex for this to be safe. | |
355da1eb | 1151 | */ |
2bf996ac | 1152 | static int splice_dentry(struct dentry **pdn, struct inode *in) |
355da1eb | 1153 | { |
2bf996ac | 1154 | struct dentry *dn = *pdn; |
355da1eb SW |
1155 | struct dentry *realdn; |
1156 | ||
2b0143b5 | 1157 | BUG_ON(d_inode(dn)); |
1cd3935b | 1158 | |
5495c2d0 YZ |
1159 | if (S_ISDIR(in->i_mode)) { |
1160 | /* If inode is directory, d_splice_alias() below will remove | |
1161 | * 'realdn' from its origin parent. We need to ensure that | |
1162 | * origin parent's readdir cache will not reference 'realdn' | |
1163 | */ | |
1164 | realdn = d_find_any_alias(in); | |
1165 | if (realdn) { | |
1166 | struct ceph_dentry_info *di = ceph_dentry(realdn); | |
1167 | spin_lock(&realdn->d_lock); | |
1168 | ||
1169 | realdn->d_op->d_prune(realdn); | |
1170 | ||
1171 | di->time = jiffies; | |
1172 | di->lease_shared_gen = 0; | |
1173 | di->offset = 0; | |
1174 | ||
1175 | spin_unlock(&realdn->d_lock); | |
1176 | dput(realdn); | |
1177 | } | |
1178 | } | |
1179 | ||
355da1eb SW |
1180 | /* dn must be unhashed */ |
1181 | if (!d_unhashed(dn)) | |
1182 | d_drop(dn); | |
41d28bca | 1183 | realdn = d_splice_alias(in, dn); |
355da1eb | 1184 | if (IS_ERR(realdn)) { |
d69ed05a SW |
1185 | pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n", |
1186 | PTR_ERR(realdn), dn, in, ceph_vinop(in)); | |
2bf996ac YZ |
1187 | return PTR_ERR(realdn); |
1188 | } | |
1189 | ||
1190 | if (realdn) { | |
355da1eb SW |
1191 | dout("dn %p (%d) spliced with %p (%d) " |
1192 | "inode %p ino %llx.%llx\n", | |
84d08fa8 AV |
1193 | dn, d_count(dn), |
1194 | realdn, d_count(realdn), | |
2b0143b5 | 1195 | d_inode(realdn), ceph_vinop(d_inode(realdn))); |
355da1eb | 1196 | dput(dn); |
2bf996ac | 1197 | *pdn = realdn; |
355da1eb SW |
1198 | } else { |
1199 | BUG_ON(!ceph_dentry(dn)); | |
355da1eb | 1200 | dout("dn %p attached to %p ino %llx.%llx\n", |
2b0143b5 | 1201 | dn, d_inode(dn), ceph_vinop(d_inode(dn))); |
355da1eb | 1202 | } |
2bf996ac | 1203 | return 0; |
355da1eb SW |
1204 | } |
1205 | ||
1206 | /* | |
1207 | * Incorporate results into the local cache. This is either just | |
1208 | * one inode, or a directory, dentry, and possibly linked-to inode (e.g., | |
1209 | * after a lookup). | |
1210 | * | |
1211 | * A reply may contain | |
1212 | * a directory inode along with a dentry. | |
1213 | * and/or a target inode | |
1214 | * | |
1215 | * Called with snap_rwsem (read). | |
1216 | */ | |
f5a03b08 | 1217 | int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req) |
355da1eb | 1218 | { |
f5a03b08 | 1219 | struct ceph_mds_session *session = req->r_session; |
355da1eb SW |
1220 | struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info; |
1221 | struct inode *in = NULL; | |
f5d55f03 | 1222 | struct ceph_vino tvino, dvino; |
3d14c5d2 | 1223 | struct ceph_fs_client *fsc = ceph_sb_to_client(sb); |
355da1eb SW |
1224 | int err = 0; |
1225 | ||
1226 | dout("fill_trace %p is_dentry %d is_target %d\n", req, | |
1227 | rinfo->head->is_dentry, rinfo->head->is_target); | |
1228 | ||
355da1eb SW |
1229 | if (!rinfo->head->is_target && !rinfo->head->is_dentry) { |
1230 | dout("fill_trace reply is empty!\n"); | |
3dd69aab | 1231 | if (rinfo->head->result == 0 && req->r_parent) |
167c9e35 | 1232 | ceph_invalidate_dir_request(req); |
355da1eb SW |
1233 | return 0; |
1234 | } | |
1235 | ||
1236 | if (rinfo->head->is_dentry) { | |
3dd69aab | 1237 | struct inode *dir = req->r_parent; |
5b1daecd | 1238 | |
6c5e50fa | 1239 | if (dir) { |
01deead0 YZ |
1240 | err = fill_inode(dir, NULL, |
1241 | &rinfo->diri, rinfo->dirfrag, | |
6c5e50fa SW |
1242 | session, req->r_request_started, -1, |
1243 | &req->r_caps_reservation); | |
1244 | if (err < 0) | |
19913b4e | 1245 | goto done; |
6c5e50fa SW |
1246 | } else { |
1247 | WARN_ON_ONCE(1); | |
1248 | } | |
19913b4e | 1249 | |
74c9e6bf YZ |
1250 | if (dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME && |
1251 | test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) && | |
1252 | !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) { | |
19913b4e YZ |
1253 | struct qstr dname; |
1254 | struct dentry *dn, *parent; | |
1255 | ||
1256 | BUG_ON(!rinfo->head->is_target); | |
1257 | BUG_ON(req->r_dentry); | |
1258 | ||
1259 | parent = d_find_any_alias(dir); | |
1260 | BUG_ON(!parent); | |
1261 | ||
1262 | dname.name = rinfo->dname; | |
1263 | dname.len = rinfo->dname_len; | |
8387ff25 | 1264 | dname.hash = full_name_hash(parent, dname.name, dname.len); |
f5d55f03 JL |
1265 | tvino.ino = le64_to_cpu(rinfo->targeti.in->ino); |
1266 | tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid); | |
19913b4e YZ |
1267 | retry_lookup: |
1268 | dn = d_lookup(parent, &dname); | |
1269 | dout("d_lookup on parent=%p name=%.*s got %p\n", | |
1270 | parent, dname.len, dname.name, dn); | |
1271 | ||
1272 | if (!dn) { | |
1273 | dn = d_alloc(parent, &dname); | |
1274 | dout("d_alloc %p '%.*s' = %p\n", parent, | |
1275 | dname.len, dname.name, dn); | |
d37b1d99 | 1276 | if (!dn) { |
19913b4e YZ |
1277 | dput(parent); |
1278 | err = -ENOMEM; | |
1279 | goto done; | |
1280 | } | |
ad5cb123 | 1281 | err = 0; |
2b0143b5 | 1282 | } else if (d_really_is_positive(dn) && |
f5d55f03 JL |
1283 | (ceph_ino(d_inode(dn)) != tvino.ino || |
1284 | ceph_snap(d_inode(dn)) != tvino.snap)) { | |
19913b4e | 1285 | dout(" dn %p points to wrong inode %p\n", |
2b0143b5 | 1286 | dn, d_inode(dn)); |
933ad2c9 | 1287 | ceph_dir_clear_ordered(dir); |
19913b4e YZ |
1288 | d_delete(dn); |
1289 | dput(dn); | |
1290 | goto retry_lookup; | |
1291 | } | |
1292 | ||
1293 | req->r_dentry = dn; | |
1294 | dput(parent); | |
1295 | } | |
5b1daecd SW |
1296 | } |
1297 | ||
86b58d13 | 1298 | if (rinfo->head->is_target) { |
f5d55f03 JL |
1299 | tvino.ino = le64_to_cpu(rinfo->targeti.in->ino); |
1300 | tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid); | |
86b58d13 | 1301 | |
f5d55f03 | 1302 | in = ceph_get_inode(sb, tvino); |
86b58d13 YZ |
1303 | if (IS_ERR(in)) { |
1304 | err = PTR_ERR(in); | |
1305 | goto done; | |
1306 | } | |
1307 | req->r_target_inode = in; | |
1308 | ||
01deead0 | 1309 | err = fill_inode(in, req->r_locked_page, &rinfo->targeti, NULL, |
86b58d13 | 1310 | session, req->r_request_started, |
bc2de10d JL |
