linux/fs/ceph/cache.c
<<
>>
Prefs
   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        struct timespec mtime;
  29        loff_t          size;
  30};
  31
  32struct fscache_netfs ceph_cache_netfs = {
  33        .name           = "ceph",
  34        .version        = 0,
  35};
  36
  37static uint16_t ceph_fscache_session_get_key(const void *cookie_netfs_data,
  38                                             void *buffer, uint16_t maxbuf)
  39{
  40        const struct ceph_fs_client* fsc = cookie_netfs_data;
  41        uint16_t klen;
  42
  43        klen = sizeof(fsc->client->fsid);
  44        if (klen > maxbuf)
  45                return 0;
  46
  47        memcpy(buffer, &fsc->client->fsid, klen);
  48        return klen;
  49}
  50
  51static const struct fscache_cookie_def ceph_fscache_fsid_object_def = {
  52        .name           = "CEPH.fsid",
  53        .type           = FSCACHE_COOKIE_TYPE_INDEX,
  54        .get_key        = ceph_fscache_session_get_key,
  55};
  56
  57int ceph_fscache_register(void)
  58{
  59        return fscache_register_netfs(&ceph_cache_netfs);
  60}
  61
  62void ceph_fscache_unregister(void)
  63{
  64        fscache_unregister_netfs(&ceph_cache_netfs);
  65}
  66
  67int ceph_fscache_register_fs(struct ceph_fs_client* fsc)
  68{
  69        fsc->fscache = fscache_acquire_cookie(ceph_cache_netfs.primary_index,
  70                                              &ceph_fscache_fsid_object_def,
  71                                              fsc, true);
  72
  73        if (fsc->fscache == NULL) {
  74                pr_err("Unable to resgister fsid: %p fscache cookie", fsc);
  75                return 0;
  76        }
  77
  78        fsc->revalidate_wq = alloc_workqueue("ceph-revalidate", 0, 1);
  79        if (fsc->revalidate_wq == NULL)
  80                return -ENOMEM;
  81
  82        return 0;
  83}
  84
  85static uint16_t ceph_fscache_inode_get_key(const void *cookie_netfs_data,
  86                                           void *buffer, uint16_t maxbuf)
  87{
  88        const struct ceph_inode_info* ci = cookie_netfs_data;
  89        uint16_t klen;
  90
  91        /* use ceph virtual inode (id + snapshot) */
  92        klen = sizeof(ci->i_vino);
  93        if (klen > maxbuf)
  94                return 0;
  95
  96        memcpy(buffer, &ci->i_vino, klen);
  97        return klen;
  98}
  99
 100static uint16_t ceph_fscache_inode_get_aux(const void *cookie_netfs_data,
 101                                           void *buffer, uint16_t bufmax)
 102{
 103        struct ceph_aux_inode aux;
 104        const struct ceph_inode_info* ci = cookie_netfs_data;
 105        const struct inode* inode = &ci->vfs_inode;
 106
 107        memset(&aux, 0, sizeof(aux));
 108        aux.mtime = inode->i_mtime;
 109        aux.size = i_size_read(inode);
 110
 111        memcpy(buffer, &aux, sizeof(aux));
 112
 113        return sizeof(aux);
 114}
 115
 116static void ceph_fscache_inode_get_attr(const void *cookie_netfs_data,
 117                                        uint64_t *size)
 118{
 119        const struct ceph_inode_info* ci = cookie_netfs_data;
 120        *size = i_size_read(&ci->vfs_inode);
 121}
 122
 123static enum fscache_checkaux ceph_fscache_inode_check_aux(
 124        void *cookie_netfs_data, const void *data, uint16_t dlen)
 125{
 126        struct ceph_aux_inode aux;
 127        struct ceph_inode_info* ci = cookie_netfs_data;
 128        struct inode* inode = &ci->vfs_inode;
 129
 130        if (dlen != sizeof(aux))
 131                return FSCACHE_CHECKAUX_OBSOLETE;
 132
 133        memset(&aux, 0, sizeof(aux));
 134        aux.mtime = inode->i_mtime;
 135        aux.size = i_size_read(inode);
 136
