| 1 | // SPDX-License-Identifier: GPL-2.0+ | 
|---|
| 2 | /* | 
|---|
| 3 | * linux/fs/jbd2/revoke.c | 
|---|
| 4 | * | 
|---|
| 5 | * Written by Stephen C. Tweedie <sct@redhat.com>, 2000 | 
|---|
| 6 | * | 
|---|
| 7 | * Copyright 2000 Red Hat corp --- All Rights Reserved | 
|---|
| 8 | * | 
|---|
| 9 | * Journal revoke routines for the generic filesystem journaling code; | 
|---|
| 10 | * part of the ext2fs journaling system. | 
|---|
| 11 | * | 
|---|
| 12 | * Revoke is the mechanism used to prevent old log records for deleted | 
|---|
| 13 | * metadata from being replayed on top of newer data using the same | 
|---|
| 14 | * blocks.  The revoke mechanism is used in two separate places: | 
|---|
| 15 | * | 
|---|
| 16 | * + Commit: during commit we write the entire list of the current | 
|---|
| 17 | *   transaction's revoked blocks to the journal | 
|---|
| 18 | * | 
|---|
| 19 | * + Recovery: during recovery we record the transaction ID of all | 
|---|
| 20 | *   revoked blocks.  If there are multiple revoke records in the log | 
|---|
| 21 | *   for a single block, only the last one counts, and if there is a log | 
|---|
| 22 | *   entry for a block beyond the last revoke, then that log entry still | 
|---|
| 23 | *   gets replayed. | 
|---|
| 24 | * | 
|---|
| 25 | * We can get interactions between revokes and new log data within a | 
|---|
| 26 | * single transaction: | 
|---|
| 27 | * | 
|---|
| 28 | * Block is revoked and then journaled: | 
|---|
| 29 | *   The desired end result is the journaling of the new block, so we | 
|---|
| 30 | *   cancel the revoke before the transaction commits. | 
|---|
| 31 | * | 
|---|
| 32 | * Block is journaled and then revoked: | 
|---|
| 33 | *   The revoke must take precedence over the write of the block, so we | 
|---|
| 34 | *   need either to cancel the journal entry or to write the revoke | 
|---|
| 35 | *   later in the log than the log block.  In this case, we choose the | 
|---|
| 36 | *   latter: journaling a block cancels any revoke record for that block | 
|---|
| 37 | *   in the current transaction, so any revoke for that block in the | 
|---|
| 38 | *   transaction must have happened after the block was journaled and so | 
|---|
| 39 | *   the revoke must take precedence. | 
|---|
| 40 | * | 
|---|
| 41 | * Block is revoked and then written as data: | 
|---|
| 42 | *   The data write is allowed to succeed, but the revoke is _not_ | 
|---|
| 43 | *   cancelled.  We still need to prevent old log records from | 
|---|
| 44 | *   overwriting the new data.  We don't even need to clear the revoke | 
|---|
| 45 | *   bit here. | 
|---|
| 46 | * | 
|---|
| 47 | * We cache revoke status of a buffer in the current transaction in b_states | 
|---|
| 48 | * bits.  As the name says, revokevalid flag indicates that the cached revoke | 
|---|
| 49 | * status of a buffer is valid and we can rely on the cached status. | 
|---|
| 50 | * | 
|---|
| 51 | * Revoke information on buffers is a tri-state value: | 
|---|
| 52 | * | 
|---|
| 53 | * RevokeValid clear:	no cached revoke status, need to look it up | 
|---|
| 54 | * RevokeValid set, Revoked clear: | 
|---|
| 55 | *			buffer has not been revoked, and cancel_revoke | 
|---|
| 56 | *			need do nothing. | 
|---|
| 57 | * RevokeValid set, Revoked set: | 
|---|
| 58 | *			buffer has been revoked. | 
|---|
| 59 | * | 
|---|
| 60 | * Locking rules: | 
|---|
| 61 | * We keep two hash tables of revoke records. One hashtable belongs to the | 
|---|
| 62 | * running transaction (is pointed to by journal->j_revoke), the other one | 
|---|
| 63 | * belongs to the committing transaction. Accesses to the second hash table | 
|---|
| 64 | * happen only from the kjournald and no other thread touches this table.  Also | 
|---|
| 65 | * journal_switch_revoke_table() which switches which hashtable belongs to the | 
|---|
| 66 | * running and which to the committing transaction is called only from | 
|---|
| 67 | * kjournald. Therefore we need no locks when accessing the hashtable belonging | 
|---|
| 68 | * to the committing transaction. | 
|---|
| 69 | * | 
|---|
| 70 | * All users operating on the hash table belonging to the running transaction | 
|---|
