blob: 60d27bc9eb83d82f61e448031495625e42c752d3 [file] [log] [blame]
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001#include <linux/ceph/ceph_debug.h>
Sage Weila8599bd2009-10-06 11:31:12 -07002
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"
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -070012#include "mds_client.h"
13#include <linux/ceph/decode.h>
14#include <linux/ceph/messenger.h>
Sage Weila8599bd2009-10-06 11:31:12 -070015
16/*
17 * Capability management
18 *
19 * The Ceph metadata servers control client access to inode metadata
20 * and file data by issuing capabilities, granting clients permission
21 * to read and/or write both inode field and file data to OSDs
22 * (storage nodes). Each capability consists of a set of bits
23 * indicating which operations are allowed.
24 *
25 * If the client holds a *_SHARED cap, the client has a coherent value
26 * that can be safely read from the cached inode.
27 *
28 * In the case of a *_EXCL (exclusive) or FILE_WR capabilities, the
29 * client is allowed to change inode attributes (e.g., file size,
30 * mtime), note its dirty state in the ceph_cap, and asynchronously
31 * flush that metadata change to the MDS.
32 *
33 * In the event of a conflicting operation (perhaps by another
34 * client), the MDS will revoke the conflicting client capabilities.
35 *
36 * In order for a client to cache an inode, it must hold a capability
37 * with at least one MDS server. When inodes are released, release
38 * notifications are batched and periodically sent en masse to the MDS
39 * cluster to release server state.
40 */
41
42
43/*
44 * Generate readable cap strings for debugging output.
45 */
46#define MAX_CAP_STR 20
47static char cap_str[MAX_CAP_STR][40];
48static DEFINE_SPINLOCK(cap_str_lock);
49static int last_cap_str;
50
51static char *gcap_string(char *s, int c)
52{
53 if (c & CEPH_CAP_GSHARED)
54 *s++ = 's';
55 if (c & CEPH_CAP_GEXCL)
56 *s++ = 'x';
57 if (c & CEPH_CAP_GCACHE)
58 *s++ = 'c';
59 if (c & CEPH_CAP_GRD)
60 *s++ = 'r';
61 if (c & CEPH_CAP_GWR)
62 *s++ = 'w';
63 if (c & CEPH_CAP_GBUFFER)
64 *s++ = 'b';
65 if (c & CEPH_CAP_GLAZYIO)
66 *s++ = 'l';
67 return s;
68}
69
70const char *ceph_cap_string(int caps)
71{
72 int i;
73 char *s;
74 int c;
75
76 spin_lock(&cap_str_lock);
77 i = last_cap_str++;
78 if (last_cap_str == MAX_CAP_STR)
79 last_cap_str = 0;
80 spin_unlock(&cap_str_lock);
81
82 s = cap_str[i];
83
84 if (caps & CEPH_CAP_PIN)
85 *s++ = 'p';
86
87 c = (caps >> CEPH_CAP_SAUTH) & 3;
88 if (c) {
89 *s++ = 'A';
90 s = gcap_string(s, c);
91 }
92
93 c = (caps >> CEPH_CAP_SLINK) & 3;
94 if (c) {
95 *s++ = 'L';
96 s = gcap_string(s, c);
97 }
98
99 c = (caps >> CEPH_CAP_SXATTR) & 3;
100 if (c) {
101 *s++ = 'X';
102 s = gcap_string(s, c);
103 }
104
105 c = caps >> CEPH_CAP_SFILE;
106 if (c) {
107 *s++ = 'F';
108 s = gcap_string(s, c);
109 }
110
111 if (s == cap_str[i])
112 *s++ = '-';
113 *s = 0;
114 return cap_str[i];
115}
116
Yehuda Sadeh37151662010-06-17 16:16:12 -0700117void ceph_caps_init(struct ceph_mds_client *mdsc)
Sage Weila8599bd2009-10-06 11:31:12 -0700118{
Yehuda Sadeh37151662010-06-17 16:16:12 -0700119 INIT_LIST_HEAD(&mdsc->caps_list);
120 spin_lock_init(&mdsc->caps_list_lock);
Sage Weila8599bd2009-10-06 11:31:12 -0700121}
122
Yehuda Sadeh37151662010-06-17 16:16:12 -0700123void ceph_caps_finalize(struct ceph_mds_client *mdsc)
Sage Weila8599bd2009-10-06 11:31:12 -0700124{
125 struct ceph_cap *cap;
126
Yehuda Sadeh37151662010-06-17 16:16:12 -0700127 spin_lock(&mdsc->caps_list_lock);
128 while (!list_empty(&mdsc->caps_list)) {
129 cap = list_first_entry(&mdsc->caps_list,
130 struct ceph_cap, caps_item);
Sage Weila8599bd2009-10-06 11:31:12 -0700131 list_del(&cap->caps_item);
132 kmem_cache_free(ceph_cap_cachep, cap);
133 }
Yehuda Sadeh37151662010-06-17 16:16:12 -0700134 mdsc->caps_total_count = 0;
135 mdsc->caps_avail_count = 0;
136 mdsc->caps_use_count = 0;
137 mdsc->caps_reserve_count = 0;
138 mdsc->caps_min_count = 0;
139 spin_unlock(&mdsc->caps_list_lock);
Sage Weil85ccce42010-02-17 10:02:43 -0800140}
141
Yehuda Sadeh37151662010-06-17 16:16:12 -0700142void ceph_adjust_min_caps(struct ceph_mds_client *mdsc, int delta)
Sage Weil85ccce42010-02-17 10:02:43 -0800143{
Yehuda Sadeh37151662010-06-17 16:16:12 -0700144 spin_lock(&mdsc->caps_list_lock);
145 mdsc->caps_min_count += delta;
146 BUG_ON(mdsc->caps_min_count < 0);
147 spin_unlock(&mdsc->caps_list_lock);
Sage Weila8599bd2009-10-06 11:31:12 -0700148}
149
Yehuda Sadeh37151662010-06-17 16:16:12 -0700150int ceph_reserve_caps(struct ceph_mds_client *mdsc,
151 struct ceph_cap_reservation *ctx, int need)
Sage Weila8599bd2009-10-06 11:31:12 -0700152{
153 int i;
154 struct ceph_cap *cap;
155 int have;
156 int alloc = 0;
157 LIST_HEAD(newcaps);
158 int ret = 0;
159
160 dout("reserve caps ctx=%p need=%d\n", ctx, need);
161
162 /* first reserve any caps that are already allocated */
Yehuda Sadeh37151662010-06-17 16:16:12 -0700163 spin_lock(&mdsc->caps_list_lock);
164 if (mdsc->caps_avail_count >= need)
Sage Weila8599bd2009-10-06 11:31:12 -0700165 have = need;
166 else
Yehuda Sadeh37151662010-06-17 16:16:12 -0700167 have = mdsc->caps_avail_count;
168 mdsc->caps_avail_count -= have;
169 mdsc->caps_reserve_count += have;
170 BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count +
171 mdsc->caps_reserve_count +
172 mdsc->caps_avail_count);
173 spin_unlock(&mdsc->caps_list_lock);
Sage Weila8599bd2009-10-06 11:31:12 -0700174
175 for (i = have; i < need; i++) {
176 cap = kmem_cache_alloc(ceph_cap_cachep, GFP_NOFS);
177 if (!cap) {
178 ret = -ENOMEM;
179 goto out_alloc_count;
180 }
181 list_add(&cap->caps_item, &newcaps);
182 alloc++;
183 }
184 BUG_ON(have + alloc != need);
185
Yehuda Sadeh37151662010-06-17 16:16:12 -0700186 spin_lock(&mdsc->caps_list_lock);
187 mdsc->caps_total_count += alloc;
188 mdsc->caps_reserve_count += alloc;
189 list_splice(&newcaps, &mdsc->caps_list);
Sage Weila8599bd2009-10-06 11:31:12 -0700190
Yehuda Sadeh37151662010-06-17 16:16:12 -0700191 BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count +
192 mdsc->caps_reserve_count +
193 mdsc->caps_avail_count);
194 spin_unlock(&mdsc->caps_list_lock);
Sage Weila8599bd2009-10-06 11:31:12 -0700195
196 ctx->count = need;
197 dout("reserve caps ctx=%p %d = %d used + %d resv + %d avail\n",
Yehuda Sadeh37151662010-06-17 16:16:12 -0700198 ctx, mdsc->caps_total_count, mdsc->caps_use_count,
199 mdsc->caps_reserve_count, mdsc->caps_avail_count);
Sage Weila8599bd2009-10-06 11:31:12 -0700200 return 0;
201
202out_alloc_count:
203 /* we didn't manage to reserve as much as we needed */
204 pr_warning("reserve caps ctx=%p ENOMEM need=%d got=%d\n",
205 ctx, need, have);
206 return ret;
207}
208
Yehuda Sadeh37151662010-06-17 16:16:12 -0700209int ceph_unreserve_caps(struct ceph_mds_client *mdsc,
210 struct ceph_cap_reservation *ctx)
Sage Weila8599bd2009-10-06 11:31:12 -0700211{
212 dout("unreserve caps ctx=%p count=%d\n", ctx, ctx->count);
213 if (ctx->count) {
Yehuda Sadeh37151662010-06-17 16:16:12 -0700214 spin_lock(&mdsc->caps_list_lock);
215 BUG_ON(mdsc->caps_reserve_count < ctx->count);
216 mdsc->caps_reserve_count -= ctx->count;
217 mdsc->caps_avail_count += ctx->count;
Sage Weila8599bd2009-10-06 11:31:12 -0700218 ctx->count = 0;
219 dout("unreserve caps %d = %d used + %d resv + %d avail\n",
Yehuda Sadeh37151662010-06-17 16:16:12 -0700220 mdsc->caps_total_count, mdsc->caps_use_count,
221 mdsc->caps_reserve_count, mdsc->caps_avail_count);
222 BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count +
223 mdsc->caps_reserve_count +
224 mdsc->caps_avail_count);
225 spin_unlock(&mdsc->caps_list_lock);
Sage Weila8599bd2009-10-06 11:31:12 -0700226 }
227 return 0;
228}
229
Yehuda Sadeh37151662010-06-17 16:16:12 -0700230static struct ceph_cap *get_cap(struct ceph_mds_client *mdsc,
231 struct ceph_cap_reservation *ctx)
Sage Weila8599bd2009-10-06 11:31:12 -0700232{
233 struct ceph_cap *cap = NULL;
234
235 /* temporary, until we do something about cap import/export */
Sage Weil443b3762010-06-29 09:28:39 -0700236 if (!ctx) {
237 cap = kmem_cache_alloc(ceph_cap_cachep, GFP_NOFS);
238 if (cap) {
Yehuda Sadeh37151662010-06-17 16:16:12 -0700239 mdsc->caps_use_count++;
240 mdsc->caps_total_count++;
Sage Weil443b3762010-06-29 09:28:39 -0700241 }
242 return cap;
243 }
Sage Weila8599bd2009-10-06 11:31:12 -0700244
Yehuda Sadeh37151662010-06-17 16:16:12 -0700245 spin_lock(&mdsc->caps_list_lock);
Sage Weila8599bd2009-10-06 11:31:12 -0700246 dout("get_cap ctx=%p (%d) %d = %d used + %d resv + %d avail\n",
Yehuda Sadeh37151662010-06-17 16:16:12 -0700247 ctx, ctx->count, mdsc->caps_total_count, mdsc->caps_use_count,
248 mdsc->caps_reserve_count, mdsc->caps_avail_count);
Sage Weila8599bd2009-10-06 11:31:12 -0700249 BUG_ON(!ctx->count);
Yehuda Sadeh37151662010-06-17 16:16:12 -0700250 BUG_ON(ctx->count > mdsc->caps_reserve_count);
251 BUG_ON(list_empty(&mdsc->caps_list));
Sage Weila8599bd2009-10-06 11:31:12 -0700252
253 ctx->count--;
Yehuda Sadeh37151662010-06-17 16:16:12 -0700254 mdsc->caps_reserve_count--;
255 mdsc->caps_use_count++;
Sage Weila8599bd2009-10-06 11:31:12 -0700256
Yehuda Sadeh37151662010-06-17 16:16:12 -0700257 cap = list_first_entry(&mdsc->caps_list, struct ceph_cap, caps_item);
Sage Weila8599bd2009-10-06 11:31:12 -0700258 list_del(&cap->caps_item);
259
Yehuda Sadeh37151662010-06-17 16:16:12 -0700260 BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count +
261 mdsc->caps_reserve_count + mdsc->caps_avail_count);
262 spin_unlock(&mdsc->caps_list_lock);
Sage Weila8599bd2009-10-06 11:31:12 -0700263 return cap;
264}
265
Yehuda Sadeh37151662010-06-17 16:16:12 -0700266void ceph_put_cap(struct ceph_mds_client *mdsc, struct ceph_cap *cap)
Sage Weila8599bd2009-10-06 11:31:12 -0700267{
Yehuda Sadeh37151662010-06-17 16:16:12 -0700268 spin_lock(&mdsc->caps_list_lock);
Sage Weil7c1332b2010-02-16 11:39:45 -0800269 dout("put_cap %p %d = %d used + %d resv + %d avail\n",
Yehuda Sadeh37151662010-06-17 16:16:12 -0700270 cap, mdsc->caps_total_count, mdsc->caps_use_count,
271 mdsc->caps_reserve_count, mdsc->caps_avail_count);
272 mdsc->caps_use_count--;
Sage Weila8599bd2009-10-06 11:31:12 -0700273 /*
Sage Weil85ccce42010-02-17 10:02:43 -0800274 * Keep some preallocated caps around (ceph_min_count), to
275 * avoid lots of free/alloc churn.
Sage Weila8599bd2009-10-06 11:31:12 -0700276 */
Yehuda Sadeh37151662010-06-17 16:16:12 -0700277 if (mdsc->caps_avail_count >= mdsc->caps_reserve_count +
278 mdsc->caps_min_count) {
279 mdsc->caps_total_count--;
Sage Weila8599bd2009-10-06 11:31:12 -0700280 kmem_cache_free(ceph_cap_cachep, cap);
281 } else {
Yehuda Sadeh37151662010-06-17 16:16:12 -0700282 mdsc->caps_avail_count++;
283 list_add(&cap->caps_item, &mdsc->caps_list);
Sage Weila8599bd2009-10-06 11:31:12 -0700284 }
285
Yehuda Sadeh37151662010-06-17 16:16:12 -0700286 BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count +
287 mdsc->caps_reserve_count + mdsc->caps_avail_count);
288 spin_unlock(&mdsc->caps_list_lock);
Sage Weila8599bd2009-10-06 11:31:12 -0700289}
290
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700291void ceph_reservation_status(struct ceph_fs_client *fsc,
Sage Weil85ccce42010-02-17 10:02:43 -0800292 int *total, int *avail, int *used, int *reserved,
293 int *min)
Sage Weila8599bd2009-10-06 11:31:12 -0700294{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700295 struct ceph_mds_client *mdsc = fsc->mdsc;
Yehuda Sadeh37151662010-06-17 16:16:12 -0700296
Sage Weila8599bd2009-10-06 11:31:12 -0700297 if (total)
Yehuda Sadeh37151662010-06-17 16:16:12 -0700298 *total = mdsc->caps_total_count;
Sage Weila8599bd2009-10-06 11:31:12 -0700299 if (avail)
Yehuda Sadeh37151662010-06-17 16:16:12 -0700300 *avail = mdsc->caps_avail_count;
Sage Weila8599bd2009-10-06 11:31:12 -0700301 if (used)
Yehuda Sadeh37151662010-06-17 16:16:12 -0700302 *used = mdsc->caps_use_count;
Sage Weila8599bd2009-10-06 11:31:12 -0700303 if (reserved)
Yehuda Sadeh37151662010-06-17 16:16:12 -0700304 *reserved = mdsc->caps_reserve_count;
Sage Weil85ccce42010-02-17 10:02:43 -0800305 if (min)
Yehuda Sadeh37151662010-06-17 16:16:12 -0700306 *min = mdsc->caps_min_count;
Sage Weila8599bd2009-10-06 11:31:12 -0700307}
308
309/*
310 * Find ceph_cap for given mds, if any.
311 *
312 * Called with i_lock held.
313 */
314static struct ceph_cap *__get_cap_for_mds(struct ceph_inode_info *ci, int mds)
315{
316 struct ceph_cap *cap;
317 struct rb_node *n = ci->i_caps.rb_node;
318
319 while (n) {
320 cap = rb_entry(n, struct ceph_cap, ci_node);
321 if (mds < cap->mds)
322 n = n->rb_left;
323 else if (mds > cap->mds)
324 n = n->rb_right;
325 else
326 return cap;
327 }
328 return NULL;
329}
330
Greg Farnum2bc50252010-06-30 12:44:34 -0700331struct ceph_cap *ceph_get_cap_for_mds(struct ceph_inode_info *ci, int mds)
332{
333 struct ceph_cap *cap;
334
335 spin_lock(&ci->vfs_inode.i_lock);
336 cap = __get_cap_for_mds(ci, mds);
337 spin_unlock(&ci->vfs_inode.i_lock);
338 return cap;
339}
340
Sage Weila8599bd2009-10-06 11:31:12 -0700341/*
Sage Weil33caad32010-05-26 14:31:27 -0700342 * Return id of any MDS with a cap, preferably FILE_WR|BUFFER|EXCL, else -1.
Sage Weila8599bd2009-10-06 11:31:12 -0700343 */
Sage Weilca81f3f2010-06-10 14:21:36 -0700344static int __ceph_get_cap_mds(struct ceph_inode_info *ci)
Sage Weila8599bd2009-10-06 11:31:12 -0700345{
346 struct ceph_cap *cap;
347 int mds = -1;
348 struct rb_node *p;
349
Sage Weil33caad32010-05-26 14:31:27 -0700350 /* prefer mds with WR|BUFFER|EXCL caps */
Sage Weila8599bd2009-10-06 11:31:12 -0700351 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
352 cap = rb_entry(p, struct ceph_cap, ci_node);
353 mds = cap->mds;
Sage Weila8599bd2009-10-06 11:31:12 -0700354 if (cap->issued & (CEPH_CAP_FILE_WR |
355 CEPH_CAP_FILE_BUFFER |
356 CEPH_CAP_FILE_EXCL))
357 break;
358 }
359 return mds;
360}
361
362int ceph_get_cap_mds(struct inode *inode)
363{
364 int mds;
365 spin_lock(&inode->i_lock);
Sage Weilca81f3f2010-06-10 14:21:36 -0700366 mds = __ceph_get_cap_mds(ceph_inode(inode));
Sage Weila8599bd2009-10-06 11:31:12 -0700367 spin_unlock(&inode->i_lock);
368 return mds;
369}
370
371/*
372 * Called under i_lock.
