Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
Nathan Scott | 7b71876 | 2005-11-02 14:58:39 +1100 | [diff] [blame] | 2 | * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc. |
| 3 | * All Rights Reserved. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4 | * |
Nathan Scott | 7b71876 | 2005-11-02 14:58:39 +1100 | [diff] [blame] | 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 |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7 | * published by the Free Software Foundation. |
| 8 | * |
Nathan Scott | 7b71876 | 2005-11-02 14:58:39 +1100 | [diff] [blame] | 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. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 13 | * |
Nathan Scott | 7b71876 | 2005-11-02 14:58:39 +1100 | [diff] [blame] | 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 |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 17 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 18 | #include "xfs.h" |
Nathan Scott | a844f45 | 2005-11-02 14:38:42 +1100 | [diff] [blame] | 19 | #include "xfs_fs.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 20 | #include "xfs_types.h" |
Nathan Scott | a844f45 | 2005-11-02 14:38:42 +1100 | [diff] [blame] | 21 | #include "xfs_bit.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 22 | #include "xfs_log.h" |
Nathan Scott | a844f45 | 2005-11-02 14:38:42 +1100 | [diff] [blame] | 23 | #include "xfs_inum.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 24 | #include "xfs_trans.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 25 | #include "xfs_sb.h" |
| 26 | #include "xfs_ag.h" |
Nathan Scott | a844f45 | 2005-11-02 14:38:42 +1100 | [diff] [blame] | 27 | #include "xfs_dir2.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 28 | #include "xfs_dmapi.h" |
| 29 | #include "xfs_mount.h" |
Nathan Scott | a844f45 | 2005-11-02 14:38:42 +1100 | [diff] [blame] | 30 | #include "xfs_bmap_btree.h" |
| 31 | #include "xfs_alloc_btree.h" |
| 32 | #include "xfs_ialloc_btree.h" |
Nathan Scott | a844f45 | 2005-11-02 14:38:42 +1100 | [diff] [blame] | 33 | #include "xfs_dir2_sf.h" |
| 34 | #include "xfs_attr_sf.h" |
| 35 | #include "xfs_dinode.h" |
| 36 | #include "xfs_inode.h" |
| 37 | #include "xfs_buf_item.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 38 | #include "xfs_trans_priv.h" |
| 39 | #include "xfs_error.h" |
| 40 | #include "xfs_rw.h" |
| 41 | |
| 42 | |
| 43 | STATIC xfs_buf_t *xfs_trans_buf_item_match(xfs_trans_t *, xfs_buftarg_t *, |
| 44 | xfs_daddr_t, int); |
| 45 | STATIC xfs_buf_t *xfs_trans_buf_item_match_all(xfs_trans_t *, xfs_buftarg_t *, |
| 46 | xfs_daddr_t, int); |
| 47 | |
| 48 | |
| 49 | /* |
| 50 | * Get and lock the buffer for the caller if it is not already |
| 51 | * locked within the given transaction. If it is already locked |
| 52 | * within the transaction, just increment its lock recursion count |
| 53 | * and return a pointer to it. |
| 54 | * |
| 55 | * Use the fast path function xfs_trans_buf_item_match() or the buffer |
| 56 | * cache routine incore_match() to find the buffer |
| 57 | * if it is already owned by this transaction. |
| 58 | * |
| 59 | * If we don't already own the buffer, use get_buf() to get it. |
| 60 | * If it doesn't yet have an associated xfs_buf_log_item structure, |
| 61 | * then allocate one and add the item to this transaction. |
| 62 | * |
| 63 | * If the transaction pointer is NULL, make this just a normal |
| 64 | * get_buf() call. |
| 65 | */ |
| 66 | xfs_buf_t * |
| 67 | xfs_trans_get_buf(xfs_trans_t *tp, |
| 68 | xfs_buftarg_t *target_dev, |
| 69 | xfs_daddr_t blkno, |
| 70 | int len, |
| 71 | uint flags) |
| 72 | { |
| 73 | xfs_buf_t *bp; |
| 74 | xfs_buf_log_item_t *bip; |
| 75 | |
| 76 | if (flags == 0) |
| 77 | flags = XFS_BUF_LOCK | XFS_BUF_MAPPED; |
| 78 | |
| 79 | /* |
| 80 | * Default to a normal get_buf() call if the tp is NULL. |
| 81 | */ |
| 82 | if (tp == NULL) { |
| 83 | bp = xfs_buf_get_flags(target_dev, blkno, len, |
| 84 | flags | BUF_BUSY); |
| 85 | return(bp); |
| 86 | } |
| 87 | |
| 88 | /* |
| 89 | * If we find the buffer in the cache with this transaction |
| 90 | * pointer in its b_fsprivate2 field, then we know we already |
| 91 | * have it locked. In this case we just increment the lock |
| 92 | * recursion count and return the buffer to the caller. |
| 93 | */ |
| 94 | if (tp->t_items.lic_next == NULL) { |
| 95 | bp = xfs_trans_buf_item_match(tp, target_dev, blkno, len); |
| 96 | } else { |
| 97 | bp = xfs_trans_buf_item_match_all(tp, target_dev, blkno, len); |
| 98 | } |
| 99 | if (bp != NULL) { |
| 100 | ASSERT(XFS_BUF_VALUSEMA(bp) <= 0); |
| 101 | if (XFS_FORCED_SHUTDOWN(tp->t_mountp)) { |
| 102 | xfs_buftrace("TRANS GET RECUR SHUT", bp); |
| 103 | XFS_BUF_SUPER_STALE(bp); |
| 104 | } |
| 105 | /* |
| 106 | * If the buffer is stale then it was binval'ed |
| 107 | * since last read. This doesn't matter since the |
| 108 | * caller isn't allowed to use the data anyway. |
| 109 | */ |
| 110 | else if (XFS_BUF_ISSTALE(bp)) { |
| 111 | xfs_buftrace("TRANS GET RECUR STALE", bp); |
| 112 | ASSERT(!XFS_BUF_ISDELAYWRITE(bp)); |
| 113 | } |
| 114 | ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp); |
| 115 | bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *); |
| 116 | ASSERT(bip != NULL); |
| 117 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
| 118 | bip->bli_recur++; |
| 119 | xfs_buftrace("TRANS GET RECUR", bp); |
| 120 | xfs_buf_item_trace("GET RECUR", bip); |
| 121 | return (bp); |
| 122 | } |
| 123 | |
| 124 | /* |
| 125 | * We always specify the BUF_BUSY flag within a transaction so |
