blob: 0f327c6c967954ec329f25bb44f45a1e4219cd71 [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 /*
Sage Weilc6ffe102011-11-03 09:23:36 -0700490 * if we are newly issued FILE_SHARED, clear D_COMPLETE; we
Sage Weila8599bd2009-10-06 11:31:12 -0700491 * 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++;
Sage Weilc6ffe102011-11-03 09:23:36 -0700497 if (S_ISDIR(ci->vfs_inode.i_mode))
498 ceph_dir_clear_complete(&ci->vfs_inode);
Sage Weila8599bd2009-10-06 11:31:12 -0700499 }
500}
501
502/*
503 * Add a capability under the given MDS session.
504 *
505 * Caller should hold session snap_rwsem (read) and s_mutex.
506 *
507 * @fmode is the open file mode, if we are opening a file, otherwise
508 * it is < 0. (This is so we can atomically add the cap and add an
509 * open file reference to it.)
510 */
511int ceph_add_cap(struct inode *inode,
512 struct ceph_mds_session *session, u64 cap_id,
513 int fmode, unsigned issued, unsigned wanted,
514 unsigned seq, unsigned mseq, u64 realmino, int flags,
515 struct ceph_cap_reservation *caps_reservation)
516{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700517 struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
Sage Weila8599bd2009-10-06 11:31:12 -0700518 struct ceph_inode_info *ci = ceph_inode(inode);
519 struct ceph_cap *new_cap = NULL;
520 struct ceph_cap *cap;
521 int mds = session->s_mds;
522 int actual_wanted;
523
524 dout("add_cap %p mds%d cap %llx %s seq %d\n", inode,
525 session->s_mds, cap_id, ceph_cap_string(issued), seq);
526
527 /*
528 * If we are opening the file, include file mode wanted bits
529 * in wanted.
530 */
531 if (fmode >= 0)
532 wanted |= ceph_caps_for_mode(fmode);
533
534retry:
535 spin_lock(&inode->i_lock);
536 cap = __get_cap_for_mds(ci, mds);
537 if (!cap) {
538 if (new_cap) {
539 cap = new_cap;
540 new_cap = NULL;
541 } else {
542 spin_unlock(&inode->i_lock);
Yehuda Sadeh37151662010-06-17 16:16:12 -0700543 new_cap = get_cap(mdsc, caps_reservation);
Sage Weila8599bd2009-10-06 11:31:12 -0700544 if (new_cap == NULL)
545 return -ENOMEM;
546 goto retry;
547 }
548
549 cap->issued = 0;
550 cap->implemented = 0;
551 cap->mds = mds;
552 cap->mds_wanted = 0;
553
554 cap->ci = ci;
555 __insert_cap_node(ci, cap);
556
557 /* clear out old exporting info? (i.e. on cap import) */
558 if (ci->i_cap_exporting_mds == mds) {
559 ci->i_cap_exporting_issued = 0;
560 ci->i_cap_exporting_mseq = 0;
561 ci->i_cap_exporting_mds = -1;
562 }
563
564 /* add to session cap list */
565 cap->session = session;
566 spin_lock(&session->s_cap_lock);
567 list_add_tail(&cap->session_caps, &session->s_caps);
568 session->s_nr_caps++;
569 spin_unlock(&session->s_cap_lock);
Sage Weil35403032011-05-12 15:13:23 -0700570 } else if (new_cap)
571 ceph_put_cap(mdsc, new_cap);
Sage Weila8599bd2009-10-06 11:31:12 -0700572
573 if (!ci->i_snap_realm) {
574 /*
575 * add this inode to the appropriate snap realm
576 */
577 struct ceph_snap_realm *realm = ceph_lookup_snap_realm(mdsc,
578 realmino);
579 if (realm) {
580 ceph_get_snap_realm(mdsc, realm);
581 spin_lock(&realm->inodes_with_caps_lock);
582 ci->i_snap_realm = realm;
583 list_add(&ci->i_snap_realm_item,
584 &realm->inodes_with_caps);
585 spin_unlock(&realm->inodes_with_caps_lock);
586 } else {
587 pr_err("ceph_add_cap: couldn't find snap realm %llx\n",
588 realmino);
Sage Weilb8cd07e2010-07-16 12:00:02 -0700589 WARN_ON(!realm);
Sage Weila8599bd2009-10-06 11:31:12 -0700590 }
591 }
592
593 __check_cap_issue(ci, cap, issued);
594
595 /*
596 * If we are issued caps we don't want, or the mds' wanted
597 * value appears to be off, queue a check so we'll release
598 * later and/or update the mds wanted value.
599 */
600 actual_wanted = __ceph_caps_wanted(ci);
601 if ((wanted & ~actual_wanted) ||
602 (issued & ~actual_wanted & CEPH_CAP_ANY_WR)) {
603 dout(" issued %s, mds wanted %s, actual %s, queueing\n",
604 ceph_cap_string(issued), ceph_cap_string(wanted),
605 ceph_cap_string(actual_wanted));
606 __cap_delay_requeue(mdsc, ci);
607 }
608
609 if (flags & CEPH_CAP_FLAG_AUTH)
610 ci->i_auth_cap = cap;
611 else if (ci->i_auth_cap == cap)
612 ci->i_auth_cap = NULL;
613
614 dout("add_cap inode %p (%llx.%llx) cap %p %s now %s seq %d mds%d\n",
615 inode, ceph_vinop(inode), cap, ceph_cap_string(issued),
616 ceph_cap_string(issued|cap->issued), seq, mds);
617 cap->cap_id = cap_id;
618 cap->issued = issued;
619 cap->implemented |= issued;
620 cap->mds_wanted |= wanted;
621 cap->seq = seq;
622 cap->issue_seq = seq;
623 cap->mseq = mseq;
Sage Weil685f9a5d2009-11-09 12:05:48 -0800624 cap->cap_gen = session->s_cap_gen;
Sage Weila8599bd2009-10-06 11:31:12 -0700625
626 if (fmode >= 0)
627 __ceph_get_fmode(ci, fmode);
628 spin_unlock(&inode->i_lock);
Yehuda Sadeh03066f22010-07-27 13:11:08 -0700629 wake_up_all(&ci->i_cap_wq);
Sage Weila8599bd2009-10-06 11:31:12 -0700630 return 0;
631}
632
633/*
634 * Return true if cap has not timed out and belongs to the current
635 * generation of the MDS session (i.e. has not gone 'stale' due to
636 * us losing touch with the mds).
637 */
638static int __cap_is_valid(struct ceph_cap *cap)
639{
640 unsigned long ttl;
Sage Weilcdac8302009-11-10 16:02:23 -0800641 u32 gen;
Sage Weila8599bd2009-10-06 11:31:12 -0700642
643 spin_lock(&cap->session->s_cap_lock);
644 gen = cap->session->s_cap_gen;
645 ttl = cap->session->s_cap_ttl;
646 spin_unlock(&cap->session->s_cap_lock);
647
Sage Weil685f9a5d2009-11-09 12:05:48 -0800648 if (cap->cap_gen < gen || time_after_eq(jiffies, ttl)) {
Sage Weila8599bd2009-10-06 11:31:12 -0700649 dout("__cap_is_valid %p cap %p issued %s "
650 "but STALE (gen %u vs %u)\n", &cap->ci->vfs_inode,
Sage Weil685f9a5d2009-11-09 12:05:48 -0800651 cap, ceph_cap_string(cap->issued), cap->cap_gen, gen);
Sage Weila8599bd2009-10-06 11:31:12 -0700652 return 0;
653 }
654
655 return 1;
656}
657
658/*
659 * Return set of valid cap bits issued to us. Note that caps time
660 * out, and may be invalidated in bulk if the client session times out
661 * and session->s_cap_gen is bumped.
662 */
663int __ceph_caps_issued(struct ceph_inode_info *ci, int *implemented)
664{
Sage Weil7af8f1e2010-03-01 15:17:34 -0800665 int have = ci->i_snap_caps | ci->i_cap_exporting_issued;
Sage Weila8599bd2009-10-06 11:31:12 -0700666 struct ceph_cap *cap;
667 struct rb_node *p;
668
669 if (implemented)
670 *implemented = 0;
671 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
672 cap = rb_entry(p, struct ceph_cap, ci_node);
673 if (!__cap_is_valid(cap))
674 continue;
675 dout("__ceph_caps_issued %p cap %p issued %s\n",
676 &ci->vfs_inode, cap, ceph_cap_string(cap->issued));
677 have |= cap->issued;
678 if (implemented)
679 *implemented |= cap->implemented;
680 }
681 return have;
682}
683
684/*
685 * Get cap bits issued by caps other than @ocap
686 */
687int __ceph_caps_issued_other(struct ceph_inode_info *ci, struct ceph_cap *ocap)
688{
689 int have = ci->i_snap_caps;
690 struct ceph_cap *cap;
691 struct rb_node *p;
692
693 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
694 cap = rb_entry(p, struct ceph_cap, ci_node);
695 if (cap == ocap)
696 continue;
697 if (!__cap_is_valid(cap))
698 continue;
699 have |= cap->issued;
700 }
701 return have;
702}
703
704/*
705 * Move a cap to the end of the LRU (oldest caps at list head, newest
706 * at list tail).
707 */
708static void __touch_cap(struct ceph_cap *cap)
709{
710 struct ceph_mds_session *s = cap->session;
711
Sage Weila8599bd2009-10-06 11:31:12 -0700712 spin_lock(&s->s_cap_lock);
Sage Weil7c1332b2010-02-16 11:39:45 -0800713 if (s->s_cap_iterator == NULL) {
Sage Weil5dacf092009-12-21 20:40:34 -0800714 dout("__touch_cap %p cap %p mds%d\n", &cap->ci->vfs_inode, cap,
715 s->s_mds);
716 list_move_tail(&cap->session_caps, &s->s_caps);
717 } else {
718 dout("__touch_cap %p cap %p mds%d NOP, iterating over caps\n",
719 &cap->ci->vfs_inode, cap, s->s_mds);
720 }
Sage Weila8599bd2009-10-06 11:31:12 -0700721 spin_unlock(&s->s_cap_lock);
722}
723
724/*
725 * Check if we hold the given mask. If so, move the cap(s) to the
726 * front of their respective LRUs. (This is the preferred way for
727 * callers to check for caps they want.)
728 */
729int __ceph_caps_issued_mask(struct ceph_inode_info *ci, int mask, int touch)
730{
731 struct ceph_cap *cap;
732 struct rb_node *p;
733 int have = ci->i_snap_caps;
734
735 if ((have & mask) == mask) {
736 dout("__ceph_caps_issued_mask %p snap issued %s"
737 " (mask %s)\n", &ci->vfs_inode,
738 ceph_cap_string(have),
739 ceph_cap_string(mask));
740 return 1;
741 }
742
743 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
744 cap = rb_entry(p, struct ceph_cap, ci_node);
745 if (!__cap_is_valid(cap))
746 continue;
747 if ((cap->issued & mask) == mask) {
748 dout("__ceph_caps_issued_mask %p cap %p issued %s"
749 " (mask %s)\n", &ci->vfs_inode, cap,
750 ceph_cap_string(cap->issued),
751 ceph_cap_string(mask));
752 if (touch)
753 __touch_cap(cap);
754 return 1;
755 }
756
757 /* does a combination of caps satisfy mask? */
758 have |= cap->issued;
759 if ((have & mask) == mask) {
760 dout("__ceph_caps_issued_mask %p combo issued %s"
761 " (mask %s)\n", &ci->vfs_inode,
762 ceph_cap_string(cap->issued),
763 ceph_cap_string(mask));
764 if (touch) {
765 struct rb_node *q;
766
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300767 /* touch this + preceding caps */
Sage Weila8599bd2009-10-06 11:31:12 -0700768 __touch_cap(cap);
769 for (q = rb_first(&ci->i_caps); q != p;
770 q = rb_next(q)) {
771 cap = rb_entry(q, struct ceph_cap,
772 ci_node);
773 if (!__cap_is_valid(cap))
774 continue;
775 __touch_cap(cap);
776 }
777 }
778 return 1;
779 }
780 }
781
782 return 0;
783}
784
785/*
786 * Return true if mask caps are currently being revoked by an MDS.
787 */
788int ceph_caps_revoking(struct ceph_inode_info *ci, int mask)
789{
790 struct inode *inode = &ci->vfs_inode;
791 struct ceph_cap *cap;
792 struct rb_node *p;
793 int ret = 0;
794
795 spin_lock(&inode->i_lock);
796 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
797 cap = rb_entry(p, struct ceph_cap, ci_node);
798 if (__cap_is_valid(cap) &&
799 (cap->implemented & ~cap->issued & mask)) {
800 ret = 1;
801 break;
802 }
803 }
804 spin_unlock(&inode->i_lock);
805 dout("ceph_caps_revoking %p %s = %d\n", inode,
806 ceph_cap_string(mask), ret);
807 return ret;
808}
809
810int __ceph_caps_used(struct ceph_inode_info *ci)
811{
812 int used = 0;
813 if (ci->i_pin_ref)
814 used |= CEPH_CAP_PIN;
815 if (ci->i_rd_ref)
816 used |= CEPH_CAP_FILE_RD;
Sage Weila43fb732010-09-17 09:54:08 -0700817 if (ci->i_rdcache_ref || ci->vfs_inode.i_data.nrpages)
Sage Weila8599bd2009-10-06 11:31:12 -0700818 used |= CEPH_CAP_FILE_CACHE;
819 if (ci->i_wr_ref)
820 used |= CEPH_CAP_FILE_WR;
Henry C Changd3d07202011-05-11 10:29:54 +0000821 if (ci->i_wb_ref || ci->i_wrbuffer_ref)
Sage Weila8599bd2009-10-06 11:31:12 -0700822 used |= CEPH_CAP_FILE_BUFFER;
823 return used;
824}
825
826/*
827 * wanted, by virtue of open file modes
828 */
829int __ceph_caps_file_wanted(struct ceph_inode_info *ci)
830{
831 int want = 0;
832 int mode;
Sage Weil33caad32010-05-26 14:31:27 -0700833 for (mode = 0; mode < CEPH_FILE_MODE_NUM; mode++)
Sage Weila8599bd2009-10-06 11:31:12 -0700834 if (ci->i_nr_by_mode[mode])
835 want |= ceph_caps_for_mode(mode);
836 return want;
837}
838
839/*
840 * Return caps we have registered with the MDS(s) as 'wanted'.
841 */
842int __ceph_caps_mds_wanted(struct ceph_inode_info *ci)
843{
844 struct ceph_cap *cap;
845 struct rb_node *p;
846 int mds_wanted = 0;
847
848 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
849 cap = rb_entry(p, struct ceph_cap, ci_node);
850 if (!__cap_is_valid(cap))
851 continue;
852 mds_wanted |= cap->mds_wanted;
853 }
854 return mds_wanted;
855}
856
857/*
858 * called under i_lock
859 */
860static int __ceph_is_any_caps(struct ceph_inode_info *ci)
861{
862 return !RB_EMPTY_ROOT(&ci->i_caps) || ci->i_cap_exporting_mds >= 0;
863}
864
865/*
Sage Weilf818a732010-05-11 20:56:31 -0700866 * Remove a cap. Take steps to deal with a racing iterate_session_caps.
