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) 2000-2005 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" |
David Chinner | da353b0 | 2007-08-28 14:00:13 +1000 | [diff] [blame] | 26 | #include "xfs_ag.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 27 | #include "xfs_dmapi.h" |
| 28 | #include "xfs_mount.h" |
Nathan Scott | a844f45 | 2005-11-02 14:38:42 +1100 | [diff] [blame] | 29 | #include "xfs_buf_item.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 30 | #include "xfs_trans_priv.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 31 | #include "xfs_error.h" |
Christoph Hellwig | 0b1b213 | 2009-12-14 23:14:59 +0000 | [diff] [blame] | 32 | #include "xfs_trace.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 33 | |
| 34 | |
| 35 | kmem_zone_t *xfs_buf_item_zone; |
| 36 | |
| 37 | #ifdef XFS_TRANS_DEBUG |
| 38 | /* |
| 39 | * This function uses an alternate strategy for tracking the bytes |
| 40 | * that the user requests to be logged. This can then be used |
| 41 | * in conjunction with the bli_orig array in the buf log item to |
| 42 | * catch bugs in our callers' code. |
| 43 | * |
| 44 | * We also double check the bits set in xfs_buf_item_log using a |
| 45 | * simple algorithm to check that every byte is accounted for. |
| 46 | */ |
| 47 | STATIC void |
| 48 | xfs_buf_item_log_debug( |
| 49 | xfs_buf_log_item_t *bip, |
| 50 | uint first, |
| 51 | uint last) |
| 52 | { |
| 53 | uint x; |
| 54 | uint byte; |
| 55 | uint nbytes; |
| 56 | uint chunk_num; |
| 57 | uint word_num; |
| 58 | uint bit_num; |
| 59 | uint bit_set; |
| 60 | uint *wordp; |
| 61 | |
| 62 | ASSERT(bip->bli_logged != NULL); |
| 63 | byte = first; |
| 64 | nbytes = last - first + 1; |
| 65 | bfset(bip->bli_logged, first, nbytes); |
| 66 | for (x = 0; x < nbytes; x++) { |
| 67 | chunk_num = byte >> XFS_BLI_SHIFT; |
| 68 | word_num = chunk_num >> BIT_TO_WORD_SHIFT; |
| 69 | bit_num = chunk_num & (NBWORD - 1); |
| 70 | wordp = &(bip->bli_format.blf_data_map[word_num]); |
| 71 | bit_set = *wordp & (1 << bit_num); |
| 72 | ASSERT(bit_set); |
| 73 | byte++; |
| 74 | } |
| 75 | } |
| 76 | |
| 77 | /* |
| 78 | * This function is called when we flush something into a buffer without |
| 79 | * logging it. This happens for things like inodes which are logged |
| 80 | * separately from the buffer. |
| 81 | */ |
| 82 | void |
| 83 | xfs_buf_item_flush_log_debug( |
| 84 | xfs_buf_t *bp, |
| 85 | uint first, |
| 86 | uint last) |
| 87 | { |
| 88 | xfs_buf_log_item_t *bip; |
| 89 | uint nbytes; |
| 90 | |
| 91 | bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t*); |
| 92 | if ((bip == NULL) || (bip->bli_item.li_type != XFS_LI_BUF)) { |
| 93 | return; |
| 94 | } |
| 95 | |
| 96 | ASSERT(bip->bli_logged != NULL); |
| 97 | nbytes = last - first + 1; |
| 98 | bfset(bip->bli_logged, first, nbytes); |
| 99 | } |
| 100 | |
| 101 | /* |
Nathan Scott | c41564b | 2006-03-29 08:55:14 +1000 | [diff] [blame] | 102 | * This function is called to verify that our callers have logged |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 103 | * all the bytes that they changed. |
| 104 | * |
| 105 | * It does this by comparing the original copy of the buffer stored in |
| 106 | * the buf log item's bli_orig array to the current copy of the buffer |
Nathan Scott | c41564b | 2006-03-29 08:55:14 +1000 | [diff] [blame] | 107 | * and ensuring that all bytes which mismatch are set in the bli_logged |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 108 | * array of the buf log item. |
| 109 | */ |
| 110 | STATIC void |
| 111 | xfs_buf_item_log_check( |
| 112 | xfs_buf_log_item_t *bip) |
| 113 | { |
| 114 | char *orig; |
| 115 | char *buffer; |
| 116 | int x; |
| 117 | xfs_buf_t *bp; |
| 118 | |
| 119 | ASSERT(bip->bli_orig != NULL); |
| 120 | ASSERT(bip->bli_logged != NULL); |
| 121 | |
| 122 | bp = bip->bli_buf; |
| 123 | ASSERT(XFS_BUF_COUNT(bp) > 0); |
| 124 | ASSERT(XFS_BUF_PTR(bp) != NULL); |
| 125 | orig = bip->bli_orig; |
| 126 | buffer = XFS_BUF_PTR(bp); |
| 127 | for (x = 0; x < XFS_BUF_COUNT(bp); x++) { |
| 128 | if (orig[x] != buffer[x] && !btst(bip->bli_logged, x)) |
| 129 | cmn_err(CE_PANIC, |
| 130 | "xfs_buf_item_log_check bip %x buffer %x orig %x index %d", |
| 131 | bip, bp, orig, x); |
| 132 | } |
| 133 | } |
| 134 | #else |
| 135 | #define xfs_buf_item_log_debug(x,y,z) |
| 136 | #define xfs_buf_item_log_check(x) |
| 137 | #endif |
| 138 | |
| 139 | STATIC void xfs_buf_error_relse(xfs_buf_t *bp); |
| 140 | STATIC void xfs_buf_do_callbacks(xfs_buf_t *bp, xfs_log_item_t *lip); |
| 141 | |
| 142 | /* |
| 143 | * This returns the number of log iovecs needed to log the |
| 144 | * given buf log item. |
| 145 | * |
| 146 | * It calculates this as 1 iovec for the buf log format structure |
| 147 | * and 1 for each stretch of non-contiguous chunks to be logged. |
| 148 | * Contiguous chunks are logged in a single iovec. |
| 149 | * |
| 150 | * If the XFS_BLI_STALE flag has been set, then log nothing. |
| 151 | */ |
Christoph Hellwig | ba0f32d | 2005-06-21 15:36:52 +1000 | [diff] [blame] | 152 | STATIC uint |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 153 | xfs_buf_item_size( |
| 154 | xfs_buf_log_item_t *bip) |
| 155 | { |
| 156 | uint nvecs; |
| 157 | int next_bit; |
| 158 | int last_bit; |
| 159 | xfs_buf_t *bp; |
