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 | { | |
ebce3eb2 | 44 | struct ceph_inode_info *ci = ceph_inode(inode); |
1dd8d470 | 45 | struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb); |
ebce3eb2 JL |
46 | |
47 | ci->i_vino = *(struct ceph_vino *)data; | |
48 | inode->i_ino = ceph_vino_to_ino_t(ci->i_vino); | |
a35ead31 | 49 | inode_set_iversion_raw(inode, 0); |
1dd8d470 XL |
50 | percpu_counter_inc(&mdsc->metric.total_inodes); |
51 | ||
ad1fee96 YS |
52 | return 0; |
53 | } | |
54 | ||
355da1eb SW |
55 | struct inode *ceph_get_inode(struct super_block *sb, struct ceph_vino vino) |
56 | { | |
57 | struct inode *inode; | |
355da1eb | 58 | |
d4f6b31d JL |
59 | if (ceph_vino_is_reserved(vino)) |
60 | return ERR_PTR(-EREMOTEIO); | |
61 | ||
ebce3eb2 JL |
62 | inode = iget5_locked(sb, (unsigned long)vino.ino, ceph_ino_compare, |
63 | ceph_set_ino_cb, &vino); | |
d37b1d99 | 64 | if (!inode) |
355da1eb | 65 | return ERR_PTR(-ENOMEM); |
355da1eb | 66 | |
ebce3eb2 JL |
67 | dout("get_inode on %llu=%llx.%llx got %p new %d\n", ceph_present_inode(inode), |
68 | ceph_vinop(inode), inode, !!(inode->i_state & I_NEW)); | |
355da1eb SW |
69 | return inode; |
70 | } | |
71 | ||
72 | /* | |
73 | * get/constuct snapdir inode for a given directory | |
74 | */ | |
75 | struct inode *ceph_get_snapdir(struct inode *parent) | |
76 | { | |
77 | struct ceph_vino vino = { | |
78 | .ino = ceph_ino(parent), | |
79 | .snap = CEPH_SNAPDIR, | |
80 | }; | |
81 | struct inode *inode = ceph_get_inode(parent->i_sb, vino); | |
b377ff13 | 82 | struct ceph_inode_info *ci = ceph_inode(inode); |
355da1eb | 83 | |
355da1eb | 84 | if (IS_ERR(inode)) |
7e34bc52 | 85 | return inode; |
3e10a15f JL |
86 | |
87 | if (!S_ISDIR(parent->i_mode)) { | |
88 | pr_warn_once("bad snapdir parent type (mode=0%o)\n", | |
89 | parent->i_mode); | |
90 | return ERR_PTR(-ENOTDIR); | |
91 | } | |
92 | ||
93 | if (!(inode->i_state & I_NEW) && !S_ISDIR(inode->i_mode)) { | |
94 | pr_warn_once("bad snapdir inode type (mode=0%o)\n", | |
95 | inode->i_mode); | |
96 | return ERR_PTR(-ENOTDIR); | |
97 | } | |
98 | ||
355da1eb SW |
99 | inode->i_mode = parent->i_mode; |
100 | inode->i_uid = parent->i_uid; | |
101 | inode->i_gid = parent->i_gid; | |
ef915725 LH |
102 | inode->i_mtime = parent->i_mtime; |
103 | inode->i_ctime = parent->i_ctime; | |
104 | inode->i_atime = parent->i_atime; | |
b377ff13 | 105 | ci->i_rbytes = 0; |
ef915725 | 106 | ci->i_btime = ceph_inode(parent)->i_btime; |
893e456b | 107 | |
d3c51ae1 JL |
108 | if (inode->i_state & I_NEW) { |
109 | inode->i_op = &ceph_snapdir_iops; | |
110 | inode->i_fop = &ceph_snapdir_fops; | |
111 | ci->i_snap_caps = CEPH_CAP_PIN; /* so we can open */ | |
893e456b | 112 | unlock_new_inode(inode); |
d3c51ae1 | 113 | } |
893e456b | 114 | |
355da1eb SW |
115 | return inode; |
116 | } | |
117 | ||
118 | const struct inode_operations ceph_file_iops = { | |
119 | .permission = ceph_permission, | |
120 | .setattr = ceph_setattr, | |
121 | .getattr = ceph_getattr, | |
355da1eb | 122 | .listxattr = ceph_listxattr, |
7221fe4c | 123 | .get_acl = ceph_get_acl, |
72466d0b | 124 | .set_acl = ceph_set_acl, |
355da1eb SW |
125 | }; |
126 | ||
127 | ||
128 | /* | |
129 | * We use a 'frag tree' to keep track of the MDS's directory fragments | |
130 | * for a given inode (usually there is just a single fragment). We | |
131 | * need to know when a child frag is delegated to a new MDS, or when | |
132 | * it is flagged as replicated, so we can direct our requests | |
133 | * accordingly. | |
134 | */ | |
135 | ||
136 | /* | |
137 | * find/create a frag in the tree | |
138 | */ | |
139 | static struct ceph_inode_frag *__get_or_create_frag(struct ceph_inode_info *ci, | |
140 | u32 f) | |
141 | { | |
142 | struct rb_node **p; | |
143 | struct rb_node *parent = NULL; | |
144 | struct ceph_inode_frag *frag; | |
145 | int c; | |
146 | ||
147 | p = &ci->i_fragtree.rb_node; | |
148 | while (*p) { | |
149 | parent = *p; | |
150 | frag = rb_entry(parent, struct ceph_inode_frag, node); | |
151 | c = ceph_frag_compare(f, frag->frag); | |
152 | if (c < 0) | |
153 | p = &(*p)->rb_left; | |
154 | else if (c > 0) | |
155 | p = &(*p)->rb_right; | |
156 | else | |
157 | return frag; | |
158 | } | |
159 | ||
160 | frag = kmalloc(sizeof(*frag), GFP_NOFS); | |
51308806 | 161 | if (!frag) |
355da1eb | 162 | return ERR_PTR(-ENOMEM); |
51308806 | 163 | |
355da1eb SW |
164 | frag->frag = f; |
165 | frag->split_by = 0; | |
166 | frag->mds = -1; | |
167 | frag->ndist = 0; | |
168 | ||
169 | rb_link_node(&frag->node, parent, p); | |
170 | rb_insert_color(&frag->node, &ci->i_fragtree); | |
171 | ||
172 | dout("get_or_create_frag added %llx.%llx frag %x\n", | |
173 | ceph_vinop(&ci->vfs_inode), f); | |
174 | return frag; | |
175 | } | |
176 | ||
177 | /* | |
178 | * find a specific frag @f | |
179 | */ | |
180 | struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci, u32 f) | |
181 | { | |
182 | struct rb_node *n = ci->i_fragtree.rb_node; | |
183 | ||
184 | while (n) { | |
185 | struct ceph_inode_frag *frag = | |
186 | rb_entry(n, struct ceph_inode_frag, node); | |
187 | int c = ceph_frag_compare(f, frag->frag); | |
188 | if (c < 0) | |
189 | n = n->rb_left; | |
190 | else if (c > 0) | |
191 | n = n->rb_right; | |
192 | else | |
193 | return frag; | |
194 | } | |
195 | return NULL; | |
196 | } | |
197 | ||
198 | /* | |
199 | * Choose frag containing the given value @v. If @pfrag is | |
200 | * specified, copy the frag delegation info to the caller if | |
201 | * it is present. | |
202 | */ | |
3e7fbe9c YZ |
203 | static u32 __ceph_choose_frag(struct ceph_inode_info *ci, u32 v, |
204 | struct ceph_inode_frag *pfrag, int *found) | |
355da1eb SW |
205 | { |
206 | u32 t = ceph_frag_make(0, 0); | |
207 | struct ceph_inode_frag *frag; | |
208 | unsigned nway, i; | |
209 | u32 n; | |
210 | ||
211 | if (found) | |
212 | *found = 0; | |
213 | ||
355da1eb SW |
214 | while (1) { |
215 | WARN_ON(!ceph_frag_contains_value(t, v)); | |
216 | frag = __ceph_find_frag(ci, t); | |
217 | if (!frag) | |
218 | break; /* t is a leaf */ | |
219 | if (frag->split_by == 0) { | |
220 | if (pfrag) | |
221 | memcpy(pfrag, frag, sizeof(*pfrag)); | |
222 | if (found) | |
223 | *found = 1; | |
224 | break; | |
225 | } | |
226 | ||
227 | /* choose child */ | |
228 | nway = 1 << frag->split_by; | |
229 | dout("choose_frag(%x) %x splits by %d (%d ways)\n", v, t, | |
230 | frag->split_by, nway); | |
231 | for (i = 0; i < nway; i++) { | |
232 | n = ceph_frag_make_child(t, frag->split_by, i); | |
233 | if (ceph_frag_contains_value(n, v)) { | |
234 | t = n; | |
235 | break; | |
236 | } | |
237 | } | |
238 | BUG_ON(i == nway); | |
239 | } | |
240 | dout("choose_frag(%x) = %x\n", v, t); | |
241 | ||
355da1eb SW |
242 | return t; |
243 | } | |
244 | ||
3e7fbe9c YZ |
245 | u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v, |
246 | struct ceph_inode_frag *pfrag, int *found) | |
247 | { | |
248 | u32 ret; | |
249 | mutex_lock(&ci->i_fragtree_mutex); | |
250 | ret = __ceph_choose_frag(ci, v, pfrag, found); | |
251 | mutex_unlock(&ci->i_fragtree_mutex); | |
252 | return ret; | |
253 | } | |
254 | ||
355da1eb SW |
255 | /* |
256 | * Process dirfrag (delegation) info from the mds. Include leaf | |
257 | * fragment in tree ONLY if ndist > 0. Otherwise, only | |
258 | * branches/splits are included in i_fragtree) | |
259 | */ | |
260 | static int ceph_fill_dirfrag(struct inode *inode, | |
261 | struct ceph_mds_reply_dirfrag *dirinfo) | |
262 | { | |
263 | struct ceph_inode_info *ci = ceph_inode(inode); | |
264 | struct ceph_inode_frag *frag; | |
265 | u32 id = le32_to_cpu(dirinfo->frag); | |
266 | int mds = le32_to_cpu(dirinfo->auth); | |
267 | int ndist = le32_to_cpu(dirinfo->ndist); | |
8d08503c | 268 | int diri_auth = -1; |
355da1eb SW |
269 | int i; |
270 | int err = 0; | |
271 | ||
8d08503c YZ |
272 | spin_lock(&ci->i_ceph_lock); |
273 | if (ci->i_auth_cap) | |
274 | diri_auth = ci->i_auth_cap->mds; | |
275 | spin_unlock(&ci->i_ceph_lock); | |
276 | ||
42172119 YZ |
277 | if (mds == -1) /* CDIR_AUTH_PARENT */ |
278 | mds = diri_auth; | |
279 | ||
355da1eb | 280 | mutex_lock(&ci->i_fragtree_mutex); |
8d08503c | 281 | if (ndist == 0 && mds == diri_auth) { |
355da1eb SW |
282 | /* no delegation info needed. */ |
283 | frag = __ceph_find_frag(ci, id); | |
284 | if (!frag) | |
285 | goto out; | |
286 | if (frag->split_by == 0) { | |
287 | /* tree leaf, remove */ | |
288 | dout("fill_dirfrag removed %llx.%llx frag %x" | |
289 | " (no ref)\n", ceph_vinop(inode), id); | |
290 | rb_erase(&frag->node, &ci->i_fragtree); | |
291 | kfree(frag); | |
292 | } else { | |
293 | /* tree branch, keep and clear */ | |
294 | dout("fill_dirfrag cleared %llx.%llx frag %x" | |
295 | " referral\n", ceph_vinop(inode), id); | |
296 | frag->mds = -1; | |
297 | frag->ndist = 0; | |
298 | } | |
299 | goto out; | |
300 | } | |
301 | ||
302 | ||
303 | /* find/add this frag to store mds delegation info */ | |
304 | frag = __get_or_create_frag(ci, id); | |
305 | if (IS_ERR(frag)) { | |
306 | /* this is not the end of the world; we can continue | |
307 | with bad/inaccurate delegation info */ | |
308 | pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n", | |
309 | ceph_vinop(inode), le32_to_cpu(dirinfo->frag)); | |
310 | err = -ENOMEM; | |
311 | goto out; | |
312 | } | |
313 | ||
314 | frag->mds = mds; | |
315 | frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP); | |
316 | for (i = 0; i < frag->ndist; i++) | |
317 | frag->dist[i] = le32_to_cpu(dirinfo->dist[i]); | |
318 | dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n", | |
319 | ceph_vinop(inode), frag->frag, frag->ndist); | |
320 | ||
321 | out: | |
322 | mutex_unlock(&ci->i_fragtree_mutex); | |
323 | return err; | |
324 | } | |
325 | ||
a407846e YZ |
326 | static int frag_tree_split_cmp(const void *l, const void *r) |
327 | { | |
328 | struct ceph_frag_tree_split *ls = (struct ceph_frag_tree_split*)l; | |
329 | struct ceph_frag_tree_split *rs = (struct ceph_frag_tree_split*)r; | |
fe2ed425 JL |
330 | return ceph_frag_compare(le32_to_cpu(ls->frag), |
331 | le32_to_cpu(rs->frag)); | |
a407846e YZ |
332 | } |
333 | ||
a4b7431f YZ |
334 | static bool is_frag_child(u32 f, struct ceph_inode_frag *frag) |
335 | { | |
336 | if (!frag) | |
337 | return f == ceph_frag_make(0, 0); | |
338 | if (ceph_frag_bits(f) != ceph_frag_bits(frag->frag) + frag->split_by) | |
339 | return false; | |
340 | return ceph_frag_contains_value(frag->frag, ceph_frag_value(f)); | |
341 | } | |
342 | ||
3e7fbe9c YZ |
343 | static int ceph_fill_fragtree(struct inode *inode, |
344 | struct ceph_frag_tree_head *fragtree, | |
345 | struct ceph_mds_reply_dirfrag *dirinfo) | |
346 | { | |
347 | struct ceph_inode_info *ci = ceph_inode(inode); | |
a4b7431f | 348 | struct ceph_inode_frag *frag, *prev_frag = NULL; |
3e7fbe9c | 349 | struct rb_node *rb_node; |
1b1bc16d YZ |
350 | unsigned i, split_by, nsplits; |
351 | u32 id; | |
3e7fbe9c YZ |
352 | bool update = false; |
353 | ||
354 | mutex_lock(&ci->i_fragtree_mutex); | |
355 | nsplits = le32_to_cpu(fragtree->nsplits); | |
1b1bc16d YZ |
356 | if (nsplits != ci->i_fragtree_nsplits) { |
357 | update = true; | |
358 | } else if (nsplits) { | |
3e7fbe9c YZ |
359 | i = prandom_u32() % nsplits; |
360 | id = le32_to_cpu(fragtree->splits[i].frag); | |
361 | if (!__ceph_find_frag(ci, id)) | |
362 | update = true; | |
363 | } else if (!RB_EMPTY_ROOT(&ci->i_fragtree)) { | |
364 | rb_node = rb_first(&ci->i_fragtree); | |
365 | frag = rb_entry(rb_node, struct ceph_inode_frag, node); | |
366 | if (frag->frag != ceph_frag_make(0, 0) || rb_next(rb_node)) | |
367 | update = true; | |
368 | } | |
369 | if (!update && dirinfo) { | |
370 | id = le32_to_cpu(dirinfo->frag); | |
371 | if (id != __ceph_choose_frag(ci, id, NULL, NULL)) | |
372 | update = true; | |
373 | } | |
374 | if (!update) | |
375 | goto out_unlock; | |
376 | ||
a407846e YZ |
377 | if (nsplits > 1) { |
378 | sort(fragtree->splits, nsplits, sizeof(fragtree->splits[0]), | |
379 | frag_tree_split_cmp, NULL); | |
