blob: e3b848d7698de2dd65b2a69490937500d4394599 [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);
589 }
590 }
591
592 __check_cap_issue(ci, cap, issued);
593
594 /*
595 * If we are issued caps we don't want, or the mds' wanted
596 * value appears to be off, queue a check so we'll release
597 * later and/or update the mds wanted value.
598 */
599 actual_wanted = __ceph_caps_wanted(ci);
600 if ((wanted & ~actual_wanted) ||
601 (issued & ~actual_wanted & CEPH_CAP_ANY_WR)) {
602 dout(" issued %s, mds wanted %s, actual %s, queueing\n",
603 ceph_cap_string(issued), ceph_cap_string(wanted),
604 ceph_cap_string(actual_wanted));
605 __cap_delay_requeue(mdsc, ci);
606 }
607
608 if (flags & CEPH_CAP_FLAG_AUTH)
609 ci->i_auth_cap = cap;
610 else if (ci->i_auth_cap == cap)
611 ci->i_auth_cap = NULL;
612
613 dout("add_cap inode %p (%llx.%llx) cap %p %s now %s seq %d mds%d\n",
614 inode, ceph_vinop(inode), cap, ceph_cap_string(issued),
615 ceph_cap_string(issued|cap->issued), seq, mds);
616 cap->cap_id = cap_id;
617 cap->issued = issued;
618 cap->implemented |= issued;
619 cap->mds_wanted |= wanted;
620 cap->seq = seq;
621 cap->issue_seq = seq;
622 cap->mseq = mseq;
Sage Weil685f9a5d2009-11-09 12:05:48 -0800623 cap->cap_gen = session->s_cap_gen;
Sage Weila8599bd2009-10-06 11:31:12 -0700624
625 if (fmode >= 0)
626 __ceph_get_fmode(ci, fmode);
627 spin_unlock(&inode->i_lock);
Yehuda Sadeh03066f22010-07-27 13:11:08 -0700628 wake_up_all(&ci->i_cap_wq);
Sage Weila8599bd2009-10-06 11:31:12 -0700629 return 0;
630}
631
632/*
633 * Return true if cap has not timed out and belongs to the current
634 * generation of the MDS session (i.e. has not gone 'stale' due to
635 * us losing touch with the mds).
636 */
637static int __cap_is_valid(struct ceph_cap *cap)
638{
639 unsigned long ttl;
Sage Weilcdac8302009-11-10 16:02:23 -0800640 u32 gen;
Sage Weila8599bd2009-10-06 11:31:12 -0700641
642 spin_lock(&cap->session->s_cap_lock);
643 gen = cap->session->s_cap_gen;
644 ttl = cap->session->s_cap_ttl;
645 spin_unlock(&cap->session->s_cap_lock);
646
Sage Weil685f9a5d2009-11-09 12:05:48 -0800647 if (cap->cap_gen < gen || time_after_eq(jiffies, ttl)) {
Sage Weila8599bd2009-10-06 11:31:12 -0700648 dout("__cap_is_valid %p cap %p issued %s "
649 "but STALE (gen %u vs %u)\n", &cap->ci->vfs_inode,
Sage Weil685f9a5d2009-11-09 12:05:48 -0800650 cap, ceph_cap_string(cap->issued), cap->cap_gen, gen);
Sage Weila8599bd2009-10-06 11:31:12 -0700651 return 0;
652 }
653
654 return 1;
655}
656
657/*
658 * Return set of valid cap bits issued to us. Note that caps time
659 * out, and may be invalidated in bulk if the client session times out
660 * and session->s_cap_gen is bumped.
661 */
662int __ceph_caps_issued(struct ceph_inode_info *ci, int *implemented)
663{
Sage Weil7af8f1e2010-03-01 15:17:34 -0800664 int have = ci->i_snap_caps | ci->i_cap_exporting_issued;
Sage Weila8599bd2009-10-06 11:31:12 -0700665 struct ceph_cap *cap;
666 struct rb_node *p;
667
668 if (implemented)
669 *implemented = 0;
670 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
671 cap = rb_entry(p, struct ceph_cap, ci_node);
672 if (!__cap_is_valid(cap))
673 continue;
674 dout("__ceph_caps_issued %p cap %p issued %s\n",
675 &ci->vfs_inode, cap, ceph_cap_string(cap->issued));
676 have |= cap->issued;
677 if (implemented)
678 *implemented |= cap->implemented;
679 }
680 return have;
681}
682
683/*
684 * Get cap bits issued by caps other than @ocap
685 */
686int __ceph_caps_issued_other(struct ceph_inode_info *ci, struct ceph_cap *ocap)
687{
688 int have = ci->i_snap_caps;
689 struct ceph_cap *cap;
690 struct rb_node *p;
691
692 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
693 cap = rb_entry(p, struct ceph_cap, ci_node);
694 if (cap == ocap)
695 continue;
696 if (!__cap_is_valid(cap))
697 continue;
698 have |= cap->issued;
699 }
700 return have;
701}
702
703/*
704 * Move a cap to the end of the LRU (oldest caps at list head, newest
705 * at list tail).
706 */
707static void __touch_cap(struct ceph_cap *cap)
708{
709 struct ceph_mds_session *s = cap->session;
710
Sage Weila8599bd2009-10-06 11:31:12 -0700711 spin_lock(&s->s_cap_lock);
Sage Weil7c1332b2010-02-16 11:39:45 -0800712 if (s->s_cap_iterator == NULL) {
Sage Weil5dacf092009-12-21 20:40:34 -0800713 dout("__touch_cap %p cap %p mds%d\n", &cap->ci->vfs_inode, cap,
714 s->s_mds);
715 list_move_tail(&cap->session_caps, &s->s_caps);
716 } else {
717 dout("__touch_cap %p cap %p mds%d NOP, iterating over caps\n",
718 &cap->ci->vfs_inode, cap, s->s_mds);
719 }
Sage Weila8599bd2009-10-06 11:31:12 -0700720 spin_unlock(&s->s_cap_lock);
721}
722
723/*
724 * Check if we hold the given mask. If so, move the cap(s) to the
725 * front of their respective LRUs. (This is the preferred way for
726 * callers to check for caps they want.)
727 */
728int __ceph_caps_issued_mask(struct ceph_inode_info *ci, int mask, int touch)
729{
730 struct ceph_cap *cap;
731 struct rb_node *p;
732 int have = ci->i_snap_caps;
733
734 if ((have & mask) == mask) {
735 dout("__ceph_caps_issued_mask %p snap issued %s"
736 " (mask %s)\n", &ci->vfs_inode,
737 ceph_cap_string(have),
738 ceph_cap_string(mask));
739 return 1;
740 }
741
742 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
743 cap = rb_entry(p, struct ceph_cap, ci_node);
744 if (!__cap_is_valid(cap))
745 continue;
746 if ((cap->issued & mask) == mask) {
747 dout("__ceph_caps_issued_mask %p cap %p issued %s"
748 " (mask %s)\n", &ci->vfs_inode, cap,
749 ceph_cap_string(cap->issued),
750 ceph_cap_string(mask));
751 if (touch)
752 __touch_cap(cap);
753 return 1;
754 }
755
756 /* does a combination of caps satisfy mask? */
757 have |= cap->issued;
758 if ((have & mask) == mask) {
759 dout("__ceph_caps_issued_mask %p combo issued %s"
760 " (mask %s)\n", &ci->vfs_inode,
761 ceph_cap_string(cap->issued),
762 ceph_cap_string(mask));
763 if (touch) {
764 struct rb_node *q;
765
766 /* touch this + preceeding caps */
767 __touch_cap(cap);
768 for (q = rb_first(&ci->i_caps); q != p;
769 q = rb_next(q)) {
770 cap = rb_entry(q, struct ceph_cap,
771 ci_node);
772 if (!__cap_is_valid(cap))
773 continue;
774 __touch_cap(cap);
775 }
776 }
777 return 1;
778 }
779 }
780
781 return 0;
782}
783
784/*
785 * Return true if mask caps are currently being revoked by an MDS.
786 */
787int ceph_caps_revoking(struct ceph_inode_info *ci, int mask)
788{
789 struct inode *inode = &ci->vfs_inode;
790 struct ceph_cap *cap;
791 struct rb_node *p;
792 int ret = 0;
793
794 spin_lock(&inode->i_lock);
795 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
796 cap = rb_entry(p, struct ceph_cap, ci_node);
797 if (__cap_is_valid(cap) &&
798 (cap->implemented & ~cap->issued & mask)) {
799 ret = 1;
800 break;
801 }
802 }
803 spin_unlock(&inode->i_lock);
804 dout("ceph_caps_revoking %p %s = %d\n", inode,
805 ceph_cap_string(mask), ret);
806 return ret;
807}
808
809int __ceph_caps_used(struct ceph_inode_info *ci)
810{
811 int used = 0;
812 if (ci->i_pin_ref)
813 used |= CEPH_CAP_PIN;
814 if (ci->i_rd_ref)
815 used |= CEPH_CAP_FILE_RD;
816 if (ci->i_rdcache_ref || ci->i_rdcache_gen)
817 used |= CEPH_CAP_FILE_CACHE;
818 if (ci->i_wr_ref)
819 used |= CEPH_CAP_FILE_WR;
820 if (ci->i_wrbuffer_ref)
821 used |= CEPH_CAP_FILE_BUFFER;
822 return used;
823}
824
825/*
826 * wanted, by virtue of open file modes
827 */
828int __ceph_caps_file_wanted(struct ceph_inode_info *ci)
829{
830 int want = 0;
831 int mode;
Sage Weil33caad32010-05-26 14:31:27 -0700832 for (mode = 0; mode < CEPH_FILE_MODE_NUM; mode++)
Sage Weila8599bd2009-10-06 11:31:12 -0700833 if (ci->i_nr_by_mode[mode])
834 want |= ceph_caps_for_mode(mode);
835 return want;
836}
837
838/*
839 * Return caps we have registered with the MDS(s) as 'wanted'.
840 */
841int __ceph_caps_mds_wanted(struct ceph_inode_info *ci)
842{
843 struct ceph_cap *cap;
844 struct rb_node *p;
845 int mds_wanted = 0;
846
847 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
848 cap = rb_entry(p, struct ceph_cap, ci_node);
849 if (!__cap_is_valid(cap))
850 continue;
851 mds_wanted |= cap->mds_wanted;
852 }
853 return mds_wanted;
854}
855
856/*
857 * called under i_lock
858 */
859static int __ceph_is_any_caps(struct ceph_inode_info *ci)
860{
861 return !RB_EMPTY_ROOT(&ci->i_caps) || ci->i_cap_exporting_mds >= 0;
862}
863
864/*
Sage Weilf818a732010-05-11 20:56:31 -0700865 * Remove a cap. Take steps to deal with a racing iterate_session_caps.
866 *
Sage Weila6369742010-02-22 13:59:00 -0800867 * caller should hold i_lock.
868 * caller will not hold session s_mutex if called from destroy_inode.
Sage Weila8599bd2009-10-06 11:31:12 -0700869 */
Sage Weil7c1332b2010-02-16 11:39:45 -0800870void __ceph_remove_cap(struct ceph_cap *cap)
Sage Weila8599bd2009-10-06 11:31:12 -0700871{
872 struct ceph_mds_session *session = cap->session;
873 struct ceph_inode_info *ci = cap->ci;
Cheng Renquan640ef792010-03-26 17:40:33 +0800874 struct ceph_mds_client *mdsc =
875 &ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc;
Sage Weilf818a732010-05-11 20:56:31 -0700876 int removed = 0;
Sage Weila8599bd2009-10-06 11:31:12 -0700877
878 dout("__ceph_remove_cap %p from %p\n", cap, &ci->vfs_inode);
879
Sage Weil7c1332b2010-02-16 11:39:45 -0800880 /* remove from session list */
881 spin_lock(&session->s_cap_lock);
882 if (session->s_cap_iterator == cap) {
883 /* not yet, we are iterating over this very cap */
884 dout("__ceph_remove_cap delaying %p removal from session %p\n",
885 cap, cap->session);
886 } else {
887 list_del_init(&cap->session_caps);
888 session->s_nr_caps--;
889 cap->session = NULL;
Sage Weilf818a732010-05-11 20:56:31 -0700890 removed = 1;
Sage Weil7c1332b2010-02-16 11:39:45 -0800891 }
Sage Weilf818a732010-05-11 20:56:31 -0700892 /* protect backpointer with s_cap_lock: see iterate_session_caps */
893 cap->ci = NULL;
Sage Weil7c1332b2010-02-16 11:39:45 -0800894 spin_unlock(&session->s_cap_lock);
895
Sage Weilf818a732010-05-11 20:56:31 -0700896 /* remove from inode list */
897 rb_erase(&cap->ci_node, &ci->i_caps);
898 if (ci->i_auth_cap == cap)
899 ci->i_auth_cap = NULL;
900
901 if (removed)
Yehuda Sadeh37151662010-06-17 16:16:12 -0700902 ceph_put_cap(mdsc, cap);
Sage Weila8599bd2009-10-06 11:31:12 -0700903
904 if (!__ceph_is_any_caps(ci) && ci->i_snap_realm) {
905 struct ceph_snap_realm *realm = ci->i_snap_realm;
906 spin_lock(&realm->inodes_with_caps_lock);
907 list_del_init(&ci->i_snap_realm_item);
908 ci->i_snap_realm_counter++;
909 ci->i_snap_realm = NULL;
910 spin_unlock(&realm->inodes_with_caps_lock);
911 ceph_put_snap_realm(mdsc, realm);
912 }
913 if (!__ceph_is_any_real_caps(ci))
914 __cap_delay_cancel(mdsc, ci);
915}
916
917/*
918 * Build and send a cap message to the given MDS.
919 *
920 * Caller should be holding s_mutex.