867 *
Sage Weila6369742010-02-22 13:59:00 -0800868 * caller should hold i_lock.
869 * caller will not hold session s_mutex if called from destroy_inode.
Sage Weila8599bd2009-10-06 11:31:12 -0700870 */
Sage Weil7c1332b2010-02-16 11:39:45 -0800871void __ceph_remove_cap(struct ceph_cap *cap)
Sage Weila8599bd2009-10-06 11:31:12 -0700872{
873 struct ceph_mds_session *session = cap->session;
874 struct ceph_inode_info *ci = cap->ci;
Cheng Renquan640ef792010-03-26 17:40:33 +0800875 struct ceph_mds_client *mdsc =
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700876 ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc;
Sage Weilf818a732010-05-11 20:56:31 -0700877 int removed = 0;
Sage Weila8599bd2009-10-06 11:31:12 -0700878
879 dout("__ceph_remove_cap %p from %p\n", cap, &ci->vfs_inode);
880
Sage Weil7c1332b2010-02-16 11:39:45 -0800881 /* remove from session list */
882 spin_lock(&session->s_cap_lock);
883 if (session->s_cap_iterator == cap) {
884 /* not yet, we are iterating over this very cap */
885 dout("__ceph_remove_cap delaying %p removal from session %p\n",
886 cap, cap->session);
887 } else {
888 list_del_init(&cap->session_caps);
889 session->s_nr_caps--;
890 cap->session = NULL;
Sage Weilf818a732010-05-11 20:56:31 -0700891 removed = 1;
Sage Weil7c1332b2010-02-16 11:39:45 -0800892 }
Sage Weilf818a732010-05-11 20:56:31 -0700893 /* protect backpointer with s_cap_lock: see iterate_session_caps */
894 cap->ci = NULL;
Sage Weil7c1332b2010-02-16 11:39:45 -0800895 spin_unlock(&session->s_cap_lock);
896
Sage Weilf818a732010-05-11 20:56:31 -0700897 /* remove from inode list */
898 rb_erase(&cap->ci_node, &ci->i_caps);
899 if (ci->i_auth_cap == cap)
900 ci->i_auth_cap = NULL;
901
902 if (removed)
Yehuda Sadeh37151662010-06-17 16:16:12 -0700903 ceph_put_cap(mdsc, cap);
Sage Weila8599bd2009-10-06 11:31:12 -0700904
905 if (!__ceph_is_any_caps(ci) && ci->i_snap_realm) {
906 struct ceph_snap_realm *realm = ci->i_snap_realm;
907 spin_lock(&realm->inodes_with_caps_lock);
908 list_del_init(&ci->i_snap_realm_item);
909 ci->i_snap_realm_counter++;
910 ci->i_snap_realm = NULL;
911 spin_unlock(&realm->inodes_with_caps_lock);
912 ceph_put_snap_realm(mdsc, realm);
913 }
914 if (!__ceph_is_any_real_caps(ci))
915 __cap_delay_cancel(mdsc, ci);
916}
917
918/*
919 * Build and send a cap message to the given MDS.
920 *
921 * Caller should be holding s_mutex.
922 */
923static int send_cap_msg(struct ceph_mds_session *session,
924 u64 ino, u64 cid, int op,
925 int caps, int wanted, int dirty,
926 u32 seq, u64 flush_tid, u32 issue_seq, u32 mseq,
927 u64 size, u64 max_size,
928 struct timespec *mtime, struct timespec *atime,
929 u64 time_warp_seq,
930 uid_t uid, gid_t gid, mode_t mode,
931 u64 xattr_version,
932 struct ceph_buffer *xattrs_buf,
933 u64 follows)
934{
935 struct ceph_mds_caps *fc;
936 struct ceph_msg *msg;
937
938 dout("send_cap_msg %s %llx %llx caps %s wanted %s dirty %s"
939 " seq %u/%u mseq %u follows %lld size %llu/%llu"
940 " xattr_ver %llu xattr_len %d\n", ceph_cap_op_name(op),
941 cid, ino, ceph_cap_string(caps), ceph_cap_string(wanted),
942 ceph_cap_string(dirty),
943 seq, issue_seq, mseq, follows, size, max_size,
944 xattr_version, xattrs_buf ? (int)xattrs_buf->vec.iov_len : 0);
945
Sage Weilb61c2762011-08-09 15:03:46 -0700946 msg = ceph_msg_new(CEPH_MSG_CLIENT_CAPS, sizeof(*fc), GFP_NOFS, false);
Sage Weila79832f2010-04-01 16:06:19 -0700947 if (!msg)
948 return -ENOMEM;
Sage Weila8599bd2009-10-06 11:31:12 -0700949
Sage Weil6df058c2009-12-22 11:24:33 -0800950 msg->hdr.tid = cpu_to_le64(flush_tid);
Sage Weila8599bd2009-10-06 11:31:12 -0700951
Sage Weil6df058c2009-12-22 11:24:33 -0800952 fc = msg->front.iov_base;
Sage Weila8599bd2009-10-06 11:31:12 -0700953 memset(fc, 0, sizeof(*fc));
954
955 fc->cap_id = cpu_to_le64(cid);
956 fc->op = cpu_to_le32(op);
957 fc->seq = cpu_to_le32(seq);
Sage Weila8599bd2009-10-06 11:31:12 -0700958 fc->issue_seq = cpu_to_le32(issue_seq);
959 fc->migrate_seq = cpu_to_le32(mseq);
960 fc->caps = cpu_to_le32(caps);
961 fc->wanted = cpu_to_le32(wanted);
962 fc->dirty = cpu_to_le32(dirty);
963 fc->ino = cpu_to_le64(ino);
964 fc->snap_follows = cpu_to_le64(follows);
965
966 fc->size = cpu_to_le64(size);
967 fc->max_size = cpu_to_le64(max_size);
968 if (mtime)
969 ceph_encode_timespec(&fc->mtime, mtime);
970 if (atime)
971 ceph_encode_timespec(&fc->atime, atime);
972 fc->time_warp_seq = cpu_to_le32(time_warp_seq);
973
974 fc->uid = cpu_to_le32(uid);
975 fc->gid = cpu_to_le32(gid);
976 fc->mode = cpu_to_le32(mode);
977
978 fc->xattr_version = cpu_to_le64(xattr_version);
979 if (xattrs_buf) {
980 msg->middle = ceph_buffer_get(xattrs_buf);
981 fc->xattr_len = cpu_to_le32(xattrs_buf->vec.iov_len);
982 msg->hdr.middle_len = cpu_to_le32(xattrs_buf->vec.iov_len);
983 }
984
985 ceph_con_send(&session->s_con, msg);
986 return 0;
987}
988
Sage Weil3d7ded42010-06-09 16:47:10 -0700989static void __queue_cap_release(struct ceph_mds_session *session,
990 u64 ino, u64 cap_id, u32 migrate_seq,
991 u32 issue_seq)
992{
993 struct ceph_msg *msg;
994 struct ceph_mds_cap_release *head;
995 struct ceph_mds_cap_item *item;
996
997 spin_lock(&session->s_cap_lock);
998 BUG_ON(!session->s_num_cap_releases);
999 msg = list_first_entry(&session->s_cap_releases,
1000 struct ceph_msg, list_head);
1001
1002 dout(" adding %llx release to mds%d msg %p (%d left)\n",
1003 ino, session->s_mds, msg, session->s_num_cap_releases);
1004
1005 BUG_ON(msg->front.iov_len + sizeof(*item) > PAGE_CACHE_SIZE);
1006 head = msg->front.iov_base;
1007 head->num = cpu_to_le32(le32_to_cpu(head->num) + 1);
1008 item = msg->front.iov_base + msg->front.iov_len;
1009 item->ino = cpu_to_le64(ino);
1010 item->cap_id = cpu_to_le64(cap_id);
1011 item->migrate_seq = cpu_to_le32(migrate_seq);
1012 item->seq = cpu_to_le32(issue_seq);
1013
1014 session->s_num_cap_releases--;
1015
1016 msg->front.iov_len += sizeof(*item);
1017 if (le32_to_cpu(head->num) == CEPH_CAPS_PER_RELEASE) {
1018 dout(" release msg %p full\n", msg);
1019 list_move_tail(&msg->list_head, &session->s_cap_releases_done);
1020 } else {
1021 dout(" release msg %p at %d/%d (%d)\n", msg,
1022 (int)le32_to_cpu(head->num),
1023 (int)CEPH_CAPS_PER_RELEASE,
1024 (int)msg->front.iov_len);
1025 }
1026 spin_unlock(&session->s_cap_lock);
1027}
1028
Sage Weila8599bd2009-10-06 11:31:12 -07001029/*
Sage Weila6369742010-02-22 13:59:00 -08001030 * Queue cap releases when an inode is dropped from our cache. Since
1031 * inode is about to be destroyed, there is no need for i_lock.
Sage Weila8599bd2009-10-06 11:31:12 -07001032 */
1033void ceph_queue_caps_release(struct inode *inode)
1034{
1035 struct ceph_inode_info *ci = ceph_inode(inode);
1036 struct rb_node *p;
1037
Sage Weila8599bd2009-10-06 11:31:12 -07001038 p = rb_first(&ci->i_caps);
1039 while (p) {
1040 struct ceph_cap *cap = rb_entry(p, struct ceph_cap, ci_node);
1041 struct ceph_mds_session *session = cap->session;
Sage Weila8599bd2009-10-06 11:31:12 -07001042
Sage Weil3d7ded42010-06-09 16:47:10 -07001043 __queue_cap_release(session, ceph_ino(inode), cap->cap_id,
1044 cap->mseq, cap->issue_seq);
Sage Weila8599bd2009-10-06 11:31:12 -07001045 p = rb_next(p);
Sage Weil7c1332b2010-02-16 11:39:45 -08001046 __ceph_remove_cap(cap);
Sage Weila8599bd2009-10-06 11:31:12 -07001047 }
Sage Weila8599bd2009-10-06 11:31:12 -07001048}
1049
1050/*
1051 * Send a cap msg on the given inode. Update our caps state, then
1052 * drop i_lock and send the message.
1053 *
1054 * Make note of max_size reported/requested from mds, revoked caps
1055 * that have now been implemented.
1056 *
1057 * Make half-hearted attempt ot to invalidate page cache if we are
1058 * dropping RDCACHE. Note that this will leave behind locked pages
1059 * that we'll then need to deal with elsewhere.
1060 *
1061 * Return non-zero if delayed release, or we experienced an error
1062 * such that the caller should requeue + retry later.
1063 *
1064 * called with i_lock, then drops it.
1065 * caller should hold snap_rwsem (read), s_mutex.
1066 */
1067static int __send_cap(struct ceph_mds_client *mdsc, struct ceph_cap *cap,
1068 int op, int used, int want, int retain, int flushing,
1069 unsigned *pflush_tid)
1070 __releases(cap->ci->vfs_inode->i_lock)
1071{
1072 struct ceph_inode_info *ci = cap->ci;
1073 struct inode *inode = &ci->vfs_inode;
1074 u64 cap_id = cap->cap_id;
Sage Weil68c28322010-02-09 13:41:47 -08001075 int held, revoking, dropping, keep;
Sage Weila8599bd2009-10-06 11:31:12 -07001076 u64 seq, issue_seq, mseq, time_warp_seq, follows;
1077 u64 size, max_size;
1078 struct timespec mtime, atime;
1079 int wake = 0;
1080 mode_t mode;
1081 uid_t uid;
1082 gid_t gid;
1083 struct ceph_mds_session *session;
1084 u64 xattr_version = 0;
Sage Weil082afec2010-08-22 15:16:41 -07001085 struct ceph_buffer *xattr_blob = NULL;
Sage Weila8599bd2009-10-06 11:31:12 -07001086 int delayed = 0;
1087 u64 flush_tid = 0;
1088 int i;
1089 int ret;
1090
Sage Weil68c28322010-02-09 13:41:47 -08001091 held = cap->issued | cap->implemented;
1092 revoking = cap->implemented & ~cap->issued;
1093 retain &= ~revoking;
1094 dropping = cap->issued & ~retain;
1095
Sage Weila8599bd2009-10-06 11:31:12 -07001096 dout("__send_cap %p cap %p session %p %s -> %s (revoking %s)\n",
1097 inode, cap, cap->session,
1098 ceph_cap_string(held), ceph_cap_string(held & retain),
1099 ceph_cap_string(revoking));
1100 BUG_ON((retain & CEPH_CAP_PIN) == 0);
1101
1102 session = cap->session;
1103
1104 /* don't release wanted unless we've waited a bit. */
1105 if ((ci->i_ceph_flags & CEPH_I_NODELAY) == 0 &&
1106 time_before(jiffies, ci->i_hold_caps_min)) {
1107 dout(" delaying issued %s -> %s, wanted %s -> %s on send\n",
1108 ceph_cap_string(cap->issued),
1109 ceph_cap_string(cap->issued & retain),
1110 ceph_cap_string(cap->mds_wanted),
1111 ceph_cap_string(want));
1112 want |= cap->mds_wanted;
1113 retain |= cap->issued;
1114 delayed = 1;
1115 }
1116 ci->i_ceph_flags &= ~(CEPH_I_NODELAY | CEPH_I_FLUSH);
1117
1118 cap->issued &= retain; /* drop bits we don't want */
1119 if (cap->implemented & ~cap->issued) {
1120 /*
1121 * Wake up any waiters on wanted -> needed transition.
1122 * This is due to the weird transition from buffered
1123 * to sync IO... we need to flush dirty pages _before_
1124 * allowing sync writes to avoid reordering.
1125 */
1126 wake = 1;
1127 }
1128 cap->implemented &= cap->issued | used;
1129 cap->mds_wanted = want;
1130
1131 if (flushing) {
1132 /*
1133 * assign a tid for flush operations so we can avoid
1134 * flush1 -> dirty1 -> flush2 -> flushack1 -> mark
1135 * clean type races. track latest tid for every bit
1136 * so we can handle flush AxFw, flush Fw, and have the
1137 * first ack clean Ax.