380 | } | |
381 | ||
3e7fbe9c YZ |
382 | dout("fill_fragtree %llx.%llx\n", ceph_vinop(inode)); |
383 | rb_node = rb_first(&ci->i_fragtree); | |
384 | for (i = 0; i < nsplits; i++) { | |
385 | id = le32_to_cpu(fragtree->splits[i].frag); | |
1b1bc16d YZ |
386 | split_by = le32_to_cpu(fragtree->splits[i].by); |
387 | if (split_by == 0 || ceph_frag_bits(id) + split_by > 24) { | |
388 | pr_err("fill_fragtree %llx.%llx invalid split %d/%u, " | |
389 | "frag %x split by %d\n", ceph_vinop(inode), | |
390 | i, nsplits, id, split_by); | |
391 | continue; | |
392 | } | |
3e7fbe9c YZ |
393 | frag = NULL; |
394 | while (rb_node) { | |
395 | frag = rb_entry(rb_node, struct ceph_inode_frag, node); | |
396 | if (ceph_frag_compare(frag->frag, id) >= 0) { | |
397 | if (frag->frag != id) | |
398 | frag = NULL; | |
399 | else | |
400 | rb_node = rb_next(rb_node); | |
401 | break; | |
402 | } | |
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 | frag = NULL; |
413 | } | |
414 | if (!frag) { | |
415 | frag = __get_or_create_frag(ci, id); | |
416 | if (IS_ERR(frag)) | |
417 | continue; | |
418 | } | |
1b1bc16d YZ |
419 | if (frag->split_by == 0) |
420 | ci->i_fragtree_nsplits++; | |
421 | frag->split_by = split_by; | |
3e7fbe9c | 422 | dout(" frag %x split by %d\n", frag->frag, frag->split_by); |
a4b7431f | 423 | prev_frag = frag; |
3e7fbe9c YZ |
424 | } |
425 | while (rb_node) { | |
426 | frag = rb_entry(rb_node, struct ceph_inode_frag, node); | |
427 | rb_node = rb_next(rb_node); | |
a4b7431f YZ |
428 | /* delete stale split/leaf node */ |
429 | if (frag->split_by > 0 || | |
430 | !is_frag_child(frag->frag, prev_frag)) { | |
431 | rb_erase(&frag->node, &ci->i_fragtree); | |
1b1bc16d YZ |
432 | if (frag->split_by > 0) |
433 | ci->i_fragtree_nsplits--; | |
a4b7431f YZ |
434 | kfree(frag); |
435 | } | |
3e7fbe9c YZ |
436 | } |
437 | out_unlock: | |
438 | mutex_unlock(&ci->i_fragtree_mutex); | |
439 | return 0; | |
440 | } | |
355da1eb SW |
441 | |
442 | /* | |
443 | * initialize a newly allocated inode. | |
444 | */ | |
445 | struct inode *ceph_alloc_inode(struct super_block *sb) | |
446 | { | |
447 | struct ceph_inode_info *ci; | |
448 | int i; | |
449 | ||
fd60b288 | 450 | ci = alloc_inode_sb(sb, ceph_inode_cachep, GFP_NOFS); |
355da1eb SW |
451 | if (!ci) |
452 | return NULL; | |
453 | ||
454 | dout("alloc_inode %p\n", &ci->vfs_inode); | |
455 | ||
be655596 SW |
456 | spin_lock_init(&ci->i_ceph_lock); |
457 | ||
355da1eb | 458 | ci->i_version = 0; |
31c542a1 | 459 | ci->i_inline_version = 0; |
355da1eb SW |
460 | ci->i_time_warp_seq = 0; |
461 | ci->i_ceph_flags = 0; | |
fdd4e158 YZ |
462 | atomic64_set(&ci->i_ordered_count, 1); |
463 | atomic64_set(&ci->i_release_count, 1); | |
464 | atomic64_set(&ci->i_complete_seq[0], 0); | |
465 | atomic64_set(&ci->i_complete_seq[1], 0); | |
355da1eb SW |
466 | ci->i_symlink = NULL; |
467 | ||
fb18a575 LH |
468 | ci->i_max_bytes = 0; |
469 | ci->i_max_files = 0; | |
470 | ||
6c0f3af7 | 471 | memset(&ci->i_dir_layout, 0, sizeof(ci->i_dir_layout)); |
785892fe | 472 | memset(&ci->i_cached_layout, 0, sizeof(ci->i_cached_layout)); |
30c156d9 | 473 | RCU_INIT_POINTER(ci->i_layout.pool_ns, NULL); |
6c0f3af7 | 474 | |
355da1eb SW |
475 | ci->i_fragtree = RB_ROOT; |
476 | mutex_init(&ci->i_fragtree_mutex); | |
477 | ||
478 | ci->i_xattrs.blob = NULL; | |
479 | ci->i_xattrs.prealloc_blob = NULL; | |
480 | ci->i_xattrs.dirty = false; | |
481 | ci->i_xattrs.index = RB_ROOT; | |
482 | ci->i_xattrs.count = 0; | |
483 | ci->i_xattrs.names_size = 0; | |
484 | ci->i_xattrs.vals_size = 0; | |
485 | ci->i_xattrs.version = 0; | |
486 | ci->i_xattrs.index_version = 0; | |
487 | ||
488 | ci->i_caps = RB_ROOT; | |
489 | ci->i_auth_cap = NULL; | |
490 | ci->i_dirty_caps = 0; | |
491 | ci->i_flushing_caps = 0; | |
492 | INIT_LIST_HEAD(&ci->i_dirty_item); | |
493 | INIT_LIST_HEAD(&ci->i_flushing_item); | |
f66fd9f0 | 494 | ci->i_prealloc_cap_flush = NULL; |
e4500b5e | 495 | INIT_LIST_HEAD(&ci->i_cap_flush_list); |
355da1eb | 496 | init_waitqueue_head(&ci->i_cap_wq); |
355da1eb SW |
497 | ci->i_hold_caps_max = 0; |
498 | INIT_LIST_HEAD(&ci->i_cap_delay_list); | |
355da1eb SW |
499 | INIT_LIST_HEAD(&ci->i_cap_snaps); |
500 | ci->i_head_snapc = NULL; | |
501 | ci->i_snap_caps = 0; | |
502 | ||
719a2514 | 503 | ci->i_last_rd = ci->i_last_wr = jiffies - 3600 * HZ; |
774a6a11 | 504 | for (i = 0; i < CEPH_FILE_MODE_BITS; i++) |
355da1eb SW |
505 | ci->i_nr_by_mode[i] = 0; |
506 | ||
b0d7c223 | 507 | mutex_init(&ci->i_truncate_mutex); |
355da1eb SW |
508 | ci->i_truncate_seq = 0; |
509 | ci->i_truncate_size = 0; | |
510 | ci->i_truncate_pending = 0; | |
511 | ||
512 | ci->i_max_size = 0; | |
513 | ci->i_reported_size = 0; | |
514 | ci->i_wanted_max_size = 0; | |
515 | ci->i_requested_max_size = 0; | |
516 | ||
517 | ci->i_pin_ref = 0; | |
518 | ci->i_rd_ref = 0; | |
519 | ci->i_rdcache_ref = 0; | |
520 | ci->i_wr_ref = 0; | |
d3d0720d | 521 | ci->i_wb_ref = 0; |
f85122af | 522 | ci->i_fx_ref = 0; |
355da1eb SW |
523 | ci->i_wrbuffer_ref = 0; |
524 | ci->i_wrbuffer_ref_head = 0; | |
89aa5930 | 525 | atomic_set(&ci->i_filelock_ref, 0); |
1e9c2eb6 | 526 | atomic_set(&ci->i_shared_gen, 1); |
355da1eb SW |
527 | ci->i_rdcache_gen = 0; |
528 | ci->i_rdcache_revoking = 0; | |
529 | ||
355da1eb | 530 | INIT_LIST_HEAD(&ci->i_unsafe_dirops); |
68cd5b4b | 531 | INIT_LIST_HEAD(&ci->i_unsafe_iops); |
355da1eb SW |
532 | spin_lock_init(&ci->i_unsafe_lock); |
533 | ||
534 | ci->i_snap_realm = NULL; | |
535 | INIT_LIST_HEAD(&ci->i_snap_realm_item); | |
536 | INIT_LIST_HEAD(&ci->i_snap_flush_item); | |
537 | ||
1cf89a8d YZ |
538 | INIT_WORK(&ci->i_work, ceph_inode_work); |
539 | ci->i_work_mask = 0; | |
245ce991 | 540 | memset(&ci->i_btime, '\0', sizeof(ci->i_btime)); |
355da1eb | 541 | |
99ccbd22 MT |
542 | ceph_fscache_inode_init(ci); |
543 | ||
355da1eb SW |
544 | return &ci->vfs_inode; |
545 | } | |
546 | ||
cfa6d412 | 547 | void ceph_free_inode(struct inode *inode) |
fa0d7e3d | 548 | { |
fa0d7e3d NP |
549 | struct ceph_inode_info *ci = ceph_inode(inode); |
550 | ||
daf5cc27 | 551 | kfree(ci->i_symlink); |
fa0d7e3d NP |
552 | kmem_cache_free(ceph_inode_cachep, ci); |
553 | } | |
554 | ||
87bc5b89 | 555 | void ceph_evict_inode(struct inode *inode) |
355da1eb SW |
556 | { |
557 | struct ceph_inode_info *ci = ceph_inode(inode); | |
1dd8d470 | 558 | struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb); |
355da1eb SW |
559 | struct ceph_inode_frag *frag; |
560 | struct rb_node *n; | |
561 | ||
87bc5b89 YZ |
562 | dout("evict_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode)); |
563 | ||
1dd8d470 XL |
564 | percpu_counter_dec(&mdsc->metric.total_inodes); |
565 | ||
87bc5b89 | 566 | truncate_inode_pages_final(&inode->i_data); |
400e1286 JL |
567 | if (inode->i_state & I_PINNING_FSCACHE_WB) |
568 | ceph_fscache_unuse_cookie(inode, true); | |
87bc5b89 | 569 | clear_inode(inode); |
355da1eb | 570 | |
99ccbd22 MT |
571 | ceph_fscache_unregister_inode_cookie(ci); |
572 | ||
d6e47819 | 573 | __ceph_remove_caps(ci); |
355da1eb | 574 | |
d557c48d LH |
575 | if (__ceph_has_any_quota(ci)) |
576 | ceph_adjust_quota_realms_count(inode, false); | |
577 | ||
8b218b8a SW |
578 | /* |
579 | * we may still have a snap_realm reference if there are stray | |
d9df2783 | 580 | * caps in i_snap_caps. |
8b218b8a SW |
581 | */ |
582 | if (ci->i_snap_realm) { | |
75c9627e | 583 | if (ceph_snap(inode) == CEPH_NOSNAP) { |
75c9627e | 584 | dout(" dropping residual ref to snap realm %p\n", |
0ba92e1c JL |
585 | ci->i_snap_realm); |
586 | ceph_change_snap_realm(inode, NULL); | |
75c9627e YZ |
587 | } else { |
588 | ceph_put_snapid_map(mdsc, ci->i_snapid_map); | |
589 | ci->i_snap_realm = NULL; | |
590 | } | |
8b218b8a SW |
591 | } |
592 | ||
355da1eb SW |
593 | while ((n = rb_first(&ci->i_fragtree)) != NULL) { |
594 | frag = rb_entry(n, struct ceph_inode_frag, node); | |
595 | rb_erase(n, &ci->i_fragtree); | |
596 | kfree(frag); | |
597 | } | |
1b1bc16d | 598 | ci->i_fragtree_nsplits = 0; |
355da1eb SW |
599 | |
600 | __ceph_destroy_xattrs(ci); | |
b6c1d5b8 SW |
601 | if (ci->i_xattrs.blob) |
602 | ceph_buffer_put(ci->i_xattrs.blob); | |
603 | if (ci->i_xattrs.prealloc_blob) | |
604 | ceph_buffer_put(ci->i_xattrs.prealloc_blob); | |
355da1eb | 605 | |
779fe0fb | 606 | ceph_put_string(rcu_dereference_raw(ci->i_layout.pool_ns)); |
785892fe | 607 | ceph_put_string(rcu_dereference_raw(ci->i_cached_layout.pool_ns)); |
355da1eb SW |
608 | } |
609 | ||
224a7542 YZ |
610 | static inline blkcnt_t calc_inode_blocks(u64 size) |
611 | { | |
612 | return (size + (1<<9) - 1) >> 9; | |
613 | } | |
614 | ||
355da1eb SW |
615 | /* |
616 | * Helpers to fill in size, ctime, mtime, and atime. We have to be | |
617 | * careful because either the client or MDS may have more up to date | |
618 | * info, depending on which capabilities are held, and whether | |
619 | * time_warp_seq or truncate_seq have increased. (Ordinarily, mtime | |
620 | * and size are monotonically increasing, except when utimes() or | |
621 | * truncate() increments the corresponding _seq values.) | |
622 | */ | |
623 | int ceph_fill_file_size(struct inode *inode, int issued, | |
624 | u32 truncate_seq, u64 truncate_size, u64 size) | |
625 | { | |
626 | struct ceph_inode_info *ci = ceph_inode(inode); | |
627 | int queue_trunc = 0; | |
2d6795fb | 628 | loff_t isize = i_size_read(inode); |
355da1eb SW |
629 | |
630 | if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 || | |
2d6795fb JL |
631 | (truncate_seq == ci->i_truncate_seq && size > isize)) { |
632 | dout("size %lld -> %llu\n", isize, size); | |
a3d714c3 YZ |
633 | if (size > 0 && S_ISDIR(inode->i_mode)) { |
634 | pr_err("fill_file_size non-zero size for directory\n"); | |
635 | size = 0; | |
636 | } | |
99c88e69 | 637 | i_size_write(inode, size); |
224a7542 | 638 | inode->i_blocks = calc_inode_blocks(size); |
400e1286 JL |
639 | /* |
640 | * If we're expanding, then we should be able to just update | |
641 | * the existing cookie. | |
642 | */ | |
643 | if (size > isize) | |
644 | ceph_fscache_update(inode); | |
355da1eb SW |
645 | ci->i_reported_size = size; |
646 | if (truncate_seq != ci->i_truncate_seq) { | |
647 | dout("truncate_seq %u -> %u\n", | |
648 | ci->i_truncate_seq, truncate_seq); | |
649 | ci->i_truncate_seq = truncate_seq; | |
b0d7c223 YZ |
650 | |
651 | /* the MDS should have revoked these caps */ | |
652 | WARN_ON_ONCE(issued & (CEPH_CAP_FILE_EXCL | | |
653 | CEPH_CAP_FILE_RD | | |
654 | CEPH_CAP_FILE_WR | | |
655 | CEPH_CAP_FILE_LAZYIO)); | |
3d497d85 YS |
656 | /* |
657 | * If we hold relevant caps, or in the case where we're | |
658 | * not the only client referencing this file and we | |
659 | * don't hold those caps, then we need to check whether | |
660 | * the file is either opened or mmaped | |
661 | */ | |
b0d7c223 YZ |
662 | if ((issued & (CEPH_CAP_FILE_CACHE| |
663 | CEPH_CAP_FILE_BUFFER)) || | |
3d497d85 | 664 | mapping_mapped(inode->i_mapping) || |
719a2514 | 665 | __ceph_is_file_opened(ci)) { |
355da1eb SW |
666 | ci->i_truncate_pending++; |
667 | queue_trunc = 1; | |
668 | } | |
669 | } | |
670 | } | |
671 | if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 && | |
672 | ci->i_truncate_size != truncate_size) { | |
673 | dout("truncate_size %lld -> %llu\n", ci->i_truncate_size, | |
674 | truncate_size); | |
675 | ci->i_truncate_size = truncate_size; | |
676 | } | |
677 | return queue_trunc; | |
678 | } | |
679 | ||
680 | void ceph_fill_file_time(struct inode *inode, int issued, | |
9bbeab41 AB |
681 | u64 time_warp_seq, struct timespec64 *ctime, |
682 | struct timespec64 *mtime, struct timespec64 *atime) | |
355da1eb SW |
683 | { |
684 | struct ceph_inode_info *ci = ceph_inode(inode); | |
685 | int warn = 0; | |
686 | ||
687 | if (issued & (CEPH_CAP_FILE_EXCL| | |
688 | CEPH_CAP_FILE_WR| | |
d8672d64 SW |
689 | CEPH_CAP_FILE_BUFFER| |
690 | CEPH_CAP_AUTH_EXCL| | |
691 | CEPH_CAP_XATTR_EXCL)) { | |
ffdeec7a | 692 | if (ci->i_version == 0 || |
9bbeab41 | 693 | timespec64_compare(ctime, &inode->i_ctime) > 0) { |