| 160 | |
| 161 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
| 162 | if (bip->bli_flags & XFS_BLI_STALE) { |
| 163 | /* |
| 164 | * The buffer is stale, so all we need to log |
| 165 | * is the buf log format structure with the |
| 166 | * cancel flag in it. |
| 167 | */ |
Christoph Hellwig | 0b1b213 | 2009-12-14 23:14:59 +0000 | [diff] [blame] | 168 | trace_xfs_buf_item_size_stale(bip); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 169 | ASSERT(bip->bli_format.blf_flags & XFS_BLI_CANCEL); |
| 170 | return 1; |
| 171 | } |
| 172 | |
| 173 | bp = bip->bli_buf; |
| 174 | ASSERT(bip->bli_flags & XFS_BLI_LOGGED); |
| 175 | nvecs = 1; |
| 176 | last_bit = xfs_next_bit(bip->bli_format.blf_data_map, |
| 177 | bip->bli_format.blf_map_size, 0); |
| 178 | ASSERT(last_bit != -1); |
| 179 | nvecs++; |
| 180 | while (last_bit != -1) { |
| 181 | /* |
| 182 | * This takes the bit number to start looking from and |
| 183 | * returns the next set bit from there. It returns -1 |
| 184 | * if there are no more bits set or the start bit is |
| 185 | * beyond the end of the bitmap. |
| 186 | */ |
| 187 | next_bit = xfs_next_bit(bip->bli_format.blf_data_map, |
| 188 | bip->bli_format.blf_map_size, |
| 189 | last_bit + 1); |
| 190 | /* |
| 191 | * If we run out of bits, leave the loop, |
| 192 | * else if we find a new set of bits bump the number of vecs, |
| 193 | * else keep scanning the current set of bits. |
| 194 | */ |
| 195 | if (next_bit == -1) { |
| 196 | last_bit = -1; |
| 197 | } else if (next_bit != last_bit + 1) { |
| 198 | last_bit = next_bit; |
| 199 | nvecs++; |
| 200 | } else if (xfs_buf_offset(bp, next_bit * XFS_BLI_CHUNK) != |
| 201 | (xfs_buf_offset(bp, last_bit * XFS_BLI_CHUNK) + |
| 202 | XFS_BLI_CHUNK)) { |
| 203 | last_bit = next_bit; |
| 204 | nvecs++; |
| 205 | } else { |
| 206 | last_bit++; |
| 207 | } |
| 208 | } |
| 209 | |
Christoph Hellwig | 0b1b213 | 2009-12-14 23:14:59 +0000 | [diff] [blame] | 210 | trace_xfs_buf_item_size(bip); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 211 | return nvecs; |
| 212 | } |
| 213 | |
| 214 | /* |
| 215 | * This is called to fill in the vector of log iovecs for the |
| 216 | * given log buf item. It fills the first entry with a buf log |
| 217 | * format structure, and the rest point to contiguous chunks |
| 218 | * within the buffer. |
| 219 | */ |
Christoph Hellwig | ba0f32d | 2005-06-21 15:36:52 +1000 | [diff] [blame] | 220 | STATIC void |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 221 | xfs_buf_item_format( |
| 222 | xfs_buf_log_item_t *bip, |
| 223 | xfs_log_iovec_t *log_vector) |
| 224 | { |
| 225 | uint base_size; |
| 226 | uint nvecs; |
| 227 | xfs_log_iovec_t *vecp; |
| 228 | xfs_buf_t *bp; |
| 229 | int first_bit; |
| 230 | int last_bit; |
| 231 | int next_bit; |
| 232 | uint nbits; |
| 233 | uint buffer_offset; |
| 234 | |
| 235 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
| 236 | ASSERT((bip->bli_flags & XFS_BLI_LOGGED) || |
| 237 | (bip->bli_flags & XFS_BLI_STALE)); |
| 238 | bp = bip->bli_buf; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 239 | vecp = log_vector; |
| 240 | |
| 241 | /* |
| 242 | * The size of the base structure is the size of the |
| 243 | * declared structure plus the space for the extra words |
| 244 | * of the bitmap. We subtract one from the map size, because |
| 245 | * the first element of the bitmap is accounted for in the |
| 246 | * size of the base structure. |
| 247 | */ |
| 248 | base_size = |
| 249 | (uint)(sizeof(xfs_buf_log_format_t) + |
| 250 | ((bip->bli_format.blf_map_size - 1) * sizeof(uint))); |
| 251 | vecp->i_addr = (xfs_caddr_t)&bip->bli_format; |
| 252 | vecp->i_len = base_size; |
Christoph Hellwig | 4139b3b | 2010-01-19 09:56:45 +0000 | [diff] [blame] | 253 | vecp->i_type = XLOG_REG_TYPE_BFORMAT; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 254 | vecp++; |
| 255 | nvecs = 1; |
| 256 | |
| 257 | if (bip->bli_flags & XFS_BLI_STALE) { |
| 258 | /* |
| 259 | * The buffer is stale, so all we need to log |
| 260 | * is the buf log format structure with the |
| 261 | * cancel flag in it. |
| 262 | */ |
Christoph Hellwig | 0b1b213 | 2009-12-14 23:14:59 +0000 | [diff] [blame] | 263 | trace_xfs_buf_item_format_stale(bip); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 264 | ASSERT(bip->bli_format.blf_flags & XFS_BLI_CANCEL); |
| 265 | bip->bli_format.blf_size = nvecs; |
| 266 | return; |
| 267 | } |
| 268 | |
| 269 | /* |
| 270 | * Fill in an iovec for each set of contiguous chunks. |
| 271 | */ |
| 272 | first_bit = xfs_next_bit(bip->bli_format.blf_data_map, |
| 273 | bip->bli_format.blf_map_size, 0); |
| 274 | ASSERT(first_bit != -1); |
| 275 | last_bit = first_bit; |
| 276 | nbits = 1; |
| 277 | for (;;) { |
| 278 | /* |
| 279 | * This takes the bit number to start looking from and |
| 280 | * returns the next set bit from there. It returns -1 |
| 281 | * if there are no more bits set or the start bit is |
| 282 | * beyond the end of the bitmap. |
| 283 | */ |
| 284 | next_bit = xfs_next_bit(bip->bli_format.blf_data_map, |
| 285 | bip->bli_format.blf_map_size, |
| 286 | (uint)last_bit + 1); |
| 287 | /* |
| 288 | * If we run out of bits fill in the last iovec and get |
| 289 | * out of the loop. |
| 290 | * Else if we start a new set of bits then fill in the |
