blob: e9b03b51b874331fc0f5b424b361e82c208bed80 [file] [log] [blame]
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001#include <linux/ceph/ceph_debug.h>
Sage Weila8599bd2009-10-06 11:31:12 -07002
3#include <linux/fs.h>
4#include <linux/kernel.h>
5#include <linux/sched.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +09006#include <linux/slab.h>
Sage Weila8599bd2009-10-06 11:31:12 -07007#include <linux/vmalloc.h>
8#include <linux/wait.h>
Stephen Rothwellf1a3d572010-01-18 11:53:08 +11009#include <linux/writeback.h>
Sage Weila8599bd2009-10-06 11:31:12 -070010
11#include "super.h"
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -070012#include "mds_client.h"
Milosz Tanski99ccbd22013-08-21 17:29:54 -040013#include "cache.h"
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -070014#include <linux/ceph/decode.h>
15#include <linux/ceph/messenger.h>
Sage Weila8599bd2009-10-06 11:31:12 -070016
17/*
18 * Capability management
19 *
20 * The Ceph metadata servers control client access to inode metadata
21 * and file data by issuing capabilities, granting clients permission
22 * to read and/or write both inode field and file data to OSDs
23 * (storage nodes). Each capability consists of a set of bits
24 * indicating which operations are allowed.
25 *
26 * If the client holds a *_SHARED cap, the client has a coherent value
27 * that can be safely read from the cached inode.
28 *
29 * In the case of a *_EXCL (exclusive) or FILE_WR capabilities, the
30 * client is allowed to change inode attributes (e.g., file size,
31 * mtime), note its dirty state in the ceph_cap, and asynchronously
32 * flush that metadata change to the MDS.
33 *
34 * In the event of a conflicting operation (perhaps by another
35 * client), the MDS will revoke the conflicting client capabilities.
36 *
37 * In order for a client to cache an inode, it must hold a capability
38 * with at least one MDS server. When inodes are released, release
39 * notifications are batched and periodically sent en masse to the MDS
40 * cluster to release server state.
41 */
42
43
44/*
45 * Generate readable cap strings for debugging output.
46 */
47#define MAX_CAP_STR 20
48static char cap_str[MAX_CAP_STR][40];
49static DEFINE_SPINLOCK(cap_str_lock);
50static int last_cap_str;
51
52static char *gcap_string(char *s, int c)
53{
54 if (c & CEPH_CAP_GSHARED)
55 *s++ = 's';
56 if (c & CEPH_CAP_GEXCL)
57 *s++ = 'x';
58 if (c & CEPH_CAP_GCACHE)
59 *s++ = 'c';
60 if (c & CEPH_CAP_GRD)
61 *s++ = 'r';
62 if (c & CEPH_CAP_GWR)
63 *s++ = 'w';
64 if (c & CEPH_CAP_GBUFFER)
65 *s++ = 'b';
66 if (c & CEPH_CAP_GLAZYIO)
67 *s++ = 'l';
68 return s;
69}
70
71const char *ceph_cap_string(int caps)
72{
73 int i;
74 char *s;
75 int c;
76
77 spin_lock(&cap_str_lock);
78 i = last_cap_str++;
79 if (last_cap_str == MAX_CAP_STR)
80 last_cap_str = 0;
81 spin_unlock(&cap_str_lock);
82
83 s = cap_str[i];
84
85 if (caps & CEPH_CAP_PIN)
86 *s++ = 'p';
87
88 c = (caps >> CEPH_CAP_SAUTH) & 3;
89 if (c) {
90 *s++ = 'A';
91 s = gcap_string(s, c);
92 }
93
94 c = (caps >> CEPH_CAP_SLINK) & 3;
95 if (c) {
96 *s++ = 'L';
97 s = gcap_string(s, c);
98 }
99
100 c = (caps >> CEPH_CAP_SXATTR) & 3;
101 if (c) {
102 *s++ = 'X';
103 s = gcap_string(s, c);
104 }
105
106 c = caps >> CEPH_CAP_SFILE;
107 if (c) {
108 *s++ = 'F';
109 s = gcap_string(s, c);
110 }
111
112 if (s == cap_str[i])
113 *s++ = '-';
114 *s = 0;
115 return cap_str[i];
116}
117
Yehuda Sadeh37151662010-06-17 16:16:12 -0700118void ceph_caps_init(struct ceph_mds_client *mdsc)
Sage Weila8599bd2009-10-06 11:31:12 -0700119{
Yehuda Sadeh37151662010-06-17 16:16:12 -0700120 INIT_LIST_HEAD(&mdsc->caps_list);
121 spin_lock_init(&mdsc->caps_list_lock);
Sage Weila8599bd2009-10-06 11:31:12 -0700122}
123
Yehuda Sadeh37151662010-06-17 16:16:12 -0700124void ceph_caps_finalize(struct ceph_mds_client *mdsc)
Sage Weila8599bd2009-10-06 11:31:12 -0700125{
126 struct ceph_cap *cap;
127
Yehuda Sadeh37151662010-06-17 16:16:12 -0700128 spin_lock(&mdsc->caps_list_lock);
129 while (!list_empty(&mdsc->caps_list)) {
130 cap = list_first_entry(&mdsc->caps_list,
131 struct ceph_cap, caps_item);
Sage Weila8599bd2009-10-06 11:31:12 -0700132 list_del(&cap->caps_item);
133 kmem_cache_free(ceph_cap_cachep, cap);
134 }
Yehuda Sadeh37151662010-06-17 16:16:12 -0700135 mdsc->caps_total_count = 0;
136 mdsc->caps_avail_count = 0;
137 mdsc->caps_use_count = 0;
138 mdsc->caps_reserve_count = 0;
139 mdsc->caps_min_count = 0;
140 spin_unlock(&mdsc->caps_list_lock);
Sage Weil85ccce42010-02-17 10:02:43 -0800141}
142
Yehuda Sadeh37151662010-06-17 16:16:12 -0700143void ceph_adjust_min_caps(struct ceph_mds_client *mdsc, int delta)
Sage Weil85ccce42010-02-17 10:02:43 -0800144{
Yehuda Sadeh37151662010-06-17 16:16:12 -0700145 spin_lock(&mdsc->caps_list_lock);
146 mdsc->caps_min_count += delta;
147 BUG_ON(mdsc->caps_min_count < 0);
148 spin_unlock(&mdsc->caps_list_lock);
Sage Weila8599bd2009-10-06 11:31:12 -0700149}
150
majianpeng93faca62013-06-26 11:15:27 +0800151void ceph_reserve_caps(struct ceph_mds_client *mdsc,
Yehuda Sadeh37151662010-06-17 16:16:12 -0700152 struct ceph_cap_reservation *ctx, int need)
Sage Weila8599bd2009-10-06 11:31:12 -0700153{
154 int i;
155 struct ceph_cap *cap;
156 int have;
157 int alloc = 0;
158 LIST_HEAD(newcaps);
Sage Weila8599bd2009-10-06 11:31:12 -0700159
160 dout("reserve caps ctx=%p need=%d\n", ctx, need);
161
162 /* first reserve any caps that are already allocated */
Yehuda Sadeh37151662010-06-17 16:16:12 -0700163 spin_lock(&mdsc->caps_list_lock);
164 if (mdsc->caps_avail_count >= need)
Sage Weila8599bd2009-10-06 11:31:12 -0700165 have = need;
166 else
Yehuda Sadeh37151662010-06-17 16:16:12 -0700167 have = mdsc->caps_avail_count;
168 mdsc->caps_avail_count -= have;
169 mdsc->caps_reserve_count += have;
170 BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count +
171 mdsc->caps_reserve_count +
172 mdsc->caps_avail_count);
173 spin_unlock(&mdsc->caps_list_lock);
Sage Weila8599bd2009-10-06 11:31:12 -0700174
175 for (i = have; i < need; i++) {
176 cap = kmem_cache_alloc(ceph_cap_cachep, GFP_NOFS);
majianpeng93faca62013-06-26 11:15:27 +0800177 if (!cap)
178 break;
Sage Weila8599bd2009-10-06 11:31:12 -0700179 list_add(&cap->caps_item, &newcaps);
180 alloc++;
181 }
majianpeng93faca62013-06-26 11:15:27 +0800182 /* we didn't manage to reserve as much as we needed */
183 if (have + alloc != need)
184 pr_warn("reserve caps ctx=%p ENOMEM need=%d got=%d\n",
185 ctx, need, have + alloc);
Sage Weila8599bd2009-10-06 11:31:12 -0700186
Yehuda Sadeh37151662010-06-17 16:16:12 -0700187 spin_lock(&mdsc->caps_list_lock);
188 mdsc->caps_total_count += alloc;
189 mdsc->caps_reserve_count += alloc;
190 list_splice(&newcaps, &mdsc->caps_list);
Sage Weila8599bd2009-10-06 11:31:12 -0700191
Yehuda Sadeh37151662010-06-17 16:16:12 -0700192 BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count +
193 mdsc->caps_reserve_count +
194 mdsc->caps_avail_count);
195 spin_unlock(&mdsc->caps_list_lock);
Sage Weila8599bd2009-10-06 11:31:12 -0700196
197 ctx->count = need;
198 dout("reserve caps ctx=%p %d = %d used + %d resv + %d avail\n",
Yehuda Sadeh37151662010-06-17 16:16:12 -0700199 ctx, mdsc->caps_total_count, mdsc->caps_use_count,
200 mdsc->caps_reserve_count, mdsc->caps_avail_count);
Sage Weila8599bd2009-10-06 11:31:12 -0700201}
202
Yehuda Sadeh37151662010-06-17 16:16:12 -0700203int ceph_unreserve_caps(struct ceph_mds_client *mdsc,
204 struct ceph_cap_reservation *ctx)
Sage Weila8599bd2009-10-06 11:31:12 -0700205{
206 dout("unreserve caps ctx=%p count=%d\n", ctx, ctx->count);
207 if (ctx->count) {
Yehuda Sadeh37151662010-06-17 16:16:12 -0700208 spin_lock(&mdsc->caps_list_lock);
209 BUG_ON(mdsc->caps_reserve_count < ctx->count);
210 mdsc->caps_reserve_count -= ctx->count;
211 mdsc->caps_avail_count += ctx->count;
Sage Weila8599bd2009-10-06 11:31:12 -0700212 ctx->count = 0;
213 dout("unreserve caps %d = %d used + %d resv + %d avail\n",
Yehuda Sadeh37151662010-06-17 16:16:12 -0700214 mdsc->caps_total_count, mdsc->caps_use_count,
215 mdsc->caps_reserve_count, mdsc->caps_avail_count);
216 BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count +
217 mdsc->caps_reserve_count +
218 mdsc->caps_avail_count);
219 spin_unlock(&mdsc->caps_list_lock);
Sage Weila8599bd2009-10-06 11:31:12 -0700220 }
221 return 0;
222}
223
Yan, Zhengd9df2782014-04-18 09:57:11 +0800224struct ceph_cap *ceph_get_cap(struct ceph_mds_client *mdsc,
225 struct ceph_cap_reservation *ctx)
Sage Weila8599bd2009-10-06 11:31:12 -0700226{
227 struct ceph_cap *cap = NULL;
228
229 /* temporary, until we do something about cap import/export */
Sage Weil443b3762010-06-29 09:28:39 -0700230 if (!ctx) {
231 cap = kmem_cache_alloc(ceph_cap_cachep, GFP_NOFS);
232 if (cap) {
Yan, Zheng4d1d0532012-11-03 10:32:37 +0800233 spin_lock(&mdsc->caps_list_lock);
Yehuda Sadeh37151662010-06-17 16:16:12 -0700234 mdsc->caps_use_count++;
235 mdsc->caps_total_count++;
Yan, Zheng4d1d0532012-11-03 10:32:37 +0800236 spin_unlock(&mdsc->caps_list_lock);
Sage Weil443b3762010-06-29 09:28:39 -0700237 }
238 return cap;
239 }
Sage Weila8599bd2009-10-06 11:31:12 -0700240
Yehuda Sadeh37151662010-06-17 16:16:12 -0700241 spin_lock(&mdsc->caps_list_lock);
Sage Weila8599bd2009-10-06 11:31:12 -0700242 dout("get_cap ctx=%p (%d) %d = %d used + %d resv + %d avail\n",
Yehuda Sadeh37151662010-06-17 16:16:12 -0700243 ctx, ctx->count, mdsc->caps_total_count, mdsc->caps_use_count,
244 mdsc->caps_reserve_count, mdsc->caps_avail_count);
Sage Weila8599bd2009-10-06 11:31:12 -0700245 BUG_ON(!ctx->count);
Yehuda Sadeh37151662010-06-17 16:16:12 -0700246 BUG_ON(ctx->count > mdsc->caps_reserve_count);
247 BUG_ON(list_empty(&mdsc->caps_list));
Sage Weila8599bd2009-10-06 11:31:12 -0700248
249 ctx->count--;
Yehuda Sadeh37151662010-06-17 16:16:12 -0700250 mdsc->caps_reserve_count--;
251 mdsc->caps_use_count++;
Sage Weila8599bd2009-10-06 11:31:12 -0700252
Yehuda Sadeh37151662010-06-17 16:16:12 -0700253 cap = list_first_entry(&mdsc->caps_list, struct ceph_cap, caps_item);
Sage Weila8599bd2009-10-06 11:31:12 -0700254 list_del(&cap->caps_item);
255
Yehuda Sadeh37151662010-06-17 16:16:12 -0700256 BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count +
257 mdsc->caps_reserve_count + mdsc->caps_avail_count);
258 spin_unlock(&mdsc->caps_list_lock);
Sage Weila8599bd2009-10-06 11:31:12 -0700259 return cap;
260}
261
Yehuda Sadeh37151662010-06-17 16:16:12 -0700262void ceph_put_cap(struct ceph_mds_client *mdsc, struct ceph_cap *cap)
Sage Weila8599bd2009-10-06 11:31:12 -0700263{
Yehuda Sadeh37151662010-06-17 16:16:12 -0700264 spin_lock(&mdsc->caps_list_lock);
Sage Weil7c1332b2010-02-16 11:39:45 -0800265 dout("put_cap %p %d = %d used + %d resv + %d avail\n",
Yehuda Sadeh37151662010-06-17 16:16:12 -0700266 cap, mdsc->caps_total_count, mdsc->caps_use_count,
267 mdsc->caps_reserve_count, mdsc->caps_avail_count);
268 mdsc->caps_use_count--;
Sage Weila8599bd2009-10-06 11:31:12 -0700269 /*
Sage Weil85ccce42010-02-17 10:02:43 -0800270 * Keep some preallocated caps around (ceph_min_count), to
271 * avoid lots of free/alloc churn.
Sage Weila8599bd2009-10-06 11:31:12 -0700272 */
Yehuda Sadeh37151662010-06-17 16:16:12 -0700273 if (mdsc->caps_avail_count >= mdsc->caps_reserve_count +
274 mdsc->caps_min_count) {
275 mdsc->caps_total_count--;
Sage Weila8599bd2009-10-06 11:31:12 -0700276 kmem_cache_free(ceph_cap_cachep, cap);
277 } else {
Yehuda Sadeh37151662010-06-17 16:16:12 -0700278 mdsc->caps_avail_count++;
279 list_add(&cap->caps_item, &mdsc->caps_list);
Sage Weila8599bd2009-10-06 11:31:12 -0700280 }
281
Yehuda Sadeh37151662010-06-17 16:16:12 -0700282 BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count +
283 mdsc->caps_reserve_count + mdsc->caps_avail_count);
284 spin_unlock(&mdsc->caps_list_lock);
Sage Weila8599bd2009-10-06 11:31:12 -0700285}
286
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700287void ceph_reservation_status(struct ceph_fs_client *fsc,
Sage Weil85ccce42010-02-17 10:02:43 -0800288 int *total, int *avail, int *used, int *reserved,
289 int *min)
Sage Weila8599bd2009-10-06 11:31:12 -0700290{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700291 struct ceph_mds_client *mdsc = fsc->mdsc;
Yehuda Sadeh37151662010-06-17 16:16:12 -0700292
Sage Weila8599bd2009-10-06 11:31:12 -0700293 if (total)
Yehuda Sadeh37151662010-06-17 16:16:12 -0700294 *total = mdsc->caps_total_count;
Sage Weila8599bd2009-10-06 11:31:12 -0700295 if (avail)
Yehuda Sadeh37151662010-06-17 16:16:12 -0700296 *avail = mdsc->caps_avail_count;
Sage Weila8599bd2009-10-06 11:31:12 -0700297 if (used)
Yehuda Sadeh37151662010-06-17 16:16:12 -0700298 *used = mdsc->caps_use_count;
Sage Weila8599bd2009-10-06 11:31:12 -0700299 if (reserved)
Yehuda Sadeh37151662010-06-17 16:16:12 -0700300 *reserved = mdsc->caps_reserve_count;
Sage Weil85ccce42010-02-17 10:02:43 -0800301 if (min)
Yehuda Sadeh37151662010-06-17 16:16:12 -0700302 *min = mdsc->caps_min_count;
Sage Weila8599bd2009-10-06 11:31:12 -0700303}
304
305/*
306 * Find ceph_cap for given mds, if any.
307 *
Sage Weilbe655592011-11-30 09:47:09 -0800308 * Called with i_ceph_lock held.
Sage Weila8599bd2009-10-06 11:31:12 -0700309 */
310static struct ceph_cap *__get_cap_for_mds(struct ceph_inode_info *ci, int mds)
311{
312 struct ceph_cap *cap;
313 struct rb_node *n = ci->i_caps.rb_node;
314
315 while (n) {
316 cap = rb_entry(n, struct ceph_cap, ci_node);
317 if (mds < cap->mds)
318 n = n->rb_left;
319 else if (mds > cap->mds)
320 n = n->rb_right;
321 else
322 return cap;
323 }
324 return NULL;
325}
326
Greg Farnum2bc50252010-06-30 12:44:34 -0700327struct ceph_cap *ceph_get_cap_for_mds(struct ceph_inode_info *ci, int mds)
328{
329 struct ceph_cap *cap;
330
Sage Weilbe655592011-11-30 09:47:09 -0800331 spin_lock(&ci->i_ceph_lock);
Greg Farnum2bc50252010-06-30 12:44:34 -0700332 cap = __get_cap_for_mds(ci, mds);
Sage Weilbe655592011-11-30 09:47:09 -0800333 spin_unlock(&ci->i_ceph_lock);
Greg Farnum2bc50252010-06-30 12:44:34 -0700334 return cap;
335}
336
Sage Weila8599bd2009-10-06 11:31:12 -0700337/*
Sage Weil33caad32010-05-26 14:31:27 -0700338 * Return id of any MDS with a cap, preferably FILE_WR|BUFFER|EXCL, else -1.
Sage Weila8599bd2009-10-06 11:31:12 -0700339 */
Sage Weilca81f3f2010-06-10 14:21:36 -0700340static int __ceph_get_cap_mds(struct ceph_inode_info *ci)
Sage Weila8599bd2009-10-06 11:31:12 -0700341{
342 struct ceph_cap *cap;
343 int mds = -1;
344 struct rb_node *p;
345
Sage Weil33caad32010-05-26 14:31:27 -0700346 /* prefer mds with WR|BUFFER|EXCL caps */
Sage Weila8599bd2009-10-06 11:31:12 -0700347 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
348 cap = rb_entry(p, struct ceph_cap, ci_node);
349 mds = cap->mds;
Sage Weila8599bd2009-10-06 11:31:12 -0700350 if (cap->issued & (CEPH_CAP_FILE_WR |
351 CEPH_CAP_FILE_BUFFER |
352 CEPH_CAP_FILE_EXCL))
353 break;
354 }
355 return mds;
356}
357
358int ceph_get_cap_mds(struct inode *inode)
359{
Sage Weilbe655592011-11-30 09:47:09 -0800360 struct ceph_inode_info *ci = ceph_inode(inode);
Sage Weila8599bd2009-10-06 11:31:12 -0700361 int mds;
Sage Weilbe655592011-11-30 09:47:09 -0800362 spin_lock(&ci->i_ceph_lock);
Sage Weilca81f3f2010-06-10 14:21:36 -0700363 mds = __ceph_get_cap_mds(ceph_inode(inode));
Sage Weilbe655592011-11-30 09:47:09 -0800364 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -0700365 return mds;
366}
367
368/*
Sage Weilbe655592011-11-30 09:47:09 -0800369 * Called under i_ceph_lock.
Sage Weila8599bd2009-10-06 11:31:12 -0700370 */
371static void __insert_cap_node(struct ceph_inode_info *ci,
372 struct ceph_cap *new)
373{
374 struct rb_node **p = &ci->i_caps.rb_node;
375 struct rb_node *parent = NULL;
376 struct ceph_cap *cap = NULL;
377
378 while (*p) {
379 parent = *p;
380 cap = rb_entry(parent, struct ceph_cap, ci_node);
381 if (new->mds < cap->mds)
382 p = &(*p)->rb_left;
383 else if (new->mds > cap->mds)
384 p = &(*p)->rb_right;
385 else
386 BUG();
387 }
388
389 rb_link_node(&new->ci_node, parent, p);
390 rb_insert_color(&new->ci_node, &ci->i_caps);
391}
392
393/*
394 * (re)set cap hold timeouts, which control the delayed release
395 * of unused caps back to the MDS. Should be called on cap use.
396 */
397static void __cap_set_timeouts(struct ceph_mds_client *mdsc,
398 struct ceph_inode_info *ci)
399{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700400 struct ceph_mount_options *ma = mdsc->fsc->mount_options;
Sage Weila8599bd2009-10-06 11:31:12 -0700401
402 ci->i_hold_caps_min = round_jiffies(jiffies +
403 ma->caps_wanted_delay_min * HZ);
404 ci->i_hold_caps_max = round_jiffies(jiffies +
405 ma->caps_wanted_delay_max * HZ);
406 dout("__cap_set_timeouts %p min %lu max %lu\n", &ci->vfs_inode,
407 ci->i_hold_caps_min - jiffies, ci->i_hold_caps_max - jiffies);
408}
409
410/*
411 * (Re)queue cap at the end of the delayed cap release list.
412 *
413 * If I_FLUSH is set, leave the inode at the front of the list.
414 *
Sage Weilbe655592011-11-30 09:47:09 -0800415 * Caller holds i_ceph_lock
Sage Weila8599bd2009-10-06 11:31:12 -0700416 * -> we take mdsc->cap_delay_lock
417 */
418static void __cap_delay_requeue(struct ceph_mds_client *mdsc,
419 struct ceph_inode_info *ci)
420{
421 __cap_set_timeouts(mdsc, ci);
422 dout("__cap_delay_requeue %p flags %d at %lu\n", &ci->vfs_inode,
423 ci->i_ceph_flags, ci->i_hold_caps_max);
424 if (!mdsc->stopping) {
425 spin_lock(&mdsc->cap_delay_lock);
426 if (!list_empty(&ci->i_cap_delay_list)) {
427 if (ci->i_ceph_flags & CEPH_I_FLUSH)
428 goto no_change;
429 list_del_init(&ci->i_cap_delay_list);
430 }
431 list_add_tail(&ci->i_cap_delay_list, &mdsc->cap_delay_list);
432no_change:
433 spin_unlock(&mdsc->cap_delay_lock);
434 }
435}
436
437/*
438 * Queue an inode for immediate writeback. Mark inode with I_FLUSH,
439 * indicating we should send a cap message to flush dirty metadata
440 * asap, and move to the front of the delayed cap list.
441 */
442static void __cap_delay_requeue_front(struct ceph_mds_client *mdsc,
443 struct ceph_inode_info *ci)
444{
445 dout("__cap_delay_requeue_front %p\n", &ci->vfs_inode);
446 spin_lock(&mdsc->cap_delay_lock);
447 ci->i_ceph_flags |= CEPH_I_FLUSH;
448 if (!list_empty(&ci->i_cap_delay_list))
449 list_del_init(&ci->i_cap_delay_list);
450 list_add(&ci->i_cap_delay_list, &mdsc->cap_delay_list);
451 spin_unlock(&mdsc->cap_delay_lock);
452}
453
454/*
455 * Cancel delayed work on cap.
456 *
Sage Weilbe655592011-11-30 09:47:09 -0800457 * Caller must hold i_ceph_lock.
Sage Weila8599bd2009-10-06 11:31:12 -0700458 */
459static void __cap_delay_cancel(struct ceph_mds_client *mdsc,
460 struct ceph_inode_info *ci)
461{
462 dout("__cap_delay_cancel %p\n", &ci->vfs_inode);
463 if (list_empty(&ci->i_cap_delay_list))
464 return;
465 spin_lock(&mdsc->cap_delay_lock);
466 list_del_init(&ci->i_cap_delay_list);
467 spin_unlock(&mdsc->cap_delay_lock);
468}
469
470/*
471 * Common issue checks for add_cap, handle_cap_grant.
