blob: 30acc7b046eeb0a5f0afb035148cb5ae7d524c23 [file] [log] [blame]
Sage Weila8599bd2009-10-06 11:31:12 -07001#include "ceph_debug.h"
2
3#include <linux/fs.h>
4#include <linux/kernel.h>
5#include <linux/sched.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +09006#include <linux/slab.h>
Sage Weila8599bd2009-10-06 11:31:12 -07007#include <linux/vmalloc.h>
8#include <linux/wait.h>
Stephen Rothwellf1a3d572010-01-18 11:53:08 +11009#include <linux/writeback.h>
Sage Weila8599bd2009-10-06 11:31:12 -070010
11#include "super.h"
12#include "decode.h"
13#include "messenger.h"
14
15/*
16 * Capability management
17 *
18 * The Ceph metadata servers control client access to inode metadata
19 * and file data by issuing capabilities, granting clients permission
20 * to read and/or write both inode field and file data to OSDs
21 * (storage nodes). Each capability consists of a set of bits
22 * indicating which operations are allowed.
23 *
24 * If the client holds a *_SHARED cap, the client has a coherent value
25 * that can be safely read from the cached inode.
26 *
27 * In the case of a *_EXCL (exclusive) or FILE_WR capabilities, the
28 * client is allowed to change inode attributes (e.g., file size,
29 * mtime), note its dirty state in the ceph_cap, and asynchronously
30 * flush that metadata change to the MDS.
31 *
32 * In the event of a conflicting operation (perhaps by another
33 * client), the MDS will revoke the conflicting client capabilities.
34 *
35 * In order for a client to cache an inode, it must hold a capability
36 * with at least one MDS server. When inodes are released, release
37 * notifications are batched and periodically sent en masse to the MDS
38 * cluster to release server state.
39 */
40
41
42/*
43 * Generate readable cap strings for debugging output.
44 */
45#define MAX_CAP_STR 20
46static char cap_str[MAX_CAP_STR][40];
47static DEFINE_SPINLOCK(cap_str_lock);
48static int last_cap_str;
49
50static char *gcap_string(char *s, int c)
51{
52 if (c & CEPH_CAP_GSHARED)
53 *s++ = 's';
54 if (c & CEPH_CAP_GEXCL)
55 *s++ = 'x';
56 if (c & CEPH_CAP_GCACHE)
57 *s++ = 'c';
58 if (c & CEPH_CAP_GRD)
59 *s++ = 'r';
60 if (c & CEPH_CAP_GWR)
61 *s++ = 'w';
62 if (c & CEPH_CAP_GBUFFER)
63 *s++ = 'b';
64 if (c & CEPH_CAP_GLAZYIO)
65 *s++ = 'l';
66 return s;
67}
68
69const char *ceph_cap_string(int caps)
70{
71 int i;
72 char *s;
73 int c;
74
75 spin_lock(&cap_str_lock);
76 i = last_cap_str++;
77 if (last_cap_str == MAX_CAP_STR)
78 last_cap_str = 0;
79 spin_unlock(&cap_str_lock);
80
81 s = cap_str[i];
82
83 if (caps & CEPH_CAP_PIN)
84 *s++ = 'p';
85
86 c = (caps >> CEPH_CAP_SAUTH) & 3;
87 if (c) {
88 *s++ = 'A';
89 s = gcap_string(s, c);
90 }
91
92 c = (caps >> CEPH_CAP_SLINK) & 3;
93 if (c) {
94 *s++ = 'L';
95 s = gcap_string(s, c);
96 }
97
98 c = (caps >> CEPH_CAP_SXATTR) & 3;
99 if (c) {
100 *s++ = 'X';
101 s = gcap_string(s, c);
102 }
103
104 c = caps >> CEPH_CAP_SFILE;
105 if (c) {
106 *s++ = 'F';
107 s = gcap_string(s, c);
108 }
109
110 if (s == cap_str[i])
111 *s++ = '-';
112 *s = 0;
113 return cap_str[i];
114}
115
Yehuda Sadeh37151662010-06-17 16:16:12 -0700116void ceph_caps_init(struct ceph_mds_client *mdsc)
Sage Weila8599bd2009-10-06 11:31:12 -0700117{
Yehuda Sadeh37151662010-06-17 16:16:12 -0700118 INIT_LIST_HEAD(&mdsc->caps_list);
119 spin_lock_init(&mdsc->caps_list_lock);
Sage Weila8599bd2009-10-06 11:31:12 -0700120}
121
Yehuda Sadeh37151662010-06-17 16:16:12 -0700122void ceph_caps_finalize(struct ceph_mds_client *mdsc)
Sage Weila8599bd2009-10-06 11:31:12 -0700123{
124 struct ceph_cap *cap;
125
Yehuda Sadeh37151662010-06-17 16:16:12 -0700126 spin_lock(&mdsc->caps_list_lock);
127 while (!list_empty(&mdsc->caps_list)) {
128 cap = list_first_entry(&mdsc->caps_list,
129 struct ceph_cap, caps_item);
Sage Weila8599bd2009-10-06 11:31:12 -0700130 list_del(&cap->caps_item);
131 kmem_cache_free(ceph_cap_cachep, cap);
132 }
Yehuda Sadeh37151662010-06-17 16:16:12 -0700133 mdsc->caps_total_count = 0;
134 mdsc->caps_avail_count = 0;
135 mdsc->caps_use_count = 0;
136 mdsc->caps_reserve_count = 0;
137 mdsc->caps_min_count = 0;
138 spin_unlock(&mdsc->caps_list_lock);
Sage Weil85ccce42010-02-17 10:02:43 -0800139}
140
Yehuda Sadeh37151662010-06-17 16:16:12 -0700141void ceph_adjust_min_caps(struct ceph_mds_client *mdsc, int delta)
Sage Weil85ccce42010-02-17 10:02:43 -0800142{
Yehuda Sadeh37151662010-06-17 16:16:12 -0700143 spin_lock(&mdsc->caps_list_lock);
144 mdsc->caps_min_count += delta;
145 BUG_ON(mdsc->caps_min_count < 0);
146 spin_unlock(&mdsc->caps_list_lock);
Sage Weila8599bd2009-10-06 11:31:12 -0700147}
148
Yehuda Sadeh37151662010-06-17 16:16:12 -0700149int ceph_reserve_caps(struct ceph_mds_client *mdsc,
150 struct ceph_cap_reservation *ctx, int need)
Sage Weila8599bd2009-10-06 11:31:12 -0700151{
152 int i;
153 struct ceph_cap *cap;
154 int have;
155 int alloc = 0;
156 LIST_HEAD(newcaps);
157 int ret = 0;
158
159 dout("reserve caps ctx=%p need=%d\n", ctx, need);
160
161 /* first reserve any caps that are already allocated */
Yehuda Sadeh37151662010-06-17 16:16:12 -0700162 spin_lock(&mdsc->caps_list_lock);
163 if (mdsc->caps_avail_count >= need)
Sage Weila8599bd2009-10-06 11:31:12 -0700164 have = need;
165 else
Yehuda Sadeh37151662010-06-17 16:16:12 -0700166 have = mdsc->caps_avail_count;
167 mdsc->caps_avail_count -= have;
168 mdsc->caps_reserve_count += have;
169 BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count +
170 mdsc->caps_reserve_count +
171 mdsc->caps_avail_count);
172 spin_unlock(&mdsc->caps_list_lock);
Sage Weila8599bd2009-10-06 11:31:12 -0700173
174 for (i = have; i < need; i++) {
175 cap = kmem_cache_alloc(ceph_cap_cachep, GFP_NOFS);
176 if (!cap) {
177 ret = -ENOMEM;
178 goto out_alloc_count;
179 }
180 list_add(&cap->caps_item, &newcaps);
181 alloc++;
182 }
183 BUG_ON(have + alloc != need);
184
Yehuda Sadeh37151662010-06-17 16:16:12 -0700185 spin_lock(&mdsc->caps_list_lock);
186 mdsc->caps_total_count += alloc;
187 mdsc->caps_reserve_count += alloc;
188 list_splice(&newcaps, &mdsc->caps_list);
Sage Weila8599bd2009-10-06 11:31:12 -0700189
Yehuda Sadeh37151662010-06-17 16:16:12 -0700190 BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count +
191 mdsc->caps_reserve_count +
192 mdsc->caps_avail_count);
193 spin_unlock(&mdsc->caps_list_lock);
Sage Weila8599bd2009-10-06 11:31:12 -0700194
195 ctx->count = need;
196 dout("reserve caps ctx=%p %d = %d used + %d resv + %d avail\n",
Yehuda Sadeh37151662010-06-17 16:16:12 -0700197 ctx, mdsc->caps_total_count, mdsc->caps_use_count,
198 mdsc->caps_reserve_count, mdsc->caps_avail_count);
Sage Weila8599bd2009-10-06 11:31:12 -0700199 return 0;
200
201out_alloc_count:
202 /* we didn't manage to reserve as much as we needed */
203 pr_warning("reserve caps ctx=%p ENOMEM need=%d got=%d\n",
204 ctx, need, have);
205 return ret;
206}
207
Yehuda Sadeh37151662010-06-17 16:16:12 -0700208int ceph_unreserve_caps(struct ceph_mds_client *mdsc,
209 struct ceph_cap_reservation *ctx)
Sage Weila8599bd2009-10-06 11:31:12 -0700210{
211 dout("unreserve caps ctx=%p count=%d\n", ctx, ctx->count);
212 if (ctx->count) {
Yehuda Sadeh37151662010-06-17 16:16:12 -0700213 spin_lock(&mdsc->caps_list_lock);
214 BUG_ON(mdsc->caps_reserve_count < ctx->count);
215 mdsc->caps_reserve_count -= ctx->count;
216 mdsc->caps_avail_count += ctx->count;
Sage Weila8599bd2009-10-06 11:31:12 -0700217 ctx->count = 0;
218 dout("unreserve caps %d = %d used + %d resv + %d avail\n",
Yehuda Sadeh37151662010-06-17 16:16:12 -0700219 mdsc->caps_total_count, mdsc->caps_use_count,
220 mdsc->caps_reserve_count, mdsc->caps_avail_count);
221 BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count +
222 mdsc->caps_reserve_count +
223 mdsc->caps_avail_count);
224 spin_unlock(&mdsc->caps_list_lock);
Sage Weila8599bd2009-10-06 11:31:12 -0700225 }
226 return 0;
227}
228
Yehuda Sadeh37151662010-06-17 16:16:12 -0700229static struct ceph_cap *get_cap(struct ceph_mds_client *mdsc,
230 struct ceph_cap_reservation *ctx)
Sage Weila8599bd2009-10-06 11:31:12 -0700231{
232 struct ceph_cap *cap = NULL;
233
234 /* temporary, until we do something about cap import/export */
Sage Weil443b3762010-06-29 09:28:39 -0700235 if (!ctx) {
236 cap = kmem_cache_alloc(ceph_cap_cachep, GFP_NOFS);
237 if (cap) {
Yehuda Sadeh37151662010-06-17 16:16:12 -0700238 mdsc->caps_use_count++;
239 mdsc->caps_total_count++;
Sage Weil443b3762010-06-29 09:28:39 -0700240 }
241 return cap;
242 }
Sage Weila8599bd2009-10-06 11:31:12 -0700243
Yehuda Sadeh37151662010-06-17 16:16:12 -0700244 spin_lock(&mdsc->caps_list_lock);
Sage Weila8599bd2009-10-06 11:31:12 -0700245 dout("get_cap ctx=%p (%d) %d = %d used + %d resv + %d avail\n",
Yehuda Sadeh37151662010-06-17 16:16:12 -0700246 ctx, ctx->count, mdsc->caps_total_count, mdsc->caps_use_count,
247 mdsc->caps_reserve_count, mdsc->caps_avail_count);
Sage Weila8599bd2009-10-06 11:31:12 -0700248 BUG_ON(!ctx->count);
Yehuda Sadeh37151662010-06-17 16:16:12 -0700249 BUG_ON(ctx->count > mdsc->caps_reserve_count);
250 BUG_ON(list_empty(&mdsc->caps_list));
Sage Weila8599bd2009-10-06 11:31:12 -0700251
252 ctx->count--;
Yehuda Sadeh37151662010-06-17 16:16:12 -0700253 mdsc->caps_reserve_count--;
254 mdsc->caps_use_count++;
Sage Weila8599bd2009-10-06 11:31:12 -0700255
Yehuda Sadeh37151662010-06-17 16:16:12 -0700256 cap = list_first_entry(&mdsc->caps_list, struct ceph_cap, caps_item);
Sage Weila8599bd2009-10-06 11:31:12 -0700257 list_del(&cap->caps_item);
258
Yehuda Sadeh37151662010-06-17 16:16:12 -0700259 BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count +
260 mdsc->caps_reserve_count + mdsc->caps_avail_count);
261 spin_unlock(&mdsc->caps_list_lock);
Sage Weila8599bd2009-10-06 11:31:12 -0700262 return cap;
263}
264
Yehuda Sadeh37151662010-06-17 16:16:12 -0700265void ceph_put_cap(struct ceph_mds_client *mdsc, struct ceph_cap *cap)
Sage Weila8599bd2009-10-06 11:31:12 -0700266{
Yehuda Sadeh37151662010-06-17 16:16:12 -0700267 spin_lock(&mdsc->caps_list_lock);
Sage Weil7c1332b2010-02-16 11:39:45 -0800268 dout("put_cap %p %d = %d used + %d resv + %d avail\n",
Yehuda Sadeh37151662010-06-17 16:16:12 -0700269 cap, mdsc->caps_total_count, mdsc->caps_use_count,
270 mdsc->caps_reserve_count, mdsc->caps_avail_count);
271 mdsc->caps_use_count--;
Sage Weila8599bd2009-10-06 11:31:12 -0700272 /*
Sage Weil85ccce42010-02-17 10:02:43 -0800273 * Keep some preallocated caps around (ceph_min_count), to
274 * avoid lots of free/alloc churn.
Sage Weila8599bd2009-10-06 11:31:12 -0700275 */
Yehuda Sadeh37151662010-06-17 16:16:12 -0700276 if (mdsc->caps_avail_count >= mdsc->caps_reserve_count +
277 mdsc->caps_min_count) {
278 mdsc->caps_total_count--;
Sage Weila8599bd2009-10-06 11:31:12 -0700279 kmem_cache_free(ceph_cap_cachep, cap);
280 } else {
Yehuda Sadeh37151662010-06-17 16:16:12 -0700281 mdsc->caps_avail_count++;
282 list_add(&cap->caps_item, &mdsc->caps_list);
Sage Weila8599bd2009-10-06 11:31:12 -0700283 }
284
Yehuda Sadeh37151662010-06-17 16:16:12 -0700285 BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count +
286 mdsc->caps_reserve_count + mdsc->caps_avail_count);
287 spin_unlock(&mdsc->caps_list_lock);
Sage Weila8599bd2009-10-06 11:31:12 -0700288}
289
290void ceph_reservation_status(struct ceph_client *client,
Sage Weil85ccce42010-02-17 10:02:43 -0800291 int *total, int *avail, int *used, int *reserved,
292 int *min)
Sage Weila8599bd2009-10-06 11:31:12 -0700293{
Yehuda Sadeh37151662010-06-17 16:16:12 -0700294 struct ceph_mds_client *mdsc = &client->mdsc;
295
Sage Weila8599bd2009-10-06 11:31:12 -0700296 if (total)
Yehuda Sadeh37151662010-06-17 16:16:12 -0700297 *total = mdsc->caps_total_count;
Sage Weila8599bd2009-10-06 11:31:12 -0700298 if (avail)
Yehuda Sadeh37151662010-06-17 16:16:12 -0700299 *avail = mdsc->caps_avail_count;
Sage Weila8599bd2009-10-06 11:31:12 -0700300 if (used)
Yehuda Sadeh37151662010-06-17 16:16:12 -0700301 *used = mdsc->caps_use_count;
Sage Weila8599bd2009-10-06 11:31:12 -0700302 if (reserved)
Yehuda Sadeh37151662010-06-17 16:16:12 -0700303 *reserved = mdsc->caps_reserve_count;
Sage Weil85ccce42010-02-17 10:02:43 -0800304 if (min)
Yehuda Sadeh37151662010-06-17 16:16:12 -0700305 *min = mdsc->caps_min_count;
Sage Weila8599bd2009-10-06 11:31:12 -0700306}
307
308/*
309 * Find ceph_cap for given mds, if any.
310 *
311 * Called with i_lock held.
312 */
313static struct ceph_cap *__get_cap_for_mds(struct ceph_inode_info *ci, int mds)
314{
315 struct ceph_cap *cap;
316 struct rb_node *n = ci->i_caps.rb_node;
317
318 while (n) {
319 cap = rb_entry(n, struct ceph_cap, ci_node);
320 if (mds < cap->mds)
321 n = n->rb_left;
322 else if (mds > cap->mds)
323 n = n->rb_right;
324 else
325 return cap;
326 }
327 return NULL;
328}
329
Greg Farnum2bc50252010-06-30 12:44:34 -0700330struct ceph_cap *ceph_get_cap_for_mds(struct ceph_inode_info *ci, int mds)
331{
332 struct ceph_cap *cap;
333
334 spin_lock(&ci->vfs_inode.i_lock);
335 cap = __get_cap_for_mds(ci, mds);
336 spin_unlock(&ci->vfs_inode.i_lock);
337 return cap;
338}
339
Sage Weila8599bd2009-10-06 11:31:12 -0700340/*
Sage Weil33caad32010-05-26 14:31:27 -0700341 * Return id of any MDS with a cap, preferably FILE_WR|BUFFER|EXCL, else -1.