| 291 | * iovec for the series we were looking at and start |
| 292 | * counting the bits in the new one. |
| 293 | * Else we're still in the same set of bits so just |
| 294 | * keep counting and scanning. |
| 295 | */ |
| 296 | if (next_bit == -1) { |
| 297 | buffer_offset = first_bit * XFS_BLI_CHUNK; |
| 298 | vecp->i_addr = xfs_buf_offset(bp, buffer_offset); |
| 299 | vecp->i_len = nbits * XFS_BLI_CHUNK; |
Christoph Hellwig | 4139b3b | 2010-01-19 09:56:45 +0000 | [diff] [blame] | 300 | vecp->i_type = XLOG_REG_TYPE_BCHUNK; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 301 | nvecs++; |
| 302 | break; |
| 303 | } else if (next_bit != last_bit + 1) { |
| 304 | buffer_offset = first_bit * XFS_BLI_CHUNK; |
| 305 | vecp->i_addr = xfs_buf_offset(bp, buffer_offset); |
| 306 | vecp->i_len = nbits * XFS_BLI_CHUNK; |
Christoph Hellwig | 4139b3b | 2010-01-19 09:56:45 +0000 | [diff] [blame] | 307 | vecp->i_type = XLOG_REG_TYPE_BCHUNK; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 308 | nvecs++; |
| 309 | vecp++; |
| 310 | first_bit = next_bit; |
| 311 | last_bit = next_bit; |
| 312 | nbits = 1; |
| 313 | } else if (xfs_buf_offset(bp, next_bit << XFS_BLI_SHIFT) != |
| 314 | (xfs_buf_offset(bp, last_bit << XFS_BLI_SHIFT) + |
| 315 | XFS_BLI_CHUNK)) { |
| 316 | buffer_offset = first_bit * XFS_BLI_CHUNK; |
| 317 | vecp->i_addr = xfs_buf_offset(bp, buffer_offset); |
| 318 | vecp->i_len = nbits * XFS_BLI_CHUNK; |
Christoph Hellwig | 4139b3b | 2010-01-19 09:56:45 +0000 | [diff] [blame] | 319 | vecp->i_type = XLOG_REG_TYPE_BCHUNK; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 320 | /* You would think we need to bump the nvecs here too, but we do not |
| 321 | * this number is used by recovery, and it gets confused by the boundary |
| 322 | * split here |
| 323 | * nvecs++; |
| 324 | */ |
| 325 | vecp++; |
| 326 | first_bit = next_bit; |
| 327 | last_bit = next_bit; |
| 328 | nbits = 1; |
| 329 | } else { |
| 330 | last_bit++; |
| 331 | nbits++; |
| 332 | } |
| 333 | } |
| 334 | bip->bli_format.blf_size = nvecs; |
| 335 | |
| 336 | /* |
| 337 | * Check to make sure everything is consistent. |
| 338 | */ |
Christoph Hellwig | 0b1b213 | 2009-12-14 23:14:59 +0000 | [diff] [blame] | 339 | trace_xfs_buf_item_format(bip); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 340 | xfs_buf_item_log_check(bip); |
| 341 | } |
| 342 | |
| 343 | /* |
| 344 | * This is called to pin the buffer associated with the buf log |
| 345 | * item in memory so it cannot be written out. Simply call bpin() |
| 346 | * on the buffer to do this. |
| 347 | */ |
Christoph Hellwig | ba0f32d | 2005-06-21 15:36:52 +1000 | [diff] [blame] | 348 | STATIC void |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 349 | xfs_buf_item_pin( |
| 350 | xfs_buf_log_item_t *bip) |
| 351 | { |
| 352 | xfs_buf_t *bp; |
| 353 | |
| 354 | bp = bip->bli_buf; |
| 355 | ASSERT(XFS_BUF_ISBUSY(bp)); |
| 356 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
| 357 | ASSERT((bip->bli_flags & XFS_BLI_LOGGED) || |
| 358 | (bip->bli_flags & XFS_BLI_STALE)); |
Christoph Hellwig | 0b1b213 | 2009-12-14 23:14:59 +0000 | [diff] [blame] | 359 | trace_xfs_buf_item_pin(bip); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 360 | xfs_bpin(bp); |
| 361 | } |
| 362 | |
| 363 | |
| 364 | /* |
| 365 | * This is called to unpin the buffer associated with the buf log |
| 366 | * item which was previously pinned with a call to xfs_buf_item_pin(). |
| 367 | * Just call bunpin() on the buffer to do this. |
| 368 | * |
| 369 | * Also drop the reference to the buf item for the current transaction. |
| 370 | * If the XFS_BLI_STALE flag is set and we are the last reference, |
| 371 | * then free up the buf log item and unlock the buffer. |
| 372 | */ |
Christoph Hellwig | ba0f32d | 2005-06-21 15:36:52 +1000 | [diff] [blame] | 373 | STATIC void |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 374 | xfs_buf_item_unpin( |
| 375 | xfs_buf_log_item_t *bip, |
| 376 | int stale) |
| 377 | { |
David Chinner | 783a2f6 | 2008-10-30 17:39:58 +1100 | [diff] [blame] | 378 | struct xfs_ail *ailp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 379 | xfs_buf_t *bp; |
| 380 | int freed; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 381 | |
| 382 | bp = bip->bli_buf; |
| 383 | ASSERT(bp != NULL); |
| 384 | ASSERT(XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *) == bip); |
| 385 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
Christoph Hellwig | 0b1b213 | 2009-12-14 23:14:59 +0000 | [diff] [blame] | 386 | trace_xfs_buf_item_unpin(bip); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 387 | |
| 388 | freed = atomic_dec_and_test(&bip->bli_refcount); |
David Chinner | 783a2f6 | 2008-10-30 17:39:58 +1100 | [diff] [blame] | 389 | ailp = bip->bli_item.li_ailp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 390 | xfs_bunpin(bp); |
| 391 | if (freed && stale) { |
| 392 | ASSERT(bip->bli_flags & XFS_BLI_STALE); |
| 393 | ASSERT(XFS_BUF_VALUSEMA(bp) <= 0); |
| 394 | ASSERT(!(XFS_BUF_ISDELAYWRITE(bp))); |
| 395 | ASSERT(XFS_BUF_ISSTALE(bp)); |
| 396 | ASSERT(bip->bli_format.blf_flags & XFS_BLI_CANCEL); |
Christoph Hellwig | 0b1b213 | 2009-12-14 23:14:59 +0000 | [diff] [blame] | 397 | trace_xfs_buf_item_unpin_stale(bip); |
