| 1 | /* -*- mode: C++; c-basic-offset: 4; indent-tabs-mode: nil -*- */ |
| 2 | // vim: ft=cpp:expandtab:ts=8:sw=4:softtabstop=4: |
| 3 | #ident "$Id$" |
| 4 | /*====== |
| 5 | This file is part of PerconaFT. |
| 6 | |
| 7 | |
| 8 | Copyright (c) 2006, 2015, Percona and/or its affiliates. All rights reserved. |
| 9 | |
| 10 | PerconaFT is free software: you can redistribute it and/or modify |
| 11 | it under the terms of the GNU General Public License, version 2, |
| 12 | as published by the Free Software Foundation. |
| 13 | |
| 14 | PerconaFT is distributed in the hope that it will be useful, |
| 15 | but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 16 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 17 | GNU General Public License for more details. |
| 18 | |
| 19 | You should have received a copy of the GNU General Public License |
| 20 | along with PerconaFT. If not, see <http://www.gnu.org/licenses/>. |
| 21 | |
| 22 | ---------------------------------------- |
| 23 | |
| 24 | PerconaFT is free software: you can redistribute it and/or modify |
| 25 | it under the terms of the GNU Affero General Public License, version 3, |
| 26 | as published by the Free Software Foundation. |
| 27 | |
| 28 | PerconaFT is distributed in the hope that it will be useful, |
| 29 | but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 30 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 31 | GNU Affero General Public License for more details. |
| 32 | |
| 33 | You should have received a copy of the GNU Affero General Public License |
| 34 | along with PerconaFT. If not, see <http://www.gnu.org/licenses/>. |
| 35 | ======= */ |
| 36 | |
| 37 | #ident "Copyright (c) 2006, 2015, Percona and/or its affiliates. All rights reserved." |
| 38 | |
| 39 | /* rollback and rollforward routines. */ |
| 40 | |
| 41 | #include <memory> |
| 42 | #include "ft/ft-ops.h" |
| 43 | #include "ft/ft.h" |
| 44 | #include "ft/log_header.h" |
| 45 | #include "ft/logger/log-internal.h" |
| 46 | #include "ft/txn/rollback-apply.h" |
| 47 | #include "ft/txn/xids.h" |
| 48 | |
| 49 | // functionality provided by roll.c is exposed by an autogenerated |
| 50 | // header file, logheader.h |
| 51 | // |
| 52 | // this (poorly) explains the absence of "roll.h" |
| 53 | |
| 54 | // these flags control whether or not we send commit messages for |
| 55 | // various operations |
| 56 | |
| 57 | // When a transaction is committed, should we send a FT_COMMIT message |
| 58 | // for each FT_INSERT message sent earlier by the transaction? |
| 59 | #define TOKU_DO_COMMIT_CMD_INSERT 0 |
| 60 | |
| 61 | // When a transaction is committed, should we send a FT_COMMIT message |
| 62 | // for each FT_DELETE_ANY message sent earlier by the transaction? |
| 63 | #define TOKU_DO_COMMIT_CMD_DELETE 1 |
| 64 | |
| 65 | // When a transaction is committed, should we send a FT_COMMIT message |
| 66 | // for each FT_UPDATE message sent earlier by the transaction? |
| 67 | #define TOKU_DO_COMMIT_CMD_UPDATE 0 |
| 68 | |
| 69 | int |
| 70 | toku_commit_fdelete (FILENUM filenum, |
| 71 | TOKUTXN txn, |
| 72 | LSN UU(oplsn)) //oplsn is the lsn of the commit |
| 73 | { |
| 74 | int r; |
| 75 | CACHEFILE cf; |
| 76 | CACHETABLE ct = txn->logger->ct; |
