blob: be5ea6af8366479b675e81e1d13e96139abed2c4 [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
Yan, Zhengd9df2782014-04-18 09:57:11 +0800224struct ceph_cap *ceph_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 */
Yan, Zhengd9df2782014-04-18 09:57:11 +0800511void 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 **new_cap)
Sage Weila8599bd2009-10-06 11:31:12 -0700516{
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);
Sage Weila8599bd2009-10-06 11:31:12 -0700519 struct ceph_cap *cap;
520 int mds = session->s_mds;
521 int actual_wanted;
522
523 dout("add_cap %p mds%d cap %llx %s seq %d\n", inode,
524 session->s_mds, cap_id, ceph_cap_string(issued), seq);
525
526 /*
527 * If we are opening the file, include file mode wanted bits
528 * in wanted.
529 */
530 if (fmode >= 0)
531 wanted |= ceph_caps_for_mode(fmode);
532
Sage Weila8599bd2009-10-06 11:31:12 -0700533 cap = __get_cap_for_mds(ci, mds);
534 if (!cap) {
Yan, Zhengd9df2782014-04-18 09:57:11 +0800535 cap = *new_cap;
536 *new_cap = NULL;
Sage Weila8599bd2009-10-06 11:31:12 -0700537
538 cap->issued = 0;
539 cap->implemented = 0;
540 cap->mds = mds;
541 cap->mds_wanted = 0;
Yan, Zheng964266c2013-02-27 09:26:09 +0800542 cap->mseq = 0;
Sage Weila8599bd2009-10-06 11:31:12 -0700543
544 cap->ci = ci;
545 __insert_cap_node(ci, cap);
546
Sage Weila8599bd2009-10-06 11:31:12 -0700547 /* add to session cap list */
548 cap->session = session;
549 spin_lock(&session->s_cap_lock);
550 list_add_tail(&cap->session_caps, &session->s_caps);
551 session->s_nr_caps++;
552 spin_unlock(&session->s_cap_lock);
Yan, Zheng11df2df2013-11-24 14:44:38 +0800553 } else {
Yan, Zheng11df2df2013-11-24 14:44:38 +0800554 /*
555 * auth mds of the inode changed. we received the cap export
556 * message, but still haven't received the cap import message.
557 * handle_cap_export() updated the new auth MDS' cap.
558 *
559 * "ceph_seq_cmp(seq, cap->seq) <= 0" means we are processing
560 * a message that was send before the cap import message. So
561 * don't remove caps.
562 */
563 if (ceph_seq_cmp(seq, cap->seq) <= 0) {
564 WARN_ON(cap != ci->i_auth_cap);
565 WARN_ON(cap->cap_id != cap_id);
566 seq = cap->seq;
567 mseq = cap->mseq;
568 issued |= cap->issued;
569 flags |= CEPH_CAP_FLAG_AUTH;
570 }
571 }
Sage Weila8599bd2009-10-06 11:31:12 -0700572
573 if (!ci->i_snap_realm) {
574 /*
575 * add this inode to the appropriate snap realm
576 */
577 struct ceph_snap_realm *realm = ceph_lookup_snap_realm(mdsc,
578 realmino);
579 if (realm) {
Sage Weila8599bd2009-10-06 11:31:12 -0700580 spin_lock(&realm->inodes_with_caps_lock);
581 ci->i_snap_realm = realm;
582 list_add(&ci->i_snap_realm_item,
583 &realm->inodes_with_caps);
584 spin_unlock(&realm->inodes_with_caps_lock);
585 } else {
586 pr_err("ceph_add_cap: couldn't find snap realm %llx\n",
587 realmino);
Sage Weilb8cd07e2010-07-16 12:00:02 -0700588 WARN_ON(!realm);
Sage Weila8599bd2009-10-06 11:31:12 -0700589 }
590 }
591
592 __check_cap_issue(ci, cap, issued);
593
594 /*
595 * If we are issued caps we don't want, or the mds' wanted
596 * value appears to be off, queue a check so we'll release
597 * later and/or update the mds wanted value.
598 */
599 actual_wanted = __ceph_caps_wanted(ci);
600 if ((wanted & ~actual_wanted) ||
601 (issued & ~actual_wanted & CEPH_CAP_ANY_WR)) {
602 dout(" issued %s, mds wanted %s, actual %s, queueing\n",
603 ceph_cap_string(issued), ceph_cap_string(wanted),
604 ceph_cap_string(actual_wanted));
605 __cap_delay_requeue(mdsc, ci);
606 }
607
Yan, Zhengb8c2f3a2013-05-31 16:37:11 +0800608 if (flags & CEPH_CAP_FLAG_AUTH) {
609 if (ci->i_auth_cap == NULL ||
Yan, Zhengd9ffc4f2014-03-18 10:15:29 +0800610 ceph_seq_cmp(ci->i_auth_cap->mseq, mseq) < 0) {
Yan, Zhengb8c2f3a2013-05-31 16:37:11 +0800611 ci->i_auth_cap = cap;
Yan, Zhengd9ffc4f2014-03-18 10:15:29 +0800612 cap->mds_wanted = wanted;
613 }
Yan, Zheng11df2df2013-11-24 14:44:38 +0800614 } else {
615 WARN_ON(ci->i_auth_cap == cap);
Yan, Zheng8a92a112013-01-04 14:28:07 +0800616 }
Sage Weila8599bd2009-10-06 11:31:12 -0700617
618 dout("add_cap inode %p (%llx.%llx) cap %p %s now %s seq %d mds%d\n",
619 inode, ceph_vinop(inode), cap, ceph_cap_string(issued),
620 ceph_cap_string(issued|cap->issued), seq, mds);
621 cap->cap_id = cap_id;
622 cap->issued = issued;
623 cap->implemented |= issued;
Yan, Zhengd1b87802013-11-13 14:47:19 +0800624 if (ceph_seq_cmp(mseq, cap->mseq) > 0)
Yan, Zheng964266c2013-02-27 09:26:09 +0800625 cap->mds_wanted = wanted;
626 else
627 cap->mds_wanted |= wanted;
Sage Weila8599bd2009-10-06 11:31:12 -0700628 cap->seq = seq;
629 cap->issue_seq = seq;
630 cap->mseq = mseq;
Sage Weil685f9a5d2009-11-09 12:05:48 -0800631 cap->cap_gen = session->s_cap_gen;
Sage Weila8599bd2009-10-06 11:31:12 -0700632
633 if (fmode >= 0)
634 __ceph_get_fmode(ci, fmode);
Sage Weila8599bd2009-10-06 11:31:12 -0700635}
636
637/*
638 * Return true if cap has not timed out and belongs to the current
639 * generation of the MDS session (i.e. has not gone 'stale' due to
640 * us losing touch with the mds).
641 */
642static int __cap_is_valid(struct ceph_cap *cap)
643{
644 unsigned long ttl;
Sage Weilcdac8302009-11-10 16:02:23 -0800645 u32 gen;
Sage Weila8599bd2009-10-06 11:31:12 -0700646
Alex Elderd8fb02a2012-01-12 17:48:10 -0800647 spin_lock(&cap->session->s_gen_ttl_lock);
Sage Weila8599bd2009-10-06 11:31:12 -0700648 gen = cap->session->s_cap_gen;
649 ttl = cap->session->s_cap_ttl;
Alex Elderd8fb02a2012-01-12 17:48:10 -0800650 spin_unlock(&cap->session->s_gen_ttl_lock);
Sage Weila8599bd2009-10-06 11:31:12 -0700651
Sage Weil685f9a5d2009-11-09 12:05:48 -0800652 if (cap->cap_gen < gen || time_after_eq(jiffies, ttl)) {
Sage Weila8599bd2009-10-06 11:31:12 -0700653 dout("__cap_is_valid %p cap %p issued %s "
654 "but STALE (gen %u vs %u)\n", &cap->ci->vfs_inode,
Sage Weil685f9a5d2009-11-09 12:05:48 -0800655 cap, ceph_cap_string(cap->issued), cap->cap_gen, gen);
Sage Weila8599bd2009-10-06 11:31:12 -0700656 return 0;
657 }
658
659 return 1;
660}
661
662/*
663 * Return set of valid cap bits issued to us. Note that caps time
664 * out, and may be invalidated in bulk if the client session times out
665 * and session->s_cap_gen is bumped.
666 */
667int __ceph_caps_issued(struct ceph_inode_info *ci, int *implemented)
668{
Yan, Zhengd9df2782014-04-18 09:57:11 +0800669 int have = ci->i_snap_caps;
Sage Weila8599bd2009-10-06 11:31:12 -0700670 struct ceph_cap *cap;
671 struct rb_node *p;
672
673 if (implemented)
674 *implemented = 0;
675 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
676 cap = rb_entry(p, struct ceph_cap, ci_node);
677 if (!__cap_is_valid(cap))
678 continue;
679 dout("__ceph_caps_issued %p cap %p issued %s\n",
680 &ci->vfs_inode, cap, ceph_cap_string(cap->issued));
681 have |= cap->issued;
682 if (implemented)
683 *implemented |= cap->implemented;
684 }
Yan, Zhengb1530f52013-07-02 12:40:20 +0800685 /*
686 * exclude caps issued by non-auth MDS, but are been revoking
687 * by the auth MDS. The non-auth MDS should be revoking/exporting
688 * these caps, but the message is delayed.
689 */
690 if (ci->i_auth_cap) {
691 cap = ci->i_auth_cap;
692 have &= ~cap->implemented | cap->issued;
693 }
Sage Weila8599bd2009-10-06 11:31:12 -0700694 return have;
695}
696
697/*
698 * Get cap bits issued by caps other than @ocap
699 */
700int __ceph_caps_issued_other(struct ceph_inode_info *ci, struct ceph_cap *ocap)
701{
702 int have = ci->i_snap_caps;
703 struct ceph_cap *cap;
704 struct rb_node *p;
705
706 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
707 cap = rb_entry(p, struct ceph_cap, ci_node);
708 if (cap == ocap)
709 continue;
710 if (!__cap_is_valid(cap))
711 continue;
712 have |= cap->issued;
713 }
714 return have;
715}
716
717/*
718 * Move a cap to the end of the LRU (oldest caps at list head, newest
719 * at list tail).
720 */
721static void __touch_cap(struct ceph_cap *cap)
722{
723 struct ceph_mds_session *s = cap->session;
724
Sage Weila8599bd2009-10-06 11:31:12 -0700725 spin_lock(&s->s_cap_lock);
Sage Weil7c1332b2010-02-16 11:39:45 -0800726 if (s->s_cap_iterator == NULL) {
Sage Weil5dacf092009-12-21 20:40:34 -0800727 dout("__touch_cap %p cap %p mds%d\n", &cap->ci->vfs_inode, cap,
728 s->s_mds);
729 list_move_tail(&cap->session_caps, &s->s_caps);
730 } else {
731 dout("__touch_cap %p cap %p mds%d NOP, iterating over caps\n",
732 &cap->ci->vfs_inode, cap, s->s_mds);
733 }
Sage Weila8599bd2009-10-06 11:31:12 -0700734 spin_unlock(&s->s_cap_lock);
735}
736
737/*
738 * Check if we hold the given mask. If so, move the cap(s) to the
739 * front of their respective LRUs. (This is the preferred way for
740 * callers to check for caps they want.)
741 */
742int __ceph_caps_issued_mask(struct ceph_inode_info *ci, int mask, int touch)
743{
744 struct ceph_cap *cap;
745 struct rb_node *p;
746 int have = ci->i_snap_caps;
747
748 if ((have & mask) == mask) {
749 dout("__ceph_caps_issued_mask %p snap issued %s"
750 " (mask %s)\n", &ci->vfs_inode,
751 ceph_cap_string(have),
752 ceph_cap_string(mask));
753 return 1;
754 }
755
756 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
757 cap = rb_entry(p, struct ceph_cap, ci_node);
758 if (!__cap_is_valid(cap))
759 continue;
760 if ((cap->issued & mask) == mask) {
761 dout("__ceph_caps_issued_mask %p cap %p issued %s"
762 " (mask %s)\n", &ci->vfs_inode, cap,
763 ceph_cap_string(cap->issued),
764 ceph_cap_string(mask));
765 if (touch)
766 __touch_cap(cap);
767 return 1;
768 }
769
770 /* does a combination of caps satisfy mask? */
771 have |= cap->issued;
772 if ((have & mask) == mask) {
773 dout("__ceph_caps_issued_mask %p combo issued %s"
774 " (mask %s)\n", &ci->vfs_inode,
775 ceph_cap_string(cap->issued),
776 ceph_cap_string(mask));
777 if (touch) {
778 struct rb_node *q;
779
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300780 /* touch this + preceding caps */
Sage Weila8599bd2009-10-06 11:31:12 -0700781 __touch_cap(cap);
782 for (q = rb_first(&ci->i_caps); q != p;
783 q = rb_next(q)) {
784 cap = rb_entry(q, struct ceph_cap,
785 ci_node);
786 if (!__cap_is_valid(cap))
787 continue;
788 __touch_cap(cap);
789 }
790 }
791 return 1;
792 }
793 }
794
795 return 0;
796}
797
798/*
799 * Return true if mask caps are currently being revoked by an MDS.
800 */
Yan, Zheng6ee6b9532013-07-02 12:40:21 +0800801int __ceph_caps_revoking_other(struct ceph_inode_info *ci,
802 struct ceph_cap *ocap, int mask)
803{
804 struct ceph_cap *cap;
805 struct rb_node *p;
806
807 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
808 cap = rb_entry(p, struct ceph_cap, ci_node);
Yan, Zheng9563f882013-11-22 13:50:45 +0800809 if (cap != ocap &&
Yan, Zheng6ee6b9532013-07-02 12:40:21 +0800810 (cap->implemented & ~cap->issued & mask))
811 return 1;
812 }
813 return 0;
814}
815
Sage Weila8599bd2009-10-06 11:31:12 -0700816int ceph_caps_revoking(struct ceph_inode_info *ci, int mask)
817{
818 struct inode *inode = &ci->vfs_inode;
Yan, Zheng6ee6b9532013-07-02 12:40:21 +0800819 int ret;
Sage Weila8599bd2009-10-06 11:31:12 -0700820
Sage Weilbe655592011-11-30 09:47:09 -0800821 spin_lock(&ci->i_ceph_lock);
Yan, Zheng6ee6b9532013-07-02 12:40:21 +0800822 ret = __ceph_caps_revoking_other(ci, NULL, mask);
Sage Weilbe655592011-11-30 09:47:09 -0800823 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -0700824 dout("ceph_caps_revoking %p %s = %d\n", inode,
825 ceph_cap_string(mask), ret);
826 return ret;
827}
828
829int __ceph_caps_used(struct ceph_inode_info *ci)
830{
831 int used = 0;
832 if (ci->i_pin_ref)
833 used |= CEPH_CAP_PIN;
834 if (ci->i_rd_ref)
835 used |= CEPH_CAP_FILE_RD;
Sage Weila43fb732010-09-17 09:54:08 -0700836 if (ci->i_rdcache_ref || ci->vfs_inode.i_data.nrpages)
Sage Weila8599bd2009-10-06 11:31:12 -0700837 used |= CEPH_CAP_FILE_CACHE;
838 if (ci->i_wr_ref)
839 used |= CEPH_CAP_FILE_WR;
Henry C Changd3d07202011-05-11 10:29:54 +0000840 if (ci->i_wb_ref || ci->i_wrbuffer_ref)
Sage Weila8599bd2009-10-06 11:31:12 -0700841 used |= CEPH_CAP_FILE_BUFFER;
842 return used;
843}
844
845/*
846 * wanted, by virtue of open file modes
847 */
848int __ceph_caps_file_wanted(struct ceph_inode_info *ci)
849{
850 int want = 0;
851 int mode;
Sage Weil33caad32010-05-26 14:31:27 -0700852 for (mode = 0; mode < CEPH_FILE_MODE_NUM; mode++)
Sage Weila8599bd2009-10-06 11:31:12 -0700853 if (ci->i_nr_by_mode[mode])
854 want |= ceph_caps_for_mode(mode);
855 return want;
856}
857
858/*
859 * Return caps we have registered with the MDS(s) as 'wanted'.
860 */
861int __ceph_caps_mds_wanted(struct ceph_inode_info *ci)
862{
863 struct ceph_cap *cap;
864 struct rb_node *p;
865 int mds_wanted = 0;
866
867 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
868 cap = rb_entry(p, struct ceph_cap, ci_node);
869 if (!__cap_is_valid(cap))
870 continue;
Yan, Zhenga2550602014-03-08 09:51:45 +0800871 if (cap == ci->i_auth_cap)
872 mds_wanted |= cap->mds_wanted;
873 else
874 mds_wanted |= (cap->mds_wanted & ~CEPH_CAP_ANY_FILE_WR);
Sage Weila8599bd2009-10-06 11:31:12 -0700875 }
876 return mds_wanted;
877}
878
879/*
Sage Weilbe655592011-11-30 09:47:09 -0800880 * called under i_ceph_lock
Sage Weila8599bd2009-10-06 11:31:12 -0700881 */
882static int __ceph_is_any_caps(struct ceph_inode_info *ci)
883{
Yan, Zhengd9df2782014-04-18 09:57:11 +0800884 return !RB_EMPTY_ROOT(&ci->i_caps);
Sage Weila8599bd2009-10-06 11:31:12 -0700885}
886
Yan, Zheng9215aee2013-11-30 12:47:41 +0800887int ceph_is_any_caps(struct inode *inode)
888{
889 struct ceph_inode_info *ci = ceph_inode(inode);
890 int ret;
891
892 spin_lock(&ci->i_ceph_lock);
893 ret = __ceph_is_any_caps(ci);
894 spin_unlock(&ci->i_ceph_lock);
895
896 return ret;
897}
898
Yan, Zhengdb40cc12015-03-23 20:12:20 +0800899static void drop_inode_snap_realm(struct ceph_inode_info *ci)
900{
901 struct ceph_snap_realm *realm = ci->i_snap_realm;
902 spin_lock(&realm->inodes_with_caps_lock);
903 list_del_init(&ci->i_snap_realm_item);
904 ci->i_snap_realm_counter++;
905 ci->i_snap_realm = NULL;
906 spin_unlock(&realm->inodes_with_caps_lock);
907 ceph_put_snap_realm(ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc,
908 realm);
909}
910
Sage Weila8599bd2009-10-06 11:31:12 -0700911/*
Sage Weilf818a732010-05-11 20:56:31 -0700912 * Remove a cap. Take steps to deal with a racing iterate_session_caps.
913 *
Sage Weilbe655592011-11-30 09:47:09 -0800914 * caller should hold i_ceph_lock.
Sage Weila6369742010-02-22 13:59:00 -0800915 * caller will not hold session s_mutex if called from destroy_inode.
Sage Weila8599bd2009-10-06 11:31:12 -0700916 */
Yan, Zhenga096b092013-09-22 10:15:58 +0800917void __ceph_remove_cap(struct ceph_cap *cap, bool queue_release)
Sage Weila8599bd2009-10-06 11:31:12 -0700918{
919 struct ceph_mds_session *session = cap->session;
920 struct ceph_inode_info *ci = cap->ci;
Cheng Renquan640ef792010-03-26 17:40:33 +0800921 struct ceph_mds_client *mdsc =
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700922 ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc;
Sage Weilf818a732010-05-11 20:56:31 -0700923 int removed = 0;
Sage Weila8599bd2009-10-06 11:31:12 -0700924
925 dout("__ceph_remove_cap %p from %p\n", cap, &ci->vfs_inode);
926
Sage Weil7c1332b2010-02-16 11:39:45 -0800927 /* remove from session list */
928 spin_lock(&session->s_cap_lock);
Yan, Zheng99a9c272013-09-22 11:08:14 +0800929 /*
930 * s_cap_reconnect is protected by s_cap_lock. no one changes
931 * s_cap_gen while session is in the reconnect state.
