| 1 | /* |
| 2 | * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc. |
| 3 | * All Rights Reserved. |
| 4 | * |
| 5 | * This program is free software; you can redistribute it and/or |
| 6 | * modify it under the terms of the GNU General Public License as |
| 7 | * published by the Free Software Foundation. |
| 8 | * |
| 9 | * This program is distributed in the hope that it would be useful, |
| 10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 12 | * GNU General Public License for more details. |
| 13 | * |
| 14 | * You should have received a copy of the GNU General Public License |
| 15 | * along with this program; if not, write the Free Software Foundation, |
| 16 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA |
| 17 | */ |
| 18 | #include "xfs.h" |
| 19 | #include "xfs_fs.h" |
| 20 | #include "xfs_types.h" |
| 21 | #include "xfs_log.h" |
| 22 | #include "xfs_trans.h" |
| 23 | #include "xfs_sb.h" |
| 24 | #include "xfs_ag.h" |
| 25 | #include "xfs_mount.h" |
| 26 | #include "xfs_bmap_btree.h" |
| 27 | #include "xfs_alloc_btree.h" |
| 28 | #include "xfs_ialloc_btree.h" |
| 29 | #include "xfs_dinode.h" |
| 30 | #include "xfs_inode.h" |
| 31 | #include "xfs_buf_item.h" |
| 32 | #include "xfs_trans_priv.h" |
| 33 | #include "xfs_error.h" |
| 34 | #include "xfs_trace.h" |
| 35 | |
| 36 | /* |
| 37 | * Check to see if a buffer matching the given parameters is already |
| 38 | * a part of the given transaction. |
| 39 | */ |
| 40 | STATIC struct xfs_buf * |
| 41 | xfs_trans_buf_item_match( |
| 42 | struct xfs_trans *tp, |
| 43 | struct xfs_buftarg *target, |
| 44 | struct xfs_buf_map *map, |
| 45 | int nmaps) |
| 46 | { |
| 47 | struct xfs_log_item_desc *lidp; |
| 48 | struct xfs_buf_log_item *blip; |
| 49 | int len = 0; |
| 50 | int i; |
| 51 | |
| 52 | for (i = 0; i < nmaps; i++) |
| 53 | len += map[i].bm_len; |
| 54 | |
| 55 | list_for_each_entry(lidp, &tp->t_items, lid_trans) { |
| 56 | blip = (struct xfs_buf_log_item *)lidp->lid_item; |
| 57 | if (blip->bli_item.li_type == XFS_LI_BUF && |
| 58 | blip->bli_buf->b_target == target && |
| 59 | XFS_BUF_ADDR(blip->bli_buf) == map[0].bm_bn && |
| 60 | blip->bli_buf->b_length == len) { |
| 61 | ASSERT(blip->bli_buf->b_map_count == nmaps); |
| 62 | return blip->bli_buf; |
| 63 | } |
| 64 | } |
| 65 | |
| 66 | return NULL; |
| 67 | } |
| 68 | |
| 69 | /* |
| 70 | * Add the locked buffer to the transaction. |
| 71 | * |
| 72 | * The buffer must be locked, and it cannot be associated with any |
| 73 | * transaction. |
| 74 | * |
| 75 | * If the buffer does not yet have a buf log item associated with it, |
| 76 | * then allocate one for it. Then add the buf item to the transaction. |
| 77 | */ |
| 78 | STATIC void |
| 79 | _xfs_trans_bjoin( |
| 80 | struct xfs_trans *tp, |
| 81 | struct xfs_buf *bp, |
| 82 | int reset_recur) |
| 83 | { |
| 84 | struct xfs_buf_log_item *bip; |
| 85 | |
| 86 | ASSERT(bp->b_transp == NULL); |
| 87 | |
| 88 | /* |
| 89 | * The xfs_buf_log_item pointer is stored in b_fsprivate. If |
| 90 | * it doesn't have one yet, then allocate one and initialize it. |
| 91 | * The checks to see if one is there are in xfs_buf_item_init(). |
| 92 | */ |
| 93 | xfs_buf_item_init(bp, tp->t_mountp); |
| 94 | bip = bp->b_fspriv; |
| 95 | ASSERT(!(bip->bli_flags & XFS_BLI_STALE)); |
| 96 | ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_CANCEL)); |
| 97 | ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED)); |
| 98 | if (reset_recur) |
| 99 | bip->bli_recur = 0; |
| 100 | |
| 101 | /* |
