mm/cleancache.c

   1 /*
   2  * Cleancache frontend
   3  *
   4  * This code provides the generic "frontend" layer to call a matching
   5  * "backend" driver implementation of cleancache.  See
   6  * Documentation/vm/cleancache.txt for more information.
   7  *
   8  * Copyright (C) 2009-2010 Oracle Corp. All rights reserved.
   9  * Author: Dan Magenheimer
  10  *
  11  * This work is licensed under the terms of the GNU GPL, version 2.
  12  */
  13
  14 #include <linux/module.h>
  15 #include <linux/fs.h>
  16 #include <linux/exportfs.h>
  17 #include <linux/mm.h>
  18 #include <linux/debugfs.h>
  19 #include <linux/cleancache.h>
  20
  21 /*
  22  * cleancache_ops is set by cleancache_ops_register to contain the pointers
  23  * to the cleancache "backend" implementation functions.
  24  */
  25 static struct cleancache_ops *cleancache_ops __read_mostly;
  26
  27 /*
  28  * Counters available via /sys/kernel/debug/cleancache (if debugfs is
  29  * properly configured.  These are for information only so are not protected
  30  * against increment races.
  31  */
  32 static u64 cleancache_succ_gets;
  33 static u64 cleancache_failed_gets;
  34 static u64 cleancache_puts;
  35 static u64 cleancache_invalidates;
  36
  37 /*
  38  * When no backend is registered all calls to init_fs and init_shared_fs
  39  * are registered and fake poolids (FAKE_FS_POOLID_OFFSET or
  40  * FAKE_SHARED_FS_POOLID_OFFSET, plus offset in the respective array
  41  * [shared_|]fs_poolid_map) are given to the respective super block
  42  * (sb->cleancache_poolid) and no tmem_pools are created. When a backend
  43  * registers with cleancache the previous calls to init_fs and init_shared_fs
  44  * are executed to create tmem_pools and set the respective poolids. While no
  45  * backend is registered all "puts", "gets" and "flushes" are ignored or failed.
  46  */
  47 #define MAX_INITIALIZABLE_FS 32
  48 #define FAKE_FS_POOLID_OFFSET 1000
  49 #define FAKE_SHARED_FS_POOLID_OFFSET 2000
  50
  51 #define FS_NO_BACKEND (-1)
  52 #define FS_UNKNOWN (-2)
  53 static int fs_poolid_map[MAX_INITIALIZABLE_FS];
  54 static int shared_fs_poolid_map[MAX_INITIALIZABLE_FS];
  55 static char *uuids[MAX_INITIALIZABLE_FS];
  56 /*
  57  * Mutex for the [shared_|]fs_poolid_map to guard against multiple threads
  58  * invoking umount (and ending in __cleancache_invalidate_fs) and also multiple
  59  * threads calling mount (and ending up in __cleancache_init_[shared|]fs).
  60  */
  61 static DEFINE_MUTEX(poolid_mutex);
  62 /*
  63  * When set to false (default) all calls to the cleancache functions, except
  64  * the __cleancache_invalidate_fs and __cleancache_init_[shared|]fs are guarded
  65  * by the if (!cleancache_ops) return. This means multiple threads (from
  66  * different filesystems) will be checking cleancache_ops. The usage of a
  67  * bool instead of a atomic_t or a bool guarded by a spinlock is OK - we are
  68  * OK if the time between the backend's have been initialized (and
  69  * cleancache_ops has been set to not NULL) and when the filesystems start
  70  * actually calling the backends. The inverse (when unloading) is obviously
  71  * not good - but this shim does not do that (yet).
  72  */
  73
  74 /*
  75  * The backends and filesystems work all asynchronously. This is b/c the
  76  * backends can be built as modules.
  77  * The usual sequence of events is:
  78  *      a) mount /      -> __cleancache_init_fs is called. We set the
  79  *              [shared_|]fs_poolid_map and uuids for.
  80  *
  81  *      b). user does I/Os -> we call the rest of __cleancache_* functions
  82  *              which return immediately as cleancache_ops is false.
  83  *
  84  *      c). modprobe zcache -> cleancache_register_ops. We init the backend
  85  *              and set cleancache_ops to true, and for any fs_poolid_map
  86  *              (which is set by __cleancache_init_fs) we initialize the poolid.
  87  *
  88  *      d). user does I/Os -> now that cleancache_ops is true all the
  89  *              __cleancache_* functions can call the backend. They all check
  90  *              that fs_poolid_map is valid and if so invoke the backend.
