]>
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> |
a8599bd8 SW |
11 | |
12 | #include "super.h" | |
3d14c5d2 | 13 | #include "mds_client.h" |
99ccbd22 | 14 | #include "cache.h" |
3d14c5d2 YS |
15 | #include <linux/ceph/decode.h> |
16 | #include <linux/ceph/messenger.h> | |
a8599bd8 SW |
17 | |
18 | /* | |
19 | * Capability management | |
20 | * | |
21 | * The Ceph metadata servers control client access to inode metadata | |
22 | * and file data by issuing capabilities, granting clients permission | |
23 | * to read and/or write both inode field and file data to OSDs | |
24 | * (storage nodes). Each capability consists of a set of bits | |
25 | * indicating which operations are allowed. | |
26 | * | |
27 | * If the client holds a *_SHARED cap, the client has a coherent value | |
28 | * that can be safely read from the cached inode. | |
29 | * | |
30 | * In the case of a *_EXCL (exclusive) or FILE_WR capabilities, the | |
31 | * client is allowed to change inode attributes (e.g., file size, | |
32 | * mtime), note its dirty state in the ceph_cap, and asynchronously | |
33 | * flush that metadata change to the MDS. | |
34 | * | |
35 | * In the event of a conflicting operation (perhaps by another | |
36 | * client), the MDS will revoke the conflicting client capabilities. | |
37 | * | |
38 | * In order for a client to cache an inode, it must hold a capability | |
39 | * with at least one MDS server. When inodes are released, release | |
40 | * notifications are batched and periodically sent en masse to the MDS | |
41 | * cluster to release server state. | |
42 | */ | |
43 | ||
0e294387 | 44 | static u64 __get_oldest_flush_tid(struct ceph_mds_client *mdsc); |
7bc00fdd YZ |
45 | static void __kick_flushing_caps(struct ceph_mds_client *mdsc, |
46 | struct ceph_mds_session *session, | |
47 | struct ceph_inode_info *ci, | |
48 | u64 oldest_flush_tid); | |
a8599bd8 SW |
49 | |
50 | /* | |
51 | * Generate readable cap strings for debugging output. | |
52 | */ | |
53 | #define MAX_CAP_STR 20 | |
54 | static char cap_str[MAX_CAP_STR][40]; | |
55 | static DEFINE_SPINLOCK(cap_str_lock); | |
56 | static int last_cap_str; | |
57 | ||
58 | static char *gcap_string(char *s, int c) | |
59 | { | |
60 | if (c & CEPH_CAP_GSHARED) | |
61 | *s++ = 's'; | |
62 | if (c & CEPH_CAP_GEXCL) | |
63 | *s++ = 'x'; | |
64 | if (c & CEPH_CAP_GCACHE) | |
65 | *s++ = 'c'; | |
66 | if (c & CEPH_CAP_GRD) | |
67 | *s++ = 'r'; | |
68 | if (c & CEPH_CAP_GWR) | |
69 | *s++ = 'w'; | |
70 | if (c & CEPH_CAP_GBUFFER) | |
71 | *s++ = 'b'; | |
49a9f4f6 YZ |
72 | if (c & CEPH_CAP_GWREXTEND) |
73 | *s++ = 'a'; | |
a8599bd8 SW |
74 | if (c & CEPH_CAP_GLAZYIO) |
75 | *s++ = 'l'; | |
76 | return s; | |
77 | } | |
78 | ||
79 | const char *ceph_cap_string(int caps) | |
80 | { | |
81 | int i; | |
82 | char *s; | |
83 | int c; | |
84 | ||
85 | spin_lock(&cap_str_lock); | |
86 | i = last_cap_str++; | |
87 | if (last_cap_str == MAX_CAP_STR) | |
88 | last_cap_str = 0; | |
89 | spin_unlock(&cap_str_lock); | |
90 | ||
91 | s = cap_str[i]; | |
92 | ||
93 | if (caps & CEPH_CAP_PIN) | |
94 | *s++ = 'p'; | |
95 | ||
96 | c = (caps >> CEPH_CAP_SAUTH) & 3; | |
97 | if (c) { | |
98 | *s++ = 'A'; | |
99 | s = gcap_string(s, c); | |
100 | } | |
101 | ||
102 | c = (caps >> CEPH_CAP_SLINK) & 3; | |
103 | if (c) { | |
104 | *s++ = 'L'; | |
105 | s = gcap_string(s, c); | |
106 | } | |
107 | ||
108 | c = (caps >> CEPH_CAP_SXATTR) & 3; | |
109 | if (c) { | |
110 | *s++ = 'X'; | |
111 | s = gcap_string(s, c); | |
112 | } | |
113 | ||
114 | c = caps >> CEPH_CAP_SFILE; | |
115 | if (c) { | |
116 | *s++ = 'F'; | |
117 | s = gcap_string(s, c); | |
118 | } | |
119 | ||
120 | if (s == cap_str[i]) | |
121 | *s++ = '-'; | |
122 | *s = 0; | |
123 | return cap_str[i]; | |
124 | } | |
125 | ||
37151668 | 126 | void ceph_caps_init(struct ceph_mds_client *mdsc) |
a8599bd8 | 127 | { |
37151668 YS |
128 | INIT_LIST_HEAD(&mdsc->caps_list); |
129 | spin_lock_init(&mdsc->caps_list_lock); | |
a8599bd8 SW |
130 | } |
131 | ||
37151668 | 132 | void ceph_caps_finalize(struct ceph_mds_client *mdsc) |
a8599bd8 SW |
133 | { |
134 | struct ceph_cap *cap; | |
135 | ||
37151668 YS |
136 | spin_lock(&mdsc->caps_list_lock); |
137 | while (!list_empty(&mdsc->caps_list)) { | |
138 | cap = list_first_entry(&mdsc->caps_list, | |
139 | struct ceph_cap, caps_item); | |
a8599bd8 SW |
140 | list_del(&cap->caps_item); |
141 | kmem_cache_free(ceph_cap_cachep, cap); | |
142 | } | |
37151668 YS |
143 | mdsc->caps_total_count = 0; |
144 | mdsc->caps_avail_count = 0; | |
145 | mdsc->caps_use_count = 0; | |
146 | mdsc->caps_reserve_count = 0; | |
147 | mdsc->caps_min_count = 0; | |
148 | spin_unlock(&mdsc->caps_list_lock); | |
85ccce43 SW |
149 | } |
150 | ||
37151668 | 151 | void ceph_adjust_min_caps(struct ceph_mds_client *mdsc, int delta) |
85ccce43 | 152 | { |
37151668 YS |
153 | spin_lock(&mdsc->caps_list_lock); |
154 | mdsc->caps_min_count += delta; | |
155 | BUG_ON(mdsc->caps_min_count < 0); | |
156 | spin_unlock(&mdsc->caps_list_lock); | |
a8599bd8 SW |
157 | } |
158 | ||
7bf8f736 CX |
159 | static void __ceph_unreserve_caps(struct ceph_mds_client *mdsc, int nr_caps) |
160 | { | |
161 | struct ceph_cap *cap; | |
162 | int i; | |
163 | ||
164 | if (nr_caps) { | |
165 | BUG_ON(mdsc->caps_reserve_count < nr_caps); | |
166 | mdsc->caps_reserve_count -= nr_caps; | |
167 | if (mdsc->caps_avail_count >= | |
168 | mdsc->caps_reserve_count + mdsc->caps_min_count) { | |
169 | mdsc->caps_total_count -= nr_caps; | |
170 | for (i = 0; i < nr_caps; i++) { | |
171 | cap = list_first_entry(&mdsc->caps_list, | |
172 | struct ceph_cap, caps_item); | |
173 | list_del(&cap->caps_item); | |
174 | kmem_cache_free(ceph_cap_cachep, cap); | |
175 | } | |
176 | } else { | |
177 | mdsc->caps_avail_count += nr_caps; | |
178 | } | |
179 | ||
180 | dout("%s: caps %d = %d used + %d resv + %d avail\n", | |
181 | __func__, | |
182 | mdsc->caps_total_count, mdsc->caps_use_count, | |
183 | mdsc->caps_reserve_count, mdsc->caps_avail_count); | |
184 | BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count + | |
185 | mdsc->caps_reserve_count + | |
186 | mdsc->caps_avail_count); | |
187 | } | |
188 | } | |
189 | ||
e30ee581 ZZ |
190 | /* |
191 | * Called under mdsc->mutex. | |
192 | */ | |
193 | int ceph_reserve_caps(struct ceph_mds_client *mdsc, | |
37151668 | 194 | struct ceph_cap_reservation *ctx, int need) |
a8599bd8 | 195 | { |
e30ee581 | 196 | int i, j; |
a8599bd8 SW |
197 | struct ceph_cap *cap; |
198 | int have; | |
199 | int alloc = 0; | |
e30ee581 | 200 | int max_caps; |
e5bc08d0 | 201 | int err = 0; |
e30ee581 ZZ |
202 | bool trimmed = false; |
203 | struct ceph_mds_session *s; | |
a8599bd8 | 204 | LIST_HEAD(newcaps); |
a8599bd8 SW |
205 | |
206 | dout("reserve caps ctx=%p need=%d\n", ctx, need); | |
207 | ||
208 | /* first reserve any caps that are already allocated */ | |
37151668 YS |
209 | spin_lock(&mdsc->caps_list_lock); |
210 | if (mdsc->caps_avail_count >= need) | |
a8599bd8 SW |
211 | have = need; |
212 | else | |
37151668 YS |
213 | have = mdsc->caps_avail_count; |
214 | mdsc->caps_avail_count -= have; | |
215 | mdsc->caps_reserve_count += have; | |
216 | BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count + | |
217 | mdsc->caps_reserve_count + | |
218 | mdsc->caps_avail_count); | |
219 | spin_unlock(&mdsc->caps_list_lock); | |
a8599bd8 | 220 | |
79cd674a | 221 | for (i = have; i < need; ) { |
a8599bd8 | 222 | cap = kmem_cache_alloc(ceph_cap_cachep, GFP_NOFS); |
79cd674a CX |
223 | if (cap) { |
224 | list_add(&cap->caps_item, &newcaps); | |
225 | alloc++; | |
226 | i++; | |
227 | continue; | |
228 | } | |
229 | ||
230 | if (!trimmed) { | |
231 | for (j = 0; j < mdsc->max_sessions; j++) { | |
232 | s = __ceph_lookup_mds_session(mdsc, j); | |
233 | if (!s) | |
234 | continue; | |
235 | mutex_unlock(&mdsc->mutex); | |
236 | ||
237 | mutex_lock(&s->s_mutex); | |
238 | max_caps = s->s_nr_caps - (need - i); | |
239 | ceph_trim_caps(mdsc, s, max_caps); | |
240 | mutex_unlock(&s->s_mutex); | |
241 | ||
242 | ceph_put_mds_session(s); | |
243 | mutex_lock(&mdsc->mutex); | |
e30ee581 | 244 | } |
79cd674a CX |
245 | trimmed = true; |
246 | ||
247 | spin_lock(&mdsc->caps_list_lock); | |
248 | if (mdsc->caps_avail_count) { | |
249 | int more_have; | |
250 | if (mdsc->caps_avail_count >= need - i) | |
251 | more_have = need - i; | |
252 | else | |
253 | more_have = mdsc->caps_avail_count; | |
254 | ||
255 | i += more_have; | |
256 | have += more_have; | |
257 | mdsc->caps_avail_count -= more_have; | |
258 | mdsc->caps_reserve_count += more_have; | |
259 | ||
260 | } | |
261 | spin_unlock(&mdsc->caps_list_lock); | |
262 | ||
263 | continue; | |
e30ee581 | 264 | } |
79cd674a CX |
265 | |
266 | pr_warn("reserve caps ctx=%p ENOMEM need=%d got=%d\n", | |
267 | ctx, need, have + alloc); | |
e5bc08d0 CX |
268 | err = -ENOMEM; |
269 | break; | |
270 | } | |
271 | ||
272 | if (!err) { | |
273 | BUG_ON(have + alloc != need); | |
274 | ctx->count = need; | |
a8599bd8 | 275 | } |
a8599bd8 | 276 | |
37151668 YS |
277 | spin_lock(&mdsc->caps_list_lock); |
278 | mdsc->caps_total_count += alloc; | |
279 | mdsc->caps_reserve_count += alloc; | |
280 | list_splice(&newcaps, &mdsc->caps_list); | |
a8599bd8 | 281 | |
37151668 YS |
282 | BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count + |
283 | mdsc->caps_reserve_count + | |
284 | mdsc->caps_avail_count); | |
e5bc08d0 CX |
285 | |
286 | if (err) | |
287 | __ceph_unreserve_caps(mdsc, have + alloc); | |
288 | ||
37151668 | 289 | spin_unlock(&mdsc->caps_list_lock); |
a8599bd8 | 290 | |
a8599bd8 | 291 | dout("reserve caps ctx=%p %d = %d used + %d resv + %d avail\n", |
37151668 YS |
292 | ctx, mdsc->caps_total_count, mdsc->caps_use_count, |
293 | mdsc->caps_reserve_count, mdsc->caps_avail_count); | |
e5bc08d0 | 294 | return err; |
a8599bd8 SW |
295 | } |
296 | ||
7bf8f736 | 297 | void ceph_unreserve_caps(struct ceph_mds_client *mdsc, |
37151668 | 298 | struct ceph_cap_reservation *ctx) |
a8599bd8 SW |
299 | { |
300 | dout("unreserve caps ctx=%p count=%d\n", ctx, ctx->count); | |
7bf8f736 CX |
301 | spin_lock(&mdsc->caps_list_lock); |
302 | __ceph_unreserve_caps(mdsc, ctx->count); | |
303 | ctx->count = 0; | |
304 | spin_unlock(&mdsc->caps_list_lock); | |
a8599bd8 SW |
305 | } |
306 | ||
d9df2783 YZ |
307 | struct ceph_cap *ceph_get_cap(struct ceph_mds_client *mdsc, |
308 | struct ceph_cap_reservation *ctx) | |
a8599bd8 SW |
309 | { |
310 | struct ceph_cap *cap = NULL; | |
311 | ||
312 | /* temporary, until we do something about cap import/export */ | |
443b3760 SW |
313 | if (!ctx) { |
314 | cap = kmem_cache_alloc(ceph_cap_cachep, GFP_NOFS); | |
315 | if (cap) { | |
4d1d0534 | 316 | spin_lock(&mdsc->caps_list_lock); |
37151668 YS |
317 | mdsc->caps_use_count++; |
318 | mdsc->caps_total_count++; | |
4d1d0534 | 319 | spin_unlock(&mdsc->caps_list_lock); |
e327ce06 CX |
320 | } else { |
321 | spin_lock(&mdsc->caps_list_lock); | |
322 | if (mdsc->caps_avail_count) { | |
323 | BUG_ON(list_empty(&mdsc->caps_list)); | |
324 | ||
325 | mdsc->caps_avail_count--; | |
326 | mdsc->caps_use_count++; | |
327 | cap = list_first_entry(&mdsc->caps_list, | |
328 | struct ceph_cap, caps_item); | |
329 | list_del(&cap->caps_item); | |
330 | ||
331 | BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count + | |
332 | mdsc->caps_reserve_count + mdsc->caps_avail_count); | |
333 | } | |
334 | spin_unlock(&mdsc->caps_list_lock); | |
443b3760 | 335 | } |
e327ce06 | 336 | |
443b3760 SW |
337 | return cap; |
338 | } | |
a8599bd8 | 339 | |
37151668 | 340 | spin_lock(&mdsc->caps_list_lock); |
a8599bd8 | 341 | dout("get_cap ctx=%p (%d) %d = %d used + %d resv + %d avail\n", |
37151668 YS |
342 | ctx, ctx->count, mdsc->caps_total_count, mdsc->caps_use_count, |
343 | mdsc->caps_reserve_count, mdsc->caps_avail_count); | |
a8599bd8 | 344 | BUG_ON(!ctx->count); |
37151668 YS |
345 | BUG_ON(ctx->count > mdsc->caps_reserve_count); |
346 | BUG_ON(list_empty(&mdsc->caps_list)); | |
a8599bd8 SW |
347 | |
348 | ctx->count--; | |
37151668 YS |
349 | mdsc->caps_reserve_count--; |
350 | mdsc->caps_use_count++; | |
a8599bd8 | 351 | |
37151668 | 352 | cap = list_first_entry(&mdsc->caps_list, struct ceph_cap, caps_item); |
a8599bd8 SW |
353 | list_del(&cap->caps_item); |
354 | ||
37151668 YS |
355 | BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count + |
356 | mdsc->caps_reserve_count + mdsc->caps_avail_count); | |
357 | spin_unlock(&mdsc->caps_list_lock); | |
a8599bd8 SW |
358 | return cap; |
359 | } | |
360 | ||
37151668 | 361 | void ceph_put_cap(struct ceph_mds_client *mdsc, struct ceph_cap *cap) |
a8599bd8 | 362 | { |
37151668 | 363 | spin_lock(&mdsc->caps_list_lock); |
7c1332b8 | 364 | dout("put_cap %p %d = %d used + %d resv + %d avail\n", |
37151668 YS |
365 | cap, mdsc->caps_total_count, mdsc->caps_use_count, |
366 | mdsc->caps_reserve_count, mdsc->caps_avail_count); | |
367 | mdsc->caps_use_count--; | |
a8599bd8 | 368 | /* |
85ccce43 SW |
369 | * Keep some preallocated caps around (ceph_min_count), to |
370 | * avoid lots of free/alloc churn. | |
a8599bd8 | 371 | */ |
37151668 YS |
372 | if (mdsc->caps_avail_count >= mdsc->caps_reserve_count + |
373 | mdsc->caps_min_count) { | |
374 | mdsc->caps_total_count--; | |
a8599bd8 SW |
375 | kmem_cache_free(ceph_cap_cachep, cap); |
376 | } else { | |
37151668 YS |
377 | mdsc->caps_avail_count++; |
378 | list_add(&cap->caps_item, &mdsc->caps_list); | |
a8599bd8 SW |
379 | } |
380 | ||
37151668 YS |
381 | BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count + |
382 | mdsc->caps_reserve_count + mdsc->caps_avail_count); | |
383 | spin_unlock(&mdsc->caps_list_lock); | |
a8599bd8 SW |
384 | } |
385 | ||
3d14c5d2 | 386 | void ceph_reservation_status(struct ceph_fs_client *fsc, |
85ccce43 SW |
387 | int *total, int *avail, int *used, int *reserved, |
388 | int *min) | |
a8599bd8 | 389 | { |
3d14c5d2 | 390 | struct ceph_mds_client *mdsc = fsc->mdsc; |
37151668 | 391 | |
b884014a CX |
392 | spin_lock(&mdsc->caps_list_lock); |
393 | ||
a8599bd8 | 394 | if (total) |
37151668 | 395 | *total = mdsc->caps_total_count; |
a8599bd8 | 396 | if (avail) |
37151668 | 397 | *avail = mdsc->caps_avail_count; |
a8599bd8 | 398 | if (used) |
37151668 | 399 | *used = mdsc->caps_use_count; |
a8599bd8 | 400 | if (reserved) |
37151668 | 401 | *reserved = mdsc->caps_reserve_count; |
85ccce43 | 402 | if (min) |
37151668 | 403 | *min = mdsc->caps_min_count; |
b884014a CX |
404 | |
405 | spin_unlock(&mdsc->caps_list_lock); | |
a8599bd8 SW |
406 | } |
407 | ||
408 | /* | |
409 | * Find ceph_cap for given mds, if any. | |
410 | * | |
be655596 | 411 | * Called with i_ceph_lock held. |
a8599bd8 SW |
412 | */ |
413 | static struct ceph_cap *__get_cap_for_mds(struct ceph_inode_info *ci, int mds) | |
414 | { | |
415 | struct ceph_cap *cap; | |
416 | struct rb_node *n = ci->i_caps.rb_node; | |
417 | ||
418 | while (n) { | |
419 | cap = rb_entry(n, struct ceph_cap, ci_node); | |
420 | if (mds < cap->mds) | |
421 | n = n->rb_left; | |
422 | else if (mds > cap->mds) | |
423 | n = n->rb_right; | |
424 | else | |
425 | return cap; | |
426 | } | |
427 | return NULL; | |
428 | } | |
429 | ||
2bc50259 GF |
430 | struct ceph_cap *ceph_get_cap_for_mds(struct ceph_inode_info *ci, int mds) |
431 | { | |
432 | struct ceph_cap *cap; | |
433 | ||
be655596 | 434 | spin_lock(&ci->i_ceph_lock); |
2bc50259 | 435 | cap = __get_cap_for_mds(ci, mds); |
be655596 | 436 | spin_unlock(&ci->i_ceph_lock); |
2bc50259 GF |
437 | return cap; |
438 | } | |
439 | ||
a8599bd8 | 440 | /* |
33caad32 | 441 | * Return id of any MDS with a cap, preferably FILE_WR|BUFFER|EXCL, else -1. |
a8599bd8 | 442 | */ |
ca81f3f6 | 443 | static int __ceph_get_cap_mds(struct ceph_inode_info *ci) |
a8599bd8 SW |
444 | { |
445 | struct ceph_cap *cap; | |
446 | int mds = -1; | |
447 | struct rb_node *p; | |
448 | ||
33caad32 | 449 | /* prefer mds with WR|BUFFER|EXCL caps */ |
a8599bd8 SW |
450 | for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) { |
451 | cap = rb_entry(p, struct ceph_cap, ci_node); | |
452 | mds = cap->mds; | |
a8599bd8 SW |
453 | if (cap->issued & (CEPH_CAP_FILE_WR | |
454 | CEPH_CAP_FILE_BUFFER | | |
455 | CEPH_CAP_FILE_EXCL)) | |
456 | break; | |
457 | } | |
458 | return mds; | |
459 | } | |
460 | ||
461 | int ceph_get_cap_mds(struct inode *inode) | |
462 | { | |
be655596 | 463 | struct ceph_inode_info *ci = ceph_inode(inode); |
a8599bd8 | 464 | int mds; |
be655596 | 465 | spin_lock(&ci->i_ceph_lock); |
ca81f3f6 | 466 | mds = __ceph_get_cap_mds(ceph_inode(inode)); |
be655596 | 467 | spin_unlock(&ci->i_ceph_lock); |
a8599bd8 SW |
468 | return mds; |
469 | } | |
470 | ||
471 | /* | |
be655596 | 472 | * Called under i_ceph_lock. |
a8599bd8 SW |
473 | */ |
474 | static void __insert_cap_node(struct ceph_inode_info *ci, | |
475 | struct ceph_cap *new) | |
476 | { | |
477 | struct rb_node **p = &ci->i_caps.rb_node; | |
478 | struct rb_node *parent = NULL; | |
479 | struct ceph_cap *cap = NULL; | |
480 | ||
481 | while (*p) { | |
482 | parent = *p; | |
483 | cap = rb_entry(parent, struct ceph_cap, ci_node); | |
484 | if (new->mds < cap->mds) | |
485 | p = &(*p)->rb_left; | |
486 | else if (new->mds > cap->mds) | |
487 | p = &(*p)->rb_right; | |
488 | else | |
489 | BUG(); | |
490 | } | |
491 | ||
492 | rb_link_node(&new->ci_node, parent, p); | |
493 | rb_insert_color(&new->ci_node, &ci->i_caps); | |
494 | } | |
495 | ||
496 | /* | |
497 | * (re)set cap hold timeouts, which control the delayed release | |
498 | * of unused caps back to the MDS. Should be called on cap use. | |
499 | */ | |
500 | static void __cap_set_timeouts(struct ceph_mds_client *mdsc, | |
501 | struct ceph_inode_info *ci) | |
502 | { | |
3d14c5d2 | 503 | struct ceph_mount_options *ma = mdsc->fsc->mount_options; |
a8599bd8 SW |
504 | |
505 | ci->i_hold_caps_min = round_jiffies(jiffies + | |
506 | ma->caps_wanted_delay_min * HZ); | |
507 | ci->i_hold_caps_max = round_jiffies(jiffies + | |
508 | ma->caps_wanted_delay_max * HZ); | |
509 | dout("__cap_set_timeouts %p min %lu max %lu\n", &ci->vfs_inode, | |
510 | ci->i_hold_caps_min - jiffies, ci->i_hold_caps_max - jiffies); | |
511 | } | |
512 | ||
513 | /* | |
514 | * (Re)queue cap at the end of the delayed cap release list. | |
515 | * | |
516 | * If I_FLUSH is set, leave the inode at the front of the list. | |
517 | * | |
be655596 | 518 | * Caller holds i_ceph_lock |
a8599bd8 SW |
519 | * -> we take mdsc->cap_delay_lock |
520 | */ | |
521 | static void __cap_delay_requeue(struct ceph_mds_client *mdsc, | |
522 | struct ceph_inode_info *ci) | |
523 | { | |
524 | __cap_set_timeouts(mdsc, ci); | |
525 | dout("__cap_delay_requeue %p flags %d at %lu\n", &ci->vfs_inode, | |
526 | ci->i_ceph_flags, ci->i_hold_caps_max); | |
527 | if (!mdsc->stopping) { | |
528 | spin_lock(&mdsc->cap_delay_lock); | |
529 | if (!list_empty(&ci->i_cap_delay_list)) { | |
530 | if (ci->i_ceph_flags & CEPH_I_FLUSH) | |
531 | goto no_change; | |
532 | list_del_init(&ci->i_cap_delay_list); | |
533 | } | |
534 | list_add_tail(&ci->i_cap_delay_list, &mdsc->cap_delay_list); | |
535 | no_change: | |
536 | spin_unlock(&mdsc->cap_delay_lock); | |
537 | } | |
538 | } | |
539 | ||
540 | /* | |
541 | * Queue an inode for immediate writeback. Mark inode with I_FLUSH, | |
542 | * indicating we should send a cap message to flush dirty metadata | |
543 | * asap, and move to the front of the delayed cap list. | |
544 | */ | |
545 | static void __cap_delay_requeue_front(struct ceph_mds_client *mdsc, | |
546 | struct ceph_inode_info *ci) | |
547 | { | |
548 | dout("__cap_delay_requeue_front %p\n", &ci->vfs_inode); | |
549 | spin_lock(&mdsc->cap_delay_lock); | |
550 | ci->i_ceph_flags |= CEPH_I_FLUSH; | |
551 | if (!list_empty(&ci->i_cap_delay_list)) | |
552 | list_del_init(&ci->i_cap_delay_list); | |
553 | list_add(&ci->i_cap_delay_list, &mdsc->cap_delay_list); | |
554 | spin_unlock(&mdsc->cap_delay_lock); | |
555 | } | |
556 | ||
557 | /* | |
558 | * Cancel delayed work on cap. | |
559 | * | |
be655596 | 560 | * Caller must hold i_ceph_lock. |
a8599bd8 SW |
561 | */ |
562 | static void __cap_delay_cancel(struct ceph_mds_client *mdsc, | |
563 | struct ceph_inode_info *ci) | |
564 | { | |
565 | dout("__cap_delay_cancel %p\n", &ci->vfs_inode); | |
566 | if (list_empty(&ci->i_cap_delay_list)) | |
567 | return; | |
568 | spin_lock(&mdsc->cap_delay_lock); | |
569 | list_del_init(&ci->i_cap_delay_list); | |
570 | spin_unlock(&mdsc->cap_delay_lock); | |
571 | } | |
572 | ||
573 | /* | |
574 | * Common issue checks for add_cap, handle_cap_grant. | |
575 | */ | |
576 | static void __check_cap_issue(struct ceph_inode_info *ci, struct ceph_cap *cap, | |
577 | unsigned issued) | |
578 | { | |