921 */
922static int send_cap_msg(struct ceph_mds_session *session,
923 u64 ino, u64 cid, int op,
924 int caps, int wanted, int dirty,
925 u32 seq, u64 flush_tid, u32 issue_seq, u32 mseq,
926 u64 size, u64 max_size,
927 struct timespec *mtime, struct timespec *atime,
928 u64 time_warp_seq,
929 uid_t uid, gid_t gid, mode_t mode,
930 u64 xattr_version,
931 struct ceph_buffer *xattrs_buf,
932 u64 follows)
933{
934 struct ceph_mds_caps *fc;
935 struct ceph_msg *msg;
936
937 dout("send_cap_msg %s %llx %llx caps %s wanted %s dirty %s"
938 " seq %u/%u mseq %u follows %lld size %llu/%llu"
939 " xattr_ver %llu xattr_len %d\n", ceph_cap_op_name(op),
940 cid, ino, ceph_cap_string(caps), ceph_cap_string(wanted),
941 ceph_cap_string(dirty),
942 seq, issue_seq, mseq, follows, size, max_size,
943 xattr_version, xattrs_buf ? (int)xattrs_buf->vec.iov_len : 0);
944
Yehuda Sadeh34d23762010-04-06 14:33:58 -0700945 msg = ceph_msg_new(CEPH_MSG_CLIENT_CAPS, sizeof(*fc), GFP_NOFS);
Sage Weila79832f2010-04-01 16:06:19 -0700946 if (!msg)
947 return -ENOMEM;
Sage Weila8599bd2009-10-06 11:31:12 -0700948
Sage Weil6df058c2009-12-22 11:24:33 -0800949 msg->hdr.tid = cpu_to_le64(flush_tid);
Sage Weila8599bd2009-10-06 11:31:12 -0700950
Sage Weil6df058c2009-12-22 11:24:33 -0800951 fc = msg->front.iov_base;
Sage Weila8599bd2009-10-06 11:31:12 -0700952 memset(fc, 0, sizeof(*fc));
953
954 fc->cap_id = cpu_to_le64(cid);
955 fc->op = cpu_to_le32(op);
956 fc->seq = cpu_to_le32(seq);
Sage Weila8599bd2009-10-06 11:31:12 -0700957 fc->issue_seq = cpu_to_le32(issue_seq);
958 fc->migrate_seq = cpu_to_le32(mseq);
959 fc->caps = cpu_to_le32(caps);
960 fc->wanted = cpu_to_le32(wanted);
961 fc->dirty = cpu_to_le32(dirty);
962 fc->ino = cpu_to_le64(ino);
963 fc->snap_follows = cpu_to_le64(follows);
964
965 fc->size = cpu_to_le64(size);
966 fc->max_size = cpu_to_le64(max_size);
967 if (mtime)
968 ceph_encode_timespec(&fc->mtime, mtime);
969 if (atime)
970 ceph_encode_timespec(&fc->atime, atime);
971 fc->time_warp_seq = cpu_to_le32(time_warp_seq);
972
973 fc->uid = cpu_to_le32(uid);
974 fc->gid = cpu_to_le32(gid);
975 fc->mode = cpu_to_le32(mode);
976
977 fc->xattr_version = cpu_to_le64(xattr_version);
978 if (xattrs_buf) {
979 msg->middle = ceph_buffer_get(xattrs_buf);
980 fc->xattr_len = cpu_to_le32(xattrs_buf->vec.iov_len);
981 msg->hdr.middle_len = cpu_to_le32(xattrs_buf->vec.iov_len);
982 }
983
984 ceph_con_send(&session->s_con, msg);
985 return 0;
986}
987
Sage Weil3d7ded42010-06-09 16:47:10 -0700988static void __queue_cap_release(struct ceph_mds_session *session,
989 u64 ino, u64 cap_id, u32 migrate_seq,
990 u32 issue_seq)
991{
992 struct ceph_msg *msg;
993 struct ceph_mds_cap_release *head;
994 struct ceph_mds_cap_item *item;
995
996 spin_lock(&session->s_cap_lock);
997 BUG_ON(!session->s_num_cap_releases);
998 msg = list_first_entry(&session->s_cap_releases,
999 struct ceph_msg, list_head);
1000
1001 dout(" adding %llx release to mds%d msg %p (%d left)\n",
1002 ino, session->s_mds, msg, session->s_num_cap_releases);
1003
1004 BUG_ON(msg->front.iov_len + sizeof(*item) > PAGE_CACHE_SIZE);
1005 head = msg->front.iov_base;
1006 head->num = cpu_to_le32(le32_to_cpu(head->num) + 1);
1007 item = msg->front.iov_base + msg->front.iov_len;
1008 item->ino = cpu_to_le64(ino);
1009 item->cap_id = cpu_to_le64(cap_id);
1010 item->migrate_seq = cpu_to_le32(migrate_seq);
1011 item->seq = cpu_to_le32(issue_seq);
1012
1013 session->s_num_cap_releases--;
1014
1015 msg->front.iov_len += sizeof(*item);
1016 if (le32_to_cpu(head->num) == CEPH_CAPS_PER_RELEASE) {
1017 dout(" release msg %p full\n", msg);
1018 list_move_tail(&msg->list_head, &session->s_cap_releases_done);
1019 } else {
1020 dout(" release msg %p at %d/%d (%d)\n", msg,
1021 (int)le32_to_cpu(head->num),
1022 (int)CEPH_CAPS_PER_RELEASE,
1023 (int)msg->front.iov_len);
1024 }
1025 spin_unlock(&session->s_cap_lock);
1026}
1027
Sage Weila8599bd2009-10-06 11:31:12 -07001028/*
Sage Weila6369742010-02-22 13:59:00 -08001029 * Queue cap releases when an inode is dropped from our cache. Since
1030 * inode is about to be destroyed, there is no need for i_lock.
Sage Weila8599bd2009-10-06 11:31:12 -07001031 */
1032void ceph_queue_caps_release(struct inode *inode)
1033{
1034 struct ceph_inode_info *ci = ceph_inode(inode);
1035 struct rb_node *p;
1036
Sage Weila8599bd2009-10-06 11:31:12 -07001037 p = rb_first(&ci->i_caps);
1038 while (p) {
1039 struct ceph_cap *cap = rb_entry(p, struct ceph_cap, ci_node);
1040 struct ceph_mds_session *session = cap->session;
Sage Weila8599bd2009-10-06 11:31:12 -07001041
Sage Weil3d7ded42010-06-09 16:47:10 -07001042 __queue_cap_release(session, ceph_ino(inode), cap->cap_id,
1043 cap->mseq, cap->issue_seq);
Sage Weila8599bd2009-10-06 11:31:12 -07001044 p = rb_next(p);
Sage Weil7c1332b2010-02-16 11:39:45 -08001045 __ceph_remove_cap(cap);
Sage Weila8599bd2009-10-06 11:31:12 -07001046 }
Sage Weila8599bd2009-10-06 11:31:12 -07001047}
1048
1049/*
1050 * Send a cap msg on the given inode. Update our caps state, then
1051 * drop i_lock and send the message.
1052 *
1053 * Make note of max_size reported/requested from mds, revoked caps
1054 * that have now been implemented.
1055 *
1056 * Make half-hearted attempt ot to invalidate page cache if we are
1057 * dropping RDCACHE. Note that this will leave behind locked pages
1058 * that we'll then need to deal with elsewhere.
1059 *
1060 * Return non-zero if delayed release, or we experienced an error
1061 * such that the caller should requeue + retry later.
1062 *
1063 * called with i_lock, then drops it.
1064 * caller should hold snap_rwsem (read), s_mutex.
1065 */
1066static int __send_cap(struct ceph_mds_client *mdsc, struct ceph_cap *cap,
1067 int op, int used, int want, int retain, int flushing,
1068 unsigned *pflush_tid)
1069 __releases(cap->ci->vfs_inode->i_lock)
1070{
1071 struct ceph_inode_info *ci = cap->ci;
1072 struct inode *inode = &ci->vfs_inode;
1073 u64 cap_id = cap->cap_id;
Sage Weil68c28322010-02-09 13:41:47 -08001074 int held, revoking, dropping, keep;
Sage Weila8599bd2009-10-06 11:31:12 -07001075 u64 seq, issue_seq, mseq, time_warp_seq, follows;
1076 u64 size, max_size;
1077 struct timespec mtime, atime;
1078 int wake = 0;
1079 mode_t mode;
1080 uid_t uid;
1081 gid_t gid;
1082 struct ceph_mds_session *session;
1083 u64 xattr_version = 0;
1084 int delayed = 0;
1085 u64 flush_tid = 0;
1086 int i;
1087 int ret;
1088
Sage Weil68c28322010-02-09 13:41:47 -08001089 held = cap->issued | cap->implemented;
1090 revoking = cap->implemented & ~cap->issued;
1091 retain &= ~revoking;
1092 dropping = cap->issued & ~retain;
1093
Sage Weila8599bd2009-10-06 11:31:12 -07001094 dout("__send_cap %p cap %p session %p %s -> %s (revoking %s)\n",
1095 inode, cap, cap->session,
1096 ceph_cap_string(held), ceph_cap_string(held & retain),
1097 ceph_cap_string(revoking));
1098 BUG_ON((retain & CEPH_CAP_PIN) == 0);
1099
1100 session = cap->session;
1101
1102 /* don't release wanted unless we've waited a bit. */
1103 if ((ci->i_ceph_flags & CEPH_I_NODELAY) == 0 &&
1104 time_before(jiffies, ci->i_hold_caps_min)) {
1105 dout(" delaying issued %s -> %s, wanted %s -> %s on send\n",
1106 ceph_cap_string(cap->issued),
1107 ceph_cap_string(cap->issued & retain),
1108 ceph_cap_string(cap->mds_wanted),
1109 ceph_cap_string(want));
1110 want |= cap->mds_wanted;
1111 retain |= cap->issued;
1112 delayed = 1;
1113 }
1114 ci->i_ceph_flags &= ~(CEPH_I_NODELAY | CEPH_I_FLUSH);
1115
1116 cap->issued &= retain; /* drop bits we don't want */
1117 if (cap->implemented & ~cap->issued) {
1118 /*
1119 * Wake up any waiters on wanted -> needed transition.
1120 * This is due to the weird transition from buffered
1121 * to sync IO... we need to flush dirty pages _before_
1122 * allowing sync writes to avoid reordering.
1123 */
1124 wake = 1;
1125 }
1126 cap->implemented &= cap->issued | used;
1127 cap->mds_wanted = want;
1128
1129 if (flushing) {
1130 /*
1131 * assign a tid for flush operations so we can avoid
1132 * flush1 -> dirty1 -> flush2 -> flushack1 -> mark
1133 * clean type races. track latest tid for every bit
1134 * so we can handle flush AxFw, flush Fw, and have the
1135 * first ack clean Ax.
1136 */
1137 flush_tid = ++ci->i_cap_flush_last_tid;
1138 if (pflush_tid)
1139 *pflush_tid = flush_tid;
1140 dout(" cap_flush_tid %d\n", (int)flush_tid);
1141 for (i = 0; i < CEPH_CAP_BITS; i++)
1142 if (flushing & (1 << i))
1143 ci->i_cap_flush_tid[i] = flush_tid;
1144 }
1145
1146 keep = cap->implemented;
1147 seq = cap->seq;
1148 issue_seq = cap->issue_seq;
1149 mseq = cap->mseq;
1150 size = inode->i_size;
1151 ci->i_reported_size = size;
1152 max_size = ci->i_wanted_max_size;
1153 ci->i_requested_max_size = max_size;
1154 mtime = inode->i_mtime;
1155 atime = inode->i_atime;
1156 time_warp_seq = ci->i_time_warp_seq;
1157 follows = ci->i_snap_realm->cached_context->seq;
1158 uid = inode->i_uid;
1159 gid = inode->i_gid;
1160 mode = inode->i_mode;
1161
1162 if (dropping & CEPH_CAP_XATTR_EXCL) {
1163 __ceph_build_xattrs_blob(ci);
1164 xattr_version = ci->i_xattrs.version + 1;
1165 }
1166
1167 spin_unlock(&inode->i_lock);
1168
Sage Weila8599bd2009-10-06 11:31:12 -07001169 ret = send_cap_msg(session, ceph_vino(inode).ino, cap_id,
1170 op, keep, want, flushing, seq, flush_tid, issue_seq, mseq,
1171 size, max_size, &mtime, &atime, time_warp_seq,
1172 uid, gid, mode,
1173 xattr_version,
1174 (flushing & CEPH_CAP_XATTR_EXCL) ? ci->i_xattrs.blob : NULL,
1175 follows);
1176 if (ret < 0) {
1177 dout("error sending cap msg, must requeue %p\n", inode);
1178 delayed = 1;
1179 }
1180
1181 if (wake)
Yehuda Sadeh03066f22010-07-27 13:11:08 -07001182 wake_up_all(&ci->i_cap_wq);
Sage Weila8599bd2009-10-06 11:31:12 -07001183
1184 return delayed;
1185}
1186
1187/*
1188 * When a snapshot is taken, clients accumulate dirty metadata on
1189 * inodes with capabilities in ceph_cap_snaps to describe the file
1190 * state at the time the snapshot was taken. This must be flushed
1191 * asynchronously back to the MDS once sync writes complete and dirty
1192 * data is written out.
1193 *
1194 * Called under i_lock. Takes s_mutex as needed.
1195 */
1196void __ceph_flush_snaps(struct ceph_inode_info *ci,
1197 struct ceph_mds_session **psession)
Yehuda Sadehcd84db62010-06-11 16:58:48 -07001198 __releases(ci->vfs_inode->i_lock)
1199 __acquires(ci->vfs_inode->i_lock)
Sage Weila8599bd2009-10-06 11:31:12 -07001200{
1201 struct inode *inode = &ci->vfs_inode;
1202 int mds;
1203 struct ceph_cap_snap *capsnap;
1204 u32 mseq;
1205 struct ceph_mds_client *mdsc = &ceph_inode_to_client(inode)->mdsc;
1206 struct ceph_mds_session *session = NULL; /* if session != NULL, we hold
1207 session->s_mutex */
1208 u64 next_follows = 0; /* keep track of how far we've gotten through the
1209 i_cap_snaps list, and skip these entries next time
1210 around to avoid an infinite loop */
1211
1212 if (psession)
1213 session = *psession;
1214
1215 dout("__flush_snaps %p\n", inode);
1216retry:
1217 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
1218 /* avoid an infiniute loop after retry */
1219 if (capsnap->follows < next_follows)
1220 continue;
1221 /*
1222 * we need to wait for sync writes to complete and for dirty
1223 * pages to be written out.
1224 */
1225 if (capsnap->dirty_pages || capsnap->writing)
1226 continue;
1227
Sage Weil819ccbf2010-04-01 09:33:46 -07001228 /*
1229 * if cap writeback already occurred, we should have dropped
1230 * the capsnap in ceph_put_wrbuffer_cap_refs.