| 126 | * that get_buf does not try to push out a delayed write buffer |
| 127 | * which might cause another transaction to take place (if the |
| 128 | * buffer was delayed alloc). Such recursive transactions can |
| 129 | * easily deadlock with our current transaction as well as cause |
| 130 | * us to run out of stack space. |
| 131 | */ |
| 132 | bp = xfs_buf_get_flags(target_dev, blkno, len, flags | BUF_BUSY); |
| 133 | if (bp == NULL) { |
| 134 | return NULL; |
| 135 | } |
| 136 | |
| 137 | ASSERT(!XFS_BUF_GETERROR(bp)); |
| 138 | |
| 139 | /* |
| 140 | * The xfs_buf_log_item pointer is stored in b_fsprivate. If |
| 141 | * it doesn't have one yet, then allocate one and initialize it. |
| 142 | * The checks to see if one is there are in xfs_buf_item_init(). |
| 143 | */ |
| 144 | xfs_buf_item_init(bp, tp->t_mountp); |
| 145 | |
| 146 | /* |
| 147 | * Set the recursion count for the buffer within this transaction |
| 148 | * to 0. |
| 149 | */ |
| 150 | bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t*); |
| 151 | ASSERT(!(bip->bli_flags & XFS_BLI_STALE)); |
| 152 | ASSERT(!(bip->bli_format.blf_flags & XFS_BLI_CANCEL)); |
| 153 | ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED)); |
| 154 | bip->bli_recur = 0; |
| 155 | |
| 156 | /* |
| 157 | * Take a reference for this transaction on the buf item. |
| 158 | */ |
| 159 | atomic_inc(&bip->bli_refcount); |
| 160 | |
| 161 | /* |
| 162 | * Get a log_item_desc to point at the new item. |
| 163 | */ |
| 164 | (void) xfs_trans_add_item(tp, (xfs_log_item_t*)bip); |
| 165 | |
| 166 | /* |
| 167 | * Initialize b_fsprivate2 so we can find it with incore_match() |
| 168 | * above. |
| 169 | */ |
| 170 | XFS_BUF_SET_FSPRIVATE2(bp, tp); |
| 171 | |
| 172 | xfs_buftrace("TRANS GET", bp); |
| 173 | xfs_buf_item_trace("GET", bip); |
| 174 | return (bp); |
| 175 | } |
| 176 | |
| 177 | /* |
| 178 | * Get and lock the superblock buffer of this file system for the |
| 179 | * given transaction. |
| 180 | * |
| 181 | * We don't need to use incore_match() here, because the superblock |
| 182 | * buffer is a private buffer which we keep a pointer to in the |
| 183 | * mount structure. |
| 184 | */ |
| 185 | xfs_buf_t * |
| 186 | xfs_trans_getsb(xfs_trans_t *tp, |
| 187 | struct xfs_mount *mp, |
| 188 | int flags) |
| 189 | { |
| 190 | xfs_buf_t *bp; |
| 191 | xfs_buf_log_item_t *bip; |
| 192 | |
| 193 | /* |
| 194 | * Default to just trying to lock the superblock buffer |
| 195 | * if tp is NULL. |
| 196 | */ |
| 197 | if (tp == NULL) { |
| 198 | return (xfs_getsb(mp, flags)); |
| 199 | } |
| 200 | |
| 201 | /* |
| 202 | * If the superblock buffer already has this transaction |
| 203 | * pointer in its b_fsprivate2 field, then we know we already |
| 204 | * have it locked. In this case we just increment the lock |
| 205 | * recursion count and return the buffer to the caller. |
| 206 | */ |
| 207 | bp = mp->m_sb_bp; |
| 208 | if (XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp) { |
| 209 | bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t*); |
| 210 | ASSERT(bip != NULL); |
| 211 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
| 212 | bip->bli_recur++; |
| 213 | xfs_buf_item_trace("GETSB RECUR", bip); |
| 214 | return (bp); |
| 215 | } |
| 216 | |
| 217 | bp = xfs_getsb(mp, flags); |
| 218 | if (bp == NULL) { |
| 219 | return NULL; |
| 220 | } |
| 221 | |
| 222 | /* |
| 223 | * The xfs_buf_log_item pointer is stored in b_fsprivate. If |
| 224 | * it doesn't have one yet, then allocate one and initialize it. |
| 225 | * The checks to see if one is there are in xfs_buf_item_init(). |
| 226 | */ |
| 227 | xfs_buf_item_init(bp, mp); |
| 228 | |
| 229 | /* |
| 230 | * Set the recursion count for the buffer within this transaction |
| 231 | * to 0. |
| 232 | */ |
| 233 | bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t*); |
| 234 | ASSERT(!(bip->bli_flags & XFS_BLI_STALE)); |
| 235 | ASSERT(!(bip->bli_format.blf_flags & XFS_BLI_CANCEL)); |
| 236 | ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED)); |
| 237 | bip->bli_recur = 0; |
| 238 | |
| 239 | /* |
| 240 | * Take a reference for this transaction on the buf item. |
| 241 | */ |
| 242 | atomic_inc(&bip->bli_refcount); |
| 243 | |
| 244 | /* |
| 245 | * Get a log_item_desc to point at the new item. |
| 246 | */ |
| 247 | (void) xfs_trans_add_item(tp, (xfs_log_item_t*)bip); |
| 248 | |
| 249 | /* |
| 250 | * Initialize b_fsprivate2 so we can find it with incore_match() |
| 251 | * above. |
| 252 | */ |
| 253 | XFS_BUF_SET_FSPRIVATE2(bp, tp); |
| 254 | |
| 255 | xfs_buf_item_trace("GETSB", bip); |
| 256 | return (bp); |
| 257 | } |
| 258 | |
| 259 | #ifdef DEBUG |
| 260 | xfs_buftarg_t *xfs_error_target; |
| 261 | int xfs_do_error; |
| 262 | int xfs_req_num; |
| 263 | int xfs_error_mod = 33; |
| 264 | #endif |
| 265 | |
| 266 | /* |
| 267 | * Get and lock the buffer for the caller if it is not already |
| 268 | * locked within the given transaction. If it has not yet been |
| 269 | * read in, read it from disk. If it is already locked |
| 270 | * within the transaction and already read in, just increment its |
| 271 | * lock recursion count and return a pointer to it. |
| 272 | * |
| 273 | * Use the fast path function xfs_trans_buf_item_match() or the buffer |
| 274 | * cache routine incore_match() to find the buffer |
| 275 | * if it is already owned by this transaction. |
| 276 | * |
| 277 | * If we don't already own the buffer, use read_buf() to get it. |
| 278 | * If it doesn't yet have an associated xfs_buf_log_item structure, |