  91  *
  92  *      e). umount /    -> __cleancache_invalidate_fs, the fs_poolid_map is
  93  *              reset (which is the second check in the __cleancache_* ops
  94  *              to call the backend).
  95  *
  96  * The sequence of event could also be c), followed by a), and d). and e). The
  97  * c) would not happen anymore. There is also the chance of c), and one thread
  98  * doing a) + d), and another doing e). For that case we depend on the
  99  * filesystem calling __cleancache_invalidate_fs in the proper sequence (so
 100  * that it handles all I/Os before it invalidates the fs (which is last part
 101  * of unmounting process).
 102  *
 103  * Note: The acute reader will notice that there is no "rmmod zcache" case.
 104  * This is b/c the functionality for that is not yet implemented and when
 105  * done, will require some extra locking not yet devised.
 106  */
 107
 108 /*
 109  * Register operations for cleancache, returning previous thus allowing
 110  * detection of multiple backends and possible nesting.
 111  */
 112 struct cleancache_ops *cleancache_register_ops(struct cleancache_ops *ops)
 113 {
 114         struct cleancache_ops *old = cleancache_ops;
 115         int i;
 116
 117         mutex_lock(&poolid_mutex);
 118         for (i = 0; i < MAX_INITIALIZABLE_FS; i++) {
 119                 if (fs_poolid_map[i] == FS_NO_BACKEND)
 120                         fs_poolid_map[i] = ops->init_fs(PAGE_SIZE);
 121                 if (shared_fs_poolid_map[i] == FS_NO_BACKEND)
 122                         shared_fs_poolid_map[i] = ops->init_shared_fs
 123                                         (uuids[i], PAGE_SIZE);
 124         }
 125         /*
 126          * We MUST set cleancache_ops _after_ we have called the backends
 127          * init_fs or init_shared_fs functions. Otherwise the compiler might
 128          * re-order where cleancache_ops is set in this function.
 129          */
 130         barrier();
 131         cleancache_ops = ops;
 132         mutex_unlock(&poolid_mutex);
 133         return old;
 134 }
 135 EXPORT_SYMBOL(cleancache_register_ops);
 136
 137 /* Called by a cleancache-enabled filesystem at time of mount */
 138 void __cleancache_init_fs(struct super_block *sb)
 139 {
 140         int i;
 141
 142         mutex_lock(&poolid_mutex);
 143         for (i = 0; i < MAX_INITIALIZABLE_FS; i++) {
 144                 if (fs_poolid_map[i] == FS_UNKNOWN) {
 145                         sb->cleancache_poolid = i + FAKE_FS_POOLID_OFFSET;
 146                         if (cleancache_ops)
 147                                 fs_poolid_map[i] = cleancache_ops->init_fs(PAGE_SIZE);
 148                         else
 149                                 fs_poolid_map[i] = FS_NO_BACKEND;
 150                         break;
 151                 }
 152         }
 153         mutex_unlock(&poolid_mutex);
 154 }
 155 EXPORT_SYMBOL(__cleancache_init_fs);
 156
 157 /* Called by a cleancache-enabled clustered filesystem at time of mount */
 158 void __cleancache_init_shared_fs(char *uuid, struct super_block *sb)
 159 {
 160         int i;
 161
 162         mutex_lock(&poolid_mutex);
 163         for (i = 0; i < MAX_INITIALIZABLE_FS; i++) {
 164                 if (shared_fs_poolid_map[i] == FS_UNKNOWN) {
 165                         sb->cleancache_poolid = i + FAKE_SHARED_FS_POOLID_OFFSET;
 166                         uuids[i] = uuid;
 167                         if (cleancache_ops)
 168                                 shared_fs_poolid_map[i] = cleancache_ops->init_shared_fs
 169                                                 (uuid, PAGE_SIZE);
 170                         else
 171                                 shared_fs_poolid_map[i] = FS_NO_BACKEND;
 172                         break;
 173                 }
 174         }
 175         mutex_unlock(&poolid_mutex);
 176 }
 177 EXPORT_SYMBOL(__cleancache_init_shared_fs);
 178
 179 /*
 180  * If the filesystem uses exportable filehandles, use the filehandle as
 181  * the key, else use the inode number.
