blob: 13976c33332ec1fd7ca3999053b15b7079c5ab31 [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"
Milosz Tanski99ccbd22013-08-21 17:29:54 -040013#include "cache.h"
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -070014#include <linux/ceph/decode.h>
15#include <linux/ceph/messenger.h>
Sage Weila8599bd2009-10-06 11:31:12 -070016
17/*
18 * Capability management
19 *
20 * The Ceph metadata servers control client access to inode metadata
21 * and file data by issuing capabilities, granting clients permission
22 * to read and/or write both inode field and file data to OSDs
23 * (storage nodes). Each capability consists of a set of bits
24 * indicating which operations are allowed.
25 *
26 * If the client holds a *_SHARED cap, the client has a coherent value
27 * that can be safely read from the cached inode.
28 *
29 * In the case of a *_EXCL (exclusive) or FILE_WR capabilities, the
30 * client is allowed to change inode attributes (e.g., file size,
31 * mtime), note its dirty state in the ceph_cap, and asynchronously
32 * flush that metadata change to the MDS.
33 *
34 * In the event of a conflicting operation (perhaps by another
35 * client), the MDS will revoke the conflicting client capabilities.
36 *
37 * In order for a client to cache an inode, it must hold a capability
38 * with at least one MDS server. When inodes are released, release
39 * notifications are batched and periodically sent en masse to the MDS
40 * cluster to release server state.
41 */
42
43
44/*
45 * Generate readable cap strings for debugging output.
46 */
47#define MAX_CAP_STR 20
48static char cap_str[MAX_CAP_STR][40];
49static DEFINE_SPINLOCK(cap_str_lock);
50static int last_cap_str;
51
52static char *gcap_string(char *s, int c)
53{
54 if (c & CEPH_CAP_GSHARED)
55 *s++ = 's';
56 if (c & CEPH_CAP_GEXCL)
57 *s++ = 'x';
58 if (c & CEPH_CAP_GCACHE)
59 *s++ = 'c';
60 if (c & CEPH_CAP_GRD)
61 *s++ = 'r';
62 if (c & CEPH_CAP_GWR)
63 *s++ = 'w';
64 if (c & CEPH_CAP_GBUFFER)
65 *s++ = 'b';
66 if (c & CEPH_CAP_GLAZYIO)
67 *s++ = 'l';
68 return s;
69}
70
71const char *ceph_cap_string(int caps)
72{
73 int i;
74 char *s;
75 int c;
76
77 spin_lock(&cap_str_lock);
78 i = last_cap_str++;
79 if (last_cap_str == MAX_CAP_STR)
80 last_cap_str = 0;
81 spin_unlock(&cap_str_lock);
82
83 s = cap_str[i];
84
85 if (caps & CEPH_CAP_PIN)
86 *s++ = 'p';
87
88 c = (caps >> CEPH_CAP_SAUTH) & 3;
89 if (c) {
90 *s++ = 'A';
91 s = gcap_string(s, c);
92 }
93
94 c = (caps >> CEPH_CAP_SLINK) & 3;
95 if (c) {
96 *s++ = 'L';
97 s = gcap_string(s, c);
98 }
99
100 c = (caps >> CEPH_CAP_SXATTR) & 3;
101 if (c) {
102 *s++ = 'X';
103 s = gcap_string(s, c);
104 }
105
106 c = caps >> CEPH_CAP_SFILE;
107 if (c) {
108 *s++ = 'F';
109 s = gcap_string(s, c);
110 }
111
112 if (s == cap_str[i])
113 *s++ = '-';
114 *s = 0;
115 return cap_str[i];
116}
117
Yehuda Sadeh37151662010-06-17 16:16:12 -0700118void ceph_caps_init(struct ceph_mds_client *mdsc)
Sage Weila8599bd2009-10-06 11:31:12 -0700119{
Yehuda Sadeh37151662010-06-17 16:16:12 -0700120 INIT_LIST_HEAD(&mdsc->caps_list);
121 spin_lock_init(&mdsc->caps_list_lock);
Sage Weila8599bd2009-10-06 11:31:12 -0700122}
123
Yehuda Sadeh37151662010-06-17 16:16:12 -0700124void ceph_caps_finalize(struct ceph_mds_client *mdsc)
Sage Weila8599bd2009-10-06 11:31:12 -0700125{
126 struct ceph_cap *cap;
127
Yehuda Sadeh37151662010-06-17 16:16:12 -0700128 spin_lock(&mdsc->caps_list_lock);
129 while (!list_empty(&mdsc->caps_list)) {
130 cap = list_first_entry(&mdsc->caps_list,
131 struct ceph_cap, caps_item);
Sage Weila8599bd2009-10-06 11:31:12 -0700132 list_del(&cap->caps_item);
133 kmem_cache_free(ceph_cap_cachep, cap);
134 }
Yehuda Sadeh37151662010-06-17 16:16:12 -0700135 mdsc->caps_total_count = 0;
136 mdsc->caps_avail_count = 0;
137 mdsc->caps_use_count = 0;
138 mdsc->caps_reserve_count = 0;
139 mdsc->caps_min_count = 0;
140 spin_unlock(&mdsc->caps_list_lock);
Sage Weil85ccce42010-02-17 10:02:43 -0800141}
142
Yehuda Sadeh37151662010-06-17 16:16:12 -0700143void ceph_adjust_min_caps(struct ceph_mds_client *mdsc, int delta)
Sage Weil85ccce42010-02-17 10:02:43 -0800144{
Yehuda Sadeh37151662010-06-17 16:16:12 -0700145 spin_lock(&mdsc->caps_list_lock);
146 mdsc->caps_min_count += delta;
147 BUG_ON(mdsc->caps_min_count < 0);
148 spin_unlock(&mdsc->caps_list_lock);
Sage Weila8599bd2009-10-06 11:31:12 -0700149}
150
majianpeng93faca62013-06-26 11:15:27 +0800151void ceph_reserve_caps(struct ceph_mds_client *mdsc,
Yehuda Sadeh37151662010-06-17 16:16:12 -0700152 struct ceph_cap_reservation *ctx, int need)
Sage Weila8599bd2009-10-06 11:31:12 -0700153{
154 int i;
155 struct ceph_cap *cap;
156 int have;
157 int alloc = 0;
158 LIST_HEAD(newcaps);
Sage Weila8599bd2009-10-06 11:31:12 -0700159
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);
majianpeng93faca62013-06-26 11:15:27 +0800177 if (!cap)
178 break;
Sage Weila8599bd2009-10-06 11:31:12 -0700179 list_add(&cap->caps_item, &newcaps);
180 alloc++;
181 }
majianpeng93faca62013-06-26 11:15:27 +0800182 /* we didn't manage to reserve as much as we needed */
183 if (have + alloc != need)
184 pr_warn("reserve caps ctx=%p ENOMEM need=%d got=%d\n",
185 ctx, need, have + alloc);
Sage Weila8599bd2009-10-06 11:31:12 -0700186
Yehuda Sadeh37151662010-06-17 16:16:12 -0700187 spin_lock(&mdsc->caps_list_lock);
188 mdsc->caps_total_count += alloc;
189 mdsc->caps_reserve_count += alloc;
190 list_splice(&newcaps, &mdsc->caps_list);
Sage Weila8599bd2009-10-06 11:31:12 -0700191
Yehuda Sadeh37151662010-06-17 16:16:12 -0700192 BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count +
193 mdsc->caps_reserve_count +
194 mdsc->caps_avail_count);
195 spin_unlock(&mdsc->caps_list_lock);
Sage Weila8599bd2009-10-06 11:31:12 -0700196
197 ctx->count = need;
198 dout("reserve caps ctx=%p %d = %d used + %d resv + %d avail\n",
Yehuda Sadeh37151662010-06-17 16:16:12 -0700199 ctx, mdsc->caps_total_count, mdsc->caps_use_count,
200 mdsc->caps_reserve_count, mdsc->caps_avail_count);
Sage Weila8599bd2009-10-06 11:31:12 -0700201}
202
Yehuda Sadeh37151662010-06-17 16:16:12 -0700203int ceph_unreserve_caps(struct ceph_mds_client *mdsc,
204 struct ceph_cap_reservation *ctx)
Sage Weila8599bd2009-10-06 11:31:12 -0700205{
206 dout("unreserve caps ctx=%p count=%d\n", ctx, ctx->count);
207 if (ctx->count) {
Yehuda Sadeh37151662010-06-17 16:16:12 -0700208 spin_lock(&mdsc->caps_list_lock);
209 BUG_ON(mdsc->caps_reserve_count < ctx->count);
210 mdsc->caps_reserve_count -= ctx->count;
211 mdsc->caps_avail_count += ctx->count;
Sage Weila8599bd2009-10-06 11:31:12 -0700212 ctx->count = 0;
213 dout("unreserve caps %d = %d used + %d resv + %d avail\n",
Yehuda Sadeh37151662010-06-17 16:16:12 -0700214 mdsc->caps_total_count, mdsc->caps_use_count,
215 mdsc->caps_reserve_count, mdsc->caps_avail_count);
216 BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count +
217 mdsc->caps_reserve_count +
218 mdsc->caps_avail_count);
219 spin_unlock(&mdsc->caps_list_lock);
Sage Weila8599bd2009-10-06 11:31:12 -0700220 }
221 return 0;
222}
223
Yehuda Sadeh37151662010-06-17 16:16:12 -0700224static struct ceph_cap *get_cap(struct ceph_mds_client *mdsc,
225 struct ceph_cap_reservation *ctx)
Sage Weila8599bd2009-10-06 11:31:12 -0700226{
227 struct ceph_cap *cap = NULL;
228
229 /* temporary, until we do something about cap import/export */
Sage Weil443b3762010-06-29 09:28:39 -0700230 if (!ctx) {
231 cap = kmem_cache_alloc(ceph_cap_cachep, GFP_NOFS);
232 if (cap) {
Yan, Zheng4d1d0532012-11-03 10:32:37 +0800233 spin_lock(&mdsc->caps_list_lock);
Yehuda Sadeh37151662010-06-17 16:16:12 -0700234 mdsc->caps_use_count++;
235 mdsc->caps_total_count++;
Yan, Zheng4d1d0532012-11-03 10:32:37 +0800236 spin_unlock(&mdsc->caps_list_lock);
Sage Weil443b3762010-06-29 09:28:39 -0700237 }
238 return cap;
239 }
Sage Weila8599bd2009-10-06 11:31:12 -0700240
Yehuda Sadeh37151662010-06-17 16:16:12 -0700241 spin_lock(&mdsc->caps_list_lock);
Sage Weila8599bd2009-10-06 11:31:12 -0700242 dout("get_cap ctx=%p (%d) %d = %d used + %d resv + %d avail\n",
Yehuda Sadeh37151662010-06-17 16:16:12 -0700243 ctx, ctx->count, mdsc->caps_total_count, mdsc->caps_use_count,
244 mdsc->caps_reserve_count, mdsc->caps_avail_count);
Sage Weila8599bd2009-10-06 11:31:12 -0700245 BUG_ON(!ctx->count);
Yehuda Sadeh37151662010-06-17 16:16:12 -0700246 BUG_ON(ctx->count > mdsc->caps_reserve_count);
247 BUG_ON(list_empty(&mdsc->caps_list));
Sage Weila8599bd2009-10-06 11:31:12 -0700248
249 ctx->count--;
Yehuda Sadeh37151662010-06-17 16:16:12 -0700250 mdsc->caps_reserve_count--;
251 mdsc->caps_use_count++;
Sage Weila8599bd2009-10-06 11:31:12 -0700252
Yehuda Sadeh37151662010-06-17 16:16:12 -0700253 cap = list_first_entry(&mdsc->caps_list, struct ceph_cap, caps_item);
Sage Weila8599bd2009-10-06 11:31:12 -0700254 list_del(&cap->caps_item);
255
Yehuda Sadeh37151662010-06-17 16:16:12 -0700256 BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count +
257 mdsc->caps_reserve_count + mdsc->caps_avail_count);
258 spin_unlock(&mdsc->caps_list_lock);
Sage Weila8599bd2009-10-06 11:31:12 -0700259 return cap;
260}
261
Yehuda Sadeh37151662010-06-17 16:16:12 -0700262void ceph_put_cap(struct ceph_mds_client *mdsc, struct ceph_cap *cap)
Sage Weila8599bd2009-10-06 11:31:12 -0700263{
Yehuda Sadeh37151662010-06-17 16:16:12 -0700264 spin_lock(&mdsc->caps_list_lock);
Sage Weil7c1332b2010-02-16 11:39:45 -0800265 dout("put_cap %p %d = %d used + %d resv + %d avail\n",
Yehuda Sadeh37151662010-06-17 16:16:12 -0700266 cap, mdsc->caps_total_count, mdsc->caps_use_count,
267 mdsc->caps_reserve_count, mdsc->caps_avail_count);
268 mdsc->caps_use_count--;
Sage Weila8599bd2009-10-06 11:31:12 -0700269 /*
Sage Weil85ccce42010-02-17 10:02:43 -0800270 * Keep some preallocated caps around (ceph_min_count), to
271 * avoid lots of free/alloc churn.
Sage Weila8599bd2009-10-06 11:31:12 -0700272 */
Yehuda Sadeh37151662010-06-17 16:16:12 -0700273 if (mdsc->caps_avail_count >= mdsc->caps_reserve_count +
274 mdsc->caps_min_count) {
275 mdsc->caps_total_count--;
Sage Weila8599bd2009-10-06 11:31:12 -0700276 kmem_cache_free(ceph_cap_cachep, cap);
277 } else {
Yehuda Sadeh37151662010-06-17 16:16:12 -0700278 mdsc->caps_avail_count++;
279 list_add(&cap->caps_item, &mdsc->caps_list);
Sage Weila8599bd2009-10-06 11:31:12 -0700280 }
281
Yehuda Sadeh37151662010-06-17 16:16:12 -0700282 BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count +
283 mdsc->caps_reserve_count + mdsc->caps_avail_count);
284 spin_unlock(&mdsc->caps_list_lock);
Sage Weila8599bd2009-10-06 11:31:12 -0700285}
286
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700287void ceph_reservation_status(struct ceph_fs_client *fsc,
Sage Weil85ccce42010-02-17 10:02:43 -0800288 int *total, int *avail, int *used, int *reserved,
289 int *min)
Sage Weila8599bd2009-10-06 11:31:12 -0700290{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700291 struct ceph_mds_client *mdsc = fsc->mdsc;
Yehuda Sadeh37151662010-06-17 16:16:12 -0700292
Sage Weila8599bd2009-10-06 11:31:12 -0700293 if (total)
Yehuda Sadeh37151662010-06-17 16:16:12 -0700294 *total = mdsc->caps_total_count;
Sage Weila8599bd2009-10-06 11:31:12 -0700295 if (avail)
Yehuda Sadeh37151662010-06-17 16:16:12 -0700296 *avail = mdsc->caps_avail_count;
Sage Weila8599bd2009-10-06 11:31:12 -0700297 if (used)
Yehuda Sadeh37151662010-06-17 16:16:12 -0700298 *used = mdsc->caps_use_count;
Sage Weila8599bd2009-10-06 11:31:12 -0700299 if (reserved)
Yehuda Sadeh37151662010-06-17 16:16:12 -0700300 *reserved = mdsc->caps_reserve_count;
Sage Weil85ccce42010-02-17 10:02:43 -0800301 if (min)
Yehuda Sadeh37151662010-06-17 16:16:12 -0700302 *min = mdsc->caps_min_count;
Sage Weila8599bd2009-10-06 11:31:12 -0700303}
304
305/*
306 * Find ceph_cap for given mds, if any.
307 *
Sage Weilbe655592011-11-30 09:47:09 -0800308 * Called with i_ceph_lock held.
Sage Weila8599bd2009-10-06 11:31:12 -0700309 */
310static struct ceph_cap *__get_cap_for_mds(struct ceph_inode_info *ci, int mds)
311{
312 struct ceph_cap *cap;
313 struct rb_node *n = ci->i_caps.rb_node;
314
315 while (n) {
316 cap = rb_entry(n, struct ceph_cap, ci_node);
317 if (mds < cap->mds)
318 n = n->rb_left;
319 else if (mds > cap->mds)
320 n = n->rb_right;
321 else
322 return cap;
323 }
324 return NULL;
325}
326
Greg Farnum2bc50252010-06-30 12:44:34 -0700327struct ceph_cap *ceph_get_cap_for_mds(struct ceph_inode_info *ci, int mds)
328{
329 struct ceph_cap *cap;
330
Sage Weilbe655592011-11-30 09:47:09 -0800331 spin_lock(&ci->i_ceph_lock);
Greg Farnum2bc50252010-06-30 12:44:34 -0700332 cap = __get_cap_for_mds(ci, mds);
Sage Weilbe655592011-11-30 09:47:09 -0800333 spin_unlock(&ci->i_ceph_lock);
Greg Farnum2bc50252010-06-30 12:44:34 -0700334 return cap;
335}
336
Sage Weila8599bd2009-10-06 11:31:12 -0700337/*
Sage Weil33caad32010-05-26 14:31:27 -0700338 * Return id of any MDS with a cap, preferably FILE_WR|BUFFER|EXCL, else -1.
Sage Weila8599bd2009-10-06 11:31:12 -0700339 */
Sage Weilca81f3f2010-06-10 14:21:36 -0700340static int __ceph_get_cap_mds(struct ceph_inode_info *ci)
Sage Weila8599bd2009-10-06 11:31:12 -0700341{
342 struct ceph_cap *cap;
343 int mds = -1;
344 struct rb_node *p;
345
Sage Weil33caad32010-05-26 14:31:27 -0700346 /* prefer mds with WR|BUFFER|EXCL caps */
Sage Weila8599bd2009-10-06 11:31:12 -0700347 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
348 cap = rb_entry(p, struct ceph_cap, ci_node);
349 mds = cap->mds;
Sage Weila8599bd2009-10-06 11:31:12 -0700350 if (cap->issued & (CEPH_CAP_FILE_WR |
351 CEPH_CAP_FILE_BUFFER |
352 CEPH_CAP_FILE_EXCL))
353 break;
354 }
355 return mds;
356}
357
358int ceph_get_cap_mds(struct inode *inode)
359{
Sage Weilbe655592011-11-30 09:47:09 -0800360 struct ceph_inode_info *ci = ceph_inode(inode);
Sage Weila8599bd2009-10-06 11:31:12 -0700361 int mds;
Sage Weilbe655592011-11-30 09:47:09 -0800362 spin_lock(&ci->i_ceph_lock);
Sage Weilca81f3f2010-06-10 14:21:36 -0700363 mds = __ceph_get_cap_mds(ceph_inode(inode));
Sage Weilbe655592011-11-30 09:47:09 -0800364 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -0700365 return mds;
366}
367
368/*
Sage Weilbe655592011-11-30 09:47:09 -0800369 * Called under i_ceph_lock.
Sage Weila8599bd2009-10-06 11:31:12 -0700370 */
371static void __insert_cap_node(struct ceph_inode_info *ci,
372 struct ceph_cap *new)
373{
374 struct rb_node **p = &ci->i_caps.rb_node;
375 struct rb_node *parent = NULL;
376 struct ceph_cap *cap = NULL;
377
378 while (*p) {
379 parent = *p;
380 cap = rb_entry(parent, struct ceph_cap, ci_node);
381 if (new->mds < cap->mds)
382 p = &(*p)->rb_left;
383 else if (new->mds > cap->mds)
384 p = &(*p)->rb_right;
385 else
386 BUG();
387 }
388
389 rb_link_node(&new->ci_node, parent, p);
390 rb_insert_color(&new->ci_node, &ci->i_caps);
391}
392
393/*
394 * (re)set cap hold timeouts, which control the delayed release
395 * of unused caps back to the MDS. Should be called on cap use.
396 */
397static void __cap_set_timeouts(struct ceph_mds_client *mdsc,
398 struct ceph_inode_info *ci)
399{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700400 struct ceph_mount_options *ma = mdsc->fsc->mount_options;
Sage Weila8599bd2009-10-06 11:31:12 -0700401
402 ci->i_hold_caps_min = round_jiffies(jiffies +
403 ma->caps_wanted_delay_min * HZ);
404 ci->i_hold_caps_max = round_jiffies(jiffies +
405 ma->caps_wanted_delay_max * HZ);
406 dout("__cap_set_timeouts %p min %lu max %lu\n", &ci->vfs_inode,
407 ci->i_hold_caps_min - jiffies, ci->i_hold_caps_max - jiffies);
408}
409
410/*
411 * (Re)queue cap at the end of the delayed cap release list.
412 *
413 * If I_FLUSH is set, leave the inode at the front of the list.