932 */
933 if (queue_release &&
934 (!session->s_cap_reconnect ||
935 cap->cap_gen == session->s_cap_gen))
Yan, Zhenga096b092013-09-22 10:15:58 +0800936 __queue_cap_release(session, ci->i_vino.ino, cap->cap_id,
937 cap->mseq, cap->issue_seq);
938
Sage Weil7c1332b2010-02-16 11:39:45 -0800939 if (session->s_cap_iterator == cap) {
940 /* not yet, we are iterating over this very cap */
941 dout("__ceph_remove_cap delaying %p removal from session %p\n",
942 cap, cap->session);
943 } else {
944 list_del_init(&cap->session_caps);
945 session->s_nr_caps--;
946 cap->session = NULL;
Sage Weilf818a732010-05-11 20:56:31 -0700947 removed = 1;
Sage Weil7c1332b2010-02-16 11:39:45 -0800948 }
Sage Weilf818a732010-05-11 20:56:31 -0700949 /* protect backpointer with s_cap_lock: see iterate_session_caps */
950 cap->ci = NULL;
Sage Weil7c1332b2010-02-16 11:39:45 -0800951 spin_unlock(&session->s_cap_lock);
952
Sage Weilf818a732010-05-11 20:56:31 -0700953 /* remove from inode list */
954 rb_erase(&cap->ci_node, &ci->i_caps);
955 if (ci->i_auth_cap == cap)
956 ci->i_auth_cap = NULL;
957
958 if (removed)
Yehuda Sadeh37151662010-06-17 16:16:12 -0700959 ceph_put_cap(mdsc, cap);
Sage Weila8599bd2009-10-06 11:31:12 -0700960
Yan, Zhengdb40cc12015-03-23 20:12:20 +0800961 /* when reconnect denied, we remove session caps forcibly,
962 * i_wr_ref can be non-zero. If there are ongoing write,
963 * keep i_snap_realm.
964 */
965 if (!__ceph_is_any_caps(ci) && ci->i_wr_ref == 0 && ci->i_snap_realm)
966 drop_inode_snap_realm(ci);
967
Sage Weila8599bd2009-10-06 11:31:12 -0700968 if (!__ceph_is_any_real_caps(ci))
969 __cap_delay_cancel(mdsc, ci);
970}
971
972/*
973 * Build and send a cap message to the given MDS.
974 *
975 * Caller should be holding s_mutex.
976 */
977static int send_cap_msg(struct ceph_mds_session *session,
978 u64 ino, u64 cid, int op,
979 int caps, int wanted, int dirty,
980 u32 seq, u64 flush_tid, u32 issue_seq, u32 mseq,
981 u64 size, u64 max_size,
982 struct timespec *mtime, struct timespec *atime,
983 u64 time_warp_seq,
Eric W. Biederman05cb11c2013-01-31 02:56:19 -0800984 kuid_t uid, kgid_t gid, umode_t mode,
Sage Weila8599bd2009-10-06 11:31:12 -0700985 u64 xattr_version,
986 struct ceph_buffer *xattrs_buf,
Yan, Zhenge20d2582014-11-14 22:39:13 +0800987 u64 follows, bool inline_data)
Sage Weila8599bd2009-10-06 11:31:12 -0700988{
989 struct ceph_mds_caps *fc;
990 struct ceph_msg *msg;
Yan, Zhenge20d2582014-11-14 22:39:13 +0800991 void *p;
992 size_t extra_len;
Sage Weila8599bd2009-10-06 11:31:12 -0700993
994 dout("send_cap_msg %s %llx %llx caps %s wanted %s dirty %s"
995 " seq %u/%u mseq %u follows %lld size %llu/%llu"
996 " xattr_ver %llu xattr_len %d\n", ceph_cap_op_name(op),
997 cid, ino, ceph_cap_string(caps), ceph_cap_string(wanted),
998 ceph_cap_string(dirty),
999 seq, issue_seq, mseq, follows, size, max_size,
1000 xattr_version, xattrs_buf ? (int)xattrs_buf->vec.iov_len : 0);
1001
Yan, Zhenge20d2582014-11-14 22:39:13 +08001002 /* flock buffer size + inline version + inline data size */
1003 extra_len = 4 + 8 + 4;
1004 msg = ceph_msg_new(CEPH_MSG_CLIENT_CAPS, sizeof(*fc) + extra_len,
1005 GFP_NOFS, false);
Sage Weila79832f2010-04-01 16:06:19 -07001006 if (!msg)
1007 return -ENOMEM;
Sage Weila8599bd2009-10-06 11:31:12 -07001008
Sage Weil6df058c2009-12-22 11:24:33 -08001009 msg->hdr.tid = cpu_to_le64(flush_tid);
Sage Weila8599bd2009-10-06 11:31:12 -07001010
Sage Weil6df058c2009-12-22 11:24:33 -08001011 fc = msg->front.iov_base;
Sage Weila8599bd2009-10-06 11:31:12 -07001012 memset(fc, 0, sizeof(*fc));
1013
1014 fc->cap_id = cpu_to_le64(cid);
1015 fc->op = cpu_to_le32(op);
1016 fc->seq = cpu_to_le32(seq);
Sage Weila8599bd2009-10-06 11:31:12 -07001017 fc->issue_seq = cpu_to_le32(issue_seq);
1018 fc->migrate_seq = cpu_to_le32(mseq);
1019 fc->caps = cpu_to_le32(caps);
1020 fc->wanted = cpu_to_le32(wanted);
1021 fc->dirty = cpu_to_le32(dirty);
1022 fc->ino = cpu_to_le64(ino);
1023 fc->snap_follows = cpu_to_le64(follows);
1024
1025 fc->size = cpu_to_le64(size);
1026 fc->max_size = cpu_to_le64(max_size);
1027 if (mtime)
1028 ceph_encode_timespec(&fc->mtime, mtime);
1029 if (atime)
1030 ceph_encode_timespec(&fc->atime, atime);
1031 fc->time_warp_seq = cpu_to_le32(time_warp_seq);
1032
Eric W. Biederman05cb11c2013-01-31 02:56:19 -08001033 fc->uid = cpu_to_le32(from_kuid(&init_user_ns, uid));
1034 fc->gid = cpu_to_le32(from_kgid(&init_user_ns, gid));
Sage Weila8599bd2009-10-06 11:31:12 -07001035 fc->mode = cpu_to_le32(mode);
1036
Yan, Zhenge20d2582014-11-14 22:39:13 +08001037 p = fc + 1;
1038 /* flock buffer size */
1039 ceph_encode_32(&p, 0);
1040 /* inline version */
1041 ceph_encode_64(&p, inline_data ? 0 : CEPH_INLINE_NONE);
1042 /* inline data size */
1043 ceph_encode_32(&p, 0);
1044
Sage Weila8599bd2009-10-06 11:31:12 -07001045 fc->xattr_version = cpu_to_le64(xattr_version);
1046 if (xattrs_buf) {
1047 msg->middle = ceph_buffer_get(xattrs_buf);
1048 fc->xattr_len = cpu_to_le32(xattrs_buf->vec.iov_len);
1049 msg->hdr.middle_len = cpu_to_le32(xattrs_buf->vec.iov_len);
1050 }
1051
1052 ceph_con_send(&session->s_con, msg);
1053 return 0;
1054}
1055
Yan, Zhengd40ee0d2013-02-18 13:43:43 +08001056void __queue_cap_release(struct ceph_mds_session *session,
1057 u64 ino, u64 cap_id, u32 migrate_seq,
1058 u32 issue_seq)
Sage Weil3d7ded42010-06-09 16:47:10 -07001059{
1060 struct ceph_msg *msg;
1061 struct ceph_mds_cap_release *head;
1062 struct ceph_mds_cap_item *item;
1063
Sage Weil3d7ded42010-06-09 16:47:10 -07001064 BUG_ON(!session->s_num_cap_releases);
1065 msg = list_first_entry(&session->s_cap_releases,
1066 struct ceph_msg, list_head);
1067
1068 dout(" adding %llx release to mds%d msg %p (%d left)\n",
1069 ino, session->s_mds, msg, session->s_num_cap_releases);
1070
1071 BUG_ON(msg->front.iov_len + sizeof(*item) > PAGE_CACHE_SIZE);
1072 head = msg->front.iov_base;
Wei Yongjunb905a7f2012-09-28 12:59:16 +08001073 le32_add_cpu(&head->num, 1);
Sage Weil3d7ded42010-06-09 16:47:10 -07001074 item = msg->front.iov_base + msg->front.iov_len;
1075 item->ino = cpu_to_le64(ino);
1076 item->cap_id = cpu_to_le64(cap_id);
1077 item->migrate_seq = cpu_to_le32(migrate_seq);
1078 item->seq = cpu_to_le32(issue_seq);
1079
1080 session->s_num_cap_releases--;
1081
1082 msg->front.iov_len += sizeof(*item);
1083 if (le32_to_cpu(head->num) == CEPH_CAPS_PER_RELEASE) {
1084 dout(" release msg %p full\n", msg);
1085 list_move_tail(&msg->list_head, &session->s_cap_releases_done);
1086 } else {
1087 dout(" release msg %p at %d/%d (%d)\n", msg,
1088 (int)le32_to_cpu(head->num),
1089 (int)CEPH_CAPS_PER_RELEASE,
1090 (int)msg->front.iov_len);
1091 }
Sage Weil3d7ded42010-06-09 16:47:10 -07001092}
1093
Sage Weila8599bd2009-10-06 11:31:12 -07001094/*
Sage Weila6369742010-02-22 13:59:00 -08001095 * Queue cap releases when an inode is dropped from our cache. Since
Sage Weilbe655592011-11-30 09:47:09 -08001096 * inode is about to be destroyed, there is no need for i_ceph_lock.
Sage Weila8599bd2009-10-06 11:31:12 -07001097 */
1098void ceph_queue_caps_release(struct inode *inode)
1099{
1100 struct ceph_inode_info *ci = ceph_inode(inode);
1101 struct rb_node *p;
1102
Sage Weila8599bd2009-10-06 11:31:12 -07001103 p = rb_first(&ci->i_caps);
1104 while (p) {
1105 struct ceph_cap *cap = rb_entry(p, struct ceph_cap, ci_node);
Sage Weila8599bd2009-10-06 11:31:12 -07001106 p = rb_next(p);
Yan, Zhenga096b092013-09-22 10:15:58 +08001107 __ceph_remove_cap(cap, true);
Sage Weila8599bd2009-10-06 11:31:12 -07001108 }
Sage Weila8599bd2009-10-06 11:31:12 -07001109}
1110
1111/*
1112 * Send a cap msg on the given inode. Update our caps state, then
Sage Weilbe655592011-11-30 09:47:09 -08001113 * drop i_ceph_lock and send the message.
Sage Weila8599bd2009-10-06 11:31:12 -07001114 *
1115 * Make note of max_size reported/requested from mds, revoked caps
1116 * that have now been implemented.
1117 *
1118 * Make half-hearted attempt ot to invalidate page cache if we are
1119 * dropping RDCACHE. Note that this will leave behind locked pages
1120 * that we'll then need to deal with elsewhere.
1121 *
1122 * Return non-zero if delayed release, or we experienced an error
1123 * such that the caller should requeue + retry later.
1124 *
Sage Weilbe655592011-11-30 09:47:09 -08001125 * called with i_ceph_lock, then drops it.
Sage Weila8599bd2009-10-06 11:31:12 -07001126 * caller should hold snap_rwsem (read), s_mutex.
1127 */
1128static int __send_cap(struct ceph_mds_client *mdsc, struct ceph_cap *cap,
1129 int op, int used, int want, int retain, int flushing,
1130 unsigned *pflush_tid)
Sage Weilbe655592011-11-30 09:47:09 -08001131 __releases(cap->ci->i_ceph_lock)
Sage Weila8599bd2009-10-06 11:31:12 -07001132{
1133 struct ceph_inode_info *ci = cap->ci;
1134 struct inode *inode = &ci->vfs_inode;
1135 u64 cap_id = cap->cap_id;
Sage Weil68c28322010-02-09 13:41:47 -08001136 int held, revoking, dropping, keep;
Sage Weila8599bd2009-10-06 11:31:12 -07001137 u64 seq, issue_seq, mseq, time_warp_seq, follows;
1138 u64 size, max_size;
1139 struct timespec mtime, atime;
1140 int wake = 0;
Al Viro5706b272011-07-26 04:52:22 -04001141 umode_t mode;
Eric W. Biederman05cb11c2013-01-31 02:56:19 -08001142 kuid_t uid;
1143 kgid_t gid;
Sage Weila8599bd2009-10-06 11:31:12 -07001144 struct ceph_mds_session *session;
1145 u64 xattr_version = 0;
Sage Weil082afec2010-08-22 15:16:41 -07001146 struct ceph_buffer *xattr_blob = NULL;
Sage Weila8599bd2009-10-06 11:31:12 -07001147 int delayed = 0;
1148 u64 flush_tid = 0;
1149 int i;
1150 int ret;
Yan, Zhenge20d2582014-11-14 22:39:13 +08001151 bool inline_data;
Sage Weila8599bd2009-10-06 11:31:12 -07001152
Sage Weil68c28322010-02-09 13:41:47 -08001153 held = cap->issued | cap->implemented;
1154 revoking = cap->implemented & ~cap->issued;
1155 retain &= ~revoking;
1156 dropping = cap->issued & ~retain;
1157
Sage Weila8599bd2009-10-06 11:31:12 -07001158 dout("__send_cap %p cap %p session %p %s -> %s (revoking %s)\n",
1159 inode, cap, cap->session,
1160 ceph_cap_string(held), ceph_cap_string(held & retain),
1161 ceph_cap_string(revoking));
1162 BUG_ON((retain & CEPH_CAP_PIN) == 0);
1163
1164 session = cap->session;
1165
1166 /* don't release wanted unless we've waited a bit. */
1167 if ((ci->i_ceph_flags & CEPH_I_NODELAY) == 0 &&
1168 time_before(jiffies, ci->i_hold_caps_min)) {
1169 dout(" delaying issued %s -> %s, wanted %s -> %s on send\n",
1170 ceph_cap_string(cap->issued),
1171 ceph_cap_string(cap->issued & retain),
1172 ceph_cap_string(cap->mds_wanted),
1173 ceph_cap_string(want));
1174 want |= cap->mds_wanted;
1175 retain |= cap->issued;
1176 delayed = 1;
1177 }
1178 ci->i_ceph_flags &= ~(CEPH_I_NODELAY | CEPH_I_FLUSH);
1179
1180 cap->issued &= retain; /* drop bits we don't want */
1181 if (cap->implemented & ~cap->issued) {
1182 /*
1183 * Wake up any waiters on wanted -> needed transition.
1184 * This is due to the weird transition from buffered
1185 * to sync IO... we need to flush dirty pages _before_
1186 * allowing sync writes to avoid reordering.
1187 */
1188 wake = 1;
1189 }
1190 cap->implemented &= cap->issued | used;
1191 cap->mds_wanted = want;
1192
1193 if (flushing) {
1194 /*
1195 * assign a tid for flush operations so we can avoid
1196 * flush1 -> dirty1 -> flush2 -> flushack1 -> mark
1197 * clean type races. track latest tid for every bit
1198 * so we can handle flush AxFw, flush Fw, and have the
1199 * first ack clean Ax.
1200 */
1201 flush_tid = ++ci->i_cap_flush_last_tid;
1202 if (pflush_tid)
1203 *pflush_tid = flush_tid;
1204 dout(" cap_flush_tid %d\n", (int)flush_tid);
1205 for (i = 0; i < CEPH_CAP_BITS; i++)
1206 if (flushing & (1 << i))
1207 ci->i_cap_flush_tid[i] = flush_tid;
Sage Weil7d8cb262010-08-24 08:44:16 -07001208
1209 follows = ci->i_head_snapc->seq;
1210 } else {
1211 follows = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07001212 }
1213
1214 keep = cap->implemented;
1215 seq = cap->seq;
1216 issue_seq = cap->issue_seq;
1217 mseq = cap->mseq;
1218 size = inode->i_size;
1219 ci->i_reported_size = size;
1220 max_size = ci->i_wanted_max_size;
1221 ci->i_requested_max_size = max_size;
1222 mtime = inode->i_mtime;
1223 atime = inode->i_atime;
1224 time_warp_seq = ci->i_time_warp_seq;
Sage Weila8599bd2009-10-06 11:31:12 -07001225 uid = inode->i_uid;
1226 gid = inode->i_gid;
1227 mode = inode->i_mode;
1228
Sage Weil082afec2010-08-22 15:16:41 -07001229 if (flushing & CEPH_CAP_XATTR_EXCL) {
Sage Weila8599bd2009-10-06 11:31:12 -07001230 __ceph_build_xattrs_blob(ci);
Sage Weil082afec2010-08-22 15:16:41 -07001231 xattr_blob = ci->i_xattrs.blob;
1232 xattr_version = ci->i_xattrs.version;
Sage Weila8599bd2009-10-06 11:31:12 -07001233 }
1234
Yan, Zhenge20d2582014-11-14 22:39:13 +08001235 inline_data = ci->i_inline_version != CEPH_INLINE_NONE;
1236
Sage Weilbe655592011-11-30 09:47:09 -08001237 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001238
Sage Weila8599bd2009-10-06 11:31:12 -07001239 ret = send_cap_msg(session, ceph_vino(inode).ino, cap_id,
1240 op, keep, want, flushing, seq, flush_tid, issue_seq, mseq,
1241 size, max_size, &mtime, &atime, time_warp_seq,
Sage Weil082afec2010-08-22 15:16:41 -07001242 uid, gid, mode, xattr_version, xattr_blob,
Yan, Zhenge20d2582014-11-14 22:39:13 +08001243 follows, inline_data);
Sage Weila8599bd2009-10-06 11:31:12 -07001244 if (ret < 0) {
1245 dout("error sending cap msg, must requeue %p\n", inode);
1246 delayed = 1;
1247 }
1248
1249 if (wake)
Yehuda Sadeh03066f22010-07-27 13:11:08 -07001250 wake_up_all(&ci->i_cap_wq);
Sage Weila8599bd2009-10-06 11:31:12 -07001251
1252 return delayed;
1253}
1254
1255/*
1256 * When a snapshot is taken, clients accumulate dirty metadata on
1257 * inodes with capabilities in ceph_cap_snaps to describe the file
1258 * state at the time the snapshot was taken. This must be flushed
1259 * asynchronously back to the MDS once sync writes complete and dirty
1260 * data is written out.
1261 *
Sage Weile8351242010-09-17 08:03:08 -07001262 * Unless @again is true, skip cap_snaps that were already sent to
1263 * the MDS (i.e., during this session).
1264 *
Sage Weilbe655592011-11-30 09:47:09 -08001265 * Called under i_ceph_lock. Takes s_mutex as needed.
Sage Weila8599bd2009-10-06 11:31:12 -07001266 */
1267void __ceph_flush_snaps(struct ceph_inode_info *ci,
Sage Weile8351242010-09-17 08:03:08 -07001268 struct ceph_mds_session **psession,
1269 int again)
Sage Weilbe655592011-11-30 09:47:09 -08001270 __releases(ci->i_ceph_lock)
1271 __acquires(ci->i_ceph_lock)
Sage Weila8599bd2009-10-06 11:31:12 -07001272{
1273 struct inode *inode = &ci->vfs_inode;
1274 int mds;
1275 struct ceph_cap_snap *capsnap;
1276 u32 mseq;
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001277 struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
Sage Weila8599bd2009-10-06 11:31:12 -07001278 struct ceph_mds_session *session = NULL; /* if session != NULL, we hold
1279 session->s_mutex */
1280 u64 next_follows = 0; /* keep track of how far we've gotten through the
1281 i_cap_snaps list, and skip these entries next time
1282 around to avoid an infinite loop */
1283
1284 if (psession)
1285 session = *psession;
1286
1287 dout("__flush_snaps %p\n", inode);
1288retry:
1289 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
1290 /* avoid an infiniute loop after retry */
1291 if (capsnap->follows < next_follows)
1292 continue;
1293 /*
1294 * we need to wait for sync writes to complete and for dirty
1295 * pages to be written out.
1296 */
1297 if (capsnap->dirty_pages || capsnap->writing)
Sage Weilcfc0bf62010-09-14 15:50:59 -07001298 break;
Sage Weila8599bd2009-10-06 11:31:12 -07001299
Sage Weil819ccbf2010-04-01 09:33:46 -07001300 /*
1301 * if cap writeback already occurred, we should have dropped
1302 * the capsnap in ceph_put_wrbuffer_cap_refs.
