fs/ceph/cache.c

   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
  27 struct ceph_aux_inode {
  28         u64             version;
  29         struct timespec mtime;
  30         loff_t          size;
  31 };
  32
  33 struct fscache_netfs ceph_cache_netfs = {
  34         .name           = "ceph",
  35         .version        = 0,
  36 };
  37
  38 static DEFINE_MUTEX(ceph_fscache_lock);
  39 static LIST_HEAD(ceph_fscache_list);
  40
  41 struct ceph_fscache_entry {
  42         struct list_head list;
  43         struct fscache_cookie *fscache;
  44         struct ceph_fsid fsid;
  45         size_t uniq_len;
  46         char uniquifier[0];
  47 };
  48
  49 static uint16_t ceph_fscache_session_get_key(const void *cookie_netfs_data,
  50                                              void *buffer, uint16_t maxbuf)
  51 {
  52         const struct ceph_fs_client* fsc = cookie_netfs_data;
  53         const char *fscache_uniq = fsc->mount_options->fscache_uniq;
  54         uint16_t fsid_len, uniq_len;
  55
  56         fsid_len = sizeof(fsc->client->fsid);
  57         uniq_len = fscache_uniq ? strlen(fscache_uniq) : 0;
  58         if (fsid_len + uniq_len > maxbuf)
  59                 return 0;
  60
  61         memcpy(buffer, &fsc->client->fsid, fsid_len);
  62         if (uniq_len)
  63                 memcpy(buffer + fsid_len, fscache_uniq, uniq_len);
  64
  65         return fsid_len + uniq_len;
  66 }
  67
  68 static const struct fscache_cookie_def ceph_fscache_fsid_object_def = {
  69         .name           = "CEPH.fsid",
  70         .type           = FSCACHE_COOKIE_TYPE_INDEX,
  71         .get_key        = ceph_fscache_session_get_key,
  72 };
  73
  74 int ceph_fscache_register(void)
  75 {
  76         return fscache_register_netfs(&ceph_cache_netfs);
  77 }
  78
  79 void ceph_fscache_unregister(void)
  80 {
  81         fscache_unregister_netfs(&ceph_cache_netfs);
  82 }
  83
  84 int ceph_fscache_register_fs(struct ceph_fs_client* fsc)
  85 {
  86         const struct ceph_fsid *fsid = &fsc->client->fsid;
  87         const char *fscache_uniq = fsc->mount_options->fscache_uniq;
  88         size_t uniq_len = fscache_uniq ? strlen(fscache_uniq) : 0;
  89         struct ceph_fscache_entry *ent;
  90         int err = 0;
  91
  92         mutex_lock(&ceph_fscache_lock);
  93         list_for_each_entry(ent, &ceph_fscache_list, list) {
  94                 if (memcmp(&ent->fsid, fsid, sizeof(*fsid)))
  95                         continue;
  96                 if (ent->uniq_len != uniq_len)
  97                         continue;
  98                 if (uniq_len && memcmp(ent->uniquifier, fscache_uniq, uniq_len))
  99                         continue;
 100
 101                 pr_err("fscache cookie already registered for fsid %pU\n", fsid);
 102                 pr_err("  use fsc=%%s mount option to specify a uniquifier\n");
 103                 err = -EBUSY;
 104                 goto out_unlock;
 105         }
 106
 107         ent = kzalloc(sizeof(*ent) + uniq_len, GFP_KERNEL);
 108         if (!ent) {
 109                 err = -ENOMEM;
 110                 goto out_unlock;
 111         }
 112
 113         fsc->fscache = fscache_acquire_cookie(ceph_cache_netfs.primary_index,
 114                                               &ceph_fscache_fsid_object_def,
 115                                               fsc, true);
 116
 117         if (fsc->fscache) {
 118                 memcpy(&ent->fsid, fsid, sizeof(*fsid));
 119                 if (uniq_len > 0) {
 120                         memcpy(&ent->uniquifier, fscache_uniq, uniq_len);
 121                         ent->uniq_len = uniq_len;
 122                 }
 123                 ent->fscache = fsc->fscache;
 124                 list_add_tail(&ent->list, &ceph_fscache_list);
 125         } else {
 126                 kfree(ent);
 127                 pr_err("unable to register fscache cookie for fsid %pU\n",
 128                        fsid);
 129                 /* all other fs ignore this error */
 130         }
 131 out_unlock:
 132         mutex_unlock(&ceph_fscache_lock);
 133         return err;
 134 }
 135
 136 static uint16_t ceph_fscache_inode_get_key(const void *cookie_netfs_data,
 137                                            void *buffer, uint16_t maxbuf)
 138 {
 139         const struct ceph_inode_info* ci = cookie_netfs_data;
 140         uint16_t klen;
 141
 142         /* use ceph virtual inode (id + snapshot) */
 143         klen = sizeof(ci->i_vino);
 144         if (klen > maxbuf)
 145                 return 0;
 146
 147         memcpy(buffer, &ci->i_vino, klen);
 148         return klen;
 149 }
 150
 151 static uint16_t ceph_fscache_inode_get_aux(const void *cookie_netfs_data,
 152                                            void *buffer, uint16_t bufmax)
