Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
3d14c5d2 | 2 | #include <linux/ceph/ceph_debug.h> |
a8599bd8 SW |
3 | |
4 | #include <linux/fs.h> | |
5 | #include <linux/kernel.h> | |
174cd4b1 | 6 | #include <linux/sched/signal.h> |
5a0e3ad6 | 7 | #include <linux/slab.h> |
a8599bd8 SW |
8 | #include <linux/vmalloc.h> |
9 | #include <linux/wait.h> | |
f1a3d572 | 10 | #include <linux/writeback.h> |
176c77c9 | 11 | #include <linux/iversion.h> |
a8599bd8 SW |
12 | |
13 | #include "super.h" | |
3d14c5d2 | 14 | #include "mds_client.h" |
99ccbd22 | 15 | #include "cache.h" |
3d14c5d2 YS |
16 | #include <linux/ceph/decode.h> |
17 | #include <linux/ceph/messenger.h> | |
a8599bd8 SW |
18 | |
19 | /* | |
20 | * Capability management | |
21 | * | |
22 | * The Ceph metadata servers control client access to inode metadata | |
23 | * and file data by issuing capabilities, granting clients permission | |
24 | * to read and/or write both inode field and file data to OSDs | |
25 | * (storage nodes). Each capability consists of a set of bits | |
26 | * indicating which operations are allowed. | |
27 | * | |
28 | * If the client holds a *_SHARED cap, the client has a coherent value | |
29 | * that can be safely read from the cached inode. | |
30 | * | |
31 | * In the case of a *_EXCL (exclusive) or FILE_WR capabilities, the | |
32 | * client is allowed to change inode attributes (e.g., file size, | |
33 | * mtime), note its dirty state in the ceph_cap, and asynchronously | |
34 | * flush that metadata change to the MDS. | |
35 | * | |
36 | * In the event of a conflicting operation (perhaps by another | |
37 | * client), the MDS will revoke the conflicting client capabilities. | |
38 | * | |
39 | * In order for a client to cache an inode, it must hold a capability | |
40 | * with at least one MDS server. When inodes are released, release | |
41 | * notifications are batched and periodically sent en masse to the MDS | |
42 | * cluster to release server state. | |
43 | */ | |
44 | ||
0e294387 | 45 | static u64 __get_oldest_flush_tid(struct ceph_mds_client *mdsc); |
7bc00fdd YZ |
46 | static void __kick_flushing_caps(struct ceph_mds_client *mdsc, |
47 | struct ceph_mds_session *session, | |
48 | struct ceph_inode_info *ci, | |
49 | u64 oldest_flush_tid); | |
a8599bd8 SW |
50 | |
51 | /* | |
52 | * Generate readable cap strings for debugging output. | |
53 | */ | |
54 | #define MAX_CAP_STR 20 | |
55 | static char cap_str[MAX_CAP_STR][40]; | |
56 | static DEFINE_SPINLOCK(cap_str_lock); | |
57 | static int last_cap_str; | |
58 | ||
59 | static char *gcap_string(char *s, int c) | |
60 | { | |
61 | if (c & CEPH_CAP_GSHARED) | |
62 | *s++ = 's'; | |
63 | if (c & CEPH_CAP_GEXCL) | |
64 | *s++ = 'x'; | |
65 | if (c & CEPH_CAP_GCACHE) | |
66 | *s++ = 'c'; | |
67 | if (c & CEPH_CAP_GRD) | |
68 | *s++ = 'r'; | |
69 | if (c & CEPH_CAP_GWR) | |
70 | *s++ = 'w'; | |
71 | if (c & CEPH_CAP_GBUFFER) | |
72 | *s++ = 'b'; | |
49a9f4f6 YZ |
73 | if (c & CEPH_CAP_GWREXTEND) |
74 | *s++ = 'a'; | |
a8599bd8 SW |
75 | if (c & CEPH_CAP_GLAZYIO) |
76 | *s++ = 'l'; | |
77 | return s; | |
78 | } | |
79 | ||
80 | const char *ceph_cap_string(int caps) | |
81 | { | |
82 | int i; | |
83 | char *s; | |
84 | int c; | |
85 | ||
86 | spin_lock(&cap_str_lock); | |
87 | i = last_cap_str++; | |
88 | if (last_cap_str == MAX_CAP_STR) | |
89 | last_cap_str = 0; | |
90 | spin_unlock(&cap_str_lock); | |
91 | ||
92 | s = cap_str[i]; | |
93 | ||
94 | if (caps & CEPH_CAP_PIN) | |
95 | *s++ = 'p'; | |
96 | ||
97 | c = (caps >> CEPH_CAP_SAUTH) & 3; | |
98 | if (c) { | |
99 | *s++ = 'A'; | |
100 | s = gcap_string(s, c); | |
101 | } | |
102 | ||
103 | c = (caps >> CEPH_CAP_SLINK) & 3; | |
104 | if (c) { | |
105 | *s++ = 'L'; | |
106 | s = gcap_string(s, c); | |
107 | } | |
108 | ||
109 | c = (caps >> CEPH_CAP_SXATTR) & 3; | |
110 | if (c) { | |
111 | *s++ = 'X'; | |
112 | s = gcap_string(s, c); | |
113 | } | |
114 | ||
115 | c = caps >> CEPH_CAP_SFILE; | |
116 | if (c) { | |
117 | *s++ = 'F'; | |
118 | s = gcap_string(s, c); | |
119 | } | |
120 | ||
121 | if (s == cap_str[i]) | |
122 | *s++ = '-'; | |
123 | *s = 0; | |
124 | return cap_str[i]; | |
125 | } | |
126 | ||
37151668 | 127 | void ceph_caps_init(struct ceph_mds_client *mdsc) |
a8599bd8 | 128 | { |
37151668 YS |
129 | INIT_LIST_HEAD(&mdsc->caps_list); |
130 | spin_lock_init(&mdsc->caps_list_lock); | |
a8599bd8 SW |
131 | } |
132 | ||
37151668 | 133 | void ceph_caps_finalize(struct ceph_mds_client *mdsc) |
a8599bd8 SW |
134 | { |
135 | struct ceph_cap *cap; | |
136 | ||
37151668 YS |
137 | spin_lock(&mdsc->caps_list_lock); |
138 | while (!list_empty(&mdsc->caps_list)) { | |
139 | cap = list_first_entry(&mdsc->caps_list, | |
140 | struct ceph_cap, caps_item); | |
a8599bd8 SW |
141 | list_del(&cap->caps_item); |
142 | kmem_cache_free(ceph_cap_cachep, cap); | |
143 | } | |
37151668 YS |
144 | mdsc->caps_total_count = 0; |
145 | mdsc->caps_avail_count = 0; | |
146 | mdsc->caps_use_count = 0; | |
147 | mdsc->caps_reserve_count = 0; | |
148 | mdsc->caps_min_count = 0; | |
149 | spin_unlock(&mdsc->caps_list_lock); | |
85ccce43 SW |
150 | } |
151 | ||
fe33032d YZ |
152 | void ceph_adjust_caps_max_min(struct ceph_mds_client *mdsc, |
153 | struct ceph_mount_options *fsopt) | |
85ccce43 | 154 | { |
37151668 | 155 | spin_lock(&mdsc->caps_list_lock); |
fe33032d YZ |
156 | mdsc->caps_min_count = fsopt->max_readdir; |
157 | if (mdsc->caps_min_count < 1024) | |
158 | mdsc->caps_min_count = 1024; | |
159 | mdsc->caps_use_max = fsopt->caps_max; | |
160 | if (mdsc->caps_use_max > 0 && | |
161 | mdsc->caps_use_max < mdsc->caps_min_count) | |
162 | mdsc->caps_use_max = mdsc->caps_min_count; | |
37151668 | 163 | spin_unlock(&mdsc->caps_list_lock); |
a8599bd8 SW |
164 | } |
165 | ||
7bf8f736 CX |
166 | static void __ceph_unreserve_caps(struct ceph_mds_client *mdsc, int nr_caps) |
167 | { | |
168 | struct ceph_cap *cap; | |
169 | int i; | |
170 | ||
171 | if (nr_caps) { | |
172 | BUG_ON(mdsc->caps_reserve_count < nr_caps); | |
173 | mdsc->caps_reserve_count -= nr_caps; | |
174 | if (mdsc->caps_avail_count >= | |
175 | mdsc->caps_reserve_count + mdsc->caps_min_count) { | |
176 | mdsc->caps_total_count -= nr_caps; | |
177 | for (i = 0; i < nr_caps; i++) { | |
178 | cap = list_first_entry(&mdsc->caps_list, | |
179 | struct ceph_cap, caps_item); | |
180 | list_del(&cap->caps_item); | |
181 | kmem_cache_free(ceph_cap_cachep, cap); | |
182 | } | |
183 | } else { | |
184 | mdsc->caps_avail_count += nr_caps; | |
185 | } | |
186 | ||
187 | dout("%s: caps %d = %d used + %d resv + %d avail\n", | |
188 | __func__, | |
189 | mdsc->caps_total_count, mdsc->caps_use_count, | |
190 | mdsc->caps_reserve_count, mdsc->caps_avail_count); | |
191 | BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count + | |
192 | mdsc->caps_reserve_count + | |
193 | mdsc->caps_avail_count); | |
194 | } | |
195 | } | |
196 | ||
e30ee581 ZZ |
197 | /* |
198 | * Called under mdsc->mutex. | |
199 | */ | |
200 | int ceph_reserve_caps(struct ceph_mds_client *mdsc, | |
37151668 | 201 | struct ceph_cap_reservation *ctx, int need) |
a8599bd8 | 202 | { |
e30ee581 | 203 | int i, j; |
a8599bd8 SW |
204 | struct ceph_cap *cap; |
205 | int have; | |
206 | int alloc = 0; | |
e30ee581 | 207 | int max_caps; |
e5bc08d0 | 208 | int err = 0; |
e30ee581 ZZ |
209 | bool trimmed = false; |
210 | struct ceph_mds_session *s; | |
a8599bd8 | 211 | LIST_HEAD(newcaps); |
a8599bd8 SW |
212 | |
213 | dout("reserve caps ctx=%p need=%d\n", ctx, need); | |
214 | ||
215 | /* first reserve any caps that are already allocated */ | |
37151668 YS |
216 | spin_lock(&mdsc->caps_list_lock); |
217 | if (mdsc->caps_avail_count >= need) | |
a8599bd8 SW |
218 | have = need; |
219 | else | |
37151668 YS |
220 | have = mdsc->caps_avail_count; |
221 | mdsc->caps_avail_count -= have; | |
222 | mdsc->caps_reserve_count += have; | |
223 | BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count + | |
224 | mdsc->caps_reserve_count + | |
225 | mdsc->caps_avail_count); | |
226 | spin_unlock(&mdsc->caps_list_lock); | |
a8599bd8 | 227 | |
79cd674a | 228 | for (i = have; i < need; ) { |
a8599bd8 | 229 | cap = kmem_cache_alloc(ceph_cap_cachep, GFP_NOFS); |
79cd674a CX |
230 | if (cap) { |
231 | list_add(&cap->caps_item, &newcaps); | |
232 | alloc++; | |
233 | i++; | |
234 | continue; | |
235 | } | |
236 | ||
237 | if (!trimmed) { | |
238 | for (j = 0; j < mdsc->max_sessions; j++) { | |
239 | s = __ceph_lookup_mds_session(mdsc, j); | |
240 | if (!s) | |
241 | continue; | |
242 | mutex_unlock(&mdsc->mutex); | |
243 | ||
244 | mutex_lock(&s->s_mutex); | |
245 | max_caps = s->s_nr_caps - (need - i); | |
246 | ceph_trim_caps(mdsc, s, max_caps); | |
247 | mutex_unlock(&s->s_mutex); | |
248 | ||
249 | ceph_put_mds_session(s); | |
250 | mutex_lock(&mdsc->mutex); | |
e30ee581 | 251 | } |
79cd674a CX |
252 | trimmed = true; |
253 | ||
254 | spin_lock(&mdsc->caps_list_lock); | |
255 | if (mdsc->caps_avail_count) { | |
256 | int more_have; | |
257 | if (mdsc->caps_avail_count >= need - i) | |
258 | more_have = need - i; | |
259 | else | |
260 | more_have = mdsc->caps_avail_count; | |
261 | ||
262 | i += more_have; | |
263 | have += more_have; | |
264 | mdsc->caps_avail_count -= more_have; | |
265 | mdsc->caps_reserve_count += more_have; | |
266 | ||
267 | } | |
268 | spin_unlock(&mdsc->caps_list_lock); | |
269 | ||
270 | continue; | |
e30ee581 | 271 | } |
79cd674a CX |
272 | |
273 | pr_warn("reserve caps ctx=%p ENOMEM need=%d got=%d\n", | |
274 | ctx, need, have + alloc); | |
e5bc08d0 CX |
275 | err = -ENOMEM; |
276 | break; | |
277 | } | |
278 | ||
279 | if (!err) { | |
280 | BUG_ON(have + alloc != need); | |
281 | ctx->count = need; | |
fe33032d | 282 | ctx->used = 0; |
a8599bd8 | 283 | } |
a8599bd8 | 284 | |
37151668 YS |
285 | spin_lock(&mdsc->caps_list_lock); |
286 | mdsc->caps_total_count += alloc; | |
287 | mdsc->caps_reserve_count += alloc; | |
288 | list_splice(&newcaps, &mdsc->caps_list); | |
a8599bd8 | 289 | |
37151668 YS |
290 | BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count + |
291 | mdsc->caps_reserve_count + | |
292 | mdsc->caps_avail_count); | |
e5bc08d0 CX |
293 | |
294 | if (err) | |
295 | __ceph_unreserve_caps(mdsc, have + alloc); | |
296 | ||
37151668 | 297 | spin_unlock(&mdsc->caps_list_lock); |
a8599bd8 | 298 | |
a8599bd8 | 299 | dout("reserve caps ctx=%p %d = %d used + %d resv + %d avail\n", |
37151668 YS |
300 | ctx, mdsc->caps_total_count, mdsc->caps_use_count, |
301 | mdsc->caps_reserve_count, mdsc->caps_avail_count); | |
e5bc08d0 | 302 | return err; |
a8599bd8 SW |
303 | } |
304 | ||
7bf8f736 | 305 | void ceph_unreserve_caps(struct ceph_mds_client *mdsc, |
fe33032d | 306 | struct ceph_cap_reservation *ctx) |
a8599bd8 | 307 | { |
fe33032d YZ |
308 | bool reclaim = false; |
309 | if (!ctx->count) | |
310 | return; | |
311 | ||
a8599bd8 | 312 | dout("unreserve caps ctx=%p count=%d\n", ctx, ctx->count); |
7bf8f736 CX |
313 | spin_lock(&mdsc->caps_list_lock); |
314 | __ceph_unreserve_caps(mdsc, ctx->count); | |
315 | ctx->count = 0; | |
fe33032d YZ |
316 | |
317 | if (mdsc->caps_use_max > 0 && | |
318 | mdsc->caps_use_count > mdsc->caps_use_max) | |
319 | reclaim = true; | |
7bf8f736 | 320 | spin_unlock(&mdsc->caps_list_lock); |
fe33032d YZ |
321 | |
322 | if (reclaim) | |
323 | ceph_reclaim_caps_nr(mdsc, ctx->used); | |
a8599bd8 SW |
324 | } |
325 | ||
d9df2783 YZ |
326 | struct ceph_cap *ceph_get_cap(struct ceph_mds_client *mdsc, |
327 | struct ceph_cap_reservation *ctx) | |
a8599bd8 SW |
328 | { |
329 | struct ceph_cap *cap = NULL; | |
330 | ||
331 | /* temporary, until we do something about cap import/export */ | |
443b3760 SW |
332 | if (!ctx) { |
333 | cap = kmem_cache_alloc(ceph_cap_cachep, GFP_NOFS); | |
334 | if (cap) { | |
4d1d0534 | 335 | spin_lock(&mdsc->caps_list_lock); |
37151668 YS |
336 | mdsc->caps_use_count++; |
337 | mdsc->caps_total_count++; | |
4d1d0534 | 338 | spin_unlock(&mdsc->caps_list_lock); |
e327ce06 CX |
339 | } else { |
340 | spin_lock(&mdsc->caps_list_lock); | |
341 | if (mdsc->caps_avail_count) { | |
342 | BUG_ON(list_empty(&mdsc->caps_list)); | |
343 | ||
344 | mdsc->caps_avail_count--; | |
345 | mdsc->caps_use_count++; | |
346 | cap = list_first_entry(&mdsc->caps_list, | |
347 | struct ceph_cap, caps_item); | |
348 | list_del(&cap->caps_item); | |
349 | ||
350 | BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count + | |
351 | mdsc->caps_reserve_count + mdsc->caps_avail_count); | |
352 | } | |
353 | spin_unlock(&mdsc->caps_list_lock); | |
443b3760 | 354 | } |
e327ce06 | 355 | |
443b3760 SW |
356 | return cap; |
357 | } | |
a8599bd8 | 358 | |
37151668 | 359 | spin_lock(&mdsc->caps_list_lock); |
a8599bd8 | 360 | dout("get_cap ctx=%p (%d) %d = %d used + %d resv + %d avail\n", |
37151668 YS |
361 | ctx, ctx->count, mdsc->caps_total_count, mdsc->caps_use_count, |
362 | mdsc->caps_reserve_count, mdsc->caps_avail_count); | |
a8599bd8 | 363 | BUG_ON(!ctx->count); |
37151668 YS |
364 | BUG_ON(ctx->count > mdsc->caps_reserve_count); |
365 | BUG_ON(list_empty(&mdsc->caps_list)); | |
a8599bd8 SW |
366 | |
367 | ctx->count--; | |
fe33032d | 368 | ctx->used++; |
37151668 YS |
369 | mdsc->caps_reserve_count--; |
370 | mdsc->caps_use_count++; | |
a8599bd8 | 371 | |
37151668 | 372 | cap = list_first_entry(&mdsc->caps_list, struct ceph_cap, caps_item); |
a8599bd8 SW |
373 | list_del(&cap->caps_item); |
374 | ||
37151668 YS |
375 | BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count + |
376 | mdsc->caps_reserve_count + mdsc->caps_avail_count); | |
377 | spin_unlock(&mdsc->caps_list_lock); | |
a8599bd8 SW |
378 | return cap; |
379 | } | |
380 | ||
37151668 | 381 | void ceph_put_cap(struct ceph_mds_client *mdsc, struct ceph_cap *cap) |
a8599bd8 | 382 | { |
37151668 | 383 | spin_lock(&mdsc->caps_list_lock); |
7c1332b8 | 384 | dout("put_cap %p %d = %d used + %d resv + %d avail\n", |
37151668 YS |
385 | cap, mdsc->caps_total_count, mdsc->caps_use_count, |
386 | mdsc->caps_reserve_count, mdsc->caps_avail_count); | |
387 | mdsc->caps_use_count--; | |
a8599bd8 | 388 | /* |
85ccce43 SW |
389 | * Keep some preallocated caps around (ceph_min_count), to |
390 | * avoid lots of free/alloc churn. | |
a8599bd8 | 391 | */ |
37151668 YS |
392 | if (mdsc->caps_avail_count >= mdsc->caps_reserve_count + |
393 | mdsc->caps_min_count) { | |
394 | mdsc->caps_total_count--; | |
a8599bd8 SW |
395 | kmem_cache_free(ceph_cap_cachep, cap); |
396 | } else { | |
37151668 YS |
397 | mdsc->caps_avail_count++; |
398 | list_add(&cap->caps_item, &mdsc->caps_list); | |
a8599bd8 SW |
399 | } |
400 | ||
37151668 YS |
401 | BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count + |
402 | mdsc->caps_reserve_count + mdsc->caps_avail_count); | |
403 | spin_unlock(&mdsc->caps_list_lock); | |
a8599bd8 SW |
404 | } |
405 | ||
3d14c5d2 | 406 | void ceph_reservation_status(struct ceph_fs_client *fsc, |
85ccce43 SW |
407 | int *total, int *avail, int *used, int *reserved, |
408 | int *min) | |
a8599bd8 | 409 | { |
3d14c5d2 | 410 | struct ceph_mds_client *mdsc = fsc->mdsc; |
37151668 | 411 | |
b884014a CX |
412 | spin_lock(&mdsc->caps_list_lock); |
413 | ||
a8599bd8 | 414 | if (total) |
37151668 | 415 | *total = mdsc->caps_total_count; |
a8599bd8 | 416 | if (avail) |
37151668 | 417 | *avail = mdsc->caps_avail_count; |
a8599bd8 | 418 | if (used) |
37151668 | 419 | *used = mdsc->caps_use_count; |
a8599bd8 | 420 | if (reserved) |
37151668 | 421 | *reserved = mdsc->caps_reserve_count; |
85ccce43 | 422 | if (min) |
37151668 | 423 | *min = mdsc->caps_min_count; |
b884014a CX |
424 | |
425 | spin_unlock(&mdsc->caps_list_lock); | |
a8599bd8 SW |
426 | } |
427 | ||
428 | /* | |
429 | * Find ceph_cap for given mds, if any. | |
430 | * | |
be655596 | 431 | * Called with i_ceph_lock held. |
a8599bd8 SW |
432 | */ |
433 | static struct ceph_cap *__get_cap_for_mds(struct ceph_inode_info *ci, int mds) | |
434 | { | |
435 | struct ceph_cap *cap; | |
436 | struct rb_node *n = ci->i_caps.rb_node; | |
437 | ||
438 | while (n) { | |
439 | cap = rb_entry(n, struct ceph_cap, ci_node); | |
440 | if (mds < cap->mds) | |
441 | n = n->rb_left; | |
442 | else if (mds > cap->mds) | |
443 | n = n->rb_right; | |
444 | else | |
445 | return cap; | |
446 | } | |
447 | return NULL; | |
448 | } | |
449 | ||
2bc50259 GF |
450 | struct ceph_cap *ceph_get_cap_for_mds(struct ceph_inode_info *ci, int mds) |
451 | { | |
452 | struct ceph_cap *cap; | |
453 | ||
be655596 | 454 | spin_lock(&ci->i_ceph_lock); |
2bc50259 | 455 | cap = __get_cap_for_mds(ci, mds); |
be655596 | 456 | spin_unlock(&ci->i_ceph_lock); |
2bc50259 GF |
457 | return cap; |
458 | } | |
459 | ||
a8599bd8 | 460 | /* |
be655596 | 461 | * Called under i_ceph_lock. |
a8599bd8 SW |
462 | */ |
463 | static void __insert_cap_node(struct ceph_inode_info *ci, | |
464 | struct ceph_cap *new) | |
465 | { | |
466 | struct rb_node **p = &ci->i_caps.rb_node; | |
467 | struct rb_node *parent = NULL; | |
468 | struct ceph_cap *cap = NULL; | |
469 | ||
470 | while (*p) { | |
471 | parent = *p; | |
472 | cap = rb_entry(parent, struct ceph_cap, ci_node); | |
473 | if (new->mds < cap->mds) | |
474 | p = &(*p)->rb_left; | |
475 | else if (new->mds > cap->mds) | |
476 | p = &(*p)->rb_right; | |
477 | else | |
478 | BUG(); | |
479 | } | |
480 | ||
481 | rb_link_node(&new->ci_node, parent, p); | |
482 | rb_insert_color(&new->ci_node, &ci->i_caps); | |
483 | } | |
484 | ||
485 | /* | |
486 | * (re)set cap hold timeouts, which control the delayed release | |
487 | * of unused caps back to the MDS. Should be called on cap use. | |
488 | */ | |
489 | static void __cap_set_timeouts(struct ceph_mds_client *mdsc, | |
490 | struct ceph_inode_info *ci) | |
491 | { | |
fe33032d | 492 | struct ceph_mount_options *opt = mdsc->fsc->mount_options; |
a8599bd8 SW |
493 | |
494 | ci->i_hold_caps_min = round_jiffies(jiffies + | |
fe33032d | 495 | opt->caps_wanted_delay_min * HZ); |
a8599bd8 | 496 | ci->i_hold_caps_max = round_jiffies(jiffies + |
fe33032d | 497 | opt->caps_wanted_delay_max * HZ); |
a8599bd8 SW |
498 | dout("__cap_set_timeouts %p min %lu max %lu\n", &ci->vfs_inode, |
499 | ci->i_hold_caps_min - jiffies, ci->i_hold_caps_max - jiffies); | |
500 | } | |
501 | ||
502 | /* | |
503 | * (Re)queue cap at the end of the delayed cap release list. | |
504 | * | |
505 | * If I_FLUSH is set, leave the inode at the front of the list. | |
506 | * | |
be655596 | 507 | * Caller holds i_ceph_lock |
a8599bd8 SW |
508 | * -> we take mdsc->cap_delay_lock |
509 | */ | |
510 | static void __cap_delay_requeue(struct ceph_mds_client *mdsc, | |
66802884 XX |
511 | struct ceph_inode_info *ci, |
512 | bool set_timeout) | |
a8599bd8 | 513 | { |
a8599bd8 SW |
514 | dout("__cap_delay_requeue %p flags %d at %lu\n", &ci->vfs_inode, |
515 | ci->i_ceph_flags, ci->i_hold_caps_max); | |
516 | if (!mdsc->stopping) { | |
517 | spin_lock(&mdsc->cap_delay_lock); | |
518 | if (!list_empty(&ci->i_cap_delay_list)) { | |
519 | if (ci->i_ceph_flags & CEPH_I_FLUSH) | |
520 | goto no_change; | |
521 | list_del_init(&ci->i_cap_delay_list); | |
522 | } | |
66802884 XX |
523 | if (set_timeout) |
524 | __cap_set_timeouts(mdsc, ci); | |
a8599bd8 SW |
525 | list_add_tail(&ci->i_cap_delay_list, &mdsc->cap_delay_list); |
526 | no_change: | |
527 | spin_unlock(&mdsc->cap_delay_lock); | |
528 | } | |
529 | } | |
530 | ||
531 | /* | |
532 | * Queue an inode for immediate writeback. Mark inode with I_FLUSH, | |
533 | * indicating we should send a cap message to flush dirty metadata | |
534 | * asap, and move to the front of the delayed cap list. | |
535 | */ | |
536 | static void __cap_delay_requeue_front(struct ceph_mds_client *mdsc, | |
537 | struct ceph_inode_info *ci) | |
538 | { | |
539 | dout("__cap_delay_requeue_front %p\n", &ci->vfs_inode); | |
540 | spin_lock(&mdsc->cap_delay_lock); | |
541 | ci->i_ceph_flags |= CEPH_I_FLUSH; | |
542 | if (!list_empty(&ci->i_cap_delay_list)) | |
543 | list_del_init(&ci->i_cap_delay_list); | |
544 | list_add(&ci->i_cap_delay_list, &mdsc->cap_delay_list); | |
545 | spin_unlock(&mdsc->cap_delay_lock); | |
546 | } | |
547 | ||
548 | /* | |
549 | * Cancel delayed work on cap. | |
550 | * | |
be655596 | 551 | * Caller must hold i_ceph_lock. |
a8599bd8 SW |
552 | */ |
553 | static void __cap_delay_cancel(struct ceph_mds_client *mdsc, | |
554 | struct ceph_inode_info *ci) | |
555 | { | |
556 | dout("__cap_delay_cancel %p\n", &ci->vfs_inode); | |
557 | if (list_empty(&ci->i_cap_delay_list)) | |
558 | return; | |
559 | spin_lock(&mdsc->cap_delay_lock); | |
560 | list_del_init(&ci->i_cap_delay_list); | |
561 | spin_unlock(&mdsc->cap_delay_lock); | |
562 | } | |
563 | ||
564 | /* | |
565 | * Common issue checks for add_cap, handle_cap_grant. | |
566 | */ | |
567 | static void __check_cap_issue(struct ceph_inode_info *ci, struct ceph_cap *cap, | |
568 | unsigned issued) | |
569 | { | |
570 | unsigned had = __ceph_caps_issued(ci, NULL); | |
571 | ||
572 | /* | |
573 | * Each time we receive FILE_CACHE anew, we increment | |
574 | * i_rdcache_gen. | |
575 | */ | |
2962507c | 576 | if ((issued & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) && |
99ccbd22 | 577 | (had & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) == 0) { |
a8599bd8 | 578 | ci->i_rdcache_gen++; |
99ccbd22 | 579 | } |
a8599bd8 SW |
580 | |
581 | /* | |
15b51bd6 YZ |
582 | * If FILE_SHARED is newly issued, mark dir not complete. We don't |