| 77 | |
| 78 | // Try to get the cachefile for this filenum. A missing file on recovery |
| 79 | // is not an error, but a missing file outside of recovery is. |
| 80 | r = toku_cachefile_of_filenum(ct, filenum, &cf); |
| 81 | if (r == ENOENT) { |
| 82 | assert(txn->for_recovery); |
| 83 | r = 0; |
| 84 | goto done; |
| 85 | } |
| 86 | assert_zero(r); |
| 87 | |
| 88 | // bug fix for #4718 |
| 89 | // bug was introduced in with fix for #3590 |
| 90 | // Before Maxwell (and fix for #3590), |
| 91 | // the recovery log was fsynced after the xcommit was loged but |
| 92 | // before we processed rollback entries and before we released |
| 93 | // the row locks (in the lock tree). Due to performance concerns, |
| 94 | // the fsync was moved to after the release of row locks, which comes |
| 95 | // after processing rollback entries. As a result, we may be unlinking a file |
| 96 | // here as part of a transactoin that may abort if we do not fsync the log. |
| 97 | // So, we fsync the log here. |
| 98 | if (txn->logger) { |
| 99 | toku_logger_fsync_if_lsn_not_fsynced(txn->logger, txn->do_fsync_lsn); |
| 100 | } |
| 101 | |
| 102 | // Mark the cachefile as unlink on close. There are two ways for close |
| 103 | // to be eventually called on the cachefile: |
| 104 | // |
| 105 | // - when this txn completes, it will release a reference on the |
| 106 | // ft and close it, UNLESS it was pinned by checkpoint |
| 107 | // - if the cf was pinned by checkpoint, an unpin will release the |
| 108 | // final reference and call close. it must be the final reference |
| 109 | // since this txn has exclusive access to dictionary (by the |
| 110 | // directory row lock for its dname) and we would not get this |
| 111 | // far if there were other live handles. |
| 112 | toku_cachefile_unlink_on_close(cf); |
| 113 | done: |
| 114 | return r; |
| 115 | } |
| 116 | |
| 117 | int |
| 118 | toku_rollback_fdelete (FILENUM UU(filenum), |
| 119 | TOKUTXN UU(txn), |
| 120 | LSN UU(oplsn)) //oplsn is the lsn of the abort |
| 121 | { |
| 122 | //Rolling back an fdelete is an no-op. |
| 123 | return 0; |
| 124 | } |
| 125 | |
| 126 | int |
| 127 | toku_commit_fcreate (FILENUM UU(filenum), |
| 128 | BYTESTRING UU(bs_fname), |
| 129 | TOKUTXN UU(txn), |
| 130 | LSN UU(oplsn)) |
| 131 | { |
| 132 | return 0; |
| 133 | } |
| 134 | |
| 135 | int |
| 136 | toku_rollback_fcreate (FILENUM filenum, |
| 137 | BYTESTRING UU(bs_fname), |
| 138 | TOKUTXN txn, |
| 139 | LSN UU(oplsn)) |
| 140 | { |
| 141 | int r; |
| 142 | CACHEFILE cf; |
| 143 | CACHETABLE ct = txn->logger->ct; |
| 144 | |
| 145 | // Try to get the cachefile for this filenum. A missing file on recovery |
| 146 | // is not an error, but a missing file outside of recovery is. |
| 147 | r = toku_cachefile_of_filenum(ct, filenum, &cf); |
| 148 | if (r == ENOENT) { |
| 149 | r = 0; |
| 150 | goto done; |
| 151 | } |
| 152 | assert_zero(r); |
| 153 | |
| 154 | // Mark the cachefile as unlink on close. There are two ways for close |
| 155 | // to be eventually called on the cachefile: |
| 156 | // |
| 157 | // - when this txn completes, it will release a reference on the |
| 158 | // ft and close it, UNLESS it was pinned by checkpoint |