579 | unsigned had = __ceph_caps_issued(ci, NULL); | |
580 | ||
581 | /* | |
582 | * Each time we receive FILE_CACHE anew, we increment | |
583 | * i_rdcache_gen. | |
584 | */ | |
2962507c | 585 | if ((issued & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) && |
99ccbd22 | 586 | (had & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) == 0) { |
a8599bd8 | 587 | ci->i_rdcache_gen++; |
99ccbd22 | 588 | } |
a8599bd8 SW |
589 | |
590 | /* | |
15b51bd6 YZ |
591 | * If FILE_SHARED is newly issued, mark dir not complete. We don't |
592 | * know what happened to this directory while we didn't have the cap. | |
593 | * If FILE_SHARED is being revoked, also mark dir not complete. It | |
594 | * stops on-going cached readdir. | |
a8599bd8 | 595 | */ |
15b51bd6 YZ |
596 | if ((issued & CEPH_CAP_FILE_SHARED) != (had & CEPH_CAP_FILE_SHARED)) { |
597 | if (issued & CEPH_CAP_FILE_SHARED) | |
97aeb6bf | 598 | atomic_inc(&ci->i_shared_gen); |
a8673d61 YZ |
599 | if (S_ISDIR(ci->vfs_inode.i_mode)) { |
600 | dout(" marking %p NOT complete\n", &ci->vfs_inode); | |
2f276c51 | 601 | __ceph_dir_clear_complete(ci); |
a8673d61 | 602 | } |
a8599bd8 SW |
603 | } |
604 | } | |
605 | ||
606 | /* | |
607 | * Add a capability under the given MDS session. | |
608 | * | |
609 | * Caller should hold session snap_rwsem (read) and s_mutex. | |
610 | * | |
611 | * @fmode is the open file mode, if we are opening a file, otherwise | |
612 | * it is < 0. (This is so we can atomically add the cap and add an | |
613 | * open file reference to it.) | |
614 | */ | |
d9df2783 YZ |
615 | void ceph_add_cap(struct inode *inode, |
616 | struct ceph_mds_session *session, u64 cap_id, | |
617 | int fmode, unsigned issued, unsigned wanted, | |
618 | unsigned seq, unsigned mseq, u64 realmino, int flags, | |
619 | struct ceph_cap **new_cap) | |
a8599bd8 | 620 | { |
3d14c5d2 | 621 | struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc; |
a8599bd8 | 622 | struct ceph_inode_info *ci = ceph_inode(inode); |
a8599bd8 SW |
623 | struct ceph_cap *cap; |
624 | int mds = session->s_mds; | |
625 | int actual_wanted; | |
626 | ||
627 | dout("add_cap %p mds%d cap %llx %s seq %d\n", inode, | |
628 | session->s_mds, cap_id, ceph_cap_string(issued), seq); | |
629 | ||
630 | /* | |
631 | * If we are opening the file, include file mode wanted bits | |
632 | * in wanted. | |
633 | */ | |
634 | if (fmode >= 0) | |
635 | wanted |= ceph_caps_for_mode(fmode); | |
636 | ||
a8599bd8 SW |
637 | cap = __get_cap_for_mds(ci, mds); |
638 | if (!cap) { | |
d9df2783 YZ |
639 | cap = *new_cap; |
640 | *new_cap = NULL; | |
a8599bd8 SW |
641 | |
642 | cap->issued = 0; | |
643 | cap->implemented = 0; | |
644 | cap->mds = mds; | |
645 | cap->mds_wanted = 0; | |
964266cc | 646 | cap->mseq = 0; |
a8599bd8 SW |
647 | |
648 | cap->ci = ci; | |
649 | __insert_cap_node(ci, cap); | |
650 | ||
a8599bd8 SW |
651 | /* add to session cap list */ |
652 | cap->session = session; | |
653 | spin_lock(&session->s_cap_lock); | |
654 | list_add_tail(&cap->session_caps, &session->s_caps); | |
655 | session->s_nr_caps++; | |
656 | spin_unlock(&session->s_cap_lock); | |
11df2dfb | 657 | } else { |
11df2dfb YZ |
658 | /* |
659 | * auth mds of the inode changed. we received the cap export | |
660 | * message, but still haven't received the cap import message. | |
661 | * handle_cap_export() updated the new auth MDS' cap. | |
662 | * | |
663 | * "ceph_seq_cmp(seq, cap->seq) <= 0" means we are processing | |
664 | * a message that was send before the cap import message. So | |
665 | * don't remove caps. | |
666 | */ | |
667 | if (ceph_seq_cmp(seq, cap->seq) <= 0) { | |
668 | WARN_ON(cap != ci->i_auth_cap); | |
669 | WARN_ON(cap->cap_id != cap_id); | |
670 | seq = cap->seq; | |
671 | mseq = cap->mseq; | |
672 | issued |= cap->issued; | |
673 | flags |= CEPH_CAP_FLAG_AUTH; | |
674 | } | |
675 | } | |
a8599bd8 | 676 | |
7d9c9193 YZ |
677 | if (!ci->i_snap_realm || |
678 | ((flags & CEPH_CAP_FLAG_AUTH) && | |
679 | realmino != (u64)-1 && ci->i_snap_realm->ino != realmino)) { | |
a8599bd8 SW |
680 | /* |
681 | * add this inode to the appropriate snap realm | |
682 | */ | |
683 | struct ceph_snap_realm *realm = ceph_lookup_snap_realm(mdsc, | |
684 | realmino); | |
685 | if (realm) { | |
7d9c9193 YZ |
686 | struct ceph_snap_realm *oldrealm = ci->i_snap_realm; |
687 | if (oldrealm) { | |
688 | spin_lock(&oldrealm->inodes_with_caps_lock); | |
689 | list_del_init(&ci->i_snap_realm_item); | |
690 | spin_unlock(&oldrealm->inodes_with_caps_lock); | |
691 | } | |
692 | ||
a8599bd8 | 693 | spin_lock(&realm->inodes_with_caps_lock); |
a8599bd8 SW |
694 | list_add(&ci->i_snap_realm_item, |
695 | &realm->inodes_with_caps); | |
e3161f17 LH |
696 | ci->i_snap_realm = realm; |
697 | if (realm->ino == ci->i_vino.ino) | |
698 | realm->inode = inode; | |
a8599bd8 | 699 | spin_unlock(&realm->inodes_with_caps_lock); |
7d9c9193 YZ |
700 | |
701 | if (oldrealm) | |
702 | ceph_put_snap_realm(mdsc, oldrealm); | |
a8599bd8 SW |
703 | } else { |
704 | pr_err("ceph_add_cap: couldn't find snap realm %llx\n", | |
705 | realmino); | |
b8cd07e7 | 706 | WARN_ON(!realm); |
a8599bd8 SW |
707 | } |
708 | } | |
709 | ||
710 | __check_cap_issue(ci, cap, issued); | |
711 | ||
712 | /* | |
713 | * If we are issued caps we don't want, or the mds' wanted | |
714 | * value appears to be off, queue a check so we'll release | |
715 | * later and/or update the mds wanted value. | |
716 | */ | |
717 | actual_wanted = __ceph_caps_wanted(ci); | |
718 | if ((wanted & ~actual_wanted) || | |
719 | (issued & ~actual_wanted & CEPH_CAP_ANY_WR)) { | |
720 | dout(" issued %s, mds wanted %s, actual %s, queueing\n", | |
721 | ceph_cap_string(issued), ceph_cap_string(wanted), | |
722 | ceph_cap_string(actual_wanted)); | |
723 | __cap_delay_requeue(mdsc, ci); | |
724 | } | |
725 | ||
b8c2f3ae | 726 | if (flags & CEPH_CAP_FLAG_AUTH) { |
d37b1d99 | 727 | if (!ci->i_auth_cap || |
d9ffc4f7 | 728 | ceph_seq_cmp(ci->i_auth_cap->mseq, mseq) < 0) { |
b8c2f3ae | 729 | ci->i_auth_cap = cap; |
d9ffc4f7 YZ |
730 | cap->mds_wanted = wanted; |
731 | } | |
11df2dfb YZ |
732 | } else { |
733 | WARN_ON(ci->i_auth_cap == cap); | |
8a92a119 | 734 | } |
a8599bd8 SW |
735 | |
736 | dout("add_cap inode %p (%llx.%llx) cap %p %s now %s seq %d mds%d\n", | |
737 | inode, ceph_vinop(inode), cap, ceph_cap_string(issued), | |
738 | ceph_cap_string(issued|cap->issued), seq, mds); | |
739 | cap->cap_id = cap_id; | |
740 | cap->issued = issued; | |
741 | cap->implemented |= issued; | |
d1b87809 | 742 | if (ceph_seq_cmp(mseq, cap->mseq) > 0) |
964266cc YZ |
743 | cap->mds_wanted = wanted; |
744 | else | |
745 | cap->mds_wanted |= wanted; | |
a8599bd8 SW |
746 | cap->seq = seq; |
747 | cap->issue_seq = seq; | |
748 | cap->mseq = mseq; | |
685f9a5d | 749 | cap->cap_gen = session->s_cap_gen; |
a8599bd8 SW |
750 | |
751 | if (fmode >= 0) | |
752 | __ceph_get_fmode(ci, fmode); | |
a8599bd8 SW |
753 | } |
754 | ||
755 | /* | |
756 | * Return true if cap has not timed out and belongs to the current | |
757 | * generation of the MDS session (i.e. has not gone 'stale' due to | |
758 | * us losing touch with the mds). | |
759 | */ | |
760 | static int __cap_is_valid(struct ceph_cap *cap) | |
761 | { | |
762 | unsigned long ttl; | |
cdac8303 | 763 | u32 gen; |
a8599bd8 | 764 | |
d8fb02ab | 765 | spin_lock(&cap->session->s_gen_ttl_lock); |
a8599bd8 SW |
766 | gen = cap->session->s_cap_gen; |
767 | ttl = cap->session->s_cap_ttl; | |
d8fb02ab | 768 | spin_unlock(&cap->session->s_gen_ttl_lock); |
a8599bd8 | 769 | |
685f9a5d | 770 | if (cap->cap_gen < gen || time_after_eq(jiffies, ttl)) { |
a8599bd8 SW |
771 | dout("__cap_is_valid %p cap %p issued %s " |
772 | "but STALE (gen %u vs %u)\n", &cap->ci->vfs_inode, | |
685f9a5d | 773 | cap, ceph_cap_string(cap->issued), cap->cap_gen, gen); |
a8599bd8 SW |
774 | return 0; |
775 | } | |
776 | ||
777 | return 1; | |
778 | } | |
779 | ||
780 | /* | |
781 | * Return set of valid cap bits issued to us. Note that caps time | |
782 | * out, and may be invalidated in bulk if the client session times out | |
783 | * and session->s_cap_gen is bumped. | |
784 | */ | |
785 | int __ceph_caps_issued(struct ceph_inode_info *ci, int *implemented) | |
786 | { | |
d9df2783 | 787 | int have = ci->i_snap_caps; |
a8599bd8 SW |
788 | struct ceph_cap *cap; |
789 | struct rb_node *p; | |
790 | ||
791 | if (implemented) | |
792 | *implemented = 0; | |
793 | for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) { | |
794 | cap = rb_entry(p, struct ceph_cap, ci_node); | |
795 | if (!__cap_is_valid(cap)) | |
796 | continue; | |
797 | dout("__ceph_caps_issued %p cap %p issued %s\n", | |
798 | &ci->vfs_inode, cap, ceph_cap_string(cap->issued)); | |
799 | have |= cap->issued; | |
800 | if (implemented) | |
801 | *implemented |= cap->implemented; | |
802 | } | |
b1530f57 YZ |
803 | /* |
804 | * exclude caps issued by non-auth MDS, but are been revoking | |
805 | * by the auth MDS. The non-auth MDS should be revoking/exporting | |
806 | * these caps, but the message is delayed. | |
807 | */ | |
808 | if (ci->i_auth_cap) { | |
809 | cap = ci->i_auth_cap; | |
810 | have &= ~cap->implemented | cap->issued; | |
811 | } | |
a8599bd8 SW |
812 | return have; |
813 | } | |
814 | ||
815 | /* | |
816 | * Get cap bits issued by caps other than @ocap | |
817 | */ | |
818 | int __ceph_caps_issued_other(struct ceph_inode_info *ci, struct ceph_cap *ocap) | |
819 | { | |
820 | int have = ci->i_snap_caps; | |
821 | struct ceph_cap *cap; | |
822 | struct rb_node *p; | |
823 | ||
824 | for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) { | |
825 | cap = rb_entry(p, struct ceph_cap, ci_node); | |
826 | if (cap == ocap) | |
827 | continue; | |
828 | if (!__cap_is_valid(cap)) | |
829 | continue; | |
830 | have |= cap->issued; | |
831 | } | |
832 | return have; | |
833 | } | |
834 | ||
835 | /* | |
836 | * Move a cap to the end of the LRU (oldest caps at list head, newest | |
837 | * at list tail). | |
838 | */ | |
839 | static void __touch_cap(struct ceph_cap *cap) | |
840 | { | |
841 | struct ceph_mds_session *s = cap->session; | |
842 | ||
a8599bd8 | 843 | spin_lock(&s->s_cap_lock); |
d37b1d99 | 844 | if (!s->s_cap_iterator) { |
5dacf091 SW |
845 | dout("__touch_cap %p cap %p mds%d\n", &cap->ci->vfs_inode, cap, |
846 | s->s_mds); | |
847 | list_move_tail(&cap->session_caps, &s->s_caps); | |
848 | } else { | |
849 | dout("__touch_cap %p cap %p mds%d NOP, iterating over caps\n", | |
850 | &cap->ci->vfs_inode, cap, s->s_mds); | |
851 | } | |
a8599bd8 SW |
852 | spin_unlock(&s->s_cap_lock); |
853 | } | |
854 | ||
855 | /* | |
856 | * Check if we hold the given mask. If so, move the cap(s) to the | |
857 | * front of their respective LRUs. (This is the preferred way for | |
858 | * callers to check for caps they want.) | |
859 | */ | |
860 | int __ceph_caps_issued_mask(struct ceph_inode_info *ci, int mask, int touch) | |
861 | { | |
862 | struct ceph_cap *cap; | |
863 | struct rb_node *p; | |
864 | int have = ci->i_snap_caps; | |
865 | ||
866 | if ((have & mask) == mask) { | |
867 | dout("__ceph_caps_issued_mask %p snap issued %s" | |
868 | " (mask %s)\n", &ci->vfs_inode, | |
869 | ceph_cap_string(have), | |
870 | ceph_cap_string(mask)); | |
871 | return 1; | |
872 | } | |
873 | ||
874 | for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) { | |
875 | cap = rb_entry(p, struct ceph_cap, ci_node); | |
876 | if (!__cap_is_valid(cap)) | |
877 | continue; | |
878 | if ((cap->issued & mask) == mask) { | |
879 | dout("__ceph_caps_issued_mask %p cap %p issued %s" | |
880 | " (mask %s)\n", &ci->vfs_inode, cap, | |
881 | ceph_cap_string(cap->issued), | |
882 | ceph_cap_string(mask)); | |
883 | if (touch) | |
884 | __touch_cap(cap); | |
885 | return 1; | |
886 | } | |
887 | ||
888 | /* does a combination of caps satisfy mask? */ | |
889 | have |= cap->issued; | |
890 | if ((have & mask) == mask) { | |
891 | dout("__ceph_caps_issued_mask %p combo issued %s" | |
892 | " (mask %s)\n", &ci->vfs_inode, | |
893 | ceph_cap_string(cap->issued), | |
894 | ceph_cap_string(mask)); | |
895 | if (touch) { | |
896 | struct rb_node *q; | |
897 | ||
25985edc | 898 | /* touch this + preceding caps */ |
a8599bd8 SW |
899 | __touch_cap(cap); |
900 | for (q = rb_first(&ci->i_caps); q != p; | |
901 | q = rb_next(q)) { | |
902 | cap = rb_entry(q, struct ceph_cap, | |
903 | ci_node); | |
904 | if (!__cap_is_valid(cap)) | |
905 | continue; | |
906 | __touch_cap(cap); | |
907 | } | |
908 | } | |
909 | return 1; | |
910 | } | |
911 | } | |
912 | ||
913 | return 0; | |
914 | } | |
915 | ||
916 | /* | |
917 | * Return true if mask caps are currently being revoked by an MDS. | |
918 | */ | |
6ee6b953 YZ |
919 | int __ceph_caps_revoking_other(struct ceph_inode_info *ci, |
920 | struct ceph_cap *ocap, int mask) | |
a8599bd8 | 921 | { |
a8599bd8 SW |
922 | struct ceph_cap *cap; |
923 | struct rb_node *p; | |
a8599bd8 | 924 | |
a8599bd8 SW |
925 | for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) { |
926 | cap = rb_entry(p, struct ceph_cap, ci_node); | |
9563f88c | 927 | if (cap != ocap && |
6ee6b953 YZ |
928 | (cap->implemented & ~cap->issued & mask)) |
929 | return 1; | |
a8599bd8 | 930 | } |
6ee6b953 YZ |
931 | return 0; |
932 | } | |
933 | ||
934 | int ceph_caps_revoking(struct ceph_inode_info *ci, int mask) | |
935 | { | |
936 | struct inode *inode = &ci->vfs_inode; | |
937 | int ret; | |
938 | ||
939 | spin_lock(&ci->i_ceph_lock); | |
940 | ret = __ceph_caps_revoking_other(ci, NULL, mask); | |
be655596 | 941 | spin_unlock(&ci->i_ceph_lock); |
a8599bd8 SW |
942 | dout("ceph_caps_revoking %p %s = %d\n", inode, |
943 | ceph_cap_string(mask), ret); | |
944 | return ret; | |
945 | } | |
946 | ||
947 | int __ceph_caps_used(struct ceph_inode_info *ci) | |
948 | { | |
949 | int used = 0; | |
950 | if (ci->i_pin_ref) | |
951 | used |= CEPH_CAP_PIN; | |
952 | if (ci->i_rd_ref) | |
953 | used |= CEPH_CAP_FILE_RD; | |
fdd4e158 YZ |
954 | if (ci->i_rdcache_ref || |
955 | (!S_ISDIR(ci->vfs_inode.i_mode) && /* ignore readdir cache */ | |
956 | ci->vfs_inode.i_data.nrpages)) | |
a8599bd8 SW |
957 | used |= CEPH_CAP_FILE_CACHE; |
958 | if (ci->i_wr_ref) | |
959 | used |= CEPH_CAP_FILE_WR; | |
d3d0720d | 960 | if (ci->i_wb_ref || ci->i_wrbuffer_ref) |
a8599bd8 SW |
961 | used |= CEPH_CAP_FILE_BUFFER; |
962 | return used; | |
963 | } | |
964 | ||
965 | /* | |
966 | * wanted, by virtue of open file modes | |
967 | */ | |
968 | int __ceph_caps_file_wanted(struct ceph_inode_info *ci) | |
969 | { | |
774a6a11 YZ |
970 | int i, bits = 0; |
971 | for (i = 0; i < CEPH_FILE_MODE_BITS; i++) { | |
972 | if (ci->i_nr_by_mode[i]) | |
973 | bits |= 1 << i; | |
974 | } | |
975 | if (bits == 0) | |
976 | return 0; | |
977 | return ceph_caps_for_mode(bits >> 1); | |
a8599bd8 SW |
978 | } |
979 | ||
980 | /* | |
981 | * Return caps we have registered with the MDS(s) as 'wanted'. | |
982 | */ | |
c1944fed | 983 | int __ceph_caps_mds_wanted(struct ceph_inode_info *ci, bool check) |
a8599bd8 SW |
984 | { |
985 | struct ceph_cap *cap; | |
986 | struct rb_node *p; | |
987 | int mds_wanted = 0; | |
988 | ||
989 | for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) { | |
990 | cap = rb_entry(p, struct ceph_cap, ci_node); | |
c1944fed | 991 | if (check && !__cap_is_valid(cap)) |
a8599bd8 | 992 | continue; |
a2550604 YZ |
993 | if (cap == ci->i_auth_cap) |
994 | mds_wanted |= cap->mds_wanted; | |
995 | else | |
996 | mds_wanted |= (cap->mds_wanted & ~CEPH_CAP_ANY_FILE_WR); | |
a8599bd8 SW |
997 | } |
998 | return mds_wanted; | |
999 | } | |
1000 | ||
1001 | /* | |
be655596 | 1002 | * called under i_ceph_lock |
a8599bd8 | 1003 | */ |
0f439c74 YZ |
1004 | static int __ceph_is_single_caps(struct ceph_inode_info *ci) |
1005 | { | |
1006 | return rb_first(&ci->i_caps) == rb_last(&ci->i_caps); | |
1007 | } | |
1008 | ||
a8599bd8 SW |
1009 | static int __ceph_is_any_caps(struct ceph_inode_info *ci) |
1010 | { | |
d9df2783 | 1011 | return !RB_EMPTY_ROOT(&ci->i_caps); |
a8599bd8 SW |
1012 | } |
1013 | ||
9215aeea YZ |
1014 | int ceph_is_any_caps(struct inode *inode) |
1015 | { | |
1016 | struct ceph_inode_info *ci = ceph_inode(inode); | |
1017 | int ret; | |
1018 | ||
1019 | spin_lock(&ci->i_ceph_lock); | |
1020 | ret = __ceph_is_any_caps(ci); | |
1021 | spin_unlock(&ci->i_ceph_lock); | |
1022 | ||
1023 | return ret; | |
1024 | } | |
1025 | ||
db40cc17 YZ |
1026 | static void drop_inode_snap_realm(struct ceph_inode_info *ci) |
1027 | { | |
1028 | struct ceph_snap_realm *realm = ci->i_snap_realm; | |
1029 | spin_lock(&realm->inodes_with_caps_lock); | |
1030 | list_del_init(&ci->i_snap_realm_item); | |
1031 | ci->i_snap_realm_counter++; | |
1032 | ci->i_snap_realm = NULL; | |
1033 | spin_unlock(&realm->inodes_with_caps_lock); | |
1034 | ceph_put_snap_realm(ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc, | |
1035 | realm); | |
1036 | } | |
1037 | ||
a8599bd8 | 1038 | /* |
f818a736 SW |
1039 | * Remove a cap. Take steps to deal with a racing iterate_session_caps. |
1040 | * | |
be655596 | 1041 | * caller should hold i_ceph_lock. |
a6369741 | 1042 | * caller will not hold session s_mutex if called from destroy_inode. |
a8599bd8 | 1043 | */ |
a096b09a | 1044 | void __ceph_remove_cap(struct ceph_cap *cap, bool queue_release) |
a8599bd8 SW |
1045 | { |
1046 | struct ceph_mds_session *session = cap->session; | |
1047 | struct ceph_inode_info *ci = cap->ci; | |
640ef79d | 1048 | struct ceph_mds_client *mdsc = |
3d14c5d2 | 1049 | ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc; |
f818a736 | 1050 | int removed = 0; |
a8599bd8 SW |
1051 | |
1052 | dout("__ceph_remove_cap %p from %p\n", cap, &ci->vfs_inode); | |
1053 | ||
7c1332b8 SW |
1054 | /* remove from session list */ |
1055 | spin_lock(&session->s_cap_lock); | |
1056 | if (session->s_cap_iterator == cap) { | |
1057 | /* not yet, we are iterating over this very cap */ | |
1058 | dout("__ceph_remove_cap delaying %p removal from session %p\n", | |
1059 | cap, cap->session); | |
1060 | } else { | |
1061 | list_del_init(&cap->session_caps); | |
1062 | session->s_nr_caps--; | |
1063 | cap->session = NULL; | |
f818a736 | 1064 | removed = 1; |
7c1332b8 | 1065 | } |
f818a736 SW |
1066 | /* protect backpointer with s_cap_lock: see iterate_session_caps */ |
1067 | cap->ci = NULL; | |
745a8e3b YZ |
1068 | |
1069 | /* | |
1070 | * s_cap_reconnect is protected by s_cap_lock. no one changes | |
1071 | * s_cap_gen while session is in the reconnect state. | |
1072 | */ | |
1073 | if (queue_release && | |
1074 | (!session->s_cap_reconnect || cap->cap_gen == session->s_cap_gen)) { | |
1075 | cap->queue_release = 1; | |
1076 | if (removed) { | |
1077 | list_add_tail(&cap->session_caps, | |
1078 | &session->s_cap_releases); | |
1079 | session->s_num_cap_releases++; | |
1080 | removed = 0; | |
1081 | } | |
1082 | } else { | |
1083 | cap->queue_release = 0; | |
1084 | } | |
1085 | cap->cap_ino = ci->i_vino.ino; | |
1086 | ||
7c1332b8 SW |
1087 | spin_unlock(&session->s_cap_lock); |
1088 | ||
f818a736 SW |
1089 | /* remove from inode list */ |
1090 | rb_erase(&cap->ci_node, &ci->i_caps); | |
1091 | if (ci->i_auth_cap == cap) | |
1092 | ci->i_auth_cap = NULL; | |
1093 | ||
1094 | if (removed) | |
37151668 | 1095 | ceph_put_cap(mdsc, cap); |
a8599bd8 | 1096 | |
db40cc17 YZ |
1097 | /* when reconnect denied, we remove session caps forcibly, |
1098 | * i_wr_ref can be non-zero. If there are ongoing write, | |
1099 | * keep i_snap_realm. | |
1100 | */ | |
1101 | if (!__ceph_is_any_caps(ci) && ci->i_wr_ref == 0 && ci->i_snap_realm) | |
1102 | drop_inode_snap_realm(ci); | |
1103 | ||
a8599bd8 SW |
1104 | if (!__ceph_is_any_real_caps(ci)) |
1105 | __cap_delay_cancel(mdsc, ci); | |
1106 | } | |
1107 | ||
0ff8bfb3 JL |
1108 | struct cap_msg_args { |
1109 | struct ceph_mds_session *session; | |
1110 | u64 ino, cid, follows; | |
1111 | u64 flush_tid, oldest_flush_tid, size, max_size; | |
1112 | u64 xattr_version; | |
1113 | struct ceph_buffer *xattr_buf; | |
9bbeab41 | 1114 | struct timespec64 atime, mtime, ctime; |
0ff8bfb3 JL |
1115 | int op, caps, wanted, dirty; |
1116 | u32 seq, issue_seq, mseq, time_warp_seq; | |
1e4ef0c6 | 1117 | u32 flags; |
0ff8bfb3 JL |
1118 | kuid_t uid; |
1119 | kgid_t gid; | |
1120 | umode_t mode; | |
1121 | bool inline_data; | |
1122 | }; | |
1123 | ||
a8599bd8 SW |
1124 | /* |
1125 | * Build and send a cap message to the given MDS. | |
1126 | * | |
1127 | * Caller should be holding s_mutex. | |
1128 | */ | |
0ff8bfb3 | 1129 | static int send_cap_msg(struct cap_msg_args *arg) |
a8599bd8 SW |
1130 | { |
1131 | struct ceph_mds_caps *fc; | |
1132 | struct ceph_msg *msg; | |
e20d258d YZ |
1133 | void *p; |
1134 | size_t extra_len; | |
