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 | ||
f4b97866 YZ |
518 | ci->i_meta_err = 0; |
519 | ||
355da1eb SW |
520 | return &ci->vfs_inode; |
521 | } | |
522 | ||
cfa6d412 | 523 | void ceph_free_inode(struct inode *inode) |
fa0d7e3d | 524 | { |
fa0d7e3d NP |
525 | struct ceph_inode_info *ci = ceph_inode(inode); |
526 | ||
daf5cc27 | 527 | kfree(ci->i_symlink); |
fa0d7e3d NP |
528 | kmem_cache_free(ceph_inode_cachep, ci); |
529 | } | |
530 | ||
87bc5b89 | 531 | void ceph_evict_inode(struct inode *inode) |
355da1eb SW |
532 | { |
533 | struct ceph_inode_info *ci = ceph_inode(inode); | |
534 | struct ceph_inode_frag *frag; | |
535 | struct rb_node *n; | |
536 | ||
87bc5b89 YZ |
537 | dout("evict_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode)); |
538 | ||
539 | truncate_inode_pages_final(&inode->i_data); | |
540 | clear_inode(inode); | |
355da1eb | 541 | |
99ccbd22 MT |
542 | ceph_fscache_unregister_inode_cookie(ci); |
543 | ||
d6e47819 | 544 | __ceph_remove_caps(ci); |
355da1eb | 545 | |
d557c48d LH |
546 | if (__ceph_has_any_quota(ci)) |
547 | ceph_adjust_quota_realms_count(inode, false); | |
548 | ||
8b218b8a SW |
549 | /* |
550 | * we may still have a snap_realm reference if there are stray | |
d9df2783 | 551 | * caps in i_snap_caps. |
8b218b8a SW |
552 | */ |
553 | if (ci->i_snap_realm) { | |
554 | struct ceph_mds_client *mdsc = | |
75c9627e YZ |
555 | ceph_inode_to_client(inode)->mdsc; |
556 | if (ceph_snap(inode) == CEPH_NOSNAP) { | |
557 | struct ceph_snap_realm *realm = ci->i_snap_realm; | |
558 | dout(" dropping residual ref to snap realm %p\n", | |
559 | realm); | |
560 | spin_lock(&realm->inodes_with_caps_lock); | |
561 | list_del_init(&ci->i_snap_realm_item); | |
562 | ci->i_snap_realm = NULL; | |
563 | if (realm->ino == ci->i_vino.ino) | |
564 | realm->inode = NULL; | |
565 | spin_unlock(&realm->inodes_with_caps_lock); | |
566 | ceph_put_snap_realm(mdsc, realm); | |
567 | } else { | |
568 | ceph_put_snapid_map(mdsc, ci->i_snapid_map); | |
569 | ci->i_snap_realm = NULL; | |
570 | } | |
8b218b8a SW |
571 | } |
572 | ||
355da1eb SW |
573 | while ((n = rb_first(&ci->i_fragtree)) != NULL) { |
574 | frag = rb_entry(n, struct ceph_inode_frag, node); | |
575 | rb_erase(n, &ci->i_fragtree); | |
576 | kfree(frag); | |
577 | } | |
1b1bc16d | 578 | ci->i_fragtree_nsplits = 0; |
355da1eb SW |
579 | |
580 | __ceph_destroy_xattrs(ci); | |
b6c1d5b8 SW |
581 | if (ci->i_xattrs.blob) |
582 | ceph_buffer_put(ci->i_xattrs.blob); | |
583 | if (ci->i_xattrs.prealloc_blob) | |
584 | ceph_buffer_put(ci->i_xattrs.prealloc_blob); | |
355da1eb | 585 | |
779fe0fb | 586 | ceph_put_string(rcu_dereference_raw(ci->i_layout.pool_ns)); |
355da1eb SW |
587 | } |
588 | ||
224a7542 YZ |
589 | static inline blkcnt_t calc_inode_blocks(u64 size) |
590 | { | |
591 | return (size + (1<<9) - 1) >> 9; | |
592 | } | |
593 | ||
355da1eb SW |
594 | /* |
595 | * Helpers to fill in size, ctime, mtime, and atime. We have to be | |
596 | * careful because either the client or MDS may have more up to date | |
597 | * info, depending on which capabilities are held, and whether | |
598 | * time_warp_seq or truncate_seq have increased. (Ordinarily, mtime | |
599 | * and size are monotonically increasing, except when utimes() or | |
600 | * truncate() increments the corresponding _seq values.) | |
601 | */ | |
602 | int ceph_fill_file_size(struct inode *inode, int issued, | |
603 | u32 truncate_seq, u64 truncate_size, u64 size) | |
604 | { | |
605 | struct ceph_inode_info *ci = ceph_inode(inode); | |
606 | int queue_trunc = 0; | |
607 | ||
608 | if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 || | |
609 | (truncate_seq == ci->i_truncate_seq && size > inode->i_size)) { | |
610 | dout("size %lld -> %llu\n", inode->i_size, size); | |
a3d714c3 YZ |
611 | if (size > 0 && S_ISDIR(inode->i_mode)) { |
612 | pr_err("fill_file_size non-zero size for directory\n"); | |
613 | size = 0; | |
614 | } | |
99c88e69 | 615 | i_size_write(inode, size); |
224a7542 | 616 | inode->i_blocks = calc_inode_blocks(size); |
355da1eb SW |
617 | ci->i_reported_size = size; |
618 | if (truncate_seq != ci->i_truncate_seq) { | |
619 | dout("truncate_seq %u -> %u\n", | |
620 | ci->i_truncate_seq, truncate_seq); | |
621 | ci->i_truncate_seq = truncate_seq; | |
b0d7c223 YZ |
622 | |
623 | /* the MDS should have revoked these caps */ | |
624 | WARN_ON_ONCE(issued & (CEPH_CAP_FILE_EXCL | | |
625 | CEPH_CAP_FILE_RD | | |
626 | CEPH_CAP_FILE_WR | | |
627 | CEPH_CAP_FILE_LAZYIO)); | |
3d497d85 YS |
628 | /* |
629 | * If we hold relevant caps, or in the case where we're | |
630 | * not the only client referencing this file and we | |
631 | * don't hold those caps, then we need to check whether | |
632 | * the file is either opened or mmaped | |
633 | */ | |
b0d7c223 YZ |
634 | if ((issued & (CEPH_CAP_FILE_CACHE| |
635 | CEPH_CAP_FILE_BUFFER)) || | |
3d497d85 YS |
636 | mapping_mapped(inode->i_mapping) || |
637 | __ceph_caps_file_wanted(ci)) { | |
355da1eb SW |
638 | ci->i_truncate_pending++; |
639 | queue_trunc = 1; | |
640 | } | |
641 | } | |
642 | } | |
643 | if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 && | |
644 | ci->i_truncate_size != truncate_size) { | |
645 | dout("truncate_size %lld -> %llu\n", ci->i_truncate_size, | |
646 | truncate_size); | |
647 | ci->i_truncate_size = truncate_size; | |
648 | } | |
99ccbd22 MT |
649 | |
650 | if (queue_trunc) | |
651 | ceph_fscache_invalidate(inode); | |
652 | ||
355da1eb SW |
653 | return queue_trunc; |
654 | } | |
655 | ||
656 | void ceph_fill_file_time(struct inode *inode, int issued, | |
9bbeab41 AB |
657 | u64 time_warp_seq, struct timespec64 *ctime, |
658 | struct timespec64 *mtime, struct timespec64 *atime) | |
355da1eb SW |
659 | { |
660 | struct ceph_inode_info *ci = ceph_inode(inode); | |
661 | int warn = 0; | |
662 | ||
663 | if (issued & (CEPH_CAP_FILE_EXCL| | |
664 | CEPH_CAP_FILE_WR| | |
d8672d64 SW |
665 | CEPH_CAP_FILE_BUFFER| |
666 | CEPH_CAP_AUTH_EXCL| | |
667 | CEPH_CAP_XATTR_EXCL)) { | |
ffdeec7a | 668 | if (ci->i_version == 0 || |
9bbeab41 | 669 | timespec64_compare(ctime, &inode->i_ctime) > 0) { |
13442b03 | 670 | dout("ctime %lld.%09ld -> %lld.%09ld inc w/ cap\n", |
9bbeab41 AB |
671 | inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec, |
672 | ctime->tv_sec, ctime->tv_nsec); | |
673 | inode->i_ctime = *ctime; | |
355da1eb | 674 | } |
ffdeec7a YZ |
675 | if (ci->i_version == 0 || |
676 | ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) { | |
355da1eb | 677 | /* the MDS did a utimes() */ |
13442b03 | 678 | dout("mtime %lld.%09ld -> %lld.%09ld " |
355da1eb | 679 | "tw %d -> %d\n", |
9bbeab41 AB |
680 | inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec, |
681 | mtime->tv_sec, mtime->tv_nsec, | |
355da1eb SW |
682 | ci->i_time_warp_seq, (int)time_warp_seq); |
683 | ||
9bbeab41 AB |
684 | inode->i_mtime = *mtime; |
685 | inode->i_atime = *atime; | |
355da1eb SW |
686 | ci->i_time_warp_seq = time_warp_seq; |
687 | } else if (time_warp_seq == ci->i_time_warp_seq) { | |
688 | /* nobody did utimes(); take the max */ | |
9bbeab41 | 689 | if (timespec64_compare(mtime, &inode->i_mtime) > 0) { |
13442b03 | 690 | dout("mtime %lld.%09ld -> %lld.%09ld inc\n", |
9bbeab41 | 691 | inode->i_mtime.tv_sec, |
355da1eb | 692 | inode->i_mtime.tv_nsec, |
9bbeab41 AB |
693 | mtime->tv_sec, mtime->tv_nsec); |
694 | inode->i_mtime = *mtime; | |
355da1eb | 695 | } |
9bbeab41 | 696 | if (timespec64_compare(atime, &inode->i_atime) > 0) { |
13442b03 | 697 | dout("atime %lld.%09ld -> %lld.%09ld inc\n", |
9bbeab41 | 698 | inode->i_atime.tv_sec, |
355da1eb | 699 | inode->i_atime.tv_nsec, |
9bbeab41 AB |
700 | atime->tv_sec, atime->tv_nsec); |
701 | inode->i_atime = *atime; | |
355da1eb SW |
702 | } |
703 | } else if (issued & CEPH_CAP_FILE_EXCL) { | |
704 | /* we did a utimes(); ignore mds values */ | |
705 | } else { | |
706 | warn = 1; | |
707 | } | |
708 | } else { | |
d8672d64 | 709 | /* we have no write|excl caps; whatever the MDS says is true */ |
355da1eb | 710 | if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) { |
9bbeab41 AB |
711 | inode->i_ctime = *ctime; |
712 | inode->i_mtime = *mtime; | |
713 | inode->i_atime = *atime; | |
355da1eb SW |
714 | ci->i_time_warp_seq = time_warp_seq; |
715 | } else { | |
716 | warn = 1; | |
717 | } | |
718 | } | |
719 | if (warn) /* time_warp_seq shouldn't go backwards */ | |
720 | dout("%p mds time_warp_seq %llu < %u\n", | |
721 | inode, time_warp_seq, ci->i_time_warp_seq); | |
722 | } | |
723 | ||
724 | /* | |
725 | * Populate an inode based on info from mds. May be called on new or | |
726 | * existing inodes. | |
727 | */ | |
01deead0 | 728 | static int fill_inode(struct inode *inode, struct page *locked_page, |
355da1eb SW |
729 | struct ceph_mds_reply_info_in *iinfo, |
730 | struct ceph_mds_reply_dirfrag *dirinfo, | |
731 | struct ceph_mds_session *session, | |
732 | unsigned long ttl_from, int cap_fmode, | |
733 | struct ceph_cap_reservation *caps_reservation) | |
734 | { | |
d9df2783 | 735 | struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc; |
355da1eb SW |
736 | struct ceph_mds_reply_inode *info = iinfo->in; |
737 | struct ceph_inode_info *ci = ceph_inode(inode); | |
2af54a72 | 738 | int issued, new_issued, info_caps; |
9bbeab41 | 739 | struct timespec64 mtime, atime, ctime; |
355da1eb | 740 | struct ceph_buffer *xattr_blob = NULL; |
af8a85a4 | 741 | struct ceph_buffer *old_blob = NULL; |
779fe0fb | 742 | struct ceph_string *pool_ns = NULL; |
d9df2783 | 743 | struct ceph_cap *new_cap = NULL; |
355da1eb | 744 | int err = 0; |
