1#include <linux/ceph/ceph_debug.h>
2
3#include <linux/fs.h>
4#include <linux/kernel.h>
5#include <linux/sched.h>
6#include <linux/slab.h>
7#include <linux/vmalloc.h>
8#include <linux/wait.h>
9#include <linux/writeback.h>
10
11#include "super.h"
12#include "mds_client.h"
13#include "cache.h"
14#include <linux/ceph/decode.h>
15#include <linux/ceph/messenger.h>
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47#define MAX_CAP_STR 20
48static char cap_str[MAX_CAP_STR][40];
49static DEFINE_SPINLOCK(cap_str_lock);
50static int last_cap_str;
51
52static char *gcap_string(char *s, int c)
53{
54 if (c & CEPH_CAP_GSHARED)
55 *s++ = 's';
56 if (c & CEPH_CAP_GEXCL)
57 *s++ = 'x';
58 if (c & CEPH_CAP_GCACHE)
59 *s++ = 'c';
60 if (c & CEPH_CAP_GRD)
61 *s++ = 'r';
62 if (c & CEPH_CAP_GWR)
63 *s++ = 'w';
64 if (c & CEPH_CAP_GBUFFER)
65 *s++ = 'b';
66 if (c & CEPH_CAP_GLAZYIO)
67 *s++ = 'l';
68 return s;
69}
70
71const char *ceph_cap_string(int caps)
72{
73 int i;
74 char *s;
75 int c;
76
77 spin_lock(&cap_str_lock);
78 i = last_cap_str++;
79 if (last_cap_str == MAX_CAP_STR)
80 last_cap_str = 0;
81 spin_unlock(&cap_str_lock);
82
83 s = cap_str[i];
84
85 if (caps & CEPH_CAP_PIN)
86 *s++ = 'p';
87
88 c = (caps >> CEPH_CAP_SAUTH) & 3;
89 if (c) {
90 *s++ = 'A';
91 s = gcap_string(s, c);
92 }
93
94 c = (caps >> CEPH_CAP_SLINK) & 3;
95 if (c) {
96 *s++ = 'L';
97 s = gcap_string(s, c);
98 }
99
100 c = (caps >> CEPH_CAP_SXATTR) & 3;
101 if (c) {
102 *s++ = 'X';
103 s = gcap_string(s, c);
104 }
105
106 c = caps >> CEPH_CAP_SFILE;
107 if (c) {
108 *s++ = 'F';
109 s = gcap_string(s, c);
110 }
111
112 if (s == cap_str[i])
113 *s++ = '-';
114 *s = 0;
115 return cap_str[i];
116}
117
118void ceph_caps_init(struct ceph_mds_client *mdsc)
119{
120 INIT_LIST_HEAD(&mdsc->caps_list);
121 spin_lock_init(&mdsc->caps_list_lock);
122}
123
124void ceph_caps_finalize(struct ceph_mds_client *mdsc)
125{
126 struct ceph_cap *cap;
127
128 spin_lock(&mdsc->caps_list_lock);
129 while (!list_empty(&mdsc->caps_list)) {
130 cap = list_first_entry(&mdsc->caps_list,
131 struct ceph_cap, caps_item);
132 list_del(&cap->caps_item);
133 kmem_cache_free(ceph_cap_cachep, cap);
134 }
135 mdsc->caps_total_count = 0;
136 mdsc->caps_avail_count = 0;
137 mdsc->caps_use_count = 0;
138 mdsc->caps_reserve_count = 0;
139 mdsc->caps_min_count = 0;
140 spin_unlock(&mdsc->caps_list_lock);
141}
142
143void ceph_adjust_min_caps(struct ceph_mds_client *mdsc, int delta)
144{
145 spin_lock(&mdsc->caps_list_lock);
146 mdsc->caps_min_count += delta;
147 BUG_ON(mdsc->caps_min_count < 0);
148 spin_unlock(&mdsc->caps_list_lock);
149}
150
151void ceph_reserve_caps(struct ceph_mds_client *mdsc,
152 struct ceph_cap_reservation *ctx, int need)
153{
154 int i;
155 struct ceph_cap *cap;
156 int have;
157 int alloc = 0;
158 LIST_HEAD(newcaps);
159
160 dout("reserve caps ctx=%p need=%d\n", ctx, need);
161
162
163 spin_lock(&mdsc->caps_list_lock);
164 if (mdsc->caps_avail_count >= need)
165 have = need;
166 else
167 have = mdsc->caps_avail_count;
168 mdsc->caps_avail_count -= have;
169 mdsc->caps_reserve_count += have;
170 BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count +
171 mdsc->caps_reserve_count +
172 mdsc->caps_avail_count);
173 spin_unlock(&mdsc->caps_list_lock);
174
175 for (i = have; i < need; i++) {
176 cap = kmem_cache_alloc(ceph_cap_cachep, GFP_NOFS);
177 if (!cap)
178 break;
179 list_add(&cap->caps_item, &newcaps);
180 alloc++;
181 }
182
183 if (have + alloc != need)
184 pr_warn("reserve caps ctx=%p ENOMEM need=%d got=%d\n",
185 ctx, need, have + alloc);
186
187 spin_lock(&mdsc->caps_list_lock);
188 mdsc->caps_total_count += alloc;
189 mdsc->caps_reserve_count += alloc;
190 list_splice(&newcaps, &mdsc->caps_list);
191
192 BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count +
193 mdsc->caps_reserve_count +
194 mdsc->caps_avail_count);
195 spin_unlock(&mdsc->caps_list_lock);
196
197 ctx->count = need;
198 dout("reserve caps ctx=%p %d = %d used + %d resv + %d avail\n",
199 ctx, mdsc->caps_total_count, mdsc->caps_use_count,
200 mdsc->caps_reserve_count, mdsc->caps_avail_count);
201}
202
203int ceph_unreserve_caps(struct ceph_mds_client *mdsc,
204 struct ceph_cap_reservation *ctx)
205{
206 dout("unreserve caps ctx=%p count=%d\n", ctx, ctx->count);
207 if (ctx->count) {
208 spin_lock(&mdsc->caps_list_lock);
209 BUG_ON(mdsc->caps_reserve_count < ctx->count);
210 mdsc->caps_reserve_count -= ctx->count;
211 mdsc->caps_avail_count += ctx->count;
212 ctx->count = 0;
213 dout("unreserve caps %d = %d used + %d resv + %d avail\n",
214 mdsc->caps_total_count, mdsc->caps_use_count,
215 mdsc->caps_reserve_count, mdsc->caps_avail_count);
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);
220 }
221 return 0;
222}
223
224static struct ceph_cap *get_cap(struct ceph_mds_client *mdsc,
225 struct ceph_cap_reservation *ctx)
226{
227 struct ceph_cap *cap = NULL;
228
229
230 if (!ctx) {
231 cap = kmem_cache_alloc(ceph_cap_cachep, GFP_NOFS);
232 if (cap) {
233 spin_lock(&mdsc->caps_list_lock);
234 mdsc->caps_use_count++;
235 mdsc->caps_total_count++;
236 spin_unlock(&mdsc->caps_list_lock);
237 }
238 return cap;
239 }
240
241 spin_lock(&mdsc->caps_list_lock);
242 dout("get_cap ctx=%p (%d) %d = %d used + %d resv + %d avail\n",
243 ctx, ctx->count, mdsc->caps_total_count, mdsc->caps_use_count,
244 mdsc->caps_reserve_count, mdsc->caps_avail_count);
245 BUG_ON(!ctx->count);
246 BUG_ON(ctx->count > mdsc->caps_reserve_count);
247 BUG_ON(list_empty(&mdsc->caps_list));
248
249 ctx->count--;
250 mdsc->caps_reserve_count--;
251 mdsc->caps_use_count++;
252
253 cap = list_first_entry(&mdsc->caps_list, struct ceph_cap, caps_item);
254 list_del(&cap->caps_item);
255
256 BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count +
257 mdsc->caps_reserve_count + mdsc->caps_avail_count);
258 spin_unlock(&mdsc->caps_list_lock);
259 return cap;
260}
261
262void ceph_put_cap(struct ceph_mds_client *mdsc, struct ceph_cap *cap)
263{
264 spin_lock(&mdsc->caps_list_lock);
265 dout("put_cap %p %d = %d used + %d resv + %d avail\n",
266 cap, mdsc->caps_total_count, mdsc->caps_use_count,
267 mdsc->caps_reserve_count, mdsc->caps_avail_count);
268 mdsc->caps_use_count--;
269
270
271
272
273 if (mdsc->caps_avail_count >= mdsc->caps_reserve_count +
274 mdsc->caps_min_count) {
275 mdsc->caps_total_count--;
276 kmem_cache_free(ceph_cap_cachep, cap);
277 } else {
278 mdsc->caps_avail_count++;
279 list_add(&cap->caps_item, &mdsc->caps_list);
280 }
281
282 BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count +
283 mdsc->caps_reserve_count + mdsc->caps_avail_count);
284 spin_unlock(&mdsc->caps_list_lock);
285}
286
287void ceph_reservation_status(struct ceph_fs_client *fsc,
288 int *total, int *avail, int *used, int *reserved,
289 int *min)
290{
291 struct ceph_mds_client *mdsc = fsc->mdsc;
292
293 if (total)
294 *total = mdsc->caps_total_count;
295 if (avail)
296 *avail = mdsc->caps_avail_count;
297 if (used)
298 *used = mdsc->caps_use_count;
299 if (reserved)
300 *reserved = mdsc->caps_reserve_count;
301 if (min)
302 *min = mdsc->caps_min_count;
303}
304
305
306
307
308
309
310static struct ceph_cap *__get_cap_for_mds(struct ceph_inode_info *ci, int mds)
311{
312 struct ceph_cap *cap;
313 struct rb_node *n = ci->i_caps.rb_node;
314
315 while (n) {
316 cap = rb_entry(n, struct ceph_cap, ci_node);
317 if (mds < cap->mds)
318 n = n->rb_left;
319 else if (mds > cap->mds)
320 n = n->rb_right;
321 else
322 return cap;
323 }
324 return NULL;
325}
326
327struct ceph_cap *ceph_get_cap_for_mds(struct ceph_inode_info *ci, int mds)
328{
329 struct ceph_cap *cap;
330
331 spin_lock(&ci->i_ceph_lock);
332 cap = __get_cap_for_mds(ci, mds);
333 spin_unlock(&ci->i_ceph_lock);
334 return cap;
335}
336
337
338
339
340static int __ceph_get_cap_mds(struct ceph_inode_info *ci)
341{
342 struct ceph_cap *cap;
343 int mds = -1;
344 struct rb_node *p;
345
346
347 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
348 cap = rb_entry(p, struct ceph_cap, ci_node);
349 mds = cap->mds;
350 if (cap->issued & (CEPH_CAP_FILE_WR |
351 CEPH_CAP_FILE_BUFFER |
352 CEPH_CAP_FILE_EXCL))
353 break;
354 }
355 return mds;
356}
357
358int ceph_get_cap_mds(struct inode *inode)
359{
360 struct ceph_inode_info *ci = ceph_inode(inode);
361 int mds;
362 spin_lock(&ci->i_ceph_lock);
363 mds = __ceph_get_cap_mds(ceph_inode(inode));
364 spin_unlock(&ci->i_ceph_lock);
365 return mds;
366}
367
368
369
370
371static void __insert_cap_node(struct ceph_inode_info *ci,
372 struct ceph_cap *new)
373{
374 struct rb_node **p = &ci->i_caps.rb_node;
375 struct rb_node *parent = NULL;
376 struct ceph_cap *cap = NULL;
377
378 while (*p) {
379 parent = *p;
380 cap = rb_entry(parent, struct ceph_cap, ci_node);
381 if (new->mds < cap->mds)
382 p = &(*p)->rb_left;
383 else if (new->mds > cap->mds)
384 p = &(*p)->rb_right;
385 else
386 BUG();
387 }
388
389 rb_link_node(&new->ci_node, parent, p);
390 rb_insert_color(&new->ci_node, &ci->i_caps);
391}
392
393
394
395
396
397static void __cap_set_timeouts(struct ceph_mds_client *mdsc,
398 struct ceph_inode_info *ci)
399{
400 struct ceph_mount_options *ma = mdsc->fsc->mount_options;
401
402 ci->i_hold_caps_min = round_jiffies(jiffies +
403 ma->caps_wanted_delay_min * HZ);
404 ci->i_hold_caps_max = round_jiffies(jiffies +
405 ma->caps_wanted_delay_max * HZ);
406 dout("__cap_set_timeouts %p min %lu max %lu\n", &ci->vfs_inode,
407 ci->i_hold_caps_min - jiffies, ci->i_hold_caps_max - jiffies);
408}
409
410
411
412
413
414
415
416
417
418static void __cap_delay_requeue(struct ceph_mds_client *mdsc,
419 struct ceph_inode_info *ci)
420{
421 __cap_set_timeouts(mdsc, ci);
422 dout("__cap_delay_requeue %p flags %d at %lu\n", &ci->vfs_inode,
423 ci->i_ceph_flags, ci->i_hold_caps_max);
424 if (!mdsc->stopping) {
425 spin_lock(&mdsc->cap_delay_lock);
426 if (!list_empty(&ci->i_cap_delay_list)) {
427 if (ci->i_ceph_flags & CEPH_I_FLUSH)
428 goto no_change;
429 list_del_init(&ci->i_cap_delay_list);
430 }
431 list_add_tail(&ci->i_cap_delay_list, &mdsc->cap_delay_list);
432no_change:
433 spin_unlock(&mdsc->cap_delay_lock);
434 }
435}
436
437
438
439
440
441