 137        if (memcmp(data, &aux, sizeof(aux)) != 0)
 138                return FSCACHE_CHECKAUX_OBSOLETE;
 139
 140        dout("ceph inode 0x%p cached okay", ci);
 141        return FSCACHE_CHECKAUX_OKAY;
 142}
 143
 144static void ceph_fscache_inode_now_uncached(void* cookie_netfs_data)
 145{
 146        struct ceph_inode_info* ci = cookie_netfs_data;
 147        struct pagevec pvec;
 148        pgoff_t first;
 149        int loop, nr_pages;
 150
 151        pagevec_init(&pvec, 0);
 152        first = 0;
 153
 154        dout("ceph inode 0x%p now uncached", ci);
 155
 156        while (1) {
 157                nr_pages = pagevec_lookup(&pvec, ci->vfs_inode.i_mapping, first,
 158                                          PAGEVEC_SIZE - pagevec_count(&pvec));
 159
 160                if (!nr_pages)
 161                        break;
 162
 163                for (loop = 0; loop < nr_pages; loop++)
 164                        ClearPageFsCache(pvec.pages[loop]);
 165
 166                first = pvec.pages[nr_pages - 1]->index + 1;
 167
 168                pvec.nr = nr_pages;
 169                pagevec_release(&pvec);
 170                cond_resched();
 171        }
 172}
 173
 174static const struct fscache_cookie_def ceph_fscache_inode_object_def = {
 175        .name           = "CEPH.inode",
 176        .type           = FSCACHE_COOKIE_TYPE_DATAFILE,
 177        .get_key        = ceph_fscache_inode_get_key,
 178        .get_attr       = ceph_fscache_inode_get_attr,
 179        .get_aux        = ceph_fscache_inode_get_aux,
 180        .check_aux      = ceph_fscache_inode_check_aux,
 181        .now_uncached   = ceph_fscache_inode_now_uncached,
 182};
 183
 184void ceph_fscache_register_inode_cookie(struct ceph_fs_client* fsc,
 185                                        struct ceph_inode_info* ci)
 186{
 187        struct inode* inode = &ci->vfs_inode;
 188
 189        /* No caching for filesystem */
 190        if (fsc->fscache == NULL)
 191                return;
 192
 193        /* Only cache for regular files that are read only */
 194        if ((ci->vfs_inode.i_mode & S_IFREG) == 0)
 195                return;
 196
 197        /* Avoid multiple racing open requests */
 198        inode_lock(inode);
 199
 200        if (ci->fscache)
 201                goto done;
 202
 203        ci->fscache = fscache_acquire_cookie(fsc->fscache,
 204                                             &ceph_fscache_inode_object_def,
 205                                             ci, true);
 206        fscache_check_consistency(ci->fscache);
 207done:
 208        inode_unlock(inode);
 209
 210}
 211
 212void ceph_fscache_unregister_inode_cookie(struct ceph_inode_info* ci)
 213{
 214        struct fscache_cookie* cookie;
 215
 216        if ((cookie = ci->fscache) == NULL)
 217                return;
 218
 219        ci->fscache = NULL;
 220
 221        fscache_uncache_all_inode_pages(cookie, &ci->vfs_inode);
 222        fscache_relinquish_cookie(cookie, 0);
 223}
 224
 225static void ceph_vfs_readpage_complete(struct page *page, void *data, int error)
 226{
 227        if (!error)
 228                SetPageUptodate(page);
 229}
 230
 231static void ceph_vfs_readpage_complete_unlock(struct page *page, void *data, int error)
 232{
 233        if (!error)
 234                SetPageUptodate(page);
 235
 236        unlock_page(page);
 237}
 238
 239static inline int cache_valid(struct ceph_inode_info *ci)
 240{
 241        return ((ceph_caps_issued(ci) & CEPH_CAP_FILE_CACHE) &&
 242                (ci->i_fscache_gen == ci->i_rdcache_gen));
 243}
 244
 245
 246/* Atempt to read from the fscache,
 247 *
 248 * This function is called from the readpage_nounlock context. DO NOT attempt to
 249 * unlock the page here (or in the callback).
