blob: 27b566874bc1d2b0494b49ebe1769f380ccf8648 [file] [log] [blame]
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001#include <linux/ceph/ceph_debug.h>
Sage Weila8599bd2009-10-06 11:31:12 -07002
3#include <linux/fs.h>
4#include <linux/kernel.h>
5#include <linux/sched.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +09006#include <linux/slab.h>
Sage Weila8599bd2009-10-06 11:31:12 -07007#include <linux/vmalloc.h>
8#include <linux/wait.h>
Stephen Rothwellf1a3d572010-01-18 11:53:08 +11009#include <linux/writeback.h>
Sage Weila8599bd2009-10-06 11:31:12 -070010
11#include "super.h"
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -070012#include "mds_client.h"
Milosz Tanski99ccbd22013-08-21 17:29:54 -040013#include "cache.h"
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -070014#include <linux/ceph/decode.h>
15#include <linux/ceph/messenger.h>
Sage Weila8599bd2009-10-06 11:31:12 -070016
17/*
18 * Capability management
19 *
20 * The Ceph metadata servers control client access to inode metadata
21 * and file data by issuing capabilities, granting clients permission
22 * to read and/or write both inode field and file data to OSDs
23 * (storage nodes). Each capability consists of a set of bits
24 * indicating which operations are allowed.
25 *
26 * If the client holds a *_SHARED cap, the client has a coherent value
27 * that can be safely read from the cached inode.
28 *
29 * In the case of a *_EXCL (exclusive) or FILE_WR capabilities, the
30 * client is allowed to change inode attributes (e.g., file size,
31 * mtime), note its dirty state in the ceph_cap, and asynchronously
32 * flush that metadata change to the MDS.
33 *
34 * In the event of a conflicting operation (perhaps by another
35 * client), the MDS will revoke the conflicting client capabilities.
36 *
37 * In order for a client to cache an inode, it must hold a capability
38 * with at least one MDS server. When inodes are released, release
39 * notifications are batched and periodically sent en masse to the MDS
40 * cluster to release server state.
41 */
42
43
44/*
45 * Generate readable cap strings for debugging output.
46 */
47#define MAX_CAP_STR 20
48static char cap_str[MAX_CAP_STR][40];
49static DEFINE_SPINLOCK(cap_str_lock);
50static int last_cap_str;
51
52static char *gcap_string(char *s, int c)
53{
54 if (c & CEPH_CAP_GSHARED)
55 *s++ = 's';
56 if (c & CEPH_CAP_GEXCL)
57 *s++ = 'x';
58 if (c & CEPH_CAP_GCACHE)
59 *s++ = 'c';
60 if (c & CEPH_CAP_GRD)
61 *s++ = 'r';
62 if (c & CEPH_CAP_GWR)
63 *s++ = 'w';
64 if (c & CEPH_CAP_GBUFFER)
65 *s++ = 'b';
66 if (c & CEPH_CAP_GLAZYIO)
67 *s++ = 'l';
68 return s;
69}
70
71const char *ceph_cap_string(int caps)
72{
73 int i;
74 char *s;
75 int c;
76
77 spin_lock(&cap_str_lock);
78 i = last_cap_str++;
79 if (last_cap_str == MAX_CAP_STR)
80 last_cap_str = 0;
81 spin_unlock(&cap_str_lock);
82
83 s = cap_str[i];
84
85 if (caps & CEPH_CAP_PIN)
86 *s++ = 'p';
87
88 c = (caps >> CEPH_CAP_SAUTH) & 3;
89 if (c) {
90 *s++ = 'A';
91 s = gcap_string(s, c);
92 }
93
94 c = (caps >> CEPH_CAP_SLINK) & 3;
95 if (c) {
96 *s++ = 'L';
97 s = gcap_string(s, c);
98 }
99
100 c = (caps >> CEPH_CAP_SXATTR) & 3;
101 if (c) {
102 *s++ = 'X';
103 s = gcap_string(s, c);
104 }
105
106 c = caps >> CEPH_CAP_SFILE;
107 if (c) {
108 *s++ = 'F';
109 s = gcap_string(s, c);
110 }
111
112 if (s == cap_str[i])
113 *s++ = '-';
114 *s = 0;
115 return cap_str[i];
116}
117
Yehuda Sadeh37151662010-06-17 16:16:12 -0700118void ceph_caps_init(struct ceph_mds_client *mdsc)
Sage Weila8599bd2009-10-06 11:31:12 -0700119{
Yehuda Sadeh37151662010-06-17 16:16:12 -0700120 INIT_LIST_HEAD(&mdsc->caps_list);
121 spin_lock_init(&mdsc->caps_list_lock);
Sage Weila8599bd2009-10-06 11:31:12 -0700122}
123
Yehuda Sadeh37151662010-06-17 16:16:12 -0700124void ceph_caps_finalize(struct ceph_mds_client *mdsc)
Sage Weila8599bd2009-10-06 11:31:12 -0700125{
126 struct ceph_cap *cap;
127
Yehuda Sadeh37151662010-06-17 16:16:12 -0700128 spin_lock(&mdsc->caps_list_lock);
129 while (!list_empty(&mdsc->caps_list)) {
130 cap = list_first_entry(&mdsc->caps_list,
131 struct ceph_cap, caps_item);
Sage Weila8599bd2009-10-06 11:31:12 -0700132 list_del(&cap->caps_item);
133 kmem_cache_free(ceph_cap_cachep, cap);
134 }
Yehuda Sadeh37151662010-06-17 16:16:12 -0700135 mdsc->caps_total_count = 0;
136 mdsc->caps_avail_count = 0;
137 mdsc->caps_use_count = 0;
138 mdsc->caps_reserve_count = 0;
139 mdsc->caps_min_count = 0;
140 spin_unlock(&mdsc->caps_list_lock);
Sage Weil85ccce42010-02-17 10:02:43 -0800141}
142
Yehuda Sadeh37151662010-06-17 16:16:12 -0700143void ceph_adjust_min_caps(struct ceph_mds_client *mdsc, int delta)
Sage Weil85ccce42010-02-17 10:02:43 -0800144{
Yehuda Sadeh37151662010-06-17 16:16:12 -0700145 spin_lock(&mdsc->caps_list_lock);
146 mdsc->caps_min_count += delta;
147 BUG_ON(mdsc->caps_min_count < 0);
148 spin_unlock(&mdsc->caps_list_lock);
Sage Weila8599bd2009-10-06 11:31:12 -0700149}
150
majianpeng93faca62013-06-26 11:15:27 +0800151void ceph_reserve_caps(struct ceph_mds_client *mdsc,
Yehuda Sadeh37151662010-06-17 16:16:12 -0700152 struct ceph_cap_reservation *ctx, int need)
Sage Weila8599bd2009-10-06 11:31:12 -0700153{
154 int i;
155 struct ceph_cap *cap;
156 int have;
157 int alloc = 0;
158 LIST_HEAD(newcaps);
Sage Weila8599bd2009-10-06 11:31:12 -0700159
160 dout("reserve caps ctx=%p need=%d\n", ctx, need);
161
162 /* first reserve any caps that are already allocated */
Yehuda Sadeh37151662010-06-17 16:16:12 -0700163 spin_lock(&mdsc->caps_list_lock);
164 if (mdsc->caps_avail_count >= need)
Sage Weila8599bd2009-10-06 11:31:12 -0700165 have = need;
166 else
Yehuda Sadeh37151662010-06-17 16:16:12 -0700167 have = mdsc->caps_avail_count;
168 mdsc->caps_avail_count -= have;
169 mdsc->caps_reserve_count += have;
170 BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count +
171 mdsc->caps_reserve_count +
172 mdsc->caps_avail_count);
173 spin_unlock(&mdsc->caps_list_lock);
Sage Weila8599bd2009-10-06 11:31:12 -0700174
175 for (i = have; i < need; i++) {
176 cap = kmem_cache_alloc(ceph_cap_cachep, GFP_NOFS);
majianpeng93faca62013-06-26 11:15:27 +0800177 if (!cap)
178 break;
Sage Weila8599bd2009-10-06 11:31:12 -0700179 list_add(&cap->caps_item, &newcaps);
180 alloc++;
181 }
majianpeng93faca62013-06-26 11:15:27 +0800182 /* we didn't manage to reserve as much as we needed */
183 if (have + alloc != need)
184 pr_warn("reserve caps ctx=%p ENOMEM need=%d got=%d\n",
185 ctx, need, have + alloc);
Sage Weila8599bd2009-10-06 11:31:12 -0700186
Yehuda Sadeh37151662010-06-17 16:16:12 -0700187 spin_lock(&mdsc->caps_list_lock);
188 mdsc->caps_total_count += alloc;
189 mdsc->caps_reserve_count += alloc;
190 list_splice(&newcaps, &mdsc->caps_list);
Sage Weila8599bd2009-10-06 11:31:12 -0700191
Yehuda Sadeh37151662010-06-17 16:16:12 -0700192 BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count +
193 mdsc->caps_reserve_count +
194 mdsc->caps_avail_count);
195 spin_unlock(&mdsc->caps_list_lock);
Sage Weila8599bd2009-10-06 11:31:12 -0700196
197 ctx->count = need;
198 dout("reserve caps ctx=%p %d = %d used + %d resv + %d avail\n",
Yehuda Sadeh37151662010-06-17 16:16:12 -0700199 ctx, mdsc->caps_total_count, mdsc->caps_use_count,
200 mdsc->caps_reserve_count, mdsc->caps_avail_count);
Sage Weila8599bd2009-10-06 11:31:12 -0700201}
202
Yehuda Sadeh37151662010-06-17 16:16:12 -0700203int ceph_unreserve_caps(struct ceph_mds_client *mdsc,
204 struct ceph_cap_reservation *ctx)
Sage Weila8599bd2009-10-06 11:31:12 -0700205{
206 dout("unreserve caps ctx=%p count=%d\n", ctx, ctx->count);
207 if (ctx->count) {
Yehuda Sadeh37151662010-06-17 16:16:12 -0700208 spin_lock(&mdsc->caps_list_lock);
209 BUG_ON(mdsc->caps_reserve_count < ctx->count);
210 mdsc->caps_reserve_count -= ctx->count;
211 mdsc->caps_avail_count += ctx->count;
Sage Weila8599bd2009-10-06 11:31:12 -0700212 ctx->count = 0;
213 dout("unreserve caps %d = %d used + %d resv + %d avail\n",
Yehuda Sadeh37151662010-06-17 16:16:12 -0700214 mdsc->caps_total_count, mdsc->caps_use_count,
215 mdsc->caps_reserve_count, mdsc->caps_avail_count);
216 BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count +
217 mdsc->caps_reserve_count +
218 mdsc->caps_avail_count);
219 spin_unlock(&mdsc->caps_list_lock);
Sage Weila8599bd2009-10-06 11:31:12 -0700220 }
221 return 0;
222}
223
Yan, Zhengd9df2782014-04-18 09:57:11 +0800224struct ceph_cap *ceph_get_cap(struct ceph_mds_client *mdsc,
225 struct ceph_cap_reservation *ctx)
Sage Weila8599bd2009-10-06 11:31:12 -0700226{
227 struct ceph_cap *cap = NULL;
228
229 /* temporary, until we do something about cap import/export */
Sage Weil443b3762010-06-29 09:28:39 -0700230 if (!ctx) {
231 cap = kmem_cache_alloc(ceph_cap_cachep, GFP_NOFS);
232 if (cap) {
Yan, Zheng4d1d0532012-11-03 10:32:37 +0800233 spin_lock(&mdsc->caps_list_lock);
Yehuda Sadeh37151662010-06-17 16:16:12 -0700234 mdsc->caps_use_count++;
235 mdsc->caps_total_count++;
Yan, Zheng4d1d0532012-11-03 10:32:37 +0800236 spin_unlock(&mdsc->caps_list_lock);
Sage Weil443b3762010-06-29 09:28:39 -0700237 }
238 return cap;
239 }
Sage Weila8599bd2009-10-06 11:31:12 -0700240
Yehuda Sadeh37151662010-06-17 16:16:12 -0700241 spin_lock(&mdsc->caps_list_lock);
Sage Weila8599bd2009-10-06 11:31:12 -0700242 dout("get_cap ctx=%p (%d) %d = %d used + %d resv + %d avail\n",
Yehuda Sadeh37151662010-06-17 16:16:12 -0700243 ctx, ctx->count, mdsc->caps_total_count, mdsc->caps_use_count,
244 mdsc->caps_reserve_count, mdsc->caps_avail_count);
Sage Weila8599bd2009-10-06 11:31:12 -0700245 BUG_ON(!ctx->count);
Yehuda Sadeh37151662010-06-17 16:16:12 -0700246 BUG_ON(ctx->count > mdsc->caps_reserve_count);
247 BUG_ON(list_empty(&mdsc->caps_list));
Sage Weila8599bd2009-10-06 11:31:12 -0700248
249 ctx->count--;
Yehuda Sadeh37151662010-06-17 16:16:12 -0700250 mdsc->caps_reserve_count--;
251 mdsc->caps_use_count++;
Sage Weila8599bd2009-10-06 11:31:12 -0700252
Yehuda Sadeh37151662010-06-17 16:16:12 -0700253 cap = list_first_entry(&mdsc->caps_list, struct ceph_cap, caps_item);
Sage Weila8599bd2009-10-06 11:31:12 -0700254 list_del(&cap->caps_item);
255
Yehuda Sadeh37151662010-06-17 16:16:12 -0700256 BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count +
257 mdsc->caps_reserve_count + mdsc->caps_avail_count);
258 spin_unlock(&mdsc->caps_list_lock);
Sage Weila8599bd2009-10-06 11:31:12 -0700259 return cap;
260}
261
Yehuda Sadeh37151662010-06-17 16:16:12 -0700262void ceph_put_cap(struct ceph_mds_client *mdsc, struct ceph_cap *cap)
Sage Weila8599bd2009-10-06 11:31:12 -0700263{
Yehuda Sadeh37151662010-06-17 16:16:12 -0700264 spin_lock(&mdsc->caps_list_lock);
Sage Weil7c1332b2010-02-16 11:39:45 -0800265 dout("put_cap %p %d = %d used + %d resv + %d avail\n",
Yehuda Sadeh37151662010-06-17 16:16:12 -0700266 cap, mdsc->caps_total_count, mdsc->caps_use_count,
267 mdsc->caps_reserve_count, mdsc->caps_avail_count);
268 mdsc->caps_use_count--;
Sage Weila8599bd2009-10-06 11:31:12 -0700269 /*
Sage Weil85ccce42010-02-17 10:02:43 -0800270 * Keep some preallocated caps around (ceph_min_count), to
271 * avoid lots of free/alloc churn.
Sage Weila8599bd2009-10-06 11:31:12 -0700272 */
Yehuda Sadeh37151662010-06-17 16:16:12 -0700273 if (mdsc->caps_avail_count >= mdsc->caps_reserve_count +
274 mdsc->caps_min_count) {
275 mdsc->caps_total_count--;
Sage Weila8599bd2009-10-06 11:31:12 -0700276 kmem_cache_free(ceph_cap_cachep, cap);
277 } else {
Yehuda Sadeh37151662010-06-17 16:16:12 -0700278 mdsc->caps_avail_count++;
279 list_add(&cap->caps_item, &mdsc->caps_list);
Sage Weila8599bd2009-10-06 11:31:12 -0700280 }
281
Yehuda Sadeh37151662010-06-17 16:16:12 -0700282 BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count +
283 mdsc->caps_reserve_count + mdsc->caps_avail_count);
284 spin_unlock(&mdsc->caps_list_lock);
Sage Weila8599bd2009-10-06 11:31:12 -0700285}
286
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700287void ceph_reservation_status(struct ceph_fs_client *fsc,
Sage Weil85ccce42010-02-17 10:02:43 -0800288 int *total, int *avail, int *used, int *reserved,
289 int *min)
Sage Weila8599bd2009-10-06 11:31:12 -0700290{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700291 struct ceph_mds_client *mdsc = fsc->mdsc;
Yehuda Sadeh37151662010-06-17 16:16:12 -0700292
Sage Weila8599bd2009-10-06 11:31:12 -0700293 if (total)
Yehuda Sadeh37151662010-06-17 16:16:12 -0700294 *total = mdsc->caps_total_count;
Sage Weila8599bd2009-10-06 11:31:12 -0700295 if (avail)
Yehuda Sadeh37151662010-06-17 16:16:12 -0700296 *avail = mdsc->caps_avail_count;
Sage Weila8599bd2009-10-06 11:31:12 -0700297 if (used)
Yehuda Sadeh37151662010-06-17 16:16:12 -0700298 *used = mdsc->caps_use_count;
Sage Weila8599bd2009-10-06 11:31:12 -0700299 if (reserved)
Yehuda Sadeh37151662010-06-17 16:16:12 -0700300 *reserved = mdsc->caps_reserve_count;
Sage Weil85ccce42010-02-17 10:02:43 -0800301 if (min)
Yehuda Sadeh37151662010-06-17 16:16:12 -0700302 *min = mdsc->caps_min_count;
Sage Weila8599bd2009-10-06 11:31:12 -0700303}
304
305/*
306 * Find ceph_cap for given mds, if any.
307 *
Sage Weilbe655592011-11-30 09:47:09 -0800308 * Called with i_ceph_lock held.
Sage Weila8599bd2009-10-06 11:31:12 -0700309 */
310static struct ceph_cap *__get_cap_for_mds(struct ceph_inode_info *ci, int mds)
311{
312 struct ceph_cap *cap;
313 struct rb_node *n = ci->i_caps.rb_node;
314
315 while (n) {
316 cap = rb_entry(n, struct ceph_cap, ci_node);
317 if (mds < cap->mds)
318 n = n->rb_left;
319 else if (mds > cap->mds)
320 n = n->rb_right;
321 else
322 return cap;
323 }
324 return NULL;
325}
326
Greg Farnum2bc50252010-06-30 12:44:34 -0700327struct ceph_cap *ceph_get_cap_for_mds(struct ceph_inode_info *ci, int mds)
328{
329 struct ceph_cap *cap;
330
Sage Weilbe655592011-11-30 09:47:09 -0800331 spin_lock(&ci->i_ceph_lock);
Greg Farnum2bc50252010-06-30 12:44:34 -0700332 cap = __get_cap_for_mds(ci, mds);
Sage Weilbe655592011-11-30 09:47:09 -0800333 spin_unlock(&ci->i_ceph_lock);
Greg Farnum2bc50252010-06-30 12:44:34 -0700334 return cap;
335}
336
Sage Weila8599bd2009-10-06 11:31:12 -0700337/*
Sage Weil33caad32010-05-26 14:31:27 -0700338 * Return id of any MDS with a cap, preferably FILE_WR|BUFFER|EXCL, else -1.
Sage Weila8599bd2009-10-06 11:31:12 -0700339 */
Sage Weilca81f3f2010-06-10 14:21:36 -0700340static int __ceph_get_cap_mds(struct ceph_inode_info *ci)
Sage Weila8599bd2009-10-06 11:31:12 -0700341{
342 struct ceph_cap *cap;
343 int mds = -1;
344 struct rb_node *p;
345
Sage Weil33caad32010-05-26 14:31:27 -0700346 /* prefer mds with WR|BUFFER|EXCL caps */
Sage Weila8599bd2009-10-06 11:31:12 -0700347 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
348 cap = rb_entry(p, struct ceph_cap, ci_node);
349 mds = cap->mds;
Sage Weila8599bd2009-10-06 11:31:12 -0700350 if (cap->issued & (CEPH_CAP_FILE_WR |
351 CEPH_CAP_FILE_BUFFER |
352 CEPH_CAP_FILE_EXCL))
353 break;
354 }
355 return mds;
356}
357
358int ceph_get_cap_mds(struct inode *inode)
359{
Sage Weilbe655592011-11-30 09:47:09 -0800360 struct ceph_inode_info *ci = ceph_inode(inode);
Sage Weila8599bd2009-10-06 11:31:12 -0700361 int mds;
Sage Weilbe655592011-11-30 09:47:09 -0800362 spin_lock(&ci->i_ceph_lock);
Sage Weilca81f3f2010-06-10 14:21:36 -0700363 mds = __ceph_get_cap_mds(ceph_inode(inode));
Sage Weilbe655592011-11-30 09:47:09 -0800364 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -0700365 return mds;
366}
367
368/*
Sage Weilbe655592011-11-30 09:47:09 -0800369 * Called under i_ceph_lock.
Sage Weila8599bd2009-10-06 11:31:12 -0700370 */
371static void __insert_cap_node(struct ceph_inode_info *ci,
372 struct ceph_cap *new)
373{
374 struct rb_node **p = &ci->i_caps.rb_node;
375 struct rb_node *parent = NULL;
376 struct ceph_cap *cap = NULL;
377
378 while (*p) {
379 parent = *p;
380 cap = rb_entry(parent, struct ceph_cap, ci_node);
381 if (new->mds < cap->mds)
382 p = &(*p)->rb_left;
383 else if (new->mds > cap->mds)
384 p = &(*p)->rb_right;
385 else
386 BUG();
387 }
388
389 rb_link_node(&new->ci_node, parent, p);
390 rb_insert_color(&new->ci_node, &ci->i_caps);
391}
392
393/*
394 * (re)set cap hold timeouts, which control the delayed release
395 * of unused caps back to the MDS. Should be called on cap use.
396 */
397static void __cap_set_timeouts(struct ceph_mds_client *mdsc,
398 struct ceph_inode_info *ci)
399{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700400 struct ceph_mount_options *ma = mdsc->fsc->mount_options;
Sage Weila8599bd2009-10-06 11:31:12 -0700401
402 ci->i_hold_caps_min = round_jiffies(jiffies +
403 ma->caps_wanted_delay_min * HZ);
404 ci->i_hold_caps_max = round_jiffies(jiffies +
405 ma->caps_wanted_delay_max * HZ);
406 dout("__cap_set_timeouts %p min %lu max %lu\n", &ci->vfs_inode,
407 ci->i_hold_caps_min - jiffies, ci->i_hold_caps_max - jiffies);
408}
409
410/*
411 * (Re)queue cap at the end of the delayed cap release list.
412 *
413 * If I_FLUSH is set, leave the inode at the front of the list.
414 *
Sage Weilbe655592011-11-30 09:47:09 -0800415 * Caller holds i_ceph_lock
Sage Weila8599bd2009-10-06 11:31:12 -0700416 * -> we take mdsc->cap_delay_lock
417 */
418static void __cap_delay_requeue(struct ceph_mds_client *mdsc,
419 struct ceph_inode_info *ci)
420{
421 __cap_set_timeouts(mdsc, ci);
422 dout("__cap_delay_requeue %p flags %d at %lu\n", &ci->vfs_inode,
423 ci->i_ceph_flags, ci->i_hold_caps_max);
424 if (!mdsc->stopping) {
425 spin_lock(&mdsc->cap_delay_lock);
426 if (!list_empty(&ci->i_cap_delay_list)) {
427 if (ci->i_ceph_flags & CEPH_I_FLUSH)
428 goto no_change;
429 list_del_init(&ci->i_cap_delay_list);
430 }
431 list_add_tail(&ci->i_cap_delay_list, &mdsc->cap_delay_list);
432no_change:
433 spin_unlock(&mdsc->cap_delay_lock);
434 }
435}
436
437/*
438 * Queue an inode for immediate writeback. Mark inode with I_FLUSH,
439 * indicating we should send a cap message to flush dirty metadata
440 * asap, and move to the front of the delayed cap list.
441 */
442static void __cap_delay_requeue_front(struct ceph_mds_client *mdsc,
443 struct ceph_inode_info *ci)
444{
445 dout("__cap_delay_requeue_front %p\n", &ci->vfs_inode);
446 spin_lock(&mdsc->cap_delay_lock);
447 ci->i_ceph_flags |= CEPH_I_FLUSH;
448 if (!list_empty(&ci->i_cap_delay_list))
449 list_del_init(&ci->i_cap_delay_list);
450 list_add(&ci->i_cap_delay_list, &mdsc->cap_delay_list);
451 spin_unlock(&mdsc->cap_delay_lock);
452}
453
454/*
455 * Cancel delayed work on cap.
456 *
Sage Weilbe655592011-11-30 09:47:09 -0800457 * Caller must hold i_ceph_lock.
Sage Weila8599bd2009-10-06 11:31:12 -0700458 */
459static void __cap_delay_cancel(struct ceph_mds_client *mdsc,
460 struct ceph_inode_info *ci)
461{
462 dout("__cap_delay_cancel %p\n", &ci->vfs_inode);
463 if (list_empty(&ci->i_cap_delay_list))
464 return;
465 spin_lock(&mdsc->cap_delay_lock);
466 list_del_init(&ci->i_cap_delay_list);
467 spin_unlock(&mdsc->cap_delay_lock);
468}
469
470/*
471 * Common issue checks for add_cap, handle_cap_grant.
472 */
473static void __check_cap_issue(struct ceph_inode_info *ci, struct ceph_cap *cap,
474 unsigned issued)
475{
476 unsigned had = __ceph_caps_issued(ci, NULL);
477
478 /*
479 * Each time we receive FILE_CACHE anew, we increment
480 * i_rdcache_gen.