| 102 | * Take a reference for this transaction on the buf item. |
| 103 | */ |
| 104 | atomic_inc(&bip->bli_refcount); |
| 105 | |
| 106 | /* |
| 107 | * Get a log_item_desc to point at the new item. |
| 108 | */ |
| 109 | xfs_trans_add_item(tp, &bip->bli_item); |
| 110 | |
| 111 | /* |
| 112 | * Initialize b_fsprivate2 so we can find it with incore_match() |
| 113 | * in xfs_trans_get_buf() and friends above. |
| 114 | */ |
| 115 | bp->b_transp = tp; |
| 116 | |
| 117 | } |
| 118 | |
| 119 | void |
| 120 | xfs_trans_bjoin( |
| 121 | struct xfs_trans *tp, |
| 122 | struct xfs_buf *bp) |
| 123 | { |
| 124 | _xfs_trans_bjoin(tp, bp, 0); |
| 125 | trace_xfs_trans_bjoin(bp->b_fspriv); |
| 126 | } |
| 127 | |
| 128 | /* |
| 129 | * Get and lock the buffer for the caller if it is not already |
| 130 | * locked within the given transaction. If it is already locked |
| 131 | * within the transaction, just increment its lock recursion count |
| 132 | * and return a pointer to it. |
| 133 | * |
| 134 | * If the transaction pointer is NULL, make this just a normal |
| 135 | * get_buf() call. |
| 136 | */ |
| 137 | struct xfs_buf * |
| 138 | xfs_trans_get_buf_map( |
| 139 | struct xfs_trans *tp, |
| 140 | struct xfs_buftarg *target, |
| 141 | struct xfs_buf_map *map, |
| 142 | int nmaps, |
| 143 | xfs_buf_flags_t flags) |
| 144 | { |
| 145 | xfs_buf_t *bp; |
| 146 | xfs_buf_log_item_t *bip; |
| 147 | |
| 148 | if (!tp) |
| 149 | return xfs_buf_get_map(target, map, nmaps, flags); |
| 150 | |
| 151 | /* |
| 152 | * If we find the buffer in the cache with this transaction |
| 153 | * pointer in its b_fsprivate2 field, then we know we already |
| 154 | * have it locked. In this case we just increment the lock |
| 155 | * recursion count and return the buffer to the caller. |
| 156 | */ |
| 157 | bp = xfs_trans_buf_item_match(tp, target, map, nmaps); |
| 158 | if (bp != NULL) { |
| 159 | ASSERT(xfs_buf_islocked(bp)); |
| 160 | if (XFS_FORCED_SHUTDOWN(tp->t_mountp)) { |
| 161 | xfs_buf_stale(bp); |
| 162 | XFS_BUF_DONE(bp); |
| 163 | } |
| 164 | |
| 165 | ASSERT(bp->b_transp == tp); |
| 166 | bip = bp->b_fspriv; |
| 167 | ASSERT(bip != NULL); |
| 168 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
| 169 | bip->bli_recur++; |
| 170 | trace_xfs_trans_get_buf_recur(bip); |
| 171 | return (bp); |
| 172 | } |
| 173 | |
| 174 | bp = xfs_buf_get_map(target, map, nmaps, flags); |
| 175 | if (bp == NULL) { |
| 176 | return NULL; |
| 177 | } |
| 178 | |
| 179 | ASSERT(!bp->b_error); |
| 180 | |
| 181 | _xfs_trans_bjoin(tp, bp, 1); |
| 182 | trace_xfs_trans_get_buf(bp->b_fspriv); |
| 183 | return (bp); |
| 184 | } |
| 185 | |
| 186 | /* |
| 187 | * Get and lock the superblock buffer of this file system for the |
| 188 | * given transaction. |
| 189 | * |
| 190 | * We don't need to use incore_match() here, because the superblock |
| 191 | * buffer is a private buffer which we keep a pointer to in the |
| 192 | * mount structure. |
| 193 | */ |
| 194 | xfs_buf_t * |
| 195 | xfs_trans_getsb(xfs_trans_t *tp, |
| 196 | struct xfs_mount *mp, |
| 197 | int flags) |
| 198 | { |
| 199 | xfs_buf_t *bp; |
| 200 | xfs_buf_log_item_t *bip; |
| 201 | |
| 202 | /* |
| 203 | * Default to just trying to lock the superblock buffer |
| 204 | * if tp is NULL. |
| 205 | */ |
| 206 | if (tp == NULL) { |
| 207 | return (xfs_getsb(mp, flags)); |
| 208 | } |
| 209 | |
| 210 | /* |
| 211 | * If the superblock buffer already has this transaction |
| 212 | * pointer in its b_fsprivate2 field, then we know we already |