13442b03 | 694 | dout("ctime %lld.%09ld -> %lld.%09ld inc w/ cap\n", |
9bbeab41 AB |
695 | inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec, |
696 | ctime->tv_sec, ctime->tv_nsec); | |
697 | inode->i_ctime = *ctime; | |
355da1eb | 698 | } |
ffdeec7a YZ |
699 | if (ci->i_version == 0 || |
700 | ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) { | |
355da1eb | 701 | /* the MDS did a utimes() */ |
13442b03 | 702 | dout("mtime %lld.%09ld -> %lld.%09ld " |
355da1eb | 703 | "tw %d -> %d\n", |
9bbeab41 AB |
704 | inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec, |
705 | mtime->tv_sec, mtime->tv_nsec, | |
355da1eb SW |
706 | ci->i_time_warp_seq, (int)time_warp_seq); |
707 | ||
9bbeab41 AB |
708 | inode->i_mtime = *mtime; |
709 | inode->i_atime = *atime; | |
355da1eb SW |
710 | ci->i_time_warp_seq = time_warp_seq; |
711 | } else if (time_warp_seq == ci->i_time_warp_seq) { | |
712 | /* nobody did utimes(); take the max */ | |
9bbeab41 | 713 | if (timespec64_compare(mtime, &inode->i_mtime) > 0) { |
13442b03 | 714 | dout("mtime %lld.%09ld -> %lld.%09ld inc\n", |
9bbeab41 | 715 | inode->i_mtime.tv_sec, |
355da1eb | 716 | inode->i_mtime.tv_nsec, |
9bbeab41 AB |
717 | mtime->tv_sec, mtime->tv_nsec); |
718 | inode->i_mtime = *mtime; | |
355da1eb | 719 | } |
9bbeab41 | 720 | if (timespec64_compare(atime, &inode->i_atime) > 0) { |
13442b03 | 721 | dout("atime %lld.%09ld -> %lld.%09ld inc\n", |
9bbeab41 | 722 | inode->i_atime.tv_sec, |
355da1eb | 723 | inode->i_atime.tv_nsec, |
9bbeab41 AB |
724 | atime->tv_sec, atime->tv_nsec); |
725 | inode->i_atime = *atime; | |
355da1eb SW |
726 | } |
727 | } else if (issued & CEPH_CAP_FILE_EXCL) { | |
728 | /* we did a utimes(); ignore mds values */ | |
729 | } else { | |
730 | warn = 1; | |
731 | } | |
732 | } else { | |
d8672d64 | 733 | /* we have no write|excl caps; whatever the MDS says is true */ |
355da1eb | 734 | if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) { |
9bbeab41 AB |
735 | inode->i_ctime = *ctime; |
736 | inode->i_mtime = *mtime; | |
737 | inode->i_atime = *atime; | |
355da1eb SW |
738 | ci->i_time_warp_seq = time_warp_seq; |
739 | } else { | |
740 | warn = 1; | |
741 | } | |
742 | } | |
743 | if (warn) /* time_warp_seq shouldn't go backwards */ | |
744 | dout("%p mds time_warp_seq %llu < %u\n", | |
745 | inode, time_warp_seq, ci->i_time_warp_seq); | |
746 | } | |
747 | ||
748 | /* | |
749 | * Populate an inode based on info from mds. May be called on new or | |
750 | * existing inodes. | |
751 | */ | |
966c7160 JL |
752 | int ceph_fill_inode(struct inode *inode, struct page *locked_page, |
753 | struct ceph_mds_reply_info_in *iinfo, | |
754 | struct ceph_mds_reply_dirfrag *dirinfo, | |
755 | struct ceph_mds_session *session, int cap_fmode, | |
756 | struct ceph_cap_reservation *caps_reservation) | |
355da1eb | 757 | { |
2678da88 | 758 | struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb); |
355da1eb SW |
759 | struct ceph_mds_reply_inode *info = iinfo->in; |
760 | struct ceph_inode_info *ci = ceph_inode(inode); | |
2af54a72 | 761 | int issued, new_issued, info_caps; |
9bbeab41 | 762 | struct timespec64 mtime, atime, ctime; |
355da1eb | 763 | struct ceph_buffer *xattr_blob = NULL; |
af8a85a4 | 764 | struct ceph_buffer *old_blob = NULL; |
779fe0fb | 765 | struct ceph_string *pool_ns = NULL; |
d9df2783 | 766 | struct ceph_cap *new_cap = NULL; |
355da1eb | 767 | int err = 0; |
d9df2783 | 768 | bool wake = false; |
f98a128a YZ |
769 | bool queue_trunc = false; |
770 | bool new_version = false; | |
31c542a1 | 771 | bool fill_inline = false; |
ed94f87c JL |
772 | umode_t mode = le32_to_cpu(info->mode); |
773 | dev_t rdev = le32_to_cpu(info->rdev); | |
355da1eb | 774 | |
27171ae6 JL |
775 | lockdep_assert_held(&mdsc->snap_rwsem); |
776 | ||
966c7160 | 777 | dout("%s %p ino %llx.%llx v %llu had %llu\n", __func__, |
355da1eb SW |
778 | inode, ceph_vinop(inode), le64_to_cpu(info->version), |
779 | ci->i_version); | |
780 | ||
ed94f87c JL |
781 | /* Once I_NEW is cleared, we can't change type or dev numbers */ |
782 | if (inode->i_state & I_NEW) { | |
783 | inode->i_mode = mode; | |
784 | } else { | |
785 | if (inode_wrong_type(inode, mode)) { | |
786 | pr_warn_once("inode type changed! (ino %llx.%llx is 0%o, mds says 0%o)\n", | |
787 | ceph_vinop(inode), inode->i_mode, mode); | |
788 | return -ESTALE; | |
789 | } | |
790 | ||
791 | if ((S_ISCHR(mode) || S_ISBLK(mode)) && inode->i_rdev != rdev) { | |
792 | pr_warn_once("dev inode rdev changed! (ino %llx.%llx is %u:%u, mds says %u:%u)\n", | |
793 | ceph_vinop(inode), MAJOR(inode->i_rdev), | |
794 | MINOR(inode->i_rdev), MAJOR(rdev), | |
795 | MINOR(rdev)); | |
796 | return -ESTALE; | |
797 | } | |
798 | } | |
799 | ||
2af54a72 YZ |
800 | info_caps = le32_to_cpu(info->cap.caps); |
801 | ||
d9df2783 | 802 | /* prealloc new cap struct */ |
9a6bed4f | 803 | if (info_caps && ceph_snap(inode) == CEPH_NOSNAP) { |
d9df2783 | 804 | new_cap = ceph_get_cap(mdsc, caps_reservation); |
9a6bed4f JL |
805 | if (!new_cap) |
806 | return -ENOMEM; | |
807 | } | |
d9df2783 | 808 | |
355da1eb SW |
809 | /* |
810 | * prealloc xattr data, if it looks like we'll need it. only | |
811 | * if len > 4 (meaning there are actually xattrs; the first 4 | |
812 | * bytes are the xattr count). | |
813 | */ | |
814 | if (iinfo->xattr_len > 4) { | |
b6c1d5b8 | 815 | xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS); |
355da1eb | 816 | if (!xattr_blob) |
966c7160 | 817 | pr_err("%s ENOMEM xattr blob %d bytes\n", __func__, |
355da1eb SW |
818 | iinfo->xattr_len); |
819 | } | |
820 | ||
779fe0fb YZ |
821 | if (iinfo->pool_ns_len > 0) |
822 | pool_ns = ceph_find_or_create_string(iinfo->pool_ns_data, | |
823 | iinfo->pool_ns_len); | |
824 | ||
75c9627e YZ |
825 | if (ceph_snap(inode) != CEPH_NOSNAP && !ci->i_snapid_map) |
826 | ci->i_snapid_map = ceph_get_snapid_map(mdsc, ceph_snap(inode)); | |
827 | ||
be655596 | 828 | spin_lock(&ci->i_ceph_lock); |
355da1eb SW |
829 | |
830 | /* | |
831 | * provided version will be odd if inode value is projected, | |
8bd59e01 SW |
832 | * even if stable. skip the update if we have newer stable |
833 | * info (ours>=theirs, e.g. due to racing mds replies), unless | |
834 | * we are getting projected (unstable) info (in which case the | |
835 | * version is odd, and we want ours>theirs). | |
836 | * us them | |
837 | * 2 2 skip | |
838 | * 3 2 skip | |
839 | * 3 3 update | |
355da1eb | 840 | */ |
f98a128a YZ |
841 | if (ci->i_version == 0 || |
842 | ((info->cap.flags & CEPH_CAP_FLAG_AUTH) && | |
843 | le64_to_cpu(info->version) > (ci->i_version & ~1))) | |
844 | new_version = true; | |
845 | ||
a35ead31 JL |
846 | /* Update change_attribute */ |
847 | inode_set_max_iversion_raw(inode, iinfo->change_attr); | |
848 | ||
2af54a72 YZ |
849 | __ceph_caps_issued(ci, &issued); |
850 | issued |= __ceph_caps_dirty(ci); | |
851 | new_issued = ~issued & info_caps; | |
355da1eb | 852 | |
75067034 LH |
853 | /* directories have fl_stripe_unit set to zero */ |
854 | if (le32_to_cpu(info->layout.fl_stripe_unit)) | |
855 | inode->i_blkbits = | |
856 | fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1; | |
857 | else | |
858 | inode->i_blkbits = CEPH_BLOCK_SHIFT; | |
355da1eb | 859 | |
d557c48d | 860 | __ceph_update_quota(ci, iinfo->max_bytes, iinfo->max_files); |
fb18a575 | 861 | |
f98a128a YZ |
862 | if ((new_version || (new_issued & CEPH_CAP_AUTH_SHARED)) && |
863 | (issued & CEPH_CAP_AUTH_EXCL) == 0) { | |
ed94f87c | 864 | inode->i_mode = mode; |
ab871b90 EB |
865 | inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(info->uid)); |
866 | inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(info->gid)); | |
355da1eb | 867 | dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode, |
bd2bae6a EB |
868 | from_kuid(&init_user_ns, inode->i_uid), |
869 | from_kgid(&init_user_ns, inode->i_gid)); | |
245ce991 | 870 | ceph_decode_timespec64(&ci->i_btime, &iinfo->btime); |
193e7b37 | 871 | ceph_decode_timespec64(&ci->i_snap_btime, &iinfo->snap_btime); |
355da1eb SW |
872 | } |
873 | ||
f98a128a YZ |
874 | if ((new_version || (new_issued & CEPH_CAP_LINK_SHARED)) && |
875 | (issued & CEPH_CAP_LINK_EXCL) == 0) | |
bfe86848 | 876 | set_nlink(inode, le32_to_cpu(info->nlink)); |
355da1eb | 877 | |
f98a128a YZ |
878 | if (new_version || (new_issued & CEPH_CAP_ANY_RD)) { |
879 | /* be careful with mtime, atime, size */ | |
9bbeab41 AB |
880 | ceph_decode_timespec64(&atime, &info->atime); |
881 | ceph_decode_timespec64(&mtime, &info->mtime); | |
882 | ceph_decode_timespec64(&ctime, &info->ctime); | |
f98a128a YZ |
883 | ceph_fill_file_time(inode, issued, |
884 | le32_to_cpu(info->time_warp_seq), | |
885 | &ctime, &mtime, &atime); | |
886 | } | |
887 | ||
2af54a72 YZ |
888 | if (new_version || (info_caps & CEPH_CAP_FILE_SHARED)) { |
889 | ci->i_files = le64_to_cpu(info->files); | |
890 | ci->i_subdirs = le64_to_cpu(info->subdirs); | |
891 | } | |
892 | ||
f98a128a YZ |
893 | if (new_version || |
894 | (new_issued & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR))) { | |
7627151e | 895 | s64 old_pool = ci->i_layout.pool_id; |
779fe0fb YZ |
896 | struct ceph_string *old_ns; |
897 | ||
7627151e | 898 | ceph_file_layout_from_legacy(&ci->i_layout, &info->layout); |
779fe0fb YZ |
899 | old_ns = rcu_dereference_protected(ci->i_layout.pool_ns, |
900 | lockdep_is_held(&ci->i_ceph_lock)); | |
901 | rcu_assign_pointer(ci->i_layout.pool_ns, pool_ns); | |
902 | ||
903 | if (ci->i_layout.pool_id != old_pool || pool_ns != old_ns) | |
10183a69 | 904 | ci->i_ceph_flags &= ~CEPH_I_POOL_PERM; |
10183a69 | 905 | |
779fe0fb | 906 | pool_ns = old_ns; |
10183a69 | 907 | |
f98a128a YZ |
908 | queue_trunc = ceph_fill_file_size(inode, issued, |
909 | le32_to_cpu(info->truncate_seq), | |
910 | le64_to_cpu(info->truncate_size), | |
911 | le64_to_cpu(info->size)); | |
912 | /* only update max_size on auth cap */ | |
913 | if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) && | |
914 | ci->i_max_size != le64_to_cpu(info->max_size)) { | |
915 | dout("max_size %lld -> %llu\n", ci->i_max_size, | |
916 | le64_to_cpu(info->max_size)); | |
917 | ci->i_max_size = le64_to_cpu(info->max_size); | |
918 | } | |
919 | } | |
355da1eb | 920 | |
49a9f4f6 YZ |
921 | /* layout and rstat are not tracked by capability, update them if |
922 | * the inode info is from auth mds */ | |
923 | if (new_version || (info->cap.flags & CEPH_CAP_FLAG_AUTH)) { | |
924 | if (S_ISDIR(inode->i_mode)) { | |
925 | ci->i_dir_layout = iinfo->dir_layout; | |
926 | ci->i_rbytes = le64_to_cpu(info->rbytes); | |
927 | ci->i_rfiles = le64_to_cpu(info->rfiles); | |
928 | ci->i_rsubdirs = le64_to_cpu(info->rsubdirs); | |
08796873 | 929 | ci->i_dir_pin = iinfo->dir_pin; |
e7f72952 | 930 | ci->i_rsnaps = iinfo->rsnaps; |
9bbeab41 | 931 | ceph_decode_timespec64(&ci->i_rctime, &info->rctime); |
49a9f4f6 YZ |
932 | } |
933 | } | |
934 | ||
355da1eb SW |
935 | /* xattrs */ |
936 | /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */ | |
508b32d8 | 937 | if ((ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL)) && |
355da1eb SW |
938 | le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) { |
939 | if (ci->i_xattrs.blob) | |
af8a85a4 | 940 | old_blob = ci->i_xattrs.blob; |
355da1eb SW |
941 | ci->i_xattrs.blob = xattr_blob; |
942 | if (xattr_blob) | |
943 | memcpy(ci->i_xattrs.blob->vec.iov_base, | |
944 | iinfo->xattr_data, iinfo->xattr_len); | |
945 | ci->i_xattrs.version = le64_to_cpu(info->xattr_version); | |
7221fe4c | 946 | ceph_forget_all_cached_acls(inode); |
ac6713cc | 947 | ceph_security_invalidate_secctx(inode); |
a6424e48 | 948 | xattr_blob = NULL; |
355da1eb SW |
949 | } |
950 | ||
ffdeec7a | 951 | /* finally update i_version */ |
aae1a442 YZ |
952 | if (le64_to_cpu(info->version) > ci->i_version) |
953 | ci->i_version = le64_to_cpu(info->version); | |
ffdeec7a | 954 | |
355da1eb | 955 | inode->i_mapping->a_ops = &ceph_aops; |
355da1eb SW |
956 | |
957 | switch (inode->i_mode & S_IFMT) { | |
958 | case S_IFIFO: | |
959 | case S_IFBLK: | |