| 159 | // - if the cf was pinned by checkpoint, an unpin will release the |
| 160 | // final reference and call close. it must be the final reference |
| 161 | // since this txn has exclusive access to dictionary (by the |
| 162 | // directory row lock for its dname) and we would not get this |
| 163 | // far if there were other live handles. |
| 164 | toku_cachefile_unlink_on_close(cf); |
| 165 | toku_cachefile_skip_log_recover_on_close(cf); |
| 166 | done: |
| 167 | return 0; |
| 168 | } |
| 169 | |
| 170 | int toku_commit_frename(BYTESTRING /* old_name */, |
| 171 | BYTESTRING /* new_iname */, |
| 172 | TOKUTXN /* txn */, |
| 173 | LSN UU(oplsn)) { |
| 174 | return 0; |
| 175 | } |
| 176 | |
| 177 | int toku_rollback_frename(BYTESTRING old_iname, |
| 178 | BYTESTRING new_iname, |
| 179 | TOKUTXN txn, |
| 180 | LSN UU(oplsn)) { |
| 181 | assert(txn); |
| 182 | assert(txn->logger); |
| 183 | assert(txn->logger->ct); |
| 184 | |
| 185 | CACHETABLE cachetable = txn->logger->ct; |
| 186 | |
| 187 | toku_struct_stat stat; |
| 188 | bool old_exist = true; |
| 189 | bool new_exist = true; |
| 190 | |
| 191 | std::unique_ptr<char[], decltype(&toku_free)> old_iname_full( |
| 192 | toku_cachetable_get_fname_in_cwd(cachetable, old_iname.data), |
| 193 | &toku_free); |
| 194 | std::unique_ptr<char[], decltype(&toku_free)> new_iname_full( |
| 195 | toku_cachetable_get_fname_in_cwd(cachetable, new_iname.data), |
| 196 | &toku_free); |
| 197 | |
| 198 | if (toku_stat(old_iname_full.get(), &stat, toku_uninstrumented) == -1) { |
| 199 | if (ENOENT == errno) |
| 200 | old_exist = false; |
| 201 | else |
| 202 | return 1; |
| 203 | } |
| 204 | |
| 205 | if (toku_stat(new_iname_full.get(), &stat, toku_uninstrumented) == -1) { |
| 206 | if (ENOENT == errno || ENAMETOOLONG == errno) |
| 207 | new_exist = false; |
| 208 | else |
| 209 | return 1; |
| 210 | } |
| 211 | |
| 212 | // Both old and new files can exist if: |
| 213 | // - rename() is not completed |
| 214 | // - fcreate was replayed during recovery |
| 215 | // 'Stalled cachefiles' container cachefile_list::m_stale_fileid contains |
| 216 | // closed but not yet evicted cachefiles and the key of this container is |
| 217 | // fs-dependent file id - (device id, inode number) pair. To preserve the |
| 218 | // new cachefile |
| 219 | // file's id and keep it in 'stalled cachefiles' container the old file is |
| 220 | // removed |
| 221 | // and the new file is renamed. |
| 222 | if (old_exist && new_exist && |
| 223 | (toku_os_delete(old_iname_full.get()) == -1 || |
| 224 | toku_os_rename(new_iname_full.get(), old_iname_full.get()) == -1 || |
| 225 | toku_fsync_directory(new_iname_full.get()) == -1 || |
| 226 | toku_fsync_directory(old_iname_full.get()) == -1)) |
| 227 | return 1; |
| 228 | |
| 229 | if (!old_exist && new_exist && |
| 230 | (!toku_create_subdirs_if_needed(old_iname_full.get()) || |
| 231 | toku_os_rename(new_iname_full.get(), old_iname_full.get()) == -1 || |
| 232 | toku_fsync_directory(new_iname_full.get()) == -1 || |
| 233 | toku_fsync_directory(old_iname_full.get()) == -1)) |
| 234 | return 1; |
| 235 | |
| 236 | // it's ok if both files do not exist on recovery |
| 237 | if (!old_exist && !new_exist) |
