blob: 16e6ded0b7f281bf72e8074b9d2896713fe4aef2 [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
Yan, Zheng0e294382016-07-04 18:06:41 +080043static u64 __get_oldest_flush_tid(struct ceph_mds_client *mdsc);
Yan, Zheng7bc00fd2016-07-07 18:34:45 +080044static void __kick_flushing_caps(struct ceph_mds_client *mdsc,
45 struct ceph_mds_session *session,
46 struct ceph_inode_info *ci,
47 u64 oldest_flush_tid);
Sage Weila8599bd2009-10-06 11:31:12 -070048
49/*
50 * Generate readable cap strings for debugging output.
51 */
52#define MAX_CAP_STR 20
53static char cap_str[MAX_CAP_STR][40];
54static DEFINE_SPINLOCK(cap_str_lock);
55static int last_cap_str;
56
57static char *gcap_string(char *s, int c)
58{
59 if (c & CEPH_CAP_GSHARED)
60 *s++ = 's';
61 if (c & CEPH_CAP_GEXCL)
62 *s++ = 'x';
63 if (c & CEPH_CAP_GCACHE)
64 *s++ = 'c';
65 if (c & CEPH_CAP_GRD)
66 *s++ = 'r';
67 if (c & CEPH_CAP_GWR)
68 *s++ = 'w';
69 if (c & CEPH_CAP_GBUFFER)
70 *s++ = 'b';
71 if (c & CEPH_CAP_GLAZYIO)
72 *s++ = 'l';
73 return s;
74}
75
76const char *ceph_cap_string(int caps)
77{
78 int i;
79 char *s;
80 int c;
81
82 spin_lock(&cap_str_lock);
83 i = last_cap_str++;
84 if (last_cap_str == MAX_CAP_STR)
85 last_cap_str = 0;
86 spin_unlock(&cap_str_lock);
87
88 s = cap_str[i];
89
90 if (caps & CEPH_CAP_PIN)
91 *s++ = 'p';
92
93 c = (caps >> CEPH_CAP_SAUTH) & 3;
94 if (c) {
95 *s++ = 'A';
96 s = gcap_string(s, c);
97 }
98
99 c = (caps >> CEPH_CAP_SLINK) & 3;
100 if (c) {
101 *s++ = 'L';
102 s = gcap_string(s, c);
103 }
104
105 c = (caps >> CEPH_CAP_SXATTR) & 3;
106 if (c) {
107 *s++ = 'X';
108 s = gcap_string(s, c);
109 }
110
111 c = caps >> CEPH_CAP_SFILE;
112 if (c) {
113 *s++ = 'F';
114 s = gcap_string(s, c);
115 }
116
117 if (s == cap_str[i])
118 *s++ = '-';
119 *s = 0;
120 return cap_str[i];
121}
122
Yehuda Sadeh37151662010-06-17 16:16:12 -0700123void ceph_caps_init(struct ceph_mds_client *mdsc)
Sage Weila8599bd2009-10-06 11:31:12 -0700124{
Yehuda Sadeh37151662010-06-17 16:16:12 -0700125 INIT_LIST_HEAD(&mdsc->caps_list);
126 spin_lock_init(&mdsc->caps_list_lock);
Sage Weila8599bd2009-10-06 11:31:12 -0700127}
128
Yehuda Sadeh37151662010-06-17 16:16:12 -0700129void ceph_caps_finalize(struct ceph_mds_client *mdsc)
Sage Weila8599bd2009-10-06 11:31:12 -0700130{
131 struct ceph_cap *cap;
132
Yehuda Sadeh37151662010-06-17 16:16:12 -0700133 spin_lock(&mdsc->caps_list_lock);
134 while (!list_empty(&mdsc->caps_list)) {
135 cap = list_first_entry(&mdsc->caps_list,
136 struct ceph_cap, caps_item);
Sage Weila8599bd2009-10-06 11:31:12 -0700137 list_del(&cap->caps_item);
138 kmem_cache_free(ceph_cap_cachep, cap);
139 }
Yehuda Sadeh37151662010-06-17 16:16:12 -0700140 mdsc->caps_total_count = 0;
141 mdsc->caps_avail_count = 0;
142 mdsc->caps_use_count = 0;
143 mdsc->caps_reserve_count = 0;
144 mdsc->caps_min_count = 0;
145 spin_unlock(&mdsc->caps_list_lock);
Sage Weil85ccce42010-02-17 10:02:43 -0800146}
147
Yehuda Sadeh37151662010-06-17 16:16:12 -0700148void ceph_adjust_min_caps(struct ceph_mds_client *mdsc, int delta)
Sage Weil85ccce42010-02-17 10:02:43 -0800149{
Yehuda Sadeh37151662010-06-17 16:16:12 -0700150 spin_lock(&mdsc->caps_list_lock);
151 mdsc->caps_min_count += delta;
152 BUG_ON(mdsc->caps_min_count < 0);
153 spin_unlock(&mdsc->caps_list_lock);
Sage Weila8599bd2009-10-06 11:31:12 -0700154}
155
majianpeng93faca62013-06-26 11:15:27 +0800156void ceph_reserve_caps(struct ceph_mds_client *mdsc,
Yehuda Sadeh37151662010-06-17 16:16:12 -0700157 struct ceph_cap_reservation *ctx, int need)
Sage Weila8599bd2009-10-06 11:31:12 -0700158{
159 int i;
160 struct ceph_cap *cap;
161 int have;
162 int alloc = 0;
163 LIST_HEAD(newcaps);
Sage Weila8599bd2009-10-06 11:31:12 -0700164
165 dout("reserve caps ctx=%p need=%d\n", ctx, need);
166
167 /* first reserve any caps that are already allocated */
Yehuda Sadeh37151662010-06-17 16:16:12 -0700168 spin_lock(&mdsc->caps_list_lock);
169 if (mdsc->caps_avail_count >= need)
Sage Weila8599bd2009-10-06 11:31:12 -0700170 have = need;
171 else
Yehuda Sadeh37151662010-06-17 16:16:12 -0700172 have = mdsc->caps_avail_count;
173 mdsc->caps_avail_count -= have;
174 mdsc->caps_reserve_count += have;
175 BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count +
176 mdsc->caps_reserve_count +
177 mdsc->caps_avail_count);
178 spin_unlock(&mdsc->caps_list_lock);
Sage Weila8599bd2009-10-06 11:31:12 -0700179
180 for (i = have; i < need; i++) {
181 cap = kmem_cache_alloc(ceph_cap_cachep, GFP_NOFS);
majianpeng93faca62013-06-26 11:15:27 +0800182 if (!cap)
183 break;
Sage Weila8599bd2009-10-06 11:31:12 -0700184 list_add(&cap->caps_item, &newcaps);
185 alloc++;
186 }
majianpeng93faca62013-06-26 11:15:27 +0800187 /* we didn't manage to reserve as much as we needed */
188 if (have + alloc != need)
189 pr_warn("reserve caps ctx=%p ENOMEM need=%d got=%d\n",
190 ctx, need, have + alloc);
Sage Weila8599bd2009-10-06 11:31:12 -0700191
Yehuda Sadeh37151662010-06-17 16:16:12 -0700192 spin_lock(&mdsc->caps_list_lock);
193 mdsc->caps_total_count += alloc;
194 mdsc->caps_reserve_count += alloc;
195 list_splice(&newcaps, &mdsc->caps_list);
Sage Weila8599bd2009-10-06 11:31:12 -0700196
Yehuda Sadeh37151662010-06-17 16:16:12 -0700197 BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count +
198 mdsc->caps_reserve_count +
199 mdsc->caps_avail_count);
200 spin_unlock(&mdsc->caps_list_lock);
Sage Weila8599bd2009-10-06 11:31:12 -0700201
202 ctx->count = need;
203 dout("reserve caps ctx=%p %d = %d used + %d resv + %d avail\n",
Yehuda Sadeh37151662010-06-17 16:16:12 -0700204 ctx, mdsc->caps_total_count, mdsc->caps_use_count,
205 mdsc->caps_reserve_count, mdsc->caps_avail_count);
Sage Weila8599bd2009-10-06 11:31:12 -0700206}
207
Yehuda Sadeh37151662010-06-17 16:16:12 -0700208int ceph_unreserve_caps(struct ceph_mds_client *mdsc,
209 struct ceph_cap_reservation *ctx)
Sage Weila8599bd2009-10-06 11:31:12 -0700210{
211 dout("unreserve caps ctx=%p count=%d\n", ctx, ctx->count);
212 if (ctx->count) {
Yehuda Sadeh37151662010-06-17 16:16:12 -0700213 spin_lock(&mdsc->caps_list_lock);
214 BUG_ON(mdsc->caps_reserve_count < ctx->count);
215 mdsc->caps_reserve_count -= ctx->count;
216 mdsc->caps_avail_count += ctx->count;
Sage Weila8599bd2009-10-06 11:31:12 -0700217 ctx->count = 0;
218 dout("unreserve caps %d = %d used + %d resv + %d avail\n",
Yehuda Sadeh37151662010-06-17 16:16:12 -0700219 mdsc->caps_total_count, mdsc->caps_use_count,
220 mdsc->caps_reserve_count, mdsc->caps_avail_count);
221 BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count +
222 mdsc->caps_reserve_count +
223 mdsc->caps_avail_count);
224 spin_unlock(&mdsc->caps_list_lock);
Sage Weila8599bd2009-10-06 11:31:12 -0700225 }
226 return 0;
227}
228
Yan, Zhengd9df2782014-04-18 09:57:11 +0800229struct ceph_cap *ceph_get_cap(struct ceph_mds_client *mdsc,
230 struct ceph_cap_reservation *ctx)
Sage Weila8599bd2009-10-06 11:31:12 -0700231{
232 struct ceph_cap *cap = NULL;
233
234 /* temporary, until we do something about cap import/export */
Sage Weil443b3762010-06-29 09:28:39 -0700235 if (!ctx) {
236 cap = kmem_cache_alloc(ceph_cap_cachep, GFP_NOFS);
237 if (cap) {
Yan, Zheng4d1d0532012-11-03 10:32:37 +0800238 spin_lock(&mdsc->caps_list_lock);
Yehuda Sadeh37151662010-06-17 16:16:12 -0700239 mdsc->caps_use_count++;
240 mdsc->caps_total_count++;
Yan, Zheng4d1d0532012-11-03 10:32:37 +0800241 spin_unlock(&mdsc->caps_list_lock);
Sage Weil443b3762010-06-29 09:28:39 -0700242 }
243 return cap;
244 }
Sage Weila8599bd2009-10-06 11:31:12 -0700245
Yehuda Sadeh37151662010-06-17 16:16:12 -0700246 spin_lock(&mdsc->caps_list_lock);
Sage Weila8599bd2009-10-06 11:31:12 -0700247 dout("get_cap ctx=%p (%d) %d = %d used + %d resv + %d avail\n",
Yehuda Sadeh37151662010-06-17 16:16:12 -0700248 ctx, ctx->count, mdsc->caps_total_count, mdsc->caps_use_count,
249 mdsc->caps_reserve_count, mdsc->caps_avail_count);
Sage Weila8599bd2009-10-06 11:31:12 -0700250 BUG_ON(!ctx->count);
Yehuda Sadeh37151662010-06-17 16:16:12 -0700251 BUG_ON(ctx->count > mdsc->caps_reserve_count);
252 BUG_ON(list_empty(&mdsc->caps_list));
Sage Weila8599bd2009-10-06 11:31:12 -0700253
254 ctx->count--;
Yehuda Sadeh37151662010-06-17 16:16:12 -0700255 mdsc->caps_reserve_count--;
256 mdsc->caps_use_count++;
Sage Weila8599bd2009-10-06 11:31:12 -0700257
Yehuda Sadeh37151662010-06-17 16:16:12 -0700258 cap = list_first_entry(&mdsc->caps_list, struct ceph_cap, caps_item);
Sage Weila8599bd2009-10-06 11:31:12 -0700259 list_del(&cap->caps_item);
260
Yehuda Sadeh37151662010-06-17 16:16:12 -0700261 BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count +
262 mdsc->caps_reserve_count + mdsc->caps_avail_count);
263 spin_unlock(&mdsc->caps_list_lock);
Sage Weila8599bd2009-10-06 11:31:12 -0700264 return cap;
265}
266
Yehuda Sadeh37151662010-06-17 16:16:12 -0700267void ceph_put_cap(struct ceph_mds_client *mdsc, struct ceph_cap *cap)
Sage Weila8599bd2009-10-06 11:31:12 -0700268{
Yehuda Sadeh37151662010-06-17 16:16:12 -0700269 spin_lock(&mdsc->caps_list_lock);
Sage Weil7c1332b2010-02-16 11:39:45 -0800270 dout("put_cap %p %d = %d used + %d resv + %d avail\n",
Yehuda Sadeh37151662010-06-17 16:16:12 -0700271 cap, mdsc->caps_total_count, mdsc->caps_use_count,
272 mdsc->caps_reserve_count, mdsc->caps_avail_count);
273 mdsc->caps_use_count--;
Sage Weila8599bd2009-10-06 11:31:12 -0700274 /*
Sage Weil85ccce42010-02-17 10:02:43 -0800275 * Keep some preallocated caps around (ceph_min_count), to
276 * avoid lots of free/alloc churn.
Sage Weila8599bd2009-10-06 11:31:12 -0700277 */
Yehuda Sadeh37151662010-06-17 16:16:12 -0700278 if (mdsc->caps_avail_count >= mdsc->caps_reserve_count +
279 mdsc->caps_min_count) {
280 mdsc->caps_total_count--;
Sage Weila8599bd2009-10-06 11:31:12 -0700281 kmem_cache_free(ceph_cap_cachep, cap);
282 } else {
Yehuda Sadeh37151662010-06-17 16:16:12 -0700283 mdsc->caps_avail_count++;
284 list_add(&cap->caps_item, &mdsc->caps_list);
Sage Weila8599bd2009-10-06 11:31:12 -0700285 }
286
Yehuda Sadeh37151662010-06-17 16:16:12 -0700287 BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count +
288 mdsc->caps_reserve_count + mdsc->caps_avail_count);
289 spin_unlock(&mdsc->caps_list_lock);
Sage Weila8599bd2009-10-06 11:31:12 -0700290}
291
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700292void ceph_reservation_status(struct ceph_fs_client *fsc,
Sage Weil85ccce42010-02-17 10:02:43 -0800293 int *total, int *avail, int *used, int *reserved,
294 int *min)
Sage Weila8599bd2009-10-06 11:31:12 -0700295{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700296 struct ceph_mds_client *mdsc = fsc->mdsc;
Yehuda Sadeh37151662010-06-17 16:16:12 -0700297
Sage Weila8599bd2009-10-06 11:31:12 -0700298 if (total)
Yehuda Sadeh37151662010-06-17 16:16:12 -0700299 *total = mdsc->caps_total_count;
Sage Weila8599bd2009-10-06 11:31:12 -0700300 if (avail)
Yehuda Sadeh37151662010-06-17 16:16:12 -0700301 *avail = mdsc->caps_avail_count;
Sage Weila8599bd2009-10-06 11:31:12 -0700302 if (used)
Yehuda Sadeh37151662010-06-17 16:16:12 -0700303 *used = mdsc->caps_use_count;
Sage Weila8599bd2009-10-06 11:31:12 -0700304 if (reserved)
Yehuda Sadeh37151662010-06-17 16:16:12 -0700305 *reserved = mdsc->caps_reserve_count;
Sage Weil85ccce42010-02-17 10:02:43 -0800306 if (min)
Yehuda Sadeh37151662010-06-17 16:16:12 -0700307 *min = mdsc->caps_min_count;
Sage Weila8599bd2009-10-06 11:31:12 -0700308}
309
310/*
311 * Find ceph_cap for given mds, if any.
312 *
Sage Weilbe655592011-11-30 09:47:09 -0800313 * Called with i_ceph_lock held.
Sage Weila8599bd2009-10-06 11:31:12 -0700314 */
315static struct ceph_cap *__get_cap_for_mds(struct ceph_inode_info *ci, int mds)
316{
317 struct ceph_cap *cap;
318 struct rb_node *n = ci->i_caps.rb_node;
319
320 while (n) {
321 cap = rb_entry(n, struct ceph_cap, ci_node);
322 if (mds < cap->mds)
323 n = n->rb_left;
324 else if (mds > cap->mds)
325 n = n->rb_right;
326 else
327 return cap;
328 }
329 return NULL;
330}
331
Greg Farnum2bc50252010-06-30 12:44:34 -0700332struct ceph_cap *ceph_get_cap_for_mds(struct ceph_inode_info *ci, int mds)
333{
334 struct ceph_cap *cap;
335
Sage Weilbe655592011-11-30 09:47:09 -0800336 spin_lock(&ci->i_ceph_lock);
Greg Farnum2bc50252010-06-30 12:44:34 -0700337 cap = __get_cap_for_mds(ci, mds);
Sage Weilbe655592011-11-30 09:47:09 -0800338 spin_unlock(&ci->i_ceph_lock);
Greg Farnum2bc50252010-06-30 12:44:34 -0700339 return cap;
340}
341
Sage Weila8599bd2009-10-06 11:31:12 -0700342/*
Sage Weil33caad32010-05-26 14:31:27 -0700343 * Return id of any MDS with a cap, preferably FILE_WR|BUFFER|EXCL, else -1.
Sage Weila8599bd2009-10-06 11:31:12 -0700344 */
Sage Weilca81f3f2010-06-10 14:21:36 -0700345static int __ceph_get_cap_mds(struct ceph_inode_info *ci)
Sage Weila8599bd2009-10-06 11:31:12 -0700346{
347 struct ceph_cap *cap;
348 int mds = -1;
349 struct rb_node *p;
350
Sage Weil33caad32010-05-26 14:31:27 -0700351 /* prefer mds with WR|BUFFER|EXCL caps */
Sage Weila8599bd2009-10-06 11:31:12 -0700352 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
353 cap = rb_entry(p, struct ceph_cap, ci_node);
354 mds = cap->mds;
Sage Weila8599bd2009-10-06 11:31:12 -0700355 if (cap->issued & (CEPH_CAP_FILE_WR |
356 CEPH_CAP_FILE_BUFFER |
357 CEPH_CAP_FILE_EXCL))
358 break;
359 }
360 return mds;
361}
362
363int ceph_get_cap_mds(struct inode *inode)
364{
Sage Weilbe655592011-11-30 09:47:09 -0800365 struct ceph_inode_info *ci = ceph_inode(inode);
Sage Weila8599bd2009-10-06 11:31:12 -0700366 int mds;
Sage Weilbe655592011-11-30 09:47:09 -0800367 spin_lock(&ci->i_ceph_lock);
Sage Weilca81f3f2010-06-10 14:21:36 -0700368 mds = __ceph_get_cap_mds(ceph_inode(inode));
Sage Weilbe655592011-11-30 09:47:09 -0800369 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -0700370 return mds;
371}
372
373/*
Sage Weilbe655592011-11-30 09:47:09 -0800374 * Called under i_ceph_lock.
Sage Weila8599bd2009-10-06 11:31:12 -0700375 */
376static void __insert_cap_node(struct ceph_inode_info *ci,
377 struct ceph_cap *new)
378{
379 struct rb_node **p = &ci->i_caps.rb_node;
380 struct rb_node *parent = NULL;
381 struct ceph_cap *cap = NULL;
382
383 while (*p) {
384 parent = *p;
385 cap = rb_entry(parent, struct ceph_cap, ci_node);
386 if (new->mds < cap->mds)
387 p = &(*p)->rb_left;
388 else if (new->mds > cap->mds)
389 p = &(*p)->rb_right;
390 else
391 BUG();
392 }
393
394 rb_link_node(&new->ci_node, parent, p);
395 rb_insert_color(&new->ci_node, &ci->i_caps);
396}
397
398/*
399 * (re)set cap hold timeouts, which control the delayed release
400 * of unused caps back to the MDS. Should be called on cap use.
401 */
402static void __cap_set_timeouts(struct ceph_mds_client *mdsc,
403 struct ceph_inode_info *ci)
404{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700405 struct ceph_mount_options *ma = mdsc->fsc->mount_options;
Sage Weila8599bd2009-10-06 11:31:12 -0700406
407 ci->i_hold_caps_min = round_jiffies(jiffies +
408 ma->caps_wanted_delay_min * HZ);
409 ci->i_hold_caps_max = round_jiffies(jiffies +
410 ma->caps_wanted_delay_max * HZ);
411 dout("__cap_set_timeouts %p min %lu max %lu\n", &ci->vfs_inode,
412 ci->i_hold_caps_min - jiffies, ci->i_hold_caps_max - jiffies);
413}
414
415/*
416 * (Re)queue cap at the end of the delayed cap release list.
417 *
418 * If I_FLUSH is set, leave the inode at the front of the list.
419 *
Sage Weilbe655592011-11-30 09:47:09 -0800420 * Caller holds i_ceph_lock
Sage Weila8599bd2009-10-06 11:31:12 -0700421 * -> we take mdsc->cap_delay_lock
422 */
423static void __cap_delay_requeue(struct ceph_mds_client *mdsc,
424 struct ceph_inode_info *ci)
425{
426 __cap_set_timeouts(mdsc, ci);
427 dout("__cap_delay_requeue %p flags %d at %lu\n", &ci->vfs_inode,
428 ci->i_ceph_flags, ci->i_hold_caps_max);
429 if (!mdsc->stopping) {
430 spin_lock(&mdsc->cap_delay_lock);
431 if (!list_empty(&ci->i_cap_delay_list)) {
432 if (ci->i_ceph_flags & CEPH_I_FLUSH)
433 goto no_change;
434 list_del_init(&ci->i_cap_delay_list);
435 }
436 list_add_tail(&ci->i_cap_delay_list, &mdsc->cap_delay_list);
437no_change:
438 spin_unlock(&mdsc->cap_delay_lock);
439 }
440}
441
442/*
443 * Queue an inode for immediate writeback. Mark inode with I_FLUSH,
444 * indicating we should send a cap message to flush dirty metadata
445 * asap, and move to the front of the delayed cap list.
446 */
447static void __cap_delay_requeue_front(struct ceph_mds_client *mdsc,
448 struct ceph_inode_info *ci)
449{
450 dout("__cap_delay_requeue_front %p\n", &ci->vfs_inode);
451 spin_lock(&mdsc->cap_delay_lock);
452 ci->i_ceph_flags |= CEPH_I_FLUSH;
453 if (!list_empty(&ci->i_cap_delay_list))
454 list_del_init(&ci->i_cap_delay_list);
455 list_add(&ci->i_cap_delay_list, &mdsc->cap_delay_list);
456 spin_unlock(&mdsc->cap_delay_lock);
457}
458
459/*
460 * Cancel delayed work on cap.
461 *
Sage Weilbe655592011-11-30 09:47:09 -0800462 * Caller must hold i_ceph_lock.
Sage Weila8599bd2009-10-06 11:31:12 -0700463 */
464static void __cap_delay_cancel(struct ceph_mds_client *mdsc,
465 struct ceph_inode_info *ci)
466{
467 dout("__cap_delay_cancel %p\n", &ci->vfs_inode);
468 if (list_empty(&ci->i_cap_delay_list))
469 return;
470 spin_lock(&mdsc->cap_delay_lock);
471 list_del_init(&ci->i_cap_delay_list);
472 spin_unlock(&mdsc->cap_delay_lock);
473}
474
475/*
476 * Common issue checks for add_cap, handle_cap_grant.
477 */
478static void __check_cap_issue(struct ceph_inode_info *ci, struct ceph_cap *cap,
479 unsigned issued)
480{
481 unsigned had = __ceph_caps_issued(ci, NULL);
482
483 /*
484 * Each time we receive FILE_CACHE anew, we increment
485 * i_rdcache_gen.
486 */
Sage Weil29625072010-05-27 10:40:43 -0700487 if ((issued & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) &&
Milosz Tanski99ccbd22013-08-21 17:29:54 -0400488 (had & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) == 0) {
Sage Weila8599bd2009-10-06 11:31:12 -0700489 ci->i_rdcache_gen++;
Milosz Tanski99ccbd22013-08-21 17:29:54 -0400490 }
Sage Weila8599bd2009-10-06 11:31:12 -0700491
492 /*
Yan, Zheng2f276c52013-03-13 19:44:32 +0800493 * if we are newly issued FILE_SHARED, mark dir not complete; we
Sage Weila8599bd2009-10-06 11:31:12 -0700494 * don't know what happened to this directory while we didn't
495 * have the cap.
496 */
497 if ((issued & CEPH_CAP_FILE_SHARED) &&
498 (had & CEPH_CAP_FILE_SHARED) == 0) {
499 ci->i_shared_gen++;
Yan, Zhenga8673d62013-02-18 16:38:14 +0800500 if (S_ISDIR(ci->vfs_inode.i_mode)) {
501 dout(" marking %p NOT complete\n", &ci->vfs_inode);
Yan, Zheng2f276c52013-03-13 19:44:32 +0800502 __ceph_dir_clear_complete(ci);
Yan, Zhenga8673d62013-02-18 16:38:14 +0800503 }
Sage Weila8599bd2009-10-06 11:31:12 -0700504 }
505}
506
507/*
508 * Add a capability under the given MDS session.
509 *
510 * Caller should hold session snap_rwsem (read) and s_mutex.
511 *
512 * @fmode is the open file mode, if we are opening a file, otherwise
513 * it is < 0. (This is so we can atomically add the cap and add an
514 * open file reference to it.)
515 */
Yan, Zhengd9df2782014-04-18 09:57:11 +0800516void ceph_add_cap(struct inode *inode,
517 struct ceph_mds_session *session, u64 cap_id,
518 int fmode, unsigned issued, unsigned wanted,
519 unsigned seq, unsigned mseq, u64 realmino, int flags,
520 struct ceph_cap **new_cap)
Sage Weila8599bd2009-10-06 11:31:12 -0700521{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700522 struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
Sage Weila8599bd2009-10-06 11:31:12 -0700523 struct ceph_inode_info *ci = ceph_inode(inode);
Sage Weila8599bd2009-10-06 11:31:12 -0700524 struct ceph_cap *cap;
525 int mds = session->s_mds;
526 int actual_wanted;
527
528 dout("add_cap %p mds%d cap %llx %s seq %d\n", inode,
529 session->s_mds, cap_id, ceph_cap_string(issued), seq);
530
531 /*
532 * If we are opening the file, include file mode wanted bits
533 * in wanted.
534 */
535 if (fmode >= 0)
536 wanted |= ceph_caps_for_mode(fmode);
537
Sage Weila8599bd2009-10-06 11:31:12 -0700538 cap = __get_cap_for_mds(ci, mds);
539 if (!cap) {
Yan, Zhengd9df2782014-04-18 09:57:11 +0800540 cap = *new_cap;
541 *new_cap = NULL;
Sage Weila8599bd2009-10-06 11:31:12 -0700542
543 cap->issued = 0;
544 cap->implemented = 0;
545 cap->mds = mds;
546 cap->mds_wanted = 0;
Yan, Zheng964266c2013-02-27 09:26:09 +0800547 cap->mseq = 0;
Sage Weila8599bd2009-10-06 11:31:12 -0700548
549 cap->ci = ci;
550 __insert_cap_node(ci, cap);
551
Sage Weila8599bd2009-10-06 11:31:12 -0700552 /* add to session cap list */
553 cap->session = session;
554 spin_lock(&session->s_cap_lock);
555 list_add_tail(&cap->session_caps, &session->s_caps);
556 session->s_nr_caps++;
557 spin_unlock(&session->s_cap_lock);
Yan, Zheng11df2df2013-11-24 14:44:38 +0800558 } else {
Yan, Zheng11df2df2013-11-24 14:44:38 +0800559 /*
560 * auth mds of the inode changed. we received the cap export
561 * message, but still haven't received the cap import message.
562 * handle_cap_export() updated the new auth MDS' cap.
563 *
564 * "ceph_seq_cmp(seq, cap->seq) <= 0" means we are processing
565 * a message that was send before the cap import message. So
566 * don't remove caps.
