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> |
2d332d5b | 17 | #include <linux/fscrypt.h> |
355da1eb SW |
18 | |
19 | #include "super.h" | |
3d14c5d2 | 20 | #include "mds_client.h" |
99ccbd22 | 21 | #include "cache.h" |
2d332d5b | 22 | #include "crypto.h" |
3d14c5d2 | 23 | #include <linux/ceph/decode.h> |
355da1eb SW |
24 | |
25 | /* | |
26 | * Ceph inode operations | |
27 | * | |
28 | * Implement basic inode helpers (get, alloc) and inode ops (getattr, | |
29 | * setattr, etc.), xattr helpers, and helpers for assimilating | |
30 | * metadata returned by the MDS into our cache. | |
31 | * | |
32 | * Also define helpers for doing asynchronous writeback, invalidation, | |
33 | * and truncation for the benefit of those who can't afford to block | |
34 | * (typically because they are in the message handler path). | |
35 | */ | |
36 | ||
37 | static const struct inode_operations ceph_symlink_iops; | |
38 | ||
1cf89a8d | 39 | static void ceph_inode_work(struct work_struct *work); |
355da1eb SW |
40 | |
41 | /* | |
42 | * find or create an inode, given the ceph ino number | |
43 | */ | |
ad1fee96 YS |
44 | static int ceph_set_ino_cb(struct inode *inode, void *data) |
45 | { | |
ebce3eb2 | 46 | struct ceph_inode_info *ci = ceph_inode(inode); |
1dd8d470 | 47 | struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb); |
ebce3eb2 JL |
48 | |
49 | ci->i_vino = *(struct ceph_vino *)data; | |
50 | inode->i_ino = ceph_vino_to_ino_t(ci->i_vino); | |
a35ead31 | 51 | inode_set_iversion_raw(inode, 0); |
1dd8d470 XL |
52 | percpu_counter_inc(&mdsc->metric.total_inodes); |
53 | ||
ad1fee96 YS |
54 | return 0; |
55 | } | |
56 | ||
ec9595c0 JL |
57 | /** |
58 | * ceph_new_inode - allocate a new inode in advance of an expected create | |
59 | * @dir: parent directory for new inode | |
60 | * @dentry: dentry that may eventually point to new inode | |
61 | * @mode: mode of new inode | |
62 | * @as_ctx: pointer to inherited security context | |
63 | * | |
64 | * Allocate a new inode in advance of an operation to create a new inode. | |
65 | * This allocates the inode and sets up the acl_sec_ctx with appropriate | |
66 | * info for the new inode. | |
67 | * | |
68 | * Returns a pointer to the new inode or an ERR_PTR. | |
69 | */ | |
70 | struct inode *ceph_new_inode(struct inode *dir, struct dentry *dentry, | |
71 | umode_t *mode, struct ceph_acl_sec_ctx *as_ctx) | |
72 | { | |
73 | int err; | |
74 | struct inode *inode; | |
75 | ||
76 | inode = new_inode(dir->i_sb); | |
77 | if (!inode) | |
78 | return ERR_PTR(-ENOMEM); | |
79 | ||
80 | if (!S_ISLNK(*mode)) { | |
81 | err = ceph_pre_init_acls(dir, mode, as_ctx); | |
82 | if (err < 0) | |
83 | goto out_err; | |
84 | } | |
85 | ||
6b5717bd JL |
86 | inode->i_state = 0; |
87 | inode->i_mode = *mode; | |
88 | ||
ec9595c0 JL |
89 | err = ceph_security_init_secctx(dentry, *mode, as_ctx); |
90 | if (err < 0) | |
91 | goto out_err; | |
92 | ||
6b5717bd JL |
93 | err = ceph_fscrypt_prepare_context(dir, inode, as_ctx); |
94 | if (err) | |
95 | goto out_err; | |
96 | ||
ec9595c0 JL |
97 | return inode; |
98 | out_err: | |
99 | iput(inode); | |
100 | return ERR_PTR(err); | |
101 | } | |
102 | ||
103 | void ceph_as_ctx_to_req(struct ceph_mds_request *req, | |
104 | struct ceph_acl_sec_ctx *as_ctx) | |
105 | { | |
106 | if (as_ctx->pagelist) { | |
107 | req->r_pagelist = as_ctx->pagelist; | |
108 | as_ctx->pagelist = NULL; | |
109 | } | |
6b5717bd | 110 | ceph_fscrypt_as_ctx_to_req(req, as_ctx); |
ec9595c0 JL |
111 | } |
112 | ||
113 | /** | |
114 | * ceph_get_inode - find or create/hash a new inode | |
115 | * @sb: superblock to search and allocate in | |
116 | * @vino: vino to search for | |
117 | * @newino: optional new inode to insert if one isn't found (may be NULL) | |
118 | * | |
119 | * Search for or insert a new inode into the hash for the given vino, and | |
120 | * return a reference to it. If new is non-NULL, its reference is consumed. | |
121 | */ | |
122 | struct inode *ceph_get_inode(struct super_block *sb, struct ceph_vino vino, | |
123 | struct inode *newino) | |
355da1eb SW |
124 | { |
125 | struct inode *inode; | |
355da1eb | 126 | |
d4f6b31d JL |
127 | if (ceph_vino_is_reserved(vino)) |
128 | return ERR_PTR(-EREMOTEIO); | |
129 | ||
ec9595c0 JL |
130 | if (newino) { |
131 | inode = inode_insert5(newino, (unsigned long)vino.ino, | |
132 | ceph_ino_compare, ceph_set_ino_cb, &vino); | |
133 | if (inode != newino) | |
134 | iput(newino); | |
135 | } else { | |
136 | inode = iget5_locked(sb, (unsigned long)vino.ino, | |
137 | ceph_ino_compare, ceph_set_ino_cb, &vino); | |
138 | } | |
139 | ||
140 | if (!inode) { | |
141 | dout("No inode found for %llx.%llx\n", vino.ino, vino.snap); | |
355da1eb | 142 | return ERR_PTR(-ENOMEM); |
ec9595c0 | 143 | } |
355da1eb | 144 | |
ebce3eb2 JL |
145 | dout("get_inode on %llu=%llx.%llx got %p new %d\n", ceph_present_inode(inode), |
146 | ceph_vinop(inode), inode, !!(inode->i_state & I_NEW)); | |
355da1eb SW |
147 | return inode; |
148 | } | |
149 | ||
150 | /* | |
151 | * get/constuct snapdir inode for a given directory | |
152 | */ | |
153 | struct inode *ceph_get_snapdir(struct inode *parent) | |
154 | { | |
155 | struct ceph_vino vino = { | |
156 | .ino = ceph_ino(parent), | |
157 | .snap = CEPH_SNAPDIR, | |
158 | }; | |
ec9595c0 | 159 | struct inode *inode = ceph_get_inode(parent->i_sb, vino, NULL); |
b377ff13 | 160 | struct ceph_inode_info *ci = ceph_inode(inode); |
355da1eb | 161 | |
355da1eb | 162 | if (IS_ERR(inode)) |
7e34bc52 | 163 | return inode; |
3e10a15f JL |
164 | |
165 | if (!S_ISDIR(parent->i_mode)) { | |
166 | pr_warn_once("bad snapdir parent type (mode=0%o)\n", | |
167 | parent->i_mode); | |
322794d3 | 168 | goto err; |
3e10a15f JL |
169 | } |
170 | ||
171 | if (!(inode->i_state & I_NEW) && !S_ISDIR(inode->i_mode)) { | |
172 | pr_warn_once("bad snapdir inode type (mode=0%o)\n", | |
173 | inode->i_mode); | |
322794d3 | 174 | goto err; |
3e10a15f JL |
175 | } |
176 | ||
355da1eb SW |
177 | inode->i_mode = parent->i_mode; |
178 | inode->i_uid = parent->i_uid; | |
179 | inode->i_gid = parent->i_gid; | |
ef915725 LH |
180 | inode->i_mtime = parent->i_mtime; |
181 | inode->i_ctime = parent->i_ctime; | |
182 | inode->i_atime = parent->i_atime; | |
b377ff13 | 183 | ci->i_rbytes = 0; |
ef915725 | 184 | ci->i_btime = ceph_inode(parent)->i_btime; |
893e456b | 185 | |
d3c51ae1 JL |
186 | if (inode->i_state & I_NEW) { |
187 | inode->i_op = &ceph_snapdir_iops; | |
188 | inode->i_fop = &ceph_snapdir_fops; | |
189 | ci->i_snap_caps = CEPH_CAP_PIN; /* so we can open */ | |
893e456b | 190 | unlock_new_inode(inode); |
d3c51ae1 | 191 | } |
893e456b | 192 | |
355da1eb | 193 | return inode; |
322794d3 XL |
194 | err: |
195 | if ((inode->i_state & I_NEW)) | |
196 | discard_new_inode(inode); | |
197 | else | |
198 | iput(inode); | |
199 | return ERR_PTR(-ENOTDIR); | |
355da1eb SW |
200 | } |
201 | ||
202 | const struct inode_operations ceph_file_iops = { | |
203 | .permission = ceph_permission, | |
204 | .setattr = ceph_setattr, | |
205 | .getattr = ceph_getattr, | |
355da1eb | 206 | .listxattr = ceph_listxattr, |
cac2f8b8 | 207 | .get_inode_acl = ceph_get_acl, |
72466d0b | 208 | .set_acl = ceph_set_acl, |
355da1eb SW |
209 | }; |
210 | ||
211 | ||
212 | /* | |
213 | * We use a 'frag tree' to keep track of the MDS's directory fragments | |
214 | * for a given inode (usually there is just a single fragment). We | |
215 | * need to know when a child frag is delegated to a new MDS, or when | |
216 | * it is flagged as replicated, so we can direct our requests | |
217 | * accordingly. | |
218 | */ | |
219 | ||
220 | /* | |
221 | * find/create a frag in the tree | |
222 | */ | |
223 | static struct ceph_inode_frag *__get_or_create_frag(struct ceph_inode_info *ci, | |
224 | u32 f) | |
225 | { | |
226 | struct rb_node **p; | |
227 | struct rb_node *parent = NULL; | |
228 | struct ceph_inode_frag *frag; | |
229 | int c; | |
230 | ||
231 | p = &ci->i_fragtree.rb_node; | |
232 | while (*p) { | |
233 | parent = *p; | |
234 | frag = rb_entry(parent, struct ceph_inode_frag, node); | |
235 | c = ceph_frag_compare(f, frag->frag); | |
236 | if (c < 0) | |
237 | p = &(*p)->rb_left; | |
238 | else if (c > 0) | |
239 | p = &(*p)->rb_right; | |
240 | else | |
241 | return frag; | |
242 | } | |
243 | ||
244 | frag = kmalloc(sizeof(*frag), GFP_NOFS); | |
51308806 | 245 | if (!frag) |
355da1eb | 246 | return ERR_PTR(-ENOMEM); |
51308806 | 247 | |
355da1eb SW |
248 | frag->frag = f; |
249 | frag->split_by = 0; | |
250 | frag->mds = -1; | |
251 | frag->ndist = 0; | |
252 | ||
253 | rb_link_node(&frag->node, parent, p); | |
254 | rb_insert_color(&frag->node, &ci->i_fragtree); | |
255 | ||
256 | dout("get_or_create_frag added %llx.%llx frag %x\n", | |
874c8ca1 | 257 | ceph_vinop(&ci->netfs.inode), f); |
355da1eb SW |
258 | return frag; |
259 | } | |
260 | ||
261 | /* | |
262 | * find a specific frag @f | |
263 | */ | |
264 | struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci, u32 f) | |
265 | { | |
266 | struct rb_node *n = ci->i_fragtree.rb_node; | |
267 | ||
268 | while (n) { | |
269 | struct ceph_inode_frag *frag = | |
270 | rb_entry(n, struct ceph_inode_frag, node); | |
271 | int c = ceph_frag_compare(f, frag->frag); | |
272 | if (c < 0) | |
273 | n = n->rb_left; | |
274 | else if (c > 0) | |
275 | n = n->rb_right; | |
276 | else | |
277 | return frag; | |
278 | } | |
279 | return NULL; | |
280 | } | |
281 | ||
282 | /* | |
283 | * Choose frag containing the given value @v. If @pfrag is | |
284 | * specified, copy the frag delegation info to the caller if | |
285 | * it is present. | |
286 | */ | |
3e7fbe9c YZ |
287 | static u32 __ceph_choose_frag(struct ceph_inode_info *ci, u32 v, |
288 | struct ceph_inode_frag *pfrag, int *found) | |
355da1eb SW |
289 | { |
290 | u32 t = ceph_frag_make(0, 0); | |
291 | struct ceph_inode_frag *frag; | |
292 | unsigned nway, i; | |
293 | u32 n; | |
294 | ||
295 | if (found) | |
296 | *found = 0; | |
297 | ||
355da1eb SW |
298 | while (1) { |
299 | WARN_ON(!ceph_frag_contains_value(t, v)); | |
300 | frag = __ceph_find_frag(ci, t); | |
301 | if (!frag) | |
302 | break; /* t is a leaf */ | |
303 | if (frag->split_by == 0) { | |
304 | if (pfrag) | |
305 | memcpy(pfrag, frag, sizeof(*pfrag)); | |
306 | if (found) | |
307 | *found = 1; | |
308 | break; | |
309 | } | |
310 | ||
311 | /* choose child */ | |
312 | nway = 1 << frag->split_by; | |
313 | dout("choose_frag(%x) %x splits by %d (%d ways)\n", v, t, | |
314 | frag->split_by, nway); | |
315 | for (i = 0; i < nway; i++) { | |
316 | n = ceph_frag_make_child(t, frag->split_by, i); | |
317 | if (ceph_frag_contains_value(n, v)) { | |
318 | t = n; | |
319 | break; | |
320 | } | |
321 | } | |
322 | BUG_ON(i == nway); | |
323 | } | |
324 | dout("choose_frag(%x) = %x\n", v, t); | |
325 | ||
355da1eb SW |
326 | return t; |
327 | } | |
328 | ||
3e7fbe9c YZ |
329 | u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v, |
330 | struct ceph_inode_frag *pfrag, int *found) | |
331 | { | |
332 | u32 ret; | |
333 | mutex_lock(&ci->i_fragtree_mutex); | |
334 | ret = __ceph_choose_frag(ci, v, pfrag, found); | |
335 | mutex_unlock(&ci->i_fragtree_mutex); | |
336 | return ret; | |
337 | } | |
338 | ||
355da1eb SW |
339 | /* |
340 | * Process dirfrag (delegation) info from the mds. Include leaf | |
341 | * fragment in tree ONLY if ndist > 0. Otherwise, only | |
342 | * branches/splits are included in i_fragtree) | |
343 | */ | |
344 | static int ceph_fill_dirfrag(struct inode *inode, | |
345 | struct ceph_mds_reply_dirfrag *dirinfo) | |
346 | { | |
347 | struct ceph_inode_info *ci = ceph_inode(inode); | |
348 | struct ceph_inode_frag *frag; | |
349 | u32 id = le32_to_cpu(dirinfo->frag); | |
350 | int mds = le32_to_cpu(dirinfo->auth); | |
351 | int ndist = le32_to_cpu(dirinfo->ndist); | |
8d08503c | 352 | int diri_auth = -1; |
355da1eb SW |
353 | int i; |
354 | int err = 0; | |
355 | ||
8d08503c YZ |
356 | spin_lock(&ci->i_ceph_lock); |
357 | if (ci->i_auth_cap) | |
358 | diri_auth = ci->i_auth_cap->mds; | |
359 | spin_unlock(&ci->i_ceph_lock); | |
360 | ||
42172119 YZ |
361 | if (mds == -1) /* CDIR_AUTH_PARENT */ |
362 | mds = diri_auth; | |
363 | ||
355da1eb | 364 | mutex_lock(&ci->i_fragtree_mutex); |
8d08503c | 365 | if (ndist == 0 && mds == diri_auth) { |
355da1eb SW |
366 | /* no delegation info needed. */ |
367 | frag = __ceph_find_frag(ci, id); | |
368 | if (!frag) | |
369 | goto out; | |
370 | if (frag->split_by == 0) { | |
371 | /* tree leaf, remove */ | |
372 | dout("fill_dirfrag removed %llx.%llx frag %x" | |
373 | " (no ref)\n", ceph_vinop(inode), id); | |
374 | rb_erase(&frag->node, &ci->i_fragtree); | |
375 | kfree(frag); | |
376 | } else { | |
377 | /* tree branch, keep and clear */ | |
378 | dout("fill_dirfrag cleared %llx.%llx frag %x" | |
379 | " referral\n", ceph_vinop(inode), id); | |
380 | frag->mds = -1; | |
381 | frag->ndist = 0; | |
382 | } | |
383 | goto out; | |
384 | } | |
385 | ||
386 | ||
387 | /* find/add this frag to store mds delegation info */ | |
388 | frag = __get_or_create_frag(ci, id); | |
389 | if (IS_ERR(frag)) { | |
390 | /* this is not the end of the world; we can continue | |
391 | with bad/inaccurate delegation info */ | |
392 | pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n", | |
393 | ceph_vinop(inode), le32_to_cpu(dirinfo->frag)); | |
394 | err = -ENOMEM; | |
395 | goto out; | |
396 | } | |
397 | ||
398 | frag->mds = mds; | |