d9df2783 | 745 | bool wake = false; |
f98a128a YZ |
746 | bool queue_trunc = false; |
747 | bool new_version = false; | |
31c542a1 | 748 | bool fill_inline = false; |
355da1eb SW |
749 | |
750 | dout("fill_inode %p ino %llx.%llx v %llu had %llu\n", | |
751 | inode, ceph_vinop(inode), le64_to_cpu(info->version), | |
752 | ci->i_version); | |
753 | ||
2af54a72 YZ |
754 | info_caps = le32_to_cpu(info->cap.caps); |
755 | ||
d9df2783 | 756 | /* prealloc new cap struct */ |
2af54a72 | 757 | if (info_caps && ceph_snap(inode) == CEPH_NOSNAP) |
d9df2783 YZ |
758 | new_cap = ceph_get_cap(mdsc, caps_reservation); |
759 | ||
355da1eb SW |
760 | /* |
761 | * prealloc xattr data, if it looks like we'll need it. only | |
762 | * if len > 4 (meaning there are actually xattrs; the first 4 | |
763 | * bytes are the xattr count). | |
764 | */ | |
765 | if (iinfo->xattr_len > 4) { | |
b6c1d5b8 | 766 | xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS); |
355da1eb SW |
767 | if (!xattr_blob) |
768 | pr_err("fill_inode ENOMEM xattr blob %d bytes\n", | |
769 | iinfo->xattr_len); | |
770 | } | |
771 | ||
779fe0fb YZ |
772 | if (iinfo->pool_ns_len > 0) |
773 | pool_ns = ceph_find_or_create_string(iinfo->pool_ns_data, | |
774 | iinfo->pool_ns_len); | |
775 | ||
75c9627e YZ |
776 | if (ceph_snap(inode) != CEPH_NOSNAP && !ci->i_snapid_map) |
777 | ci->i_snapid_map = ceph_get_snapid_map(mdsc, ceph_snap(inode)); | |
778 | ||
be655596 | 779 | spin_lock(&ci->i_ceph_lock); |
355da1eb SW |
780 | |
781 | /* | |
782 | * provided version will be odd if inode value is projected, | |
8bd59e01 SW |
783 | * even if stable. skip the update if we have newer stable |
784 | * info (ours>=theirs, e.g. due to racing mds replies), unless | |
785 | * we are getting projected (unstable) info (in which case the | |
786 | * version is odd, and we want ours>theirs). | |
787 | * us them | |
788 | * 2 2 skip | |
789 | * 3 2 skip | |
790 | * 3 3 update | |
355da1eb | 791 | */ |
f98a128a YZ |
792 | if (ci->i_version == 0 || |
793 | ((info->cap.flags & CEPH_CAP_FLAG_AUTH) && | |
794 | le64_to_cpu(info->version) > (ci->i_version & ~1))) | |
795 | new_version = true; | |
796 | ||
a35ead31 JL |
797 | /* Update change_attribute */ |
798 | inode_set_max_iversion_raw(inode, iinfo->change_attr); | |
799 | ||
2af54a72 YZ |
800 | __ceph_caps_issued(ci, &issued); |
801 | issued |= __ceph_caps_dirty(ci); | |
802 | new_issued = ~issued & info_caps; | |
355da1eb SW |
803 | |
804 | /* update inode */ | |
355da1eb | 805 | inode->i_rdev = le32_to_cpu(info->rdev); |
75067034 LH |
806 | /* directories have fl_stripe_unit set to zero */ |
807 | if (le32_to_cpu(info->layout.fl_stripe_unit)) | |
808 | inode->i_blkbits = | |
809 | fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1; | |
810 | else | |
811 | inode->i_blkbits = CEPH_BLOCK_SHIFT; | |
355da1eb | 812 | |
d557c48d | 813 | __ceph_update_quota(ci, iinfo->max_bytes, iinfo->max_files); |
fb18a575 | 814 | |
f98a128a YZ |
815 | if ((new_version || (new_issued & CEPH_CAP_AUTH_SHARED)) && |
816 | (issued & CEPH_CAP_AUTH_EXCL) == 0) { | |
355da1eb | 817 | inode->i_mode = le32_to_cpu(info->mode); |
ab871b90 EB |
818 | inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(info->uid)); |
819 | inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(info->gid)); | |
355da1eb | 820 | dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode, |
bd2bae6a EB |
821 | from_kuid(&init_user_ns, inode->i_uid), |
822 | from_kgid(&init_user_ns, inode->i_gid)); | |
245ce991 | 823 | ceph_decode_timespec64(&ci->i_btime, &iinfo->btime); |
193e7b37 | 824 | ceph_decode_timespec64(&ci->i_snap_btime, &iinfo->snap_btime); |
355da1eb SW |
825 | } |
826 | ||
f98a128a YZ |
827 | if ((new_version || (new_issued & CEPH_CAP_LINK_SHARED)) && |
828 | (issued & CEPH_CAP_LINK_EXCL) == 0) | |
bfe86848 | 829 | set_nlink(inode, le32_to_cpu(info->nlink)); |
355da1eb | 830 | |
f98a128a YZ |
831 | if (new_version || (new_issued & CEPH_CAP_ANY_RD)) { |
832 | /* be careful with mtime, atime, size */ | |
9bbeab41 AB |
833 | ceph_decode_timespec64(&atime, &info->atime); |
834 | ceph_decode_timespec64(&mtime, &info->mtime); | |
835 | ceph_decode_timespec64(&ctime, &info->ctime); | |
f98a128a YZ |
836 | ceph_fill_file_time(inode, issued, |
837 | le32_to_cpu(info->time_warp_seq), | |
838 | &ctime, &mtime, &atime); | |
839 | } | |
840 | ||
2af54a72 YZ |
841 | if (new_version || (info_caps & CEPH_CAP_FILE_SHARED)) { |
842 | ci->i_files = le64_to_cpu(info->files); | |
843 | ci->i_subdirs = le64_to_cpu(info->subdirs); | |
844 | } | |
845 | ||
f98a128a YZ |
846 | if (new_version || |
847 | (new_issued & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR))) { | |
7627151e | 848 | s64 old_pool = ci->i_layout.pool_id; |
779fe0fb YZ |
849 | struct ceph_string *old_ns; |
850 | ||
7627151e | 851 | ceph_file_layout_from_legacy(&ci->i_layout, &info->layout); |
779fe0fb YZ |
852 | old_ns = rcu_dereference_protected(ci->i_layout.pool_ns, |
853 | lockdep_is_held(&ci->i_ceph_lock)); | |
854 | rcu_assign_pointer(ci->i_layout.pool_ns, pool_ns); | |
855 | ||
856 | if (ci->i_layout.pool_id != old_pool || pool_ns != old_ns) | |
10183a69 | 857 | ci->i_ceph_flags &= ~CEPH_I_POOL_PERM; |
10183a69 | 858 | |
779fe0fb | 859 | pool_ns = old_ns; |
10183a69 | 860 | |
f98a128a YZ |
861 | queue_trunc = ceph_fill_file_size(inode, issued, |
862 | le32_to_cpu(info->truncate_seq), | |
863 | le64_to_cpu(info->truncate_size), | |
864 | le64_to_cpu(info->size)); | |
865 | /* only update max_size on auth cap */ | |
866 | if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) && | |
867 | ci->i_max_size != le64_to_cpu(info->max_size)) { | |
868 | dout("max_size %lld -> %llu\n", ci->i_max_size, | |
869 | le64_to_cpu(info->max_size)); | |
870 | ci->i_max_size = le64_to_cpu(info->max_size); | |
871 | } | |
872 | } | |
355da1eb | 873 | |
49a9f4f6 YZ |
874 | /* layout and rstat are not tracked by capability, update them if |
875 | * the inode info is from auth mds */ | |
876 | if (new_version || (info->cap.flags & CEPH_CAP_FLAG_AUTH)) { | |
877 | if (S_ISDIR(inode->i_mode)) { | |
878 | ci->i_dir_layout = iinfo->dir_layout; | |
879 | ci->i_rbytes = le64_to_cpu(info->rbytes); | |
880 | ci->i_rfiles = le64_to_cpu(info->rfiles); | |
881 | ci->i_rsubdirs = le64_to_cpu(info->rsubdirs); | |
08796873 | 882 | ci->i_dir_pin = iinfo->dir_pin; |
9bbeab41 | 883 | ceph_decode_timespec64(&ci->i_rctime, &info->rctime); |
49a9f4f6 YZ |
884 | } |
885 | } | |
886 | ||
355da1eb SW |
887 | /* xattrs */ |
888 | /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */ | |
508b32d8 | 889 | if ((ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL)) && |
355da1eb SW |
890 | le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) { |
891 | if (ci->i_xattrs.blob) | |
af8a85a4 | 892 | old_blob = ci->i_xattrs.blob; |
355da1eb SW |
893 | ci->i_xattrs.blob = xattr_blob; |
894 | if (xattr_blob) | |
895 | memcpy(ci->i_xattrs.blob->vec.iov_base, | |
896 | iinfo->xattr_data, iinfo->xattr_len); | |
897 | ci->i_xattrs.version = le64_to_cpu(info->xattr_version); | |
7221fe4c | 898 | ceph_forget_all_cached_acls(inode); |
ac6713cc | 899 | ceph_security_invalidate_secctx(inode); |
a6424e48 | 900 | xattr_blob = NULL; |
355da1eb SW |
901 | } |
902 | ||
ffdeec7a | 903 | /* finally update i_version */ |
aae1a442 YZ |
904 | if (le64_to_cpu(info->version) > ci->i_version) |
905 | ci->i_version = le64_to_cpu(info->version); | |
ffdeec7a | 906 | |
355da1eb | 907 | inode->i_mapping->a_ops = &ceph_aops; |
355da1eb SW |
908 | |
909 | switch (inode->i_mode & S_IFMT) { | |
910 | case S_IFIFO: | |
911 | case S_IFBLK: | |
912 | case S_IFCHR: | |
913 | case S_IFSOCK: | |
5ba72e60 | 914 | inode->i_blkbits = PAGE_SHIFT; |
355da1eb SW |
915 | init_special_inode(inode, inode->i_mode, inode->i_rdev); |
916 | inode->i_op = &ceph_file_iops; | |
917 | break; | |
918 | case S_IFREG: | |
919 | inode->i_op = &ceph_file_iops; | |
920 | inode->i_fop = &ceph_file_fops; | |
921 | break; | |
922 | case S_IFLNK: | |
923 | inode->i_op = &ceph_symlink_iops; | |
924 | if (!ci->i_symlink) { | |
810339ec | 925 | u32 symlen = iinfo->symlink_len; |
355da1eb SW |
926 | char *sym; |
927 | ||
be655596 | 928 | spin_unlock(&ci->i_ceph_lock); |
355da1eb | 929 | |
224a7542 YZ |
930 | if (symlen != i_size_read(inode)) { |
931 | pr_err("fill_inode %llx.%llx BAD symlink " | |
932 | "size %lld\n", ceph_vinop(inode), | |
933 | i_size_read(inode)); | |
934 | i_size_write(inode, symlen); | |
935 | inode->i_blocks = calc_inode_blocks(symlen); | |
936 | } | |
810339ec | 937 | |
355da1eb | 938 | err = -ENOMEM; |
810339ec | 939 | sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS); |
355da1eb SW |
940 | if (!sym) |
941 | goto out; | |
355da1eb | 942 | |
be655596 | 943 | spin_lock(&ci->i_ceph_lock); |
355da1eb SW |
944 | if (!ci->i_symlink) |
945 | ci->i_symlink = sym; | |
946 | else | |
947 | kfree(sym); /* lost a race */ | |
948 | } | |
ac194dcc | 949 | inode->i_link = ci->i_symlink; |
355da1eb SW |
950 | break; |
951 | case S_IFDIR: | |
952 | inode->i_op = &ceph_dir_iops; | |
953 | inode->i_fop = &ceph_dir_fops; | |
355da1eb SW |
954 | break; |
955 | default: | |
956 | pr_err("fill_inode %llx.%llx BAD mode 0%o\n", | |
957 | ceph_vinop(inode), inode->i_mode); | |
958 | } | |
959 | ||
355da1eb | 960 | /* were we issued a capability? */ |
2af54a72 | 961 | if (info_caps) { |
355da1eb SW |
962 | if (ceph_snap(inode) == CEPH_NOSNAP) { |
963 | ceph_add_cap(inode, session, | |
964 | le64_to_cpu(info->cap.cap_id), | |
2af54a72 | 965 | cap_fmode, info_caps, |
355da1eb SW |
966 | le32_to_cpu(info->cap.wanted), |