373 */
374static void __insert_cap_node(struct ceph_inode_info *ci,
375 struct ceph_cap *new)
376{
377 struct rb_node **p = &ci->i_caps.rb_node;
378 struct rb_node *parent = NULL;
379 struct ceph_cap *cap = NULL;
380
381 while (*p) {
382 parent = *p;
383 cap = rb_entry(parent, struct ceph_cap, ci_node);
384 if (new->mds < cap->mds)
385 p = &(*p)->rb_left;
386 else if (new->mds > cap->mds)
387 p = &(*p)->rb_right;
388 else
389 BUG();
390 }
391
392 rb_link_node(&new->ci_node, parent, p);
393 rb_insert_color(&new->ci_node, &ci->i_caps);
394}
395
396/*
397 * (re)set cap hold timeouts, which control the delayed release
398 * of unused caps back to the MDS. Should be called on cap use.
399 */
400static void __cap_set_timeouts(struct ceph_mds_client *mdsc,
401 struct ceph_inode_info *ci)
402{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700403 struct ceph_mount_options *ma = mdsc->fsc->mount_options;
Sage Weila8599bd2009-10-06 11:31:12 -0700404
405 ci->i_hold_caps_min = round_jiffies(jiffies +
406 ma->caps_wanted_delay_min * HZ);
407 ci->i_hold_caps_max = round_jiffies(jiffies +
408 ma->caps_wanted_delay_max * HZ);
409 dout("__cap_set_timeouts %p min %lu max %lu\n", &ci->vfs_inode,
410 ci->i_hold_caps_min - jiffies, ci->i_hold_caps_max - jiffies);
411}
412
413/*
414 * (Re)queue cap at the end of the delayed cap release list.
415 *
416 * If I_FLUSH is set, leave the inode at the front of the list.
417 *
418 * Caller holds i_lock
419 * -> we take mdsc->cap_delay_lock
420 */
421static void __cap_delay_requeue(struct ceph_mds_client *mdsc,
422 struct ceph_inode_info *ci)
423{
424 __cap_set_timeouts(mdsc, ci);
425 dout("__cap_delay_requeue %p flags %d at %lu\n", &ci->vfs_inode,
426 ci->i_ceph_flags, ci->i_hold_caps_max);
427 if (!mdsc->stopping) {
428 spin_lock(&mdsc->cap_delay_lock);
429 if (!list_empty(&ci->i_cap_delay_list)) {
430 if (ci->i_ceph_flags & CEPH_I_FLUSH)
431 goto no_change;
432 list_del_init(&ci->i_cap_delay_list);
433 }
434 list_add_tail(&ci->i_cap_delay_list, &mdsc->cap_delay_list);
435no_change:
436 spin_unlock(&mdsc->cap_delay_lock);
437 }
438}
439
440/*
441 * Queue an inode for immediate writeback. Mark inode with I_FLUSH,
442 * indicating we should send a cap message to flush dirty metadata
443 * asap, and move to the front of the delayed cap list.
444 */
445static void __cap_delay_requeue_front(struct ceph_mds_client *mdsc,
446 struct ceph_inode_info *ci)
447{
448 dout("__cap_delay_requeue_front %p\n", &ci->vfs_inode);
449 spin_lock(&mdsc->cap_delay_lock);
450 ci->i_ceph_flags |= CEPH_I_FLUSH;
451 if (!list_empty(&ci->i_cap_delay_list))
452 list_del_init(&ci->i_cap_delay_list);
453 list_add(&ci->i_cap_delay_list, &mdsc->cap_delay_list);
454 spin_unlock(&mdsc->cap_delay_lock);
455}
456
457/*
458 * Cancel delayed work on cap.
459 *
460 * Caller must hold i_lock.
461 */
462static void __cap_delay_cancel(struct ceph_mds_client *mdsc,
463 struct ceph_inode_info *ci)
464{
465 dout("__cap_delay_cancel %p\n", &ci->vfs_inode);
466 if (list_empty(&ci->i_cap_delay_list))
467 return;
468 spin_lock(&mdsc->cap_delay_lock);
469 list_del_init(&ci->i_cap_delay_list);
470 spin_unlock(&mdsc->cap_delay_lock);
471}
472
473/*
474 * Common issue checks for add_cap, handle_cap_grant.
475 */
476static void __check_cap_issue(struct ceph_inode_info *ci, struct ceph_cap *cap,
477 unsigned issued)
478{
479 unsigned had = __ceph_caps_issued(ci, NULL);
480
481 /*
482 * Each time we receive FILE_CACHE anew, we increment
483 * i_rdcache_gen.
484 */
Sage Weil29625072010-05-27 10:40:43 -0700485 if ((issued & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) &&
486 (had & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) == 0)
Sage Weila8599bd2009-10-06 11:31:12 -0700487 ci->i_rdcache_gen++;
488
489 /*
490 * if we are newly issued FILE_SHARED, clear I_COMPLETE; we
491 * don't know what happened to this directory while we didn't
492 * have the cap.
493 */
494 if ((issued & CEPH_CAP_FILE_SHARED) &&
495 (had & CEPH_CAP_FILE_SHARED) == 0) {
496 ci->i_shared_gen++;
497 if (S_ISDIR(ci->vfs_inode.i_mode)) {
498 dout(" marking %p NOT complete\n", &ci->vfs_inode);
499 ci->i_ceph_flags &= ~CEPH_I_COMPLETE;
500 }
501 }
502}
503
504/*
505 * Add a capability under the given MDS session.
506 *
507 * Caller should hold session snap_rwsem (read) and s_mutex.
508 *
509 * @fmode is the open file mode, if we are opening a file, otherwise
510 * it is < 0. (This is so we can atomically add the cap and add an
511 * open file reference to it.)
512 */
513int ceph_add_cap(struct inode *inode,
514 struct ceph_mds_session *session, u64 cap_id,
515 int fmode, unsigned issued, unsigned wanted,
516 unsigned seq, unsigned mseq, u64 realmino, int flags,
517 struct ceph_cap_reservation *caps_reservation)
518{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700519 struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
Sage Weila8599bd2009-10-06 11:31:12 -0700520 struct ceph_inode_info *ci = ceph_inode(inode);
521 struct ceph_cap *new_cap = NULL;
522 struct ceph_cap *cap;
523 int mds = session->s_mds;
524 int actual_wanted;
525
526 dout("add_cap %p mds%d cap %llx %s seq %d\n", inode,
527 session->s_mds, cap_id, ceph_cap_string(issued), seq);
528
529 /*
530 * If we are opening the file, include file mode wanted bits
531 * in wanted.
532 */
533 if (fmode >= 0)
534 wanted |= ceph_caps_for_mode(fmode);
535
536retry:
537 spin_lock(&inode->i_lock);
538 cap = __get_cap_for_mds(ci, mds);
539 if (!cap) {
540 if (new_cap) {
541 cap = new_cap;
542 new_cap = NULL;
543 } else {
544 spin_unlock(&inode->i_lock);
Yehuda Sadeh37151662010-06-17 16:16:12 -0700545 new_cap = get_cap(mdsc, caps_reservation);
Sage Weila8599bd2009-10-06 11:31:12 -0700546 if (new_cap == NULL)
547 return -ENOMEM;
548 goto retry;
549 }
550
551 cap->issued = 0;
552 cap->implemented = 0;
553 cap->mds = mds;
554 cap->mds_wanted = 0;
555
556 cap->ci = ci;
557 __insert_cap_node(ci, cap);
558
559 /* clear out old exporting info? (i.e. on cap import) */
560 if (ci->i_cap_exporting_mds == mds) {
561 ci->i_cap_exporting_issued = 0;
562 ci->i_cap_exporting_mseq = 0;
563 ci->i_cap_exporting_mds = -1;
564 }
565
566 /* add to session cap list */
567 cap->session = session;
568 spin_lock(&session->s_cap_lock);
569 list_add_tail(&cap->session_caps, &session->s_caps);
570 session->s_nr_caps++;
571 spin_unlock(&session->s_cap_lock);
572 }
573
574 if (!ci->i_snap_realm) {
575 /*
576 * add this inode to the appropriate snap realm
577 */
578 struct ceph_snap_realm *realm = ceph_lookup_snap_realm(mdsc,
579 realmino);
580 if (realm) {
581 ceph_get_snap_realm(mdsc, realm);
582 spin_lock(&realm->inodes_with_caps_lock);
583 ci->i_snap_realm = realm;
584 list_add(&ci->i_snap_realm_item,
585 &realm->inodes_with_caps);
586 spin_unlock(&realm->inodes_with_caps_lock);
587 } else {
588 pr_err("ceph_add_cap: couldn't find snap realm %llx\n",
589 realmino);
Sage Weilb8cd07e2010-07-16 12:00:02 -0700590 WARN_ON(!realm);
Sage Weila8599bd2009-10-06 11:31:12 -0700591 }
592 }
593
594 __check_cap_issue(ci, cap, issued);
595
596 /*
597 * If we are issued caps we don't want, or the mds' wanted
598 * value appears to be off, queue a check so we'll release
599 * later and/or update the mds wanted value.
600 */
601 actual_wanted = __ceph_caps_wanted(ci);
602 if ((wanted & ~actual_wanted) ||
603 (issued & ~actual_wanted & CEPH_CAP_ANY_WR)) {
604 dout(" issued %s, mds wanted %s, actual %s, queueing\n",
605 ceph_cap_string(issued), ceph_cap_string(wanted),
606 ceph_cap_string(actual_wanted));
607 __cap_delay_requeue(mdsc, ci);
608 }
609
610 if (flags & CEPH_CAP_FLAG_AUTH)
611 ci->i_auth_cap = cap;
612 else if (ci->i_auth_cap == cap)
613 ci->i_auth_cap = NULL;
614
615 dout("add_cap inode %p (%llx.%llx) cap %p %s now %s seq %d mds%d\n",
616 inode, ceph_vinop(inode), cap, ceph_cap_string(issued),
617 ceph_cap_string(issued|cap->issued), seq, mds);
618 cap->cap_id = cap_id;
619 cap->issued = issued;
620 cap->implemented |= issued;
621 cap->mds_wanted |= wanted;
622 cap->seq = seq;
623 cap->issue_seq = seq;
624 cap->mseq = mseq;
Sage Weil685f9a5d2009-11-09 12:05:48 -0800625 cap->cap_gen = session->s_cap_gen;
Sage Weila8599bd2009-10-06 11:31:12 -0700626
627 if (fmode >= 0)
628 __ceph_get_fmode(ci, fmode);
629 spin_unlock(&inode->i_lock);
Yehuda Sadeh03066f22010-07-27 13:11:08 -0700630 wake_up_all(&ci->i_cap_wq);
Sage Weila8599bd2009-10-06 11:31:12 -0700631 return 0;
632}
633
634/*
635 * Return true if cap has not timed out and belongs to the current
636 * generation of the MDS session (i.e. has not gone 'stale' due to
637 * us losing touch with the mds).
638 */
639static int __cap_is_valid(struct ceph_cap *cap)
640{
641 unsigned long ttl;
Sage Weilcdac8302009-11-10 16:02:23 -0800642 u32 gen;
Sage Weila8599bd2009-10-06 11:31:12 -0700643
644 spin_lock(&cap->session->s_cap_lock);
645 gen = cap->session->s_cap_gen;
646 ttl = cap->session->s_cap_ttl;
647 spin_unlock(&cap->session->s_cap_lock);
648
Sage Weil685f9a5d2009-11-09 12:05:48 -0800649 if (cap->cap_gen < gen || time_after_eq(jiffies, ttl)) {
Sage Weila8599bd2009-10-06 11:31:12 -0700650 dout("__cap_is_valid %p cap %p issued %s "
651 "but STALE (gen %u vs %u)\n", &cap->ci->vfs_inode,
Sage Weil685f9a5d2009-11-09 12:05:48 -0800652 cap, ceph_cap_string(cap->issued), cap->cap_gen, gen);
Sage Weila8599bd2009-10-06 11:31:12 -0700653 return 0;
654 }
655
656 return 1;
657}
658
659/*
660 * Return set of valid cap bits issued to us. Note that caps time
661 * out, and may be invalidated in bulk if the client session times out
662 * and session->s_cap_gen is bumped.
663 */
664int __ceph_caps_issued(struct ceph_inode_info *ci, int *implemented)
665{
Sage Weil7af8f1e2010-03-01 15:17:34 -0800666 int have = ci->i_snap_caps | ci->i_cap_exporting_issued;
Sage Weila8599bd2009-10-06 11:31:12 -0700667 struct ceph_cap *cap;
668 struct rb_node *p;
669
670 if (implemented)
671 *implemented = 0;
672 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
673 cap = rb_entry(p, struct ceph_cap, ci_node);
674 if (!__cap_is_valid(cap))
675 continue;
676 dout("__ceph_caps_issued %p cap %p issued %s\n",
677 &ci->vfs_inode, cap, ceph_cap_string(cap->issued));
678 have |= cap->issued;
679 if (implemented)
680 *implemented |= cap->implemented;
681 }
682 return have;
683}
684
685/*
686 * Get cap bits issued by caps other than @ocap
687 */
688int __ceph_caps_issued_other(struct ceph_inode_info *ci, struct ceph_cap *ocap)
689{
690 int have = ci->i_snap_caps;
691 struct ceph_cap *cap;
692 struct rb_node *p;
693
694 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
695 cap = rb_entry(p, struct ceph_cap, ci_node);
696 if (cap == ocap)
697 continue;
698 if (!__cap_is_valid(cap))
699 continue;
700 have |= cap->issued;
701 }
702 return have;
703}
704
705/*
706 * Move a cap to the end of the LRU (oldest caps at list head, newest
707 * at list tail).
708 */
709static void __touch_cap(struct ceph_cap *cap)
710{
711 struct ceph_mds_session *s = cap->session;
712
Sage Weila8599bd2009-10-06 11:31:12 -0700713 spin_lock(&s->s_cap_lock);
Sage Weil7c1332b2010-02-16 11:39:45 -0800714 if (s->s_cap_iterator == NULL) {
Sage Weil5dacf092009-12-21 20:40:34 -0800715 dout("__touch_cap %p cap %p mds%d\n", &cap->ci->vfs_inode, cap,
716 s->s_mds);
717 list_move_tail(&cap->session_caps, &s->s_caps);
718 } else {
719 dout("__touch_cap %p cap %p mds%d NOP, iterating over caps\n",
720 &cap->ci->vfs_inode, cap, s->s_mds);
721 }
Sage Weila8599bd2009-10-06 11:31:12 -0700722 spin_unlock(&s->s_cap_lock);
723}
724
725/*
726 * Check if we hold the given mask. If so, move the cap(s) to the
727 * front of their respective LRUs. (This is the preferred way for
728 * callers to check for caps they want.)
729 */
730int __ceph_caps_issued_mask(struct ceph_inode_info *ci, int mask, int touch)
731{
732 struct ceph_cap *cap;
733 struct rb_node *p;
734 int have = ci->i_snap_caps;
735
736 if ((have & mask) == mask) {
737 dout("__ceph_caps_issued_mask %p snap issued %s"
738 " (mask %s)\n", &ci->vfs_inode,
739 ceph_cap_string(have),
740 ceph_cap_string(mask));
741 return 1;
742 }
743
744 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
745 cap = rb_entry(p, struct ceph_cap, ci_node);
746 if (!__cap_is_valid(cap))
747 continue;
748 if ((cap->issued & mask) == mask) {
749 dout("__ceph_caps_issued_mask %p cap %p issued %s"
750 " (mask %s)\n", &ci->vfs_inode, cap,
751 ceph_cap_string(cap->issued),
752 ceph_cap_string(mask));
753 if (touch)
754 __touch_cap(cap);
755 return 1;
756 }
757
758 /* does a combination of caps satisfy mask? */
759 have |= cap->issued;
760 if ((have & mask) == mask) {
761 dout("__ceph_caps_issued_mask %p combo issued %s"
762 " (mask %s)\n", &ci->vfs_inode,
763 ceph_cap_string(cap->issued),
764 ceph_cap_string(mask));
765 if (touch) {
766 struct rb_node *q;
767
768 /* touch this + preceeding caps */
769 __touch_cap(cap);
770 for (q = rb_first(&ci->i_caps); q != p;
771 q = rb_next(q)) {
772 cap = rb_entry(q, struct ceph_cap,
773 ci_node);
774 if (!__cap_is_valid(cap))
775 continue;
776 __touch_cap(cap);
777 }
778 }
779 return 1;
780 }
781 }
782
783 return 0;
784}
785
786/*
787 * Return true if mask caps are currently being revoked by an MDS.
788 */
789int ceph_caps_revoking(struct ceph_inode_info *ci, int mask)
790{
791 struct inode *inode = &ci->vfs_inode;
792 struct ceph_cap *cap;
793 struct rb_node *p;
794 int ret = 0;
795
796 spin_lock(&inode->i_lock);
797 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
798 cap = rb_entry(p, struct ceph_cap, ci_node);
799 if (__cap_is_valid(cap) &&
800 (cap->implemented & ~cap->issued & mask)) {
801 ret = 1;
802 break;
803 }
804 }
805 spin_unlock(&inode->i_lock);
806 dout("ceph_caps_revoking %p %s = %d\n", inode,
807 ceph_cap_string(mask), ret);
808 return ret;
809}
810
811int __ceph_caps_used(struct ceph_inode_info *ci)
812{
813 int used = 0;
814 if (ci->i_pin_ref)
815 used |= CEPH_CAP_PIN;
816 if (ci->i_rd_ref)
817 used |= CEPH_CAP_FILE_RD;
Sage Weila43fb732010-09-17 09:54:08 -0700818 if (ci->i_rdcache_ref || ci->vfs_inode.i_data.nrpages)
Sage Weila8599bd2009-10-06 11:31:12 -0700819 used |= CEPH_CAP_FILE_CACHE;
820 if (ci->i_wr_ref)
821 used |= CEPH_CAP_FILE_WR;
822 if (ci->i_wrbuffer_ref)
823 used |= CEPH_CAP_FILE_BUFFER;
824 return used;
825}
826
827/*
828 * wanted, by virtue of open file modes
829 */
830int __ceph_caps_file_wanted(struct ceph_inode_info *ci)
831{
832 int want = 0;
833 int mode;
Sage Weil33caad32010-05-26 14:31:27 -0700834 for (mode = 0; mode < CEPH_FILE_MODE_NUM; mode++)
Sage Weila8599bd2009-10-06 11:31:12 -0700835 if (ci->i_nr_by_mode[mode])
836 want |= ceph_caps_for_mode(mode);
837 return want;
838}
839
840/*
841 * Return caps we have registered with the MDS(s) as 'wanted'.