472 */
473static void __check_cap_issue(struct ceph_inode_info *ci, struct ceph_cap *cap,
474 unsigned issued)
475{
476 unsigned had = __ceph_caps_issued(ci, NULL);
477
478 /*
479 * Each time we receive FILE_CACHE anew, we increment
480 * i_rdcache_gen.
481 */
Sage Weil29625072010-05-27 10:40:43 -0700482 if ((issued & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) &&
Milosz Tanski99ccbd22013-08-21 17:29:54 -0400483 (had & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) == 0) {
Sage Weila8599bd2009-10-06 11:31:12 -0700484 ci->i_rdcache_gen++;
Milosz Tanski99ccbd22013-08-21 17:29:54 -0400485 }
Sage Weila8599bd2009-10-06 11:31:12 -0700486
487 /*
Yan, Zheng2f276c52013-03-13 19:44:32 +0800488 * if we are newly issued FILE_SHARED, mark dir not complete; we
Sage Weila8599bd2009-10-06 11:31:12 -0700489 * don't know what happened to this directory while we didn't
490 * have the cap.
491 */
492 if ((issued & CEPH_CAP_FILE_SHARED) &&
493 (had & CEPH_CAP_FILE_SHARED) == 0) {
494 ci->i_shared_gen++;
Yan, Zhenga8673d62013-02-18 16:38:14 +0800495 if (S_ISDIR(ci->vfs_inode.i_mode)) {
496 dout(" marking %p NOT complete\n", &ci->vfs_inode);
Yan, Zheng2f276c52013-03-13 19:44:32 +0800497 __ceph_dir_clear_complete(ci);
Yan, Zhenga8673d62013-02-18 16:38:14 +0800498 }
Sage Weila8599bd2009-10-06 11:31:12 -0700499 }
500}
501
502/*
503 * Add a capability under the given MDS session.
504 *
505 * Caller should hold session snap_rwsem (read) and s_mutex.
506 *
507 * @fmode is the open file mode, if we are opening a file, otherwise
508 * it is < 0. (This is so we can atomically add the cap and add an
509 * open file reference to it.)
510 */
Yan, Zhengd9df2782014-04-18 09:57:11 +0800511void ceph_add_cap(struct inode *inode,
512 struct ceph_mds_session *session, u64 cap_id,
513 int fmode, unsigned issued, unsigned wanted,
514 unsigned seq, unsigned mseq, u64 realmino, int flags,
515 struct ceph_cap **new_cap)
Sage Weila8599bd2009-10-06 11:31:12 -0700516{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700517 struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
Sage Weila8599bd2009-10-06 11:31:12 -0700518 struct ceph_inode_info *ci = ceph_inode(inode);
Sage Weila8599bd2009-10-06 11:31:12 -0700519 struct ceph_cap *cap;
520 int mds = session->s_mds;
521 int actual_wanted;
522
523 dout("add_cap %p mds%d cap %llx %s seq %d\n", inode,
524 session->s_mds, cap_id, ceph_cap_string(issued), seq);
525
526 /*
527 * If we are opening the file, include file mode wanted bits
528 * in wanted.
529 */
530 if (fmode >= 0)
531 wanted |= ceph_caps_for_mode(fmode);
532
Sage Weila8599bd2009-10-06 11:31:12 -0700533 cap = __get_cap_for_mds(ci, mds);
534 if (!cap) {
Yan, Zhengd9df2782014-04-18 09:57:11 +0800535 cap = *new_cap;
536 *new_cap = NULL;
Sage Weila8599bd2009-10-06 11:31:12 -0700537
538 cap->issued = 0;
539 cap->implemented = 0;
540 cap->mds = mds;
541 cap->mds_wanted = 0;
Yan, Zheng964266c2013-02-27 09:26:09 +0800542 cap->mseq = 0;
Sage Weila8599bd2009-10-06 11:31:12 -0700543
544 cap->ci = ci;
545 __insert_cap_node(ci, cap);
546
Sage Weila8599bd2009-10-06 11:31:12 -0700547 /* add to session cap list */
548 cap->session = session;
549 spin_lock(&session->s_cap_lock);
550 list_add_tail(&cap->session_caps, &session->s_caps);
551 session->s_nr_caps++;
552 spin_unlock(&session->s_cap_lock);
Yan, Zheng11df2df2013-11-24 14:44:38 +0800553 } else {
Yan, Zheng11df2df2013-11-24 14:44:38 +0800554 /*
555 * auth mds of the inode changed. we received the cap export
556 * message, but still haven't received the cap import message.
557 * handle_cap_export() updated the new auth MDS' cap.
558 *
559 * "ceph_seq_cmp(seq, cap->seq) <= 0" means we are processing
560 * a message that was send before the cap import message. So
561 * don't remove caps.
562 */
563 if (ceph_seq_cmp(seq, cap->seq) <= 0) {
564 WARN_ON(cap != ci->i_auth_cap);
565 WARN_ON(cap->cap_id != cap_id);
566 seq = cap->seq;
567 mseq = cap->mseq;
568 issued |= cap->issued;
569 flags |= CEPH_CAP_FLAG_AUTH;
570 }
571 }
Sage Weila8599bd2009-10-06 11:31:12 -0700572
573 if (!ci->i_snap_realm) {
574 /*
575 * add this inode to the appropriate snap realm
576 */
577 struct ceph_snap_realm *realm = ceph_lookup_snap_realm(mdsc,
578 realmino);
579 if (realm) {
Sage Weila8599bd2009-10-06 11:31:12 -0700580 spin_lock(&realm->inodes_with_caps_lock);
581 ci->i_snap_realm = realm;
582 list_add(&ci->i_snap_realm_item,
583 &realm->inodes_with_caps);
584 spin_unlock(&realm->inodes_with_caps_lock);
585 } else {
586 pr_err("ceph_add_cap: couldn't find snap realm %llx\n",
587 realmino);
Sage Weilb8cd07e2010-07-16 12:00:02 -0700588 WARN_ON(!realm);
Sage Weila8599bd2009-10-06 11:31:12 -0700589 }
590 }
591
592 __check_cap_issue(ci, cap, issued);
593
594 /*
595 * If we are issued caps we don't want, or the mds' wanted
596 * value appears to be off, queue a check so we'll release
597 * later and/or update the mds wanted value.
598 */
599 actual_wanted = __ceph_caps_wanted(ci);
600 if ((wanted & ~actual_wanted) ||
601 (issued & ~actual_wanted & CEPH_CAP_ANY_WR)) {
602 dout(" issued %s, mds wanted %s, actual %s, queueing\n",
603 ceph_cap_string(issued), ceph_cap_string(wanted),
604 ceph_cap_string(actual_wanted));
605 __cap_delay_requeue(mdsc, ci);
606 }
607
Yan, Zhengb8c2f3a2013-05-31 16:37:11 +0800608 if (flags & CEPH_CAP_FLAG_AUTH) {
609 if (ci->i_auth_cap == NULL ||
Yan, Zhengd9ffc4f2014-03-18 10:15:29 +0800610 ceph_seq_cmp(ci->i_auth_cap->mseq, mseq) < 0) {
Yan, Zhengb8c2f3a2013-05-31 16:37:11 +0800611 ci->i_auth_cap = cap;
Yan, Zhengd9ffc4f2014-03-18 10:15:29 +0800612 cap->mds_wanted = wanted;
613 }
Yan, Zheng11df2df2013-11-24 14:44:38 +0800614 } else {
615 WARN_ON(ci->i_auth_cap == cap);
Yan, Zheng8a92a112013-01-04 14:28:07 +0800616 }
Sage Weila8599bd2009-10-06 11:31:12 -0700617
618 dout("add_cap inode %p (%llx.%llx) cap %p %s now %s seq %d mds%d\n",
619 inode, ceph_vinop(inode), cap, ceph_cap_string(issued),
620 ceph_cap_string(issued|cap->issued), seq, mds);
621 cap->cap_id = cap_id;
622 cap->issued = issued;
623 cap->implemented |= issued;
Yan, Zhengd1b87802013-11-13 14:47:19 +0800624 if (ceph_seq_cmp(mseq, cap->mseq) > 0)
Yan, Zheng964266c2013-02-27 09:26:09 +0800625 cap->mds_wanted = wanted;
626 else
627 cap->mds_wanted |= wanted;
Sage Weila8599bd2009-10-06 11:31:12 -0700628 cap->seq = seq;
629 cap->issue_seq = seq;
630 cap->mseq = mseq;
Sage Weil685f9a5d2009-11-09 12:05:48 -0800631 cap->cap_gen = session->s_cap_gen;
Sage Weila8599bd2009-10-06 11:31:12 -0700632
633 if (fmode >= 0)
634 __ceph_get_fmode(ci, fmode);
Sage Weila8599bd2009-10-06 11:31:12 -0700635}
636
637/*
638 * Return true if cap has not timed out and belongs to the current
639 * generation of the MDS session (i.e. has not gone 'stale' due to
640 * us losing touch with the mds).
641 */
642static int __cap_is_valid(struct ceph_cap *cap)
643{
644 unsigned long ttl;
Sage Weilcdac8302009-11-10 16:02:23 -0800645 u32 gen;
Sage Weila8599bd2009-10-06 11:31:12 -0700646
Alex Elderd8fb02a2012-01-12 17:48:10 -0800647 spin_lock(&cap->session->s_gen_ttl_lock);
Sage Weila8599bd2009-10-06 11:31:12 -0700648 gen = cap->session->s_cap_gen;
649 ttl = cap->session->s_cap_ttl;
Alex Elderd8fb02a2012-01-12 17:48:10 -0800650 spin_unlock(&cap->session->s_gen_ttl_lock);
Sage Weila8599bd2009-10-06 11:31:12 -0700651
Sage Weil685f9a5d2009-11-09 12:05:48 -0800652 if (cap->cap_gen < gen || time_after_eq(jiffies, ttl)) {
Sage Weila8599bd2009-10-06 11:31:12 -0700653 dout("__cap_is_valid %p cap %p issued %s "
654 "but STALE (gen %u vs %u)\n", &cap->ci->vfs_inode,
Sage Weil685f9a5d2009-11-09 12:05:48 -0800655 cap, ceph_cap_string(cap->issued), cap->cap_gen, gen);
Sage Weila8599bd2009-10-06 11:31:12 -0700656 return 0;
657 }
658
659 return 1;
660}
661
662/*
663 * Return set of valid cap bits issued to us. Note that caps time
664 * out, and may be invalidated in bulk if the client session times out
665 * and session->s_cap_gen is bumped.
666 */
667int __ceph_caps_issued(struct ceph_inode_info *ci, int *implemented)
668{
Yan, Zhengd9df2782014-04-18 09:57:11 +0800669 int have = ci->i_snap_caps;
Sage Weila8599bd2009-10-06 11:31:12 -0700670 struct ceph_cap *cap;
671 struct rb_node *p;
672
673 if (implemented)
674 *implemented = 0;
675 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
676 cap = rb_entry(p, struct ceph_cap, ci_node);
677 if (!__cap_is_valid(cap))
678 continue;
679 dout("__ceph_caps_issued %p cap %p issued %s\n",
680 &ci->vfs_inode, cap, ceph_cap_string(cap->issued));
681 have |= cap->issued;
682 if (implemented)
683 *implemented |= cap->implemented;
684 }
Yan, Zhengb1530f52013-07-02 12:40:20 +0800685 /*
686 * exclude caps issued by non-auth MDS, but are been revoking
687 * by the auth MDS. The non-auth MDS should be revoking/exporting
688 * these caps, but the message is delayed.
689 */
690 if (ci->i_auth_cap) {
691 cap = ci->i_auth_cap;
692 have &= ~cap->implemented | cap->issued;
693 }
Sage Weila8599bd2009-10-06 11:31:12 -0700694 return have;
695}
696
697/*
698 * Get cap bits issued by caps other than @ocap
699 */
700int __ceph_caps_issued_other(struct ceph_inode_info *ci, struct ceph_cap *ocap)
701{
702 int have = ci->i_snap_caps;
703 struct ceph_cap *cap;
704 struct rb_node *p;
705
706 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
707 cap = rb_entry(p, struct ceph_cap, ci_node);
708 if (cap == ocap)
709 continue;
710 if (!__cap_is_valid(cap))
711 continue;
712 have |= cap->issued;
713 }
714 return have;
715}
716
717/*
718 * Move a cap to the end of the LRU (oldest caps at list head, newest
719 * at list tail).
720 */
721static void __touch_cap(struct ceph_cap *cap)
722{
723 struct ceph_mds_session *s = cap->session;
724
Sage Weila8599bd2009-10-06 11:31:12 -0700725 spin_lock(&s->s_cap_lock);
Sage Weil7c1332b2010-02-16 11:39:45 -0800726 if (s->s_cap_iterator == NULL) {
Sage Weil5dacf092009-12-21 20:40:34 -0800727 dout("__touch_cap %p cap %p mds%d\n", &cap->ci->vfs_inode, cap,
728 s->s_mds);
729 list_move_tail(&cap->session_caps, &s->s_caps);
730 } else {
731 dout("__touch_cap %p cap %p mds%d NOP, iterating over caps\n",
732 &cap->ci->vfs_inode, cap, s->s_mds);
733 }
Sage Weila8599bd2009-10-06 11:31:12 -0700734 spin_unlock(&s->s_cap_lock);
735}
736
737/*
738 * Check if we hold the given mask. If so, move the cap(s) to the
739 * front of their respective LRUs. (This is the preferred way for
740 * callers to check for caps they want.)
741 */
742int __ceph_caps_issued_mask(struct ceph_inode_info *ci, int mask, int touch)
743{
744 struct ceph_cap *cap;
745 struct rb_node *p;
746 int have = ci->i_snap_caps;
747
748 if ((have & mask) == mask) {
749 dout("__ceph_caps_issued_mask %p snap issued %s"
750 " (mask %s)\n", &ci->vfs_inode,
751 ceph_cap_string(have),
752 ceph_cap_string(mask));
753 return 1;
754 }
755
756 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
757 cap = rb_entry(p, struct ceph_cap, ci_node);
758 if (!__cap_is_valid(cap))
759 continue;
760 if ((cap->issued & mask) == mask) {
761 dout("__ceph_caps_issued_mask %p cap %p issued %s"
762 " (mask %s)\n", &ci->vfs_inode, cap,
763 ceph_cap_string(cap->issued),
764 ceph_cap_string(mask));
765 if (touch)
766 __touch_cap(cap);
767 return 1;
768 }
769
770 /* does a combination of caps satisfy mask? */
771 have |= cap->issued;
772 if ((have & mask) == mask) {
773 dout("__ceph_caps_issued_mask %p combo issued %s"
774 " (mask %s)\n", &ci->vfs_inode,
775 ceph_cap_string(cap->issued),
776 ceph_cap_string(mask));
777 if (touch) {
778 struct rb_node *q;
779
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300780 /* touch this + preceding caps */
Sage Weila8599bd2009-10-06 11:31:12 -0700781 __touch_cap(cap);
782 for (q = rb_first(&ci->i_caps); q != p;
783 q = rb_next(q)) {
784 cap = rb_entry(q, struct ceph_cap,
785 ci_node);
786 if (!__cap_is_valid(cap))
787 continue;
788 __touch_cap(cap);
789 }
790 }
791 return 1;
792 }
793 }
794
795 return 0;
796}
797
798/*
799 * Return true if mask caps are currently being revoked by an MDS.
800 */
Yan, Zheng6ee6b9532013-07-02 12:40:21 +0800801int __ceph_caps_revoking_other(struct ceph_inode_info *ci,
802 struct ceph_cap *ocap, int mask)
803{
804 struct ceph_cap *cap;
805 struct rb_node *p;
806
807 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
808 cap = rb_entry(p, struct ceph_cap, ci_node);
Yan, Zheng9563f882013-11-22 13:50:45 +0800809 if (cap != ocap &&
Yan, Zheng6ee6b9532013-07-02 12:40:21 +0800810 (cap->implemented & ~cap->issued & mask))
811 return 1;
812 }
813 return 0;
814}
815
Sage Weila8599bd2009-10-06 11:31:12 -0700816int ceph_caps_revoking(struct ceph_inode_info *ci, int mask)
817{
818 struct inode *inode = &ci->vfs_inode;
Yan, Zheng6ee6b9532013-07-02 12:40:21 +0800819 int ret;
Sage Weila8599bd2009-10-06 11:31:12 -0700820
Sage Weilbe655592011-11-30 09:47:09 -0800821 spin_lock(&ci->i_ceph_lock);
Yan, Zheng6ee6b9532013-07-02 12:40:21 +0800822 ret = __ceph_caps_revoking_other(ci, NULL, mask);
Sage Weilbe655592011-11-30 09:47:09 -0800823 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -0700824 dout("ceph_caps_revoking %p %s = %d\n", inode,
825 ceph_cap_string(mask), ret);
826 return ret;
827}
828
829int __ceph_caps_used(struct ceph_inode_info *ci)
830{
831 int used = 0;
832 if (ci->i_pin_ref)
833 used |= CEPH_CAP_PIN;
834 if (ci->i_rd_ref)
835 used |= CEPH_CAP_FILE_RD;
Sage Weila43fb732010-09-17 09:54:08 -0700836 if (ci->i_rdcache_ref || ci->vfs_inode.i_data.nrpages)
Sage Weila8599bd2009-10-06 11:31:12 -0700837 used |= CEPH_CAP_FILE_CACHE;
838 if (ci->i_wr_ref)
839 used |= CEPH_CAP_FILE_WR;
Henry C Changd3d07202011-05-11 10:29:54 +0000840 if (ci->i_wb_ref || ci->i_wrbuffer_ref)
Sage Weila8599bd2009-10-06 11:31:12 -0700841 used |= CEPH_CAP_FILE_BUFFER;
842 return used;
843}
844
845/*
846 * wanted, by virtue of open file modes
847 */
848int __ceph_caps_file_wanted(struct ceph_inode_info *ci)
849{
850 int want = 0;
851 int mode;
Sage Weil33caad32010-05-26 14:31:27 -0700852 for (mode = 0; mode < CEPH_FILE_MODE_NUM; mode++)
Sage Weila8599bd2009-10-06 11:31:12 -0700853 if (ci->i_nr_by_mode[mode])
854 want |= ceph_caps_for_mode(mode);
855 return want;
856}
857
858/*
859 * Return caps we have registered with the MDS(s) as 'wanted'.
860 */
861int __ceph_caps_mds_wanted(struct ceph_inode_info *ci)
862{
863 struct ceph_cap *cap;
864 struct rb_node *p;
865 int mds_wanted = 0;
866
867 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
868 cap = rb_entry(p, struct ceph_cap, ci_node);
869 if (!__cap_is_valid(cap))
870 continue;
Yan, Zhenga2550602014-03-08 09:51:45 +0800871 if (cap == ci->i_auth_cap)
872 mds_wanted |= cap->mds_wanted;
873 else
874 mds_wanted |= (cap->mds_wanted & ~CEPH_CAP_ANY_FILE_WR);
Sage Weila8599bd2009-10-06 11:31:12 -0700875 }
876 return mds_wanted;
877}
878
879/*
Sage Weilbe655592011-11-30 09:47:09 -0800880 * called under i_ceph_lock
Sage Weila8599bd2009-10-06 11:31:12 -0700881 */
882static int __ceph_is_any_caps(struct ceph_inode_info *ci)
883{
Yan, Zhengd9df2782014-04-18 09:57:11 +0800884 return !RB_EMPTY_ROOT(&ci->i_caps);
Sage Weila8599bd2009-10-06 11:31:12 -0700885}
886
Yan, Zheng9215aee2013-11-30 12:47:41 +0800887int ceph_is_any_caps(struct inode *inode)
888{
889 struct ceph_inode_info *ci = ceph_inode(inode);
890 int ret;
891
892 spin_lock(&ci->i_ceph_lock);
893 ret = __ceph_is_any_caps(ci);
894 spin_unlock(&ci->i_ceph_lock);
895
896 return ret;
897}
898
Yan, Zhengdb40cc12015-03-23 20:12:20 +0800899static void drop_inode_snap_realm(struct ceph_inode_info *ci)
900{
901 struct ceph_snap_realm *realm = ci->i_snap_realm;
902 spin_lock(&realm->inodes_with_caps_lock);
903 list_del_init(&ci->i_snap_realm_item);
904 ci->i_snap_realm_counter++;
905 ci->i_snap_realm = NULL;
906 spin_unlock(&realm->inodes_with_caps_lock);
907 ceph_put_snap_realm(ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc,
908 realm);
909}
910
Sage Weila8599bd2009-10-06 11:31:12 -0700911/*
Sage Weilf818a732010-05-11 20:56:31 -0700912 * Remove a cap. Take steps to deal with a racing iterate_session_caps.
913 *
Sage Weilbe655592011-11-30 09:47:09 -0800914 * caller should hold i_ceph_lock.
Sage Weila6369742010-02-22 13:59:00 -0800915 * caller will not hold session s_mutex if called from destroy_inode.
Sage Weila8599bd2009-10-06 11:31:12 -0700916 */
Yan, Zhenga096b092013-09-22 10:15:58 +0800917void __ceph_remove_cap(struct ceph_cap *cap, bool queue_release)
Sage Weila8599bd2009-10-06 11:31:12 -0700918{
919 struct ceph_mds_session *session = cap->session;
920 struct ceph_inode_info *ci = cap->ci;
Cheng Renquan640ef792010-03-26 17:40:33 +0800921 struct ceph_mds_client *mdsc =
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700922 ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc;
Sage Weilf818a732010-05-11 20:56:31 -0700923 int removed = 0;
Sage Weila8599bd2009-10-06 11:31:12 -0700924
925 dout("__ceph_remove_cap %p from %p\n", cap, &ci->vfs_inode);
926
Sage Weil7c1332b2010-02-16 11:39:45 -0800927 /* remove from session list */
928 spin_lock(&session->s_cap_lock);
929 if (session->s_cap_iterator == cap) {
930 /* not yet, we are iterating over this very cap */
931 dout("__ceph_remove_cap delaying %p removal from session %p\n",
932 cap, cap->session);
933 } else {
934 list_del_init(&cap->session_caps);
935 session->s_nr_caps--;
936 cap->session = NULL;
Sage Weilf818a732010-05-11 20:56:31 -0700937 removed = 1;
Sage Weil7c1332b2010-02-16 11:39:45 -0800938 }
Sage Weilf818a732010-05-11 20:56:31 -0700939 /* protect backpointer with s_cap_lock: see iterate_session_caps */
940 cap->ci = NULL;
Yan, Zheng745a8e32015-05-14 17:22:42 +0800941
942 /*
943 * s_cap_reconnect is protected by s_cap_lock. no one changes
944 * s_cap_gen while session is in the reconnect state.
945 */
946 if (queue_release &&
947 (!session->s_cap_reconnect || cap->cap_gen == session->s_cap_gen)) {
948 cap->queue_release = 1;
949 if (removed) {
950 list_add_tail(&cap->session_caps,
951 &session->s_cap_releases);
952 session->s_num_cap_releases++;
953 removed = 0;
954 }
955 } else {
956 cap->queue_release = 0;
957 }
958 cap->cap_ino = ci->i_vino.ino;
959
Sage Weil7c1332b2010-02-16 11:39:45 -0800960 spin_unlock(&session->s_cap_lock);
961
Sage Weilf818a732010-05-11 20:56:31 -0700962 /* remove from inode list */
963 rb_erase(&cap->ci_node, &ci->i_caps);
964 if (ci->i_auth_cap == cap)
965 ci->i_auth_cap = NULL;
966
967 if (removed)
Yehuda Sadeh37151662010-06-17 16:16:12 -0700968 ceph_put_cap(mdsc, cap);
Sage Weila8599bd2009-10-06 11:31:12 -0700969
Yan, Zhengdb40cc12015-03-23 20:12:20 +0800970 /* when reconnect denied, we remove session caps forcibly,
971 * i_wr_ref can be non-zero. If there are ongoing write,
972 * keep i_snap_realm.
973 */
974 if (!__ceph_is_any_caps(ci) && ci->i_wr_ref == 0 && ci->i_snap_realm)
975 drop_inode_snap_realm(ci);
976
Sage Weila8599bd2009-10-06 11:31:12 -0700977 if (!__ceph_is_any_real_caps(ci))
978 __cap_delay_cancel(mdsc, ci);
979}
980
981/*
982 * Build and send a cap message to the given MDS.
983 *
984 * Caller should be holding s_mutex.