| 398 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 399 | /* |
| 400 | * If we get called here because of an IO error, we may |
David Chinner | 783a2f6 | 2008-10-30 17:39:58 +1100 | [diff] [blame] | 401 | * or may not have the item on the AIL. xfs_trans_ail_delete() |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 402 | * will take care of that situation. |
David Chinner | 783a2f6 | 2008-10-30 17:39:58 +1100 | [diff] [blame] | 403 | * xfs_trans_ail_delete() drops the AIL lock. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 404 | */ |
| 405 | if (bip->bli_flags & XFS_BLI_STALE_INODE) { |
| 406 | xfs_buf_do_callbacks(bp, (xfs_log_item_t *)bip); |
| 407 | XFS_BUF_SET_FSPRIVATE(bp, NULL); |
| 408 | XFS_BUF_CLR_IODONE_FUNC(bp); |
| 409 | } else { |
David Chinner | 783a2f6 | 2008-10-30 17:39:58 +1100 | [diff] [blame] | 410 | spin_lock(&ailp->xa_lock); |
| 411 | xfs_trans_ail_delete(ailp, (xfs_log_item_t *)bip); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 412 | xfs_buf_item_relse(bp); |
| 413 | ASSERT(XFS_BUF_FSPRIVATE(bp, void *) == NULL); |
| 414 | } |
| 415 | xfs_buf_relse(bp); |
| 416 | } |
| 417 | } |
| 418 | |
| 419 | /* |
| 420 | * this is called from uncommit in the forced-shutdown path. |
| 421 | * we need to check to see if the reference count on the log item |
| 422 | * is going to drop to zero. If so, unpin will free the log item |
| 423 | * so we need to free the item's descriptor (that points to the item) |
| 424 | * in the transaction. |
| 425 | */ |
Christoph Hellwig | ba0f32d | 2005-06-21 15:36:52 +1000 | [diff] [blame] | 426 | STATIC void |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 427 | xfs_buf_item_unpin_remove( |
| 428 | xfs_buf_log_item_t *bip, |
| 429 | xfs_trans_t *tp) |
| 430 | { |
| 431 | xfs_buf_t *bp; |
| 432 | xfs_log_item_desc_t *lidp; |
| 433 | int stale = 0; |
| 434 | |
| 435 | bp = bip->bli_buf; |
| 436 | /* |
| 437 | * will xfs_buf_item_unpin() call xfs_buf_item_relse()? |
| 438 | */ |
| 439 | if ((atomic_read(&bip->bli_refcount) == 1) && |
| 440 | (bip->bli_flags & XFS_BLI_STALE)) { |
| 441 | ASSERT(XFS_BUF_VALUSEMA(bip->bli_buf) <= 0); |
Christoph Hellwig | 0b1b213 | 2009-12-14 23:14:59 +0000 | [diff] [blame] | 442 | trace_xfs_buf_item_unpin_stale(bip); |
| 443 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 444 | /* |
| 445 | * yes -- clear the xaction descriptor in-use flag |
| 446 | * and free the chunk if required. We can safely |
| 447 | * do some work here and then call buf_item_unpin |
| 448 | * to do the rest because if the if is true, then |
| 449 | * we are holding the buffer locked so no one else |
| 450 | * will be able to bump up the refcount. |
| 451 | */ |
| 452 | lidp = xfs_trans_find_item(tp, (xfs_log_item_t *) bip); |
| 453 | stale = lidp->lid_flags & XFS_LID_BUF_STALE; |
| 454 | xfs_trans_free_item(tp, lidp); |
| 455 | /* |
| 456 | * Since the transaction no longer refers to the buffer, |
| 457 | * the buffer should no longer refer to the transaction. |
| 458 | */ |
| 459 | XFS_BUF_SET_FSPRIVATE2(bp, NULL); |
| 460 | } |
| 461 | |
| 462 | xfs_buf_item_unpin(bip, stale); |
| 463 | |
| 464 | return; |
| 465 | } |
| 466 | |
| 467 | /* |
| 468 | * This is called to attempt to lock the buffer associated with this |
| 469 | * buf log item. Don't sleep on the buffer lock. If we can't get |
Dave Chinner | d808f61 | 2010-02-02 10:13:42 +1100 | [diff] [blame] | 470 | * the lock right away, return 0. If we can get the lock, take a |
| 471 | * reference to the buffer. If this is a delayed write buffer that |
| 472 | * needs AIL help to be written back, invoke the pushbuf routine |
| 473 | * rather than the normal success path. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 474 | */ |
Christoph Hellwig | ba0f32d | 2005-06-21 15:36:52 +1000 | [diff] [blame] | 475 | STATIC uint |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 476 | xfs_buf_item_trylock( |
| 477 | xfs_buf_log_item_t *bip) |
| 478 | { |
| 479 | xfs_buf_t *bp; |
| 480 | |
| 481 | bp = bip->bli_buf; |
Dave Chinner | d808f61 | 2010-02-02 10:13:42 +1100 | [diff] [blame] | 482 | if (XFS_BUF_ISPINNED(bp)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 483 | return XFS_ITEM_PINNED; |
Dave Chinner | d808f61 | 2010-02-02 10:13:42 +1100 | [diff] [blame] | 484 | if (!XFS_BUF_CPSEMA(bp)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 485 | return XFS_ITEM_LOCKED; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 486 | |
Dave Chinner | d808f61 | 2010-02-02 10:13:42 +1100 | [diff] [blame] | 487 | /* take a reference to the buffer. */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 488 | XFS_BUF_HOLD(bp); |
| 489 | |
| 490 | ASSERT(!(bip->bli_flags & XFS_BLI_STALE)); |
Christoph Hellwig | 0b1b213 | 2009-12-14 23:14:59 +0000 | [diff] [blame] | 491 | trace_xfs_buf_item_trylock(bip); |
Dave Chinner | d808f61 | 2010-02-02 10:13:42 +1100 | [diff] [blame] | 492 | if (XFS_BUF_ISDELAYWRITE(bp)) |
| 493 | return XFS_ITEM_PUSHBUF; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 494 | return XFS_ITEM_SUCCESS; |
| 495 | } |
| 496 | |
| 497 | /* |
| 498 | * Release the buffer associated with the buf log item. |
| 499 | * If there is no dirty logged data associated with the |
| 500 | * buffer recorded in the buf log item, then free the |
| 501 | * buf log item and remove the reference to it in the |