| 71 | * have a handle to the transaction. Therefore they are safe from kjournald | 
|---|
| 72 | * switching hash tables under them. For operations on the lists of entries in | 
|---|
| 73 | * the hash table j_revoke_lock is used. | 
|---|
| 74 | * | 
|---|
| 75 | * Finally, also replay code uses the hash tables but at this moment no one else | 
|---|
| 76 | * can touch them (filesystem isn't mounted yet) and hence no locking is | 
|---|
| 77 | * needed. | 
|---|
| 78 | */ | 
|---|
| 79 |  | 
|---|
| 80 | #ifndef __KERNEL__ | 
|---|
| 81 | #include "jfs_user.h" | 
|---|
| 82 | #else | 
|---|
| 83 | #include <linux/time.h> | 
|---|
| 84 | #include <linux/fs.h> | 
|---|
| 85 | #include <linux/jbd2.h> | 
|---|
| 86 | #include <linux/errno.h> | 
|---|
| 87 | #include <linux/slab.h> | 
|---|
| 88 | #include <linux/list.h> | 
|---|
| 89 | #include <linux/init.h> | 
|---|
| 90 | #include <linux/bio.h> | 
|---|
| 91 | #include <linux/log2.h> | 
|---|
| 92 | #include <linux/hash.h> | 
|---|
| 93 | #endif | 
|---|
| 94 |  | 
|---|
| 95 | static struct kmem_cache *jbd2_revoke_record_cache; | 
|---|
| 96 | static struct kmem_cache *jbd2_revoke_table_cache; | 
|---|
| 97 |  | 
|---|
| 98 | /* Each revoke record represents one single revoked block.  During | 
|---|
| 99 | journal replay, this involves recording the transaction ID of the | 
|---|
| 100 | last transaction to revoke this block. */ | 
|---|
| 101 |  | 
|---|
| 102 | struct jbd2_revoke_record_s | 
|---|
| 103 | { | 
|---|
| 104 | struct list_head  hash; | 
|---|
| 105 | tid_t		  sequence;	/* Used for recovery only */ | 
|---|
| 106 | unsigned long long	  blocknr; | 
|---|
| 107 | }; | 
|---|
| 108 |  | 
|---|
| 109 |  | 
|---|
| 110 | /* The revoke table is just a simple hash table of revoke records. */ | 
|---|
| 111 | struct jbd2_revoke_table_s | 
|---|
| 112 | { | 
|---|
| 113 | /* It is conceivable that we might want a larger hash table | 
|---|
| 114 | * for recovery.  Must be a power of two. */ | 
|---|
| 115 | int		  hash_size; | 
|---|
| 116 | int		  hash_shift; | 
|---|
| 117 | struct list_head *hash_table; | 
|---|
| 118 | }; | 
|---|
| 119 |  | 
|---|
| 120 |  | 
|---|
| 121 | #ifdef __KERNEL__ | 
|---|
| 122 | static void write_one_revoke_record(transaction_t *, | 
|---|
| 123 | struct list_head *, | 
|---|
| 124 | struct buffer_head **, int *, | 
|---|
| 125 | struct jbd2_revoke_record_s *); | 
|---|
| 126 | static void flush_descriptor(journal_t *, struct buffer_head *, int); | 
|---|
| 127 | #endif | 
|---|
| 128 |  | 
|---|
| 129 | /* Utility functions to maintain the revoke table */ | 
|---|
| 130 |  | 
|---|
| 131 | static inline int hash(journal_t *journal, unsigned long long block) | 
|---|
| 132 | { | 
|---|
| 133 | return hash_64(val: block, bits: journal->j_revoke->hash_shift); | 
|---|
| 134 | } | 
|---|
| 135 |  | 
|---|
| 136 | static int insert_revoke_hash(journal_t *journal, unsigned long long blocknr, | 
|---|
| 137 | tid_t seq) | 
|---|
| 138 | { | 
|---|
| 139 | struct list_head *hash_list; | 
|---|
| 140 | struct jbd2_revoke_record_s *record; | 
|---|
| 141 | gfp_t gfp_mask = GFP_NOFS; | 
|---|
| 142 |  | 
|---|
| 143 | if (journal_oom_retry) | 
|---|
| 144 | gfp_mask |= __GFP_NOFAIL; | 
|---|
| 145 | record = kmem_cache_alloc(jbd2_revoke_record_cache, gfp_mask); | 
|---|
| 146 | if (!record) | 
|---|
| 147 | return -ENOMEM; | 
|---|
| 148 |  | 
|---|
| 149 | record->sequence = seq; | 
|---|
| 150 | record->blocknr = blocknr; | 
|---|
| 151 | hash_list = &journal->j_revoke->hash_table[hash(journal, block: blocknr)]; | 
|---|
| 152 | spin_lock(lock: &journal->j_revoke_lock); | 
|---|
| 153 | list_add(new: &record->hash, head: hash_list); | 
|---|
| 154 | spin_unlock(lock: &journal->j_revoke_lock); | 
|---|
| 155 | return 0; | 
|---|
| 156 | } | 
|---|
| 157 |  | 
|---|
| 158 | /* Find a revoke record in the journal's hash table. */ | 
|---|
| 159 |  | 
|---|
| 160 | static struct jbd2_revoke_record_s *find_revoke_record(journal_t *journal, | 
|---|
| 161 | unsigned long long blocknr) | 
|---|
| 162 | { | 
|---|
| 163 | struct list_head *hash_list; | 
|---|
| 164 | struct jbd2_revoke_record_s *record; | 
|---|
| 165 |  | 
|---|
| 166 | hash_list = &journal->j_revoke->hash_table[hash(journal, block: blocknr)]; | 