1311 | (!test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) && |
1312 | rinfo->head->result == 0) ? req->r_fmode : -1, | |
86b58d13 YZ |
1313 | &req->r_caps_reservation); |
1314 | if (err < 0) { | |
1315 | pr_err("fill_inode badness %p %llx.%llx\n", | |
1316 | in, ceph_vinop(in)); | |
1317 | goto done; | |
1318 | } | |
1319 | } | |
1320 | ||
9358c6d4 SW |
1321 | /* |
1322 | * ignore null lease/binding on snapdir ENOENT, or else we | |
1323 | * will have trouble splicing in the virtual snapdir later | |
1324 | */ | |
3dd69aab JL |
1325 | if (rinfo->head->is_dentry && |
1326 | !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) && | |
1327 | test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) && | |
9358c6d4 | 1328 | (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name, |
3d14c5d2 | 1329 | fsc->mount_options->snapdir_name, |
9358c6d4 | 1330 | req->r_dentry->d_name.len))) { |
355da1eb SW |
1331 | /* |
1332 | * lookup link rename : null -> possibly existing inode | |
1333 | * mknod symlink mkdir : null -> new inode | |
1334 | * unlink : linked -> null | |
1335 | */ | |
3dd69aab | 1336 | struct inode *dir = req->r_parent; |
355da1eb SW |
1337 | struct dentry *dn = req->r_dentry; |
1338 | bool have_dir_cap, have_lease; | |
1339 | ||
1340 | BUG_ON(!dn); | |
1341 | BUG_ON(!dir); | |
2b0143b5 | 1342 | BUG_ON(d_inode(dn->d_parent) != dir); |
f5d55f03 JL |
1343 | |
1344 | dvino.ino = le64_to_cpu(rinfo->diri.in->ino); | |
1345 | dvino.snap = le64_to_cpu(rinfo->diri.in->snapid); | |
1346 | ||
1347 | BUG_ON(ceph_ino(dir) != dvino.ino); | |
1348 | BUG_ON(ceph_snap(dir) != dvino.snap); | |
355da1eb | 1349 | |
355da1eb SW |
1350 | /* do we have a lease on the whole dir? */ |
1351 | have_dir_cap = | |
1352 | (le32_to_cpu(rinfo->diri.in->cap.caps) & | |
1353 | CEPH_CAP_FILE_SHARED); | |
1354 | ||
1355 | /* do we have a dn lease? */ | |
1356 | have_lease = have_dir_cap || | |
2f90b852 | 1357 | le32_to_cpu(rinfo->dlease->duration_ms); |
355da1eb SW |
1358 | if (!have_lease) |
1359 | dout("fill_trace no dentry lease or dir cap\n"); | |
1360 | ||
1361 | /* rename? */ | |
1362 | if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) { | |
0a8a70f9 YZ |
1363 | struct inode *olddir = req->r_old_dentry_dir; |
1364 | BUG_ON(!olddir); | |
1365 | ||
a455589f | 1366 | dout(" src %p '%pd' dst %p '%pd'\n", |
355da1eb | 1367 | req->r_old_dentry, |
a455589f AV |
1368 | req->r_old_dentry, |
1369 | dn, dn); | |
355da1eb SW |
1370 | dout("fill_trace doing d_move %p -> %p\n", |
1371 | req->r_old_dentry, dn); | |
c10f5e12 | 1372 | |
fdd4e158 YZ |
1373 | /* d_move screws up sibling dentries' offsets */ |
1374 | ceph_dir_clear_ordered(dir); | |
1375 | ceph_dir_clear_ordered(olddir); | |
1376 | ||
355da1eb | 1377 | d_move(req->r_old_dentry, dn); |
a455589f AV |
1378 | dout(" src %p '%pd' dst %p '%pd'\n", |
1379 | req->r_old_dentry, | |
355da1eb | 1380 | req->r_old_dentry, |
a455589f | 1381 | dn, dn); |
81a6cf2d | 1382 | |
c4a29f26 SW |
1383 | /* ensure target dentry is invalidated, despite |
1384 | rehashing bug in vfs_rename_dir */ | |
81a6cf2d SW |
1385 | ceph_invalidate_dentry_lease(dn); |
1386 | ||
99ccbd22 | 1387 | dout("dn %p gets new offset %lld\n", req->r_old_dentry, |
1cd3935b | 1388 | ceph_dentry(req->r_old_dentry)->offset); |
81a6cf2d | 1389 | |
2bf996ac YZ |
1390 | /* swap r_dentry and r_old_dentry in case that |
1391 | * splice_dentry() gets called later. This is safe | |
1392 | * because no other place will use them */ | |
1393 | req->r_dentry = req->r_old_dentry; | |
1394 | req->r_old_dentry = dn; | |
1395 | dn = req->r_dentry; | |
355da1eb SW |
1396 | } |
1397 | ||
1398 | /* null dentry? */ | |
1399 | if (!rinfo->head->is_target) { | |
1400 | dout("fill_trace null dentry\n"); | |
2b0143b5 | 1401 | if (d_really_is_positive(dn)) { |
355da1eb | 1402 | dout("d_delete %p\n", dn); |
5495c2d0 | 1403 | ceph_dir_clear_ordered(dir); |
355da1eb | 1404 | d_delete(dn); |
80d025ff JL |
1405 | } else if (have_lease) { |
1406 | if (d_unhashed(dn)) | |
f8b31710 | 1407 | d_add(dn, NULL); |
543212b3 YZ |
1408 | update_dentry_lease(dir, dn, |
1409 | rinfo->dlease, session, | |
1410 | req->r_request_started); | |
355da1eb SW |
1411 | } |
1412 | goto done; | |
1413 | } | |
1414 | ||
1415 | /* attach proper inode */ | |
2b0143b5 | 1416 | if (d_really_is_negative(dn)) { |
70db4f36 | 1417 | ceph_dir_clear_ordered(dir); |
86b58d13 | 1418 | ihold(in); |
2bf996ac YZ |
1419 | err = splice_dentry(&req->r_dentry, in); |
1420 | if (err < 0) | |
355da1eb | 1421 | goto done; |
2bf996ac | 1422 | dn = req->r_dentry; /* may have spliced */ |
2b0143b5 | 1423 | } else if (d_really_is_positive(dn) && d_inode(dn) != in) { |
355da1eb | 1424 | dout(" %p links to %p %llx.%llx, not %llx.%llx\n", |
2b0143b5 | 1425 | dn, d_inode(dn), ceph_vinop(d_inode(dn)), |
86b58d13 | 1426 | ceph_vinop(in)); |
200fd27c | 1427 | d_invalidate(dn); |
355da1eb | 1428 | have_lease = false; |
355da1eb SW |
1429 | } |
1430 | ||
f5d55f03 | 1431 | if (have_lease) { |
543212b3 YZ |
1432 | update_dentry_lease(dir, dn, |
1433 | rinfo->dlease, session, | |
1434 | req->r_request_started); | |
f5d55f03 | 1435 | } |
355da1eb | 1436 | dout(" final dn %p\n", dn); |
bc2de10d JL |
1437 | } else if ((req->r_op == CEPH_MDS_OP_LOOKUPSNAP || |
1438 | req->r_op == CEPH_MDS_OP_MKSNAP) && | |
1439 | !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) { | |
3dd69aab | 1440 | struct inode *dir = req->r_parent; |
355da1eb SW |
1441 | |
1442 | /* fill out a snapdir LOOKUPSNAP dentry */ | |
0a8a70f9 YZ |
1443 | BUG_ON(!dir); |
1444 | BUG_ON(ceph_snap(dir) != CEPH_SNAPDIR); | |
2bf996ac YZ |
1445 | BUG_ON(!req->r_dentry); |
1446 | dout(" linking snapped dir %p to dn %p\n", in, req->r_dentry); | |
70db4f36 | 1447 | ceph_dir_clear_ordered(dir); |
86b58d13 | 1448 | ihold(in); |
2bf996ac YZ |
1449 | err = splice_dentry(&req->r_dentry, in); |
1450 | if (err < 0) | |
355da1eb | 1451 | goto done; |
543212b3 YZ |
1452 | } else if (rinfo->head->is_dentry && req->r_dentry) { |
1453 | /* parent inode is not locked, be carefull */ | |
cdde7c43 | 1454 | struct ceph_vino *ptvino = NULL; |
543212b3 YZ |
1455 | dvino.ino = le64_to_cpu(rinfo->diri.in->ino); |
1456 | dvino.snap = le64_to_cpu(rinfo->diri.in->snapid); | |
1457 | if (rinfo->head->is_target) { | |
1458 | tvino.ino = le64_to_cpu(rinfo->targeti.in->ino); | |