| 279 | * then allocate one and add the item to this transaction. |
| 280 | * |
| 281 | * If the transaction pointer is NULL, make this just a normal |
| 282 | * read_buf() call. |
| 283 | */ |
| 284 | int |
| 285 | xfs_trans_read_buf( |
| 286 | xfs_mount_t *mp, |
| 287 | xfs_trans_t *tp, |
| 288 | xfs_buftarg_t *target, |
| 289 | xfs_daddr_t blkno, |
| 290 | int len, |
| 291 | uint flags, |
| 292 | xfs_buf_t **bpp) |
| 293 | { |
| 294 | xfs_buf_t *bp; |
| 295 | xfs_buf_log_item_t *bip; |
| 296 | int error; |
| 297 | |
| 298 | if (flags == 0) |
| 299 | flags = XFS_BUF_LOCK | XFS_BUF_MAPPED; |
| 300 | |
| 301 | /* |
| 302 | * Default to a normal get_buf() call if the tp is NULL. |
| 303 | */ |
| 304 | if (tp == NULL) { |
| 305 | bp = xfs_buf_read_flags(target, blkno, len, flags | BUF_BUSY); |
| 306 | if (!bp) |
David Chinner | a3f74ff | 2008-03-06 13:43:42 +1100 | [diff] [blame] | 307 | return (flags & XFS_BUF_TRYLOCK) ? |
| 308 | EAGAIN : XFS_ERROR(ENOMEM); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 309 | |
Julia Lawall | a0f7bfd | 2009-07-27 18:15:25 +0200 | [diff] [blame] | 310 | if (XFS_BUF_GETERROR(bp) != 0) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 311 | xfs_ioerror_alert("xfs_trans_read_buf", mp, |
| 312 | bp, blkno); |
| 313 | error = XFS_BUF_GETERROR(bp); |
| 314 | xfs_buf_relse(bp); |
| 315 | return error; |
| 316 | } |
| 317 | #ifdef DEBUG |
Julia Lawall | a0f7bfd | 2009-07-27 18:15:25 +0200 | [diff] [blame] | 318 | if (xfs_do_error) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 319 | if (xfs_error_target == target) { |
| 320 | if (((xfs_req_num++) % xfs_error_mod) == 0) { |
| 321 | xfs_buf_relse(bp); |
Nathan Scott | b657452 | 2006-06-09 15:29:40 +1000 | [diff] [blame] | 322 | cmn_err(CE_DEBUG, "Returning error!\n"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 323 | return XFS_ERROR(EIO); |
| 324 | } |
| 325 | } |
| 326 | } |
| 327 | #endif |
| 328 | if (XFS_FORCED_SHUTDOWN(mp)) |
| 329 | goto shutdown_abort; |
| 330 | *bpp = bp; |
| 331 | return 0; |
| 332 | } |
| 333 | |
| 334 | /* |
| 335 | * If we find the buffer in the cache with this transaction |
| 336 | * pointer in its b_fsprivate2 field, then we know we already |
| 337 | * have it locked. If it is already read in we just increment |
| 338 | * the lock recursion count and return the buffer to the caller. |
| 339 | * If the buffer is not yet read in, then we read it in, increment |
| 340 | * the lock recursion count, and return it to the caller. |
| 341 | */ |
| 342 | if (tp->t_items.lic_next == NULL) { |
| 343 | bp = xfs_trans_buf_item_match(tp, target, blkno, len); |
| 344 | } else { |
| 345 | bp = xfs_trans_buf_item_match_all(tp, target, blkno, len); |
| 346 | } |
| 347 | if (bp != NULL) { |
| 348 | ASSERT(XFS_BUF_VALUSEMA(bp) <= 0); |
| 349 | ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp); |
| 350 | ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL); |
| 351 | ASSERT((XFS_BUF_ISERROR(bp)) == 0); |
| 352 | if (!(XFS_BUF_ISDONE(bp))) { |
| 353 | xfs_buftrace("READ_BUF_INCORE !DONE", bp); |
| 354 | ASSERT(!XFS_BUF_ISASYNC(bp)); |
| 355 | XFS_BUF_READ(bp); |
| 356 | xfsbdstrat(tp->t_mountp, bp); |
David Chinner | d64e31a | 2008-04-10 12:22:17 +1000 | [diff] [blame] | 357 | error = xfs_iowait(bp); |
| 358 | if (error) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 359 | xfs_ioerror_alert("xfs_trans_read_buf", mp, |
| 360 | bp, blkno); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 361 | xfs_buf_relse(bp); |
| 362 | /* |
David Chinner | d64e31a | 2008-04-10 12:22:17 +1000 | [diff] [blame] | 363 | * We can gracefully recover from most read |
| 364 | * errors. Ones we can't are those that happen |
| 365 | * after the transaction's already dirty. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 366 | */ |
| 367 | if (tp->t_flags & XFS_TRANS_DIRTY) |
| 368 | xfs_force_shutdown(tp->t_mountp, |
Nathan Scott | 7d04a33 | 2006-06-09 14:58:38 +1000 | [diff] [blame] | 369 | SHUTDOWN_META_IO_ERROR); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 370 | return error; |
| 371 | } |
| 372 | } |
| 373 | /* |
| 374 | * We never locked this buf ourselves, so we shouldn't |
| 375 | * brelse it either. Just get out. |
| 376 | */ |
| 377 | if (XFS_FORCED_SHUTDOWN(mp)) { |
| 378 | xfs_buftrace("READ_BUF_INCORE XFSSHUTDN", bp); |
| 379 | *bpp = NULL; |
| 380 | return XFS_ERROR(EIO); |
| 381 | } |
| 382 | |
| 383 | |
| 384 | bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t*); |
| 385 | bip->bli_recur++; |
| 386 | |
| 387 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
| 388 | xfs_buf_item_trace("READ RECUR", bip); |
| 389 | *bpp = bp; |
| 390 | return 0; |
| 391 | } |
| 392 | |
| 393 | /* |
| 394 | * We always specify the BUF_BUSY flag within a transaction so |
| 395 | * that get_buf does not try to push out a delayed write buffer |
| 396 | * which might cause another transaction to take place (if the |
| 397 | * buffer was delayed alloc). Such recursive transactions can |
| 398 | * easily deadlock with our current transaction as well as cause |
| 399 | * us to run out of stack space. |
| 400 | */ |
| 401 | bp = xfs_buf_read_flags(target, blkno, len, flags | BUF_BUSY); |
| 402 | if (bp == NULL) { |
| 403 | *bpp = NULL; |
| 404 | return 0; |
| 405 | } |
| 406 | if (XFS_BUF_GETERROR(bp) != 0) { |
| 407 | XFS_BUF_SUPER_STALE(bp); |
| 408 | xfs_buftrace("READ ERROR", bp); |
| 409 | error = XFS_BUF_GETERROR(bp); |
| 410 | |
| 411 | xfs_ioerror_alert("xfs_trans_read_buf", mp, |
| 412 | bp, blkno); |
| 413 | if (tp->t_flags & XFS_TRANS_DIRTY) |
