1 // SPDX-License-Identifier: GPL-2.0+ 2 /* 3 * Copyright (C) 2016 Oracle. All Rights Reserved. 4 * Author: Darrick J. Wong <darrick.wong@oracle.com> 5 */ 6 #include "xfs.h" 7 #include "xfs_fs.h" 8 #include "xfs_format.h" 9 #include "xfs_log_format.h" 10 #include "xfs_trans_resv.h" 11 #include "xfs_bit.h" 12 #include "xfs_shared.h" 13 #include "xfs_mount.h" 14 #include "xfs_defer.h" 15 #include "xfs_trans.h" 16 #include "xfs_trans_priv.h" 17 #include "xfs_buf_item.h" 18 #include "xfs_rmap_item.h" 19 #include "xfs_log.h" 20 #include "xfs_rmap.h" 21 22 23 kmem_zone_t *xfs_rui_zone; 24 kmem_zone_t *xfs_rud_zone; 25 26 static inline struct xfs_rui_log_item *RUI_ITEM(struct xfs_log_item *lip) 27 { 28 return container_of(lip, struct xfs_rui_log_item, rui_item); 29 } 30 31 void 32 xfs_rui_item_free( 33 struct xfs_rui_log_item *ruip) 34 { 35 if (ruip->rui_format.rui_nextents > XFS_RUI_MAX_FAST_EXTENTS) 36 kmem_free(ruip); 37 else 38 kmem_zone_free(xfs_rui_zone, ruip); 39 } 40 41 /* 42 * Freeing the RUI requires that we remove it from the AIL if it has already 43 * been placed there. However, the RUI may not yet have been placed in the AIL 44 * when called by xfs_rui_release() from RUD processing due to the ordering of 45 * committed vs unpin operations in bulk insert operations. Hence the reference 46 * count to ensure only the last caller frees the RUI. 47 */ 48 void 49 xfs_rui_release( 50 struct xfs_rui_log_item *ruip) 51 { 52 ASSERT(atomic_read(&ruip->rui_refcount) > 0); 53 if (atomic_dec_and_test(&ruip->rui_refcount)) { 54 xfs_trans_ail_remove(&ruip->rui_item, SHUTDOWN_LOG_IO_ERROR); 55 xfs_rui_item_free(ruip); 56 } 57 } 58 59 STATIC void 60 xfs_rui_item_size( 61 struct xfs_log_item *lip, 62 int *nvecs, 63 int *nbytes) 64 { 65 struct xfs_rui_log_item *ruip = RUI_ITEM(lip); 66 67 *nvecs += 1; 68 *nbytes += xfs_rui_log_format_sizeof(ruip->rui_format.rui_nextents); 69 } 70 71 /* 72 * This is called to fill in the vector of log iovecs for the 73 * given rui log item. We use only 1 iovec, and we point that 74 * at the rui_log_format structure embedded in the rui item. 75 * It is at this point that we assert that all of the extent 76 * slots in the rui item have been filled. 77 */ 78 STATIC void 79 xfs_rui_item_format( 80 struct xfs_log_item *lip, 81 struct xfs_log_vec *lv) 82 { 83 struct xfs_rui_log_item *ruip = RUI_ITEM(lip); 84 struct xfs_log_iovec *vecp = NULL; 85 86 ASSERT(atomic_read(&ruip->rui_next_extent) == 87 ruip->rui_format.rui_nextents); 88 89 ruip->rui_format.rui_type = XFS_LI_RUI; 90 ruip->rui_format.rui_size = 1; 91 92 xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_RUI_FORMAT, &ruip->rui_format, 93 xfs_rui_log_format_sizeof(ruip->rui_format.rui_nextents)); 94 } 95 96 /* 97 * Pinning has no meaning for an rui item, so just return. 