481 */
Sage Weil29625072010-05-27 10:40:43 -0700482 if ((issued & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) &&
Milosz Tanski99ccbd22013-08-21 17:29:54 -0400483 (had & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) == 0) {
Sage Weila8599bd2009-10-06 11:31:12 -0700484 ci->i_rdcache_gen++;
Milosz Tanski99ccbd22013-08-21 17:29:54 -0400485 }
Sage Weila8599bd2009-10-06 11:31:12 -0700486
487 /*
Yan, Zheng2f276c52013-03-13 19:44:32 +0800488 * if we are newly issued FILE_SHARED, mark dir not complete; we
Sage Weila8599bd2009-10-06 11:31:12 -0700489 * don't know what happened to this directory while we didn't
490 * have the cap.
491 */
492 if ((issued & CEPH_CAP_FILE_SHARED) &&
493 (had & CEPH_CAP_FILE_SHARED) == 0) {
494 ci->i_shared_gen++;
Yan, Zhenga8673d62013-02-18 16:38:14 +0800495 if (S_ISDIR(ci->vfs_inode.i_mode)) {
496 dout(" marking %p NOT complete\n", &ci->vfs_inode);
Yan, Zheng2f276c52013-03-13 19:44:32 +0800497 __ceph_dir_clear_complete(ci);
Yan, Zhenga8673d62013-02-18 16:38:14 +0800498 }
Sage Weila8599bd2009-10-06 11:31:12 -0700499 }
500}
501
502/*
503 * Add a capability under the given MDS session.
504 *
505 * Caller should hold session snap_rwsem (read) and s_mutex.
506 *
507 * @fmode is the open file mode, if we are opening a file, otherwise
508 * it is < 0. (This is so we can atomically add the cap and add an
509 * open file reference to it.)
510 */
Yan, Zhengd9df2782014-04-18 09:57:11 +0800511void ceph_add_cap(struct inode *inode,
512 struct ceph_mds_session *session, u64 cap_id,
513 int fmode, unsigned issued, unsigned wanted,
514 unsigned seq, unsigned mseq, u64 realmino, int flags,
515 struct ceph_cap **new_cap)
Sage Weila8599bd2009-10-06 11:31:12 -0700516{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700517 struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
Sage Weila8599bd2009-10-06 11:31:12 -0700518 struct ceph_inode_info *ci = ceph_inode(inode);
Sage Weila8599bd2009-10-06 11:31:12 -0700519 struct ceph_cap *cap;
520 int mds = session->s_mds;
521 int actual_wanted;
522
523 dout("add_cap %p mds%d cap %llx %s seq %d\n", inode,
524 session->s_mds, cap_id, ceph_cap_string(issued), seq);
525
526 /*
527 * If we are opening the file, include file mode wanted bits
528 * in wanted.
529 */
530 if (fmode >= 0)
531 wanted |= ceph_caps_for_mode(fmode);
532
Sage Weila8599bd2009-10-06 11:31:12 -0700533 cap = __get_cap_for_mds(ci, mds);
534 if (!cap) {
Yan, Zhengd9df2782014-04-18 09:57:11 +0800535 cap = *new_cap;
536 *new_cap = NULL;
Sage Weila8599bd2009-10-06 11:31:12 -0700537
538 cap->issued = 0;
539 cap->implemented = 0;
540 cap->mds = mds;
541 cap->mds_wanted = 0;
Yan, Zheng964266c2013-02-27 09:26:09 +0800542 cap->mseq = 0;
Sage Weila8599bd2009-10-06 11:31:12 -0700543
544 cap->ci = ci;
545 __insert_cap_node(ci, cap);
546
Sage Weila8599bd2009-10-06 11:31:12 -0700547 /* add to session cap list */
548 cap->session = session;
549 spin_lock(&session->s_cap_lock);
550 list_add_tail(&cap->session_caps, &session->s_caps);
551 session->s_nr_caps++;
552 spin_unlock(&session->s_cap_lock);
Yan, Zheng11df2df2013-11-24 14:44:38 +0800553 } else {
Yan, Zheng11df2df2013-11-24 14:44:38 +0800554 /*
555 * auth mds of the inode changed. we received the cap export
556 * message, but still haven't received the cap import message.
557 * handle_cap_export() updated the new auth MDS' cap.
558 *
559 * "ceph_seq_cmp(seq, cap->seq) <= 0" means we are processing
560 * a message that was send before the cap import message. So
561 * don't remove caps.
562 */
563 if (ceph_seq_cmp(seq, cap->seq) <= 0) {
564 WARN_ON(cap != ci->i_auth_cap);
565 WARN_ON(cap->cap_id != cap_id);
566 seq = cap->seq;
567 mseq = cap->mseq;
568 issued |= cap->issued;
569 flags |= CEPH_CAP_FLAG_AUTH;
570 }
571 }
Sage Weila8599bd2009-10-06 11:31:12 -0700572
573 if (!ci->i_snap_realm) {
574 /*
575 * add this inode to the appropriate snap realm
576 */
577 struct ceph_snap_realm *realm = ceph_lookup_snap_realm(mdsc,
578 realmino);
579 if (realm) {
Sage Weila8599bd2009-10-06 11:31:12 -0700580 spin_lock(&realm->inodes_with_caps_lock);
581 ci->i_snap_realm = realm;
582 list_add(&ci->i_snap_realm_item,
583 &realm->inodes_with_caps);
584 spin_unlock(&realm->inodes_with_caps_lock);
585 } else {
586 pr_err("ceph_add_cap: couldn't find snap realm %llx\n",
587 realmino);
Sage Weilb8cd07e2010-07-16 12:00:02 -0700588 WARN_ON(!realm);
Sage Weila8599bd2009-10-06 11:31:12 -0700589 }
590 }
591
592 __check_cap_issue(ci, cap, issued);
593
594 /*
595 * If we are issued caps we don't want, or the mds' wanted
596 * value appears to be off, queue a check so we'll release
597 * later and/or update the mds wanted value.
598 */
599 actual_wanted = __ceph_caps_wanted(ci);
600 if ((wanted & ~actual_wanted) ||
601 (issued & ~actual_wanted & CEPH_CAP_ANY_WR)) {
602 dout(" issued %s, mds wanted %s, actual %s, queueing\n",
603 ceph_cap_string(issued), ceph_cap_string(wanted),
604 ceph_cap_string(actual_wanted));
605 __cap_delay_requeue(mdsc, ci);
606 }
607
Yan, Zhengb8c2f3a2013-05-31 16:37:11 +0800608 if (flags & CEPH_CAP_FLAG_AUTH) {
609 if (ci->i_auth_cap == NULL ||
Yan, Zhengd9ffc4f2014-03-18 10:15:29 +0800610 ceph_seq_cmp(ci->i_auth_cap->mseq, mseq) < 0) {
Yan, Zhengb8c2f3a2013-05-31 16:37:11 +0800611 ci->i_auth_cap = cap;
Yan, Zhengd9ffc4f2014-03-18 10:15:29 +0800612 cap->mds_wanted = wanted;
613 }
Yan, Zheng11df2df2013-11-24 14:44:38 +0800614 } else {
615 WARN_ON(ci->i_auth_cap == cap);
Yan, Zheng8a92a112013-01-04 14:28:07 +0800616 }
Sage Weila8599bd2009-10-06 11:31:12 -0700617
618 dout("add_cap inode %p (%llx.%llx) cap %p %s now %s seq %d mds%d\n",
619 inode, ceph_vinop(inode), cap, ceph_cap_string(issued),
620 ceph_cap_string(issued|cap->issued), seq, mds);
621 cap->cap_id = cap_id;
622 cap->issued = issued;
623 cap->implemented |= issued;
Yan, Zhengd1b87802013-11-13 14:47:19 +0800624 if (ceph_seq_cmp(mseq, cap->mseq) > 0)
Yan, Zheng964266c2013-02-27 09:26:09 +0800625 cap->mds_wanted = wanted;
626 else
627 cap->mds_wanted |= wanted;
Sage Weila8599bd2009-10-06 11:31:12 -0700628 cap->seq = seq;
629 cap->issue_seq = seq;
630 cap->mseq = mseq;
Sage Weil685f9a5d2009-11-09 12:05:48 -0800631 cap->cap_gen = session->s_cap_gen;
Sage Weila8599bd2009-10-06 11:31:12 -0700632
633 if (fmode >= 0)
634 __ceph_get_fmode(ci, fmode);
Sage Weila8599bd2009-10-06 11:31:12 -0700635}
636
637/*
638 * Return true if cap has not timed out and belongs to the current
639 * generation of the MDS session (i.e. has not gone 'stale' due to
640 * us losing touch with the mds).
641 */
642static int __cap_is_valid(struct ceph_cap *cap)
643{
644 unsigned long ttl;
Sage Weilcdac8302009-11-10 16:02:23 -0800645 u32 gen;
Sage Weila8599bd2009-10-06 11:31:12 -0700646
Alex Elderd8fb02a2012-01-12 17:48:10 -0800647 spin_lock(&cap->session->s_gen_ttl_lock);
Sage Weila8599bd2009-10-06 11:31:12 -0700648 gen = cap->session->s_cap_gen;
649 ttl = cap->session->s_cap_ttl;
Alex Elderd8fb02a2012-01-12 17:48:10 -0800650 spin_unlock(&cap->session->s_gen_ttl_lock);
Sage Weila8599bd2009-10-06 11:31:12 -0700651
Sage Weil685f9a5d2009-11-09 12:05:48 -0800652 if (cap->cap_gen < gen || time_after_eq(jiffies, ttl)) {
Sage Weila8599bd2009-10-06 11:31:12 -0700653 dout("__cap_is_valid %p cap %p issued %s "
654 "but STALE (gen %u vs %u)\n", &cap->ci->vfs_inode,
Sage Weil685f9a5d2009-11-09 12:05:48 -0800655 cap, ceph_cap_string(cap->issued), cap->cap_gen, gen);
Sage Weila8599bd2009-10-06 11:31:12 -0700656 return 0;
657 }
658
659 return 1;
660}
661
662/*
663 * Return set of valid cap bits issued to us. Note that caps time
664 * out, and may be invalidated in bulk if the client session times out
665 * and session->s_cap_gen is bumped.
666 */
667int __ceph_caps_issued(struct ceph_inode_info *ci, int *implemented)
668{
Yan, Zhengd9df2782014-04-18 09:57:11 +0800669 int have = ci->i_snap_caps;
Sage Weila8599bd2009-10-06 11:31:12 -0700670 struct ceph_cap *cap;
671 struct rb_node *p;
672
673 if (implemented)
674 *implemented = 0;
675 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
676 cap = rb_entry(p, struct ceph_cap, ci_node);
677 if (!__cap_is_valid(cap))
678 continue;
679 dout("__ceph_caps_issued %p cap %p issued %s\n",
680 &ci->vfs_inode, cap, ceph_cap_string(cap->issued));
681 have |= cap->issued;
682 if (implemented)
683 *implemented |= cap->implemented;
684 }
Yan, Zhengb1530f52013-07-02 12:40:20 +0800685 /*
686 * exclude caps issued by non-auth MDS, but are been revoking
687 * by the auth MDS. The non-auth MDS should be revoking/exporting
688 * these caps, but the message is delayed.
689 */
690 if (ci->i_auth_cap) {
691 cap = ci->i_auth_cap;
692 have &= ~cap->implemented | cap->issued;
693 }
Sage Weila8599bd2009-10-06 11:31:12 -0700694 return have;
695}
696
697/*
698 * Get cap bits issued by caps other than @ocap
699 */
700int __ceph_caps_issued_other(struct ceph_inode_info *ci, struct ceph_cap *ocap)
701{
702 int have = ci->i_snap_caps;
703 struct ceph_cap *cap;
704 struct rb_node *p;
705
706 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
707 cap = rb_entry(p, struct ceph_cap, ci_node);
708 if (cap == ocap)
709 continue;
710 if (!__cap_is_valid(cap))
711 continue;
712 have |= cap->issued;
713 }
714 return have;
715}
716
717/*
718 * Move a cap to the end of the LRU (oldest caps at list head, newest
719 * at list tail).
720 */
721static void __touch_cap(struct ceph_cap *cap)
722{
723 struct ceph_mds_session *s = cap->session;
724
Sage Weila8599bd2009-10-06 11:31:12 -0700725 spin_lock(&s->s_cap_lock);
Sage Weil7c1332b2010-02-16 11:39:45 -0800726 if (s->s_cap_iterator == NULL) {
Sage Weil5dacf092009-12-21 20:40:34 -0800727 dout("__touch_cap %p cap %p mds%d\n", &cap->ci->vfs_inode, cap,
728 s->s_mds);
729 list_move_tail(&cap->session_caps, &s->s_caps);
730 } else {
731 dout("__touch_cap %p cap %p mds%d NOP, iterating over caps\n",
732 &cap->ci->vfs_inode, cap, s->s_mds);
733 }
Sage Weila8599bd2009-10-06 11:31:12 -0700734 spin_unlock(&s->s_cap_lock);
735}
736
737/*
738 * Check if we hold the given mask. If so, move the cap(s) to the
739 * front of their respective LRUs. (This is the preferred way for
740 * callers to check for caps they want.)
741 */
742int __ceph_caps_issued_mask(struct ceph_inode_info *ci, int mask, int touch)
743{
744 struct ceph_cap *cap;
745 struct rb_node *p;
746 int have = ci->i_snap_caps;
747
748 if ((have & mask) == mask) {
749 dout("__ceph_caps_issued_mask %p snap issued %s"
750 " (mask %s)\n", &ci->vfs_inode,
751 ceph_cap_string(have),
752 ceph_cap_string(mask));
753 return 1;
754 }
755
756 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
757 cap = rb_entry(p, struct ceph_cap, ci_node);
758 if (!__cap_is_valid(cap))
759 continue;
760 if ((cap->issued & mask) == mask) {
761 dout("__ceph_caps_issued_mask %p cap %p issued %s"
762 " (mask %s)\n", &ci->vfs_inode, cap,
763 ceph_cap_string(cap->issued),
764 ceph_cap_string(mask));
765 if (touch)
766 __touch_cap(cap);
767 return 1;
768 }
769
770 /* does a combination of caps satisfy mask? */
771 have |= cap->issued;
772 if ((have & mask) == mask) {
773 dout("__ceph_caps_issued_mask %p combo issued %s"
774 " (mask %s)\n", &ci->vfs_inode,
775 ceph_cap_string(cap->issued),
776 ceph_cap_string(mask));
777 if (touch) {
778 struct rb_node *q;
779
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300780 /* touch this + preceding caps */
Sage Weila8599bd2009-10-06 11:31:12 -0700781 __touch_cap(cap);
782 for (q = rb_first(&ci->i_caps); q != p;
783 q = rb_next(q)) {
784 cap = rb_entry(q, struct ceph_cap,
785 ci_node);
786 if (!__cap_is_valid(cap))
787 continue;
788 __touch_cap(cap);
789 }
790 }
791 return 1;
792 }
793 }
794
795 return 0;
796}
797
798/*
799 * Return true if mask caps are currently being revoked by an MDS.
800 */
Yan, Zheng6ee6b9532013-07-02 12:40:21 +0800801int __ceph_caps_revoking_other(struct ceph_inode_info *ci,
802 struct ceph_cap *ocap, int mask)
803{
804 struct ceph_cap *cap;
805 struct rb_node *p;
806
807 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
808 cap = rb_entry(p, struct ceph_cap, ci_node);
Yan, Zheng9563f882013-11-22 13:50:45 +0800809 if (cap != ocap &&
Yan, Zheng6ee6b9532013-07-02 12:40:21 +0800810 (cap->implemented & ~cap->issued & mask))
811 return 1;
812 }
813 return 0;
814}
815
Sage Weila8599bd2009-10-06 11:31:12 -0700816int ceph_caps_revoking(struct ceph_inode_info *ci, int mask)
817{
818 struct inode *inode = &ci->vfs_inode;
Yan, Zheng6ee6b9532013-07-02 12:40:21 +0800819 int ret;
Sage Weila8599bd2009-10-06 11:31:12 -0700820
Sage Weilbe655592011-11-30 09:47:09 -0800821 spin_lock(&ci->i_ceph_lock);
Yan, Zheng6ee6b9532013-07-02 12:40:21 +0800822 ret = __ceph_caps_revoking_other(ci, NULL, mask);
Sage Weilbe655592011-11-30 09:47:09 -0800823 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -0700824 dout("ceph_caps_revoking %p %s = %d\n", inode,
825 ceph_cap_string(mask), ret);
826 return ret;
827}
828
829int __ceph_caps_used(struct ceph_inode_info *ci)
830{
831 int used = 0;
832 if (ci->i_pin_ref)
833 used |= CEPH_CAP_PIN;
834 if (ci->i_rd_ref)
835 used |= CEPH_CAP_FILE_RD;
Yan, Zhengfdd4e152015-06-16 20:48:56 +0800836 if (ci->i_rdcache_ref ||
837 (!S_ISDIR(ci->vfs_inode.i_mode) && /* ignore readdir cache */
838 ci->vfs_inode.i_data.nrpages))
Sage Weila8599bd2009-10-06 11:31:12 -0700839 used |= CEPH_CAP_FILE_CACHE;
840 if (ci->i_wr_ref)
841 used |= CEPH_CAP_FILE_WR;
Henry C Changd3d07202011-05-11 10:29:54 +0000842 if (ci->i_wb_ref || ci->i_wrbuffer_ref)
Sage Weila8599bd2009-10-06 11:31:12 -0700843 used |= CEPH_CAP_FILE_BUFFER;
844 return used;
845}
846
847/*
848 * wanted, by virtue of open file modes
849 */
850int __ceph_caps_file_wanted(struct ceph_inode_info *ci)
851{
852 int want = 0;
853 int mode;
Sage Weil33caad32010-05-26 14:31:27 -0700854 for (mode = 0; mode < CEPH_FILE_MODE_NUM; mode++)
Sage Weila8599bd2009-10-06 11:31:12 -0700855 if (ci->i_nr_by_mode[mode])
856 want |= ceph_caps_for_mode(mode);
857 return want;
858}
859
860/*
861 * Return caps we have registered with the MDS(s) as 'wanted'.
862 */
863int __ceph_caps_mds_wanted(struct ceph_inode_info *ci)
864{
865 struct ceph_cap *cap;
866 struct rb_node *p;
867 int mds_wanted = 0;
868
869 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
870 cap = rb_entry(p, struct ceph_cap, ci_node);
871 if (!__cap_is_valid(cap))
872 continue;
Yan, Zhenga2550602014-03-08 09:51:45 +0800873 if (cap == ci->i_auth_cap)
874 mds_wanted |= cap->mds_wanted;
875 else
876 mds_wanted |= (cap->mds_wanted & ~CEPH_CAP_ANY_FILE_WR);
Sage Weila8599bd2009-10-06 11:31:12 -0700877 }
878 return mds_wanted;
879}
880
881/*
Sage Weilbe655592011-11-30 09:47:09 -0800882 * called under i_ceph_lock
Sage Weila8599bd2009-10-06 11:31:12 -0700883 */
884static int __ceph_is_any_caps(struct ceph_inode_info *ci)
885{
Yan, Zhengd9df2782014-04-18 09:57:11 +0800886 return !RB_EMPTY_ROOT(&ci->i_caps);
Sage Weila8599bd2009-10-06 11:31:12 -0700887}
888
Yan, Zheng9215aee2013-11-30 12:47:41 +0800889int ceph_is_any_caps(struct inode *inode)
890{
891 struct ceph_inode_info *ci = ceph_inode(inode);
892 int ret;
893
894 spin_lock(&ci->i_ceph_lock);
895 ret = __ceph_is_any_caps(ci);
896 spin_unlock(&ci->i_ceph_lock);
897
898 return ret;
899}
900
Yan, Zhengdb40cc12015-03-23 20:12:20 +0800901static void drop_inode_snap_realm(struct ceph_inode_info *ci)
902{
903 struct ceph_snap_realm *realm = ci->i_snap_realm;
904 spin_lock(&realm->inodes_with_caps_lock);
905 list_del_init(&ci->i_snap_realm_item);
906 ci->i_snap_realm_counter++;
907 ci->i_snap_realm = NULL;
908 spin_unlock(&realm->inodes_with_caps_lock);
909 ceph_put_snap_realm(ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc,
910 realm);
911}
912
Sage Weila8599bd2009-10-06 11:31:12 -0700913/*
Sage Weilf818a732010-05-11 20:56:31 -0700914 * Remove a cap. Take steps to deal with a racing iterate_session_caps.
915 *
Sage Weilbe655592011-11-30 09:47:09 -0800916 * caller should hold i_ceph_lock.
Sage Weila6369742010-02-22 13:59:00 -0800917 * caller will not hold session s_mutex if called from destroy_inode.
Sage Weila8599bd2009-10-06 11:31:12 -0700918 */
Yan, Zhenga096b092013-09-22 10:15:58 +0800919void __ceph_remove_cap(struct ceph_cap *cap, bool queue_release)
Sage Weila8599bd2009-10-06 11:31:12 -0700920{
921 struct ceph_mds_session *session = cap->session;
922 struct ceph_inode_info *ci = cap->ci;
Cheng Renquan640ef792010-03-26 17:40:33 +0800923 struct ceph_mds_client *mdsc =
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700924 ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc;
Sage Weilf818a732010-05-11 20:56:31 -0700925 int removed = 0;
Sage Weila8599bd2009-10-06 11:31:12 -0700926
927 dout("__ceph_remove_cap %p from %p\n", cap, &ci->vfs_inode);
928
Sage Weil7c1332b2010-02-16 11:39:45 -0800929 /* remove from session list */
930 spin_lock(&session->s_cap_lock);
931 if (session->s_cap_iterator == cap) {
932 /* not yet, we are iterating over this very cap */
933 dout("__ceph_remove_cap delaying %p removal from session %p\n",
934 cap, cap->session);
935 } else {
936 list_del_init(&cap->session_caps);
937 session->s_nr_caps--;
938 cap->session = NULL;
Sage Weilf818a732010-05-11 20:56:31 -0700939 removed = 1;
Sage Weil7c1332b2010-02-16 11:39:45 -0800940 }
Sage Weilf818a732010-05-11 20:56:31 -0700941 /* protect backpointer with s_cap_lock: see iterate_session_caps */
942 cap->ci = NULL;
Yan, Zheng745a8e32015-05-14 17:22:42 +0800943
944 /*
945 * s_cap_reconnect is protected by s_cap_lock. no one changes
946 * s_cap_gen while session is in the reconnect state.
947 */
948 if (queue_release &&
949 (!session->s_cap_reconnect || cap->cap_gen == session->s_cap_gen)) {
950 cap->queue_release = 1;
951 if (removed) {
952 list_add_tail(&cap->session_caps,
953 &session->s_cap_releases);
954 session->s_num_cap_releases++;
955 removed = 0;
956 }
957 } else {
958 cap->queue_release = 0;
959 }
960 cap->cap_ino = ci->i_vino.ino;
961
Sage Weil7c1332b2010-02-16 11:39:45 -0800962 spin_unlock(&session->s_cap_lock);
963
Sage Weilf818a732010-05-11 20:56:31 -0700964 /* remove from inode list */
965 rb_erase(&cap->ci_node, &ci->i_caps);
966 if (ci->i_auth_cap == cap)
967 ci->i_auth_cap = NULL;
968
969 if (removed)
Yehuda Sadeh37151662010-06-17 16:16:12 -0700970 ceph_put_cap(mdsc, cap);
Sage Weila8599bd2009-10-06 11:31:12 -0700971
Yan, Zhengdb40cc12015-03-23 20:12:20 +0800972 /* when reconnect denied, we remove session caps forcibly,
973 * i_wr_ref can be non-zero. If there are ongoing write,
974 * keep i_snap_realm.
975 */
976 if (!__ceph_is_any_caps(ci) && ci->i_wr_ref == 0 && ci->i_snap_realm)
977 drop_inode_snap_realm(ci);
978
Sage Weila8599bd2009-10-06 11:31:12 -0700979 if (!__ceph_is_any_real_caps(ci))
980 __cap_delay_cancel(mdsc, ci);
981}
982
983/*
984 * Build and send a cap message to the given MDS.
985 *
986 * Caller should be holding s_mutex.