967 | le32_to_cpu(info->cap.seq), | |
968 | le32_to_cpu(info->cap.mseq), | |
969 | le64_to_cpu(info->cap.realm), | |
d9df2783 | 970 | info->cap.flags, &new_cap); |
2f92b3d0 YZ |
971 | |
972 | /* set dir completion flag? */ | |
973 | if (S_ISDIR(inode->i_mode) && | |
974 | ci->i_files == 0 && ci->i_subdirs == 0 && | |
2af54a72 | 975 | (info_caps & CEPH_CAP_FILE_SHARED) && |
2f92b3d0 YZ |
976 | (issued & CEPH_CAP_FILE_EXCL) == 0 && |
977 | !__ceph_dir_is_complete(ci)) { | |
978 | dout(" marking %p complete (empty)\n", inode); | |
fdd4e158 | 979 | i_size_write(inode, 0); |
2f92b3d0 | 980 | __ceph_dir_set_complete(ci, |
fdd4e158 YZ |
981 | atomic64_read(&ci->i_release_count), |
982 | atomic64_read(&ci->i_ordered_count)); | |
2f92b3d0 YZ |
983 | } |
984 | ||
d9df2783 | 985 | wake = true; |
355da1eb | 986 | } else { |
355da1eb | 987 | dout(" %p got snap_caps %s\n", inode, |
2af54a72 YZ |
988 | ceph_cap_string(info_caps)); |
989 | ci->i_snap_caps |= info_caps; | |
355da1eb SW |
990 | if (cap_fmode >= 0) |
991 | __ceph_get_fmode(ci, cap_fmode); | |
355da1eb | 992 | } |
04d000eb | 993 | } else if (cap_fmode >= 0) { |
f3ae1b97 | 994 | pr_warn("mds issued no caps on %llx.%llx\n", |
04d000eb SW |
995 | ceph_vinop(inode)); |
996 | __ceph_get_fmode(ci, cap_fmode); | |
355da1eb | 997 | } |
31c542a1 YZ |
998 | |
999 | if (iinfo->inline_version > 0 && | |
1000 | iinfo->inline_version >= ci->i_inline_version) { | |
1001 | int cache_caps = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO; | |
1002 | ci->i_inline_version = iinfo->inline_version; | |
1003 | if (ci->i_inline_version != CEPH_INLINE_NONE && | |
2af54a72 | 1004 | (locked_page || (info_caps & cache_caps))) |
31c542a1 YZ |
1005 | fill_inline = true; |
1006 | } | |
1007 | ||
be655596 | 1008 | spin_unlock(&ci->i_ceph_lock); |
355da1eb | 1009 | |
31c542a1 | 1010 | if (fill_inline) |
01deead0 | 1011 | ceph_fill_inline_data(inode, locked_page, |
31c542a1 YZ |
1012 | iinfo->inline_data, iinfo->inline_len); |
1013 | ||
d9df2783 YZ |
1014 | if (wake) |
1015 | wake_up_all(&ci->i_cap_wq); | |
1016 | ||
355da1eb SW |
1017 | /* queue truncate if we saw i_size decrease */ |
1018 | if (queue_trunc) | |
3c6f6b79 | 1019 | ceph_queue_vmtruncate(inode); |
355da1eb SW |
1020 | |
1021 | /* populate frag tree */ | |
3e7fbe9c YZ |
1022 | if (S_ISDIR(inode->i_mode)) |
1023 | ceph_fill_fragtree(inode, &info->fragtree, dirinfo); | |
355da1eb SW |
1024 | |
1025 | /* update delegation info? */ | |
1026 | if (dirinfo) | |
1027 | ceph_fill_dirfrag(inode, dirinfo); | |
1028 | ||
1029 | err = 0; | |
355da1eb | 1030 | out: |
d9df2783 YZ |
1031 | if (new_cap) |
1032 | ceph_put_cap(mdsc, new_cap); | |
af8a85a4 LH |
1033 | ceph_buffer_put(old_blob); |
1034 | ceph_buffer_put(xattr_blob); | |
779fe0fb | 1035 | ceph_put_string(pool_ns); |
355da1eb SW |
1036 | return err; |
1037 | } | |
1038 | ||
1039 | /* | |
543212b3 | 1040 | * caller should hold session s_mutex and dentry->d_lock. |
355da1eb | 1041 | */ |
543212b3 YZ |
1042 | static void __update_dentry_lease(struct inode *dir, struct dentry *dentry, |
1043 | struct ceph_mds_reply_lease *lease, | |
1044 | struct ceph_mds_session *session, | |
1045 | unsigned long from_time, | |
1046 | struct ceph_mds_session **old_lease_session) | |
355da1eb SW |
1047 | { |
1048 | struct ceph_dentry_info *di = ceph_dentry(dentry); | |
1049 | long unsigned duration = le32_to_cpu(lease->duration_ms); | |
1050 | long unsigned ttl = from_time + (duration * HZ) / 1000; | |
1051 | long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000; | |
f5d55f03 | 1052 | |
2f90b852 SW |
1053 | dout("update_dentry_lease %p duration %lu ms ttl %lu\n", |
1054 | dentry, duration, ttl); | |
355da1eb | 1055 | |
18fc8abd AV |
1056 | /* only track leases on regular dentries */ |
1057 | if (ceph_snap(dir) != CEPH_NOSNAP) | |
543212b3 | 1058 | return; |
18fc8abd | 1059 | |
97aeb6bf | 1060 | di->lease_shared_gen = atomic_read(&ceph_inode(dir)->i_shared_gen); |
37c4efc1 YZ |
1061 | if (duration == 0) { |
1062 | __ceph_dentry_dir_lease_touch(di); | |
543212b3 | 1063 | return; |
37c4efc1 | 1064 | } |
355da1eb SW |
1065 | |
1066 | if (di->lease_gen == session->s_cap_gen && | |
9b16f03c | 1067 | time_before(ttl, di->time)) |
543212b3 | 1068 | return; /* we already have a newer lease. */ |
355da1eb | 1069 | |
481f001f | 1070 | if (di->lease_session && di->lease_session != session) { |
543212b3 | 1071 | *old_lease_session = di->lease_session; |
481f001f YZ |
1072 | di->lease_session = NULL; |
1073 | } | |
355da1eb | 1074 | |
355da1eb SW |
1075 | if (!di->lease_session) |
1076 | di->lease_session = ceph_get_mds_session(session); | |
1077 | di->lease_gen = session->s_cap_gen; | |
1078 | di->lease_seq = le32_to_cpu(lease->seq); | |
1079 | di->lease_renew_after = half_ttl; | |
1080 | di->lease_renew_from = 0; | |
9b16f03c | 1081 | di->time = ttl; |
37c4efc1 YZ |
1082 | |
1083 | __ceph_dentry_lease_touch(di); | |
543212b3 YZ |
1084 | } |
1085 | ||
1086 | static inline void update_dentry_lease(struct inode *dir, struct dentry *dentry, | |
1087 | struct ceph_mds_reply_lease *lease, | |
1088 | struct ceph_mds_session *session, | |
1089 | unsigned long from_time) | |
1090 | { | |
1091 | struct ceph_mds_session *old_lease_session = NULL; | |
1092 | spin_lock(&dentry->d_lock); | |
1093 | __update_dentry_lease(dir, dentry, lease, session, from_time, | |
1094 | &old_lease_session); | |
1095 | spin_unlock(&dentry->d_lock); | |
1096 | if (old_lease_session) | |
1097 | ceph_put_mds_session(old_lease_session); | |
1098 | } | |
1099 | ||
1100 | /* | |
1101 | * update dentry lease without having parent inode locked | |
1102 | */ | |
1103 | static void update_dentry_lease_careful(struct dentry *dentry, | |
1104 | struct ceph_mds_reply_lease *lease, | |
1105 | struct ceph_mds_session *session, | |
1106 | unsigned long from_time, | |
1107 | char *dname, u32 dname_len, | |
1108 | struct ceph_vino *pdvino, | |
1109 | struct ceph_vino *ptvino) | |
1110 | ||
1111 | { | |
1112 | struct inode *dir; | |
1113 | struct ceph_mds_session *old_lease_session = NULL; | |
1114 | ||
1115 | spin_lock(&dentry->d_lock); | |
1116 | /* make sure dentry's name matches target */ | |
1117 | if (dentry->d_name.len != dname_len || | |
1118 | memcmp(dentry->d_name.name, dname, dname_len)) | |
1119 | goto out_unlock; | |
1120 | ||
1121 | dir = d_inode(dentry->d_parent); | |
1122 | /* make sure parent matches dvino */ | |
1123 | if (!ceph_ino_compare(dir, pdvino)) | |
1124 | goto out_unlock; | |
1125 | ||
1126 | /* make sure dentry's inode matches target. NULL ptvino means that | |
1127 | * we expect a negative dentry */ | |
1128 | if (ptvino) { | |
1129 | if (d_really_is_negative(dentry)) | |
1130 | goto out_unlock; | |
1131 | if (!ceph_ino_compare(d_inode(dentry), ptvino)) | |
1132 | goto out_unlock; | |
1133 | } else { | |
1134 | if (d_really_is_positive(dentry)) | |
1135 | goto out_unlock; | |
1136 | } | |
1137 | ||
1138 | __update_dentry_lease(dir, dentry, lease, session, | |
1139 | from_time, &old_lease_session); | |
355da1eb SW |
1140 | out_unlock: |
1141 | spin_unlock(&dentry->d_lock); | |
481f001f YZ |
1142 | if (old_lease_session) |
1143 | ceph_put_mds_session(old_lease_session); | |
355da1eb SW |
1144 | } |
1145 | ||
1146 | /* | |
1147 | * splice a dentry to an inode. | |
1148 | * caller must hold directory i_mutex for this to be safe. | |
355da1eb | 1149 | */ |
2bf996ac | 1150 | static int splice_dentry(struct dentry **pdn, struct inode *in) |
355da1eb | 1151 | { |
2bf996ac | 1152 | struct dentry *dn = *pdn; |
355da1eb SW |
1153 | struct dentry *realdn; |
1154 | ||
2b0143b5 | 1155 | BUG_ON(d_inode(dn)); |
1cd3935b | 1156 | |
5495c2d0 YZ |
1157 | if (S_ISDIR(in->i_mode)) { |
1158 | /* If inode is directory, d_splice_alias() below will remove | |
1159 | * 'realdn' from its origin parent. We need to ensure that | |
1160 | * origin parent's readdir cache will not reference 'realdn' | |
1161 | */ | |
1162 | realdn = d_find_any_alias(in); | |
1163 | if (realdn) { | |
1164 | struct ceph_dentry_info *di = ceph_dentry(realdn); | |
1165 | spin_lock(&realdn->d_lock); | |
1166 | ||
1167 | realdn->d_op->d_prune(realdn); | |
1168 | ||
1169 | di->time = jiffies; | |
1170 | di->lease_shared_gen = 0; | |
1171 | di->offset = 0; | |
1172 | ||
1173 | spin_unlock(&realdn->d_lock); | |
1174 | dput(realdn); | |
1175 | } | |
1176 | } | |
1177 | ||
355da1eb SW |
1178 | /* dn must be unhashed */ |
1179 | if (!d_unhashed(dn)) | |
1180 | d_drop(dn); | |
41d28bca | 1181 | realdn = d_splice_alias(in, dn); |
355da1eb | 1182 | if (IS_ERR(realdn)) { |
d69ed05a SW |
1183 | pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n", |
1184 | PTR_ERR(realdn), dn, in, ceph_vinop(in)); | |
2bf996ac YZ |
1185 | return PTR_ERR(realdn); |
1186 | } | |
1187 | ||
1188 | if (realdn) { | |
355da1eb SW |
1189 | dout("dn %p (%d) spliced with %p (%d) " |
1190 | "inode %p ino %llx.%llx\n", | |
84d08fa8 AV |
1191 | dn, d_count(dn), |
1192 | realdn, d_count(realdn), | |
2b0143b5 | 1193 | d_inode(realdn), ceph_vinop(d_inode(realdn))); |
355da1eb | 1194 | dput(dn); |
2bf996ac | 1195 | *pdn = realdn; |
355da1eb SW |
1196 | } else { |
1197 | BUG_ON(!ceph_dentry(dn)); | |
355da1eb | 1198 | dout("dn %p attached to %p ino %llx.%llx\n", |
2b0143b5 | 1199 | dn, d_inode(dn), ceph_vinop(d_inode(dn))); |
355da1eb | 1200 | } |
2bf996ac | 1201 | return 0; |
355da1eb SW |
1202 | } |
1203 | ||
1204 | /* | |
1205 | * Incorporate results into the local cache. This is either just | |
1206 | * one inode, or a directory, dentry, and possibly linked-to inode (e.g., | |
1207 | * after a lookup). | |
1208 | * | |