583 | * know what happened to this directory while we didn't have the cap. | |
584 | * If FILE_SHARED is being revoked, also mark dir not complete. It | |
585 | * stops on-going cached readdir. | |
a8599bd8 | 586 | */ |
15b51bd6 YZ |
587 | if ((issued & CEPH_CAP_FILE_SHARED) != (had & CEPH_CAP_FILE_SHARED)) { |
588 | if (issued & CEPH_CAP_FILE_SHARED) | |
97aeb6bf | 589 | atomic_inc(&ci->i_shared_gen); |
a8673d61 YZ |
590 | if (S_ISDIR(ci->vfs_inode.i_mode)) { |
591 | dout(" marking %p NOT complete\n", &ci->vfs_inode); | |
2f276c51 | 592 | __ceph_dir_clear_complete(ci); |
a8673d61 | 593 | } |
a8599bd8 SW |
594 | } |
595 | } | |
596 | ||
597 | /* | |
598 | * Add a capability under the given MDS session. | |
599 | * | |
354c63a0 | 600 | * Caller should hold session snap_rwsem (read) and ci->i_ceph_lock |
a8599bd8 SW |
601 | * |
602 | * @fmode is the open file mode, if we are opening a file, otherwise | |
603 | * it is < 0. (This is so we can atomically add the cap and add an | |
604 | * open file reference to it.) | |
605 | */ | |
d9df2783 YZ |
606 | void ceph_add_cap(struct inode *inode, |
607 | struct ceph_mds_session *session, u64 cap_id, | |
608 | int fmode, unsigned issued, unsigned wanted, | |
609 | unsigned seq, unsigned mseq, u64 realmino, int flags, | |
610 | struct ceph_cap **new_cap) | |
a8599bd8 | 611 | { |
3d14c5d2 | 612 | struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc; |
a8599bd8 | 613 | struct ceph_inode_info *ci = ceph_inode(inode); |
a8599bd8 SW |
614 | struct ceph_cap *cap; |
615 | int mds = session->s_mds; | |
616 | int actual_wanted; | |
606d1023 | 617 | u32 gen; |
a8599bd8 | 618 | |
354c63a0 JL |
619 | lockdep_assert_held(&ci->i_ceph_lock); |
620 | ||
a8599bd8 SW |
621 | dout("add_cap %p mds%d cap %llx %s seq %d\n", inode, |
622 | session->s_mds, cap_id, ceph_cap_string(issued), seq); | |
623 | ||
624 | /* | |
625 | * If we are opening the file, include file mode wanted bits | |
626 | * in wanted. | |
627 | */ | |
628 | if (fmode >= 0) | |
629 | wanted |= ceph_caps_for_mode(fmode); | |
630 | ||
606d1023 JL |
631 | spin_lock(&session->s_gen_ttl_lock); |
632 | gen = session->s_cap_gen; | |
633 | spin_unlock(&session->s_gen_ttl_lock); | |
634 | ||
a8599bd8 SW |
635 | cap = __get_cap_for_mds(ci, mds); |
636 | if (!cap) { | |
d9df2783 YZ |
637 | cap = *new_cap; |
638 | *new_cap = NULL; | |
a8599bd8 SW |
639 | |
640 | cap->issued = 0; | |
641 | cap->implemented = 0; | |
642 | cap->mds = mds; | |
643 | cap->mds_wanted = 0; | |
964266cc | 644 | cap->mseq = 0; |
a8599bd8 SW |
645 | |
646 | cap->ci = ci; | |
647 | __insert_cap_node(ci, cap); | |
648 | ||
a8599bd8 SW |
649 | /* add to session cap list */ |
650 | cap->session = session; | |
651 | spin_lock(&session->s_cap_lock); | |
652 | list_add_tail(&cap->session_caps, &session->s_caps); | |
653 | session->s_nr_caps++; | |
654 | spin_unlock(&session->s_cap_lock); | |
11df2dfb | 655 | } else { |
32f6511a YZ |
656 | spin_lock(&session->s_cap_lock); |
657 | list_move_tail(&cap->session_caps, &session->s_caps); | |
658 | spin_unlock(&session->s_cap_lock); | |
659 | ||
606d1023 | 660 | if (cap->cap_gen < gen) |
d2f8bb27 YZ |
661 | cap->issued = cap->implemented = CEPH_CAP_PIN; |
662 | ||
11df2dfb YZ |
663 | /* |
664 | * auth mds of the inode changed. we received the cap export | |
665 | * message, but still haven't received the cap import message. | |
666 | * handle_cap_export() updated the new auth MDS' cap. | |
667 | * | |
668 | * "ceph_seq_cmp(seq, cap->seq) <= 0" means we are processing | |
669 | * a message that was send before the cap import message. So | |
670 | * don't remove caps. | |
671 | */ | |
672 | if (ceph_seq_cmp(seq, cap->seq) <= 0) { | |
673 | WARN_ON(cap != ci->i_auth_cap); | |
674 | WARN_ON(cap->cap_id != cap_id); | |
675 | seq = cap->seq; | |
676 | mseq = cap->mseq; | |
677 | issued |= cap->issued; | |
678 | flags |= CEPH_CAP_FLAG_AUTH; | |
679 | } | |
680 | } | |
a8599bd8 | 681 | |
7d9c9193 YZ |
682 | if (!ci->i_snap_realm || |
683 | ((flags & CEPH_CAP_FLAG_AUTH) && | |
684 | realmino != (u64)-1 && ci->i_snap_realm->ino != realmino)) { | |
a8599bd8 SW |
685 | /* |
686 | * add this inode to the appropriate snap realm | |
687 | */ | |
688 | struct ceph_snap_realm *realm = ceph_lookup_snap_realm(mdsc, | |
689 | realmino); | |
690 | if (realm) { | |
7d9c9193 YZ |
691 | struct ceph_snap_realm *oldrealm = ci->i_snap_realm; |
692 | if (oldrealm) { | |
693 | spin_lock(&oldrealm->inodes_with_caps_lock); | |
694 | list_del_init(&ci->i_snap_realm_item); | |
695 | spin_unlock(&oldrealm->inodes_with_caps_lock); | |
696 | } | |
697 | ||
a8599bd8 | 698 | spin_lock(&realm->inodes_with_caps_lock); |
a8599bd8 SW |
699 | list_add(&ci->i_snap_realm_item, |
700 | &realm->inodes_with_caps); | |
e3161f17 LH |
701 | ci->i_snap_realm = realm; |
702 | if (realm->ino == ci->i_vino.ino) | |
703 | realm->inode = inode; | |
a8599bd8 | 704 | spin_unlock(&realm->inodes_with_caps_lock); |
7d9c9193 YZ |
705 | |
706 | if (oldrealm) | |
707 | ceph_put_snap_realm(mdsc, oldrealm); | |
a8599bd8 SW |
708 | } else { |
709 | pr_err("ceph_add_cap: couldn't find snap realm %llx\n", | |
710 | realmino); | |
b8cd07e7 | 711 | WARN_ON(!realm); |
a8599bd8 SW |
712 | } |
713 | } | |
714 | ||
715 | __check_cap_issue(ci, cap, issued); | |
716 | ||
717 | /* | |
718 | * If we are issued caps we don't want, or the mds' wanted | |
719 | * value appears to be off, queue a check so we'll release | |
720 | * later and/or update the mds wanted value. | |
721 | */ | |
722 | actual_wanted = __ceph_caps_wanted(ci); | |
723 | if ((wanted & ~actual_wanted) || | |
724 | (issued & ~actual_wanted & CEPH_CAP_ANY_WR)) { | |
725 | dout(" issued %s, mds wanted %s, actual %s, queueing\n", | |
726 | ceph_cap_string(issued), ceph_cap_string(wanted), | |
727 | ceph_cap_string(actual_wanted)); | |
66802884 | 728 | __cap_delay_requeue(mdsc, ci, true); |
a8599bd8 SW |
729 | } |
730 | ||
b8c2f3ae | 731 | if (flags & CEPH_CAP_FLAG_AUTH) { |
d37b1d99 | 732 | if (!ci->i_auth_cap || |
d9ffc4f7 | 733 | ceph_seq_cmp(ci->i_auth_cap->mseq, mseq) < 0) { |
b8c2f3ae | 734 | ci->i_auth_cap = cap; |
d9ffc4f7 YZ |
735 | cap->mds_wanted = wanted; |
736 | } | |
11df2dfb YZ |
737 | } else { |
738 | WARN_ON(ci->i_auth_cap == cap); | |
8a92a119 | 739 | } |
a8599bd8 SW |
740 | |
741 | dout("add_cap inode %p (%llx.%llx) cap %p %s now %s seq %d mds%d\n", | |
742 | inode, ceph_vinop(inode), cap, ceph_cap_string(issued), | |
743 | ceph_cap_string(issued|cap->issued), seq, mds); | |
744 | cap->cap_id = cap_id; | |
745 | cap->issued = issued; | |
746 | cap->implemented |= issued; | |
d1b87809 | 747 | if (ceph_seq_cmp(mseq, cap->mseq) > 0) |
964266cc YZ |
748 | cap->mds_wanted = wanted; |
749 | else | |
750 | cap->mds_wanted |= wanted; | |
a8599bd8 SW |
751 | cap->seq = seq; |
752 | cap->issue_seq = seq; | |
753 | cap->mseq = mseq; | |
606d1023 | 754 | cap->cap_gen = gen; |
a8599bd8 SW |
755 | |
756 | if (fmode >= 0) | |
757 | __ceph_get_fmode(ci, fmode); | |
a8599bd8 SW |
758 | } |
759 | ||
760 | /* | |
761 | * Return true if cap has not timed out and belongs to the current | |
762 | * generation of the MDS session (i.e. has not gone 'stale' due to | |
763 | * us losing touch with the mds). | |
764 | */ | |
765 | static int __cap_is_valid(struct ceph_cap *cap) | |
766 | { | |
767 | unsigned long ttl; | |
cdac8303 | 768 | u32 gen; |
a8599bd8 | 769 | |
d8fb02ab | 770 | spin_lock(&cap->session->s_gen_ttl_lock); |
a8599bd8 SW |
771 | gen = cap->session->s_cap_gen; |
772 | ttl = cap->session->s_cap_ttl; | |
d8fb02ab | 773 | spin_unlock(&cap->session->s_gen_ttl_lock); |
a8599bd8 | 774 | |
685f9a5d | 775 | if (cap->cap_gen < gen || time_after_eq(jiffies, ttl)) { |
a8599bd8 SW |
776 | dout("__cap_is_valid %p cap %p issued %s " |
777 | "but STALE (gen %u vs %u)\n", &cap->ci->vfs_inode, | |
685f9a5d | 778 | cap, ceph_cap_string(cap->issued), cap->cap_gen, gen); |
a8599bd8 SW |
779 | return 0; |
780 | } | |
781 | ||
782 | return 1; | |
783 | } | |
784 | ||
785 | /* | |
786 | * Return set of valid cap bits issued to us. Note that caps time | |
787 | * out, and may be invalidated in bulk if the client session times out | |
788 | * and session->s_cap_gen is bumped. | |
789 | */ | |
790 | int __ceph_caps_issued(struct ceph_inode_info *ci, int *implemented) | |
791 | { | |
d9df2783 | 792 | int have = ci->i_snap_caps; |
a8599bd8 SW |
793 | struct ceph_cap *cap; |
794 | struct rb_node *p; | |
795 | ||
796 | if (implemented) | |
797 | *implemented = 0; | |
798 | for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) { | |
799 | cap = rb_entry(p, struct ceph_cap, ci_node); | |
800 | if (!__cap_is_valid(cap)) | |
801 | continue; | |
802 | dout("__ceph_caps_issued %p cap %p issued %s\n", | |
803 | &ci->vfs_inode, cap, ceph_cap_string(cap->issued)); | |
804 | have |= cap->issued; | |
805 | if (implemented) | |
806 | *implemented |= cap->implemented; | |
807 | } | |
b1530f57 YZ |
808 | /* |
809 | * exclude caps issued by non-auth MDS, but are been revoking | |
810 | * by the auth MDS. The non-auth MDS should be revoking/exporting | |
811 | * these caps, but the message is delayed. | |
812 | */ | |
813 | if (ci->i_auth_cap) { | |
814 | cap = ci->i_auth_cap; | |
815 | have &= ~cap->implemented | cap->issued; | |
816 | } | |
a8599bd8 SW |
817 | return have; |
818 | } | |
819 | ||
820 | /* | |
821 | * Get cap bits issued by caps other than @ocap | |
822 | */ | |
823 | int __ceph_caps_issued_other(struct ceph_inode_info *ci, struct ceph_cap *ocap) | |
824 | { | |
825 | int have = ci->i_snap_caps; | |
826 | struct ceph_cap *cap; | |
827 | struct rb_node *p; | |
828 | ||
829 | for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) { | |
830 | cap = rb_entry(p, struct ceph_cap, ci_node); | |
831 | if (cap == ocap) | |
832 | continue; | |
833 | if (!__cap_is_valid(cap)) | |
834 | continue; | |
835 | have |= cap->issued; | |
836 | } | |
837 | return have; | |
838 | } | |
839 | ||
840 | /* | |
841 | * Move a cap to the end of the LRU (oldest caps at list head, newest | |
842 | * at list tail). | |
843 | */ | |
844 | static void __touch_cap(struct ceph_cap *cap) | |
845 | { | |
846 | struct ceph_mds_session *s = cap->session; | |
847 | ||
a8599bd8 | 848 | spin_lock(&s->s_cap_lock); |
d37b1d99 | 849 | if (!s->s_cap_iterator) { |
5dacf091 SW |
850 | dout("__touch_cap %p cap %p mds%d\n", &cap->ci->vfs_inode, cap, |
851 | s->s_mds); | |
852 | list_move_tail(&cap->session_caps, &s->s_caps); | |
853 | } else { | |
854 | dout("__touch_cap %p cap %p mds%d NOP, iterating over caps\n", | |
855 | &cap->ci->vfs_inode, cap, s->s_mds); | |
856 | } | |
a8599bd8 SW |
857 | spin_unlock(&s->s_cap_lock); |
858 | } | |
859 | ||
860 | /* | |
861 | * Check if we hold the given mask. If so, move the cap(s) to the | |
862 | * front of their respective LRUs. (This is the preferred way for | |
863 | * callers to check for caps they want.) | |
864 | */ | |
865 | int __ceph_caps_issued_mask(struct ceph_inode_info *ci, int mask, int touch) | |
866 | { | |
867 | struct ceph_cap *cap; | |
868 | struct rb_node *p; | |
869 | int have = ci->i_snap_caps; | |
870 | ||
871 | if ((have & mask) == mask) { | |
5ddc61fc JL |
872 | dout("__ceph_caps_issued_mask ino 0x%lx snap issued %s" |
873 | " (mask %s)\n", ci->vfs_inode.i_ino, | |
a8599bd8 SW |
874 | ceph_cap_string(have), |
875 | ceph_cap_string(mask)); | |
876 | return 1; | |
877 | } | |
878 | ||
879 | for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) { | |
880 | cap = rb_entry(p, struct ceph_cap, ci_node); | |
881 | if (!__cap_is_valid(cap)) | |
882 | continue; | |
883 | if ((cap->issued & mask) == mask) { | |
5ddc61fc JL |
884 | dout("__ceph_caps_issued_mask ino 0x%lx cap %p issued %s" |
885 | " (mask %s)\n", ci->vfs_inode.i_ino, cap, | |
a8599bd8 SW |
886 | ceph_cap_string(cap->issued), |
887 | ceph_cap_string(mask)); | |
888 | if (touch) | |
889 | __touch_cap(cap); | |
890 | return 1; | |
891 | } | |
892 | ||
893 | /* does a combination of caps satisfy mask? */ | |
894 | have |= cap->issued; | |
895 | if ((have & mask) == mask) { | |
5ddc61fc JL |
896 | dout("__ceph_caps_issued_mask ino 0x%lx combo issued %s" |
897 | " (mask %s)\n", ci->vfs_inode.i_ino, | |
a8599bd8 SW |
898 | ceph_cap_string(cap->issued), |
899 | ceph_cap_string(mask)); | |
900 | if (touch) { | |
901 | struct rb_node *q; | |
902 | ||
25985edc | 903 | /* touch this + preceding caps */ |
a8599bd8 SW |
904 | __touch_cap(cap); |
905 | for (q = rb_first(&ci->i_caps); q != p; | |
906 | q = rb_next(q)) { | |
907 | cap = rb_entry(q, struct ceph_cap, | |
908 | ci_node); | |
909 | if (!__cap_is_valid(cap)) | |
910 | continue; | |
9f8b72b3 XL |
911 | if (cap->issued & mask) |
912 | __touch_cap(cap); | |
a8599bd8 SW |
913 | } |
914 | } | |
915 | return 1; | |
916 | } | |
917 | } | |
918 | ||
919 | return 0; | |
920 | } | |
921 | ||
922 | /* | |
923 | * Return true if mask caps are currently being revoked by an MDS. | |
924 | */ | |
6ee6b953 YZ |
925 | int __ceph_caps_revoking_other(struct ceph_inode_info *ci, |
926 | struct ceph_cap *ocap, int mask) | |
a8599bd8 | 927 | { |
a8599bd8 SW |
928 | struct ceph_cap *cap; |
929 | struct rb_node *p; | |
a8599bd8 | 930 | |
a8599bd8 SW |
931 | for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) { |
932 | cap = rb_entry(p, struct ceph_cap, ci_node); | |
9563f88c | 933 | if (cap != ocap && |
6ee6b953 YZ |
934 | (cap->implemented & ~cap->issued & mask)) |
935 | return 1; | |
a8599bd8 | 936 | } |
6ee6b953 YZ |
937 | return 0; |
938 | } | |
939 | ||
940 | int ceph_caps_revoking(struct ceph_inode_info *ci, int mask) | |
941 | { | |
942 | struct inode *inode = &ci->vfs_inode; | |
943 | int ret; | |
944 | ||
945 | spin_lock(&ci->i_ceph_lock); | |
946 | ret = __ceph_caps_revoking_other(ci, NULL, mask); | |
be655596 | 947 | spin_unlock(&ci->i_ceph_lock); |
a8599bd8 SW |
948 | dout("ceph_caps_revoking %p %s = %d\n", inode, |
949 | ceph_cap_string(mask), ret); | |
950 | return ret; | |
951 | } | |
952 | ||
953 | int __ceph_caps_used(struct ceph_inode_info *ci) | |
954 | { | |
955 | int used = 0; | |
956 | if (ci->i_pin_ref) | |
957 | used |= CEPH_CAP_PIN; | |
958 | if (ci->i_rd_ref) | |
959 | used |= CEPH_CAP_FILE_RD; | |
fdd4e158 YZ |
960 | if (ci->i_rdcache_ref || |
961 | (!S_ISDIR(ci->vfs_inode.i_mode) && /* ignore readdir cache */ | |
962 | ci->vfs_inode.i_data.nrpages)) | |
a8599bd8 SW |
963 | used |= CEPH_CAP_FILE_CACHE; |
964 | if (ci->i_wr_ref) | |
965 | used |= CEPH_CAP_FILE_WR; | |
d3d0720d | 966 | if (ci->i_wb_ref || ci->i_wrbuffer_ref) |
a8599bd8 SW |
967 | used |= CEPH_CAP_FILE_BUFFER; |
968 | return used; | |
969 | } | |
970 | ||
971 | /* | |
972 | * wanted, by virtue of open file modes | |
973 | */ | |
974 | int __ceph_caps_file_wanted(struct ceph_inode_info *ci) | |
975 | { | |
774a6a11 YZ |
976 | int i, bits = 0; |
977 | for (i = 0; i < CEPH_FILE_MODE_BITS; i++) { | |
978 | if (ci->i_nr_by_mode[i]) | |
979 | bits |= 1 << i; | |
980 | } | |
981 | if (bits == 0) | |
982 | return 0; | |
983 | return ceph_caps_for_mode(bits >> 1); | |
a8599bd8 SW |
984 | } |
985 | ||
986 | /* | |
987 | * Return caps we have registered with the MDS(s) as 'wanted'. | |
988 | */ | |
c1944fed | 989 | int __ceph_caps_mds_wanted(struct ceph_inode_info *ci, bool check) |
a8599bd8 SW |
990 | { |
991 | struct ceph_cap *cap; | |
992 | struct rb_node *p; | |
993 | int mds_wanted = 0; | |
994 | ||
995 | for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) { | |
996 | cap = rb_entry(p, struct ceph_cap, ci_node); | |
c1944fed | 997 | if (check && !__cap_is_valid(cap)) |
a8599bd8 | 998 | continue; |
a2550604 YZ |
999 | if (cap == ci->i_auth_cap) |
1000 | mds_wanted |= cap->mds_wanted; | |
1001 | else | |
1002 | mds_wanted |= (cap->mds_wanted & ~CEPH_CAP_ANY_FILE_WR); | |
a8599bd8 SW |
1003 | } |
1004 | return mds_wanted; | |
1005 | } | |
1006 | ||
1007 | /* | |
be655596 | 1008 | * called under i_ceph_lock |
a8599bd8 | 1009 | */ |
0f439c74 YZ |
1010 | static int __ceph_is_single_caps(struct ceph_inode_info *ci) |
1011 | { | |
1012 | return rb_first(&ci->i_caps) == rb_last(&ci->i_caps); | |
1013 | } | |
1014 | ||
9215aeea YZ |
1015 | int ceph_is_any_caps(struct inode *inode) |
1016 | { | |
1017 | struct ceph_inode_info *ci = ceph_inode(inode); | |
1018 | int ret; | |
1019 | ||
1020 | spin_lock(&ci->i_ceph_lock); | |
bd84fbcb | 1021 | ret = __ceph_is_any_real_caps(ci); |
9215aeea YZ |
1022 | spin_unlock(&ci->i_ceph_lock); |
1023 | ||
1024 | return ret; | |
1025 | } | |
1026 | ||
db40cc17 YZ |
1027 | static void drop_inode_snap_realm(struct ceph_inode_info *ci) |
1028 | { | |
1029 | struct ceph_snap_realm *realm = ci->i_snap_realm; | |
1030 | spin_lock(&realm->inodes_with_caps_lock); | |
1031 | list_del_init(&ci->i_snap_realm_item); | |
1032 | ci->i_snap_realm_counter++; | |
1033 | ci->i_snap_realm = NULL; | |
d95e674c YZ |
1034 | if (realm->ino == ci->i_vino.ino) |
1035 | realm->inode = NULL; | |
db40cc17 YZ |
1036 | spin_unlock(&realm->inodes_with_caps_lock); |
1037 | ceph_put_snap_realm(ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc, | |
1038 | realm); | |
1039 | } | |
1040 | ||
a8599bd8 | 1041 | /* |
f818a736 SW |
1042 | * Remove a cap. Take steps to deal with a racing iterate_session_caps. |
1043 | * | |
be655596 | 1044 | * caller should hold i_ceph_lock. |
a6369741 | 1045 | * caller will not hold session s_mutex if called from destroy_inode. |
a8599bd8 | 1046 | */ |
a096b09a | 1047 | void __ceph_remove_cap(struct ceph_cap *cap, bool queue_release) |
a8599bd8 SW |
1048 | { |
1049 | struct ceph_mds_session *session = cap->session; | |
1050 | struct ceph_inode_info *ci = cap->ci; | |
640ef79d | 1051 | struct ceph_mds_client *mdsc = |
3d14c5d2 | 1052 | ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc; |
f818a736 | 1053 | int removed = 0; |
a8599bd8 SW |
1054 | |
1055 | dout("__ceph_remove_cap %p from %p\n", cap, &ci->vfs_inode); | |
1056 | ||
ea60ed6f LH |
1057 | /* remove from inode's cap rbtree, and clear auth cap */ |
1058 | rb_erase(&cap->ci_node, &ci->i_caps); | |
1059 | if (ci->i_auth_cap == cap) | |
1060 | ci->i_auth_cap = NULL; | |
1061 | ||
7c1332b8 SW |
1062 | /* remove from session list */ |
1063 | spin_lock(&session->s_cap_lock); | |
1064 | if (session->s_cap_iterator == cap) { | |
1065 | /* not yet, we are iterating over this very cap */ | |
1066 | dout("__ceph_remove_cap delaying %p removal from session %p\n", | |
1067 | cap, cap->session); | |
1068 | } else { | |
1069 | list_del_init(&cap->session_caps); | |
1070 | session->s_nr_caps--; | |
1071 | cap->session = NULL; | |
f818a736 | 1072 | removed = 1; |
7c1332b8 | 1073 | } |
f818a736 SW |
1074 | /* protect backpointer with s_cap_lock: see iterate_session_caps */ |
1075 | cap->ci = NULL; | |
745a8e3b YZ |
1076 | |
1077 | /* | |
1078 | * s_cap_reconnect is protected by s_cap_lock. no one changes | |
1079 | * s_cap_gen while session is in the reconnect state. | |
1080 | */ | |
1081 | if (queue_release && | |
1082 | (!session->s_cap_reconnect || cap->cap_gen == session->s_cap_gen)) { | |
1083 | cap->queue_release = 1; | |
1084 | if (removed) { | |
e3ec8d68 | 1085 | __ceph_queue_cap_release(session, cap); |
745a8e3b YZ |
1086 | removed = 0; |
1087 | } | |
1088 | } else { | |
1089 | cap->queue_release = 0; | |
1090 | } | |
1091 | cap->cap_ino = ci->i_vino.ino; | |
1092 | ||
7c1332b8 SW |
1093 | spin_unlock(&session->s_cap_lock); |
1094 | ||
f818a736 | 1095 | if (removed) |
37151668 | 1096 | ceph_put_cap(mdsc, cap); |
a8599bd8 | 1097 | |
bd84fbcb XL |
1098 | if (!__ceph_is_any_real_caps(ci)) { |
1099 | /* when reconnect denied, we remove session caps forcibly, | |
1100 | * i_wr_ref can be non-zero. If there are ongoing write, | |
1101 | * keep i_snap_realm. | |
1102 | */ | |
1103 | if (ci->i_wr_ref == 0 && ci->i_snap_realm) | |
1104 | drop_inode_snap_realm(ci); | |
db40cc17 | 1105 | |
a8599bd8 | 1106 | __cap_delay_cancel(mdsc, ci); |
bd84fbcb | 1107 | } |
a8599bd8 SW |
1108 | } |
1109 | ||
0ff8bfb3 JL |
1110 | struct cap_msg_args { |
1111 | struct ceph_mds_session *session; | |
1112 | u64 ino, cid, follows; | |
1113 | u64 flush_tid, oldest_flush_tid, size, max_size; | |
1114 | u64 xattr_version; | |
176c77c9 | 1115 | u64 change_attr; |
0ff8bfb3 | 1116 | struct ceph_buffer *xattr_buf; |
ec62b894 | 1117 | struct timespec64 atime, mtime, ctime, btime; |
0ff8bfb3 JL |
1118 | int op, caps, wanted, dirty; |
1119 | u32 seq, issue_seq, mseq, time_warp_seq; | |
1e4ef0c6 | 1120 | u32 flags; |
0ff8bfb3 JL |
1121 | kuid_t uid; |
1122 | kgid_t gid; | |
1123 | umode_t mode; | |
1124 | bool inline_data; | |
1125 | }; | |
1126 | ||
a8599bd8 SW |
1127 | /* |
1128 | * Build and send a cap message to the given MDS. | |
1129 | * | |
1130 | * Caller should be holding s_mutex. | |
1131 | */ | |
0ff8bfb3 | 1132 | static int send_cap_msg(struct cap_msg_args *arg) |
a8599bd8 SW |
1133 | { |
1134 | struct ceph_mds_caps *fc; | |
1135 | struct ceph_msg *msg; | |
e20d258d YZ |
1136 | void *p; |
1137 | size_t extra_len; | |
92475f05 | 1138 | struct ceph_osd_client *osdc = &arg->session->s_mdsc->fsc->client->osdc; |
a8599bd8 SW |
1139 | |
1140 | dout("send_cap_msg %s %llx %llx caps %s wanted %s dirty %s" | |
a2971c8c | 1141 | " seq %u/%u tid %llu/%llu mseq %u follows %lld size %llu/%llu" |
0ff8bfb3 JL |
1142 | " xattr_ver %llu xattr_len %d\n", ceph_cap_op_name(arg->op), |
1143 | arg->cid, arg->ino, ceph_cap_string(arg->caps), | |