Nathan Scott | 7d04a33 | 2006-06-09 14:58:38 +1000 | [diff] [blame] | 414 | xfs_force_shutdown(tp->t_mountp, SHUTDOWN_META_IO_ERROR); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 415 | xfs_buf_relse(bp); |
| 416 | return error; |
| 417 | } |
| 418 | #ifdef DEBUG |
| 419 | if (xfs_do_error && !(tp->t_flags & XFS_TRANS_DIRTY)) { |
| 420 | if (xfs_error_target == target) { |
| 421 | if (((xfs_req_num++) % xfs_error_mod) == 0) { |
| 422 | xfs_force_shutdown(tp->t_mountp, |
Nathan Scott | 7d04a33 | 2006-06-09 14:58:38 +1000 | [diff] [blame] | 423 | SHUTDOWN_META_IO_ERROR); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 424 | xfs_buf_relse(bp); |
Nathan Scott | b657452 | 2006-06-09 15:29:40 +1000 | [diff] [blame] | 425 | cmn_err(CE_DEBUG, "Returning trans error!\n"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 426 | return XFS_ERROR(EIO); |
| 427 | } |
| 428 | } |
| 429 | } |
| 430 | #endif |
| 431 | if (XFS_FORCED_SHUTDOWN(mp)) |
| 432 | goto shutdown_abort; |
| 433 | |
| 434 | /* |
| 435 | * The xfs_buf_log_item pointer is stored in b_fsprivate. If |
| 436 | * it doesn't have one yet, then allocate one and initialize it. |
| 437 | * The checks to see if one is there are in xfs_buf_item_init(). |
| 438 | */ |
| 439 | xfs_buf_item_init(bp, tp->t_mountp); |
| 440 | |
| 441 | /* |
| 442 | * Set the recursion count for the buffer within this transaction |
| 443 | * to 0. |
| 444 | */ |
| 445 | bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t*); |
| 446 | ASSERT(!(bip->bli_flags & XFS_BLI_STALE)); |
| 447 | ASSERT(!(bip->bli_format.blf_flags & XFS_BLI_CANCEL)); |
| 448 | ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED)); |
| 449 | bip->bli_recur = 0; |
| 450 | |
| 451 | /* |
| 452 | * Take a reference for this transaction on the buf item. |
| 453 | */ |
| 454 | atomic_inc(&bip->bli_refcount); |
| 455 | |
| 456 | /* |
| 457 | * Get a log_item_desc to point at the new item. |
| 458 | */ |
| 459 | (void) xfs_trans_add_item(tp, (xfs_log_item_t*)bip); |
| 460 | |
| 461 | /* |
| 462 | * Initialize b_fsprivate2 so we can find it with incore_match() |
| 463 | * above. |
| 464 | */ |
| 465 | XFS_BUF_SET_FSPRIVATE2(bp, tp); |
| 466 | |
| 467 | xfs_buftrace("TRANS READ", bp); |
| 468 | xfs_buf_item_trace("READ", bip); |
| 469 | *bpp = bp; |
| 470 | return 0; |
| 471 | |
| 472 | shutdown_abort: |
| 473 | /* |
| 474 | * the theory here is that buffer is good but we're |
| 475 | * bailing out because the filesystem is being forcibly |
| 476 | * shut down. So we should leave the b_flags alone since |
| 477 | * the buffer's not staled and just get out. |
| 478 | */ |
| 479 | #if defined(DEBUG) |
| 480 | if (XFS_BUF_ISSTALE(bp) && XFS_BUF_ISDELAYWRITE(bp)) |
| 481 | cmn_err(CE_NOTE, "about to pop assert, bp == 0x%p", bp); |
| 482 | #endif |
| 483 | ASSERT((XFS_BUF_BFLAGS(bp) & (XFS_B_STALE|XFS_B_DELWRI)) != |
| 484 | (XFS_B_STALE|XFS_B_DELWRI)); |
| 485 | |
| 486 | xfs_buftrace("READ_BUF XFSSHUTDN", bp); |
| 487 | xfs_buf_relse(bp); |
| 488 | *bpp = NULL; |
| 489 | return XFS_ERROR(EIO); |
| 490 | } |
| 491 | |
| 492 | |
| 493 | /* |
| 494 | * Release the buffer bp which was previously acquired with one of the |
| 495 | * xfs_trans_... buffer allocation routines if the buffer has not |
| 496 | * been modified within this transaction. If the buffer is modified |
| 497 | * within this transaction, do decrement the recursion count but do |
| 498 | * not release the buffer even if the count goes to 0. If the buffer is not |
| 499 | * modified within the transaction, decrement the recursion count and |
| 500 | * release the buffer if the recursion count goes to 0. |
| 501 | * |
| 502 | * If the buffer is to be released and it was not modified before |
| 503 | * this transaction began, then free the buf_log_item associated with it. |
| 504 | * |
| 505 | * If the transaction pointer is NULL, make this just a normal |
| 506 | * brelse() call. |
| 507 | */ |
| 508 | void |
| 509 | xfs_trans_brelse(xfs_trans_t *tp, |
| 510 | xfs_buf_t *bp) |
| 511 | { |
| 512 | xfs_buf_log_item_t *bip; |
| 513 | xfs_log_item_t *lip; |
| 514 | xfs_log_item_desc_t *lidp; |
| 515 | |
| 516 | /* |
| 517 | * Default to a normal brelse() call if the tp is NULL. |
| 518 | */ |
| 519 | if (tp == NULL) { |
| 520 | ASSERT(XFS_BUF_FSPRIVATE2(bp, void *) == NULL); |
| 521 | /* |
| 522 | * If there's a buf log item attached to the buffer, |
| 523 | * then let the AIL know that the buffer is being |
| 524 | * unlocked. |
| 525 | */ |
| 526 | if (XFS_BUF_FSPRIVATE(bp, void *) != NULL) { |
| 527 | lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *); |
| 528 | if (lip->li_type == XFS_LI_BUF) { |
| 529 | bip = XFS_BUF_FSPRIVATE(bp,xfs_buf_log_item_t*); |
David Chinner | 783a2f6 | 2008-10-30 17:39:58 +1100 | [diff] [blame] | 530 | xfs_trans_unlocked_item(bip->bli_item.li_ailp, |
| 531 | lip); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 532 | } |
| 533 | } |
| 534 | xfs_buf_relse(bp); |
| 535 | return; |
| 536 | } |
| 537 | |
| 538 | ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp); |
| 539 | bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *); |
| 540 | ASSERT(bip->bli_item.li_type == XFS_LI_BUF); |
| 541 | ASSERT(!(bip->bli_flags & XFS_BLI_STALE)); |
| 542 | ASSERT(!(bip->bli_format.blf_flags & XFS_BLI_CANCEL)); |
| 543 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
| 544 | |
| 545 | /* |
| 546 | * Find the item descriptor pointing to this buffer's |
| 547 | * log item. It must be there. |
| 548 | */ |
| 549 | lidp = xfs_trans_find_item(tp, (xfs_log_item_t*)bip); |