9bbeab41 | 1135 | struct timespec64 zerotime = {0}; |
92475f05 | 1136 | struct ceph_osd_client *osdc = &arg->session->s_mdsc->fsc->client->osdc; |
a8599bd8 SW |
1137 | |
1138 | dout("send_cap_msg %s %llx %llx caps %s wanted %s dirty %s" | |
a2971c8c | 1139 | " seq %u/%u tid %llu/%llu mseq %u follows %lld size %llu/%llu" |
0ff8bfb3 JL |
1140 | " xattr_ver %llu xattr_len %d\n", ceph_cap_op_name(arg->op), |
1141 | arg->cid, arg->ino, ceph_cap_string(arg->caps), | |
1142 | ceph_cap_string(arg->wanted), ceph_cap_string(arg->dirty), | |
1143 | arg->seq, arg->issue_seq, arg->flush_tid, arg->oldest_flush_tid, | |
1144 | arg->mseq, arg->follows, arg->size, arg->max_size, | |
1145 | arg->xattr_version, | |
1146 | arg->xattr_buf ? (int)arg->xattr_buf->vec.iov_len : 0); | |
a8599bd8 | 1147 | |
a2971c8c YZ |
1148 | /* flock buffer size + inline version + inline data size + |
1149 | * osd_epoch_barrier + oldest_flush_tid */ | |
43b29673 | 1150 | extra_len = 4 + 8 + 4 + 4 + 8 + 4 + 4 + 4 + 8 + 8 + 4; |
e20d258d YZ |
1151 | msg = ceph_msg_new(CEPH_MSG_CLIENT_CAPS, sizeof(*fc) + extra_len, |
1152 | GFP_NOFS, false); | |
a79832f2 SW |
1153 | if (!msg) |
1154 | return -ENOMEM; | |
a8599bd8 | 1155 | |
43b29673 | 1156 | msg->hdr.version = cpu_to_le16(10); |
0ff8bfb3 | 1157 | msg->hdr.tid = cpu_to_le64(arg->flush_tid); |
a8599bd8 | 1158 | |
6df058c0 | 1159 | fc = msg->front.iov_base; |
a8599bd8 SW |
1160 | memset(fc, 0, sizeof(*fc)); |
1161 | ||
0ff8bfb3 JL |
1162 | fc->cap_id = cpu_to_le64(arg->cid); |
1163 | fc->op = cpu_to_le32(arg->op); | |
1164 | fc->seq = cpu_to_le32(arg->seq); | |
1165 | fc->issue_seq = cpu_to_le32(arg->issue_seq); | |
1166 | fc->migrate_seq = cpu_to_le32(arg->mseq); | |
1167 | fc->caps = cpu_to_le32(arg->caps); | |
1168 | fc->wanted = cpu_to_le32(arg->wanted); | |
1169 | fc->dirty = cpu_to_le32(arg->dirty); | |
1170 | fc->ino = cpu_to_le64(arg->ino); | |
1171 | fc->snap_follows = cpu_to_le64(arg->follows); | |
1172 | ||
1173 | fc->size = cpu_to_le64(arg->size); | |
1174 | fc->max_size = cpu_to_le64(arg->max_size); | |
9bbeab41 AB |
1175 | ceph_encode_timespec64(&fc->mtime, &arg->mtime); |
1176 | ceph_encode_timespec64(&fc->atime, &arg->atime); | |
1177 | ceph_encode_timespec64(&fc->ctime, &arg->ctime); | |
0ff8bfb3 JL |
1178 | fc->time_warp_seq = cpu_to_le32(arg->time_warp_seq); |
1179 | ||
1180 | fc->uid = cpu_to_le32(from_kuid(&init_user_ns, arg->uid)); | |
1181 | fc->gid = cpu_to_le32(from_kgid(&init_user_ns, arg->gid)); | |
1182 | fc->mode = cpu_to_le32(arg->mode); | |
1183 | ||
1184 | fc->xattr_version = cpu_to_le64(arg->xattr_version); | |
1185 | if (arg->xattr_buf) { | |
1186 | msg->middle = ceph_buffer_get(arg->xattr_buf); | |
1187 | fc->xattr_len = cpu_to_le32(arg->xattr_buf->vec.iov_len); | |
1188 | msg->hdr.middle_len = cpu_to_le32(arg->xattr_buf->vec.iov_len); | |
9670079f JL |
1189 | } |
1190 | ||
e20d258d | 1191 | p = fc + 1; |
43b29673 | 1192 | /* flock buffer size (version 2) */ |
e20d258d | 1193 | ceph_encode_32(&p, 0); |
43b29673 | 1194 | /* inline version (version 4) */ |
0ff8bfb3 | 1195 | ceph_encode_64(&p, arg->inline_data ? 0 : CEPH_INLINE_NONE); |
e20d258d YZ |
1196 | /* inline data size */ |
1197 | ceph_encode_32(&p, 0); | |
92475f05 JL |
1198 | /* |
1199 | * osd_epoch_barrier (version 5) | |
1200 | * The epoch_barrier is protected osdc->lock, so READ_ONCE here in | |
1201 | * case it was recently changed | |
1202 | */ | |
1203 | ceph_encode_32(&p, READ_ONCE(osdc->epoch_barrier)); | |
43b29673 | 1204 | /* oldest_flush_tid (version 6) */ |
0ff8bfb3 | 1205 | ceph_encode_64(&p, arg->oldest_flush_tid); |
e20d258d | 1206 | |
43b29673 JL |
1207 | /* |
1208 | * caller_uid/caller_gid (version 7) | |
1209 | * | |
1210 | * Currently, we don't properly track which caller dirtied the caps | |
1211 | * last, and force a flush of them when there is a conflict. For now, | |
1212 | * just set this to 0:0, to emulate how the MDS has worked up to now. | |
1213 | */ | |
1214 | ceph_encode_32(&p, 0); | |
1215 | ceph_encode_32(&p, 0); | |
1216 | ||
1217 | /* pool namespace (version 8) (mds always ignores this) */ | |
1218 | ceph_encode_32(&p, 0); | |
1219 | ||
1220 | /* | |
1221 | * btime and change_attr (version 9) | |
1222 | * | |
1223 | * We just zero these out for now, as the MDS ignores them unless | |
1224 | * the requisite feature flags are set (which we don't do yet). | |
1225 | */ | |
9bbeab41 | 1226 | ceph_encode_timespec64(p, &zerotime); |
43b29673 JL |
1227 | p += sizeof(struct ceph_timespec); |
1228 | ceph_encode_64(&p, 0); | |
1229 | ||
1230 | /* Advisory flags (version 10) */ | |
1e4ef0c6 | 1231 | ceph_encode_32(&p, arg->flags); |
43b29673 | 1232 | |
0ff8bfb3 | 1233 | ceph_con_send(&arg->session->s_con, msg); |
a8599bd8 SW |
1234 | return 0; |
1235 | } | |
1236 | ||
1237 | /* | |
a6369741 | 1238 | * Queue cap releases when an inode is dropped from our cache. Since |
be655596 | 1239 | * inode is about to be destroyed, there is no need for i_ceph_lock. |
a8599bd8 SW |
1240 | */ |
1241 | void ceph_queue_caps_release(struct inode *inode) | |
1242 | { | |
1243 | struct ceph_inode_info *ci = ceph_inode(inode); | |
1244 | struct rb_node *p; | |
1245 | ||
a8599bd8 SW |
1246 | p = rb_first(&ci->i_caps); |
1247 | while (p) { | |
1248 | struct ceph_cap *cap = rb_entry(p, struct ceph_cap, ci_node); | |
a8599bd8 | 1249 | p = rb_next(p); |
a096b09a | 1250 | __ceph_remove_cap(cap, true); |
a8599bd8 | 1251 | } |
a8599bd8 SW |
1252 | } |
1253 | ||
1254 | /* | |
1255 | * Send a cap msg on the given inode. Update our caps state, then | |
be655596 | 1256 | * drop i_ceph_lock and send the message. |
a8599bd8 SW |
1257 | * |
1258 | * Make note of max_size reported/requested from mds, revoked caps | |
1259 | * that have now been implemented. | |
1260 | * | |
1261 | * Make half-hearted attempt ot to invalidate page cache if we are | |
1262 | * dropping RDCACHE. Note that this will leave behind locked pages | |
1263 | * that we'll then need to deal with elsewhere. | |
1264 | * | |
1265 | * Return non-zero if delayed release, or we experienced an error | |
1266 | * such that the caller should requeue + retry later. | |
1267 | * | |
be655596 | 1268 | * called with i_ceph_lock, then drops it. |
a8599bd8 SW |
1269 | * caller should hold snap_rwsem (read), s_mutex. |
1270 | */ | |
1271 | static int __send_cap(struct ceph_mds_client *mdsc, struct ceph_cap *cap, | |
1e4ef0c6 JL |
1272 | int op, bool sync, int used, int want, int retain, |
1273 | int flushing, u64 flush_tid, u64 oldest_flush_tid) | |
be655596 | 1274 | __releases(cap->ci->i_ceph_lock) |
a8599bd8 SW |
1275 | { |
1276 | struct ceph_inode_info *ci = cap->ci; | |
1277 | struct inode *inode = &ci->vfs_inode; | |
0ff8bfb3 | 1278 | struct cap_msg_args arg; |
bb0581f0 | 1279 | int held, revoking; |
a8599bd8 | 1280 | int wake = 0; |
a8599bd8 | 1281 | int delayed = 0; |
a8599bd8 SW |
1282 | int ret; |
1283 | ||
68c28323 SW |
1284 | held = cap->issued | cap->implemented; |
1285 | revoking = cap->implemented & ~cap->issued; | |
1286 | retain &= ~revoking; | |
68c28323 | 1287 | |
a8599bd8 SW |
1288 | dout("__send_cap %p cap %p session %p %s -> %s (revoking %s)\n", |
1289 | inode, cap, cap->session, | |
1290 | ceph_cap_string(held), ceph_cap_string(held & retain), | |
1291 | ceph_cap_string(revoking)); | |
1292 | BUG_ON((retain & CEPH_CAP_PIN) == 0); | |
1293 | ||
0ff8bfb3 | 1294 | arg.session = cap->session; |
a8599bd8 SW |
1295 | |
1296 | /* don't release wanted unless we've waited a bit. */ | |
1297 | if ((ci->i_ceph_flags & CEPH_I_NODELAY) == 0 && | |
1298 | time_before(jiffies, ci->i_hold_caps_min)) { | |
1299 | dout(" delaying issued %s -> %s, wanted %s -> %s on send\n", | |
1300 | ceph_cap_string(cap->issued), | |
1301 | ceph_cap_string(cap->issued & retain), | |
1302 | ceph_cap_string(cap->mds_wanted), | |
1303 | ceph_cap_string(want)); | |
1304 | want |= cap->mds_wanted; | |
1305 | retain |= cap->issued; | |
1306 | delayed = 1; | |
1307 | } | |
1308 | ci->i_ceph_flags &= ~(CEPH_I_NODELAY | CEPH_I_FLUSH); | |
eb65b919 YZ |
1309 | if (want & ~cap->mds_wanted) { |
1310 | /* user space may open/close single file frequently. | |
1311 | * This avoids droping mds_wanted immediately after | |
1312 | * requesting new mds_wanted. | |
1313 | */ | |
1314 | __cap_set_timeouts(mdsc, ci); | |
1315 | } | |
a8599bd8 SW |
1316 | |
1317 | cap->issued &= retain; /* drop bits we don't want */ | |
1318 | if (cap->implemented & ~cap->issued) { | |
1319 | /* | |
1320 | * Wake up any waiters on wanted -> needed transition. | |
1321 | * This is due to the weird transition from buffered | |
1322 | * to sync IO... we need to flush dirty pages _before_ | |
1323 | * allowing sync writes to avoid reordering. | |
1324 | */ | |
1325 | wake = 1; | |
1326 | } | |
1327 | cap->implemented &= cap->issued | used; | |
1328 | cap->mds_wanted = want; | |
1329 | ||
0ff8bfb3 JL |
1330 | arg.ino = ceph_vino(inode).ino; |
1331 | arg.cid = cap->cap_id; | |
1332 | arg.follows = flushing ? ci->i_head_snapc->seq : 0; | |
1333 | arg.flush_tid = flush_tid; | |
1334 | arg.oldest_flush_tid = oldest_flush_tid; | |
1335 | ||
1336 | arg.size = inode->i_size; | |
1337 | ci->i_reported_size = arg.size; | |
1338 | arg.max_size = ci->i_wanted_max_size; | |
1339 | ci->i_requested_max_size = arg.max_size; | |
a8599bd8 | 1340 | |
082afec9 | 1341 | if (flushing & CEPH_CAP_XATTR_EXCL) { |
a8599bd8 | 1342 | __ceph_build_xattrs_blob(ci); |
0ff8bfb3 JL |
1343 | arg.xattr_version = ci->i_xattrs.version; |
1344 | arg.xattr_buf = ci->i_xattrs.blob; | |
1345 | } else { | |
1346 | arg.xattr_buf = NULL; | |
a8599bd8 SW |
1347 | } |
1348 | ||
9bbeab41 AB |
1349 | arg.mtime = inode->i_mtime; |
1350 | arg.atime = inode->i_atime; | |
1351 | arg.ctime = inode->i_ctime; | |
0ff8bfb3 JL |
1352 | |
1353 | arg.op = op; | |
1354 | arg.caps = cap->implemented; | |
1355 | arg.wanted = want; | |
1356 | arg.dirty = flushing; | |
1357 | ||
1358 | arg.seq = cap->seq; | |
1359 | arg.issue_seq = cap->issue_seq; | |
1360 | arg.mseq = cap->mseq; | |
1361 | arg.time_warp_seq = ci->i_time_warp_seq; | |
1362 | ||
1363 | arg.uid = inode->i_uid; | |
1364 | arg.gid = inode->i_gid; | |
1365 | arg.mode = inode->i_mode; | |
1366 | ||
1367 | arg.inline_data = ci->i_inline_version != CEPH_INLINE_NONE; | |
95569713 YZ |
1368 | if (list_empty(&ci->i_cap_snaps)) |
1369 | arg.flags = CEPH_CLIENT_CAPS_NO_CAPSNAP; | |
1370 | else | |
1371 | arg.flags = CEPH_CLIENT_CAPS_PENDING_CAPSNAP; | |
1e4ef0c6 JL |
1372 | if (sync) |
1373 | arg.flags |= CEPH_CLIENT_CAPS_SYNC; | |
e20d258d | 1374 | |
be655596 | 1375 | spin_unlock(&ci->i_ceph_lock); |
a8599bd8 | 1376 | |
0ff8bfb3 | 1377 | ret = send_cap_msg(&arg); |
a8599bd8 SW |
1378 | if (ret < 0) { |
1379 | dout("error sending cap msg, must requeue %p\n", inode); | |
1380 | delayed = 1; | |
1381 | } | |
1382 | ||
1383 | if (wake) | |
03066f23 | 1384 | wake_up_all(&ci->i_cap_wq); |
a8599bd8 SW |
1385 | |
1386 | return delayed; | |
1387 | } | |
1388 | ||
0e294387 YZ |
1389 | static inline int __send_flush_snap(struct inode *inode, |
1390 | struct ceph_mds_session *session, | |
1391 | struct ceph_cap_snap *capsnap, | |
1392 | u32 mseq, u64 oldest_flush_tid) | |
1393 | { | |
0ff8bfb3 JL |
1394 | struct cap_msg_args arg; |
1395 | ||
1396 | arg.session = session; | |
1397 | arg.ino = ceph_vino(inode).ino; | |
1398 | arg.cid = 0; | |
1399 | arg.follows = capsnap->follows; | |
1400 | arg.flush_tid = capsnap->cap_flush.tid; | |
1401 | arg.oldest_flush_tid = oldest_flush_tid; | |
1402 | ||
1403 | arg.size = capsnap->size; | |
1404 | arg.max_size = 0; | |
1405 | arg.xattr_version = capsnap->xattr_version; | |
1406 | arg.xattr_buf = capsnap->xattr_blob; | |
1407 | ||
1408 | arg.atime = capsnap->atime; | |
1409 | arg.mtime = capsnap->mtime; | |
1410 | arg.ctime = capsnap->ctime; | |
1411 | ||
1412 | arg.op = CEPH_CAP_OP_FLUSHSNAP; | |
1413 | arg.caps = capsnap->issued; | |
1414 | arg.wanted = 0; | |
1415 | arg.dirty = capsnap->dirty; | |
1416 | ||
1417 | arg.seq = 0; | |
1418 | arg.issue_seq = 0; | |
1419 | arg.mseq = mseq; | |
1420 | arg.time_warp_seq = capsnap->time_warp_seq; | |
1421 | ||
1422 | arg.uid = capsnap->uid; | |
1423 | arg.gid = capsnap->gid; | |
1424 | arg.mode = capsnap->mode; | |
1425 | ||
1426 | arg.inline_data = capsnap->inline_data; | |
1e4ef0c6 | 1427 | arg.flags = 0; |
0ff8bfb3 JL |
1428 | |
1429 | return send_cap_msg(&arg); | |
0e294387 YZ |
1430 | } |
1431 | ||
a8599bd8 SW |
1432 | /* |
1433 | * When a snapshot is taken, clients accumulate dirty metadata on | |
1434 | * inodes with capabilities in ceph_cap_snaps to describe the file | |
1435 | * state at the time the snapshot was taken. This must be flushed | |
1436 | * asynchronously back to the MDS once sync writes complete and dirty | |
1437 | * data is written out. | |
1438 | * | |
be655596 | 1439 | * Called under i_ceph_lock. Takes s_mutex as needed. |
a8599bd8 | 1440 | */ |
ed9b430c YZ |
1441 | static void __ceph_flush_snaps(struct ceph_inode_info *ci, |
1442 | struct ceph_mds_session *session) | |
be655596 SW |
1443 | __releases(ci->i_ceph_lock) |
1444 | __acquires(ci->i_ceph_lock) | |
a8599bd8 SW |
1445 | { |
1446 | struct inode *inode = &ci->vfs_inode; | |
ed9b430c | 1447 | struct ceph_mds_client *mdsc = session->s_mdsc; |
a8599bd8 | 1448 | struct ceph_cap_snap *capsnap; |
ed9b430c YZ |
1449 | u64 oldest_flush_tid = 0; |
1450 | u64 first_tid = 1, last_tid = 0; | |
a8599bd8 | 1451 | |
ed9b430c | 1452 | dout("__flush_snaps %p session %p\n", inode, session); |
a8599bd8 | 1453 | |
a8599bd8 | 1454 | list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) { |
a8599bd8 SW |
1455 | /* |
1456 | * we need to wait for sync writes to complete and for dirty | |
1457 | * pages to be written out. | |
1458 | */ | |
1459 | if (capsnap->dirty_pages || capsnap->writing) | |
cfc0bf66 | 1460 | break; |
a8599bd8 | 1461 | |
86056090 YZ |
1462 | /* should be removed by ceph_try_drop_cap_snap() */ |
1463 | BUG_ON(!capsnap->need_flush); | |
819ccbfa | 1464 | |
e835124c | 1465 | /* only flush each capsnap once */ |
0e294387 | 1466 | if (capsnap->cap_flush.tid > 0) { |
ed9b430c | 1467 | dout(" already flushed %p, skipping\n", capsnap); |
e835124c SW |
1468 | continue; |
1469 | } | |
1470 | ||
553adfd9 | 1471 | spin_lock(&mdsc->cap_dirty_lock); |
0e294387 YZ |
1472 | capsnap->cap_flush.tid = ++mdsc->last_cap_flush_tid; |
1473 | list_add_tail(&capsnap->cap_flush.g_list, | |
1474 | &mdsc->cap_flush_list); | |
ed9b430c YZ |
1475 | if (oldest_flush_tid == 0) |
1476 | oldest_flush_tid = __get_oldest_flush_tid(mdsc); | |
0e294387 YZ |
1477 | if (list_empty(&ci->i_flushing_item)) { |
1478 | list_add_tail(&ci->i_flushing_item, | |
1479 | &session->s_cap_flushing); | |
1480 | } | |
553adfd9 YZ |
1481 | spin_unlock(&mdsc->cap_dirty_lock); |
1482 | ||
0e294387 YZ |
1483 | list_add_tail(&capsnap->cap_flush.i_list, |
1484 | &ci->i_cap_flush_list); | |
1485 | ||
ed9b430c YZ |
1486 | if (first_tid == 1) |
1487 | first_tid = capsnap->cap_flush.tid; | |
1488 | last_tid = capsnap->cap_flush.tid; | |
1489 | } | |
1490 | ||
1491 | ci->i_ceph_flags &= ~CEPH_I_FLUSH_SNAPS; | |
1492 | ||
1493 | while (first_tid <= last_tid) { | |
1494 | struct ceph_cap *cap = ci->i_auth_cap; | |
1495 | struct ceph_cap_flush *cf; | |
1496 | int ret; | |
1497 | ||
1498 | if (!(cap && cap->session == session)) { | |
1499 | dout("__flush_snaps %p auth cap %p not mds%d, " | |
1500 | "stop\n", inode, cap, session->s_mds); | |
1501 | break; | |
1502 | } | |
1503 | ||
1504 | ret = -ENOENT; | |
1505 | list_for_each_entry(cf, &ci->i_cap_flush_list, i_list) { | |
1506 | if (cf->tid >= first_tid) { | |
1507 | ret = 0; | |
1508 | break; | |
1509 | } | |
1510 | } | |
1511 | if (ret < 0) | |
1512 | break; | |
1513 | ||
1514 | first_tid = cf->tid + 1; | |
1515 | ||
1516 | capsnap = container_of(cf, struct ceph_cap_snap, cap_flush); | |
805692d0 | 1517 | refcount_inc(&capsnap->nref); |
be655596 | 1518 | spin_unlock(&ci->i_ceph_lock); |
a8599bd8 | 1519 | |
ed9b430c YZ |
1520 | dout("__flush_snaps %p capsnap %p tid %llu %s\n", |
1521 | inode, capsnap, cf->tid, ceph_cap_string(capsnap->dirty)); | |
a8599bd8 | 1522 | |
ed9b430c YZ |
1523 | ret = __send_flush_snap(inode, session, capsnap, cap->mseq, |
1524 | oldest_flush_tid); | |
1525 | if (ret < 0) { | |
1526 | pr_err("__flush_snaps: error sending cap flushsnap, " | |
1527 | "ino (%llx.%llx) tid %llu follows %llu\n", | |
1528 | ceph_vinop(inode), cf->tid, capsnap->follows); | |
1529 | } | |
a8599bd8 | 1530 | |
ed9b430c | 1531 | ceph_put_cap_snap(capsnap); |
be655596 | 1532 | spin_lock(&ci->i_ceph_lock); |
a8599bd8 | 1533 | } |
ed9b430c | 1534 | } |
a8599bd8 | 1535 | |
ed9b430c YZ |
1536 | void ceph_flush_snaps(struct ceph_inode_info *ci, |
1537 | struct ceph_mds_session **psession) | |
1538 | { | |
1539 | struct inode *inode = &ci->vfs_inode; | |
1540 | struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc; | |
e4d2b16a | 1541 | struct ceph_mds_session *session = NULL; |
ed9b430c | 1542 | int mds; |
e4d2b16a | 1543 | |
ed9b430c | 1544 | dout("ceph_flush_snaps %p\n", inode); |
e4d2b16a YZ |
1545 | if (psession) |
1546 | session = *psession; | |
ed9b430c YZ |
1547 | retry: |
1548 | spin_lock(&ci->i_ceph_lock); | |
1549 | if (!(ci->i_ceph_flags & CEPH_I_FLUSH_SNAPS)) { | |
1550 | dout(" no capsnap needs flush, doing nothing\n"); | |
1551 | goto out; | |
1552 | } | |
1553 | if (!ci->i_auth_cap) { | |
1554 | dout(" no auth cap (migrating?), doing nothing\n"); | |
1555 | goto out; | |
1556 | } | |
a8599bd8 | 1557 | |
ed9b430c YZ |
1558 | mds = ci->i_auth_cap->session->s_mds; |
1559 | if (session && session->s_mds != mds) { | |
1560 | dout(" oops, wrong session %p mutex\n", session); | |
a8599bd8 SW |
1561 | mutex_unlock(&session->s_mutex); |
1562 | ceph_put_mds_session(session); | |
ed9b430c YZ |
1563 | session = NULL; |
1564 | } | |
1565 | if (!session) { | |
1566 | spin_unlock(&ci->i_ceph_lock); | |
1567 | mutex_lock(&mdsc->mutex); | |
1568 | session = __ceph_lookup_mds_session(mdsc, mds); | |
1569 | mutex_unlock(&mdsc->mutex); | |
1570 | if (session) { | |
1571 | dout(" inverting session/ino locks on %p\n", session); | |
1572 | mutex_lock(&session->s_mutex); | |
1573 | } | |
1574 | goto retry; | |
a8599bd8 | 1575 | } |
a8599bd8 | 1576 | |
24d063ac YZ |
1577 | // make sure flushsnap messages are sent in proper order. |
1578 | if (ci->i_ceph_flags & CEPH_I_KICK_FLUSH) { | |
1579 | __kick_flushing_caps(mdsc, session, ci, 0); | |
1580 | ci->i_ceph_flags &= ~CEPH_I_KICK_FLUSH; | |
1581 | } | |
1582 | ||
ed9b430c YZ |
1583 | __ceph_flush_snaps(ci, session); |
1584 | out: | |
be655596 | 1585 | spin_unlock(&ci->i_ceph_lock); |
ed9b430c YZ |
1586 | |
1587 | if (psession) { | |
1588 | *psession = session; | |
c858a070 | 1589 | } else if (session) { |
ed9b430c YZ |
1590 | mutex_unlock(&session->s_mutex); |
1591 | ceph_put_mds_session(session); | |
1592 | } | |
1593 | /* we flushed them all; remove this inode from the queue */ | |
1594 | spin_lock(&mdsc->snap_flush_lock); | |
1595 | list_del_init(&ci->i_snap_flush_item); | |
1596 | spin_unlock(&mdsc->snap_flush_lock); | |
a8599bd8 SW |
1597 | } |
1598 | ||
76e3b390 | 1599 | /* |
fca65b4a SW |
1600 | * Mark caps dirty. If inode is newly dirty, return the dirty flags. |
1601 | * Caller is then responsible for calling __mark_inode_dirty with the | |
1602 | * returned flags value. | |
76e3b390 | 1603 | */ |
f66fd9f0 YZ |
1604 | int __ceph_mark_dirty_caps(struct ceph_inode_info *ci, int mask, |
1605 | struct ceph_cap_flush **pcf) | |
76e3b390 | 1606 | { |
640ef79d | 1607 | struct ceph_mds_client *mdsc = |
3d14c5d2 | 1608 | ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc; |
76e3b390 SW |
1609 | struct inode *inode = &ci->vfs_inode; |
1610 | int was = ci->i_dirty_caps; | |
1611 | int dirty = 0; | |
1612 | ||
571ade33 YZ |
1613 | if (!ci->i_auth_cap) { |
1614 | pr_warn("__mark_dirty_caps %p %llx mask %s, " | |
1615 | "but no auth cap (session was closed?)\n", | |
1616 | inode, ceph_ino(inode), ceph_cap_string(mask)); | |
1617 | return 0; | |
1618 | } | |
1619 | ||
76e3b390 SW |
1620 | dout("__mark_dirty_caps %p %s dirty %s -> %s\n", &ci->vfs_inode, |
1621 | ceph_cap_string(mask), ceph_cap_string(was), | |
1622 | ceph_cap_string(was | mask)); | |
1623 | ci->i_dirty_caps |= mask; | |
1624 | if (was == 0) { | |
f66fd9f0 YZ |
1625 | WARN_ON_ONCE(ci->i_prealloc_cap_flush); |
1626 | swap(ci->i_prealloc_cap_flush, *pcf); | |
1627 | ||
604d1b02 YZ |
1628 | if (!ci->i_head_snapc) { |
1629 | WARN_ON_ONCE(!rwsem_is_locked(&mdsc->snap_rwsem)); | |
7d8cb26d SW |
1630 | ci->i_head_snapc = ceph_get_snap_context( |
1631 | ci->i_snap_realm->cached_context); | |
604d1b02 | 1632 | } |
0685235f YZ |
1633 | dout(" inode %p now dirty snapc %p auth cap %p\n", |
