]>
Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
3d14c5d2 | 2 | #include <linux/ceph/ceph_debug.h> |
355da1eb SW |
3 | |
4 | #include <linux/module.h> | |
5 | #include <linux/fs.h> | |
355da1eb SW |
6 | #include <linux/slab.h> |
7 | #include <linux/string.h> | |
8 | #include <linux/uaccess.h> | |
9 | #include <linux/kernel.h> | |
355da1eb SW |
10 | #include <linux/writeback.h> |
11 | #include <linux/vmalloc.h> | |
2cdeb1e4 | 12 | #include <linux/xattr.h> |
4db658ea | 13 | #include <linux/posix_acl.h> |
3e7fbe9c | 14 | #include <linux/random.h> |
a407846e | 15 | #include <linux/sort.h> |
355da1eb SW |
16 | |
17 | #include "super.h" | |
3d14c5d2 | 18 | #include "mds_client.h" |
99ccbd22 | 19 | #include "cache.h" |
3d14c5d2 | 20 | #include <linux/ceph/decode.h> |
355da1eb SW |
21 | |
22 | /* | |
23 | * Ceph inode operations | |
24 | * | |
25 | * Implement basic inode helpers (get, alloc) and inode ops (getattr, | |
26 | * setattr, etc.), xattr helpers, and helpers for assimilating | |
27 | * metadata returned by the MDS into our cache. | |
28 | * | |
29 | * Also define helpers for doing asynchronous writeback, invalidation, | |
30 | * and truncation for the benefit of those who can't afford to block | |
31 | * (typically because they are in the message handler path). | |
32 | */ | |
33 | ||
34 | static const struct inode_operations ceph_symlink_iops; | |
35 | ||
3c6f6b79 SW |
36 | static void ceph_invalidate_work(struct work_struct *work); |
37 | static void ceph_writeback_work(struct work_struct *work); | |
38 | static void ceph_vmtruncate_work(struct work_struct *work); | |
355da1eb SW |
39 | |
40 | /* | |
41 | * find or create an inode, given the ceph ino number | |
42 | */ | |
ad1fee96 YS |
43 | static int ceph_set_ino_cb(struct inode *inode, void *data) |
44 | { | |
45 | ceph_inode(inode)->i_vino = *(struct ceph_vino *)data; | |
46 | inode->i_ino = ceph_vino_to_ino(*(struct ceph_vino *)data); | |
47 | return 0; | |
48 | } | |
49 | ||
355da1eb SW |
50 | struct inode *ceph_get_inode(struct super_block *sb, struct ceph_vino vino) |
51 | { | |
52 | struct inode *inode; | |
53 | ino_t t = ceph_vino_to_ino(vino); | |
54 | ||
55 | inode = iget5_locked(sb, t, ceph_ino_compare, ceph_set_ino_cb, &vino); | |
d37b1d99 | 56 | if (!inode) |
355da1eb SW |
57 | return ERR_PTR(-ENOMEM); |
58 | if (inode->i_state & I_NEW) { | |
59 | dout("get_inode created new inode %p %llx.%llx ino %llx\n", | |
60 | inode, ceph_vinop(inode), (u64)inode->i_ino); | |
61 | unlock_new_inode(inode); | |
62 | } | |
63 | ||
64 | dout("get_inode on %lu=%llx.%llx got %p\n", inode->i_ino, vino.ino, | |
65 | vino.snap, inode); | |
66 | return inode; | |
67 | } | |
68 | ||
69 | /* | |
70 | * get/constuct snapdir inode for a given directory | |
71 | */ | |
72 | struct inode *ceph_get_snapdir(struct inode *parent) | |
73 | { | |
74 | struct ceph_vino vino = { | |
75 | .ino = ceph_ino(parent), | |
76 | .snap = CEPH_SNAPDIR, | |
77 | }; | |
78 | struct inode *inode = ceph_get_inode(parent->i_sb, vino); | |
b377ff13 | 79 | struct ceph_inode_info *ci = ceph_inode(inode); |
355da1eb SW |
80 | |
81 | BUG_ON(!S_ISDIR(parent->i_mode)); | |
82 | if (IS_ERR(inode)) | |
7e34bc52 | 83 | return inode; |
355da1eb SW |
84 | inode->i_mode = parent->i_mode; |
85 | inode->i_uid = parent->i_uid; | |
86 | inode->i_gid = parent->i_gid; | |
38c48b5f YZ |
87 | inode->i_op = &ceph_snapdir_iops; |
88 | inode->i_fop = &ceph_snapdir_fops; | |
b377ff13 SW |
89 | ci->i_snap_caps = CEPH_CAP_PIN; /* so we can open */ |
90 | ci->i_rbytes = 0; | |
355da1eb SW |
91 | return inode; |
92 | } | |
93 | ||
94 | const struct inode_operations ceph_file_iops = { | |
95 | .permission = ceph_permission, | |
96 | .setattr = ceph_setattr, | |
97 | .getattr = ceph_getattr, | |
355da1eb | 98 | .listxattr = ceph_listxattr, |
7221fe4c | 99 | .get_acl = ceph_get_acl, |
72466d0b | 100 | .set_acl = ceph_set_acl, |
355da1eb SW |
101 | }; |
102 | ||
103 | ||
104 | /* | |
105 | * We use a 'frag tree' to keep track of the MDS's directory fragments | |
106 | * for a given inode (usually there is just a single fragment). We | |
107 | * need to know when a child frag is delegated to a new MDS, or when | |
108 | * it is flagged as replicated, so we can direct our requests | |
109 | * accordingly. | |
110 | */ | |
111 | ||
112 | /* | |
113 | * find/create a frag in the tree | |
114 | */ | |
115 | static struct ceph_inode_frag *__get_or_create_frag(struct ceph_inode_info *ci, | |
116 | u32 f) | |
117 | { | |
118 | struct rb_node **p; | |
119 | struct rb_node *parent = NULL; | |
120 | struct ceph_inode_frag *frag; | |
121 | int c; | |
122 | ||
123 | p = &ci->i_fragtree.rb_node; | |
124 | while (*p) { | |
125 | parent = *p; | |
126 | frag = rb_entry(parent, struct ceph_inode_frag, node); | |
127 | c = ceph_frag_compare(f, frag->frag); | |
128 | if (c < 0) | |
129 | p = &(*p)->rb_left; | |
130 | else if (c > 0) | |
131 | p = &(*p)->rb_right; | |
132 | else | |
133 | return frag; | |
134 | } | |
135 | ||
136 | frag = kmalloc(sizeof(*frag), GFP_NOFS); | |
51308806 | 137 | if (!frag) |
355da1eb | 138 | return ERR_PTR(-ENOMEM); |
51308806 | 139 | |
355da1eb SW |
140 | frag->frag = f; |
141 | frag->split_by = 0; | |
142 | frag->mds = -1; | |
143 | frag->ndist = 0; | |
144 | ||
145 | rb_link_node(&frag->node, parent, p); | |
146 | rb_insert_color(&frag->node, &ci->i_fragtree); | |
147 | ||
148 | dout("get_or_create_frag added %llx.%llx frag %x\n", | |
149 | ceph_vinop(&ci->vfs_inode), f); | |
150 | return frag; | |
151 | } | |
152 | ||
153 | /* | |
154 | * find a specific frag @f | |
155 | */ | |
156 | struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci, u32 f) | |
157 | { | |
158 | struct rb_node *n = ci->i_fragtree.rb_node; | |
159 | ||
160 | while (n) { | |
161 | struct ceph_inode_frag *frag = | |
162 | rb_entry(n, struct ceph_inode_frag, node); | |
163 | int c = ceph_frag_compare(f, frag->frag); | |
164 | if (c < 0) | |
165 | n = n->rb_left; | |
166 | else if (c > 0) | |
167 | n = n->rb_right; | |
168 | else | |
169 | return frag; | |
170 | } | |
171 | return NULL; | |
172 | } | |
173 | ||
174 | /* | |
175 | * Choose frag containing the given value @v. If @pfrag is | |
176 | * specified, copy the frag delegation info to the caller if | |
177 | * it is present. | |
178 | */ | |
3e7fbe9c YZ |
179 | static u32 __ceph_choose_frag(struct ceph_inode_info *ci, u32 v, |
180 | struct ceph_inode_frag *pfrag, int *found) | |
355da1eb SW |
181 | { |
182 | u32 t = ceph_frag_make(0, 0); | |
183 | struct ceph_inode_frag *frag; | |
184 | unsigned nway, i; | |
185 | u32 n; | |
186 | ||
187 | if (found) | |
188 | *found = 0; | |
189 | ||
355da1eb SW |
190 | while (1) { |
191 | WARN_ON(!ceph_frag_contains_value(t, v)); | |
192 | frag = __ceph_find_frag(ci, t); | |
193 | if (!frag) | |
194 | break; /* t is a leaf */ | |
195 | if (frag->split_by == 0) { | |
196 | if (pfrag) | |
197 | memcpy(pfrag, frag, sizeof(*pfrag)); | |
198 | if (found) | |
199 | *found = 1; | |
200 | break; | |
201 | } | |
202 | ||
203 | /* choose child */ | |
204 | nway = 1 << frag->split_by; | |
205 | dout("choose_frag(%x) %x splits by %d (%d ways)\n", v, t, | |
206 | frag->split_by, nway); | |
207 | for (i = 0; i < nway; i++) { | |
208 | n = ceph_frag_make_child(t, frag->split_by, i); | |
209 | if (ceph_frag_contains_value(n, v)) { | |
210 | t = n; | |
211 | break; | |
212 | } | |
213 | } | |
214 | BUG_ON(i == nway); | |
215 | } | |
216 | dout("choose_frag(%x) = %x\n", v, t); | |
217 | ||
355da1eb SW |
218 | return t; |
219 | } | |
220 | ||
3e7fbe9c YZ |
221 | u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v, |
222 | struct ceph_inode_frag *pfrag, int *found) | |
223 | { | |
224 | u32 ret; | |
225 | mutex_lock(&ci->i_fragtree_mutex); | |
226 | ret = __ceph_choose_frag(ci, v, pfrag, found); | |
227 | mutex_unlock(&ci->i_fragtree_mutex); | |
228 | return ret; | |
229 | } | |
230 | ||
355da1eb SW |
231 | /* |
232 | * Process dirfrag (delegation) info from the mds. Include leaf | |
233 | * fragment in tree ONLY if ndist > 0. Otherwise, only | |
234 | * branches/splits are included in i_fragtree) | |
235 | */ | |
236 | static int ceph_fill_dirfrag(struct inode *inode, | |
237 | struct ceph_mds_reply_dirfrag *dirinfo) | |
238 | { | |
239 | struct ceph_inode_info *ci = ceph_inode(inode); | |
240 | struct ceph_inode_frag *frag; | |
241 | u32 id = le32_to_cpu(dirinfo->frag); | |
242 | int mds = le32_to_cpu(dirinfo->auth); | |
243 | int ndist = le32_to_cpu(dirinfo->ndist); | |
8d08503c | 244 | int diri_auth = -1; |
355da1eb SW |
245 | int i; |
246 | int err = 0; | |
247 | ||
8d08503c YZ |
248 | spin_lock(&ci->i_ceph_lock); |
249 | if (ci->i_auth_cap) | |
250 | diri_auth = ci->i_auth_cap->mds; | |
251 | spin_unlock(&ci->i_ceph_lock); | |
252 | ||
42172119 YZ |
253 | if (mds == -1) /* CDIR_AUTH_PARENT */ |
254 | mds = diri_auth; | |
255 | ||
355da1eb | 256 | mutex_lock(&ci->i_fragtree_mutex); |
8d08503c | 257 | if (ndist == 0 && mds == diri_auth) { |
355da1eb SW |
258 | /* no delegation info needed. */ |
259 | frag = __ceph_find_frag(ci, id); | |
260 | if (!frag) | |
261 | goto out; | |
262 | if (frag->split_by == 0) { | |
263 | /* tree leaf, remove */ | |
264 | dout("fill_dirfrag removed %llx.%llx frag %x" | |
265 | " (no ref)\n", ceph_vinop(inode), id); | |
266 | rb_erase(&frag->node, &ci->i_fragtree); | |
267 | kfree(frag); | |
268 | } else { | |
269 | /* tree branch, keep and clear */ | |
270 | dout("fill_dirfrag cleared %llx.%llx frag %x" | |
271 | " referral\n", ceph_vinop(inode), id); | |
272 | frag->mds = -1; | |
273 | frag->ndist = 0; | |
274 | } | |
275 | goto out; | |
276 | } | |
277 | ||
278 | ||
279 | /* find/add this frag to store mds delegation info */ | |
280 | frag = __get_or_create_frag(ci, id); | |
281 | if (IS_ERR(frag)) { | |
282 | /* this is not the end of the world; we can continue | |
283 | with bad/inaccurate delegation info */ | |
284 | pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n", | |
285 | ceph_vinop(inode), le32_to_cpu(dirinfo->frag)); | |
286 | err = -ENOMEM; | |
287 | goto out; | |
288 | } | |
289 | ||
290 | frag->mds = mds; | |
291 | frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP); | |
292 | for (i = 0; i < frag->ndist; i++) | |
293 | frag->dist[i] = le32_to_cpu(dirinfo->dist[i]); | |
294 | dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n", | |
295 | ceph_vinop(inode), frag->frag, frag->ndist); | |
296 | ||
297 | out: | |
298 | mutex_unlock(&ci->i_fragtree_mutex); | |
299 | return err; | |
300 | } | |
301 | ||
a407846e YZ |
302 | static int frag_tree_split_cmp(const void *l, const void *r) |
303 | { | |
304 | struct ceph_frag_tree_split *ls = (struct ceph_frag_tree_split*)l; | |
305 | struct ceph_frag_tree_split *rs = (struct ceph_frag_tree_split*)r; | |
fe2ed425 JL |
306 | return ceph_frag_compare(le32_to_cpu(ls->frag), |
307 | le32_to_cpu(rs->frag)); | |
a407846e YZ |
308 | } |
309 | ||
a4b7431f YZ |
310 | static bool is_frag_child(u32 f, struct ceph_inode_frag *frag) |
311 | { | |
312 | if (!frag) | |