960 | case S_IFCHR: | |
961 | case S_IFSOCK: | |
5ba72e60 | 962 | inode->i_blkbits = PAGE_SHIFT; |
ed94f87c | 963 | init_special_inode(inode, inode->i_mode, rdev); |
355da1eb SW |
964 | inode->i_op = &ceph_file_iops; |
965 | break; | |
966 | case S_IFREG: | |
967 | inode->i_op = &ceph_file_iops; | |
968 | inode->i_fop = &ceph_file_fops; | |
969 | break; | |
970 | case S_IFLNK: | |
971 | inode->i_op = &ceph_symlink_iops; | |
972 | if (!ci->i_symlink) { | |
810339ec | 973 | u32 symlen = iinfo->symlink_len; |
355da1eb SW |
974 | char *sym; |
975 | ||
be655596 | 976 | spin_unlock(&ci->i_ceph_lock); |
355da1eb | 977 | |
224a7542 | 978 | if (symlen != i_size_read(inode)) { |
966c7160 JL |
979 | pr_err("%s %llx.%llx BAD symlink " |
980 | "size %lld\n", __func__, | |
981 | ceph_vinop(inode), | |
224a7542 YZ |
982 | i_size_read(inode)); |
983 | i_size_write(inode, symlen); | |
984 | inode->i_blocks = calc_inode_blocks(symlen); | |
985 | } | |
810339ec | 986 | |
355da1eb | 987 | err = -ENOMEM; |
810339ec | 988 | sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS); |
355da1eb SW |
989 | if (!sym) |
990 | goto out; | |
355da1eb | 991 | |
be655596 | 992 | spin_lock(&ci->i_ceph_lock); |
355da1eb SW |
993 | if (!ci->i_symlink) |
994 | ci->i_symlink = sym; | |
995 | else | |
996 | kfree(sym); /* lost a race */ | |
997 | } | |
ac194dcc | 998 | inode->i_link = ci->i_symlink; |
355da1eb SW |
999 | break; |
1000 | case S_IFDIR: | |
1001 | inode->i_op = &ceph_dir_iops; | |
1002 | inode->i_fop = &ceph_dir_fops; | |
355da1eb SW |
1003 | break; |
1004 | default: | |
966c7160 | 1005 | pr_err("%s %llx.%llx BAD mode 0%o\n", __func__, |
355da1eb SW |
1006 | ceph_vinop(inode), inode->i_mode); |
1007 | } | |
1008 | ||
355da1eb | 1009 | /* were we issued a capability? */ |
2af54a72 | 1010 | if (info_caps) { |
355da1eb SW |
1011 | if (ceph_snap(inode) == CEPH_NOSNAP) { |
1012 | ceph_add_cap(inode, session, | |
1013 | le64_to_cpu(info->cap.cap_id), | |
135e671e | 1014 | info_caps, |
355da1eb SW |
1015 | le32_to_cpu(info->cap.wanted), |
1016 | le32_to_cpu(info->cap.seq), | |
1017 | le32_to_cpu(info->cap.mseq), | |
1018 | le64_to_cpu(info->cap.realm), | |
d9df2783 | 1019 | info->cap.flags, &new_cap); |
2f92b3d0 YZ |
1020 | |
1021 | /* set dir completion flag? */ | |
1022 | if (S_ISDIR(inode->i_mode) && | |
1023 | ci->i_files == 0 && ci->i_subdirs == 0 && | |
2af54a72 | 1024 | (info_caps & CEPH_CAP_FILE_SHARED) && |
2f92b3d0 YZ |
1025 | (issued & CEPH_CAP_FILE_EXCL) == 0 && |
1026 | !__ceph_dir_is_complete(ci)) { | |
1027 | dout(" marking %p complete (empty)\n", inode); | |
fdd4e158 | 1028 | i_size_write(inode, 0); |
2f92b3d0 | 1029 | __ceph_dir_set_complete(ci, |
fdd4e158 YZ |
1030 | atomic64_read(&ci->i_release_count), |
1031 | atomic64_read(&ci->i_ordered_count)); | |
2f92b3d0 YZ |
1032 | } |
1033 | ||
d9df2783 | 1034 | wake = true; |
355da1eb | 1035 | } else { |
355da1eb | 1036 | dout(" %p got snap_caps %s\n", inode, |
2af54a72 YZ |
1037 | ceph_cap_string(info_caps)); |
1038 | ci->i_snap_caps |= info_caps; | |
355da1eb SW |
1039 | } |
1040 | } | |
31c542a1 YZ |
1041 | |
1042 | if (iinfo->inline_version > 0 && | |
1043 | iinfo->inline_version >= ci->i_inline_version) { | |
1044 | int cache_caps = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO; | |
1045 | ci->i_inline_version = iinfo->inline_version; | |
1046 | if (ci->i_inline_version != CEPH_INLINE_NONE && | |
2af54a72 | 1047 | (locked_page || (info_caps & cache_caps))) |
31c542a1 YZ |
1048 | fill_inline = true; |
1049 | } | |
1050 | ||
719a2514 YZ |
1051 | if (cap_fmode >= 0) { |
1052 | if (!info_caps) | |
1053 | pr_warn("mds issued no caps on %llx.%llx\n", | |
1054 | ceph_vinop(inode)); | |
1055 | __ceph_touch_fmode(ci, mdsc, cap_fmode); | |
1056 | } | |
1057 | ||
be655596 | 1058 | spin_unlock(&ci->i_ceph_lock); |
355da1eb | 1059 | |
400e1286 JL |
1060 | ceph_fscache_register_inode_cookie(inode); |
1061 | ||
31c542a1 | 1062 | if (fill_inline) |
01deead0 | 1063 | ceph_fill_inline_data(inode, locked_page, |
31c542a1 YZ |
1064 | iinfo->inline_data, iinfo->inline_len); |
1065 | ||
d9df2783 YZ |
1066 | if (wake) |
1067 | wake_up_all(&ci->i_cap_wq); | |
1068 | ||
355da1eb SW |
1069 | /* queue truncate if we saw i_size decrease */ |
1070 | if (queue_trunc) | |
3c6f6b79 | 1071 | ceph_queue_vmtruncate(inode); |
355da1eb SW |
1072 | |
1073 | /* populate frag tree */ | |
3e7fbe9c YZ |
1074 | if (S_ISDIR(inode->i_mode)) |
1075 | ceph_fill_fragtree(inode, &info->fragtree, dirinfo); | |
355da1eb SW |
1076 | |
1077 | /* update delegation info? */ | |
1078 | if (dirinfo) | |
1079 | ceph_fill_dirfrag(inode, dirinfo); | |
1080 | ||
1081 | err = 0; | |
355da1eb | 1082 | out: |
d9df2783 YZ |
1083 | if (new_cap) |
1084 | ceph_put_cap(mdsc, new_cap); | |
af8a85a4 LH |
1085 | ceph_buffer_put(old_blob); |
1086 | ceph_buffer_put(xattr_blob); | |
779fe0fb | 1087 | ceph_put_string(pool_ns); |
355da1eb SW |
1088 | return err; |
1089 | } | |
1090 | ||
1091 | /* | |
543212b3 | 1092 | * caller should hold session s_mutex and dentry->d_lock. |
355da1eb | 1093 | */ |
543212b3 YZ |
1094 | static void __update_dentry_lease(struct inode *dir, struct dentry *dentry, |
1095 | struct ceph_mds_reply_lease *lease, | |
1096 | struct ceph_mds_session *session, | |
1097 | unsigned long from_time, | |
1098 | struct ceph_mds_session **old_lease_session) | |
355da1eb SW |
1099 | { |
1100 | struct ceph_dentry_info *di = ceph_dentry(dentry); | |
f5e17aed | 1101 | unsigned mask = le16_to_cpu(lease->mask); |
355da1eb SW |
1102 | long unsigned duration = le32_to_cpu(lease->duration_ms); |
1103 | long unsigned ttl = from_time + (duration * HZ) / 1000; | |
1104 | long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000; | |
f5d55f03 | 1105 | |
2f90b852 SW |
1106 | dout("update_dentry_lease %p duration %lu ms ttl %lu\n", |
1107 | dentry, duration, ttl); | |
355da1eb | 1108 | |
18fc8abd AV |
1109 | /* only track leases on regular dentries */ |
1110 | if (ceph_snap(dir) != CEPH_NOSNAP) | |
543212b3 | 1111 | return; |
18fc8abd | 1112 | |
f5e17aed JL |
1113 | if (mask & CEPH_LEASE_PRIMARY_LINK) |
1114 | di->flags |= CEPH_DENTRY_PRIMARY_LINK; | |
1115 | else | |
1116 | di->flags &= ~CEPH_DENTRY_PRIMARY_LINK; | |
1117 | ||
97aeb6bf | 1118 | di->lease_shared_gen = atomic_read(&ceph_inode(dir)->i_shared_gen); |
f5e17aed | 1119 | if (!(mask & CEPH_LEASE_VALID)) { |
37c4efc1 | 1120 | __ceph_dentry_dir_lease_touch(di); |
543212b3 | 1121 | return; |
37c4efc1 | 1122 | } |
355da1eb | 1123 | |
52d60f8e | 1124 | if (di->lease_gen == atomic_read(&session->s_cap_gen) && |
9b16f03c | 1125 | time_before(ttl, di->time)) |
543212b3 | 1126 | return; /* we already have a newer lease. */ |
355da1eb | 1127 | |
481f001f | 1128 | if (di->lease_session && di->lease_session != session) { |
543212b3 | 1129 | *old_lease_session = di->lease_session; |
481f001f YZ |
1130 | di->lease_session = NULL; |
1131 | } | |
355da1eb | 1132 | |
355da1eb SW |
1133 | if (!di->lease_session) |
1134 | di->lease_session = ceph_get_mds_session(session); | |
52d60f8e | 1135 | di->lease_gen = atomic_read(&session->s_cap_gen); |
355da1eb SW |
1136 | di->lease_seq = le32_to_cpu(lease->seq); |
1137 | di->lease_renew_after = half_ttl; | |
1138 | di->lease_renew_from = 0; | |
9b16f03c | 1139 | di->time = ttl; |
37c4efc1 YZ |
1140 | |
1141 | __ceph_dentry_lease_touch(di); | |
543212b3 YZ |
1142 | } |
1143 | ||
1144 | static inline void update_dentry_lease(struct inode *dir, struct dentry *dentry, | |
1145 | struct ceph_mds_reply_lease *lease, | |
1146 | struct ceph_mds_session *session, | |
1147 | unsigned long from_time) | |
1148 | { | |
1149 | struct ceph_mds_session *old_lease_session = NULL; | |
1150 | spin_lock(&dentry->d_lock); | |
1151 | __update_dentry_lease(dir, dentry, lease, session, from_time, | |
1152 | &old_lease_session); | |
1153 | spin_unlock(&dentry->d_lock); | |
7e65624d | 1154 | ceph_put_mds_session(old_lease_session); |
543212b3 YZ |
1155 | } |
1156 | ||
1157 | /* | |
1158 | * update dentry lease without having parent inode locked | |
1159 | */ | |
1160 | static void update_dentry_lease_careful(struct dentry *dentry, | |
1161 | struct ceph_mds_reply_lease *lease, | |
1162 | struct ceph_mds_session *session, | |
1163 | unsigned long from_time, | |
1164 | char *dname, u32 dname_len, | |
1165 | struct ceph_vino *pdvino, | |
1166 | struct ceph_vino *ptvino) | |
1167 | ||
1168 | { | |
1169 | struct inode *dir; | |
1170 | struct ceph_mds_session *old_lease_session = NULL; | |
1171 | ||
1172 | spin_lock(&dentry->d_lock); | |
1173 | /* make sure dentry's name matches target */ | |
1174 | if (dentry->d_name.len != dname_len || | |
1175 | memcmp(dentry->d_name.name, dname, dname_len)) | |
1176 | goto out_unlock; | |
1177 | ||
1178 | dir = d_inode(dentry->d_parent); | |
1179 | /* make sure parent matches dvino */ | |
1180 | if (!ceph_ino_compare(dir, pdvino)) | |
1181 | goto out_unlock; | |
1182 | ||
1183 | /* make sure dentry's inode matches target. NULL ptvino means that | |
1184 | * we expect a negative dentry */ | |
1185 | if (ptvino) { | |
1186 | if (d_really_is_negative(dentry)) | |
1187 | goto out_unlock; | |
1188 | if (!ceph_ino_compare(d_inode(dentry), ptvino)) | |
1189 | goto out_unlock; | |
1190 | } else { | |
1191 | if (d_really_is_positive(dentry)) | |
1192 | goto out_unlock; | |
1193 | } | |
1194 | ||
1195 | __update_dentry_lease(dir, dentry, lease, session, | |
1196 | from_time, &old_lease_session); | |
355da1eb SW |
1197 | out_unlock: |
1198 | spin_unlock(&dentry->d_lock); | |
7e65624d | 1199 | ceph_put_mds_session(old_lease_session); |
355da1eb SW |
1200 | } |
1201 | ||
1202 | /* | |
1203 | * splice a dentry to an inode. | |
1204 | * caller must hold directory i_mutex for this to be safe. | |
355da1eb | 1205 | */ |
2bf996ac | 1206 | static int splice_dentry(struct dentry **pdn, struct inode *in) |
355da1eb | 1207 | { |
2bf996ac | 1208 | struct dentry *dn = *pdn; |
355da1eb SW |
1209 | struct dentry *realdn; |
1210 | ||
2b0143b5 | 1211 | BUG_ON(d_inode(dn)); |
1cd3935b | 1212 | |
5495c2d0 YZ |
1213 | if (S_ISDIR(in->i_mode)) { |
1214 | /* If inode is directory, d_splice_alias() below will remove | |
1215 | * 'realdn' from its origin parent. We need to ensure that | |
1216 | * origin parent's readdir cache will not reference 'realdn' | |
1217 | */ | |
1218 | realdn = d_find_any_alias(in); | |
1219 | if (realdn) { | |
1220 | struct ceph_dentry_info *di = ceph_dentry(realdn); | |
1221 | spin_lock(&realdn->d_lock); | |
1222 | ||
1223 | realdn->d_op->d_prune(realdn); | |
1224 | ||
1225 | di->time = jiffies; | |
1226 | di->lease_shared_gen = 0; | |
1227 | di->offset = 0; | |
1228 | ||
1229 | spin_unlock(&realdn->d_lock); | |
1230 | dput(realdn); | |
1231 | } | |
1232 | } | |
1233 | ||
355da1eb SW |
1234 | /* dn must be unhashed */ |
1235 | if (!d_unhashed(dn)) | |
1236 | d_drop(dn); | |
41d28bca | 1237 | realdn = d_splice_alias(in, dn); |
355da1eb | 1238 | if (IS_ERR(realdn)) { |
d69ed05a SW |
1239 | pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n", |
1240 | PTR_ERR(realdn), dn, in, ceph_vinop(in)); | |
2bf996ac YZ |
1241 | return PTR_ERR(realdn); |
1242 | } | |
1243 | ||
1244 | if (realdn) { | |
355da1eb SW |
1245 | dout("dn %p (%d) spliced with %p (%d) " |
1246 | "inode %p ino %llx.%llx\n", | |
84d08fa8 AV |
1247 | dn, d_count(dn), |
1248 | realdn, d_count(realdn), | |
2b0143b5 | 1249 | d_inode(realdn), ceph_vinop(d_inode(realdn))); |
355da1eb | 1250 | dput(dn); |
2bf996ac | 1251 | *pdn = realdn; |
355da1eb SW |
1252 | } else { |
1253 | BUG_ON(!ceph_dentry(dn)); | |
355da1eb | 1254 | dout("dn %p attached to %p ino %llx.%llx\n", |
2b0143b5 | 1255 | dn, d_inode(dn), ceph_vinop(d_inode(dn))); |
355da1eb | 1256 | } |
2bf996ac | 1257 | return 0; |
355da1eb SW |
1258 | } |
1259 | ||
1260 | /* | |
1261 | * Incorporate results into the local cache. This is either just | |
1262 | * one inode, or a directory, dentry, and possibly linked-to inode (e.g., | |
1263 | * after a lookup). | |
1264 | * | |
1265 | * A reply may contain | |
1266 | * a directory inode along with a dentry. | |