567 */
568 if (ceph_seq_cmp(seq, cap->seq) <= 0) {
569 WARN_ON(cap != ci->i_auth_cap);
570 WARN_ON(cap->cap_id != cap_id);
571 seq = cap->seq;
572 mseq = cap->mseq;
573 issued |= cap->issued;
574 flags |= CEPH_CAP_FLAG_AUTH;
575 }
576 }
Sage Weila8599bd2009-10-06 11:31:12 -0700577
578 if (!ci->i_snap_realm) {
579 /*
580 * add this inode to the appropriate snap realm
581 */
582 struct ceph_snap_realm *realm = ceph_lookup_snap_realm(mdsc,
583 realmino);
584 if (realm) {
Sage Weila8599bd2009-10-06 11:31:12 -0700585 spin_lock(&realm->inodes_with_caps_lock);
586 ci->i_snap_realm = realm;
587 list_add(&ci->i_snap_realm_item,
588 &realm->inodes_with_caps);
589 spin_unlock(&realm->inodes_with_caps_lock);
590 } else {
591 pr_err("ceph_add_cap: couldn't find snap realm %llx\n",
592 realmino);
Sage Weilb8cd07e2010-07-16 12:00:02 -0700593 WARN_ON(!realm);
Sage Weila8599bd2009-10-06 11:31:12 -0700594 }
595 }
596
597 __check_cap_issue(ci, cap, issued);
598
599 /*
600 * If we are issued caps we don't want, or the mds' wanted
601 * value appears to be off, queue a check so we'll release
602 * later and/or update the mds wanted value.
603 */
604 actual_wanted = __ceph_caps_wanted(ci);
605 if ((wanted & ~actual_wanted) ||
606 (issued & ~actual_wanted & CEPH_CAP_ANY_WR)) {
607 dout(" issued %s, mds wanted %s, actual %s, queueing\n",
608 ceph_cap_string(issued), ceph_cap_string(wanted),
609 ceph_cap_string(actual_wanted));
610 __cap_delay_requeue(mdsc, ci);
611 }
612
Yan, Zhengb8c2f3a2013-05-31 16:37:11 +0800613 if (flags & CEPH_CAP_FLAG_AUTH) {
614 if (ci->i_auth_cap == NULL ||
Yan, Zhengd9ffc4f2014-03-18 10:15:29 +0800615 ceph_seq_cmp(ci->i_auth_cap->mseq, mseq) < 0) {
Yan, Zhengb8c2f3a2013-05-31 16:37:11 +0800616 ci->i_auth_cap = cap;
Yan, Zhengd9ffc4f2014-03-18 10:15:29 +0800617 cap->mds_wanted = wanted;
618 }
Yan, Zheng11df2df2013-11-24 14:44:38 +0800619 } else {
620 WARN_ON(ci->i_auth_cap == cap);
Yan, Zheng8a92a112013-01-04 14:28:07 +0800621 }
Sage Weila8599bd2009-10-06 11:31:12 -0700622
623 dout("add_cap inode %p (%llx.%llx) cap %p %s now %s seq %d mds%d\n",
624 inode, ceph_vinop(inode), cap, ceph_cap_string(issued),
625 ceph_cap_string(issued|cap->issued), seq, mds);
626 cap->cap_id = cap_id;
627 cap->issued = issued;
628 cap->implemented |= issued;
Yan, Zhengd1b87802013-11-13 14:47:19 +0800629 if (ceph_seq_cmp(mseq, cap->mseq) > 0)
Yan, Zheng964266c2013-02-27 09:26:09 +0800630 cap->mds_wanted = wanted;
631 else
632 cap->mds_wanted |= wanted;
Sage Weila8599bd2009-10-06 11:31:12 -0700633 cap->seq = seq;
634 cap->issue_seq = seq;
635 cap->mseq = mseq;
Sage Weil685f9a5d2009-11-09 12:05:48 -0800636 cap->cap_gen = session->s_cap_gen;
Sage Weila8599bd2009-10-06 11:31:12 -0700637
638 if (fmode >= 0)
639 __ceph_get_fmode(ci, fmode);
Sage Weila8599bd2009-10-06 11:31:12 -0700640}
641
642/*
643 * Return true if cap has not timed out and belongs to the current
644 * generation of the MDS session (i.e. has not gone 'stale' due to
645 * us losing touch with the mds).
646 */
647static int __cap_is_valid(struct ceph_cap *cap)
648{
649 unsigned long ttl;
Sage Weilcdac8302009-11-10 16:02:23 -0800650 u32 gen;
Sage Weila8599bd2009-10-06 11:31:12 -0700651
Alex Elderd8fb02a2012-01-12 17:48:10 -0800652 spin_lock(&cap->session->s_gen_ttl_lock);
Sage Weila8599bd2009-10-06 11:31:12 -0700653 gen = cap->session->s_cap_gen;
654 ttl = cap->session->s_cap_ttl;
Alex Elderd8fb02a2012-01-12 17:48:10 -0800655 spin_unlock(&cap->session->s_gen_ttl_lock);
Sage Weila8599bd2009-10-06 11:31:12 -0700656
Sage Weil685f9a5d2009-11-09 12:05:48 -0800657 if (cap->cap_gen < gen || time_after_eq(jiffies, ttl)) {
Sage Weila8599bd2009-10-06 11:31:12 -0700658 dout("__cap_is_valid %p cap %p issued %s "
659 "but STALE (gen %u vs %u)\n", &cap->ci->vfs_inode,
Sage Weil685f9a5d2009-11-09 12:05:48 -0800660 cap, ceph_cap_string(cap->issued), cap->cap_gen, gen);
Sage Weila8599bd2009-10-06 11:31:12 -0700661 return 0;
662 }
663
664 return 1;
665}
666
667/*
668 * Return set of valid cap bits issued to us. Note that caps time
669 * out, and may be invalidated in bulk if the client session times out
670 * and session->s_cap_gen is bumped.
671 */
672int __ceph_caps_issued(struct ceph_inode_info *ci, int *implemented)
673{
Yan, Zhengd9df2782014-04-18 09:57:11 +0800674 int have = ci->i_snap_caps;
Sage Weila8599bd2009-10-06 11:31:12 -0700675 struct ceph_cap *cap;
676 struct rb_node *p;
677
678 if (implemented)
679 *implemented = 0;
680 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
681 cap = rb_entry(p, struct ceph_cap, ci_node);
682 if (!__cap_is_valid(cap))
683 continue;
684 dout("__ceph_caps_issued %p cap %p issued %s\n",
685 &ci->vfs_inode, cap, ceph_cap_string(cap->issued));
686 have |= cap->issued;
687 if (implemented)
688 *implemented |= cap->implemented;
689 }
Yan, Zhengb1530f52013-07-02 12:40:20 +0800690 /*
691 * exclude caps issued by non-auth MDS, but are been revoking
692 * by the auth MDS. The non-auth MDS should be revoking/exporting
693 * these caps, but the message is delayed.
694 */
695 if (ci->i_auth_cap) {
696 cap = ci->i_auth_cap;
697 have &= ~cap->implemented | cap->issued;
698 }
Sage Weila8599bd2009-10-06 11:31:12 -0700699 return have;
700}
701
702/*
703 * Get cap bits issued by caps other than @ocap
704 */
705int __ceph_caps_issued_other(struct ceph_inode_info *ci, struct ceph_cap *ocap)
706{
707 int have = ci->i_snap_caps;
708 struct ceph_cap *cap;
709 struct rb_node *p;
710
711 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
712 cap = rb_entry(p, struct ceph_cap, ci_node);
713 if (cap == ocap)
714 continue;
715 if (!__cap_is_valid(cap))
716 continue;
717 have |= cap->issued;
718 }
719 return have;
720}
721
722/*
723 * Move a cap to the end of the LRU (oldest caps at list head, newest
724 * at list tail).
725 */
726static void __touch_cap(struct ceph_cap *cap)
727{
728 struct ceph_mds_session *s = cap->session;
729
Sage Weila8599bd2009-10-06 11:31:12 -0700730 spin_lock(&s->s_cap_lock);
Sage Weil7c1332b2010-02-16 11:39:45 -0800731 if (s->s_cap_iterator == NULL) {
Sage Weil5dacf092009-12-21 20:40:34 -0800732 dout("__touch_cap %p cap %p mds%d\n", &cap->ci->vfs_inode, cap,
733 s->s_mds);
734 list_move_tail(&cap->session_caps, &s->s_caps);
735 } else {
736 dout("__touch_cap %p cap %p mds%d NOP, iterating over caps\n",
737 &cap->ci->vfs_inode, cap, s->s_mds);
738 }
Sage Weila8599bd2009-10-06 11:31:12 -0700739 spin_unlock(&s->s_cap_lock);
740}
741
742/*
743 * Check if we hold the given mask. If so, move the cap(s) to the
744 * front of their respective LRUs. (This is the preferred way for
745 * callers to check for caps they want.)
746 */
747int __ceph_caps_issued_mask(struct ceph_inode_info *ci, int mask, int touch)
748{
749 struct ceph_cap *cap;
750 struct rb_node *p;
751 int have = ci->i_snap_caps;
752
753 if ((have & mask) == mask) {
754 dout("__ceph_caps_issued_mask %p snap issued %s"
755 " (mask %s)\n", &ci->vfs_inode,
756 ceph_cap_string(have),
757 ceph_cap_string(mask));
758 return 1;
759 }
760
761 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
762 cap = rb_entry(p, struct ceph_cap, ci_node);
763 if (!__cap_is_valid(cap))
764 continue;
765 if ((cap->issued & mask) == mask) {
766 dout("__ceph_caps_issued_mask %p cap %p issued %s"
767 " (mask %s)\n", &ci->vfs_inode, cap,
768 ceph_cap_string(cap->issued),
769 ceph_cap_string(mask));
770 if (touch)
771 __touch_cap(cap);
772 return 1;
773 }
774
775 /* does a combination of caps satisfy mask? */
776 have |= cap->issued;
777 if ((have & mask) == mask) {
778 dout("__ceph_caps_issued_mask %p combo issued %s"
779 " (mask %s)\n", &ci->vfs_inode,
780 ceph_cap_string(cap->issued),
781 ceph_cap_string(mask));
782 if (touch) {
783 struct rb_node *q;
784
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300785 /* touch this + preceding caps */
Sage Weila8599bd2009-10-06 11:31:12 -0700786 __touch_cap(cap);
787 for (q = rb_first(&ci->i_caps); q != p;
788 q = rb_next(q)) {
789 cap = rb_entry(q, struct ceph_cap,
790 ci_node);
791 if (!__cap_is_valid(cap))
792 continue;
793 __touch_cap(cap);
794 }
795 }
796 return 1;
797 }
798 }
799
800 return 0;
801}
802
803/*
804 * Return true if mask caps are currently being revoked by an MDS.
805 */
Yan, Zheng6ee6b9532013-07-02 12:40:21 +0800806int __ceph_caps_revoking_other(struct ceph_inode_info *ci,
807 struct ceph_cap *ocap, int mask)
808{
809 struct ceph_cap *cap;
810 struct rb_node *p;
811
812 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
813 cap = rb_entry(p, struct ceph_cap, ci_node);
Yan, Zheng9563f882013-11-22 13:50:45 +0800814 if (cap != ocap &&
Yan, Zheng6ee6b9532013-07-02 12:40:21 +0800815 (cap->implemented & ~cap->issued & mask))
816 return 1;
817 }
818 return 0;
819}
820
Sage Weila8599bd2009-10-06 11:31:12 -0700821int ceph_caps_revoking(struct ceph_inode_info *ci, int mask)
822{
823 struct inode *inode = &ci->vfs_inode;
Yan, Zheng6ee6b9532013-07-02 12:40:21 +0800824 int ret;
Sage Weila8599bd2009-10-06 11:31:12 -0700825
Sage Weilbe655592011-11-30 09:47:09 -0800826 spin_lock(&ci->i_ceph_lock);
Yan, Zheng6ee6b9532013-07-02 12:40:21 +0800827 ret = __ceph_caps_revoking_other(ci, NULL, mask);
Sage Weilbe655592011-11-30 09:47:09 -0800828 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -0700829 dout("ceph_caps_revoking %p %s = %d\n", inode,
830 ceph_cap_string(mask), ret);
831 return ret;
832}
833
834int __ceph_caps_used(struct ceph_inode_info *ci)
835{
836 int used = 0;
837 if (ci->i_pin_ref)
838 used |= CEPH_CAP_PIN;
839 if (ci->i_rd_ref)
840 used |= CEPH_CAP_FILE_RD;
Yan, Zhengfdd4e152015-06-16 20:48:56 +0800841 if (ci->i_rdcache_ref ||
842 (!S_ISDIR(ci->vfs_inode.i_mode) && /* ignore readdir cache */
843 ci->vfs_inode.i_data.nrpages))
Sage Weila8599bd2009-10-06 11:31:12 -0700844 used |= CEPH_CAP_FILE_CACHE;
845 if (ci->i_wr_ref)
846 used |= CEPH_CAP_FILE_WR;
Henry C Changd3d07202011-05-11 10:29:54 +0000847 if (ci->i_wb_ref || ci->i_wrbuffer_ref)
Sage Weila8599bd2009-10-06 11:31:12 -0700848 used |= CEPH_CAP_FILE_BUFFER;
849 return used;
850}
851
852/*
853 * wanted, by virtue of open file modes
854 */
855int __ceph_caps_file_wanted(struct ceph_inode_info *ci)
856{
Yan, Zheng774a6a12016-06-06 16:01:39 +0800857 int i, bits = 0;
858 for (i = 0; i < CEPH_FILE_MODE_BITS; i++) {
859 if (ci->i_nr_by_mode[i])
860 bits |= 1 << i;
861 }
862 if (bits == 0)
863 return 0;
864 return ceph_caps_for_mode(bits >> 1);
Sage Weila8599bd2009-10-06 11:31:12 -0700865}
866
867/*
868 * Return caps we have registered with the MDS(s) as 'wanted'.
869 */
870int __ceph_caps_mds_wanted(struct ceph_inode_info *ci)
871{
872 struct ceph_cap *cap;
873 struct rb_node *p;
874 int mds_wanted = 0;
875
876 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
877 cap = rb_entry(p, struct ceph_cap, ci_node);
878 if (!__cap_is_valid(cap))
879 continue;
Yan, Zhenga2550602014-03-08 09:51:45 +0800880 if (cap == ci->i_auth_cap)
881 mds_wanted |= cap->mds_wanted;
882 else
883 mds_wanted |= (cap->mds_wanted & ~CEPH_CAP_ANY_FILE_WR);
Sage Weila8599bd2009-10-06 11:31:12 -0700884 }
885 return mds_wanted;
886}
887
888/*
Sage Weilbe655592011-11-30 09:47:09 -0800889 * called under i_ceph_lock
Sage Weila8599bd2009-10-06 11:31:12 -0700890 */
891static int __ceph_is_any_caps(struct ceph_inode_info *ci)
892{
Yan, Zhengd9df2782014-04-18 09:57:11 +0800893 return !RB_EMPTY_ROOT(&ci->i_caps);
Sage Weila8599bd2009-10-06 11:31:12 -0700894}
895
Yan, Zheng9215aee2013-11-30 12:47:41 +0800896int ceph_is_any_caps(struct inode *inode)
897{
898 struct ceph_inode_info *ci = ceph_inode(inode);
899 int ret;
900
901 spin_lock(&ci->i_ceph_lock);
902 ret = __ceph_is_any_caps(ci);
903 spin_unlock(&ci->i_ceph_lock);
904
905 return ret;
906}
907
Yan, Zhengdb40cc12015-03-23 20:12:20 +0800908static void drop_inode_snap_realm(struct ceph_inode_info *ci)
909{
910 struct ceph_snap_realm *realm = ci->i_snap_realm;
911 spin_lock(&realm->inodes_with_caps_lock);
912 list_del_init(&ci->i_snap_realm_item);
913 ci->i_snap_realm_counter++;
914 ci->i_snap_realm = NULL;
915 spin_unlock(&realm->inodes_with_caps_lock);
916 ceph_put_snap_realm(ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc,
917 realm);
918}
919
Sage Weila8599bd2009-10-06 11:31:12 -0700920/*
Sage Weilf818a732010-05-11 20:56:31 -0700921 * Remove a cap. Take steps to deal with a racing iterate_session_caps.
922 *
Sage Weilbe655592011-11-30 09:47:09 -0800923 * caller should hold i_ceph_lock.
Sage Weila6369742010-02-22 13:59:00 -0800924 * caller will not hold session s_mutex if called from destroy_inode.
Sage Weila8599bd2009-10-06 11:31:12 -0700925 */
Yan, Zhenga096b092013-09-22 10:15:58 +0800926void __ceph_remove_cap(struct ceph_cap *cap, bool queue_release)
Sage Weila8599bd2009-10-06 11:31:12 -0700927{
928 struct ceph_mds_session *session = cap->session;
929 struct ceph_inode_info *ci = cap->ci;
Cheng Renquan640ef792010-03-26 17:40:33 +0800930 struct ceph_mds_client *mdsc =
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700931 ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc;
Sage Weilf818a732010-05-11 20:56:31 -0700932 int removed = 0;
Sage Weila8599bd2009-10-06 11:31:12 -0700933
934 dout("__ceph_remove_cap %p from %p\n", cap, &ci->vfs_inode);
935
Sage Weil7c1332b2010-02-16 11:39:45 -0800936 /* remove from session list */
937 spin_lock(&session->s_cap_lock);
938 if (session->s_cap_iterator == cap) {
939 /* not yet, we are iterating over this very cap */
940 dout("__ceph_remove_cap delaying %p removal from session %p\n",
941 cap, cap->session);
942 } else {
943 list_del_init(&cap->session_caps);
944 session->s_nr_caps--;
945 cap->session = NULL;
Sage Weilf818a732010-05-11 20:56:31 -0700946 removed = 1;
Sage Weil7c1332b2010-02-16 11:39:45 -0800947 }
Sage Weilf818a732010-05-11 20:56:31 -0700948 /* protect backpointer with s_cap_lock: see iterate_session_caps */
949 cap->ci = NULL;
Yan, Zheng745a8e32015-05-14 17:22:42 +0800950
951 /*
952 * s_cap_reconnect is protected by s_cap_lock. no one changes
953 * s_cap_gen while session is in the reconnect state.
954 */
955 if (queue_release &&
956 (!session->s_cap_reconnect || cap->cap_gen == session->s_cap_gen)) {
957 cap->queue_release = 1;
958 if (removed) {
959 list_add_tail(&cap->session_caps,
960 &session->s_cap_releases);
961 session->s_num_cap_releases++;
962 removed = 0;
963 }
964 } else {
965 cap->queue_release = 0;
966 }
967 cap->cap_ino = ci->i_vino.ino;
968
Sage Weil7c1332b2010-02-16 11:39:45 -0800969 spin_unlock(&session->s_cap_lock);
970
Sage Weilf818a732010-05-11 20:56:31 -0700971 /* remove from inode list */
972 rb_erase(&cap->ci_node, &ci->i_caps);
973 if (ci->i_auth_cap == cap)
974 ci->i_auth_cap = NULL;
975
976 if (removed)
Yehuda Sadeh37151662010-06-17 16:16:12 -0700977 ceph_put_cap(mdsc, cap);
Sage Weila8599bd2009-10-06 11:31:12 -0700978
Yan, Zhengdb40cc12015-03-23 20:12:20 +0800979 /* when reconnect denied, we remove session caps forcibly,
980 * i_wr_ref can be non-zero. If there are ongoing write,
981 * keep i_snap_realm.
982 */
983 if (!__ceph_is_any_caps(ci) && ci->i_wr_ref == 0 && ci->i_snap_realm)
984 drop_inode_snap_realm(ci);
985
Sage Weila8599bd2009-10-06 11:31:12 -0700986 if (!__ceph_is_any_real_caps(ci))
987 __cap_delay_cancel(mdsc, ci);
988}
989
990/*
991 * Build and send a cap message to the given MDS.
992 *
993 * Caller should be holding s_mutex.
994 */
995static int send_cap_msg(struct ceph_mds_session *session,
996 u64 ino, u64 cid, int op,
997 int caps, int wanted, int dirty,
Yan, Zhenga2971c82015-06-10 11:09:32 +0800998 u32 seq, u64 flush_tid, u64 oldest_flush_tid,
999 u32 issue_seq, u32 mseq, u64 size, u64 max_size,
Sage Weila8599bd2009-10-06 11:31:12 -07001000 struct timespec *mtime, struct timespec *atime,
Yan, Zheng36094042016-07-06 15:37:42 +08001001 struct timespec *ctime, u32 time_warp_seq,
Eric W. Biederman05cb11c2013-01-31 02:56:19 -08001002 kuid_t uid, kgid_t gid, umode_t mode,
Sage Weila8599bd2009-10-06 11:31:12 -07001003 u64 xattr_version,
1004 struct ceph_buffer *xattrs_buf,
Yan, Zhenge20d2582014-11-14 22:39:13 +08001005 u64 follows, bool inline_data)
Sage Weila8599bd2009-10-06 11:31:12 -07001006{
1007 struct ceph_mds_caps *fc;
1008 struct ceph_msg *msg;
Yan, Zhenge20d2582014-11-14 22:39:13 +08001009 void *p;
1010 size_t extra_len;
Sage Weila8599bd2009-10-06 11:31:12 -07001011
1012 dout("send_cap_msg %s %llx %llx caps %s wanted %s dirty %s"
Yan, Zhenga2971c82015-06-10 11:09:32 +08001013 " seq %u/%u tid %llu/%llu mseq %u follows %lld size %llu/%llu"
Sage Weila8599bd2009-10-06 11:31:12 -07001014 " xattr_ver %llu xattr_len %d\n", ceph_cap_op_name(op),
1015 cid, ino, ceph_cap_string(caps), ceph_cap_string(wanted),
1016 ceph_cap_string(dirty),
Yan, Zhenga2971c82015-06-10 11:09:32 +08001017 seq, issue_seq, flush_tid, oldest_flush_tid,
1018 mseq, follows, size, max_size,
Sage Weila8599bd2009-10-06 11:31:12 -07001019 xattr_version, xattrs_buf ? (int)xattrs_buf->vec.iov_len : 0);
1020
Yan, Zhenga2971c82015-06-10 11:09:32 +08001021 /* flock buffer size + inline version + inline data size +
1022 * osd_epoch_barrier + oldest_flush_tid */
1023 extra_len = 4 + 8 + 4 + 4 + 8;
Yan, Zhenge20d2582014-11-14 22:39:13 +08001024 msg = ceph_msg_new(CEPH_MSG_CLIENT_CAPS, sizeof(*fc) + extra_len,
1025 GFP_NOFS, false);
Sage Weila79832f2010-04-01 16:06:19 -07001026 if (!msg)
1027 return -ENOMEM;
Sage Weila8599bd2009-10-06 11:31:12 -07001028
Yan, Zhenga2971c82015-06-10 11:09:32 +08001029 msg->hdr.version = cpu_to_le16(6);
Sage Weil6df058c2009-12-22 11:24:33 -08001030 msg->hdr.tid = cpu_to_le64(flush_tid);
Sage Weila8599bd2009-10-06 11:31:12 -07001031
Sage Weil6df058c2009-12-22 11:24:33 -08001032 fc = msg->front.iov_base;
Sage Weila8599bd2009-10-06 11:31:12 -07001033 memset(fc, 0, sizeof(*fc));
1034
1035 fc->cap_id = cpu_to_le64(cid);
1036 fc->op = cpu_to_le32(op);
1037 fc->seq = cpu_to_le32(seq);
Sage Weila8599bd2009-10-06 11:31:12 -07001038 fc->issue_seq = cpu_to_le32(issue_seq);
1039 fc->migrate_seq = cpu_to_le32(mseq);
1040 fc->caps = cpu_to_le32(caps);
1041 fc->wanted = cpu_to_le32(wanted);
1042 fc->dirty = cpu_to_le32(dirty);
1043 fc->ino = cpu_to_le64(ino);
1044 fc->snap_follows = cpu_to_le64(follows);
1045
1046 fc->size = cpu_to_le64(size);
1047 fc->max_size = cpu_to_le64(max_size);
1048 if (mtime)
1049 ceph_encode_timespec(&fc->mtime, mtime);
1050 if (atime)
1051 ceph_encode_timespec(&fc->atime, atime);
Yan, Zhengd1eee0c2016-01-22 18:30:12 +08001052 if (ctime)
1053 ceph_encode_timespec(&fc->ctime, ctime);
Sage Weila8599bd2009-10-06 11:31:12 -07001054 fc->time_warp_seq = cpu_to_le32(time_warp_seq);
1055
Eric W. Biederman05cb11c2013-01-31 02:56:19 -08001056 fc->uid = cpu_to_le32(from_kuid(&init_user_ns, uid));
1057 fc->gid = cpu_to_le32(from_kgid(&init_user_ns, gid));
Sage Weila8599bd2009-10-06 11:31:12 -07001058 fc->mode = cpu_to_le32(mode);
1059
Yan, Zhenge20d2582014-11-14 22:39:13 +08001060 p = fc + 1;
1061 /* flock buffer size */
1062 ceph_encode_32(&p, 0);
1063 /* inline version */
1064 ceph_encode_64(&p, inline_data ? 0 : CEPH_INLINE_NONE);
1065 /* inline data size */
1066 ceph_encode_32(&p, 0);
Yan, Zhenga2971c82015-06-10 11:09:32 +08001067 /* osd_epoch_barrier */
1068 ceph_encode_32(&p, 0);
1069 /* oldest_flush_tid */
1070 ceph_encode_64(&p, oldest_flush_tid);
Yan, Zhenge20d2582014-11-14 22:39:13 +08001071
Sage Weila8599bd2009-10-06 11:31:12 -07001072 fc->xattr_version = cpu_to_le64(xattr_version);
1073 if (xattrs_buf) {
1074 msg->middle = ceph_buffer_get(xattrs_buf);
1075 fc->xattr_len = cpu_to_le32(xattrs_buf->vec.iov_len);
1076 msg->hdr.middle_len = cpu_to_le32(xattrs_buf->vec.iov_len);
1077 }
1078
1079 ceph_con_send(&session->s_con, msg);
1080 return 0;
1081}
1082
1083/*
Sage Weila6369742010-02-22 13:59:00 -08001084 * Queue cap releases when an inode is dropped from our cache. Since
Sage Weilbe655592011-11-30 09:47:09 -08001085 * inode is about to be destroyed, there is no need for i_ceph_lock.
Sage Weila8599bd2009-10-06 11:31:12 -07001086 */
1087void ceph_queue_caps_release(struct inode *inode)
1088{
1089 struct ceph_inode_info *ci = ceph_inode(inode);
1090 struct rb_node *p;
1091
Sage Weila8599bd2009-10-06 11:31:12 -07001092 p = rb_first(&ci->i_caps);
1093 while (p) {
1094 struct ceph_cap *cap = rb_entry(p, struct ceph_cap, ci_node);
Sage Weila8599bd2009-10-06 11:31:12 -07001095 p = rb_next(p);
Yan, Zhenga096b092013-09-22 10:15:58 +08001096 __ceph_remove_cap(cap, true);
Sage Weila8599bd2009-10-06 11:31:12 -07001097 }
Sage Weila8599bd2009-10-06 11:31:12 -07001098}
1099
1100/*
1101 * Send a cap msg on the given inode. Update our caps state, then
Sage Weilbe655592011-11-30 09:47:09 -08001102 * drop i_ceph_lock and send the message.
Sage Weila8599bd2009-10-06 11:31:12 -07001103 *
1104 * Make note of max_size reported/requested from mds, revoked caps
1105 * that have now been implemented.
1106 *
1107 * Make half-hearted attempt ot to invalidate page cache if we are
1108 * dropping RDCACHE. Note that this will leave behind locked pages
1109 * that we'll then need to deal with elsewhere.
1110 *
1111 * Return non-zero if delayed release, or we experienced an error
1112 * such that the caller should requeue + retry later.
1113 *
Sage Weilbe655592011-11-30 09:47:09 -08001114 * called with i_ceph_lock, then drops it.
Sage Weila8599bd2009-10-06 11:31:12 -07001115 * caller should hold snap_rwsem (read), s_mutex.