1303 */
1304 BUG_ON(capsnap->dirty == 0);
1305
Sage Weila8599bd2009-10-06 11:31:12 -07001306 /* pick mds, take s_mutex */
Sage Weilca81f3f2010-06-10 14:21:36 -07001307 if (ci->i_auth_cap == NULL) {
1308 dout("no auth cap (migrating?), doing nothing\n");
1309 goto out;
1310 }
Sage Weile8351242010-09-17 08:03:08 -07001311
1312 /* only flush each capsnap once */
1313 if (!again && !list_empty(&capsnap->flushing_item)) {
1314 dout("already flushed %p, skipping\n", capsnap);
1315 continue;
1316 }
1317
Sage Weilca81f3f2010-06-10 14:21:36 -07001318 mds = ci->i_auth_cap->session->s_mds;
1319 mseq = ci->i_auth_cap->mseq;
1320
Sage Weila8599bd2009-10-06 11:31:12 -07001321 if (session && session->s_mds != mds) {
1322 dout("oops, wrong session %p mutex\n", session);
1323 mutex_unlock(&session->s_mutex);
1324 ceph_put_mds_session(session);
1325 session = NULL;
1326 }
1327 if (!session) {
Sage Weilbe655592011-11-30 09:47:09 -08001328 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001329 mutex_lock(&mdsc->mutex);
1330 session = __ceph_lookup_mds_session(mdsc, mds);
1331 mutex_unlock(&mdsc->mutex);
1332 if (session) {
1333 dout("inverting session/ino locks on %p\n",
1334 session);
1335 mutex_lock(&session->s_mutex);
1336 }
1337 /*
1338 * if session == NULL, we raced against a cap
Sage Weilca81f3f2010-06-10 14:21:36 -07001339 * deletion or migration. retry, and we'll
1340 * get a better @mds value next time.
Sage Weila8599bd2009-10-06 11:31:12 -07001341 */
Sage Weilbe655592011-11-30 09:47:09 -08001342 spin_lock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001343 goto retry;
1344 }
1345
1346 capsnap->flush_tid = ++ci->i_cap_flush_last_tid;
1347 atomic_inc(&capsnap->nref);
1348 if (!list_empty(&capsnap->flushing_item))
1349 list_del_init(&capsnap->flushing_item);
1350 list_add_tail(&capsnap->flushing_item,
1351 &session->s_cap_snaps_flushing);
Sage Weilbe655592011-11-30 09:47:09 -08001352 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001353
Sage Weilcfc0bf62010-09-14 15:50:59 -07001354 dout("flush_snaps %p cap_snap %p follows %lld tid %llu\n",
1355 inode, capsnap, capsnap->follows, capsnap->flush_tid);
Sage Weila8599bd2009-10-06 11:31:12 -07001356 send_cap_msg(session, ceph_vino(inode).ino, 0,
1357 CEPH_CAP_OP_FLUSHSNAP, capsnap->issued, 0,
1358 capsnap->dirty, 0, capsnap->flush_tid, 0, mseq,
1359 capsnap->size, 0,
1360 &capsnap->mtime, &capsnap->atime,
1361 capsnap->time_warp_seq,
1362 capsnap->uid, capsnap->gid, capsnap->mode,
Sage Weil4a625be2010-08-22 15:03:56 -07001363 capsnap->xattr_version, capsnap->xattr_blob,
Yan, Zhenge20d2582014-11-14 22:39:13 +08001364 capsnap->follows, capsnap->inline_data);
Sage Weila8599bd2009-10-06 11:31:12 -07001365
1366 next_follows = capsnap->follows + 1;
1367 ceph_put_cap_snap(capsnap);
1368
Sage Weilbe655592011-11-30 09:47:09 -08001369 spin_lock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001370 goto retry;
1371 }
1372
1373 /* we flushed them all; remove this inode from the queue */
1374 spin_lock(&mdsc->snap_flush_lock);
1375 list_del_init(&ci->i_snap_flush_item);
1376 spin_unlock(&mdsc->snap_flush_lock);
1377
Sage Weilca81f3f2010-06-10 14:21:36 -07001378out:
Sage Weila8599bd2009-10-06 11:31:12 -07001379 if (psession)
1380 *psession = session;
1381 else if (session) {
1382 mutex_unlock(&session->s_mutex);
1383 ceph_put_mds_session(session);
1384 }
1385}
1386
1387static void ceph_flush_snaps(struct ceph_inode_info *ci)
1388{
Sage Weilbe655592011-11-30 09:47:09 -08001389 spin_lock(&ci->i_ceph_lock);
Sage Weile8351242010-09-17 08:03:08 -07001390 __ceph_flush_snaps(ci, NULL, 0);
Sage Weilbe655592011-11-30 09:47:09 -08001391 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001392}
1393
1394/*
Sage Weilfca65b42011-05-04 11:33:47 -07001395 * Mark caps dirty. If inode is newly dirty, return the dirty flags.
1396 * Caller is then responsible for calling __mark_inode_dirty with the
1397 * returned flags value.
Sage Weil76e3b392009-10-15 18:13:53 -07001398 */
Sage Weilfca65b42011-05-04 11:33:47 -07001399int __ceph_mark_dirty_caps(struct ceph_inode_info *ci, int mask)
Sage Weil76e3b392009-10-15 18:13:53 -07001400{
Cheng Renquan640ef792010-03-26 17:40:33 +08001401 struct ceph_mds_client *mdsc =
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001402 ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc;
Sage Weil76e3b392009-10-15 18:13:53 -07001403 struct inode *inode = &ci->vfs_inode;
1404 int was = ci->i_dirty_caps;
1405 int dirty = 0;
1406
Yan, Zheng571ade32015-03-24 11:36:08 +08001407 if (!ci->i_auth_cap) {
1408 pr_warn("__mark_dirty_caps %p %llx mask %s, "
1409 "but no auth cap (session was closed?)\n",
1410 inode, ceph_ino(inode), ceph_cap_string(mask));
1411 return 0;
1412 }
1413
Sage Weil76e3b392009-10-15 18:13:53 -07001414 dout("__mark_dirty_caps %p %s dirty %s -> %s\n", &ci->vfs_inode,
1415 ceph_cap_string(mask), ceph_cap_string(was),
1416 ceph_cap_string(was | mask));
1417 ci->i_dirty_caps |= mask;
1418 if (was == 0) {
Sage Weil7d8cb262010-08-24 08:44:16 -07001419 if (!ci->i_head_snapc)
1420 ci->i_head_snapc = ceph_get_snap_context(
1421 ci->i_snap_realm->cached_context);
Yan, Zheng06852352012-11-19 10:49:07 +08001422 dout(" inode %p now dirty snapc %p auth cap %p\n",
1423 &ci->vfs_inode, ci->i_head_snapc, ci->i_auth_cap);
Sage Weil76e3b392009-10-15 18:13:53 -07001424 BUG_ON(!list_empty(&ci->i_dirty_item));
1425 spin_lock(&mdsc->cap_dirty_lock);
Yan, Zheng11df2df2013-11-24 14:44:38 +08001426 list_add(&ci->i_dirty_item, &mdsc->cap_dirty);
Sage Weil76e3b392009-10-15 18:13:53 -07001427 spin_unlock(&mdsc->cap_dirty_lock);
1428 if (ci->i_flushing_caps == 0) {
Sage Weil3772d262011-05-03 09:28:08 -07001429 ihold(inode);
Sage Weil76e3b392009-10-15 18:13:53 -07001430 dirty |= I_DIRTY_SYNC;
1431 }
1432 }
1433 BUG_ON(list_empty(&ci->i_dirty_item));
1434 if (((was | ci->i_flushing_caps) & CEPH_CAP_FILE_BUFFER) &&
1435 (mask & CEPH_CAP_FILE_BUFFER))
1436 dirty |= I_DIRTY_DATASYNC;
Sage Weil76e3b392009-10-15 18:13:53 -07001437 __cap_delay_requeue(mdsc, ci);
Sage Weilfca65b42011-05-04 11:33:47 -07001438 return dirty;
Sage Weil76e3b392009-10-15 18:13:53 -07001439}
1440
1441/*
Sage Weila8599bd2009-10-06 11:31:12 -07001442 * Add dirty inode to the flushing list. Assigned a seq number so we
1443 * can wait for caps to flush without starving.
Sage Weilcdc35f92009-10-14 14:24:19 -07001444 *
Sage Weilbe655592011-11-30 09:47:09 -08001445 * Called under i_ceph_lock.
Sage Weila8599bd2009-10-06 11:31:12 -07001446 */
Sage Weilcdc35f92009-10-14 14:24:19 -07001447static int __mark_caps_flushing(struct inode *inode,
Sage Weila8599bd2009-10-06 11:31:12 -07001448 struct ceph_mds_session *session)
1449{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001450 struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
Sage Weila8599bd2009-10-06 11:31:12 -07001451 struct ceph_inode_info *ci = ceph_inode(inode);
Sage Weilcdc35f92009-10-14 14:24:19 -07001452 int flushing;
Sage Weil50b885b2009-12-01 14:12:07 -08001453
Sage Weilcdc35f92009-10-14 14:24:19 -07001454 BUG_ON(ci->i_dirty_caps == 0);
Sage Weila8599bd2009-10-06 11:31:12 -07001455 BUG_ON(list_empty(&ci->i_dirty_item));
Sage Weilcdc35f92009-10-14 14:24:19 -07001456
1457 flushing = ci->i_dirty_caps;
1458 dout("__mark_caps_flushing flushing %s, flushing_caps %s -> %s\n",
1459 ceph_cap_string(flushing),
1460 ceph_cap_string(ci->i_flushing_caps),
1461 ceph_cap_string(ci->i_flushing_caps | flushing));
1462 ci->i_flushing_caps |= flushing;
1463 ci->i_dirty_caps = 0;
Sage Weilafcdaea2009-10-14 14:27:38 -07001464 dout(" inode %p now !dirty\n", inode);
Sage Weilcdc35f92009-10-14 14:24:19 -07001465
Sage Weila8599bd2009-10-06 11:31:12 -07001466 spin_lock(&mdsc->cap_dirty_lock);
Sage Weilafcdaea2009-10-14 14:27:38 -07001467 list_del_init(&ci->i_dirty_item);
1468
Sage Weila8599bd2009-10-06 11:31:12 -07001469 if (list_empty(&ci->i_flushing_item)) {
Yan, Zhengd3383a82015-01-08 21:30:12 +08001470 ci->i_cap_flush_seq = ++mdsc->cap_flush_seq;
Sage Weila8599bd2009-10-06 11:31:12 -07001471 list_add_tail(&ci->i_flushing_item, &session->s_cap_flushing);
1472 mdsc->num_cap_flushing++;
Sage Weilafcdaea2009-10-14 14:27:38 -07001473 dout(" inode %p now flushing seq %lld\n", inode,
1474 ci->i_cap_flush_seq);
1475 } else {
1476 list_move_tail(&ci->i_flushing_item, &session->s_cap_flushing);
1477 dout(" inode %p now flushing (more) seq %lld\n", inode,
Sage Weila8599bd2009-10-06 11:31:12 -07001478 ci->i_cap_flush_seq);
1479 }
1480 spin_unlock(&mdsc->cap_dirty_lock);
Sage Weilcdc35f92009-10-14 14:24:19 -07001481
1482 return flushing;
Sage Weila8599bd2009-10-06 11:31:12 -07001483}
1484
1485/*
Sage Weil5ecad6f2010-02-17 10:43:37 -08001486 * try to invalidate mapping pages without blocking.
1487 */
Sage Weil5ecad6f2010-02-17 10:43:37 -08001488static int try_nonblocking_invalidate(struct inode *inode)
1489{
1490 struct ceph_inode_info *ci = ceph_inode(inode);
1491 u32 invalidating_gen = ci->i_rdcache_gen;
1492
Sage Weilbe655592011-11-30 09:47:09 -08001493 spin_unlock(&ci->i_ceph_lock);
Sage Weil5ecad6f2010-02-17 10:43:37 -08001494 invalidate_mapping_pages(&inode->i_data, 0, -1);
Sage Weilbe655592011-11-30 09:47:09 -08001495 spin_lock(&ci->i_ceph_lock);
Sage Weil5ecad6f2010-02-17 10:43:37 -08001496
Sage Weil18a38192010-09-17 10:46:44 -07001497 if (inode->i_data.nrpages == 0 &&
Sage Weil5ecad6f2010-02-17 10:43:37 -08001498 invalidating_gen == ci->i_rdcache_gen) {
1499 /* success. */
1500 dout("try_nonblocking_invalidate %p success\n", inode);
Sage Weilcd045cb2010-11-04 11:05:05 -07001501 /* save any racing async invalidate some trouble */
1502 ci->i_rdcache_revoking = ci->i_rdcache_gen - 1;
Sage Weil5ecad6f2010-02-17 10:43:37 -08001503 return 0;
1504 }
1505 dout("try_nonblocking_invalidate %p failed\n", inode);
1506 return -1;
1507}
1508
1509/*
Sage Weila8599bd2009-10-06 11:31:12 -07001510 * Swiss army knife function to examine currently used and wanted
1511 * versus held caps. Release, flush, ack revoked caps to mds as
1512 * appropriate.
1513 *
1514 * CHECK_CAPS_NODELAY - caller is delayed work and we should not delay
1515 * cap release further.
1516 * CHECK_CAPS_AUTHONLY - we should only check the auth cap
1517 * CHECK_CAPS_FLUSH - we should flush any dirty caps immediately, without
1518 * further delay.
1519 */
1520void ceph_check_caps(struct ceph_inode_info *ci, int flags,
1521 struct ceph_mds_session *session)
1522{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001523 struct ceph_fs_client *fsc = ceph_inode_to_client(&ci->vfs_inode);
1524 struct ceph_mds_client *mdsc = fsc->mdsc;
Sage Weila8599bd2009-10-06 11:31:12 -07001525 struct inode *inode = &ci->vfs_inode;
1526 struct ceph_cap *cap;
Yan, Zheng395c3122013-01-04 14:37:57 +08001527 int file_wanted, used, cap_used;
Sage Weila8599bd2009-10-06 11:31:12 -07001528 int took_snap_rwsem = 0; /* true if mdsc->snap_rwsem held */
Sage Weilcbd03632010-02-09 13:41:18 -08001529 int issued, implemented, want, retain, revoking, flushing = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07001530 int mds = -1; /* keep track of how far we've gone through i_caps list
1531 to avoid an infinite loop on retry */
1532 struct rb_node *p;
1533 int tried_invalidate = 0;
1534 int delayed = 0, sent = 0, force_requeue = 0, num;
Sage Weilcbd03632010-02-09 13:41:18 -08001535 int queue_invalidate = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07001536 int is_delayed = flags & CHECK_CAPS_NODELAY;
1537
1538 /* if we are unmounting, flush any unused caps immediately. */
1539 if (mdsc->stopping)
1540 is_delayed = 1;
1541
Sage Weilbe655592011-11-30 09:47:09 -08001542 spin_lock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001543
1544 if (ci->i_ceph_flags & CEPH_I_FLUSH)
1545 flags |= CHECK_CAPS_FLUSH;
1546
1547 /* flush snaps first time around only */
1548 if (!list_empty(&ci->i_cap_snaps))
Sage Weile8351242010-09-17 08:03:08 -07001549 __ceph_flush_snaps(ci, &session, 0);
Sage Weila8599bd2009-10-06 11:31:12 -07001550 goto retry_locked;
1551retry:
Sage Weilbe655592011-11-30 09:47:09 -08001552 spin_lock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001553retry_locked:
1554 file_wanted = __ceph_caps_file_wanted(ci);
1555 used = __ceph_caps_used(ci);
1556 want = file_wanted | used;
Sage Weilcbd03632010-02-09 13:41:18 -08001557 issued = __ceph_caps_issued(ci, &implemented);
1558 revoking = implemented & ~issued;
Sage Weila8599bd2009-10-06 11:31:12 -07001559
1560 retain = want | CEPH_CAP_PIN;
1561 if (!mdsc->stopping && inode->i_nlink > 0) {
1562 if (want) {
1563 retain |= CEPH_CAP_ANY; /* be greedy */
Yan, Zheng32ec4392015-03-26 19:06:00 +08001564 } else if (S_ISDIR(inode->i_mode) &&
1565 (issued & CEPH_CAP_FILE_SHARED) &&
1566 __ceph_dir_is_complete(ci)) {
1567 /*
1568 * If a directory is complete, we want to keep
1569 * the exclusive cap. So that MDS does not end up
1570 * revoking the shared cap on every create/unlink
1571 * operation.
1572 */
1573 want = CEPH_CAP_ANY_SHARED | CEPH_CAP_FILE_EXCL;
1574 retain |= want;
Sage Weila8599bd2009-10-06 11:31:12 -07001575 } else {
Yan, Zheng32ec4392015-03-26 19:06:00 +08001576
Sage Weila8599bd2009-10-06 11:31:12 -07001577 retain |= CEPH_CAP_ANY_SHARED;
1578 /*
1579 * keep RD only if we didn't have the file open RW,
1580 * because then the mds would revoke it anyway to
1581 * journal max_size=0.
1582 */
1583 if (ci->i_max_size == 0)
1584 retain |= CEPH_CAP_ANY_RD;
1585 }
1586 }
1587
1588 dout("check_caps %p file_want %s used %s dirty %s flushing %s"
Sage Weilcbd03632010-02-09 13:41:18 -08001589 " issued %s revoking %s retain %s %s%s%s\n", inode,
Sage Weila8599bd2009-10-06 11:31:12 -07001590 ceph_cap_string(file_wanted),
1591 ceph_cap_string(used), ceph_cap_string(ci->i_dirty_caps),
1592 ceph_cap_string(ci->i_flushing_caps),
Sage Weilcbd03632010-02-09 13:41:18 -08001593 ceph_cap_string(issued), ceph_cap_string(revoking),
Sage Weila8599bd2009-10-06 11:31:12 -07001594 ceph_cap_string(retain),
1595 (flags & CHECK_CAPS_AUTHONLY) ? " AUTHONLY" : "",
1596 (flags & CHECK_CAPS_NODELAY) ? " NODELAY" : "",
1597 (flags & CHECK_CAPS_FLUSH) ? " FLUSH" : "");
1598
1599 /*
1600 * If we no longer need to hold onto old our caps, and we may
1601 * have cached pages, but don't want them, then try to invalidate.
1602 * If we fail, it's because pages are locked.... try again later.