1459 | tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid); | |
1460 | ptvino = &tvino; | |
cdde7c43 | 1461 | } |
543212b3 YZ |
1462 | update_dentry_lease_careful(req->r_dentry, rinfo->dlease, |
1463 | session, req->r_request_started, | |
1464 | rinfo->dname, rinfo->dname_len, | |
1465 | &dvino, ptvino); | |
355da1eb | 1466 | } |
355da1eb SW |
1467 | done: |
1468 | dout("fill_trace done err=%d\n", err); | |
1469 | return err; | |
1470 | } | |
1471 | ||
1472 | /* | |
1473 | * Prepopulate our cache with readdir results, leases, etc. | |
1474 | */ | |
79f9f99a SW |
1475 | static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req, |
1476 | struct ceph_mds_session *session) | |
1477 | { | |
1478 | struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info; | |
1479 | int i, err = 0; | |
1480 | ||
1481 | for (i = 0; i < rinfo->dir_nr; i++) { | |
2a5beea3 | 1482 | struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i; |
79f9f99a SW |
1483 | struct ceph_vino vino; |
1484 | struct inode *in; | |
1485 | int rc; | |
1486 | ||
2a5beea3 YZ |
1487 | vino.ino = le64_to_cpu(rde->inode.in->ino); |
1488 | vino.snap = le64_to_cpu(rde->inode.in->snapid); | |
79f9f99a SW |
1489 | |
1490 | in = ceph_get_inode(req->r_dentry->d_sb, vino); | |
1491 | if (IS_ERR(in)) { | |
1492 | err = PTR_ERR(in); | |
1493 | dout("new_inode badness got %d\n", err); | |
1494 | continue; | |
1495 | } | |
2a5beea3 | 1496 | rc = fill_inode(in, NULL, &rde->inode, NULL, session, |
79f9f99a SW |
1497 | req->r_request_started, -1, |
1498 | &req->r_caps_reservation); | |
1499 | if (rc < 0) { | |
1500 | pr_err("fill_inode badness on %p got %d\n", in, rc); | |
1501 | err = rc; | |
79f9f99a | 1502 | } |
3e1d0452 YZ |
1503 | /* avoid calling iput_final() in mds dispatch threads */ |
1504 | ceph_async_iput(in); | |
79f9f99a SW |
1505 | } |
1506 | ||
1507 | return err; | |
1508 | } | |
1509 | ||
fdd4e158 YZ |
1510 | void ceph_readdir_cache_release(struct ceph_readdir_cache_control *ctl) |
1511 | { | |
1512 | if (ctl->page) { | |
1513 | kunmap(ctl->page); | |
09cbfeaf | 1514 | put_page(ctl->page); |
fdd4e158 YZ |
1515 | ctl->page = NULL; |
1516 | } | |
1517 | } | |
1518 | ||
1519 | static int fill_readdir_cache(struct inode *dir, struct dentry *dn, | |
1520 | struct ceph_readdir_cache_control *ctl, | |
1521 | struct ceph_mds_request *req) | |
1522 | { | |
1523 | struct ceph_inode_info *ci = ceph_inode(dir); | |
09cbfeaf | 1524 | unsigned nsize = PAGE_SIZE / sizeof(struct dentry*); |
fdd4e158 YZ |
1525 | unsigned idx = ctl->index % nsize; |
1526 | pgoff_t pgoff = ctl->index / nsize; | |
1527 | ||
1528 | if (!ctl->page || pgoff != page_index(ctl->page)) { | |
1529 | ceph_readdir_cache_release(ctl); | |
af5e5eb5 YZ |
1530 | if (idx == 0) |
1531 | ctl->page = grab_cache_page(&dir->i_data, pgoff); | |
1532 | else | |
1533 | ctl->page = find_lock_page(&dir->i_data, pgoff); | |
fdd4e158 YZ |
1534 | if (!ctl->page) { |
1535 | ctl->index = -1; | |
af5e5eb5 | 1536 | return idx == 0 ? -ENOMEM : 0; |
fdd4e158 YZ |
1537 | } |
1538 | /* reading/filling the cache are serialized by | |
1539 | * i_mutex, no need to use page lock */ | |
1540 | unlock_page(ctl->page); | |
1541 | ctl->dentries = kmap(ctl->page); | |
af5e5eb5 | 1542 | if (idx == 0) |
09cbfeaf | 1543 | memset(ctl->dentries, 0, PAGE_SIZE); |
fdd4e158 YZ |
1544 | } |
1545 | ||
1546 | if (req->r_dir_release_cnt == atomic64_read(&ci->i_release_count) && | |
1547 | req->r_dir_ordered_cnt == atomic64_read(&ci->i_ordered_count)) { | |
1548 | dout("readdir cache dn %p idx %d\n", dn, ctl->index); | |
1549 | ctl->dentries[idx] = dn; | |
1550 | ctl->index++; | |
1551 | } else { | |
1552 | dout("disable readdir cache\n"); | |
1553 | ctl->index = -1; | |
1554 | } | |
1555 | return 0; | |
1556 | } | |
1557 | ||
355da1eb SW |
1558 | int ceph_readdir_prepopulate(struct ceph_mds_request *req, |
1559 | struct ceph_mds_session *session) | |
1560 | { | |
1561 | struct dentry *parent = req->r_dentry; | |
f3c4ebe6 | 1562 | struct ceph_inode_info *ci = ceph_inode(d_inode(parent)); |
355da1eb SW |
1563 | struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info; |
1564 | struct qstr dname; | |
1565 | struct dentry *dn; | |
1566 | struct inode *in; | |
315f2408 | 1567 | int err = 0, skipped = 0, ret, i; |
355da1eb | 1568 | struct ceph_mds_request_head *rhead = req->r_request->front.iov_base; |
81c6aea5 | 1569 | u32 frag = le32_to_cpu(rhead->args.readdir.frag); |
f3c4ebe6 YZ |
1570 | u32 last_hash = 0; |
1571 | u32 fpos_offset; | |
fdd4e158 YZ |
1572 | struct ceph_readdir_cache_control cache_ctl = {}; |
1573 | ||
bc2de10d | 1574 | if (test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) |
fdd4e158 | 1575 | return readdir_prepopulate_inodes_only(req, session); |
81c6aea5 | 1576 | |
79162547 YZ |
1577 | if (rinfo->hash_order) { |
1578 | if (req->r_path2) { | |
1579 | last_hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash, | |
1580 | req->r_path2, | |
1581 | strlen(req->r_path2)); | |
1582 | last_hash = ceph_frag_value(last_hash); | |
1583 | } else if (rinfo->offset_hash) { | |
1584 | /* mds understands offset_hash */ | |
1585 | WARN_ON_ONCE(req->r_readdir_offset != 2); | |
1586 | last_hash = le32_to_cpu(rhead->args.readdir.offset_hash); | |
1587 | } | |
f3c4ebe6 YZ |
1588 | } |
1589 | ||
81c6aea5 YZ |
1590 | if (rinfo->dir_dir && |
1591 | le32_to_cpu(rinfo->dir_dir->frag) != frag) { | |
1592 | dout("readdir_prepopulate got new frag %x -> %x\n", | |
1593 | frag, le32_to_cpu(rinfo->dir_dir->frag)); | |
1594 | frag = le32_to_cpu(rinfo->dir_dir->frag); | |
f3c4ebe6 | 1595 | if (!rinfo->hash_order) |
fdd4e158 | 1596 | req->r_readdir_offset = 2; |
81c6aea5 | 1597 | } |
355da1eb SW |
1598 | |
1599 | if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) { | |
355da1eb SW |
1600 | dout("readdir_prepopulate %d items under SNAPDIR dn %p\n", |
1601 | rinfo->dir_nr, parent); | |
1602 | } else { | |
1603 | dout("readdir_prepopulate %d items under dn %p\n", | |
1604 | rinfo->dir_nr, parent); | |
1605 | if (rinfo->dir_dir) | |
2b0143b5 | 1606 | ceph_fill_dirfrag(d_inode(parent), rinfo->dir_dir); |
355da1eb | 1607 | |
8d45b911 YZ |
1608 | if (ceph_frag_is_leftmost(frag) && |
1609 | req->r_readdir_offset == 2 && | |
1610 | !(rinfo->hash_order && last_hash)) { | |