842 */
843int __ceph_caps_mds_wanted(struct ceph_inode_info *ci)
844{
845 struct ceph_cap *cap;
846 struct rb_node *p;
847 int mds_wanted = 0;
848
849 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
850 cap = rb_entry(p, struct ceph_cap, ci_node);
851 if (!__cap_is_valid(cap))
852 continue;
853 mds_wanted |= cap->mds_wanted;
854 }
855 return mds_wanted;
856}
857
858/*
859 * called under i_lock
860 */
861static int __ceph_is_any_caps(struct ceph_inode_info *ci)
862{
863 return !RB_EMPTY_ROOT(&ci->i_caps) || ci->i_cap_exporting_mds >= 0;
864}
865
866/*
Sage Weilf818a732010-05-11 20:56:31 -0700867 * Remove a cap. Take steps to deal with a racing iterate_session_caps.
868 *
Sage Weila6369742010-02-22 13:59:00 -0800869 * caller should hold i_lock.
870 * caller will not hold session s_mutex if called from destroy_inode.
Sage Weila8599bd2009-10-06 11:31:12 -0700871 */
Sage Weil7c1332b2010-02-16 11:39:45 -0800872void __ceph_remove_cap(struct ceph_cap *cap)
Sage Weila8599bd2009-10-06 11:31:12 -0700873{
874 struct ceph_mds_session *session = cap->session;
875 struct ceph_inode_info *ci = cap->ci;
Cheng Renquan640ef792010-03-26 17:40:33 +0800876 struct ceph_mds_client *mdsc =
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700877 ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc;
Sage Weilf818a732010-05-11 20:56:31 -0700878 int removed = 0;
Sage Weila8599bd2009-10-06 11:31:12 -0700879
880 dout("__ceph_remove_cap %p from %p\n", cap, &ci->vfs_inode);
881
Sage Weil7c1332b2010-02-16 11:39:45 -0800882 /* remove from session list */
883 spin_lock(&session->s_cap_lock);
884 if (session->s_cap_iterator == cap) {
885 /* not yet, we are iterating over this very cap */
886 dout("__ceph_remove_cap delaying %p removal from session %p\n",
887 cap, cap->session);
888 } else {
889 list_del_init(&cap->session_caps);
890 session->s_nr_caps--;
891 cap->session = NULL;
Sage Weilf818a732010-05-11 20:56:31 -0700892 removed = 1;
Sage Weil7c1332b2010-02-16 11:39:45 -0800893 }
Sage Weilf818a732010-05-11 20:56:31 -0700894 /* protect backpointer with s_cap_lock: see iterate_session_caps */
895 cap->ci = NULL;
Sage Weil7c1332b2010-02-16 11:39:45 -0800896 spin_unlock(&session->s_cap_lock);
897
Sage Weilf818a732010-05-11 20:56:31 -0700898 /* remove from inode list */
899 rb_erase(&cap->ci_node, &ci->i_caps);
900 if (ci->i_auth_cap == cap)
901 ci->i_auth_cap = NULL;
902
903 if (removed)
Yehuda Sadeh37151662010-06-17 16:16:12 -0700904 ceph_put_cap(mdsc, cap);
Sage Weila8599bd2009-10-06 11:31:12 -0700905
906 if (!__ceph_is_any_caps(ci) && ci->i_snap_realm) {
907 struct ceph_snap_realm *realm = ci->i_snap_realm;
908 spin_lock(&realm->inodes_with_caps_lock);
909 list_del_init(&ci->i_snap_realm_item);
910 ci->i_snap_realm_counter++;
911 ci->i_snap_realm = NULL;
912 spin_unlock(&realm->inodes_with_caps_lock);
913 ceph_put_snap_realm(mdsc, realm);
914 }
915 if (!__ceph_is_any_real_caps(ci))
916 __cap_delay_cancel(mdsc, ci);
917}
918
919/*
920 * Build and send a cap message to the given MDS.
921 *
922 * Caller should be holding s_mutex.
923 */
924static int send_cap_msg(struct ceph_mds_session *session,
925 u64 ino, u64 cid, int op,
926 int caps, int wanted, int dirty,
927 u32 seq, u64 flush_tid, u32 issue_seq, u32 mseq,
928 u64 size, u64 max_size,
929 struct timespec *mtime, struct timespec *atime,
930 u64 time_warp_seq,
931 uid_t uid, gid_t gid, mode_t mode,
932 u64 xattr_version,
933 struct ceph_buffer *xattrs_buf,
934 u64 follows)
935{
936 struct ceph_mds_caps *fc;
937 struct ceph_msg *msg;
938
939 dout("send_cap_msg %s %llx %llx caps %s wanted %s dirty %s"
940 " seq %u/%u mseq %u follows %lld size %llu/%llu"
941 " xattr_ver %llu xattr_len %d\n", ceph_cap_op_name(op),
942 cid, ino, ceph_cap_string(caps), ceph_cap_string(wanted),
943 ceph_cap_string(dirty),
944 seq, issue_seq, mseq, follows, size, max_size,
945 xattr_version, xattrs_buf ? (int)xattrs_buf->vec.iov_len : 0);
946
Yehuda Sadeh34d23762010-04-06 14:33:58 -0700947 msg = ceph_msg_new(CEPH_MSG_CLIENT_CAPS, sizeof(*fc), GFP_NOFS);
Sage Weila79832f2010-04-01 16:06:19 -0700948 if (!msg)
949 return -ENOMEM;
Sage Weila8599bd2009-10-06 11:31:12 -0700950
Sage Weil6df058c2009-12-22 11:24:33 -0800951 msg->hdr.tid = cpu_to_le64(flush_tid);
Sage Weila8599bd2009-10-06 11:31:12 -0700952
Sage Weil6df058c2009-12-22 11:24:33 -0800953 fc = msg->front.iov_base;
Sage Weila8599bd2009-10-06 11:31:12 -0700954 memset(fc, 0, sizeof(*fc));
955
956 fc->cap_id = cpu_to_le64(cid);
957 fc->op = cpu_to_le32(op);
958 fc->seq = cpu_to_le32(seq);
Sage Weila8599bd2009-10-06 11:31:12 -0700959 fc->issue_seq = cpu_to_le32(issue_seq);
960 fc->migrate_seq = cpu_to_le32(mseq);
961 fc->caps = cpu_to_le32(caps);
962 fc->wanted = cpu_to_le32(wanted);
963 fc->dirty = cpu_to_le32(dirty);
964 fc->ino = cpu_to_le64(ino);
965 fc->snap_follows = cpu_to_le64(follows);
966
967 fc->size = cpu_to_le64(size);
968 fc->max_size = cpu_to_le64(max_size);
969 if (mtime)
970 ceph_encode_timespec(&fc->mtime, mtime);
971 if (atime)
972 ceph_encode_timespec(&fc->atime, atime);
973 fc->time_warp_seq = cpu_to_le32(time_warp_seq);
974
975 fc->uid = cpu_to_le32(uid);
976 fc->gid = cpu_to_le32(gid);
977 fc->mode = cpu_to_le32(mode);
978
979 fc->xattr_version = cpu_to_le64(xattr_version);
980 if (xattrs_buf) {
981 msg->middle = ceph_buffer_get(xattrs_buf);
982 fc->xattr_len = cpu_to_le32(xattrs_buf->vec.iov_len);
983 msg->hdr.middle_len = cpu_to_le32(xattrs_buf->vec.iov_len);
984 }
985
986 ceph_con_send(&session->s_con, msg);
987 return 0;
988}
989
Sage Weil3d7ded42010-06-09 16:47:10 -0700990static void __queue_cap_release(struct ceph_mds_session *session,
991 u64 ino, u64 cap_id, u32 migrate_seq,
992 u32 issue_seq)
993{
994 struct ceph_msg *msg;
995 struct ceph_mds_cap_release *head;
996 struct ceph_mds_cap_item *item;
997
998 spin_lock(&session->s_cap_lock);
999 BUG_ON(!session->s_num_cap_releases);
1000 msg = list_first_entry(&session->s_cap_releases,
1001 struct ceph_msg, list_head);
1002
1003 dout(" adding %llx release to mds%d msg %p (%d left)\n",
1004 ino, session->s_mds, msg, session->s_num_cap_releases);
1005
1006 BUG_ON(msg->front.iov_len + sizeof(*item) > PAGE_CACHE_SIZE);
1007 head = msg->front.iov_base;
1008 head->num = cpu_to_le32(le32_to_cpu(head->num) + 1);
1009 item = msg->front.iov_base + msg->front.iov_len;
1010 item->ino = cpu_to_le64(ino);
1011 item->cap_id = cpu_to_le64(cap_id);
1012 item->migrate_seq = cpu_to_le32(migrate_seq);
1013 item->seq = cpu_to_le32(issue_seq);
1014
1015 session->s_num_cap_releases--;
1016
1017 msg->front.iov_len += sizeof(*item);
1018 if (le32_to_cpu(head->num) == CEPH_CAPS_PER_RELEASE) {
1019 dout(" release msg %p full\n", msg);
1020 list_move_tail(&msg->list_head, &session->s_cap_releases_done);
1021 } else {
1022 dout(" release msg %p at %d/%d (%d)\n", msg,
1023 (int)le32_to_cpu(head->num),
1024 (int)CEPH_CAPS_PER_RELEASE,
1025 (int)msg->front.iov_len);
1026 }
1027 spin_unlock(&session->s_cap_lock);
1028}
1029
Sage Weila8599bd2009-10-06 11:31:12 -07001030/*
Sage Weila6369742010-02-22 13:59:00 -08001031 * Queue cap releases when an inode is dropped from our cache. Since
1032 * inode is about to be destroyed, there is no need for i_lock.
Sage Weila8599bd2009-10-06 11:31:12 -07001033 */
1034void ceph_queue_caps_release(struct inode *inode)
1035{
1036 struct ceph_inode_info *ci = ceph_inode(inode);
1037 struct rb_node *p;
1038
Sage Weila8599bd2009-10-06 11:31:12 -07001039 p = rb_first(&ci->i_caps);
1040 while (p) {
1041 struct ceph_cap *cap = rb_entry(p, struct ceph_cap, ci_node);
1042 struct ceph_mds_session *session = cap->session;
Sage Weila8599bd2009-10-06 11:31:12 -07001043
Sage Weil3d7ded42010-06-09 16:47:10 -07001044 __queue_cap_release(session, ceph_ino(inode), cap->cap_id,
1045 cap->mseq, cap->issue_seq);
Sage Weila8599bd2009-10-06 11:31:12 -07001046 p = rb_next(p);
Sage Weil7c1332b2010-02-16 11:39:45 -08001047 __ceph_remove_cap(cap);
Sage Weila8599bd2009-10-06 11:31:12 -07001048 }
Sage Weila8599bd2009-10-06 11:31:12 -07001049}
1050
1051/*
1052 * Send a cap msg on the given inode. Update our caps state, then
1053 * drop i_lock and send the message.
1054 *
1055 * Make note of max_size reported/requested from mds, revoked caps
1056 * that have now been implemented.
1057 *
1058 * Make half-hearted attempt ot to invalidate page cache if we are
1059 * dropping RDCACHE. Note that this will leave behind locked pages
1060 * that we'll then need to deal with elsewhere.
1061 *
1062 * Return non-zero if delayed release, or we experienced an error
1063 * such that the caller should requeue + retry later.
1064 *
1065 * called with i_lock, then drops it.
1066 * caller should hold snap_rwsem (read), s_mutex.
1067 */
1068static int __send_cap(struct ceph_mds_client *mdsc, struct ceph_cap *cap,
1069 int op, int used, int want, int retain, int flushing,
1070 unsigned *pflush_tid)
1071 __releases(cap->ci->vfs_inode->i_lock)
1072{
1073 struct ceph_inode_info *ci = cap->ci;
1074 struct inode *inode = &ci->vfs_inode;
1075 u64 cap_id = cap->cap_id;
Sage Weil68c28322010-02-09 13:41:47 -08001076 int held, revoking, dropping, keep;
Sage Weila8599bd2009-10-06 11:31:12 -07001077 u64 seq, issue_seq, mseq, time_warp_seq, follows;
1078 u64 size, max_size;
1079 struct timespec mtime, atime;
1080 int wake = 0;
1081 mode_t mode;
1082 uid_t uid;
1083 gid_t gid;
1084 struct ceph_mds_session *session;
1085 u64 xattr_version = 0;
Sage Weil082afec2010-08-22 15:16:41 -07001086 struct ceph_buffer *xattr_blob = NULL;
Sage Weila8599bd2009-10-06 11:31:12 -07001087 int delayed = 0;
1088 u64 flush_tid = 0;
1089 int i;
1090 int ret;
1091
Sage Weil68c28322010-02-09 13:41:47 -08001092 held = cap->issued | cap->implemented;
1093 revoking = cap->implemented & ~cap->issued;
1094 retain &= ~revoking;
1095 dropping = cap->issued & ~retain;
1096
Sage Weila8599bd2009-10-06 11:31:12 -07001097 dout("__send_cap %p cap %p session %p %s -> %s (revoking %s)\n",
1098 inode, cap, cap->session,
1099 ceph_cap_string(held), ceph_cap_string(held & retain),
1100 ceph_cap_string(revoking));
1101 BUG_ON((retain & CEPH_CAP_PIN) == 0);
1102
1103 session = cap->session;
1104
1105 /* don't release wanted unless we've waited a bit. */
1106 if ((ci->i_ceph_flags & CEPH_I_NODELAY) == 0 &&
1107 time_before(jiffies, ci->i_hold_caps_min)) {
1108 dout(" delaying issued %s -> %s, wanted %s -> %s on send\n",
1109 ceph_cap_string(cap->issued),
1110 ceph_cap_string(cap->issued & retain),
1111 ceph_cap_string(cap->mds_wanted),
1112 ceph_cap_string(want));
1113 want |= cap->mds_wanted;
1114 retain |= cap->issued;
1115 delayed = 1;
1116 }
1117 ci->i_ceph_flags &= ~(CEPH_I_NODELAY | CEPH_I_FLUSH);
1118
1119 cap->issued &= retain; /* drop bits we don't want */
1120 if (cap->implemented & ~cap->issued) {
1121 /*
1122 * Wake up any waiters on wanted -> needed transition.
1123 * This is due to the weird transition from buffered
1124 * to sync IO... we need to flush dirty pages _before_
1125 * allowing sync writes to avoid reordering.
1126 */
1127 wake = 1;
1128 }
1129 cap->implemented &= cap->issued | used;
1130 cap->mds_wanted = want;
1131
1132 if (flushing) {
1133 /*
1134 * assign a tid for flush operations so we can avoid
1135 * flush1 -> dirty1 -> flush2 -> flushack1 -> mark
1136 * clean type races. track latest tid for every bit
1137 * so we can handle flush AxFw, flush Fw, and have the
1138 * first ack clean Ax.
1139 */
1140 flush_tid = ++ci->i_cap_flush_last_tid;
1141 if (pflush_tid)
1142 *pflush_tid = flush_tid;
1143 dout(" cap_flush_tid %d\n", (int)flush_tid);
1144 for (i = 0; i < CEPH_CAP_BITS; i++)
1145 if (flushing & (1 << i))
1146 ci->i_cap_flush_tid[i] = flush_tid;
Sage Weil7d8cb262010-08-24 08:44:16 -07001147
1148 follows = ci->i_head_snapc->seq;
1149 } else {
1150 follows = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07001151 }
1152
1153 keep = cap->implemented;
1154 seq = cap->seq;
1155 issue_seq = cap->issue_seq;
1156 mseq = cap->mseq;
1157 size = inode->i_size;
1158 ci->i_reported_size = size;
1159 max_size = ci->i_wanted_max_size;
1160 ci->i_requested_max_size = max_size;
1161 mtime = inode->i_mtime;
1162 atime = inode->i_atime;
1163 time_warp_seq = ci->i_time_warp_seq;
Sage Weila8599bd2009-10-06 11:31:12 -07001164 uid = inode->i_uid;
1165 gid = inode->i_gid;
1166 mode = inode->i_mode;
1167
Sage Weil082afec2010-08-22 15:16:41 -07001168 if (flushing & CEPH_CAP_XATTR_EXCL) {
Sage Weila8599bd2009-10-06 11:31:12 -07001169 __ceph_build_xattrs_blob(ci);
Sage Weil082afec2010-08-22 15:16:41 -07001170 xattr_blob = ci->i_xattrs.blob;
1171 xattr_version = ci->i_xattrs.version;
Sage Weila8599bd2009-10-06 11:31:12 -07001172 }
1173
1174 spin_unlock(&inode->i_lock);
1175
Sage Weila8599bd2009-10-06 11:31:12 -07001176 ret = send_cap_msg(session, ceph_vino(inode).ino, cap_id,
1177 op, keep, want, flushing, seq, flush_tid, issue_seq, mseq,
1178 size, max_size, &mtime, &atime, time_warp_seq,
Sage Weil082afec2010-08-22 15:16:41 -07001179 uid, gid, mode, xattr_version, xattr_blob,
Sage Weila8599bd2009-10-06 11:31:12 -07001180 follows);
1181 if (ret < 0) {
1182 dout("error sending cap msg, must requeue %p\n", inode);
1183 delayed = 1;
1184 }
1185
1186 if (wake)
Yehuda Sadeh03066f22010-07-27 13:11:08 -07001187 wake_up_all(&ci->i_cap_wq);
Sage Weila8599bd2009-10-06 11:31:12 -07001188
1189 return delayed;
1190}
1191
1192/*
1193 * When a snapshot is taken, clients accumulate dirty metadata on
1194 * inodes with capabilities in ceph_cap_snaps to describe the file
1195 * state at the time the snapshot was taken. This must be flushed
1196 * asynchronously back to the MDS once sync writes complete and dirty
1197 * data is written out.