| 213 | * have it locked. In this case we just increment the lock |
| 214 | * recursion count and return the buffer to the caller. |
| 215 | */ |
| 216 | bp = mp->m_sb_bp; |
| 217 | if (bp->b_transp == tp) { |
| 218 | bip = bp->b_fspriv; |
| 219 | ASSERT(bip != NULL); |
| 220 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
| 221 | bip->bli_recur++; |
| 222 | trace_xfs_trans_getsb_recur(bip); |
| 223 | return (bp); |
| 224 | } |
| 225 | |
| 226 | bp = xfs_getsb(mp, flags); |
| 227 | if (bp == NULL) |
| 228 | return NULL; |
| 229 | |
| 230 | _xfs_trans_bjoin(tp, bp, 1); |
| 231 | trace_xfs_trans_getsb(bp->b_fspriv); |
| 232 | return (bp); |
| 233 | } |
| 234 | |
| 235 | #ifdef DEBUG |
| 236 | xfs_buftarg_t *xfs_error_target; |
| 237 | int xfs_do_error; |
| 238 | int xfs_req_num; |
| 239 | int xfs_error_mod = 33; |
| 240 | #endif |
| 241 | |
| 242 | /* |
| 243 | * Get and lock the buffer for the caller if it is not already |
| 244 | * locked within the given transaction. If it has not yet been |
| 245 | * read in, read it from disk. If it is already locked |
| 246 | * within the transaction and already read in, just increment its |
| 247 | * lock recursion count and return a pointer to it. |
| 248 | * |
| 249 | * If the transaction pointer is NULL, make this just a normal |
| 250 | * read_buf() call. |
| 251 | */ |
| 252 | int |
| 253 | xfs_trans_read_buf_map( |
| 254 | struct xfs_mount *mp, |
| 255 | struct xfs_trans *tp, |
| 256 | struct xfs_buftarg *target, |
| 257 | struct xfs_buf_map *map, |
| 258 | int nmaps, |
| 259 | xfs_buf_flags_t flags, |
| 260 | struct xfs_buf **bpp, |
| 261 | const struct xfs_buf_ops *ops) |
| 262 | { |
| 263 | xfs_buf_t *bp; |
| 264 | xfs_buf_log_item_t *bip; |
| 265 | int error; |
| 266 | |
| 267 | *bpp = NULL; |
| 268 | if (!tp) { |
| 269 | bp = xfs_buf_read_map(target, map, nmaps, flags, ops); |
| 270 | if (!bp) |
| 271 | return (flags & XBF_TRYLOCK) ? |
| 272 | EAGAIN : XFS_ERROR(ENOMEM); |
| 273 | |
| 274 | if (bp->b_error) { |
| 275 | error = bp->b_error; |
| 276 | xfs_buf_ioerror_alert(bp, __func__); |
| 277 | XFS_BUF_UNDONE(bp); |
| 278 | xfs_buf_stale(bp); |
| 279 | xfs_buf_relse(bp); |
| 280 | return error; |
| 281 | } |
| 282 | #ifdef DEBUG |
| 283 | if (xfs_do_error) { |
| 284 | if (xfs_error_target == target) { |
| 285 | if (((xfs_req_num++) % xfs_error_mod) == 0) { |
| 286 | xfs_buf_relse(bp); |
| 287 | xfs_debug(mp, "Returning error!"); |
| 288 | return XFS_ERROR(EIO); |
| 289 | } |
| 290 | } |
| 291 | } |
| 292 | #endif |
| 293 | if (XFS_FORCED_SHUTDOWN(mp)) |
| 294 | goto shutdown_abort; |
| 295 | *bpp = bp; |
| 296 | return 0; |
| 297 | } |
| 298 | |
| 299 | /* |
| 300 | * If we find the buffer in the cache with this transaction |
| 301 | * pointer in its b_fsprivate2 field, then we know we already |
| 302 | * have it locked. If it is already read in we just increment |
| 303 | * the lock recursion count and return the buffer to the caller. |
| 304 | * If the buffer is not yet read in, then we read it in, increment |
| 305 | * the lock recursion count, and return it to the caller. |
| 306 | */ |
| 307 | bp = xfs_trans_buf_item_match(tp, target, map, nmaps); |
| 308 | if (bp != NULL) { |
| 309 | ASSERT(xfs_buf_islocked(bp)); |
| 310 | ASSERT(bp->b_transp == tp); |
| 311 | ASSERT(bp->b_fspriv != NULL); |
| 312 | ASSERT(!bp->b_error); |
| 313 | if (!(XFS_BUF_ISDONE(bp))) { |
| 314 | trace_xfs_trans_read_buf_io(bp, _RET_IP_); |
| 315 | ASSERT(!XFS_BUF_ISASYNC(bp)); |
| 316 | ASSERT(bp->b_iodone == NULL); |