1603 */
1604 if ((!is_delayed || mdsc->stopping) &&
1605 ci->i_wrbuffer_ref == 0 && /* no dirty pages... */
Sage Weil93afd442010-09-17 08:38:25 -07001606 inode->i_data.nrpages && /* have cached pages */
Sage Weilcbd03632010-02-09 13:41:18 -08001607 (file_wanted == 0 || /* no open files */
Sage Weil29625072010-05-27 10:40:43 -07001608 (revoking & (CEPH_CAP_FILE_CACHE|
1609 CEPH_CAP_FILE_LAZYIO))) && /* or revoking cache */
Sage Weila8599bd2009-10-06 11:31:12 -07001610 !tried_invalidate) {
Sage Weila8599bd2009-10-06 11:31:12 -07001611 dout("check_caps trying to invalidate on %p\n", inode);
Sage Weil5ecad6f2010-02-17 10:43:37 -08001612 if (try_nonblocking_invalidate(inode) < 0) {
Sage Weil29625072010-05-27 10:40:43 -07001613 if (revoking & (CEPH_CAP_FILE_CACHE|
1614 CEPH_CAP_FILE_LAZYIO)) {
Sage Weil5ecad6f2010-02-17 10:43:37 -08001615 dout("check_caps queuing invalidate\n");
1616 queue_invalidate = 1;
1617 ci->i_rdcache_revoking = ci->i_rdcache_gen;
1618 } else {
1619 dout("check_caps failed to invalidate pages\n");
1620 /* we failed to invalidate pages. check these
1621 caps again later. */
1622 force_requeue = 1;
1623 __cap_set_timeouts(mdsc, ci);
1624 }
Sage Weila8599bd2009-10-06 11:31:12 -07001625 }
1626 tried_invalidate = 1;
1627 goto retry_locked;
1628 }
1629
1630 num = 0;
1631 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
1632 cap = rb_entry(p, struct ceph_cap, ci_node);
1633 num++;
1634
1635 /* avoid looping forever */
1636 if (mds >= cap->mds ||
1637 ((flags & CHECK_CAPS_AUTHONLY) && cap != ci->i_auth_cap))
1638 continue;
1639
1640 /* NOTE: no side-effects allowed, until we take s_mutex */
1641
Yan, Zheng395c3122013-01-04 14:37:57 +08001642 cap_used = used;
1643 if (ci->i_auth_cap && cap != ci->i_auth_cap)
1644 cap_used &= ~ci->i_auth_cap->issued;
1645
Sage Weila8599bd2009-10-06 11:31:12 -07001646 revoking = cap->implemented & ~cap->issued;
Yan, Zheng395c3122013-01-04 14:37:57 +08001647 dout(" mds%d cap %p used %s issued %s implemented %s revoking %s\n",
Sage Weil088b3f52011-01-18 08:56:01 -08001648 cap->mds, cap, ceph_cap_string(cap->issued),
Yan, Zheng395c3122013-01-04 14:37:57 +08001649 ceph_cap_string(cap_used),
Sage Weil088b3f52011-01-18 08:56:01 -08001650 ceph_cap_string(cap->implemented),
1651 ceph_cap_string(revoking));
Sage Weila8599bd2009-10-06 11:31:12 -07001652
1653 if (cap == ci->i_auth_cap &&
1654 (cap->issued & CEPH_CAP_FILE_WR)) {
1655 /* request larger max_size from MDS? */
1656 if (ci->i_wanted_max_size > ci->i_max_size &&
1657 ci->i_wanted_max_size > ci->i_requested_max_size) {
1658 dout("requesting new max_size\n");
1659 goto ack;
1660 }
1661
1662 /* approaching file_max? */
1663 if ((inode->i_size << 1) >= ci->i_max_size &&
1664 (ci->i_reported_size << 1) < ci->i_max_size) {
1665 dout("i_size approaching max_size\n");
1666 goto ack;
1667 }
1668 }
1669 /* flush anything dirty? */
1670 if (cap == ci->i_auth_cap && (flags & CHECK_CAPS_FLUSH) &&
1671 ci->i_dirty_caps) {
1672 dout("flushing dirty caps\n");
1673 goto ack;
1674 }
1675
1676 /* completed revocation? going down and there are no caps? */
Yan, Zheng395c3122013-01-04 14:37:57 +08001677 if (revoking && (revoking & cap_used) == 0) {
Sage Weila8599bd2009-10-06 11:31:12 -07001678 dout("completed revocation of %s\n",
1679 ceph_cap_string(cap->implemented & ~cap->issued));
1680 goto ack;
1681 }
1682
1683 /* want more caps from mds? */
1684 if (want & ~(cap->mds_wanted | cap->issued))
1685 goto ack;
1686
1687 /* things we might delay */
1688 if ((cap->issued & ~retain) == 0 &&
1689 cap->mds_wanted == want)
1690 continue; /* nope, all good */
1691
1692 if (is_delayed)
1693 goto ack;
1694
1695 /* delay? */
1696 if ((ci->i_ceph_flags & CEPH_I_NODELAY) == 0 &&
1697 time_before(jiffies, ci->i_hold_caps_max)) {
1698 dout(" delaying issued %s -> %s, wanted %s -> %s\n",
1699 ceph_cap_string(cap->issued),
1700 ceph_cap_string(cap->issued & retain),
1701 ceph_cap_string(cap->mds_wanted),
1702 ceph_cap_string(want));
1703 delayed++;
1704 continue;
1705 }
1706
1707ack:
Sage Weile9964c12010-03-01 15:16:56 -08001708 if (ci->i_ceph_flags & CEPH_I_NOFLUSH) {
1709 dout(" skipping %p I_NOFLUSH set\n", inode);
1710 continue;
1711 }
1712
Sage Weila8599bd2009-10-06 11:31:12 -07001713 if (session && session != cap->session) {
1714 dout("oops, wrong session %p mutex\n", session);
1715 mutex_unlock(&session->s_mutex);
1716 session = NULL;
1717 }
1718 if (!session) {
1719 session = cap->session;
1720 if (mutex_trylock(&session->s_mutex) == 0) {
1721 dout("inverting session/ino locks on %p\n",
1722 session);
Sage Weilbe655592011-11-30 09:47:09 -08001723 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001724 if (took_snap_rwsem) {
1725 up_read(&mdsc->snap_rwsem);
1726 took_snap_rwsem = 0;
1727 }
1728 mutex_lock(&session->s_mutex);
1729 goto retry;
1730 }
1731 }
1732 /* take snap_rwsem after session mutex */
1733 if (!took_snap_rwsem) {
1734 if (down_read_trylock(&mdsc->snap_rwsem) == 0) {
1735 dout("inverting snap/in locks on %p\n",
1736 inode);
Sage Weilbe655592011-11-30 09:47:09 -08001737 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001738 down_read(&mdsc->snap_rwsem);
1739 took_snap_rwsem = 1;
1740 goto retry;
1741 }
1742 took_snap_rwsem = 1;
1743 }
1744
Sage Weilcdc35f92009-10-14 14:24:19 -07001745 if (cap == ci->i_auth_cap && ci->i_dirty_caps)
1746 flushing = __mark_caps_flushing(inode, session);
Sage Weil24be0c42011-01-18 08:48:06 -08001747 else
1748 flushing = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07001749
1750 mds = cap->mds; /* remember mds, so we don't repeat */
1751 sent++;
1752
Sage Weilbe655592011-11-30 09:47:09 -08001753 /* __send_cap drops i_ceph_lock */
Yan, Zheng395c3122013-01-04 14:37:57 +08001754 delayed += __send_cap(mdsc, cap, CEPH_CAP_OP_UPDATE, cap_used,
1755 want, retain, flushing, NULL);
Sage Weilbe655592011-11-30 09:47:09 -08001756 goto retry; /* retake i_ceph_lock and restart our cap scan. */
Sage Weila8599bd2009-10-06 11:31:12 -07001757 }
1758
1759 /*
1760 * Reschedule delayed caps release if we delayed anything,
1761 * otherwise cancel.
1762 */
1763 if (delayed && is_delayed)
1764 force_requeue = 1; /* __send_cap delayed release; requeue */
1765 if (!delayed && !is_delayed)
1766 __cap_delay_cancel(mdsc, ci);
1767 else if (!is_delayed || force_requeue)
1768 __cap_delay_requeue(mdsc, ci);
1769
Sage Weilbe655592011-11-30 09:47:09 -08001770 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001771
Sage Weilcbd03632010-02-09 13:41:18 -08001772 if (queue_invalidate)
Sage Weil3c6f6b72010-02-09 15:24:44 -08001773 ceph_queue_invalidate(inode);
Sage Weilcbd03632010-02-09 13:41:18 -08001774
Sage Weilcdc2ce02010-03-16 13:39:28 -07001775 if (session)
Sage Weila8599bd2009-10-06 11:31:12 -07001776 mutex_unlock(&session->s_mutex);
1777 if (took_snap_rwsem)
1778 up_read(&mdsc->snap_rwsem);
1779}
1780
1781/*
Sage Weila8599bd2009-10-06 11:31:12 -07001782 * Try to flush dirty caps back to the auth mds.
1783 */
Yan, Zheng4fe59782013-10-31 16:44:14 +08001784static int try_flush_caps(struct inode *inode, unsigned *flush_tid)
Sage Weila8599bd2009-10-06 11:31:12 -07001785{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001786 struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
Sage Weila8599bd2009-10-06 11:31:12 -07001787 struct ceph_inode_info *ci = ceph_inode(inode);
Sage Weila8599bd2009-10-06 11:31:12 -07001788 int flushing = 0;
Yan, Zheng4fe59782013-10-31 16:44:14 +08001789 struct ceph_mds_session *session = NULL;
Sage Weila8599bd2009-10-06 11:31:12 -07001790
1791retry:
Sage Weilbe655592011-11-30 09:47:09 -08001792 spin_lock(&ci->i_ceph_lock);
Sage Weile9964c12010-03-01 15:16:56 -08001793 if (ci->i_ceph_flags & CEPH_I_NOFLUSH) {
1794 dout("try_flush_caps skipping %p I_NOFLUSH set\n", inode);
1795 goto out;
1796 }
Sage Weila8599bd2009-10-06 11:31:12 -07001797 if (ci->i_dirty_caps && ci->i_auth_cap) {
1798 struct ceph_cap *cap = ci->i_auth_cap;
1799 int used = __ceph_caps_used(ci);
1800 int want = __ceph_caps_wanted(ci);
1801 int delayed;
1802
Yan, Zheng4fe59782013-10-31 16:44:14 +08001803 if (!session || session != cap->session) {
Sage Weilbe655592011-11-30 09:47:09 -08001804 spin_unlock(&ci->i_ceph_lock);
Yan, Zheng4fe59782013-10-31 16:44:14 +08001805 if (session)
1806 mutex_unlock(&session->s_mutex);
Sage Weila8599bd2009-10-06 11:31:12 -07001807 session = cap->session;
1808 mutex_lock(&session->s_mutex);
1809 goto retry;
1810 }
Sage Weila8599bd2009-10-06 11:31:12 -07001811 if (cap->session->s_state < CEPH_MDS_SESSION_OPEN)
1812 goto out;
1813
Sage Weilcdc35f92009-10-14 14:24:19 -07001814 flushing = __mark_caps_flushing(inode, session);
Sage Weila8599bd2009-10-06 11:31:12 -07001815
Sage Weilbe655592011-11-30 09:47:09 -08001816 /* __send_cap drops i_ceph_lock */
Sage Weila8599bd2009-10-06 11:31:12 -07001817 delayed = __send_cap(mdsc, cap, CEPH_CAP_OP_FLUSH, used, want,
1818 cap->issued | cap->implemented, flushing,
1819 flush_tid);
1820 if (!delayed)
1821 goto out_unlocked;
1822
Sage Weilbe655592011-11-30 09:47:09 -08001823 spin_lock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001824 __cap_delay_requeue(mdsc, ci);
1825 }
1826out:
Sage Weilbe655592011-11-30 09:47:09 -08001827 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001828out_unlocked:
Yan, Zheng4fe59782013-10-31 16:44:14 +08001829 if (session)
Sage Weila8599bd2009-10-06 11:31:12 -07001830 mutex_unlock(&session->s_mutex);
1831 return flushing;
1832}
1833
1834/*
1835 * Return true if we've flushed caps through the given flush_tid.
1836 */
1837static int caps_are_flushed(struct inode *inode, unsigned tid)
1838{
1839 struct ceph_inode_info *ci = ceph_inode(inode);
Dan Carpentera5ee7512010-05-07 10:27:14 +02001840 int i, ret = 1;
Sage Weila8599bd2009-10-06 11:31:12 -07001841
Sage Weilbe655592011-11-30 09:47:09 -08001842 spin_lock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001843 for (i = 0; i < CEPH_CAP_BITS; i++)
1844 if ((ci->i_flushing_caps & (1 << i)) &&
1845 ci->i_cap_flush_tid[i] <= tid) {
1846 /* still flushing this bit */
1847 ret = 0;
1848 break;
1849 }
Sage Weilbe655592011-11-30 09:47:09 -08001850 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001851 return ret;
1852}
1853
1854/*
1855 * Wait on any unsafe replies for the given inode. First wait on the
1856 * newest request, and make that the upper bound. Then, if there are
1857 * more requests, keep waiting on the oldest as long as it is still older
1858 * than the original request.
1859 */
1860static void sync_write_wait(struct inode *inode)
1861{
1862 struct ceph_inode_info *ci = ceph_inode(inode);
1863 struct list_head *head = &ci->i_unsafe_writes;
1864 struct ceph_osd_request *req;
1865 u64 last_tid;
1866
1867 spin_lock(&ci->i_unsafe_lock);
1868 if (list_empty(head))
1869 goto out;
1870
1871 /* set upper bound as _last_ entry in chain */
1872 req = list_entry(head->prev, struct ceph_osd_request,
1873 r_unsafe_item);
1874 last_tid = req->r_tid;
1875
1876 do {
1877 ceph_osdc_get_request(req);
1878 spin_unlock(&ci->i_unsafe_lock);
1879 dout("sync_write_wait on tid %llu (until %llu)\n",
1880 req->r_tid, last_tid);
1881 wait_for_completion(&req->r_safe_completion);
1882 spin_lock(&ci->i_unsafe_lock);
1883 ceph_osdc_put_request(req);
1884
1885 /*
1886 * from here on look at first entry in chain, since we
1887 * only want to wait for anything older than last_tid
1888 */
1889 if (list_empty(head))
1890 break;
1891 req = list_entry(head->next, struct ceph_osd_request,
1892 r_unsafe_item);
1893 } while (req->r_tid < last_tid);
1894out:
1895 spin_unlock(&ci->i_unsafe_lock);
1896}
1897
Josef Bacik02c24a82011-07-16 20:44:56 -04001898int ceph_fsync(struct file *file, loff_t start, loff_t end, int datasync)
Sage Weila8599bd2009-10-06 11:31:12 -07001899{
Christoph Hellwig7ea80852010-05-26 17:53:25 +02001900 struct inode *inode = file->f_mapping->host;
Sage Weila8599bd2009-10-06 11:31:12 -07001901 struct ceph_inode_info *ci = ceph_inode(inode);
1902 unsigned flush_tid;
1903 int ret;
1904 int dirty;
1905
1906 dout("fsync %p%s\n", inode, datasync ? " datasync" : "");
1907 sync_write_wait(inode);
1908
Josef Bacik02c24a82011-07-16 20:44:56 -04001909 ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
Sage Weila8599bd2009-10-06 11:31:12 -07001910 if (ret < 0)
1911 return ret;
Josef Bacik02c24a82011-07-16 20:44:56 -04001912 mutex_lock(&inode->i_mutex);
Sage Weila8599bd2009-10-06 11:31:12 -07001913
Yan, Zheng4fe59782013-10-31 16:44:14 +08001914 dirty = try_flush_caps(inode, &flush_tid);
Sage Weila8599bd2009-10-06 11:31:12 -07001915 dout("fsync dirty caps are %s\n", ceph_cap_string(dirty));
1916
1917 /*
1918 * only wait on non-file metadata writeback (the mds
1919 * can recover size and mtime, so we don't need to
1920 * wait for that)
1921 */
1922 if (!datasync && (dirty & ~CEPH_CAP_ANY_FILE_WR)) {
1923 dout("fsync waiting for flush_tid %u\n", flush_tid);
1924 ret = wait_event_interruptible(ci->i_cap_wq,
1925 caps_are_flushed(inode, flush_tid));
1926 }
1927
1928 dout("fsync %p%s done\n", inode, datasync ? " datasync" : "");
Josef Bacik02c24a82011-07-16 20:44:56 -04001929 mutex_unlock(&inode->i_mutex);
Sage Weila8599bd2009-10-06 11:31:12 -07001930 return ret;
1931}
1932
1933/*
1934 * Flush any dirty caps back to the mds. If we aren't asked to wait,
1935 * queue inode for flush but don't do so immediately, because we can
1936 * get by with fewer MDS messages if we wait for data writeback to
1937 * complete first.
1938 */
Stephen Rothwellf1a3d572010-01-18 11:53:08 +11001939int ceph_write_inode(struct inode *inode, struct writeback_control *wbc)
Sage Weila8599bd2009-10-06 11:31:12 -07001940{
1941 struct ceph_inode_info *ci = ceph_inode(inode);
1942 unsigned flush_tid;
1943 int err = 0;
1944 int dirty;
Stephen Rothwellf1a3d572010-01-18 11:53:08 +11001945 int wait = wbc->sync_mode == WB_SYNC_ALL;
Sage Weila8599bd2009-10-06 11:31:12 -07001946
1947 dout("write_inode %p wait=%d\n", inode, wait);
1948 if (wait) {
Yan, Zheng4fe59782013-10-31 16:44:14 +08001949 dirty = try_flush_caps(inode, &flush_tid);
Sage Weila8599bd2009-10-06 11:31:12 -07001950 if (dirty)
1951 err = wait_event_interruptible(ci->i_cap_wq,
1952 caps_are_flushed(inode, flush_tid));
1953 } else {
Cheng Renquan640ef792010-03-26 17:40:33 +08001954 struct ceph_mds_client *mdsc =
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001955 ceph_sb_to_client(inode->i_sb)->mdsc;
Sage Weila8599bd2009-10-06 11:31:12 -07001956
Sage Weilbe655592011-11-30 09:47:09 -08001957 spin_lock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001958 if (__ceph_caps_dirty(ci))
1959 __cap_delay_requeue_front(mdsc, ci);
Sage Weilbe655592011-11-30 09:47:09 -08001960 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001961 }
1962 return err;
1963}
1964
1965/*
1966 * After a recovering MDS goes active, we need to resend any caps
1967 * we were flushing.
1968 *
1969 * Caller holds session->s_mutex.
1970 */
1971static void kick_flushing_capsnaps(struct ceph_mds_client *mdsc,
1972 struct ceph_mds_session *session)
1973{
1974 struct ceph_cap_snap *capsnap;
1975
1976 dout("kick_flushing_capsnaps mds%d\n", session->s_mds);
1977 list_for_each_entry(capsnap, &session->s_cap_snaps_flushing,
1978 flushing_item) {
1979 struct ceph_inode_info *ci = capsnap->ci;
1980 struct inode *inode = &ci->vfs_inode;
1981 struct ceph_cap *cap;
1982
Sage Weilbe655592011-11-30 09:47:09 -08001983 spin_lock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001984 cap = ci->i_auth_cap;
1985 if (cap && cap->session == session) {
1986 dout("kick_flushing_caps %p cap %p capsnap %p\n", inode,
1987 cap, capsnap);
Sage Weile8351242010-09-17 08:03:08 -07001988 __ceph_flush_snaps(ci, &session, 1);
Sage Weila8599bd2009-10-06 11:31:12 -07001989 } else {
1990 pr_err("%p auth cap %p not mds%d ???\n", inode,
1991 cap, session->s_mds);
Sage Weila8599bd2009-10-06 11:31:12 -07001992 }
Sage Weilbe655592011-11-30 09:47:09 -08001993 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001994 }
1995}
1996
1997void ceph_kick_flushing_caps(struct ceph_mds_client *mdsc,
1998 struct ceph_mds_session *session)
1999{
2000 struct ceph_inode_info *ci;
2001
2002 kick_flushing_capsnaps(mdsc, session);
2003
2004 dout("kick_flushing_caps mds%d\n", session->s_mds);
2005 list_for_each_entry(ci, &session->s_cap_flushing, i_flushing_item) {
2006 struct inode *inode = &ci->vfs_inode;
2007 struct ceph_cap *cap;
2008 int delayed = 0;
2009
Sage Weilbe655592011-11-30 09:47:09 -08002010 spin_lock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07002011 cap = ci->i_auth_cap;
2012 if (cap && cap->session == session) {
2013 dout("kick_flushing_caps %p cap %p %s\n", inode,
2014 cap, ceph_cap_string(ci->i_flushing_caps));
2015 delayed = __send_cap(mdsc, cap, CEPH_CAP_OP_FLUSH,
2016 __ceph_caps_used(ci),
2017 __ceph_caps_wanted(ci),
2018 cap->issued | cap->implemented,
2019 ci->i_flushing_caps, NULL);
2020 if (delayed) {
Sage Weilbe655592011-11-30 09:47:09 -08002021 spin_lock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07002022 __cap_delay_requeue(mdsc, ci);
Sage Weilbe655592011-11-30 09:47:09 -08002023 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07002024 }
2025 } else {
2026 pr_err("%p auth cap %p not mds%d ???\n", inode,
2027 cap, session->s_mds);
Sage Weilbe655592011-11-30 09:47:09 -08002028 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07002029 }
2030 }
2031}
2032
Sage Weil088b3f52011-01-18 08:56:01 -08002033static void kick_flushing_inode_caps(struct ceph_mds_client *mdsc,
2034 struct ceph_mds_session *session,
2035 struct inode *inode)
2036{
2037 struct ceph_inode_info *ci = ceph_inode(inode);
2038 struct ceph_cap *cap;
2039 int delayed = 0;
2040
Sage Weilbe655592011-11-30 09:47:09 -08002041 spin_lock(&ci->i_ceph_lock);
Sage Weil088b3f52011-01-18 08:56:01 -08002042 cap = ci->i_auth_cap;
2043 dout("kick_flushing_inode_caps %p flushing %s flush_seq %lld\n", inode,
2044 ceph_cap_string(ci->i_flushing_caps), ci->i_cap_flush_seq);
Yan, Zheng005c4692013-05-31 16:40:24 +08002045
Sage Weil088b3f52011-01-18 08:56:01 -08002046 __ceph_flush_snaps(ci, &session, 1);
Yan, Zheng005c4692013-05-31 16:40:24 +08002047
Sage Weil088b3f52011-01-18 08:56:01 -08002048 if (ci->i_flushing_caps) {
Yan, Zheng005c4692013-05-31 16:40:24 +08002049 spin_lock(&mdsc->cap_dirty_lock);
2050 list_move_tail(&ci->i_flushing_item,
2051 &cap->session->s_cap_flushing);
2052 spin_unlock(&mdsc->cap_dirty_lock);
2053
Sage Weil088b3f52011-01-18 08:56:01 -08002054 delayed = __send_cap(mdsc, cap, CEPH_CAP_OP_FLUSH,
2055 __ceph_caps_used(ci),
2056 __ceph_caps_wanted(ci),
2057 cap->issued | cap->implemented,
2058 ci->i_flushing_caps, NULL);
2059 if (delayed) {
Sage Weilbe655592011-11-30 09:47:09 -08002060 spin_lock(&ci->i_ceph_lock);
Sage Weil088b3f52011-01-18 08:56:01 -08002061 __cap_delay_requeue(mdsc, ci);
Sage Weilbe655592011-11-30 09:47:09 -08002062 spin_unlock(&ci->i_ceph_lock);
Sage Weil088b3f52011-01-18 08:56:01 -08002063 }
2064 } else {
Sage Weilbe655592011-11-30 09:47:09 -08002065 spin_unlock(&ci->i_ceph_lock);
Sage Weil088b3f52011-01-18 08:56:01 -08002066 }
2067}
2068
Sage Weila8599bd2009-10-06 11:31:12 -07002069
2070/*
2071 * Take references to capabilities we hold, so that we don't release
2072 * them to the MDS prematurely.