1267 | * and/or a target inode | |
1268 | * | |
1269 | * Called with snap_rwsem (read). | |
1270 | */ | |
f5a03b08 | 1271 | int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req) |
355da1eb | 1272 | { |
f5a03b08 | 1273 | struct ceph_mds_session *session = req->r_session; |
355da1eb SW |
1274 | struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info; |
1275 | struct inode *in = NULL; | |
f5d55f03 | 1276 | struct ceph_vino tvino, dvino; |
3d14c5d2 | 1277 | struct ceph_fs_client *fsc = ceph_sb_to_client(sb); |
355da1eb SW |
1278 | int err = 0; |
1279 | ||
1280 | dout("fill_trace %p is_dentry %d is_target %d\n", req, | |
1281 | rinfo->head->is_dentry, rinfo->head->is_target); | |
1282 | ||
355da1eb SW |
1283 | if (!rinfo->head->is_target && !rinfo->head->is_dentry) { |
1284 | dout("fill_trace reply is empty!\n"); | |
3dd69aab | 1285 | if (rinfo->head->result == 0 && req->r_parent) |
167c9e35 | 1286 | ceph_invalidate_dir_request(req); |
355da1eb SW |
1287 | return 0; |
1288 | } | |
1289 | ||
1290 | if (rinfo->head->is_dentry) { | |
3dd69aab | 1291 | struct inode *dir = req->r_parent; |
5b1daecd | 1292 | |
6c5e50fa | 1293 | if (dir) { |
966c7160 JL |
1294 | err = ceph_fill_inode(dir, NULL, &rinfo->diri, |
1295 | rinfo->dirfrag, session, -1, | |
1296 | &req->r_caps_reservation); | |
6c5e50fa | 1297 | if (err < 0) |
19913b4e | 1298 | goto done; |
6c5e50fa SW |
1299 | } else { |
1300 | WARN_ON_ONCE(1); | |
1301 | } | |
19913b4e | 1302 | |
74c9e6bf YZ |
1303 | if (dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME && |
1304 | test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) && | |
1305 | !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) { | |
19913b4e YZ |
1306 | struct qstr dname; |
1307 | struct dentry *dn, *parent; | |
1308 | ||
1309 | BUG_ON(!rinfo->head->is_target); | |
1310 | BUG_ON(req->r_dentry); | |
1311 | ||
1312 | parent = d_find_any_alias(dir); | |
1313 | BUG_ON(!parent); | |
1314 | ||
1315 | dname.name = rinfo->dname; | |
1316 | dname.len = rinfo->dname_len; | |
8387ff25 | 1317 | dname.hash = full_name_hash(parent, dname.name, dname.len); |
f5d55f03 JL |
1318 | tvino.ino = le64_to_cpu(rinfo->targeti.in->ino); |
1319 | tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid); | |
19913b4e YZ |
1320 | retry_lookup: |
1321 | dn = d_lookup(parent, &dname); | |
1322 | dout("d_lookup on parent=%p name=%.*s got %p\n", | |
1323 | parent, dname.len, dname.name, dn); | |
1324 | ||
1325 | if (!dn) { | |
1326 | dn = d_alloc(parent, &dname); | |
1327 | dout("d_alloc %p '%.*s' = %p\n", parent, | |
1328 | dname.len, dname.name, dn); | |
d37b1d99 | 1329 | if (!dn) { |
19913b4e YZ |
1330 | dput(parent); |
1331 | err = -ENOMEM; | |
1332 | goto done; | |
1333 | } | |
ad5cb123 | 1334 | err = 0; |
2b0143b5 | 1335 | } else if (d_really_is_positive(dn) && |
f5d55f03 JL |
1336 | (ceph_ino(d_inode(dn)) != tvino.ino || |
1337 | ceph_snap(d_inode(dn)) != tvino.snap)) { | |
19913b4e | 1338 | dout(" dn %p points to wrong inode %p\n", |
2b0143b5 | 1339 | dn, d_inode(dn)); |
933ad2c9 | 1340 | ceph_dir_clear_ordered(dir); |
19913b4e YZ |
1341 | d_delete(dn); |
1342 | dput(dn); | |
1343 | goto retry_lookup; | |
1344 | } | |
1345 | ||
1346 | req->r_dentry = dn; | |
1347 | dput(parent); | |
1348 | } | |
5b1daecd SW |
1349 | } |
1350 | ||
86b58d13 | 1351 | if (rinfo->head->is_target) { |
bca9fc14 JL |
1352 | /* Should be filled in by handle_reply */ |
1353 | BUG_ON(!req->r_target_inode); | |
86b58d13 | 1354 | |
bca9fc14 | 1355 | in = req->r_target_inode; |
966c7160 JL |
1356 | err = ceph_fill_inode(in, req->r_locked_page, &rinfo->targeti, |
1357 | NULL, session, | |
bc2de10d | 1358 | (!test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) && |
3bb48b41 | 1359 | !test_bit(CEPH_MDS_R_ASYNC, &req->r_req_flags) && |
57c21994 | 1360 | rinfo->head->result == 0) ? req->r_fmode : -1, |
86b58d13 YZ |
1361 | &req->r_caps_reservation); |
1362 | if (err < 0) { | |
966c7160 | 1363 | pr_err("ceph_fill_inode badness %p %llx.%llx\n", |
86b58d13 | 1364 | in, ceph_vinop(in)); |
bca9fc14 | 1365 | req->r_target_inode = NULL; |
893e456b JL |
1366 | if (in->i_state & I_NEW) |
1367 | discard_new_inode(in); | |
68cbb805 JL |
1368 | else |
1369 | iput(in); | |
86b58d13 YZ |
1370 | goto done; |
1371 | } | |
893e456b JL |
1372 | if (in->i_state & I_NEW) |
1373 | unlock_new_inode(in); | |
86b58d13 YZ |
1374 | } |
1375 | ||
9358c6d4 SW |
1376 | /* |
1377 | * ignore null lease/binding on snapdir ENOENT, or else we | |
1378 | * will have trouble splicing in the virtual snapdir later | |
1379 | */ | |
3dd69aab JL |
1380 | if (rinfo->head->is_dentry && |
1381 | !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) && | |
1382 | test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) && | |
9358c6d4 | 1383 | (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name, |
3d14c5d2 | 1384 | fsc->mount_options->snapdir_name, |
9358c6d4 | 1385 | req->r_dentry->d_name.len))) { |
355da1eb SW |
1386 | /* |
1387 | * lookup link rename : null -> possibly existing inode | |
1388 | * mknod symlink mkdir : null -> new inode | |
1389 | * unlink : linked -> null | |
1390 | */ | |
3dd69aab | 1391 | struct inode *dir = req->r_parent; |
355da1eb SW |
1392 | struct dentry *dn = req->r_dentry; |
1393 | bool have_dir_cap, have_lease; | |
1394 | ||
1395 | BUG_ON(!dn); | |
1396 | BUG_ON(!dir); | |
2b0143b5 | 1397 | BUG_ON(d_inode(dn->d_parent) != dir); |
f5d55f03 JL |
1398 | |
1399 | dvino.ino = le64_to_cpu(rinfo->diri.in->ino); | |
1400 | dvino.snap = le64_to_cpu(rinfo->diri.in->snapid); | |
1401 | ||
1402 | BUG_ON(ceph_ino(dir) != dvino.ino); | |
1403 | BUG_ON(ceph_snap(dir) != dvino.snap); | |
355da1eb | 1404 | |
355da1eb SW |
1405 | /* do we have a lease on the whole dir? */ |
1406 | have_dir_cap = | |
1407 | (le32_to_cpu(rinfo->diri.in->cap.caps) & | |
1408 | CEPH_CAP_FILE_SHARED); | |
1409 | ||
1410 | /* do we have a dn lease? */ | |
1411 | have_lease = have_dir_cap || | |
2f90b852 | 1412 | le32_to_cpu(rinfo->dlease->duration_ms); |
355da1eb SW |
1413 | if (!have_lease) |
1414 | dout("fill_trace no dentry lease or dir cap\n"); | |
1415 | ||
1416 | /* rename? */ | |
1417 | if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) { | |
0a8a70f9 YZ |
1418 | struct inode *olddir = req->r_old_dentry_dir; |
1419 | BUG_ON(!olddir); | |
1420 | ||
a455589f | 1421 | dout(" src %p '%pd' dst %p '%pd'\n", |
355da1eb | 1422 | req->r_old_dentry, |
a455589f AV |
1423 | req->r_old_dentry, |
1424 | dn, dn); | |
355da1eb SW |
1425 | dout("fill_trace doing d_move %p -> %p\n", |
1426 | req->r_old_dentry, dn); | |
c10f5e12 | 1427 | |
fdd4e158 YZ |
1428 | /* d_move screws up sibling dentries' offsets */ |
1429 | ceph_dir_clear_ordered(dir); | |
1430 | ceph_dir_clear_ordered(olddir); | |
1431 | ||
355da1eb | 1432 | d_move(req->r_old_dentry, dn); |
a455589f AV |
1433 | dout(" src %p '%pd' dst %p '%pd'\n", |
1434 | req->r_old_dentry, | |
355da1eb | 1435 | req->r_old_dentry, |
a455589f | 1436 | dn, dn); |
81a6cf2d | 1437 | |
c4a29f26 SW |
1438 | /* ensure target dentry is invalidated, despite |
1439 | rehashing bug in vfs_rename_dir */ | |
81a6cf2d SW |
1440 | ceph_invalidate_dentry_lease(dn); |
1441 | ||
99ccbd22 | 1442 | dout("dn %p gets new offset %lld\n", req->r_old_dentry, |
1cd3935b | 1443 | ceph_dentry(req->r_old_dentry)->offset); |
81a6cf2d | 1444 | |
2bf996ac YZ |
1445 | /* swap r_dentry and r_old_dentry in case that |
1446 | * splice_dentry() gets called later. This is safe | |
1447 | * because no other place will use them */ | |
1448 | req->r_dentry = req->r_old_dentry; | |
1449 | req->r_old_dentry = dn; | |
1450 | dn = req->r_dentry; | |
355da1eb SW |
1451 | } |
1452 | ||
1453 | /* null dentry? */ | |
1454 | if (!rinfo->head->is_target) { | |
1455 | dout("fill_trace null dentry\n"); | |
2b0143b5 | 1456 | if (d_really_is_positive(dn)) { |
355da1eb | 1457 | dout("d_delete %p\n", dn); |
5495c2d0 | 1458 | ceph_dir_clear_ordered(dir); |
355da1eb | 1459 | d_delete(dn); |
80d025ff JL |
1460 | } else if (have_lease) { |
1461 | if (d_unhashed(dn)) | |
f8b31710 | 1462 | d_add(dn, NULL); |
543212b3 YZ |
1463 | update_dentry_lease(dir, dn, |
1464 | rinfo->dlease, session, | |
1465 | req->r_request_started); | |
355da1eb SW |
1466 | } |
1467 | goto done; | |
1468 | } | |
1469 | ||
1470 | /* attach proper inode */ | |
2b0143b5 | 1471 | if (d_really_is_negative(dn)) { |
70db4f36 | 1472 | ceph_dir_clear_ordered(dir); |
86b58d13 | 1473 | ihold(in); |
2bf996ac YZ |
1474 | err = splice_dentry(&req->r_dentry, in); |
1475 | if (err < 0) | |
355da1eb | 1476 | goto done; |
2bf996ac | 1477 | dn = req->r_dentry; /* may have spliced */ |
2b0143b5 | 1478 | } else if (d_really_is_positive(dn) && d_inode(dn) != in) { |
355da1eb | 1479 | dout(" %p links to %p %llx.%llx, not %llx.%llx\n", |
2b0143b5 | 1480 | dn, d_inode(dn), ceph_vinop(d_inode(dn)), |
86b58d13 | 1481 | ceph_vinop(in)); |
200fd27c | 1482 | d_invalidate(dn); |
355da1eb | 1483 | have_lease = false; |
355da1eb SW |
1484 | } |
1485 | ||
f5d55f03 | 1486 | if (have_lease) { |
543212b3 YZ |
1487 | update_dentry_lease(dir, dn, |
1488 | rinfo->dlease, session, | |
1489 | req->r_request_started); | |
f5d55f03 | 1490 | } |
355da1eb | 1491 | dout(" final dn %p\n", dn); |
bc2de10d JL |
1492 | } else if ((req->r_op == CEPH_MDS_OP_LOOKUPSNAP || |
1493 | req->r_op == CEPH_MDS_OP_MKSNAP) && | |
1f08529c | 1494 | test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) && |
bc2de10d | 1495 | !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) { |
3dd69aab | 1496 | struct inode *dir = req->r_parent; |
355da1eb SW |
1497 | |
1498 | /* fill out a snapdir LOOKUPSNAP dentry */ | |
0a8a70f9 YZ |
1499 | BUG_ON(!dir); |
1500 | BUG_ON(ceph_snap(dir) != CEPH_SNAPDIR); | |
2bf996ac YZ |
1501 | BUG_ON(!req->r_dentry); |
1502 | dout(" linking snapped dir %p to dn %p\n", in, req->r_dentry); | |
70db4f36 | 1503 | ceph_dir_clear_ordered(dir); |
86b58d13 | 1504 | ihold(in); |
2bf996ac YZ |
1505 | err = splice_dentry(&req->r_dentry, in); |
1506 | if (err < 0) | |
355da1eb | 1507 | goto done; |
543212b3 YZ |
1508 | } else if (rinfo->head->is_dentry && req->r_dentry) { |
1509 | /* parent inode is not locked, be carefull */ | |
cdde7c43 | 1510 | struct ceph_vino *ptvino = NULL; |
543212b3 YZ |
1511 | dvino.ino = le64_to_cpu(rinfo->diri.in->ino); |
1512 | dvino.snap = le64_to_cpu(rinfo->diri.in->snapid); | |
1513 | if (rinfo->head->is_target) { | |
1514 | tvino.ino = le64_to_cpu(rinfo->targeti.in->ino); | |
1515 | tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid); | |
1516 | ptvino = &tvino; | |
cdde7c43 | 1517 | } |
543212b3 YZ |
1518 | update_dentry_lease_careful(req->r_dentry, rinfo->dlease, |
1519 | session, req->r_request_started, | |
1520 | rinfo->dname, rinfo->dname_len, | |
1521 | &dvino, ptvino); | |
355da1eb | 1522 | } |
355da1eb SW |
1523 | done: |
1524 | dout("fill_trace done err=%d\n", err); | |
1525 | return err; | |
1526 | } | |
1527 | ||
1528 | /* | |
1529 | * Prepopulate our cache with readdir results, leases, etc. | |
1530 | */ | |
79f9f99a SW |
1531 | static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req, |
1532 | struct ceph_mds_session *session) | |
1533 | { | |
1534 | struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info; | |
1535 | int i, err = 0; | |
1536 | ||
1537 | for (i = 0; i < rinfo->dir_nr; i++) { | |
2a5beea3 | 1538 | struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i; |
79f9f99a SW |
1539 | struct ceph_vino vino; |
1540 | struct inode *in; | |
1541 | int rc; | |
1542 | ||
2a5beea3 YZ |
1543 | vino.ino = le64_to_cpu(rde->inode.in->ino); |
1544 | vino.snap = le64_to_cpu(rde->inode.in->snapid); | |
79f9f99a SW |
1545 | |
1546 | in = ceph_get_inode(req->r_dentry->d_sb, vino); | |
1547 | if (IS_ERR(in)) { | |
1548 | err = PTR_ERR(in); | |
1549 | dout("new_inode badness got %d\n", err); | |
1550 | continue; | |
1551 | } | |
966c7160 JL |
1552 | rc = ceph_fill_inode(in, NULL, &rde->inode, NULL, session, |
1553 | -1, &req->r_caps_reservation); | |
79f9f99a | 1554 | if (rc < 0) { |