1209 | * A reply may contain | |
1210 | * a directory inode along with a dentry. | |
1211 | * and/or a target inode | |
1212 | * | |
1213 | * Called with snap_rwsem (read). | |
1214 | */ | |
f5a03b08 | 1215 | int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req) |
355da1eb | 1216 | { |
f5a03b08 | 1217 | struct ceph_mds_session *session = req->r_session; |
355da1eb SW |
1218 | struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info; |
1219 | struct inode *in = NULL; | |
f5d55f03 | 1220 | struct ceph_vino tvino, dvino; |
3d14c5d2 | 1221 | struct ceph_fs_client *fsc = ceph_sb_to_client(sb); |
355da1eb SW |
1222 | int err = 0; |
1223 | ||
1224 | dout("fill_trace %p is_dentry %d is_target %d\n", req, | |
1225 | rinfo->head->is_dentry, rinfo->head->is_target); | |
1226 | ||
355da1eb SW |
1227 | if (!rinfo->head->is_target && !rinfo->head->is_dentry) { |
1228 | dout("fill_trace reply is empty!\n"); | |
3dd69aab | 1229 | if (rinfo->head->result == 0 && req->r_parent) |
167c9e35 | 1230 | ceph_invalidate_dir_request(req); |
355da1eb SW |
1231 | return 0; |
1232 | } | |
1233 | ||
1234 | if (rinfo->head->is_dentry) { | |
3dd69aab | 1235 | struct inode *dir = req->r_parent; |
5b1daecd | 1236 | |
6c5e50fa | 1237 | if (dir) { |
01deead0 YZ |
1238 | err = fill_inode(dir, NULL, |
1239 | &rinfo->diri, rinfo->dirfrag, | |
6c5e50fa SW |
1240 | session, req->r_request_started, -1, |
1241 | &req->r_caps_reservation); | |
1242 | if (err < 0) | |
19913b4e | 1243 | goto done; |
6c5e50fa SW |
1244 | } else { |
1245 | WARN_ON_ONCE(1); | |
1246 | } | |
19913b4e | 1247 | |
74c9e6bf YZ |
1248 | if (dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME && |
1249 | test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) && | |
1250 | !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) { | |
19913b4e YZ |
1251 | struct qstr dname; |
1252 | struct dentry *dn, *parent; | |
1253 | ||
1254 | BUG_ON(!rinfo->head->is_target); | |
1255 | BUG_ON(req->r_dentry); | |
1256 | ||
1257 | parent = d_find_any_alias(dir); | |
1258 | BUG_ON(!parent); | |
1259 | ||
1260 | dname.name = rinfo->dname; | |
1261 | dname.len = rinfo->dname_len; | |
8387ff25 | 1262 | dname.hash = full_name_hash(parent, dname.name, dname.len); |
f5d55f03 JL |
1263 | tvino.ino = le64_to_cpu(rinfo->targeti.in->ino); |
1264 | tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid); | |
19913b4e YZ |
1265 | retry_lookup: |
1266 | dn = d_lookup(parent, &dname); | |
1267 | dout("d_lookup on parent=%p name=%.*s got %p\n", | |
1268 | parent, dname.len, dname.name, dn); | |
1269 | ||
1270 | if (!dn) { | |
1271 | dn = d_alloc(parent, &dname); | |
1272 | dout("d_alloc %p '%.*s' = %p\n", parent, | |
1273 | dname.len, dname.name, dn); | |
d37b1d99 | 1274 | if (!dn) { |
19913b4e YZ |
1275 | dput(parent); |
1276 | err = -ENOMEM; | |
1277 | goto done; | |
1278 | } | |
ad5cb123 | 1279 | err = 0; |
2b0143b5 | 1280 | } else if (d_really_is_positive(dn) && |
f5d55f03 JL |
1281 | (ceph_ino(d_inode(dn)) != tvino.ino || |
1282 | ceph_snap(d_inode(dn)) != tvino.snap)) { | |
19913b4e | 1283 | dout(" dn %p points to wrong inode %p\n", |
2b0143b5 | 1284 | dn, d_inode(dn)); |
933ad2c9 | 1285 | ceph_dir_clear_ordered(dir); |
19913b4e YZ |
1286 | d_delete(dn); |
1287 | dput(dn); | |
1288 | goto retry_lookup; | |
1289 | } | |
1290 | ||
1291 | req->r_dentry = dn; | |
1292 | dput(parent); | |
1293 | } | |
5b1daecd SW |
1294 | } |
1295 | ||
86b58d13 | 1296 | if (rinfo->head->is_target) { |
f5d55f03 JL |
1297 | tvino.ino = le64_to_cpu(rinfo->targeti.in->ino); |
1298 | tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid); | |
86b58d13 | 1299 | |
f5d55f03 | 1300 | in = ceph_get_inode(sb, tvino); |
86b58d13 YZ |
1301 | if (IS_ERR(in)) { |
1302 | err = PTR_ERR(in); | |
1303 | goto done; | |
1304 | } | |
1305 | req->r_target_inode = in; | |
1306 | ||
01deead0 | 1307 | err = fill_inode(in, req->r_locked_page, &rinfo->targeti, NULL, |
86b58d13 | 1308 | session, req->r_request_started, |
bc2de10d JL |
1309 | (!test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) && |
1310 | rinfo->head->result == 0) ? req->r_fmode : -1, | |
86b58d13 YZ |
1311 | &req->r_caps_reservation); |
1312 | if (err < 0) { | |
1313 | pr_err("fill_inode badness %p %llx.%llx\n", | |
1314 | in, ceph_vinop(in)); | |
1315 | goto done; | |
1316 | } | |
1317 | } | |
1318 | ||
9358c6d4 SW |
1319 | /* |
1320 | * ignore null lease/binding on snapdir ENOENT, or else we | |
1321 | * will have trouble splicing in the virtual snapdir later | |
1322 | */ | |
3dd69aab JL |
1323 | if (rinfo->head->is_dentry && |
1324 | !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) && | |
1325 | test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) && | |
9358c6d4 | 1326 | (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name, |
3d14c5d2 | 1327 | fsc->mount_options->snapdir_name, |
9358c6d4 | 1328 | req->r_dentry->d_name.len))) { |
355da1eb SW |
1329 | /* |
1330 | * lookup link rename : null -> possibly existing inode | |
1331 | * mknod symlink mkdir : null -> new inode | |
1332 | * unlink : linked -> null | |
1333 | */ | |
3dd69aab | 1334 | struct inode *dir = req->r_parent; |
355da1eb SW |
1335 | struct dentry *dn = req->r_dentry; |
1336 | bool have_dir_cap, have_lease; | |
1337 | ||
1338 | BUG_ON(!dn); | |
1339 | BUG_ON(!dir); | |
2b0143b5 | 1340 | BUG_ON(d_inode(dn->d_parent) != dir); |
f5d55f03 JL |
1341 | |
1342 | dvino.ino = le64_to_cpu(rinfo->diri.in->ino); | |
1343 | dvino.snap = le64_to_cpu(rinfo->diri.in->snapid); | |
1344 | ||
1345 | BUG_ON(ceph_ino(dir) != dvino.ino); | |
1346 | BUG_ON(ceph_snap(dir) != dvino.snap); | |
355da1eb | 1347 | |
355da1eb SW |
1348 | /* do we have a lease on the whole dir? */ |
1349 | have_dir_cap = | |
1350 | (le32_to_cpu(rinfo->diri.in->cap.caps) & | |
1351 | CEPH_CAP_FILE_SHARED); | |
1352 | ||
1353 | /* do we have a dn lease? */ | |
1354 | have_lease = have_dir_cap || | |
2f90b852 | 1355 | le32_to_cpu(rinfo->dlease->duration_ms); |
355da1eb SW |
1356 | if (!have_lease) |
1357 | dout("fill_trace no dentry lease or dir cap\n"); | |
1358 | ||
1359 | /* rename? */ | |
1360 | if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) { | |
0a8a70f9 YZ |
1361 | struct inode *olddir = req->r_old_dentry_dir; |
1362 | BUG_ON(!olddir); | |
1363 | ||
a455589f | 1364 | dout(" src %p '%pd' dst %p '%pd'\n", |
355da1eb | 1365 | req->r_old_dentry, |
a455589f AV |
1366 | req->r_old_dentry, |
1367 | dn, dn); | |
355da1eb SW |
1368 | dout("fill_trace doing d_move %p -> %p\n", |
1369 | req->r_old_dentry, dn); | |
c10f5e12 | 1370 | |
fdd4e158 YZ |
1371 | /* d_move screws up sibling dentries' offsets */ |
1372 | ceph_dir_clear_ordered(dir); | |
1373 | ceph_dir_clear_ordered(olddir); | |
1374 | ||
355da1eb | 1375 | d_move(req->r_old_dentry, dn); |
a455589f AV |
1376 | dout(" src %p '%pd' dst %p '%pd'\n", |
1377 | req->r_old_dentry, | |
355da1eb | 1378 | req->r_old_dentry, |
a455589f | 1379 | dn, dn); |
81a6cf2d | 1380 | |
c4a29f26 SW |
1381 | /* ensure target dentry is invalidated, despite |
1382 | rehashing bug in vfs_rename_dir */ | |
81a6cf2d SW |
1383 | ceph_invalidate_dentry_lease(dn); |
1384 | ||
99ccbd22 | 1385 | dout("dn %p gets new offset %lld\n", req->r_old_dentry, |
1cd3935b | 1386 | ceph_dentry(req->r_old_dentry)->offset); |
81a6cf2d | 1387 | |
2bf996ac YZ |
1388 | /* swap r_dentry and r_old_dentry in case that |
1389 | * splice_dentry() gets called later. This is safe | |
1390 | * because no other place will use them */ | |
1391 | req->r_dentry = req->r_old_dentry; | |
1392 | req->r_old_dentry = dn; | |
1393 | dn = req->r_dentry; | |
355da1eb SW |
1394 | } |
1395 | ||
1396 | /* null dentry? */ | |
1397 | if (!rinfo->head->is_target) { | |
1398 | dout("fill_trace null dentry\n"); | |
2b0143b5 | 1399 | if (d_really_is_positive(dn)) { |
355da1eb | 1400 | dout("d_delete %p\n", dn); |
5495c2d0 | 1401 | ceph_dir_clear_ordered(dir); |
355da1eb | 1402 | d_delete(dn); |
80d025ff JL |
1403 | } else if (have_lease) { |
1404 | if (d_unhashed(dn)) | |
f8b31710 | 1405 | d_add(dn, NULL); |
543212b3 YZ |
1406 | update_dentry_lease(dir, dn, |
1407 | rinfo->dlease, session, | |
1408 | req->r_request_started); | |
355da1eb SW |
1409 | } |
1410 | goto done; | |
1411 | } | |
1412 | ||
1413 | /* attach proper inode */ | |
2b0143b5 | 1414 | if (d_really_is_negative(dn)) { |
70db4f36 | 1415 | ceph_dir_clear_ordered(dir); |
86b58d13 | 1416 | ihold(in); |
2bf996ac YZ |
1417 | err = splice_dentry(&req->r_dentry, in); |
1418 | if (err < 0) | |
355da1eb | 1419 | goto done; |
2bf996ac | 1420 | dn = req->r_dentry; /* may have spliced */ |
2b0143b5 | 1421 | } else if (d_really_is_positive(dn) && d_inode(dn) != in) { |
355da1eb | 1422 | dout(" %p links to %p %llx.%llx, not %llx.%llx\n", |
2b0143b5 | 1423 | dn, d_inode(dn), ceph_vinop(d_inode(dn)), |
86b58d13 | 1424 | ceph_vinop(in)); |
200fd27c | 1425 | d_invalidate(dn); |
355da1eb | 1426 | have_lease = false; |
355da1eb SW |
1427 | } |
1428 | ||
f5d55f03 | 1429 | if (have_lease) { |
543212b3 YZ |
1430 | update_dentry_lease(dir, dn, |
1431 | rinfo->dlease, session, | |
1432 | req->r_request_started); | |
f5d55f03 | 1433 | } |
355da1eb | 1434 | dout(" final dn %p\n", dn); |
bc2de10d JL |
1435 | } else if ((req->r_op == CEPH_MDS_OP_LOOKUPSNAP || |
1436 | req->r_op == CEPH_MDS_OP_MKSNAP) && | |
1437 | !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) { | |
3dd69aab | 1438 | struct inode *dir = req->r_parent; |
355da1eb SW |
1439 | |
1440 | /* fill out a snapdir LOOKUPSNAP dentry */ | |
0a8a70f9 YZ |
1441 | BUG_ON(!dir); |
1442 | BUG_ON(ceph_snap(dir) != CEPH_SNAPDIR); | |
2bf996ac YZ |