985 */
986static int send_cap_msg(struct ceph_mds_session *session,
987 u64 ino, u64 cid, int op,
988 int caps, int wanted, int dirty,
989 u32 seq, u64 flush_tid, u32 issue_seq, u32 mseq,
990 u64 size, u64 max_size,
991 struct timespec *mtime, struct timespec *atime,
992 u64 time_warp_seq,
Eric W. Biederman05cb11c2013-01-31 02:56:19 -0800993 kuid_t uid, kgid_t gid, umode_t mode,
Sage Weila8599bd2009-10-06 11:31:12 -0700994 u64 xattr_version,
995 struct ceph_buffer *xattrs_buf,
Yan, Zhenge20d2582014-11-14 22:39:13 +0800996 u64 follows, bool inline_data)
Sage Weila8599bd2009-10-06 11:31:12 -0700997{
998 struct ceph_mds_caps *fc;
999 struct ceph_msg *msg;
Yan, Zhenge20d2582014-11-14 22:39:13 +08001000 void *p;
1001 size_t extra_len;
Sage Weila8599bd2009-10-06 11:31:12 -07001002
1003 dout("send_cap_msg %s %llx %llx caps %s wanted %s dirty %s"
1004 " seq %u/%u mseq %u follows %lld size %llu/%llu"
1005 " xattr_ver %llu xattr_len %d\n", ceph_cap_op_name(op),
1006 cid, ino, ceph_cap_string(caps), ceph_cap_string(wanted),
1007 ceph_cap_string(dirty),
1008 seq, issue_seq, mseq, follows, size, max_size,
1009 xattr_version, xattrs_buf ? (int)xattrs_buf->vec.iov_len : 0);
1010
Yan, Zhenge20d2582014-11-14 22:39:13 +08001011 /* flock buffer size + inline version + inline data size */
1012 extra_len = 4 + 8 + 4;
1013 msg = ceph_msg_new(CEPH_MSG_CLIENT_CAPS, sizeof(*fc) + extra_len,
1014 GFP_NOFS, false);
Sage Weila79832f2010-04-01 16:06:19 -07001015 if (!msg)
1016 return -ENOMEM;
Sage Weila8599bd2009-10-06 11:31:12 -07001017
Sage Weil6df058c2009-12-22 11:24:33 -08001018 msg->hdr.tid = cpu_to_le64(flush_tid);
Sage Weila8599bd2009-10-06 11:31:12 -07001019
Sage Weil6df058c2009-12-22 11:24:33 -08001020 fc = msg->front.iov_base;
Sage Weila8599bd2009-10-06 11:31:12 -07001021 memset(fc, 0, sizeof(*fc));
1022
1023 fc->cap_id = cpu_to_le64(cid);
1024 fc->op = cpu_to_le32(op);
1025 fc->seq = cpu_to_le32(seq);
Sage Weila8599bd2009-10-06 11:31:12 -07001026 fc->issue_seq = cpu_to_le32(issue_seq);
1027 fc->migrate_seq = cpu_to_le32(mseq);
1028 fc->caps = cpu_to_le32(caps);
1029 fc->wanted = cpu_to_le32(wanted);
1030 fc->dirty = cpu_to_le32(dirty);
1031 fc->ino = cpu_to_le64(ino);
1032 fc->snap_follows = cpu_to_le64(follows);
1033
1034 fc->size = cpu_to_le64(size);
1035 fc->max_size = cpu_to_le64(max_size);
1036 if (mtime)
1037 ceph_encode_timespec(&fc->mtime, mtime);
1038 if (atime)
1039 ceph_encode_timespec(&fc->atime, atime);
1040 fc->time_warp_seq = cpu_to_le32(time_warp_seq);
1041
Eric W. Biederman05cb11c2013-01-31 02:56:19 -08001042 fc->uid = cpu_to_le32(from_kuid(&init_user_ns, uid));
1043 fc->gid = cpu_to_le32(from_kgid(&init_user_ns, gid));
Sage Weila8599bd2009-10-06 11:31:12 -07001044 fc->mode = cpu_to_le32(mode);
1045
Yan, Zhenge20d2582014-11-14 22:39:13 +08001046 p = fc + 1;
1047 /* flock buffer size */
1048 ceph_encode_32(&p, 0);
1049 /* inline version */
1050 ceph_encode_64(&p, inline_data ? 0 : CEPH_INLINE_NONE);
1051 /* inline data size */
1052 ceph_encode_32(&p, 0);
1053
Sage Weila8599bd2009-10-06 11:31:12 -07001054 fc->xattr_version = cpu_to_le64(xattr_version);
1055 if (xattrs_buf) {
1056 msg->middle = ceph_buffer_get(xattrs_buf);
1057 fc->xattr_len = cpu_to_le32(xattrs_buf->vec.iov_len);
1058 msg->hdr.middle_len = cpu_to_le32(xattrs_buf->vec.iov_len);
1059 }
1060
1061 ceph_con_send(&session->s_con, msg);
1062 return 0;
1063}
1064
1065/*
Sage Weila6369742010-02-22 13:59:00 -08001066 * Queue cap releases when an inode is dropped from our cache. Since
Sage Weilbe655592011-11-30 09:47:09 -08001067 * inode is about to be destroyed, there is no need for i_ceph_lock.
Sage Weila8599bd2009-10-06 11:31:12 -07001068 */
1069void ceph_queue_caps_release(struct inode *inode)
1070{
1071 struct ceph_inode_info *ci = ceph_inode(inode);
1072 struct rb_node *p;
1073
Sage Weila8599bd2009-10-06 11:31:12 -07001074 p = rb_first(&ci->i_caps);
1075 while (p) {
1076 struct ceph_cap *cap = rb_entry(p, struct ceph_cap, ci_node);
Sage Weila8599bd2009-10-06 11:31:12 -07001077 p = rb_next(p);
Yan, Zhenga096b092013-09-22 10:15:58 +08001078 __ceph_remove_cap(cap, true);
Sage Weila8599bd2009-10-06 11:31:12 -07001079 }
Sage Weila8599bd2009-10-06 11:31:12 -07001080}
1081
1082/*
1083 * Send a cap msg on the given inode. Update our caps state, then
Sage Weilbe655592011-11-30 09:47:09 -08001084 * drop i_ceph_lock and send the message.
Sage Weila8599bd2009-10-06 11:31:12 -07001085 *
1086 * Make note of max_size reported/requested from mds, revoked caps
1087 * that have now been implemented.
1088 *
1089 * Make half-hearted attempt ot to invalidate page cache if we are
1090 * dropping RDCACHE. Note that this will leave behind locked pages
1091 * that we'll then need to deal with elsewhere.
1092 *
1093 * Return non-zero if delayed release, or we experienced an error
1094 * such that the caller should requeue + retry later.
1095 *
Sage Weilbe655592011-11-30 09:47:09 -08001096 * called with i_ceph_lock, then drops it.
Sage Weila8599bd2009-10-06 11:31:12 -07001097 * caller should hold snap_rwsem (read), s_mutex.
1098 */
1099static int __send_cap(struct ceph_mds_client *mdsc, struct ceph_cap *cap,
Yan, Zheng89b52fe2015-05-27 09:59:48 +08001100 int op, int used, int want, int retain, int flushing)
Sage Weilbe655592011-11-30 09:47:09 -08001101 __releases(cap->ci->i_ceph_lock)
Sage Weila8599bd2009-10-06 11:31:12 -07001102{
1103 struct ceph_inode_info *ci = cap->ci;
1104 struct inode *inode = &ci->vfs_inode;
1105 u64 cap_id = cap->cap_id;
Sage Weil68c28322010-02-09 13:41:47 -08001106 int held, revoking, dropping, keep;
Sage Weila8599bd2009-10-06 11:31:12 -07001107 u64 seq, issue_seq, mseq, time_warp_seq, follows;
1108 u64 size, max_size;
1109 struct timespec mtime, atime;
1110 int wake = 0;
Al Viro5706b272011-07-26 04:52:22 -04001111 umode_t mode;
Eric W. Biederman05cb11c2013-01-31 02:56:19 -08001112 kuid_t uid;
1113 kgid_t gid;
Sage Weila8599bd2009-10-06 11:31:12 -07001114 struct ceph_mds_session *session;
1115 u64 xattr_version = 0;
Sage Weil082afec2010-08-22 15:16:41 -07001116 struct ceph_buffer *xattr_blob = NULL;
Sage Weila8599bd2009-10-06 11:31:12 -07001117 int delayed = 0;
1118 u64 flush_tid = 0;
1119 int i;
1120 int ret;
Yan, Zhenge20d2582014-11-14 22:39:13 +08001121 bool inline_data;
Sage Weila8599bd2009-10-06 11:31:12 -07001122
Sage Weil68c28322010-02-09 13:41:47 -08001123 held = cap->issued | cap->implemented;
1124 revoking = cap->implemented & ~cap->issued;
1125 retain &= ~revoking;
1126 dropping = cap->issued & ~retain;
1127
Sage Weila8599bd2009-10-06 11:31:12 -07001128 dout("__send_cap %p cap %p session %p %s -> %s (revoking %s)\n",
1129 inode, cap, cap->session,
1130 ceph_cap_string(held), ceph_cap_string(held & retain),
1131 ceph_cap_string(revoking));
1132 BUG_ON((retain & CEPH_CAP_PIN) == 0);
1133
1134 session = cap->session;
1135
1136 /* don't release wanted unless we've waited a bit. */
1137 if ((ci->i_ceph_flags & CEPH_I_NODELAY) == 0 &&
1138 time_before(jiffies, ci->i_hold_caps_min)) {
1139 dout(" delaying issued %s -> %s, wanted %s -> %s on send\n",
1140 ceph_cap_string(cap->issued),
1141 ceph_cap_string(cap->issued & retain),
1142 ceph_cap_string(cap->mds_wanted),
1143 ceph_cap_string(want));
1144 want |= cap->mds_wanted;
1145 retain |= cap->issued;
1146 delayed = 1;
1147 }
1148 ci->i_ceph_flags &= ~(CEPH_I_NODELAY | CEPH_I_FLUSH);
1149
1150 cap->issued &= retain; /* drop bits we don't want */
1151 if (cap->implemented & ~cap->issued) {
1152 /*
1153 * Wake up any waiters on wanted -> needed transition.
1154 * This is due to the weird transition from buffered
1155 * to sync IO... we need to flush dirty pages _before_
1156 * allowing sync writes to avoid reordering.
1157 */
1158 wake = 1;
1159 }
1160 cap->implemented &= cap->issued | used;
1161 cap->mds_wanted = want;
1162
1163 if (flushing) {
1164 /*
1165 * assign a tid for flush operations so we can avoid
1166 * flush1 -> dirty1 -> flush2 -> flushack1 -> mark
1167 * clean type races. track latest tid for every bit
1168 * so we can handle flush AxFw, flush Fw, and have the
1169 * first ack clean Ax.
1170 */
1171 flush_tid = ++ci->i_cap_flush_last_tid;
Sage Weila8599bd2009-10-06 11:31:12 -07001172 dout(" cap_flush_tid %d\n", (int)flush_tid);
1173 for (i = 0; i < CEPH_CAP_BITS; i++)
1174 if (flushing & (1 << i))
1175 ci->i_cap_flush_tid[i] = flush_tid;
Sage Weil7d8cb262010-08-24 08:44:16 -07001176
1177 follows = ci->i_head_snapc->seq;
1178 } else {
1179 follows = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07001180 }
1181
1182 keep = cap->implemented;
1183 seq = cap->seq;
1184 issue_seq = cap->issue_seq;
1185 mseq = cap->mseq;
1186 size = inode->i_size;
1187 ci->i_reported_size = size;
1188 max_size = ci->i_wanted_max_size;
1189 ci->i_requested_max_size = max_size;
1190 mtime = inode->i_mtime;
1191 atime = inode->i_atime;
1192 time_warp_seq = ci->i_time_warp_seq;
Sage Weila8599bd2009-10-06 11:31:12 -07001193 uid = inode->i_uid;
1194 gid = inode->i_gid;
1195 mode = inode->i_mode;
1196
Sage Weil082afec2010-08-22 15:16:41 -07001197 if (flushing & CEPH_CAP_XATTR_EXCL) {
Sage Weila8599bd2009-10-06 11:31:12 -07001198 __ceph_build_xattrs_blob(ci);
Sage Weil082afec2010-08-22 15:16:41 -07001199 xattr_blob = ci->i_xattrs.blob;
1200 xattr_version = ci->i_xattrs.version;
Sage Weila8599bd2009-10-06 11:31:12 -07001201 }
1202
Yan, Zhenge20d2582014-11-14 22:39:13 +08001203 inline_data = ci->i_inline_version != CEPH_INLINE_NONE;
1204
Sage Weilbe655592011-11-30 09:47:09 -08001205 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001206
Sage Weila8599bd2009-10-06 11:31:12 -07001207 ret = send_cap_msg(session, ceph_vino(inode).ino, cap_id,
1208 op, keep, want, flushing, seq, flush_tid, issue_seq, mseq,
1209 size, max_size, &mtime, &atime, time_warp_seq,
Sage Weil082afec2010-08-22 15:16:41 -07001210 uid, gid, mode, xattr_version, xattr_blob,
Yan, Zhenge20d2582014-11-14 22:39:13 +08001211 follows, inline_data);
Sage Weila8599bd2009-10-06 11:31:12 -07001212 if (ret < 0) {
1213 dout("error sending cap msg, must requeue %p\n", inode);
1214 delayed = 1;
1215 }
1216
1217 if (wake)
Yehuda Sadeh03066f22010-07-27 13:11:08 -07001218 wake_up_all(&ci->i_cap_wq);
Sage Weila8599bd2009-10-06 11:31:12 -07001219
1220 return delayed;
1221}
1222
1223/*
1224 * When a snapshot is taken, clients accumulate dirty metadata on
1225 * inodes with capabilities in ceph_cap_snaps to describe the file
1226 * state at the time the snapshot was taken. This must be flushed
1227 * asynchronously back to the MDS once sync writes complete and dirty
1228 * data is written out.
1229 *
Yan, Zhengaffbc192015-05-05 21:22:13 +08001230 * Unless @kick is true, skip cap_snaps that were already sent to
Sage Weile8351242010-09-17 08:03:08 -07001231 * the MDS (i.e., during this session).
1232 *
Sage Weilbe655592011-11-30 09:47:09 -08001233 * Called under i_ceph_lock. Takes s_mutex as needed.
Sage Weila8599bd2009-10-06 11:31:12 -07001234 */
1235void __ceph_flush_snaps(struct ceph_inode_info *ci,
Sage Weile8351242010-09-17 08:03:08 -07001236 struct ceph_mds_session **psession,
Yan, Zhengaffbc192015-05-05 21:22:13 +08001237 int kick)
Sage Weilbe655592011-11-30 09:47:09 -08001238 __releases(ci->i_ceph_lock)
1239 __acquires(ci->i_ceph_lock)
Sage Weila8599bd2009-10-06 11:31:12 -07001240{
1241 struct inode *inode = &ci->vfs_inode;
1242 int mds;
1243 struct ceph_cap_snap *capsnap;
1244 u32 mseq;
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001245 struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
Sage Weila8599bd2009-10-06 11:31:12 -07001246 struct ceph_mds_session *session = NULL; /* if session != NULL, we hold
1247 session->s_mutex */
1248 u64 next_follows = 0; /* keep track of how far we've gotten through the
1249 i_cap_snaps list, and skip these entries next time
1250 around to avoid an infinite loop */
1251
1252 if (psession)
1253 session = *psession;
1254
1255 dout("__flush_snaps %p\n", inode);
1256retry:
1257 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
1258 /* avoid an infiniute loop after retry */
1259 if (capsnap->follows < next_follows)
1260 continue;
1261 /*
1262 * we need to wait for sync writes to complete and for dirty
1263 * pages to be written out.
1264 */
1265 if (capsnap->dirty_pages || capsnap->writing)
Sage Weilcfc0bf62010-09-14 15:50:59 -07001266 break;
Sage Weila8599bd2009-10-06 11:31:12 -07001267
Yan, Zheng86056092015-05-01 16:57:16 +08001268 /* should be removed by ceph_try_drop_cap_snap() */
1269 BUG_ON(!capsnap->need_flush);
Sage Weil819ccbf2010-04-01 09:33:46 -07001270
Sage Weila8599bd2009-10-06 11:31:12 -07001271 /* pick mds, take s_mutex */
Sage Weilca81f3f2010-06-10 14:21:36 -07001272 if (ci->i_auth_cap == NULL) {
1273 dout("no auth cap (migrating?), doing nothing\n");
1274 goto out;
1275 }
Sage Weile8351242010-09-17 08:03:08 -07001276
1277 /* only flush each capsnap once */
Yan, Zhengaffbc192015-05-05 21:22:13 +08001278 if (!kick && !list_empty(&capsnap->flushing_item)) {
Sage Weile8351242010-09-17 08:03:08 -07001279 dout("already flushed %p, skipping\n", capsnap);
1280 continue;
1281 }
1282
Sage Weilca81f3f2010-06-10 14:21:36 -07001283 mds = ci->i_auth_cap->session->s_mds;
1284 mseq = ci->i_auth_cap->mseq;
1285
Sage Weila8599bd2009-10-06 11:31:12 -07001286 if (session && session->s_mds != mds) {
1287 dout("oops, wrong session %p mutex\n", session);
Yan, Zhengaffbc192015-05-05 21:22:13 +08001288 if (kick)
1289 goto out;
1290
Sage Weila8599bd2009-10-06 11:31:12 -07001291 mutex_unlock(&session->s_mutex);
1292 ceph_put_mds_session(session);
1293 session = NULL;
1294 }
1295 if (!session) {
Sage Weilbe655592011-11-30 09:47:09 -08001296 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001297 mutex_lock(&mdsc->mutex);
1298 session = __ceph_lookup_mds_session(mdsc, mds);
1299 mutex_unlock(&mdsc->mutex);
1300 if (session) {
1301 dout("inverting session/ino locks on %p\n",
1302 session);
1303 mutex_lock(&session->s_mutex);
1304 }
1305 /*
1306 * if session == NULL, we raced against a cap
Sage Weilca81f3f2010-06-10 14:21:36 -07001307 * deletion or migration. retry, and we'll
1308 * get a better @mds value next time.
Sage Weila8599bd2009-10-06 11:31:12 -07001309 */
Sage Weilbe655592011-11-30 09:47:09 -08001310 spin_lock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001311 goto retry;
1312 }
1313
1314 capsnap->flush_tid = ++ci->i_cap_flush_last_tid;
1315 atomic_inc(&capsnap->nref);
Yan, Zhengaffbc192015-05-05 21:22:13 +08001316 if (list_empty(&capsnap->flushing_item))
1317 list_add_tail(&capsnap->flushing_item,
1318 &session->s_cap_snaps_flushing);
Sage Weilbe655592011-11-30 09:47:09 -08001319 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001320
Sage Weilcfc0bf62010-09-14 15:50:59 -07001321 dout("flush_snaps %p cap_snap %p follows %lld tid %llu\n",
1322 inode, capsnap, capsnap->follows, capsnap->flush_tid);
Sage Weila8599bd2009-10-06 11:31:12 -07001323 send_cap_msg(session, ceph_vino(inode).ino, 0,
1324 CEPH_CAP_OP_FLUSHSNAP, capsnap->issued, 0,
1325 capsnap->dirty, 0, capsnap->flush_tid, 0, mseq,
1326 capsnap->size, 0,
1327 &capsnap->mtime, &capsnap->atime,
1328 capsnap->time_warp_seq,
1329 capsnap->uid, capsnap->gid, capsnap->mode,
Sage Weil4a625be2010-08-22 15:03:56 -07001330 capsnap->xattr_version, capsnap->xattr_blob,
Yan, Zhenge20d2582014-11-14 22:39:13 +08001331 capsnap->follows, capsnap->inline_data);
Sage Weila8599bd2009-10-06 11:31:12 -07001332
1333 next_follows = capsnap->follows + 1;
1334 ceph_put_cap_snap(capsnap);
1335
Sage Weilbe655592011-11-30 09:47:09 -08001336 spin_lock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001337 goto retry;
1338 }
1339
1340 /* we flushed them all; remove this inode from the queue */
1341 spin_lock(&mdsc->snap_flush_lock);
1342 list_del_init(&ci->i_snap_flush_item);
1343 spin_unlock(&mdsc->snap_flush_lock);
1344
Sage Weilca81f3f2010-06-10 14:21:36 -07001345out:
Sage Weila8599bd2009-10-06 11:31:12 -07001346 if (psession)
1347 *psession = session;
1348 else if (session) {
1349 mutex_unlock(&session->s_mutex);
1350 ceph_put_mds_session(session);
1351 }
1352}
1353
1354static void ceph_flush_snaps(struct ceph_inode_info *ci)
1355{
Sage Weilbe655592011-11-30 09:47:09 -08001356 spin_lock(&ci->i_ceph_lock);
Sage Weile8351242010-09-17 08:03:08 -07001357 __ceph_flush_snaps(ci, NULL, 0);
Sage Weilbe655592011-11-30 09:47:09 -08001358 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001359}
1360
1361/*
Sage Weilfca65b42011-05-04 11:33:47 -07001362 * Mark caps dirty. If inode is newly dirty, return the dirty flags.
1363 * Caller is then responsible for calling __mark_inode_dirty with the
1364 * returned flags value.
Sage Weil76e3b392009-10-15 18:13:53 -07001365 */
Sage Weilfca65b42011-05-04 11:33:47 -07001366int __ceph_mark_dirty_caps(struct ceph_inode_info *ci, int mask)
Sage Weil76e3b392009-10-15 18:13:53 -07001367{
Cheng Renquan640ef792010-03-26 17:40:33 +08001368 struct ceph_mds_client *mdsc =
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001369 ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc;
Sage Weil76e3b392009-10-15 18:13:53 -07001370 struct inode *inode = &ci->vfs_inode;
1371 int was = ci->i_dirty_caps;
1372 int dirty = 0;
1373
Yan, Zheng571ade32015-03-24 11:36:08 +08001374 if (!ci->i_auth_cap) {
1375 pr_warn("__mark_dirty_caps %p %llx mask %s, "
1376 "but no auth cap (session was closed?)\n",
1377 inode, ceph_ino(inode), ceph_cap_string(mask));
1378 return 0;
1379 }
1380
Sage Weil76e3b392009-10-15 18:13:53 -07001381 dout("__mark_dirty_caps %p %s dirty %s -> %s\n", &ci->vfs_inode,
1382 ceph_cap_string(mask), ceph_cap_string(was),
1383 ceph_cap_string(was | mask));
1384 ci->i_dirty_caps |= mask;
1385 if (was == 0) {
Yan, Zheng604d1b02015-05-01 17:49:16 +08001386 if (!ci->i_head_snapc) {
1387 WARN_ON_ONCE(!rwsem_is_locked(&mdsc->snap_rwsem));
Sage Weil7d8cb262010-08-24 08:44:16 -07001388 ci->i_head_snapc = ceph_get_snap_context(
1389 ci->i_snap_realm->cached_context);
Yan, Zheng604d1b02015-05-01 17:49:16 +08001390 }
Yan, Zheng06852352012-11-19 10:49:07 +08001391 dout(" inode %p now dirty snapc %p auth cap %p\n",
1392 &ci->vfs_inode, ci->i_head_snapc, ci->i_auth_cap);
Sage Weil76e3b392009-10-15 18:13:53 -07001393 BUG_ON(!list_empty(&ci->i_dirty_item));
1394 spin_lock(&mdsc->cap_dirty_lock);
Yan, Zheng11df2df2013-11-24 14:44:38 +08001395 list_add(&ci->i_dirty_item, &mdsc->cap_dirty);
Sage Weil76e3b392009-10-15 18:13:53 -07001396 spin_unlock(&mdsc->cap_dirty_lock);
1397 if (ci->i_flushing_caps == 0) {
Sage Weil3772d262011-05-03 09:28:08 -07001398 ihold(inode);
Sage Weil76e3b392009-10-15 18:13:53 -07001399 dirty |= I_DIRTY_SYNC;
1400 }
1401 }
1402 BUG_ON(list_empty(&ci->i_dirty_item));
1403 if (((was | ci->i_flushing_caps) & CEPH_CAP_FILE_BUFFER) &&
1404 (mask & CEPH_CAP_FILE_BUFFER))
1405 dirty |= I_DIRTY_DATASYNC;
Sage Weil76e3b392009-10-15 18:13:53 -07001406 __cap_delay_requeue(mdsc, ci);
Sage Weilfca65b42011-05-04 11:33:47 -07001407 return dirty;
Sage Weil76e3b392009-10-15 18:13:53 -07001408}
1409
1410/*
Sage Weila8599bd2009-10-06 11:31:12 -07001411 * Add dirty inode to the flushing list. Assigned a seq number so we
1412 * can wait for caps to flush without starving.