 153 {
 154         struct ceph_aux_inode aux;
 155         const struct ceph_inode_info* ci = cookie_netfs_data;
 156         const struct inode* inode = &ci->vfs_inode;
 157
 158         memset(&aux, 0, sizeof(aux));
 159         aux.version = ci->i_version;
 160         aux.mtime = inode->i_mtime;
 161         aux.size = i_size_read(inode);
 162
 163         memcpy(buffer, &aux, sizeof(aux));
 164
 165         return sizeof(aux);
 166 }
 167
 168 static void ceph_fscache_inode_get_attr(const void *cookie_netfs_data,
 169                                         uint64_t *size)
 170 {
 171         const struct ceph_inode_info* ci = cookie_netfs_data;
 172         *size = i_size_read(&ci->vfs_inode);
 173 }
 174
 175 static enum fscache_checkaux ceph_fscache_inode_check_aux(
 176         void *cookie_netfs_data, const void *data, uint16_t dlen)
 177 {
 178         struct ceph_aux_inode aux;
 179         struct ceph_inode_info* ci = cookie_netfs_data;
 180         struct inode* inode = &ci->vfs_inode;
 181
 182         if (dlen != sizeof(aux))
 183                 return FSCACHE_CHECKAUX_OBSOLETE;
 184
 185         memset(&aux, 0, sizeof(aux));
 186         aux.version = ci->i_version;
 187         aux.mtime = inode->i_mtime;
 188         aux.size = i_size_read(inode);
 189
 190         if (memcmp(data, &aux, sizeof(aux)) != 0)
 191                 return FSCACHE_CHECKAUX_OBSOLETE;
 192
 193         dout("ceph inode 0x%p cached okay", ci);
 194         return FSCACHE_CHECKAUX_OKAY;
 195 }
 196
 197 static const struct fscache_cookie_def ceph_fscache_inode_object_def = {
 198         .name           = "CEPH.inode",
 199         .type           = FSCACHE_COOKIE_TYPE_DATAFILE,
 200         .get_key        = ceph_fscache_inode_get_key,
 201         .get_attr       = ceph_fscache_inode_get_attr,
 202         .get_aux        = ceph_fscache_inode_get_aux,
 203         .check_aux      = ceph_fscache_inode_check_aux,
 204 };
 205
 206 void ceph_fscache_register_inode_cookie(struct inode *inode)
 207 {
 208         struct ceph_inode_info *ci = ceph_inode(inode);
 209         struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
 210
 211         /* No caching for filesystem */
 212         if (!fsc->fscache)
 213                 return;
 214
 215         /* Only cache for regular files that are read only */
 216         if (!S_ISREG(inode->i_mode))
 217                 return;
 218
 219         inode_lock_nested(inode, I_MUTEX_CHILD);
 220         if (!ci->fscache) {
 221                 ci->fscache = fscache_acquire_cookie(fsc->fscache,
 222                                         &ceph_fscache_inode_object_def,
 223                                         ci, false);
 224         }
 225         inode_unlock(inode);
 226 }
 227
 228 void ceph_fscache_unregister_inode_cookie(struct ceph_inode_info* ci)
 229 {
 230         struct fscache_cookie* cookie;
 231
 232         if ((cookie = ci->fscache) == NULL)
 233                 return;
 234
 235         ci->fscache = NULL;
 236
 237         fscache_uncache_all_inode_pages(cookie, &ci->vfs_inode);
 238         fscache_relinquish_cookie(cookie, 0);
 239 }
 240
 241 static bool ceph_fscache_can_enable(void *data)
 242 {
 243         struct inode *inode = data;
 244         return !inode_is_open_for_write(inode);
 245 }
 246
 247 void ceph_fscache_file_set_cookie(struct inode *inode, struct file *filp)
 248 {
 249         struct ceph_inode_info *ci = ceph_inode(inode);
 250
 251         if (!fscache_cookie_valid(ci->fscache))
 252                 return;
 253
 254         if (inode_is_open_for_write(inode)) {
 255                 dout("fscache_file_set_cookie %p %p disabling cache\n",
 256                      inode, filp);
 257                 fscache_disable_cookie(ci->fscache, false);
 258                 fscache_uncache_all_inode_pages(ci->fscache, inode);
 259         } else {
 260                 fscache_enable_cookie(ci->fscache, ceph_fscache_can_enable,
 261                                 inode);
 262                 if (fscache_cookie_enabled(ci->fscache)) {
 263                         dout("fscache_file_set_cookie %p %p enabling cache\n",
 264                              inode, filp);
 265                 }
 266         }
 267 }
 268
 269 static void ceph_readpage_from_fscache_complete(struct page *page, void *data, int error)
 270 {
 271         if (!error)
 272                 SetPageUptodate(page);
 273
 274         unlock_page(page);
 275 }
 276
 277 static inline bool cache_valid(struct ceph_inode_info *ci)
 278 {
 279         return ci->i_fscache_gen == ci->i_rdcache_gen;
 280 }
 281
 282
 283 /* Atempt to read from the fscache,
 284  *
 285  * This function is called from the readpage_nounlock context. DO NOT attempt to
 286  * unlock the page here (or in the callback).