| 550 | ASSERT(lidp != NULL); |
| 551 | |
| 552 | /* |
| 553 | * If the release is just for a recursive lock, |
| 554 | * then decrement the count and return. |
| 555 | */ |
| 556 | if (bip->bli_recur > 0) { |
| 557 | bip->bli_recur--; |
| 558 | xfs_buf_item_trace("RELSE RECUR", bip); |
| 559 | return; |
| 560 | } |
| 561 | |
| 562 | /* |
| 563 | * If the buffer is dirty within this transaction, we can't |
| 564 | * release it until we commit. |
| 565 | */ |
| 566 | if (lidp->lid_flags & XFS_LID_DIRTY) { |
| 567 | xfs_buf_item_trace("RELSE DIRTY", bip); |
| 568 | return; |
| 569 | } |
| 570 | |
| 571 | /* |
| 572 | * If the buffer has been invalidated, then we can't release |
| 573 | * it until the transaction commits to disk unless it is re-dirtied |
| 574 | * as part of this transaction. This prevents us from pulling |
| 575 | * the item from the AIL before we should. |
| 576 | */ |
| 577 | if (bip->bli_flags & XFS_BLI_STALE) { |
| 578 | xfs_buf_item_trace("RELSE STALE", bip); |
| 579 | return; |
| 580 | } |
| 581 | |
| 582 | ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED)); |
| 583 | xfs_buf_item_trace("RELSE", bip); |
| 584 | |
| 585 | /* |
| 586 | * Free up the log item descriptor tracking the released item. |
| 587 | */ |
| 588 | xfs_trans_free_item(tp, lidp); |
| 589 | |
| 590 | /* |
| 591 | * Clear the hold flag in the buf log item if it is set. |
| 592 | * We wouldn't want the next user of the buffer to |
| 593 | * get confused. |
| 594 | */ |
| 595 | if (bip->bli_flags & XFS_BLI_HOLD) { |
| 596 | bip->bli_flags &= ~XFS_BLI_HOLD; |
| 597 | } |
| 598 | |
| 599 | /* |
| 600 | * Drop our reference to the buf log item. |
| 601 | */ |
| 602 | atomic_dec(&bip->bli_refcount); |
| 603 | |
| 604 | /* |
| 605 | * If the buf item is not tracking data in the log, then |
| 606 | * we must free it before releasing the buffer back to the |
| 607 | * free pool. Before releasing the buffer to the free pool, |
| 608 | * clear the transaction pointer in b_fsprivate2 to dissolve |
| 609 | * its relation to this transaction. |
| 610 | */ |
| 611 | if (!xfs_buf_item_dirty(bip)) { |
| 612 | /*** |
| 613 | ASSERT(bp->b_pincount == 0); |
| 614 | ***/ |
| 615 | ASSERT(atomic_read(&bip->bli_refcount) == 0); |
| 616 | ASSERT(!(bip->bli_item.li_flags & XFS_LI_IN_AIL)); |
| 617 | ASSERT(!(bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF)); |
| 618 | xfs_buf_item_relse(bp); |
| 619 | bip = NULL; |
| 620 | } |
| 621 | XFS_BUF_SET_FSPRIVATE2(bp, NULL); |
| 622 | |
| 623 | /* |
| 624 | * If we've still got a buf log item on the buffer, then |
| 625 | * tell the AIL that the buffer is being unlocked. |
| 626 | */ |
| 627 | if (bip != NULL) { |
David Chinner | 783a2f6 | 2008-10-30 17:39:58 +1100 | [diff] [blame] | 628 | xfs_trans_unlocked_item(bip->bli_item.li_ailp, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 629 | (xfs_log_item_t*)bip); |
| 630 | } |
| 631 | |
| 632 | xfs_buf_relse(bp); |
| 633 | return; |
| 634 | } |
| 635 | |
| 636 | /* |
| 637 | * Add the locked buffer to the transaction. |
| 638 | * The buffer must be locked, and it cannot be associated with any |
| 639 | * transaction. |
| 640 | * |
| 641 | * If the buffer does not yet have a buf log item associated with it, |
| 642 | * then allocate one for it. Then add the buf item to the transaction. |
| 643 | */ |
| 644 | void |
| 645 | xfs_trans_bjoin(xfs_trans_t *tp, |
| 646 | xfs_buf_t *bp) |
| 647 | { |
| 648 | xfs_buf_log_item_t *bip; |
| 649 | |
| 650 | ASSERT(XFS_BUF_ISBUSY(bp)); |
| 651 | ASSERT(XFS_BUF_FSPRIVATE2(bp, void *) == NULL); |
| 652 | |
| 653 | /* |
| 654 | * The xfs_buf_log_item pointer is stored in b_fsprivate. If |
| 655 | * it doesn't have one yet, then allocate one and initialize it. |
| 656 | * The checks to see if one is there are in xfs_buf_item_init(). |
| 657 | */ |
| 658 | xfs_buf_item_init(bp, tp->t_mountp); |
| 659 | bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *); |
| 660 | ASSERT(!(bip->bli_flags & XFS_BLI_STALE)); |
| 661 | ASSERT(!(bip->bli_format.blf_flags & XFS_BLI_CANCEL)); |
| 662 | ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED)); |
| 663 | |
| 664 | /* |
| 665 | * Take a reference for this transaction on the buf item. |
| 666 | */ |
| 667 | atomic_inc(&bip->bli_refcount); |
| 668 | |
| 669 | /* |
| 670 | * Get a log_item_desc to point at the new item. |
| 671 | */ |
| 672 | (void) xfs_trans_add_item(tp, (xfs_log_item_t *)bip); |
| 673 | |
| 674 | /* |
| 675 | * Initialize b_fsprivate2 so we can find it with incore_match() |
| 676 | * in xfs_trans_get_buf() and friends above. |
| 677 | */ |
| 678 | XFS_BUF_SET_FSPRIVATE2(bp, tp); |
| 679 | |
| 680 | xfs_buf_item_trace("BJOIN", bip); |
| 681 | } |
| 682 | |
| 683 | /* |
| 684 | * Mark the buffer as not needing to be unlocked when the buf item's |
| 685 | * IOP_UNLOCK() routine is called. The buffer must already be locked |
| 686 | * and associated with the given transaction. |
| 687 | */ |
| 688 | /* ARGSUSED */ |
| 689 | void |
| 690 | xfs_trans_bhold(xfs_trans_t *tp, |
| 691 | xfs_buf_t *bp) |
| 692 | { |
| 693 | xfs_buf_log_item_t *bip; |
| 694 | |
| 695 | ASSERT(XFS_BUF_ISBUSY(bp)); |
| 696 | ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp); |
| 697 | ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL); |
| 698 | |
| 699 | bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *); |
| 700 | ASSERT(!(bip->bli_flags & XFS_BLI_STALE)); |
| 701 | ASSERT(!(bip->bli_format.blf_flags & XFS_BLI_CANCEL)); |
| 702 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
| 703 | bip->bli_flags |= XFS_BLI_HOLD; |