1634 | &ci->vfs_inode, ci->i_head_snapc, ci->i_auth_cap); | |
76e3b390 SW |
1635 | BUG_ON(!list_empty(&ci->i_dirty_item)); |
1636 | spin_lock(&mdsc->cap_dirty_lock); | |
11df2dfb | 1637 | list_add(&ci->i_dirty_item, &mdsc->cap_dirty); |
76e3b390 SW |
1638 | spin_unlock(&mdsc->cap_dirty_lock); |
1639 | if (ci->i_flushing_caps == 0) { | |
3772d26d | 1640 | ihold(inode); |
76e3b390 SW |
1641 | dirty |= I_DIRTY_SYNC; |
1642 | } | |
f66fd9f0 YZ |
1643 | } else { |
1644 | WARN_ON_ONCE(!ci->i_prealloc_cap_flush); | |
76e3b390 SW |
1645 | } |
1646 | BUG_ON(list_empty(&ci->i_dirty_item)); | |
1647 | if (((was | ci->i_flushing_caps) & CEPH_CAP_FILE_BUFFER) && | |
1648 | (mask & CEPH_CAP_FILE_BUFFER)) | |
1649 | dirty |= I_DIRTY_DATASYNC; | |
76e3b390 | 1650 | __cap_delay_requeue(mdsc, ci); |
fca65b4a | 1651 | return dirty; |
76e3b390 SW |
1652 | } |
1653 | ||
f66fd9f0 YZ |
1654 | struct ceph_cap_flush *ceph_alloc_cap_flush(void) |
1655 | { | |
1656 | return kmem_cache_alloc(ceph_cap_flush_cachep, GFP_KERNEL); | |
1657 | } | |
1658 | ||
1659 | void ceph_free_cap_flush(struct ceph_cap_flush *cf) | |
1660 | { | |
1661 | if (cf) | |
1662 | kmem_cache_free(ceph_cap_flush_cachep, cf); | |
1663 | } | |
1664 | ||
a2971c8c YZ |
1665 | static u64 __get_oldest_flush_tid(struct ceph_mds_client *mdsc) |
1666 | { | |
e4500b5e | 1667 | if (!list_empty(&mdsc->cap_flush_list)) { |
a2971c8c | 1668 | struct ceph_cap_flush *cf = |
e4500b5e YZ |
1669 | list_first_entry(&mdsc->cap_flush_list, |
1670 | struct ceph_cap_flush, g_list); | |
a2971c8c YZ |
1671 | return cf->tid; |
1672 | } | |
1673 | return 0; | |
1674 | } | |
1675 | ||
c8799fc4 YZ |
1676 | /* |
1677 | * Remove cap_flush from the mdsc's or inode's flushing cap list. | |
1678 | * Return true if caller needs to wake up flush waiters. | |
1679 | */ | |
1680 | static bool __finish_cap_flush(struct ceph_mds_client *mdsc, | |
1681 | struct ceph_inode_info *ci, | |
1682 | struct ceph_cap_flush *cf) | |
1683 | { | |
1684 | struct ceph_cap_flush *prev; | |
1685 | bool wake = cf->wake; | |
1686 | if (mdsc) { | |
1687 | /* are there older pending cap flushes? */ | |
1688 | if (wake && cf->g_list.prev != &mdsc->cap_flush_list) { | |
1689 | prev = list_prev_entry(cf, g_list); | |
1690 | prev->wake = true; | |
1691 | wake = false; | |
1692 | } | |
1693 | list_del(&cf->g_list); | |
1694 | } else if (ci) { | |
1695 | if (wake && cf->i_list.prev != &ci->i_cap_flush_list) { | |
1696 | prev = list_prev_entry(cf, i_list); | |
1697 | prev->wake = true; | |
1698 | wake = false; | |
1699 | } | |
1700 | list_del(&cf->i_list); | |
1701 | } else { | |
1702 | BUG_ON(1); | |
1703 | } | |
1704 | return wake; | |
1705 | } | |
1706 | ||
a8599bd8 SW |
1707 | /* |
1708 | * Add dirty inode to the flushing list. Assigned a seq number so we | |
1709 | * can wait for caps to flush without starving. | |
cdc35f96 | 1710 | * |
be655596 | 1711 | * Called under i_ceph_lock. |
a8599bd8 | 1712 | */ |
cdc35f96 | 1713 | static int __mark_caps_flushing(struct inode *inode, |
c8799fc4 | 1714 | struct ceph_mds_session *session, bool wake, |
a2971c8c | 1715 | u64 *flush_tid, u64 *oldest_flush_tid) |
a8599bd8 | 1716 | { |
3d14c5d2 | 1717 | struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc; |
a8599bd8 | 1718 | struct ceph_inode_info *ci = ceph_inode(inode); |
f66fd9f0 | 1719 | struct ceph_cap_flush *cf = NULL; |
cdc35f96 | 1720 | int flushing; |
50b885b9 | 1721 | |
cdc35f96 | 1722 | BUG_ON(ci->i_dirty_caps == 0); |
a8599bd8 | 1723 | BUG_ON(list_empty(&ci->i_dirty_item)); |
f66fd9f0 | 1724 | BUG_ON(!ci->i_prealloc_cap_flush); |
cdc35f96 SW |
1725 | |
1726 | flushing = ci->i_dirty_caps; | |
1727 | dout("__mark_caps_flushing flushing %s, flushing_caps %s -> %s\n", | |
1728 | ceph_cap_string(flushing), | |
1729 | ceph_cap_string(ci->i_flushing_caps), | |
1730 | ceph_cap_string(ci->i_flushing_caps | flushing)); | |
1731 | ci->i_flushing_caps |= flushing; | |
1732 | ci->i_dirty_caps = 0; | |
afcdaea3 | 1733 | dout(" inode %p now !dirty\n", inode); |
cdc35f96 | 1734 | |
f66fd9f0 | 1735 | swap(cf, ci->i_prealloc_cap_flush); |
553adfd9 | 1736 | cf->caps = flushing; |
c8799fc4 | 1737 | cf->wake = wake; |
553adfd9 | 1738 | |
a8599bd8 | 1739 | spin_lock(&mdsc->cap_dirty_lock); |
afcdaea3 SW |
1740 | list_del_init(&ci->i_dirty_item); |
1741 | ||
553adfd9 | 1742 | cf->tid = ++mdsc->last_cap_flush_tid; |
e4500b5e | 1743 | list_add_tail(&cf->g_list, &mdsc->cap_flush_list); |
a2971c8c | 1744 | *oldest_flush_tid = __get_oldest_flush_tid(mdsc); |
553adfd9 | 1745 | |
a8599bd8 SW |
1746 | if (list_empty(&ci->i_flushing_item)) { |
1747 | list_add_tail(&ci->i_flushing_item, &session->s_cap_flushing); | |
1748 | mdsc->num_cap_flushing++; | |
a8599bd8 SW |
1749 | } |
1750 | spin_unlock(&mdsc->cap_dirty_lock); | |
cdc35f96 | 1751 | |
e4500b5e | 1752 | list_add_tail(&cf->i_list, &ci->i_cap_flush_list); |
553adfd9 YZ |
1753 | |
1754 | *flush_tid = cf->tid; | |
cdc35f96 | 1755 | return flushing; |
a8599bd8 SW |
1756 | } |
1757 | ||
5ecad6fd SW |
1758 | /* |
1759 | * try to invalidate mapping pages without blocking. | |
1760 | */ | |
5ecad6fd SW |
1761 | static int try_nonblocking_invalidate(struct inode *inode) |
1762 | { | |
1763 | struct ceph_inode_info *ci = ceph_inode(inode); | |
1764 | u32 invalidating_gen = ci->i_rdcache_gen; | |
1765 | ||
be655596 | 1766 | spin_unlock(&ci->i_ceph_lock); |
5ecad6fd | 1767 | invalidate_mapping_pages(&inode->i_data, 0, -1); |
be655596 | 1768 | spin_lock(&ci->i_ceph_lock); |
5ecad6fd | 1769 | |
18a38193 | 1770 | if (inode->i_data.nrpages == 0 && |
5ecad6fd SW |
1771 | invalidating_gen == ci->i_rdcache_gen) { |
1772 | /* success. */ | |
1773 | dout("try_nonblocking_invalidate %p success\n", inode); | |
cd045cb4 SW |
1774 | /* save any racing async invalidate some trouble */ |
1775 | ci->i_rdcache_revoking = ci->i_rdcache_gen - 1; | |
5ecad6fd SW |
1776 | return 0; |
1777 | } | |
1778 | dout("try_nonblocking_invalidate %p failed\n", inode); | |
1779 | return -1; | |
1780 | } | |
1781 | ||
efb0ca76 YZ |
1782 | bool __ceph_should_report_size(struct ceph_inode_info *ci) |
1783 | { | |
1784 | loff_t size = ci->vfs_inode.i_size; | |
1785 | /* mds will adjust max size according to the reported size */ | |
1786 | if (ci->i_flushing_caps & CEPH_CAP_FILE_WR) | |
1787 | return false; | |
1788 | if (size >= ci->i_max_size) | |
1789 | return true; | |
1790 | /* half of previous max_size increment has been used */ | |
1791 | if (ci->i_max_size > ci->i_reported_size && | |
1792 | (size << 1) >= ci->i_max_size + ci->i_reported_size) | |
1793 | return true; | |
1794 | return false; | |
1795 | } | |
1796 | ||
a8599bd8 SW |
1797 | /* |
1798 | * Swiss army knife function to examine currently used and wanted | |
1799 | * versus held caps. Release, flush, ack revoked caps to mds as | |
1800 | * appropriate. | |
1801 | * | |
1802 | * CHECK_CAPS_NODELAY - caller is delayed work and we should not delay | |
1803 | * cap release further. | |
1804 | * CHECK_CAPS_AUTHONLY - we should only check the auth cap | |
1805 | * CHECK_CAPS_FLUSH - we should flush any dirty caps immediately, without | |
1806 | * further delay. | |
1807 | */ | |
1808 | void ceph_check_caps(struct ceph_inode_info *ci, int flags, | |
1809 | struct ceph_mds_session *session) | |
1810 | { | |
3d14c5d2 YS |
1811 | struct ceph_fs_client *fsc = ceph_inode_to_client(&ci->vfs_inode); |
1812 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
a8599bd8 SW |
1813 | struct inode *inode = &ci->vfs_inode; |
1814 | struct ceph_cap *cap; | |
a2971c8c | 1815 | u64 flush_tid, oldest_flush_tid; |
395c312b | 1816 | int file_wanted, used, cap_used; |
a8599bd8 | 1817 | int took_snap_rwsem = 0; /* true if mdsc->snap_rwsem held */ |
cbd03635 | 1818 | int issued, implemented, want, retain, revoking, flushing = 0; |
a8599bd8 SW |
1819 | int mds = -1; /* keep track of how far we've gone through i_caps list |
1820 | to avoid an infinite loop on retry */ | |
1821 | struct rb_node *p; | |
0f439c74 YZ |
1822 | int delayed = 0, sent = 0; |
1823 | bool no_delay = flags & CHECK_CAPS_NODELAY; | |
3609404f | 1824 | bool queue_invalidate = false; |
3609404f | 1825 | bool tried_invalidate = false; |
a8599bd8 SW |
1826 | |
1827 | /* if we are unmounting, flush any unused caps immediately. */ | |
1828 | if (mdsc->stopping) | |
0f439c74 | 1829 | no_delay = true; |
a8599bd8 | 1830 | |
be655596 | 1831 | spin_lock(&ci->i_ceph_lock); |
a8599bd8 SW |
1832 | |
1833 | if (ci->i_ceph_flags & CEPH_I_FLUSH) | |
1834 | flags |= CHECK_CAPS_FLUSH; | |
1835 | ||
0f439c74 YZ |
1836 | if (!(flags & CHECK_CAPS_AUTHONLY) || |
1837 | (ci->i_auth_cap && __ceph_is_single_caps(ci))) | |
1838 | __cap_delay_cancel(mdsc, ci); | |
1839 | ||
a8599bd8 SW |
1840 | goto retry_locked; |
1841 | retry: | |
be655596 | 1842 | spin_lock(&ci->i_ceph_lock); |
a8599bd8 SW |
1843 | retry_locked: |
1844 | file_wanted = __ceph_caps_file_wanted(ci); | |
1845 | used = __ceph_caps_used(ci); | |
cbd03635 SW |
1846 | issued = __ceph_caps_issued(ci, &implemented); |
1847 | revoking = implemented & ~issued; | |
a8599bd8 | 1848 | |
41445999 YZ |
1849 | want = file_wanted; |
1850 | retain = file_wanted | used | CEPH_CAP_PIN; | |
a8599bd8 | 1851 | if (!mdsc->stopping && inode->i_nlink > 0) { |
41445999 | 1852 | if (file_wanted) { |
a8599bd8 | 1853 | retain |= CEPH_CAP_ANY; /* be greedy */ |
32ec4397 YZ |
1854 | } else if (S_ISDIR(inode->i_mode) && |
1855 | (issued & CEPH_CAP_FILE_SHARED) && | |
1856 | __ceph_dir_is_complete(ci)) { | |
1857 | /* | |
1858 | * If a directory is complete, we want to keep | |
1859 | * the exclusive cap. So that MDS does not end up | |
1860 | * revoking the shared cap on every create/unlink | |
1861 | * operation. | |
1862 | */ | |
1863 | want = CEPH_CAP_ANY_SHARED | CEPH_CAP_FILE_EXCL; | |
1864 | retain |= want; | |
a8599bd8 | 1865 | } else { |
32ec4397 | 1866 | |
a8599bd8 SW |
1867 | retain |= CEPH_CAP_ANY_SHARED; |
1868 | /* | |
1869 | * keep RD only if we didn't have the file open RW, | |
1870 | * because then the mds would revoke it anyway to | |
1871 | * journal max_size=0. | |
1872 | */ | |
1873 | if (ci->i_max_size == 0) | |
1874 | retain |= CEPH_CAP_ANY_RD; | |
1875 | } | |
1876 | } | |
1877 | ||
1878 | dout("check_caps %p file_want %s used %s dirty %s flushing %s" | |
cbd03635 | 1879 | " issued %s revoking %s retain %s %s%s%s\n", inode, |
a8599bd8 SW |
1880 | ceph_cap_string(file_wanted), |
1881 | ceph_cap_string(used), ceph_cap_string(ci->i_dirty_caps), | |
1882 | ceph_cap_string(ci->i_flushing_caps), | |
cbd03635 | 1883 | ceph_cap_string(issued), ceph_cap_string(revoking), |
a8599bd8 SW |
1884 | ceph_cap_string(retain), |
1885 | (flags & CHECK_CAPS_AUTHONLY) ? " AUTHONLY" : "", | |
1886 | (flags & CHECK_CAPS_NODELAY) ? " NODELAY" : "", | |
1887 | (flags & CHECK_CAPS_FLUSH) ? " FLUSH" : ""); | |
1888 | ||
1889 | /* | |
1890 | * If we no longer need to hold onto old our caps, and we may | |
1891 | * have cached pages, but don't want them, then try to invalidate. | |
1892 | * If we fail, it's because pages are locked.... try again later. | |
1893 | */ | |
0f439c74 | 1894 | if ((!no_delay || mdsc->stopping) && |
fdd4e158 | 1895 | !S_ISDIR(inode->i_mode) && /* ignore readdir cache */ |
9abd4db7 | 1896 | !(ci->i_wb_ref || ci->i_wrbuffer_ref) && /* no dirty pages... */ |
fdd4e158 | 1897 | inode->i_data.nrpages && /* have cached pages */ |
5e804ac4 YZ |
1898 | (revoking & (CEPH_CAP_FILE_CACHE| |
1899 | CEPH_CAP_FILE_LAZYIO)) && /* or revoking cache */ | |
a8599bd8 | 1900 | !tried_invalidate) { |
a8599bd8 | 1901 | dout("check_caps trying to invalidate on %p\n", inode); |
5ecad6fd | 1902 | if (try_nonblocking_invalidate(inode) < 0) { |
ee612d95 YZ |
1903 | dout("check_caps queuing invalidate\n"); |
1904 | queue_invalidate = true; | |
1905 | ci->i_rdcache_revoking = ci->i_rdcache_gen; | |
a8599bd8 | 1906 | } |
3609404f | 1907 | tried_invalidate = true; |
a8599bd8 SW |
1908 | goto retry_locked; |
1909 | } | |
1910 | ||
a8599bd8 SW |
1911 | for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) { |
1912 | cap = rb_entry(p, struct ceph_cap, ci_node); | |
a8599bd8 SW |
1913 | |
1914 | /* avoid looping forever */ | |
1915 | if (mds >= cap->mds || | |
1916 | ((flags & CHECK_CAPS_AUTHONLY) && cap != ci->i_auth_cap)) | |
1917 | continue; | |
1918 | ||
1919 | /* NOTE: no side-effects allowed, until we take s_mutex */ | |
1920 | ||
395c312b YZ |
1921 | cap_used = used; |
1922 | if (ci->i_auth_cap && cap != ci->i_auth_cap) | |
1923 | cap_used &= ~ci->i_auth_cap->issued; | |
1924 | ||
a8599bd8 | 1925 | revoking = cap->implemented & ~cap->issued; |
395c312b | 1926 | dout(" mds%d cap %p used %s issued %s implemented %s revoking %s\n", |
9abd4db7 YZ |
1927 | cap->mds, cap, ceph_cap_string(cap_used), |
1928 | ceph_cap_string(cap->issued), | |
088b3f5e SW |
1929 | ceph_cap_string(cap->implemented), |
1930 | ceph_cap_string(revoking)); | |
a8599bd8 SW |
1931 | |
1932 | if (cap == ci->i_auth_cap && | |
1933 | (cap->issued & CEPH_CAP_FILE_WR)) { | |
1934 | /* request larger max_size from MDS? */ | |
1935 | if (ci->i_wanted_max_size > ci->i_max_size && | |
1936 | ci->i_wanted_max_size > ci->i_requested_max_size) { | |
1937 | dout("requesting new max_size\n"); | |
1938 | goto ack; | |
1939 | } | |
1940 | ||
1941 | /* approaching file_max? */ | |
efb0ca76 | 1942 | if (__ceph_should_report_size(ci)) { |
a8599bd8 SW |
1943 | dout("i_size approaching max_size\n"); |
1944 | goto ack; | |
1945 | } | |
1946 | } | |
1947 | /* flush anything dirty? */ | |
7bc00fdd YZ |
1948 | if (cap == ci->i_auth_cap) { |
1949 | if ((flags & CHECK_CAPS_FLUSH) && ci->i_dirty_caps) { | |
1950 | dout("flushing dirty caps\n"); | |
1951 | goto ack; | |
1952 | } | |
1953 | if (ci->i_ceph_flags & CEPH_I_FLUSH_SNAPS) { | |
1954 | dout("flushing snap caps\n"); | |
1955 | goto ack; | |
1956 | } | |
a8599bd8 SW |
1957 | } |
1958 | ||
1959 | /* completed revocation? going down and there are no caps? */ | |
395c312b | 1960 | if (revoking && (revoking & cap_used) == 0) { |
a8599bd8 SW |
1961 | dout("completed revocation of %s\n", |
1962 | ceph_cap_string(cap->implemented & ~cap->issued)); | |
1963 | goto ack; | |
1964 | } | |
1965 | ||
1966 | /* want more caps from mds? */ | |
1967 | if (want & ~(cap->mds_wanted | cap->issued)) | |
1968 | goto ack; | |
1969 | ||
1970 | /* things we might delay */ | |
1971 | if ((cap->issued & ~retain) == 0 && | |
1972 | cap->mds_wanted == want) | |
1973 | continue; /* nope, all good */ | |
1974 | ||
0f439c74 | 1975 | if (no_delay) |
a8599bd8 SW |
1976 | goto ack; |
1977 | ||
1978 | /* delay? */ | |
1979 | if ((ci->i_ceph_flags & CEPH_I_NODELAY) == 0 && | |
1980 | time_before(jiffies, ci->i_hold_caps_max)) { | |
1981 | dout(" delaying issued %s -> %s, wanted %s -> %s\n", | |
1982 | ceph_cap_string(cap->issued), | |
1983 | ceph_cap_string(cap->issued & retain), | |
1984 | ceph_cap_string(cap->mds_wanted), | |
1985 | ceph_cap_string(want)); | |
1986 | delayed++; | |
1987 | continue; | |
1988 | } | |
1989 | ||
1990 | ack: | |
e9964c10 SW |
1991 | if (ci->i_ceph_flags & CEPH_I_NOFLUSH) { |
1992 | dout(" skipping %p I_NOFLUSH set\n", inode); | |
1993 | continue; | |
1994 | } | |
1995 | ||
a8599bd8 SW |
1996 | if (session && session != cap->session) { |
1997 | dout("oops, wrong session %p mutex\n", session); | |
1998 | mutex_unlock(&session->s_mutex); | |
1999 | session = NULL; | |
2000 | } | |
2001 | if (!session) { | |
2002 | session = cap->session; | |
2003 | if (mutex_trylock(&session->s_mutex) == 0) { | |
2004 | dout("inverting session/ino locks on %p\n", | |
2005 | session); | |
be655596 | 2006 | spin_unlock(&ci->i_ceph_lock); |
a8599bd8 SW |
2007 | if (took_snap_rwsem) { |
2008 | up_read(&mdsc->snap_rwsem); | |
2009 | took_snap_rwsem = 0; | |
2010 | } | |
2011 | mutex_lock(&session->s_mutex); | |
2012 | goto retry; | |
2013 | } | |
2014 | } | |
7bc00fdd YZ |
2015 | |
2016 | /* kick flushing and flush snaps before sending normal | |
2017 | * cap message */ | |
2018 | if (cap == ci->i_auth_cap && | |
2019 | (ci->i_ceph_flags & | |
2020 | (CEPH_I_KICK_FLUSH | CEPH_I_FLUSH_SNAPS))) { | |
2021 | if (ci->i_ceph_flags & CEPH_I_KICK_FLUSH) { | |
24d063ac | 2022 | __kick_flushing_caps(mdsc, session, ci, 0); |
7bc00fdd YZ |
2023 | ci->i_ceph_flags &= ~CEPH_I_KICK_FLUSH; |
2024 | } | |
ed9b430c YZ |
2025 | if (ci->i_ceph_flags & CEPH_I_FLUSH_SNAPS) |
2026 | __ceph_flush_snaps(ci, session); | |
2027 | ||
7bc00fdd YZ |
2028 | goto retry_locked; |
2029 | } | |
2030 | ||
a8599bd8 SW |
2031 | /* take snap_rwsem after session mutex */ |
2032 | if (!took_snap_rwsem) { | |
2033 | if (down_read_trylock(&mdsc->snap_rwsem) == 0) { | |
2034 | dout("inverting snap/in locks on %p\n", | |
2035 | inode); | |
be655596 | 2036 | spin_unlock(&ci->i_ceph_lock); |
a8599bd8 SW |
2037 | down_read(&mdsc->snap_rwsem); |
2038 | took_snap_rwsem = 1; | |
2039 | goto retry; | |
2040 | } | |
2041 | took_snap_rwsem = 1; | |
2042 | } | |
2043 | ||
553adfd9 | 2044 | if (cap == ci->i_auth_cap && ci->i_dirty_caps) { |
c8799fc4 | 2045 | flushing = __mark_caps_flushing(inode, session, false, |
a2971c8c YZ |
2046 | &flush_tid, |
2047 | &oldest_flush_tid); | |
553adfd9 | 2048 | } else { |
24be0c48 | 2049 | flushing = 0; |
553adfd9 | 2050 | flush_tid = 0; |
a2971c8c YZ |
2051 | spin_lock(&mdsc->cap_dirty_lock); |
2052 | oldest_flush_tid = __get_oldest_flush_tid(mdsc); | |
2053 | spin_unlock(&mdsc->cap_dirty_lock); | |
553adfd9 | 2054 | } |
a8599bd8 SW |
2055 | |
2056 | mds = cap->mds; /* remember mds, so we don't repeat */ | |
2057 | sent++; | |
2058 | ||
be655596 | 2059 | /* __send_cap drops i_ceph_lock */ |
1e4ef0c6 JL |
2060 | delayed += __send_cap(mdsc, cap, CEPH_CAP_OP_UPDATE, false, |
2061 | cap_used, want, retain, flushing, | |
2062 | flush_tid, oldest_flush_tid); | |
be655596 | 2063 | goto retry; /* retake i_ceph_lock and restart our cap scan. */ |
a8599bd8 SW |
2064 | } |
2065 | ||
0f439c74 YZ |
2066 | /* Reschedule delayed caps release if we delayed anything */ |
2067 | if (delayed) | |
a8599bd8 SW |
2068 | __cap_delay_requeue(mdsc, ci); |
2069 | ||
be655596 | 2070 | spin_unlock(&ci->i_ceph_lock); |
a8599bd8 | 2071 | |
cbd03635 | 2072 | if (queue_invalidate) |
3c6f6b79 | 2073 | ceph_queue_invalidate(inode); |
cbd03635 | 2074 | |
cdc2ce05 | 2075 | if (session) |
a8599bd8 SW |
2076 | mutex_unlock(&session->s_mutex); |
2077 | if (took_snap_rwsem) | |
2078 | up_read(&mdsc->snap_rwsem); | |
2079 | } | |
2080 | ||
a8599bd8 SW |
2081 | /* |
2082 | * Try to flush dirty caps back to the auth mds. | |
2083 | */ | |
553adfd9 | 2084 | static int try_flush_caps(struct inode *inode, u64 *ptid) |
a8599bd8 | 2085 | { |
3d14c5d2 | 2086 | struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc; |
a8599bd8 | 2087 | struct ceph_inode_info *ci = ceph_inode(inode); |
4fe59789 | 2088 | struct ceph_mds_session *session = NULL; |
89b52fe1 | 2089 | int flushing = 0; |
a2971c8c | 2090 | u64 flush_tid = 0, oldest_flush_tid = 0; |
a8599bd8 SW |
2091 | |
2092 | retry: | |
be655596 | 2093 | spin_lock(&ci->i_ceph_lock); |
e9964c10 | 2094 | if (ci->i_ceph_flags & CEPH_I_NOFLUSH) { |
6c2838fb | 2095 | spin_unlock(&ci->i_ceph_lock); |
e9964c10 SW |
2096 | dout("try_flush_caps skipping %p I_NOFLUSH set\n", inode); |
2097 | goto out; | |
2098 | } | |
a8599bd8 SW |
2099 | if (ci->i_dirty_caps && ci->i_auth_cap) { |
2100 | struct ceph_cap *cap = ci->i_auth_cap; | |
2101 | int used = __ceph_caps_used(ci); | |
2102 | int want = __ceph_caps_wanted(ci); | |
2103 | int delayed; | |
2104 | ||
4fe59789 | 2105 | if (!session || session != cap->session) { |
be655596 | 2106 | spin_unlock(&ci->i_ceph_lock); |
4fe59789 YZ |
2107 | if (session) |
2108 | mutex_unlock(&session->s_mutex); | |
a8599bd8 SW |
2109 | session = cap->session; |
2110 | mutex_lock(&session->s_mutex); | |
2111 | goto retry; | |
2112 | } | |
6c2838fb JL |
2113 | if (cap->session->s_state < CEPH_MDS_SESSION_OPEN) { |
2114 | spin_unlock(&ci->i_ceph_lock); | |
a8599bd8 | 2115 | goto out; |
6c2838fb | 2116 | } |
a8599bd8 | 2117 | |
c8799fc4 YZ |
2118 | flushing = __mark_caps_flushing(inode, session, true, |
2119 | &flush_tid, &oldest_flush_tid); | |
a8599bd8 | 2120 | |
be655596 | 2121 | /* __send_cap drops i_ceph_lock */ |
1e4ef0c6 JL |
2122 | delayed = __send_cap(mdsc, cap, CEPH_CAP_OP_FLUSH, true, |
2123 | used, want, (cap->issued | cap->implemented), | |
2124 | flushing, flush_tid, oldest_flush_tid); | |
a8599bd8 | 2125 | |
553adfd9 YZ |
2126 | if (delayed) { |
2127 | spin_lock(&ci->i_ceph_lock); | |
89b52fe1 | 2128 | __cap_delay_requeue(mdsc, ci); |
553adfd9 YZ |
2129 | spin_unlock(&ci->i_ceph_lock); |
2130 | } | |
2131 | } else { | |
e4500b5e | 2132 | if (!list_empty(&ci->i_cap_flush_list)) { |
553adfd9 | 2133 | struct ceph_cap_flush *cf = |
e4500b5e | 2134 | list_last_entry(&ci->i_cap_flush_list, |