2073 *
Sage Weilbe655592011-11-30 09:47:09 -08002074 * Protected by i_ceph_lock.
Sage Weila8599bd2009-10-06 11:31:12 -07002075 */
2076static void __take_cap_refs(struct ceph_inode_info *ci, int got)
2077{
2078 if (got & CEPH_CAP_PIN)
2079 ci->i_pin_ref++;
2080 if (got & CEPH_CAP_FILE_RD)
2081 ci->i_rd_ref++;
2082 if (got & CEPH_CAP_FILE_CACHE)
2083 ci->i_rdcache_ref++;
2084 if (got & CEPH_CAP_FILE_WR)
2085 ci->i_wr_ref++;
2086 if (got & CEPH_CAP_FILE_BUFFER) {
Henry C Changd3d07202011-05-11 10:29:54 +00002087 if (ci->i_wb_ref == 0)
Sage Weil3772d262011-05-03 09:28:08 -07002088 ihold(&ci->vfs_inode);
Henry C Changd3d07202011-05-11 10:29:54 +00002089 ci->i_wb_ref++;
2090 dout("__take_cap_refs %p wb %d -> %d (?)\n",
2091 &ci->vfs_inode, ci->i_wb_ref-1, ci->i_wb_ref);
Sage Weila8599bd2009-10-06 11:31:12 -07002092 }
2093}
2094
2095/*
2096 * Try to grab cap references. Specify those refs we @want, and the
2097 * minimal set we @need. Also include the larger offset we are writing
2098 * to (when applicable), and check against max_size here as well.
2099 * Note that caller is responsible for ensuring max_size increases are
2100 * requested from the MDS.
2101 */
2102static int try_get_cap_refs(struct ceph_inode_info *ci, int need, int want,
Yan, Zhengc4d4a582015-01-09 15:56:18 +08002103 loff_t endoff, int *got, int *check_max, int *err)
Sage Weila8599bd2009-10-06 11:31:12 -07002104{
2105 struct inode *inode = &ci->vfs_inode;
2106 int ret = 0;
Yan, Zhengc4d4a582015-01-09 15:56:18 +08002107 int have, implemented;
Sage Weil195d3ce2010-03-01 09:57:54 -08002108 int file_wanted;
Sage Weila8599bd2009-10-06 11:31:12 -07002109
2110 dout("get_cap_refs %p need %s want %s\n", inode,
2111 ceph_cap_string(need), ceph_cap_string(want));
Yan, Zhengc4d4a582015-01-09 15:56:18 +08002112
Sage Weilbe655592011-11-30 09:47:09 -08002113 spin_lock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07002114
Sage Weil195d3ce2010-03-01 09:57:54 -08002115 /* make sure file is actually open */
2116 file_wanted = __ceph_caps_file_wanted(ci);
2117 if ((file_wanted & need) == 0) {
2118 dout("try_get_cap_refs need %s file_wanted %s, EBADF\n",
2119 ceph_cap_string(need), ceph_cap_string(file_wanted));
Sage Weila8599bd2009-10-06 11:31:12 -07002120 *err = -EBADF;
2121 ret = 1;
Yan, Zheng3738daa2014-11-14 22:10:07 +08002122 goto out_unlock;
Sage Weila8599bd2009-10-06 11:31:12 -07002123 }
2124
Yan, Zheng37505d52013-04-12 16:11:10 +08002125 /* finish pending truncate */
2126 while (ci->i_truncate_pending) {
2127 spin_unlock(&ci->i_ceph_lock);
Yan, Zhengb415bf42013-07-02 12:40:19 +08002128 __ceph_do_pending_vmtruncate(inode);
Yan, Zheng37505d52013-04-12 16:11:10 +08002129 spin_lock(&ci->i_ceph_lock);
2130 }
2131
Yan, Zheng3871cbb2013-08-05 14:10:29 +08002132 have = __ceph_caps_issued(ci, &implemented);
2133
2134 if (have & need & CEPH_CAP_FILE_WR) {
Sage Weila8599bd2009-10-06 11:31:12 -07002135 if (endoff >= 0 && endoff > (loff_t)ci->i_max_size) {
2136 dout("get_cap_refs %p endoff %llu > maxsize %llu\n",
2137 inode, endoff, ci->i_max_size);
Yan, Zheng3871cbb2013-08-05 14:10:29 +08002138 if (endoff > ci->i_requested_max_size) {
Sage Weila8599bd2009-10-06 11:31:12 -07002139 *check_max = 1;
2140 ret = 1;
2141 }
Yan, Zheng3738daa2014-11-14 22:10:07 +08002142 goto out_unlock;
Sage Weila8599bd2009-10-06 11:31:12 -07002143 }
2144 /*
2145 * If a sync write is in progress, we must wait, so that we
2146 * can get a final snapshot value for size+mtime.
2147 */
2148 if (__ceph_have_pending_cap_snap(ci)) {
2149 dout("get_cap_refs %p cap_snap_pending\n", inode);
Yan, Zheng3738daa2014-11-14 22:10:07 +08002150 goto out_unlock;
Sage Weila8599bd2009-10-06 11:31:12 -07002151 }
2152 }
Sage Weila8599bd2009-10-06 11:31:12 -07002153
Sage Weila8599bd2009-10-06 11:31:12 -07002154 if ((have & need) == need) {
2155 /*
2156 * Look at (implemented & ~have & not) so that we keep waiting
2157 * on transition from wanted -> needed caps. This is needed
2158 * for WRBUFFER|WR -> WR to avoid a new WR sync write from
2159 * going before a prior buffered writeback happens.
2160 */
2161 int not = want & ~(have & need);
2162 int revoking = implemented & ~have;
2163 dout("get_cap_refs %p have %s but not %s (revoking %s)\n",
2164 inode, ceph_cap_string(have), ceph_cap_string(not),
2165 ceph_cap_string(revoking));
2166 if ((revoking & not) == 0) {
Yan, Zhengc4d4a582015-01-09 15:56:18 +08002167 *got = need | (have & want);
2168 __take_cap_refs(ci, *got);
Sage Weila8599bd2009-10-06 11:31:12 -07002169 ret = 1;
2170 }
2171 } else {
Yan, Zheng03f4fcb2015-01-05 11:04:04 +08002172 int session_readonly = false;
2173 if ((need & CEPH_CAP_FILE_WR) && ci->i_auth_cap) {
2174 struct ceph_mds_session *s = ci->i_auth_cap->session;
2175 spin_lock(&s->s_cap_lock);
2176 session_readonly = s->s_readonly;
2177 spin_unlock(&s->s_cap_lock);
2178 }
2179 if (session_readonly) {
2180 dout("get_cap_refs %p needed %s but mds%d readonly\n",
2181 inode, ceph_cap_string(need), ci->i_auth_cap->mds);
2182 *err = -EROFS;
2183 ret = 1;
2184 goto out_unlock;
2185 }
2186
Sage Weila8599bd2009-10-06 11:31:12 -07002187 dout("get_cap_refs %p have %s needed %s\n", inode,
2188 ceph_cap_string(have), ceph_cap_string(need));
2189 }
Yan, Zheng3738daa2014-11-14 22:10:07 +08002190out_unlock:
Sage Weilbe655592011-11-30 09:47:09 -08002191 spin_unlock(&ci->i_ceph_lock);
Yan, Zheng3738daa2014-11-14 22:10:07 +08002192
Sage Weila8599bd2009-10-06 11:31:12 -07002193 dout("get_cap_refs %p ret %d got %s\n", inode,
Yan, Zhengc4d4a582015-01-09 15:56:18 +08002194 ret, ceph_cap_string(*got));
Sage Weila8599bd2009-10-06 11:31:12 -07002195 return ret;
2196}
2197
2198/*
2199 * Check the offset we are writing up to against our current
2200 * max_size. If necessary, tell the MDS we want to write to
2201 * a larger offset.
2202 */
2203static void check_max_size(struct inode *inode, loff_t endoff)
2204{
2205 struct ceph_inode_info *ci = ceph_inode(inode);
2206 int check = 0;
2207
2208 /* do we need to explicitly request a larger max_size? */
Sage Weilbe655592011-11-30 09:47:09 -08002209 spin_lock(&ci->i_ceph_lock);
Yan, Zheng3871cbb2013-08-05 14:10:29 +08002210 if (endoff >= ci->i_max_size && endoff > ci->i_wanted_max_size) {
Sage Weila8599bd2009-10-06 11:31:12 -07002211 dout("write %p at large endoff %llu, req max_size\n",
2212 inode, endoff);
2213 ci->i_wanted_max_size = endoff;
Sage Weila8599bd2009-10-06 11:31:12 -07002214 }
Yan, Zheng3871cbb2013-08-05 14:10:29 +08002215 /* duplicate ceph_check_caps()'s logic */
2216 if (ci->i_auth_cap &&
2217 (ci->i_auth_cap->issued & CEPH_CAP_FILE_WR) &&
2218 ci->i_wanted_max_size > ci->i_max_size &&
2219 ci->i_wanted_max_size > ci->i_requested_max_size)
2220 check = 1;
Sage Weilbe655592011-11-30 09:47:09 -08002221 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07002222 if (check)
2223 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
2224}
2225
2226/*
2227 * Wait for caps, and take cap references. If we can't get a WR cap
2228 * due to a small max_size, make sure we check_max_size (and possibly
2229 * ask the mds) so we don't get hung up indefinitely.
2230 */
Yan, Zheng3738daa2014-11-14 22:10:07 +08002231int ceph_get_caps(struct ceph_inode_info *ci, int need, int want,
2232 loff_t endoff, int *got, struct page **pinned_page)
Sage Weila8599bd2009-10-06 11:31:12 -07002233{
Yan, Zhengc4d4a582015-01-09 15:56:18 +08002234 int _got, check_max, ret, err = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07002235
2236retry:
2237 if (endoff > 0)
2238 check_max_size(&ci->vfs_inode, endoff);
Yan, Zhengc4d4a582015-01-09 15:56:18 +08002239 _got = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07002240 check_max = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07002241 ret = wait_event_interruptible(ci->i_cap_wq,
Yan, Zhengc4d4a582015-01-09 15:56:18 +08002242 try_get_cap_refs(ci, need, want, endoff,
2243 &_got, &check_max, &err));
Sage Weila8599bd2009-10-06 11:31:12 -07002244 if (err)
2245 ret = err;
Yan, Zhengc4d4a582015-01-09 15:56:18 +08002246 if (ret < 0)
2247 return ret;
2248
Sage Weila8599bd2009-10-06 11:31:12 -07002249 if (check_max)
2250 goto retry;
Yan, Zhengc4d4a582015-01-09 15:56:18 +08002251
2252 if (ci->i_inline_version != CEPH_INLINE_NONE &&
2253 (_got & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) &&
2254 i_size_read(&ci->vfs_inode) > 0) {
2255 struct page *page = find_get_page(ci->vfs_inode.i_mapping, 0);
2256 if (page) {
2257 if (PageUptodate(page)) {
2258 *pinned_page = page;
2259 goto out;
2260 }
2261 page_cache_release(page);
2262 }
2263 /*
2264 * drop cap refs first because getattr while holding
2265 * caps refs can cause deadlock.
2266 */
2267 ceph_put_cap_refs(ci, _got);
2268 _got = 0;
2269
2270 /* getattr request will bring inline data into page cache */
2271 ret = __ceph_do_getattr(&ci->vfs_inode, NULL,
2272 CEPH_STAT_CAP_INLINE_DATA, true);
2273 if (ret < 0)
2274 return ret;
2275 goto retry;
2276 }
2277out:
2278 *got = _got;
2279 return 0;
Sage Weila8599bd2009-10-06 11:31:12 -07002280}
2281
2282/*
2283 * Take cap refs. Caller must already know we hold at least one ref
2284 * on the caps in question or we don't know this is safe.
2285 */
2286void ceph_get_cap_refs(struct ceph_inode_info *ci, int caps)
2287{
Sage Weilbe655592011-11-30 09:47:09 -08002288 spin_lock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07002289 __take_cap_refs(ci, caps);
Sage Weilbe655592011-11-30 09:47:09 -08002290 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07002291}
2292
2293/*
2294 * Release cap refs.
2295 *
2296 * If we released the last ref on any given cap, call ceph_check_caps
2297 * to release (or schedule a release).
2298 *
2299 * If we are releasing a WR cap (from a sync write), finalize any affected
2300 * cap_snap, and wake up any waiters.
2301 */
2302void ceph_put_cap_refs(struct ceph_inode_info *ci, int had)
2303{
2304 struct inode *inode = &ci->vfs_inode;
2305 int last = 0, put = 0, flushsnaps = 0, wake = 0;
2306 struct ceph_cap_snap *capsnap;
2307
Sage Weilbe655592011-11-30 09:47:09 -08002308 spin_lock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07002309 if (had & CEPH_CAP_PIN)
2310 --ci->i_pin_ref;
2311 if (had & CEPH_CAP_FILE_RD)
2312 if (--ci->i_rd_ref == 0)
2313 last++;
2314 if (had & CEPH_CAP_FILE_CACHE)
2315 if (--ci->i_rdcache_ref == 0)
2316 last++;
2317 if (had & CEPH_CAP_FILE_BUFFER) {
Henry C Changd3d07202011-05-11 10:29:54 +00002318 if (--ci->i_wb_ref == 0) {
Sage Weila8599bd2009-10-06 11:31:12 -07002319 last++;
2320 put++;
2321 }
Henry C Changd3d07202011-05-11 10:29:54 +00002322 dout("put_cap_refs %p wb %d -> %d (?)\n",
2323 inode, ci->i_wb_ref+1, ci->i_wb_ref);
Sage Weila8599bd2009-10-06 11:31:12 -07002324 }
2325 if (had & CEPH_CAP_FILE_WR)
2326 if (--ci->i_wr_ref == 0) {
2327 last++;
2328 if (!list_empty(&ci->i_cap_snaps)) {
2329 capsnap = list_first_entry(&ci->i_cap_snaps,
2330 struct ceph_cap_snap,
2331 ci_item);
2332 if (capsnap->writing) {
2333 capsnap->writing = 0;
2334 flushsnaps =
2335 __ceph_finish_cap_snap(ci,
2336 capsnap);
2337 wake = 1;
2338 }
2339 }
Yan, Zhengdb40cc12015-03-23 20:12:20 +08002340 /* see comment in __ceph_remove_cap() */
2341 if (!__ceph_is_any_caps(ci) && ci->i_snap_realm)
2342 drop_inode_snap_realm(ci);
Sage Weila8599bd2009-10-06 11:31:12 -07002343 }
Sage Weilbe655592011-11-30 09:47:09 -08002344 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07002345
Sage Weil819ccbf2010-04-01 09:33:46 -07002346 dout("put_cap_refs %p had %s%s%s\n", inode, ceph_cap_string(had),
2347 last ? " last" : "", put ? " put" : "");
Sage Weila8599bd2009-10-06 11:31:12 -07002348
2349 if (last && !flushsnaps)
2350 ceph_check_caps(ci, 0, NULL);
2351 else if (flushsnaps)
2352 ceph_flush_snaps(ci);
2353 if (wake)
Yehuda Sadeh03066f22010-07-27 13:11:08 -07002354 wake_up_all(&ci->i_cap_wq);
Sage Weila8599bd2009-10-06 11:31:12 -07002355 if (put)
2356 iput(inode);
2357}
2358
2359/*
2360 * Release @nr WRBUFFER refs on dirty pages for the given @snapc snap
2361 * context. Adjust per-snap dirty page accounting as appropriate.
2362 * Once all dirty data for a cap_snap is flushed, flush snapped file
2363 * metadata back to the MDS. If we dropped the last ref, call
2364 * ceph_check_caps.
2365 */
2366void ceph_put_wrbuffer_cap_refs(struct ceph_inode_info *ci, int nr,
2367 struct ceph_snap_context *snapc)
2368{
2369 struct inode *inode = &ci->vfs_inode;
2370 int last = 0;
Sage Weil819ccbf2010-04-01 09:33:46 -07002371 int complete_capsnap = 0;
2372 int drop_capsnap = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07002373 int found = 0;
2374 struct ceph_cap_snap *capsnap = NULL;
2375
Sage Weilbe655592011-11-30 09:47:09 -08002376 spin_lock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07002377 ci->i_wrbuffer_ref -= nr;
2378 last = !ci->i_wrbuffer_ref;
2379
2380 if (ci->i_head_snapc == snapc) {
2381 ci->i_wrbuffer_ref_head -= nr;
Sage Weil7d8cb262010-08-24 08:44:16 -07002382 if (ci->i_wrbuffer_ref_head == 0 &&
2383 ci->i_dirty_caps == 0 && ci->i_flushing_caps == 0) {
2384 BUG_ON(!ci->i_head_snapc);
Sage Weila8599bd2009-10-06 11:31:12 -07002385 ceph_put_snap_context(ci->i_head_snapc);
2386 ci->i_head_snapc = NULL;
2387 }
2388 dout("put_wrbuffer_cap_refs on %p head %d/%d -> %d/%d %s\n",
2389 inode,
2390 ci->i_wrbuffer_ref+nr, ci->i_wrbuffer_ref_head+nr,
2391 ci->i_wrbuffer_ref, ci->i_wrbuffer_ref_head,
2392 last ? " LAST" : "");
2393 } else {
2394 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
2395 if (capsnap->context == snapc) {
2396 found = 1;
Sage Weila8599bd2009-10-06 11:31:12 -07002397 break;
2398 }
2399 }
2400 BUG_ON(!found);
Sage Weil819ccbf2010-04-01 09:33:46 -07002401 capsnap->dirty_pages -= nr;
2402 if (capsnap->dirty_pages == 0) {
2403 complete_capsnap = 1;
2404 if (capsnap->dirty == 0)
2405 /* cap writeback completed before we created
2406 * the cap_snap; no FLUSHSNAP is needed */
2407 drop_capsnap = 1;
2408 }
Sage Weila8599bd2009-10-06 11:31:12 -07002409 dout("put_wrbuffer_cap_refs on %p cap_snap %p "
Sage Weil819ccbf2010-04-01 09:33:46 -07002410 " snap %lld %d/%d -> %d/%d %s%s%s\n",
Sage Weila8599bd2009-10-06 11:31:12 -07002411 inode, capsnap, capsnap->context->seq,
2412 ci->i_wrbuffer_ref+nr, capsnap->dirty_pages + nr,
2413 ci->i_wrbuffer_ref, capsnap->dirty_pages,
2414 last ? " (wrbuffer last)" : "",
Sage Weil819ccbf2010-04-01 09:33:46 -07002415 complete_capsnap ? " (complete capsnap)" : "",
2416 drop_capsnap ? " (drop capsnap)" : "");
2417 if (drop_capsnap) {
2418 ceph_put_snap_context(capsnap->context);
2419 list_del(&capsnap->ci_item);
2420 list_del(&capsnap->flushing_item);
2421 ceph_put_cap_snap(capsnap);
2422 }
Sage Weila8599bd2009-10-06 11:31:12 -07002423 }
2424
Sage Weilbe655592011-11-30 09:47:09 -08002425 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07002426
2427 if (last) {
2428 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
2429 iput(inode);
Sage Weil819ccbf2010-04-01 09:33:46 -07002430 } else if (complete_capsnap) {
Sage Weila8599bd2009-10-06 11:31:12 -07002431 ceph_flush_snaps(ci);
Yehuda Sadeh03066f22010-07-27 13:11:08 -07002432 wake_up_all(&ci->i_cap_wq);
Sage Weila8599bd2009-10-06 11:31:12 -07002433 }
Sage Weil819ccbf2010-04-01 09:33:46 -07002434 if (drop_capsnap)
2435 iput(inode);
Sage Weila8599bd2009-10-06 11:31:12 -07002436}
2437
2438/*
Yan, Zhengca20c992013-07-21 10:07:51 +08002439 * Invalidate unlinked inode's aliases, so we can drop the inode ASAP.