987 */
988static int send_cap_msg(struct ceph_mds_session *session,
989 u64 ino, u64 cid, int op,
990 int caps, int wanted, int dirty,
Yan, Zhenga2971c82015-06-10 11:09:32 +0800991 u32 seq, u64 flush_tid, u64 oldest_flush_tid,
992 u32 issue_seq, u32 mseq, u64 size, u64 max_size,
Sage Weila8599bd2009-10-06 11:31:12 -0700993 struct timespec *mtime, struct timespec *atime,
994 u64 time_warp_seq,
Eric W. Biederman05cb11c2013-01-31 02:56:19 -0800995 kuid_t uid, kgid_t gid, umode_t mode,
Sage Weila8599bd2009-10-06 11:31:12 -0700996 u64 xattr_version,
997 struct ceph_buffer *xattrs_buf,
Yan, Zhenge20d2582014-11-14 22:39:13 +0800998 u64 follows, bool inline_data)
Sage Weila8599bd2009-10-06 11:31:12 -0700999{
1000 struct ceph_mds_caps *fc;
1001 struct ceph_msg *msg;
Yan, Zhenge20d2582014-11-14 22:39:13 +08001002 void *p;
1003 size_t extra_len;
Sage Weila8599bd2009-10-06 11:31:12 -07001004
1005 dout("send_cap_msg %s %llx %llx caps %s wanted %s dirty %s"
Yan, Zhenga2971c82015-06-10 11:09:32 +08001006 " seq %u/%u tid %llu/%llu mseq %u follows %lld size %llu/%llu"
Sage Weila8599bd2009-10-06 11:31:12 -07001007 " xattr_ver %llu xattr_len %d\n", ceph_cap_op_name(op),
1008 cid, ino, ceph_cap_string(caps), ceph_cap_string(wanted),
1009 ceph_cap_string(dirty),
Yan, Zhenga2971c82015-06-10 11:09:32 +08001010 seq, issue_seq, flush_tid, oldest_flush_tid,
1011 mseq, follows, size, max_size,
Sage Weila8599bd2009-10-06 11:31:12 -07001012 xattr_version, xattrs_buf ? (int)xattrs_buf->vec.iov_len : 0);
1013
Yan, Zhenga2971c82015-06-10 11:09:32 +08001014 /* flock buffer size + inline version + inline data size +
1015 * osd_epoch_barrier + oldest_flush_tid */
1016 extra_len = 4 + 8 + 4 + 4 + 8;
Yan, Zhenge20d2582014-11-14 22:39:13 +08001017 msg = ceph_msg_new(CEPH_MSG_CLIENT_CAPS, sizeof(*fc) + extra_len,
1018 GFP_NOFS, false);
Sage Weila79832f2010-04-01 16:06:19 -07001019 if (!msg)
1020 return -ENOMEM;
Sage Weila8599bd2009-10-06 11:31:12 -07001021
Yan, Zhenga2971c82015-06-10 11:09:32 +08001022 msg->hdr.version = cpu_to_le16(6);
Sage Weil6df058c2009-12-22 11:24:33 -08001023 msg->hdr.tid = cpu_to_le64(flush_tid);
Sage Weila8599bd2009-10-06 11:31:12 -07001024
Sage Weil6df058c2009-12-22 11:24:33 -08001025 fc = msg->front.iov_base;
Sage Weila8599bd2009-10-06 11:31:12 -07001026 memset(fc, 0, sizeof(*fc));
1027
1028 fc->cap_id = cpu_to_le64(cid);
1029 fc->op = cpu_to_le32(op);
1030 fc->seq = cpu_to_le32(seq);
Sage Weila8599bd2009-10-06 11:31:12 -07001031 fc->issue_seq = cpu_to_le32(issue_seq);
1032 fc->migrate_seq = cpu_to_le32(mseq);
1033 fc->caps = cpu_to_le32(caps);
1034 fc->wanted = cpu_to_le32(wanted);
1035 fc->dirty = cpu_to_le32(dirty);
1036 fc->ino = cpu_to_le64(ino);
1037 fc->snap_follows = cpu_to_le64(follows);
1038
1039 fc->size = cpu_to_le64(size);
1040 fc->max_size = cpu_to_le64(max_size);
1041 if (mtime)
1042 ceph_encode_timespec(&fc->mtime, mtime);
1043 if (atime)
1044 ceph_encode_timespec(&fc->atime, atime);
1045 fc->time_warp_seq = cpu_to_le32(time_warp_seq);
1046
Eric W. Biederman05cb11c2013-01-31 02:56:19 -08001047 fc->uid = cpu_to_le32(from_kuid(&init_user_ns, uid));
1048 fc->gid = cpu_to_le32(from_kgid(&init_user_ns, gid));
Sage Weila8599bd2009-10-06 11:31:12 -07001049 fc->mode = cpu_to_le32(mode);
1050
Yan, Zhenge20d2582014-11-14 22:39:13 +08001051 p = fc + 1;
1052 /* flock buffer size */
1053 ceph_encode_32(&p, 0);
1054 /* inline version */
1055 ceph_encode_64(&p, inline_data ? 0 : CEPH_INLINE_NONE);
1056 /* inline data size */
1057 ceph_encode_32(&p, 0);
Yan, Zhenga2971c82015-06-10 11:09:32 +08001058 /* osd_epoch_barrier */
1059 ceph_encode_32(&p, 0);
1060 /* oldest_flush_tid */
1061 ceph_encode_64(&p, oldest_flush_tid);
Yan, Zhenge20d2582014-11-14 22:39:13 +08001062
Sage Weila8599bd2009-10-06 11:31:12 -07001063 fc->xattr_version = cpu_to_le64(xattr_version);
1064 if (xattrs_buf) {
1065 msg->middle = ceph_buffer_get(xattrs_buf);
1066 fc->xattr_len = cpu_to_le32(xattrs_buf->vec.iov_len);
1067 msg->hdr.middle_len = cpu_to_le32(xattrs_buf->vec.iov_len);
1068 }
1069
1070 ceph_con_send(&session->s_con, msg);
1071 return 0;
1072}
1073
1074/*
Sage Weila6369742010-02-22 13:59:00 -08001075 * Queue cap releases when an inode is dropped from our cache. Since
Sage Weilbe655592011-11-30 09:47:09 -08001076 * inode is about to be destroyed, there is no need for i_ceph_lock.
Sage Weila8599bd2009-10-06 11:31:12 -07001077 */
1078void ceph_queue_caps_release(struct inode *inode)
1079{
1080 struct ceph_inode_info *ci = ceph_inode(inode);
1081 struct rb_node *p;
1082
Sage Weila8599bd2009-10-06 11:31:12 -07001083 p = rb_first(&ci->i_caps);
1084 while (p) {
1085 struct ceph_cap *cap = rb_entry(p, struct ceph_cap, ci_node);
Sage Weila8599bd2009-10-06 11:31:12 -07001086 p = rb_next(p);
Yan, Zhenga096b092013-09-22 10:15:58 +08001087 __ceph_remove_cap(cap, true);
Sage Weila8599bd2009-10-06 11:31:12 -07001088 }
Sage Weila8599bd2009-10-06 11:31:12 -07001089}
1090
1091/*
1092 * Send a cap msg on the given inode. Update our caps state, then
Sage Weilbe655592011-11-30 09:47:09 -08001093 * drop i_ceph_lock and send the message.
Sage Weila8599bd2009-10-06 11:31:12 -07001094 *
1095 * Make note of max_size reported/requested from mds, revoked caps
1096 * that have now been implemented.
1097 *
1098 * Make half-hearted attempt ot to invalidate page cache if we are
1099 * dropping RDCACHE. Note that this will leave behind locked pages
1100 * that we'll then need to deal with elsewhere.
1101 *
1102 * Return non-zero if delayed release, or we experienced an error
1103 * such that the caller should requeue + retry later.
1104 *
Sage Weilbe655592011-11-30 09:47:09 -08001105 * called with i_ceph_lock, then drops it.
Sage Weila8599bd2009-10-06 11:31:12 -07001106 * caller should hold snap_rwsem (read), s_mutex.
1107 */
1108static int __send_cap(struct ceph_mds_client *mdsc, struct ceph_cap *cap,
Yan, Zheng553adfd2015-06-09 15:48:57 +08001109 int op, int used, int want, int retain, int flushing,
Yan, Zhenga2971c82015-06-10 11:09:32 +08001110 u64 flush_tid, u64 oldest_flush_tid)
Sage Weilbe655592011-11-30 09:47:09 -08001111 __releases(cap->ci->i_ceph_lock)
Sage Weila8599bd2009-10-06 11:31:12 -07001112{
1113 struct ceph_inode_info *ci = cap->ci;
1114 struct inode *inode = &ci->vfs_inode;
1115 u64 cap_id = cap->cap_id;
Sage Weil68c28322010-02-09 13:41:47 -08001116 int held, revoking, dropping, keep;
Sage Weila8599bd2009-10-06 11:31:12 -07001117 u64 seq, issue_seq, mseq, time_warp_seq, follows;
1118 u64 size, max_size;
1119 struct timespec mtime, atime;
1120 int wake = 0;
Al Viro5706b272011-07-26 04:52:22 -04001121 umode_t mode;
Eric W. Biederman05cb11c2013-01-31 02:56:19 -08001122 kuid_t uid;
1123 kgid_t gid;
Sage Weila8599bd2009-10-06 11:31:12 -07001124 struct ceph_mds_session *session;
1125 u64 xattr_version = 0;
Sage Weil082afec2010-08-22 15:16:41 -07001126 struct ceph_buffer *xattr_blob = NULL;
Sage Weila8599bd2009-10-06 11:31:12 -07001127 int delayed = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07001128 int ret;
Yan, Zhenge20d2582014-11-14 22:39:13 +08001129 bool inline_data;
Sage Weila8599bd2009-10-06 11:31:12 -07001130
Sage Weil68c28322010-02-09 13:41:47 -08001131 held = cap->issued | cap->implemented;
1132 revoking = cap->implemented & ~cap->issued;
1133 retain &= ~revoking;
1134 dropping = cap->issued & ~retain;
1135
Sage Weila8599bd2009-10-06 11:31:12 -07001136 dout("__send_cap %p cap %p session %p %s -> %s (revoking %s)\n",
1137 inode, cap, cap->session,
1138 ceph_cap_string(held), ceph_cap_string(held & retain),
1139 ceph_cap_string(revoking));
1140 BUG_ON((retain & CEPH_CAP_PIN) == 0);
1141
1142 session = cap->session;
1143
1144 /* don't release wanted unless we've waited a bit. */
1145 if ((ci->i_ceph_flags & CEPH_I_NODELAY) == 0 &&
1146 time_before(jiffies, ci->i_hold_caps_min)) {
1147 dout(" delaying issued %s -> %s, wanted %s -> %s on send\n",
1148 ceph_cap_string(cap->issued),
1149 ceph_cap_string(cap->issued & retain),
1150 ceph_cap_string(cap->mds_wanted),
1151 ceph_cap_string(want));
1152 want |= cap->mds_wanted;
1153 retain |= cap->issued;
1154 delayed = 1;
1155 }
1156 ci->i_ceph_flags &= ~(CEPH_I_NODELAY | CEPH_I_FLUSH);
1157
1158 cap->issued &= retain; /* drop bits we don't want */
1159 if (cap->implemented & ~cap->issued) {
1160 /*
1161 * Wake up any waiters on wanted -> needed transition.
1162 * This is due to the weird transition from buffered
1163 * to sync IO... we need to flush dirty pages _before_
1164 * allowing sync writes to avoid reordering.
1165 */
1166 wake = 1;
1167 }
1168 cap->implemented &= cap->issued | used;
1169 cap->mds_wanted = want;
1170
Yan, Zheng553adfd2015-06-09 15:48:57 +08001171 follows = flushing ? ci->i_head_snapc->seq : 0;
Sage Weila8599bd2009-10-06 11:31:12 -07001172
1173 keep = cap->implemented;
1174 seq = cap->seq;
1175 issue_seq = cap->issue_seq;
1176 mseq = cap->mseq;
1177 size = inode->i_size;
1178 ci->i_reported_size = size;
1179 max_size = ci->i_wanted_max_size;
1180 ci->i_requested_max_size = max_size;
1181 mtime = inode->i_mtime;
1182 atime = inode->i_atime;
1183 time_warp_seq = ci->i_time_warp_seq;
Sage Weila8599bd2009-10-06 11:31:12 -07001184 uid = inode->i_uid;
1185 gid = inode->i_gid;
1186 mode = inode->i_mode;
1187
Sage Weil082afec2010-08-22 15:16:41 -07001188 if (flushing & CEPH_CAP_XATTR_EXCL) {
Sage Weila8599bd2009-10-06 11:31:12 -07001189 __ceph_build_xattrs_blob(ci);
Sage Weil082afec2010-08-22 15:16:41 -07001190 xattr_blob = ci->i_xattrs.blob;
1191 xattr_version = ci->i_xattrs.version;
Sage Weila8599bd2009-10-06 11:31:12 -07001192 }
1193
Yan, Zhenge20d2582014-11-14 22:39:13 +08001194 inline_data = ci->i_inline_version != CEPH_INLINE_NONE;
1195
Sage Weilbe655592011-11-30 09:47:09 -08001196 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001197
Sage Weila8599bd2009-10-06 11:31:12 -07001198 ret = send_cap_msg(session, ceph_vino(inode).ino, cap_id,
Yan, Zhenga2971c82015-06-10 11:09:32 +08001199 op, keep, want, flushing, seq,
1200 flush_tid, oldest_flush_tid, issue_seq, mseq,
Sage Weila8599bd2009-10-06 11:31:12 -07001201 size, max_size, &mtime, &atime, time_warp_seq,
Sage Weil082afec2010-08-22 15:16:41 -07001202 uid, gid, mode, xattr_version, xattr_blob,
Yan, Zhenge20d2582014-11-14 22:39:13 +08001203 follows, inline_data);
Sage Weila8599bd2009-10-06 11:31:12 -07001204 if (ret < 0) {
1205 dout("error sending cap msg, must requeue %p\n", inode);
1206 delayed = 1;
1207 }
1208
1209 if (wake)
Yehuda Sadeh03066f22010-07-27 13:11:08 -07001210 wake_up_all(&ci->i_cap_wq);
Sage Weila8599bd2009-10-06 11:31:12 -07001211
1212 return delayed;
1213}
1214
1215/*
1216 * When a snapshot is taken, clients accumulate dirty metadata on
1217 * inodes with capabilities in ceph_cap_snaps to describe the file
1218 * state at the time the snapshot was taken. This must be flushed
1219 * asynchronously back to the MDS once sync writes complete and dirty
1220 * data is written out.
1221 *
Yan, Zhengaffbc192015-05-05 21:22:13 +08001222 * Unless @kick is true, skip cap_snaps that were already sent to
Sage Weile8351242010-09-17 08:03:08 -07001223 * the MDS (i.e., during this session).
1224 *
Sage Weilbe655592011-11-30 09:47:09 -08001225 * Called under i_ceph_lock. Takes s_mutex as needed.
Sage Weila8599bd2009-10-06 11:31:12 -07001226 */
1227void __ceph_flush_snaps(struct ceph_inode_info *ci,
Sage Weile8351242010-09-17 08:03:08 -07001228 struct ceph_mds_session **psession,
Yan, Zhengaffbc192015-05-05 21:22:13 +08001229 int kick)
Sage Weilbe655592011-11-30 09:47:09 -08001230 __releases(ci->i_ceph_lock)
1231 __acquires(ci->i_ceph_lock)
Sage Weila8599bd2009-10-06 11:31:12 -07001232{
1233 struct inode *inode = &ci->vfs_inode;
1234 int mds;
1235 struct ceph_cap_snap *capsnap;
1236 u32 mseq;
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001237 struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
Sage Weila8599bd2009-10-06 11:31:12 -07001238 struct ceph_mds_session *session = NULL; /* if session != NULL, we hold
1239 session->s_mutex */
1240 u64 next_follows = 0; /* keep track of how far we've gotten through the
1241 i_cap_snaps list, and skip these entries next time
1242 around to avoid an infinite loop */
1243
1244 if (psession)
1245 session = *psession;
1246
1247 dout("__flush_snaps %p\n", inode);
1248retry:
1249 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
1250 /* avoid an infiniute loop after retry */
1251 if (capsnap->follows < next_follows)
1252 continue;
1253 /*
1254 * we need to wait for sync writes to complete and for dirty
1255 * pages to be written out.
1256 */
1257 if (capsnap->dirty_pages || capsnap->writing)
Sage Weilcfc0bf62010-09-14 15:50:59 -07001258 break;
Sage Weila8599bd2009-10-06 11:31:12 -07001259
Yan, Zheng86056092015-05-01 16:57:16 +08001260 /* should be removed by ceph_try_drop_cap_snap() */
1261 BUG_ON(!capsnap->need_flush);
Sage Weil819ccbf2010-04-01 09:33:46 -07001262
Sage Weila8599bd2009-10-06 11:31:12 -07001263 /* pick mds, take s_mutex */
Sage Weilca81f3f2010-06-10 14:21:36 -07001264 if (ci->i_auth_cap == NULL) {
1265 dout("no auth cap (migrating?), doing nothing\n");
1266 goto out;
1267 }
Sage Weile8351242010-09-17 08:03:08 -07001268
1269 /* only flush each capsnap once */
Yan, Zhengaffbc192015-05-05 21:22:13 +08001270 if (!kick && !list_empty(&capsnap->flushing_item)) {
Sage Weile8351242010-09-17 08:03:08 -07001271 dout("already flushed %p, skipping\n", capsnap);
1272 continue;
1273 }
1274
Sage Weilca81f3f2010-06-10 14:21:36 -07001275 mds = ci->i_auth_cap->session->s_mds;
1276 mseq = ci->i_auth_cap->mseq;
1277
Sage Weila8599bd2009-10-06 11:31:12 -07001278 if (session && session->s_mds != mds) {
1279 dout("oops, wrong session %p mutex\n", session);
Yan, Zhengaffbc192015-05-05 21:22:13 +08001280 if (kick)
1281 goto out;
1282
Sage Weila8599bd2009-10-06 11:31:12 -07001283 mutex_unlock(&session->s_mutex);
1284 ceph_put_mds_session(session);
1285 session = NULL;
1286 }
1287 if (!session) {
Sage Weilbe655592011-11-30 09:47:09 -08001288 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001289 mutex_lock(&mdsc->mutex);
1290 session = __ceph_lookup_mds_session(mdsc, mds);
1291 mutex_unlock(&mdsc->mutex);
1292 if (session) {
1293 dout("inverting session/ino locks on %p\n",
1294 session);
1295 mutex_lock(&session->s_mutex);
1296 }
1297 /*
1298 * if session == NULL, we raced against a cap
Sage Weilca81f3f2010-06-10 14:21:36 -07001299 * deletion or migration. retry, and we'll
1300 * get a better @mds value next time.
Sage Weila8599bd2009-10-06 11:31:12 -07001301 */
Sage Weilbe655592011-11-30 09:47:09 -08001302 spin_lock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001303 goto retry;
1304 }
1305
Yan, Zheng553adfd2015-06-09 15:48:57 +08001306 spin_lock(&mdsc->cap_dirty_lock);
1307 capsnap->flush_tid = ++mdsc->last_cap_flush_tid;
1308 spin_unlock(&mdsc->cap_dirty_lock);
1309
Sage Weila8599bd2009-10-06 11:31:12 -07001310 atomic_inc(&capsnap->nref);
Yan, Zhengaffbc192015-05-05 21:22:13 +08001311 if (list_empty(&capsnap->flushing_item))
1312 list_add_tail(&capsnap->flushing_item,
1313 &session->s_cap_snaps_flushing);
Sage Weilbe655592011-11-30 09:47:09 -08001314 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001315
Sage Weilcfc0bf62010-09-14 15:50:59 -07001316 dout("flush_snaps %p cap_snap %p follows %lld tid %llu\n",
1317 inode, capsnap, capsnap->follows, capsnap->flush_tid);
Sage Weila8599bd2009-10-06 11:31:12 -07001318 send_cap_msg(session, ceph_vino(inode).ino, 0,
1319 CEPH_CAP_OP_FLUSHSNAP, capsnap->issued, 0,
Yan, Zhenga2971c82015-06-10 11:09:32 +08001320 capsnap->dirty, 0, capsnap->flush_tid, 0,
1321 0, mseq, capsnap->size, 0,
Sage Weila8599bd2009-10-06 11:31:12 -07001322 &capsnap->mtime, &capsnap->atime,
1323 capsnap->time_warp_seq,
1324 capsnap->uid, capsnap->gid, capsnap->mode,
Sage Weil4a625be2010-08-22 15:03:56 -07001325 capsnap->xattr_version, capsnap->xattr_blob,
Yan, Zhenge20d2582014-11-14 22:39:13 +08001326 capsnap->follows, capsnap->inline_data);
Sage Weila8599bd2009-10-06 11:31:12 -07001327
1328 next_follows = capsnap->follows + 1;
1329 ceph_put_cap_snap(capsnap);
1330
Sage Weilbe655592011-11-30 09:47:09 -08001331 spin_lock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001332 goto retry;
1333 }
1334
1335 /* we flushed them all; remove this inode from the queue */
1336 spin_lock(&mdsc->snap_flush_lock);
1337 list_del_init(&ci->i_snap_flush_item);
1338 spin_unlock(&mdsc->snap_flush_lock);
1339
Sage Weilca81f3f2010-06-10 14:21:36 -07001340out:
Sage Weila8599bd2009-10-06 11:31:12 -07001341 if (psession)
1342 *psession = session;
1343 else if (session) {
1344 mutex_unlock(&session->s_mutex);
1345 ceph_put_mds_session(session);
1346 }
1347}
1348
1349static void ceph_flush_snaps(struct ceph_inode_info *ci)
1350{
Sage Weilbe655592011-11-30 09:47:09 -08001351 spin_lock(&ci->i_ceph_lock);
Sage Weile8351242010-09-17 08:03:08 -07001352 __ceph_flush_snaps(ci, NULL, 0);
Sage Weilbe655592011-11-30 09:47:09 -08001353 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001354}
1355
1356/*
Sage Weilfca65b42011-05-04 11:33:47 -07001357 * Mark caps dirty. If inode is newly dirty, return the dirty flags.
1358 * Caller is then responsible for calling __mark_inode_dirty with the
1359 * returned flags value.
Sage Weil76e3b392009-10-15 18:13:53 -07001360 */
Yan, Zhengf66fd9f2015-06-10 17:26:13 +08001361int __ceph_mark_dirty_caps(struct ceph_inode_info *ci, int mask,
1362 struct ceph_cap_flush **pcf)
Sage Weil76e3b392009-10-15 18:13:53 -07001363{
Cheng Renquan640ef792010-03-26 17:40:33 +08001364 struct ceph_mds_client *mdsc =
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001365 ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc;
Sage Weil76e3b392009-10-15 18:13:53 -07001366 struct inode *inode = &ci->vfs_inode;
1367 int was = ci->i_dirty_caps;
1368 int dirty = 0;
1369
Yan, Zheng571ade32015-03-24 11:36:08 +08001370 if (!ci->i_auth_cap) {
1371 pr_warn("__mark_dirty_caps %p %llx mask %s, "
1372 "but no auth cap (session was closed?)\n",
1373 inode, ceph_ino(inode), ceph_cap_string(mask));
1374 return 0;
1375 }
1376
Sage Weil76e3b392009-10-15 18:13:53 -07001377 dout("__mark_dirty_caps %p %s dirty %s -> %s\n", &ci->vfs_inode,
1378 ceph_cap_string(mask), ceph_cap_string(was),
1379 ceph_cap_string(was | mask));
1380 ci->i_dirty_caps |= mask;
1381 if (was == 0) {
Yan, Zhengf66fd9f2015-06-10 17:26:13 +08001382 WARN_ON_ONCE(ci->i_prealloc_cap_flush);
1383 swap(ci->i_prealloc_cap_flush, *pcf);
1384
Yan, Zheng604d1b02015-05-01 17:49:16 +08001385 if (!ci->i_head_snapc) {
1386 WARN_ON_ONCE(!rwsem_is_locked(&mdsc->snap_rwsem));
Sage Weil7d8cb262010-08-24 08:44:16 -07001387 ci->i_head_snapc = ceph_get_snap_context(
1388 ci->i_snap_realm->cached_context);
Yan, Zheng604d1b02015-05-01 17:49:16 +08001389 }
Yan, Zheng06852352012-11-19 10:49:07 +08001390 dout(" inode %p now dirty snapc %p auth cap %p\n",
1391 &ci->vfs_inode, ci->i_head_snapc, ci->i_auth_cap);
Sage Weil76e3b392009-10-15 18:13:53 -07001392 BUG_ON(!list_empty(&ci->i_dirty_item));
1393 spin_lock(&mdsc->cap_dirty_lock);
Yan, Zheng11df2df2013-11-24 14:44:38 +08001394 list_add(&ci->i_dirty_item, &mdsc->cap_dirty);
Sage Weil76e3b392009-10-15 18:13:53 -07001395 spin_unlock(&mdsc->cap_dirty_lock);
1396 if (ci->i_flushing_caps == 0) {
Sage Weil3772d262011-05-03 09:28:08 -07001397 ihold(inode);
Sage Weil76e3b392009-10-15 18:13:53 -07001398 dirty |= I_DIRTY_SYNC;
1399 }
Yan, Zhengf66fd9f2015-06-10 17:26:13 +08001400 } else {
1401 WARN_ON_ONCE(!ci->i_prealloc_cap_flush);
Sage Weil76e3b392009-10-15 18:13:53 -07001402 }
1403 BUG_ON(list_empty(&ci->i_dirty_item));
1404 if (((was | ci->i_flushing_caps) & CEPH_CAP_FILE_BUFFER) &&
1405 (mask & CEPH_CAP_FILE_BUFFER))
1406 dirty |= I_DIRTY_DATASYNC;
Sage Weil76e3b392009-10-15 18:13:53 -07001407 __cap_delay_requeue(mdsc, ci);
Sage Weilfca65b42011-05-04 11:33:47 -07001408 return dirty;
Sage Weil76e3b392009-10-15 18:13:53 -07001409}
1410
Yan, Zheng553adfd2015-06-09 15:48:57 +08001411static void __add_cap_flushing_to_inode(struct ceph_inode_info *ci,
1412 struct ceph_cap_flush *cf)
1413{
1414 struct rb_node **p = &ci->i_cap_flush_tree.rb_node;
1415 struct rb_node *parent = NULL;
1416 struct ceph_cap_flush *other = NULL;
1417
1418 while (*p) {
1419 parent = *p;
1420 other = rb_entry(parent, struct ceph_cap_flush, i_node);
1421
1422 if (cf->tid < other->tid)
1423 p = &(*p)->rb_left;
1424 else if (cf->tid > other->tid)
1425 p = &(*p)->rb_right;
1426 else
1427 BUG();
1428 }
1429
1430 rb_link_node(&cf->i_node, parent, p);
1431 rb_insert_color(&cf->i_node, &ci->i_cap_flush_tree);
1432}
1433
Yan, Zheng8310b082015-06-09 17:20:12 +08001434static void __add_cap_flushing_to_mdsc(struct ceph_mds_client *mdsc,
1435 struct ceph_cap_flush *cf)
1436{
1437 struct rb_node **p = &mdsc->cap_flush_tree.rb_node;
1438 struct rb_node *parent = NULL;
1439 struct ceph_cap_flush *other = NULL;
1440
1441 while (*p) {
1442 parent = *p;
1443 other = rb_entry(parent, struct ceph_cap_flush, g_node);
1444
1445 if (cf->tid < other->tid)
1446 p = &(*p)->rb_left;
1447 else if (cf->tid > other->tid)
1448 p = &(*p)->rb_right;
1449 else
1450 BUG();
1451 }
1452
1453 rb_link_node(&cf->g_node, parent, p);
1454 rb_insert_color(&cf->g_node, &mdsc->cap_flush_tree);
1455}
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{
1470 struct rb_node *n = rb_first(&mdsc->cap_flush_tree);
1471 if (n) {
1472 struct ceph_cap_flush *cf =
1473 rb_entry(n, struct ceph_cap_flush, g_node);
1474 return cf->tid;
1475 }
1476 return 0;
1477}
1478
Sage Weil76e3b392009-10-15 18:13:53 -07001479/*
Sage Weila8599bd2009-10-06 11:31:12 -07001480 * Add dirty inode to the flushing list. Assigned a seq number so we
1481 * can wait for caps to flush without starving.