| 238 | assert(txn->for_recovery); |
| 239 | |
| 240 | CACHEFILE cf; |
| 241 | int r = toku_cachefile_of_iname_in_env(cachetable, new_iname.data, &cf); |
| 242 | if (r != ENOENT) { |
| 243 | char *old_fname_in_cf = toku_cachefile_fname_in_env(cf); |
| 244 | toku_cachefile_set_fname_in_env(cf, toku_xstrdup(old_iname.data)); |
| 245 | toku_free(old_fname_in_cf); |
| 246 | // There is at least one case when fclose logging cause error: |
| 247 | // 1) start transaction |
| 248 | // 2) create ft 'a'(write "fcreate" in recovery log) |
| 249 | // 3) rename ft 'a' to 'b'(write "frename" in recovery log) |
| 250 | // 4) abort transaction: |
| 251 | // a) rollback rename ft (renames 'b' to 'a') |
| 252 | // b) rollback create ft (removes 'a'): |
| 253 | // invokes toku_cachefile_unlink_on_close - lazy unlink on file |
| 254 | // close, |
| 255 | // it just sets corresponding flag in cachefile object |
| 256 | // c) write "unlink" for 'a' in recovery log |
| 257 | // (when transaction is aborted all locks are released, |
| 258 | // when file lock is released the file is closed and unlinked if |
| 259 | // corresponding flag is set in cachefile object) |
| 260 | // 5) crash |
| 261 | // |
| 262 | // After this we have the following records in recovery log: |
| 263 | // - create ft 'a', |
| 264 | // - rename 'a' to 'b', |
| 265 | // - unlink 'a' |
| 266 | // |
| 267 | // On recovery: |
| 268 | // - create 'a' |
| 269 | // - rename 'a' to 'b' |
| 270 | // - unlink 'a' - as 'a' file does not exist we have crash on assert |
| 271 | // here |
| 272 | // |
| 273 | // There is no need to write "unlink" in recovery log in (4a) because |
| 274 | // 'a' will be removed |
| 275 | // on transaction rollback on recovery. |
| 276 | toku_cachefile_skip_log_recover_on_close(cf); |
| 277 | } |
| 278 | |
| 279 | return 0; |
| 280 | } |
| 281 | |
| 282 | int find_ft_from_filenum (const FT &ft, const FILENUM &filenum); |
| 283 | int find_ft_from_filenum (const FT &ft, const FILENUM &filenum) { |
| 284 | FILENUM thisfnum = toku_cachefile_filenum(ft->cf); |
| 285 | if (thisfnum.fileid<filenum.fileid) return -1; |
| 286 | if (thisfnum.fileid>filenum.fileid) return +1; |
| 287 | return 0; |
| 288 | } |
| 289 | |
| 290 | // Input arg reset_root_xid_that_created true means that this operation has changed the definition of this dictionary. |
| 291 | // (Example use is for schema change committed with txn that inserted cmdupdatebroadcast message.) |
| 292 | // The oplsn argument is ZERO_LSN for normal operation. When this function is called for recovery, it has the LSN of |
| 293 | // the operation (insert, delete, update, etc). |
| 294 | static int do_insertion (enum ft_msg_type type, FILENUM filenum, BYTESTRING key, BYTESTRING *data, TOKUTXN txn, LSN oplsn, |
| 295 | bool reset_root_xid_that_created) { |
| 296 | int r = 0; |
| 297 | //printf("%s:%d committing insert %s %s\n", __FILE__, __LINE__, key.data, data.data); |
| 298 | FT ft = nullptr; |
| 299 | r = txn->open_fts.find_zero<FILENUM, find_ft_from_filenum>(filenum, &ft, NULL); |
| 300 | if (r == DB_NOTFOUND) { |
| 301 | assert(txn->for_recovery); |
| 302 | r = 0; |
| 303 | goto done; |
| 304 | } |
| 305 | assert(r==0); |
| 306 | |
| 307 | if (oplsn.lsn != 0) { // if we are executing the recovery algorithm |
