blob: 5bc5d37b121712a2f288ede38b46420d13a2f0e5 [file] [log] [blame]
Milosz Tanski99ccbd22013-08-21 17:29:54 -04001/*
2 * Ceph cache definitions.
3 *
4 * Copyright (C) 2013 by Adfin Solutions, Inc. All Rights Reserved.
5 * Written by Milosz Tanski (milosz@adfin.com)
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License version 2
9 * as published by the Free Software Foundation.
10 *
11 * This program is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 * GNU General Public License for more details.
15 *
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write to:
18 * Free Software Foundation
19 * 51 Franklin Street, Fifth Floor
20 * Boston, MA 02111-1301 USA
21 *
22 */
23
Milosz Tanski99ccbd22013-08-21 17:29:54 -040024#include "super.h"
25#include "cache.h"
26
27struct ceph_aux_inode {
Yan, Zhengf6973c02016-05-20 16:57:29 +080028 u64 version;
Milosz Tanski99ccbd22013-08-21 17:29:54 -040029 struct timespec mtime;
30 loff_t size;
31};
32
33struct fscache_netfs ceph_cache_netfs = {
34 .name = "ceph",
35 .version = 0,
36};
37
38static uint16_t ceph_fscache_session_get_key(const void *cookie_netfs_data,
39 void *buffer, uint16_t maxbuf)
40{
41 const struct ceph_fs_client* fsc = cookie_netfs_data;
42 uint16_t klen;
43
44 klen = sizeof(fsc->client->fsid);
45 if (klen > maxbuf)
46 return 0;
47
48 memcpy(buffer, &fsc->client->fsid, klen);
49 return klen;
50}
51
52static const struct fscache_cookie_def ceph_fscache_fsid_object_def = {
53 .name = "CEPH.fsid",
54 .type = FSCACHE_COOKIE_TYPE_INDEX,
55 .get_key = ceph_fscache_session_get_key,
56};
57
Milosz Tanski971f0bd2013-09-06 15:13:18 +000058int ceph_fscache_register(void)
Milosz Tanski99ccbd22013-08-21 17:29:54 -040059{
60 return fscache_register_netfs(&ceph_cache_netfs);
61}
62
Milosz Tanski971f0bd2013-09-06 15:13:18 +000063void ceph_fscache_unregister(void)
Milosz Tanski99ccbd22013-08-21 17:29:54 -040064{
65 fscache_unregister_netfs(&ceph_cache_netfs);
66}
67
68int ceph_fscache_register_fs(struct ceph_fs_client* fsc)
69{
70 fsc->fscache = fscache_acquire_cookie(ceph_cache_netfs.primary_index,
71 &ceph_fscache_fsid_object_def,
David Howells94d30ae2013-09-21 00:09:31 +010072 fsc, true);
Yan, Zhengf7f7e7a2016-05-18 20:31:55 +080073 if (!fsc->fscache)
Colin Ian King679f0b82016-06-23 14:45:24 +010074 pr_err("Unable to register fsid: %p fscache cookie\n", fsc);
Milosz Tanski99ccbd22013-08-21 17:29:54 -040075
76 return 0;
77}
78
79static uint16_t ceph_fscache_inode_get_key(const void *cookie_netfs_data,
80 void *buffer, uint16_t maxbuf)
81{
82 const struct ceph_inode_info* ci = cookie_netfs_data;
83 uint16_t klen;
84
Geliang Tang1291fb92015-09-30 11:41:05 +080085 /* use ceph virtual inode (id + snapshot) */
Milosz Tanski99ccbd22013-08-21 17:29:54 -040086 klen = sizeof(ci->i_vino);
87 if (klen > maxbuf)
88 return 0;
89
90 memcpy(buffer, &ci->i_vino, klen);
91 return klen;
92}
93