Sage Weilcdc35f92009-10-14 14:24:19 -07001413 *
Sage Weilbe655592011-11-30 09:47:09 -08001414 * Called under i_ceph_lock.
Sage Weila8599bd2009-10-06 11:31:12 -07001415 */
Sage Weilcdc35f92009-10-14 14:24:19 -07001416static int __mark_caps_flushing(struct inode *inode,
Sage Weila8599bd2009-10-06 11:31:12 -07001417 struct ceph_mds_session *session)
1418{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001419 struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
Sage Weila8599bd2009-10-06 11:31:12 -07001420 struct ceph_inode_info *ci = ceph_inode(inode);
Sage Weilcdc35f92009-10-14 14:24:19 -07001421 int flushing;
Sage Weil50b885b2009-12-01 14:12:07 -08001422
Sage Weilcdc35f92009-10-14 14:24:19 -07001423 BUG_ON(ci->i_dirty_caps == 0);
Sage Weila8599bd2009-10-06 11:31:12 -07001424 BUG_ON(list_empty(&ci->i_dirty_item));
Sage Weilcdc35f92009-10-14 14:24:19 -07001425
1426 flushing = ci->i_dirty_caps;
1427 dout("__mark_caps_flushing flushing %s, flushing_caps %s -> %s\n",
1428 ceph_cap_string(flushing),
1429 ceph_cap_string(ci->i_flushing_caps),
1430 ceph_cap_string(ci->i_flushing_caps | flushing));
1431 ci->i_flushing_caps |= flushing;
1432 ci->i_dirty_caps = 0;
Sage Weilafcdaea2009-10-14 14:27:38 -07001433 dout(" inode %p now !dirty\n", inode);
Sage Weilcdc35f92009-10-14 14:24:19 -07001434
Sage Weila8599bd2009-10-06 11:31:12 -07001435 spin_lock(&mdsc->cap_dirty_lock);
Sage Weilafcdaea2009-10-14 14:27:38 -07001436 list_del_init(&ci->i_dirty_item);
1437
Sage Weila8599bd2009-10-06 11:31:12 -07001438 if (list_empty(&ci->i_flushing_item)) {
Yan, Zhengd3383a82015-01-08 21:30:12 +08001439 ci->i_cap_flush_seq = ++mdsc->cap_flush_seq;
Sage Weila8599bd2009-10-06 11:31:12 -07001440 list_add_tail(&ci->i_flushing_item, &session->s_cap_flushing);
1441 mdsc->num_cap_flushing++;
Sage Weilafcdaea2009-10-14 14:27:38 -07001442 dout(" inode %p now flushing seq %lld\n", inode,
1443 ci->i_cap_flush_seq);
1444 } else {
1445 list_move_tail(&ci->i_flushing_item, &session->s_cap_flushing);
1446 dout(" inode %p now flushing (more) seq %lld\n", inode,
Sage Weila8599bd2009-10-06 11:31:12 -07001447 ci->i_cap_flush_seq);
1448 }
1449 spin_unlock(&mdsc->cap_dirty_lock);
Sage Weilcdc35f92009-10-14 14:24:19 -07001450
1451 return flushing;
Sage Weila8599bd2009-10-06 11:31:12 -07001452}
1453
1454/*
Sage Weil5ecad6f2010-02-17 10:43:37 -08001455 * try to invalidate mapping pages without blocking.
1456 */
Sage Weil5ecad6f2010-02-17 10:43:37 -08001457static int try_nonblocking_invalidate(struct inode *inode)
1458{
1459 struct ceph_inode_info *ci = ceph_inode(inode);
1460 u32 invalidating_gen = ci->i_rdcache_gen;
1461
Sage Weilbe655592011-11-30 09:47:09 -08001462 spin_unlock(&ci->i_ceph_lock);
Sage Weil5ecad6f2010-02-17 10:43:37 -08001463 invalidate_mapping_pages(&inode->i_data, 0, -1);
Sage Weilbe655592011-11-30 09:47:09 -08001464 spin_lock(&ci->i_ceph_lock);
Sage Weil5ecad6f2010-02-17 10:43:37 -08001465
Sage Weil18a38192010-09-17 10:46:44 -07001466 if (inode->i_data.nrpages == 0 &&
Sage Weil5ecad6f2010-02-17 10:43:37 -08001467 invalidating_gen == ci->i_rdcache_gen) {
1468 /* success. */
1469 dout("try_nonblocking_invalidate %p success\n", inode);
Sage Weilcd045cb2010-11-04 11:05:05 -07001470 /* save any racing async invalidate some trouble */
1471 ci->i_rdcache_revoking = ci->i_rdcache_gen - 1;
Sage Weil5ecad6f2010-02-17 10:43:37 -08001472 return 0;
1473 }
1474 dout("try_nonblocking_invalidate %p failed\n", inode);
1475 return -1;
1476}
1477
1478/*
Sage Weila8599bd2009-10-06 11:31:12 -07001479 * Swiss army knife function to examine currently used and wanted
1480 * versus held caps. Release, flush, ack revoked caps to mds as
1481 * appropriate.
1482 *
1483 * CHECK_CAPS_NODELAY - caller is delayed work and we should not delay
1484 * cap release further.
1485 * CHECK_CAPS_AUTHONLY - we should only check the auth cap
1486 * CHECK_CAPS_FLUSH - we should flush any dirty caps immediately, without
1487 * further delay.
1488 */
1489void ceph_check_caps(struct ceph_inode_info *ci, int flags,
1490 struct ceph_mds_session *session)
1491{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001492 struct ceph_fs_client *fsc = ceph_inode_to_client(&ci->vfs_inode);
1493 struct ceph_mds_client *mdsc = fsc->mdsc;
Sage Weila8599bd2009-10-06 11:31:12 -07001494 struct inode *inode = &ci->vfs_inode;
1495 struct ceph_cap *cap;
Yan, Zheng395c3122013-01-04 14:37:57 +08001496 int file_wanted, used, cap_used;
Sage Weila8599bd2009-10-06 11:31:12 -07001497 int took_snap_rwsem = 0; /* true if mdsc->snap_rwsem held */
Sage Weilcbd03632010-02-09 13:41:18 -08001498 int issued, implemented, want, retain, revoking, flushing = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07001499 int mds = -1; /* keep track of how far we've gone through i_caps list
1500 to avoid an infinite loop on retry */
1501 struct rb_node *p;
1502 int tried_invalidate = 0;
1503 int delayed = 0, sent = 0, force_requeue = 0, num;
Sage Weilcbd03632010-02-09 13:41:18 -08001504 int queue_invalidate = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07001505 int is_delayed = flags & CHECK_CAPS_NODELAY;
1506
1507 /* if we are unmounting, flush any unused caps immediately. */
1508 if (mdsc->stopping)
1509 is_delayed = 1;
1510
Sage Weilbe655592011-11-30 09:47:09 -08001511 spin_lock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001512
1513 if (ci->i_ceph_flags & CEPH_I_FLUSH)
1514 flags |= CHECK_CAPS_FLUSH;
1515
1516 /* flush snaps first time around only */
1517 if (!list_empty(&ci->i_cap_snaps))
Sage Weile8351242010-09-17 08:03:08 -07001518 __ceph_flush_snaps(ci, &session, 0);
Sage Weila8599bd2009-10-06 11:31:12 -07001519 goto retry_locked;
1520retry:
Sage Weilbe655592011-11-30 09:47:09 -08001521 spin_lock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001522retry_locked:
1523 file_wanted = __ceph_caps_file_wanted(ci);
1524 used = __ceph_caps_used(ci);
Sage Weilcbd03632010-02-09 13:41:18 -08001525 issued = __ceph_caps_issued(ci, &implemented);
1526 revoking = implemented & ~issued;
Sage Weila8599bd2009-10-06 11:31:12 -07001527
Yan, Zheng41445992015-05-25 17:36:42 +08001528 want = file_wanted;
1529 retain = file_wanted | used | CEPH_CAP_PIN;
Sage Weila8599bd2009-10-06 11:31:12 -07001530 if (!mdsc->stopping && inode->i_nlink > 0) {
Yan, Zheng41445992015-05-25 17:36:42 +08001531 if (file_wanted) {
Sage Weila8599bd2009-10-06 11:31:12 -07001532 retain |= CEPH_CAP_ANY; /* be greedy */
Yan, Zheng32ec4392015-03-26 19:06:00 +08001533 } else if (S_ISDIR(inode->i_mode) &&
1534 (issued & CEPH_CAP_FILE_SHARED) &&
1535 __ceph_dir_is_complete(ci)) {
1536 /*
1537 * If a directory is complete, we want to keep
1538 * the exclusive cap. So that MDS does not end up
1539 * revoking the shared cap on every create/unlink
1540 * operation.
1541 */
1542 want = CEPH_CAP_ANY_SHARED | CEPH_CAP_FILE_EXCL;
1543 retain |= want;
Sage Weila8599bd2009-10-06 11:31:12 -07001544 } else {
Yan, Zheng32ec4392015-03-26 19:06:00 +08001545
Sage Weila8599bd2009-10-06 11:31:12 -07001546 retain |= CEPH_CAP_ANY_SHARED;
1547 /*
1548 * keep RD only if we didn't have the file open RW,
1549 * because then the mds would revoke it anyway to
1550 * journal max_size=0.
1551 */
1552 if (ci->i_max_size == 0)
1553 retain |= CEPH_CAP_ANY_RD;
1554 }
1555 }
1556
1557 dout("check_caps %p file_want %s used %s dirty %s flushing %s"
Sage Weilcbd03632010-02-09 13:41:18 -08001558 " issued %s revoking %s retain %s %s%s%s\n", inode,
Sage Weila8599bd2009-10-06 11:31:12 -07001559 ceph_cap_string(file_wanted),
1560 ceph_cap_string(used), ceph_cap_string(ci->i_dirty_caps),
1561 ceph_cap_string(ci->i_flushing_caps),
Sage Weilcbd03632010-02-09 13:41:18 -08001562 ceph_cap_string(issued), ceph_cap_string(revoking),
Sage Weila8599bd2009-10-06 11:31:12 -07001563 ceph_cap_string(retain),
1564 (flags & CHECK_CAPS_AUTHONLY) ? " AUTHONLY" : "",
1565 (flags & CHECK_CAPS_NODELAY) ? " NODELAY" : "",
1566 (flags & CHECK_CAPS_FLUSH) ? " FLUSH" : "");
1567
1568 /*
1569 * If we no longer need to hold onto old our caps, and we may
1570 * have cached pages, but don't want them, then try to invalidate.
1571 * If we fail, it's because pages are locked.... try again later.
1572 */
1573 if ((!is_delayed || mdsc->stopping) &&
1574 ci->i_wrbuffer_ref == 0 && /* no dirty pages... */
Sage Weil93afd442010-09-17 08:38:25 -07001575 inode->i_data.nrpages && /* have cached pages */
Sage Weilcbd03632010-02-09 13:41:18 -08001576 (file_wanted == 0 || /* no open files */
Sage Weil29625072010-05-27 10:40:43 -07001577 (revoking & (CEPH_CAP_FILE_CACHE|
1578 CEPH_CAP_FILE_LAZYIO))) && /* or revoking cache */
Sage Weila8599bd2009-10-06 11:31:12 -07001579 !tried_invalidate) {
Sage Weila8599bd2009-10-06 11:31:12 -07001580 dout("check_caps trying to invalidate on %p\n", inode);
Sage Weil5ecad6f2010-02-17 10:43:37 -08001581 if (try_nonblocking_invalidate(inode) < 0) {
Sage Weil29625072010-05-27 10:40:43 -07001582 if (revoking & (CEPH_CAP_FILE_CACHE|
1583 CEPH_CAP_FILE_LAZYIO)) {
Sage Weil5ecad6f2010-02-17 10:43:37 -08001584 dout("check_caps queuing invalidate\n");
1585 queue_invalidate = 1;
1586 ci->i_rdcache_revoking = ci->i_rdcache_gen;
1587 } else {
1588 dout("check_caps failed to invalidate pages\n");
1589 /* we failed to invalidate pages. check these
1590 caps again later. */
1591 force_requeue = 1;
1592 __cap_set_timeouts(mdsc, ci);
1593 }
Sage Weila8599bd2009-10-06 11:31:12 -07001594 }
1595 tried_invalidate = 1;
1596 goto retry_locked;
1597 }
1598
1599 num = 0;
1600 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
1601 cap = rb_entry(p, struct ceph_cap, ci_node);
1602 num++;
1603
1604 /* avoid looping forever */
1605 if (mds >= cap->mds ||
1606 ((flags & CHECK_CAPS_AUTHONLY) && cap != ci->i_auth_cap))
1607 continue;
1608
1609 /* NOTE: no side-effects allowed, until we take s_mutex */
1610
Yan, Zheng395c3122013-01-04 14:37:57 +08001611 cap_used = used;
1612 if (ci->i_auth_cap && cap != ci->i_auth_cap)
1613 cap_used &= ~ci->i_auth_cap->issued;
1614
Sage Weila8599bd2009-10-06 11:31:12 -07001615 revoking = cap->implemented & ~cap->issued;
Yan, Zheng395c3122013-01-04 14:37:57 +08001616 dout(" mds%d cap %p used %s issued %s implemented %s revoking %s\n",
Sage Weil088b3f52011-01-18 08:56:01 -08001617 cap->mds, cap, ceph_cap_string(cap->issued),
Yan, Zheng395c3122013-01-04 14:37:57 +08001618 ceph_cap_string(cap_used),
Sage Weil088b3f52011-01-18 08:56:01 -08001619 ceph_cap_string(cap->implemented),
1620 ceph_cap_string(revoking));
Sage Weila8599bd2009-10-06 11:31:12 -07001621
1622 if (cap == ci->i_auth_cap &&
1623 (cap->issued & CEPH_CAP_FILE_WR)) {
1624 /* request larger max_size from MDS? */
1625 if (ci->i_wanted_max_size > ci->i_max_size &&
1626 ci->i_wanted_max_size > ci->i_requested_max_size) {
1627 dout("requesting new max_size\n");
1628 goto ack;
1629 }
1630
1631 /* approaching file_max? */
1632 if ((inode->i_size << 1) >= ci->i_max_size &&
1633 (ci->i_reported_size << 1) < ci->i_max_size) {
1634 dout("i_size approaching max_size\n");
1635 goto ack;
1636 }
1637 }
1638 /* flush anything dirty? */
1639 if (cap == ci->i_auth_cap && (flags & CHECK_CAPS_FLUSH) &&
1640 ci->i_dirty_caps) {
1641 dout("flushing dirty caps\n");
1642 goto ack;
1643 }
1644
1645 /* completed revocation? going down and there are no caps? */
Yan, Zheng395c3122013-01-04 14:37:57 +08001646 if (revoking && (revoking & cap_used) == 0) {
Sage Weila8599bd2009-10-06 11:31:12 -07001647 dout("completed revocation of %s\n",
1648 ceph_cap_string(cap->implemented & ~cap->issued));
1649 goto ack;
1650 }
1651
1652 /* want more caps from mds? */
1653 if (want & ~(cap->mds_wanted | cap->issued))
1654 goto ack;
1655
1656 /* things we might delay */
1657 if ((cap->issued & ~retain) == 0 &&
1658 cap->mds_wanted == want)
1659 continue; /* nope, all good */
1660
1661 if (is_delayed)
1662 goto ack;
1663
1664 /* delay? */
1665 if ((ci->i_ceph_flags & CEPH_I_NODELAY) == 0 &&
1666 time_before(jiffies, ci->i_hold_caps_max)) {
1667 dout(" delaying issued %s -> %s, wanted %s -> %s\n",
1668 ceph_cap_string(cap->issued),
1669 ceph_cap_string(cap->issued & retain),
1670 ceph_cap_string(cap->mds_wanted),
1671 ceph_cap_string(want));
1672 delayed++;
1673 continue;
1674 }
1675
1676ack:
Sage Weile9964c12010-03-01 15:16:56 -08001677 if (ci->i_ceph_flags & CEPH_I_NOFLUSH) {
1678 dout(" skipping %p I_NOFLUSH set\n", inode);
1679 continue;
1680 }
1681
Sage Weila8599bd2009-10-06 11:31:12 -07001682 if (session && session != cap->session) {
1683 dout("oops, wrong session %p mutex\n", session);
1684 mutex_unlock(&session->s_mutex);
1685 session = NULL;
1686 }
1687 if (!session) {
1688 session = cap->session;
1689 if (mutex_trylock(&session->s_mutex) == 0) {
1690 dout("inverting session/ino locks on %p\n",
1691 session);
Sage Weilbe655592011-11-30 09:47:09 -08001692 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001693 if (took_snap_rwsem) {
1694 up_read(&mdsc->snap_rwsem);
1695 took_snap_rwsem = 0;
1696 }
1697 mutex_lock(&session->s_mutex);
1698 goto retry;
1699 }
1700 }
1701 /* take snap_rwsem after session mutex */
1702 if (!took_snap_rwsem) {
1703 if (down_read_trylock(&mdsc->snap_rwsem) == 0) {
1704 dout("inverting snap/in locks on %p\n",
1705 inode);
Sage Weilbe655592011-11-30 09:47:09 -08001706 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001707 down_read(&mdsc->snap_rwsem);
1708 took_snap_rwsem = 1;
1709 goto retry;
1710 }
1711 took_snap_rwsem = 1;
1712 }
1713
Sage Weilcdc35f92009-10-14 14:24:19 -07001714 if (cap == ci->i_auth_cap && ci->i_dirty_caps)
1715 flushing = __mark_caps_flushing(inode, session);
Sage Weil24be0c42011-01-18 08:48:06 -08001716 else
1717 flushing = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07001718
1719 mds = cap->mds; /* remember mds, so we don't repeat */
1720 sent++;
1721
Sage Weilbe655592011-11-30 09:47:09 -08001722 /* __send_cap drops i_ceph_lock */
Yan, Zheng395c3122013-01-04 14:37:57 +08001723 delayed += __send_cap(mdsc, cap, CEPH_CAP_OP_UPDATE, cap_used,
Yan, Zheng89b52fe2015-05-27 09:59:48 +08001724 want, retain, flushing);
Sage Weilbe655592011-11-30 09:47:09 -08001725 goto retry; /* retake i_ceph_lock and restart our cap scan. */
Sage Weila8599bd2009-10-06 11:31:12 -07001726 }
1727
1728 /*
1729 * Reschedule delayed caps release if we delayed anything,
1730 * otherwise cancel.
1731 */
1732 if (delayed && is_delayed)
1733 force_requeue = 1; /* __send_cap delayed release; requeue */
1734 if (!delayed && !is_delayed)
1735 __cap_delay_cancel(mdsc, ci);
1736 else if (!is_delayed || force_requeue)
1737 __cap_delay_requeue(mdsc, ci);
1738
Sage Weilbe655592011-11-30 09:47:09 -08001739 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001740
Sage Weilcbd03632010-02-09 13:41:18 -08001741 if (queue_invalidate)
Sage Weil3c6f6b792010-02-09 15:24:44 -08001742 ceph_queue_invalidate(inode);
Sage Weilcbd03632010-02-09 13:41:18 -08001743
Sage Weilcdc2ce02010-03-16 13:39:28 -07001744 if (session)
Sage Weila8599bd2009-10-06 11:31:12 -07001745 mutex_unlock(&session->s_mutex);
1746 if (took_snap_rwsem)
1747 up_read(&mdsc->snap_rwsem);
1748}
1749
1750/*
Sage Weila8599bd2009-10-06 11:31:12 -07001751 * Try to flush dirty caps back to the auth mds.
1752 */
Yan, Zheng89b52fe2015-05-27 09:59:48 +08001753static int try_flush_caps(struct inode *inode, u16 flush_tid[])
Sage Weila8599bd2009-10-06 11:31:12 -07001754{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001755 struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
Sage Weila8599bd2009-10-06 11:31:12 -07001756 struct ceph_inode_info *ci = ceph_inode(inode);
Yan, Zheng4fe59782013-10-31 16:44:14 +08001757 struct ceph_mds_session *session = NULL;
Yan, Zheng89b52fe2015-05-27 09:59:48 +08001758 int flushing = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07001759
1760retry:
Sage Weilbe655592011-11-30 09:47:09 -08001761 spin_lock(&ci->i_ceph_lock);
Sage Weile9964c12010-03-01 15:16:56 -08001762 if (ci->i_ceph_flags & CEPH_I_NOFLUSH) {
1763 dout("try_flush_caps skipping %p I_NOFLUSH set\n", inode);
1764 goto out;
1765 }
Sage Weila8599bd2009-10-06 11:31:12 -07001766 if (ci->i_dirty_caps && ci->i_auth_cap) {
1767 struct ceph_cap *cap = ci->i_auth_cap;
1768 int used = __ceph_caps_used(ci);
1769 int want = __ceph_caps_wanted(ci);
1770 int delayed;
1771
Yan, Zheng4fe59782013-10-31 16:44:14 +08001772 if (!session || session != cap->session) {
Sage Weilbe655592011-11-30 09:47:09 -08001773 spin_unlock(&ci->i_ceph_lock);
Yan, Zheng4fe59782013-10-31 16:44:14 +08001774 if (session)
1775 mutex_unlock(&session->s_mutex);
Sage Weila8599bd2009-10-06 11:31:12 -07001776 session = cap->session;
1777 mutex_lock(&session->s_mutex);
1778 goto retry;
1779 }
Sage Weila8599bd2009-10-06 11:31:12 -07001780 if (cap->session->s_state < CEPH_MDS_SESSION_OPEN)
1781 goto out;
1782
Sage Weilcdc35f92009-10-14 14:24:19 -07001783 flushing = __mark_caps_flushing(inode, session);
Sage Weila8599bd2009-10-06 11:31:12 -07001784
Sage Weilbe655592011-11-30 09:47:09 -08001785 /* __send_cap drops i_ceph_lock */
Sage Weila8599bd2009-10-06 11:31:12 -07001786 delayed = __send_cap(mdsc, cap, CEPH_CAP_OP_FLUSH, used, want,
Yan, Zheng89b52fe2015-05-27 09:59:48 +08001787 cap->issued | cap->implemented, flushing);
Sage Weila8599bd2009-10-06 11:31:12 -07001788
Sage Weilbe655592011-11-30 09:47:09 -08001789 spin_lock(&ci->i_ceph_lock);
Yan, Zheng89b52fe2015-05-27 09:59:48 +08001790 if (delayed)
1791 __cap_delay_requeue(mdsc, ci);
Sage Weila8599bd2009-10-06 11:31:12 -07001792 }
Yan, Zheng89b52fe2015-05-27 09:59:48 +08001793
1794 flushing = ci->i_flushing_caps;
1795 if (flushing)
1796 memcpy(flush_tid, ci->i_cap_flush_tid,
1797 sizeof(ci->i_cap_flush_tid));
Sage Weila8599bd2009-10-06 11:31:12 -07001798out:
Sage Weilbe655592011-11-30 09:47:09 -08001799 spin_unlock(&ci->i_ceph_lock);
Yan, Zheng4fe59782013-10-31 16:44:14 +08001800 if (session)
Sage Weila8599bd2009-10-06 11:31:12 -07001801 mutex_unlock(&session->s_mutex);
1802 return flushing;
1803}
1804
1805/*
1806 * Return true if we've flushed caps through the given flush_tid.
1807 */
Yan, Zheng89b52fe2015-05-27 09:59:48 +08001808static int caps_are_flushed(struct inode *inode, u16 flush_tid[])
Sage Weila8599bd2009-10-06 11:31:12 -07001809{
1810 struct ceph_inode_info *ci = ceph_inode(inode);
Dan Carpentera5ee7512010-05-07 10:27:14 +02001811 int i, ret = 1;
Sage Weila8599bd2009-10-06 11:31:12 -07001812
Sage Weilbe655592011-11-30 09:47:09 -08001813 spin_lock(&ci->i_ceph_lock);
Yan, Zheng89b52fe2015-05-27 09:59:48 +08001814 for (i = 0; i < CEPH_CAP_BITS; i++) {
1815 if (!(ci->i_flushing_caps & (1 << i)))
1816 continue;
1817 // tid only has 16 bits. we need to handle wrapping
1818 if ((s16)(ci->i_cap_flush_tid[i] - flush_tid[i]) <= 0) {
Sage Weila8599bd2009-10-06 11:31:12 -07001819 /* still flushing this bit */
1820 ret = 0;
1821 break;
1822 }
Yan, Zheng89b52fe2015-05-27 09:59:48 +08001823 }
Sage Weilbe655592011-11-30 09:47:09 -08001824 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001825 return ret;
1826}
1827
1828/*
1829 * Wait on any unsafe replies for the given inode. First wait on the
1830 * newest request, and make that the upper bound. Then, if there are
1831 * more requests, keep waiting on the oldest as long as it is still older
1832 * than the original request.