|---|
| 167 |  | 
|---|
| 168 | spin_lock(lock: &journal->j_revoke_lock); | 
|---|
| 169 | record = (struct jbd2_revoke_record_s *) hash_list->next; | 
|---|
| 170 | while (&(record->hash) != hash_list) { | 
|---|
| 171 | if (record->blocknr == blocknr) { | 
|---|
| 172 | spin_unlock(lock: &journal->j_revoke_lock); | 
|---|
| 173 | return record; | 
|---|
| 174 | } | 
|---|
| 175 | record = (struct jbd2_revoke_record_s *) record->hash.next; | 
|---|
| 176 | } | 
|---|
| 177 | spin_unlock(lock: &journal->j_revoke_lock); | 
|---|
| 178 | return NULL; | 
|---|
| 179 | } | 
|---|
| 180 |  | 
|---|
| 181 | void jbd2_journal_destroy_revoke_record_cache(void) | 
|---|
| 182 | { | 
|---|
| 183 | kmem_cache_destroy(s: jbd2_revoke_record_cache); | 
|---|
| 184 | jbd2_revoke_record_cache = NULL; | 
|---|
| 185 | } | 
|---|
| 186 |  | 
|---|
| 187 | void jbd2_journal_destroy_revoke_table_cache(void) | 
|---|
| 188 | { | 
|---|
| 189 | kmem_cache_destroy(s: jbd2_revoke_table_cache); | 
|---|
| 190 | jbd2_revoke_table_cache = NULL; | 
|---|
| 191 | } | 
|---|
| 192 |  | 
|---|
| 193 | int __init jbd2_journal_init_revoke_record_cache(void) | 
|---|
| 194 | { | 
|---|
| 195 | J_ASSERT(!jbd2_revoke_record_cache); | 
|---|
| 196 | jbd2_revoke_record_cache = KMEM_CACHE(jbd2_revoke_record_s, | 
|---|
| 197 | SLAB_HWCACHE_ALIGN|SLAB_TEMPORARY); | 
|---|
| 198 |  | 
|---|
| 199 | if (!jbd2_revoke_record_cache) { | 
|---|
| 200 | pr_emerg( "JBD2: failed to create revoke_record cache\n"); | 
|---|
| 201 | return -ENOMEM; | 
|---|
| 202 | } | 
|---|
| 203 | return 0; | 
|---|
| 204 | } | 
|---|
| 205 |  | 
|---|
| 206 | int __init jbd2_journal_init_revoke_table_cache(void) | 
|---|
| 207 | { | 
|---|
| 208 | J_ASSERT(!jbd2_revoke_table_cache); | 
|---|
| 209 | jbd2_revoke_table_cache = KMEM_CACHE(jbd2_revoke_table_s, | 
|---|
| 210 | SLAB_TEMPORARY); | 
|---|
| 211 | if (!jbd2_revoke_table_cache) { | 
|---|
| 212 | pr_emerg( "JBD2: failed to create revoke_table cache\n"); | 
|---|
| 213 | return -ENOMEM; | 
|---|
| 214 | } | 
|---|
| 215 | return 0; | 
|---|
| 216 | } | 
|---|
| 217 |  | 
|---|
| 218 | struct jbd2_revoke_table_s *jbd2_journal_init_revoke_table(int hash_size) | 
|---|
| 219 | { | 
|---|
| 220 | int shift = 0; | 
|---|
| 221 | int tmp = hash_size; | 
|---|
| 222 | struct jbd2_revoke_table_s *table; | 
|---|
| 223 |  | 
|---|
| 224 | table = kmem_cache_alloc(jbd2_revoke_table_cache, GFP_KERNEL); | 
|---|
| 225 | if (!table) | 
|---|
| 226 | goto out; | 
|---|
| 227 |  | 
|---|
| 228 | while((tmp >>= 1UL) != 0UL) | 
|---|
| 229 | shift++; | 
|---|
| 230 |  | 
|---|
| 231 | table->hash_size = hash_size; | 
|---|
| 232 | table->hash_shift = shift; | 
|---|
| 233 | table->hash_table = | 
|---|
| 234 | kvmalloc_array(hash_size, sizeof(struct list_head), GFP_KERNEL); | 
|---|
| 235 | if (!table->hash_table) { | 
|---|
| 236 | kmem_cache_free(s: jbd2_revoke_table_cache, objp: table); | 
|---|
| 237 | table = NULL; | 
|---|
| 238 | goto out; | 
|---|
| 239 | } | 
|---|
| 240 |  | 
|---|
| 241 | for (tmp = 0; tmp < hash_size; tmp++) | 
|---|
| 242 | INIT_LIST_HEAD(list: &table->hash_table[tmp]); | 
|---|
| 243 |  | 
|---|
| 244 | out: | 
|---|
| 245 | return table; | 
|---|
| 246 | } | 
|---|
| 247 |  | 
|---|
| 248 | void jbd2_journal_destroy_revoke_table(struct jbd2_revoke_table_s *table) | 
|---|
| 249 | { | 
|---|
| 250 | int i; | 
|---|
| 251 | struct list_head *hash_list; | 
|---|
| 252 |  | 
|---|
| 253 | for (i = 0; i < table->hash_size; i++) { | 
|---|
| 254 | hash_list = &table->hash_table[i]; | 
|---|
| 255 | J_ASSERT(list_empty(hash_list)); | 
|---|
| 256 | } | 
|---|
| 257 |  | 
|---|
| 258 | kvfree(addr: table->hash_table); | 
|---|
| 259 | kmem_cache_free(s: jbd2_revoke_table_cache, objp: table); | 
|---|
| 260 | } | 
|---|
| 261 |  | 
|---|
| 262 | /* Initialise the revoke table for a given journal to a given size. */ | 
|---|
| 263 | int jbd2_journal_init_revoke(journal_t *journal, int hash_size) | 
|---|
| 264 | { | 
|---|
| 265 | J_ASSERT(journal->j_revoke_table[0] == NULL); | 
|---|
| 266 | J_ASSERT(is_power_of_2(hash_size)); | 
|---|
| 267 |  | 
|---|
| 268 | journal->j_revoke_table[0] = jbd2_journal_init_revoke_table(hash_size); | 
|---|