399 | frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP); | |
400 | for (i = 0; i < frag->ndist; i++) | |
401 | frag->dist[i] = le32_to_cpu(dirinfo->dist[i]); | |
402 | dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n", | |
403 | ceph_vinop(inode), frag->frag, frag->ndist); | |
404 | ||
405 | out: | |
406 | mutex_unlock(&ci->i_fragtree_mutex); | |
407 | return err; | |
408 | } | |
409 | ||
a407846e YZ |
410 | static int frag_tree_split_cmp(const void *l, const void *r) |
411 | { | |
412 | struct ceph_frag_tree_split *ls = (struct ceph_frag_tree_split*)l; | |
413 | struct ceph_frag_tree_split *rs = (struct ceph_frag_tree_split*)r; | |
fe2ed425 JL |
414 | return ceph_frag_compare(le32_to_cpu(ls->frag), |
415 | le32_to_cpu(rs->frag)); | |
a407846e YZ |
416 | } |
417 | ||
a4b7431f YZ |
418 | static bool is_frag_child(u32 f, struct ceph_inode_frag *frag) |
419 | { | |
420 | if (!frag) | |
421 | return f == ceph_frag_make(0, 0); | |
422 | if (ceph_frag_bits(f) != ceph_frag_bits(frag->frag) + frag->split_by) | |
423 | return false; | |
424 | return ceph_frag_contains_value(frag->frag, ceph_frag_value(f)); | |
425 | } | |
426 | ||
3e7fbe9c YZ |
427 | static int ceph_fill_fragtree(struct inode *inode, |
428 | struct ceph_frag_tree_head *fragtree, | |
429 | struct ceph_mds_reply_dirfrag *dirinfo) | |
430 | { | |
431 | struct ceph_inode_info *ci = ceph_inode(inode); | |
a4b7431f | 432 | struct ceph_inode_frag *frag, *prev_frag = NULL; |
3e7fbe9c | 433 | struct rb_node *rb_node; |
1b1bc16d YZ |
434 | unsigned i, split_by, nsplits; |
435 | u32 id; | |
3e7fbe9c YZ |
436 | bool update = false; |
437 | ||
438 | mutex_lock(&ci->i_fragtree_mutex); | |
439 | nsplits = le32_to_cpu(fragtree->nsplits); | |
1b1bc16d YZ |
440 | if (nsplits != ci->i_fragtree_nsplits) { |
441 | update = true; | |
442 | } else if (nsplits) { | |
8032bf12 | 443 | i = get_random_u32_below(nsplits); |
3e7fbe9c YZ |
444 | id = le32_to_cpu(fragtree->splits[i].frag); |
445 | if (!__ceph_find_frag(ci, id)) | |
446 | update = true; | |
447 | } else if (!RB_EMPTY_ROOT(&ci->i_fragtree)) { | |
448 | rb_node = rb_first(&ci->i_fragtree); | |
449 | frag = rb_entry(rb_node, struct ceph_inode_frag, node); | |
450 | if (frag->frag != ceph_frag_make(0, 0) || rb_next(rb_node)) | |
451 | update = true; | |
452 | } | |
453 | if (!update && dirinfo) { | |
454 | id = le32_to_cpu(dirinfo->frag); | |
455 | if (id != __ceph_choose_frag(ci, id, NULL, NULL)) | |
456 | update = true; | |
457 | } | |
458 | if (!update) | |
459 | goto out_unlock; | |
460 | ||
a407846e YZ |
461 | if (nsplits > 1) { |
462 | sort(fragtree->splits, nsplits, sizeof(fragtree->splits[0]), | |
463 | frag_tree_split_cmp, NULL); | |
464 | } | |
465 | ||
3e7fbe9c YZ |
466 | dout("fill_fragtree %llx.%llx\n", ceph_vinop(inode)); |
467 | rb_node = rb_first(&ci->i_fragtree); | |
468 | for (i = 0; i < nsplits; i++) { | |
469 | id = le32_to_cpu(fragtree->splits[i].frag); | |
1b1bc16d YZ |
470 | split_by = le32_to_cpu(fragtree->splits[i].by); |
471 | if (split_by == 0 || ceph_frag_bits(id) + split_by > 24) { | |
472 | pr_err("fill_fragtree %llx.%llx invalid split %d/%u, " | |
473 | "frag %x split by %d\n", ceph_vinop(inode), | |
474 | i, nsplits, id, split_by); | |
475 | continue; | |
476 | } | |
3e7fbe9c YZ |
477 | frag = NULL; |
478 | while (rb_node) { | |
479 | frag = rb_entry(rb_node, struct ceph_inode_frag, node); | |
480 | if (ceph_frag_compare(frag->frag, id) >= 0) { | |
481 | if (frag->frag != id) | |
482 | frag = NULL; | |
483 | else | |
484 | rb_node = rb_next(rb_node); | |
485 | break; | |
486 | } | |
487 | rb_node = rb_next(rb_node); | |
a4b7431f YZ |
488 | /* delete stale split/leaf node */ |
489 | if (frag->split_by > 0 || | |
490 | !is_frag_child(frag->frag, prev_frag)) { | |
491 | rb_erase(&frag->node, &ci->i_fragtree); | |
1b1bc16d YZ |
492 | if (frag->split_by > 0) |
493 | ci->i_fragtree_nsplits--; | |
a4b7431f YZ |
494 | kfree(frag); |
495 | } | |
3e7fbe9c YZ |
496 | frag = NULL; |
497 | } | |
498 | if (!frag) { | |
499 | frag = __get_or_create_frag(ci, id); | |
500 | if (IS_ERR(frag)) | |
501 | continue; | |
502 | } | |
1b1bc16d YZ |
503 | if (frag->split_by == 0) |
504 | ci->i_fragtree_nsplits++; | |
505 | frag->split_by = split_by; | |
3e7fbe9c | 506 | dout(" frag %x split by %d\n", frag->frag, frag->split_by); |
a4b7431f | 507 | prev_frag = frag; |
3e7fbe9c YZ |
508 | } |
509 | while (rb_node) { | |
510 | frag = rb_entry(rb_node, struct ceph_inode_frag, node); | |
511 | rb_node = rb_next(rb_node); | |
a4b7431f YZ |
512 | /* delete stale split/leaf node */ |
513 | if (frag->split_by > 0 || | |
514 | !is_frag_child(frag->frag, prev_frag)) { | |
515 | rb_erase(&frag->node, &ci->i_fragtree); | |
1b1bc16d YZ |
516 | if (frag->split_by > 0) |
517 | ci->i_fragtree_nsplits--; | |
a4b7431f YZ |
518 | kfree(frag); |
519 | } | |
3e7fbe9c YZ |
520 | } |
521 | out_unlock: | |
522 | mutex_unlock(&ci->i_fragtree_mutex); | |
523 | return 0; | |
524 | } | |
355da1eb SW |
525 | |
526 | /* | |
527 | * initialize a newly allocated inode. | |
528 | */ | |
529 | struct inode *ceph_alloc_inode(struct super_block *sb) | |
530 | { | |
531 | struct ceph_inode_info *ci; | |
532 | int i; | |
533 | ||
fd60b288 | 534 | ci = alloc_inode_sb(sb, ceph_inode_cachep, GFP_NOFS); |
355da1eb SW |
535 | if (!ci) |
536 | return NULL; | |
537 | ||
874c8ca1 | 538 | dout("alloc_inode %p\n", &ci->netfs.inode); |
355da1eb | 539 | |
bc899ee1 | 540 | /* Set parameters for the netfs library */ |
e81fb419 | 541 | netfs_inode_init(&ci->netfs, &ceph_netfs_ops); |
bc899ee1 | 542 | |
be655596 SW |
543 | spin_lock_init(&ci->i_ceph_lock); |
544 | ||
355da1eb | 545 | ci->i_version = 0; |
31c542a1 | 546 | ci->i_inline_version = 0; |
355da1eb SW |
547 | ci->i_time_warp_seq = 0; |
548 | ci->i_ceph_flags = 0; | |
fdd4e158 YZ |
549 | atomic64_set(&ci->i_ordered_count, 1); |
550 | atomic64_set(&ci->i_release_count, 1); | |
551 | atomic64_set(&ci->i_complete_seq[0], 0); | |
552 | atomic64_set(&ci->i_complete_seq[1], 0); | |
355da1eb SW |
553 | ci->i_symlink = NULL; |
554 | ||
fb18a575 LH |
555 | ci->i_max_bytes = 0; |
556 | ci->i_max_files = 0; | |
557 | ||
6c0f3af7 | 558 | memset(&ci->i_dir_layout, 0, sizeof(ci->i_dir_layout)); |
785892fe | 559 | memset(&ci->i_cached_layout, 0, sizeof(ci->i_cached_layout)); |
30c156d9 | 560 | RCU_INIT_POINTER(ci->i_layout.pool_ns, NULL); |
6c0f3af7 | 561 | |
355da1eb SW |
562 | ci->i_fragtree = RB_ROOT; |
563 | mutex_init(&ci->i_fragtree_mutex); | |
564 | ||
565 | ci->i_xattrs.blob = NULL; | |
566 | ci->i_xattrs.prealloc_blob = NULL; | |
567 | ci->i_xattrs.dirty = false; | |
568 | ci->i_xattrs.index = RB_ROOT; | |
569 | ci->i_xattrs.count = 0; | |
570 | ci->i_xattrs.names_size = 0; | |
571 | ci->i_xattrs.vals_size = 0; | |
572 | ci->i_xattrs.version = 0; | |
573 | ci->i_xattrs.index_version = 0; | |
574 | ||
575 | ci->i_caps = RB_ROOT; | |
576 | ci->i_auth_cap = NULL; | |
577 | ci->i_dirty_caps = 0; | |
578 | ci->i_flushing_caps = 0; | |
579 | INIT_LIST_HEAD(&ci->i_dirty_item); | |
580 | INIT_LIST_HEAD(&ci->i_flushing_item); | |
f66fd9f0 | 581 | ci->i_prealloc_cap_flush = NULL; |
e4500b5e | 582 | INIT_LIST_HEAD(&ci->i_cap_flush_list); |
355da1eb | 583 | init_waitqueue_head(&ci->i_cap_wq); |
355da1eb SW |
584 | ci->i_hold_caps_max = 0; |
585 | INIT_LIST_HEAD(&ci->i_cap_delay_list); | |
355da1eb SW |
586 | INIT_LIST_HEAD(&ci->i_cap_snaps); |
587 | ci->i_head_snapc = NULL; | |
588 | ci->i_snap_caps = 0; | |
589 | ||
719a2514 | 590 | ci->i_last_rd = ci->i_last_wr = jiffies - 3600 * HZ; |
774a6a11 | 591 | for (i = 0; i < CEPH_FILE_MODE_BITS; i++) |
355da1eb SW |
592 | ci->i_nr_by_mode[i] = 0; |
593 | ||
b0d7c223 | 594 | mutex_init(&ci->i_truncate_mutex); |
355da1eb SW |
595 | ci->i_truncate_seq = 0; |
596 | ci->i_truncate_size = 0; | |
597 | ci->i_truncate_pending = 0; | |
598 | ||
599 | ci->i_max_size = 0; | |
600 | ci->i_reported_size = 0; | |
601 | ci->i_wanted_max_size = 0; | |
602 | ci->i_requested_max_size = 0; | |
603 | ||
604 | ci->i_pin_ref = 0; | |
605 | ci->i_rd_ref = 0; | |
606 | ci->i_rdcache_ref = 0; | |
607 | ci->i_wr_ref = 0; | |
d3d0720d | 608 | ci->i_wb_ref = 0; |
f85122af | 609 | ci->i_fx_ref = 0; |
355da1eb SW |
610 | ci->i_wrbuffer_ref = 0; |
611 | ci->i_wrbuffer_ref_head = 0; | |
89aa5930 | 612 | atomic_set(&ci->i_filelock_ref, 0); |
1e9c2eb6 | 613 | atomic_set(&ci->i_shared_gen, 1); |
355da1eb SW |
614 | ci->i_rdcache_gen = 0; |
615 | ci->i_rdcache_revoking = 0; | |
616 | ||
355da1eb | 617 | INIT_LIST_HEAD(&ci->i_unsafe_dirops); |
68cd5b4b | 618 | INIT_LIST_HEAD(&ci->i_unsafe_iops); |
355da1eb SW |
619 | spin_lock_init(&ci->i_unsafe_lock); |
620 | ||
621 | ci->i_snap_realm = NULL; | |
622 | INIT_LIST_HEAD(&ci->i_snap_realm_item); | |
623 | INIT_LIST_HEAD(&ci->i_snap_flush_item); | |
624 | ||
1cf89a8d YZ |
625 | INIT_WORK(&ci->i_work, ceph_inode_work); |
626 | ci->i_work_mask = 0; | |
245ce991 | 627 | memset(&ci->i_btime, '\0', sizeof(ci->i_btime)); |
2d332d5b JL |
628 | #ifdef CONFIG_FS_ENCRYPTION |
629 | ci->fscrypt_auth = NULL; | |
630 | ci->fscrypt_auth_len = 0; | |
631 | #endif | |
874c8ca1 | 632 | return &ci->netfs.inode; |
355da1eb SW |
633 | } |
634 | ||
cfa6d412 | 635 | void ceph_free_inode(struct inode *inode) |
fa0d7e3d | 636 | { |
fa0d7e3d NP |
637 | struct ceph_inode_info *ci = ceph_inode(inode); |
638 | ||
daf5cc27 | 639 | kfree(ci->i_symlink); |
2d332d5b JL |
640 | #ifdef CONFIG_FS_ENCRYPTION |
641 | kfree(ci->fscrypt_auth); | |
642 | #endif | |
fa0d7e3d NP |
643 | kmem_cache_free(ceph_inode_cachep, ci); |
644 | } | |
645 | ||
87bc5b89 | 646 | void ceph_evict_inode(struct inode *inode) |
355da1eb SW |
647 | { |
648 | struct ceph_inode_info *ci = ceph_inode(inode); | |
1dd8d470 | 649 | struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb); |
355da1eb SW |
650 | struct ceph_inode_frag *frag; |
651 | struct rb_node *n; | |
652 | ||
87bc5b89 YZ |
653 | dout("evict_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode)); |
654 | ||
1dd8d470 XL |
655 | percpu_counter_dec(&mdsc->metric.total_inodes); |
656 | ||
87bc5b89 | 657 | truncate_inode_pages_final(&inode->i_data); |
400e1286 JL |
658 | if (inode->i_state & I_PINNING_FSCACHE_WB) |
659 | ceph_fscache_unuse_cookie(inode, true); | |
87bc5b89 | 660 | clear_inode(inode); |
355da1eb | 661 | |
99ccbd22 | 662 | ceph_fscache_unregister_inode_cookie(ci); |
2d332d5b | 663 | fscrypt_put_encryption_info(inode); |
99ccbd22 | 664 | |
d6e47819 | 665 | __ceph_remove_caps(ci); |
355da1eb | 666 | |
55ab5520 | 667 | if (__ceph_has_quota(ci, QUOTA_GET_ANY)) |
d557c48d LH |
668 | ceph_adjust_quota_realms_count(inode, false); |
669 | ||
8b218b8a SW |
670 | /* |
671 | * we may still have a snap_realm reference if there are stray | |
d9df2783 | 672 | * caps in i_snap_caps. |
8b218b8a SW |
673 | */ |
674 | if (ci->i_snap_realm) { | |
75c9627e | 675 | if (ceph_snap(inode) == CEPH_NOSNAP) { |
75c9627e | 676 | dout(" dropping residual ref to snap realm %p\n", |
0ba92e1c JL |
677 | ci->i_snap_realm); |
678 | ceph_change_snap_realm(inode, NULL); | |
75c9627e YZ |
679 | } else { |
680 | ceph_put_snapid_map(mdsc, ci->i_snapid_map); | |
681 | ci->i_snap_realm = NULL; | |
682 | } | |
8b218b8a SW |
683 | } |
684 | ||
355da1eb SW |
685 | while ((n = rb_first(&ci->i_fragtree)) != NULL) { |
686 | frag = rb_entry(n, struct ceph_inode_frag, node); | |
687 | rb_erase(n, &ci->i_fragtree); | |
688 | kfree(frag); | |
689 | } | |
1b1bc16d | 690 | ci->i_fragtree_nsplits = 0; |
355da1eb SW |
691 | |
692 | __ceph_destroy_xattrs(ci); | |
b6c1d5b8 SW |
693 | if (ci->i_xattrs.blob) |
694 | ceph_buffer_put(ci->i_xattrs.blob); | |
695 | if (ci->i_xattrs.prealloc_blob) | |
696 | ceph_buffer_put(ci->i_xattrs.prealloc_blob); | |
355da1eb | 697 | |
779fe0fb | 698 | ceph_put_string(rcu_dereference_raw(ci->i_layout.pool_ns)); |
785892fe | 699 | ceph_put_string(rcu_dereference_raw(ci->i_cached_layout.pool_ns)); |
355da1eb SW |
700 | } |
701 | ||
224a7542 YZ |
702 | static inline blkcnt_t calc_inode_blocks(u64 size) |
703 | { | |
704 | return (size + (1<<9) - 1) >> 9; | |
705 | } | |
706 | ||
355da1eb SW |
707 | /* |
708 | * Helpers to fill in size, ctime, mtime, and atime. We have to be | |
709 | * careful because either the client or MDS may have more up to date | |
710 | * info, depending on which capabilities are held, and whether | |
711 | * time_warp_seq or truncate_seq have increased. (Ordinarily, mtime | |
712 | * and size are monotonically increasing, except when utimes() or | |
713 | * truncate() increments the corresponding _seq values.) | |
714 | */ | |
715 | int ceph_fill_file_size(struct inode *inode, int issued, | |
716 | u32 truncate_seq, u64 truncate_size, u64 size) | |
717 | { | |
718 | struct ceph_inode_info *ci = ceph_inode(inode); | |
719 | int queue_trunc = 0; | |
2d6795fb | 720 | loff_t isize = i_size_read(inode); |
355da1eb SW |
721 | |
722 | if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 || | |
2d6795fb JL |
723 | (truncate_seq == ci->i_truncate_seq && size > isize)) { |
724 | dout("size %lld -> %llu\n", isize, size); | |
a3d714c3 YZ |
725 | if (size > 0 && S_ISDIR(inode->i_mode)) { |
726 | pr_err("fill_file_size non-zero size for directory\n"); | |
727 | size = 0; | |
728 | } | |
99c88e69 | 729 | i_size_write(inode, size); |
224a7542 | 730 | inode->i_blocks = calc_inode_blocks(size); |
400e1286 JL |
731 | /* |
732 | * If we're expanding, then we should be able to just update | |
733 | * the existing cookie. | |
734 | */ | |
735 | if (size > isize) | |
736 | ceph_fscache_update(inode); | |