2440 */
2441static void invalidate_aliases(struct inode *inode)
2442{
2443 struct dentry *dn, *prev = NULL;
2444
2445 dout("invalidate_aliases inode %p\n", inode);
2446 d_prune_aliases(inode);
2447 /*
2448 * For non-directory inode, d_find_alias() only returns
J. Bruce Fieldsfc12c802014-01-16 17:42:53 -05002449 * hashed dentry. After calling d_invalidate(), the
2450 * dentry becomes unhashed.
Yan, Zhengca20c992013-07-21 10:07:51 +08002451 *
Yan, Zhenga8d436f2013-09-02 15:19:54 +08002452 * For directory inode, d_find_alias() can return
J. Bruce Fieldsfc12c802014-01-16 17:42:53 -05002453 * unhashed dentry. But directory inode should have
Yan, Zhengca20c992013-07-21 10:07:51 +08002454 * one alias at most.
2455 */
2456 while ((dn = d_find_alias(inode))) {
2457 if (dn == prev) {
2458 dput(dn);
2459 break;
2460 }
Yan, Zhenga8d436f2013-09-02 15:19:54 +08002461 d_invalidate(dn);
Yan, Zhengca20c992013-07-21 10:07:51 +08002462 if (prev)
2463 dput(prev);
2464 prev = dn;
2465 }
2466 if (prev)
2467 dput(prev);
2468}
2469
2470/*
Sage Weila8599bd2009-10-06 11:31:12 -07002471 * Handle a cap GRANT message from the MDS. (Note that a GRANT may
2472 * actually be a revocation if it specifies a smaller cap set.)
2473 *
Sage Weilbe655592011-11-30 09:47:09 -08002474 * caller holds s_mutex and i_ceph_lock, we drop both.
Sage Weila8599bd2009-10-06 11:31:12 -07002475 */
Yan, Zheng2cd698b2014-04-18 13:20:27 +08002476static void handle_cap_grant(struct ceph_mds_client *mdsc,
2477 struct inode *inode, struct ceph_mds_caps *grant,
Yan, Zhengfb01d1f2014-11-14 21:29:55 +08002478 u64 inline_version,
2479 void *inline_data, int inline_len,
Yan, Zheng2cd698b2014-04-18 13:20:27 +08002480 struct ceph_buffer *xattr_buf,
Sage Weil15637c82010-03-16 13:42:00 -07002481 struct ceph_mds_session *session,
Yan, Zheng2cd698b2014-04-18 13:20:27 +08002482 struct ceph_cap *cap, int issued)
2483 __releases(ci->i_ceph_lock)
Yan, Zheng982d6012014-12-23 15:30:54 +08002484 __releases(mdsc->snap_rwsem)
Sage Weila8599bd2009-10-06 11:31:12 -07002485{
2486 struct ceph_inode_info *ci = ceph_inode(inode);
2487 int mds = session->s_mds;
Sage Weil2f56f562010-10-27 20:59:49 -07002488 int seq = le32_to_cpu(grant->seq);
Sage Weila8599bd2009-10-06 11:31:12 -07002489 int newcaps = le32_to_cpu(grant->caps);
Yan, Zheng2cd698b2014-04-18 13:20:27 +08002490 int used, wanted, dirty;
Sage Weila8599bd2009-10-06 11:31:12 -07002491 u64 size = le64_to_cpu(grant->size);
2492 u64 max_size = le64_to_cpu(grant->max_size);
2493 struct timespec mtime, atime, ctime;
Sage Weil15637c82010-03-16 13:42:00 -07002494 int check_caps = 0;
Fabian Frederickab6c2c32014-10-09 23:16:35 +02002495 bool wake = false;
2496 bool writeback = false;
2497 bool queue_trunc = false;
2498 bool queue_invalidate = false;
2499 bool queue_revalidate = false;
2500 bool deleted_inode = false;
Yan, Zheng31c542a2014-11-14 21:41:55 +08002501 bool fill_inline = false;
Sage Weila8599bd2009-10-06 11:31:12 -07002502
Sage Weil2f56f562010-10-27 20:59:49 -07002503 dout("handle_cap_grant inode %p cap %p mds%d seq %d %s\n",
2504 inode, cap, mds, seq, ceph_cap_string(newcaps));
Sage Weila8599bd2009-10-06 11:31:12 -07002505 dout(" size %llu max_size %llu, i_size %llu\n", size, max_size,
2506 inode->i_size);
2507
Yan, Zheng11df2df2013-11-24 14:44:38 +08002508
2509 /*
2510 * auth mds of the inode changed. we received the cap export message,
2511 * but still haven't received the cap import message. handle_cap_export
2512 * updated the new auth MDS' cap.
2513 *
2514 * "ceph_seq_cmp(seq, cap->seq) <= 0" means we are processing a message
2515 * that was sent before the cap import message. So don't remove caps.
2516 */
2517 if (ceph_seq_cmp(seq, cap->seq) <= 0) {
2518 WARN_ON(cap != ci->i_auth_cap);
2519 WARN_ON(cap->cap_id != le64_to_cpu(grant->cap_id));
2520 seq = cap->seq;
2521 newcaps |= cap->issued;
2522 }
2523
Sage Weila8599bd2009-10-06 11:31:12 -07002524 /*
2525 * If CACHE is being revoked, and we have no dirty buffers,
2526 * try to invalidate (once). (If there are dirty buffers, we
2527 * will invalidate _after_ writeback.)
2528 */
Sage Weil3b454c42010-06-10 13:20:33 -07002529 if (((cap->issued & ~newcaps) & CEPH_CAP_FILE_CACHE) &&
2530 (newcaps & CEPH_CAP_FILE_LAZYIO) == 0 &&
Yehuda Sadehbcd2cbd2010-02-19 00:12:21 +00002531 !ci->i_wrbuffer_ref) {
Li Wange9075742013-08-15 22:00:25 -07002532 if (try_nonblocking_invalidate(inode)) {
Sage Weila8599bd2009-10-06 11:31:12 -07002533 /* there were locked pages.. invalidate later
2534 in a separate thread. */
2535 if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
Fabian Frederickab6c2c32014-10-09 23:16:35 +02002536 queue_invalidate = true;
Sage Weila8599bd2009-10-06 11:31:12 -07002537 ci->i_rdcache_revoking = ci->i_rdcache_gen;
2538 }
Sage Weila8599bd2009-10-06 11:31:12 -07002539 }
Milosz Tanski99ccbd22013-08-21 17:29:54 -04002540
2541 ceph_fscache_invalidate(inode);
Sage Weila8599bd2009-10-06 11:31:12 -07002542 }
2543
2544 /* side effects now are allowed */
Sage Weil685f9a5d2009-11-09 12:05:48 -08002545 cap->cap_gen = session->s_cap_gen;
Yan, Zheng11df2df2013-11-24 14:44:38 +08002546 cap->seq = seq;
Sage Weila8599bd2009-10-06 11:31:12 -07002547
2548 __check_cap_issue(ci, cap, newcaps);
2549
Yan, Zhengf98a1282014-04-17 08:55:50 +08002550 if ((newcaps & CEPH_CAP_AUTH_SHARED) &&
2551 (issued & CEPH_CAP_AUTH_EXCL) == 0) {
Sage Weila8599bd2009-10-06 11:31:12 -07002552 inode->i_mode = le32_to_cpu(grant->mode);
Eric W. Biederman05cb11c2013-01-31 02:56:19 -08002553 inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(grant->uid));
2554 inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(grant->gid));
Sage Weila8599bd2009-10-06 11:31:12 -07002555 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
Eric W. Biedermanbd2bae62013-01-31 04:05:39 -08002556 from_kuid(&init_user_ns, inode->i_uid),
2557 from_kgid(&init_user_ns, inode->i_gid));
Sage Weila8599bd2009-10-06 11:31:12 -07002558 }
2559
Yan, Zhengf98a1282014-04-17 08:55:50 +08002560 if ((newcaps & CEPH_CAP_AUTH_SHARED) &&
2561 (issued & CEPH_CAP_LINK_EXCL) == 0) {
Miklos Szeredibfe86842011-10-28 14:13:29 +02002562 set_nlink(inode, le32_to_cpu(grant->nlink));
Yan, Zhengca20c992013-07-21 10:07:51 +08002563 if (inode->i_nlink == 0 &&
2564 (newcaps & (CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL)))
Fabian Frederickab6c2c32014-10-09 23:16:35 +02002565 deleted_inode = true;
Yan, Zhengca20c992013-07-21 10:07:51 +08002566 }
Sage Weila8599bd2009-10-06 11:31:12 -07002567
2568 if ((issued & CEPH_CAP_XATTR_EXCL) == 0 && grant->xattr_len) {
2569 int len = le32_to_cpu(grant->xattr_len);
2570 u64 version = le64_to_cpu(grant->xattr_version);
2571
2572 if (version > ci->i_xattrs.version) {
2573 dout(" got new xattrs v%llu on %p len %d\n",
2574 version, inode, len);
2575 if (ci->i_xattrs.blob)
2576 ceph_buffer_put(ci->i_xattrs.blob);
2577 ci->i_xattrs.blob = ceph_buffer_get(xattr_buf);
2578 ci->i_xattrs.version = version;
Guangliang Zhao7221fe42013-11-11 15:18:03 +08002579 ceph_forget_all_cached_acls(inode);
Sage Weila8599bd2009-10-06 11:31:12 -07002580 }
2581 }
2582
Milosz Tanski99ccbd22013-08-21 17:29:54 -04002583 /* Do we need to revalidate our fscache cookie. Don't bother on the
2584 * first cache cap as we already validate at cookie creation time. */
2585 if ((issued & CEPH_CAP_FILE_CACHE) && ci->i_rdcache_gen > 1)
Fabian Frederickab6c2c32014-10-09 23:16:35 +02002586 queue_revalidate = true;
Milosz Tanski99ccbd22013-08-21 17:29:54 -04002587
Yan, Zhengf98a1282014-04-17 08:55:50 +08002588 if (newcaps & CEPH_CAP_ANY_RD) {
2589 /* ctime/mtime/atime? */
2590 ceph_decode_timespec(&mtime, &grant->mtime);
2591 ceph_decode_timespec(&atime, &grant->atime);
2592 ceph_decode_timespec(&ctime, &grant->ctime);
2593 ceph_fill_file_time(inode, issued,
2594 le32_to_cpu(grant->time_warp_seq),
2595 &ctime, &mtime, &atime);
2596 }
Sage Weila8599bd2009-10-06 11:31:12 -07002597
Yan, Zhengf98a1282014-04-17 08:55:50 +08002598 if (newcaps & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR)) {
2599 /* file layout may have changed */
2600 ci->i_layout = grant->layout;
2601 /* size/truncate_seq? */
2602 queue_trunc = ceph_fill_file_size(inode, issued,
2603 le32_to_cpu(grant->truncate_seq),
2604 le64_to_cpu(grant->truncate_size),
2605 size);
2606 /* max size increase? */
2607 if (ci->i_auth_cap == cap && max_size != ci->i_max_size) {
2608 dout("max_size %lld -> %llu\n",
2609 ci->i_max_size, max_size);
2610 ci->i_max_size = max_size;
2611 if (max_size >= ci->i_wanted_max_size) {
2612 ci->i_wanted_max_size = 0; /* reset */
2613 ci->i_requested_max_size = 0;
2614 }
Fabian Frederickab6c2c32014-10-09 23:16:35 +02002615 wake = true;
Sage Weila8599bd2009-10-06 11:31:12 -07002616 }
Sage Weila8599bd2009-10-06 11:31:12 -07002617 }
2618
2619 /* check cap bits */
2620 wanted = __ceph_caps_wanted(ci);
2621 used = __ceph_caps_used(ci);
2622 dirty = __ceph_caps_dirty(ci);
2623 dout(" my wanted = %s, used = %s, dirty %s\n",
2624 ceph_cap_string(wanted),
2625 ceph_cap_string(used),
2626 ceph_cap_string(dirty));
2627 if (wanted != le32_to_cpu(grant->wanted)) {
2628 dout("mds wanted %s -> %s\n",
2629 ceph_cap_string(le32_to_cpu(grant->wanted)),
2630 ceph_cap_string(wanted));
Yan, Zheng390306c2013-01-04 15:30:10 +08002631 /* imported cap may not have correct mds_wanted */
2632 if (le32_to_cpu(grant->op) == CEPH_CAP_OP_IMPORT)
2633 check_caps = 1;
Sage Weila8599bd2009-10-06 11:31:12 -07002634 }
2635
Sage Weila8599bd2009-10-06 11:31:12 -07002636 /* revocation, grant, or no-op? */
2637 if (cap->issued & ~newcaps) {
Sage Weil3b454c42010-06-10 13:20:33 -07002638 int revoking = cap->issued & ~newcaps;
2639
2640 dout("revocation: %s -> %s (revoking %s)\n",
2641 ceph_cap_string(cap->issued),
2642 ceph_cap_string(newcaps),
2643 ceph_cap_string(revoking));
Sage Weil0eb6cd42010-08-05 13:53:18 -07002644 if (revoking & used & CEPH_CAP_FILE_BUFFER)
Fabian Frederickab6c2c32014-10-09 23:16:35 +02002645 writeback = true; /* initiate writeback; will delay ack */
Sage Weil3b454c42010-06-10 13:20:33 -07002646 else if (revoking == CEPH_CAP_FILE_CACHE &&
2647 (newcaps & CEPH_CAP_FILE_LAZYIO) == 0 &&
2648 queue_invalidate)
2649 ; /* do nothing yet, invalidation will be queued */
2650 else if (cap == ci->i_auth_cap)
2651 check_caps = 1; /* check auth cap only */
2652 else
2653 check_caps = 2; /* check all caps */
Sage Weila8599bd2009-10-06 11:31:12 -07002654 cap->issued = newcaps;
Sage Weil978097c2010-03-08 15:27:53 -08002655 cap->implemented |= newcaps;
Sage Weila8599bd2009-10-06 11:31:12 -07002656 } else if (cap->issued == newcaps) {
2657 dout("caps unchanged: %s -> %s\n",
2658 ceph_cap_string(cap->issued), ceph_cap_string(newcaps));
2659 } else {
2660 dout("grant: %s -> %s\n", ceph_cap_string(cap->issued),
2661 ceph_cap_string(newcaps));
Yan, Zheng6ee6b9532013-07-02 12:40:21 +08002662 /* non-auth MDS is revoking the newly grant caps ? */
2663 if (cap == ci->i_auth_cap &&
2664 __ceph_caps_revoking_other(ci, cap, newcaps))
2665 check_caps = 2;
2666
Sage Weila8599bd2009-10-06 11:31:12 -07002667 cap->issued = newcaps;
2668 cap->implemented |= newcaps; /* add bits only, to
2669 * avoid stepping on a
2670 * pending revocation */
Fabian Frederickab6c2c32014-10-09 23:16:35 +02002671 wake = true;
Sage Weila8599bd2009-10-06 11:31:12 -07002672 }
Sage Weil978097c2010-03-08 15:27:53 -08002673 BUG_ON(cap->issued & ~cap->implemented);
Sage Weila8599bd2009-10-06 11:31:12 -07002674
Yan, Zheng31c542a2014-11-14 21:41:55 +08002675 if (inline_version > 0 && inline_version >= ci->i_inline_version) {
2676 ci->i_inline_version = inline_version;
2677 if (ci->i_inline_version != CEPH_INLINE_NONE &&
2678 (newcaps & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)))
2679 fill_inline = true;
2680 }
2681
Sage Weilbe655592011-11-30 09:47:09 -08002682 spin_unlock(&ci->i_ceph_lock);
Milosz Tanski99ccbd22013-08-21 17:29:54 -04002683
Yan, Zheng2cd698b2014-04-18 13:20:27 +08002684 if (le32_to_cpu(grant->op) == CEPH_CAP_OP_IMPORT) {
Yan, Zheng2cd698b2014-04-18 13:20:27 +08002685 kick_flushing_inode_caps(mdsc, session, inode);
2686 up_read(&mdsc->snap_rwsem);
2687 if (newcaps & ~issued)
Fabian Frederickab6c2c32014-10-09 23:16:35 +02002688 wake = true;
Yan, Zheng2cd698b2014-04-18 13:20:27 +08002689 }
2690
Yan, Zheng31c542a2014-11-14 21:41:55 +08002691 if (fill_inline)
2692 ceph_fill_inline_data(inode, NULL, inline_data, inline_len);
2693
Yan, Zhengc6bcda62014-04-11 10:18:07 +08002694 if (queue_trunc) {
2695 ceph_queue_vmtruncate(inode);
2696 ceph_queue_revalidate(inode);
2697 } else if (queue_revalidate)
2698 ceph_queue_revalidate(inode);
2699
Sage Weil3c6f6b72010-02-09 15:24:44 -08002700 if (writeback)
Sage Weila8599bd2009-10-06 11:31:12 -07002701 /*
2702 * queue inode for writeback: we can't actually call
2703 * filemap_write_and_wait, etc. from message handler
2704 * context.
2705 */
Sage Weil3c6f6b72010-02-09 15:24:44 -08002706 ceph_queue_writeback(inode);
2707 if (queue_invalidate)
2708 ceph_queue_invalidate(inode);
Yan, Zhengca20c992013-07-21 10:07:51 +08002709 if (deleted_inode)
2710 invalidate_aliases(inode);
Sage Weila8599bd2009-10-06 11:31:12 -07002711 if (wake)
Yehuda Sadeh03066f22010-07-27 13:11:08 -07002712 wake_up_all(&ci->i_cap_wq);
Sage Weil15637c82010-03-16 13:42:00 -07002713
2714 if (check_caps == 1)
2715 ceph_check_caps(ci, CHECK_CAPS_NODELAY|CHECK_CAPS_AUTHONLY,
2716 session);
2717 else if (check_caps == 2)
2718 ceph_check_caps(ci, CHECK_CAPS_NODELAY, session);
2719 else
2720 mutex_unlock(&session->s_mutex);
Sage Weila8599bd2009-10-06 11:31:12 -07002721}
2722
2723/*
2724 * Handle FLUSH_ACK from MDS, indicating that metadata we sent to the
2725 * MDS has been safely committed.
2726 */
Sage Weil6df058c2009-12-22 11:24:33 -08002727static void handle_cap_flush_ack(struct inode *inode, u64 flush_tid,
Sage Weila8599bd2009-10-06 11:31:12 -07002728 struct ceph_mds_caps *m,
2729 struct ceph_mds_session *session,
2730 struct ceph_cap *cap)
Sage Weilbe655592011-11-30 09:47:09 -08002731 __releases(ci->i_ceph_lock)
Sage Weila8599bd2009-10-06 11:31:12 -07002732{
2733 struct ceph_inode_info *ci = ceph_inode(inode);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002734 struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
Sage Weila8599bd2009-10-06 11:31:12 -07002735 unsigned seq = le32_to_cpu(m->seq);
2736 int dirty = le32_to_cpu(m->dirty);
2737 int cleaned = 0;
Sage Weilafcdaea2009-10-14 14:27:38 -07002738 int drop = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07002739 int i;
2740
2741 for (i = 0; i < CEPH_CAP_BITS; i++)
2742 if ((dirty & (1 << i)) &&
Yan, Zheng32313002014-10-22 17:13:26 -07002743 (u16)flush_tid == ci->i_cap_flush_tid[i])
Sage Weila8599bd2009-10-06 11:31:12 -07002744 cleaned |= 1 << i;
2745
2746 dout("handle_cap_flush_ack inode %p mds%d seq %d on %s cleaned %s,"
2747 " flushing %s -> %s\n",
2748 inode, session->s_mds, seq, ceph_cap_string(dirty),
2749 ceph_cap_string(cleaned), ceph_cap_string(ci->i_flushing_caps),
2750 ceph_cap_string(ci->i_flushing_caps & ~cleaned));
2751
2752 if (ci->i_flushing_caps == (ci->i_flushing_caps & ~cleaned))
2753 goto out;
2754
Sage Weila8599bd2009-10-06 11:31:12 -07002755 ci->i_flushing_caps &= ~cleaned;
Sage Weila8599bd2009-10-06 11:31:12 -07002756
2757 spin_lock(&mdsc->cap_dirty_lock);
2758 if (ci->i_flushing_caps == 0) {
2759 list_del_init(&ci->i_flushing_item);
2760 if (!list_empty(&session->s_cap_flushing))
2761 dout(" mds%d still flushing cap on %p\n",
2762 session->s_mds,
2763 &list_entry(session->s_cap_flushing.next,
2764 struct ceph_inode_info,
2765 i_flushing_item)->vfs_inode);
2766 mdsc->num_cap_flushing--;
Yehuda Sadeh03066f22010-07-27 13:11:08 -07002767 wake_up_all(&mdsc->cap_flushing_wq);
Sage Weila8599bd2009-10-06 11:31:12 -07002768 dout(" inode %p now !flushing\n", inode);
Sage Weilafcdaea2009-10-14 14:27:38 -07002769
2770 if (ci->i_dirty_caps == 0) {
2771 dout(" inode %p now clean\n", inode);
2772 BUG_ON(!list_empty(&ci->i_dirty_item));
2773 drop = 1;
Sage Weil7d8cb262010-08-24 08:44:16 -07002774 if (ci->i_wrbuffer_ref_head == 0) {
2775 BUG_ON(!ci->i_head_snapc);
2776 ceph_put_snap_context(ci->i_head_snapc);
2777 ci->i_head_snapc = NULL;
2778 }
Sage Weil76e3b392009-10-15 18:13:53 -07002779 } else {
2780 BUG_ON(list_empty(&ci->i_dirty_item));
Sage Weilafcdaea2009-10-14 14:27:38 -07002781 }
Sage Weila8599bd2009-10-06 11:31:12 -07002782 }
2783 spin_unlock(&mdsc->cap_dirty_lock);
Yehuda Sadeh03066f22010-07-27 13:11:08 -07002784 wake_up_all(&ci->i_cap_wq);
Sage Weila8599bd2009-10-06 11:31:12 -07002785
2786out:
Sage Weilbe655592011-11-30 09:47:09 -08002787 spin_unlock(&ci->i_ceph_lock);
Sage Weilafcdaea2009-10-14 14:27:38 -07002788 if (drop)
Sage Weila8599bd2009-10-06 11:31:12 -07002789 iput(inode);
2790}
2791
2792/*
2793 * Handle FLUSHSNAP_ACK. MDS has flushed snap data to disk and we can
2794 * throw away our cap_snap.