98 */ 99 STATIC void 100 xfs_rui_item_pin( 101 struct xfs_log_item *lip) 102 { 103 } 104 105 /* 106 * The unpin operation is the last place an RUI is manipulated in the log. It is 107 * either inserted in the AIL or aborted in the event of a log I/O error. In 108 * either case, the RUI transaction has been successfully committed to make it 109 * this far. Therefore, we expect whoever committed the RUI to either construct 110 * and commit the RUD or drop the RUD's reference in the event of error. Simply 111 * drop the log's RUI reference now that the log is done with it. 112 */ 113 STATIC void 114 xfs_rui_item_unpin( 115 struct xfs_log_item *lip, 116 int remove) 117 { 118 struct xfs_rui_log_item *ruip = RUI_ITEM(lip); 119 120 xfs_rui_release(ruip); 121 } 122 123 /* 124 * RUI items have no locking or pushing. However, since RUIs are pulled from 125 * the AIL when their corresponding RUDs are committed to disk, their situation 126 * is very similar to being pinned. Return XFS_ITEM_PINNED so that the caller 127 * will eventually flush the log. This should help in getting the RUI out of 128 * the AIL. 129 */ 130 STATIC uint 131 xfs_rui_item_push( 132 struct xfs_log_item *lip, 133 struct list_head *buffer_list) 134 { 135 return XFS_ITEM_PINNED; 136 } 137 138 /* 139 * The RUI has been either committed or aborted if the transaction has been 140 * cancelled. If the transaction was cancelled, an RUD isn't going to be 141 * constructed and thus we free the RUI here directly. 142 */ 143 STATIC void 144 xfs_rui_item_unlock( 145 struct xfs_log_item *lip) 146 { 147 if (test_bit(XFS_LI_ABORTED, &lip->li_flags)) 148 xfs_rui_release(RUI_ITEM(lip)); 149 } 150 151 /* 152 * The RUI is logged only once and cannot be moved in the log, so simply return 153 * the lsn at which it's been logged. 154 */ 155 STATIC xfs_lsn_t 156 xfs_rui_item_committed( 157 struct xfs_log_item *lip, 158 xfs_lsn_t lsn) 159 { 160 return lsn; 161 } 162 163 /* 164 * The RUI dependency tracking op doesn't do squat. It can't because 165 * it doesn't know where the free extent is coming from. The dependency 166 * tracking has to be handled by the "enclosing" metadata object. For 167 * example, for inodes, the inode is locked throughout the extent freeing 168 * so the dependency should be recorded there. 169 */ 170 STATIC void 171 xfs_rui_item_committing( 172 struct xfs_log_item *lip, 173 xfs_lsn_t lsn) 174 { 175 } 176 177 /* 178 * This is the ops vector shared by all rui log items. 179 */ 180 static const struct xfs_item_ops xfs_rui_item_ops = { 181 .iop_size = xfs_rui_item_size, 182 .iop_format = xfs_rui_item_format, 183 .iop_pin = xfs_rui_item_pin, 184 .iop_unpin = xfs_rui_item_unpin, 185 .iop_unlock = xfs_rui_item_unlock, 186 .iop_committed = xfs_rui_item_committed, 187 .iop_push = xfs_rui_item_push, 188 .iop_committing = xfs_rui_item_committing, 189 }; 190 191 /* 192 * Allocate and initialize an rui item with the given number of extents. 