| 269 | if (!journal->j_revoke_table[0]) | 
|---|
| 270 | goto fail0; | 
|---|
| 271 |  | 
|---|
| 272 | journal->j_revoke_table[1] = jbd2_journal_init_revoke_table(hash_size); | 
|---|
| 273 | if (!journal->j_revoke_table[1]) | 
|---|
| 274 | goto fail1; | 
|---|
| 275 |  | 
|---|
| 276 | journal->j_revoke = journal->j_revoke_table[1]; | 
|---|
| 277 |  | 
|---|
| 278 | spin_lock_init(&journal->j_revoke_lock); | 
|---|
| 279 |  | 
|---|
| 280 | return 0; | 
|---|
| 281 |  | 
|---|
| 282 | fail1: | 
|---|
| 283 | jbd2_journal_destroy_revoke_table(table: journal->j_revoke_table[0]); | 
|---|
| 284 | journal->j_revoke_table[0] = NULL; | 
|---|
| 285 | fail0: | 
|---|
| 286 | return -ENOMEM; | 
|---|
| 287 | } | 
|---|
| 288 |  | 
|---|
| 289 | /* Destroy a journal's revoke table.  The table must already be empty! */ | 
|---|
| 290 | void jbd2_journal_destroy_revoke(journal_t *journal) | 
|---|
| 291 | { | 
|---|
| 292 | journal->j_revoke = NULL; | 
|---|
| 293 | if (journal->j_revoke_table[0]) | 
|---|
| 294 | jbd2_journal_destroy_revoke_table(table: journal->j_revoke_table[0]); | 
|---|
| 295 | if (journal->j_revoke_table[1]) | 
|---|
| 296 | jbd2_journal_destroy_revoke_table(table: journal->j_revoke_table[1]); | 
|---|
| 297 | } | 
|---|
| 298 |  | 
|---|
| 299 |  | 
|---|
| 300 | #ifdef __KERNEL__ | 
|---|
| 301 |  | 
|---|
| 302 | /* | 
|---|
| 303 | * jbd2_journal_revoke: revoke a given buffer_head from the journal.  This | 
|---|
| 304 | * prevents the block from being replayed during recovery if we take a | 
|---|
| 305 | * crash after this current transaction commits.  Any subsequent | 
|---|
| 306 | * metadata writes of the buffer in this transaction cancel the | 
|---|
| 307 | * revoke. | 
|---|
| 308 | * | 
|---|
| 309 | * Note that this call may block --- it is up to the caller to make | 
|---|
| 310 | * sure that there are no further calls to journal_write_metadata | 
|---|
| 311 | * before the revoke is complete.  In ext3, this implies calling the | 
|---|
| 312 | * revoke before clearing the block bitmap when we are deleting | 
|---|
| 313 | * metadata. | 
|---|
| 314 | * | 
|---|
| 315 | * Revoke performs a jbd2_journal_forget on any buffer_head passed in as a | 
|---|
| 316 | * parameter, but does _not_ forget the buffer_head if the bh was only | 
|---|
| 317 | * found implicitly. | 
|---|
| 318 | * | 
|---|
| 319 | * bh_in may not be a journalled buffer - it may have come off | 
|---|
| 320 | * the hash tables without an attached journal_head. | 
|---|
| 321 | * | 
|---|
| 322 | * If bh_in is non-zero, jbd2_journal_revoke() will decrement its b_count | 
|---|
| 323 | * by one. | 
|---|
| 324 | */ | 
|---|
| 325 |  | 
|---|
| 326 | int jbd2_journal_revoke(handle_t *handle, unsigned long long blocknr, | 
|---|
| 327 | struct buffer_head *bh_in) | 
|---|
| 328 | { | 
|---|
| 329 | struct buffer_head *bh = NULL; | 
|---|
| 330 | journal_t *journal; | 
|---|
| 331 | struct block_device *bdev; | 
|---|
| 332 | int err; | 
|---|
| 333 |  | 
|---|
| 334 | might_sleep(); | 
|---|
| 335 | if (bh_in) | 
|---|
| 336 | BUFFER_TRACE(bh_in, "enter"); | 
|---|
| 337 |  | 
|---|
| 338 | journal = handle->h_transaction->t_journal; | 
|---|
| 339 | if (!jbd2_journal_set_features(journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)){ | 
|---|
| 340 | J_ASSERT (! "Cannot set revoke feature!"); | 
|---|
| 341 | return -EINVAL; | 
|---|
| 342 | } | 
|---|
| 343 |  | 
|---|
| 344 | bdev = journal->j_fs_dev; | 
|---|
| 345 | bh = bh_in; | 
|---|
| 346 |  | 
|---|
| 347 | if (!bh) { | 
|---|
| 348 | bh = __find_get_block_nonatomic(bdev, block: blocknr, | 
|---|
| 349 | size: journal->j_blocksize); | 
|---|
| 350 | if (bh) | 
|---|
| 351 | BUFFER_TRACE(bh, "found on hash"); | 
|---|
| 352 | } | 
|---|
| 353 | #ifdef JBD2_EXPENSIVE_CHECKING | 
|---|
| 354 | else { | 
|---|
| 355 | struct buffer_head *bh2; | 
|---|
| 356 |  | 
|---|
| 357 | /* If there is a different buffer_head lying around in | 
|---|
| 358 | * memory anywhere... */ | 
|---|
| 359 | bh2 = __find_get_block_nonatomic(bdev, blocknr, | 
|---|
| 360 | journal->j_blocksize); | 
|---|
| 361 | if (bh2) { | 
|---|
| 362 | /* ... and it has RevokeValid status... */ | 
|---|
| 363 | if (bh2 != bh && buffer_revokevalid(bh2)) | 