442static void __cap_delay_requeue_front(struct ceph_mds_client *mdsc,
443 struct ceph_inode_info *ci)
444{
445 dout("__cap_delay_requeue_front %p\n", &ci->vfs_inode);
446 spin_lock(&mdsc->cap_delay_lock);
447 ci->i_ceph_flags |= CEPH_I_FLUSH;
448 if (!list_empty(&ci->i_cap_delay_list))
449 list_del_init(&ci->i_cap_delay_list);
450 list_add(&ci->i_cap_delay_list, &mdsc->cap_delay_list);
451 spin_unlock(&mdsc->cap_delay_lock);
452}
453
454
455
456
457
458
459static void __cap_delay_cancel(struct ceph_mds_client *mdsc,
460 struct ceph_inode_info *ci)
461{
462 dout("__cap_delay_cancel %p\n", &ci->vfs_inode);
463 if (list_empty(&ci->i_cap_delay_list))
464 return;
465 spin_lock(&mdsc->cap_delay_lock);
466 list_del_init(&ci->i_cap_delay_list);
467 spin_unlock(&mdsc->cap_delay_lock);
468}
469
470
471
472
473static void __check_cap_issue(struct ceph_inode_info *ci, struct ceph_cap *cap,
474 unsigned issued)
475{
476 unsigned had = __ceph_caps_issued(ci, NULL);
477
478
479
480
481
482 if ((issued & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) &&
483 (had & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) == 0) {
484 ci->i_rdcache_gen++;
485 }
486
487
488
489
490
491
492 if ((issued & CEPH_CAP_FILE_SHARED) &&
493 (had & CEPH_CAP_FILE_SHARED) == 0) {
494 ci->i_shared_gen++;
495 if (S_ISDIR(ci->vfs_inode.i_mode)) {
496 dout(" marking %p NOT complete\n", &ci->vfs_inode);
497 __ceph_dir_clear_complete(ci);
498 }
499 }
500}
501
502
503
504
505
506
507
508
509
510
511int ceph_add_cap(struct inode *inode,
512 struct ceph_mds_session *session, u64 cap_id,
513 int fmode, unsigned issued, unsigned wanted,
514 unsigned seq, unsigned mseq, u64 realmino, int flags,
515 struct ceph_cap_reservation *caps_reservation)
516{
517 struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
518 struct ceph_inode_info *ci = ceph_inode(inode);
519 struct ceph_cap *new_cap = NULL;
520 struct ceph_cap *cap;
521 int mds = session->s_mds;
522 int actual_wanted;
523
524 dout("add_cap %p mds%d cap %llx %s seq %d\n", inode,
525 session->s_mds, cap_id, ceph_cap_string(issued), seq);
526
527
528
529
530
531 if (fmode >= 0)
532 wanted |= ceph_caps_for_mode(fmode);
533
534retry:
535 spin_lock(&ci->i_ceph_lock);
536 cap = __get_cap_for_mds(ci, mds);
537 if (!cap) {
538 if (new_cap) {
539 cap = new_cap;
540 new_cap = NULL;
541 } else {
542 spin_unlock(&ci->i_ceph_lock);
543 new_cap = get_cap(mdsc, caps_reservation);
544 if (new_cap == NULL)
545 return -ENOMEM;
546 goto retry;
547 }
548
549 cap->issued = 0;
550 cap->implemented = 0;
551 cap->mds = mds;
552 cap->mds_wanted = 0;
553 cap->mseq = 0;
554
555 cap->ci = ci;
556 __insert_cap_node(ci, cap);
557
558
559 if (ci->i_cap_exporting_mds == mds) {
560 ci->i_cap_exporting_issued = 0;
561 ci->i_cap_exporting_mseq = 0;
562 ci->i_cap_exporting_mds = -1;
563 }
564
565
566 cap->session = session;
567 spin_lock(&session->s_cap_lock);
568 list_add_tail(&cap->session_caps, &session->s_caps);
569 session->s_nr_caps++;
570 spin_unlock(&session->s_cap_lock);
571 } else if (new_cap)
572 ceph_put_cap(mdsc, new_cap);
573
574 if (!ci->i_snap_realm) {
575
576
577
578 struct ceph_snap_realm *realm = ceph_lookup_snap_realm(mdsc,
579 realmino);
580 if (realm) {
581 ceph_get_snap_realm(mdsc, realm);
582 spin_lock(&realm->inodes_with_caps_lock);
583 ci->i_snap_realm = realm;
584 list_add(&ci->i_snap_realm_item,
585 &realm->inodes_with_caps);
586 spin_unlock(&realm->inodes_with_caps_lock);
587 } else {
588 pr_err("ceph_add_cap: couldn't find snap realm %llx\n",
589 realmino);
590 WARN_ON(!realm);
591 }
592 }
593
594 __check_cap_issue(ci, cap, issued);
595
596
597
598
599
600
601 actual_wanted = __ceph_caps_wanted(ci);
602 if ((wanted & ~actual_wanted) ||
603 (issued & ~actual_wanted & CEPH_CAP_ANY_WR)) {
604 dout(" issued %s, mds wanted %s, actual %s, queueing\n",
605 ceph_cap_string(issued), ceph_cap_string(wanted),
606 ceph_cap_string(actual_wanted));
607 __cap_delay_requeue(mdsc, ci);
608 }
609
610 if (flags & CEPH_CAP_FLAG_AUTH) {
611 if (ci->i_auth_cap == NULL ||
612 ceph_seq_cmp(ci->i_auth_cap->mseq, mseq) < 0)
613 ci->i_auth_cap = cap;
614 } else if (ci->i_auth_cap == cap) {
615 ci->i_auth_cap = NULL;
616 spin_lock(&mdsc->cap_dirty_lock);
617 if (!list_empty(&ci->i_dirty_item)) {
618 dout(" moving %p to cap_dirty_migrating\n", inode);
619 list_move(&ci->i_dirty_item,
620 &mdsc->cap_dirty_migrating);
621 }
622 spin_unlock(&mdsc->cap_dirty_lock);
623 }
624
625 dout("add_cap inode %p (%llx.%llx) cap %p %s now %s seq %d mds%d\n",
626 inode, ceph_vinop(inode), cap, ceph_cap_string(issued),
627 ceph_cap_string(issued|cap->issued), seq, mds);
628 cap->cap_id = cap_id;
629 cap->issued = issued;
630 cap->implemented |= issued;
631 if (mseq > cap->mseq)
632 cap->mds_wanted = wanted;
633 else
634 cap->mds_wanted |= wanted;
635 cap->seq = seq;
636 cap->issue_seq = seq;
637 cap->mseq = mseq;
638 cap->cap_gen = session->s_cap_gen;
639
640 if (fmode >= 0)
641 __ceph_get_fmode(ci, fmode);
642 spin_unlock(&ci->i_ceph_lock);
643 wake_up_all(&ci->i_cap_wq);
644 return 0;
645}
646
647
648
649
650
651
652static int __cap_is_valid(struct ceph_cap *cap)
653{
654 unsigned long ttl;
655 u32 gen;
656
657 spin_lock(&cap->session->s_gen_ttl_lock);
658 gen = cap->session->s_cap_gen;
659 ttl = cap->session->s_cap_ttl;
660 spin_unlock(&cap->session->s_gen_ttl_lock);
661
662 if (cap->cap_gen < gen || time_after_eq(jiffies, ttl)) {
663 dout("__cap_is_valid %p cap %p issued %s "
664 "but STALE (gen %u vs %u)\n", &cap->ci->vfs_inode,
665 cap, ceph_cap_string(cap->issued), cap->cap_gen, gen);
666 return 0;
667 }
668
669 return 1;
670}
671
672
673
674
675
676
677int __ceph_caps_issued(struct ceph_inode_info *ci, int *implemented)
678{
679 int have = ci->i_snap_caps | ci->i_cap_exporting_issued;
680 struct ceph_cap *cap;
681 struct rb_node *p;
682
683 if (implemented)
684 *implemented = 0;
685 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
686 cap = rb_entry(p, struct ceph_cap, ci_node);
687 if (!__cap_is_valid(cap))
688 continue;
689 dout("__ceph_caps_issued %p cap %p issued %s\n",
690 &ci->vfs_inode, cap, ceph_cap_string(cap->issued));
691 have |= cap->issued;
692 if (implemented)
693 *implemented |= cap->implemented;
694 }
695
696
697
698
699
700 if (ci->i_auth_cap) {
701 cap = ci->i_auth_cap;
702 have &= ~cap->implemented | cap->issued;
703 }
704 return have;
705}
706
707
708
709
710int __ceph_caps_issued_other(struct ceph_inode_info *ci, struct ceph_cap *ocap)
711{
712 int have = ci->i_snap_caps;
713 struct ceph_cap *cap;
714 struct rb_node *p;
715
716 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
717 cap = rb_entry(p, struct ceph_cap, ci_node);
718 if (cap == ocap)
719 continue;
720 if (!__cap_is_valid(cap))
721 continue;
722 have |= cap->issued;
723 }
724 return have;
725}
726
727
728
729
730
731static void __touch_cap(struct ceph_cap *cap)
732{
733 struct ceph_mds_session *s = cap->session;
734
735 spin_lock(&s->s_cap_lock);
736 if (s->s_cap_iterator == NULL) {
737 dout("__touch_cap %p cap %p mds%d\n", &cap->ci->vfs_inode, cap,
738 s->s_mds);
739 list_move_tail(&cap->session_caps, &s->s_caps);
740 } else {
741 dout("__touch_cap %p cap %p mds%d NOP, iterating over caps\n",
742 &cap->ci->vfs_inode, cap, s->s_mds);
743 }
744 spin_unlock(&s->s_cap_lock);
745}
746
747
748
749
750
751
752int __ceph_caps_issued_mask(struct ceph_inode_info *ci, int mask, int touch)
753{
754 struct ceph_cap *cap;
755 struct rb_node *p;
756 int have = ci->i_snap_caps;
757
758 if ((have & mask) == mask) {
759 dout("__ceph_caps_issued_mask %p snap issued %s"
760 " (mask %s)\n", &ci->vfs_inode,
761 ceph_cap_string(have),
762 ceph_cap_string(mask));
763 return 1;
764 }
765
766 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
767 cap = rb_entry(p, struct ceph_cap, ci_node);
768 if (!__cap_is_valid(cap))
769 continue;
770 if ((cap->issued & mask) == mask) {
771 dout("__ceph_caps_issued_mask %p cap %p issued %s"
772 " (mask %s)\n", &ci->vfs_inode, cap,
773 ceph_cap_string(cap->issued),
774 ceph_cap_string(mask));
775 if (touch)
776 __touch_cap(cap);
777 return 1;
778 }
779
780
781 have |= cap->issued;
782 if ((have & mask) == mask) {
783 dout("__ceph_caps_issued_mask %p combo issued %s"
784 " (mask %s)\n", &ci->vfs_inode,
785 ceph_cap_string(cap->issued),
786 ceph_cap_string(mask));
787 if (touch) {
788 struct rb_node *q;
789
790
791 __touch_cap(cap);
792 for (q = rb_first(&ci->i_caps); q != p;
793 q = rb_next(q)) {
794 cap = rb_entry(q, struct ceph_cap,
795 ci_node);
796 if (!__cap_is_valid(cap))
797 continue;
798 __touch_cap(cap);
799 }
800 }
801 return 1;
802 }
803 }
804
805 return 0;
806}
807
808
809
810
811int __ceph_caps_revoking_other(struct ceph_inode_info *ci,
812 struct ceph_cap *ocap, int mask)
813{
814 struct ceph_cap *cap;
815 struct rb_node *p;
816
817 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
818 cap = rb_entry(p, struct ceph_cap, ci_node);
819 if (cap != ocap && __cap_is_valid(cap) &&
820 (cap->implemented & ~cap->issued & mask))
821 return 1;
822 }
823 return 0;
824}
825
826int ceph_caps_revoking(struct ceph_inode_info *ci, int mask)
827{
828 struct inode *inode = &ci->vfs_inode;
829 int ret;
830
831 spin_lock(&ci->i_ceph_lock);
832 ret = __ceph_caps_revoking_other(ci, NULL, mask);
833 spin_unlock(&ci->i_ceph_lock);
834 dout("ceph_caps_revoking %p %s = %d\n", inode,
835 ceph_cap_string(mask), ret);
836 return ret;
837}
838
839int __ceph_caps_used(struct ceph_inode_info *ci)
840{
841 int used = 0;
842 if (ci->i_pin_ref)
843 used |= CEPH_CAP_PIN;
844 if (ci->i_rd_ref)
845 used |= CEPH_CAP_FILE_RD;
846 if (ci->i_rdcache_ref || ci->vfs_inode.i_data.nrpages)
847 used |= CEPH_CAP_FILE_CACHE;
848 if (ci->i_wr_ref)
849 used |= CEPH_CAP_FILE_WR;
850 if (ci->i_wb_ref || ci->i_wrbuffer_ref)
851 used |= CEPH_CAP_FILE_BUFFER;
852 return used;
853}
854
855
856
857
858int __ceph_caps_file_wanted(struct ceph_inode_info *ci)
859{
860 int want = 0;
861 int mode;
862 for (mode = 0; mode < CEPH_FILE_MODE_NUM; mode++)
863 if (ci->i_nr_by_mode[mode])
864 want |= ceph_caps_for_mode(mode);
865 return want;
866}
867
868
869
870
871int __ceph_caps_mds_wanted(struct ceph_inode_info *ci)
872{