355da1eb SW |
737 | ci->i_reported_size = size; |
738 | if (truncate_seq != ci->i_truncate_seq) { | |
739 | dout("truncate_seq %u -> %u\n", | |
740 | ci->i_truncate_seq, truncate_seq); | |
741 | ci->i_truncate_seq = truncate_seq; | |
b0d7c223 YZ |
742 | |
743 | /* the MDS should have revoked these caps */ | |
744 | WARN_ON_ONCE(issued & (CEPH_CAP_FILE_EXCL | | |
745 | CEPH_CAP_FILE_RD | | |
746 | CEPH_CAP_FILE_WR | | |
747 | CEPH_CAP_FILE_LAZYIO)); | |
3d497d85 YS |
748 | /* |
749 | * If we hold relevant caps, or in the case where we're | |
750 | * not the only client referencing this file and we | |
751 | * don't hold those caps, then we need to check whether | |
752 | * the file is either opened or mmaped | |
753 | */ | |
b0d7c223 YZ |
754 | if ((issued & (CEPH_CAP_FILE_CACHE| |
755 | CEPH_CAP_FILE_BUFFER)) || | |
3d497d85 | 756 | mapping_mapped(inode->i_mapping) || |
719a2514 | 757 | __ceph_is_file_opened(ci)) { |
355da1eb SW |
758 | ci->i_truncate_pending++; |
759 | queue_trunc = 1; | |
760 | } | |
761 | } | |
762 | } | |
763 | if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 && | |
764 | ci->i_truncate_size != truncate_size) { | |
765 | dout("truncate_size %lld -> %llu\n", ci->i_truncate_size, | |
766 | truncate_size); | |
767 | ci->i_truncate_size = truncate_size; | |
768 | } | |
769 | return queue_trunc; | |
770 | } | |
771 | ||
772 | void ceph_fill_file_time(struct inode *inode, int issued, | |
9bbeab41 AB |
773 | u64 time_warp_seq, struct timespec64 *ctime, |
774 | struct timespec64 *mtime, struct timespec64 *atime) | |
355da1eb SW |
775 | { |
776 | struct ceph_inode_info *ci = ceph_inode(inode); | |
777 | int warn = 0; | |
778 | ||
779 | if (issued & (CEPH_CAP_FILE_EXCL| | |
780 | CEPH_CAP_FILE_WR| | |
d8672d64 SW |
781 | CEPH_CAP_FILE_BUFFER| |
782 | CEPH_CAP_AUTH_EXCL| | |
783 | CEPH_CAP_XATTR_EXCL)) { | |
ffdeec7a | 784 | if (ci->i_version == 0 || |
9bbeab41 | 785 | timespec64_compare(ctime, &inode->i_ctime) > 0) { |
13442b03 | 786 | dout("ctime %lld.%09ld -> %lld.%09ld inc w/ cap\n", |
9bbeab41 AB |
787 | inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec, |
788 | ctime->tv_sec, ctime->tv_nsec); | |
789 | inode->i_ctime = *ctime; | |
355da1eb | 790 | } |
ffdeec7a YZ |
791 | if (ci->i_version == 0 || |
792 | ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) { | |
355da1eb | 793 | /* the MDS did a utimes() */ |
13442b03 | 794 | dout("mtime %lld.%09ld -> %lld.%09ld " |
355da1eb | 795 | "tw %d -> %d\n", |
9bbeab41 AB |
796 | inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec, |
797 | mtime->tv_sec, mtime->tv_nsec, | |
355da1eb SW |
798 | ci->i_time_warp_seq, (int)time_warp_seq); |
799 | ||
9bbeab41 AB |
800 | inode->i_mtime = *mtime; |
801 | inode->i_atime = *atime; | |
355da1eb SW |
802 | ci->i_time_warp_seq = time_warp_seq; |
803 | } else if (time_warp_seq == ci->i_time_warp_seq) { | |
804 | /* nobody did utimes(); take the max */ | |
9bbeab41 | 805 | if (timespec64_compare(mtime, &inode->i_mtime) > 0) { |
13442b03 | 806 | dout("mtime %lld.%09ld -> %lld.%09ld inc\n", |
9bbeab41 | 807 | inode->i_mtime.tv_sec, |
355da1eb | 808 | inode->i_mtime.tv_nsec, |
9bbeab41 AB |
809 | mtime->tv_sec, mtime->tv_nsec); |
810 | inode->i_mtime = *mtime; | |
355da1eb | 811 | } |
9bbeab41 | 812 | if (timespec64_compare(atime, &inode->i_atime) > 0) { |
13442b03 | 813 | dout("atime %lld.%09ld -> %lld.%09ld inc\n", |
9bbeab41 | 814 | inode->i_atime.tv_sec, |
355da1eb | 815 | inode->i_atime.tv_nsec, |
9bbeab41 AB |
816 | atime->tv_sec, atime->tv_nsec); |
817 | inode->i_atime = *atime; | |
355da1eb SW |
818 | } |
819 | } else if (issued & CEPH_CAP_FILE_EXCL) { | |
820 | /* we did a utimes(); ignore mds values */ | |
821 | } else { | |
822 | warn = 1; | |
823 | } | |
824 | } else { | |
d8672d64 | 825 | /* we have no write|excl caps; whatever the MDS says is true */ |
355da1eb | 826 | if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) { |
9bbeab41 AB |
827 | inode->i_ctime = *ctime; |
828 | inode->i_mtime = *mtime; | |
829 | inode->i_atime = *atime; | |
355da1eb SW |
830 | ci->i_time_warp_seq = time_warp_seq; |
831 | } else { | |
832 | warn = 1; | |
833 | } | |
834 | } | |
835 | if (warn) /* time_warp_seq shouldn't go backwards */ | |
836 | dout("%p mds time_warp_seq %llu < %u\n", | |
837 | inode, time_warp_seq, ci->i_time_warp_seq); | |
838 | } | |
839 | ||
840 | /* | |
841 | * Populate an inode based on info from mds. May be called on new or | |
842 | * existing inodes. | |
843 | */ | |
966c7160 JL |
844 | int ceph_fill_inode(struct inode *inode, struct page *locked_page, |
845 | struct ceph_mds_reply_info_in *iinfo, | |
846 | struct ceph_mds_reply_dirfrag *dirinfo, | |
847 | struct ceph_mds_session *session, int cap_fmode, | |
848 | struct ceph_cap_reservation *caps_reservation) | |
355da1eb | 849 | { |
2678da88 | 850 | struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb); |
355da1eb SW |
851 | struct ceph_mds_reply_inode *info = iinfo->in; |
852 | struct ceph_inode_info *ci = ceph_inode(inode); | |
2af54a72 | 853 | int issued, new_issued, info_caps; |
9bbeab41 | 854 | struct timespec64 mtime, atime, ctime; |
355da1eb | 855 | struct ceph_buffer *xattr_blob = NULL; |
af8a85a4 | 856 | struct ceph_buffer *old_blob = NULL; |
779fe0fb | 857 | struct ceph_string *pool_ns = NULL; |
d9df2783 | 858 | struct ceph_cap *new_cap = NULL; |
355da1eb | 859 | int err = 0; |
d9df2783 | 860 | bool wake = false; |
f98a128a YZ |
861 | bool queue_trunc = false; |
862 | bool new_version = false; | |
31c542a1 | 863 | bool fill_inline = false; |
ed94f87c JL |
864 | umode_t mode = le32_to_cpu(info->mode); |
865 | dev_t rdev = le32_to_cpu(info->rdev); | |
355da1eb | 866 | |
27171ae6 JL |
867 | lockdep_assert_held(&mdsc->snap_rwsem); |
868 | ||
966c7160 | 869 | dout("%s %p ino %llx.%llx v %llu had %llu\n", __func__, |
355da1eb SW |
870 | inode, ceph_vinop(inode), le64_to_cpu(info->version), |
871 | ci->i_version); | |
872 | ||
ed94f87c JL |
873 | /* Once I_NEW is cleared, we can't change type or dev numbers */ |
874 | if (inode->i_state & I_NEW) { | |
875 | inode->i_mode = mode; | |
876 | } else { | |
877 | if (inode_wrong_type(inode, mode)) { | |
878 | pr_warn_once("inode type changed! (ino %llx.%llx is 0%o, mds says 0%o)\n", | |
879 | ceph_vinop(inode), inode->i_mode, mode); | |
880 | return -ESTALE; | |
881 | } | |
882 | ||
883 | if ((S_ISCHR(mode) || S_ISBLK(mode)) && inode->i_rdev != rdev) { | |
884 | pr_warn_once("dev inode rdev changed! (ino %llx.%llx is %u:%u, mds says %u:%u)\n", | |
885 | ceph_vinop(inode), MAJOR(inode->i_rdev), | |
886 | MINOR(inode->i_rdev), MAJOR(rdev), | |
887 | MINOR(rdev)); | |
888 | return -ESTALE; | |
889 | } | |
890 | } | |
891 | ||
2af54a72 YZ |
892 | info_caps = le32_to_cpu(info->cap.caps); |
893 | ||
d9df2783 | 894 | /* prealloc new cap struct */ |
9a6bed4f | 895 | if (info_caps && ceph_snap(inode) == CEPH_NOSNAP) { |
d9df2783 | 896 | new_cap = ceph_get_cap(mdsc, caps_reservation); |
9a6bed4f JL |
897 | if (!new_cap) |
898 | return -ENOMEM; | |
899 | } | |
d9df2783 | 900 | |
355da1eb SW |
901 | /* |
902 | * prealloc xattr data, if it looks like we'll need it. only | |
903 | * if len > 4 (meaning there are actually xattrs; the first 4 | |
904 | * bytes are the xattr count). | |
905 | */ | |
906 | if (iinfo->xattr_len > 4) { | |
b6c1d5b8 | 907 | xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS); |
355da1eb | 908 | if (!xattr_blob) |
966c7160 | 909 | pr_err("%s ENOMEM xattr blob %d bytes\n", __func__, |
355da1eb SW |
910 | iinfo->xattr_len); |
911 | } | |
912 | ||
779fe0fb YZ |
913 | if (iinfo->pool_ns_len > 0) |
914 | pool_ns = ceph_find_or_create_string(iinfo->pool_ns_data, | |
915 | iinfo->pool_ns_len); | |
916 | ||
75c9627e YZ |
917 | if (ceph_snap(inode) != CEPH_NOSNAP && !ci->i_snapid_map) |
918 | ci->i_snapid_map = ceph_get_snapid_map(mdsc, ceph_snap(inode)); | |
919 | ||
be655596 | 920 | spin_lock(&ci->i_ceph_lock); |
355da1eb SW |
921 | |
922 | /* | |
923 | * provided version will be odd if inode value is projected, | |
8bd59e01 SW |
924 | * even if stable. skip the update if we have newer stable |
925 | * info (ours>=theirs, e.g. due to racing mds replies), unless | |
926 | * we are getting projected (unstable) info (in which case the | |
927 | * version is odd, and we want ours>theirs). | |
928 | * us them | |
929 | * 2 2 skip | |
930 | * 3 2 skip | |
931 | * 3 3 update | |
355da1eb | 932 | */ |
f98a128a YZ |
933 | if (ci->i_version == 0 || |
934 | ((info->cap.flags & CEPH_CAP_FLAG_AUTH) && | |
935 | le64_to_cpu(info->version) > (ci->i_version & ~1))) | |
936 | new_version = true; | |
937 | ||
a35ead31 JL |
938 | /* Update change_attribute */ |
939 | inode_set_max_iversion_raw(inode, iinfo->change_attr); | |
940 | ||
2af54a72 YZ |
941 | __ceph_caps_issued(ci, &issued); |
942 | issued |= __ceph_caps_dirty(ci); | |
943 | new_issued = ~issued & info_caps; | |
355da1eb | 944 | |
75067034 LH |
945 | /* directories have fl_stripe_unit set to zero */ |
946 | if (le32_to_cpu(info->layout.fl_stripe_unit)) | |
947 | inode->i_blkbits = | |
948 | fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1; | |
949 | else | |
950 | inode->i_blkbits = CEPH_BLOCK_SHIFT; | |
355da1eb | 951 | |
d557c48d | 952 | __ceph_update_quota(ci, iinfo->max_bytes, iinfo->max_files); |
fb18a575 | 953 | |
2d332d5b JL |
954 | #ifdef CONFIG_FS_ENCRYPTION |
955 | if (iinfo->fscrypt_auth_len && (inode->i_state & I_NEW)) { | |
956 | kfree(ci->fscrypt_auth); | |
957 | ci->fscrypt_auth_len = iinfo->fscrypt_auth_len; | |
958 | ci->fscrypt_auth = iinfo->fscrypt_auth; | |
959 | iinfo->fscrypt_auth = NULL; | |
960 | iinfo->fscrypt_auth_len = 0; | |
961 | inode_set_flags(inode, S_ENCRYPTED, S_ENCRYPTED); | |
962 | } | |
963 | #endif | |
964 | ||
f98a128a YZ |
965 | if ((new_version || (new_issued & CEPH_CAP_AUTH_SHARED)) && |
966 | (issued & CEPH_CAP_AUTH_EXCL) == 0) { | |
ed94f87c | 967 | inode->i_mode = mode; |
ab871b90 EB |
968 | inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(info->uid)); |
969 | inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(info->gid)); | |
355da1eb | 970 | dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode, |
bd2bae6a EB |
971 | from_kuid(&init_user_ns, inode->i_uid), |
972 | from_kgid(&init_user_ns, inode->i_gid)); | |
245ce991 | 973 | ceph_decode_timespec64(&ci->i_btime, &iinfo->btime); |
193e7b37 | 974 | ceph_decode_timespec64(&ci->i_snap_btime, &iinfo->snap_btime); |
355da1eb SW |
975 | } |
976 | ||
f98a128a YZ |
977 | if ((new_version || (new_issued & CEPH_CAP_LINK_SHARED)) && |
978 | (issued & CEPH_CAP_LINK_EXCL) == 0) | |
bfe86848 | 979 | set_nlink(inode, le32_to_cpu(info->nlink)); |
355da1eb | 980 | |
f98a128a YZ |
981 | if (new_version || (new_issued & CEPH_CAP_ANY_RD)) { |
982 | /* be careful with mtime, atime, size */ | |
9bbeab41 AB |
983 | ceph_decode_timespec64(&atime, &info->atime); |
984 | ceph_decode_timespec64(&mtime, &info->mtime); | |
985 | ceph_decode_timespec64(&ctime, &info->ctime); | |
f98a128a YZ |
986 | ceph_fill_file_time(inode, issued, |
987 | le32_to_cpu(info->time_warp_seq), | |
988 | &ctime, &mtime, &atime); | |
989 | } | |
990 | ||
2af54a72 YZ |
991 | if (new_version || (info_caps & CEPH_CAP_FILE_SHARED)) { |
992 | ci->i_files = le64_to_cpu(info->files); | |
993 | ci->i_subdirs = le64_to_cpu(info->subdirs); | |
994 | } | |
995 | ||
f98a128a YZ |
996 | if (new_version || |
997 | (new_issued & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR))) { | |
7627151e | 998 | s64 old_pool = ci->i_layout.pool_id; |
779fe0fb YZ |
999 | struct ceph_string *old_ns; |
1000 | ||
7627151e | 1001 | ceph_file_layout_from_legacy(&ci->i_layout, &info->layout); |
779fe0fb YZ |
1002 | old_ns = rcu_dereference_protected(ci->i_layout.pool_ns, |
1003 | lockdep_is_held(&ci->i_ceph_lock)); | |
1004 | rcu_assign_pointer(ci->i_layout.pool_ns, pool_ns); | |
1005 | ||
1006 | if (ci->i_layout.pool_id != old_pool || pool_ns != old_ns) | |
10183a69 | 1007 | ci->i_ceph_flags &= ~CEPH_I_POOL_PERM; |
10183a69 | 1008 | |
779fe0fb | 1009 | pool_ns = old_ns; |
10183a69 | 1010 | |
f98a128a YZ |
1011 | queue_trunc = ceph_fill_file_size(inode, issued, |
1012 | le32_to_cpu(info->truncate_seq), | |
1013 | le64_to_cpu(info->truncate_size), | |
1014 | le64_to_cpu(info->size)); | |
1015 | /* only update max_size on auth cap */ | |
1016 | if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) && | |
1017 | ci->i_max_size != le64_to_cpu(info->max_size)) { | |
1018 | dout("max_size %lld -> %llu\n", ci->i_max_size, | |
1019 | le64_to_cpu(info->max_size)); | |
1020 | ci->i_max_size = le64_to_cpu(info->max_size); | |
1021 | } | |
1022 | } | |
355da1eb | 1023 | |
49a9f4f6 YZ |
1024 | /* layout and rstat are not tracked by capability, update them if |
1025 | * the inode info is from auth mds */ | |
1026 | if (new_version || (info->cap.flags & CEPH_CAP_FLAG_AUTH)) { | |
1027 | if (S_ISDIR(inode->i_mode)) { | |
1028 | ci->i_dir_layout = iinfo->dir_layout; | |
1029 | ci->i_rbytes = le64_to_cpu(info->rbytes); | |
1030 | ci->i_rfiles = le64_to_cpu(info->rfiles); | |
1031 | ci->i_rsubdirs = le64_to_cpu(info->rsubdirs); | |
08796873 | 1032 | ci->i_dir_pin = iinfo->dir_pin; |
e7f72952 | 1033 | ci->i_rsnaps = iinfo->rsnaps; |
9bbeab41 | 1034 | ceph_decode_timespec64(&ci->i_rctime, &info->rctime); |
49a9f4f6 YZ |
1035 | } |
1036 | } | |
1037 | ||
355da1eb SW |
1038 | /* xattrs */ |
1039 | /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */ | |
508b32d8 | 1040 | if ((ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL)) && |
355da1eb SW |
1041 | le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) { |
1042 | if (ci->i_xattrs.blob) | |
af8a85a4 | 1043 | old_blob = ci->i_xattrs.blob; |
355da1eb SW |
1044 | ci->i_xattrs.blob = xattr_blob; |
1045 | if (xattr_blob) | |
1046 | memcpy(ci->i_xattrs.blob->vec.iov_base, | |
1047 | iinfo->xattr_data, iinfo->xattr_len); | |
1048 | ci->i_xattrs.version = le64_to_cpu(info->xattr_version); | |
7221fe4c | 1049 | ceph_forget_all_cached_acls(inode); |
ac6713cc | 1050 | ceph_security_invalidate_secctx(inode); |
a6424e48 | 1051 | xattr_blob = NULL; |
355da1eb SW |
1052 | } |
1053 | ||
ffdeec7a | 1054 | /* finally update i_version */ |
aae1a442 YZ |
1055 | if (le64_to_cpu(info->version) > ci->i_version) |
1056 | ci->i_version = le64_to_cpu(info->version); | |
ffdeec7a | 1057 | |
355da1eb | 1058 | inode->i_mapping->a_ops = &ceph_aops; |
355da1eb SW |
1059 | |
1060 | switch (inode->i_mode & S_IFMT) { | |
1061 | case S_IFIFO: | |
1062 | case S_IFBLK: | |
1063 | case S_IFCHR: | |
1064 | case S_IFSOCK: | |
5ba72e60 | 1065 | inode->i_blkbits = PAGE_SHIFT; |
ed94f87c | 1066 | init_special_inode(inode, inode->i_mode, rdev); |
355da1eb SW |
1067 | inode->i_op = &ceph_file_iops; |
1068 | break; | |
1069 | case S_IFREG: | |
1070 | inode->i_op = &ceph_file_iops; | |
1071 | inode->i_fop = &ceph_file_fops; | |
1072 | break; | |
1073 | case S_IFLNK: | |
1074 | inode->i_op = &ceph_symlink_iops; | |
1075 | if (!ci->i_symlink) { | |
810339ec | 1076 | u32 symlen = iinfo->symlink_len; |
355da1eb SW |
1077 | char *sym; |
1078 | ||
be655596 | 1079 | spin_unlock(&ci->i_ceph_lock); |
355da1eb | 1080 | |
224a7542 | 1081 | if (symlen != i_size_read(inode)) { |
966c7160 JL |
1082 | pr_err("%s %llx.%llx BAD symlink " |
1083 | "size %lld\n", __func__, | |
1084 | ceph_vinop(inode), | |
224a7542 YZ |
1085 | i_size_read(inode)); |
1086 | i_size_write(inode, symlen); | |
1087 | inode->i_blocks = calc_inode_blocks(symlen); | |
1088 | } | |
810339ec | 1089 | |
355da1eb | 1090 | err = -ENOMEM; |
810339ec | 1091 | sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS); |
355da1eb SW |
1092 | if (!sym) |
1093 | goto out; | |
355da1eb | 1094 | |
be655596 | 1095 | spin_lock(&ci->i_ceph_lock); |
355da1eb SW |
1096 | if (!ci->i_symlink) |
1097 | ci->i_symlink = sym; | |
1098 | else | |
1099 | kfree(sym); /* lost a race */ | |
1100 | } | |
ac194dcc | 1101 | inode->i_link = ci->i_symlink; |
355da1eb SW |
1102 | break; |
1103 | case S_IFDIR: | |
1104 | inode->i_op = &ceph_dir_iops; | |
1105 | inode->i_fop = &ceph_dir_fops; | |
355da1eb SW |
1106 | break; |
1107 | default: | |
966c7160 | 1108 | pr_err("%s %llx.%llx BAD mode 0%o\n", __func__, |
355da1eb SW |
1109 | ceph_vinop(inode), inode->i_mode); |
1110 | } | |
1111 | ||
355da1eb | 1112 | /* were we issued a capability? */ |
2af54a72 | 1113 | if (info_caps) { |
355da1eb SW |
1114 | if (ceph_snap(inode) == CEPH_NOSNAP) { |
1115 | ceph_add_cap(inode, session, | |
1116 | le64_to_cpu(info->cap.cap_id), | |
135e671e | 1117 | info_caps, |
355da1eb SW |
1118 | le32_to_cpu(info->cap.wanted), |
1119 | le32_to_cpu(info->cap.seq), | |
1120 | le32_to_cpu(info->cap.mseq), | |
1121 | le64_to_cpu(info->cap.realm), | |
d9df2783 | 1122 | info->cap.flags, &new_cap); |
2f92b3d0 YZ |
1123 | |
1124 | /* set dir completion flag? */ | |
1125 | if (S_ISDIR(inode->i_mode) && | |
1126 | ci->i_files == 0 && ci->i_subdirs == 0 && | |
2af54a72 | 1127 | (info_caps & CEPH_CAP_FILE_SHARED) && |
2f92b3d0 YZ |
1128 | (issued & CEPH_CAP_FILE_EXCL) == 0 && |
1129 | !__ceph_dir_is_complete(ci)) { | |
1130 | dout(" marking %p complete (empty)\n", inode); | |
fdd4e158 | 1131 | i_size_write(inode, 0); |
2f92b3d0 | 1132 | __ceph_dir_set_complete(ci, |
fdd4e158 YZ |
1133 | atomic64_read(&ci->i_release_count), |
1134 | atomic64_read(&ci->i_ordered_count)); | |
2f92b3d0 YZ |
1135 | } |
1136 | ||
d9df2783 | 1137 | wake = true; |
355da1eb | 1138 | } else { |
355da1eb | 1139 | dout(" %p got snap_caps %s\n", inode, |
2af54a72 YZ |
1140 | ceph_cap_string(info_caps)); |
1141 | ci->i_snap_caps |= info_caps; | |
355da1eb SW |
1142 | } |
1143 | } | |
31c542a1 YZ |
1144 | |
1145 | if (iinfo->inline_version > 0 && | |
1146 | iinfo->inline_version >= ci->i_inline_version) { | |
1147 | int cache_caps = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO; | |
1148 | ci->i_inline_version = iinfo->inline_version; | |
48490776 | 1149 | if (ceph_has_inline_data(ci) && |
2af54a72 | 1150 | (locked_page || (info_caps & cache_caps))) |
31c542a1 YZ |
1151 | fill_inline = true; |
1152 | } | |
1153 | ||
719a2514 YZ |
1154 | if (cap_fmode >= 0) { |
1155 | if (!info_caps) | |
1156 | pr_warn("mds issued no caps on %llx.%llx\n", | |
1157 | ceph_vinop(inode)); | |
1158 | __ceph_touch_fmode(ci, mdsc, cap_fmode); | |
1159 | } | |
1160 | ||
be655596 | 1161 | spin_unlock(&ci->i_ceph_lock); |
355da1eb | 1162 | |
400e1286 JL |
1163 | ceph_fscache_register_inode_cookie(inode); |
1164 | ||
31c542a1 | 1165 | if (fill_inline) |
01deead0 | 1166 | ceph_fill_inline_data(inode, locked_page, |
31c542a1 YZ |
1167 | iinfo->inline_data, iinfo->inline_len); |
1168 | ||
d9df2783 YZ |
1169 | if (wake) |
1170 | wake_up_all(&ci->i_cap_wq); | |
1171 | ||
355da1eb SW |
1172 | /* queue truncate if we saw i_size decrease */ |
1173 | if (queue_trunc) | |
3c6f6b79 | 1174 | ceph_queue_vmtruncate(inode); |
355da1eb SW |
1175 | |
1176 | /* populate frag tree */ | |
3e7fbe9c YZ |
1177 | if (S_ISDIR(inode->i_mode)) |
1178 | ceph_fill_fragtree(inode, &info->fragtree, dirinfo); | |
355da1eb SW |
1179 | |
1180 | /* update delegation info? */ | |
1181 | if (dirinfo) | |
1182 | ceph_fill_dirfrag(inode, dirinfo); | |
1183 | ||
1184 | err = 0; | |
355da1eb | 1185 | out: |
d9df2783 YZ |
1186 | if (new_cap) |
1187 | ceph_put_cap(mdsc, new_cap); | |
af8a85a4 LH |
1188 | ceph_buffer_put(old_blob); |
1189 | ceph_buffer_put(xattr_blob); | |
779fe0fb | 1190 | ceph_put_string(pool_ns); |
355da1eb SW |
1191 | return err; |
1192 | } | |
1193 | ||
1194 | /* | |
543212b3 | 1195 | * caller should hold session s_mutex and dentry->d_lock. |
355da1eb | 1196 | */ |
543212b3 YZ |
1197 | static void __update_dentry_lease(struct inode *dir, struct dentry *dentry, |
1198 | struct ceph_mds_reply_lease *lease, | |
1199 | struct ceph_mds_session *session, | |
1200 | unsigned long from_time, | |
1201 | struct ceph_mds_session **old_lease_session) | |
355da1eb SW |
1202 | { |
1203 | struct ceph_dentry_info *di = ceph_dentry(dentry); | |
f5e17aed | 1204 | unsigned mask = le16_to_cpu(lease->mask); |
355da1eb SW |
1205 | long unsigned duration = le32_to_cpu(lease->duration_ms); |
1206 | long unsigned ttl = from_time + (duration * HZ) / 1000; | |
1207 | long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000; | |
f5d55f03 | 1208 | |
2f90b852 SW |
1209 | dout("update_dentry_lease %p duration %lu ms ttl %lu\n", |
1210 | dentry, duration, ttl); | |
355da1eb | 1211 | |
18fc8abd AV |
1212 | /* only track leases on regular dentries */ |
1213 | if (ceph_snap(dir) != CEPH_NOSNAP) | |
543212b3 | 1214 | return; |
18fc8abd | 1215 | |
f5e17aed JL |
1216 | if (mask & CEPH_LEASE_PRIMARY_LINK) |
1217 | di->flags |= CEPH_DENTRY_PRIMARY_LINK; | |
1218 | else | |
1219 | di->flags &= ~CEPH_DENTRY_PRIMARY_LINK; | |
1220 | ||
97aeb6bf | 1221 | di->lease_shared_gen = atomic_read(&ceph_inode(dir)->i_shared_gen); |
f5e17aed | 1222 | if (!(mask & CEPH_LEASE_VALID)) { |
37c4efc1 | 1223 | __ceph_dentry_dir_lease_touch(di); |
543212b3 | 1224 | return; |
37c4efc1 | 1225 | } |
355da1eb | 1226 | |
52d60f8e | 1227 | if (di->lease_gen == atomic_read(&session->s_cap_gen) && |
9b16f03c | 1228 | time_before(ttl, di->time)) |
543212b3 | 1229 | return; /* we already have a newer lease. */ |
355da1eb | 1230 | |
481f001f | 1231 | if (di->lease_session && di->lease_session != session) { |
543212b3 | 1232 | *old_lease_session = di->lease_session; |
481f001f YZ |
1233 | di->lease_session = NULL; |
1234 | } | |
355da1eb | 1235 | |
355da1eb SW |
1236 | if (!di->lease_session) |
1237 | di->lease_session = ceph_get_mds_session(session); | |
52d60f8e | 1238 | di->lease_gen = atomic_read(&session->s_cap_gen); |
355da1eb SW |
1239 | di->lease_seq = le32_to_cpu(lease->seq); |
1240 | di->lease_renew_after = half_ttl; | |
1241 | di->lease_renew_from = 0; | |
9b16f03c | 1242 | di->time = ttl; |
37c4efc1 YZ |
1243 | |
1244 | __ceph_dentry_lease_touch(di); | |
543212b3 YZ |
1245 | } |
1246 | ||
1247 | static inline void update_dentry_lease(struct inode *dir, struct dentry *dentry, | |
1248 | struct ceph_mds_reply_lease *lease, | |
1249 | struct ceph_mds_session *session, | |
1250 | unsigned long from_time) | |
1251 | { | |
1252 | struct ceph_mds_session *old_lease_session = NULL; | |
1253 | spin_lock(&dentry->d_lock); | |
1254 | __update_dentry_lease(dir, dentry, lease, session, from_time, | |
1255 | &old_lease_session); | |
1256 | spin_unlock(&dentry->d_lock); | |
7e65624d | 1257 | ceph_put_mds_session(old_lease_session); |
543212b3 YZ |
1258 | } |
1259 | ||
1260 | /* | |
1261 | * update dentry lease without having parent inode locked | |
1262 | */ | |
1263 | static void update_dentry_lease_careful(struct dentry *dentry, | |
1264 | struct ceph_mds_reply_lease *lease, | |
1265 | struct ceph_mds_session *session, | |
1266 | unsigned long from_time, | |
1267 | char *dname, u32 dname_len, | |
1268 | struct ceph_vino *pdvino, | |
1269 | struct ceph_vino *ptvino) | |
1270 | ||
1271 | { | |
1272 | struct inode *dir; | |
1273 | struct ceph_mds_session *old_lease_session = NULL; | |
1274 | ||
1275 | spin_lock(&dentry->d_lock); | |
1276 | /* make sure dentry's name matches target */ | |
1277 | if (dentry->d_name.len != dname_len || | |
1278 | memcmp(dentry->d_name.name, dname, dname_len)) | |
1279 | goto out_unlock; | |
1280 | ||
1281 | dir = d_inode(dentry->d_parent); | |
1282 | /* make sure parent matches dvino */ | |
1283 | if (!ceph_ino_compare(dir, pdvino)) | |
1284 | goto out_unlock; | |
1285 | ||
1286 | /* make sure dentry's inode matches target. NULL ptvino means that | |
1287 | * we expect a negative dentry */ | |
1288 | if (ptvino) { | |
1289 | if (d_really_is_negative(dentry)) | |
1290 | goto out_unlock; | |
1291 | if (!ceph_ino_compare(d_inode(dentry), ptvino)) | |
1292 | goto out_unlock; | |
1293 | } else { | |
1294 | if (d_really_is_positive(dentry)) | |
1295 | goto out_unlock; | |
1296 | } | |
1297 | ||
1298 | __update_dentry_lease(dir, dentry, lease, session, | |
1299 | from_time, &old_lease_session); | |
355da1eb SW |
1300 | out_unlock: |
1301 | spin_unlock(&dentry->d_lock); | |
7e65624d | 1302 | ceph_put_mds_session(old_lease_session); |
355da1eb SW |
1303 | } |
1304 | ||
1305 | /* | |
1306 | * splice a dentry to an inode. | |
810313c5 | 1307 | * caller must hold directory i_rwsem for this to be safe. |
355da1eb | 1308 | */ |
2bf996ac | 1309 | static int splice_dentry(struct dentry **pdn, struct inode *in) |
355da1eb | 1310 | { |
2bf996ac | 1311 | struct dentry *dn = *pdn; |
355da1eb SW |
1312 | struct dentry *realdn; |
1313 | ||
2b0143b5 | 1314 | BUG_ON(d_inode(dn)); |
1cd3935b | 1315 | |
5495c2d0 YZ |
1316 | if (S_ISDIR(in->i_mode)) { |
1317 | /* If inode is directory, d_splice_alias() below will remove | |
1318 | * 'realdn' from its origin parent. We need to ensure that | |
1319 | * origin parent's readdir cache will not reference 'realdn' | |
1320 | */ | |
1321 | realdn = d_find_any_alias(in); | |
1322 | if (realdn) { | |
1323 | struct ceph_dentry_info *di = ceph_dentry(realdn); | |
1324 | spin_lock(&realdn->d_lock); | |
1325 | ||
1326 | realdn->d_op->d_prune(realdn); | |
1327 | ||
1328 | di->time = jiffies; | |
1329 | di->lease_shared_gen = 0; | |
1330 | di->offset = 0; | |
1331 | ||
1332 | spin_unlock(&realdn->d_lock); | |
1333 | dput(realdn); | |
1334 | } | |
1335 | } | |
1336 | ||
355da1eb SW |
1337 | /* dn must be unhashed */ |
1338 | if (!d_unhashed(dn)) | |
1339 | d_drop(dn); | |
41d28bca | 1340 | realdn = d_splice_alias(in, dn); |
355da1eb | 1341 | if (IS_ERR(realdn)) { |
d69ed05a SW |
1342 | pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n", |
1343 | PTR_ERR(realdn), dn, in, ceph_vinop(in)); | |
2bf996ac YZ |
1344 | return PTR_ERR(realdn); |
1345 | } | |
1346 | ||
1347 | if (realdn) { | |
355da1eb SW |
1348 | dout("dn %p (%d) spliced with %p (%d) " |
1349 | "inode %p ino %llx.%llx\n", | |
84d08fa8 AV |
1350 | dn, d_count(dn), |
1351 | realdn, d_count(realdn), | |
2b0143b5 | 1352 | d_inode(realdn), ceph_vinop(d_inode(realdn))); |
355da1eb | 1353 | dput(dn); |
2bf996ac | 1354 | *pdn = realdn; |
355da1eb SW |
1355 | } else { |
1356 | BUG_ON(!ceph_dentry(dn)); | |
355da1eb | 1357 | dout("dn %p attached to %p ino %llx.%llx\n", |
2b0143b5 | 1358 | dn, d_inode(dn), ceph_vinop(d_inode(dn))); |
355da1eb | 1359 | } |
2bf996ac | 1360 | return 0; |
355da1eb SW |
1361 | } |
1362 | ||
1363 | /* | |
1364 | * Incorporate results into the local cache. This is either just | |
1365 | * one inode, or a directory, dentry, and possibly linked-to inode (e.g., | |
1366 | * after a lookup). | |
1367 | * | |
1368 | * A reply may contain | |
1369 | * a directory inode along with a dentry. | |
1370 | * and/or a target inode | |
1371 | * | |
1372 | * Called with snap_rwsem (read). | |
1373 | */ | |
f5a03b08 | 1374 | int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req) |
355da1eb | 1375 | { |
f5a03b08 | 1376 | struct ceph_mds_session *session = req->r_session; |
355da1eb SW |
1377 | struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info; |