1144 | ceph_cap_string(arg->wanted), ceph_cap_string(arg->dirty), | |
1145 | arg->seq, arg->issue_seq, arg->flush_tid, arg->oldest_flush_tid, | |
1146 | arg->mseq, arg->follows, arg->size, arg->max_size, | |
1147 | arg->xattr_version, | |
1148 | arg->xattr_buf ? (int)arg->xattr_buf->vec.iov_len : 0); | |
a8599bd8 | 1149 | |
a2971c8c YZ |
1150 | /* flock buffer size + inline version + inline data size + |
1151 | * osd_epoch_barrier + oldest_flush_tid */ | |
43b29673 | 1152 | extra_len = 4 + 8 + 4 + 4 + 8 + 4 + 4 + 4 + 8 + 8 + 4; |
e20d258d YZ |
1153 | msg = ceph_msg_new(CEPH_MSG_CLIENT_CAPS, sizeof(*fc) + extra_len, |
1154 | GFP_NOFS, false); | |
a79832f2 SW |
1155 | if (!msg) |
1156 | return -ENOMEM; | |
a8599bd8 | 1157 | |
43b29673 | 1158 | msg->hdr.version = cpu_to_le16(10); |
0ff8bfb3 | 1159 | msg->hdr.tid = cpu_to_le64(arg->flush_tid); |
a8599bd8 | 1160 | |
6df058c0 | 1161 | fc = msg->front.iov_base; |
a8599bd8 SW |
1162 | memset(fc, 0, sizeof(*fc)); |
1163 | ||
0ff8bfb3 JL |
1164 | fc->cap_id = cpu_to_le64(arg->cid); |
1165 | fc->op = cpu_to_le32(arg->op); | |
1166 | fc->seq = cpu_to_le32(arg->seq); | |
1167 | fc->issue_seq = cpu_to_le32(arg->issue_seq); | |
1168 | fc->migrate_seq = cpu_to_le32(arg->mseq); | |
1169 | fc->caps = cpu_to_le32(arg->caps); | |
1170 | fc->wanted = cpu_to_le32(arg->wanted); | |
1171 | fc->dirty = cpu_to_le32(arg->dirty); | |
1172 | fc->ino = cpu_to_le64(arg->ino); | |
1173 | fc->snap_follows = cpu_to_le64(arg->follows); | |
1174 | ||
1175 | fc->size = cpu_to_le64(arg->size); | |
1176 | fc->max_size = cpu_to_le64(arg->max_size); | |
9bbeab41 AB |
1177 | ceph_encode_timespec64(&fc->mtime, &arg->mtime); |
1178 | ceph_encode_timespec64(&fc->atime, &arg->atime); | |
1179 | ceph_encode_timespec64(&fc->ctime, &arg->ctime); | |
0ff8bfb3 JL |
1180 | fc->time_warp_seq = cpu_to_le32(arg->time_warp_seq); |
1181 | ||
1182 | fc->uid = cpu_to_le32(from_kuid(&init_user_ns, arg->uid)); | |
1183 | fc->gid = cpu_to_le32(from_kgid(&init_user_ns, arg->gid)); | |
1184 | fc->mode = cpu_to_le32(arg->mode); | |
1185 | ||
1186 | fc->xattr_version = cpu_to_le64(arg->xattr_version); | |
1187 | if (arg->xattr_buf) { | |
1188 | msg->middle = ceph_buffer_get(arg->xattr_buf); | |
1189 | fc->xattr_len = cpu_to_le32(arg->xattr_buf->vec.iov_len); | |
1190 | msg->hdr.middle_len = cpu_to_le32(arg->xattr_buf->vec.iov_len); | |
9670079f JL |
1191 | } |
1192 | ||
e20d258d | 1193 | p = fc + 1; |
43b29673 | 1194 | /* flock buffer size (version 2) */ |
e20d258d | 1195 | ceph_encode_32(&p, 0); |
43b29673 | 1196 | /* inline version (version 4) */ |
0ff8bfb3 | 1197 | ceph_encode_64(&p, arg->inline_data ? 0 : CEPH_INLINE_NONE); |
e20d258d YZ |
1198 | /* inline data size */ |
1199 | ceph_encode_32(&p, 0); | |
92475f05 JL |
1200 | /* |
1201 | * osd_epoch_barrier (version 5) | |
1202 | * The epoch_barrier is protected osdc->lock, so READ_ONCE here in | |
1203 | * case it was recently changed | |
1204 | */ | |
1205 | ceph_encode_32(&p, READ_ONCE(osdc->epoch_barrier)); | |
43b29673 | 1206 | /* oldest_flush_tid (version 6) */ |
0ff8bfb3 | 1207 | ceph_encode_64(&p, arg->oldest_flush_tid); |
e20d258d | 1208 | |
43b29673 JL |
1209 | /* |
1210 | * caller_uid/caller_gid (version 7) | |
1211 | * | |
1212 | * Currently, we don't properly track which caller dirtied the caps | |
1213 | * last, and force a flush of them when there is a conflict. For now, | |
1214 | * just set this to 0:0, to emulate how the MDS has worked up to now. | |
1215 | */ | |
1216 | ceph_encode_32(&p, 0); | |
1217 | ceph_encode_32(&p, 0); | |
1218 | ||
1219 | /* pool namespace (version 8) (mds always ignores this) */ | |
1220 | ceph_encode_32(&p, 0); | |
1221 | ||
176c77c9 | 1222 | /* btime and change_attr (version 9) */ |
ec62b894 | 1223 | ceph_encode_timespec64(p, &arg->btime); |
43b29673 | 1224 | p += sizeof(struct ceph_timespec); |
176c77c9 | 1225 | ceph_encode_64(&p, arg->change_attr); |
43b29673 JL |
1226 | |
1227 | /* Advisory flags (version 10) */ | |
1e4ef0c6 | 1228 | ceph_encode_32(&p, arg->flags); |
43b29673 | 1229 | |
0ff8bfb3 | 1230 | ceph_con_send(&arg->session->s_con, msg); |
a8599bd8 SW |
1231 | return 0; |
1232 | } | |
1233 | ||
1234 | /* | |
d6e47819 | 1235 | * Queue cap releases when an inode is dropped from our cache. |
a8599bd8 | 1236 | */ |
d6e47819 | 1237 | void __ceph_remove_caps(struct ceph_inode_info *ci) |
a8599bd8 | 1238 | { |
a8599bd8 SW |
1239 | struct rb_node *p; |
1240 | ||
d6e47819 YZ |
1241 | /* lock i_ceph_lock, because ceph_d_revalidate(..., LOOKUP_RCU) |
1242 | * may call __ceph_caps_issued_mask() on a freeing inode. */ | |
1243 | spin_lock(&ci->i_ceph_lock); | |
a8599bd8 SW |
1244 | p = rb_first(&ci->i_caps); |
1245 | while (p) { | |
1246 | struct ceph_cap *cap = rb_entry(p, struct ceph_cap, ci_node); | |
a8599bd8 | 1247 | p = rb_next(p); |
a096b09a | 1248 | __ceph_remove_cap(cap, true); |
a8599bd8 | 1249 | } |
d6e47819 | 1250 | spin_unlock(&ci->i_ceph_lock); |
a8599bd8 SW |
1251 | } |
1252 | ||
1253 | /* | |
1254 | * Send a cap msg on the given inode. Update our caps state, then | |
be655596 | 1255 | * drop i_ceph_lock and send the message. |
a8599bd8 SW |
1256 | * |
1257 | * Make note of max_size reported/requested from mds, revoked caps | |
1258 | * that have now been implemented. | |
1259 | * | |
a8599bd8 SW |
1260 | * Return non-zero if delayed release, or we experienced an error |
1261 | * such that the caller should requeue + retry later. | |
1262 | * | |
be655596 | 1263 | * called with i_ceph_lock, then drops it. |
a8599bd8 SW |
1264 | * caller should hold snap_rwsem (read), s_mutex. |
1265 | */ | |
1266 | static int __send_cap(struct ceph_mds_client *mdsc, struct ceph_cap *cap, | |
49ada6e8 | 1267 | int op, int flags, int used, int want, int retain, |
1e4ef0c6 | 1268 | int flushing, u64 flush_tid, u64 oldest_flush_tid) |
be655596 | 1269 | __releases(cap->ci->i_ceph_lock) |
a8599bd8 SW |
1270 | { |
1271 | struct ceph_inode_info *ci = cap->ci; | |
1272 | struct inode *inode = &ci->vfs_inode; | |
12fe3dda | 1273 | struct ceph_buffer *old_blob = NULL; |
0ff8bfb3 | 1274 | struct cap_msg_args arg; |
bb0581f0 | 1275 | int held, revoking; |
a8599bd8 | 1276 | int wake = 0; |
a8599bd8 | 1277 | int delayed = 0; |
a8599bd8 SW |
1278 | int ret; |
1279 | ||
68c28323 SW |
1280 | held = cap->issued | cap->implemented; |
1281 | revoking = cap->implemented & ~cap->issued; | |
1282 | retain &= ~revoking; | |
68c28323 | 1283 | |
a8599bd8 SW |
1284 | dout("__send_cap %p cap %p session %p %s -> %s (revoking %s)\n", |
1285 | inode, cap, cap->session, | |
1286 | ceph_cap_string(held), ceph_cap_string(held & retain), | |
1287 | ceph_cap_string(revoking)); | |
1288 | BUG_ON((retain & CEPH_CAP_PIN) == 0); | |
1289 | ||
0ff8bfb3 | 1290 | arg.session = cap->session; |
a8599bd8 SW |
1291 | |
1292 | /* don't release wanted unless we've waited a bit. */ | |
1293 | if ((ci->i_ceph_flags & CEPH_I_NODELAY) == 0 && | |
1294 | time_before(jiffies, ci->i_hold_caps_min)) { | |
1295 | dout(" delaying issued %s -> %s, wanted %s -> %s on send\n", | |
1296 | ceph_cap_string(cap->issued), | |
1297 | ceph_cap_string(cap->issued & retain), | |
1298 | ceph_cap_string(cap->mds_wanted), | |
1299 | ceph_cap_string(want)); | |
1300 | want |= cap->mds_wanted; | |
1301 | retain |= cap->issued; | |
1302 | delayed = 1; | |
1303 | } | |
1304 | ci->i_ceph_flags &= ~(CEPH_I_NODELAY | CEPH_I_FLUSH); | |
eb65b919 YZ |
1305 | if (want & ~cap->mds_wanted) { |
1306 | /* user space may open/close single file frequently. | |
1307 | * This avoids droping mds_wanted immediately after | |
1308 | * requesting new mds_wanted. | |
1309 | */ | |
1310 | __cap_set_timeouts(mdsc, ci); | |
1311 | } | |
a8599bd8 SW |
1312 | |
1313 | cap->issued &= retain; /* drop bits we don't want */ | |
1314 | if (cap->implemented & ~cap->issued) { | |
1315 | /* | |
1316 | * Wake up any waiters on wanted -> needed transition. | |
1317 | * This is due to the weird transition from buffered | |
1318 | * to sync IO... we need to flush dirty pages _before_ | |
1319 | * allowing sync writes to avoid reordering. | |
1320 | */ | |
1321 | wake = 1; | |
1322 | } | |
1323 | cap->implemented &= cap->issued | used; | |
1324 | cap->mds_wanted = want; | |
1325 | ||
0ff8bfb3 JL |
1326 | arg.ino = ceph_vino(inode).ino; |
1327 | arg.cid = cap->cap_id; | |
1328 | arg.follows = flushing ? ci->i_head_snapc->seq : 0; | |
1329 | arg.flush_tid = flush_tid; | |
1330 | arg.oldest_flush_tid = oldest_flush_tid; | |
1331 | ||
1332 | arg.size = inode->i_size; | |
1333 | ci->i_reported_size = arg.size; | |
1334 | arg.max_size = ci->i_wanted_max_size; | |
1335 | ci->i_requested_max_size = arg.max_size; | |
a8599bd8 | 1336 | |
082afec9 | 1337 | if (flushing & CEPH_CAP_XATTR_EXCL) { |
12fe3dda | 1338 | old_blob = __ceph_build_xattrs_blob(ci); |
0ff8bfb3 JL |
1339 | arg.xattr_version = ci->i_xattrs.version; |
1340 | arg.xattr_buf = ci->i_xattrs.blob; | |
1341 | } else { | |
1342 | arg.xattr_buf = NULL; | |
a8599bd8 SW |
1343 | } |
1344 | ||
9bbeab41 AB |
1345 | arg.mtime = inode->i_mtime; |
1346 | arg.atime = inode->i_atime; | |
1347 | arg.ctime = inode->i_ctime; | |
ec62b894 | 1348 | arg.btime = ci->i_btime; |
176c77c9 | 1349 | arg.change_attr = inode_peek_iversion_raw(inode); |
0ff8bfb3 JL |
1350 | |
1351 | arg.op = op; | |
1352 | arg.caps = cap->implemented; | |
1353 | arg.wanted = want; | |
1354 | arg.dirty = flushing; | |
1355 | ||
1356 | arg.seq = cap->seq; | |
1357 | arg.issue_seq = cap->issue_seq; | |
1358 | arg.mseq = cap->mseq; | |
1359 | arg.time_warp_seq = ci->i_time_warp_seq; | |
1360 | ||
1361 | arg.uid = inode->i_uid; | |
1362 | arg.gid = inode->i_gid; | |
1363 | arg.mode = inode->i_mode; | |
1364 | ||
1365 | arg.inline_data = ci->i_inline_version != CEPH_INLINE_NONE; | |
49ada6e8 YZ |
1366 | if (!(flags & CEPH_CLIENT_CAPS_PENDING_CAPSNAP) && |
1367 | !list_empty(&ci->i_cap_snaps)) { | |
1368 | struct ceph_cap_snap *capsnap; | |
1369 | list_for_each_entry_reverse(capsnap, &ci->i_cap_snaps, ci_item) { | |
1370 | if (capsnap->cap_flush.tid) | |
1371 | break; | |
1372 | if (capsnap->need_flush) { | |
1373 | flags |= CEPH_CLIENT_CAPS_PENDING_CAPSNAP; | |
1374 | break; | |
1375 | } | |
1376 | } | |
1377 | } | |
1378 | arg.flags = flags; | |
e20d258d | 1379 | |
be655596 | 1380 | spin_unlock(&ci->i_ceph_lock); |
a8599bd8 | 1381 | |
12fe3dda LH |
1382 | ceph_buffer_put(old_blob); |
1383 | ||
0ff8bfb3 | 1384 | ret = send_cap_msg(&arg); |
a8599bd8 SW |
1385 | if (ret < 0) { |
1386 | dout("error sending cap msg, must requeue %p\n", inode); | |
1387 | delayed = 1; | |
1388 | } | |
1389 | ||
1390 | if (wake) | |
03066f23 | 1391 | wake_up_all(&ci->i_cap_wq); |
a8599bd8 SW |
1392 | |
1393 | return delayed; | |
1394 | } | |
1395 | ||
0e294387 YZ |
1396 | static inline int __send_flush_snap(struct inode *inode, |
1397 | struct ceph_mds_session *session, | |
1398 | struct ceph_cap_snap *capsnap, | |
1399 | u32 mseq, u64 oldest_flush_tid) | |
1400 | { | |
0ff8bfb3 JL |
1401 | struct cap_msg_args arg; |
1402 | ||
1403 | arg.session = session; | |
1404 | arg.ino = ceph_vino(inode).ino; | |
1405 | arg.cid = 0; | |
1406 | arg.follows = capsnap->follows; | |
1407 | arg.flush_tid = capsnap->cap_flush.tid; | |
1408 | arg.oldest_flush_tid = oldest_flush_tid; | |
1409 | ||
1410 | arg.size = capsnap->size; | |
1411 | arg.max_size = 0; | |
1412 | arg.xattr_version = capsnap->xattr_version; | |
1413 | arg.xattr_buf = capsnap->xattr_blob; | |
1414 | ||
1415 | arg.atime = capsnap->atime; | |
1416 | arg.mtime = capsnap->mtime; | |
1417 | arg.ctime = capsnap->ctime; | |
ec62b894 | 1418 | arg.btime = capsnap->btime; |
176c77c9 | 1419 | arg.change_attr = capsnap->change_attr; |
0ff8bfb3 JL |
1420 | |
1421 | arg.op = CEPH_CAP_OP_FLUSHSNAP; | |
1422 | arg.caps = capsnap->issued; | |
1423 | arg.wanted = 0; | |
1424 | arg.dirty = capsnap->dirty; | |
1425 | ||
1426 | arg.seq = 0; | |
1427 | arg.issue_seq = 0; | |
1428 | arg.mseq = mseq; | |
1429 | arg.time_warp_seq = capsnap->time_warp_seq; | |
1430 | ||
1431 | arg.uid = capsnap->uid; | |
1432 | arg.gid = capsnap->gid; | |
1433 | arg.mode = capsnap->mode; | |
1434 | ||
1435 | arg.inline_data = capsnap->inline_data; | |
1e4ef0c6 | 1436 | arg.flags = 0; |
0ff8bfb3 JL |
1437 | |
1438 | return send_cap_msg(&arg); | |
0e294387 YZ |
1439 | } |
1440 | ||
a8599bd8 SW |
1441 | /* |
1442 | * When a snapshot is taken, clients accumulate dirty metadata on | |
1443 | * inodes with capabilities in ceph_cap_snaps to describe the file | |
1444 | * state at the time the snapshot was taken. This must be flushed | |
1445 | * asynchronously back to the MDS once sync writes complete and dirty | |
1446 | * data is written out. | |
1447 | * | |
be655596 | 1448 | * Called under i_ceph_lock. Takes s_mutex as needed. |
a8599bd8 | 1449 | */ |
ed9b430c YZ |
1450 | static void __ceph_flush_snaps(struct ceph_inode_info *ci, |
1451 | struct ceph_mds_session *session) | |
be655596 SW |
1452 | __releases(ci->i_ceph_lock) |
1453 | __acquires(ci->i_ceph_lock) | |
a8599bd8 SW |
1454 | { |
1455 | struct inode *inode = &ci->vfs_inode; | |
ed9b430c | 1456 | struct ceph_mds_client *mdsc = session->s_mdsc; |
a8599bd8 | 1457 | struct ceph_cap_snap *capsnap; |
ed9b430c YZ |
1458 | u64 oldest_flush_tid = 0; |
1459 | u64 first_tid = 1, last_tid = 0; | |
a8599bd8 | 1460 | |
ed9b430c | 1461 | dout("__flush_snaps %p session %p\n", inode, session); |
a8599bd8 | 1462 | |
a8599bd8 | 1463 | list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) { |
a8599bd8 SW |
1464 | /* |
1465 | * we need to wait for sync writes to complete and for dirty | |
1466 | * pages to be written out. | |
1467 | */ | |
1468 | if (capsnap->dirty_pages || capsnap->writing) | |
cfc0bf66 | 1469 | break; |
a8599bd8 | 1470 | |
86056090 YZ |
1471 | /* should be removed by ceph_try_drop_cap_snap() */ |
1472 | BUG_ON(!capsnap->need_flush); | |
819ccbfa | 1473 | |
e835124c | 1474 | /* only flush each capsnap once */ |
0e294387 | 1475 | if (capsnap->cap_flush.tid > 0) { |
ed9b430c | 1476 | dout(" already flushed %p, skipping\n", capsnap); |
e835124c SW |
1477 | continue; |
1478 | } | |
1479 | ||
553adfd9 | 1480 | spin_lock(&mdsc->cap_dirty_lock); |
0e294387 YZ |
1481 | capsnap->cap_flush.tid = ++mdsc->last_cap_flush_tid; |
1482 | list_add_tail(&capsnap->cap_flush.g_list, | |
1483 | &mdsc->cap_flush_list); | |
ed9b430c YZ |
1484 | if (oldest_flush_tid == 0) |
1485 | oldest_flush_tid = __get_oldest_flush_tid(mdsc); | |
0e294387 YZ |
1486 | if (list_empty(&ci->i_flushing_item)) { |
1487 | list_add_tail(&ci->i_flushing_item, | |
1488 | &session->s_cap_flushing); | |
1489 | } | |
553adfd9 YZ |
1490 | spin_unlock(&mdsc->cap_dirty_lock); |
1491 | ||
0e294387 YZ |
1492 | list_add_tail(&capsnap->cap_flush.i_list, |
1493 | &ci->i_cap_flush_list); | |
1494 | ||
ed9b430c YZ |
1495 | if (first_tid == 1) |
1496 | first_tid = capsnap->cap_flush.tid; | |
1497 | last_tid = capsnap->cap_flush.tid; | |
1498 | } | |
1499 | ||
1500 | ci->i_ceph_flags &= ~CEPH_I_FLUSH_SNAPS; | |
1501 | ||
1502 | while (first_tid <= last_tid) { | |
1503 | struct ceph_cap *cap = ci->i_auth_cap; | |
1504 | struct ceph_cap_flush *cf; | |
1505 | int ret; | |
1506 | ||
1507 | if (!(cap && cap->session == session)) { | |
1508 | dout("__flush_snaps %p auth cap %p not mds%d, " | |
1509 | "stop\n", inode, cap, session->s_mds); | |
1510 | break; | |
1511 | } | |
1512 | ||
1513 | ret = -ENOENT; | |
1514 | list_for_each_entry(cf, &ci->i_cap_flush_list, i_list) { | |
1515 | if (cf->tid >= first_tid) { | |
1516 | ret = 0; | |
1517 | break; | |
1518 | } | |
1519 | } | |
1520 | if (ret < 0) | |
1521 | break; | |
1522 | ||
1523 | first_tid = cf->tid + 1; | |
1524 | ||
1525 | capsnap = container_of(cf, struct ceph_cap_snap, cap_flush); | |
805692d0 | 1526 | refcount_inc(&capsnap->nref); |
be655596 | 1527 | spin_unlock(&ci->i_ceph_lock); |
a8599bd8 | 1528 | |
ed9b430c YZ |
1529 | dout("__flush_snaps %p capsnap %p tid %llu %s\n", |
1530 | inode, capsnap, cf->tid, ceph_cap_string(capsnap->dirty)); | |
a8599bd8 | 1531 | |
ed9b430c YZ |
1532 | ret = __send_flush_snap(inode, session, capsnap, cap->mseq, |
1533 | oldest_flush_tid); | |
1534 | if (ret < 0) { | |
1535 | pr_err("__flush_snaps: error sending cap flushsnap, " | |
1536 | "ino (%llx.%llx) tid %llu follows %llu\n", | |
1537 | ceph_vinop(inode), cf->tid, capsnap->follows); | |
1538 | } | |
a8599bd8 | 1539 | |
ed9b430c | 1540 | ceph_put_cap_snap(capsnap); |
be655596 | 1541 | spin_lock(&ci->i_ceph_lock); |
a8599bd8 | 1542 | } |
ed9b430c | 1543 | } |
a8599bd8 | 1544 | |
ed9b430c YZ |
1545 | void ceph_flush_snaps(struct ceph_inode_info *ci, |
1546 | struct ceph_mds_session **psession) | |
1547 | { | |
1548 | struct inode *inode = &ci->vfs_inode; | |
1549 | struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc; | |
e4d2b16a | 1550 | struct ceph_mds_session *session = NULL; |
ed9b430c | 1551 | int mds; |
e4d2b16a | 1552 | |
ed9b430c | 1553 | dout("ceph_flush_snaps %p\n", inode); |
e4d2b16a YZ |
1554 | if (psession) |
1555 | session = *psession; | |
ed9b430c YZ |
1556 | retry: |
1557 | spin_lock(&ci->i_ceph_lock); | |
1558 | if (!(ci->i_ceph_flags & CEPH_I_FLUSH_SNAPS)) { | |
1559 | dout(" no capsnap needs flush, doing nothing\n"); | |
1560 | goto out; | |
1561 | } | |
1562 | if (!ci->i_auth_cap) { | |
1563 | dout(" no auth cap (migrating?), doing nothing\n"); | |
1564 | goto out; | |
1565 | } | |
a8599bd8 | 1566 | |
ed9b430c YZ |
1567 | mds = ci->i_auth_cap->session->s_mds; |
1568 | if (session && session->s_mds != mds) { | |
1569 | dout(" oops, wrong session %p mutex\n", session); | |
a8599bd8 SW |
1570 | mutex_unlock(&session->s_mutex); |
1571 | ceph_put_mds_session(session); | |
ed9b430c YZ |
1572 | session = NULL; |
1573 | } | |
1574 | if (!session) { | |
1575 | spin_unlock(&ci->i_ceph_lock); | |
1576 | mutex_lock(&mdsc->mutex); | |
1577 | session = __ceph_lookup_mds_session(mdsc, mds); | |
1578 | mutex_unlock(&mdsc->mutex); | |
1579 | if (session) { | |
1580 | dout(" inverting session/ino locks on %p\n", session); | |
1581 | mutex_lock(&session->s_mutex); | |
1582 | } | |
1583 | goto retry; | |
a8599bd8 | 1584 | } |
a8599bd8 | 1585 | |
24d063ac | 1586 | // make sure flushsnap messages are sent in proper order. |
054f8d41 | 1587 | if (ci->i_ceph_flags & CEPH_I_KICK_FLUSH) |
24d063ac | 1588 | __kick_flushing_caps(mdsc, session, ci, 0); |
24d063ac | 1589 | |
ed9b430c YZ |
1590 | __ceph_flush_snaps(ci, session); |
1591 | out: | |
be655596 | 1592 | spin_unlock(&ci->i_ceph_lock); |
ed9b430c YZ |
1593 | |
1594 | if (psession) { | |
1595 | *psession = session; | |
c858a070 | 1596 | } else if (session) { |
ed9b430c YZ |
1597 | mutex_unlock(&session->s_mutex); |
1598 | ceph_put_mds_session(session); | |
1599 | } | |
1600 | /* we flushed them all; remove this inode from the queue */ | |
1601 | spin_lock(&mdsc->snap_flush_lock); | |
1602 | list_del_init(&ci->i_snap_flush_item); | |
1603 | spin_unlock(&mdsc->snap_flush_lock); | |
a8599bd8 SW |
1604 | } |
1605 | ||
76e3b390 | 1606 | /* |
fca65b4a SW |
1607 | * Mark caps dirty. If inode is newly dirty, return the dirty flags. |
1608 | * Caller is then responsible for calling __mark_inode_dirty with the | |
1609 | * returned flags value. | |
76e3b390 | 1610 | */ |
f66fd9f0 YZ |
1611 | int __ceph_mark_dirty_caps(struct ceph_inode_info *ci, int mask, |
1612 | struct ceph_cap_flush **pcf) | |
76e3b390 | 1613 | { |
640ef79d | 1614 | struct ceph_mds_client *mdsc = |
3d14c5d2 | 1615 | ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc; |
76e3b390 SW |
1616 | struct inode *inode = &ci->vfs_inode; |
1617 | int was = ci->i_dirty_caps; | |
1618 | int dirty = 0; | |
1619 | ||
571ade33 YZ |
1620 | if (!ci->i_auth_cap) { |
1621 | pr_warn("__mark_dirty_caps %p %llx mask %s, " | |
1622 | "but no auth cap (session was closed?)\n", | |
1623 | inode, ceph_ino(inode), ceph_cap_string(mask)); | |
1624 | return 0; | |
1625 | } | |
1626 | ||
76e3b390 SW |
1627 | dout("__mark_dirty_caps %p %s dirty %s -> %s\n", &ci->vfs_inode, |
1628 | ceph_cap_string(mask), ceph_cap_string(was), | |
1629 | ceph_cap_string(was | mask)); | |
1630 | ci->i_dirty_caps |= mask; | |
1631 | if (was == 0) { | |
f66fd9f0 YZ |
1632 | WARN_ON_ONCE(ci->i_prealloc_cap_flush); |
1633 | swap(ci->i_prealloc_cap_flush, *pcf); | |
1634 | ||
604d1b02 YZ |
1635 | if (!ci->i_head_snapc) { |
1636 | WARN_ON_ONCE(!rwsem_is_locked(&mdsc->snap_rwsem)); | |
7d8cb26d SW |
1637 | ci->i_head_snapc = ceph_get_snap_context( |
1638 | ci->i_snap_realm->cached_context); | |
604d1b02 | 1639 | } |
0685235f YZ |
1640 | dout(" inode %p now dirty snapc %p auth cap %p\n", |
1641 | &ci->vfs_inode, ci->i_head_snapc, ci->i_auth_cap); | |
76e3b390 SW |
1642 | BUG_ON(!list_empty(&ci->i_dirty_item)); |
1643 | spin_lock(&mdsc->cap_dirty_lock); | |
11df2dfb | 1644 | list_add(&ci->i_dirty_item, &mdsc->cap_dirty); |
76e3b390 SW |
1645 | spin_unlock(&mdsc->cap_dirty_lock); |
1646 | if (ci->i_flushing_caps == 0) { | |
3772d26d | 1647 | ihold(inode); |
76e3b390 SW |
1648 | dirty |= I_DIRTY_SYNC; |
1649 | } | |
f66fd9f0 YZ |
1650 | } else { |
1651 | WARN_ON_ONCE(!ci->i_prealloc_cap_flush); | |
76e3b390 SW |
1652 | } |
1653 | BUG_ON(list_empty(&ci->i_dirty_item)); | |
1654 | if (((was | ci->i_flushing_caps) & CEPH_CAP_FILE_BUFFER) && | |
1655 | (mask & CEPH_CAP_FILE_BUFFER)) | |
1656 | dirty |= I_DIRTY_DATASYNC; | |
66802884 | 1657 | __cap_delay_requeue(mdsc, ci, true); |