1443 | BUG_ON(!req->r_dentry); |
1444 | dout(" linking snapped dir %p to dn %p\n", in, req->r_dentry); | |
70db4f36 | 1445 | ceph_dir_clear_ordered(dir); |
86b58d13 | 1446 | ihold(in); |
2bf996ac YZ |
1447 | err = splice_dentry(&req->r_dentry, in); |
1448 | if (err < 0) | |
355da1eb | 1449 | goto done; |
543212b3 YZ |
1450 | } else if (rinfo->head->is_dentry && req->r_dentry) { |
1451 | /* parent inode is not locked, be carefull */ | |
cdde7c43 | 1452 | struct ceph_vino *ptvino = NULL; |
543212b3 YZ |
1453 | dvino.ino = le64_to_cpu(rinfo->diri.in->ino); |
1454 | dvino.snap = le64_to_cpu(rinfo->diri.in->snapid); | |
1455 | if (rinfo->head->is_target) { | |
1456 | tvino.ino = le64_to_cpu(rinfo->targeti.in->ino); | |
1457 | tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid); | |
1458 | ptvino = &tvino; | |
cdde7c43 | 1459 | } |
543212b3 YZ |
1460 | update_dentry_lease_careful(req->r_dentry, rinfo->dlease, |
1461 | session, req->r_request_started, | |
1462 | rinfo->dname, rinfo->dname_len, | |
1463 | &dvino, ptvino); | |
355da1eb | 1464 | } |
355da1eb SW |
1465 | done: |
1466 | dout("fill_trace done err=%d\n", err); | |
1467 | return err; | |
1468 | } | |
1469 | ||
1470 | /* | |
1471 | * Prepopulate our cache with readdir results, leases, etc. | |
1472 | */ | |
79f9f99a SW |
1473 | static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req, |
1474 | struct ceph_mds_session *session) | |
1475 | { | |
1476 | struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info; | |
1477 | int i, err = 0; | |
1478 | ||
1479 | for (i = 0; i < rinfo->dir_nr; i++) { | |
2a5beea3 | 1480 | struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i; |
79f9f99a SW |
1481 | struct ceph_vino vino; |
1482 | struct inode *in; | |
1483 | int rc; | |
1484 | ||
2a5beea3 YZ |
1485 | vino.ino = le64_to_cpu(rde->inode.in->ino); |
1486 | vino.snap = le64_to_cpu(rde->inode.in->snapid); | |
79f9f99a SW |
1487 | |
1488 | in = ceph_get_inode(req->r_dentry->d_sb, vino); | |
1489 | if (IS_ERR(in)) { | |
1490 | err = PTR_ERR(in); | |
1491 | dout("new_inode badness got %d\n", err); | |
1492 | continue; | |
1493 | } | |
2a5beea3 | 1494 | rc = fill_inode(in, NULL, &rde->inode, NULL, session, |
79f9f99a SW |
1495 | req->r_request_started, -1, |
1496 | &req->r_caps_reservation); | |
1497 | if (rc < 0) { | |
1498 | pr_err("fill_inode badness on %p got %d\n", in, rc); | |
1499 | err = rc; | |
79f9f99a | 1500 | } |
3e1d0452 YZ |
1501 | /* avoid calling iput_final() in mds dispatch threads */ |
1502 | ceph_async_iput(in); | |
79f9f99a SW |
1503 | } |
1504 | ||
1505 | return err; | |
1506 | } | |
1507 | ||
fdd4e158 YZ |
1508 | void ceph_readdir_cache_release(struct ceph_readdir_cache_control *ctl) |
1509 | { | |
1510 | if (ctl->page) { | |
1511 | kunmap(ctl->page); | |
09cbfeaf | 1512 | put_page(ctl->page); |
fdd4e158 YZ |
1513 | ctl->page = NULL; |
1514 | } | |
1515 | } | |
1516 | ||
1517 | static int fill_readdir_cache(struct inode *dir, struct dentry *dn, | |
1518 | struct ceph_readdir_cache_control *ctl, | |
1519 | struct ceph_mds_request *req) | |
1520 | { | |
1521 | struct ceph_inode_info *ci = ceph_inode(dir); | |
09cbfeaf | 1522 | unsigned nsize = PAGE_SIZE / sizeof(struct dentry*); |
fdd4e158 YZ |
1523 | unsigned idx = ctl->index % nsize; |
1524 | pgoff_t pgoff = ctl->index / nsize; | |
1525 | ||
1526 | if (!ctl->page || pgoff != page_index(ctl->page)) { | |
1527 | ceph_readdir_cache_release(ctl); | |
af5e5eb5 YZ |
1528 | if (idx == 0) |
1529 | ctl->page = grab_cache_page(&dir->i_data, pgoff); | |
1530 | else | |
1531 | ctl->page = find_lock_page(&dir->i_data, pgoff); | |
fdd4e158 YZ |
1532 | if (!ctl->page) { |
1533 | ctl->index = -1; | |
af5e5eb5 | 1534 | return idx == 0 ? -ENOMEM : 0; |
fdd4e158 YZ |
1535 | } |
1536 | /* reading/filling the cache are serialized by | |
1537 | * i_mutex, no need to use page lock */ | |
1538 | unlock_page(ctl->page); | |
1539 | ctl->dentries = kmap(ctl->page); | |
af5e5eb5 | 1540 | if (idx == 0) |
09cbfeaf | 1541 | memset(ctl->dentries, 0, PAGE_SIZE); |
fdd4e158 YZ |
1542 | } |
1543 | ||
1544 | if (req->r_dir_release_cnt == atomic64_read(&ci->i_release_count) && | |
1545 | req->r_dir_ordered_cnt == atomic64_read(&ci->i_ordered_count)) { | |
1546 | dout("readdir cache dn %p idx %d\n", dn, ctl->index); | |
1547 | ctl->dentries[idx] = dn; | |
1548 | ctl->index++; | |
1549 | } else { | |
1550 | dout("disable readdir cache\n"); | |
1551 | ctl->index = -1; | |
1552 | } | |
1553 | return 0; | |
1554 | } | |
1555 | ||
355da1eb SW |
1556 | int ceph_readdir_prepopulate(struct ceph_mds_request *req, |
1557 | struct ceph_mds_session *session) | |
1558 | { | |
1559 | struct dentry *parent = req->r_dentry; | |
f3c4ebe6 | 1560 | struct ceph_inode_info *ci = ceph_inode(d_inode(parent)); |
355da1eb SW |
1561 | struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info; |
1562 | struct qstr dname; | |
1563 | struct dentry *dn; | |
1564 | struct inode *in; | |
315f2408 | 1565 | int err = 0, skipped = 0, ret, i; |
355da1eb | 1566 | struct ceph_mds_request_head *rhead = req->r_request->front.iov_base; |
81c6aea5 | 1567 | u32 frag = le32_to_cpu(rhead->args.readdir.frag); |
f3c4ebe6 YZ |
1568 | u32 last_hash = 0; |
1569 | u32 fpos_offset; | |
fdd4e158 YZ |
1570 | struct ceph_readdir_cache_control cache_ctl = {}; |
1571 | ||
bc2de10d | 1572 | if (test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) |
fdd4e158 | 1573 | return readdir_prepopulate_inodes_only(req, session); |
81c6aea5 | 1574 | |
79162547 YZ |
1575 | if (rinfo->hash_order) { |
1576 | if (req->r_path2) { | |
1577 | last_hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash, | |
1578 | req->r_path2, | |
1579 | strlen(req->r_path2)); | |
1580 | last_hash = ceph_frag_value(last_hash); | |
1581 | } else if (rinfo->offset_hash) { | |
1582 | /* mds understands offset_hash */ | |
1583 | WARN_ON_ONCE(req->r_readdir_offset != 2); | |
1584 | last_hash = le32_to_cpu(rhead->args.readdir.offset_hash); | |
1585 | } | |
f3c4ebe6 YZ |
1586 | } |
1587 | ||
81c6aea5 YZ |
1588 | if (rinfo->dir_dir && |
1589 | le32_to_cpu(rinfo->dir_dir->frag) != frag) { | |
1590 | dout("readdir_prepopulate got new frag %x -> %x\n", | |
1591 | frag, le32_to_cpu(rinfo->dir_dir->frag)); | |
1592 | frag = le32_to_cpu(rinfo->dir_dir->frag); | |
f3c4ebe6 | 1593 | if (!rinfo->hash_order) |
fdd4e158 | 1594 | req->r_readdir_offset = 2; |
81c6aea5 | 1595 | } |
355da1eb SW |
1596 | |
1597 | if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) { | |
355da1eb SW |
1598 | dout("readdir_prepopulate %d items under SNAPDIR dn %p\n", |
1599 | rinfo->dir_nr, parent); | |
1600 | } else { | |
1601 | dout("readdir_prepopulate %d items under dn %p\n", | |
1602 | rinfo->dir_nr, parent); | |
1603 | if (rinfo->dir_dir) | |
2b0143b5 | 1604 | ceph_fill_dirfrag(d_inode(parent), rinfo->dir_dir); |
355da1eb | 1605 | |
8d45b911 YZ |
1606 | if (ceph_frag_is_leftmost(frag) && |
1607 | req->r_readdir_offset == 2 && | |
1608 | !(rinfo->hash_order && last_hash)) { | |
1609 | /* note dir version at start of readdir so we can | |
1610 | * tell if any dentries get dropped */ | |
1611 | req->r_dir_release_cnt = | |
1612 | atomic64_read(&ci->i_release_count); | |
1613 | req->r_dir_ordered_cnt = | |
1614 | atomic64_read(&ci->i_ordered_count); | |
1615 | req->r_readdir_cache_idx = 0; | |
1616 | } | |
fdd4e158 YZ |
1617 | } |
1618 | ||
1619 | cache_ctl.index = req->r_readdir_cache_idx; | |
f3c4ebe6 | 1620 | fpos_offset = req->r_readdir_offset; |
fdd4e158 | 1621 | |
86b58d13 | 1622 | /* FIXME: release caps/leases if error occurs */ |
355da1eb | 1623 | for (i = 0; i < rinfo->dir_nr; i++) { |
2a5beea3 | 1624 | struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i; |
543212b3 | 1625 | struct ceph_vino tvino; |
355da1eb | 1626 | |
2a5beea3 YZ |
1627 | dname.name = rde->name; |
1628 | dname.len = rde->name_len; | |
8387ff25 | 1629 | dname.hash = full_name_hash(parent, dname.name, dname.len); |
355da1eb | 1630 | |
f5d55f03 JL |
1631 | tvino.ino = le64_to_cpu(rde->inode.in->ino); |
1632 | tvino.snap = le64_to_cpu(rde->inode.in->snapid); | |
355da1eb | 1633 | |
f3c4ebe6 YZ |
1634 | if (rinfo->hash_order) { |
1635 | u32 hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash, | |
1636 | rde->name, rde->name_len); | |
1637 | hash = ceph_frag_value(hash); | |
1638 | if (hash != last_hash) | |
1639 | fpos_offset = 2; | |
1640 | last_hash = hash; | |
1641 | rde->offset = ceph_make_fpos(hash, fpos_offset++, true); | |
1642 | } else { | |
1643 | rde->offset = ceph_make_fpos(frag, fpos_offset++, false); | |
1644 | } | |
355da1eb SW |
1645 | |
1646 | retry_lookup: | |
1647 | dn = d_lookup(parent, &dname); | |
1648 | dout("d_lookup on parent=%p name=%.*s got %p\n", | |
1649 | parent, dname.len, dname.name, dn); | |
1650 | ||
1651 | if (!dn) { | |
1652 | dn = d_alloc(parent, &dname); | |
1653 | dout("d_alloc %p '%.*s' = %p\n", parent, | |
1654 | dname.len, dname.name, dn); | |
d37b1d99 | 1655 | if (!dn) { |
355da1eb SW |
1656 | dout("d_alloc badness\n"); |
1657 | err = -ENOMEM; | |
1658 | goto out; | |
1659 | } | |
2b0143b5 | 1660 | } else if (d_really_is_positive(dn) && |
f5d55f03 JL |
1661 | (ceph_ino(d_inode(dn)) != tvino.ino || |
1662 | ceph_snap(d_inode(dn)) != tvino.snap)) { | |
5495c2d0 | 1663 | struct ceph_dentry_info *di = ceph_dentry(dn); |
355da1eb | 1664 | dout(" dn %p points to wrong inode %p\n", |
2b0143b5 | 1665 | dn, d_inode(dn)); |
5495c2d0 YZ |
1666 | |
1667 | spin_lock(&dn->d_lock); | |
1668 | if (di->offset > 0 && | |
1669 | di->lease_shared_gen == | |
1670 | atomic_read(&ci->i_shared_gen)) { | |
1671 | __ceph_dir_clear_ordered(ci); | |