Sage Weila8599bd2009-10-06 11:31:12 -0700342 */
Sage Weilca81f3f2010-06-10 14:21:36 -0700343static int __ceph_get_cap_mds(struct ceph_inode_info *ci)
Sage Weila8599bd2009-10-06 11:31:12 -0700344{
345 struct ceph_cap *cap;
346 int mds = -1;
347 struct rb_node *p;
348
Sage Weil33caad32010-05-26 14:31:27 -0700349 /* prefer mds with WR|BUFFER|EXCL caps */
Sage Weila8599bd2009-10-06 11:31:12 -0700350 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
351 cap = rb_entry(p, struct ceph_cap, ci_node);
352 mds = cap->mds;
Sage Weila8599bd2009-10-06 11:31:12 -0700353 if (cap->issued & (CEPH_CAP_FILE_WR |
354 CEPH_CAP_FILE_BUFFER |
355 CEPH_CAP_FILE_EXCL))
356 break;
357 }
358 return mds;
359}
360
361int ceph_get_cap_mds(struct inode *inode)
362{
363 int mds;
364 spin_lock(&inode->i_lock);
Sage Weilca81f3f2010-06-10 14:21:36 -0700365 mds = __ceph_get_cap_mds(ceph_inode(inode));
Sage Weila8599bd2009-10-06 11:31:12 -0700366 spin_unlock(&inode->i_lock);
367 return mds;
368}
369
370/*
371 * Called under i_lock.
372 */
373static void __insert_cap_node(struct ceph_inode_info *ci,
374 struct ceph_cap *new)
375{
376 struct rb_node **p = &ci->i_caps.rb_node;
377 struct rb_node *parent = NULL;
378 struct ceph_cap *cap = NULL;
379
380 while (*p) {
381 parent = *p;
382 cap = rb_entry(parent, struct ceph_cap, ci_node);
383 if (new->mds < cap->mds)
384 p = &(*p)->rb_left;
385 else if (new->mds > cap->mds)
386 p = &(*p)->rb_right;
387 else
388 BUG();
389 }
390
391 rb_link_node(&new->ci_node, parent, p);
392 rb_insert_color(&new->ci_node, &ci->i_caps);
393}
394
395/*
396 * (re)set cap hold timeouts, which control the delayed release
397 * of unused caps back to the MDS. Should be called on cap use.
398 */
399static void __cap_set_timeouts(struct ceph_mds_client *mdsc,
400 struct ceph_inode_info *ci)
401{
Sage Weil6b805182009-10-27 11:50:50 -0700402 struct ceph_mount_args *ma = mdsc->client->mount_args;
Sage Weila8599bd2009-10-06 11:31:12 -0700403
404 ci->i_hold_caps_min = round_jiffies(jiffies +
405 ma->caps_wanted_delay_min * HZ);
406 ci->i_hold_caps_max = round_jiffies(jiffies +
407 ma->caps_wanted_delay_max * HZ);
408 dout("__cap_set_timeouts %p min %lu max %lu\n", &ci->vfs_inode,
409 ci->i_hold_caps_min - jiffies, ci->i_hold_caps_max - jiffies);
410}
411
412/*
413 * (Re)queue cap at the end of the delayed cap release list.
414 *
415 * If I_FLUSH is set, leave the inode at the front of the list.
416 *
417 * Caller holds i_lock
418 * -> we take mdsc->cap_delay_lock
419 */
420static void __cap_delay_requeue(struct ceph_mds_client *mdsc,
421 struct ceph_inode_info *ci)
422{
423 __cap_set_timeouts(mdsc, ci);
424 dout("__cap_delay_requeue %p flags %d at %lu\n", &ci->vfs_inode,
425 ci->i_ceph_flags, ci->i_hold_caps_max);
426 if (!mdsc->stopping) {
427 spin_lock(&mdsc->cap_delay_lock);
428 if (!list_empty(&ci->i_cap_delay_list)) {
429 if (ci->i_ceph_flags & CEPH_I_FLUSH)
430 goto no_change;
431 list_del_init(&ci->i_cap_delay_list);
432 }
433 list_add_tail(&ci->i_cap_delay_list, &mdsc->cap_delay_list);
434no_change:
435 spin_unlock(&mdsc->cap_delay_lock);
436 }
437}
438
439/*
440 * Queue an inode for immediate writeback. Mark inode with I_FLUSH,
441 * indicating we should send a cap message to flush dirty metadata
442 * asap, and move to the front of the delayed cap list.
443 */
444static void __cap_delay_requeue_front(struct ceph_mds_client *mdsc,
445 struct ceph_inode_info *ci)
446{
447 dout("__cap_delay_requeue_front %p\n", &ci->vfs_inode);
448 spin_lock(&mdsc->cap_delay_lock);
449 ci->i_ceph_flags |= CEPH_I_FLUSH;
450 if (!list_empty(&ci->i_cap_delay_list))
451 list_del_init(&ci->i_cap_delay_list);
452 list_add(&ci->i_cap_delay_list, &mdsc->cap_delay_list);
453 spin_unlock(&mdsc->cap_delay_lock);
454}
455
456/*
457 * Cancel delayed work on cap.
458 *
459 * Caller must hold i_lock.
460 */
461static void __cap_delay_cancel(struct ceph_mds_client *mdsc,
462 struct ceph_inode_info *ci)
463{
464 dout("__cap_delay_cancel %p\n", &ci->vfs_inode);
465 if (list_empty(&ci->i_cap_delay_list))
466 return;
467 spin_lock(&mdsc->cap_delay_lock);
468 list_del_init(&ci->i_cap_delay_list);
469 spin_unlock(&mdsc->cap_delay_lock);
470}
471
472/*
473 * Common issue checks for add_cap, handle_cap_grant.
474 */
475static void __check_cap_issue(struct ceph_inode_info *ci, struct ceph_cap *cap,
476 unsigned issued)
477{
478 unsigned had = __ceph_caps_issued(ci, NULL);
479
480 /*
481 * Each time we receive FILE_CACHE anew, we increment
482 * i_rdcache_gen.
483 */
Sage Weil29625072010-05-27 10:40:43 -0700484 if ((issued & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) &&
485 (had & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) == 0)
Sage Weila8599bd2009-10-06 11:31:12 -0700486 ci->i_rdcache_gen++;
487
488 /*
489 * if we are newly issued FILE_SHARED, clear I_COMPLETE; we
490 * don't know what happened to this directory while we didn't
491 * have the cap.
492 */
493 if ((issued & CEPH_CAP_FILE_SHARED) &&
494 (had & CEPH_CAP_FILE_SHARED) == 0) {
495 ci->i_shared_gen++;
496 if (S_ISDIR(ci->vfs_inode.i_mode)) {
497 dout(" marking %p NOT complete\n", &ci->vfs_inode);
498 ci->i_ceph_flags &= ~CEPH_I_COMPLETE;
499 }
500 }
501}
502
503/*
504 * Add a capability under the given MDS session.
505 *
506 * Caller should hold session snap_rwsem (read) and s_mutex.
507 *
508 * @fmode is the open file mode, if we are opening a file, otherwise
509 * it is < 0. (This is so we can atomically add the cap and add an
510 * open file reference to it.)
511 */
512int ceph_add_cap(struct inode *inode,
513 struct ceph_mds_session *session, u64 cap_id,
514 int fmode, unsigned issued, unsigned wanted,
515 unsigned seq, unsigned mseq, u64 realmino, int flags,
516 struct ceph_cap_reservation *caps_reservation)
517{
518 struct ceph_mds_client *mdsc = &ceph_inode_to_client(inode)->mdsc;
519 struct ceph_inode_info *ci = ceph_inode(inode);
520 struct ceph_cap *new_cap = NULL;
521 struct ceph_cap *cap;
522 int mds = session->s_mds;
523 int actual_wanted;
524
525 dout("add_cap %p mds%d cap %llx %s seq %d\n", inode,
526 session->s_mds, cap_id, ceph_cap_string(issued), seq);
527
528 /*
529 * If we are opening the file, include file mode wanted bits
530 * in wanted.
531 */
532 if (fmode >= 0)
533 wanted |= ceph_caps_for_mode(fmode);
534
535retry:
536 spin_lock(&inode->i_lock);
537 cap = __get_cap_for_mds(ci, mds);
538 if (!cap) {
539 if (new_cap) {
540 cap = new_cap;
541 new_cap = NULL;
542 } else {
543 spin_unlock(&inode->i_lock);
Yehuda Sadeh37151662010-06-17 16:16:12 -0700544 new_cap = get_cap(mdsc, caps_reservation);
Sage Weila8599bd2009-10-06 11:31:12 -0700545 if (new_cap == NULL)
546 return -ENOMEM;
547 goto retry;
548 }
549
550 cap->issued = 0;
551 cap->implemented = 0;
552 cap->mds = mds;
553 cap->mds_wanted = 0;
554
555 cap->ci = ci;
556 __insert_cap_node(ci, cap);
557
558 /* clear out old exporting info? (i.e. on cap import) */
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 /* add to session cap list */
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 }
572
573 if (!ci->i_snap_realm) {
574 /*
575 * add this inode to the appropriate snap realm
576 */
577 struct ceph_snap_realm *realm = ceph_lookup_snap_realm(mdsc,
578 realmino);
579 if (realm) {
580 ceph_get_snap_realm(mdsc, realm);
581 spin_lock(&realm->inodes_with_caps_lock);
582 ci->i_snap_realm = realm;
583 list_add(&ci->i_snap_realm_item,
584 &realm->inodes_with_caps);
585 spin_unlock(&realm->inodes_with_caps_lock);
586 } else {
587 pr_err("ceph_add_cap: couldn't find snap realm %llx\n",
588 realmino);
Sage Weilb8cd07e2010-07-16 12:00:02 -0700589 WARN_ON(!realm);
Sage Weila8599bd2009-10-06 11:31:12 -0700590 }
591 }
592
593 __check_cap_issue(ci, cap, issued);
594
595 /*
596 * If we are issued caps we don't want, or the mds' wanted
597 * value appears to be off, queue a check so we'll release
598 * later and/or update the mds wanted value.
599 */
600 actual_wanted = __ceph_caps_wanted(ci);
601 if ((wanted & ~actual_wanted) ||
602 (issued & ~actual_wanted & CEPH_CAP_ANY_WR)) {
603 dout(" issued %s, mds wanted %s, actual %s, queueing\n",
604 ceph_cap_string(issued), ceph_cap_string(wanted),
605 ceph_cap_string(actual_wanted));
606 __cap_delay_requeue(mdsc, ci);
607 }
608
609 if (flags & CEPH_CAP_FLAG_AUTH)
610 ci->i_auth_cap = cap;
611 else if (ci->i_auth_cap == cap)
612 ci->i_auth_cap = NULL;
613
614 dout("add_cap inode %p (%llx.%llx) cap %p %s now %s seq %d mds%d\n",
615 inode, ceph_vinop(inode), cap, ceph_cap_string(issued),
616 ceph_cap_string(issued|cap->issued), seq, mds);
617 cap->cap_id = cap_id;
618 cap->issued = issued;
619 cap->implemented |= issued;
620 cap->mds_wanted |= wanted;
621 cap->seq = seq;
622 cap->issue_seq = seq;
623 cap->mseq = mseq;
Sage Weil685f9a5d2009-11-09 12:05:48 -0800624 cap->cap_gen = session->s_cap_gen;
Sage Weila8599bd2009-10-06 11:31:12 -0700625
626 if (fmode >= 0)
627 __ceph_get_fmode(ci, fmode);
628 spin_unlock(&inode->i_lock);
Yehuda Sadeh03066f22010-07-27 13:11:08 -0700629 wake_up_all(&ci->i_cap_wq);
Sage Weila8599bd2009-10-06 11:31:12 -0700630 return 0;
631}
632
633/*
634 * Return true if cap has not timed out and belongs to the current
635 * generation of the MDS session (i.e. has not gone 'stale' due to
636 * us losing touch with the mds).
637 */
638static int __cap_is_valid(struct ceph_cap *cap)
639{
640 unsigned long ttl;
Sage Weilcdac8302009-11-10 16:02:23 -0800641 u32 gen;
Sage Weila8599bd2009-10-06 11:31:12 -0700642
643 spin_lock(&cap->session->s_cap_lock);
644 gen = cap->session->s_cap_gen;
645 ttl = cap->session->s_cap_ttl;
646 spin_unlock(&cap->session->s_cap_lock);
647
Sage Weil685f9a5d2009-11-09 12:05:48 -0800648 if (cap->cap_gen < gen || time_after_eq(jiffies, ttl)) {
Sage Weila8599bd2009-10-06 11:31:12 -0700649 dout("__cap_is_valid %p cap %p issued %s "
650 "but STALE (gen %u vs %u)\n", &cap->ci->vfs_inode,
Sage Weil685f9a5d2009-11-09 12:05:48 -0800651 cap, ceph_cap_string(cap->issued), cap->cap_gen, gen);
Sage Weila8599bd2009-10-06 11:31:12 -0700652 return 0;
653 }
654
655 return 1;
656}
657
658/*
659 * Return set of valid cap bits issued to us. Note that caps time
660 * out, and may be invalidated in bulk if the client session times out
661 * and session->s_cap_gen is bumped.
662 */
663int __ceph_caps_issued(struct ceph_inode_info *ci, int *implemented)
664{
Sage Weil7af8f1e2010-03-01 15:17:34 -0800665 int have = ci->i_snap_caps | ci->i_cap_exporting_issued;
Sage Weila8599bd2009-10-06 11:31:12 -0700666 struct ceph_cap *cap;
667 struct rb_node *p;
668
669 if (implemented)
670 *implemented = 0;
671 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
672 cap = rb_entry(p, struct ceph_cap, ci_node);
673 if (!__cap_is_valid(cap))
674 continue;
675 dout("__ceph_caps_issued %p cap %p issued %s\n",
676 &ci->vfs_inode, cap, ceph_cap_string(cap->issued));
677 have |= cap->issued;
678 if (implemented)
679 *implemented |= cap->implemented;
680 }
681 return have;
682}
683
684/*
685 * Get cap bits issued by caps other than @ocap
686 */
687int __ceph_caps_issued_other(struct ceph_inode_info *ci, struct ceph_cap *ocap)
688{
689 int have = ci->i_snap_caps;
690 struct ceph_cap *cap;
691 struct rb_node *p;
692
693 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
694 cap = rb_entry(p, struct ceph_cap, ci_node);
695 if (cap == ocap)
696 continue;
697 if (!__cap_is_valid(cap))
698 continue;
699 have |= cap->issued;
700 }
701 return have;
702}
703
704/*
705 * Move a cap to the end of the LRU (oldest caps at list head, newest
706 * at list tail).
707 */
708static void __touch_cap(struct ceph_cap *cap)
709{
710 struct ceph_mds_session *s = cap->session;
711
Sage Weila8599bd2009-10-06 11:31:12 -0700712 spin_lock(&s->s_cap_lock);
Sage Weil7c1332b2010-02-16 11:39:45 -0800713 if (s->s_cap_iterator == NULL) {
Sage Weil5dacf092009-12-21 20:40:34 -0800714 dout("__touch_cap %p cap %p mds%d\n", &cap->ci->vfs_inode, cap,
715 s->s_mds);
716 list_move_tail(&cap->session_caps, &s->s_caps);
717 } else {
718 dout("__touch_cap %p cap %p mds%d NOP, iterating over caps\n",
719 &cap->ci->vfs_inode, cap, s->s_mds);
720 }
Sage Weila8599bd2009-10-06 11:31:12 -0700721 spin_unlock(&s->s_cap_lock);
722}
723
724/*
725 * Check if we hold the given mask. If so, move the cap(s) to the
726 * front of their respective LRUs. (This is the preferred way for
727 * callers to check for caps they want.)
728 */
729int __ceph_caps_issued_mask(struct ceph_inode_info *ci, int mask, int touch)
730{
731 struct ceph_cap *cap;
732 struct rb_node *p;
733 int have = ci->i_snap_caps;
734
735 if ((have & mask) == mask) {
736 dout("__ceph_caps_issued_mask %p snap issued %s"
737 " (mask %s)\n", &ci->vfs_inode,
738 ceph_cap_string(have),
739 ceph_cap_string(mask));
740 return 1;
741 }
742
743 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
744 cap = rb_entry(p, struct ceph_cap, ci_node);
745 if (!__cap_is_valid(cap))
746 continue;
747 if ((cap->issued & mask) == mask) {
748 dout("__ceph_caps_issued_mask %p cap %p issued %s"
749 " (mask %s)\n", &ci->vfs_inode, cap,
750 ceph_cap_string(cap->issued),
751 ceph_cap_string(mask));
752 if (touch)
753 __touch_cap(cap);
754 return 1;
755 }
756
757 /* does a combination of caps satisfy mask? */
758 have |= cap->issued;
759 if ((have & mask) == mask) {
760 dout("__ceph_caps_issued_mask %p combo issued %s"
761 " (mask %s)\n", &ci->vfs_inode,
762 ceph_cap_string(cap->issued),
763 ceph_cap_string(mask));
764 if (touch) {
765 struct rb_node *q;
766
767 /* touch this + preceeding caps */
768 __touch_cap(cap);
769 for (q = rb_first(&ci->i_caps); q != p;
770 q = rb_next(q)) {
771 cap = rb_entry(q, struct ceph_cap,
772 ci_node);
773 if (!__cap_is_valid(cap))
774 continue;
775 __touch_cap(cap);
776 }
777 }
778 return 1;
779 }
780 }
781
782 return 0;
783}
784
785/*
786 * Return true if mask caps are currently being revoked by an MDS.
