Loading...
1/*
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
24#include "super.h"
25#include "cache.h"
26
27struct ceph_aux_inode {
28 u64 version;
29 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
58int ceph_fscache_register(void)
59{
60 return fscache_register_netfs(&ceph_cache_netfs);
61}
62
63void ceph_fscache_unregister(void)
64{
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,
72 fsc, true);
73 if (!fsc->fscache)
74 pr_err("Unable to register fsid: %p fscache cookie\n", fsc);
75
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
85 /* use ceph virtual inode (id + snapshot) */
86 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));
102 aux.version = ci->i_version;
103 aux.mtime = inode->i_mtime;
104 aux.size = i_size_read(inode);
105
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;
115 *size = i_size_read(&ci->vfs_inode);
116}
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));
129 aux.version = ci->i_version;
130 aux.mtime = inode->i_mtime;
131 aux.size = i_size_read(inode);
132
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
180void ceph_fscache_register_inode_cookie(struct inode *inode)
181{
182 struct ceph_inode_info *ci = ceph_inode(inode);
183 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
184
185 /* No caching for filesystem */
186 if (fsc->fscache == NULL)
187 return;
188
189 /* Only cache for regular files that are read only */
190 if (!S_ISREG(inode->i_mode))
191 return;
192
193 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 }
199 inode_unlock(inode);
200}
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
215static 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
243static 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
257static inline bool cache_valid(struct ceph_inode_info *ci)
258{
259 return ci->i_fscache_gen == ci->i_rdcache_gen;
260}
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
328 if (!PageFsCache(page))
329 return;
330
331 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
343 if (!PageFsCache(page))
344 return;
345
346 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{
352 fscache_relinquish_cookie(fsc->fscache, 0);
353 fsc->fscache = NULL;
354}
355
356/*
357 * caller should hold CEPH_CAP_FILE_{RD,CACHE}
358 */
359void ceph_fscache_revalidate_cookie(struct ceph_inode_info *ci)
360{
361 if (cache_valid(ci))
362 return;
363
364 /* 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);
373 }
374 mutex_unlock(&ci->i_truncate_mutex);
375}
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Ceph cache definitions.
4 *
5 * Copyright (C) 2013 by Adfin Solutions, Inc. All Rights Reserved.
6 * Written by Milosz Tanski (milosz@adfin.com)
7 */
8
9#include <linux/ceph/ceph_debug.h>
10
11#include "super.h"
12#include "cache.h"
13
14struct ceph_aux_inode {
15 u64 version;
16 u64 mtime_sec;
17 u64 mtime_nsec;
18};
19
20struct fscache_netfs ceph_cache_netfs = {
21 .name = "ceph",
22 .version = 0,
23};
24
25static DEFINE_MUTEX(ceph_fscache_lock);
26static LIST_HEAD(ceph_fscache_list);
27
28struct ceph_fscache_entry {
29 struct list_head list;
30 struct fscache_cookie *fscache;
31 size_t uniq_len;
32 /* The following members must be last */
33 struct ceph_fsid fsid;
34 char uniquifier[0];
35};
36
37static const struct fscache_cookie_def ceph_fscache_fsid_object_def = {
38 .name = "CEPH.fsid",
39 .type = FSCACHE_COOKIE_TYPE_INDEX,
40};
41
42int __init ceph_fscache_register(void)
43{
44 return fscache_register_netfs(&ceph_cache_netfs);
45}
46
47void ceph_fscache_unregister(void)
48{
49 fscache_unregister_netfs(&ceph_cache_netfs);
50}
51
52int ceph_fscache_register_fs(struct ceph_fs_client* fsc)
53{
54 const struct ceph_fsid *fsid = &fsc->client->fsid;
55 const char *fscache_uniq = fsc->mount_options->fscache_uniq;
56 size_t uniq_len = fscache_uniq ? strlen(fscache_uniq) : 0;
57 struct ceph_fscache_entry *ent;
58 int err = 0;
59
60 mutex_lock(&ceph_fscache_lock);
61 list_for_each_entry(ent, &ceph_fscache_list, list) {
62 if (memcmp(&ent->fsid, fsid, sizeof(*fsid)))
63 continue;
64 if (ent->uniq_len != uniq_len)
65 continue;
66 if (uniq_len && memcmp(ent->uniquifier, fscache_uniq, uniq_len))
67 continue;
68
69 pr_err("fscache cookie already registered for fsid %pU\n", fsid);
70 pr_err(" use fsc=%%s mount option to specify a uniquifier\n");