966c7160 JL |
1555 | pr_err("ceph_fill_inode badness on %p got %d\n", |
1556 | in, rc); | |
79f9f99a | 1557 | err = rc; |
893e456b JL |
1558 | if (in->i_state & I_NEW) { |
1559 | ihold(in); | |
1560 | discard_new_inode(in); | |
1561 | } | |
1562 | } else if (in->i_state & I_NEW) { | |
1563 | unlock_new_inode(in); | |
79f9f99a | 1564 | } |
893e456b | 1565 | |
23c2c76e | 1566 | iput(in); |
79f9f99a SW |
1567 | } |
1568 | ||
1569 | return err; | |
1570 | } | |
1571 | ||
fdd4e158 YZ |
1572 | void ceph_readdir_cache_release(struct ceph_readdir_cache_control *ctl) |
1573 | { | |
1574 | if (ctl->page) { | |
1575 | kunmap(ctl->page); | |
09cbfeaf | 1576 | put_page(ctl->page); |
fdd4e158 YZ |
1577 | ctl->page = NULL; |
1578 | } | |
1579 | } | |
1580 | ||
1581 | static int fill_readdir_cache(struct inode *dir, struct dentry *dn, | |
1582 | struct ceph_readdir_cache_control *ctl, | |
1583 | struct ceph_mds_request *req) | |
1584 | { | |
1585 | struct ceph_inode_info *ci = ceph_inode(dir); | |
09cbfeaf | 1586 | unsigned nsize = PAGE_SIZE / sizeof(struct dentry*); |
fdd4e158 YZ |
1587 | unsigned idx = ctl->index % nsize; |
1588 | pgoff_t pgoff = ctl->index / nsize; | |
1589 | ||
1590 | if (!ctl->page || pgoff != page_index(ctl->page)) { | |
1591 | ceph_readdir_cache_release(ctl); | |
af5e5eb5 YZ |
1592 | if (idx == 0) |
1593 | ctl->page = grab_cache_page(&dir->i_data, pgoff); | |
1594 | else | |
1595 | ctl->page = find_lock_page(&dir->i_data, pgoff); | |
fdd4e158 YZ |
1596 | if (!ctl->page) { |
1597 | ctl->index = -1; | |
af5e5eb5 | 1598 | return idx == 0 ? -ENOMEM : 0; |
fdd4e158 YZ |
1599 | } |
1600 | /* reading/filling the cache are serialized by | |
1601 | * i_mutex, no need to use page lock */ | |
1602 | unlock_page(ctl->page); | |
1603 | ctl->dentries = kmap(ctl->page); | |
af5e5eb5 | 1604 | if (idx == 0) |
09cbfeaf | 1605 | memset(ctl->dentries, 0, PAGE_SIZE); |
fdd4e158 YZ |
1606 | } |
1607 | ||
1608 | if (req->r_dir_release_cnt == atomic64_read(&ci->i_release_count) && | |
1609 | req->r_dir_ordered_cnt == atomic64_read(&ci->i_ordered_count)) { | |
1610 | dout("readdir cache dn %p idx %d\n", dn, ctl->index); | |
1611 | ctl->dentries[idx] = dn; | |
1612 | ctl->index++; | |
1613 | } else { | |
1614 | dout("disable readdir cache\n"); | |
1615 | ctl->index = -1; | |
1616 | } | |
1617 | return 0; | |
1618 | } | |
1619 | ||
355da1eb SW |
1620 | int ceph_readdir_prepopulate(struct ceph_mds_request *req, |
1621 | struct ceph_mds_session *session) | |
1622 | { | |
1623 | struct dentry *parent = req->r_dentry; | |
f3c4ebe6 | 1624 | struct ceph_inode_info *ci = ceph_inode(d_inode(parent)); |
355da1eb SW |
1625 | struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info; |
1626 | struct qstr dname; | |
1627 | struct dentry *dn; | |
1628 | struct inode *in; | |
315f2408 | 1629 | int err = 0, skipped = 0, ret, i; |
0f51a983 | 1630 | u32 frag = le32_to_cpu(req->r_args.readdir.frag); |
f3c4ebe6 YZ |
1631 | u32 last_hash = 0; |
1632 | u32 fpos_offset; | |
fdd4e158 YZ |
1633 | struct ceph_readdir_cache_control cache_ctl = {}; |
1634 | ||
bc2de10d | 1635 | if (test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) |
fdd4e158 | 1636 | return readdir_prepopulate_inodes_only(req, session); |
81c6aea5 | 1637 | |
79162547 YZ |
1638 | if (rinfo->hash_order) { |
1639 | if (req->r_path2) { | |
1640 | last_hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash, | |
1641 | req->r_path2, | |
1642 | strlen(req->r_path2)); | |
1643 | last_hash = ceph_frag_value(last_hash); | |
1644 | } else if (rinfo->offset_hash) { | |
1645 | /* mds understands offset_hash */ | |
1646 | WARN_ON_ONCE(req->r_readdir_offset != 2); | |
0f51a983 | 1647 | last_hash = le32_to_cpu(req->r_args.readdir.offset_hash); |
79162547 | 1648 | } |
f3c4ebe6 YZ |
1649 | } |
1650 | ||
81c6aea5 YZ |
1651 | if (rinfo->dir_dir && |
1652 | le32_to_cpu(rinfo->dir_dir->frag) != frag) { | |
1653 | dout("readdir_prepopulate got new frag %x -> %x\n", | |
1654 | frag, le32_to_cpu(rinfo->dir_dir->frag)); | |
1655 | frag = le32_to_cpu(rinfo->dir_dir->frag); | |
f3c4ebe6 | 1656 | if (!rinfo->hash_order) |
fdd4e158 | 1657 | req->r_readdir_offset = 2; |
81c6aea5 | 1658 | } |
355da1eb SW |
1659 | |
1660 | if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) { | |
355da1eb SW |
1661 | dout("readdir_prepopulate %d items under SNAPDIR dn %p\n", |
1662 | rinfo->dir_nr, parent); | |
1663 | } else { | |
1664 | dout("readdir_prepopulate %d items under dn %p\n", | |
1665 | rinfo->dir_nr, parent); | |
1666 | if (rinfo->dir_dir) | |
2b0143b5 | 1667 | ceph_fill_dirfrag(d_inode(parent), rinfo->dir_dir); |
355da1eb | 1668 | |
8d45b911 YZ |
1669 | if (ceph_frag_is_leftmost(frag) && |
1670 | req->r_readdir_offset == 2 && | |
1671 | !(rinfo->hash_order && last_hash)) { | |
1672 | /* note dir version at start of readdir so we can | |
1673 | * tell if any dentries get dropped */ | |
1674 | req->r_dir_release_cnt = | |
1675 | atomic64_read(&ci->i_release_count); | |
1676 | req->r_dir_ordered_cnt = | |
1677 | atomic64_read(&ci->i_ordered_count); | |
1678 | req->r_readdir_cache_idx = 0; | |
1679 | } | |
fdd4e158 YZ |
1680 | } |
1681 | ||
1682 | cache_ctl.index = req->r_readdir_cache_idx; | |
f3c4ebe6 | 1683 | fpos_offset = req->r_readdir_offset; |
fdd4e158 | 1684 | |
86b58d13 | 1685 | /* FIXME: release caps/leases if error occurs */ |
355da1eb | 1686 | for (i = 0; i < rinfo->dir_nr; i++) { |
2a5beea3 | 1687 | struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i; |
543212b3 | 1688 | struct ceph_vino tvino; |
355da1eb | 1689 | |
2a5beea3 YZ |
1690 | dname.name = rde->name; |
1691 | dname.len = rde->name_len; | |
8387ff25 | 1692 | dname.hash = full_name_hash(parent, dname.name, dname.len); |
355da1eb | 1693 | |
f5d55f03 JL |
1694 | tvino.ino = le64_to_cpu(rde->inode.in->ino); |
1695 | tvino.snap = le64_to_cpu(rde->inode.in->snapid); | |
355da1eb | 1696 | |
f3c4ebe6 YZ |
1697 | if (rinfo->hash_order) { |
1698 | u32 hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash, | |
1699 | rde->name, rde->name_len); | |
1700 | hash = ceph_frag_value(hash); | |
1701 | if (hash != last_hash) | |
1702 | fpos_offset = 2; | |
1703 | last_hash = hash; | |
1704 | rde->offset = ceph_make_fpos(hash, fpos_offset++, true); | |
1705 | } else { | |
1706 | rde->offset = ceph_make_fpos(frag, fpos_offset++, false); | |
1707 | } | |
355da1eb SW |
1708 | |
1709 | retry_lookup: | |
1710 | dn = d_lookup(parent, &dname); | |
1711 | dout("d_lookup on parent=%p name=%.*s got %p\n", | |
1712 | parent, dname.len, dname.name, dn); | |
1713 | ||
1714 | if (!dn) { | |
1715 | dn = d_alloc(parent, &dname); | |
1716 | dout("d_alloc %p '%.*s' = %p\n", parent, | |
1717 | dname.len, dname.name, dn); | |
d37b1d99 | 1718 | if (!dn) { |
355da1eb SW |
1719 | dout("d_alloc badness\n"); |
1720 | err = -ENOMEM; | |
1721 | goto out; | |
1722 | } | |
2b0143b5 | 1723 | } else if (d_really_is_positive(dn) && |
f5d55f03 JL |
1724 | (ceph_ino(d_inode(dn)) != tvino.ino || |
1725 | ceph_snap(d_inode(dn)) != tvino.snap)) { | |
5495c2d0 | 1726 | struct ceph_dentry_info *di = ceph_dentry(dn); |
355da1eb | 1727 | dout(" dn %p points to wrong inode %p\n", |
2b0143b5 | 1728 | dn, d_inode(dn)); |
5495c2d0 YZ |
1729 | |
1730 | spin_lock(&dn->d_lock); | |
1731 | if (di->offset > 0 && | |
1732 | di->lease_shared_gen == | |
1733 | atomic_read(&ci->i_shared_gen)) { | |
1734 | __ceph_dir_clear_ordered(ci); | |
1735 | di->offset = 0; | |
1736 | } | |
1737 | spin_unlock(&dn->d_lock); | |
1738 | ||
355da1eb SW |
1739 | d_delete(dn); |
1740 | dput(dn); | |
1741 | goto retry_lookup; | |
355da1eb SW |
1742 | } |
1743 | ||
355da1eb | 1744 | /* inode */ |
2b0143b5 DH |
1745 | if (d_really_is_positive(dn)) { |
1746 | in = d_inode(dn); | |
355da1eb | 1747 | } else { |
f5d55f03 | 1748 | in = ceph_get_inode(parent->d_sb, tvino); |
ac1f12ef | 1749 | if (IS_ERR(in)) { |
355da1eb | 1750 | dout("new_inode badness\n"); |
2744c171 | 1751 | d_drop(dn); |
355da1eb | 1752 | dput(dn); |
ac1f12ef | 1753 | err = PTR_ERR(in); |
355da1eb SW |
1754 | goto out; |
1755 | } | |
355da1eb SW |
1756 | } |
1757 | ||
966c7160 JL |
1758 | ret = ceph_fill_inode(in, NULL, &rde->inode, NULL, session, |
1759 | -1, &req->r_caps_reservation); | |
fdd4e158 | 1760 | if (ret < 0) { |
966c7160 | 1761 | pr_err("ceph_fill_inode badness on %p\n", in); |
3e1d0452 | 1762 | if (d_really_is_negative(dn)) { |
893e456b JL |
1763 | if (in->i_state & I_NEW) { |
1764 | ihold(in); | |
1765 | discard_new_inode(in); | |
1766 | } | |
23c2c76e | 1767 | iput(in); |
3e1d0452 | 1768 | } |
86b58d13 | 1769 | d_drop(dn); |
fdd4e158 | 1770 | err = ret; |
d69ed05a | 1771 | goto next_item; |
355da1eb | 1772 | } |
893e456b JL |
1773 | if (in->i_state & I_NEW) |
1774 | unlock_new_inode(in); | |
86b58d13 | 1775 | |
2b0143b5 | 1776 | if (d_really_is_negative(dn)) { |
315f2408 YZ |
1777 | if (ceph_security_xattr_deadlock(in)) { |
1778 | dout(" skip splicing dn %p to inode %p" | |
1779 | " (security xattr deadlock)\n", dn, in); | |
23c2c76e | 1780 | iput(in); |
315f2408 YZ |
1781 | skipped++; |
1782 | goto next_item; | |
1783 | } | |
1784 | ||
2bf996ac YZ |
1785 | err = splice_dentry(&dn, in); |
1786 | if (err < 0) | |
86b58d13 | 1787 | goto next_item; |
86b58d13 YZ |
1788 | } |
1789 | ||
f3c4ebe6 | 1790 | ceph_dentry(dn)->offset = rde->offset; |
86b58d13 | 1791 | |
543212b3 YZ |
1792 | update_dentry_lease(d_inode(parent), dn, |
1793 | rde->lease, req->r_session, | |
1794 | req->r_request_started); | |
fdd4e158 | 1795 | |
315f2408 | 1796 | if (err == 0 && skipped == 0 && cache_ctl.index >= 0) { |
fdd4e158 YZ |
1797 | ret = fill_readdir_cache(d_inode(parent), dn, |
1798 | &cache_ctl, req); | |
1799 | if (ret < 0) | |
1800 | err = ret; | |
1801 | } | |
d69ed05a | 1802 | next_item: |
2bf996ac | 1803 | dput(dn); |
355da1eb | 1804 | } |
355da1eb | 1805 | out: |
315f2408 | 1806 | if (err == 0 && skipped == 0) { |
bc2de10d | 1807 | set_bit(CEPH_MDS_R_DID_PREPOPULATE, &req->r_req_flags); |
fdd4e158 YZ |
1808 | req->r_readdir_cache_idx = cache_ctl.index; |
1809 | } | |
1810 | ceph_readdir_cache_release(&cache_ctl); | |
355da1eb SW |
1811 | dout("readdir_prepopulate done\n"); |
1812 | return err; | |
1813 | } | |
1814 | ||
efb0ca76 | 1815 | bool ceph_inode_set_size(struct inode *inode, loff_t size) |
355da1eb SW |
1816 | { |
1817 | struct ceph_inode_info *ci = ceph_inode(inode); | |
efb0ca76 | 1818 | bool ret; |
355da1eb | 1819 | |
be655596 | 1820 | spin_lock(&ci->i_ceph_lock); |
2d6795fb | 1821 | dout("set_size %p %llu -> %llu\n", inode, i_size_read(inode), size); |
99c88e69 | 1822 | i_size_write(inode, size); |
400e1286 | 1823 | ceph_fscache_update(inode); |
224a7542 | 1824 | inode->i_blocks = calc_inode_blocks(size); |
355da1eb | 1825 | |
efb0ca76 | 1826 | ret = __ceph_should_report_size(ci); |
355da1eb | 1827 | |
be655596 | 1828 | spin_unlock(&ci->i_ceph_lock); |
400e1286 | 1829 | |
355da1eb SW |
1830 | return ret; |
1831 | } | |
1832 | ||
64f28c62 | 1833 | void ceph_queue_inode_work(struct inode *inode, int work_bit) |
355da1eb | 1834 | { |
64f28c62 | 1835 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); |
1cf89a8d | 1836 | struct ceph_inode_info *ci = ceph_inode(inode); |
64f28c62 | 1837 | set_bit(work_bit, &ci->i_work_mask); |
1cf89a8d YZ |
1838 | |
1839 | ihold(inode); | |
64f28c62 JL |
1840 | if (queue_work(fsc->inode_wq, &ci->i_work)) { |
1841 | dout("queue_inode_work %p, mask=%lx\n", inode, ci->i_work_mask); | |
1cf89a8d | 1842 | } else { |
64f28c62 | 1843 | dout("queue_inode_work %p already queued, mask=%lx\n", |
1cf89a8d YZ |
1844 | inode, ci->i_work_mask); |
1845 | iput(inode); | |
1846 | } | |
1847 | } | |
1848 | ||
1849 | static void ceph_do_invalidate_pages(struct inode *inode) | |
1850 | { | |
1851 | struct ceph_inode_info *ci = ceph_inode(inode); | |
355da1eb SW |
1852 | u32 orig_gen; |
1853 | int check = 0; | |
1854 | ||
400e1286 JL |
1855 | ceph_fscache_invalidate(inode, false); |
1856 | ||
b0d7c223 | 1857 | mutex_lock(&ci->i_truncate_mutex); |
6c93df5d | 1858 | |
5d6451b1 | 1859 | if (ceph_inode_is_shutdown(inode)) { |