| 317 | XFS_BUF_READ(bp); |
| 318 | bp->b_ops = ops; |
| 319 | xfsbdstrat(tp->t_mountp, bp); |
| 320 | error = xfs_buf_iowait(bp); |
| 321 | if (error) { |
| 322 | xfs_buf_ioerror_alert(bp, __func__); |
| 323 | xfs_buf_relse(bp); |
| 324 | /* |
| 325 | * We can gracefully recover from most read |
| 326 | * errors. Ones we can't are those that happen |
| 327 | * after the transaction's already dirty. |
| 328 | */ |
| 329 | if (tp->t_flags & XFS_TRANS_DIRTY) |
| 330 | xfs_force_shutdown(tp->t_mountp, |
| 331 | SHUTDOWN_META_IO_ERROR); |
| 332 | return error; |
| 333 | } |
| 334 | } |
| 335 | /* |
| 336 | * We never locked this buf ourselves, so we shouldn't |
| 337 | * brelse it either. Just get out. |
| 338 | */ |
| 339 | if (XFS_FORCED_SHUTDOWN(mp)) { |
| 340 | trace_xfs_trans_read_buf_shut(bp, _RET_IP_); |
| 341 | *bpp = NULL; |
| 342 | return XFS_ERROR(EIO); |
| 343 | } |
| 344 | |
| 345 | |
| 346 | bip = bp->b_fspriv; |
| 347 | bip->bli_recur++; |
| 348 | |
| 349 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
| 350 | trace_xfs_trans_read_buf_recur(bip); |
| 351 | *bpp = bp; |
| 352 | return 0; |
| 353 | } |
| 354 | |
| 355 | bp = xfs_buf_read_map(target, map, nmaps, flags, ops); |
| 356 | if (bp == NULL) { |
| 357 | *bpp = NULL; |
| 358 | return (flags & XBF_TRYLOCK) ? |
| 359 | 0 : XFS_ERROR(ENOMEM); |
| 360 | } |
| 361 | if (bp->b_error) { |
| 362 | error = bp->b_error; |
| 363 | xfs_buf_stale(bp); |
| 364 | XFS_BUF_DONE(bp); |
| 365 | xfs_buf_ioerror_alert(bp, __func__); |
| 366 | if (tp->t_flags & XFS_TRANS_DIRTY) |
| 367 | xfs_force_shutdown(tp->t_mountp, SHUTDOWN_META_IO_ERROR); |
| 368 | xfs_buf_relse(bp); |
| 369 | return error; |
| 370 | } |
| 371 | #ifdef DEBUG |
| 372 | if (xfs_do_error && !(tp->t_flags & XFS_TRANS_DIRTY)) { |
| 373 | if (xfs_error_target == target) { |
| 374 | if (((xfs_req_num++) % xfs_error_mod) == 0) { |
| 375 | xfs_force_shutdown(tp->t_mountp, |
| 376 | SHUTDOWN_META_IO_ERROR); |
| 377 | xfs_buf_relse(bp); |
| 378 | xfs_debug(mp, "Returning trans error!"); |
| 379 | return XFS_ERROR(EIO); |
| 380 | } |
| 381 | } |
| 382 | } |
| 383 | #endif |
| 384 | if (XFS_FORCED_SHUTDOWN(mp)) |
| 385 | goto shutdown_abort; |
| 386 | |
| 387 | _xfs_trans_bjoin(tp, bp, 1); |
| 388 | trace_xfs_trans_read_buf(bp->b_fspriv); |
| 389 | |
| 390 | *bpp = bp; |
| 391 | return 0; |
| 392 | |
| 393 | shutdown_abort: |
| 394 | trace_xfs_trans_read_buf_shut(bp, _RET_IP_); |
| 395 | xfs_buf_relse(bp); |
| 396 | *bpp = NULL; |
| 397 | return XFS_ERROR(EIO); |
| 398 | } |
| 399 | |
| 400 | |
| 401 | /* |
| 402 | * Release the buffer bp which was previously acquired with one of the |
| 403 | * xfs_trans_... buffer allocation routines if the buffer has not |
| 404 | * been modified within this transaction. If the buffer is modified |
| 405 | * within this transaction, do decrement the recursion count but do |
| 406 | * not release the buffer even if the count goes to 0. If the buffer is not |
| 407 | * modified within the transaction, decrement the recursion count and |
| 408 | * release the buffer if the recursion count goes to 0. |
| 409 | * |
| 410 | * If the buffer is to be released and it was not modified before |
| 411 | * this transaction began, then free the buf_log_item associated with it. |
| 412 | * |
| 413 | * If the transaction pointer is NULL, make this just a normal |
| 414 | * brelse() call. |
| 415 | */ |
| 416 | void |
| 417 | xfs_trans_brelse(xfs_trans_t *tp, |
| 418 | xfs_buf_t *bp) |
| 419 | { |