1611 | /* note dir version at start of readdir so we can | |
1612 | * tell if any dentries get dropped */ | |
1613 | req->r_dir_release_cnt = | |
1614 | atomic64_read(&ci->i_release_count); | |
1615 | req->r_dir_ordered_cnt = | |
1616 | atomic64_read(&ci->i_ordered_count); | |
1617 | req->r_readdir_cache_idx = 0; | |
1618 | } | |
fdd4e158 YZ |
1619 | } |
1620 | ||
1621 | cache_ctl.index = req->r_readdir_cache_idx; | |
f3c4ebe6 | 1622 | fpos_offset = req->r_readdir_offset; |
fdd4e158 | 1623 | |
86b58d13 | 1624 | /* FIXME: release caps/leases if error occurs */ |
355da1eb | 1625 | for (i = 0; i < rinfo->dir_nr; i++) { |
2a5beea3 | 1626 | struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i; |
543212b3 | 1627 | struct ceph_vino tvino; |
355da1eb | 1628 | |
2a5beea3 YZ |
1629 | dname.name = rde->name; |
1630 | dname.len = rde->name_len; | |
8387ff25 | 1631 | dname.hash = full_name_hash(parent, dname.name, dname.len); |
355da1eb | 1632 | |
f5d55f03 JL |
1633 | tvino.ino = le64_to_cpu(rde->inode.in->ino); |
1634 | tvino.snap = le64_to_cpu(rde->inode.in->snapid); | |
355da1eb | 1635 | |
f3c4ebe6 YZ |
1636 | if (rinfo->hash_order) { |
1637 | u32 hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash, | |
1638 | rde->name, rde->name_len); | |
1639 | hash = ceph_frag_value(hash); | |
1640 | if (hash != last_hash) | |
1641 | fpos_offset = 2; | |
1642 | last_hash = hash; | |
1643 | rde->offset = ceph_make_fpos(hash, fpos_offset++, true); | |
1644 | } else { | |
1645 | rde->offset = ceph_make_fpos(frag, fpos_offset++, false); | |
1646 | } | |
355da1eb SW |
1647 | |
1648 | retry_lookup: | |
1649 | dn = d_lookup(parent, &dname); | |
1650 | dout("d_lookup on parent=%p name=%.*s got %p\n", | |
1651 | parent, dname.len, dname.name, dn); | |
1652 | ||
1653 | if (!dn) { | |
1654 | dn = d_alloc(parent, &dname); | |
1655 | dout("d_alloc %p '%.*s' = %p\n", parent, | |
1656 | dname.len, dname.name, dn); | |
d37b1d99 | 1657 | if (!dn) { |
355da1eb SW |
1658 | dout("d_alloc badness\n"); |
1659 | err = -ENOMEM; | |
1660 | goto out; | |
1661 | } | |
2b0143b5 | 1662 | } else if (d_really_is_positive(dn) && |
f5d55f03 JL |
1663 | (ceph_ino(d_inode(dn)) != tvino.ino || |
1664 | ceph_snap(d_inode(dn)) != tvino.snap)) { | |
5495c2d0 | 1665 | struct ceph_dentry_info *di = ceph_dentry(dn); |
355da1eb | 1666 | dout(" dn %p points to wrong inode %p\n", |
2b0143b5 | 1667 | dn, d_inode(dn)); |
5495c2d0 YZ |
1668 | |
1669 | spin_lock(&dn->d_lock); | |
1670 | if (di->offset > 0 && | |
1671 | di->lease_shared_gen == | |
1672 | atomic_read(&ci->i_shared_gen)) { | |
1673 | __ceph_dir_clear_ordered(ci); | |
1674 | di->offset = 0; | |
1675 | } | |
1676 | spin_unlock(&dn->d_lock); | |
1677 | ||
355da1eb SW |
1678 | d_delete(dn); |
1679 | dput(dn); | |
1680 | goto retry_lookup; | |
355da1eb SW |
1681 | } |
1682 | ||
355da1eb | 1683 | /* inode */ |
2b0143b5 DH |
1684 | if (d_really_is_positive(dn)) { |
1685 | in = d_inode(dn); | |
355da1eb | 1686 | } else { |
f5d55f03 | 1687 | in = ceph_get_inode(parent->d_sb, tvino); |
ac1f12ef | 1688 | if (IS_ERR(in)) { |
355da1eb | 1689 | dout("new_inode badness\n"); |
2744c171 | 1690 | d_drop(dn); |
355da1eb | 1691 | dput(dn); |
ac1f12ef | 1692 | err = PTR_ERR(in); |
355da1eb SW |
1693 | goto out; |
1694 | } | |
355da1eb SW |
1695 | } |
1696 | ||
2a5beea3 | 1697 | ret = fill_inode(in, NULL, &rde->inode, NULL, session, |
fdd4e158 YZ |
1698 | req->r_request_started, -1, |
1699 | &req->r_caps_reservation); | |
1700 | if (ret < 0) { | |
355da1eb | 1701 | pr_err("fill_inode badness on %p\n", in); |
3e1d0452 YZ |
1702 | if (d_really_is_negative(dn)) { |
1703 | /* avoid calling iput_final() in mds | |
1704 | * dispatch threads */ | |
1705 | ceph_async_iput(in); | |
1706 | } | |
86b58d13 | 1707 | d_drop(dn); |
fdd4e158 | 1708 | err = ret; |
d69ed05a | 1709 | goto next_item; |
355da1eb | 1710 | } |
86b58d13 | 1711 | |
2b0143b5 | 1712 | if (d_really_is_negative(dn)) { |
315f2408 YZ |
1713 | if (ceph_security_xattr_deadlock(in)) { |
1714 | dout(" skip splicing dn %p to inode %p" | |
1715 | " (security xattr deadlock)\n", dn, in); | |
3e1d0452 | 1716 | ceph_async_iput(in); |
315f2408 YZ |
1717 | skipped++; |
1718 | goto next_item; | |
1719 | } | |
1720 | ||
2bf996ac YZ |
1721 | err = splice_dentry(&dn, in); |
1722 | if (err < 0) | |
86b58d13 | 1723 | goto next_item; |
86b58d13 YZ |
1724 | } |
1725 | ||
f3c4ebe6 | 1726 | ceph_dentry(dn)->offset = rde->offset; |
86b58d13 | 1727 | |
543212b3 YZ |
1728 | update_dentry_lease(d_inode(parent), dn, |
1729 | rde->lease, req->r_session, | |
1730 | req->r_request_started); | |
fdd4e158 | 1731 | |
315f2408 | 1732 | if (err == 0 && skipped == 0 && cache_ctl.index >= 0) { |
fdd4e158 YZ |
1733 | ret = fill_readdir_cache(d_inode(parent), dn, |
1734 | &cache_ctl, req); | |
1735 | if (ret < 0) | |
1736 | err = ret; | |
1737 | } | |
d69ed05a | 1738 | next_item: |
2bf996ac | 1739 | dput(dn); |
355da1eb | 1740 | } |
355da1eb | 1741 | out: |
315f2408 | 1742 | if (err == 0 && skipped == 0) { |
bc2de10d | 1743 | set_bit(CEPH_MDS_R_DID_PREPOPULATE, &req->r_req_flags); |
fdd4e158 YZ |
1744 | req->r_readdir_cache_idx = cache_ctl.index; |
1745 | } | |
1746 | ceph_readdir_cache_release(&cache_ctl); | |
355da1eb SW |
1747 | dout("readdir_prepopulate done\n"); |
1748 | return err; | |
1749 | } | |
1750 | ||
efb0ca76 | 1751 | bool ceph_inode_set_size(struct inode *inode, loff_t size) |
355da1eb SW |
1752 | { |
1753 | struct ceph_inode_info *ci = ceph_inode(inode); | |
efb0ca76 | 1754 | bool ret; |
355da1eb | 1755 | |
be655596 | 1756 | spin_lock(&ci->i_ceph_lock); |
355da1eb | 1757 | dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size); |
99c88e69 | 1758 | i_size_write(inode, size); |
224a7542 | 1759 | inode->i_blocks = calc_inode_blocks(size); |
355da1eb | 1760 | |
efb0ca76 | 1761 | ret = __ceph_should_report_size(ci); |
355da1eb | 1762 | |
be655596 | 1763 | spin_unlock(&ci->i_ceph_lock); |
355da1eb SW |
1764 | return ret; |
1765 | } | |
1766 | ||
3e1d0452 YZ |
1767 | /* |
1768 | * Put reference to inode, but avoid calling iput_final() in current thread. | |
1769 | * iput_final() may wait for reahahead pages. The wait can cause deadlock in | |
1770 | * some contexts. | |
1771 | */ | |
1772 | void ceph_async_iput(struct inode *inode) | |
1773 | { | |