c8799fc4 YZ |
2135 | struct ceph_cap_flush, i_list); |
2136 | cf->wake = true; | |
553adfd9 YZ |
2137 | flush_tid = cf->tid; |
2138 | } | |
2139 | flushing = ci->i_flushing_caps; | |
2140 | spin_unlock(&ci->i_ceph_lock); | |
a8599bd8 SW |
2141 | } |
2142 | out: | |
4fe59789 | 2143 | if (session) |
a8599bd8 | 2144 | mutex_unlock(&session->s_mutex); |
553adfd9 YZ |
2145 | |
2146 | *ptid = flush_tid; | |
a8599bd8 SW |
2147 | return flushing; |
2148 | } | |
2149 | ||
2150 | /* | |
2151 | * Return true if we've flushed caps through the given flush_tid. | |
2152 | */ | |
553adfd9 | 2153 | static int caps_are_flushed(struct inode *inode, u64 flush_tid) |
a8599bd8 SW |
2154 | { |
2155 | struct ceph_inode_info *ci = ceph_inode(inode); | |
553adfd9 | 2156 | int ret = 1; |
a8599bd8 | 2157 | |
be655596 | 2158 | spin_lock(&ci->i_ceph_lock); |
e4500b5e YZ |
2159 | if (!list_empty(&ci->i_cap_flush_list)) { |
2160 | struct ceph_cap_flush * cf = | |
2161 | list_first_entry(&ci->i_cap_flush_list, | |
2162 | struct ceph_cap_flush, i_list); | |
553adfd9 | 2163 | if (cf->tid <= flush_tid) |
a8599bd8 | 2164 | ret = 0; |
89b52fe1 | 2165 | } |
be655596 | 2166 | spin_unlock(&ci->i_ceph_lock); |
a8599bd8 SW |
2167 | return ret; |
2168 | } | |
2169 | ||
da819c81 | 2170 | /* |
68cd5b4b | 2171 | * wait for any unsafe requests to complete. |
da819c81 | 2172 | */ |
68cd5b4b | 2173 | static int unsafe_request_wait(struct inode *inode) |
da819c81 YZ |
2174 | { |
2175 | struct ceph_inode_info *ci = ceph_inode(inode); | |
68cd5b4b YZ |
2176 | struct ceph_mds_request *req1 = NULL, *req2 = NULL; |
2177 | int ret, err = 0; | |
da819c81 YZ |
2178 | |
2179 | spin_lock(&ci->i_unsafe_lock); | |
68cd5b4b YZ |
2180 | if (S_ISDIR(inode->i_mode) && !list_empty(&ci->i_unsafe_dirops)) { |
2181 | req1 = list_last_entry(&ci->i_unsafe_dirops, | |
2182 | struct ceph_mds_request, | |
2183 | r_unsafe_dir_item); | |
2184 | ceph_mdsc_get_request(req1); | |
2185 | } | |
2186 | if (!list_empty(&ci->i_unsafe_iops)) { | |
2187 | req2 = list_last_entry(&ci->i_unsafe_iops, | |
2188 | struct ceph_mds_request, | |
2189 | r_unsafe_target_item); | |
2190 | ceph_mdsc_get_request(req2); | |
2191 | } | |
2192 | spin_unlock(&ci->i_unsafe_lock); | |
da819c81 | 2193 | |
4945a084 | 2194 | dout("unsafe_request_wait %p wait on tid %llu %llu\n", |
68cd5b4b YZ |
2195 | inode, req1 ? req1->r_tid : 0ULL, req2 ? req2->r_tid : 0ULL); |
2196 | if (req1) { | |
2197 | ret = !wait_for_completion_timeout(&req1->r_safe_completion, | |
2198 | ceph_timeout_jiffies(req1->r_timeout)); | |
da819c81 | 2199 | if (ret) |
68cd5b4b YZ |
2200 | err = -EIO; |
2201 | ceph_mdsc_put_request(req1); | |
2202 | } | |
2203 | if (req2) { | |
2204 | ret = !wait_for_completion_timeout(&req2->r_safe_completion, | |
2205 | ceph_timeout_jiffies(req2->r_timeout)); | |
2206 | if (ret) | |
2207 | err = -EIO; | |
2208 | ceph_mdsc_put_request(req2); | |
2209 | } | |
2210 | return err; | |
da819c81 YZ |
2211 | } |
2212 | ||
02c24a82 | 2213 | int ceph_fsync(struct file *file, loff_t start, loff_t end, int datasync) |
a8599bd8 | 2214 | { |
7ea80859 | 2215 | struct inode *inode = file->f_mapping->host; |
a8599bd8 | 2216 | struct ceph_inode_info *ci = ceph_inode(inode); |
553adfd9 | 2217 | u64 flush_tid; |
a8599bd8 SW |
2218 | int ret; |
2219 | int dirty; | |
2220 | ||
2221 | dout("fsync %p%s\n", inode, datasync ? " datasync" : ""); | |
9a5530c6 | 2222 | |
b74fceae | 2223 | ret = file_write_and_wait_range(file, start, end); |
a8599bd8 | 2224 | if (ret < 0) |
da819c81 YZ |
2225 | goto out; |
2226 | ||
2227 | if (datasync) | |
2228 | goto out; | |
2229 | ||
5955102c | 2230 | inode_lock(inode); |
a8599bd8 | 2231 | |
553adfd9 | 2232 | dirty = try_flush_caps(inode, &flush_tid); |
a8599bd8 SW |
2233 | dout("fsync dirty caps are %s\n", ceph_cap_string(dirty)); |
2234 | ||
68cd5b4b | 2235 | ret = unsafe_request_wait(inode); |
da819c81 | 2236 | |
a8599bd8 SW |
2237 | /* |
2238 | * only wait on non-file metadata writeback (the mds | |
2239 | * can recover size and mtime, so we don't need to | |
2240 | * wait for that) | |
2241 | */ | |
da819c81 | 2242 | if (!ret && (dirty & ~CEPH_CAP_ANY_FILE_WR)) { |
a8599bd8 | 2243 | ret = wait_event_interruptible(ci->i_cap_wq, |
da819c81 | 2244 | caps_are_flushed(inode, flush_tid)); |
a8599bd8 | 2245 | } |
5955102c | 2246 | inode_unlock(inode); |
da819c81 YZ |
2247 | out: |
2248 | dout("fsync %p%s result=%d\n", inode, datasync ? " datasync" : "", ret); | |
a8599bd8 SW |
2249 | return ret; |
2250 | } | |
2251 | ||
2252 | /* | |
2253 | * Flush any dirty caps back to the mds. If we aren't asked to wait, | |
2254 | * queue inode for flush but don't do so immediately, because we can | |
2255 | * get by with fewer MDS messages if we wait for data writeback to | |
2256 | * complete first. | |
2257 | */ | |
f1a3d572 | 2258 | int ceph_write_inode(struct inode *inode, struct writeback_control *wbc) |
a8599bd8 SW |
2259 | { |
2260 | struct ceph_inode_info *ci = ceph_inode(inode); | |
553adfd9 | 2261 | u64 flush_tid; |
a8599bd8 SW |
2262 | int err = 0; |
2263 | int dirty; | |
16515a6d | 2264 | int wait = (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync); |
a8599bd8 SW |
2265 | |
2266 | dout("write_inode %p wait=%d\n", inode, wait); | |
2267 | if (wait) { | |
553adfd9 | 2268 | dirty = try_flush_caps(inode, &flush_tid); |
a8599bd8 SW |
2269 | if (dirty) |
2270 | err = wait_event_interruptible(ci->i_cap_wq, | |
2271 | caps_are_flushed(inode, flush_tid)); | |
2272 | } else { | |
640ef79d | 2273 | struct ceph_mds_client *mdsc = |
3d14c5d2 | 2274 | ceph_sb_to_client(inode->i_sb)->mdsc; |
a8599bd8 | 2275 | |
be655596 | 2276 | spin_lock(&ci->i_ceph_lock); |
a8599bd8 SW |
2277 | if (__ceph_caps_dirty(ci)) |
2278 | __cap_delay_requeue_front(mdsc, ci); | |
be655596 | 2279 | spin_unlock(&ci->i_ceph_lock); |
a8599bd8 SW |
2280 | } |
2281 | return err; | |
2282 | } | |
2283 | ||
0e294387 YZ |
2284 | static void __kick_flushing_caps(struct ceph_mds_client *mdsc, |
2285 | struct ceph_mds_session *session, | |
2286 | struct ceph_inode_info *ci, | |
2287 | u64 oldest_flush_tid) | |
2288 | __releases(ci->i_ceph_lock) | |
2289 | __acquires(ci->i_ceph_lock) | |
553adfd9 YZ |
2290 | { |
2291 | struct inode *inode = &ci->vfs_inode; | |
2292 | struct ceph_cap *cap; | |
2293 | struct ceph_cap_flush *cf; | |
0e294387 | 2294 | int ret; |
553adfd9 YZ |
2295 | u64 first_tid = 0; |
2296 | ||
e4500b5e YZ |
2297 | list_for_each_entry(cf, &ci->i_cap_flush_list, i_list) { |
2298 | if (cf->tid < first_tid) | |
2299 | continue; | |
2300 | ||
553adfd9 YZ |
2301 | cap = ci->i_auth_cap; |
2302 | if (!(cap && cap->session == session)) { | |
0e294387 YZ |
2303 | pr_err("%p auth cap %p not mds%d ???\n", |
2304 | inode, cap, session->s_mds); | |
553adfd9 YZ |
2305 | break; |
2306 | } | |
2307 | ||
553adfd9 YZ |
2308 | first_tid = cf->tid + 1; |
2309 | ||
0e294387 YZ |
2310 | if (cf->caps) { |
2311 | dout("kick_flushing_caps %p cap %p tid %llu %s\n", | |
2312 | inode, cap, cf->tid, ceph_cap_string(cf->caps)); | |
2313 | ci->i_ceph_flags |= CEPH_I_NODELAY; | |
2314 | ret = __send_cap(mdsc, cap, CEPH_CAP_OP_FLUSH, | |
1e4ef0c6 | 2315 | false, __ceph_caps_used(ci), |
0e294387 YZ |
2316 | __ceph_caps_wanted(ci), |
2317 | cap->issued | cap->implemented, | |
2318 | cf->caps, cf->tid, oldest_flush_tid); | |
2319 | if (ret) { | |
2320 | pr_err("kick_flushing_caps: error sending " | |
2321 | "cap flush, ino (%llx.%llx) " | |
2322 | "tid %llu flushing %s\n", | |
2323 | ceph_vinop(inode), cf->tid, | |
2324 | ceph_cap_string(cf->caps)); | |
2325 | } | |
2326 | } else { | |
2327 | struct ceph_cap_snap *capsnap = | |
2328 | container_of(cf, struct ceph_cap_snap, | |
2329 | cap_flush); | |
2330 | dout("kick_flushing_caps %p capsnap %p tid %llu %s\n", | |
2331 | inode, capsnap, cf->tid, | |
2332 | ceph_cap_string(capsnap->dirty)); | |
2333 | ||
805692d0 | 2334 | refcount_inc(&capsnap->nref); |
0e294387 YZ |
2335 | spin_unlock(&ci->i_ceph_lock); |
2336 | ||
2337 | ret = __send_flush_snap(inode, session, capsnap, cap->mseq, | |
2338 | oldest_flush_tid); | |
2339 | if (ret < 0) { | |
2340 | pr_err("kick_flushing_caps: error sending " | |
2341 | "cap flushsnap, ino (%llx.%llx) " | |
2342 | "tid %llu follows %llu\n", | |
2343 | ceph_vinop(inode), cf->tid, | |
2344 | capsnap->follows); | |
2345 | } | |
2346 | ||
2347 | ceph_put_cap_snap(capsnap); | |
2348 | } | |
e4500b5e YZ |
2349 | |
2350 | spin_lock(&ci->i_ceph_lock); | |
553adfd9 | 2351 | } |
553adfd9 YZ |
2352 | } |
2353 | ||
e548e9b9 YZ |
2354 | void ceph_early_kick_flushing_caps(struct ceph_mds_client *mdsc, |
2355 | struct ceph_mds_session *session) | |
2356 | { | |
2357 | struct ceph_inode_info *ci; | |
2358 | struct ceph_cap *cap; | |
0e294387 | 2359 | u64 oldest_flush_tid; |
e548e9b9 YZ |
2360 | |
2361 | dout("early_kick_flushing_caps mds%d\n", session->s_mds); | |
0e294387 YZ |
2362 | |
2363 | spin_lock(&mdsc->cap_dirty_lock); | |
2364 | oldest_flush_tid = __get_oldest_flush_tid(mdsc); | |
2365 | spin_unlock(&mdsc->cap_dirty_lock); | |
2366 | ||
e548e9b9 YZ |
2367 | list_for_each_entry(ci, &session->s_cap_flushing, i_flushing_item) { |
2368 | spin_lock(&ci->i_ceph_lock); | |
2369 | cap = ci->i_auth_cap; | |
2370 | if (!(cap && cap->session == session)) { | |
2371 | pr_err("%p auth cap %p not mds%d ???\n", | |
2372 | &ci->vfs_inode, cap, session->s_mds); | |
2373 | spin_unlock(&ci->i_ceph_lock); | |
2374 | continue; | |
2375 | } | |
2376 | ||
2377 | ||
2378 | /* | |
2379 | * if flushing caps were revoked, we re-send the cap flush | |
2380 | * in client reconnect stage. This guarantees MDS * processes | |
2381 | * the cap flush message before issuing the flushing caps to | |
2382 | * other client. | |
2383 | */ | |
2384 | if ((cap->issued & ci->i_flushing_caps) != | |
2385 | ci->i_flushing_caps) { | |
13c2b57d | 2386 | ci->i_ceph_flags &= ~CEPH_I_KICK_FLUSH; |
0e294387 YZ |
2387 | __kick_flushing_caps(mdsc, session, ci, |
2388 | oldest_flush_tid); | |
13c2b57d YZ |
2389 | } else { |
2390 | ci->i_ceph_flags |= CEPH_I_KICK_FLUSH; | |
e548e9b9 YZ |
2391 | } |
2392 | ||
e548e9b9 YZ |
2393 | spin_unlock(&ci->i_ceph_lock); |
2394 | } | |
2395 | } | |
2396 | ||
a8599bd8 SW |
2397 | void ceph_kick_flushing_caps(struct ceph_mds_client *mdsc, |
2398 | struct ceph_mds_session *session) | |
2399 | { | |
2400 | struct ceph_inode_info *ci; | |
13c2b57d | 2401 | struct ceph_cap *cap; |
0e294387 | 2402 | u64 oldest_flush_tid; |
a8599bd8 SW |
2403 | |
2404 | dout("kick_flushing_caps mds%d\n", session->s_mds); | |
0e294387 YZ |
2405 | |
2406 | spin_lock(&mdsc->cap_dirty_lock); | |
2407 | oldest_flush_tid = __get_oldest_flush_tid(mdsc); | |
2408 | spin_unlock(&mdsc->cap_dirty_lock); | |
2409 | ||
a8599bd8 | 2410 | list_for_each_entry(ci, &session->s_cap_flushing, i_flushing_item) { |
0e294387 | 2411 | spin_lock(&ci->i_ceph_lock); |
13c2b57d YZ |
2412 | cap = ci->i_auth_cap; |
2413 | if (!(cap && cap->session == session)) { | |
2414 | pr_err("%p auth cap %p not mds%d ???\n", | |
2415 | &ci->vfs_inode, cap, session->s_mds); | |
2416 | spin_unlock(&ci->i_ceph_lock); | |
2417 | continue; | |
2418 | } | |
2419 | if (ci->i_ceph_flags & CEPH_I_KICK_FLUSH) { | |
2420 | ci->i_ceph_flags &= ~CEPH_I_KICK_FLUSH; | |
2421 | __kick_flushing_caps(mdsc, session, ci, | |
2422 | oldest_flush_tid); | |
2423 | } | |
0e294387 | 2424 | spin_unlock(&ci->i_ceph_lock); |
a8599bd8 SW |
2425 | } |
2426 | } | |
2427 | ||
088b3f5e SW |
2428 | static void kick_flushing_inode_caps(struct ceph_mds_client *mdsc, |
2429 | struct ceph_mds_session *session, | |
2430 | struct inode *inode) | |
0e294387 | 2431 | __releases(ci->i_ceph_lock) |
088b3f5e SW |
2432 | { |
2433 | struct ceph_inode_info *ci = ceph_inode(inode); | |
2434 | struct ceph_cap *cap; | |
088b3f5e | 2435 | |
088b3f5e | 2436 | cap = ci->i_auth_cap; |
8310b089 YZ |
2437 | dout("kick_flushing_inode_caps %p flushing %s\n", inode, |
2438 | ceph_cap_string(ci->i_flushing_caps)); | |
005c4697 | 2439 | |
0e294387 YZ |
2440 | if (!list_empty(&ci->i_cap_flush_list)) { |
2441 | u64 oldest_flush_tid; | |
005c4697 YZ |
2442 | spin_lock(&mdsc->cap_dirty_lock); |
2443 | list_move_tail(&ci->i_flushing_item, | |
2444 | &cap->session->s_cap_flushing); | |
0e294387 | 2445 | oldest_flush_tid = __get_oldest_flush_tid(mdsc); |
005c4697 YZ |
2446 | spin_unlock(&mdsc->cap_dirty_lock); |
2447 | ||
13c2b57d | 2448 | ci->i_ceph_flags &= ~CEPH_I_KICK_FLUSH; |
0e294387 | 2449 | __kick_flushing_caps(mdsc, session, ci, oldest_flush_tid); |
553adfd9 | 2450 | spin_unlock(&ci->i_ceph_lock); |
088b3f5e | 2451 | } else { |
be655596 | 2452 | spin_unlock(&ci->i_ceph_lock); |
088b3f5e SW |
2453 | } |
2454 | } | |
2455 | ||
a8599bd8 SW |
2456 | |
2457 | /* | |
2458 | * Take references to capabilities we hold, so that we don't release | |
2459 | * them to the MDS prematurely. | |
2460 | * | |
be655596 | 2461 | * Protected by i_ceph_lock. |
a8599bd8 | 2462 | */ |
5dda377c YZ |
2463 | static void __take_cap_refs(struct ceph_inode_info *ci, int got, |
2464 | bool snap_rwsem_locked) | |
a8599bd8 SW |
2465 | { |
2466 | if (got & CEPH_CAP_PIN) | |
2467 | ci->i_pin_ref++; | |
2468 | if (got & CEPH_CAP_FILE_RD) | |
2469 | ci->i_rd_ref++; | |
2470 | if (got & CEPH_CAP_FILE_CACHE) | |
2471 | ci->i_rdcache_ref++; | |
5dda377c YZ |
2472 | if (got & CEPH_CAP_FILE_WR) { |
2473 | if (ci->i_wr_ref == 0 && !ci->i_head_snapc) { | |
2474 | BUG_ON(!snap_rwsem_locked); | |
2475 | ci->i_head_snapc = ceph_get_snap_context( | |
2476 | ci->i_snap_realm->cached_context); | |
2477 | } | |
a8599bd8 | 2478 | ci->i_wr_ref++; |
5dda377c | 2479 | } |
a8599bd8 | 2480 | if (got & CEPH_CAP_FILE_BUFFER) { |
d3d0720d | 2481 | if (ci->i_wb_ref == 0) |
3772d26d | 2482 | ihold(&ci->vfs_inode); |
d3d0720d HC |
2483 | ci->i_wb_ref++; |
2484 | dout("__take_cap_refs %p wb %d -> %d (?)\n", | |
2485 | &ci->vfs_inode, ci->i_wb_ref-1, ci->i_wb_ref); | |
a8599bd8 SW |
2486 | } |
2487 | } | |
2488 | ||
2489 | /* | |
2490 | * Try to grab cap references. Specify those refs we @want, and the | |
2491 | * minimal set we @need. Also include the larger offset we are writing | |
2492 | * to (when applicable), and check against max_size here as well. | |
2493 | * Note that caller is responsible for ensuring max_size increases are | |
2494 | * requested from the MDS. | |
2495 | */ | |
2496 | static int try_get_cap_refs(struct ceph_inode_info *ci, int need, int want, | |
5dda377c | 2497 | loff_t endoff, bool nonblock, int *got, int *err) |
a8599bd8 SW |
2498 | { |
2499 | struct inode *inode = &ci->vfs_inode; | |
5dda377c | 2500 | struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc; |
a8599bd8 | 2501 | int ret = 0; |
c4d4a582 | 2502 | int have, implemented; |
195d3ce2 | 2503 | int file_wanted; |
5dda377c | 2504 | bool snap_rwsem_locked = false; |
a8599bd8 SW |
2505 | |
2506 | dout("get_cap_refs %p need %s want %s\n", inode, | |
2507 | ceph_cap_string(need), ceph_cap_string(want)); | |
c4d4a582 | 2508 | |
5dda377c | 2509 | again: |
be655596 | 2510 | spin_lock(&ci->i_ceph_lock); |
a8599bd8 | 2511 | |
195d3ce2 SW |
2512 | /* make sure file is actually open */ |
2513 | file_wanted = __ceph_caps_file_wanted(ci); | |
77310320 | 2514 | if ((file_wanted & need) != need) { |
195d3ce2 SW |
2515 | dout("try_get_cap_refs need %s file_wanted %s, EBADF\n", |
2516 | ceph_cap_string(need), ceph_cap_string(file_wanted)); | |
a8599bd8 SW |
2517 | *err = -EBADF; |
2518 | ret = 1; | |
3738daa6 | 2519 | goto out_unlock; |
a8599bd8 SW |
2520 | } |
2521 | ||
37505d57 YZ |
2522 | /* finish pending truncate */ |
2523 | while (ci->i_truncate_pending) { | |
2524 | spin_unlock(&ci->i_ceph_lock); | |
5dda377c YZ |
2525 | if (snap_rwsem_locked) { |
2526 | up_read(&mdsc->snap_rwsem); | |
2527 | snap_rwsem_locked = false; | |
2528 | } | |
b415bf4f | 2529 | __ceph_do_pending_vmtruncate(inode); |
37505d57 YZ |
2530 | spin_lock(&ci->i_ceph_lock); |
2531 | } | |
2532 | ||
3871cbb9 YZ |
2533 | have = __ceph_caps_issued(ci, &implemented); |
2534 | ||
2535 | if (have & need & CEPH_CAP_FILE_WR) { | |
a8599bd8 SW |
2536 | if (endoff >= 0 && endoff > (loff_t)ci->i_max_size) { |
2537 | dout("get_cap_refs %p endoff %llu > maxsize %llu\n", | |
2538 | inode, endoff, ci->i_max_size); | |
3871cbb9 | 2539 | if (endoff > ci->i_requested_max_size) { |
5dda377c | 2540 | *err = -EAGAIN; |
a8599bd8 SW |
2541 | ret = 1; |
2542 | } | |
3738daa6 | 2543 | goto out_unlock; |
a8599bd8 SW |
2544 | } |
2545 | /* | |
2546 | * If a sync write is in progress, we must wait, so that we | |
2547 | * can get a final snapshot value for size+mtime. | |
2548 | */ | |
2549 | if (__ceph_have_pending_cap_snap(ci)) { | |
2550 | dout("get_cap_refs %p cap_snap_pending\n", inode); | |
3738daa6 | 2551 | goto out_unlock; |
a8599bd8 SW |
2552 | } |
2553 | } | |
a8599bd8 | 2554 | |
a8599bd8 SW |
2555 | if ((have & need) == need) { |
2556 | /* | |
2557 | * Look at (implemented & ~have & not) so that we keep waiting | |
2558 | * on transition from wanted -> needed caps. This is needed | |
2559 | * for WRBUFFER|WR -> WR to avoid a new WR sync write from | |
2560 | * going before a prior buffered writeback happens. | |
2561 | */ | |
2562 | int not = want & ~(have & need); | |
2563 | int revoking = implemented & ~have; | |
2564 | dout("get_cap_refs %p have %s but not %s (revoking %s)\n", | |
2565 | inode, ceph_cap_string(have), ceph_cap_string(not), | |
2566 | ceph_cap_string(revoking)); | |
2567 | if ((revoking & not) == 0) { | |
5dda377c YZ |
2568 | if (!snap_rwsem_locked && |
2569 | !ci->i_head_snapc && | |
2570 | (need & CEPH_CAP_FILE_WR)) { | |
2571 | if (!down_read_trylock(&mdsc->snap_rwsem)) { | |
2572 | /* | |
2573 | * we can not call down_read() when | |
2574 | * task isn't in TASK_RUNNING state | |
2575 | */ | |
2576 | if (nonblock) { | |
2577 | *err = -EAGAIN; | |
2578 | ret = 1; | |
2579 | goto out_unlock; | |
2580 | } | |
2581 | ||
2582 | spin_unlock(&ci->i_ceph_lock); | |
2583 | down_read(&mdsc->snap_rwsem); | |
2584 | snap_rwsem_locked = true; | |
2585 | goto again; | |
2586 | } | |
2587 | snap_rwsem_locked = true; | |
2588 | } | |
c4d4a582 | 2589 | *got = need | (have & want); |
f7f7e7a0 YZ |
2590 | if ((need & CEPH_CAP_FILE_RD) && |
2591 | !(*got & CEPH_CAP_FILE_CACHE)) | |
2592 | ceph_disable_fscache_readpage(ci); | |
5dda377c | 2593 | __take_cap_refs(ci, *got, true); |
a8599bd8 SW |
2594 | ret = 1; |
2595 | } | |
2596 | } else { | |
03f4fcb0 YZ |
2597 | int session_readonly = false; |
2598 | if ((need & CEPH_CAP_FILE_WR) && ci->i_auth_cap) { | |
2599 | struct ceph_mds_session *s = ci->i_auth_cap->session; | |
2600 | spin_lock(&s->s_cap_lock); | |
2601 | session_readonly = s->s_readonly; | |
2602 | spin_unlock(&s->s_cap_lock); | |
2603 | } | |
2604 | if (session_readonly) { | |
2605 | dout("get_cap_refs %p needed %s but mds%d readonly\n", | |
2606 | inode, ceph_cap_string(need), ci->i_auth_cap->mds); | |
2607 | *err = -EROFS; | |
2608 | ret = 1; | |
2609 | goto out_unlock; | |
2610 | } | |
2611 | ||
77310320 YZ |
2612 | if (ci->i_ceph_flags & CEPH_I_CAP_DROPPED) { |
2613 | int mds_wanted; | |
52953d55 | 2614 | if (READ_ONCE(mdsc->fsc->mount_state) == |
77310320 YZ |
2615 | CEPH_MOUNT_SHUTDOWN) { |
2616 | dout("get_cap_refs %p forced umount\n", inode); | |
2617 | *err = -EIO; | |
2618 | ret = 1; | |
2619 | goto out_unlock; | |
2620 | } | |
c1944fed | 2621 | mds_wanted = __ceph_caps_mds_wanted(ci, false); |
eb65b919 | 2622 | if (need & ~(mds_wanted & need)) { |
77310320 YZ |
2623 | dout("get_cap_refs %p caps were dropped" |
2624 | " (session killed?)\n", inode); | |
2625 | *err = -ESTALE; | |
2626 | ret = 1; | |
2627 | goto out_unlock; | |
2628 | } | |
eb65b919 | 2629 | if (!(file_wanted & ~mds_wanted)) |
77310320 | 2630 | ci->i_ceph_flags &= ~CEPH_I_CAP_DROPPED; |
48fec5d0 YZ |
2631 | } |
2632 | ||
a8599bd8 SW |
2633 | dout("get_cap_refs %p have %s needed %s\n", inode, |
2634 | ceph_cap_string(have), ceph_cap_string(need)); | |
2635 | } | |
3738daa6 | 2636 | out_unlock: |
be655596 | 2637 | spin_unlock(&ci->i_ceph_lock); |
5dda377c YZ |
2638 | if (snap_rwsem_locked) |
2639 | up_read(&mdsc->snap_rwsem); | |
3738daa6 | 2640 | |
a8599bd8 | 2641 | dout("get_cap_refs %p ret %d got %s\n", inode, |
c4d4a582 | 2642 | ret, ceph_cap_string(*got)); |
a8599bd8 SW |
2643 | return ret; |
2644 | } | |
2645 | ||
2646 | /* | |
2647 | * Check the offset we are writing up to against our current | |
2648 | * max_size. If necessary, tell the MDS we want to write to | |
2649 | * a larger offset. | |
2650 | */ | |
2651 | static void check_max_size(struct inode *inode, loff_t endoff) | |
2652 | { | |
2653 | struct ceph_inode_info *ci = ceph_inode(inode); | |
2654 | int check = 0; | |
2655 | ||
2656 | /* do we need to explicitly request a larger max_size? */ | |