| 308 | LSN treelsn = toku_ft_checkpoint_lsn(ft); |
| 309 | if (oplsn.lsn <= treelsn.lsn) { // if operation was already applied to tree ... |
| 310 | r = 0; // ... do not apply it again. |
| 311 | goto done; |
| 312 | } |
| 313 | } |
| 314 | |
| 315 | DBT key_dbt,data_dbt; |
| 316 | XIDS xids; |
| 317 | xids = toku_txn_get_xids(txn); |
| 318 | { |
| 319 | const DBT *kdbt = key.len > 0 ? toku_fill_dbt(&key_dbt, key.data, key.len) : |
| 320 | toku_init_dbt(&key_dbt); |
| 321 | const DBT *vdbt = data ? toku_fill_dbt(&data_dbt, data->data, data->len) : |
| 322 | toku_init_dbt(&data_dbt); |
| 323 | ft_msg msg(kdbt, vdbt, type, ZERO_MSN, xids); |
| 324 | |
| 325 | TXN_MANAGER txn_manager = toku_logger_get_txn_manager(txn->logger); |
| 326 | txn_manager_state txn_state_for_gc(txn_manager); |
| 327 | |
| 328 | TXNID oldest_referenced_xid_estimate = toku_txn_manager_get_oldest_referenced_xid_estimate(txn_manager); |
| 329 | txn_gc_info gc_info(&txn_state_for_gc, |
| 330 | oldest_referenced_xid_estimate, |
| 331 | // no messages above us, we can implicitly promote uxrs based on this xid |
| 332 | oldest_referenced_xid_estimate, |
| 333 | !txn->for_recovery); |
| 334 | toku_ft_root_put_msg(ft, msg, &gc_info); |
| 335 | if (reset_root_xid_that_created) { |
| 336 | TXNID new_root_xid_that_created = toku_xids_get_outermost_xid(xids); |
| 337 | toku_reset_root_xid_that_created(ft, new_root_xid_that_created); |
| 338 | } |
| 339 | } |
| 340 | done: |
| 341 | return r; |
| 342 | } |
| 343 | |
| 344 | |
| 345 | static int do_nothing_with_filenum(TOKUTXN UU(txn), FILENUM UU(filenum)) { |
| 346 | return 0; |
| 347 | } |
| 348 | |
| 349 | |
| 350 | int toku_commit_cmdinsert (FILENUM filenum, BYTESTRING UU(key), TOKUTXN txn, LSN UU(oplsn)) { |
| 351 | #if TOKU_DO_COMMIT_CMD_INSERT |
| 352 | return do_insertion (FT_COMMIT_ANY, filenum, key, 0, txn, oplsn, false); |
| 353 | #else |
| 354 | return do_nothing_with_filenum(txn, filenum); |
| 355 | #endif |
| 356 | } |
| 357 | |
| 358 | int |
| 359 | toku_rollback_cmdinsert (FILENUM filenum, |
| 360 | BYTESTRING key, |
| 361 | TOKUTXN txn, |
| 362 | LSN oplsn) |
| 363 | { |
| 364 | return do_insertion (FT_ABORT_ANY, filenum, key, 0, txn, oplsn, false); |
| 365 | } |
| 366 | |
| 367 | int |
| 368 | toku_commit_cmdupdate(FILENUM filenum, |
| 369 | BYTESTRING UU(key), |
| 370 | TOKUTXN txn, |
| 371 | LSN UU(oplsn)) |
| 372 | { |
| 373 | #if TOKU_DO_COMMIT_CMD_UPDATE |
| 374 | return do_insertion(FT_COMMIT_ANY, filenum, key, 0, txn, oplsn, false); |
| 375 | #else |
| 376 | return do_nothing_with_filenum(txn, filenum); |
| 377 | #endif |
| 378 | } |
| 379 | |
| 380 | int |
| 381 | toku_rollback_cmdupdate(FILENUM filenum, |
| 382 | BYTESTRING key, |
| 383 | TOKUTXN txn, |
| 384 | LSN oplsn) |
| 385 | { |
| 386 | return do_insertion(FT_ABORT_ANY, filenum, key, 0, txn, oplsn, false); |
| 387 | } |
| 388 | |
| 389 | int |
| 390 | toku_commit_cmdupdatebroadcast(FILENUM filenum, |
| 391 | bool is_resetting_op, |
| 392 | TOKUTXN txn, |
| 393 | LSN oplsn) |
| 394 | { |
| 395 | // if is_resetting_op, reset root_xid_that_created in |
| 396 | // relevant ft. |
| 397 | bool reset_root_xid_that_created = (is_resetting_op ? true : false); |