6407fbb9 JL |
1860 | pr_warn_ratelimited("%s: inode %llx.%llx is shut down\n", |
1861 | __func__, ceph_vinop(inode)); | |
6c93df5d YZ |
1862 | mapping_set_error(inode->i_mapping, -EIO); |
1863 | truncate_pagecache(inode, 0); | |
1864 | mutex_unlock(&ci->i_truncate_mutex); | |
1865 | goto out; | |
1866 | } | |
1867 | ||
be655596 | 1868 | spin_lock(&ci->i_ceph_lock); |
355da1eb SW |
1869 | dout("invalidate_pages %p gen %d revoking %d\n", inode, |
1870 | ci->i_rdcache_gen, ci->i_rdcache_revoking); | |
cd045cb4 | 1871 | if (ci->i_rdcache_revoking != ci->i_rdcache_gen) { |
9563f88c YZ |
1872 | if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE)) |
1873 | check = 1; | |
be655596 | 1874 | spin_unlock(&ci->i_ceph_lock); |
b0d7c223 | 1875 | mutex_unlock(&ci->i_truncate_mutex); |
355da1eb SW |
1876 | goto out; |
1877 | } | |
1878 | orig_gen = ci->i_rdcache_gen; | |
be655596 | 1879 | spin_unlock(&ci->i_ceph_lock); |
355da1eb | 1880 | |
400e1286 | 1881 | ceph_fscache_invalidate(inode, false); |
9abd4db7 | 1882 | if (invalidate_inode_pages2(inode->i_mapping) < 0) { |
6407fbb9 JL |
1883 | pr_err("invalidate_inode_pages2 %llx.%llx failed\n", |
1884 | ceph_vinop(inode)); | |
9abd4db7 | 1885 | } |
355da1eb | 1886 | |
be655596 | 1887 | spin_lock(&ci->i_ceph_lock); |
cd045cb4 SW |
1888 | if (orig_gen == ci->i_rdcache_gen && |
1889 | orig_gen == ci->i_rdcache_revoking) { | |
355da1eb SW |
1890 | dout("invalidate_pages %p gen %d successful\n", inode, |
1891 | ci->i_rdcache_gen); | |
cd045cb4 | 1892 | ci->i_rdcache_revoking--; |
355da1eb SW |
1893 | check = 1; |
1894 | } else { | |
cd045cb4 SW |
1895 | dout("invalidate_pages %p gen %d raced, now %d revoking %d\n", |
1896 | inode, orig_gen, ci->i_rdcache_gen, | |
1897 | ci->i_rdcache_revoking); | |
9563f88c YZ |
1898 | if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE)) |
1899 | check = 1; | |
355da1eb | 1900 | } |
be655596 | 1901 | spin_unlock(&ci->i_ceph_lock); |
b0d7c223 | 1902 | mutex_unlock(&ci->i_truncate_mutex); |
9563f88c | 1903 | out: |
355da1eb SW |
1904 | if (check) |
1905 | ceph_check_caps(ci, 0, NULL); | |
3c6f6b79 SW |
1906 | } |
1907 | ||
355da1eb | 1908 | /* |
355da1eb SW |
1909 | * Make sure any pending truncation is applied before doing anything |
1910 | * that may depend on it. | |
1911 | */ | |
b415bf4f | 1912 | void __ceph_do_pending_vmtruncate(struct inode *inode) |
355da1eb SW |
1913 | { |
1914 | struct ceph_inode_info *ci = ceph_inode(inode); | |
1915 | u64 to; | |
a85f50b6 | 1916 | int wrbuffer_refs, finish = 0; |
355da1eb | 1917 | |
b0d7c223 | 1918 | mutex_lock(&ci->i_truncate_mutex); |
355da1eb | 1919 | retry: |
be655596 | 1920 | spin_lock(&ci->i_ceph_lock); |
355da1eb SW |
1921 | if (ci->i_truncate_pending == 0) { |
1922 | dout("__do_pending_vmtruncate %p none pending\n", inode); | |
be655596 | 1923 | spin_unlock(&ci->i_ceph_lock); |
b0d7c223 | 1924 | mutex_unlock(&ci->i_truncate_mutex); |
355da1eb SW |
1925 | return; |
1926 | } | |
1927 | ||
1928 | /* | |
1929 | * make sure any dirty snapped pages are flushed before we | |
1930 | * possibly truncate them.. so write AND block! | |
1931 | */ | |
1932 | if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) { | |
c8fd0d37 | 1933 | spin_unlock(&ci->i_ceph_lock); |
355da1eb SW |
1934 | dout("__do_pending_vmtruncate %p flushing snaps first\n", |
1935 | inode); | |
355da1eb SW |
1936 | filemap_write_and_wait_range(&inode->i_data, 0, |
1937 | inode->i_sb->s_maxbytes); | |
1938 | goto retry; | |
1939 | } | |
1940 | ||
b0d7c223 YZ |
1941 | /* there should be no reader or writer */ |
1942 | WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref); | |
1943 | ||
355da1eb SW |
1944 | to = ci->i_truncate_size; |
1945 | wrbuffer_refs = ci->i_wrbuffer_ref; | |
1946 | dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode, | |
1947 | ci->i_truncate_pending, to); | |
be655596 | 1948 | spin_unlock(&ci->i_ceph_lock); |
355da1eb | 1949 | |
400e1286 | 1950 | ceph_fscache_resize(inode, to); |
4e217b5d | 1951 | truncate_pagecache(inode, to); |
355da1eb | 1952 | |
be655596 | 1953 | spin_lock(&ci->i_ceph_lock); |
a85f50b6 YZ |
1954 | if (to == ci->i_truncate_size) { |
1955 | ci->i_truncate_pending = 0; | |
1956 | finish = 1; | |
1957 | } | |
be655596 | 1958 | spin_unlock(&ci->i_ceph_lock); |
a85f50b6 YZ |
1959 | if (!finish) |
1960 | goto retry; | |
355da1eb | 1961 | |
b0d7c223 YZ |
1962 | mutex_unlock(&ci->i_truncate_mutex); |
1963 | ||
355da1eb | 1964 | if (wrbuffer_refs == 0) |
bf73c62e | 1965 | ceph_check_caps(ci, 0, NULL); |
a85f50b6 YZ |
1966 | |
1967 | wake_up_all(&ci->i_cap_wq); | |
355da1eb SW |
1968 | } |
1969 | ||
1cf89a8d YZ |
1970 | static void ceph_inode_work(struct work_struct *work) |
1971 | { | |
1972 | struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info, | |
1973 | i_work); | |
1974 | struct inode *inode = &ci->vfs_inode; | |
1975 | ||
1976 | if (test_and_clear_bit(CEPH_I_WORK_WRITEBACK, &ci->i_work_mask)) { | |
1977 | dout("writeback %p\n", inode); | |
1978 | filemap_fdatawrite(&inode->i_data); | |
1979 | } | |
1980 | if (test_and_clear_bit(CEPH_I_WORK_INVALIDATE_PAGES, &ci->i_work_mask)) | |
1981 | ceph_do_invalidate_pages(inode); | |
1982 | ||
1983 | if (test_and_clear_bit(CEPH_I_WORK_VMTRUNCATE, &ci->i_work_mask)) | |
1984 | __ceph_do_pending_vmtruncate(inode); | |
1985 | ||
a8810cdc JL |
1986 | if (test_and_clear_bit(CEPH_I_WORK_CHECK_CAPS, &ci->i_work_mask)) |
1987 | ceph_check_caps(ci, 0, NULL); | |
1988 | ||
1989 | if (test_and_clear_bit(CEPH_I_WORK_FLUSH_SNAPS, &ci->i_work_mask)) | |
1990 | ceph_flush_snaps(ci, NULL); | |
1991 | ||
1cf89a8d YZ |
1992 | iput(inode); |
1993 | } | |
1994 | ||
355da1eb SW |
1995 | /* |
1996 | * symlinks | |
1997 | */ | |
355da1eb | 1998 | static const struct inode_operations ceph_symlink_iops = { |
6b255391 | 1999 | .get_link = simple_get_link, |
0b932672 YZ |
2000 | .setattr = ceph_setattr, |
2001 | .getattr = ceph_getattr, | |
0b932672 | 2002 | .listxattr = ceph_listxattr, |
355da1eb SW |
2003 | }; |
2004 | ||
a26fecca | 2005 | int __ceph_setattr(struct inode *inode, struct iattr *attr) |
355da1eb | 2006 | { |
355da1eb | 2007 | struct ceph_inode_info *ci = ceph_inode(inode); |
c62498d7 | 2008 | unsigned int ia_valid = attr->ia_valid; |
355da1eb | 2009 | struct ceph_mds_request *req; |
a26fecca | 2010 | struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc; |
f66fd9f0 | 2011 | struct ceph_cap_flush *prealloc_cf; |
355da1eb SW |
2012 | int issued; |
2013 | int release = 0, dirtied = 0; | |
2014 | int mask = 0; | |
2015 | int err = 0; | |
fca65b4a | 2016 | int inode_dirty_flags = 0; |
604d1b02 | 2017 | bool lock_snap_rwsem = false; |
355da1eb | 2018 | |
f66fd9f0 YZ |
2019 | prealloc_cf = ceph_alloc_cap_flush(); |
2020 | if (!prealloc_cf) | |
2021 | return -ENOMEM; | |
2022 | ||
355da1eb SW |
2023 | req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR, |
2024 | USE_AUTH_MDS); | |
f66fd9f0 YZ |
2025 | if (IS_ERR(req)) { |
2026 | ceph_free_cap_flush(prealloc_cf); | |
355da1eb | 2027 | return PTR_ERR(req); |
f66fd9f0 | 2028 | } |
355da1eb | 2029 | |
be655596 | 2030 | spin_lock(&ci->i_ceph_lock); |
355da1eb | 2031 | issued = __ceph_caps_issued(ci, NULL); |
604d1b02 YZ |
2032 | |
2033 | if (!ci->i_head_snapc && | |
2034 | (issued & (CEPH_CAP_ANY_EXCL | CEPH_CAP_FILE_WR))) { | |
2035 | lock_snap_rwsem = true; | |
2036 | if (!down_read_trylock(&mdsc->snap_rwsem)) { | |
2037 | spin_unlock(&ci->i_ceph_lock); | |
2038 | down_read(&mdsc->snap_rwsem); | |
2039 | spin_lock(&ci->i_ceph_lock); | |
2040 | issued = __ceph_caps_issued(ci, NULL); | |
2041 | } | |
2042 | } | |
2043 | ||
355da1eb SW |
2044 | dout("setattr %p issued %s\n", inode, ceph_cap_string(issued)); |
2045 | ||
2046 | if (ia_valid & ATTR_UID) { | |
2047 | dout("setattr %p uid %d -> %d\n", inode, | |
bd2bae6a EB |
2048 | from_kuid(&init_user_ns, inode->i_uid), |
2049 | from_kuid(&init_user_ns, attr->ia_uid)); | |
355da1eb SW |
2050 | if (issued & CEPH_CAP_AUTH_EXCL) { |
2051 | inode->i_uid = attr->ia_uid; | |
2052 | dirtied |= CEPH_CAP_AUTH_EXCL; | |
2053 | } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || | |
ab871b90 EB |
2054 | !uid_eq(attr->ia_uid, inode->i_uid)) { |
2055 | req->r_args.setattr.uid = cpu_to_le32( | |
2056 | from_kuid(&init_user_ns, attr->ia_uid)); | |
355da1eb SW |
2057 | mask |= CEPH_SETATTR_UID; |
2058 | release |= CEPH_CAP_AUTH_SHARED; | |
2059 | } | |
2060 | } | |
2061 | if (ia_valid & ATTR_GID) { | |
2062 | dout("setattr %p gid %d -> %d\n", inode, | |
bd2bae6a EB |
2063 | from_kgid(&init_user_ns, inode->i_gid), |
2064 | from_kgid(&init_user_ns, attr->ia_gid)); | |
355da1eb SW |
2065 | if (issued & CEPH_CAP_AUTH_EXCL) { |
2066 | inode->i_gid = attr->ia_gid; | |
2067 | dirtied |= CEPH_CAP_AUTH_EXCL; | |
2068 | } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || | |
ab871b90 EB |
2069 | !gid_eq(attr->ia_gid, inode->i_gid)) { |
2070 | req->r_args.setattr.gid = cpu_to_le32( | |
2071 | from_kgid(&init_user_ns, attr->ia_gid)); | |
355da1eb SW |
2072 | mask |= CEPH_SETATTR_GID; |
2073 | release |= CEPH_CAP_AUTH_SHARED; | |
2074 | } | |
2075 | } | |
2076 | if (ia_valid & ATTR_MODE) { | |
2077 | dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode, | |
2078 | attr->ia_mode); | |
2079 | if (issued & CEPH_CAP_AUTH_EXCL) { | |
2080 | inode->i_mode = attr->ia_mode; | |
2081 | dirtied |= CEPH_CAP_AUTH_EXCL; | |
2082 | } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || | |
2083 | attr->ia_mode != inode->i_mode) { | |
7221fe4c | 2084 | inode->i_mode = attr->ia_mode; |
355da1eb SW |
2085 | req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode); |
2086 | mask |= CEPH_SETATTR_MODE; | |
2087 | release |= CEPH_CAP_AUTH_SHARED; | |
2088 | } | |
2089 | } | |
2090 | ||
2091 | if (ia_valid & ATTR_ATIME) { | |
13442b03 | 2092 | dout("setattr %p atime %lld.%ld -> %lld.%ld\n", inode, |
9bbeab41 AB |
2093 | inode->i_atime.tv_sec, inode->i_atime.tv_nsec, |
2094 | attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec); | |
355da1eb SW |
2095 | if (issued & CEPH_CAP_FILE_EXCL) { |
2096 | ci->i_time_warp_seq++; | |
2097 | inode->i_atime = attr->ia_atime; | |
2098 | dirtied |= CEPH_CAP_FILE_EXCL; | |
2099 | } else if ((issued & CEPH_CAP_FILE_WR) && | |
95582b00 | 2100 | timespec64_compare(&inode->i_atime, |
355da1eb SW |
2101 | &attr->ia_atime) < 0) { |
2102 | inode->i_atime = attr->ia_atime; | |
2103 | dirtied |= CEPH_CAP_FILE_WR; | |
2104 | } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || | |
95582b00 | 2105 | !timespec64_equal(&inode->i_atime, &attr->ia_atime)) { |
9bbeab41 AB |
2106 | ceph_encode_timespec64(&req->r_args.setattr.atime, |
2107 | &attr->ia_atime); | |
355da1eb | 2108 | mask |= CEPH_SETATTR_ATIME; |
be70489e YZ |
2109 | release |= CEPH_CAP_FILE_SHARED | |
2110 | CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR; | |
355da1eb SW |
2111 | } |
2112 | } | |
c62498d7 | 2113 | if (ia_valid & ATTR_SIZE) { |
2d6795fb JL |
2114 | loff_t isize = i_size_read(inode); |
2115 | ||
2116 | dout("setattr %p size %lld -> %lld\n", inode, isize, attr->ia_size); | |
e90334e8 XL |
2117 | if ((issued & CEPH_CAP_FILE_EXCL) && attr->ia_size >= isize) { |
2118 | if (attr->ia_size > isize) { | |
2119 | i_size_write(inode, attr->ia_size); | |
2120 | inode->i_blocks = calc_inode_blocks(attr->ia_size); | |
2121 | ci->i_reported_size = attr->ia_size; | |
2122 | dirtied |= CEPH_CAP_FILE_EXCL; | |
2123 | ia_valid |= ATTR_MTIME; | |
2124 | } | |
c62498d7 | 2125 | } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || |
2d6795fb | 2126 | attr->ia_size != isize) { |
c62498d7 | 2127 | req->r_args.setattr.size = cpu_to_le64(attr->ia_size); |
2d6795fb | 2128 | req->r_args.setattr.old_size = cpu_to_le64(isize); |
c62498d7 JL |
2129 | mask |= CEPH_SETATTR_SIZE; |
2130 | release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_EXCL | | |
2131 | CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR; | |
2132 | } | |
2133 | } | |
355da1eb | 2134 | if (ia_valid & ATTR_MTIME) { |
13442b03 | 2135 | dout("setattr %p mtime %lld.%ld -> %lld.%ld\n", inode, |