1138 */
1139 flush_tid = ++ci->i_cap_flush_last_tid;
1140 if (pflush_tid)
1141 *pflush_tid = flush_tid;
1142 dout(" cap_flush_tid %d\n", (int)flush_tid);
1143 for (i = 0; i < CEPH_CAP_BITS; i++)
1144 if (flushing & (1 << i))
1145 ci->i_cap_flush_tid[i] = flush_tid;
Sage Weil7d8cb262010-08-24 08:44:16 -07001146
1147 follows = ci->i_head_snapc->seq;
1148 } else {
1149 follows = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07001150 }
1151
1152 keep = cap->implemented;
1153 seq = cap->seq;
1154 issue_seq = cap->issue_seq;
1155 mseq = cap->mseq;
1156 size = inode->i_size;
1157 ci->i_reported_size = size;
1158 max_size = ci->i_wanted_max_size;
1159 ci->i_requested_max_size = max_size;
1160 mtime = inode->i_mtime;
1161 atime = inode->i_atime;
1162 time_warp_seq = ci->i_time_warp_seq;
Sage Weila8599bd2009-10-06 11:31:12 -07001163 uid = inode->i_uid;
1164 gid = inode->i_gid;
1165 mode = inode->i_mode;
1166
Sage Weil082afec2010-08-22 15:16:41 -07001167 if (flushing & CEPH_CAP_XATTR_EXCL) {
Sage Weila8599bd2009-10-06 11:31:12 -07001168 __ceph_build_xattrs_blob(ci);
Sage Weil082afec2010-08-22 15:16:41 -07001169 xattr_blob = ci->i_xattrs.blob;
1170 xattr_version = ci->i_xattrs.version;
Sage Weila8599bd2009-10-06 11:31:12 -07001171 }
1172
1173 spin_unlock(&inode->i_lock);
1174
Sage Weila8599bd2009-10-06 11:31:12 -07001175 ret = send_cap_msg(session, ceph_vino(inode).ino, cap_id,
1176 op, keep, want, flushing, seq, flush_tid, issue_seq, mseq,
1177 size, max_size, &mtime, &atime, time_warp_seq,
Sage Weil082afec2010-08-22 15:16:41 -07001178 uid, gid, mode, xattr_version, xattr_blob,
Sage Weila8599bd2009-10-06 11:31:12 -07001179 follows);
1180 if (ret < 0) {
1181 dout("error sending cap msg, must requeue %p\n", inode);
1182 delayed = 1;
1183 }
1184
1185 if (wake)
Yehuda Sadeh03066f22010-07-27 13:11:08 -07001186 wake_up_all(&ci->i_cap_wq);
Sage Weila8599bd2009-10-06 11:31:12 -07001187
1188 return delayed;
1189}
1190
1191/*
1192 * When a snapshot is taken, clients accumulate dirty metadata on
1193 * inodes with capabilities in ceph_cap_snaps to describe the file
1194 * state at the time the snapshot was taken. This must be flushed
1195 * asynchronously back to the MDS once sync writes complete and dirty
1196 * data is written out.
1197 *
Sage Weile8351242010-09-17 08:03:08 -07001198 * Unless @again is true, skip cap_snaps that were already sent to
1199 * the MDS (i.e., during this session).
1200 *
Sage Weila8599bd2009-10-06 11:31:12 -07001201 * Called under i_lock. Takes s_mutex as needed.
1202 */
1203void __ceph_flush_snaps(struct ceph_inode_info *ci,
Sage Weile8351242010-09-17 08:03:08 -07001204 struct ceph_mds_session **psession,
1205 int again)
Yehuda Sadehcd84db62010-06-11 16:58:48 -07001206 __releases(ci->vfs_inode->i_lock)
1207 __acquires(ci->vfs_inode->i_lock)
Sage Weila8599bd2009-10-06 11:31:12 -07001208{
1209 struct inode *inode = &ci->vfs_inode;
1210 int mds;
1211 struct ceph_cap_snap *capsnap;
1212 u32 mseq;
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001213 struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
Sage Weila8599bd2009-10-06 11:31:12 -07001214 struct ceph_mds_session *session = NULL; /* if session != NULL, we hold
1215 session->s_mutex */
1216 u64 next_follows = 0; /* keep track of how far we've gotten through the
1217 i_cap_snaps list, and skip these entries next time
1218 around to avoid an infinite loop */
1219
1220 if (psession)
1221 session = *psession;
1222
1223 dout("__flush_snaps %p\n", inode);
1224retry:
1225 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
1226 /* avoid an infiniute loop after retry */
1227 if (capsnap->follows < next_follows)
1228 continue;
1229 /*
1230 * we need to wait for sync writes to complete and for dirty
1231 * pages to be written out.
1232 */
1233 if (capsnap->dirty_pages || capsnap->writing)
Sage Weilcfc0bf62010-09-14 15:50:59 -07001234 break;
Sage Weila8599bd2009-10-06 11:31:12 -07001235
Sage Weil819ccbf2010-04-01 09:33:46 -07001236 /*
1237 * if cap writeback already occurred, we should have dropped
1238 * the capsnap in ceph_put_wrbuffer_cap_refs.
1239 */
1240 BUG_ON(capsnap->dirty == 0);
1241
Sage Weila8599bd2009-10-06 11:31:12 -07001242 /* pick mds, take s_mutex */
Sage Weilca81f3f2010-06-10 14:21:36 -07001243 if (ci->i_auth_cap == NULL) {
1244 dout("no auth cap (migrating?), doing nothing\n");
1245 goto out;
1246 }
Sage Weile8351242010-09-17 08:03:08 -07001247
1248 /* only flush each capsnap once */
1249 if (!again && !list_empty(&capsnap->flushing_item)) {
1250 dout("already flushed %p, skipping\n", capsnap);
1251 continue;
1252 }
1253
Sage Weilca81f3f2010-06-10 14:21:36 -07001254 mds = ci->i_auth_cap->session->s_mds;
1255 mseq = ci->i_auth_cap->mseq;
1256
Sage Weila8599bd2009-10-06 11:31:12 -07001257 if (session && session->s_mds != mds) {
1258 dout("oops, wrong session %p mutex\n", session);
1259 mutex_unlock(&session->s_mutex);
1260 ceph_put_mds_session(session);
1261 session = NULL;
1262 }
1263 if (!session) {
1264 spin_unlock(&inode->i_lock);
1265 mutex_lock(&mdsc->mutex);
1266 session = __ceph_lookup_mds_session(mdsc, mds);
1267 mutex_unlock(&mdsc->mutex);
1268 if (session) {
1269 dout("inverting session/ino locks on %p\n",
1270 session);
1271 mutex_lock(&session->s_mutex);
1272 }
1273 /*
1274 * if session == NULL, we raced against a cap
Sage Weilca81f3f2010-06-10 14:21:36 -07001275 * deletion or migration. retry, and we'll
1276 * get a better @mds value next time.
Sage Weila8599bd2009-10-06 11:31:12 -07001277 */
1278 spin_lock(&inode->i_lock);
1279 goto retry;
1280 }
1281
1282 capsnap->flush_tid = ++ci->i_cap_flush_last_tid;
1283 atomic_inc(&capsnap->nref);
1284 if (!list_empty(&capsnap->flushing_item))
1285 list_del_init(&capsnap->flushing_item);
1286 list_add_tail(&capsnap->flushing_item,
1287 &session->s_cap_snaps_flushing);
1288 spin_unlock(&inode->i_lock);
1289
Sage Weilcfc0bf62010-09-14 15:50:59 -07001290 dout("flush_snaps %p cap_snap %p follows %lld tid %llu\n",
1291 inode, capsnap, capsnap->follows, capsnap->flush_tid);
Sage Weila8599bd2009-10-06 11:31:12 -07001292 send_cap_msg(session, ceph_vino(inode).ino, 0,
1293 CEPH_CAP_OP_FLUSHSNAP, capsnap->issued, 0,
1294 capsnap->dirty, 0, capsnap->flush_tid, 0, mseq,
1295 capsnap->size, 0,
1296 &capsnap->mtime, &capsnap->atime,
1297 capsnap->time_warp_seq,
1298 capsnap->uid, capsnap->gid, capsnap->mode,
Sage Weil4a625be2010-08-22 15:03:56 -07001299 capsnap->xattr_version, capsnap->xattr_blob,
Sage Weila8599bd2009-10-06 11:31:12 -07001300 capsnap->follows);
1301
1302 next_follows = capsnap->follows + 1;
1303 ceph_put_cap_snap(capsnap);
1304
1305 spin_lock(&inode->i_lock);
1306 goto retry;
1307 }
1308
1309 /* we flushed them all; remove this inode from the queue */
1310 spin_lock(&mdsc->snap_flush_lock);
1311 list_del_init(&ci->i_snap_flush_item);
1312 spin_unlock(&mdsc->snap_flush_lock);
1313
Sage Weilca81f3f2010-06-10 14:21:36 -07001314out:
Sage Weila8599bd2009-10-06 11:31:12 -07001315 if (psession)
1316 *psession = session;
1317 else if (session) {
1318 mutex_unlock(&session->s_mutex);
1319 ceph_put_mds_session(session);
1320 }
1321}
1322
1323static void ceph_flush_snaps(struct ceph_inode_info *ci)
1324{
1325 struct inode *inode = &ci->vfs_inode;
1326
1327 spin_lock(&inode->i_lock);
Sage Weile8351242010-09-17 08:03:08 -07001328 __ceph_flush_snaps(ci, NULL, 0);
Sage Weila8599bd2009-10-06 11:31:12 -07001329 spin_unlock(&inode->i_lock);
1330}
1331
1332/*
Sage Weilfca65b42011-05-04 11:33:47 -07001333 * Mark caps dirty. If inode is newly dirty, return the dirty flags.
1334 * Caller is then responsible for calling __mark_inode_dirty with the
1335 * returned flags value.
Sage Weil76e3b392009-10-15 18:13:53 -07001336 */
Sage Weilfca65b42011-05-04 11:33:47 -07001337int __ceph_mark_dirty_caps(struct ceph_inode_info *ci, int mask)
Sage Weil76e3b392009-10-15 18:13:53 -07001338{
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) {
Sage Weil3772d262011-05-03 09:28:08 -07001360 ihold(inode);
Sage Weil76e3b392009-10-15 18:13:53 -07001361 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;
Sage Weil76e3b392009-10-15 18:13:53 -07001368 __cap_delay_requeue(mdsc, ci);
Sage Weilfca65b42011-05-04 11:33:47 -07001369 return dirty;
Sage Weil76e3b392009-10-15 18:13:53 -07001370}
1371
1372/*
Sage Weila8599bd2009-10-06 11:31:12 -07001373 * Add dirty inode to the flushing list. Assigned a seq number so we
1374 * can wait for caps to flush without starving.
Sage Weilcdc35f92009-10-14 14:24:19 -07001375 *
1376 * Called under i_lock.
Sage Weila8599bd2009-10-06 11:31:12 -07001377 */
Sage Weilcdc35f92009-10-14 14:24:19 -07001378static int __mark_caps_flushing(struct inode *inode,
Sage Weila8599bd2009-10-06 11:31:12 -07001379 struct ceph_mds_session *session)
1380{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001381 struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
Sage Weila8599bd2009-10-06 11:31:12 -07001382 struct ceph_inode_info *ci = ceph_inode(inode);
Sage Weilcdc35f92009-10-14 14:24:19 -07001383 int flushing;
Sage Weil50b885b2009-12-01 14:12:07 -08001384
Sage Weilcdc35f92009-10-14 14:24:19 -07001385 BUG_ON(ci->i_dirty_caps == 0);
Sage Weila8599bd2009-10-06 11:31:12 -07001386 BUG_ON(list_empty(&ci->i_dirty_item));
Sage Weilcdc35f92009-10-14 14:24:19 -07001387
1388 flushing = ci->i_dirty_caps;
1389 dout("__mark_caps_flushing flushing %s, flushing_caps %s -> %s\n",
1390 ceph_cap_string(flushing),
1391 ceph_cap_string(ci->i_flushing_caps),
1392 ceph_cap_string(ci->i_flushing_caps | flushing));
1393 ci->i_flushing_caps |= flushing;
1394 ci->i_dirty_caps = 0;
Sage Weilafcdaea2009-10-14 14:27:38 -07001395 dout(" inode %p now !dirty\n", inode);
Sage Weilcdc35f92009-10-14 14:24:19 -07001396
Sage Weila8599bd2009-10-06 11:31:12 -07001397 spin_lock(&mdsc->cap_dirty_lock);
Sage Weilafcdaea2009-10-14 14:27:38 -07001398 list_del_init(&ci->i_dirty_item);
1399
1400 ci->i_cap_flush_seq = ++mdsc->cap_flush_seq;
Sage Weila8599bd2009-10-06 11:31:12 -07001401 if (list_empty(&ci->i_flushing_item)) {
1402 list_add_tail(&ci->i_flushing_item, &session->s_cap_flushing);
1403 mdsc->num_cap_flushing++;
Sage Weilafcdaea2009-10-14 14:27:38 -07001404 dout(" inode %p now flushing seq %lld\n", inode,
1405 ci->i_cap_flush_seq);
1406 } else {
1407 list_move_tail(&ci->i_flushing_item, &session->s_cap_flushing);
1408 dout(" inode %p now flushing (more) seq %lld\n", inode,
Sage Weila8599bd2009-10-06 11:31:12 -07001409 ci->i_cap_flush_seq);
1410 }
1411 spin_unlock(&mdsc->cap_dirty_lock);
Sage Weilcdc35f92009-10-14 14:24:19 -07001412
1413 return flushing;
Sage Weila8599bd2009-10-06 11:31:12 -07001414}
1415
1416/*
Sage Weil5ecad6f2010-02-17 10:43:37 -08001417 * try to invalidate mapping pages without blocking.
1418 */
Sage Weil5ecad6f2010-02-17 10:43:37 -08001419static int try_nonblocking_invalidate(struct inode *inode)
1420{
1421 struct ceph_inode_info *ci = ceph_inode(inode);
1422 u32 invalidating_gen = ci->i_rdcache_gen;
1423
1424 spin_unlock(&inode->i_lock);
1425 invalidate_mapping_pages(&inode->i_data, 0, -1);
1426 spin_lock(&inode->i_lock);
1427
Sage Weil18a38192010-09-17 10:46:44 -07001428 if (inode->i_data.nrpages == 0 &&
Sage Weil5ecad6f2010-02-17 10:43:37 -08001429 invalidating_gen == ci->i_rdcache_gen) {
1430 /* success. */
1431 dout("try_nonblocking_invalidate %p success\n", inode);
Sage Weilcd045cb2010-11-04 11:05:05 -07001432 /* save any racing async invalidate some trouble */
1433 ci->i_rdcache_revoking = ci->i_rdcache_gen - 1;
Sage Weil5ecad6f2010-02-17 10:43:37 -08001434 return 0;
1435 }
1436 dout("try_nonblocking_invalidate %p failed\n", inode);
1437 return -1;
1438}
1439
1440/*
Sage Weila8599bd2009-10-06 11:31:12 -07001441 * Swiss army knife function to examine currently used and wanted
1442 * versus held caps. Release, flush, ack revoked caps to mds as
1443 * appropriate.
1444 *
1445 * CHECK_CAPS_NODELAY - caller is delayed work and we should not delay
1446 * cap release further.
1447 * CHECK_CAPS_AUTHONLY - we should only check the auth cap
1448 * CHECK_CAPS_FLUSH - we should flush any dirty caps immediately, without
1449 * further delay.