fca65b4a | 1658 | return dirty; |
76e3b390 SW |
1659 | } |
1660 | ||
f66fd9f0 YZ |
1661 | struct ceph_cap_flush *ceph_alloc_cap_flush(void) |
1662 | { | |
1663 | return kmem_cache_alloc(ceph_cap_flush_cachep, GFP_KERNEL); | |
1664 | } | |
1665 | ||
1666 | void ceph_free_cap_flush(struct ceph_cap_flush *cf) | |
1667 | { | |
1668 | if (cf) | |
1669 | kmem_cache_free(ceph_cap_flush_cachep, cf); | |
1670 | } | |
1671 | ||
a2971c8c YZ |
1672 | static u64 __get_oldest_flush_tid(struct ceph_mds_client *mdsc) |
1673 | { | |
e4500b5e | 1674 | if (!list_empty(&mdsc->cap_flush_list)) { |
a2971c8c | 1675 | struct ceph_cap_flush *cf = |
e4500b5e YZ |
1676 | list_first_entry(&mdsc->cap_flush_list, |
1677 | struct ceph_cap_flush, g_list); | |
a2971c8c YZ |
1678 | return cf->tid; |
1679 | } | |
1680 | return 0; | |
1681 | } | |
1682 | ||
c8799fc4 YZ |
1683 | /* |
1684 | * Remove cap_flush from the mdsc's or inode's flushing cap list. | |
1685 | * Return true if caller needs to wake up flush waiters. | |
1686 | */ | |
1687 | static bool __finish_cap_flush(struct ceph_mds_client *mdsc, | |
1688 | struct ceph_inode_info *ci, | |
1689 | struct ceph_cap_flush *cf) | |
1690 | { | |
1691 | struct ceph_cap_flush *prev; | |
1692 | bool wake = cf->wake; | |
1693 | if (mdsc) { | |
1694 | /* are there older pending cap flushes? */ | |
1695 | if (wake && cf->g_list.prev != &mdsc->cap_flush_list) { | |
1696 | prev = list_prev_entry(cf, g_list); | |
1697 | prev->wake = true; | |
1698 | wake = false; | |
1699 | } | |
1700 | list_del(&cf->g_list); | |
1701 | } else if (ci) { | |
1702 | if (wake && cf->i_list.prev != &ci->i_cap_flush_list) { | |
1703 | prev = list_prev_entry(cf, i_list); | |
1704 | prev->wake = true; | |
1705 | wake = false; | |
1706 | } | |
1707 | list_del(&cf->i_list); | |
1708 | } else { | |
1709 | BUG_ON(1); | |
1710 | } | |
1711 | return wake; | |
1712 | } | |
1713 | ||
a8599bd8 SW |
1714 | /* |
1715 | * Add dirty inode to the flushing list. Assigned a seq number so we | |
1716 | * can wait for caps to flush without starving. | |
cdc35f96 | 1717 | * |
9f3345d8 | 1718 | * Called under i_ceph_lock. Returns the flush tid. |
a8599bd8 | 1719 | */ |
9f3345d8 | 1720 | static u64 __mark_caps_flushing(struct inode *inode, |
c8799fc4 | 1721 | struct ceph_mds_session *session, bool wake, |
9f3345d8 | 1722 | u64 *oldest_flush_tid) |
a8599bd8 | 1723 | { |
3d14c5d2 | 1724 | struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc; |
a8599bd8 | 1725 | struct ceph_inode_info *ci = ceph_inode(inode); |
f66fd9f0 | 1726 | struct ceph_cap_flush *cf = NULL; |
cdc35f96 | 1727 | int flushing; |
50b885b9 | 1728 | |
cdc35f96 | 1729 | BUG_ON(ci->i_dirty_caps == 0); |
a8599bd8 | 1730 | BUG_ON(list_empty(&ci->i_dirty_item)); |
f66fd9f0 | 1731 | BUG_ON(!ci->i_prealloc_cap_flush); |
cdc35f96 SW |
1732 | |
1733 | flushing = ci->i_dirty_caps; | |
1734 | dout("__mark_caps_flushing flushing %s, flushing_caps %s -> %s\n", | |
1735 | ceph_cap_string(flushing), | |
1736 | ceph_cap_string(ci->i_flushing_caps), | |
1737 | ceph_cap_string(ci->i_flushing_caps | flushing)); | |
1738 | ci->i_flushing_caps |= flushing; | |
1739 | ci->i_dirty_caps = 0; | |
afcdaea3 | 1740 | dout(" inode %p now !dirty\n", inode); |
cdc35f96 | 1741 | |
f66fd9f0 | 1742 | swap(cf, ci->i_prealloc_cap_flush); |
553adfd9 | 1743 | cf->caps = flushing; |
c8799fc4 | 1744 | cf->wake = wake; |
553adfd9 | 1745 | |
a8599bd8 | 1746 | spin_lock(&mdsc->cap_dirty_lock); |
afcdaea3 SW |
1747 | list_del_init(&ci->i_dirty_item); |
1748 | ||
553adfd9 | 1749 | cf->tid = ++mdsc->last_cap_flush_tid; |
e4500b5e | 1750 | list_add_tail(&cf->g_list, &mdsc->cap_flush_list); |
a2971c8c | 1751 | *oldest_flush_tid = __get_oldest_flush_tid(mdsc); |
553adfd9 | 1752 | |
a8599bd8 SW |
1753 | if (list_empty(&ci->i_flushing_item)) { |
1754 | list_add_tail(&ci->i_flushing_item, &session->s_cap_flushing); | |
1755 | mdsc->num_cap_flushing++; | |
a8599bd8 SW |
1756 | } |
1757 | spin_unlock(&mdsc->cap_dirty_lock); | |
cdc35f96 | 1758 | |
e4500b5e | 1759 | list_add_tail(&cf->i_list, &ci->i_cap_flush_list); |
553adfd9 | 1760 | |
9f3345d8 | 1761 | return cf->tid; |
a8599bd8 SW |
1762 | } |
1763 | ||
5ecad6fd SW |
1764 | /* |
1765 | * try to invalidate mapping pages without blocking. | |
1766 | */ | |
5ecad6fd SW |
1767 | static int try_nonblocking_invalidate(struct inode *inode) |
1768 | { | |
1769 | struct ceph_inode_info *ci = ceph_inode(inode); | |
1770 | u32 invalidating_gen = ci->i_rdcache_gen; | |
1771 | ||
be655596 | 1772 | spin_unlock(&ci->i_ceph_lock); |
5ecad6fd | 1773 | invalidate_mapping_pages(&inode->i_data, 0, -1); |
be655596 | 1774 | spin_lock(&ci->i_ceph_lock); |
5ecad6fd | 1775 | |
18a38193 | 1776 | if (inode->i_data.nrpages == 0 && |
5ecad6fd SW |
1777 | invalidating_gen == ci->i_rdcache_gen) { |
1778 | /* success. */ | |
1779 | dout("try_nonblocking_invalidate %p success\n", inode); | |
cd045cb4 SW |
1780 | /* save any racing async invalidate some trouble */ |
1781 | ci->i_rdcache_revoking = ci->i_rdcache_gen - 1; | |
5ecad6fd SW |
1782 | return 0; |
1783 | } | |
1784 | dout("try_nonblocking_invalidate %p failed\n", inode); | |
1785 | return -1; | |
1786 | } | |
1787 | ||
efb0ca76 YZ |
1788 | bool __ceph_should_report_size(struct ceph_inode_info *ci) |
1789 | { | |
1790 | loff_t size = ci->vfs_inode.i_size; | |
1791 | /* mds will adjust max size according to the reported size */ | |
1792 | if (ci->i_flushing_caps & CEPH_CAP_FILE_WR) | |
1793 | return false; | |
1794 | if (size >= ci->i_max_size) | |
1795 | return true; | |
1796 | /* half of previous max_size increment has been used */ | |
1797 | if (ci->i_max_size > ci->i_reported_size && | |
1798 | (size << 1) >= ci->i_max_size + ci->i_reported_size) | |
1799 | return true; | |
1800 | return false; | |
1801 | } | |
1802 | ||
a8599bd8 SW |
1803 | /* |
1804 | * Swiss army knife function to examine currently used and wanted | |
1805 | * versus held caps. Release, flush, ack revoked caps to mds as | |
1806 | * appropriate. | |
1807 | * | |
1808 | * CHECK_CAPS_NODELAY - caller is delayed work and we should not delay | |
1809 | * cap release further. | |
1810 | * CHECK_CAPS_AUTHONLY - we should only check the auth cap | |
1811 | * CHECK_CAPS_FLUSH - we should flush any dirty caps immediately, without | |
1812 | * further delay. | |
1813 | */ | |
1814 | void ceph_check_caps(struct ceph_inode_info *ci, int flags, | |
1815 | struct ceph_mds_session *session) | |
1816 | { | |
3d14c5d2 YS |
1817 | struct ceph_fs_client *fsc = ceph_inode_to_client(&ci->vfs_inode); |
1818 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
a8599bd8 SW |
1819 | struct inode *inode = &ci->vfs_inode; |
1820 | struct ceph_cap *cap; | |
a2971c8c | 1821 | u64 flush_tid, oldest_flush_tid; |
395c312b | 1822 | int file_wanted, used, cap_used; |
a8599bd8 | 1823 | int took_snap_rwsem = 0; /* true if mdsc->snap_rwsem held */ |
cbd03635 | 1824 | int issued, implemented, want, retain, revoking, flushing = 0; |
a8599bd8 SW |
1825 | int mds = -1; /* keep track of how far we've gone through i_caps list |
1826 | to avoid an infinite loop on retry */ | |
1827 | struct rb_node *p; | |
0f439c74 YZ |
1828 | int delayed = 0, sent = 0; |
1829 | bool no_delay = flags & CHECK_CAPS_NODELAY; | |
3609404f | 1830 | bool queue_invalidate = false; |
3609404f | 1831 | bool tried_invalidate = false; |
a8599bd8 SW |
1832 | |
1833 | /* if we are unmounting, flush any unused caps immediately. */ | |
1834 | if (mdsc->stopping) | |
0f439c74 | 1835 | no_delay = true; |
a8599bd8 | 1836 | |
be655596 | 1837 | spin_lock(&ci->i_ceph_lock); |
a8599bd8 SW |
1838 | |
1839 | if (ci->i_ceph_flags & CEPH_I_FLUSH) | |
1840 | flags |= CHECK_CAPS_FLUSH; | |
1841 | ||
0f439c74 YZ |
1842 | if (!(flags & CHECK_CAPS_AUTHONLY) || |
1843 | (ci->i_auth_cap && __ceph_is_single_caps(ci))) | |
1844 | __cap_delay_cancel(mdsc, ci); | |
1845 | ||
a8599bd8 SW |
1846 | goto retry_locked; |
1847 | retry: | |
be655596 | 1848 | spin_lock(&ci->i_ceph_lock); |
a8599bd8 SW |
1849 | retry_locked: |
1850 | file_wanted = __ceph_caps_file_wanted(ci); | |
1851 | used = __ceph_caps_used(ci); | |
cbd03635 SW |
1852 | issued = __ceph_caps_issued(ci, &implemented); |
1853 | revoking = implemented & ~issued; | |
a8599bd8 | 1854 | |
41445999 YZ |
1855 | want = file_wanted; |
1856 | retain = file_wanted | used | CEPH_CAP_PIN; | |
a8599bd8 | 1857 | if (!mdsc->stopping && inode->i_nlink > 0) { |
41445999 | 1858 | if (file_wanted) { |
a8599bd8 | 1859 | retain |= CEPH_CAP_ANY; /* be greedy */ |
32ec4397 YZ |
1860 | } else if (S_ISDIR(inode->i_mode) && |
1861 | (issued & CEPH_CAP_FILE_SHARED) && | |
8a2ac3a8 | 1862 | __ceph_dir_is_complete(ci)) { |
32ec4397 YZ |
1863 | /* |
1864 | * If a directory is complete, we want to keep | |
1865 | * the exclusive cap. So that MDS does not end up | |
1866 | * revoking the shared cap on every create/unlink | |
1867 | * operation. | |
1868 | */ | |
8a2ac3a8 YZ |
1869 | if (IS_RDONLY(inode)) |
1870 | want = CEPH_CAP_ANY_SHARED; | |
1871 | else | |
1872 | want = CEPH_CAP_ANY_SHARED | CEPH_CAP_FILE_EXCL; | |
32ec4397 | 1873 | retain |= want; |
a8599bd8 | 1874 | } else { |
32ec4397 | 1875 | |
a8599bd8 SW |
1876 | retain |= CEPH_CAP_ANY_SHARED; |
1877 | /* | |
1878 | * keep RD only if we didn't have the file open RW, | |
1879 | * because then the mds would revoke it anyway to | |
1880 | * journal max_size=0. | |
1881 | */ | |
1882 | if (ci->i_max_size == 0) | |
1883 | retain |= CEPH_CAP_ANY_RD; | |
1884 | } | |
1885 | } | |
1886 | ||
1887 | dout("check_caps %p file_want %s used %s dirty %s flushing %s" | |
cbd03635 | 1888 | " issued %s revoking %s retain %s %s%s%s\n", inode, |
a8599bd8 SW |
1889 | ceph_cap_string(file_wanted), |
1890 | ceph_cap_string(used), ceph_cap_string(ci->i_dirty_caps), | |
1891 | ceph_cap_string(ci->i_flushing_caps), | |
cbd03635 | 1892 | ceph_cap_string(issued), ceph_cap_string(revoking), |
a8599bd8 SW |
1893 | ceph_cap_string(retain), |
1894 | (flags & CHECK_CAPS_AUTHONLY) ? " AUTHONLY" : "", | |
1895 | (flags & CHECK_CAPS_NODELAY) ? " NODELAY" : "", | |
1896 | (flags & CHECK_CAPS_FLUSH) ? " FLUSH" : ""); | |
1897 | ||
1898 | /* | |
1899 | * If we no longer need to hold onto old our caps, and we may | |
1900 | * have cached pages, but don't want them, then try to invalidate. | |
1901 | * If we fail, it's because pages are locked.... try again later. | |
1902 | */ | |
0f439c74 | 1903 | if ((!no_delay || mdsc->stopping) && |
fdd4e158 | 1904 | !S_ISDIR(inode->i_mode) && /* ignore readdir cache */ |
9abd4db7 | 1905 | !(ci->i_wb_ref || ci->i_wrbuffer_ref) && /* no dirty pages... */ |
fdd4e158 | 1906 | inode->i_data.nrpages && /* have cached pages */ |
5e804ac4 YZ |
1907 | (revoking & (CEPH_CAP_FILE_CACHE| |
1908 | CEPH_CAP_FILE_LAZYIO)) && /* or revoking cache */ | |
a8599bd8 | 1909 | !tried_invalidate) { |
a8599bd8 | 1910 | dout("check_caps trying to invalidate on %p\n", inode); |
5ecad6fd | 1911 | if (try_nonblocking_invalidate(inode) < 0) { |
ee612d95 YZ |
1912 | dout("check_caps queuing invalidate\n"); |
1913 | queue_invalidate = true; | |
1914 | ci->i_rdcache_revoking = ci->i_rdcache_gen; | |
a8599bd8 | 1915 | } |
3609404f | 1916 | tried_invalidate = true; |
a8599bd8 SW |
1917 | goto retry_locked; |
1918 | } | |
1919 | ||
a8599bd8 SW |
1920 | for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) { |
1921 | cap = rb_entry(p, struct ceph_cap, ci_node); | |
a8599bd8 SW |
1922 | |
1923 | /* avoid looping forever */ | |
1924 | if (mds >= cap->mds || | |
1925 | ((flags & CHECK_CAPS_AUTHONLY) && cap != ci->i_auth_cap)) | |
1926 | continue; | |
1927 | ||
1928 | /* NOTE: no side-effects allowed, until we take s_mutex */ | |
1929 | ||
395c312b YZ |
1930 | cap_used = used; |
1931 | if (ci->i_auth_cap && cap != ci->i_auth_cap) | |
1932 | cap_used &= ~ci->i_auth_cap->issued; | |
1933 | ||
a8599bd8 | 1934 | revoking = cap->implemented & ~cap->issued; |
395c312b | 1935 | dout(" mds%d cap %p used %s issued %s implemented %s revoking %s\n", |
9abd4db7 YZ |
1936 | cap->mds, cap, ceph_cap_string(cap_used), |
1937 | ceph_cap_string(cap->issued), | |
088b3f5e SW |
1938 | ceph_cap_string(cap->implemented), |
1939 | ceph_cap_string(revoking)); | |
a8599bd8 SW |
1940 | |
1941 | if (cap == ci->i_auth_cap && | |
1942 | (cap->issued & CEPH_CAP_FILE_WR)) { | |
1943 | /* request larger max_size from MDS? */ | |
1944 | if (ci->i_wanted_max_size > ci->i_max_size && | |
1945 | ci->i_wanted_max_size > ci->i_requested_max_size) { | |
1946 | dout("requesting new max_size\n"); | |
1947 | goto ack; | |
1948 | } | |
1949 | ||
1950 | /* approaching file_max? */ | |
efb0ca76 | 1951 | if (__ceph_should_report_size(ci)) { |
a8599bd8 SW |
1952 | dout("i_size approaching max_size\n"); |
1953 | goto ack; | |
1954 | } | |
1955 | } | |
1956 | /* flush anything dirty? */ | |
7bc00fdd YZ |
1957 | if (cap == ci->i_auth_cap) { |
1958 | if ((flags & CHECK_CAPS_FLUSH) && ci->i_dirty_caps) { | |
1959 | dout("flushing dirty caps\n"); | |
1960 | goto ack; | |
1961 | } | |
1962 | if (ci->i_ceph_flags & CEPH_I_FLUSH_SNAPS) { | |
1963 | dout("flushing snap caps\n"); | |
1964 | goto ack; | |
1965 | } | |
a8599bd8 SW |
1966 | } |
1967 | ||
1968 | /* completed revocation? going down and there are no caps? */ | |
395c312b | 1969 | if (revoking && (revoking & cap_used) == 0) { |
a8599bd8 SW |
1970 | dout("completed revocation of %s\n", |
1971 | ceph_cap_string(cap->implemented & ~cap->issued)); | |
1972 | goto ack; | |
1973 | } | |
1974 | ||
1975 | /* want more caps from mds? */ | |
1976 | if (want & ~(cap->mds_wanted | cap->issued)) | |
1977 | goto ack; | |
1978 | ||
1979 | /* things we might delay */ | |
fdac94fa | 1980 | if ((cap->issued & ~retain) == 0) |
a8599bd8 SW |
1981 | continue; /* nope, all good */ |
1982 | ||
0f439c74 | 1983 | if (no_delay) |
a8599bd8 SW |
1984 | goto ack; |
1985 | ||
1986 | /* delay? */ | |
1987 | if ((ci->i_ceph_flags & CEPH_I_NODELAY) == 0 && | |
1988 | time_before(jiffies, ci->i_hold_caps_max)) { | |
1989 | dout(" delaying issued %s -> %s, wanted %s -> %s\n", | |
1990 | ceph_cap_string(cap->issued), | |
1991 | ceph_cap_string(cap->issued & retain), | |
1992 | ceph_cap_string(cap->mds_wanted), | |
1993 | ceph_cap_string(want)); | |
1994 | delayed++; | |
1995 | continue; | |
1996 | } | |
1997 | ||
1998 | ack: | |
1999 | if (session && session != cap->session) { | |
2000 | dout("oops, wrong session %p mutex\n", session); | |
2001 | mutex_unlock(&session->s_mutex); | |
2002 | session = NULL; | |
2003 | } | |
2004 | if (!session) { | |
2005 | session = cap->session; | |
2006 | if (mutex_trylock(&session->s_mutex) == 0) { | |
2007 | dout("inverting session/ino locks on %p\n", | |
2008 | session); | |
be655596 | 2009 | spin_unlock(&ci->i_ceph_lock); |
a8599bd8 SW |
2010 | if (took_snap_rwsem) { |
2011 | up_read(&mdsc->snap_rwsem); | |
2012 | took_snap_rwsem = 0; | |
2013 | } | |
2014 | mutex_lock(&session->s_mutex); | |
2015 | goto retry; | |
2016 | } | |
2017 | } | |
7bc00fdd YZ |
2018 | |
2019 | /* kick flushing and flush snaps before sending normal | |
2020 | * cap message */ | |
2021 | if (cap == ci->i_auth_cap && | |
2022 | (ci->i_ceph_flags & | |
2023 | (CEPH_I_KICK_FLUSH | CEPH_I_FLUSH_SNAPS))) { | |
054f8d41 | 2024 | if (ci->i_ceph_flags & CEPH_I_KICK_FLUSH) |
24d063ac | 2025 | __kick_flushing_caps(mdsc, session, ci, 0); |
ed9b430c YZ |
2026 | if (ci->i_ceph_flags & CEPH_I_FLUSH_SNAPS) |
2027 | __ceph_flush_snaps(ci, session); | |
2028 | ||
7bc00fdd YZ |
2029 | goto retry_locked; |
2030 | } | |
2031 | ||
a8599bd8 SW |
2032 | /* take snap_rwsem after session mutex */ |
2033 | if (!took_snap_rwsem) { | |
2034 | if (down_read_trylock(&mdsc->snap_rwsem) == 0) { | |
2035 | dout("inverting snap/in locks on %p\n", | |
2036 | inode); | |
be655596 | 2037 | spin_unlock(&ci->i_ceph_lock); |
a8599bd8 SW |
2038 | down_read(&mdsc->snap_rwsem); |
2039 | took_snap_rwsem = 1; | |
2040 | goto retry; | |
2041 | } | |
2042 | took_snap_rwsem = 1; | |
2043 | } | |
2044 | ||
553adfd9 | 2045 | if (cap == ci->i_auth_cap && ci->i_dirty_caps) { |
9f3345d8 JL |
2046 | flushing = ci->i_dirty_caps; |
2047 | flush_tid = __mark_caps_flushing(inode, session, false, | |
2048 | &oldest_flush_tid); | |
553adfd9 | 2049 | } else { |
24be0c48 | 2050 | flushing = 0; |
553adfd9 | 2051 | flush_tid = 0; |
a2971c8c YZ |
2052 | spin_lock(&mdsc->cap_dirty_lock); |
2053 | oldest_flush_tid = __get_oldest_flush_tid(mdsc); | |
2054 | spin_unlock(&mdsc->cap_dirty_lock); | |
553adfd9 | 2055 | } |
a8599bd8 SW |
2056 | |
2057 | mds = cap->mds; /* remember mds, so we don't repeat */ | |
2058 | sent++; | |
2059 | ||
be655596 | 2060 | /* __send_cap drops i_ceph_lock */ |
49ada6e8 | 2061 | delayed += __send_cap(mdsc, cap, CEPH_CAP_OP_UPDATE, 0, |
1e4ef0c6 JL |
2062 | cap_used, want, retain, flushing, |
2063 | flush_tid, oldest_flush_tid); | |
be655596 | 2064 | goto retry; /* retake i_ceph_lock and restart our cap scan. */ |
a8599bd8 SW |
2065 | } |
2066 | ||
0f439c74 YZ |
2067 | /* Reschedule delayed caps release if we delayed anything */ |
2068 | if (delayed) | |
66802884 | 2069 | __cap_delay_requeue(mdsc, ci, false); |
a8599bd8 | 2070 | |
be655596 | 2071 | spin_unlock(&ci->i_ceph_lock); |
a8599bd8 | 2072 | |
cbd03635 | 2073 | if (queue_invalidate) |
3c6f6b79 | 2074 | ceph_queue_invalidate(inode); |
cbd03635 | 2075 | |
cdc2ce05 | 2076 | if (session) |
a8599bd8 SW |
2077 | mutex_unlock(&session->s_mutex); |
2078 | if (took_snap_rwsem) | |
2079 | up_read(&mdsc->snap_rwsem); | |
2080 | } | |
2081 | ||
a8599bd8 SW |
2082 | /* |
2083 | * Try to flush dirty caps back to the auth mds. | |
2084 | */ | |
553adfd9 | 2085 | static int try_flush_caps(struct inode *inode, u64 *ptid) |
a8599bd8 | 2086 | { |
3d14c5d2 | 2087 | struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc; |
a8599bd8 | 2088 | struct ceph_inode_info *ci = ceph_inode(inode); |
4fe59789 | 2089 | struct ceph_mds_session *session = NULL; |
89b52fe1 | 2090 | int flushing = 0; |
a2971c8c | 2091 | u64 flush_tid = 0, oldest_flush_tid = 0; |
a8599bd8 SW |
2092 | |
2093 | retry: | |
be655596 | 2094 | spin_lock(&ci->i_ceph_lock); |
d6cee9db | 2095 | retry_locked: |
a8599bd8 SW |
2096 | if (ci->i_dirty_caps && ci->i_auth_cap) { |
2097 | struct ceph_cap *cap = ci->i_auth_cap; | |
a8599bd8 SW |
2098 | int delayed; |
2099 | ||
27b0a392 | 2100 | if (session != cap->session) { |
be655596 | 2101 | spin_unlock(&ci->i_ceph_lock); |
4fe59789 YZ |
2102 | if (session) |
2103 | mutex_unlock(&session->s_mutex); | |
a8599bd8 SW |
2104 | session = cap->session; |
2105 | mutex_lock(&session->s_mutex); | |
2106 | goto retry; | |
2107 | } | |
6c2838fb JL |
2108 | if (cap->session->s_state < CEPH_MDS_SESSION_OPEN) { |
2109 | spin_unlock(&ci->i_ceph_lock); | |
a8599bd8 | 2110 | goto out; |
6c2838fb | 2111 | } |
a8599bd8 | 2112 | |
d6cee9db YZ |
2113 | if (ci->i_ceph_flags & |
2114 | (CEPH_I_KICK_FLUSH | CEPH_I_FLUSH_SNAPS)) { | |
2115 | if (ci->i_ceph_flags & CEPH_I_KICK_FLUSH) | |
2116 | __kick_flushing_caps(mdsc, session, ci, 0); | |
2117 | if (ci->i_ceph_flags & CEPH_I_FLUSH_SNAPS) | |
2118 | __ceph_flush_snaps(ci, session); | |
2119 | goto retry_locked; | |
2120 | } | |
2121 | ||
9f3345d8 JL |
2122 | flushing = ci->i_dirty_caps; |
2123 | flush_tid = __mark_caps_flushing(inode, session, true, | |
2124 | &oldest_flush_tid); | |
a8599bd8 | 2125 | |
be655596 | 2126 | /* __send_cap drops i_ceph_lock */ |
49ada6e8 YZ |
2127 | delayed = __send_cap(mdsc, cap, CEPH_CAP_OP_FLUSH, |
2128 | CEPH_CLIENT_CAPS_SYNC, | |
d6cee9db YZ |
2129 | __ceph_caps_used(ci), |
2130 | __ceph_caps_wanted(ci), | |
2131 | (cap->issued | cap->implemented), | |
2132 | flushing, flush_tid, oldest_flush_tid); | |
a8599bd8 | 2133 | |
553adfd9 YZ |
2134 | if (delayed) { |
2135 | spin_lock(&ci->i_ceph_lock); | |
66802884 | 2136 | __cap_delay_requeue(mdsc, ci, true); |
553adfd9 YZ |
2137 | spin_unlock(&ci->i_ceph_lock); |
2138 | } | |
2139 | } else { | |
e4500b5e | 2140 | if (!list_empty(&ci->i_cap_flush_list)) { |
553adfd9 | 2141 | struct ceph_cap_flush *cf = |
e4500b5e | 2142 | list_last_entry(&ci->i_cap_flush_list, |
c8799fc4 YZ |
2143 | struct ceph_cap_flush, i_list); |
2144 | cf->wake = true; | |
553adfd9 YZ |
2145 | flush_tid = cf->tid; |
2146 | } | |
2147 | flushing = ci->i_flushing_caps; | |
2148 | spin_unlock(&ci->i_ceph_lock); | |
a8599bd8 SW |
2149 | } |
2150 | out: | |
4fe59789 | 2151 | if (session) |
a8599bd8 | 2152 | mutex_unlock(&session->s_mutex); |
553adfd9 YZ |
2153 | |
2154 | *ptid = flush_tid; | |
a8599bd8 SW |
2155 | return flushing; |
2156 | } | |
2157 | ||
2158 | /* | |
2159 | * Return true if we've flushed caps through the given flush_tid. | |
2160 | */ | |
553adfd9 | 2161 | static int caps_are_flushed(struct inode *inode, u64 flush_tid) |
a8599bd8 SW |
2162 | { |
2163 | struct ceph_inode_info *ci = ceph_inode(inode); | |
553adfd9 | 2164 | int ret = 1; |
a8599bd8 | 2165 | |
be655596 | 2166 | spin_lock(&ci->i_ceph_lock); |
e4500b5e YZ |
2167 | if (!list_empty(&ci->i_cap_flush_list)) { |
2168 | struct ceph_cap_flush * cf = | |
2169 | list_first_entry(&ci->i_cap_flush_list, | |
2170 | struct ceph_cap_flush, i_list); | |
553adfd9 | 2171 | if (cf->tid <= flush_tid) |
a8599bd8 | 2172 | ret = 0; |
89b52fe1 | 2173 | } |
be655596 | 2174 | spin_unlock(&ci->i_ceph_lock); |
a8599bd8 SW |
2175 | return ret; |
2176 | } | |
2177 | ||
da819c81 | 2178 | /* |
68cd5b4b | 2179 | * wait for any unsafe requests to complete. |