1378 | struct inode *in = NULL; | |
f5d55f03 | 1379 | struct ceph_vino tvino, dvino; |
3d14c5d2 | 1380 | struct ceph_fs_client *fsc = ceph_sb_to_client(sb); |
355da1eb SW |
1381 | int err = 0; |
1382 | ||
1383 | dout("fill_trace %p is_dentry %d is_target %d\n", req, | |
1384 | rinfo->head->is_dentry, rinfo->head->is_target); | |
1385 | ||
355da1eb SW |
1386 | if (!rinfo->head->is_target && !rinfo->head->is_dentry) { |
1387 | dout("fill_trace reply is empty!\n"); | |
3dd69aab | 1388 | if (rinfo->head->result == 0 && req->r_parent) |
167c9e35 | 1389 | ceph_invalidate_dir_request(req); |
355da1eb SW |
1390 | return 0; |
1391 | } | |
1392 | ||
1393 | if (rinfo->head->is_dentry) { | |
3dd69aab | 1394 | struct inode *dir = req->r_parent; |
5b1daecd | 1395 | |
6c5e50fa | 1396 | if (dir) { |
966c7160 JL |
1397 | err = ceph_fill_inode(dir, NULL, &rinfo->diri, |
1398 | rinfo->dirfrag, session, -1, | |
1399 | &req->r_caps_reservation); | |
6c5e50fa | 1400 | if (err < 0) |
19913b4e | 1401 | goto done; |
6c5e50fa SW |
1402 | } else { |
1403 | WARN_ON_ONCE(1); | |
1404 | } | |
19913b4e | 1405 | |
74c9e6bf YZ |
1406 | if (dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME && |
1407 | test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) && | |
1408 | !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) { | |
19913b4e YZ |
1409 | struct qstr dname; |
1410 | struct dentry *dn, *parent; | |
1411 | ||
1412 | BUG_ON(!rinfo->head->is_target); | |
1413 | BUG_ON(req->r_dentry); | |
1414 | ||
1415 | parent = d_find_any_alias(dir); | |
1416 | BUG_ON(!parent); | |
1417 | ||
1418 | dname.name = rinfo->dname; | |
1419 | dname.len = rinfo->dname_len; | |
8387ff25 | 1420 | dname.hash = full_name_hash(parent, dname.name, dname.len); |
f5d55f03 JL |
1421 | tvino.ino = le64_to_cpu(rinfo->targeti.in->ino); |
1422 | tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid); | |
19913b4e YZ |
1423 | retry_lookup: |
1424 | dn = d_lookup(parent, &dname); | |
1425 | dout("d_lookup on parent=%p name=%.*s got %p\n", | |
1426 | parent, dname.len, dname.name, dn); | |
1427 | ||
1428 | if (!dn) { | |
1429 | dn = d_alloc(parent, &dname); | |
1430 | dout("d_alloc %p '%.*s' = %p\n", parent, | |
1431 | dname.len, dname.name, dn); | |
d37b1d99 | 1432 | if (!dn) { |
19913b4e YZ |
1433 | dput(parent); |
1434 | err = -ENOMEM; | |
1435 | goto done; | |
1436 | } | |
ad5cb123 | 1437 | err = 0; |
2b0143b5 | 1438 | } else if (d_really_is_positive(dn) && |
f5d55f03 JL |
1439 | (ceph_ino(d_inode(dn)) != tvino.ino || |
1440 | ceph_snap(d_inode(dn)) != tvino.snap)) { | |
19913b4e | 1441 | dout(" dn %p points to wrong inode %p\n", |
2b0143b5 | 1442 | dn, d_inode(dn)); |
933ad2c9 | 1443 | ceph_dir_clear_ordered(dir); |
19913b4e YZ |
1444 | d_delete(dn); |
1445 | dput(dn); | |
1446 | goto retry_lookup; | |
1447 | } | |
1448 | ||
1449 | req->r_dentry = dn; | |
1450 | dput(parent); | |
1451 | } | |
5b1daecd SW |
1452 | } |
1453 | ||
86b58d13 | 1454 | if (rinfo->head->is_target) { |
bca9fc14 JL |
1455 | /* Should be filled in by handle_reply */ |
1456 | BUG_ON(!req->r_target_inode); | |
86b58d13 | 1457 | |
bca9fc14 | 1458 | in = req->r_target_inode; |
966c7160 JL |
1459 | err = ceph_fill_inode(in, req->r_locked_page, &rinfo->targeti, |
1460 | NULL, session, | |
bc2de10d | 1461 | (!test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) && |
3bb48b41 | 1462 | !test_bit(CEPH_MDS_R_ASYNC, &req->r_req_flags) && |
57c21994 | 1463 | rinfo->head->result == 0) ? req->r_fmode : -1, |
86b58d13 YZ |
1464 | &req->r_caps_reservation); |
1465 | if (err < 0) { | |
966c7160 | 1466 | pr_err("ceph_fill_inode badness %p %llx.%llx\n", |
86b58d13 | 1467 | in, ceph_vinop(in)); |
bca9fc14 | 1468 | req->r_target_inode = NULL; |
893e456b JL |
1469 | if (in->i_state & I_NEW) |
1470 | discard_new_inode(in); | |
68cbb805 JL |
1471 | else |
1472 | iput(in); | |
86b58d13 YZ |
1473 | goto done; |
1474 | } | |
893e456b JL |
1475 | if (in->i_state & I_NEW) |
1476 | unlock_new_inode(in); | |
86b58d13 YZ |
1477 | } |
1478 | ||
9358c6d4 SW |
1479 | /* |
1480 | * ignore null lease/binding on snapdir ENOENT, or else we | |
1481 | * will have trouble splicing in the virtual snapdir later | |
1482 | */ | |
3dd69aab JL |
1483 | if (rinfo->head->is_dentry && |
1484 | !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) && | |
1485 | test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) && | |
9358c6d4 | 1486 | (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name, |
3d14c5d2 | 1487 | fsc->mount_options->snapdir_name, |
9358c6d4 | 1488 | req->r_dentry->d_name.len))) { |
355da1eb SW |
1489 | /* |
1490 | * lookup link rename : null -> possibly existing inode | |
1491 | * mknod symlink mkdir : null -> new inode | |
1492 | * unlink : linked -> null | |
1493 | */ | |
3dd69aab | 1494 | struct inode *dir = req->r_parent; |
355da1eb SW |
1495 | struct dentry *dn = req->r_dentry; |
1496 | bool have_dir_cap, have_lease; | |
1497 | ||
1498 | BUG_ON(!dn); | |
1499 | BUG_ON(!dir); | |
2b0143b5 | 1500 | BUG_ON(d_inode(dn->d_parent) != dir); |
f5d55f03 JL |
1501 | |
1502 | dvino.ino = le64_to_cpu(rinfo->diri.in->ino); | |
1503 | dvino.snap = le64_to_cpu(rinfo->diri.in->snapid); | |
1504 | ||
1505 | BUG_ON(ceph_ino(dir) != dvino.ino); | |
1506 | BUG_ON(ceph_snap(dir) != dvino.snap); | |
355da1eb | 1507 | |
355da1eb SW |
1508 | /* do we have a lease on the whole dir? */ |
1509 | have_dir_cap = | |
1510 | (le32_to_cpu(rinfo->diri.in->cap.caps) & | |
1511 | CEPH_CAP_FILE_SHARED); | |
1512 | ||
1513 | /* do we have a dn lease? */ | |
1514 | have_lease = have_dir_cap || | |
2f90b852 | 1515 | le32_to_cpu(rinfo->dlease->duration_ms); |
355da1eb SW |
1516 | if (!have_lease) |
1517 | dout("fill_trace no dentry lease or dir cap\n"); | |
1518 | ||
1519 | /* rename? */ | |
1520 | if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) { | |
0a8a70f9 YZ |
1521 | struct inode *olddir = req->r_old_dentry_dir; |
1522 | BUG_ON(!olddir); | |
1523 | ||
a455589f | 1524 | dout(" src %p '%pd' dst %p '%pd'\n", |
355da1eb | 1525 | req->r_old_dentry, |
a455589f AV |
1526 | req->r_old_dentry, |
1527 | dn, dn); | |
355da1eb SW |
1528 | dout("fill_trace doing d_move %p -> %p\n", |
1529 | req->r_old_dentry, dn); | |
c10f5e12 | 1530 | |
fdd4e158 YZ |
1531 | /* d_move screws up sibling dentries' offsets */ |
1532 | ceph_dir_clear_ordered(dir); | |
1533 | ceph_dir_clear_ordered(olddir); | |
1534 | ||
355da1eb | 1535 | d_move(req->r_old_dentry, dn); |
a455589f AV |
1536 | dout(" src %p '%pd' dst %p '%pd'\n", |
1537 | req->r_old_dentry, | |
355da1eb | 1538 | req->r_old_dentry, |
a455589f | 1539 | dn, dn); |
81a6cf2d | 1540 | |
c4a29f26 SW |
1541 | /* ensure target dentry is invalidated, despite |
1542 | rehashing bug in vfs_rename_dir */ | |
81a6cf2d SW |
1543 | ceph_invalidate_dentry_lease(dn); |
1544 | ||
99ccbd22 | 1545 | dout("dn %p gets new offset %lld\n", req->r_old_dentry, |
1cd3935b | 1546 | ceph_dentry(req->r_old_dentry)->offset); |
81a6cf2d | 1547 | |
2bf996ac YZ |
1548 | /* swap r_dentry and r_old_dentry in case that |
1549 | * splice_dentry() gets called later. This is safe | |
1550 | * because no other place will use them */ | |
1551 | req->r_dentry = req->r_old_dentry; | |
1552 | req->r_old_dentry = dn; | |
1553 | dn = req->r_dentry; | |
355da1eb SW |
1554 | } |
1555 | ||
1556 | /* null dentry? */ | |
1557 | if (!rinfo->head->is_target) { | |
1558 | dout("fill_trace null dentry\n"); | |
2b0143b5 | 1559 | if (d_really_is_positive(dn)) { |
355da1eb | 1560 | dout("d_delete %p\n", dn); |
5495c2d0 | 1561 | ceph_dir_clear_ordered(dir); |
355da1eb | 1562 | d_delete(dn); |
80d025ff JL |
1563 | } else if (have_lease) { |
1564 | if (d_unhashed(dn)) | |
f8b31710 | 1565 | d_add(dn, NULL); |
7ffe4fce XL |
1566 | } |
1567 | ||
1568 | if (!d_unhashed(dn) && have_lease) | |
543212b3 YZ |
1569 | update_dentry_lease(dir, dn, |
1570 | rinfo->dlease, session, | |
1571 | req->r_request_started); | |
355da1eb SW |
1572 | goto done; |
1573 | } | |
1574 | ||
1575 | /* attach proper inode */ | |
2b0143b5 | 1576 | if (d_really_is_negative(dn)) { |
70db4f36 | 1577 | ceph_dir_clear_ordered(dir); |
86b58d13 | 1578 | ihold(in); |
2bf996ac YZ |
1579 | err = splice_dentry(&req->r_dentry, in); |
1580 | if (err < 0) | |
355da1eb | 1581 | goto done; |
2bf996ac | 1582 | dn = req->r_dentry; /* may have spliced */ |
2b0143b5 | 1583 | } else if (d_really_is_positive(dn) && d_inode(dn) != in) { |
355da1eb | 1584 | dout(" %p links to %p %llx.%llx, not %llx.%llx\n", |
2b0143b5 | 1585 | dn, d_inode(dn), ceph_vinop(d_inode(dn)), |
86b58d13 | 1586 | ceph_vinop(in)); |
200fd27c | 1587 | d_invalidate(dn); |
355da1eb | 1588 | have_lease = false; |
355da1eb SW |
1589 | } |
1590 | ||
f5d55f03 | 1591 | if (have_lease) { |
543212b3 YZ |
1592 | update_dentry_lease(dir, dn, |
1593 | rinfo->dlease, session, | |
1594 | req->r_request_started); | |
f5d55f03 | 1595 | } |
355da1eb | 1596 | dout(" final dn %p\n", dn); |
bc2de10d JL |
1597 | } else if ((req->r_op == CEPH_MDS_OP_LOOKUPSNAP || |
1598 | req->r_op == CEPH_MDS_OP_MKSNAP) && | |
1f08529c | 1599 | test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) && |
bc2de10d | 1600 | !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) { |
3dd69aab | 1601 | struct inode *dir = req->r_parent; |
355da1eb SW |
1602 | |
1603 | /* fill out a snapdir LOOKUPSNAP dentry */ | |
0a8a70f9 YZ |
1604 | BUG_ON(!dir); |
1605 | BUG_ON(ceph_snap(dir) != CEPH_SNAPDIR); | |
2bf996ac YZ |
1606 | BUG_ON(!req->r_dentry); |
1607 | dout(" linking snapped dir %p to dn %p\n", in, req->r_dentry); | |
70db4f36 | 1608 | ceph_dir_clear_ordered(dir); |
86b58d13 | 1609 | ihold(in); |
2bf996ac YZ |
1610 | err = splice_dentry(&req->r_dentry, in); |
1611 | if (err < 0) | |
355da1eb | 1612 | goto done; |
543212b3 YZ |
1613 | } else if (rinfo->head->is_dentry && req->r_dentry) { |
1614 | /* parent inode is not locked, be carefull */ | |
cdde7c43 | 1615 | struct ceph_vino *ptvino = NULL; |
543212b3 YZ |
1616 | dvino.ino = le64_to_cpu(rinfo->diri.in->ino); |
1617 | dvino.snap = le64_to_cpu(rinfo->diri.in->snapid); | |
1618 | if (rinfo->head->is_target) { | |
1619 | tvino.ino = le64_to_cpu(rinfo->targeti.in->ino); | |
1620 | tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid); | |
1621 | ptvino = &tvino; | |
cdde7c43 | 1622 | } |
543212b3 YZ |
1623 | update_dentry_lease_careful(req->r_dentry, rinfo->dlease, |
1624 | session, req->r_request_started, | |
1625 | rinfo->dname, rinfo->dname_len, | |
1626 | &dvino, ptvino); | |
355da1eb | 1627 | } |
355da1eb SW |
1628 | done: |
1629 | dout("fill_trace done err=%d\n", err); | |
1630 | return err; | |
1631 | } | |
1632 | ||
1633 | /* | |
1634 | * Prepopulate our cache with readdir results, leases, etc. | |
1635 | */ | |
79f9f99a SW |
1636 | static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req, |
1637 | struct ceph_mds_session *session) | |
1638 | { | |
1639 | struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info; | |
1640 | int i, err = 0; | |
1641 | ||
1642 | for (i = 0; i < rinfo->dir_nr; i++) { | |
2a5beea3 | 1643 | struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i; |
79f9f99a SW |
1644 | struct ceph_vino vino; |
1645 | struct inode *in; | |
1646 | int rc; | |
1647 | ||
2a5beea3 YZ |
1648 | vino.ino = le64_to_cpu(rde->inode.in->ino); |
1649 | vino.snap = le64_to_cpu(rde->inode.in->snapid); | |
79f9f99a | 1650 | |
ec9595c0 | 1651 | in = ceph_get_inode(req->r_dentry->d_sb, vino, NULL); |
79f9f99a SW |
1652 | if (IS_ERR(in)) { |
1653 | err = PTR_ERR(in); | |
1654 | dout("new_inode badness got %d\n", err); | |
1655 | continue; | |
1656 | } | |
966c7160 JL |
1657 | rc = ceph_fill_inode(in, NULL, &rde->inode, NULL, session, |
1658 | -1, &req->r_caps_reservation); | |
79f9f99a | 1659 | if (rc < 0) { |
966c7160 JL |
1660 | pr_err("ceph_fill_inode badness on %p got %d\n", |
1661 | in, rc); | |
79f9f99a | 1662 | err = rc; |
893e456b JL |
1663 | if (in->i_state & I_NEW) { |
1664 | ihold(in); | |
1665 | discard_new_inode(in); | |
1666 | } | |
1667 | } else if (in->i_state & I_NEW) { | |
1668 | unlock_new_inode(in); | |
79f9f99a | 1669 | } |
893e456b | 1670 | |
23c2c76e | 1671 | iput(in); |
79f9f99a SW |
1672 | } |
1673 | ||
1674 | return err; | |
1675 | } | |
1676 | ||
fdd4e158 YZ |
1677 | void ceph_readdir_cache_release(struct ceph_readdir_cache_control *ctl) |
1678 | { | |
1679 | if (ctl->page) { | |
1680 | kunmap(ctl->page); | |
09cbfeaf | 1681 | put_page(ctl->page); |
fdd4e158 YZ |
1682 | ctl->page = NULL; |
1683 | } | |
1684 | } | |
1685 | ||
1686 | static int fill_readdir_cache(struct inode *dir, struct dentry *dn, | |
1687 | struct ceph_readdir_cache_control *ctl, | |
1688 | struct ceph_mds_request *req) | |
1689 | { | |
1690 | struct ceph_inode_info *ci = ceph_inode(dir); | |
09cbfeaf | 1691 | unsigned nsize = PAGE_SIZE / sizeof(struct dentry*); |
fdd4e158 YZ |
1692 | unsigned idx = ctl->index % nsize; |
1693 | pgoff_t pgoff = ctl->index / nsize; | |
1694 | ||
1695 | if (!ctl->page || pgoff != page_index(ctl->page)) { | |
1696 | ceph_readdir_cache_release(ctl); | |
af5e5eb5 YZ |
1697 | if (idx == 0) |
1698 | ctl->page = grab_cache_page(&dir->i_data, pgoff); | |
1699 | else | |
1700 | ctl->page = find_lock_page(&dir->i_data, pgoff); | |
fdd4e158 YZ |
1701 | if (!ctl->page) { |
1702 | ctl->index = -1; | |
af5e5eb5 | 1703 | return idx == 0 ? -ENOMEM : 0; |
fdd4e158 YZ |
1704 | } |
1705 | /* reading/filling the cache are serialized by | |
810313c5 | 1706 | * i_rwsem, no need to use page lock */ |