1774 | if (!inode) | |
1775 | return; | |
1776 | for (;;) { | |
1777 | if (atomic_add_unless(&inode->i_count, -1, 1)) | |
1778 | break; | |
1779 | if (queue_work(ceph_inode_to_client(inode)->inode_wq, | |
1780 | &ceph_inode(inode)->i_work)) | |
1781 | break; | |
1782 | /* queue work failed, i_count must be at least 2 */ | |
1783 | } | |
1784 | } | |
1785 | ||
355da1eb SW |
1786 | /* |
1787 | * Write back inode data in a worker thread. (This can't be done | |
1788 | * in the message handler context.) | |
1789 | */ | |
3c6f6b79 SW |
1790 | void ceph_queue_writeback(struct inode *inode) |
1791 | { | |
1cf89a8d YZ |
1792 | struct ceph_inode_info *ci = ceph_inode(inode); |
1793 | set_bit(CEPH_I_WORK_WRITEBACK, &ci->i_work_mask); | |
1794 | ||
15a2015f | 1795 | ihold(inode); |
1cf89a8d YZ |
1796 | if (queue_work(ceph_inode_to_client(inode)->inode_wq, |
1797 | &ci->i_work)) { | |
2c27c9a5 | 1798 | dout("ceph_queue_writeback %p\n", inode); |
3c6f6b79 | 1799 | } else { |
1cf89a8d YZ |
1800 | dout("ceph_queue_writeback %p already queued, mask=%lx\n", |
1801 | inode, ci->i_work_mask); | |
15a2015f | 1802 | iput(inode); |
3c6f6b79 SW |
1803 | } |
1804 | } | |
1805 | ||
3c6f6b79 SW |
1806 | /* |
1807 | * queue an async invalidation | |
1808 | */ | |
1809 | void ceph_queue_invalidate(struct inode *inode) | |
1810 | { | |
1cf89a8d YZ |
1811 | struct ceph_inode_info *ci = ceph_inode(inode); |
1812 | set_bit(CEPH_I_WORK_INVALIDATE_PAGES, &ci->i_work_mask); | |
1813 | ||
15a2015f | 1814 | ihold(inode); |
1cf89a8d YZ |
1815 | if (queue_work(ceph_inode_to_client(inode)->inode_wq, |
1816 | &ceph_inode(inode)->i_work)) { | |
3c6f6b79 | 1817 | dout("ceph_queue_invalidate %p\n", inode); |
3c6f6b79 | 1818 | } else { |
1cf89a8d YZ |
1819 | dout("ceph_queue_invalidate %p already queued, mask=%lx\n", |
1820 | inode, ci->i_work_mask); | |
15a2015f | 1821 | iput(inode); |
3c6f6b79 SW |
1822 | } |
1823 | } | |
1824 | ||
355da1eb | 1825 | /* |
1cf89a8d YZ |
1826 | * Queue an async vmtruncate. If we fail to queue work, we will handle |
1827 | * the truncation the next time we call __ceph_do_pending_vmtruncate. | |
355da1eb | 1828 | */ |
1cf89a8d | 1829 | void ceph_queue_vmtruncate(struct inode *inode) |
355da1eb | 1830 | { |
1cf89a8d YZ |
1831 | struct ceph_inode_info *ci = ceph_inode(inode); |
1832 | set_bit(CEPH_I_WORK_VMTRUNCATE, &ci->i_work_mask); | |
1833 | ||
1834 | ihold(inode); | |
1835 | if (queue_work(ceph_inode_to_client(inode)->inode_wq, | |
1836 | &ci->i_work)) { | |
1837 | dout("ceph_queue_vmtruncate %p\n", inode); | |
1838 | } else { | |
1839 | dout("ceph_queue_vmtruncate %p already queued, mask=%lx\n", | |
1840 | inode, ci->i_work_mask); | |
1841 | iput(inode); | |
1842 | } | |
1843 | } | |
1844 | ||
1845 | static void ceph_do_invalidate_pages(struct inode *inode) | |
1846 | { | |
1847 | struct ceph_inode_info *ci = ceph_inode(inode); | |
6c93df5d | 1848 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); |
355da1eb SW |
1849 | u32 orig_gen; |
1850 | int check = 0; | |
1851 | ||
b0d7c223 | 1852 | mutex_lock(&ci->i_truncate_mutex); |
6c93df5d | 1853 | |
52953d55 | 1854 | if (READ_ONCE(fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) { |
6c93df5d YZ |
1855 | pr_warn_ratelimited("invalidate_pages %p %lld forced umount\n", |
1856 | inode, ceph_ino(inode)); | |
1857 | mapping_set_error(inode->i_mapping, -EIO); | |
1858 | truncate_pagecache(inode, 0); | |
1859 | mutex_unlock(&ci->i_truncate_mutex); | |
1860 | goto out; | |
1861 | } | |
1862 | ||
be655596 | 1863 | spin_lock(&ci->i_ceph_lock); |
355da1eb SW |
1864 | dout("invalidate_pages %p gen %d revoking %d\n", inode, |
1865 | ci->i_rdcache_gen, ci->i_rdcache_revoking); | |
cd045cb4 | 1866 | if (ci->i_rdcache_revoking != ci->i_rdcache_gen) { |
9563f88c YZ |
1867 | if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE)) |
1868 | check = 1; | |
be655596 | 1869 | spin_unlock(&ci->i_ceph_lock); |
b0d7c223 | 1870 | mutex_unlock(&ci->i_truncate_mutex); |
355da1eb SW |
1871 | goto out; |
1872 | } | |
1873 | orig_gen = ci->i_rdcache_gen; | |
be655596 | 1874 | spin_unlock(&ci->i_ceph_lock); |
355da1eb | 1875 | |
9abd4db7 YZ |
1876 | if (invalidate_inode_pages2(inode->i_mapping) < 0) { |
1877 | pr_err("invalidate_pages %p fails\n", inode); | |
1878 | } | |
355da1eb | 1879 | |
be655596 | 1880 | spin_lock(&ci->i_ceph_lock); |
cd045cb4 SW |
1881 | if (orig_gen == ci->i_rdcache_gen && |
1882 | orig_gen == ci->i_rdcache_revoking) { | |
355da1eb SW |
1883 | dout("invalidate_pages %p gen %d successful\n", inode, |
1884 | ci->i_rdcache_gen); | |
cd045cb4 | 1885 | ci->i_rdcache_revoking--; |
355da1eb SW |
1886 | check = 1; |
1887 | } else { | |
cd045cb4 SW |
1888 | dout("invalidate_pages %p gen %d raced, now %d revoking %d\n", |
1889 | inode, orig_gen, ci->i_rdcache_gen, | |
1890 | ci->i_rdcache_revoking); | |
9563f88c YZ |
1891 | if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE)) |
1892 | check = 1; | |
355da1eb | 1893 | } |
be655596 | 1894 | spin_unlock(&ci->i_ceph_lock); |
b0d7c223 | 1895 | mutex_unlock(&ci->i_truncate_mutex); |
9563f88c | 1896 | out: |
355da1eb SW |
1897 | if (check) |
1898 | ceph_check_caps(ci, 0, NULL); | |
3c6f6b79 SW |
1899 | } |
1900 | ||
355da1eb | 1901 | /* |
355da1eb SW |
1902 | * Make sure any pending truncation is applied before doing anything |
1903 | * that may depend on it. | |
1904 | */ | |
b415bf4f | 1905 | void __ceph_do_pending_vmtruncate(struct inode *inode) |
355da1eb SW |
1906 | { |
1907 | struct ceph_inode_info *ci = ceph_inode(inode); | |
1908 | u64 to; | |
a85f50b6 | 1909 | int wrbuffer_refs, finish = 0; |
355da1eb | 1910 | |
b0d7c223 | 1911 | mutex_lock(&ci->i_truncate_mutex); |
355da1eb | 1912 | retry: |
be655596 | 1913 | spin_lock(&ci->i_ceph_lock); |
355da1eb SW |
1914 | if (ci->i_truncate_pending == 0) { |
1915 | dout("__do_pending_vmtruncate %p none pending\n", inode); | |
be655596 | 1916 | spin_unlock(&ci->i_ceph_lock); |
b0d7c223 | 1917 | mutex_unlock(&ci->i_truncate_mutex); |
355da1eb SW |
1918 | return; |
1919 | } | |
1920 | ||
1921 | /* | |
1922 | * make sure any dirty snapped pages are flushed before we | |
1923 | * possibly truncate them.. so write AND block! | |
1924 | */ | |
1925 | if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) { | |