873 struct ceph_cap *cap;
874 struct rb_node *p;
875 int mds_wanted = 0;
876
877 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
878 cap = rb_entry(p, struct ceph_cap, ci_node);
879 if (!__cap_is_valid(cap))
880 continue;
881 mds_wanted |= cap->mds_wanted;
882 }
883 return mds_wanted;
884}
885
886
887
888
889static int __ceph_is_any_caps(struct ceph_inode_info *ci)
890{
891 return !RB_EMPTY_ROOT(&ci->i_caps) || ci->i_cap_exporting_mds >= 0;
892}
893
894
895
896
897
898
899
900void __ceph_remove_cap(struct ceph_cap *cap)
901{
902 struct ceph_mds_session *session = cap->session;
903 struct ceph_inode_info *ci = cap->ci;
904 struct ceph_mds_client *mdsc =
905 ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc;
906 int removed = 0;
907
908 dout("__ceph_remove_cap %p from %p\n", cap, &ci->vfs_inode);
909
910
911 spin_lock(&session->s_cap_lock);
912 if (session->s_cap_iterator == cap) {
913
914 dout("__ceph_remove_cap delaying %p removal from session %p\n",
915 cap, cap->session);
916 } else {
917 list_del_init(&cap->session_caps);
918 session->s_nr_caps--;
919 cap->session = NULL;
920 removed = 1;
921 }
922
923 cap->ci = NULL;
924 spin_unlock(&session->s_cap_lock);
925
926
927 rb_erase(&cap->ci_node, &ci->i_caps);
928 if (ci->i_auth_cap == cap)
929 ci->i_auth_cap = NULL;
930
931 if (removed)
932 ceph_put_cap(mdsc, cap);
933
934 if (!__ceph_is_any_caps(ci) && ci->i_snap_realm) {
935 struct ceph_snap_realm *realm = ci->i_snap_realm;
936 spin_lock(&realm->inodes_with_caps_lock);
937 list_del_init(&ci->i_snap_realm_item);
938 ci->i_snap_realm_counter++;
939 ci->i_snap_realm = NULL;
940 spin_unlock(&realm->inodes_with_caps_lock);
941 ceph_put_snap_realm(mdsc, realm);
942 }
943 if (!__ceph_is_any_real_caps(ci))
944 __cap_delay_cancel(mdsc, ci);
945}
946
947
948
949
950
951
952static int send_cap_msg(struct ceph_mds_session *session,
953 u64 ino, u64 cid, int op,
954 int caps, int wanted, int dirty,
955 u32 seq, u64 flush_tid, u32 issue_seq, u32 mseq,
956 u64 size, u64 max_size,
957 struct timespec *mtime, struct timespec *atime,
958 u64 time_warp_seq,
959 kuid_t uid, kgid_t gid, umode_t mode,
960 u64 xattr_version,
961 struct ceph_buffer *xattrs_buf,
962 u64 follows)
963{
964 struct ceph_mds_caps *fc;
965 struct ceph_msg *msg;
966
967 dout("send_cap_msg %s %llx %llx caps %s wanted %s dirty %s"
968 " seq %u/%u mseq %u follows %lld size %llu/%llu"
969 " xattr_ver %llu xattr_len %d\n", ceph_cap_op_name(op),
970 cid, ino, ceph_cap_string(caps), ceph_cap_string(wanted),
971 ceph_cap_string(dirty),
972 seq, issue_seq, mseq, follows, size, max_size,
973 xattr_version, xattrs_buf ? (int)xattrs_buf->vec.iov_len : 0);
974
975 msg = ceph_msg_new(CEPH_MSG_CLIENT_CAPS, sizeof(*fc), GFP_NOFS, false);
976 if (!msg)
977 return -ENOMEM;
978
979 msg->hdr.tid = cpu_to_le64(flush_tid);
980
981 fc = msg->front.iov_base;
982 memset(fc, 0, sizeof(*fc));
983
984 fc->cap_id = cpu_to_le64(cid);
985 fc->op = cpu_to_le32(op);
986 fc->seq = cpu_to_le32(seq);
987 fc->issue_seq = cpu_to_le32(issue_seq);
988 fc->migrate_seq = cpu_to_le32(mseq);
989 fc->caps = cpu_to_le32(caps);
990 fc->wanted = cpu_to_le32(wanted);
991 fc->dirty = cpu_to_le32(dirty);
992 fc->ino = cpu_to_le64(ino);
993 fc->snap_follows = cpu_to_le64(follows);
994
995 fc->size = cpu_to_le64(size);
996 fc->max_size = cpu_to_le64(max_size);
997 if (mtime)
998 ceph_encode_timespec(&fc->mtime, mtime);
999 if (atime)
1000 ceph_encode_timespec(&fc->atime, atime);
1001 fc->time_warp_seq = cpu_to_le32(time_warp_seq);
1002
1003 fc->uid = cpu_to_le32(from_kuid(&init_user_ns, uid));
1004 fc->gid = cpu_to_le32(from_kgid(&init_user_ns, gid));
1005 fc->mode = cpu_to_le32(mode);
1006
1007 fc->xattr_version = cpu_to_le64(xattr_version);
1008 if (xattrs_buf) {
1009 msg->middle = ceph_buffer_get(xattrs_buf);
1010 fc->xattr_len = cpu_to_le32(xattrs_buf->vec.iov_len);
1011 msg->hdr.middle_len = cpu_to_le32(xattrs_buf->vec.iov_len);
1012 }
1013
1014 ceph_con_send(&session->s_con, msg);
1015 return 0;
1016}
1017
1018void __queue_cap_release(struct ceph_mds_session *session,
1019 u64 ino, u64 cap_id, u32 migrate_seq,
1020 u32 issue_seq)
1021{
1022 struct ceph_msg *msg;
1023 struct ceph_mds_cap_release *head;
1024 struct ceph_mds_cap_item *item;
1025
1026 spin_lock(&session->s_cap_lock);
1027 BUG_ON(!session->s_num_cap_releases);
1028 msg = list_first_entry(&session->s_cap_releases,
1029 struct ceph_msg, list_head);
1030
1031 dout(" adding %llx release to mds%d msg %p (%d left)\n",
1032 ino, session->s_mds, msg, session->s_num_cap_releases);
1033
1034 BUG_ON(msg->front.iov_len + sizeof(*item) > PAGE_CACHE_SIZE);
1035 head = msg->front.iov_base;
1036 le32_add_cpu(&head->num, 1);
1037 item = msg->front.iov_base + msg->front.iov_len;
1038 item->ino = cpu_to_le64(ino);
1039 item->cap_id = cpu_to_le64(cap_id);
1040 item->migrate_seq = cpu_to_le32(migrate_seq);
1041 item->seq = cpu_to_le32(issue_seq);
1042
1043 session->s_num_cap_releases--;
1044
1045 msg->front.iov_len += sizeof(*item);
1046 if (le32_to_cpu(head->num) == CEPH_CAPS_PER_RELEASE) {
1047 dout(" release msg %p full\n", msg);
1048 list_move_tail(&msg->list_head, &session->s_cap_releases_done);
1049 } else {
1050 dout(" release msg %p at %d/%d (%d)\n", msg,
1051 (int)le32_to_cpu(head->num),
1052 (int)CEPH_CAPS_PER_RELEASE,
1053 (int)msg->front.iov_len);
1054 }
1055 spin_unlock(&session->s_cap_lock);
1056}
1057
1058
1059
1060
1061
1062void ceph_queue_caps_release(struct inode *inode)
1063{
1064 struct ceph_inode_info *ci = ceph_inode(inode);
1065 struct rb_node *p;
1066
1067 p = rb_first(&ci->i_caps);
1068 while (p) {
1069 struct ceph_cap *cap = rb_entry(p, struct ceph_cap, ci_node);
1070 struct ceph_mds_session *session = cap->session;
1071
1072 __queue_cap_release(session, ceph_ino(inode), cap->cap_id,
1073 cap->mseq, cap->issue_seq);
1074 p = rb_next(p);
1075 __ceph_remove_cap(cap);
1076 }
1077}
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096static int __send_cap(struct ceph_mds_client *mdsc, struct ceph_cap *cap,
1097 int op, int used, int want, int retain, int flushing,
1098 unsigned *pflush_tid)
1099 __releases(cap->ci->i_ceph_lock)
1100{
1101 struct ceph_inode_info *ci = cap->ci;
1102 struct inode *inode = &ci->vfs_inode;
1103 u64 cap_id = cap->cap_id;
1104 int held, revoking, dropping, keep;
1105 u64 seq, issue_seq, mseq, time_warp_seq, follows;
1106 u64 size, max_size;
1107 struct timespec mtime, atime;
1108 int wake = 0;
1109 umode_t mode;
1110 kuid_t uid;
1111 kgid_t gid;
1112 struct ceph_mds_session *session;
1113 u64 xattr_version = 0;
1114 struct ceph_buffer *xattr_blob = NULL;
1115 int delayed = 0;
1116 u64 flush_tid = 0;
1117 int i;
1118 int ret;
1119
1120 held = cap->issued | cap->implemented;
1121 revoking = cap->implemented & ~cap->issued;
1122 retain &= ~revoking;
1123 dropping = cap->issued & ~retain;
1124
1125 dout("__send_cap %p cap %p session %p %s -> %s (revoking %s)\n",
1126 inode, cap, cap->session,
1127 ceph_cap_string(held), ceph_cap_string(held & retain),
1128 ceph_cap_string(revoking));
1129 BUG_ON((retain & CEPH_CAP_PIN) == 0);
1130
1131 session = cap->session;
1132
1133
1134 if ((ci->i_ceph_flags & CEPH_I_NODELAY) == 0 &&
1135 time_before(jiffies, ci->i_hold_caps_min)) {
1136 dout(" delaying issued %s -> %s, wanted %s -> %s on send\n",
1137 ceph_cap_string(cap->issued),
1138 ceph_cap_string(cap->issued & retain),
1139 ceph_cap_string(cap->mds_wanted),
1140 ceph_cap_string(want));
1141 want |= cap->mds_wanted;
1142 retain |= cap->issued;
1143 delayed = 1;
1144 }
1145 ci->i_ceph_flags &= ~(CEPH_I_NODELAY | CEPH_I_FLUSH);
1146
1147 cap->issued &= retain;
1148 if (cap->implemented & ~cap->issued) {
1149
1150
1151
1152
1153
1154
1155 wake = 1;
1156 }
1157 cap->implemented &= cap->issued | used;
1158 cap->mds_wanted = want;
1159
1160 if (flushing) {
1161
1162
1163
1164
1165
1166
1167
1168 flush_tid = ++ci->i_cap_flush_last_tid;
1169 if (pflush_tid)
1170 *pflush_tid = flush_tid;
1171 dout(" cap_flush_tid %d\n", (int)flush_tid);
1172 for (i = 0; i < CEPH_CAP_BITS; i++)
1173 if (flushing & (1 << i))
1174 ci->i_cap_flush_tid[i] = flush_tid;
1175
1176 follows = ci->i_head_snapc->seq;
1177 } else {
1178 follows = 0;
1179 }
1180
1181 keep = cap->implemented;
1182 seq = cap->seq;
1183 issue_seq = cap->issue_seq;
1184 mseq = cap->mseq;
1185 size = inode->i_size;
1186 ci->i_reported_size = size;
1187 max_size = ci->i_wanted_max_size;
1188 ci->i_requested_max_size = max_size;
1189 mtime = inode->i_mtime;
1190 atime = inode->i_atime;
1191 time_warp_seq = ci->i_time_warp_seq;
1192 uid = inode->i_uid;
1193 gid = inode->i_gid;
1194 mode = inode->i_mode;
1195
1196 if (flushing & CEPH_CAP_XATTR_EXCL) {
1197 __ceph_build_xattrs_blob(ci);
1198 xattr_blob = ci->i_xattrs.blob;
1199 xattr_version = ci->i_xattrs.version;
1200 }
1201
1202 spin_unlock(&ci->i_ceph_lock);
1203
1204 ret = send_cap_msg(session, ceph_vino(inode).ino, cap_id,
1205 op, keep, want, flushing, seq, flush_tid, issue_seq, mseq,
1206 size, max_size, &mtime, &atime, time_warp_seq,
1207 uid, gid, mode, xattr_version, xattr_blob,
1208 follows);
1209 if (ret < 0) {
1210 dout("error sending cap msg, must requeue %p\n", inode);
1211 delayed = 1;
1212 }
1213
1214 if (wake)
1215 wake_up_all(&ci->i_cap_wq);
1216
1217 return delayed;
1218}
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232void __ceph_flush_snaps(struct ceph_inode_info *ci,
1233 struct ceph_mds_session **psession,
1234 int again)
1235 __releases(ci->i_ceph_lock)
1236 __acquires(ci->i_ceph_lock)
1237{
1238 struct inode *inode = &ci->vfs_inode;
1239 int mds;
1240 struct ceph_cap_snap *capsnap;
1241 u32 mseq;
1242 struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
1243 struct ceph_mds_session *session = NULL;
1244
1245 u64 next_follows = 0;
1246
1247
1248
1249 if (psession)
1250 session = *psession;
1251
1252 dout("__flush_snaps %p\n", inode);
1253retry:
1254 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
1255
1256 if (capsnap->follows < next_follows)
1257 continue;
1258
1259
1260
1261
1262 if (capsnap->dirty_pages || capsnap->writing)
1263 break;
1264
1265
1266
1267
1268
1269 BUG_ON(capsnap->dirty == 0);
1270
1271