| 502 | * buffer. |
| 503 | * |
| 504 | * This call ignores the recursion count. It is only called |
| 505 | * when the buffer should REALLY be unlocked, regardless |
| 506 | * of the recursion count. |
| 507 | * |
| 508 | * If the XFS_BLI_HOLD flag is set in the buf log item, then |
| 509 | * free the log item if necessary but do not unlock the buffer. |
| 510 | * This is for support of xfs_trans_bhold(). Make sure the |
| 511 | * XFS_BLI_HOLD field is cleared if we don't free the item. |
| 512 | */ |
Christoph Hellwig | ba0f32d | 2005-06-21 15:36:52 +1000 | [diff] [blame] | 513 | STATIC void |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 514 | xfs_buf_item_unlock( |
| 515 | xfs_buf_log_item_t *bip) |
| 516 | { |
| 517 | int aborted; |
| 518 | xfs_buf_t *bp; |
| 519 | uint hold; |
| 520 | |
| 521 | bp = bip->bli_buf; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 522 | |
| 523 | /* |
| 524 | * Clear the buffer's association with this transaction. |
| 525 | */ |
| 526 | XFS_BUF_SET_FSPRIVATE2(bp, NULL); |
| 527 | |
| 528 | /* |
| 529 | * If this is a transaction abort, don't return early. |
| 530 | * Instead, allow the brelse to happen. |
| 531 | * Normally it would be done for stale (cancelled) buffers |
| 532 | * at unpin time, but we'll never go through the pin/unpin |
| 533 | * cycle if we abort inside commit. |
| 534 | */ |
| 535 | aborted = (bip->bli_item.li_flags & XFS_LI_ABORTED) != 0; |
| 536 | |
| 537 | /* |
| 538 | * If the buf item is marked stale, then don't do anything. |
| 539 | * We'll unlock the buffer and free the buf item when the |
| 540 | * buffer is unpinned for the last time. |
| 541 | */ |
| 542 | if (bip->bli_flags & XFS_BLI_STALE) { |
| 543 | bip->bli_flags &= ~XFS_BLI_LOGGED; |
Christoph Hellwig | 0b1b213 | 2009-12-14 23:14:59 +0000 | [diff] [blame] | 544 | trace_xfs_buf_item_unlock_stale(bip); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 545 | ASSERT(bip->bli_format.blf_flags & XFS_BLI_CANCEL); |
| 546 | if (!aborted) |
| 547 | return; |
| 548 | } |
| 549 | |
| 550 | /* |
| 551 | * Drop the transaction's reference to the log item if |
| 552 | * it was not logged as part of the transaction. Otherwise |
| 553 | * we'll drop the reference in xfs_buf_item_unpin() when |
| 554 | * the transaction is really through with the buffer. |
| 555 | */ |
| 556 | if (!(bip->bli_flags & XFS_BLI_LOGGED)) { |
| 557 | atomic_dec(&bip->bli_refcount); |
| 558 | } else { |
| 559 | /* |
| 560 | * Clear the logged flag since this is per |
| 561 | * transaction state. |
| 562 | */ |
| 563 | bip->bli_flags &= ~XFS_BLI_LOGGED; |
| 564 | } |
| 565 | |
| 566 | /* |
| 567 | * Before possibly freeing the buf item, determine if we should |
| 568 | * release the buffer at the end of this routine. |
| 569 | */ |
| 570 | hold = bip->bli_flags & XFS_BLI_HOLD; |
Christoph Hellwig | 0b1b213 | 2009-12-14 23:14:59 +0000 | [diff] [blame] | 571 | trace_xfs_buf_item_unlock(bip); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 572 | |
| 573 | /* |
| 574 | * If the buf item isn't tracking any data, free it. |
| 575 | * Otherwise, if XFS_BLI_HOLD is set clear it. |
| 576 | */ |
Eric Sandeen | 24ad33f | 2007-06-28 16:43:30 +1000 | [diff] [blame] | 577 | if (xfs_bitmap_empty(bip->bli_format.blf_data_map, |
| 578 | bip->bli_format.blf_map_size)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 579 | xfs_buf_item_relse(bp); |
| 580 | } else if (hold) { |
| 581 | bip->bli_flags &= ~XFS_BLI_HOLD; |
| 582 | } |
| 583 | |
| 584 | /* |
| 585 | * Release the buffer if XFS_BLI_HOLD was not set. |
| 586 | */ |
| 587 | if (!hold) { |
| 588 | xfs_buf_relse(bp); |
| 589 | } |
| 590 | } |
| 591 | |
| 592 | /* |
| 593 | * This is called to find out where the oldest active copy of the |
| 594 | * buf log item in the on disk log resides now that the last log |
| 595 | * write of it completed at the given lsn. |
| 596 | * We always re-log all the dirty data in a buffer, so usually the |
| 597 | * latest copy in the on disk log is the only one that matters. For |
| 598 | * those cases we simply return the given lsn. |
| 599 | * |
| 600 | * The one exception to this is for buffers full of newly allocated |
| 601 | * inodes. These buffers are only relogged with the XFS_BLI_INODE_BUF |
| 602 | * flag set, indicating that only the di_next_unlinked fields from the |
| 603 | * inodes in the buffers will be replayed during recovery. If the |
| 604 | * original newly allocated inode images have not yet been flushed |
| 605 | * when the buffer is so relogged, then we need to make sure that we |
| 606 | * keep the old images in the 'active' portion of the log. We do this |
| 607 | * by returning the original lsn of that transaction here rather than |
| 608 | * the current one. |
| 609 | */ |
Christoph Hellwig | ba0f32d | 2005-06-21 15:36:52 +1000 | [diff] [blame] | 610 | STATIC xfs_lsn_t |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 611 | xfs_buf_item_committed( |
| 612 | xfs_buf_log_item_t *bip, |
| 613 | xfs_lsn_t lsn) |
| 614 | { |
Christoph Hellwig | 0b1b213 | 2009-12-14 23:14:59 +0000 | [diff] [blame] | 615 | trace_xfs_buf_item_committed(bip); |
| 616 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 617 | if ((bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF) && |