2795 *
2796 * Caller hold s_mutex.
2797 */
Sage Weil6df058c2009-12-22 11:24:33 -08002798static void handle_cap_flushsnap_ack(struct inode *inode, u64 flush_tid,
Sage Weila8599bd2009-10-06 11:31:12 -07002799 struct ceph_mds_caps *m,
2800 struct ceph_mds_session *session)
2801{
2802 struct ceph_inode_info *ci = ceph_inode(inode);
2803 u64 follows = le64_to_cpu(m->snap_follows);
Sage Weila8599bd2009-10-06 11:31:12 -07002804 struct ceph_cap_snap *capsnap;
2805 int drop = 0;
2806
2807 dout("handle_cap_flushsnap_ack inode %p ci %p mds%d follows %lld\n",
2808 inode, ci, session->s_mds, follows);
2809
Sage Weilbe655592011-11-30 09:47:09 -08002810 spin_lock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07002811 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
2812 if (capsnap->follows == follows) {
2813 if (capsnap->flush_tid != flush_tid) {
2814 dout(" cap_snap %p follows %lld tid %lld !="
2815 " %lld\n", capsnap, follows,
2816 flush_tid, capsnap->flush_tid);
2817 break;
2818 }
2819 WARN_ON(capsnap->dirty_pages || capsnap->writing);
Sage Weil819ccbf2010-04-01 09:33:46 -07002820 dout(" removing %p cap_snap %p follows %lld\n",
2821 inode, capsnap, follows);
Sage Weila8599bd2009-10-06 11:31:12 -07002822 ceph_put_snap_context(capsnap->context);
2823 list_del(&capsnap->ci_item);
2824 list_del(&capsnap->flushing_item);
2825 ceph_put_cap_snap(capsnap);
2826 drop = 1;
2827 break;
2828 } else {
2829 dout(" skipping cap_snap %p follows %lld\n",
2830 capsnap, capsnap->follows);
2831 }
2832 }
Sage Weilbe655592011-11-30 09:47:09 -08002833 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07002834 if (drop)
2835 iput(inode);
2836}
2837
2838/*
2839 * Handle TRUNC from MDS, indicating file truncation.
2840 *
2841 * caller hold s_mutex.
2842 */
2843static void handle_cap_trunc(struct inode *inode,
2844 struct ceph_mds_caps *trunc,
2845 struct ceph_mds_session *session)
Sage Weilbe655592011-11-30 09:47:09 -08002846 __releases(ci->i_ceph_lock)
Sage Weila8599bd2009-10-06 11:31:12 -07002847{
2848 struct ceph_inode_info *ci = ceph_inode(inode);
2849 int mds = session->s_mds;
2850 int seq = le32_to_cpu(trunc->seq);
2851 u32 truncate_seq = le32_to_cpu(trunc->truncate_seq);
2852 u64 truncate_size = le64_to_cpu(trunc->truncate_size);
2853 u64 size = le64_to_cpu(trunc->size);
2854 int implemented = 0;
2855 int dirty = __ceph_caps_dirty(ci);
2856 int issued = __ceph_caps_issued(ceph_inode(inode), &implemented);
2857 int queue_trunc = 0;
2858
2859 issued |= implemented | dirty;
2860
2861 dout("handle_cap_trunc inode %p mds%d seq %d to %lld seq %d\n",
2862 inode, mds, seq, truncate_size, truncate_seq);
2863 queue_trunc = ceph_fill_file_size(inode, issued,
2864 truncate_seq, truncate_size, size);
Sage Weilbe655592011-11-30 09:47:09 -08002865 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07002866
Milosz Tanski99ccbd22013-08-21 17:29:54 -04002867 if (queue_trunc) {
Sage Weil3c6f6b72010-02-09 15:24:44 -08002868 ceph_queue_vmtruncate(inode);
Milosz Tanski99ccbd22013-08-21 17:29:54 -04002869 ceph_fscache_invalidate(inode);
2870 }
Sage Weila8599bd2009-10-06 11:31:12 -07002871}
2872
2873/*
2874 * Handle EXPORT from MDS. Cap is being migrated _from_ this mds to a
2875 * different one. If we are the most recent migration we've seen (as
2876 * indicated by mseq), make note of the migrating cap bits for the
2877 * duration (until we see the corresponding IMPORT).
2878 *
2879 * caller holds s_mutex
2880 */
2881static void handle_cap_export(struct inode *inode, struct ceph_mds_caps *ex,
Yan, Zheng11df2df2013-11-24 14:44:38 +08002882 struct ceph_mds_cap_peer *ph,
2883 struct ceph_mds_session *session)
Sage Weila8599bd2009-10-06 11:31:12 -07002884{
Sage Weildb354052011-05-24 11:46:31 -07002885 struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
Yan, Zheng11df2df2013-11-24 14:44:38 +08002886 struct ceph_mds_session *tsession = NULL;
Yan, Zhengd9df2782014-04-18 09:57:11 +08002887 struct ceph_cap *cap, *tcap, *new_cap = NULL;
Sage Weila8599bd2009-10-06 11:31:12 -07002888 struct ceph_inode_info *ci = ceph_inode(inode);
Yan, Zheng11df2df2013-11-24 14:44:38 +08002889 u64 t_cap_id;
Sage Weila8599bd2009-10-06 11:31:12 -07002890 unsigned mseq = le32_to_cpu(ex->migrate_seq);
Yan, Zheng11df2df2013-11-24 14:44:38 +08002891 unsigned t_seq, t_mseq;
2892 int target, issued;
2893 int mds = session->s_mds;
Sage Weila8599bd2009-10-06 11:31:12 -07002894
Yan, Zheng11df2df2013-11-24 14:44:38 +08002895 if (ph) {
2896 t_cap_id = le64_to_cpu(ph->cap_id);
2897 t_seq = le32_to_cpu(ph->seq);
2898 t_mseq = le32_to_cpu(ph->mseq);
2899 target = le32_to_cpu(ph->mds);
2900 } else {
2901 t_cap_id = t_seq = t_mseq = 0;
2902 target = -1;
Sage Weila8599bd2009-10-06 11:31:12 -07002903 }
2904
Yan, Zheng11df2df2013-11-24 14:44:38 +08002905 dout("handle_cap_export inode %p ci %p mds%d mseq %d target %d\n",
2906 inode, ci, mds, mseq, target);
2907retry:
2908 spin_lock(&ci->i_ceph_lock);
2909 cap = __get_cap_for_mds(ci, mds);
Yan, Zhengca665e02014-04-21 15:46:37 +08002910 if (!cap || cap->cap_id != le64_to_cpu(ex->cap_id))
Yan, Zheng11df2df2013-11-24 14:44:38 +08002911 goto out_unlock;
Sage Weil154f42c2010-06-21 13:45:04 -07002912
Yan, Zheng11df2df2013-11-24 14:44:38 +08002913 if (target < 0) {
2914 __ceph_remove_cap(cap, false);
2915 goto out_unlock;
2916 }
Sage Weildb354052011-05-24 11:46:31 -07002917
Yan, Zheng11df2df2013-11-24 14:44:38 +08002918 /*
2919 * now we know we haven't received the cap import message yet
2920 * because the exported cap still exist.
2921 */
2922
2923 issued = cap->issued;
2924 WARN_ON(issued != cap->implemented);
2925
2926 tcap = __get_cap_for_mds(ci, target);
2927 if (tcap) {
2928 /* already have caps from the target */
2929 if (tcap->cap_id != t_cap_id ||
2930 ceph_seq_cmp(tcap->seq, t_seq) < 0) {
2931 dout(" updating import cap %p mds%d\n", tcap, target);
2932 tcap->cap_id = t_cap_id;
2933 tcap->seq = t_seq - 1;
2934 tcap->issue_seq = t_seq - 1;
2935 tcap->mseq = t_mseq;
2936 tcap->issued |= issued;
2937 tcap->implemented |= issued;
2938 if (cap == ci->i_auth_cap)
2939 ci->i_auth_cap = tcap;
2940 if (ci->i_flushing_caps && ci->i_auth_cap == tcap) {
2941 spin_lock(&mdsc->cap_dirty_lock);
2942 list_move_tail(&ci->i_flushing_item,
2943 &tcap->session->s_cap_flushing);
2944 spin_unlock(&mdsc->cap_dirty_lock);
Sage Weildb354052011-05-24 11:46:31 -07002945 }
Sage Weila8599bd2009-10-06 11:31:12 -07002946 }
Yan, Zhenga096b092013-09-22 10:15:58 +08002947 __ceph_remove_cap(cap, false);
Yan, Zheng11df2df2013-11-24 14:44:38 +08002948 goto out_unlock;
Yan, Zhengd9df2782014-04-18 09:57:11 +08002949 } else if (tsession) {
Yan, Zheng11df2df2013-11-24 14:44:38 +08002950 /* add placeholder for the export tagert */
Yan, Zhengd9df2782014-04-18 09:57:11 +08002951 int flag = (cap == ci->i_auth_cap) ? CEPH_CAP_FLAG_AUTH : 0;
Yan, Zheng11df2df2013-11-24 14:44:38 +08002952 ceph_add_cap(inode, tsession, t_cap_id, -1, issued, 0,
Yan, Zhengd9df2782014-04-18 09:57:11 +08002953 t_seq - 1, t_mseq, (u64)-1, flag, &new_cap);
2954
2955 __ceph_remove_cap(cap, false);
2956 goto out_unlock;
Yan, Zheng11df2df2013-11-24 14:44:38 +08002957 }
Sage Weila8599bd2009-10-06 11:31:12 -07002958
Sage Weilbe655592011-11-30 09:47:09 -08002959 spin_unlock(&ci->i_ceph_lock);
Yan, Zheng11df2df2013-11-24 14:44:38 +08002960 mutex_unlock(&session->s_mutex);
2961
2962 /* open target session */
2963 tsession = ceph_mdsc_open_export_target_session(mdsc, target);
2964 if (!IS_ERR(tsession)) {
2965 if (mds > target) {
2966 mutex_lock(&session->s_mutex);
2967 mutex_lock_nested(&tsession->s_mutex,
2968 SINGLE_DEPTH_NESTING);
2969 } else {
2970 mutex_lock(&tsession->s_mutex);
2971 mutex_lock_nested(&session->s_mutex,
2972 SINGLE_DEPTH_NESTING);
2973 }
2974 ceph_add_cap_releases(mdsc, tsession);
Yan, Zhengd9df2782014-04-18 09:57:11 +08002975 new_cap = ceph_get_cap(mdsc, NULL);
Yan, Zheng11df2df2013-11-24 14:44:38 +08002976 } else {
2977 WARN_ON(1);
2978 tsession = NULL;
2979 target = -1;
2980 }
2981 goto retry;
2982
2983out_unlock:
2984 spin_unlock(&ci->i_ceph_lock);
2985 mutex_unlock(&session->s_mutex);
2986 if (tsession) {
2987 mutex_unlock(&tsession->s_mutex);
2988 ceph_put_mds_session(tsession);
2989 }
Yan, Zhengd9df2782014-04-18 09:57:11 +08002990 if (new_cap)
2991 ceph_put_cap(mdsc, new_cap);
Sage Weila8599bd2009-10-06 11:31:12 -07002992}
2993
2994/*
Yan, Zheng2cd698b2014-04-18 13:20:27 +08002995 * Handle cap IMPORT.
Sage Weila8599bd2009-10-06 11:31:12 -07002996 *
Yan, Zheng2cd698b2014-04-18 13:20:27 +08002997 * caller holds s_mutex. acquires i_ceph_lock
Sage Weila8599bd2009-10-06 11:31:12 -07002998 */
2999static void handle_cap_import(struct ceph_mds_client *mdsc,
3000 struct inode *inode, struct ceph_mds_caps *im,
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003001 struct ceph_mds_cap_peer *ph,
Sage Weila8599bd2009-10-06 11:31:12 -07003002 struct ceph_mds_session *session,
Yan, Zheng2cd698b2014-04-18 13:20:27 +08003003 struct ceph_cap **target_cap, int *old_issued)
3004 __acquires(ci->i_ceph_lock)
Sage Weila8599bd2009-10-06 11:31:12 -07003005{
3006 struct ceph_inode_info *ci = ceph_inode(inode);
Yan, Zheng2cd698b2014-04-18 13:20:27 +08003007 struct ceph_cap *cap, *ocap, *new_cap = NULL;
Sage Weila8599bd2009-10-06 11:31:12 -07003008 int mds = session->s_mds;
Yan, Zheng2cd698b2014-04-18 13:20:27 +08003009 int issued;
3010 unsigned caps = le32_to_cpu(im->caps);
Sage Weila8599bd2009-10-06 11:31:12 -07003011 unsigned wanted = le32_to_cpu(im->wanted);
3012 unsigned seq = le32_to_cpu(im->seq);
3013 unsigned mseq = le32_to_cpu(im->migrate_seq);
3014 u64 realmino = le64_to_cpu(im->realm);
3015 u64 cap_id = le64_to_cpu(im->cap_id);
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003016 u64 p_cap_id;
3017 int peer;
Sage Weila8599bd2009-10-06 11:31:12 -07003018
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003019 if (ph) {
3020 p_cap_id = le64_to_cpu(ph->cap_id);
3021 peer = le32_to_cpu(ph->mds);
Sage Weila8599bd2009-10-06 11:31:12 -07003022 } else {
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003023 p_cap_id = 0;
3024 peer = -1;
Sage Weila8599bd2009-10-06 11:31:12 -07003025 }
3026
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003027 dout("handle_cap_import inode %p ci %p mds%d mseq %d peer %d\n",
3028 inode, ci, mds, mseq, peer);
3029
Yan, Zhengd9df2782014-04-18 09:57:11 +08003030retry:
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003031 spin_lock(&ci->i_ceph_lock);
Yan, Zhengd9df2782014-04-18 09:57:11 +08003032 cap = __get_cap_for_mds(ci, mds);
3033 if (!cap) {
3034 if (!new_cap) {
3035 spin_unlock(&ci->i_ceph_lock);
3036 new_cap = ceph_get_cap(mdsc, NULL);
3037 goto retry;
3038 }
Yan, Zheng2cd698b2014-04-18 13:20:27 +08003039 cap = new_cap;
3040 } else {
3041 if (new_cap) {
3042 ceph_put_cap(mdsc, new_cap);
3043 new_cap = NULL;
3044 }
Yan, Zhengd9df2782014-04-18 09:57:11 +08003045 }
3046
Yan, Zheng2cd698b2014-04-18 13:20:27 +08003047 __ceph_caps_issued(ci, &issued);
3048 issued |= __ceph_caps_dirty(ci);
3049
3050 ceph_add_cap(inode, session, cap_id, -1, caps, wanted, seq, mseq,
Yan, Zhengd9df2782014-04-18 09:57:11 +08003051 realmino, CEPH_CAP_FLAG_AUTH, &new_cap);
3052
Yan, Zheng2cd698b2014-04-18 13:20:27 +08003053 ocap = peer >= 0 ? __get_cap_for_mds(ci, peer) : NULL;
3054 if (ocap && ocap->cap_id == p_cap_id) {
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003055 dout(" remove export cap %p mds%d flags %d\n",
Yan, Zheng2cd698b2014-04-18 13:20:27 +08003056 ocap, peer, ph->flags);
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003057 if ((ph->flags & CEPH_CAP_FLAG_AUTH) &&
Yan, Zheng2cd698b2014-04-18 13:20:27 +08003058 (ocap->seq != le32_to_cpu(ph->seq) ||
3059 ocap->mseq != le32_to_cpu(ph->mseq))) {
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003060 pr_err("handle_cap_import: mismatched seq/mseq: "
3061 "ino (%llx.%llx) mds%d seq %d mseq %d "
3062 "importer mds%d has peer seq %d mseq %d\n",
Yan, Zheng2cd698b2014-04-18 13:20:27 +08003063 ceph_vinop(inode), peer, ocap->seq,
3064 ocap->mseq, mds, le32_to_cpu(ph->seq),
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003065 le32_to_cpu(ph->mseq));
3066 }
Yan, Zheng2cd698b2014-04-18 13:20:27 +08003067 __ceph_remove_cap(ocap, (ph->flags & CEPH_CAP_FLAG_RELEASE));
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003068 }
3069
3070 /* make sure we re-request max_size, if necessary */
3071 ci->i_wanted_max_size = 0;
3072 ci->i_requested_max_size = 0;
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003073
Yan, Zheng2cd698b2014-04-18 13:20:27 +08003074 *old_issued = issued;
3075 *target_cap = cap;
Sage Weila8599bd2009-10-06 11:31:12 -07003076}
3077
3078/*
3079 * Handle a caps message from the MDS.
3080 *
3081 * Identify the appropriate session, inode, and call the right handler
3082 * based on the cap op.