1116 */
1117static int __send_cap(struct ceph_mds_client *mdsc, struct ceph_cap *cap,
Yan, Zheng553adfd2015-06-09 15:48:57 +08001118 int op, int used, int want, int retain, int flushing,
Yan, Zhenga2971c82015-06-10 11:09:32 +08001119 u64 flush_tid, u64 oldest_flush_tid)
Sage Weilbe655592011-11-30 09:47:09 -08001120 __releases(cap->ci->i_ceph_lock)
Sage Weila8599bd2009-10-06 11:31:12 -07001121{
1122 struct ceph_inode_info *ci = cap->ci;
1123 struct inode *inode = &ci->vfs_inode;
1124 u64 cap_id = cap->cap_id;
Sage Weil68c28322010-02-09 13:41:47 -08001125 int held, revoking, dropping, keep;
Yan, Zheng36094042016-07-06 15:37:42 +08001126 u64 follows, size, max_size;
1127 u32 seq, issue_seq, mseq, time_warp_seq;
Yan, Zhengd1eee0c2016-01-22 18:30:12 +08001128 struct timespec mtime, atime, ctime;
Sage Weila8599bd2009-10-06 11:31:12 -07001129 int wake = 0;
Al Viro5706b272011-07-26 04:52:22 -04001130 umode_t mode;
Eric W. Biederman05cb11c2013-01-31 02:56:19 -08001131 kuid_t uid;
1132 kgid_t gid;
Sage Weila8599bd2009-10-06 11:31:12 -07001133 struct ceph_mds_session *session;
1134 u64 xattr_version = 0;
Sage Weil082afec2010-08-22 15:16:41 -07001135 struct ceph_buffer *xattr_blob = NULL;
Sage Weila8599bd2009-10-06 11:31:12 -07001136 int delayed = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07001137 int ret;
Yan, Zhenge20d2582014-11-14 22:39:13 +08001138 bool inline_data;
Sage Weila8599bd2009-10-06 11:31:12 -07001139
Sage Weil68c28322010-02-09 13:41:47 -08001140 held = cap->issued | cap->implemented;
1141 revoking = cap->implemented & ~cap->issued;
1142 retain &= ~revoking;
1143 dropping = cap->issued & ~retain;
1144
Sage Weila8599bd2009-10-06 11:31:12 -07001145 dout("__send_cap %p cap %p session %p %s -> %s (revoking %s)\n",
1146 inode, cap, cap->session,
1147 ceph_cap_string(held), ceph_cap_string(held & retain),
1148 ceph_cap_string(revoking));
1149 BUG_ON((retain & CEPH_CAP_PIN) == 0);
1150
1151 session = cap->session;
1152
1153 /* don't release wanted unless we've waited a bit. */
1154 if ((ci->i_ceph_flags & CEPH_I_NODELAY) == 0 &&
1155 time_before(jiffies, ci->i_hold_caps_min)) {
1156 dout(" delaying issued %s -> %s, wanted %s -> %s on send\n",
1157 ceph_cap_string(cap->issued),
1158 ceph_cap_string(cap->issued & retain),
1159 ceph_cap_string(cap->mds_wanted),
1160 ceph_cap_string(want));
1161 want |= cap->mds_wanted;
1162 retain |= cap->issued;
1163 delayed = 1;
1164 }
1165 ci->i_ceph_flags &= ~(CEPH_I_NODELAY | CEPH_I_FLUSH);
1166
1167 cap->issued &= retain; /* drop bits we don't want */
1168 if (cap->implemented & ~cap->issued) {
1169 /*
1170 * Wake up any waiters on wanted -> needed transition.
1171 * This is due to the weird transition from buffered
1172 * to sync IO... we need to flush dirty pages _before_
1173 * allowing sync writes to avoid reordering.
1174 */
1175 wake = 1;
1176 }
1177 cap->implemented &= cap->issued | used;
1178 cap->mds_wanted = want;
1179
Yan, Zheng553adfd2015-06-09 15:48:57 +08001180 follows = flushing ? ci->i_head_snapc->seq : 0;
Sage Weila8599bd2009-10-06 11:31:12 -07001181
1182 keep = cap->implemented;
1183 seq = cap->seq;
1184 issue_seq = cap->issue_seq;
1185 mseq = cap->mseq;
1186 size = inode->i_size;
1187 ci->i_reported_size = size;
1188 max_size = ci->i_wanted_max_size;
1189 ci->i_requested_max_size = max_size;
1190 mtime = inode->i_mtime;
1191 atime = inode->i_atime;
Yan, Zhengd1eee0c2016-01-22 18:30:12 +08001192 ctime = inode->i_ctime;
Sage Weila8599bd2009-10-06 11:31:12 -07001193 time_warp_seq = ci->i_time_warp_seq;
Sage Weila8599bd2009-10-06 11:31:12 -07001194 uid = inode->i_uid;
1195 gid = inode->i_gid;
1196 mode = inode->i_mode;
1197
Sage Weil082afec2010-08-22 15:16:41 -07001198 if (flushing & CEPH_CAP_XATTR_EXCL) {
Sage Weila8599bd2009-10-06 11:31:12 -07001199 __ceph_build_xattrs_blob(ci);
Sage Weil082afec2010-08-22 15:16:41 -07001200 xattr_blob = ci->i_xattrs.blob;
1201 xattr_version = ci->i_xattrs.version;
Sage Weila8599bd2009-10-06 11:31:12 -07001202 }
1203
Yan, Zhenge20d2582014-11-14 22:39:13 +08001204 inline_data = ci->i_inline_version != CEPH_INLINE_NONE;
1205
Sage Weilbe655592011-11-30 09:47:09 -08001206 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001207
Sage Weila8599bd2009-10-06 11:31:12 -07001208 ret = send_cap_msg(session, ceph_vino(inode).ino, cap_id,
Yan, Zhenga2971c82015-06-10 11:09:32 +08001209 op, keep, want, flushing, seq,
1210 flush_tid, oldest_flush_tid, issue_seq, mseq,
Yan, Zhengd1eee0c2016-01-22 18:30:12 +08001211 size, max_size, &mtime, &atime, &ctime, time_warp_seq,
Sage Weil082afec2010-08-22 15:16:41 -07001212 uid, gid, mode, xattr_version, xattr_blob,
Yan, Zhenge20d2582014-11-14 22:39:13 +08001213 follows, inline_data);
Sage Weila8599bd2009-10-06 11:31:12 -07001214 if (ret < 0) {
1215 dout("error sending cap msg, must requeue %p\n", inode);
1216 delayed = 1;
1217 }
1218
1219 if (wake)
Yehuda Sadeh03066f22010-07-27 13:11:08 -07001220 wake_up_all(&ci->i_cap_wq);
Sage Weila8599bd2009-10-06 11:31:12 -07001221
1222 return delayed;
1223}
1224
Yan, Zheng0e294382016-07-04 18:06:41 +08001225static inline int __send_flush_snap(struct inode *inode,
1226 struct ceph_mds_session *session,
1227 struct ceph_cap_snap *capsnap,
1228 u32 mseq, u64 oldest_flush_tid)
1229{
1230 return send_cap_msg(session, ceph_vino(inode).ino, 0,
1231 CEPH_CAP_OP_FLUSHSNAP, capsnap->issued, 0,
1232 capsnap->dirty, 0, capsnap->cap_flush.tid,
1233 oldest_flush_tid, 0, mseq, capsnap->size, 0,
1234 &capsnap->mtime, &capsnap->atime,
1235 &capsnap->ctime, capsnap->time_warp_seq,
1236 capsnap->uid, capsnap->gid, capsnap->mode,
1237 capsnap->xattr_version, capsnap->xattr_blob,
1238 capsnap->follows, capsnap->inline_data);
1239}
1240
Sage Weila8599bd2009-10-06 11:31:12 -07001241/*
1242 * When a snapshot is taken, clients accumulate dirty metadata on
1243 * inodes with capabilities in ceph_cap_snaps to describe the file
1244 * state at the time the snapshot was taken. This must be flushed
1245 * asynchronously back to the MDS once sync writes complete and dirty
1246 * data is written out.
1247 *
Sage Weilbe655592011-11-30 09:47:09 -08001248 * Called under i_ceph_lock. Takes s_mutex as needed.
Sage Weila8599bd2009-10-06 11:31:12 -07001249 */
Yan, Zhenged9b4302016-07-05 21:08:07 +08001250static void __ceph_flush_snaps(struct ceph_inode_info *ci,
1251 struct ceph_mds_session *session)
Sage Weilbe655592011-11-30 09:47:09 -08001252 __releases(ci->i_ceph_lock)
1253 __acquires(ci->i_ceph_lock)
Sage Weila8599bd2009-10-06 11:31:12 -07001254{
1255 struct inode *inode = &ci->vfs_inode;
Yan, Zhenged9b4302016-07-05 21:08:07 +08001256 struct ceph_mds_client *mdsc = session->s_mdsc;
Sage Weila8599bd2009-10-06 11:31:12 -07001257 struct ceph_cap_snap *capsnap;
Yan, Zhenged9b4302016-07-05 21:08:07 +08001258 u64 oldest_flush_tid = 0;
1259 u64 first_tid = 1, last_tid = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07001260
Yan, Zhenged9b4302016-07-05 21:08:07 +08001261 dout("__flush_snaps %p session %p\n", inode, session);
Sage Weila8599bd2009-10-06 11:31:12 -07001262
Sage Weila8599bd2009-10-06 11:31:12 -07001263 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
Sage Weila8599bd2009-10-06 11:31:12 -07001264 /*
1265 * we need to wait for sync writes to complete and for dirty
1266 * pages to be written out.
1267 */
1268 if (capsnap->dirty_pages || capsnap->writing)
Sage Weilcfc0bf62010-09-14 15:50:59 -07001269 break;
Sage Weila8599bd2009-10-06 11:31:12 -07001270
Yan, Zheng86056092015-05-01 16:57:16 +08001271 /* should be removed by ceph_try_drop_cap_snap() */
1272 BUG_ON(!capsnap->need_flush);
Sage Weil819ccbf2010-04-01 09:33:46 -07001273
Sage Weile8351242010-09-17 08:03:08 -07001274 /* only flush each capsnap once */
Yan, Zheng0e294382016-07-04 18:06:41 +08001275 if (capsnap->cap_flush.tid > 0) {
Yan, Zhenged9b4302016-07-05 21:08:07 +08001276 dout(" already flushed %p, skipping\n", capsnap);
Sage Weile8351242010-09-17 08:03:08 -07001277 continue;
1278 }
1279
Yan, Zheng553adfd2015-06-09 15:48:57 +08001280 spin_lock(&mdsc->cap_dirty_lock);
Yan, Zheng0e294382016-07-04 18:06:41 +08001281 capsnap->cap_flush.tid = ++mdsc->last_cap_flush_tid;
1282 list_add_tail(&capsnap->cap_flush.g_list,
1283 &mdsc->cap_flush_list);
Yan, Zhenged9b4302016-07-05 21:08:07 +08001284 if (oldest_flush_tid == 0)
1285 oldest_flush_tid = __get_oldest_flush_tid(mdsc);
Yan, Zheng0e294382016-07-04 18:06:41 +08001286 if (list_empty(&ci->i_flushing_item)) {
1287 list_add_tail(&ci->i_flushing_item,
1288 &session->s_cap_flushing);
1289 }
Yan, Zheng553adfd2015-06-09 15:48:57 +08001290 spin_unlock(&mdsc->cap_dirty_lock);
1291
Yan, Zheng0e294382016-07-04 18:06:41 +08001292 list_add_tail(&capsnap->cap_flush.i_list,
1293 &ci->i_cap_flush_list);
1294
Yan, Zhenged9b4302016-07-05 21:08:07 +08001295 if (first_tid == 1)
1296 first_tid = capsnap->cap_flush.tid;
1297 last_tid = capsnap->cap_flush.tid;
1298 }
1299
1300 ci->i_ceph_flags &= ~CEPH_I_FLUSH_SNAPS;
1301
1302 while (first_tid <= last_tid) {
1303 struct ceph_cap *cap = ci->i_auth_cap;
1304 struct ceph_cap_flush *cf;
1305 int ret;
1306
1307 if (!(cap && cap->session == session)) {
1308 dout("__flush_snaps %p auth cap %p not mds%d, "
1309 "stop\n", inode, cap, session->s_mds);
1310 break;
1311 }
1312
1313 ret = -ENOENT;
1314 list_for_each_entry(cf, &ci->i_cap_flush_list, i_list) {
1315 if (cf->tid >= first_tid) {
1316 ret = 0;
1317 break;
1318 }
1319 }
1320 if (ret < 0)
1321 break;
1322
1323 first_tid = cf->tid + 1;
1324
1325 capsnap = container_of(cf, struct ceph_cap_snap, cap_flush);
Sage Weila8599bd2009-10-06 11:31:12 -07001326 atomic_inc(&capsnap->nref);
Sage Weilbe655592011-11-30 09:47:09 -08001327 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001328
Yan, Zhenged9b4302016-07-05 21:08:07 +08001329 dout("__flush_snaps %p capsnap %p tid %llu %s\n",
1330 inode, capsnap, cf->tid, ceph_cap_string(capsnap->dirty));
Sage Weila8599bd2009-10-06 11:31:12 -07001331
Yan, Zhenged9b4302016-07-05 21:08:07 +08001332 ret = __send_flush_snap(inode, session, capsnap, cap->mseq,
1333 oldest_flush_tid);
1334 if (ret < 0) {
1335 pr_err("__flush_snaps: error sending cap flushsnap, "
1336 "ino (%llx.%llx) tid %llu follows %llu\n",
1337 ceph_vinop(inode), cf->tid, capsnap->follows);
1338 }
1339
Sage Weila8599bd2009-10-06 11:31:12 -07001340 ceph_put_cap_snap(capsnap);
Sage Weilbe655592011-11-30 09:47:09 -08001341 spin_lock(&ci->i_ceph_lock);
Yan, Zhenged9b4302016-07-05 21:08:07 +08001342 }
1343}
1344
1345void ceph_flush_snaps(struct ceph_inode_info *ci,
1346 struct ceph_mds_session **psession)
1347{
1348 struct inode *inode = &ci->vfs_inode;
1349 struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
Yan, Zhenge4d2b162016-08-04 08:43:33 +08001350 struct ceph_mds_session *session = NULL;
Yan, Zhenged9b4302016-07-05 21:08:07 +08001351 int mds;
Yan, Zhenge4d2b162016-08-04 08:43:33 +08001352
Yan, Zhenged9b4302016-07-05 21:08:07 +08001353 dout("ceph_flush_snaps %p\n", inode);
Yan, Zhenge4d2b162016-08-04 08:43:33 +08001354 if (psession)
1355 session = *psession;
Yan, Zhenged9b4302016-07-05 21:08:07 +08001356retry:
1357 spin_lock(&ci->i_ceph_lock);
1358 if (!(ci->i_ceph_flags & CEPH_I_FLUSH_SNAPS)) {
1359 dout(" no capsnap needs flush, doing nothing\n");
1360 goto out;
1361 }
1362 if (!ci->i_auth_cap) {
1363 dout(" no auth cap (migrating?), doing nothing\n");
1364 goto out;
1365 }
1366
1367 mds = ci->i_auth_cap->session->s_mds;
1368 if (session && session->s_mds != mds) {
1369 dout(" oops, wrong session %p mutex\n", session);
1370 mutex_unlock(&session->s_mutex);
1371 ceph_put_mds_session(session);
1372 session = NULL;
1373 }
1374 if (!session) {
1375 spin_unlock(&ci->i_ceph_lock);
1376 mutex_lock(&mdsc->mutex);
1377 session = __ceph_lookup_mds_session(mdsc, mds);
1378 mutex_unlock(&mdsc->mutex);
1379 if (session) {
1380 dout(" inverting session/ino locks on %p\n", session);
1381 mutex_lock(&session->s_mutex);
1382 }
Sage Weila8599bd2009-10-06 11:31:12 -07001383 goto retry;
1384 }
1385
Yan, Zhenged9b4302016-07-05 21:08:07 +08001386 __ceph_flush_snaps(ci, session);
1387out:
1388 spin_unlock(&ci->i_ceph_lock);
1389
1390 if (psession) {
1391 *psession = session;
1392 } else {
1393 mutex_unlock(&session->s_mutex);
1394 ceph_put_mds_session(session);
1395 }
Sage Weila8599bd2009-10-06 11:31:12 -07001396 /* we flushed them all; remove this inode from the queue */
1397 spin_lock(&mdsc->snap_flush_lock);
1398 list_del_init(&ci->i_snap_flush_item);
1399 spin_unlock(&mdsc->snap_flush_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001400}
1401
1402/*
Sage Weilfca65b42011-05-04 11:33:47 -07001403 * Mark caps dirty. If inode is newly dirty, return the dirty flags.
1404 * Caller is then responsible for calling __mark_inode_dirty with the
1405 * returned flags value.
Sage Weil76e3b392009-10-15 18:13:53 -07001406 */
Yan, Zhengf66fd9f2015-06-10 17:26:13 +08001407int __ceph_mark_dirty_caps(struct ceph_inode_info *ci, int mask,
1408 struct ceph_cap_flush **pcf)
Sage Weil76e3b392009-10-15 18:13:53 -07001409{
Cheng Renquan640ef792010-03-26 17:40:33 +08001410 struct ceph_mds_client *mdsc =
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001411 ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc;
Sage Weil76e3b392009-10-15 18:13:53 -07001412 struct inode *inode = &ci->vfs_inode;
1413 int was = ci->i_dirty_caps;
1414 int dirty = 0;
1415
Yan, Zheng571ade32015-03-24 11:36:08 +08001416 if (!ci->i_auth_cap) {
1417 pr_warn("__mark_dirty_caps %p %llx mask %s, "
1418 "but no auth cap (session was closed?)\n",
1419 inode, ceph_ino(inode), ceph_cap_string(mask));
1420 return 0;
1421 }
1422
Sage Weil76e3b392009-10-15 18:13:53 -07001423 dout("__mark_dirty_caps %p %s dirty %s -> %s\n", &ci->vfs_inode,
1424 ceph_cap_string(mask), ceph_cap_string(was),
1425 ceph_cap_string(was | mask));
1426 ci->i_dirty_caps |= mask;
1427 if (was == 0) {
Yan, Zhengf66fd9f2015-06-10 17:26:13 +08001428 WARN_ON_ONCE(ci->i_prealloc_cap_flush);
1429 swap(ci->i_prealloc_cap_flush, *pcf);
1430
Yan, Zheng604d1b02015-05-01 17:49:16 +08001431 if (!ci->i_head_snapc) {
1432 WARN_ON_ONCE(!rwsem_is_locked(&mdsc->snap_rwsem));
Sage Weil7d8cb262010-08-24 08:44:16 -07001433 ci->i_head_snapc = ceph_get_snap_context(
1434 ci->i_snap_realm->cached_context);
Yan, Zheng604d1b02015-05-01 17:49:16 +08001435 }
Yan, Zheng06852352012-11-19 10:49:07 +08001436 dout(" inode %p now dirty snapc %p auth cap %p\n",
1437 &ci->vfs_inode, ci->i_head_snapc, ci->i_auth_cap);
Sage Weil76e3b392009-10-15 18:13:53 -07001438 BUG_ON(!list_empty(&ci->i_dirty_item));
1439 spin_lock(&mdsc->cap_dirty_lock);
Yan, Zheng11df2df2013-11-24 14:44:38 +08001440 list_add(&ci->i_dirty_item, &mdsc->cap_dirty);
Sage Weil76e3b392009-10-15 18:13:53 -07001441 spin_unlock(&mdsc->cap_dirty_lock);
1442 if (ci->i_flushing_caps == 0) {
Sage Weil3772d262011-05-03 09:28:08 -07001443 ihold(inode);
Sage Weil76e3b392009-10-15 18:13:53 -07001444 dirty |= I_DIRTY_SYNC;
1445 }
Yan, Zhengf66fd9f2015-06-10 17:26:13 +08001446 } else {
1447 WARN_ON_ONCE(!ci->i_prealloc_cap_flush);
Sage Weil76e3b392009-10-15 18:13:53 -07001448 }
1449 BUG_ON(list_empty(&ci->i_dirty_item));
1450 if (((was | ci->i_flushing_caps) & CEPH_CAP_FILE_BUFFER) &&
1451 (mask & CEPH_CAP_FILE_BUFFER))
1452 dirty |= I_DIRTY_DATASYNC;
Sage Weil76e3b392009-10-15 18:13:53 -07001453 __cap_delay_requeue(mdsc, ci);
Sage Weilfca65b42011-05-04 11:33:47 -07001454 return dirty;
Sage Weil76e3b392009-10-15 18:13:53 -07001455}
1456
Yan, Zhengf66fd9f2015-06-10 17:26:13 +08001457struct ceph_cap_flush *ceph_alloc_cap_flush(void)
1458{
1459 return kmem_cache_alloc(ceph_cap_flush_cachep, GFP_KERNEL);
1460}
1461
1462void ceph_free_cap_flush(struct ceph_cap_flush *cf)
1463{
1464 if (cf)
1465 kmem_cache_free(ceph_cap_flush_cachep, cf);
1466}
1467
Yan, Zhenga2971c82015-06-10 11:09:32 +08001468static u64 __get_oldest_flush_tid(struct ceph_mds_client *mdsc)
1469{
Yan, Zhenge4500b52016-07-06 11:12:56 +08001470 if (!list_empty(&mdsc->cap_flush_list)) {
Yan, Zhenga2971c82015-06-10 11:09:32 +08001471 struct ceph_cap_flush *cf =
Yan, Zhenge4500b52016-07-06 11:12:56 +08001472 list_first_entry(&mdsc->cap_flush_list,
1473 struct ceph_cap_flush, g_list);
Yan, Zhenga2971c82015-06-10 11:09:32 +08001474 return cf->tid;
1475 }
1476 return 0;
1477}
1478
Sage Weil76e3b392009-10-15 18:13:53 -07001479/*
Yan, Zhengc8799fc2016-07-07 15:22:38 +08001480 * Remove cap_flush from the mdsc's or inode's flushing cap list.
1481 * Return true if caller needs to wake up flush waiters.
1482 */
1483static bool __finish_cap_flush(struct ceph_mds_client *mdsc,
1484 struct ceph_inode_info *ci,
1485 struct ceph_cap_flush *cf)
1486{
1487 struct ceph_cap_flush *prev;
1488 bool wake = cf->wake;
1489 if (mdsc) {
1490 /* are there older pending cap flushes? */
1491 if (wake && cf->g_list.prev != &mdsc->cap_flush_list) {
1492 prev = list_prev_entry(cf, g_list);
1493 prev->wake = true;
1494 wake = false;
1495 }
1496 list_del(&cf->g_list);
1497 } else if (ci) {
1498 if (wake && cf->i_list.prev != &ci->i_cap_flush_list) {
1499 prev = list_prev_entry(cf, i_list);
1500 prev->wake = true;
1501 wake = false;
1502 }
1503 list_del(&cf->i_list);
1504 } else {
1505 BUG_ON(1);
1506 }
1507 return wake;
1508}
1509
1510/*
Sage Weila8599bd2009-10-06 11:31:12 -07001511 * Add dirty inode to the flushing list. Assigned a seq number so we
1512 * can wait for caps to flush without starving.
Sage Weilcdc35f92009-10-14 14:24:19 -07001513 *
Sage Weilbe655592011-11-30 09:47:09 -08001514 * Called under i_ceph_lock.
Sage Weila8599bd2009-10-06 11:31:12 -07001515 */
Sage Weilcdc35f92009-10-14 14:24:19 -07001516static int __mark_caps_flushing(struct inode *inode,
Yan, Zhengc8799fc2016-07-07 15:22:38 +08001517 struct ceph_mds_session *session, bool wake,
Yan, Zhenga2971c82015-06-10 11:09:32 +08001518 u64 *flush_tid, u64 *oldest_flush_tid)
Sage Weila8599bd2009-10-06 11:31:12 -07001519{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001520 struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
Sage Weila8599bd2009-10-06 11:31:12 -07001521 struct ceph_inode_info *ci = ceph_inode(inode);
Yan, Zhengf66fd9f2015-06-10 17:26:13 +08001522 struct ceph_cap_flush *cf = NULL;
Sage Weilcdc35f92009-10-14 14:24:19 -07001523 int flushing;
Sage Weil50b885b2009-12-01 14:12:07 -08001524
Sage Weilcdc35f92009-10-14 14:24:19 -07001525 BUG_ON(ci->i_dirty_caps == 0);
Sage Weila8599bd2009-10-06 11:31:12 -07001526 BUG_ON(list_empty(&ci->i_dirty_item));
Yan, Zhengf66fd9f2015-06-10 17:26:13 +08001527 BUG_ON(!ci->i_prealloc_cap_flush);
Sage Weilcdc35f92009-10-14 14:24:19 -07001528
1529 flushing = ci->i_dirty_caps;
1530 dout("__mark_caps_flushing flushing %s, flushing_caps %s -> %s\n",
1531 ceph_cap_string(flushing),
1532 ceph_cap_string(ci->i_flushing_caps),
1533 ceph_cap_string(ci->i_flushing_caps | flushing));
1534 ci->i_flushing_caps |= flushing;
1535 ci->i_dirty_caps = 0;
Sage Weilafcdaea2009-10-14 14:27:38 -07001536 dout(" inode %p now !dirty\n", inode);
Sage Weilcdc35f92009-10-14 14:24:19 -07001537
Yan, Zhengf66fd9f2015-06-10 17:26:13 +08001538 swap(cf, ci->i_prealloc_cap_flush);
Yan, Zheng553adfd2015-06-09 15:48:57 +08001539 cf->caps = flushing;
Yan, Zhengc8799fc2016-07-07 15:22:38 +08001540 cf->wake = wake;
Yan, Zheng553adfd2015-06-09 15:48:57 +08001541
Sage Weila8599bd2009-10-06 11:31:12 -07001542 spin_lock(&mdsc->cap_dirty_lock);
Sage Weilafcdaea2009-10-14 14:27:38 -07001543 list_del_init(&ci->i_dirty_item);
1544
Yan, Zheng553adfd2015-06-09 15:48:57 +08001545 cf->tid = ++mdsc->last_cap_flush_tid;
Yan, Zhenge4500b52016-07-06 11:12:56 +08001546 list_add_tail(&cf->g_list, &mdsc->cap_flush_list);
Yan, Zhenga2971c82015-06-10 11:09:32 +08001547 *oldest_flush_tid = __get_oldest_flush_tid(mdsc);
Yan, Zheng553adfd2015-06-09 15:48:57 +08001548
Sage Weila8599bd2009-10-06 11:31:12 -07001549 if (list_empty(&ci->i_flushing_item)) {
1550 list_add_tail(&ci->i_flushing_item, &session->s_cap_flushing);
1551 mdsc->num_cap_flushing++;
Sage Weila8599bd2009-10-06 11:31:12 -07001552 }
1553 spin_unlock(&mdsc->cap_dirty_lock);
Sage Weilcdc35f92009-10-14 14:24:19 -07001554
Yan, Zhenge4500b52016-07-06 11:12:56 +08001555 list_add_tail(&cf->i_list, &ci->i_cap_flush_list);
Yan, Zheng553adfd2015-06-09 15:48:57 +08001556
1557 *flush_tid = cf->tid;
Sage Weilcdc35f92009-10-14 14:24:19 -07001558 return flushing;
Sage Weila8599bd2009-10-06 11:31:12 -07001559}
1560
1561/*
Sage Weil5ecad6f2010-02-17 10:43:37 -08001562 * try to invalidate mapping pages without blocking.