1272 if (ci->i_auth_cap == NULL) {
1273 dout("no auth cap (migrating?), doing nothing\n");
1274 goto out;
1275 }
1276
1277
1278 if (!again && !list_empty(&capsnap->flushing_item)) {
1279 dout("already flushed %p, skipping\n", capsnap);
1280 continue;
1281 }
1282
1283 mds = ci->i_auth_cap->session->s_mds;
1284 mseq = ci->i_auth_cap->mseq;
1285
1286 if (session && session->s_mds != mds) {
1287 dout("oops, wrong session %p mutex\n", session);
1288 mutex_unlock(&session->s_mutex);
1289 ceph_put_mds_session(session);
1290 session = NULL;
1291 }
1292 if (!session) {
1293 spin_unlock(&ci->i_ceph_lock);
1294 mutex_lock(&mdsc->mutex);
1295 session = __ceph_lookup_mds_session(mdsc, mds);
1296 mutex_unlock(&mdsc->mutex);
1297 if (session) {
1298 dout("inverting session/ino locks on %p\n",
1299 session);
1300 mutex_lock(&session->s_mutex);
1301 }
1302
1303
1304
1305
1306
1307 spin_lock(&ci->i_ceph_lock);
1308 goto retry;
1309 }
1310
1311 capsnap->flush_tid = ++ci->i_cap_flush_last_tid;
1312 atomic_inc(&capsnap->nref);
1313 if (!list_empty(&capsnap->flushing_item))
1314 list_del_init(&capsnap->flushing_item);
1315 list_add_tail(&capsnap->flushing_item,
1316 &session->s_cap_snaps_flushing);
1317 spin_unlock(&ci->i_ceph_lock);
1318
1319 dout("flush_snaps %p cap_snap %p follows %lld tid %llu\n",
1320 inode, capsnap, capsnap->follows, capsnap->flush_tid);
1321 send_cap_msg(session, ceph_vino(inode).ino, 0,
1322 CEPH_CAP_OP_FLUSHSNAP, capsnap->issued, 0,
1323 capsnap->dirty, 0, capsnap->flush_tid, 0, mseq,
1324 capsnap->size, 0,
1325 &capsnap->mtime, &capsnap->atime,
1326 capsnap->time_warp_seq,
1327 capsnap->uid, capsnap->gid, capsnap->mode,
1328 capsnap->xattr_version, capsnap->xattr_blob,
1329 capsnap->follows);
1330
1331 next_follows = capsnap->follows + 1;
1332 ceph_put_cap_snap(capsnap);
1333
1334 spin_lock(&ci->i_ceph_lock);
1335 goto retry;
1336 }
1337
1338
1339 spin_lock(&mdsc->snap_flush_lock);
1340 list_del_init(&ci->i_snap_flush_item);
1341 spin_unlock(&mdsc->snap_flush_lock);
1342
1343out:
1344 if (psession)
1345 *psession = session;
1346 else if (session) {
1347 mutex_unlock(&session->s_mutex);
1348 ceph_put_mds_session(session);
1349 }
1350}
1351
1352static void ceph_flush_snaps(struct ceph_inode_info *ci)
1353{
1354 spin_lock(&ci->i_ceph_lock);
1355 __ceph_flush_snaps(ci, NULL, 0);
1356 spin_unlock(&ci->i_ceph_lock);
1357}
1358
1359
1360
1361
1362
1363
1364int __ceph_mark_dirty_caps(struct ceph_inode_info *ci, int mask)
1365{
1366 struct ceph_mds_client *mdsc =
1367 ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc;
1368 struct inode *inode = &ci->vfs_inode;
1369 int was = ci->i_dirty_caps;
1370 int dirty = 0;
1371
1372 dout("__mark_dirty_caps %p %s dirty %s -> %s\n", &ci->vfs_inode,
1373 ceph_cap_string(mask), ceph_cap_string(was),
1374 ceph_cap_string(was | mask));
1375 ci->i_dirty_caps |= mask;
1376 if (was == 0) {
1377 if (!ci->i_head_snapc)
1378 ci->i_head_snapc = ceph_get_snap_context(
1379 ci->i_snap_realm->cached_context);
1380 dout(" inode %p now dirty snapc %p auth cap %p\n",
1381 &ci->vfs_inode, ci->i_head_snapc, ci->i_auth_cap);
1382 BUG_ON(!list_empty(&ci->i_dirty_item));
1383 spin_lock(&mdsc->cap_dirty_lock);
1384 if (ci->i_auth_cap)
1385 list_add(&ci->i_dirty_item, &mdsc->cap_dirty);
1386 else
1387 list_add(&ci->i_dirty_item,
1388 &mdsc->cap_dirty_migrating);
1389 spin_unlock(&mdsc->cap_dirty_lock);
1390 if (ci->i_flushing_caps == 0) {
1391 ihold(inode);
1392 dirty |= I_DIRTY_SYNC;
1393 }
1394 }
1395 BUG_ON(list_empty(&ci->i_dirty_item));
1396 if (((was | ci->i_flushing_caps) & CEPH_CAP_FILE_BUFFER) &&
1397 (mask & CEPH_CAP_FILE_BUFFER))
1398 dirty |= I_DIRTY_DATASYNC;
1399 __cap_delay_requeue(mdsc, ci);
1400 return dirty;
1401}
1402
1403
1404
1405
1406
1407
1408
1409static int __mark_caps_flushing(struct inode *inode,
1410 struct ceph_mds_session *session)
1411{
1412 struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
1413 struct ceph_inode_info *ci = ceph_inode(inode);
1414 int flushing;
1415
1416 BUG_ON(ci->i_dirty_caps == 0);
1417 BUG_ON(list_empty(&ci->i_dirty_item));
1418
1419 flushing = ci->i_dirty_caps;
1420 dout("__mark_caps_flushing flushing %s, flushing_caps %s -> %s\n",
1421 ceph_cap_string(flushing),
1422 ceph_cap_string(ci->i_flushing_caps),
1423 ceph_cap_string(ci->i_flushing_caps | flushing));
1424 ci->i_flushing_caps |= flushing;
1425 ci->i_dirty_caps = 0;
1426 dout(" inode %p now !dirty\n", inode);
1427
1428 spin_lock(&mdsc->cap_dirty_lock);
1429 list_del_init(&ci->i_dirty_item);
1430
1431 ci->i_cap_flush_seq = ++mdsc->cap_flush_seq;
1432 if (list_empty(&ci->i_flushing_item)) {
1433 list_add_tail(&ci->i_flushing_item, &session->s_cap_flushing);
1434 mdsc->num_cap_flushing++;
1435 dout(" inode %p now flushing seq %lld\n", inode,
1436 ci->i_cap_flush_seq);
1437 } else {
1438 list_move_tail(&ci->i_flushing_item, &session->s_cap_flushing);
1439 dout(" inode %p now flushing (more) seq %lld\n", inode,
1440 ci->i_cap_flush_seq);
1441 }
1442 spin_unlock(&mdsc->cap_dirty_lock);
1443
1444 return flushing;
1445}
1446
1447
1448
1449
1450static int try_nonblocking_invalidate(struct inode *inode)
1451{
1452 struct ceph_inode_info *ci = ceph_inode(inode);
1453 u32 invalidating_gen = ci->i_rdcache_gen;
1454
1455 spin_unlock(&ci->i_ceph_lock);
1456 invalidate_mapping_pages(&inode->i_data, 0, -1);
1457 spin_lock(&ci->i_ceph_lock);
1458
1459 if (inode->i_data.nrpages == 0 &&
1460 invalidating_gen == ci->i_rdcache_gen) {
1461
1462 dout("try_nonblocking_invalidate %p success\n", inode);
1463
1464 ci->i_rdcache_revoking = ci->i_rdcache_gen - 1;
1465 return 0;
1466 }
1467 dout("try_nonblocking_invalidate %p failed\n", inode);
1468 return -1;
1469}
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482void ceph_check_caps(struct ceph_inode_info *ci, int flags,
1483 struct ceph_mds_session *session)
1484{
1485 struct ceph_fs_client *fsc = ceph_inode_to_client(&ci->vfs_inode);
1486 struct ceph_mds_client *mdsc = fsc->mdsc;
1487 struct inode *inode = &ci->vfs_inode;
1488 struct ceph_cap *cap;
1489 int file_wanted, used, cap_used;
1490 int took_snap_rwsem = 0;
1491 int issued, implemented, want, retain, revoking, flushing = 0;
1492 int mds = -1;
1493
1494 struct rb_node *p;
1495 int tried_invalidate = 0;
1496 int delayed = 0, sent = 0, force_requeue = 0, num;
1497 int queue_invalidate = 0;
1498 int is_delayed = flags & CHECK_CAPS_NODELAY;
1499
1500
1501 if (mdsc->stopping)
1502 is_delayed = 1;
1503
1504 spin_lock(&ci->i_ceph_lock);
1505
1506 if (ci->i_ceph_flags & CEPH_I_FLUSH)
1507 flags |= CHECK_CAPS_FLUSH;
1508
1509
1510 if (!list_empty(&ci->i_cap_snaps))
1511 __ceph_flush_snaps(ci, &session, 0);
1512 goto retry_locked;
1513retry:
1514 spin_lock(&ci->i_ceph_lock);
1515retry_locked:
1516 file_wanted = __ceph_caps_file_wanted(ci);
1517 used = __ceph_caps_used(ci);
1518 want = file_wanted | used;
1519 issued = __ceph_caps_issued(ci, &implemented);
1520 revoking = implemented & ~issued;
1521
1522 retain = want | CEPH_CAP_PIN;
1523 if (!mdsc->stopping && inode->i_nlink > 0) {
1524 if (want) {
1525 retain |= CEPH_CAP_ANY;
1526 } else {
1527 retain |= CEPH_CAP_ANY_SHARED;
1528
1529
1530
1531
1532
1533 if (ci->i_max_size == 0)
1534 retain |= CEPH_CAP_ANY_RD;
1535 }
1536 }
1537
1538 dout("check_caps %p file_want %s used %s dirty %s flushing %s"
1539 " issued %s revoking %s retain %s %s%s%s\n", inode,
1540 ceph_cap_string(file_wanted),
1541 ceph_cap_string(used), ceph_cap_string(ci->i_dirty_caps),
1542 ceph_cap_string(ci->i_flushing_caps),
1543 ceph_cap_string(issued), ceph_cap_string(revoking),
1544 ceph_cap_string(retain),
1545 (flags & CHECK_CAPS_AUTHONLY) ? " AUTHONLY" : "",
1546 (flags & CHECK_CAPS_NODELAY) ? " NODELAY" : "",
1547 (flags & CHECK_CAPS_FLUSH) ? " FLUSH" : "");
1548
1549
1550
1551
1552
1553
1554 if ((!is_delayed || mdsc->stopping) &&
1555 ci->i_wrbuffer_ref == 0 &&
1556 inode->i_data.nrpages &&
1557 (file_wanted == 0 ||
1558 (revoking & (CEPH_CAP_FILE_CACHE|
1559 CEPH_CAP_FILE_LAZYIO))) &&
1560 !tried_invalidate) {
1561 dout("check_caps trying to invalidate on %p\n", inode);
1562 if (try_nonblocking_invalidate(inode) < 0) {
1563 if (revoking & (CEPH_CAP_FILE_CACHE|
1564 CEPH_CAP_FILE_LAZYIO)) {
1565 dout("check_caps queuing invalidate\n");
1566 queue_invalidate = 1;
1567 ci->i_rdcache_revoking = ci->i_rdcache_gen;
1568 } else {
1569 dout("check_caps failed to invalidate pages\n");
1570
1571
1572 force_requeue = 1;
1573 __cap_set_timeouts(mdsc, ci);
1574 }
1575 }
1576 tried_invalidate = 1;
1577 goto retry_locked;
1578 }
1579
1580 num = 0;
1581 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
1582 cap = rb_entry(p, struct ceph_cap, ci_node);
1583 num++;
1584
1585
1586 if (mds >= cap->mds ||
1587 ((flags & CHECK_CAPS_AUTHONLY) && cap != ci->i_auth_cap))
1588 continue;
1589
1590
1591
1592 cap_used = used;
1593 if (ci->i_auth_cap && cap != ci->i_auth_cap)
1594 cap_used &= ~ci->i_auth_cap->issued;
1595
1596 revoking = cap->implemented & ~cap->issued;
1597 dout(" mds%d cap %p used %s issued %s implemented %s revoking %s\n",
1598 cap->mds, cap, ceph_cap_string(cap->issued),
1599 ceph_cap_string(cap_used),
1600 ceph_cap_string(cap->implemented),
1601 ceph_cap_string(revoking));
1602
1603 if (cap == ci->i_auth_cap &&
1604 (cap->issued & CEPH_CAP_FILE_WR)) {
1605
1606 if (ci->i_wanted_max_size > ci->i_max_size &&
1607 ci->i_wanted_max_size > ci->i_requested_max_size) {
1608 dout("requesting new max_size\n");
1609 goto ack;
1610 }
1611
1612
1613 if ((inode->i_size << 1) >= ci->i_max_size &&
1614 (ci->i_reported_size << 1) < ci->i_max_size) {
1615 dout("i_size approaching max_size\n");
1616 goto ack;
1617 }
1618 }
1619
1620 if (cap == ci->i_auth_cap && (flags & CHECK_CAPS_FLUSH) &&
1621 ci->i_dirty_caps) {
1622 dout("flushing dirty caps\n");
1623 goto ack;
1624 }
1625
1626
1627 if (revoking && (revoking & cap_used) == 0) {
1628 dout("completed revocation of %s\n",
1629 ceph_cap_string(cap->implemented & ~cap->issued));
1630 goto ack;
1631 }
1632
1633
1634 if (want & ~(cap->mds_wanted | cap->issued))
1635 goto ack;
1636
1637
1638 if ((cap->issued & ~retain) == 0 &&
1639 cap->mds_wanted == want)
1640 continue;
1641