|---|
| 364 | /* ...then it better be revoked too, | 
|---|
| 365 | * since it's illegal to create a revoke | 
|---|
| 366 | * record against a buffer_head which is | 
|---|
| 367 | * not marked revoked --- that would | 
|---|
| 368 | * risk missing a subsequent revoke | 
|---|
| 369 | * cancel. */ | 
|---|
| 370 | J_ASSERT_BH(bh2, buffer_revoked(bh2)); | 
|---|
| 371 | put_bh(bh2); | 
|---|
| 372 | } | 
|---|
| 373 | } | 
|---|
| 374 | #endif | 
|---|
| 375 |  | 
|---|
| 376 | if (WARN_ON_ONCE(handle->h_revoke_credits <= 0)) { | 
|---|
| 377 | if (!bh_in) | 
|---|
| 378 | brelse(bh); | 
|---|
| 379 | return -EIO; | 
|---|
| 380 | } | 
|---|
| 381 | /* We really ought not ever to revoke twice in a row without | 
|---|
| 382 | first having the revoke cancelled: it's illegal to free a | 
|---|
| 383 | block twice without allocating it in between! */ | 
|---|
| 384 | if (bh) { | 
|---|
| 385 | if (!J_EXPECT_BH(bh, !buffer_revoked(bh), | 
|---|
| 386 | "inconsistent data on disk")) { | 
|---|
| 387 | if (!bh_in) | 
|---|
| 388 | brelse(bh); | 
|---|
| 389 | return -EIO; | 
|---|
| 390 | } | 
|---|
| 391 | set_buffer_revoked(bh); | 
|---|
| 392 | set_buffer_revokevalid(bh); | 
|---|
| 393 | if (bh_in) { | 
|---|
| 394 | BUFFER_TRACE(bh_in, "call jbd2_journal_forget"); | 
|---|
| 395 | jbd2_journal_forget(handle, bh_in); | 
|---|
| 396 | } else { | 
|---|
| 397 | BUFFER_TRACE(bh, "call brelse"); | 
|---|
| 398 | __brelse(bh); | 
|---|
| 399 | } | 
|---|
| 400 | } | 
|---|
| 401 | handle->h_revoke_credits--; | 
|---|
| 402 |  | 
|---|
| 403 | jbd2_debug(2, "insert revoke for block %llu, bh_in=%p\n",blocknr, bh_in); | 
|---|
| 404 | err = insert_revoke_hash(journal, blocknr, | 
|---|
| 405 | seq: handle->h_transaction->t_tid); | 
|---|
| 406 | BUFFER_TRACE(bh_in, "exit"); | 
|---|
| 407 | return err; | 
|---|
| 408 | } | 
|---|
| 409 |  | 
|---|
| 410 | /* | 
|---|
| 411 | * Cancel an outstanding revoke.  For use only internally by the | 
|---|
| 412 | * journaling code (called from jbd2_journal_get_write_access). | 
|---|
| 413 | * | 
|---|
| 414 | * We trust buffer_revoked() on the buffer if the buffer is already | 
|---|
| 415 | * being journaled: if there is no revoke pending on the buffer, then we | 
|---|
| 416 | * don't do anything here. | 
|---|
| 417 | * | 
|---|
| 418 | * This would break if it were possible for a buffer to be revoked and | 
|---|
| 419 | * discarded, and then reallocated within the same transaction.  In such | 
|---|
| 420 | * a case we would have lost the revoked bit, but when we arrived here | 
|---|
| 421 | * the second time we would still have a pending revoke to cancel.  So, | 
|---|
| 422 | * do not trust the Revoked bit on buffers unless RevokeValid is also | 
|---|
| 423 | * set. | 
|---|
| 424 | */ | 
|---|
| 425 | void jbd2_journal_cancel_revoke(handle_t *handle, struct journal_head *jh) | 
|---|
| 426 | { | 
|---|
| 427 | struct jbd2_revoke_record_s *record; | 
|---|
| 428 | journal_t *journal = handle->h_transaction->t_journal; | 
|---|
| 429 | int need_cancel; | 
|---|
| 430 | struct buffer_head *bh = jh2bh(jh); | 
|---|
| 431 |  | 
|---|
| 432 | jbd2_debug(4, "journal_head %p, cancelling revoke\n", jh); | 
|---|
| 433 |  | 
|---|
| 434 | /* Is the existing Revoke bit valid?  If so, we trust it, and | 
|---|
| 435 | * only perform the full cancel if the revoke bit is set.  If | 
|---|
| 436 | * not, we can't trust the revoke bit, and we need to do the | 
|---|
| 437 | * full search for a revoke record. */ | 
|---|
| 438 | if (test_set_buffer_revokevalid(bh)) { | 
|---|
| 439 | need_cancel = test_clear_buffer_revoked(bh); | 
|---|
| 440 | } else { | 
|---|
| 441 | need_cancel = 1; | 
|---|
| 442 | clear_buffer_revoked(bh); | 
|---|
| 443 | } | 
|---|
| 444 |  | 
|---|
| 445 | if (need_cancel) { | 
|---|
| 446 | record = find_revoke_record(journal, blocknr: bh->b_blocknr); | 
|---|
| 447 | if (record) { | 
|---|
| 448 | jbd2_debug(4, "cancelled existing revoke on " | 
|---|
| 449 | "blocknr %llu\n", (unsigned long long)bh->b_blocknr); | 
|---|
| 450 | spin_lock(lock: &journal->j_revoke_lock); | 
|---|
| 451 | list_del(entry: &record->hash); | 
|---|
| 452 | spin_unlock(lock: &journal->j_revoke_lock); | 