1450 */
1451void ceph_check_caps(struct ceph_inode_info *ci, int flags,
1452 struct ceph_mds_session *session)
1453{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001454 struct ceph_fs_client *fsc = ceph_inode_to_client(&ci->vfs_inode);
1455 struct ceph_mds_client *mdsc = fsc->mdsc;
Sage Weila8599bd2009-10-06 11:31:12 -07001456 struct inode *inode = &ci->vfs_inode;
1457 struct ceph_cap *cap;
1458 int file_wanted, used;
1459 int took_snap_rwsem = 0; /* true if mdsc->snap_rwsem held */
Sage Weilcbd03632010-02-09 13:41:18 -08001460 int issued, implemented, want, retain, revoking, flushing = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07001461 int mds = -1; /* keep track of how far we've gone through i_caps list
1462 to avoid an infinite loop on retry */
1463 struct rb_node *p;
1464 int tried_invalidate = 0;
1465 int delayed = 0, sent = 0, force_requeue = 0, num;
Sage Weilcbd03632010-02-09 13:41:18 -08001466 int queue_invalidate = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07001467 int is_delayed = flags & CHECK_CAPS_NODELAY;
1468
1469 /* if we are unmounting, flush any unused caps immediately. */
1470 if (mdsc->stopping)
1471 is_delayed = 1;
1472
1473 spin_lock(&inode->i_lock);
1474
1475 if (ci->i_ceph_flags & CEPH_I_FLUSH)
1476 flags |= CHECK_CAPS_FLUSH;
1477
1478 /* flush snaps first time around only */
1479 if (!list_empty(&ci->i_cap_snaps))
Sage Weile8351242010-09-17 08:03:08 -07001480 __ceph_flush_snaps(ci, &session, 0);
Sage Weila8599bd2009-10-06 11:31:12 -07001481 goto retry_locked;
1482retry:
1483 spin_lock(&inode->i_lock);
1484retry_locked:
1485 file_wanted = __ceph_caps_file_wanted(ci);
1486 used = __ceph_caps_used(ci);
1487 want = file_wanted | used;
Sage Weilcbd03632010-02-09 13:41:18 -08001488 issued = __ceph_caps_issued(ci, &implemented);
1489 revoking = implemented & ~issued;
Sage Weila8599bd2009-10-06 11:31:12 -07001490
1491 retain = want | CEPH_CAP_PIN;
1492 if (!mdsc->stopping && inode->i_nlink > 0) {
1493 if (want) {
1494 retain |= CEPH_CAP_ANY; /* be greedy */
1495 } else {
1496 retain |= CEPH_CAP_ANY_SHARED;
1497 /*
1498 * keep RD only if we didn't have the file open RW,
1499 * because then the mds would revoke it anyway to
1500 * journal max_size=0.
1501 */
1502 if (ci->i_max_size == 0)
1503 retain |= CEPH_CAP_ANY_RD;
1504 }
1505 }
1506
1507 dout("check_caps %p file_want %s used %s dirty %s flushing %s"
Sage Weilcbd03632010-02-09 13:41:18 -08001508 " issued %s revoking %s retain %s %s%s%s\n", inode,
Sage Weila8599bd2009-10-06 11:31:12 -07001509 ceph_cap_string(file_wanted),
1510 ceph_cap_string(used), ceph_cap_string(ci->i_dirty_caps),
1511 ceph_cap_string(ci->i_flushing_caps),
Sage Weilcbd03632010-02-09 13:41:18 -08001512 ceph_cap_string(issued), ceph_cap_string(revoking),
Sage Weila8599bd2009-10-06 11:31:12 -07001513 ceph_cap_string(retain),
1514 (flags & CHECK_CAPS_AUTHONLY) ? " AUTHONLY" : "",
1515 (flags & CHECK_CAPS_NODELAY) ? " NODELAY" : "",
1516 (flags & CHECK_CAPS_FLUSH) ? " FLUSH" : "");
1517
1518 /*
1519 * If we no longer need to hold onto old our caps, and we may
1520 * have cached pages, but don't want them, then try to invalidate.
1521 * If we fail, it's because pages are locked.... try again later.
1522 */
1523 if ((!is_delayed || mdsc->stopping) &&
1524 ci->i_wrbuffer_ref == 0 && /* no dirty pages... */
Sage Weil93afd442010-09-17 08:38:25 -07001525 inode->i_data.nrpages && /* have cached pages */
Sage Weilcbd03632010-02-09 13:41:18 -08001526 (file_wanted == 0 || /* no open files */
Sage Weil29625072010-05-27 10:40:43 -07001527 (revoking & (CEPH_CAP_FILE_CACHE|
1528 CEPH_CAP_FILE_LAZYIO))) && /* or revoking cache */
Sage Weila8599bd2009-10-06 11:31:12 -07001529 !tried_invalidate) {
Sage Weila8599bd2009-10-06 11:31:12 -07001530 dout("check_caps trying to invalidate on %p\n", inode);
Sage Weil5ecad6f2010-02-17 10:43:37 -08001531 if (try_nonblocking_invalidate(inode) < 0) {
Sage Weil29625072010-05-27 10:40:43 -07001532 if (revoking & (CEPH_CAP_FILE_CACHE|
1533 CEPH_CAP_FILE_LAZYIO)) {
Sage Weil5ecad6f2010-02-17 10:43:37 -08001534 dout("check_caps queuing invalidate\n");
1535 queue_invalidate = 1;
1536 ci->i_rdcache_revoking = ci->i_rdcache_gen;
1537 } else {
1538 dout("check_caps failed to invalidate pages\n");
1539 /* we failed to invalidate pages. check these
1540 caps again later. */
1541 force_requeue = 1;
1542 __cap_set_timeouts(mdsc, ci);
1543 }
Sage Weila8599bd2009-10-06 11:31:12 -07001544 }
1545 tried_invalidate = 1;
1546 goto retry_locked;
1547 }
1548
1549 num = 0;
1550 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
1551 cap = rb_entry(p, struct ceph_cap, ci_node);
1552 num++;
1553
1554 /* avoid looping forever */
1555 if (mds >= cap->mds ||
1556 ((flags & CHECK_CAPS_AUTHONLY) && cap != ci->i_auth_cap))
1557 continue;
1558
1559 /* NOTE: no side-effects allowed, until we take s_mutex */
1560
1561 revoking = cap->implemented & ~cap->issued;
Sage Weil088b3f52011-01-18 08:56:01 -08001562 dout(" mds%d cap %p issued %s implemented %s revoking %s\n",
1563 cap->mds, cap, ceph_cap_string(cap->issued),
1564 ceph_cap_string(cap->implemented),
1565 ceph_cap_string(revoking));
Sage Weila8599bd2009-10-06 11:31:12 -07001566
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 Weil24be0c42011-01-18 08:48:06 -08001661 else
1662 flushing = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07001663
1664 mds = cap->mds; /* remember mds, so we don't repeat */
1665 sent++;
1666
1667 /* __send_cap drops i_lock */
1668 delayed += __send_cap(mdsc, cap, CEPH_CAP_OP_UPDATE, used, want,
1669 retain, flushing, NULL);
1670 goto retry; /* retake i_lock and restart our cap scan. */
1671 }
1672
1673 /*
1674 * Reschedule delayed caps release if we delayed anything,
1675 * otherwise cancel.
1676 */
1677 if (delayed && is_delayed)
1678 force_requeue = 1; /* __send_cap delayed release; requeue */
1679 if (!delayed && !is_delayed)
1680 __cap_delay_cancel(mdsc, ci);
1681 else if (!is_delayed || force_requeue)
1682 __cap_delay_requeue(mdsc, ci);
1683
1684 spin_unlock(&inode->i_lock);
1685
Sage Weilcbd03632010-02-09 13:41:18 -08001686 if (queue_invalidate)
Sage Weil3c6f6b72010-02-09 15:24:44 -08001687 ceph_queue_invalidate(inode);
Sage Weilcbd03632010-02-09 13:41:18 -08001688
Sage Weilcdc2ce02010-03-16 13:39:28 -07001689 if (session)
Sage Weila8599bd2009-10-06 11:31:12 -07001690 mutex_unlock(&session->s_mutex);
1691 if (took_snap_rwsem)
1692 up_read(&mdsc->snap_rwsem);
1693}
1694
1695/*
Sage Weila8599bd2009-10-06 11:31:12 -07001696 * Try to flush dirty caps back to the auth mds.
1697 */
1698static int try_flush_caps(struct inode *inode, struct ceph_mds_session *session,
1699 unsigned *flush_tid)
1700{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001701 struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
Sage Weila8599bd2009-10-06 11:31:12 -07001702 struct ceph_inode_info *ci = ceph_inode(inode);
1703 int unlock_session = session ? 0 : 1;
1704 int flushing = 0;
1705
1706retry:
1707 spin_lock(&inode->i_lock);
Sage Weile9964c12010-03-01 15:16:56 -08001708 if (ci->i_ceph_flags & CEPH_I_NOFLUSH) {
1709 dout("try_flush_caps skipping %p I_NOFLUSH set\n", inode);
1710 goto out;
1711 }
Sage Weila8599bd2009-10-06 11:31:12 -07001712 if (ci->i_dirty_caps && ci->i_auth_cap) {
1713 struct ceph_cap *cap = ci->i_auth_cap;
1714 int used = __ceph_caps_used(ci);
1715 int want = __ceph_caps_wanted(ci);
1716 int delayed;
1717
1718 if (!session) {
1719 spin_unlock(&inode->i_lock);
1720 session = cap->session;
1721 mutex_lock(&session->s_mutex);
1722 goto retry;
1723 }
1724 BUG_ON(session != cap->session);
1725 if (cap->session->s_state < CEPH_MDS_SESSION_OPEN)
1726 goto out;
1727
Sage Weilcdc35f92009-10-14 14:24:19 -07001728 flushing = __mark_caps_flushing(inode, session);
Sage Weila8599bd2009-10-06 11:31:12 -07001729
1730 /* __send_cap drops i_lock */
1731 delayed = __send_cap(mdsc, cap, CEPH_CAP_OP_FLUSH, used, want,
1732 cap->issued | cap->implemented, flushing,
1733 flush_tid);
1734 if (!delayed)
1735 goto out_unlocked;
1736
1737 spin_lock(&inode->i_lock);
1738 __cap_delay_requeue(mdsc, ci);
1739 }
1740out:
1741 spin_unlock(&inode->i_lock);
1742out_unlocked:
1743 if (session && unlock_session)
1744 mutex_unlock(&session->s_mutex);
1745 return flushing;
1746}
1747
1748/*
1749 * Return true if we've flushed caps through the given flush_tid.
1750 */
1751static int caps_are_flushed(struct inode *inode, unsigned tid)
1752{
1753 struct ceph_inode_info *ci = ceph_inode(inode);
Dan Carpentera5ee7512010-05-07 10:27:14 +02001754 int i, ret = 1;
Sage Weila8599bd2009-10-06 11:31:12 -07001755
1756 spin_lock(&inode->i_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001757 for (i = 0; i < CEPH_CAP_BITS; i++)
1758 if ((ci->i_flushing_caps & (1 << i)) &&
1759 ci->i_cap_flush_tid[i] <= tid) {
1760 /* still flushing this bit */
1761 ret = 0;
1762 break;
1763 }
1764 spin_unlock(&inode->i_lock);
1765 return ret;
1766}
1767
1768/*
1769 * Wait on any unsafe replies for the given inode. First wait on the
1770 * newest request, and make that the upper bound. Then, if there are
1771 * more requests, keep waiting on the oldest as long as it is still older
1772 * than the original request.
1773 */
1774static void sync_write_wait(struct inode *inode)
1775{
1776 struct ceph_inode_info *ci = ceph_inode(inode);
1777 struct list_head *head = &ci->i_unsafe_writes;
1778 struct ceph_osd_request *req;
1779 u64 last_tid;
1780
1781 spin_lock(&ci->i_unsafe_lock);
1782 if (list_empty(head))
1783 goto out;
1784
1785 /* set upper bound as _last_ entry in chain */
1786 req = list_entry(head->prev, struct ceph_osd_request,
1787 r_unsafe_item);
1788 last_tid = req->r_tid;
1789
1790 do {
1791 ceph_osdc_get_request(req);
1792 spin_unlock(&ci->i_unsafe_lock);
1793 dout("sync_write_wait on tid %llu (until %llu)\n",
1794 req->r_tid, last_tid);
1795 wait_for_completion(&req->r_safe_completion);
1796 spin_lock(&ci->i_unsafe_lock);
1797 ceph_osdc_put_request(req);
1798
1799 /*
1800 * from here on look at first entry in chain, since we
1801 * only want to wait for anything older than last_tid
1802 */
1803 if (list_empty(head))
1804 break;
1805 req = list_entry(head->next, struct ceph_osd_request,
1806 r_unsafe_item);
1807 } while (req->r_tid < last_tid);
1808out:
1809 spin_unlock(&ci->i_unsafe_lock);
1810}
1811
Josef Bacik02c24a82011-07-16 20:44:56 -04001812int ceph_fsync(struct file *file, loff_t start, loff_t end, int datasync)
Sage Weila8599bd2009-10-06 11:31:12 -07001813{
Christoph Hellwig7ea80852010-05-26 17:53:25 +02001814 struct inode *inode = file->f_mapping->host;
Sage Weila8599bd2009-10-06 11:31:12 -07001815 struct ceph_inode_info *ci = ceph_inode(inode);
1816 unsigned flush_tid;
1817 int ret;
1818 int dirty;
1819
1820 dout("fsync %p%s\n", inode, datasync ? " datasync" : "");
1821 sync_write_wait(inode);
1822
Josef Bacik02c24a82011-07-16 20:44:56 -04001823 ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
Sage Weila8599bd2009-10-06 11:31:12 -07001824 if (ret < 0)
1825 return ret;
Josef Bacik02c24a82011-07-16 20:44:56 -04001826 mutex_lock(&inode->i_mutex);
Sage Weila8599bd2009-10-06 11:31:12 -07001827
1828 dirty = try_flush_caps(inode, NULL, &flush_tid);
1829 dout("fsync dirty caps are %s\n", ceph_cap_string(dirty));
1830
1831 /*
1832 * only wait on non-file metadata writeback (the mds
1833 * can recover size and mtime, so we don't need to
1834 * wait for that)
1835 */
1836 if (!datasync && (dirty & ~CEPH_CAP_ANY_FILE_WR)) {
1837 dout("fsync waiting for flush_tid %u\n", flush_tid);
1838 ret = wait_event_interruptible(ci->i_cap_wq,
1839 caps_are_flushed(inode, flush_tid));
1840 }
1841
1842 dout("fsync %p%s done\n", inode, datasync ? " datasync" : "");
Josef Bacik02c24a82011-07-16 20:44:56 -04001843 mutex_unlock(&inode->i_mutex);
Sage Weila8599bd2009-10-06 11:31:12 -07001844 return ret;
1845}
1846
1847/*
1848 * Flush any dirty caps back to the mds. If we aren't asked to wait,
1849 * queue inode for flush but don't do so immediately, because we can
1850 * get by with fewer MDS messages if we wait for data writeback to
1851 * complete first.
1852 */
Stephen Rothwellf1a3d572010-01-18 11:53:08 +11001853int ceph_write_inode(struct inode *inode, struct writeback_control *wbc)
Sage Weila8599bd2009-10-06 11:31:12 -07001854{
1855 struct ceph_inode_info *ci = ceph_inode(inode);
1856 unsigned flush_tid;
1857 int err = 0;
1858 int dirty;
Stephen Rothwellf1a3d572010-01-18 11:53:08 +11001859 int wait = wbc->sync_mode == WB_SYNC_ALL;
Sage Weila8599bd2009-10-06 11:31:12 -07001860
1861 dout("write_inode %p wait=%d\n", inode, wait);
1862 if (wait) {
1863 dirty = try_flush_caps(inode, NULL, &flush_tid);
1864 if (dirty)
1865 err = wait_event_interruptible(ci->i_cap_wq,
1866 caps_are_flushed(inode, flush_tid));
1867 } else {
Cheng Renquan640ef792010-03-26 17:40:33 +08001868 struct ceph_mds_client *mdsc =
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001869 ceph_sb_to_client(inode->i_sb)->mdsc;
Sage Weila8599bd2009-10-06 11:31:12 -07001870
1871 spin_lock(&inode->i_lock);
1872 if (__ceph_caps_dirty(ci))
1873 __cap_delay_requeue_front(mdsc, ci);
1874 spin_unlock(&inode->i_lock);
1875 }
1876 return err;
1877}
1878
1879/*
1880 * After a recovering MDS goes active, we need to resend any caps
1881 * we were flushing.