 182  */
 183 static int cleancache_get_key(struct inode *inode,
 184                               struct cleancache_filekey *key)
 185 {
 186         int (*fhfn)(struct inode *, __u32 *fh, int *, struct inode *);
 187         int len = 0, maxlen = CLEANCACHE_KEY_MAX;
 188         struct super_block *sb = inode->i_sb;
 189
 190         key->u.ino = inode->i_ino;
 191         if (sb->s_export_op != NULL) {
 192                 fhfn = sb->s_export_op->encode_fh;
 193                 if  (fhfn) {
 194                         len = (*fhfn)(inode, &key->u.fh[0], &maxlen, NULL);
 195                         if (len <= FILEID_ROOT || len == FILEID_INVALID)
 196                                 return -1;
 197                         if (maxlen > CLEANCACHE_KEY_MAX)
 198                                 return -1;
 199                 }
 200         }
 201         return 0;
 202 }
 203
 204 /*
 205  * Returns a pool_id that is associated with a given fake poolid.
 206  */
 207 static int get_poolid_from_fake(int fake_pool_id)
 208 {
 209         if (fake_pool_id >= FAKE_SHARED_FS_POOLID_OFFSET)
 210                 return shared_fs_poolid_map[fake_pool_id -
 211                         FAKE_SHARED_FS_POOLID_OFFSET];
 212         else if (fake_pool_id >= FAKE_FS_POOLID_OFFSET)
 213                 return fs_poolid_map[fake_pool_id - FAKE_FS_POOLID_OFFSET];
 214         return FS_NO_BACKEND;
 215 }
 216
 217 /*
 218  * "Get" data from cleancache associated with the poolid/inode/index
 219  * that were specified when the data was put to cleanache and, if
 220  * successful, use it to fill the specified page with data and return 0.
 221  * The pageframe is unchanged and returns -1 if the get fails.
 222  * Page must be locked by caller.
 223  *
 224  * The function has two checks before any action is taken - whether
 225  * a backend is registered and whether the sb->cleancache_poolid
 226  * is correct.
 227  */
 228 int __cleancache_get_page(struct page *page)
 229 {
 230         int ret = -1;
 231         int pool_id;
 232         int fake_pool_id;
 233         struct cleancache_filekey key = { .u.key = { 0 } };
 234
 235         if (!cleancache_ops) {
 236                 cleancache_failed_gets++;
 237                 goto out;
 238         }
 239
 240         VM_BUG_ON_PAGE(!PageLocked(page), page);
 241         fake_pool_id = page->mapping->host->i_sb->cleancache_poolid;
 242         if (fake_pool_id < 0)
 243                 goto out;
 244         pool_id = get_poolid_from_fake(fake_pool_id);
 245
 246         if (cleancache_get_key(page->mapping->host, &key) < 0)
 247                 goto out;
 248
 249         if (pool_id >= 0)
 250                 ret = cleancache_ops->get_page(pool_id,
 251                                 key, page->index, page);
 252         if (ret == 0)
 253                 cleancache_succ_gets++;
 254         else
 255                 cleancache_failed_gets++;
 256 out:
 257         return ret;
 258 }
 259 EXPORT_SYMBOL(__cleancache_get_page);
 260
 261 /*
 262  * "Put" data from a page to cleancache and associate it with the
 263  * (previously-obtained per-filesystem) poolid and the page's,
 264  * inode and page index.  Page must be locked.  Note that a put_page
 265  * always "succeeds", though a subsequent get_page may succeed or fail.
 266  *
 267  * The function has two checks before any action is taken - whether
 268  * a backend is registered and whether the sb->cleancache_poolid
 269  * is correct.
 270  */
 271 void __cleancache_put_page(struct page *page)
 272 {
 273         int pool_id;
 274         int fake_pool_id;
 275         struct cleancache_filekey key = { .u.key = { 0 } };
 276
 277         if (!cleancache_ops) {
 278                 cleancache_puts++;
 279                 return;
 280         }
 281
 282         VM_BUG_ON_PAGE(!PageLocked(page), page);
 283         fake_pool_id = page->mapping->host->i_sb->cleancache_poolid;
 284         if (fake_pool_id < 0)
 285                 return;
 286
 287         pool_id = get_poolid_from_fake(fake_pool_id);
 288
 289         if (pool_id >= 0 &&
 290                 cleancache_get_key(page->mapping->host, &key) >= 0) {
 291                 cleancache_ops->put_page(pool_id, key, page->index, page);
 292                 cleancache_puts++;
 293         }
 294 }
 295 EXPORT_SYMBOL(__cleancache_put_page);
 296
 297 /*
 298  * Invalidate any data from cleancache associated with the poolid and the
 299  * page's inode and page index so that a subsequent "get" will fail.
 300  *
 301  * The function has two checks before any action is taken - whether
 302  * a backend is registered and whether the sb->cleancache_poolid
 303  * is correct.