Sage Weilcdc35f92009-10-14 14:24:19 -07001482 *
Sage Weilbe655592011-11-30 09:47:09 -08001483 * Called under i_ceph_lock.
Sage Weila8599bd2009-10-06 11:31:12 -07001484 */
Sage Weilcdc35f92009-10-14 14:24:19 -07001485static int __mark_caps_flushing(struct inode *inode,
Yan, Zheng553adfd2015-06-09 15:48:57 +08001486 struct ceph_mds_session *session,
Yan, Zhenga2971c82015-06-10 11:09:32 +08001487 u64 *flush_tid, u64 *oldest_flush_tid)
Sage Weila8599bd2009-10-06 11:31:12 -07001488{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001489 struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
Sage Weila8599bd2009-10-06 11:31:12 -07001490 struct ceph_inode_info *ci = ceph_inode(inode);
Yan, Zhengf66fd9f2015-06-10 17:26:13 +08001491 struct ceph_cap_flush *cf = NULL;
Sage Weilcdc35f92009-10-14 14:24:19 -07001492 int flushing;
Sage Weil50b885b2009-12-01 14:12:07 -08001493
Sage Weilcdc35f92009-10-14 14:24:19 -07001494 BUG_ON(ci->i_dirty_caps == 0);
Sage Weila8599bd2009-10-06 11:31:12 -07001495 BUG_ON(list_empty(&ci->i_dirty_item));
Yan, Zhengf66fd9f2015-06-10 17:26:13 +08001496 BUG_ON(!ci->i_prealloc_cap_flush);
Sage Weilcdc35f92009-10-14 14:24:19 -07001497
1498 flushing = ci->i_dirty_caps;
1499 dout("__mark_caps_flushing flushing %s, flushing_caps %s -> %s\n",
1500 ceph_cap_string(flushing),
1501 ceph_cap_string(ci->i_flushing_caps),
1502 ceph_cap_string(ci->i_flushing_caps | flushing));
1503 ci->i_flushing_caps |= flushing;
1504 ci->i_dirty_caps = 0;
Sage Weilafcdaea2009-10-14 14:27:38 -07001505 dout(" inode %p now !dirty\n", inode);
Sage Weilcdc35f92009-10-14 14:24:19 -07001506
Yan, Zhengf66fd9f2015-06-10 17:26:13 +08001507 swap(cf, ci->i_prealloc_cap_flush);
Yan, Zheng553adfd2015-06-09 15:48:57 +08001508 cf->caps = flushing;
1509
Sage Weila8599bd2009-10-06 11:31:12 -07001510 spin_lock(&mdsc->cap_dirty_lock);
Sage Weilafcdaea2009-10-14 14:27:38 -07001511 list_del_init(&ci->i_dirty_item);
1512
Yan, Zheng553adfd2015-06-09 15:48:57 +08001513 cf->tid = ++mdsc->last_cap_flush_tid;
Yan, Zheng8310b082015-06-09 17:20:12 +08001514 __add_cap_flushing_to_mdsc(mdsc, cf);
Yan, Zhenga2971c82015-06-10 11:09:32 +08001515 *oldest_flush_tid = __get_oldest_flush_tid(mdsc);
Yan, Zheng553adfd2015-06-09 15:48:57 +08001516
Sage Weila8599bd2009-10-06 11:31:12 -07001517 if (list_empty(&ci->i_flushing_item)) {
1518 list_add_tail(&ci->i_flushing_item, &session->s_cap_flushing);
1519 mdsc->num_cap_flushing++;
Yan, Zheng8310b082015-06-09 17:20:12 +08001520 dout(" inode %p now flushing tid %llu\n", inode, cf->tid);
Sage Weilafcdaea2009-10-14 14:27:38 -07001521 } else {
1522 list_move_tail(&ci->i_flushing_item, &session->s_cap_flushing);
Yan, Zheng8310b082015-06-09 17:20:12 +08001523 dout(" inode %p now flushing (more) tid %llu\n",
1524 inode, cf->tid);
Sage Weila8599bd2009-10-06 11:31:12 -07001525 }
1526 spin_unlock(&mdsc->cap_dirty_lock);
Sage Weilcdc35f92009-10-14 14:24:19 -07001527
Yan, Zheng553adfd2015-06-09 15:48:57 +08001528 __add_cap_flushing_to_inode(ci, cf);
1529
1530 *flush_tid = cf->tid;
Sage Weilcdc35f92009-10-14 14:24:19 -07001531 return flushing;
Sage Weila8599bd2009-10-06 11:31:12 -07001532}
1533
1534/*
Sage Weil5ecad6f2010-02-17 10:43:37 -08001535 * try to invalidate mapping pages without blocking.
1536 */
Sage Weil5ecad6f2010-02-17 10:43:37 -08001537static int try_nonblocking_invalidate(struct inode *inode)
1538{
1539 struct ceph_inode_info *ci = ceph_inode(inode);
1540 u32 invalidating_gen = ci->i_rdcache_gen;
1541
Sage Weilbe655592011-11-30 09:47:09 -08001542 spin_unlock(&ci->i_ceph_lock);
Sage Weil5ecad6f2010-02-17 10:43:37 -08001543 invalidate_mapping_pages(&inode->i_data, 0, -1);
Sage Weilbe655592011-11-30 09:47:09 -08001544 spin_lock(&ci->i_ceph_lock);
Sage Weil5ecad6f2010-02-17 10:43:37 -08001545
Sage Weil18a38192010-09-17 10:46:44 -07001546 if (inode->i_data.nrpages == 0 &&
Sage Weil5ecad6f2010-02-17 10:43:37 -08001547 invalidating_gen == ci->i_rdcache_gen) {
1548 /* success. */
1549 dout("try_nonblocking_invalidate %p success\n", inode);
Sage Weilcd045cb2010-11-04 11:05:05 -07001550 /* save any racing async invalidate some trouble */
1551 ci->i_rdcache_revoking = ci->i_rdcache_gen - 1;
Sage Weil5ecad6f2010-02-17 10:43:37 -08001552 return 0;
1553 }
1554 dout("try_nonblocking_invalidate %p failed\n", inode);
1555 return -1;
1556}
1557
1558/*
Sage Weila8599bd2009-10-06 11:31:12 -07001559 * Swiss army knife function to examine currently used and wanted
1560 * versus held caps. Release, flush, ack revoked caps to mds as
1561 * appropriate.
1562 *
1563 * CHECK_CAPS_NODELAY - caller is delayed work and we should not delay
1564 * cap release further.
1565 * CHECK_CAPS_AUTHONLY - we should only check the auth cap
1566 * CHECK_CAPS_FLUSH - we should flush any dirty caps immediately, without
1567 * further delay.
1568 */
1569void ceph_check_caps(struct ceph_inode_info *ci, int flags,
1570 struct ceph_mds_session *session)
1571{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001572 struct ceph_fs_client *fsc = ceph_inode_to_client(&ci->vfs_inode);
1573 struct ceph_mds_client *mdsc = fsc->mdsc;
Sage Weila8599bd2009-10-06 11:31:12 -07001574 struct inode *inode = &ci->vfs_inode;
1575 struct ceph_cap *cap;
Yan, Zhenga2971c82015-06-10 11:09:32 +08001576 u64 flush_tid, oldest_flush_tid;
Yan, Zheng395c3122013-01-04 14:37:57 +08001577 int file_wanted, used, cap_used;
Sage Weila8599bd2009-10-06 11:31:12 -07001578 int took_snap_rwsem = 0; /* true if mdsc->snap_rwsem held */
Sage Weilcbd03632010-02-09 13:41:18 -08001579 int issued, implemented, want, retain, revoking, flushing = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07001580 int mds = -1; /* keep track of how far we've gone through i_caps list
1581 to avoid an infinite loop on retry */
1582 struct rb_node *p;
1583 int tried_invalidate = 0;
1584 int delayed = 0, sent = 0, force_requeue = 0, num;
Sage Weilcbd03632010-02-09 13:41:18 -08001585 int queue_invalidate = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07001586 int is_delayed = flags & CHECK_CAPS_NODELAY;
1587
1588 /* if we are unmounting, flush any unused caps immediately. */
1589 if (mdsc->stopping)
1590 is_delayed = 1;
1591
Sage Weilbe655592011-11-30 09:47:09 -08001592 spin_lock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001593
1594 if (ci->i_ceph_flags & CEPH_I_FLUSH)
1595 flags |= CHECK_CAPS_FLUSH;
1596
1597 /* flush snaps first time around only */
1598 if (!list_empty(&ci->i_cap_snaps))
Sage Weile8351242010-09-17 08:03:08 -07001599 __ceph_flush_snaps(ci, &session, 0);
Sage Weila8599bd2009-10-06 11:31:12 -07001600 goto retry_locked;
1601retry:
Sage Weilbe655592011-11-30 09:47:09 -08001602 spin_lock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001603retry_locked:
1604 file_wanted = __ceph_caps_file_wanted(ci);
1605 used = __ceph_caps_used(ci);
Sage Weilcbd03632010-02-09 13:41:18 -08001606 issued = __ceph_caps_issued(ci, &implemented);
1607 revoking = implemented & ~issued;
Sage Weila8599bd2009-10-06 11:31:12 -07001608
Yan, Zheng41445992015-05-25 17:36:42 +08001609 want = file_wanted;
1610 retain = file_wanted | used | CEPH_CAP_PIN;
Sage Weila8599bd2009-10-06 11:31:12 -07001611 if (!mdsc->stopping && inode->i_nlink > 0) {
Yan, Zheng41445992015-05-25 17:36:42 +08001612 if (file_wanted) {
Sage Weila8599bd2009-10-06 11:31:12 -07001613 retain |= CEPH_CAP_ANY; /* be greedy */
Yan, Zheng32ec4392015-03-26 19:06:00 +08001614 } else if (S_ISDIR(inode->i_mode) &&
1615 (issued & CEPH_CAP_FILE_SHARED) &&
1616 __ceph_dir_is_complete(ci)) {
1617 /*
1618 * If a directory is complete, we want to keep
1619 * the exclusive cap. So that MDS does not end up
1620 * revoking the shared cap on every create/unlink
1621 * operation.
1622 */
1623 want = CEPH_CAP_ANY_SHARED | CEPH_CAP_FILE_EXCL;
1624 retain |= want;
Sage Weila8599bd2009-10-06 11:31:12 -07001625 } else {
Yan, Zheng32ec4392015-03-26 19:06:00 +08001626
Sage Weila8599bd2009-10-06 11:31:12 -07001627 retain |= CEPH_CAP_ANY_SHARED;
1628 /*
1629 * keep RD only if we didn't have the file open RW,
1630 * because then the mds would revoke it anyway to
1631 * journal max_size=0.
1632 */
1633 if (ci->i_max_size == 0)
1634 retain |= CEPH_CAP_ANY_RD;
1635 }
1636 }
1637
1638 dout("check_caps %p file_want %s used %s dirty %s flushing %s"
Sage Weilcbd03632010-02-09 13:41:18 -08001639 " issued %s revoking %s retain %s %s%s%s\n", inode,
Sage Weila8599bd2009-10-06 11:31:12 -07001640 ceph_cap_string(file_wanted),
1641 ceph_cap_string(used), ceph_cap_string(ci->i_dirty_caps),
1642 ceph_cap_string(ci->i_flushing_caps),
Sage Weilcbd03632010-02-09 13:41:18 -08001643 ceph_cap_string(issued), ceph_cap_string(revoking),
Sage Weila8599bd2009-10-06 11:31:12 -07001644 ceph_cap_string(retain),
1645 (flags & CHECK_CAPS_AUTHONLY) ? " AUTHONLY" : "",
1646 (flags & CHECK_CAPS_NODELAY) ? " NODELAY" : "",
1647 (flags & CHECK_CAPS_FLUSH) ? " FLUSH" : "");
1648
1649 /*
1650 * If we no longer need to hold onto old our caps, and we may
1651 * have cached pages, but don't want them, then try to invalidate.
1652 * If we fail, it's because pages are locked.... try again later.
1653 */
1654 if ((!is_delayed || mdsc->stopping) &&
Yan, Zhengfdd4e152015-06-16 20:48:56 +08001655 !S_ISDIR(inode->i_mode) && /* ignore readdir cache */
1656 ci->i_wrbuffer_ref == 0 && /* no dirty pages... */
1657 inode->i_data.nrpages && /* have cached pages */
1658 (file_wanted == 0 || /* no open files */
Sage Weil29625072010-05-27 10:40:43 -07001659 (revoking & (CEPH_CAP_FILE_CACHE|
1660 CEPH_CAP_FILE_LAZYIO))) && /* or revoking cache */
Sage Weila8599bd2009-10-06 11:31:12 -07001661 !tried_invalidate) {
Sage Weila8599bd2009-10-06 11:31:12 -07001662 dout("check_caps trying to invalidate on %p\n", inode);
Sage Weil5ecad6f2010-02-17 10:43:37 -08001663 if (try_nonblocking_invalidate(inode) < 0) {
Sage Weil29625072010-05-27 10:40:43 -07001664 if (revoking & (CEPH_CAP_FILE_CACHE|
1665 CEPH_CAP_FILE_LAZYIO)) {
Sage Weil5ecad6f2010-02-17 10:43:37 -08001666 dout("check_caps queuing invalidate\n");
1667 queue_invalidate = 1;
1668 ci->i_rdcache_revoking = ci->i_rdcache_gen;
1669 } else {
1670 dout("check_caps failed to invalidate pages\n");
1671 /* we failed to invalidate pages. check these
1672 caps again later. */
1673 force_requeue = 1;
1674 __cap_set_timeouts(mdsc, ci);
1675 }
Sage Weila8599bd2009-10-06 11:31:12 -07001676 }
1677 tried_invalidate = 1;
1678 goto retry_locked;
1679 }
1680
1681 num = 0;
1682 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
1683 cap = rb_entry(p, struct ceph_cap, ci_node);
1684 num++;
1685
1686 /* avoid looping forever */
1687 if (mds >= cap->mds ||
1688 ((flags & CHECK_CAPS_AUTHONLY) && cap != ci->i_auth_cap))
1689 continue;
1690
1691 /* NOTE: no side-effects allowed, until we take s_mutex */
1692
Yan, Zheng395c3122013-01-04 14:37:57 +08001693 cap_used = used;
1694 if (ci->i_auth_cap && cap != ci->i_auth_cap)
1695 cap_used &= ~ci->i_auth_cap->issued;
1696
Sage Weila8599bd2009-10-06 11:31:12 -07001697 revoking = cap->implemented & ~cap->issued;
Yan, Zheng395c3122013-01-04 14:37:57 +08001698 dout(" mds%d cap %p used %s issued %s implemented %s revoking %s\n",
Sage Weil088b3f52011-01-18 08:56:01 -08001699 cap->mds, cap, ceph_cap_string(cap->issued),
Yan, Zheng395c3122013-01-04 14:37:57 +08001700 ceph_cap_string(cap_used),
Sage Weil088b3f52011-01-18 08:56:01 -08001701 ceph_cap_string(cap->implemented),
1702 ceph_cap_string(revoking));
Sage Weila8599bd2009-10-06 11:31:12 -07001703
1704 if (cap == ci->i_auth_cap &&
1705 (cap->issued & CEPH_CAP_FILE_WR)) {
1706 /* request larger max_size from MDS? */
1707 if (ci->i_wanted_max_size > ci->i_max_size &&
1708 ci->i_wanted_max_size > ci->i_requested_max_size) {
1709 dout("requesting new max_size\n");
1710 goto ack;
1711 }
1712
1713 /* approaching file_max? */
1714 if ((inode->i_size << 1) >= ci->i_max_size &&
1715 (ci->i_reported_size << 1) < ci->i_max_size) {
1716 dout("i_size approaching max_size\n");
1717 goto ack;
1718 }
1719 }
1720 /* flush anything dirty? */
1721 if (cap == ci->i_auth_cap && (flags & CHECK_CAPS_FLUSH) &&
1722 ci->i_dirty_caps) {
1723 dout("flushing dirty caps\n");
1724 goto ack;
1725 }
1726
1727 /* completed revocation? going down and there are no caps? */
Yan, Zheng395c3122013-01-04 14:37:57 +08001728 if (revoking && (revoking & cap_used) == 0) {
Sage Weila8599bd2009-10-06 11:31:12 -07001729 dout("completed revocation of %s\n",
1730 ceph_cap_string(cap->implemented & ~cap->issued));
1731 goto ack;
1732 }
1733
1734 /* want more caps from mds? */
1735 if (want & ~(cap->mds_wanted | cap->issued))
1736 goto ack;
1737
1738 /* things we might delay */
1739 if ((cap->issued & ~retain) == 0 &&
1740 cap->mds_wanted == want)
1741 continue; /* nope, all good */
1742
1743 if (is_delayed)
1744 goto ack;
1745
1746 /* delay? */
1747 if ((ci->i_ceph_flags & CEPH_I_NODELAY) == 0 &&
1748 time_before(jiffies, ci->i_hold_caps_max)) {
1749 dout(" delaying issued %s -> %s, wanted %s -> %s\n",
1750 ceph_cap_string(cap->issued),
1751 ceph_cap_string(cap->issued & retain),
1752 ceph_cap_string(cap->mds_wanted),
1753 ceph_cap_string(want));
1754 delayed++;
1755 continue;
1756 }
1757
1758ack:
Sage Weile9964c12010-03-01 15:16:56 -08001759 if (ci->i_ceph_flags & CEPH_I_NOFLUSH) {
1760 dout(" skipping %p I_NOFLUSH set\n", inode);
1761 continue;
1762 }
1763
Sage Weila8599bd2009-10-06 11:31:12 -07001764 if (session && session != cap->session) {
1765 dout("oops, wrong session %p mutex\n", session);
1766 mutex_unlock(&session->s_mutex);
1767 session = NULL;
1768 }
1769 if (!session) {
1770 session = cap->session;
1771 if (mutex_trylock(&session->s_mutex) == 0) {
1772 dout("inverting session/ino locks on %p\n",
1773 session);
Sage Weilbe655592011-11-30 09:47:09 -08001774 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001775 if (took_snap_rwsem) {
1776 up_read(&mdsc->snap_rwsem);
1777 took_snap_rwsem = 0;
1778 }
1779 mutex_lock(&session->s_mutex);
1780 goto retry;
1781 }
1782 }
1783 /* take snap_rwsem after session mutex */
1784 if (!took_snap_rwsem) {
1785 if (down_read_trylock(&mdsc->snap_rwsem) == 0) {
1786 dout("inverting snap/in locks on %p\n",
1787 inode);
Sage Weilbe655592011-11-30 09:47:09 -08001788 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001789 down_read(&mdsc->snap_rwsem);
1790 took_snap_rwsem = 1;
1791 goto retry;
1792 }
1793 took_snap_rwsem = 1;
1794 }
1795
Yan, Zheng553adfd2015-06-09 15:48:57 +08001796 if (cap == ci->i_auth_cap && ci->i_dirty_caps) {
1797 flushing = __mark_caps_flushing(inode, session,
Yan, Zhenga2971c82015-06-10 11:09:32 +08001798 &flush_tid,
1799 &oldest_flush_tid);
Yan, Zheng553adfd2015-06-09 15:48:57 +08001800 } else {
Sage Weil24be0c42011-01-18 08:48:06 -08001801 flushing = 0;
Yan, Zheng553adfd2015-06-09 15:48:57 +08001802 flush_tid = 0;
Yan, Zhenga2971c82015-06-10 11:09:32 +08001803 spin_lock(&mdsc->cap_dirty_lock);
1804 oldest_flush_tid = __get_oldest_flush_tid(mdsc);
1805 spin_unlock(&mdsc->cap_dirty_lock);
Yan, Zheng553adfd2015-06-09 15:48:57 +08001806 }
Sage Weila8599bd2009-10-06 11:31:12 -07001807
1808 mds = cap->mds; /* remember mds, so we don't repeat */
1809 sent++;
1810
Sage Weilbe655592011-11-30 09:47:09 -08001811 /* __send_cap drops i_ceph_lock */
Yan, Zheng395c3122013-01-04 14:37:57 +08001812 delayed += __send_cap(mdsc, cap, CEPH_CAP_OP_UPDATE, cap_used,
Yan, Zhenga2971c82015-06-10 11:09:32 +08001813 want, retain, flushing,
1814 flush_tid, oldest_flush_tid);
Sage Weilbe655592011-11-30 09:47:09 -08001815 goto retry; /* retake i_ceph_lock and restart our cap scan. */
Sage Weila8599bd2009-10-06 11:31:12 -07001816 }
1817
1818 /*
1819 * Reschedule delayed caps release if we delayed anything,
1820 * otherwise cancel.
1821 */
1822 if (delayed && is_delayed)
1823 force_requeue = 1; /* __send_cap delayed release; requeue */
1824 if (!delayed && !is_delayed)
1825 __cap_delay_cancel(mdsc, ci);
1826 else if (!is_delayed || force_requeue)
1827 __cap_delay_requeue(mdsc, ci);
1828
Sage Weilbe655592011-11-30 09:47:09 -08001829 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001830
Sage Weilcbd03632010-02-09 13:41:18 -08001831 if (queue_invalidate)
Sage Weil3c6f6b72010-02-09 15:24:44 -08001832 ceph_queue_invalidate(inode);
Sage Weilcbd03632010-02-09 13:41:18 -08001833
Sage Weilcdc2ce02010-03-16 13:39:28 -07001834 if (session)
Sage Weila8599bd2009-10-06 11:31:12 -07001835 mutex_unlock(&session->s_mutex);
1836 if (took_snap_rwsem)
1837 up_read(&mdsc->snap_rwsem);
1838}
1839
1840/*
Sage Weila8599bd2009-10-06 11:31:12 -07001841 * Try to flush dirty caps back to the auth mds.