da819c81 | 2180 | */ |
68cd5b4b | 2181 | static int unsafe_request_wait(struct inode *inode) |
da819c81 YZ |
2182 | { |
2183 | struct ceph_inode_info *ci = ceph_inode(inode); | |
68cd5b4b YZ |
2184 | struct ceph_mds_request *req1 = NULL, *req2 = NULL; |
2185 | int ret, err = 0; | |
da819c81 YZ |
2186 | |
2187 | spin_lock(&ci->i_unsafe_lock); | |
68cd5b4b YZ |
2188 | if (S_ISDIR(inode->i_mode) && !list_empty(&ci->i_unsafe_dirops)) { |
2189 | req1 = list_last_entry(&ci->i_unsafe_dirops, | |
2190 | struct ceph_mds_request, | |
2191 | r_unsafe_dir_item); | |
2192 | ceph_mdsc_get_request(req1); | |
2193 | } | |
2194 | if (!list_empty(&ci->i_unsafe_iops)) { | |
2195 | req2 = list_last_entry(&ci->i_unsafe_iops, | |
2196 | struct ceph_mds_request, | |
2197 | r_unsafe_target_item); | |
2198 | ceph_mdsc_get_request(req2); | |
2199 | } | |
2200 | spin_unlock(&ci->i_unsafe_lock); | |
da819c81 | 2201 | |
4945a084 | 2202 | dout("unsafe_request_wait %p wait on tid %llu %llu\n", |
68cd5b4b YZ |
2203 | inode, req1 ? req1->r_tid : 0ULL, req2 ? req2->r_tid : 0ULL); |
2204 | if (req1) { | |
2205 | ret = !wait_for_completion_timeout(&req1->r_safe_completion, | |
2206 | ceph_timeout_jiffies(req1->r_timeout)); | |
da819c81 | 2207 | if (ret) |
68cd5b4b YZ |
2208 | err = -EIO; |
2209 | ceph_mdsc_put_request(req1); | |
2210 | } | |
2211 | if (req2) { | |
2212 | ret = !wait_for_completion_timeout(&req2->r_safe_completion, | |
2213 | ceph_timeout_jiffies(req2->r_timeout)); | |
2214 | if (ret) | |
2215 | err = -EIO; | |
2216 | ceph_mdsc_put_request(req2); | |
2217 | } | |
2218 | return err; | |
da819c81 YZ |
2219 | } |
2220 | ||
02c24a82 | 2221 | int ceph_fsync(struct file *file, loff_t start, loff_t end, int datasync) |
a8599bd8 | 2222 | { |
f4b97866 | 2223 | struct ceph_file_info *fi = file->private_data; |
7ea80859 | 2224 | struct inode *inode = file->f_mapping->host; |
a8599bd8 | 2225 | struct ceph_inode_info *ci = ceph_inode(inode); |
553adfd9 | 2226 | u64 flush_tid; |
f4b97866 | 2227 | int ret, err; |
a8599bd8 SW |
2228 | int dirty; |
2229 | ||
2230 | dout("fsync %p%s\n", inode, datasync ? " datasync" : ""); | |
9a5530c6 | 2231 | |
b74fceae | 2232 | ret = file_write_and_wait_range(file, start, end); |
da819c81 YZ |
2233 | if (datasync) |
2234 | goto out; | |
2235 | ||
553adfd9 | 2236 | dirty = try_flush_caps(inode, &flush_tid); |
a8599bd8 SW |
2237 | dout("fsync dirty caps are %s\n", ceph_cap_string(dirty)); |
2238 | ||
f4b97866 | 2239 | err = unsafe_request_wait(inode); |
da819c81 | 2240 | |
a8599bd8 SW |
2241 | /* |
2242 | * only wait on non-file metadata writeback (the mds | |
2243 | * can recover size and mtime, so we don't need to | |
2244 | * wait for that) | |
2245 | */ | |
f4b97866 YZ |
2246 | if (!err && (dirty & ~CEPH_CAP_ANY_FILE_WR)) { |
2247 | err = wait_event_interruptible(ci->i_cap_wq, | |
da819c81 | 2248 | caps_are_flushed(inode, flush_tid)); |
a8599bd8 | 2249 | } |
f4b97866 YZ |
2250 | |
2251 | if (err < 0) | |
2252 | ret = err; | |
2253 | ||
2254 | if (errseq_check(&ci->i_meta_err, READ_ONCE(fi->meta_err))) { | |
2255 | spin_lock(&file->f_lock); | |
2256 | err = errseq_check_and_advance(&ci->i_meta_err, | |
2257 | &fi->meta_err); | |
2258 | spin_unlock(&file->f_lock); | |
2259 | if (err < 0) | |
2260 | ret = err; | |
2261 | } | |
da819c81 YZ |
2262 | out: |
2263 | dout("fsync %p%s result=%d\n", inode, datasync ? " datasync" : "", ret); | |
a8599bd8 SW |
2264 | return ret; |
2265 | } | |
2266 | ||
2267 | /* | |
2268 | * Flush any dirty caps back to the mds. If we aren't asked to wait, | |
2269 | * queue inode for flush but don't do so immediately, because we can | |
2270 | * get by with fewer MDS messages if we wait for data writeback to | |
2271 | * complete first. | |
2272 | */ | |
f1a3d572 | 2273 | int ceph_write_inode(struct inode *inode, struct writeback_control *wbc) |
a8599bd8 SW |
2274 | { |
2275 | struct ceph_inode_info *ci = ceph_inode(inode); | |
553adfd9 | 2276 | u64 flush_tid; |
a8599bd8 SW |
2277 | int err = 0; |
2278 | int dirty; | |
16515a6d | 2279 | int wait = (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync); |
a8599bd8 SW |
2280 | |
2281 | dout("write_inode %p wait=%d\n", inode, wait); | |
2282 | if (wait) { | |
553adfd9 | 2283 | dirty = try_flush_caps(inode, &flush_tid); |
a8599bd8 SW |
2284 | if (dirty) |
2285 | err = wait_event_interruptible(ci->i_cap_wq, | |
2286 | caps_are_flushed(inode, flush_tid)); | |
2287 | } else { | |
640ef79d | 2288 | struct ceph_mds_client *mdsc = |
3d14c5d2 | 2289 | ceph_sb_to_client(inode->i_sb)->mdsc; |
a8599bd8 | 2290 | |
be655596 | 2291 | spin_lock(&ci->i_ceph_lock); |
a8599bd8 SW |
2292 | if (__ceph_caps_dirty(ci)) |
2293 | __cap_delay_requeue_front(mdsc, ci); | |
be655596 | 2294 | spin_unlock(&ci->i_ceph_lock); |
a8599bd8 SW |
2295 | } |
2296 | return err; | |
2297 | } | |
2298 | ||
0e294387 YZ |
2299 | static void __kick_flushing_caps(struct ceph_mds_client *mdsc, |
2300 | struct ceph_mds_session *session, | |
2301 | struct ceph_inode_info *ci, | |
2302 | u64 oldest_flush_tid) | |
2303 | __releases(ci->i_ceph_lock) | |
2304 | __acquires(ci->i_ceph_lock) | |
553adfd9 YZ |
2305 | { |
2306 | struct inode *inode = &ci->vfs_inode; | |
2307 | struct ceph_cap *cap; | |
2308 | struct ceph_cap_flush *cf; | |
0e294387 | 2309 | int ret; |
553adfd9 | 2310 | u64 first_tid = 0; |
49ada6e8 | 2311 | u64 last_snap_flush = 0; |
553adfd9 | 2312 | |
054f8d41 YZ |
2313 | ci->i_ceph_flags &= ~CEPH_I_KICK_FLUSH; |
2314 | ||
49ada6e8 YZ |
2315 | list_for_each_entry_reverse(cf, &ci->i_cap_flush_list, i_list) { |
2316 | if (!cf->caps) { | |
2317 | last_snap_flush = cf->tid; | |
2318 | break; | |
2319 | } | |
2320 | } | |
2321 | ||
e4500b5e YZ |
2322 | list_for_each_entry(cf, &ci->i_cap_flush_list, i_list) { |
2323 | if (cf->tid < first_tid) | |
2324 | continue; | |
2325 | ||
553adfd9 YZ |
2326 | cap = ci->i_auth_cap; |
2327 | if (!(cap && cap->session == session)) { | |
0e294387 YZ |
2328 | pr_err("%p auth cap %p not mds%d ???\n", |
2329 | inode, cap, session->s_mds); | |
553adfd9 YZ |
2330 | break; |
2331 | } | |
2332 | ||
553adfd9 YZ |
2333 | first_tid = cf->tid + 1; |
2334 | ||
0e294387 YZ |
2335 | if (cf->caps) { |
2336 | dout("kick_flushing_caps %p cap %p tid %llu %s\n", | |
2337 | inode, cap, cf->tid, ceph_cap_string(cf->caps)); | |
2338 | ci->i_ceph_flags |= CEPH_I_NODELAY; | |
49ada6e8 | 2339 | |
0e294387 | 2340 | ret = __send_cap(mdsc, cap, CEPH_CAP_OP_FLUSH, |
49ada6e8 YZ |
2341 | (cf->tid < last_snap_flush ? |
2342 | CEPH_CLIENT_CAPS_PENDING_CAPSNAP : 0), | |
2343 | __ceph_caps_used(ci), | |
0e294387 | 2344 | __ceph_caps_wanted(ci), |
49ada6e8 | 2345 | (cap->issued | cap->implemented), |
0e294387 YZ |
2346 | cf->caps, cf->tid, oldest_flush_tid); |
2347 | if (ret) { | |
2348 | pr_err("kick_flushing_caps: error sending " | |
2349 | "cap flush, ino (%llx.%llx) " | |
2350 | "tid %llu flushing %s\n", | |
2351 | ceph_vinop(inode), cf->tid, | |
2352 | ceph_cap_string(cf->caps)); | |
2353 | } | |
2354 | } else { | |
2355 | struct ceph_cap_snap *capsnap = | |
2356 | container_of(cf, struct ceph_cap_snap, | |
2357 | cap_flush); | |
2358 | dout("kick_flushing_caps %p capsnap %p tid %llu %s\n", | |
2359 | inode, capsnap, cf->tid, | |
2360 | ceph_cap_string(capsnap->dirty)); | |
2361 | ||
805692d0 | 2362 | refcount_inc(&capsnap->nref); |
0e294387 YZ |
2363 | spin_unlock(&ci->i_ceph_lock); |
2364 | ||
2365 | ret = __send_flush_snap(inode, session, capsnap, cap->mseq, | |
2366 | oldest_flush_tid); | |
2367 | if (ret < 0) { | |
2368 | pr_err("kick_flushing_caps: error sending " | |
2369 | "cap flushsnap, ino (%llx.%llx) " | |
2370 | "tid %llu follows %llu\n", | |
2371 | ceph_vinop(inode), cf->tid, | |
2372 | capsnap->follows); | |
2373 | } | |
2374 | ||
2375 | ceph_put_cap_snap(capsnap); | |
2376 | } | |
e4500b5e YZ |
2377 | |
2378 | spin_lock(&ci->i_ceph_lock); | |
553adfd9 | 2379 | } |
553adfd9 YZ |
2380 | } |
2381 | ||
e548e9b9 YZ |
2382 | void ceph_early_kick_flushing_caps(struct ceph_mds_client *mdsc, |
2383 | struct ceph_mds_session *session) | |
2384 | { | |
2385 | struct ceph_inode_info *ci; | |
2386 | struct ceph_cap *cap; | |
0e294387 | 2387 | u64 oldest_flush_tid; |
e548e9b9 YZ |
2388 | |
2389 | dout("early_kick_flushing_caps mds%d\n", session->s_mds); | |
0e294387 YZ |
2390 | |
2391 | spin_lock(&mdsc->cap_dirty_lock); | |
2392 | oldest_flush_tid = __get_oldest_flush_tid(mdsc); | |
2393 | spin_unlock(&mdsc->cap_dirty_lock); | |
2394 | ||
e548e9b9 YZ |
2395 | list_for_each_entry(ci, &session->s_cap_flushing, i_flushing_item) { |
2396 | spin_lock(&ci->i_ceph_lock); | |
2397 | cap = ci->i_auth_cap; | |
2398 | if (!(cap && cap->session == session)) { | |
2399 | pr_err("%p auth cap %p not mds%d ???\n", | |
2400 | &ci->vfs_inode, cap, session->s_mds); | |
2401 | spin_unlock(&ci->i_ceph_lock); | |
2402 | continue; | |
2403 | } | |
2404 | ||
2405 | ||
2406 | /* | |
2407 | * if flushing caps were revoked, we re-send the cap flush | |
2408 | * in client reconnect stage. This guarantees MDS * processes | |
2409 | * the cap flush message before issuing the flushing caps to | |
2410 | * other client. | |
2411 | */ | |
2412 | if ((cap->issued & ci->i_flushing_caps) != | |
2413 | ci->i_flushing_caps) { | |
81c5a148 YZ |
2414 | /* encode_caps_cb() also will reset these sequence |
2415 | * numbers. make sure sequence numbers in cap flush | |
2416 | * message match later reconnect message */ | |
2417 | cap->seq = 0; | |
2418 | cap->issue_seq = 0; | |
2419 | cap->mseq = 0; | |
0e294387 YZ |
2420 | __kick_flushing_caps(mdsc, session, ci, |
2421 | oldest_flush_tid); | |
13c2b57d YZ |
2422 | } else { |
2423 | ci->i_ceph_flags |= CEPH_I_KICK_FLUSH; | |
e548e9b9 YZ |
2424 | } |
2425 | ||
e548e9b9 YZ |
2426 | spin_unlock(&ci->i_ceph_lock); |
2427 | } | |
2428 | } | |
2429 | ||
a8599bd8 SW |
2430 | void ceph_kick_flushing_caps(struct ceph_mds_client *mdsc, |
2431 | struct ceph_mds_session *session) | |
2432 | { | |
2433 | struct ceph_inode_info *ci; | |
13c2b57d | 2434 | struct ceph_cap *cap; |
0e294387 | 2435 | u64 oldest_flush_tid; |
a8599bd8 SW |
2436 | |
2437 | dout("kick_flushing_caps mds%d\n", session->s_mds); | |
0e294387 YZ |
2438 | |
2439 | spin_lock(&mdsc->cap_dirty_lock); | |
2440 | oldest_flush_tid = __get_oldest_flush_tid(mdsc); | |
2441 | spin_unlock(&mdsc->cap_dirty_lock); | |
2442 | ||
a8599bd8 | 2443 | list_for_each_entry(ci, &session->s_cap_flushing, i_flushing_item) { |
0e294387 | 2444 | spin_lock(&ci->i_ceph_lock); |
13c2b57d YZ |
2445 | cap = ci->i_auth_cap; |
2446 | if (!(cap && cap->session == session)) { | |
2447 | pr_err("%p auth cap %p not mds%d ???\n", | |
2448 | &ci->vfs_inode, cap, session->s_mds); | |
2449 | spin_unlock(&ci->i_ceph_lock); | |
2450 | continue; | |
2451 | } | |
2452 | if (ci->i_ceph_flags & CEPH_I_KICK_FLUSH) { | |
13c2b57d YZ |
2453 | __kick_flushing_caps(mdsc, session, ci, |
2454 | oldest_flush_tid); | |
2455 | } | |
0e294387 | 2456 | spin_unlock(&ci->i_ceph_lock); |
a8599bd8 SW |
2457 | } |
2458 | } | |
2459 | ||
088b3f5e SW |
2460 | static void kick_flushing_inode_caps(struct ceph_mds_client *mdsc, |
2461 | struct ceph_mds_session *session, | |
2462 | struct inode *inode) | |
0e294387 | 2463 | __releases(ci->i_ceph_lock) |
088b3f5e SW |
2464 | { |
2465 | struct ceph_inode_info *ci = ceph_inode(inode); | |
2466 | struct ceph_cap *cap; | |
088b3f5e | 2467 | |
088b3f5e | 2468 | cap = ci->i_auth_cap; |
8310b089 YZ |
2469 | dout("kick_flushing_inode_caps %p flushing %s\n", inode, |
2470 | ceph_cap_string(ci->i_flushing_caps)); | |
005c4697 | 2471 | |
0e294387 YZ |
2472 | if (!list_empty(&ci->i_cap_flush_list)) { |
2473 | u64 oldest_flush_tid; | |
005c4697 YZ |
2474 | spin_lock(&mdsc->cap_dirty_lock); |
2475 | list_move_tail(&ci->i_flushing_item, | |
2476 | &cap->session->s_cap_flushing); | |
0e294387 | 2477 | oldest_flush_tid = __get_oldest_flush_tid(mdsc); |
005c4697 YZ |
2478 | spin_unlock(&mdsc->cap_dirty_lock); |
2479 | ||
0e294387 | 2480 | __kick_flushing_caps(mdsc, session, ci, oldest_flush_tid); |
553adfd9 | 2481 | spin_unlock(&ci->i_ceph_lock); |
088b3f5e | 2482 | } else { |
be655596 | 2483 | spin_unlock(&ci->i_ceph_lock); |
088b3f5e SW |
2484 | } |
2485 | } | |
2486 | ||
a8599bd8 SW |
2487 | |
2488 | /* | |
2489 | * Take references to capabilities we hold, so that we don't release | |
2490 | * them to the MDS prematurely. | |
2491 | * | |
be655596 | 2492 | * Protected by i_ceph_lock. |
a8599bd8 | 2493 | */ |
5dda377c YZ |
2494 | static void __take_cap_refs(struct ceph_inode_info *ci, int got, |
2495 | bool snap_rwsem_locked) | |
a8599bd8 SW |
2496 | { |
2497 | if (got & CEPH_CAP_PIN) | |
2498 | ci->i_pin_ref++; | |
2499 | if (got & CEPH_CAP_FILE_RD) | |
2500 | ci->i_rd_ref++; | |
2501 | if (got & CEPH_CAP_FILE_CACHE) | |
2502 | ci->i_rdcache_ref++; | |
5dda377c YZ |
2503 | if (got & CEPH_CAP_FILE_WR) { |
2504 | if (ci->i_wr_ref == 0 && !ci->i_head_snapc) { | |
2505 | BUG_ON(!snap_rwsem_locked); | |
2506 | ci->i_head_snapc = ceph_get_snap_context( | |
2507 | ci->i_snap_realm->cached_context); | |
2508 | } | |
a8599bd8 | 2509 | ci->i_wr_ref++; |
5dda377c | 2510 | } |
a8599bd8 | 2511 | if (got & CEPH_CAP_FILE_BUFFER) { |
d3d0720d | 2512 | if (ci->i_wb_ref == 0) |
3772d26d | 2513 | ihold(&ci->vfs_inode); |
d3d0720d HC |
2514 | ci->i_wb_ref++; |
2515 | dout("__take_cap_refs %p wb %d -> %d (?)\n", | |
2516 | &ci->vfs_inode, ci->i_wb_ref-1, ci->i_wb_ref); | |
a8599bd8 SW |
2517 | } |
2518 | } | |
2519 | ||
2520 | /* | |
2521 | * Try to grab cap references. Specify those refs we @want, and the | |
2522 | * minimal set we @need. Also include the larger offset we are writing | |
2523 | * to (when applicable), and check against max_size here as well. | |
2524 | * Note that caller is responsible for ensuring max_size increases are | |
2525 | * requested from the MDS. | |
1199d7da JL |
2526 | * |
2527 | * Returns 0 if caps were not able to be acquired (yet), a 1 if they were, | |
2528 | * or a negative error code. | |
2529 | * | |
2530 | * FIXME: how does a 0 return differ from -EAGAIN? | |
a8599bd8 | 2531 | */ |
ff5d913d YZ |
2532 | enum { |
2533 | NON_BLOCKING = 1, | |
2534 | CHECK_FILELOCK = 2, | |
2535 | }; | |
2536 | ||
5e3ded1b | 2537 | static int try_get_cap_refs(struct inode *inode, int need, int want, |
ff5d913d | 2538 | loff_t endoff, int flags, int *got) |
a8599bd8 | 2539 | { |
5e3ded1b | 2540 | struct ceph_inode_info *ci = ceph_inode(inode); |
5dda377c | 2541 | struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc; |
a8599bd8 | 2542 | int ret = 0; |
c4d4a582 | 2543 | int have, implemented; |
195d3ce2 | 2544 | int file_wanted; |
5dda377c | 2545 | bool snap_rwsem_locked = false; |
a8599bd8 SW |
2546 | |
2547 | dout("get_cap_refs %p need %s want %s\n", inode, | |
2548 | ceph_cap_string(need), ceph_cap_string(want)); | |
c4d4a582 | 2549 | |
5dda377c | 2550 | again: |
be655596 | 2551 | spin_lock(&ci->i_ceph_lock); |
a8599bd8 | 2552 | |
ff5d913d YZ |
2553 | if ((flags & CHECK_FILELOCK) && |
2554 | (ci->i_ceph_flags & CEPH_I_ERROR_FILELOCK)) { | |
2555 | dout("try_get_cap_refs %p error filelock\n", inode); | |
2556 | ret = -EIO; | |
2557 | goto out_unlock; | |
2558 | } | |
2559 | ||
195d3ce2 SW |
2560 | /* make sure file is actually open */ |
2561 | file_wanted = __ceph_caps_file_wanted(ci); | |
77310320 | 2562 | if ((file_wanted & need) != need) { |
195d3ce2 SW |
2563 | dout("try_get_cap_refs need %s file_wanted %s, EBADF\n", |
2564 | ceph_cap_string(need), ceph_cap_string(file_wanted)); | |
1199d7da | 2565 | ret = -EBADF; |
3738daa6 | 2566 | goto out_unlock; |
a8599bd8 SW |
2567 | } |
2568 | ||
37505d57 YZ |
2569 | /* finish pending truncate */ |
2570 | while (ci->i_truncate_pending) { | |
2571 | spin_unlock(&ci->i_ceph_lock); | |
5dda377c YZ |
2572 | if (snap_rwsem_locked) { |
2573 | up_read(&mdsc->snap_rwsem); | |
2574 | snap_rwsem_locked = false; | |
2575 | } | |
b415bf4f | 2576 | __ceph_do_pending_vmtruncate(inode); |
37505d57 YZ |
2577 | spin_lock(&ci->i_ceph_lock); |
2578 | } | |
2579 | ||
3871cbb9 YZ |
2580 | have = __ceph_caps_issued(ci, &implemented); |
2581 | ||
2582 | if (have & need & CEPH_CAP_FILE_WR) { | |
a8599bd8 SW |
2583 | if (endoff >= 0 && endoff > (loff_t)ci->i_max_size) { |
2584 | dout("get_cap_refs %p endoff %llu > maxsize %llu\n", | |
2585 | inode, endoff, ci->i_max_size); | |
1199d7da JL |
2586 | if (endoff > ci->i_requested_max_size) |
2587 | ret = -EAGAIN; | |
3738daa6 | 2588 | goto out_unlock; |
a8599bd8 SW |
2589 | } |
2590 | /* | |
2591 | * If a sync write is in progress, we must wait, so that we | |
2592 | * can get a final snapshot value for size+mtime. | |
2593 | */ | |
2594 | if (__ceph_have_pending_cap_snap(ci)) { | |
2595 | dout("get_cap_refs %p cap_snap_pending\n", inode); | |
3738daa6 | 2596 | goto out_unlock; |
a8599bd8 SW |
2597 | } |
2598 | } | |
a8599bd8 | 2599 | |
a8599bd8 SW |
2600 | if ((have & need) == need) { |
2601 | /* | |
2602 | * Look at (implemented & ~have & not) so that we keep waiting | |
2603 | * on transition from wanted -> needed caps. This is needed | |
2604 | * for WRBUFFER|WR -> WR to avoid a new WR sync write from | |
2605 | * going before a prior buffered writeback happens. | |
2606 | */ | |
2607 | int not = want & ~(have & need); | |
2608 | int revoking = implemented & ~have; | |
2609 | dout("get_cap_refs %p have %s but not %s (revoking %s)\n", | |
2610 | inode, ceph_cap_string(have), ceph_cap_string(not), | |
2611 | ceph_cap_string(revoking)); | |
2612 | if ((revoking & not) == 0) { | |
5dda377c YZ |
2613 | if (!snap_rwsem_locked && |
2614 | !ci->i_head_snapc && | |
2615 | (need & CEPH_CAP_FILE_WR)) { | |
2616 | if (!down_read_trylock(&mdsc->snap_rwsem)) { | |
2617 | /* | |
2618 | * we can not call down_read() when | |
2619 | * task isn't in TASK_RUNNING state | |
2620 | */ | |
ff5d913d | 2621 | if (flags & NON_BLOCKING) { |
1199d7da | 2622 | ret = -EAGAIN; |
5dda377c YZ |
2623 | goto out_unlock; |
2624 | } | |
2625 | ||
2626 | spin_unlock(&ci->i_ceph_lock); | |
2627 | down_read(&mdsc->snap_rwsem); | |
2628 | snap_rwsem_locked = true; | |
2629 | goto again; | |
2630 | } | |
2631 | snap_rwsem_locked = true; | |
2632 | } | |
c4d4a582 | 2633 | *got = need | (have & want); |
f7f7e7a0 YZ |
2634 | if ((need & CEPH_CAP_FILE_RD) && |
2635 | !(*got & CEPH_CAP_FILE_CACHE)) | |
2636 | ceph_disable_fscache_readpage(ci); | |
5dda377c | 2637 | __take_cap_refs(ci, *got, true); |
a8599bd8 SW |
2638 | ret = 1; |
2639 | } | |
2640 | } else { | |
03f4fcb0 YZ |
2641 | int session_readonly = false; |
2642 | if ((need & CEPH_CAP_FILE_WR) && ci->i_auth_cap) { | |
2643 | struct ceph_mds_session *s = ci->i_auth_cap->session; | |
2644 | spin_lock(&s->s_cap_lock); | |
2645 | session_readonly = s->s_readonly; | |
2646 | spin_unlock(&s->s_cap_lock); | |
2647 | } | |
2648 | if (session_readonly) { | |
2649 | dout("get_cap_refs %p needed %s but mds%d readonly\n", | |
2650 | inode, ceph_cap_string(need), ci->i_auth_cap->mds); | |
1199d7da | 2651 | ret = -EROFS; |
03f4fcb0 YZ |
2652 | goto out_unlock; |
2653 | } | |
2654 | ||
77310320 YZ |
2655 | if (ci->i_ceph_flags & CEPH_I_CAP_DROPPED) { |
2656 | int mds_wanted; | |
52953d55 | 2657 | if (READ_ONCE(mdsc->fsc->mount_state) == |
77310320 YZ |
2658 | CEPH_MOUNT_SHUTDOWN) { |
2659 | dout("get_cap_refs %p forced umount\n", inode); | |
1199d7da | 2660 | ret = -EIO; |
77310320 YZ |
2661 | goto out_unlock; |
2662 | } | |
c1944fed | 2663 | mds_wanted = __ceph_caps_mds_wanted(ci, false); |
eb65b919 | 2664 | if (need & ~(mds_wanted & need)) { |
77310320 YZ |
2665 | dout("get_cap_refs %p caps were dropped" |
2666 | " (session killed?)\n", inode); | |
1199d7da | 2667 | ret = -ESTALE; |
77310320 YZ |
2668 | goto out_unlock; |
2669 | } | |
eb65b919 | 2670 | if (!(file_wanted & ~mds_wanted)) |
77310320 | 2671 | ci->i_ceph_flags &= ~CEPH_I_CAP_DROPPED; |
48fec5d0 YZ |
2672 | } |
2673 | ||
a8599bd8 SW |
2674 | dout("get_cap_refs %p have %s needed %s\n", inode, |
2675 | ceph_cap_string(have), ceph_cap_string(need)); | |
2676 | } | |
3738daa6 | 2677 | out_unlock: |
be655596 | 2678 | spin_unlock(&ci->i_ceph_lock); |
5dda377c YZ |
2679 | if (snap_rwsem_locked) |
2680 | up_read(&mdsc->snap_rwsem); | |
3738daa6 | 2681 | |
a8599bd8 | 2682 | dout("get_cap_refs %p ret %d got %s\n", inode, |
c4d4a582 | 2683 | ret, ceph_cap_string(*got)); |
a8599bd8 SW |
2684 | return ret; |
2685 | } | |
2686 | ||
2687 | /* | |
2688 | * Check the offset we are writing up to against our current | |
2689 | * max_size. If necessary, tell the MDS we want to write to | |
2690 | * a larger offset. | |
2691 | */ | |
2692 | static void check_max_size(struct inode *inode, loff_t endoff) | |
2693 | { | |
2694 | struct ceph_inode_info *ci = ceph_inode(inode); | |
2695 | int check = 0; | |
2696 | ||
2697 | /* do we need to explicitly request a larger max_size? */ | |
be655596 | 2698 | spin_lock(&ci->i_ceph_lock); |
3871cbb9 | 2699 | if (endoff >= ci->i_max_size && endoff > ci->i_wanted_max_size) { |
a8599bd8 SW |
2700 | dout("write %p at large endoff %llu, req max_size\n", |
2701 | inode, endoff); | |
2702 | ci->i_wanted_max_size = endoff; | |
a8599bd8 | 2703 | } |
3871cbb9 YZ |
2704 | /* duplicate ceph_check_caps()'s logic */ |
2705 | if (ci->i_auth_cap && | |
2706 | (ci->i_auth_cap->issued & CEPH_CAP_FILE_WR) && | |
2707 | ci->i_wanted_max_size > ci->i_max_size && | |
2708 | ci->i_wanted_max_size > ci->i_requested_max_size) | |
2709 | check = 1; | |