94static uint16_t ceph_fscache_inode_get_aux(const void *cookie_netfs_data,
95 void *buffer, uint16_t bufmax)
96{
97 struct ceph_aux_inode aux;
98 const struct ceph_inode_info* ci = cookie_netfs_data;
99 const struct inode* inode = &ci->vfs_inode;
100
101 memset(&aux, 0, sizeof(aux));
Yan, Zhengf6973c02016-05-20 16:57:29 +0800102 aux.version = ci->i_version;
Milosz Tanski99ccbd22013-08-21 17:29:54 -0400103 aux.mtime = inode->i_mtime;
Yan, Zheng99c88e62015-12-30 11:32:46 +0800104 aux.size = i_size_read(inode);
Milosz Tanski99ccbd22013-08-21 17:29:54 -0400105
106 memcpy(buffer, &aux, sizeof(aux));
107
108 return sizeof(aux);
109}
110
111static void ceph_fscache_inode_get_attr(const void *cookie_netfs_data,
112 uint64_t *size)
113{
114 const struct ceph_inode_info* ci = cookie_netfs_data;
Yan, Zheng99c88e62015-12-30 11:32:46 +0800115 *size = i_size_read(&ci->vfs_inode);
Milosz Tanski99ccbd22013-08-21 17:29:54 -0400116}
117
118static enum fscache_checkaux ceph_fscache_inode_check_aux(
119 void *cookie_netfs_data, const void *data, uint16_t dlen)
120{
121 struct ceph_aux_inode aux;
122 struct ceph_inode_info* ci = cookie_netfs_data;
123 struct inode* inode = &ci->vfs_inode;
124
125 if (dlen != sizeof(aux))
126 return FSCACHE_CHECKAUX_OBSOLETE;
127
128 memset(&aux, 0, sizeof(aux));
Yan, Zhengf6973c02016-05-20 16:57:29 +0800129 aux.version = ci->i_version;
Milosz Tanski99ccbd22013-08-21 17:29:54 -0400130 aux.mtime = inode->i_mtime;
Yan, Zheng99c88e62015-12-30 11:32:46 +0800131 aux.size = i_size_read(inode);
Milosz Tanski99ccbd22013-08-21 17:29:54 -0400132
133 if (memcmp(data, &aux, sizeof(aux)) != 0)
134 return FSCACHE_CHECKAUX_OBSOLETE;
135
136 dout("ceph inode 0x%p cached okay", ci);
137 return FSCACHE_CHECKAUX_OKAY;
138}
139
140static void ceph_fscache_inode_now_uncached(void* cookie_netfs_data)
141{
142 struct ceph_inode_info* ci = cookie_netfs_data;
143 struct pagevec pvec;
144 pgoff_t first;
145 int loop, nr_pages;
146
147 pagevec_init(&pvec, 0);
148 first = 0;
149
150 dout("ceph inode 0x%p now uncached", ci);
151
152 while (1) {
153 nr_pages = pagevec_lookup(&pvec, ci->vfs_inode.i_mapping, first,
154 PAGEVEC_SIZE - pagevec_count(&pvec));
155
156 if (!nr_pages)
157 break;
158
159 for (loop = 0; loop < nr_pages; loop++)
160 ClearPageFsCache(pvec.pages[loop]);
161
162 first = pvec.pages[nr_pages - 1]->index + 1;
163
164 pvec.nr = nr_pages;
165 pagevec_release(&pvec);
166 cond_resched();
167 }
168}
169
170static const struct fscache_cookie_def ceph_fscache_inode_object_def = {
171 .name = "CEPH.inode",
172 .type = FSCACHE_COOKIE_TYPE_DATAFILE,
173 .get_key = ceph_fscache_inode_get_key,
174 .get_attr = ceph_fscache_inode_get_attr,
175 .get_aux = ceph_fscache_inode_get_aux,
176 .check_aux = ceph_fscache_inode_check_aux,
177 .now_uncached = ceph_fscache_inode_now_uncached,
178};
179
Yan, Zheng46b59b22016-05-18 15:25:03 +0800180void ceph_fscache_register_inode_cookie(struct inode *inode)