fdd4e158 YZ |
1707 | unlock_page(ctl->page); |
1708 | ctl->dentries = kmap(ctl->page); | |
af5e5eb5 | 1709 | if (idx == 0) |
09cbfeaf | 1710 | memset(ctl->dentries, 0, PAGE_SIZE); |
fdd4e158 YZ |
1711 | } |
1712 | ||
1713 | if (req->r_dir_release_cnt == atomic64_read(&ci->i_release_count) && | |
1714 | req->r_dir_ordered_cnt == atomic64_read(&ci->i_ordered_count)) { | |
1715 | dout("readdir cache dn %p idx %d\n", dn, ctl->index); | |
1716 | ctl->dentries[idx] = dn; | |
1717 | ctl->index++; | |
1718 | } else { | |
1719 | dout("disable readdir cache\n"); | |
1720 | ctl->index = -1; | |
1721 | } | |
1722 | return 0; | |
1723 | } | |
1724 | ||
355da1eb SW |
1725 | int ceph_readdir_prepopulate(struct ceph_mds_request *req, |
1726 | struct ceph_mds_session *session) | |
1727 | { | |
1728 | struct dentry *parent = req->r_dentry; | |
f3c4ebe6 | 1729 | struct ceph_inode_info *ci = ceph_inode(d_inode(parent)); |
355da1eb SW |
1730 | struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info; |
1731 | struct qstr dname; | |
1732 | struct dentry *dn; | |
1733 | struct inode *in; | |
315f2408 | 1734 | int err = 0, skipped = 0, ret, i; |
0f51a983 | 1735 | u32 frag = le32_to_cpu(req->r_args.readdir.frag); |
f3c4ebe6 YZ |
1736 | u32 last_hash = 0; |
1737 | u32 fpos_offset; | |
fdd4e158 YZ |
1738 | struct ceph_readdir_cache_control cache_ctl = {}; |
1739 | ||
bc2de10d | 1740 | if (test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) |
fdd4e158 | 1741 | return readdir_prepopulate_inodes_only(req, session); |
81c6aea5 | 1742 | |
79162547 YZ |
1743 | if (rinfo->hash_order) { |
1744 | if (req->r_path2) { | |
1745 | last_hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash, | |
1746 | req->r_path2, | |
1747 | strlen(req->r_path2)); | |
1748 | last_hash = ceph_frag_value(last_hash); | |
1749 | } else if (rinfo->offset_hash) { | |
1750 | /* mds understands offset_hash */ | |
1751 | WARN_ON_ONCE(req->r_readdir_offset != 2); | |
0f51a983 | 1752 | last_hash = le32_to_cpu(req->r_args.readdir.offset_hash); |
79162547 | 1753 | } |
f3c4ebe6 YZ |
1754 | } |
1755 | ||
81c6aea5 YZ |
1756 | if (rinfo->dir_dir && |
1757 | le32_to_cpu(rinfo->dir_dir->frag) != frag) { | |
1758 | dout("readdir_prepopulate got new frag %x -> %x\n", | |
1759 | frag, le32_to_cpu(rinfo->dir_dir->frag)); | |
1760 | frag = le32_to_cpu(rinfo->dir_dir->frag); | |
f3c4ebe6 | 1761 | if (!rinfo->hash_order) |
fdd4e158 | 1762 | req->r_readdir_offset = 2; |
81c6aea5 | 1763 | } |
355da1eb SW |
1764 | |
1765 | if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) { | |
355da1eb SW |
1766 | dout("readdir_prepopulate %d items under SNAPDIR dn %p\n", |
1767 | rinfo->dir_nr, parent); | |
1768 | } else { | |
1769 | dout("readdir_prepopulate %d items under dn %p\n", | |
1770 | rinfo->dir_nr, parent); | |
1771 | if (rinfo->dir_dir) | |
2b0143b5 | 1772 | ceph_fill_dirfrag(d_inode(parent), rinfo->dir_dir); |
355da1eb | 1773 | |
8d45b911 YZ |
1774 | if (ceph_frag_is_leftmost(frag) && |
1775 | req->r_readdir_offset == 2 && | |
1776 | !(rinfo->hash_order && last_hash)) { | |
1777 | /* note dir version at start of readdir so we can | |
1778 | * tell if any dentries get dropped */ | |
1779 | req->r_dir_release_cnt = | |
1780 | atomic64_read(&ci->i_release_count); | |
1781 | req->r_dir_ordered_cnt = | |
1782 | atomic64_read(&ci->i_ordered_count); | |
1783 | req->r_readdir_cache_idx = 0; | |
1784 | } | |
fdd4e158 YZ |
1785 | } |
1786 | ||
1787 | cache_ctl.index = req->r_readdir_cache_idx; | |
f3c4ebe6 | 1788 | fpos_offset = req->r_readdir_offset; |
fdd4e158 | 1789 | |
86b58d13 | 1790 | /* FIXME: release caps/leases if error occurs */ |
355da1eb | 1791 | for (i = 0; i < rinfo->dir_nr; i++) { |
2a5beea3 | 1792 | struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i; |
543212b3 | 1793 | struct ceph_vino tvino; |
355da1eb | 1794 | |
2a5beea3 YZ |
1795 | dname.name = rde->name; |
1796 | dname.len = rde->name_len; | |
8387ff25 | 1797 | dname.hash = full_name_hash(parent, dname.name, dname.len); |
355da1eb | 1798 | |
f5d55f03 JL |
1799 | tvino.ino = le64_to_cpu(rde->inode.in->ino); |
1800 | tvino.snap = le64_to_cpu(rde->inode.in->snapid); | |
355da1eb | 1801 | |
f3c4ebe6 YZ |
1802 | if (rinfo->hash_order) { |
1803 | u32 hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash, | |
1804 | rde->name, rde->name_len); | |
1805 | hash = ceph_frag_value(hash); | |
1806 | if (hash != last_hash) | |
1807 | fpos_offset = 2; | |
1808 | last_hash = hash; | |
1809 | rde->offset = ceph_make_fpos(hash, fpos_offset++, true); | |
1810 | } else { | |
1811 | rde->offset = ceph_make_fpos(frag, fpos_offset++, false); | |
1812 | } | |
355da1eb SW |
1813 | |
1814 | retry_lookup: | |
1815 | dn = d_lookup(parent, &dname); | |
1816 | dout("d_lookup on parent=%p name=%.*s got %p\n", | |
1817 | parent, dname.len, dname.name, dn); | |
1818 | ||
1819 | if (!dn) { | |
1820 | dn = d_alloc(parent, &dname); | |
1821 | dout("d_alloc %p '%.*s' = %p\n", parent, | |
1822 | dname.len, dname.name, dn); | |
d37b1d99 | 1823 | if (!dn) { |
355da1eb SW |
1824 | dout("d_alloc badness\n"); |
1825 | err = -ENOMEM; | |
1826 | goto out; | |
1827 | } | |
2b0143b5 | 1828 | } else if (d_really_is_positive(dn) && |
f5d55f03 JL |
1829 | (ceph_ino(d_inode(dn)) != tvino.ino || |
1830 | ceph_snap(d_inode(dn)) != tvino.snap)) { | |
5495c2d0 | 1831 | struct ceph_dentry_info *di = ceph_dentry(dn); |
355da1eb | 1832 | dout(" dn %p points to wrong inode %p\n", |
2b0143b5 | 1833 | dn, d_inode(dn)); |
5495c2d0 YZ |
1834 | |
1835 | spin_lock(&dn->d_lock); | |
1836 | if (di->offset > 0 && | |
1837 | di->lease_shared_gen == | |
1838 | atomic_read(&ci->i_shared_gen)) { | |
1839 | __ceph_dir_clear_ordered(ci); | |
1840 | di->offset = 0; | |
1841 | } | |
1842 | spin_unlock(&dn->d_lock); | |
1843 | ||
355da1eb SW |
1844 | d_delete(dn); |
1845 | dput(dn); | |
1846 | goto retry_lookup; | |
355da1eb SW |
1847 | } |
1848 | ||
355da1eb | 1849 | /* inode */ |
2b0143b5 DH |
1850 | if (d_really_is_positive(dn)) { |
1851 | in = d_inode(dn); | |
355da1eb | 1852 | } else { |
ec9595c0 | 1853 | in = ceph_get_inode(parent->d_sb, tvino, NULL); |
ac1f12ef | 1854 | if (IS_ERR(in)) { |
355da1eb | 1855 | dout("new_inode badness\n"); |
2744c171 | 1856 | d_drop(dn); |
355da1eb | 1857 | dput(dn); |
ac1f12ef | 1858 | err = PTR_ERR(in); |
355da1eb SW |
1859 | goto out; |
1860 | } | |
355da1eb SW |
1861 | } |
1862 | ||
966c7160 JL |
1863 | ret = ceph_fill_inode(in, NULL, &rde->inode, NULL, session, |
1864 | -1, &req->r_caps_reservation); | |
fdd4e158 | 1865 | if (ret < 0) { |
966c7160 | 1866 | pr_err("ceph_fill_inode badness on %p\n", in); |
3e1d0452 | 1867 | if (d_really_is_negative(dn)) { |
893e456b JL |
1868 | if (in->i_state & I_NEW) { |
1869 | ihold(in); | |
1870 | discard_new_inode(in); | |
1871 | } | |
23c2c76e | 1872 | iput(in); |
3e1d0452 | 1873 | } |
86b58d13 | 1874 | d_drop(dn); |
fdd4e158 | 1875 | err = ret; |
d69ed05a | 1876 | goto next_item; |
355da1eb | 1877 | } |
893e456b JL |
1878 | if (in->i_state & I_NEW) |
1879 | unlock_new_inode(in); | |
86b58d13 | 1880 | |
2b0143b5 | 1881 | if (d_really_is_negative(dn)) { |
315f2408 YZ |
1882 | if (ceph_security_xattr_deadlock(in)) { |
1883 | dout(" skip splicing dn %p to inode %p" | |
1884 | " (security xattr deadlock)\n", dn, in); | |
23c2c76e | 1885 | iput(in); |
315f2408 YZ |
1886 | skipped++; |
1887 | goto next_item; | |
1888 | } | |
1889 | ||
2bf996ac YZ |
1890 | err = splice_dentry(&dn, in); |
1891 | if (err < 0) | |
86b58d13 | 1892 | goto next_item; |
86b58d13 YZ |
1893 | } |
1894 | ||
f3c4ebe6 | 1895 | ceph_dentry(dn)->offset = rde->offset; |
86b58d13 | 1896 | |
543212b3 YZ |
1897 | update_dentry_lease(d_inode(parent), dn, |
1898 | rde->lease, req->r_session, | |
1899 | req->r_request_started); | |
fdd4e158 | 1900 | |
315f2408 | 1901 | if (err == 0 && skipped == 0 && cache_ctl.index >= 0) { |
fdd4e158 YZ |
1902 | ret = fill_readdir_cache(d_inode(parent), dn, |
1903 | &cache_ctl, req); | |
1904 | if (ret < 0) | |
1905 | err = ret; | |
1906 | } | |
d69ed05a | 1907 | next_item: |
2bf996ac | 1908 | dput(dn); |
355da1eb | 1909 | } |
355da1eb | 1910 | out: |
315f2408 | 1911 | if (err == 0 && skipped == 0) { |
bc2de10d | 1912 | set_bit(CEPH_MDS_R_DID_PREPOPULATE, &req->r_req_flags); |
fdd4e158 YZ |
1913 | req->r_readdir_cache_idx = cache_ctl.index; |
1914 | } | |
1915 | ceph_readdir_cache_release(&cache_ctl); | |
355da1eb SW |
1916 | dout("readdir_prepopulate done\n"); |
1917 | return err; | |
1918 | } | |
1919 | ||
efb0ca76 | 1920 | bool ceph_inode_set_size(struct inode *inode, loff_t size) |
355da1eb SW |
1921 | { |
1922 | struct ceph_inode_info *ci = ceph_inode(inode); | |
efb0ca76 | 1923 | bool ret; |
355da1eb | 1924 | |
be655596 | 1925 | spin_lock(&ci->i_ceph_lock); |
2d6795fb | 1926 | dout("set_size %p %llu -> %llu\n", inode, i_size_read(inode), size); |
99c88e69 | 1927 | i_size_write(inode, size); |
400e1286 | 1928 | ceph_fscache_update(inode); |
224a7542 | 1929 | inode->i_blocks = calc_inode_blocks(size); |
355da1eb | 1930 | |
efb0ca76 | 1931 | ret = __ceph_should_report_size(ci); |
355da1eb | 1932 | |
be655596 | 1933 | spin_unlock(&ci->i_ceph_lock); |
400e1286 | 1934 | |
355da1eb SW |
1935 | return ret; |
1936 | } | |
1937 | ||
64f28c62 | 1938 | void ceph_queue_inode_work(struct inode *inode, int work_bit) |
355da1eb | 1939 | { |
64f28c62 | 1940 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); |
1cf89a8d | 1941 | struct ceph_inode_info *ci = ceph_inode(inode); |
64f28c62 | 1942 | set_bit(work_bit, &ci->i_work_mask); |
1cf89a8d YZ |
1943 | |
1944 | ihold(inode); | |
64f28c62 JL |
1945 | if (queue_work(fsc->inode_wq, &ci->i_work)) { |
1946 | dout("queue_inode_work %p, mask=%lx\n", inode, ci->i_work_mask); | |
1cf89a8d | 1947 | } else { |
64f28c62 | 1948 | dout("queue_inode_work %p already queued, mask=%lx\n", |
1cf89a8d YZ |
1949 | inode, ci->i_work_mask); |
1950 | iput(inode); | |
1951 | } | |
1952 | } | |
1953 | ||
1954 | static void ceph_do_invalidate_pages(struct inode *inode) | |
1955 | { | |
1956 | struct ceph_inode_info *ci = ceph_inode(inode); | |
355da1eb SW |
1957 | u32 orig_gen; |
1958 | int check = 0; | |
1959 | ||
400e1286 JL |
1960 | ceph_fscache_invalidate(inode, false); |
1961 | ||
b0d7c223 | 1962 | mutex_lock(&ci->i_truncate_mutex); |
6c93df5d | 1963 | |
5d6451b1 | 1964 | if (ceph_inode_is_shutdown(inode)) { |
6407fbb9 JL |
1965 | pr_warn_ratelimited("%s: inode %llx.%llx is shut down\n", |
1966 | __func__, ceph_vinop(inode)); | |
6c93df5d YZ |
1967 | mapping_set_error(inode->i_mapping, -EIO); |
1968 | truncate_pagecache(inode, 0); | |
1969 | mutex_unlock(&ci->i_truncate_mutex); | |
1970 | goto out; | |
1971 | } | |
1972 | ||
be655596 | 1973 | spin_lock(&ci->i_ceph_lock); |
355da1eb SW |
1974 | dout("invalidate_pages %p gen %d revoking %d\n", inode, |
1975 | ci->i_rdcache_gen, ci->i_rdcache_revoking); | |
cd045cb4 | 1976 | if (ci->i_rdcache_revoking != ci->i_rdcache_gen) { |
9563f88c YZ |
1977 | if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE)) |
1978 | check = 1; | |
be655596 | 1979 | spin_unlock(&ci->i_ceph_lock); |
b0d7c223 | 1980 | mutex_unlock(&ci->i_truncate_mutex); |
355da1eb SW |
1981 | goto out; |
1982 | } | |
1983 | orig_gen = ci->i_rdcache_gen; | |
be655596 | 1984 | spin_unlock(&ci->i_ceph_lock); |
355da1eb | 1985 | |
9abd4db7 | 1986 | if (invalidate_inode_pages2(inode->i_mapping) < 0) { |
6407fbb9 JL |
1987 | pr_err("invalidate_inode_pages2 %llx.%llx failed\n", |
1988 | ceph_vinop(inode)); | |
9abd4db7 | 1989 | } |
355da1eb | 1990 | |
be655596 | 1991 | spin_lock(&ci->i_ceph_lock); |
cd045cb4 SW |
1992 | if (orig_gen == ci->i_rdcache_gen && |
1993 | orig_gen == ci->i_rdcache_revoking) { | |
355da1eb SW |
1994 | dout("invalidate_pages %p gen %d successful\n", inode, |
1995 | ci->i_rdcache_gen); | |
cd045cb4 | 1996 | ci->i_rdcache_revoking--; |
355da1eb SW |
1997 | check = 1; |
1998 | } else { | |
cd045cb4 SW |
1999 | dout("invalidate_pages %p gen %d raced, now %d revoking %d\n", |
2000 | inode, orig_gen, ci->i_rdcache_gen, | |
2001 | ci->i_rdcache_revoking); | |
9563f88c YZ |
2002 | if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE)) |
2003 | check = 1; | |
355da1eb | 2004 | } |
be655596 | 2005 | spin_unlock(&ci->i_ceph_lock); |
b0d7c223 | 2006 | mutex_unlock(&ci->i_truncate_mutex); |
9563f88c | 2007 | out: |
355da1eb | 2008 | if (check) |
e4b731cc | 2009 | ceph_check_caps(ci, 0); |
3c6f6b79 SW |
2010 | } |
2011 | ||
355da1eb | 2012 | /* |
355da1eb SW |
2013 | * Make sure any pending truncation is applied before doing anything |
2014 | * that may depend on it. | |
2015 | */ | |
b415bf4f | 2016 | void __ceph_do_pending_vmtruncate(struct inode *inode) |
355da1eb SW |
2017 | { |
2018 | struct ceph_inode_info *ci = ceph_inode(inode); | |
2019 | u64 to; | |
a85f50b6 | 2020 | int wrbuffer_refs, finish = 0; |
355da1eb | 2021 | |
b0d7c223 | 2022 | mutex_lock(&ci->i_truncate_mutex); |
355da1eb | 2023 | retry: |
be655596 | 2024 | spin_lock(&ci->i_ceph_lock); |
355da1eb SW |
2025 | if (ci->i_truncate_pending == 0) { |
2026 | dout("__do_pending_vmtruncate %p none pending\n", inode); | |
be655596 | 2027 | spin_unlock(&ci->i_ceph_lock); |
b0d7c223 | 2028 | mutex_unlock(&ci->i_truncate_mutex); |
355da1eb SW |
2029 | return; |
2030 | } | |
2031 | ||
2032 | /* | |
2033 | * make sure any dirty snapped pages are flushed before we | |
2034 | * possibly truncate them.. so write AND block! | |
2035 | */ | |
2036 | if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) { | |