9bbeab41 AB |
2136 | inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec, |
2137 | attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec); | |
355da1eb SW |
2138 | if (issued & CEPH_CAP_FILE_EXCL) { |
2139 | ci->i_time_warp_seq++; | |
2140 | inode->i_mtime = attr->ia_mtime; | |
2141 | dirtied |= CEPH_CAP_FILE_EXCL; | |
2142 | } else if ((issued & CEPH_CAP_FILE_WR) && | |
95582b00 | 2143 | timespec64_compare(&inode->i_mtime, |
355da1eb SW |
2144 | &attr->ia_mtime) < 0) { |
2145 | inode->i_mtime = attr->ia_mtime; | |
2146 | dirtied |= CEPH_CAP_FILE_WR; | |
2147 | } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || | |
95582b00 | 2148 | !timespec64_equal(&inode->i_mtime, &attr->ia_mtime)) { |
9bbeab41 AB |
2149 | ceph_encode_timespec64(&req->r_args.setattr.mtime, |
2150 | &attr->ia_mtime); | |
355da1eb | 2151 | mask |= CEPH_SETATTR_MTIME; |
be70489e YZ |
2152 | release |= CEPH_CAP_FILE_SHARED | |
2153 | CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR; | |
355da1eb SW |
2154 | } |
2155 | } | |
355da1eb SW |
2156 | |
2157 | /* these do nothing */ | |
2158 | if (ia_valid & ATTR_CTIME) { | |
2159 | bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME| | |
2160 | ATTR_MODE|ATTR_UID|ATTR_GID)) == 0; | |
13442b03 | 2161 | dout("setattr %p ctime %lld.%ld -> %lld.%ld (%s)\n", inode, |
9bbeab41 AB |
2162 | inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec, |
2163 | attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec, | |
355da1eb | 2164 | only ? "ctime only" : "ignored"); |
355da1eb SW |
2165 | if (only) { |
2166 | /* | |
2167 | * if kernel wants to dirty ctime but nothing else, | |
2168 | * we need to choose a cap to dirty under, or do | |
2169 | * a almost-no-op setattr | |
2170 | */ | |
2171 | if (issued & CEPH_CAP_AUTH_EXCL) | |
2172 | dirtied |= CEPH_CAP_AUTH_EXCL; | |
2173 | else if (issued & CEPH_CAP_FILE_EXCL) | |
2174 | dirtied |= CEPH_CAP_FILE_EXCL; | |
2175 | else if (issued & CEPH_CAP_XATTR_EXCL) | |
2176 | dirtied |= CEPH_CAP_XATTR_EXCL; | |
2177 | else | |
2178 | mask |= CEPH_SETATTR_CTIME; | |
2179 | } | |
2180 | } | |
2181 | if (ia_valid & ATTR_FILE) | |
2182 | dout("setattr %p ATTR_FILE ... hrm!\n", inode); | |
2183 | ||
2184 | if (dirtied) { | |
f66fd9f0 YZ |
2185 | inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied, |
2186 | &prealloc_cf); | |
4ca2fea6 | 2187 | inode->i_ctime = attr->ia_ctime; |
355da1eb SW |
2188 | } |
2189 | ||
2190 | release &= issued; | |
be655596 | 2191 | spin_unlock(&ci->i_ceph_lock); |
604d1b02 YZ |
2192 | if (lock_snap_rwsem) |
2193 | up_read(&mdsc->snap_rwsem); | |
355da1eb | 2194 | |
fca65b4a SW |
2195 | if (inode_dirty_flags) |
2196 | __mark_inode_dirty(inode, inode_dirty_flags); | |
2197 | ||
355da1eb | 2198 | if (mask) { |
70b666c3 SW |
2199 | req->r_inode = inode; |
2200 | ihold(inode); | |
355da1eb SW |
2201 | req->r_inode_drop = release; |
2202 | req->r_args.setattr.mask = cpu_to_le32(mask); | |
2203 | req->r_num_caps = 1; | |
0ed1e90a | 2204 | req->r_stamp = attr->ia_ctime; |
752c8bdc | 2205 | err = ceph_mdsc_do_request(mdsc, NULL, req); |
355da1eb SW |
2206 | } |
2207 | dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err, | |
2208 | ceph_cap_string(dirtied), mask); | |
2209 | ||
355da1eb | 2210 | ceph_mdsc_put_request(req); |
f66fd9f0 | 2211 | ceph_free_cap_flush(prealloc_cf); |
8179a101 YZ |
2212 | |
2213 | if (err >= 0 && (mask & CEPH_SETATTR_SIZE)) | |
2214 | __ceph_do_pending_vmtruncate(inode); | |
2215 | ||
355da1eb SW |
2216 | return err; |
2217 | } | |
2218 | ||
a26fecca AG |
2219 | /* |
2220 | * setattr | |
2221 | */ | |
549c7297 CB |
2222 | int ceph_setattr(struct user_namespace *mnt_userns, struct dentry *dentry, |
2223 | struct iattr *attr) | |
a26fecca | 2224 | { |
fd5472ed | 2225 | struct inode *inode = d_inode(dentry); |
36a4c72d | 2226 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); |
fd5472ed JK |
2227 | int err; |
2228 | ||
2229 | if (ceph_snap(inode) != CEPH_NOSNAP) | |
2230 | return -EROFS; | |
2231 | ||
5d6451b1 JL |
2232 | if (ceph_inode_is_shutdown(inode)) |
2233 | return -ESTALE; | |
2234 | ||
2f221d6f | 2235 | err = setattr_prepare(&init_user_ns, dentry, attr); |
fd5472ed JK |
2236 | if (err != 0) |
2237 | return err; | |
2238 | ||
36a4c72d | 2239 | if ((attr->ia_valid & ATTR_SIZE) && |
2d6795fb | 2240 | attr->ia_size > max(i_size_read(inode), fsc->max_file_size)) |
36a4c72d CX |
2241 | return -EFBIG; |
2242 | ||
2b83845f LH |
2243 | if ((attr->ia_valid & ATTR_SIZE) && |
2244 | ceph_quota_is_max_bytes_exceeded(inode, attr->ia_size)) | |
2245 | return -EDQUOT; | |
2246 | ||
8179a101 YZ |
2247 | err = __ceph_setattr(inode, attr); |
2248 | ||
2249 | if (err >= 0 && (attr->ia_valid & ATTR_MODE)) | |
e65ce2a5 | 2250 | err = posix_acl_chmod(&init_user_ns, inode, attr->ia_mode); |
8179a101 YZ |
2251 | |
2252 | return err; | |
a26fecca AG |
2253 | } |
2254 | ||
355da1eb SW |
2255 | /* |
2256 | * Verify that we have a lease on the given mask. If not, | |
2257 | * do a getattr against an mds. | |
2258 | */ | |
01deead0 YZ |
2259 | int __ceph_do_getattr(struct inode *inode, struct page *locked_page, |
2260 | int mask, bool force) | |
355da1eb | 2261 | { |
3d14c5d2 YS |
2262 | struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb); |
2263 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
355da1eb | 2264 | struct ceph_mds_request *req; |
49a9f4f6 | 2265 | int mode; |
355da1eb SW |
2266 | int err; |
2267 | ||
2268 | if (ceph_snap(inode) == CEPH_SNAPDIR) { | |
2269 | dout("do_getattr inode %p SNAPDIR\n", inode); | |
2270 | return 0; | |
2271 | } | |
2272 | ||
01deead0 YZ |
2273 | dout("do_getattr inode %p mask %s mode 0%o\n", |
2274 | inode, ceph_cap_string(mask), inode->i_mode); | |
1af16d54 XL |
2275 | if (!force && ceph_caps_issued_mask_metric(ceph_inode(inode), mask, 1)) |
2276 | return 0; | |
355da1eb | 2277 | |
49a9f4f6 YZ |
2278 | mode = (mask & CEPH_STAT_RSTAT) ? USE_AUTH_MDS : USE_ANY_MDS; |
2279 | req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, mode); | |
355da1eb SW |
2280 | if (IS_ERR(req)) |
2281 | return PTR_ERR(req); | |
70b666c3 SW |
2282 | req->r_inode = inode; |
2283 | ihold(inode); | |
355da1eb SW |
2284 | req->r_num_caps = 1; |
2285 | req->r_args.getattr.mask = cpu_to_le32(mask); | |
01deead0 | 2286 | req->r_locked_page = locked_page; |
355da1eb | 2287 | err = ceph_mdsc_do_request(mdsc, NULL, req); |
01deead0 YZ |
2288 | if (locked_page && err == 0) { |
2289 | u64 inline_version = req->r_reply_info.targeti.inline_version; | |
2290 | if (inline_version == 0) { | |
2291 | /* the reply is supposed to contain inline data */ | |
2292 | err = -EINVAL; | |
2293 | } else if (inline_version == CEPH_INLINE_NONE) { | |
2294 | err = -ENODATA; | |
2295 | } else { | |
2296 | err = req->r_reply_info.targeti.inline_len; | |
2297 | } | |
2298 | } | |
355da1eb SW |
2299 | ceph_mdsc_put_request(req); |
2300 | dout("do_getattr result=%d\n", err); | |
2301 | return err; | |
2302 | } | |
2303 | ||
2304 | ||
2305 | /* | |
2306 | * Check inode permissions. We verify we have a valid value for | |
2307 | * the AUTH cap, then call the generic handler. | |
2308 | */ | |
549c7297 CB |
2309 | int ceph_permission(struct user_namespace *mnt_userns, struct inode *inode, |
2310 | int mask) | |
355da1eb | 2311 | { |
b74c79e9 NP |
2312 | int err; |
2313 | ||
10556cb2 | 2314 | if (mask & MAY_NOT_BLOCK) |
b74c79e9 NP |
2315 | return -ECHILD; |
2316 | ||
508b32d8 | 2317 | err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED, false); |
355da1eb SW |
2318 | |
2319 | if (!err) | |
47291baa | 2320 | err = generic_permission(&init_user_ns, inode, mask); |
355da1eb SW |
2321 | return err; |
2322 | } | |
2323 | ||
428bb68a | 2324 | /* Craft a mask of needed caps given a set of requested statx attrs. */ |
04fabb11 | 2325 | static int statx_to_caps(u32 want, umode_t mode) |
428bb68a JL |
2326 | { |
2327 | int mask = 0; | |
2328 | ||
58981784 | 2329 | if (want & (STATX_MODE|STATX_UID|STATX_GID|STATX_CTIME|STATX_BTIME)) |
428bb68a JL |
2330 | mask |= CEPH_CAP_AUTH_SHARED; |
2331 | ||
04fabb11 JL |
2332 | if (want & (STATX_NLINK|STATX_CTIME)) { |
2333 | /* | |
2334 | * The link count for directories depends on inode->i_subdirs, | |
2335 | * and that is only updated when Fs caps are held. | |
2336 | */ | |
2337 | if (S_ISDIR(mode)) | |
2338 | mask |= CEPH_CAP_FILE_SHARED; | |
2339 | else | |
2340 | mask |= CEPH_CAP_LINK_SHARED; | |
2341 | } | |
428bb68a JL |
2342 | |
2343 | if (want & (STATX_ATIME|STATX_MTIME|STATX_CTIME|STATX_SIZE| | |
2344 | STATX_BLOCKS)) | |
2345 | mask |= CEPH_CAP_FILE_SHARED; | |
2346 | ||
2347 | if (want & (STATX_CTIME)) | |
2348 | mask |= CEPH_CAP_XATTR_SHARED; | |
2349 | ||
2350 | return mask; | |
2351 | } | |
2352 | ||
355da1eb | 2353 | /* |
428bb68a JL |
2354 | * Get all the attributes. If we have sufficient caps for the requested attrs, |
2355 | * then we can avoid talking to the MDS at all. | |
355da1eb | 2356 | */ |
549c7297 CB |
2357 | int ceph_getattr(struct user_namespace *mnt_userns, const struct path *path, |
2358 | struct kstat *stat, u32 request_mask, unsigned int flags) | |
355da1eb | 2359 | { |
a528d35e | 2360 | struct inode *inode = d_inode(path->dentry); |
232d4b01 | 2361 | struct ceph_inode_info *ci = ceph_inode(inode); |
58981784 | 2362 | u32 valid_mask = STATX_BASIC_STATS; |
428bb68a | 2363 | int err = 0; |
355da1eb | 2364 | |
5d6451b1 JL |
2365 | if (ceph_inode_is_shutdown(inode)) |
2366 | return -ESTALE; | |
2367 | ||
428bb68a JL |
2368 | /* Skip the getattr altogether if we're asked not to sync */ |
2369 | if (!(flags & AT_STATX_DONT_SYNC)) { | |
04fabb11 JL |
2370 | err = ceph_do_getattr(inode, |
2371 | statx_to_caps(request_mask, inode->i_mode), | |
2372 | flags & AT_STATX_FORCE_SYNC); | |
428bb68a JL |
2373 | if (err) |
2374 | return err; | |
2375 | } | |
2376 | ||
0d56a451 | 2377 | generic_fillattr(&init_user_ns, inode, stat); |
ebce3eb2 | 2378 | stat->ino = ceph_present_inode(inode); |
58981784 JL |
2379 | |
2380 | /* | |
2381 | * btime on newly-allocated inodes is 0, so if this is still set to | |
2382 | * that, then assume that it's not valid. | |
2383 | */ | |
2384 | if (ci->i_btime.tv_sec || ci->i_btime.tv_nsec) { | |
2385 | stat->btime = ci->i_btime; | |
2386 | valid_mask |= STATX_BTIME; | |
2387 | } | |
2388 | ||
428bb68a JL |
2389 | if (ceph_snap(inode) == CEPH_NOSNAP) |
2390 | stat->dev = inode->i_sb->s_dev; | |
2391 | else | |
2392 | stat->dev = ci->i_snapid_map ? ci->i_snapid_map->dev : 0; | |
2393 | ||
2394 | if (S_ISDIR(inode->i_mode)) { | |
2395 | if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb), | |
2396 | RBYTES)) | |
2397 | stat->size = ci->i_rbytes; | |
355da1eb | 2398 | else |
428bb68a JL |
2399 | stat->size = ci->i_files + ci->i_subdirs; |
2400 | stat->blocks = 0; | |
2401 | stat->blksize = 65536; | |
2402 | /* | |
2403 | * Some applications rely on the number of st_nlink | |
2404 | * value on directories to be either 0 (if unlinked) | |
2405 | * or 2 + number of subdirectories. | |
2406 | */ | |
2407 | if (stat->nlink == 1) | |
2408 | /* '.' + '..' + subdirs */ | |
2409 | stat->nlink = 1 + 1 + ci->i_subdirs; | |
355da1eb | 2410 | } |
428bb68a | 2411 | |
58981784 | 2412 | stat->result_mask = request_mask & valid_mask; |
355da1eb SW |
2413 | return err; |
2414 | } | |
5d6451b1 JL |
2415 | |
2416 | void ceph_inode_shutdown(struct inode *inode) | |
2417 | { | |
2418 | struct ceph_inode_info *ci = ceph_inode(inode); | |
2419 | struct rb_node *p; | |
2420 | int iputs = 0; | |
2421 | bool invalidate = false; | |
2422 | ||
2423 | spin_lock(&ci->i_ceph_lock); | |
2424 | ci->i_ceph_flags |= CEPH_I_SHUTDOWN; | |
2425 | p = rb_first(&ci->i_caps); | |
2426 | while (p) { | |
2427 | struct ceph_cap *cap = rb_entry(p, struct ceph_cap, ci_node); | |
2428 | ||
2429 | p = rb_next(p); | |
2430 | iputs += ceph_purge_inode_cap(inode, cap, &invalidate); | |
2431 | } | |
2432 | spin_unlock(&ci->i_ceph_lock); | |
2433 | ||
2434 | if (invalidate) | |
2435 | ceph_queue_invalidate(inode); | |
2436 | while (iputs--) | |
2437 | iput(inode); | |
2438 | } |