3083 */
3084void ceph_handle_caps(struct ceph_mds_session *session,
3085 struct ceph_msg *msg)
3086{
3087 struct ceph_mds_client *mdsc = session->s_mdsc;
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07003088 struct super_block *sb = mdsc->fsc->sb;
Sage Weila8599bd2009-10-06 11:31:12 -07003089 struct inode *inode;
Sage Weilbe655592011-11-30 09:47:09 -08003090 struct ceph_inode_info *ci;
Sage Weila8599bd2009-10-06 11:31:12 -07003091 struct ceph_cap *cap;
3092 struct ceph_mds_caps *h;
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003093 struct ceph_mds_cap_peer *peer = NULL;
Yan, Zheng982d6012014-12-23 15:30:54 +08003094 struct ceph_snap_realm *realm;
Sage Weil2600d2d2010-02-22 15:12:16 -08003095 int mds = session->s_mds;
Yan, Zheng2cd698b2014-04-18 13:20:27 +08003096 int op, issued;
Sage Weil3d7ded42010-06-09 16:47:10 -07003097 u32 seq, mseq;
Sage Weila8599bd2009-10-06 11:31:12 -07003098 struct ceph_vino vino;
3099 u64 cap_id;
3100 u64 size, max_size;
Sage Weil6df058c2009-12-22 11:24:33 -08003101 u64 tid;
Yan, Zhengfb01d1f2014-11-14 21:29:55 +08003102 u64 inline_version = 0;
3103 void *inline_data = NULL;
3104 u32 inline_len = 0;
Sage Weil70edb552010-03-01 13:20:50 -08003105 void *snaptrace;
Sage Weilce1fbc82010-08-02 15:09:39 -07003106 size_t snaptrace_len;
Yan, Zhengfb01d1f2014-11-14 21:29:55 +08003107 void *p, *end;
Sage Weila8599bd2009-10-06 11:31:12 -07003108
3109 dout("handle_caps from mds%d\n", mds);
3110
3111 /* decode */
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003112 end = msg->front.iov_base + msg->front.iov_len;
Sage Weil6df058c2009-12-22 11:24:33 -08003113 tid = le64_to_cpu(msg->hdr.tid);
Sage Weila8599bd2009-10-06 11:31:12 -07003114 if (msg->front.iov_len < sizeof(*h))
3115 goto bad;
3116 h = msg->front.iov_base;
3117 op = le32_to_cpu(h->op);
3118 vino.ino = le64_to_cpu(h->ino);
3119 vino.snap = CEPH_NOSNAP;
3120 cap_id = le64_to_cpu(h->cap_id);
3121 seq = le32_to_cpu(h->seq);
Sage Weil3d7ded42010-06-09 16:47:10 -07003122 mseq = le32_to_cpu(h->migrate_seq);
Sage Weila8599bd2009-10-06 11:31:12 -07003123 size = le64_to_cpu(h->size);
3124 max_size = le64_to_cpu(h->max_size);
3125
Sage Weilce1fbc82010-08-02 15:09:39 -07003126 snaptrace = h + 1;
3127 snaptrace_len = le32_to_cpu(h->snap_trace_len);
Yan, Zhengfb01d1f2014-11-14 21:29:55 +08003128 p = snaptrace + snaptrace_len;
Sage Weilce1fbc82010-08-02 15:09:39 -07003129
3130 if (le16_to_cpu(msg->hdr.version) >= 2) {
Yan, Zhengfb01d1f2014-11-14 21:29:55 +08003131 u32 flock_len;
Sage Weilce1fbc82010-08-02 15:09:39 -07003132 ceph_decode_32_safe(&p, end, flock_len, bad);
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003133 if (p + flock_len > end)
3134 goto bad;
Yan, Zhengfb01d1f2014-11-14 21:29:55 +08003135 p += flock_len;
Sage Weilce1fbc82010-08-02 15:09:39 -07003136 }
3137
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003138 if (le16_to_cpu(msg->hdr.version) >= 3) {
3139 if (op == CEPH_CAP_OP_IMPORT) {
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003140 if (p + sizeof(*peer) > end)
3141 goto bad;
3142 peer = p;
Yan, Zhengfb01d1f2014-11-14 21:29:55 +08003143 p += sizeof(*peer);
Yan, Zheng11df2df2013-11-24 14:44:38 +08003144 } else if (op == CEPH_CAP_OP_EXPORT) {
3145 /* recorded in unused fields */
3146 peer = (void *)&h->size;
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003147 }
3148 }
3149
Yan, Zhengfb01d1f2014-11-14 21:29:55 +08003150 if (le16_to_cpu(msg->hdr.version) >= 4) {
3151 ceph_decode_64_safe(&p, end, inline_version, bad);
3152 ceph_decode_32_safe(&p, end, inline_len, bad);
3153 if (p + inline_len > end)
3154 goto bad;
3155 inline_data = p;
3156 p += inline_len;
3157 }
3158
Yan, Zheng6cd3bca2014-09-17 07:45:12 +08003159 /* lookup ino */
3160 inode = ceph_find_inode(sb, vino);
3161 ci = ceph_inode(inode);
3162 dout(" op %s ino %llx.%llx inode %p\n", ceph_cap_op_name(op), vino.ino,
3163 vino.snap, inode);
3164
Sage Weila8599bd2009-10-06 11:31:12 -07003165 mutex_lock(&session->s_mutex);
3166 session->s_seq++;
3167 dout(" mds%d seq %lld cap seq %u\n", session->s_mds, session->s_seq,
3168 (unsigned)seq);
3169
Yan, Zheng66f58692013-01-04 14:45:18 +08003170 if (op == CEPH_CAP_OP_IMPORT)
3171 ceph_add_cap_releases(mdsc, session);
3172
Sage Weila8599bd2009-10-06 11:31:12 -07003173 if (!inode) {
3174 dout(" i don't have ino %llx\n", vino.ino);
Sage Weil3d7ded42010-06-09 16:47:10 -07003175
Yan, Zhenga096b092013-09-22 10:15:58 +08003176 if (op == CEPH_CAP_OP_IMPORT) {
3177 spin_lock(&session->s_cap_lock);
Sage Weil3d7ded42010-06-09 16:47:10 -07003178 __queue_cap_release(session, vino.ino, cap_id,
3179 mseq, seq);
Yan, Zhenga096b092013-09-22 10:15:58 +08003180 spin_unlock(&session->s_cap_lock);
3181 }
Greg Farnum21b559d2010-10-06 15:46:30 -07003182 goto flush_cap_releases;
Sage Weila8599bd2009-10-06 11:31:12 -07003183 }
3184
3185 /* these will work even if we don't have a cap yet */
3186 switch (op) {
3187 case CEPH_CAP_OP_FLUSHSNAP_ACK:
Sage Weil6df058c2009-12-22 11:24:33 -08003188 handle_cap_flushsnap_ack(inode, tid, h, session);
Sage Weila8599bd2009-10-06 11:31:12 -07003189 goto done;
3190
3191 case CEPH_CAP_OP_EXPORT:
Yan, Zheng11df2df2013-11-24 14:44:38 +08003192 handle_cap_export(inode, h, peer, session);
3193 goto done_unlocked;
Sage Weila8599bd2009-10-06 11:31:12 -07003194
3195 case CEPH_CAP_OP_IMPORT:
Yan, Zheng982d6012014-12-23 15:30:54 +08003196 realm = NULL;
3197 if (snaptrace_len) {
3198 down_write(&mdsc->snap_rwsem);
3199 ceph_update_snap_trace(mdsc, snaptrace,
3200 snaptrace + snaptrace_len,
3201 false, &realm);
3202 downgrade_write(&mdsc->snap_rwsem);
3203 } else {
3204 down_read(&mdsc->snap_rwsem);
3205 }
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003206 handle_cap_import(mdsc, inode, h, peer, session,
Yan, Zheng2cd698b2014-04-18 13:20:27 +08003207 &cap, &issued);
Yan, Zheng982d6012014-12-23 15:30:54 +08003208 handle_cap_grant(mdsc, inode, h,
Yan, Zhengfb01d1f2014-11-14 21:29:55 +08003209 inline_version, inline_data, inline_len,
Yan, Zheng2cd698b2014-04-18 13:20:27 +08003210 msg->middle, session, cap, issued);
Yan, Zheng982d6012014-12-23 15:30:54 +08003211 if (realm)
3212 ceph_put_snap_realm(mdsc, realm);
Yan, Zheng2cd698b2014-04-18 13:20:27 +08003213 goto done_unlocked;
Sage Weila8599bd2009-10-06 11:31:12 -07003214 }
3215
3216 /* the rest require a cap */
Sage Weilbe655592011-11-30 09:47:09 -08003217 spin_lock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07003218 cap = __get_cap_for_mds(ceph_inode(inode), mds);
3219 if (!cap) {
Sage Weil9dbd4122010-06-10 13:21:20 -07003220 dout(" no cap on %p ino %llx.%llx from mds%d\n",
Sage Weila8599bd2009-10-06 11:31:12 -07003221 inode, ceph_ino(inode), ceph_snap(inode), mds);
Sage Weilbe655592011-11-30 09:47:09 -08003222 spin_unlock(&ci->i_ceph_lock);
Greg Farnum21b559d2010-10-06 15:46:30 -07003223 goto flush_cap_releases;
Sage Weila8599bd2009-10-06 11:31:12 -07003224 }
3225
Sage Weilbe655592011-11-30 09:47:09 -08003226 /* note that each of these drops i_ceph_lock for us */
Sage Weila8599bd2009-10-06 11:31:12 -07003227 switch (op) {
3228 case CEPH_CAP_OP_REVOKE:
3229 case CEPH_CAP_OP_GRANT:
Yan, Zheng2cd698b2014-04-18 13:20:27 +08003230 __ceph_caps_issued(ci, &issued);
3231 issued |= __ceph_caps_dirty(ci);
Yan, Zheng982d6012014-12-23 15:30:54 +08003232 handle_cap_grant(mdsc, inode, h,
Yan, Zhengfb01d1f2014-11-14 21:29:55 +08003233 inline_version, inline_data, inline_len,
3234 msg->middle, session, cap, issued);
Sage Weil15637c82010-03-16 13:42:00 -07003235 goto done_unlocked;
Sage Weila8599bd2009-10-06 11:31:12 -07003236
3237 case CEPH_CAP_OP_FLUSH_ACK:
Sage Weil6df058c2009-12-22 11:24:33 -08003238 handle_cap_flush_ack(inode, tid, h, session, cap);
Sage Weila8599bd2009-10-06 11:31:12 -07003239 break;
3240
3241 case CEPH_CAP_OP_TRUNC:
3242 handle_cap_trunc(inode, h, session);
3243 break;
3244
3245 default:
Sage Weilbe655592011-11-30 09:47:09 -08003246 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07003247 pr_err("ceph_handle_caps: unknown cap op %d %s\n", op,
3248 ceph_cap_op_name(op));
3249 }
3250
Greg Farnum21b559d2010-10-06 15:46:30 -07003251 goto done;
3252
3253flush_cap_releases:
3254 /*
3255 * send any full release message to try to move things
3256 * along for the mds (who clearly thinks we still have this
3257 * cap).
3258 */
3259 ceph_add_cap_releases(mdsc, session);
3260 ceph_send_cap_releases(mdsc, session);
3261
Sage Weila8599bd2009-10-06 11:31:12 -07003262done:
Sage Weil15637c82010-03-16 13:42:00 -07003263 mutex_unlock(&session->s_mutex);
3264done_unlocked:
SF Markus Elfringe96a6502014-11-02 15:20:59 +01003265 iput(inode);
Sage Weila8599bd2009-10-06 11:31:12 -07003266 return;
3267
3268bad:
3269 pr_err("ceph_handle_caps: corrupt message\n");
Sage Weil9ec7cab2009-12-14 15:13:47 -08003270 ceph_msg_dump(msg);
Sage Weila8599bd2009-10-06 11:31:12 -07003271 return;
3272}
3273
3274/*
3275 * Delayed work handler to process end of delayed cap release LRU list.
3276 */
Sage Weilafcdaea2009-10-14 14:27:38 -07003277void ceph_check_delayed_caps(struct ceph_mds_client *mdsc)
Sage Weila8599bd2009-10-06 11:31:12 -07003278{
3279 struct ceph_inode_info *ci;
3280 int flags = CHECK_CAPS_NODELAY;
3281
Sage Weila8599bd2009-10-06 11:31:12 -07003282 dout("check_delayed_caps\n");
3283 while (1) {
3284 spin_lock(&mdsc->cap_delay_lock);
3285 if (list_empty(&mdsc->cap_delay_list))
3286 break;
3287 ci = list_first_entry(&mdsc->cap_delay_list,
3288 struct ceph_inode_info,
3289 i_cap_delay_list);
3290 if ((ci->i_ceph_flags & CEPH_I_FLUSH) == 0 &&
3291 time_before(jiffies, ci->i_hold_caps_max))
3292 break;
3293 list_del_init(&ci->i_cap_delay_list);
3294 spin_unlock(&mdsc->cap_delay_lock);
3295 dout("check_delayed_caps on %p\n", &ci->vfs_inode);
3296 ceph_check_caps(ci, flags, NULL);
3297 }
3298 spin_unlock(&mdsc->cap_delay_lock);
3299}
3300
3301/*
Sage Weilafcdaea2009-10-14 14:27:38 -07003302 * Flush all dirty caps to the mds
3303 */
3304void ceph_flush_dirty_caps(struct ceph_mds_client *mdsc)
3305{
Sage Weildb354052011-05-24 11:46:31 -07003306 struct ceph_inode_info *ci;
3307 struct inode *inode;
Sage Weilafcdaea2009-10-14 14:27:38 -07003308
3309 dout("flush_dirty_caps\n");
3310 spin_lock(&mdsc->cap_dirty_lock);
Sage Weildb354052011-05-24 11:46:31 -07003311 while (!list_empty(&mdsc->cap_dirty)) {
3312 ci = list_first_entry(&mdsc->cap_dirty, struct ceph_inode_info,
3313 i_dirty_item);
Sage Weil70b666c2011-05-27 09:24:26 -07003314 inode = &ci->vfs_inode;
3315 ihold(inode);
Sage Weildb354052011-05-24 11:46:31 -07003316 dout("flush_dirty_caps %p\n", inode);
Sage Weilafcdaea2009-10-14 14:27:38 -07003317 spin_unlock(&mdsc->cap_dirty_lock);
Sage Weil70b666c2011-05-27 09:24:26 -07003318 ceph_check_caps(ci, CHECK_CAPS_NODELAY|CHECK_CAPS_FLUSH, NULL);
3319 iput(inode);
Sage Weilafcdaea2009-10-14 14:27:38 -07003320 spin_lock(&mdsc->cap_dirty_lock);
3321 }
3322 spin_unlock(&mdsc->cap_dirty_lock);
Sage Weildb354052011-05-24 11:46:31 -07003323 dout("flush_dirty_caps done\n");
Sage Weilafcdaea2009-10-14 14:27:38 -07003324}
3325
3326/*
Sage Weila8599bd2009-10-06 11:31:12 -07003327 * Drop open file reference. If we were the last open file,
3328 * we may need to release capabilities to the MDS (or schedule
3329 * their delayed release).
3330 */
3331void ceph_put_fmode(struct ceph_inode_info *ci, int fmode)
3332{
3333 struct inode *inode = &ci->vfs_inode;
3334 int last = 0;
3335
Sage Weilbe655592011-11-30 09:47:09 -08003336 spin_lock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07003337 dout("put_fmode %p fmode %d %d -> %d\n", inode, fmode,
3338 ci->i_nr_by_mode[fmode], ci->i_nr_by_mode[fmode]-1);
3339 BUG_ON(ci->i_nr_by_mode[fmode] == 0);
3340 if (--ci->i_nr_by_mode[fmode] == 0)
3341 last++;
Sage Weilbe655592011-11-30 09:47:09 -08003342 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07003343
3344 if (last && ci->i_vino.snap == CEPH_NOSNAP)
3345 ceph_check_caps(ci, 0, NULL);
3346}
3347
3348/*
3349 * Helpers for embedding cap and dentry lease releases into mds
3350 * requests.
3351 *
3352 * @force is used by dentry_release (below) to force inclusion of a
3353 * record for the directory inode, even when there aren't any caps to
3354 * drop.
3355 */
3356int ceph_encode_inode_release(void **p, struct inode *inode,
3357 int mds, int drop, int unless, int force)
3358{
3359 struct ceph_inode_info *ci = ceph_inode(inode);
3360 struct ceph_cap *cap;
3361 struct ceph_mds_request_release *rel = *p;
Sage Weilec97f882010-06-24 15:12:37 -07003362 int used, dirty;
Sage Weila8599bd2009-10-06 11:31:12 -07003363 int ret = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07003364
Sage Weilbe655592011-11-30 09:47:09 -08003365 spin_lock(&ci->i_ceph_lock);
Sage Weil916623d2010-03-16 15:01:07 -07003366 used = __ceph_caps_used(ci);
Sage Weilec97f882010-06-24 15:12:37 -07003367 dirty = __ceph_caps_dirty(ci);
Sage Weil916623d2010-03-16 15:01:07 -07003368
Sage Weilec97f882010-06-24 15:12:37 -07003369 dout("encode_inode_release %p mds%d used|dirty %s drop %s unless %s\n",
3370 inode, mds, ceph_cap_string(used|dirty), ceph_cap_string(drop),
Sage Weil916623d2010-03-16 15:01:07 -07003371 ceph_cap_string(unless));
3372
Sage Weilec97f882010-06-24 15:12:37 -07003373 /* only drop unused, clean caps */
3374 drop &= ~(used | dirty);
Sage Weil916623d2010-03-16 15:01:07 -07003375
Sage Weila8599bd2009-10-06 11:31:12 -07003376 cap = __get_cap_for_mds(ci, mds);
3377 if (cap && __cap_is_valid(cap)) {
3378 if (force ||
3379 ((cap->issued & drop) &&
3380 (cap->issued & unless) == 0)) {
3381 if ((cap->issued & drop) &&
3382 (cap->issued & unless) == 0) {
Yan, Zhengbb137f82013-06-03 18:22:17 +08003383 int wanted = __ceph_caps_wanted(ci);
3384 if ((ci->i_ceph_flags & CEPH_I_NODELAY) == 0)
3385 wanted |= cap->mds_wanted;
3386 dout("encode_inode_release %p cap %p "
3387 "%s -> %s, wanted %s -> %s\n", inode, cap,
Sage Weila8599bd2009-10-06 11:31:12 -07003388 ceph_cap_string(cap->issued),
Yan, Zhengbb137f82013-06-03 18:22:17 +08003389 ceph_cap_string(cap->issued & ~drop),
3390 ceph_cap_string(cap->mds_wanted),
3391 ceph_cap_string(wanted));
3392
Sage Weila8599bd2009-10-06 11:31:12 -07003393 cap->issued &= ~drop;
3394 cap->implemented &= ~drop;
Yan, Zhengbb137f82013-06-03 18:22:17 +08003395 cap->mds_wanted = wanted;
Sage Weila8599bd2009-10-06 11:31:12 -07003396 } else {
3397 dout("encode_inode_release %p cap %p %s"
3398 " (force)\n", inode, cap,
3399 ceph_cap_string(cap->issued));
3400 }
3401
3402 rel->ino = cpu_to_le64(ceph_ino(inode));
3403 rel->cap_id = cpu_to_le64(cap->cap_id);
3404 rel->seq = cpu_to_le32(cap->seq);
Himangi Saraogi08a0f242014-07-23 20:11:11 +05303405 rel->issue_seq = cpu_to_le32(cap->issue_seq);
Sage Weila8599bd2009-10-06 11:31:12 -07003406 rel->mseq = cpu_to_le32(cap->mseq);
Yan, Zhengfd7b95c2014-04-17 08:02:02 +08003407 rel->caps = cpu_to_le32(cap->implemented);
Sage Weila8599bd2009-10-06 11:31:12 -07003408 rel->wanted = cpu_to_le32(cap->mds_wanted);
3409 rel->dname_len = 0;
3410 rel->dname_seq = 0;
3411 *p += sizeof(*rel);
3412 ret = 1;
3413 } else {
3414 dout("encode_inode_release %p cap %p %s\n",
3415 inode, cap, ceph_cap_string(cap->issued));
3416 }
3417 }
Sage Weilbe655592011-11-30 09:47:09 -08003418 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07003419 return ret;
3420}
3421
3422int ceph_encode_dentry_release(void **p, struct dentry *dentry,
3423 int mds, int drop, int unless)
3424{
David Howells2b0143b2015-03-17 22:25:59 +00003425 struct inode *dir = d_inode(dentry->d_parent);
Sage Weila8599bd2009-10-06 11:31:12 -07003426 struct ceph_mds_request_release *rel = *p;
3427 struct ceph_dentry_info *di = ceph_dentry(dentry);
3428 int force = 0;
3429 int ret;
3430
3431 /*
3432 * force an record for the directory caps if we have a dentry lease.
Sage Weilbe655592011-11-30 09:47:09 -08003433 * this is racy (can't take i_ceph_lock and d_lock together), but it
Sage Weila8599bd2009-10-06 11:31:12 -07003434 * doesn't have to be perfect; the mds will revoke anything we don't
3435 * release.
3436 */
3437 spin_lock(&dentry->d_lock);
3438 if (di->lease_session && di->lease_session->s_mds == mds)
3439 force = 1;
3440 spin_unlock(&dentry->d_lock);
3441
3442 ret = ceph_encode_inode_release(p, dir, mds, drop, unless, force);
3443
3444 spin_lock(&dentry->d_lock);
3445 if (ret && di->lease_session && di->lease_session->s_mds == mds) {
3446 dout("encode_dentry_release %p mds%d seq %d\n",
3447 dentry, mds, (int)di->lease_seq);
3448 rel->dname_len = cpu_to_le32(dentry->d_name.len);
3449 memcpy(*p, dentry->d_name.name, dentry->d_name.len);
3450 *p += dentry->d_name.len;
3451 rel->dname_seq = cpu_to_le32(di->lease_seq);
Sage Weil1dadcce2010-07-23 13:54:21 -07003452 __ceph_mdsc_drop_dentry_lease(dentry);
Sage Weila8599bd2009-10-06 11:31:12 -07003453 }
3454 spin_unlock(&dentry->d_lock);
3455 return ret;
3456}