1642 if (is_delayed)
1643 goto ack;
1644
1645
1646 if ((ci->i_ceph_flags & CEPH_I_NODELAY) == 0 &&
1647 time_before(jiffies, ci->i_hold_caps_max)) {
1648 dout(" delaying issued %s -> %s, wanted %s -> %s\n",
1649 ceph_cap_string(cap->issued),
1650 ceph_cap_string(cap->issued & retain),
1651 ceph_cap_string(cap->mds_wanted),
1652 ceph_cap_string(want));
1653 delayed++;
1654 continue;
1655 }
1656
1657ack:
1658 if (ci->i_ceph_flags & CEPH_I_NOFLUSH) {
1659 dout(" skipping %p I_NOFLUSH set\n", inode);
1660 continue;
1661 }
1662
1663 if (session && session != cap->session) {
1664 dout("oops, wrong session %p mutex\n", session);
1665 mutex_unlock(&session->s_mutex);
1666 session = NULL;
1667 }
1668 if (!session) {
1669 session = cap->session;
1670 if (mutex_trylock(&session->s_mutex) == 0) {
1671 dout("inverting session/ino locks on %p\n",
1672 session);
1673 spin_unlock(&ci->i_ceph_lock);
1674 if (took_snap_rwsem) {
1675 up_read(&mdsc->snap_rwsem);
1676 took_snap_rwsem = 0;
1677 }
1678 mutex_lock(&session->s_mutex);
1679 goto retry;
1680 }
1681 }
1682
1683 if (!took_snap_rwsem) {
1684 if (down_read_trylock(&mdsc->snap_rwsem) == 0) {
1685 dout("inverting snap/in locks on %p\n",
1686 inode);
1687 spin_unlock(&ci->i_ceph_lock);
1688 down_read(&mdsc->snap_rwsem);
1689 took_snap_rwsem = 1;
1690 goto retry;
1691 }
1692 took_snap_rwsem = 1;
1693 }
1694
1695 if (cap == ci->i_auth_cap && ci->i_dirty_caps)
1696 flushing = __mark_caps_flushing(inode, session);
1697 else
1698 flushing = 0;
1699
1700 mds = cap->mds;
1701 sent++;
1702
1703
1704 delayed += __send_cap(mdsc, cap, CEPH_CAP_OP_UPDATE, cap_used,
1705 want, retain, flushing, NULL);
1706 goto retry;
1707 }
1708
1709
1710
1711
1712
1713 if (delayed && is_delayed)
1714 force_requeue = 1;
1715 if (!delayed && !is_delayed)
1716 __cap_delay_cancel(mdsc, ci);
1717 else if (!is_delayed || force_requeue)
1718 __cap_delay_requeue(mdsc, ci);
1719
1720 spin_unlock(&ci->i_ceph_lock);
1721
1722 if (queue_invalidate)
1723 ceph_queue_invalidate(inode);
1724
1725 if (session)
1726 mutex_unlock(&session->s_mutex);
1727 if (took_snap_rwsem)
1728 up_read(&mdsc->snap_rwsem);
1729}
1730
1731
1732
1733
1734static int try_flush_caps(struct inode *inode, struct ceph_mds_session *session,
1735 unsigned *flush_tid)
1736{
1737 struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
1738 struct ceph_inode_info *ci = ceph_inode(inode);
1739 int unlock_session = session ? 0 : 1;
1740 int flushing = 0;
1741
1742retry:
1743 spin_lock(&ci->i_ceph_lock);
1744 if (ci->i_ceph_flags & CEPH_I_NOFLUSH) {
1745 dout("try_flush_caps skipping %p I_NOFLUSH set\n", inode);
1746 goto out;
1747 }
1748 if (ci->i_dirty_caps && ci->i_auth_cap) {
1749 struct ceph_cap *cap = ci->i_auth_cap;
1750 int used = __ceph_caps_used(ci);
1751 int want = __ceph_caps_wanted(ci);
1752 int delayed;
1753
1754 if (!session) {
1755 spin_unlock(&ci->i_ceph_lock);
1756 session = cap->session;
1757 mutex_lock(&session->s_mutex);
1758 goto retry;
1759 }
1760 BUG_ON(session != cap->session);
1761 if (cap->session->s_state < CEPH_MDS_SESSION_OPEN)
1762 goto out;
1763
1764 flushing = __mark_caps_flushing(inode, session);
1765
1766
1767 delayed = __send_cap(mdsc, cap, CEPH_CAP_OP_FLUSH, used, want,
1768 cap->issued | cap->implemented, flushing,
1769 flush_tid);
1770 if (!delayed)
1771 goto out_unlocked;
1772
1773 spin_lock(&ci->i_ceph_lock);
1774 __cap_delay_requeue(mdsc, ci);
1775 }
1776out:
1777 spin_unlock(&ci->i_ceph_lock);
1778out_unlocked:
1779 if (session && unlock_session)
1780 mutex_unlock(&session->s_mutex);
1781 return flushing;
1782}
1783
1784
1785
1786
1787static int caps_are_flushed(struct inode *inode, unsigned tid)
1788{
1789 struct ceph_inode_info *ci = ceph_inode(inode);
1790 int i, ret = 1;
1791
1792 spin_lock(&ci->i_ceph_lock);
1793 for (i = 0; i < CEPH_CAP_BITS; i++)
1794 if ((ci->i_flushing_caps & (1 << i)) &&
1795 ci->i_cap_flush_tid[i] <= tid) {
1796
1797 ret = 0;
1798 break;
1799 }
1800 spin_unlock(&ci->i_ceph_lock);
1801 return ret;
1802}
1803
1804
1805
1806
1807
1808
1809
1810static void sync_write_wait(struct inode *inode)
1811{
1812 struct ceph_inode_info *ci = ceph_inode(inode);
1813 struct list_head *head = &ci->i_unsafe_writes;
1814 struct ceph_osd_request *req;
1815 u64 last_tid;
1816
1817 spin_lock(&ci->i_unsafe_lock);
1818 if (list_empty(head))
1819 goto out;
1820
1821
1822 req = list_entry(head->prev, struct ceph_osd_request,
1823 r_unsafe_item);
1824 last_tid = req->r_tid;
1825
1826 do {
1827 ceph_osdc_get_request(req);
1828 spin_unlock(&ci->i_unsafe_lock);
1829 dout("sync_write_wait on tid %llu (until %llu)\n",
1830 req->r_tid, last_tid);
1831 wait_for_completion(&req->r_safe_completion);
1832 spin_lock(&ci->i_unsafe_lock);
1833 ceph_osdc_put_request(req);
1834
1835
1836
1837
1838
1839 if (list_empty(head))
1840 break;
1841 req = list_entry(head->next, struct ceph_osd_request,
1842 r_unsafe_item);
1843 } while (req->r_tid < last_tid);
1844out:
1845 spin_unlock(&ci->i_unsafe_lock);
1846}
1847
1848int ceph_fsync(struct file *file, loff_t start, loff_t end, int datasync)
1849{
1850 struct inode *inode = file->f_mapping->host;
1851 struct ceph_inode_info *ci = ceph_inode(inode);
1852 unsigned flush_tid;
1853 int ret;
1854 int dirty;
1855
1856 dout("fsync %p%s\n", inode, datasync ? " datasync" : "");
1857 sync_write_wait(inode);
1858
1859 ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
1860 if (ret < 0)
1861 return ret;
1862 mutex_lock(&inode->i_mutex);
1863
1864 dirty = try_flush_caps(inode, NULL, &flush_tid);
1865 dout("fsync dirty caps are %s\n", ceph_cap_string(dirty));
1866
1867
1868
1869
1870
1871
1872 if (!datasync && (dirty & ~CEPH_CAP_ANY_FILE_WR)) {
1873 dout("fsync waiting for flush_tid %u\n", flush_tid);
1874 ret = wait_event_interruptible(ci->i_cap_wq,
1875 caps_are_flushed(inode, flush_tid));
1876 }
1877
1878 dout("fsync %p%s done\n", inode, datasync ? " datasync" : "");
1879 mutex_unlock(&inode->i_mutex);
1880 return ret;
1881}
1882
1883
1884
1885
1886
1887
1888
1889int ceph_write_inode(struct inode *inode, struct writeback_control *wbc)
1890{
1891 struct ceph_inode_info *ci = ceph_inode(inode);
1892 unsigned flush_tid;
1893 int err = 0;
1894 int dirty;
1895 int wait = wbc->sync_mode == WB_SYNC_ALL;
1896
1897 dout("write_inode %p wait=%d\n", inode, wait);
1898 if (wait) {
1899 dirty = try_flush_caps(inode, NULL, &flush_tid);
1900 if (dirty)
1901 err = wait_event_interruptible(ci->i_cap_wq,
1902 caps_are_flushed(inode, flush_tid));
1903 } else {
1904 struct ceph_mds_client *mdsc =
1905 ceph_sb_to_client(inode->i_sb)->mdsc;
1906
1907 spin_lock(&ci->i_ceph_lock);
1908 if (__ceph_caps_dirty(ci))
1909 __cap_delay_requeue_front(mdsc, ci);
1910 spin_unlock(&ci->i_ceph_lock);
1911 }
1912 return err;
1913}
1914
1915
1916
1917
1918
1919
1920
1921static void kick_flushing_capsnaps(struct ceph_mds_client *mdsc,
1922 struct ceph_mds_session *session)
1923{
1924 struct ceph_cap_snap *capsnap;
1925
1926 dout("kick_flushing_capsnaps mds%d\n", session->s_mds);
1927 list_for_each_entry(capsnap, &session->s_cap_snaps_flushing,
1928 flushing_item) {
1929 struct ceph_inode_info *ci = capsnap->ci;
1930 struct inode *inode = &ci->vfs_inode;
1931 struct ceph_cap *cap;
1932
1933 spin_lock(&ci->i_ceph_lock);
1934 cap = ci->i_auth_cap;
1935 if (cap && cap->session == session) {
1936 dout("kick_flushing_caps %p cap %p capsnap %p\n", inode,
1937 cap, capsnap);
1938 __ceph_flush_snaps(ci, &session, 1);
1939 } else {
1940 pr_err("%p auth cap %p not mds%d ???\n", inode,
1941 cap, session->s_mds);
1942 }
1943 spin_unlock(&ci->i_ceph_lock);
1944 }
1945}
1946
1947void ceph_kick_flushing_caps(struct ceph_mds_client *mdsc,
1948 struct ceph_mds_session *session)
1949{
1950 struct ceph_inode_info *ci;
1951
1952 kick_flushing_capsnaps(mdsc, session);
1953
1954 dout("kick_flushing_caps mds%d\n", session->s_mds);
1955 list_for_each_entry(ci, &session->s_cap_flushing, i_flushing_item) {
1956 struct inode *inode = &ci->vfs_inode;
1957 struct ceph_cap *cap;
1958 int delayed = 0;
1959
1960 spin_lock(&ci->i_ceph_lock);
1961 cap = ci->i_auth_cap;
1962 if (cap && cap->session == session) {
1963 dout("kick_flushing_caps %p cap %p %s\n", inode,
1964 cap, ceph_cap_string(ci->i_flushing_caps));
1965 delayed = __send_cap(mdsc, cap, CEPH_CAP_OP_FLUSH,
1966 __ceph_caps_used(ci),
1967 __ceph_caps_wanted(ci),
1968 cap->issued | cap->implemented,
1969 ci->i_flushing_caps, NULL);
1970 if (delayed) {
1971 spin_lock(&ci->i_ceph_lock);
1972 __cap_delay_requeue(mdsc, ci);
1973 spin_unlock(&ci->i_ceph_lock);
1974 }
1975 } else {
1976 pr_err("%p auth cap %p not mds%d ???\n", inode,
1977 cap, session->s_mds);
1978 spin_unlock(&ci->i_ceph_lock);
1979 }
1980 }
1981}
1982
1983static void kick_flushing_inode_caps(struct ceph_mds_client *mdsc,
1984 struct ceph_mds_session *session,
1985 struct inode *inode)
1986{
1987 struct ceph_inode_info *ci = ceph_inode(inode);
1988 struct ceph_cap *cap;
1989 int delayed = 0;
1990
1991 spin_lock(&ci->i_ceph_lock);
1992 cap = ci->i_auth_cap;
1993 dout("kick_flushing_inode_caps %p flushing %s flush_seq %lld\n", inode,
1994 ceph_cap_string(ci->i_flushing_caps), ci->i_cap_flush_seq);
1995
1996 __ceph_flush_snaps(ci, &session, 1);
1997
1998 if (ci->i_flushing_caps) {
1999 spin_lock(&mdsc->cap_dirty_lock);
2000 list_move_tail(&ci->i_flushing_item,
2001 &cap->session->s_cap_flushing);
2002 spin_unlock(&mdsc->cap_dirty_lock);
2003
2004 delayed = __send_cap(mdsc, cap, CEPH_CAP_OP_FLUSH,
2005 __ceph_caps_used(ci),
2006 __ceph_caps_wanted(ci),
2007 cap->issued | cap->implemented,
2008 ci->i_flushing_caps, NULL);
2009 if (delayed) {
2010 spin_lock(&ci->i_ceph_lock);
2011 __cap_delay_requeue(mdsc, ci);
2012 spin_unlock(&ci->i_ceph_lock);
2013 }
2014 } else {
2015 spin_unlock(&ci->i_ceph_lock);
2016 }
2017}
2018
2019
2020
2021
2022
2023
2024
2025
2026static void __take_cap_refs(struct ceph_inode_info *ci, int got)
2027{
2028 if (got & CEPH_CAP_PIN)
2029 ci->i_pin_ref++;
2030 if (got & CEPH_CAP_FILE_RD)
2031 ci->i_rd_ref++;
2032 if (got & CEPH_CAP_FILE_CACHE)
2033 ci->i_rdcache_ref++;
2034 if (got & CEPH_CAP_FILE_WR)
2035 ci->i_wr_ref++;
2036 if (got & CEPH_CAP_FILE_BUFFER) {