1842 */
Yan, Zheng553adfd2015-06-09 15:48:57 +08001843static int try_flush_caps(struct inode *inode, u64 *ptid)
Sage Weila8599bd2009-10-06 11:31:12 -07001844{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001845 struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
Sage Weila8599bd2009-10-06 11:31:12 -07001846 struct ceph_inode_info *ci = ceph_inode(inode);
Yan, Zheng4fe59782013-10-31 16:44:14 +08001847 struct ceph_mds_session *session = NULL;
Yan, Zheng89b52fe2015-05-27 09:59:48 +08001848 int flushing = 0;
Yan, Zhenga2971c82015-06-10 11:09:32 +08001849 u64 flush_tid = 0, oldest_flush_tid = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07001850
1851retry:
Sage Weilbe655592011-11-30 09:47:09 -08001852 spin_lock(&ci->i_ceph_lock);
Sage Weile9964c12010-03-01 15:16:56 -08001853 if (ci->i_ceph_flags & CEPH_I_NOFLUSH) {
1854 dout("try_flush_caps skipping %p I_NOFLUSH set\n", inode);
1855 goto out;
1856 }
Sage Weila8599bd2009-10-06 11:31:12 -07001857 if (ci->i_dirty_caps && ci->i_auth_cap) {
1858 struct ceph_cap *cap = ci->i_auth_cap;
1859 int used = __ceph_caps_used(ci);
1860 int want = __ceph_caps_wanted(ci);
1861 int delayed;
1862
Yan, Zheng4fe59782013-10-31 16:44:14 +08001863 if (!session || session != cap->session) {
Sage Weilbe655592011-11-30 09:47:09 -08001864 spin_unlock(&ci->i_ceph_lock);
Yan, Zheng4fe59782013-10-31 16:44:14 +08001865 if (session)
1866 mutex_unlock(&session->s_mutex);
Sage Weila8599bd2009-10-06 11:31:12 -07001867 session = cap->session;
1868 mutex_lock(&session->s_mutex);
1869 goto retry;
1870 }
Sage Weila8599bd2009-10-06 11:31:12 -07001871 if (cap->session->s_state < CEPH_MDS_SESSION_OPEN)
1872 goto out;
1873
Yan, Zhenga2971c82015-06-10 11:09:32 +08001874 flushing = __mark_caps_flushing(inode, session, &flush_tid,
1875 &oldest_flush_tid);
Sage Weila8599bd2009-10-06 11:31:12 -07001876
Sage Weilbe655592011-11-30 09:47:09 -08001877 /* __send_cap drops i_ceph_lock */
Sage Weila8599bd2009-10-06 11:31:12 -07001878 delayed = __send_cap(mdsc, cap, CEPH_CAP_OP_FLUSH, used, want,
Yan, Zheng553adfd2015-06-09 15:48:57 +08001879 (cap->issued | cap->implemented),
Yan, Zhenga2971c82015-06-10 11:09:32 +08001880 flushing, flush_tid, oldest_flush_tid);
Sage Weila8599bd2009-10-06 11:31:12 -07001881
Yan, Zheng553adfd2015-06-09 15:48:57 +08001882 if (delayed) {
1883 spin_lock(&ci->i_ceph_lock);
Yan, Zheng89b52fe2015-05-27 09:59:48 +08001884 __cap_delay_requeue(mdsc, ci);
Yan, Zheng553adfd2015-06-09 15:48:57 +08001885 spin_unlock(&ci->i_ceph_lock);
1886 }
1887 } else {
1888 struct rb_node *n = rb_last(&ci->i_cap_flush_tree);
1889 if (n) {
1890 struct ceph_cap_flush *cf =
1891 rb_entry(n, struct ceph_cap_flush, i_node);
1892 flush_tid = cf->tid;
1893 }
1894 flushing = ci->i_flushing_caps;
1895 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001896 }
1897out:
Yan, Zheng4fe59782013-10-31 16:44:14 +08001898 if (session)
Sage Weila8599bd2009-10-06 11:31:12 -07001899 mutex_unlock(&session->s_mutex);
Yan, Zheng553adfd2015-06-09 15:48:57 +08001900
1901 *ptid = flush_tid;
Sage Weila8599bd2009-10-06 11:31:12 -07001902 return flushing;
1903}
1904
1905/*
1906 * Return true if we've flushed caps through the given flush_tid.
1907 */
Yan, Zheng553adfd2015-06-09 15:48:57 +08001908static int caps_are_flushed(struct inode *inode, u64 flush_tid)
Sage Weila8599bd2009-10-06 11:31:12 -07001909{
1910 struct ceph_inode_info *ci = ceph_inode(inode);
Yan, Zheng553adfd2015-06-09 15:48:57 +08001911 struct ceph_cap_flush *cf;
1912 struct rb_node *n;
1913 int ret = 1;
Sage Weila8599bd2009-10-06 11:31:12 -07001914
Sage Weilbe655592011-11-30 09:47:09 -08001915 spin_lock(&ci->i_ceph_lock);
Yan, Zheng553adfd2015-06-09 15:48:57 +08001916 n = rb_first(&ci->i_cap_flush_tree);
1917 if (n) {
1918 cf = rb_entry(n, struct ceph_cap_flush, i_node);
1919 if (cf->tid <= flush_tid)
Sage Weila8599bd2009-10-06 11:31:12 -07001920 ret = 0;
Yan, Zheng89b52fe2015-05-27 09:59:48 +08001921 }
Sage Weilbe655592011-11-30 09:47:09 -08001922 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001923 return ret;
1924}
1925
1926/*
1927 * Wait on any unsafe replies for the given inode. First wait on the
1928 * newest request, and make that the upper bound. Then, if there are
1929 * more requests, keep waiting on the oldest as long as it is still older
1930 * than the original request.
1931 */
1932static void sync_write_wait(struct inode *inode)
1933{
1934 struct ceph_inode_info *ci = ceph_inode(inode);
1935 struct list_head *head = &ci->i_unsafe_writes;
1936 struct ceph_osd_request *req;
1937 u64 last_tid;
1938
Yan, Zhengda819c82015-05-27 11:19:34 +08001939 if (!S_ISREG(inode->i_mode))
1940 return;
1941
Sage Weila8599bd2009-10-06 11:31:12 -07001942 spin_lock(&ci->i_unsafe_lock);
1943 if (list_empty(head))
1944 goto out;
1945
1946 /* set upper bound as _last_ entry in chain */
Yan, Zhengda819c82015-05-27 11:19:34 +08001947 req = list_last_entry(head, struct ceph_osd_request,
1948 r_unsafe_item);
Sage Weila8599bd2009-10-06 11:31:12 -07001949 last_tid = req->r_tid;
1950
1951 do {
1952 ceph_osdc_get_request(req);
1953 spin_unlock(&ci->i_unsafe_lock);
1954 dout("sync_write_wait on tid %llu (until %llu)\n",
1955 req->r_tid, last_tid);
1956 wait_for_completion(&req->r_safe_completion);
1957 spin_lock(&ci->i_unsafe_lock);
1958 ceph_osdc_put_request(req);
1959
1960 /*
1961 * from here on look at first entry in chain, since we
1962 * only want to wait for anything older than last_tid
1963 */
1964 if (list_empty(head))
1965 break;
Yan, Zhengda819c82015-05-27 11:19:34 +08001966 req = list_first_entry(head, struct ceph_osd_request,
1967 r_unsafe_item);
Sage Weila8599bd2009-10-06 11:31:12 -07001968 } while (req->r_tid < last_tid);
1969out:
1970 spin_unlock(&ci->i_unsafe_lock);
1971}
1972
Yan, Zhengda819c82015-05-27 11:19:34 +08001973/*
1974 * wait for any uncommitted directory operations to commit.
1975 */
1976static int unsafe_dirop_wait(struct inode *inode)
1977{
1978 struct ceph_inode_info *ci = ceph_inode(inode);
1979 struct list_head *head = &ci->i_unsafe_dirops;
1980 struct ceph_mds_request *req;
1981 u64 last_tid;
1982 int ret = 0;
1983
1984 if (!S_ISDIR(inode->i_mode))
1985 return 0;
1986
1987 spin_lock(&ci->i_unsafe_lock);
1988 if (list_empty(head))
1989 goto out;
1990
1991 req = list_last_entry(head, struct ceph_mds_request,
1992 r_unsafe_dir_item);
1993 last_tid = req->r_tid;
1994
1995 do {
1996 ceph_mdsc_get_request(req);
1997 spin_unlock(&ci->i_unsafe_lock);
1998
1999 dout("unsafe_dirop_wait %p wait on tid %llu (until %llu)\n",
2000 inode, req->r_tid, last_tid);
2001 ret = !wait_for_completion_timeout(&req->r_safe_completion,
2002 ceph_timeout_jiffies(req->r_timeout));
2003 if (ret)
2004 ret = -EIO; /* timed out */
2005
2006 ceph_mdsc_put_request(req);
2007
2008 spin_lock(&ci->i_unsafe_lock);
2009 if (ret || list_empty(head))
2010 break;
2011 req = list_first_entry(head, struct ceph_mds_request,
2012 r_unsafe_dir_item);
2013 } while (req->r_tid < last_tid);
2014out:
2015 spin_unlock(&ci->i_unsafe_lock);
2016 return ret;
2017}
2018
Josef Bacik02c24a82011-07-16 20:44:56 -04002019int ceph_fsync(struct file *file, loff_t start, loff_t end, int datasync)
Sage Weila8599bd2009-10-06 11:31:12 -07002020{
Christoph Hellwig7ea80852010-05-26 17:53:25 +02002021 struct inode *inode = file->f_mapping->host;
Sage Weila8599bd2009-10-06 11:31:12 -07002022 struct ceph_inode_info *ci = ceph_inode(inode);
Yan, Zheng553adfd2015-06-09 15:48:57 +08002023 u64 flush_tid;
Sage Weila8599bd2009-10-06 11:31:12 -07002024 int ret;
2025 int dirty;
2026
2027 dout("fsync %p%s\n", inode, datasync ? " datasync" : "");
2028 sync_write_wait(inode);
2029
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
Josef Bacik02c24a82011-07-16 20:44:56 -04002037 mutex_lock(&inode->i_mutex);
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, Zhengda819c82015-05-27 11:19:34 +08002042 ret = unsafe_dirop_wait(inode);
2043
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 }
Josef Bacik02c24a82011-07-16 20:44:56 -04002053 mutex_unlock(&inode->i_mutex);
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
2091/*
2092 * After a recovering MDS goes active, we need to resend any caps
2093 * we were flushing.
2094 *
2095 * Caller holds session->s_mutex.
2096 */
2097static void kick_flushing_capsnaps(struct ceph_mds_client *mdsc,
2098 struct ceph_mds_session *session)
2099{
2100 struct ceph_cap_snap *capsnap;
2101
2102 dout("kick_flushing_capsnaps mds%d\n", session->s_mds);
2103 list_for_each_entry(capsnap, &session->s_cap_snaps_flushing,
2104 flushing_item) {
2105 struct ceph_inode_info *ci = capsnap->ci;
2106 struct inode *inode = &ci->vfs_inode;
2107 struct ceph_cap *cap;
2108
Sage Weilbe655592011-11-30 09:47:09 -08002109 spin_lock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07002110 cap = ci->i_auth_cap;
2111 if (cap && cap->session == session) {
2112 dout("kick_flushing_caps %p cap %p capsnap %p\n", inode,
2113 cap, capsnap);
Sage Weile8351242010-09-17 08:03:08 -07002114 __ceph_flush_snaps(ci, &session, 1);
Sage Weila8599bd2009-10-06 11:31:12 -07002115 } else {
2116 pr_err("%p auth cap %p not mds%d ???\n", inode,
2117 cap, session->s_mds);
Sage Weila8599bd2009-10-06 11:31:12 -07002118 }
Sage Weilbe655592011-11-30 09:47:09 -08002119 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07002120 }
2121}
2122
Yan, Zheng553adfd2015-06-09 15:48:57 +08002123static int __kick_flushing_caps(struct ceph_mds_client *mdsc,
2124 struct ceph_mds_session *session,
Yan, Zhengfc927cd2015-07-20 09:50:58 +08002125 struct ceph_inode_info *ci)
Yan, Zheng553adfd2015-06-09 15:48:57 +08002126{
2127 struct inode *inode = &ci->vfs_inode;
2128 struct ceph_cap *cap;
2129 struct ceph_cap_flush *cf;
2130 struct rb_node *n;
2131 int delayed = 0;
2132 u64 first_tid = 0;
Yan, Zhenga2971c82015-06-10 11:09:32 +08002133 u64 oldest_flush_tid;
2134
2135 spin_lock(&mdsc->cap_dirty_lock);
2136 oldest_flush_tid = __get_oldest_flush_tid(mdsc);
2137 spin_unlock(&mdsc->cap_dirty_lock);
Yan, Zheng553adfd2015-06-09 15:48:57 +08002138
2139 while (true) {
2140 spin_lock(&ci->i_ceph_lock);
2141 cap = ci->i_auth_cap;
2142 if (!(cap && cap->session == session)) {
2143 pr_err("%p auth cap %p not mds%d ???\n", inode,
2144 cap, session->s_mds);
2145 spin_unlock(&ci->i_ceph_lock);
2146 break;
2147 }
2148
2149 for (n = rb_first(&ci->i_cap_flush_tree); n; n = rb_next(n)) {
2150 cf = rb_entry(n, struct ceph_cap_flush, i_node);
Yan, Zhengfc927cd2015-07-20 09:50:58 +08002151 if (cf->tid >= first_tid)
Yan, Zheng553adfd2015-06-09 15:48:57 +08002152 break;
2153 }
2154 if (!n) {
2155 spin_unlock(&ci->i_ceph_lock);
2156 break;
2157 }
2158
2159 cf = rb_entry(n, struct ceph_cap_flush, i_node);
Yan, Zhenge548e9b2015-06-10 15:17:56 +08002160
Yan, Zheng553adfd2015-06-09 15:48:57 +08002161 first_tid = cf->tid + 1;
2162
2163 dout("kick_flushing_caps %p cap %p tid %llu %s\n", inode,
2164 cap, cf->tid, ceph_cap_string(cf->caps));
2165 delayed |= __send_cap(mdsc, cap, CEPH_CAP_OP_FLUSH,
2166 __ceph_caps_used(ci),
2167 __ceph_caps_wanted(ci),
2168 cap->issued | cap->implemented,
Yan, Zhenga2971c82015-06-10 11:09:32 +08002169 cf->caps, cf->tid, oldest_flush_tid);
Yan, Zheng553adfd2015-06-09 15:48:57 +08002170 }
2171 return delayed;
2172}
2173
Yan, Zhenge548e9b2015-06-10 15:17:56 +08002174void ceph_early_kick_flushing_caps(struct ceph_mds_client *mdsc,
2175 struct ceph_mds_session *session)
2176{
2177 struct ceph_inode_info *ci;
2178 struct ceph_cap *cap;
Yan, Zhenge548e9b2015-06-10 15:17:56 +08002179
2180 dout("early_kick_flushing_caps mds%d\n", session->s_mds);
2181 list_for_each_entry(ci, &session->s_cap_flushing, i_flushing_item) {
2182 spin_lock(&ci->i_ceph_lock);
2183 cap = ci->i_auth_cap;
2184 if (!(cap && cap->session == session)) {
2185 pr_err("%p auth cap %p not mds%d ???\n",
2186 &ci->vfs_inode, cap, session->s_mds);
2187 spin_unlock(&ci->i_ceph_lock);
2188 continue;
2189 }
2190
2191
2192 /*
2193 * if flushing caps were revoked, we re-send the cap flush
2194 * in client reconnect stage. This guarantees MDS * processes
2195 * the cap flush message before issuing the flushing caps to
2196 * other client.
2197 */
2198 if ((cap->issued & ci->i_flushing_caps) !=
2199 ci->i_flushing_caps) {
2200 spin_unlock(&ci->i_ceph_lock);
Yan, Zhengfc927cd2015-07-20 09:50:58 +08002201 if (!__kick_flushing_caps(mdsc, session, ci))
Yan, Zhenge548e9b2015-06-10 15:17:56 +08002202 continue;
2203 spin_lock(&ci->i_ceph_lock);
2204 }
2205
Yan, Zhenge548e9b2015-06-10 15:17:56 +08002206 spin_unlock(&ci->i_ceph_lock);
2207 }
2208}
2209
Sage Weila8599bd2009-10-06 11:31:12 -07002210void ceph_kick_flushing_caps(struct ceph_mds_client *mdsc,
2211 struct ceph_mds_session *session)
2212{
2213 struct ceph_inode_info *ci;
2214
2215 kick_flushing_capsnaps(mdsc, session);
2216
2217 dout("kick_flushing_caps mds%d\n", session->s_mds);
2218 list_for_each_entry(ci, &session->s_cap_flushing, i_flushing_item) {
Yan, Zhengfc927cd2015-07-20 09:50:58 +08002219 int delayed = __kick_flushing_caps(mdsc, session, ci);
Yan, Zheng553adfd2015-06-09 15:48:57 +08002220 if (delayed) {
2221 spin_lock(&ci->i_ceph_lock);
2222 __cap_delay_requeue(mdsc, ci);
Sage Weilbe655592011-11-30 09:47:09 -08002223 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07002224 }
2225 }
2226}
2227
Sage Weil088b3f52011-01-18 08:56:01 -08002228static void kick_flushing_inode_caps(struct ceph_mds_client *mdsc,
2229 struct ceph_mds_session *session,
2230 struct inode *inode)
2231{
2232 struct ceph_inode_info *ci = ceph_inode(inode);
2233 struct ceph_cap *cap;
Sage Weil088b3f52011-01-18 08:56:01 -08002234
Sage Weilbe655592011-11-30 09:47:09 -08002235 spin_lock(&ci->i_ceph_lock);
Sage Weil088b3f52011-01-18 08:56:01 -08002236 cap = ci->i_auth_cap;
Yan, Zheng8310b082015-06-09 17:20:12 +08002237 dout("kick_flushing_inode_caps %p flushing %s\n", inode,
2238 ceph_cap_string(ci->i_flushing_caps));
Yan, Zheng005c4692013-05-31 16:40:24 +08002239
Sage Weil088b3f52011-01-18 08:56:01 -08002240 __ceph_flush_snaps(ci, &session, 1);
Yan, Zheng005c4692013-05-31 16:40:24 +08002241
Sage Weil088b3f52011-01-18 08:56:01 -08002242 if (ci->i_flushing_caps) {
Yan, Zheng553adfd2015-06-09 15:48:57 +08002243 int delayed;
2244
Yan, Zheng005c4692013-05-31 16:40:24 +08002245 spin_lock(&mdsc->cap_dirty_lock);
2246 list_move_tail(&ci->i_flushing_item,
2247 &cap->session->s_cap_flushing);
2248 spin_unlock(&mdsc->cap_dirty_lock);
2249
Yan, Zheng553adfd2015-06-09 15:48:57 +08002250 spin_unlock(&ci->i_ceph_lock);
2251
Yan, Zhengfc927cd2015-07-20 09:50:58 +08002252 delayed = __kick_flushing_caps(mdsc, session, ci);
Sage Weil088b3f52011-01-18 08:56:01 -08002253 if (delayed) {
Sage Weilbe655592011-11-30 09:47:09 -08002254 spin_lock(&ci->i_ceph_lock);
Sage Weil088b3f52011-01-18 08:56:01 -08002255 __cap_delay_requeue(mdsc, ci);
Sage Weilbe655592011-11-30 09:47:09 -08002256 spin_unlock(&ci->i_ceph_lock);
Sage Weil088b3f52011-01-18 08:56:01 -08002257 }
2258 } 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);
2321 if ((file_wanted & need) == 0) {
2322 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, Zheng5dda377c2015-04-30 14:40:54 +08002397 __take_cap_refs(ci, *got, true);
Sage Weila8599bd2009-10-06 11:31:12 -07002398 ret = 1;
2399 }
2400 } else {
Yan, Zheng03f4fcb2015-01-05 11:04:04 +08002401 int session_readonly = false;
2402 if ((need & CEPH_CAP_FILE_WR) && ci->i_auth_cap) {
2403 struct ceph_mds_session *s = ci->i_auth_cap->session;
2404 spin_lock(&s->s_cap_lock);
2405 session_readonly = s->s_readonly;
2406 spin_unlock(&s->s_cap_lock);
2407 }
2408 if (session_readonly) {
2409 dout("get_cap_refs %p needed %s but mds%d readonly\n",
2410 inode, ceph_cap_string(need), ci->i_auth_cap->mds);
2411 *err = -EROFS;
2412 ret = 1;
2413 goto out_unlock;
2414 }
2415
Yan, Zheng48fec5d2015-07-01 16:27:46 +08002416 if (!__ceph_is_any_caps(ci) &&
2417 ACCESS_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) {
2418 dout("get_cap_refs %p forced umount\n", inode);
2419 *err = -EIO;
2420 ret = 1;
2421 goto out_unlock;
2422 }
2423
Sage Weila8599bd2009-10-06 11:31:12 -07002424 dout("get_cap_refs %p have %s needed %s\n", inode,
2425 ceph_cap_string(have), ceph_cap_string(need));
2426 }
Yan, Zheng3738daa2014-11-14 22:10:07 +08002427out_unlock:
Sage Weilbe655592011-11-30 09:47:09 -08002428 spin_unlock(&ci->i_ceph_lock);
Yan, Zheng5dda377c2015-04-30 14:40:54 +08002429 if (snap_rwsem_locked)
2430 up_read(&mdsc->snap_rwsem);
Yan, Zheng3738daa2014-11-14 22:10:07 +08002431
Sage Weila8599bd2009-10-06 11:31:12 -07002432 dout("get_cap_refs %p ret %d got %s\n", inode,
Yan, Zhengc4d4a582015-01-09 15:56:18 +08002433 ret, ceph_cap_string(*got));
Sage Weila8599bd2009-10-06 11:31:12 -07002434 return ret;
2435}
2436
2437/*
2438 * Check the offset we are writing up to against our current
2439 * max_size. If necessary, tell the MDS we want to write to
2440 * a larger offset.
2441 */
2442static void check_max_size(struct inode *inode, loff_t endoff)
2443{
2444 struct ceph_inode_info *ci = ceph_inode(inode);
2445 int check = 0;
2446
2447 /* do we need to explicitly request a larger max_size? */
Sage Weilbe655592011-11-30 09:47:09 -08002448 spin_lock(&ci->i_ceph_lock);
Yan, Zheng3871cbb2013-08-05 14:10:29 +08002449 if (endoff >= ci->i_max_size && endoff > ci->i_wanted_max_size) {
Sage Weila8599bd2009-10-06 11:31:12 -07002450 dout("write %p at large endoff %llu, req max_size\n",
2451 inode, endoff);
2452 ci->i_wanted_max_size = endoff;
Sage Weila8599bd2009-10-06 11:31:12 -07002453 }
Yan, Zheng3871cbb2013-08-05 14:10:29 +08002454 /* duplicate ceph_check_caps()'s logic */
2455 if (ci->i_auth_cap &&
2456 (ci->i_auth_cap->issued & CEPH_CAP_FILE_WR) &&
2457 ci->i_wanted_max_size > ci->i_max_size &&
2458 ci->i_wanted_max_size > ci->i_requested_max_size)
2459 check = 1;
Sage Weilbe655592011-11-30 09:47:09 -08002460 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07002461 if (check)
2462 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
2463}
2464
2465/*
2466 * Wait for caps, and take cap references. If we can't get a WR cap
2467 * due to a small max_size, make sure we check_max_size (and possibly
2468 * ask the mds) so we don't get hung up indefinitely.
2469 */
Yan, Zheng3738daa2014-11-14 22:10:07 +08002470int ceph_get_caps(struct ceph_inode_info *ci, int need, int want,
2471 loff_t endoff, int *got, struct page **pinned_page)
Sage Weila8599bd2009-10-06 11:31:12 -07002472{
Yan, Zheng5dda377c2015-04-30 14:40:54 +08002473 int _got, ret, err = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07002474
Yan, Zheng10183a62015-04-27 15:33:28 +08002475 ret = ceph_pool_perm_check(ci, need);
2476 if (ret < 0)
2477 return ret;
2478
Yan, Zheng5dda377c2015-04-30 14:40:54 +08002479 while (true) {
2480 if (endoff > 0)
2481 check_max_size(&ci->vfs_inode, endoff);
Yan, Zhengc4d4a582015-01-09 15:56:18 +08002482
Yan, Zheng5dda377c2015-04-30 14:40:54 +08002483 err = 0;
Yan, Zhengc4d4a582015-01-09 15:56:18 +08002484 _got = 0;
Yan, Zheng5dda377c2015-04-30 14:40:54 +08002485 ret = try_get_cap_refs(ci, need, want, endoff,
2486 false, &_got, &err);
2487 if (ret) {
2488 if (err == -EAGAIN)
2489 continue;
2490 if (err < 0)
2491 return err;
2492 } else {
2493 ret = wait_event_interruptible(ci->i_cap_wq,
2494 try_get_cap_refs(ci, need, want, endoff,
2495 true, &_got, &err));
2496 if (err == -EAGAIN)
2497 continue;
2498 if (err < 0)
2499 ret = err;
2500 if (ret < 0)
2501 return ret;
2502 }
Yan, Zhengc4d4a582015-01-09 15:56:18 +08002503
Yan, Zheng5dda377c2015-04-30 14:40:54 +08002504 if (ci->i_inline_version != CEPH_INLINE_NONE &&
2505 (_got & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) &&
2506 i_size_read(&ci->vfs_inode) > 0) {
2507 struct page *page =
2508 find_get_page(ci->vfs_inode.i_mapping, 0);
2509 if (page) {
2510 if (PageUptodate(page)) {
2511 *pinned_page = page;
2512 break;
2513 }
2514 page_cache_release(page);
2515 }
2516 /*
2517 * drop cap refs first because getattr while
2518 * holding * caps refs can cause deadlock.