Milosz Tanski99ccbd22013-08-21 17:29:54 -0400181{
Yan, Zheng46b59b22016-05-18 15:25:03 +0800182 struct ceph_inode_info *ci = ceph_inode(inode);
183 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
Milosz Tanski99ccbd22013-08-21 17:29:54 -0400184
185 /* No caching for filesystem */
186 if (fsc->fscache == NULL)
187 return;
188
189 /* Only cache for regular files that are read only */
Yan, Zheng46b59b22016-05-18 15:25:03 +0800190 if (!S_ISREG(inode->i_mode))
Milosz Tanski99ccbd22013-08-21 17:29:54 -0400191 return;
192
Yan, Zheng46b59b22016-05-18 15:25:03 +0800193 inode_lock_nested(inode, I_MUTEX_CHILD);
194 if (!ci->fscache) {
195 ci->fscache = fscache_acquire_cookie(fsc->fscache,
196 &ceph_fscache_inode_object_def,
197 ci, false);
198 }
Al Viro59551022016-01-22 15:40:57 -0500199 inode_unlock(inode);
Milosz Tanski99ccbd22013-08-21 17:29:54 -0400200}
201
202void ceph_fscache_unregister_inode_cookie(struct ceph_inode_info* ci)
203{
204 struct fscache_cookie* cookie;
205
206 if ((cookie = ci->fscache) == NULL)
207 return;
208
209 ci->fscache = NULL;
210
211 fscache_uncache_all_inode_pages(cookie, &ci->vfs_inode);
212 fscache_relinquish_cookie(cookie, 0);
213}
214
Yan, Zheng46b59b22016-05-18 15:25:03 +0800215static bool ceph_fscache_can_enable(void *data)
216{
217 struct inode *inode = data;
218 return !inode_is_open_for_write(inode);
219}
220
221void ceph_fscache_file_set_cookie(struct inode *inode, struct file *filp)
222{
223 struct ceph_inode_info *ci = ceph_inode(inode);
224
225 if (!fscache_cookie_valid(ci->fscache))
226 return;
227
228 if (inode_is_open_for_write(inode)) {
229 dout("fscache_file_set_cookie %p %p disabling cache\n",
230 inode, filp);
231 fscache_disable_cookie(ci->fscache, false);
232 fscache_uncache_all_inode_pages(ci->fscache, inode);
233 } else {
234 fscache_enable_cookie(ci->fscache, ceph_fscache_can_enable,
235 inode);
236 if (fscache_cookie_enabled(ci->fscache)) {
237 dout("fscache_file_set_cookie %p %p enabing cache\n",
238 inode, filp);
239 }
240 }
241}
242
Milosz Tanski99ccbd22013-08-21 17:29:54 -0400243static void ceph_vfs_readpage_complete(struct page *page, void *data, int error)
244{
245 if (!error)
246 SetPageUptodate(page);
247}
248
249static void ceph_vfs_readpage_complete_unlock(struct page *page, void *data, int error)
250{
251 if (!error)
252 SetPageUptodate(page);
253
254 unlock_page(page);
255}
256
Zhang Zhuoyu3b33f692016-03-25 05:18:39 -0400257static inline bool cache_valid(struct ceph_inode_info *ci)
Milosz Tanski99ccbd22013-08-21 17:29:54 -0400258{
Yan, Zhengf7f7e7a2016-05-18 20:31:55 +0800259 return ci->i_fscache_gen == ci->i_rdcache_gen;
Milosz Tanski99ccbd22013-08-21 17:29:54 -0400260}
261
262
263/* Atempt to read from the fscache,
264 *
265 * This function is called from the readpage_nounlock context. DO NOT attempt to
266 * unlock the page here (or in the callback).