1833 */
1834static void sync_write_wait(struct inode *inode)
1835{
1836 struct ceph_inode_info *ci = ceph_inode(inode);
1837 struct list_head *head = &ci->i_unsafe_writes;
1838 struct ceph_osd_request *req;
1839 u64 last_tid;
1840
1841 spin_lock(&ci->i_unsafe_lock);
1842 if (list_empty(head))
1843 goto out;
1844
1845 /* set upper bound as _last_ entry in chain */
1846 req = list_entry(head->prev, struct ceph_osd_request,
1847 r_unsafe_item);
1848 last_tid = req->r_tid;
1849
1850 do {
1851 ceph_osdc_get_request(req);
1852 spin_unlock(&ci->i_unsafe_lock);
1853 dout("sync_write_wait on tid %llu (until %llu)\n",
1854 req->r_tid, last_tid);
1855 wait_for_completion(&req->r_safe_completion);
1856 spin_lock(&ci->i_unsafe_lock);
1857 ceph_osdc_put_request(req);
1858
1859 /*
1860 * from here on look at first entry in chain, since we
1861 * only want to wait for anything older than last_tid
1862 */
1863 if (list_empty(head))
1864 break;
1865 req = list_entry(head->next, struct ceph_osd_request,
1866 r_unsafe_item);
1867 } while (req->r_tid < last_tid);
1868out:
1869 spin_unlock(&ci->i_unsafe_lock);
1870}
1871
Josef Bacik02c24a82011-07-16 20:44:56 -04001872int ceph_fsync(struct file *file, loff_t start, loff_t end, int datasync)
Sage Weila8599bd2009-10-06 11:31:12 -07001873{
Christoph Hellwig7ea80852010-05-26 17:53:25 +02001874 struct inode *inode = file->f_mapping->host;
Sage Weila8599bd2009-10-06 11:31:12 -07001875 struct ceph_inode_info *ci = ceph_inode(inode);
Yan, Zheng89b52fe2015-05-27 09:59:48 +08001876 u16 flush_tid[CEPH_CAP_BITS];
Sage Weila8599bd2009-10-06 11:31:12 -07001877 int ret;
1878 int dirty;
1879
1880 dout("fsync %p%s\n", inode, datasync ? " datasync" : "");
1881 sync_write_wait(inode);
1882
Josef Bacik02c24a82011-07-16 20:44:56 -04001883 ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
Sage Weila8599bd2009-10-06 11:31:12 -07001884 if (ret < 0)
1885 return ret;
Josef Bacik02c24a82011-07-16 20:44:56 -04001886 mutex_lock(&inode->i_mutex);
Sage Weila8599bd2009-10-06 11:31:12 -07001887
Yan, Zheng89b52fe2015-05-27 09:59:48 +08001888 dirty = try_flush_caps(inode, flush_tid);
Sage Weila8599bd2009-10-06 11:31:12 -07001889 dout("fsync dirty caps are %s\n", ceph_cap_string(dirty));
1890
1891 /*
1892 * only wait on non-file metadata writeback (the mds
1893 * can recover size and mtime, so we don't need to
1894 * wait for that)
1895 */
1896 if (!datasync && (dirty & ~CEPH_CAP_ANY_FILE_WR)) {
Sage Weila8599bd2009-10-06 11:31:12 -07001897 ret = wait_event_interruptible(ci->i_cap_wq,
1898 caps_are_flushed(inode, flush_tid));
1899 }
1900
1901 dout("fsync %p%s done\n", inode, datasync ? " datasync" : "");
Josef Bacik02c24a82011-07-16 20:44:56 -04001902 mutex_unlock(&inode->i_mutex);
Sage Weila8599bd2009-10-06 11:31:12 -07001903 return ret;
1904}
1905
1906/*
1907 * Flush any dirty caps back to the mds. If we aren't asked to wait,
1908 * queue inode for flush but don't do so immediately, because we can
1909 * get by with fewer MDS messages if we wait for data writeback to
1910 * complete first.
1911 */
Stephen Rothwellf1a3d572010-01-18 11:53:08 +11001912int ceph_write_inode(struct inode *inode, struct writeback_control *wbc)
Sage Weila8599bd2009-10-06 11:31:12 -07001913{
1914 struct ceph_inode_info *ci = ceph_inode(inode);
Yan, Zheng89b52fe2015-05-27 09:59:48 +08001915 u16 flush_tid[CEPH_CAP_BITS];
Sage Weila8599bd2009-10-06 11:31:12 -07001916 int err = 0;
1917 int dirty;
Stephen Rothwellf1a3d572010-01-18 11:53:08 +11001918 int wait = wbc->sync_mode == WB_SYNC_ALL;
Sage Weila8599bd2009-10-06 11:31:12 -07001919
1920 dout("write_inode %p wait=%d\n", inode, wait);
1921 if (wait) {
Yan, Zheng89b52fe2015-05-27 09:59:48 +08001922 dirty = try_flush_caps(inode, flush_tid);
Sage Weila8599bd2009-10-06 11:31:12 -07001923 if (dirty)
1924 err = wait_event_interruptible(ci->i_cap_wq,
1925 caps_are_flushed(inode, flush_tid));
1926 } else {
Cheng Renquan640ef792010-03-26 17:40:33 +08001927 struct ceph_mds_client *mdsc =
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001928 ceph_sb_to_client(inode->i_sb)->mdsc;
Sage Weila8599bd2009-10-06 11:31:12 -07001929
Sage Weilbe655592011-11-30 09:47:09 -08001930 spin_lock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001931 if (__ceph_caps_dirty(ci))
1932 __cap_delay_requeue_front(mdsc, ci);
Sage Weilbe655592011-11-30 09:47:09 -08001933 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001934 }
1935 return err;
1936}
1937
1938/*
1939 * After a recovering MDS goes active, we need to resend any caps
1940 * we were flushing.
1941 *
1942 * Caller holds session->s_mutex.
1943 */
1944static void kick_flushing_capsnaps(struct ceph_mds_client *mdsc,
1945 struct ceph_mds_session *session)
1946{
1947 struct ceph_cap_snap *capsnap;
1948
1949 dout("kick_flushing_capsnaps mds%d\n", session->s_mds);
1950 list_for_each_entry(capsnap, &session->s_cap_snaps_flushing,
1951 flushing_item) {
1952 struct ceph_inode_info *ci = capsnap->ci;
1953 struct inode *inode = &ci->vfs_inode;
1954 struct ceph_cap *cap;
1955
Sage Weilbe655592011-11-30 09:47:09 -08001956 spin_lock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001957 cap = ci->i_auth_cap;
1958 if (cap && cap->session == session) {
1959 dout("kick_flushing_caps %p cap %p capsnap %p\n", inode,
1960 cap, capsnap);
Sage Weile8351242010-09-17 08:03:08 -07001961 __ceph_flush_snaps(ci, &session, 1);
Sage Weila8599bd2009-10-06 11:31:12 -07001962 } else {
1963 pr_err("%p auth cap %p not mds%d ???\n", inode,
1964 cap, session->s_mds);
Sage Weila8599bd2009-10-06 11:31:12 -07001965 }
Sage Weilbe655592011-11-30 09:47:09 -08001966 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001967 }
1968}
1969
1970void ceph_kick_flushing_caps(struct ceph_mds_client *mdsc,
1971 struct ceph_mds_session *session)
1972{
1973 struct ceph_inode_info *ci;
1974
1975 kick_flushing_capsnaps(mdsc, session);
1976
1977 dout("kick_flushing_caps mds%d\n", session->s_mds);
1978 list_for_each_entry(ci, &session->s_cap_flushing, i_flushing_item) {
1979 struct inode *inode = &ci->vfs_inode;
1980 struct ceph_cap *cap;
1981 int delayed = 0;
1982
Sage Weilbe655592011-11-30 09:47:09 -08001983 spin_lock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001984 cap = ci->i_auth_cap;
1985 if (cap && cap->session == session) {
1986 dout("kick_flushing_caps %p cap %p %s\n", inode,
1987 cap, ceph_cap_string(ci->i_flushing_caps));
1988 delayed = __send_cap(mdsc, cap, CEPH_CAP_OP_FLUSH,
1989 __ceph_caps_used(ci),
1990 __ceph_caps_wanted(ci),
1991 cap->issued | cap->implemented,
Yan, Zheng89b52fe2015-05-27 09:59:48 +08001992 ci->i_flushing_caps);
Sage Weila8599bd2009-10-06 11:31:12 -07001993 if (delayed) {
Sage Weilbe655592011-11-30 09:47:09 -08001994 spin_lock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001995 __cap_delay_requeue(mdsc, ci);
Sage Weilbe655592011-11-30 09:47:09 -08001996 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07001997 }
1998 } else {
1999 pr_err("%p auth cap %p not mds%d ???\n", inode,
2000 cap, session->s_mds);
Sage Weilbe655592011-11-30 09:47:09 -08002001 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07002002 }
2003 }
2004}
2005
Sage Weil088b3f52011-01-18 08:56:01 -08002006static void kick_flushing_inode_caps(struct ceph_mds_client *mdsc,
2007 struct ceph_mds_session *session,
2008 struct inode *inode)
2009{
2010 struct ceph_inode_info *ci = ceph_inode(inode);
2011 struct ceph_cap *cap;
2012 int delayed = 0;
2013
Sage Weilbe655592011-11-30 09:47:09 -08002014 spin_lock(&ci->i_ceph_lock);
Sage Weil088b3f52011-01-18 08:56:01 -08002015 cap = ci->i_auth_cap;
2016 dout("kick_flushing_inode_caps %p flushing %s flush_seq %lld\n", inode,
2017 ceph_cap_string(ci->i_flushing_caps), ci->i_cap_flush_seq);
Yan, Zheng005c4692013-05-31 16:40:24 +08002018
Sage Weil088b3f52011-01-18 08:56:01 -08002019 __ceph_flush_snaps(ci, &session, 1);
Yan, Zheng005c4692013-05-31 16:40:24 +08002020
Sage Weil088b3f52011-01-18 08:56:01 -08002021 if (ci->i_flushing_caps) {
Yan, Zheng005c4692013-05-31 16:40:24 +08002022 spin_lock(&mdsc->cap_dirty_lock);
2023 list_move_tail(&ci->i_flushing_item,
2024 &cap->session->s_cap_flushing);
2025 spin_unlock(&mdsc->cap_dirty_lock);
2026
Sage Weil088b3f52011-01-18 08:56:01 -08002027 delayed = __send_cap(mdsc, cap, CEPH_CAP_OP_FLUSH,
2028 __ceph_caps_used(ci),
2029 __ceph_caps_wanted(ci),
2030 cap->issued | cap->implemented,
Yan, Zheng89b52fe2015-05-27 09:59:48 +08002031 ci->i_flushing_caps);
Sage Weil088b3f52011-01-18 08:56:01 -08002032 if (delayed) {
Sage Weilbe655592011-11-30 09:47:09 -08002033 spin_lock(&ci->i_ceph_lock);
Sage Weil088b3f52011-01-18 08:56:01 -08002034 __cap_delay_requeue(mdsc, ci);
Sage Weilbe655592011-11-30 09:47:09 -08002035 spin_unlock(&ci->i_ceph_lock);
Sage Weil088b3f52011-01-18 08:56:01 -08002036 }
2037 } else {
Sage Weilbe655592011-11-30 09:47:09 -08002038 spin_unlock(&ci->i_ceph_lock);
Sage Weil088b3f52011-01-18 08:56:01 -08002039 }
2040}
2041
Sage Weila8599bd2009-10-06 11:31:12 -07002042
2043/*
2044 * Take references to capabilities we hold, so that we don't release
2045 * them to the MDS prematurely.
2046 *
Sage Weilbe655592011-11-30 09:47:09 -08002047 * Protected by i_ceph_lock.
Sage Weila8599bd2009-10-06 11:31:12 -07002048 */
Yan, Zheng5dda377c2015-04-30 14:40:54 +08002049static void __take_cap_refs(struct ceph_inode_info *ci, int got,
2050 bool snap_rwsem_locked)
Sage Weila8599bd2009-10-06 11:31:12 -07002051{
2052 if (got & CEPH_CAP_PIN)
2053 ci->i_pin_ref++;
2054 if (got & CEPH_CAP_FILE_RD)
2055 ci->i_rd_ref++;
2056 if (got & CEPH_CAP_FILE_CACHE)
2057 ci->i_rdcache_ref++;
Yan, Zheng5dda377c2015-04-30 14:40:54 +08002058 if (got & CEPH_CAP_FILE_WR) {
2059 if (ci->i_wr_ref == 0 && !ci->i_head_snapc) {
2060 BUG_ON(!snap_rwsem_locked);
2061 ci->i_head_snapc = ceph_get_snap_context(
2062 ci->i_snap_realm->cached_context);
2063 }
Sage Weila8599bd2009-10-06 11:31:12 -07002064 ci->i_wr_ref++;
Yan, Zheng5dda377c2015-04-30 14:40:54 +08002065 }
Sage Weila8599bd2009-10-06 11:31:12 -07002066 if (got & CEPH_CAP_FILE_BUFFER) {
Henry C Changd3d07202011-05-11 10:29:54 +00002067 if (ci->i_wb_ref == 0)
Sage Weil3772d262011-05-03 09:28:08 -07002068 ihold(&ci->vfs_inode);
Henry C Changd3d07202011-05-11 10:29:54 +00002069 ci->i_wb_ref++;
2070 dout("__take_cap_refs %p wb %d -> %d (?)\n",
2071 &ci->vfs_inode, ci->i_wb_ref-1, ci->i_wb_ref);
Sage Weila8599bd2009-10-06 11:31:12 -07002072 }
2073}
2074
2075/*
2076 * Try to grab cap references. Specify those refs we @want, and the
2077 * minimal set we @need. Also include the larger offset we are writing
2078 * to (when applicable), and check against max_size here as well.
2079 * Note that caller is responsible for ensuring max_size increases are
2080 * requested from the MDS.
2081 */
2082static int try_get_cap_refs(struct ceph_inode_info *ci, int need, int want,
Yan, Zheng5dda377c2015-04-30 14:40:54 +08002083 loff_t endoff, bool nonblock, int *got, int *err)
Sage Weila8599bd2009-10-06 11:31:12 -07002084{
2085 struct inode *inode = &ci->vfs_inode;
Yan, Zheng5dda377c2015-04-30 14:40:54 +08002086 struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
Sage Weila8599bd2009-10-06 11:31:12 -07002087 int ret = 0;
Yan, Zhengc4d4a582015-01-09 15:56:18 +08002088 int have, implemented;
Sage Weil195d3ce2010-03-01 09:57:54 -08002089 int file_wanted;
Yan, Zheng5dda377c2015-04-30 14:40:54 +08002090 bool snap_rwsem_locked = false;
Sage Weila8599bd2009-10-06 11:31:12 -07002091
2092 dout("get_cap_refs %p need %s want %s\n", inode,
2093 ceph_cap_string(need), ceph_cap_string(want));
Yan, Zhengc4d4a582015-01-09 15:56:18 +08002094
Yan, Zheng5dda377c2015-04-30 14:40:54 +08002095again:
Sage Weilbe655592011-11-30 09:47:09 -08002096 spin_lock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07002097
Sage Weil195d3ce2010-03-01 09:57:54 -08002098 /* make sure file is actually open */
2099 file_wanted = __ceph_caps_file_wanted(ci);
2100 if ((file_wanted & need) == 0) {
2101 dout("try_get_cap_refs need %s file_wanted %s, EBADF\n",
2102 ceph_cap_string(need), ceph_cap_string(file_wanted));
Sage Weila8599bd2009-10-06 11:31:12 -07002103 *err = -EBADF;
2104 ret = 1;
Yan, Zheng3738daa2014-11-14 22:10:07 +08002105 goto out_unlock;
Sage Weila8599bd2009-10-06 11:31:12 -07002106 }
2107
Yan, Zheng37505d52013-04-12 16:11:10 +08002108 /* finish pending truncate */
2109 while (ci->i_truncate_pending) {
2110 spin_unlock(&ci->i_ceph_lock);
Yan, Zheng5dda377c2015-04-30 14:40:54 +08002111 if (snap_rwsem_locked) {
2112 up_read(&mdsc->snap_rwsem);
2113 snap_rwsem_locked = false;
2114 }
Yan, Zhengb415bf42013-07-02 12:40:19 +08002115 __ceph_do_pending_vmtruncate(inode);
Yan, Zheng37505d52013-04-12 16:11:10 +08002116 spin_lock(&ci->i_ceph_lock);
2117 }
2118
Yan, Zheng3871cbb2013-08-05 14:10:29 +08002119 have = __ceph_caps_issued(ci, &implemented);
2120
2121 if (have & need & CEPH_CAP_FILE_WR) {
Sage Weila8599bd2009-10-06 11:31:12 -07002122 if (endoff >= 0 && endoff > (loff_t)ci->i_max_size) {
2123 dout("get_cap_refs %p endoff %llu > maxsize %llu\n",
2124 inode, endoff, ci->i_max_size);
Yan, Zheng3871cbb2013-08-05 14:10:29 +08002125 if (endoff > ci->i_requested_max_size) {
Yan, Zheng5dda377c2015-04-30 14:40:54 +08002126 *err = -EAGAIN;
Sage Weila8599bd2009-10-06 11:31:12 -07002127 ret = 1;
2128 }
Yan, Zheng3738daa2014-11-14 22:10:07 +08002129 goto out_unlock;
Sage Weila8599bd2009-10-06 11:31:12 -07002130 }
2131 /*
2132 * If a sync write is in progress, we must wait, so that we
2133 * can get a final snapshot value for size+mtime.
2134 */
2135 if (__ceph_have_pending_cap_snap(ci)) {
2136 dout("get_cap_refs %p cap_snap_pending\n", inode);
Yan, Zheng3738daa2014-11-14 22:10:07 +08002137 goto out_unlock;
Sage Weila8599bd2009-10-06 11:31:12 -07002138 }
2139 }
Sage Weila8599bd2009-10-06 11:31:12 -07002140
Sage Weila8599bd2009-10-06 11:31:12 -07002141 if ((have & need) == need) {
2142 /*
2143 * Look at (implemented & ~have & not) so that we keep waiting
2144 * on transition from wanted -> needed caps. This is needed
2145 * for WRBUFFER|WR -> WR to avoid a new WR sync write from
2146 * going before a prior buffered writeback happens.
2147 */
2148 int not = want & ~(have & need);
2149 int revoking = implemented & ~have;
2150 dout("get_cap_refs %p have %s but not %s (revoking %s)\n",
2151 inode, ceph_cap_string(have), ceph_cap_string(not),
2152 ceph_cap_string(revoking));
2153 if ((revoking & not) == 0) {
Yan, Zheng5dda377c2015-04-30 14:40:54 +08002154 if (!snap_rwsem_locked &&
2155 !ci->i_head_snapc &&
2156 (need & CEPH_CAP_FILE_WR)) {
2157 if (!down_read_trylock(&mdsc->snap_rwsem)) {
2158 /*
2159 * we can not call down_read() when
2160 * task isn't in TASK_RUNNING state
2161 */
2162 if (nonblock) {
2163 *err = -EAGAIN;
2164 ret = 1;
2165 goto out_unlock;
2166 }
2167
2168 spin_unlock(&ci->i_ceph_lock);
2169 down_read(&mdsc->snap_rwsem);
2170 snap_rwsem_locked = true;
2171 goto again;
2172 }
2173 snap_rwsem_locked = true;
2174 }
Yan, Zhengc4d4a582015-01-09 15:56:18 +08002175 *got = need | (have & want);
Yan, Zheng5dda377c2015-04-30 14:40:54 +08002176 __take_cap_refs(ci, *got, true);
Sage Weila8599bd2009-10-06 11:31:12 -07002177 ret = 1;
2178 }
2179 } else {
Yan, Zheng03f4fcb2015-01-05 11:04:04 +08002180 int session_readonly = false;
2181 if ((need & CEPH_CAP_FILE_WR) && ci->i_auth_cap) {
2182 struct ceph_mds_session *s = ci->i_auth_cap->session;
2183 spin_lock(&s->s_cap_lock);
2184 session_readonly = s->s_readonly;
2185 spin_unlock(&s->s_cap_lock);
2186 }
2187 if (session_readonly) {
2188 dout("get_cap_refs %p needed %s but mds%d readonly\n",
2189 inode, ceph_cap_string(need), ci->i_auth_cap->mds);
2190 *err = -EROFS;
2191 ret = 1;
2192 goto out_unlock;
2193 }
2194
Sage Weila8599bd2009-10-06 11:31:12 -07002195 dout("get_cap_refs %p have %s needed %s\n", inode,
2196 ceph_cap_string(have), ceph_cap_string(need));
2197 }
Yan, Zheng3738daa2014-11-14 22:10:07 +08002198out_unlock:
Sage Weilbe655592011-11-30 09:47:09 -08002199 spin_unlock(&ci->i_ceph_lock);
Yan, Zheng5dda377c2015-04-30 14:40:54 +08002200 if (snap_rwsem_locked)
2201 up_read(&mdsc->snap_rwsem);
Yan, Zheng3738daa2014-11-14 22:10:07 +08002202
Sage Weila8599bd2009-10-06 11:31:12 -07002203 dout("get_cap_refs %p ret %d got %s\n", inode,
Yan, Zhengc4d4a582015-01-09 15:56:18 +08002204 ret, ceph_cap_string(*got));
Sage Weila8599bd2009-10-06 11:31:12 -07002205 return ret;
2206}
2207
2208/*
2209 * Check the offset we are writing up to against our current
2210 * max_size. If necessary, tell the MDS we want to write to
2211 * a larger offset.
2212 */
2213static void check_max_size(struct inode *inode, loff_t endoff)
2214{
2215 struct ceph_inode_info *ci = ceph_inode(inode);
2216 int check = 0;
2217
2218 /* do we need to explicitly request a larger max_size? */
Sage Weilbe655592011-11-30 09:47:09 -08002219 spin_lock(&ci->i_ceph_lock);
Yan, Zheng3871cbb2013-08-05 14:10:29 +08002220 if (endoff >= ci->i_max_size && endoff > ci->i_wanted_max_size) {
Sage Weila8599bd2009-10-06 11:31:12 -07002221 dout("write %p at large endoff %llu, req max_size\n",
2222 inode, endoff);
2223 ci->i_wanted_max_size = endoff;
Sage Weila8599bd2009-10-06 11:31:12 -07002224 }
Yan, Zheng3871cbb2013-08-05 14:10:29 +08002225 /* duplicate ceph_check_caps()'s logic */
2226 if (ci->i_auth_cap &&
2227 (ci->i_auth_cap->issued & CEPH_CAP_FILE_WR) &&
2228 ci->i_wanted_max_size > ci->i_max_size &&
2229 ci->i_wanted_max_size > ci->i_requested_max_size)
2230 check = 1;
Sage Weilbe655592011-11-30 09:47:09 -08002231 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07002232 if (check)
2233 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
2234}
2235
2236/*
2237 * Wait for caps, and take cap references. If we can't get a WR cap
2238 * due to a small max_size, make sure we check_max_size (and possibly
2239 * ask the mds) so we don't get hung up indefinitely.
2240 */
Yan, Zheng3738daa2014-11-14 22:10:07 +08002241int ceph_get_caps(struct ceph_inode_info *ci, int need, int want,
2242 loff_t endoff, int *got, struct page **pinned_page)
Sage Weila8599bd2009-10-06 11:31:12 -07002243{
Yan, Zheng5dda377c2015-04-30 14:40:54 +08002244 int _got, ret, err = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07002245
Yan, Zheng10183a62015-04-27 15:33:28 +08002246 ret = ceph_pool_perm_check(ci, need);
2247 if (ret < 0)
2248 return ret;
2249
Yan, Zheng5dda377c2015-04-30 14:40:54 +08002250 while (true) {
2251 if (endoff > 0)
2252 check_max_size(&ci->vfs_inode, endoff);
Yan, Zhengc4d4a582015-01-09 15:56:18 +08002253
Yan, Zheng5dda377c2015-04-30 14:40:54 +08002254 err = 0;
Yan, Zhengc4d4a582015-01-09 15:56:18 +08002255 _got = 0;
Yan, Zheng5dda377c2015-04-30 14:40:54 +08002256 ret = try_get_cap_refs(ci, need, want, endoff,
2257 false, &_got, &err);
2258 if (ret) {
2259 if (err == -EAGAIN)
2260 continue;
2261 if (err < 0)
2262 return err;
2263 } else {
2264 ret = wait_event_interruptible(ci->i_cap_wq,
2265 try_get_cap_refs(ci, need, want, endoff,
2266 true, &_got, &err));
2267 if (err == -EAGAIN)
2268 continue;
2269 if (err < 0)
2270 ret = err;
2271 if (ret < 0)
2272 return ret;
2273 }
Yan, Zhengc4d4a582015-01-09 15:56:18 +08002274
Yan, Zheng5dda377c2015-04-30 14:40:54 +08002275 if (ci->i_inline_version != CEPH_INLINE_NONE &&
2276 (_got & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) &&
2277 i_size_read(&ci->vfs_inode) > 0) {
2278 struct page *page =
2279 find_get_page(ci->vfs_inode.i_mapping, 0);
2280 if (page) {
2281 if (PageUptodate(page)) {
2282 *pinned_page = page;
2283 break;
2284 }
2285 page_cache_release(page);
2286 }
2287 /*
2288 * drop cap refs first because getattr while
2289 * holding * caps refs can cause deadlock.