be655596 | 2710 | spin_unlock(&ci->i_ceph_lock); |
a8599bd8 SW |
2711 | if (check) |
2712 | ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL); | |
2713 | } | |
2714 | ||
5e3ded1b | 2715 | int ceph_try_get_caps(struct inode *inode, int need, int want, |
2ee9dd95 | 2716 | bool nonblock, int *got) |
2b1ac852 | 2717 | { |
1199d7da | 2718 | int ret; |
2b1ac852 YZ |
2719 | |
2720 | BUG_ON(need & ~CEPH_CAP_FILE_RD); | |
2ee9dd95 | 2721 | BUG_ON(want & ~(CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO|CEPH_CAP_FILE_SHARED)); |
5e3ded1b | 2722 | ret = ceph_pool_perm_check(inode, need); |
2b1ac852 YZ |
2723 | if (ret < 0) |
2724 | return ret; | |
2725 | ||
ff5d913d YZ |
2726 | ret = try_get_cap_refs(inode, need, want, 0, |
2727 | (nonblock ? NON_BLOCKING : 0), got); | |
1199d7da | 2728 | return ret == -EAGAIN ? 0 : ret; |
2b1ac852 YZ |
2729 | } |
2730 | ||
a8599bd8 SW |
2731 | /* |
2732 | * Wait for caps, and take cap references. If we can't get a WR cap | |
2733 | * due to a small max_size, make sure we check_max_size (and possibly | |
2734 | * ask the mds) so we don't get hung up indefinitely. | |
2735 | */ | |
5e3ded1b | 2736 | int ceph_get_caps(struct file *filp, int need, int want, |
3738daa6 | 2737 | loff_t endoff, int *got, struct page **pinned_page) |
a8599bd8 | 2738 | { |
ff5d913d | 2739 | struct ceph_file_info *fi = filp->private_data; |
5e3ded1b YZ |
2740 | struct inode *inode = file_inode(filp); |
2741 | struct ceph_inode_info *ci = ceph_inode(inode); | |
81f148a9 | 2742 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); |
ff5d913d | 2743 | int ret, _got, flags; |
a8599bd8 | 2744 | |
5e3ded1b | 2745 | ret = ceph_pool_perm_check(inode, need); |
10183a69 YZ |
2746 | if (ret < 0) |
2747 | return ret; | |
2748 | ||
81f148a9 YZ |
2749 | if ((fi->fmode & CEPH_FILE_MODE_WR) && |
2750 | fi->filp_gen != READ_ONCE(fsc->filp_gen)) | |
2751 | return -EBADF; | |
2752 | ||
5dda377c YZ |
2753 | while (true) { |
2754 | if (endoff > 0) | |
5e3ded1b | 2755 | check_max_size(inode, endoff); |
c4d4a582 | 2756 | |
ff5d913d | 2757 | flags = atomic_read(&fi->num_locks) ? CHECK_FILELOCK : 0; |
5dda377c | 2758 | _got = 0; |
5e3ded1b | 2759 | ret = try_get_cap_refs(inode, need, want, endoff, |
ff5d913d | 2760 | flags, &_got); |
7b2f936f | 2761 | if (ret == -EAGAIN) |
1199d7da | 2762 | continue; |
7b2f936f | 2763 | if (!ret) { |
3a3430af JL |
2764 | struct ceph_mds_client *mdsc = fsc->mdsc; |
2765 | struct cap_wait cw; | |
5c341ee3 | 2766 | DEFINE_WAIT_FUNC(wait, woken_wake_function); |
3a3430af JL |
2767 | |
2768 | cw.ino = inode->i_ino; | |
2769 | cw.tgid = current->tgid; | |
2770 | cw.need = need; | |
2771 | cw.want = want; | |
2772 | ||
2773 | spin_lock(&mdsc->caps_list_lock); | |
2774 | list_add(&cw.list, &mdsc->cap_wait_list); | |
2775 | spin_unlock(&mdsc->caps_list_lock); | |
2776 | ||
5c341ee3 NB |
2777 | add_wait_queue(&ci->i_cap_wq, &wait); |
2778 | ||
ff5d913d | 2779 | flags |= NON_BLOCKING; |
5e3ded1b | 2780 | while (!(ret = try_get_cap_refs(inode, need, want, |
ff5d913d | 2781 | endoff, flags, &_got))) { |
6e09d0fb YZ |
2782 | if (signal_pending(current)) { |
2783 | ret = -ERESTARTSYS; | |
2784 | break; | |
2785 | } | |
5c341ee3 | 2786 | wait_woken(&wait, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
6e09d0fb | 2787 | } |
5c341ee3 NB |
2788 | |
2789 | remove_wait_queue(&ci->i_cap_wq, &wait); | |
3a3430af JL |
2790 | |
2791 | spin_lock(&mdsc->caps_list_lock); | |
2792 | list_del(&cw.list); | |
2793 | spin_unlock(&mdsc->caps_list_lock); | |
2794 | ||
7b2f936f | 2795 | if (ret == -EAGAIN) |
5dda377c | 2796 | continue; |
77310320 | 2797 | } |
81f148a9 YZ |
2798 | |
2799 | if ((fi->fmode & CEPH_FILE_MODE_WR) && | |
2800 | fi->filp_gen != READ_ONCE(fsc->filp_gen)) { | |
2801 | if (ret >= 0 && _got) | |
2802 | ceph_put_cap_refs(ci, _got); | |
2803 | return -EBADF; | |
2804 | } | |
2805 | ||
7b2f936f YZ |
2806 | if (ret < 0) { |
2807 | if (ret == -ESTALE) { | |
2808 | /* session was killed, try renew caps */ | |
5e3ded1b | 2809 | ret = ceph_renew_caps(inode); |
7b2f936f YZ |
2810 | if (ret == 0) |
2811 | continue; | |
2812 | } | |
77310320 | 2813 | return ret; |
5dda377c | 2814 | } |
c4d4a582 | 2815 | |
5dda377c YZ |
2816 | if (ci->i_inline_version != CEPH_INLINE_NONE && |
2817 | (_got & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) && | |
5e3ded1b | 2818 | i_size_read(inode) > 0) { |
5dda377c | 2819 | struct page *page = |
5e3ded1b | 2820 | find_get_page(inode->i_mapping, 0); |
5dda377c YZ |
2821 | if (page) { |
2822 | if (PageUptodate(page)) { | |
2823 | *pinned_page = page; | |
2824 | break; | |
2825 | } | |
09cbfeaf | 2826 | put_page(page); |
c4d4a582 | 2827 | } |
5dda377c YZ |
2828 | /* |
2829 | * drop cap refs first because getattr while | |
2830 | * holding * caps refs can cause deadlock. | |
2831 | */ | |
2832 | ceph_put_cap_refs(ci, _got); | |
2833 | _got = 0; | |
c4d4a582 | 2834 | |
5dda377c YZ |
2835 | /* |
2836 | * getattr request will bring inline data into | |
2837 | * page cache | |
2838 | */ | |
5e3ded1b | 2839 | ret = __ceph_do_getattr(inode, NULL, |
5dda377c YZ |
2840 | CEPH_STAT_CAP_INLINE_DATA, |
2841 | true); | |
2842 | if (ret < 0) | |
2843 | return ret; | |
2844 | continue; | |
2845 | } | |
2846 | break; | |
c4d4a582 | 2847 | } |
5dda377c | 2848 | |
f7f7e7a0 YZ |
2849 | if ((_got & CEPH_CAP_FILE_RD) && (_got & CEPH_CAP_FILE_CACHE)) |
2850 | ceph_fscache_revalidate_cookie(ci); | |
2851 | ||
c4d4a582 YZ |
2852 | *got = _got; |
2853 | return 0; | |
a8599bd8 SW |
2854 | } |
2855 | ||
2856 | /* | |
2857 | * Take cap refs. Caller must already know we hold at least one ref | |
2858 | * on the caps in question or we don't know this is safe. | |
2859 | */ | |
2860 | void ceph_get_cap_refs(struct ceph_inode_info *ci, int caps) | |
2861 | { | |
be655596 | 2862 | spin_lock(&ci->i_ceph_lock); |
5dda377c | 2863 | __take_cap_refs(ci, caps, false); |
be655596 | 2864 | spin_unlock(&ci->i_ceph_lock); |
a8599bd8 SW |
2865 | } |
2866 | ||
86056090 YZ |
2867 | |
2868 | /* | |
2869 | * drop cap_snap that is not associated with any snapshot. | |
2870 | * we don't need to send FLUSHSNAP message for it. | |
2871 | */ | |
70220ac8 YZ |
2872 | static int ceph_try_drop_cap_snap(struct ceph_inode_info *ci, |
2873 | struct ceph_cap_snap *capsnap) | |
86056090 YZ |
2874 | { |
2875 | if (!capsnap->need_flush && | |
2876 | !capsnap->writing && !capsnap->dirty_pages) { | |
86056090 YZ |
2877 | dout("dropping cap_snap %p follows %llu\n", |
2878 | capsnap, capsnap->follows); | |
0e294387 | 2879 | BUG_ON(capsnap->cap_flush.tid > 0); |
86056090 | 2880 | ceph_put_snap_context(capsnap->context); |
70220ac8 YZ |
2881 | if (!list_is_last(&capsnap->ci_item, &ci->i_cap_snaps)) |
2882 | ci->i_ceph_flags |= CEPH_I_FLUSH_SNAPS; | |
2883 | ||
86056090 | 2884 | list_del(&capsnap->ci_item); |
86056090 YZ |
2885 | ceph_put_cap_snap(capsnap); |
2886 | return 1; | |
2887 | } | |
2888 | return 0; | |
2889 | } | |
2890 | ||
a8599bd8 SW |
2891 | /* |
2892 | * Release cap refs. | |
2893 | * | |
2894 | * If we released the last ref on any given cap, call ceph_check_caps | |
2895 | * to release (or schedule a release). | |
2896 | * | |
2897 | * If we are releasing a WR cap (from a sync write), finalize any affected | |
2898 | * cap_snap, and wake up any waiters. | |
2899 | */ | |
2900 | void ceph_put_cap_refs(struct ceph_inode_info *ci, int had) | |
2901 | { | |
2902 | struct inode *inode = &ci->vfs_inode; | |
2903 | int last = 0, put = 0, flushsnaps = 0, wake = 0; | |
a8599bd8 | 2904 | |
be655596 | 2905 | spin_lock(&ci->i_ceph_lock); |
a8599bd8 SW |
2906 | if (had & CEPH_CAP_PIN) |
2907 | --ci->i_pin_ref; | |
2908 | if (had & CEPH_CAP_FILE_RD) | |
2909 | if (--ci->i_rd_ref == 0) | |
2910 | last++; | |
2911 | if (had & CEPH_CAP_FILE_CACHE) | |
2912 | if (--ci->i_rdcache_ref == 0) | |
2913 | last++; | |
2914 | if (had & CEPH_CAP_FILE_BUFFER) { | |
d3d0720d | 2915 | if (--ci->i_wb_ref == 0) { |
a8599bd8 SW |
2916 | last++; |
2917 | put++; | |
2918 | } | |
d3d0720d HC |
2919 | dout("put_cap_refs %p wb %d -> %d (?)\n", |
2920 | inode, ci->i_wb_ref+1, ci->i_wb_ref); | |
a8599bd8 SW |
2921 | } |
2922 | if (had & CEPH_CAP_FILE_WR) | |
2923 | if (--ci->i_wr_ref == 0) { | |
2924 | last++; | |
86056090 YZ |
2925 | if (__ceph_have_pending_cap_snap(ci)) { |
2926 | struct ceph_cap_snap *capsnap = | |
2927 | list_last_entry(&ci->i_cap_snaps, | |
2928 | struct ceph_cap_snap, | |
2929 | ci_item); | |
2930 | capsnap->writing = 0; | |
70220ac8 | 2931 | if (ceph_try_drop_cap_snap(ci, capsnap)) |
86056090 YZ |
2932 | put++; |
2933 | else if (__ceph_finish_cap_snap(ci, capsnap)) | |
2934 | flushsnaps = 1; | |
2935 | wake = 1; | |
a8599bd8 | 2936 | } |
5dda377c YZ |
2937 | if (ci->i_wrbuffer_ref_head == 0 && |
2938 | ci->i_dirty_caps == 0 && | |
2939 | ci->i_flushing_caps == 0) { | |
2940 | BUG_ON(!ci->i_head_snapc); | |
2941 | ceph_put_snap_context(ci->i_head_snapc); | |
2942 | ci->i_head_snapc = NULL; | |
2943 | } | |
db40cc17 | 2944 | /* see comment in __ceph_remove_cap() */ |
bd84fbcb | 2945 | if (!__ceph_is_any_real_caps(ci) && ci->i_snap_realm) |
db40cc17 | 2946 | drop_inode_snap_realm(ci); |
a8599bd8 | 2947 | } |
be655596 | 2948 | spin_unlock(&ci->i_ceph_lock); |
a8599bd8 | 2949 | |
819ccbfa SW |
2950 | dout("put_cap_refs %p had %s%s%s\n", inode, ceph_cap_string(had), |
2951 | last ? " last" : "", put ? " put" : ""); | |
a8599bd8 SW |
2952 | |
2953 | if (last && !flushsnaps) | |
2954 | ceph_check_caps(ci, 0, NULL); | |
2955 | else if (flushsnaps) | |
ed9b430c | 2956 | ceph_flush_snaps(ci, NULL); |
a8599bd8 | 2957 | if (wake) |
03066f23 | 2958 | wake_up_all(&ci->i_cap_wq); |
86056090 | 2959 | while (put-- > 0) |
a8599bd8 SW |
2960 | iput(inode); |
2961 | } | |
2962 | ||
2963 | /* | |
2964 | * Release @nr WRBUFFER refs on dirty pages for the given @snapc snap | |
2965 | * context. Adjust per-snap dirty page accounting as appropriate. | |
2966 | * Once all dirty data for a cap_snap is flushed, flush snapped file | |
2967 | * metadata back to the MDS. If we dropped the last ref, call | |
2968 | * ceph_check_caps. | |
2969 | */ | |
2970 | void ceph_put_wrbuffer_cap_refs(struct ceph_inode_info *ci, int nr, | |
2971 | struct ceph_snap_context *snapc) | |
2972 | { | |
2973 | struct inode *inode = &ci->vfs_inode; | |
a8599bd8 | 2974 | struct ceph_cap_snap *capsnap = NULL; |
70220ac8 YZ |
2975 | int put = 0; |
2976 | bool last = false; | |
2977 | bool found = false; | |
2978 | bool flush_snaps = false; | |
2979 | bool complete_capsnap = false; | |
a8599bd8 | 2980 | |
be655596 | 2981 | spin_lock(&ci->i_ceph_lock); |
a8599bd8 | 2982 | ci->i_wrbuffer_ref -= nr; |
70220ac8 YZ |
2983 | if (ci->i_wrbuffer_ref == 0) { |
2984 | last = true; | |
2985 | put++; | |
2986 | } | |
a8599bd8 SW |
2987 | |
2988 | if (ci->i_head_snapc == snapc) { | |
2989 | ci->i_wrbuffer_ref_head -= nr; | |
7d8cb26d | 2990 | if (ci->i_wrbuffer_ref_head == 0 && |
5dda377c YZ |
2991 | ci->i_wr_ref == 0 && |
2992 | ci->i_dirty_caps == 0 && | |
2993 | ci->i_flushing_caps == 0) { | |
7d8cb26d | 2994 | BUG_ON(!ci->i_head_snapc); |
a8599bd8 SW |
2995 | ceph_put_snap_context(ci->i_head_snapc); |
2996 | ci->i_head_snapc = NULL; | |
2997 | } | |
2998 | dout("put_wrbuffer_cap_refs on %p head %d/%d -> %d/%d %s\n", | |
2999 | inode, | |
3000 | ci->i_wrbuffer_ref+nr, ci->i_wrbuffer_ref_head+nr, | |
3001 | ci->i_wrbuffer_ref, ci->i_wrbuffer_ref_head, | |
3002 | last ? " LAST" : ""); | |
3003 | } else { | |
3004 | list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) { | |
3005 | if (capsnap->context == snapc) { | |
70220ac8 | 3006 | found = true; |
a8599bd8 SW |
3007 | break; |
3008 | } | |
3009 | } | |
3010 | BUG_ON(!found); | |
819ccbfa SW |
3011 | capsnap->dirty_pages -= nr; |
3012 | if (capsnap->dirty_pages == 0) { | |
70220ac8 YZ |
3013 | complete_capsnap = true; |
3014 | if (!capsnap->writing) { | |
3015 | if (ceph_try_drop_cap_snap(ci, capsnap)) { | |
3016 | put++; | |
3017 | } else { | |
3018 | ci->i_ceph_flags |= CEPH_I_FLUSH_SNAPS; | |
3019 | flush_snaps = true; | |
3020 | } | |
3021 | } | |
819ccbfa | 3022 | } |
a8599bd8 | 3023 | dout("put_wrbuffer_cap_refs on %p cap_snap %p " |
86056090 | 3024 | " snap %lld %d/%d -> %d/%d %s%s\n", |
a8599bd8 SW |
3025 | inode, capsnap, capsnap->context->seq, |
3026 | ci->i_wrbuffer_ref+nr, capsnap->dirty_pages + nr, | |
3027 | ci->i_wrbuffer_ref, capsnap->dirty_pages, | |
3028 | last ? " (wrbuffer last)" : "", | |
86056090 | 3029 | complete_capsnap ? " (complete capsnap)" : ""); |
a8599bd8 SW |
3030 | } |
3031 | ||
be655596 | 3032 | spin_unlock(&ci->i_ceph_lock); |
a8599bd8 SW |
3033 | |
3034 | if (last) { | |
3035 | ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL); | |
70220ac8 | 3036 | } else if (flush_snaps) { |
ed9b430c | 3037 | ceph_flush_snaps(ci, NULL); |
a8599bd8 | 3038 | } |
70220ac8 YZ |
3039 | if (complete_capsnap) |
3040 | wake_up_all(&ci->i_cap_wq); | |
3e1d0452 YZ |
3041 | while (put-- > 0) { |
3042 | /* avoid calling iput_final() in osd dispatch threads */ | |
3043 | ceph_async_iput(inode); | |
3044 | } | |
a8599bd8 SW |
3045 | } |
3046 | ||
ca20c991 YZ |
3047 | /* |
3048 | * Invalidate unlinked inode's aliases, so we can drop the inode ASAP. | |
3049 | */ | |
3050 | static void invalidate_aliases(struct inode *inode) | |
3051 | { | |
3052 | struct dentry *dn, *prev = NULL; | |
3053 | ||
3054 | dout("invalidate_aliases inode %p\n", inode); | |
3055 | d_prune_aliases(inode); | |
3056 | /* | |
3057 | * For non-directory inode, d_find_alias() only returns | |
fc12c80a BF |
3058 | * hashed dentry. After calling d_invalidate(), the |
3059 | * dentry becomes unhashed. | |
ca20c991 | 3060 | * |
a8d436f0 | 3061 | * For directory inode, d_find_alias() can return |
fc12c80a | 3062 | * unhashed dentry. But directory inode should have |
ca20c991 YZ |
3063 | * one alias at most. |
3064 | */ | |
3065 | while ((dn = d_find_alias(inode))) { | |
3066 | if (dn == prev) { | |
3067 | dput(dn); | |
3068 | break; | |
3069 | } | |
a8d436f0 | 3070 | d_invalidate(dn); |
ca20c991 YZ |
3071 | if (prev) |
3072 | dput(prev); | |
3073 | prev = dn; | |
3074 | } | |
3075 | if (prev) | |
3076 | dput(prev); | |
3077 | } | |
3078 | ||
a1c6b835 YZ |
3079 | struct cap_extra_info { |
3080 | struct ceph_string *pool_ns; | |
3081 | /* inline data */ | |
3082 | u64 inline_version; | |
3083 | void *inline_data; | |
3084 | u32 inline_len; | |
4985d6f9 YZ |
3085 | /* dirstat */ |
3086 | bool dirstat_valid; | |
3087 | u64 nfiles; | |
3088 | u64 nsubdirs; | |
176c77c9 | 3089 | u64 change_attr; |
a1c6b835 YZ |
3090 | /* currently issued */ |
3091 | int issued; | |
ec62b894 | 3092 | struct timespec64 btime; |
a1c6b835 YZ |
3093 | }; |
3094 | ||
a8599bd8 SW |
3095 | /* |
3096 | * Handle a cap GRANT message from the MDS. (Note that a GRANT may | |
3097 | * actually be a revocation if it specifies a smaller cap set.) | |
3098 | * | |
be655596 | 3099 | * caller holds s_mutex and i_ceph_lock, we drop both. |
a8599bd8 | 3100 | */ |
a1c6b835 | 3101 | static void handle_cap_grant(struct inode *inode, |
15637c8b | 3102 | struct ceph_mds_session *session, |
a1c6b835 YZ |
3103 | struct ceph_cap *cap, |
3104 | struct ceph_mds_caps *grant, | |
3105 | struct ceph_buffer *xattr_buf, | |
3106 | struct cap_extra_info *extra_info) | |
2cd698be | 3107 | __releases(ci->i_ceph_lock) |
a1c6b835 | 3108 | __releases(session->s_mdsc->snap_rwsem) |
a8599bd8 SW |
3109 | { |
3110 | struct ceph_inode_info *ci = ceph_inode(inode); | |
2f56f56a | 3111 | int seq = le32_to_cpu(grant->seq); |
a8599bd8 | 3112 | int newcaps = le32_to_cpu(grant->caps); |
2cd698be | 3113 | int used, wanted, dirty; |
a8599bd8 SW |
3114 | u64 size = le64_to_cpu(grant->size); |
3115 | u64 max_size = le64_to_cpu(grant->max_size); | |
fdac94fa YZ |
3116 | unsigned char check_caps = 0; |
3117 | bool was_stale = cap->cap_gen < session->s_cap_gen; | |
ab6c2c3e FF |
3118 | bool wake = false; |
3119 | bool writeback = false; | |
3120 | bool queue_trunc = false; | |
3121 | bool queue_invalidate = false; | |
ab6c2c3e | 3122 | bool deleted_inode = false; |
31c542a1 | 3123 | bool fill_inline = false; |
a8599bd8 | 3124 | |
2f56f56a | 3125 | dout("handle_cap_grant inode %p cap %p mds%d seq %d %s\n", |
a1c6b835 | 3126 | inode, cap, session->s_mds, seq, ceph_cap_string(newcaps)); |
a8599bd8 SW |
3127 | dout(" size %llu max_size %llu, i_size %llu\n", size, max_size, |
3128 | inode->i_size); | |
3129 | ||
11df2dfb | 3130 | |
a8599bd8 SW |
3131 | /* |
3132 | * If CACHE is being revoked, and we have no dirty buffers, | |
3133 | * try to invalidate (once). (If there are dirty buffers, we | |
3134 | * will invalidate _after_ writeback.) | |
3135 | */ | |
fdd4e158 YZ |
3136 | if (!S_ISDIR(inode->i_mode) && /* don't invalidate readdir cache */ |
3137 | ((cap->issued & ~newcaps) & CEPH_CAP_FILE_CACHE) && | |
3b454c49 | 3138 | (newcaps & CEPH_CAP_FILE_LAZYIO) == 0 && |
9abd4db7 | 3139 | !(ci->i_wrbuffer_ref || ci->i_wb_ref)) { |
e9075743 | 3140 | if (try_nonblocking_invalidate(inode)) { |
a8599bd8 SW |
3141 | /* there were locked pages.. invalidate later |
3142 | in a separate thread. */ | |
3143 | if (ci->i_rdcache_revoking != ci->i_rdcache_gen) { | |
ab6c2c3e | 3144 | queue_invalidate = true; |
a8599bd8 SW |
3145 | ci->i_rdcache_revoking = ci->i_rdcache_gen; |
3146 | } | |
a8599bd8 | 3147 | } |
a8599bd8 SW |
3148 | } |
3149 | ||
d2f8bb27 YZ |
3150 | if (was_stale) |
3151 | cap->issued = cap->implemented = CEPH_CAP_PIN; | |
3152 | ||
3153 | /* | |
3154 | * auth mds of the inode changed. we received the cap export message, | |
3155 | * but still haven't received the cap import message. handle_cap_export | |
3156 | * updated the new auth MDS' cap. | |
3157 | * | |
3158 | * "ceph_seq_cmp(seq, cap->seq) <= 0" means we are processing a message | |
3159 | * that was sent before the cap import message. So don't remove caps. | |
3160 | */ | |
3161 | if (ceph_seq_cmp(seq, cap->seq) <= 0) { | |
3162 | WARN_ON(cap != ci->i_auth_cap); | |
3163 | WARN_ON(cap->cap_id != le64_to_cpu(grant->cap_id)); | |
3164 | seq = cap->seq; | |
3165 | newcaps |= cap->issued; | |
3166 | } | |
3167 | ||
a8599bd8 | 3168 | /* side effects now are allowed */ |
685f9a5d | 3169 | cap->cap_gen = session->s_cap_gen; |
11df2dfb | 3170 | cap->seq = seq; |
a8599bd8 SW |
3171 | |
3172 | __check_cap_issue(ci, cap, newcaps); | |
3173 | ||
176c77c9 JL |
3174 | inode_set_max_iversion_raw(inode, extra_info->change_attr); |
3175 | ||
f98a128a | 3176 | if ((newcaps & CEPH_CAP_AUTH_SHARED) && |
a1c6b835 | 3177 | (extra_info->issued & CEPH_CAP_AUTH_EXCL) == 0) { |
a8599bd8 | 3178 | inode->i_mode = le32_to_cpu(grant->mode); |
05cb11c1 EB |
3179 | inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(grant->uid)); |
3180 | inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(grant->gid)); | |
ec62b894 | 3181 | ci->i_btime = extra_info->btime; |
a8599bd8 | 3182 | dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode, |
bd2bae6a EB |
3183 | from_kuid(&init_user_ns, inode->i_uid), |
3184 | from_kgid(&init_user_ns, inode->i_gid)); | |
a8599bd8 SW |
3185 | } |
3186 | ||
fa466743 | 3187 | if ((newcaps & CEPH_CAP_LINK_SHARED) && |
a1c6b835 | 3188 | (extra_info->issued & CEPH_CAP_LINK_EXCL) == 0) { |
bfe86848 | 3189 | set_nlink(inode, le32_to_cpu(grant->nlink)); |
ca20c991 YZ |
3190 | if (inode->i_nlink == 0 && |
3191 | (newcaps & (CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL))) | |
ab6c2c3e | 3192 | deleted_inode = true; |
ca20c991 | 3193 | } |
a8599bd8 | 3194 | |
a1c6b835 YZ |
3195 | if ((extra_info->issued & CEPH_CAP_XATTR_EXCL) == 0 && |
3196 | grant->xattr_len) { | |
a8599bd8 SW |
3197 | int len = le32_to_cpu(grant->xattr_len); |
3198 | u64 version = le64_to_cpu(grant->xattr_version); | |
3199 | ||
3200 | if (version > ci->i_xattrs.version) { | |
3201 | dout(" got new xattrs v%llu on %p len %d\n", | |
3202 | version, inode, len); | |
3203 | if (ci->i_xattrs.blob) | |
3204 | ceph_buffer_put(ci->i_xattrs.blob); | |
3205 | ci->i_xattrs.blob = ceph_buffer_get(xattr_buf); | |
3206 | ci->i_xattrs.version = version; | |
7221fe4c | 3207 | ceph_forget_all_cached_acls(inode); |
ac6713cc | 3208 | ceph_security_invalidate_secctx(inode); |
a8599bd8 SW |
3209 | } |
3210 | } | |
3211 | ||
f98a128a | 3212 | if (newcaps & CEPH_CAP_ANY_RD) { |
9bbeab41 | 3213 | struct timespec64 mtime, atime, ctime; |
f98a128a | 3214 | /* ctime/mtime/atime? */ |
9bbeab41 AB |
3215 | ceph_decode_timespec64(&mtime, &grant->mtime); |
3216 | ceph_decode_timespec64(&atime, &grant->atime); | |
3217 | ceph_decode_timespec64(&ctime, &grant->ctime); | |
a1c6b835 | 3218 | ceph_fill_file_time(inode, extra_info->issued, |
f98a128a YZ |
3219 | le32_to_cpu(grant->time_warp_seq), |
3220 | &ctime, &mtime, &atime); | |
3221 | } | |
3222 | ||
4985d6f9 YZ |
3223 | if ((newcaps & CEPH_CAP_FILE_SHARED) && extra_info->dirstat_valid) { |
3224 | ci->i_files = extra_info->nfiles; | |
3225 | ci->i_subdirs = extra_info->nsubdirs; | |
3226 | } | |
3227 | ||
f98a128a YZ |
3228 | if (newcaps & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR)) { |
3229 | /* file layout may have changed */ | |
7627151e | 3230 | s64 old_pool = ci->i_layout.pool_id; |
779fe0fb YZ |
3231 | struct ceph_string *old_ns; |
3232 | ||
7627151e | 3233 | ceph_file_layout_from_legacy(&ci->i_layout, &grant->layout); |
779fe0fb YZ |
3234 | old_ns = rcu_dereference_protected(ci->i_layout.pool_ns, |