1672 | di->offset = 0; | |
1673 | } | |
1674 | spin_unlock(&dn->d_lock); | |
1675 | ||
355da1eb SW |
1676 | d_delete(dn); |
1677 | dput(dn); | |
1678 | goto retry_lookup; | |
355da1eb SW |
1679 | } |
1680 | ||
355da1eb | 1681 | /* inode */ |
2b0143b5 DH |
1682 | if (d_really_is_positive(dn)) { |
1683 | in = d_inode(dn); | |
355da1eb | 1684 | } else { |
f5d55f03 | 1685 | in = ceph_get_inode(parent->d_sb, tvino); |
ac1f12ef | 1686 | if (IS_ERR(in)) { |
355da1eb | 1687 | dout("new_inode badness\n"); |
2744c171 | 1688 | d_drop(dn); |
355da1eb | 1689 | dput(dn); |
ac1f12ef | 1690 | err = PTR_ERR(in); |
355da1eb SW |
1691 | goto out; |
1692 | } | |
355da1eb SW |
1693 | } |
1694 | ||
2a5beea3 | 1695 | ret = fill_inode(in, NULL, &rde->inode, NULL, session, |
fdd4e158 YZ |
1696 | req->r_request_started, -1, |
1697 | &req->r_caps_reservation); | |
1698 | if (ret < 0) { | |
355da1eb | 1699 | pr_err("fill_inode badness on %p\n", in); |
3e1d0452 YZ |
1700 | if (d_really_is_negative(dn)) { |
1701 | /* avoid calling iput_final() in mds | |
1702 | * dispatch threads */ | |
1703 | ceph_async_iput(in); | |
1704 | } | |
86b58d13 | 1705 | d_drop(dn); |
fdd4e158 | 1706 | err = ret; |
d69ed05a | 1707 | goto next_item; |
355da1eb | 1708 | } |
86b58d13 | 1709 | |
2b0143b5 | 1710 | if (d_really_is_negative(dn)) { |
315f2408 YZ |
1711 | if (ceph_security_xattr_deadlock(in)) { |
1712 | dout(" skip splicing dn %p to inode %p" | |
1713 | " (security xattr deadlock)\n", dn, in); | |
3e1d0452 | 1714 | ceph_async_iput(in); |
315f2408 YZ |
1715 | skipped++; |
1716 | goto next_item; | |
1717 | } | |
1718 | ||
2bf996ac YZ |
1719 | err = splice_dentry(&dn, in); |
1720 | if (err < 0) | |
86b58d13 | 1721 | goto next_item; |
86b58d13 YZ |
1722 | } |
1723 | ||
f3c4ebe6 | 1724 | ceph_dentry(dn)->offset = rde->offset; |
86b58d13 | 1725 | |
543212b3 YZ |
1726 | update_dentry_lease(d_inode(parent), dn, |
1727 | rde->lease, req->r_session, | |
1728 | req->r_request_started); | |
fdd4e158 | 1729 | |
315f2408 | 1730 | if (err == 0 && skipped == 0 && cache_ctl.index >= 0) { |
fdd4e158 YZ |
1731 | ret = fill_readdir_cache(d_inode(parent), dn, |
1732 | &cache_ctl, req); | |
1733 | if (ret < 0) | |
1734 | err = ret; | |
1735 | } | |
d69ed05a | 1736 | next_item: |
2bf996ac | 1737 | dput(dn); |
355da1eb | 1738 | } |
355da1eb | 1739 | out: |
315f2408 | 1740 | if (err == 0 && skipped == 0) { |
bc2de10d | 1741 | set_bit(CEPH_MDS_R_DID_PREPOPULATE, &req->r_req_flags); |
fdd4e158 YZ |
1742 | req->r_readdir_cache_idx = cache_ctl.index; |
1743 | } | |
1744 | ceph_readdir_cache_release(&cache_ctl); | |
355da1eb SW |
1745 | dout("readdir_prepopulate done\n"); |
1746 | return err; | |
1747 | } | |
1748 | ||
efb0ca76 | 1749 | bool ceph_inode_set_size(struct inode *inode, loff_t size) |
355da1eb SW |
1750 | { |
1751 | struct ceph_inode_info *ci = ceph_inode(inode); | |
efb0ca76 | 1752 | bool ret; |
355da1eb | 1753 | |
be655596 | 1754 | spin_lock(&ci->i_ceph_lock); |
355da1eb | 1755 | dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size); |
99c88e69 | 1756 | i_size_write(inode, size); |
224a7542 | 1757 | inode->i_blocks = calc_inode_blocks(size); |
355da1eb | 1758 | |
efb0ca76 | 1759 | ret = __ceph_should_report_size(ci); |
355da1eb | 1760 | |
be655596 | 1761 | spin_unlock(&ci->i_ceph_lock); |
355da1eb SW |
1762 | return ret; |
1763 | } | |
1764 | ||
3e1d0452 YZ |
1765 | /* |
1766 | * Put reference to inode, but avoid calling iput_final() in current thread. | |
1767 | * iput_final() may wait for reahahead pages. The wait can cause deadlock in | |
1768 | * some contexts. | |
1769 | */ | |
1770 | void ceph_async_iput(struct inode *inode) | |
1771 | { | |
1772 | if (!inode) | |
1773 | return; | |
1774 | for (;;) { | |
1775 | if (atomic_add_unless(&inode->i_count, -1, 1)) | |
1776 | break; | |
1777 | if (queue_work(ceph_inode_to_client(inode)->inode_wq, | |
1778 | &ceph_inode(inode)->i_work)) | |
1779 | break; | |
1780 | /* queue work failed, i_count must be at least 2 */ | |
1781 | } | |
1782 | } | |
1783 | ||
355da1eb SW |
1784 | /* |
1785 | * Write back inode data in a worker thread. (This can't be done | |
1786 | * in the message handler context.) | |
1787 | */ | |
3c6f6b79 SW |
1788 | void ceph_queue_writeback(struct inode *inode) |
1789 | { | |
1cf89a8d YZ |
1790 | struct ceph_inode_info *ci = ceph_inode(inode); |
1791 | set_bit(CEPH_I_WORK_WRITEBACK, &ci->i_work_mask); | |
1792 | ||
15a2015f | 1793 | ihold(inode); |
1cf89a8d YZ |
1794 | if (queue_work(ceph_inode_to_client(inode)->inode_wq, |
1795 | &ci->i_work)) { | |
2c27c9a5 | 1796 | dout("ceph_queue_writeback %p\n", inode); |
3c6f6b79 | 1797 | } else { |
1cf89a8d YZ |
1798 | dout("ceph_queue_writeback %p already queued, mask=%lx\n", |
1799 | inode, ci->i_work_mask); | |
15a2015f | 1800 | iput(inode); |
3c6f6b79 SW |
1801 | } |
1802 | } | |
1803 | ||
3c6f6b79 SW |
1804 | /* |
1805 | * queue an async invalidation | |
1806 | */ | |
1807 | void ceph_queue_invalidate(struct inode *inode) | |
1808 | { | |
1cf89a8d YZ |
1809 | struct ceph_inode_info *ci = ceph_inode(inode); |
1810 | set_bit(CEPH_I_WORK_INVALIDATE_PAGES, &ci->i_work_mask); | |
1811 | ||
15a2015f | 1812 | ihold(inode); |
1cf89a8d YZ |
1813 | if (queue_work(ceph_inode_to_client(inode)->inode_wq, |
1814 | &ceph_inode(inode)->i_work)) { | |
3c6f6b79 | 1815 | dout("ceph_queue_invalidate %p\n", inode); |
3c6f6b79 | 1816 | } else { |
1cf89a8d YZ |
1817 | dout("ceph_queue_invalidate %p already queued, mask=%lx\n", |
1818 | inode, ci->i_work_mask); | |
15a2015f | 1819 | iput(inode); |
3c6f6b79 SW |
1820 | } |
1821 | } | |
1822 | ||
355da1eb | 1823 | /* |
1cf89a8d YZ |
1824 | * Queue an async vmtruncate. If we fail to queue work, we will handle |
1825 | * the truncation the next time we call __ceph_do_pending_vmtruncate. | |
355da1eb | 1826 | */ |
1cf89a8d | 1827 | void ceph_queue_vmtruncate(struct inode *inode) |
355da1eb | 1828 | { |
1cf89a8d YZ |
1829 | struct ceph_inode_info *ci = ceph_inode(inode); |
1830 | set_bit(CEPH_I_WORK_VMTRUNCATE, &ci->i_work_mask); | |
1831 | ||
1832 | ihold(inode); | |
1833 | if (queue_work(ceph_inode_to_client(inode)->inode_wq, | |
1834 | &ci->i_work)) { | |
1835 | dout("ceph_queue_vmtruncate %p\n", inode); | |
1836 | } else { | |
1837 | dout("ceph_queue_vmtruncate %p already queued, mask=%lx\n", | |
1838 | inode, ci->i_work_mask); | |
1839 | iput(inode); | |
1840 | } | |
1841 | } | |
1842 | ||
1843 | static void ceph_do_invalidate_pages(struct inode *inode) | |
1844 | { | |
1845 | struct ceph_inode_info *ci = ceph_inode(inode); | |
6c93df5d | 1846 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); |
355da1eb SW |
1847 | u32 orig_gen; |
1848 | int check = 0; | |
1849 | ||
b0d7c223 | 1850 | mutex_lock(&ci->i_truncate_mutex); |
6c93df5d | 1851 | |
52953d55 | 1852 | if (READ_ONCE(fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) { |
6c93df5d YZ |
1853 | pr_warn_ratelimited("invalidate_pages %p %lld forced umount\n", |
1854 | inode, ceph_ino(inode)); | |
1855 | mapping_set_error(inode->i_mapping, -EIO); | |
1856 | truncate_pagecache(inode, 0); | |
1857 | mutex_unlock(&ci->i_truncate_mutex); | |
1858 | goto out; | |
1859 | } | |
1860 | ||
be655596 | 1861 | spin_lock(&ci->i_ceph_lock); |
355da1eb SW |
1862 | dout("invalidate_pages %p gen %d revoking %d\n", inode, |
1863 | ci->i_rdcache_gen, ci->i_rdcache_revoking); | |
cd045cb4 | 1864 | if (ci->i_rdcache_revoking != ci->i_rdcache_gen) { |
9563f88c YZ |
1865 | if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE)) |
1866 | check = 1; | |
be655596 | 1867 | spin_unlock(&ci->i_ceph_lock); |
b0d7c223 | 1868 | mutex_unlock(&ci->i_truncate_mutex); |
355da1eb SW |
1869 | goto out; |
1870 | } | |
1871 | orig_gen = ci->i_rdcache_gen; | |
be655596 | 1872 | spin_unlock(&ci->i_ceph_lock); |
355da1eb | 1873 | |
9abd4db7 YZ |
1874 | if (invalidate_inode_pages2(inode->i_mapping) < 0) { |
1875 | pr_err("invalidate_pages %p fails\n", inode); | |
1876 | } | |
355da1eb | 1877 | |
be655596 | 1878 | spin_lock(&ci->i_ceph_lock); |
cd045cb4 SW |
1879 | if (orig_gen == ci->i_rdcache_gen && |
1880 | orig_gen == ci->i_rdcache_revoking) { | |
355da1eb SW |
1881 | dout("invalidate_pages %p gen %d successful\n", inode, |
1882 | ci->i_rdcache_gen); | |
cd045cb4 | 1883 | ci->i_rdcache_revoking--; |
355da1eb SW |
1884 | check = 1; |
1885 | } else { | |
cd045cb4 SW |
1886 | dout("invalidate_pages %p gen %d raced, now %d revoking %d\n", |
1887 | inode, orig_gen, ci->i_rdcache_gen, | |
1888 | ci->i_rdcache_revoking); | |
9563f88c YZ |
1889 | if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE)) |
1890 | check = 1; | |
355da1eb | 1891 | } |
be655596 | 1892 | spin_unlock(&ci->i_ceph_lock); |
b0d7c223 | 1893 | mutex_unlock(&ci->i_truncate_mutex); |
9563f88c | 1894 | out: |
355da1eb SW |
1895 | if (check) |
1896 | ceph_check_caps(ci, 0, NULL); | |
3c6f6b79 SW |
1897 | } |
1898 | ||
355da1eb | 1899 | /* |
355da1eb SW |
1900 | * Make sure any pending truncation is applied before doing anything |
1901 | * that may depend on it. | |
1902 | */ | |
b415bf4f | 1903 | void __ceph_do_pending_vmtruncate(struct inode *inode) |
355da1eb SW |
1904 | { |
1905 | struct ceph_inode_info *ci = ceph_inode(inode); | |
1906 | u64 to; | |
a85f50b6 | 1907 | int wrbuffer_refs, finish = 0; |
355da1eb | 1908 | |
b0d7c223 | 1909 | mutex_lock(&ci->i_truncate_mutex); |
355da1eb | 1910 | retry: |
be655596 | 1911 | spin_lock(&ci->i_ceph_lock); |
355da1eb SW |
1912 | if (ci->i_truncate_pending == 0) { |
1913 | dout("__do_pending_vmtruncate %p none pending\n", inode); | |
be655596 | 1914 | spin_unlock(&ci->i_ceph_lock); |