1231 */
1232 BUG_ON(capsnap->dirty == 0);
1233
Sage Weila8599bd2009-10-06 11:31:12 -07001234 /* pick mds, take s_mutex */
Sage Weilca81f3f2010-06-10 14:21:36 -07001235 if (ci->i_auth_cap == NULL) {
1236 dout("no auth cap (migrating?), doing nothing\n");
1237 goto out;
1238 }
1239 mds = ci->i_auth_cap->session->s_mds;
1240 mseq = ci->i_auth_cap->mseq;
1241
Sage Weila8599bd2009-10-06 11:31:12 -07001242 if (session && session->s_mds != mds) {
1243 dout("oops, wrong session %p mutex\n", session);
1244 mutex_unlock(&session->s_mutex);
1245 ceph_put_mds_session(session);
1246 session = NULL;
1247 }
1248 if (!session) {
1249 spin_unlock(&inode->i_lock);
1250 mutex_lock(&mdsc->mutex);
1251 session = __ceph_lookup_mds_session(mdsc, mds);
1252 mutex_unlock(&mdsc->mutex);
1253 if (session) {
1254 dout("inverting session/ino locks on %p\n",
1255 session);
1256 mutex_lock(&session->s_mutex);
1257 }
1258 /*
1259 * if session == NULL, we raced against a cap
Sage Weilca81f3f2010-06-10 14:21:36 -07001260 * deletion or migration. retry, and we'll
1261 * get a better @mds value next time.
Sage Weila8599bd2009-10-06 11:31:12 -07001262 */
1263 spin_lock(&inode->i_lock);
1264 goto retry;
1265 }
1266
1267 capsnap->flush_tid = ++ci->i_cap_flush_last_tid;
1268 atomic_inc(&capsnap->nref);
1269 if (!list_empty(&capsnap->flushing_item))
1270 list_del_init(&capsnap->flushing_item);
1271 list_add_tail(&capsnap->flushing_item,
1272 &session->s_cap_snaps_flushing);
1273 spin_unlock(&inode->i_lock);
1274
1275 dout("flush_snaps %p cap_snap %p follows %lld size %llu\n",
1276 inode, capsnap, next_follows, capsnap->size);
1277 send_cap_msg(session, ceph_vino(inode).ino, 0,
1278 CEPH_CAP_OP_FLUSHSNAP, capsnap->issued, 0,
1279 capsnap->dirty, 0, capsnap->flush_tid, 0, mseq,
1280 capsnap->size, 0,
1281 &capsnap->mtime, &capsnap->atime,
1282 capsnap->time_warp_seq,
1283 capsnap->uid, capsnap->gid, capsnap->mode,
1284 0, NULL,
1285 capsnap->follows);
1286
1287 next_follows = capsnap->follows + 1;
1288 ceph_put_cap_snap(capsnap);
1289
1290 spin_lock(&inode->i_lock);
1291 goto retry;
1292 }
1293
1294 /* we flushed them all; remove this inode from the queue */
1295 spin_lock(&mdsc->snap_flush_lock);
1296 list_del_init(&ci->i_snap_flush_item);
1297 spin_unlock(&mdsc->snap_flush_lock);
1298
Sage Weilca81f3f2010-06-10 14:21:36 -07001299out:
Sage Weila8599bd2009-10-06 11:31:12 -07001300 if (psession)
1301 *psession = session;
1302 else if (session) {
1303 mutex_unlock(&session->s_mutex);
1304 ceph_put_mds_session(session);
1305 }
1306}
1307
1308static void ceph_flush_snaps(struct ceph_inode_info *ci)
1309{
1310 struct inode *inode = &ci->vfs_inode;
1311
1312 spin_lock(&inode->i_lock);
1313 __ceph_flush_snaps(ci, NULL);
1314 spin_unlock(&inode->i_lock);
1315}
1316
1317/*
Sage Weil76e3b392009-10-15 18:13:53 -07001318 * Mark caps dirty. If inode is newly dirty, add to the global dirty
1319 * list.
1320 */
1321void __ceph_mark_dirty_caps(struct ceph_inode_info *ci, int mask)
1322{
Cheng Renquan640ef792010-03-26 17:40:33 +08001323 struct ceph_mds_client *mdsc =
1324 &ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc;
Sage Weil76e3b392009-10-15 18:13:53 -07001325 struct inode *inode = &ci->vfs_inode;
1326 int was = ci->i_dirty_caps;
1327 int dirty = 0;
1328
1329 dout("__mark_dirty_caps %p %s dirty %s -> %s\n", &ci->vfs_inode,
1330 ceph_cap_string(mask), ceph_cap_string(was),
1331 ceph_cap_string(was | mask));
1332 ci->i_dirty_caps |= mask;
1333 if (was == 0) {
1334 dout(" inode %p now dirty\n", &ci->vfs_inode);
1335 BUG_ON(!list_empty(&ci->i_dirty_item));
1336 spin_lock(&mdsc->cap_dirty_lock);
1337 list_add(&ci->i_dirty_item, &mdsc->cap_dirty);
1338 spin_unlock(&mdsc->cap_dirty_lock);
1339 if (ci->i_flushing_caps == 0) {
1340 igrab(inode);
1341 dirty |= I_DIRTY_SYNC;
1342 }
1343 }
1344 BUG_ON(list_empty(&ci->i_dirty_item));
1345 if (((was | ci->i_flushing_caps) & CEPH_CAP_FILE_BUFFER) &&
1346 (mask & CEPH_CAP_FILE_BUFFER))
1347 dirty |= I_DIRTY_DATASYNC;
1348 if (dirty)
1349 __mark_inode_dirty(inode, dirty);
1350 __cap_delay_requeue(mdsc, ci);
1351}
1352
1353/*
Sage Weila8599bd2009-10-06 11:31:12 -07001354 * Add dirty inode to the flushing list. Assigned a seq number so we
1355 * can wait for caps to flush without starving.
Sage Weilcdc35f92009-10-14 14:24:19 -07001356 *
1357 * Called under i_lock.
Sage Weila8599bd2009-10-06 11:31:12 -07001358 */
Sage Weilcdc35f92009-10-14 14:24:19 -07001359static int __mark_caps_flushing(struct inode *inode,
Sage Weila8599bd2009-10-06 11:31:12 -07001360 struct ceph_mds_session *session)
1361{
Cheng Renquan640ef792010-03-26 17:40:33 +08001362 struct ceph_mds_client *mdsc = &ceph_sb_to_client(inode->i_sb)->mdsc;
Sage Weila8599bd2009-10-06 11:31:12 -07001363 struct ceph_inode_info *ci = ceph_inode(inode);
Sage Weilcdc35f92009-10-14 14:24:19 -07001364 int flushing;
Sage Weil50b885b2009-12-01 14:12:07 -08001365
Sage Weilcdc35f92009-10-14 14:24:19 -07001366 BUG_ON(ci->i_dirty_caps == 0);
Sage Weila8599bd2009-10-06 11:31:12 -07001367 BUG_ON(list_empty(&ci->i_dirty_item));
Sage Weilcdc35f92009-10-14 14:24:19 -07001368
1369 flushing = ci->i_dirty_caps;
1370 dout("__mark_caps_flushing flushing %s, flushing_caps %s -> %s\n",
1371 ceph_cap_string(flushing),
1372 ceph_cap_string(ci->i_flushing_caps),
1373 ceph_cap_string(ci->i_flushing_caps | flushing));
1374 ci->i_flushing_caps |= flushing;
1375 ci->i_dirty_caps = 0;
Sage Weilafcdaea2009-10-14 14:27:38 -07001376 dout(" inode %p now !dirty\n", inode);
Sage Weilcdc35f92009-10-14 14:24:19 -07001377
Sage Weila8599bd2009-10-06 11:31:12 -07001378 spin_lock(&mdsc->cap_dirty_lock);
Sage Weilafcdaea2009-10-14 14:27:38 -07001379 list_del_init(&ci->i_dirty_item);
1380
1381 ci->i_cap_flush_seq = ++mdsc->cap_flush_seq;
Sage Weila8599bd2009-10-06 11:31:12 -07001382 if (list_empty(&ci->i_flushing_item)) {
1383 list_add_tail(&ci->i_flushing_item, &session->s_cap_flushing);
1384 mdsc->num_cap_flushing++;
Sage Weilafcdaea2009-10-14 14:27:38 -07001385 dout(" inode %p now flushing seq %lld\n", inode,
1386 ci->i_cap_flush_seq);
1387 } else {
1388 list_move_tail(&ci->i_flushing_item, &session->s_cap_flushing);
1389 dout(" inode %p now flushing (more) seq %lld\n", inode,
Sage Weila8599bd2009-10-06 11:31:12 -07001390 ci->i_cap_flush_seq);
1391 }
1392 spin_unlock(&mdsc->cap_dirty_lock);
Sage Weilcdc35f92009-10-14 14:24:19 -07001393
1394 return flushing;
Sage Weila8599bd2009-10-06 11:31:12 -07001395}
1396
1397/*
Sage Weil5ecad6f2010-02-17 10:43:37 -08001398 * try to invalidate mapping pages without blocking.
1399 */
1400static int mapping_is_empty(struct address_space *mapping)
1401{
1402 struct page *page = find_get_page(mapping, 0);
1403
1404 if (!page)
1405 return 1;
1406
1407 put_page(page);
1408 return 0;
1409}
1410
1411static int try_nonblocking_invalidate(struct inode *inode)
1412{
1413 struct ceph_inode_info *ci = ceph_inode(inode);
1414 u32 invalidating_gen = ci->i_rdcache_gen;
1415
1416 spin_unlock(&inode->i_lock);
1417 invalidate_mapping_pages(&inode->i_data, 0, -1);
1418 spin_lock(&inode->i_lock);
1419
1420 if (mapping_is_empty(&inode->i_data) &&
1421 invalidating_gen == ci->i_rdcache_gen) {
1422 /* success. */
1423 dout("try_nonblocking_invalidate %p success\n", inode);
1424 ci->i_rdcache_gen = 0;
1425 ci->i_rdcache_revoking = 0;
1426 return 0;
1427 }
1428 dout("try_nonblocking_invalidate %p failed\n", inode);
1429 return -1;
1430}
1431
1432/*
Sage Weila8599bd2009-10-06 11:31:12 -07001433 * Swiss army knife function to examine currently used and wanted
1434 * versus held caps. Release, flush, ack revoked caps to mds as
1435 * appropriate.
1436 *
1437 * CHECK_CAPS_NODELAY - caller is delayed work and we should not delay
1438 * cap release further.
1439 * CHECK_CAPS_AUTHONLY - we should only check the auth cap
1440 * CHECK_CAPS_FLUSH - we should flush any dirty caps immediately, without
1441 * further delay.
1442 */
1443void ceph_check_caps(struct ceph_inode_info *ci, int flags,
1444 struct ceph_mds_session *session)
1445{
1446 struct ceph_client *client = ceph_inode_to_client(&ci->vfs_inode);
1447 struct ceph_mds_client *mdsc = &client->mdsc;
1448 struct inode *inode = &ci->vfs_inode;
1449 struct ceph_cap *cap;
1450 int file_wanted, used;
1451 int took_snap_rwsem = 0; /* true if mdsc->snap_rwsem held */
Sage Weilcbd03632010-02-09 13:41:18 -08001452 int issued, implemented, want, retain, revoking, flushing = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07001453 int mds = -1; /* keep track of how far we've gone through i_caps list
1454 to avoid an infinite loop on retry */
1455 struct rb_node *p;
1456 int tried_invalidate = 0;
1457 int delayed = 0, sent = 0, force_requeue = 0, num;
Sage Weilcbd03632010-02-09 13:41:18 -08001458 int queue_invalidate = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07001459 int is_delayed = flags & CHECK_CAPS_NODELAY;
1460
1461 /* if we are unmounting, flush any unused caps immediately. */
1462 if (mdsc->stopping)
1463 is_delayed = 1;
1464
1465 spin_lock(&inode->i_lock);
1466
1467 if (ci->i_ceph_flags & CEPH_I_FLUSH)
1468 flags |= CHECK_CAPS_FLUSH;
1469
1470 /* flush snaps first time around only */
1471 if (!list_empty(&ci->i_cap_snaps))
1472 __ceph_flush_snaps(ci, &session);
1473 goto retry_locked;
1474retry:
1475 spin_lock(&inode->i_lock);
1476retry_locked:
1477 file_wanted = __ceph_caps_file_wanted(ci);
1478 used = __ceph_caps_used(ci);
1479 want = file_wanted | used;
Sage Weilcbd03632010-02-09 13:41:18 -08001480 issued = __ceph_caps_issued(ci, &implemented);
1481 revoking = implemented & ~issued;
Sage Weila8599bd2009-10-06 11:31:12 -07001482
1483 retain = want | CEPH_CAP_PIN;
1484 if (!mdsc->stopping && inode->i_nlink > 0) {
1485 if (want) {
1486 retain |= CEPH_CAP_ANY; /* be greedy */
1487 } else {
1488 retain |= CEPH_CAP_ANY_SHARED;
1489 /*
1490 * keep RD only if we didn't have the file open RW,
1491 * because then the mds would revoke it anyway to
1492 * journal max_size=0.
1493 */
1494 if (ci->i_max_size == 0)
1495 retain |= CEPH_CAP_ANY_RD;
1496 }
1497 }
1498
1499 dout("check_caps %p file_want %s used %s dirty %s flushing %s"
Sage Weilcbd03632010-02-09 13:41:18 -08001500 " issued %s revoking %s retain %s %s%s%s\n", inode,
Sage Weila8599bd2009-10-06 11:31:12 -07001501 ceph_cap_string(file_wanted),
1502 ceph_cap_string(used), ceph_cap_string(ci->i_dirty_caps),
1503 ceph_cap_string(ci->i_flushing_caps),
Sage Weilcbd03632010-02-09 13:41:18 -08001504 ceph_cap_string(issued), ceph_cap_string(revoking),
Sage Weila8599bd2009-10-06 11:31:12 -07001505 ceph_cap_string(retain),
1506 (flags & CHECK_CAPS_AUTHONLY) ? " AUTHONLY" : "",
1507 (flags & CHECK_CAPS_NODELAY) ? " NODELAY" : "",
1508 (flags & CHECK_CAPS_FLUSH) ? " FLUSH" : "");
1509
1510 /*
1511 * If we no longer need to hold onto old our caps, and we may
1512 * have cached pages, but don't want them, then try to invalidate.