| 398 | const enum ft_msg_type msg_type = (is_resetting_op |
| 399 | ? FT_COMMIT_BROADCAST_ALL |
| 400 | : FT_COMMIT_BROADCAST_TXN); |
| 401 | BYTESTRING nullkey = { 0, NULL }; |
| 402 | return do_insertion(msg_type, filenum, nullkey, 0, txn, oplsn, reset_root_xid_that_created); |
| 403 | } |
| 404 | |
| 405 | int |
| 406 | toku_rollback_cmdupdatebroadcast(FILENUM filenum, |
| 407 | bool UU(is_resetting_op), |
| 408 | TOKUTXN txn, |
| 409 | LSN oplsn) |
| 410 | { |
| 411 | BYTESTRING nullkey = { 0, NULL }; |
| 412 | return do_insertion(FT_ABORT_BROADCAST_TXN, filenum, nullkey, 0, txn, oplsn, false); |
| 413 | } |
| 414 | |
| 415 | int |
| 416 | toku_commit_cmddelete (FILENUM filenum, |
| 417 | BYTESTRING key, |
| 418 | TOKUTXN txn, |
| 419 | LSN oplsn) |
| 420 | { |
| 421 | #if TOKU_DO_COMMIT_CMD_DELETE |
| 422 | return do_insertion (FT_COMMIT_ANY, filenum, key, 0, txn, oplsn, false); |
| 423 | #else |
| 424 | key = key; oplsn = oplsn; |
| 425 | return do_nothing_with_filenum(txn, filenum); |
| 426 | #endif |
| 427 | } |
| 428 | |
| 429 | int |
| 430 | toku_rollback_cmddelete (FILENUM filenum, |
| 431 | BYTESTRING key, |
| 432 | TOKUTXN txn, |
| 433 | LSN oplsn) |
| 434 | { |
| 435 | return do_insertion (FT_ABORT_ANY, filenum, key, 0, txn, oplsn, false); |
| 436 | } |
| 437 | |
| 438 | static int |
| 439 | toku_apply_rollinclude (TXNID_PAIR xid, |
| 440 | uint64_t num_nodes, |
| 441 | BLOCKNUM spilled_head, |
| 442 | BLOCKNUM spilled_tail, |
| 443 | TOKUTXN txn, |
| 444 | LSN oplsn, |
| 445 | apply_rollback_item func) { |
| 446 | int r = 0; |
| 447 | struct roll_entry *item; |
| 448 | |
| 449 | BLOCKNUM next_log = spilled_tail; |
| 450 | uint64_t last_sequence = num_nodes; |
| 451 | |
| 452 | bool found_head = false; |
| 453 | assert(next_log.b != ROLLBACK_NONE.b); |
| 454 | while (next_log.b != ROLLBACK_NONE.b) { |
| 455 | //pin log |
| 456 | ROLLBACK_LOG_NODE log; |
| 457 | toku_get_and_pin_rollback_log(txn, next_log, &log); |
| 458 | toku_rollback_verify_contents(log, xid, last_sequence - 1); |
| 459 | last_sequence = log->sequence; |
| 460 | |
| 461 | toku_maybe_prefetch_previous_rollback_log(txn, log); |
| 462 | |
| 463 | while ((item=log->newest_logentry)) { |
| 464 | log->newest_logentry = item->prev; |
| 465 | r = func(txn, item, oplsn); |
| 466 | if (r!=0) return r; |
| 467 | } |
| 468 | if (next_log.b == spilled_head.b) { |
| 469 | assert(!found_head); |
| 470 | found_head = true; |
| 471 | assert(log->sequence == 0); |
| 472 | } |
| 473 | next_log = log->previous; |
| 474 | { |
| 475 | //Clean up transaction structure to prevent |
| 476 | //toku_txn_close from double-freeing |
| 477 | spilled_tail = next_log; |
| 478 | if (found_head) { |
| 479 | assert(next_log.b == ROLLBACK_NONE.b); |
| 480 | spilled_head = next_log; |
| 481 | } |
| 482 | } |
| 483 | toku_rollback_log_unpin_and_remove(txn, log); |
| 484 | } |
| 485 | return r; |
| 486 | } |
| 487 | |
| 488 | int |
| 489 | toku_commit_rollinclude (TXNID_PAIR xid, |
| 490 | uint64_t num_nodes, |
| 491 | BLOCKNUM spilled_head, |
| 492 | BLOCKNUM spilled_tail, |
| 493 | TOKUTXN txn, |
| 494 | LSN oplsn) { |
| 495 | int r; |
| 496 | r = toku_apply_rollinclude(xid, num_nodes, |
| 497 | spilled_head, |
