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 void ceph_fscache_inode_now_uncached(void* cookie_netfs_data)
 198 {
 199         struct ceph_inode_info* ci = cookie_netfs_data;
 200         struct pagevec pvec;
 201         pgoff_t first;
 202         int loop, nr_pages;
 203
 204         pagevec_init(&pvec, 0);
 205         first = 0;
 206
 207         dout("ceph inode 0x%p now uncached", ci);
 208
 209         while (1) {
 210                 nr_pages = pagevec_lookup(&pvec, ci->vfs_inode.i_mapping, first,
 211                                           PAGEVEC_SIZE - pagevec_count(&pvec));
 212
 213                 if (!nr_pages)
 214                         break;
 215
 216                 for (loop = 0; loop < nr_pages; loop++)
 217                         ClearPageFsCache(pvec.pages[loop]);
 218
 219                 first = pvec.pages[nr_pages - 1]->index + 1;
 220
 221                 pvec.nr = nr_pages;
 222                 pagevec_release(&pvec);
 223                 cond_resched();
 224         }
 225 }
 226
 227 static const struct fscache_cookie_def ceph_fscache_inode_object_def = {
 228         .name           = "CEPH.inode",
 229         .type           = FSCACHE_COOKIE_TYPE_DATAFILE,
 230         .get_key        = ceph_fscache_inode_get_key,
 231         .get_attr       = ceph_fscache_inode_get_attr,
 232         .get_aux        = ceph_fscache_inode_get_aux,
 233         .check_aux      = ceph_fscache_inode_check_aux,
 234         .now_uncached   = ceph_fscache_inode_now_uncached,
 235 };
 236
 237 void ceph_fscache_register_inode_cookie(struct inode *inode)
 238 {
 239         struct ceph_inode_info *ci = ceph_inode(inode);
 240         struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
 241
 242         /* No caching for filesystem */
 243         if (fsc->fscache == NULL)
 244                 return;
 245
 246         /* Only cache for regular files that are read only */
 247         if (!S_ISREG(inode->i_mode))
 248                 return;
 249
 250         inode_lock_nested(inode, I_MUTEX_CHILD);
 251         if (!ci->fscache) {
 252                 ci->fscache = fscache_acquire_cookie(fsc->fscache,
 253                                         &ceph_fscache_inode_object_def,
 254                                         ci, false);
 255         }
 256         inode_unlock(inode);
 257 }
 258
 259 void ceph_fscache_unregister_inode_cookie(struct ceph_inode_info* ci)
 260 {
 261         struct fscache_cookie* cookie;
 262
 263         if ((cookie = ci->fscache) == NULL)
 264                 return;
 265
 266         ci->fscache = NULL;
 267
 268         fscache_uncache_all_inode_pages(cookie, &ci->vfs_inode);
 269         fscache_relinquish_cookie(cookie, 0);
 270 }
 271
 272 static bool ceph_fscache_can_enable(void *data)
 273 {
 274         struct inode *inode = data;
 275         return !inode_is_open_for_write(inode);
 276 }
 277
 278 void ceph_fscache_file_set_cookie(struct inode *inode, struct file *filp)
 279 {
 280         struct ceph_inode_info *ci = ceph_inode(inode);
 281
 282         if (!fscache_cookie_valid(ci->fscache))
 283                 return;
 284
 285         if (inode_is_open_for_write(inode)) {
 286                 dout("fscache_file_set_cookie %p %p disabling cache\n",
 287                      inode, filp);
 288                 fscache_disable_cookie(ci->fscache, false);
 289                 fscache_uncache_all_inode_pages(ci->fscache, inode);
 290         } else {
 291                 fscache_enable_cookie(ci->fscache, ceph_fscache_can_enable,
 292                                 inode);
 293                 if (fscache_cookie_enabled(ci->fscache)) {
 294                         dout("fscache_file_set_cookie %p %p enabling cache\n",
 295                              inode, filp);
 296                 }
 297         }
 298 }
 299
 300 static void ceph_readpage_from_fscache_complete(struct page *page, void *data, int error)
 301 {
 302         if (!error)
 303                 SetPageUptodate(page);
 304
 305         unlock_page(page);
 306 }
 307
 308 static inline bool cache_valid(struct ceph_inode_info *ci)
 309 {
 310         return ci->i_fscache_gen == ci->i_rdcache_gen;
 311 }
 312
 313
 314 /* Atempt to read from the fscache,
 315  *
 316  * This function is called from the readpage_nounlock context. DO NOT attempt to
 317  * unlock the page here (or in the callback).