| 704 | xfs_buf_item_trace("BHOLD", bip); |
| 705 | } |
| 706 | |
| 707 | /* |
Tim Shimmin | efa092f | 2005-09-05 08:29:01 +1000 | [diff] [blame] | 708 | * Cancel the previous buffer hold request made on this buffer |
| 709 | * for this transaction. |
| 710 | */ |
| 711 | void |
| 712 | xfs_trans_bhold_release(xfs_trans_t *tp, |
| 713 | xfs_buf_t *bp) |
| 714 | { |
| 715 | xfs_buf_log_item_t *bip; |
| 716 | |
| 717 | ASSERT(XFS_BUF_ISBUSY(bp)); |
| 718 | ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp); |
| 719 | ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL); |
| 720 | |
| 721 | bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *); |
| 722 | ASSERT(!(bip->bli_flags & XFS_BLI_STALE)); |
| 723 | ASSERT(!(bip->bli_format.blf_flags & XFS_BLI_CANCEL)); |
| 724 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
| 725 | ASSERT(bip->bli_flags & XFS_BLI_HOLD); |
| 726 | bip->bli_flags &= ~XFS_BLI_HOLD; |
| 727 | xfs_buf_item_trace("BHOLD RELEASE", bip); |
| 728 | } |
| 729 | |
| 730 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 731 | * This is called to mark bytes first through last inclusive of the given |
| 732 | * buffer as needing to be logged when the transaction is committed. |
| 733 | * The buffer must already be associated with the given transaction. |
| 734 | * |
| 735 | * First and last are numbers relative to the beginning of this buffer, |
| 736 | * so the first byte in the buffer is numbered 0 regardless of the |
| 737 | * value of b_blkno. |
| 738 | */ |
| 739 | void |
| 740 | xfs_trans_log_buf(xfs_trans_t *tp, |
| 741 | xfs_buf_t *bp, |
| 742 | uint first, |
| 743 | uint last) |
| 744 | { |
| 745 | xfs_buf_log_item_t *bip; |
| 746 | xfs_log_item_desc_t *lidp; |
| 747 | |
| 748 | ASSERT(XFS_BUF_ISBUSY(bp)); |
| 749 | ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp); |
| 750 | ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL); |
| 751 | ASSERT((first <= last) && (last < XFS_BUF_COUNT(bp))); |
| 752 | ASSERT((XFS_BUF_IODONE_FUNC(bp) == NULL) || |
| 753 | (XFS_BUF_IODONE_FUNC(bp) == xfs_buf_iodone_callbacks)); |
| 754 | |
| 755 | /* |
| 756 | * Mark the buffer as needing to be written out eventually, |
| 757 | * and set its iodone function to remove the buffer's buf log |
| 758 | * item from the AIL and free it when the buffer is flushed |
| 759 | * to disk. See xfs_buf_attach_iodone() for more details |
| 760 | * on li_cb and xfs_buf_iodone_callbacks(). |
| 761 | * If we end up aborting this transaction, we trap this buffer |
| 762 | * inside the b_bdstrat callback so that this won't get written to |
| 763 | * disk. |
| 764 | */ |
| 765 | XFS_BUF_DELAYWRITE(bp); |
| 766 | XFS_BUF_DONE(bp); |
| 767 | |
| 768 | bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *); |
| 769 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
| 770 | XFS_BUF_SET_IODONE_FUNC(bp, xfs_buf_iodone_callbacks); |
| 771 | bip->bli_item.li_cb = (void(*)(xfs_buf_t*,xfs_log_item_t*))xfs_buf_iodone; |
| 772 | |
| 773 | /* |
| 774 | * If we invalidated the buffer within this transaction, then |
| 775 | * cancel the invalidation now that we're dirtying the buffer |
| 776 | * again. There are no races with the code in xfs_buf_item_unpin(), |
| 777 | * because we have a reference to the buffer this entire time. |
| 778 | */ |
| 779 | if (bip->bli_flags & XFS_BLI_STALE) { |
| 780 | xfs_buf_item_trace("BLOG UNSTALE", bip); |
| 781 | bip->bli_flags &= ~XFS_BLI_STALE; |
| 782 | ASSERT(XFS_BUF_ISSTALE(bp)); |
| 783 | XFS_BUF_UNSTALE(bp); |
| 784 | bip->bli_format.blf_flags &= ~XFS_BLI_CANCEL; |
| 785 | } |
| 786 | |
| 787 | lidp = xfs_trans_find_item(tp, (xfs_log_item_t*)bip); |
| 788 | ASSERT(lidp != NULL); |
| 789 | |
| 790 | tp->t_flags |= XFS_TRANS_DIRTY; |
| 791 | lidp->lid_flags |= XFS_LID_DIRTY; |
| 792 | lidp->lid_flags &= ~XFS_LID_BUF_STALE; |
| 793 | bip->bli_flags |= XFS_BLI_LOGGED; |
| 794 | xfs_buf_item_log(bip, first, last); |
| 795 | xfs_buf_item_trace("BLOG", bip); |
| 796 | } |
| 797 | |
| 798 | |
| 799 | /* |
| 800 | * This called to invalidate a buffer that is being used within |
| 801 | * a transaction. Typically this is because the blocks in the |
| 802 | * buffer are being freed, so we need to prevent it from being |
| 803 | * written out when we're done. Allowing it to be written again |
| 804 | * might overwrite data in the free blocks if they are reallocated |
| 805 | * to a file. |
| 806 | * |
| 807 | * We prevent the buffer from being written out by clearing the |
| 808 | * B_DELWRI flag. We can't always |
| 809 | * get rid of the buf log item at this point, though, because |
| 810 | * the buffer may still be pinned by another transaction. If that |
| 811 | * is the case, then we'll wait until the buffer is committed to |
| 812 | * disk for the last time (we can tell by the ref count) and |
| 813 | * free it in xfs_buf_item_unpin(). Until it is cleaned up we |
| 814 | * will keep the buffer locked so that the buffer and buf log item |
| 815 | * are not reused. |
| 816 | */ |
| 817 | void |
| 818 | xfs_trans_binval( |
| 819 | xfs_trans_t *tp, |
| 820 | xfs_buf_t *bp) |
| 821 | { |
| 822 | xfs_log_item_desc_t *lidp; |
| 823 | xfs_buf_log_item_t *bip; |
| 824 | |
| 825 | ASSERT(XFS_BUF_ISBUSY(bp)); |
| 826 | ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp); |
| 827 | ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL); |
| 828 | |
| 829 | bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *); |
| 830 | lidp = xfs_trans_find_item(tp, (xfs_log_item_t*)bip); |
| 831 | ASSERT(lidp != NULL); |