| 420 | xfs_buf_log_item_t *bip; |
| 421 | |
| 422 | /* |
| 423 | * Default to a normal brelse() call if the tp is NULL. |
| 424 | */ |
| 425 | if (tp == NULL) { |
| 426 | ASSERT(bp->b_transp == NULL); |
| 427 | xfs_buf_relse(bp); |
| 428 | return; |
| 429 | } |
| 430 | |
| 431 | ASSERT(bp->b_transp == tp); |
| 432 | bip = bp->b_fspriv; |
| 433 | ASSERT(bip->bli_item.li_type == XFS_LI_BUF); |
| 434 | ASSERT(!(bip->bli_flags & XFS_BLI_STALE)); |
| 435 | ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_CANCEL)); |
| 436 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
| 437 | |
| 438 | trace_xfs_trans_brelse(bip); |
| 439 | |
| 440 | /* |
| 441 | * If the release is just for a recursive lock, |
| 442 | * then decrement the count and return. |
| 443 | */ |
| 444 | if (bip->bli_recur > 0) { |
| 445 | bip->bli_recur--; |
| 446 | return; |
| 447 | } |
| 448 | |
| 449 | /* |
| 450 | * If the buffer is dirty within this transaction, we can't |
| 451 | * release it until we commit. |
| 452 | */ |
| 453 | if (bip->bli_item.li_desc->lid_flags & XFS_LID_DIRTY) |
| 454 | return; |
| 455 | |
| 456 | /* |
| 457 | * If the buffer has been invalidated, then we can't release |
| 458 | * it until the transaction commits to disk unless it is re-dirtied |
| 459 | * as part of this transaction. This prevents us from pulling |
| 460 | * the item from the AIL before we should. |
| 461 | */ |
| 462 | if (bip->bli_flags & XFS_BLI_STALE) |
| 463 | return; |
| 464 | |
| 465 | ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED)); |
| 466 | |
| 467 | /* |
| 468 | * Free up the log item descriptor tracking the released item. |
| 469 | */ |
| 470 | xfs_trans_del_item(&bip->bli_item); |
| 471 | |
| 472 | /* |
| 473 | * Clear the hold flag in the buf log item if it is set. |
| 474 | * We wouldn't want the next user of the buffer to |
| 475 | * get confused. |
| 476 | */ |
| 477 | if (bip->bli_flags & XFS_BLI_HOLD) { |
| 478 | bip->bli_flags &= ~XFS_BLI_HOLD; |
| 479 | } |
| 480 | |
| 481 | /* |
| 482 | * Drop our reference to the buf log item. |
| 483 | */ |
| 484 | atomic_dec(&bip->bli_refcount); |
| 485 | |
| 486 | /* |
| 487 | * If the buf item is not tracking data in the log, then |
| 488 | * we must free it before releasing the buffer back to the |
| 489 | * free pool. Before releasing the buffer to the free pool, |
| 490 | * clear the transaction pointer in b_fsprivate2 to dissolve |
| 491 | * its relation to this transaction. |
| 492 | */ |
| 493 | if (!xfs_buf_item_dirty(bip)) { |
| 494 | /*** |
| 495 | ASSERT(bp->b_pincount == 0); |
| 496 | ***/ |
| 497 | ASSERT(atomic_read(&bip->bli_refcount) == 0); |
| 498 | ASSERT(!(bip->bli_item.li_flags & XFS_LI_IN_AIL)); |
| 499 | ASSERT(!(bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF)); |
| 500 | xfs_buf_item_relse(bp); |
| 501 | } |
| 502 | |
| 503 | bp->b_transp = NULL; |
| 504 | xfs_buf_relse(bp); |
| 505 | } |
| 506 | |
| 507 | /* |
| 508 | * Mark the buffer as not needing to be unlocked when the buf item's |
| 509 | * IOP_UNLOCK() routine is called. The buffer must already be locked |
| 510 | * and associated with the given transaction. |
| 511 | */ |
| 512 | /* ARGSUSED */ |
| 513 | void |
| 514 | xfs_trans_bhold(xfs_trans_t *tp, |
| 515 | xfs_buf_t *bp) |
| 516 | { |
| 517 | xfs_buf_log_item_t *bip = bp->b_fspriv; |
| 518 | |
| 519 | ASSERT(bp->b_transp == tp); |
| 520 | ASSERT(bip != NULL); |
| 521 | ASSERT(!(bip->bli_flags & XFS_BLI_STALE)); |
| 522 | ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_CANCEL)); |