1198 *
Sage Weile8351242010-09-17 08:03:08 -07001199 * Unless @again is true, skip cap_snaps that were already sent to
1200 * the MDS (i.e., during this session).
1201 *
Sage Weila8599bd2009-10-06 11:31:12 -07001202 * Called under i_lock. Takes s_mutex as needed.
1203 */
1204void __ceph_flush_snaps(struct ceph_inode_info *ci,
Sage Weile8351242010-09-17 08:03:08 -07001205 struct ceph_mds_session **psession,
1206 int again)
Yehuda Sadehcd84db62010-06-11 16:58:48 -07001207 __releases(ci->vfs_inode->i_lock)
1208 __acquires(ci->vfs_inode->i_lock)
Sage Weila8599bd2009-10-06 11:31:12 -07001209{
1210 struct inode *inode = &ci->vfs_inode;
1211 int mds;
1212 struct ceph_cap_snap *capsnap;
1213 u32 mseq;
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001214 struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
Sage Weila8599bd2009-10-06 11:31:12 -07001215 struct ceph_mds_session *session = NULL; /* if session != NULL, we hold
1216 session->s_mutex */
1217 u64 next_follows = 0; /* keep track of how far we've gotten through the
1218 i_cap_snaps list, and skip these entries next time
1219 around to avoid an infinite loop */
1220
1221 if (psession)
1222 session = *psession;
1223
1224 dout("__flush_snaps %p\n", inode);
1225retry:
1226 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
1227 /* avoid an infiniute loop after retry */
1228 if (capsnap->follows < next_follows)
1229 continue;
1230 /*
1231 * we need to wait for sync writes to complete and for dirty
1232 * pages to be written out.
1233 */
1234 if (capsnap->dirty_pages || capsnap->writing)
Sage Weilcfc0bf62010-09-14 15:50:59 -07001235 break;
Sage Weila8599bd2009-10-06 11:31:12 -07001236
Sage Weil819ccbf2010-04-01 09:33:46 -07001237 /*
1238 * if cap writeback already occurred, we should have dropped
1239 * the capsnap in ceph_put_wrbuffer_cap_refs.
1240 */
1241 BUG_ON(capsnap->dirty == 0);
1242
Sage Weila8599bd2009-10-06 11:31:12 -07001243 /* pick mds, take s_mutex */
Sage Weilca81f3f2010-06-10 14:21:36 -07001244 if (ci->i_auth_cap == NULL) {
1245 dout("no auth cap (migrating?), doing nothing\n");
1246 goto out;
1247 }
Sage Weile8351242010-09-17 08:03:08 -07001248
1249 /* only flush each capsnap once */
1250 if (!again && !list_empty(&capsnap->flushing_item)) {
1251 dout("already flushed %p, skipping\n", capsnap);
1252 continue;
1253 }
1254
Sage Weilca81f3f2010-06-10 14:21:36 -07001255 mds = ci->i_auth_cap->session->s_mds;
1256 mseq = ci->i_auth_cap->mseq;
1257
Sage Weila8599bd2009-10-06 11:31:12 -07001258 if (session && session->s_mds != mds) {
1259 dout("oops, wrong session %p mutex\n", session);
1260 mutex_unlock(&session->s_mutex);
1261 ceph_put_mds_session(session);
1262 session = NULL;
1263 }
1264 if (!session) {
1265 spin_unlock(&inode->i_lock);
1266 mutex_lock(&mdsc->mutex);
1267 session = __ceph_lookup_mds_session(mdsc, mds);
1268 mutex_unlock(&mdsc->mutex);
1269 if (session) {
1270 dout("inverting session/ino locks on %p\n",
1271 session);
1272 mutex_lock(&session->s_mutex);
1273 }
1274 /*
1275 * if session == NULL, we raced against a cap
Sage Weilca81f3f2010-06-10 14:21:36 -07001276 * deletion or migration. retry, and we'll
1277 * get a better @mds value next time.
Sage Weila8599bd2009-10-06 11:31:12 -07001278 */
1279 spin_lock(&inode->i_lock);
1280 goto retry;
1281 }
1282
1283 capsnap->flush_tid = ++ci->i_cap_flush_last_tid;
1284 atomic_inc(&capsnap->nref);
1285 if (!list_empty(&capsnap->flushing_item))
1286 list_del_init(&capsnap->flushing_item);
1287 list_add_tail(&capsnap->flushing_item,
1288 &session->s_cap_snaps_flushing);
1289 spin_unlock(&inode->i_lock);
1290
Sage Weilcfc0bf62010-09-14 15:50:59 -07001291 dout("flush_snaps %p cap_snap %p follows %lld tid %llu\n",
1292 inode, capsnap, capsnap->follows, capsnap->flush_tid);
Sage Weila8599bd2009-10-06 11:31:12 -07001293 send_cap_msg(session, ceph_vino(inode).ino, 0,
1294 CEPH_CAP_OP_FLUSHSNAP, capsnap->issued, 0,
1295 capsnap->dirty, 0, capsnap->flush_tid, 0, mseq,
1296 capsnap->size, 0,
1297 &capsnap->mtime, &capsnap->atime,
1298 capsnap->time_warp_seq,
1299 capsnap->uid, capsnap->gid, capsnap->mode,
Sage Weil4a625be2010-08-22 15:03:56 -07001300 capsnap->xattr_version, capsnap->xattr_blob,
Sage Weila8599bd2009-10-06 11:31:12 -07001301 capsnap->follows);
1302
1303 next_follows = capsnap->follows + 1;
1304 ceph_put_cap_snap(capsnap);
1305
1306 spin_lock(&inode->i_lock);
1307 goto retry;
1308 }
1309
1310 /* we flushed them all; remove this inode from the queue */
1311 spin_lock(&mdsc->snap_flush_lock);
1312 list_del_init(&ci->i_snap_flush_item);
1313 spin_unlock(&mdsc->snap_flush_lock);
1314
Sage Weilca81f3f2010-06-10 14:21:36 -07001315out:
Sage Weila8599bd2009-10-06 11:31:12 -07001316 if (psession)
1317 *psession = session;
1318 else if (session) {
1319 mutex_unlock(&session->s_mutex);
1320 ceph_put_mds_session(session);
1321 }
1322}
1323
1324static void ceph_flush_snaps(struct ceph_inode_info *ci)
1325{
1326 struct inode *inode = &ci->vfs_inode;
1327
1328 spin_lock(&inode->i_lock);
Sage Weile8351242010-09-17 08:03:08 -07001329 __ceph_flush_snaps(ci, NULL, 0);
Sage Weila8599bd2009-10-06 11:31:12 -07001330 spin_unlock(&inode->i_lock);
1331}
1332
1333/*
Sage Weil76e3b392009-10-15 18:13:53 -07001334 * Mark caps dirty. If inode is newly dirty, add to the global dirty
1335 * list.
1336 */
1337void __ceph_mark_dirty_caps(struct ceph_inode_info *ci, int mask)
1338{
Cheng Renquan640ef792010-03-26 17:40:33 +08001339 struct ceph_mds_client *mdsc =
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001340 ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc;
Sage Weil76e3b392009-10-15 18:13:53 -07001341 struct inode *inode = &ci->vfs_inode;
1342 int was = ci->i_dirty_caps;
1343 int dirty = 0;
1344
1345 dout("__mark_dirty_caps %p %s dirty %s -> %s\n", &ci->vfs_inode,
1346 ceph_cap_string(mask), ceph_cap_string(was),
1347 ceph_cap_string(was | mask));
1348 ci->i_dirty_caps |= mask;
1349 if (was == 0) {
Sage Weil7d8cb262010-08-24 08:44:16 -07001350 if (!ci->i_head_snapc)
1351 ci->i_head_snapc = ceph_get_snap_context(
1352 ci->i_snap_realm->cached_context);
1353 dout(" inode %p now dirty snapc %p\n", &ci->vfs_inode,
1354 ci->i_head_snapc);
Sage Weil76e3b392009-10-15 18:13:53 -07001355 BUG_ON(!list_empty(&ci->i_dirty_item));
1356 spin_lock(&mdsc->cap_dirty_lock);
1357 list_add(&ci->i_dirty_item, &mdsc->cap_dirty);
1358 spin_unlock(&mdsc->cap_dirty_lock);
1359 if (ci->i_flushing_caps == 0) {
1360 igrab(inode);
1361 dirty |= I_DIRTY_SYNC;
1362 }
1363 }
1364 BUG_ON(list_empty(&ci->i_dirty_item));
1365 if (((was | ci->i_flushing_caps) & CEPH_CAP_FILE_BUFFER) &&
1366 (mask & CEPH_CAP_FILE_BUFFER))
1367 dirty |= I_DIRTY_DATASYNC;
1368 if (dirty)
1369 __mark_inode_dirty(inode, dirty);
1370 __cap_delay_requeue(mdsc, ci);
1371}
1372
1373/*
Sage Weila8599bd2009-10-06 11:31:12 -07001374 * Add dirty inode to the flushing list. Assigned a seq number so we
1375 * can wait for caps to flush without starving.
Sage Weilcdc35f92009-10-14 14:24:19 -07001376 *
1377 * Called under i_lock.
Sage Weila8599bd2009-10-06 11:31:12 -07001378 */
Sage Weilcdc35f92009-10-14 14:24:19 -07001379static int __mark_caps_flushing(struct inode *inode,
Sage Weila8599bd2009-10-06 11:31:12 -07001380 struct ceph_mds_session *session)
1381{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001382 struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
Sage Weila8599bd2009-10-06 11:31:12 -07001383 struct ceph_inode_info *ci = ceph_inode(inode);
Sage Weilcdc35f92009-10-14 14:24:19 -07001384 int flushing;
Sage Weil50b885b2009-12-01 14:12:07 -08001385
Sage Weilcdc35f92009-10-14 14:24:19 -07001386 BUG_ON(ci->i_dirty_caps == 0);
Sage Weila8599bd2009-10-06 11:31:12 -07001387 BUG_ON(list_empty(&ci->i_dirty_item));
Sage Weilcdc35f92009-10-14 14:24:19 -07001388
1389 flushing = ci->i_dirty_caps;
1390 dout("__mark_caps_flushing flushing %s, flushing_caps %s -> %s\n",
1391 ceph_cap_string(flushing),
1392 ceph_cap_string(ci->i_flushing_caps),
1393 ceph_cap_string(ci->i_flushing_caps | flushing));
1394 ci->i_flushing_caps |= flushing;
1395 ci->i_dirty_caps = 0;
Sage Weilafcdaea2009-10-14 14:27:38 -07001396 dout(" inode %p now !dirty\n", inode);
Sage Weilcdc35f92009-10-14 14:24:19 -07001397
Sage Weila8599bd2009-10-06 11:31:12 -07001398 spin_lock(&mdsc->cap_dirty_lock);
Sage Weilafcdaea2009-10-14 14:27:38 -07001399 list_del_init(&ci->i_dirty_item);
1400
1401 ci->i_cap_flush_seq = ++mdsc->cap_flush_seq;
Sage Weila8599bd2009-10-06 11:31:12 -07001402 if (list_empty(&ci->i_flushing_item)) {
1403 list_add_tail(&ci->i_flushing_item, &session->s_cap_flushing);
1404 mdsc->num_cap_flushing++;
Sage Weilafcdaea2009-10-14 14:27:38 -07001405 dout(" inode %p now flushing seq %lld\n", inode,
1406 ci->i_cap_flush_seq);
1407 } else {
1408 list_move_tail(&ci->i_flushing_item, &session->s_cap_flushing);
1409 dout(" inode %p now flushing (more) seq %lld\n", inode,
Sage Weila8599bd2009-10-06 11:31:12 -07001410 ci->i_cap_flush_seq);
1411 }
1412 spin_unlock(&mdsc->cap_dirty_lock);
Sage Weilcdc35f92009-10-14 14:24:19 -07001413
1414 return flushing;
Sage Weila8599bd2009-10-06 11:31:12 -07001415}
1416
1417/*
Sage Weil5ecad6f2010-02-17 10:43:37 -08001418 * try to invalidate mapping pages without blocking.
1419 */
Sage Weil5ecad6f2010-02-17 10:43:37 -08001420static int try_nonblocking_invalidate(struct inode *inode)
1421{
1422 struct ceph_inode_info *ci = ceph_inode(inode);
1423 u32 invalidating_gen = ci->i_rdcache_gen;
1424
1425 spin_unlock(&inode->i_lock);
1426 invalidate_mapping_pages(&inode->i_data, 0, -1);
1427 spin_lock(&inode->i_lock);
1428
Sage Weil18a38192010-09-17 10:46:44 -07001429 if (inode->i_data.nrpages == 0 &&
Sage Weil5ecad6f2010-02-17 10:43:37 -08001430 invalidating_gen == ci->i_rdcache_gen) {
1431 /* success. */
1432 dout("try_nonblocking_invalidate %p success\n", inode);
Sage Weilcd045cb2010-11-04 11:05:05 -07001433 /* save any racing async invalidate some trouble */
1434 ci->i_rdcache_revoking = ci->i_rdcache_gen - 1;
Sage Weil5ecad6f2010-02-17 10:43:37 -08001435 return 0;
1436 }
1437 dout("try_nonblocking_invalidate %p failed\n", inode);
1438 return -1;
1439}
1440
1441/*
Sage Weila8599bd2009-10-06 11:31:12 -07001442 * Swiss army knife function to examine currently used and wanted
1443 * versus held caps. Release, flush, ack revoked caps to mds as
1444 * appropriate.
1445 *
1446 * CHECK_CAPS_NODELAY - caller is delayed work and we should not delay
1447 * cap release further.
1448 * CHECK_CAPS_AUTHONLY - we should only check the auth cap
1449 * CHECK_CAPS_FLUSH - we should flush any dirty caps immediately, without
1450 * further delay.
1451 */
1452void ceph_check_caps(struct ceph_inode_info *ci, int flags,
1453 struct ceph_mds_session *session)
1454{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001455 struct ceph_fs_client *fsc = ceph_inode_to_client(&ci->vfs_inode);
1456 struct ceph_mds_client *mdsc = fsc->mdsc;
Sage Weila8599bd2009-10-06 11:31:12 -07001457 struct inode *inode = &ci->vfs_inode;
1458 struct ceph_cap *cap;
1459 int file_wanted, used;
1460 int took_snap_rwsem = 0; /* true if mdsc->snap_rwsem held */
Sage Weilcbd03632010-02-09 13:41:18 -08001461 int issued, implemented, want, retain, revoking, flushing = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07001462 int mds = -1; /* keep track of how far we've gone through i_caps list
1463 to avoid an infinite loop on retry */
1464 struct rb_node *p;
1465 int tried_invalidate = 0;
1466 int delayed = 0, sent = 0, force_requeue = 0, num;
Sage Weilcbd03632010-02-09 13:41:18 -08001467 int queue_invalidate = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07001468 int is_delayed = flags & CHECK_CAPS_NODELAY;
1469
1470 /* if we are unmounting, flush any unused caps immediately. */
1471 if (mdsc->stopping)
1472 is_delayed = 1;
1473
1474 spin_lock(&inode->i_lock);
1475
1476 if (ci->i_ceph_flags & CEPH_I_FLUSH)
1477 flags |= CHECK_CAPS_FLUSH;
1478
1479 /* flush snaps first time around only */
1480 if (!list_empty(&ci->i_cap_snaps))
Sage Weile8351242010-09-17 08:03:08 -07001481 __ceph_flush_snaps(ci, &session, 0);
Sage Weila8599bd2009-10-06 11:31:12 -07001482 goto retry_locked;
1483retry:
1484 spin_lock(&inode->i_lock);
1485retry_locked:
1486 file_wanted = __ceph_caps_file_wanted(ci);
1487 used = __ceph_caps_used(ci);
1488 want = file_wanted | used;
Sage Weilcbd03632010-02-09 13:41:18 -08001489 issued = __ceph_caps_issued(ci, &implemented);
1490 revoking = implemented & ~issued;
Sage Weila8599bd2009-10-06 11:31:12 -07001491
1492 retain = want | CEPH_CAP_PIN;
1493 if (!mdsc->stopping && inode->i_nlink > 0) {
1494 if (want) {
1495 retain |= CEPH_CAP_ANY; /* be greedy */
1496 } else {
1497 retain |= CEPH_CAP_ANY_SHARED;
1498 /*
1499 * keep RD only if we didn't have the file open RW,
1500 * because then the mds would revoke it anyway to
1501 * journal max_size=0.
1502 */
1503 if (ci->i_max_size == 0)
1504 retain |= CEPH_CAP_ANY_RD;
1505 }
1506 }
1507
1508 dout("check_caps %p file_want %s used %s dirty %s flushing %s"
Sage Weilcbd03632010-02-09 13:41:18 -08001509 " issued %s revoking %s retain %s %s%s%s\n", inode,
Sage Weila8599bd2009-10-06 11:31:12 -07001510 ceph_cap_string(file_wanted),
1511 ceph_cap_string(used), ceph_cap_string(ci->i_dirty_caps),
1512 ceph_cap_string(ci->i_flushing_caps),
Sage Weilcbd03632010-02-09 13:41:18 -08001513 ceph_cap_string(issued), ceph_cap_string(revoking),
Sage Weila8599bd2009-10-06 11:31:12 -07001514 ceph_cap_string(retain),
1515 (flags & CHECK_CAPS_AUTHONLY) ? " AUTHONLY" : "",
1516 (flags & CHECK_CAPS_NODELAY) ? " NODELAY" : "",
1517 (flags & CHECK_CAPS_FLUSH) ? " FLUSH" : "");
1518
1519 /*
1520 * If we no longer need to hold onto old our caps, and we may
1521 * have cached pages, but don't want them, then try to invalidate.
1522 * If we fail, it's because pages are locked.... try again later.