193 */ 194 struct xfs_rui_log_item * 195 xfs_rui_init( 196 struct xfs_mount *mp, 197 uint nextents) 198 199 { 200 struct xfs_rui_log_item *ruip; 201 202 ASSERT(nextents > 0); 203 if (nextents > XFS_RUI_MAX_FAST_EXTENTS) 204 ruip = kmem_zalloc(xfs_rui_log_item_sizeof(nextents), KM_SLEEP); 205 else 206 ruip = kmem_zone_zalloc(xfs_rui_zone, KM_SLEEP); 207 208 xfs_log_item_init(mp, &ruip->rui_item, XFS_LI_RUI, &xfs_rui_item_ops); 209 ruip->rui_format.rui_nextents = nextents; 210 ruip->rui_format.rui_id = (uintptr_t)(void *)ruip; 211 atomic_set(&ruip->rui_next_extent, 0); 212 atomic_set(&ruip->rui_refcount, 2); 213 214 return ruip; 215 } 216 217 /* 218 * Copy an RUI format buffer from the given buf, and into the destination 219 * RUI format structure. The RUI/RUD items were designed not to need any 220 * special alignment handling. 221 */ 222 int 223 xfs_rui_copy_format( 224 struct xfs_log_iovec *buf, 225 struct xfs_rui_log_format *dst_rui_fmt) 226 { 227 struct xfs_rui_log_format *src_rui_fmt; 228 uint len; 229 230 src_rui_fmt = buf->i_addr; 231 len = xfs_rui_log_format_sizeof(src_rui_fmt->rui_nextents); 232 233 if (buf->i_len != len) 234 return -EFSCORRUPTED; 235 236 memcpy(dst_rui_fmt, src_rui_fmt, len); 237 return 0; 238 } 239 240 static inline struct xfs_rud_log_item *RUD_ITEM(struct xfs_log_item *lip) 241 { 242 return container_of(lip, struct xfs_rud_log_item, rud_item); 243 } 244 245 STATIC void 246 xfs_rud_item_size( 247 struct xfs_log_item *lip, 248 int *nvecs, 249 int *nbytes) 250 { 251 *nvecs += 1; 252 *nbytes += sizeof(struct xfs_rud_log_format); 253 } 254 255 /* 256 * This is called to fill in the vector of log iovecs for the 257 * given rud log item. We use only 1 iovec, and we point that 258 * at the rud_log_format structure embedded in the rud item. 259 * It is at this point that we assert that all of the extent 260 * slots in the rud item have been filled. 261 */ 262 STATIC void 263 xfs_rud_item_format( 264 struct xfs_log_item *lip, 265 struct xfs_log_vec *lv) 266 { 267 struct xfs_rud_log_item *rudp = RUD_ITEM(lip); 268 struct xfs_log_iovec *vecp = NULL; 269 270 rudp->rud_format.rud_type = XFS_LI_RUD; 271 rudp->rud_format.rud_size = 1; 272 273 xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_RUD_FORMAT, &rudp->rud_format, 274 sizeof(struct xfs_rud_log_format)); 275 } 276 277 /* 278 * Pinning has no meaning for an rud item, so just return. 279 */ 280 STATIC void 281 xfs_rud_item_pin( 282 struct xfs_log_item *lip) 283 { 284 } 285 286 /* 287 * Since pinning has no meaning for an rud item, unpinning does 288 * not either. 289 */ 290 STATIC void 291 xfs_rud_item_unpin( 292 struct xfs_log_item *lip, 293 int remove) 294 { 295 } 296 297 /* 298 * There isn't much you can do to push on an rud item. It is simply stuck 299 * waiting for the log to be flushed to disk. 300 */ 301 STATIC uint 302 xfs_rud_item_push( 303 struct xfs_log_item *lip, 304 struct list_head *buffer_list) 305 { 306 return XFS_ITEM_PINNED; 307 } 308 309 /* 310 * The RUD is either committed or aborted if the transaction is cancelled. If 311 * the transaction is cancelled, drop our reference to the RUI and free the 312 * RUD. 313 */ 314 STATIC void 315 