1563 */
Sage Weil5ecad6f2010-02-17 10:43:37 -08001564static int try_nonblocking_invalidate(struct inode *inode)
1565{
1566 struct ceph_inode_info *ci = ceph_inode(inode);
1567 u32 invalidating_gen = ci->i_rdcache_gen;
1568
Sage Weilbe655592011-11-30 09:47:09 -08001569 spin_unlock(&ci->i_ceph_lock);
Sage Weil5ecad6f2010-02-17 10:43:37 -08001570 invalidate_mapping_pages(&inode->i_data, 0, -1);
Sage Weilbe655592011-11-30 09:47:09 -08001571 spin_lock(&ci->i_ceph_lock);
Sage Weil5ecad6f2010-02-17 10:43:37 -08001572
Sage Weil18a38192010-09-17 10:46:44 -07001573 if (inode->i_data.nrpages == 0 &&
Sage Weil5ecad6f2010-02-17 10:43:37 -08001574 invalidating_gen == ci->i_rdcache_gen) {
1575 /* success. */
1576 dout("try_nonblocking_invalidate %p success\n", inode);
Sage Weilcd045cb2010-11-04 11:05:05 -07001577 /* save any racing async invalidate some trouble */
1578 ci->i_rdcache_revoking = ci->i_rdcache_gen - 1;
Sage Weil5ecad6f2010-02-17 10:43:37 -08001579 return 0;
1580 }
1581 dout("try_nonblocking_invalidate %p failed\n", inode);
1582 return -1;
1583}
1584
1585/*
Sage Weila8599bd2009-10-06 11:31:12 -07001586 * Swiss army knife function to examine currently used and wanted
1587 * versus held caps. Release, flush, ack revoked caps to mds as
1588 * appropriate.
1589 *
1590 * CHECK_CAPS_NODELAY - caller is delayed work and we should not delay
1591 * cap release further.
1592 * CHECK_CAPS_AUTHONLY - we should only check the auth cap
1593 * CHECK_CAPS_FLUSH - we should flush any dirty caps immediately, without
1594 * further delay.
1595 */
1596void ceph_check_caps(struct ceph_inode_info *ci, int flags,
1597 struct ceph_mds_session *session)
1598{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001599 struct ceph_fs_client *fsc = ceph_inode_to_client(&ci->vfs_inode);
1600 struct ceph_mds_client *mdsc = fsc->mdsc;
Sage Weila8599bd2009-10-06 11:31:12 -07001601 struct inode *inode = &ci->vfs_inode;
1602 struct ceph_cap *cap;
Yan, Zhenga2971c82015-06-10 11:09:32 +08001603 u64 flush_tid, oldest_flush_tid;
Yan, Zheng395c3122013-01-04 14:37:57 +08001604 int file_wanted, used, cap_used;
Sage Weila8599bd2009-10-06 11:31:12 -07001605 int took_snap_rwsem = 0; /* true if mdsc->snap_rwsem held */
Sage Weilcbd03632010-02-09 13:41:18 -08001606 int issued, implemented, want, retain, revoking, flushing = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07001607 int mds = -1; /* keep track of how far we've gone through i_caps list
1608 to avoid an infinite loop on retry */
1609 struct rb_node *p;
Yan, Zheng36094042016-07-06 15:37:42 +08001610 int delayed = 0, sent = 0, num;
1611 bool is_delayed = flags & CHECK_CAPS_NODELAY;
1612 bool queue_invalidate = false;
1613 bool force_requeue = false;
1614 bool tried_invalidate = false;
Sage Weila8599bd2009-10-06 11:31:12 -07001615
1616 /* if we are unmounting, flush any unused caps immediately. */
1617 if (mdsc->stopping)
1618 is_delayed = 1;
1619
Sage Weilbe655592011-11-30 09:47:09 -08001620 spin_lock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001621
1622 if (ci->i_ceph_flags & CEPH_I_FLUSH)
1623 flags |= CHECK_CAPS_FLUSH;
1624
Sage Weila8599bd2009-10-06 11:31:12 -07001625 goto retry_locked;
1626retry:
Sage Weilbe655592011-11-30 09:47:09 -08001627 spin_lock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001628retry_locked:
1629 file_wanted = __ceph_caps_file_wanted(ci);
1630 used = __ceph_caps_used(ci);
Sage Weilcbd03632010-02-09 13:41:18 -08001631 issued = __ceph_caps_issued(ci, &implemented);
1632 revoking = implemented & ~issued;
Sage Weila8599bd2009-10-06 11:31:12 -07001633
Yan, Zheng41445992015-05-25 17:36:42 +08001634 want = file_wanted;
1635 retain = file_wanted | used | CEPH_CAP_PIN;
Sage Weila8599bd2009-10-06 11:31:12 -07001636 if (!mdsc->stopping && inode->i_nlink > 0) {
Yan, Zheng41445992015-05-25 17:36:42 +08001637 if (file_wanted) {
Sage Weila8599bd2009-10-06 11:31:12 -07001638 retain |= CEPH_CAP_ANY; /* be greedy */
Yan, Zheng32ec4392015-03-26 19:06:00 +08001639 } else if (S_ISDIR(inode->i_mode) &&
1640 (issued & CEPH_CAP_FILE_SHARED) &&
1641 __ceph_dir_is_complete(ci)) {
1642 /*
1643 * If a directory is complete, we want to keep
1644 * the exclusive cap. So that MDS does not end up
1645 * revoking the shared cap on every create/unlink
1646 * operation.
1647 */
1648 want = CEPH_CAP_ANY_SHARED | CEPH_CAP_FILE_EXCL;
1649 retain |= want;
Sage Weila8599bd2009-10-06 11:31:12 -07001650 } else {
Yan, Zheng32ec4392015-03-26 19:06:00 +08001651
Sage Weila8599bd2009-10-06 11:31:12 -07001652 retain |= CEPH_CAP_ANY_SHARED;
1653 /*
1654 * keep RD only if we didn't have the file open RW,
1655 * because then the mds would revoke it anyway to
1656 * journal max_size=0.
1657 */
1658 if (ci->i_max_size == 0)
1659 retain |= CEPH_CAP_ANY_RD;
1660 }
1661 }
1662
1663 dout("check_caps %p file_want %s used %s dirty %s flushing %s"
Sage Weilcbd03632010-02-09 13:41:18 -08001664 " issued %s revoking %s retain %s %s%s%s\n", inode,
Sage Weila8599bd2009-10-06 11:31:12 -07001665 ceph_cap_string(file_wanted),
1666 ceph_cap_string(used), ceph_cap_string(ci->i_dirty_caps),
1667 ceph_cap_string(ci->i_flushing_caps),
Sage Weilcbd03632010-02-09 13:41:18 -08001668 ceph_cap_string(issued), ceph_cap_string(revoking),
Sage Weila8599bd2009-10-06 11:31:12 -07001669 ceph_cap_string(retain),
1670 (flags & CHECK_CAPS_AUTHONLY) ? " AUTHONLY" : "",
1671 (flags & CHECK_CAPS_NODELAY) ? " NODELAY" : "",
1672 (flags & CHECK_CAPS_FLUSH) ? " FLUSH" : "");
1673
1674 /*
1675 * If we no longer need to hold onto old our caps, and we may
1676 * have cached pages, but don't want them, then try to invalidate.
1677 * If we fail, it's because pages are locked.... try again later.
1678 */
1679 if ((!is_delayed || mdsc->stopping) &&
Yan, Zhengfdd4e152015-06-16 20:48:56 +08001680 !S_ISDIR(inode->i_mode) && /* ignore readdir cache */
Yan, Zheng9abd4db2016-05-18 20:58:26 +08001681 !(ci->i_wb_ref || ci->i_wrbuffer_ref) && /* no dirty pages... */
Yan, Zhengfdd4e152015-06-16 20:48:56 +08001682 inode->i_data.nrpages && /* have cached pages */
Yan, Zheng5e804ac2015-10-26 16:08:43 +08001683 (revoking & (CEPH_CAP_FILE_CACHE|
1684 CEPH_CAP_FILE_LAZYIO)) && /* or revoking cache */
Sage Weila8599bd2009-10-06 11:31:12 -07001685 !tried_invalidate) {
Sage Weila8599bd2009-10-06 11:31:12 -07001686 dout("check_caps trying to invalidate on %p\n", inode);
Sage Weil5ecad6f2010-02-17 10:43:37 -08001687 if (try_nonblocking_invalidate(inode) < 0) {
Sage Weil29625072010-05-27 10:40:43 -07001688 if (revoking & (CEPH_CAP_FILE_CACHE|
1689 CEPH_CAP_FILE_LAZYIO)) {
Sage Weil5ecad6f2010-02-17 10:43:37 -08001690 dout("check_caps queuing invalidate\n");
Yan, Zheng36094042016-07-06 15:37:42 +08001691 queue_invalidate = true;
Sage Weil5ecad6f2010-02-17 10:43:37 -08001692 ci->i_rdcache_revoking = ci->i_rdcache_gen;
1693 } else {
1694 dout("check_caps failed to invalidate pages\n");
1695 /* we failed to invalidate pages. check these
1696 caps again later. */
Yan, Zheng36094042016-07-06 15:37:42 +08001697 force_requeue = true;
Sage Weil5ecad6f2010-02-17 10:43:37 -08001698 __cap_set_timeouts(mdsc, ci);
1699 }
Sage Weila8599bd2009-10-06 11:31:12 -07001700 }
Yan, Zheng36094042016-07-06 15:37:42 +08001701 tried_invalidate = true;
Sage Weila8599bd2009-10-06 11:31:12 -07001702 goto retry_locked;
1703 }
1704
1705 num = 0;
1706 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
1707 cap = rb_entry(p, struct ceph_cap, ci_node);
1708 num++;
1709
1710 /* avoid looping forever */
1711 if (mds >= cap->mds ||
1712 ((flags & CHECK_CAPS_AUTHONLY) && cap != ci->i_auth_cap))
1713 continue;
1714
1715 /* NOTE: no side-effects allowed, until we take s_mutex */
1716
Yan, Zheng395c3122013-01-04 14:37:57 +08001717 cap_used = used;
1718 if (ci->i_auth_cap && cap != ci->i_auth_cap)
1719 cap_used &= ~ci->i_auth_cap->issued;
1720
Sage Weila8599bd2009-10-06 11:31:12 -07001721 revoking = cap->implemented & ~cap->issued;
Yan, Zheng395c3122013-01-04 14:37:57 +08001722 dout(" mds%d cap %p used %s issued %s implemented %s revoking %s\n",
Yan, Zheng9abd4db2016-05-18 20:58:26 +08001723 cap->mds, cap, ceph_cap_string(cap_used),
1724 ceph_cap_string(cap->issued),
Sage Weil088b3f52011-01-18 08:56:01 -08001725 ceph_cap_string(cap->implemented),
1726 ceph_cap_string(revoking));
Sage Weila8599bd2009-10-06 11:31:12 -07001727
1728 if (cap == ci->i_auth_cap &&
1729 (cap->issued & CEPH_CAP_FILE_WR)) {
1730 /* request larger max_size from MDS? */
1731 if (ci->i_wanted_max_size > ci->i_max_size &&
1732 ci->i_wanted_max_size > ci->i_requested_max_size) {
1733 dout("requesting new max_size\n");
1734 goto ack;
1735 }
1736
1737 /* approaching file_max? */
1738 if ((inode->i_size << 1) >= ci->i_max_size &&
1739 (ci->i_reported_size << 1) < ci->i_max_size) {
1740 dout("i_size approaching max_size\n");
1741 goto ack;
1742 }
1743 }
1744 /* flush anything dirty? */
Yan, Zheng7bc00fd2016-07-07 18:34:45 +08001745 if (cap == ci->i_auth_cap) {
1746 if ((flags & CHECK_CAPS_FLUSH) && ci->i_dirty_caps) {
1747 dout("flushing dirty caps\n");
1748 goto ack;
1749 }
1750 if (ci->i_ceph_flags & CEPH_I_FLUSH_SNAPS) {
1751 dout("flushing snap caps\n");
1752 goto ack;
1753 }
Sage Weila8599bd2009-10-06 11:31:12 -07001754 }
1755
1756 /* completed revocation? going down and there are no caps? */
Yan, Zheng395c3122013-01-04 14:37:57 +08001757 if (revoking && (revoking & cap_used) == 0) {
Sage Weila8599bd2009-10-06 11:31:12 -07001758 dout("completed revocation of %s\n",
1759 ceph_cap_string(cap->implemented & ~cap->issued));
1760 goto ack;
1761 }
1762
1763 /* want more caps from mds? */
1764 if (want & ~(cap->mds_wanted | cap->issued))
1765 goto ack;
1766
1767 /* things we might delay */
1768 if ((cap->issued & ~retain) == 0 &&
1769 cap->mds_wanted == want)
1770 continue; /* nope, all good */
1771
1772 if (is_delayed)
1773 goto ack;
1774
1775 /* delay? */
1776 if ((ci->i_ceph_flags & CEPH_I_NODELAY) == 0 &&
1777 time_before(jiffies, ci->i_hold_caps_max)) {
1778 dout(" delaying issued %s -> %s, wanted %s -> %s\n",
1779 ceph_cap_string(cap->issued),
1780 ceph_cap_string(cap->issued & retain),
1781 ceph_cap_string(cap->mds_wanted),
1782 ceph_cap_string(want));
1783 delayed++;
1784 continue;
1785 }
1786
1787ack:
Sage Weile9964c12010-03-01 15:16:56 -08001788 if (ci->i_ceph_flags & CEPH_I_NOFLUSH) {
1789 dout(" skipping %p I_NOFLUSH set\n", inode);
1790 continue;
1791 }
1792
Sage Weila8599bd2009-10-06 11:31:12 -07001793 if (session && session != cap->session) {
1794 dout("oops, wrong session %p mutex\n", session);
1795 mutex_unlock(&session->s_mutex);
1796 session = NULL;
1797 }
1798 if (!session) {
1799 session = cap->session;
1800 if (mutex_trylock(&session->s_mutex) == 0) {
1801 dout("inverting session/ino locks on %p\n",
1802 session);
Sage Weilbe655592011-11-30 09:47:09 -08001803 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001804 if (took_snap_rwsem) {
1805 up_read(&mdsc->snap_rwsem);
1806 took_snap_rwsem = 0;
1807 }
1808 mutex_lock(&session->s_mutex);
1809 goto retry;
1810 }
1811 }
Yan, Zheng7bc00fd2016-07-07 18:34:45 +08001812
1813 /* kick flushing and flush snaps before sending normal
1814 * cap message */
1815 if (cap == ci->i_auth_cap &&
1816 (ci->i_ceph_flags &
1817 (CEPH_I_KICK_FLUSH | CEPH_I_FLUSH_SNAPS))) {
1818 if (ci->i_ceph_flags & CEPH_I_KICK_FLUSH) {
1819 spin_lock(&mdsc->cap_dirty_lock);
1820 oldest_flush_tid = __get_oldest_flush_tid(mdsc);
1821 spin_unlock(&mdsc->cap_dirty_lock);
1822 __kick_flushing_caps(mdsc, session, ci,
1823 oldest_flush_tid);
1824 ci->i_ceph_flags &= ~CEPH_I_KICK_FLUSH;
1825 }
Yan, Zhenged9b4302016-07-05 21:08:07 +08001826 if (ci->i_ceph_flags & CEPH_I_FLUSH_SNAPS)
1827 __ceph_flush_snaps(ci, session);
1828
Yan, Zheng7bc00fd2016-07-07 18:34:45 +08001829 goto retry_locked;
1830 }
1831
Sage Weila8599bd2009-10-06 11:31:12 -07001832 /* take snap_rwsem after session mutex */
1833 if (!took_snap_rwsem) {
1834 if (down_read_trylock(&mdsc->snap_rwsem) == 0) {
1835 dout("inverting snap/in locks on %p\n",
1836 inode);
Sage Weilbe655592011-11-30 09:47:09 -08001837 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001838 down_read(&mdsc->snap_rwsem);
1839 took_snap_rwsem = 1;
1840 goto retry;
1841 }
1842 took_snap_rwsem = 1;
1843 }
1844
Yan, Zheng553adfd2015-06-09 15:48:57 +08001845 if (cap == ci->i_auth_cap && ci->i_dirty_caps) {
Yan, Zhengc8799fc2016-07-07 15:22:38 +08001846 flushing = __mark_caps_flushing(inode, session, false,
Yan, Zhenga2971c82015-06-10 11:09:32 +08001847 &flush_tid,
1848 &oldest_flush_tid);
Yan, Zheng553adfd2015-06-09 15:48:57 +08001849 } else {
Sage Weil24be0c42011-01-18 08:48:06 -08001850 flushing = 0;
Yan, Zheng553adfd2015-06-09 15:48:57 +08001851 flush_tid = 0;
Yan, Zhenga2971c82015-06-10 11:09:32 +08001852 spin_lock(&mdsc->cap_dirty_lock);
1853 oldest_flush_tid = __get_oldest_flush_tid(mdsc);
1854 spin_unlock(&mdsc->cap_dirty_lock);
Yan, Zheng553adfd2015-06-09 15:48:57 +08001855 }
Sage Weila8599bd2009-10-06 11:31:12 -07001856
1857 mds = cap->mds; /* remember mds, so we don't repeat */
1858 sent++;
1859
Sage Weilbe655592011-11-30 09:47:09 -08001860 /* __send_cap drops i_ceph_lock */
Yan, Zheng395c3122013-01-04 14:37:57 +08001861 delayed += __send_cap(mdsc, cap, CEPH_CAP_OP_UPDATE, cap_used,
Yan, Zhenga2971c82015-06-10 11:09:32 +08001862 want, retain, flushing,
1863 flush_tid, oldest_flush_tid);
Sage Weilbe655592011-11-30 09:47:09 -08001864 goto retry; /* retake i_ceph_lock and restart our cap scan. */
Sage Weila8599bd2009-10-06 11:31:12 -07001865 }
1866
1867 /*
1868 * Reschedule delayed caps release if we delayed anything,
1869 * otherwise cancel.
1870 */
1871 if (delayed && is_delayed)
Yan, Zheng36094042016-07-06 15:37:42 +08001872 force_requeue = true; /* __send_cap delayed release; requeue */
Sage Weila8599bd2009-10-06 11:31:12 -07001873 if (!delayed && !is_delayed)
1874 __cap_delay_cancel(mdsc, ci);
1875 else if (!is_delayed || force_requeue)
1876 __cap_delay_requeue(mdsc, ci);
1877
Sage Weilbe655592011-11-30 09:47:09 -08001878 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001879
Sage Weilcbd03632010-02-09 13:41:18 -08001880 if (queue_invalidate)
Sage Weil3c6f6b72010-02-09 15:24:44 -08001881 ceph_queue_invalidate(inode);
Sage Weilcbd03632010-02-09 13:41:18 -08001882
Sage Weilcdc2ce02010-03-16 13:39:28 -07001883 if (session)
Sage Weila8599bd2009-10-06 11:31:12 -07001884 mutex_unlock(&session->s_mutex);
1885 if (took_snap_rwsem)
1886 up_read(&mdsc->snap_rwsem);
1887}
1888
1889/*
Sage Weila8599bd2009-10-06 11:31:12 -07001890 * Try to flush dirty caps back to the auth mds.
1891 */
Yan, Zheng553adfd2015-06-09 15:48:57 +08001892static int try_flush_caps(struct inode *inode, u64 *ptid)
Sage Weila8599bd2009-10-06 11:31:12 -07001893{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001894 struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
Sage Weila8599bd2009-10-06 11:31:12 -07001895 struct ceph_inode_info *ci = ceph_inode(inode);
Yan, Zheng4fe59782013-10-31 16:44:14 +08001896 struct ceph_mds_session *session = NULL;
Yan, Zheng89b52fe2015-05-27 09:59:48 +08001897 int flushing = 0;
Yan, Zhenga2971c82015-06-10 11:09:32 +08001898 u64 flush_tid = 0, oldest_flush_tid = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07001899
1900retry:
Sage Weilbe655592011-11-30 09:47:09 -08001901 spin_lock(&ci->i_ceph_lock);
Sage Weile9964c12010-03-01 15:16:56 -08001902 if (ci->i_ceph_flags & CEPH_I_NOFLUSH) {
1903 dout("try_flush_caps skipping %p I_NOFLUSH set\n", inode);
1904 goto out;
1905 }
Sage Weila8599bd2009-10-06 11:31:12 -07001906 if (ci->i_dirty_caps && ci->i_auth_cap) {
1907 struct ceph_cap *cap = ci->i_auth_cap;
1908 int used = __ceph_caps_used(ci);
1909 int want = __ceph_caps_wanted(ci);
1910 int delayed;
1911
Yan, Zheng4fe59782013-10-31 16:44:14 +08001912 if (!session || session != cap->session) {
Sage Weilbe655592011-11-30 09:47:09 -08001913 spin_unlock(&ci->i_ceph_lock);
Yan, Zheng4fe59782013-10-31 16:44:14 +08001914 if (session)
1915 mutex_unlock(&session->s_mutex);
Sage Weila8599bd2009-10-06 11:31:12 -07001916 session = cap->session;
1917 mutex_lock(&session->s_mutex);
1918 goto retry;
1919 }
Sage Weila8599bd2009-10-06 11:31:12 -07001920 if (cap->session->s_state < CEPH_MDS_SESSION_OPEN)
1921 goto out;
1922
Yan, Zhengc8799fc2016-07-07 15:22:38 +08001923 flushing = __mark_caps_flushing(inode, session, true,
1924 &flush_tid, &oldest_flush_tid);
Sage Weila8599bd2009-10-06 11:31:12 -07001925
Sage Weilbe655592011-11-30 09:47:09 -08001926 /* __send_cap drops i_ceph_lock */
Sage Weila8599bd2009-10-06 11:31:12 -07001927 delayed = __send_cap(mdsc, cap, CEPH_CAP_OP_FLUSH, used, want,
Yan, Zheng553adfd2015-06-09 15:48:57 +08001928 (cap->issued | cap->implemented),
Yan, Zhenga2971c82015-06-10 11:09:32 +08001929 flushing, flush_tid, oldest_flush_tid);
Sage Weila8599bd2009-10-06 11:31:12 -07001930
Yan, Zheng553adfd2015-06-09 15:48:57 +08001931 if (delayed) {
1932 spin_lock(&ci->i_ceph_lock);
Yan, Zheng89b52fe2015-05-27 09:59:48 +08001933 __cap_delay_requeue(mdsc, ci);
Yan, Zheng553adfd2015-06-09 15:48:57 +08001934 spin_unlock(&ci->i_ceph_lock);
1935 }
1936 } else {
Yan, Zhenge4500b52016-07-06 11:12:56 +08001937 if (!list_empty(&ci->i_cap_flush_list)) {
Yan, Zheng553adfd2015-06-09 15:48:57 +08001938 struct ceph_cap_flush *cf =
Yan, Zhenge4500b52016-07-06 11:12:56 +08001939 list_last_entry(&ci->i_cap_flush_list,
Yan, Zhengc8799fc2016-07-07 15:22:38 +08001940 struct ceph_cap_flush, i_list);
1941 cf->wake = true;
Yan, Zheng553adfd2015-06-09 15:48:57 +08001942 flush_tid = cf->tid;
1943 }
1944 flushing = ci->i_flushing_caps;
1945 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001946 }
1947out:
Yan, Zheng4fe59782013-10-31 16:44:14 +08001948 if (session)
Sage Weila8599bd2009-10-06 11:31:12 -07001949 mutex_unlock(&session->s_mutex);
Yan, Zheng553adfd2015-06-09 15:48:57 +08001950
1951 *ptid = flush_tid;
Sage Weila8599bd2009-10-06 11:31:12 -07001952 return flushing;
1953}
1954
1955/*
1956 * Return true if we've flushed caps through the given flush_tid.
1957 */
Yan, Zheng553adfd2015-06-09 15:48:57 +08001958static int caps_are_flushed(struct inode *inode, u64 flush_tid)
Sage Weila8599bd2009-10-06 11:31:12 -07001959{
1960 struct ceph_inode_info *ci = ceph_inode(inode);
Yan, Zheng553adfd2015-06-09 15:48:57 +08001961 int ret = 1;
Sage Weila8599bd2009-10-06 11:31:12 -07001962
Sage Weilbe655592011-11-30 09:47:09 -08001963 spin_lock(&ci->i_ceph_lock);
Yan, Zhenge4500b52016-07-06 11:12:56 +08001964 if (!list_empty(&ci->i_cap_flush_list)) {
1965 struct ceph_cap_flush * cf =
1966 list_first_entry(&ci->i_cap_flush_list,
1967 struct ceph_cap_flush, i_list);
Yan, Zheng553adfd2015-06-09 15:48:57 +08001968 if (cf->tid <= flush_tid)
Sage Weila8599bd2009-10-06 11:31:12 -07001969 ret = 0;
Yan, Zheng89b52fe2015-05-27 09:59:48 +08001970 }
Sage Weilbe655592011-11-30 09:47:09 -08001971 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001972 return ret;
1973}
1974
1975/*
Yan, Zheng68cd5b42015-10-27 18:36:06 +08001976 * wait for any unsafe requests to complete.
Yan, Zhengda819c82015-05-27 11:19:34 +08001977 */
Yan, Zheng68cd5b42015-10-27 18:36:06 +08001978static int unsafe_request_wait(struct inode *inode)
Yan, Zhengda819c82015-05-27 11:19:34 +08001979{
1980 struct ceph_inode_info *ci = ceph_inode(inode);
Yan, Zheng68cd5b42015-10-27 18:36:06 +08001981 struct ceph_mds_request *req1 = NULL, *req2 = NULL;
1982 int ret, err = 0;
Yan, Zhengda819c82015-05-27 11:19:34 +08001983
1984 spin_lock(&ci->i_unsafe_lock);
Yan, Zheng68cd5b42015-10-27 18:36:06 +08001985 if (S_ISDIR(inode->i_mode) && !list_empty(&ci->i_unsafe_dirops)) {
1986 req1 = list_last_entry(&ci->i_unsafe_dirops,
1987 struct ceph_mds_request,
1988 r_unsafe_dir_item);
1989 ceph_mdsc_get_request(req1);
1990 }
1991 if (!list_empty(&ci->i_unsafe_iops)) {
1992 req2 = list_last_entry(&ci->i_unsafe_iops,
1993 struct ceph_mds_request,
1994 r_unsafe_target_item);
1995 ceph_mdsc_get_request(req2);
1996 }
Yan, Zhengda819c82015-05-27 11:19:34 +08001997 spin_unlock(&ci->i_unsafe_lock);
Yan, Zheng68cd5b42015-10-27 18:36:06 +08001998
1999 dout("unsafe_requeset_wait %p wait on tid %llu %llu\n",
2000 inode, req1 ? req1->r_tid : 0ULL, req2 ? req2->r_tid : 0ULL);
2001 if (req1) {
2002 ret = !wait_for_completion_timeout(&req1->r_safe_completion,
2003 ceph_timeout_jiffies(req1->r_timeout));
2004 if (ret)
2005 err = -EIO;
2006 ceph_mdsc_put_request(req1);
2007 }
2008 if (req2) {
2009 ret = !wait_for_completion_timeout(&req2->r_safe_completion,
2010 ceph_timeout_jiffies(req2->r_timeout));
2011 if (ret)
2012 err = -EIO;
2013 ceph_mdsc_put_request(req2);
2014 }
2015 return err;
Yan, Zhengda819c82015-05-27 11:19:34 +08002016}
2017
Josef Bacik02c24a82011-07-16 20:44:56 -04002018int ceph_fsync(struct file *file, loff_t start, loff_t end, int datasync)
Sage Weila8599bd2009-10-06 11:31:12 -07002019{
Christoph Hellwig7ea80852010-05-26 17:53:25 +02002020 struct inode *inode = file->f_mapping->host;
Sage Weila8599bd2009-10-06 11:31:12 -07002021 struct ceph_inode_info *ci = ceph_inode(inode);
Yan, Zheng553adfd2015-06-09 15:48:57 +08002022 u64 flush_tid;
Sage Weila8599bd2009-10-06 11:31:12 -07002023 int ret;
2024 int dirty;
2025
2026 dout("fsync %p%s\n", inode, datasync ? " datasync" : "");
Yan, Zheng9a5530c2016-06-15 16:29:18 +08002027
2028 ceph_sync_write_wait(inode);
Sage Weila8599bd2009-10-06 11:31:12 -07002029
Josef Bacik02c24a82011-07-16 20:44:56 -04002030 ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
Sage Weila8599bd2009-10-06 11:31:12 -07002031 if (ret < 0)
Yan, Zhengda819c82015-05-27 11:19:34 +08002032 goto out;
2033
2034 if (datasync)
2035 goto out;
2036
Al Viro59551022016-01-22 15:40:57 -05002037 inode_lock(inode);
Sage Weila8599bd2009-10-06 11:31:12 -07002038
Yan, Zheng553adfd2015-06-09 15:48:57 +08002039 dirty = try_flush_caps(inode, &flush_tid);
Sage Weila8599bd2009-10-06 11:31:12 -07002040 dout("fsync dirty caps are %s\n", ceph_cap_string(dirty));
2041
Yan, Zheng68cd5b42015-10-27 18:36:06 +08002042 ret = unsafe_request_wait(inode);
Yan, Zhengda819c82015-05-27 11:19:34 +08002043
Sage Weila8599bd2009-10-06 11:31:12 -07002044 /*
2045 * only wait on non-file metadata writeback (the mds
2046 * can recover size and mtime, so we don't need to
2047 * wait for that)
2048 */
Yan, Zhengda819c82015-05-27 11:19:34 +08002049 if (!ret && (dirty & ~CEPH_CAP_ANY_FILE_WR)) {
Sage Weila8599bd2009-10-06 11:31:12 -07002050 ret = wait_event_interruptible(ci->i_cap_wq,
Yan, Zhengda819c82015-05-27 11:19:34 +08002051 caps_are_flushed(inode, flush_tid));
Sage Weila8599bd2009-10-06 11:31:12 -07002052 }
Al Viro59551022016-01-22 15:40:57 -05002053 inode_unlock(inode);
Yan, Zhengda819c82015-05-27 11:19:34 +08002054out:
2055 dout("fsync %p%s result=%d\n", inode, datasync ? " datasync" : "", ret);
Sage Weila8599bd2009-10-06 11:31:12 -07002056 return ret;
2057}
2058
2059/*
2060 * Flush any dirty caps back to the mds. If we aren't asked to wait,
2061 * queue inode for flush but don't do so immediately, because we can
2062 * get by with fewer MDS messages if we wait for data writeback to
2063 * complete first.