313 | return f == ceph_frag_make(0, 0); | |
314 | if (ceph_frag_bits(f) != ceph_frag_bits(frag->frag) + frag->split_by) | |
315 | return false; | |
316 | return ceph_frag_contains_value(frag->frag, ceph_frag_value(f)); | |
317 | } | |
318 | ||
3e7fbe9c YZ |
319 | static int ceph_fill_fragtree(struct inode *inode, |
320 | struct ceph_frag_tree_head *fragtree, | |
321 | struct ceph_mds_reply_dirfrag *dirinfo) | |
322 | { | |
323 | struct ceph_inode_info *ci = ceph_inode(inode); | |
a4b7431f | 324 | struct ceph_inode_frag *frag, *prev_frag = NULL; |
3e7fbe9c | 325 | struct rb_node *rb_node; |
1b1bc16d YZ |
326 | unsigned i, split_by, nsplits; |
327 | u32 id; | |
3e7fbe9c YZ |
328 | bool update = false; |
329 | ||
330 | mutex_lock(&ci->i_fragtree_mutex); | |
331 | nsplits = le32_to_cpu(fragtree->nsplits); | |
1b1bc16d YZ |
332 | if (nsplits != ci->i_fragtree_nsplits) { |
333 | update = true; | |
334 | } else if (nsplits) { | |
3e7fbe9c YZ |
335 | i = prandom_u32() % nsplits; |
336 | id = le32_to_cpu(fragtree->splits[i].frag); | |
337 | if (!__ceph_find_frag(ci, id)) | |
338 | update = true; | |
339 | } else if (!RB_EMPTY_ROOT(&ci->i_fragtree)) { | |
340 | rb_node = rb_first(&ci->i_fragtree); | |
341 | frag = rb_entry(rb_node, struct ceph_inode_frag, node); | |
342 | if (frag->frag != ceph_frag_make(0, 0) || rb_next(rb_node)) | |
343 | update = true; | |
344 | } | |
345 | if (!update && dirinfo) { | |
346 | id = le32_to_cpu(dirinfo->frag); | |
347 | if (id != __ceph_choose_frag(ci, id, NULL, NULL)) | |
348 | update = true; | |
349 | } | |
350 | if (!update) | |
351 | goto out_unlock; | |
352 | ||
a407846e YZ |
353 | if (nsplits > 1) { |
354 | sort(fragtree->splits, nsplits, sizeof(fragtree->splits[0]), | |
355 | frag_tree_split_cmp, NULL); | |
356 | } | |
357 | ||
3e7fbe9c YZ |
358 | dout("fill_fragtree %llx.%llx\n", ceph_vinop(inode)); |
359 | rb_node = rb_first(&ci->i_fragtree); | |
360 | for (i = 0; i < nsplits; i++) { | |
361 | id = le32_to_cpu(fragtree->splits[i].frag); | |
1b1bc16d YZ |
362 | split_by = le32_to_cpu(fragtree->splits[i].by); |
363 | if (split_by == 0 || ceph_frag_bits(id) + split_by > 24) { | |
364 | pr_err("fill_fragtree %llx.%llx invalid split %d/%u, " | |
365 | "frag %x split by %d\n", ceph_vinop(inode), | |
366 | i, nsplits, id, split_by); | |
367 | continue; | |
368 | } | |
3e7fbe9c YZ |
369 | frag = NULL; |
370 | while (rb_node) { | |
371 | frag = rb_entry(rb_node, struct ceph_inode_frag, node); | |
372 | if (ceph_frag_compare(frag->frag, id) >= 0) { | |
373 | if (frag->frag != id) | |
374 | frag = NULL; | |
375 | else | |
376 | rb_node = rb_next(rb_node); | |
377 | break; | |
378 | } | |
379 | rb_node = rb_next(rb_node); | |
a4b7431f YZ |
380 | /* delete stale split/leaf node */ |
381 | if (frag->split_by > 0 || | |
382 | !is_frag_child(frag->frag, prev_frag)) { | |
383 | rb_erase(&frag->node, &ci->i_fragtree); | |
1b1bc16d YZ |
384 | if (frag->split_by > 0) |
385 | ci->i_fragtree_nsplits--; | |
a4b7431f YZ |
386 | kfree(frag); |
387 | } | |
3e7fbe9c YZ |
388 | frag = NULL; |
389 | } | |
390 | if (!frag) { | |
391 | frag = __get_or_create_frag(ci, id); | |
392 | if (IS_ERR(frag)) | |
393 | continue; | |
394 | } | |
1b1bc16d YZ |
395 | if (frag->split_by == 0) |
396 | ci->i_fragtree_nsplits++; | |
397 | frag->split_by = split_by; | |
3e7fbe9c | 398 | dout(" frag %x split by %d\n", frag->frag, frag->split_by); |
a4b7431f | 399 | prev_frag = frag; |
3e7fbe9c YZ |
400 | } |
401 | while (rb_node) { | |
402 | frag = rb_entry(rb_node, struct ceph_inode_frag, node); | |
403 | rb_node = rb_next(rb_node); | |
a4b7431f YZ |
404 | /* delete stale split/leaf node */ |
405 | if (frag->split_by > 0 || | |
406 | !is_frag_child(frag->frag, prev_frag)) { | |
407 | rb_erase(&frag->node, &ci->i_fragtree); | |
1b1bc16d YZ |
408 | if (frag->split_by > 0) |
409 | ci->i_fragtree_nsplits--; | |
a4b7431f YZ |
410 | kfree(frag); |
411 | } | |
3e7fbe9c YZ |
412 | } |
413 | out_unlock: | |
414 | mutex_unlock(&ci->i_fragtree_mutex); | |
415 | return 0; | |
416 | } | |
355da1eb SW |
417 | |
418 | /* | |
419 | * initialize a newly allocated inode. | |
420 | */ | |
421 | struct inode *ceph_alloc_inode(struct super_block *sb) | |
422 | { | |
423 | struct ceph_inode_info *ci; | |
424 | int i; | |
425 | ||
426 | ci = kmem_cache_alloc(ceph_inode_cachep, GFP_NOFS); | |
427 | if (!ci) | |
428 | return NULL; | |
429 | ||
430 | dout("alloc_inode %p\n", &ci->vfs_inode); | |
431 | ||
be655596 SW |
432 | spin_lock_init(&ci->i_ceph_lock); |
433 | ||
355da1eb | 434 | ci->i_version = 0; |
31c542a1 | 435 | ci->i_inline_version = 0; |
355da1eb SW |
436 | ci->i_time_warp_seq = 0; |
437 | ci->i_ceph_flags = 0; | |
fdd4e158 YZ |
438 | atomic64_set(&ci->i_ordered_count, 1); |
439 | atomic64_set(&ci->i_release_count, 1); | |
440 | atomic64_set(&ci->i_complete_seq[0], 0); | |
441 | atomic64_set(&ci->i_complete_seq[1], 0); | |
355da1eb SW |
442 | ci->i_symlink = NULL; |
443 | ||
6c0f3af7 | 444 | memset(&ci->i_dir_layout, 0, sizeof(ci->i_dir_layout)); |
30c156d9 | 445 | RCU_INIT_POINTER(ci->i_layout.pool_ns, NULL); |
6c0f3af7 | 446 | |
355da1eb SW |
447 | ci->i_fragtree = RB_ROOT; |
448 | mutex_init(&ci->i_fragtree_mutex); | |
449 | ||
450 | ci->i_xattrs.blob = NULL; | |
451 | ci->i_xattrs.prealloc_blob = NULL; | |
452 | ci->i_xattrs.dirty = false; | |
453 | ci->i_xattrs.index = RB_ROOT; | |
454 | ci->i_xattrs.count = 0; | |
455 | ci->i_xattrs.names_size = 0; | |
456 | ci->i_xattrs.vals_size = 0; | |
457 | ci->i_xattrs.version = 0; | |
458 | ci->i_xattrs.index_version = 0; | |
459 | ||
460 | ci->i_caps = RB_ROOT; | |
461 | ci->i_auth_cap = NULL; | |
462 | ci->i_dirty_caps = 0; | |
463 | ci->i_flushing_caps = 0; | |
464 | INIT_LIST_HEAD(&ci->i_dirty_item); | |
465 | INIT_LIST_HEAD(&ci->i_flushing_item); | |
f66fd9f0 | 466 | ci->i_prealloc_cap_flush = NULL; |
e4500b5e | 467 | INIT_LIST_HEAD(&ci->i_cap_flush_list); |
355da1eb SW |
468 | init_waitqueue_head(&ci->i_cap_wq); |
469 | ci->i_hold_caps_min = 0; | |
470 | ci->i_hold_caps_max = 0; | |
471 | INIT_LIST_HEAD(&ci->i_cap_delay_list); | |
355da1eb SW |
472 | INIT_LIST_HEAD(&ci->i_cap_snaps); |
473 | ci->i_head_snapc = NULL; | |
474 | ci->i_snap_caps = 0; | |
475 | ||
774a6a11 | 476 | for (i = 0; i < CEPH_FILE_MODE_BITS; i++) |
355da1eb SW |
477 | ci->i_nr_by_mode[i] = 0; |
478 | ||
b0d7c223 | 479 | mutex_init(&ci->i_truncate_mutex); |
355da1eb SW |
480 | ci->i_truncate_seq = 0; |
481 | ci->i_truncate_size = 0; | |
482 | ci->i_truncate_pending = 0; | |
483 | ||
484 | ci->i_max_size = 0; | |
485 | ci->i_reported_size = 0; | |
486 | ci->i_wanted_max_size = 0; | |
487 | ci->i_requested_max_size = 0; | |
488 | ||
489 | ci->i_pin_ref = 0; | |
490 | ci->i_rd_ref = 0; | |
491 | ci->i_rdcache_ref = 0; | |
492 | ci->i_wr_ref = 0; | |
d3d0720d | 493 | ci->i_wb_ref = 0; |
355da1eb SW |
494 | ci->i_wrbuffer_ref = 0; |
495 | ci->i_wrbuffer_ref_head = 0; | |
89aa5930 | 496 | atomic_set(&ci->i_filelock_ref, 0); |
355da1eb SW |
497 | ci->i_shared_gen = 0; |
498 | ci->i_rdcache_gen = 0; | |
499 | ci->i_rdcache_revoking = 0; | |
500 | ||
355da1eb | 501 | INIT_LIST_HEAD(&ci->i_unsafe_dirops); |
68cd5b4b | 502 | INIT_LIST_HEAD(&ci->i_unsafe_iops); |
355da1eb SW |
503 | spin_lock_init(&ci->i_unsafe_lock); |
504 | ||
505 | ci->i_snap_realm = NULL; | |
506 | INIT_LIST_HEAD(&ci->i_snap_realm_item); | |
507 | INIT_LIST_HEAD(&ci->i_snap_flush_item); | |
508 | ||
3c6f6b79 SW |
509 | INIT_WORK(&ci->i_wb_work, ceph_writeback_work); |
510 | INIT_WORK(&ci->i_pg_inv_work, ceph_invalidate_work); | |
355da1eb SW |
511 | |
512 | INIT_WORK(&ci->i_vmtruncate_work, ceph_vmtruncate_work); | |
513 | ||
99ccbd22 MT |
514 | ceph_fscache_inode_init(ci); |
515 | ||
355da1eb SW |
516 | return &ci->vfs_inode; |
517 | } | |
518 | ||
fa0d7e3d NP |
519 | static void ceph_i_callback(struct rcu_head *head) |
520 | { | |
521 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
522 | struct ceph_inode_info *ci = ceph_inode(inode); | |
523 | ||
fa0d7e3d NP |
524 | kmem_cache_free(ceph_inode_cachep, ci); |
525 | } | |
526 | ||
355da1eb SW |
527 | void ceph_destroy_inode(struct inode *inode) |
528 | { | |
529 | struct ceph_inode_info *ci = ceph_inode(inode); | |
530 | struct ceph_inode_frag *frag; | |
531 | struct rb_node *n; | |
532 | ||
533 | dout("destroy_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode)); | |
534 | ||
99ccbd22 MT |
535 | ceph_fscache_unregister_inode_cookie(ci); |
536 | ||
355da1eb SW |
537 | ceph_queue_caps_release(inode); |
538 | ||
8b218b8a SW |
539 | /* |
540 | * we may still have a snap_realm reference if there are stray | |
d9df2783 | 541 | * caps in i_snap_caps. |
8b218b8a SW |
542 | */ |
543 | if (ci->i_snap_realm) { | |
544 | struct ceph_mds_client *mdsc = | |
3d14c5d2 | 545 | ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc; |
8b218b8a SW |
546 | struct ceph_snap_realm *realm = ci->i_snap_realm; |
547 | ||
548 | dout(" dropping residual ref to snap realm %p\n", realm); | |
549 | spin_lock(&realm->inodes_with_caps_lock); | |
550 | list_del_init(&ci->i_snap_realm_item); | |
551 | spin_unlock(&realm->inodes_with_caps_lock); | |
552 | ceph_put_snap_realm(mdsc, realm); | |
553 | } | |
554 | ||
355da1eb SW |
555 | kfree(ci->i_symlink); |
556 | while ((n = rb_first(&ci->i_fragtree)) != NULL) { | |
557 | frag = rb_entry(n, struct ceph_inode_frag, node); | |
558 | rb_erase(n, &ci->i_fragtree); | |
559 | kfree(frag); | |
560 | } | |
1b1bc16d | 561 | ci->i_fragtree_nsplits = 0; |
355da1eb SW |
562 | |
563 | __ceph_destroy_xattrs(ci); | |
b6c1d5b8 SW |
564 | if (ci->i_xattrs.blob) |
565 | ceph_buffer_put(ci->i_xattrs.blob); | |
566 | if (ci->i_xattrs.prealloc_blob) | |
567 | ceph_buffer_put(ci->i_xattrs.prealloc_blob); | |
355da1eb | 568 | |
779fe0fb | 569 | ceph_put_string(rcu_dereference_raw(ci->i_layout.pool_ns)); |
30c156d9 | 570 | |
fa0d7e3d | 571 | call_rcu(&inode->i_rcu, ceph_i_callback); |
355da1eb SW |
572 | } |
573 | ||
9f12bd11 YZ |
574 | int ceph_drop_inode(struct inode *inode) |
575 | { | |
576 | /* | |
577 | * Positve dentry and corresponding inode are always accompanied | |
578 | * in MDS reply. So no need to keep inode in the cache after | |
579 | * dropping all its aliases. | |
580 | */ | |
581 | return 1; | |
582 | } | |
583 | ||
224a7542 YZ |
584 | static inline blkcnt_t calc_inode_blocks(u64 size) |
585 | { | |
586 | return (size + (1<<9) - 1) >> 9; | |
587 | } | |
588 | ||
355da1eb SW |
589 | /* |
590 | * Helpers to fill in size, ctime, mtime, and atime. We have to be | |
591 | * careful because either the client or MDS may have more up to date | |
592 | * info, depending on which capabilities are held, and whether | |
593 | * time_warp_seq or truncate_seq have increased. (Ordinarily, mtime | |
594 | * and size are monotonically increasing, except when utimes() or | |
595 | * truncate() increments the corresponding _seq values.) | |
596 | */ | |
597 | int ceph_fill_file_size(struct inode *inode, int issued, | |
598 | u32 truncate_seq, u64 truncate_size, u64 size) | |
599 | { | |
600 | struct ceph_inode_info *ci = ceph_inode(inode); | |