 318  */
 319 int ceph_readpage_from_fscache(struct inode *inode, struct page *page)
 320 {
 321         struct ceph_inode_info *ci = ceph_inode(inode);
 322         int ret;
 323
 324         if (!cache_valid(ci))
 325                 return -ENOBUFS;
 326
 327         ret = fscache_read_or_alloc_page(ci->fscache, page,
 328                                          ceph_readpage_from_fscache_complete, NULL,
 329                                          GFP_KERNEL);
 330
 331         switch (ret) {
 332                 case 0: /* Page found */
 333                         dout("page read submitted\n");
 334                         return 0;
 335                 case -ENOBUFS: /* Pages were not found, and can't be */
 336                 case -ENODATA: /* Pages were not found */
 337                         dout("page/inode not in cache\n");
 338                         return ret;
 339                 default:
 340                         dout("%s: unknown error ret = %i\n", __func__, ret);
 341                         return ret;
 342         }
 343 }
 344
 345 int ceph_readpages_from_fscache(struct inode *inode,
 346                                   struct address_space *mapping,
 347                                   struct list_head *pages,
 348                                   unsigned *nr_pages)
 349 {
 350         struct ceph_inode_info *ci = ceph_inode(inode);
 351         int ret;
 352
 353         if (!cache_valid(ci))
 354                 return -ENOBUFS;
 355
 356         ret = fscache_read_or_alloc_pages(ci->fscache, mapping, pages, nr_pages,
 357                                           ceph_readpage_from_fscache_complete,
 358                                           NULL, mapping_gfp_mask(mapping));
 359
 360         switch (ret) {
 361                 case 0: /* All pages found */
 362                         dout("all-page read submitted\n");
 363                         return 0;
 364                 case -ENOBUFS: /* Some pages were not found, and can't be */
 365                 case -ENODATA: /* some pages were not found */
 366                         dout("page/inode not in cache\n");
 367                         return ret;
 368                 default:
 369                         dout("%s: unknown error ret = %i\n", __func__, ret);
 370                         return ret;
 371         }
 372 }
 373
 374 void ceph_readpage_to_fscache(struct inode *inode, struct page *page)
 375 {
 376         struct ceph_inode_info *ci = ceph_inode(inode);
 377         int ret;
 378
 379         if (!PageFsCache(page))
 380                 return;
 381
 382         if (!cache_valid(ci))
 383                 return;
 384
 385         ret = fscache_write_page(ci->fscache, page, GFP_KERNEL);
 386         if (ret)
 387                  fscache_uncache_page(ci->fscache, page);
 388 }
 389
 390 void ceph_invalidate_fscache_page(struct inode* inode, struct page *page)
 391 {
 392         struct ceph_inode_info *ci = ceph_inode(inode);
 393
 394         if (!PageFsCache(page))
 395                 return;
 396
 397         fscache_wait_on_page_write(ci->fscache, page);
 398         fscache_uncache_page(ci->fscache, page);
 399 }
 400
 401 void ceph_fscache_unregister_fs(struct ceph_fs_client* fsc)
 402 {
 403         if (fscache_cookie_valid(fsc->fscache)) {
 404                 struct ceph_fscache_entry *ent;
 405                 bool found = false;
 406
 407                 mutex_lock(&ceph_fscache_lock);
 408                 list_for_each_entry(ent, &ceph_fscache_list, list) {
 409                         if (ent->fscache == fsc->fscache) {
 410                                 list_del(&ent->list);
 411                                 kfree(ent);
 412                                 found = true;
 413                                 break;
 414                         }
 415                 }
 416                 WARN_ON_ONCE(!found);
 417                 mutex_unlock(&ceph_fscache_lock);
 418
 419                 __fscache_relinquish_cookie(fsc->fscache, 0);
 420         }
 421         fsc->fscache = NULL;
 422 }
 423
 424 /*
 425  * caller should hold CEPH_CAP_FILE_{RD,CACHE}
 426  */
 427 void ceph_fscache_revalidate_cookie(struct ceph_inode_info *ci)
 428 {
 429         if (cache_valid(ci))
 430                 return;
 431
 432         /* resue i_truncate_mutex. There should be no pending
 433          * truncate while the caller holds CEPH_CAP_FILE_RD */
 434         mutex_lock(&ci->i_truncate_mutex);
 435         if (!cache_valid(ci)) {
 436                 if (fscache_check_consistency(ci->fscache))
 437                         fscache_invalidate(ci->fscache);
 438                 spin_lock(&ci->i_ceph_lock);
 439                 ci->i_fscache_gen = ci->i_rdcache_gen;
 440                 spin_unlock(&ci->i_ceph_lock);
 441         }
 442         mutex_unlock(&ci->i_truncate_mutex);
 443 }