267 */
268int ceph_readpage_from_fscache(struct inode *inode, struct page *page)
269{
270 struct ceph_inode_info *ci = ceph_inode(inode);
271 int ret;
272
273 if (!cache_valid(ci))
274 return -ENOBUFS;
275
276 ret = fscache_read_or_alloc_page(ci->fscache, page,
277 ceph_vfs_readpage_complete, NULL,
278 GFP_KERNEL);
279
280 switch (ret) {
281 case 0: /* Page found */
282 dout("page read submitted\n");
283 return 0;
284 case -ENOBUFS: /* Pages were not found, and can't be */
285 case -ENODATA: /* Pages were not found */
286 dout("page/inode not in cache\n");
287 return ret;
288 default:
289 dout("%s: unknown error ret = %i\n", __func__, ret);
290 return ret;
291 }
292}
293
294int ceph_readpages_from_fscache(struct inode *inode,
295 struct address_space *mapping,
296 struct list_head *pages,
297 unsigned *nr_pages)
298{
299 struct ceph_inode_info *ci = ceph_inode(inode);
300 int ret;
301
302 if (!cache_valid(ci))
303 return -ENOBUFS;
304
305 ret = fscache_read_or_alloc_pages(ci->fscache, mapping, pages, nr_pages,
306 ceph_vfs_readpage_complete_unlock,
307 NULL, mapping_gfp_mask(mapping));
308
309 switch (ret) {
310 case 0: /* All pages found */
311 dout("all-page read submitted\n");
312 return 0;
313 case -ENOBUFS: /* Some pages were not found, and can't be */
314 case -ENODATA: /* some pages were not found */
315 dout("page/inode not in cache\n");
316 return ret;
317 default:
318 dout("%s: unknown error ret = %i\n", __func__, ret);
319 return ret;
320 }
321}
322
323void ceph_readpage_to_fscache(struct inode *inode, struct page *page)
324{
325 struct ceph_inode_info *ci = ceph_inode(inode);
326 int ret;
327
Milosz Tanski9b8dd1e2013-09-03 19:11:01 -0400328 if (!PageFsCache(page))
329 return;
330
Milosz Tanski99ccbd22013-08-21 17:29:54 -0400331 if (!cache_valid(ci))
332 return;
333
334 ret = fscache_write_page(ci->fscache, page, GFP_KERNEL);
335 if (ret)
336 fscache_uncache_page(ci->fscache, page);
337}
338
339void ceph_invalidate_fscache_page(struct inode* inode, struct page *page)
340{
341 struct ceph_inode_info *ci = ceph_inode(inode);
342
Milosz Tanskiffc79662013-09-25 11:18:14 -0400343 if (!PageFsCache(page))
344 return;
345
Milosz Tanski99ccbd22013-08-21 17:29:54 -0400346 fscache_wait_on_page_write(ci->fscache, page);
347 fscache_uncache_page(ci->fscache, page);
348}
349
350void ceph_fscache_unregister_fs(struct ceph_fs_client* fsc)
351{
Milosz Tanski99ccbd22013-08-21 17:29:54 -0400352 fscache_relinquish_cookie(fsc->fscache, 0);
353 fsc->fscache = NULL;
354}
355
Yan, Zhengf7f7e7a2016-05-18 20:31:55 +0800356/*
357 * caller should hold CEPH_CAP_FILE_{RD,CACHE}
358 */
359void ceph_fscache_revalidate_cookie(struct ceph_inode_info *ci)
Milosz Tanski99ccbd22013-08-21 17:29:54 -0400360{
Yan, Zhengf7f7e7a2016-05-18 20:31:55 +0800361 if (cache_valid(ci))
Milosz Tanskie81568e2013-09-05 18:29:03 +0000362 return;
363
Yan, Zhengf7f7e7a2016-05-18 20:31:55 +0800364 /* resue i_truncate_mutex. There should be no pending
365 * truncate while the caller holds CEPH_CAP_FILE_RD */
366 mutex_lock(&ci->i_truncate_mutex);
367 if (!cache_valid(ci)) {
368 if (fscache_check_consistency(ci->fscache))
369 fscache_invalidate(ci->fscache);
370 spin_lock(&ci->i_ceph_lock);
371 ci->i_fscache_gen = ci->i_rdcache_gen;
372 spin_unlock(&ci->i_ceph_lock);
Milosz Tanski99ccbd22013-08-21 17:29:54 -0400373 }
Yan, Zhengf7f7e7a2016-05-18 20:31:55 +0800374 mutex_unlock(&ci->i_truncate_mutex);
Milosz Tanski99ccbd22013-08-21 17:29:54 -0400375}