1882 *
1883 * Caller holds session->s_mutex.
1884 */
1885static void kick_flushing_capsnaps(struct ceph_mds_client *mdsc,
1886 struct ceph_mds_session *session)
1887{
1888 struct ceph_cap_snap *capsnap;
1889
1890 dout("kick_flushing_capsnaps mds%d\n", session->s_mds);
1891 list_for_each_entry(capsnap, &session->s_cap_snaps_flushing,
1892 flushing_item) {
1893 struct ceph_inode_info *ci = capsnap->ci;
1894 struct inode *inode = &ci->vfs_inode;
1895 struct ceph_cap *cap;
1896
1897 spin_lock(&inode->i_lock);
1898 cap = ci->i_auth_cap;
1899 if (cap && cap->session == session) {
1900 dout("kick_flushing_caps %p cap %p capsnap %p\n", inode,
1901 cap, capsnap);
Sage Weile8351242010-09-17 08:03:08 -07001902 __ceph_flush_snaps(ci, &session, 1);
Sage Weila8599bd2009-10-06 11:31:12 -07001903 } else {
1904 pr_err("%p auth cap %p not mds%d ???\n", inode,
1905 cap, session->s_mds);
Sage Weila8599bd2009-10-06 11:31:12 -07001906 }
Sage Weil0b0c06d2010-04-20 10:27:13 -07001907 spin_unlock(&inode->i_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001908 }
1909}
1910
1911void ceph_kick_flushing_caps(struct ceph_mds_client *mdsc,
1912 struct ceph_mds_session *session)
1913{
1914 struct ceph_inode_info *ci;
1915
1916 kick_flushing_capsnaps(mdsc, session);
1917
1918 dout("kick_flushing_caps mds%d\n", session->s_mds);
1919 list_for_each_entry(ci, &session->s_cap_flushing, i_flushing_item) {
1920 struct inode *inode = &ci->vfs_inode;
1921 struct ceph_cap *cap;
1922 int delayed = 0;
1923
1924 spin_lock(&inode->i_lock);
1925 cap = ci->i_auth_cap;
1926 if (cap && cap->session == session) {
1927 dout("kick_flushing_caps %p cap %p %s\n", inode,
1928 cap, ceph_cap_string(ci->i_flushing_caps));
1929 delayed = __send_cap(mdsc, cap, CEPH_CAP_OP_FLUSH,
1930 __ceph_caps_used(ci),
1931 __ceph_caps_wanted(ci),
1932 cap->issued | cap->implemented,
1933 ci->i_flushing_caps, NULL);
1934 if (delayed) {
1935 spin_lock(&inode->i_lock);
1936 __cap_delay_requeue(mdsc, ci);
1937 spin_unlock(&inode->i_lock);
1938 }
1939 } else {
1940 pr_err("%p auth cap %p not mds%d ???\n", inode,
1941 cap, session->s_mds);
1942 spin_unlock(&inode->i_lock);
1943 }
1944 }
1945}
1946
Sage Weil088b3f52011-01-18 08:56:01 -08001947static void kick_flushing_inode_caps(struct ceph_mds_client *mdsc,
1948 struct ceph_mds_session *session,
1949 struct inode *inode)
1950{
1951 struct ceph_inode_info *ci = ceph_inode(inode);
1952 struct ceph_cap *cap;
1953 int delayed = 0;
1954
1955 spin_lock(&inode->i_lock);
1956 cap = ci->i_auth_cap;
1957 dout("kick_flushing_inode_caps %p flushing %s flush_seq %lld\n", inode,
1958 ceph_cap_string(ci->i_flushing_caps), ci->i_cap_flush_seq);
1959 __ceph_flush_snaps(ci, &session, 1);
1960 if (ci->i_flushing_caps) {
1961 delayed = __send_cap(mdsc, cap, CEPH_CAP_OP_FLUSH,
1962 __ceph_caps_used(ci),
1963 __ceph_caps_wanted(ci),
1964 cap->issued | cap->implemented,
1965 ci->i_flushing_caps, NULL);
1966 if (delayed) {
1967 spin_lock(&inode->i_lock);
1968 __cap_delay_requeue(mdsc, ci);
1969 spin_unlock(&inode->i_lock);
1970 }
1971 } else {
1972 spin_unlock(&inode->i_lock);
1973 }
1974}
1975
Sage Weila8599bd2009-10-06 11:31:12 -07001976
1977/*
1978 * Take references to capabilities we hold, so that we don't release
1979 * them to the MDS prematurely.
1980 *
1981 * Protected by i_lock.
1982 */
1983static void __take_cap_refs(struct ceph_inode_info *ci, int got)
1984{
1985 if (got & CEPH_CAP_PIN)
1986 ci->i_pin_ref++;
1987 if (got & CEPH_CAP_FILE_RD)
1988 ci->i_rd_ref++;
1989 if (got & CEPH_CAP_FILE_CACHE)
1990 ci->i_rdcache_ref++;
1991 if (got & CEPH_CAP_FILE_WR)
1992 ci->i_wr_ref++;
1993 if (got & CEPH_CAP_FILE_BUFFER) {
Henry C Changd3d07202011-05-11 10:29:54 +00001994 if (ci->i_wb_ref == 0)
Sage Weil3772d262011-05-03 09:28:08 -07001995 ihold(&ci->vfs_inode);
Henry C Changd3d07202011-05-11 10:29:54 +00001996 ci->i_wb_ref++;
1997 dout("__take_cap_refs %p wb %d -> %d (?)\n",
1998 &ci->vfs_inode, ci->i_wb_ref-1, ci->i_wb_ref);
Sage Weila8599bd2009-10-06 11:31:12 -07001999 }
2000}
2001
2002/*
2003 * Try to grab cap references. Specify those refs we @want, and the
2004 * minimal set we @need. Also include the larger offset we are writing
2005 * to (when applicable), and check against max_size here as well.
2006 * Note that caller is responsible for ensuring max_size increases are
2007 * requested from the MDS.
2008 */
2009static int try_get_cap_refs(struct ceph_inode_info *ci, int need, int want,
2010 int *got, loff_t endoff, int *check_max, int *err)
2011{
2012 struct inode *inode = &ci->vfs_inode;
2013 int ret = 0;
2014 int have, implemented;
Sage Weil195d3ce2010-03-01 09:57:54 -08002015 int file_wanted;
Sage Weila8599bd2009-10-06 11:31:12 -07002016
2017 dout("get_cap_refs %p need %s want %s\n", inode,
2018 ceph_cap_string(need), ceph_cap_string(want));
2019 spin_lock(&inode->i_lock);
2020
Sage Weil195d3ce2010-03-01 09:57:54 -08002021 /* make sure file is actually open */
2022 file_wanted = __ceph_caps_file_wanted(ci);
2023 if ((file_wanted & need) == 0) {
2024 dout("try_get_cap_refs need %s file_wanted %s, EBADF\n",
2025 ceph_cap_string(need), ceph_cap_string(file_wanted));
Sage Weila8599bd2009-10-06 11:31:12 -07002026 *err = -EBADF;
2027 ret = 1;
2028 goto out;
2029 }
2030
2031 if (need & CEPH_CAP_FILE_WR) {
2032 if (endoff >= 0 && endoff > (loff_t)ci->i_max_size) {
2033 dout("get_cap_refs %p endoff %llu > maxsize %llu\n",
2034 inode, endoff, ci->i_max_size);
2035 if (endoff > ci->i_wanted_max_size) {
2036 *check_max = 1;
2037 ret = 1;
2038 }
2039 goto out;
2040 }
2041 /*
2042 * If a sync write is in progress, we must wait, so that we
2043 * can get a final snapshot value for size+mtime.
2044 */
2045 if (__ceph_have_pending_cap_snap(ci)) {
2046 dout("get_cap_refs %p cap_snap_pending\n", inode);
2047 goto out;
2048 }
2049 }
2050 have = __ceph_caps_issued(ci, &implemented);
2051
2052 /*
2053 * disallow writes while a truncate is pending
2054 */
2055 if (ci->i_truncate_pending)
2056 have &= ~CEPH_CAP_FILE_WR;
2057
2058 if ((have & need) == need) {
2059 /*
2060 * Look at (implemented & ~have & not) so that we keep waiting
2061 * on transition from wanted -> needed caps. This is needed
2062 * for WRBUFFER|WR -> WR to avoid a new WR sync write from
2063 * going before a prior buffered writeback happens.
2064 */
2065 int not = want & ~(have & need);
2066 int revoking = implemented & ~have;
2067 dout("get_cap_refs %p have %s but not %s (revoking %s)\n",
2068 inode, ceph_cap_string(have), ceph_cap_string(not),
2069 ceph_cap_string(revoking));
2070 if ((revoking & not) == 0) {
2071 *got = need | (have & want);
2072 __take_cap_refs(ci, *got);
2073 ret = 1;
2074 }
2075 } else {
2076 dout("get_cap_refs %p have %s needed %s\n", inode,
2077 ceph_cap_string(have), ceph_cap_string(need));
2078 }
2079out:
2080 spin_unlock(&inode->i_lock);
2081 dout("get_cap_refs %p ret %d got %s\n", inode,
2082 ret, ceph_cap_string(*got));
2083 return ret;
2084}
2085
2086/*
2087 * Check the offset we are writing up to against our current
2088 * max_size. If necessary, tell the MDS we want to write to
2089 * a larger offset.
2090 */
2091static void check_max_size(struct inode *inode, loff_t endoff)
2092{
2093 struct ceph_inode_info *ci = ceph_inode(inode);
2094 int check = 0;
2095
2096 /* do we need to explicitly request a larger max_size? */
2097 spin_lock(&inode->i_lock);
2098 if ((endoff >= ci->i_max_size ||
2099 endoff > (inode->i_size << 1)) &&
2100 endoff > ci->i_wanted_max_size) {
2101 dout("write %p at large endoff %llu, req max_size\n",
2102 inode, endoff);
2103 ci->i_wanted_max_size = endoff;
2104 check = 1;
2105 }
2106 spin_unlock(&inode->i_lock);
2107 if (check)
2108 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
2109}
2110
2111/*
2112 * Wait for caps, and take cap references. If we can't get a WR cap
2113 * due to a small max_size, make sure we check_max_size (and possibly
2114 * ask the mds) so we don't get hung up indefinitely.
2115 */
2116int ceph_get_caps(struct ceph_inode_info *ci, int need, int want, int *got,
2117 loff_t endoff)
2118{
2119 int check_max, ret, err;
2120
2121retry:
2122 if (endoff > 0)
2123 check_max_size(&ci->vfs_inode, endoff);
2124 check_max = 0;
2125 err = 0;
2126 ret = wait_event_interruptible(ci->i_cap_wq,
2127 try_get_cap_refs(ci, need, want,
2128 got, endoff,
2129 &check_max, &err));
2130 if (err)
2131 ret = err;
2132 if (check_max)
2133 goto retry;
2134 return ret;
2135}
2136
2137/*
2138 * Take cap refs. Caller must already know we hold at least one ref
2139 * on the caps in question or we don't know this is safe.
2140 */
2141void ceph_get_cap_refs(struct ceph_inode_info *ci, int caps)
2142{
2143 spin_lock(&ci->vfs_inode.i_lock);
2144 __take_cap_refs(ci, caps);
2145 spin_unlock(&ci->vfs_inode.i_lock);
2146}
2147
2148/*
2149 * Release cap refs.
2150 *
2151 * If we released the last ref on any given cap, call ceph_check_caps
2152 * to release (or schedule a release).
2153 *
2154 * If we are releasing a WR cap (from a sync write), finalize any affected
2155 * cap_snap, and wake up any waiters.
2156 */
2157void ceph_put_cap_refs(struct ceph_inode_info *ci, int had)
2158{
2159 struct inode *inode = &ci->vfs_inode;
2160 int last = 0, put = 0, flushsnaps = 0, wake = 0;
2161 struct ceph_cap_snap *capsnap;
2162
2163 spin_lock(&inode->i_lock);
2164 if (had & CEPH_CAP_PIN)
2165 --ci->i_pin_ref;
2166 if (had & CEPH_CAP_FILE_RD)
2167 if (--ci->i_rd_ref == 0)
2168 last++;
2169 if (had & CEPH_CAP_FILE_CACHE)
2170 if (--ci->i_rdcache_ref == 0)
2171 last++;
2172 if (had & CEPH_CAP_FILE_BUFFER) {
Henry C Changd3d07202011-05-11 10:29:54 +00002173 if (--ci->i_wb_ref == 0) {
Sage Weila8599bd2009-10-06 11:31:12 -07002174 last++;
2175 put++;
2176 }
Henry C Changd3d07202011-05-11 10:29:54 +00002177 dout("put_cap_refs %p wb %d -> %d (?)\n",
2178 inode, ci->i_wb_ref+1, ci->i_wb_ref);
Sage Weila8599bd2009-10-06 11:31:12 -07002179 }
2180 if (had & CEPH_CAP_FILE_WR)
2181 if (--ci->i_wr_ref == 0) {
2182 last++;
2183 if (!list_empty(&ci->i_cap_snaps)) {
2184 capsnap = list_first_entry(&ci->i_cap_snaps,
2185 struct ceph_cap_snap,
2186 ci_item);
2187 if (capsnap->writing) {
2188 capsnap->writing = 0;
2189 flushsnaps =
2190 __ceph_finish_cap_snap(ci,
2191 capsnap);
2192 wake = 1;
2193 }
2194 }
2195 }
2196 spin_unlock(&inode->i_lock);
2197
Sage Weil819ccbf2010-04-01 09:33:46 -07002198 dout("put_cap_refs %p had %s%s%s\n", inode, ceph_cap_string(had),
2199 last ? " last" : "", put ? " put" : "");
Sage Weila8599bd2009-10-06 11:31:12 -07002200
2201 if (last && !flushsnaps)
2202 ceph_check_caps(ci, 0, NULL);
2203 else if (flushsnaps)
2204 ceph_flush_snaps(ci);
2205 if (wake)
Yehuda Sadeh03066f22010-07-27 13:11:08 -07002206 wake_up_all(&ci->i_cap_wq);
Sage Weila8599bd2009-10-06 11:31:12 -07002207 if (put)
2208 iput(inode);
2209}
2210
2211/*
2212 * Release @nr WRBUFFER refs on dirty pages for the given @snapc snap
2213 * context. Adjust per-snap dirty page accounting as appropriate.
2214 * Once all dirty data for a cap_snap is flushed, flush snapped file
2215 * metadata back to the MDS. If we dropped the last ref, call
2216 * ceph_check_caps.