| 832 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
| 833 | |
| 834 | if (bip->bli_flags & XFS_BLI_STALE) { |
| 835 | /* |
| 836 | * If the buffer is already invalidated, then |
| 837 | * just return. |
| 838 | */ |
| 839 | ASSERT(!(XFS_BUF_ISDELAYWRITE(bp))); |
| 840 | ASSERT(XFS_BUF_ISSTALE(bp)); |
| 841 | ASSERT(!(bip->bli_flags & (XFS_BLI_LOGGED | XFS_BLI_DIRTY))); |
| 842 | ASSERT(!(bip->bli_format.blf_flags & XFS_BLI_INODE_BUF)); |
| 843 | ASSERT(bip->bli_format.blf_flags & XFS_BLI_CANCEL); |
| 844 | ASSERT(lidp->lid_flags & XFS_LID_DIRTY); |
| 845 | ASSERT(tp->t_flags & XFS_TRANS_DIRTY); |
| 846 | xfs_buftrace("XFS_BINVAL RECUR", bp); |
| 847 | xfs_buf_item_trace("BINVAL RECUR", bip); |
| 848 | return; |
| 849 | } |
| 850 | |
| 851 | /* |
| 852 | * Clear the dirty bit in the buffer and set the STALE flag |
| 853 | * in the buf log item. The STALE flag will be used in |
| 854 | * xfs_buf_item_unpin() to determine if it should clean up |
| 855 | * when the last reference to the buf item is given up. |
| 856 | * We set the XFS_BLI_CANCEL flag in the buf log format structure |
| 857 | * and log the buf item. This will be used at recovery time |
| 858 | * to determine that copies of the buffer in the log before |
| 859 | * this should not be replayed. |
| 860 | * We mark the item descriptor and the transaction dirty so |
| 861 | * that we'll hold the buffer until after the commit. |
| 862 | * |
| 863 | * Since we're invalidating the buffer, we also clear the state |
| 864 | * about which parts of the buffer have been logged. We also |
| 865 | * clear the flag indicating that this is an inode buffer since |
| 866 | * the data in the buffer will no longer be valid. |
| 867 | * |
| 868 | * We set the stale bit in the buffer as well since we're getting |
| 869 | * rid of it. |
| 870 | */ |
| 871 | XFS_BUF_UNDELAYWRITE(bp); |
| 872 | XFS_BUF_STALE(bp); |
| 873 | bip->bli_flags |= XFS_BLI_STALE; |
| 874 | bip->bli_flags &= ~(XFS_BLI_LOGGED | XFS_BLI_DIRTY); |
| 875 | bip->bli_format.blf_flags &= ~XFS_BLI_INODE_BUF; |
| 876 | bip->bli_format.blf_flags |= XFS_BLI_CANCEL; |
| 877 | memset((char *)(bip->bli_format.blf_data_map), 0, |
| 878 | (bip->bli_format.blf_map_size * sizeof(uint))); |
| 879 | lidp->lid_flags |= XFS_LID_DIRTY|XFS_LID_BUF_STALE; |
| 880 | tp->t_flags |= XFS_TRANS_DIRTY; |
| 881 | xfs_buftrace("XFS_BINVAL", bp); |
| 882 | xfs_buf_item_trace("BINVAL", bip); |
| 883 | } |
| 884 | |
| 885 | /* |
| 886 | * This call is used to indicate that the buffer contains on-disk |
| 887 | * inodes which must be handled specially during recovery. They |
| 888 | * require special handling because only the di_next_unlinked from |
| 889 | * the inodes in the buffer should be recovered. The rest of the |
| 890 | * data in the buffer is logged via the inodes themselves. |
| 891 | * |
| 892 | * All we do is set the XFS_BLI_INODE_BUF flag in the buffer's log |
| 893 | * format structure so that we'll know what to do at recovery time. |
| 894 | */ |
| 895 | /* ARGSUSED */ |
| 896 | void |
| 897 | xfs_trans_inode_buf( |
| 898 | xfs_trans_t *tp, |
| 899 | xfs_buf_t *bp) |
| 900 | { |
| 901 | xfs_buf_log_item_t *bip; |
| 902 | |
| 903 | ASSERT(XFS_BUF_ISBUSY(bp)); |
| 904 | ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp); |
| 905 | ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL); |
| 906 | |
| 907 | bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *); |
| 908 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
| 909 | |
| 910 | bip->bli_format.blf_flags |= XFS_BLI_INODE_BUF; |
| 911 | } |
| 912 | |
| 913 | /* |
| 914 | * This call is used to indicate that the buffer is going to |
| 915 | * be staled and was an inode buffer. This means it gets |
| 916 | * special processing during unpin - where any inodes |
| 917 | * associated with the buffer should be removed from ail. |
| 918 | * There is also special processing during recovery, |
| 919 | * any replay of the inodes in the buffer needs to be |
| 920 | * prevented as the buffer may have been reused. |
| 921 | */ |
| 922 | void |
| 923 | xfs_trans_stale_inode_buf( |
| 924 | xfs_trans_t *tp, |
| 925 | xfs_buf_t *bp) |
| 926 | { |
| 927 | xfs_buf_log_item_t *bip; |
| 928 | |
| 929 | ASSERT(XFS_BUF_ISBUSY(bp)); |
| 930 | ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp); |
| 931 | ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL); |
| 932 | |
| 933 | bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *); |
| 934 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
| 935 | |
| 936 | bip->bli_flags |= XFS_BLI_STALE_INODE; |
| 937 | bip->bli_item.li_cb = (void(*)(xfs_buf_t*,xfs_log_item_t*)) |
| 938 | xfs_buf_iodone; |
| 939 | } |
| 940 | |
| 941 | |
| 942 | |
| 943 | /* |
| 944 | * Mark the buffer as being one which contains newly allocated |
| 945 | * inodes. We need to make sure that even if this buffer is |
| 946 | * relogged as an 'inode buf' we still recover all of the inode |
| 947 | * images in the face of a crash. This works in coordination with |
| 948 | * xfs_buf_item_committed() to ensure that the buffer remains in the |
| 949 | * AIL at its original location even after it has been relogged. |
| 950 | */ |
| 951 | /* ARGSUSED */ |
| 952 | void |
| 953 | xfs_trans_inode_alloc_buf( |
| 954 | xfs_trans_t *tp, |
| 955 | xfs_buf_t *bp) |
| 956 | { |
| 957 | xfs_buf_log_item_t *bip; |
| 958 | |
| 959 | ASSERT(XFS_BUF_ISBUSY(bp)); |
| 960 | ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp); |
| 961 | ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL); |
| 962 | |