601 | int queue_trunc = 0; | |
602 | ||
603 | if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 || | |
604 | (truncate_seq == ci->i_truncate_seq && size > inode->i_size)) { | |
605 | dout("size %lld -> %llu\n", inode->i_size, size); | |
a3d714c3 YZ |
606 | if (size > 0 && S_ISDIR(inode->i_mode)) { |
607 | pr_err("fill_file_size non-zero size for directory\n"); | |
608 | size = 0; | |
609 | } | |
99c88e69 | 610 | i_size_write(inode, size); |
224a7542 | 611 | inode->i_blocks = calc_inode_blocks(size); |
355da1eb SW |
612 | ci->i_reported_size = size; |
613 | if (truncate_seq != ci->i_truncate_seq) { | |
614 | dout("truncate_seq %u -> %u\n", | |
615 | ci->i_truncate_seq, truncate_seq); | |
616 | ci->i_truncate_seq = truncate_seq; | |
b0d7c223 YZ |
617 | |
618 | /* the MDS should have revoked these caps */ | |
619 | WARN_ON_ONCE(issued & (CEPH_CAP_FILE_EXCL | | |
620 | CEPH_CAP_FILE_RD | | |
621 | CEPH_CAP_FILE_WR | | |
622 | CEPH_CAP_FILE_LAZYIO)); | |
3d497d85 YS |
623 | /* |
624 | * If we hold relevant caps, or in the case where we're | |
625 | * not the only client referencing this file and we | |
626 | * don't hold those caps, then we need to check whether | |
627 | * the file is either opened or mmaped | |
628 | */ | |
b0d7c223 YZ |
629 | if ((issued & (CEPH_CAP_FILE_CACHE| |
630 | CEPH_CAP_FILE_BUFFER)) || | |
3d497d85 YS |
631 | mapping_mapped(inode->i_mapping) || |
632 | __ceph_caps_file_wanted(ci)) { | |
355da1eb SW |
633 | ci->i_truncate_pending++; |
634 | queue_trunc = 1; | |
635 | } | |
636 | } | |
637 | } | |
638 | if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 && | |
639 | ci->i_truncate_size != truncate_size) { | |
640 | dout("truncate_size %lld -> %llu\n", ci->i_truncate_size, | |
641 | truncate_size); | |
642 | ci->i_truncate_size = truncate_size; | |
643 | } | |
99ccbd22 MT |
644 | |
645 | if (queue_trunc) | |
646 | ceph_fscache_invalidate(inode); | |
647 | ||
355da1eb SW |
648 | return queue_trunc; |
649 | } | |
650 | ||
651 | void ceph_fill_file_time(struct inode *inode, int issued, | |
652 | u64 time_warp_seq, struct timespec *ctime, | |
653 | struct timespec *mtime, struct timespec *atime) | |
654 | { | |
655 | struct ceph_inode_info *ci = ceph_inode(inode); | |
656 | int warn = 0; | |
657 | ||
658 | if (issued & (CEPH_CAP_FILE_EXCL| | |
659 | CEPH_CAP_FILE_WR| | |
d8672d64 SW |
660 | CEPH_CAP_FILE_BUFFER| |
661 | CEPH_CAP_AUTH_EXCL| | |
662 | CEPH_CAP_XATTR_EXCL)) { | |
fb64b047 YZ |
663 | if (ci->i_version == 0 || |
664 | timespec_compare(ctime, &inode->i_ctime) > 0) { | |
355da1eb SW |
665 | dout("ctime %ld.%09ld -> %ld.%09ld inc w/ cap\n", |
666 | inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec, | |
667 | ctime->tv_sec, ctime->tv_nsec); | |
668 | inode->i_ctime = *ctime; | |
669 | } | |
fb64b047 YZ |
670 | if (ci->i_version == 0 || |
671 | ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) { | |
355da1eb SW |
672 | /* the MDS did a utimes() */ |
673 | dout("mtime %ld.%09ld -> %ld.%09ld " | |
674 | "tw %d -> %d\n", | |
675 | inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec, | |
676 | mtime->tv_sec, mtime->tv_nsec, | |
677 | ci->i_time_warp_seq, (int)time_warp_seq); | |
678 | ||
679 | inode->i_mtime = *mtime; | |
680 | inode->i_atime = *atime; | |
681 | ci->i_time_warp_seq = time_warp_seq; | |
682 | } else if (time_warp_seq == ci->i_time_warp_seq) { | |
683 | /* nobody did utimes(); take the max */ | |
684 | if (timespec_compare(mtime, &inode->i_mtime) > 0) { | |
685 | dout("mtime %ld.%09ld -> %ld.%09ld inc\n", | |
686 | inode->i_mtime.tv_sec, | |
687 | inode->i_mtime.tv_nsec, | |
688 | mtime->tv_sec, mtime->tv_nsec); | |
689 | inode->i_mtime = *mtime; | |
690 | } | |
691 | if (timespec_compare(atime, &inode->i_atime) > 0) { | |
692 | dout("atime %ld.%09ld -> %ld.%09ld inc\n", | |
693 | inode->i_atime.tv_sec, | |
694 | inode->i_atime.tv_nsec, | |
695 | atime->tv_sec, atime->tv_nsec); | |
696 | inode->i_atime = *atime; | |
697 | } | |
698 | } else if (issued & CEPH_CAP_FILE_EXCL) { | |
699 | /* we did a utimes(); ignore mds values */ | |
700 | } else { | |
701 | warn = 1; | |
702 | } | |
703 | } else { | |
d8672d64 | 704 | /* we have no write|excl caps; whatever the MDS says is true */ |
355da1eb SW |
705 | if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) { |
706 | inode->i_ctime = *ctime; | |
707 | inode->i_mtime = *mtime; | |
708 | inode->i_atime = *atime; | |
709 | ci->i_time_warp_seq = time_warp_seq; | |
710 | } else { | |
711 | warn = 1; | |
712 | } | |
713 | } | |
714 | if (warn) /* time_warp_seq shouldn't go backwards */ | |
715 | dout("%p mds time_warp_seq %llu < %u\n", | |
716 | inode, time_warp_seq, ci->i_time_warp_seq); | |
717 | } | |
718 | ||
719 | /* | |
720 | * Populate an inode based on info from mds. May be called on new or | |
721 | * existing inodes. | |
722 | */ | |
01deead0 | 723 | static int fill_inode(struct inode *inode, struct page *locked_page, |
355da1eb SW |
724 | struct ceph_mds_reply_info_in *iinfo, |
725 | struct ceph_mds_reply_dirfrag *dirinfo, | |
726 | struct ceph_mds_session *session, | |
727 | unsigned long ttl_from, int cap_fmode, | |
728 | struct ceph_cap_reservation *caps_reservation) | |
729 | { | |
d9df2783 | 730 | struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc; |
355da1eb SW |
731 | struct ceph_mds_reply_inode *info = iinfo->in; |
732 | struct ceph_inode_info *ci = ceph_inode(inode); | |
f98a128a | 733 | int issued = 0, implemented, new_issued; |
355da1eb | 734 | struct timespec mtime, atime, ctime; |
355da1eb | 735 | struct ceph_buffer *xattr_blob = NULL; |
779fe0fb | 736 | struct ceph_string *pool_ns = NULL; |
d9df2783 | 737 | struct ceph_cap *new_cap = NULL; |
355da1eb | 738 | int err = 0; |
d9df2783 | 739 | bool wake = false; |
f98a128a YZ |
740 | bool queue_trunc = false; |
741 | bool new_version = false; | |
31c542a1 | 742 | bool fill_inline = false; |
355da1eb SW |
743 | |
744 | dout("fill_inode %p ino %llx.%llx v %llu had %llu\n", | |
745 | inode, ceph_vinop(inode), le64_to_cpu(info->version), | |
746 | ci->i_version); | |
747 | ||
d9df2783 YZ |
748 | /* prealloc new cap struct */ |
749 | if (info->cap.caps && ceph_snap(inode) == CEPH_NOSNAP) | |
750 | new_cap = ceph_get_cap(mdsc, caps_reservation); | |
751 | ||
355da1eb SW |
752 | /* |
753 | * prealloc xattr data, if it looks like we'll need it. only | |
754 | * if len > 4 (meaning there are actually xattrs; the first 4 | |
755 | * bytes are the xattr count). | |
756 | */ | |
757 | if (iinfo->xattr_len > 4) { | |
b6c1d5b8 | 758 | xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS); |
355da1eb SW |
759 | if (!xattr_blob) |
760 | pr_err("fill_inode ENOMEM xattr blob %d bytes\n", | |
761 | iinfo->xattr_len); | |
762 | } | |
763 | ||
779fe0fb YZ |
764 | if (iinfo->pool_ns_len > 0) |
765 | pool_ns = ceph_find_or_create_string(iinfo->pool_ns_data, | |
766 | iinfo->pool_ns_len); | |
767 | ||
be655596 | 768 | spin_lock(&ci->i_ceph_lock); |
355da1eb SW |
769 | |
770 | /* | |
771 | * provided version will be odd if inode value is projected, | |
8bd59e01 SW |
772 | * even if stable. skip the update if we have newer stable |
773 | * info (ours>=theirs, e.g. due to racing mds replies), unless | |
774 | * we are getting projected (unstable) info (in which case the | |
775 | * version is odd, and we want ours>theirs). | |
776 | * us them | |
777 | * 2 2 skip | |
778 | * 3 2 skip | |
779 | * 3 3 update | |
355da1eb | 780 | */ |
f98a128a YZ |
781 | if (ci->i_version == 0 || |
782 | ((info->cap.flags & CEPH_CAP_FLAG_AUTH) && | |
783 | le64_to_cpu(info->version) > (ci->i_version & ~1))) | |
784 | new_version = true; | |
785 | ||
355da1eb SW |
786 | issued = __ceph_caps_issued(ci, &implemented); |
787 | issued |= implemented | __ceph_caps_dirty(ci); | |
f98a128a | 788 | new_issued = ~issued & le32_to_cpu(info->cap.caps); |
355da1eb SW |
789 | |
790 | /* update inode */ | |
355da1eb | 791 | inode->i_rdev = le32_to_cpu(info->rdev); |
f98a128a | 792 | inode->i_blkbits = fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1; |
355da1eb | 793 | |
f98a128a YZ |
794 | if ((new_version || (new_issued & CEPH_CAP_AUTH_SHARED)) && |
795 | (issued & CEPH_CAP_AUTH_EXCL) == 0) { | |
355da1eb | 796 | inode->i_mode = le32_to_cpu(info->mode); |
ab871b90 EB |
797 | inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(info->uid)); |
798 | inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(info->gid)); | |
355da1eb | 799 | dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode, |
bd2bae6a EB |
800 | from_kuid(&init_user_ns, inode->i_uid), |
801 | from_kgid(&init_user_ns, inode->i_gid)); | |
355da1eb SW |
802 | } |
803 | ||
f98a128a YZ |
804 | if ((new_version || (new_issued & CEPH_CAP_LINK_SHARED)) && |
805 | (issued & CEPH_CAP_LINK_EXCL) == 0) | |
bfe86848 | 806 | set_nlink(inode, le32_to_cpu(info->nlink)); |
355da1eb | 807 | |
f98a128a YZ |
808 | if (new_version || (new_issued & CEPH_CAP_ANY_RD)) { |
809 | /* be careful with mtime, atime, size */ | |
810 | ceph_decode_timespec(&atime, &info->atime); | |
811 | ceph_decode_timespec(&mtime, &info->mtime); | |
812 | ceph_decode_timespec(&ctime, &info->ctime); | |
813 | ceph_fill_file_time(inode, issued, | |
814 | le32_to_cpu(info->time_warp_seq), | |
815 | &ctime, &mtime, &atime); | |
816 | } | |
817 | ||
818 | if (new_version || | |
819 | (new_issued & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR))) { | |
7627151e | 820 | s64 old_pool = ci->i_layout.pool_id; |
779fe0fb YZ |
821 | struct ceph_string *old_ns; |
822 | ||
7627151e | 823 | ceph_file_layout_from_legacy(&ci->i_layout, &info->layout); |
779fe0fb YZ |
824 | old_ns = rcu_dereference_protected(ci->i_layout.pool_ns, |
825 | lockdep_is_held(&ci->i_ceph_lock)); | |
826 | rcu_assign_pointer(ci->i_layout.pool_ns, pool_ns); | |
827 | ||
828 | if (ci->i_layout.pool_id != old_pool || pool_ns != old_ns) | |
10183a69 | 829 | ci->i_ceph_flags &= ~CEPH_I_POOL_PERM; |
10183a69 | 830 | |
779fe0fb | 831 | pool_ns = old_ns; |
10183a69 | 832 | |
f98a128a YZ |
833 | queue_trunc = ceph_fill_file_size(inode, issued, |
834 | le32_to_cpu(info->truncate_seq), | |
835 | le64_to_cpu(info->truncate_size), | |
836 | le64_to_cpu(info->size)); | |
837 | /* only update max_size on auth cap */ | |
838 | if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) && | |
839 | ci->i_max_size != le64_to_cpu(info->max_size)) { | |
840 | dout("max_size %lld -> %llu\n", ci->i_max_size, | |
841 | le64_to_cpu(info->max_size)); | |
842 | ci->i_max_size = le64_to_cpu(info->max_size); | |
843 | } | |
844 | } | |
355da1eb SW |
845 | |
846 | /* xattrs */ | |
847 | /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */ | |
508b32d8 | 848 | if ((ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL)) && |
355da1eb SW |
849 | le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) { |
850 | if (ci->i_xattrs.blob) | |
851 | ceph_buffer_put(ci->i_xattrs.blob); | |
852 | ci->i_xattrs.blob = xattr_blob; | |
853 | if (xattr_blob) | |
854 | memcpy(ci->i_xattrs.blob->vec.iov_base, | |
855 | iinfo->xattr_data, iinfo->xattr_len); | |