be655596 | 2657 | spin_lock(&ci->i_ceph_lock); |
3871cbb9 | 2658 | if (endoff >= ci->i_max_size && endoff > ci->i_wanted_max_size) { |
a8599bd8 SW |
2659 | dout("write %p at large endoff %llu, req max_size\n", |
2660 | inode, endoff); | |
2661 | ci->i_wanted_max_size = endoff; | |
a8599bd8 | 2662 | } |
3871cbb9 YZ |
2663 | /* duplicate ceph_check_caps()'s logic */ |
2664 | if (ci->i_auth_cap && | |
2665 | (ci->i_auth_cap->issued & CEPH_CAP_FILE_WR) && | |
2666 | ci->i_wanted_max_size > ci->i_max_size && | |
2667 | ci->i_wanted_max_size > ci->i_requested_max_size) | |
2668 | check = 1; | |
be655596 | 2669 | spin_unlock(&ci->i_ceph_lock); |
a8599bd8 SW |
2670 | if (check) |
2671 | ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL); | |
2672 | } | |
2673 | ||
2b1ac852 YZ |
2674 | int ceph_try_get_caps(struct ceph_inode_info *ci, int need, int want, int *got) |
2675 | { | |
2676 | int ret, err = 0; | |
2677 | ||
2678 | BUG_ON(need & ~CEPH_CAP_FILE_RD); | |
2679 | BUG_ON(want & ~(CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)); | |
2680 | ret = ceph_pool_perm_check(ci, need); | |
2681 | if (ret < 0) | |
2682 | return ret; | |
2683 | ||
2684 | ret = try_get_cap_refs(ci, need, want, 0, true, got, &err); | |
2685 | if (ret) { | |
2686 | if (err == -EAGAIN) { | |
2687 | ret = 0; | |
2688 | } else if (err < 0) { | |
2689 | ret = err; | |
2690 | } | |
2691 | } | |
2692 | return ret; | |
2693 | } | |
2694 | ||
a8599bd8 SW |
2695 | /* |
2696 | * Wait for caps, and take cap references. If we can't get a WR cap | |
2697 | * due to a small max_size, make sure we check_max_size (and possibly | |
2698 | * ask the mds) so we don't get hung up indefinitely. | |
2699 | */ | |
3738daa6 YZ |
2700 | int ceph_get_caps(struct ceph_inode_info *ci, int need, int want, |
2701 | loff_t endoff, int *got, struct page **pinned_page) | |
a8599bd8 | 2702 | { |
5dda377c | 2703 | int _got, ret, err = 0; |
a8599bd8 | 2704 | |
10183a69 YZ |
2705 | ret = ceph_pool_perm_check(ci, need); |
2706 | if (ret < 0) | |
2707 | return ret; | |
2708 | ||
5dda377c YZ |
2709 | while (true) { |
2710 | if (endoff > 0) | |
2711 | check_max_size(&ci->vfs_inode, endoff); | |
c4d4a582 | 2712 | |
5dda377c YZ |
2713 | err = 0; |
2714 | _got = 0; | |
2715 | ret = try_get_cap_refs(ci, need, want, endoff, | |
2716 | false, &_got, &err); | |
2717 | if (ret) { | |
2718 | if (err == -EAGAIN) | |
2719 | continue; | |
2720 | if (err < 0) | |
77310320 | 2721 | ret = err; |
5dda377c | 2722 | } else { |
5c341ee3 NB |
2723 | DEFINE_WAIT_FUNC(wait, woken_wake_function); |
2724 | add_wait_queue(&ci->i_cap_wq, &wait); | |
2725 | ||
2726 | while (!try_get_cap_refs(ci, need, want, endoff, | |
6e09d0fb YZ |
2727 | true, &_got, &err)) { |
2728 | if (signal_pending(current)) { | |
2729 | ret = -ERESTARTSYS; | |
2730 | break; | |
2731 | } | |
5c341ee3 | 2732 | wait_woken(&wait, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
6e09d0fb | 2733 | } |
5c341ee3 NB |
2734 | |
2735 | remove_wait_queue(&ci->i_cap_wq, &wait); | |
2736 | ||
5dda377c YZ |
2737 | if (err == -EAGAIN) |
2738 | continue; | |
2739 | if (err < 0) | |
2740 | ret = err; | |
77310320 YZ |
2741 | } |
2742 | if (ret < 0) { | |
2743 | if (err == -ESTALE) { | |
2744 | /* session was killed, try renew caps */ | |
2745 | ret = ceph_renew_caps(&ci->vfs_inode); | |
2746 | if (ret == 0) | |
2747 | continue; | |
2748 | } | |
2749 | return ret; | |
5dda377c | 2750 | } |
c4d4a582 | 2751 | |
5dda377c YZ |
2752 | if (ci->i_inline_version != CEPH_INLINE_NONE && |
2753 | (_got & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) && | |
2754 | i_size_read(&ci->vfs_inode) > 0) { | |
2755 | struct page *page = | |
2756 | find_get_page(ci->vfs_inode.i_mapping, 0); | |
2757 | if (page) { | |
2758 | if (PageUptodate(page)) { | |
2759 | *pinned_page = page; | |
2760 | break; | |
2761 | } | |
09cbfeaf | 2762 | put_page(page); |
c4d4a582 | 2763 | } |
5dda377c YZ |
2764 | /* |
2765 | * drop cap refs first because getattr while | |
2766 | * holding * caps refs can cause deadlock. | |
2767 | */ | |
2768 | ceph_put_cap_refs(ci, _got); | |
2769 | _got = 0; | |
c4d4a582 | 2770 | |
5dda377c YZ |
2771 | /* |
2772 | * getattr request will bring inline data into | |
2773 | * page cache | |
2774 | */ | |
2775 | ret = __ceph_do_getattr(&ci->vfs_inode, NULL, | |
2776 | CEPH_STAT_CAP_INLINE_DATA, | |
2777 | true); | |
2778 | if (ret < 0) | |
2779 | return ret; | |
2780 | continue; | |
2781 | } | |
2782 | break; | |
c4d4a582 | 2783 | } |
5dda377c | 2784 | |
f7f7e7a0 YZ |
2785 | if ((_got & CEPH_CAP_FILE_RD) && (_got & CEPH_CAP_FILE_CACHE)) |
2786 | ceph_fscache_revalidate_cookie(ci); | |
2787 | ||
c4d4a582 YZ |
2788 | *got = _got; |
2789 | return 0; | |
a8599bd8 SW |
2790 | } |
2791 | ||
2792 | /* | |
2793 | * Take cap refs. Caller must already know we hold at least one ref | |
2794 | * on the caps in question or we don't know this is safe. | |
2795 | */ | |
2796 | void ceph_get_cap_refs(struct ceph_inode_info *ci, int caps) | |
2797 | { | |
be655596 | 2798 | spin_lock(&ci->i_ceph_lock); |
5dda377c | 2799 | __take_cap_refs(ci, caps, false); |
be655596 | 2800 | spin_unlock(&ci->i_ceph_lock); |
a8599bd8 SW |
2801 | } |
2802 | ||
86056090 YZ |
2803 | |
2804 | /* | |
2805 | * drop cap_snap that is not associated with any snapshot. | |
2806 | * we don't need to send FLUSHSNAP message for it. | |
2807 | */ | |
70220ac8 YZ |
2808 | static int ceph_try_drop_cap_snap(struct ceph_inode_info *ci, |
2809 | struct ceph_cap_snap *capsnap) | |
86056090 YZ |
2810 | { |
2811 | if (!capsnap->need_flush && | |
2812 | !capsnap->writing && !capsnap->dirty_pages) { | |
86056090 YZ |
2813 | dout("dropping cap_snap %p follows %llu\n", |
2814 | capsnap, capsnap->follows); | |
0e294387 | 2815 | BUG_ON(capsnap->cap_flush.tid > 0); |
86056090 | 2816 | ceph_put_snap_context(capsnap->context); |
70220ac8 YZ |
2817 | if (!list_is_last(&capsnap->ci_item, &ci->i_cap_snaps)) |
2818 | ci->i_ceph_flags |= CEPH_I_FLUSH_SNAPS; | |
2819 | ||
86056090 | 2820 | list_del(&capsnap->ci_item); |
86056090 YZ |
2821 | ceph_put_cap_snap(capsnap); |
2822 | return 1; | |
2823 | } | |
2824 | return 0; | |
2825 | } | |
2826 | ||
a8599bd8 SW |
2827 | /* |
2828 | * Release cap refs. | |
2829 | * | |
2830 | * If we released the last ref on any given cap, call ceph_check_caps | |
2831 | * to release (or schedule a release). | |
2832 | * | |
2833 | * If we are releasing a WR cap (from a sync write), finalize any affected | |
2834 | * cap_snap, and wake up any waiters. | |
2835 | */ | |
2836 | void ceph_put_cap_refs(struct ceph_inode_info *ci, int had) | |
2837 | { | |
2838 | struct inode *inode = &ci->vfs_inode; | |
2839 | int last = 0, put = 0, flushsnaps = 0, wake = 0; | |
a8599bd8 | 2840 | |
be655596 | 2841 | spin_lock(&ci->i_ceph_lock); |
a8599bd8 SW |
2842 | if (had & CEPH_CAP_PIN) |
2843 | --ci->i_pin_ref; | |
2844 | if (had & CEPH_CAP_FILE_RD) | |
2845 | if (--ci->i_rd_ref == 0) | |
2846 | last++; | |
2847 | if (had & CEPH_CAP_FILE_CACHE) | |
2848 | if (--ci->i_rdcache_ref == 0) | |
2849 | last++; | |
2850 | if (had & CEPH_CAP_FILE_BUFFER) { | |
d3d0720d | 2851 | if (--ci->i_wb_ref == 0) { |
a8599bd8 SW |
2852 | last++; |
2853 | put++; | |
2854 | } | |
d3d0720d HC |
2855 | dout("put_cap_refs %p wb %d -> %d (?)\n", |
2856 | inode, ci->i_wb_ref+1, ci->i_wb_ref); | |
a8599bd8 SW |
2857 | } |
2858 | if (had & CEPH_CAP_FILE_WR) | |
2859 | if (--ci->i_wr_ref == 0) { | |
2860 | last++; | |
86056090 YZ |
2861 | if (__ceph_have_pending_cap_snap(ci)) { |
2862 | struct ceph_cap_snap *capsnap = | |
2863 | list_last_entry(&ci->i_cap_snaps, | |
2864 | struct ceph_cap_snap, | |
2865 | ci_item); | |
2866 | capsnap->writing = 0; | |
70220ac8 | 2867 | if (ceph_try_drop_cap_snap(ci, capsnap)) |
86056090 YZ |
2868 | put++; |
2869 | else if (__ceph_finish_cap_snap(ci, capsnap)) | |
2870 | flushsnaps = 1; | |
2871 | wake = 1; | |
a8599bd8 | 2872 | } |
5dda377c YZ |
2873 | if (ci->i_wrbuffer_ref_head == 0 && |
2874 | ci->i_dirty_caps == 0 && | |
2875 | ci->i_flushing_caps == 0) { | |
2876 | BUG_ON(!ci->i_head_snapc); | |
2877 | ceph_put_snap_context(ci->i_head_snapc); | |
2878 | ci->i_head_snapc = NULL; | |
2879 | } | |
db40cc17 YZ |
2880 | /* see comment in __ceph_remove_cap() */ |
2881 | if (!__ceph_is_any_caps(ci) && ci->i_snap_realm) | |
2882 | drop_inode_snap_realm(ci); | |
a8599bd8 | 2883 | } |
be655596 | 2884 | spin_unlock(&ci->i_ceph_lock); |
a8599bd8 | 2885 | |
819ccbfa SW |
2886 | dout("put_cap_refs %p had %s%s%s\n", inode, ceph_cap_string(had), |
2887 | last ? " last" : "", put ? " put" : ""); | |
a8599bd8 SW |
2888 | |
2889 | if (last && !flushsnaps) | |
2890 | ceph_check_caps(ci, 0, NULL); | |
2891 | else if (flushsnaps) | |
ed9b430c | 2892 | ceph_flush_snaps(ci, NULL); |
a8599bd8 | 2893 | if (wake) |
03066f23 | 2894 | wake_up_all(&ci->i_cap_wq); |
86056090 | 2895 | while (put-- > 0) |
a8599bd8 SW |
2896 | iput(inode); |
2897 | } | |
2898 | ||
2899 | /* | |
2900 | * Release @nr WRBUFFER refs on dirty pages for the given @snapc snap | |
2901 | * context. Adjust per-snap dirty page accounting as appropriate. | |
2902 | * Once all dirty data for a cap_snap is flushed, flush snapped file | |
2903 | * metadata back to the MDS. If we dropped the last ref, call | |
2904 | * ceph_check_caps. | |
2905 | */ | |
2906 | void ceph_put_wrbuffer_cap_refs(struct ceph_inode_info *ci, int nr, | |
2907 | struct ceph_snap_context *snapc) | |
2908 | { | |
2909 | struct inode *inode = &ci->vfs_inode; | |
a8599bd8 | 2910 | struct ceph_cap_snap *capsnap = NULL; |
70220ac8 YZ |
2911 | int put = 0; |
2912 | bool last = false; | |
2913 | bool found = false; | |
2914 | bool flush_snaps = false; | |
2915 | bool complete_capsnap = false; | |
a8599bd8 | 2916 | |
be655596 | 2917 | spin_lock(&ci->i_ceph_lock); |
a8599bd8 | 2918 | ci->i_wrbuffer_ref -= nr; |
70220ac8 YZ |
2919 | if (ci->i_wrbuffer_ref == 0) { |
2920 | last = true; | |
2921 | put++; | |
2922 | } | |
a8599bd8 SW |
2923 | |
2924 | if (ci->i_head_snapc == snapc) { | |
2925 | ci->i_wrbuffer_ref_head -= nr; | |
7d8cb26d | 2926 | if (ci->i_wrbuffer_ref_head == 0 && |
5dda377c YZ |
2927 | ci->i_wr_ref == 0 && |
2928 | ci->i_dirty_caps == 0 && | |
2929 | ci->i_flushing_caps == 0) { | |
7d8cb26d | 2930 | BUG_ON(!ci->i_head_snapc); |
a8599bd8 SW |
2931 | ceph_put_snap_context(ci->i_head_snapc); |
2932 | ci->i_head_snapc = NULL; | |
2933 | } | |
2934 | dout("put_wrbuffer_cap_refs on %p head %d/%d -> %d/%d %s\n", | |
2935 | inode, | |
2936 | ci->i_wrbuffer_ref+nr, ci->i_wrbuffer_ref_head+nr, | |
2937 | ci->i_wrbuffer_ref, ci->i_wrbuffer_ref_head, | |
2938 | last ? " LAST" : ""); | |
2939 | } else { | |
2940 | list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) { | |
2941 | if (capsnap->context == snapc) { | |
70220ac8 | 2942 | found = true; |
a8599bd8 SW |
2943 | break; |
2944 | } | |
2945 | } | |
2946 | BUG_ON(!found); | |
819ccbfa SW |
2947 | capsnap->dirty_pages -= nr; |
2948 | if (capsnap->dirty_pages == 0) { | |
70220ac8 YZ |
2949 | complete_capsnap = true; |
2950 | if (!capsnap->writing) { | |
2951 | if (ceph_try_drop_cap_snap(ci, capsnap)) { | |
2952 | put++; | |
2953 | } else { | |
2954 | ci->i_ceph_flags |= CEPH_I_FLUSH_SNAPS; | |
2955 | flush_snaps = true; | |
2956 | } | |
2957 | } | |
819ccbfa | 2958 | } |
a8599bd8 | 2959 | dout("put_wrbuffer_cap_refs on %p cap_snap %p " |
86056090 | 2960 | " snap %lld %d/%d -> %d/%d %s%s\n", |
a8599bd8 SW |
2961 | inode, capsnap, capsnap->context->seq, |
2962 | ci->i_wrbuffer_ref+nr, capsnap->dirty_pages + nr, | |
2963 | ci->i_wrbuffer_ref, capsnap->dirty_pages, | |
2964 | last ? " (wrbuffer last)" : "", | |
86056090 | 2965 | complete_capsnap ? " (complete capsnap)" : ""); |
a8599bd8 SW |
2966 | } |
2967 | ||
be655596 | 2968 | spin_unlock(&ci->i_ceph_lock); |
a8599bd8 SW |
2969 | |
2970 | if (last) { | |
2971 | ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL); | |
70220ac8 | 2972 | } else if (flush_snaps) { |
ed9b430c | 2973 | ceph_flush_snaps(ci, NULL); |
a8599bd8 | 2974 | } |
70220ac8 YZ |
2975 | if (complete_capsnap) |
2976 | wake_up_all(&ci->i_cap_wq); | |
2977 | while (put-- > 0) | |
819ccbfa | 2978 | iput(inode); |
a8599bd8 SW |
2979 | } |
2980 | ||
ca20c991 YZ |
2981 | /* |
2982 | * Invalidate unlinked inode's aliases, so we can drop the inode ASAP. | |
2983 | */ | |
2984 | static void invalidate_aliases(struct inode *inode) | |
2985 | { | |
2986 | struct dentry *dn, *prev = NULL; | |
2987 | ||
2988 | dout("invalidate_aliases inode %p\n", inode); | |
2989 | d_prune_aliases(inode); | |
2990 | /* | |
2991 | * For non-directory inode, d_find_alias() only returns | |
fc12c80a BF |
2992 | * hashed dentry. After calling d_invalidate(), the |
2993 | * dentry becomes unhashed. | |
ca20c991 | 2994 | * |
a8d436f0 | 2995 | * For directory inode, d_find_alias() can return |
fc12c80a | 2996 | * unhashed dentry. But directory inode should have |
ca20c991 YZ |
2997 | * one alias at most. |
2998 | */ | |
2999 | while ((dn = d_find_alias(inode))) { | |
3000 | if (dn == prev) { | |
3001 | dput(dn); | |
3002 | break; | |
3003 | } | |
a8d436f0 | 3004 | d_invalidate(dn); |
ca20c991 YZ |
3005 | if (prev) |
3006 | dput(prev); | |
3007 | prev = dn; | |
3008 | } | |
3009 | if (prev) | |
3010 | dput(prev); | |
3011 | } | |
3012 | ||
a1c6b835 YZ |
3013 | struct cap_extra_info { |
3014 | struct ceph_string *pool_ns; | |
3015 | /* inline data */ | |
3016 | u64 inline_version; | |
3017 | void *inline_data; | |
3018 | u32 inline_len; | |
4985d6f9 YZ |
3019 | /* dirstat */ |
3020 | bool dirstat_valid; | |
3021 | u64 nfiles; | |
3022 | u64 nsubdirs; | |
a1c6b835 YZ |
3023 | /* currently issued */ |
3024 | int issued; | |
3025 | }; | |
3026 | ||
a8599bd8 SW |
3027 | /* |
3028 | * Handle a cap GRANT message from the MDS. (Note that a GRANT may | |
3029 | * actually be a revocation if it specifies a smaller cap set.) | |
3030 | * | |
be655596 | 3031 | * caller holds s_mutex and i_ceph_lock, we drop both. |
a8599bd8 | 3032 | */ |
a1c6b835 | 3033 | static void handle_cap_grant(struct inode *inode, |
15637c8b | 3034 | struct ceph_mds_session *session, |
a1c6b835 YZ |
3035 | struct ceph_cap *cap, |
3036 | struct ceph_mds_caps *grant, | |
3037 | struct ceph_buffer *xattr_buf, | |
3038 | struct cap_extra_info *extra_info) | |
2cd698be | 3039 | __releases(ci->i_ceph_lock) |
a1c6b835 | 3040 | __releases(session->s_mdsc->snap_rwsem) |
a8599bd8 SW |
3041 | { |
3042 | struct ceph_inode_info *ci = ceph_inode(inode); | |
2f56f56a | 3043 | int seq = le32_to_cpu(grant->seq); |
a8599bd8 | 3044 | int newcaps = le32_to_cpu(grant->caps); |
2cd698be | 3045 | int used, wanted, dirty; |
a8599bd8 SW |
3046 | u64 size = le64_to_cpu(grant->size); |
3047 | u64 max_size = le64_to_cpu(grant->max_size); | |
15637c8b | 3048 | int check_caps = 0; |
ab6c2c3e FF |
3049 | bool wake = false; |
3050 | bool writeback = false; | |
3051 | bool queue_trunc = false; | |
3052 | bool queue_invalidate = false; | |
ab6c2c3e | 3053 | bool deleted_inode = false; |
31c542a1 | 3054 | bool fill_inline = false; |
a8599bd8 | 3055 | |
2f56f56a | 3056 | dout("handle_cap_grant inode %p cap %p mds%d seq %d %s\n", |
a1c6b835 | 3057 | inode, cap, session->s_mds, seq, ceph_cap_string(newcaps)); |
a8599bd8 SW |
3058 | dout(" size %llu max_size %llu, i_size %llu\n", size, max_size, |
3059 | inode->i_size); | |
3060 | ||
11df2dfb YZ |
3061 | |
3062 | /* | |
3063 | * auth mds of the inode changed. we received the cap export message, | |
3064 | * but still haven't received the cap import message. handle_cap_export | |
3065 | * updated the new auth MDS' cap. | |
3066 | * | |
3067 | * "ceph_seq_cmp(seq, cap->seq) <= 0" means we are processing a message | |
3068 | * that was sent before the cap import message. So don't remove caps. | |
3069 | */ | |
3070 | if (ceph_seq_cmp(seq, cap->seq) <= 0) { | |
3071 | WARN_ON(cap != ci->i_auth_cap); | |
3072 | WARN_ON(cap->cap_id != le64_to_cpu(grant->cap_id)); | |
3073 | seq = cap->seq; | |
3074 | newcaps |= cap->issued; | |
3075 | } | |
3076 | ||
a8599bd8 SW |
3077 | /* |
3078 | * If CACHE is being revoked, and we have no dirty buffers, | |
3079 | * try to invalidate (once). (If there are dirty buffers, we | |
3080 | * will invalidate _after_ writeback.) | |
3081 | */ | |
fdd4e158 YZ |
3082 | if (!S_ISDIR(inode->i_mode) && /* don't invalidate readdir cache */ |
3083 | ((cap->issued & ~newcaps) & CEPH_CAP_FILE_CACHE) && | |
3b454c49 | 3084 | (newcaps & CEPH_CAP_FILE_LAZYIO) == 0 && |
9abd4db7 | 3085 | !(ci->i_wrbuffer_ref || ci->i_wb_ref)) { |
e9075743 | 3086 | if (try_nonblocking_invalidate(inode)) { |
a8599bd8 SW |
3087 | /* there were locked pages.. invalidate later |
3088 | in a separate thread. */ | |
3089 | if (ci->i_rdcache_revoking != ci->i_rdcache_gen) { | |
ab6c2c3e | 3090 | queue_invalidate = true; |
a8599bd8 SW |
3091 | ci->i_rdcache_revoking = ci->i_rdcache_gen; |
3092 | } | |
a8599bd8 | 3093 | } |
a8599bd8 SW |
3094 | } |
3095 | ||
3096 | /* side effects now are allowed */ | |
685f9a5d | 3097 | cap->cap_gen = session->s_cap_gen; |
11df2dfb | 3098 | cap->seq = seq; |
a8599bd8 SW |
3099 | |
3100 | __check_cap_issue(ci, cap, newcaps); | |
3101 | ||
f98a128a | 3102 | if ((newcaps & CEPH_CAP_AUTH_SHARED) && |
a1c6b835 | 3103 | (extra_info->issued & CEPH_CAP_AUTH_EXCL) == 0) { |
a8599bd8 | 3104 | inode->i_mode = le32_to_cpu(grant->mode); |
05cb11c1 EB |
3105 | inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(grant->uid)); |
3106 | inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(grant->gid)); | |
a8599bd8 | 3107 | dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode, |
bd2bae6a EB |
3108 | from_kuid(&init_user_ns, inode->i_uid), |
3109 | from_kgid(&init_user_ns, inode->i_gid)); | |
a8599bd8 SW |
3110 | } |
3111 | ||
fa466743 | 3112 | if ((newcaps & CEPH_CAP_LINK_SHARED) && |
a1c6b835 | 3113 | (extra_info->issued & CEPH_CAP_LINK_EXCL) == 0) { |
bfe86848 | 3114 | set_nlink(inode, le32_to_cpu(grant->nlink)); |
ca20c991 YZ |
3115 | if (inode->i_nlink == 0 && |
3116 | (newcaps & (CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL))) | |
ab6c2c3e | 3117 | deleted_inode = true; |
ca20c991 | 3118 | } |
a8599bd8 | 3119 | |
a1c6b835 YZ |
3120 | if ((extra_info->issued & CEPH_CAP_XATTR_EXCL) == 0 && |
3121 | grant->xattr_len) { | |
a8599bd8 SW |
3122 | int len = le32_to_cpu(grant->xattr_len); |
3123 | u64 version = le64_to_cpu(grant->xattr_version); | |
3124 | ||
3125 | if (version > ci->i_xattrs.version) { | |
3126 | dout(" got new xattrs v%llu on %p len %d\n", | |
3127 | version, inode, len); | |
3128 | if (ci->i_xattrs.blob) | |
3129 | ceph_buffer_put(ci->i_xattrs.blob); | |
3130 | ci->i_xattrs.blob = ceph_buffer_get(xattr_buf); | |
3131 | ci->i_xattrs.version = version; | |
7221fe4c | 3132 | ceph_forget_all_cached_acls(inode); |
a8599bd8 SW |
3133 | } |
3134 | } | |
3135 | ||
f98a128a | 3136 | if (newcaps & CEPH_CAP_ANY_RD) { |
9bbeab41 | 3137 | struct timespec64 mtime, atime, ctime; |
f98a128a | 3138 | /* ctime/mtime/atime? */ |
9bbeab41 AB |
3139 | ceph_decode_timespec64(&mtime, &grant->mtime); |
3140 | ceph_decode_timespec64(&atime, &grant->atime); | |
3141 | ceph_decode_timespec64(&ctime, &grant->ctime); | |
a1c6b835 | 3142 | ceph_fill_file_time(inode, extra_info->issued, |
f98a128a YZ |
3143 | le32_to_cpu(grant->time_warp_seq), |
3144 | &ctime, &mtime, &atime); | |
3145 | } | |
3146 | ||
4985d6f9 YZ |
3147 | if ((newcaps & CEPH_CAP_FILE_SHARED) && extra_info->dirstat_valid) { |
3148 | ci->i_files = extra_info->nfiles; | |
3149 | ci->i_subdirs = extra_info->nsubdirs; | |
3150 | } | |
3151 | ||
f98a128a YZ |
3152 | if (newcaps & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR)) { |
3153 | /* file layout may have changed */ | |
7627151e | 3154 | s64 old_pool = ci->i_layout.pool_id; |
779fe0fb YZ |
3155 | struct ceph_string *old_ns; |
3156 | ||
7627151e | 3157 | ceph_file_layout_from_legacy(&ci->i_layout, &grant->layout); |
779fe0fb YZ |
3158 | old_ns = rcu_dereference_protected(ci->i_layout.pool_ns, |
3159 | lockdep_is_held(&ci->i_ceph_lock)); | |
a1c6b835 | 3160 | rcu_assign_pointer(ci->i_layout.pool_ns, extra_info->pool_ns); |
779fe0fb | 3161 | |
a1c6b835 YZ |
3162 | if (ci->i_layout.pool_id != old_pool || |
3163 | extra_info->pool_ns != old_ns) | |
7627151e | 3164 | ci->i_ceph_flags &= ~CEPH_I_POOL_PERM; |
5ea5c5e0 | 3165 | |
a1c6b835 | 3166 | extra_info->pool_ns = old_ns; |
779fe0fb | 3167 | |
f98a128a | 3168 | /* size/truncate_seq? */ |
a1c6b835 | 3169 | queue_trunc = ceph_fill_file_size(inode, extra_info->issued, |
f98a128a YZ |
3170 | le32_to_cpu(grant->truncate_seq), |
3171 | le64_to_cpu(grant->truncate_size), | |
3172 | size); | |
84eea8c7 YZ |
3173 | } |
3174 | ||
3175 | if (ci->i_auth_cap == cap && (newcaps & CEPH_CAP_ANY_FILE_WR)) { | |