71 err = -EBUSY;
72 goto out_unlock;
73 }
74
75 ent = kzalloc(sizeof(*ent) + uniq_len, GFP_KERNEL);
76 if (!ent) {
77 err = -ENOMEM;
78 goto out_unlock;
79 }
80
81 memcpy(&ent->fsid, fsid, sizeof(*fsid));
82 if (uniq_len > 0) {
83 memcpy(&ent->uniquifier, fscache_uniq, uniq_len);
84 ent->uniq_len = uniq_len;
85 }
86
87 fsc->fscache = fscache_acquire_cookie(ceph_cache_netfs.primary_index,
88 &ceph_fscache_fsid_object_def,
89 &ent->fsid, sizeof(ent->fsid) + uniq_len,
90 NULL, 0,
91 fsc, 0, true);
92
93 if (fsc->fscache) {
94 ent->fscache = fsc->fscache;
95 list_add_tail(&ent->list, &ceph_fscache_list);
96 } else {
97 kfree(ent);
98 pr_err("unable to register fscache cookie for fsid %pU\n",
99 fsid);
100 /* all other fs ignore this error */
101 }
102out_unlock:
103 mutex_unlock(&ceph_fscache_lock);
104 return err;
105}
106
107static enum fscache_checkaux ceph_fscache_inode_check_aux(
108 void *cookie_netfs_data, const void *data, uint16_t dlen,
109 loff_t object_size)
110{
111 struct ceph_aux_inode aux;
112 struct ceph_inode_info* ci = cookie_netfs_data;
113 struct inode* inode = &ci->vfs_inode;
114
115 if (dlen != sizeof(aux) ||
116 i_size_read(inode) != object_size)
117 return FSCACHE_CHECKAUX_OBSOLETE;
118
119 memset(&aux, 0, sizeof(aux));
120 aux.version = ci->i_version;
121 aux.mtime_sec = inode->i_mtime.tv_sec;
122 aux.mtime_nsec = inode->i_mtime.tv_nsec;
123
124 if (memcmp(data, &aux, sizeof(aux)) != 0)
125 return FSCACHE_CHECKAUX_OBSOLETE;
126
127 dout("ceph inode 0x%p cached okay\n", ci);
128 return FSCACHE_CHECKAUX_OKAY;
129}
130
131static const struct fscache_cookie_def ceph_fscache_inode_object_def = {
132 .name = "CEPH.inode",
133 .type = FSCACHE_COOKIE_TYPE_DATAFILE,
134 .check_aux = ceph_fscache_inode_check_aux,
135};
136
137void ceph_fscache_register_inode_cookie(struct inode *inode)
138{
139 struct ceph_inode_info *ci = ceph_inode(inode);
140 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
141 struct ceph_aux_inode aux;
142
143 /* No caching for filesystem */
144 if (!fsc->fscache)
145 return;
146
147 /* Only cache for regular files that are read only */
148 if (!S_ISREG(inode->i_mode))
149 return;
150
151 inode_lock_nested(inode, I_MUTEX_CHILD);
152 if (!ci->fscache) {
153 memset(&aux, 0, sizeof(aux));
154 aux.version = ci->i_version;
155 aux.mtime_sec = inode->i_mtime.tv_sec;
156 aux.mtime_nsec = inode->i_mtime.tv_nsec;
157 ci->fscache = fscache_acquire_cookie(fsc->fscache,
158 &ceph_fscache_inode_object_def,
159 &ci->i_vino, sizeof(ci->i_vino),
160 &aux, sizeof(aux),
161 ci, i_size_read(inode), false);
162 }
163 inode_unlock(inode);
164}
165
166void ceph_fscache_unregister_inode_cookie(struct ceph_inode_info* ci)
167{
168 struct fscache_cookie* cookie;
169
170 if ((cookie = ci->fscache) == NULL)
171 return;
172
173 ci->fscache = NULL;
174
175 fscache_uncache_all_inode_pages(cookie, &ci->vfs_inode);
176 fscache_relinquish_cookie(cookie, &ci->i_vino, false);
177}
178
179static bool ceph_fscache_can_enable(void *data)
180{
181 struct inode *inode = data;
182 return !inode_is_open_for_write(inode);
183}
184
185void ceph_fscache_file_set_cookie(struct inode *inode, struct file *filp)
186{
187 struct ceph_inode_info *ci = ceph_inode(inode);
188
189 if (!fscache_cookie_valid(ci->fscache))
190 return;
191
192 if (inode_is_open_for_write(inode)) {
193 dout("fscache_file_set_cookie %p %p disabling cache\n",
194 inode, filp);
195 fscache_disable_cookie(ci->fscache, &ci->i_vino, false);
196 fscache_uncache_all_inode_pages(ci->fscache, inode);
197 } else {
198 fscache_enable_cookie(ci->fscache, &ci->i_vino, i_size_read(inode),
199 ceph_fscache_can_enable, inode);
200 if (fscache_cookie_enabled(ci->fscache)) {
201 dout("fscache_file_set_cookie %p %p enabling cache\n",
202 inode, filp);
203 }
204 }
205}
206
207static void ceph_readpage_from_fscache_complete(struct page *page, void *data, int error)
208{
209 if (!error)
210 SetPageUptodate(page);
211
212 unlock_page(page);
213}
214
215static inline bool cache_valid(struct ceph_inode_info *ci)
216{
217 return ci->i_fscache_gen == ci->i_rdcache_gen;
218}
219
220
221/* Atempt to read from the fscache,
222 *
223 * This function is called from the readpage_nounlock context. DO NOT attempt to
224 * unlock the page here (or in the callback).