856 | ci->i_xattrs.version = le64_to_cpu(info->xattr_version); | |
7221fe4c | 857 | ceph_forget_all_cached_acls(inode); |
a6424e48 | 858 | xattr_blob = NULL; |
355da1eb SW |
859 | } |
860 | ||
fb64b047 YZ |
861 | /* finally update i_version */ |
862 | ci->i_version = le64_to_cpu(info->version); | |
863 | ||
355da1eb | 864 | inode->i_mapping->a_ops = &ceph_aops; |
355da1eb SW |
865 | |
866 | switch (inode->i_mode & S_IFMT) { | |
867 | case S_IFIFO: | |
868 | case S_IFBLK: | |
869 | case S_IFCHR: | |
870 | case S_IFSOCK: | |
871 | init_special_inode(inode, inode->i_mode, inode->i_rdev); | |
872 | inode->i_op = &ceph_file_iops; | |
873 | break; | |
874 | case S_IFREG: | |
875 | inode->i_op = &ceph_file_iops; | |
876 | inode->i_fop = &ceph_file_fops; | |
877 | break; | |
878 | case S_IFLNK: | |
879 | inode->i_op = &ceph_symlink_iops; | |
880 | if (!ci->i_symlink) { | |
810339ec | 881 | u32 symlen = iinfo->symlink_len; |
355da1eb SW |
882 | char *sym; |
883 | ||
be655596 | 884 | spin_unlock(&ci->i_ceph_lock); |
355da1eb | 885 | |
224a7542 YZ |
886 | if (symlen != i_size_read(inode)) { |
887 | pr_err("fill_inode %llx.%llx BAD symlink " | |
888 | "size %lld\n", ceph_vinop(inode), | |
889 | i_size_read(inode)); | |
890 | i_size_write(inode, symlen); | |
891 | inode->i_blocks = calc_inode_blocks(symlen); | |
892 | } | |
810339ec | 893 | |
355da1eb | 894 | err = -ENOMEM; |
810339ec | 895 | sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS); |
355da1eb SW |
896 | if (!sym) |
897 | goto out; | |
355da1eb | 898 | |
be655596 | 899 | spin_lock(&ci->i_ceph_lock); |
355da1eb SW |
900 | if (!ci->i_symlink) |
901 | ci->i_symlink = sym; | |
902 | else | |
903 | kfree(sym); /* lost a race */ | |
904 | } | |
ac194dcc | 905 | inode->i_link = ci->i_symlink; |
355da1eb SW |
906 | break; |
907 | case S_IFDIR: | |
908 | inode->i_op = &ceph_dir_iops; | |
909 | inode->i_fop = &ceph_dir_fops; | |
910 | ||
14303d20 SW |
911 | ci->i_dir_layout = iinfo->dir_layout; |
912 | ||
355da1eb SW |
913 | ci->i_files = le64_to_cpu(info->files); |
914 | ci->i_subdirs = le64_to_cpu(info->subdirs); | |
915 | ci->i_rbytes = le64_to_cpu(info->rbytes); | |
916 | ci->i_rfiles = le64_to_cpu(info->rfiles); | |
917 | ci->i_rsubdirs = le64_to_cpu(info->rsubdirs); | |
918 | ceph_decode_timespec(&ci->i_rctime, &info->rctime); | |
355da1eb SW |
919 | break; |
920 | default: | |
921 | pr_err("fill_inode %llx.%llx BAD mode 0%o\n", | |
922 | ceph_vinop(inode), inode->i_mode); | |
923 | } | |
924 | ||
355da1eb SW |
925 | /* were we issued a capability? */ |
926 | if (info->cap.caps) { | |
927 | if (ceph_snap(inode) == CEPH_NOSNAP) { | |
2f92b3d0 | 928 | unsigned caps = le32_to_cpu(info->cap.caps); |
355da1eb SW |
929 | ceph_add_cap(inode, session, |
930 | le64_to_cpu(info->cap.cap_id), | |
2f92b3d0 | 931 | cap_fmode, caps, |
355da1eb SW |
932 | le32_to_cpu(info->cap.wanted), |
933 | le32_to_cpu(info->cap.seq), | |
934 | le32_to_cpu(info->cap.mseq), | |
935 | le64_to_cpu(info->cap.realm), | |
d9df2783 | 936 | info->cap.flags, &new_cap); |
2f92b3d0 YZ |
937 | |
938 | /* set dir completion flag? */ | |
939 | if (S_ISDIR(inode->i_mode) && | |
940 | ci->i_files == 0 && ci->i_subdirs == 0 && | |
941 | (caps & CEPH_CAP_FILE_SHARED) && | |
942 | (issued & CEPH_CAP_FILE_EXCL) == 0 && | |
943 | !__ceph_dir_is_complete(ci)) { | |
944 | dout(" marking %p complete (empty)\n", inode); | |
fdd4e158 | 945 | i_size_write(inode, 0); |
2f92b3d0 | 946 | __ceph_dir_set_complete(ci, |
fdd4e158 YZ |
947 | atomic64_read(&ci->i_release_count), |
948 | atomic64_read(&ci->i_ordered_count)); | |
2f92b3d0 YZ |
949 | } |
950 | ||
d9df2783 | 951 | wake = true; |
355da1eb | 952 | } else { |
355da1eb SW |
953 | dout(" %p got snap_caps %s\n", inode, |
954 | ceph_cap_string(le32_to_cpu(info->cap.caps))); | |
955 | ci->i_snap_caps |= le32_to_cpu(info->cap.caps); | |
956 | if (cap_fmode >= 0) | |
957 | __ceph_get_fmode(ci, cap_fmode); | |
355da1eb | 958 | } |
04d000eb | 959 | } else if (cap_fmode >= 0) { |
f3ae1b97 | 960 | pr_warn("mds issued no caps on %llx.%llx\n", |
04d000eb SW |
961 | ceph_vinop(inode)); |
962 | __ceph_get_fmode(ci, cap_fmode); | |
355da1eb | 963 | } |
31c542a1 YZ |
964 | |
965 | if (iinfo->inline_version > 0 && | |
966 | iinfo->inline_version >= ci->i_inline_version) { | |
967 | int cache_caps = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO; | |
968 | ci->i_inline_version = iinfo->inline_version; | |
969 | if (ci->i_inline_version != CEPH_INLINE_NONE && | |
01deead0 YZ |
970 | (locked_page || |
971 | (le32_to_cpu(info->cap.caps) & cache_caps))) | |
31c542a1 YZ |
972 | fill_inline = true; |
973 | } | |
974 | ||
be655596 | 975 | spin_unlock(&ci->i_ceph_lock); |
355da1eb | 976 | |
31c542a1 | 977 | if (fill_inline) |
01deead0 | 978 | ceph_fill_inline_data(inode, locked_page, |
31c542a1 YZ |
979 | iinfo->inline_data, iinfo->inline_len); |
980 | ||
d9df2783 YZ |
981 | if (wake) |
982 | wake_up_all(&ci->i_cap_wq); | |
983 | ||
355da1eb SW |
984 | /* queue truncate if we saw i_size decrease */ |
985 | if (queue_trunc) | |
3c6f6b79 | 986 | ceph_queue_vmtruncate(inode); |
355da1eb SW |
987 | |
988 | /* populate frag tree */ | |
3e7fbe9c YZ |
989 | if (S_ISDIR(inode->i_mode)) |
990 | ceph_fill_fragtree(inode, &info->fragtree, dirinfo); | |
355da1eb SW |
991 | |
992 | /* update delegation info? */ | |
993 | if (dirinfo) | |
994 | ceph_fill_dirfrag(inode, dirinfo); | |
995 | ||
996 | err = 0; | |
355da1eb | 997 | out: |
d9df2783 YZ |
998 | if (new_cap) |
999 | ceph_put_cap(mdsc, new_cap); | |
b6c1d5b8 SW |
1000 | if (xattr_blob) |
1001 | ceph_buffer_put(xattr_blob); | |
779fe0fb | 1002 | ceph_put_string(pool_ns); |
355da1eb SW |
1003 | return err; |
1004 | } | |
1005 | ||
1006 | /* | |
1007 | * caller should hold session s_mutex. | |
1008 | */ | |
1009 | static void update_dentry_lease(struct dentry *dentry, | |
1010 | struct ceph_mds_reply_lease *lease, | |
1011 | struct ceph_mds_session *session, | |
f5d55f03 JL |
1012 | unsigned long from_time, |
1013 | struct ceph_vino *tgt_vino, | |
1014 | struct ceph_vino *dir_vino) | |
355da1eb SW |
1015 | { |
1016 | struct ceph_dentry_info *di = ceph_dentry(dentry); | |
1017 | long unsigned duration = le32_to_cpu(lease->duration_ms); | |
1018 | long unsigned ttl = from_time + (duration * HZ) / 1000; | |
1019 | long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000; | |
1020 | struct inode *dir; | |
481f001f | 1021 | struct ceph_mds_session *old_lease_session = NULL; |
355da1eb | 1022 | |
f5d55f03 JL |
1023 | /* |
1024 | * Make sure dentry's inode matches tgt_vino. NULL tgt_vino means that | |
1025 | * we expect a negative dentry. | |
1026 | */ | |
1027 | if (!tgt_vino && d_really_is_positive(dentry)) | |
1028 | return; | |
1029 | ||
1030 | if (tgt_vino && (d_really_is_negative(dentry) || | |
1031 | !ceph_ino_compare(d_inode(dentry), tgt_vino))) | |
1032 | return; | |
1033 | ||
355da1eb | 1034 | spin_lock(&dentry->d_lock); |
2f90b852 SW |
1035 | dout("update_dentry_lease %p duration %lu ms ttl %lu\n", |
1036 | dentry, duration, ttl); | |
355da1eb | 1037 | |
2b0143b5 | 1038 | dir = d_inode(dentry->d_parent); |
18fc8abd | 1039 | |
f5d55f03 JL |
1040 | /* make sure parent matches dir_vino */ |
1041 | if (!ceph_ino_compare(dir, dir_vino)) | |
1042 | goto out_unlock; | |
1043 | ||
18fc8abd AV |
1044 | /* only track leases on regular dentries */ |
1045 | if (ceph_snap(dir) != CEPH_NOSNAP) | |
1046 | goto out_unlock; | |
1047 | ||
355da1eb SW |
1048 | di->lease_shared_gen = ceph_inode(dir)->i_shared_gen; |
1049 | ||
2f90b852 | 1050 | if (duration == 0) |
355da1eb SW |
1051 | goto out_unlock; |
1052 | ||
1053 | if (di->lease_gen == session->s_cap_gen && | |
9b16f03c | 1054 | time_before(ttl, di->time)) |
355da1eb SW |
1055 | goto out_unlock; /* we already have a newer lease. */ |
1056 | ||
481f001f YZ |
1057 | if (di->lease_session && di->lease_session != session) { |
1058 | old_lease_session = di->lease_session; | |
1059 | di->lease_session = NULL; | |
1060 | } | |
355da1eb SW |
1061 | |
1062 | ceph_dentry_lru_touch(dentry); | |
1063 | ||
1064 | if (!di->lease_session) | |
1065 | di->lease_session = ceph_get_mds_session(session); | |
1066 | di->lease_gen = session->s_cap_gen; | |
1067 | di->lease_seq = le32_to_cpu(lease->seq); | |
1068 | di->lease_renew_after = half_ttl; | |
1069 | di->lease_renew_from = 0; | |
9b16f03c | 1070 | di->time = ttl; |
355da1eb SW |
1071 | out_unlock: |
1072 | spin_unlock(&dentry->d_lock); | |
481f001f YZ |
1073 | if (old_lease_session) |
1074 | ceph_put_mds_session(old_lease_session); | |
355da1eb SW |
1075 | } |
1076 | ||
1077 | /* | |
1078 | * splice a dentry to an inode. | |
1079 | * caller must hold directory i_mutex for this to be safe. | |
355da1eb | 1080 | */ |
f7380af0 | 1081 | static struct dentry *splice_dentry(struct dentry *dn, struct inode *in) |
355da1eb SW |
1082 | { |
1083 | struct dentry *realdn; | |
1084 | ||
2b0143b5 | 1085 | BUG_ON(d_inode(dn)); |
1cd3935b | 1086 | |
355da1eb SW |
1087 | /* dn must be unhashed */ |
1088 | if (!d_unhashed(dn)) | |
1089 | d_drop(dn); | |
41d28bca | 1090 | realdn = d_splice_alias(in, dn); |
355da1eb | 1091 | if (IS_ERR(realdn)) { |
d69ed05a SW |
1092 | pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n", |
1093 | PTR_ERR(realdn), dn, in, ceph_vinop(in)); | |
355da1eb SW |
1094 | dn = realdn; /* note realdn contains the error */ |
1095 | goto out; | |
1096 | } else if (realdn) { | |
1097 | dout("dn %p (%d) spliced with %p (%d) " | |
1098 | "inode %p ino %llx.%llx\n", | |
84d08fa8 AV |
1099 | dn, d_count(dn), |
1100 | realdn, d_count(realdn), | |
2b0143b5 | 1101 | d_inode(realdn), ceph_vinop(d_inode(realdn))); |
355da1eb SW |
1102 | dput(dn); |
1103 | dn = realdn; | |
1104 | } else { | |
1105 | BUG_ON(!ceph_dentry(dn)); | |
355da1eb | 1106 | dout("dn %p attached to %p ino %llx.%llx\n", |
2b0143b5 | 1107 | dn, d_inode(dn), ceph_vinop(d_inode(dn))); |
355da1eb | 1108 | } |
355da1eb SW |
1109 | out: |
1110 | return dn; | |
1111 | } | |
1112 | ||
1113 | /* | |
1114 | * Incorporate results into the local cache. This is either just | |
1115 | * one inode, or a directory, dentry, and possibly linked-to inode (e.g., | |
1116 | * after a lookup). | |
1117 | * | |
1118 | * A reply may contain | |
1119 | * a directory inode along with a dentry. | |
1120 | * and/or a target inode | |
1121 | * | |
1122 | * Called with snap_rwsem (read). | |
1123 | */ | |
f5a03b08 | 1124 | int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req) |
355da1eb | 1125 | { |
f5a03b08 | 1126 | struct ceph_mds_session *session = req->r_session; |
355da1eb SW |
1127 | struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info; |
1128 | struct inode *in = NULL; | |
f5d55f03 | 1129 | struct ceph_vino tvino, dvino; |
3d14c5d2 | 1130 | struct ceph_fs_client *fsc = ceph_sb_to_client(sb); |
355da1eb SW |
1131 | int err = 0; |
1132 | ||
1133 | dout("fill_trace %p is_dentry %d is_target %d\n", req, | |
1134 | rinfo->head->is_dentry, rinfo->head->is_target); | |
1135 | ||
355da1eb SW |
1136 | if (!rinfo->head->is_target && !rinfo->head->is_dentry) { |