3235 | lockdep_is_held(&ci->i_ceph_lock)); | |
a1c6b835 | 3236 | rcu_assign_pointer(ci->i_layout.pool_ns, extra_info->pool_ns); |
779fe0fb | 3237 | |
a1c6b835 YZ |
3238 | if (ci->i_layout.pool_id != old_pool || |
3239 | extra_info->pool_ns != old_ns) | |
7627151e | 3240 | ci->i_ceph_flags &= ~CEPH_I_POOL_PERM; |
5ea5c5e0 | 3241 | |
a1c6b835 | 3242 | extra_info->pool_ns = old_ns; |
779fe0fb | 3243 | |
f98a128a | 3244 | /* size/truncate_seq? */ |
a1c6b835 | 3245 | queue_trunc = ceph_fill_file_size(inode, extra_info->issued, |
f98a128a YZ |
3246 | le32_to_cpu(grant->truncate_seq), |
3247 | le64_to_cpu(grant->truncate_size), | |
3248 | size); | |
84eea8c7 YZ |
3249 | } |
3250 | ||
3251 | if (ci->i_auth_cap == cap && (newcaps & CEPH_CAP_ANY_FILE_WR)) { | |
3252 | if (max_size != ci->i_max_size) { | |
f98a128a YZ |
3253 | dout("max_size %lld -> %llu\n", |
3254 | ci->i_max_size, max_size); | |
3255 | ci->i_max_size = max_size; | |
3256 | if (max_size >= ci->i_wanted_max_size) { | |
3257 | ci->i_wanted_max_size = 0; /* reset */ | |
3258 | ci->i_requested_max_size = 0; | |
3259 | } | |
ab6c2c3e | 3260 | wake = true; |
84eea8c7 YZ |
3261 | } else if (ci->i_wanted_max_size > ci->i_max_size && |
3262 | ci->i_wanted_max_size > ci->i_requested_max_size) { | |
3263 | /* CEPH_CAP_OP_IMPORT */ | |
3264 | wake = true; | |
a8599bd8 | 3265 | } |
a8599bd8 SW |
3266 | } |
3267 | ||
3268 | /* check cap bits */ | |
3269 | wanted = __ceph_caps_wanted(ci); | |
3270 | used = __ceph_caps_used(ci); | |
3271 | dirty = __ceph_caps_dirty(ci); | |
3272 | dout(" my wanted = %s, used = %s, dirty %s\n", | |
3273 | ceph_cap_string(wanted), | |
3274 | ceph_cap_string(used), | |
3275 | ceph_cap_string(dirty)); | |
fdac94fa YZ |
3276 | |
3277 | if ((was_stale || le32_to_cpu(grant->op) == CEPH_CAP_OP_IMPORT) && | |
3278 | (wanted & ~(cap->mds_wanted | newcaps))) { | |
3279 | /* | |
3280 | * If mds is importing cap, prior cap messages that update | |
3281 | * 'wanted' may get dropped by mds (migrate seq mismatch). | |
3282 | * | |
3283 | * We don't send cap message to update 'wanted' if what we | |
3284 | * want are already issued. If mds revokes caps, cap message | |
3285 | * that releases caps also tells mds what we want. But if | |
3286 | * caps got revoked by mds forcedly (session stale). We may | |
3287 | * haven't told mds what we want. | |
3288 | */ | |
3289 | check_caps = 1; | |
a8599bd8 SW |
3290 | } |
3291 | ||
a8599bd8 SW |
3292 | /* revocation, grant, or no-op? */ |
3293 | if (cap->issued & ~newcaps) { | |
3b454c49 SW |
3294 | int revoking = cap->issued & ~newcaps; |
3295 | ||
3296 | dout("revocation: %s -> %s (revoking %s)\n", | |
3297 | ceph_cap_string(cap->issued), | |
3298 | ceph_cap_string(newcaps), | |
3299 | ceph_cap_string(revoking)); | |
0eb6cd49 | 3300 | if (revoking & used & CEPH_CAP_FILE_BUFFER) |
ab6c2c3e | 3301 | writeback = true; /* initiate writeback; will delay ack */ |
3b454c49 SW |
3302 | else if (revoking == CEPH_CAP_FILE_CACHE && |
3303 | (newcaps & CEPH_CAP_FILE_LAZYIO) == 0 && | |
3304 | queue_invalidate) | |
3305 | ; /* do nothing yet, invalidation will be queued */ | |
3306 | else if (cap == ci->i_auth_cap) | |
3307 | check_caps = 1; /* check auth cap only */ | |
3308 | else | |
3309 | check_caps = 2; /* check all caps */ | |
a8599bd8 | 3310 | cap->issued = newcaps; |
978097c9 | 3311 | cap->implemented |= newcaps; |
a8599bd8 SW |
3312 | } else if (cap->issued == newcaps) { |
3313 | dout("caps unchanged: %s -> %s\n", | |
3314 | ceph_cap_string(cap->issued), ceph_cap_string(newcaps)); | |
3315 | } else { | |
3316 | dout("grant: %s -> %s\n", ceph_cap_string(cap->issued), | |
3317 | ceph_cap_string(newcaps)); | |
6ee6b953 YZ |
3318 | /* non-auth MDS is revoking the newly grant caps ? */ |
3319 | if (cap == ci->i_auth_cap && | |
3320 | __ceph_caps_revoking_other(ci, cap, newcaps)) | |
3321 | check_caps = 2; | |
3322 | ||
a8599bd8 SW |
3323 | cap->issued = newcaps; |
3324 | cap->implemented |= newcaps; /* add bits only, to | |
3325 | * avoid stepping on a | |
3326 | * pending revocation */ | |
ab6c2c3e | 3327 | wake = true; |
a8599bd8 | 3328 | } |
978097c9 | 3329 | BUG_ON(cap->issued & ~cap->implemented); |
a8599bd8 | 3330 | |
a1c6b835 YZ |
3331 | if (extra_info->inline_version > 0 && |
3332 | extra_info->inline_version >= ci->i_inline_version) { | |
3333 | ci->i_inline_version = extra_info->inline_version; | |
31c542a1 YZ |
3334 | if (ci->i_inline_version != CEPH_INLINE_NONE && |
3335 | (newcaps & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO))) | |
3336 | fill_inline = true; | |
3337 | } | |
3338 | ||
2cd698be | 3339 | if (le32_to_cpu(grant->op) == CEPH_CAP_OP_IMPORT) { |
a1c6b835 | 3340 | if (newcaps & ~extra_info->issued) |
ab6c2c3e | 3341 | wake = true; |
a1c6b835 YZ |
3342 | kick_flushing_inode_caps(session->s_mdsc, session, inode); |
3343 | up_read(&session->s_mdsc->snap_rwsem); | |
0e294387 YZ |
3344 | } else { |
3345 | spin_unlock(&ci->i_ceph_lock); | |
2cd698be YZ |
3346 | } |
3347 | ||
31c542a1 | 3348 | if (fill_inline) |
a1c6b835 YZ |
3349 | ceph_fill_inline_data(inode, NULL, extra_info->inline_data, |
3350 | extra_info->inline_len); | |
31c542a1 | 3351 | |
14649758 | 3352 | if (queue_trunc) |
c6bcda6f | 3353 | ceph_queue_vmtruncate(inode); |
c6bcda6f | 3354 | |
3c6f6b79 | 3355 | if (writeback) |
a8599bd8 SW |
3356 | /* |
3357 | * queue inode for writeback: we can't actually call | |
3358 | * filemap_write_and_wait, etc. from message handler | |
3359 | * context. | |
3360 | */ | |
3c6f6b79 SW |
3361 | ceph_queue_writeback(inode); |
3362 | if (queue_invalidate) | |
3363 | ceph_queue_invalidate(inode); | |
ca20c991 YZ |
3364 | if (deleted_inode) |
3365 | invalidate_aliases(inode); | |
a8599bd8 | 3366 | if (wake) |
03066f23 | 3367 | wake_up_all(&ci->i_cap_wq); |
15637c8b SW |
3368 | |
3369 | if (check_caps == 1) | |
3370 | ceph_check_caps(ci, CHECK_CAPS_NODELAY|CHECK_CAPS_AUTHONLY, | |
3371 | session); | |
3372 | else if (check_caps == 2) | |
3373 | ceph_check_caps(ci, CHECK_CAPS_NODELAY, session); | |
3374 | else | |
3375 | mutex_unlock(&session->s_mutex); | |
a8599bd8 SW |
3376 | } |
3377 | ||
3378 | /* | |
3379 | * Handle FLUSH_ACK from MDS, indicating that metadata we sent to the | |
3380 | * MDS has been safely committed. | |
3381 | */ | |
6df058c0 | 3382 | static void handle_cap_flush_ack(struct inode *inode, u64 flush_tid, |
a8599bd8 SW |
3383 | struct ceph_mds_caps *m, |
3384 | struct ceph_mds_session *session, | |
3385 | struct ceph_cap *cap) | |
be655596 | 3386 | __releases(ci->i_ceph_lock) |
a8599bd8 SW |
3387 | { |
3388 | struct ceph_inode_info *ci = ceph_inode(inode); | |
3d14c5d2 | 3389 | struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc; |
e4500b5e | 3390 | struct ceph_cap_flush *cf, *tmp_cf; |
553adfd9 | 3391 | LIST_HEAD(to_remove); |
a8599bd8 SW |
3392 | unsigned seq = le32_to_cpu(m->seq); |
3393 | int dirty = le32_to_cpu(m->dirty); | |
3394 | int cleaned = 0; | |
c8799fc4 | 3395 | bool drop = false; |
7271efa7 TM |
3396 | bool wake_ci = false; |
3397 | bool wake_mdsc = false; | |
a8599bd8 | 3398 | |
e4500b5e | 3399 | list_for_each_entry_safe(cf, tmp_cf, &ci->i_cap_flush_list, i_list) { |
553adfd9 YZ |
3400 | if (cf->tid == flush_tid) |
3401 | cleaned = cf->caps; | |
0e294387 YZ |
3402 | if (cf->caps == 0) /* capsnap */ |
3403 | continue; | |
553adfd9 | 3404 | if (cf->tid <= flush_tid) { |
c8799fc4 YZ |
3405 | if (__finish_cap_flush(NULL, ci, cf)) |
3406 | wake_ci = true; | |
e4500b5e | 3407 | list_add_tail(&cf->i_list, &to_remove); |
553adfd9 YZ |
3408 | } else { |
3409 | cleaned &= ~cf->caps; | |
3410 | if (!cleaned) | |
3411 | break; | |
3412 | } | |
3413 | } | |
a8599bd8 SW |
3414 | |
3415 | dout("handle_cap_flush_ack inode %p mds%d seq %d on %s cleaned %s," | |
3416 | " flushing %s -> %s\n", | |
3417 | inode, session->s_mds, seq, ceph_cap_string(dirty), | |
3418 | ceph_cap_string(cleaned), ceph_cap_string(ci->i_flushing_caps), | |
3419 | ceph_cap_string(ci->i_flushing_caps & ~cleaned)); | |
3420 | ||
8310b089 | 3421 | if (list_empty(&to_remove) && !cleaned) |
a8599bd8 SW |
3422 | goto out; |
3423 | ||
a8599bd8 | 3424 | ci->i_flushing_caps &= ~cleaned; |
a8599bd8 SW |
3425 | |
3426 | spin_lock(&mdsc->cap_dirty_lock); | |
8310b089 | 3427 | |
c8799fc4 YZ |
3428 | list_for_each_entry(cf, &to_remove, i_list) { |
3429 | if (__finish_cap_flush(mdsc, NULL, cf)) | |
3430 | wake_mdsc = true; | |
8310b089 YZ |
3431 | } |
3432 | ||
a8599bd8 | 3433 | if (ci->i_flushing_caps == 0) { |
0e294387 YZ |
3434 | if (list_empty(&ci->i_cap_flush_list)) { |
3435 | list_del_init(&ci->i_flushing_item); | |
3436 | if (!list_empty(&session->s_cap_flushing)) { | |
3437 | dout(" mds%d still flushing cap on %p\n", | |
3438 | session->s_mds, | |
3439 | &list_first_entry(&session->s_cap_flushing, | |
3440 | struct ceph_inode_info, | |
3441 | i_flushing_item)->vfs_inode); | |
3442 | } | |
3443 | } | |
a8599bd8 | 3444 | mdsc->num_cap_flushing--; |
a8599bd8 | 3445 | dout(" inode %p now !flushing\n", inode); |
afcdaea3 SW |
3446 | |
3447 | if (ci->i_dirty_caps == 0) { | |
3448 | dout(" inode %p now clean\n", inode); | |
3449 | BUG_ON(!list_empty(&ci->i_dirty_item)); | |
c8799fc4 | 3450 | drop = true; |
5dda377c YZ |
3451 | if (ci->i_wr_ref == 0 && |
3452 | ci->i_wrbuffer_ref_head == 0) { | |
7d8cb26d SW |
3453 | BUG_ON(!ci->i_head_snapc); |
3454 | ceph_put_snap_context(ci->i_head_snapc); | |
3455 | ci->i_head_snapc = NULL; | |
3456 | } | |
76e3b390 SW |
3457 | } else { |
3458 | BUG_ON(list_empty(&ci->i_dirty_item)); | |
afcdaea3 | 3459 | } |
a8599bd8 SW |
3460 | } |
3461 | spin_unlock(&mdsc->cap_dirty_lock); | |
a8599bd8 SW |
3462 | |
3463 | out: | |
be655596 | 3464 | spin_unlock(&ci->i_ceph_lock); |
553adfd9 YZ |
3465 | |
3466 | while (!list_empty(&to_remove)) { | |
3467 | cf = list_first_entry(&to_remove, | |
e4500b5e YZ |
3468 | struct ceph_cap_flush, i_list); |
3469 | list_del(&cf->i_list); | |
f66fd9f0 | 3470 | ceph_free_cap_flush(cf); |
553adfd9 | 3471 | } |
c8799fc4 YZ |
3472 | |
3473 | if (wake_ci) | |
3474 | wake_up_all(&ci->i_cap_wq); | |
3475 | if (wake_mdsc) | |
3476 | wake_up_all(&mdsc->cap_flushing_wq); | |
afcdaea3 | 3477 | if (drop) |
a8599bd8 SW |
3478 | iput(inode); |
3479 | } | |
3480 | ||
3481 | /* | |
3482 | * Handle FLUSHSNAP_ACK. MDS has flushed snap data to disk and we can | |
3483 | * throw away our cap_snap. | |
3484 | * | |
3485 | * Caller hold s_mutex. | |
3486 | */ | |
6df058c0 | 3487 | static void handle_cap_flushsnap_ack(struct inode *inode, u64 flush_tid, |
a8599bd8 SW |
3488 | struct ceph_mds_caps *m, |
3489 | struct ceph_mds_session *session) | |
3490 | { | |
3491 | struct ceph_inode_info *ci = ceph_inode(inode); | |
affbc19a | 3492 | struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc; |
a8599bd8 | 3493 | u64 follows = le64_to_cpu(m->snap_follows); |
a8599bd8 | 3494 | struct ceph_cap_snap *capsnap; |
c8799fc4 YZ |
3495 | bool flushed = false; |
3496 | bool wake_ci = false; | |
3497 | bool wake_mdsc = false; | |
a8599bd8 SW |
3498 | |
3499 | dout("handle_cap_flushsnap_ack inode %p ci %p mds%d follows %lld\n", | |
3500 | inode, ci, session->s_mds, follows); | |
3501 | ||
be655596 | 3502 | spin_lock(&ci->i_ceph_lock); |
a8599bd8 SW |
3503 | list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) { |
3504 | if (capsnap->follows == follows) { | |
0e294387 | 3505 | if (capsnap->cap_flush.tid != flush_tid) { |
a8599bd8 SW |
3506 | dout(" cap_snap %p follows %lld tid %lld !=" |
3507 | " %lld\n", capsnap, follows, | |
0e294387 | 3508 | flush_tid, capsnap->cap_flush.tid); |
a8599bd8 SW |
3509 | break; |
3510 | } | |
c8799fc4 | 3511 | flushed = true; |
a8599bd8 SW |
3512 | break; |
3513 | } else { | |
3514 | dout(" skipping cap_snap %p follows %lld\n", | |
3515 | capsnap, capsnap->follows); | |
3516 | } | |
3517 | } | |
0e294387 | 3518 | if (flushed) { |
0e294387 YZ |
3519 | WARN_ON(capsnap->dirty_pages || capsnap->writing); |
3520 | dout(" removing %p cap_snap %p follows %lld\n", | |
3521 | inode, capsnap, follows); | |
3522 | list_del(&capsnap->ci_item); | |
c8799fc4 YZ |
3523 | if (__finish_cap_flush(NULL, ci, &capsnap->cap_flush)) |
3524 | wake_ci = true; | |
0e294387 YZ |
3525 | |
3526 | spin_lock(&mdsc->cap_dirty_lock); | |
3527 | ||
3528 | if (list_empty(&ci->i_cap_flush_list)) | |
3529 | list_del_init(&ci->i_flushing_item); | |
3530 | ||
c8799fc4 YZ |
3531 | if (__finish_cap_flush(mdsc, NULL, &capsnap->cap_flush)) |
3532 | wake_mdsc = true; | |
0e294387 YZ |
3533 | |
3534 | spin_unlock(&mdsc->cap_dirty_lock); | |
0e294387 | 3535 | } |
be655596 | 3536 | spin_unlock(&ci->i_ceph_lock); |
0e294387 YZ |
3537 | if (flushed) { |
3538 | ceph_put_snap_context(capsnap->context); | |
3539 | ceph_put_cap_snap(capsnap); | |
c8799fc4 YZ |
3540 | if (wake_ci) |
3541 | wake_up_all(&ci->i_cap_wq); | |
3542 | if (wake_mdsc) | |
3543 | wake_up_all(&mdsc->cap_flushing_wq); | |
a8599bd8 | 3544 | iput(inode); |
0e294387 | 3545 | } |
a8599bd8 SW |
3546 | } |
3547 | ||
3548 | /* | |
3549 | * Handle TRUNC from MDS, indicating file truncation. | |
3550 | * | |
3551 | * caller hold s_mutex. | |
3552 | */ | |
3553 | static void handle_cap_trunc(struct inode *inode, | |
3554 | struct ceph_mds_caps *trunc, | |
3555 | struct ceph_mds_session *session) | |
be655596 | 3556 | __releases(ci->i_ceph_lock) |
a8599bd8 SW |
3557 | { |
3558 | struct ceph_inode_info *ci = ceph_inode(inode); | |
3559 | int mds = session->s_mds; | |
3560 | int seq = le32_to_cpu(trunc->seq); | |
3561 | u32 truncate_seq = le32_to_cpu(trunc->truncate_seq); | |
3562 | u64 truncate_size = le64_to_cpu(trunc->truncate_size); | |
3563 | u64 size = le64_to_cpu(trunc->size); | |
3564 | int implemented = 0; | |
3565 | int dirty = __ceph_caps_dirty(ci); | |
3566 | int issued = __ceph_caps_issued(ceph_inode(inode), &implemented); | |
3567 | int queue_trunc = 0; | |
3568 | ||
3569 | issued |= implemented | dirty; | |
3570 | ||
3571 | dout("handle_cap_trunc inode %p mds%d seq %d to %lld seq %d\n", | |
3572 | inode, mds, seq, truncate_size, truncate_seq); | |
3573 | queue_trunc = ceph_fill_file_size(inode, issued, | |
3574 | truncate_seq, truncate_size, size); | |
be655596 | 3575 | spin_unlock(&ci->i_ceph_lock); |
a8599bd8 | 3576 | |
14649758 | 3577 | if (queue_trunc) |
3c6f6b79 | 3578 | ceph_queue_vmtruncate(inode); |
a8599bd8 SW |
3579 | } |
3580 | ||
3581 | /* | |
3582 | * Handle EXPORT from MDS. Cap is being migrated _from_ this mds to a | |
3583 | * different one. If we are the most recent migration we've seen (as | |
3584 | * indicated by mseq), make note of the migrating cap bits for the | |
3585 | * duration (until we see the corresponding IMPORT). | |
3586 | * | |
3587 | * caller holds s_mutex | |
3588 | */ | |
3589 | static void handle_cap_export(struct inode *inode, struct ceph_mds_caps *ex, | |
11df2dfb YZ |
3590 | struct ceph_mds_cap_peer *ph, |
3591 | struct ceph_mds_session *session) | |
a8599bd8 | 3592 | { |
db354052 | 3593 | struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc; |
11df2dfb | 3594 | struct ceph_mds_session *tsession = NULL; |
d9df2783 | 3595 | struct ceph_cap *cap, *tcap, *new_cap = NULL; |
a8599bd8 | 3596 | struct ceph_inode_info *ci = ceph_inode(inode); |
11df2dfb | 3597 | u64 t_cap_id; |
a8599bd8 | 3598 | unsigned mseq = le32_to_cpu(ex->migrate_seq); |
11df2dfb YZ |
3599 | unsigned t_seq, t_mseq; |
3600 | int target, issued; | |
3601 | int mds = session->s_mds; | |
a8599bd8 | 3602 | |
11df2dfb YZ |
3603 | if (ph) { |
3604 | t_cap_id = le64_to_cpu(ph->cap_id); | |
3605 | t_seq = le32_to_cpu(ph->seq); | |
3606 | t_mseq = le32_to_cpu(ph->mseq); | |
3607 | target = le32_to_cpu(ph->mds); | |
3608 | } else { | |
3609 | t_cap_id = t_seq = t_mseq = 0; | |
3610 | target = -1; | |
3611 | } | |
a8599bd8 | 3612 | |
11df2dfb YZ |
3613 | dout("handle_cap_export inode %p ci %p mds%d mseq %d target %d\n", |
3614 | inode, ci, mds, mseq, target); | |
3615 | retry: | |
be655596 | 3616 | spin_lock(&ci->i_ceph_lock); |
11df2dfb | 3617 | cap = __get_cap_for_mds(ci, mds); |
ca665e02 | 3618 | if (!cap || cap->cap_id != le64_to_cpu(ex->cap_id)) |
11df2dfb | 3619 | goto out_unlock; |
a8599bd8 | 3620 | |
11df2dfb | 3621 | if (target < 0) { |
d2f8bb27 | 3622 | if (cap->mds_wanted | cap->issued) |
77310320 | 3623 | ci->i_ceph_flags |= CEPH_I_CAP_DROPPED; |
d2f8bb27 | 3624 | __ceph_remove_cap(cap, false); |
11df2dfb | 3625 | goto out_unlock; |
a8599bd8 SW |
3626 | } |
3627 | ||
11df2dfb YZ |
3628 | /* |
3629 | * now we know we haven't received the cap import message yet | |
3630 | * because the exported cap still exist. | |
3631 | */ | |
db354052 | 3632 | |
11df2dfb | 3633 | issued = cap->issued; |
d84b37f9 YZ |
3634 | if (issued != cap->implemented) |
3635 | pr_err_ratelimited("handle_cap_export: issued != implemented: " | |
3636 | "ino (%llx.%llx) mds%d seq %d mseq %d " | |
3637 | "issued %s implemented %s\n", | |
3638 | ceph_vinop(inode), mds, cap->seq, cap->mseq, | |
3639 | ceph_cap_string(issued), | |
3640 | ceph_cap_string(cap->implemented)); | |
3641 | ||
11df2dfb YZ |
3642 | |
3643 | tcap = __get_cap_for_mds(ci, target); | |
3644 | if (tcap) { | |
3645 | /* already have caps from the target */ | |
fa0aa3b8 | 3646 | if (tcap->cap_id == t_cap_id && |
11df2dfb YZ |
3647 | ceph_seq_cmp(tcap->seq, t_seq) < 0) { |
3648 | dout(" updating import cap %p mds%d\n", tcap, target); | |
3649 | tcap->cap_id = t_cap_id; | |
3650 | tcap->seq = t_seq - 1; | |
3651 | tcap->issue_seq = t_seq - 1; | |
11df2dfb YZ |
3652 | tcap->issued |= issued; |
3653 | tcap->implemented |= issued; | |
3654 | if (cap == ci->i_auth_cap) | |
3655 | ci->i_auth_cap = tcap; | |
00f06cba | 3656 | |
0e294387 YZ |
3657 | if (!list_empty(&ci->i_cap_flush_list) && |
3658 | ci->i_auth_cap == tcap) { | |
11df2dfb YZ |
3659 | spin_lock(&mdsc->cap_dirty_lock); |
3660 | list_move_tail(&ci->i_flushing_item, | |
3661 | &tcap->session->s_cap_flushing); | |
3662 | spin_unlock(&mdsc->cap_dirty_lock); | |
db354052 | 3663 | } |
a8599bd8 | 3664 | } |
a096b09a | 3665 | __ceph_remove_cap(cap, false); |
11df2dfb | 3666 | goto out_unlock; |
d9df2783 | 3667 | } else if (tsession) { |
11df2dfb | 3668 | /* add placeholder for the export tagert */ |
d9df2783 | 3669 | int flag = (cap == ci->i_auth_cap) ? CEPH_CAP_FLAG_AUTH : 0; |
00f06cba | 3670 | tcap = new_cap; |
11df2dfb | 3671 | ceph_add_cap(inode, tsession, t_cap_id, -1, issued, 0, |
d9df2783 YZ |
3672 | t_seq - 1, t_mseq, (u64)-1, flag, &new_cap); |
3673 | ||
00f06cba YZ |
3674 | if (!list_empty(&ci->i_cap_flush_list) && |
3675 | ci->i_auth_cap == tcap) { | |
3676 | spin_lock(&mdsc->cap_dirty_lock); | |
3677 | list_move_tail(&ci->i_flushing_item, | |
3678 | &tcap->session->s_cap_flushing); | |
3679 | spin_unlock(&mdsc->cap_dirty_lock); | |
3680 | } | |
3681 | ||
d9df2783 YZ |
3682 | __ceph_remove_cap(cap, false); |
3683 | goto out_unlock; | |
a8599bd8 SW |
3684 | } |
3685 | ||
be655596 | 3686 | spin_unlock(&ci->i_ceph_lock); |
11df2dfb YZ |
3687 | mutex_unlock(&session->s_mutex); |
3688 | ||
3689 | /* open target session */ | |
3690 | tsession = ceph_mdsc_open_export_target_session(mdsc, target); | |
3691 | if (!IS_ERR(tsession)) { | |
3692 | if (mds > target) { | |
3693 | mutex_lock(&session->s_mutex); | |
3694 | mutex_lock_nested(&tsession->s_mutex, | |
3695 | SINGLE_DEPTH_NESTING); | |
3696 | } else { | |
3697 | mutex_lock(&tsession->s_mutex); | |
3698 | mutex_lock_nested(&session->s_mutex, | |
3699 | SINGLE_DEPTH_NESTING); | |
3700 | } | |
d9df2783 | 3701 | new_cap = ceph_get_cap(mdsc, NULL); |
11df2dfb YZ |
3702 | } else { |
3703 | WARN_ON(1); | |
3704 | tsession = NULL; | |
3705 | target = -1; | |
3706 | } | |
3707 | goto retry; | |
3708 | ||
3709 | out_unlock: | |
3710 | spin_unlock(&ci->i_ceph_lock); | |
3711 | mutex_unlock(&session->s_mutex); | |
3712 | if (tsession) { | |
3713 | mutex_unlock(&tsession->s_mutex); | |
3714 | ceph_put_mds_session(tsession); | |
3715 | } | |
d9df2783 YZ |
3716 | if (new_cap) |
3717 | ceph_put_cap(mdsc, new_cap); | |
a8599bd8 SW |
3718 | } |
3719 | ||
3720 | /* | |
2cd698be | 3721 | * Handle cap IMPORT. |
a8599bd8 | 3722 | * |
2cd698be | 3723 | * caller holds s_mutex. acquires i_ceph_lock |
a8599bd8 SW |
3724 | */ |
3725 | static void handle_cap_import(struct ceph_mds_client *mdsc, | |
3726 | struct inode *inode, struct ceph_mds_caps *im, | |
4ee6a914 | 3727 | struct ceph_mds_cap_peer *ph, |
a8599bd8 | 3728 | struct ceph_mds_session *session, |
2cd698be YZ |
3729 | struct ceph_cap **target_cap, int *old_issued) |
3730 | __acquires(ci->i_ceph_lock) | |
a8599bd8 SW |
3731 | { |
3732 | struct ceph_inode_info *ci = ceph_inode(inode); | |
2cd698be | 3733 | struct ceph_cap *cap, *ocap, *new_cap = NULL; |
a8599bd8 | 3734 | int mds = session->s_mds; |
2cd698be YZ |
3735 | int issued; |
3736 | unsigned caps = le32_to_cpu(im->caps); | |
a8599bd8 SW |
3737 | unsigned wanted = le32_to_cpu(im->wanted); |
3738 | unsigned seq = le32_to_cpu(im->seq); | |
3739 | unsigned mseq = le32_to_cpu(im->migrate_seq); | |
3740 | u64 realmino = le64_to_cpu(im->realm); | |
3741 | u64 cap_id = le64_to_cpu(im->cap_id); | |
4ee6a914 YZ |
3742 | u64 p_cap_id; |
3743 | int peer; | |
a8599bd8 | 3744 | |
4ee6a914 YZ |
3745 | if (ph) { |
3746 | p_cap_id = le64_to_cpu(ph->cap_id); | |
3747 | peer = le32_to_cpu(ph->mds); | |
3748 | } else { | |
3749 | p_cap_id = 0; | |
3750 | peer = -1; | |
3751 | } | |
db354052 | 3752 | |
4ee6a914 YZ |
3753 | dout("handle_cap_import inode %p ci %p mds%d mseq %d peer %d\n", |