2519 */
2520 ceph_put_cap_refs(ci, _got);
2521 _got = 0;
2522
2523 /*
2524 * getattr request will bring inline data into
2525 * page cache
2526 */
2527 ret = __ceph_do_getattr(&ci->vfs_inode, NULL,
2528 CEPH_STAT_CAP_INLINE_DATA,
2529 true);
2530 if (ret < 0)
2531 return ret;
2532 continue;
2533 }
2534 break;
Yan, Zhengc4d4a582015-01-09 15:56:18 +08002535 }
Yan, Zheng5dda377c2015-04-30 14:40:54 +08002536
Yan, Zhengc4d4a582015-01-09 15:56:18 +08002537 *got = _got;
2538 return 0;
Sage Weila8599bd2009-10-06 11:31:12 -07002539}
2540
2541/*
2542 * Take cap refs. Caller must already know we hold at least one ref
2543 * on the caps in question or we don't know this is safe.
2544 */
2545void ceph_get_cap_refs(struct ceph_inode_info *ci, int caps)
2546{
Sage Weilbe655592011-11-30 09:47:09 -08002547 spin_lock(&ci->i_ceph_lock);
Yan, Zheng5dda377c2015-04-30 14:40:54 +08002548 __take_cap_refs(ci, caps, false);
Sage Weilbe655592011-11-30 09:47:09 -08002549 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07002550}
2551
Yan, Zheng86056092015-05-01 16:57:16 +08002552
2553/*
2554 * drop cap_snap that is not associated with any snapshot.
2555 * we don't need to send FLUSHSNAP message for it.
2556 */
2557static int ceph_try_drop_cap_snap(struct ceph_cap_snap *capsnap)
2558{
2559 if (!capsnap->need_flush &&
2560 !capsnap->writing && !capsnap->dirty_pages) {
2561
2562 dout("dropping cap_snap %p follows %llu\n",
2563 capsnap, capsnap->follows);
2564 ceph_put_snap_context(capsnap->context);
2565 list_del(&capsnap->ci_item);
2566 list_del(&capsnap->flushing_item);
2567 ceph_put_cap_snap(capsnap);
2568 return 1;
2569 }
2570 return 0;
2571}
2572
Sage Weila8599bd2009-10-06 11:31:12 -07002573/*
2574 * Release cap refs.
2575 *
2576 * If we released the last ref on any given cap, call ceph_check_caps
2577 * to release (or schedule a release).
2578 *
2579 * If we are releasing a WR cap (from a sync write), finalize any affected
2580 * cap_snap, and wake up any waiters.
2581 */
2582void ceph_put_cap_refs(struct ceph_inode_info *ci, int had)
2583{
2584 struct inode *inode = &ci->vfs_inode;
2585 int last = 0, put = 0, flushsnaps = 0, wake = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07002586
Sage Weilbe655592011-11-30 09:47:09 -08002587 spin_lock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07002588 if (had & CEPH_CAP_PIN)
2589 --ci->i_pin_ref;
2590 if (had & CEPH_CAP_FILE_RD)
2591 if (--ci->i_rd_ref == 0)
2592 last++;
2593 if (had & CEPH_CAP_FILE_CACHE)
2594 if (--ci->i_rdcache_ref == 0)
2595 last++;
2596 if (had & CEPH_CAP_FILE_BUFFER) {
Henry C Changd3d07202011-05-11 10:29:54 +00002597 if (--ci->i_wb_ref == 0) {
Sage Weila8599bd2009-10-06 11:31:12 -07002598 last++;
2599 put++;
2600 }
Henry C Changd3d07202011-05-11 10:29:54 +00002601 dout("put_cap_refs %p wb %d -> %d (?)\n",
2602 inode, ci->i_wb_ref+1, ci->i_wb_ref);
Sage Weila8599bd2009-10-06 11:31:12 -07002603 }
2604 if (had & CEPH_CAP_FILE_WR)
2605 if (--ci->i_wr_ref == 0) {
2606 last++;
Yan, Zheng86056092015-05-01 16:57:16 +08002607 if (__ceph_have_pending_cap_snap(ci)) {
2608 struct ceph_cap_snap *capsnap =
2609 list_last_entry(&ci->i_cap_snaps,
2610 struct ceph_cap_snap,
2611 ci_item);
2612 capsnap->writing = 0;
2613 if (ceph_try_drop_cap_snap(capsnap))
2614 put++;
2615 else if (__ceph_finish_cap_snap(ci, capsnap))
2616 flushsnaps = 1;
2617 wake = 1;
Sage Weila8599bd2009-10-06 11:31:12 -07002618 }
Yan, Zheng5dda377c2015-04-30 14:40:54 +08002619 if (ci->i_wrbuffer_ref_head == 0 &&
2620 ci->i_dirty_caps == 0 &&
2621 ci->i_flushing_caps == 0) {
2622 BUG_ON(!ci->i_head_snapc);
2623 ceph_put_snap_context(ci->i_head_snapc);
2624 ci->i_head_snapc = NULL;
2625 }
Yan, Zhengdb40cc12015-03-23 20:12:20 +08002626 /* see comment in __ceph_remove_cap() */
2627 if (!__ceph_is_any_caps(ci) && ci->i_snap_realm)
2628 drop_inode_snap_realm(ci);
Sage Weila8599bd2009-10-06 11:31:12 -07002629 }
Sage Weilbe655592011-11-30 09:47:09 -08002630 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07002631
Sage Weil819ccbf2010-04-01 09:33:46 -07002632 dout("put_cap_refs %p had %s%s%s\n", inode, ceph_cap_string(had),
2633 last ? " last" : "", put ? " put" : "");
Sage Weila8599bd2009-10-06 11:31:12 -07002634
2635 if (last && !flushsnaps)
2636 ceph_check_caps(ci, 0, NULL);
2637 else if (flushsnaps)
2638 ceph_flush_snaps(ci);
2639 if (wake)
Yehuda Sadeh03066f22010-07-27 13:11:08 -07002640 wake_up_all(&ci->i_cap_wq);
Yan, Zheng86056092015-05-01 16:57:16 +08002641 while (put-- > 0)
Sage Weila8599bd2009-10-06 11:31:12 -07002642 iput(inode);
2643}
2644
2645/*
2646 * Release @nr WRBUFFER refs on dirty pages for the given @snapc snap
2647 * context. Adjust per-snap dirty page accounting as appropriate.
2648 * Once all dirty data for a cap_snap is flushed, flush snapped file
2649 * metadata back to the MDS. If we dropped the last ref, call
2650 * ceph_check_caps.
2651 */
2652void ceph_put_wrbuffer_cap_refs(struct ceph_inode_info *ci, int nr,
2653 struct ceph_snap_context *snapc)
2654{
2655 struct inode *inode = &ci->vfs_inode;
2656 int last = 0;
Sage Weil819ccbf2010-04-01 09:33:46 -07002657 int complete_capsnap = 0;
2658 int drop_capsnap = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07002659 int found = 0;
2660 struct ceph_cap_snap *capsnap = NULL;
2661
Sage Weilbe655592011-11-30 09:47:09 -08002662 spin_lock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07002663 ci->i_wrbuffer_ref -= nr;
2664 last = !ci->i_wrbuffer_ref;
2665
2666 if (ci->i_head_snapc == snapc) {
2667 ci->i_wrbuffer_ref_head -= nr;
Sage Weil7d8cb262010-08-24 08:44:16 -07002668 if (ci->i_wrbuffer_ref_head == 0 &&
Yan, Zheng5dda377c2015-04-30 14:40:54 +08002669 ci->i_wr_ref == 0 &&
2670 ci->i_dirty_caps == 0 &&
2671 ci->i_flushing_caps == 0) {
Sage Weil7d8cb262010-08-24 08:44:16 -07002672 BUG_ON(!ci->i_head_snapc);
Sage Weila8599bd2009-10-06 11:31:12 -07002673 ceph_put_snap_context(ci->i_head_snapc);
2674 ci->i_head_snapc = NULL;
2675 }
2676 dout("put_wrbuffer_cap_refs on %p head %d/%d -> %d/%d %s\n",
2677 inode,
2678 ci->i_wrbuffer_ref+nr, ci->i_wrbuffer_ref_head+nr,
2679 ci->i_wrbuffer_ref, ci->i_wrbuffer_ref_head,
2680 last ? " LAST" : "");
2681 } else {
2682 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
2683 if (capsnap->context == snapc) {
2684 found = 1;
Sage Weila8599bd2009-10-06 11:31:12 -07002685 break;
2686 }
2687 }
2688 BUG_ON(!found);
Sage Weil819ccbf2010-04-01 09:33:46 -07002689 capsnap->dirty_pages -= nr;
2690 if (capsnap->dirty_pages == 0) {
2691 complete_capsnap = 1;
Yan, Zheng86056092015-05-01 16:57:16 +08002692 drop_capsnap = ceph_try_drop_cap_snap(capsnap);
Sage Weil819ccbf2010-04-01 09:33:46 -07002693 }
Sage Weila8599bd2009-10-06 11:31:12 -07002694 dout("put_wrbuffer_cap_refs on %p cap_snap %p "
Yan, Zheng86056092015-05-01 16:57:16 +08002695 " snap %lld %d/%d -> %d/%d %s%s\n",
Sage Weila8599bd2009-10-06 11:31:12 -07002696 inode, capsnap, capsnap->context->seq,
2697 ci->i_wrbuffer_ref+nr, capsnap->dirty_pages + nr,
2698 ci->i_wrbuffer_ref, capsnap->dirty_pages,
2699 last ? " (wrbuffer last)" : "",
Yan, Zheng86056092015-05-01 16:57:16 +08002700 complete_capsnap ? " (complete capsnap)" : "");
Sage Weila8599bd2009-10-06 11:31:12 -07002701 }
2702
Sage Weilbe655592011-11-30 09:47:09 -08002703 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07002704
2705 if (last) {
2706 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
2707 iput(inode);
Sage Weil819ccbf2010-04-01 09:33:46 -07002708 } else if (complete_capsnap) {
Sage Weila8599bd2009-10-06 11:31:12 -07002709 ceph_flush_snaps(ci);
Yehuda Sadeh03066f22010-07-27 13:11:08 -07002710 wake_up_all(&ci->i_cap_wq);
Sage Weila8599bd2009-10-06 11:31:12 -07002711 }
Sage Weil819ccbf2010-04-01 09:33:46 -07002712 if (drop_capsnap)
2713 iput(inode);
Sage Weila8599bd2009-10-06 11:31:12 -07002714}
2715
2716/*
Yan, Zhengca20c992013-07-21 10:07:51 +08002717 * Invalidate unlinked inode's aliases, so we can drop the inode ASAP.
2718 */
2719static void invalidate_aliases(struct inode *inode)
2720{
2721 struct dentry *dn, *prev = NULL;
2722
2723 dout("invalidate_aliases inode %p\n", inode);
2724 d_prune_aliases(inode);
2725 /*
2726 * For non-directory inode, d_find_alias() only returns
J. Bruce Fieldsfc12c802014-01-16 17:42:53 -05002727 * hashed dentry. After calling d_invalidate(), the
2728 * dentry becomes unhashed.
Yan, Zhengca20c992013-07-21 10:07:51 +08002729 *
Yan, Zhenga8d436f2013-09-02 15:19:54 +08002730 * For directory inode, d_find_alias() can return
J. Bruce Fieldsfc12c802014-01-16 17:42:53 -05002731 * unhashed dentry. But directory inode should have
Yan, Zhengca20c992013-07-21 10:07:51 +08002732 * one alias at most.
2733 */
2734 while ((dn = d_find_alias(inode))) {
2735 if (dn == prev) {
2736 dput(dn);
2737 break;
2738 }
Yan, Zhenga8d436f2013-09-02 15:19:54 +08002739 d_invalidate(dn);
Yan, Zhengca20c992013-07-21 10:07:51 +08002740 if (prev)
2741 dput(prev);
2742 prev = dn;
2743 }
2744 if (prev)
2745 dput(prev);
2746}
2747
2748/*
Sage Weila8599bd2009-10-06 11:31:12 -07002749 * Handle a cap GRANT message from the MDS. (Note that a GRANT may
2750 * actually be a revocation if it specifies a smaller cap set.)
2751 *
Sage Weilbe655592011-11-30 09:47:09 -08002752 * caller holds s_mutex and i_ceph_lock, we drop both.
Sage Weila8599bd2009-10-06 11:31:12 -07002753 */
Yan, Zheng2cd698b2014-04-18 13:20:27 +08002754static void handle_cap_grant(struct ceph_mds_client *mdsc,
2755 struct inode *inode, struct ceph_mds_caps *grant,
Yan, Zhengfb01d1f2014-11-14 21:29:55 +08002756 u64 inline_version,
2757 void *inline_data, int inline_len,
Yan, Zheng2cd698b2014-04-18 13:20:27 +08002758 struct ceph_buffer *xattr_buf,
Sage Weil15637c82010-03-16 13:42:00 -07002759 struct ceph_mds_session *session,
Yan, Zheng2cd698b2014-04-18 13:20:27 +08002760 struct ceph_cap *cap, int issued)
2761 __releases(ci->i_ceph_lock)
Yan, Zheng982d6012014-12-23 15:30:54 +08002762 __releases(mdsc->snap_rwsem)
Sage Weila8599bd2009-10-06 11:31:12 -07002763{
2764 struct ceph_inode_info *ci = ceph_inode(inode);
2765 int mds = session->s_mds;
Sage Weil2f56f562010-10-27 20:59:49 -07002766 int seq = le32_to_cpu(grant->seq);
Sage Weila8599bd2009-10-06 11:31:12 -07002767 int newcaps = le32_to_cpu(grant->caps);
Yan, Zheng2cd698b2014-04-18 13:20:27 +08002768 int used, wanted, dirty;
Sage Weila8599bd2009-10-06 11:31:12 -07002769 u64 size = le64_to_cpu(grant->size);
2770 u64 max_size = le64_to_cpu(grant->max_size);
2771 struct timespec mtime, atime, ctime;
Sage Weil15637c82010-03-16 13:42:00 -07002772 int check_caps = 0;
Fabian Frederickab6c2c32014-10-09 23:16:35 +02002773 bool wake = false;
2774 bool writeback = false;
2775 bool queue_trunc = false;
2776 bool queue_invalidate = false;
2777 bool queue_revalidate = false;
2778 bool deleted_inode = false;
Yan, Zheng31c542a2014-11-14 21:41:55 +08002779 bool fill_inline = false;
Sage Weila8599bd2009-10-06 11:31:12 -07002780
Sage Weil2f56f562010-10-27 20:59:49 -07002781 dout("handle_cap_grant inode %p cap %p mds%d seq %d %s\n",
2782 inode, cap, mds, seq, ceph_cap_string(newcaps));
Sage Weila8599bd2009-10-06 11:31:12 -07002783 dout(" size %llu max_size %llu, i_size %llu\n", size, max_size,
2784 inode->i_size);
2785
Yan, Zheng11df2df2013-11-24 14:44:38 +08002786
2787 /*
2788 * auth mds of the inode changed. we received the cap export message,
2789 * but still haven't received the cap import message. handle_cap_export
2790 * updated the new auth MDS' cap.
2791 *
2792 * "ceph_seq_cmp(seq, cap->seq) <= 0" means we are processing a message
2793 * that was sent before the cap import message. So don't remove caps.
2794 */
2795 if (ceph_seq_cmp(seq, cap->seq) <= 0) {
2796 WARN_ON(cap != ci->i_auth_cap);
2797 WARN_ON(cap->cap_id != le64_to_cpu(grant->cap_id));
2798 seq = cap->seq;
2799 newcaps |= cap->issued;
2800 }
2801
Sage Weila8599bd2009-10-06 11:31:12 -07002802 /*
2803 * If CACHE is being revoked, and we have no dirty buffers,
2804 * try to invalidate (once). (If there are dirty buffers, we
2805 * will invalidate _after_ writeback.)
2806 */
Yan, Zhengfdd4e152015-06-16 20:48:56 +08002807 if (!S_ISDIR(inode->i_mode) && /* don't invalidate readdir cache */
2808 ((cap->issued & ~newcaps) & CEPH_CAP_FILE_CACHE) &&
Sage Weil3b454c42010-06-10 13:20:33 -07002809 (newcaps & CEPH_CAP_FILE_LAZYIO) == 0 &&
Yehuda Sadehbcd2cbd2010-02-19 00:12:21 +00002810 !ci->i_wrbuffer_ref) {
Li Wange9075742013-08-15 22:00:25 -07002811 if (try_nonblocking_invalidate(inode)) {
Sage Weila8599bd2009-10-06 11:31:12 -07002812 /* there were locked pages.. invalidate later
2813 in a separate thread. */
2814 if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
Fabian Frederickab6c2c32014-10-09 23:16:35 +02002815 queue_invalidate = true;
Sage Weila8599bd2009-10-06 11:31:12 -07002816 ci->i_rdcache_revoking = ci->i_rdcache_gen;
2817 }
Sage Weila8599bd2009-10-06 11:31:12 -07002818 }
Milosz Tanski99ccbd22013-08-21 17:29:54 -04002819
2820 ceph_fscache_invalidate(inode);
Sage Weila8599bd2009-10-06 11:31:12 -07002821 }
2822
2823 /* side effects now are allowed */
Sage Weil685f9a5d2009-11-09 12:05:48 -08002824 cap->cap_gen = session->s_cap_gen;
Yan, Zheng11df2df2013-11-24 14:44:38 +08002825 cap->seq = seq;
Sage Weila8599bd2009-10-06 11:31:12 -07002826
2827 __check_cap_issue(ci, cap, newcaps);
2828
Yan, Zhengf98a1282014-04-17 08:55:50 +08002829 if ((newcaps & CEPH_CAP_AUTH_SHARED) &&
2830 (issued & CEPH_CAP_AUTH_EXCL) == 0) {
Sage Weila8599bd2009-10-06 11:31:12 -07002831 inode->i_mode = le32_to_cpu(grant->mode);
Eric W. Biederman05cb11c2013-01-31 02:56:19 -08002832 inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(grant->uid));
2833 inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(grant->gid));
Sage Weila8599bd2009-10-06 11:31:12 -07002834 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
Eric W. Biedermanbd2bae62013-01-31 04:05:39 -08002835 from_kuid(&init_user_ns, inode->i_uid),
2836 from_kgid(&init_user_ns, inode->i_gid));
Sage Weila8599bd2009-10-06 11:31:12 -07002837 }
2838
Yan, Zhengf98a1282014-04-17 08:55:50 +08002839 if ((newcaps & CEPH_CAP_AUTH_SHARED) &&
2840 (issued & CEPH_CAP_LINK_EXCL) == 0) {
Miklos Szeredibfe86842011-10-28 14:13:29 +02002841 set_nlink(inode, le32_to_cpu(grant->nlink));
Yan, Zhengca20c992013-07-21 10:07:51 +08002842 if (inode->i_nlink == 0 &&
2843 (newcaps & (CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL)))
Fabian Frederickab6c2c32014-10-09 23:16:35 +02002844 deleted_inode = true;
Yan, Zhengca20c992013-07-21 10:07:51 +08002845 }
Sage Weila8599bd2009-10-06 11:31:12 -07002846
2847 if ((issued & CEPH_CAP_XATTR_EXCL) == 0 && grant->xattr_len) {
2848 int len = le32_to_cpu(grant->xattr_len);
2849 u64 version = le64_to_cpu(grant->xattr_version);
2850
2851 if (version > ci->i_xattrs.version) {
2852 dout(" got new xattrs v%llu on %p len %d\n",
2853 version, inode, len);
2854 if (ci->i_xattrs.blob)
2855 ceph_buffer_put(ci->i_xattrs.blob);
2856 ci->i_xattrs.blob = ceph_buffer_get(xattr_buf);
2857 ci->i_xattrs.version = version;
Guangliang Zhao7221fe42013-11-11 15:18:03 +08002858 ceph_forget_all_cached_acls(inode);
Sage Weila8599bd2009-10-06 11:31:12 -07002859 }
2860 }
2861
Milosz Tanski99ccbd22013-08-21 17:29:54 -04002862 /* Do we need to revalidate our fscache cookie. Don't bother on the
2863 * first cache cap as we already validate at cookie creation time. */
2864 if ((issued & CEPH_CAP_FILE_CACHE) && ci->i_rdcache_gen > 1)
Fabian Frederickab6c2c32014-10-09 23:16:35 +02002865 queue_revalidate = true;
Milosz Tanski99ccbd22013-08-21 17:29:54 -04002866
Yan, Zhengf98a1282014-04-17 08:55:50 +08002867 if (newcaps & CEPH_CAP_ANY_RD) {
2868 /* ctime/mtime/atime? */
2869 ceph_decode_timespec(&mtime, &grant->mtime);
2870 ceph_decode_timespec(&atime, &grant->atime);
2871 ceph_decode_timespec(&ctime, &grant->ctime);
2872 ceph_fill_file_time(inode, issued,
2873 le32_to_cpu(grant->time_warp_seq),
2874 &ctime, &mtime, &atime);
2875 }
Sage Weila8599bd2009-10-06 11:31:12 -07002876
Yan, Zhengf98a1282014-04-17 08:55:50 +08002877 if (newcaps & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR)) {
2878 /* file layout may have changed */
2879 ci->i_layout = grant->layout;
2880 /* size/truncate_seq? */
2881 queue_trunc = ceph_fill_file_size(inode, issued,
2882 le32_to_cpu(grant->truncate_seq),
2883 le64_to_cpu(grant->truncate_size),
2884 size);
2885 /* max size increase? */
2886 if (ci->i_auth_cap == cap && max_size != ci->i_max_size) {
2887 dout("max_size %lld -> %llu\n",
2888 ci->i_max_size, max_size);
2889 ci->i_max_size = max_size;
2890 if (max_size >= ci->i_wanted_max_size) {
2891 ci->i_wanted_max_size = 0; /* reset */
2892 ci->i_requested_max_size = 0;
2893 }
Fabian Frederickab6c2c32014-10-09 23:16:35 +02002894 wake = true;
Sage Weila8599bd2009-10-06 11:31:12 -07002895 }
Sage Weila8599bd2009-10-06 11:31:12 -07002896 }
2897
2898 /* check cap bits */
2899 wanted = __ceph_caps_wanted(ci);
2900 used = __ceph_caps_used(ci);
2901 dirty = __ceph_caps_dirty(ci);
2902 dout(" my wanted = %s, used = %s, dirty %s\n",
2903 ceph_cap_string(wanted),
2904 ceph_cap_string(used),
2905 ceph_cap_string(dirty));
2906 if (wanted != le32_to_cpu(grant->wanted)) {
2907 dout("mds wanted %s -> %s\n",
2908 ceph_cap_string(le32_to_cpu(grant->wanted)),
2909 ceph_cap_string(wanted));
Yan, Zheng390306c2013-01-04 15:30:10 +08002910 /* imported cap may not have correct mds_wanted */
2911 if (le32_to_cpu(grant->op) == CEPH_CAP_OP_IMPORT)
2912 check_caps = 1;
Sage Weila8599bd2009-10-06 11:31:12 -07002913 }
2914
Sage Weila8599bd2009-10-06 11:31:12 -07002915 /* revocation, grant, or no-op? */
2916 if (cap->issued & ~newcaps) {
Sage Weil3b454c42010-06-10 13:20:33 -07002917 int revoking = cap->issued & ~newcaps;
2918
2919 dout("revocation: %s -> %s (revoking %s)\n",
2920 ceph_cap_string(cap->issued),
2921 ceph_cap_string(newcaps),
2922 ceph_cap_string(revoking));
Sage Weil0eb6cd42010-08-05 13:53:18 -07002923 if (revoking & used & CEPH_CAP_FILE_BUFFER)
Fabian Frederickab6c2c32014-10-09 23:16:35 +02002924 writeback = true; /* initiate writeback; will delay ack */
Sage Weil3b454c42010-06-10 13:20:33 -07002925 else if (revoking == CEPH_CAP_FILE_CACHE &&
2926 (newcaps & CEPH_CAP_FILE_LAZYIO) == 0 &&
2927 queue_invalidate)
2928 ; /* do nothing yet, invalidation will be queued */
2929 else if (cap == ci->i_auth_cap)
2930 check_caps = 1; /* check auth cap only */
2931 else
2932 check_caps = 2; /* check all caps */
Sage Weila8599bd2009-10-06 11:31:12 -07002933 cap->issued = newcaps;
Sage Weil978097c2010-03-08 15:27:53 -08002934 cap->implemented |= newcaps;
Sage Weila8599bd2009-10-06 11:31:12 -07002935 } else if (cap->issued == newcaps) {
2936 dout("caps unchanged: %s -> %s\n",
2937 ceph_cap_string(cap->issued), ceph_cap_string(newcaps));
2938 } else {
2939 dout("grant: %s -> %s\n", ceph_cap_string(cap->issued),
2940 ceph_cap_string(newcaps));
Yan, Zheng6ee6b9532013-07-02 12:40:21 +08002941 /* non-auth MDS is revoking the newly grant caps ? */
2942 if (cap == ci->i_auth_cap &&
2943 __ceph_caps_revoking_other(ci, cap, newcaps))
2944 check_caps = 2;
2945
Sage Weila8599bd2009-10-06 11:31:12 -07002946 cap->issued = newcaps;
2947 cap->implemented |= newcaps; /* add bits only, to
2948 * avoid stepping on a
2949 * pending revocation */
Fabian Frederickab6c2c32014-10-09 23:16:35 +02002950 wake = true;
Sage Weila8599bd2009-10-06 11:31:12 -07002951 }
Sage Weil978097c2010-03-08 15:27:53 -08002952 BUG_ON(cap->issued & ~cap->implemented);
Sage Weila8599bd2009-10-06 11:31:12 -07002953
Yan, Zheng31c542a2014-11-14 21:41:55 +08002954 if (inline_version > 0 && inline_version >= ci->i_inline_version) {
2955 ci->i_inline_version = inline_version;
2956 if (ci->i_inline_version != CEPH_INLINE_NONE &&
2957 (newcaps & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)))
2958 fill_inline = true;
2959 }
2960
Sage Weilbe655592011-11-30 09:47:09 -08002961 spin_unlock(&ci->i_ceph_lock);
Milosz Tanski99ccbd22013-08-21 17:29:54 -04002962
Yan, Zheng2cd698b2014-04-18 13:20:27 +08002963 if (le32_to_cpu(grant->op) == CEPH_CAP_OP_IMPORT) {
Yan, Zheng2cd698b2014-04-18 13:20:27 +08002964 kick_flushing_inode_caps(mdsc, session, inode);
2965 up_read(&mdsc->snap_rwsem);
2966 if (newcaps & ~issued)
Fabian Frederickab6c2c32014-10-09 23:16:35 +02002967 wake = true;
Yan, Zheng2cd698b2014-04-18 13:20:27 +08002968 }
2969
Yan, Zheng31c542a2014-11-14 21:41:55 +08002970 if (fill_inline)
2971 ceph_fill_inline_data(inode, NULL, inline_data, inline_len);
2972
Yan, Zhengc6bcda62014-04-11 10:18:07 +08002973 if (queue_trunc) {
2974 ceph_queue_vmtruncate(inode);
2975 ceph_queue_revalidate(inode);
2976 } else if (queue_revalidate)
2977 ceph_queue_revalidate(inode);
2978
Sage Weil3c6f6b72010-02-09 15:24:44 -08002979 if (writeback)
Sage Weila8599bd2009-10-06 11:31:12 -07002980 /*
2981 * queue inode for writeback: we can't actually call
2982 * filemap_write_and_wait, etc. from message handler
2983 * context.