2290 */
2291 ceph_put_cap_refs(ci, _got);
2292 _got = 0;
2293
2294 /*
2295 * getattr request will bring inline data into
2296 * page cache
2297 */
2298 ret = __ceph_do_getattr(&ci->vfs_inode, NULL,
2299 CEPH_STAT_CAP_INLINE_DATA,
2300 true);
2301 if (ret < 0)
2302 return ret;
2303 continue;
2304 }
2305 break;
Yan, Zhengc4d4a582015-01-09 15:56:18 +08002306 }
Yan, Zheng5dda377c2015-04-30 14:40:54 +08002307
Yan, Zhengc4d4a582015-01-09 15:56:18 +08002308 *got = _got;
2309 return 0;
Sage Weila8599bd2009-10-06 11:31:12 -07002310}
2311
2312/*
2313 * Take cap refs. Caller must already know we hold at least one ref
2314 * on the caps in question or we don't know this is safe.
2315 */
2316void ceph_get_cap_refs(struct ceph_inode_info *ci, int caps)
2317{
Sage Weilbe655592011-11-30 09:47:09 -08002318 spin_lock(&ci->i_ceph_lock);
Yan, Zheng5dda377c2015-04-30 14:40:54 +08002319 __take_cap_refs(ci, caps, false);
Sage Weilbe655592011-11-30 09:47:09 -08002320 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07002321}
2322
Yan, Zheng86056092015-05-01 16:57:16 +08002323
2324/*
2325 * drop cap_snap that is not associated with any snapshot.
2326 * we don't need to send FLUSHSNAP message for it.
2327 */
2328static int ceph_try_drop_cap_snap(struct ceph_cap_snap *capsnap)
2329{
2330 if (!capsnap->need_flush &&
2331 !capsnap->writing && !capsnap->dirty_pages) {
2332
2333 dout("dropping cap_snap %p follows %llu\n",
2334 capsnap, capsnap->follows);
2335 ceph_put_snap_context(capsnap->context);
2336 list_del(&capsnap->ci_item);
2337 list_del(&capsnap->flushing_item);
2338 ceph_put_cap_snap(capsnap);
2339 return 1;
2340 }
2341 return 0;
2342}
2343
Sage Weila8599bd2009-10-06 11:31:12 -07002344/*
2345 * Release cap refs.
2346 *
2347 * If we released the last ref on any given cap, call ceph_check_caps
2348 * to release (or schedule a release).
2349 *
2350 * If we are releasing a WR cap (from a sync write), finalize any affected
2351 * cap_snap, and wake up any waiters.
2352 */
2353void ceph_put_cap_refs(struct ceph_inode_info *ci, int had)
2354{
2355 struct inode *inode = &ci->vfs_inode;
2356 int last = 0, put = 0, flushsnaps = 0, wake = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07002357
Sage Weilbe655592011-11-30 09:47:09 -08002358 spin_lock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07002359 if (had & CEPH_CAP_PIN)
2360 --ci->i_pin_ref;
2361 if (had & CEPH_CAP_FILE_RD)
2362 if (--ci->i_rd_ref == 0)
2363 last++;
2364 if (had & CEPH_CAP_FILE_CACHE)
2365 if (--ci->i_rdcache_ref == 0)
2366 last++;
2367 if (had & CEPH_CAP_FILE_BUFFER) {
Henry C Changd3d07202011-05-11 10:29:54 +00002368 if (--ci->i_wb_ref == 0) {
Sage Weila8599bd2009-10-06 11:31:12 -07002369 last++;
2370 put++;
2371 }
Henry C Changd3d07202011-05-11 10:29:54 +00002372 dout("put_cap_refs %p wb %d -> %d (?)\n",
2373 inode, ci->i_wb_ref+1, ci->i_wb_ref);
Sage Weila8599bd2009-10-06 11:31:12 -07002374 }
2375 if (had & CEPH_CAP_FILE_WR)
2376 if (--ci->i_wr_ref == 0) {
2377 last++;
Yan, Zheng86056092015-05-01 16:57:16 +08002378 if (__ceph_have_pending_cap_snap(ci)) {
2379 struct ceph_cap_snap *capsnap =
2380 list_last_entry(&ci->i_cap_snaps,
2381 struct ceph_cap_snap,
2382 ci_item);
2383 capsnap->writing = 0;
2384 if (ceph_try_drop_cap_snap(capsnap))
2385 put++;
2386 else if (__ceph_finish_cap_snap(ci, capsnap))
2387 flushsnaps = 1;
2388 wake = 1;
Sage Weila8599bd2009-10-06 11:31:12 -07002389 }
Yan, Zheng5dda377c2015-04-30 14:40:54 +08002390 if (ci->i_wrbuffer_ref_head == 0 &&
2391 ci->i_dirty_caps == 0 &&
2392 ci->i_flushing_caps == 0) {
2393 BUG_ON(!ci->i_head_snapc);
2394 ceph_put_snap_context(ci->i_head_snapc);
2395 ci->i_head_snapc = NULL;
2396 }
Yan, Zhengdb40cc12015-03-23 20:12:20 +08002397 /* see comment in __ceph_remove_cap() */
2398 if (!__ceph_is_any_caps(ci) && ci->i_snap_realm)
2399 drop_inode_snap_realm(ci);
Sage Weila8599bd2009-10-06 11:31:12 -07002400 }
Sage Weilbe655592011-11-30 09:47:09 -08002401 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07002402
Sage Weil819ccbf2010-04-01 09:33:46 -07002403 dout("put_cap_refs %p had %s%s%s\n", inode, ceph_cap_string(had),
2404 last ? " last" : "", put ? " put" : "");
Sage Weila8599bd2009-10-06 11:31:12 -07002405
2406 if (last && !flushsnaps)
2407 ceph_check_caps(ci, 0, NULL);
2408 else if (flushsnaps)
2409 ceph_flush_snaps(ci);
2410 if (wake)
Yehuda Sadeh03066f22010-07-27 13:11:08 -07002411 wake_up_all(&ci->i_cap_wq);
Yan, Zheng86056092015-05-01 16:57:16 +08002412 while (put-- > 0)
Sage Weila8599bd2009-10-06 11:31:12 -07002413 iput(inode);
2414}
2415
2416/*
2417 * Release @nr WRBUFFER refs on dirty pages for the given @snapc snap
2418 * context. Adjust per-snap dirty page accounting as appropriate.
2419 * Once all dirty data for a cap_snap is flushed, flush snapped file
2420 * metadata back to the MDS. If we dropped the last ref, call
2421 * ceph_check_caps.
2422 */
2423void ceph_put_wrbuffer_cap_refs(struct ceph_inode_info *ci, int nr,
2424 struct ceph_snap_context *snapc)
2425{
2426 struct inode *inode = &ci->vfs_inode;
2427 int last = 0;
Sage Weil819ccbf2010-04-01 09:33:46 -07002428 int complete_capsnap = 0;
2429 int drop_capsnap = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07002430 int found = 0;
2431 struct ceph_cap_snap *capsnap = NULL;
2432
Sage Weilbe655592011-11-30 09:47:09 -08002433 spin_lock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07002434 ci->i_wrbuffer_ref -= nr;
2435 last = !ci->i_wrbuffer_ref;
2436
2437 if (ci->i_head_snapc == snapc) {
2438 ci->i_wrbuffer_ref_head -= nr;
Sage Weil7d8cb262010-08-24 08:44:16 -07002439 if (ci->i_wrbuffer_ref_head == 0 &&
Yan, Zheng5dda377c2015-04-30 14:40:54 +08002440 ci->i_wr_ref == 0 &&
2441 ci->i_dirty_caps == 0 &&
2442 ci->i_flushing_caps == 0) {
Sage Weil7d8cb262010-08-24 08:44:16 -07002443 BUG_ON(!ci->i_head_snapc);
Sage Weila8599bd2009-10-06 11:31:12 -07002444 ceph_put_snap_context(ci->i_head_snapc);
2445 ci->i_head_snapc = NULL;
2446 }
2447 dout("put_wrbuffer_cap_refs on %p head %d/%d -> %d/%d %s\n",
2448 inode,
2449 ci->i_wrbuffer_ref+nr, ci->i_wrbuffer_ref_head+nr,
2450 ci->i_wrbuffer_ref, ci->i_wrbuffer_ref_head,
2451 last ? " LAST" : "");
2452 } else {
2453 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
2454 if (capsnap->context == snapc) {
2455 found = 1;
Sage Weila8599bd2009-10-06 11:31:12 -07002456 break;
2457 }
2458 }
2459 BUG_ON(!found);
Sage Weil819ccbf2010-04-01 09:33:46 -07002460 capsnap->dirty_pages -= nr;
2461 if (capsnap->dirty_pages == 0) {
2462 complete_capsnap = 1;
Yan, Zheng86056092015-05-01 16:57:16 +08002463 drop_capsnap = ceph_try_drop_cap_snap(capsnap);
Sage Weil819ccbf2010-04-01 09:33:46 -07002464 }
Sage Weila8599bd2009-10-06 11:31:12 -07002465 dout("put_wrbuffer_cap_refs on %p cap_snap %p "
Yan, Zheng86056092015-05-01 16:57:16 +08002466 " snap %lld %d/%d -> %d/%d %s%s\n",
Sage Weila8599bd2009-10-06 11:31:12 -07002467 inode, capsnap, capsnap->context->seq,
2468 ci->i_wrbuffer_ref+nr, capsnap->dirty_pages + nr,
2469 ci->i_wrbuffer_ref, capsnap->dirty_pages,
2470 last ? " (wrbuffer last)" : "",
Yan, Zheng86056092015-05-01 16:57:16 +08002471 complete_capsnap ? " (complete capsnap)" : "");
Sage Weila8599bd2009-10-06 11:31:12 -07002472 }
2473
Sage Weilbe655592011-11-30 09:47:09 -08002474 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07002475
2476 if (last) {
2477 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
2478 iput(inode);
Sage Weil819ccbf2010-04-01 09:33:46 -07002479 } else if (complete_capsnap) {
Sage Weila8599bd2009-10-06 11:31:12 -07002480 ceph_flush_snaps(ci);
Yehuda Sadeh03066f22010-07-27 13:11:08 -07002481 wake_up_all(&ci->i_cap_wq);
Sage Weila8599bd2009-10-06 11:31:12 -07002482 }
Sage Weil819ccbf2010-04-01 09:33:46 -07002483 if (drop_capsnap)
2484 iput(inode);
Sage Weila8599bd2009-10-06 11:31:12 -07002485}
2486
2487/*
Yan, Zhengca20c992013-07-21 10:07:51 +08002488 * Invalidate unlinked inode's aliases, so we can drop the inode ASAP.
2489 */
2490static void invalidate_aliases(struct inode *inode)
2491{
2492 struct dentry *dn, *prev = NULL;
2493
2494 dout("invalidate_aliases inode %p\n", inode);
2495 d_prune_aliases(inode);
2496 /*
2497 * For non-directory inode, d_find_alias() only returns
J. Bruce Fieldsfc12c802014-01-16 17:42:53 -05002498 * hashed dentry. After calling d_invalidate(), the
2499 * dentry becomes unhashed.
Yan, Zhengca20c992013-07-21 10:07:51 +08002500 *
Yan, Zhenga8d436f2013-09-02 15:19:54 +08002501 * For directory inode, d_find_alias() can return
J. Bruce Fieldsfc12c802014-01-16 17:42:53 -05002502 * unhashed dentry. But directory inode should have
Yan, Zhengca20c992013-07-21 10:07:51 +08002503 * one alias at most.
2504 */
2505 while ((dn = d_find_alias(inode))) {
2506 if (dn == prev) {
2507 dput(dn);
2508 break;
2509 }
Yan, Zhenga8d436f2013-09-02 15:19:54 +08002510 d_invalidate(dn);
Yan, Zhengca20c992013-07-21 10:07:51 +08002511 if (prev)
2512 dput(prev);
2513 prev = dn;
2514 }
2515 if (prev)
2516 dput(prev);
2517}
2518
2519/*
Sage Weila8599bd2009-10-06 11:31:12 -07002520 * Handle a cap GRANT message from the MDS. (Note that a GRANT may
2521 * actually be a revocation if it specifies a smaller cap set.)
2522 *
Sage Weilbe655592011-11-30 09:47:09 -08002523 * caller holds s_mutex and i_ceph_lock, we drop both.
Sage Weila8599bd2009-10-06 11:31:12 -07002524 */
Yan, Zheng2cd698b2014-04-18 13:20:27 +08002525static void handle_cap_grant(struct ceph_mds_client *mdsc,
2526 struct inode *inode, struct ceph_mds_caps *grant,
Yan, Zhengfb01d1f2014-11-14 21:29:55 +08002527 u64 inline_version,
2528 void *inline_data, int inline_len,
Yan, Zheng2cd698b2014-04-18 13:20:27 +08002529 struct ceph_buffer *xattr_buf,
Sage Weil15637c82010-03-16 13:42:00 -07002530 struct ceph_mds_session *session,
Yan, Zheng2cd698b2014-04-18 13:20:27 +08002531 struct ceph_cap *cap, int issued)
2532 __releases(ci->i_ceph_lock)
Yan, Zheng982d6012014-12-23 15:30:54 +08002533 __releases(mdsc->snap_rwsem)
Sage Weila8599bd2009-10-06 11:31:12 -07002534{
2535 struct ceph_inode_info *ci = ceph_inode(inode);
2536 int mds = session->s_mds;
Sage Weil2f56f562010-10-27 20:59:49 -07002537 int seq = le32_to_cpu(grant->seq);
Sage Weila8599bd2009-10-06 11:31:12 -07002538 int newcaps = le32_to_cpu(grant->caps);
Yan, Zheng2cd698b2014-04-18 13:20:27 +08002539 int used, wanted, dirty;
Sage Weila8599bd2009-10-06 11:31:12 -07002540 u64 size = le64_to_cpu(grant->size);
2541 u64 max_size = le64_to_cpu(grant->max_size);
2542 struct timespec mtime, atime, ctime;
Sage Weil15637c82010-03-16 13:42:00 -07002543 int check_caps = 0;
Fabian Frederickab6c2c32014-10-09 23:16:35 +02002544 bool wake = false;
2545 bool writeback = false;
2546 bool queue_trunc = false;
2547 bool queue_invalidate = false;
2548 bool queue_revalidate = false;
2549 bool deleted_inode = false;
Yan, Zheng31c542a2014-11-14 21:41:55 +08002550 bool fill_inline = false;
Sage Weila8599bd2009-10-06 11:31:12 -07002551
Sage Weil2f56f562010-10-27 20:59:49 -07002552 dout("handle_cap_grant inode %p cap %p mds%d seq %d %s\n",
2553 inode, cap, mds, seq, ceph_cap_string(newcaps));
Sage Weila8599bd2009-10-06 11:31:12 -07002554 dout(" size %llu max_size %llu, i_size %llu\n", size, max_size,
2555 inode->i_size);
2556
Yan, Zheng11df2df2013-11-24 14:44:38 +08002557
2558 /*
2559 * auth mds of the inode changed. we received the cap export message,
2560 * but still haven't received the cap import message. handle_cap_export
2561 * updated the new auth MDS' cap.
2562 *
2563 * "ceph_seq_cmp(seq, cap->seq) <= 0" means we are processing a message
2564 * that was sent before the cap import message. So don't remove caps.
2565 */
2566 if (ceph_seq_cmp(seq, cap->seq) <= 0) {
2567 WARN_ON(cap != ci->i_auth_cap);
2568 WARN_ON(cap->cap_id != le64_to_cpu(grant->cap_id));
2569 seq = cap->seq;
2570 newcaps |= cap->issued;
2571 }
2572
Sage Weila8599bd2009-10-06 11:31:12 -07002573 /*
2574 * If CACHE is being revoked, and we have no dirty buffers,
2575 * try to invalidate (once). (If there are dirty buffers, we
2576 * will invalidate _after_ writeback.)
2577 */
Sage Weil3b454c42010-06-10 13:20:33 -07002578 if (((cap->issued & ~newcaps) & CEPH_CAP_FILE_CACHE) &&
2579 (newcaps & CEPH_CAP_FILE_LAZYIO) == 0 &&
Yehuda Sadehbcd2cbd2010-02-19 00:12:21 +00002580 !ci->i_wrbuffer_ref) {
Li Wange9075742013-08-15 22:00:25 -07002581 if (try_nonblocking_invalidate(inode)) {
Sage Weila8599bd2009-10-06 11:31:12 -07002582 /* there were locked pages.. invalidate later
2583 in a separate thread. */
2584 if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
Fabian Frederickab6c2c32014-10-09 23:16:35 +02002585 queue_invalidate = true;
Sage Weila8599bd2009-10-06 11:31:12 -07002586 ci->i_rdcache_revoking = ci->i_rdcache_gen;
2587 }
Sage Weila8599bd2009-10-06 11:31:12 -07002588 }
Milosz Tanski99ccbd22013-08-21 17:29:54 -04002589
2590 ceph_fscache_invalidate(inode);
Sage Weila8599bd2009-10-06 11:31:12 -07002591 }
2592
2593 /* side effects now are allowed */
Sage Weil685f9a5d2009-11-09 12:05:48 -08002594 cap->cap_gen = session->s_cap_gen;
Yan, Zheng11df2df2013-11-24 14:44:38 +08002595 cap->seq = seq;
Sage Weila8599bd2009-10-06 11:31:12 -07002596
2597 __check_cap_issue(ci, cap, newcaps);
2598
Yan, Zhengf98a1282014-04-17 08:55:50 +08002599 if ((newcaps & CEPH_CAP_AUTH_SHARED) &&
2600 (issued & CEPH_CAP_AUTH_EXCL) == 0) {
Sage Weila8599bd2009-10-06 11:31:12 -07002601 inode->i_mode = le32_to_cpu(grant->mode);
Eric W. Biederman05cb11c2013-01-31 02:56:19 -08002602 inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(grant->uid));
2603 inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(grant->gid));
Sage Weila8599bd2009-10-06 11:31:12 -07002604 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
Eric W. Biedermanbd2bae62013-01-31 04:05:39 -08002605 from_kuid(&init_user_ns, inode->i_uid),
2606 from_kgid(&init_user_ns, inode->i_gid));
Sage Weila8599bd2009-10-06 11:31:12 -07002607 }
2608
Yan, Zhengf98a1282014-04-17 08:55:50 +08002609 if ((newcaps & CEPH_CAP_AUTH_SHARED) &&
2610 (issued & CEPH_CAP_LINK_EXCL) == 0) {
Miklos Szeredibfe86842011-10-28 14:13:29 +02002611 set_nlink(inode, le32_to_cpu(grant->nlink));
Yan, Zhengca20c992013-07-21 10:07:51 +08002612 if (inode->i_nlink == 0 &&
2613 (newcaps & (CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL)))
Fabian Frederickab6c2c32014-10-09 23:16:35 +02002614 deleted_inode = true;
Yan, Zhengca20c992013-07-21 10:07:51 +08002615 }
Sage Weila8599bd2009-10-06 11:31:12 -07002616
2617 if ((issued & CEPH_CAP_XATTR_EXCL) == 0 && grant->xattr_len) {
2618 int len = le32_to_cpu(grant->xattr_len);
2619 u64 version = le64_to_cpu(grant->xattr_version);
2620
2621 if (version > ci->i_xattrs.version) {
2622 dout(" got new xattrs v%llu on %p len %d\n",
2623 version, inode, len);
2624 if (ci->i_xattrs.blob)
2625 ceph_buffer_put(ci->i_xattrs.blob);
2626 ci->i_xattrs.blob = ceph_buffer_get(xattr_buf);
2627 ci->i_xattrs.version = version;
Guangliang Zhao7221fe42013-11-11 15:18:03 +08002628 ceph_forget_all_cached_acls(inode);
Sage Weila8599bd2009-10-06 11:31:12 -07002629 }
2630 }
2631
Milosz Tanski99ccbd22013-08-21 17:29:54 -04002632 /* Do we need to revalidate our fscache cookie. Don't bother on the
2633 * first cache cap as we already validate at cookie creation time. */
2634 if ((issued & CEPH_CAP_FILE_CACHE) && ci->i_rdcache_gen > 1)
Fabian Frederickab6c2c32014-10-09 23:16:35 +02002635 queue_revalidate = true;
Milosz Tanski99ccbd22013-08-21 17:29:54 -04002636
Yan, Zhengf98a1282014-04-17 08:55:50 +08002637 if (newcaps & CEPH_CAP_ANY_RD) {
2638 /* ctime/mtime/atime? */
2639 ceph_decode_timespec(&mtime, &grant->mtime);
2640 ceph_decode_timespec(&atime, &grant->atime);
2641 ceph_decode_timespec(&ctime, &grant->ctime);
2642 ceph_fill_file_time(inode, issued,
2643 le32_to_cpu(grant->time_warp_seq),
2644 &ctime, &mtime, &atime);
2645 }
Sage Weila8599bd2009-10-06 11:31:12 -07002646
Yan, Zhengf98a1282014-04-17 08:55:50 +08002647 if (newcaps & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR)) {
2648 /* file layout may have changed */
2649 ci->i_layout = grant->layout;
2650 /* size/truncate_seq? */
2651 queue_trunc = ceph_fill_file_size(inode, issued,
2652 le32_to_cpu(grant->truncate_seq),
2653 le64_to_cpu(grant->truncate_size),
2654 size);
2655 /* max size increase? */
2656 if (ci->i_auth_cap == cap && max_size != ci->i_max_size) {
2657 dout("max_size %lld -> %llu\n",
2658 ci->i_max_size, max_size);
2659 ci->i_max_size = max_size;
2660 if (max_size >= ci->i_wanted_max_size) {
2661 ci->i_wanted_max_size = 0; /* reset */
2662 ci->i_requested_max_size = 0;
2663 }
Fabian Frederickab6c2c32014-10-09 23:16:35 +02002664 wake = true;
Sage Weila8599bd2009-10-06 11:31:12 -07002665 }
Sage Weila8599bd2009-10-06 11:31:12 -07002666 }
2667
2668 /* check cap bits */
2669 wanted = __ceph_caps_wanted(ci);
2670 used = __ceph_caps_used(ci);
2671 dirty = __ceph_caps_dirty(ci);
2672 dout(" my wanted = %s, used = %s, dirty %s\n",
2673 ceph_cap_string(wanted),
2674 ceph_cap_string(used),
2675 ceph_cap_string(dirty));
2676 if (wanted != le32_to_cpu(grant->wanted)) {
2677 dout("mds wanted %s -> %s\n",
2678 ceph_cap_string(le32_to_cpu(grant->wanted)),
2679 ceph_cap_string(wanted));
Yan, Zheng390306c2013-01-04 15:30:10 +08002680 /* imported cap may not have correct mds_wanted */
2681 if (le32_to_cpu(grant->op) == CEPH_CAP_OP_IMPORT)
2682 check_caps = 1;
Sage Weila8599bd2009-10-06 11:31:12 -07002683 }
2684
Sage Weila8599bd2009-10-06 11:31:12 -07002685 /* revocation, grant, or no-op? */
2686 if (cap->issued & ~newcaps) {
Sage Weil3b454c42010-06-10 13:20:33 -07002687 int revoking = cap->issued & ~newcaps;
2688
2689 dout("revocation: %s -> %s (revoking %s)\n",
2690 ceph_cap_string(cap->issued),
2691 ceph_cap_string(newcaps),
2692 ceph_cap_string(revoking));
Sage Weil0eb6cd42010-08-05 13:53:18 -07002693 if (revoking & used & CEPH_CAP_FILE_BUFFER)
Fabian Frederickab6c2c32014-10-09 23:16:35 +02002694 writeback = true; /* initiate writeback; will delay ack */
Sage Weil3b454c42010-06-10 13:20:33 -07002695 else if (revoking == CEPH_CAP_FILE_CACHE &&
2696 (newcaps & CEPH_CAP_FILE_LAZYIO) == 0 &&
2697 queue_invalidate)
2698 ; /* do nothing yet, invalidation will be queued */
2699 else if (cap == ci->i_auth_cap)
2700 check_caps = 1; /* check auth cap only */
2701 else
2702 check_caps = 2; /* check all caps */
Sage Weila8599bd2009-10-06 11:31:12 -07002703 cap->issued = newcaps;
Sage Weil978097c2010-03-08 15:27:53 -08002704 cap->implemented |= newcaps;
Sage Weila8599bd2009-10-06 11:31:12 -07002705 } else if (cap->issued == newcaps) {
2706 dout("caps unchanged: %s -> %s\n",
2707 ceph_cap_string(cap->issued), ceph_cap_string(newcaps));
2708 } else {
2709 dout("grant: %s -> %s\n", ceph_cap_string(cap->issued),
2710 ceph_cap_string(newcaps));
Yan, Zheng6ee6b9532013-07-02 12:40:21 +08002711 /* non-auth MDS is revoking the newly grant caps ? */
2712 if (cap == ci->i_auth_cap &&
2713 __ceph_caps_revoking_other(ci, cap, newcaps))
2714 check_caps = 2;
2715
Sage Weila8599bd2009-10-06 11:31:12 -07002716 cap->issued = newcaps;
2717 cap->implemented |= newcaps; /* add bits only, to
2718 * avoid stepping on a
2719 * pending revocation */
Fabian Frederickab6c2c32014-10-09 23:16:35 +02002720 wake = true;
Sage Weila8599bd2009-10-06 11:31:12 -07002721 }
Sage Weil978097c2010-03-08 15:27:53 -08002722 BUG_ON(cap->issued & ~cap->implemented);
Sage Weila8599bd2009-10-06 11:31:12 -07002723
Yan, Zheng31c542a2014-11-14 21:41:55 +08002724 if (inline_version > 0 && inline_version >= ci->i_inline_version) {
2725 ci->i_inline_version = inline_version;
2726 if (ci->i_inline_version != CEPH_INLINE_NONE &&
2727 (newcaps & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)))
2728 fill_inline = true;
2729 }
2730
Sage Weilbe655592011-11-30 09:47:09 -08002731 spin_unlock(&ci->i_ceph_lock);
Milosz Tanski99ccbd22013-08-21 17:29:54 -04002732
Yan, Zheng2cd698b2014-04-18 13:20:27 +08002733 if (le32_to_cpu(grant->op) == CEPH_CAP_OP_IMPORT) {
Yan, Zheng2cd698b2014-04-18 13:20:27 +08002734 kick_flushing_inode_caps(mdsc, session, inode);
2735 up_read(&mdsc->snap_rwsem);
2736 if (newcaps & ~issued)
Fabian Frederickab6c2c32014-10-09 23:16:35 +02002737 wake = true;
Yan, Zheng2cd698b2014-04-18 13:20:27 +08002738 }
2739
Yan, Zheng31c542a2014-11-14 21:41:55 +08002740 if (fill_inline)
2741 ceph_fill_inline_data(inode, NULL, inline_data, inline_len);
2742
Yan, Zhengc6bcda6f2014-04-11 10:18:07 +08002743 if (queue_trunc) {
2744 ceph_queue_vmtruncate(inode);
2745 ceph_queue_revalidate(inode);
2746 } else if (queue_revalidate)
2747 ceph_queue_revalidate(inode);
2748
Sage Weil3c6f6b792010-02-09 15:24:44 -08002749 if (writeback)
Sage Weila8599bd2009-10-06 11:31:12 -07002750 /*
2751 * queue inode for writeback: we can't actually call
2752 * filemap_write_and_wait, etc. from message handler
2753 * context.