1513 * If we fail, it's because pages are locked.... try again later.
1514 */
1515 if ((!is_delayed || mdsc->stopping) &&
1516 ci->i_wrbuffer_ref == 0 && /* no dirty pages... */
1517 ci->i_rdcache_gen && /* may have cached pages */
Sage Weilcbd03632010-02-09 13:41:18 -08001518 (file_wanted == 0 || /* no open files */
Sage Weil29625072010-05-27 10:40:43 -07001519 (revoking & (CEPH_CAP_FILE_CACHE|
1520 CEPH_CAP_FILE_LAZYIO))) && /* or revoking cache */
Sage Weila8599bd2009-10-06 11:31:12 -07001521 !tried_invalidate) {
Sage Weila8599bd2009-10-06 11:31:12 -07001522 dout("check_caps trying to invalidate on %p\n", inode);
Sage Weil5ecad6f2010-02-17 10:43:37 -08001523 if (try_nonblocking_invalidate(inode) < 0) {
Sage Weil29625072010-05-27 10:40:43 -07001524 if (revoking & (CEPH_CAP_FILE_CACHE|
1525 CEPH_CAP_FILE_LAZYIO)) {
Sage Weil5ecad6f2010-02-17 10:43:37 -08001526 dout("check_caps queuing invalidate\n");
1527 queue_invalidate = 1;
1528 ci->i_rdcache_revoking = ci->i_rdcache_gen;
1529 } else {
1530 dout("check_caps failed to invalidate pages\n");
1531 /* we failed to invalidate pages. check these
1532 caps again later. */
1533 force_requeue = 1;
1534 __cap_set_timeouts(mdsc, ci);
1535 }
Sage Weila8599bd2009-10-06 11:31:12 -07001536 }
1537 tried_invalidate = 1;
1538 goto retry_locked;
1539 }
1540
1541 num = 0;
1542 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
1543 cap = rb_entry(p, struct ceph_cap, ci_node);
1544 num++;
1545
1546 /* avoid looping forever */
1547 if (mds >= cap->mds ||
1548 ((flags & CHECK_CAPS_AUTHONLY) && cap != ci->i_auth_cap))
1549 continue;
1550
1551 /* NOTE: no side-effects allowed, until we take s_mutex */
1552
1553 revoking = cap->implemented & ~cap->issued;
1554 if (revoking)
Sage Weilcbd03632010-02-09 13:41:18 -08001555 dout(" mds%d revoking %s\n", cap->mds,
Sage Weila8599bd2009-10-06 11:31:12 -07001556 ceph_cap_string(revoking));
1557
1558 if (cap == ci->i_auth_cap &&
1559 (cap->issued & CEPH_CAP_FILE_WR)) {
1560 /* request larger max_size from MDS? */
1561 if (ci->i_wanted_max_size > ci->i_max_size &&
1562 ci->i_wanted_max_size > ci->i_requested_max_size) {
1563 dout("requesting new max_size\n");
1564 goto ack;
1565 }
1566
1567 /* approaching file_max? */
1568 if ((inode->i_size << 1) >= ci->i_max_size &&
1569 (ci->i_reported_size << 1) < ci->i_max_size) {
1570 dout("i_size approaching max_size\n");
1571 goto ack;
1572 }
1573 }
1574 /* flush anything dirty? */
1575 if (cap == ci->i_auth_cap && (flags & CHECK_CAPS_FLUSH) &&
1576 ci->i_dirty_caps) {
1577 dout("flushing dirty caps\n");
1578 goto ack;
1579 }
1580
1581 /* completed revocation? going down and there are no caps? */
1582 if (revoking && (revoking & used) == 0) {
1583 dout("completed revocation of %s\n",
1584 ceph_cap_string(cap->implemented & ~cap->issued));
1585 goto ack;
1586 }
1587
1588 /* want more caps from mds? */
1589 if (want & ~(cap->mds_wanted | cap->issued))
1590 goto ack;
1591
1592 /* things we might delay */
1593 if ((cap->issued & ~retain) == 0 &&
1594 cap->mds_wanted == want)
1595 continue; /* nope, all good */
1596
1597 if (is_delayed)
1598 goto ack;
1599
1600 /* delay? */
1601 if ((ci->i_ceph_flags & CEPH_I_NODELAY) == 0 &&
1602 time_before(jiffies, ci->i_hold_caps_max)) {
1603 dout(" delaying issued %s -> %s, wanted %s -> %s\n",
1604 ceph_cap_string(cap->issued),
1605 ceph_cap_string(cap->issued & retain),
1606 ceph_cap_string(cap->mds_wanted),
1607 ceph_cap_string(want));
1608 delayed++;
1609 continue;
1610 }
1611
1612ack:
Sage Weile9964c12010-03-01 15:16:56 -08001613 if (ci->i_ceph_flags & CEPH_I_NOFLUSH) {
1614 dout(" skipping %p I_NOFLUSH set\n", inode);
1615 continue;
1616 }
1617
Sage Weila8599bd2009-10-06 11:31:12 -07001618 if (session && session != cap->session) {
1619 dout("oops, wrong session %p mutex\n", session);
1620 mutex_unlock(&session->s_mutex);
1621 session = NULL;
1622 }
1623 if (!session) {
1624 session = cap->session;
1625 if (mutex_trylock(&session->s_mutex) == 0) {
1626 dout("inverting session/ino locks on %p\n",
1627 session);
1628 spin_unlock(&inode->i_lock);
1629 if (took_snap_rwsem) {
1630 up_read(&mdsc->snap_rwsem);
1631 took_snap_rwsem = 0;
1632 }
1633 mutex_lock(&session->s_mutex);
1634 goto retry;
1635 }
1636 }
1637 /* take snap_rwsem after session mutex */
1638 if (!took_snap_rwsem) {
1639 if (down_read_trylock(&mdsc->snap_rwsem) == 0) {
1640 dout("inverting snap/in locks on %p\n",
1641 inode);
1642 spin_unlock(&inode->i_lock);
1643 down_read(&mdsc->snap_rwsem);
1644 took_snap_rwsem = 1;
1645 goto retry;
1646 }
1647 took_snap_rwsem = 1;
1648 }
1649
Sage Weilcdc35f92009-10-14 14:24:19 -07001650 if (cap == ci->i_auth_cap && ci->i_dirty_caps)
1651 flushing = __mark_caps_flushing(inode, session);
Sage Weila8599bd2009-10-06 11:31:12 -07001652
1653 mds = cap->mds; /* remember mds, so we don't repeat */
1654 sent++;
1655
1656 /* __send_cap drops i_lock */
1657 delayed += __send_cap(mdsc, cap, CEPH_CAP_OP_UPDATE, used, want,
1658 retain, flushing, NULL);
1659 goto retry; /* retake i_lock and restart our cap scan. */
1660 }
1661
1662 /*
1663 * Reschedule delayed caps release if we delayed anything,
1664 * otherwise cancel.
1665 */
1666 if (delayed && is_delayed)
1667 force_requeue = 1; /* __send_cap delayed release; requeue */
1668 if (!delayed && !is_delayed)
1669 __cap_delay_cancel(mdsc, ci);
1670 else if (!is_delayed || force_requeue)
1671 __cap_delay_requeue(mdsc, ci);
1672
1673 spin_unlock(&inode->i_lock);
1674
Sage Weilcbd03632010-02-09 13:41:18 -08001675 if (queue_invalidate)
Sage Weil3c6f6b72010-02-09 15:24:44 -08001676 ceph_queue_invalidate(inode);
Sage Weilcbd03632010-02-09 13:41:18 -08001677
Sage Weilcdc2ce02010-03-16 13:39:28 -07001678 if (session)
Sage Weila8599bd2009-10-06 11:31:12 -07001679 mutex_unlock(&session->s_mutex);
1680 if (took_snap_rwsem)
1681 up_read(&mdsc->snap_rwsem);
1682}
1683
1684/*
Sage Weila8599bd2009-10-06 11:31:12 -07001685 * Try to flush dirty caps back to the auth mds.
1686 */
1687static int try_flush_caps(struct inode *inode, struct ceph_mds_session *session,
1688 unsigned *flush_tid)
1689{
Cheng Renquan640ef792010-03-26 17:40:33 +08001690 struct ceph_mds_client *mdsc = &ceph_sb_to_client(inode->i_sb)->mdsc;
Sage Weila8599bd2009-10-06 11:31:12 -07001691 struct ceph_inode_info *ci = ceph_inode(inode);
1692 int unlock_session = session ? 0 : 1;
1693 int flushing = 0;
1694
1695retry:
1696 spin_lock(&inode->i_lock);
Sage Weile9964c12010-03-01 15:16:56 -08001697 if (ci->i_ceph_flags & CEPH_I_NOFLUSH) {
1698 dout("try_flush_caps skipping %p I_NOFLUSH set\n", inode);
1699 goto out;
1700 }
Sage Weila8599bd2009-10-06 11:31:12 -07001701 if (ci->i_dirty_caps && ci->i_auth_cap) {
1702 struct ceph_cap *cap = ci->i_auth_cap;
1703 int used = __ceph_caps_used(ci);
1704 int want = __ceph_caps_wanted(ci);
1705 int delayed;
1706
1707 if (!session) {
1708 spin_unlock(&inode->i_lock);
1709 session = cap->session;
1710 mutex_lock(&session->s_mutex);
1711 goto retry;
1712 }
1713 BUG_ON(session != cap->session);
1714 if (cap->session->s_state < CEPH_MDS_SESSION_OPEN)
1715 goto out;
1716
Sage Weilcdc35f92009-10-14 14:24:19 -07001717 flushing = __mark_caps_flushing(inode, session);
Sage Weila8599bd2009-10-06 11:31:12 -07001718
1719 /* __send_cap drops i_lock */
1720 delayed = __send_cap(mdsc, cap, CEPH_CAP_OP_FLUSH, used, want,
1721 cap->issued | cap->implemented, flushing,
1722 flush_tid);
1723 if (!delayed)
1724 goto out_unlocked;
1725
1726 spin_lock(&inode->i_lock);
1727 __cap_delay_requeue(mdsc, ci);
1728 }
1729out:
1730 spin_unlock(&inode->i_lock);
1731out_unlocked:
1732 if (session && unlock_session)
1733 mutex_unlock(&session->s_mutex);
1734 return flushing;
1735}
1736
1737/*
1738 * Return true if we've flushed caps through the given flush_tid.
1739 */
1740static int caps_are_flushed(struct inode *inode, unsigned tid)
1741{
1742 struct ceph_inode_info *ci = ceph_inode(inode);
Dan Carpentera5ee7512010-05-07 10:27:14 +02001743 int i, ret = 1;
Sage Weila8599bd2009-10-06 11:31:12 -07001744
1745 spin_lock(&inode->i_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001746 for (i = 0; i < CEPH_CAP_BITS; i++)
1747 if ((ci->i_flushing_caps & (1 << i)) &&
1748 ci->i_cap_flush_tid[i] <= tid) {
1749 /* still flushing this bit */
1750 ret = 0;
1751 break;
1752 }
1753 spin_unlock(&inode->i_lock);
1754 return ret;
1755}
1756
1757/*
1758 * Wait on any unsafe replies for the given inode. First wait on the
1759 * newest request, and make that the upper bound. Then, if there are
1760 * more requests, keep waiting on the oldest as long as it is still older
1761 * than the original request.
1762 */
1763static void sync_write_wait(struct inode *inode)
1764{
1765 struct ceph_inode_info *ci = ceph_inode(inode);
1766 struct list_head *head = &ci->i_unsafe_writes;
1767 struct ceph_osd_request *req;
1768 u64 last_tid;
1769
1770 spin_lock(&ci->i_unsafe_lock);
1771 if (list_empty(head))
1772 goto out;
1773
1774 /* set upper bound as _last_ entry in chain */
1775 req = list_entry(head->prev, struct ceph_osd_request,
1776 r_unsafe_item);
1777 last_tid = req->r_tid;
1778
1779 do {
1780 ceph_osdc_get_request(req);
1781 spin_unlock(&ci->i_unsafe_lock);
1782 dout("sync_write_wait on tid %llu (until %llu)\n",
1783 req->r_tid, last_tid);
1784 wait_for_completion(&req->r_safe_completion);
1785 spin_lock(&ci->i_unsafe_lock);
1786 ceph_osdc_put_request(req);
1787
1788 /*
1789 * from here on look at first entry in chain, since we
1790 * only want to wait for anything older than last_tid
1791 */
1792 if (list_empty(head))
1793 break;
1794 req = list_entry(head->next, struct ceph_osd_request,
1795 r_unsafe_item);
1796 } while (req->r_tid < last_tid);
1797out:
1798 spin_unlock(&ci->i_unsafe_lock);
1799}
1800
Christoph Hellwig7ea80852010-05-26 17:53:25 +02001801int ceph_fsync(struct file *file, int datasync)
Sage Weila8599bd2009-10-06 11:31:12 -07001802{
Christoph Hellwig7ea80852010-05-26 17:53:25 +02001803 struct inode *inode = file->f_mapping->host;
Sage Weila8599bd2009-10-06 11:31:12 -07001804 struct ceph_inode_info *ci = ceph_inode(inode);
1805 unsigned flush_tid;
1806 int ret;
1807 int dirty;
1808
1809 dout("fsync %p%s\n", inode, datasync ? " datasync" : "");
1810 sync_write_wait(inode);
1811
1812 ret = filemap_write_and_wait(inode->i_mapping);
1813 if (ret < 0)
1814 return ret;
1815
1816 dirty = try_flush_caps(inode, NULL, &flush_tid);
1817 dout("fsync dirty caps are %s\n", ceph_cap_string(dirty));
1818
1819 /*
1820 * only wait on non-file metadata writeback (the mds
1821 * can recover size and mtime, so we don't need to
1822 * wait for that)
1823 */
1824 if (!datasync && (dirty & ~CEPH_CAP_ANY_FILE_WR)) {
1825 dout("fsync waiting for flush_tid %u\n", flush_tid);
1826 ret = wait_event_interruptible(ci->i_cap_wq,
1827 caps_are_flushed(inode, flush_tid));
1828 }
1829
1830 dout("fsync %p%s done\n", inode, datasync ? " datasync" : "");
1831 return ret;
1832}
1833
1834/*
1835 * Flush any dirty caps back to the mds. If we aren't asked to wait,
1836 * queue inode for flush but don't do so immediately, because we can
1837 * get by with fewer MDS messages if we wait for data writeback to
1838 * complete first.