c8fd0d37 | 1926 | spin_unlock(&ci->i_ceph_lock); |
355da1eb SW |
1927 | dout("__do_pending_vmtruncate %p flushing snaps first\n", |
1928 | inode); | |
355da1eb SW |
1929 | filemap_write_and_wait_range(&inode->i_data, 0, |
1930 | inode->i_sb->s_maxbytes); | |
1931 | goto retry; | |
1932 | } | |
1933 | ||
b0d7c223 YZ |
1934 | /* there should be no reader or writer */ |
1935 | WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref); | |
1936 | ||
355da1eb SW |
1937 | to = ci->i_truncate_size; |
1938 | wrbuffer_refs = ci->i_wrbuffer_ref; | |
1939 | dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode, | |
1940 | ci->i_truncate_pending, to); | |
be655596 | 1941 | spin_unlock(&ci->i_ceph_lock); |
355da1eb | 1942 | |
4e217b5d | 1943 | truncate_pagecache(inode, to); |
355da1eb | 1944 | |
be655596 | 1945 | spin_lock(&ci->i_ceph_lock); |
a85f50b6 YZ |
1946 | if (to == ci->i_truncate_size) { |
1947 | ci->i_truncate_pending = 0; | |
1948 | finish = 1; | |
1949 | } | |
be655596 | 1950 | spin_unlock(&ci->i_ceph_lock); |
a85f50b6 YZ |
1951 | if (!finish) |
1952 | goto retry; | |
355da1eb | 1953 | |
b0d7c223 YZ |
1954 | mutex_unlock(&ci->i_truncate_mutex); |
1955 | ||
355da1eb SW |
1956 | if (wrbuffer_refs == 0) |
1957 | ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL); | |
a85f50b6 YZ |
1958 | |
1959 | wake_up_all(&ci->i_cap_wq); | |
355da1eb SW |
1960 | } |
1961 | ||
1cf89a8d YZ |
1962 | static void ceph_inode_work(struct work_struct *work) |
1963 | { | |
1964 | struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info, | |
1965 | i_work); | |
1966 | struct inode *inode = &ci->vfs_inode; | |
1967 | ||
1968 | if (test_and_clear_bit(CEPH_I_WORK_WRITEBACK, &ci->i_work_mask)) { | |
1969 | dout("writeback %p\n", inode); | |
1970 | filemap_fdatawrite(&inode->i_data); | |
1971 | } | |
1972 | if (test_and_clear_bit(CEPH_I_WORK_INVALIDATE_PAGES, &ci->i_work_mask)) | |
1973 | ceph_do_invalidate_pages(inode); | |
1974 | ||
1975 | if (test_and_clear_bit(CEPH_I_WORK_VMTRUNCATE, &ci->i_work_mask)) | |
1976 | __ceph_do_pending_vmtruncate(inode); | |
1977 | ||
1978 | iput(inode); | |
1979 | } | |
1980 | ||
355da1eb SW |
1981 | /* |
1982 | * symlinks | |
1983 | */ | |
355da1eb | 1984 | static const struct inode_operations ceph_symlink_iops = { |
6b255391 | 1985 | .get_link = simple_get_link, |
0b932672 YZ |
1986 | .setattr = ceph_setattr, |
1987 | .getattr = ceph_getattr, | |
0b932672 | 1988 | .listxattr = ceph_listxattr, |
355da1eb SW |
1989 | }; |
1990 | ||
a26fecca | 1991 | int __ceph_setattr(struct inode *inode, struct iattr *attr) |
355da1eb | 1992 | { |
355da1eb | 1993 | struct ceph_inode_info *ci = ceph_inode(inode); |
355da1eb SW |
1994 | const unsigned int ia_valid = attr->ia_valid; |
1995 | struct ceph_mds_request *req; | |
a26fecca | 1996 | struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc; |
f66fd9f0 | 1997 | struct ceph_cap_flush *prealloc_cf; |
355da1eb SW |
1998 | int issued; |
1999 | int release = 0, dirtied = 0; | |
2000 | int mask = 0; | |
2001 | int err = 0; | |
fca65b4a | 2002 | int inode_dirty_flags = 0; |
604d1b02 | 2003 | bool lock_snap_rwsem = false; |
355da1eb | 2004 | |
f66fd9f0 YZ |
2005 | prealloc_cf = ceph_alloc_cap_flush(); |
2006 | if (!prealloc_cf) | |
2007 | return -ENOMEM; | |
2008 | ||
355da1eb SW |
2009 | req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR, |
2010 | USE_AUTH_MDS); | |
f66fd9f0 YZ |
2011 | if (IS_ERR(req)) { |
2012 | ceph_free_cap_flush(prealloc_cf); | |
355da1eb | 2013 | return PTR_ERR(req); |
f66fd9f0 | 2014 | } |
355da1eb | 2015 | |
be655596 | 2016 | spin_lock(&ci->i_ceph_lock); |
355da1eb | 2017 | issued = __ceph_caps_issued(ci, NULL); |
604d1b02 YZ |
2018 | |
2019 | if (!ci->i_head_snapc && | |
2020 | (issued & (CEPH_CAP_ANY_EXCL | CEPH_CAP_FILE_WR))) { | |
2021 | lock_snap_rwsem = true; | |
2022 | if (!down_read_trylock(&mdsc->snap_rwsem)) { | |
2023 | spin_unlock(&ci->i_ceph_lock); | |
2024 | down_read(&mdsc->snap_rwsem); | |
2025 | spin_lock(&ci->i_ceph_lock); | |
2026 | issued = __ceph_caps_issued(ci, NULL); | |
2027 | } | |
2028 | } | |
2029 | ||
355da1eb SW |
2030 | dout("setattr %p issued %s\n", inode, ceph_cap_string(issued)); |
2031 | ||
2032 | if (ia_valid & ATTR_UID) { | |
2033 | dout("setattr %p uid %d -> %d\n", inode, | |
bd2bae6a EB |
2034 | from_kuid(&init_user_ns, inode->i_uid), |
2035 | from_kuid(&init_user_ns, attr->ia_uid)); | |
355da1eb SW |
2036 | if (issued & CEPH_CAP_AUTH_EXCL) { |
2037 | inode->i_uid = attr->ia_uid; | |
2038 | dirtied |= CEPH_CAP_AUTH_EXCL; | |
2039 | } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || | |
ab871b90 EB |
2040 | !uid_eq(attr->ia_uid, inode->i_uid)) { |
2041 | req->r_args.setattr.uid = cpu_to_le32( | |
2042 | from_kuid(&init_user_ns, attr->ia_uid)); | |
355da1eb SW |
2043 | mask |= CEPH_SETATTR_UID; |
2044 | release |= CEPH_CAP_AUTH_SHARED; | |
2045 | } | |
2046 | } | |
2047 | if (ia_valid & ATTR_GID) { | |
2048 | dout("setattr %p gid %d -> %d\n", inode, | |
bd2bae6a EB |
2049 | from_kgid(&init_user_ns, inode->i_gid), |
2050 | from_kgid(&init_user_ns, attr->ia_gid)); | |
355da1eb SW |
2051 | if (issued & CEPH_CAP_AUTH_EXCL) { |
2052 | inode->i_gid = attr->ia_gid; | |
2053 | dirtied |= CEPH_CAP_AUTH_EXCL; | |
2054 | } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || | |
ab871b90 EB |
2055 | !gid_eq(attr->ia_gid, inode->i_gid)) { |
2056 | req->r_args.setattr.gid = cpu_to_le32( | |
2057 | from_kgid(&init_user_ns, attr->ia_gid)); | |
355da1eb SW |
2058 | mask |= CEPH_SETATTR_GID; |
2059 | release |= CEPH_CAP_AUTH_SHARED; | |
2060 | } | |
2061 | } | |
2062 | if (ia_valid & ATTR_MODE) { | |
2063 | dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode, | |
2064 | attr->ia_mode); | |
2065 | if (issued & CEPH_CAP_AUTH_EXCL) { | |
2066 | inode->i_mode = attr->ia_mode; | |
2067 | dirtied |= CEPH_CAP_AUTH_EXCL; | |
2068 | } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || | |
2069 | attr->ia_mode != inode->i_mode) { | |
7221fe4c | 2070 | inode->i_mode = attr->ia_mode; |
355da1eb SW |
2071 | req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode); |
2072 | mask |= CEPH_SETATTR_MODE; | |
2073 | release |= CEPH_CAP_AUTH_SHARED; | |
2074 | } | |
2075 | } | |
2076 | ||
2077 | if (ia_valid & ATTR_ATIME) { | |
13442b03 | 2078 | dout("setattr %p atime %lld.%ld -> %lld.%ld\n", inode, |