 304  */
 305 void __cleancache_invalidate_page(struct address_space *mapping,
 306                                         struct page *page)
 307 {
 308         /* careful... page->mapping is NULL sometimes when this is called */
 309         int pool_id;
 310         int fake_pool_id = mapping->host->i_sb->cleancache_poolid;
 311         struct cleancache_filekey key = { .u.key = { 0 } };
 312
 313         if (!cleancache_ops)
 314                 return;
 315
 316         if (fake_pool_id >= 0) {
 317                 pool_id = get_poolid_from_fake(fake_pool_id);
 318                 if (pool_id < 0)
 319                         return;
 320
 321                 VM_BUG_ON_PAGE(!PageLocked(page), page);
 322                 if (cleancache_get_key(mapping->host, &key) >= 0) {
 323                         cleancache_ops->invalidate_page(pool_id,
 324                                         key, page->index);
 325                         cleancache_invalidates++;
 326                 }
 327         }
 328 }
 329 EXPORT_SYMBOL(__cleancache_invalidate_page);
 330
 331 /*
 332  * Invalidate all data from cleancache associated with the poolid and the
 333  * mappings's inode so that all subsequent gets to this poolid/inode
 334  * will fail.
 335  *
 336  * The function has two checks before any action is taken - whether
 337  * a backend is registered and whether the sb->cleancache_poolid
 338  * is correct.
 339  */
 340 void __cleancache_invalidate_inode(struct address_space *mapping)
 341 {
 342         int pool_id;
 343         int fake_pool_id = mapping->host->i_sb->cleancache_poolid;
 344         struct cleancache_filekey key = { .u.key = { 0 } };
 345
 346         if (!cleancache_ops)
 347                 return;
 348
 349         if (fake_pool_id < 0)
 350                 return;
 351
 352         pool_id = get_poolid_from_fake(fake_pool_id);
 353
 354         if (pool_id >= 0 && cleancache_get_key(mapping->host, &key) >= 0)
 355                 cleancache_ops->invalidate_inode(pool_id, key);
 356 }
 357 EXPORT_SYMBOL(__cleancache_invalidate_inode);
 358
 359 /*
 360  * Called by any cleancache-enabled filesystem at time of unmount;
 361  * note that pool_id is surrendered and may be returned by a subsequent
 362  * cleancache_init_fs or cleancache_init_shared_fs.
 363  */
 364 void __cleancache_invalidate_fs(struct super_block *sb)
 365 {
 366         int index;
 367         int fake_pool_id = sb->cleancache_poolid;
 368         int old_poolid = fake_pool_id;
 369
 370         mutex_lock(&poolid_mutex);
 371         if (fake_pool_id >= FAKE_SHARED_FS_POOLID_OFFSET) {
 372                 index = fake_pool_id - FAKE_SHARED_FS_POOLID_OFFSET;
 373                 old_poolid = shared_fs_poolid_map[index];
 374                 shared_fs_poolid_map[index] = FS_UNKNOWN;
 375                 uuids[index] = NULL;
 376         } else if (fake_pool_id >= FAKE_FS_POOLID_OFFSET) {
 377                 index = fake_pool_id - FAKE_FS_POOLID_OFFSET;
 378                 old_poolid = fs_poolid_map[index];
 379                 fs_poolid_map[index] = FS_UNKNOWN;
 380         }
 381         sb->cleancache_poolid = -1;
 382         if (cleancache_ops)
 383                 cleancache_ops->invalidate_fs(old_poolid);
 384         mutex_unlock(&poolid_mutex);
 385 }
 386 EXPORT_SYMBOL(__cleancache_invalidate_fs);
 387
 388 static int __init init_cleancache(void)
 389 {
 390         int i;
 391
 392 #ifdef CONFIG_DEBUG_FS
 393         struct dentry *root = debugfs_create_dir("cleancache", NULL);
 394         if (root == NULL)
 395                 return -ENXIO;
 396         debugfs_create_u64("succ_gets", S_IRUGO, root, &cleancache_succ_gets);
 397         debugfs_create_u64("failed_gets", S_IRUGO,
 398                                 root, &cleancache_failed_gets);
 399         debugfs_create_u64("puts", S_IRUGO, root, &cleancache_puts);
 400         debugfs_create_u64("invalidates", S_IRUGO,
 401                                 root, &cleancache_invalidates);
 402 #endif
 403         for (i = 0; i < MAX_INITIALIZABLE_FS; i++) {
 404                 fs_poolid_map[i] = FS_UNKNOWN;
 405                 shared_fs_poolid_map[i] = FS_UNKNOWN;
 406         }
 407         return 0;
 408 }
 409 module_init(init_cleancache)