787 */
788int ceph_caps_revoking(struct ceph_inode_info *ci, int mask)
789{
790 struct inode *inode = &ci->vfs_inode;
791 struct ceph_cap *cap;
792 struct rb_node *p;
793 int ret = 0;
794
795 spin_lock(&inode->i_lock);
796 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
797 cap = rb_entry(p, struct ceph_cap, ci_node);
798 if (__cap_is_valid(cap) &&
799 (cap->implemented & ~cap->issued & mask)) {
800 ret = 1;
801 break;
802 }
803 }
804 spin_unlock(&inode->i_lock);
805 dout("ceph_caps_revoking %p %s = %d\n", inode,
806 ceph_cap_string(mask), ret);
807 return ret;
808}
809
810int __ceph_caps_used(struct ceph_inode_info *ci)
811{
812 int used = 0;
813 if (ci->i_pin_ref)
814 used |= CEPH_CAP_PIN;
815 if (ci->i_rd_ref)
816 used |= CEPH_CAP_FILE_RD;
817 if (ci->i_rdcache_ref || ci->i_rdcache_gen)
818 used |= CEPH_CAP_FILE_CACHE;
819 if (ci->i_wr_ref)
820 used |= CEPH_CAP_FILE_WR;
821 if (ci->i_wrbuffer_ref)
822 used |= CEPH_CAP_FILE_BUFFER;
823 return used;
824}
825
826/*
827 * wanted, by virtue of open file modes
828 */
829int __ceph_caps_file_wanted(struct ceph_inode_info *ci)
830{
831 int want = 0;
832 int mode;
Sage Weil33caad32010-05-26 14:31:27 -0700833 for (mode = 0; mode < CEPH_FILE_MODE_NUM; mode++)
Sage Weila8599bd2009-10-06 11:31:12 -0700834 if (ci->i_nr_by_mode[mode])
835 want |= ceph_caps_for_mode(mode);
836 return want;
837}
838
839/*
840 * Return caps we have registered with the MDS(s) as 'wanted'.
841 */
842int __ceph_caps_mds_wanted(struct ceph_inode_info *ci)
843{
844 struct ceph_cap *cap;
845 struct rb_node *p;
846 int mds_wanted = 0;
847
848 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
849 cap = rb_entry(p, struct ceph_cap, ci_node);
850 if (!__cap_is_valid(cap))
851 continue;
852 mds_wanted |= cap->mds_wanted;
853 }
854 return mds_wanted;
855}
856
857/*
858 * called under i_lock
859 */
860static int __ceph_is_any_caps(struct ceph_inode_info *ci)
861{
862 return !RB_EMPTY_ROOT(&ci->i_caps) || ci->i_cap_exporting_mds >= 0;
863}
864
865/*
Sage Weilf818a732010-05-11 20:56:31 -0700866 * Remove a cap. Take steps to deal with a racing iterate_session_caps.
867 *
Sage Weila6369742010-02-22 13:59:00 -0800868 * caller should hold i_lock.
869 * caller will not hold session s_mutex if called from destroy_inode.
Sage Weila8599bd2009-10-06 11:31:12 -0700870 */
Sage Weil7c1332b2010-02-16 11:39:45 -0800871void __ceph_remove_cap(struct ceph_cap *cap)
Sage Weila8599bd2009-10-06 11:31:12 -0700872{
873 struct ceph_mds_session *session = cap->session;
874 struct ceph_inode_info *ci = cap->ci;
Cheng Renquan640ef792010-03-26 17:40:33 +0800875 struct ceph_mds_client *mdsc =
876 &ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc;
Sage Weilf818a732010-05-11 20:56:31 -0700877 int removed = 0;
Sage Weila8599bd2009-10-06 11:31:12 -0700878
879 dout("__ceph_remove_cap %p from %p\n", cap, &ci->vfs_inode);
880
Sage Weil7c1332b2010-02-16 11:39:45 -0800881 /* remove from session list */
882 spin_lock(&session->s_cap_lock);
883 if (session->s_cap_iterator == cap) {
884 /* not yet, we are iterating over this very cap */
885 dout("__ceph_remove_cap delaying %p removal from session %p\n",
886 cap, cap->session);
887 } else {
888 list_del_init(&cap->session_caps);
889 session->s_nr_caps--;
890 cap->session = NULL;
Sage Weilf818a732010-05-11 20:56:31 -0700891 removed = 1;
Sage Weil7c1332b2010-02-16 11:39:45 -0800892 }
Sage Weilf818a732010-05-11 20:56:31 -0700893 /* protect backpointer with s_cap_lock: see iterate_session_caps */
894 cap->ci = NULL;
Sage Weil7c1332b2010-02-16 11:39:45 -0800895 spin_unlock(&session->s_cap_lock);
896
Sage Weilf818a732010-05-11 20:56:31 -0700897 /* remove from inode list */
898 rb_erase(&cap->ci_node, &ci->i_caps);
899 if (ci->i_auth_cap == cap)
900 ci->i_auth_cap = NULL;
901
902 if (removed)
Yehuda Sadeh37151662010-06-17 16:16:12 -0700903 ceph_put_cap(mdsc, cap);
Sage Weila8599bd2009-10-06 11:31:12 -0700904
905 if (!__ceph_is_any_caps(ci) && ci->i_snap_realm) {
906 struct ceph_snap_realm *realm = ci->i_snap_realm;
907 spin_lock(&realm->inodes_with_caps_lock);
908 list_del_init(&ci->i_snap_realm_item);
909 ci->i_snap_realm_counter++;
910 ci->i_snap_realm = NULL;
911 spin_unlock(&realm->inodes_with_caps_lock);
912 ceph_put_snap_realm(mdsc, realm);
913 }
914 if (!__ceph_is_any_real_caps(ci))
915 __cap_delay_cancel(mdsc, ci);
916}
917
918/*
919 * Build and send a cap message to the given MDS.
920 *
921 * Caller should be holding s_mutex.
922 */
923static int send_cap_msg(struct ceph_mds_session *session,
924 u64 ino, u64 cid, int op,
925 int caps, int wanted, int dirty,
926 u32 seq, u64 flush_tid, u32 issue_seq, u32 mseq,
927 u64 size, u64 max_size,
928 struct timespec *mtime, struct timespec *atime,
929 u64 time_warp_seq,
930 uid_t uid, gid_t gid, mode_t mode,
931 u64 xattr_version,
932 struct ceph_buffer *xattrs_buf,
933 u64 follows)
934{
935 struct ceph_mds_caps *fc;
936 struct ceph_msg *msg;
937
938 dout("send_cap_msg %s %llx %llx caps %s wanted %s dirty %s"
939 " seq %u/%u mseq %u follows %lld size %llu/%llu"
940 " xattr_ver %llu xattr_len %d\n", ceph_cap_op_name(op),
941 cid, ino, ceph_cap_string(caps), ceph_cap_string(wanted),
942 ceph_cap_string(dirty),
943 seq, issue_seq, mseq, follows, size, max_size,
944 xattr_version, xattrs_buf ? (int)xattrs_buf->vec.iov_len : 0);
945
Yehuda Sadeh34d23762010-04-06 14:33:58 -0700946 msg = ceph_msg_new(CEPH_MSG_CLIENT_CAPS, sizeof(*fc), GFP_NOFS);
Sage Weila79832f2010-04-01 16:06:19 -0700947 if (!msg)
948 return -ENOMEM;
Sage Weila8599bd2009-10-06 11:31:12 -0700949
Sage Weil6df058c2009-12-22 11:24:33 -0800950 msg->hdr.tid = cpu_to_le64(flush_tid);
Sage Weila8599bd2009-10-06 11:31:12 -0700951
Sage Weil6df058c2009-12-22 11:24:33 -0800952 fc = msg->front.iov_base;
Sage Weila8599bd2009-10-06 11:31:12 -0700953 memset(fc, 0, sizeof(*fc));
954
955 fc->cap_id = cpu_to_le64(cid);
956 fc->op = cpu_to_le32(op);
957 fc->seq = cpu_to_le32(seq);
Sage Weila8599bd2009-10-06 11:31:12 -0700958 fc->issue_seq = cpu_to_le32(issue_seq);
959 fc->migrate_seq = cpu_to_le32(mseq);
960 fc->caps = cpu_to_le32(caps);
961 fc->wanted = cpu_to_le32(wanted);
962 fc->dirty = cpu_to_le32(dirty);
963 fc->ino = cpu_to_le64(ino);
964 fc->snap_follows = cpu_to_le64(follows);
965
966 fc->size = cpu_to_le64(size);
967 fc->max_size = cpu_to_le64(max_size);
968 if (mtime)
969 ceph_encode_timespec(&fc->mtime, mtime);
970 if (atime)
971 ceph_encode_timespec(&fc->atime, atime);
972 fc->time_warp_seq = cpu_to_le32(time_warp_seq);
973
974 fc->uid = cpu_to_le32(uid);
975 fc->gid = cpu_to_le32(gid);
976 fc->mode = cpu_to_le32(mode);
977
978 fc->xattr_version = cpu_to_le64(xattr_version);
979 if (xattrs_buf) {
980 msg->middle = ceph_buffer_get(xattrs_buf);
981 fc->xattr_len = cpu_to_le32(xattrs_buf->vec.iov_len);
982 msg->hdr.middle_len = cpu_to_le32(xattrs_buf->vec.iov_len);
983 }
984
985 ceph_con_send(&session->s_con, msg);
986 return 0;
987}
988
Sage Weil3d7ded42010-06-09 16:47:10 -0700989static void __queue_cap_release(struct ceph_mds_session *session,
990 u64 ino, u64 cap_id, u32 migrate_seq,
991 u32 issue_seq)
992{
993 struct ceph_msg *msg;
994 struct ceph_mds_cap_release *head;
995 struct ceph_mds_cap_item *item;
996
997 spin_lock(&session->s_cap_lock);
998 BUG_ON(!session->s_num_cap_releases);
999 msg = list_first_entry(&session->s_cap_releases,
1000 struct ceph_msg, list_head);
1001
1002 dout(" adding %llx release to mds%d msg %p (%d left)\n",
1003 ino, session->s_mds, msg, session->s_num_cap_releases);
1004
1005 BUG_ON(msg->front.iov_len + sizeof(*item) > PAGE_CACHE_SIZE);
1006 head = msg->front.iov_base;
1007 head->num = cpu_to_le32(le32_to_cpu(head->num) + 1);
1008 item = msg->front.iov_base + msg->front.iov_len;
1009 item->ino = cpu_to_le64(ino);
1010 item->cap_id = cpu_to_le64(cap_id);
1011 item->migrate_seq = cpu_to_le32(migrate_seq);
1012 item->seq = cpu_to_le32(issue_seq);
1013
1014 session->s_num_cap_releases--;
1015
1016 msg->front.iov_len += sizeof(*item);
1017 if (le32_to_cpu(head->num) == CEPH_CAPS_PER_RELEASE) {
1018 dout(" release msg %p full\n", msg);
1019 list_move_tail(&msg->list_head, &session->s_cap_releases_done);
1020 } else {
1021 dout(" release msg %p at %d/%d (%d)\n", msg,
1022 (int)le32_to_cpu(head->num),
1023 (int)CEPH_CAPS_PER_RELEASE,
1024 (int)msg->front.iov_len);
1025 }
1026 spin_unlock(&session->s_cap_lock);
1027}
1028
Sage Weila8599bd2009-10-06 11:31:12 -07001029/*
Sage Weila6369742010-02-22 13:59:00 -08001030 * Queue cap releases when an inode is dropped from our cache. Since
1031 * inode is about to be destroyed, there is no need for i_lock.
Sage Weila8599bd2009-10-06 11:31:12 -07001032 */
1033void ceph_queue_caps_release(struct inode *inode)
1034{
1035 struct ceph_inode_info *ci = ceph_inode(inode);
1036 struct rb_node *p;
1037
Sage Weila8599bd2009-10-06 11:31:12 -07001038 p = rb_first(&ci->i_caps);
1039 while (p) {
1040 struct ceph_cap *cap = rb_entry(p, struct ceph_cap, ci_node);
1041 struct ceph_mds_session *session = cap->session;
Sage Weila8599bd2009-10-06 11:31:12 -07001042
Sage Weil3d7ded42010-06-09 16:47:10 -07001043 __queue_cap_release(session, ceph_ino(inode), cap->cap_id,
1044 cap->mseq, cap->issue_seq);
Sage Weila8599bd2009-10-06 11:31:12 -07001045 p = rb_next(p);
Sage Weil7c1332b2010-02-16 11:39:45 -08001046 __ceph_remove_cap(cap);
Sage Weila8599bd2009-10-06 11:31:12 -07001047 }
Sage Weila8599bd2009-10-06 11:31:12 -07001048}
1049
1050/*
1051 * Send a cap msg on the given inode. Update our caps state, then
1052 * drop i_lock and send the message.
1053 *
1054 * Make note of max_size reported/requested from mds, revoked caps
1055 * that have now been implemented.
1056 *
1057 * Make half-hearted attempt ot to invalidate page cache if we are
1058 * dropping RDCACHE. Note that this will leave behind locked pages
1059 * that we'll then need to deal with elsewhere.
1060 *
1061 * Return non-zero if delayed release, or we experienced an error
1062 * such that the caller should requeue + retry later.
1063 *
1064 * called with i_lock, then drops it.
1065 * caller should hold snap_rwsem (read), s_mutex.
1066 */
1067static int __send_cap(struct ceph_mds_client *mdsc, struct ceph_cap *cap,
1068 int op, int used, int want, int retain, int flushing,
1069 unsigned *pflush_tid)
1070 __releases(cap->ci->vfs_inode->i_lock)
1071{
1072 struct ceph_inode_info *ci = cap->ci;
1073 struct inode *inode = &ci->vfs_inode;
1074 u64 cap_id = cap->cap_id;
Sage Weil68c28322010-02-09 13:41:47 -08001075 int held, revoking, dropping, keep;
Sage Weila8599bd2009-10-06 11:31:12 -07001076 u64 seq, issue_seq, mseq, time_warp_seq, follows;
1077 u64 size, max_size;
1078 struct timespec mtime, atime;
1079 int wake = 0;
1080 mode_t mode;
1081 uid_t uid;
1082 gid_t gid;
1083 struct ceph_mds_session *session;
1084 u64 xattr_version = 0;
1085 int delayed = 0;
1086 u64 flush_tid = 0;
1087 int i;
1088 int ret;
1089
Sage Weil68c28322010-02-09 13:41:47 -08001090 held = cap->issued | cap->implemented;
1091 revoking = cap->implemented & ~cap->issued;
1092 retain &= ~revoking;
1093 dropping = cap->issued & ~retain;
1094
Sage Weila8599bd2009-10-06 11:31:12 -07001095 dout("__send_cap %p cap %p session %p %s -> %s (revoking %s)\n",
1096 inode, cap, cap->session,
1097 ceph_cap_string(held), ceph_cap_string(held & retain),
1098 ceph_cap_string(revoking));
1099 BUG_ON((retain & CEPH_CAP_PIN) == 0);
1100
1101 session = cap->session;
1102
1103 /* don't release wanted unless we've waited a bit. */
1104 if ((ci->i_ceph_flags & CEPH_I_NODELAY) == 0 &&
1105 time_before(jiffies, ci->i_hold_caps_min)) {
1106 dout(" delaying issued %s -> %s, wanted %s -> %s on send\n",
1107 ceph_cap_string(cap->issued),
1108 ceph_cap_string(cap->issued & retain),
1109 ceph_cap_string(cap->mds_wanted),
1110 ceph_cap_string(want));
1111 want |= cap->mds_wanted;
1112 retain |= cap->issued;
1113 delayed = 1;
1114 }
1115 ci->i_ceph_flags &= ~(CEPH_I_NODELAY | CEPH_I_FLUSH);
1116
1117 cap->issued &= retain; /* drop bits we don't want */
1118 if (cap->implemented & ~cap->issued) {
1119 /*
1120 * Wake up any waiters on wanted -> needed transition.
1121 * This is due to the weird transition from buffered
1122 * to sync IO... we need to flush dirty pages _before_
1123 * allowing sync writes to avoid reordering.
1124 */
1125 wake = 1;
1126 }
1127 cap->implemented &= cap->issued | used;
1128 cap->mds_wanted = want;
1129
1130 if (flushing) {
1131 /*
1132 * assign a tid for flush operations so we can avoid
1133 * flush1 -> dirty1 -> flush2 -> flushack1 -> mark
1134 * clean type races. track latest tid for every bit
1135 * so we can handle flush AxFw, flush Fw, and have the
1136 * first ack clean Ax.
1137 */
1138 flush_tid = ++ci->i_cap_flush_last_tid;
1139 if (pflush_tid)
1140 *pflush_tid = flush_tid;
1141 dout(" cap_flush_tid %d\n", (int)flush_tid);
1142 for (i = 0; i < CEPH_CAP_BITS; i++)
1143 if (flushing & (1 << i))
1144 ci->i_cap_flush_tid[i] = flush_tid;
1145 }
1146
1147 keep = cap->implemented;
1148 seq = cap->seq;
1149 issue_seq = cap->issue_seq;
1150 mseq = cap->mseq;
1151 size = inode->i_size;
1152 ci->i_reported_size = size;
1153 max_size = ci->i_wanted_max_size;
1154 ci->i_requested_max_size = max_size;
1155 mtime = inode->i_mtime;
1156 atime = inode->i_atime;
1157 time_warp_seq = ci->i_time_warp_seq;
1158 follows = ci->i_snap_realm->cached_context->seq;
1159 uid = inode->i_uid;
1160 gid = inode->i_gid;
1161 mode = inode->i_mode;
1162
1163 if (dropping & CEPH_CAP_XATTR_EXCL) {
1164 __ceph_build_xattrs_blob(ci);
1165 xattr_version = ci->i_xattrs.version + 1;
1166 }
1167
1168 spin_unlock(&inode->i_lock);
1169
Sage Weila8599bd2009-10-06 11:31:12 -07001170 ret = send_cap_msg(session, ceph_vino(inode).ino, cap_id,
1171 op, keep, want, flushing, seq, flush_tid, issue_seq, mseq,
1172 size, max_size, &mtime, &atime, time_warp_seq,
1173 uid, gid, mode,
1174 xattr_version,
1175 (flushing & CEPH_CAP_XATTR_EXCL) ? ci->i_xattrs.blob : NULL,
1176 follows);
1177 if (ret < 0) {
1178 dout("error sending cap msg, must requeue %p\n", inode);
1179 delayed = 1;
1180 }
1181
1182 if (wake)
Yehuda Sadeh03066f22010-07-27 13:11:08 -07001183 wake_up_all(&ci->i_cap_wq);
Sage Weila8599bd2009-10-06 11:31:12 -07001184
1185 return delayed;
1186}
1187
1188/*
1189 * When a snapshot is taken, clients accumulate dirty metadata on
1190 * inodes with capabilities in ceph_cap_snaps to describe the file
1191 * state at the time the snapshot was taken. This must be flushed
1192 * asynchronously back to the MDS once sync writes complete and dirty
1193 * data is written out.