1137 | dout("fill_trace reply is empty!\n"); | |
3dd69aab | 1138 | if (rinfo->head->result == 0 && req->r_parent) |
167c9e35 | 1139 | ceph_invalidate_dir_request(req); |
355da1eb SW |
1140 | return 0; |
1141 | } | |
1142 | ||
1143 | if (rinfo->head->is_dentry) { | |
3dd69aab | 1144 | struct inode *dir = req->r_parent; |
5b1daecd | 1145 | |
6c5e50fa | 1146 | if (dir) { |
01deead0 YZ |
1147 | err = fill_inode(dir, NULL, |
1148 | &rinfo->diri, rinfo->dirfrag, | |
6c5e50fa SW |
1149 | session, req->r_request_started, -1, |
1150 | &req->r_caps_reservation); | |
1151 | if (err < 0) | |
19913b4e | 1152 | goto done; |
6c5e50fa SW |
1153 | } else { |
1154 | WARN_ON_ONCE(1); | |
1155 | } | |
19913b4e YZ |
1156 | |
1157 | if (dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME) { | |
1158 | struct qstr dname; | |
1159 | struct dentry *dn, *parent; | |
1160 | ||
1161 | BUG_ON(!rinfo->head->is_target); | |
1162 | BUG_ON(req->r_dentry); | |
1163 | ||
1164 | parent = d_find_any_alias(dir); | |
1165 | BUG_ON(!parent); | |
1166 | ||
1167 | dname.name = rinfo->dname; | |
1168 | dname.len = rinfo->dname_len; | |
8387ff25 | 1169 | dname.hash = full_name_hash(parent, dname.name, dname.len); |
f5d55f03 JL |
1170 | tvino.ino = le64_to_cpu(rinfo->targeti.in->ino); |
1171 | tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid); | |
19913b4e YZ |
1172 | retry_lookup: |
1173 | dn = d_lookup(parent, &dname); | |
1174 | dout("d_lookup on parent=%p name=%.*s got %p\n", | |
1175 | parent, dname.len, dname.name, dn); | |
1176 | ||
1177 | if (!dn) { | |
1178 | dn = d_alloc(parent, &dname); | |
1179 | dout("d_alloc %p '%.*s' = %p\n", parent, | |
1180 | dname.len, dname.name, dn); | |
d37b1d99 | 1181 | if (!dn) { |
19913b4e YZ |
1182 | dput(parent); |
1183 | err = -ENOMEM; | |
1184 | goto done; | |
1185 | } | |
ad5cb123 | 1186 | err = 0; |
2b0143b5 | 1187 | } else if (d_really_is_positive(dn) && |
f5d55f03 JL |
1188 | (ceph_ino(d_inode(dn)) != tvino.ino || |
1189 | ceph_snap(d_inode(dn)) != tvino.snap)) { | |
19913b4e | 1190 | dout(" dn %p points to wrong inode %p\n", |
2b0143b5 | 1191 | dn, d_inode(dn)); |
933ad2c9 | 1192 | ceph_dir_clear_ordered(dir); |
19913b4e YZ |
1193 | d_delete(dn); |
1194 | dput(dn); | |
1195 | goto retry_lookup; | |
1196 | } | |
1197 | ||
1198 | req->r_dentry = dn; | |
1199 | dput(parent); | |
1200 | } | |
5b1daecd SW |
1201 | } |
1202 | ||
86b58d13 | 1203 | if (rinfo->head->is_target) { |
f5d55f03 JL |
1204 | tvino.ino = le64_to_cpu(rinfo->targeti.in->ino); |
1205 | tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid); | |
86b58d13 | 1206 | |
f5d55f03 | 1207 | in = ceph_get_inode(sb, tvino); |
86b58d13 YZ |
1208 | if (IS_ERR(in)) { |
1209 | err = PTR_ERR(in); | |
1210 | goto done; | |
1211 | } | |
1212 | req->r_target_inode = in; | |
1213 | ||
01deead0 | 1214 | err = fill_inode(in, req->r_locked_page, &rinfo->targeti, NULL, |
86b58d13 | 1215 | session, req->r_request_started, |
bc2de10d JL |
1216 | (!test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) && |
1217 | rinfo->head->result == 0) ? req->r_fmode : -1, | |
86b58d13 YZ |
1218 | &req->r_caps_reservation); |
1219 | if (err < 0) { | |
1220 | pr_err("fill_inode badness %p %llx.%llx\n", | |
1221 | in, ceph_vinop(in)); | |
1222 | goto done; | |
1223 | } | |
1224 | } | |
1225 | ||
9358c6d4 SW |
1226 | /* |
1227 | * ignore null lease/binding on snapdir ENOENT, or else we | |
1228 | * will have trouble splicing in the virtual snapdir later | |
1229 | */ | |
3dd69aab JL |
1230 | if (rinfo->head->is_dentry && |
1231 | !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) && | |
1232 | test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) && | |
9358c6d4 | 1233 | (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name, |
3d14c5d2 | 1234 | fsc->mount_options->snapdir_name, |
9358c6d4 | 1235 | req->r_dentry->d_name.len))) { |
355da1eb SW |
1236 | /* |
1237 | * lookup link rename : null -> possibly existing inode | |
1238 | * mknod symlink mkdir : null -> new inode | |
1239 | * unlink : linked -> null | |
1240 | */ | |
3dd69aab | 1241 | struct inode *dir = req->r_parent; |
355da1eb SW |
1242 | struct dentry *dn = req->r_dentry; |
1243 | bool have_dir_cap, have_lease; | |
1244 | ||
1245 | BUG_ON(!dn); | |
1246 | BUG_ON(!dir); | |
2b0143b5 | 1247 | BUG_ON(d_inode(dn->d_parent) != dir); |
f5d55f03 JL |
1248 | |
1249 | dvino.ino = le64_to_cpu(rinfo->diri.in->ino); | |
1250 | dvino.snap = le64_to_cpu(rinfo->diri.in->snapid); | |
1251 | ||
1252 | BUG_ON(ceph_ino(dir) != dvino.ino); | |
1253 | BUG_ON(ceph_snap(dir) != dvino.snap); | |
355da1eb | 1254 | |
355da1eb SW |
1255 | /* do we have a lease on the whole dir? */ |
1256 | have_dir_cap = | |
1257 | (le32_to_cpu(rinfo->diri.in->cap.caps) & | |
1258 | CEPH_CAP_FILE_SHARED); | |
1259 | ||
1260 | /* do we have a dn lease? */ | |
1261 | have_lease = have_dir_cap || | |
2f90b852 | 1262 | le32_to_cpu(rinfo->dlease->duration_ms); |
355da1eb SW |
1263 | if (!have_lease) |
1264 | dout("fill_trace no dentry lease or dir cap\n"); | |
1265 | ||
1266 | /* rename? */ | |
1267 | if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) { | |
0a8a70f9 YZ |
1268 | struct inode *olddir = req->r_old_dentry_dir; |
1269 | BUG_ON(!olddir); | |
1270 | ||
a455589f | 1271 | dout(" src %p '%pd' dst %p '%pd'\n", |
355da1eb | 1272 | req->r_old_dentry, |
a455589f AV |
1273 | req->r_old_dentry, |
1274 | dn, dn); | |
355da1eb SW |
1275 | dout("fill_trace doing d_move %p -> %p\n", |
1276 | req->r_old_dentry, dn); | |
c10f5e12 | 1277 | |
fdd4e158 YZ |
1278 | /* d_move screws up sibling dentries' offsets */ |
1279 | ceph_dir_clear_ordered(dir); | |
1280 | ceph_dir_clear_ordered(olddir); | |
1281 | ||
355da1eb | 1282 | d_move(req->r_old_dentry, dn); |
a455589f AV |
1283 | dout(" src %p '%pd' dst %p '%pd'\n", |
1284 | req->r_old_dentry, | |
355da1eb | 1285 | req->r_old_dentry, |
a455589f | 1286 | dn, dn); |
81a6cf2d | 1287 | |
c4a29f26 SW |
1288 | /* ensure target dentry is invalidated, despite |
1289 | rehashing bug in vfs_rename_dir */ | |
81a6cf2d SW |
1290 | ceph_invalidate_dentry_lease(dn); |
1291 | ||
99ccbd22 | 1292 | dout("dn %p gets new offset %lld\n", req->r_old_dentry, |
1cd3935b | 1293 | ceph_dentry(req->r_old_dentry)->offset); |
81a6cf2d | 1294 | |
355da1eb | 1295 | dn = req->r_old_dentry; /* use old_dentry */ |
355da1eb SW |
1296 | } |
1297 | ||
1298 | /* null dentry? */ | |
1299 | if (!rinfo->head->is_target) { | |
1300 | dout("fill_trace null dentry\n"); | |
2b0143b5 | 1301 | if (d_really_is_positive(dn)) { |
70db4f36 | 1302 | ceph_dir_clear_ordered(dir); |
355da1eb SW |
1303 | dout("d_delete %p\n", dn); |
1304 | d_delete(dn); | |
80d025ff JL |
1305 | } else if (have_lease) { |
1306 | if (d_unhashed(dn)) | |
f8b31710 | 1307 | d_add(dn, NULL); |
355da1eb SW |
1308 | update_dentry_lease(dn, rinfo->dlease, |
1309 | session, | |
f5d55f03 JL |
1310 | req->r_request_started, |
1311 | NULL, &dvino); | |
355da1eb SW |
1312 | } |
1313 | goto done; | |
1314 | } | |
1315 | ||
1316 | /* attach proper inode */ | |
2b0143b5 | 1317 | if (d_really_is_negative(dn)) { |
70db4f36 | 1318 | ceph_dir_clear_ordered(dir); |
86b58d13 | 1319 | ihold(in); |
f7380af0 | 1320 | dn = splice_dentry(dn, in); |
355da1eb SW |
1321 | if (IS_ERR(dn)) { |
1322 | err = PTR_ERR(dn); | |
1323 | goto done; | |
1324 | } | |
1325 | req->r_dentry = dn; /* may have spliced */ | |
2b0143b5 | 1326 | } else if (d_really_is_positive(dn) && d_inode(dn) != in) { |
355da1eb | 1327 | dout(" %p links to %p %llx.%llx, not %llx.%llx\n", |
2b0143b5 | 1328 | dn, d_inode(dn), ceph_vinop(d_inode(dn)), |
86b58d13 | 1329 | ceph_vinop(in)); |
933ad2c9 | 1330 | ceph_dir_clear_ordered(dir); |
200fd27c | 1331 | d_invalidate(dn); |
355da1eb | 1332 | have_lease = false; |
355da1eb SW |
1333 | } |
1334 | ||
f5d55f03 JL |
1335 | if (have_lease) { |
1336 | tvino.ino = le64_to_cpu(rinfo->targeti.in->ino); | |
1337 | tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid); | |
355da1eb | 1338 | update_dentry_lease(dn, rinfo->dlease, session, |
f5d55f03 JL |
1339 | req->r_request_started, |
1340 | &tvino, &dvino); | |
1341 | } | |
355da1eb | 1342 | dout(" final dn %p\n", dn); |
bc2de10d JL |
1343 | } else if ((req->r_op == CEPH_MDS_OP_LOOKUPSNAP || |
1344 | req->r_op == CEPH_MDS_OP_MKSNAP) && | |
1345 | !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) { | |
355da1eb | 1346 | struct dentry *dn = req->r_dentry; |
3dd69aab | 1347 | struct inode *dir = req->r_parent; |
355da1eb SW |
1348 | |
1349 | /* fill out a snapdir LOOKUPSNAP dentry */ | |
1350 | BUG_ON(!dn); | |
0a8a70f9 YZ |
1351 | BUG_ON(!dir); |
1352 | BUG_ON(ceph_snap(dir) != CEPH_SNAPDIR); | |
355da1eb | 1353 | dout(" linking snapped dir %p to dn %p\n", in, dn); |
70db4f36 | 1354 | ceph_dir_clear_ordered(dir); |
86b58d13 | 1355 | ihold(in); |
f7380af0 | 1356 | dn = splice_dentry(dn, in); |
355da1eb SW |
1357 | if (IS_ERR(dn)) { |
1358 | err = PTR_ERR(dn); | |
1359 | goto done; | |
1360 | } | |
1361 | req->r_dentry = dn; /* may have spliced */ | |
cdde7c43 JL |
1362 | } else if (rinfo->head->is_dentry) { |
1363 | struct ceph_vino *ptvino = NULL; | |
1364 | ||
1365 | if ((le32_to_cpu(rinfo->diri.in->cap.caps) & CEPH_CAP_FILE_SHARED) || | |
1366 | le32_to_cpu(rinfo->dlease->duration_ms)) { | |
1367 | dvino.ino = le64_to_cpu(rinfo->diri.in->ino); | |
1368 | dvino.snap = le64_to_cpu(rinfo->diri.in->snapid); | |
1369 | ||
1370 | if (rinfo->head->is_target) { | |
1371 | tvino.ino = le64_to_cpu(rinfo->targeti.in->ino); | |
1372 | tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid); | |
1373 | ptvino = &tvino; | |
1374 | } | |
1375 | ||
1376 | update_dentry_lease(req->r_dentry, rinfo->dlease, | |
1377 | session, req->r_request_started, ptvino, | |
1378 | &dvino); | |
1379 | } else { | |
1380 | dout("%s: no dentry lease or dir cap\n", __func__); | |
1381 | } | |
355da1eb | 1382 | } |
355da1eb SW |
1383 | done: |
1384 | dout("fill_trace done err=%d\n", err); | |
1385 | return err; | |
1386 | } | |
1387 | ||
1388 | /* | |
1389 | * Prepopulate our cache with readdir results, leases, etc. | |
1390 | */ | |
79f9f99a SW |
1391 | static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req, |
1392 | struct ceph_mds_session *session) | |
1393 | { | |
1394 | struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info; | |
1395 | int i, err = 0; | |
1396 | ||
1397 | for (i = 0; i < rinfo->dir_nr; i++) { | |
2a5beea3 | 1398 | struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i; |
79f9f99a SW |
1399 | struct ceph_vino vino; |
1400 | struct inode *in; | |
1401 | int rc; | |
1402 | ||
2a5beea3 YZ |
1403 | vino.ino = le64_to_cpu(rde->inode.in->ino); |
1404 | vino.snap = le64_to_cpu(rde->inode.in->snapid); | |
79f9f99a SW |
1405 | |
1406 | in = ceph_get_inode(req->r_dentry->d_sb, vino); | |
1407 | if (IS_ERR(in)) { | |
1408 | err = PTR_ERR(in); | |
1409 | dout("new_inode badness got %d\n", err); | |
1410 | continue; | |
1411 | } | |
2a5beea3 | 1412 | rc = fill_inode(in, NULL, &rde->inode, NULL, session, |
79f9f99a SW |
1413 | req->r_request_started, -1, |