3176 | if (max_size != ci->i_max_size) { | |
f98a128a YZ |
3177 | dout("max_size %lld -> %llu\n", |
3178 | ci->i_max_size, max_size); | |
3179 | ci->i_max_size = max_size; | |
3180 | if (max_size >= ci->i_wanted_max_size) { | |
3181 | ci->i_wanted_max_size = 0; /* reset */ | |
3182 | ci->i_requested_max_size = 0; | |
3183 | } | |
ab6c2c3e | 3184 | wake = true; |
84eea8c7 YZ |
3185 | } else if (ci->i_wanted_max_size > ci->i_max_size && |
3186 | ci->i_wanted_max_size > ci->i_requested_max_size) { | |
3187 | /* CEPH_CAP_OP_IMPORT */ | |
3188 | wake = true; | |
a8599bd8 | 3189 | } |
a8599bd8 SW |
3190 | } |
3191 | ||
3192 | /* check cap bits */ | |
3193 | wanted = __ceph_caps_wanted(ci); | |
3194 | used = __ceph_caps_used(ci); | |
3195 | dirty = __ceph_caps_dirty(ci); | |
3196 | dout(" my wanted = %s, used = %s, dirty %s\n", | |
3197 | ceph_cap_string(wanted), | |
3198 | ceph_cap_string(used), | |
3199 | ceph_cap_string(dirty)); | |
3200 | if (wanted != le32_to_cpu(grant->wanted)) { | |
3201 | dout("mds wanted %s -> %s\n", | |
3202 | ceph_cap_string(le32_to_cpu(grant->wanted)), | |
3203 | ceph_cap_string(wanted)); | |
390306c3 YZ |
3204 | /* imported cap may not have correct mds_wanted */ |
3205 | if (le32_to_cpu(grant->op) == CEPH_CAP_OP_IMPORT) | |
3206 | check_caps = 1; | |
a8599bd8 SW |
3207 | } |
3208 | ||
a8599bd8 SW |
3209 | /* revocation, grant, or no-op? */ |
3210 | if (cap->issued & ~newcaps) { | |
3b454c49 SW |
3211 | int revoking = cap->issued & ~newcaps; |
3212 | ||
3213 | dout("revocation: %s -> %s (revoking %s)\n", | |
3214 | ceph_cap_string(cap->issued), | |
3215 | ceph_cap_string(newcaps), | |
3216 | ceph_cap_string(revoking)); | |
0eb6cd49 | 3217 | if (revoking & used & CEPH_CAP_FILE_BUFFER) |
ab6c2c3e | 3218 | writeback = true; /* initiate writeback; will delay ack */ |
3b454c49 SW |
3219 | else if (revoking == CEPH_CAP_FILE_CACHE && |
3220 | (newcaps & CEPH_CAP_FILE_LAZYIO) == 0 && | |
3221 | queue_invalidate) | |
3222 | ; /* do nothing yet, invalidation will be queued */ | |
3223 | else if (cap == ci->i_auth_cap) | |
3224 | check_caps = 1; /* check auth cap only */ | |
3225 | else | |
3226 | check_caps = 2; /* check all caps */ | |
a8599bd8 | 3227 | cap->issued = newcaps; |
978097c9 | 3228 | cap->implemented |= newcaps; |
a8599bd8 SW |
3229 | } else if (cap->issued == newcaps) { |
3230 | dout("caps unchanged: %s -> %s\n", | |
3231 | ceph_cap_string(cap->issued), ceph_cap_string(newcaps)); | |
3232 | } else { | |
3233 | dout("grant: %s -> %s\n", ceph_cap_string(cap->issued), | |
3234 | ceph_cap_string(newcaps)); | |
6ee6b953 YZ |
3235 | /* non-auth MDS is revoking the newly grant caps ? */ |
3236 | if (cap == ci->i_auth_cap && | |
3237 | __ceph_caps_revoking_other(ci, cap, newcaps)) | |
3238 | check_caps = 2; | |
3239 | ||
a8599bd8 SW |
3240 | cap->issued = newcaps; |
3241 | cap->implemented |= newcaps; /* add bits only, to | |
3242 | * avoid stepping on a | |
3243 | * pending revocation */ | |
ab6c2c3e | 3244 | wake = true; |
a8599bd8 | 3245 | } |
978097c9 | 3246 | BUG_ON(cap->issued & ~cap->implemented); |
a8599bd8 | 3247 | |
a1c6b835 YZ |
3248 | if (extra_info->inline_version > 0 && |
3249 | extra_info->inline_version >= ci->i_inline_version) { | |
3250 | ci->i_inline_version = extra_info->inline_version; | |
31c542a1 YZ |
3251 | if (ci->i_inline_version != CEPH_INLINE_NONE && |
3252 | (newcaps & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO))) | |
3253 | fill_inline = true; | |
3254 | } | |
3255 | ||
2cd698be | 3256 | if (le32_to_cpu(grant->op) == CEPH_CAP_OP_IMPORT) { |
a1c6b835 | 3257 | if (newcaps & ~extra_info->issued) |
ab6c2c3e | 3258 | wake = true; |
a1c6b835 YZ |
3259 | kick_flushing_inode_caps(session->s_mdsc, session, inode); |
3260 | up_read(&session->s_mdsc->snap_rwsem); | |
0e294387 YZ |
3261 | } else { |
3262 | spin_unlock(&ci->i_ceph_lock); | |
2cd698be YZ |
3263 | } |
3264 | ||
31c542a1 | 3265 | if (fill_inline) |
a1c6b835 YZ |
3266 | ceph_fill_inline_data(inode, NULL, extra_info->inline_data, |
3267 | extra_info->inline_len); | |
31c542a1 | 3268 | |
14649758 | 3269 | if (queue_trunc) |
c6bcda6f | 3270 | ceph_queue_vmtruncate(inode); |
c6bcda6f | 3271 | |
3c6f6b79 | 3272 | if (writeback) |
a8599bd8 SW |
3273 | /* |
3274 | * queue inode for writeback: we can't actually call | |
3275 | * filemap_write_and_wait, etc. from message handler | |
3276 | * context. | |
3277 | */ | |
3c6f6b79 SW |
3278 | ceph_queue_writeback(inode); |
3279 | if (queue_invalidate) | |
3280 | ceph_queue_invalidate(inode); | |
ca20c991 YZ |
3281 | if (deleted_inode) |
3282 | invalidate_aliases(inode); | |
a8599bd8 | 3283 | if (wake) |
03066f23 | 3284 | wake_up_all(&ci->i_cap_wq); |
15637c8b SW |
3285 | |
3286 | if (check_caps == 1) | |
3287 | ceph_check_caps(ci, CHECK_CAPS_NODELAY|CHECK_CAPS_AUTHONLY, | |
3288 | session); | |
3289 | else if (check_caps == 2) | |
3290 | ceph_check_caps(ci, CHECK_CAPS_NODELAY, session); | |
3291 | else | |
3292 | mutex_unlock(&session->s_mutex); | |
a8599bd8 SW |
3293 | } |
3294 | ||
3295 | /* | |
3296 | * Handle FLUSH_ACK from MDS, indicating that metadata we sent to the | |
3297 | * MDS has been safely committed. | |
3298 | */ | |
6df058c0 | 3299 | static void handle_cap_flush_ack(struct inode *inode, u64 flush_tid, |
a8599bd8 SW |
3300 | struct ceph_mds_caps *m, |
3301 | struct ceph_mds_session *session, | |
3302 | struct ceph_cap *cap) | |
be655596 | 3303 | __releases(ci->i_ceph_lock) |
a8599bd8 SW |
3304 | { |
3305 | struct ceph_inode_info *ci = ceph_inode(inode); | |
3d14c5d2 | 3306 | struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc; |
e4500b5e | 3307 | struct ceph_cap_flush *cf, *tmp_cf; |
553adfd9 | 3308 | LIST_HEAD(to_remove); |
a8599bd8 SW |
3309 | unsigned seq = le32_to_cpu(m->seq); |
3310 | int dirty = le32_to_cpu(m->dirty); | |
3311 | int cleaned = 0; | |
c8799fc4 | 3312 | bool drop = false; |
7271efa7 TM |
3313 | bool wake_ci = false; |
3314 | bool wake_mdsc = false; | |
a8599bd8 | 3315 | |
e4500b5e | 3316 | list_for_each_entry_safe(cf, tmp_cf, &ci->i_cap_flush_list, i_list) { |
553adfd9 YZ |
3317 | if (cf->tid == flush_tid) |
3318 | cleaned = cf->caps; | |
0e294387 YZ |
3319 | if (cf->caps == 0) /* capsnap */ |
3320 | continue; | |
553adfd9 | 3321 | if (cf->tid <= flush_tid) { |
c8799fc4 YZ |
3322 | if (__finish_cap_flush(NULL, ci, cf)) |
3323 | wake_ci = true; | |
e4500b5e | 3324 | list_add_tail(&cf->i_list, &to_remove); |
553adfd9 YZ |
3325 | } else { |
3326 | cleaned &= ~cf->caps; | |
3327 | if (!cleaned) | |
3328 | break; | |
3329 | } | |
3330 | } | |
a8599bd8 SW |
3331 | |
3332 | dout("handle_cap_flush_ack inode %p mds%d seq %d on %s cleaned %s," | |
3333 | " flushing %s -> %s\n", | |
3334 | inode, session->s_mds, seq, ceph_cap_string(dirty), | |
3335 | ceph_cap_string(cleaned), ceph_cap_string(ci->i_flushing_caps), | |
3336 | ceph_cap_string(ci->i_flushing_caps & ~cleaned)); | |
3337 | ||
8310b089 | 3338 | if (list_empty(&to_remove) && !cleaned) |
a8599bd8 SW |
3339 | goto out; |
3340 | ||
a8599bd8 | 3341 | ci->i_flushing_caps &= ~cleaned; |
a8599bd8 SW |
3342 | |
3343 | spin_lock(&mdsc->cap_dirty_lock); | |
8310b089 | 3344 | |
c8799fc4 YZ |
3345 | list_for_each_entry(cf, &to_remove, i_list) { |
3346 | if (__finish_cap_flush(mdsc, NULL, cf)) | |
3347 | wake_mdsc = true; | |
8310b089 YZ |
3348 | } |
3349 | ||
a8599bd8 | 3350 | if (ci->i_flushing_caps == 0) { |
0e294387 YZ |
3351 | if (list_empty(&ci->i_cap_flush_list)) { |
3352 | list_del_init(&ci->i_flushing_item); | |
3353 | if (!list_empty(&session->s_cap_flushing)) { | |
3354 | dout(" mds%d still flushing cap on %p\n", | |
3355 | session->s_mds, | |
3356 | &list_first_entry(&session->s_cap_flushing, | |
3357 | struct ceph_inode_info, | |
3358 | i_flushing_item)->vfs_inode); | |
3359 | } | |
3360 | } | |
a8599bd8 | 3361 | mdsc->num_cap_flushing--; |
a8599bd8 | 3362 | dout(" inode %p now !flushing\n", inode); |
afcdaea3 SW |
3363 | |
3364 | if (ci->i_dirty_caps == 0) { | |
3365 | dout(" inode %p now clean\n", inode); | |
3366 | BUG_ON(!list_empty(&ci->i_dirty_item)); | |
c8799fc4 | 3367 | drop = true; |
5dda377c YZ |
3368 | if (ci->i_wr_ref == 0 && |
3369 | ci->i_wrbuffer_ref_head == 0) { | |
7d8cb26d SW |
3370 | BUG_ON(!ci->i_head_snapc); |
3371 | ceph_put_snap_context(ci->i_head_snapc); | |
3372 | ci->i_head_snapc = NULL; | |
3373 | } | |
76e3b390 SW |
3374 | } else { |
3375 | BUG_ON(list_empty(&ci->i_dirty_item)); | |
afcdaea3 | 3376 | } |
a8599bd8 SW |
3377 | } |
3378 | spin_unlock(&mdsc->cap_dirty_lock); | |
a8599bd8 SW |
3379 | |
3380 | out: | |
be655596 | 3381 | spin_unlock(&ci->i_ceph_lock); |
553adfd9 YZ |
3382 | |
3383 | while (!list_empty(&to_remove)) { | |
3384 | cf = list_first_entry(&to_remove, | |
e4500b5e YZ |
3385 | struct ceph_cap_flush, i_list); |
3386 | list_del(&cf->i_list); | |
f66fd9f0 | 3387 | ceph_free_cap_flush(cf); |
553adfd9 | 3388 | } |
c8799fc4 YZ |
3389 | |
3390 | if (wake_ci) | |
3391 | wake_up_all(&ci->i_cap_wq); | |
3392 | if (wake_mdsc) | |
3393 | wake_up_all(&mdsc->cap_flushing_wq); | |
afcdaea3 | 3394 | if (drop) |
a8599bd8 SW |
3395 | iput(inode); |
3396 | } | |
3397 | ||
3398 | /* | |
3399 | * Handle FLUSHSNAP_ACK. MDS has flushed snap data to disk and we can | |
3400 | * throw away our cap_snap. | |
3401 | * | |
3402 | * Caller hold s_mutex. | |
3403 | */ | |
6df058c0 | 3404 | static void handle_cap_flushsnap_ack(struct inode *inode, u64 flush_tid, |
a8599bd8 SW |
3405 | struct ceph_mds_caps *m, |
3406 | struct ceph_mds_session *session) | |
3407 | { | |
3408 | struct ceph_inode_info *ci = ceph_inode(inode); | |
affbc19a | 3409 | struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc; |
a8599bd8 | 3410 | u64 follows = le64_to_cpu(m->snap_follows); |
a8599bd8 | 3411 | struct ceph_cap_snap *capsnap; |
c8799fc4 YZ |
3412 | bool flushed = false; |
3413 | bool wake_ci = false; | |
3414 | bool wake_mdsc = false; | |
a8599bd8 SW |
3415 | |
3416 | dout("handle_cap_flushsnap_ack inode %p ci %p mds%d follows %lld\n", | |
3417 | inode, ci, session->s_mds, follows); | |
3418 | ||
be655596 | 3419 | spin_lock(&ci->i_ceph_lock); |
a8599bd8 SW |
3420 | list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) { |
3421 | if (capsnap->follows == follows) { | |
0e294387 | 3422 | if (capsnap->cap_flush.tid != flush_tid) { |
a8599bd8 SW |
3423 | dout(" cap_snap %p follows %lld tid %lld !=" |
3424 | " %lld\n", capsnap, follows, | |
0e294387 | 3425 | flush_tid, capsnap->cap_flush.tid); |
a8599bd8 SW |
3426 | break; |
3427 | } | |
c8799fc4 | 3428 | flushed = true; |
a8599bd8 SW |
3429 | break; |
3430 | } else { | |
3431 | dout(" skipping cap_snap %p follows %lld\n", | |
3432 | capsnap, capsnap->follows); | |
3433 | } | |
3434 | } | |
0e294387 | 3435 | if (flushed) { |
0e294387 YZ |
3436 | WARN_ON(capsnap->dirty_pages || capsnap->writing); |
3437 | dout(" removing %p cap_snap %p follows %lld\n", | |
3438 | inode, capsnap, follows); | |
3439 | list_del(&capsnap->ci_item); | |
c8799fc4 YZ |
3440 | if (__finish_cap_flush(NULL, ci, &capsnap->cap_flush)) |
3441 | wake_ci = true; | |
0e294387 YZ |
3442 | |
3443 | spin_lock(&mdsc->cap_dirty_lock); | |
3444 | ||
3445 | if (list_empty(&ci->i_cap_flush_list)) | |
3446 | list_del_init(&ci->i_flushing_item); | |
3447 | ||
c8799fc4 YZ |
3448 | if (__finish_cap_flush(mdsc, NULL, &capsnap->cap_flush)) |
3449 | wake_mdsc = true; | |
0e294387 YZ |
3450 | |
3451 | spin_unlock(&mdsc->cap_dirty_lock); | |
0e294387 | 3452 | } |
be655596 | 3453 | spin_unlock(&ci->i_ceph_lock); |
0e294387 YZ |
3454 | if (flushed) { |
3455 | ceph_put_snap_context(capsnap->context); | |
3456 | ceph_put_cap_snap(capsnap); | |
c8799fc4 YZ |
3457 | if (wake_ci) |
3458 | wake_up_all(&ci->i_cap_wq); | |
3459 | if (wake_mdsc) | |
3460 | wake_up_all(&mdsc->cap_flushing_wq); | |
a8599bd8 | 3461 | iput(inode); |
0e294387 | 3462 | } |
a8599bd8 SW |
3463 | } |
3464 | ||
3465 | /* | |
3466 | * Handle TRUNC from MDS, indicating file truncation. | |
3467 | * | |
3468 | * caller hold s_mutex. | |
3469 | */ | |
3470 | static void handle_cap_trunc(struct inode *inode, | |
3471 | struct ceph_mds_caps *trunc, | |
3472 | struct ceph_mds_session *session) | |
be655596 | 3473 | __releases(ci->i_ceph_lock) |
a8599bd8 SW |
3474 | { |
3475 | struct ceph_inode_info *ci = ceph_inode(inode); | |
3476 | int mds = session->s_mds; | |
3477 | int seq = le32_to_cpu(trunc->seq); | |
3478 | u32 truncate_seq = le32_to_cpu(trunc->truncate_seq); | |
3479 | u64 truncate_size = le64_to_cpu(trunc->truncate_size); | |
3480 | u64 size = le64_to_cpu(trunc->size); | |
3481 | int implemented = 0; | |
3482 | int dirty = __ceph_caps_dirty(ci); | |
3483 | int issued = __ceph_caps_issued(ceph_inode(inode), &implemented); | |
3484 | int queue_trunc = 0; | |
3485 | ||
3486 | issued |= implemented | dirty; | |
3487 | ||
3488 | dout("handle_cap_trunc inode %p mds%d seq %d to %lld seq %d\n", | |
3489 | inode, mds, seq, truncate_size, truncate_seq); | |
3490 | queue_trunc = ceph_fill_file_size(inode, issued, | |
3491 | truncate_seq, truncate_size, size); | |
be655596 | 3492 | spin_unlock(&ci->i_ceph_lock); |
a8599bd8 | 3493 | |
14649758 | 3494 | if (queue_trunc) |
3c6f6b79 | 3495 | ceph_queue_vmtruncate(inode); |
a8599bd8 SW |
3496 | } |
3497 | ||
3498 | /* | |
3499 | * Handle EXPORT from MDS. Cap is being migrated _from_ this mds to a | |
3500 | * different one. If we are the most recent migration we've seen (as | |
3501 | * indicated by mseq), make note of the migrating cap bits for the | |
3502 | * duration (until we see the corresponding IMPORT). | |
3503 | * | |
3504 | * caller holds s_mutex | |
3505 | */ | |
3506 | static void handle_cap_export(struct inode *inode, struct ceph_mds_caps *ex, | |
11df2dfb YZ |
3507 | struct ceph_mds_cap_peer *ph, |
3508 | struct ceph_mds_session *session) | |
a8599bd8 | 3509 | { |
db354052 | 3510 | struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc; |
11df2dfb | 3511 | struct ceph_mds_session *tsession = NULL; |
d9df2783 | 3512 | struct ceph_cap *cap, *tcap, *new_cap = NULL; |
a8599bd8 | 3513 | struct ceph_inode_info *ci = ceph_inode(inode); |
11df2dfb | 3514 | u64 t_cap_id; |
a8599bd8 | 3515 | unsigned mseq = le32_to_cpu(ex->migrate_seq); |
11df2dfb YZ |
3516 | unsigned t_seq, t_mseq; |
3517 | int target, issued; | |
3518 | int mds = session->s_mds; | |
a8599bd8 | 3519 | |
11df2dfb YZ |
3520 | if (ph) { |
3521 | t_cap_id = le64_to_cpu(ph->cap_id); | |
3522 | t_seq = le32_to_cpu(ph->seq); | |
3523 | t_mseq = le32_to_cpu(ph->mseq); | |
3524 | target = le32_to_cpu(ph->mds); | |
3525 | } else { | |
3526 | t_cap_id = t_seq = t_mseq = 0; | |
3527 | target = -1; | |
3528 | } | |
a8599bd8 | 3529 | |
11df2dfb YZ |
3530 | dout("handle_cap_export inode %p ci %p mds%d mseq %d target %d\n", |
3531 | inode, ci, mds, mseq, target); | |
3532 | retry: | |
be655596 | 3533 | spin_lock(&ci->i_ceph_lock); |
11df2dfb | 3534 | cap = __get_cap_for_mds(ci, mds); |
ca665e02 | 3535 | if (!cap || cap->cap_id != le64_to_cpu(ex->cap_id)) |
11df2dfb | 3536 | goto out_unlock; |
a8599bd8 | 3537 | |
11df2dfb YZ |
3538 | if (target < 0) { |
3539 | __ceph_remove_cap(cap, false); | |
77310320 YZ |
3540 | if (!ci->i_auth_cap) |
3541 | ci->i_ceph_flags |= CEPH_I_CAP_DROPPED; | |
11df2dfb | 3542 | goto out_unlock; |
a8599bd8 SW |
3543 | } |
3544 | ||
11df2dfb YZ |
3545 | /* |
3546 | * now we know we haven't received the cap import message yet | |
3547 | * because the exported cap still exist. | |
3548 | */ | |
db354052 | 3549 | |
11df2dfb | 3550 | issued = cap->issued; |
d84b37f9 YZ |
3551 | if (issued != cap->implemented) |
3552 | pr_err_ratelimited("handle_cap_export: issued != implemented: " | |
3553 | "ino (%llx.%llx) mds%d seq %d mseq %d " | |
3554 | "issued %s implemented %s\n", | |
3555 | ceph_vinop(inode), mds, cap->seq, cap->mseq, | |
3556 | ceph_cap_string(issued), | |
3557 | ceph_cap_string(cap->implemented)); | |
3558 | ||
11df2dfb YZ |
3559 | |
3560 | tcap = __get_cap_for_mds(ci, target); | |
3561 | if (tcap) { | |
3562 | /* already have caps from the target */ | |
fa0aa3b8 | 3563 | if (tcap->cap_id == t_cap_id && |
11df2dfb YZ |
3564 | ceph_seq_cmp(tcap->seq, t_seq) < 0) { |
3565 | dout(" updating import cap %p mds%d\n", tcap, target); | |
3566 | tcap->cap_id = t_cap_id; | |
3567 | tcap->seq = t_seq - 1; | |
3568 | tcap->issue_seq = t_seq - 1; | |
3569 | tcap->mseq = t_mseq; | |
3570 | tcap->issued |= issued; | |
3571 | tcap->implemented |= issued; | |
3572 | if (cap == ci->i_auth_cap) | |
3573 | ci->i_auth_cap = tcap; | |
00f06cba | 3574 | |
0e294387 YZ |
3575 | if (!list_empty(&ci->i_cap_flush_list) && |
3576 | ci->i_auth_cap == tcap) { | |
11df2dfb YZ |
3577 | spin_lock(&mdsc->cap_dirty_lock); |
3578 | list_move_tail(&ci->i_flushing_item, | |
3579 | &tcap->session->s_cap_flushing); | |
3580 | spin_unlock(&mdsc->cap_dirty_lock); | |
db354052 | 3581 | } |
a8599bd8 | 3582 | } |
a096b09a | 3583 | __ceph_remove_cap(cap, false); |
11df2dfb | 3584 | goto out_unlock; |
d9df2783 | 3585 | } else if (tsession) { |
11df2dfb | 3586 | /* add placeholder for the export tagert */ |
d9df2783 | 3587 | int flag = (cap == ci->i_auth_cap) ? CEPH_CAP_FLAG_AUTH : 0; |
00f06cba | 3588 | tcap = new_cap; |
11df2dfb | 3589 | ceph_add_cap(inode, tsession, t_cap_id, -1, issued, 0, |
d9df2783 YZ |
3590 | t_seq - 1, t_mseq, (u64)-1, flag, &new_cap); |
3591 | ||
00f06cba YZ |
3592 | if (!list_empty(&ci->i_cap_flush_list) && |
3593 | ci->i_auth_cap == tcap) { | |
3594 | spin_lock(&mdsc->cap_dirty_lock); | |
3595 | list_move_tail(&ci->i_flushing_item, | |
3596 | &tcap->session->s_cap_flushing); | |
3597 | spin_unlock(&mdsc->cap_dirty_lock); | |
3598 | } | |
3599 | ||
d9df2783 YZ |
3600 | __ceph_remove_cap(cap, false); |
3601 | goto out_unlock; | |
a8599bd8 SW |
3602 | } |
3603 | ||
be655596 | 3604 | spin_unlock(&ci->i_ceph_lock); |
11df2dfb YZ |
3605 | mutex_unlock(&session->s_mutex); |
3606 | ||
3607 | /* open target session */ | |
3608 | tsession = ceph_mdsc_open_export_target_session(mdsc, target); | |
3609 | if (!IS_ERR(tsession)) { | |
3610 | if (mds > target) { | |
3611 | mutex_lock(&session->s_mutex); | |
3612 | mutex_lock_nested(&tsession->s_mutex, | |
3613 | SINGLE_DEPTH_NESTING); | |
3614 | } else { | |
3615 | mutex_lock(&tsession->s_mutex); | |
3616 | mutex_lock_nested(&session->s_mutex, | |
3617 | SINGLE_DEPTH_NESTING); | |
3618 | } | |
d9df2783 | 3619 | new_cap = ceph_get_cap(mdsc, NULL); |
11df2dfb YZ |
3620 | } else { |
3621 | WARN_ON(1); | |
3622 | tsession = NULL; | |
3623 | target = -1; | |
3624 | } | |
3625 | goto retry; | |
3626 | ||
3627 | out_unlock: | |
3628 | spin_unlock(&ci->i_ceph_lock); | |
3629 | mutex_unlock(&session->s_mutex); | |
3630 | if (tsession) { | |
3631 | mutex_unlock(&tsession->s_mutex); | |
3632 | ceph_put_mds_session(tsession); | |
3633 | } | |
d9df2783 YZ |
3634 | if (new_cap) |
3635 | ceph_put_cap(mdsc, new_cap); | |
a8599bd8 SW |
3636 | } |
3637 | ||
3638 | /* | |
2cd698be | 3639 | * Handle cap IMPORT. |
a8599bd8 | 3640 | * |
2cd698be | 3641 | * caller holds s_mutex. acquires i_ceph_lock |
a8599bd8 SW |
3642 | */ |
3643 | static void handle_cap_import(struct ceph_mds_client *mdsc, | |
3644 | struct inode *inode, struct ceph_mds_caps *im, | |
4ee6a914 | 3645 | struct ceph_mds_cap_peer *ph, |
a8599bd8 | 3646 | struct ceph_mds_session *session, |
2cd698be YZ |
3647 | struct ceph_cap **target_cap, int *old_issued) |
3648 | __acquires(ci->i_ceph_lock) | |
a8599bd8 SW |
3649 | { |
3650 | struct ceph_inode_info *ci = ceph_inode(inode); | |
2cd698be | 3651 | struct ceph_cap *cap, *ocap, *new_cap = NULL; |
a8599bd8 | 3652 | int mds = session->s_mds; |
2cd698be YZ |
3653 | int issued; |
3654 | unsigned caps = le32_to_cpu(im->caps); | |
a8599bd8 SW |
3655 | unsigned wanted = le32_to_cpu(im->wanted); |
3656 | unsigned seq = le32_to_cpu(im->seq); | |
3657 | unsigned mseq = le32_to_cpu(im->migrate_seq); | |
3658 | u64 realmino = le64_to_cpu(im->realm); | |
3659 | u64 cap_id = le64_to_cpu(im->cap_id); | |
4ee6a914 YZ |
3660 | u64 p_cap_id; |
3661 | int peer; | |
a8599bd8 | 3662 | |
4ee6a914 YZ |
3663 | if (ph) { |
3664 | p_cap_id = le64_to_cpu(ph->cap_id); | |
3665 | peer = le32_to_cpu(ph->mds); | |
3666 | } else { | |
3667 | p_cap_id = 0; | |
3668 | peer = -1; | |
3669 | } | |
db354052 | 3670 | |
4ee6a914 YZ |
3671 | dout("handle_cap_import inode %p ci %p mds%d mseq %d peer %d\n", |
3672 | inode, ci, mds, mseq, peer); | |
3673 | ||
d9df2783 | 3674 | retry: |
4ee6a914 | 3675 | spin_lock(&ci->i_ceph_lock); |
d9df2783 YZ |
3676 | cap = __get_cap_for_mds(ci, mds); |
3677 | if (!cap) { | |
3678 | if (!new_cap) { | |
3679 | spin_unlock(&ci->i_ceph_lock); | |
3680 | new_cap = ceph_get_cap(mdsc, NULL); | |
3681 | goto retry; | |
3682 | } | |
2cd698be YZ |
3683 | cap = new_cap; |