225 */
226int ceph_readpage_from_fscache(struct inode *inode, struct page *page)
227{
228 struct ceph_inode_info *ci = ceph_inode(inode);
229 int ret;
230
231 if (!cache_valid(ci))
232 return -ENOBUFS;
233
234 ret = fscache_read_or_alloc_page(ci->fscache, page,
235 ceph_readpage_from_fscache_complete, NULL,
236 GFP_KERNEL);
237
238 switch (ret) {
239 case 0: /* Page found */
240 dout("page read submitted\n");
241 return 0;
242 case -ENOBUFS: /* Pages were not found, and can't be */
243 case -ENODATA: /* Pages were not found */
244 dout("page/inode not in cache\n");
245 return ret;
246 default:
247 dout("%s: unknown error ret = %i\n", __func__, ret);
248 return ret;
249 }
250}
251
252int ceph_readpages_from_fscache(struct inode *inode,
253 struct address_space *mapping,
254 struct list_head *pages,
255 unsigned *nr_pages)
256{
257 struct ceph_inode_info *ci = ceph_inode(inode);
258 int ret;
259
260 if (!cache_valid(ci))
261 return -ENOBUFS;
262
263 ret = fscache_read_or_alloc_pages(ci->fscache, mapping, pages, nr_pages,
264 ceph_readpage_from_fscache_complete,
265 NULL, mapping_gfp_mask(mapping));
266
267 switch (ret) {
268 case 0: /* All pages found */
269 dout("all-page read submitted\n");
270 return 0;
271 case -ENOBUFS: /* Some pages were not found, and can't be */
272 case -ENODATA: /* some pages were not found */
273 dout("page/inode not in cache\n");
274 return ret;
275 default:
276 dout("%s: unknown error ret = %i\n", __func__, ret);
277 return ret;
278 }
279}
280
281void ceph_readpage_to_fscache(struct inode *inode, struct page *page)
282{
283 struct ceph_inode_info *ci = ceph_inode(inode);
284 int ret;
285
286 if (!PageFsCache(page))
287 return;
288
289 if (!cache_valid(ci))
290 return;
291
292 ret = fscache_write_page(ci->fscache, page, i_size_read(inode),
293 GFP_KERNEL);
294 if (ret)
295 fscache_uncache_page(ci->fscache, page);
296}
297
298void ceph_invalidate_fscache_page(struct inode* inode, struct page *page)
299{
300 struct ceph_inode_info *ci = ceph_inode(inode);
301
302 if (!PageFsCache(page))
303 return;
304
305 fscache_wait_on_page_write(ci->fscache, page);
306 fscache_uncache_page(ci->fscache, page);
307}
308
309void ceph_fscache_unregister_fs(struct ceph_fs_client* fsc)
310{
311 if (fscache_cookie_valid(fsc->fscache)) {
312 struct ceph_fscache_entry *ent;
313 bool found = false;
314
315 mutex_lock(&ceph_fscache_lock);
316 list_for_each_entry(ent, &ceph_fscache_list, list) {
317 if (ent->fscache == fsc->fscache) {
318 list_del(&ent->list);
319 kfree(ent);
320 found = true;
321 break;
322 }
323 }
324 WARN_ON_ONCE(!found);
325 mutex_unlock(&ceph_fscache_lock);
326
327 __fscache_relinquish_cookie(fsc->fscache, NULL, false);
328 }
329 fsc->fscache = NULL;
330}
331
332/*
333 * caller should hold CEPH_CAP_FILE_{RD,CACHE}
334 */
335void ceph_fscache_revalidate_cookie(struct ceph_inode_info *ci)
336{
337 if (cache_valid(ci))
338 return;
339
340 /* resue i_truncate_mutex. There should be no pending
341 * truncate while the caller holds CEPH_CAP_FILE_RD */
342 mutex_lock(&ci->i_truncate_mutex);
343 if (!cache_valid(ci)) {
344 if (fscache_check_consistency(ci->fscache, &ci->i_vino))
345 fscache_invalidate(ci->fscache);
346 spin_lock(&ci->i_ceph_lock);
347 ci->i_fscache_gen = ci->i_rdcache_gen;
348 spin_unlock(&ci->i_ceph_lock);
349 }
350 mutex_unlock(&ci->i_truncate_mutex);
351}