| 498 | spilled_tail, |
| 499 | txn, oplsn, |
| 500 | toku_commit_rollback_item); |
| 501 | return r; |
| 502 | } |
| 503 | |
| 504 | int |
| 505 | toku_rollback_rollinclude (TXNID_PAIR xid, |
| 506 | uint64_t num_nodes, |
| 507 | BLOCKNUM spilled_head, |
| 508 | BLOCKNUM spilled_tail, |
| 509 | TOKUTXN txn, |
| 510 | LSN oplsn) { |
| 511 | int r; |
| 512 | r = toku_apply_rollinclude(xid, num_nodes, |
| 513 | spilled_head, |
| 514 | spilled_tail, |
| 515 | txn, oplsn, |
| 516 | toku_abort_rollback_item); |
| 517 | return r; |
| 518 | } |
| 519 | |
| 520 | int |
| 521 | toku_commit_load (FILENUM old_filenum, |
| 522 | BYTESTRING UU(new_iname), |
| 523 | TOKUTXN txn, |
| 524 | LSN UU(oplsn)) |
| 525 | { |
| 526 | int r; |
| 527 | CACHEFILE old_cf; |
| 528 | CACHETABLE ct = txn->logger->ct; |
| 529 | |
| 530 | // To commit a dictionary load, we delete the old file |
| 531 | // |
| 532 | // Try to get the cachefile for the old filenum. A missing file on recovery |
| 533 | // is not an error, but a missing file outside of recovery is. |
| 534 | r = toku_cachefile_of_filenum(ct, old_filenum, &old_cf); |
| 535 | if (r == ENOENT) { |
| 536 | invariant(txn->for_recovery); |
| 537 | r = 0; |
| 538 | goto done; |
| 539 | } |
| 540 | lazy_assert(r == 0); |
| 541 | |
| 542 | // bug fix for #4718 |
| 543 | // bug was introduced in with fix for #3590 |
| 544 | // Before Maxwell (and fix for #3590), |
| 545 | // the recovery log was fsynced after the xcommit was loged but |
| 546 | // before we processed rollback entries and before we released |
| 547 | // the row locks (in the lock tree). Due to performance concerns, |
| 548 | // the fsync was moved to after the release of row locks, which comes |
| 549 | // after processing rollback entries. As a result, we may be unlinking a file |
| 550 | // here as part of a transactoin that may abort if we do not fsync the log. |
| 551 | // So, we fsync the log here. |
| 552 | if (txn->logger) { |
| 553 | toku_logger_fsync_if_lsn_not_fsynced(txn->logger, txn->do_fsync_lsn); |
| 554 | } |
| 555 | |
| 556 | // TODO: Zardosht |
| 557 | // Explain why this condition is valid, because I forget. |
| 558 | if (!toku_cachefile_is_unlink_on_close(old_cf)) { |
| 559 | toku_cachefile_unlink_on_close(old_cf); |
| 560 | } |
| 561 | done: |
| 562 | return r; |
| 563 | } |
| 564 | |
| 565 | int |
| 566 | toku_rollback_load (FILENUM UU(old_filenum), |
| 567 | BYTESTRING new_iname, |
| 568 | TOKUTXN txn, |
| 569 | LSN UU(oplsn)) |
| 570 | { |
| 571 | int r; |
| 572 | CACHEFILE new_cf; |
| 573 | CACHETABLE ct = txn->logger->ct; |
| 574 | |
| 575 | // To rollback a dictionary load, we delete the new file. |
| 576 | // Try to get the cachefile for the new fname. |
| 577 | char *fname_in_env = fixup_fname(&new_iname); |
| 578 | r = toku_cachefile_of_iname_in_env(ct, fname_in_env, &new_cf); |
| 579 | if (r == ENOENT) { |
| 580 | // It's possible the new iname was never created, so just try to |
| 581 | // unlink it if it's there and ignore the error if it's not. |
| 582 | char *fname_in_cwd = toku_cachetable_get_fname_in_cwd(ct, fname_in_env); |
| 583 | r = unlink(fname_in_cwd); |
| 584 | assert(r == 0 || get_error_errno() == ENOENT); |
| 585 | toku_free(fname_in_cwd); |
| 586 | r = 0; |