414 *
Sage Weilbe655592011-11-30 09:47:09 -0800415 * Caller holds i_ceph_lock
Sage Weila8599bd2009-10-06 11:31:12 -0700416 * -> we take mdsc->cap_delay_lock
417 */
418static void __cap_delay_requeue(struct ceph_mds_client *mdsc,
419 struct ceph_inode_info *ci)
420{
421 __cap_set_timeouts(mdsc, ci);
422 dout("__cap_delay_requeue %p flags %d at %lu\n", &ci->vfs_inode,
423 ci->i_ceph_flags, ci->i_hold_caps_max);
424 if (!mdsc->stopping) {
425 spin_lock(&mdsc->cap_delay_lock);
426 if (!list_empty(&ci->i_cap_delay_list)) {
427 if (ci->i_ceph_flags & CEPH_I_FLUSH)
428 goto no_change;
429 list_del_init(&ci->i_cap_delay_list);
430 }
431 list_add_tail(&ci->i_cap_delay_list, &mdsc->cap_delay_list);
432no_change:
433 spin_unlock(&mdsc->cap_delay_lock);
434 }
435}
436
437/*
438 * Queue an inode for immediate writeback. Mark inode with I_FLUSH,
439 * indicating we should send a cap message to flush dirty metadata
440 * asap, and move to the front of the delayed cap list.
441 */
442static void __cap_delay_requeue_front(struct ceph_mds_client *mdsc,
443 struct ceph_inode_info *ci)
444{
445 dout("__cap_delay_requeue_front %p\n", &ci->vfs_inode);
446 spin_lock(&mdsc->cap_delay_lock);
447 ci->i_ceph_flags |= CEPH_I_FLUSH;
448 if (!list_empty(&ci->i_cap_delay_list))
449 list_del_init(&ci->i_cap_delay_list);
450 list_add(&ci->i_cap_delay_list, &mdsc->cap_delay_list);
451 spin_unlock(&mdsc->cap_delay_lock);
452}
453
454/*
455 * Cancel delayed work on cap.
456 *
Sage Weilbe655592011-11-30 09:47:09 -0800457 * Caller must hold i_ceph_lock.
Sage Weila8599bd2009-10-06 11:31:12 -0700458 */
459static void __cap_delay_cancel(struct ceph_mds_client *mdsc,
460 struct ceph_inode_info *ci)
461{
462 dout("__cap_delay_cancel %p\n", &ci->vfs_inode);
463 if (list_empty(&ci->i_cap_delay_list))
464 return;
465 spin_lock(&mdsc->cap_delay_lock);
466 list_del_init(&ci->i_cap_delay_list);
467 spin_unlock(&mdsc->cap_delay_lock);
468}
469
470/*
471 * Common issue checks for add_cap, handle_cap_grant.
472 */
473static void __check_cap_issue(struct ceph_inode_info *ci, struct ceph_cap *cap,
474 unsigned issued)
475{
476 unsigned had = __ceph_caps_issued(ci, NULL);
477
478 /*
479 * Each time we receive FILE_CACHE anew, we increment
480 * i_rdcache_gen.
481 */
Sage Weil29625072010-05-27 10:40:43 -0700482 if ((issued & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) &&
Milosz Tanski99ccbd22013-08-21 17:29:54 -0400483 (had & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) == 0) {
Sage Weila8599bd2009-10-06 11:31:12 -0700484 ci->i_rdcache_gen++;
Milosz Tanski99ccbd22013-08-21 17:29:54 -0400485 }
Sage Weila8599bd2009-10-06 11:31:12 -0700486
487 /*
Yan, Zheng2f276c52013-03-13 19:44:32 +0800488 * if we are newly issued FILE_SHARED, mark dir not complete; we
Sage Weila8599bd2009-10-06 11:31:12 -0700489 * don't know what happened to this directory while we didn't
490 * have the cap.
491 */
492 if ((issued & CEPH_CAP_FILE_SHARED) &&
493 (had & CEPH_CAP_FILE_SHARED) == 0) {
494 ci->i_shared_gen++;
Yan, Zhenga8673d62013-02-18 16:38:14 +0800495 if (S_ISDIR(ci->vfs_inode.i_mode)) {
496 dout(" marking %p NOT complete\n", &ci->vfs_inode);
Yan, Zheng2f276c52013-03-13 19:44:32 +0800497 __ceph_dir_clear_complete(ci);
Yan, Zhenga8673d62013-02-18 16:38:14 +0800498 }
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:
Sage Weilbe655592011-11-30 09:47:09 -0800535 spin_lock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -0700536 cap = __get_cap_for_mds(ci, mds);
537 if (!cap) {
538 if (new_cap) {
539 cap = new_cap;
540 new_cap = NULL;
541 } else {
Sage Weilbe655592011-11-30 09:47:09 -0800542 spin_unlock(&ci->i_ceph_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;
Yan, Zheng964266c2013-02-27 09:26:09 +0800553 cap->mseq = 0;
Sage Weila8599bd2009-10-06 11:31:12 -0700554
555 cap->ci = ci;
556 __insert_cap_node(ci, cap);
557
558 /* clear out old exporting info? (i.e. on cap import) */
559 if (ci->i_cap_exporting_mds == mds) {
560 ci->i_cap_exporting_issued = 0;
561 ci->i_cap_exporting_mseq = 0;
562 ci->i_cap_exporting_mds = -1;
563 }
564
565 /* add to session cap list */
566 cap->session = session;
567 spin_lock(&session->s_cap_lock);
568 list_add_tail(&cap->session_caps, &session->s_caps);
569 session->s_nr_caps++;
570 spin_unlock(&session->s_cap_lock);
Sage Weil35403032011-05-12 15:13:23 -0700571 } else if (new_cap)
572 ceph_put_cap(mdsc, new_cap);
Sage Weila8599bd2009-10-06 11:31:12 -0700573
574 if (!ci->i_snap_realm) {
575 /*
576 * add this inode to the appropriate snap realm
577 */
578 struct ceph_snap_realm *realm = ceph_lookup_snap_realm(mdsc,
579 realmino);
580 if (realm) {
581 ceph_get_snap_realm(mdsc, realm);
582 spin_lock(&realm->inodes_with_caps_lock);
583 ci->i_snap_realm = realm;
584 list_add(&ci->i_snap_realm_item,
585 &realm->inodes_with_caps);
586 spin_unlock(&realm->inodes_with_caps_lock);
587 } else {
588 pr_err("ceph_add_cap: couldn't find snap realm %llx\n",
589 realmino);
Sage Weilb8cd07e2010-07-16 12:00:02 -0700590 WARN_ON(!realm);
Sage Weila8599bd2009-10-06 11:31:12 -0700591 }
592 }
593
594 __check_cap_issue(ci, cap, issued);
595
596 /*
597 * If we are issued caps we don't want, or the mds' wanted
598 * value appears to be off, queue a check so we'll release
599 * later and/or update the mds wanted value.
600 */
601 actual_wanted = __ceph_caps_wanted(ci);
602 if ((wanted & ~actual_wanted) ||
603 (issued & ~actual_wanted & CEPH_CAP_ANY_WR)) {
604 dout(" issued %s, mds wanted %s, actual %s, queueing\n",
605 ceph_cap_string(issued), ceph_cap_string(wanted),
606 ceph_cap_string(actual_wanted));
607 __cap_delay_requeue(mdsc, ci);
608 }
609
Yan, Zhengb8c2f3a2013-05-31 16:37:11 +0800610 if (flags & CEPH_CAP_FLAG_AUTH) {
611 if (ci->i_auth_cap == NULL ||
612 ceph_seq_cmp(ci->i_auth_cap->mseq, mseq) < 0)
613 ci->i_auth_cap = cap;
614 } else if (ci->i_auth_cap == cap) {
Sage Weila8599bd2009-10-06 11:31:12 -0700615 ci->i_auth_cap = NULL;
Yan, Zheng8a92a112013-01-04 14:28:07 +0800616 spin_lock(&mdsc->cap_dirty_lock);
617 if (!list_empty(&ci->i_dirty_item)) {
618 dout(" moving %p to cap_dirty_migrating\n", inode);
619 list_move(&ci->i_dirty_item,
620 &mdsc->cap_dirty_migrating);
621 }
622 spin_unlock(&mdsc->cap_dirty_lock);
623 }
Sage Weila8599bd2009-10-06 11:31:12 -0700624
625 dout("add_cap inode %p (%llx.%llx) cap %p %s now %s seq %d mds%d\n",
626 inode, ceph_vinop(inode), cap, ceph_cap_string(issued),
627 ceph_cap_string(issued|cap->issued), seq, mds);
628 cap->cap_id = cap_id;
629 cap->issued = issued;
630 cap->implemented |= issued;
Yan, Zheng964266c2013-02-27 09:26:09 +0800631 if (mseq > cap->mseq)
632 cap->mds_wanted = wanted;
633 else
634 cap->mds_wanted |= wanted;
Sage Weila8599bd2009-10-06 11:31:12 -0700635 cap->seq = seq;
636 cap->issue_seq = seq;
637 cap->mseq = mseq;
Sage Weil685f9a5d2009-11-09 12:05:48 -0800638 cap->cap_gen = session->s_cap_gen;
Sage Weila8599bd2009-10-06 11:31:12 -0700639
640 if (fmode >= 0)
641 __ceph_get_fmode(ci, fmode);
Sage Weilbe655592011-11-30 09:47:09 -0800642 spin_unlock(&ci->i_ceph_lock);
Yehuda Sadeh03066f22010-07-27 13:11:08 -0700643 wake_up_all(&ci->i_cap_wq);
Sage Weila8599bd2009-10-06 11:31:12 -0700644 return 0;
645}
646
647/*
648 * Return true if cap has not timed out and belongs to the current
649 * generation of the MDS session (i.e. has not gone 'stale' due to
650 * us losing touch with the mds).
651 */
652static int __cap_is_valid(struct ceph_cap *cap)
653{
654 unsigned long ttl;
Sage Weilcdac8302009-11-10 16:02:23 -0800655 u32 gen;
Sage Weila8599bd2009-10-06 11:31:12 -0700656
Alex Elderd8fb02a2012-01-12 17:48:10 -0800657 spin_lock(&cap->session->s_gen_ttl_lock);
Sage Weila8599bd2009-10-06 11:31:12 -0700658 gen = cap->session->s_cap_gen;
659 ttl = cap->session->s_cap_ttl;
Alex Elderd8fb02a2012-01-12 17:48:10 -0800660 spin_unlock(&cap->session->s_gen_ttl_lock);
Sage Weila8599bd2009-10-06 11:31:12 -0700661
Sage Weil685f9a5d2009-11-09 12:05:48 -0800662 if (cap->cap_gen < gen || time_after_eq(jiffies, ttl)) {
Sage Weila8599bd2009-10-06 11:31:12 -0700663 dout("__cap_is_valid %p cap %p issued %s "
664 "but STALE (gen %u vs %u)\n", &cap->ci->vfs_inode,
Sage Weil685f9a5d2009-11-09 12:05:48 -0800665 cap, ceph_cap_string(cap->issued), cap->cap_gen, gen);
Sage Weila8599bd2009-10-06 11:31:12 -0700666 return 0;
667 }
668
669 return 1;
670}
671
672/*
673 * Return set of valid cap bits issued to us. Note that caps time
674 * out, and may be invalidated in bulk if the client session times out
675 * and session->s_cap_gen is bumped.
676 */
677int __ceph_caps_issued(struct ceph_inode_info *ci, int *implemented)
678{
Sage Weil7af8f1e2010-03-01 15:17:34 -0800679 int have = ci->i_snap_caps | ci->i_cap_exporting_issued;
Sage Weila8599bd2009-10-06 11:31:12 -0700680 struct ceph_cap *cap;
681 struct rb_node *p;
682
683 if (implemented)
684 *implemented = 0;
685 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
686 cap = rb_entry(p, struct ceph_cap, ci_node);
687 if (!__cap_is_valid(cap))
688 continue;
689 dout("__ceph_caps_issued %p cap %p issued %s\n",
690 &ci->vfs_inode, cap, ceph_cap_string(cap->issued));
691 have |= cap->issued;
692 if (implemented)
693 *implemented |= cap->implemented;
694 }
Yan, Zhengb1530f52013-07-02 12:40:20 +0800695 /*
696 * exclude caps issued by non-auth MDS, but are been revoking
697 * by the auth MDS. The non-auth MDS should be revoking/exporting
698 * these caps, but the message is delayed.
699 */
700 if (ci->i_auth_cap) {
701 cap = ci->i_auth_cap;
702 have &= ~cap->implemented | cap->issued;
703 }
Sage Weila8599bd2009-10-06 11:31:12 -0700704 return have;
705}
706
707/*
708 * Get cap bits issued by caps other than @ocap
709 */
710int __ceph_caps_issued_other(struct ceph_inode_info *ci, struct ceph_cap *ocap)
711{
712 int have = ci->i_snap_caps;
713 struct ceph_cap *cap;
714 struct rb_node *p;
715
716 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
717 cap = rb_entry(p, struct ceph_cap, ci_node);
718 if (cap == ocap)
719 continue;
720 if (!__cap_is_valid(cap))
721 continue;
722 have |= cap->issued;
723 }
724 return have;
725}
726
727/*
728 * Move a cap to the end of the LRU (oldest caps at list head, newest
729 * at list tail).
730 */
731static void __touch_cap(struct ceph_cap *cap)
732{
733 struct ceph_mds_session *s = cap->session;
734
Sage Weila8599bd2009-10-06 11:31:12 -0700735 spin_lock(&s->s_cap_lock);
Sage Weil7c1332b2010-02-16 11:39:45 -0800736 if (s->s_cap_iterator == NULL) {
Sage Weil5dacf092009-12-21 20:40:34 -0800737 dout("__touch_cap %p cap %p mds%d\n", &cap->ci->vfs_inode, cap,
738 s->s_mds);
739 list_move_tail(&cap->session_caps, &s->s_caps);
740 } else {
741 dout("__touch_cap %p cap %p mds%d NOP, iterating over caps\n",
742 &cap->ci->vfs_inode, cap, s->s_mds);
743 }
Sage Weila8599bd2009-10-06 11:31:12 -0700744 spin_unlock(&s->s_cap_lock);
745}
746
747/*
748 * Check if we hold the given mask. If so, move the cap(s) to the
749 * front of their respective LRUs. (This is the preferred way for
750 * callers to check for caps they want.)
751 */
752int __ceph_caps_issued_mask(struct ceph_inode_info *ci, int mask, int touch)
753{
754 struct ceph_cap *cap;
755 struct rb_node *p;
756 int have = ci->i_snap_caps;
757
758 if ((have & mask) == mask) {
759 dout("__ceph_caps_issued_mask %p snap issued %s"
760 " (mask %s)\n", &ci->vfs_inode,
761 ceph_cap_string(have),
762 ceph_cap_string(mask));
763 return 1;
764 }
765
766 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
767 cap = rb_entry(p, struct ceph_cap, ci_node);
768 if (!__cap_is_valid(cap))
769 continue;
770 if ((cap->issued & mask) == mask) {
771 dout("__ceph_caps_issued_mask %p cap %p issued %s"
772 " (mask %s)\n", &ci->vfs_inode, cap,
773 ceph_cap_string(cap->issued),
774 ceph_cap_string(mask));
775 if (touch)
776 __touch_cap(cap);
777 return 1;
778 }
779
780 /* does a combination of caps satisfy mask? */
781 have |= cap->issued;
782 if ((have & mask) == mask) {
783 dout("__ceph_caps_issued_mask %p combo issued %s"
784 " (mask %s)\n", &ci->vfs_inode,
785 ceph_cap_string(cap->issued),
786 ceph_cap_string(mask));
787 if (touch) {
788 struct rb_node *q;
789
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300790 /* touch this + preceding caps */
Sage Weila8599bd2009-10-06 11:31:12 -0700791 __touch_cap(cap);
792 for (q = rb_first(&ci->i_caps); q != p;
793 q = rb_next(q)) {
794 cap = rb_entry(q, struct ceph_cap,
795 ci_node);
796 if (!__cap_is_valid(cap))
797 continue;
798 __touch_cap(cap);
799 }
800 }
801 return 1;
802 }
803 }
804
805 return 0;
806}
807
808/*
809 * Return true if mask caps are currently being revoked by an MDS.
810 */
Yan, Zheng6ee6b9532013-07-02 12:40:21 +0800811int __ceph_caps_revoking_other(struct ceph_inode_info *ci,
812 struct ceph_cap *ocap, int mask)
813{
814 struct ceph_cap *cap;
815 struct rb_node *p;
816
817 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
818 cap = rb_entry(p, struct ceph_cap, ci_node);
819 if (cap != ocap && __cap_is_valid(cap) &&
820 (cap->implemented & ~cap->issued & mask))
821 return 1;
822 }
823 return 0;
824}
825
Sage Weila8599bd2009-10-06 11:31:12 -0700826int ceph_caps_revoking(struct ceph_inode_info *ci, int mask)
827{
828 struct inode *inode = &ci->vfs_inode;
Yan, Zheng6ee6b9532013-07-02 12:40:21 +0800829 int ret;
Sage Weila8599bd2009-10-06 11:31:12 -0700830
Sage Weilbe655592011-11-30 09:47:09 -0800831 spin_lock(&ci->i_ceph_lock);
Yan, Zheng6ee6b9532013-07-02 12:40:21 +0800832 ret = __ceph_caps_revoking_other(ci, NULL, mask);
Sage Weilbe655592011-11-30 09:47:09 -0800833 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -0700834 dout("ceph_caps_revoking %p %s = %d\n", inode,
835 ceph_cap_string(mask), ret);
836 return ret;
837}
838
839int __ceph_caps_used(struct ceph_inode_info *ci)
840{
841 int used = 0;
842 if (ci->i_pin_ref)
843 used |= CEPH_CAP_PIN;
844 if (ci->i_rd_ref)
845 used |= CEPH_CAP_FILE_RD;
Sage Weila43fb732010-09-17 09:54:08 -0700846 if (ci->i_rdcache_ref || ci->vfs_inode.i_data.nrpages)
Sage Weila8599bd2009-10-06 11:31:12 -0700847 used |= CEPH_CAP_FILE_CACHE;
848 if (ci->i_wr_ref)
849 used |= CEPH_CAP_FILE_WR;
Henry C Changd3d07202011-05-11 10:29:54 +0000850 if (ci->i_wb_ref || ci->i_wrbuffer_ref)
Sage Weila8599bd2009-10-06 11:31:12 -0700851 used |= CEPH_CAP_FILE_BUFFER;
852 return used;
853}
854
855/*
856 * wanted, by virtue of open file modes
857 */
858int __ceph_caps_file_wanted(struct ceph_inode_info *ci)
859{
860 int want = 0;
861 int mode;
Sage Weil33caad32010-05-26 14:31:27 -0700862 for (mode = 0; mode < CEPH_FILE_MODE_NUM; mode++)
Sage Weila8599bd2009-10-06 11:31:12 -0700863 if (ci->i_nr_by_mode[mode])
864 want |= ceph_caps_for_mode(mode);
865 return want;
866}
867
868/*
869 * Return caps we have registered with the MDS(s) as 'wanted'.
870 */
871int __ceph_caps_mds_wanted(struct ceph_inode_info *ci)
872{
873 struct ceph_cap *cap;
874 struct rb_node *p;
875 int mds_wanted = 0;
876
877 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
878 cap = rb_entry(p, struct ceph_cap, ci_node);
879 if (!__cap_is_valid(cap))
880 continue;
881 mds_wanted |= cap->mds_wanted;
882 }
883 return mds_wanted;
884}
885
886/*
Sage Weilbe655592011-11-30 09:47:09 -0800887 * called under i_ceph_lock
Sage Weila8599bd2009-10-06 11:31:12 -0700888 */
889static int __ceph_is_any_caps(struct ceph_inode_info *ci)
890{
891 return !RB_EMPTY_ROOT(&ci->i_caps) || ci->i_cap_exporting_mds >= 0;
892}
893
894/*
Sage Weilf818a732010-05-11 20:56:31 -0700895 * Remove a cap. Take steps to deal with a racing iterate_session_caps.
896 *
Sage Weilbe655592011-11-30 09:47:09 -0800897 * caller should hold i_ceph_lock.
Sage Weila6369742010-02-22 13:59:00 -0800898 * caller will not hold session s_mutex if called from destroy_inode.