2064 */
Stephen Rothwellf1a3d572010-01-18 11:53:08 +11002065int ceph_write_inode(struct inode *inode, struct writeback_control *wbc)
Sage Weila8599bd2009-10-06 11:31:12 -07002066{
2067 struct ceph_inode_info *ci = ceph_inode(inode);
Yan, Zheng553adfd2015-06-09 15:48:57 +08002068 u64 flush_tid;
Sage Weila8599bd2009-10-06 11:31:12 -07002069 int err = 0;
2070 int dirty;
Stephen Rothwellf1a3d572010-01-18 11:53:08 +11002071 int wait = wbc->sync_mode == WB_SYNC_ALL;
Sage Weila8599bd2009-10-06 11:31:12 -07002072
2073 dout("write_inode %p wait=%d\n", inode, wait);
2074 if (wait) {
Yan, Zheng553adfd2015-06-09 15:48:57 +08002075 dirty = try_flush_caps(inode, &flush_tid);
Sage Weila8599bd2009-10-06 11:31:12 -07002076 if (dirty)
2077 err = wait_event_interruptible(ci->i_cap_wq,
2078 caps_are_flushed(inode, flush_tid));
2079 } else {
Cheng Renquan640ef792010-03-26 17:40:33 +08002080 struct ceph_mds_client *mdsc =
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002081 ceph_sb_to_client(inode->i_sb)->mdsc;
Sage Weila8599bd2009-10-06 11:31:12 -07002082
Sage Weilbe655592011-11-30 09:47:09 -08002083 spin_lock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07002084 if (__ceph_caps_dirty(ci))
2085 __cap_delay_requeue_front(mdsc, ci);
Sage Weilbe655592011-11-30 09:47:09 -08002086 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07002087 }
2088 return err;
2089}
2090
Yan, Zheng0e294382016-07-04 18:06:41 +08002091static void __kick_flushing_caps(struct ceph_mds_client *mdsc,
2092 struct ceph_mds_session *session,
2093 struct ceph_inode_info *ci,
2094 u64 oldest_flush_tid)
2095 __releases(ci->i_ceph_lock)
2096 __acquires(ci->i_ceph_lock)
Yan, Zheng553adfd2015-06-09 15:48:57 +08002097{
2098 struct inode *inode = &ci->vfs_inode;
2099 struct ceph_cap *cap;
2100 struct ceph_cap_flush *cf;
Yan, Zheng0e294382016-07-04 18:06:41 +08002101 int ret;
Yan, Zheng553adfd2015-06-09 15:48:57 +08002102 u64 first_tid = 0;
Yan, Zhenga2971c82015-06-10 11:09:32 +08002103
Yan, Zhenge4500b52016-07-06 11:12:56 +08002104 list_for_each_entry(cf, &ci->i_cap_flush_list, i_list) {
2105 if (cf->tid < first_tid)
2106 continue;
2107
Yan, Zheng553adfd2015-06-09 15:48:57 +08002108 cap = ci->i_auth_cap;
2109 if (!(cap && cap->session == session)) {
Yan, Zheng0e294382016-07-04 18:06:41 +08002110 pr_err("%p auth cap %p not mds%d ???\n",
2111 inode, cap, session->s_mds);
Yan, Zheng553adfd2015-06-09 15:48:57 +08002112 break;
2113 }
2114
Yan, Zheng553adfd2015-06-09 15:48:57 +08002115 first_tid = cf->tid + 1;
2116
Yan, Zheng0e294382016-07-04 18:06:41 +08002117 if (cf->caps) {
2118 dout("kick_flushing_caps %p cap %p tid %llu %s\n",
2119 inode, cap, cf->tid, ceph_cap_string(cf->caps));
2120 ci->i_ceph_flags |= CEPH_I_NODELAY;
2121 ret = __send_cap(mdsc, cap, CEPH_CAP_OP_FLUSH,
2122 __ceph_caps_used(ci),
2123 __ceph_caps_wanted(ci),
2124 cap->issued | cap->implemented,
2125 cf->caps, cf->tid, oldest_flush_tid);
2126 if (ret) {
2127 pr_err("kick_flushing_caps: error sending "
2128 "cap flush, ino (%llx.%llx) "
2129 "tid %llu flushing %s\n",
2130 ceph_vinop(inode), cf->tid,
2131 ceph_cap_string(cf->caps));
2132 }
2133 } else {
2134 struct ceph_cap_snap *capsnap =
2135 container_of(cf, struct ceph_cap_snap,
2136 cap_flush);
2137 dout("kick_flushing_caps %p capsnap %p tid %llu %s\n",
2138 inode, capsnap, cf->tid,
2139 ceph_cap_string(capsnap->dirty));
2140
2141 atomic_inc(&capsnap->nref);
2142 spin_unlock(&ci->i_ceph_lock);
2143
2144 ret = __send_flush_snap(inode, session, capsnap, cap->mseq,
2145 oldest_flush_tid);
2146 if (ret < 0) {
2147 pr_err("kick_flushing_caps: error sending "
2148 "cap flushsnap, ino (%llx.%llx) "
2149 "tid %llu follows %llu\n",
2150 ceph_vinop(inode), cf->tid,
2151 capsnap->follows);
2152 }
2153
2154 ceph_put_cap_snap(capsnap);
2155 }
Yan, Zhenge4500b52016-07-06 11:12:56 +08002156
2157 spin_lock(&ci->i_ceph_lock);
Yan, Zheng553adfd2015-06-09 15:48:57 +08002158 }
Yan, Zheng553adfd2015-06-09 15:48:57 +08002159}
2160
Yan, Zhenge548e9b2015-06-10 15:17:56 +08002161void ceph_early_kick_flushing_caps(struct ceph_mds_client *mdsc,
2162 struct ceph_mds_session *session)
2163{
2164 struct ceph_inode_info *ci;
2165 struct ceph_cap *cap;
Yan, Zheng0e294382016-07-04 18:06:41 +08002166 u64 oldest_flush_tid;
Yan, Zhenge548e9b2015-06-10 15:17:56 +08002167
2168 dout("early_kick_flushing_caps mds%d\n", session->s_mds);
Yan, Zheng0e294382016-07-04 18:06:41 +08002169
2170 spin_lock(&mdsc->cap_dirty_lock);
2171 oldest_flush_tid = __get_oldest_flush_tid(mdsc);
2172 spin_unlock(&mdsc->cap_dirty_lock);
2173
Yan, Zhenge548e9b2015-06-10 15:17:56 +08002174 list_for_each_entry(ci, &session->s_cap_flushing, i_flushing_item) {
2175 spin_lock(&ci->i_ceph_lock);
2176 cap = ci->i_auth_cap;
2177 if (!(cap && cap->session == session)) {
2178 pr_err("%p auth cap %p not mds%d ???\n",
2179 &ci->vfs_inode, cap, session->s_mds);
2180 spin_unlock(&ci->i_ceph_lock);
2181 continue;
2182 }
2183
2184
2185 /*
2186 * if flushing caps were revoked, we re-send the cap flush
2187 * in client reconnect stage. This guarantees MDS * processes
2188 * the cap flush message before issuing the flushing caps to
2189 * other client.
2190 */
2191 if ((cap->issued & ci->i_flushing_caps) !=
2192 ci->i_flushing_caps) {
Yan, Zheng13c2b572016-07-05 16:45:21 +08002193 ci->i_ceph_flags &= ~CEPH_I_KICK_FLUSH;
Yan, Zheng0e294382016-07-04 18:06:41 +08002194 __kick_flushing_caps(mdsc, session, ci,
2195 oldest_flush_tid);
Yan, Zheng13c2b572016-07-05 16:45:21 +08002196 } else {
2197 ci->i_ceph_flags |= CEPH_I_KICK_FLUSH;
Yan, Zhenge548e9b2015-06-10 15:17:56 +08002198 }
2199
Yan, Zhenge548e9b2015-06-10 15:17:56 +08002200 spin_unlock(&ci->i_ceph_lock);
2201 }
2202}
2203
Sage Weila8599bd2009-10-06 11:31:12 -07002204void ceph_kick_flushing_caps(struct ceph_mds_client *mdsc,
2205 struct ceph_mds_session *session)
2206{
2207 struct ceph_inode_info *ci;
Yan, Zheng13c2b572016-07-05 16:45:21 +08002208 struct ceph_cap *cap;
Yan, Zheng0e294382016-07-04 18:06:41 +08002209 u64 oldest_flush_tid;
Sage Weila8599bd2009-10-06 11:31:12 -07002210
2211 dout("kick_flushing_caps mds%d\n", session->s_mds);
Yan, Zheng0e294382016-07-04 18:06:41 +08002212
2213 spin_lock(&mdsc->cap_dirty_lock);
2214 oldest_flush_tid = __get_oldest_flush_tid(mdsc);
2215 spin_unlock(&mdsc->cap_dirty_lock);
2216
Sage Weila8599bd2009-10-06 11:31:12 -07002217 list_for_each_entry(ci, &session->s_cap_flushing, i_flushing_item) {
Yan, Zheng0e294382016-07-04 18:06:41 +08002218 spin_lock(&ci->i_ceph_lock);
Yan, Zheng13c2b572016-07-05 16:45:21 +08002219 cap = ci->i_auth_cap;
2220 if (!(cap && cap->session == session)) {
2221 pr_err("%p auth cap %p not mds%d ???\n",
2222 &ci->vfs_inode, cap, session->s_mds);
2223 spin_unlock(&ci->i_ceph_lock);
2224 continue;
2225 }
2226 if (ci->i_ceph_flags & CEPH_I_KICK_FLUSH) {
2227 ci->i_ceph_flags &= ~CEPH_I_KICK_FLUSH;
2228 __kick_flushing_caps(mdsc, session, ci,
2229 oldest_flush_tid);
2230 }
Yan, Zheng0e294382016-07-04 18:06:41 +08002231 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07002232 }
2233}
2234
Sage Weil088b3f52011-01-18 08:56:01 -08002235static void kick_flushing_inode_caps(struct ceph_mds_client *mdsc,
2236 struct ceph_mds_session *session,
2237 struct inode *inode)
Yan, Zheng0e294382016-07-04 18:06:41 +08002238 __releases(ci->i_ceph_lock)
Sage Weil088b3f52011-01-18 08:56:01 -08002239{
2240 struct ceph_inode_info *ci = ceph_inode(inode);
2241 struct ceph_cap *cap;
Sage Weil088b3f52011-01-18 08:56:01 -08002242
Sage Weil088b3f52011-01-18 08:56:01 -08002243 cap = ci->i_auth_cap;
Yan, Zheng8310b082015-06-09 17:20:12 +08002244 dout("kick_flushing_inode_caps %p flushing %s\n", inode,
2245 ceph_cap_string(ci->i_flushing_caps));
Yan, Zheng005c4692013-05-31 16:40:24 +08002246
Yan, Zheng0e294382016-07-04 18:06:41 +08002247 if (!list_empty(&ci->i_cap_flush_list)) {
2248 u64 oldest_flush_tid;
Yan, Zheng005c4692013-05-31 16:40:24 +08002249 spin_lock(&mdsc->cap_dirty_lock);
2250 list_move_tail(&ci->i_flushing_item,
2251 &cap->session->s_cap_flushing);
Yan, Zheng0e294382016-07-04 18:06:41 +08002252 oldest_flush_tid = __get_oldest_flush_tid(mdsc);
Yan, Zheng005c4692013-05-31 16:40:24 +08002253 spin_unlock(&mdsc->cap_dirty_lock);
2254
Yan, Zheng13c2b572016-07-05 16:45:21 +08002255 ci->i_ceph_flags &= ~CEPH_I_KICK_FLUSH;
Yan, Zheng0e294382016-07-04 18:06:41 +08002256 __kick_flushing_caps(mdsc, session, ci, oldest_flush_tid);
Yan, Zheng553adfd2015-06-09 15:48:57 +08002257 spin_unlock(&ci->i_ceph_lock);
Sage Weil088b3f52011-01-18 08:56:01 -08002258 } else {
Sage Weilbe655592011-11-30 09:47:09 -08002259 spin_unlock(&ci->i_ceph_lock);
Sage Weil088b3f52011-01-18 08:56:01 -08002260 }
2261}
2262
Sage Weila8599bd2009-10-06 11:31:12 -07002263
2264/*
2265 * Take references to capabilities we hold, so that we don't release
2266 * them to the MDS prematurely.
2267 *
Sage Weilbe655592011-11-30 09:47:09 -08002268 * Protected by i_ceph_lock.
Sage Weila8599bd2009-10-06 11:31:12 -07002269 */
Yan, Zheng5dda377c2015-04-30 14:40:54 +08002270static void __take_cap_refs(struct ceph_inode_info *ci, int got,
2271 bool snap_rwsem_locked)
Sage Weila8599bd2009-10-06 11:31:12 -07002272{
2273 if (got & CEPH_CAP_PIN)
2274 ci->i_pin_ref++;
2275 if (got & CEPH_CAP_FILE_RD)
2276 ci->i_rd_ref++;
2277 if (got & CEPH_CAP_FILE_CACHE)
2278 ci->i_rdcache_ref++;
Yan, Zheng5dda377c2015-04-30 14:40:54 +08002279 if (got & CEPH_CAP_FILE_WR) {
2280 if (ci->i_wr_ref == 0 && !ci->i_head_snapc) {
2281 BUG_ON(!snap_rwsem_locked);
2282 ci->i_head_snapc = ceph_get_snap_context(
2283 ci->i_snap_realm->cached_context);
2284 }
Sage Weila8599bd2009-10-06 11:31:12 -07002285 ci->i_wr_ref++;
Yan, Zheng5dda377c2015-04-30 14:40:54 +08002286 }
Sage Weila8599bd2009-10-06 11:31:12 -07002287 if (got & CEPH_CAP_FILE_BUFFER) {
Henry C Changd3d07202011-05-11 10:29:54 +00002288 if (ci->i_wb_ref == 0)
Sage Weil3772d262011-05-03 09:28:08 -07002289 ihold(&ci->vfs_inode);
Henry C Changd3d07202011-05-11 10:29:54 +00002290 ci->i_wb_ref++;
2291 dout("__take_cap_refs %p wb %d -> %d (?)\n",
2292 &ci->vfs_inode, ci->i_wb_ref-1, ci->i_wb_ref);
Sage Weila8599bd2009-10-06 11:31:12 -07002293 }
2294}
2295
2296/*
2297 * Try to grab cap references. Specify those refs we @want, and the
2298 * minimal set we @need. Also include the larger offset we are writing
2299 * to (when applicable), and check against max_size here as well.
2300 * Note that caller is responsible for ensuring max_size increases are
2301 * requested from the MDS.
2302 */
2303static int try_get_cap_refs(struct ceph_inode_info *ci, int need, int want,
Yan, Zheng5dda377c2015-04-30 14:40:54 +08002304 loff_t endoff, bool nonblock, int *got, int *err)
Sage Weila8599bd2009-10-06 11:31:12 -07002305{
2306 struct inode *inode = &ci->vfs_inode;
Yan, Zheng5dda377c2015-04-30 14:40:54 +08002307 struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
Sage Weila8599bd2009-10-06 11:31:12 -07002308 int ret = 0;
Yan, Zhengc4d4a582015-01-09 15:56:18 +08002309 int have, implemented;
Sage Weil195d3ce2010-03-01 09:57:54 -08002310 int file_wanted;
Yan, Zheng5dda377c2015-04-30 14:40:54 +08002311 bool snap_rwsem_locked = false;
Sage Weila8599bd2009-10-06 11:31:12 -07002312
2313 dout("get_cap_refs %p need %s want %s\n", inode,
2314 ceph_cap_string(need), ceph_cap_string(want));
Yan, Zhengc4d4a582015-01-09 15:56:18 +08002315
Yan, Zheng5dda377c2015-04-30 14:40:54 +08002316again:
Sage Weilbe655592011-11-30 09:47:09 -08002317 spin_lock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07002318
Sage Weil195d3ce2010-03-01 09:57:54 -08002319 /* make sure file is actually open */
2320 file_wanted = __ceph_caps_file_wanted(ci);
Yan, Zheng77310322016-04-08 15:27:16 +08002321 if ((file_wanted & need) != need) {
Sage Weil195d3ce2010-03-01 09:57:54 -08002322 dout("try_get_cap_refs need %s file_wanted %s, EBADF\n",
2323 ceph_cap_string(need), ceph_cap_string(file_wanted));
Sage Weila8599bd2009-10-06 11:31:12 -07002324 *err = -EBADF;
2325 ret = 1;
Yan, Zheng3738daa2014-11-14 22:10:07 +08002326 goto out_unlock;
Sage Weila8599bd2009-10-06 11:31:12 -07002327 }
2328
Yan, Zheng37505d52013-04-12 16:11:10 +08002329 /* finish pending truncate */
2330 while (ci->i_truncate_pending) {
2331 spin_unlock(&ci->i_ceph_lock);
Yan, Zheng5dda377c2015-04-30 14:40:54 +08002332 if (snap_rwsem_locked) {
2333 up_read(&mdsc->snap_rwsem);
2334 snap_rwsem_locked = false;
2335 }
Yan, Zhengb415bf42013-07-02 12:40:19 +08002336 __ceph_do_pending_vmtruncate(inode);
Yan, Zheng37505d52013-04-12 16:11:10 +08002337 spin_lock(&ci->i_ceph_lock);
2338 }
2339
Yan, Zheng3871cbb2013-08-05 14:10:29 +08002340 have = __ceph_caps_issued(ci, &implemented);
2341
2342 if (have & need & CEPH_CAP_FILE_WR) {
Sage Weila8599bd2009-10-06 11:31:12 -07002343 if (endoff >= 0 && endoff > (loff_t)ci->i_max_size) {
2344 dout("get_cap_refs %p endoff %llu > maxsize %llu\n",
2345 inode, endoff, ci->i_max_size);
Yan, Zheng3871cbb2013-08-05 14:10:29 +08002346 if (endoff > ci->i_requested_max_size) {
Yan, Zheng5dda377c2015-04-30 14:40:54 +08002347 *err = -EAGAIN;
Sage Weila8599bd2009-10-06 11:31:12 -07002348 ret = 1;
2349 }
Yan, Zheng3738daa2014-11-14 22:10:07 +08002350 goto out_unlock;
Sage Weila8599bd2009-10-06 11:31:12 -07002351 }
2352 /*
2353 * If a sync write is in progress, we must wait, so that we
2354 * can get a final snapshot value for size+mtime.
2355 */
2356 if (__ceph_have_pending_cap_snap(ci)) {
2357 dout("get_cap_refs %p cap_snap_pending\n", inode);
Yan, Zheng3738daa2014-11-14 22:10:07 +08002358 goto out_unlock;
Sage Weila8599bd2009-10-06 11:31:12 -07002359 }
2360 }
Sage Weila8599bd2009-10-06 11:31:12 -07002361
Sage Weila8599bd2009-10-06 11:31:12 -07002362 if ((have & need) == need) {
2363 /*
2364 * Look at (implemented & ~have & not) so that we keep waiting
2365 * on transition from wanted -> needed caps. This is needed
2366 * for WRBUFFER|WR -> WR to avoid a new WR sync write from
2367 * going before a prior buffered writeback happens.
2368 */
2369 int not = want & ~(have & need);
2370 int revoking = implemented & ~have;
2371 dout("get_cap_refs %p have %s but not %s (revoking %s)\n",
2372 inode, ceph_cap_string(have), ceph_cap_string(not),
2373 ceph_cap_string(revoking));
2374 if ((revoking & not) == 0) {
Yan, Zheng5dda377c2015-04-30 14:40:54 +08002375 if (!snap_rwsem_locked &&
2376 !ci->i_head_snapc &&
2377 (need & CEPH_CAP_FILE_WR)) {
2378 if (!down_read_trylock(&mdsc->snap_rwsem)) {
2379 /*
2380 * we can not call down_read() when
2381 * task isn't in TASK_RUNNING state
2382 */
2383 if (nonblock) {
2384 *err = -EAGAIN;
2385 ret = 1;
2386 goto out_unlock;
2387 }
2388
2389 spin_unlock(&ci->i_ceph_lock);
2390 down_read(&mdsc->snap_rwsem);
2391 snap_rwsem_locked = true;
2392 goto again;
2393 }
2394 snap_rwsem_locked = true;
2395 }
Yan, Zhengc4d4a582015-01-09 15:56:18 +08002396 *got = need | (have & want);
Yan, Zhengf7f7e7a2016-05-18 20:31:55 +08002397 if ((need & CEPH_CAP_FILE_RD) &&
2398 !(*got & CEPH_CAP_FILE_CACHE))
2399 ceph_disable_fscache_readpage(ci);
Yan, Zheng5dda377c2015-04-30 14:40:54 +08002400 __take_cap_refs(ci, *got, true);
Sage Weila8599bd2009-10-06 11:31:12 -07002401 ret = 1;
2402 }
2403 } else {
Yan, Zheng03f4fcb2015-01-05 11:04:04 +08002404 int session_readonly = false;
2405 if ((need & CEPH_CAP_FILE_WR) && ci->i_auth_cap) {
2406 struct ceph_mds_session *s = ci->i_auth_cap->session;
2407 spin_lock(&s->s_cap_lock);
2408 session_readonly = s->s_readonly;
2409 spin_unlock(&s->s_cap_lock);
2410 }
2411 if (session_readonly) {
2412 dout("get_cap_refs %p needed %s but mds%d readonly\n",
2413 inode, ceph_cap_string(need), ci->i_auth_cap->mds);
2414 *err = -EROFS;
2415 ret = 1;
2416 goto out_unlock;
2417 }
2418
Yan, Zheng77310322016-04-08 15:27:16 +08002419 if (ci->i_ceph_flags & CEPH_I_CAP_DROPPED) {
2420 int mds_wanted;
2421 if (ACCESS_ONCE(mdsc->fsc->mount_state) ==
2422 CEPH_MOUNT_SHUTDOWN) {
2423 dout("get_cap_refs %p forced umount\n", inode);
2424 *err = -EIO;
2425 ret = 1;
2426 goto out_unlock;
2427 }
2428 mds_wanted = __ceph_caps_mds_wanted(ci);
2429 if ((mds_wanted & need) != need) {
2430 dout("get_cap_refs %p caps were dropped"
2431 " (session killed?)\n", inode);
2432 *err = -ESTALE;
2433 ret = 1;
2434 goto out_unlock;
2435 }
2436 if ((mds_wanted & file_wanted) ==
2437 (file_wanted & (CEPH_CAP_FILE_RD|CEPH_CAP_FILE_WR)))
2438 ci->i_ceph_flags &= ~CEPH_I_CAP_DROPPED;
Yan, Zheng48fec5d2015-07-01 16:27:46 +08002439 }
2440
Sage Weila8599bd2009-10-06 11:31:12 -07002441 dout("get_cap_refs %p have %s needed %s\n", inode,
2442 ceph_cap_string(have), ceph_cap_string(need));
2443 }
Yan, Zheng3738daa2014-11-14 22:10:07 +08002444out_unlock:
Sage Weilbe655592011-11-30 09:47:09 -08002445 spin_unlock(&ci->i_ceph_lock);
Yan, Zheng5dda377c2015-04-30 14:40:54 +08002446 if (snap_rwsem_locked)
2447 up_read(&mdsc->snap_rwsem);
Yan, Zheng3738daa2014-11-14 22:10:07 +08002448
Sage Weila8599bd2009-10-06 11:31:12 -07002449 dout("get_cap_refs %p ret %d got %s\n", inode,
Yan, Zhengc4d4a582015-01-09 15:56:18 +08002450 ret, ceph_cap_string(*got));
Sage Weila8599bd2009-10-06 11:31:12 -07002451 return ret;
2452}
2453
2454/*
2455 * Check the offset we are writing up to against our current
2456 * max_size. If necessary, tell the MDS we want to write to
2457 * a larger offset.
2458 */
2459static void check_max_size(struct inode *inode, loff_t endoff)
2460{
2461 struct ceph_inode_info *ci = ceph_inode(inode);
2462 int check = 0;
2463
2464 /* do we need to explicitly request a larger max_size? */
Sage Weilbe655592011-11-30 09:47:09 -08002465 spin_lock(&ci->i_ceph_lock);
Yan, Zheng3871cbb2013-08-05 14:10:29 +08002466 if (endoff >= ci->i_max_size && endoff > ci->i_wanted_max_size) {
Sage Weila8599bd2009-10-06 11:31:12 -07002467 dout("write %p at large endoff %llu, req max_size\n",
2468 inode, endoff);
2469 ci->i_wanted_max_size = endoff;
Sage Weila8599bd2009-10-06 11:31:12 -07002470 }
Yan, Zheng3871cbb2013-08-05 14:10:29 +08002471 /* duplicate ceph_check_caps()'s logic */
2472 if (ci->i_auth_cap &&
2473 (ci->i_auth_cap->issued & CEPH_CAP_FILE_WR) &&
2474 ci->i_wanted_max_size > ci->i_max_size &&
2475 ci->i_wanted_max_size > ci->i_requested_max_size)
2476 check = 1;
Sage Weilbe655592011-11-30 09:47:09 -08002477 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07002478 if (check)
2479 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
2480}
2481
2482/*
2483 * Wait for caps, and take cap references. If we can't get a WR cap
2484 * due to a small max_size, make sure we check_max_size (and possibly
2485 * ask the mds) so we don't get hung up indefinitely.