2217 */
2218void ceph_put_wrbuffer_cap_refs(struct ceph_inode_info *ci, int nr,
2219 struct ceph_snap_context *snapc)
2220{
2221 struct inode *inode = &ci->vfs_inode;
2222 int last = 0;
Sage Weil819ccbf2010-04-01 09:33:46 -07002223 int complete_capsnap = 0;
2224 int drop_capsnap = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07002225 int found = 0;
2226 struct ceph_cap_snap *capsnap = NULL;
2227
2228 spin_lock(&inode->i_lock);
2229 ci->i_wrbuffer_ref -= nr;
2230 last = !ci->i_wrbuffer_ref;
2231
2232 if (ci->i_head_snapc == snapc) {
2233 ci->i_wrbuffer_ref_head -= nr;
Sage Weil7d8cb262010-08-24 08:44:16 -07002234 if (ci->i_wrbuffer_ref_head == 0 &&
2235 ci->i_dirty_caps == 0 && ci->i_flushing_caps == 0) {
2236 BUG_ON(!ci->i_head_snapc);
Sage Weila8599bd2009-10-06 11:31:12 -07002237 ceph_put_snap_context(ci->i_head_snapc);
2238 ci->i_head_snapc = NULL;
2239 }
2240 dout("put_wrbuffer_cap_refs on %p head %d/%d -> %d/%d %s\n",
2241 inode,
2242 ci->i_wrbuffer_ref+nr, ci->i_wrbuffer_ref_head+nr,
2243 ci->i_wrbuffer_ref, ci->i_wrbuffer_ref_head,
2244 last ? " LAST" : "");
2245 } else {
2246 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
2247 if (capsnap->context == snapc) {
2248 found = 1;
Sage Weila8599bd2009-10-06 11:31:12 -07002249 break;
2250 }
2251 }
2252 BUG_ON(!found);
Sage Weil819ccbf2010-04-01 09:33:46 -07002253 capsnap->dirty_pages -= nr;
2254 if (capsnap->dirty_pages == 0) {
2255 complete_capsnap = 1;
2256 if (capsnap->dirty == 0)
2257 /* cap writeback completed before we created
2258 * the cap_snap; no FLUSHSNAP is needed */
2259 drop_capsnap = 1;
2260 }
Sage Weila8599bd2009-10-06 11:31:12 -07002261 dout("put_wrbuffer_cap_refs on %p cap_snap %p "
Sage Weil819ccbf2010-04-01 09:33:46 -07002262 " snap %lld %d/%d -> %d/%d %s%s%s\n",
Sage Weila8599bd2009-10-06 11:31:12 -07002263 inode, capsnap, capsnap->context->seq,
2264 ci->i_wrbuffer_ref+nr, capsnap->dirty_pages + nr,
2265 ci->i_wrbuffer_ref, capsnap->dirty_pages,
2266 last ? " (wrbuffer last)" : "",
Sage Weil819ccbf2010-04-01 09:33:46 -07002267 complete_capsnap ? " (complete capsnap)" : "",
2268 drop_capsnap ? " (drop capsnap)" : "");
2269 if (drop_capsnap) {
2270 ceph_put_snap_context(capsnap->context);
2271 list_del(&capsnap->ci_item);
2272 list_del(&capsnap->flushing_item);
2273 ceph_put_cap_snap(capsnap);
2274 }
Sage Weila8599bd2009-10-06 11:31:12 -07002275 }
2276
2277 spin_unlock(&inode->i_lock);
2278
2279 if (last) {
2280 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
2281 iput(inode);
Sage Weil819ccbf2010-04-01 09:33:46 -07002282 } else if (complete_capsnap) {
Sage Weila8599bd2009-10-06 11:31:12 -07002283 ceph_flush_snaps(ci);
Yehuda Sadeh03066f22010-07-27 13:11:08 -07002284 wake_up_all(&ci->i_cap_wq);
Sage Weila8599bd2009-10-06 11:31:12 -07002285 }
Sage Weil819ccbf2010-04-01 09:33:46 -07002286 if (drop_capsnap)
2287 iput(inode);
Sage Weila8599bd2009-10-06 11:31:12 -07002288}
2289
2290/*
2291 * Handle a cap GRANT message from the MDS. (Note that a GRANT may
2292 * actually be a revocation if it specifies a smaller cap set.)
2293 *
Sage Weil15637c82010-03-16 13:42:00 -07002294 * caller holds s_mutex and i_lock, we drop both.
2295 *
Sage Weila8599bd2009-10-06 11:31:12 -07002296 * return value:
2297 * 0 - ok
2298 * 1 - check_caps on auth cap only (writeback)
2299 * 2 - check_caps (ack revoke)
2300 */
Sage Weil15637c82010-03-16 13:42:00 -07002301static void handle_cap_grant(struct inode *inode, struct ceph_mds_caps *grant,
2302 struct ceph_mds_session *session,
2303 struct ceph_cap *cap,
2304 struct ceph_buffer *xattr_buf)
Yehuda Sadehcd84db62010-06-11 16:58:48 -07002305 __releases(inode->i_lock)
Sage Weila8599bd2009-10-06 11:31:12 -07002306{
2307 struct ceph_inode_info *ci = ceph_inode(inode);
2308 int mds = session->s_mds;
Sage Weil2f56f562010-10-27 20:59:49 -07002309 int seq = le32_to_cpu(grant->seq);
Sage Weila8599bd2009-10-06 11:31:12 -07002310 int newcaps = le32_to_cpu(grant->caps);
2311 int issued, implemented, used, wanted, dirty;
2312 u64 size = le64_to_cpu(grant->size);
2313 u64 max_size = le64_to_cpu(grant->max_size);
2314 struct timespec mtime, atime, ctime;
Sage Weil15637c82010-03-16 13:42:00 -07002315 int check_caps = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07002316 int wake = 0;
2317 int writeback = 0;
2318 int revoked_rdcache = 0;
Sage Weil3c6f6b72010-02-09 15:24:44 -08002319 int queue_invalidate = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07002320
Sage Weil2f56f562010-10-27 20:59:49 -07002321 dout("handle_cap_grant inode %p cap %p mds%d seq %d %s\n",
2322 inode, cap, mds, seq, ceph_cap_string(newcaps));
Sage Weila8599bd2009-10-06 11:31:12 -07002323 dout(" size %llu max_size %llu, i_size %llu\n", size, max_size,
2324 inode->i_size);
2325
2326 /*
2327 * If CACHE is being revoked, and we have no dirty buffers,
2328 * try to invalidate (once). (If there are dirty buffers, we
2329 * will invalidate _after_ writeback.)
2330 */
Sage Weil3b454c42010-06-10 13:20:33 -07002331 if (((cap->issued & ~newcaps) & CEPH_CAP_FILE_CACHE) &&
2332 (newcaps & CEPH_CAP_FILE_LAZYIO) == 0 &&
Yehuda Sadehbcd2cbd2010-02-19 00:12:21 +00002333 !ci->i_wrbuffer_ref) {
Sage Weil5ecad6f2010-02-17 10:43:37 -08002334 if (try_nonblocking_invalidate(inode) == 0) {
2335 revoked_rdcache = 1;
2336 } else {
Sage Weila8599bd2009-10-06 11:31:12 -07002337 /* there were locked pages.. invalidate later
2338 in a separate thread. */
2339 if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
Sage Weil3c6f6b72010-02-09 15:24:44 -08002340 queue_invalidate = 1;
Sage Weila8599bd2009-10-06 11:31:12 -07002341 ci->i_rdcache_revoking = ci->i_rdcache_gen;
2342 }
Sage Weila8599bd2009-10-06 11:31:12 -07002343 }
Sage Weila8599bd2009-10-06 11:31:12 -07002344 }
2345
2346 /* side effects now are allowed */
2347
2348 issued = __ceph_caps_issued(ci, &implemented);
2349 issued |= implemented | __ceph_caps_dirty(ci);
2350
Sage Weil685f9a5d2009-11-09 12:05:48 -08002351 cap->cap_gen = session->s_cap_gen;
Sage Weila8599bd2009-10-06 11:31:12 -07002352
2353 __check_cap_issue(ci, cap, newcaps);
2354
2355 if ((issued & CEPH_CAP_AUTH_EXCL) == 0) {
2356 inode->i_mode = le32_to_cpu(grant->mode);
2357 inode->i_uid = le32_to_cpu(grant->uid);
2358 inode->i_gid = le32_to_cpu(grant->gid);
2359 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
2360 inode->i_uid, inode->i_gid);
2361 }
2362
2363 if ((issued & CEPH_CAP_LINK_EXCL) == 0)
Miklos Szeredibfe86842011-10-28 14:13:29 +02002364 set_nlink(inode, le32_to_cpu(grant->nlink));
Sage Weila8599bd2009-10-06 11:31:12 -07002365
2366 if ((issued & CEPH_CAP_XATTR_EXCL) == 0 && grant->xattr_len) {
2367 int len = le32_to_cpu(grant->xattr_len);
2368 u64 version = le64_to_cpu(grant->xattr_version);
2369
2370 if (version > ci->i_xattrs.version) {
2371 dout(" got new xattrs v%llu on %p len %d\n",
2372 version, inode, len);
2373 if (ci->i_xattrs.blob)
2374 ceph_buffer_put(ci->i_xattrs.blob);
2375 ci->i_xattrs.blob = ceph_buffer_get(xattr_buf);
2376 ci->i_xattrs.version = version;
2377 }
2378 }
2379
2380 /* size/ctime/mtime/atime? */
2381 ceph_fill_file_size(inode, issued,
2382 le32_to_cpu(grant->truncate_seq),
2383 le64_to_cpu(grant->truncate_size), size);
2384 ceph_decode_timespec(&mtime, &grant->mtime);
2385 ceph_decode_timespec(&atime, &grant->atime);
2386 ceph_decode_timespec(&ctime, &grant->ctime);
2387 ceph_fill_file_time(inode, issued,
2388 le32_to_cpu(grant->time_warp_seq), &ctime, &mtime,
2389 &atime);
2390
2391 /* max size increase? */
2392 if (max_size != ci->i_max_size) {
2393 dout("max_size %lld -> %llu\n", ci->i_max_size, max_size);
2394 ci->i_max_size = max_size;
2395 if (max_size >= ci->i_wanted_max_size) {
2396 ci->i_wanted_max_size = 0; /* reset */
2397 ci->i_requested_max_size = 0;
2398 }
2399 wake = 1;
2400 }
2401
2402 /* check cap bits */
2403 wanted = __ceph_caps_wanted(ci);
2404 used = __ceph_caps_used(ci);
2405 dirty = __ceph_caps_dirty(ci);
2406 dout(" my wanted = %s, used = %s, dirty %s\n",
2407 ceph_cap_string(wanted),
2408 ceph_cap_string(used),
2409 ceph_cap_string(dirty));
2410 if (wanted != le32_to_cpu(grant->wanted)) {
2411 dout("mds wanted %s -> %s\n",
2412 ceph_cap_string(le32_to_cpu(grant->wanted)),
2413 ceph_cap_string(wanted));
2414 grant->wanted = cpu_to_le32(wanted);
2415 }
2416
2417 cap->seq = seq;
2418
2419 /* file layout may have changed */
2420 ci->i_layout = grant->layout;
2421
2422 /* revocation, grant, or no-op? */
2423 if (cap->issued & ~newcaps) {
Sage Weil3b454c42010-06-10 13:20:33 -07002424 int revoking = cap->issued & ~newcaps;
2425
2426 dout("revocation: %s -> %s (revoking %s)\n",
2427 ceph_cap_string(cap->issued),
2428 ceph_cap_string(newcaps),
2429 ceph_cap_string(revoking));
Sage Weil0eb6cd42010-08-05 13:53:18 -07002430 if (revoking & used & CEPH_CAP_FILE_BUFFER)
Sage Weil3b454c42010-06-10 13:20:33 -07002431 writeback = 1; /* initiate writeback; will delay ack */
2432 else if (revoking == CEPH_CAP_FILE_CACHE &&
2433 (newcaps & CEPH_CAP_FILE_LAZYIO) == 0 &&
2434 queue_invalidate)
2435 ; /* do nothing yet, invalidation will be queued */
2436 else if (cap == ci->i_auth_cap)
2437 check_caps = 1; /* check auth cap only */
2438 else
2439 check_caps = 2; /* check all caps */
Sage Weila8599bd2009-10-06 11:31:12 -07002440 cap->issued = newcaps;
Sage Weil978097c2010-03-08 15:27:53 -08002441 cap->implemented |= newcaps;
Sage Weila8599bd2009-10-06 11:31:12 -07002442 } else if (cap->issued == newcaps) {
2443 dout("caps unchanged: %s -> %s\n",
2444 ceph_cap_string(cap->issued), ceph_cap_string(newcaps));
2445 } else {
2446 dout("grant: %s -> %s\n", ceph_cap_string(cap->issued),
2447 ceph_cap_string(newcaps));
2448 cap->issued = newcaps;
2449 cap->implemented |= newcaps; /* add bits only, to
2450 * avoid stepping on a
2451 * pending revocation */
2452 wake = 1;
2453 }
Sage Weil978097c2010-03-08 15:27:53 -08002454 BUG_ON(cap->issued & ~cap->implemented);
Sage Weila8599bd2009-10-06 11:31:12 -07002455
2456 spin_unlock(&inode->i_lock);
Sage Weil3c6f6b72010-02-09 15:24:44 -08002457 if (writeback)
Sage Weila8599bd2009-10-06 11:31:12 -07002458 /*
2459 * queue inode for writeback: we can't actually call
2460 * filemap_write_and_wait, etc. from message handler
2461 * context.
2462 */
Sage Weil3c6f6b72010-02-09 15:24:44 -08002463 ceph_queue_writeback(inode);
2464 if (queue_invalidate)
2465 ceph_queue_invalidate(inode);
Sage Weila8599bd2009-10-06 11:31:12 -07002466 if (wake)
Yehuda Sadeh03066f22010-07-27 13:11:08 -07002467 wake_up_all(&ci->i_cap_wq);
Sage Weil15637c82010-03-16 13:42:00 -07002468
2469 if (check_caps == 1)
2470 ceph_check_caps(ci, CHECK_CAPS_NODELAY|CHECK_CAPS_AUTHONLY,
2471 session);
2472 else if (check_caps == 2)
2473 ceph_check_caps(ci, CHECK_CAPS_NODELAY, session);
2474 else
2475 mutex_unlock(&session->s_mutex);
Sage Weila8599bd2009-10-06 11:31:12 -07002476}
2477
2478/*
2479 * Handle FLUSH_ACK from MDS, indicating that metadata we sent to the
2480 * MDS has been safely committed.