2037 if (ci->i_wb_ref == 0)
2038 ihold(&ci->vfs_inode);
2039 ci->i_wb_ref++;
2040 dout("__take_cap_refs %p wb %d -> %d (?)\n",
2041 &ci->vfs_inode, ci->i_wb_ref-1, ci->i_wb_ref);
2042 }
2043}
2044
2045
2046
2047
2048
2049
2050
2051
2052static int try_get_cap_refs(struct ceph_inode_info *ci, int need, int want,
2053 int *got, loff_t endoff, int *check_max, int *err)
2054{
2055 struct inode *inode = &ci->vfs_inode;
2056 int ret = 0;
2057 int have, implemented;
2058 int file_wanted;
2059
2060 dout("get_cap_refs %p need %s want %s\n", inode,
2061 ceph_cap_string(need), ceph_cap_string(want));
2062 spin_lock(&ci->i_ceph_lock);
2063
2064
2065 file_wanted = __ceph_caps_file_wanted(ci);
2066 if ((file_wanted & need) == 0) {
2067 dout("try_get_cap_refs need %s file_wanted %s, EBADF\n",
2068 ceph_cap_string(need), ceph_cap_string(file_wanted));
2069 *err = -EBADF;
2070 ret = 1;
2071 goto out;
2072 }
2073
2074
2075 while (ci->i_truncate_pending) {
2076 spin_unlock(&ci->i_ceph_lock);
2077 __ceph_do_pending_vmtruncate(inode);
2078 spin_lock(&ci->i_ceph_lock);
2079 }
2080
2081 have = __ceph_caps_issued(ci, &implemented);
2082
2083 if (have & need & CEPH_CAP_FILE_WR) {
2084 if (endoff >= 0 && endoff > (loff_t)ci->i_max_size) {
2085 dout("get_cap_refs %p endoff %llu > maxsize %llu\n",
2086 inode, endoff, ci->i_max_size);
2087 if (endoff > ci->i_requested_max_size) {
2088 *check_max = 1;
2089 ret = 1;
2090 }
2091 goto out;
2092 }
2093
2094
2095
2096
2097 if (__ceph_have_pending_cap_snap(ci)) {
2098 dout("get_cap_refs %p cap_snap_pending\n", inode);
2099 goto out;
2100 }
2101 }
2102
2103 if ((have & need) == need) {
2104
2105
2106
2107
2108
2109
2110 int not = want & ~(have & need);
2111 int revoking = implemented & ~have;
2112 dout("get_cap_refs %p have %s but not %s (revoking %s)\n",
2113 inode, ceph_cap_string(have), ceph_cap_string(not),
2114 ceph_cap_string(revoking));
2115 if ((revoking & not) == 0) {
2116 *got = need | (have & want);
2117 __take_cap_refs(ci, *got);
2118 ret = 1;
2119 }
2120 } else {
2121 dout("get_cap_refs %p have %s needed %s\n", inode,
2122 ceph_cap_string(have), ceph_cap_string(need));
2123 }
2124out:
2125 spin_unlock(&ci->i_ceph_lock);
2126 dout("get_cap_refs %p ret %d got %s\n", inode,
2127 ret, ceph_cap_string(*got));
2128 return ret;
2129}
2130
2131
2132
2133
2134
2135
2136static void check_max_size(struct inode *inode, loff_t endoff)
2137{
2138 struct ceph_inode_info *ci = ceph_inode(inode);
2139 int check = 0;
2140
2141
2142 spin_lock(&ci->i_ceph_lock);
2143 if (endoff >= ci->i_max_size && endoff > ci->i_wanted_max_size) {
2144 dout("write %p at large endoff %llu, req max_size\n",
2145 inode, endoff);
2146 ci->i_wanted_max_size = endoff;
2147 }
2148
2149 if (ci->i_auth_cap &&
2150 (ci->i_auth_cap->issued & CEPH_CAP_FILE_WR) &&
2151 ci->i_wanted_max_size > ci->i_max_size &&
2152 ci->i_wanted_max_size > ci->i_requested_max_size)
2153 check = 1;
2154 spin_unlock(&ci->i_ceph_lock);
2155 if (check)
2156 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
2157}
2158
2159
2160
2161
2162
2163
2164int ceph_get_caps(struct ceph_inode_info *ci, int need, int want, int *got,
2165 loff_t endoff)
2166{
2167 int check_max, ret, err;
2168
2169retry:
2170 if (endoff > 0)
2171 check_max_size(&ci->vfs_inode, endoff);
2172 check_max = 0;
2173 err = 0;
2174 ret = wait_event_interruptible(ci->i_cap_wq,
2175 try_get_cap_refs(ci, need, want,
2176 got, endoff,
2177 &check_max, &err));
2178 if (err)
2179 ret = err;
2180 if (check_max)
2181 goto retry;
2182 return ret;
2183}
2184
2185
2186
2187
2188
2189void ceph_get_cap_refs(struct ceph_inode_info *ci, int caps)
2190{
2191 spin_lock(&ci->i_ceph_lock);
2192 __take_cap_refs(ci, caps);
2193 spin_unlock(&ci->i_ceph_lock);
2194}
2195
2196
2197
2198
2199
2200
2201
2202
2203
2204
2205void ceph_put_cap_refs(struct ceph_inode_info *ci, int had)
2206{
2207 struct inode *inode = &ci->vfs_inode;
2208 int last = 0, put = 0, flushsnaps = 0, wake = 0;
2209 struct ceph_cap_snap *capsnap;
2210
2211 spin_lock(&ci->i_ceph_lock);
2212 if (had & CEPH_CAP_PIN)
2213 --ci->i_pin_ref;
2214 if (had & CEPH_CAP_FILE_RD)
2215 if (--ci->i_rd_ref == 0)
2216 last++;
2217 if (had & CEPH_CAP_FILE_CACHE)
2218 if (--ci->i_rdcache_ref == 0)
2219 last++;
2220 if (had & CEPH_CAP_FILE_BUFFER) {
2221 if (--ci->i_wb_ref == 0) {
2222 last++;
2223 put++;
2224 }
2225 dout("put_cap_refs %p wb %d -> %d (?)\n",
2226 inode, ci->i_wb_ref+1, ci->i_wb_ref);
2227 }
2228 if (had & CEPH_CAP_FILE_WR)
2229 if (--ci->i_wr_ref == 0) {
2230 last++;
2231 if (!list_empty(&ci->i_cap_snaps)) {
2232 capsnap = list_first_entry(&ci->i_cap_snaps,
2233 struct ceph_cap_snap,
2234 ci_item);
2235 if (capsnap->writing) {
2236 capsnap->writing = 0;
2237 flushsnaps =
2238 __ceph_finish_cap_snap(ci,
2239 capsnap);
2240 wake = 1;
2241 }
2242 }
2243 }
2244 spin_unlock(&ci->i_ceph_lock);
2245
2246 dout("put_cap_refs %p had %s%s%s\n", inode, ceph_cap_string(had),
2247 last ? " last" : "", put ? " put" : "");
2248
2249 if (last && !flushsnaps)
2250 ceph_check_caps(ci, 0, NULL);
2251 else if (flushsnaps)
2252 ceph_flush_snaps(ci);
2253 if (wake)
2254 wake_up_all(&ci->i_cap_wq);
2255 if (put)
2256 iput(inode);
2257}
2258
2259
2260
2261
2262
2263
2264
2265
2266void ceph_put_wrbuffer_cap_refs(struct ceph_inode_info *ci, int nr,
2267 struct ceph_snap_context *snapc)
2268{
2269 struct inode *inode = &ci->vfs_inode;
2270 int last = 0;
2271 int complete_capsnap = 0;
2272 int drop_capsnap = 0;
2273 int found = 0;
2274 struct ceph_cap_snap *capsnap = NULL;
2275
2276 spin_lock(&ci->i_ceph_lock);
2277 ci->i_wrbuffer_ref -= nr;
2278 last = !ci->i_wrbuffer_ref;
2279
2280 if (ci->i_head_snapc == snapc) {
2281 ci->i_wrbuffer_ref_head -= nr;
2282 if (ci->i_wrbuffer_ref_head == 0 &&
2283 ci->i_dirty_caps == 0 && ci->i_flushing_caps == 0) {
2284 BUG_ON(!ci->i_head_snapc);
2285 ceph_put_snap_context(ci->i_head_snapc);
2286 ci->i_head_snapc = NULL;
2287 }
2288 dout("put_wrbuffer_cap_refs on %p head %d/%d -> %d/%d %s\n",
2289 inode,
2290 ci->i_wrbuffer_ref+nr, ci->i_wrbuffer_ref_head+nr,
2291 ci->i_wrbuffer_ref, ci->i_wrbuffer_ref_head,
2292 last ? " LAST" : "");
2293 } else {
2294 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
2295 if (capsnap->context == snapc) {
2296 found = 1;
2297 break;
2298 }
2299 }
2300 BUG_ON(!found);
2301 capsnap->dirty_pages -= nr;
2302 if (capsnap->dirty_pages == 0) {
2303 complete_capsnap = 1;
2304 if (capsnap->dirty == 0)
2305
2306
2307 drop_capsnap = 1;
2308 }
2309 dout("put_wrbuffer_cap_refs on %p cap_snap %p "
2310 " snap %lld %d/%d -> %d/%d %s%s%s\n",
2311 inode, capsnap, capsnap->context->seq,
2312 ci->i_wrbuffer_ref+nr, capsnap->dirty_pages + nr,
2313 ci->i_wrbuffer_ref, capsnap->dirty_pages,
2314 last ? " (wrbuffer last)" : "",
2315 complete_capsnap ? " (complete capsnap)" : "",
2316 drop_capsnap ? " (drop capsnap)" : "");
2317 if (drop_capsnap) {
2318 ceph_put_snap_context(capsnap->context);
2319 list_del(&capsnap->ci_item);
2320 list_del(&capsnap->flushing_item);
2321 ceph_put_cap_snap(capsnap);
2322 }
2323 }
2324
2325 spin_unlock(&ci->i_ceph_lock);
2326
2327 if (last) {
2328 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
2329 iput(inode);
2330 } else if (complete_capsnap) {
2331 ceph_flush_snaps(ci);
2332 wake_up_all(&ci->i_cap_wq);
2333 }
2334 if (drop_capsnap)
2335 iput(inode);
2336}
2337
2338
2339
2340
2341static void invalidate_aliases(struct inode *inode)
2342{
2343 struct dentry *dn, *prev = NULL;
2344
2345 dout("invalidate_aliases inode %p\n", inode);
2346 d_prune_aliases(inode);
2347
2348
2349
2350
2351
2352
2353
2354
2355
2356 while ((dn = d_find_alias(inode))) {
2357 if (dn == prev) {
2358 dput(dn);
2359 break;
2360 }
2361 d_invalidate(dn);
2362 if (prev)
2363 dput(prev);
2364 prev = dn;
2365 }
2366 if (prev)
2367 dput(prev);
2368}
2369
2370
2371
2372
2373
2374
2375
2376
2377
2378
2379
2380
2381static void handle_cap_grant(struct inode *inode, struct ceph_mds_caps *grant,
2382 struct ceph_mds_session *session,
2383 struct ceph_cap *cap,
2384 struct ceph_buffer *xattr_buf)
2385 __releases(ci->i_ceph_lock)
2386{
2387 struct ceph_inode_info *ci = ceph_inode(inode);
2388 int mds = session->s_mds;
2389 int seq = le32_to_cpu(grant->seq);
2390 int newcaps = le32_to_cpu(grant->caps);
2391 int issued, implemented, used, wanted, dirty;
2392 u64 size = le64_to_cpu(grant->size);
2393 u64 max_size = le64_to_cpu(grant->max_size);
2394 struct timespec mtime, atime, ctime;
2395 int check_caps = 0;
2396 int wake = 0;
2397 int writeback = 0;
2398 int queue_invalidate = 0;
2399 int deleted_inode = 0;
2400 int queue_revalidate = 0;
2401
2402 dout("handle_cap_grant inode %p cap %p mds%d seq %d %s\n",
2403 inode, cap, mds, seq, ceph_cap_string(newcaps));
2404 dout(" size %llu max_size %llu, i_size %llu\n", size, max_size,
2405 inode->i_size);
2406
2407
2408
2409
2410
2411
2412 if (((cap->issued & ~newcaps) & CEPH_CAP_FILE_CACHE) &&
2413 (newcaps & CEPH_CAP_FILE_LAZYIO) == 0 &&
2414 !ci->i_wrbuffer_ref) {
2415 if (try_nonblocking_invalidate(inode)) {
2416
2417
2418 if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
2419 queue_invalidate = 1;
2420 ci->i_rdcache_revoking = ci->i_rdcache_gen;
2421 }
2422 }
2423
2424 ceph_fscache_invalidate(inode);
2425 }
2426
2427
2428
2429 issued = __ceph_caps_issued(ci, &implemented);
2430 issued |= implemented | __ceph_caps_dirty(ci);
2431
2432 cap->cap_gen = session->s_cap_gen;
2433
2434 __check_cap_issue(ci, cap, newcaps);
2435
2436 if ((issued & CEPH_CAP_AUTH_EXCL) == 0) {
2437 inode->i_mode = le32_to_cpu(grant->mode);
2438 inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(grant->uid));
2439 inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(grant->gid));
2440 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
2441 from_kuid(&init_user_ns, inode->i_uid),
2442 from_kgid(&init_user_ns, inode->i_gid));
2443 }
2444
2445 if ((issued & CEPH_CAP_LINK_EXCL) == 0) {
2446 set_nlink(inode, le32_to_cpu(grant->nlink));
2447 if (inode->i_nlink == 0 &&
2448 (newcaps & (CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL)))
2449 deleted_inode = 1;
2450 }
2451
2452 if ((issued & CEPH_CAP_XATTR_EXCL) == 0 && grant->xattr_len) {