 287  */
 288 int ceph_readpage_from_fscache(struct inode *inode, struct page *page)
 289 {
 290         struct ceph_inode_info *ci = ceph_inode(inode);
 291         int ret;
 292
 293         if (!cache_valid(ci))
 294                 return -ENOBUFS;
 295
 296         ret = fscache_read_or_alloc_page(ci->fscache, page,
 297                                          ceph_readpage_from_fscache_complete, NULL,
 298                                          GFP_KERNEL);
 299
 300         switch (ret) {
 301                 case 0: /* Page found */
 302                         dout("page read submitted\n");
 303                         return 0;
 304                 case -ENOBUFS: /* Pages were not found, and can't be */
 305                 case -ENODATA: /* Pages were not found */
 306                         dout("page/inode not in cache\n");
 307                         return ret;
 308                 default:
 309                         dout("%s: unknown error ret = %i\n", __func__, ret);
 310                         return ret;
 311         }
 312 }
 313
 314 int ceph_readpages_from_fscache(struct inode *inode,
 315                                   struct address_space *mapping,
 316                                   struct list_head *pages,
 317                                   unsigned *nr_pages)
 318 {
 319         struct ceph_inode_info *ci = ceph_inode(inode);
 320         int ret;
 321
 322         if (!cache_valid(ci))
 323                 return -ENOBUFS;
 324
 325         ret = fscache_read_or_alloc_pages(ci->fscache, mapping, pages, nr_pages,
 326                                           ceph_readpage_from_fscache_complete,
 327                                           NULL, mapping_gfp_mask(mapping));
 328
 329         switch (ret) {
 330                 case 0: /* All pages found */
 331                         dout("all-page read submitted\n");
 332                         return 0;
 333                 case -ENOBUFS: /* Some pages were not found, and can't be */
 334                 case -ENODATA: /* some pages were not found */
 335                         dout("page/inode not in cache\n");
 336                         return ret;
 337                 default:
 338                         dout("%s: unknown error ret = %i\n", __func__, ret);
 339                         return ret;
 340         }
 341 }
 342
 343 void ceph_readpage_to_fscache(struct inode *inode, struct page *page)
 344 {
 345         struct ceph_inode_info *ci = ceph_inode(inode);
 346         int ret;
 347
 348         if (!PageFsCache(page))
 349                 return;
 350
 351         if (!cache_valid(ci))
 352                 return;
 353
 354         ret = fscache_write_page(ci->fscache, page, GFP_KERNEL);
 355         if (ret)
 356                  fscache_uncache_page(ci->fscache, page);
 357 }
 358
 359 void ceph_invalidate_fscache_page(struct inode* inode, struct page *page)
 360 {
 361         struct ceph_inode_info *ci = ceph_inode(inode);
 362
 363         if (!PageFsCache(page))
 364                 return;
 365
 366         fscache_wait_on_page_write(ci->fscache, page);
 367         fscache_uncache_page(ci->fscache, page);
 368 }
 369
 370 void ceph_fscache_unregister_fs(struct ceph_fs_client* fsc)
 371 {
 372         if (fscache_cookie_valid(fsc->fscache)) {
 373                 struct ceph_fscache_entry *ent;
 374                 bool found = false;
 375
 376                 mutex_lock(&ceph_fscache_lock);
 377                 list_for_each_entry(ent, &ceph_fscache_list, list) {
 378                         if (ent->fscache == fsc->fscache) {
 379                                 list_del(&ent->list);
 380                                 kfree(ent);
 381                                 found = true;
 382                                 break;
 383                         }
 384                 }
 385                 WARN_ON_ONCE(!found);
 386                 mutex_unlock(&ceph_fscache_lock);
 387
 388                 __fscache_relinquish_cookie(fsc->fscache, 0);
 389         }
 390         fsc->fscache = NULL;
 391 }
 392
 393 /*
 394  * caller should hold CEPH_CAP_FILE_{RD,CACHE}
 395  */
 396 void ceph_fscache_revalidate_cookie(struct ceph_inode_info *ci)
 397 {
 398         if (cache_valid(ci))
 399                 return;
 400
 401         /* resue i_truncate_mutex. There should be no pending
 402          * truncate while the caller holds CEPH_CAP_FILE_RD */
 403         mutex_lock(&ci->i_truncate_mutex);
 404         if (!cache_valid(ci)) {
 405                 if (fscache_check_consistency(ci->fscache))
 406                         fscache_invalidate(ci->fscache);
 407                 spin_lock(&ci->i_ceph_lock);
 408                 ci->i_fscache_gen = ci->i_rdcache_gen;
 409                 spin_unlock(&ci->i_ceph_lock);
 410         }
 411         mutex_unlock(&ci->i_truncate_mutex);
 412 }