Sage Weila8599bd2009-10-06 11:31:12 -0700899 */
Sage Weil7c1332b2010-02-16 11:39:45 -0800900void __ceph_remove_cap(struct ceph_cap *cap)
Sage Weila8599bd2009-10-06 11:31:12 -0700901{
902 struct ceph_mds_session *session = cap->session;
903 struct ceph_inode_info *ci = cap->ci;
Cheng Renquan640ef792010-03-26 17:40:33 +0800904 struct ceph_mds_client *mdsc =
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700905 ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc;
Sage Weilf818a732010-05-11 20:56:31 -0700906 int removed = 0;
Sage Weila8599bd2009-10-06 11:31:12 -0700907
908 dout("__ceph_remove_cap %p from %p\n", cap, &ci->vfs_inode);
909
Sage Weil7c1332b2010-02-16 11:39:45 -0800910 /* remove from session list */
911 spin_lock(&session->s_cap_lock);
912 if (session->s_cap_iterator == cap) {
913 /* not yet, we are iterating over this very cap */
914 dout("__ceph_remove_cap delaying %p removal from session %p\n",
915 cap, cap->session);
916 } else {
917 list_del_init(&cap->session_caps);
918 session->s_nr_caps--;
919 cap->session = NULL;
Sage Weilf818a732010-05-11 20:56:31 -0700920 removed = 1;
Sage Weil7c1332b2010-02-16 11:39:45 -0800921 }
Sage Weilf818a732010-05-11 20:56:31 -0700922 /* protect backpointer with s_cap_lock: see iterate_session_caps */
923 cap->ci = NULL;
Sage Weil7c1332b2010-02-16 11:39:45 -0800924 spin_unlock(&session->s_cap_lock);
925
Sage Weilf818a732010-05-11 20:56:31 -0700926 /* remove from inode list */
927 rb_erase(&cap->ci_node, &ci->i_caps);
928 if (ci->i_auth_cap == cap)
929 ci->i_auth_cap = NULL;
930
931 if (removed)
Yehuda Sadeh37151662010-06-17 16:16:12 -0700932 ceph_put_cap(mdsc, cap);
Sage Weila8599bd2009-10-06 11:31:12 -0700933
934 if (!__ceph_is_any_caps(ci) && ci->i_snap_realm) {
935 struct ceph_snap_realm *realm = ci->i_snap_realm;
936 spin_lock(&realm->inodes_with_caps_lock);
937 list_del_init(&ci->i_snap_realm_item);
938 ci->i_snap_realm_counter++;
939 ci->i_snap_realm = NULL;
940 spin_unlock(&realm->inodes_with_caps_lock);
941 ceph_put_snap_realm(mdsc, realm);
942 }
943 if (!__ceph_is_any_real_caps(ci))
944 __cap_delay_cancel(mdsc, ci);
945}
946
947/*
948 * Build and send a cap message to the given MDS.
949 *
950 * Caller should be holding s_mutex.
951 */
952static int send_cap_msg(struct ceph_mds_session *session,
953 u64 ino, u64 cid, int op,
954 int caps, int wanted, int dirty,
955 u32 seq, u64 flush_tid, u32 issue_seq, u32 mseq,
956 u64 size, u64 max_size,
957 struct timespec *mtime, struct timespec *atime,
958 u64 time_warp_seq,
Eric W. Biederman05cb11c2013-01-31 02:56:19 -0800959 kuid_t uid, kgid_t gid, umode_t mode,
Sage Weila8599bd2009-10-06 11:31:12 -0700960 u64 xattr_version,
961 struct ceph_buffer *xattrs_buf,
962 u64 follows)
963{
964 struct ceph_mds_caps *fc;
965 struct ceph_msg *msg;
966
967 dout("send_cap_msg %s %llx %llx caps %s wanted %s dirty %s"
968 " seq %u/%u mseq %u follows %lld size %llu/%llu"
969 " xattr_ver %llu xattr_len %d\n", ceph_cap_op_name(op),
970 cid, ino, ceph_cap_string(caps), ceph_cap_string(wanted),
971 ceph_cap_string(dirty),
972 seq, issue_seq, mseq, follows, size, max_size,
973 xattr_version, xattrs_buf ? (int)xattrs_buf->vec.iov_len : 0);
974
Sage Weilb61c2762011-08-09 15:03:46 -0700975 msg = ceph_msg_new(CEPH_MSG_CLIENT_CAPS, sizeof(*fc), GFP_NOFS, false);
Sage Weila79832f2010-04-01 16:06:19 -0700976 if (!msg)
977 return -ENOMEM;
Sage Weila8599bd2009-10-06 11:31:12 -0700978
Sage Weil6df058c2009-12-22 11:24:33 -0800979 msg->hdr.tid = cpu_to_le64(flush_tid);
Sage Weila8599bd2009-10-06 11:31:12 -0700980
Sage Weil6df058c2009-12-22 11:24:33 -0800981 fc = msg->front.iov_base;
Sage Weila8599bd2009-10-06 11:31:12 -0700982 memset(fc, 0, sizeof(*fc));
983
984 fc->cap_id = cpu_to_le64(cid);
985 fc->op = cpu_to_le32(op);
986 fc->seq = cpu_to_le32(seq);
Sage Weila8599bd2009-10-06 11:31:12 -0700987 fc->issue_seq = cpu_to_le32(issue_seq);
988 fc->migrate_seq = cpu_to_le32(mseq);
989 fc->caps = cpu_to_le32(caps);
990 fc->wanted = cpu_to_le32(wanted);
991 fc->dirty = cpu_to_le32(dirty);
992 fc->ino = cpu_to_le64(ino);
993 fc->snap_follows = cpu_to_le64(follows);
994
995 fc->size = cpu_to_le64(size);
996 fc->max_size = cpu_to_le64(max_size);
997 if (mtime)
998 ceph_encode_timespec(&fc->mtime, mtime);
999 if (atime)
1000 ceph_encode_timespec(&fc->atime, atime);
1001 fc->time_warp_seq = cpu_to_le32(time_warp_seq);
1002
Eric W. Biederman05cb11c2013-01-31 02:56:19 -08001003 fc->uid = cpu_to_le32(from_kuid(&init_user_ns, uid));
1004 fc->gid = cpu_to_le32(from_kgid(&init_user_ns, gid));
Sage Weila8599bd2009-10-06 11:31:12 -07001005 fc->mode = cpu_to_le32(mode);
1006
1007 fc->xattr_version = cpu_to_le64(xattr_version);
1008 if (xattrs_buf) {
1009 msg->middle = ceph_buffer_get(xattrs_buf);
1010 fc->xattr_len = cpu_to_le32(xattrs_buf->vec.iov_len);
1011 msg->hdr.middle_len = cpu_to_le32(xattrs_buf->vec.iov_len);
1012 }
1013
1014 ceph_con_send(&session->s_con, msg);
1015 return 0;
1016}
1017
Yan, Zhengd40ee0d2013-02-18 13:43:43 +08001018void __queue_cap_release(struct ceph_mds_session *session,
1019 u64 ino, u64 cap_id, u32 migrate_seq,
1020 u32 issue_seq)
Sage Weil3d7ded42010-06-09 16:47:10 -07001021{
1022 struct ceph_msg *msg;
1023 struct ceph_mds_cap_release *head;
1024 struct ceph_mds_cap_item *item;
1025
1026 spin_lock(&session->s_cap_lock);
1027 BUG_ON(!session->s_num_cap_releases);
1028 msg = list_first_entry(&session->s_cap_releases,
1029 struct ceph_msg, list_head);
1030
1031 dout(" adding %llx release to mds%d msg %p (%d left)\n",
1032 ino, session->s_mds, msg, session->s_num_cap_releases);
1033
1034 BUG_ON(msg->front.iov_len + sizeof(*item) > PAGE_CACHE_SIZE);
1035 head = msg->front.iov_base;
Wei Yongjunb905a7f2012-09-28 12:59:16 +08001036 le32_add_cpu(&head->num, 1);
Sage Weil3d7ded42010-06-09 16:47:10 -07001037 item = msg->front.iov_base + msg->front.iov_len;
1038 item->ino = cpu_to_le64(ino);
1039 item->cap_id = cpu_to_le64(cap_id);
1040 item->migrate_seq = cpu_to_le32(migrate_seq);
1041 item->seq = cpu_to_le32(issue_seq);
1042
1043 session->s_num_cap_releases--;
1044
1045 msg->front.iov_len += sizeof(*item);
1046 if (le32_to_cpu(head->num) == CEPH_CAPS_PER_RELEASE) {
1047 dout(" release msg %p full\n", msg);
1048 list_move_tail(&msg->list_head, &session->s_cap_releases_done);
1049 } else {
1050 dout(" release msg %p at %d/%d (%d)\n", msg,
1051 (int)le32_to_cpu(head->num),
1052 (int)CEPH_CAPS_PER_RELEASE,
1053 (int)msg->front.iov_len);
1054 }
1055 spin_unlock(&session->s_cap_lock);
1056}
1057
Sage Weila8599bd2009-10-06 11:31:12 -07001058/*
Sage Weila6369742010-02-22 13:59:00 -08001059 * Queue cap releases when an inode is dropped from our cache. Since
Sage Weilbe655592011-11-30 09:47:09 -08001060 * inode is about to be destroyed, there is no need for i_ceph_lock.
Sage Weila8599bd2009-10-06 11:31:12 -07001061 */
1062void ceph_queue_caps_release(struct inode *inode)
1063{
1064 struct ceph_inode_info *ci = ceph_inode(inode);
1065 struct rb_node *p;
1066
Sage Weila8599bd2009-10-06 11:31:12 -07001067 p = rb_first(&ci->i_caps);
1068 while (p) {
1069 struct ceph_cap *cap = rb_entry(p, struct ceph_cap, ci_node);
1070 struct ceph_mds_session *session = cap->session;
Sage Weila8599bd2009-10-06 11:31:12 -07001071
Sage Weil3d7ded42010-06-09 16:47:10 -07001072 __queue_cap_release(session, ceph_ino(inode), cap->cap_id,
1073 cap->mseq, cap->issue_seq);
Sage Weila8599bd2009-10-06 11:31:12 -07001074 p = rb_next(p);
Sage Weil7c1332b2010-02-16 11:39:45 -08001075 __ceph_remove_cap(cap);
Sage Weila8599bd2009-10-06 11:31:12 -07001076 }
Sage Weila8599bd2009-10-06 11:31:12 -07001077}
1078
1079/*
1080 * Send a cap msg on the given inode. Update our caps state, then
Sage Weilbe655592011-11-30 09:47:09 -08001081 * drop i_ceph_lock and send the message.
Sage Weila8599bd2009-10-06 11:31:12 -07001082 *
1083 * Make note of max_size reported/requested from mds, revoked caps
1084 * that have now been implemented.
1085 *
1086 * Make half-hearted attempt ot to invalidate page cache if we are
1087 * dropping RDCACHE. Note that this will leave behind locked pages
1088 * that we'll then need to deal with elsewhere.
1089 *
1090 * Return non-zero if delayed release, or we experienced an error
1091 * such that the caller should requeue + retry later.
1092 *
Sage Weilbe655592011-11-30 09:47:09 -08001093 * called with i_ceph_lock, then drops it.
Sage Weila8599bd2009-10-06 11:31:12 -07001094 * caller should hold snap_rwsem (read), s_mutex.
1095 */
1096static int __send_cap(struct ceph_mds_client *mdsc, struct ceph_cap *cap,
1097 int op, int used, int want, int retain, int flushing,
1098 unsigned *pflush_tid)
Sage Weilbe655592011-11-30 09:47:09 -08001099 __releases(cap->ci->i_ceph_lock)
Sage Weila8599bd2009-10-06 11:31:12 -07001100{
1101 struct ceph_inode_info *ci = cap->ci;
1102 struct inode *inode = &ci->vfs_inode;
1103 u64 cap_id = cap->cap_id;
Sage Weil68c28322010-02-09 13:41:47 -08001104 int held, revoking, dropping, keep;
Sage Weila8599bd2009-10-06 11:31:12 -07001105 u64 seq, issue_seq, mseq, time_warp_seq, follows;
1106 u64 size, max_size;
1107 struct timespec mtime, atime;
1108 int wake = 0;
Al Viro5706b272011-07-26 04:52:22 -04001109 umode_t mode;
Eric W. Biederman05cb11c2013-01-31 02:56:19 -08001110 kuid_t uid;
1111 kgid_t gid;
Sage Weila8599bd2009-10-06 11:31:12 -07001112 struct ceph_mds_session *session;
1113 u64 xattr_version = 0;
Sage Weil082afec2010-08-22 15:16:41 -07001114 struct ceph_buffer *xattr_blob = NULL;
Sage Weila8599bd2009-10-06 11:31:12 -07001115 int delayed = 0;
1116 u64 flush_tid = 0;
1117 int i;
1118 int ret;
1119
Sage Weil68c28322010-02-09 13:41:47 -08001120 held = cap->issued | cap->implemented;
1121 revoking = cap->implemented & ~cap->issued;
1122 retain &= ~revoking;
1123 dropping = cap->issued & ~retain;
1124
Sage Weila8599bd2009-10-06 11:31:12 -07001125 dout("__send_cap %p cap %p session %p %s -> %s (revoking %s)\n",
1126 inode, cap, cap->session,
1127 ceph_cap_string(held), ceph_cap_string(held & retain),
1128 ceph_cap_string(revoking));
1129 BUG_ON((retain & CEPH_CAP_PIN) == 0);
1130
1131 session = cap->session;
1132
1133 /* don't release wanted unless we've waited a bit. */
1134 if ((ci->i_ceph_flags & CEPH_I_NODELAY) == 0 &&
1135 time_before(jiffies, ci->i_hold_caps_min)) {
1136 dout(" delaying issued %s -> %s, wanted %s -> %s on send\n",
1137 ceph_cap_string(cap->issued),
1138 ceph_cap_string(cap->issued & retain),
1139 ceph_cap_string(cap->mds_wanted),
1140 ceph_cap_string(want));
1141 want |= cap->mds_wanted;
1142 retain |= cap->issued;
1143 delayed = 1;
1144 }
1145 ci->i_ceph_flags &= ~(CEPH_I_NODELAY | CEPH_I_FLUSH);
1146
1147 cap->issued &= retain; /* drop bits we don't want */
1148 if (cap->implemented & ~cap->issued) {
1149 /*
1150 * Wake up any waiters on wanted -> needed transition.
1151 * This is due to the weird transition from buffered
1152 * to sync IO... we need to flush dirty pages _before_
1153 * allowing sync writes to avoid reordering.
1154 */
1155 wake = 1;
1156 }
1157 cap->implemented &= cap->issued | used;
1158 cap->mds_wanted = want;
1159
1160 if (flushing) {
1161 /*
1162 * assign a tid for flush operations so we can avoid
1163 * flush1 -> dirty1 -> flush2 -> flushack1 -> mark
1164 * clean type races. track latest tid for every bit
1165 * so we can handle flush AxFw, flush Fw, and have the
1166 * first ack clean Ax.
1167 */
1168 flush_tid = ++ci->i_cap_flush_last_tid;
1169 if (pflush_tid)
1170 *pflush_tid = flush_tid;
1171 dout(" cap_flush_tid %d\n", (int)flush_tid);
1172 for (i = 0; i < CEPH_CAP_BITS; i++)
1173 if (flushing & (1 << i))
1174 ci->i_cap_flush_tid[i] = flush_tid;
Sage Weil7d8cb262010-08-24 08:44:16 -07001175
1176 follows = ci->i_head_snapc->seq;
1177 } else {
1178 follows = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07001179 }
1180
1181 keep = cap->implemented;
1182 seq = cap->seq;
1183 issue_seq = cap->issue_seq;
1184 mseq = cap->mseq;
1185 size = inode->i_size;
1186 ci->i_reported_size = size;
1187 max_size = ci->i_wanted_max_size;
1188 ci->i_requested_max_size = max_size;
1189 mtime = inode->i_mtime;
1190 atime = inode->i_atime;
1191 time_warp_seq = ci->i_time_warp_seq;
Sage Weila8599bd2009-10-06 11:31:12 -07001192 uid = inode->i_uid;
1193 gid = inode->i_gid;
1194 mode = inode->i_mode;
1195
Sage Weil082afec2010-08-22 15:16:41 -07001196 if (flushing & CEPH_CAP_XATTR_EXCL) {
Sage Weila8599bd2009-10-06 11:31:12 -07001197 __ceph_build_xattrs_blob(ci);
Sage Weil082afec2010-08-22 15:16:41 -07001198 xattr_blob = ci->i_xattrs.blob;
1199 xattr_version = ci->i_xattrs.version;
Sage Weila8599bd2009-10-06 11:31:12 -07001200 }
1201
Sage Weilbe655592011-11-30 09:47:09 -08001202 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001203
Sage Weila8599bd2009-10-06 11:31:12 -07001204 ret = send_cap_msg(session, ceph_vino(inode).ino, cap_id,
1205 op, keep, want, flushing, seq, flush_tid, issue_seq, mseq,
1206 size, max_size, &mtime, &atime, time_warp_seq,
Sage Weil082afec2010-08-22 15:16:41 -07001207 uid, gid, mode, xattr_version, xattr_blob,
Sage Weila8599bd2009-10-06 11:31:12 -07001208 follows);
1209 if (ret < 0) {
1210 dout("error sending cap msg, must requeue %p\n", inode);
1211 delayed = 1;
1212 }
1213
1214 if (wake)
Yehuda Sadeh03066f22010-07-27 13:11:08 -07001215 wake_up_all(&ci->i_cap_wq);
Sage Weila8599bd2009-10-06 11:31:12 -07001216
1217 return delayed;
1218}
1219
1220/*
1221 * When a snapshot is taken, clients accumulate dirty metadata on
1222 * inodes with capabilities in ceph_cap_snaps to describe the file
1223 * state at the time the snapshot was taken. This must be flushed
1224 * asynchronously back to the MDS once sync writes complete and dirty
1225 * data is written out.
1226 *
Sage Weile8351242010-09-17 08:03:08 -07001227 * Unless @again is true, skip cap_snaps that were already sent to
1228 * the MDS (i.e., during this session).
1229 *
Sage Weilbe655592011-11-30 09:47:09 -08001230 * Called under i_ceph_lock. Takes s_mutex as needed.
Sage Weila8599bd2009-10-06 11:31:12 -07001231 */
1232void __ceph_flush_snaps(struct ceph_inode_info *ci,
Sage Weile8351242010-09-17 08:03:08 -07001233 struct ceph_mds_session **psession,
1234 int again)
Sage Weilbe655592011-11-30 09:47:09 -08001235 __releases(ci->i_ceph_lock)
1236 __acquires(ci->i_ceph_lock)
Sage Weila8599bd2009-10-06 11:31:12 -07001237{
1238 struct inode *inode = &ci->vfs_inode;
1239 int mds;
1240 struct ceph_cap_snap *capsnap;
1241 u32 mseq;
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001242 struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
Sage Weila8599bd2009-10-06 11:31:12 -07001243 struct ceph_mds_session *session = NULL; /* if session != NULL, we hold
1244 session->s_mutex */
1245 u64 next_follows = 0; /* keep track of how far we've gotten through the
1246 i_cap_snaps list, and skip these entries next time
1247 around to avoid an infinite loop */
1248
1249 if (psession)
1250 session = *psession;
1251
1252 dout("__flush_snaps %p\n", inode);
1253retry:
1254 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
1255 /* avoid an infiniute loop after retry */
1256 if (capsnap->follows < next_follows)
1257 continue;
1258 /*
1259 * we need to wait for sync writes to complete and for dirty
1260 * pages to be written out.
1261 */
1262 if (capsnap->dirty_pages || capsnap->writing)
Sage Weilcfc0bf62010-09-14 15:50:59 -07001263 break;
Sage Weila8599bd2009-10-06 11:31:12 -07001264
Sage Weil819ccbf2010-04-01 09:33:46 -07001265 /*
1266 * if cap writeback already occurred, we should have dropped
1267 * the capsnap in ceph_put_wrbuffer_cap_refs.
1268 */
1269 BUG_ON(capsnap->dirty == 0);
1270
Sage Weila8599bd2009-10-06 11:31:12 -07001271 /* pick mds, take s_mutex */
Sage Weilca81f3f2010-06-10 14:21:36 -07001272 if (ci->i_auth_cap == NULL) {
1273 dout("no auth cap (migrating?), doing nothing\n");
1274 goto out;
1275 }
Sage Weile8351242010-09-17 08:03:08 -07001276
1277 /* only flush each capsnap once */
1278 if (!again && !list_empty(&capsnap->flushing_item)) {
1279 dout("already flushed %p, skipping\n", capsnap);
1280 continue;
1281 }
1282
Sage Weilca81f3f2010-06-10 14:21:36 -07001283 mds = ci->i_auth_cap->session->s_mds;
1284 mseq = ci->i_auth_cap->mseq;
1285
Sage Weila8599bd2009-10-06 11:31:12 -07001286 if (session && session->s_mds != mds) {
1287 dout("oops, wrong session %p mutex\n", session);
1288 mutex_unlock(&session->s_mutex);
1289 ceph_put_mds_session(session);
1290 session = NULL;
1291 }
1292 if (!session) {
Sage Weilbe655592011-11-30 09:47:09 -08001293 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001294 mutex_lock(&mdsc->mutex);
1295 session = __ceph_lookup_mds_session(mdsc, mds);
1296 mutex_unlock(&mdsc->mutex);
1297 if (session) {
1298 dout("inverting session/ino locks on %p\n",
1299 session);
1300 mutex_lock(&session->s_mutex);
1301 }
1302 /*
1303 * if session == NULL, we raced against a cap
Sage Weilca81f3f2010-06-10 14:21:36 -07001304 * deletion or migration. retry, and we'll
1305 * get a better @mds value next time.