1839 */
Stephen Rothwellf1a3d572010-01-18 11:53:08 +11001840int ceph_write_inode(struct inode *inode, struct writeback_control *wbc)
Sage Weila8599bd2009-10-06 11:31:12 -07001841{
1842 struct ceph_inode_info *ci = ceph_inode(inode);
1843 unsigned flush_tid;
1844 int err = 0;
1845 int dirty;
Stephen Rothwellf1a3d572010-01-18 11:53:08 +11001846 int wait = wbc->sync_mode == WB_SYNC_ALL;
Sage Weila8599bd2009-10-06 11:31:12 -07001847
1848 dout("write_inode %p wait=%d\n", inode, wait);
1849 if (wait) {
1850 dirty = try_flush_caps(inode, NULL, &flush_tid);
1851 if (dirty)
1852 err = wait_event_interruptible(ci->i_cap_wq,
1853 caps_are_flushed(inode, flush_tid));
1854 } else {
Cheng Renquan640ef792010-03-26 17:40:33 +08001855 struct ceph_mds_client *mdsc =
1856 &ceph_sb_to_client(inode->i_sb)->mdsc;
Sage Weila8599bd2009-10-06 11:31:12 -07001857
1858 spin_lock(&inode->i_lock);
1859 if (__ceph_caps_dirty(ci))
1860 __cap_delay_requeue_front(mdsc, ci);
1861 spin_unlock(&inode->i_lock);
1862 }
1863 return err;
1864}
1865
1866/*
1867 * After a recovering MDS goes active, we need to resend any caps
1868 * we were flushing.
1869 *
1870 * Caller holds session->s_mutex.
1871 */
1872static void kick_flushing_capsnaps(struct ceph_mds_client *mdsc,
1873 struct ceph_mds_session *session)
1874{
1875 struct ceph_cap_snap *capsnap;
1876
1877 dout("kick_flushing_capsnaps mds%d\n", session->s_mds);
1878 list_for_each_entry(capsnap, &session->s_cap_snaps_flushing,
1879 flushing_item) {
1880 struct ceph_inode_info *ci = capsnap->ci;
1881 struct inode *inode = &ci->vfs_inode;
1882 struct ceph_cap *cap;
1883
1884 spin_lock(&inode->i_lock);
1885 cap = ci->i_auth_cap;
1886 if (cap && cap->session == session) {
1887 dout("kick_flushing_caps %p cap %p capsnap %p\n", inode,
1888 cap, capsnap);
1889 __ceph_flush_snaps(ci, &session);
1890 } else {
1891 pr_err("%p auth cap %p not mds%d ???\n", inode,
1892 cap, session->s_mds);
Sage Weila8599bd2009-10-06 11:31:12 -07001893 }
Sage Weil0b0c06d2010-04-20 10:27:13 -07001894 spin_unlock(&inode->i_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001895 }
1896}
1897
1898void ceph_kick_flushing_caps(struct ceph_mds_client *mdsc,
1899 struct ceph_mds_session *session)
1900{
1901 struct ceph_inode_info *ci;
1902
1903 kick_flushing_capsnaps(mdsc, session);
1904
1905 dout("kick_flushing_caps mds%d\n", session->s_mds);
1906 list_for_each_entry(ci, &session->s_cap_flushing, i_flushing_item) {
1907 struct inode *inode = &ci->vfs_inode;
1908 struct ceph_cap *cap;
1909 int delayed = 0;
1910
1911 spin_lock(&inode->i_lock);
1912 cap = ci->i_auth_cap;
1913 if (cap && cap->session == session) {
1914 dout("kick_flushing_caps %p cap %p %s\n", inode,
1915 cap, ceph_cap_string(ci->i_flushing_caps));
1916 delayed = __send_cap(mdsc, cap, CEPH_CAP_OP_FLUSH,
1917 __ceph_caps_used(ci),
1918 __ceph_caps_wanted(ci),
1919 cap->issued | cap->implemented,
1920 ci->i_flushing_caps, NULL);
1921 if (delayed) {
1922 spin_lock(&inode->i_lock);
1923 __cap_delay_requeue(mdsc, ci);
1924 spin_unlock(&inode->i_lock);
1925 }
1926 } else {
1927 pr_err("%p auth cap %p not mds%d ???\n", inode,
1928 cap, session->s_mds);
1929 spin_unlock(&inode->i_lock);
1930 }
1931 }
1932}
1933
1934
1935/*
1936 * Take references to capabilities we hold, so that we don't release
1937 * them to the MDS prematurely.
1938 *
1939 * Protected by i_lock.
1940 */
1941static void __take_cap_refs(struct ceph_inode_info *ci, int got)
1942{
1943 if (got & CEPH_CAP_PIN)
1944 ci->i_pin_ref++;
1945 if (got & CEPH_CAP_FILE_RD)
1946 ci->i_rd_ref++;
1947 if (got & CEPH_CAP_FILE_CACHE)
1948 ci->i_rdcache_ref++;
1949 if (got & CEPH_CAP_FILE_WR)
1950 ci->i_wr_ref++;
1951 if (got & CEPH_CAP_FILE_BUFFER) {
1952 if (ci->i_wrbuffer_ref == 0)
1953 igrab(&ci->vfs_inode);
1954 ci->i_wrbuffer_ref++;
1955 dout("__take_cap_refs %p wrbuffer %d -> %d (?)\n",
1956 &ci->vfs_inode, ci->i_wrbuffer_ref-1, ci->i_wrbuffer_ref);
1957 }
1958}
1959
1960/*
1961 * Try to grab cap references. Specify those refs we @want, and the
1962 * minimal set we @need. Also include the larger offset we are writing
1963 * to (when applicable), and check against max_size here as well.
1964 * Note that caller is responsible for ensuring max_size increases are
1965 * requested from the MDS.
1966 */
1967static int try_get_cap_refs(struct ceph_inode_info *ci, int need, int want,
1968 int *got, loff_t endoff, int *check_max, int *err)
1969{
1970 struct inode *inode = &ci->vfs_inode;
1971 int ret = 0;
1972 int have, implemented;
Sage Weil195d3ce2010-03-01 09:57:54 -08001973 int file_wanted;
Sage Weila8599bd2009-10-06 11:31:12 -07001974
1975 dout("get_cap_refs %p need %s want %s\n", inode,
1976 ceph_cap_string(need), ceph_cap_string(want));
1977 spin_lock(&inode->i_lock);
1978
Sage Weil195d3ce2010-03-01 09:57:54 -08001979 /* make sure file is actually open */
1980 file_wanted = __ceph_caps_file_wanted(ci);
1981 if ((file_wanted & need) == 0) {
1982 dout("try_get_cap_refs need %s file_wanted %s, EBADF\n",
1983 ceph_cap_string(need), ceph_cap_string(file_wanted));
Sage Weila8599bd2009-10-06 11:31:12 -07001984 *err = -EBADF;
1985 ret = 1;
1986 goto out;
1987 }
1988
1989 if (need & CEPH_CAP_FILE_WR) {
1990 if (endoff >= 0 && endoff > (loff_t)ci->i_max_size) {
1991 dout("get_cap_refs %p endoff %llu > maxsize %llu\n",
1992 inode, endoff, ci->i_max_size);
1993 if (endoff > ci->i_wanted_max_size) {
1994 *check_max = 1;
1995 ret = 1;
1996 }
1997 goto out;
1998 }
1999 /*
2000 * If a sync write is in progress, we must wait, so that we
2001 * can get a final snapshot value for size+mtime.
2002 */
2003 if (__ceph_have_pending_cap_snap(ci)) {
2004 dout("get_cap_refs %p cap_snap_pending\n", inode);
2005 goto out;
2006 }
2007 }
2008 have = __ceph_caps_issued(ci, &implemented);
2009
2010 /*
2011 * disallow writes while a truncate is pending
2012 */
2013 if (ci->i_truncate_pending)
2014 have &= ~CEPH_CAP_FILE_WR;
2015
2016 if ((have & need) == need) {
2017 /*
2018 * Look at (implemented & ~have & not) so that we keep waiting
2019 * on transition from wanted -> needed caps. This is needed
2020 * for WRBUFFER|WR -> WR to avoid a new WR sync write from
2021 * going before a prior buffered writeback happens.
2022 */
2023 int not = want & ~(have & need);
2024 int revoking = implemented & ~have;
2025 dout("get_cap_refs %p have %s but not %s (revoking %s)\n",
2026 inode, ceph_cap_string(have), ceph_cap_string(not),
2027 ceph_cap_string(revoking));
2028 if ((revoking & not) == 0) {
2029 *got = need | (have & want);
2030 __take_cap_refs(ci, *got);
2031 ret = 1;
2032 }
2033 } else {
2034 dout("get_cap_refs %p have %s needed %s\n", inode,
2035 ceph_cap_string(have), ceph_cap_string(need));
2036 }
2037out:
2038 spin_unlock(&inode->i_lock);
2039 dout("get_cap_refs %p ret %d got %s\n", inode,
2040 ret, ceph_cap_string(*got));
2041 return ret;
2042}
2043
2044/*
2045 * Check the offset we are writing up to against our current
2046 * max_size. If necessary, tell the MDS we want to write to
2047 * a larger offset.
2048 */
2049static void check_max_size(struct inode *inode, loff_t endoff)
2050{
2051 struct ceph_inode_info *ci = ceph_inode(inode);
2052 int check = 0;
2053
2054 /* do we need to explicitly request a larger max_size? */
2055 spin_lock(&inode->i_lock);
2056 if ((endoff >= ci->i_max_size ||
2057 endoff > (inode->i_size << 1)) &&
2058 endoff > ci->i_wanted_max_size) {
2059 dout("write %p at large endoff %llu, req max_size\n",
2060 inode, endoff);
2061 ci->i_wanted_max_size = endoff;
2062 check = 1;
2063 }
2064 spin_unlock(&inode->i_lock);
2065 if (check)
2066 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
2067}
2068
2069/*
2070 * Wait for caps, and take cap references. If we can't get a WR cap
2071 * due to a small max_size, make sure we check_max_size (and possibly
2072 * ask the mds) so we don't get hung up indefinitely.
2073 */
2074int ceph_get_caps(struct ceph_inode_info *ci, int need, int want, int *got,
2075 loff_t endoff)
2076{
2077 int check_max, ret, err;
2078
2079retry:
2080 if (endoff > 0)
2081 check_max_size(&ci->vfs_inode, endoff);
2082 check_max = 0;
2083 err = 0;
2084 ret = wait_event_interruptible(ci->i_cap_wq,
2085 try_get_cap_refs(ci, need, want,
2086 got, endoff,
2087 &check_max, &err));
2088 if (err)
2089 ret = err;
2090 if (check_max)
2091 goto retry;
2092 return ret;
2093}
2094
2095/*
2096 * Take cap refs. Caller must already know we hold at least one ref
2097 * on the caps in question or we don't know this is safe.
2098 */
2099void ceph_get_cap_refs(struct ceph_inode_info *ci, int caps)
2100{
2101 spin_lock(&ci->vfs_inode.i_lock);
2102 __take_cap_refs(ci, caps);
2103 spin_unlock(&ci->vfs_inode.i_lock);
2104}
2105
2106/*
2107 * Release cap refs.
2108 *
2109 * If we released the last ref on any given cap, call ceph_check_caps
2110 * to release (or schedule a release).
2111 *
2112 * If we are releasing a WR cap (from a sync write), finalize any affected
2113 * cap_snap, and wake up any waiters.
2114 */
2115void ceph_put_cap_refs(struct ceph_inode_info *ci, int had)
2116{
2117 struct inode *inode = &ci->vfs_inode;
2118 int last = 0, put = 0, flushsnaps = 0, wake = 0;
2119 struct ceph_cap_snap *capsnap;
2120
2121 spin_lock(&inode->i_lock);
2122 if (had & CEPH_CAP_PIN)
2123 --ci->i_pin_ref;
2124 if (had & CEPH_CAP_FILE_RD)
2125 if (--ci->i_rd_ref == 0)
2126 last++;
2127 if (had & CEPH_CAP_FILE_CACHE)
2128 if (--ci->i_rdcache_ref == 0)
2129 last++;
2130 if (had & CEPH_CAP_FILE_BUFFER) {
2131 if (--ci->i_wrbuffer_ref == 0) {
2132 last++;
2133 put++;
2134 }
2135 dout("put_cap_refs %p wrbuffer %d -> %d (?)\n",
2136 inode, ci->i_wrbuffer_ref+1, ci->i_wrbuffer_ref);
2137 }
2138 if (had & CEPH_CAP_FILE_WR)
2139 if (--ci->i_wr_ref == 0) {
2140 last++;
2141 if (!list_empty(&ci->i_cap_snaps)) {
2142 capsnap = list_first_entry(&ci->i_cap_snaps,
2143 struct ceph_cap_snap,
2144 ci_item);
2145 if (capsnap->writing) {
2146 capsnap->writing = 0;
2147 flushsnaps =
2148 __ceph_finish_cap_snap(ci,
2149 capsnap);
2150 wake = 1;
2151 }
2152 }
2153 }
2154 spin_unlock(&inode->i_lock);
2155
Sage Weil819ccbf2010-04-01 09:33:46 -07002156 dout("put_cap_refs %p had %s%s%s\n", inode, ceph_cap_string(had),
2157 last ? " last" : "", put ? " put" : "");
Sage Weila8599bd2009-10-06 11:31:12 -07002158
2159 if (last && !flushsnaps)
2160 ceph_check_caps(ci, 0, NULL);
2161 else if (flushsnaps)
2162 ceph_flush_snaps(ci);
2163 if (wake)
Yehuda Sadeh03066f22010-07-27 13:11:08 -07002164 wake_up_all(&ci->i_cap_wq);
Sage Weila8599bd2009-10-06 11:31:12 -07002165 if (put)
2166 iput(inode);
2167}
2168
2169/*
2170 * Release @nr WRBUFFER refs on dirty pages for the given @snapc snap
2171 * context. Adjust per-snap dirty page accounting as appropriate.
2172 * Once all dirty data for a cap_snap is flushed, flush snapped file
2173 * metadata back to the MDS. If we dropped the last ref, call
2174 * ceph_check_caps.