 250 */
 251int ceph_readpage_from_fscache(struct inode *inode, struct page *page)
 252{
 253        struct ceph_inode_info *ci = ceph_inode(inode);
 254        int ret;
 255
 256        if (!cache_valid(ci))
 257                return -ENOBUFS;
 258
 259        ret = fscache_read_or_alloc_page(ci->fscache, page,
 260                                         ceph_vfs_readpage_complete, NULL,
 261                                         GFP_KERNEL);
 262
 263        switch (ret) {
 264                case 0: /* Page found */
 265                        dout("page read submitted\n");
 266                        return 0;
 267                case -ENOBUFS: /* Pages were not found, and can't be */
 268                case -ENODATA: /* Pages were not found */
 269                        dout("page/inode not in cache\n");
 270                        return ret;
 271                default:
 272                        dout("%s: unknown error ret = %i\n", __func__, ret);
 273                        return ret;
 274        }
 275}
 276
 277int ceph_readpages_from_fscache(struct inode *inode,
 278                                  struct address_space *mapping,
 279                                  struct list_head *pages,
 280                                  unsigned *nr_pages)
 281{
 282        struct ceph_inode_info *ci = ceph_inode(inode);
 283        int ret;
 284
 285        if (!cache_valid(ci))
 286                return -ENOBUFS;
 287
 288        ret = fscache_read_or_alloc_pages(ci->fscache, mapping, pages, nr_pages,
 289                                          ceph_vfs_readpage_complete_unlock,
 290                                          NULL, mapping_gfp_mask(mapping));
 291
 292        switch (ret) {
 293                case 0: /* All pages found */
 294                        dout("all-page read submitted\n");
 295                        return 0;
 296                case -ENOBUFS: /* Some pages were not found, and can't be */
 297                case -ENODATA: /* some pages were not found */
 298                        dout("page/inode not in cache\n");
 299                        return ret;
 300                default:
 301                        dout("%s: unknown error ret = %i\n", __func__, ret);
 302                        return ret;
 303        }
 304}
 305
 306void ceph_readpage_to_fscache(struct inode *inode, struct page *page)
 307{
 308        struct ceph_inode_info *ci = ceph_inode(inode);
 309        int ret;
 310
 311        if (!PageFsCache(page))
 312                return;
 313
 314        if (!cache_valid(ci))
 315                return;
 316
 317        ret = fscache_write_page(ci->fscache, page, GFP_KERNEL);
 318        if (ret)
 319                 fscache_uncache_page(ci->fscache, page);
 320}
 321
 322void ceph_invalidate_fscache_page(struct inode* inode, struct page *page)
 323{
 324        struct ceph_inode_info *ci = ceph_inode(inode);
 325
 326        if (!PageFsCache(page))
 327                return;
 328
 329        fscache_wait_on_page_write(ci->fscache, page);
 330        fscache_uncache_page(ci->fscache, page);
 331}
 332
 333void ceph_fscache_unregister_fs(struct ceph_fs_client* fsc)
 334{
 335        if (fsc->revalidate_wq)
 336                destroy_workqueue(fsc->revalidate_wq);
 337
 338        fscache_relinquish_cookie(fsc->fscache, 0);
 339        fsc->fscache = NULL;
 340}
 341
 342static void ceph_revalidate_work(struct work_struct *work)
 343{
 344        int issued;
 345        u32 orig_gen;
 346        struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
 347                                                  i_revalidate_work);
 348        struct inode *inode = &ci->vfs_inode;
 349
 350        spin_lock(&ci->i_ceph_lock);
 351        issued = __ceph_caps_issued(ci, NULL);
 352        orig_gen = ci->i_rdcache_gen;
 353        spin_unlock(&ci->i_ceph_lock);
 354
 355        if (!(issued & CEPH_CAP_FILE_CACHE)) {
 356                dout("revalidate_work lost cache before validation %p\n",
 357                     inode);
 358                goto out;
 359        }
 360
 361        if (!fscache_check_consistency(ci->fscache))
 362                fscache_invalidate(ci->fscache);
 363
 364        spin_lock(&ci->i_ceph_lock);
 365        /* Update the new valid generation (backwards sanity check too) */
 366        if (orig_gen > ci->i_fscache_gen) {
 367                ci->i_fscache_gen = orig_gen;
 368        }
 369        spin_unlock(&ci->i_ceph_lock);
 370
 371out:
 372        iput(&ci->vfs_inode);
 373}
 374
 375void ceph_queue_revalidate(struct inode *inode)
 376{
 377        struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
 378        struct ceph_inode_info *ci = ceph_inode(inode);
 379
 380        if (fsc->revalidate_wq == NULL || ci->fscache == NULL)
 381                return;
 382
 383        ihold(inode);
 384
 385        if (queue_work(ceph_sb_to_client(inode->i_sb)->revalidate_wq,
 386                       &ci->i_revalidate_work)) {
 387                dout("ceph_queue_revalidate %p\n", inode);
 388        } else {
 389                dout("ceph_queue_revalidate %p failed\n)", inode);
 390                iput(inode);
 391        }
 392}
 393
 394void ceph_fscache_inode_init(struct ceph_inode_info *ci)
 395{
 396        ci->fscache = NULL;
 397        /* The first load is verifed cookie open time */
 398        ci->i_fscache_gen = 1;
 399        INIT_WORK(&ci->i_revalidate_work, ceph_revalidate_work);
 400}
 401