Sage Weila8599bd2009-10-06 11:31:12 -07001306 */
Sage Weilbe655592011-11-30 09:47:09 -08001307 spin_lock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001308 goto retry;
1309 }
1310
1311 capsnap->flush_tid = ++ci->i_cap_flush_last_tid;
1312 atomic_inc(&capsnap->nref);
1313 if (!list_empty(&capsnap->flushing_item))
1314 list_del_init(&capsnap->flushing_item);
1315 list_add_tail(&capsnap->flushing_item,
1316 &session->s_cap_snaps_flushing);
Sage Weilbe655592011-11-30 09:47:09 -08001317 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001318
Sage Weilcfc0bf62010-09-14 15:50:59 -07001319 dout("flush_snaps %p cap_snap %p follows %lld tid %llu\n",
1320 inode, capsnap, capsnap->follows, capsnap->flush_tid);
Sage Weila8599bd2009-10-06 11:31:12 -07001321 send_cap_msg(session, ceph_vino(inode).ino, 0,
1322 CEPH_CAP_OP_FLUSHSNAP, capsnap->issued, 0,
1323 capsnap->dirty, 0, capsnap->flush_tid, 0, mseq,
1324 capsnap->size, 0,
1325 &capsnap->mtime, &capsnap->atime,
1326 capsnap->time_warp_seq,
1327 capsnap->uid, capsnap->gid, capsnap->mode,
Sage Weil4a625be2010-08-22 15:03:56 -07001328 capsnap->xattr_version, capsnap->xattr_blob,
Sage Weila8599bd2009-10-06 11:31:12 -07001329 capsnap->follows);
1330
1331 next_follows = capsnap->follows + 1;
1332 ceph_put_cap_snap(capsnap);
1333
Sage Weilbe655592011-11-30 09:47:09 -08001334 spin_lock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001335 goto retry;
1336 }
1337
1338 /* we flushed them all; remove this inode from the queue */
1339 spin_lock(&mdsc->snap_flush_lock);
1340 list_del_init(&ci->i_snap_flush_item);
1341 spin_unlock(&mdsc->snap_flush_lock);
1342
Sage Weilca81f3f2010-06-10 14:21:36 -07001343out:
Sage Weila8599bd2009-10-06 11:31:12 -07001344 if (psession)
1345 *psession = session;
1346 else if (session) {
1347 mutex_unlock(&session->s_mutex);
1348 ceph_put_mds_session(session);
1349 }
1350}
1351
1352static void ceph_flush_snaps(struct ceph_inode_info *ci)
1353{
Sage Weilbe655592011-11-30 09:47:09 -08001354 spin_lock(&ci->i_ceph_lock);
Sage Weile8351242010-09-17 08:03:08 -07001355 __ceph_flush_snaps(ci, NULL, 0);
Sage Weilbe655592011-11-30 09:47:09 -08001356 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001357}
1358
1359/*
Sage Weilfca65b42011-05-04 11:33:47 -07001360 * Mark caps dirty. If inode is newly dirty, return the dirty flags.
1361 * Caller is then responsible for calling __mark_inode_dirty with the
1362 * returned flags value.
Sage Weil76e3b392009-10-15 18:13:53 -07001363 */
Sage Weilfca65b42011-05-04 11:33:47 -07001364int __ceph_mark_dirty_caps(struct ceph_inode_info *ci, int mask)
Sage Weil76e3b392009-10-15 18:13:53 -07001365{
Cheng Renquan640ef792010-03-26 17:40:33 +08001366 struct ceph_mds_client *mdsc =
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001367 ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc;
Sage Weil76e3b392009-10-15 18:13:53 -07001368 struct inode *inode = &ci->vfs_inode;
1369 int was = ci->i_dirty_caps;
1370 int dirty = 0;
1371
1372 dout("__mark_dirty_caps %p %s dirty %s -> %s\n", &ci->vfs_inode,
1373 ceph_cap_string(mask), ceph_cap_string(was),
1374 ceph_cap_string(was | mask));
1375 ci->i_dirty_caps |= mask;
1376 if (was == 0) {
Sage Weil7d8cb262010-08-24 08:44:16 -07001377 if (!ci->i_head_snapc)
1378 ci->i_head_snapc = ceph_get_snap_context(
1379 ci->i_snap_realm->cached_context);
Yan, Zheng06852352012-11-19 10:49:07 +08001380 dout(" inode %p now dirty snapc %p auth cap %p\n",
1381 &ci->vfs_inode, ci->i_head_snapc, ci->i_auth_cap);
Sage Weil76e3b392009-10-15 18:13:53 -07001382 BUG_ON(!list_empty(&ci->i_dirty_item));
1383 spin_lock(&mdsc->cap_dirty_lock);
Yan, Zheng06852352012-11-19 10:49:07 +08001384 if (ci->i_auth_cap)
1385 list_add(&ci->i_dirty_item, &mdsc->cap_dirty);
1386 else
1387 list_add(&ci->i_dirty_item,
1388 &mdsc->cap_dirty_migrating);
Sage Weil76e3b392009-10-15 18:13:53 -07001389 spin_unlock(&mdsc->cap_dirty_lock);
1390 if (ci->i_flushing_caps == 0) {
Sage Weil3772d262011-05-03 09:28:08 -07001391 ihold(inode);
Sage Weil76e3b392009-10-15 18:13:53 -07001392 dirty |= I_DIRTY_SYNC;
1393 }
1394 }
1395 BUG_ON(list_empty(&ci->i_dirty_item));
1396 if (((was | ci->i_flushing_caps) & CEPH_CAP_FILE_BUFFER) &&
1397 (mask & CEPH_CAP_FILE_BUFFER))
1398 dirty |= I_DIRTY_DATASYNC;
Sage Weil76e3b392009-10-15 18:13:53 -07001399 __cap_delay_requeue(mdsc, ci);
Sage Weilfca65b42011-05-04 11:33:47 -07001400 return dirty;
Sage Weil76e3b392009-10-15 18:13:53 -07001401}
1402
1403/*
Sage Weila8599bd2009-10-06 11:31:12 -07001404 * Add dirty inode to the flushing list. Assigned a seq number so we
1405 * can wait for caps to flush without starving.
Sage Weilcdc35f92009-10-14 14:24:19 -07001406 *
Sage Weilbe655592011-11-30 09:47:09 -08001407 * Called under i_ceph_lock.
Sage Weila8599bd2009-10-06 11:31:12 -07001408 */
Sage Weilcdc35f92009-10-14 14:24:19 -07001409static int __mark_caps_flushing(struct inode *inode,
Sage Weila8599bd2009-10-06 11:31:12 -07001410 struct ceph_mds_session *session)
1411{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001412 struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
Sage Weila8599bd2009-10-06 11:31:12 -07001413 struct ceph_inode_info *ci = ceph_inode(inode);
Sage Weilcdc35f92009-10-14 14:24:19 -07001414 int flushing;
Sage Weil50b885b2009-12-01 14:12:07 -08001415
Sage Weilcdc35f92009-10-14 14:24:19 -07001416 BUG_ON(ci->i_dirty_caps == 0);
Sage Weila8599bd2009-10-06 11:31:12 -07001417 BUG_ON(list_empty(&ci->i_dirty_item));
Sage Weilcdc35f92009-10-14 14:24:19 -07001418
1419 flushing = ci->i_dirty_caps;
1420 dout("__mark_caps_flushing flushing %s, flushing_caps %s -> %s\n",
1421 ceph_cap_string(flushing),
1422 ceph_cap_string(ci->i_flushing_caps),
1423 ceph_cap_string(ci->i_flushing_caps | flushing));
1424 ci->i_flushing_caps |= flushing;
1425 ci->i_dirty_caps = 0;
Sage Weilafcdaea2009-10-14 14:27:38 -07001426 dout(" inode %p now !dirty\n", inode);
Sage Weilcdc35f92009-10-14 14:24:19 -07001427
Sage Weila8599bd2009-10-06 11:31:12 -07001428 spin_lock(&mdsc->cap_dirty_lock);
Sage Weilafcdaea2009-10-14 14:27:38 -07001429 list_del_init(&ci->i_dirty_item);
1430
1431 ci->i_cap_flush_seq = ++mdsc->cap_flush_seq;
Sage Weila8599bd2009-10-06 11:31:12 -07001432 if (list_empty(&ci->i_flushing_item)) {
1433 list_add_tail(&ci->i_flushing_item, &session->s_cap_flushing);
1434 mdsc->num_cap_flushing++;
Sage Weilafcdaea2009-10-14 14:27:38 -07001435 dout(" inode %p now flushing seq %lld\n", inode,
1436 ci->i_cap_flush_seq);
1437 } else {
1438 list_move_tail(&ci->i_flushing_item, &session->s_cap_flushing);
1439 dout(" inode %p now flushing (more) seq %lld\n", inode,
Sage Weila8599bd2009-10-06 11:31:12 -07001440 ci->i_cap_flush_seq);
1441 }
1442 spin_unlock(&mdsc->cap_dirty_lock);
Sage Weilcdc35f92009-10-14 14:24:19 -07001443
1444 return flushing;
Sage Weila8599bd2009-10-06 11:31:12 -07001445}
1446
1447/*
Sage Weil5ecad6f2010-02-17 10:43:37 -08001448 * try to invalidate mapping pages without blocking.
1449 */
Sage Weil5ecad6f2010-02-17 10:43:37 -08001450static int try_nonblocking_invalidate(struct inode *inode)
1451{
1452 struct ceph_inode_info *ci = ceph_inode(inode);
1453 u32 invalidating_gen = ci->i_rdcache_gen;
1454
Sage Weilbe655592011-11-30 09:47:09 -08001455 spin_unlock(&ci->i_ceph_lock);
Sage Weil5ecad6f2010-02-17 10:43:37 -08001456 invalidate_mapping_pages(&inode->i_data, 0, -1);
Sage Weilbe655592011-11-30 09:47:09 -08001457 spin_lock(&ci->i_ceph_lock);
Sage Weil5ecad6f2010-02-17 10:43:37 -08001458
Sage Weil18a38192010-09-17 10:46:44 -07001459 if (inode->i_data.nrpages == 0 &&
Sage Weil5ecad6f2010-02-17 10:43:37 -08001460 invalidating_gen == ci->i_rdcache_gen) {
1461 /* success. */
1462 dout("try_nonblocking_invalidate %p success\n", inode);
Sage Weilcd045cb2010-11-04 11:05:05 -07001463 /* save any racing async invalidate some trouble */
1464 ci->i_rdcache_revoking = ci->i_rdcache_gen - 1;
Sage Weil5ecad6f2010-02-17 10:43:37 -08001465 return 0;
1466 }
1467 dout("try_nonblocking_invalidate %p failed\n", inode);
1468 return -1;
1469}
1470
1471/*
Sage Weila8599bd2009-10-06 11:31:12 -07001472 * Swiss army knife function to examine currently used and wanted
1473 * versus held caps. Release, flush, ack revoked caps to mds as
1474 * appropriate.
1475 *
1476 * CHECK_CAPS_NODELAY - caller is delayed work and we should not delay
1477 * cap release further.
1478 * CHECK_CAPS_AUTHONLY - we should only check the auth cap
1479 * CHECK_CAPS_FLUSH - we should flush any dirty caps immediately, without
1480 * further delay.
1481 */
1482void ceph_check_caps(struct ceph_inode_info *ci, int flags,
1483 struct ceph_mds_session *session)
1484{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001485 struct ceph_fs_client *fsc = ceph_inode_to_client(&ci->vfs_inode);
1486 struct ceph_mds_client *mdsc = fsc->mdsc;
Sage Weila8599bd2009-10-06 11:31:12 -07001487 struct inode *inode = &ci->vfs_inode;
1488 struct ceph_cap *cap;
Yan, Zheng395c3122013-01-04 14:37:57 +08001489 int file_wanted, used, cap_used;
Sage Weila8599bd2009-10-06 11:31:12 -07001490 int took_snap_rwsem = 0; /* true if mdsc->snap_rwsem held */
Sage Weilcbd03632010-02-09 13:41:18 -08001491 int issued, implemented, want, retain, revoking, flushing = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07001492 int mds = -1; /* keep track of how far we've gone through i_caps list
1493 to avoid an infinite loop on retry */
1494 struct rb_node *p;
1495 int tried_invalidate = 0;
1496 int delayed = 0, sent = 0, force_requeue = 0, num;
Sage Weilcbd03632010-02-09 13:41:18 -08001497 int queue_invalidate = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07001498 int is_delayed = flags & CHECK_CAPS_NODELAY;
1499
1500 /* if we are unmounting, flush any unused caps immediately. */
1501 if (mdsc->stopping)
1502 is_delayed = 1;
1503
Sage Weilbe655592011-11-30 09:47:09 -08001504 spin_lock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001505
1506 if (ci->i_ceph_flags & CEPH_I_FLUSH)
1507 flags |= CHECK_CAPS_FLUSH;
1508
1509 /* flush snaps first time around only */
1510 if (!list_empty(&ci->i_cap_snaps))
Sage Weile8351242010-09-17 08:03:08 -07001511 __ceph_flush_snaps(ci, &session, 0);
Sage Weila8599bd2009-10-06 11:31:12 -07001512 goto retry_locked;
1513retry:
Sage Weilbe655592011-11-30 09:47:09 -08001514 spin_lock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001515retry_locked:
1516 file_wanted = __ceph_caps_file_wanted(ci);
1517 used = __ceph_caps_used(ci);
1518 want = file_wanted | used;
Sage Weilcbd03632010-02-09 13:41:18 -08001519 issued = __ceph_caps_issued(ci, &implemented);
1520 revoking = implemented & ~issued;
Sage Weila8599bd2009-10-06 11:31:12 -07001521
1522 retain = want | CEPH_CAP_PIN;
1523 if (!mdsc->stopping && inode->i_nlink > 0) {
1524 if (want) {
1525 retain |= CEPH_CAP_ANY; /* be greedy */
1526 } else {
1527 retain |= CEPH_CAP_ANY_SHARED;
1528 /*
1529 * keep RD only if we didn't have the file open RW,
1530 * because then the mds would revoke it anyway to
1531 * journal max_size=0.
1532 */
1533 if (ci->i_max_size == 0)
1534 retain |= CEPH_CAP_ANY_RD;
1535 }
1536 }
1537
1538 dout("check_caps %p file_want %s used %s dirty %s flushing %s"
Sage Weilcbd03632010-02-09 13:41:18 -08001539 " issued %s revoking %s retain %s %s%s%s\n", inode,
Sage Weila8599bd2009-10-06 11:31:12 -07001540 ceph_cap_string(file_wanted),
1541 ceph_cap_string(used), ceph_cap_string(ci->i_dirty_caps),
1542 ceph_cap_string(ci->i_flushing_caps),
Sage Weilcbd03632010-02-09 13:41:18 -08001543 ceph_cap_string(issued), ceph_cap_string(revoking),
Sage Weila8599bd2009-10-06 11:31:12 -07001544 ceph_cap_string(retain),
1545 (flags & CHECK_CAPS_AUTHONLY) ? " AUTHONLY" : "",
1546 (flags & CHECK_CAPS_NODELAY) ? " NODELAY" : "",
1547 (flags & CHECK_CAPS_FLUSH) ? " FLUSH" : "");
1548
1549 /*
1550 * If we no longer need to hold onto old our caps, and we may
1551 * have cached pages, but don't want them, then try to invalidate.
1552 * If we fail, it's because pages are locked.... try again later.
1553 */
1554 if ((!is_delayed || mdsc->stopping) &&
1555 ci->i_wrbuffer_ref == 0 && /* no dirty pages... */
Sage Weil93afd442010-09-17 08:38:25 -07001556 inode->i_data.nrpages && /* have cached pages */
Sage Weilcbd03632010-02-09 13:41:18 -08001557 (file_wanted == 0 || /* no open files */
Sage Weil29625072010-05-27 10:40:43 -07001558 (revoking & (CEPH_CAP_FILE_CACHE|
1559 CEPH_CAP_FILE_LAZYIO))) && /* or revoking cache */
Sage Weila8599bd2009-10-06 11:31:12 -07001560 !tried_invalidate) {
Sage Weila8599bd2009-10-06 11:31:12 -07001561 dout("check_caps trying to invalidate on %p\n", inode);
Sage Weil5ecad6f2010-02-17 10:43:37 -08001562 if (try_nonblocking_invalidate(inode) < 0) {
Sage Weil29625072010-05-27 10:40:43 -07001563 if (revoking & (CEPH_CAP_FILE_CACHE|
1564 CEPH_CAP_FILE_LAZYIO)) {
Sage Weil5ecad6f2010-02-17 10:43:37 -08001565 dout("check_caps queuing invalidate\n");
1566 queue_invalidate = 1;
1567 ci->i_rdcache_revoking = ci->i_rdcache_gen;
1568 } else {
1569 dout("check_caps failed to invalidate pages\n");
1570 /* we failed to invalidate pages. check these
1571 caps again later. */
1572 force_requeue = 1;
1573 __cap_set_timeouts(mdsc, ci);
1574 }
Sage Weila8599bd2009-10-06 11:31:12 -07001575 }
1576 tried_invalidate = 1;
1577 goto retry_locked;
1578 }
1579
1580 num = 0;
1581 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
1582 cap = rb_entry(p, struct ceph_cap, ci_node);
1583 num++;
1584
1585 /* avoid looping forever */
1586 if (mds >= cap->mds ||
1587 ((flags & CHECK_CAPS_AUTHONLY) && cap != ci->i_auth_cap))
1588 continue;
1589
1590 /* NOTE: no side-effects allowed, until we take s_mutex */
1591
Yan, Zheng395c3122013-01-04 14:37:57 +08001592 cap_used = used;
1593 if (ci->i_auth_cap && cap != ci->i_auth_cap)
1594 cap_used &= ~ci->i_auth_cap->issued;
1595
Sage Weila8599bd2009-10-06 11:31:12 -07001596 revoking = cap->implemented & ~cap->issued;
Yan, Zheng395c3122013-01-04 14:37:57 +08001597 dout(" mds%d cap %p used %s issued %s implemented %s revoking %s\n",
Sage Weil088b3f52011-01-18 08:56:01 -08001598 cap->mds, cap, ceph_cap_string(cap->issued),
Yan, Zheng395c3122013-01-04 14:37:57 +08001599 ceph_cap_string(cap_used),
Sage Weil088b3f52011-01-18 08:56:01 -08001600 ceph_cap_string(cap->implemented),
1601 ceph_cap_string(revoking));
Sage Weila8599bd2009-10-06 11:31:12 -07001602
1603 if (cap == ci->i_auth_cap &&
1604 (cap->issued & CEPH_CAP_FILE_WR)) {
1605 /* request larger max_size from MDS? */
1606 if (ci->i_wanted_max_size > ci->i_max_size &&
1607 ci->i_wanted_max_size > ci->i_requested_max_size) {
1608 dout("requesting new max_size\n");
1609 goto ack;
1610 }
1611
1612 /* approaching file_max? */
1613 if ((inode->i_size << 1) >= ci->i_max_size &&
1614 (ci->i_reported_size << 1) < ci->i_max_size) {
1615 dout("i_size approaching max_size\n");
1616 goto ack;
1617 }
1618 }
1619 /* flush anything dirty? */
1620 if (cap == ci->i_auth_cap && (flags & CHECK_CAPS_FLUSH) &&
1621 ci->i_dirty_caps) {
1622 dout("flushing dirty caps\n");
1623 goto ack;
1624 }
1625
1626 /* completed revocation? going down and there are no caps? */
Yan, Zheng395c3122013-01-04 14:37:57 +08001627 if (revoking && (revoking & cap_used) == 0) {
Sage Weila8599bd2009-10-06 11:31:12 -07001628 dout("completed revocation of %s\n",
1629 ceph_cap_string(cap->implemented & ~cap->issued));
1630 goto ack;
1631 }
1632
1633 /* want more caps from mds? */
1634 if (want & ~(cap->mds_wanted | cap->issued))
1635 goto ack;
1636
1637 /* things we might delay */
1638 if ((cap->issued & ~retain) == 0 &&
1639 cap->mds_wanted == want)
1640 continue; /* nope, all good */
1641
1642 if (is_delayed)
1643 goto ack;
1644
1645 /* delay? */
1646 if ((ci->i_ceph_flags & CEPH_I_NODELAY) == 0 &&
1647 time_before(jiffies, ci->i_hold_caps_max)) {
1648 dout(" delaying issued %s -> %s, wanted %s -> %s\n",
1649 ceph_cap_string(cap->issued),
1650 ceph_cap_string(cap->issued & retain),
1651 ceph_cap_string(cap->mds_wanted),
1652 ceph_cap_string(want));
1653 delayed++;
1654 continue;
1655 }
1656
1657ack:
Sage Weile9964c12010-03-01 15:16:56 -08001658 if (ci->i_ceph_flags & CEPH_I_NOFLUSH) {
1659 dout(" skipping %p I_NOFLUSH set\n", inode);
1660 continue;
1661 }
1662
Sage Weila8599bd2009-10-06 11:31:12 -07001663 if (session && session != cap->session) {
1664 dout("oops, wrong session %p mutex\n", session);
1665 mutex_unlock(&session->s_mutex);
1666 session = NULL;
1667 }
1668 if (!session) {
1669 session = cap->session;
1670 if (mutex_trylock(&session->s_mutex) == 0) {
1671 dout("inverting session/ino locks on %p\n",
1672 session);
Sage Weilbe655592011-11-30 09:47:09 -08001673 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001674 if (took_snap_rwsem) {
1675 up_read(&mdsc->snap_rwsem);
1676 took_snap_rwsem = 0;
1677 }
1678 mutex_lock(&session->s_mutex);
1679 goto retry;
1680 }
1681 }
1682 /* take snap_rwsem after session mutex */
1683 if (!took_snap_rwsem) {
1684 if (down_read_trylock(&mdsc->snap_rwsem) == 0) {
1685 dout("inverting snap/in locks on %p\n",
1686 inode);
Sage Weilbe655592011-11-30 09:47:09 -08001687 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001688 down_read(&mdsc->snap_rwsem);
1689 took_snap_rwsem = 1;
1690 goto retry;
1691 }
1692 took_snap_rwsem = 1;
1693 }
1694
Sage Weilcdc35f92009-10-14 14:24:19 -07001695 if (cap == ci->i_auth_cap && ci->i_dirty_caps)
1696 flushing = __mark_caps_flushing(inode, session);
Sage Weil24be0c42011-01-18 08:48:06 -08001697 else
1698 flushing = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07001699
1700 mds = cap->mds; /* remember mds, so we don't repeat */
1701 sent++;
1702
Sage Weilbe655592011-11-30 09:47:09 -08001703 /* __send_cap drops i_ceph_lock */
Yan, Zheng395c3122013-01-04 14:37:57 +08001704 delayed += __send_cap(mdsc, cap, CEPH_CAP_OP_UPDATE, cap_used,
1705 want, retain, flushing, NULL);
Sage Weilbe655592011-11-30 09:47:09 -08001706 goto retry; /* retake i_ceph_lock and restart our cap scan. */
Sage Weila8599bd2009-10-06 11:31:12 -07001707 }
1708
1709 /*
1710 * Reschedule delayed caps release if we delayed anything,
1711 * otherwise cancel.