2486 */
Yan, Zheng3738daa2014-11-14 22:10:07 +08002487int ceph_get_caps(struct ceph_inode_info *ci, int need, int want,
2488 loff_t endoff, int *got, struct page **pinned_page)
Sage Weila8599bd2009-10-06 11:31:12 -07002489{
Yan, Zheng5dda377c2015-04-30 14:40:54 +08002490 int _got, ret, err = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07002491
Yan, Zheng10183a62015-04-27 15:33:28 +08002492 ret = ceph_pool_perm_check(ci, need);
2493 if (ret < 0)
2494 return ret;
2495
Yan, Zheng5dda377c2015-04-30 14:40:54 +08002496 while (true) {
2497 if (endoff > 0)
2498 check_max_size(&ci->vfs_inode, endoff);
Yan, Zhengc4d4a582015-01-09 15:56:18 +08002499
Yan, Zheng5dda377c2015-04-30 14:40:54 +08002500 err = 0;
Yan, Zhengc4d4a582015-01-09 15:56:18 +08002501 _got = 0;
Yan, Zheng5dda377c2015-04-30 14:40:54 +08002502 ret = try_get_cap_refs(ci, need, want, endoff,
2503 false, &_got, &err);
2504 if (ret) {
2505 if (err == -EAGAIN)
2506 continue;
2507 if (err < 0)
Yan, Zheng77310322016-04-08 15:27:16 +08002508 ret = err;
Yan, Zheng5dda377c2015-04-30 14:40:54 +08002509 } else {
2510 ret = wait_event_interruptible(ci->i_cap_wq,
2511 try_get_cap_refs(ci, need, want, endoff,
2512 true, &_got, &err));
2513 if (err == -EAGAIN)
2514 continue;
2515 if (err < 0)
2516 ret = err;
Yan, Zheng77310322016-04-08 15:27:16 +08002517 }
2518 if (ret < 0) {
2519 if (err == -ESTALE) {
2520 /* session was killed, try renew caps */
2521 ret = ceph_renew_caps(&ci->vfs_inode);
2522 if (ret == 0)
2523 continue;
2524 }
2525 return ret;
Yan, Zheng5dda377c2015-04-30 14:40:54 +08002526 }
Yan, Zhengc4d4a582015-01-09 15:56:18 +08002527
Yan, Zheng5dda377c2015-04-30 14:40:54 +08002528 if (ci->i_inline_version != CEPH_INLINE_NONE &&
2529 (_got & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) &&
2530 i_size_read(&ci->vfs_inode) > 0) {
2531 struct page *page =
2532 find_get_page(ci->vfs_inode.i_mapping, 0);
2533 if (page) {
2534 if (PageUptodate(page)) {
2535 *pinned_page = page;
2536 break;
2537 }
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002538 put_page(page);
Yan, Zheng5dda377c2015-04-30 14:40:54 +08002539 }
2540 /*
2541 * drop cap refs first because getattr while
2542 * holding * caps refs can cause deadlock.
2543 */
2544 ceph_put_cap_refs(ci, _got);
2545 _got = 0;
2546
2547 /*
2548 * getattr request will bring inline data into
2549 * page cache
2550 */
2551 ret = __ceph_do_getattr(&ci->vfs_inode, NULL,
2552 CEPH_STAT_CAP_INLINE_DATA,
2553 true);
2554 if (ret < 0)
2555 return ret;
2556 continue;
2557 }
2558 break;
Yan, Zhengc4d4a582015-01-09 15:56:18 +08002559 }
Yan, Zheng5dda377c2015-04-30 14:40:54 +08002560
Yan, Zhengf7f7e7a2016-05-18 20:31:55 +08002561 if ((_got & CEPH_CAP_FILE_RD) && (_got & CEPH_CAP_FILE_CACHE))
2562 ceph_fscache_revalidate_cookie(ci);
2563
Yan, Zhengc4d4a582015-01-09 15:56:18 +08002564 *got = _got;
2565 return 0;
Sage Weila8599bd2009-10-06 11:31:12 -07002566}
2567
2568/*
2569 * Take cap refs. Caller must already know we hold at least one ref
2570 * on the caps in question or we don't know this is safe.
2571 */
2572void ceph_get_cap_refs(struct ceph_inode_info *ci, int caps)
2573{
Sage Weilbe655592011-11-30 09:47:09 -08002574 spin_lock(&ci->i_ceph_lock);
Yan, Zheng5dda377c2015-04-30 14:40:54 +08002575 __take_cap_refs(ci, caps, false);
Sage Weilbe655592011-11-30 09:47:09 -08002576 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07002577}
2578
Yan, Zheng86056092015-05-01 16:57:16 +08002579
2580/*
2581 * drop cap_snap that is not associated with any snapshot.
2582 * we don't need to send FLUSHSNAP message for it.
2583 */
Yan, Zheng70220ac2016-07-06 16:21:30 +08002584static int ceph_try_drop_cap_snap(struct ceph_inode_info *ci,
2585 struct ceph_cap_snap *capsnap)
Yan, Zheng86056092015-05-01 16:57:16 +08002586{
2587 if (!capsnap->need_flush &&
2588 !capsnap->writing && !capsnap->dirty_pages) {
Yan, Zheng86056092015-05-01 16:57:16 +08002589 dout("dropping cap_snap %p follows %llu\n",
2590 capsnap, capsnap->follows);
Yan, Zheng0e294382016-07-04 18:06:41 +08002591 BUG_ON(capsnap->cap_flush.tid > 0);
Yan, Zheng86056092015-05-01 16:57:16 +08002592 ceph_put_snap_context(capsnap->context);
Yan, Zheng70220ac2016-07-06 16:21:30 +08002593 if (!list_is_last(&capsnap->ci_item, &ci->i_cap_snaps))
2594 ci->i_ceph_flags |= CEPH_I_FLUSH_SNAPS;
2595
Yan, Zheng86056092015-05-01 16:57:16 +08002596 list_del(&capsnap->ci_item);
Yan, Zheng86056092015-05-01 16:57:16 +08002597 ceph_put_cap_snap(capsnap);
2598 return 1;
2599 }
2600 return 0;
2601}
2602
Sage Weila8599bd2009-10-06 11:31:12 -07002603/*
2604 * Release cap refs.
2605 *
2606 * If we released the last ref on any given cap, call ceph_check_caps
2607 * to release (or schedule a release).
2608 *
2609 * If we are releasing a WR cap (from a sync write), finalize any affected
2610 * cap_snap, and wake up any waiters.
2611 */
2612void ceph_put_cap_refs(struct ceph_inode_info *ci, int had)
2613{
2614 struct inode *inode = &ci->vfs_inode;
2615 int last = 0, put = 0, flushsnaps = 0, wake = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07002616
Sage Weilbe655592011-11-30 09:47:09 -08002617 spin_lock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07002618 if (had & CEPH_CAP_PIN)
2619 --ci->i_pin_ref;
2620 if (had & CEPH_CAP_FILE_RD)
2621 if (--ci->i_rd_ref == 0)
2622 last++;
2623 if (had & CEPH_CAP_FILE_CACHE)
2624 if (--ci->i_rdcache_ref == 0)
2625 last++;
2626 if (had & CEPH_CAP_FILE_BUFFER) {
Henry C Changd3d07202011-05-11 10:29:54 +00002627 if (--ci->i_wb_ref == 0) {
Sage Weila8599bd2009-10-06 11:31:12 -07002628 last++;
2629 put++;
2630 }
Henry C Changd3d07202011-05-11 10:29:54 +00002631 dout("put_cap_refs %p wb %d -> %d (?)\n",
2632 inode, ci->i_wb_ref+1, ci->i_wb_ref);
Sage Weila8599bd2009-10-06 11:31:12 -07002633 }
2634 if (had & CEPH_CAP_FILE_WR)
2635 if (--ci->i_wr_ref == 0) {
2636 last++;
Yan, Zheng86056092015-05-01 16:57:16 +08002637 if (__ceph_have_pending_cap_snap(ci)) {
2638 struct ceph_cap_snap *capsnap =
2639 list_last_entry(&ci->i_cap_snaps,
2640 struct ceph_cap_snap,
2641 ci_item);
2642 capsnap->writing = 0;
Yan, Zheng70220ac2016-07-06 16:21:30 +08002643 if (ceph_try_drop_cap_snap(ci, capsnap))
Yan, Zheng86056092015-05-01 16:57:16 +08002644 put++;
2645 else if (__ceph_finish_cap_snap(ci, capsnap))
2646 flushsnaps = 1;
2647 wake = 1;
Sage Weila8599bd2009-10-06 11:31:12 -07002648 }
Yan, Zheng5dda377c2015-04-30 14:40:54 +08002649 if (ci->i_wrbuffer_ref_head == 0 &&
2650 ci->i_dirty_caps == 0 &&
2651 ci->i_flushing_caps == 0) {
2652 BUG_ON(!ci->i_head_snapc);
2653 ceph_put_snap_context(ci->i_head_snapc);
2654 ci->i_head_snapc = NULL;
2655 }
Yan, Zhengdb40cc12015-03-23 20:12:20 +08002656 /* see comment in __ceph_remove_cap() */
2657 if (!__ceph_is_any_caps(ci) && ci->i_snap_realm)
2658 drop_inode_snap_realm(ci);
Sage Weila8599bd2009-10-06 11:31:12 -07002659 }
Sage Weilbe655592011-11-30 09:47:09 -08002660 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07002661
Sage Weil819ccbf2010-04-01 09:33:46 -07002662 dout("put_cap_refs %p had %s%s%s\n", inode, ceph_cap_string(had),
2663 last ? " last" : "", put ? " put" : "");
Sage Weila8599bd2009-10-06 11:31:12 -07002664
2665 if (last && !flushsnaps)
2666 ceph_check_caps(ci, 0, NULL);
2667 else if (flushsnaps)
Yan, Zhenged9b4302016-07-05 21:08:07 +08002668 ceph_flush_snaps(ci, NULL);
Sage Weila8599bd2009-10-06 11:31:12 -07002669 if (wake)
Yehuda Sadeh03066f22010-07-27 13:11:08 -07002670 wake_up_all(&ci->i_cap_wq);
Yan, Zheng86056092015-05-01 16:57:16 +08002671 while (put-- > 0)
Sage Weila8599bd2009-10-06 11:31:12 -07002672 iput(inode);
2673}
2674
2675/*
2676 * Release @nr WRBUFFER refs on dirty pages for the given @snapc snap
2677 * context. Adjust per-snap dirty page accounting as appropriate.
2678 * Once all dirty data for a cap_snap is flushed, flush snapped file
2679 * metadata back to the MDS. If we dropped the last ref, call
2680 * ceph_check_caps.
2681 */
2682void ceph_put_wrbuffer_cap_refs(struct ceph_inode_info *ci, int nr,
2683 struct ceph_snap_context *snapc)
2684{
2685 struct inode *inode = &ci->vfs_inode;
Sage Weila8599bd2009-10-06 11:31:12 -07002686 struct ceph_cap_snap *capsnap = NULL;
Yan, Zheng70220ac2016-07-06 16:21:30 +08002687 int put = 0;
2688 bool last = false;
2689 bool found = false;
2690 bool flush_snaps = false;
2691 bool complete_capsnap = false;
Sage Weila8599bd2009-10-06 11:31:12 -07002692
Sage Weilbe655592011-11-30 09:47:09 -08002693 spin_lock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07002694 ci->i_wrbuffer_ref -= nr;
Yan, Zheng70220ac2016-07-06 16:21:30 +08002695 if (ci->i_wrbuffer_ref == 0) {
2696 last = true;
2697 put++;
2698 }
Sage Weila8599bd2009-10-06 11:31:12 -07002699
2700 if (ci->i_head_snapc == snapc) {
2701 ci->i_wrbuffer_ref_head -= nr;
Sage Weil7d8cb262010-08-24 08:44:16 -07002702 if (ci->i_wrbuffer_ref_head == 0 &&
Yan, Zheng5dda377c2015-04-30 14:40:54 +08002703 ci->i_wr_ref == 0 &&
2704 ci->i_dirty_caps == 0 &&
2705 ci->i_flushing_caps == 0) {
Sage Weil7d8cb262010-08-24 08:44:16 -07002706 BUG_ON(!ci->i_head_snapc);
Sage Weila8599bd2009-10-06 11:31:12 -07002707 ceph_put_snap_context(ci->i_head_snapc);
2708 ci->i_head_snapc = NULL;
2709 }
2710 dout("put_wrbuffer_cap_refs on %p head %d/%d -> %d/%d %s\n",
2711 inode,
2712 ci->i_wrbuffer_ref+nr, ci->i_wrbuffer_ref_head+nr,
2713 ci->i_wrbuffer_ref, ci->i_wrbuffer_ref_head,
2714 last ? " LAST" : "");
2715 } else {
2716 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
2717 if (capsnap->context == snapc) {
Yan, Zheng70220ac2016-07-06 16:21:30 +08002718 found = true;
Sage Weila8599bd2009-10-06 11:31:12 -07002719 break;
2720 }
2721 }
2722 BUG_ON(!found);
Sage Weil819ccbf2010-04-01 09:33:46 -07002723 capsnap->dirty_pages -= nr;
2724 if (capsnap->dirty_pages == 0) {
Yan, Zheng70220ac2016-07-06 16:21:30 +08002725 complete_capsnap = true;
2726 if (!capsnap->writing) {
2727 if (ceph_try_drop_cap_snap(ci, capsnap)) {
2728 put++;
2729 } else {
2730 ci->i_ceph_flags |= CEPH_I_FLUSH_SNAPS;
2731 flush_snaps = true;
2732 }
2733 }
Sage Weil819ccbf2010-04-01 09:33:46 -07002734 }
Sage Weila8599bd2009-10-06 11:31:12 -07002735 dout("put_wrbuffer_cap_refs on %p cap_snap %p "
Yan, Zheng86056092015-05-01 16:57:16 +08002736 " snap %lld %d/%d -> %d/%d %s%s\n",
Sage Weila8599bd2009-10-06 11:31:12 -07002737 inode, capsnap, capsnap->context->seq,
2738 ci->i_wrbuffer_ref+nr, capsnap->dirty_pages + nr,
2739 ci->i_wrbuffer_ref, capsnap->dirty_pages,
2740 last ? " (wrbuffer last)" : "",
Yan, Zheng86056092015-05-01 16:57:16 +08002741 complete_capsnap ? " (complete capsnap)" : "");
Sage Weila8599bd2009-10-06 11:31:12 -07002742 }
2743
Sage Weilbe655592011-11-30 09:47:09 -08002744 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07002745
2746 if (last) {
2747 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
Yan, Zheng70220ac2016-07-06 16:21:30 +08002748 } else if (flush_snaps) {
Yan, Zhenged9b4302016-07-05 21:08:07 +08002749 ceph_flush_snaps(ci, NULL);
Sage Weila8599bd2009-10-06 11:31:12 -07002750 }
Yan, Zheng70220ac2016-07-06 16:21:30 +08002751 if (complete_capsnap)
2752 wake_up_all(&ci->i_cap_wq);
2753 while (put-- > 0)
Sage Weil819ccbf2010-04-01 09:33:46 -07002754 iput(inode);
Sage Weila8599bd2009-10-06 11:31:12 -07002755}
2756
2757/*
Yan, Zhengca20c992013-07-21 10:07:51 +08002758 * Invalidate unlinked inode's aliases, so we can drop the inode ASAP.
2759 */
2760static void invalidate_aliases(struct inode *inode)
2761{
2762 struct dentry *dn, *prev = NULL;
2763
2764 dout("invalidate_aliases inode %p\n", inode);
2765 d_prune_aliases(inode);
2766 /*
2767 * For non-directory inode, d_find_alias() only returns
J. Bruce Fieldsfc12c802014-01-16 17:42:53 -05002768 * hashed dentry. After calling d_invalidate(), the
2769 * dentry becomes unhashed.
Yan, Zhengca20c992013-07-21 10:07:51 +08002770 *
Yan, Zhenga8d436f2013-09-02 15:19:54 +08002771 * For directory inode, d_find_alias() can return
J. Bruce Fieldsfc12c802014-01-16 17:42:53 -05002772 * unhashed dentry. But directory inode should have
Yan, Zhengca20c992013-07-21 10:07:51 +08002773 * one alias at most.
2774 */
2775 while ((dn = d_find_alias(inode))) {
2776 if (dn == prev) {
2777 dput(dn);
2778 break;
2779 }
Yan, Zhenga8d436f2013-09-02 15:19:54 +08002780 d_invalidate(dn);
Yan, Zhengca20c992013-07-21 10:07:51 +08002781 if (prev)
2782 dput(prev);
2783 prev = dn;
2784 }
2785 if (prev)
2786 dput(prev);
2787}
2788
2789/*
Sage Weila8599bd2009-10-06 11:31:12 -07002790 * Handle a cap GRANT message from the MDS. (Note that a GRANT may
2791 * actually be a revocation if it specifies a smaller cap set.)
2792 *
Sage Weilbe655592011-11-30 09:47:09 -08002793 * caller holds s_mutex and i_ceph_lock, we drop both.
Sage Weila8599bd2009-10-06 11:31:12 -07002794 */
Yan, Zheng2cd698b2014-04-18 13:20:27 +08002795static void handle_cap_grant(struct ceph_mds_client *mdsc,
2796 struct inode *inode, struct ceph_mds_caps *grant,
Yan, Zheng779fe0f2016-03-07 09:35:06 +08002797 struct ceph_string **pns, u64 inline_version,
2798 void *inline_data, u32 inline_len,
Yan, Zheng2cd698b2014-04-18 13:20:27 +08002799 struct ceph_buffer *xattr_buf,
Sage Weil15637c82010-03-16 13:42:00 -07002800 struct ceph_mds_session *session,
Yan, Zheng779fe0f2016-03-07 09:35:06 +08002801 struct ceph_cap *cap, int issued)
Yan, Zheng2cd698b2014-04-18 13:20:27 +08002802 __releases(ci->i_ceph_lock)
Yan, Zheng982d6012014-12-23 15:30:54 +08002803 __releases(mdsc->snap_rwsem)
Sage Weila8599bd2009-10-06 11:31:12 -07002804{
2805 struct ceph_inode_info *ci = ceph_inode(inode);
2806 int mds = session->s_mds;
Sage Weil2f56f562010-10-27 20:59:49 -07002807 int seq = le32_to_cpu(grant->seq);
Sage Weila8599bd2009-10-06 11:31:12 -07002808 int newcaps = le32_to_cpu(grant->caps);
Yan, Zheng2cd698b2014-04-18 13:20:27 +08002809 int used, wanted, dirty;
Sage Weila8599bd2009-10-06 11:31:12 -07002810 u64 size = le64_to_cpu(grant->size);
2811 u64 max_size = le64_to_cpu(grant->max_size);
2812 struct timespec mtime, atime, ctime;
Sage Weil15637c82010-03-16 13:42:00 -07002813 int check_caps = 0;
Fabian Frederickab6c2c32014-10-09 23:16:35 +02002814 bool wake = false;
2815 bool writeback = false;
2816 bool queue_trunc = false;
2817 bool queue_invalidate = false;
Fabian Frederickab6c2c32014-10-09 23:16:35 +02002818 bool deleted_inode = false;
Yan, Zheng31c542a2014-11-14 21:41:55 +08002819 bool fill_inline = false;
Sage Weila8599bd2009-10-06 11:31:12 -07002820
Sage Weil2f56f562010-10-27 20:59:49 -07002821 dout("handle_cap_grant inode %p cap %p mds%d seq %d %s\n",
2822 inode, cap, mds, seq, ceph_cap_string(newcaps));
Sage Weila8599bd2009-10-06 11:31:12 -07002823 dout(" size %llu max_size %llu, i_size %llu\n", size, max_size,
2824 inode->i_size);
2825
Yan, Zheng11df2df2013-11-24 14:44:38 +08002826
2827 /*
2828 * auth mds of the inode changed. we received the cap export message,
2829 * but still haven't received the cap import message. handle_cap_export
2830 * updated the new auth MDS' cap.
2831 *
2832 * "ceph_seq_cmp(seq, cap->seq) <= 0" means we are processing a message
2833 * that was sent before the cap import message. So don't remove caps.
2834 */
2835 if (ceph_seq_cmp(seq, cap->seq) <= 0) {
2836 WARN_ON(cap != ci->i_auth_cap);
2837 WARN_ON(cap->cap_id != le64_to_cpu(grant->cap_id));
2838 seq = cap->seq;
2839 newcaps |= cap->issued;
2840 }
2841
Sage Weila8599bd2009-10-06 11:31:12 -07002842 /*
2843 * If CACHE is being revoked, and we have no dirty buffers,
2844 * try to invalidate (once). (If there are dirty buffers, we
2845 * will invalidate _after_ writeback.)
2846 */
Yan, Zhengfdd4e152015-06-16 20:48:56 +08002847 if (!S_ISDIR(inode->i_mode) && /* don't invalidate readdir cache */
2848 ((cap->issued & ~newcaps) & CEPH_CAP_FILE_CACHE) &&
Sage Weil3b454c42010-06-10 13:20:33 -07002849 (newcaps & CEPH_CAP_FILE_LAZYIO) == 0 &&
Yan, Zheng9abd4db2016-05-18 20:58:26 +08002850 !(ci->i_wrbuffer_ref || ci->i_wb_ref)) {
Li Wange9075742013-08-15 22:00:25 -07002851 if (try_nonblocking_invalidate(inode)) {
Sage Weila8599bd2009-10-06 11:31:12 -07002852 /* there were locked pages.. invalidate later
2853 in a separate thread. */
2854 if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
Fabian Frederickab6c2c32014-10-09 23:16:35 +02002855 queue_invalidate = true;
Sage Weila8599bd2009-10-06 11:31:12 -07002856 ci->i_rdcache_revoking = ci->i_rdcache_gen;
2857 }
Sage Weila8599bd2009-10-06 11:31:12 -07002858 }
Sage Weila8599bd2009-10-06 11:31:12 -07002859 }
2860
2861 /* side effects now are allowed */
Sage Weil685f9a5d2009-11-09 12:05:48 -08002862 cap->cap_gen = session->s_cap_gen;
Yan, Zheng11df2df2013-11-24 14:44:38 +08002863 cap->seq = seq;
Sage Weila8599bd2009-10-06 11:31:12 -07002864
2865 __check_cap_issue(ci, cap, newcaps);
2866
Yan, Zhengf98a1282014-04-17 08:55:50 +08002867 if ((newcaps & CEPH_CAP_AUTH_SHARED) &&
2868 (issued & CEPH_CAP_AUTH_EXCL) == 0) {
Sage Weila8599bd2009-10-06 11:31:12 -07002869 inode->i_mode = le32_to_cpu(grant->mode);
Eric W. Biederman05cb11c2013-01-31 02:56:19 -08002870 inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(grant->uid));
2871 inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(grant->gid));
Sage Weila8599bd2009-10-06 11:31:12 -07002872 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
Eric W. Biedermanbd2bae62013-01-31 04:05:39 -08002873 from_kuid(&init_user_ns, inode->i_uid),
2874 from_kgid(&init_user_ns, inode->i_gid));
Sage Weila8599bd2009-10-06 11:31:12 -07002875 }
2876
Yan, Zhengf98a1282014-04-17 08:55:50 +08002877 if ((newcaps & CEPH_CAP_AUTH_SHARED) &&
2878 (issued & CEPH_CAP_LINK_EXCL) == 0) {
Miklos Szeredibfe86842011-10-28 14:13:29 +02002879 set_nlink(inode, le32_to_cpu(grant->nlink));
Yan, Zhengca20c992013-07-21 10:07:51 +08002880 if (inode->i_nlink == 0 &&
2881 (newcaps & (CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL)))
Fabian Frederickab6c2c32014-10-09 23:16:35 +02002882 deleted_inode = true;
Yan, Zhengca20c992013-07-21 10:07:51 +08002883 }
Sage Weila8599bd2009-10-06 11:31:12 -07002884
2885 if ((issued & CEPH_CAP_XATTR_EXCL) == 0 && grant->xattr_len) {
2886 int len = le32_to_cpu(grant->xattr_len);
2887 u64 version = le64_to_cpu(grant->xattr_version);
2888
2889 if (version > ci->i_xattrs.version) {
2890 dout(" got new xattrs v%llu on %p len %d\n",
2891 version, inode, len);
2892 if (ci->i_xattrs.blob)
2893 ceph_buffer_put(ci->i_xattrs.blob);
2894 ci->i_xattrs.blob = ceph_buffer_get(xattr_buf);
2895 ci->i_xattrs.version = version;
Guangliang Zhao7221fe42013-11-11 15:18:03 +08002896 ceph_forget_all_cached_acls(inode);
Sage Weila8599bd2009-10-06 11:31:12 -07002897 }
2898 }
2899
Yan, Zhengf98a1282014-04-17 08:55:50 +08002900 if (newcaps & CEPH_CAP_ANY_RD) {
2901 /* ctime/mtime/atime? */
2902 ceph_decode_timespec(&mtime, &grant->mtime);
2903 ceph_decode_timespec(&atime, &grant->atime);
2904 ceph_decode_timespec(&ctime, &grant->ctime);
2905 ceph_fill_file_time(inode, issued,
2906 le32_to_cpu(grant->time_warp_seq),
2907 &ctime, &mtime, &atime);
2908 }
Sage Weila8599bd2009-10-06 11:31:12 -07002909
Yan, Zhengf98a1282014-04-17 08:55:50 +08002910 if (newcaps & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR)) {
2911 /* file layout may have changed */
Yan, Zheng76271512016-02-03 21:24:49 +08002912 s64 old_pool = ci->i_layout.pool_id;
Yan, Zheng779fe0f2016-03-07 09:35:06 +08002913 struct ceph_string *old_ns;
2914
Yan, Zheng76271512016-02-03 21:24:49 +08002915 ceph_file_layout_from_legacy(&ci->i_layout, &grant->layout);
Yan, Zheng779fe0f2016-03-07 09:35:06 +08002916 old_ns = rcu_dereference_protected(ci->i_layout.pool_ns,
2917 lockdep_is_held(&ci->i_ceph_lock));
2918 rcu_assign_pointer(ci->i_layout.pool_ns, *pns);
2919
2920 if (ci->i_layout.pool_id != old_pool || *pns != old_ns)
Yan, Zheng76271512016-02-03 21:24:49 +08002921 ci->i_ceph_flags &= ~CEPH_I_POOL_PERM;
Yan, Zheng5ea5c5e2016-02-14 18:06:41 +08002922
Yan, Zheng779fe0f2016-03-07 09:35:06 +08002923 *pns = old_ns;
2924
Yan, Zhengf98a1282014-04-17 08:55:50 +08002925 /* size/truncate_seq? */
2926 queue_trunc = ceph_fill_file_size(inode, issued,
2927 le32_to_cpu(grant->truncate_seq),
2928 le64_to_cpu(grant->truncate_size),
2929 size);
2930 /* max size increase? */
2931 if (ci->i_auth_cap == cap && max_size != ci->i_max_size) {
2932 dout("max_size %lld -> %llu\n",
2933 ci->i_max_size, max_size);
2934 ci->i_max_size = max_size;
2935 if (max_size >= ci->i_wanted_max_size) {
2936 ci->i_wanted_max_size = 0; /* reset */
2937 ci->i_requested_max_size = 0;
2938 }
Fabian Frederickab6c2c32014-10-09 23:16:35 +02002939 wake = true;
Sage Weila8599bd2009-10-06 11:31:12 -07002940 }
Sage Weila8599bd2009-10-06 11:31:12 -07002941 }
2942
2943 /* check cap bits */
2944 wanted = __ceph_caps_wanted(ci);
2945 used = __ceph_caps_used(ci);
2946 dirty = __ceph_caps_dirty(ci);
2947 dout(" my wanted = %s, used = %s, dirty %s\n",
2948 ceph_cap_string(wanted),
2949 ceph_cap_string(used),
2950 ceph_cap_string(dirty));
2951 if (wanted != le32_to_cpu(grant->wanted)) {
2952 dout("mds wanted %s -> %s\n",
2953 ceph_cap_string(le32_to_cpu(grant->wanted)),
2954 ceph_cap_string(wanted));
Yan, Zheng390306c2013-01-04 15:30:10 +08002955 /* imported cap may not have correct mds_wanted */
2956 if (le32_to_cpu(grant->op) == CEPH_CAP_OP_IMPORT)
2957 check_caps = 1;
Sage Weila8599bd2009-10-06 11:31:12 -07002958 }
2959
Sage Weila8599bd2009-10-06 11:31:12 -07002960 /* revocation, grant, or no-op? */
2961 if (cap->issued & ~newcaps) {
Sage Weil3b454c42010-06-10 13:20:33 -07002962 int revoking = cap->issued & ~newcaps;
2963
2964 dout("revocation: %s -> %s (revoking %s)\n",
2965 ceph_cap_string(cap->issued),
2966 ceph_cap_string(newcaps),
2967 ceph_cap_string(revoking));
Sage Weil0eb6cd42010-08-05 13:53:18 -07002968 if (revoking & used & CEPH_CAP_FILE_BUFFER)
Fabian Frederickab6c2c32014-10-09 23:16:35 +02002969 writeback = true; /* initiate writeback; will delay ack */
Sage Weil3b454c42010-06-10 13:20:33 -07002970 else if (revoking == CEPH_CAP_FILE_CACHE &&
2971 (newcaps & CEPH_CAP_FILE_LAZYIO) == 0 &&
2972 queue_invalidate)
2973 ; /* do nothing yet, invalidation will be queued */
2974 else if (cap == ci->i_auth_cap)
2975 check_caps = 1; /* check auth cap only */
2976 else
2977 check_caps = 2; /* check all caps */
Sage Weila8599bd2009-10-06 11:31:12 -07002978 cap->issued = newcaps;
Sage Weil978097c2010-03-08 15:27:53 -08002979 cap->implemented |= newcaps;
Sage Weila8599bd2009-10-06 11:31:12 -07002980 } else if (cap->issued == newcaps) {
2981 dout("caps unchanged: %s -> %s\n",
2982 ceph_cap_string(cap->issued), ceph_cap_string(newcaps));
2983 } else {
2984 dout("grant: %s -> %s\n", ceph_cap_string(cap->issued),
2985 ceph_cap_string(newcaps));
Yan, Zheng6ee6b9532013-07-02 12:40:21 +08002986 /* non-auth MDS is revoking the newly grant caps ? */
2987 if (cap == ci->i_auth_cap &&
2988 __ceph_caps_revoking_other(ci, cap, newcaps))
2989 check_caps = 2;
2990
Sage Weila8599bd2009-10-06 11:31:12 -07002991 cap->issued = newcaps;
2992 cap->implemented |= newcaps; /* add bits only, to
2993 * avoid stepping on a
2994 * pending revocation */
Fabian Frederickab6c2c32014-10-09 23:16:35 +02002995 wake = true;
Sage Weila8599bd2009-10-06 11:31:12 -07002996 }
Sage Weil978097c2010-03-08 15:27:53 -08002997 BUG_ON(cap->issued & ~cap->implemented);
Sage Weila8599bd2009-10-06 11:31:12 -07002998
Yan, Zheng31c542a2014-11-14 21:41:55 +08002999 if (inline_version > 0 && inline_version >= ci->i_inline_version) {
3000 ci->i_inline_version = inline_version;
3001 if (ci->i_inline_version != CEPH_INLINE_NONE &&
3002 (newcaps & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)))
3003 fill_inline = true;
3004 }
3005
Yan, Zheng2cd698b2014-04-18 13:20:27 +08003006 if (le32_to_cpu(grant->op) == CEPH_CAP_OP_IMPORT) {
Yan, Zheng2cd698b2014-04-18 13:20:27 +08003007 if (newcaps & ~issued)
Fabian Frederickab6c2c32014-10-09 23:16:35 +02003008 wake = true;
Yan, Zheng0e294382016-07-04 18:06:41 +08003009 kick_flushing_inode_caps(mdsc, session, inode);
3010 up_read(&mdsc->snap_rwsem);
3011 } else {
3012 spin_unlock(&ci->i_ceph_lock);
Yan, Zheng2cd698b2014-04-18 13:20:27 +08003013 }
3014
Yan, Zheng31c542a2014-11-14 21:41:55 +08003015 if (fill_inline)
3016 ceph_fill_inline_data(inode, NULL, inline_data, inline_len);
3017
Yan, Zheng14649752016-05-20 15:41:20 +08003018 if (queue_trunc)
Yan, Zhengc6bcda62014-04-11 10:18:07 +08003019 ceph_queue_vmtruncate(inode);
Yan, Zhengc6bcda62014-04-11 10:18:07 +08003020
Sage Weil3c6f6b72010-02-09 15:24:44 -08003021 if (writeback)
Sage Weila8599bd2009-10-06 11:31:12 -07003022 /*
3023 * queue inode for writeback: we can't actually call
3024 * filemap_write_and_wait, etc. from message handler
3025 * context.