xfs_rud_item_unlock( 316 struct xfs_log_item *lip) 317 { 318 struct xfs_rud_log_item *rudp = RUD_ITEM(lip); 319 320 if (test_bit(XFS_LI_ABORTED, &lip->li_flags)) { 321 xfs_rui_release(rudp->rud_ruip); 322 kmem_zone_free(xfs_rud_zone, rudp); 323 } 324 } 325 326 /* 327 * When the rud item is committed to disk, all we need to do is delete our 328 * reference to our partner rui item and then free ourselves. Since we're 329 * freeing ourselves we must return -1 to keep the transaction code from 330 * further referencing this item. 331 */ 332 STATIC xfs_lsn_t 333 xfs_rud_item_committed( 334 struct xfs_log_item *lip, 335 xfs_lsn_t lsn) 336 { 337 struct xfs_rud_log_item *rudp = RUD_ITEM(lip); 338 339 /* 340 * Drop the RUI reference regardless of whether the RUD has been 341 * aborted. Once the RUD transaction is constructed, it is the sole 342 * responsibility of the RUD to release the RUI (even if the RUI is 343 * aborted due to log I/O error). 344 */ 345 xfs_rui_release(rudp->rud_ruip); 346 kmem_zone_free(xfs_rud_zone, rudp); 347 348 return (xfs_lsn_t)-1; 349 } 350 351 /* 352 * The RUD dependency tracking op doesn't do squat. It can't because 353 * it doesn't know where the free extent is coming from. The dependency 354 * tracking has to be handled by the "enclosing" metadata object. For 355 * example, for inodes, the inode is locked throughout the extent freeing 356 * so the dependency should be recorded there. 357 */ 358 STATIC void 359 xfs_rud_item_committing( 360 struct xfs_log_item *lip, 361 xfs_lsn_t lsn) 362 { 363 } 364 365 /* 366 * This is the ops vector shared by all rud log items. 367 */ 368 static const struct xfs_item_ops xfs_rud_item_ops = { 369 .iop_size = xfs_rud_item_size, 370 .iop_format = xfs_rud_item_format, 371 .iop_pin = xfs_rud_item_pin, 372 .iop_unpin = xfs_rud_item_unpin, 373 .iop_unlock = xfs_rud_item_unlock, 374 .iop_committed = xfs_rud_item_committed, 375 .iop_push = xfs_rud_item_push, 376 .iop_committing = xfs_rud_item_committing, 377 }; 378 379 /* 380 * Allocate and initialize an rud item with the given number of extents. 381 */ 382 struct xfs_rud_log_item * 383 xfs_rud_init( 384 struct xfs_mount *mp, 385 struct xfs_rui_log_item *ruip) 386 387 { 388 struct xfs_rud_log_item *rudp; 389 390 rudp = kmem_zone_zalloc(xfs_rud_zone, KM_SLEEP); 391 xfs_log_item_init(mp, &rudp->rud_item, XFS_LI_RUD, &xfs_rud_item_ops); 392 rudp->rud_ruip = ruip; 393 rudp->rud_format.rud_rui_id = ruip->rui_format.rui_id; 394 395 return rudp; 396 } 397 398 /* 399 * Process an rmap update intent item that was recovered from the log. 400 * We need to update the rmapbt. 