| 587 | } else { |
| 588 | assert_zero(r); |
| 589 | toku_cachefile_unlink_on_close(new_cf); |
| 590 | } |
| 591 | toku_free(fname_in_env); |
| 592 | return r; |
| 593 | } |
| 594 | |
| 595 | //2954 |
| 596 | int |
| 597 | toku_commit_hot_index (FILENUMS UU(hot_index_filenums), |
| 598 | TOKUTXN UU(txn), |
| 599 | LSN UU(oplsn)) |
| 600 | { |
| 601 | // nothing |
| 602 | return 0; |
| 603 | } |
| 604 | |
| 605 | int |
| 606 | toku_rollback_hot_index (FILENUMS UU(hot_index_filenums), |
| 607 | TOKUTXN UU(txn), |
| 608 | LSN UU(oplsn)) |
| 609 | { |
| 610 | return 0; |
| 611 | } |
| 612 | |
| 613 | int |
| 614 | toku_commit_dictionary_redirect (FILENUM UU(old_filenum), |
| 615 | FILENUM UU(new_filenum), |
| 616 | TOKUTXN UU(txn), |
| 617 | LSN UU(oplsn)) //oplsn is the lsn of the commit |
| 618 | { |
| 619 | //Redirect only has meaning during normal operation (NOT during recovery). |
| 620 | if (!txn->for_recovery) { |
| 621 | //NO-OP |
| 622 | } |
| 623 | return 0; |
| 624 | } |
| 625 | |
| 626 | int |
| 627 | toku_rollback_dictionary_redirect (FILENUM old_filenum, |
| 628 | FILENUM new_filenum, |
| 629 | TOKUTXN txn, |
| 630 | LSN UU(oplsn)) //oplsn is the lsn of the abort |
| 631 | { |
| 632 | int r = 0; |
| 633 | //Redirect only has meaning during normal operation (NOT during recovery). |
| 634 | if (!txn->for_recovery) { |
| 635 | CACHEFILE new_cf = NULL; |
| 636 | r = toku_cachefile_of_filenum(txn->logger->ct, new_filenum, &new_cf); |
| 637 | assert(r == 0); |
| 638 | FT CAST_FROM_VOIDP(new_ft, toku_cachefile_get_userdata(new_cf)); |
| 639 | |
| 640 | CACHEFILE old_cf = NULL; |
| 641 | r = toku_cachefile_of_filenum(txn->logger->ct, old_filenum, &old_cf); |
| 642 | assert(r == 0); |
| 643 | FT CAST_FROM_VOIDP(old_ft, toku_cachefile_get_userdata(old_cf)); |
| 644 | |
| 645 | //Redirect back from new to old. |
| 646 | r = toku_dictionary_redirect_abort(old_ft, new_ft, txn); |
| 647 | assert(r==0); |
| 648 | } |
| 649 | return r; |
| 650 | } |
| 651 | |
| 652 | int |
| 653 | toku_commit_change_fdescriptor(FILENUM filenum, |
| 654 | BYTESTRING UU(old_descriptor), |
| 655 | TOKUTXN txn, |
| 656 | LSN UU(oplsn)) |
| 657 | { |
| 658 | return do_nothing_with_filenum(txn, filenum); |
| 659 | } |
| 660 | |
| 661 | int |
| 662 | toku_rollback_change_fdescriptor(FILENUM filenum, |
| 663 | BYTESTRING old_descriptor, |
| 664 | TOKUTXN txn, |
| 665 | LSN UU(oplsn)) |
| 666 | { |
| 667 | CACHEFILE cf; |
| 668 | int r; |
| 669 | r = toku_cachefile_of_filenum(txn->logger->ct, filenum, &cf); |
| 670 | if (r == ENOENT) { //Missing file on recovered transaction is not an error |
| 671 | assert(txn->for_recovery); |
| 672 | r = 0; |
| 673 | goto done; |
| 674 | } |
| 675 | // file must be open, because the txn that created it opened it and |
| 676 | // noted it, |
| 677 | assert(r == 0); |
| 678 | |
| 679 | FT ft; |
| 680 | ft = NULL; |
| 681 | r = txn->open_fts.find_zero<FILENUM, find_ft_from_filenum>(filenum, &ft, NULL); |
| 682 | assert(r == 0); |
| 683 | |
| 684 | DESCRIPTOR_S d; |
| 685 | toku_fill_dbt(&d.dbt, old_descriptor.data, old_descriptor.len); |
| 686 | toku_ft_update_descriptor(ft, &d); |
| 687 | done: |
| 688 | return r; |
| 689 | } |
| 690 | |
| 691 | |
| 692 | |
| 693 | |