2175 */
2176void ceph_put_wrbuffer_cap_refs(struct ceph_inode_info *ci, int nr,
2177 struct ceph_snap_context *snapc)
2178{
2179 struct inode *inode = &ci->vfs_inode;
2180 int last = 0;
Sage Weil819ccbf2010-04-01 09:33:46 -07002181 int complete_capsnap = 0;
2182 int drop_capsnap = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07002183 int found = 0;
2184 struct ceph_cap_snap *capsnap = NULL;
2185
2186 spin_lock(&inode->i_lock);
2187 ci->i_wrbuffer_ref -= nr;
2188 last = !ci->i_wrbuffer_ref;
2189
2190 if (ci->i_head_snapc == snapc) {
2191 ci->i_wrbuffer_ref_head -= nr;
2192 if (!ci->i_wrbuffer_ref_head) {
2193 ceph_put_snap_context(ci->i_head_snapc);
2194 ci->i_head_snapc = NULL;
2195 }
2196 dout("put_wrbuffer_cap_refs on %p head %d/%d -> %d/%d %s\n",
2197 inode,
2198 ci->i_wrbuffer_ref+nr, ci->i_wrbuffer_ref_head+nr,
2199 ci->i_wrbuffer_ref, ci->i_wrbuffer_ref_head,
2200 last ? " LAST" : "");
2201 } else {
2202 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
2203 if (capsnap->context == snapc) {
2204 found = 1;
Sage Weila8599bd2009-10-06 11:31:12 -07002205 break;
2206 }
2207 }
2208 BUG_ON(!found);
Sage Weil819ccbf2010-04-01 09:33:46 -07002209 capsnap->dirty_pages -= nr;
2210 if (capsnap->dirty_pages == 0) {
2211 complete_capsnap = 1;
2212 if (capsnap->dirty == 0)
2213 /* cap writeback completed before we created
2214 * the cap_snap; no FLUSHSNAP is needed */
2215 drop_capsnap = 1;
2216 }
Sage Weila8599bd2009-10-06 11:31:12 -07002217 dout("put_wrbuffer_cap_refs on %p cap_snap %p "
Sage Weil819ccbf2010-04-01 09:33:46 -07002218 " snap %lld %d/%d -> %d/%d %s%s%s\n",
Sage Weila8599bd2009-10-06 11:31:12 -07002219 inode, capsnap, capsnap->context->seq,
2220 ci->i_wrbuffer_ref+nr, capsnap->dirty_pages + nr,
2221 ci->i_wrbuffer_ref, capsnap->dirty_pages,
2222 last ? " (wrbuffer last)" : "",
Sage Weil819ccbf2010-04-01 09:33:46 -07002223 complete_capsnap ? " (complete capsnap)" : "",
2224 drop_capsnap ? " (drop capsnap)" : "");
2225 if (drop_capsnap) {
2226 ceph_put_snap_context(capsnap->context);
2227 list_del(&capsnap->ci_item);
2228 list_del(&capsnap->flushing_item);
2229 ceph_put_cap_snap(capsnap);
2230 }
Sage Weila8599bd2009-10-06 11:31:12 -07002231 }
2232
2233 spin_unlock(&inode->i_lock);
2234
2235 if (last) {
2236 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
2237 iput(inode);
Sage Weil819ccbf2010-04-01 09:33:46 -07002238 } else if (complete_capsnap) {
Sage Weila8599bd2009-10-06 11:31:12 -07002239 ceph_flush_snaps(ci);
Yehuda Sadeh03066f22010-07-27 13:11:08 -07002240 wake_up_all(&ci->i_cap_wq);
Sage Weila8599bd2009-10-06 11:31:12 -07002241 }
Sage Weil819ccbf2010-04-01 09:33:46 -07002242 if (drop_capsnap)
2243 iput(inode);
Sage Weila8599bd2009-10-06 11:31:12 -07002244}
2245
2246/*
2247 * Handle a cap GRANT message from the MDS. (Note that a GRANT may
2248 * actually be a revocation if it specifies a smaller cap set.)
2249 *
Sage Weil15637c82010-03-16 13:42:00 -07002250 * caller holds s_mutex and i_lock, we drop both.
2251 *
Sage Weila8599bd2009-10-06 11:31:12 -07002252 * return value:
2253 * 0 - ok
2254 * 1 - check_caps on auth cap only (writeback)
2255 * 2 - check_caps (ack revoke)
2256 */
Sage Weil15637c82010-03-16 13:42:00 -07002257static void handle_cap_grant(struct inode *inode, struct ceph_mds_caps *grant,
2258 struct ceph_mds_session *session,
2259 struct ceph_cap *cap,
2260 struct ceph_buffer *xattr_buf)
Yehuda Sadehcd84db62010-06-11 16:58:48 -07002261 __releases(inode->i_lock)
Sage Weila8599bd2009-10-06 11:31:12 -07002262{
2263 struct ceph_inode_info *ci = ceph_inode(inode);
2264 int mds = session->s_mds;
2265 int seq = le32_to_cpu(grant->seq);
2266 int newcaps = le32_to_cpu(grant->caps);
2267 int issued, implemented, used, wanted, dirty;
2268 u64 size = le64_to_cpu(grant->size);
2269 u64 max_size = le64_to_cpu(grant->max_size);
2270 struct timespec mtime, atime, ctime;
Sage Weil15637c82010-03-16 13:42:00 -07002271 int check_caps = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07002272 int wake = 0;
2273 int writeback = 0;
2274 int revoked_rdcache = 0;
Sage Weil3c6f6b72010-02-09 15:24:44 -08002275 int queue_invalidate = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07002276
2277 dout("handle_cap_grant inode %p cap %p mds%d seq %d %s\n",
2278 inode, cap, mds, seq, ceph_cap_string(newcaps));
2279 dout(" size %llu max_size %llu, i_size %llu\n", size, max_size,
2280 inode->i_size);
2281
2282 /*
2283 * If CACHE is being revoked, and we have no dirty buffers,
2284 * try to invalidate (once). (If there are dirty buffers, we
2285 * will invalidate _after_ writeback.)
2286 */
Sage Weil3b454c42010-06-10 13:20:33 -07002287 if (((cap->issued & ~newcaps) & CEPH_CAP_FILE_CACHE) &&
2288 (newcaps & CEPH_CAP_FILE_LAZYIO) == 0 &&
Yehuda Sadehbcd2cbd2010-02-19 00:12:21 +00002289 !ci->i_wrbuffer_ref) {
Sage Weil5ecad6f2010-02-17 10:43:37 -08002290 if (try_nonblocking_invalidate(inode) == 0) {
2291 revoked_rdcache = 1;
2292 } else {
Sage Weila8599bd2009-10-06 11:31:12 -07002293 /* there were locked pages.. invalidate later
2294 in a separate thread. */
2295 if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
Sage Weil3c6f6b72010-02-09 15:24:44 -08002296 queue_invalidate = 1;
Sage Weila8599bd2009-10-06 11:31:12 -07002297 ci->i_rdcache_revoking = ci->i_rdcache_gen;
2298 }
Sage Weila8599bd2009-10-06 11:31:12 -07002299 }
Sage Weila8599bd2009-10-06 11:31:12 -07002300 }
2301
2302 /* side effects now are allowed */
2303
2304 issued = __ceph_caps_issued(ci, &implemented);
2305 issued |= implemented | __ceph_caps_dirty(ci);
2306
Sage Weil685f9a5d2009-11-09 12:05:48 -08002307 cap->cap_gen = session->s_cap_gen;
Sage Weila8599bd2009-10-06 11:31:12 -07002308
2309 __check_cap_issue(ci, cap, newcaps);
2310
2311 if ((issued & CEPH_CAP_AUTH_EXCL) == 0) {
2312 inode->i_mode = le32_to_cpu(grant->mode);
2313 inode->i_uid = le32_to_cpu(grant->uid);
2314 inode->i_gid = le32_to_cpu(grant->gid);
2315 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
2316 inode->i_uid, inode->i_gid);
2317 }
2318
2319 if ((issued & CEPH_CAP_LINK_EXCL) == 0)
2320 inode->i_nlink = le32_to_cpu(grant->nlink);
2321
2322 if ((issued & CEPH_CAP_XATTR_EXCL) == 0 && grant->xattr_len) {
2323 int len = le32_to_cpu(grant->xattr_len);
2324 u64 version = le64_to_cpu(grant->xattr_version);
2325
2326 if (version > ci->i_xattrs.version) {
2327 dout(" got new xattrs v%llu on %p len %d\n",
2328 version, inode, len);
2329 if (ci->i_xattrs.blob)
2330 ceph_buffer_put(ci->i_xattrs.blob);
2331 ci->i_xattrs.blob = ceph_buffer_get(xattr_buf);
2332 ci->i_xattrs.version = version;
2333 }
2334 }
2335
2336 /* size/ctime/mtime/atime? */
2337 ceph_fill_file_size(inode, issued,
2338 le32_to_cpu(grant->truncate_seq),
2339 le64_to_cpu(grant->truncate_size), size);
2340 ceph_decode_timespec(&mtime, &grant->mtime);
2341 ceph_decode_timespec(&atime, &grant->atime);
2342 ceph_decode_timespec(&ctime, &grant->ctime);
2343 ceph_fill_file_time(inode, issued,
2344 le32_to_cpu(grant->time_warp_seq), &ctime, &mtime,
2345 &atime);
2346
2347 /* max size increase? */
2348 if (max_size != ci->i_max_size) {
2349 dout("max_size %lld -> %llu\n", ci->i_max_size, max_size);
2350 ci->i_max_size = max_size;
2351 if (max_size >= ci->i_wanted_max_size) {
2352 ci->i_wanted_max_size = 0; /* reset */
2353 ci->i_requested_max_size = 0;
2354 }
2355 wake = 1;
2356 }
2357
2358 /* check cap bits */
2359 wanted = __ceph_caps_wanted(ci);
2360 used = __ceph_caps_used(ci);
2361 dirty = __ceph_caps_dirty(ci);
2362 dout(" my wanted = %s, used = %s, dirty %s\n",
2363 ceph_cap_string(wanted),
2364 ceph_cap_string(used),
2365 ceph_cap_string(dirty));
2366 if (wanted != le32_to_cpu(grant->wanted)) {
2367 dout("mds wanted %s -> %s\n",
2368 ceph_cap_string(le32_to_cpu(grant->wanted)),
2369 ceph_cap_string(wanted));
2370 grant->wanted = cpu_to_le32(wanted);
2371 }
2372
2373 cap->seq = seq;
2374
2375 /* file layout may have changed */
2376 ci->i_layout = grant->layout;
2377
2378 /* revocation, grant, or no-op? */
2379 if (cap->issued & ~newcaps) {
Sage Weil3b454c42010-06-10 13:20:33 -07002380 int revoking = cap->issued & ~newcaps;
2381
2382 dout("revocation: %s -> %s (revoking %s)\n",
2383 ceph_cap_string(cap->issued),
2384 ceph_cap_string(newcaps),
2385 ceph_cap_string(revoking));
2386 if (revoking & CEPH_CAP_FILE_BUFFER)
2387 writeback = 1; /* initiate writeback; will delay ack */
2388 else if (revoking == CEPH_CAP_FILE_CACHE &&
2389 (newcaps & CEPH_CAP_FILE_LAZYIO) == 0 &&
2390 queue_invalidate)
2391 ; /* do nothing yet, invalidation will be queued */
2392 else if (cap == ci->i_auth_cap)
2393 check_caps = 1; /* check auth cap only */
2394 else
2395 check_caps = 2; /* check all caps */
Sage Weila8599bd2009-10-06 11:31:12 -07002396 cap->issued = newcaps;
Sage Weil978097c2010-03-08 15:27:53 -08002397 cap->implemented |= newcaps;
Sage Weila8599bd2009-10-06 11:31:12 -07002398 } else if (cap->issued == newcaps) {
2399 dout("caps unchanged: %s -> %s\n",
2400 ceph_cap_string(cap->issued), ceph_cap_string(newcaps));
2401 } else {
2402 dout("grant: %s -> %s\n", ceph_cap_string(cap->issued),
2403 ceph_cap_string(newcaps));
2404 cap->issued = newcaps;
2405 cap->implemented |= newcaps; /* add bits only, to
2406 * avoid stepping on a
2407 * pending revocation */
2408 wake = 1;
2409 }
Sage Weil978097c2010-03-08 15:27:53 -08002410 BUG_ON(cap->issued & ~cap->implemented);
Sage Weila8599bd2009-10-06 11:31:12 -07002411
2412 spin_unlock(&inode->i_lock);
Sage Weil3c6f6b72010-02-09 15:24:44 -08002413 if (writeback)
Sage Weila8599bd2009-10-06 11:31:12 -07002414 /*
2415 * queue inode for writeback: we can't actually call
2416 * filemap_write_and_wait, etc. from message handler
2417 * context.
2418 */
Sage Weil3c6f6b72010-02-09 15:24:44 -08002419 ceph_queue_writeback(inode);
2420 if (queue_invalidate)
2421 ceph_queue_invalidate(inode);
Sage Weila8599bd2009-10-06 11:31:12 -07002422 if (wake)
Yehuda Sadeh03066f22010-07-27 13:11:08 -07002423 wake_up_all(&ci->i_cap_wq);
Sage Weil15637c82010-03-16 13:42:00 -07002424
2425 if (check_caps == 1)
2426 ceph_check_caps(ci, CHECK_CAPS_NODELAY|CHECK_CAPS_AUTHONLY,
2427 session);
2428 else if (check_caps == 2)
2429 ceph_check_caps(ci, CHECK_CAPS_NODELAY, session);
2430 else
2431 mutex_unlock(&session->s_mutex);
Sage Weila8599bd2009-10-06 11:31:12 -07002432}
2433
2434/*
2435 * Handle FLUSH_ACK from MDS, indicating that metadata we sent to the
2436 * MDS has been safely committed.