c8fd0d37 | 2037 | spin_unlock(&ci->i_ceph_lock); |
355da1eb SW |
2038 | dout("__do_pending_vmtruncate %p flushing snaps first\n", |
2039 | inode); | |
355da1eb SW |
2040 | filemap_write_and_wait_range(&inode->i_data, 0, |
2041 | inode->i_sb->s_maxbytes); | |
2042 | goto retry; | |
2043 | } | |
2044 | ||
b0d7c223 YZ |
2045 | /* there should be no reader or writer */ |
2046 | WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref); | |
2047 | ||
355da1eb SW |
2048 | to = ci->i_truncate_size; |
2049 | wrbuffer_refs = ci->i_wrbuffer_ref; | |
2050 | dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode, | |
2051 | ci->i_truncate_pending, to); | |
be655596 | 2052 | spin_unlock(&ci->i_ceph_lock); |
355da1eb | 2053 | |
400e1286 | 2054 | ceph_fscache_resize(inode, to); |
4e217b5d | 2055 | truncate_pagecache(inode, to); |
355da1eb | 2056 | |
be655596 | 2057 | spin_lock(&ci->i_ceph_lock); |
a85f50b6 YZ |
2058 | if (to == ci->i_truncate_size) { |
2059 | ci->i_truncate_pending = 0; | |
2060 | finish = 1; | |
2061 | } | |
be655596 | 2062 | spin_unlock(&ci->i_ceph_lock); |
a85f50b6 YZ |
2063 | if (!finish) |
2064 | goto retry; | |
355da1eb | 2065 | |
b0d7c223 YZ |
2066 | mutex_unlock(&ci->i_truncate_mutex); |
2067 | ||
355da1eb | 2068 | if (wrbuffer_refs == 0) |
e4b731cc | 2069 | ceph_check_caps(ci, 0); |
a85f50b6 YZ |
2070 | |
2071 | wake_up_all(&ci->i_cap_wq); | |
355da1eb SW |
2072 | } |
2073 | ||
1cf89a8d YZ |
2074 | static void ceph_inode_work(struct work_struct *work) |
2075 | { | |
2076 | struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info, | |
2077 | i_work); | |
874c8ca1 | 2078 | struct inode *inode = &ci->netfs.inode; |
1cf89a8d YZ |
2079 | |
2080 | if (test_and_clear_bit(CEPH_I_WORK_WRITEBACK, &ci->i_work_mask)) { | |
2081 | dout("writeback %p\n", inode); | |
2082 | filemap_fdatawrite(&inode->i_data); | |
2083 | } | |
2084 | if (test_and_clear_bit(CEPH_I_WORK_INVALIDATE_PAGES, &ci->i_work_mask)) | |
2085 | ceph_do_invalidate_pages(inode); | |
2086 | ||
2087 | if (test_and_clear_bit(CEPH_I_WORK_VMTRUNCATE, &ci->i_work_mask)) | |
2088 | __ceph_do_pending_vmtruncate(inode); | |
2089 | ||
a8810cdc | 2090 | if (test_and_clear_bit(CEPH_I_WORK_CHECK_CAPS, &ci->i_work_mask)) |
e4b731cc | 2091 | ceph_check_caps(ci, 0); |
a8810cdc JL |
2092 | |
2093 | if (test_and_clear_bit(CEPH_I_WORK_FLUSH_SNAPS, &ci->i_work_mask)) | |
2094 | ceph_flush_snaps(ci, NULL); | |
2095 | ||
1cf89a8d YZ |
2096 | iput(inode); |
2097 | } | |
2098 | ||
355da1eb SW |
2099 | /* |
2100 | * symlinks | |
2101 | */ | |
355da1eb | 2102 | static const struct inode_operations ceph_symlink_iops = { |
6b255391 | 2103 | .get_link = simple_get_link, |
0b932672 YZ |
2104 | .setattr = ceph_setattr, |
2105 | .getattr = ceph_getattr, | |
0b932672 | 2106 | .listxattr = ceph_listxattr, |
355da1eb SW |
2107 | }; |
2108 | ||
2d332d5b JL |
2109 | int __ceph_setattr(struct inode *inode, struct iattr *attr, |
2110 | struct ceph_iattr *cia) | |
355da1eb | 2111 | { |
355da1eb | 2112 | struct ceph_inode_info *ci = ceph_inode(inode); |
c62498d7 | 2113 | unsigned int ia_valid = attr->ia_valid; |
355da1eb | 2114 | struct ceph_mds_request *req; |
a26fecca | 2115 | struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc; |
f66fd9f0 | 2116 | struct ceph_cap_flush *prealloc_cf; |
355da1eb SW |
2117 | int issued; |
2118 | int release = 0, dirtied = 0; | |
2119 | int mask = 0; | |
2120 | int err = 0; | |
fca65b4a | 2121 | int inode_dirty_flags = 0; |
604d1b02 | 2122 | bool lock_snap_rwsem = false; |
355da1eb | 2123 | |
f66fd9f0 YZ |
2124 | prealloc_cf = ceph_alloc_cap_flush(); |
2125 | if (!prealloc_cf) | |
2126 | return -ENOMEM; | |
2127 | ||
355da1eb SW |
2128 | req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR, |
2129 | USE_AUTH_MDS); | |
f66fd9f0 YZ |
2130 | if (IS_ERR(req)) { |
2131 | ceph_free_cap_flush(prealloc_cf); | |
355da1eb | 2132 | return PTR_ERR(req); |
f66fd9f0 | 2133 | } |
355da1eb | 2134 | |
be655596 | 2135 | spin_lock(&ci->i_ceph_lock); |
355da1eb | 2136 | issued = __ceph_caps_issued(ci, NULL); |
604d1b02 YZ |
2137 | |
2138 | if (!ci->i_head_snapc && | |
2139 | (issued & (CEPH_CAP_ANY_EXCL | CEPH_CAP_FILE_WR))) { | |
2140 | lock_snap_rwsem = true; | |
2141 | if (!down_read_trylock(&mdsc->snap_rwsem)) { | |
2142 | spin_unlock(&ci->i_ceph_lock); | |
2143 | down_read(&mdsc->snap_rwsem); | |
2144 | spin_lock(&ci->i_ceph_lock); | |
2145 | issued = __ceph_caps_issued(ci, NULL); | |
2146 | } | |
2147 | } | |
2148 | ||
355da1eb | 2149 | dout("setattr %p issued %s\n", inode, ceph_cap_string(issued)); |
2d332d5b JL |
2150 | #if IS_ENABLED(CONFIG_FS_ENCRYPTION) |
2151 | if (cia && cia->fscrypt_auth) { | |
2152 | u32 len = ceph_fscrypt_auth_len(cia->fscrypt_auth); | |
2153 | ||
2154 | if (len > sizeof(*cia->fscrypt_auth)) { | |
2155 | err = -EINVAL; | |
2156 | spin_unlock(&ci->i_ceph_lock); | |
2157 | goto out; | |
2158 | } | |
2159 | ||
2160 | dout("setattr %llx:%llx fscrypt_auth len %u to %u)\n", | |
2161 | ceph_vinop(inode), ci->fscrypt_auth_len, len); | |
2162 | ||
2163 | /* It should never be re-set once set */ | |
2164 | WARN_ON_ONCE(ci->fscrypt_auth); | |
2165 | ||
2166 | if (issued & CEPH_CAP_AUTH_EXCL) { | |
2167 | dirtied |= CEPH_CAP_AUTH_EXCL; | |
2168 | kfree(ci->fscrypt_auth); | |
2169 | ci->fscrypt_auth = (u8 *)cia->fscrypt_auth; | |
2170 | ci->fscrypt_auth_len = len; | |
2171 | } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || | |
2172 | ci->fscrypt_auth_len != len || | |
2173 | memcmp(ci->fscrypt_auth, cia->fscrypt_auth, len)) { | |
2174 | req->r_fscrypt_auth = cia->fscrypt_auth; | |
2175 | mask |= CEPH_SETATTR_FSCRYPT_AUTH; | |
2176 | release |= CEPH_CAP_AUTH_SHARED; | |
2177 | } | |
2178 | cia->fscrypt_auth = NULL; | |
2179 | } | |
2180 | #else | |
2181 | if (cia && cia->fscrypt_auth) { | |
2182 | err = -EINVAL; | |
2183 | spin_unlock(&ci->i_ceph_lock); | |
2184 | goto out; | |
2185 | } | |
2186 | #endif /* CONFIG_FS_ENCRYPTION */ | |
355da1eb SW |
2187 | |
2188 | if (ia_valid & ATTR_UID) { | |
2189 | dout("setattr %p uid %d -> %d\n", inode, | |
bd2bae6a EB |
2190 | from_kuid(&init_user_ns, inode->i_uid), |
2191 | from_kuid(&init_user_ns, attr->ia_uid)); | |
355da1eb SW |
2192 | if (issued & CEPH_CAP_AUTH_EXCL) { |
2193 | inode->i_uid = attr->ia_uid; | |
2194 | dirtied |= CEPH_CAP_AUTH_EXCL; | |
2195 | } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || | |
ab871b90 EB |
2196 | !uid_eq(attr->ia_uid, inode->i_uid)) { |
2197 | req->r_args.setattr.uid = cpu_to_le32( | |
2198 | from_kuid(&init_user_ns, attr->ia_uid)); | |
355da1eb SW |
2199 | mask |= CEPH_SETATTR_UID; |
2200 | release |= CEPH_CAP_AUTH_SHARED; | |
2201 | } | |
2202 | } | |
2203 | if (ia_valid & ATTR_GID) { | |
2204 | dout("setattr %p gid %d -> %d\n", inode, | |
bd2bae6a EB |
2205 | from_kgid(&init_user_ns, inode->i_gid), |
2206 | from_kgid(&init_user_ns, attr->ia_gid)); | |
355da1eb SW |
2207 | if (issued & CEPH_CAP_AUTH_EXCL) { |
2208 | inode->i_gid = attr->ia_gid; | |
2209 | dirtied |= CEPH_CAP_AUTH_EXCL; | |
2210 | } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || | |
ab871b90 EB |
2211 | !gid_eq(attr->ia_gid, inode->i_gid)) { |
2212 | req->r_args.setattr.gid = cpu_to_le32( | |
2213 | from_kgid(&init_user_ns, attr->ia_gid)); | |
355da1eb SW |
2214 | mask |= CEPH_SETATTR_GID; |
2215 | release |= CEPH_CAP_AUTH_SHARED; | |
2216 | } | |
2217 | } | |
2218 | if (ia_valid & ATTR_MODE) { | |
2219 | dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode, | |
2220 | attr->ia_mode); | |
2221 | if (issued & CEPH_CAP_AUTH_EXCL) { | |
2222 | inode->i_mode = attr->ia_mode; | |
2223 | dirtied |= CEPH_CAP_AUTH_EXCL; | |
2224 | } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || | |
2225 | attr->ia_mode != inode->i_mode) { | |
7221fe4c | 2226 | inode->i_mode = attr->ia_mode; |
355da1eb SW |
2227 | req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode); |
2228 | mask |= CEPH_SETATTR_MODE; | |
2229 | release |= CEPH_CAP_AUTH_SHARED; | |
2230 | } | |
2231 | } | |
2232 | ||
2233 | if (ia_valid & ATTR_ATIME) { | |
13442b03 | 2234 | dout("setattr %p atime %lld.%ld -> %lld.%ld\n", inode, |
9bbeab41 AB |
2235 | inode->i_atime.tv_sec, inode->i_atime.tv_nsec, |
2236 | attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec); | |
355da1eb SW |
2237 | if (issued & CEPH_CAP_FILE_EXCL) { |
2238 | ci->i_time_warp_seq++; | |
2239 | inode->i_atime = attr->ia_atime; | |
2240 | dirtied |= CEPH_CAP_FILE_EXCL; | |
2241 | } else if ((issued & CEPH_CAP_FILE_WR) && | |
95582b00 | 2242 | timespec64_compare(&inode->i_atime, |
355da1eb SW |
2243 | &attr->ia_atime) < 0) { |
2244 | inode->i_atime = attr->ia_atime; | |
2245 | dirtied |= CEPH_CAP_FILE_WR; | |
2246 | } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || | |
95582b00 | 2247 | !timespec64_equal(&inode->i_atime, &attr->ia_atime)) { |
9bbeab41 AB |
2248 | ceph_encode_timespec64(&req->r_args.setattr.atime, |
2249 | &attr->ia_atime); | |
355da1eb | 2250 | mask |= CEPH_SETATTR_ATIME; |
be70489e YZ |
2251 | release |= CEPH_CAP_FILE_SHARED | |
2252 | CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR; | |
355da1eb SW |
2253 | } |
2254 | } | |
c62498d7 | 2255 | if (ia_valid & ATTR_SIZE) { |
2d6795fb JL |
2256 | loff_t isize = i_size_read(inode); |
2257 | ||
2258 | dout("setattr %p size %lld -> %lld\n", inode, isize, attr->ia_size); | |
e90334e8 XL |
2259 | if ((issued & CEPH_CAP_FILE_EXCL) && attr->ia_size >= isize) { |
2260 | if (attr->ia_size > isize) { | |
2261 | i_size_write(inode, attr->ia_size); | |
2262 | inode->i_blocks = calc_inode_blocks(attr->ia_size); | |
2263 | ci->i_reported_size = attr->ia_size; | |
2264 | dirtied |= CEPH_CAP_FILE_EXCL; | |
2265 | ia_valid |= ATTR_MTIME; | |
2266 | } | |
c62498d7 | 2267 | } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || |
2d6795fb | 2268 | attr->ia_size != isize) { |
c62498d7 | 2269 | req->r_args.setattr.size = cpu_to_le64(attr->ia_size); |
2d6795fb | 2270 | req->r_args.setattr.old_size = cpu_to_le64(isize); |
c62498d7 JL |
2271 | mask |= CEPH_SETATTR_SIZE; |
2272 | release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_EXCL | | |
2273 | CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR; | |
2274 | } | |
2275 | } | |
355da1eb | 2276 | if (ia_valid & ATTR_MTIME) { |
13442b03 | 2277 | dout("setattr %p mtime %lld.%ld -> %lld.%ld\n", inode, |
9bbeab41 AB |
2278 | inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec, |
2279 | attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec); | |
355da1eb SW |
2280 | if (issued & CEPH_CAP_FILE_EXCL) { |
2281 | ci->i_time_warp_seq++; | |
2282 | inode->i_mtime = attr->ia_mtime; | |
2283 | dirtied |= CEPH_CAP_FILE_EXCL; | |
2284 | } else if ((issued & CEPH_CAP_FILE_WR) && | |
95582b00 | 2285 | timespec64_compare(&inode->i_mtime, |
355da1eb SW |
2286 | &attr->ia_mtime) < 0) { |
2287 | inode->i_mtime = attr->ia_mtime; | |
2288 | dirtied |= CEPH_CAP_FILE_WR; | |
2289 | } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || | |
95582b00 | 2290 | !timespec64_equal(&inode->i_mtime, &attr->ia_mtime)) { |
9bbeab41 AB |
2291 | ceph_encode_timespec64(&req->r_args.setattr.mtime, |
2292 | &attr->ia_mtime); | |
355da1eb | 2293 | mask |= CEPH_SETATTR_MTIME; |
be70489e YZ |
2294 | release |= CEPH_CAP_FILE_SHARED | |
2295 | CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR; | |
355da1eb SW |
2296 | } |
2297 | } | |
355da1eb SW |
2298 | |
2299 | /* these do nothing */ | |
2300 | if (ia_valid & ATTR_CTIME) { | |
2301 | bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME| | |
2302 | ATTR_MODE|ATTR_UID|ATTR_GID)) == 0; | |
13442b03 | 2303 | dout("setattr %p ctime %lld.%ld -> %lld.%ld (%s)\n", inode, |
9bbeab41 AB |
2304 | inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec, |
2305 | attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec, | |
355da1eb | 2306 | only ? "ctime only" : "ignored"); |
355da1eb SW |
2307 | if (only) { |
2308 | /* | |
2309 | * if kernel wants to dirty ctime but nothing else, | |
2310 | * we need to choose a cap to dirty under, or do | |
2311 | * a almost-no-op setattr | |
2312 | */ | |
2313 | if (issued & CEPH_CAP_AUTH_EXCL) | |
2314 | dirtied |= CEPH_CAP_AUTH_EXCL; | |
2315 | else if (issued & CEPH_CAP_FILE_EXCL) | |
2316 | dirtied |= CEPH_CAP_FILE_EXCL; | |
2317 | else if (issued & CEPH_CAP_XATTR_EXCL) | |
2318 | dirtied |= CEPH_CAP_XATTR_EXCL; | |
2319 | else | |
2320 | mask |= CEPH_SETATTR_CTIME; | |
2321 | } | |
2322 | } | |
2323 | if (ia_valid & ATTR_FILE) | |
2324 | dout("setattr %p ATTR_FILE ... hrm!\n", inode); | |
2325 | ||
2326 | if (dirtied) { | |
f66fd9f0 YZ |
2327 | inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied, |
2328 | &prealloc_cf); | |
4ca2fea6 | 2329 | inode->i_ctime = attr->ia_ctime; |
b4b924c7 | 2330 | inode_inc_iversion_raw(inode); |
355da1eb SW |
2331 | } |
2332 | ||
2333 | release &= issued; | |
be655596 | 2334 | spin_unlock(&ci->i_ceph_lock); |
604d1b02 YZ |
2335 | if (lock_snap_rwsem) |
2336 | up_read(&mdsc->snap_rwsem); | |
355da1eb | 2337 | |
fca65b4a SW |
2338 | if (inode_dirty_flags) |
2339 | __mark_inode_dirty(inode, inode_dirty_flags); | |
2340 | ||
355da1eb | 2341 | if (mask) { |
70b666c3 SW |
2342 | req->r_inode = inode; |
2343 | ihold(inode); | |
355da1eb SW |
2344 | req->r_inode_drop = release; |
2345 | req->r_args.setattr.mask = cpu_to_le32(mask); | |
2346 | req->r_num_caps = 1; | |
0ed1e90a | 2347 | req->r_stamp = attr->ia_ctime; |
752c8bdc | 2348 | err = ceph_mdsc_do_request(mdsc, NULL, req); |
355da1eb | 2349 | } |
2d332d5b | 2350 | out: |