| 523 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
| 524 | |
| 525 | bip->bli_flags |= XFS_BLI_HOLD; |
| 526 | trace_xfs_trans_bhold(bip); |
| 527 | } |
| 528 | |
| 529 | /* |
| 530 | * Cancel the previous buffer hold request made on this buffer |
| 531 | * for this transaction. |
| 532 | */ |
| 533 | void |
| 534 | xfs_trans_bhold_release(xfs_trans_t *tp, |
| 535 | xfs_buf_t *bp) |
| 536 | { |
| 537 | xfs_buf_log_item_t *bip = bp->b_fspriv; |
| 538 | |
| 539 | ASSERT(bp->b_transp == tp); |
| 540 | ASSERT(bip != NULL); |
| 541 | ASSERT(!(bip->bli_flags & XFS_BLI_STALE)); |
| 542 | ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_CANCEL)); |
| 543 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
| 544 | ASSERT(bip->bli_flags & XFS_BLI_HOLD); |
| 545 | |
| 546 | bip->bli_flags &= ~XFS_BLI_HOLD; |
| 547 | trace_xfs_trans_bhold_release(bip); |
| 548 | } |
| 549 | |
| 550 | /* |
| 551 | * This is called to mark bytes first through last inclusive of the given |
| 552 | * buffer as needing to be logged when the transaction is committed. |
| 553 | * The buffer must already be associated with the given transaction. |
| 554 | * |
| 555 | * First and last are numbers relative to the beginning of this buffer, |
| 556 | * so the first byte in the buffer is numbered 0 regardless of the |
| 557 | * value of b_blkno. |
| 558 | */ |
| 559 | void |
| 560 | xfs_trans_log_buf(xfs_trans_t *tp, |
| 561 | xfs_buf_t *bp, |
| 562 | uint first, |
| 563 | uint last) |
| 564 | { |
| 565 | xfs_buf_log_item_t *bip = bp->b_fspriv; |
| 566 | |
| 567 | ASSERT(bp->b_transp == tp); |
| 568 | ASSERT(bip != NULL); |
| 569 | ASSERT(first <= last && last < BBTOB(bp->b_length)); |
| 570 | ASSERT(bp->b_iodone == NULL || |
| 571 | bp->b_iodone == xfs_buf_iodone_callbacks); |
| 572 | |
| 573 | /* |
| 574 | * Mark the buffer as needing to be written out eventually, |
| 575 | * and set its iodone function to remove the buffer's buf log |
| 576 | * item from the AIL and free it when the buffer is flushed |
| 577 | * to disk. See xfs_buf_attach_iodone() for more details |
| 578 | * on li_cb and xfs_buf_iodone_callbacks(). |
| 579 | * If we end up aborting this transaction, we trap this buffer |
| 580 | * inside the b_bdstrat callback so that this won't get written to |
| 581 | * disk. |
| 582 | */ |
| 583 | XFS_BUF_DONE(bp); |
| 584 | |
| 585 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
| 586 | bp->b_iodone = xfs_buf_iodone_callbacks; |
| 587 | bip->bli_item.li_cb = xfs_buf_iodone; |
| 588 | |
| 589 | trace_xfs_trans_log_buf(bip); |
| 590 | |
| 591 | /* |
| 592 | * If we invalidated the buffer within this transaction, then |
| 593 | * cancel the invalidation now that we're dirtying the buffer |
| 594 | * again. There are no races with the code in xfs_buf_item_unpin(), |
| 595 | * because we have a reference to the buffer this entire time. |
| 596 | */ |
| 597 | if (bip->bli_flags & XFS_BLI_STALE) { |
| 598 | bip->bli_flags &= ~XFS_BLI_STALE; |
| 599 | ASSERT(XFS_BUF_ISSTALE(bp)); |
| 600 | XFS_BUF_UNSTALE(bp); |
| 601 | bip->__bli_format.blf_flags &= ~XFS_BLF_CANCEL; |
| 602 | } |
| 603 | |
| 604 | tp->t_flags |= XFS_TRANS_DIRTY; |
| 605 | bip->bli_item.li_desc->lid_flags |= XFS_LID_DIRTY; |
| 606 | bip->bli_flags |= XFS_BLI_LOGGED; |
| 607 | xfs_buf_item_log(bip, first, last); |
| 608 | } |
| 609 | |
| 610 | |
| 611 | /* |
| 612 | * Invalidate a buffer that is being used within a transaction. |
| 613 | * |
| 614 | * Typically this is because the blocks in the buffer are being freed, so we |