2437 */
Sage Weil6df058c2009-12-22 11:24:33 -08002438static void handle_cap_flush_ack(struct inode *inode, u64 flush_tid,
Sage Weila8599bd2009-10-06 11:31:12 -07002439 struct ceph_mds_caps *m,
2440 struct ceph_mds_session *session,
2441 struct ceph_cap *cap)
2442 __releases(inode->i_lock)
2443{
2444 struct ceph_inode_info *ci = ceph_inode(inode);
Cheng Renquan640ef792010-03-26 17:40:33 +08002445 struct ceph_mds_client *mdsc = &ceph_sb_to_client(inode->i_sb)->mdsc;
Sage Weila8599bd2009-10-06 11:31:12 -07002446 unsigned seq = le32_to_cpu(m->seq);
2447 int dirty = le32_to_cpu(m->dirty);
2448 int cleaned = 0;
Sage Weilafcdaea2009-10-14 14:27:38 -07002449 int drop = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07002450 int i;
2451
2452 for (i = 0; i < CEPH_CAP_BITS; i++)
2453 if ((dirty & (1 << i)) &&
2454 flush_tid == ci->i_cap_flush_tid[i])
2455 cleaned |= 1 << i;
2456
2457 dout("handle_cap_flush_ack inode %p mds%d seq %d on %s cleaned %s,"
2458 " flushing %s -> %s\n",
2459 inode, session->s_mds, seq, ceph_cap_string(dirty),
2460 ceph_cap_string(cleaned), ceph_cap_string(ci->i_flushing_caps),
2461 ceph_cap_string(ci->i_flushing_caps & ~cleaned));
2462
2463 if (ci->i_flushing_caps == (ci->i_flushing_caps & ~cleaned))
2464 goto out;
2465
Sage Weila8599bd2009-10-06 11:31:12 -07002466 ci->i_flushing_caps &= ~cleaned;
Sage Weila8599bd2009-10-06 11:31:12 -07002467
2468 spin_lock(&mdsc->cap_dirty_lock);
2469 if (ci->i_flushing_caps == 0) {
2470 list_del_init(&ci->i_flushing_item);
2471 if (!list_empty(&session->s_cap_flushing))
2472 dout(" mds%d still flushing cap on %p\n",
2473 session->s_mds,
2474 &list_entry(session->s_cap_flushing.next,
2475 struct ceph_inode_info,
2476 i_flushing_item)->vfs_inode);
2477 mdsc->num_cap_flushing--;
Yehuda Sadeh03066f22010-07-27 13:11:08 -07002478 wake_up_all(&mdsc->cap_flushing_wq);
Sage Weila8599bd2009-10-06 11:31:12 -07002479 dout(" inode %p now !flushing\n", inode);
Sage Weilafcdaea2009-10-14 14:27:38 -07002480
2481 if (ci->i_dirty_caps == 0) {
2482 dout(" inode %p now clean\n", inode);
2483 BUG_ON(!list_empty(&ci->i_dirty_item));
2484 drop = 1;
Sage Weil76e3b392009-10-15 18:13:53 -07002485 } else {
2486 BUG_ON(list_empty(&ci->i_dirty_item));
Sage Weilafcdaea2009-10-14 14:27:38 -07002487 }
Sage Weila8599bd2009-10-06 11:31:12 -07002488 }
2489 spin_unlock(&mdsc->cap_dirty_lock);
Yehuda Sadeh03066f22010-07-27 13:11:08 -07002490 wake_up_all(&ci->i_cap_wq);
Sage Weila8599bd2009-10-06 11:31:12 -07002491
2492out:
2493 spin_unlock(&inode->i_lock);
Sage Weilafcdaea2009-10-14 14:27:38 -07002494 if (drop)
Sage Weila8599bd2009-10-06 11:31:12 -07002495 iput(inode);
2496}
2497
2498/*
2499 * Handle FLUSHSNAP_ACK. MDS has flushed snap data to disk and we can
2500 * throw away our cap_snap.
2501 *
2502 * Caller hold s_mutex.
2503 */
Sage Weil6df058c2009-12-22 11:24:33 -08002504static void handle_cap_flushsnap_ack(struct inode *inode, u64 flush_tid,
Sage Weila8599bd2009-10-06 11:31:12 -07002505 struct ceph_mds_caps *m,
2506 struct ceph_mds_session *session)
2507{
2508 struct ceph_inode_info *ci = ceph_inode(inode);
2509 u64 follows = le64_to_cpu(m->snap_follows);
Sage Weila8599bd2009-10-06 11:31:12 -07002510 struct ceph_cap_snap *capsnap;
2511 int drop = 0;
2512
2513 dout("handle_cap_flushsnap_ack inode %p ci %p mds%d follows %lld\n",
2514 inode, ci, session->s_mds, follows);
2515
2516 spin_lock(&inode->i_lock);
2517 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
2518 if (capsnap->follows == follows) {
2519 if (capsnap->flush_tid != flush_tid) {
2520 dout(" cap_snap %p follows %lld tid %lld !="
2521 " %lld\n", capsnap, follows,
2522 flush_tid, capsnap->flush_tid);
2523 break;
2524 }
2525 WARN_ON(capsnap->dirty_pages || capsnap->writing);
Sage Weil819ccbf2010-04-01 09:33:46 -07002526 dout(" removing %p cap_snap %p follows %lld\n",
2527 inode, capsnap, follows);
Sage Weila8599bd2009-10-06 11:31:12 -07002528 ceph_put_snap_context(capsnap->context);
2529 list_del(&capsnap->ci_item);
2530 list_del(&capsnap->flushing_item);
2531 ceph_put_cap_snap(capsnap);
2532 drop = 1;
2533 break;
2534 } else {
2535 dout(" skipping cap_snap %p follows %lld\n",
2536 capsnap, capsnap->follows);
2537 }
2538 }
2539 spin_unlock(&inode->i_lock);
2540 if (drop)
2541 iput(inode);
2542}
2543
2544/*
2545 * Handle TRUNC from MDS, indicating file truncation.
2546 *
2547 * caller hold s_mutex.
2548 */
2549static void handle_cap_trunc(struct inode *inode,
2550 struct ceph_mds_caps *trunc,
2551 struct ceph_mds_session *session)
2552 __releases(inode->i_lock)
2553{
2554 struct ceph_inode_info *ci = ceph_inode(inode);
2555 int mds = session->s_mds;
2556 int seq = le32_to_cpu(trunc->seq);
2557 u32 truncate_seq = le32_to_cpu(trunc->truncate_seq);
2558 u64 truncate_size = le64_to_cpu(trunc->truncate_size);
2559 u64 size = le64_to_cpu(trunc->size);
2560 int implemented = 0;
2561 int dirty = __ceph_caps_dirty(ci);
2562 int issued = __ceph_caps_issued(ceph_inode(inode), &implemented);
2563 int queue_trunc = 0;
2564
2565 issued |= implemented | dirty;
2566
2567 dout("handle_cap_trunc inode %p mds%d seq %d to %lld seq %d\n",
2568 inode, mds, seq, truncate_size, truncate_seq);
2569 queue_trunc = ceph_fill_file_size(inode, issued,
2570 truncate_seq, truncate_size, size);
2571 spin_unlock(&inode->i_lock);
2572
2573 if (queue_trunc)
Sage Weil3c6f6b72010-02-09 15:24:44 -08002574 ceph_queue_vmtruncate(inode);
Sage Weila8599bd2009-10-06 11:31:12 -07002575}
2576
2577/*
2578 * Handle EXPORT from MDS. Cap is being migrated _from_ this mds to a
2579 * different one. If we are the most recent migration we've seen (as
2580 * indicated by mseq), make note of the migrating cap bits for the
2581 * duration (until we see the corresponding IMPORT).
2582 *
2583 * caller holds s_mutex
2584 */
2585static void handle_cap_export(struct inode *inode, struct ceph_mds_caps *ex,
Sage Weil154f42c2010-06-21 13:45:04 -07002586 struct ceph_mds_session *session,
2587 int *open_target_sessions)
Sage Weila8599bd2009-10-06 11:31:12 -07002588{
2589 struct ceph_inode_info *ci = ceph_inode(inode);
2590 int mds = session->s_mds;
2591 unsigned mseq = le32_to_cpu(ex->migrate_seq);
2592 struct ceph_cap *cap = NULL, *t;
2593 struct rb_node *p;
2594 int remember = 1;
2595
2596 dout("handle_cap_export inode %p ci %p mds%d mseq %d\n",
2597 inode, ci, mds, mseq);
2598
2599 spin_lock(&inode->i_lock);
2600
2601 /* make sure we haven't seen a higher mseq */
2602 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
2603 t = rb_entry(p, struct ceph_cap, ci_node);
2604 if (ceph_seq_cmp(t->mseq, mseq) > 0) {
2605 dout(" higher mseq on cap from mds%d\n",
2606 t->session->s_mds);
2607 remember = 0;
2608 }
2609 if (t->session->s_mds == mds)
2610 cap = t;
2611 }
2612
2613 if (cap) {
2614 if (remember) {
2615 /* make note */
2616 ci->i_cap_exporting_mds = mds;
2617 ci->i_cap_exporting_mseq = mseq;
2618 ci->i_cap_exporting_issued = cap->issued;
Sage Weil154f42c2010-06-21 13:45:04 -07002619
2620 /*
2621 * make sure we have open sessions with all possible
2622 * export targets, so that we get the matching IMPORT
2623 */
2624 *open_target_sessions = 1;
Sage Weila8599bd2009-10-06 11:31:12 -07002625 }
Sage Weil7c1332b2010-02-16 11:39:45 -08002626 __ceph_remove_cap(cap);
Sage Weila8599bd2009-10-06 11:31:12 -07002627 }
Sage Weil4ea00432010-03-16 10:36:40 -07002628 /* else, we already released it */
Sage Weila8599bd2009-10-06 11:31:12 -07002629
2630 spin_unlock(&inode->i_lock);
2631}
2632
2633/*
2634 * Handle cap IMPORT. If there are temp bits from an older EXPORT,
2635 * clean them up.
2636 *
2637 * caller holds s_mutex.
2638 */
2639static void handle_cap_import(struct ceph_mds_client *mdsc,
2640 struct inode *inode, struct ceph_mds_caps *im,
2641 struct ceph_mds_session *session,
2642 void *snaptrace, int snaptrace_len)
2643{
2644 struct ceph_inode_info *ci = ceph_inode(inode);
2645 int mds = session->s_mds;
2646 unsigned issued = le32_to_cpu(im->caps);
2647 unsigned wanted = le32_to_cpu(im->wanted);
2648 unsigned seq = le32_to_cpu(im->seq);
2649 unsigned mseq = le32_to_cpu(im->migrate_seq);
2650 u64 realmino = le64_to_cpu(im->realm);
2651 u64 cap_id = le64_to_cpu(im->cap_id);
2652
2653 if (ci->i_cap_exporting_mds >= 0 &&
2654 ceph_seq_cmp(ci->i_cap_exporting_mseq, mseq) < 0) {
2655 dout("handle_cap_import inode %p ci %p mds%d mseq %d"
2656 " - cleared exporting from mds%d\n",
2657 inode, ci, mds, mseq,
2658 ci->i_cap_exporting_mds);
2659 ci->i_cap_exporting_issued = 0;
2660 ci->i_cap_exporting_mseq = 0;
2661 ci->i_cap_exporting_mds = -1;
2662 } else {
2663 dout("handle_cap_import inode %p ci %p mds%d mseq %d\n",
2664 inode, ci, mds, mseq);
2665 }
2666
2667 down_write(&mdsc->snap_rwsem);
2668 ceph_update_snap_trace(mdsc, snaptrace, snaptrace+snaptrace_len,
2669 false);
2670 downgrade_write(&mdsc->snap_rwsem);
2671 ceph_add_cap(inode, session, cap_id, -1,
2672 issued, wanted, seq, mseq, realmino, CEPH_CAP_FLAG_AUTH,
2673 NULL /* no caps context */);
2674 try_flush_caps(inode, session, NULL);
2675 up_read(&mdsc->snap_rwsem);
2676}
2677
2678/*
2679 * Handle a caps message from the MDS.
2680 *
2681 * Identify the appropriate session, inode, and call the right handler
2682 * based on the cap op.
2683 */
2684void ceph_handle_caps(struct ceph_mds_session *session,
2685 struct ceph_msg *msg)
2686{
2687 struct ceph_mds_client *mdsc = session->s_mdsc;
2688 struct super_block *sb = mdsc->client->sb;
2689 struct inode *inode;
2690 struct ceph_cap *cap;
2691 struct ceph_mds_caps *h;
Sage Weil2600d2d2010-02-22 15:12:16 -08002692 int mds = session->s_mds;
Sage Weila8599bd2009-10-06 11:31:12 -07002693 int op;
Sage Weil3d7ded42010-06-09 16:47:10 -07002694 u32 seq, mseq;
Sage Weila8599bd2009-10-06 11:31:12 -07002695 struct ceph_vino vino;
2696 u64 cap_id;
2697 u64 size, max_size;
Sage Weil6df058c2009-12-22 11:24:33 -08002698 u64 tid;
Sage Weil70edb552010-03-01 13:20:50 -08002699 void *snaptrace;
Sage Weil154f42c2010-06-21 13:45:04 -07002700 int open_target_sessions = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07002701
2702 dout("handle_caps from mds%d\n", mds);
2703
2704 /* decode */
Sage Weil6df058c2009-12-22 11:24:33 -08002705 tid = le64_to_cpu(msg->hdr.tid);
Sage Weila8599bd2009-10-06 11:31:12 -07002706 if (msg->front.iov_len < sizeof(*h))
2707 goto bad;
2708 h = msg->front.iov_base;
Sage Weil70edb552010-03-01 13:20:50 -08002709 snaptrace = h + 1;
Sage Weila8599bd2009-10-06 11:31:12 -07002710 op = le32_to_cpu(h->op);
2711 vino.ino = le64_to_cpu(h->ino);
2712 vino.snap = CEPH_NOSNAP;
2713 cap_id = le64_to_cpu(h->cap_id);
2714 seq = le32_to_cpu(h->seq);
Sage Weil3d7ded42010-06-09 16:47:10 -07002715 mseq = le32_to_cpu(h->migrate_seq);
Sage Weila8599bd2009-10-06 11:31:12 -07002716 size = le64_to_cpu(h->size);
2717 max_size = le64_to_cpu(h->max_size);
2718
2719 mutex_lock(&session->s_mutex);
2720 session->s_seq++;
2721 dout(" mds%d seq %lld cap seq %u\n", session->s_mds, session->s_seq,
2722 (unsigned)seq);
2723
2724 /* lookup ino */
2725 inode = ceph_find_inode(sb, vino);
2726 dout(" op %s ino %llx.%llx inode %p\n", ceph_cap_op_name(op), vino.ino,
2727 vino.snap, inode);
2728 if (!inode) {
2729 dout(" i don't have ino %llx\n", vino.ino);
Sage Weil3d7ded42010-06-09 16:47:10 -07002730
2731 if (op == CEPH_CAP_OP_IMPORT)
2732 __queue_cap_release(session, vino.ino, cap_id,
2733 mseq, seq);
2734
2735 /*
2736 * send any full release message to try to move things
2737 * along for the mds (who clearly thinks we still have this
2738 * cap).