1712 */
1713 if (delayed && is_delayed)
1714 force_requeue = 1; /* __send_cap delayed release; requeue */
1715 if (!delayed && !is_delayed)
1716 __cap_delay_cancel(mdsc, ci);
1717 else if (!is_delayed || force_requeue)
1718 __cap_delay_requeue(mdsc, ci);
1719
Sage Weilbe655592011-11-30 09:47:09 -08001720 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001721
Sage Weilcbd03632010-02-09 13:41:18 -08001722 if (queue_invalidate)
Sage Weil3c6f6b72010-02-09 15:24:44 -08001723 ceph_queue_invalidate(inode);
Sage Weilcbd03632010-02-09 13:41:18 -08001724
Sage Weilcdc2ce02010-03-16 13:39:28 -07001725 if (session)
Sage Weila8599bd2009-10-06 11:31:12 -07001726 mutex_unlock(&session->s_mutex);
1727 if (took_snap_rwsem)
1728 up_read(&mdsc->snap_rwsem);
1729}
1730
1731/*
Sage Weila8599bd2009-10-06 11:31:12 -07001732 * Try to flush dirty caps back to the auth mds.
1733 */
1734static int try_flush_caps(struct inode *inode, struct ceph_mds_session *session,
1735 unsigned *flush_tid)
1736{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001737 struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
Sage Weila8599bd2009-10-06 11:31:12 -07001738 struct ceph_inode_info *ci = ceph_inode(inode);
1739 int unlock_session = session ? 0 : 1;
1740 int flushing = 0;
1741
1742retry:
Sage Weilbe655592011-11-30 09:47:09 -08001743 spin_lock(&ci->i_ceph_lock);
Sage Weile9964c12010-03-01 15:16:56 -08001744 if (ci->i_ceph_flags & CEPH_I_NOFLUSH) {
1745 dout("try_flush_caps skipping %p I_NOFLUSH set\n", inode);
1746 goto out;
1747 }
Sage Weila8599bd2009-10-06 11:31:12 -07001748 if (ci->i_dirty_caps && ci->i_auth_cap) {
1749 struct ceph_cap *cap = ci->i_auth_cap;
1750 int used = __ceph_caps_used(ci);
1751 int want = __ceph_caps_wanted(ci);
1752 int delayed;
1753
1754 if (!session) {
Sage Weilbe655592011-11-30 09:47:09 -08001755 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001756 session = cap->session;
1757 mutex_lock(&session->s_mutex);
1758 goto retry;
1759 }
1760 BUG_ON(session != cap->session);
1761 if (cap->session->s_state < CEPH_MDS_SESSION_OPEN)
1762 goto out;
1763
Sage Weilcdc35f92009-10-14 14:24:19 -07001764 flushing = __mark_caps_flushing(inode, session);
Sage Weila8599bd2009-10-06 11:31:12 -07001765
Sage Weilbe655592011-11-30 09:47:09 -08001766 /* __send_cap drops i_ceph_lock */
Sage Weila8599bd2009-10-06 11:31:12 -07001767 delayed = __send_cap(mdsc, cap, CEPH_CAP_OP_FLUSH, used, want,
1768 cap->issued | cap->implemented, flushing,
1769 flush_tid);
1770 if (!delayed)
1771 goto out_unlocked;
1772
Sage Weilbe655592011-11-30 09:47:09 -08001773 spin_lock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001774 __cap_delay_requeue(mdsc, ci);
1775 }
1776out:
Sage Weilbe655592011-11-30 09:47:09 -08001777 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001778out_unlocked:
1779 if (session && unlock_session)
1780 mutex_unlock(&session->s_mutex);
1781 return flushing;
1782}
1783
1784/*
1785 * Return true if we've flushed caps through the given flush_tid.
1786 */
1787static int caps_are_flushed(struct inode *inode, unsigned tid)
1788{
1789 struct ceph_inode_info *ci = ceph_inode(inode);
Dan Carpentera5ee7512010-05-07 10:27:14 +02001790 int i, ret = 1;
Sage Weila8599bd2009-10-06 11:31:12 -07001791
Sage Weilbe655592011-11-30 09:47:09 -08001792 spin_lock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001793 for (i = 0; i < CEPH_CAP_BITS; i++)
1794 if ((ci->i_flushing_caps & (1 << i)) &&
1795 ci->i_cap_flush_tid[i] <= tid) {
1796 /* still flushing this bit */
1797 ret = 0;
1798 break;
1799 }
Sage Weilbe655592011-11-30 09:47:09 -08001800 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001801 return ret;
1802}
1803
1804/*
1805 * Wait on any unsafe replies for the given inode. First wait on the
1806 * newest request, and make that the upper bound. Then, if there are
1807 * more requests, keep waiting on the oldest as long as it is still older
1808 * than the original request.
1809 */
1810static void sync_write_wait(struct inode *inode)
1811{
1812 struct ceph_inode_info *ci = ceph_inode(inode);
1813 struct list_head *head = &ci->i_unsafe_writes;
1814 struct ceph_osd_request *req;
1815 u64 last_tid;
1816
1817 spin_lock(&ci->i_unsafe_lock);
1818 if (list_empty(head))
1819 goto out;
1820
1821 /* set upper bound as _last_ entry in chain */
1822 req = list_entry(head->prev, struct ceph_osd_request,
1823 r_unsafe_item);
1824 last_tid = req->r_tid;
1825
1826 do {
1827 ceph_osdc_get_request(req);
1828 spin_unlock(&ci->i_unsafe_lock);
1829 dout("sync_write_wait on tid %llu (until %llu)\n",
1830 req->r_tid, last_tid);
1831 wait_for_completion(&req->r_safe_completion);
1832 spin_lock(&ci->i_unsafe_lock);
1833 ceph_osdc_put_request(req);
1834
1835 /*
1836 * from here on look at first entry in chain, since we
1837 * only want to wait for anything older than last_tid
1838 */
1839 if (list_empty(head))
1840 break;
1841 req = list_entry(head->next, struct ceph_osd_request,
1842 r_unsafe_item);
1843 } while (req->r_tid < last_tid);
1844out:
1845 spin_unlock(&ci->i_unsafe_lock);
1846}
1847
Josef Bacik02c24a82011-07-16 20:44:56 -04001848int ceph_fsync(struct file *file, loff_t start, loff_t end, int datasync)
Sage Weila8599bd2009-10-06 11:31:12 -07001849{
Christoph Hellwig7ea80852010-05-26 17:53:25 +02001850 struct inode *inode = file->f_mapping->host;
Sage Weila8599bd2009-10-06 11:31:12 -07001851 struct ceph_inode_info *ci = ceph_inode(inode);
1852 unsigned flush_tid;
1853 int ret;
1854 int dirty;
1855
1856 dout("fsync %p%s\n", inode, datasync ? " datasync" : "");
1857 sync_write_wait(inode);
1858
Josef Bacik02c24a82011-07-16 20:44:56 -04001859 ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
Sage Weila8599bd2009-10-06 11:31:12 -07001860 if (ret < 0)
1861 return ret;
Josef Bacik02c24a82011-07-16 20:44:56 -04001862 mutex_lock(&inode->i_mutex);
Sage Weila8599bd2009-10-06 11:31:12 -07001863
1864 dirty = try_flush_caps(inode, NULL, &flush_tid);
1865 dout("fsync dirty caps are %s\n", ceph_cap_string(dirty));
1866
1867 /*
1868 * only wait on non-file metadata writeback (the mds
1869 * can recover size and mtime, so we don't need to
1870 * wait for that)
1871 */
1872 if (!datasync && (dirty & ~CEPH_CAP_ANY_FILE_WR)) {
1873 dout("fsync waiting for flush_tid %u\n", flush_tid);
1874 ret = wait_event_interruptible(ci->i_cap_wq,
1875 caps_are_flushed(inode, flush_tid));
1876 }
1877
1878 dout("fsync %p%s done\n", inode, datasync ? " datasync" : "");
Josef Bacik02c24a82011-07-16 20:44:56 -04001879 mutex_unlock(&inode->i_mutex);
Sage Weila8599bd2009-10-06 11:31:12 -07001880 return ret;
1881}
1882
1883/*
1884 * Flush any dirty caps back to the mds. If we aren't asked to wait,
1885 * queue inode for flush but don't do so immediately, because we can
1886 * get by with fewer MDS messages if we wait for data writeback to
1887 * complete first.
1888 */
Stephen Rothwellf1a3d572010-01-18 11:53:08 +11001889int ceph_write_inode(struct inode *inode, struct writeback_control *wbc)
Sage Weila8599bd2009-10-06 11:31:12 -07001890{
1891 struct ceph_inode_info *ci = ceph_inode(inode);
1892 unsigned flush_tid;
1893 int err = 0;
1894 int dirty;
Stephen Rothwellf1a3d572010-01-18 11:53:08 +11001895 int wait = wbc->sync_mode == WB_SYNC_ALL;
Sage Weila8599bd2009-10-06 11:31:12 -07001896
1897 dout("write_inode %p wait=%d\n", inode, wait);
1898 if (wait) {
1899 dirty = try_flush_caps(inode, NULL, &flush_tid);
1900 if (dirty)
1901 err = wait_event_interruptible(ci->i_cap_wq,
1902 caps_are_flushed(inode, flush_tid));
1903 } else {
Cheng Renquan640ef792010-03-26 17:40:33 +08001904 struct ceph_mds_client *mdsc =
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001905 ceph_sb_to_client(inode->i_sb)->mdsc;
Sage Weila8599bd2009-10-06 11:31:12 -07001906
Sage Weilbe655592011-11-30 09:47:09 -08001907 spin_lock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001908 if (__ceph_caps_dirty(ci))
1909 __cap_delay_requeue_front(mdsc, ci);
Sage Weilbe655592011-11-30 09:47:09 -08001910 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001911 }
1912 return err;
1913}
1914
1915/*
1916 * After a recovering MDS goes active, we need to resend any caps
1917 * we were flushing.
1918 *
1919 * Caller holds session->s_mutex.
1920 */
1921static void kick_flushing_capsnaps(struct ceph_mds_client *mdsc,
1922 struct ceph_mds_session *session)
1923{
1924 struct ceph_cap_snap *capsnap;
1925
1926 dout("kick_flushing_capsnaps mds%d\n", session->s_mds);
1927 list_for_each_entry(capsnap, &session->s_cap_snaps_flushing,
1928 flushing_item) {
1929 struct ceph_inode_info *ci = capsnap->ci;
1930 struct inode *inode = &ci->vfs_inode;
1931 struct ceph_cap *cap;
1932
Sage Weilbe655592011-11-30 09:47:09 -08001933 spin_lock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001934 cap = ci->i_auth_cap;
1935 if (cap && cap->session == session) {
1936 dout("kick_flushing_caps %p cap %p capsnap %p\n", inode,
1937 cap, capsnap);
Sage Weile8351242010-09-17 08:03:08 -07001938 __ceph_flush_snaps(ci, &session, 1);
Sage Weila8599bd2009-10-06 11:31:12 -07001939 } else {
1940 pr_err("%p auth cap %p not mds%d ???\n", inode,
1941 cap, session->s_mds);
Sage Weila8599bd2009-10-06 11:31:12 -07001942 }
Sage Weilbe655592011-11-30 09:47:09 -08001943 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001944 }
1945}
1946
1947void ceph_kick_flushing_caps(struct ceph_mds_client *mdsc,
1948 struct ceph_mds_session *session)
1949{
1950 struct ceph_inode_info *ci;
1951
1952 kick_flushing_capsnaps(mdsc, session);
1953
1954 dout("kick_flushing_caps mds%d\n", session->s_mds);
1955 list_for_each_entry(ci, &session->s_cap_flushing, i_flushing_item) {
1956 struct inode *inode = &ci->vfs_inode;
1957 struct ceph_cap *cap;
1958 int delayed = 0;
1959
Sage Weilbe655592011-11-30 09:47:09 -08001960 spin_lock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001961 cap = ci->i_auth_cap;
1962 if (cap && cap->session == session) {
1963 dout("kick_flushing_caps %p cap %p %s\n", inode,
1964 cap, ceph_cap_string(ci->i_flushing_caps));
1965 delayed = __send_cap(mdsc, cap, CEPH_CAP_OP_FLUSH,
1966 __ceph_caps_used(ci),
1967 __ceph_caps_wanted(ci),
1968 cap->issued | cap->implemented,
1969 ci->i_flushing_caps, NULL);
1970 if (delayed) {
Sage Weilbe655592011-11-30 09:47:09 -08001971 spin_lock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001972 __cap_delay_requeue(mdsc, ci);
Sage Weilbe655592011-11-30 09:47:09 -08001973 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001974 }
1975 } else {
1976 pr_err("%p auth cap %p not mds%d ???\n", inode,
1977 cap, session->s_mds);
Sage Weilbe655592011-11-30 09:47:09 -08001978 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001979 }
1980 }
1981}
1982
Sage Weil088b3f52011-01-18 08:56:01 -08001983static void kick_flushing_inode_caps(struct ceph_mds_client *mdsc,
1984 struct ceph_mds_session *session,
1985 struct inode *inode)
1986{
1987 struct ceph_inode_info *ci = ceph_inode(inode);
1988 struct ceph_cap *cap;
1989 int delayed = 0;
1990
Sage Weilbe655592011-11-30 09:47:09 -08001991 spin_lock(&ci->i_ceph_lock);
Sage Weil088b3f52011-01-18 08:56:01 -08001992 cap = ci->i_auth_cap;
1993 dout("kick_flushing_inode_caps %p flushing %s flush_seq %lld\n", inode,
1994 ceph_cap_string(ci->i_flushing_caps), ci->i_cap_flush_seq);
Yan, Zheng005c4692013-05-31 16:40:24 +08001995
Sage Weil088b3f52011-01-18 08:56:01 -08001996 __ceph_flush_snaps(ci, &session, 1);
Yan, Zheng005c4692013-05-31 16:40:24 +08001997
Sage Weil088b3f52011-01-18 08:56:01 -08001998 if (ci->i_flushing_caps) {
Yan, Zheng005c4692013-05-31 16:40:24 +08001999 spin_lock(&mdsc->cap_dirty_lock);
2000 list_move_tail(&ci->i_flushing_item,
2001 &cap->session->s_cap_flushing);
2002 spin_unlock(&mdsc->cap_dirty_lock);
2003
Sage Weil088b3f52011-01-18 08:56:01 -08002004 delayed = __send_cap(mdsc, cap, CEPH_CAP_OP_FLUSH,
2005 __ceph_caps_used(ci),
2006 __ceph_caps_wanted(ci),
2007 cap->issued | cap->implemented,
2008 ci->i_flushing_caps, NULL);
2009 if (delayed) {
Sage Weilbe655592011-11-30 09:47:09 -08002010 spin_lock(&ci->i_ceph_lock);
Sage Weil088b3f52011-01-18 08:56:01 -08002011 __cap_delay_requeue(mdsc, ci);
Sage Weilbe655592011-11-30 09:47:09 -08002012 spin_unlock(&ci->i_ceph_lock);
Sage Weil088b3f52011-01-18 08:56:01 -08002013 }
2014 } else {
Sage Weilbe655592011-11-30 09:47:09 -08002015 spin_unlock(&ci->i_ceph_lock);
Sage Weil088b3f52011-01-18 08:56:01 -08002016 }
2017}
2018
Sage Weila8599bd2009-10-06 11:31:12 -07002019
2020/*
2021 * Take references to capabilities we hold, so that we don't release
2022 * them to the MDS prematurely.
2023 *
Sage Weilbe655592011-11-30 09:47:09 -08002024 * Protected by i_ceph_lock.
Sage Weila8599bd2009-10-06 11:31:12 -07002025 */
2026static void __take_cap_refs(struct ceph_inode_info *ci, int got)
2027{
2028 if (got & CEPH_CAP_PIN)
2029 ci->i_pin_ref++;
2030 if (got & CEPH_CAP_FILE_RD)
2031 ci->i_rd_ref++;
2032 if (got & CEPH_CAP_FILE_CACHE)
2033 ci->i_rdcache_ref++;
2034 if (got & CEPH_CAP_FILE_WR)
2035 ci->i_wr_ref++;
2036 if (got & CEPH_CAP_FILE_BUFFER) {
Henry C Changd3d07202011-05-11 10:29:54 +00002037 if (ci->i_wb_ref == 0)
Sage Weil3772d262011-05-03 09:28:08 -07002038 ihold(&ci->vfs_inode);
Henry C Changd3d07202011-05-11 10:29:54 +00002039 ci->i_wb_ref++;
2040 dout("__take_cap_refs %p wb %d -> %d (?)\n",
2041 &ci->vfs_inode, ci->i_wb_ref-1, ci->i_wb_ref);
Sage Weila8599bd2009-10-06 11:31:12 -07002042 }
2043}
2044
2045/*
2046 * Try to grab cap references. Specify those refs we @want, and the
2047 * minimal set we @need. Also include the larger offset we are writing
2048 * to (when applicable), and check against max_size here as well.
2049 * Note that caller is responsible for ensuring max_size increases are
2050 * requested from the MDS.