401 */ 402 int 403 xfs_rui_recover( 404 struct xfs_mount *mp, 405 struct xfs_rui_log_item *ruip) 406 { 407 int i; 408 int error = 0; 409 struct xfs_map_extent *rmap; 410 xfs_fsblock_t startblock_fsb; 411 bool op_ok; 412 struct xfs_rud_log_item *rudp; 413 enum xfs_rmap_intent_type type; 414 int whichfork; 415 xfs_exntst_t state; 416 struct xfs_trans *tp; 417 struct xfs_btree_cur *rcur = NULL; 418 419 ASSERT(!test_bit(XFS_RUI_RECOVERED, &ruip->rui_flags)); 420 421 /* 422 * First check the validity of the extents described by the 423 * RUI. If any are bad, then assume that all are bad and 424 * just toss the RUI. 425 */ 426 for (i = 0; i < ruip->rui_format.rui_nextents; i++) { 427 rmap = &ruip->rui_format.rui_extents[i]; 428 startblock_fsb = XFS_BB_TO_FSB(mp, 429 XFS_FSB_TO_DADDR(mp, rmap->me_startblock)); 430 switch (rmap->me_flags & XFS_RMAP_EXTENT_TYPE_MASK) { 431 case XFS_RMAP_EXTENT_MAP: 432 case XFS_RMAP_EXTENT_MAP_SHARED: 433 case XFS_RMAP_EXTENT_UNMAP: 434 case XFS_RMAP_EXTENT_UNMAP_SHARED: 435 case XFS_RMAP_EXTENT_CONVERT: 436 case XFS_RMAP_EXTENT_CONVERT_SHARED: 437 case XFS_RMAP_EXTENT_ALLOC: 438 case XFS_RMAP_EXTENT_FREE: 439 op_ok = true; 440 break; 441 default: 442 op_ok = false; 443 break; 444 } 445 if (!op_ok || startblock_fsb == 0 || 446 rmap->me_len == 0 || 447 startblock_fsb >= mp->m_sb.sb_dblocks || 448 rmap->me_len >= mp->m_sb.sb_agblocks || 449 (rmap->me_flags & ~XFS_RMAP_EXTENT_FLAGS)) { 450 /* 451 * This will pull the RUI from the AIL and 452 * free the memory associated with it. 453 */ 454 set_bit(XFS_RUI_RECOVERED, &ruip->rui_flags); 455 xfs_rui_release(ruip); 456 return -EIO; 457 } 458 } 459 460 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 461 mp->m_rmap_maxlevels, 0, XFS_TRANS_RESERVE, &tp); 462 if (error) 463 return error; 464 rudp = xfs_trans_get_rud(tp, ruip); 465 466 for (i = 0; i < ruip->rui_format.rui_nextents; i++) { 467 rmap = &ruip->rui_format.rui_extents[i]; 468 state = (rmap->me_flags & XFS_RMAP_EXTENT_UNWRITTEN) ? 469 XFS_EXT_UNWRITTEN : XFS_EXT_NORM; 470 whichfork = (rmap->me_flags & XFS_RMAP_EXTENT_ATTR_FORK) ? 471 XFS_ATTR_FORK : XFS_DATA_FORK; 472 switch (rmap->me_flags & XFS_RMAP_EXTENT_TYPE_MASK) { 473 case XFS_RMAP_EXTENT_MAP: 474 type = XFS_RMAP_MAP; 475 break; 476 case XFS_RMAP_EXTENT_MAP_SHARED: 477 type = XFS_RMAP_MAP_SHARED; 478 break; 479 case XFS_RMAP_EXTENT_UNMAP: 480 type = XFS_RMAP_UNMAP; 481 break; 482 case XFS_RMAP_EXTENT_UNMAP_SHARED: 483 type = XFS_RMAP_UNMAP_SHARED; 484 break; 485 case XFS_RMAP_EXTENT_CONVERT: 486 type = XFS_RMAP_CONVERT; 487 break; 488 case XFS_RMAP_EXTENT_CONVERT_SHARED: 489 type = XFS_RMAP_CONVERT_SHARED; 490 break; 491 case XFS_RMAP_EXTENT_ALLOC: 492 type = XFS_RMAP_ALLOC; 493 break; 494 case XFS_RMAP_EXTENT_FREE: 495 type = XFS_RMAP_FREE; 496 break; 497 default: 498 error = -EFSCORRUPTED; 499 goto abort_error; 500 } 501 error = xfs_trans_log_finish_rmap_update(tp, rudp, type, 502 rmap->me_owner, whichfork, 503 rmap->me_startoff, rmap->me_startblock, 504 rmap->me_len, state, &rcur); 505 if (error) 506 goto abort_error; 507 508 } 509 510 xfs_rmap_finish_one_cleanup(tp, rcur, error); 511 set_bit(XFS_RUI_RECOVERED, &ruip->rui_flags); 512 error = xfs_trans_commit(tp); 513 return error; 514 515 abort_error: 516 xfs_rmap_finish_one_cleanup(tp, rcur, error); 517 xfs_trans_cancel(tp); 518 return error; 519 } 520