2453 int len = le32_to_cpu(grant->xattr_len);
2454 u64 version = le64_to_cpu(grant->xattr_version);
2455
2456 if (version > ci->i_xattrs.version) {
2457 dout(" got new xattrs v%llu on %p len %d\n",
2458 version, inode, len);
2459 if (ci->i_xattrs.blob)
2460 ceph_buffer_put(ci->i_xattrs.blob);
2461 ci->i_xattrs.blob = ceph_buffer_get(xattr_buf);
2462 ci->i_xattrs.version = version;
2463 }
2464 }
2465
2466
2467
2468 if ((issued & CEPH_CAP_FILE_CACHE) && ci->i_rdcache_gen > 1)
2469 queue_revalidate = 1;
2470
2471
2472 ceph_fill_file_size(inode, issued,
2473 le32_to_cpu(grant->truncate_seq),
2474 le64_to_cpu(grant->truncate_size), size);
2475 ceph_decode_timespec(&mtime, &grant->mtime);
2476 ceph_decode_timespec(&atime, &grant->atime);
2477 ceph_decode_timespec(&ctime, &grant->ctime);
2478 ceph_fill_file_time(inode, issued,
2479 le32_to_cpu(grant->time_warp_seq), &ctime, &mtime,
2480 &atime);
2481
2482
2483 if (ci->i_auth_cap == cap && max_size != ci->i_max_size) {
2484 dout("max_size %lld -> %llu\n", ci->i_max_size, max_size);
2485 ci->i_max_size = max_size;
2486 if (max_size >= ci->i_wanted_max_size) {
2487 ci->i_wanted_max_size = 0;
2488 ci->i_requested_max_size = 0;
2489 }
2490 wake = 1;
2491 }
2492
2493
2494 wanted = __ceph_caps_wanted(ci);
2495 used = __ceph_caps_used(ci);
2496 dirty = __ceph_caps_dirty(ci);
2497 dout(" my wanted = %s, used = %s, dirty %s\n",
2498 ceph_cap_string(wanted),
2499 ceph_cap_string(used),
2500 ceph_cap_string(dirty));
2501 if (wanted != le32_to_cpu(grant->wanted)) {
2502 dout("mds wanted %s -> %s\n",
2503 ceph_cap_string(le32_to_cpu(grant->wanted)),
2504 ceph_cap_string(wanted));
2505
2506 if (le32_to_cpu(grant->op) == CEPH_CAP_OP_IMPORT)
2507 check_caps = 1;
2508 }
2509
2510 cap->seq = seq;
2511
2512
2513 ci->i_layout = grant->layout;
2514
2515
2516 if (cap->issued & ~newcaps) {
2517 int revoking = cap->issued & ~newcaps;
2518
2519 dout("revocation: %s -> %s (revoking %s)\n",
2520 ceph_cap_string(cap->issued),
2521 ceph_cap_string(newcaps),
2522 ceph_cap_string(revoking));
2523 if (revoking & used & CEPH_CAP_FILE_BUFFER)
2524 writeback = 1;
2525 else if (revoking == CEPH_CAP_FILE_CACHE &&
2526 (newcaps & CEPH_CAP_FILE_LAZYIO) == 0 &&
2527 queue_invalidate)
2528 ;
2529 else if (cap == ci->i_auth_cap)
2530 check_caps = 1;
2531 else
2532 check_caps = 2;
2533 cap->issued = newcaps;
2534 cap->implemented |= newcaps;
2535 } else if (cap->issued == newcaps) {
2536 dout("caps unchanged: %s -> %s\n",
2537 ceph_cap_string(cap->issued), ceph_cap_string(newcaps));
2538 } else {
2539 dout("grant: %s -> %s\n", ceph_cap_string(cap->issued),
2540 ceph_cap_string(newcaps));
2541
2542 if (cap == ci->i_auth_cap &&
2543 __ceph_caps_revoking_other(ci, cap, newcaps))
2544 check_caps = 2;
2545
2546 cap->issued = newcaps;
2547 cap->implemented |= newcaps;
2548
2549
2550 wake = 1;
2551 }
2552 BUG_ON(cap->issued & ~cap->implemented);
2553
2554 spin_unlock(&ci->i_ceph_lock);
2555
2556 if (writeback)
2557
2558
2559
2560
2561
2562 ceph_queue_writeback(inode);
2563 if (queue_invalidate)
2564 ceph_queue_invalidate(inode);
2565 if (deleted_inode)
2566 invalidate_aliases(inode);
2567 if (queue_revalidate)
2568 ceph_queue_revalidate(inode);
2569 if (wake)
2570 wake_up_all(&ci->i_cap_wq);
2571
2572 if (check_caps == 1)
2573 ceph_check_caps(ci, CHECK_CAPS_NODELAY|CHECK_CAPS_AUTHONLY,
2574 session);
2575 else if (check_caps == 2)
2576 ceph_check_caps(ci, CHECK_CAPS_NODELAY, session);
2577 else
2578 mutex_unlock(&session->s_mutex);
2579}
2580
2581
2582
2583
2584
2585static void handle_cap_flush_ack(struct inode *inode, u64 flush_tid,
2586 struct ceph_mds_caps *m,
2587 struct ceph_mds_session *session,
2588 struct ceph_cap *cap)
2589 __releases(ci->i_ceph_lock)
2590{
2591 struct ceph_inode_info *ci = ceph_inode(inode);
2592 struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
2593 unsigned seq = le32_to_cpu(m->seq);
2594 int dirty = le32_to_cpu(m->dirty);
2595 int cleaned = 0;
2596 int drop = 0;
2597 int i;
2598
2599 for (i = 0; i < CEPH_CAP_BITS; i++)
2600 if ((dirty & (1 << i)) &&
2601 flush_tid == ci->i_cap_flush_tid[i])
2602 cleaned |= 1 << i;
2603
2604 dout("handle_cap_flush_ack inode %p mds%d seq %d on %s cleaned %s,"
2605 " flushing %s -> %s\n",
2606 inode, session->s_mds, seq, ceph_cap_string(dirty),
2607 ceph_cap_string(cleaned), ceph_cap_string(ci->i_flushing_caps),
2608 ceph_cap_string(ci->i_flushing_caps & ~cleaned));
2609
2610 if (ci->i_flushing_caps == (ci->i_flushing_caps & ~cleaned))
2611 goto out;
2612
2613 ci->i_flushing_caps &= ~cleaned;
2614
2615 spin_lock(&mdsc->cap_dirty_lock);
2616 if (ci->i_flushing_caps == 0) {
2617 list_del_init(&ci->i_flushing_item);
2618 if (!list_empty(&session->s_cap_flushing))
2619 dout(" mds%d still flushing cap on %p\n",
2620 session->s_mds,
2621 &list_entry(session->s_cap_flushing.next,
2622 struct ceph_inode_info,
2623 i_flushing_item)->vfs_inode);
2624 mdsc->num_cap_flushing--;
2625 wake_up_all(&mdsc->cap_flushing_wq);
2626 dout(" inode %p now !flushing\n", inode);
2627
2628 if (ci->i_dirty_caps == 0) {
2629 dout(" inode %p now clean\n", inode);
2630 BUG_ON(!list_empty(&ci->i_dirty_item));
2631 drop = 1;
2632 if (ci->i_wrbuffer_ref_head == 0) {
2633 BUG_ON(!ci->i_head_snapc);
2634 ceph_put_snap_context(ci->i_head_snapc);
2635 ci->i_head_snapc = NULL;
2636 }
2637 } else {
2638 BUG_ON(list_empty(&ci->i_dirty_item));
2639 }
2640 }
2641 spin_unlock(&mdsc->cap_dirty_lock);
2642 wake_up_all(&ci->i_cap_wq);
2643
2644out:
2645 spin_unlock(&ci->i_ceph_lock);
2646 if (drop)
2647 iput(inode);
2648}
2649
2650
2651
2652
2653
2654
2655
2656static void handle_cap_flushsnap_ack(struct inode *inode, u64 flush_tid,
2657 struct ceph_mds_caps *m,
2658 struct ceph_mds_session *session)
2659{
2660 struct ceph_inode_info *ci = ceph_inode(inode);
2661 u64 follows = le64_to_cpu(m->snap_follows);
2662 struct ceph_cap_snap *capsnap;
2663 int drop = 0;
2664
2665 dout("handle_cap_flushsnap_ack inode %p ci %p mds%d follows %lld\n",
2666 inode, ci, session->s_mds, follows);
2667
2668 spin_lock(&ci->i_ceph_lock);
2669 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
2670 if (capsnap->follows == follows) {
2671 if (capsnap->flush_tid != flush_tid) {
2672 dout(" cap_snap %p follows %lld tid %lld !="
2673 " %lld\n", capsnap, follows,
2674 flush_tid, capsnap->flush_tid);
2675 break;
2676 }
2677 WARN_ON(capsnap->dirty_pages || capsnap->writing);
2678 dout(" removing %p cap_snap %p follows %lld\n",
2679 inode, capsnap, follows);
2680 ceph_put_snap_context(capsnap->context);
2681 list_del(&capsnap->ci_item);
2682 list_del(&capsnap->flushing_item);
2683 ceph_put_cap_snap(capsnap);
2684 drop = 1;
2685 break;
2686 } else {
2687 dout(" skipping cap_snap %p follows %lld\n",
2688 capsnap, capsnap->follows);
2689 }
2690 }
2691 spin_unlock(&ci->i_ceph_lock);
2692 if (drop)
2693 iput(inode);
2694}
2695
2696
2697
2698
2699
2700
2701static void handle_cap_trunc(struct inode *inode,
2702 struct ceph_mds_caps *trunc,
2703 struct ceph_mds_session *session)
2704 __releases(ci->i_ceph_lock)
2705{
2706 struct ceph_inode_info *ci = ceph_inode(inode);
2707 int mds = session->s_mds;
2708 int seq = le32_to_cpu(trunc->seq);
2709 u32 truncate_seq = le32_to_cpu(trunc->truncate_seq);
2710 u64 truncate_size = le64_to_cpu(trunc->truncate_size);
2711 u64 size = le64_to_cpu(trunc->size);
2712 int implemented = 0;
2713 int dirty = __ceph_caps_dirty(ci);
2714 int issued = __ceph_caps_issued(ceph_inode(inode), &implemented);
2715 int queue_trunc = 0;
2716
2717 issued |= implemented | dirty;
2718
2719 dout("handle_cap_trunc inode %p mds%d seq %d to %lld seq %d\n",
2720 inode, mds, seq, truncate_size, truncate_seq);
2721 queue_trunc = ceph_fill_file_size(inode, issued,
2722 truncate_seq, truncate_size, size);
2723 spin_unlock(&ci->i_ceph_lock);
2724
2725 if (queue_trunc) {
2726 ceph_queue_vmtruncate(inode);
2727 ceph_fscache_invalidate(inode);
2728 }
2729}
2730
2731
2732
2733
2734
2735
2736
2737
2738
2739static void handle_cap_export(struct inode *inode, struct ceph_mds_caps *ex,
2740 struct ceph_mds_session *session,
2741 int *open_target_sessions)
2742{
2743 struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
2744 struct ceph_inode_info *ci = ceph_inode(inode);
2745 int mds = session->s_mds;
2746 unsigned mseq = le32_to_cpu(ex->migrate_seq);
2747 struct ceph_cap *cap = NULL, *t;
2748 struct rb_node *p;
2749 int remember = 1;
2750
2751 dout("handle_cap_export inode %p ci %p mds%d mseq %d\n",
2752 inode, ci, mds, mseq);
2753
2754 spin_lock(&ci->i_ceph_lock);
2755
2756
2757 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
2758 t = rb_entry(p, struct ceph_cap, ci_node);
2759 if (ceph_seq_cmp(t->mseq, mseq) > 0) {
2760 dout(" higher mseq on cap from mds%d\n",
2761 t->session->s_mds);
2762 remember = 0;
2763 }
2764 if (t->session->s_mds == mds)
2765 cap = t;
2766 }
2767
2768 if (cap) {
2769 if (remember) {
2770
2771 ci->i_cap_exporting_mds = mds;
2772 ci->i_cap_exporting_mseq = mseq;
2773 ci->i_cap_exporting_issued = cap->issued;
2774
2775
2776
2777
2778
2779 *open_target_sessions = 1;
2780
2781
2782
2783
2784
2785 spin_lock(&mdsc->cap_dirty_lock);
2786 if (!list_empty(&ci->i_dirty_item)) {
2787 dout(" moving %p to cap_dirty_migrating\n",
2788 inode);
2789 list_move(&ci->i_dirty_item,
2790 &mdsc->cap_dirty_migrating);
2791 }
2792 spin_unlock(&mdsc->cap_dirty_lock);
2793 }
2794 __ceph_remove_cap(cap);
2795 }
2796
2797
2798 spin_unlock(&ci->i_ceph_lock);
2799}
2800
2801
2802
2803
2804
2805
2806
2807static void handle_cap_import(struct ceph_mds_client *mdsc,
2808 struct inode *inode, struct ceph_mds_caps *im,
2809 struct ceph_mds_session *session,
2810 void *snaptrace, int snaptrace_len)
2811{
2812 struct ceph_inode_info *ci = ceph_inode(inode);
2813 int mds = session->s_mds;
2814 unsigned issued = le32_to_cpu(im->caps);
2815 unsigned wanted = le32_to_cpu(im->wanted);
2816 unsigned seq = le32_to_cpu(im->seq);
2817 unsigned mseq = le32_to_cpu(im->migrate_seq);
2818 u64 realmino = le64_to_cpu(im->realm);
2819 u64 cap_id = le64_to_cpu(im->cap_id);
2820
2821 if (ci->i_cap_exporting_mds >= 0 &&
2822 ceph_seq_cmp(ci->i_cap_exporting_mseq, mseq) < 0) {