2754 */
Sage Weil3c6f6b792010-02-09 15:24:44 -08002755 ceph_queue_writeback(inode);
2756 if (queue_invalidate)
2757 ceph_queue_invalidate(inode);
Yan, Zhengca20c992013-07-21 10:07:51 +08002758 if (deleted_inode)
2759 invalidate_aliases(inode);
Sage Weila8599bd2009-10-06 11:31:12 -07002760 if (wake)
Yehuda Sadeh03066f22010-07-27 13:11:08 -07002761 wake_up_all(&ci->i_cap_wq);
Sage Weil15637c82010-03-16 13:42:00 -07002762
2763 if (check_caps == 1)
2764 ceph_check_caps(ci, CHECK_CAPS_NODELAY|CHECK_CAPS_AUTHONLY,
2765 session);
2766 else if (check_caps == 2)
2767 ceph_check_caps(ci, CHECK_CAPS_NODELAY, session);
2768 else
2769 mutex_unlock(&session->s_mutex);
Sage Weila8599bd2009-10-06 11:31:12 -07002770}
2771
2772/*
2773 * Handle FLUSH_ACK from MDS, indicating that metadata we sent to the
2774 * MDS has been safely committed.
2775 */
Sage Weil6df058c2009-12-22 11:24:33 -08002776static void handle_cap_flush_ack(struct inode *inode, u64 flush_tid,
Sage Weila8599bd2009-10-06 11:31:12 -07002777 struct ceph_mds_caps *m,
2778 struct ceph_mds_session *session,
2779 struct ceph_cap *cap)
Sage Weilbe655592011-11-30 09:47:09 -08002780 __releases(ci->i_ceph_lock)
Sage Weila8599bd2009-10-06 11:31:12 -07002781{
2782 struct ceph_inode_info *ci = ceph_inode(inode);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002783 struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
Sage Weila8599bd2009-10-06 11:31:12 -07002784 unsigned seq = le32_to_cpu(m->seq);
2785 int dirty = le32_to_cpu(m->dirty);
2786 int cleaned = 0;
Sage Weilafcdaea2009-10-14 14:27:38 -07002787 int drop = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07002788 int i;
2789
2790 for (i = 0; i < CEPH_CAP_BITS; i++)
2791 if ((dirty & (1 << i)) &&
Yan, Zheng32313002014-10-22 17:13:26 -07002792 (u16)flush_tid == ci->i_cap_flush_tid[i])
Sage Weila8599bd2009-10-06 11:31:12 -07002793 cleaned |= 1 << i;
2794
2795 dout("handle_cap_flush_ack inode %p mds%d seq %d on %s cleaned %s,"
2796 " flushing %s -> %s\n",
2797 inode, session->s_mds, seq, ceph_cap_string(dirty),
2798 ceph_cap_string(cleaned), ceph_cap_string(ci->i_flushing_caps),
2799 ceph_cap_string(ci->i_flushing_caps & ~cleaned));
2800
2801 if (ci->i_flushing_caps == (ci->i_flushing_caps & ~cleaned))
2802 goto out;
2803
Sage Weila8599bd2009-10-06 11:31:12 -07002804 ci->i_flushing_caps &= ~cleaned;
Sage Weila8599bd2009-10-06 11:31:12 -07002805
2806 spin_lock(&mdsc->cap_dirty_lock);
2807 if (ci->i_flushing_caps == 0) {
2808 list_del_init(&ci->i_flushing_item);
2809 if (!list_empty(&session->s_cap_flushing))
2810 dout(" mds%d still flushing cap on %p\n",
2811 session->s_mds,
2812 &list_entry(session->s_cap_flushing.next,
2813 struct ceph_inode_info,
2814 i_flushing_item)->vfs_inode);
2815 mdsc->num_cap_flushing--;
Yehuda Sadeh03066f22010-07-27 13:11:08 -07002816 wake_up_all(&mdsc->cap_flushing_wq);
Sage Weila8599bd2009-10-06 11:31:12 -07002817 dout(" inode %p now !flushing\n", inode);
Sage Weilafcdaea2009-10-14 14:27:38 -07002818
2819 if (ci->i_dirty_caps == 0) {
2820 dout(" inode %p now clean\n", inode);
2821 BUG_ON(!list_empty(&ci->i_dirty_item));
2822 drop = 1;
Yan, Zheng5dda377c2015-04-30 14:40:54 +08002823 if (ci->i_wr_ref == 0 &&
2824 ci->i_wrbuffer_ref_head == 0) {
Sage Weil7d8cb262010-08-24 08:44:16 -07002825 BUG_ON(!ci->i_head_snapc);
2826 ceph_put_snap_context(ci->i_head_snapc);
2827 ci->i_head_snapc = NULL;
2828 }
Sage Weil76e3b392009-10-15 18:13:53 -07002829 } else {
2830 BUG_ON(list_empty(&ci->i_dirty_item));
Sage Weilafcdaea2009-10-14 14:27:38 -07002831 }
Sage Weila8599bd2009-10-06 11:31:12 -07002832 }
2833 spin_unlock(&mdsc->cap_dirty_lock);
Yehuda Sadeh03066f22010-07-27 13:11:08 -07002834 wake_up_all(&ci->i_cap_wq);
Sage Weila8599bd2009-10-06 11:31:12 -07002835
2836out:
Sage Weilbe655592011-11-30 09:47:09 -08002837 spin_unlock(&ci->i_ceph_lock);
Sage Weilafcdaea2009-10-14 14:27:38 -07002838 if (drop)
Sage Weila8599bd2009-10-06 11:31:12 -07002839 iput(inode);
2840}
2841
2842/*
2843 * Handle FLUSHSNAP_ACK. MDS has flushed snap data to disk and we can
2844 * throw away our cap_snap.
2845 *
2846 * Caller hold s_mutex.
2847 */
Sage Weil6df058c2009-12-22 11:24:33 -08002848static void handle_cap_flushsnap_ack(struct inode *inode, u64 flush_tid,
Sage Weila8599bd2009-10-06 11:31:12 -07002849 struct ceph_mds_caps *m,
2850 struct ceph_mds_session *session)
2851{
2852 struct ceph_inode_info *ci = ceph_inode(inode);
Yan, Zhengaffbc192015-05-05 21:22:13 +08002853 struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
Sage Weila8599bd2009-10-06 11:31:12 -07002854 u64 follows = le64_to_cpu(m->snap_follows);
Sage Weila8599bd2009-10-06 11:31:12 -07002855 struct ceph_cap_snap *capsnap;
2856 int drop = 0;
2857
2858 dout("handle_cap_flushsnap_ack inode %p ci %p mds%d follows %lld\n",
2859 inode, ci, session->s_mds, follows);
2860
Sage Weilbe655592011-11-30 09:47:09 -08002861 spin_lock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07002862 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
2863 if (capsnap->follows == follows) {
2864 if (capsnap->flush_tid != flush_tid) {
2865 dout(" cap_snap %p follows %lld tid %lld !="
2866 " %lld\n", capsnap, follows,
2867 flush_tid, capsnap->flush_tid);
2868 break;
2869 }
2870 WARN_ON(capsnap->dirty_pages || capsnap->writing);
Sage Weil819ccbf2010-04-01 09:33:46 -07002871 dout(" removing %p cap_snap %p follows %lld\n",
2872 inode, capsnap, follows);
Sage Weila8599bd2009-10-06 11:31:12 -07002873 ceph_put_snap_context(capsnap->context);
2874 list_del(&capsnap->ci_item);
2875 list_del(&capsnap->flushing_item);
2876 ceph_put_cap_snap(capsnap);
Yan, Zhengaffbc192015-05-05 21:22:13 +08002877 wake_up_all(&mdsc->cap_flushing_wq);
Sage Weila8599bd2009-10-06 11:31:12 -07002878 drop = 1;
2879 break;
2880 } else {
2881 dout(" skipping cap_snap %p follows %lld\n",
2882 capsnap, capsnap->follows);
2883 }
2884 }
Sage Weilbe655592011-11-30 09:47:09 -08002885 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07002886 if (drop)
2887 iput(inode);
2888}
2889
2890/*
2891 * Handle TRUNC from MDS, indicating file truncation.
2892 *
2893 * caller hold s_mutex.
2894 */
2895static void handle_cap_trunc(struct inode *inode,
2896 struct ceph_mds_caps *trunc,
2897 struct ceph_mds_session *session)
Sage Weilbe655592011-11-30 09:47:09 -08002898 __releases(ci->i_ceph_lock)
Sage Weila8599bd2009-10-06 11:31:12 -07002899{
2900 struct ceph_inode_info *ci = ceph_inode(inode);
2901 int mds = session->s_mds;
2902 int seq = le32_to_cpu(trunc->seq);
2903 u32 truncate_seq = le32_to_cpu(trunc->truncate_seq);
2904 u64 truncate_size = le64_to_cpu(trunc->truncate_size);
2905 u64 size = le64_to_cpu(trunc->size);
2906 int implemented = 0;
2907 int dirty = __ceph_caps_dirty(ci);
2908 int issued = __ceph_caps_issued(ceph_inode(inode), &implemented);
2909 int queue_trunc = 0;
2910
2911 issued |= implemented | dirty;
2912
2913 dout("handle_cap_trunc inode %p mds%d seq %d to %lld seq %d\n",
2914 inode, mds, seq, truncate_size, truncate_seq);
2915 queue_trunc = ceph_fill_file_size(inode, issued,
2916 truncate_seq, truncate_size, size);
Sage Weilbe655592011-11-30 09:47:09 -08002917 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07002918
Milosz Tanski99ccbd22013-08-21 17:29:54 -04002919 if (queue_trunc) {
Sage Weil3c6f6b792010-02-09 15:24:44 -08002920 ceph_queue_vmtruncate(inode);
Milosz Tanski99ccbd22013-08-21 17:29:54 -04002921 ceph_fscache_invalidate(inode);
2922 }
Sage Weila8599bd2009-10-06 11:31:12 -07002923}
2924
2925/*
2926 * Handle EXPORT from MDS. Cap is being migrated _from_ this mds to a
2927 * different one. If we are the most recent migration we've seen (as
2928 * indicated by mseq), make note of the migrating cap bits for the
2929 * duration (until we see the corresponding IMPORT).
2930 *
2931 * caller holds s_mutex
2932 */
2933static void handle_cap_export(struct inode *inode, struct ceph_mds_caps *ex,
Yan, Zheng11df2df2013-11-24 14:44:38 +08002934 struct ceph_mds_cap_peer *ph,
2935 struct ceph_mds_session *session)
Sage Weila8599bd2009-10-06 11:31:12 -07002936{
Sage Weildb354052011-05-24 11:46:31 -07002937 struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
Yan, Zheng11df2df2013-11-24 14:44:38 +08002938 struct ceph_mds_session *tsession = NULL;
Yan, Zhengd9df2782014-04-18 09:57:11 +08002939 struct ceph_cap *cap, *tcap, *new_cap = NULL;
Sage Weila8599bd2009-10-06 11:31:12 -07002940 struct ceph_inode_info *ci = ceph_inode(inode);
Yan, Zheng11df2df2013-11-24 14:44:38 +08002941 u64 t_cap_id;
Sage Weila8599bd2009-10-06 11:31:12 -07002942 unsigned mseq = le32_to_cpu(ex->migrate_seq);
Yan, Zheng11df2df2013-11-24 14:44:38 +08002943 unsigned t_seq, t_mseq;
2944 int target, issued;
2945 int mds = session->s_mds;
Sage Weila8599bd2009-10-06 11:31:12 -07002946
Yan, Zheng11df2df2013-11-24 14:44:38 +08002947 if (ph) {
2948 t_cap_id = le64_to_cpu(ph->cap_id);
2949 t_seq = le32_to_cpu(ph->seq);
2950 t_mseq = le32_to_cpu(ph->mseq);
2951 target = le32_to_cpu(ph->mds);
2952 } else {
2953 t_cap_id = t_seq = t_mseq = 0;
2954 target = -1;
Sage Weila8599bd2009-10-06 11:31:12 -07002955 }
2956
Yan, Zheng11df2df2013-11-24 14:44:38 +08002957 dout("handle_cap_export inode %p ci %p mds%d mseq %d target %d\n",
2958 inode, ci, mds, mseq, target);
2959retry:
2960 spin_lock(&ci->i_ceph_lock);
2961 cap = __get_cap_for_mds(ci, mds);
Yan, Zhengca665e02014-04-21 15:46:37 +08002962 if (!cap || cap->cap_id != le64_to_cpu(ex->cap_id))
Yan, Zheng11df2df2013-11-24 14:44:38 +08002963 goto out_unlock;
Sage Weil154f42c2010-06-21 13:45:04 -07002964
Yan, Zheng11df2df2013-11-24 14:44:38 +08002965 if (target < 0) {
2966 __ceph_remove_cap(cap, false);
2967 goto out_unlock;
2968 }
Sage Weildb354052011-05-24 11:46:31 -07002969
Yan, Zheng11df2df2013-11-24 14:44:38 +08002970 /*
2971 * now we know we haven't received the cap import message yet
2972 * because the exported cap still exist.
2973 */
2974
2975 issued = cap->issued;
2976 WARN_ON(issued != cap->implemented);
2977
2978 tcap = __get_cap_for_mds(ci, target);
2979 if (tcap) {
2980 /* already have caps from the target */
2981 if (tcap->cap_id != t_cap_id ||
2982 ceph_seq_cmp(tcap->seq, t_seq) < 0) {
2983 dout(" updating import cap %p mds%d\n", tcap, target);
2984 tcap->cap_id = t_cap_id;
2985 tcap->seq = t_seq - 1;
2986 tcap->issue_seq = t_seq - 1;
2987 tcap->mseq = t_mseq;
2988 tcap->issued |= issued;
2989 tcap->implemented |= issued;
2990 if (cap == ci->i_auth_cap)
2991 ci->i_auth_cap = tcap;
2992 if (ci->i_flushing_caps && ci->i_auth_cap == tcap) {
2993 spin_lock(&mdsc->cap_dirty_lock);
2994 list_move_tail(&ci->i_flushing_item,
2995 &tcap->session->s_cap_flushing);
2996 spin_unlock(&mdsc->cap_dirty_lock);
Sage Weildb354052011-05-24 11:46:31 -07002997 }
Sage Weila8599bd2009-10-06 11:31:12 -07002998 }
Yan, Zhenga096b092013-09-22 10:15:58 +08002999 __ceph_remove_cap(cap, false);
Yan, Zheng11df2df2013-11-24 14:44:38 +08003000 goto out_unlock;
Yan, Zhengd9df2782014-04-18 09:57:11 +08003001 } else if (tsession) {
Yan, Zheng11df2df2013-11-24 14:44:38 +08003002 /* add placeholder for the export tagert */
Yan, Zhengd9df2782014-04-18 09:57:11 +08003003 int flag = (cap == ci->i_auth_cap) ? CEPH_CAP_FLAG_AUTH : 0;
Yan, Zheng11df2df2013-11-24 14:44:38 +08003004 ceph_add_cap(inode, tsession, t_cap_id, -1, issued, 0,
Yan, Zhengd9df2782014-04-18 09:57:11 +08003005 t_seq - 1, t_mseq, (u64)-1, flag, &new_cap);
3006
3007 __ceph_remove_cap(cap, false);
3008 goto out_unlock;
Yan, Zheng11df2df2013-11-24 14:44:38 +08003009 }
Sage Weila8599bd2009-10-06 11:31:12 -07003010
Sage Weilbe655592011-11-30 09:47:09 -08003011 spin_unlock(&ci->i_ceph_lock);
Yan, Zheng11df2df2013-11-24 14:44:38 +08003012 mutex_unlock(&session->s_mutex);
3013
3014 /* open target session */
3015 tsession = ceph_mdsc_open_export_target_session(mdsc, target);
3016 if (!IS_ERR(tsession)) {
3017 if (mds > target) {
3018 mutex_lock(&session->s_mutex);
3019 mutex_lock_nested(&tsession->s_mutex,
3020 SINGLE_DEPTH_NESTING);
3021 } else {
3022 mutex_lock(&tsession->s_mutex);
3023 mutex_lock_nested(&session->s_mutex,
3024 SINGLE_DEPTH_NESTING);
3025 }
Yan, Zhengd9df2782014-04-18 09:57:11 +08003026 new_cap = ceph_get_cap(mdsc, NULL);
Yan, Zheng11df2df2013-11-24 14:44:38 +08003027 } else {
3028 WARN_ON(1);
3029 tsession = NULL;
3030 target = -1;
3031 }
3032 goto retry;
3033
3034out_unlock:
3035 spin_unlock(&ci->i_ceph_lock);
3036 mutex_unlock(&session->s_mutex);
3037 if (tsession) {
3038 mutex_unlock(&tsession->s_mutex);
3039 ceph_put_mds_session(tsession);
3040 }
Yan, Zhengd9df2782014-04-18 09:57:11 +08003041 if (new_cap)
3042 ceph_put_cap(mdsc, new_cap);
Sage Weila8599bd2009-10-06 11:31:12 -07003043}
3044
3045/*
Yan, Zheng2cd698b2014-04-18 13:20:27 +08003046 * Handle cap IMPORT.
Sage Weila8599bd2009-10-06 11:31:12 -07003047 *
Yan, Zheng2cd698b2014-04-18 13:20:27 +08003048 * caller holds s_mutex. acquires i_ceph_lock
Sage Weila8599bd2009-10-06 11:31:12 -07003049 */
3050static void handle_cap_import(struct ceph_mds_client *mdsc,
3051 struct inode *inode, struct ceph_mds_caps *im,
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003052 struct ceph_mds_cap_peer *ph,
Sage Weila8599bd2009-10-06 11:31:12 -07003053 struct ceph_mds_session *session,
Yan, Zheng2cd698b2014-04-18 13:20:27 +08003054 struct ceph_cap **target_cap, int *old_issued)
3055 __acquires(ci->i_ceph_lock)
Sage Weila8599bd2009-10-06 11:31:12 -07003056{
3057 struct ceph_inode_info *ci = ceph_inode(inode);
Yan, Zheng2cd698b2014-04-18 13:20:27 +08003058 struct ceph_cap *cap, *ocap, *new_cap = NULL;
Sage Weila8599bd2009-10-06 11:31:12 -07003059 int mds = session->s_mds;
Yan, Zheng2cd698b2014-04-18 13:20:27 +08003060 int issued;
3061 unsigned caps = le32_to_cpu(im->caps);
Sage Weila8599bd2009-10-06 11:31:12 -07003062 unsigned wanted = le32_to_cpu(im->wanted);
3063 unsigned seq = le32_to_cpu(im->seq);
3064 unsigned mseq = le32_to_cpu(im->migrate_seq);
3065 u64 realmino = le64_to_cpu(im->realm);
3066 u64 cap_id = le64_to_cpu(im->cap_id);
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003067 u64 p_cap_id;
3068 int peer;
Sage Weila8599bd2009-10-06 11:31:12 -07003069
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003070 if (ph) {
3071 p_cap_id = le64_to_cpu(ph->cap_id);
3072 peer = le32_to_cpu(ph->mds);
Sage Weila8599bd2009-10-06 11:31:12 -07003073 } else {
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003074 p_cap_id = 0;
3075 peer = -1;
Sage Weila8599bd2009-10-06 11:31:12 -07003076 }
3077
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003078 dout("handle_cap_import inode %p ci %p mds%d mseq %d peer %d\n",
3079 inode, ci, mds, mseq, peer);
3080
Yan, Zhengd9df2782014-04-18 09:57:11 +08003081retry:
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003082 spin_lock(&ci->i_ceph_lock);
Yan, Zhengd9df2782014-04-18 09:57:11 +08003083 cap = __get_cap_for_mds(ci, mds);
3084 if (!cap) {
3085 if (!new_cap) {
3086 spin_unlock(&ci->i_ceph_lock);
3087 new_cap = ceph_get_cap(mdsc, NULL);
3088 goto retry;
3089 }
Yan, Zheng2cd698b2014-04-18 13:20:27 +08003090 cap = new_cap;
3091 } else {
3092 if (new_cap) {
3093 ceph_put_cap(mdsc, new_cap);
3094 new_cap = NULL;
3095 }
Yan, Zhengd9df2782014-04-18 09:57:11 +08003096 }
3097
Yan, Zheng2cd698b2014-04-18 13:20:27 +08003098 __ceph_caps_issued(ci, &issued);
3099 issued |= __ceph_caps_dirty(ci);
3100
3101 ceph_add_cap(inode, session, cap_id, -1, caps, wanted, seq, mseq,
Yan, Zhengd9df2782014-04-18 09:57:11 +08003102 realmino, CEPH_CAP_FLAG_AUTH, &new_cap);
3103
Yan, Zheng2cd698b2014-04-18 13:20:27 +08003104 ocap = peer >= 0 ? __get_cap_for_mds(ci, peer) : NULL;
3105 if (ocap && ocap->cap_id == p_cap_id) {
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003106 dout(" remove export cap %p mds%d flags %d\n",
Yan, Zheng2cd698b2014-04-18 13:20:27 +08003107 ocap, peer, ph->flags);
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003108 if ((ph->flags & CEPH_CAP_FLAG_AUTH) &&
Yan, Zheng2cd698b2014-04-18 13:20:27 +08003109 (ocap->seq != le32_to_cpu(ph->seq) ||
3110 ocap->mseq != le32_to_cpu(ph->mseq))) {
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003111 pr_err("handle_cap_import: mismatched seq/mseq: "
3112 "ino (%llx.%llx) mds%d seq %d mseq %d "
3113 "importer mds%d has peer seq %d mseq %d\n",
Yan, Zheng2cd698b2014-04-18 13:20:27 +08003114 ceph_vinop(inode), peer, ocap->seq,
3115 ocap->mseq, mds, le32_to_cpu(ph->seq),
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003116 le32_to_cpu(ph->mseq));
3117 }
Yan, Zheng2cd698b2014-04-18 13:20:27 +08003118 __ceph_remove_cap(ocap, (ph->flags & CEPH_CAP_FLAG_RELEASE));
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003119 }
3120
3121 /* make sure we re-request max_size, if necessary */
3122 ci->i_wanted_max_size = 0;
3123 ci->i_requested_max_size = 0;
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003124
Yan, Zheng2cd698b2014-04-18 13:20:27 +08003125 *old_issued = issued;
3126 *target_cap = cap;
Sage Weila8599bd2009-10-06 11:31:12 -07003127}
3128
3129/*
3130 * Handle a caps message from the MDS.