b0d7c223 | 1915 | mutex_unlock(&ci->i_truncate_mutex); |
355da1eb SW |
1916 | return; |
1917 | } | |
1918 | ||
1919 | /* | |
1920 | * make sure any dirty snapped pages are flushed before we | |
1921 | * possibly truncate them.. so write AND block! | |
1922 | */ | |
1923 | if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) { | |
c8fd0d37 | 1924 | spin_unlock(&ci->i_ceph_lock); |
355da1eb SW |
1925 | dout("__do_pending_vmtruncate %p flushing snaps first\n", |
1926 | inode); | |
355da1eb SW |
1927 | filemap_write_and_wait_range(&inode->i_data, 0, |
1928 | inode->i_sb->s_maxbytes); | |
1929 | goto retry; | |
1930 | } | |
1931 | ||
b0d7c223 YZ |
1932 | /* there should be no reader or writer */ |
1933 | WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref); | |
1934 | ||
355da1eb SW |
1935 | to = ci->i_truncate_size; |
1936 | wrbuffer_refs = ci->i_wrbuffer_ref; | |
1937 | dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode, | |
1938 | ci->i_truncate_pending, to); | |
be655596 | 1939 | spin_unlock(&ci->i_ceph_lock); |
355da1eb | 1940 | |
4e217b5d | 1941 | truncate_pagecache(inode, to); |
355da1eb | 1942 | |
be655596 | 1943 | spin_lock(&ci->i_ceph_lock); |
a85f50b6 YZ |
1944 | if (to == ci->i_truncate_size) { |
1945 | ci->i_truncate_pending = 0; | |
1946 | finish = 1; | |
1947 | } | |
be655596 | 1948 | spin_unlock(&ci->i_ceph_lock); |
a85f50b6 YZ |
1949 | if (!finish) |
1950 | goto retry; | |
355da1eb | 1951 | |
b0d7c223 YZ |
1952 | mutex_unlock(&ci->i_truncate_mutex); |
1953 | ||
355da1eb SW |
1954 | if (wrbuffer_refs == 0) |
1955 | ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL); | |
a85f50b6 YZ |
1956 | |
1957 | wake_up_all(&ci->i_cap_wq); | |
355da1eb SW |
1958 | } |
1959 | ||
1cf89a8d YZ |
1960 | static void ceph_inode_work(struct work_struct *work) |
1961 | { | |
1962 | struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info, | |
1963 | i_work); | |
1964 | struct inode *inode = &ci->vfs_inode; | |
1965 | ||
1966 | if (test_and_clear_bit(CEPH_I_WORK_WRITEBACK, &ci->i_work_mask)) { | |
1967 | dout("writeback %p\n", inode); | |
1968 | filemap_fdatawrite(&inode->i_data); | |
1969 | } | |
1970 | if (test_and_clear_bit(CEPH_I_WORK_INVALIDATE_PAGES, &ci->i_work_mask)) | |
1971 | ceph_do_invalidate_pages(inode); | |
1972 | ||
1973 | if (test_and_clear_bit(CEPH_I_WORK_VMTRUNCATE, &ci->i_work_mask)) | |
1974 | __ceph_do_pending_vmtruncate(inode); | |
1975 | ||
1976 | iput(inode); | |
1977 | } | |
1978 | ||
355da1eb SW |
1979 | /* |
1980 | * symlinks | |
1981 | */ | |
355da1eb | 1982 | static const struct inode_operations ceph_symlink_iops = { |
6b255391 | 1983 | .get_link = simple_get_link, |
0b932672 YZ |
1984 | .setattr = ceph_setattr, |
1985 | .getattr = ceph_getattr, | |
0b932672 | 1986 | .listxattr = ceph_listxattr, |
355da1eb SW |
1987 | }; |
1988 | ||
a26fecca | 1989 | int __ceph_setattr(struct inode *inode, struct iattr *attr) |
355da1eb | 1990 | { |
355da1eb | 1991 | struct ceph_inode_info *ci = ceph_inode(inode); |
c62498d7 | 1992 | unsigned int ia_valid = attr->ia_valid; |
355da1eb | 1993 | struct ceph_mds_request *req; |
a26fecca | 1994 | struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc; |
f66fd9f0 | 1995 | struct ceph_cap_flush *prealloc_cf; |
355da1eb SW |
1996 | int issued; |
1997 | int release = 0, dirtied = 0; | |
1998 | int mask = 0; | |
1999 | int err = 0; | |
fca65b4a | 2000 | int inode_dirty_flags = 0; |
604d1b02 | 2001 | bool lock_snap_rwsem = false; |
355da1eb | 2002 | |
f66fd9f0 YZ |
2003 | prealloc_cf = ceph_alloc_cap_flush(); |
2004 | if (!prealloc_cf) | |
2005 | return -ENOMEM; | |
2006 | ||
355da1eb SW |
2007 | req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR, |
2008 | USE_AUTH_MDS); | |
f66fd9f0 YZ |
2009 | if (IS_ERR(req)) { |
2010 | ceph_free_cap_flush(prealloc_cf); | |
355da1eb | 2011 | return PTR_ERR(req); |
f66fd9f0 | 2012 | } |
355da1eb | 2013 | |
be655596 | 2014 | spin_lock(&ci->i_ceph_lock); |
355da1eb | 2015 | issued = __ceph_caps_issued(ci, NULL); |
604d1b02 YZ |
2016 | |
2017 | if (!ci->i_head_snapc && | |
2018 | (issued & (CEPH_CAP_ANY_EXCL | CEPH_CAP_FILE_WR))) { | |
2019 | lock_snap_rwsem = true; | |
2020 | if (!down_read_trylock(&mdsc->snap_rwsem)) { | |
2021 | spin_unlock(&ci->i_ceph_lock); | |
2022 | down_read(&mdsc->snap_rwsem); | |
2023 | spin_lock(&ci->i_ceph_lock); | |
2024 | issued = __ceph_caps_issued(ci, NULL); | |
2025 | } | |
2026 | } | |
2027 | ||
355da1eb SW |
2028 | dout("setattr %p issued %s\n", inode, ceph_cap_string(issued)); |
2029 | ||
2030 | if (ia_valid & ATTR_UID) { | |
2031 | dout("setattr %p uid %d -> %d\n", inode, | |
bd2bae6a EB |
2032 | from_kuid(&init_user_ns, inode->i_uid), |
2033 | from_kuid(&init_user_ns, attr->ia_uid)); | |
355da1eb SW |
2034 | if (issued & CEPH_CAP_AUTH_EXCL) { |
2035 | inode->i_uid = attr->ia_uid; | |
2036 | dirtied |= CEPH_CAP_AUTH_EXCL; | |
2037 | } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || | |
ab871b90 EB |
2038 | !uid_eq(attr->ia_uid, inode->i_uid)) { |
2039 | req->r_args.setattr.uid = cpu_to_le32( | |
2040 | from_kuid(&init_user_ns, attr->ia_uid)); | |
355da1eb SW |
2041 | mask |= CEPH_SETATTR_UID; |
2042 | release |= CEPH_CAP_AUTH_SHARED; | |
2043 | } | |
2044 | } | |
2045 | if (ia_valid & ATTR_GID) { | |
2046 | dout("setattr %p gid %d -> %d\n", inode, | |
bd2bae6a EB |
2047 | from_kgid(&init_user_ns, inode->i_gid), |
2048 | from_kgid(&init_user_ns, attr->ia_gid)); | |
355da1eb SW |
2049 | if (issued & CEPH_CAP_AUTH_EXCL) { |
2050 | inode->i_gid = attr->ia_gid; | |
2051 | dirtied |= CEPH_CAP_AUTH_EXCL; | |
2052 | } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || | |
ab871b90 EB |
2053 | !gid_eq(attr->ia_gid, inode->i_gid)) { |
2054 | req->r_args.setattr.gid = cpu_to_le32( | |
2055 | from_kgid(&init_user_ns, attr->ia_gid)); | |
355da1eb SW |
2056 | mask |= CEPH_SETATTR_GID; |
2057 | release |= CEPH_CAP_AUTH_SHARED; | |
2058 | } | |
2059 | } | |
2060 | if (ia_valid & ATTR_MODE) { | |
2061 | dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode, | |
2062 | attr->ia_mode); | |
2063 | if (issued & CEPH_CAP_AUTH_EXCL) { | |
2064 | inode->i_mode = attr->ia_mode; | |
2065 | dirtied |= CEPH_CAP_AUTH_EXCL; | |
2066 | } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || | |
2067 | attr->ia_mode != inode->i_mode) { | |
7221fe4c | 2068 | inode->i_mode = attr->ia_mode; |
355da1eb SW |
2069 | req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode); |
2070 | mask |= CEPH_SETATTR_MODE; | |
2071 | release |= CEPH_CAP_AUTH_SHARED; | |
2072 | } | |
2073 | } | |
2074 | ||
2075 | if (ia_valid & ATTR_ATIME) { | |
13442b03 | 2076 | dout("setattr %p atime %lld.%ld -> %lld.%ld\n", inode, |
9bbeab41 AB |
2077 | inode->i_atime.tv_sec, inode->i_atime.tv_nsec, |
2078 | attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec); | |
355da1eb SW |
2079 | if (issued & CEPH_CAP_FILE_EXCL) { |
2080 | ci->i_time_warp_seq++; | |
2081 | inode->i_atime = attr->ia_atime; | |
2082 | dirtied |= CEPH_CAP_FILE_EXCL; | |
2083 | } else if ((issued & CEPH_CAP_FILE_WR) && | |
95582b00 | 2084 | timespec64_compare(&inode->i_atime, |
355da1eb SW |
2085 | &attr->ia_atime) < 0) { |
2086 | inode->i_atime = attr->ia_atime; | |
2087 | dirtied |= CEPH_CAP_FILE_WR; | |
2088 | } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || | |
95582b00 | 2089 | !timespec64_equal(&inode->i_atime, &attr->ia_atime)) { |
9bbeab41 AB |
2090 | ceph_encode_timespec64(&req->r_args.setattr.atime, |
2091 | &attr->ia_atime); | |
355da1eb | 2092 | mask |= CEPH_SETATTR_ATIME; |
be70489e YZ |
2093 | release |= CEPH_CAP_FILE_SHARED | |
2094 | CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR; | |
355da1eb SW |
2095 | } |
2096 | } | |
c62498d7 JL |
2097 | if (ia_valid & ATTR_SIZE) { |
2098 | dout("setattr %p size %lld -> %lld\n", inode, | |
2099 | inode->i_size, attr->ia_size); | |
2100 | if ((issued & CEPH_CAP_FILE_EXCL) && | |
2101 | attr->ia_size > inode->i_size) { | |
2102 | i_size_write(inode, attr->ia_size); | |
2103 | inode->i_blocks = calc_inode_blocks(attr->ia_size); | |
2104 | ci->i_reported_size = attr->ia_size; | |
2105 | dirtied |= CEPH_CAP_FILE_EXCL; | |
2106 | ia_valid |= ATTR_MTIME; | |
2107 | } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || | |
2108 | attr->ia_size != inode->i_size) { | |
2109 | req->r_args.setattr.size = cpu_to_le64(attr->ia_size); | |
2110 | req->r_args.setattr.old_size = | |
2111 | cpu_to_le64(inode->i_size); | |
2112 | mask |= CEPH_SETATTR_SIZE; | |
2113 | release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_EXCL | | |
2114 | CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR; | |
2115 | } | |
2116 | } | |
355da1eb | 2117 | if (ia_valid & ATTR_MTIME) { |
13442b03 | 2118 | dout("setattr %p mtime %lld.%ld -> %lld.%ld\n", inode, |
9bbeab41 AB |
2119 | inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec, |
2120 | attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec); | |
355da1eb SW |
2121 | if (issued & CEPH_CAP_FILE_EXCL) { |
2122 | ci->i_time_warp_seq++; | |
2123 | inode->i_mtime = attr->ia_mtime; | |
2124 | dirtied |= CEPH_CAP_FILE_EXCL; | |
2125 | } else if ((issued & CEPH_CAP_FILE_WR) && | |
95582b00 | 2126 | timespec64_compare(&inode->i_mtime, |
355da1eb SW |
2127 | &attr->ia_mtime) < 0) { |
2128 | inode->i_mtime = attr->ia_mtime; | |
2129 | dirtied |= CEPH_CAP_FILE_WR; | |
2130 | } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || | |
95582b00 | 2131 | !timespec64_equal(&inode->i_mtime, &attr->ia_mtime)) { |
9bbeab41 AB |
2132 | ceph_encode_timespec64(&req->r_args.setattr.mtime, |
2133 | &attr->ia_mtime); | |
355da1eb | 2134 | mask |= CEPH_SETATTR_MTIME; |