2051 */
2052static int try_get_cap_refs(struct ceph_inode_info *ci, int need, int want,
2053 int *got, loff_t endoff, int *check_max, int *err)
2054{
2055 struct inode *inode = &ci->vfs_inode;
2056 int ret = 0;
2057 int have, implemented;
Sage Weil195d3ce2010-03-01 09:57:54 -08002058 int file_wanted;
Sage Weila8599bd2009-10-06 11:31:12 -07002059
2060 dout("get_cap_refs %p need %s want %s\n", inode,
2061 ceph_cap_string(need), ceph_cap_string(want));
Sage Weilbe655592011-11-30 09:47:09 -08002062 spin_lock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07002063
Sage Weil195d3ce2010-03-01 09:57:54 -08002064 /* make sure file is actually open */
2065 file_wanted = __ceph_caps_file_wanted(ci);
2066 if ((file_wanted & need) == 0) {
2067 dout("try_get_cap_refs need %s file_wanted %s, EBADF\n",
2068 ceph_cap_string(need), ceph_cap_string(file_wanted));
Sage Weila8599bd2009-10-06 11:31:12 -07002069 *err = -EBADF;
2070 ret = 1;
2071 goto out;
2072 }
2073
Yan, Zheng37505d52013-04-12 16:11:10 +08002074 /* finish pending truncate */
2075 while (ci->i_truncate_pending) {
2076 spin_unlock(&ci->i_ceph_lock);
Yan, Zhengb415bf42013-07-02 12:40:19 +08002077 __ceph_do_pending_vmtruncate(inode);
Yan, Zheng37505d52013-04-12 16:11:10 +08002078 spin_lock(&ci->i_ceph_lock);
2079 }
2080
Yan, Zheng3871cbb2013-08-05 14:10:29 +08002081 have = __ceph_caps_issued(ci, &implemented);
2082
2083 if (have & need & CEPH_CAP_FILE_WR) {
Sage Weila8599bd2009-10-06 11:31:12 -07002084 if (endoff >= 0 && endoff > (loff_t)ci->i_max_size) {
2085 dout("get_cap_refs %p endoff %llu > maxsize %llu\n",
2086 inode, endoff, ci->i_max_size);
Yan, Zheng3871cbb2013-08-05 14:10:29 +08002087 if (endoff > ci->i_requested_max_size) {
Sage Weila8599bd2009-10-06 11:31:12 -07002088 *check_max = 1;
2089 ret = 1;
2090 }
2091 goto out;
2092 }
2093 /*
2094 * If a sync write is in progress, we must wait, so that we
2095 * can get a final snapshot value for size+mtime.
2096 */
2097 if (__ceph_have_pending_cap_snap(ci)) {
2098 dout("get_cap_refs %p cap_snap_pending\n", inode);
2099 goto out;
2100 }
2101 }
Sage Weila8599bd2009-10-06 11:31:12 -07002102
Sage Weila8599bd2009-10-06 11:31:12 -07002103 if ((have & need) == need) {
2104 /*
2105 * Look at (implemented & ~have & not) so that we keep waiting
2106 * on transition from wanted -> needed caps. This is needed
2107 * for WRBUFFER|WR -> WR to avoid a new WR sync write from
2108 * going before a prior buffered writeback happens.
2109 */
2110 int not = want & ~(have & need);
2111 int revoking = implemented & ~have;
2112 dout("get_cap_refs %p have %s but not %s (revoking %s)\n",
2113 inode, ceph_cap_string(have), ceph_cap_string(not),
2114 ceph_cap_string(revoking));
2115 if ((revoking & not) == 0) {
2116 *got = need | (have & want);
2117 __take_cap_refs(ci, *got);
2118 ret = 1;
2119 }
2120 } else {
2121 dout("get_cap_refs %p have %s needed %s\n", inode,
2122 ceph_cap_string(have), ceph_cap_string(need));
2123 }
2124out:
Sage Weilbe655592011-11-30 09:47:09 -08002125 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07002126 dout("get_cap_refs %p ret %d got %s\n", inode,
2127 ret, ceph_cap_string(*got));
2128 return ret;
2129}
2130
2131/*
2132 * Check the offset we are writing up to against our current
2133 * max_size. If necessary, tell the MDS we want to write to
2134 * a larger offset.
2135 */
2136static void check_max_size(struct inode *inode, loff_t endoff)
2137{
2138 struct ceph_inode_info *ci = ceph_inode(inode);
2139 int check = 0;
2140
2141 /* do we need to explicitly request a larger max_size? */
Sage Weilbe655592011-11-30 09:47:09 -08002142 spin_lock(&ci->i_ceph_lock);
Yan, Zheng3871cbb2013-08-05 14:10:29 +08002143 if (endoff >= ci->i_max_size && endoff > ci->i_wanted_max_size) {
Sage Weila8599bd2009-10-06 11:31:12 -07002144 dout("write %p at large endoff %llu, req max_size\n",
2145 inode, endoff);
2146 ci->i_wanted_max_size = endoff;
Sage Weila8599bd2009-10-06 11:31:12 -07002147 }
Yan, Zheng3871cbb2013-08-05 14:10:29 +08002148 /* duplicate ceph_check_caps()'s logic */
2149 if (ci->i_auth_cap &&
2150 (ci->i_auth_cap->issued & CEPH_CAP_FILE_WR) &&
2151 ci->i_wanted_max_size > ci->i_max_size &&
2152 ci->i_wanted_max_size > ci->i_requested_max_size)
2153 check = 1;
Sage Weilbe655592011-11-30 09:47:09 -08002154 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07002155 if (check)
2156 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
2157}
2158
2159/*
2160 * Wait for caps, and take cap references. If we can't get a WR cap
2161 * due to a small max_size, make sure we check_max_size (and possibly
2162 * ask the mds) so we don't get hung up indefinitely.
2163 */
2164int ceph_get_caps(struct ceph_inode_info *ci, int need, int want, int *got,
2165 loff_t endoff)
2166{
2167 int check_max, ret, err;
2168
2169retry:
2170 if (endoff > 0)
2171 check_max_size(&ci->vfs_inode, endoff);
2172 check_max = 0;
2173 err = 0;
2174 ret = wait_event_interruptible(ci->i_cap_wq,
2175 try_get_cap_refs(ci, need, want,
2176 got, endoff,
2177 &check_max, &err));
2178 if (err)
2179 ret = err;
2180 if (check_max)
2181 goto retry;
2182 return ret;
2183}
2184
2185/*
2186 * Take cap refs. Caller must already know we hold at least one ref
2187 * on the caps in question or we don't know this is safe.
2188 */
2189void ceph_get_cap_refs(struct ceph_inode_info *ci, int caps)
2190{
Sage Weilbe655592011-11-30 09:47:09 -08002191 spin_lock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07002192 __take_cap_refs(ci, caps);
Sage Weilbe655592011-11-30 09:47:09 -08002193 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07002194}
2195
2196/*
2197 * Release cap refs.
2198 *
2199 * If we released the last ref on any given cap, call ceph_check_caps
2200 * to release (or schedule a release).
2201 *
2202 * If we are releasing a WR cap (from a sync write), finalize any affected
2203 * cap_snap, and wake up any waiters.
2204 */
2205void ceph_put_cap_refs(struct ceph_inode_info *ci, int had)
2206{
2207 struct inode *inode = &ci->vfs_inode;
2208 int last = 0, put = 0, flushsnaps = 0, wake = 0;
2209 struct ceph_cap_snap *capsnap;
2210
Sage Weilbe655592011-11-30 09:47:09 -08002211 spin_lock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07002212 if (had & CEPH_CAP_PIN)
2213 --ci->i_pin_ref;
2214 if (had & CEPH_CAP_FILE_RD)
2215 if (--ci->i_rd_ref == 0)
2216 last++;
2217 if (had & CEPH_CAP_FILE_CACHE)
2218 if (--ci->i_rdcache_ref == 0)
2219 last++;
2220 if (had & CEPH_CAP_FILE_BUFFER) {
Henry C Changd3d07202011-05-11 10:29:54 +00002221 if (--ci->i_wb_ref == 0) {
Sage Weila8599bd2009-10-06 11:31:12 -07002222 last++;
2223 put++;
2224 }
Henry C Changd3d07202011-05-11 10:29:54 +00002225 dout("put_cap_refs %p wb %d -> %d (?)\n",
2226 inode, ci->i_wb_ref+1, ci->i_wb_ref);
Sage Weila8599bd2009-10-06 11:31:12 -07002227 }
2228 if (had & CEPH_CAP_FILE_WR)
2229 if (--ci->i_wr_ref == 0) {
2230 last++;
2231 if (!list_empty(&ci->i_cap_snaps)) {
2232 capsnap = list_first_entry(&ci->i_cap_snaps,
2233 struct ceph_cap_snap,
2234 ci_item);
2235 if (capsnap->writing) {
2236 capsnap->writing = 0;
2237 flushsnaps =
2238 __ceph_finish_cap_snap(ci,
2239 capsnap);
2240 wake = 1;
2241 }
2242 }
2243 }
Sage Weilbe655592011-11-30 09:47:09 -08002244 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07002245
Sage Weil819ccbf2010-04-01 09:33:46 -07002246 dout("put_cap_refs %p had %s%s%s\n", inode, ceph_cap_string(had),
2247 last ? " last" : "", put ? " put" : "");
Sage Weila8599bd2009-10-06 11:31:12 -07002248
2249 if (last && !flushsnaps)
2250 ceph_check_caps(ci, 0, NULL);
2251 else if (flushsnaps)
2252 ceph_flush_snaps(ci);
2253 if (wake)
Yehuda Sadeh03066f22010-07-27 13:11:08 -07002254 wake_up_all(&ci->i_cap_wq);
Sage Weila8599bd2009-10-06 11:31:12 -07002255 if (put)
2256 iput(inode);
2257}
2258
2259/*
2260 * Release @nr WRBUFFER refs on dirty pages for the given @snapc snap
2261 * context. Adjust per-snap dirty page accounting as appropriate.
2262 * Once all dirty data for a cap_snap is flushed, flush snapped file
2263 * metadata back to the MDS. If we dropped the last ref, call
2264 * ceph_check_caps.
2265 */
2266void ceph_put_wrbuffer_cap_refs(struct ceph_inode_info *ci, int nr,
2267 struct ceph_snap_context *snapc)
2268{
2269 struct inode *inode = &ci->vfs_inode;
2270 int last = 0;
Sage Weil819ccbf2010-04-01 09:33:46 -07002271 int complete_capsnap = 0;
2272 int drop_capsnap = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07002273 int found = 0;
2274 struct ceph_cap_snap *capsnap = NULL;
2275
Sage Weilbe655592011-11-30 09:47:09 -08002276 spin_lock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07002277 ci->i_wrbuffer_ref -= nr;
2278 last = !ci->i_wrbuffer_ref;
2279
2280 if (ci->i_head_snapc == snapc) {
2281 ci->i_wrbuffer_ref_head -= nr;
Sage Weil7d8cb262010-08-24 08:44:16 -07002282 if (ci->i_wrbuffer_ref_head == 0 &&
2283 ci->i_dirty_caps == 0 && ci->i_flushing_caps == 0) {
2284 BUG_ON(!ci->i_head_snapc);
Sage Weila8599bd2009-10-06 11:31:12 -07002285 ceph_put_snap_context(ci->i_head_snapc);
2286 ci->i_head_snapc = NULL;
2287 }
2288 dout("put_wrbuffer_cap_refs on %p head %d/%d -> %d/%d %s\n",
2289 inode,
2290 ci->i_wrbuffer_ref+nr, ci->i_wrbuffer_ref_head+nr,
2291 ci->i_wrbuffer_ref, ci->i_wrbuffer_ref_head,
2292 last ? " LAST" : "");
2293 } else {
2294 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
2295 if (capsnap->context == snapc) {
2296 found = 1;
Sage Weila8599bd2009-10-06 11:31:12 -07002297 break;
2298 }
2299 }
2300 BUG_ON(!found);
Sage Weil819ccbf2010-04-01 09:33:46 -07002301 capsnap->dirty_pages -= nr;
2302 if (capsnap->dirty_pages == 0) {
2303 complete_capsnap = 1;
2304 if (capsnap->dirty == 0)
2305 /* cap writeback completed before we created
2306 * the cap_snap; no FLUSHSNAP is needed */
2307 drop_capsnap = 1;
2308 }
Sage Weila8599bd2009-10-06 11:31:12 -07002309 dout("put_wrbuffer_cap_refs on %p cap_snap %p "
Sage Weil819ccbf2010-04-01 09:33:46 -07002310 " snap %lld %d/%d -> %d/%d %s%s%s\n",
Sage Weila8599bd2009-10-06 11:31:12 -07002311 inode, capsnap, capsnap->context->seq,
2312 ci->i_wrbuffer_ref+nr, capsnap->dirty_pages + nr,
2313 ci->i_wrbuffer_ref, capsnap->dirty_pages,
2314 last ? " (wrbuffer last)" : "",
Sage Weil819ccbf2010-04-01 09:33:46 -07002315 complete_capsnap ? " (complete capsnap)" : "",
2316 drop_capsnap ? " (drop capsnap)" : "");
2317 if (drop_capsnap) {
2318 ceph_put_snap_context(capsnap->context);
2319 list_del(&capsnap->ci_item);
2320 list_del(&capsnap->flushing_item);
2321 ceph_put_cap_snap(capsnap);
2322 }
Sage Weila8599bd2009-10-06 11:31:12 -07002323 }
2324
Sage Weilbe655592011-11-30 09:47:09 -08002325 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07002326
2327 if (last) {
2328 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
2329 iput(inode);
Sage Weil819ccbf2010-04-01 09:33:46 -07002330 } else if (complete_capsnap) {
Sage Weila8599bd2009-10-06 11:31:12 -07002331 ceph_flush_snaps(ci);
Yehuda Sadeh03066f22010-07-27 13:11:08 -07002332 wake_up_all(&ci->i_cap_wq);
Sage Weila8599bd2009-10-06 11:31:12 -07002333 }
Sage Weil819ccbf2010-04-01 09:33:46 -07002334 if (drop_capsnap)
2335 iput(inode);
Sage Weila8599bd2009-10-06 11:31:12 -07002336}
2337
2338/*
Yan, Zhengca20c992013-07-21 10:07:51 +08002339 * Invalidate unlinked inode's aliases, so we can drop the inode ASAP.
2340 */
2341static void invalidate_aliases(struct inode *inode)
2342{
2343 struct dentry *dn, *prev = NULL;
2344
2345 dout("invalidate_aliases inode %p\n", inode);
2346 d_prune_aliases(inode);
2347 /*
2348 * For non-directory inode, d_find_alias() only returns
Yan, Zhenga8d436f2013-09-02 15:19:54 +08002349 * connected dentry. After calling d_invalidate(), the
2350 * dentry become disconnected.
Yan, Zhengca20c992013-07-21 10:07:51 +08002351 *
Yan, Zhenga8d436f2013-09-02 15:19:54 +08002352 * For directory inode, d_find_alias() can return
Yan, Zhengca20c992013-07-21 10:07:51 +08002353 * disconnected dentry. But directory inode should have
2354 * one alias at most.
2355 */
2356 while ((dn = d_find_alias(inode))) {
2357 if (dn == prev) {
2358 dput(dn);
2359 break;
2360 }
Yan, Zhenga8d436f2013-09-02 15:19:54 +08002361 d_invalidate(dn);
Yan, Zhengca20c992013-07-21 10:07:51 +08002362 if (prev)
2363 dput(prev);
2364 prev = dn;
2365 }
2366 if (prev)
2367 dput(prev);
2368}
2369
2370/*
Sage Weila8599bd2009-10-06 11:31:12 -07002371 * Handle a cap GRANT message from the MDS. (Note that a GRANT may
2372 * actually be a revocation if it specifies a smaller cap set.)
2373 *
Sage Weilbe655592011-11-30 09:47:09 -08002374 * caller holds s_mutex and i_ceph_lock, we drop both.
Sage Weil15637c82010-03-16 13:42:00 -07002375 *
Sage Weila8599bd2009-10-06 11:31:12 -07002376 * return value:
2377 * 0 - ok
2378 * 1 - check_caps on auth cap only (writeback)
2379 * 2 - check_caps (ack revoke)
2380 */
Sage Weil15637c82010-03-16 13:42:00 -07002381static void handle_cap_grant(struct inode *inode, struct ceph_mds_caps *grant,
2382 struct ceph_mds_session *session,
2383 struct ceph_cap *cap,
2384 struct ceph_buffer *xattr_buf)
Sage Weilbe655592011-11-30 09:47:09 -08002385 __releases(ci->i_ceph_lock)
Sage Weila8599bd2009-10-06 11:31:12 -07002386{
2387 struct ceph_inode_info *ci = ceph_inode(inode);
2388 int mds = session->s_mds;
Sage Weil2f56f562010-10-27 20:59:49 -07002389 int seq = le32_to_cpu(grant->seq);
Sage Weila8599bd2009-10-06 11:31:12 -07002390 int newcaps = le32_to_cpu(grant->caps);
2391 int issued, implemented, used, wanted, dirty;
2392 u64 size = le64_to_cpu(grant->size);
2393 u64 max_size = le64_to_cpu(grant->max_size);
2394 struct timespec mtime, atime, ctime;
Sage Weil15637c82010-03-16 13:42:00 -07002395 int check_caps = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07002396 int wake = 0;
2397 int writeback = 0;
Sage Weil3c6f6b72010-02-09 15:24:44 -08002398 int queue_invalidate = 0;
Yan, Zhengca20c992013-07-21 10:07:51 +08002399 int deleted_inode = 0;
Milosz Tanski99ccbd22013-08-21 17:29:54 -04002400 int queue_revalidate = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07002401
Sage Weil2f56f562010-10-27 20:59:49 -07002402 dout("handle_cap_grant inode %p cap %p mds%d seq %d %s\n",
2403 inode, cap, mds, seq, ceph_cap_string(newcaps));
Sage Weila8599bd2009-10-06 11:31:12 -07002404 dout(" size %llu max_size %llu, i_size %llu\n", size, max_size,
2405 inode->i_size);
2406
2407 /*
2408 * If CACHE is being revoked, and we have no dirty buffers,
2409 * try to invalidate (once). (If there are dirty buffers, we
2410 * will invalidate _after_ writeback.)