355da1eb SW |
2351 | dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err, |
2352 | ceph_cap_string(dirtied), mask); | |
2353 | ||
355da1eb | 2354 | ceph_mdsc_put_request(req); |
f66fd9f0 | 2355 | ceph_free_cap_flush(prealloc_cf); |
8179a101 YZ |
2356 | |
2357 | if (err >= 0 && (mask & CEPH_SETATTR_SIZE)) | |
2358 | __ceph_do_pending_vmtruncate(inode); | |
2359 | ||
355da1eb SW |
2360 | return err; |
2361 | } | |
2362 | ||
a26fecca AG |
2363 | /* |
2364 | * setattr | |
2365 | */ | |
c1632a0f | 2366 | int ceph_setattr(struct mnt_idmap *idmap, struct dentry *dentry, |
549c7297 | 2367 | struct iattr *attr) |
a26fecca | 2368 | { |
fd5472ed | 2369 | struct inode *inode = d_inode(dentry); |
36a4c72d | 2370 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); |
fd5472ed JK |
2371 | int err; |
2372 | ||
2373 | if (ceph_snap(inode) != CEPH_NOSNAP) | |
2374 | return -EROFS; | |
2375 | ||
5d6451b1 JL |
2376 | if (ceph_inode_is_shutdown(inode)) |
2377 | return -ESTALE; | |
2378 | ||
c1632a0f | 2379 | err = setattr_prepare(&nop_mnt_idmap, dentry, attr); |
fd5472ed JK |
2380 | if (err != 0) |
2381 | return err; | |
2382 | ||
36a4c72d | 2383 | if ((attr->ia_valid & ATTR_SIZE) && |
2d6795fb | 2384 | attr->ia_size > max(i_size_read(inode), fsc->max_file_size)) |
36a4c72d CX |
2385 | return -EFBIG; |
2386 | ||
2b83845f LH |
2387 | if ((attr->ia_valid & ATTR_SIZE) && |
2388 | ceph_quota_is_max_bytes_exceeded(inode, attr->ia_size)) | |
2389 | return -EDQUOT; | |
2390 | ||
2d332d5b | 2391 | err = __ceph_setattr(inode, attr, NULL); |
8179a101 YZ |
2392 | |
2393 | if (err >= 0 && (attr->ia_valid & ATTR_MODE)) | |
13e83a49 | 2394 | err = posix_acl_chmod(&nop_mnt_idmap, dentry, attr->ia_mode); |
8179a101 YZ |
2395 | |
2396 | return err; | |
a26fecca AG |
2397 | } |
2398 | ||
5eed80fb XL |
2399 | int ceph_try_to_choose_auth_mds(struct inode *inode, int mask) |
2400 | { | |
2401 | int issued = ceph_caps_issued(ceph_inode(inode)); | |
2402 | ||
2403 | /* | |
2404 | * If any 'x' caps is issued we can just choose the auth MDS | |
2405 | * instead of the random replica MDSes. Because only when the | |
2406 | * Locker is in LOCK_EXEC state will the loner client could | |
2407 | * get the 'x' caps. And if we send the getattr requests to | |
2408 | * any replica MDS it must auth pin and tries to rdlock from | |
2409 | * the auth MDS, and then the auth MDS need to do the Locker | |
2410 | * state transition to LOCK_SYNC. And after that the lock state | |
2411 | * will change back. | |
2412 | * | |
2413 | * This cost much when doing the Locker state transition and | |
2414 | * usually will need to revoke caps from clients. | |
8266c4d7 XL |
2415 | * |
2416 | * And for the 'Xs' caps for getxattr we will also choose the | |
2417 | * auth MDS, because the MDS side code is buggy due to setxattr | |
2418 | * won't notify the replica MDSes when the values changed and | |
2419 | * the replica MDS will return the old values. Though we will | |
2420 | * fix it in MDS code, but this still makes sense for old ceph. | |
5eed80fb XL |
2421 | */ |
2422 | if (((mask & CEPH_CAP_ANY_SHARED) && (issued & CEPH_CAP_ANY_EXCL)) | |
8266c4d7 | 2423 | || (mask & (CEPH_STAT_RSTAT | CEPH_STAT_CAP_XATTR))) |
5eed80fb XL |
2424 | return USE_AUTH_MDS; |
2425 | else | |
2426 | return USE_ANY_MDS; | |
2427 | } | |
2428 | ||
355da1eb SW |
2429 | /* |
2430 | * Verify that we have a lease on the given mask. If not, | |
2431 | * do a getattr against an mds. | |
2432 | */ | |
01deead0 YZ |
2433 | int __ceph_do_getattr(struct inode *inode, struct page *locked_page, |
2434 | int mask, bool force) | |
355da1eb | 2435 | { |
3d14c5d2 YS |
2436 | struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb); |
2437 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
355da1eb | 2438 | struct ceph_mds_request *req; |
49a9f4f6 | 2439 | int mode; |
355da1eb SW |
2440 | int err; |
2441 | ||
2442 | if (ceph_snap(inode) == CEPH_SNAPDIR) { | |
2443 | dout("do_getattr inode %p SNAPDIR\n", inode); | |
2444 | return 0; | |
2445 | } | |
2446 | ||
01deead0 YZ |
2447 | dout("do_getattr inode %p mask %s mode 0%o\n", |
2448 | inode, ceph_cap_string(mask), inode->i_mode); | |
1af16d54 XL |
2449 | if (!force && ceph_caps_issued_mask_metric(ceph_inode(inode), mask, 1)) |
2450 | return 0; | |
355da1eb | 2451 | |
5eed80fb | 2452 | mode = ceph_try_to_choose_auth_mds(inode, mask); |
49a9f4f6 | 2453 | req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, mode); |
355da1eb SW |
2454 | if (IS_ERR(req)) |
2455 | return PTR_ERR(req); | |
70b666c3 SW |
2456 | req->r_inode = inode; |
2457 | ihold(inode); | |
355da1eb SW |
2458 | req->r_num_caps = 1; |
2459 | req->r_args.getattr.mask = cpu_to_le32(mask); | |
01deead0 | 2460 | req->r_locked_page = locked_page; |
355da1eb | 2461 | err = ceph_mdsc_do_request(mdsc, NULL, req); |
01deead0 YZ |
2462 | if (locked_page && err == 0) { |
2463 | u64 inline_version = req->r_reply_info.targeti.inline_version; | |
2464 | if (inline_version == 0) { | |
2465 | /* the reply is supposed to contain inline data */ | |
2466 | err = -EINVAL; | |
48490776 XL |
2467 | } else if (inline_version == CEPH_INLINE_NONE || |
2468 | inline_version == 1) { | |
01deead0 YZ |
2469 | err = -ENODATA; |
2470 | } else { | |
2471 | err = req->r_reply_info.targeti.inline_len; | |
2472 | } | |
2473 | } | |
355da1eb SW |
2474 | ceph_mdsc_put_request(req); |
2475 | dout("do_getattr result=%d\n", err); | |
2476 | return err; | |
2477 | } | |
2478 | ||
6ddf5f16 MC |
2479 | int ceph_do_getvxattr(struct inode *inode, const char *name, void *value, |
2480 | size_t size) | |
2481 | { | |
2482 | struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb); | |
2483 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
2484 | struct ceph_mds_request *req; | |
2485 | int mode = USE_AUTH_MDS; | |
2486 | int err; | |
2487 | char *xattr_value; | |
2488 | size_t xattr_value_len; | |
2489 | ||
2490 | req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETVXATTR, mode); | |
2491 | if (IS_ERR(req)) { | |
2492 | err = -ENOMEM; | |
2493 | goto out; | |
2494 | } | |
2495 | ||
6eb06c46 | 2496 | req->r_feature_needed = CEPHFS_FEATURE_OP_GETVXATTR; |
6ddf5f16 MC |
2497 | req->r_path2 = kstrdup(name, GFP_NOFS); |
2498 | if (!req->r_path2) { | |
2499 | err = -ENOMEM; | |
2500 | goto put; | |
2501 | } | |
2502 | ||
2503 | ihold(inode); | |
2504 | req->r_inode = inode; | |
2505 | err = ceph_mdsc_do_request(mdsc, NULL, req); | |
2506 | if (err < 0) | |
2507 | goto put; | |
2508 | ||
2509 | xattr_value = req->r_reply_info.xattr_info.xattr_value; | |
2510 | xattr_value_len = req->r_reply_info.xattr_info.xattr_value_len; | |
2511 | ||
2512 | dout("do_getvxattr xattr_value_len:%zu, size:%zu\n", xattr_value_len, size); | |
2513 | ||
2514 | err = (int)xattr_value_len; | |
2515 | if (size == 0) | |
2516 | goto put; | |
2517 | ||
2518 | if (xattr_value_len > size) { | |
2519 | err = -ERANGE; | |
2520 | goto put; | |
2521 | } | |
2522 | ||
2523 | memcpy(value, xattr_value, xattr_value_len); | |
2524 | put: | |
2525 | ceph_mdsc_put_request(req); | |
2526 | out: | |
2527 | dout("do_getvxattr result=%d\n", err); | |
2528 | return err; | |
2529 | } | |
2530 | ||
355da1eb SW |
2531 | |
2532 | /* | |
2533 | * Check inode permissions. We verify we have a valid value for | |
2534 | * the AUTH cap, then call the generic handler. | |
2535 | */ | |
4609e1f1 | 2536 | int ceph_permission(struct mnt_idmap *idmap, struct inode *inode, |
549c7297 | 2537 | int mask) |
355da1eb | 2538 | { |
b74c79e9 NP |
2539 | int err; |
2540 | ||
10556cb2 | 2541 | if (mask & MAY_NOT_BLOCK) |
b74c79e9 NP |
2542 | return -ECHILD; |
2543 | ||
508b32d8 | 2544 | err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED, false); |
355da1eb SW |
2545 | |
2546 | if (!err) | |
4609e1f1 | 2547 | err = generic_permission(&nop_mnt_idmap, inode, mask); |
355da1eb SW |
2548 | return err; |
2549 | } | |
2550 | ||
428bb68a | 2551 | /* Craft a mask of needed caps given a set of requested statx attrs. */ |
04fabb11 | 2552 | static int statx_to_caps(u32 want, umode_t mode) |
428bb68a JL |
2553 | { |
2554 | int mask = 0; | |
2555 | ||
f6102994 | 2556 | if (want & (STATX_MODE|STATX_UID|STATX_GID|STATX_CTIME|STATX_BTIME|STATX_CHANGE_COOKIE)) |
428bb68a JL |
2557 | mask |= CEPH_CAP_AUTH_SHARED; |
2558 | ||
f6102994 | 2559 | if (want & (STATX_NLINK|STATX_CTIME|STATX_CHANGE_COOKIE)) { |
04fabb11 JL |
2560 | /* |
2561 | * The link count for directories depends on inode->i_subdirs, | |
2562 | * and that is only updated when Fs caps are held. | |
2563 | */ | |
2564 | if (S_ISDIR(mode)) | |
2565 | mask |= CEPH_CAP_FILE_SHARED; | |
2566 | else | |
2567 | mask |= CEPH_CAP_LINK_SHARED; | |
2568 | } | |
428bb68a | 2569 | |
f6102994 | 2570 | if (want & (STATX_ATIME|STATX_MTIME|STATX_CTIME|STATX_SIZE|STATX_BLOCKS|STATX_CHANGE_COOKIE)) |
428bb68a JL |
2571 | mask |= CEPH_CAP_FILE_SHARED; |
2572 | ||
f6102994 | 2573 | if (want & (STATX_CTIME|STATX_CHANGE_COOKIE)) |
428bb68a JL |
2574 | mask |= CEPH_CAP_XATTR_SHARED; |
2575 | ||
2576 | return mask; | |
2577 | } | |
2578 | ||
355da1eb | 2579 | /* |
428bb68a JL |
2580 | * Get all the attributes. If we have sufficient caps for the requested attrs, |
2581 | * then we can avoid talking to the MDS at all. | |
355da1eb | 2582 | */ |
b74d24f7 | 2583 | int ceph_getattr(struct mnt_idmap *idmap, const struct path *path, |
549c7297 | 2584 | struct kstat *stat, u32 request_mask, unsigned int flags) |
355da1eb | 2585 | { |
a528d35e | 2586 | struct inode *inode = d_inode(path->dentry); |
aa87052d | 2587 | struct super_block *sb = inode->i_sb; |
232d4b01 | 2588 | struct ceph_inode_info *ci = ceph_inode(inode); |
58981784 | 2589 | u32 valid_mask = STATX_BASIC_STATS; |
428bb68a | 2590 | int err = 0; |
355da1eb | 2591 | |
5d6451b1 JL |
2592 | if (ceph_inode_is_shutdown(inode)) |
2593 | return -ESTALE; | |
2594 | ||
428bb68a | 2595 | /* Skip the getattr altogether if we're asked not to sync */ |
261998c3 | 2596 | if ((flags & AT_STATX_SYNC_TYPE) != AT_STATX_DONT_SYNC) { |
04fabb11 JL |
2597 | err = ceph_do_getattr(inode, |
2598 | statx_to_caps(request_mask, inode->i_mode), | |
2599 | flags & AT_STATX_FORCE_SYNC); | |
428bb68a JL |
2600 | if (err) |
2601 | return err; | |
2602 | } | |
2603 | ||
b74d24f7 | 2604 | generic_fillattr(&nop_mnt_idmap, inode, stat); |
ebce3eb2 | 2605 | stat->ino = ceph_present_inode(inode); |
58981784 JL |
2606 | |
2607 | /* | |
2608 | * btime on newly-allocated inodes is 0, so if this is still set to | |
2609 | * that, then assume that it's not valid. | |
2610 | */ | |
2611 | if (ci->i_btime.tv_sec || ci->i_btime.tv_nsec) { | |
2612 | stat->btime = ci->i_btime; | |
2613 | valid_mask |= STATX_BTIME; | |
2614 | } | |
2615 | ||
f6102994 JL |
2616 | if (request_mask & STATX_CHANGE_COOKIE) { |
2617 | stat->change_cookie = inode_peek_iversion_raw(inode); | |
2618 | valid_mask |= STATX_CHANGE_COOKIE; | |
2619 | } | |
2620 | ||
428bb68a | 2621 | if (ceph_snap(inode) == CEPH_NOSNAP) |
aa87052d | 2622 | stat->dev = sb->s_dev; |
428bb68a JL |
2623 | else |
2624 | stat->dev = ci->i_snapid_map ? ci->i_snapid_map->dev : 0; | |
2625 | ||
2626 | if (S_ISDIR(inode->i_mode)) { | |
aa87052d | 2627 | if (ceph_test_mount_opt(ceph_sb_to_client(sb), RBYTES)) { |
428bb68a | 2628 | stat->size = ci->i_rbytes; |
aa87052d XL |
2629 | } else if (ceph_snap(inode) == CEPH_SNAPDIR) { |
2630 | struct ceph_inode_info *pci; | |
2631 | struct ceph_snap_realm *realm; | |
2632 | struct inode *parent; | |
2633 | ||
2634 | parent = ceph_lookup_inode(sb, ceph_ino(inode)); | |
f86a4866 | 2635 | if (IS_ERR(parent)) |
aa87052d XL |
2636 | return PTR_ERR(parent); |
2637 | ||
2638 | pci = ceph_inode(parent); | |
2639 | spin_lock(&pci->i_ceph_lock); | |
2640 | realm = pci->i_snap_realm; | |
2641 | if (realm) | |
2642 | stat->size = realm->num_snaps; | |
2643 | else | |
2644 | stat->size = 0; | |
2645 | spin_unlock(&pci->i_ceph_lock); | |
2646 | iput(parent); | |
2647 | } else { | |
428bb68a | 2648 | stat->size = ci->i_files + ci->i_subdirs; |
aa87052d | 2649 | } |
428bb68a JL |
2650 | stat->blocks = 0; |
2651 | stat->blksize = 65536; | |
2652 | /* | |
2653 | * Some applications rely on the number of st_nlink | |
2654 | * value on directories to be either 0 (if unlinked) | |
2655 | * or 2 + number of subdirectories. | |
2656 | */ | |
2657 | if (stat->nlink == 1) | |
2658 | /* '.' + '..' + subdirs */ | |
2659 | stat->nlink = 1 + 1 + ci->i_subdirs; | |
355da1eb | 2660 | } |
428bb68a | 2661 | |
f6102994 JL |
2662 | stat->attributes_mask |= STATX_ATTR_CHANGE_MONOTONIC; |
2663 | stat->attributes |= STATX_ATTR_CHANGE_MONOTONIC; | |
58981784 | 2664 | stat->result_mask = request_mask & valid_mask; |
355da1eb SW |
2665 | return err; |
2666 | } | |
5d6451b1 JL |
2667 | |
2668 | void ceph_inode_shutdown(struct inode *inode) | |
2669 | { | |
2670 | struct ceph_inode_info *ci = ceph_inode(inode); | |
2671 | struct rb_node *p; | |
2672 | int iputs = 0; | |
2673 | bool invalidate = false; | |
2674 | ||
2675 | spin_lock(&ci->i_ceph_lock); | |
2676 | ci->i_ceph_flags |= CEPH_I_SHUTDOWN; | |
2677 | p = rb_first(&ci->i_caps); | |
2678 | while (p) { | |
2679 | struct ceph_cap *cap = rb_entry(p, struct ceph_cap, ci_node); | |
2680 | ||
2681 | p = rb_next(p); | |
2682 | iputs += ceph_purge_inode_cap(inode, cap, &invalidate); | |
2683 | } | |
2684 | spin_unlock(&ci->i_ceph_lock); | |
2685 | ||
2686 | if (invalidate) | |
2687 | ceph_queue_invalidate(inode); | |
2688 | while (iputs--) | |
2689 | iput(inode); | |
2690 | } |