| 615 | * need to prevent it from being written out when we're done. Allowing it |
| 616 | * to be written again might overwrite data in the free blocks if they are |
| 617 | * reallocated to a file. |
| 618 | * |
| 619 | * We prevent the buffer from being written out by marking it stale. We can't |
| 620 | * get rid of the buf log item at this point because the buffer may still be |
| 621 | * pinned by another transaction. If that is the case, then we'll wait until |
| 622 | * the buffer is committed to disk for the last time (we can tell by the ref |
| 623 | * count) and free it in xfs_buf_item_unpin(). Until that happens we will |
| 624 | * keep the buffer locked so that the buffer and buf log item are not reused. |
| 625 | * |
| 626 | * We also set the XFS_BLF_CANCEL flag in the buf log format structure and log |
| 627 | * the buf item. This will be used at recovery time to determine that copies |
| 628 | * of the buffer in the log before this should not be replayed. |
| 629 | * |
| 630 | * We mark the item descriptor and the transaction dirty so that we'll hold |
| 631 | * the buffer until after the commit. |
| 632 | * |
| 633 | * Since we're invalidating the buffer, we also clear the state about which |
| 634 | * parts of the buffer have been logged. We also clear the flag indicating |
| 635 | * that this is an inode buffer since the data in the buffer will no longer |
| 636 | * be valid. |
| 637 | * |
| 638 | * We set the stale bit in the buffer as well since we're getting rid of it. |
| 639 | */ |
| 640 | void |
| 641 | xfs_trans_binval( |
| 642 | xfs_trans_t *tp, |
| 643 | xfs_buf_t *bp) |
| 644 | { |
| 645 | xfs_buf_log_item_t *bip = bp->b_fspriv; |
| 646 | int i; |
| 647 | |
| 648 | ASSERT(bp->b_transp == tp); |
| 649 | ASSERT(bip != NULL); |
| 650 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
| 651 | |
| 652 | trace_xfs_trans_binval(bip); |
| 653 | |
| 654 | if (bip->bli_flags & XFS_BLI_STALE) { |
| 655 | /* |
| 656 | * If the buffer is already invalidated, then |
| 657 | * just return. |
| 658 | */ |
| 659 | ASSERT(XFS_BUF_ISSTALE(bp)); |
| 660 | ASSERT(!(bip->bli_flags & (XFS_BLI_LOGGED | XFS_BLI_DIRTY))); |
| 661 | ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_INODE_BUF)); |
| 662 | ASSERT(bip->__bli_format.blf_flags & XFS_BLF_CANCEL); |
| 663 | ASSERT(bip->bli_item.li_desc->lid_flags & XFS_LID_DIRTY); |
| 664 | ASSERT(tp->t_flags & XFS_TRANS_DIRTY); |
| 665 | return; |
| 666 | } |
| 667 | |
| 668 | xfs_buf_stale(bp); |
| 669 | |
| 670 | bip->bli_flags |= XFS_BLI_STALE; |
| 671 | bip->bli_flags &= ~(XFS_BLI_INODE_BUF | XFS_BLI_LOGGED | XFS_BLI_DIRTY); |
| 672 | bip->__bli_format.blf_flags &= ~XFS_BLF_INODE_BUF; |
| 673 | bip->__bli_format.blf_flags |= XFS_BLF_CANCEL; |
| 674 | for (i = 0; i < bip->bli_format_count; i++) { |
| 675 | memset(bip->bli_formats[i].blf_data_map, 0, |
| 676 | (bip->bli_formats[i].blf_map_size * sizeof(uint))); |
| 677 | } |
| 678 | bip->bli_item.li_desc->lid_flags |= XFS_LID_DIRTY; |
| 679 | tp->t_flags |= XFS_TRANS_DIRTY; |
| 680 | } |
| 681 | |
| 682 | /* |
| 683 | * This call is used to indicate that the buffer contains on-disk inodes which |
| 684 | * must be handled specially during recovery. They require special handling |
| 685 | * because only the di_next_unlinked from the inodes in the buffer should be |
| 686 | * recovered. The rest of the data in the buffer is logged via the inodes |
| 687 | * themselves. |
| 688 | * |
| 689 | * All we do is set the XFS_BLI_INODE_BUF flag in the items flags so it can be |