9bbeab41 AB |
2079 | inode->i_atime.tv_sec, inode->i_atime.tv_nsec, |
2080 | attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec); | |
355da1eb SW |
2081 | if (issued & CEPH_CAP_FILE_EXCL) { |
2082 | ci->i_time_warp_seq++; | |
2083 | inode->i_atime = attr->ia_atime; | |
2084 | dirtied |= CEPH_CAP_FILE_EXCL; | |
2085 | } else if ((issued & CEPH_CAP_FILE_WR) && | |
95582b00 | 2086 | timespec64_compare(&inode->i_atime, |
355da1eb SW |
2087 | &attr->ia_atime) < 0) { |
2088 | inode->i_atime = attr->ia_atime; | |
2089 | dirtied |= CEPH_CAP_FILE_WR; | |
2090 | } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || | |
95582b00 | 2091 | !timespec64_equal(&inode->i_atime, &attr->ia_atime)) { |
9bbeab41 AB |
2092 | ceph_encode_timespec64(&req->r_args.setattr.atime, |
2093 | &attr->ia_atime); | |
355da1eb | 2094 | mask |= CEPH_SETATTR_ATIME; |
be70489e YZ |
2095 | release |= CEPH_CAP_FILE_SHARED | |
2096 | CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR; | |
355da1eb SW |
2097 | } |
2098 | } | |
2099 | if (ia_valid & ATTR_MTIME) { | |
13442b03 | 2100 | dout("setattr %p mtime %lld.%ld -> %lld.%ld\n", inode, |
9bbeab41 AB |
2101 | inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec, |
2102 | attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec); | |
355da1eb SW |
2103 | if (issued & CEPH_CAP_FILE_EXCL) { |
2104 | ci->i_time_warp_seq++; | |
2105 | inode->i_mtime = attr->ia_mtime; | |
2106 | dirtied |= CEPH_CAP_FILE_EXCL; | |
2107 | } else if ((issued & CEPH_CAP_FILE_WR) && | |
95582b00 | 2108 | timespec64_compare(&inode->i_mtime, |
355da1eb SW |
2109 | &attr->ia_mtime) < 0) { |
2110 | inode->i_mtime = attr->ia_mtime; | |
2111 | dirtied |= CEPH_CAP_FILE_WR; | |
2112 | } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || | |
95582b00 | 2113 | !timespec64_equal(&inode->i_mtime, &attr->ia_mtime)) { |
9bbeab41 AB |
2114 | ceph_encode_timespec64(&req->r_args.setattr.mtime, |
2115 | &attr->ia_mtime); | |
355da1eb | 2116 | mask |= CEPH_SETATTR_MTIME; |
be70489e YZ |
2117 | release |= CEPH_CAP_FILE_SHARED | |
2118 | CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR; | |
355da1eb SW |
2119 | } |
2120 | } | |
2121 | if (ia_valid & ATTR_SIZE) { | |
2122 | dout("setattr %p size %lld -> %lld\n", inode, | |
2123 | inode->i_size, attr->ia_size); | |
355da1eb SW |
2124 | if ((issued & CEPH_CAP_FILE_EXCL) && |
2125 | attr->ia_size > inode->i_size) { | |
99c88e69 | 2126 | i_size_write(inode, attr->ia_size); |
224a7542 | 2127 | inode->i_blocks = calc_inode_blocks(attr->ia_size); |
355da1eb SW |
2128 | ci->i_reported_size = attr->ia_size; |
2129 | dirtied |= CEPH_CAP_FILE_EXCL; | |
2130 | } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || | |
2131 | attr->ia_size != inode->i_size) { | |
2132 | req->r_args.setattr.size = cpu_to_le64(attr->ia_size); | |
2133 | req->r_args.setattr.old_size = | |
2134 | cpu_to_le64(inode->i_size); | |
2135 | mask |= CEPH_SETATTR_SIZE; | |
be70489e YZ |
2136 | release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_EXCL | |
2137 | CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR; | |
355da1eb SW |
2138 | } |
2139 | } | |
2140 | ||
2141 | /* these do nothing */ | |
2142 | if (ia_valid & ATTR_CTIME) { | |
2143 | bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME| | |
2144 | ATTR_MODE|ATTR_UID|ATTR_GID)) == 0; | |
13442b03 | 2145 | dout("setattr %p ctime %lld.%ld -> %lld.%ld (%s)\n", inode, |
9bbeab41 AB |
2146 | inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec, |
2147 | attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec, | |
355da1eb | 2148 | only ? "ctime only" : "ignored"); |
355da1eb SW |
2149 | if (only) { |
2150 | /* | |
2151 | * if kernel wants to dirty ctime but nothing else, | |
2152 | * we need to choose a cap to dirty under, or do | |
2153 | * a almost-no-op setattr | |
2154 | */ | |
2155 | if (issued & CEPH_CAP_AUTH_EXCL) | |
2156 | dirtied |= CEPH_CAP_AUTH_EXCL; | |
2157 | else if (issued & CEPH_CAP_FILE_EXCL) | |
2158 | dirtied |= CEPH_CAP_FILE_EXCL; | |
2159 | else if (issued & CEPH_CAP_XATTR_EXCL) | |
2160 | dirtied |= CEPH_CAP_XATTR_EXCL; | |
2161 | else | |
2162 | mask |= CEPH_SETATTR_CTIME; | |
2163 | } | |
2164 | } | |
2165 | if (ia_valid & ATTR_FILE) | |
2166 | dout("setattr %p ATTR_FILE ... hrm!\n", inode); | |
2167 | ||
2168 | if (dirtied) { | |
f66fd9f0 YZ |
2169 | inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied, |
2170 | &prealloc_cf); | |
4ca2fea6 | 2171 | inode->i_ctime = attr->ia_ctime; |
355da1eb SW |
2172 | } |
2173 | ||
2174 | release &= issued; | |
be655596 | 2175 | spin_unlock(&ci->i_ceph_lock); |
604d1b02 YZ |
2176 | if (lock_snap_rwsem) |
2177 | up_read(&mdsc->snap_rwsem); | |
355da1eb | 2178 | |
fca65b4a SW |
2179 | if (inode_dirty_flags) |
2180 | __mark_inode_dirty(inode, inode_dirty_flags); | |
2181 | ||
7221fe4c | 2182 | |
355da1eb | 2183 | if (mask) { |
70b666c3 SW |
2184 | req->r_inode = inode; |
2185 | ihold(inode); | |
355da1eb SW |
2186 | req->r_inode_drop = release; |
2187 | req->r_args.setattr.mask = cpu_to_le32(mask); | |
2188 | req->r_num_caps = 1; | |
0ed1e90a | 2189 | req->r_stamp = attr->ia_ctime; |
752c8bdc | 2190 | err = ceph_mdsc_do_request(mdsc, NULL, req); |
355da1eb SW |
2191 | } |
2192 | dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err, | |
2193 | ceph_cap_string(dirtied), mask); | |
2194 | ||
355da1eb | 2195 | ceph_mdsc_put_request(req); |
f66fd9f0 | 2196 | ceph_free_cap_flush(prealloc_cf); |
8179a101 YZ |
2197 | |
2198 | if (err >= 0 && (mask & CEPH_SETATTR_SIZE)) | |
2199 | __ceph_do_pending_vmtruncate(inode); | |
2200 | ||
355da1eb SW |
2201 | return err; |
2202 | } | |
2203 | ||
a26fecca AG |
2204 | /* |
2205 | * setattr | |
2206 | */ | |
2207 | int ceph_setattr(struct dentry *dentry, struct iattr *attr) | |
2208 | { | |
fd5472ed | 2209 | struct inode *inode = d_inode(dentry); |
36a4c72d | 2210 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); |
fd5472ed JK |
2211 | int err; |
2212 | ||
2213 | if (ceph_snap(inode) != CEPH_NOSNAP) | |
2214 | return -EROFS; | |
2215 | ||
31051c85 | 2216 | err = setattr_prepare(dentry, attr); |
fd5472ed JK |
2217 | if (err != 0) |
2218 | return err; | |
2219 | ||
36a4c72d CX |
2220 | if ((attr->ia_valid & ATTR_SIZE) && |
2221 | attr->ia_size > max(inode->i_size, fsc->max_file_size)) | |