3131 *
3132 * Identify the appropriate session, inode, and call the right handler
3133 * based on the cap op.
3134 */
3135void ceph_handle_caps(struct ceph_mds_session *session,
3136 struct ceph_msg *msg)
3137{
3138 struct ceph_mds_client *mdsc = session->s_mdsc;
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07003139 struct super_block *sb = mdsc->fsc->sb;
Sage Weila8599bd2009-10-06 11:31:12 -07003140 struct inode *inode;
Sage Weilbe655592011-11-30 09:47:09 -08003141 struct ceph_inode_info *ci;
Sage Weila8599bd2009-10-06 11:31:12 -07003142 struct ceph_cap *cap;
3143 struct ceph_mds_caps *h;
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003144 struct ceph_mds_cap_peer *peer = NULL;
Yan, Zheng982d6012014-12-23 15:30:54 +08003145 struct ceph_snap_realm *realm;
Sage Weil2600d2d2010-02-22 15:12:16 -08003146 int mds = session->s_mds;
Yan, Zheng2cd698b2014-04-18 13:20:27 +08003147 int op, issued;
Sage Weil3d7ded42010-06-09 16:47:10 -07003148 u32 seq, mseq;
Sage Weila8599bd2009-10-06 11:31:12 -07003149 struct ceph_vino vino;
3150 u64 cap_id;
3151 u64 size, max_size;
Sage Weil6df058c2009-12-22 11:24:33 -08003152 u64 tid;
Yan, Zhengfb01d1f2014-11-14 21:29:55 +08003153 u64 inline_version = 0;
3154 void *inline_data = NULL;
3155 u32 inline_len = 0;
Sage Weil70edb552010-03-01 13:20:50 -08003156 void *snaptrace;
Sage Weilce1fbc82010-08-02 15:09:39 -07003157 size_t snaptrace_len;
Yan, Zhengfb01d1f2014-11-14 21:29:55 +08003158 void *p, *end;
Sage Weila8599bd2009-10-06 11:31:12 -07003159
3160 dout("handle_caps from mds%d\n", mds);
3161
3162 /* decode */
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003163 end = msg->front.iov_base + msg->front.iov_len;
Sage Weil6df058c2009-12-22 11:24:33 -08003164 tid = le64_to_cpu(msg->hdr.tid);
Sage Weila8599bd2009-10-06 11:31:12 -07003165 if (msg->front.iov_len < sizeof(*h))
3166 goto bad;
3167 h = msg->front.iov_base;
3168 op = le32_to_cpu(h->op);
3169 vino.ino = le64_to_cpu(h->ino);
3170 vino.snap = CEPH_NOSNAP;
3171 cap_id = le64_to_cpu(h->cap_id);
3172 seq = le32_to_cpu(h->seq);
Sage Weil3d7ded42010-06-09 16:47:10 -07003173 mseq = le32_to_cpu(h->migrate_seq);
Sage Weila8599bd2009-10-06 11:31:12 -07003174 size = le64_to_cpu(h->size);
3175 max_size = le64_to_cpu(h->max_size);
3176
Sage Weilce1fbc82010-08-02 15:09:39 -07003177 snaptrace = h + 1;
3178 snaptrace_len = le32_to_cpu(h->snap_trace_len);
Yan, Zhengfb01d1f2014-11-14 21:29:55 +08003179 p = snaptrace + snaptrace_len;
Sage Weilce1fbc82010-08-02 15:09:39 -07003180
3181 if (le16_to_cpu(msg->hdr.version) >= 2) {
Yan, Zhengfb01d1f2014-11-14 21:29:55 +08003182 u32 flock_len;
Sage Weilce1fbc82010-08-02 15:09:39 -07003183 ceph_decode_32_safe(&p, end, flock_len, bad);
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003184 if (p + flock_len > end)
3185 goto bad;
Yan, Zhengfb01d1f2014-11-14 21:29:55 +08003186 p += flock_len;
Sage Weilce1fbc82010-08-02 15:09:39 -07003187 }
3188
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003189 if (le16_to_cpu(msg->hdr.version) >= 3) {
3190 if (op == CEPH_CAP_OP_IMPORT) {
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003191 if (p + sizeof(*peer) > end)
3192 goto bad;
3193 peer = p;
Yan, Zhengfb01d1f2014-11-14 21:29:55 +08003194 p += sizeof(*peer);
Yan, Zheng11df2df2013-11-24 14:44:38 +08003195 } else if (op == CEPH_CAP_OP_EXPORT) {
3196 /* recorded in unused fields */
3197 peer = (void *)&h->size;
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003198 }
3199 }
3200
Yan, Zhengfb01d1f2014-11-14 21:29:55 +08003201 if (le16_to_cpu(msg->hdr.version) >= 4) {
3202 ceph_decode_64_safe(&p, end, inline_version, bad);
3203 ceph_decode_32_safe(&p, end, inline_len, bad);
3204 if (p + inline_len > end)
3205 goto bad;
3206 inline_data = p;
3207 p += inline_len;
3208 }
3209
Yan, Zheng6cd3bca2014-09-17 07:45:12 +08003210 /* lookup ino */
3211 inode = ceph_find_inode(sb, vino);
3212 ci = ceph_inode(inode);
3213 dout(" op %s ino %llx.%llx inode %p\n", ceph_cap_op_name(op), vino.ino,
3214 vino.snap, inode);
3215
Sage Weila8599bd2009-10-06 11:31:12 -07003216 mutex_lock(&session->s_mutex);
3217 session->s_seq++;
3218 dout(" mds%d seq %lld cap seq %u\n", session->s_mds, session->s_seq,
3219 (unsigned)seq);
3220
Sage Weila8599bd2009-10-06 11:31:12 -07003221 if (!inode) {
3222 dout(" i don't have ino %llx\n", vino.ino);
Sage Weil3d7ded42010-06-09 16:47:10 -07003223
Yan, Zhenga096b092013-09-22 10:15:58 +08003224 if (op == CEPH_CAP_OP_IMPORT) {
Yan, Zheng745a8e32015-05-14 17:22:42 +08003225 cap = ceph_get_cap(mdsc, NULL);
3226 cap->cap_ino = vino.ino;
3227 cap->queue_release = 1;
3228 cap->cap_id = cap_id;
3229 cap->mseq = mseq;
3230 cap->seq = seq;
Yan, Zhenga096b092013-09-22 10:15:58 +08003231 spin_lock(&session->s_cap_lock);
Yan, Zheng745a8e32015-05-14 17:22:42 +08003232 list_add_tail(&cap->session_caps,
3233 &session->s_cap_releases);
3234 session->s_num_cap_releases++;
Yan, Zhenga096b092013-09-22 10:15:58 +08003235 spin_unlock(&session->s_cap_lock);
3236 }
Greg Farnum21b559d2010-10-06 15:46:30 -07003237 goto flush_cap_releases;
Sage Weila8599bd2009-10-06 11:31:12 -07003238 }
3239
3240 /* these will work even if we don't have a cap yet */
3241 switch (op) {
3242 case CEPH_CAP_OP_FLUSHSNAP_ACK:
Sage Weil6df058c2009-12-22 11:24:33 -08003243 handle_cap_flushsnap_ack(inode, tid, h, session);
Sage Weila8599bd2009-10-06 11:31:12 -07003244 goto done;
3245
3246 case CEPH_CAP_OP_EXPORT:
Yan, Zheng11df2df2013-11-24 14:44:38 +08003247 handle_cap_export(inode, h, peer, session);
3248 goto done_unlocked;
Sage Weila8599bd2009-10-06 11:31:12 -07003249
3250 case CEPH_CAP_OP_IMPORT:
Yan, Zheng982d6012014-12-23 15:30:54 +08003251 realm = NULL;
3252 if (snaptrace_len) {
3253 down_write(&mdsc->snap_rwsem);
3254 ceph_update_snap_trace(mdsc, snaptrace,
3255 snaptrace + snaptrace_len,
3256 false, &realm);
3257 downgrade_write(&mdsc->snap_rwsem);
3258 } else {
3259 down_read(&mdsc->snap_rwsem);
3260 }
Yan, Zheng4ee6a912013-11-24 14:43:46 +08003261 handle_cap_import(mdsc, inode, h, peer, session,
Yan, Zheng2cd698b2014-04-18 13:20:27 +08003262 &cap, &issued);
Yan, Zheng982d6012014-12-23 15:30:54 +08003263 handle_cap_grant(mdsc, inode, h,
Yan, Zhengfb01d1f2014-11-14 21:29:55 +08003264 inline_version, inline_data, inline_len,
Yan, Zheng2cd698b2014-04-18 13:20:27 +08003265 msg->middle, session, cap, issued);
Yan, Zheng982d6012014-12-23 15:30:54 +08003266 if (realm)
3267 ceph_put_snap_realm(mdsc, realm);
Yan, Zheng2cd698b2014-04-18 13:20:27 +08003268 goto done_unlocked;
Sage Weila8599bd2009-10-06 11:31:12 -07003269 }
3270
3271 /* the rest require a cap */
Sage Weilbe655592011-11-30 09:47:09 -08003272 spin_lock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07003273 cap = __get_cap_for_mds(ceph_inode(inode), mds);
3274 if (!cap) {
Sage Weil9dbd4122010-06-10 13:21:20 -07003275 dout(" no cap on %p ino %llx.%llx from mds%d\n",
Sage Weila8599bd2009-10-06 11:31:12 -07003276 inode, ceph_ino(inode), ceph_snap(inode), mds);
Sage Weilbe655592011-11-30 09:47:09 -08003277 spin_unlock(&ci->i_ceph_lock);
Greg Farnum21b559d2010-10-06 15:46:30 -07003278 goto flush_cap_releases;
Sage Weila8599bd2009-10-06 11:31:12 -07003279 }
3280
Sage Weilbe655592011-11-30 09:47:09 -08003281 /* note that each of these drops i_ceph_lock for us */
Sage Weila8599bd2009-10-06 11:31:12 -07003282 switch (op) {
3283 case CEPH_CAP_OP_REVOKE:
3284 case CEPH_CAP_OP_GRANT:
Yan, Zheng2cd698b2014-04-18 13:20:27 +08003285 __ceph_caps_issued(ci, &issued);
3286 issued |= __ceph_caps_dirty(ci);
Yan, Zheng982d6012014-12-23 15:30:54 +08003287 handle_cap_grant(mdsc, inode, h,
Yan, Zhengfb01d1f2014-11-14 21:29:55 +08003288 inline_version, inline_data, inline_len,
3289 msg->middle, session, cap, issued);
Sage Weil15637c82010-03-16 13:42:00 -07003290 goto done_unlocked;
Sage Weila8599bd2009-10-06 11:31:12 -07003291
3292 case CEPH_CAP_OP_FLUSH_ACK:
Sage Weil6df058c2009-12-22 11:24:33 -08003293 handle_cap_flush_ack(inode, tid, h, session, cap);
Sage Weila8599bd2009-10-06 11:31:12 -07003294 break;
3295
3296 case CEPH_CAP_OP_TRUNC:
3297 handle_cap_trunc(inode, h, session);
3298 break;
3299
3300 default:
Sage Weilbe655592011-11-30 09:47:09 -08003301 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07003302 pr_err("ceph_handle_caps: unknown cap op %d %s\n", op,
3303 ceph_cap_op_name(op));
3304 }
3305
Greg Farnum21b559d2010-10-06 15:46:30 -07003306 goto done;
3307
3308flush_cap_releases:
3309 /*
Yan, Zheng745a8e32015-05-14 17:22:42 +08003310 * send any cap release message to try to move things
Greg Farnum21b559d2010-10-06 15:46:30 -07003311 * along for the mds (who clearly thinks we still have this
3312 * cap).
3313 */
Greg Farnum21b559d2010-10-06 15:46:30 -07003314 ceph_send_cap_releases(mdsc, session);
3315
Sage Weila8599bd2009-10-06 11:31:12 -07003316done:
Sage Weil15637c82010-03-16 13:42:00 -07003317 mutex_unlock(&session->s_mutex);
3318done_unlocked:
SF Markus Elfringe96a6502014-11-02 15:20:59 +01003319 iput(inode);
Sage Weila8599bd2009-10-06 11:31:12 -07003320 return;
3321
3322bad:
3323 pr_err("ceph_handle_caps: corrupt message\n");
Sage Weil9ec7cab2009-12-14 15:13:47 -08003324 ceph_msg_dump(msg);
Sage Weila8599bd2009-10-06 11:31:12 -07003325 return;
3326}
3327
3328/*
3329 * Delayed work handler to process end of delayed cap release LRU list.
3330 */
Sage Weilafcdaea2009-10-14 14:27:38 -07003331void ceph_check_delayed_caps(struct ceph_mds_client *mdsc)
Sage Weila8599bd2009-10-06 11:31:12 -07003332{
3333 struct ceph_inode_info *ci;
3334 int flags = CHECK_CAPS_NODELAY;
3335
Sage Weila8599bd2009-10-06 11:31:12 -07003336 dout("check_delayed_caps\n");
3337 while (1) {
3338 spin_lock(&mdsc->cap_delay_lock);
3339 if (list_empty(&mdsc->cap_delay_list))
3340 break;
3341 ci = list_first_entry(&mdsc->cap_delay_list,
3342 struct ceph_inode_info,
3343 i_cap_delay_list);
3344 if ((ci->i_ceph_flags & CEPH_I_FLUSH) == 0 &&
3345 time_before(jiffies, ci->i_hold_caps_max))
3346 break;
3347 list_del_init(&ci->i_cap_delay_list);
3348 spin_unlock(&mdsc->cap_delay_lock);
3349 dout("check_delayed_caps on %p\n", &ci->vfs_inode);
3350 ceph_check_caps(ci, flags, NULL);
3351 }
3352 spin_unlock(&mdsc->cap_delay_lock);
3353}
3354
3355/*
Sage Weilafcdaea2009-10-14 14:27:38 -07003356 * Flush all dirty caps to the mds
3357 */
3358void ceph_flush_dirty_caps(struct ceph_mds_client *mdsc)
3359{
Sage Weildb354052011-05-24 11:46:31 -07003360 struct ceph_inode_info *ci;
3361 struct inode *inode;
Sage Weilafcdaea2009-10-14 14:27:38 -07003362
3363 dout("flush_dirty_caps\n");
3364 spin_lock(&mdsc->cap_dirty_lock);
Sage Weildb354052011-05-24 11:46:31 -07003365 while (!list_empty(&mdsc->cap_dirty)) {
3366 ci = list_first_entry(&mdsc->cap_dirty, struct ceph_inode_info,
3367 i_dirty_item);
Sage Weil70b666c2011-05-27 09:24:26 -07003368 inode = &ci->vfs_inode;
3369 ihold(inode);
Sage Weildb354052011-05-24 11:46:31 -07003370 dout("flush_dirty_caps %p\n", inode);
Sage Weilafcdaea2009-10-14 14:27:38 -07003371 spin_unlock(&mdsc->cap_dirty_lock);
Sage Weil70b666c2011-05-27 09:24:26 -07003372 ceph_check_caps(ci, CHECK_CAPS_NODELAY|CHECK_CAPS_FLUSH, NULL);
3373 iput(inode);
Sage Weilafcdaea2009-10-14 14:27:38 -07003374 spin_lock(&mdsc->cap_dirty_lock);
3375 }
3376 spin_unlock(&mdsc->cap_dirty_lock);
Sage Weildb354052011-05-24 11:46:31 -07003377 dout("flush_dirty_caps done\n");
Sage Weilafcdaea2009-10-14 14:27:38 -07003378}
3379
3380/*
Sage Weila8599bd2009-10-06 11:31:12 -07003381 * Drop open file reference. If we were the last open file,
3382 * we may need to release capabilities to the MDS (or schedule
3383 * their delayed release).
3384 */
3385void ceph_put_fmode(struct ceph_inode_info *ci, int fmode)
3386{
3387 struct inode *inode = &ci->vfs_inode;
3388 int last = 0;
3389
Sage Weilbe655592011-11-30 09:47:09 -08003390 spin_lock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07003391 dout("put_fmode %p fmode %d %d -> %d\n", inode, fmode,
3392 ci->i_nr_by_mode[fmode], ci->i_nr_by_mode[fmode]-1);
3393 BUG_ON(ci->i_nr_by_mode[fmode] == 0);
3394 if (--ci->i_nr_by_mode[fmode] == 0)
3395 last++;
Sage Weilbe655592011-11-30 09:47:09 -08003396 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07003397
3398 if (last && ci->i_vino.snap == CEPH_NOSNAP)
3399 ceph_check_caps(ci, 0, NULL);
3400}
3401
3402/*
3403 * Helpers for embedding cap and dentry lease releases into mds
3404 * requests.
3405 *
3406 * @force is used by dentry_release (below) to force inclusion of a
3407 * record for the directory inode, even when there aren't any caps to
3408 * drop.
3409 */
3410int ceph_encode_inode_release(void **p, struct inode *inode,
3411 int mds, int drop, int unless, int force)
3412{
3413 struct ceph_inode_info *ci = ceph_inode(inode);
3414 struct ceph_cap *cap;
3415 struct ceph_mds_request_release *rel = *p;
Sage Weilec97f882010-06-24 15:12:37 -07003416 int used, dirty;
Sage Weila8599bd2009-10-06 11:31:12 -07003417 int ret = 0;
Sage Weila8599bd2009-10-06 11:31:12 -07003418
Sage Weilbe655592011-11-30 09:47:09 -08003419 spin_lock(&ci->i_ceph_lock);
Sage Weil916623d2010-03-16 15:01:07 -07003420 used = __ceph_caps_used(ci);
Sage Weilec97f882010-06-24 15:12:37 -07003421 dirty = __ceph_caps_dirty(ci);
Sage Weil916623d2010-03-16 15:01:07 -07003422
Sage Weilec97f882010-06-24 15:12:37 -07003423 dout("encode_inode_release %p mds%d used|dirty %s drop %s unless %s\n",
3424 inode, mds, ceph_cap_string(used|dirty), ceph_cap_string(drop),
Sage Weil916623d2010-03-16 15:01:07 -07003425 ceph_cap_string(unless));
3426
Sage Weilec97f882010-06-24 15:12:37 -07003427 /* only drop unused, clean caps */
3428 drop &= ~(used | dirty);
Sage Weil916623d2010-03-16 15:01:07 -07003429
Sage Weila8599bd2009-10-06 11:31:12 -07003430 cap = __get_cap_for_mds(ci, mds);
3431 if (cap && __cap_is_valid(cap)) {
3432 if (force ||
3433 ((cap->issued & drop) &&
3434 (cap->issued & unless) == 0)) {
3435 if ((cap->issued & drop) &&
3436 (cap->issued & unless) == 0) {
Yan, Zhengbb137f82013-06-03 18:22:17 +08003437 int wanted = __ceph_caps_wanted(ci);
3438 if ((ci->i_ceph_flags & CEPH_I_NODELAY) == 0)
3439 wanted |= cap->mds_wanted;
3440 dout("encode_inode_release %p cap %p "
3441 "%s -> %s, wanted %s -> %s\n", inode, cap,
Sage Weila8599bd2009-10-06 11:31:12 -07003442 ceph_cap_string(cap->issued),
Yan, Zhengbb137f82013-06-03 18:22:17 +08003443 ceph_cap_string(cap->issued & ~drop),
3444 ceph_cap_string(cap->mds_wanted),
3445 ceph_cap_string(wanted));
3446
Sage Weila8599bd2009-10-06 11:31:12 -07003447 cap->issued &= ~drop;
3448 cap->implemented &= ~drop;
Yan, Zhengbb137f82013-06-03 18:22:17 +08003449 cap->mds_wanted = wanted;
Sage Weila8599bd2009-10-06 11:31:12 -07003450 } else {
3451 dout("encode_inode_release %p cap %p %s"
3452 " (force)\n", inode, cap,
3453 ceph_cap_string(cap->issued));
3454 }
3455
3456 rel->ino = cpu_to_le64(ceph_ino(inode));
3457 rel->cap_id = cpu_to_le64(cap->cap_id);
3458 rel->seq = cpu_to_le32(cap->seq);
Himangi Saraogi08a0f242014-07-23 20:11:11 +05303459 rel->issue_seq = cpu_to_le32(cap->issue_seq);
Sage Weila8599bd2009-10-06 11:31:12 -07003460 rel->mseq = cpu_to_le32(cap->mseq);
Yan, Zhengfd7b95c2014-04-17 08:02:02 +08003461 rel->caps = cpu_to_le32(cap->implemented);
Sage Weila8599bd2009-10-06 11:31:12 -07003462 rel->wanted = cpu_to_le32(cap->mds_wanted);
3463 rel->dname_len = 0;
3464 rel->dname_seq = 0;
3465 *p += sizeof(*rel);
3466 ret = 1;
3467 } else {
3468 dout("encode_inode_release %p cap %p %s\n",
3469 inode, cap, ceph_cap_string(cap->issued));
3470 }
3471 }
Sage Weilbe655592011-11-30 09:47:09 -08003472 spin_unlock(&ci->i_ceph_lock);
Sage Weila8599bd2009-10-06 11:31:12 -07003473 return ret;
3474}
3475
3476int ceph_encode_dentry_release(void **p, struct dentry *dentry,
3477 int mds, int drop, int unless)
3478{
David Howells2b0143b2015-03-17 22:25:59 +00003479 struct inode *dir = d_inode(dentry->d_parent);
Sage Weila8599bd2009-10-06 11:31:12 -07003480 struct ceph_mds_request_release *rel = *p;
3481 struct ceph_dentry_info *di = ceph_dentry(dentry);
3482 int force = 0;
3483 int ret;
3484
3485 /*
3486 * force an record for the directory caps if we have a dentry lease.
Sage Weilbe655592011-11-30 09:47:09 -08003487 * this is racy (can't take i_ceph_lock and d_lock together), but it
Sage Weila8599bd2009-10-06 11:31:12 -07003488 * doesn't have to be perfect; the mds will revoke anything we don't
3489 * release.
3490 */
3491 spin_lock(&dentry->d_lock);
3492 if (di->lease_session && di->lease_session->s_mds == mds)
3493 force = 1;
3494 spin_unlock(&dentry->d_lock);
3495
3496 ret = ceph_encode_inode_release(p, dir, mds, drop, unless, force);
3497
3498 spin_lock(&dentry->d_lock);
3499 if (ret && di->lease_session && di->lease_session->s_mds == mds) {
3500 dout("encode_dentry_release %p mds%d seq %d\n",
3501 dentry, mds, (int)di->lease_seq);
3502 rel->dname_len = cpu_to_le32(dentry->d_name.len);
3503 memcpy(*p, dentry->d_name.name, dentry->d_name.len);
3504 *p += dentry->d_name.len;
3505 rel->dname_seq = cpu_to_le32(di->lease_seq);
Sage Weil1dadcce2010-07-23 13:54:21 -07003506 __ceph_mdsc_drop_dentry_lease(dentry);
Sage Weila8599bd2009-10-06 11:31:12 -07003507 }
3508 spin_unlock(&dentry->d_lock);
3509 return ret;
3510}