be70489e YZ |
2135 | release |= CEPH_CAP_FILE_SHARED | |
2136 | CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR; | |
355da1eb SW |
2137 | } |
2138 | } | |
355da1eb SW |
2139 | |
2140 | /* these do nothing */ | |
2141 | if (ia_valid & ATTR_CTIME) { | |
2142 | bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME| | |
2143 | ATTR_MODE|ATTR_UID|ATTR_GID)) == 0; | |
13442b03 | 2144 | dout("setattr %p ctime %lld.%ld -> %lld.%ld (%s)\n", inode, |
9bbeab41 AB |
2145 | inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec, |
2146 | attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec, | |
355da1eb | 2147 | only ? "ctime only" : "ignored"); |
355da1eb SW |
2148 | if (only) { |
2149 | /* | |
2150 | * if kernel wants to dirty ctime but nothing else, | |
2151 | * we need to choose a cap to dirty under, or do | |
2152 | * a almost-no-op setattr | |
2153 | */ | |
2154 | if (issued & CEPH_CAP_AUTH_EXCL) | |
2155 | dirtied |= CEPH_CAP_AUTH_EXCL; | |
2156 | else if (issued & CEPH_CAP_FILE_EXCL) | |
2157 | dirtied |= CEPH_CAP_FILE_EXCL; | |
2158 | else if (issued & CEPH_CAP_XATTR_EXCL) | |
2159 | dirtied |= CEPH_CAP_XATTR_EXCL; | |
2160 | else | |
2161 | mask |= CEPH_SETATTR_CTIME; | |
2162 | } | |
2163 | } | |
2164 | if (ia_valid & ATTR_FILE) | |
2165 | dout("setattr %p ATTR_FILE ... hrm!\n", inode); | |
2166 | ||
2167 | if (dirtied) { | |
f66fd9f0 YZ |
2168 | inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied, |
2169 | &prealloc_cf); | |
4ca2fea6 | 2170 | inode->i_ctime = attr->ia_ctime; |
355da1eb SW |
2171 | } |
2172 | ||
2173 | release &= issued; | |
be655596 | 2174 | spin_unlock(&ci->i_ceph_lock); |
604d1b02 YZ |
2175 | if (lock_snap_rwsem) |
2176 | up_read(&mdsc->snap_rwsem); | |
355da1eb | 2177 | |
fca65b4a SW |
2178 | if (inode_dirty_flags) |
2179 | __mark_inode_dirty(inode, inode_dirty_flags); | |
2180 | ||
7221fe4c | 2181 | |
355da1eb | 2182 | if (mask) { |
70b666c3 SW |
2183 | req->r_inode = inode; |
2184 | ihold(inode); | |
355da1eb SW |
2185 | req->r_inode_drop = release; |
2186 | req->r_args.setattr.mask = cpu_to_le32(mask); | |
2187 | req->r_num_caps = 1; | |
0ed1e90a | 2188 | req->r_stamp = attr->ia_ctime; |
752c8bdc | 2189 | err = ceph_mdsc_do_request(mdsc, NULL, req); |
355da1eb SW |
2190 | } |
2191 | dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err, | |
2192 | ceph_cap_string(dirtied), mask); | |
2193 | ||
355da1eb | 2194 | ceph_mdsc_put_request(req); |
f66fd9f0 | 2195 | ceph_free_cap_flush(prealloc_cf); |
8179a101 YZ |
2196 | |
2197 | if (err >= 0 && (mask & CEPH_SETATTR_SIZE)) | |
2198 | __ceph_do_pending_vmtruncate(inode); | |
2199 | ||
355da1eb SW |
2200 | return err; |
2201 | } | |
2202 | ||
a26fecca AG |
2203 | /* |
2204 | * setattr | |
2205 | */ | |
2206 | int ceph_setattr(struct dentry *dentry, struct iattr *attr) | |
2207 | { | |
fd5472ed | 2208 | struct inode *inode = d_inode(dentry); |
36a4c72d | 2209 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); |
fd5472ed JK |
2210 | int err; |
2211 | ||
2212 | if (ceph_snap(inode) != CEPH_NOSNAP) | |
2213 | return -EROFS; | |
2214 | ||
31051c85 | 2215 | err = setattr_prepare(dentry, attr); |
fd5472ed JK |
2216 | if (err != 0) |
2217 | return err; | |
2218 | ||
36a4c72d CX |
2219 | if ((attr->ia_valid & ATTR_SIZE) && |
2220 | attr->ia_size > max(inode->i_size, fsc->max_file_size)) | |
2221 | return -EFBIG; | |
2222 | ||
2b83845f LH |
2223 | if ((attr->ia_valid & ATTR_SIZE) && |
2224 | ceph_quota_is_max_bytes_exceeded(inode, attr->ia_size)) | |
2225 | return -EDQUOT; | |
2226 | ||
8179a101 YZ |
2227 | err = __ceph_setattr(inode, attr); |
2228 | ||
2229 | if (err >= 0 && (attr->ia_valid & ATTR_MODE)) | |
2230 | err = posix_acl_chmod(inode, attr->ia_mode); | |
2231 | ||
2232 | return err; | |
a26fecca AG |
2233 | } |
2234 | ||
355da1eb SW |
2235 | /* |
2236 | * Verify that we have a lease on the given mask. If not, | |
2237 | * do a getattr against an mds. | |
2238 | */ | |
01deead0 YZ |
2239 | int __ceph_do_getattr(struct inode *inode, struct page *locked_page, |
2240 | int mask, bool force) | |
355da1eb | 2241 | { |
3d14c5d2 YS |
2242 | struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb); |
2243 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
355da1eb | 2244 | struct ceph_mds_request *req; |
49a9f4f6 | 2245 | int mode; |
355da1eb SW |
2246 | int err; |
2247 | ||
2248 | if (ceph_snap(inode) == CEPH_SNAPDIR) { | |
2249 | dout("do_getattr inode %p SNAPDIR\n", inode); | |
2250 | return 0; | |
2251 | } | |
2252 | ||
01deead0 YZ |
2253 | dout("do_getattr inode %p mask %s mode 0%o\n", |
2254 | inode, ceph_cap_string(mask), inode->i_mode); | |
508b32d8 | 2255 | if (!force && ceph_caps_issued_mask(ceph_inode(inode), mask, 1)) |
355da1eb SW |
2256 | return 0; |
2257 | ||
49a9f4f6 YZ |
2258 | mode = (mask & CEPH_STAT_RSTAT) ? USE_AUTH_MDS : USE_ANY_MDS; |
2259 | req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, mode); | |
355da1eb SW |
2260 | if (IS_ERR(req)) |
2261 | return PTR_ERR(req); | |
70b666c3 SW |
2262 | req->r_inode = inode; |
2263 | ihold(inode); | |
355da1eb SW |
2264 | req->r_num_caps = 1; |
2265 | req->r_args.getattr.mask = cpu_to_le32(mask); | |
01deead0 | 2266 | req->r_locked_page = locked_page; |
355da1eb | 2267 | err = ceph_mdsc_do_request(mdsc, NULL, req); |
01deead0 YZ |
2268 | if (locked_page && err == 0) { |
2269 | u64 inline_version = req->r_reply_info.targeti.inline_version; | |
2270 | if (inline_version == 0) { | |
2271 | /* the reply is supposed to contain inline data */ | |
2272 | err = -EINVAL; | |
2273 | } else if (inline_version == CEPH_INLINE_NONE) { | |
2274 | err = -ENODATA; | |
2275 | } else { | |
2276 | err = req->r_reply_info.targeti.inline_len; | |
2277 | } | |
2278 | } | |
355da1eb SW |
2279 | ceph_mdsc_put_request(req); |
2280 | dout("do_getattr result=%d\n", err); | |
2281 | return err; | |
2282 | } | |
2283 | ||
2284 | ||
2285 | /* | |
2286 | * Check inode permissions. We verify we have a valid value for | |
2287 | * the AUTH cap, then call the generic handler. | |
2288 | */ | |
10556cb2 | 2289 | int ceph_permission(struct inode *inode, int mask) |
355da1eb | 2290 | { |
b74c79e9 NP |
2291 | int err; |
2292 | ||
10556cb2 | 2293 | if (mask & MAY_NOT_BLOCK) |
b74c79e9 NP |
2294 | return -ECHILD; |
2295 | ||
508b32d8 | 2296 | err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED, false); |
355da1eb SW |
2297 | |
2298 | if (!err) | |
2830ba7f | 2299 | err = generic_permission(inode, mask); |
355da1eb SW |
2300 | return err; |
2301 | } | |
2302 | ||
428bb68a JL |
2303 | /* Craft a mask of needed caps given a set of requested statx attrs. */ |
2304 | static int statx_to_caps(u32 want) | |
2305 | { | |
2306 | int mask = 0; | |
2307 | ||
58981784 | 2308 | if (want & (STATX_MODE|STATX_UID|STATX_GID|STATX_CTIME|STATX_BTIME)) |
428bb68a JL |
2309 | mask |= CEPH_CAP_AUTH_SHARED; |
2310 | ||
2311 | if (want & (STATX_NLINK|STATX_CTIME)) | |
2312 | mask |= CEPH_CAP_LINK_SHARED; | |
2313 | ||
2314 | if (want & (STATX_ATIME|STATX_MTIME|STATX_CTIME|STATX_SIZE| | |
2315 | STATX_BLOCKS)) | |
2316 | mask |= CEPH_CAP_FILE_SHARED; | |
2317 | ||
2318 | if (want & (STATX_CTIME)) | |
2319 | mask |= CEPH_CAP_XATTR_SHARED; | |
2320 | ||
2321 | return mask; | |
2322 | } | |
2323 | ||
355da1eb | 2324 | /* |
428bb68a JL |
2325 | * Get all the attributes. If we have sufficient caps for the requested attrs, |
2326 | * then we can avoid talking to the MDS at all. | |
355da1eb | 2327 | */ |
a528d35e DH |
2328 | int ceph_getattr(const struct path *path, struct kstat *stat, |
2329 | u32 request_mask, unsigned int flags) | |
355da1eb | 2330 | { |
a528d35e | 2331 | struct inode *inode = d_inode(path->dentry); |
232d4b01 | 2332 | struct ceph_inode_info *ci = ceph_inode(inode); |
58981784 | 2333 | u32 valid_mask = STATX_BASIC_STATS; |
428bb68a | 2334 | int err = 0; |
355da1eb | 2335 | |
428bb68a JL |
2336 | /* Skip the getattr altogether if we're asked not to sync */ |
2337 | if (!(flags & AT_STATX_DONT_SYNC)) { | |
2338 | err = ceph_do_getattr(inode, statx_to_caps(request_mask), | |
2339 | flags & AT_STATX_FORCE_SYNC); | |
2340 | if (err) | |
2341 | return err; | |
2342 | } | |
2343 | ||
2344 | generic_fillattr(inode, stat); | |
2345 | stat->ino = ceph_translate_ino(inode->i_sb, inode->i_ino); | |
58981784 JL |
2346 | |
2347 | /* | |
2348 | * btime on newly-allocated inodes is 0, so if this is still set to | |
2349 | * that, then assume that it's not valid. | |
2350 | */ | |
2351 | if (ci->i_btime.tv_sec || ci->i_btime.tv_nsec) { | |
2352 | stat->btime = ci->i_btime; | |
2353 | valid_mask |= STATX_BTIME; | |
2354 | } | |
2355 | ||
428bb68a JL |
2356 | if (ceph_snap(inode) == CEPH_NOSNAP) |
2357 | stat->dev = inode->i_sb->s_dev; | |
2358 | else | |
2359 | stat->dev = ci->i_snapid_map ? ci->i_snapid_map->dev : 0; | |
2360 | ||
2361 | if (S_ISDIR(inode->i_mode)) { | |
2362 | if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb), | |
2363 | RBYTES)) | |
2364 | stat->size = ci->i_rbytes; | |
355da1eb | 2365 | else |
428bb68a JL |
2366 | stat->size = ci->i_files + ci->i_subdirs; |
2367 | stat->blocks = 0; | |
2368 | stat->blksize = 65536; | |
2369 | /* | |
2370 | * Some applications rely on the number of st_nlink | |
2371 | * value on directories to be either 0 (if unlinked) | |
2372 | * or 2 + number of subdirectories. | |
2373 | */ | |
2374 | if (stat->nlink == 1) | |
2375 | /* '.' + '..' + subdirs */ | |
2376 | stat->nlink = 1 + 1 + ci->i_subdirs; | |
355da1eb | 2377 | } |
428bb68a | 2378 | |
58981784 | 2379 | stat->result_mask = request_mask & valid_mask; |
355da1eb SW |
2380 | return err; |
2381 | } |