1194 *
1195 * Called under i_lock. Takes s_mutex as needed.
1196 */
1197void __ceph_flush_snaps(struct ceph_inode_info *ci,
1198 struct ceph_mds_session **psession)
Yehuda Sadehcd84db62010-06-11 16:58:48 -07001199 __releases(ci->vfs_inode->i_lock)
1200 __acquires(ci->vfs_inode->i_lock)
Sage Weila8599bd2009-10-06 11:31:12 -07001201{
1202 struct inode *inode = &ci->vfs_inode;
1203 int mds;
1204 struct ceph_cap_snap *capsnap;
1205 u32 mseq;
1206 struct ceph_mds_client *mdsc = &ceph_inode_to_client(inode)->mdsc;
1207 struct ceph_mds_session *session = NULL; /* if session != NULL, we hold
1208 session->s_mutex */
1209 u64 next_follows = 0; /* keep track of how far we've gotten through the
1210 i_cap_snaps list, and skip these entries next time
1211 around to avoid an infinite loop */
1212
1213 if (psession)
1214 session = *psession;
1215
1216 dout("__flush_snaps %p\n", inode);
1217retry:
1218 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
1219 /* avoid an infiniute loop after retry */
1220 if (capsnap->follows < next_follows)
1221 continue;
1222 /*
1223 * we need to wait for sync writes to complete and for dirty
1224 * pages to be written out.
1225 */
1226 if (capsnap->dirty_pages || capsnap->writing)
1227 continue;
1228
Sage Weil819ccbf2010-04-01 09:33:46 -07001229 /*
1230 * if cap writeback already occurred, we should have dropped
1231 * the capsnap in ceph_put_wrbuffer_cap_refs.
1232 */
1233 BUG_ON(capsnap->dirty == 0);
1234
Sage Weila8599bd2009-10-06 11:31:12 -07001235 /* pick mds, take s_mutex */
Sage Weilca81f3f2010-06-10 14:21:36 -07001236 if (ci->i_auth_cap == NULL) {
1237 dout("no auth cap (migrating?), doing nothing\n");
1238 goto out;
1239 }
1240 mds = ci->i_auth_cap->session->s_mds;
1241 mseq = ci->i_auth_cap->mseq;
1242
Sage Weila8599bd2009-10-06 11:31:12 -07001243 if (session && session->s_mds != mds) {
1244 dout("oops, wrong session %p mutex\n", session);
1245 mutex_unlock(&session->s_mutex);
1246 ceph_put_mds_session(session);
1247 session = NULL;
1248 }
1249 if (!session) {
1250 spin_unlock(&inode->i_lock);
1251 mutex_lock(&mdsc->mutex);
1252 session = __ceph_lookup_mds_session(mdsc, mds);
1253 mutex_unlock(&mdsc->mutex);
1254 if (session) {
1255 dout("inverting session/ino locks on %p\n",
1256 session);
1257 mutex_lock(&session->s_mutex);
1258 }
1259 /*
1260 * if session == NULL, we raced against a cap
Sage Weilca81f3f2010-06-10 14:21:36 -07001261 * deletion or migration. retry, and we'll
1262 * get a better @mds value next time.
Sage Weila8599bd2009-10-06 11:31:12 -07001263 */
1264 spin_lock(&inode->i_lock);
1265 goto retry;
1266 }
1267
1268 capsnap->flush_tid = ++ci->i_cap_flush_last_tid;
1269 atomic_inc(&capsnap->nref);
1270 if (!list_empty(&capsnap->flushing_item))
1271 list_del_init(&capsnap->flushing_item);
1272 list_add_tail(&capsnap->flushing_item,
1273 &session->s_cap_snaps_flushing);
1274 spin_unlock(&inode->i_lock);
1275
1276 dout("flush_snaps %p cap_snap %p follows %lld size %llu\n",
1277 inode, capsnap, next_follows, capsnap->size);
1278 send_cap_msg(session, ceph_vino(inode).ino, 0,
1279 CEPH_CAP_OP_FLUSHSNAP, capsnap->issued, 0,
1280 capsnap->dirty, 0, capsnap->flush_tid, 0, mseq,
1281 capsnap->size, 0,
1282 &capsnap->mtime, &capsnap->atime,
1283 capsnap->time_warp_seq,
1284 capsnap->uid, capsnap->gid, capsnap->mode,
1285 0, NULL,
1286 capsnap->follows);
1287
1288 next_follows = capsnap->follows + 1;
1289 ceph_put_cap_snap(capsnap);
1290
1291 spin_lock(&inode->i_lock);
1292 goto retry;
1293 }
1294
1295 /* we flushed them all; remove this inode from the queue */
1296 spin_lock(&mdsc->snap_flush_lock);
1297 list_del_init(&ci->i_snap_flush_item);
1298 spin_unlock(&mdsc->snap_flush_lock);
1299
Sage Weilca81f3f2010-06-10 14:21:36 -07001300out:
Sage Weila8599bd2009-10-06 11:31:12 -07001301 if (psession)
1302 *psession = session;
1303 else if (session) {
1304 mutex_unlock(&session->s_mutex);
1305 ceph_put_mds_session(session);
1306 }
1307}
1308
1309static void ceph_flush_snaps(struct ceph_inode_info *ci)
1310{
1311 struct inode *inode = &ci->vfs_inode;
1312
1313 spin_lock(&inode->i_lock);
1314 __ceph_flush_snaps(ci, NULL);
1315 spin_unlock(&inode->i_lock);
1316}
1317
1318/*
Sage Weil76e3b392009-10-15 18:13:53 -07001319 * Mark caps dirty. If inode is newly dirty, add to the global dirty
1320 * list.
1321 */
1322void __ceph_mark_dirty_caps(struct ceph_inode_info *ci, int mask)
1323{
Cheng Renquan640ef792010-03-26 17:40:33 +08001324 struct ceph_mds_client *mdsc =
1325 &ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc;
Sage Weil76e3b392009-10-15 18:13:53 -07001326 struct inode *inode = &ci->vfs_inode;
1327 int was = ci->i_dirty_caps;
1328 int dirty = 0;
1329
1330 dout("__mark_dirty_caps %p %s dirty %s -> %s\n", &ci->vfs_inode,
1331 ceph_cap_string(mask), ceph_cap_string(was),
1332 ceph_cap_string(was | mask));
1333 ci->i_dirty_caps |= mask;
1334 if (was == 0) {
1335 dout(" inode %p now dirty\n", &ci->vfs_inode);
1336 BUG_ON(!list_empty(&ci->i_dirty_item));
1337 spin_lock(&mdsc->cap_dirty_lock);
1338 list_add(&ci->i_dirty_item, &mdsc->cap_dirty);
1339 spin_unlock(&mdsc->cap_dirty_lock);
1340 if (ci->i_flushing_caps == 0) {
1341 igrab(inode);
1342 dirty |= I_DIRTY_SYNC;
1343 }
1344 }
1345 BUG_ON(list_empty(&ci->i_dirty_item));
1346 if (((was | ci->i_flushing_caps) & CEPH_CAP_FILE_BUFFER) &&
1347 (mask & CEPH_CAP_FILE_BUFFER))
1348 dirty |= I_DIRTY_DATASYNC;
1349 if (dirty)
1350 __mark_inode_dirty(inode, dirty);
1351 __cap_delay_requeue(mdsc, ci);
1352}
1353
1354/*
Sage Weila8599bd2009-10-06 11:31:12 -07001355 * Add dirty inode to the flushing list. Assigned a seq number so we
1356 * can wait for caps to flush without starving.
Sage Weilcdc35f92009-10-14 14:24:19 -07001357 *
1358 * Called under i_lock.
Sage Weila8599bd2009-10-06 11:31:12 -07001359 */
Sage Weilcdc35f92009-10-14 14:24:19 -07001360static int __mark_caps_flushing(struct inode *inode,
Sage Weila8599bd2009-10-06 11:31:12 -07001361 struct ceph_mds_session *session)
1362{
Cheng Renquan640ef792010-03-26 17:40:33 +08001363 struct ceph_mds_client *mdsc = &ceph_sb_to_client(inode->i_sb)->mdsc;
Sage Weila8599bd2009-10-06 11:31:12 -07001364 struct ceph_inode_info *ci = ceph_inode(inode);
Sage Weilcdc35f92009-10-14 14:24:19 -07001365 int flushing;
Sage Weil50b885b2009-12-01 14:12:07 -08001366
Sage Weilcdc35f92009-10-14 14:24:19 -07001367 BUG_ON(ci->i_dirty_caps == 0);
Sage Weila8599bd2009-10-06 11:31:12 -07001368 BUG_ON(list_empty(&ci->i_dirty_item));
Sage Weilcdc35f92009-10-14 14:24:19 -07001369
1370 flushing = ci->i_dirty_caps;
1371 dout("__mark_caps_flushing flushing %s, flushing_caps %s -> %s\n",
1372 ceph_cap_string(flushing),
1373 ceph_cap_string(ci->i_flushing_caps),
1374 ceph_cap_string(ci->i_flushing_caps | flushing));
1375 ci->i_flushing_caps |= flushing;
1376 ci->i_dirty_caps = 0;
Sage Weilafcdaea2009-10-14 14:27:38 -07001377 dout(" inode %p now !dirty\n", inode);
Sage Weilcdc35f92009-10-14 14:24:19 -07001378
Sage Weila8599bd2009-10-06 11:31:12 -07001379 spin_lock(&mdsc->cap_dirty_lock);
Sage Weilafcdaea2009-10-14 14:27:38 -07001380 list_del_init(&ci->i_dirty_item);
1381
1382 ci->i_cap_flush_seq = ++mdsc->cap_flush_seq;
Sage Weila8599bd2009-10-06 11:31:12 -07001383 if (list_empty(&ci->i_flushing_item)) {
1384 list_add_tail(&ci->i_flushing_item, &session->s_cap_flushing);
1385 mdsc->num_cap_flushing++;
Sage Weilafcdaea2009-10-14 14:27:38 -07001386 dout(" inode %p now flushing seq %lld\n", inode,
1387 ci->i_cap_flush_seq);
1388 } else {
1389 list_move_tail(&ci->i_flushing_item, &session->s_cap_flushing);
1390 dout(" inode %p now flushing (more) seq %lld\n", inode,
Sage Weila8599bd2009-10-06 11:31:12 -07001391 ci->i_cap_flush_seq);
1392 }
1393 spin_unlock(&mdsc->cap_dirty_lock);
Sage Weilcdc35f92009-10-14 14:24:19 -07001394
1395 return flushing;
Sage Weila8599bd2009-10-06 11:31:12 -07001396}
1397
1398/*
Sage Weil5ecad6f2010-02-17 10:43:37 -08001399 * try to invalidate mapping pages without blocking.
1400 */
1401static int mapping_is_empty(struct address_space *mapping)
1402{
1403 struct page *page = find_get_page(mapping, 0);
1404
1405 if (!page)
1406 return 1;
1407
1408 put_page(page);
1409 return 0;
1410}
1411
1412static int try_nonblocking_invalidate(struct inode *inode)
1413{
1414 struct ceph_inode_info *ci = ceph_inode(inode);
1415 u32 invalidating_gen = ci->i_rdcache_gen;
1416
1417 spin_unlock(&inode->i_lock);
1418 invalidate_mapping_pages(&inode->i_data, 0, -1);
1419 spin_lock(&inode->i_lock);
1420
1421 if (mapping_is_empty(&inode->i_data) &&
1422 invalidating_gen == ci->i_rdcache_gen) {
1423 /* success. */
1424 dout("try_nonblocking_invalidate %p success\n", inode);
1425 ci->i_rdcache_gen = 0;
1426 ci->i_rdcache_revoking = 0;
1427 return 0;
1428 }
1429 dout("try_nonblocking_invalidate %p failed\n", inode);
1430 return -1;
1431}
1432
1433/*
Sage Weila8599bd2009-10-06 11:31:12 -07001434 * Swiss army knife function to examine currently used and wanted
1435 * versus held caps. Release, flush, ack revoked caps to mds as
1436 * appropriate.
1437 *
1438 * CHECK_CAPS_NODELAY - caller is delayed work and we should not delay
1439 * cap release further.
1440 * CHECK_CAPS_AUTHONLY - we should only check the auth cap
1441 * CHECK_CAPS_FLUSH - we should flush any dirty caps immediately, without
1442 * further delay.
1443 */
1444void ceph_check_caps(struct ceph_inode_info *ci, int flags,
1445 struct ceph_mds_session *session)
1446{
1447 struct ceph_client *client = ceph_inode_to_client(&ci->vfs_inode);
1448 struct ceph_mds_client *mdsc = &client->mdsc;
1449 struct inode *inode = &ci->vfs_inode;
1450 struct ceph_cap *cap;
1451 int file_wanted, used;
1452 int took_snap_rwsem = 0; /* true if mdsc->snap_rwsem held */
Sage Weilcbd03632010-02-09 13:41:18 -08001453 int issued, implemented, want, retain, revoking, flushing = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07001454 int mds = -1; /* keep track of how far we've gone through i_caps list
1455 to avoid an infinite loop on retry */
1456 struct rb_node *p;
1457 int tried_invalidate = 0;
1458 int delayed = 0, sent = 0, force_requeue = 0, num;
Sage Weilcbd03632010-02-09 13:41:18 -08001459 int queue_invalidate = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07001460 int is_delayed = flags & CHECK_CAPS_NODELAY;
1461
1462 /* if we are unmounting, flush any unused caps immediately. */
1463 if (mdsc->stopping)
1464 is_delayed = 1;
1465
1466 spin_lock(&inode->i_lock);
1467
1468 if (ci->i_ceph_flags & CEPH_I_FLUSH)
1469 flags |= CHECK_CAPS_FLUSH;
1470
1471 /* flush snaps first time around only */
1472 if (!list_empty(&ci->i_cap_snaps))
1473 __ceph_flush_snaps(ci, &session);
1474 goto retry_locked;
1475retry:
1476 spin_lock(&inode->i_lock);
1477retry_locked:
1478 file_wanted = __ceph_caps_file_wanted(ci);
1479 used = __ceph_caps_used(ci);
1480 want = file_wanted | used;
Sage Weilcbd03632010-02-09 13:41:18 -08001481 issued = __ceph_caps_issued(ci, &implemented);
1482 revoking = implemented & ~issued;
Sage Weila8599bd2009-10-06 11:31:12 -07001483
1484 retain = want | CEPH_CAP_PIN;
1485 if (!mdsc->stopping && inode->i_nlink > 0) {
1486 if (want) {
1487 retain |= CEPH_CAP_ANY; /* be greedy */
1488 } else {
1489 retain |= CEPH_CAP_ANY_SHARED;
1490 /*
1491 * keep RD only if we didn't have the file open RW,
1492 * because then the mds would revoke it anyway to
1493 * journal max_size=0.
1494 */
1495 if (ci->i_max_size == 0)
1496 retain |= CEPH_CAP_ANY_RD;
1497 }
1498 }
1499
1500 dout("check_caps %p file_want %s used %s dirty %s flushing %s"
Sage Weilcbd03632010-02-09 13:41:18 -08001501 " issued %s revoking %s retain %s %s%s%s\n", inode,
Sage Weila8599bd2009-10-06 11:31:12 -07001502 ceph_cap_string(file_wanted),
1503 ceph_cap_string(used), ceph_cap_string(ci->i_dirty_caps),
1504 ceph_cap_string(ci->i_flushing_caps),
Sage Weilcbd03632010-02-09 13:41:18 -08001505 ceph_cap_string(issued), ceph_cap_string(revoking),
Sage Weila8599bd2009-10-06 11:31:12 -07001506 ceph_cap_string(retain),
1507 (flags & CHECK_CAPS_AUTHONLY) ? " AUTHONLY" : "",
1508 (flags & CHECK_CAPS_NODELAY) ? " NODELAY" : "",
1509 (flags & CHECK_CAPS_FLUSH) ? " FLUSH" : "");
1510
1511 /*
1512 * If we no longer need to hold onto old our caps, and we may
1513 * have cached pages, but don't want them, then try to invalidate.
1514 * If we fail, it's because pages are locked.... try again later.