2823 dout("handle_cap_import inode %p ci %p mds%d mseq %d"
2824 " - cleared exporting from mds%d\n",
2825 inode, ci, mds, mseq,
2826 ci->i_cap_exporting_mds);
2827 ci->i_cap_exporting_issued = 0;
2828 ci->i_cap_exporting_mseq = 0;
2829 ci->i_cap_exporting_mds = -1;
2830
2831 spin_lock(&mdsc->cap_dirty_lock);
2832 if (!list_empty(&ci->i_dirty_item)) {
2833 dout(" moving %p back to cap_dirty\n", inode);
2834 list_move(&ci->i_dirty_item, &mdsc->cap_dirty);
2835 }
2836 spin_unlock(&mdsc->cap_dirty_lock);
2837 } else {
2838 dout("handle_cap_import inode %p ci %p mds%d mseq %d\n",
2839 inode, ci, mds, mseq);
2840 }
2841
2842 down_write(&mdsc->snap_rwsem);
2843 ceph_update_snap_trace(mdsc, snaptrace, snaptrace+snaptrace_len,
2844 false);
2845 downgrade_write(&mdsc->snap_rwsem);
2846 ceph_add_cap(inode, session, cap_id, -1,
2847 issued, wanted, seq, mseq, realmino, CEPH_CAP_FLAG_AUTH,
2848 NULL );
2849 kick_flushing_inode_caps(mdsc, session, inode);
2850 up_read(&mdsc->snap_rwsem);
2851
2852
2853 spin_lock(&ci->i_ceph_lock);
2854 ci->i_wanted_max_size = 0;
2855 ci->i_requested_max_size = 0;
2856 spin_unlock(&ci->i_ceph_lock);
2857}
2858
2859
2860
2861
2862
2863
2864
2865void ceph_handle_caps(struct ceph_mds_session *session,
2866 struct ceph_msg *msg)
2867{
2868 struct ceph_mds_client *mdsc = session->s_mdsc;
2869 struct super_block *sb = mdsc->fsc->sb;
2870 struct inode *inode;
2871 struct ceph_inode_info *ci;
2872 struct ceph_cap *cap;
2873 struct ceph_mds_caps *h;
2874 int mds = session->s_mds;
2875 int op;
2876 u32 seq, mseq;
2877 struct ceph_vino vino;
2878 u64 cap_id;
2879 u64 size, max_size;
2880 u64 tid;
2881 void *snaptrace;
2882 size_t snaptrace_len;
2883 void *flock;
2884 u32 flock_len;
2885 int open_target_sessions = 0;
2886
2887 dout("handle_caps from mds%d\n", mds);
2888
2889
2890 tid = le64_to_cpu(msg->hdr.tid);
2891 if (msg->front.iov_len < sizeof(*h))
2892 goto bad;
2893 h = msg->front.iov_base;
2894 op = le32_to_cpu(h->op);
2895 vino.ino = le64_to_cpu(h->ino);
2896 vino.snap = CEPH_NOSNAP;
2897 cap_id = le64_to_cpu(h->cap_id);
2898 seq = le32_to_cpu(h->seq);
2899 mseq = le32_to_cpu(h->migrate_seq);
2900 size = le64_to_cpu(h->size);
2901 max_size = le64_to_cpu(h->max_size);
2902
2903 snaptrace = h + 1;
2904 snaptrace_len = le32_to_cpu(h->snap_trace_len);
2905
2906 if (le16_to_cpu(msg->hdr.version) >= 2) {
2907 void *p, *end;
2908
2909 p = snaptrace + snaptrace_len;
2910 end = msg->front.iov_base + msg->front.iov_len;
2911 ceph_decode_32_safe(&p, end, flock_len, bad);
2912 flock = p;
2913 } else {
2914 flock = NULL;
2915 flock_len = 0;
2916 }
2917
2918 mutex_lock(&session->s_mutex);
2919 session->s_seq++;
2920 dout(" mds%d seq %lld cap seq %u\n", session->s_mds, session->s_seq,
2921 (unsigned)seq);
2922
2923 if (op == CEPH_CAP_OP_IMPORT)
2924 ceph_add_cap_releases(mdsc, session);
2925
2926
2927 inode = ceph_find_inode(sb, vino);
2928 ci = ceph_inode(inode);
2929 dout(" op %s ino %llx.%llx inode %p\n", ceph_cap_op_name(op), vino.ino,
2930 vino.snap, inode);
2931 if (!inode) {
2932 dout(" i don't have ino %llx\n", vino.ino);
2933
2934 if (op == CEPH_CAP_OP_IMPORT)
2935 __queue_cap_release(session, vino.ino, cap_id,
2936 mseq, seq);
2937 goto flush_cap_releases;
2938 }
2939
2940
2941 switch (op) {
2942 case CEPH_CAP_OP_FLUSHSNAP_ACK:
2943 handle_cap_flushsnap_ack(inode, tid, h, session);
2944 goto done;
2945
2946 case CEPH_CAP_OP_EXPORT:
2947 handle_cap_export(inode, h, session, &open_target_sessions);
2948 goto done;
2949
2950 case CEPH_CAP_OP_IMPORT:
2951 handle_cap_import(mdsc, inode, h, session,
2952 snaptrace, snaptrace_len);
2953 }
2954
2955
2956 spin_lock(&ci->i_ceph_lock);
2957 cap = __get_cap_for_mds(ceph_inode(inode), mds);
2958 if (!cap) {
2959 dout(" no cap on %p ino %llx.%llx from mds%d\n",
2960 inode, ceph_ino(inode), ceph_snap(inode), mds);
2961 spin_unlock(&ci->i_ceph_lock);
2962 goto flush_cap_releases;
2963 }
2964
2965
2966 switch (op) {
2967 case CEPH_CAP_OP_REVOKE:
2968 case CEPH_CAP_OP_GRANT:
2969 case CEPH_CAP_OP_IMPORT:
2970 handle_cap_grant(inode, h, session, cap, msg->middle);
2971 goto done_unlocked;
2972
2973 case CEPH_CAP_OP_FLUSH_ACK:
2974 handle_cap_flush_ack(inode, tid, h, session, cap);
2975 break;
2976
2977 case CEPH_CAP_OP_TRUNC:
2978 handle_cap_trunc(inode, h, session);
2979 break;
2980
2981 default:
2982 spin_unlock(&ci->i_ceph_lock);
2983 pr_err("ceph_handle_caps: unknown cap op %d %s\n", op,
2984 ceph_cap_op_name(op));
2985 }
2986
2987 goto done;
2988
2989flush_cap_releases:
2990
2991
2992
2993
2994
2995 ceph_add_cap_releases(mdsc, session);
2996 ceph_send_cap_releases(mdsc, session);
2997
2998done:
2999 mutex_unlock(&session->s_mutex);
3000done_unlocked:
3001 if (inode)
3002 iput(inode);
3003 if (open_target_sessions)
3004 ceph_mdsc_open_export_target_sessions(mdsc, session);
3005 return;
3006
3007bad:
3008 pr_err("ceph_handle_caps: corrupt message\n");
3009 ceph_msg_dump(msg);
3010 return;
3011}
3012
3013
3014
3015
3016void ceph_check_delayed_caps(struct ceph_mds_client *mdsc)
3017{
3018 struct ceph_inode_info *ci;
3019 int flags = CHECK_CAPS_NODELAY;
3020
3021 dout("check_delayed_caps\n");
3022 while (1) {
3023 spin_lock(&mdsc->cap_delay_lock);
3024 if (list_empty(&mdsc->cap_delay_list))
3025 break;
3026 ci = list_first_entry(&mdsc->cap_delay_list,
3027 struct ceph_inode_info,
3028 i_cap_delay_list);
3029 if ((ci->i_ceph_flags & CEPH_I_FLUSH) == 0 &&
3030 time_before(jiffies, ci->i_hold_caps_max))
3031 break;
3032 list_del_init(&ci->i_cap_delay_list);
3033 spin_unlock(&mdsc->cap_delay_lock);
3034 dout("check_delayed_caps on %p\n", &ci->vfs_inode);
3035 ceph_check_caps(ci, flags, NULL);
3036 }
3037 spin_unlock(&mdsc->cap_delay_lock);
3038}
3039
3040
3041
3042
3043void ceph_flush_dirty_caps(struct ceph_mds_client *mdsc)
3044{
3045 struct ceph_inode_info *ci;
3046 struct inode *inode;
3047
3048 dout("flush_dirty_caps\n");
3049 spin_lock(&mdsc->cap_dirty_lock);
3050 while (!list_empty(&mdsc->cap_dirty)) {
3051 ci = list_first_entry(&mdsc->cap_dirty, struct ceph_inode_info,
3052 i_dirty_item);
3053 inode = &ci->vfs_inode;
3054 ihold(inode);
3055 dout("flush_dirty_caps %p\n", inode);
3056 spin_unlock(&mdsc->cap_dirty_lock);
3057 ceph_check_caps(ci, CHECK_CAPS_NODELAY|CHECK_CAPS_FLUSH, NULL);
3058 iput(inode);
3059 spin_lock(&mdsc->cap_dirty_lock);
3060 }
3061 spin_unlock(&mdsc->cap_dirty_lock);
3062 dout("flush_dirty_caps done\n");
3063}
3064
3065
3066
3067
3068
3069
3070void ceph_put_fmode(struct ceph_inode_info *ci, int fmode)
3071{
3072 struct inode *inode = &ci->vfs_inode;
3073 int last = 0;
3074
3075 spin_lock(&ci->i_ceph_lock);
3076 dout("put_fmode %p fmode %d %d -> %d\n", inode, fmode,
3077 ci->i_nr_by_mode[fmode], ci->i_nr_by_mode[fmode]-1);
3078 BUG_ON(ci->i_nr_by_mode[fmode] == 0);
3079 if (--ci->i_nr_by_mode[fmode] == 0)
3080 last++;
3081 spin_unlock(&ci->i_ceph_lock);
3082
3083 if (last && ci->i_vino.snap == CEPH_NOSNAP)
3084 ceph_check_caps(ci, 0, NULL);
3085}
3086
3087
3088
3089
3090
3091
3092
3093
3094
3095int ceph_encode_inode_release(void **p, struct inode *inode,
3096 int mds, int drop, int unless, int force)
3097{
3098 struct ceph_inode_info *ci = ceph_inode(inode);
3099 struct ceph_cap *cap;
3100 struct ceph_mds_request_release *rel = *p;
3101 int used, dirty;
3102 int ret = 0;
3103
3104 spin_lock(&ci->i_ceph_lock);
3105 used = __ceph_caps_used(ci);
3106 dirty = __ceph_caps_dirty(ci);
3107
3108 dout("encode_inode_release %p mds%d used|dirty %s drop %s unless %s\n",
3109 inode, mds, ceph_cap_string(used|dirty), ceph_cap_string(drop),
3110 ceph_cap_string(unless));
3111
3112
3113 drop &= ~(used | dirty);
3114
3115 cap = __get_cap_for_mds(ci, mds);
3116 if (cap && __cap_is_valid(cap)) {
3117 if (force ||
3118 ((cap->issued & drop) &&
3119 (cap->issued & unless) == 0)) {
3120 if ((cap->issued & drop) &&
3121 (cap->issued & unless) == 0) {
3122 int wanted = __ceph_caps_wanted(ci);
3123 if ((ci->i_ceph_flags & CEPH_I_NODELAY) == 0)
3124 wanted |= cap->mds_wanted;
3125 dout("encode_inode_release %p cap %p "
3126 "%s -> %s, wanted %s -> %s\n", inode, cap,
3127 ceph_cap_string(cap->issued),
3128 ceph_cap_string(cap->issued & ~drop),
3129 ceph_cap_string(cap->mds_wanted),
3130 ceph_cap_string(wanted));
3131
3132 cap->issued &= ~drop;
3133 cap->implemented &= ~drop;
3134 cap->mds_wanted = wanted;
3135 } else {
3136 dout("encode_inode_release %p cap %p %s"
3137 " (force)\n", inode, cap,
3138 ceph_cap_string(cap->issued));
3139 }
3140
3141 rel->ino = cpu_to_le64(ceph_ino(inode));
3142 rel->cap_id = cpu_to_le64(cap->cap_id);
3143 rel->seq = cpu_to_le32(cap->seq);
3144 rel->issue_seq = cpu_to_le32(cap->issue_seq),
3145 rel->mseq = cpu_to_le32(cap->mseq);
3146 rel->caps = cpu_to_le32(cap->issued);
3147 rel->wanted = cpu_to_le32(cap->mds_wanted);
3148 rel->dname_len = 0;
3149 rel->dname_seq = 0;
3150 *p += sizeof(*rel);
3151 ret = 1;
3152 } else {
3153 dout("encode_inode_release %p cap %p %s\n",
3154 inode, cap, ceph_cap_string(cap->issued));
3155 }
3156 }
3157 spin_unlock(&ci->i_ceph_lock);
3158 return ret;
3159}
3160
3161int ceph_encode_dentry_release(void **p, struct dentry *dentry,
3162 int mds, int drop, int unless)
3163{
3164 struct inode *dir = dentry->d_parent->d_inode;
3165 struct ceph_mds_request_release *rel = *p;
3166 struct ceph_dentry_info *di = ceph_dentry(dentry);
3167 int force = 0;
3168 int ret;
3169
3170
3171
3172
3173
3174
3175
3176 spin_lock(&dentry->d_lock);
3177 if (di->lease_session && di->lease_session->s_mds == mds)
3178 force = 1;
3179 spin_unlock(&dentry->d_lock);
3180
3181 ret = ceph_encode_inode_release(p, dir, mds, drop, unless, force);
3182
3183 spin_lock(&dentry->d_lock);
3184 if (ret && di->lease_session && di->lease_session->s_mds == mds) {
3185 dout("encode_dentry_release %p mds%d seq %d\n",
3186 dentry, mds, (int)di->lease_seq);
3187 rel->dname_len = cpu_to_le32(dentry->d_name.len);
3188 memcpy(*p, dentry->d_name.name, dentry->d_name.len);
3189 *p += dentry->d_name.len;
3190 rel->dname_seq = cpu_to_le32(di->lease_seq);
3191 __ceph_mdsc_drop_dentry_lease(dentry);
3192 }
3193 spin_unlock(&dentry->d_lock);
3194 return ret;
3195}
3196