|---|
| 453 | kmem_cache_free(s: jbd2_revoke_record_cache, objp: record); | 
|---|
| 454 | } | 
|---|
| 455 | } | 
|---|
| 456 |  | 
|---|
| 457 | #ifdef JBD2_EXPENSIVE_CHECKING | 
|---|
| 458 | /* There better not be one left behind by now! */ | 
|---|
| 459 | record = find_revoke_record(journal, bh->b_blocknr); | 
|---|
| 460 | J_ASSERT_JH(jh, record == NULL); | 
|---|
| 461 | #endif | 
|---|
| 462 |  | 
|---|
| 463 | /* Finally, have we just cleared revoke on an unhashed | 
|---|
| 464 | * buffer_head?  If so, we'd better make sure we clear the | 
|---|
| 465 | * revoked status on any hashed alias too, otherwise the revoke | 
|---|
| 466 | * state machine will get very upset later on. */ | 
|---|
| 467 | if (need_cancel) { | 
|---|
| 468 | struct buffer_head *bh2; | 
|---|
| 469 | bh2 = __find_get_block_nonatomic(bdev: bh->b_bdev, block: bh->b_blocknr, | 
|---|
| 470 | size: bh->b_size); | 
|---|
| 471 | if (bh2) { | 
|---|
| 472 | if (bh2 != bh) | 
|---|
| 473 | clear_buffer_revoked(bh: bh2); | 
|---|
| 474 | __brelse(bh2); | 
|---|
| 475 | } | 
|---|
| 476 | } | 
|---|
| 477 | } | 
|---|
| 478 |  | 
|---|
| 479 | /* | 
|---|
| 480 | * jbd2_clear_buffer_revoked_flags clears revoked flag of buffers in | 
|---|
| 481 | * revoke table to reflect there is no revoked buffers in the next | 
|---|
| 482 | * transaction which is going to be started. | 
|---|
| 483 | */ | 
|---|
| 484 | void jbd2_clear_buffer_revoked_flags(journal_t *journal) | 
|---|
| 485 | { | 
|---|
| 486 | struct jbd2_revoke_table_s *revoke = journal->j_revoke; | 
|---|
| 487 | int i = 0; | 
|---|
| 488 |  | 
|---|
| 489 | for (i = 0; i < revoke->hash_size; i++) { | 
|---|
| 490 | struct list_head *hash_list; | 
|---|
| 491 | struct list_head *list_entry; | 
|---|
| 492 | hash_list = &revoke->hash_table[i]; | 
|---|
| 493 |  | 
|---|
| 494 | list_for_each(list_entry, hash_list) { | 
|---|
| 495 | struct jbd2_revoke_record_s *record; | 
|---|
| 496 | struct buffer_head *bh; | 
|---|
| 497 | record = (struct jbd2_revoke_record_s *)list_entry; | 
|---|
| 498 | bh = __find_get_block_nonatomic(bdev: journal->j_fs_dev, | 
|---|
| 499 | block: record->blocknr, | 
|---|
| 500 | size: journal->j_blocksize); | 
|---|
| 501 | if (bh) { | 
|---|
| 502 | clear_buffer_revoked(bh); | 
|---|
| 503 | __brelse(bh); | 
|---|
| 504 | } | 
|---|
| 505 | } | 
|---|
| 506 | } | 
|---|
| 507 | } | 
|---|
| 508 |  | 
|---|
| 509 | /* jbd2_journal_switch_revoke_table table select j_revoke for next | 
|---|
| 510 | * transaction we do not want to suspend any processing until all | 
|---|
| 511 | * revokes are written -bzzz | 
|---|
| 512 | */ | 
|---|
| 513 | void jbd2_journal_switch_revoke_table(journal_t *journal) | 
|---|
| 514 | { | 
|---|
| 515 | int i; | 
|---|
| 516 |  | 
|---|
| 517 | if (journal->j_revoke == journal->j_revoke_table[0]) | 
|---|
| 518 | journal->j_revoke = journal->j_revoke_table[1]; | 
|---|
| 519 | else | 
|---|
| 520 | journal->j_revoke = journal->j_revoke_table[0]; | 
|---|
| 521 |  | 
|---|
| 522 | for (i = 0; i < journal->j_revoke->hash_size; i++) | 
|---|
| 523 | INIT_LIST_HEAD(list: &journal->j_revoke->hash_table[i]); | 
|---|
| 524 | } | 
|---|
| 525 |  | 
|---|
| 526 | /* | 
|---|
| 527 | * Write revoke records to the journal for all entries in the current | 
|---|
| 528 | * revoke hash, deleting the entries as we go. | 
|---|
| 529 | */ | 
|---|
| 530 | void jbd2_journal_write_revoke_records(transaction_t *transaction, | 
|---|
| 531 | struct list_head *log_bufs) | 
|---|
| 532 | { | 
|---|
| 533 | journal_t *journal = transaction->t_journal; | 
|---|
| 534 | struct buffer_head *descriptor; | 
|---|
| 535 | struct jbd2_revoke_record_s *record; | 
|---|
| 536 | struct jbd2_revoke_table_s *revoke; | 
|---|
| 537 | struct list_head *hash_list; | 
|---|
| 538 | int i, offset, count; | 
|---|
| 539 |  | 
|---|
| 540 | descriptor = NULL; | 
|---|
| 541 | offset = 0; | 
|---|
| 542 | count = 0; | 
|---|
| 543 |  | 
|---|
| 544 | /* select revoke table for committing transaction */ | 
|---|
| 545 | revoke = journal->j_revoke == journal->j_revoke_table[0] ? | 
|---|
| 546 | journal->j_revoke_table[1] : journal->j_revoke_table[0]; | 
|---|
| 547 |  | 
|---|
| 548 | for (i = 0; i < revoke->hash_size; i++) { | 