2739 */
Sage Weilee6b2722010-06-10 12:55:52 -07002740 ceph_add_cap_releases(mdsc, session);
Sage Weil3d7ded42010-06-09 16:47:10 -07002741 ceph_send_cap_releases(mdsc, session);
Sage Weila8599bd2009-10-06 11:31:12 -07002742 goto done;
2743 }
2744
2745 /* these will work even if we don't have a cap yet */
2746 switch (op) {
2747 case CEPH_CAP_OP_FLUSHSNAP_ACK:
Sage Weil6df058c2009-12-22 11:24:33 -08002748 handle_cap_flushsnap_ack(inode, tid, h, session);
Sage Weila8599bd2009-10-06 11:31:12 -07002749 goto done;
2750
2751 case CEPH_CAP_OP_EXPORT:
Sage Weil154f42c2010-06-21 13:45:04 -07002752 handle_cap_export(inode, h, session, &open_target_sessions);
Sage Weila8599bd2009-10-06 11:31:12 -07002753 goto done;
2754
2755 case CEPH_CAP_OP_IMPORT:
2756 handle_cap_import(mdsc, inode, h, session,
Sage Weil70edb552010-03-01 13:20:50 -08002757 snaptrace, le32_to_cpu(h->snap_trace_len));
Sage Weil15637c82010-03-16 13:42:00 -07002758 ceph_check_caps(ceph_inode(inode), CHECK_CAPS_NODELAY,
2759 session);
2760 goto done_unlocked;
Sage Weila8599bd2009-10-06 11:31:12 -07002761 }
2762
2763 /* the rest require a cap */
2764 spin_lock(&inode->i_lock);
2765 cap = __get_cap_for_mds(ceph_inode(inode), mds);
2766 if (!cap) {
Sage Weil9dbd4122010-06-10 13:21:20 -07002767 dout(" no cap on %p ino %llx.%llx from mds%d\n",
Sage Weila8599bd2009-10-06 11:31:12 -07002768 inode, ceph_ino(inode), ceph_snap(inode), mds);
2769 spin_unlock(&inode->i_lock);
2770 goto done;
2771 }
2772
2773 /* note that each of these drops i_lock for us */
2774 switch (op) {
2775 case CEPH_CAP_OP_REVOKE:
2776 case CEPH_CAP_OP_GRANT:
Sage Weil15637c82010-03-16 13:42:00 -07002777 handle_cap_grant(inode, h, session, cap, msg->middle);
2778 goto done_unlocked;
Sage Weila8599bd2009-10-06 11:31:12 -07002779
2780 case CEPH_CAP_OP_FLUSH_ACK:
Sage Weil6df058c2009-12-22 11:24:33 -08002781 handle_cap_flush_ack(inode, tid, h, session, cap);
Sage Weila8599bd2009-10-06 11:31:12 -07002782 break;
2783
2784 case CEPH_CAP_OP_TRUNC:
2785 handle_cap_trunc(inode, h, session);
2786 break;
2787
2788 default:
2789 spin_unlock(&inode->i_lock);
2790 pr_err("ceph_handle_caps: unknown cap op %d %s\n", op,
2791 ceph_cap_op_name(op));
2792 }
2793
2794done:
Sage Weil15637c82010-03-16 13:42:00 -07002795 mutex_unlock(&session->s_mutex);
2796done_unlocked:
Sage Weila8599bd2009-10-06 11:31:12 -07002797 if (inode)
2798 iput(inode);
Sage Weil154f42c2010-06-21 13:45:04 -07002799 if (open_target_sessions)
2800 ceph_mdsc_open_export_target_sessions(mdsc, session);
Sage Weila8599bd2009-10-06 11:31:12 -07002801 return;
2802
2803bad:
2804 pr_err("ceph_handle_caps: corrupt message\n");
Sage Weil9ec7cab2009-12-14 15:13:47 -08002805 ceph_msg_dump(msg);
Sage Weila8599bd2009-10-06 11:31:12 -07002806 return;
2807}
2808
2809/*
2810 * Delayed work handler to process end of delayed cap release LRU list.
2811 */
Sage Weilafcdaea2009-10-14 14:27:38 -07002812void ceph_check_delayed_caps(struct ceph_mds_client *mdsc)
Sage Weila8599bd2009-10-06 11:31:12 -07002813{
2814 struct ceph_inode_info *ci;
2815 int flags = CHECK_CAPS_NODELAY;
2816
Sage Weila8599bd2009-10-06 11:31:12 -07002817 dout("check_delayed_caps\n");
2818 while (1) {
2819 spin_lock(&mdsc->cap_delay_lock);
2820 if (list_empty(&mdsc->cap_delay_list))
2821 break;
2822 ci = list_first_entry(&mdsc->cap_delay_list,
2823 struct ceph_inode_info,
2824 i_cap_delay_list);
2825 if ((ci->i_ceph_flags & CEPH_I_FLUSH) == 0 &&
2826 time_before(jiffies, ci->i_hold_caps_max))
2827 break;
2828 list_del_init(&ci->i_cap_delay_list);
2829 spin_unlock(&mdsc->cap_delay_lock);
2830 dout("check_delayed_caps on %p\n", &ci->vfs_inode);
2831 ceph_check_caps(ci, flags, NULL);
2832 }
2833 spin_unlock(&mdsc->cap_delay_lock);
2834}
2835
2836/*
Sage Weilafcdaea2009-10-14 14:27:38 -07002837 * Flush all dirty caps to the mds
2838 */
2839void ceph_flush_dirty_caps(struct ceph_mds_client *mdsc)
2840{
Sage Weile9964c12010-03-01 15:16:56 -08002841 struct ceph_inode_info *ci, *nci = NULL;
2842 struct inode *inode, *ninode = NULL;
2843 struct list_head *p, *n;
Sage Weilafcdaea2009-10-14 14:27:38 -07002844
2845 dout("flush_dirty_caps\n");
2846 spin_lock(&mdsc->cap_dirty_lock);
Sage Weile9964c12010-03-01 15:16:56 -08002847 list_for_each_safe(p, n, &mdsc->cap_dirty) {
2848 if (nci) {
2849 ci = nci;
2850 inode = ninode;
2851 ci->i_ceph_flags &= ~CEPH_I_NOFLUSH;
2852 dout("flush_dirty_caps inode %p (was next inode)\n",
2853 inode);
2854 } else {
2855 ci = list_entry(p, struct ceph_inode_info,
2856 i_dirty_item);
2857 inode = igrab(&ci->vfs_inode);
2858 BUG_ON(!inode);
2859 dout("flush_dirty_caps inode %p\n", inode);
2860 }
2861 if (n != &mdsc->cap_dirty) {
2862 nci = list_entry(n, struct ceph_inode_info,
2863 i_dirty_item);
2864 ninode = igrab(&nci->vfs_inode);
2865 BUG_ON(!ninode);
2866 nci->i_ceph_flags |= CEPH_I_NOFLUSH;
2867 dout("flush_dirty_caps next inode %p, noflush\n",
2868 ninode);
2869 } else {
2870 nci = NULL;
2871 ninode = NULL;
2872 }
Sage Weilafcdaea2009-10-14 14:27:38 -07002873 spin_unlock(&mdsc->cap_dirty_lock);
2874 if (inode) {
2875 ceph_check_caps(ci, CHECK_CAPS_NODELAY|CHECK_CAPS_FLUSH,
2876 NULL);
2877 iput(inode);
2878 }
2879 spin_lock(&mdsc->cap_dirty_lock);
2880 }
2881 spin_unlock(&mdsc->cap_dirty_lock);
2882}
2883
2884/*
Sage Weila8599bd2009-10-06 11:31:12 -07002885 * Drop open file reference. If we were the last open file,
2886 * we may need to release capabilities to the MDS (or schedule
2887 * their delayed release).
2888 */
2889void ceph_put_fmode(struct ceph_inode_info *ci, int fmode)
2890{
2891 struct inode *inode = &ci->vfs_inode;
2892 int last = 0;
2893
2894 spin_lock(&inode->i_lock);
2895 dout("put_fmode %p fmode %d %d -> %d\n", inode, fmode,
2896 ci->i_nr_by_mode[fmode], ci->i_nr_by_mode[fmode]-1);
2897 BUG_ON(ci->i_nr_by_mode[fmode] == 0);
2898 if (--ci->i_nr_by_mode[fmode] == 0)
2899 last++;
2900 spin_unlock(&inode->i_lock);
2901
2902 if (last && ci->i_vino.snap == CEPH_NOSNAP)
2903 ceph_check_caps(ci, 0, NULL);
2904}
2905
2906/*
2907 * Helpers for embedding cap and dentry lease releases into mds
2908 * requests.
2909 *
2910 * @force is used by dentry_release (below) to force inclusion of a
2911 * record for the directory inode, even when there aren't any caps to
2912 * drop.
2913 */
2914int ceph_encode_inode_release(void **p, struct inode *inode,
2915 int mds, int drop, int unless, int force)
2916{
2917 struct ceph_inode_info *ci = ceph_inode(inode);
2918 struct ceph_cap *cap;
2919 struct ceph_mds_request_release *rel = *p;
Sage Weilec97f882010-06-24 15:12:37 -07002920 int used, dirty;
Sage Weila8599bd2009-10-06 11:31:12 -07002921 int ret = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07002922
2923 spin_lock(&inode->i_lock);
Sage Weil916623d2010-03-16 15:01:07 -07002924 used = __ceph_caps_used(ci);
Sage Weilec97f882010-06-24 15:12:37 -07002925 dirty = __ceph_caps_dirty(ci);
Sage Weil916623d2010-03-16 15:01:07 -07002926
Sage Weilec97f882010-06-24 15:12:37 -07002927 dout("encode_inode_release %p mds%d used|dirty %s drop %s unless %s\n",
2928 inode, mds, ceph_cap_string(used|dirty), ceph_cap_string(drop),
Sage Weil916623d2010-03-16 15:01:07 -07002929 ceph_cap_string(unless));
2930
Sage Weilec97f882010-06-24 15:12:37 -07002931 /* only drop unused, clean caps */
2932 drop &= ~(used | dirty);
Sage Weil916623d2010-03-16 15:01:07 -07002933
Sage Weila8599bd2009-10-06 11:31:12 -07002934 cap = __get_cap_for_mds(ci, mds);
2935 if (cap && __cap_is_valid(cap)) {
2936 if (force ||
2937 ((cap->issued & drop) &&
2938 (cap->issued & unless) == 0)) {
2939 if ((cap->issued & drop) &&
2940 (cap->issued & unless) == 0) {
2941 dout("encode_inode_release %p cap %p %s -> "
2942 "%s\n", inode, cap,
2943 ceph_cap_string(cap->issued),
2944 ceph_cap_string(cap->issued & ~drop));
2945 cap->issued &= ~drop;
2946 cap->implemented &= ~drop;
2947 if (ci->i_ceph_flags & CEPH_I_NODELAY) {
2948 int wanted = __ceph_caps_wanted(ci);
2949 dout(" wanted %s -> %s (act %s)\n",
2950 ceph_cap_string(cap->mds_wanted),
2951 ceph_cap_string(cap->mds_wanted &
2952 ~wanted),
2953 ceph_cap_string(wanted));
2954 cap->mds_wanted &= wanted;
2955 }
2956 } else {
2957 dout("encode_inode_release %p cap %p %s"
2958 " (force)\n", inode, cap,
2959 ceph_cap_string(cap->issued));
2960 }
2961
2962 rel->ino = cpu_to_le64(ceph_ino(inode));
2963 rel->cap_id = cpu_to_le64(cap->cap_id);
2964 rel->seq = cpu_to_le32(cap->seq);
2965 rel->issue_seq = cpu_to_le32(cap->issue_seq),
2966 rel->mseq = cpu_to_le32(cap->mseq);
2967 rel->caps = cpu_to_le32(cap->issued);
2968 rel->wanted = cpu_to_le32(cap->mds_wanted);
2969 rel->dname_len = 0;
2970 rel->dname_seq = 0;
2971 *p += sizeof(*rel);
2972 ret = 1;
2973 } else {
2974 dout("encode_inode_release %p cap %p %s\n",
2975 inode, cap, ceph_cap_string(cap->issued));
2976 }
2977 }
2978 spin_unlock(&inode->i_lock);
2979 return ret;
2980}
2981
2982int ceph_encode_dentry_release(void **p, struct dentry *dentry,
2983 int mds, int drop, int unless)
2984{
2985 struct inode *dir = dentry->d_parent->d_inode;
2986 struct ceph_mds_request_release *rel = *p;
2987 struct ceph_dentry_info *di = ceph_dentry(dentry);
2988 int force = 0;
2989 int ret;
2990
2991 /*
2992 * force an record for the directory caps if we have a dentry lease.
2993 * this is racy (can't take i_lock and d_lock together), but it
2994 * doesn't have to be perfect; the mds will revoke anything we don't
2995 * release.
2996 */
2997 spin_lock(&dentry->d_lock);
2998 if (di->lease_session && di->lease_session->s_mds == mds)
2999 force = 1;
3000 spin_unlock(&dentry->d_lock);
3001
3002 ret = ceph_encode_inode_release(p, dir, mds, drop, unless, force);
3003
3004 spin_lock(&dentry->d_lock);
3005 if (ret && di->lease_session && di->lease_session->s_mds == mds) {
3006 dout("encode_dentry_release %p mds%d seq %d\n",
3007 dentry, mds, (int)di->lease_seq);
3008 rel->dname_len = cpu_to_le32(dentry->d_name.len);
3009 memcpy(*p, dentry->d_name.name, dentry->d_name.len);
3010 *p += dentry->d_name.len;
3011 rel->dname_seq = cpu_to_le32(di->lease_seq);
Sage Weil1dadcce2010-07-23 13:54:21 -07003012 __ceph_mdsc_drop_dentry_lease(dentry);
Sage Weila8599bd2009-10-06 11:31:12 -07003013 }
3014 spin_unlock(&dentry->d_lock);
3015 return ret;
3016}