2984 */
Sage Weil3c6f6b72010-02-09 15:24:44 -08002985 ceph_queue_writeback(inode);
2986 if (queue_invalidate)
2987 ceph_queue_invalidate(inode);
Yan, Zhengca20c992013-07-21 10:07:51 +08002988 if (deleted_inode)
2989 invalidate_aliases(inode);
Sage Weila8599bd2009-10-06 11:31:12 -07002990 if (wake)
Yehuda Sadeh03066f22010-07-27 13:11:08 -07002991 wake_up_all(&ci->i_cap_wq);
Sage Weil15637c82010-03-16 13:42:00 -07002992
2993 if (check_caps == 1)
2994 ceph_check_caps(ci, CHECK_CAPS_NODELAY|CHECK_CAPS_AUTHONLY,
2995 session);
2996 else if (check_caps == 2)
2997 ceph_check_caps(ci, CHECK_CAPS_NODELAY, session);
2998 else
2999 mutex_unlock(&session->s_mutex);
Sage Weila8599bd2009-10-06 11:31:12 -07003000}
3001
3002/*
3003 * Handle FLUSH_ACK from MDS, indicating that metadata we sent to the
3004 * MDS has been safely committed.
3005 */
Sage Weil6df058c2009-12-22 11:24:33 -08003006static void handle_cap_flush_ack(struct inode *inode, u64 flush_tid,
Sage Weila8599bd2009-10-06 11:31:12 -07003007 struct ceph_mds_caps *m,
3008 struct ceph_mds_session *session,
3009 struct ceph_cap *cap)
Sage Weilbe655592011-11-30 09:47:09 -08003010 __releases(ci->i_ceph_lock)
Sage Weila8599bd2009-10-06 11:31:12 -07003011{
3012 struct ceph_inode_info *ci = ceph_inode(inode);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07003013 struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
Yan, Zheng553adfd2015-06-09 15:48:57 +08003014 struct ceph_cap_flush *cf;
3015 struct rb_node *n;
3016 LIST_HEAD(to_remove);
Sage Weila8599bd2009-10-06 11:31:12 -07003017 unsigned seq = le32_to_cpu(m->seq);
3018 int dirty = le32_to_cpu(m->dirty);
3019 int cleaned = 0;
Sage Weilafcdaea2009-10-14 14:27:38 -07003020 int drop = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07003021
Yan, Zheng553adfd2015-06-09 15:48:57 +08003022 n = rb_first(&ci->i_cap_flush_tree);
3023 while (n) {
3024 cf = rb_entry(n, struct ceph_cap_flush, i_node);
3025 n = rb_next(&cf->i_node);
3026 if (cf->tid == flush_tid)
3027 cleaned = cf->caps;
3028 if (cf->tid <= flush_tid) {
3029 rb_erase(&cf->i_node, &ci->i_cap_flush_tree);
3030 list_add_tail(&cf->list, &to_remove);
3031 } else {
3032 cleaned &= ~cf->caps;
3033 if (!cleaned)
3034 break;
3035 }
3036 }
Sage Weila8599bd2009-10-06 11:31:12 -07003037
3038 dout("handle_cap_flush_ack inode %p mds%d seq %d on %s cleaned %s,"
3039 " flushing %s -> %s\n",
3040 inode, session->s_mds, seq, ceph_cap_string(dirty),
3041 ceph_cap_string(cleaned), ceph_cap_string(ci->i_flushing_caps),
3042 ceph_cap_string(ci->i_flushing_caps & ~cleaned));
3043
Yan, Zheng8310b082015-06-09 17:20:12 +08003044 if (list_empty(&to_remove) && !cleaned)
Sage Weila8599bd2009-10-06 11:31:12 -07003045 goto out;
3046
Sage Weila8599bd2009-10-06 11:31:12 -07003047 ci->i_flushing_caps &= ~cleaned;
Sage Weila8599bd2009-10-06 11:31:12 -07003048
3049 spin_lock(&mdsc->cap_dirty_lock);
Yan, Zheng8310b082015-06-09 17:20:12 +08003050
3051 if (!list_empty(&to_remove)) {
3052 list_for_each_entry(cf, &to_remove, list)
3053 rb_erase(&cf->g_node, &mdsc->cap_flush_tree);
3054
3055 n = rb_first(&mdsc->cap_flush_tree);
3056 cf = n ? rb_entry(n, struct ceph_cap_flush, g_node) : NULL;
3057 if (!cf || cf->tid > flush_tid)
3058 wake_up_all(&mdsc->cap_flushing_wq);
3059 }
3060
Sage Weila8599bd2009-10-06 11:31:12 -07003061 if (ci->i_flushing_caps == 0) {
3062 list_del_init(&ci->i_flushing_item);
3063 if (!list_empty(&session->s_cap_flushing))
3064 dout(" mds%d still flushing cap on %p\n",
3065 session->s_mds,
3066 &list_entry(session->s_cap_flushing.next,
3067 struct ceph_inode_info,
3068 i_flushing_item)->vfs_inode);
3069 mdsc->num_cap_flushing--;
Sage Weila8599bd2009-10-06 11:31:12 -07003070 dout(" inode %p now !flushing\n", inode);
Sage Weilafcdaea2009-10-14 14:27:38 -07003071
3072 if (ci->i_dirty_caps == 0) {
3073 dout(" inode %p now clean\n", inode);
3074 BUG_ON(!list_empty(&ci->i_dirty_item));
3075 drop = 1;
Yan, Zheng5dda377c2015-04-30 14:40:54 +08003076 if (ci->i_wr_ref == 0 &&
3077 ci->i_wrbuffer_ref_head == 0) {
Sage Weil7d8cb262010-08-24 08:44:16 -07003078 BUG_ON(!ci->i_head_snapc);
3079 ceph_put_snap_context(ci->i_head_snapc);
3080 ci->i_head_snapc = NULL;
3081 }
Sage Weil76e3b392009-10-15 18:13:53 -07003082 } else {
3083 BUG_ON(list_empty(&ci->i_dirty_item));
Sage Weilafcdaea2009-10-14 14:27:38 -07003084 }
Sage Weila8599bd2009-10-06 11:31:12 -07003085 }
3086 spin_unlock(&mdsc->cap_dirty_lock);
Yehuda Sadeh03066f22010-07-27 13:11:08 -07003087 wake_up_all(&ci->i_cap_wq);
Sage Weila8599bd2009-10-06 11:31:12 -07003088
3089out:
Sage Weilbe655592011-11-30 09:47:09 -08003090 spin_unlock(&ci->i_ceph_lock);
Yan, Zheng553adfd2015-06-09 15:48:57 +08003091
3092 while (!list_empty(&to_remove)) {
3093 cf = list_first_entry(&to_remove,
3094 struct ceph_cap_flush, list);
3095 list_del(&cf->list);
Yan, Zhengf66fd9f2015-06-10 17:26:13 +08003096 ceph_free_cap_flush(cf);
Yan, Zheng553adfd2015-06-09 15:48:57 +08003097 }
Sage Weilafcdaea2009-10-14 14:27:38 -07003098 if (drop)
Sage Weila8599bd2009-10-06 11:31:12 -07003099 iput(inode);
3100}
3101
3102/*
3103 * Handle FLUSHSNAP_ACK. MDS has flushed snap data to disk and we can
3104 * throw away our cap_snap.
3105 *
3106 * Caller hold s_mutex.
3107 */
Sage Weil6df058c2009-12-22 11:24:33 -08003108static void handle_cap_flushsnap_ack(struct inode *inode, u64 flush_tid,
Sage Weila8599bd2009-10-06 11:31:12 -07003109 struct ceph_mds_caps *m,
3110 struct ceph_mds_session *session)
3111{
3112 struct ceph_inode_info *ci = ceph_inode(inode);
Yan, Zhengaffbc192015-05-05 21:22:13 +08003113 struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
Sage Weila8599bd2009-10-06 11:31:12 -07003114 u64 follows = le64_to_cpu(m->snap_follows);
Sage Weila8599bd2009-10-06 11:31:12 -07003115 struct ceph_cap_snap *capsnap;
3116 int drop = 0;
3117
3118 dout("handle_cap_flushsnap_ack inode %p ci %p mds%d follows %lld\n",
3119 inode, ci, session->s_mds, follows);
3120
Sage Weilbe655592011-11-30 09:47:09 -08003121 spin_lock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07003122 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
3123 if (capsnap->follows == follows) {
3124 if (capsnap->flush_tid != flush_tid) {
3125 dout(" cap_snap %p follows %lld tid %lld !="
3126 " %lld\n", capsnap, follows,
3127 flush_tid, capsnap->flush_tid);
3128 break;
3129 }
3130 WARN_ON(capsnap->dirty_pages || capsnap->writing);
Sage Weil819ccbf2010-04-01 09:33:46 -07003131 dout(" removing %p cap_snap %p follows %lld\n",
3132 inode, capsnap, follows);
Sage Weila8599bd2009-10-06 11:31:12 -07003133 ceph_put_snap_context(capsnap->context);
3134 list_del(&capsnap->ci_item);
3135 list_del(&capsnap->flushing_item);
3136 ceph_put_cap_snap(capsnap);
Yan, Zhengaffbc192015-05-05 21:22:13 +08003137 wake_up_all(&mdsc->cap_flushing_wq);
Sage Weila8599bd2009-10-06 11:31:12 -07003138 drop = 1;
3139 break;
3140 } else {
3141 dout(" skipping cap_snap %p follows %lld\n",
3142 capsnap, capsnap->follows);
3143 }
3144 }
Sage Weilbe655592011-11-30 09:47:09 -08003145 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07003146 if (drop)
3147 iput(inode);
3148}
3149
3150/*
3151 * Handle TRUNC from MDS, indicating file truncation.
3152 *
3153 * caller hold s_mutex.
3154 */
3155static void handle_cap_trunc(struct inode *inode,
3156 struct ceph_mds_caps *trunc,
3157 struct ceph_mds_session *session)
Sage Weilbe655592011-11-30 09:47:09 -08003158 __releases(ci->i_ceph_lock)
Sage Weila8599bd2009-10-06 11:31:12 -07003159{
3160 struct ceph_inode_info *ci = ceph_inode(inode);
3161 int mds = session->s_mds;
3162 int seq = le32_to_cpu(trunc->seq);
3163 u32 truncate_seq = le32_to_cpu(trunc->truncate_seq);
3164 u64 truncate_size = le64_to_cpu(trunc->truncate_size);
3165 u64 size = le64_to_cpu(trunc->size);
3166 int implemented = 0;
3167 int dirty = __ceph_caps_dirty(ci);
3168 int issued = __ceph_caps_issued(ceph_inode(inode), &implemented);
3169 int queue_trunc = 0;
3170
3171 issued |= implemented | dirty;
3172
3173 dout("handle_cap_trunc inode %p mds%d seq %d to %lld seq %d\n",
3174 inode, mds, seq, truncate_size, truncate_seq);
3175 queue_trunc = ceph_fill_file_size(inode, issued,
3176 truncate_seq, truncate_size, size);
Sage Weilbe655592011-11-30 09:47:09 -08003177 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07003178
Milosz Tanski99ccbd22013-08-21 17:29:54 -04003179 if (queue_trunc) {
Sage Weil3c6f6b72010-02-09 15:24:44 -08003180 ceph_queue_vmtruncate(inode);
Milosz Tanski99ccbd22013-08-21 17:29:54 -04003181 ceph_fscache_invalidate(inode);
3182 }
Sage Weila8599bd2009-10-06 11:31:12 -07003183}
3184
3185/*
3186 * Handle EXPORT from MDS. Cap is being migrated _from_ this mds to a
3187 * different one. If we are the most recent migration we've seen (as
3188 * indicated by mseq), make note of the migrating cap bits for the
3189 * duration (until we see the corresponding IMPORT).
3190 *
3191 * caller holds s_mutex
3192 */
3193static void handle_cap_export(struct inode *inode, struct ceph_mds_caps *ex,
Yan, Zheng11df2df2013-11-24 14:44:38 +08003194 struct ceph_mds_cap_peer *ph,
3195 struct ceph_mds_session *session)
Sage Weila8599bd2009-10-06 11:31:12 -07003196{
Sage Weildb354052011-05-24 11:46:31 -07003197 struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
Yan, Zheng11df2df2013-11-24 14:44:38 +08003198 struct ceph_mds_session *tsession = NULL;
Yan, Zhengd9df2782014-04-18 09:57:11 +08003199 struct ceph_cap *cap, *tcap, *new_cap = NULL;
Sage Weila8599bd2009-10-06 11:31:12 -07003200 struct ceph_inode_info *ci = ceph_inode(inode);
Yan, Zheng11df2df2013-11-24 14:44:38 +08003201 u64 t_cap_id;
Sage Weila8599bd2009-10-06 11:31:12 -07003202 unsigned mseq = le32_to_cpu(ex->migrate_seq);
Yan, Zheng11df2df2013-11-24 14:44:38 +08003203 unsigned t_seq, t_mseq;
3204 int target, issued;
3205 int mds = session->s_mds;
Sage Weila8599bd2009-10-06 11:31:12 -07003206
Yan, Zheng11df2df2013-11-24 14:44:38 +08003207 if (ph) {
3208 t_cap_id = le64_to_cpu(ph->cap_id);
3209 t_seq = le32_to_cpu(ph->seq);
3210 t_mseq = le32_to_cpu(ph->mseq);
3211 target = le32_to_cpu(ph->mds);
3212 } else {
3213 t_cap_id = t_seq = t_mseq = 0;
3214 target = -1;
Sage Weila8599bd2009-10-06 11:31:12 -07003215 }
3216
Yan, Zheng11df2df2013-11-24 14:44:38 +08003217 dout("handle_cap_export inode %p ci %p mds%d mseq %d target %d\n",
3218 inode, ci, mds, mseq, target);
3219retry:
3220 spin_lock(&ci->i_ceph_lock);
3221 cap = __get_cap_for_mds(ci, mds);
Yan, Zhengca665e02014-04-21 15:46:37 +08003222 if (!cap || cap->cap_id != le64_to_cpu(ex->cap_id))
Yan, Zheng11df2df2013-11-24 14:44:38 +08003223 goto out_unlock;
Sage Weil154f42c2010-06-21 13:45:04 -07003224
Yan, Zheng11df2df2013-11-24 14:44:38 +08003225 if (target < 0) {
3226 __ceph_remove_cap(cap, false);
3227 goto out_unlock;
3228 }
Sage Weildb354052011-05-24 11:46:31 -07003229
Yan, Zheng11df2df2013-11-24 14:44:38 +08003230 /*
3231 * now we know we haven't received the cap import message yet
3232 * because the exported cap still exist.
3233 */
3234
3235 issued = cap->issued;
3236 WARN_ON(issued != cap->implemented);
3237
3238 tcap = __get_cap_for_mds(ci, target);
3239 if (tcap) {
3240 /* already have caps from the target */
3241 if (tcap->cap_id != t_cap_id ||
3242 ceph_seq_cmp(tcap->seq, t_seq) < 0) {
3243 dout(" updating import cap %p mds%d\n", tcap, target);
3244 tcap->cap_id = t_cap_id;
3245 tcap->seq = t_seq - 1;
3246 tcap->issue_seq = t_seq - 1;
3247 tcap->mseq = t_mseq;
3248 tcap->issued |= issued;
3249 tcap->implemented |= issued;
3250 if (cap == ci->i_auth_cap)
3251 ci->i_auth_cap = tcap;
3252 if (ci->i_flushing_caps && ci->i_auth_cap == tcap) {
3253 spin_lock(&mdsc->cap_dirty_lock);
3254 list_move_tail(&ci->i_flushing_item,
3255 &tcap->session->s_cap_flushing);
3256 spin_unlock(&mdsc->cap_dirty_lock);
Sage Weildb354052011-05-24 11:46:31 -07003257 }
Sage Weila8599bd2009-10-06 11:31:12 -07003258 }
Yan, Zhenga096b092013-09-22 10:15:58 +08003259 __ceph_remove_cap(cap, false);
Yan, Zheng11df2df2013-11-24 14:44:38 +08003260 goto out_unlock;
Yan, Zhengd9df2782014-04-18 09:57:11 +08003261 } else if (tsession) {
Yan, Zheng11df2df2013-11-24 14:44:38 +08003262 /* add placeholder for the export tagert */
Yan, Zhengd9df2782014-04-18 09:57:11 +08003263 int flag = (cap == ci->i_auth_cap) ? CEPH_CAP_FLAG_AUTH : 0;
Yan, Zheng11df2df2013-11-24 14:44:38 +08003264 ceph_add_cap(inode, tsession, t_cap_id, -1, issued, 0,
Yan, Zhengd9df2782014-04-18 09:57:11 +08003265 t_seq - 1, t_mseq, (u64)-1, flag, &new_cap);
3266
3267 __ceph_remove_cap(cap, false);
3268 goto out_unlock;
Yan, Zheng11df2df2013-11-24 14:44:38 +08003269 }
Sage Weila8599bd2009-10-06 11:31:12 -07003270
Sage Weilbe655592011-11-30 09:47:09 -08003271 spin_unlock(&ci->i_ceph_lock);
Yan, Zheng11df2df2013-11-24 14:44:38 +08003272 mutex_unlock(&session->s_mutex);
3273
3274 /* open target session */
3275 tsession = ceph_mdsc_open_export_target_session(mdsc, target);
3276 if (!IS_ERR(tsession)) {
3277 if (mds > target) {
3278 mutex_lock(&session->s_mutex);
3279 mutex_lock_nested(&tsession->s_mutex,
3280 SINGLE_DEPTH_NESTING);
3281 } else {
3282 mutex_lock(&tsession->s_mutex);
3283 mutex_lock_nested(&session->s_mutex,
3284 SINGLE_DEPTH_NESTING);
3285 }
Yan, Zhengd9df2782014-04-18 09:57:11 +08003286 new_cap = ceph_get_cap(mdsc, NULL);
Yan, Zheng11df2df2013-11-24 14:44:38 +08003287 } else {
3288 WARN_ON(1);
3289 tsession = NULL;
3290 target = -1;
3291 }
3292 goto retry;
3293
3294out_unlock:
3295 spin_unlock(&ci->i_ceph_lock);
3296 mutex_unlock(&session->s_mutex);
3297 if (tsession) {
3298 mutex_unlock(&tsession->s_mutex);
3299 ceph_put_mds_session(tsession);
3300 }
Yan, Zhengd9df2782014-04-18 09:57:11 +08003301 if (new_cap)
3302 ceph_put_cap(mdsc, new_cap);
Sage Weila8599bd2009-10-06 11:31:12 -07003303}
3304
3305/*
Yan, Zheng2cd698b2014-04-18 13:20:27 +08003306 * Handle cap IMPORT.
Sage Weila8599bd2009-10-06 11:31:12 -07003307 *
Yan, Zheng2cd698b2014-04-18 13:20:27 +08003308 * caller holds s_mutex. acquires i_ceph_lock
Sage Weila8599bd2009-10-06 11:31:12 -07003309 */
3310static void handle_cap_import(struct ceph_mds_client *mdsc,
3311 struct inode *inode, struct ceph_mds_caps *im,
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003312 struct ceph_mds_cap_peer *ph,
Sage Weila8599bd2009-10-06 11:31:12 -07003313 struct ceph_mds_session *session,
Yan, Zheng2cd698b2014-04-18 13:20:27 +08003314 struct ceph_cap **target_cap, int *old_issued)
3315 __acquires(ci->i_ceph_lock)
Sage Weila8599bd2009-10-06 11:31:12 -07003316{
3317 struct ceph_inode_info *ci = ceph_inode(inode);
Yan, Zheng2cd698b2014-04-18 13:20:27 +08003318 struct ceph_cap *cap, *ocap, *new_cap = NULL;
Sage Weila8599bd2009-10-06 11:31:12 -07003319 int mds = session->s_mds;
Yan, Zheng2cd698b2014-04-18 13:20:27 +08003320 int issued;
3321 unsigned caps = le32_to_cpu(im->caps);
Sage Weila8599bd2009-10-06 11:31:12 -07003322 unsigned wanted = le32_to_cpu(im->wanted);
3323 unsigned seq = le32_to_cpu(im->seq);
3324 unsigned mseq = le32_to_cpu(im->migrate_seq);
3325 u64 realmino = le64_to_cpu(im->realm);
3326 u64 cap_id = le64_to_cpu(im->cap_id);
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003327 u64 p_cap_id;
3328 int peer;
Sage Weila8599bd2009-10-06 11:31:12 -07003329
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003330 if (ph) {
3331 p_cap_id = le64_to_cpu(ph->cap_id);
3332 peer = le32_to_cpu(ph->mds);
Sage Weila8599bd2009-10-06 11:31:12 -07003333 } else {
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003334 p_cap_id = 0;
3335 peer = -1;
Sage Weila8599bd2009-10-06 11:31:12 -07003336 }
3337
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003338 dout("handle_cap_import inode %p ci %p mds%d mseq %d peer %d\n",
3339 inode, ci, mds, mseq, peer);
3340
Yan, Zhengd9df2782014-04-18 09:57:11 +08003341retry:
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003342 spin_lock(&ci->i_ceph_lock);
Yan, Zhengd9df2782014-04-18 09:57:11 +08003343 cap = __get_cap_for_mds(ci, mds);
3344 if (!cap) {
3345 if (!new_cap) {
3346 spin_unlock(&ci->i_ceph_lock);
3347 new_cap = ceph_get_cap(mdsc, NULL);
3348 goto retry;
3349 }
Yan, Zheng2cd698b2014-04-18 13:20:27 +08003350 cap = new_cap;
3351 } else {
3352 if (new_cap) {
3353 ceph_put_cap(mdsc, new_cap);
3354 new_cap = NULL;
3355 }
Yan, Zhengd9df2782014-04-18 09:57:11 +08003356 }
3357
Yan, Zheng2cd698b2014-04-18 13:20:27 +08003358 __ceph_caps_issued(ci, &issued);
3359 issued |= __ceph_caps_dirty(ci);
3360
3361 ceph_add_cap(inode, session, cap_id, -1, caps, wanted, seq, mseq,
Yan, Zhengd9df2782014-04-18 09:57:11 +08003362 realmino, CEPH_CAP_FLAG_AUTH, &new_cap);
3363
Yan, Zheng2cd698b2014-04-18 13:20:27 +08003364 ocap = peer >= 0 ? __get_cap_for_mds(ci, peer) : NULL;
3365 if (ocap && ocap->cap_id == p_cap_id) {
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003366 dout(" remove export cap %p mds%d flags %d\n",
Yan, Zheng2cd698b2014-04-18 13:20:27 +08003367 ocap, peer, ph->flags);
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003368 if ((ph->flags & CEPH_CAP_FLAG_AUTH) &&
Yan, Zheng2cd698b2014-04-18 13:20:27 +08003369 (ocap->seq != le32_to_cpu(ph->seq) ||
3370 ocap->mseq != le32_to_cpu(ph->mseq))) {
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003371 pr_err("handle_cap_import: mismatched seq/mseq: "
3372 "ino (%llx.%llx) mds%d seq %d mseq %d "
3373 "importer mds%d has peer seq %d mseq %d\n",
Yan, Zheng2cd698b2014-04-18 13:20:27 +08003374 ceph_vinop(inode), peer, ocap->seq,
3375 ocap->mseq, mds, le32_to_cpu(ph->seq),
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003376 le32_to_cpu(ph->mseq));
3377 }
Yan, Zheng2cd698b2014-04-18 13:20:27 +08003378 __ceph_remove_cap(ocap, (ph->flags & CEPH_CAP_FLAG_RELEASE));
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003379 }
3380
3381 /* make sure we re-request max_size, if necessary */
3382 ci->i_wanted_max_size = 0;
3383 ci->i_requested_max_size = 0;
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003384
Yan, Zheng2cd698b2014-04-18 13:20:27 +08003385 *old_issued = issued;
3386 *target_cap = cap;
Sage Weila8599bd2009-10-06 11:31:12 -07003387}
3388
3389/*
3390 * Handle a caps message from the MDS.