|---|
| 549 | hash_list = &revoke->hash_table[i]; | 
|---|
| 550 |  | 
|---|
| 551 | while (!list_empty(head: hash_list)) { | 
|---|
| 552 | record = (struct jbd2_revoke_record_s *) | 
|---|
| 553 | hash_list->next; | 
|---|
| 554 | write_one_revoke_record(transaction, log_bufs, | 
|---|
| 555 | &descriptor, &offset, record); | 
|---|
| 556 | count++; | 
|---|
| 557 | list_del(entry: &record->hash); | 
|---|
| 558 | kmem_cache_free(s: jbd2_revoke_record_cache, objp: record); | 
|---|
| 559 | } | 
|---|
| 560 | } | 
|---|
| 561 | if (descriptor) | 
|---|
| 562 | flush_descriptor(journal, descriptor, offset); | 
|---|
| 563 | jbd2_debug(1, "Wrote %d revoke records\n", count); | 
|---|
| 564 | } | 
|---|
| 565 |  | 
|---|
| 566 | /* | 
|---|
| 567 | * Write out one revoke record.  We need to create a new descriptor | 
|---|
| 568 | * block if the old one is full or if we have not already created one. | 
|---|
| 569 | */ | 
|---|
| 570 |  | 
|---|
| 571 | static void write_one_revoke_record(transaction_t *transaction, | 
|---|
| 572 | struct list_head *log_bufs, | 
|---|
| 573 | struct buffer_head **descriptorp, | 
|---|
| 574 | int *offsetp, | 
|---|
| 575 | struct jbd2_revoke_record_s *record) | 
|---|
| 576 | { | 
|---|
| 577 | journal_t *journal = transaction->t_journal; | 
|---|
| 578 | int csum_size = 0; | 
|---|
| 579 | struct buffer_head *descriptor; | 
|---|
| 580 | int sz, offset; | 
|---|
| 581 |  | 
|---|
| 582 | /* If we are already aborting, this all becomes a noop.  We | 
|---|
| 583 | still need to go round the loop in | 
|---|
| 584 | jbd2_journal_write_revoke_records in order to free all of the | 
|---|
| 585 | revoke records: only the IO to the journal is omitted. */ | 
|---|
| 586 | if (is_journal_aborted(journal)) | 
|---|
| 587 | return; | 
|---|
| 588 |  | 
|---|
| 589 | descriptor = *descriptorp; | 
|---|
| 590 | offset = *offsetp; | 
|---|
| 591 |  | 
|---|
| 592 | /* Do we need to leave space at the end for a checksum? */ | 
|---|
| 593 | if (jbd2_journal_has_csum_v2or3(journal)) | 
|---|
| 594 | csum_size = sizeof(struct jbd2_journal_block_tail); | 
|---|
| 595 |  | 
|---|
| 596 | if (jbd2_has_feature_64bit(j: journal)) | 
|---|
| 597 | sz = 8; | 
|---|
| 598 | else | 
|---|
| 599 | sz = 4; | 
|---|
| 600 |  | 
|---|
| 601 | /* Make sure we have a descriptor with space left for the record */ | 
|---|
| 602 | if (descriptor) { | 
|---|
| 603 | if (offset + sz > journal->j_blocksize - csum_size) { | 
|---|
| 604 | flush_descriptor(journal, descriptor, offset); | 
|---|
| 605 | descriptor = NULL; | 
|---|
| 606 | } | 
|---|
| 607 | } | 
|---|
| 608 |  | 
|---|
| 609 | if (!descriptor) { | 
|---|
| 610 | descriptor = jbd2_journal_get_descriptor_buffer(transaction, | 
|---|
| 611 | JBD2_REVOKE_BLOCK); | 
|---|
| 612 | if (!descriptor) | 
|---|
| 613 | return; | 
|---|
| 614 |  | 
|---|
| 615 | /* Record it so that we can wait for IO completion later */ | 
|---|
| 616 | BUFFER_TRACE(descriptor, "file in log_bufs"); | 
|---|
| 617 | jbd2_file_log_bh(head: log_bufs, bh: descriptor); | 
|---|
| 618 |  | 
|---|
| 619 | offset = sizeof(jbd2_journal_revoke_header_t); | 
|---|
| 620 | *descriptorp = descriptor; | 
|---|
| 621 | } | 
|---|
| 622 |  | 
|---|
| 623 | if (jbd2_has_feature_64bit(j: journal)) | 
|---|
| 624 | * ((__be64 *)(&descriptor->b_data[offset])) = | 
|---|
| 625 | cpu_to_be64(record->blocknr); | 
|---|
| 626 | else | 
|---|
| 627 | * ((__be32 *)(&descriptor->b_data[offset])) = | 
|---|
| 628 | cpu_to_be32(record->blocknr); | 
|---|
| 629 | offset += sz; | 
|---|
| 630 |  | 
|---|
| 631 | *offsetp = offset; | 
|---|
| 632 | } | 
|---|
| 633 |  | 
|---|
| 634 | /* | 
|---|
| 635 | * Flush a revoke descriptor out to the journal.  If we are aborting, | 
|---|
| 636 | * this is a noop; otherwise we are generating a buffer which needs to | 
|---|
| 637 | * be waited for during commit, so it has to go onto the appropriate | 
|---|
| 638 | * journal buffer list. | 
|---|
| 639 | */ | 
|---|
| 640 |  | 
|---|
| 641 | static void flush_descriptor(journal_t *journal, | 
|---|
| 642 | struct buffer_head *descriptor, | 
|---|
| 643 | int offset) | 
|---|
| 644 | { | 
|---|
| 645 | jbd2_journal_revoke_header_t *; | 
|---|