2222 | return -EFBIG; | |
2223 | ||
2b83845f LH |
2224 | if ((attr->ia_valid & ATTR_SIZE) && |
2225 | ceph_quota_is_max_bytes_exceeded(inode, attr->ia_size)) | |
2226 | return -EDQUOT; | |
2227 | ||
8179a101 YZ |
2228 | err = __ceph_setattr(inode, attr); |
2229 | ||
2230 | if (err >= 0 && (attr->ia_valid & ATTR_MODE)) | |
2231 | err = posix_acl_chmod(inode, attr->ia_mode); | |
2232 | ||
2233 | return err; | |
a26fecca AG |
2234 | } |
2235 | ||
355da1eb SW |
2236 | /* |
2237 | * Verify that we have a lease on the given mask. If not, | |
2238 | * do a getattr against an mds. | |
2239 | */ | |
01deead0 YZ |
2240 | int __ceph_do_getattr(struct inode *inode, struct page *locked_page, |
2241 | int mask, bool force) | |
355da1eb | 2242 | { |
3d14c5d2 YS |
2243 | struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb); |
2244 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
355da1eb | 2245 | struct ceph_mds_request *req; |
49a9f4f6 | 2246 | int mode; |
355da1eb SW |
2247 | int err; |
2248 | ||
2249 | if (ceph_snap(inode) == CEPH_SNAPDIR) { | |
2250 | dout("do_getattr inode %p SNAPDIR\n", inode); | |
2251 | return 0; | |
2252 | } | |
2253 | ||
01deead0 YZ |
2254 | dout("do_getattr inode %p mask %s mode 0%o\n", |
2255 | inode, ceph_cap_string(mask), inode->i_mode); | |
508b32d8 | 2256 | if (!force && ceph_caps_issued_mask(ceph_inode(inode), mask, 1)) |
355da1eb SW |
2257 | return 0; |
2258 | ||
49a9f4f6 YZ |
2259 | mode = (mask & CEPH_STAT_RSTAT) ? USE_AUTH_MDS : USE_ANY_MDS; |
2260 | req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, mode); | |
355da1eb SW |
2261 | if (IS_ERR(req)) |
2262 | return PTR_ERR(req); | |
70b666c3 SW |
2263 | req->r_inode = inode; |
2264 | ihold(inode); | |
355da1eb SW |
2265 | req->r_num_caps = 1; |
2266 | req->r_args.getattr.mask = cpu_to_le32(mask); | |
01deead0 | 2267 | req->r_locked_page = locked_page; |
355da1eb | 2268 | err = ceph_mdsc_do_request(mdsc, NULL, req); |
01deead0 YZ |
2269 | if (locked_page && err == 0) { |
2270 | u64 inline_version = req->r_reply_info.targeti.inline_version; | |
2271 | if (inline_version == 0) { | |
2272 | /* the reply is supposed to contain inline data */ | |
2273 | err = -EINVAL; | |
2274 | } else if (inline_version == CEPH_INLINE_NONE) { | |
2275 | err = -ENODATA; | |
2276 | } else { | |
2277 | err = req->r_reply_info.targeti.inline_len; | |
2278 | } | |
2279 | } | |
355da1eb SW |
2280 | ceph_mdsc_put_request(req); |
2281 | dout("do_getattr result=%d\n", err); | |
2282 | return err; | |
2283 | } | |
2284 | ||
2285 | ||
2286 | /* | |
2287 | * Check inode permissions. We verify we have a valid value for | |
2288 | * the AUTH cap, then call the generic handler. | |
2289 | */ | |
10556cb2 | 2290 | int ceph_permission(struct inode *inode, int mask) |
355da1eb | 2291 | { |
b74c79e9 NP |
2292 | int err; |
2293 | ||
10556cb2 | 2294 | if (mask & MAY_NOT_BLOCK) |
b74c79e9 NP |
2295 | return -ECHILD; |
2296 | ||
508b32d8 | 2297 | err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED, false); |
355da1eb SW |
2298 | |
2299 | if (!err) | |
2830ba7f | 2300 | err = generic_permission(inode, mask); |
355da1eb SW |
2301 | return err; |
2302 | } | |
2303 | ||
428bb68a JL |
2304 | /* Craft a mask of needed caps given a set of requested statx attrs. */ |
2305 | static int statx_to_caps(u32 want) | |
2306 | { | |
2307 | int mask = 0; | |
2308 | ||
58981784 | 2309 | if (want & (STATX_MODE|STATX_UID|STATX_GID|STATX_CTIME|STATX_BTIME)) |
428bb68a JL |
2310 | mask |= CEPH_CAP_AUTH_SHARED; |
2311 | ||
2312 | if (want & (STATX_NLINK|STATX_CTIME)) | |
2313 | mask |= CEPH_CAP_LINK_SHARED; | |
2314 | ||
2315 | if (want & (STATX_ATIME|STATX_MTIME|STATX_CTIME|STATX_SIZE| | |
2316 | STATX_BLOCKS)) | |
2317 | mask |= CEPH_CAP_FILE_SHARED; | |
2318 | ||
2319 | if (want & (STATX_CTIME)) | |
2320 | mask |= CEPH_CAP_XATTR_SHARED; | |
2321 | ||
2322 | return mask; | |
2323 | } | |
2324 | ||
355da1eb | 2325 | /* |
428bb68a JL |
2326 | * Get all the attributes. If we have sufficient caps for the requested attrs, |
2327 | * then we can avoid talking to the MDS at all. | |
355da1eb | 2328 | */ |
a528d35e DH |
2329 | int ceph_getattr(const struct path *path, struct kstat *stat, |
2330 | u32 request_mask, unsigned int flags) | |
355da1eb | 2331 | { |
a528d35e | 2332 | struct inode *inode = d_inode(path->dentry); |
232d4b01 | 2333 | struct ceph_inode_info *ci = ceph_inode(inode); |
58981784 | 2334 | u32 valid_mask = STATX_BASIC_STATS; |
428bb68a | 2335 | int err = 0; |
355da1eb | 2336 | |
428bb68a JL |
2337 | /* Skip the getattr altogether if we're asked not to sync */ |
2338 | if (!(flags & AT_STATX_DONT_SYNC)) { | |
2339 | err = ceph_do_getattr(inode, statx_to_caps(request_mask), | |
2340 | flags & AT_STATX_FORCE_SYNC); | |
2341 | if (err) | |
2342 | return err; | |
2343 | } | |
2344 | ||
2345 | generic_fillattr(inode, stat); | |
2346 | stat->ino = ceph_translate_ino(inode->i_sb, inode->i_ino); | |
58981784 JL |
2347 | |
2348 | /* | |
2349 | * btime on newly-allocated inodes is 0, so if this is still set to | |
2350 | * that, then assume that it's not valid. | |
2351 | */ | |
2352 | if (ci->i_btime.tv_sec || ci->i_btime.tv_nsec) { | |
2353 | stat->btime = ci->i_btime; | |
2354 | valid_mask |= STATX_BTIME; | |
2355 | } | |
2356 | ||
428bb68a JL |
2357 | if (ceph_snap(inode) == CEPH_NOSNAP) |
2358 | stat->dev = inode->i_sb->s_dev; | |
2359 | else | |
2360 | stat->dev = ci->i_snapid_map ? ci->i_snapid_map->dev : 0; | |
2361 | ||
2362 | if (S_ISDIR(inode->i_mode)) { | |
2363 | if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb), | |
2364 | RBYTES)) | |
2365 | stat->size = ci->i_rbytes; | |
355da1eb | 2366 | else |
428bb68a JL |
2367 | stat->size = ci->i_files + ci->i_subdirs; |
2368 | stat->blocks = 0; | |
2369 | stat->blksize = 65536; | |
2370 | /* | |
2371 | * Some applications rely on the number of st_nlink | |
2372 | * value on directories to be either 0 (if unlinked) | |
2373 | * or 2 + number of subdirectories. | |
2374 | */ | |
2375 | if (stat->nlink == 1) | |
2376 | /* '.' + '..' + subdirs */ | |
2377 | stat->nlink = 1 + 1 + ci->i_subdirs; | |
355da1eb | 2378 | } |
428bb68a | 2379 | |
58981784 | 2380 | stat->result_mask = request_mask & valid_mask; |
355da1eb SW |
2381 | return err; |
2382 | } |