| 618 | (bip->bli_item.li_lsn != 0)) { |
| 619 | return bip->bli_item.li_lsn; |
| 620 | } |
| 621 | return (lsn); |
| 622 | } |
| 623 | |
| 624 | /* |
Dave Chinner | d808f61 | 2010-02-02 10:13:42 +1100 | [diff] [blame] | 625 | * The buffer is locked, but is not a delayed write buffer. This happens |
| 626 | * if we race with IO completion and hence we don't want to try to write it |
| 627 | * again. Just release the buffer. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 628 | */ |
Christoph Hellwig | ba0f32d | 2005-06-21 15:36:52 +1000 | [diff] [blame] | 629 | STATIC void |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 630 | xfs_buf_item_push( |
| 631 | xfs_buf_log_item_t *bip) |
| 632 | { |
| 633 | xfs_buf_t *bp; |
| 634 | |
| 635 | ASSERT(!(bip->bli_flags & XFS_BLI_STALE)); |
Christoph Hellwig | 0b1b213 | 2009-12-14 23:14:59 +0000 | [diff] [blame] | 636 | trace_xfs_buf_item_push(bip); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 637 | |
| 638 | bp = bip->bli_buf; |
Dave Chinner | d808f61 | 2010-02-02 10:13:42 +1100 | [diff] [blame] | 639 | ASSERT(!XFS_BUF_ISDELAYWRITE(bp)); |
| 640 | xfs_buf_relse(bp); |
| 641 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 642 | |
Dave Chinner | d808f61 | 2010-02-02 10:13:42 +1100 | [diff] [blame] | 643 | /* |
| 644 | * The buffer is locked and is a delayed write buffer. Promote the buffer |
| 645 | * in the delayed write queue as the caller knows that they must invoke |
| 646 | * the xfsbufd to get this buffer written. We have to unlock the buffer |
| 647 | * to allow the xfsbufd to write it, too. |
| 648 | */ |
| 649 | STATIC void |
| 650 | xfs_buf_item_pushbuf( |
| 651 | xfs_buf_log_item_t *bip) |
| 652 | { |
| 653 | xfs_buf_t *bp; |
| 654 | |
| 655 | ASSERT(!(bip->bli_flags & XFS_BLI_STALE)); |
| 656 | trace_xfs_buf_item_pushbuf(bip); |
| 657 | |
| 658 | bp = bip->bli_buf; |
| 659 | ASSERT(XFS_BUF_ISDELAYWRITE(bp)); |
| 660 | xfs_buf_delwri_promote(bp); |
| 661 | xfs_buf_relse(bp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 662 | } |
| 663 | |
| 664 | /* ARGSUSED */ |
Christoph Hellwig | ba0f32d | 2005-06-21 15:36:52 +1000 | [diff] [blame] | 665 | STATIC void |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 666 | xfs_buf_item_committing(xfs_buf_log_item_t *bip, xfs_lsn_t commit_lsn) |
| 667 | { |
| 668 | } |
| 669 | |
| 670 | /* |
| 671 | * This is the ops vector shared by all buf log items. |
| 672 | */ |
David Chinner | 7989cb8 | 2007-02-10 18:34:56 +1100 | [diff] [blame] | 673 | static struct xfs_item_ops xfs_buf_item_ops = { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 674 | .iop_size = (uint(*)(xfs_log_item_t*))xfs_buf_item_size, |
| 675 | .iop_format = (void(*)(xfs_log_item_t*, xfs_log_iovec_t*)) |
| 676 | xfs_buf_item_format, |
| 677 | .iop_pin = (void(*)(xfs_log_item_t*))xfs_buf_item_pin, |
| 678 | .iop_unpin = (void(*)(xfs_log_item_t*, int))xfs_buf_item_unpin, |
| 679 | .iop_unpin_remove = (void(*)(xfs_log_item_t*, xfs_trans_t *)) |
| 680 | xfs_buf_item_unpin_remove, |
| 681 | .iop_trylock = (uint(*)(xfs_log_item_t*))xfs_buf_item_trylock, |
| 682 | .iop_unlock = (void(*)(xfs_log_item_t*))xfs_buf_item_unlock, |
| 683 | .iop_committed = (xfs_lsn_t(*)(xfs_log_item_t*, xfs_lsn_t)) |
| 684 | xfs_buf_item_committed, |
| 685 | .iop_push = (void(*)(xfs_log_item_t*))xfs_buf_item_push, |
Dave Chinner | d808f61 | 2010-02-02 10:13:42 +1100 | [diff] [blame] | 686 | .iop_pushbuf = (void(*)(xfs_log_item_t*))xfs_buf_item_pushbuf, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 687 | .iop_committing = (void(*)(xfs_log_item_t*, xfs_lsn_t)) |
| 688 | xfs_buf_item_committing |
| 689 | }; |
| 690 | |
| 691 | |
| 692 | /* |
| 693 | * Allocate a new buf log item to go with the given buffer. |
| 694 | * Set the buffer's b_fsprivate field to point to the new |
| 695 | * buf log item. If there are other item's attached to the |
| 696 | * buffer (see xfs_buf_attach_iodone() below), then put the |
| 697 | * buf log item at the front. |
| 698 | */ |
| 699 | void |
| 700 | xfs_buf_item_init( |
| 701 | xfs_buf_t *bp, |
| 702 | xfs_mount_t *mp) |
| 703 | { |
| 704 | xfs_log_item_t *lip; |
| 705 | xfs_buf_log_item_t *bip; |
| 706 | int chunks; |
| 707 | int map_size; |
| 708 | |
| 709 | /* |
| 710 | * Check to see if there is already a buf log item for |
| 711 | * this buffer. If there is, it is guaranteed to be |
| 712 | * the first. If we do already have one, there is |
| 713 | * nothing to do here so return. |
| 714 | */ |
Christoph Hellwig | 15ac08a | 2008-12-09 04:47:30 -0500 | [diff] [blame] | 715 | if (bp->b_mount != mp) |
| 716 | bp->b_mount = mp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 717 | XFS_BUF_SET_BDSTRAT_FUNC(bp, xfs_bdstrat_cb); |
| 718 | if (XFS_BUF_FSPRIVATE(bp, void *) != NULL) { |
| 719 | lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *); |
| 720 | if (lip->li_type == XFS_LI_BUF) { |
| 721 | return; |
| 722 | } |
| 723 | } |
| 724 | |
| 725 | /* |
| 726 | * chunks is the number of XFS_BLI_CHUNK size pieces |
| 727 | * the buffer can be divided into. Make sure not to |
| 728 | * truncate any pieces. map_size is the size of the |
| 729 | * bitmap needed to describe the chunks of the buffer. |
| 730 | */ |
| 731 | chunks = (int)((XFS_BUF_COUNT(bp) + (XFS_BLI_CHUNK - 1)) >> XFS_BLI_SHIFT); |