| 646 |  | 
|---|
| 647 | if (is_journal_aborted(journal)) | 
|---|
| 648 | return; | 
|---|
| 649 |  | 
|---|
| 650 | header = (jbd2_journal_revoke_header_t *)descriptor->b_data; | 
|---|
| 651 | header->r_count = cpu_to_be32(offset); | 
|---|
| 652 | jbd2_descriptor_block_csum_set(journal, descriptor); | 
|---|
| 653 |  | 
|---|
| 654 | set_buffer_jwrite(descriptor); | 
|---|
| 655 | BUFFER_TRACE(descriptor, "write"); | 
|---|
| 656 | set_buffer_dirty(descriptor); | 
|---|
| 657 | write_dirty_buffer(bh: descriptor, JBD2_JOURNAL_REQ_FLAGS); | 
|---|
| 658 | } | 
|---|
| 659 | #endif | 
|---|
| 660 |  | 
|---|
| 661 | /* | 
|---|
| 662 | * Revoke support for recovery. | 
|---|
| 663 | * | 
|---|
| 664 | * Recovery needs to be able to: | 
|---|
| 665 | * | 
|---|
| 666 | *  record all revoke records, including the tid of the latest instance | 
|---|
| 667 | *  of each revoke in the journal | 
|---|
| 668 | * | 
|---|
| 669 | *  check whether a given block in a given transaction should be replayed | 
|---|
| 670 | *  (ie. has not been revoked by a revoke record in that or a subsequent | 
|---|
| 671 | *  transaction) | 
|---|
| 672 | * | 
|---|
| 673 | *  empty the revoke table after recovery. | 
|---|
| 674 | */ | 
|---|
| 675 |  | 
|---|
| 676 | /* | 
|---|
| 677 | * First, setting revoke records.  We create a new revoke record for | 
|---|
| 678 | * every block ever revoked in the log as we scan it for recovery, and | 
|---|
| 679 | * we update the existing records if we find multiple revokes for a | 
|---|
| 680 | * single block. | 
|---|
| 681 | */ | 
|---|
| 682 |  | 
|---|
| 683 | int jbd2_journal_set_revoke(journal_t *journal, | 
|---|
| 684 | unsigned long long blocknr, | 
|---|
| 685 | tid_t sequence) | 
|---|
| 686 | { | 
|---|
| 687 | struct jbd2_revoke_record_s *record; | 
|---|
| 688 |  | 
|---|
| 689 | record = find_revoke_record(journal, blocknr); | 
|---|
| 690 | if (record) { | 
|---|
| 691 | /* If we have multiple occurrences, only record the | 
|---|
| 692 | * latest sequence number in the hashed record */ | 
|---|
| 693 | if (tid_gt(x: sequence, y: record->sequence)) | 
|---|
| 694 | record->sequence = sequence; | 
|---|
| 695 | return 0; | 
|---|
| 696 | } | 
|---|
| 697 | return insert_revoke_hash(journal, blocknr, seq: sequence); | 
|---|
| 698 | } | 
|---|
| 699 |  | 
|---|
| 700 | /* | 
|---|
| 701 | * Test revoke records.  For a given block referenced in the log, has | 
|---|
| 702 | * that block been revoked?  A revoke record with a given transaction | 
|---|
| 703 | * sequence number revokes all blocks in that transaction and earlier | 
|---|
| 704 | * ones, but later transactions still need replayed. | 
|---|
| 705 | */ | 
|---|
| 706 |  | 
|---|
| 707 | int jbd2_journal_test_revoke(journal_t *journal, | 
|---|
| 708 | unsigned long long blocknr, | 
|---|
| 709 | tid_t sequence) | 
|---|
| 710 | { | 
|---|
| 711 | struct jbd2_revoke_record_s *record; | 
|---|
| 712 |  | 
|---|
| 713 | record = find_revoke_record(journal, blocknr); | 
|---|
| 714 | if (!record) | 
|---|
| 715 | return 0; | 
|---|
| 716 | if (tid_gt(x: sequence, y: record->sequence)) | 
|---|
| 717 | return 0; | 
|---|
| 718 | return 1; | 
|---|
| 719 | } | 
|---|
| 720 |  | 
|---|
| 721 | /* | 
|---|
| 722 | * Finally, once recovery is over, we need to clear the revoke table so | 
|---|
| 723 | * that it can be reused by the running filesystem. | 
|---|
| 724 | */ | 
|---|
| 725 |  | 
|---|
| 726 | void jbd2_journal_clear_revoke(journal_t *journal) | 
|---|
| 727 | { | 
|---|
| 728 | int i; | 
|---|
| 729 | struct list_head *hash_list; | 
|---|
| 730 | struct jbd2_revoke_record_s *record; | 
|---|
| 731 | struct jbd2_revoke_table_s *revoke; | 
|---|
| 732 |  | 
|---|
| 733 | revoke = journal->j_revoke; | 
|---|
| 734 |  | 
|---|
| 735 | for (i = 0; i < revoke->hash_size; i++) { | 
|---|
| 736 | hash_list = &revoke->hash_table[i]; | 
|---|
| 737 | while (!list_empty(head: hash_list)) { | 
|---|
| 738 | record = (struct jbd2_revoke_record_s*) hash_list->next; | 
|---|
| 739 | list_del(entry: &record->hash); | 
|---|
| 740 | kmem_cache_free(s: jbd2_revoke_record_cache, objp: record); | 
|---|
| 741 | } | 
|---|
| 742 | } | 
|---|
| 743 | } | 
|---|
| 744 |  | 
|---|