3754 | inode, ci, mds, mseq, peer); | |
3755 | ||
d9df2783 | 3756 | retry: |
4ee6a914 | 3757 | spin_lock(&ci->i_ceph_lock); |
d9df2783 YZ |
3758 | cap = __get_cap_for_mds(ci, mds); |
3759 | if (!cap) { | |
3760 | if (!new_cap) { | |
3761 | spin_unlock(&ci->i_ceph_lock); | |
3762 | new_cap = ceph_get_cap(mdsc, NULL); | |
3763 | goto retry; | |
3764 | } | |
2cd698be YZ |
3765 | cap = new_cap; |
3766 | } else { | |
3767 | if (new_cap) { | |
3768 | ceph_put_cap(mdsc, new_cap); | |
3769 | new_cap = NULL; | |
3770 | } | |
d9df2783 YZ |
3771 | } |
3772 | ||
2cd698be YZ |
3773 | __ceph_caps_issued(ci, &issued); |
3774 | issued |= __ceph_caps_dirty(ci); | |
3775 | ||
3776 | ceph_add_cap(inode, session, cap_id, -1, caps, wanted, seq, mseq, | |
d9df2783 YZ |
3777 | realmino, CEPH_CAP_FLAG_AUTH, &new_cap); |
3778 | ||
2cd698be YZ |
3779 | ocap = peer >= 0 ? __get_cap_for_mds(ci, peer) : NULL; |
3780 | if (ocap && ocap->cap_id == p_cap_id) { | |
4ee6a914 | 3781 | dout(" remove export cap %p mds%d flags %d\n", |
2cd698be | 3782 | ocap, peer, ph->flags); |
4ee6a914 | 3783 | if ((ph->flags & CEPH_CAP_FLAG_AUTH) && |
2cd698be YZ |
3784 | (ocap->seq != le32_to_cpu(ph->seq) || |
3785 | ocap->mseq != le32_to_cpu(ph->mseq))) { | |
d84b37f9 YZ |
3786 | pr_err_ratelimited("handle_cap_import: " |
3787 | "mismatched seq/mseq: ino (%llx.%llx) " | |
3788 | "mds%d seq %d mseq %d importer mds%d " | |
3789 | "has peer seq %d mseq %d\n", | |
3790 | ceph_vinop(inode), peer, ocap->seq, | |
3791 | ocap->mseq, mds, le32_to_cpu(ph->seq), | |
3792 | le32_to_cpu(ph->mseq)); | |
db354052 | 3793 | } |
2cd698be | 3794 | __ceph_remove_cap(ocap, (ph->flags & CEPH_CAP_FLAG_RELEASE)); |
a8599bd8 SW |
3795 | } |
3796 | ||
4ee6a914 | 3797 | /* make sure we re-request max_size, if necessary */ |
4ee6a914 | 3798 | ci->i_requested_max_size = 0; |
d9df2783 | 3799 | |
2cd698be YZ |
3800 | *old_issued = issued; |
3801 | *target_cap = cap; | |
a8599bd8 SW |
3802 | } |
3803 | ||
3804 | /* | |
3805 | * Handle a caps message from the MDS. | |
3806 | * | |
3807 | * Identify the appropriate session, inode, and call the right handler | |
3808 | * based on the cap op. | |
3809 | */ | |
3810 | void ceph_handle_caps(struct ceph_mds_session *session, | |
3811 | struct ceph_msg *msg) | |
3812 | { | |
3813 | struct ceph_mds_client *mdsc = session->s_mdsc; | |
a8599bd8 | 3814 | struct inode *inode; |
be655596 | 3815 | struct ceph_inode_info *ci; |
a8599bd8 SW |
3816 | struct ceph_cap *cap; |
3817 | struct ceph_mds_caps *h; | |
4ee6a914 | 3818 | struct ceph_mds_cap_peer *peer = NULL; |
779fe0fb | 3819 | struct ceph_snap_realm *realm = NULL; |
a1c6b835 | 3820 | int op; |
4985d6f9 | 3821 | int msg_version = le16_to_cpu(msg->hdr.version); |
3d7ded4d | 3822 | u32 seq, mseq; |
a8599bd8 | 3823 | struct ceph_vino vino; |
70edb55b | 3824 | void *snaptrace; |
ce1fbc8d | 3825 | size_t snaptrace_len; |
fb01d1f8 | 3826 | void *p, *end; |
a1c6b835 | 3827 | struct cap_extra_info extra_info = {}; |
a8599bd8 | 3828 | |
a1c6b835 | 3829 | dout("handle_caps from mds%d\n", session->s_mds); |
a8599bd8 SW |
3830 | |
3831 | /* decode */ | |
4ee6a914 | 3832 | end = msg->front.iov_base + msg->front.iov_len; |
a8599bd8 SW |
3833 | if (msg->front.iov_len < sizeof(*h)) |
3834 | goto bad; | |
3835 | h = msg->front.iov_base; | |
3836 | op = le32_to_cpu(h->op); | |
3837 | vino.ino = le64_to_cpu(h->ino); | |
3838 | vino.snap = CEPH_NOSNAP; | |
a8599bd8 | 3839 | seq = le32_to_cpu(h->seq); |
3d7ded4d | 3840 | mseq = le32_to_cpu(h->migrate_seq); |
a8599bd8 | 3841 | |
ce1fbc8d SW |
3842 | snaptrace = h + 1; |
3843 | snaptrace_len = le32_to_cpu(h->snap_trace_len); | |
fb01d1f8 | 3844 | p = snaptrace + snaptrace_len; |
ce1fbc8d | 3845 | |
4985d6f9 | 3846 | if (msg_version >= 2) { |
fb01d1f8 | 3847 | u32 flock_len; |
ce1fbc8d | 3848 | ceph_decode_32_safe(&p, end, flock_len, bad); |
4ee6a914 YZ |
3849 | if (p + flock_len > end) |
3850 | goto bad; | |
fb01d1f8 | 3851 | p += flock_len; |
ce1fbc8d SW |
3852 | } |
3853 | ||
4985d6f9 | 3854 | if (msg_version >= 3) { |
4ee6a914 | 3855 | if (op == CEPH_CAP_OP_IMPORT) { |
4ee6a914 YZ |
3856 | if (p + sizeof(*peer) > end) |
3857 | goto bad; | |
3858 | peer = p; | |
fb01d1f8 | 3859 | p += sizeof(*peer); |
11df2dfb YZ |
3860 | } else if (op == CEPH_CAP_OP_EXPORT) { |
3861 | /* recorded in unused fields */ | |
3862 | peer = (void *)&h->size; | |
4ee6a914 YZ |
3863 | } |
3864 | } | |
3865 | ||
4985d6f9 | 3866 | if (msg_version >= 4) { |
a1c6b835 YZ |
3867 | ceph_decode_64_safe(&p, end, extra_info.inline_version, bad); |
3868 | ceph_decode_32_safe(&p, end, extra_info.inline_len, bad); | |
3869 | if (p + extra_info.inline_len > end) | |
fb01d1f8 | 3870 | goto bad; |
a1c6b835 YZ |
3871 | extra_info.inline_data = p; |
3872 | p += extra_info.inline_len; | |
fb01d1f8 YZ |
3873 | } |
3874 | ||
4985d6f9 | 3875 | if (msg_version >= 5) { |
92475f05 JL |
3876 | struct ceph_osd_client *osdc = &mdsc->fsc->client->osdc; |
3877 | u32 epoch_barrier; | |
3878 | ||
3879 | ceph_decode_32_safe(&p, end, epoch_barrier, bad); | |
3880 | ceph_osdc_update_epoch_barrier(osdc, epoch_barrier); | |
3881 | } | |
3882 | ||
4985d6f9 | 3883 | if (msg_version >= 8) { |
5ea5c5e0 YZ |
3884 | u64 flush_tid; |
3885 | u32 caller_uid, caller_gid; | |
779fe0fb | 3886 | u32 pool_ns_len; |
92475f05 | 3887 | |
5ea5c5e0 YZ |
3888 | /* version >= 6 */ |
3889 | ceph_decode_64_safe(&p, end, flush_tid, bad); | |
3890 | /* version >= 7 */ | |
3891 | ceph_decode_32_safe(&p, end, caller_uid, bad); | |
3892 | ceph_decode_32_safe(&p, end, caller_gid, bad); | |
3893 | /* version >= 8 */ | |
3894 | ceph_decode_32_safe(&p, end, pool_ns_len, bad); | |
779fe0fb YZ |
3895 | if (pool_ns_len > 0) { |
3896 | ceph_decode_need(&p, end, pool_ns_len, bad); | |
a1c6b835 YZ |
3897 | extra_info.pool_ns = |
3898 | ceph_find_or_create_string(p, pool_ns_len); | |
779fe0fb YZ |
3899 | p += pool_ns_len; |
3900 | } | |
5ea5c5e0 YZ |
3901 | } |
3902 | ||
ec62b894 | 3903 | if (msg_version >= 9) { |
4985d6f9 | 3904 | struct ceph_timespec *btime; |
4985d6f9 | 3905 | |
4985d6f9 YZ |
3906 | if (p + sizeof(*btime) > end) |
3907 | goto bad; | |
3908 | btime = p; | |
ec62b894 | 3909 | ceph_decode_timespec64(&extra_info.btime, btime); |
4985d6f9 | 3910 | p += sizeof(*btime); |
176c77c9 | 3911 | ceph_decode_64_safe(&p, end, extra_info.change_attr, bad); |
ec62b894 JL |
3912 | } |
3913 | ||
3914 | if (msg_version >= 11) { | |
3915 | u32 flags; | |
4985d6f9 YZ |
3916 | /* version >= 10 */ |
3917 | ceph_decode_32_safe(&p, end, flags, bad); | |
3918 | /* version >= 11 */ | |
3919 | extra_info.dirstat_valid = true; | |
3920 | ceph_decode_64_safe(&p, end, extra_info.nfiles, bad); | |
3921 | ceph_decode_64_safe(&p, end, extra_info.nsubdirs, bad); | |
3922 | } | |
3923 | ||
6cd3bcad | 3924 | /* lookup ino */ |
a1c6b835 | 3925 | inode = ceph_find_inode(mdsc->fsc->sb, vino); |
6cd3bcad YZ |
3926 | ci = ceph_inode(inode); |
3927 | dout(" op %s ino %llx.%llx inode %p\n", ceph_cap_op_name(op), vino.ino, | |
3928 | vino.snap, inode); | |
3929 | ||
a8599bd8 SW |
3930 | mutex_lock(&session->s_mutex); |
3931 | session->s_seq++; | |
3932 | dout(" mds%d seq %lld cap seq %u\n", session->s_mds, session->s_seq, | |
3933 | (unsigned)seq); | |
3934 | ||
a8599bd8 SW |
3935 | if (!inode) { |
3936 | dout(" i don't have ino %llx\n", vino.ino); | |
3d7ded4d | 3937 | |
a096b09a | 3938 | if (op == CEPH_CAP_OP_IMPORT) { |
745a8e3b YZ |
3939 | cap = ceph_get_cap(mdsc, NULL); |
3940 | cap->cap_ino = vino.ino; | |
3941 | cap->queue_release = 1; | |
779fe0fb | 3942 | cap->cap_id = le64_to_cpu(h->cap_id); |
745a8e3b YZ |
3943 | cap->mseq = mseq; |
3944 | cap->seq = seq; | |
dc24de82 | 3945 | cap->issue_seq = seq; |
a096b09a | 3946 | spin_lock(&session->s_cap_lock); |
e3ec8d68 | 3947 | __ceph_queue_cap_release(session, cap); |
a096b09a YZ |
3948 | spin_unlock(&session->s_cap_lock); |
3949 | } | |
e3ec8d68 | 3950 | goto done; |
a8599bd8 SW |
3951 | } |
3952 | ||
3953 | /* these will work even if we don't have a cap yet */ | |
3954 | switch (op) { | |
3955 | case CEPH_CAP_OP_FLUSHSNAP_ACK: | |
a1c6b835 YZ |
3956 | handle_cap_flushsnap_ack(inode, le64_to_cpu(msg->hdr.tid), |
3957 | h, session); | |
a8599bd8 SW |
3958 | goto done; |
3959 | ||
3960 | case CEPH_CAP_OP_EXPORT: | |
11df2dfb YZ |
3961 | handle_cap_export(inode, h, peer, session); |
3962 | goto done_unlocked; | |
a8599bd8 SW |
3963 | |
3964 | case CEPH_CAP_OP_IMPORT: | |
982d6011 YZ |
3965 | realm = NULL; |
3966 | if (snaptrace_len) { | |
3967 | down_write(&mdsc->snap_rwsem); | |
3968 | ceph_update_snap_trace(mdsc, snaptrace, | |
3969 | snaptrace + snaptrace_len, | |
3970 | false, &realm); | |
3971 | downgrade_write(&mdsc->snap_rwsem); | |
3972 | } else { | |
3973 | down_read(&mdsc->snap_rwsem); | |
3974 | } | |
4ee6a914 | 3975 | handle_cap_import(mdsc, inode, h, peer, session, |
a1c6b835 YZ |
3976 | &cap, &extra_info.issued); |
3977 | handle_cap_grant(inode, session, cap, | |
3978 | h, msg->middle, &extra_info); | |
982d6011 YZ |
3979 | if (realm) |
3980 | ceph_put_snap_realm(mdsc, realm); | |
2cd698be | 3981 | goto done_unlocked; |
a8599bd8 SW |
3982 | } |
3983 | ||
3984 | /* the rest require a cap */ | |
be655596 | 3985 | spin_lock(&ci->i_ceph_lock); |
a1c6b835 | 3986 | cap = __get_cap_for_mds(ceph_inode(inode), session->s_mds); |
a8599bd8 | 3987 | if (!cap) { |
9dbd412f | 3988 | dout(" no cap on %p ino %llx.%llx from mds%d\n", |
a1c6b835 YZ |
3989 | inode, ceph_ino(inode), ceph_snap(inode), |
3990 | session->s_mds); | |
be655596 | 3991 | spin_unlock(&ci->i_ceph_lock); |
21b559de | 3992 | goto flush_cap_releases; |
a8599bd8 SW |
3993 | } |
3994 | ||
be655596 | 3995 | /* note that each of these drops i_ceph_lock for us */ |
a8599bd8 SW |
3996 | switch (op) { |
3997 | case CEPH_CAP_OP_REVOKE: | |
3998 | case CEPH_CAP_OP_GRANT: | |
a1c6b835 YZ |
3999 | __ceph_caps_issued(ci, &extra_info.issued); |
4000 | extra_info.issued |= __ceph_caps_dirty(ci); | |
4001 | handle_cap_grant(inode, session, cap, | |
4002 | h, msg->middle, &extra_info); | |
15637c8b | 4003 | goto done_unlocked; |
a8599bd8 SW |
4004 | |
4005 | case CEPH_CAP_OP_FLUSH_ACK: | |
a1c6b835 YZ |
4006 | handle_cap_flush_ack(inode, le64_to_cpu(msg->hdr.tid), |
4007 | h, session, cap); | |
a8599bd8 SW |
4008 | break; |
4009 | ||
4010 | case CEPH_CAP_OP_TRUNC: | |
4011 | handle_cap_trunc(inode, h, session); | |
4012 | break; | |
4013 | ||
4014 | default: | |
be655596 | 4015 | spin_unlock(&ci->i_ceph_lock); |
a8599bd8 SW |
4016 | pr_err("ceph_handle_caps: unknown cap op %d %s\n", op, |
4017 | ceph_cap_op_name(op)); | |
4018 | } | |
4019 | ||
e3ec8d68 YZ |
4020 | done: |
4021 | mutex_unlock(&session->s_mutex); | |
4022 | done_unlocked: | |
e3ec8d68 | 4023 | ceph_put_string(extra_info.pool_ns); |
3e1d0452 YZ |
4024 | /* avoid calling iput_final() in mds dispatch threads */ |
4025 | ceph_async_iput(inode); | |
e3ec8d68 | 4026 | return; |
21b559de GF |
4027 | |
4028 | flush_cap_releases: | |
4029 | /* | |
745a8e3b | 4030 | * send any cap release message to try to move things |
21b559de GF |
4031 | * along for the mds (who clearly thinks we still have this |
4032 | * cap). | |
4033 | */ | |
e3ec8d68 YZ |
4034 | ceph_flush_cap_releases(mdsc, session); |
4035 | goto done; | |
a8599bd8 SW |
4036 | |
4037 | bad: | |
4038 | pr_err("ceph_handle_caps: corrupt message\n"); | |
9ec7cab1 | 4039 | ceph_msg_dump(msg); |
a8599bd8 SW |
4040 | return; |
4041 | } | |
4042 | ||
4043 | /* | |
4044 | * Delayed work handler to process end of delayed cap release LRU list. | |
4045 | */ | |
afcdaea3 | 4046 | void ceph_check_delayed_caps(struct ceph_mds_client *mdsc) |
a8599bd8 | 4047 | { |
4b9f2042 | 4048 | struct inode *inode; |
a8599bd8 SW |
4049 | struct ceph_inode_info *ci; |
4050 | int flags = CHECK_CAPS_NODELAY; | |
4051 | ||
a8599bd8 SW |
4052 | dout("check_delayed_caps\n"); |
4053 | while (1) { | |
4054 | spin_lock(&mdsc->cap_delay_lock); | |
4055 | if (list_empty(&mdsc->cap_delay_list)) | |
4056 | break; | |
4057 | ci = list_first_entry(&mdsc->cap_delay_list, | |
4058 | struct ceph_inode_info, | |
4059 | i_cap_delay_list); | |
4060 | if ((ci->i_ceph_flags & CEPH_I_FLUSH) == 0 && | |
4061 | time_before(jiffies, ci->i_hold_caps_max)) | |
4062 | break; | |
4063 | list_del_init(&ci->i_cap_delay_list); | |
4b9f2042 YZ |
4064 | |
4065 | inode = igrab(&ci->vfs_inode); | |
a8599bd8 | 4066 | spin_unlock(&mdsc->cap_delay_lock); |
4b9f2042 YZ |
4067 | |
4068 | if (inode) { | |
4069 | dout("check_delayed_caps on %p\n", inode); | |
4070 | ceph_check_caps(ci, flags, NULL); | |
3e1d0452 YZ |
4071 | /* avoid calling iput_final() in tick thread */ |
4072 | ceph_async_iput(inode); | |
4b9f2042 | 4073 | } |
a8599bd8 SW |
4074 | } |
4075 | spin_unlock(&mdsc->cap_delay_lock); | |
4076 | } | |
4077 | ||
afcdaea3 SW |
4078 | /* |
4079 | * Flush all dirty caps to the mds | |
4080 | */ | |
4081 | void ceph_flush_dirty_caps(struct ceph_mds_client *mdsc) | |
4082 | { | |
db354052 SW |
4083 | struct ceph_inode_info *ci; |
4084 | struct inode *inode; | |
afcdaea3 SW |
4085 | |
4086 | dout("flush_dirty_caps\n"); | |
4087 | spin_lock(&mdsc->cap_dirty_lock); | |
db354052 SW |
4088 | while (!list_empty(&mdsc->cap_dirty)) { |
4089 | ci = list_first_entry(&mdsc->cap_dirty, struct ceph_inode_info, | |
4090 | i_dirty_item); | |
70b666c3 SW |
4091 | inode = &ci->vfs_inode; |
4092 | ihold(inode); | |
db354052 | 4093 | dout("flush_dirty_caps %p\n", inode); |
afcdaea3 | 4094 | spin_unlock(&mdsc->cap_dirty_lock); |
70b666c3 SW |
4095 | ceph_check_caps(ci, CHECK_CAPS_NODELAY|CHECK_CAPS_FLUSH, NULL); |
4096 | iput(inode); | |
afcdaea3 SW |
4097 | spin_lock(&mdsc->cap_dirty_lock); |
4098 | } | |
4099 | spin_unlock(&mdsc->cap_dirty_lock); | |
db354052 | 4100 | dout("flush_dirty_caps done\n"); |
afcdaea3 SW |
4101 | } |
4102 | ||
774a6a11 YZ |
4103 | void __ceph_get_fmode(struct ceph_inode_info *ci, int fmode) |
4104 | { | |
4105 | int i; | |
4106 | int bits = (fmode << 1) | 1; | |
4107 | for (i = 0; i < CEPH_FILE_MODE_BITS; i++) { | |
4108 | if (bits & (1 << i)) | |
4109 | ci->i_nr_by_mode[i]++; | |
4110 | } | |
4111 | } | |
4112 | ||
a8599bd8 SW |
4113 | /* |
4114 | * Drop open file reference. If we were the last open file, | |
4115 | * we may need to release capabilities to the MDS (or schedule | |
4116 | * their delayed release). | |
4117 | */ | |
4118 | void ceph_put_fmode(struct ceph_inode_info *ci, int fmode) | |
4119 | { | |
774a6a11 YZ |
4120 | int i, last = 0; |
4121 | int bits = (fmode << 1) | 1; | |
be655596 | 4122 | spin_lock(&ci->i_ceph_lock); |
774a6a11 YZ |
4123 | for (i = 0; i < CEPH_FILE_MODE_BITS; i++) { |
4124 | if (bits & (1 << i)) { | |
4125 | BUG_ON(ci->i_nr_by_mode[i] == 0); | |
4126 | if (--ci->i_nr_by_mode[i] == 0) | |
4127 | last++; | |
4128 | } | |
4129 | } | |
4130 | dout("put_fmode %p fmode %d {%d,%d,%d,%d}\n", | |
4131 | &ci->vfs_inode, fmode, | |
4132 | ci->i_nr_by_mode[0], ci->i_nr_by_mode[1], | |
4133 | ci->i_nr_by_mode[2], ci->i_nr_by_mode[3]); | |
be655596 | 4134 | spin_unlock(&ci->i_ceph_lock); |
a8599bd8 SW |
4135 | |
4136 | if (last && ci->i_vino.snap == CEPH_NOSNAP) | |
4137 | ceph_check_caps(ci, 0, NULL); | |
4138 | } | |
4139 | ||
6ef0bc6d | 4140 | /* |
a452bc06 | 4141 | * For a soon-to-be unlinked file, drop the LINK caps. If it |
6ef0bc6d ZZ |
4142 | * looks like the link count will hit 0, drop any other caps (other |
4143 | * than PIN) we don't specifically want (due to the file still being | |
4144 | * open). | |
4145 | */ | |
4146 | int ceph_drop_caps_for_unlink(struct inode *inode) | |
4147 | { | |
4148 | struct ceph_inode_info *ci = ceph_inode(inode); | |
4149 | int drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL; | |
4150 | ||
4151 | spin_lock(&ci->i_ceph_lock); | |
4152 | if (inode->i_nlink == 1) { | |
4153 | drop |= ~(__ceph_caps_wanted(ci) | CEPH_CAP_PIN); | |
4154 | ||
4155 | ci->i_ceph_flags |= CEPH_I_NODELAY; | |
4156 | if (__ceph_caps_dirty(ci)) { | |
4157 | struct ceph_mds_client *mdsc = | |
4158 | ceph_inode_to_client(inode)->mdsc; | |
4159 | __cap_delay_requeue_front(mdsc, ci); | |
4160 | } | |
4161 | } | |
4162 | spin_unlock(&ci->i_ceph_lock); | |
4163 | return drop; | |
4164 | } | |
4165 | ||
a8599bd8 SW |
4166 | /* |
4167 | * Helpers for embedding cap and dentry lease releases into mds | |
4168 | * requests. | |
4169 | * | |
4170 | * @force is used by dentry_release (below) to force inclusion of a | |
4171 | * record for the directory inode, even when there aren't any caps to | |
4172 | * drop. | |
4173 | */ | |
4174 | int ceph_encode_inode_release(void **p, struct inode *inode, | |
4175 | int mds, int drop, int unless, int force) | |
4176 | { | |
4177 | struct ceph_inode_info *ci = ceph_inode(inode); | |
4178 | struct ceph_cap *cap; | |
4179 | struct ceph_mds_request_release *rel = *p; | |
ec97f88b | 4180 | int used, dirty; |
a8599bd8 | 4181 | int ret = 0; |
a8599bd8 | 4182 | |
be655596 | 4183 | spin_lock(&ci->i_ceph_lock); |
916623da | 4184 | used = __ceph_caps_used(ci); |
ec97f88b | 4185 | dirty = __ceph_caps_dirty(ci); |
916623da | 4186 | |
ec97f88b SW |
4187 | dout("encode_inode_release %p mds%d used|dirty %s drop %s unless %s\n", |
4188 | inode, mds, ceph_cap_string(used|dirty), ceph_cap_string(drop), | |
916623da SW |
4189 | ceph_cap_string(unless)); |
4190 | ||
ec97f88b SW |
4191 | /* only drop unused, clean caps */ |
4192 | drop &= ~(used | dirty); | |
916623da | 4193 | |
a8599bd8 SW |
4194 | cap = __get_cap_for_mds(ci, mds); |
4195 | if (cap && __cap_is_valid(cap)) { | |
222b7f90 YZ |
4196 | unless &= cap->issued; |
4197 | if (unless) { | |
4198 | if (unless & CEPH_CAP_AUTH_EXCL) | |
4199 | drop &= ~CEPH_CAP_AUTH_SHARED; | |
4200 | if (unless & CEPH_CAP_LINK_EXCL) | |
4201 | drop &= ~CEPH_CAP_LINK_SHARED; | |
4202 | if (unless & CEPH_CAP_XATTR_EXCL) | |
4203 | drop &= ~CEPH_CAP_XATTR_SHARED; | |
4204 | if (unless & CEPH_CAP_FILE_EXCL) | |
4205 | drop &= ~CEPH_CAP_FILE_SHARED; | |
4206 | } | |
4207 | ||
4208 | if (force || (cap->issued & drop)) { | |
4209 | if (cap->issued & drop) { | |
bb137f84 YZ |
4210 | int wanted = __ceph_caps_wanted(ci); |
4211 | if ((ci->i_ceph_flags & CEPH_I_NODELAY) == 0) | |
4212 | wanted |= cap->mds_wanted; | |
4213 | dout("encode_inode_release %p cap %p " | |
4214 | "%s -> %s, wanted %s -> %s\n", inode, cap, | |
a8599bd8 | 4215 | ceph_cap_string(cap->issued), |
bb137f84 YZ |
4216 | ceph_cap_string(cap->issued & ~drop), |
4217 | ceph_cap_string(cap->mds_wanted), | |
4218 | ceph_cap_string(wanted)); | |
4219 | ||
a8599bd8 SW |
4220 | cap->issued &= ~drop; |
4221 | cap->implemented &= ~drop; | |
bb137f84 | 4222 | cap->mds_wanted = wanted; |
a8599bd8 SW |
4223 | } else { |
4224 | dout("encode_inode_release %p cap %p %s" | |
4225 | " (force)\n", inode, cap, | |
4226 | ceph_cap_string(cap->issued)); | |
4227 | } | |
4228 | ||
4229 | rel->ino = cpu_to_le64(ceph_ino(inode)); | |
4230 | rel->cap_id = cpu_to_le64(cap->cap_id); | |
4231 | rel->seq = cpu_to_le32(cap->seq); | |
08a0f24e | 4232 | rel->issue_seq = cpu_to_le32(cap->issue_seq); |
a8599bd8 | 4233 | rel->mseq = cpu_to_le32(cap->mseq); |
fd7b95cd | 4234 | rel->caps = cpu_to_le32(cap->implemented); |
a8599bd8 SW |
4235 | rel->wanted = cpu_to_le32(cap->mds_wanted); |
4236 | rel->dname_len = 0; | |
4237 | rel->dname_seq = 0; | |
4238 | *p += sizeof(*rel); | |
4239 | ret = 1; | |
4240 | } else { | |
222b7f90 | 4241 | dout("encode_inode_release %p cap %p %s (noop)\n", |
a8599bd8 SW |
4242 | inode, cap, ceph_cap_string(cap->issued)); |
4243 | } | |
4244 | } | |
be655596 | 4245 | spin_unlock(&ci->i_ceph_lock); |
a8599bd8 SW |
4246 | return ret; |
4247 | } | |
4248 | ||
4249 | int ceph_encode_dentry_release(void **p, struct dentry *dentry, | |
ca6c8ae0 | 4250 | struct inode *dir, |
a8599bd8 SW |
4251 | int mds, int drop, int unless) |
4252 | { | |
ca6c8ae0 | 4253 | struct dentry *parent = NULL; |
a8599bd8 SW |
4254 | struct ceph_mds_request_release *rel = *p; |
4255 | struct ceph_dentry_info *di = ceph_dentry(dentry); | |
4256 | int force = 0; | |
4257 | int ret; | |
4258 | ||
4259 | /* | |
4260 | * force an record for the directory caps if we have a dentry lease. | |
be655596 | 4261 | * this is racy (can't take i_ceph_lock and d_lock together), but it |
a8599bd8 SW |
4262 | * doesn't have to be perfect; the mds will revoke anything we don't |
4263 | * release. | |
4264 | */ | |
4265 | spin_lock(&dentry->d_lock); | |
4266 | if (di->lease_session && di->lease_session->s_mds == mds) | |
4267 | force = 1; | |
ca6c8ae0 JL |
4268 | if (!dir) { |
4269 | parent = dget(dentry->d_parent); | |
4270 | dir = d_inode(parent); | |
4271 | } | |
a8599bd8 SW |
4272 | spin_unlock(&dentry->d_lock); |
4273 | ||
ca6c8ae0 | 4274 | ret = ceph_encode_inode_release(p, dir, mds, drop, unless, force); |
adf0d687 | 4275 | dput(parent); |
a8599bd8 SW |
4276 | |
4277 | spin_lock(&dentry->d_lock); | |
4278 | if (ret && di->lease_session && di->lease_session->s_mds == mds) { | |
4279 | dout("encode_dentry_release %p mds%d seq %d\n", | |
4280 | dentry, mds, (int)di->lease_seq); | |
4281 | rel->dname_len = cpu_to_le32(dentry->d_name.len); | |
4282 | memcpy(*p, dentry->d_name.name, dentry->d_name.len); | |
4283 | *p += dentry->d_name.len; | |
4284 | rel->dname_seq = cpu_to_le32(di->lease_seq); | |
1dadcce3 | 4285 | __ceph_mdsc_drop_dentry_lease(dentry); |
a8599bd8 SW |
4286 | } |
4287 | spin_unlock(&dentry->d_lock); | |
4288 | return ret; | |
4289 | } |