| 732 | map_size = (int)((chunks + NBWORD) >> BIT_TO_WORD_SHIFT); |
| 733 | |
| 734 | bip = (xfs_buf_log_item_t*)kmem_zone_zalloc(xfs_buf_item_zone, |
| 735 | KM_SLEEP); |
| 736 | bip->bli_item.li_type = XFS_LI_BUF; |
| 737 | bip->bli_item.li_ops = &xfs_buf_item_ops; |
| 738 | bip->bli_item.li_mountp = mp; |
David Chinner | fc1829f | 2008-10-30 17:39:46 +1100 | [diff] [blame] | 739 | bip->bli_item.li_ailp = mp->m_ail; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 740 | bip->bli_buf = bp; |
Lachlan McIlroy | e1f5dbd | 2008-09-17 16:52:13 +1000 | [diff] [blame] | 741 | xfs_buf_hold(bp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 742 | bip->bli_format.blf_type = XFS_LI_BUF; |
| 743 | bip->bli_format.blf_blkno = (__int64_t)XFS_BUF_ADDR(bp); |
| 744 | bip->bli_format.blf_len = (ushort)BTOBB(XFS_BUF_COUNT(bp)); |
| 745 | bip->bli_format.blf_map_size = map_size; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 746 | |
| 747 | #ifdef XFS_TRANS_DEBUG |
| 748 | /* |
| 749 | * Allocate the arrays for tracking what needs to be logged |
| 750 | * and what our callers request to be logged. bli_orig |
| 751 | * holds a copy of the original, clean buffer for comparison |
| 752 | * against, and bli_logged keeps a 1 bit flag per byte in |
| 753 | * the buffer to indicate which bytes the callers have asked |
| 754 | * to have logged. |
| 755 | */ |
| 756 | bip->bli_orig = (char *)kmem_alloc(XFS_BUF_COUNT(bp), KM_SLEEP); |
| 757 | memcpy(bip->bli_orig, XFS_BUF_PTR(bp), XFS_BUF_COUNT(bp)); |
| 758 | bip->bli_logged = (char *)kmem_zalloc(XFS_BUF_COUNT(bp) / NBBY, KM_SLEEP); |
| 759 | #endif |
| 760 | |
| 761 | /* |
| 762 | * Put the buf item into the list of items attached to the |
| 763 | * buffer at the front. |
| 764 | */ |
| 765 | if (XFS_BUF_FSPRIVATE(bp, void *) != NULL) { |
| 766 | bip->bli_item.li_bio_list = |
| 767 | XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *); |
| 768 | } |
| 769 | XFS_BUF_SET_FSPRIVATE(bp, bip); |
| 770 | } |
| 771 | |
| 772 | |
| 773 | /* |
| 774 | * Mark bytes first through last inclusive as dirty in the buf |
| 775 | * item's bitmap. |
| 776 | */ |
| 777 | void |
| 778 | xfs_buf_item_log( |
| 779 | xfs_buf_log_item_t *bip, |
| 780 | uint first, |
| 781 | uint last) |
| 782 | { |
| 783 | uint first_bit; |
| 784 | uint last_bit; |
| 785 | uint bits_to_set; |
| 786 | uint bits_set; |
| 787 | uint word_num; |
| 788 | uint *wordp; |
| 789 | uint bit; |
| 790 | uint end_bit; |
| 791 | uint mask; |
| 792 | |
| 793 | /* |
| 794 | * Mark the item as having some dirty data for |
| 795 | * quick reference in xfs_buf_item_dirty. |
| 796 | */ |
| 797 | bip->bli_flags |= XFS_BLI_DIRTY; |
| 798 | |
| 799 | /* |
| 800 | * Convert byte offsets to bit numbers. |
| 801 | */ |
| 802 | first_bit = first >> XFS_BLI_SHIFT; |
| 803 | last_bit = last >> XFS_BLI_SHIFT; |
| 804 | |
| 805 | /* |
| 806 | * Calculate the total number of bits to be set. |
| 807 | */ |
| 808 | bits_to_set = last_bit - first_bit + 1; |
| 809 | |
| 810 | /* |
| 811 | * Get a pointer to the first word in the bitmap |
| 812 | * to set a bit in. |
| 813 | */ |
| 814 | word_num = first_bit >> BIT_TO_WORD_SHIFT; |
| 815 | wordp = &(bip->bli_format.blf_data_map[word_num]); |
| 816 | |
| 817 | /* |
| 818 | * Calculate the starting bit in the first word. |
| 819 | */ |
| 820 | bit = first_bit & (uint)(NBWORD - 1); |
| 821 | |
| 822 | /* |
| 823 | * First set any bits in the first word of our range. |
| 824 | * If it starts at bit 0 of the word, it will be |
| 825 | * set below rather than here. That is what the variable |
| 826 | * bit tells us. The variable bits_set tracks the number |
| 827 | * of bits that have been set so far. End_bit is the number |
| 828 | * of the last bit to be set in this word plus one. |
| 829 | */ |
| 830 | if (bit) { |
| 831 | end_bit = MIN(bit + bits_to_set, (uint)NBWORD); |
| 832 | mask = ((1 << (end_bit - bit)) - 1) << bit; |
| 833 | *wordp |= mask; |
| 834 | wordp++; |
| 835 | bits_set = end_bit - bit; |
| 836 | } else { |
| 837 | bits_set = 0; |
| 838 | } |
| 839 | |
| 840 | /* |
| 841 | * Now set bits a whole word at a time that are between |
| 842 | * first_bit and last_bit. |
| 843 | */ |
| 844 | while ((bits_to_set - bits_set) >= NBWORD) { |
| 845 | *wordp |= 0xffffffff; |
| 846 | bits_set += NBWORD; |
| 847 | wordp++; |
| 848 | } |
| 849 | |
| 850 | /* |
| 851 | * Finally, set any bits left to be set in one last partial word. |
| 852 | */ |
| 853 | end_bit = bits_to_set - bits_set; |
| 854 | if (end_bit) { |
| 855 | mask = (1 << end_bit) - 1; |
| 856 | *wordp |= mask; |
| 857 | } |
| 858 | |
| 859 | xfs_buf_item_log_debug(bip, first, last); |
| 860 | } |
| 861 | |
| 862 | |
| 863 | /* |
| 864 | * Return 1 if the buffer has some data that has been logged (at any |
| 865 | * point, not just the current transaction) and 0 if not. |
| 866 | */ |
| 867 | uint |
| 868 | xfs_buf_item_dirty( |
| 869 | xfs_buf_log_item_t *bip) |
| 870 | { |
| 871 | return (bip->bli_flags & XFS_BLI_DIRTY); |
| 872 | } |
| 873 | |
Lachlan McIlroy | e1f5dbd | 2008-09-17 16:52:13 +1000 | [diff] [blame] | 874 | STATIC void |
| 875 | xfs_buf_item_free( |
| 876 | xfs_buf_log_item_t *bip) |
| 877 | { |
| 878 | #ifdef XFS_TRANS_DEBUG |
| 879 | kmem_free(bip->bli_orig); |
| 880 | kmem_free(bip->bli_logged); |
| 881 | #endif /* XFS_TRANS_DEBUG */ |