1523 */
1524 if ((!is_delayed || mdsc->stopping) &&
1525 ci->i_wrbuffer_ref == 0 && /* no dirty pages... */
Sage Weil93afd442010-09-17 08:38:25 -07001526 inode->i_data.nrpages && /* have cached pages */
Sage Weilcbd03632010-02-09 13:41:18 -08001527 (file_wanted == 0 || /* no open files */
Sage Weil29625072010-05-27 10:40:43 -07001528 (revoking & (CEPH_CAP_FILE_CACHE|
1529 CEPH_CAP_FILE_LAZYIO))) && /* or revoking cache */
Sage Weila8599bd2009-10-06 11:31:12 -07001530 !tried_invalidate) {
Sage Weila8599bd2009-10-06 11:31:12 -07001531 dout("check_caps trying to invalidate on %p\n", inode);
Sage Weil5ecad6f2010-02-17 10:43:37 -08001532 if (try_nonblocking_invalidate(inode) < 0) {
Sage Weil29625072010-05-27 10:40:43 -07001533 if (revoking & (CEPH_CAP_FILE_CACHE|
1534 CEPH_CAP_FILE_LAZYIO)) {
Sage Weil5ecad6f2010-02-17 10:43:37 -08001535 dout("check_caps queuing invalidate\n");
1536 queue_invalidate = 1;
1537 ci->i_rdcache_revoking = ci->i_rdcache_gen;
1538 } else {
1539 dout("check_caps failed to invalidate pages\n");
1540 /* we failed to invalidate pages. check these
1541 caps again later. */
1542 force_requeue = 1;
1543 __cap_set_timeouts(mdsc, ci);
1544 }
Sage Weila8599bd2009-10-06 11:31:12 -07001545 }
1546 tried_invalidate = 1;
1547 goto retry_locked;
1548 }
1549
1550 num = 0;
1551 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
1552 cap = rb_entry(p, struct ceph_cap, ci_node);
1553 num++;
1554
1555 /* avoid looping forever */
1556 if (mds >= cap->mds ||
1557 ((flags & CHECK_CAPS_AUTHONLY) && cap != ci->i_auth_cap))
1558 continue;
1559
1560 /* NOTE: no side-effects allowed, until we take s_mutex */
1561
1562 revoking = cap->implemented & ~cap->issued;
1563 if (revoking)
Sage Weilcbd03632010-02-09 13:41:18 -08001564 dout(" mds%d revoking %s\n", cap->mds,
Sage Weila8599bd2009-10-06 11:31:12 -07001565 ceph_cap_string(revoking));
1566
1567 if (cap == ci->i_auth_cap &&
1568 (cap->issued & CEPH_CAP_FILE_WR)) {
1569 /* request larger max_size from MDS? */
1570 if (ci->i_wanted_max_size > ci->i_max_size &&
1571 ci->i_wanted_max_size > ci->i_requested_max_size) {
1572 dout("requesting new max_size\n");
1573 goto ack;
1574 }
1575
1576 /* approaching file_max? */
1577 if ((inode->i_size << 1) >= ci->i_max_size &&
1578 (ci->i_reported_size << 1) < ci->i_max_size) {
1579 dout("i_size approaching max_size\n");
1580 goto ack;
1581 }
1582 }
1583 /* flush anything dirty? */
1584 if (cap == ci->i_auth_cap && (flags & CHECK_CAPS_FLUSH) &&
1585 ci->i_dirty_caps) {
1586 dout("flushing dirty caps\n");
1587 goto ack;
1588 }
1589
1590 /* completed revocation? going down and there are no caps? */
1591 if (revoking && (revoking & used) == 0) {
1592 dout("completed revocation of %s\n",
1593 ceph_cap_string(cap->implemented & ~cap->issued));
1594 goto ack;
1595 }
1596
1597 /* want more caps from mds? */
1598 if (want & ~(cap->mds_wanted | cap->issued))
1599 goto ack;
1600
1601 /* things we might delay */
1602 if ((cap->issued & ~retain) == 0 &&
1603 cap->mds_wanted == want)
1604 continue; /* nope, all good */
1605
1606 if (is_delayed)
1607 goto ack;
1608
1609 /* delay? */
1610 if ((ci->i_ceph_flags & CEPH_I_NODELAY) == 0 &&
1611 time_before(jiffies, ci->i_hold_caps_max)) {
1612 dout(" delaying issued %s -> %s, wanted %s -> %s\n",
1613 ceph_cap_string(cap->issued),
1614 ceph_cap_string(cap->issued & retain),
1615 ceph_cap_string(cap->mds_wanted),
1616 ceph_cap_string(want));
1617 delayed++;
1618 continue;
1619 }
1620
1621ack:
Sage Weile9964c12010-03-01 15:16:56 -08001622 if (ci->i_ceph_flags & CEPH_I_NOFLUSH) {
1623 dout(" skipping %p I_NOFLUSH set\n", inode);
1624 continue;
1625 }
1626
Sage Weila8599bd2009-10-06 11:31:12 -07001627 if (session && session != cap->session) {
1628 dout("oops, wrong session %p mutex\n", session);
1629 mutex_unlock(&session->s_mutex);
1630 session = NULL;
1631 }
1632 if (!session) {
1633 session = cap->session;
1634 if (mutex_trylock(&session->s_mutex) == 0) {
1635 dout("inverting session/ino locks on %p\n",
1636 session);
1637 spin_unlock(&inode->i_lock);
1638 if (took_snap_rwsem) {
1639 up_read(&mdsc->snap_rwsem);
1640 took_snap_rwsem = 0;
1641 }
1642 mutex_lock(&session->s_mutex);
1643 goto retry;
1644 }
1645 }
1646 /* take snap_rwsem after session mutex */
1647 if (!took_snap_rwsem) {
1648 if (down_read_trylock(&mdsc->snap_rwsem) == 0) {
1649 dout("inverting snap/in locks on %p\n",
1650 inode);
1651 spin_unlock(&inode->i_lock);
1652 down_read(&mdsc->snap_rwsem);
1653 took_snap_rwsem = 1;
1654 goto retry;
1655 }
1656 took_snap_rwsem = 1;
1657 }
1658
Sage Weilcdc35f92009-10-14 14:24:19 -07001659 if (cap == ci->i_auth_cap && ci->i_dirty_caps)
1660 flushing = __mark_caps_flushing(inode, session);
Sage Weila8599bd2009-10-06 11:31:12 -07001661
1662 mds = cap->mds; /* remember mds, so we don't repeat */
1663 sent++;
1664
1665 /* __send_cap drops i_lock */
1666 delayed += __send_cap(mdsc, cap, CEPH_CAP_OP_UPDATE, used, want,
1667 retain, flushing, NULL);
1668 goto retry; /* retake i_lock and restart our cap scan. */
1669 }
1670
1671 /*
1672 * Reschedule delayed caps release if we delayed anything,
1673 * otherwise cancel.
1674 */
1675 if (delayed && is_delayed)
1676 force_requeue = 1; /* __send_cap delayed release; requeue */
1677 if (!delayed && !is_delayed)
1678 __cap_delay_cancel(mdsc, ci);
1679 else if (!is_delayed || force_requeue)
1680 __cap_delay_requeue(mdsc, ci);
1681
1682 spin_unlock(&inode->i_lock);
1683
Sage Weilcbd03632010-02-09 13:41:18 -08001684 if (queue_invalidate)
Sage Weil3c6f6b72010-02-09 15:24:44 -08001685 ceph_queue_invalidate(inode);
Sage Weilcbd03632010-02-09 13:41:18 -08001686
Sage Weilcdc2ce02010-03-16 13:39:28 -07001687 if (session)
Sage Weila8599bd2009-10-06 11:31:12 -07001688 mutex_unlock(&session->s_mutex);
1689 if (took_snap_rwsem)
1690 up_read(&mdsc->snap_rwsem);
1691}
1692
1693/*
Sage Weila8599bd2009-10-06 11:31:12 -07001694 * Try to flush dirty caps back to the auth mds.
1695 */
1696static int try_flush_caps(struct inode *inode, struct ceph_mds_session *session,
1697 unsigned *flush_tid)
1698{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001699 struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
Sage Weila8599bd2009-10-06 11:31:12 -07001700 struct ceph_inode_info *ci = ceph_inode(inode);
1701 int unlock_session = session ? 0 : 1;
1702 int flushing = 0;
1703
1704retry:
1705 spin_lock(&inode->i_lock);
Sage Weile9964c12010-03-01 15:16:56 -08001706 if (ci->i_ceph_flags & CEPH_I_NOFLUSH) {
1707 dout("try_flush_caps skipping %p I_NOFLUSH set\n", inode);
1708 goto out;
1709 }
Sage Weila8599bd2009-10-06 11:31:12 -07001710 if (ci->i_dirty_caps && ci->i_auth_cap) {
1711 struct ceph_cap *cap = ci->i_auth_cap;
1712 int used = __ceph_caps_used(ci);
1713 int want = __ceph_caps_wanted(ci);
1714 int delayed;
1715
1716 if (!session) {
1717 spin_unlock(&inode->i_lock);
1718 session = cap->session;
1719 mutex_lock(&session->s_mutex);
1720 goto retry;
1721 }
1722 BUG_ON(session != cap->session);
1723 if (cap->session->s_state < CEPH_MDS_SESSION_OPEN)
1724 goto out;
1725
Sage Weilcdc35f92009-10-14 14:24:19 -07001726 flushing = __mark_caps_flushing(inode, session);
Sage Weila8599bd2009-10-06 11:31:12 -07001727
1728 /* __send_cap drops i_lock */
1729 delayed = __send_cap(mdsc, cap, CEPH_CAP_OP_FLUSH, used, want,
1730 cap->issued | cap->implemented, flushing,
1731 flush_tid);
1732 if (!delayed)
1733 goto out_unlocked;
1734
1735 spin_lock(&inode->i_lock);
1736 __cap_delay_requeue(mdsc, ci);
1737 }
1738out:
1739 spin_unlock(&inode->i_lock);
1740out_unlocked:
1741 if (session && unlock_session)
1742 mutex_unlock(&session->s_mutex);
1743 return flushing;
1744}
1745
1746/*
1747 * Return true if we've flushed caps through the given flush_tid.
1748 */
1749static int caps_are_flushed(struct inode *inode, unsigned tid)
1750{
1751 struct ceph_inode_info *ci = ceph_inode(inode);
Dan Carpentera5ee7512010-05-07 10:27:14 +02001752 int i, ret = 1;
Sage Weila8599bd2009-10-06 11:31:12 -07001753
1754 spin_lock(&inode->i_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001755 for (i = 0; i < CEPH_CAP_BITS; i++)
1756 if ((ci->i_flushing_caps & (1 << i)) &&
1757 ci->i_cap_flush_tid[i] <= tid) {
1758 /* still flushing this bit */
1759 ret = 0;
1760 break;
1761 }
1762 spin_unlock(&inode->i_lock);
1763 return ret;
1764}
1765
1766/*
1767 * Wait on any unsafe replies for the given inode. First wait on the
1768 * newest request, and make that the upper bound. Then, if there are
1769 * more requests, keep waiting on the oldest as long as it is still older
1770 * than the original request.
1771 */
1772static void sync_write_wait(struct inode *inode)
1773{
1774 struct ceph_inode_info *ci = ceph_inode(inode);
1775 struct list_head *head = &ci->i_unsafe_writes;
1776 struct ceph_osd_request *req;
1777 u64 last_tid;
1778
1779 spin_lock(&ci->i_unsafe_lock);
1780 if (list_empty(head))
1781 goto out;
1782
1783 /* set upper bound as _last_ entry in chain */
1784 req = list_entry(head->prev, struct ceph_osd_request,
1785 r_unsafe_item);
1786 last_tid = req->r_tid;
1787
1788 do {
1789 ceph_osdc_get_request(req);
1790 spin_unlock(&ci->i_unsafe_lock);
1791 dout("sync_write_wait on tid %llu (until %llu)\n",
1792 req->r_tid, last_tid);
1793 wait_for_completion(&req->r_safe_completion);
1794 spin_lock(&ci->i_unsafe_lock);
1795 ceph_osdc_put_request(req);
1796
1797 /*
1798 * from here on look at first entry in chain, since we
1799 * only want to wait for anything older than last_tid
1800 */
1801 if (list_empty(head))
1802 break;
1803 req = list_entry(head->next, struct ceph_osd_request,
1804 r_unsafe_item);
1805 } while (req->r_tid < last_tid);
1806out:
1807 spin_unlock(&ci->i_unsafe_lock);
1808}
1809
Christoph Hellwig7ea80852010-05-26 17:53:25 +02001810int ceph_fsync(struct file *file, int datasync)
Sage Weila8599bd2009-10-06 11:31:12 -07001811{
Christoph Hellwig7ea80852010-05-26 17:53:25 +02001812 struct inode *inode = file->f_mapping->host;
Sage Weila8599bd2009-10-06 11:31:12 -07001813 struct ceph_inode_info *ci = ceph_inode(inode);
1814 unsigned flush_tid;
1815 int ret;
1816 int dirty;
1817
1818 dout("fsync %p%s\n", inode, datasync ? " datasync" : "");
1819 sync_write_wait(inode);
1820
1821 ret = filemap_write_and_wait(inode->i_mapping);
1822 if (ret < 0)
1823 return ret;
1824
1825 dirty = try_flush_caps(inode, NULL, &flush_tid);
1826 dout("fsync dirty caps are %s\n", ceph_cap_string(dirty));
1827
1828 /*
1829 * only wait on non-file metadata writeback (the mds
1830 * can recover size and mtime, so we don't need to
1831 * wait for that)
1832 */
1833 if (!datasync && (dirty & ~CEPH_CAP_ANY_FILE_WR)) {
1834 dout("fsync waiting for flush_tid %u\n", flush_tid);
1835 ret = wait_event_interruptible(ci->i_cap_wq,
1836 caps_are_flushed(inode, flush_tid));
1837 }
1838
1839 dout("fsync %p%s done\n", inode, datasync ? " datasync" : "");
1840 return ret;
1841}
1842
1843/*
1844 * Flush any dirty caps back to the mds. If we aren't asked to wait,
1845 * queue inode for flush but don't do so immediately, because we can
1846 * get by with fewer MDS messages if we wait for data writeback to
1847 * complete first.
1848 */
Stephen Rothwellf1a3d572010-01-18 11:53:08 +11001849int ceph_write_inode(struct inode *inode, struct writeback_control *wbc)
Sage Weila8599bd2009-10-06 11:31:12 -07001850{
1851 struct ceph_inode_info *ci = ceph_inode(inode);
1852 unsigned flush_tid;
1853 int err = 0;
1854 int dirty;
Stephen Rothwellf1a3d572010-01-18 11:53:08 +11001855 int wait = wbc->sync_mode == WB_SYNC_ALL;
Sage Weila8599bd2009-10-06 11:31:12 -07001856
1857 dout("write_inode %p wait=%d\n", inode, wait);
1858 if (wait) {
1859 dirty = try_flush_caps(inode, NULL, &flush_tid);
1860 if (dirty)
1861 err = wait_event_interruptible(ci->i_cap_wq,
1862 caps_are_flushed(inode, flush_tid));
1863 } else {
Cheng Renquan640ef792010-03-26 17:40:33 +08001864 struct ceph_mds_client *mdsc =
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001865 ceph_sb_to_client(inode->i_sb)->mdsc;
Sage Weila8599bd2009-10-06 11:31:12 -07001866
1867 spin_lock(&inode->i_lock);
1868 if (__ceph_caps_dirty(ci))
1869 __cap_delay_requeue_front(mdsc, ci);
1870 spin_unlock(&inode->i_lock);
1871 }
1872 return err;
1873}
1874
1875/*
1876 * After a recovering MDS goes active, we need to resend any caps
1877 * we were flushing.
1878 *
1879 * Caller holds session->s_mutex.
1880 */
1881static void kick_flushing_capsnaps(struct ceph_mds_client *mdsc,
1882 struct ceph_mds_session *session)
1883{
1884 struct ceph_cap_snap *capsnap;
1885
1886 dout("kick_flushing_capsnaps mds%d\n", session->s_mds);
1887 list_for_each_entry(capsnap, &session->s_cap_snaps_flushing,
1888 flushing_item) {
1889 struct ceph_inode_info *ci = capsnap->ci;
1890 struct inode *inode = &ci->vfs_inode;
1891 struct ceph_cap *cap;
1892
1893 spin_lock(&inode->i_lock);
1894 cap = ci->i_auth_cap;
1895 if (cap && cap->session == session) {
1896 dout("kick_flushing_caps %p cap %p capsnap %p\n", inode,
1897 cap, capsnap);
Sage Weile8351242010-09-17 08:03:08 -07001898 __ceph_flush_snaps(ci, &session, 1);
Sage Weila8599bd2009-10-06 11:31:12 -07001899 } else {
1900 pr_err("%p auth cap %p not mds%d ???\n", inode,
1901 cap, session->s_mds);
Sage Weila8599bd2009-10-06 11:31:12 -07001902 }
Sage Weil0b0c06d2010-04-20 10:27:13 -07001903 spin_unlock(&inode->i_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001904 }
1905}
1906
1907void ceph_kick_flushing_caps(struct ceph_mds_client *mdsc,
1908 struct ceph_mds_session *session)
1909{
1910 struct ceph_inode_info *ci;
1911
1912 kick_flushing_capsnaps(mdsc, session);
1913
1914 dout("kick_flushing_caps mds%d\n", session->s_mds);
1915 list_for_each_entry(ci, &session->s_cap_flushing, i_flushing_item) {
1916 struct inode *inode = &ci->vfs_inode;
1917 struct ceph_cap *cap;
1918 int delayed = 0;
1919
1920 spin_lock(&inode->i_lock);
1921 cap = ci->i_auth_cap;
1922 if (cap && cap->session == session) {
1923 dout("kick_flushing_caps %p cap %p %s\n", inode,
1924 cap, ceph_cap_string(ci->i_flushing_caps));
1925 delayed = __send_cap(mdsc, cap, CEPH_CAP_OP_FLUSH,
1926 __ceph_caps_used(ci),
1927 __ceph_caps_wanted(ci),
1928 cap->issued | cap->implemented,
1929 ci->i_flushing_caps, NULL);
1930 if (delayed) {
1931 spin_lock(&inode->i_lock);
1932 __cap_delay_requeue(mdsc, ci);
1933 spin_unlock(&inode->i_lock);
1934 }
1935 } else {
1936 pr_err("%p auth cap %p not mds%d ???\n", inode,
1937 cap, session->s_mds);
1938 spin_unlock(&inode->i_lock);
1939 }
1940 }
1941}
1942
1943
1944/*
1945 * Take references to capabilities we hold, so that we don't release
1946 * them to the MDS prematurely.