| 690 | * transferred to the buffer's log format structure so that we'll know what to |
| 691 | * do at recovery time. |
| 692 | */ |
| 693 | void |
| 694 | xfs_trans_inode_buf( |
| 695 | xfs_trans_t *tp, |
| 696 | xfs_buf_t *bp) |
| 697 | { |
| 698 | xfs_buf_log_item_t *bip = bp->b_fspriv; |
| 699 | |
| 700 | ASSERT(bp->b_transp == tp); |
| 701 | ASSERT(bip != NULL); |
| 702 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
| 703 | |
| 704 | bip->bli_flags |= XFS_BLI_INODE_BUF; |
| 705 | } |
| 706 | |
| 707 | /* |
| 708 | * This call is used to indicate that the buffer is going to |
| 709 | * be staled and was an inode buffer. This means it gets |
| 710 | * special processing during unpin - where any inodes |
| 711 | * associated with the buffer should be removed from ail. |
| 712 | * There is also special processing during recovery, |
| 713 | * any replay of the inodes in the buffer needs to be |
| 714 | * prevented as the buffer may have been reused. |
| 715 | */ |
| 716 | void |
| 717 | xfs_trans_stale_inode_buf( |
| 718 | xfs_trans_t *tp, |
| 719 | xfs_buf_t *bp) |
| 720 | { |
| 721 | xfs_buf_log_item_t *bip = bp->b_fspriv; |
| 722 | |
| 723 | ASSERT(bp->b_transp == tp); |
| 724 | ASSERT(bip != NULL); |
| 725 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
| 726 | |
| 727 | bip->bli_flags |= XFS_BLI_STALE_INODE; |
| 728 | bip->bli_item.li_cb = xfs_buf_iodone; |
| 729 | } |
| 730 | |
| 731 | /* |
| 732 | * Mark the buffer as being one which contains newly allocated |
| 733 | * inodes. We need to make sure that even if this buffer is |
| 734 | * relogged as an 'inode buf' we still recover all of the inode |
| 735 | * images in the face of a crash. This works in coordination with |
| 736 | * xfs_buf_item_committed() to ensure that the buffer remains in the |
| 737 | * AIL at its original location even after it has been relogged. |
| 738 | */ |
| 739 | /* ARGSUSED */ |
| 740 | void |
| 741 | xfs_trans_inode_alloc_buf( |
| 742 | xfs_trans_t *tp, |
| 743 | xfs_buf_t *bp) |
| 744 | { |
| 745 | xfs_buf_log_item_t *bip = bp->b_fspriv; |
| 746 | |
| 747 | ASSERT(bp->b_transp == tp); |
| 748 | ASSERT(bip != NULL); |
| 749 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
| 750 | |
| 751 | bip->bli_flags |= XFS_BLI_INODE_ALLOC_BUF; |
| 752 | } |
| 753 | |
| 754 | |
| 755 | /* |
| 756 | * Similar to xfs_trans_inode_buf(), this marks the buffer as a cluster of |
| 757 | * dquots. However, unlike in inode buffer recovery, dquot buffers get |
| 758 | * recovered in their entirety. (Hence, no XFS_BLI_DQUOT_ALLOC_BUF flag). |
| 759 | * The only thing that makes dquot buffers different from regular |
| 760 | * buffers is that we must not replay dquot bufs when recovering |
| 761 | * if a _corresponding_ quotaoff has happened. We also have to distinguish |
| 762 | * between usr dquot bufs and grp dquot bufs, because usr and grp quotas |
| 763 | * can be turned off independently. |
| 764 | */ |
| 765 | /* ARGSUSED */ |
| 766 | void |
| 767 | xfs_trans_dquot_buf( |
| 768 | xfs_trans_t *tp, |
| 769 | xfs_buf_t *bp, |
| 770 | uint type) |
| 771 | { |
| 772 | xfs_buf_log_item_t *bip = bp->b_fspriv; |
| 773 | |
| 774 | ASSERT(bp->b_transp == tp); |
| 775 | ASSERT(bip != NULL); |
| 776 | ASSERT(type == XFS_BLF_UDQUOT_BUF || |
| 777 | type == XFS_BLF_PDQUOT_BUF || |
| 778 | type == XFS_BLF_GDQUOT_BUF); |
| 779 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
| 780 | |
| 781 | bip->__bli_format.blf_flags |= type; |
| 782 | } |