2481 */
Sage Weil6df058c2009-12-22 11:24:33 -08002482static void handle_cap_flush_ack(struct inode *inode, u64 flush_tid,
Sage Weila8599bd2009-10-06 11:31:12 -07002483 struct ceph_mds_caps *m,
2484 struct ceph_mds_session *session,
2485 struct ceph_cap *cap)
2486 __releases(inode->i_lock)
2487{
2488 struct ceph_inode_info *ci = ceph_inode(inode);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002489 struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
Sage Weila8599bd2009-10-06 11:31:12 -07002490 unsigned seq = le32_to_cpu(m->seq);
2491 int dirty = le32_to_cpu(m->dirty);
2492 int cleaned = 0;
Sage Weilafcdaea2009-10-14 14:27:38 -07002493 int drop = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07002494 int i;
2495
2496 for (i = 0; i < CEPH_CAP_BITS; i++)
2497 if ((dirty & (1 << i)) &&
2498 flush_tid == ci->i_cap_flush_tid[i])
2499 cleaned |= 1 << i;
2500
2501 dout("handle_cap_flush_ack inode %p mds%d seq %d on %s cleaned %s,"
2502 " flushing %s -> %s\n",
2503 inode, session->s_mds, seq, ceph_cap_string(dirty),
2504 ceph_cap_string(cleaned), ceph_cap_string(ci->i_flushing_caps),
2505 ceph_cap_string(ci->i_flushing_caps & ~cleaned));
2506
2507 if (ci->i_flushing_caps == (ci->i_flushing_caps & ~cleaned))
2508 goto out;
2509
Sage Weila8599bd2009-10-06 11:31:12 -07002510 ci->i_flushing_caps &= ~cleaned;
Sage Weila8599bd2009-10-06 11:31:12 -07002511
2512 spin_lock(&mdsc->cap_dirty_lock);
2513 if (ci->i_flushing_caps == 0) {
2514 list_del_init(&ci->i_flushing_item);
2515 if (!list_empty(&session->s_cap_flushing))
2516 dout(" mds%d still flushing cap on %p\n",
2517 session->s_mds,
2518 &list_entry(session->s_cap_flushing.next,
2519 struct ceph_inode_info,
2520 i_flushing_item)->vfs_inode);
2521 mdsc->num_cap_flushing--;
Yehuda Sadeh03066f22010-07-27 13:11:08 -07002522 wake_up_all(&mdsc->cap_flushing_wq);
Sage Weila8599bd2009-10-06 11:31:12 -07002523 dout(" inode %p now !flushing\n", inode);
Sage Weilafcdaea2009-10-14 14:27:38 -07002524
2525 if (ci->i_dirty_caps == 0) {
2526 dout(" inode %p now clean\n", inode);
2527 BUG_ON(!list_empty(&ci->i_dirty_item));
2528 drop = 1;
Sage Weil7d8cb262010-08-24 08:44:16 -07002529 if (ci->i_wrbuffer_ref_head == 0) {
2530 BUG_ON(!ci->i_head_snapc);
2531 ceph_put_snap_context(ci->i_head_snapc);
2532 ci->i_head_snapc = NULL;
2533 }
Sage Weil76e3b392009-10-15 18:13:53 -07002534 } else {
2535 BUG_ON(list_empty(&ci->i_dirty_item));
Sage Weilafcdaea2009-10-14 14:27:38 -07002536 }
Sage Weila8599bd2009-10-06 11:31:12 -07002537 }
2538 spin_unlock(&mdsc->cap_dirty_lock);
Yehuda Sadeh03066f22010-07-27 13:11:08 -07002539 wake_up_all(&ci->i_cap_wq);
Sage Weila8599bd2009-10-06 11:31:12 -07002540
2541out:
2542 spin_unlock(&inode->i_lock);
Sage Weilafcdaea2009-10-14 14:27:38 -07002543 if (drop)
Sage Weila8599bd2009-10-06 11:31:12 -07002544 iput(inode);
2545}
2546
2547/*
2548 * Handle FLUSHSNAP_ACK. MDS has flushed snap data to disk and we can
2549 * throw away our cap_snap.
2550 *
2551 * Caller hold s_mutex.
2552 */
Sage Weil6df058c2009-12-22 11:24:33 -08002553static void handle_cap_flushsnap_ack(struct inode *inode, u64 flush_tid,
Sage Weila8599bd2009-10-06 11:31:12 -07002554 struct ceph_mds_caps *m,
2555 struct ceph_mds_session *session)
2556{
2557 struct ceph_inode_info *ci = ceph_inode(inode);
2558 u64 follows = le64_to_cpu(m->snap_follows);
Sage Weila8599bd2009-10-06 11:31:12 -07002559 struct ceph_cap_snap *capsnap;
2560 int drop = 0;
2561
2562 dout("handle_cap_flushsnap_ack inode %p ci %p mds%d follows %lld\n",
2563 inode, ci, session->s_mds, follows);
2564
2565 spin_lock(&inode->i_lock);
2566 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
2567 if (capsnap->follows == follows) {
2568 if (capsnap->flush_tid != flush_tid) {
2569 dout(" cap_snap %p follows %lld tid %lld !="
2570 " %lld\n", capsnap, follows,
2571 flush_tid, capsnap->flush_tid);
2572 break;
2573 }
2574 WARN_ON(capsnap->dirty_pages || capsnap->writing);
Sage Weil819ccbf2010-04-01 09:33:46 -07002575 dout(" removing %p cap_snap %p follows %lld\n",
2576 inode, capsnap, follows);
Sage Weila8599bd2009-10-06 11:31:12 -07002577 ceph_put_snap_context(capsnap->context);
2578 list_del(&capsnap->ci_item);
2579 list_del(&capsnap->flushing_item);
2580 ceph_put_cap_snap(capsnap);
2581 drop = 1;
2582 break;
2583 } else {
2584 dout(" skipping cap_snap %p follows %lld\n",
2585 capsnap, capsnap->follows);
2586 }
2587 }
2588 spin_unlock(&inode->i_lock);
2589 if (drop)
2590 iput(inode);
2591}
2592
2593/*
2594 * Handle TRUNC from MDS, indicating file truncation.
2595 *
2596 * caller hold s_mutex.
2597 */
2598static void handle_cap_trunc(struct inode *inode,
2599 struct ceph_mds_caps *trunc,
2600 struct ceph_mds_session *session)
2601 __releases(inode->i_lock)
2602{
2603 struct ceph_inode_info *ci = ceph_inode(inode);
2604 int mds = session->s_mds;
2605 int seq = le32_to_cpu(trunc->seq);
2606 u32 truncate_seq = le32_to_cpu(trunc->truncate_seq);
2607 u64 truncate_size = le64_to_cpu(trunc->truncate_size);
2608 u64 size = le64_to_cpu(trunc->size);
2609 int implemented = 0;
2610 int dirty = __ceph_caps_dirty(ci);
2611 int issued = __ceph_caps_issued(ceph_inode(inode), &implemented);
2612 int queue_trunc = 0;
2613
2614 issued |= implemented | dirty;
2615
2616 dout("handle_cap_trunc inode %p mds%d seq %d to %lld seq %d\n",
2617 inode, mds, seq, truncate_size, truncate_seq);
2618 queue_trunc = ceph_fill_file_size(inode, issued,
2619 truncate_seq, truncate_size, size);
2620 spin_unlock(&inode->i_lock);
2621
2622 if (queue_trunc)
Sage Weil3c6f6b72010-02-09 15:24:44 -08002623 ceph_queue_vmtruncate(inode);
Sage Weila8599bd2009-10-06 11:31:12 -07002624}
2625
2626/*
2627 * Handle EXPORT from MDS. Cap is being migrated _from_ this mds to a
2628 * different one. If we are the most recent migration we've seen (as
2629 * indicated by mseq), make note of the migrating cap bits for the
2630 * duration (until we see the corresponding IMPORT).
2631 *
2632 * caller holds s_mutex
2633 */
2634static void handle_cap_export(struct inode *inode, struct ceph_mds_caps *ex,
Sage Weil154f42c2010-06-21 13:45:04 -07002635 struct ceph_mds_session *session,
2636 int *open_target_sessions)
Sage Weila8599bd2009-10-06 11:31:12 -07002637{
Sage Weildb354052011-05-24 11:46:31 -07002638 struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
Sage Weila8599bd2009-10-06 11:31:12 -07002639 struct ceph_inode_info *ci = ceph_inode(inode);
2640 int mds = session->s_mds;
2641 unsigned mseq = le32_to_cpu(ex->migrate_seq);
2642 struct ceph_cap *cap = NULL, *t;
2643 struct rb_node *p;
2644 int remember = 1;
2645
2646 dout("handle_cap_export inode %p ci %p mds%d mseq %d\n",
2647 inode, ci, mds, mseq);
2648
2649 spin_lock(&inode->i_lock);
2650
2651 /* make sure we haven't seen a higher mseq */
2652 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
2653 t = rb_entry(p, struct ceph_cap, ci_node);
2654 if (ceph_seq_cmp(t->mseq, mseq) > 0) {
2655 dout(" higher mseq on cap from mds%d\n",
2656 t->session->s_mds);
2657 remember = 0;
2658 }
2659 if (t->session->s_mds == mds)
2660 cap = t;
2661 }
2662
2663 if (cap) {
2664 if (remember) {
2665 /* make note */
2666 ci->i_cap_exporting_mds = mds;
2667 ci->i_cap_exporting_mseq = mseq;
2668 ci->i_cap_exporting_issued = cap->issued;
Sage Weil154f42c2010-06-21 13:45:04 -07002669
2670 /*
2671 * make sure we have open sessions with all possible
2672 * export targets, so that we get the matching IMPORT
2673 */
2674 *open_target_sessions = 1;
Sage Weildb354052011-05-24 11:46:31 -07002675
2676 /*
2677 * we can't flush dirty caps that we've seen the
2678 * EXPORT but no IMPORT for
2679 */
2680 spin_lock(&mdsc->cap_dirty_lock);
2681 if (!list_empty(&ci->i_dirty_item)) {
2682 dout(" moving %p to cap_dirty_migrating\n",
2683 inode);
2684 list_move(&ci->i_dirty_item,
2685 &mdsc->cap_dirty_migrating);
2686 }
2687 spin_unlock(&mdsc->cap_dirty_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07002688 }
Sage Weil7c1332b2010-02-16 11:39:45 -08002689 __ceph_remove_cap(cap);
Sage Weila8599bd2009-10-06 11:31:12 -07002690 }
Sage Weil4ea00432010-03-16 10:36:40 -07002691 /* else, we already released it */
Sage Weila8599bd2009-10-06 11:31:12 -07002692
2693 spin_unlock(&inode->i_lock);
2694}
2695
2696/*
2697 * Handle cap IMPORT. If there are temp bits from an older EXPORT,
2698 * clean them up.
2699 *
2700 * caller holds s_mutex.
2701 */
2702static void handle_cap_import(struct ceph_mds_client *mdsc,
2703 struct inode *inode, struct ceph_mds_caps *im,
2704 struct ceph_mds_session *session,
2705 void *snaptrace, int snaptrace_len)
2706{
2707 struct ceph_inode_info *ci = ceph_inode(inode);
2708 int mds = session->s_mds;
2709 unsigned issued = le32_to_cpu(im->caps);
2710 unsigned wanted = le32_to_cpu(im->wanted);
2711 unsigned seq = le32_to_cpu(im->seq);
2712 unsigned mseq = le32_to_cpu(im->migrate_seq);
2713 u64 realmino = le64_to_cpu(im->realm);
2714 u64 cap_id = le64_to_cpu(im->cap_id);
2715
2716 if (ci->i_cap_exporting_mds >= 0 &&
2717 ceph_seq_cmp(ci->i_cap_exporting_mseq, mseq) < 0) {
2718 dout("handle_cap_import inode %p ci %p mds%d mseq %d"
2719 " - cleared exporting from mds%d\n",
2720 inode, ci, mds, mseq,
2721 ci->i_cap_exporting_mds);
2722 ci->i_cap_exporting_issued = 0;
2723 ci->i_cap_exporting_mseq = 0;
2724 ci->i_cap_exporting_mds = -1;
Sage Weildb354052011-05-24 11:46:31 -07002725
2726 spin_lock(&mdsc->cap_dirty_lock);
2727 if (!list_empty(&ci->i_dirty_item)) {
2728 dout(" moving %p back to cap_dirty\n", inode);
2729 list_move(&ci->i_dirty_item, &mdsc->cap_dirty);
2730 }
2731 spin_unlock(&mdsc->cap_dirty_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07002732 } else {
2733 dout("handle_cap_import inode %p ci %p mds%d mseq %d\n",
2734 inode, ci, mds, mseq);
2735 }
2736
2737 down_write(&mdsc->snap_rwsem);
2738 ceph_update_snap_trace(mdsc, snaptrace, snaptrace+snaptrace_len,
2739 false);
2740 downgrade_write(&mdsc->snap_rwsem);
2741 ceph_add_cap(inode, session, cap_id, -1,
2742 issued, wanted, seq, mseq, realmino, CEPH_CAP_FLAG_AUTH,
2743 NULL /* no caps context */);
Sage Weil088b3f52011-01-18 08:56:01 -08002744 kick_flushing_inode_caps(mdsc, session, inode);
Sage Weila8599bd2009-10-06 11:31:12 -07002745 up_read(&mdsc->snap_rwsem);
Sage Weilfeb4cc92010-11-07 09:39:00 -08002746
2747 /* make sure we re-request max_size, if necessary */
2748 spin_lock(&inode->i_lock);
2749 ci->i_requested_max_size = 0;
2750 spin_unlock(&inode->i_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07002751}
2752
2753/*
2754 * Handle a caps message from the MDS.
2755 *
2756 * Identify the appropriate session, inode, and call the right handler
2757 * based on the cap op.