2411 */
Sage Weil3b454c42010-06-10 13:20:33 -07002412 if (((cap->issued & ~newcaps) & CEPH_CAP_FILE_CACHE) &&
2413 (newcaps & CEPH_CAP_FILE_LAZYIO) == 0 &&
Yehuda Sadehbcd2cbd2010-02-19 00:12:21 +00002414 !ci->i_wrbuffer_ref) {
Li Wange9075742013-08-15 22:00:25 -07002415 if (try_nonblocking_invalidate(inode)) {
Sage Weila8599bd2009-10-06 11:31:12 -07002416 /* there were locked pages.. invalidate later
2417 in a separate thread. */
2418 if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
Sage Weil3c6f6b72010-02-09 15:24:44 -08002419 queue_invalidate = 1;
Sage Weila8599bd2009-10-06 11:31:12 -07002420 ci->i_rdcache_revoking = ci->i_rdcache_gen;
2421 }
Sage Weila8599bd2009-10-06 11:31:12 -07002422 }
Milosz Tanski99ccbd22013-08-21 17:29:54 -04002423
2424 ceph_fscache_invalidate(inode);
Sage Weila8599bd2009-10-06 11:31:12 -07002425 }
2426
2427 /* side effects now are allowed */
2428
2429 issued = __ceph_caps_issued(ci, &implemented);
2430 issued |= implemented | __ceph_caps_dirty(ci);
2431
Sage Weil685f9a5d2009-11-09 12:05:48 -08002432 cap->cap_gen = session->s_cap_gen;
Sage Weila8599bd2009-10-06 11:31:12 -07002433
2434 __check_cap_issue(ci, cap, newcaps);
2435
2436 if ((issued & CEPH_CAP_AUTH_EXCL) == 0) {
2437 inode->i_mode = le32_to_cpu(grant->mode);
Eric W. Biederman05cb11c2013-01-31 02:56:19 -08002438 inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(grant->uid));
2439 inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(grant->gid));
Sage Weila8599bd2009-10-06 11:31:12 -07002440 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
Eric W. Biedermanbd2bae62013-01-31 04:05:39 -08002441 from_kuid(&init_user_ns, inode->i_uid),
2442 from_kgid(&init_user_ns, inode->i_gid));
Sage Weila8599bd2009-10-06 11:31:12 -07002443 }
2444
Yan, Zhengca20c992013-07-21 10:07:51 +08002445 if ((issued & CEPH_CAP_LINK_EXCL) == 0) {
Miklos Szeredibfe86842011-10-28 14:13:29 +02002446 set_nlink(inode, le32_to_cpu(grant->nlink));
Yan, Zhengca20c992013-07-21 10:07:51 +08002447 if (inode->i_nlink == 0 &&
2448 (newcaps & (CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL)))
2449 deleted_inode = 1;
2450 }
Sage Weila8599bd2009-10-06 11:31:12 -07002451
2452 if ((issued & CEPH_CAP_XATTR_EXCL) == 0 && grant->xattr_len) {
2453 int len = le32_to_cpu(grant->xattr_len);
2454 u64 version = le64_to_cpu(grant->xattr_version);
2455
2456 if (version > ci->i_xattrs.version) {
2457 dout(" got new xattrs v%llu on %p len %d\n",
2458 version, inode, len);
2459 if (ci->i_xattrs.blob)
2460 ceph_buffer_put(ci->i_xattrs.blob);
2461 ci->i_xattrs.blob = ceph_buffer_get(xattr_buf);
2462 ci->i_xattrs.version = version;
2463 }
2464 }
2465
Milosz Tanski99ccbd22013-08-21 17:29:54 -04002466 /* Do we need to revalidate our fscache cookie. Don't bother on the
2467 * first cache cap as we already validate at cookie creation time. */
2468 if ((issued & CEPH_CAP_FILE_CACHE) && ci->i_rdcache_gen > 1)
2469 queue_revalidate = 1;
2470
Sage Weila8599bd2009-10-06 11:31:12 -07002471 /* size/ctime/mtime/atime? */
2472 ceph_fill_file_size(inode, issued,
2473 le32_to_cpu(grant->truncate_seq),
2474 le64_to_cpu(grant->truncate_size), size);
2475 ceph_decode_timespec(&mtime, &grant->mtime);
2476 ceph_decode_timespec(&atime, &grant->atime);
2477 ceph_decode_timespec(&ctime, &grant->ctime);
2478 ceph_fill_file_time(inode, issued,
2479 le32_to_cpu(grant->time_warp_seq), &ctime, &mtime,
2480 &atime);
2481
2482 /* max size increase? */
Yan, Zheng5e62ad32012-11-19 10:49:04 +08002483 if (ci->i_auth_cap == cap && max_size != ci->i_max_size) {
Sage Weila8599bd2009-10-06 11:31:12 -07002484 dout("max_size %lld -> %llu\n", ci->i_max_size, max_size);
2485 ci->i_max_size = max_size;
2486 if (max_size >= ci->i_wanted_max_size) {
2487 ci->i_wanted_max_size = 0; /* reset */
2488 ci->i_requested_max_size = 0;
2489 }
2490 wake = 1;
2491 }
2492
2493 /* check cap bits */
2494 wanted = __ceph_caps_wanted(ci);
2495 used = __ceph_caps_used(ci);
2496 dirty = __ceph_caps_dirty(ci);
2497 dout(" my wanted = %s, used = %s, dirty %s\n",
2498 ceph_cap_string(wanted),
2499 ceph_cap_string(used),
2500 ceph_cap_string(dirty));
2501 if (wanted != le32_to_cpu(grant->wanted)) {
2502 dout("mds wanted %s -> %s\n",
2503 ceph_cap_string(le32_to_cpu(grant->wanted)),
2504 ceph_cap_string(wanted));
Yan, Zheng390306c2013-01-04 15:30:10 +08002505 /* imported cap may not have correct mds_wanted */
2506 if (le32_to_cpu(grant->op) == CEPH_CAP_OP_IMPORT)
2507 check_caps = 1;
Sage Weila8599bd2009-10-06 11:31:12 -07002508 }
2509
2510 cap->seq = seq;
2511
2512 /* file layout may have changed */
2513 ci->i_layout = grant->layout;
2514
2515 /* revocation, grant, or no-op? */
2516 if (cap->issued & ~newcaps) {
Sage Weil3b454c42010-06-10 13:20:33 -07002517 int revoking = cap->issued & ~newcaps;
2518
2519 dout("revocation: %s -> %s (revoking %s)\n",
2520 ceph_cap_string(cap->issued),
2521 ceph_cap_string(newcaps),
2522 ceph_cap_string(revoking));
Sage Weil0eb6cd42010-08-05 13:53:18 -07002523 if (revoking & used & CEPH_CAP_FILE_BUFFER)
Sage Weil3b454c42010-06-10 13:20:33 -07002524 writeback = 1; /* initiate writeback; will delay ack */
2525 else if (revoking == CEPH_CAP_FILE_CACHE &&
2526 (newcaps & CEPH_CAP_FILE_LAZYIO) == 0 &&
2527 queue_invalidate)
2528 ; /* do nothing yet, invalidation will be queued */
2529 else if (cap == ci->i_auth_cap)
2530 check_caps = 1; /* check auth cap only */
2531 else
2532 check_caps = 2; /* check all caps */
Sage Weila8599bd2009-10-06 11:31:12 -07002533 cap->issued = newcaps;
Sage Weil978097c2010-03-08 15:27:53 -08002534 cap->implemented |= newcaps;
Sage Weila8599bd2009-10-06 11:31:12 -07002535 } else if (cap->issued == newcaps) {
2536 dout("caps unchanged: %s -> %s\n",
2537 ceph_cap_string(cap->issued), ceph_cap_string(newcaps));
2538 } else {
2539 dout("grant: %s -> %s\n", ceph_cap_string(cap->issued),
2540 ceph_cap_string(newcaps));
Yan, Zheng6ee6b9532013-07-02 12:40:21 +08002541 /* non-auth MDS is revoking the newly grant caps ? */
2542 if (cap == ci->i_auth_cap &&
2543 __ceph_caps_revoking_other(ci, cap, newcaps))
2544 check_caps = 2;
2545
Sage Weila8599bd2009-10-06 11:31:12 -07002546 cap->issued = newcaps;
2547 cap->implemented |= newcaps; /* add bits only, to
2548 * avoid stepping on a
2549 * pending revocation */
2550 wake = 1;
2551 }
Sage Weil978097c2010-03-08 15:27:53 -08002552 BUG_ON(cap->issued & ~cap->implemented);
Sage Weila8599bd2009-10-06 11:31:12 -07002553
Sage Weilbe655592011-11-30 09:47:09 -08002554 spin_unlock(&ci->i_ceph_lock);
Milosz Tanski99ccbd22013-08-21 17:29:54 -04002555
Sage Weil3c6f6b72010-02-09 15:24:44 -08002556 if (writeback)
Sage Weila8599bd2009-10-06 11:31:12 -07002557 /*
2558 * queue inode for writeback: we can't actually call
2559 * filemap_write_and_wait, etc. from message handler
2560 * context.
2561 */
Sage Weil3c6f6b72010-02-09 15:24:44 -08002562 ceph_queue_writeback(inode);
2563 if (queue_invalidate)
2564 ceph_queue_invalidate(inode);
Yan, Zhengca20c992013-07-21 10:07:51 +08002565 if (deleted_inode)
2566 invalidate_aliases(inode);
Milosz Tanski99ccbd22013-08-21 17:29:54 -04002567 if (queue_revalidate)
2568 ceph_queue_revalidate(inode);
Sage Weila8599bd2009-10-06 11:31:12 -07002569 if (wake)
Yehuda Sadeh03066f22010-07-27 13:11:08 -07002570 wake_up_all(&ci->i_cap_wq);
Sage Weil15637c82010-03-16 13:42:00 -07002571
2572 if (check_caps == 1)
2573 ceph_check_caps(ci, CHECK_CAPS_NODELAY|CHECK_CAPS_AUTHONLY,
2574 session);
2575 else if (check_caps == 2)
2576 ceph_check_caps(ci, CHECK_CAPS_NODELAY, session);
2577 else
2578 mutex_unlock(&session->s_mutex);
Sage Weila8599bd2009-10-06 11:31:12 -07002579}
2580
2581/*
2582 * Handle FLUSH_ACK from MDS, indicating that metadata we sent to the
2583 * MDS has been safely committed.
2584 */
Sage Weil6df058c2009-12-22 11:24:33 -08002585static void handle_cap_flush_ack(struct inode *inode, u64 flush_tid,
Sage Weila8599bd2009-10-06 11:31:12 -07002586 struct ceph_mds_caps *m,
2587 struct ceph_mds_session *session,
2588 struct ceph_cap *cap)
Sage Weilbe655592011-11-30 09:47:09 -08002589 __releases(ci->i_ceph_lock)
Sage Weila8599bd2009-10-06 11:31:12 -07002590{
2591 struct ceph_inode_info *ci = ceph_inode(inode);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002592 struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
Sage Weila8599bd2009-10-06 11:31:12 -07002593 unsigned seq = le32_to_cpu(m->seq);
2594 int dirty = le32_to_cpu(m->dirty);
2595 int cleaned = 0;
Sage Weilafcdaea2009-10-14 14:27:38 -07002596 int drop = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07002597 int i;
2598
2599 for (i = 0; i < CEPH_CAP_BITS; i++)
2600 if ((dirty & (1 << i)) &&
2601 flush_tid == ci->i_cap_flush_tid[i])
2602 cleaned |= 1 << i;
2603
2604 dout("handle_cap_flush_ack inode %p mds%d seq %d on %s cleaned %s,"
2605 " flushing %s -> %s\n",
2606 inode, session->s_mds, seq, ceph_cap_string(dirty),
2607 ceph_cap_string(cleaned), ceph_cap_string(ci->i_flushing_caps),
2608 ceph_cap_string(ci->i_flushing_caps & ~cleaned));
2609
2610 if (ci->i_flushing_caps == (ci->i_flushing_caps & ~cleaned))
2611 goto out;
2612
Sage Weila8599bd2009-10-06 11:31:12 -07002613 ci->i_flushing_caps &= ~cleaned;
Sage Weila8599bd2009-10-06 11:31:12 -07002614
2615 spin_lock(&mdsc->cap_dirty_lock);
2616 if (ci->i_flushing_caps == 0) {
2617 list_del_init(&ci->i_flushing_item);
2618 if (!list_empty(&session->s_cap_flushing))
2619 dout(" mds%d still flushing cap on %p\n",
2620 session->s_mds,
2621 &list_entry(session->s_cap_flushing.next,
2622 struct ceph_inode_info,
2623 i_flushing_item)->vfs_inode);
2624 mdsc->num_cap_flushing--;
Yehuda Sadeh03066f22010-07-27 13:11:08 -07002625 wake_up_all(&mdsc->cap_flushing_wq);
Sage Weila8599bd2009-10-06 11:31:12 -07002626 dout(" inode %p now !flushing\n", inode);
Sage Weilafcdaea2009-10-14 14:27:38 -07002627
2628 if (ci->i_dirty_caps == 0) {
2629 dout(" inode %p now clean\n", inode);
2630 BUG_ON(!list_empty(&ci->i_dirty_item));
2631 drop = 1;
Sage Weil7d8cb262010-08-24 08:44:16 -07002632 if (ci->i_wrbuffer_ref_head == 0) {
2633 BUG_ON(!ci->i_head_snapc);
2634 ceph_put_snap_context(ci->i_head_snapc);
2635 ci->i_head_snapc = NULL;
2636 }
Sage Weil76e3b392009-10-15 18:13:53 -07002637 } else {
2638 BUG_ON(list_empty(&ci->i_dirty_item));
Sage Weilafcdaea2009-10-14 14:27:38 -07002639 }
Sage Weila8599bd2009-10-06 11:31:12 -07002640 }
2641 spin_unlock(&mdsc->cap_dirty_lock);
Yehuda Sadeh03066f22010-07-27 13:11:08 -07002642 wake_up_all(&ci->i_cap_wq);
Sage Weila8599bd2009-10-06 11:31:12 -07002643
2644out:
Sage Weilbe655592011-11-30 09:47:09 -08002645 spin_unlock(&ci->i_ceph_lock);
Sage Weilafcdaea2009-10-14 14:27:38 -07002646 if (drop)
Sage Weila8599bd2009-10-06 11:31:12 -07002647 iput(inode);
2648}
2649
2650/*
2651 * Handle FLUSHSNAP_ACK. MDS has flushed snap data to disk and we can
2652 * throw away our cap_snap.
2653 *
2654 * Caller hold s_mutex.
2655 */
Sage Weil6df058c2009-12-22 11:24:33 -08002656static void handle_cap_flushsnap_ack(struct inode *inode, u64 flush_tid,
Sage Weila8599bd2009-10-06 11:31:12 -07002657 struct ceph_mds_caps *m,
2658 struct ceph_mds_session *session)
2659{
2660 struct ceph_inode_info *ci = ceph_inode(inode);
2661 u64 follows = le64_to_cpu(m->snap_follows);
Sage Weila8599bd2009-10-06 11:31:12 -07002662 struct ceph_cap_snap *capsnap;
2663 int drop = 0;
2664
2665 dout("handle_cap_flushsnap_ack inode %p ci %p mds%d follows %lld\n",
2666 inode, ci, session->s_mds, follows);
2667
Sage Weilbe655592011-11-30 09:47:09 -08002668 spin_lock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07002669 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
2670 if (capsnap->follows == follows) {
2671 if (capsnap->flush_tid != flush_tid) {
2672 dout(" cap_snap %p follows %lld tid %lld !="
2673 " %lld\n", capsnap, follows,
2674 flush_tid, capsnap->flush_tid);
2675 break;
2676 }
2677 WARN_ON(capsnap->dirty_pages || capsnap->writing);
Sage Weil819ccbf2010-04-01 09:33:46 -07002678 dout(" removing %p cap_snap %p follows %lld\n",
2679 inode, capsnap, follows);
Sage Weila8599bd2009-10-06 11:31:12 -07002680 ceph_put_snap_context(capsnap->context);
2681 list_del(&capsnap->ci_item);
2682 list_del(&capsnap->flushing_item);
2683 ceph_put_cap_snap(capsnap);
2684 drop = 1;
2685 break;
2686 } else {
2687 dout(" skipping cap_snap %p follows %lld\n",
2688 capsnap, capsnap->follows);
2689 }
2690 }
Sage Weilbe655592011-11-30 09:47:09 -08002691 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07002692 if (drop)
2693 iput(inode);
2694}
2695
2696/*
2697 * Handle TRUNC from MDS, indicating file truncation.
2698 *
2699 * caller hold s_mutex.
2700 */
2701static void handle_cap_trunc(struct inode *inode,
2702 struct ceph_mds_caps *trunc,
2703 struct ceph_mds_session *session)
Sage Weilbe655592011-11-30 09:47:09 -08002704 __releases(ci->i_ceph_lock)
Sage Weila8599bd2009-10-06 11:31:12 -07002705{
2706 struct ceph_inode_info *ci = ceph_inode(inode);
2707 int mds = session->s_mds;
2708 int seq = le32_to_cpu(trunc->seq);
2709 u32 truncate_seq = le32_to_cpu(trunc->truncate_seq);
2710 u64 truncate_size = le64_to_cpu(trunc->truncate_size);
2711 u64 size = le64_to_cpu(trunc->size);
2712 int implemented = 0;
2713 int dirty = __ceph_caps_dirty(ci);
2714 int issued = __ceph_caps_issued(ceph_inode(inode), &implemented);
2715 int queue_trunc = 0;
2716
2717 issued |= implemented | dirty;
2718
2719 dout("handle_cap_trunc inode %p mds%d seq %d to %lld seq %d\n",
2720 inode, mds, seq, truncate_size, truncate_seq);
2721 queue_trunc = ceph_fill_file_size(inode, issued,
2722 truncate_seq, truncate_size, size);
Sage Weilbe655592011-11-30 09:47:09 -08002723 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07002724
Milosz Tanski99ccbd22013-08-21 17:29:54 -04002725 if (queue_trunc) {
Sage Weil3c6f6b72010-02-09 15:24:44 -08002726 ceph_queue_vmtruncate(inode);
Milosz Tanski99ccbd22013-08-21 17:29:54 -04002727 ceph_fscache_invalidate(inode);
2728 }
Sage Weila8599bd2009-10-06 11:31:12 -07002729}
2730
2731/*
2732 * Handle EXPORT from MDS. Cap is being migrated _from_ this mds to a
2733 * different one. If we are the most recent migration we've seen (as
2734 * indicated by mseq), make note of the migrating cap bits for the
2735 * duration (until we see the corresponding IMPORT).
2736 *
2737 * caller holds s_mutex
2738 */
2739static void handle_cap_export(struct inode *inode, struct ceph_mds_caps *ex,
Sage Weil154f42c2010-06-21 13:45:04 -07002740 struct ceph_mds_session *session,
2741 int *open_target_sessions)
Sage Weila8599bd2009-10-06 11:31:12 -07002742{
Sage Weildb354052011-05-24 11:46:31 -07002743 struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
Sage Weila8599bd2009-10-06 11:31:12 -07002744 struct ceph_inode_info *ci = ceph_inode(inode);
2745 int mds = session->s_mds;
2746 unsigned mseq = le32_to_cpu(ex->migrate_seq);
2747 struct ceph_cap *cap = NULL, *t;
2748 struct rb_node *p;
2749 int remember = 1;
2750
2751 dout("handle_cap_export inode %p ci %p mds%d mseq %d\n",
2752 inode, ci, mds, mseq);
2753
Sage Weilbe655592011-11-30 09:47:09 -08002754 spin_lock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07002755
2756 /* make sure we haven't seen a higher mseq */
2757 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
2758 t = rb_entry(p, struct ceph_cap, ci_node);
2759 if (ceph_seq_cmp(t->mseq, mseq) > 0) {
2760 dout(" higher mseq on cap from mds%d\n",
2761 t->session->s_mds);
2762 remember = 0;
2763 }
2764 if (t->session->s_mds == mds)
2765 cap = t;
2766 }
2767
2768 if (cap) {
2769 if (remember) {
2770 /* make note */
2771 ci->i_cap_exporting_mds = mds;
2772 ci->i_cap_exporting_mseq = mseq;
2773 ci->i_cap_exporting_issued = cap->issued;
Sage Weil154f42c2010-06-21 13:45:04 -07002774
2775 /*
2776 * make sure we have open sessions with all possible
2777 * export targets, so that we get the matching IMPORT
2778 */
2779 *open_target_sessions = 1;
Sage Weildb354052011-05-24 11:46:31 -07002780
2781 /*
2782 * we can't flush dirty caps that we've seen the
2783 * EXPORT but no IMPORT for
2784 */
2785 spin_lock(&mdsc->cap_dirty_lock);
2786 if (!list_empty(&ci->i_dirty_item)) {
2787 dout(" moving %p to cap_dirty_migrating\n",
2788 inode);
2789 list_move(&ci->i_dirty_item,
2790 &mdsc->cap_dirty_migrating);
2791 }
2792 spin_unlock(&mdsc->cap_dirty_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07002793 }
Sage Weil7c1332b2010-02-16 11:39:45 -08002794 __ceph_remove_cap(cap);
Sage Weila8599bd2009-10-06 11:31:12 -07002795 }
Sage Weil4ea00432010-03-16 10:36:40 -07002796 /* else, we already released it */
Sage Weila8599bd2009-10-06 11:31:12 -07002797
Sage Weilbe655592011-11-30 09:47:09 -08002798 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07002799}
2800
2801/*
2802 * Handle cap IMPORT. If there are temp bits from an older EXPORT,
2803 * clean them up.
2804 *
2805 * caller holds s_mutex.