| 963 | bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *); |
| 964 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
| 965 | |
| 966 | bip->bli_flags |= XFS_BLI_INODE_ALLOC_BUF; |
| 967 | } |
| 968 | |
| 969 | |
| 970 | /* |
| 971 | * Similar to xfs_trans_inode_buf(), this marks the buffer as a cluster of |
| 972 | * dquots. However, unlike in inode buffer recovery, dquot buffers get |
| 973 | * recovered in their entirety. (Hence, no XFS_BLI_DQUOT_ALLOC_BUF flag). |
| 974 | * The only thing that makes dquot buffers different from regular |
| 975 | * buffers is that we must not replay dquot bufs when recovering |
| 976 | * if a _corresponding_ quotaoff has happened. We also have to distinguish |
| 977 | * between usr dquot bufs and grp dquot bufs, because usr and grp quotas |
| 978 | * can be turned off independently. |
| 979 | */ |
| 980 | /* ARGSUSED */ |
| 981 | void |
| 982 | xfs_trans_dquot_buf( |
| 983 | xfs_trans_t *tp, |
| 984 | xfs_buf_t *bp, |
| 985 | uint type) |
| 986 | { |
| 987 | xfs_buf_log_item_t *bip; |
| 988 | |
| 989 | ASSERT(XFS_BUF_ISBUSY(bp)); |
| 990 | ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp); |
| 991 | ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL); |
| 992 | ASSERT(type == XFS_BLI_UDQUOT_BUF || |
Nathan Scott | c8ad20f | 2005-06-21 15:38:48 +1000 | [diff] [blame] | 993 | type == XFS_BLI_PDQUOT_BUF || |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 994 | type == XFS_BLI_GDQUOT_BUF); |
| 995 | |
| 996 | bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *); |
| 997 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
| 998 | |
| 999 | bip->bli_format.blf_flags |= type; |
| 1000 | } |
| 1001 | |
| 1002 | /* |
| 1003 | * Check to see if a buffer matching the given parameters is already |
| 1004 | * a part of the given transaction. Only check the first, embedded |
| 1005 | * chunk, since we don't want to spend all day scanning large transactions. |
| 1006 | */ |
| 1007 | STATIC xfs_buf_t * |
| 1008 | xfs_trans_buf_item_match( |
| 1009 | xfs_trans_t *tp, |
| 1010 | xfs_buftarg_t *target, |
| 1011 | xfs_daddr_t blkno, |
| 1012 | int len) |
| 1013 | { |
| 1014 | xfs_log_item_chunk_t *licp; |
| 1015 | xfs_log_item_desc_t *lidp; |
| 1016 | xfs_buf_log_item_t *blip; |
| 1017 | xfs_buf_t *bp; |
| 1018 | int i; |
| 1019 | |
| 1020 | bp = NULL; |
| 1021 | len = BBTOB(len); |
| 1022 | licp = &tp->t_items; |
Eric Sandeen | 39dab9d | 2008-08-13 16:10:52 +1000 | [diff] [blame] | 1023 | if (!xfs_lic_are_all_free(licp)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1024 | for (i = 0; i < licp->lic_unused; i++) { |
| 1025 | /* |
| 1026 | * Skip unoccupied slots. |
| 1027 | */ |
Eric Sandeen | 39dab9d | 2008-08-13 16:10:52 +1000 | [diff] [blame] | 1028 | if (xfs_lic_isfree(licp, i)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1029 | continue; |
| 1030 | } |
| 1031 | |
Eric Sandeen | 39dab9d | 2008-08-13 16:10:52 +1000 | [diff] [blame] | 1032 | lidp = xfs_lic_slot(licp, i); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1033 | blip = (xfs_buf_log_item_t *)lidp->lid_item; |
| 1034 | if (blip->bli_item.li_type != XFS_LI_BUF) { |
| 1035 | continue; |
| 1036 | } |
| 1037 | |
| 1038 | bp = blip->bli_buf; |
| 1039 | if ((XFS_BUF_TARGET(bp) == target) && |
| 1040 | (XFS_BUF_ADDR(bp) == blkno) && |
| 1041 | (XFS_BUF_COUNT(bp) == len)) { |
| 1042 | /* |
| 1043 | * We found it. Break out and |
| 1044 | * return the pointer to the buffer. |
| 1045 | */ |
| 1046 | break; |
| 1047 | } else { |
| 1048 | bp = NULL; |
| 1049 | } |
| 1050 | } |
| 1051 | } |
| 1052 | return bp; |
| 1053 | } |
| 1054 | |
| 1055 | /* |
| 1056 | * Check to see if a buffer matching the given parameters is already |
| 1057 | * a part of the given transaction. Check all the chunks, we |
| 1058 | * want to be thorough. |
| 1059 | */ |
| 1060 | STATIC xfs_buf_t * |
| 1061 | xfs_trans_buf_item_match_all( |
| 1062 | xfs_trans_t *tp, |
| 1063 | xfs_buftarg_t *target, |
| 1064 | xfs_daddr_t blkno, |
| 1065 | int len) |
| 1066 | { |
| 1067 | xfs_log_item_chunk_t *licp; |
| 1068 | xfs_log_item_desc_t *lidp; |
| 1069 | xfs_buf_log_item_t *blip; |
| 1070 | xfs_buf_t *bp; |
| 1071 | int i; |
| 1072 | |
| 1073 | bp = NULL; |
| 1074 | len = BBTOB(len); |
| 1075 | for (licp = &tp->t_items; licp != NULL; licp = licp->lic_next) { |
Eric Sandeen | 39dab9d | 2008-08-13 16:10:52 +1000 | [diff] [blame] | 1076 | if (xfs_lic_are_all_free(licp)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1077 | ASSERT(licp == &tp->t_items); |
| 1078 | ASSERT(licp->lic_next == NULL); |
| 1079 | return NULL; |
| 1080 | } |
| 1081 | for (i = 0; i < licp->lic_unused; i++) { |
| 1082 | /* |
| 1083 | * Skip unoccupied slots. |
| 1084 | */ |
Eric Sandeen | 39dab9d | 2008-08-13 16:10:52 +1000 | [diff] [blame] | 1085 | if (xfs_lic_isfree(licp, i)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1086 | continue; |
| 1087 | } |
| 1088 | |
Eric Sandeen | 39dab9d | 2008-08-13 16:10:52 +1000 | [diff] [blame] | 1089 | lidp = xfs_lic_slot(licp, i); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1090 | blip = (xfs_buf_log_item_t *)lidp->lid_item; |
| 1091 | if (blip->bli_item.li_type != XFS_LI_BUF) { |
| 1092 | continue; |
| 1093 | } |
| 1094 | |
| 1095 | bp = blip->bli_buf; |
| 1096 | if ((XFS_BUF_TARGET(bp) == target) && |
| 1097 | (XFS_BUF_ADDR(bp) == blkno) && |
| 1098 | (XFS_BUF_COUNT(bp) == len)) { |
| 1099 | /* |
| 1100 | * We found it. Break out and |
| 1101 | * return the pointer to the buffer. |
| 1102 | */ |
| 1103 | return bp; |
| 1104 | } |
| 1105 | } |
| 1106 | } |
| 1107 | return NULL; |
| 1108 | } |