2758 */
2759void ceph_handle_caps(struct ceph_mds_session *session,
2760 struct ceph_msg *msg)
2761{
2762 struct ceph_mds_client *mdsc = session->s_mdsc;
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002763 struct super_block *sb = mdsc->fsc->sb;
Sage Weila8599bd2009-10-06 11:31:12 -07002764 struct inode *inode;
2765 struct ceph_cap *cap;
2766 struct ceph_mds_caps *h;
Sage Weil2600d2d2010-02-22 15:12:16 -08002767 int mds = session->s_mds;
Sage Weila8599bd2009-10-06 11:31:12 -07002768 int op;
Sage Weil3d7ded42010-06-09 16:47:10 -07002769 u32 seq, mseq;
Sage Weila8599bd2009-10-06 11:31:12 -07002770 struct ceph_vino vino;
2771 u64 cap_id;
2772 u64 size, max_size;
Sage Weil6df058c2009-12-22 11:24:33 -08002773 u64 tid;
Sage Weil70edb552010-03-01 13:20:50 -08002774 void *snaptrace;
Sage Weilce1fbc82010-08-02 15:09:39 -07002775 size_t snaptrace_len;
2776 void *flock;
2777 u32 flock_len;
Sage Weil154f42c2010-06-21 13:45:04 -07002778 int open_target_sessions = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07002779
2780 dout("handle_caps from mds%d\n", mds);
2781
2782 /* decode */
Sage Weil6df058c2009-12-22 11:24:33 -08002783 tid = le64_to_cpu(msg->hdr.tid);
Sage Weila8599bd2009-10-06 11:31:12 -07002784 if (msg->front.iov_len < sizeof(*h))
2785 goto bad;
2786 h = msg->front.iov_base;
2787 op = le32_to_cpu(h->op);
2788 vino.ino = le64_to_cpu(h->ino);
2789 vino.snap = CEPH_NOSNAP;
2790 cap_id = le64_to_cpu(h->cap_id);
2791 seq = le32_to_cpu(h->seq);
Sage Weil3d7ded42010-06-09 16:47:10 -07002792 mseq = le32_to_cpu(h->migrate_seq);
Sage Weila8599bd2009-10-06 11:31:12 -07002793 size = le64_to_cpu(h->size);
2794 max_size = le64_to_cpu(h->max_size);
2795
Sage Weilce1fbc82010-08-02 15:09:39 -07002796 snaptrace = h + 1;
2797 snaptrace_len = le32_to_cpu(h->snap_trace_len);
2798
2799 if (le16_to_cpu(msg->hdr.version) >= 2) {
2800 void *p, *end;
2801
2802 p = snaptrace + snaptrace_len;
2803 end = msg->front.iov_base + msg->front.iov_len;
2804 ceph_decode_32_safe(&p, end, flock_len, bad);
2805 flock = p;
2806 } else {
2807 flock = NULL;
2808 flock_len = 0;
2809 }
2810
Sage Weila8599bd2009-10-06 11:31:12 -07002811 mutex_lock(&session->s_mutex);
2812 session->s_seq++;
2813 dout(" mds%d seq %lld cap seq %u\n", session->s_mds, session->s_seq,
2814 (unsigned)seq);
2815
2816 /* lookup ino */
2817 inode = ceph_find_inode(sb, vino);
2818 dout(" op %s ino %llx.%llx inode %p\n", ceph_cap_op_name(op), vino.ino,
2819 vino.snap, inode);
2820 if (!inode) {
2821 dout(" i don't have ino %llx\n", vino.ino);
Sage Weil3d7ded42010-06-09 16:47:10 -07002822
2823 if (op == CEPH_CAP_OP_IMPORT)
2824 __queue_cap_release(session, vino.ino, cap_id,
2825 mseq, seq);
Greg Farnum21b559d2010-10-06 15:46:30 -07002826 goto flush_cap_releases;
Sage Weila8599bd2009-10-06 11:31:12 -07002827 }
2828
2829 /* these will work even if we don't have a cap yet */
2830 switch (op) {
2831 case CEPH_CAP_OP_FLUSHSNAP_ACK:
Sage Weil6df058c2009-12-22 11:24:33 -08002832 handle_cap_flushsnap_ack(inode, tid, h, session);
Sage Weila8599bd2009-10-06 11:31:12 -07002833 goto done;
2834
2835 case CEPH_CAP_OP_EXPORT:
Sage Weil154f42c2010-06-21 13:45:04 -07002836 handle_cap_export(inode, h, session, &open_target_sessions);
Sage Weila8599bd2009-10-06 11:31:12 -07002837 goto done;
2838
2839 case CEPH_CAP_OP_IMPORT:
2840 handle_cap_import(mdsc, inode, h, session,
Sage Weilce1fbc82010-08-02 15:09:39 -07002841 snaptrace, snaptrace_len);
Sage Weil7e57b812011-01-18 09:00:01 -08002842 ceph_check_caps(ceph_inode(inode), 0, session);
Sage Weil15637c82010-03-16 13:42:00 -07002843 goto done_unlocked;
Sage Weila8599bd2009-10-06 11:31:12 -07002844 }
2845
2846 /* the rest require a cap */
2847 spin_lock(&inode->i_lock);
2848 cap = __get_cap_for_mds(ceph_inode(inode), mds);
2849 if (!cap) {
Sage Weil9dbd4122010-06-10 13:21:20 -07002850 dout(" no cap on %p ino %llx.%llx from mds%d\n",
Sage Weila8599bd2009-10-06 11:31:12 -07002851 inode, ceph_ino(inode), ceph_snap(inode), mds);
2852 spin_unlock(&inode->i_lock);
Greg Farnum21b559d2010-10-06 15:46:30 -07002853 goto flush_cap_releases;
Sage Weila8599bd2009-10-06 11:31:12 -07002854 }
2855
2856 /* note that each of these drops i_lock for us */
2857 switch (op) {
2858 case CEPH_CAP_OP_REVOKE:
2859 case CEPH_CAP_OP_GRANT:
Sage Weil15637c82010-03-16 13:42:00 -07002860 handle_cap_grant(inode, h, session, cap, msg->middle);
2861 goto done_unlocked;
Sage Weila8599bd2009-10-06 11:31:12 -07002862
2863 case CEPH_CAP_OP_FLUSH_ACK:
Sage Weil6df058c2009-12-22 11:24:33 -08002864 handle_cap_flush_ack(inode, tid, h, session, cap);
Sage Weila8599bd2009-10-06 11:31:12 -07002865 break;
2866
2867 case CEPH_CAP_OP_TRUNC:
2868 handle_cap_trunc(inode, h, session);
2869 break;
2870
2871 default:
2872 spin_unlock(&inode->i_lock);
2873 pr_err("ceph_handle_caps: unknown cap op %d %s\n", op,
2874 ceph_cap_op_name(op));
2875 }
2876
Greg Farnum21b559d2010-10-06 15:46:30 -07002877 goto done;
2878
2879flush_cap_releases:
2880 /*
2881 * send any full release message to try to move things
2882 * along for the mds (who clearly thinks we still have this
2883 * cap).
2884 */
2885 ceph_add_cap_releases(mdsc, session);
2886 ceph_send_cap_releases(mdsc, session);
2887
Sage Weila8599bd2009-10-06 11:31:12 -07002888done:
Sage Weil15637c82010-03-16 13:42:00 -07002889 mutex_unlock(&session->s_mutex);
2890done_unlocked:
Sage Weila8599bd2009-10-06 11:31:12 -07002891 if (inode)
2892 iput(inode);
Sage Weil154f42c2010-06-21 13:45:04 -07002893 if (open_target_sessions)
2894 ceph_mdsc_open_export_target_sessions(mdsc, session);
Sage Weila8599bd2009-10-06 11:31:12 -07002895 return;
2896
2897bad:
2898 pr_err("ceph_handle_caps: corrupt message\n");
Sage Weil9ec7cab2009-12-14 15:13:47 -08002899 ceph_msg_dump(msg);
Sage Weila8599bd2009-10-06 11:31:12 -07002900 return;
2901}
2902
2903/*
2904 * Delayed work handler to process end of delayed cap release LRU list.
2905 */
Sage Weilafcdaea2009-10-14 14:27:38 -07002906void ceph_check_delayed_caps(struct ceph_mds_client *mdsc)
Sage Weila8599bd2009-10-06 11:31:12 -07002907{
2908 struct ceph_inode_info *ci;
2909 int flags = CHECK_CAPS_NODELAY;
2910
Sage Weila8599bd2009-10-06 11:31:12 -07002911 dout("check_delayed_caps\n");
2912 while (1) {
2913 spin_lock(&mdsc->cap_delay_lock);
2914 if (list_empty(&mdsc->cap_delay_list))
2915 break;
2916 ci = list_first_entry(&mdsc->cap_delay_list,
2917 struct ceph_inode_info,
2918 i_cap_delay_list);
2919 if ((ci->i_ceph_flags & CEPH_I_FLUSH) == 0 &&
2920 time_before(jiffies, ci->i_hold_caps_max))
2921 break;
2922 list_del_init(&ci->i_cap_delay_list);
2923 spin_unlock(&mdsc->cap_delay_lock);
2924 dout("check_delayed_caps on %p\n", &ci->vfs_inode);
2925 ceph_check_caps(ci, flags, NULL);
2926 }
2927 spin_unlock(&mdsc->cap_delay_lock);
2928}
2929
2930/*
Sage Weilafcdaea2009-10-14 14:27:38 -07002931 * Flush all dirty caps to the mds
2932 */
2933void ceph_flush_dirty_caps(struct ceph_mds_client *mdsc)
2934{
Sage Weildb354052011-05-24 11:46:31 -07002935 struct ceph_inode_info *ci;
2936 struct inode *inode;
Sage Weilafcdaea2009-10-14 14:27:38 -07002937
2938 dout("flush_dirty_caps\n");
2939 spin_lock(&mdsc->cap_dirty_lock);
Sage Weildb354052011-05-24 11:46:31 -07002940 while (!list_empty(&mdsc->cap_dirty)) {
2941 ci = list_first_entry(&mdsc->cap_dirty, struct ceph_inode_info,
2942 i_dirty_item);
Sage Weil70b666c2011-05-27 09:24:26 -07002943 inode = &ci->vfs_inode;
2944 ihold(inode);
Sage Weildb354052011-05-24 11:46:31 -07002945 dout("flush_dirty_caps %p\n", inode);
Sage Weilafcdaea2009-10-14 14:27:38 -07002946 spin_unlock(&mdsc->cap_dirty_lock);
Sage Weil70b666c2011-05-27 09:24:26 -07002947 ceph_check_caps(ci, CHECK_CAPS_NODELAY|CHECK_CAPS_FLUSH, NULL);
2948 iput(inode);
Sage Weilafcdaea2009-10-14 14:27:38 -07002949 spin_lock(&mdsc->cap_dirty_lock);
2950 }
2951 spin_unlock(&mdsc->cap_dirty_lock);
Sage Weildb354052011-05-24 11:46:31 -07002952 dout("flush_dirty_caps done\n");
Sage Weilafcdaea2009-10-14 14:27:38 -07002953}
2954
2955/*
Sage Weila8599bd2009-10-06 11:31:12 -07002956 * Drop open file reference. If we were the last open file,
2957 * we may need to release capabilities to the MDS (or schedule
2958 * their delayed release).
2959 */
2960void ceph_put_fmode(struct ceph_inode_info *ci, int fmode)
2961{
2962 struct inode *inode = &ci->vfs_inode;
2963 int last = 0;
2964
2965 spin_lock(&inode->i_lock);
2966 dout("put_fmode %p fmode %d %d -> %d\n", inode, fmode,
2967 ci->i_nr_by_mode[fmode], ci->i_nr_by_mode[fmode]-1);
2968 BUG_ON(ci->i_nr_by_mode[fmode] == 0);
2969 if (--ci->i_nr_by_mode[fmode] == 0)
2970 last++;
2971 spin_unlock(&inode->i_lock);
2972
2973 if (last && ci->i_vino.snap == CEPH_NOSNAP)
2974 ceph_check_caps(ci, 0, NULL);
2975}
2976
2977/*
2978 * Helpers for embedding cap and dentry lease releases into mds
2979 * requests.
2980 *
2981 * @force is used by dentry_release (below) to force inclusion of a
2982 * record for the directory inode, even when there aren't any caps to
2983 * drop.
2984 */
2985int ceph_encode_inode_release(void **p, struct inode *inode,
2986 int mds, int drop, int unless, int force)
2987{
2988 struct ceph_inode_info *ci = ceph_inode(inode);
2989 struct ceph_cap *cap;
2990 struct ceph_mds_request_release *rel = *p;
Sage Weilec97f882010-06-24 15:12:37 -07002991 int used, dirty;
Sage Weila8599bd2009-10-06 11:31:12 -07002992 int ret = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07002993
2994 spin_lock(&inode->i_lock);
Sage Weil916623d2010-03-16 15:01:07 -07002995 used = __ceph_caps_used(ci);
Sage Weilec97f882010-06-24 15:12:37 -07002996 dirty = __ceph_caps_dirty(ci);
Sage Weil916623d2010-03-16 15:01:07 -07002997
Sage Weilec97f882010-06-24 15:12:37 -07002998 dout("encode_inode_release %p mds%d used|dirty %s drop %s unless %s\n",
2999 inode, mds, ceph_cap_string(used|dirty), ceph_cap_string(drop),
Sage Weil916623d2010-03-16 15:01:07 -07003000 ceph_cap_string(unless));
3001
Sage Weilec97f882010-06-24 15:12:37 -07003002 /* only drop unused, clean caps */
3003 drop &= ~(used | dirty);
Sage Weil916623d2010-03-16 15:01:07 -07003004
Sage Weila8599bd2009-10-06 11:31:12 -07003005 cap = __get_cap_for_mds(ci, mds);
3006 if (cap && __cap_is_valid(cap)) {
3007 if (force ||
3008 ((cap->issued & drop) &&
3009 (cap->issued & unless) == 0)) {
3010 if ((cap->issued & drop) &&
3011 (cap->issued & unless) == 0) {
3012 dout("encode_inode_release %p cap %p %s -> "
3013 "%s\n", inode, cap,
3014 ceph_cap_string(cap->issued),
3015 ceph_cap_string(cap->issued & ~drop));
3016 cap->issued &= ~drop;
3017 cap->implemented &= ~drop;
3018 if (ci->i_ceph_flags & CEPH_I_NODELAY) {
3019 int wanted = __ceph_caps_wanted(ci);
3020 dout(" wanted %s -> %s (act %s)\n",
3021 ceph_cap_string(cap->mds_wanted),
3022 ceph_cap_string(cap->mds_wanted &
3023 ~wanted),
3024 ceph_cap_string(wanted));
3025 cap->mds_wanted &= wanted;
3026 }
3027 } else {
3028 dout("encode_inode_release %p cap %p %s"
3029 " (force)\n", inode, cap,
3030 ceph_cap_string(cap->issued));
3031 }
3032
3033 rel->ino = cpu_to_le64(ceph_ino(inode));
3034 rel->cap_id = cpu_to_le64(cap->cap_id);
3035 rel->seq = cpu_to_le32(cap->seq);
3036 rel->issue_seq = cpu_to_le32(cap->issue_seq),
3037 rel->mseq = cpu_to_le32(cap->mseq);
3038 rel->caps = cpu_to_le32(cap->issued);
3039 rel->wanted = cpu_to_le32(cap->mds_wanted);
3040 rel->dname_len = 0;
3041 rel->dname_seq = 0;
3042 *p += sizeof(*rel);
3043 ret = 1;
3044 } else {
3045 dout("encode_inode_release %p cap %p %s\n",
3046 inode, cap, ceph_cap_string(cap->issued));
3047 }
3048 }
3049 spin_unlock(&inode->i_lock);
3050 return ret;
3051}
3052
3053int ceph_encode_dentry_release(void **p, struct dentry *dentry,
3054 int mds, int drop, int unless)
3055{
3056 struct inode *dir = dentry->d_parent->d_inode;
3057 struct ceph_mds_request_release *rel = *p;
3058 struct ceph_dentry_info *di = ceph_dentry(dentry);
3059 int force = 0;
3060 int ret;
3061
3062 /*
3063 * force an record for the directory caps if we have a dentry lease.
3064 * this is racy (can't take i_lock and d_lock together), but it
3065 * doesn't have to be perfect; the mds will revoke anything we don't
3066 * release.
3067 */
3068 spin_lock(&dentry->d_lock);
3069 if (di->lease_session && di->lease_session->s_mds == mds)
3070 force = 1;
3071 spin_unlock(&dentry->d_lock);
3072
3073 ret = ceph_encode_inode_release(p, dir, mds, drop, unless, force);
3074
3075 spin_lock(&dentry->d_lock);
3076 if (ret && di->lease_session && di->lease_session->s_mds == mds) {
3077 dout("encode_dentry_release %p mds%d seq %d\n",
3078 dentry, mds, (int)di->lease_seq);
3079 rel->dname_len = cpu_to_le32(dentry->d_name.len);
3080 memcpy(*p, dentry->d_name.name, dentry->d_name.len);
3081 *p += dentry->d_name.len;
3082 rel->dname_seq = cpu_to_le32(di->lease_seq);
Sage Weil1dadcce2010-07-23 13:54:21 -07003083 __ceph_mdsc_drop_dentry_lease(dentry);
Sage Weila8599bd2009-10-06 11:31:12 -07003084 }
3085 spin_unlock(&dentry->d_lock);
3086 return ret;
3087}