1414 | &req->r_caps_reservation); | |
1415 | if (rc < 0) { | |
1416 | pr_err("fill_inode badness on %p got %d\n", in, rc); | |
1417 | err = rc; | |
79f9f99a | 1418 | } |
209ae762 | 1419 | iput(in); |
79f9f99a SW |
1420 | } |
1421 | ||
1422 | return err; | |
1423 | } | |
1424 | ||
fdd4e158 YZ |
1425 | void ceph_readdir_cache_release(struct ceph_readdir_cache_control *ctl) |
1426 | { | |
1427 | if (ctl->page) { | |
1428 | kunmap(ctl->page); | |
09cbfeaf | 1429 | put_page(ctl->page); |
fdd4e158 YZ |
1430 | ctl->page = NULL; |
1431 | } | |
1432 | } | |
1433 | ||
1434 | static int fill_readdir_cache(struct inode *dir, struct dentry *dn, | |
1435 | struct ceph_readdir_cache_control *ctl, | |
1436 | struct ceph_mds_request *req) | |
1437 | { | |
1438 | struct ceph_inode_info *ci = ceph_inode(dir); | |
09cbfeaf | 1439 | unsigned nsize = PAGE_SIZE / sizeof(struct dentry*); |
fdd4e158 YZ |
1440 | unsigned idx = ctl->index % nsize; |
1441 | pgoff_t pgoff = ctl->index / nsize; | |
1442 | ||
1443 | if (!ctl->page || pgoff != page_index(ctl->page)) { | |
1444 | ceph_readdir_cache_release(ctl); | |
af5e5eb5 YZ |
1445 | if (idx == 0) |
1446 | ctl->page = grab_cache_page(&dir->i_data, pgoff); | |
1447 | else | |
1448 | ctl->page = find_lock_page(&dir->i_data, pgoff); | |
fdd4e158 YZ |
1449 | if (!ctl->page) { |
1450 | ctl->index = -1; | |
af5e5eb5 | 1451 | return idx == 0 ? -ENOMEM : 0; |
fdd4e158 YZ |
1452 | } |
1453 | /* reading/filling the cache are serialized by | |
1454 | * i_mutex, no need to use page lock */ | |
1455 | unlock_page(ctl->page); | |
1456 | ctl->dentries = kmap(ctl->page); | |
af5e5eb5 | 1457 | if (idx == 0) |
09cbfeaf | 1458 | memset(ctl->dentries, 0, PAGE_SIZE); |
fdd4e158 YZ |
1459 | } |
1460 | ||
1461 | if (req->r_dir_release_cnt == atomic64_read(&ci->i_release_count) && | |
1462 | req->r_dir_ordered_cnt == atomic64_read(&ci->i_ordered_count)) { | |
1463 | dout("readdir cache dn %p idx %d\n", dn, ctl->index); | |
1464 | ctl->dentries[idx] = dn; | |
1465 | ctl->index++; | |
1466 | } else { | |
1467 | dout("disable readdir cache\n"); | |
1468 | ctl->index = -1; | |
1469 | } | |
1470 | return 0; | |
1471 | } | |
1472 | ||
355da1eb SW |
1473 | int ceph_readdir_prepopulate(struct ceph_mds_request *req, |
1474 | struct ceph_mds_session *session) | |
1475 | { | |
1476 | struct dentry *parent = req->r_dentry; | |
f3c4ebe6 | 1477 | struct ceph_inode_info *ci = ceph_inode(d_inode(parent)); |
355da1eb SW |
1478 | struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info; |
1479 | struct qstr dname; | |
1480 | struct dentry *dn; | |
1481 | struct inode *in; | |
315f2408 | 1482 | int err = 0, skipped = 0, ret, i; |
355da1eb | 1483 | struct ceph_mds_request_head *rhead = req->r_request->front.iov_base; |
81c6aea5 | 1484 | u32 frag = le32_to_cpu(rhead->args.readdir.frag); |
f3c4ebe6 YZ |
1485 | u32 last_hash = 0; |
1486 | u32 fpos_offset; | |
fdd4e158 YZ |
1487 | struct ceph_readdir_cache_control cache_ctl = {}; |
1488 | ||
bc2de10d | 1489 | if (test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) |
fdd4e158 | 1490 | return readdir_prepopulate_inodes_only(req, session); |
81c6aea5 | 1491 | |
79162547 YZ |
1492 | if (rinfo->hash_order) { |
1493 | if (req->r_path2) { | |
1494 | last_hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash, | |
1495 | req->r_path2, | |
1496 | strlen(req->r_path2)); | |
1497 | last_hash = ceph_frag_value(last_hash); | |
1498 | } else if (rinfo->offset_hash) { | |
1499 | /* mds understands offset_hash */ | |
1500 | WARN_ON_ONCE(req->r_readdir_offset != 2); | |
1501 | last_hash = le32_to_cpu(rhead->args.readdir.offset_hash); | |
1502 | } | |
f3c4ebe6 YZ |
1503 | } |
1504 | ||
81c6aea5 YZ |
1505 | if (rinfo->dir_dir && |
1506 | le32_to_cpu(rinfo->dir_dir->frag) != frag) { | |
1507 | dout("readdir_prepopulate got new frag %x -> %x\n", | |
1508 | frag, le32_to_cpu(rinfo->dir_dir->frag)); | |
1509 | frag = le32_to_cpu(rinfo->dir_dir->frag); | |
f3c4ebe6 | 1510 | if (!rinfo->hash_order) |
fdd4e158 | 1511 | req->r_readdir_offset = 2; |
81c6aea5 | 1512 | } |
355da1eb SW |
1513 | |
1514 | if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) { | |
355da1eb SW |
1515 | dout("readdir_prepopulate %d items under SNAPDIR dn %p\n", |
1516 | rinfo->dir_nr, parent); | |
1517 | } else { | |
1518 | dout("readdir_prepopulate %d items under dn %p\n", | |
1519 | rinfo->dir_nr, parent); | |
1520 | if (rinfo->dir_dir) | |
2b0143b5 | 1521 | ceph_fill_dirfrag(d_inode(parent), rinfo->dir_dir); |
355da1eb | 1522 | |
8d45b911 YZ |
1523 | if (ceph_frag_is_leftmost(frag) && |
1524 | req->r_readdir_offset == 2 && | |
1525 | !(rinfo->hash_order && last_hash)) { | |
1526 | /* note dir version at start of readdir so we can | |
1527 | * tell if any dentries get dropped */ | |
1528 | req->r_dir_release_cnt = | |
1529 | atomic64_read(&ci->i_release_count); | |
1530 | req->r_dir_ordered_cnt = | |
1531 | atomic64_read(&ci->i_ordered_count); | |
1532 | req->r_readdir_cache_idx = 0; | |
1533 | } | |
fdd4e158 YZ |
1534 | } |
1535 | ||
1536 | cache_ctl.index = req->r_readdir_cache_idx; | |
f3c4ebe6 | 1537 | fpos_offset = req->r_readdir_offset; |
fdd4e158 | 1538 | |
86b58d13 | 1539 | /* FIXME: release caps/leases if error occurs */ |
355da1eb | 1540 | for (i = 0; i < rinfo->dir_nr; i++) { |
2a5beea3 | 1541 | struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i; |
f5d55f03 | 1542 | struct ceph_vino tvino, dvino; |
355da1eb | 1543 | |
2a5beea3 YZ |
1544 | dname.name = rde->name; |
1545 | dname.len = rde->name_len; | |
8387ff25 | 1546 | dname.hash = full_name_hash(parent, dname.name, dname.len); |
355da1eb | 1547 | |
f5d55f03 JL |
1548 | tvino.ino = le64_to_cpu(rde->inode.in->ino); |
1549 | tvino.snap = le64_to_cpu(rde->inode.in->snapid); | |
355da1eb | 1550 | |
f3c4ebe6 YZ |
1551 | if (rinfo->hash_order) { |
1552 | u32 hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash, | |
1553 | rde->name, rde->name_len); | |
1554 | hash = ceph_frag_value(hash); | |
1555 | if (hash != last_hash) | |
1556 | fpos_offset = 2; | |
1557 | last_hash = hash; | |
1558 | rde->offset = ceph_make_fpos(hash, fpos_offset++, true); | |
1559 | } else { | |
1560 | rde->offset = ceph_make_fpos(frag, fpos_offset++, false); | |
1561 | } | |
355da1eb SW |
1562 | |
1563 | retry_lookup: | |
1564 | dn = d_lookup(parent, &dname); | |
1565 | dout("d_lookup on parent=%p name=%.*s got %p\n", | |
1566 | parent, dname.len, dname.name, dn); | |
1567 | ||
1568 | if (!dn) { | |
1569 | dn = d_alloc(parent, &dname); | |
1570 | dout("d_alloc %p '%.*s' = %p\n", parent, | |
1571 | dname.len, dname.name, dn); | |
d37b1d99 | 1572 | if (!dn) { |
355da1eb SW |
1573 | dout("d_alloc badness\n"); |
1574 | err = -ENOMEM; | |
1575 | goto out; | |
1576 | } | |
2b0143b5 | 1577 | } else if (d_really_is_positive(dn) && |
f5d55f03 JL |
1578 | (ceph_ino(d_inode(dn)) != tvino.ino || |
1579 | ceph_snap(d_inode(dn)) != tvino.snap)) { | |
355da1eb | 1580 | dout(" dn %p points to wrong inode %p\n", |
2b0143b5 | 1581 | dn, d_inode(dn)); |
933ad2c9 | 1582 | __ceph_dir_clear_ordered(ci); |
355da1eb SW |
1583 | d_delete(dn); |
1584 | dput(dn); | |
1585 | goto retry_lookup; | |
355da1eb SW |
1586 | } |
1587 | ||
355da1eb | 1588 | /* inode */ |
2b0143b5 DH |
1589 | if (d_really_is_positive(dn)) { |
1590 | in = d_inode(dn); | |
355da1eb | 1591 | } else { |
f5d55f03 | 1592 | in = ceph_get_inode(parent->d_sb, tvino); |
ac1f12ef | 1593 | if (IS_ERR(in)) { |
355da1eb | 1594 | dout("new_inode badness\n"); |
2744c171 | 1595 | d_drop(dn); |
355da1eb | 1596 | dput(dn); |
ac1f12ef | 1597 | err = PTR_ERR(in); |
355da1eb SW |
1598 | goto out; |
1599 | } | |
355da1eb SW |
1600 | } |
1601 | ||
2a5beea3 | 1602 | ret = fill_inode(in, NULL, &rde->inode, NULL, session, |
fdd4e158 YZ |
1603 | req->r_request_started, -1, |
1604 | &req->r_caps_reservation); | |
1605 | if (ret < 0) { | |
355da1eb | 1606 | pr_err("fill_inode badness on %p\n", in); |
933ad2c9 YZ |
1607 | if (d_really_is_positive(dn)) |
1608 | __ceph_dir_clear_ordered(ci); | |
1609 | else | |
86b58d13 YZ |
1610 | iput(in); |
1611 | d_drop(dn); | |
fdd4e158 | 1612 | err = ret; |
d69ed05a | 1613 | goto next_item; |
355da1eb | 1614 | } |
86b58d13 | 1615 | |
2b0143b5 | 1616 | if (d_really_is_negative(dn)) { |
315f2408 YZ |
1617 | struct dentry *realdn; |
1618 | ||
1619 | if (ceph_security_xattr_deadlock(in)) { | |
1620 | dout(" skip splicing dn %p to inode %p" | |
1621 | " (security xattr deadlock)\n", dn, in); | |
1622 | iput(in); | |
1623 | skipped++; | |
1624 | goto next_item; | |
1625 | } | |
1626 | ||
1627 | realdn = splice_dentry(dn, in); | |
5cba372c YZ |
1628 | if (IS_ERR(realdn)) { |
1629 | err = PTR_ERR(realdn); | |
1630 | d_drop(dn); | |
86b58d13 YZ |
1631 | dn = NULL; |
1632 | goto next_item; | |
1633 | } | |
5cba372c | 1634 | dn = realdn; |
86b58d13 YZ |
1635 | } |
1636 | ||
f3c4ebe6 | 1637 | ceph_dentry(dn)->offset = rde->offset; |
86b58d13 | 1638 | |
f5d55f03 | 1639 | dvino = ceph_vino(d_inode(parent)); |
2a5beea3 | 1640 | update_dentry_lease(dn, rde->lease, req->r_session, |
f5d55f03 | 1641 | req->r_request_started, &tvino, &dvino); |
fdd4e158 | 1642 | |
315f2408 | 1643 | if (err == 0 && skipped == 0 && cache_ctl.index >= 0) { |
fdd4e158 YZ |
1644 | ret = fill_readdir_cache(d_inode(parent), dn, |
1645 | &cache_ctl, req); | |
1646 | if (ret < 0) | |
1647 | err = ret; | |
1648 | } | |
d69ed05a SW |
1649 | next_item: |
1650 | if (dn) | |
1651 | dput(dn); | |
355da1eb | 1652 | } |
355da1eb | 1653 | out: |
315f2408 | 1654 | if (err == 0 && skipped == 0) { |
bc2de10d | 1655 | set_bit(CEPH_MDS_R_DID_PREPOPULATE, &req->r_req_flags); |
fdd4e158 YZ |
1656 | req->r_readdir_cache_idx = cache_ctl.index; |
1657 | } | |
1658 | ceph_readdir_cache_release(&cache_ctl); | |
355da1eb SW |
1659 | dout("readdir_prepopulate done\n"); |
1660 | return err; | |
1661 | } | |
1662 | ||
efb0ca76 | 1663 | bool ceph_inode_set_size(struct inode *inode, loff_t size) |
355da1eb SW |
1664 | { |
1665 | struct ceph_inode_info *ci = ceph_inode(inode); | |
efb0ca76 | 1666 | bool ret; |
355da1eb | 1667 | |
be655596 | 1668 | spin_lock(&ci->i_ceph_lock); |
355da1eb | 1669 | dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size); |
99c88e69 | 1670 | i_size_write(inode, size); |
224a7542 | 1671 | inode->i_blocks = calc_inode_blocks(size); |
355da1eb | 1672 | |
efb0ca76 | 1673 | ret = __ceph_should_report_size(ci); |
355da1eb | 1674 | |
be655596 | 1675 | spin_unlock(&ci->i_ceph_lock); |
355da1eb SW |
1676 | return ret; |
1677 | } | |
1678 | ||
1679 | /* | |
1680 | * Write back inode data in a worker thread. (This can't be done | |
1681 | * in the message handler context.) | |
1682 | */ | |
3c6f6b79 SW |
1683 | void ceph_queue_writeback(struct inode *inode) |
1684 | { | |
15a2015f | 1685 | ihold(inode); |
3c6f6b79 SW |
1686 | if (queue_work(ceph_inode_to_client(inode)->wb_wq, |
1687 | &ceph_inode(inode)->i_wb_work)) { | |
2c27c9a5 | 1688 | dout("ceph_queue_writeback %p\n", inode); |
3c6f6b79 | 1689 | } else { |
2c27c9a5 | 1690 | dout("ceph_queue_writeback %p failed\n", inode); |