3026 */
Sage Weil3c6f6b72010-02-09 15:24:44 -08003027 ceph_queue_writeback(inode);
3028 if (queue_invalidate)
3029 ceph_queue_invalidate(inode);
Yan, Zhengca20c992013-07-21 10:07:51 +08003030 if (deleted_inode)
3031 invalidate_aliases(inode);
Sage Weila8599bd2009-10-06 11:31:12 -07003032 if (wake)
Yehuda Sadeh03066f22010-07-27 13:11:08 -07003033 wake_up_all(&ci->i_cap_wq);
Sage Weil15637c82010-03-16 13:42:00 -07003034
3035 if (check_caps == 1)
3036 ceph_check_caps(ci, CHECK_CAPS_NODELAY|CHECK_CAPS_AUTHONLY,
3037 session);
3038 else if (check_caps == 2)
3039 ceph_check_caps(ci, CHECK_CAPS_NODELAY, session);
3040 else
3041 mutex_unlock(&session->s_mutex);
Sage Weila8599bd2009-10-06 11:31:12 -07003042}
3043
3044/*
3045 * Handle FLUSH_ACK from MDS, indicating that metadata we sent to the
3046 * MDS has been safely committed.
3047 */
Sage Weil6df058c2009-12-22 11:24:33 -08003048static void handle_cap_flush_ack(struct inode *inode, u64 flush_tid,
Sage Weila8599bd2009-10-06 11:31:12 -07003049 struct ceph_mds_caps *m,
3050 struct ceph_mds_session *session,
3051 struct ceph_cap *cap)
Sage Weilbe655592011-11-30 09:47:09 -08003052 __releases(ci->i_ceph_lock)
Sage Weila8599bd2009-10-06 11:31:12 -07003053{
3054 struct ceph_inode_info *ci = ceph_inode(inode);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07003055 struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
Yan, Zhenge4500b52016-07-06 11:12:56 +08003056 struct ceph_cap_flush *cf, *tmp_cf;
Yan, Zheng553adfd2015-06-09 15:48:57 +08003057 LIST_HEAD(to_remove);
Sage Weila8599bd2009-10-06 11:31:12 -07003058 unsigned seq = le32_to_cpu(m->seq);
3059 int dirty = le32_to_cpu(m->dirty);
3060 int cleaned = 0;
Yan, Zhengc8799fc2016-07-07 15:22:38 +08003061 bool drop = false;
3062 bool wake_ci = 0;
3063 bool wake_mdsc = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07003064
Yan, Zhenge4500b52016-07-06 11:12:56 +08003065 list_for_each_entry_safe(cf, tmp_cf, &ci->i_cap_flush_list, i_list) {
Yan, Zheng553adfd2015-06-09 15:48:57 +08003066 if (cf->tid == flush_tid)
3067 cleaned = cf->caps;
Yan, Zheng0e294382016-07-04 18:06:41 +08003068 if (cf->caps == 0) /* capsnap */
3069 continue;
Yan, Zheng553adfd2015-06-09 15:48:57 +08003070 if (cf->tid <= flush_tid) {
Yan, Zhengc8799fc2016-07-07 15:22:38 +08003071 if (__finish_cap_flush(NULL, ci, cf))
3072 wake_ci = true;
Yan, Zhenge4500b52016-07-06 11:12:56 +08003073 list_add_tail(&cf->i_list, &to_remove);
Yan, Zheng553adfd2015-06-09 15:48:57 +08003074 } else {
3075 cleaned &= ~cf->caps;
3076 if (!cleaned)
3077 break;
3078 }
3079 }
Sage Weila8599bd2009-10-06 11:31:12 -07003080
3081 dout("handle_cap_flush_ack inode %p mds%d seq %d on %s cleaned %s,"
3082 " flushing %s -> %s\n",
3083 inode, session->s_mds, seq, ceph_cap_string(dirty),
3084 ceph_cap_string(cleaned), ceph_cap_string(ci->i_flushing_caps),
3085 ceph_cap_string(ci->i_flushing_caps & ~cleaned));
3086
Yan, Zheng8310b082015-06-09 17:20:12 +08003087 if (list_empty(&to_remove) && !cleaned)
Sage Weila8599bd2009-10-06 11:31:12 -07003088 goto out;
3089
Sage Weila8599bd2009-10-06 11:31:12 -07003090 ci->i_flushing_caps &= ~cleaned;
Sage Weila8599bd2009-10-06 11:31:12 -07003091
3092 spin_lock(&mdsc->cap_dirty_lock);
Yan, Zheng8310b082015-06-09 17:20:12 +08003093
Yan, Zhengc8799fc2016-07-07 15:22:38 +08003094 list_for_each_entry(cf, &to_remove, i_list) {
3095 if (__finish_cap_flush(mdsc, NULL, cf))
3096 wake_mdsc = true;
Yan, Zheng8310b082015-06-09 17:20:12 +08003097 }
3098
Sage Weila8599bd2009-10-06 11:31:12 -07003099 if (ci->i_flushing_caps == 0) {
Yan, Zheng0e294382016-07-04 18:06:41 +08003100 if (list_empty(&ci->i_cap_flush_list)) {
3101 list_del_init(&ci->i_flushing_item);
3102 if (!list_empty(&session->s_cap_flushing)) {
3103 dout(" mds%d still flushing cap on %p\n",
3104 session->s_mds,
3105 &list_first_entry(&session->s_cap_flushing,
3106 struct ceph_inode_info,
3107 i_flushing_item)->vfs_inode);
3108 }
3109 }
Sage Weila8599bd2009-10-06 11:31:12 -07003110 mdsc->num_cap_flushing--;
Sage Weila8599bd2009-10-06 11:31:12 -07003111 dout(" inode %p now !flushing\n", inode);
Sage Weilafcdaea2009-10-14 14:27:38 -07003112
3113 if (ci->i_dirty_caps == 0) {
3114 dout(" inode %p now clean\n", inode);
3115 BUG_ON(!list_empty(&ci->i_dirty_item));
Yan, Zhengc8799fc2016-07-07 15:22:38 +08003116 drop = true;
Yan, Zheng5dda377c2015-04-30 14:40:54 +08003117 if (ci->i_wr_ref == 0 &&
3118 ci->i_wrbuffer_ref_head == 0) {
Sage Weil7d8cb262010-08-24 08:44:16 -07003119 BUG_ON(!ci->i_head_snapc);
3120 ceph_put_snap_context(ci->i_head_snapc);
3121 ci->i_head_snapc = NULL;
3122 }
Sage Weil76e3b392009-10-15 18:13:53 -07003123 } else {
3124 BUG_ON(list_empty(&ci->i_dirty_item));
Sage Weilafcdaea2009-10-14 14:27:38 -07003125 }
Sage Weila8599bd2009-10-06 11:31:12 -07003126 }
3127 spin_unlock(&mdsc->cap_dirty_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07003128
3129out:
Sage Weilbe655592011-11-30 09:47:09 -08003130 spin_unlock(&ci->i_ceph_lock);
Yan, Zheng553adfd2015-06-09 15:48:57 +08003131
3132 while (!list_empty(&to_remove)) {
3133 cf = list_first_entry(&to_remove,
Yan, Zhenge4500b52016-07-06 11:12:56 +08003134 struct ceph_cap_flush, i_list);
3135 list_del(&cf->i_list);
Yan, Zhengf66fd9f2015-06-10 17:26:13 +08003136 ceph_free_cap_flush(cf);
Yan, Zheng553adfd2015-06-09 15:48:57 +08003137 }
Yan, Zhengc8799fc2016-07-07 15:22:38 +08003138
3139 if (wake_ci)
3140 wake_up_all(&ci->i_cap_wq);
3141 if (wake_mdsc)
3142 wake_up_all(&mdsc->cap_flushing_wq);
Sage Weilafcdaea2009-10-14 14:27:38 -07003143 if (drop)
Sage Weila8599bd2009-10-06 11:31:12 -07003144 iput(inode);
3145}
3146
3147/*
3148 * Handle FLUSHSNAP_ACK. MDS has flushed snap data to disk and we can
3149 * throw away our cap_snap.
3150 *
3151 * Caller hold s_mutex.
3152 */
Sage Weil6df058c2009-12-22 11:24:33 -08003153static void handle_cap_flushsnap_ack(struct inode *inode, u64 flush_tid,
Sage Weila8599bd2009-10-06 11:31:12 -07003154 struct ceph_mds_caps *m,
3155 struct ceph_mds_session *session)
3156{
3157 struct ceph_inode_info *ci = ceph_inode(inode);
Yan, Zhengaffbc192015-05-05 21:22:13 +08003158 struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
Sage Weila8599bd2009-10-06 11:31:12 -07003159 u64 follows = le64_to_cpu(m->snap_follows);
Sage Weila8599bd2009-10-06 11:31:12 -07003160 struct ceph_cap_snap *capsnap;
Yan, Zhengc8799fc2016-07-07 15:22:38 +08003161 bool flushed = false;
3162 bool wake_ci = false;
3163 bool wake_mdsc = false;
Sage Weila8599bd2009-10-06 11:31:12 -07003164
3165 dout("handle_cap_flushsnap_ack inode %p ci %p mds%d follows %lld\n",
3166 inode, ci, session->s_mds, follows);
3167
Sage Weilbe655592011-11-30 09:47:09 -08003168 spin_lock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07003169 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
3170 if (capsnap->follows == follows) {
Yan, Zheng0e294382016-07-04 18:06:41 +08003171 if (capsnap->cap_flush.tid != flush_tid) {
Sage Weila8599bd2009-10-06 11:31:12 -07003172 dout(" cap_snap %p follows %lld tid %lld !="
3173 " %lld\n", capsnap, follows,
Yan, Zheng0e294382016-07-04 18:06:41 +08003174 flush_tid, capsnap->cap_flush.tid);
Sage Weila8599bd2009-10-06 11:31:12 -07003175 break;
3176 }
Yan, Zhengc8799fc2016-07-07 15:22:38 +08003177 flushed = true;
Sage Weila8599bd2009-10-06 11:31:12 -07003178 break;
3179 } else {
3180 dout(" skipping cap_snap %p follows %lld\n",
3181 capsnap, capsnap->follows);
3182 }
3183 }
Yan, Zheng0e294382016-07-04 18:06:41 +08003184 if (flushed) {
Yan, Zheng0e294382016-07-04 18:06:41 +08003185 WARN_ON(capsnap->dirty_pages || capsnap->writing);
3186 dout(" removing %p cap_snap %p follows %lld\n",
3187 inode, capsnap, follows);
3188 list_del(&capsnap->ci_item);
Yan, Zhengc8799fc2016-07-07 15:22:38 +08003189 if (__finish_cap_flush(NULL, ci, &capsnap->cap_flush))
3190 wake_ci = true;
Yan, Zheng0e294382016-07-04 18:06:41 +08003191
3192 spin_lock(&mdsc->cap_dirty_lock);
3193
3194 if (list_empty(&ci->i_cap_flush_list))
3195 list_del_init(&ci->i_flushing_item);
3196
Yan, Zhengc8799fc2016-07-07 15:22:38 +08003197 if (__finish_cap_flush(mdsc, NULL, &capsnap->cap_flush))
3198 wake_mdsc = true;
Yan, Zheng0e294382016-07-04 18:06:41 +08003199
3200 spin_unlock(&mdsc->cap_dirty_lock);
Yan, Zheng0e294382016-07-04 18:06:41 +08003201 }
Sage Weilbe655592011-11-30 09:47:09 -08003202 spin_unlock(&ci->i_ceph_lock);
Yan, Zheng0e294382016-07-04 18:06:41 +08003203 if (flushed) {
3204 ceph_put_snap_context(capsnap->context);
3205 ceph_put_cap_snap(capsnap);
Yan, Zhengc8799fc2016-07-07 15:22:38 +08003206 if (wake_ci)
3207 wake_up_all(&ci->i_cap_wq);
3208 if (wake_mdsc)
3209 wake_up_all(&mdsc->cap_flushing_wq);
Sage Weila8599bd2009-10-06 11:31:12 -07003210 iput(inode);
Yan, Zheng0e294382016-07-04 18:06:41 +08003211 }
Sage Weila8599bd2009-10-06 11:31:12 -07003212}
3213
3214/*
3215 * Handle TRUNC from MDS, indicating file truncation.
3216 *
3217 * caller hold s_mutex.
3218 */
3219static void handle_cap_trunc(struct inode *inode,
3220 struct ceph_mds_caps *trunc,
3221 struct ceph_mds_session *session)
Sage Weilbe655592011-11-30 09:47:09 -08003222 __releases(ci->i_ceph_lock)
Sage Weila8599bd2009-10-06 11:31:12 -07003223{
3224 struct ceph_inode_info *ci = ceph_inode(inode);
3225 int mds = session->s_mds;
3226 int seq = le32_to_cpu(trunc->seq);
3227 u32 truncate_seq = le32_to_cpu(trunc->truncate_seq);
3228 u64 truncate_size = le64_to_cpu(trunc->truncate_size);
3229 u64 size = le64_to_cpu(trunc->size);
3230 int implemented = 0;
3231 int dirty = __ceph_caps_dirty(ci);
3232 int issued = __ceph_caps_issued(ceph_inode(inode), &implemented);
3233 int queue_trunc = 0;
3234
3235 issued |= implemented | dirty;
3236
3237 dout("handle_cap_trunc inode %p mds%d seq %d to %lld seq %d\n",
3238 inode, mds, seq, truncate_size, truncate_seq);
3239 queue_trunc = ceph_fill_file_size(inode, issued,
3240 truncate_seq, truncate_size, size);
Sage Weilbe655592011-11-30 09:47:09 -08003241 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07003242
Yan, Zheng14649752016-05-20 15:41:20 +08003243 if (queue_trunc)
Sage Weil3c6f6b72010-02-09 15:24:44 -08003244 ceph_queue_vmtruncate(inode);
Sage Weila8599bd2009-10-06 11:31:12 -07003245}
3246
3247/*
3248 * Handle EXPORT from MDS. Cap is being migrated _from_ this mds to a
3249 * different one. If we are the most recent migration we've seen (as
3250 * indicated by mseq), make note of the migrating cap bits for the
3251 * duration (until we see the corresponding IMPORT).
3252 *
3253 * caller holds s_mutex
3254 */
3255static void handle_cap_export(struct inode *inode, struct ceph_mds_caps *ex,
Yan, Zheng11df2df2013-11-24 14:44:38 +08003256 struct ceph_mds_cap_peer *ph,
3257 struct ceph_mds_session *session)
Sage Weila8599bd2009-10-06 11:31:12 -07003258{
Sage Weildb354052011-05-24 11:46:31 -07003259 struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
Yan, Zheng11df2df2013-11-24 14:44:38 +08003260 struct ceph_mds_session *tsession = NULL;
Yan, Zhengd9df2782014-04-18 09:57:11 +08003261 struct ceph_cap *cap, *tcap, *new_cap = NULL;
Sage Weila8599bd2009-10-06 11:31:12 -07003262 struct ceph_inode_info *ci = ceph_inode(inode);
Yan, Zheng11df2df2013-11-24 14:44:38 +08003263 u64 t_cap_id;
Sage Weila8599bd2009-10-06 11:31:12 -07003264 unsigned mseq = le32_to_cpu(ex->migrate_seq);
Yan, Zheng11df2df2013-11-24 14:44:38 +08003265 unsigned t_seq, t_mseq;
3266 int target, issued;
3267 int mds = session->s_mds;
Sage Weila8599bd2009-10-06 11:31:12 -07003268
Yan, Zheng11df2df2013-11-24 14:44:38 +08003269 if (ph) {
3270 t_cap_id = le64_to_cpu(ph->cap_id);
3271 t_seq = le32_to_cpu(ph->seq);
3272 t_mseq = le32_to_cpu(ph->mseq);
3273 target = le32_to_cpu(ph->mds);
3274 } else {
3275 t_cap_id = t_seq = t_mseq = 0;
3276 target = -1;
Sage Weila8599bd2009-10-06 11:31:12 -07003277 }
3278
Yan, Zheng11df2df2013-11-24 14:44:38 +08003279 dout("handle_cap_export inode %p ci %p mds%d mseq %d target %d\n",
3280 inode, ci, mds, mseq, target);
3281retry:
3282 spin_lock(&ci->i_ceph_lock);
3283 cap = __get_cap_for_mds(ci, mds);
Yan, Zhengca665e02014-04-21 15:46:37 +08003284 if (!cap || cap->cap_id != le64_to_cpu(ex->cap_id))
Yan, Zheng11df2df2013-11-24 14:44:38 +08003285 goto out_unlock;
Sage Weil154f42c2010-06-21 13:45:04 -07003286
Yan, Zheng11df2df2013-11-24 14:44:38 +08003287 if (target < 0) {
3288 __ceph_remove_cap(cap, false);
Yan, Zheng77310322016-04-08 15:27:16 +08003289 if (!ci->i_auth_cap)
3290 ci->i_ceph_flags |= CEPH_I_CAP_DROPPED;
Yan, Zheng11df2df2013-11-24 14:44:38 +08003291 goto out_unlock;
3292 }
Sage Weildb354052011-05-24 11:46:31 -07003293
Yan, Zheng11df2df2013-11-24 14:44:38 +08003294 /*
3295 * now we know we haven't received the cap import message yet
3296 * because the exported cap still exist.
3297 */
3298
3299 issued = cap->issued;
3300 WARN_ON(issued != cap->implemented);
3301
3302 tcap = __get_cap_for_mds(ci, target);
3303 if (tcap) {
3304 /* already have caps from the target */
3305 if (tcap->cap_id != t_cap_id ||
3306 ceph_seq_cmp(tcap->seq, t_seq) < 0) {
3307 dout(" updating import cap %p mds%d\n", tcap, target);
3308 tcap->cap_id = t_cap_id;
3309 tcap->seq = t_seq - 1;
3310 tcap->issue_seq = t_seq - 1;
3311 tcap->mseq = t_mseq;
3312 tcap->issued |= issued;
3313 tcap->implemented |= issued;
3314 if (cap == ci->i_auth_cap)
3315 ci->i_auth_cap = tcap;
Yan, Zheng0e294382016-07-04 18:06:41 +08003316 if (!list_empty(&ci->i_cap_flush_list) &&
3317 ci->i_auth_cap == tcap) {
Yan, Zheng11df2df2013-11-24 14:44:38 +08003318 spin_lock(&mdsc->cap_dirty_lock);
3319 list_move_tail(&ci->i_flushing_item,
3320 &tcap->session->s_cap_flushing);
3321 spin_unlock(&mdsc->cap_dirty_lock);
Sage Weildb354052011-05-24 11:46:31 -07003322 }
Sage Weila8599bd2009-10-06 11:31:12 -07003323 }
Yan, Zhenga096b092013-09-22 10:15:58 +08003324 __ceph_remove_cap(cap, false);
Yan, Zheng11df2df2013-11-24 14:44:38 +08003325 goto out_unlock;
Yan, Zhengd9df2782014-04-18 09:57:11 +08003326 } else if (tsession) {
Yan, Zheng11df2df2013-11-24 14:44:38 +08003327 /* add placeholder for the export tagert */
Yan, Zhengd9df2782014-04-18 09:57:11 +08003328 int flag = (cap == ci->i_auth_cap) ? CEPH_CAP_FLAG_AUTH : 0;
Yan, Zheng11df2df2013-11-24 14:44:38 +08003329 ceph_add_cap(inode, tsession, t_cap_id, -1, issued, 0,
Yan, Zhengd9df2782014-04-18 09:57:11 +08003330 t_seq - 1, t_mseq, (u64)-1, flag, &new_cap);
3331
3332 __ceph_remove_cap(cap, false);
3333 goto out_unlock;
Yan, Zheng11df2df2013-11-24 14:44:38 +08003334 }
Sage Weila8599bd2009-10-06 11:31:12 -07003335
Sage Weilbe655592011-11-30 09:47:09 -08003336 spin_unlock(&ci->i_ceph_lock);
Yan, Zheng11df2df2013-11-24 14:44:38 +08003337 mutex_unlock(&session->s_mutex);
3338
3339 /* open target session */
3340 tsession = ceph_mdsc_open_export_target_session(mdsc, target);
3341 if (!IS_ERR(tsession)) {
3342 if (mds > target) {
3343 mutex_lock(&session->s_mutex);
3344 mutex_lock_nested(&tsession->s_mutex,
3345 SINGLE_DEPTH_NESTING);
3346 } else {
3347 mutex_lock(&tsession->s_mutex);
3348 mutex_lock_nested(&session->s_mutex,
3349 SINGLE_DEPTH_NESTING);
3350 }
Yan, Zhengd9df2782014-04-18 09:57:11 +08003351 new_cap = ceph_get_cap(mdsc, NULL);
Yan, Zheng11df2df2013-11-24 14:44:38 +08003352 } else {
3353 WARN_ON(1);
3354 tsession = NULL;
3355 target = -1;
3356 }
3357 goto retry;
3358
3359out_unlock:
3360 spin_unlock(&ci->i_ceph_lock);
3361 mutex_unlock(&session->s_mutex);
3362 if (tsession) {
3363 mutex_unlock(&tsession->s_mutex);
3364 ceph_put_mds_session(tsession);
3365 }
Yan, Zhengd9df2782014-04-18 09:57:11 +08003366 if (new_cap)
3367 ceph_put_cap(mdsc, new_cap);
Sage Weila8599bd2009-10-06 11:31:12 -07003368}
3369
3370/*
Yan, Zheng2cd698b2014-04-18 13:20:27 +08003371 * Handle cap IMPORT.