1947 *
1948 * Protected by i_lock.
1949 */
1950static void __take_cap_refs(struct ceph_inode_info *ci, int got)
1951{
1952 if (got & CEPH_CAP_PIN)
1953 ci->i_pin_ref++;
1954 if (got & CEPH_CAP_FILE_RD)
1955 ci->i_rd_ref++;
1956 if (got & CEPH_CAP_FILE_CACHE)
1957 ci->i_rdcache_ref++;
1958 if (got & CEPH_CAP_FILE_WR)
1959 ci->i_wr_ref++;
1960 if (got & CEPH_CAP_FILE_BUFFER) {
1961 if (ci->i_wrbuffer_ref == 0)
1962 igrab(&ci->vfs_inode);
1963 ci->i_wrbuffer_ref++;
1964 dout("__take_cap_refs %p wrbuffer %d -> %d (?)\n",
1965 &ci->vfs_inode, ci->i_wrbuffer_ref-1, ci->i_wrbuffer_ref);
1966 }
1967}
1968
1969/*
1970 * Try to grab cap references. Specify those refs we @want, and the
1971 * minimal set we @need. Also include the larger offset we are writing
1972 * to (when applicable), and check against max_size here as well.
1973 * Note that caller is responsible for ensuring max_size increases are
1974 * requested from the MDS.
1975 */
1976static int try_get_cap_refs(struct ceph_inode_info *ci, int need, int want,
1977 int *got, loff_t endoff, int *check_max, int *err)
1978{
1979 struct inode *inode = &ci->vfs_inode;
1980 int ret = 0;
1981 int have, implemented;
Sage Weil195d3ce2010-03-01 09:57:54 -08001982 int file_wanted;
Sage Weila8599bd2009-10-06 11:31:12 -07001983
1984 dout("get_cap_refs %p need %s want %s\n", inode,
1985 ceph_cap_string(need), ceph_cap_string(want));
1986 spin_lock(&inode->i_lock);
1987
Sage Weil195d3ce2010-03-01 09:57:54 -08001988 /* make sure file is actually open */
1989 file_wanted = __ceph_caps_file_wanted(ci);
1990 if ((file_wanted & need) == 0) {
1991 dout("try_get_cap_refs need %s file_wanted %s, EBADF\n",
1992 ceph_cap_string(need), ceph_cap_string(file_wanted));
Sage Weila8599bd2009-10-06 11:31:12 -07001993 *err = -EBADF;
1994 ret = 1;
1995 goto out;
1996 }
1997
1998 if (need & CEPH_CAP_FILE_WR) {
1999 if (endoff >= 0 && endoff > (loff_t)ci->i_max_size) {
2000 dout("get_cap_refs %p endoff %llu > maxsize %llu\n",
2001 inode, endoff, ci->i_max_size);
2002 if (endoff > ci->i_wanted_max_size) {
2003 *check_max = 1;
2004 ret = 1;
2005 }
2006 goto out;
2007 }
2008 /*
2009 * If a sync write is in progress, we must wait, so that we
2010 * can get a final snapshot value for size+mtime.
2011 */
2012 if (__ceph_have_pending_cap_snap(ci)) {
2013 dout("get_cap_refs %p cap_snap_pending\n", inode);
2014 goto out;
2015 }
2016 }
2017 have = __ceph_caps_issued(ci, &implemented);
2018
2019 /*
2020 * disallow writes while a truncate is pending
2021 */
2022 if (ci->i_truncate_pending)
2023 have &= ~CEPH_CAP_FILE_WR;
2024
2025 if ((have & need) == need) {
2026 /*
2027 * Look at (implemented & ~have & not) so that we keep waiting
2028 * on transition from wanted -> needed caps. This is needed
2029 * for WRBUFFER|WR -> WR to avoid a new WR sync write from
2030 * going before a prior buffered writeback happens.
2031 */
2032 int not = want & ~(have & need);
2033 int revoking = implemented & ~have;
2034 dout("get_cap_refs %p have %s but not %s (revoking %s)\n",
2035 inode, ceph_cap_string(have), ceph_cap_string(not),
2036 ceph_cap_string(revoking));
2037 if ((revoking & not) == 0) {
2038 *got = need | (have & want);
2039 __take_cap_refs(ci, *got);
2040 ret = 1;
2041 }
2042 } else {
2043 dout("get_cap_refs %p have %s needed %s\n", inode,
2044 ceph_cap_string(have), ceph_cap_string(need));
2045 }
2046out:
2047 spin_unlock(&inode->i_lock);
2048 dout("get_cap_refs %p ret %d got %s\n", inode,
2049 ret, ceph_cap_string(*got));
2050 return ret;
2051}
2052
2053/*
2054 * Check the offset we are writing up to against our current
2055 * max_size. If necessary, tell the MDS we want to write to
2056 * a larger offset.
2057 */
2058static void check_max_size(struct inode *inode, loff_t endoff)
2059{
2060 struct ceph_inode_info *ci = ceph_inode(inode);
2061 int check = 0;
2062
2063 /* do we need to explicitly request a larger max_size? */
2064 spin_lock(&inode->i_lock);
2065 if ((endoff >= ci->i_max_size ||
2066 endoff > (inode->i_size << 1)) &&
2067 endoff > ci->i_wanted_max_size) {
2068 dout("write %p at large endoff %llu, req max_size\n",
2069 inode, endoff);
2070 ci->i_wanted_max_size = endoff;
2071 check = 1;
2072 }
2073 spin_unlock(&inode->i_lock);
2074 if (check)
2075 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
2076}
2077
2078/*
2079 * Wait for caps, and take cap references. If we can't get a WR cap
2080 * due to a small max_size, make sure we check_max_size (and possibly
2081 * ask the mds) so we don't get hung up indefinitely.
2082 */
2083int ceph_get_caps(struct ceph_inode_info *ci, int need, int want, int *got,
2084 loff_t endoff)
2085{
2086 int check_max, ret, err;
2087
2088retry:
2089 if (endoff > 0)
2090 check_max_size(&ci->vfs_inode, endoff);
2091 check_max = 0;
2092 err = 0;
2093 ret = wait_event_interruptible(ci->i_cap_wq,
2094 try_get_cap_refs(ci, need, want,
2095 got, endoff,
2096 &check_max, &err));
2097 if (err)
2098 ret = err;
2099 if (check_max)
2100 goto retry;
2101 return ret;
2102}
2103
2104/*
2105 * Take cap refs. Caller must already know we hold at least one ref
2106 * on the caps in question or we don't know this is safe.
2107 */
2108void ceph_get_cap_refs(struct ceph_inode_info *ci, int caps)
2109{
2110 spin_lock(&ci->vfs_inode.i_lock);
2111 __take_cap_refs(ci, caps);
2112 spin_unlock(&ci->vfs_inode.i_lock);
2113}
2114
2115/*
2116 * Release cap refs.
2117 *
2118 * If we released the last ref on any given cap, call ceph_check_caps
2119 * to release (or schedule a release).
2120 *
2121 * If we are releasing a WR cap (from a sync write), finalize any affected
2122 * cap_snap, and wake up any waiters.
2123 */
2124void ceph_put_cap_refs(struct ceph_inode_info *ci, int had)
2125{
2126 struct inode *inode = &ci->vfs_inode;
2127 int last = 0, put = 0, flushsnaps = 0, wake = 0;
2128 struct ceph_cap_snap *capsnap;
2129
2130 spin_lock(&inode->i_lock);
2131 if (had & CEPH_CAP_PIN)
2132 --ci->i_pin_ref;
2133 if (had & CEPH_CAP_FILE_RD)
2134 if (--ci->i_rd_ref == 0)
2135 last++;
2136 if (had & CEPH_CAP_FILE_CACHE)
2137 if (--ci->i_rdcache_ref == 0)
2138 last++;
2139 if (had & CEPH_CAP_FILE_BUFFER) {
2140 if (--ci->i_wrbuffer_ref == 0) {
2141 last++;
2142 put++;
2143 }
2144 dout("put_cap_refs %p wrbuffer %d -> %d (?)\n",
2145 inode, ci->i_wrbuffer_ref+1, ci->i_wrbuffer_ref);
2146 }
2147 if (had & CEPH_CAP_FILE_WR)
2148 if (--ci->i_wr_ref == 0) {
2149 last++;
2150 if (!list_empty(&ci->i_cap_snaps)) {
2151 capsnap = list_first_entry(&ci->i_cap_snaps,
2152 struct ceph_cap_snap,
2153 ci_item);
2154 if (capsnap->writing) {
2155 capsnap->writing = 0;
2156 flushsnaps =
2157 __ceph_finish_cap_snap(ci,
2158 capsnap);
2159 wake = 1;
2160 }
2161 }
2162 }
2163 spin_unlock(&inode->i_lock);
2164
Sage Weil819ccbf2010-04-01 09:33:46 -07002165 dout("put_cap_refs %p had %s%s%s\n", inode, ceph_cap_string(had),
2166 last ? " last" : "", put ? " put" : "");
Sage Weila8599bd2009-10-06 11:31:12 -07002167
2168 if (last && !flushsnaps)
2169 ceph_check_caps(ci, 0, NULL);
2170 else if (flushsnaps)
2171 ceph_flush_snaps(ci);
2172 if (wake)
Yehuda Sadeh03066f22010-07-27 13:11:08 -07002173 wake_up_all(&ci->i_cap_wq);
Sage Weila8599bd2009-10-06 11:31:12 -07002174 if (put)
2175 iput(inode);
2176}
2177
2178/*
2179 * Release @nr WRBUFFER refs on dirty pages for the given @snapc snap
2180 * context. Adjust per-snap dirty page accounting as appropriate.
2181 * Once all dirty data for a cap_snap is flushed, flush snapped file
2182 * metadata back to the MDS. If we dropped the last ref, call
2183 * ceph_check_caps.
2184 */
2185void ceph_put_wrbuffer_cap_refs(struct ceph_inode_info *ci, int nr,
2186 struct ceph_snap_context *snapc)
2187{
2188 struct inode *inode = &ci->vfs_inode;
2189 int last = 0;
Sage Weil819ccbf2010-04-01 09:33:46 -07002190 int complete_capsnap = 0;
2191 int drop_capsnap = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07002192 int found = 0;
2193 struct ceph_cap_snap *capsnap = NULL;
2194
2195 spin_lock(&inode->i_lock);
2196 ci->i_wrbuffer_ref -= nr;
2197 last = !ci->i_wrbuffer_ref;
2198
2199 if (ci->i_head_snapc == snapc) {
2200 ci->i_wrbuffer_ref_head -= nr;
Sage Weil7d8cb262010-08-24 08:44:16 -07002201 if (ci->i_wrbuffer_ref_head == 0 &&
2202 ci->i_dirty_caps == 0 && ci->i_flushing_caps == 0) {
2203 BUG_ON(!ci->i_head_snapc);
Sage Weila8599bd2009-10-06 11:31:12 -07002204 ceph_put_snap_context(ci->i_head_snapc);
2205 ci->i_head_snapc = NULL;
2206 }
2207 dout("put_wrbuffer_cap_refs on %p head %d/%d -> %d/%d %s\n",
2208 inode,
2209 ci->i_wrbuffer_ref+nr, ci->i_wrbuffer_ref_head+nr,
2210 ci->i_wrbuffer_ref, ci->i_wrbuffer_ref_head,
2211 last ? " LAST" : "");
2212 } else {
2213 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
2214 if (capsnap->context == snapc) {
2215 found = 1;
Sage Weila8599bd2009-10-06 11:31:12 -07002216 break;
2217 }
2218 }
2219 BUG_ON(!found);
Sage Weil819ccbf2010-04-01 09:33:46 -07002220 capsnap->dirty_pages -= nr;
2221 if (capsnap->dirty_pages == 0) {
2222 complete_capsnap = 1;
2223 if (capsnap->dirty == 0)
2224 /* cap writeback completed before we created
2225 * the cap_snap; no FLUSHSNAP is needed */
2226 drop_capsnap = 1;
2227 }
Sage Weila8599bd2009-10-06 11:31:12 -07002228 dout("put_wrbuffer_cap_refs on %p cap_snap %p "
Sage Weil819ccbf2010-04-01 09:33:46 -07002229 " snap %lld %d/%d -> %d/%d %s%s%s\n",
Sage Weila8599bd2009-10-06 11:31:12 -07002230 inode, capsnap, capsnap->context->seq,
2231 ci->i_wrbuffer_ref+nr, capsnap->dirty_pages + nr,
2232 ci->i_wrbuffer_ref, capsnap->dirty_pages,
2233 last ? " (wrbuffer last)" : "",
Sage Weil819ccbf2010-04-01 09:33:46 -07002234 complete_capsnap ? " (complete capsnap)" : "",
2235 drop_capsnap ? " (drop capsnap)" : "");
2236 if (drop_capsnap) {
2237 ceph_put_snap_context(capsnap->context);
2238 list_del(&capsnap->ci_item);
2239 list_del(&capsnap->flushing_item);
2240 ceph_put_cap_snap(capsnap);
2241 }
Sage Weila8599bd2009-10-06 11:31:12 -07002242 }
2243
2244 spin_unlock(&inode->i_lock);
2245
2246 if (last) {
2247 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
2248 iput(inode);
Sage Weil819ccbf2010-04-01 09:33:46 -07002249 } else if (complete_capsnap) {
Sage Weila8599bd2009-10-06 11:31:12 -07002250 ceph_flush_snaps(ci);
Yehuda Sadeh03066f22010-07-27 13:11:08 -07002251 wake_up_all(&ci->i_cap_wq);
Sage Weila8599bd2009-10-06 11:31:12 -07002252 }
Sage Weil819ccbf2010-04-01 09:33:46 -07002253 if (drop_capsnap)
2254 iput(inode);
Sage Weila8599bd2009-10-06 11:31:12 -07002255}
2256
2257/*
2258 * Handle a cap GRANT message from the MDS. (Note that a GRANT may
2259 * actually be a revocation if it specifies a smaller cap set.)
2260 *
Sage Weil15637c82010-03-16 13:42:00 -07002261 * caller holds s_mutex and i_lock, we drop both.
2262 *
Sage Weila8599bd2009-10-06 11:31:12 -07002263 * return value:
2264 * 0 - ok
2265 * 1 - check_caps on auth cap only (writeback)
2266 * 2 - check_caps (ack revoke)
2267 */
Sage Weil15637c82010-03-16 13:42:00 -07002268static void handle_cap_grant(struct inode *inode, struct ceph_mds_caps *grant,
2269 struct ceph_mds_session *session,
2270 struct ceph_cap *cap,
2271 struct ceph_buffer *xattr_buf)
Yehuda Sadehcd84db62010-06-11 16:58:48 -07002272 __releases(inode->i_lock)
Sage Weila8599bd2009-10-06 11:31:12 -07002273{
2274 struct ceph_inode_info *ci = ceph_inode(inode);
2275 int mds = session->s_mds;
Sage Weil2f56f562010-10-27 20:59:49 -07002276 int seq = le32_to_cpu(grant->seq);
Sage Weila8599bd2009-10-06 11:31:12 -07002277 int newcaps = le32_to_cpu(grant->caps);
2278 int issued, implemented, used, wanted, dirty;
2279 u64 size = le64_to_cpu(grant->size);
2280 u64 max_size = le64_to_cpu(grant->max_size);
2281 struct timespec mtime, atime, ctime;
Sage Weil15637c82010-03-16 13:42:00 -07002282 int check_caps = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07002283 int wake = 0;
2284 int writeback = 0;
2285 int revoked_rdcache = 0;
Sage Weil3c6f6b72010-02-09 15:24:44 -08002286 int queue_invalidate = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07002287
Sage Weil2f56f562010-10-27 20:59:49 -07002288 dout("handle_cap_grant inode %p cap %p mds%d seq %d %s\n",
2289 inode, cap, mds, seq, ceph_cap_string(newcaps));
Sage Weila8599bd2009-10-06 11:31:12 -07002290 dout(" size %llu max_size %llu, i_size %llu\n", size, max_size,
2291 inode->i_size);
2292
2293 /*
2294 * If CACHE is being revoked, and we have no dirty buffers,
2295 * try to invalidate (once). (If there are dirty buffers, we
2296 * will invalidate _after_ writeback.)