3391 *
3392 * Identify the appropriate session, inode, and call the right handler
3393 * based on the cap op.
3394 */
3395void ceph_handle_caps(struct ceph_mds_session *session,
3396 struct ceph_msg *msg)
3397{
3398 struct ceph_mds_client *mdsc = session->s_mdsc;
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07003399 struct super_block *sb = mdsc->fsc->sb;
Sage Weila8599bd2009-10-06 11:31:12 -07003400 struct inode *inode;
Sage Weilbe655592011-11-30 09:47:09 -08003401 struct ceph_inode_info *ci;
Sage Weila8599bd2009-10-06 11:31:12 -07003402 struct ceph_cap *cap;
3403 struct ceph_mds_caps *h;
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003404 struct ceph_mds_cap_peer *peer = NULL;
Yan, Zheng982d6012014-12-23 15:30:54 +08003405 struct ceph_snap_realm *realm;
Sage Weil2600d2d2010-02-22 15:12:16 -08003406 int mds = session->s_mds;
Yan, Zheng2cd698b2014-04-18 13:20:27 +08003407 int op, issued;
Sage Weil3d7ded42010-06-09 16:47:10 -07003408 u32 seq, mseq;
Sage Weila8599bd2009-10-06 11:31:12 -07003409 struct ceph_vino vino;
3410 u64 cap_id;
3411 u64 size, max_size;
Sage Weil6df058c2009-12-22 11:24:33 -08003412 u64 tid;
Yan, Zhengfb01d1f2014-11-14 21:29:55 +08003413 u64 inline_version = 0;
3414 void *inline_data = NULL;
3415 u32 inline_len = 0;
Sage Weil70edb552010-03-01 13:20:50 -08003416 void *snaptrace;
Sage Weilce1fbc82010-08-02 15:09:39 -07003417 size_t snaptrace_len;
Yan, Zhengfb01d1f2014-11-14 21:29:55 +08003418 void *p, *end;
Sage Weila8599bd2009-10-06 11:31:12 -07003419
3420 dout("handle_caps from mds%d\n", mds);
3421
3422 /* decode */
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003423 end = msg->front.iov_base + msg->front.iov_len;
Sage Weil6df058c2009-12-22 11:24:33 -08003424 tid = le64_to_cpu(msg->hdr.tid);
Sage Weila8599bd2009-10-06 11:31:12 -07003425 if (msg->front.iov_len < sizeof(*h))
3426 goto bad;
3427 h = msg->front.iov_base;
3428 op = le32_to_cpu(h->op);
3429 vino.ino = le64_to_cpu(h->ino);
3430 vino.snap = CEPH_NOSNAP;
3431 cap_id = le64_to_cpu(h->cap_id);
3432 seq = le32_to_cpu(h->seq);
Sage Weil3d7ded42010-06-09 16:47:10 -07003433 mseq = le32_to_cpu(h->migrate_seq);
Sage Weila8599bd2009-10-06 11:31:12 -07003434 size = le64_to_cpu(h->size);
3435 max_size = le64_to_cpu(h->max_size);
3436
Sage Weilce1fbc82010-08-02 15:09:39 -07003437 snaptrace = h + 1;
3438 snaptrace_len = le32_to_cpu(h->snap_trace_len);
Yan, Zhengfb01d1f2014-11-14 21:29:55 +08003439 p = snaptrace + snaptrace_len;
Sage Weilce1fbc82010-08-02 15:09:39 -07003440
3441 if (le16_to_cpu(msg->hdr.version) >= 2) {
Yan, Zhengfb01d1f2014-11-14 21:29:55 +08003442 u32 flock_len;
Sage Weilce1fbc82010-08-02 15:09:39 -07003443 ceph_decode_32_safe(&p, end, flock_len, bad);
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003444 if (p + flock_len > end)
3445 goto bad;
Yan, Zhengfb01d1f2014-11-14 21:29:55 +08003446 p += flock_len;
Sage Weilce1fbc82010-08-02 15:09:39 -07003447 }
3448
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003449 if (le16_to_cpu(msg->hdr.version) >= 3) {
3450 if (op == CEPH_CAP_OP_IMPORT) {
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003451 if (p + sizeof(*peer) > end)
3452 goto bad;
3453 peer = p;
Yan, Zhengfb01d1f2014-11-14 21:29:55 +08003454 p += sizeof(*peer);
Yan, Zheng11df2df2013-11-24 14:44:38 +08003455 } else if (op == CEPH_CAP_OP_EXPORT) {
3456 /* recorded in unused fields */
3457 peer = (void *)&h->size;
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003458 }
3459 }
3460
Yan, Zhengfb01d1f2014-11-14 21:29:55 +08003461 if (le16_to_cpu(msg->hdr.version) >= 4) {
3462 ceph_decode_64_safe(&p, end, inline_version, bad);
3463 ceph_decode_32_safe(&p, end, inline_len, bad);
3464 if (p + inline_len > end)
3465 goto bad;
3466 inline_data = p;
3467 p += inline_len;
3468 }
3469
Yan, Zheng6cd3bca2014-09-17 07:45:12 +08003470 /* lookup ino */
3471 inode = ceph_find_inode(sb, vino);
3472 ci = ceph_inode(inode);
3473 dout(" op %s ino %llx.%llx inode %p\n", ceph_cap_op_name(op), vino.ino,
3474 vino.snap, inode);
3475
Sage Weila8599bd2009-10-06 11:31:12 -07003476 mutex_lock(&session->s_mutex);
3477 session->s_seq++;
3478 dout(" mds%d seq %lld cap seq %u\n", session->s_mds, session->s_seq,
3479 (unsigned)seq);
3480
Sage Weila8599bd2009-10-06 11:31:12 -07003481 if (!inode) {
3482 dout(" i don't have ino %llx\n", vino.ino);
Sage Weil3d7ded42010-06-09 16:47:10 -07003483
Yan, Zhenga096b092013-09-22 10:15:58 +08003484 if (op == CEPH_CAP_OP_IMPORT) {
Yan, Zheng745a8e32015-05-14 17:22:42 +08003485 cap = ceph_get_cap(mdsc, NULL);
3486 cap->cap_ino = vino.ino;
3487 cap->queue_release = 1;
3488 cap->cap_id = cap_id;
3489 cap->mseq = mseq;
3490 cap->seq = seq;
Yan, Zhenga096b092013-09-22 10:15:58 +08003491 spin_lock(&session->s_cap_lock);
Yan, Zheng745a8e32015-05-14 17:22:42 +08003492 list_add_tail(&cap->session_caps,
3493 &session->s_cap_releases);
3494 session->s_num_cap_releases++;
Yan, Zhenga096b092013-09-22 10:15:58 +08003495 spin_unlock(&session->s_cap_lock);
3496 }
Greg Farnum21b559d2010-10-06 15:46:30 -07003497 goto flush_cap_releases;
Sage Weila8599bd2009-10-06 11:31:12 -07003498 }
3499
3500 /* these will work even if we don't have a cap yet */
3501 switch (op) {
3502 case CEPH_CAP_OP_FLUSHSNAP_ACK:
Sage Weil6df058c2009-12-22 11:24:33 -08003503 handle_cap_flushsnap_ack(inode, tid, h, session);
Sage Weila8599bd2009-10-06 11:31:12 -07003504 goto done;
3505
3506 case CEPH_CAP_OP_EXPORT:
Yan, Zheng11df2df2013-11-24 14:44:38 +08003507 handle_cap_export(inode, h, peer, session);
3508 goto done_unlocked;
Sage Weila8599bd2009-10-06 11:31:12 -07003509
3510 case CEPH_CAP_OP_IMPORT:
Yan, Zheng982d6012014-12-23 15:30:54 +08003511 realm = NULL;
3512 if (snaptrace_len) {
3513 down_write(&mdsc->snap_rwsem);
3514 ceph_update_snap_trace(mdsc, snaptrace,
3515 snaptrace + snaptrace_len,
3516 false, &realm);
3517 downgrade_write(&mdsc->snap_rwsem);
3518 } else {
3519 down_read(&mdsc->snap_rwsem);
3520 }
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003521 handle_cap_import(mdsc, inode, h, peer, session,
Yan, Zheng2cd698b2014-04-18 13:20:27 +08003522 &cap, &issued);
Yan, Zheng982d6012014-12-23 15:30:54 +08003523 handle_cap_grant(mdsc, inode, h,
Yan, Zhengfb01d1f2014-11-14 21:29:55 +08003524 inline_version, inline_data, inline_len,
Yan, Zheng2cd698b2014-04-18 13:20:27 +08003525 msg->middle, session, cap, issued);
Yan, Zheng982d6012014-12-23 15:30:54 +08003526 if (realm)
3527 ceph_put_snap_realm(mdsc, realm);
Yan, Zheng2cd698b2014-04-18 13:20:27 +08003528 goto done_unlocked;
Sage Weila8599bd2009-10-06 11:31:12 -07003529 }
3530
3531 /* the rest require a cap */
Sage Weilbe655592011-11-30 09:47:09 -08003532 spin_lock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07003533 cap = __get_cap_for_mds(ceph_inode(inode), mds);
3534 if (!cap) {
Sage Weil9dbd4122010-06-10 13:21:20 -07003535 dout(" no cap on %p ino %llx.%llx from mds%d\n",
Sage Weila8599bd2009-10-06 11:31:12 -07003536 inode, ceph_ino(inode), ceph_snap(inode), mds);
Sage Weilbe655592011-11-30 09:47:09 -08003537 spin_unlock(&ci->i_ceph_lock);
Greg Farnum21b559d2010-10-06 15:46:30 -07003538 goto flush_cap_releases;
Sage Weila8599bd2009-10-06 11:31:12 -07003539 }
3540
Sage Weilbe655592011-11-30 09:47:09 -08003541 /* note that each of these drops i_ceph_lock for us */
Sage Weila8599bd2009-10-06 11:31:12 -07003542 switch (op) {
3543 case CEPH_CAP_OP_REVOKE:
3544 case CEPH_CAP_OP_GRANT:
Yan, Zheng2cd698b2014-04-18 13:20:27 +08003545 __ceph_caps_issued(ci, &issued);
3546 issued |= __ceph_caps_dirty(ci);
Yan, Zheng982d6012014-12-23 15:30:54 +08003547 handle_cap_grant(mdsc, inode, h,
Yan, Zhengfb01d1f2014-11-14 21:29:55 +08003548 inline_version, inline_data, inline_len,
3549 msg->middle, session, cap, issued);
Sage Weil15637c82010-03-16 13:42:00 -07003550 goto done_unlocked;
Sage Weila8599bd2009-10-06 11:31:12 -07003551
3552 case CEPH_CAP_OP_FLUSH_ACK:
Sage Weil6df058c2009-12-22 11:24:33 -08003553 handle_cap_flush_ack(inode, tid, h, session, cap);
Sage Weila8599bd2009-10-06 11:31:12 -07003554 break;
3555
3556 case CEPH_CAP_OP_TRUNC:
3557 handle_cap_trunc(inode, h, session);
3558 break;
3559
3560 default:
Sage Weilbe655592011-11-30 09:47:09 -08003561 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07003562 pr_err("ceph_handle_caps: unknown cap op %d %s\n", op,
3563 ceph_cap_op_name(op));
3564 }
3565
Greg Farnum21b559d2010-10-06 15:46:30 -07003566 goto done;
3567
3568flush_cap_releases:
3569 /*
Yan, Zheng745a8e32015-05-14 17:22:42 +08003570 * send any cap release message to try to move things
Greg Farnum21b559d2010-10-06 15:46:30 -07003571 * along for the mds (who clearly thinks we still have this
3572 * cap).
3573 */
Greg Farnum21b559d2010-10-06 15:46:30 -07003574 ceph_send_cap_releases(mdsc, session);
3575
Sage Weila8599bd2009-10-06 11:31:12 -07003576done:
Sage Weil15637c82010-03-16 13:42:00 -07003577 mutex_unlock(&session->s_mutex);
3578done_unlocked:
SF Markus Elfringe96a6502014-11-02 15:20:59 +01003579 iput(inode);
Sage Weila8599bd2009-10-06 11:31:12 -07003580 return;
3581
3582bad:
3583 pr_err("ceph_handle_caps: corrupt message\n");
Sage Weil9ec7cab2009-12-14 15:13:47 -08003584 ceph_msg_dump(msg);
Sage Weila8599bd2009-10-06 11:31:12 -07003585 return;
3586}
3587
3588/*
3589 * Delayed work handler to process end of delayed cap release LRU list.
3590 */
Sage Weilafcdaea2009-10-14 14:27:38 -07003591void ceph_check_delayed_caps(struct ceph_mds_client *mdsc)
Sage Weila8599bd2009-10-06 11:31:12 -07003592{
3593 struct ceph_inode_info *ci;
3594 int flags = CHECK_CAPS_NODELAY;
3595
Sage Weila8599bd2009-10-06 11:31:12 -07003596 dout("check_delayed_caps\n");
3597 while (1) {
3598 spin_lock(&mdsc->cap_delay_lock);
3599 if (list_empty(&mdsc->cap_delay_list))
3600 break;
3601 ci = list_first_entry(&mdsc->cap_delay_list,
3602 struct ceph_inode_info,
3603 i_cap_delay_list);
3604 if ((ci->i_ceph_flags & CEPH_I_FLUSH) == 0 &&
3605 time_before(jiffies, ci->i_hold_caps_max))
3606 break;
3607 list_del_init(&ci->i_cap_delay_list);
3608 spin_unlock(&mdsc->cap_delay_lock);
3609 dout("check_delayed_caps on %p\n", &ci->vfs_inode);
3610 ceph_check_caps(ci, flags, NULL);
3611 }
3612 spin_unlock(&mdsc->cap_delay_lock);
3613}
3614
3615/*
Sage Weilafcdaea2009-10-14 14:27:38 -07003616 * Flush all dirty caps to the mds
3617 */
3618void ceph_flush_dirty_caps(struct ceph_mds_client *mdsc)
3619{
Sage Weildb354052011-05-24 11:46:31 -07003620 struct ceph_inode_info *ci;
3621 struct inode *inode;
Sage Weilafcdaea2009-10-14 14:27:38 -07003622
3623 dout("flush_dirty_caps\n");
3624 spin_lock(&mdsc->cap_dirty_lock);
Sage Weildb354052011-05-24 11:46:31 -07003625 while (!list_empty(&mdsc->cap_dirty)) {
3626 ci = list_first_entry(&mdsc->cap_dirty, struct ceph_inode_info,
3627 i_dirty_item);
Sage Weil70b666c2011-05-27 09:24:26 -07003628 inode = &ci->vfs_inode;
3629 ihold(inode);
Sage Weildb354052011-05-24 11:46:31 -07003630 dout("flush_dirty_caps %p\n", inode);
Sage Weilafcdaea2009-10-14 14:27:38 -07003631 spin_unlock(&mdsc->cap_dirty_lock);
Sage Weil70b666c2011-05-27 09:24:26 -07003632 ceph_check_caps(ci, CHECK_CAPS_NODELAY|CHECK_CAPS_FLUSH, NULL);
3633 iput(inode);
Sage Weilafcdaea2009-10-14 14:27:38 -07003634 spin_lock(&mdsc->cap_dirty_lock);
3635 }
3636 spin_unlock(&mdsc->cap_dirty_lock);
Sage Weildb354052011-05-24 11:46:31 -07003637 dout("flush_dirty_caps done\n");
Sage Weilafcdaea2009-10-14 14:27:38 -07003638}
3639
3640/*
Sage Weila8599bd2009-10-06 11:31:12 -07003641 * Drop open file reference. If we were the last open file,
3642 * we may need to release capabilities to the MDS (or schedule
3643 * their delayed release).
3644 */
3645void ceph_put_fmode(struct ceph_inode_info *ci, int fmode)
3646{
3647 struct inode *inode = &ci->vfs_inode;
3648 int last = 0;
3649
Sage Weilbe655592011-11-30 09:47:09 -08003650 spin_lock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07003651 dout("put_fmode %p fmode %d %d -> %d\n", inode, fmode,
3652 ci->i_nr_by_mode[fmode], ci->i_nr_by_mode[fmode]-1);
3653 BUG_ON(ci->i_nr_by_mode[fmode] == 0);
3654 if (--ci->i_nr_by_mode[fmode] == 0)
3655 last++;
Sage Weilbe655592011-11-30 09:47:09 -08003656 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07003657
3658 if (last && ci->i_vino.snap == CEPH_NOSNAP)
3659 ceph_check_caps(ci, 0, NULL);
3660}
3661
3662/*
3663 * Helpers for embedding cap and dentry lease releases into mds
3664 * requests.
3665 *
3666 * @force is used by dentry_release (below) to force inclusion of a
3667 * record for the directory inode, even when there aren't any caps to
3668 * drop.
3669 */
3670int ceph_encode_inode_release(void **p, struct inode *inode,
3671 int mds, int drop, int unless, int force)
3672{
3673 struct ceph_inode_info *ci = ceph_inode(inode);
3674 struct ceph_cap *cap;
3675 struct ceph_mds_request_release *rel = *p;
Sage Weilec97f882010-06-24 15:12:37 -07003676 int used, dirty;
Sage Weila8599bd2009-10-06 11:31:12 -07003677 int ret = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07003678
Sage Weilbe655592011-11-30 09:47:09 -08003679 spin_lock(&ci->i_ceph_lock);
Sage Weil916623d2010-03-16 15:01:07 -07003680 used = __ceph_caps_used(ci);
Sage Weilec97f882010-06-24 15:12:37 -07003681 dirty = __ceph_caps_dirty(ci);
Sage Weil916623d2010-03-16 15:01:07 -07003682
Sage Weilec97f882010-06-24 15:12:37 -07003683 dout("encode_inode_release %p mds%d used|dirty %s drop %s unless %s\n",
3684 inode, mds, ceph_cap_string(used|dirty), ceph_cap_string(drop),
Sage Weil916623d2010-03-16 15:01:07 -07003685 ceph_cap_string(unless));
3686
Sage Weilec97f882010-06-24 15:12:37 -07003687 /* only drop unused, clean caps */
3688 drop &= ~(used | dirty);
Sage Weil916623d2010-03-16 15:01:07 -07003689
Sage Weila8599bd2009-10-06 11:31:12 -07003690 cap = __get_cap_for_mds(ci, mds);
3691 if (cap && __cap_is_valid(cap)) {
3692 if (force ||
3693 ((cap->issued & drop) &&
3694 (cap->issued & unless) == 0)) {
3695 if ((cap->issued & drop) &&
3696 (cap->issued & unless) == 0) {
Yan, Zhengbb137f82013-06-03 18:22:17 +08003697 int wanted = __ceph_caps_wanted(ci);
3698 if ((ci->i_ceph_flags & CEPH_I_NODELAY) == 0)
3699 wanted |= cap->mds_wanted;
3700 dout("encode_inode_release %p cap %p "
3701 "%s -> %s, wanted %s -> %s\n", inode, cap,
Sage Weila8599bd2009-10-06 11:31:12 -07003702 ceph_cap_string(cap->issued),
Yan, Zhengbb137f82013-06-03 18:22:17 +08003703 ceph_cap_string(cap->issued & ~drop),
3704 ceph_cap_string(cap->mds_wanted),
3705 ceph_cap_string(wanted));
3706
Sage Weila8599bd2009-10-06 11:31:12 -07003707 cap->issued &= ~drop;
3708 cap->implemented &= ~drop;
Yan, Zhengbb137f82013-06-03 18:22:17 +08003709 cap->mds_wanted = wanted;
Sage Weila8599bd2009-10-06 11:31:12 -07003710 } else {
3711 dout("encode_inode_release %p cap %p %s"
3712 " (force)\n", inode, cap,
3713 ceph_cap_string(cap->issued));
3714 }
3715
3716 rel->ino = cpu_to_le64(ceph_ino(inode));
3717 rel->cap_id = cpu_to_le64(cap->cap_id);
3718 rel->seq = cpu_to_le32(cap->seq);
Himangi Saraogi08a0f242014-07-23 20:11:11 +05303719 rel->issue_seq = cpu_to_le32(cap->issue_seq);
Sage Weila8599bd2009-10-06 11:31:12 -07003720 rel->mseq = cpu_to_le32(cap->mseq);
Yan, Zhengfd7b95c2014-04-17 08:02:02 +08003721 rel->caps = cpu_to_le32(cap->implemented);
Sage Weila8599bd2009-10-06 11:31:12 -07003722 rel->wanted = cpu_to_le32(cap->mds_wanted);
3723 rel->dname_len = 0;
3724 rel->dname_seq = 0;
3725 *p += sizeof(*rel);
3726 ret = 1;
3727 } else {
3728 dout("encode_inode_release %p cap %p %s\n",
3729 inode, cap, ceph_cap_string(cap->issued));
3730 }
3731 }
Sage Weilbe655592011-11-30 09:47:09 -08003732 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07003733 return ret;
3734}
3735
3736int ceph_encode_dentry_release(void **p, struct dentry *dentry,
3737 int mds, int drop, int unless)
3738{
David Howells2b0143b2015-03-17 22:25:59 +00003739 struct inode *dir = d_inode(dentry->d_parent);
Sage Weila8599bd2009-10-06 11:31:12 -07003740 struct ceph_mds_request_release *rel = *p;
3741 struct ceph_dentry_info *di = ceph_dentry(dentry);
3742 int force = 0;
3743 int ret;
3744
3745 /*
3746 * force an record for the directory caps if we have a dentry lease.
Sage Weilbe655592011-11-30 09:47:09 -08003747 * this is racy (can't take i_ceph_lock and d_lock together), but it
Sage Weila8599bd2009-10-06 11:31:12 -07003748 * doesn't have to be perfect; the mds will revoke anything we don't
3749 * release.
3750 */
3751 spin_lock(&dentry->d_lock);
3752 if (di->lease_session && di->lease_session->s_mds == mds)
3753 force = 1;
3754 spin_unlock(&dentry->d_lock);
3755
3756 ret = ceph_encode_inode_release(p, dir, mds, drop, unless, force);
3757
3758 spin_lock(&dentry->d_lock);
3759 if (ret && di->lease_session && di->lease_session->s_mds == mds) {
3760 dout("encode_dentry_release %p mds%d seq %d\n",
3761 dentry, mds, (int)di->lease_seq);
3762 rel->dname_len = cpu_to_le32(dentry->d_name.len);
3763 memcpy(*p, dentry->d_name.name, dentry->d_name.len);
3764 *p += dentry->d_name.len;
3765 rel->dname_seq = cpu_to_le32(di->lease_seq);
Sage Weil1dadcce2010-07-23 13:54:21 -07003766 __ceph_mdsc_drop_dentry_lease(dentry);
Sage Weila8599bd2009-10-06 11:31:12 -07003767 }
3768 spin_unlock(&dentry->d_lock);
3769 return ret;
3770}