15a2015f | 1691 | iput(inode); |
3c6f6b79 SW |
1692 | } |
1693 | } | |
1694 | ||
1695 | static void ceph_writeback_work(struct work_struct *work) | |
355da1eb SW |
1696 | { |
1697 | struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info, | |
1698 | i_wb_work); | |
1699 | struct inode *inode = &ci->vfs_inode; | |
1700 | ||
1701 | dout("writeback %p\n", inode); | |
1702 | filemap_fdatawrite(&inode->i_data); | |
1703 | iput(inode); | |
1704 | } | |
1705 | ||
3c6f6b79 SW |
1706 | /* |
1707 | * queue an async invalidation | |
1708 | */ | |
1709 | void ceph_queue_invalidate(struct inode *inode) | |
1710 | { | |
15a2015f | 1711 | ihold(inode); |
3c6f6b79 SW |
1712 | if (queue_work(ceph_inode_to_client(inode)->pg_inv_wq, |
1713 | &ceph_inode(inode)->i_pg_inv_work)) { | |
1714 | dout("ceph_queue_invalidate %p\n", inode); | |
3c6f6b79 SW |
1715 | } else { |
1716 | dout("ceph_queue_invalidate %p failed\n", inode); | |
15a2015f | 1717 | iput(inode); |
3c6f6b79 SW |
1718 | } |
1719 | } | |
1720 | ||
355da1eb SW |
1721 | /* |
1722 | * Invalidate inode pages in a worker thread. (This can't be done | |
1723 | * in the message handler context.) | |
1724 | */ | |
3c6f6b79 | 1725 | static void ceph_invalidate_work(struct work_struct *work) |
355da1eb SW |
1726 | { |
1727 | struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info, | |
1728 | i_pg_inv_work); | |
1729 | struct inode *inode = &ci->vfs_inode; | |
6c93df5d | 1730 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); |
355da1eb SW |
1731 | u32 orig_gen; |
1732 | int check = 0; | |
1733 | ||
b0d7c223 | 1734 | mutex_lock(&ci->i_truncate_mutex); |
6c93df5d | 1735 | |
52953d55 | 1736 | if (READ_ONCE(fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) { |
6c93df5d YZ |
1737 | pr_warn_ratelimited("invalidate_pages %p %lld forced umount\n", |
1738 | inode, ceph_ino(inode)); | |
1739 | mapping_set_error(inode->i_mapping, -EIO); | |
1740 | truncate_pagecache(inode, 0); | |
1741 | mutex_unlock(&ci->i_truncate_mutex); | |
1742 | goto out; | |
1743 | } | |
1744 | ||
be655596 | 1745 | spin_lock(&ci->i_ceph_lock); |
355da1eb SW |
1746 | dout("invalidate_pages %p gen %d revoking %d\n", inode, |
1747 | ci->i_rdcache_gen, ci->i_rdcache_revoking); | |
cd045cb4 | 1748 | if (ci->i_rdcache_revoking != ci->i_rdcache_gen) { |
9563f88c YZ |
1749 | if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE)) |
1750 | check = 1; | |
be655596 | 1751 | spin_unlock(&ci->i_ceph_lock); |
b0d7c223 | 1752 | mutex_unlock(&ci->i_truncate_mutex); |
355da1eb SW |
1753 | goto out; |
1754 | } | |
1755 | orig_gen = ci->i_rdcache_gen; | |
be655596 | 1756 | spin_unlock(&ci->i_ceph_lock); |
355da1eb | 1757 | |
9abd4db7 YZ |
1758 | if (invalidate_inode_pages2(inode->i_mapping) < 0) { |
1759 | pr_err("invalidate_pages %p fails\n", inode); | |
1760 | } | |
355da1eb | 1761 | |
be655596 | 1762 | spin_lock(&ci->i_ceph_lock); |
cd045cb4 SW |
1763 | if (orig_gen == ci->i_rdcache_gen && |
1764 | orig_gen == ci->i_rdcache_revoking) { | |
355da1eb SW |
1765 | dout("invalidate_pages %p gen %d successful\n", inode, |
1766 | ci->i_rdcache_gen); | |
cd045cb4 | 1767 | ci->i_rdcache_revoking--; |
355da1eb SW |
1768 | check = 1; |
1769 | } else { | |
cd045cb4 SW |
1770 | dout("invalidate_pages %p gen %d raced, now %d revoking %d\n", |
1771 | inode, orig_gen, ci->i_rdcache_gen, | |
1772 | ci->i_rdcache_revoking); | |
9563f88c YZ |
1773 | if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE)) |
1774 | check = 1; | |
355da1eb | 1775 | } |
be655596 | 1776 | spin_unlock(&ci->i_ceph_lock); |
b0d7c223 | 1777 | mutex_unlock(&ci->i_truncate_mutex); |
9563f88c | 1778 | out: |
355da1eb SW |
1779 | if (check) |
1780 | ceph_check_caps(ci, 0, NULL); | |
355da1eb SW |
1781 | iput(inode); |
1782 | } | |
1783 | ||
1784 | ||
1785 | /* | |
3f99969f | 1786 | * called by trunc_wq; |
355da1eb SW |
1787 | * |
1788 | * We also truncate in a separate thread as well. | |
1789 | */ | |
3c6f6b79 | 1790 | static void ceph_vmtruncate_work(struct work_struct *work) |
355da1eb SW |
1791 | { |
1792 | struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info, | |
1793 | i_vmtruncate_work); | |
1794 | struct inode *inode = &ci->vfs_inode; | |
1795 | ||
1796 | dout("vmtruncate_work %p\n", inode); | |
b415bf4f | 1797 | __ceph_do_pending_vmtruncate(inode); |
355da1eb SW |
1798 | iput(inode); |
1799 | } | |
1800 | ||
3c6f6b79 SW |
1801 | /* |
1802 | * Queue an async vmtruncate. If we fail to queue work, we will handle | |
1803 | * the truncation the next time we call __ceph_do_pending_vmtruncate. | |
1804 | */ | |
1805 | void ceph_queue_vmtruncate(struct inode *inode) | |
1806 | { | |
1807 | struct ceph_inode_info *ci = ceph_inode(inode); | |
1808 | ||
15a2015f | 1809 | ihold(inode); |
99ccbd22 | 1810 | |
640ef79d | 1811 | if (queue_work(ceph_sb_to_client(inode->i_sb)->trunc_wq, |
3c6f6b79 SW |
1812 | &ci->i_vmtruncate_work)) { |
1813 | dout("ceph_queue_vmtruncate %p\n", inode); | |
3c6f6b79 SW |
1814 | } else { |
1815 | dout("ceph_queue_vmtruncate %p failed, pending=%d\n", | |
1816 | inode, ci->i_truncate_pending); | |
15a2015f | 1817 | iput(inode); |
3c6f6b79 SW |
1818 | } |
1819 | } | |
1820 | ||
355da1eb | 1821 | /* |
355da1eb SW |
1822 | * Make sure any pending truncation is applied before doing anything |
1823 | * that may depend on it. | |
1824 | */ | |
b415bf4f | 1825 | void __ceph_do_pending_vmtruncate(struct inode *inode) |
355da1eb SW |
1826 | { |
1827 | struct ceph_inode_info *ci = ceph_inode(inode); | |
1828 | u64 to; | |
a85f50b6 | 1829 | int wrbuffer_refs, finish = 0; |
355da1eb | 1830 | |
b0d7c223 | 1831 | mutex_lock(&ci->i_truncate_mutex); |
355da1eb | 1832 | retry: |
be655596 | 1833 | spin_lock(&ci->i_ceph_lock); |
355da1eb SW |
1834 | if (ci->i_truncate_pending == 0) { |
1835 | dout("__do_pending_vmtruncate %p none pending\n", inode); | |
be655596 | 1836 | spin_unlock(&ci->i_ceph_lock); |
b0d7c223 | 1837 | mutex_unlock(&ci->i_truncate_mutex); |
355da1eb SW |
1838 | return; |
1839 | } | |
1840 | ||
1841 | /* | |
1842 | * make sure any dirty snapped pages are flushed before we | |
1843 | * possibly truncate them.. so write AND block! | |
1844 | */ | |
1845 | if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) { | |
c8fd0d37 YZ |
1846 | struct ceph_cap_snap *capsnap; |
1847 | to = ci->i_truncate_size; | |
1848 | list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) { | |
1849 | // MDS should have revoked Frw caps | |
1850 | WARN_ON_ONCE(capsnap->writing); | |
1851 | if (capsnap->dirty_pages && capsnap->size > to) | |
1852 | to = capsnap->size; | |
1853 | } | |
1854 | spin_unlock(&ci->i_ceph_lock); | |
355da1eb SW |
1855 | dout("__do_pending_vmtruncate %p flushing snaps first\n", |
1856 | inode); | |
c8fd0d37 YZ |
1857 | |
1858 | truncate_pagecache(inode, to); | |
1859 | ||
355da1eb SW |
1860 | filemap_write_and_wait_range(&inode->i_data, 0, |
1861 | inode->i_sb->s_maxbytes); | |
1862 | goto retry; | |
1863 | } | |
1864 | ||
b0d7c223 YZ |
1865 | /* there should be no reader or writer */ |
1866 | WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref); | |
1867 | ||
355da1eb SW |
1868 | to = ci->i_truncate_size; |
1869 | wrbuffer_refs = ci->i_wrbuffer_ref; | |
1870 | dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode, | |
1871 | ci->i_truncate_pending, to); | |
be655596 | 1872 | spin_unlock(&ci->i_ceph_lock); |
355da1eb | 1873 | |
4e217b5d | 1874 | truncate_pagecache(inode, to); |
355da1eb | 1875 | |
be655596 | 1876 | spin_lock(&ci->i_ceph_lock); |
a85f50b6 YZ |
1877 | if (to == ci->i_truncate_size) { |
1878 | ci->i_truncate_pending = 0; | |
1879 | finish = 1; | |
1880 | } | |
be655596 | 1881 | spin_unlock(&ci->i_ceph_lock); |
a85f50b6 YZ |
1882 | if (!finish) |
1883 | goto retry; | |
355da1eb | 1884 | |
b0d7c223 YZ |
1885 | mutex_unlock(&ci->i_truncate_mutex); |
1886 | ||
355da1eb SW |
1887 | if (wrbuffer_refs == 0) |
1888 | ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL); | |
a85f50b6 YZ |
1889 | |
1890 | wake_up_all(&ci->i_cap_wq); | |
355da1eb SW |
1891 | } |
1892 | ||
355da1eb SW |
1893 | /* |
1894 | * symlinks | |
1895 | */ | |
355da1eb | 1896 | static const struct inode_operations ceph_symlink_iops = { |
6b255391 | 1897 | .get_link = simple_get_link, |
0b932672 YZ |
1898 | .setattr = ceph_setattr, |
1899 | .getattr = ceph_getattr, | |
0b932672 | 1900 | .listxattr = ceph_listxattr, |
355da1eb SW |
1901 | }; |
1902 | ||
a26fecca | 1903 | int __ceph_setattr(struct inode *inode, struct iattr *attr) |
355da1eb | 1904 | { |
355da1eb | 1905 | struct ceph_inode_info *ci = ceph_inode(inode); |
355da1eb SW |
1906 | const unsigned int ia_valid = attr->ia_valid; |
1907 | struct ceph_mds_request *req; | |
a26fecca | 1908 | struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc; |
f66fd9f0 | 1909 | struct ceph_cap_flush *prealloc_cf; |
355da1eb SW |
1910 | int issued; |
1911 | int release = 0, dirtied = 0; | |
1912 | int mask = 0; | |
1913 | int err = 0; | |
fca65b4a | 1914 | int inode_dirty_flags = 0; |
604d1b02 | 1915 | bool lock_snap_rwsem = false; |
355da1eb | 1916 | |
f66fd9f0 YZ |
1917 | prealloc_cf = ceph_alloc_cap_flush(); |
1918 | if (!prealloc_cf) | |
1919 | return -ENOMEM; | |
1920 | ||
355da1eb SW |
1921 | req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR, |
1922 | USE_AUTH_MDS); | |
f66fd9f0 YZ |
1923 | if (IS_ERR(req)) { |
1924 | ceph_free_cap_flush(prealloc_cf); | |
355da1eb | 1925 | return PTR_ERR(req); |
f66fd9f0 | 1926 | } |
355da1eb | 1927 | |
be655596 | 1928 | spin_lock(&ci->i_ceph_lock); |
355da1eb | 1929 | issued = __ceph_caps_issued(ci, NULL); |
604d1b02 YZ |
1930 | |
1931 | if (!ci->i_head_snapc && | |
1932 | (issued & (CEPH_CAP_ANY_EXCL | CEPH_CAP_FILE_WR))) { | |
1933 | lock_snap_rwsem = true; | |
1934 | if (!down_read_trylock(&mdsc->snap_rwsem)) { | |
1935 | spin_unlock(&ci->i_ceph_lock); | |
1936 | down_read(&mdsc->snap_rwsem); | |
1937 | spin_lock(&ci->i_ceph_lock); | |
1938 | issued = __ceph_caps_issued(ci, NULL); | |
1939 | } | |
1940 | } | |
1941 | ||
355da1eb SW |
1942 | dout("setattr %p issued %s\n", inode, ceph_cap_string(issued)); |
1943 | ||
1944 | if (ia_valid & ATTR_UID) { | |
1945 | dout("setattr %p uid %d -> %d\n", inode, | |
bd2bae6a EB |
1946 | from_kuid(&init_user_ns, inode->i_uid), |
1947 | from_kuid(&init_user_ns, attr->ia_uid)); | |
355da1eb SW |
1948 | if (issued & CEPH_CAP_AUTH_EXCL) { |
1949 | inode->i_uid = attr->ia_uid; | |
1950 | dirtied |= CEPH_CAP_AUTH_EXCL; | |
1951 | } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || | |
ab871b90 EB |
1952 | !uid_eq(attr->ia_uid, inode->i_uid)) { |
1953 | req->r_args.setattr.uid = cpu_to_le32( | |
1954 | from_kuid(&init_user_ns, attr->ia_uid)); | |
355da1eb SW |
1955 | mask |= CEPH_SETATTR_UID; |
1956 | release |= CEPH_CAP_AUTH_SHARED; | |
1957 | } | |
1958 | } | |
1959 | if (ia_valid & ATTR_GID) { | |
1960 | dout("setattr %p gid %d -> %d\n", inode, | |
bd2bae6a EB |
1961 | from_kgid(&init_user_ns, inode->i_gid), |
1962 | from_kgid(&init_user_ns, attr->ia_gid)); | |
355da1eb SW |
1963 | if (issued & CEPH_CAP_AUTH_EXCL) { |
1964 | inode->i_gid = attr->ia_gid; | |