2297 */
Sage Weil3b454c42010-06-10 13:20:33 -07002298 if (((cap->issued & ~newcaps) & CEPH_CAP_FILE_CACHE) &&
2299 (newcaps & CEPH_CAP_FILE_LAZYIO) == 0 &&
Yehuda Sadehbcd2cbd2010-02-19 00:12:21 +00002300 !ci->i_wrbuffer_ref) {
Sage Weil5ecad6f2010-02-17 10:43:37 -08002301 if (try_nonblocking_invalidate(inode) == 0) {
2302 revoked_rdcache = 1;
2303 } else {
Sage Weila8599bd2009-10-06 11:31:12 -07002304 /* there were locked pages.. invalidate later
2305 in a separate thread. */
2306 if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
Sage Weil3c6f6b72010-02-09 15:24:44 -08002307 queue_invalidate = 1;
Sage Weila8599bd2009-10-06 11:31:12 -07002308 ci->i_rdcache_revoking = ci->i_rdcache_gen;
2309 }
Sage Weila8599bd2009-10-06 11:31:12 -07002310 }
Sage Weila8599bd2009-10-06 11:31:12 -07002311 }
2312
2313 /* side effects now are allowed */
2314
2315 issued = __ceph_caps_issued(ci, &implemented);
2316 issued |= implemented | __ceph_caps_dirty(ci);
2317
Sage Weil685f9a5d2009-11-09 12:05:48 -08002318 cap->cap_gen = session->s_cap_gen;
Sage Weila8599bd2009-10-06 11:31:12 -07002319
2320 __check_cap_issue(ci, cap, newcaps);
2321
2322 if ((issued & CEPH_CAP_AUTH_EXCL) == 0) {
2323 inode->i_mode = le32_to_cpu(grant->mode);
2324 inode->i_uid = le32_to_cpu(grant->uid);
2325 inode->i_gid = le32_to_cpu(grant->gid);
2326 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
2327 inode->i_uid, inode->i_gid);
2328 }
2329
2330 if ((issued & CEPH_CAP_LINK_EXCL) == 0)
2331 inode->i_nlink = le32_to_cpu(grant->nlink);
2332
2333 if ((issued & CEPH_CAP_XATTR_EXCL) == 0 && grant->xattr_len) {
2334 int len = le32_to_cpu(grant->xattr_len);
2335 u64 version = le64_to_cpu(grant->xattr_version);
2336
2337 if (version > ci->i_xattrs.version) {
2338 dout(" got new xattrs v%llu on %p len %d\n",
2339 version, inode, len);
2340 if (ci->i_xattrs.blob)
2341 ceph_buffer_put(ci->i_xattrs.blob);
2342 ci->i_xattrs.blob = ceph_buffer_get(xattr_buf);
2343 ci->i_xattrs.version = version;
2344 }
2345 }
2346
2347 /* size/ctime/mtime/atime? */
2348 ceph_fill_file_size(inode, issued,
2349 le32_to_cpu(grant->truncate_seq),
2350 le64_to_cpu(grant->truncate_size), size);
2351 ceph_decode_timespec(&mtime, &grant->mtime);
2352 ceph_decode_timespec(&atime, &grant->atime);
2353 ceph_decode_timespec(&ctime, &grant->ctime);
2354 ceph_fill_file_time(inode, issued,
2355 le32_to_cpu(grant->time_warp_seq), &ctime, &mtime,
2356 &atime);
2357
2358 /* max size increase? */
2359 if (max_size != ci->i_max_size) {
2360 dout("max_size %lld -> %llu\n", ci->i_max_size, max_size);
2361 ci->i_max_size = max_size;
2362 if (max_size >= ci->i_wanted_max_size) {
2363 ci->i_wanted_max_size = 0; /* reset */
2364 ci->i_requested_max_size = 0;
2365 }
2366 wake = 1;
2367 }
2368
2369 /* check cap bits */
2370 wanted = __ceph_caps_wanted(ci);
2371 used = __ceph_caps_used(ci);
2372 dirty = __ceph_caps_dirty(ci);
2373 dout(" my wanted = %s, used = %s, dirty %s\n",
2374 ceph_cap_string(wanted),
2375 ceph_cap_string(used),
2376 ceph_cap_string(dirty));
2377 if (wanted != le32_to_cpu(grant->wanted)) {
2378 dout("mds wanted %s -> %s\n",
2379 ceph_cap_string(le32_to_cpu(grant->wanted)),
2380 ceph_cap_string(wanted));
2381 grant->wanted = cpu_to_le32(wanted);
2382 }
2383
2384 cap->seq = seq;
2385
2386 /* file layout may have changed */
2387 ci->i_layout = grant->layout;
2388
2389 /* revocation, grant, or no-op? */
2390 if (cap->issued & ~newcaps) {
Sage Weil3b454c42010-06-10 13:20:33 -07002391 int revoking = cap->issued & ~newcaps;
2392
2393 dout("revocation: %s -> %s (revoking %s)\n",
2394 ceph_cap_string(cap->issued),
2395 ceph_cap_string(newcaps),
2396 ceph_cap_string(revoking));
Sage Weil0eb6cd42010-08-05 13:53:18 -07002397 if (revoking & used & CEPH_CAP_FILE_BUFFER)
Sage Weil3b454c42010-06-10 13:20:33 -07002398 writeback = 1; /* initiate writeback; will delay ack */
2399 else if (revoking == CEPH_CAP_FILE_CACHE &&
2400 (newcaps & CEPH_CAP_FILE_LAZYIO) == 0 &&
2401 queue_invalidate)
2402 ; /* do nothing yet, invalidation will be queued */
2403 else if (cap == ci->i_auth_cap)
2404 check_caps = 1; /* check auth cap only */
2405 else
2406 check_caps = 2; /* check all caps */
Sage Weila8599bd2009-10-06 11:31:12 -07002407 cap->issued = newcaps;
Sage Weil978097c2010-03-08 15:27:53 -08002408 cap->implemented |= newcaps;
Sage Weila8599bd2009-10-06 11:31:12 -07002409 } else if (cap->issued == newcaps) {
2410 dout("caps unchanged: %s -> %s\n",
2411 ceph_cap_string(cap->issued), ceph_cap_string(newcaps));
2412 } else {
2413 dout("grant: %s -> %s\n", ceph_cap_string(cap->issued),
2414 ceph_cap_string(newcaps));
2415 cap->issued = newcaps;
2416 cap->implemented |= newcaps; /* add bits only, to
2417 * avoid stepping on a
2418 * pending revocation */
2419 wake = 1;
2420 }
Sage Weil978097c2010-03-08 15:27:53 -08002421 BUG_ON(cap->issued & ~cap->implemented);
Sage Weila8599bd2009-10-06 11:31:12 -07002422
2423 spin_unlock(&inode->i_lock);
Sage Weil3c6f6b72010-02-09 15:24:44 -08002424 if (writeback)
Sage Weila8599bd2009-10-06 11:31:12 -07002425 /*
2426 * queue inode for writeback: we can't actually call
2427 * filemap_write_and_wait, etc. from message handler
2428 * context.
2429 */
Sage Weil3c6f6b72010-02-09 15:24:44 -08002430 ceph_queue_writeback(inode);
2431 if (queue_invalidate)
2432 ceph_queue_invalidate(inode);
Sage Weila8599bd2009-10-06 11:31:12 -07002433 if (wake)
Yehuda Sadeh03066f22010-07-27 13:11:08 -07002434 wake_up_all(&ci->i_cap_wq);
Sage Weil15637c82010-03-16 13:42:00 -07002435
2436 if (check_caps == 1)
2437 ceph_check_caps(ci, CHECK_CAPS_NODELAY|CHECK_CAPS_AUTHONLY,
2438 session);
2439 else if (check_caps == 2)
2440 ceph_check_caps(ci, CHECK_CAPS_NODELAY, session);
2441 else
2442 mutex_unlock(&session->s_mutex);
Sage Weila8599bd2009-10-06 11:31:12 -07002443}
2444
2445/*
2446 * Handle FLUSH_ACK from MDS, indicating that metadata we sent to the
2447 * MDS has been safely committed.
2448 */
Sage Weil6df058c2009-12-22 11:24:33 -08002449static void handle_cap_flush_ack(struct inode *inode, u64 flush_tid,
Sage Weila8599bd2009-10-06 11:31:12 -07002450 struct ceph_mds_caps *m,
2451 struct ceph_mds_session *session,
2452 struct ceph_cap *cap)
2453 __releases(inode->i_lock)
2454{
2455 struct ceph_inode_info *ci = ceph_inode(inode);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002456 struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
Sage Weila8599bd2009-10-06 11:31:12 -07002457 unsigned seq = le32_to_cpu(m->seq);
2458 int dirty = le32_to_cpu(m->dirty);
2459 int cleaned = 0;
Sage Weilafcdaea2009-10-14 14:27:38 -07002460 int drop = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07002461 int i;
2462
2463 for (i = 0; i < CEPH_CAP_BITS; i++)
2464 if ((dirty & (1 << i)) &&
2465 flush_tid == ci->i_cap_flush_tid[i])
2466 cleaned |= 1 << i;
2467
2468 dout("handle_cap_flush_ack inode %p mds%d seq %d on %s cleaned %s,"
2469 " flushing %s -> %s\n",
2470 inode, session->s_mds, seq, ceph_cap_string(dirty),
2471 ceph_cap_string(cleaned), ceph_cap_string(ci->i_flushing_caps),
2472 ceph_cap_string(ci->i_flushing_caps & ~cleaned));
2473
2474 if (ci->i_flushing_caps == (ci->i_flushing_caps & ~cleaned))
2475 goto out;
2476
Sage Weila8599bd2009-10-06 11:31:12 -07002477 ci->i_flushing_caps &= ~cleaned;
Sage Weila8599bd2009-10-06 11:31:12 -07002478
2479 spin_lock(&mdsc->cap_dirty_lock);
2480 if (ci->i_flushing_caps == 0) {
2481 list_del_init(&ci->i_flushing_item);
2482 if (!list_empty(&session->s_cap_flushing))
2483 dout(" mds%d still flushing cap on %p\n",
2484 session->s_mds,
2485 &list_entry(session->s_cap_flushing.next,
2486 struct ceph_inode_info,
2487 i_flushing_item)->vfs_inode);
2488 mdsc->num_cap_flushing--;
Yehuda Sadeh03066f22010-07-27 13:11:08 -07002489 wake_up_all(&mdsc->cap_flushing_wq);
Sage Weila8599bd2009-10-06 11:31:12 -07002490 dout(" inode %p now !flushing\n", inode);
Sage Weilafcdaea2009-10-14 14:27:38 -07002491
2492 if (ci->i_dirty_caps == 0) {
2493 dout(" inode %p now clean\n", inode);
2494 BUG_ON(!list_empty(&ci->i_dirty_item));
2495 drop = 1;
Sage Weil7d8cb262010-08-24 08:44:16 -07002496 if (ci->i_wrbuffer_ref_head == 0) {
2497 BUG_ON(!ci->i_head_snapc);
2498 ceph_put_snap_context(ci->i_head_snapc);
2499 ci->i_head_snapc = NULL;
2500 }
Sage Weil76e3b392009-10-15 18:13:53 -07002501 } else {
2502 BUG_ON(list_empty(&ci->i_dirty_item));
Sage Weilafcdaea2009-10-14 14:27:38 -07002503 }
Sage Weila8599bd2009-10-06 11:31:12 -07002504 }
2505 spin_unlock(&mdsc->cap_dirty_lock);
Yehuda Sadeh03066f22010-07-27 13:11:08 -07002506 wake_up_all(&ci->i_cap_wq);
Sage Weila8599bd2009-10-06 11:31:12 -07002507
2508out:
2509 spin_unlock(&inode->i_lock);
Sage Weilafcdaea2009-10-14 14:27:38 -07002510 if (drop)
Sage Weila8599bd2009-10-06 11:31:12 -07002511 iput(inode);
2512}
2513
2514/*
2515 * Handle FLUSHSNAP_ACK. MDS has flushed snap data to disk and we can
2516 * throw away our cap_snap.
2517 *
2518 * Caller hold s_mutex.
2519 */
Sage Weil6df058c2009-12-22 11:24:33 -08002520static void handle_cap_flushsnap_ack(struct inode *inode, u64 flush_tid,
Sage Weila8599bd2009-10-06 11:31:12 -07002521 struct ceph_mds_caps *m,
2522 struct ceph_mds_session *session)
2523{
2524 struct ceph_inode_info *ci = ceph_inode(inode);
2525 u64 follows = le64_to_cpu(m->snap_follows);
Sage Weila8599bd2009-10-06 11:31:12 -07002526 struct ceph_cap_snap *capsnap;
2527 int drop = 0;
2528
2529 dout("handle_cap_flushsnap_ack inode %p ci %p mds%d follows %lld\n",
2530 inode, ci, session->s_mds, follows);
2531
2532 spin_lock(&inode->i_lock);
2533 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
2534 if (capsnap->follows == follows) {
2535 if (capsnap->flush_tid != flush_tid) {
2536 dout(" cap_snap %p follows %lld tid %lld !="
2537 " %lld\n", capsnap, follows,
2538 flush_tid, capsnap->flush_tid);
2539 break;
2540 }
2541 WARN_ON(capsnap->dirty_pages || capsnap->writing);
Sage Weil819ccbf2010-04-01 09:33:46 -07002542 dout(" removing %p cap_snap %p follows %lld\n",
2543 inode, capsnap, follows);
Sage Weila8599bd2009-10-06 11:31:12 -07002544 ceph_put_snap_context(capsnap->context);
2545 list_del(&capsnap->ci_item);
2546 list_del(&capsnap->flushing_item);
2547 ceph_put_cap_snap(capsnap);
2548 drop = 1;
2549 break;
2550 } else {
2551 dout(" skipping cap_snap %p follows %lld\n",
2552 capsnap, capsnap->follows);
2553 }
2554 }
2555 spin_unlock(&inode->i_lock);
2556 if (drop)
2557 iput(inode);
2558}
2559
2560/*
2561 * Handle TRUNC from MDS, indicating file truncation.
2562 *
2563 * caller hold s_mutex.
2564 */
2565static void handle_cap_trunc(struct inode *inode,
2566 struct ceph_mds_caps *trunc,
2567 struct ceph_mds_session *session)
2568 __releases(inode->i_lock)
2569{
2570 struct ceph_inode_info *ci = ceph_inode(inode);
2571 int mds = session->s_mds;
2572 int seq = le32_to_cpu(trunc->seq);
2573 u32 truncate_seq = le32_to_cpu(trunc->truncate_seq);
2574 u64 truncate_size = le64_to_cpu(trunc->truncate_size);
2575 u64 size = le64_to_cpu(trunc->size);
2576 int implemented = 0;
2577 int dirty = __ceph_caps_dirty(ci);
2578 int issued = __ceph_caps_issued(ceph_inode(inode), &implemented);
2579 int queue_trunc = 0;
2580
2581 issued |= implemented | dirty;
2582
2583 dout("handle_cap_trunc inode %p mds%d seq %d to %lld seq %d\n",
2584 inode, mds, seq, truncate_size, truncate_seq);
2585 queue_trunc = ceph_fill_file_size(inode, issued,
2586 truncate_seq, truncate_size, size);
2587 spin_unlock(&inode->i_lock);
2588
2589 if (queue_trunc)
Sage Weil3c6f6b72010-02-09 15:24:44 -08002590 ceph_queue_vmtruncate(inode);
Sage Weila8599bd2009-10-06 11:31:12 -07002591}
2592
2593/*
2594 * Handle EXPORT from MDS. Cap is being migrated _from_ this mds to a
2595 * different one. If we are the most recent migration we've seen (as
2596 * indicated by mseq), make note of the migrating cap bits for the
2597 * duration (until we see the corresponding IMPORT).
2598 *
2599 * caller holds s_mutex
2600 */
2601static void handle_cap_export(struct inode *inode, struct ceph_mds_caps *ex,
Sage Weil154f42c2010-06-21 13:45:04 -07002602 struct ceph_mds_session *session,
2603 int *open_target_sessions)
Sage Weila8599bd2009-10-06 11:31:12 -07002604{
2605 struct ceph_inode_info *ci = ceph_inode(inode);
2606 int mds = session->s_mds;
2607 unsigned mseq = le32_to_cpu(ex->migrate_seq);
2608 struct ceph_cap *cap = NULL, *t;
2609 struct rb_node *p;
2610 int remember = 1;
2611
2612 dout("handle_cap_export inode %p ci %p mds%d mseq %d\n",
2613 inode, ci, mds, mseq);
2614
2615 spin_lock(&inode->i_lock);
2616
2617 /* make sure we haven't seen a higher mseq */
2618 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
2619 t = rb_entry(p, struct ceph_cap, ci_node);
2620 if (ceph_seq_cmp(t->mseq, mseq) > 0) {
2621 dout(" higher mseq on cap from mds%d\n",
2622 t->session->s_mds);
2623 remember = 0;
2624 }
2625 if (t->session->s_mds == mds)
2626 cap = t;
2627 }
2628
2629 if (cap) {
2630 if (remember) {
2631 /* make note */
2632 ci->i_cap_exporting_mds = mds;
2633 ci->i_cap_exporting_mseq = mseq;
2634 ci->i_cap_exporting_issued = cap->issued;
Sage Weil154f42c2010-06-21 13:45:04 -07002635
2636 /*
2637 * make sure we have open sessions with all possible
2638 * export targets, so that we get the matching IMPORT
2639 */
2640 *open_target_sessions = 1;
Sage Weila8599bd2009-10-06 11:31:12 -07002641 }
Sage Weil7c1332b2010-02-16 11:39:45 -08002642 __ceph_remove_cap(cap);
Sage Weila8599bd2009-10-06 11:31:12 -07002643 }
Sage Weil4ea00432010-03-16 10:36:40 -07002644 /* else, we already released it */
Sage Weila8599bd2009-10-06 11:31:12 -07002645
2646 spin_unlock(&inode->i_lock);
2647}
2648
2649/*
2650 * Handle cap IMPORT. If there are temp bits from an older EXPORT,
2651 * clean them up.
2652 *
2653 * caller holds s_mutex.
2654 */
2655static void handle_cap_import(struct ceph_mds_client *mdsc,
2656 struct inode *inode, struct ceph_mds_caps *im,
2657 struct ceph_mds_session *session,
2658 void *snaptrace, int snaptrace_len)
2659{
2660 struct ceph_inode_info *ci = ceph_inode(inode);
2661 int mds = session->s_mds;
2662 unsigned issued = le32_to_cpu(im->caps);
2663 unsigned wanted = le32_to_cpu(im->wanted);
2664 unsigned seq = le32_to_cpu(im->seq);
2665 unsigned mseq = le32_to_cpu(im->migrate_seq);
2666 u64 realmino = le64_to_cpu(im->realm);
2667 u64 cap_id = le64_to_cpu(im->cap_id);
2668
2669 if (ci->i_cap_exporting_mds >= 0 &&
2670 ceph_seq_cmp(ci->i_cap_exporting_mseq, mseq) < 0) {
2671 dout("handle_cap_import inode %p ci %p mds%d mseq %d"
2672 " - cleared exporting from mds%d\n",
2673 inode, ci, mds, mseq,
2674 ci->i_cap_exporting_mds);
2675 ci->i_cap_exporting_issued = 0;
2676 ci->i_cap_exporting_mseq = 0;
2677 ci->i_cap_exporting_mds = -1;
2678 } else {
2679 dout("handle_cap_import inode %p ci %p mds%d mseq %d\n",
2680 inode, ci, mds, mseq);
2681 }
2682
2683 down_write(&mdsc->snap_rwsem);
2684 ceph_update_snap_trace(mdsc, snaptrace, snaptrace+snaptrace_len,
2685 false);
2686 downgrade_write(&mdsc->snap_rwsem);
2687 ceph_add_cap(inode, session, cap_id, -1,
2688 issued, wanted, seq, mseq, realmino, CEPH_CAP_FLAG_AUTH,
2689 NULL /* no caps context */);
2690 try_flush_caps(inode, session, NULL);
2691 up_read(&mdsc->snap_rwsem);
Sage Weilfeb4cc92010-11-07 09:39:00 -08002692
2693 /* make sure we re-request max_size, if necessary */
2694 spin_lock(&inode->i_lock);
2695 ci->i_requested_max_size = 0;
2696 spin_unlock(&inode->i_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07002697}
2698
2699/*
2700 * Handle a caps message from the MDS.