3684 | } else { | |
3685 | if (new_cap) { | |
3686 | ceph_put_cap(mdsc, new_cap); | |
3687 | new_cap = NULL; | |
3688 | } | |
d9df2783 YZ |
3689 | } |
3690 | ||
2cd698be YZ |
3691 | __ceph_caps_issued(ci, &issued); |
3692 | issued |= __ceph_caps_dirty(ci); | |
3693 | ||
3694 | ceph_add_cap(inode, session, cap_id, -1, caps, wanted, seq, mseq, | |
d9df2783 YZ |
3695 | realmino, CEPH_CAP_FLAG_AUTH, &new_cap); |
3696 | ||
2cd698be YZ |
3697 | ocap = peer >= 0 ? __get_cap_for_mds(ci, peer) : NULL; |
3698 | if (ocap && ocap->cap_id == p_cap_id) { | |
4ee6a914 | 3699 | dout(" remove export cap %p mds%d flags %d\n", |
2cd698be | 3700 | ocap, peer, ph->flags); |
4ee6a914 | 3701 | if ((ph->flags & CEPH_CAP_FLAG_AUTH) && |
2cd698be YZ |
3702 | (ocap->seq != le32_to_cpu(ph->seq) || |
3703 | ocap->mseq != le32_to_cpu(ph->mseq))) { | |
d84b37f9 YZ |
3704 | pr_err_ratelimited("handle_cap_import: " |
3705 | "mismatched seq/mseq: ino (%llx.%llx) " | |
3706 | "mds%d seq %d mseq %d importer mds%d " | |
3707 | "has peer seq %d mseq %d\n", | |
3708 | ceph_vinop(inode), peer, ocap->seq, | |
3709 | ocap->mseq, mds, le32_to_cpu(ph->seq), | |
3710 | le32_to_cpu(ph->mseq)); | |
db354052 | 3711 | } |
2cd698be | 3712 | __ceph_remove_cap(ocap, (ph->flags & CEPH_CAP_FLAG_RELEASE)); |
a8599bd8 SW |
3713 | } |
3714 | ||
4ee6a914 | 3715 | /* make sure we re-request max_size, if necessary */ |
4ee6a914 | 3716 | ci->i_requested_max_size = 0; |
d9df2783 | 3717 | |
2cd698be YZ |
3718 | *old_issued = issued; |
3719 | *target_cap = cap; | |
a8599bd8 SW |
3720 | } |
3721 | ||
3722 | /* | |
3723 | * Handle a caps message from the MDS. | |
3724 | * | |
3725 | * Identify the appropriate session, inode, and call the right handler | |
3726 | * based on the cap op. | |
3727 | */ | |
3728 | void ceph_handle_caps(struct ceph_mds_session *session, | |
3729 | struct ceph_msg *msg) | |
3730 | { | |
3731 | struct ceph_mds_client *mdsc = session->s_mdsc; | |
a8599bd8 | 3732 | struct inode *inode; |
be655596 | 3733 | struct ceph_inode_info *ci; |
a8599bd8 SW |
3734 | struct ceph_cap *cap; |
3735 | struct ceph_mds_caps *h; | |
4ee6a914 | 3736 | struct ceph_mds_cap_peer *peer = NULL; |
779fe0fb | 3737 | struct ceph_snap_realm *realm = NULL; |
a1c6b835 | 3738 | int op; |
4985d6f9 | 3739 | int msg_version = le16_to_cpu(msg->hdr.version); |
3d7ded4d | 3740 | u32 seq, mseq; |
a8599bd8 | 3741 | struct ceph_vino vino; |
70edb55b | 3742 | void *snaptrace; |
ce1fbc8d | 3743 | size_t snaptrace_len; |
fb01d1f8 | 3744 | void *p, *end; |
a1c6b835 | 3745 | struct cap_extra_info extra_info = {}; |
a8599bd8 | 3746 | |
a1c6b835 | 3747 | dout("handle_caps from mds%d\n", session->s_mds); |
a8599bd8 SW |
3748 | |
3749 | /* decode */ | |
4ee6a914 | 3750 | end = msg->front.iov_base + msg->front.iov_len; |
a8599bd8 SW |
3751 | if (msg->front.iov_len < sizeof(*h)) |
3752 | goto bad; | |
3753 | h = msg->front.iov_base; | |
3754 | op = le32_to_cpu(h->op); | |
3755 | vino.ino = le64_to_cpu(h->ino); | |
3756 | vino.snap = CEPH_NOSNAP; | |
a8599bd8 | 3757 | seq = le32_to_cpu(h->seq); |
3d7ded4d | 3758 | mseq = le32_to_cpu(h->migrate_seq); |
a8599bd8 | 3759 | |
ce1fbc8d SW |
3760 | snaptrace = h + 1; |
3761 | snaptrace_len = le32_to_cpu(h->snap_trace_len); | |
fb01d1f8 | 3762 | p = snaptrace + snaptrace_len; |
ce1fbc8d | 3763 | |
4985d6f9 | 3764 | if (msg_version >= 2) { |
fb01d1f8 | 3765 | u32 flock_len; |
ce1fbc8d | 3766 | ceph_decode_32_safe(&p, end, flock_len, bad); |
4ee6a914 YZ |
3767 | if (p + flock_len > end) |
3768 | goto bad; | |
fb01d1f8 | 3769 | p += flock_len; |
ce1fbc8d SW |
3770 | } |
3771 | ||
4985d6f9 | 3772 | if (msg_version >= 3) { |
4ee6a914 | 3773 | if (op == CEPH_CAP_OP_IMPORT) { |
4ee6a914 YZ |
3774 | if (p + sizeof(*peer) > end) |
3775 | goto bad; | |
3776 | peer = p; | |
fb01d1f8 | 3777 | p += sizeof(*peer); |
11df2dfb YZ |
3778 | } else if (op == CEPH_CAP_OP_EXPORT) { |
3779 | /* recorded in unused fields */ | |
3780 | peer = (void *)&h->size; | |
4ee6a914 YZ |
3781 | } |
3782 | } | |
3783 | ||
4985d6f9 | 3784 | if (msg_version >= 4) { |
a1c6b835 YZ |
3785 | ceph_decode_64_safe(&p, end, extra_info.inline_version, bad); |
3786 | ceph_decode_32_safe(&p, end, extra_info.inline_len, bad); | |
3787 | if (p + extra_info.inline_len > end) | |
fb01d1f8 | 3788 | goto bad; |
a1c6b835 YZ |
3789 | extra_info.inline_data = p; |
3790 | p += extra_info.inline_len; | |
fb01d1f8 YZ |
3791 | } |
3792 | ||
4985d6f9 | 3793 | if (msg_version >= 5) { |
92475f05 JL |
3794 | struct ceph_osd_client *osdc = &mdsc->fsc->client->osdc; |
3795 | u32 epoch_barrier; | |
3796 | ||
3797 | ceph_decode_32_safe(&p, end, epoch_barrier, bad); | |
3798 | ceph_osdc_update_epoch_barrier(osdc, epoch_barrier); | |
3799 | } | |
3800 | ||
4985d6f9 | 3801 | if (msg_version >= 8) { |
5ea5c5e0 YZ |
3802 | u64 flush_tid; |
3803 | u32 caller_uid, caller_gid; | |
779fe0fb | 3804 | u32 pool_ns_len; |
92475f05 | 3805 | |
5ea5c5e0 YZ |
3806 | /* version >= 6 */ |
3807 | ceph_decode_64_safe(&p, end, flush_tid, bad); | |
3808 | /* version >= 7 */ | |
3809 | ceph_decode_32_safe(&p, end, caller_uid, bad); | |
3810 | ceph_decode_32_safe(&p, end, caller_gid, bad); | |
3811 | /* version >= 8 */ | |
3812 | ceph_decode_32_safe(&p, end, pool_ns_len, bad); | |
779fe0fb YZ |
3813 | if (pool_ns_len > 0) { |
3814 | ceph_decode_need(&p, end, pool_ns_len, bad); | |
a1c6b835 YZ |
3815 | extra_info.pool_ns = |
3816 | ceph_find_or_create_string(p, pool_ns_len); | |
779fe0fb YZ |
3817 | p += pool_ns_len; |
3818 | } | |
5ea5c5e0 YZ |
3819 | } |
3820 | ||
4985d6f9 YZ |
3821 | if (msg_version >= 11) { |
3822 | struct ceph_timespec *btime; | |
3823 | u64 change_attr; | |
3824 | u32 flags; | |
3825 | ||
3826 | /* version >= 9 */ | |
3827 | if (p + sizeof(*btime) > end) | |
3828 | goto bad; | |
3829 | btime = p; | |
3830 | p += sizeof(*btime); | |
3831 | ceph_decode_64_safe(&p, end, change_attr, bad); | |
3832 | /* version >= 10 */ | |
3833 | ceph_decode_32_safe(&p, end, flags, bad); | |
3834 | /* version >= 11 */ | |
3835 | extra_info.dirstat_valid = true; | |
3836 | ceph_decode_64_safe(&p, end, extra_info.nfiles, bad); | |
3837 | ceph_decode_64_safe(&p, end, extra_info.nsubdirs, bad); | |
3838 | } | |
3839 | ||
6cd3bcad | 3840 | /* lookup ino */ |
a1c6b835 | 3841 | inode = ceph_find_inode(mdsc->fsc->sb, vino); |
6cd3bcad YZ |
3842 | ci = ceph_inode(inode); |
3843 | dout(" op %s ino %llx.%llx inode %p\n", ceph_cap_op_name(op), vino.ino, | |
3844 | vino.snap, inode); | |
3845 | ||
a8599bd8 SW |
3846 | mutex_lock(&session->s_mutex); |
3847 | session->s_seq++; | |
3848 | dout(" mds%d seq %lld cap seq %u\n", session->s_mds, session->s_seq, | |
3849 | (unsigned)seq); | |
3850 | ||
a8599bd8 SW |
3851 | if (!inode) { |
3852 | dout(" i don't have ino %llx\n", vino.ino); | |
3d7ded4d | 3853 | |
a096b09a | 3854 | if (op == CEPH_CAP_OP_IMPORT) { |
745a8e3b YZ |
3855 | cap = ceph_get_cap(mdsc, NULL); |
3856 | cap->cap_ino = vino.ino; | |
3857 | cap->queue_release = 1; | |
779fe0fb | 3858 | cap->cap_id = le64_to_cpu(h->cap_id); |
745a8e3b YZ |
3859 | cap->mseq = mseq; |
3860 | cap->seq = seq; | |
dc24de82 | 3861 | cap->issue_seq = seq; |
a096b09a | 3862 | spin_lock(&session->s_cap_lock); |
745a8e3b YZ |
3863 | list_add_tail(&cap->session_caps, |
3864 | &session->s_cap_releases); | |
3865 | session->s_num_cap_releases++; | |
a096b09a YZ |
3866 | spin_unlock(&session->s_cap_lock); |
3867 | } | |
21b559de | 3868 | goto flush_cap_releases; |
a8599bd8 SW |
3869 | } |
3870 | ||
3871 | /* these will work even if we don't have a cap yet */ | |
3872 | switch (op) { | |
3873 | case CEPH_CAP_OP_FLUSHSNAP_ACK: | |
a1c6b835 YZ |
3874 | handle_cap_flushsnap_ack(inode, le64_to_cpu(msg->hdr.tid), |
3875 | h, session); | |
a8599bd8 SW |
3876 | goto done; |
3877 | ||
3878 | case CEPH_CAP_OP_EXPORT: | |
11df2dfb YZ |
3879 | handle_cap_export(inode, h, peer, session); |
3880 | goto done_unlocked; | |
a8599bd8 SW |
3881 | |
3882 | case CEPH_CAP_OP_IMPORT: | |
982d6011 YZ |
3883 | realm = NULL; |
3884 | if (snaptrace_len) { | |
3885 | down_write(&mdsc->snap_rwsem); | |
3886 | ceph_update_snap_trace(mdsc, snaptrace, | |
3887 | snaptrace + snaptrace_len, | |
3888 | false, &realm); | |
3889 | downgrade_write(&mdsc->snap_rwsem); | |
3890 | } else { | |
3891 | down_read(&mdsc->snap_rwsem); | |
3892 | } | |
4ee6a914 | 3893 | handle_cap_import(mdsc, inode, h, peer, session, |
a1c6b835 YZ |
3894 | &cap, &extra_info.issued); |
3895 | handle_cap_grant(inode, session, cap, | |
3896 | h, msg->middle, &extra_info); | |
982d6011 YZ |
3897 | if (realm) |
3898 | ceph_put_snap_realm(mdsc, realm); | |
2cd698be | 3899 | goto done_unlocked; |
a8599bd8 SW |
3900 | } |
3901 | ||
3902 | /* the rest require a cap */ | |
be655596 | 3903 | spin_lock(&ci->i_ceph_lock); |
a1c6b835 | 3904 | cap = __get_cap_for_mds(ceph_inode(inode), session->s_mds); |
a8599bd8 | 3905 | if (!cap) { |
9dbd412f | 3906 | dout(" no cap on %p ino %llx.%llx from mds%d\n", |
a1c6b835 YZ |
3907 | inode, ceph_ino(inode), ceph_snap(inode), |
3908 | session->s_mds); | |
be655596 | 3909 | spin_unlock(&ci->i_ceph_lock); |
21b559de | 3910 | goto flush_cap_releases; |
a8599bd8 SW |
3911 | } |
3912 | ||
be655596 | 3913 | /* note that each of these drops i_ceph_lock for us */ |
a8599bd8 SW |
3914 | switch (op) { |
3915 | case CEPH_CAP_OP_REVOKE: | |
3916 | case CEPH_CAP_OP_GRANT: | |
a1c6b835 YZ |
3917 | __ceph_caps_issued(ci, &extra_info.issued); |
3918 | extra_info.issued |= __ceph_caps_dirty(ci); | |
3919 | handle_cap_grant(inode, session, cap, | |
3920 | h, msg->middle, &extra_info); | |
15637c8b | 3921 | goto done_unlocked; |
a8599bd8 SW |
3922 | |
3923 | case CEPH_CAP_OP_FLUSH_ACK: | |
a1c6b835 YZ |
3924 | handle_cap_flush_ack(inode, le64_to_cpu(msg->hdr.tid), |
3925 | h, session, cap); | |
a8599bd8 SW |
3926 | break; |
3927 | ||
3928 | case CEPH_CAP_OP_TRUNC: | |
3929 | handle_cap_trunc(inode, h, session); | |
3930 | break; | |
3931 | ||
3932 | default: | |
be655596 | 3933 | spin_unlock(&ci->i_ceph_lock); |
a8599bd8 SW |
3934 | pr_err("ceph_handle_caps: unknown cap op %d %s\n", op, |
3935 | ceph_cap_op_name(op)); | |
3936 | } | |
3937 | ||
21b559de GF |
3938 | goto done; |
3939 | ||
3940 | flush_cap_releases: | |
3941 | /* | |
745a8e3b | 3942 | * send any cap release message to try to move things |
21b559de GF |
3943 | * along for the mds (who clearly thinks we still have this |
3944 | * cap). | |
3945 | */ | |
21b559de GF |
3946 | ceph_send_cap_releases(mdsc, session); |
3947 | ||
a8599bd8 | 3948 | done: |
15637c8b SW |
3949 | mutex_unlock(&session->s_mutex); |
3950 | done_unlocked: | |
e96a650a | 3951 | iput(inode); |
a1c6b835 | 3952 | ceph_put_string(extra_info.pool_ns); |
a8599bd8 SW |
3953 | return; |
3954 | ||
3955 | bad: | |
3956 | pr_err("ceph_handle_caps: corrupt message\n"); | |
9ec7cab1 | 3957 | ceph_msg_dump(msg); |
a8599bd8 SW |
3958 | return; |
3959 | } | |
3960 | ||
3961 | /* | |
3962 | * Delayed work handler to process end of delayed cap release LRU list. | |
3963 | */ | |
afcdaea3 | 3964 | void ceph_check_delayed_caps(struct ceph_mds_client *mdsc) |
a8599bd8 | 3965 | { |
4b9f2042 | 3966 | struct inode *inode; |
a8599bd8 SW |
3967 | struct ceph_inode_info *ci; |
3968 | int flags = CHECK_CAPS_NODELAY; | |
3969 | ||
a8599bd8 SW |
3970 | dout("check_delayed_caps\n"); |
3971 | while (1) { | |
3972 | spin_lock(&mdsc->cap_delay_lock); | |
3973 | if (list_empty(&mdsc->cap_delay_list)) | |
3974 | break; | |
3975 | ci = list_first_entry(&mdsc->cap_delay_list, | |
3976 | struct ceph_inode_info, | |
3977 | i_cap_delay_list); | |
3978 | if ((ci->i_ceph_flags & CEPH_I_FLUSH) == 0 && | |
3979 | time_before(jiffies, ci->i_hold_caps_max)) | |
3980 | break; | |
3981 | list_del_init(&ci->i_cap_delay_list); | |
4b9f2042 YZ |
3982 | |
3983 | inode = igrab(&ci->vfs_inode); | |
a8599bd8 | 3984 | spin_unlock(&mdsc->cap_delay_lock); |
4b9f2042 YZ |
3985 | |
3986 | if (inode) { | |
3987 | dout("check_delayed_caps on %p\n", inode); | |
3988 | ceph_check_caps(ci, flags, NULL); | |
3989 | iput(inode); | |
3990 | } | |
a8599bd8 SW |
3991 | } |
3992 | spin_unlock(&mdsc->cap_delay_lock); | |
3993 | } | |
3994 | ||
afcdaea3 SW |
3995 | /* |
3996 | * Flush all dirty caps to the mds | |
3997 | */ | |
3998 | void ceph_flush_dirty_caps(struct ceph_mds_client *mdsc) | |
3999 | { | |
db354052 SW |
4000 | struct ceph_inode_info *ci; |
4001 | struct inode *inode; | |
afcdaea3 SW |
4002 | |
4003 | dout("flush_dirty_caps\n"); | |
4004 | spin_lock(&mdsc->cap_dirty_lock); | |
db354052 SW |
4005 | while (!list_empty(&mdsc->cap_dirty)) { |
4006 | ci = list_first_entry(&mdsc->cap_dirty, struct ceph_inode_info, | |
4007 | i_dirty_item); | |
70b666c3 SW |
4008 | inode = &ci->vfs_inode; |
4009 | ihold(inode); | |
db354052 | 4010 | dout("flush_dirty_caps %p\n", inode); |
afcdaea3 | 4011 | spin_unlock(&mdsc->cap_dirty_lock); |
70b666c3 SW |
4012 | ceph_check_caps(ci, CHECK_CAPS_NODELAY|CHECK_CAPS_FLUSH, NULL); |
4013 | iput(inode); | |
afcdaea3 SW |
4014 | spin_lock(&mdsc->cap_dirty_lock); |
4015 | } | |
4016 | spin_unlock(&mdsc->cap_dirty_lock); | |
db354052 | 4017 | dout("flush_dirty_caps done\n"); |
afcdaea3 SW |
4018 | } |
4019 | ||
774a6a11 YZ |
4020 | void __ceph_get_fmode(struct ceph_inode_info *ci, int fmode) |
4021 | { | |
4022 | int i; | |
4023 | int bits = (fmode << 1) | 1; | |
4024 | for (i = 0; i < CEPH_FILE_MODE_BITS; i++) { | |
4025 | if (bits & (1 << i)) | |
4026 | ci->i_nr_by_mode[i]++; | |
4027 | } | |
4028 | } | |
4029 | ||
a8599bd8 SW |
4030 | /* |
4031 | * Drop open file reference. If we were the last open file, | |
4032 | * we may need to release capabilities to the MDS (or schedule | |
4033 | * their delayed release). | |
4034 | */ | |
4035 | void ceph_put_fmode(struct ceph_inode_info *ci, int fmode) | |
4036 | { | |
774a6a11 YZ |
4037 | int i, last = 0; |
4038 | int bits = (fmode << 1) | 1; | |
be655596 | 4039 | spin_lock(&ci->i_ceph_lock); |
774a6a11 YZ |
4040 | for (i = 0; i < CEPH_FILE_MODE_BITS; i++) { |
4041 | if (bits & (1 << i)) { | |
4042 | BUG_ON(ci->i_nr_by_mode[i] == 0); | |
4043 | if (--ci->i_nr_by_mode[i] == 0) | |
4044 | last++; | |
4045 | } | |
4046 | } | |
4047 | dout("put_fmode %p fmode %d {%d,%d,%d,%d}\n", | |
4048 | &ci->vfs_inode, fmode, | |
4049 | ci->i_nr_by_mode[0], ci->i_nr_by_mode[1], | |
4050 | ci->i_nr_by_mode[2], ci->i_nr_by_mode[3]); | |
be655596 | 4051 | spin_unlock(&ci->i_ceph_lock); |
a8599bd8 SW |
4052 | |
4053 | if (last && ci->i_vino.snap == CEPH_NOSNAP) | |
4054 | ceph_check_caps(ci, 0, NULL); | |
4055 | } | |
4056 | ||
6ef0bc6d ZZ |
4057 | /* |
4058 | * For a soon-to-be unlinked file, drop the AUTH_RDCACHE caps. If it | |
4059 | * looks like the link count will hit 0, drop any other caps (other | |
4060 | * than PIN) we don't specifically want (due to the file still being | |
4061 | * open). | |
4062 | */ | |
4063 | int ceph_drop_caps_for_unlink(struct inode *inode) | |
4064 | { | |
4065 | struct ceph_inode_info *ci = ceph_inode(inode); | |
4066 | int drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL; | |
4067 | ||
4068 | spin_lock(&ci->i_ceph_lock); | |
4069 | if (inode->i_nlink == 1) { | |
4070 | drop |= ~(__ceph_caps_wanted(ci) | CEPH_CAP_PIN); | |
4071 | ||
4072 | ci->i_ceph_flags |= CEPH_I_NODELAY; | |
4073 | if (__ceph_caps_dirty(ci)) { | |
4074 | struct ceph_mds_client *mdsc = | |
4075 | ceph_inode_to_client(inode)->mdsc; | |
4076 | __cap_delay_requeue_front(mdsc, ci); | |
4077 | } | |
4078 | } | |
4079 | spin_unlock(&ci->i_ceph_lock); | |
4080 | return drop; | |
4081 | } | |
4082 | ||
a8599bd8 SW |
4083 | /* |
4084 | * Helpers for embedding cap and dentry lease releases into mds | |
4085 | * requests. | |
4086 | * | |
4087 | * @force is used by dentry_release (below) to force inclusion of a | |
4088 | * record for the directory inode, even when there aren't any caps to | |
4089 | * drop. | |
4090 | */ | |
4091 | int ceph_encode_inode_release(void **p, struct inode *inode, | |
4092 | int mds, int drop, int unless, int force) | |
4093 | { | |
4094 | struct ceph_inode_info *ci = ceph_inode(inode); | |
4095 | struct ceph_cap *cap; | |
4096 | struct ceph_mds_request_release *rel = *p; | |
ec97f88b | 4097 | int used, dirty; |
a8599bd8 | 4098 | int ret = 0; |
a8599bd8 | 4099 | |
be655596 | 4100 | spin_lock(&ci->i_ceph_lock); |
916623da | 4101 | used = __ceph_caps_used(ci); |
ec97f88b | 4102 | dirty = __ceph_caps_dirty(ci); |
916623da | 4103 | |
ec97f88b SW |
4104 | dout("encode_inode_release %p mds%d used|dirty %s drop %s unless %s\n", |
4105 | inode, mds, ceph_cap_string(used|dirty), ceph_cap_string(drop), | |
916623da SW |
4106 | ceph_cap_string(unless)); |
4107 | ||
ec97f88b SW |
4108 | /* only drop unused, clean caps */ |
4109 | drop &= ~(used | dirty); | |
916623da | 4110 | |
a8599bd8 SW |
4111 | cap = __get_cap_for_mds(ci, mds); |
4112 | if (cap && __cap_is_valid(cap)) { | |
222b7f90 YZ |
4113 | unless &= cap->issued; |
4114 | if (unless) { | |
4115 | if (unless & CEPH_CAP_AUTH_EXCL) | |
4116 | drop &= ~CEPH_CAP_AUTH_SHARED; | |
4117 | if (unless & CEPH_CAP_LINK_EXCL) | |
4118 | drop &= ~CEPH_CAP_LINK_SHARED; | |
4119 | if (unless & CEPH_CAP_XATTR_EXCL) | |
4120 | drop &= ~CEPH_CAP_XATTR_SHARED; | |
4121 | if (unless & CEPH_CAP_FILE_EXCL) | |
4122 | drop &= ~CEPH_CAP_FILE_SHARED; | |
4123 | } | |
4124 | ||
4125 | if (force || (cap->issued & drop)) { | |
4126 | if (cap->issued & drop) { | |
bb137f84 YZ |
4127 | int wanted = __ceph_caps_wanted(ci); |
4128 | if ((ci->i_ceph_flags & CEPH_I_NODELAY) == 0) | |
4129 | wanted |= cap->mds_wanted; | |
4130 | dout("encode_inode_release %p cap %p " | |
4131 | "%s -> %s, wanted %s -> %s\n", inode, cap, | |
a8599bd8 | 4132 | ceph_cap_string(cap->issued), |
bb137f84 YZ |
4133 | ceph_cap_string(cap->issued & ~drop), |
4134 | ceph_cap_string(cap->mds_wanted), | |
4135 | ceph_cap_string(wanted)); | |
4136 | ||
a8599bd8 SW |
4137 | cap->issued &= ~drop; |
4138 | cap->implemented &= ~drop; | |
bb137f84 | 4139 | cap->mds_wanted = wanted; |
a8599bd8 SW |
4140 | } else { |
4141 | dout("encode_inode_release %p cap %p %s" | |
4142 | " (force)\n", inode, cap, | |
4143 | ceph_cap_string(cap->issued)); | |
4144 | } | |
4145 | ||
4146 | rel->ino = cpu_to_le64(ceph_ino(inode)); | |
4147 | rel->cap_id = cpu_to_le64(cap->cap_id); | |
4148 | rel->seq = cpu_to_le32(cap->seq); | |
08a0f24e | 4149 | rel->issue_seq = cpu_to_le32(cap->issue_seq); |
a8599bd8 | 4150 | rel->mseq = cpu_to_le32(cap->mseq); |
fd7b95cd | 4151 | rel->caps = cpu_to_le32(cap->implemented); |
a8599bd8 SW |
4152 | rel->wanted = cpu_to_le32(cap->mds_wanted); |
4153 | rel->dname_len = 0; | |
4154 | rel->dname_seq = 0; | |
4155 | *p += sizeof(*rel); | |
4156 | ret = 1; | |
4157 | } else { | |
222b7f90 | 4158 | dout("encode_inode_release %p cap %p %s (noop)\n", |
a8599bd8 SW |
4159 | inode, cap, ceph_cap_string(cap->issued)); |
4160 | } | |
4161 | } | |
be655596 | 4162 | spin_unlock(&ci->i_ceph_lock); |
a8599bd8 SW |
4163 | return ret; |
4164 | } | |
4165 | ||
4166 | int ceph_encode_dentry_release(void **p, struct dentry *dentry, | |
ca6c8ae0 | 4167 | struct inode *dir, |
a8599bd8 SW |
4168 | int mds, int drop, int unless) |
4169 | { | |
ca6c8ae0 | 4170 | struct dentry *parent = NULL; |
a8599bd8 SW |
4171 | struct ceph_mds_request_release *rel = *p; |
4172 | struct ceph_dentry_info *di = ceph_dentry(dentry); | |
4173 | int force = 0; | |
4174 | int ret; | |
4175 | ||
4176 | /* | |
4177 | * force an record for the directory caps if we have a dentry lease. | |
be655596 | 4178 | * this is racy (can't take i_ceph_lock and d_lock together), but it |
a8599bd8 SW |
4179 | * doesn't have to be perfect; the mds will revoke anything we don't |
4180 | * release. | |
4181 | */ | |
4182 | spin_lock(&dentry->d_lock); | |
4183 | if (di->lease_session && di->lease_session->s_mds == mds) | |
4184 | force = 1; | |
ca6c8ae0 JL |
4185 | if (!dir) { |
4186 | parent = dget(dentry->d_parent); | |
4187 | dir = d_inode(parent); | |
4188 | } | |
a8599bd8 SW |
4189 | spin_unlock(&dentry->d_lock); |
4190 | ||
ca6c8ae0 | 4191 | ret = ceph_encode_inode_release(p, dir, mds, drop, unless, force); |
adf0d687 | 4192 | dput(parent); |
a8599bd8 SW |
4193 | |
4194 | spin_lock(&dentry->d_lock); | |
4195 | if (ret && di->lease_session && di->lease_session->s_mds == mds) { | |
4196 | dout("encode_dentry_release %p mds%d seq %d\n", | |
4197 | dentry, mds, (int)di->lease_seq); | |
4198 | rel->dname_len = cpu_to_le32(dentry->d_name.len); | |
4199 | memcpy(*p, dentry->d_name.name, dentry->d_name.len); | |
4200 | *p += dentry->d_name.len; | |
4201 | rel->dname_seq = cpu_to_le32(di->lease_seq); | |
1dadcce3 | 4202 | __ceph_mdsc_drop_dentry_lease(dentry); |
a8599bd8 SW |
4203 | } |
4204 | spin_unlock(&dentry->d_lock); | |
4205 | return ret; | |
4206 | } |