2806 */
2807static void handle_cap_import(struct ceph_mds_client *mdsc,
2808 struct inode *inode, struct ceph_mds_caps *im,
2809 struct ceph_mds_session *session,
2810 void *snaptrace, int snaptrace_len)
2811{
2812 struct ceph_inode_info *ci = ceph_inode(inode);
2813 int mds = session->s_mds;
2814 unsigned issued = le32_to_cpu(im->caps);
2815 unsigned wanted = le32_to_cpu(im->wanted);
2816 unsigned seq = le32_to_cpu(im->seq);
2817 unsigned mseq = le32_to_cpu(im->migrate_seq);
2818 u64 realmino = le64_to_cpu(im->realm);
2819 u64 cap_id = le64_to_cpu(im->cap_id);
2820
2821 if (ci->i_cap_exporting_mds >= 0 &&
2822 ceph_seq_cmp(ci->i_cap_exporting_mseq, mseq) < 0) {
2823 dout("handle_cap_import inode %p ci %p mds%d mseq %d"
2824 " - cleared exporting from mds%d\n",
2825 inode, ci, mds, mseq,
2826 ci->i_cap_exporting_mds);
2827 ci->i_cap_exporting_issued = 0;
2828 ci->i_cap_exporting_mseq = 0;
2829 ci->i_cap_exporting_mds = -1;
Sage Weildb354052011-05-24 11:46:31 -07002830
2831 spin_lock(&mdsc->cap_dirty_lock);
2832 if (!list_empty(&ci->i_dirty_item)) {
2833 dout(" moving %p back to cap_dirty\n", inode);
2834 list_move(&ci->i_dirty_item, &mdsc->cap_dirty);
2835 }
2836 spin_unlock(&mdsc->cap_dirty_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07002837 } else {
2838 dout("handle_cap_import inode %p ci %p mds%d mseq %d\n",
2839 inode, ci, mds, mseq);
2840 }
2841
2842 down_write(&mdsc->snap_rwsem);
2843 ceph_update_snap_trace(mdsc, snaptrace, snaptrace+snaptrace_len,
2844 false);
2845 downgrade_write(&mdsc->snap_rwsem);
2846 ceph_add_cap(inode, session, cap_id, -1,
2847 issued, wanted, seq, mseq, realmino, CEPH_CAP_FLAG_AUTH,
2848 NULL /* no caps context */);
Sage Weil088b3f52011-01-18 08:56:01 -08002849 kick_flushing_inode_caps(mdsc, session, inode);
Sage Weila8599bd2009-10-06 11:31:12 -07002850 up_read(&mdsc->snap_rwsem);
Sage Weilfeb4cc92010-11-07 09:39:00 -08002851
2852 /* make sure we re-request max_size, if necessary */
Sage Weilbe655592011-11-30 09:47:09 -08002853 spin_lock(&ci->i_ceph_lock);
Yan, Zheng0e5e1772012-11-19 10:49:09 +08002854 ci->i_wanted_max_size = 0; /* reset */
Sage Weilfeb4cc92010-11-07 09:39:00 -08002855 ci->i_requested_max_size = 0;
Sage Weilbe655592011-11-30 09:47:09 -08002856 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07002857}
2858
2859/*
2860 * Handle a caps message from the MDS.
2861 *
2862 * Identify the appropriate session, inode, and call the right handler
2863 * based on the cap op.
2864 */
2865void ceph_handle_caps(struct ceph_mds_session *session,
2866 struct ceph_msg *msg)
2867{
2868 struct ceph_mds_client *mdsc = session->s_mdsc;
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002869 struct super_block *sb = mdsc->fsc->sb;
Sage Weila8599bd2009-10-06 11:31:12 -07002870 struct inode *inode;
Sage Weilbe655592011-11-30 09:47:09 -08002871 struct ceph_inode_info *ci;
Sage Weila8599bd2009-10-06 11:31:12 -07002872 struct ceph_cap *cap;
2873 struct ceph_mds_caps *h;
Sage Weil2600d2d2010-02-22 15:12:16 -08002874 int mds = session->s_mds;
Sage Weila8599bd2009-10-06 11:31:12 -07002875 int op;
Sage Weil3d7ded42010-06-09 16:47:10 -07002876 u32 seq, mseq;
Sage Weila8599bd2009-10-06 11:31:12 -07002877 struct ceph_vino vino;
2878 u64 cap_id;
2879 u64 size, max_size;
Sage Weil6df058c2009-12-22 11:24:33 -08002880 u64 tid;
Sage Weil70edb552010-03-01 13:20:50 -08002881 void *snaptrace;
Sage Weilce1fbc82010-08-02 15:09:39 -07002882 size_t snaptrace_len;
2883 void *flock;
2884 u32 flock_len;
Sage Weil154f42c2010-06-21 13:45:04 -07002885 int open_target_sessions = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07002886
2887 dout("handle_caps from mds%d\n", mds);
2888
2889 /* decode */
Sage Weil6df058c2009-12-22 11:24:33 -08002890 tid = le64_to_cpu(msg->hdr.tid);
Sage Weila8599bd2009-10-06 11:31:12 -07002891 if (msg->front.iov_len < sizeof(*h))
2892 goto bad;
2893 h = msg->front.iov_base;
2894 op = le32_to_cpu(h->op);
2895 vino.ino = le64_to_cpu(h->ino);
2896 vino.snap = CEPH_NOSNAP;
2897 cap_id = le64_to_cpu(h->cap_id);
2898 seq = le32_to_cpu(h->seq);
Sage Weil3d7ded42010-06-09 16:47:10 -07002899 mseq = le32_to_cpu(h->migrate_seq);
Sage Weila8599bd2009-10-06 11:31:12 -07002900 size = le64_to_cpu(h->size);
2901 max_size = le64_to_cpu(h->max_size);
2902
Sage Weilce1fbc82010-08-02 15:09:39 -07002903 snaptrace = h + 1;
2904 snaptrace_len = le32_to_cpu(h->snap_trace_len);
2905
2906 if (le16_to_cpu(msg->hdr.version) >= 2) {
2907 void *p, *end;
2908
2909 p = snaptrace + snaptrace_len;
2910 end = msg->front.iov_base + msg->front.iov_len;
2911 ceph_decode_32_safe(&p, end, flock_len, bad);
2912 flock = p;
2913 } else {
2914 flock = NULL;
2915 flock_len = 0;
2916 }
2917
Sage Weila8599bd2009-10-06 11:31:12 -07002918 mutex_lock(&session->s_mutex);
2919 session->s_seq++;
2920 dout(" mds%d seq %lld cap seq %u\n", session->s_mds, session->s_seq,
2921 (unsigned)seq);
2922
Yan, Zheng66f58692013-01-04 14:45:18 +08002923 if (op == CEPH_CAP_OP_IMPORT)
2924 ceph_add_cap_releases(mdsc, session);
2925
Sage Weila8599bd2009-10-06 11:31:12 -07002926 /* lookup ino */
2927 inode = ceph_find_inode(sb, vino);
Sage Weilbe655592011-11-30 09:47:09 -08002928 ci = ceph_inode(inode);
Sage Weila8599bd2009-10-06 11:31:12 -07002929 dout(" op %s ino %llx.%llx inode %p\n", ceph_cap_op_name(op), vino.ino,
2930 vino.snap, inode);
2931 if (!inode) {
2932 dout(" i don't have ino %llx\n", vino.ino);
Sage Weil3d7ded42010-06-09 16:47:10 -07002933
2934 if (op == CEPH_CAP_OP_IMPORT)
2935 __queue_cap_release(session, vino.ino, cap_id,
2936 mseq, seq);
Greg Farnum21b559d2010-10-06 15:46:30 -07002937 goto flush_cap_releases;
Sage Weila8599bd2009-10-06 11:31:12 -07002938 }
2939
2940 /* these will work even if we don't have a cap yet */
2941 switch (op) {
2942 case CEPH_CAP_OP_FLUSHSNAP_ACK:
Sage Weil6df058c2009-12-22 11:24:33 -08002943 handle_cap_flushsnap_ack(inode, tid, h, session);
Sage Weila8599bd2009-10-06 11:31:12 -07002944 goto done;
2945
2946 case CEPH_CAP_OP_EXPORT:
Sage Weil154f42c2010-06-21 13:45:04 -07002947 handle_cap_export(inode, h, session, &open_target_sessions);
Sage Weila8599bd2009-10-06 11:31:12 -07002948 goto done;
2949
2950 case CEPH_CAP_OP_IMPORT:
2951 handle_cap_import(mdsc, inode, h, session,
Sage Weilce1fbc82010-08-02 15:09:39 -07002952 snaptrace, snaptrace_len);
Sage Weila8599bd2009-10-06 11:31:12 -07002953 }
2954
2955 /* the rest require a cap */
Sage Weilbe655592011-11-30 09:47:09 -08002956 spin_lock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07002957 cap = __get_cap_for_mds(ceph_inode(inode), mds);
2958 if (!cap) {
Sage Weil9dbd4122010-06-10 13:21:20 -07002959 dout(" no cap on %p ino %llx.%llx from mds%d\n",
Sage Weila8599bd2009-10-06 11:31:12 -07002960 inode, ceph_ino(inode), ceph_snap(inode), mds);
Sage Weilbe655592011-11-30 09:47:09 -08002961 spin_unlock(&ci->i_ceph_lock);
Greg Farnum21b559d2010-10-06 15:46:30 -07002962 goto flush_cap_releases;
Sage Weila8599bd2009-10-06 11:31:12 -07002963 }
2964
Sage Weilbe655592011-11-30 09:47:09 -08002965 /* note that each of these drops i_ceph_lock for us */
Sage Weila8599bd2009-10-06 11:31:12 -07002966 switch (op) {
2967 case CEPH_CAP_OP_REVOKE:
2968 case CEPH_CAP_OP_GRANT:
Yan, Zheng0e5e1772012-11-19 10:49:09 +08002969 case CEPH_CAP_OP_IMPORT:
Sage Weil15637c82010-03-16 13:42:00 -07002970 handle_cap_grant(inode, h, session, cap, msg->middle);
2971 goto done_unlocked;
Sage Weila8599bd2009-10-06 11:31:12 -07002972
2973 case CEPH_CAP_OP_FLUSH_ACK:
Sage Weil6df058c2009-12-22 11:24:33 -08002974 handle_cap_flush_ack(inode, tid, h, session, cap);
Sage Weila8599bd2009-10-06 11:31:12 -07002975 break;
2976
2977 case CEPH_CAP_OP_TRUNC:
2978 handle_cap_trunc(inode, h, session);
2979 break;
2980
2981 default:
Sage Weilbe655592011-11-30 09:47:09 -08002982 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07002983 pr_err("ceph_handle_caps: unknown cap op %d %s\n", op,
2984 ceph_cap_op_name(op));
2985 }
2986
Greg Farnum21b559d2010-10-06 15:46:30 -07002987 goto done;
2988
2989flush_cap_releases:
2990 /*
2991 * send any full release message to try to move things
2992 * along for the mds (who clearly thinks we still have this
2993 * cap).
2994 */
2995 ceph_add_cap_releases(mdsc, session);
2996 ceph_send_cap_releases(mdsc, session);
2997
Sage Weila8599bd2009-10-06 11:31:12 -07002998done:
Sage Weil15637c82010-03-16 13:42:00 -07002999 mutex_unlock(&session->s_mutex);
3000done_unlocked:
Sage Weila8599bd2009-10-06 11:31:12 -07003001 if (inode)
3002 iput(inode);
Sage Weil154f42c2010-06-21 13:45:04 -07003003 if (open_target_sessions)
3004 ceph_mdsc_open_export_target_sessions(mdsc, session);
Sage Weila8599bd2009-10-06 11:31:12 -07003005 return;
3006
3007bad:
3008 pr_err("ceph_handle_caps: corrupt message\n");
Sage Weil9ec7cab2009-12-14 15:13:47 -08003009 ceph_msg_dump(msg);
Sage Weila8599bd2009-10-06 11:31:12 -07003010 return;
3011}
3012
3013/*
3014 * Delayed work handler to process end of delayed cap release LRU list.
3015 */
Sage Weilafcdaea2009-10-14 14:27:38 -07003016void ceph_check_delayed_caps(struct ceph_mds_client *mdsc)
Sage Weila8599bd2009-10-06 11:31:12 -07003017{
3018 struct ceph_inode_info *ci;
3019 int flags = CHECK_CAPS_NODELAY;
3020
Sage Weila8599bd2009-10-06 11:31:12 -07003021 dout("check_delayed_caps\n");
3022 while (1) {
3023 spin_lock(&mdsc->cap_delay_lock);
3024 if (list_empty(&mdsc->cap_delay_list))
3025 break;
3026 ci = list_first_entry(&mdsc->cap_delay_list,
3027 struct ceph_inode_info,
3028 i_cap_delay_list);
3029 if ((ci->i_ceph_flags & CEPH_I_FLUSH) == 0 &&
3030 time_before(jiffies, ci->i_hold_caps_max))
3031 break;
3032 list_del_init(&ci->i_cap_delay_list);
3033 spin_unlock(&mdsc->cap_delay_lock);
3034 dout("check_delayed_caps on %p\n", &ci->vfs_inode);
3035 ceph_check_caps(ci, flags, NULL);
3036 }
3037 spin_unlock(&mdsc->cap_delay_lock);
3038}
3039
3040/*
Sage Weilafcdaea2009-10-14 14:27:38 -07003041 * Flush all dirty caps to the mds
3042 */
3043void ceph_flush_dirty_caps(struct ceph_mds_client *mdsc)
3044{
Sage Weildb354052011-05-24 11:46:31 -07003045 struct ceph_inode_info *ci;
3046 struct inode *inode;
Sage Weilafcdaea2009-10-14 14:27:38 -07003047
3048 dout("flush_dirty_caps\n");
3049 spin_lock(&mdsc->cap_dirty_lock);
Sage Weildb354052011-05-24 11:46:31 -07003050 while (!list_empty(&mdsc->cap_dirty)) {
3051 ci = list_first_entry(&mdsc->cap_dirty, struct ceph_inode_info,
3052 i_dirty_item);
Sage Weil70b666c2011-05-27 09:24:26 -07003053 inode = &ci->vfs_inode;
3054 ihold(inode);
Sage Weildb354052011-05-24 11:46:31 -07003055 dout("flush_dirty_caps %p\n", inode);
Sage Weilafcdaea2009-10-14 14:27:38 -07003056 spin_unlock(&mdsc->cap_dirty_lock);
Sage Weil70b666c2011-05-27 09:24:26 -07003057 ceph_check_caps(ci, CHECK_CAPS_NODELAY|CHECK_CAPS_FLUSH, NULL);
3058 iput(inode);
Sage Weilafcdaea2009-10-14 14:27:38 -07003059 spin_lock(&mdsc->cap_dirty_lock);
3060 }
3061 spin_unlock(&mdsc->cap_dirty_lock);
Sage Weildb354052011-05-24 11:46:31 -07003062 dout("flush_dirty_caps done\n");
Sage Weilafcdaea2009-10-14 14:27:38 -07003063}
3064
3065/*
Sage Weila8599bd2009-10-06 11:31:12 -07003066 * Drop open file reference. If we were the last open file,
3067 * we may need to release capabilities to the MDS (or schedule
3068 * their delayed release).
3069 */
3070void ceph_put_fmode(struct ceph_inode_info *ci, int fmode)
3071{
3072 struct inode *inode = &ci->vfs_inode;
3073 int last = 0;
3074
Sage Weilbe655592011-11-30 09:47:09 -08003075 spin_lock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07003076 dout("put_fmode %p fmode %d %d -> %d\n", inode, fmode,
3077 ci->i_nr_by_mode[fmode], ci->i_nr_by_mode[fmode]-1);
3078 BUG_ON(ci->i_nr_by_mode[fmode] == 0);
3079 if (--ci->i_nr_by_mode[fmode] == 0)
3080 last++;
Sage Weilbe655592011-11-30 09:47:09 -08003081 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07003082
3083 if (last && ci->i_vino.snap == CEPH_NOSNAP)
3084 ceph_check_caps(ci, 0, NULL);
3085}
3086
3087/*
3088 * Helpers for embedding cap and dentry lease releases into mds
3089 * requests.
3090 *
3091 * @force is used by dentry_release (below) to force inclusion of a
3092 * record for the directory inode, even when there aren't any caps to
3093 * drop.
3094 */
3095int ceph_encode_inode_release(void **p, struct inode *inode,
3096 int mds, int drop, int unless, int force)
3097{
3098 struct ceph_inode_info *ci = ceph_inode(inode);
3099 struct ceph_cap *cap;
3100 struct ceph_mds_request_release *rel = *p;
Sage Weilec97f882010-06-24 15:12:37 -07003101 int used, dirty;
Sage Weila8599bd2009-10-06 11:31:12 -07003102 int ret = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07003103
Sage Weilbe655592011-11-30 09:47:09 -08003104 spin_lock(&ci->i_ceph_lock);
Sage Weil916623d2010-03-16 15:01:07 -07003105 used = __ceph_caps_used(ci);
Sage Weilec97f882010-06-24 15:12:37 -07003106 dirty = __ceph_caps_dirty(ci);
Sage Weil916623d2010-03-16 15:01:07 -07003107
Sage Weilec97f882010-06-24 15:12:37 -07003108 dout("encode_inode_release %p mds%d used|dirty %s drop %s unless %s\n",
3109 inode, mds, ceph_cap_string(used|dirty), ceph_cap_string(drop),
Sage Weil916623d2010-03-16 15:01:07 -07003110 ceph_cap_string(unless));
3111
Sage Weilec97f882010-06-24 15:12:37 -07003112 /* only drop unused, clean caps */
3113 drop &= ~(used | dirty);
Sage Weil916623d2010-03-16 15:01:07 -07003114
Sage Weila8599bd2009-10-06 11:31:12 -07003115 cap = __get_cap_for_mds(ci, mds);
3116 if (cap && __cap_is_valid(cap)) {
3117 if (force ||
3118 ((cap->issued & drop) &&
3119 (cap->issued & unless) == 0)) {
3120 if ((cap->issued & drop) &&
3121 (cap->issued & unless) == 0) {
Yan, Zhengbb137f82013-06-03 18:22:17 +08003122 int wanted = __ceph_caps_wanted(ci);
3123 if ((ci->i_ceph_flags & CEPH_I_NODELAY) == 0)
3124 wanted |= cap->mds_wanted;
3125 dout("encode_inode_release %p cap %p "
3126 "%s -> %s, wanted %s -> %s\n", inode, cap,
Sage Weila8599bd2009-10-06 11:31:12 -07003127 ceph_cap_string(cap->issued),
Yan, Zhengbb137f82013-06-03 18:22:17 +08003128 ceph_cap_string(cap->issued & ~drop),
3129 ceph_cap_string(cap->mds_wanted),
3130 ceph_cap_string(wanted));
3131
Sage Weila8599bd2009-10-06 11:31:12 -07003132 cap->issued &= ~drop;
3133 cap->implemented &= ~drop;
Yan, Zhengbb137f82013-06-03 18:22:17 +08003134 cap->mds_wanted = wanted;
Sage Weila8599bd2009-10-06 11:31:12 -07003135 } else {
3136 dout("encode_inode_release %p cap %p %s"
3137 " (force)\n", inode, cap,
3138 ceph_cap_string(cap->issued));
3139 }
3140
3141 rel->ino = cpu_to_le64(ceph_ino(inode));
3142 rel->cap_id = cpu_to_le64(cap->cap_id);
3143 rel->seq = cpu_to_le32(cap->seq);
3144 rel->issue_seq = cpu_to_le32(cap->issue_seq),
3145 rel->mseq = cpu_to_le32(cap->mseq);
3146 rel->caps = cpu_to_le32(cap->issued);
3147 rel->wanted = cpu_to_le32(cap->mds_wanted);
3148 rel->dname_len = 0;
3149 rel->dname_seq = 0;
3150 *p += sizeof(*rel);
3151 ret = 1;
3152 } else {
3153 dout("encode_inode_release %p cap %p %s\n",
3154 inode, cap, ceph_cap_string(cap->issued));
3155 }
3156 }
Sage Weilbe655592011-11-30 09:47:09 -08003157 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07003158 return ret;
3159}
3160
3161int ceph_encode_dentry_release(void **p, struct dentry *dentry,
3162 int mds, int drop, int unless)
3163{
3164 struct inode *dir = dentry->d_parent->d_inode;
3165 struct ceph_mds_request_release *rel = *p;
3166 struct ceph_dentry_info *di = ceph_dentry(dentry);
3167 int force = 0;
3168 int ret;
3169
3170 /*
3171 * force an record for the directory caps if we have a dentry lease.
Sage Weilbe655592011-11-30 09:47:09 -08003172 * this is racy (can't take i_ceph_lock and d_lock together), but it
Sage Weila8599bd2009-10-06 11:31:12 -07003173 * doesn't have to be perfect; the mds will revoke anything we don't
3174 * release.
3175 */
3176 spin_lock(&dentry->d_lock);
3177 if (di->lease_session && di->lease_session->s_mds == mds)
3178 force = 1;
3179 spin_unlock(&dentry->d_lock);
3180
3181 ret = ceph_encode_inode_release(p, dir, mds, drop, unless, force);
3182
3183 spin_lock(&dentry->d_lock);
3184 if (ret && di->lease_session && di->lease_session->s_mds == mds) {
3185 dout("encode_dentry_release %p mds%d seq %d\n",
3186 dentry, mds, (int)di->lease_seq);
3187 rel->dname_len = cpu_to_le32(dentry->d_name.len);
3188 memcpy(*p, dentry->d_name.name, dentry->d_name.len);
3189 *p += dentry->d_name.len;
3190 rel->dname_seq = cpu_to_le32(di->lease_seq);
Sage Weil1dadcce2010-07-23 13:54:21 -07003191 __ceph_mdsc_drop_dentry_lease(dentry);
Sage Weila8599bd2009-10-06 11:31:12 -07003192 }
3193 spin_unlock(&dentry->d_lock);
3194 return ret;
3195}