2701 *
2702 * Identify the appropriate session, inode, and call the right handler
2703 * based on the cap op.
2704 */
2705void ceph_handle_caps(struct ceph_mds_session *session,
2706 struct ceph_msg *msg)
2707{
2708 struct ceph_mds_client *mdsc = session->s_mdsc;
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002709 struct super_block *sb = mdsc->fsc->sb;
Sage Weila8599bd2009-10-06 11:31:12 -07002710 struct inode *inode;
2711 struct ceph_cap *cap;
2712 struct ceph_mds_caps *h;
Sage Weil2600d2d2010-02-22 15:12:16 -08002713 int mds = session->s_mds;
Sage Weila8599bd2009-10-06 11:31:12 -07002714 int op;
Sage Weil3d7ded42010-06-09 16:47:10 -07002715 u32 seq, mseq;
Sage Weila8599bd2009-10-06 11:31:12 -07002716 struct ceph_vino vino;
2717 u64 cap_id;
2718 u64 size, max_size;
Sage Weil6df058c2009-12-22 11:24:33 -08002719 u64 tid;
Sage Weil70edb552010-03-01 13:20:50 -08002720 void *snaptrace;
Sage Weilce1fbc82010-08-02 15:09:39 -07002721 size_t snaptrace_len;
2722 void *flock;
2723 u32 flock_len;
Sage Weil154f42c2010-06-21 13:45:04 -07002724 int open_target_sessions = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07002725
2726 dout("handle_caps from mds%d\n", mds);
2727
2728 /* decode */
Sage Weil6df058c2009-12-22 11:24:33 -08002729 tid = le64_to_cpu(msg->hdr.tid);
Sage Weila8599bd2009-10-06 11:31:12 -07002730 if (msg->front.iov_len < sizeof(*h))
2731 goto bad;
2732 h = msg->front.iov_base;
2733 op = le32_to_cpu(h->op);
2734 vino.ino = le64_to_cpu(h->ino);
2735 vino.snap = CEPH_NOSNAP;
2736 cap_id = le64_to_cpu(h->cap_id);
2737 seq = le32_to_cpu(h->seq);
Sage Weil3d7ded42010-06-09 16:47:10 -07002738 mseq = le32_to_cpu(h->migrate_seq);
Sage Weila8599bd2009-10-06 11:31:12 -07002739 size = le64_to_cpu(h->size);
2740 max_size = le64_to_cpu(h->max_size);
2741
Sage Weilce1fbc82010-08-02 15:09:39 -07002742 snaptrace = h + 1;
2743 snaptrace_len = le32_to_cpu(h->snap_trace_len);
2744
2745 if (le16_to_cpu(msg->hdr.version) >= 2) {
2746 void *p, *end;
2747
2748 p = snaptrace + snaptrace_len;
2749 end = msg->front.iov_base + msg->front.iov_len;
2750 ceph_decode_32_safe(&p, end, flock_len, bad);
2751 flock = p;
2752 } else {
2753 flock = NULL;
2754 flock_len = 0;
2755 }
2756
Sage Weila8599bd2009-10-06 11:31:12 -07002757 mutex_lock(&session->s_mutex);
2758 session->s_seq++;
2759 dout(" mds%d seq %lld cap seq %u\n", session->s_mds, session->s_seq,
2760 (unsigned)seq);
2761
2762 /* lookup ino */
2763 inode = ceph_find_inode(sb, vino);
2764 dout(" op %s ino %llx.%llx inode %p\n", ceph_cap_op_name(op), vino.ino,
2765 vino.snap, inode);
2766 if (!inode) {
2767 dout(" i don't have ino %llx\n", vino.ino);
Sage Weil3d7ded42010-06-09 16:47:10 -07002768
2769 if (op == CEPH_CAP_OP_IMPORT)
2770 __queue_cap_release(session, vino.ino, cap_id,
2771 mseq, seq);
Greg Farnum21b559d2010-10-06 15:46:30 -07002772 goto flush_cap_releases;
Sage Weila8599bd2009-10-06 11:31:12 -07002773 }
2774
2775 /* these will work even if we don't have a cap yet */
2776 switch (op) {
2777 case CEPH_CAP_OP_FLUSHSNAP_ACK:
Sage Weil6df058c2009-12-22 11:24:33 -08002778 handle_cap_flushsnap_ack(inode, tid, h, session);
Sage Weila8599bd2009-10-06 11:31:12 -07002779 goto done;
2780
2781 case CEPH_CAP_OP_EXPORT:
Sage Weil154f42c2010-06-21 13:45:04 -07002782 handle_cap_export(inode, h, session, &open_target_sessions);
Sage Weila8599bd2009-10-06 11:31:12 -07002783 goto done;
2784
2785 case CEPH_CAP_OP_IMPORT:
2786 handle_cap_import(mdsc, inode, h, session,
Sage Weilce1fbc82010-08-02 15:09:39 -07002787 snaptrace, snaptrace_len);
Sage Weil15637c82010-03-16 13:42:00 -07002788 ceph_check_caps(ceph_inode(inode), CHECK_CAPS_NODELAY,
2789 session);
2790 goto done_unlocked;
Sage Weila8599bd2009-10-06 11:31:12 -07002791 }
2792
2793 /* the rest require a cap */
2794 spin_lock(&inode->i_lock);
2795 cap = __get_cap_for_mds(ceph_inode(inode), mds);
2796 if (!cap) {
Sage Weil9dbd4122010-06-10 13:21:20 -07002797 dout(" no cap on %p ino %llx.%llx from mds%d\n",
Sage Weila8599bd2009-10-06 11:31:12 -07002798 inode, ceph_ino(inode), ceph_snap(inode), mds);
2799 spin_unlock(&inode->i_lock);
Greg Farnum21b559d2010-10-06 15:46:30 -07002800 goto flush_cap_releases;
Sage Weila8599bd2009-10-06 11:31:12 -07002801 }
2802
2803 /* note that each of these drops i_lock for us */
2804 switch (op) {
2805 case CEPH_CAP_OP_REVOKE:
2806 case CEPH_CAP_OP_GRANT:
Sage Weil15637c82010-03-16 13:42:00 -07002807 handle_cap_grant(inode, h, session, cap, msg->middle);
2808 goto done_unlocked;
Sage Weila8599bd2009-10-06 11:31:12 -07002809
2810 case CEPH_CAP_OP_FLUSH_ACK:
Sage Weil6df058c2009-12-22 11:24:33 -08002811 handle_cap_flush_ack(inode, tid, h, session, cap);
Sage Weila8599bd2009-10-06 11:31:12 -07002812 break;
2813
2814 case CEPH_CAP_OP_TRUNC:
2815 handle_cap_trunc(inode, h, session);
2816 break;
2817
2818 default:
2819 spin_unlock(&inode->i_lock);
2820 pr_err("ceph_handle_caps: unknown cap op %d %s\n", op,
2821 ceph_cap_op_name(op));
2822 }
2823
Greg Farnum21b559d2010-10-06 15:46:30 -07002824 goto done;
2825
2826flush_cap_releases:
2827 /*
2828 * send any full release message to try to move things
2829 * along for the mds (who clearly thinks we still have this
2830 * cap).
2831 */
2832 ceph_add_cap_releases(mdsc, session);
2833 ceph_send_cap_releases(mdsc, session);
2834
Sage Weila8599bd2009-10-06 11:31:12 -07002835done:
Sage Weil15637c82010-03-16 13:42:00 -07002836 mutex_unlock(&session->s_mutex);
2837done_unlocked:
Sage Weila8599bd2009-10-06 11:31:12 -07002838 if (inode)
2839 iput(inode);
Sage Weil154f42c2010-06-21 13:45:04 -07002840 if (open_target_sessions)
2841 ceph_mdsc_open_export_target_sessions(mdsc, session);
Sage Weila8599bd2009-10-06 11:31:12 -07002842 return;
2843
2844bad:
2845 pr_err("ceph_handle_caps: corrupt message\n");
Sage Weil9ec7cab2009-12-14 15:13:47 -08002846 ceph_msg_dump(msg);
Sage Weila8599bd2009-10-06 11:31:12 -07002847 return;
2848}
2849
2850/*
2851 * Delayed work handler to process end of delayed cap release LRU list.
2852 */
Sage Weilafcdaea2009-10-14 14:27:38 -07002853void ceph_check_delayed_caps(struct ceph_mds_client *mdsc)
Sage Weila8599bd2009-10-06 11:31:12 -07002854{
2855 struct ceph_inode_info *ci;
2856 int flags = CHECK_CAPS_NODELAY;
2857
Sage Weila8599bd2009-10-06 11:31:12 -07002858 dout("check_delayed_caps\n");
2859 while (1) {
2860 spin_lock(&mdsc->cap_delay_lock);
2861 if (list_empty(&mdsc->cap_delay_list))
2862 break;
2863 ci = list_first_entry(&mdsc->cap_delay_list,
2864 struct ceph_inode_info,
2865 i_cap_delay_list);
2866 if ((ci->i_ceph_flags & CEPH_I_FLUSH) == 0 &&
2867 time_before(jiffies, ci->i_hold_caps_max))
2868 break;
2869 list_del_init(&ci->i_cap_delay_list);
2870 spin_unlock(&mdsc->cap_delay_lock);
2871 dout("check_delayed_caps on %p\n", &ci->vfs_inode);
2872 ceph_check_caps(ci, flags, NULL);
2873 }
2874 spin_unlock(&mdsc->cap_delay_lock);
2875}
2876
2877/*
Sage Weilafcdaea2009-10-14 14:27:38 -07002878 * Flush all dirty caps to the mds
2879 */
2880void ceph_flush_dirty_caps(struct ceph_mds_client *mdsc)
2881{
Sage Weile9964c12010-03-01 15:16:56 -08002882 struct ceph_inode_info *ci, *nci = NULL;
2883 struct inode *inode, *ninode = NULL;
2884 struct list_head *p, *n;
Sage Weilafcdaea2009-10-14 14:27:38 -07002885
2886 dout("flush_dirty_caps\n");
2887 spin_lock(&mdsc->cap_dirty_lock);
Sage Weile9964c12010-03-01 15:16:56 -08002888 list_for_each_safe(p, n, &mdsc->cap_dirty) {
2889 if (nci) {
2890 ci = nci;
2891 inode = ninode;
2892 ci->i_ceph_flags &= ~CEPH_I_NOFLUSH;
2893 dout("flush_dirty_caps inode %p (was next inode)\n",
2894 inode);
2895 } else {
2896 ci = list_entry(p, struct ceph_inode_info,
2897 i_dirty_item);
2898 inode = igrab(&ci->vfs_inode);
2899 BUG_ON(!inode);
2900 dout("flush_dirty_caps inode %p\n", inode);
2901 }
2902 if (n != &mdsc->cap_dirty) {
2903 nci = list_entry(n, struct ceph_inode_info,
2904 i_dirty_item);
2905 ninode = igrab(&nci->vfs_inode);
2906 BUG_ON(!ninode);
2907 nci->i_ceph_flags |= CEPH_I_NOFLUSH;
2908 dout("flush_dirty_caps next inode %p, noflush\n",
2909 ninode);
2910 } else {
2911 nci = NULL;
2912 ninode = NULL;
2913 }
Sage Weilafcdaea2009-10-14 14:27:38 -07002914 spin_unlock(&mdsc->cap_dirty_lock);
2915 if (inode) {
2916 ceph_check_caps(ci, CHECK_CAPS_NODELAY|CHECK_CAPS_FLUSH,
2917 NULL);
2918 iput(inode);
2919 }
2920 spin_lock(&mdsc->cap_dirty_lock);
2921 }
2922 spin_unlock(&mdsc->cap_dirty_lock);
2923}
2924
2925/*
Sage Weila8599bd2009-10-06 11:31:12 -07002926 * Drop open file reference. If we were the last open file,
2927 * we may need to release capabilities to the MDS (or schedule
2928 * their delayed release).
2929 */
2930void ceph_put_fmode(struct ceph_inode_info *ci, int fmode)
2931{
2932 struct inode *inode = &ci->vfs_inode;
2933 int last = 0;
2934
2935 spin_lock(&inode->i_lock);
2936 dout("put_fmode %p fmode %d %d -> %d\n", inode, fmode,
2937 ci->i_nr_by_mode[fmode], ci->i_nr_by_mode[fmode]-1);
2938 BUG_ON(ci->i_nr_by_mode[fmode] == 0);
2939 if (--ci->i_nr_by_mode[fmode] == 0)
2940 last++;
2941 spin_unlock(&inode->i_lock);
2942
2943 if (last && ci->i_vino.snap == CEPH_NOSNAP)
2944 ceph_check_caps(ci, 0, NULL);
2945}
2946
2947/*
2948 * Helpers for embedding cap and dentry lease releases into mds
2949 * requests.
2950 *
2951 * @force is used by dentry_release (below) to force inclusion of a
2952 * record for the directory inode, even when there aren't any caps to
2953 * drop.
2954 */
2955int ceph_encode_inode_release(void **p, struct inode *inode,
2956 int mds, int drop, int unless, int force)
2957{
2958 struct ceph_inode_info *ci = ceph_inode(inode);
2959 struct ceph_cap *cap;
2960 struct ceph_mds_request_release *rel = *p;
Sage Weilec97f882010-06-24 15:12:37 -07002961 int used, dirty;
Sage Weila8599bd2009-10-06 11:31:12 -07002962 int ret = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07002963
2964 spin_lock(&inode->i_lock);
Sage Weil916623d2010-03-16 15:01:07 -07002965 used = __ceph_caps_used(ci);
Sage Weilec97f882010-06-24 15:12:37 -07002966 dirty = __ceph_caps_dirty(ci);
Sage Weil916623d2010-03-16 15:01:07 -07002967
Sage Weilec97f882010-06-24 15:12:37 -07002968 dout("encode_inode_release %p mds%d used|dirty %s drop %s unless %s\n",
2969 inode, mds, ceph_cap_string(used|dirty), ceph_cap_string(drop),
Sage Weil916623d2010-03-16 15:01:07 -07002970 ceph_cap_string(unless));
2971
Sage Weilec97f882010-06-24 15:12:37 -07002972 /* only drop unused, clean caps */
2973 drop &= ~(used | dirty);
Sage Weil916623d2010-03-16 15:01:07 -07002974
Sage Weila8599bd2009-10-06 11:31:12 -07002975 cap = __get_cap_for_mds(ci, mds);
2976 if (cap && __cap_is_valid(cap)) {
2977 if (force ||
2978 ((cap->issued & drop) &&
2979 (cap->issued & unless) == 0)) {
2980 if ((cap->issued & drop) &&
2981 (cap->issued & unless) == 0) {
2982 dout("encode_inode_release %p cap %p %s -> "
2983 "%s\n", inode, cap,
2984 ceph_cap_string(cap->issued),
2985 ceph_cap_string(cap->issued & ~drop));
2986 cap->issued &= ~drop;
2987 cap->implemented &= ~drop;
2988 if (ci->i_ceph_flags & CEPH_I_NODELAY) {
2989 int wanted = __ceph_caps_wanted(ci);
2990 dout(" wanted %s -> %s (act %s)\n",
2991 ceph_cap_string(cap->mds_wanted),
2992 ceph_cap_string(cap->mds_wanted &
2993 ~wanted),
2994 ceph_cap_string(wanted));
2995 cap->mds_wanted &= wanted;
2996 }
2997 } else {
2998 dout("encode_inode_release %p cap %p %s"
2999 " (force)\n", inode, cap,
3000 ceph_cap_string(cap->issued));
3001 }
3002
3003 rel->ino = cpu_to_le64(ceph_ino(inode));
3004 rel->cap_id = cpu_to_le64(cap->cap_id);
3005 rel->seq = cpu_to_le32(cap->seq);
3006 rel->issue_seq = cpu_to_le32(cap->issue_seq),
3007 rel->mseq = cpu_to_le32(cap->mseq);
3008 rel->caps = cpu_to_le32(cap->issued);
3009 rel->wanted = cpu_to_le32(cap->mds_wanted);
3010 rel->dname_len = 0;
3011 rel->dname_seq = 0;
3012 *p += sizeof(*rel);
3013 ret = 1;
3014 } else {
3015 dout("encode_inode_release %p cap %p %s\n",
3016 inode, cap, ceph_cap_string(cap->issued));
3017 }
3018 }
3019 spin_unlock(&inode->i_lock);
3020 return ret;
3021}
3022
3023int ceph_encode_dentry_release(void **p, struct dentry *dentry,
3024 int mds, int drop, int unless)
3025{
3026 struct inode *dir = dentry->d_parent->d_inode;
3027 struct ceph_mds_request_release *rel = *p;
3028 struct ceph_dentry_info *di = ceph_dentry(dentry);
3029 int force = 0;
3030 int ret;
3031
3032 /*
3033 * force an record for the directory caps if we have a dentry lease.
3034 * this is racy (can't take i_lock and d_lock together), but it
3035 * doesn't have to be perfect; the mds will revoke anything we don't
3036 * release.
3037 */
3038 spin_lock(&dentry->d_lock);
3039 if (di->lease_session && di->lease_session->s_mds == mds)
3040 force = 1;
3041 spin_unlock(&dentry->d_lock);
3042
3043 ret = ceph_encode_inode_release(p, dir, mds, drop, unless, force);
3044
3045 spin_lock(&dentry->d_lock);
3046 if (ret && di->lease_session && di->lease_session->s_mds == mds) {
3047 dout("encode_dentry_release %p mds%d seq %d\n",
3048 dentry, mds, (int)di->lease_seq);
3049 rel->dname_len = cpu_to_le32(dentry->d_name.len);
3050 memcpy(*p, dentry->d_name.name, dentry->d_name.len);
3051 *p += dentry->d_name.len;
3052 rel->dname_seq = cpu_to_le32(di->lease_seq);
Sage Weil1dadcce2010-07-23 13:54:21 -07003053 __ceph_mdsc_drop_dentry_lease(dentry);
Sage Weila8599bd2009-10-06 11:31:12 -07003054 }
3055 spin_unlock(&dentry->d_lock);
3056 return ret;
3057}