Sage Weila8599bd2009-10-06 11:31:12 -07003372 *
Yan, Zheng2cd698b2014-04-18 13:20:27 +08003373 * caller holds s_mutex. acquires i_ceph_lock
Sage Weila8599bd2009-10-06 11:31:12 -07003374 */
3375static void handle_cap_import(struct ceph_mds_client *mdsc,
3376 struct inode *inode, struct ceph_mds_caps *im,
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003377 struct ceph_mds_cap_peer *ph,
Sage Weila8599bd2009-10-06 11:31:12 -07003378 struct ceph_mds_session *session,
Yan, Zheng2cd698b2014-04-18 13:20:27 +08003379 struct ceph_cap **target_cap, int *old_issued)
3380 __acquires(ci->i_ceph_lock)
Sage Weila8599bd2009-10-06 11:31:12 -07003381{
3382 struct ceph_inode_info *ci = ceph_inode(inode);
Yan, Zheng2cd698b2014-04-18 13:20:27 +08003383 struct ceph_cap *cap, *ocap, *new_cap = NULL;
Sage Weila8599bd2009-10-06 11:31:12 -07003384 int mds = session->s_mds;
Yan, Zheng2cd698b2014-04-18 13:20:27 +08003385 int issued;
3386 unsigned caps = le32_to_cpu(im->caps);
Sage Weila8599bd2009-10-06 11:31:12 -07003387 unsigned wanted = le32_to_cpu(im->wanted);
3388 unsigned seq = le32_to_cpu(im->seq);
3389 unsigned mseq = le32_to_cpu(im->migrate_seq);
3390 u64 realmino = le64_to_cpu(im->realm);
3391 u64 cap_id = le64_to_cpu(im->cap_id);
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003392 u64 p_cap_id;
3393 int peer;
Sage Weila8599bd2009-10-06 11:31:12 -07003394
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003395 if (ph) {
3396 p_cap_id = le64_to_cpu(ph->cap_id);
3397 peer = le32_to_cpu(ph->mds);
Sage Weila8599bd2009-10-06 11:31:12 -07003398 } else {
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003399 p_cap_id = 0;
3400 peer = -1;
Sage Weila8599bd2009-10-06 11:31:12 -07003401 }
3402
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003403 dout("handle_cap_import inode %p ci %p mds%d mseq %d peer %d\n",
3404 inode, ci, mds, mseq, peer);
3405
Yan, Zhengd9df2782014-04-18 09:57:11 +08003406retry:
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003407 spin_lock(&ci->i_ceph_lock);
Yan, Zhengd9df2782014-04-18 09:57:11 +08003408 cap = __get_cap_for_mds(ci, mds);
3409 if (!cap) {
3410 if (!new_cap) {
3411 spin_unlock(&ci->i_ceph_lock);
3412 new_cap = ceph_get_cap(mdsc, NULL);
3413 goto retry;
3414 }
Yan, Zheng2cd698b2014-04-18 13:20:27 +08003415 cap = new_cap;
3416 } else {
3417 if (new_cap) {
3418 ceph_put_cap(mdsc, new_cap);
3419 new_cap = NULL;
3420 }
Yan, Zhengd9df2782014-04-18 09:57:11 +08003421 }
3422
Yan, Zheng2cd698b2014-04-18 13:20:27 +08003423 __ceph_caps_issued(ci, &issued);
3424 issued |= __ceph_caps_dirty(ci);
3425
3426 ceph_add_cap(inode, session, cap_id, -1, caps, wanted, seq, mseq,
Yan, Zhengd9df2782014-04-18 09:57:11 +08003427 realmino, CEPH_CAP_FLAG_AUTH, &new_cap);
3428
Yan, Zheng2cd698b2014-04-18 13:20:27 +08003429 ocap = peer >= 0 ? __get_cap_for_mds(ci, peer) : NULL;
3430 if (ocap && ocap->cap_id == p_cap_id) {
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003431 dout(" remove export cap %p mds%d flags %d\n",
Yan, Zheng2cd698b2014-04-18 13:20:27 +08003432 ocap, peer, ph->flags);
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003433 if ((ph->flags & CEPH_CAP_FLAG_AUTH) &&
Yan, Zheng2cd698b2014-04-18 13:20:27 +08003434 (ocap->seq != le32_to_cpu(ph->seq) ||
3435 ocap->mseq != le32_to_cpu(ph->mseq))) {
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003436 pr_err("handle_cap_import: mismatched seq/mseq: "
3437 "ino (%llx.%llx) mds%d seq %d mseq %d "
3438 "importer mds%d has peer seq %d mseq %d\n",
Yan, Zheng2cd698b2014-04-18 13:20:27 +08003439 ceph_vinop(inode), peer, ocap->seq,
3440 ocap->mseq, mds, le32_to_cpu(ph->seq),
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003441 le32_to_cpu(ph->mseq));
3442 }
Yan, Zheng2cd698b2014-04-18 13:20:27 +08003443 __ceph_remove_cap(ocap, (ph->flags & CEPH_CAP_FLAG_RELEASE));
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003444 }
3445
3446 /* make sure we re-request max_size, if necessary */
3447 ci->i_wanted_max_size = 0;
3448 ci->i_requested_max_size = 0;
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003449
Yan, Zheng2cd698b2014-04-18 13:20:27 +08003450 *old_issued = issued;
3451 *target_cap = cap;
Sage Weila8599bd2009-10-06 11:31:12 -07003452}
3453
3454/*
3455 * Handle a caps message from the MDS.
3456 *
3457 * Identify the appropriate session, inode, and call the right handler
3458 * based on the cap op.
3459 */
3460void ceph_handle_caps(struct ceph_mds_session *session,
3461 struct ceph_msg *msg)
3462{
3463 struct ceph_mds_client *mdsc = session->s_mdsc;
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07003464 struct super_block *sb = mdsc->fsc->sb;
Sage Weila8599bd2009-10-06 11:31:12 -07003465 struct inode *inode;
Sage Weilbe655592011-11-30 09:47:09 -08003466 struct ceph_inode_info *ci;
Sage Weila8599bd2009-10-06 11:31:12 -07003467 struct ceph_cap *cap;
3468 struct ceph_mds_caps *h;
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003469 struct ceph_mds_cap_peer *peer = NULL;
Yan, Zheng779fe0f2016-03-07 09:35:06 +08003470 struct ceph_snap_realm *realm = NULL;
3471 struct ceph_string *pool_ns = NULL;
Sage Weil2600d2d2010-02-22 15:12:16 -08003472 int mds = session->s_mds;
Yan, Zheng2cd698b2014-04-18 13:20:27 +08003473 int op, issued;
Sage Weil3d7ded42010-06-09 16:47:10 -07003474 u32 seq, mseq;
Sage Weila8599bd2009-10-06 11:31:12 -07003475 struct ceph_vino vino;
Sage Weil6df058c2009-12-22 11:24:33 -08003476 u64 tid;
Yan, Zhengfb01d1f2014-11-14 21:29:55 +08003477 u64 inline_version = 0;
3478 void *inline_data = NULL;
3479 u32 inline_len = 0;
Sage Weil70edb552010-03-01 13:20:50 -08003480 void *snaptrace;
Sage Weilce1fbc82010-08-02 15:09:39 -07003481 size_t snaptrace_len;
Yan, Zhengfb01d1f2014-11-14 21:29:55 +08003482 void *p, *end;
Sage Weila8599bd2009-10-06 11:31:12 -07003483
3484 dout("handle_caps from mds%d\n", mds);
3485
3486 /* decode */
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003487 end = msg->front.iov_base + msg->front.iov_len;
Sage Weil6df058c2009-12-22 11:24:33 -08003488 tid = le64_to_cpu(msg->hdr.tid);
Sage Weila8599bd2009-10-06 11:31:12 -07003489 if (msg->front.iov_len < sizeof(*h))
3490 goto bad;
3491 h = msg->front.iov_base;
3492 op = le32_to_cpu(h->op);
3493 vino.ino = le64_to_cpu(h->ino);
3494 vino.snap = CEPH_NOSNAP;
Sage Weila8599bd2009-10-06 11:31:12 -07003495 seq = le32_to_cpu(h->seq);
Sage Weil3d7ded42010-06-09 16:47:10 -07003496 mseq = le32_to_cpu(h->migrate_seq);
Sage Weila8599bd2009-10-06 11:31:12 -07003497
Sage Weilce1fbc82010-08-02 15:09:39 -07003498 snaptrace = h + 1;
3499 snaptrace_len = le32_to_cpu(h->snap_trace_len);
Yan, Zhengfb01d1f2014-11-14 21:29:55 +08003500 p = snaptrace + snaptrace_len;
Sage Weilce1fbc82010-08-02 15:09:39 -07003501
3502 if (le16_to_cpu(msg->hdr.version) >= 2) {
Yan, Zhengfb01d1f2014-11-14 21:29:55 +08003503 u32 flock_len;
Sage Weilce1fbc82010-08-02 15:09:39 -07003504 ceph_decode_32_safe(&p, end, flock_len, bad);
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003505 if (p + flock_len > end)
3506 goto bad;
Yan, Zhengfb01d1f2014-11-14 21:29:55 +08003507 p += flock_len;
Sage Weilce1fbc82010-08-02 15:09:39 -07003508 }
3509
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003510 if (le16_to_cpu(msg->hdr.version) >= 3) {
3511 if (op == CEPH_CAP_OP_IMPORT) {
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003512 if (p + sizeof(*peer) > end)
3513 goto bad;
3514 peer = p;
Yan, Zhengfb01d1f2014-11-14 21:29:55 +08003515 p += sizeof(*peer);
Yan, Zheng11df2df2013-11-24 14:44:38 +08003516 } else if (op == CEPH_CAP_OP_EXPORT) {
3517 /* recorded in unused fields */
3518 peer = (void *)&h->size;
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003519 }
3520 }
3521
Yan, Zhengfb01d1f2014-11-14 21:29:55 +08003522 if (le16_to_cpu(msg->hdr.version) >= 4) {
3523 ceph_decode_64_safe(&p, end, inline_version, bad);
3524 ceph_decode_32_safe(&p, end, inline_len, bad);
3525 if (p + inline_len > end)
3526 goto bad;
3527 inline_data = p;
3528 p += inline_len;
3529 }
3530
Yan, Zheng5ea5c5e2016-02-14 18:06:41 +08003531 if (le16_to_cpu(msg->hdr.version) >= 8) {
3532 u64 flush_tid;
3533 u32 caller_uid, caller_gid;
3534 u32 osd_epoch_barrier;
Yan, Zheng779fe0f2016-03-07 09:35:06 +08003535 u32 pool_ns_len;
Yan, Zheng5ea5c5e2016-02-14 18:06:41 +08003536 /* version >= 5 */
3537 ceph_decode_32_safe(&p, end, osd_epoch_barrier, bad);
3538 /* version >= 6 */
3539 ceph_decode_64_safe(&p, end, flush_tid, bad);
3540 /* version >= 7 */
3541 ceph_decode_32_safe(&p, end, caller_uid, bad);
3542 ceph_decode_32_safe(&p, end, caller_gid, bad);
3543 /* version >= 8 */
3544 ceph_decode_32_safe(&p, end, pool_ns_len, bad);
Yan, Zheng779fe0f2016-03-07 09:35:06 +08003545 if (pool_ns_len > 0) {
3546 ceph_decode_need(&p, end, pool_ns_len, bad);
3547 pool_ns = ceph_find_or_create_string(p, pool_ns_len);
3548 p += pool_ns_len;
3549 }
Yan, Zheng5ea5c5e2016-02-14 18:06:41 +08003550 }
3551
Yan, Zheng6cd3bca2014-09-17 07:45:12 +08003552 /* lookup ino */
3553 inode = ceph_find_inode(sb, vino);
3554 ci = ceph_inode(inode);
3555 dout(" op %s ino %llx.%llx inode %p\n", ceph_cap_op_name(op), vino.ino,
3556 vino.snap, inode);
3557
Sage Weila8599bd2009-10-06 11:31:12 -07003558 mutex_lock(&session->s_mutex);
3559 session->s_seq++;
3560 dout(" mds%d seq %lld cap seq %u\n", session->s_mds, session->s_seq,
3561 (unsigned)seq);
3562
Sage Weila8599bd2009-10-06 11:31:12 -07003563 if (!inode) {
3564 dout(" i don't have ino %llx\n", vino.ino);
Sage Weil3d7ded42010-06-09 16:47:10 -07003565
Yan, Zhenga096b092013-09-22 10:15:58 +08003566 if (op == CEPH_CAP_OP_IMPORT) {
Yan, Zheng745a8e32015-05-14 17:22:42 +08003567 cap = ceph_get_cap(mdsc, NULL);
3568 cap->cap_ino = vino.ino;
3569 cap->queue_release = 1;
Yan, Zheng779fe0f2016-03-07 09:35:06 +08003570 cap->cap_id = le64_to_cpu(h->cap_id);
Yan, Zheng745a8e32015-05-14 17:22:42 +08003571 cap->mseq = mseq;
3572 cap->seq = seq;
Yan, Zhenga096b092013-09-22 10:15:58 +08003573 spin_lock(&session->s_cap_lock);
Yan, Zheng745a8e32015-05-14 17:22:42 +08003574 list_add_tail(&cap->session_caps,
3575 &session->s_cap_releases);
3576 session->s_num_cap_releases++;
Yan, Zhenga096b092013-09-22 10:15:58 +08003577 spin_unlock(&session->s_cap_lock);
3578 }
Greg Farnum21b559d2010-10-06 15:46:30 -07003579 goto flush_cap_releases;
Sage Weila8599bd2009-10-06 11:31:12 -07003580 }
3581
3582 /* these will work even if we don't have a cap yet */
3583 switch (op) {
3584 case CEPH_CAP_OP_FLUSHSNAP_ACK:
Sage Weil6df058c2009-12-22 11:24:33 -08003585 handle_cap_flushsnap_ack(inode, tid, h, session);
Sage Weila8599bd2009-10-06 11:31:12 -07003586 goto done;
3587
3588 case CEPH_CAP_OP_EXPORT:
Yan, Zheng11df2df2013-11-24 14:44:38 +08003589 handle_cap_export(inode, h, peer, session);
3590 goto done_unlocked;
Sage Weila8599bd2009-10-06 11:31:12 -07003591
3592 case CEPH_CAP_OP_IMPORT:
Yan, Zheng982d6012014-12-23 15:30:54 +08003593 realm = NULL;
3594 if (snaptrace_len) {
3595 down_write(&mdsc->snap_rwsem);
3596 ceph_update_snap_trace(mdsc, snaptrace,
3597 snaptrace + snaptrace_len,
3598 false, &realm);
3599 downgrade_write(&mdsc->snap_rwsem);
3600 } else {
3601 down_read(&mdsc->snap_rwsem);
3602 }
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003603 handle_cap_import(mdsc, inode, h, peer, session,
Yan, Zheng2cd698b2014-04-18 13:20:27 +08003604 &cap, &issued);
Yan, Zheng779fe0f2016-03-07 09:35:06 +08003605 handle_cap_grant(mdsc, inode, h, &pool_ns,
Yan, Zhengfb01d1f2014-11-14 21:29:55 +08003606 inline_version, inline_data, inline_len,
Yan, Zheng779fe0f2016-03-07 09:35:06 +08003607 msg->middle, session, cap, issued);
Yan, Zheng982d6012014-12-23 15:30:54 +08003608 if (realm)
3609 ceph_put_snap_realm(mdsc, realm);
Yan, Zheng2cd698b2014-04-18 13:20:27 +08003610 goto done_unlocked;
Sage Weila8599bd2009-10-06 11:31:12 -07003611 }
3612
3613 /* the rest require a cap */
Sage Weilbe655592011-11-30 09:47:09 -08003614 spin_lock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07003615 cap = __get_cap_for_mds(ceph_inode(inode), mds);
3616 if (!cap) {
Sage Weil9dbd4122010-06-10 13:21:20 -07003617 dout(" no cap on %p ino %llx.%llx from mds%d\n",
Sage Weila8599bd2009-10-06 11:31:12 -07003618 inode, ceph_ino(inode), ceph_snap(inode), mds);
Sage Weilbe655592011-11-30 09:47:09 -08003619 spin_unlock(&ci->i_ceph_lock);
Greg Farnum21b559d2010-10-06 15:46:30 -07003620 goto flush_cap_releases;
Sage Weila8599bd2009-10-06 11:31:12 -07003621 }
3622
Sage Weilbe655592011-11-30 09:47:09 -08003623 /* note that each of these drops i_ceph_lock for us */
Sage Weila8599bd2009-10-06 11:31:12 -07003624 switch (op) {
3625 case CEPH_CAP_OP_REVOKE:
3626 case CEPH_CAP_OP_GRANT:
Yan, Zheng2cd698b2014-04-18 13:20:27 +08003627 __ceph_caps_issued(ci, &issued);
3628 issued |= __ceph_caps_dirty(ci);
Yan, Zheng779fe0f2016-03-07 09:35:06 +08003629 handle_cap_grant(mdsc, inode, h, &pool_ns,
Yan, Zhengfb01d1f2014-11-14 21:29:55 +08003630 inline_version, inline_data, inline_len,
Yan, Zheng779fe0f2016-03-07 09:35:06 +08003631 msg->middle, session, cap, issued);
Sage Weil15637c82010-03-16 13:42:00 -07003632 goto done_unlocked;
Sage Weila8599bd2009-10-06 11:31:12 -07003633
3634 case CEPH_CAP_OP_FLUSH_ACK:
Sage Weil6df058c2009-12-22 11:24:33 -08003635 handle_cap_flush_ack(inode, tid, h, session, cap);
Sage Weila8599bd2009-10-06 11:31:12 -07003636 break;
3637
3638 case CEPH_CAP_OP_TRUNC:
3639 handle_cap_trunc(inode, h, session);
3640 break;
3641
3642 default:
Sage Weilbe655592011-11-30 09:47:09 -08003643 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07003644 pr_err("ceph_handle_caps: unknown cap op %d %s\n", op,
3645 ceph_cap_op_name(op));
3646 }
3647
Greg Farnum21b559d2010-10-06 15:46:30 -07003648 goto done;
3649
3650flush_cap_releases:
3651 /*
Yan, Zheng745a8e32015-05-14 17:22:42 +08003652 * send any cap release message to try to move things
Greg Farnum21b559d2010-10-06 15:46:30 -07003653 * along for the mds (who clearly thinks we still have this
3654 * cap).
3655 */
Greg Farnum21b559d2010-10-06 15:46:30 -07003656 ceph_send_cap_releases(mdsc, session);
3657
Sage Weila8599bd2009-10-06 11:31:12 -07003658done:
Sage Weil15637c82010-03-16 13:42:00 -07003659 mutex_unlock(&session->s_mutex);
3660done_unlocked:
SF Markus Elfringe96a6502014-11-02 15:20:59 +01003661 iput(inode);
Yan, Zheng779fe0f2016-03-07 09:35:06 +08003662 ceph_put_string(pool_ns);
Sage Weila8599bd2009-10-06 11:31:12 -07003663 return;
3664
3665bad:
3666 pr_err("ceph_handle_caps: corrupt message\n");
Sage Weil9ec7cab2009-12-14 15:13:47 -08003667 ceph_msg_dump(msg);
Sage Weila8599bd2009-10-06 11:31:12 -07003668 return;
3669}
3670
3671/*
3672 * Delayed work handler to process end of delayed cap release LRU list.
3673 */
Sage Weilafcdaea2009-10-14 14:27:38 -07003674void ceph_check_delayed_caps(struct ceph_mds_client *mdsc)
Sage Weila8599bd2009-10-06 11:31:12 -07003675{
3676 struct ceph_inode_info *ci;
3677 int flags = CHECK_CAPS_NODELAY;
3678
Sage Weila8599bd2009-10-06 11:31:12 -07003679 dout("check_delayed_caps\n");
3680 while (1) {
3681 spin_lock(&mdsc->cap_delay_lock);
3682 if (list_empty(&mdsc->cap_delay_list))
3683 break;
3684 ci = list_first_entry(&mdsc->cap_delay_list,
3685 struct ceph_inode_info,
3686 i_cap_delay_list);
3687 if ((ci->i_ceph_flags & CEPH_I_FLUSH) == 0 &&
3688 time_before(jiffies, ci->i_hold_caps_max))
3689 break;
3690 list_del_init(&ci->i_cap_delay_list);
3691 spin_unlock(&mdsc->cap_delay_lock);
3692 dout("check_delayed_caps on %p\n", &ci->vfs_inode);
3693 ceph_check_caps(ci, flags, NULL);
3694 }
3695 spin_unlock(&mdsc->cap_delay_lock);
3696}
3697
3698/*
Sage Weilafcdaea2009-10-14 14:27:38 -07003699 * Flush all dirty caps to the mds
3700 */
3701void ceph_flush_dirty_caps(struct ceph_mds_client *mdsc)
3702{
Sage Weildb354052011-05-24 11:46:31 -07003703 struct ceph_inode_info *ci;
3704 struct inode *inode;
Sage Weilafcdaea2009-10-14 14:27:38 -07003705
3706 dout("flush_dirty_caps\n");
3707 spin_lock(&mdsc->cap_dirty_lock);
Sage Weildb354052011-05-24 11:46:31 -07003708 while (!list_empty(&mdsc->cap_dirty)) {
3709 ci = list_first_entry(&mdsc->cap_dirty, struct ceph_inode_info,
3710 i_dirty_item);
Sage Weil70b666c2011-05-27 09:24:26 -07003711 inode = &ci->vfs_inode;
3712 ihold(inode);
Sage Weildb354052011-05-24 11:46:31 -07003713 dout("flush_dirty_caps %p\n", inode);
Sage Weilafcdaea2009-10-14 14:27:38 -07003714 spin_unlock(&mdsc->cap_dirty_lock);
Sage Weil70b666c2011-05-27 09:24:26 -07003715 ceph_check_caps(ci, CHECK_CAPS_NODELAY|CHECK_CAPS_FLUSH, NULL);
3716 iput(inode);
Sage Weilafcdaea2009-10-14 14:27:38 -07003717 spin_lock(&mdsc->cap_dirty_lock);
3718 }
3719 spin_unlock(&mdsc->cap_dirty_lock);
Sage Weildb354052011-05-24 11:46:31 -07003720 dout("flush_dirty_caps done\n");
Sage Weilafcdaea2009-10-14 14:27:38 -07003721}
3722
Yan, Zheng774a6a12016-06-06 16:01:39 +08003723void __ceph_get_fmode(struct ceph_inode_info *ci, int fmode)
3724{
3725 int i;
3726 int bits = (fmode << 1) | 1;
3727 for (i = 0; i < CEPH_FILE_MODE_BITS; i++) {
3728 if (bits & (1 << i))
3729 ci->i_nr_by_mode[i]++;
3730 }
3731}
3732
Sage Weilafcdaea2009-10-14 14:27:38 -07003733/*
Sage Weila8599bd2009-10-06 11:31:12 -07003734 * Drop open file reference. If we were the last open file,
3735 * we may need to release capabilities to the MDS (or schedule
3736 * their delayed release).
3737 */
3738void ceph_put_fmode(struct ceph_inode_info *ci, int fmode)
3739{
Yan, Zheng774a6a12016-06-06 16:01:39 +08003740 int i, last = 0;
3741 int bits = (fmode << 1) | 1;
Sage Weilbe655592011-11-30 09:47:09 -08003742 spin_lock(&ci->i_ceph_lock);
Yan, Zheng774a6a12016-06-06 16:01:39 +08003743 for (i = 0; i < CEPH_FILE_MODE_BITS; i++) {
3744 if (bits & (1 << i)) {
3745 BUG_ON(ci->i_nr_by_mode[i] == 0);
3746 if (--ci->i_nr_by_mode[i] == 0)
3747 last++;
3748 }
3749 }
3750 dout("put_fmode %p fmode %d {%d,%d,%d,%d}\n",
3751 &ci->vfs_inode, fmode,
3752 ci->i_nr_by_mode[0], ci->i_nr_by_mode[1],
3753 ci->i_nr_by_mode[2], ci->i_nr_by_mode[3]);
Sage Weilbe655592011-11-30 09:47:09 -08003754 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07003755
3756 if (last && ci->i_vino.snap == CEPH_NOSNAP)
3757 ceph_check_caps(ci, 0, NULL);
3758}
3759
3760/*
3761 * Helpers for embedding cap and dentry lease releases into mds
3762 * requests.
3763 *
3764 * @force is used by dentry_release (below) to force inclusion of a
3765 * record for the directory inode, even when there aren't any caps to
3766 * drop.
3767 */
3768int ceph_encode_inode_release(void **p, struct inode *inode,
3769 int mds, int drop, int unless, int force)
3770{
3771 struct ceph_inode_info *ci = ceph_inode(inode);
3772 struct ceph_cap *cap;
3773 struct ceph_mds_request_release *rel = *p;
Sage Weilec97f882010-06-24 15:12:37 -07003774 int used, dirty;
Sage Weila8599bd2009-10-06 11:31:12 -07003775 int ret = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07003776
Sage Weilbe655592011-11-30 09:47:09 -08003777 spin_lock(&ci->i_ceph_lock);
Sage Weil916623d2010-03-16 15:01:07 -07003778 used = __ceph_caps_used(ci);
Sage Weilec97f882010-06-24 15:12:37 -07003779 dirty = __ceph_caps_dirty(ci);
Sage Weil916623d2010-03-16 15:01:07 -07003780
Sage Weilec97f882010-06-24 15:12:37 -07003781 dout("encode_inode_release %p mds%d used|dirty %s drop %s unless %s\n",
3782 inode, mds, ceph_cap_string(used|dirty), ceph_cap_string(drop),
Sage Weil916623d2010-03-16 15:01:07 -07003783 ceph_cap_string(unless));
3784
Sage Weilec97f882010-06-24 15:12:37 -07003785 /* only drop unused, clean caps */
3786 drop &= ~(used | dirty);
Sage Weil916623d2010-03-16 15:01:07 -07003787
Sage Weila8599bd2009-10-06 11:31:12 -07003788 cap = __get_cap_for_mds(ci, mds);
3789 if (cap && __cap_is_valid(cap)) {
3790 if (force ||
3791 ((cap->issued & drop) &&
3792 (cap->issued & unless) == 0)) {
3793 if ((cap->issued & drop) &&
3794 (cap->issued & unless) == 0) {
Yan, Zhengbb137f82013-06-03 18:22:17 +08003795 int wanted = __ceph_caps_wanted(ci);
3796 if ((ci->i_ceph_flags & CEPH_I_NODELAY) == 0)
3797 wanted |= cap->mds_wanted;
3798 dout("encode_inode_release %p cap %p "
3799 "%s -> %s, wanted %s -> %s\n", inode, cap,
Sage Weila8599bd2009-10-06 11:31:12 -07003800 ceph_cap_string(cap->issued),
Yan, Zhengbb137f82013-06-03 18:22:17 +08003801 ceph_cap_string(cap->issued & ~drop),
3802 ceph_cap_string(cap->mds_wanted),
3803 ceph_cap_string(wanted));
3804
Sage Weila8599bd2009-10-06 11:31:12 -07003805 cap->issued &= ~drop;
3806 cap->implemented &= ~drop;
Yan, Zhengbb137f82013-06-03 18:22:17 +08003807 cap->mds_wanted = wanted;
Sage Weila8599bd2009-10-06 11:31:12 -07003808 } else {
3809 dout("encode_inode_release %p cap %p %s"
3810 " (force)\n", inode, cap,
3811 ceph_cap_string(cap->issued));
3812 }
3813
3814 rel->ino = cpu_to_le64(ceph_ino(inode));
3815 rel->cap_id = cpu_to_le64(cap->cap_id);
3816 rel->seq = cpu_to_le32(cap->seq);
Himangi Saraogi08a0f242014-07-23 20:11:11 +05303817 rel->issue_seq = cpu_to_le32(cap->issue_seq);
Sage Weila8599bd2009-10-06 11:31:12 -07003818 rel->mseq = cpu_to_le32(cap->mseq);
Yan, Zhengfd7b95c2014-04-17 08:02:02 +08003819 rel->caps = cpu_to_le32(cap->implemented);
Sage Weila8599bd2009-10-06 11:31:12 -07003820 rel->wanted = cpu_to_le32(cap->mds_wanted);
3821 rel->dname_len = 0;
3822 rel->dname_seq = 0;
3823 *p += sizeof(*rel);
3824 ret = 1;
3825 } else {
3826 dout("encode_inode_release %p cap %p %s\n",
3827 inode, cap, ceph_cap_string(cap->issued));
3828 }
3829 }
Sage Weilbe655592011-11-30 09:47:09 -08003830 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07003831 return ret;
3832}
3833
3834int ceph_encode_dentry_release(void **p, struct dentry *dentry,
3835 int mds, int drop, int unless)
3836{
David Howells2b0143b2015-03-17 22:25:59 +00003837 struct inode *dir = d_inode(dentry->d_parent);
Sage Weila8599bd2009-10-06 11:31:12 -07003838 struct ceph_mds_request_release *rel = *p;
3839 struct ceph_dentry_info *di = ceph_dentry(dentry);
3840 int force = 0;
3841 int ret;
3842
3843 /*
3844 * force an record for the directory caps if we have a dentry lease.
Sage Weilbe655592011-11-30 09:47:09 -08003845 * this is racy (can't take i_ceph_lock and d_lock together), but it
Sage Weila8599bd2009-10-06 11:31:12 -07003846 * doesn't have to be perfect; the mds will revoke anything we don't
3847 * release.
3848 */
3849 spin_lock(&dentry->d_lock);
3850 if (di->lease_session && di->lease_session->s_mds == mds)
3851 force = 1;
3852 spin_unlock(&dentry->d_lock);
3853
3854 ret = ceph_encode_inode_release(p, dir, mds, drop, unless, force);
3855
3856 spin_lock(&dentry->d_lock);
3857 if (ret && di->lease_session && di->lease_session->s_mds == mds) {
3858 dout("encode_dentry_release %p mds%d seq %d\n",
3859 dentry, mds, (int)di->lease_seq);
3860 rel->dname_len = cpu_to_le32(dentry->d_name.len);
3861 memcpy(*p, dentry->d_name.name, dentry->d_name.len);
3862 *p += dentry->d_name.len;
3863 rel->dname_seq = cpu_to_le32(di->lease_seq);
Sage Weil1dadcce2010-07-23 13:54:21 -07003864 __ceph_mdsc_drop_dentry_lease(dentry);
Sage Weila8599bd2009-10-06 11:31:12 -07003865 }
3866 spin_unlock(&dentry->d_lock);
3867 return ret;
3868}