1515 */
1516 if ((!is_delayed || mdsc->stopping) &&
1517 ci->i_wrbuffer_ref == 0 && /* no dirty pages... */
1518 ci->i_rdcache_gen && /* may have cached pages */
Sage Weilcbd03632010-02-09 13:41:18 -08001519 (file_wanted == 0 || /* no open files */
Sage Weil29625072010-05-27 10:40:43 -07001520 (revoking & (CEPH_CAP_FILE_CACHE|
1521 CEPH_CAP_FILE_LAZYIO))) && /* or revoking cache */
Sage Weila8599bd2009-10-06 11:31:12 -07001522 !tried_invalidate) {
Sage Weila8599bd2009-10-06 11:31:12 -07001523 dout("check_caps trying to invalidate on %p\n", inode);
Sage Weil5ecad6f2010-02-17 10:43:37 -08001524 if (try_nonblocking_invalidate(inode) < 0) {
Sage Weil29625072010-05-27 10:40:43 -07001525 if (revoking & (CEPH_CAP_FILE_CACHE|
1526 CEPH_CAP_FILE_LAZYIO)) {
Sage Weil5ecad6f2010-02-17 10:43:37 -08001527 dout("check_caps queuing invalidate\n");
1528 queue_invalidate = 1;
1529 ci->i_rdcache_revoking = ci->i_rdcache_gen;
1530 } else {
1531 dout("check_caps failed to invalidate pages\n");
1532 /* we failed to invalidate pages. check these
1533 caps again later. */
1534 force_requeue = 1;
1535 __cap_set_timeouts(mdsc, ci);
1536 }
Sage Weila8599bd2009-10-06 11:31:12 -07001537 }
1538 tried_invalidate = 1;
1539 goto retry_locked;
1540 }
1541
1542 num = 0;
1543 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
1544 cap = rb_entry(p, struct ceph_cap, ci_node);
1545 num++;
1546
1547 /* avoid looping forever */
1548 if (mds >= cap->mds ||
1549 ((flags & CHECK_CAPS_AUTHONLY) && cap != ci->i_auth_cap))
1550 continue;
1551
1552 /* NOTE: no side-effects allowed, until we take s_mutex */
1553
1554 revoking = cap->implemented & ~cap->issued;
1555 if (revoking)
Sage Weilcbd03632010-02-09 13:41:18 -08001556 dout(" mds%d revoking %s\n", cap->mds,
Sage Weila8599bd2009-10-06 11:31:12 -07001557 ceph_cap_string(revoking));
1558
1559 if (cap == ci->i_auth_cap &&
1560 (cap->issued & CEPH_CAP_FILE_WR)) {
1561 /* request larger max_size from MDS? */
1562 if (ci->i_wanted_max_size > ci->i_max_size &&
1563 ci->i_wanted_max_size > ci->i_requested_max_size) {
1564 dout("requesting new max_size\n");
1565 goto ack;
1566 }
1567
1568 /* approaching file_max? */
1569 if ((inode->i_size << 1) >= ci->i_max_size &&
1570 (ci->i_reported_size << 1) < ci->i_max_size) {
1571 dout("i_size approaching max_size\n");
1572 goto ack;
1573 }
1574 }
1575 /* flush anything dirty? */
1576 if (cap == ci->i_auth_cap && (flags & CHECK_CAPS_FLUSH) &&
1577 ci->i_dirty_caps) {
1578 dout("flushing dirty caps\n");
1579 goto ack;
1580 }
1581
1582 /* completed revocation? going down and there are no caps? */
1583 if (revoking && (revoking & used) == 0) {
1584 dout("completed revocation of %s\n",
1585 ceph_cap_string(cap->implemented & ~cap->issued));
1586 goto ack;
1587 }
1588
1589 /* want more caps from mds? */
1590 if (want & ~(cap->mds_wanted | cap->issued))
1591 goto ack;
1592
1593 /* things we might delay */
1594 if ((cap->issued & ~retain) == 0 &&
1595 cap->mds_wanted == want)
1596 continue; /* nope, all good */
1597
1598 if (is_delayed)
1599 goto ack;
1600
1601 /* delay? */
1602 if ((ci->i_ceph_flags & CEPH_I_NODELAY) == 0 &&
1603 time_before(jiffies, ci->i_hold_caps_max)) {
1604 dout(" delaying issued %s -> %s, wanted %s -> %s\n",
1605 ceph_cap_string(cap->issued),
1606 ceph_cap_string(cap->issued & retain),
1607 ceph_cap_string(cap->mds_wanted),
1608 ceph_cap_string(want));
1609 delayed++;
1610 continue;
1611 }
1612
1613ack:
Sage Weile9964c12010-03-01 15:16:56 -08001614 if (ci->i_ceph_flags & CEPH_I_NOFLUSH) {
1615 dout(" skipping %p I_NOFLUSH set\n", inode);
1616 continue;
1617 }
1618
Sage Weila8599bd2009-10-06 11:31:12 -07001619 if (session && session != cap->session) {
1620 dout("oops, wrong session %p mutex\n", session);
1621 mutex_unlock(&session->s_mutex);
1622 session = NULL;
1623 }
1624 if (!session) {
1625 session = cap->session;
1626 if (mutex_trylock(&session->s_mutex) == 0) {
1627 dout("inverting session/ino locks on %p\n",
1628 session);
1629 spin_unlock(&inode->i_lock);
1630 if (took_snap_rwsem) {
1631 up_read(&mdsc->snap_rwsem);
1632 took_snap_rwsem = 0;
1633 }
1634 mutex_lock(&session->s_mutex);
1635 goto retry;
1636 }
1637 }
1638 /* take snap_rwsem after session mutex */
1639 if (!took_snap_rwsem) {
1640 if (down_read_trylock(&mdsc->snap_rwsem) == 0) {
1641 dout("inverting snap/in locks on %p\n",
1642 inode);
1643 spin_unlock(&inode->i_lock);
1644 down_read(&mdsc->snap_rwsem);
1645 took_snap_rwsem = 1;
1646 goto retry;
1647 }
1648 took_snap_rwsem = 1;
1649 }
1650
Sage Weilcdc35f92009-10-14 14:24:19 -07001651 if (cap == ci->i_auth_cap && ci->i_dirty_caps)
1652 flushing = __mark_caps_flushing(inode, session);
Sage Weila8599bd2009-10-06 11:31:12 -07001653
1654 mds = cap->mds; /* remember mds, so we don't repeat */
1655 sent++;
1656
1657 /* __send_cap drops i_lock */
1658 delayed += __send_cap(mdsc, cap, CEPH_CAP_OP_UPDATE, used, want,
1659 retain, flushing, NULL);
1660 goto retry; /* retake i_lock and restart our cap scan. */
1661 }
1662
1663 /*
1664 * Reschedule delayed caps release if we delayed anything,
1665 * otherwise cancel.
1666 */
1667 if (delayed && is_delayed)
1668 force_requeue = 1; /* __send_cap delayed release; requeue */
1669 if (!delayed && !is_delayed)
1670 __cap_delay_cancel(mdsc, ci);
1671 else if (!is_delayed || force_requeue)
1672 __cap_delay_requeue(mdsc, ci);
1673
1674 spin_unlock(&inode->i_lock);
1675
Sage Weilcbd03632010-02-09 13:41:18 -08001676 if (queue_invalidate)
Sage Weil3c6f6b72010-02-09 15:24:44 -08001677 ceph_queue_invalidate(inode);
Sage Weilcbd03632010-02-09 13:41:18 -08001678
Sage Weilcdc2ce02010-03-16 13:39:28 -07001679 if (session)
Sage Weila8599bd2009-10-06 11:31:12 -07001680 mutex_unlock(&session->s_mutex);
1681 if (took_snap_rwsem)
1682 up_read(&mdsc->snap_rwsem);
1683}
1684
1685/*
Sage Weila8599bd2009-10-06 11:31:12 -07001686 * Try to flush dirty caps back to the auth mds.
1687 */
1688static int try_flush_caps(struct inode *inode, struct ceph_mds_session *session,
1689 unsigned *flush_tid)
1690{
Cheng Renquan640ef792010-03-26 17:40:33 +08001691 struct ceph_mds_client *mdsc = &ceph_sb_to_client(inode->i_sb)->mdsc;
Sage Weila8599bd2009-10-06 11:31:12 -07001692 struct ceph_inode_info *ci = ceph_inode(inode);
1693 int unlock_session = session ? 0 : 1;
1694 int flushing = 0;
1695
1696retry:
1697 spin_lock(&inode->i_lock);
Sage Weile9964c12010-03-01 15:16:56 -08001698 if (ci->i_ceph_flags & CEPH_I_NOFLUSH) {
1699 dout("try_flush_caps skipping %p I_NOFLUSH set\n", inode);
1700 goto out;
1701 }
Sage Weila8599bd2009-10-06 11:31:12 -07001702 if (ci->i_dirty_caps && ci->i_auth_cap) {
1703 struct ceph_cap *cap = ci->i_auth_cap;
1704 int used = __ceph_caps_used(ci);
1705 int want = __ceph_caps_wanted(ci);
1706 int delayed;
1707
1708 if (!session) {
1709 spin_unlock(&inode->i_lock);
1710 session = cap->session;
1711 mutex_lock(&session->s_mutex);
1712 goto retry;
1713 }
1714 BUG_ON(session != cap->session);
1715 if (cap->session->s_state < CEPH_MDS_SESSION_OPEN)
1716 goto out;
1717
Sage Weilcdc35f92009-10-14 14:24:19 -07001718 flushing = __mark_caps_flushing(inode, session);
Sage Weila8599bd2009-10-06 11:31:12 -07001719
1720 /* __send_cap drops i_lock */
1721 delayed = __send_cap(mdsc, cap, CEPH_CAP_OP_FLUSH, used, want,
1722 cap->issued | cap->implemented, flushing,
1723 flush_tid);
1724 if (!delayed)
1725 goto out_unlocked;
1726
1727 spin_lock(&inode->i_lock);
1728 __cap_delay_requeue(mdsc, ci);
1729 }
1730out:
1731 spin_unlock(&inode->i_lock);
1732out_unlocked:
1733 if (session && unlock_session)
1734 mutex_unlock(&session->s_mutex);
1735 return flushing;
1736}
1737
1738/*
1739 * Return true if we've flushed caps through the given flush_tid.
1740 */
1741static int caps_are_flushed(struct inode *inode, unsigned tid)
1742{
1743 struct ceph_inode_info *ci = ceph_inode(inode);
Dan Carpentera5ee7512010-05-07 10:27:14 +02001744 int i, ret = 1;
Sage Weila8599bd2009-10-06 11:31:12 -07001745
1746 spin_lock(&inode->i_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001747 for (i = 0; i < CEPH_CAP_BITS; i++)
1748 if ((ci->i_flushing_caps & (1 << i)) &&
1749 ci->i_cap_flush_tid[i] <= tid) {
1750 /* still flushing this bit */
1751 ret = 0;
1752 break;
1753 }
1754 spin_unlock(&inode->i_lock);
1755 return ret;
1756}
1757
1758/*
1759 * Wait on any unsafe replies for the given inode. First wait on the
1760 * newest request, and make that the upper bound. Then, if there are
1761 * more requests, keep waiting on the oldest as long as it is still older
1762 * than the original request.
1763 */
1764static void sync_write_wait(struct inode *inode)
1765{
1766 struct ceph_inode_info *ci = ceph_inode(inode);
1767 struct list_head *head = &ci->i_unsafe_writes;
1768 struct ceph_osd_request *req;
1769 u64 last_tid;
1770
1771 spin_lock(&ci->i_unsafe_lock);
1772 if (list_empty(head))
1773 goto out;
1774
1775 /* set upper bound as _last_ entry in chain */
1776 req = list_entry(head->prev, struct ceph_osd_request,
1777 r_unsafe_item);
1778 last_tid = req->r_tid;
1779
1780 do {
1781 ceph_osdc_get_request(req);
1782 spin_unlock(&ci->i_unsafe_lock);
1783 dout("sync_write_wait on tid %llu (until %llu)\n",
1784 req->r_tid, last_tid);
1785 wait_for_completion(&req->r_safe_completion);
1786 spin_lock(&ci->i_unsafe_lock);
1787 ceph_osdc_put_request(req);
1788
1789 /*
1790 * from here on look at first entry in chain, since we
1791 * only want to wait for anything older than last_tid
1792 */
1793 if (list_empty(head))
1794 break;
1795 req = list_entry(head->next, struct ceph_osd_request,
1796 r_unsafe_item);
1797 } while (req->r_tid < last_tid);
1798out:
1799 spin_unlock(&ci->i_unsafe_lock);
1800}
1801
Christoph Hellwig7ea80852010-05-26 17:53:25 +02001802int ceph_fsync(struct file *file, int datasync)
Sage Weila8599bd2009-10-06 11:31:12 -07001803{
Christoph Hellwig7ea80852010-05-26 17:53:25 +02001804 struct inode *inode = file->f_mapping->host;
Sage Weila8599bd2009-10-06 11:31:12 -07001805 struct ceph_inode_info *ci = ceph_inode(inode);
1806 unsigned flush_tid;
1807 int ret;
1808 int dirty;
1809
1810 dout("fsync %p%s\n", inode, datasync ? " datasync" : "");
1811 sync_write_wait(inode);
1812
1813 ret = filemap_write_and_wait(inode->i_mapping);
1814 if (ret < 0)
1815 return ret;
1816
1817 dirty = try_flush_caps(inode, NULL, &flush_tid);
1818 dout("fsync dirty caps are %s\n", ceph_cap_string(dirty));
1819
1820 /*
1821 * only wait on non-file metadata writeback (the mds
1822 * can recover size and mtime, so we don't need to
1823 * wait for that)
1824 */
1825 if (!datasync && (dirty & ~CEPH_CAP_ANY_FILE_WR)) {
1826 dout("fsync waiting for flush_tid %u\n", flush_tid);
1827 ret = wait_event_interruptible(ci->i_cap_wq,
1828 caps_are_flushed(inode, flush_tid));
1829 }
1830
1831 dout("fsync %p%s done\n", inode, datasync ? " datasync" : "");
1832 return ret;
1833}
1834
1835/*
1836 * Flush any dirty caps back to the mds. If we aren't asked to wait,
1837 * queue inode for flush but don't do so immediately, because we can
1838 * get by with fewer MDS messages if we wait for data writeback to
1839 * complete first.
1840 */
Stephen Rothwellf1a3d572010-01-18 11:53:08 +11001841int ceph_write_inode(struct inode *inode, struct writeback_control *wbc)
Sage Weila8599bd2009-10-06 11:31:12 -07001842{
1843 struct ceph_inode_info *ci = ceph_inode(inode);
1844 unsigned flush_tid;
1845 int err = 0;
1846 int dirty;
Stephen Rothwellf1a3d572010-01-18 11:53:08 +11001847 int wait = wbc->sync_mode == WB_SYNC_ALL;
Sage Weila8599bd2009-10-06 11:31:12 -07001848
1849 dout("write_inode %p wait=%d\n", inode, wait);
1850 if (wait) {
1851 dirty = try_flush_caps(inode, NULL, &flush_tid);
1852 if (dirty)
1853 err = wait_event_interruptible(ci->i_cap_wq,
1854 caps_are_flushed(inode, flush_tid));
1855 } else {
Cheng Renquan640ef792010-03-26 17:40:33 +08001856 struct ceph_mds_client *mdsc =
1857 &ceph_sb_to_client(inode->i_sb)->mdsc;
Sage Weila8599bd2009-10-06 11:31:12 -07001858
1859 spin_lock(&inode->i_lock);
1860 if (__ceph_caps_dirty(ci))
1861 __cap_delay_requeue_front(mdsc, ci);
1862 spin_unlock(&inode->i_lock);
1863 }
1864 return err;
1865}
1866
1867/*
1868 * After a recovering MDS goes active, we need to resend any caps
1869 * we were flushing.
1870 *
1871 * Caller holds session->s_mutex.
1872 */
1873static void kick_flushing_capsnaps(struct ceph_mds_client *mdsc,
1874 struct ceph_mds_session *session)
1875{
1876 struct ceph_cap_snap *capsnap;
1877
1878 dout("kick_flushing_capsnaps mds%d\n", session->s_mds);
1879 list_for_each_entry(capsnap, &session->s_cap_snaps_flushing,
1880 flushing_item) {
1881 struct ceph_inode_info *ci = capsnap->ci;
1882 struct inode *inode = &ci->vfs_inode;
1883 struct ceph_cap *cap;
1884
1885 spin_lock(&inode->i_lock);
1886 cap = ci->i_auth_cap;
1887 if (cap && cap->session == session) {
1888 dout("kick_flushing_caps %p cap %p capsnap %p\n", inode,
1889 cap, capsnap);
1890 __ceph_flush_snaps(ci, &session);
1891 } else {
1892 pr_err("%p auth cap %p not mds%d ???\n", inode,
1893 cap, session->s_mds);
Sage Weila8599bd2009-10-06 11:31:12 -07001894 }
Sage Weil0b0c06d2010-04-20 10:27:13 -07001895 spin_unlock(&inode->i_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001896 }
1897}
1898
1899void ceph_kick_flushing_caps(struct ceph_mds_client *mdsc,
1900 struct ceph_mds_session *session)
1901{
1902 struct ceph_inode_info *ci;
1903
1904 kick_flushing_capsnaps(mdsc, session);
1905
1906 dout("kick_flushing_caps mds%d\n", session->s_mds);
1907 list_for_each_entry(ci, &session->s_cap_flushing, i_flushing_item) {
1908 struct inode *inode = &ci->vfs_inode;
1909 struct ceph_cap *cap;
1910 int delayed = 0;
1911
1912 spin_lock(&inode->i_lock);
1913 cap = ci->i_auth_cap;
1914 if (cap && cap->session == session) {
1915 dout("kick_flushing_caps %p cap %p %s\n", inode,
1916 cap, ceph_cap_string(ci->i_flushing_caps));
1917 delayed = __send_cap(mdsc, cap, CEPH_CAP_OP_FLUSH,
1918 __ceph_caps_used(ci),
1919 __ceph_caps_wanted(ci),
1920 cap->issued | cap->implemented,
1921 ci->i_flushing_caps, NULL);
1922 if (delayed) {
1923 spin_lock(&inode->i_lock);
1924 __cap_delay_requeue(mdsc, ci);
1925 spin_unlock(&inode->i_lock);
1926 }
1927 } else {
1928 pr_err("%p auth cap %p not mds%d ???\n", inode,
1929 cap, session->s_mds);
1930 spin_unlock(&inode->i_lock);
1931 }
1932 }
1933}
1934
1935
1936/*
1937 * Take references to capabilities we hold, so that we don't release
1938 * them to the MDS prematurely.