1965 | dirtied |= CEPH_CAP_AUTH_EXCL; | |
1966 | } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || | |
ab871b90 EB |
1967 | !gid_eq(attr->ia_gid, inode->i_gid)) { |
1968 | req->r_args.setattr.gid = cpu_to_le32( | |
1969 | from_kgid(&init_user_ns, attr->ia_gid)); | |
355da1eb SW |
1970 | mask |= CEPH_SETATTR_GID; |
1971 | release |= CEPH_CAP_AUTH_SHARED; | |
1972 | } | |
1973 | } | |
1974 | if (ia_valid & ATTR_MODE) { | |
1975 | dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode, | |
1976 | attr->ia_mode); | |
1977 | if (issued & CEPH_CAP_AUTH_EXCL) { | |
1978 | inode->i_mode = attr->ia_mode; | |
1979 | dirtied |= CEPH_CAP_AUTH_EXCL; | |
1980 | } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || | |
1981 | attr->ia_mode != inode->i_mode) { | |
7221fe4c | 1982 | inode->i_mode = attr->ia_mode; |
355da1eb SW |
1983 | req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode); |
1984 | mask |= CEPH_SETATTR_MODE; | |
1985 | release |= CEPH_CAP_AUTH_SHARED; | |
1986 | } | |
1987 | } | |
1988 | ||
1989 | if (ia_valid & ATTR_ATIME) { | |
1990 | dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode, | |
1991 | inode->i_atime.tv_sec, inode->i_atime.tv_nsec, | |
1992 | attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec); | |
1993 | if (issued & CEPH_CAP_FILE_EXCL) { | |
1994 | ci->i_time_warp_seq++; | |
1995 | inode->i_atime = attr->ia_atime; | |
1996 | dirtied |= CEPH_CAP_FILE_EXCL; | |
1997 | } else if ((issued & CEPH_CAP_FILE_WR) && | |
1998 | timespec_compare(&inode->i_atime, | |
1999 | &attr->ia_atime) < 0) { | |
2000 | inode->i_atime = attr->ia_atime; | |
2001 | dirtied |= CEPH_CAP_FILE_WR; | |
2002 | } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || | |
2003 | !timespec_equal(&inode->i_atime, &attr->ia_atime)) { | |
2004 | ceph_encode_timespec(&req->r_args.setattr.atime, | |
2005 | &attr->ia_atime); | |
2006 | mask |= CEPH_SETATTR_ATIME; | |
2007 | release |= CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_RD | | |
2008 | CEPH_CAP_FILE_WR; | |
2009 | } | |
2010 | } | |
2011 | if (ia_valid & ATTR_MTIME) { | |
2012 | dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode, | |
2013 | inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec, | |
2014 | attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec); | |
2015 | if (issued & CEPH_CAP_FILE_EXCL) { | |
2016 | ci->i_time_warp_seq++; | |
2017 | inode->i_mtime = attr->ia_mtime; | |
2018 | dirtied |= CEPH_CAP_FILE_EXCL; | |
2019 | } else if ((issued & CEPH_CAP_FILE_WR) && | |
2020 | timespec_compare(&inode->i_mtime, | |
2021 | &attr->ia_mtime) < 0) { | |
2022 | inode->i_mtime = attr->ia_mtime; | |
2023 | dirtied |= CEPH_CAP_FILE_WR; | |
2024 | } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || | |
2025 | !timespec_equal(&inode->i_mtime, &attr->ia_mtime)) { | |
2026 | ceph_encode_timespec(&req->r_args.setattr.mtime, | |
2027 | &attr->ia_mtime); | |
2028 | mask |= CEPH_SETATTR_MTIME; | |
2029 | release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD | | |
2030 | CEPH_CAP_FILE_WR; | |
2031 | } | |
2032 | } | |
2033 | if (ia_valid & ATTR_SIZE) { | |
2034 | dout("setattr %p size %lld -> %lld\n", inode, | |
2035 | inode->i_size, attr->ia_size); | |
355da1eb SW |
2036 | if ((issued & CEPH_CAP_FILE_EXCL) && |
2037 | attr->ia_size > inode->i_size) { | |
99c88e69 | 2038 | i_size_write(inode, attr->ia_size); |
224a7542 | 2039 | inode->i_blocks = calc_inode_blocks(attr->ia_size); |
355da1eb SW |
2040 | ci->i_reported_size = attr->ia_size; |
2041 | dirtied |= CEPH_CAP_FILE_EXCL; | |
2042 | } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || | |
2043 | attr->ia_size != inode->i_size) { | |
2044 | req->r_args.setattr.size = cpu_to_le64(attr->ia_size); | |
2045 | req->r_args.setattr.old_size = | |
2046 | cpu_to_le64(inode->i_size); | |
2047 | mask |= CEPH_SETATTR_SIZE; | |
2048 | release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD | | |
2049 | CEPH_CAP_FILE_WR; | |
2050 | } | |
2051 | } | |
2052 | ||
2053 | /* these do nothing */ | |
2054 | if (ia_valid & ATTR_CTIME) { | |
2055 | bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME| | |
2056 | ATTR_MODE|ATTR_UID|ATTR_GID)) == 0; | |
2057 | dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode, | |
2058 | inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec, | |
2059 | attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec, | |
2060 | only ? "ctime only" : "ignored"); | |
355da1eb SW |
2061 | if (only) { |
2062 | /* | |
2063 | * if kernel wants to dirty ctime but nothing else, | |
2064 | * we need to choose a cap to dirty under, or do | |
2065 | * a almost-no-op setattr | |
2066 | */ | |
2067 | if (issued & CEPH_CAP_AUTH_EXCL) | |
2068 | dirtied |= CEPH_CAP_AUTH_EXCL; | |
2069 | else if (issued & CEPH_CAP_FILE_EXCL) | |
2070 | dirtied |= CEPH_CAP_FILE_EXCL; | |
2071 | else if (issued & CEPH_CAP_XATTR_EXCL) | |
2072 | dirtied |= CEPH_CAP_XATTR_EXCL; | |
2073 | else | |
2074 | mask |= CEPH_SETATTR_CTIME; | |
2075 | } | |
2076 | } | |
2077 | if (ia_valid & ATTR_FILE) | |
2078 | dout("setattr %p ATTR_FILE ... hrm!\n", inode); | |
2079 | ||
2080 | if (dirtied) { | |
f66fd9f0 YZ |
2081 | inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied, |
2082 | &prealloc_cf); | |
4ca2fea6 | 2083 | inode->i_ctime = attr->ia_ctime; |
355da1eb SW |
2084 | } |
2085 | ||
2086 | release &= issued; | |
be655596 | 2087 | spin_unlock(&ci->i_ceph_lock); |
604d1b02 YZ |
2088 | if (lock_snap_rwsem) |
2089 | up_read(&mdsc->snap_rwsem); | |
355da1eb | 2090 | |
fca65b4a SW |
2091 | if (inode_dirty_flags) |
2092 | __mark_inode_dirty(inode, inode_dirty_flags); | |
2093 | ||
7221fe4c | 2094 | |
355da1eb | 2095 | if (mask) { |
70b666c3 SW |
2096 | req->r_inode = inode; |
2097 | ihold(inode); | |
355da1eb SW |
2098 | req->r_inode_drop = release; |
2099 | req->r_args.setattr.mask = cpu_to_le32(mask); | |
2100 | req->r_num_caps = 1; | |
4ca2fea6 | 2101 | req->r_stamp = attr->ia_ctime; |
752c8bdc | 2102 | err = ceph_mdsc_do_request(mdsc, NULL, req); |
355da1eb SW |
2103 | } |
2104 | dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err, | |
2105 | ceph_cap_string(dirtied), mask); | |
2106 | ||
355da1eb | 2107 | ceph_mdsc_put_request(req); |
f66fd9f0 | 2108 | ceph_free_cap_flush(prealloc_cf); |
8179a101 YZ |
2109 | |
2110 | if (err >= 0 && (mask & CEPH_SETATTR_SIZE)) | |
2111 | __ceph_do_pending_vmtruncate(inode); | |
2112 | ||
355da1eb SW |
2113 | return err; |
2114 | } | |
2115 | ||
a26fecca AG |
2116 | /* |
2117 | * setattr | |
2118 | */ | |
2119 | int ceph_setattr(struct dentry *dentry, struct iattr *attr) | |
2120 | { | |
fd5472ed JK |
2121 | struct inode *inode = d_inode(dentry); |
2122 | int err; | |
2123 | ||
2124 | if (ceph_snap(inode) != CEPH_NOSNAP) | |
2125 | return -EROFS; | |
2126 | ||
31051c85 | 2127 | err = setattr_prepare(dentry, attr); |
fd5472ed JK |
2128 | if (err != 0) |
2129 | return err; | |
2130 | ||
8179a101 YZ |
2131 | err = __ceph_setattr(inode, attr); |
2132 | ||
2133 | if (err >= 0 && (attr->ia_valid & ATTR_MODE)) | |
2134 | err = posix_acl_chmod(inode, attr->ia_mode); | |
2135 | ||
2136 | return err; | |
a26fecca AG |
2137 | } |
2138 | ||
355da1eb SW |
2139 | /* |
2140 | * Verify that we have a lease on the given mask. If not, | |
2141 | * do a getattr against an mds. | |
2142 | */ | |
01deead0 YZ |
2143 | int __ceph_do_getattr(struct inode *inode, struct page *locked_page, |
2144 | int mask, bool force) | |
355da1eb | 2145 | { |
3d14c5d2 YS |
2146 | struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb); |
2147 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
355da1eb SW |
2148 | struct ceph_mds_request *req; |
2149 | int err; | |
2150 | ||
2151 | if (ceph_snap(inode) == CEPH_SNAPDIR) { | |
2152 | dout("do_getattr inode %p SNAPDIR\n", inode); | |
2153 | return 0; | |
2154 | } | |
2155 | ||
01deead0 YZ |
2156 | dout("do_getattr inode %p mask %s mode 0%o\n", |
2157 | inode, ceph_cap_string(mask), inode->i_mode); | |
508b32d8 | 2158 | if (!force && ceph_caps_issued_mask(ceph_inode(inode), mask, 1)) |
355da1eb SW |
2159 | return 0; |
2160 | ||
2161 | req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS); | |
2162 | if (IS_ERR(req)) | |
2163 | return PTR_ERR(req); | |
70b666c3 SW |
2164 | req->r_inode = inode; |
2165 | ihold(inode); | |
355da1eb SW |
2166 | req->r_num_caps = 1; |
2167 | req->r_args.getattr.mask = cpu_to_le32(mask); | |
01deead0 | 2168 | req->r_locked_page = locked_page; |
355da1eb | 2169 | err = ceph_mdsc_do_request(mdsc, NULL, req); |
01deead0 YZ |
2170 | if (locked_page && err == 0) { |
2171 | u64 inline_version = req->r_reply_info.targeti.inline_version; | |
2172 | if (inline_version == 0) { | |
2173 | /* the reply is supposed to contain inline data */ | |
2174 | err = -EINVAL; | |
2175 | } else if (inline_version == CEPH_INLINE_NONE) { | |
2176 | err = -ENODATA; | |
2177 | } else { | |
2178 | err = req->r_reply_info.targeti.inline_len; | |
2179 | } | |
2180 | } | |
355da1eb SW |
2181 | ceph_mdsc_put_request(req); |
2182 | dout("do_getattr result=%d\n", err); | |
2183 | return err; | |
2184 | } | |
2185 | ||
2186 | ||
2187 | /* | |
2188 | * Check inode permissions. We verify we have a valid value for | |
2189 | * the AUTH cap, then call the generic handler. | |
2190 | */ | |
10556cb2 | 2191 | int ceph_permission(struct inode *inode, int mask) |
355da1eb | 2192 | { |
b74c79e9 NP |
2193 | int err; |
2194 | ||
10556cb2 | 2195 | if (mask & MAY_NOT_BLOCK) |
b74c79e9 NP |
2196 | return -ECHILD; |
2197 | ||
508b32d8 | 2198 | err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED, false); |
355da1eb SW |
2199 | |
2200 | if (!err) | |
2830ba7f | 2201 | err = generic_permission(inode, mask); |
355da1eb SW |
2202 | return err; |
2203 | } | |
2204 | ||
2205 | /* | |
2206 | * Get all attributes. Hopefully somedata we'll have a statlite() | |
2207 | * and can limit the fields we require to be accurate. | |
2208 | */ | |
a528d35e DH |
2209 | int ceph_getattr(const struct path *path, struct kstat *stat, |
2210 | u32 request_mask, unsigned int flags) | |
355da1eb | 2211 | { |
a528d35e | 2212 | struct inode *inode = d_inode(path->dentry); |
232d4b01 | 2213 | struct ceph_inode_info *ci = ceph_inode(inode); |
355da1eb SW |
2214 | int err; |
2215 | ||
508b32d8 | 2216 | err = ceph_do_getattr(inode, CEPH_STAT_CAP_INODE_ALL, false); |
355da1eb SW |
2217 | if (!err) { |
2218 | generic_fillattr(inode, stat); | |
ad1fee96 | 2219 | stat->ino = ceph_translate_ino(inode->i_sb, inode->i_ino); |
355da1eb SW |
2220 | if (ceph_snap(inode) != CEPH_NOSNAP) |
2221 | stat->dev = ceph_snap(inode); | |
2222 | else | |
2223 | stat->dev = 0; | |
232d4b01 | 2224 | if (S_ISDIR(inode->i_mode)) { |
1c1266bb YS |
2225 | if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb), |
2226 | RBYTES)) | |
2227 | stat->size = ci->i_rbytes; | |
2228 | else | |
2229 | stat->size = ci->i_files + ci->i_subdirs; | |
232d4b01 | 2230 | stat->blocks = 0; |
355da1eb | 2231 | stat->blksize = 65536; |
232d4b01 | 2232 | } |
355da1eb SW |
2233 | } |
2234 | return err; | |
2235 | } |