1939 *
1940 * Protected by i_lock.
1941 */
1942static void __take_cap_refs(struct ceph_inode_info *ci, int got)
1943{
1944 if (got & CEPH_CAP_PIN)
1945 ci->i_pin_ref++;
1946 if (got & CEPH_CAP_FILE_RD)
1947 ci->i_rd_ref++;
1948 if (got & CEPH_CAP_FILE_CACHE)
1949 ci->i_rdcache_ref++;
1950 if (got & CEPH_CAP_FILE_WR)
1951 ci->i_wr_ref++;
1952 if (got & CEPH_CAP_FILE_BUFFER) {
1953 if (ci->i_wrbuffer_ref == 0)
1954 igrab(&ci->vfs_inode);
1955 ci->i_wrbuffer_ref++;
1956 dout("__take_cap_refs %p wrbuffer %d -> %d (?)\n",
1957 &ci->vfs_inode, ci->i_wrbuffer_ref-1, ci->i_wrbuffer_ref);
1958 }
1959}
1960
1961/*
1962 * Try to grab cap references. Specify those refs we @want, and the
1963 * minimal set we @need. Also include the larger offset we are writing
1964 * to (when applicable), and check against max_size here as well.
1965 * Note that caller is responsible for ensuring max_size increases are
1966 * requested from the MDS.
1967 */
1968static int try_get_cap_refs(struct ceph_inode_info *ci, int need, int want,
1969 int *got, loff_t endoff, int *check_max, int *err)
1970{
1971 struct inode *inode = &ci->vfs_inode;
1972 int ret = 0;
1973 int have, implemented;
Sage Weil195d3ce2010-03-01 09:57:54 -08001974 int file_wanted;
Sage Weila8599bd2009-10-06 11:31:12 -07001975
1976 dout("get_cap_refs %p need %s want %s\n", inode,
1977 ceph_cap_string(need), ceph_cap_string(want));
1978 spin_lock(&inode->i_lock);
1979
Sage Weil195d3ce2010-03-01 09:57:54 -08001980 /* make sure file is actually open */
1981 file_wanted = __ceph_caps_file_wanted(ci);
1982 if ((file_wanted & need) == 0) {
1983 dout("try_get_cap_refs need %s file_wanted %s, EBADF\n",
1984 ceph_cap_string(need), ceph_cap_string(file_wanted));
Sage Weila8599bd2009-10-06 11:31:12 -07001985 *err = -EBADF;
1986 ret = 1;
1987 goto out;
1988 }
1989
1990 if (need & CEPH_CAP_FILE_WR) {
1991 if (endoff >= 0 && endoff > (loff_t)ci->i_max_size) {
1992 dout("get_cap_refs %p endoff %llu > maxsize %llu\n",
1993 inode, endoff, ci->i_max_size);
1994 if (endoff > ci->i_wanted_max_size) {
1995 *check_max = 1;
1996 ret = 1;
1997 }
1998 goto out;
1999 }
2000 /*
2001 * If a sync write is in progress, we must wait, so that we
2002 * can get a final snapshot value for size+mtime.
2003 */
2004 if (__ceph_have_pending_cap_snap(ci)) {
2005 dout("get_cap_refs %p cap_snap_pending\n", inode);
2006 goto out;
2007 }
2008 }
2009 have = __ceph_caps_issued(ci, &implemented);
2010
2011 /*
2012 * disallow writes while a truncate is pending
2013 */
2014 if (ci->i_truncate_pending)
2015 have &= ~CEPH_CAP_FILE_WR;
2016
2017 if ((have & need) == need) {
2018 /*
2019 * Look at (implemented & ~have & not) so that we keep waiting
2020 * on transition from wanted -> needed caps. This is needed
2021 * for WRBUFFER|WR -> WR to avoid a new WR sync write from
2022 * going before a prior buffered writeback happens.
2023 */
2024 int not = want & ~(have & need);
2025 int revoking = implemented & ~have;
2026 dout("get_cap_refs %p have %s but not %s (revoking %s)\n",
2027 inode, ceph_cap_string(have), ceph_cap_string(not),
2028 ceph_cap_string(revoking));
2029 if ((revoking & not) == 0) {
2030 *got = need | (have & want);
2031 __take_cap_refs(ci, *got);
2032 ret = 1;
2033 }
2034 } else {
2035 dout("get_cap_refs %p have %s needed %s\n", inode,
2036 ceph_cap_string(have), ceph_cap_string(need));
2037 }
2038out:
2039 spin_unlock(&inode->i_lock);
2040 dout("get_cap_refs %p ret %d got %s\n", inode,
2041 ret, ceph_cap_string(*got));
2042 return ret;
2043}
2044
2045/*
2046 * Check the offset we are writing up to against our current
2047 * max_size. If necessary, tell the MDS we want to write to
2048 * a larger offset.
2049 */
2050static void check_max_size(struct inode *inode, loff_t endoff)
2051{
2052 struct ceph_inode_info *ci = ceph_inode(inode);
2053 int check = 0;
2054
2055 /* do we need to explicitly request a larger max_size? */
2056 spin_lock(&inode->i_lock);
2057 if ((endoff >= ci->i_max_size ||
2058 endoff > (inode->i_size << 1)) &&
2059 endoff > ci->i_wanted_max_size) {
2060 dout("write %p at large endoff %llu, req max_size\n",
2061 inode, endoff);
2062 ci->i_wanted_max_size = endoff;
2063 check = 1;
2064 }
2065 spin_unlock(&inode->i_lock);
2066 if (check)
2067 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
2068}
2069
2070/*
2071 * Wait for caps, and take cap references. If we can't get a WR cap
2072 * due to a small max_size, make sure we check_max_size (and possibly
2073 * ask the mds) so we don't get hung up indefinitely.
2074 */
2075int ceph_get_caps(struct ceph_inode_info *ci, int need, int want, int *got,
2076 loff_t endoff)
2077{
2078 int check_max, ret, err;
2079
2080retry:
2081 if (endoff > 0)
2082 check_max_size(&ci->vfs_inode, endoff);
2083 check_max = 0;
2084 err = 0;
2085 ret = wait_event_interruptible(ci->i_cap_wq,
2086 try_get_cap_refs(ci, need, want,
2087 got, endoff,
2088 &check_max, &err));
2089 if (err)
2090 ret = err;
2091 if (check_max)
2092 goto retry;
2093 return ret;
2094}
2095
2096/*
2097 * Take cap refs. Caller must already know we hold at least one ref
2098 * on the caps in question or we don't know this is safe.
2099 */
2100void ceph_get_cap_refs(struct ceph_inode_info *ci, int caps)
2101{
2102 spin_lock(&ci->vfs_inode.i_lock);
2103 __take_cap_refs(ci, caps);
2104 spin_unlock(&ci->vfs_inode.i_lock);
2105}
2106
2107/*
2108 * Release cap refs.
2109 *
2110 * If we released the last ref on any given cap, call ceph_check_caps
2111 * to release (or schedule a release).
2112 *
2113 * If we are releasing a WR cap (from a sync write), finalize any affected
2114 * cap_snap, and wake up any waiters.
2115 */
2116void ceph_put_cap_refs(struct ceph_inode_info *ci, int had)
2117{
2118 struct inode *inode = &ci->vfs_inode;
2119 int last = 0, put = 0, flushsnaps = 0, wake = 0;
2120 struct ceph_cap_snap *capsnap;
2121
2122 spin_lock(&inode->i_lock);
2123 if (had & CEPH_CAP_PIN)
2124 --ci->i_pin_ref;
2125 if (had & CEPH_CAP_FILE_RD)
2126 if (--ci->i_rd_ref == 0)
2127 last++;
2128 if (had & CEPH_CAP_FILE_CACHE)
2129 if (--ci->i_rdcache_ref == 0)
2130 last++;
2131 if (had & CEPH_CAP_FILE_BUFFER) {
2132 if (--ci->i_wrbuffer_ref == 0) {
2133 last++;
2134 put++;
2135 }
2136 dout("put_cap_refs %p wrbuffer %d -> %d (?)\n",
2137 inode, ci->i_wrbuffer_ref+1, ci->i_wrbuffer_ref);
2138 }
2139 if (had & CEPH_CAP_FILE_WR)
2140 if (--ci->i_wr_ref == 0) {
2141 last++;
2142 if (!list_empty(&ci->i_cap_snaps)) {
2143 capsnap = list_first_entry(&ci->i_cap_snaps,
2144 struct ceph_cap_snap,
2145 ci_item);
2146 if (capsnap->writing) {
2147 capsnap->writing = 0;
2148 flushsnaps =
2149 __ceph_finish_cap_snap(ci,
2150 capsnap);
2151 wake = 1;
2152 }
2153 }
2154 }
2155 spin_unlock(&inode->i_lock);
2156
Sage Weil819ccbf2010-04-01 09:33:46 -07002157 dout("put_cap_refs %p had %s%s%s\n", inode, ceph_cap_string(had),
2158 last ? " last" : "", put ? " put" : "");
Sage Weila8599bd2009-10-06 11:31:12 -07002159
2160 if (last && !flushsnaps)
2161 ceph_check_caps(ci, 0, NULL);
2162 else if (flushsnaps)
2163 ceph_flush_snaps(ci);
2164 if (wake)
Yehuda Sadeh03066f22010-07-27 13:11:08 -07002165 wake_up_all(&ci->i_cap_wq);
Sage Weila8599bd2009-10-06 11:31:12 -07002166 if (put)
2167 iput(inode);
2168}
2169
2170/*
2171 * Release @nr WRBUFFER refs on dirty pages for the given @snapc snap
2172 * context. Adjust per-snap dirty page accounting as appropriate.
2173 * Once all dirty data for a cap_snap is flushed, flush snapped file
2174 * metadata back to the MDS. If we dropped the last ref, call
2175 * ceph_check_caps.
2176 */
2177void ceph_put_wrbuffer_cap_refs(struct ceph_inode_info *ci, int nr,
2178 struct ceph_snap_context *snapc)
2179{
2180 struct inode *inode = &ci->vfs_inode;
2181 int last = 0;
Sage Weil819ccbf2010-04-01 09:33:46 -07002182 int complete_capsnap = 0;
2183 int drop_capsnap = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07002184 int found = 0;
2185 struct ceph_cap_snap *capsnap = NULL;
2186
2187 spin_lock(&inode->i_lock);
2188 ci->i_wrbuffer_ref -= nr;
2189 last = !ci->i_wrbuffer_ref;
2190
2191 if (ci->i_head_snapc == snapc) {
2192 ci->i_wrbuffer_ref_head -= nr;
2193 if (!ci->i_wrbuffer_ref_head) {
2194 ceph_put_snap_context(ci->i_head_snapc);
2195 ci->i_head_snapc = NULL;
2196 }
2197 dout("put_wrbuffer_cap_refs on %p head %d/%d -> %d/%d %s\n",
2198 inode,
2199 ci->i_wrbuffer_ref+nr, ci->i_wrbuffer_ref_head+nr,
2200 ci->i_wrbuffer_ref, ci->i_wrbuffer_ref_head,
2201 last ? " LAST" : "");
2202 } else {
2203 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
2204 if (capsnap->context == snapc) {
2205 found = 1;
Sage Weila8599bd2009-10-06 11:31:12 -07002206 break;
2207 }
2208 }
2209 BUG_ON(!found);
Sage Weil819ccbf2010-04-01 09:33:46 -07002210 capsnap->dirty_pages -= nr;
2211 if (capsnap->dirty_pages == 0) {
2212 complete_capsnap = 1;
2213 if (capsnap->dirty == 0)
2214 /* cap writeback completed before we created
2215 * the cap_snap; no FLUSHSNAP is needed */
2216 drop_capsnap = 1;
2217 }
Sage Weila8599bd2009-10-06 11:31:12 -07002218 dout("put_wrbuffer_cap_refs on %p cap_snap %p "
Sage Weil819ccbf2010-04-01 09:33:46 -07002219 " snap %lld %d/%d -> %d/%d %s%s%s\n",
Sage Weila8599bd2009-10-06 11:31:12 -07002220 inode, capsnap, capsnap->context->seq,
2221 ci->i_wrbuffer_ref+nr, capsnap->dirty_pages + nr,
2222 ci->i_wrbuffer_ref, capsnap->dirty_pages,
2223 last ? " (wrbuffer last)" : "",
Sage Weil819ccbf2010-04-01 09:33:46 -07002224 complete_capsnap ? " (complete capsnap)" : "",
2225 drop_capsnap ? " (drop capsnap)" : "");
2226 if (drop_capsnap) {
2227 ceph_put_snap_context(capsnap->context);
2228 list_del(&capsnap->ci_item);
2229 list_del(&capsnap->flushing_item);
2230 ceph_put_cap_snap(capsnap);
2231 }
Sage Weila8599bd2009-10-06 11:31:12 -07002232 }
2233
2234 spin_unlock(&inode->i_lock);
2235
2236 if (last) {
2237 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
2238 iput(inode);
Sage Weil819ccbf2010-04-01 09:33:46 -07002239 } else if (complete_capsnap) {
Sage Weila8599bd2009-10-06 11:31:12 -07002240 ceph_flush_snaps(ci);
Yehuda Sadeh03066f22010-07-27 13:11:08 -07002241 wake_up_all(&ci->i_cap_wq);
Sage Weila8599bd2009-10-06 11:31:12 -07002242 }
Sage Weil819ccbf2010-04-01 09:33:46 -07002243 if (drop_capsnap)
2244 iput(inode);
Sage Weila8599bd2009-10-06 11:31:12 -07002245}
2246
2247/*
2248 * Handle a cap GRANT message from the MDS. (Note that a GRANT may
2249 * actually be a revocation if it specifies a smaller cap set.)
2250 *
Sage Weil15637c82010-03-16 13:42:00 -07002251 * caller holds s_mutex and i_lock, we drop both.
2252 *
Sage Weila8599bd2009-10-06 11:31:12 -07002253 * return value:
2254 * 0 - ok
2255 * 1 - check_caps on auth cap only (writeback)
2256 * 2 - check_caps (ack revoke)
2257 */
Sage Weil15637c82010-03-16 13:42:00 -07002258static void handle_cap_grant(struct inode *inode, struct ceph_mds_caps *grant,
2259 struct ceph_mds_session *session,
2260 struct ceph_cap *cap,
2261 struct ceph_buffer *xattr_buf)
Yehuda Sadehcd84db62010-06-11 16:58:48 -07002262 __releases(inode->i_lock)
Sage Weila8599bd2009-10-06 11:31:12 -07002263{
2264 struct ceph_inode_info *ci = ceph_inode(inode);
2265 int mds = session->s_mds;
2266 int seq = le32_to_cpu(grant->seq);
2267 int newcaps = le32_to_cpu(grant->caps);
2268 int issued, implemented, used, wanted, dirty;
2269 u64 size = le64_to_cpu(grant->size);
2270 u64 max_size = le64_to_cpu(grant->max_size);
2271 struct timespec mtime, atime, ctime;
Sage Weil15637c82010-03-16 13:42:00 -07002272 int check_caps = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07002273 int wake = 0;
2274 int writeback = 0;
2275 int revoked_rdcache = 0;
Sage Weil3c6f6b72010-02-09 15:24:44 -08002276 int queue_invalidate = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07002277
2278 dout("handle_cap_grant inode %p cap %p mds%d seq %d %s\n",
2279 inode, cap, mds, seq, ceph_cap_string(newcaps));
2280 dout(" size %llu max_size %llu, i_size %llu\n", size, max_size,
2281 inode->i_size);
2282
2283 /*
2284 * If CACHE is being revoked, and we have no dirty buffers,
2285 * try to invalidate (once). (If there are dirty buffers, we
2286 * will invalidate _after_ writeback.)
2287 */
Sage Weil3b454c42010-06-10 13:20:33 -07002288 if (((cap->issued & ~newcaps) & CEPH_CAP_FILE_CACHE) &&
2289 (newcaps & CEPH_CAP_FILE_LAZYIO) == 0 &&
Yehuda Sadehbcd2cbd2010-02-19 00:12:21 +00002290 !ci->i_wrbuffer_ref) {
Sage Weil5ecad6f2010-02-17 10:43:37 -08002291 if (try_nonblocking_invalidate(inode) == 0) {
2292 revoked_rdcache = 1;
2293 } else {
Sage Weila8599bd2009-10-06 11:31:12 -07002294 /* there were locked pages.. invalidate later
2295 in a separate thread. */
2296 if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
Sage Weil3c6f6b72010-02-09 15:24:44 -08002297 queue_invalidate = 1;
Sage Weila8599bd2009-10-06 11:31:12 -07002298 ci->i_rdcache_revoking = ci->i_rdcache_gen;
2299 }
Sage Weila8599bd2009-10-06 11:31:12 -07002300 }
Sage Weila8599bd2009-10-06 11:31:12 -07002301 }
2302
2303 /* side effects now are allowed */
2304
2305 issued = __ceph_caps_issued(ci, &implemented);
2306 issued |= implemented | __ceph_caps_dirty(ci);
2307
Sage Weil685f9a5d2009-11-09 12:05:48 -08002308 cap->cap_gen = session->s_cap_gen;
Sage Weila8599bd2009-10-06 11:31:12 -07002309
2310 __check_cap_issue(ci, cap, newcaps);
2311
2312 if ((issued & CEPH_CAP_AUTH_EXCL) == 0) {
2313 inode->i_mode = le32_to_cpu(grant->mode);
2314 inode->i_uid = le32_to_cpu(grant->uid);
2315 inode->i_gid = le32_to_cpu(grant->gid);
2316 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
2317 inode->i_uid, inode->i_gid);
2318 }
2319
2320 if ((issued & CEPH_CAP_LINK_EXCL) == 0)
2321 inode->i_nlink = le32_to_cpu(grant->nlink);
2322
2323 if ((issued & CEPH_CAP_XATTR_EXCL) == 0 && grant->xattr_len) {
2324 int len = le32_to_cpu(grant->xattr_len);
2325 u64 version = le64_to_cpu(grant->xattr_version);
2326
2327 if (version > ci->i_xattrs.version) {
2328 dout(" got new xattrs v%llu on %p len %d\n",
2329 version, inode, len);
2330 if (ci->i_xattrs.blob)
2331 ceph_buffer_put(ci->i_xattrs.blob);
2332 ci->i_xattrs.blob = ceph_buffer_get(xattr_buf);
2333 ci->i_xattrs.version = version;
2334 }
2335 }
2336
2337 /* size/ctime/mtime/atime? */
2338 ceph_fill_file_size(inode, issued,
2339 le32_to_cpu(grant->truncate_seq),
2340 le64_to_cpu(grant->truncate_size), size);
2341 ceph_decode_timespec(&mtime, &grant->mtime);
2342 ceph_decode_timespec(&atime, &grant->atime);
2343 ceph_decode_timespec(&ctime, &grant->ctime);
2344 ceph_fill_file_time(inode, issued,
2345 le32_to_cpu(grant->time_warp_seq), &ctime, &mtime,
2346 &atime);
2347
2348 /* max size increase? */
2349 if (max_size != ci->i_max_size) {
2350 dout("max_size %lld -> %llu\n", ci->i_max_size, max_size);
2351 ci->i_max_size = max_size;
2352 if (max_size >= ci->i_wanted_max_size) {
2353 ci->i_wanted_max_size = 0; /* reset */
2354 ci->i_requested_max_size = 0;
2355 }
2356 wake = 1;
2357 }
2358
2359 /* check cap bits */
2360 wanted = __ceph_caps_wanted(ci);
2361 used = __ceph_caps_used(ci);
2362 dirty = __ceph_caps_dirty(ci);
2363 dout(" my wanted = %s, used = %s, dirty %s\n",
2364 ceph_cap_string(wanted),
2365 ceph_cap_string(used),
2366 ceph_cap_string(dirty));
2367 if (wanted != le32_to_cpu(grant->wanted)) {
2368 dout("mds wanted %s -> %s\n",
2369 ceph_cap_string(le32_to_cpu(grant->wanted)),
2370 ceph_cap_string(wanted));
2371 grant->wanted = cpu_to_le32(wanted);
2372 }
2373
2374 cap->seq = seq;
2375
2376 /* file layout may have changed */
2377 ci->i_layout = grant->layout;
2378
2379 /* revocation, grant, or no-op? */
2380 if (cap->issued & ~newcaps) {
Sage Weil3b454c42010-06-10 13:20:33 -07002381 int revoking = cap->issued & ~newcaps;
2382
2383 dout("revocation: %s -> %s (revoking %s)\n",
2384 ceph_cap_string(cap->issued),
2385 ceph_cap_string(newcaps),
2386 ceph_cap_string(revoking));
2387 if (revoking & CEPH_CAP_FILE_BUFFER)
2388 writeback = 1; /* initiate writeback; will delay ack */
2389 else if (revoking == CEPH_CAP_FILE_CACHE &&
2390 (newcaps & CEPH_CAP_FILE_LAZYIO) == 0 &&
2391 queue_invalidate)
2392 ; /* do nothing yet, invalidation will be queued */
2393 else if (cap == ci->i_auth_cap)
2394 check_caps = 1; /* check auth cap only */
2395 else
2396 check_caps = 2; /* check all caps */
Sage Weila8599bd2009-10-06 11:31:12 -07002397 cap->issued = newcaps;
Sage Weil978097c2010-03-08 15:27:53 -08002398 cap->implemented |= newcaps;
Sage Weila8599bd2009-10-06 11:31:12 -07002399 } else if (cap->issued == newcaps) {
2400 dout("caps unchanged: %s -> %s\n",
2401 ceph_cap_string(cap->issued), ceph_cap_string(newcaps));
2402 } else {
2403 dout("grant: %s -> %s\n", ceph_cap_string(cap->issued),
2404 ceph_cap_string(newcaps));
2405 cap->issued = newcaps;
2406 cap->implemented |= newcaps; /* add bits only, to
2407 * avoid stepping on a
2408 * pending revocation */
2409 wake = 1;
2410 }
Sage Weil978097c2010-03-08 15:27:53 -08002411 BUG_ON(cap->issued & ~cap->implemented);
Sage Weila8599bd2009-10-06 11:31:12 -07002412
2413 spin_unlock(&inode->i_lock);
Sage Weil3c6f6b72010-02-09 15:24:44 -08002414 if (writeback)
Sage Weila8599bd2009-10-06 11:31:12 -07002415 /*
2416 * queue inode for writeback: we can't actually call
2417 * filemap_write_and_wait, etc. from message handler
2418 * context.
2419 */
Sage Weil3c6f6b72010-02-09 15:24:44 -08002420 ceph_queue_writeback(inode);
2421 if (queue_invalidate)
2422 ceph_queue_invalidate(inode);
Sage Weila8599bd2009-10-06 11:31:12 -07002423 if (wake)
Yehuda Sadeh03066f22010-07-27 13:11:08 -07002424 wake_up_all(&ci->i_cap_wq);
Sage Weil15637c82010-03-16 13:42:00 -07002425
2426 if (check_caps == 1)
2427 ceph_check_caps(ci, CHECK_CAPS_NODELAY|CHECK_CAPS_AUTHONLY,
2428 session);
2429 else if (check_caps == 2)
2430 ceph_check_caps(ci, CHECK_CAPS_NODELAY, session);
2431 else
2432 mutex_unlock(&session->s_mutex);
Sage Weila8599bd2009-10-06 11:31:12 -07002433}
2434
2435/*
2436 * Handle FLUSH_ACK from MDS, indicating that metadata we sent to the
2437 * MDS has been safely committed.
2438 */
Sage Weil6df058c2009-12-22 11:24:33 -08002439static void handle_cap_flush_ack(struct inode *inode, u64 flush_tid,
Sage Weila8599bd2009-10-06 11:31:12 -07002440 struct ceph_mds_caps *m,
2441 struct ceph_mds_session *session,
2442 struct ceph_cap *cap)
2443 __releases(inode->i_lock)
2444{
2445 struct ceph_inode_info *ci = ceph_inode(inode);
Cheng Renquan640ef792010-03-26 17:40:33 +08002446 struct ceph_mds_client *mdsc = &ceph_sb_to_client(inode->i_sb)->mdsc;
Sage Weila8599bd2009-10-06 11:31:12 -07002447 unsigned seq = le32_to_cpu(m->seq);
2448 int dirty = le32_to_cpu(m->dirty);
2449 int cleaned = 0;
Sage Weilafcdaea2009-10-14 14:27:38 -07002450 int drop = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07002451 int i;
2452
2453 for (i = 0; i < CEPH_CAP_BITS; i++)
2454 if ((dirty & (1 << i)) &&
2455 flush_tid == ci->i_cap_flush_tid[i])
2456 cleaned |= 1 << i;
2457
2458 dout("handle_cap_flush_ack inode %p mds%d seq %d on %s cleaned %s,"
2459 " flushing %s -> %s\n",
2460 inode, session->s_mds, seq, ceph_cap_string(dirty),
2461 ceph_cap_string(cleaned), ceph_cap_string(ci->i_flushing_caps),
2462 ceph_cap_string(ci->i_flushing_caps & ~cleaned));
2463
2464 if (ci->i_flushing_caps == (ci->i_flushing_caps & ~cleaned))
2465 goto out;
2466
Sage Weila8599bd2009-10-06 11:31:12 -07002467 ci->i_flushing_caps &= ~cleaned;
Sage Weila8599bd2009-10-06 11:31:12 -07002468
2469 spin_lock(&mdsc->cap_dirty_lock);
2470 if (ci->i_flushing_caps == 0) {
2471 list_del_init(&ci->i_flushing_item);
2472 if (!list_empty(&session->s_cap_flushing))
2473 dout(" mds%d still flushing cap on %p\n",
2474 session->s_mds,
2475 &list_entry(session->s_cap_flushing.next,
2476 struct ceph_inode_info,
2477 i_flushing_item)->vfs_inode);
2478 mdsc->num_cap_flushing--;
Yehuda Sadeh03066f22010-07-27 13:11:08 -07002479 wake_up_all(&mdsc->cap_flushing_wq);
Sage Weila8599bd2009-10-06 11:31:12 -07002480 dout(" inode %p now !flushing\n", inode);
Sage Weilafcdaea2009-10-14 14:27:38 -07002481
2482 if (ci->i_dirty_caps == 0) {
2483 dout(" inode %p now clean\n", inode);
2484 BUG_ON(!list_empty(&ci->i_dirty_item));
2485 drop = 1;
Sage Weil76e3b392009-10-15 18:13:53 -07002486 } else {
2487 BUG_ON(list_empty(&ci->i_dirty_item));
Sage Weilafcdaea2009-10-14 14:27:38 -07002488 }
Sage Weila8599bd2009-10-06 11:31:12 -07002489 }
2490 spin_unlock(&mdsc->cap_dirty_lock);
Yehuda Sadeh03066f22010-07-27 13:11:08 -07002491 wake_up_all(&ci->i_cap_wq);
Sage Weila8599bd2009-10-06 11:31:12 -07002492
2493out:
2494 spin_unlock(&inode->i_lock);
Sage Weilafcdaea2009-10-14 14:27:38 -07002495 if (drop)
Sage Weila8599bd2009-10-06 11:31:12 -07002496 iput(inode);
2497}
2498
2499/*
2500 * Handle FLUSHSNAP_ACK. MDS has flushed snap data to disk and we can
2501 * throw away our cap_snap.
2502 *
2503 * Caller hold s_mutex.
2504 */
Sage Weil6df058c2009-12-22 11:24:33 -08002505static void handle_cap_flushsnap_ack(struct inode *inode, u64 flush_tid,
Sage Weila8599bd2009-10-06 11:31:12 -07002506 struct ceph_mds_caps *m,
2507 struct ceph_mds_session *session)
2508{
2509 struct ceph_inode_info *ci = ceph_inode(inode);
2510 u64 follows = le64_to_cpu(m->snap_follows);
Sage Weila8599bd2009-10-06 11:31:12 -07002511 struct ceph_cap_snap *capsnap;
2512 int drop = 0;
2513
2514 dout("handle_cap_flushsnap_ack inode %p ci %p mds%d follows %lld\n",
2515 inode, ci, session->s_mds, follows);
2516
2517 spin_lock(&inode->i_lock);
2518 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
2519 if (capsnap->follows == follows) {
2520 if (capsnap->flush_tid != flush_tid) {
2521 dout(" cap_snap %p follows %lld tid %lld !="
2522 " %lld\n", capsnap, follows,
2523 flush_tid, capsnap->flush_tid);
2524 break;
2525 }
2526 WARN_ON(capsnap->dirty_pages || capsnap->writing);
Sage Weil819ccbf2010-04-01 09:33:46 -07002527 dout(" removing %p cap_snap %p follows %lld\n",
2528 inode, capsnap, follows);
Sage Weila8599bd2009-10-06 11:31:12 -07002529 ceph_put_snap_context(capsnap->context);
2530 list_del(&capsnap->ci_item);
2531 list_del(&capsnap->flushing_item);
2532 ceph_put_cap_snap(capsnap);
2533 drop = 1;
2534 break;
2535 } else {
2536 dout(" skipping cap_snap %p follows %lld\n",
2537 capsnap, capsnap->follows);
2538 }
2539 }
2540 spin_unlock(&inode->i_lock);
2541 if (drop)
2542 iput(inode);
2543}
2544
2545/*
2546 * Handle TRUNC from MDS, indicating file truncation.
2547 *
2548 * caller hold s_mutex.
2549 */
2550static void handle_cap_trunc(struct inode *inode,
2551 struct ceph_mds_caps *trunc,
2552 struct ceph_mds_session *session)
2553 __releases(inode->i_lock)
2554{
2555 struct ceph_inode_info *ci = ceph_inode(inode);
2556 int mds = session->s_mds;
2557 int seq = le32_to_cpu(trunc->seq);
2558 u32 truncate_seq = le32_to_cpu(trunc->truncate_seq);
2559 u64 truncate_size = le64_to_cpu(trunc->truncate_size);
2560 u64 size = le64_to_cpu(trunc->size);
2561 int implemented = 0;
2562 int dirty = __ceph_caps_dirty(ci);
2563 int issued = __ceph_caps_issued(ceph_inode(inode), &implemented);
2564 int queue_trunc = 0;
2565
2566 issued |= implemented | dirty;
2567
2568 dout("handle_cap_trunc inode %p mds%d seq %d to %lld seq %d\n",
2569 inode, mds, seq, truncate_size, truncate_seq);
2570 queue_trunc = ceph_fill_file_size(inode, issued,
2571 truncate_seq, truncate_size, size);
2572 spin_unlock(&inode->i_lock);
2573
2574 if (queue_trunc)
Sage Weil3c6f6b72010-02-09 15:24:44 -08002575 ceph_queue_vmtruncate(inode);
Sage Weila8599bd2009-10-06 11:31:12 -07002576}
2577
2578/*
2579 * Handle EXPORT from MDS. Cap is being migrated _from_ this mds to a
2580 * different one. If we are the most recent migration we've seen (as
2581 * indicated by mseq), make note of the migrating cap bits for the
2582 * duration (until we see the corresponding IMPORT).
2583 *
2584 * caller holds s_mutex
2585 */
2586static void handle_cap_export(struct inode *inode, struct ceph_mds_caps *ex,
Sage Weil154f42c2010-06-21 13:45:04 -07002587 struct ceph_mds_session *session,
2588 int *open_target_sessions)
Sage Weila8599bd2009-10-06 11:31:12 -07002589{
2590 struct ceph_inode_info *ci = ceph_inode(inode);
2591 int mds = session->s_mds;
2592 unsigned mseq = le32_to_cpu(ex->migrate_seq);
2593 struct ceph_cap *cap = NULL, *t;
2594 struct rb_node *p;
2595 int remember = 1;
2596
2597 dout("handle_cap_export inode %p ci %p mds%d mseq %d\n",
2598 inode, ci, mds, mseq);
2599
2600 spin_lock(&inode->i_lock);
2601
2602 /* make sure we haven't seen a higher mseq */
2603 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
2604 t = rb_entry(p, struct ceph_cap, ci_node);
2605 if (ceph_seq_cmp(t->mseq, mseq) > 0) {
2606 dout(" higher mseq on cap from mds%d\n",
2607 t->session->s_mds);
2608 remember = 0;
2609 }
2610 if (t->session->s_mds == mds)
2611 cap = t;
2612 }
2613
2614 if (cap) {
2615 if (remember) {
2616 /* make note */
2617 ci->i_cap_exporting_mds = mds;
2618 ci->i_cap_exporting_mseq = mseq;
2619 ci->i_cap_exporting_issued = cap->issued;
Sage Weil154f42c2010-06-21 13:45:04 -07002620
2621 /*
2622 * make sure we have open sessions with all possible
2623 * export targets, so that we get the matching IMPORT
2624 */
2625 *open_target_sessions = 1;
Sage Weila8599bd2009-10-06 11:31:12 -07002626 }
Sage Weil7c1332b2010-02-16 11:39:45 -08002627 __ceph_remove_cap(cap);
Sage Weila8599bd2009-10-06 11:31:12 -07002628 }
Sage Weil4ea00432010-03-16 10:36:40 -07002629 /* else, we already released it */
Sage Weila8599bd2009-10-06 11:31:12 -07002630
2631 spin_unlock(&inode->i_lock);
2632}
2633
2634/*
2635 * Handle cap IMPORT. If there are temp bits from an older EXPORT,
2636 * clean them up.
2637 *
2638 * caller holds s_mutex.
2639 */
2640static void handle_cap_import(struct ceph_mds_client *mdsc,
2641 struct inode *inode, struct ceph_mds_caps *im,
2642 struct ceph_mds_session *session,
2643 void *snaptrace, int snaptrace_len)
2644{
2645 struct ceph_inode_info *ci = ceph_inode(inode);
2646 int mds = session->s_mds;
2647 unsigned issued = le32_to_cpu(im->caps);
2648 unsigned wanted = le32_to_cpu(im->wanted);
2649 unsigned seq = le32_to_cpu(im->seq);
2650 unsigned mseq = le32_to_cpu(im->migrate_seq);
2651 u64 realmino = le64_to_cpu(im->realm);
2652 u64 cap_id = le64_to_cpu(im->cap_id);
2653
2654 if (ci->i_cap_exporting_mds >= 0 &&
2655 ceph_seq_cmp(ci->i_cap_exporting_mseq, mseq) < 0) {
2656 dout("handle_cap_import inode %p ci %p mds%d mseq %d"
2657 " - cleared exporting from mds%d\n",
2658 inode, ci, mds, mseq,
2659 ci->i_cap_exporting_mds);
2660 ci->i_cap_exporting_issued = 0;
2661 ci->i_cap_exporting_mseq = 0;
2662 ci->i_cap_exporting_mds = -1;
2663 } else {
2664 dout("handle_cap_import inode %p ci %p mds%d mseq %d\n",
2665 inode, ci, mds, mseq);
2666 }
2667
2668 down_write(&mdsc->snap_rwsem);
2669 ceph_update_snap_trace(mdsc, snaptrace, snaptrace+snaptrace_len,
2670 false);
2671 downgrade_write(&mdsc->snap_rwsem);
2672 ceph_add_cap(inode, session, cap_id, -1,
2673 issued, wanted, seq, mseq, realmino, CEPH_CAP_FLAG_AUTH,
2674 NULL /* no caps context */);
2675 try_flush_caps(inode, session, NULL);
2676 up_read(&mdsc->snap_rwsem);
2677}
2678
2679/*
2680 * Handle a caps message from the MDS.
2681 *
2682 * Identify the appropriate session, inode, and call the right handler
2683 * based on the cap op.
2684 */
2685void ceph_handle_caps(struct ceph_mds_session *session,
2686 struct ceph_msg *msg)
2687{
2688 struct ceph_mds_client *mdsc = session->s_mdsc;
2689 struct super_block *sb = mdsc->client->sb;
2690 struct inode *inode;
2691 struct ceph_cap *cap;
2692 struct ceph_mds_caps *h;
Sage Weil2600d2d2010-02-22 15:12:16 -08002693 int mds = session->s_mds;
Sage Weila8599bd2009-10-06 11:31:12 -07002694 int op;
Sage Weil3d7ded42010-06-09 16:47:10 -07002695 u32 seq, mseq;
Sage Weila8599bd2009-10-06 11:31:12 -07002696 struct ceph_vino vino;
2697 u64 cap_id;
2698 u64 size, max_size;
Sage Weil6df058c2009-12-22 11:24:33 -08002699 u64 tid;
Sage Weil70edb552010-03-01 13:20:50 -08002700 void *snaptrace;
Sage Weil154f42c2010-06-21 13:45:04 -07002701 int open_target_sessions = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07002702
2703 dout("handle_caps from mds%d\n", mds);
2704
2705 /* decode */
Sage Weil6df058c2009-12-22 11:24:33 -08002706 tid = le64_to_cpu(msg->hdr.tid);
Sage Weila8599bd2009-10-06 11:31:12 -07002707 if (msg->front.iov_len < sizeof(*h))
2708 goto bad;
2709 h = msg->front.iov_base;
Sage Weil70edb552010-03-01 13:20:50 -08002710 snaptrace = h + 1;
Sage Weila8599bd2009-10-06 11:31:12 -07002711 op = le32_to_cpu(h->op);
2712 vino.ino = le64_to_cpu(h->ino);
2713 vino.snap = CEPH_NOSNAP;
2714 cap_id = le64_to_cpu(h->cap_id);
2715 seq = le32_to_cpu(h->seq);
Sage Weil3d7ded42010-06-09 16:47:10 -07002716 mseq = le32_to_cpu(h->migrate_seq);
Sage Weila8599bd2009-10-06 11:31:12 -07002717 size = le64_to_cpu(h->size);
2718 max_size = le64_to_cpu(h->max_size);
2719
2720 mutex_lock(&session->s_mutex);
2721 session->s_seq++;
2722 dout(" mds%d seq %lld cap seq %u\n", session->s_mds, session->s_seq,
2723 (unsigned)seq);
2724
2725 /* lookup ino */
2726 inode = ceph_find_inode(sb, vino);
2727 dout(" op %s ino %llx.%llx inode %p\n", ceph_cap_op_name(op), vino.ino,
2728 vino.snap, inode);
2729 if (!inode) {
2730 dout(" i don't have ino %llx\n", vino.ino);
Sage Weil3d7ded42010-06-09 16:47:10 -07002731
2732 if (op == CEPH_CAP_OP_IMPORT)
2733 __queue_cap_release(session, vino.ino, cap_id,
2734 mseq, seq);
2735
2736 /*
2737 * send any full release message to try to move things
2738 * along for the mds (who clearly thinks we still have this
2739 * cap).
2740 */
Sage Weilee6b2722010-06-10 12:55:52 -07002741 ceph_add_cap_releases(mdsc, session);
Sage Weil3d7ded42010-06-09 16:47:10 -07002742 ceph_send_cap_releases(mdsc, session);
Sage Weila8599bd2009-10-06 11:31:12 -07002743 goto done;
2744 }
2745
2746 /* these will work even if we don't have a cap yet */
2747 switch (op) {
2748 case CEPH_CAP_OP_FLUSHSNAP_ACK:
Sage Weil6df058c2009-12-22 11:24:33 -08002749 handle_cap_flushsnap_ack(inode, tid, h, session);
Sage Weila8599bd2009-10-06 11:31:12 -07002750 goto done;
2751
2752 case CEPH_CAP_OP_EXPORT:
Sage Weil154f42c2010-06-21 13:45:04 -07002753 handle_cap_export(inode, h, session, &open_target_sessions);
Sage Weila8599bd2009-10-06 11:31:12 -07002754 goto done;
2755
2756 case CEPH_CAP_OP_IMPORT:
2757 handle_cap_import(mdsc, inode, h, session,
Sage Weil70edb552010-03-01 13:20:50 -08002758 snaptrace, le32_to_cpu(h->snap_trace_len));
Sage Weil15637c82010-03-16 13:42:00 -07002759 ceph_check_caps(ceph_inode(inode), CHECK_CAPS_NODELAY,
2760 session);
2761 goto done_unlocked;
Sage Weila8599bd2009-10-06 11:31:12 -07002762 }
2763
2764 /* the rest require a cap */
2765 spin_lock(&inode->i_lock);
2766 cap = __get_cap_for_mds(ceph_inode(inode), mds);
2767 if (!cap) {
Sage Weil9dbd4122010-06-10 13:21:20 -07002768 dout(" no cap on %p ino %llx.%llx from mds%d\n",
Sage Weila8599bd2009-10-06 11:31:12 -07002769 inode, ceph_ino(inode), ceph_snap(inode), mds);
2770 spin_unlock(&inode->i_lock);
2771 goto done;
2772 }
2773
2774 /* note that each of these drops i_lock for us */
2775 switch (op) {
2776 case CEPH_CAP_OP_REVOKE:
2777 case CEPH_CAP_OP_GRANT:
Sage Weil15637c82010-03-16 13:42:00 -07002778 handle_cap_grant(inode, h, session, cap, msg->middle);
2779 goto done_unlocked;
Sage Weila8599bd2009-10-06 11:31:12 -07002780
2781 case CEPH_CAP_OP_FLUSH_ACK:
Sage Weil6df058c2009-12-22 11:24:33 -08002782 handle_cap_flush_ack(inode, tid, h, session, cap);
Sage Weila8599bd2009-10-06 11:31:12 -07002783 break;
2784
2785 case CEPH_CAP_OP_TRUNC:
2786 handle_cap_trunc(inode, h, session);
2787 break;
2788
2789 default:
2790 spin_unlock(&inode->i_lock);
2791 pr_err("ceph_handle_caps: unknown cap op %d %s\n", op,
2792 ceph_cap_op_name(op));
2793 }
2794
2795done:
Sage Weil15637c82010-03-16 13:42:00 -07002796 mutex_unlock(&session->s_mutex);
2797done_unlocked:
Sage Weila8599bd2009-10-06 11:31:12 -07002798 if (inode)
2799 iput(inode);
Sage Weil154f42c2010-06-21 13:45:04 -07002800 if (open_target_sessions)
2801 ceph_mdsc_open_export_target_sessions(mdsc, session);
Sage Weila8599bd2009-10-06 11:31:12 -07002802 return;
2803
2804bad:
2805 pr_err("ceph_handle_caps: corrupt message\n");
Sage Weil9ec7cab2009-12-14 15:13:47 -08002806 ceph_msg_dump(msg);
Sage Weila8599bd2009-10-06 11:31:12 -07002807 return;
2808}
2809
2810/*
2811 * Delayed work handler to process end of delayed cap release LRU list.
2812 */
Sage Weilafcdaea2009-10-14 14:27:38 -07002813void ceph_check_delayed_caps(struct ceph_mds_client *mdsc)
Sage Weila8599bd2009-10-06 11:31:12 -07002814{
2815 struct ceph_inode_info *ci;
2816 int flags = CHECK_CAPS_NODELAY;
2817
Sage Weila8599bd2009-10-06 11:31:12 -07002818 dout("check_delayed_caps\n");
2819 while (1) {
2820 spin_lock(&mdsc->cap_delay_lock);
2821 if (list_empty(&mdsc->cap_delay_list))
2822 break;
2823 ci = list_first_entry(&mdsc->cap_delay_list,
2824 struct ceph_inode_info,
2825 i_cap_delay_list);
2826 if ((ci->i_ceph_flags & CEPH_I_FLUSH) == 0 &&
2827 time_before(jiffies, ci->i_hold_caps_max))
2828 break;
2829 list_del_init(&ci->i_cap_delay_list);
2830 spin_unlock(&mdsc->cap_delay_lock);
2831 dout("check_delayed_caps on %p\n", &ci->vfs_inode);
2832 ceph_check_caps(ci, flags, NULL);
2833 }
2834 spin_unlock(&mdsc->cap_delay_lock);
2835}
2836
2837/*
Sage Weilafcdaea2009-10-14 14:27:38 -07002838 * Flush all dirty caps to the mds
2839 */
2840void ceph_flush_dirty_caps(struct ceph_mds_client *mdsc)
2841{
Sage Weile9964c12010-03-01 15:16:56 -08002842 struct ceph_inode_info *ci, *nci = NULL;
2843 struct inode *inode, *ninode = NULL;
2844 struct list_head *p, *n;
Sage Weilafcdaea2009-10-14 14:27:38 -07002845
2846 dout("flush_dirty_caps\n");
2847 spin_lock(&mdsc->cap_dirty_lock);
Sage Weile9964c12010-03-01 15:16:56 -08002848 list_for_each_safe(p, n, &mdsc->cap_dirty) {
2849 if (nci) {
2850 ci = nci;
2851 inode = ninode;
2852 ci->i_ceph_flags &= ~CEPH_I_NOFLUSH;
2853 dout("flush_dirty_caps inode %p (was next inode)\n",
2854 inode);
2855 } else {
2856 ci = list_entry(p, struct ceph_inode_info,
2857 i_dirty_item);
2858 inode = igrab(&ci->vfs_inode);
2859 BUG_ON(!inode);
2860 dout("flush_dirty_caps inode %p\n", inode);
2861 }
2862 if (n != &mdsc->cap_dirty) {
2863 nci = list_entry(n, struct ceph_inode_info,
2864 i_dirty_item);
2865 ninode = igrab(&nci->vfs_inode);
2866 BUG_ON(!ninode);
2867 nci->i_ceph_flags |= CEPH_I_NOFLUSH;
2868 dout("flush_dirty_caps next inode %p, noflush\n",
2869 ninode);
2870 } else {
2871 nci = NULL;
2872 ninode = NULL;
2873 }
Sage Weilafcdaea2009-10-14 14:27:38 -07002874 spin_unlock(&mdsc->cap_dirty_lock);
2875 if (inode) {
2876 ceph_check_caps(ci, CHECK_CAPS_NODELAY|CHECK_CAPS_FLUSH,
2877 NULL);
2878 iput(inode);
2879 }
2880 spin_lock(&mdsc->cap_dirty_lock);
2881 }
2882 spin_unlock(&mdsc->cap_dirty_lock);
2883}
2884
2885/*
Sage Weila8599bd2009-10-06 11:31:12 -07002886 * Drop open file reference. If we were the last open file,
2887 * we may need to release capabilities to the MDS (or schedule
2888 * their delayed release).
2889 */
2890void ceph_put_fmode(struct ceph_inode_info *ci, int fmode)
2891{
2892 struct inode *inode = &ci->vfs_inode;
2893 int last = 0;
2894
2895 spin_lock(&inode->i_lock);
2896 dout("put_fmode %p fmode %d %d -> %d\n", inode, fmode,
2897 ci->i_nr_by_mode[fmode], ci->i_nr_by_mode[fmode]-1);
2898 BUG_ON(ci->i_nr_by_mode[fmode] == 0);
2899 if (--ci->i_nr_by_mode[fmode] == 0)
2900 last++;
2901 spin_unlock(&inode->i_lock);
2902
2903 if (last && ci->i_vino.snap == CEPH_NOSNAP)
2904 ceph_check_caps(ci, 0, NULL);
2905}
2906
2907/*
2908 * Helpers for embedding cap and dentry lease releases into mds
2909 * requests.
2910 *
2911 * @force is used by dentry_release (below) to force inclusion of a
2912 * record for the directory inode, even when there aren't any caps to
2913 * drop.
2914 */
2915int ceph_encode_inode_release(void **p, struct inode *inode,
2916 int mds, int drop, int unless, int force)
2917{
2918 struct ceph_inode_info *ci = ceph_inode(inode);
2919 struct ceph_cap *cap;
2920 struct ceph_mds_request_release *rel = *p;
Sage Weilec97f882010-06-24 15:12:37 -07002921 int used, dirty;
Sage Weila8599bd2009-10-06 11:31:12 -07002922 int ret = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07002923
2924 spin_lock(&inode->i_lock);
Sage Weil916623d2010-03-16 15:01:07 -07002925 used = __ceph_caps_used(ci);
Sage Weilec97f882010-06-24 15:12:37 -07002926 dirty = __ceph_caps_dirty(ci);
Sage Weil916623d2010-03-16 15:01:07 -07002927
Sage Weilec97f882010-06-24 15:12:37 -07002928 dout("encode_inode_release %p mds%d used|dirty %s drop %s unless %s\n",
2929 inode, mds, ceph_cap_string(used|dirty), ceph_cap_string(drop),
Sage Weil916623d2010-03-16 15:01:07 -07002930 ceph_cap_string(unless));
2931
Sage Weilec97f882010-06-24 15:12:37 -07002932 /* only drop unused, clean caps */
2933 drop &= ~(used | dirty);
Sage Weil916623d2010-03-16 15:01:07 -07002934
Sage Weila8599bd2009-10-06 11:31:12 -07002935 cap = __get_cap_for_mds(ci, mds);
2936 if (cap && __cap_is_valid(cap)) {
2937 if (force ||
2938 ((cap->issued & drop) &&
2939 (cap->issued & unless) == 0)) {
2940 if ((cap->issued & drop) &&
2941 (cap->issued & unless) == 0) {
2942 dout("encode_inode_release %p cap %p %s -> "
2943 "%s\n", inode, cap,
2944 ceph_cap_string(cap->issued),
2945 ceph_cap_string(cap->issued & ~drop));
2946 cap->issued &= ~drop;
2947 cap->implemented &= ~drop;
2948 if (ci->i_ceph_flags & CEPH_I_NODELAY) {
2949 int wanted = __ceph_caps_wanted(ci);
2950 dout(" wanted %s -> %s (act %s)\n",
2951 ceph_cap_string(cap->mds_wanted),
2952 ceph_cap_string(cap->mds_wanted &
2953 ~wanted),
2954 ceph_cap_string(wanted));
2955 cap->mds_wanted &= wanted;
2956 }
2957 } else {
2958 dout("encode_inode_release %p cap %p %s"
2959 " (force)\n", inode, cap,
2960 ceph_cap_string(cap->issued));
2961 }
2962
2963 rel->ino = cpu_to_le64(ceph_ino(inode));
2964 rel->cap_id = cpu_to_le64(cap->cap_id);
2965 rel->seq = cpu_to_le32(cap->seq);
2966 rel->issue_seq = cpu_to_le32(cap->issue_seq),
2967 rel->mseq = cpu_to_le32(cap->mseq);
2968 rel->caps = cpu_to_le32(cap->issued);
2969 rel->wanted = cpu_to_le32(cap->mds_wanted);
2970 rel->dname_len = 0;
2971 rel->dname_seq = 0;
2972 *p += sizeof(*rel);
2973 ret = 1;
2974 } else {
2975 dout("encode_inode_release %p cap %p %s\n",
2976 inode, cap, ceph_cap_string(cap->issued));
2977 }
2978 }
2979 spin_unlock(&inode->i_lock);
2980 return ret;
2981}
2982
2983int ceph_encode_dentry_release(void **p, struct dentry *dentry,
2984 int mds, int drop, int unless)
2985{
2986 struct inode *dir = dentry->d_parent->d_inode;
2987 struct ceph_mds_request_release *rel = *p;
2988 struct ceph_dentry_info *di = ceph_dentry(dentry);
2989 int force = 0;
2990 int ret;
2991
2992 /*
2993 * force an record for the directory caps if we have a dentry lease.
2994 * this is racy (can't take i_lock and d_lock together), but it
2995 * doesn't have to be perfect; the mds will revoke anything we don't
2996 * release.
2997 */
2998 spin_lock(&dentry->d_lock);
2999 if (di->lease_session && di->lease_session->s_mds == mds)
3000 force = 1;
3001 spin_unlock(&dentry->d_lock);
3002
3003 ret = ceph_encode_inode_release(p, dir, mds, drop, unless, force);
3004
3005 spin_lock(&dentry->d_lock);
3006 if (ret && di->lease_session && di->lease_session->s_mds == mds) {
3007 dout("encode_dentry_release %p mds%d seq %d\n",
3008 dentry, mds, (int)di->lease_seq);
3009 rel->dname_len = cpu_to_le32(dentry->d_name.len);
3010 memcpy(*p, dentry->d_name.name, dentry->d_name.len);
3011 *p += dentry->d_name.len;
3012 rel->dname_seq = cpu_to_le32(di->lease_seq);
Sage Weil1dadcce2010-07-23 13:54:21 -07003013 __ceph_mdsc_drop_dentry_lease(dentry);
Sage Weila8599bd2009-10-06 11:31:12 -07003014 }
3015 spin_unlock(&dentry->d_lock);
3016 return ret;
3017}