| 882 | |
Lachlan McIlroy | e1f5dbd | 2008-09-17 16:52:13 +1000 | [diff] [blame] | 883 | kmem_zone_free(xfs_buf_item_zone, bip); |
| 884 | } |
| 885 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 886 | /* |
| 887 | * This is called when the buf log item is no longer needed. It should |
| 888 | * free the buf log item associated with the given buffer and clear |
| 889 | * the buffer's pointer to the buf log item. If there are no more |
| 890 | * items in the list, clear the b_iodone field of the buffer (see |
| 891 | * xfs_buf_attach_iodone() below). |
| 892 | */ |
| 893 | void |
| 894 | xfs_buf_item_relse( |
| 895 | xfs_buf_t *bp) |
| 896 | { |
| 897 | xfs_buf_log_item_t *bip; |
| 898 | |
Christoph Hellwig | 0b1b213 | 2009-12-14 23:14:59 +0000 | [diff] [blame] | 899 | trace_xfs_buf_item_relse(bp, _RET_IP_); |
| 900 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 901 | bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t*); |
| 902 | XFS_BUF_SET_FSPRIVATE(bp, bip->bli_item.li_bio_list); |
| 903 | if ((XFS_BUF_FSPRIVATE(bp, void *) == NULL) && |
| 904 | (XFS_BUF_IODONE_FUNC(bp) != NULL)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 905 | XFS_BUF_CLR_IODONE_FUNC(bp); |
| 906 | } |
Lachlan McIlroy | e1f5dbd | 2008-09-17 16:52:13 +1000 | [diff] [blame] | 907 | xfs_buf_rele(bp); |
| 908 | xfs_buf_item_free(bip); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 909 | } |
| 910 | |
| 911 | |
| 912 | /* |
| 913 | * Add the given log item with its callback to the list of callbacks |
| 914 | * to be called when the buffer's I/O completes. If it is not set |
| 915 | * already, set the buffer's b_iodone() routine to be |
| 916 | * xfs_buf_iodone_callbacks() and link the log item into the list of |
| 917 | * items rooted at b_fsprivate. Items are always added as the second |
| 918 | * entry in the list if there is a first, because the buf item code |
| 919 | * assumes that the buf log item is first. |
| 920 | */ |
| 921 | void |
| 922 | xfs_buf_attach_iodone( |
| 923 | xfs_buf_t *bp, |
| 924 | void (*cb)(xfs_buf_t *, xfs_log_item_t *), |
| 925 | xfs_log_item_t *lip) |
| 926 | { |
| 927 | xfs_log_item_t *head_lip; |
| 928 | |
| 929 | ASSERT(XFS_BUF_ISBUSY(bp)); |
| 930 | ASSERT(XFS_BUF_VALUSEMA(bp) <= 0); |
| 931 | |
| 932 | lip->li_cb = cb; |
| 933 | if (XFS_BUF_FSPRIVATE(bp, void *) != NULL) { |
| 934 | head_lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *); |
| 935 | lip->li_bio_list = head_lip->li_bio_list; |
| 936 | head_lip->li_bio_list = lip; |
| 937 | } else { |
| 938 | XFS_BUF_SET_FSPRIVATE(bp, lip); |
| 939 | } |
| 940 | |
| 941 | ASSERT((XFS_BUF_IODONE_FUNC(bp) == xfs_buf_iodone_callbacks) || |
| 942 | (XFS_BUF_IODONE_FUNC(bp) == NULL)); |
| 943 | XFS_BUF_SET_IODONE_FUNC(bp, xfs_buf_iodone_callbacks); |
| 944 | } |
| 945 | |
| 946 | STATIC void |
| 947 | xfs_buf_do_callbacks( |
| 948 | xfs_buf_t *bp, |
| 949 | xfs_log_item_t *lip) |
| 950 | { |
| 951 | xfs_log_item_t *nlip; |
| 952 | |
| 953 | while (lip != NULL) { |
| 954 | nlip = lip->li_bio_list; |
| 955 | ASSERT(lip->li_cb != NULL); |
| 956 | /* |
| 957 | * Clear the next pointer so we don't have any |
| 958 | * confusion if the item is added to another buf. |
| 959 | * Don't touch the log item after calling its |
| 960 | * callback, because it could have freed itself. |
| 961 | */ |
| 962 | lip->li_bio_list = NULL; |
| 963 | lip->li_cb(bp, lip); |
| 964 | lip = nlip; |
| 965 | } |
| 966 | } |
| 967 | |
| 968 | /* |
| 969 | * This is the iodone() function for buffers which have had callbacks |
| 970 | * attached to them by xfs_buf_attach_iodone(). It should remove each |
| 971 | * log item from the buffer's list and call the callback of each in turn. |
| 972 | * When done, the buffer's fsprivate field is set to NULL and the buffer |
| 973 | * is unlocked with a call to iodone(). |
| 974 | */ |
| 975 | void |
| 976 | xfs_buf_iodone_callbacks( |
| 977 | xfs_buf_t *bp) |
| 978 | { |
| 979 | xfs_log_item_t *lip; |
| 980 | static ulong lasttime; |
| 981 | static xfs_buftarg_t *lasttarg; |
| 982 | xfs_mount_t *mp; |
| 983 | |
| 984 | ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL); |
| 985 | lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *); |
| 986 | |
| 987 | if (XFS_BUF_GETERROR(bp) != 0) { |
| 988 | /* |
| 989 | * If we've already decided to shutdown the filesystem |
| 990 | * because of IO errors, there's no point in giving this |
| 991 | * a retry. |
| 992 | */ |
| 993 | mp = lip->li_mountp; |
| 994 | if (XFS_FORCED_SHUTDOWN(mp)) { |
| 995 | ASSERT(XFS_BUF_TARGET(bp) == mp->m_ddev_targp); |
| 996 | XFS_BUF_SUPER_STALE(bp); |
Christoph Hellwig | 0b1b213 | 2009-12-14 23:14:59 +0000 | [diff] [blame] | 997 | trace_xfs_buf_item_iodone(bp, _RET_IP_); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 998 | xfs_buf_do_callbacks(bp, lip); |
| 999 | XFS_BUF_SET_FSPRIVATE(bp, NULL); |
| 1000 | XFS_BUF_CLR_IODONE_FUNC(bp); |
Lachlan McIlroy | 4fdc778 | 2008-12-22 17:52:58 +1100 | [diff] [blame] | 1001 | xfs_biodone(bp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1002 | return; |
| 1003 | } |
| 1004 | |
| 1005 | if ((XFS_BUF_TARGET(bp) != lasttarg) || |
| 1006 | (time_after(jiffies, (lasttime + 5*HZ)))) { |
| 1007 | lasttime = jiffies; |
Nathan Scott | b657452 | 2006-06-09 15:29:40 +1000 | [diff] [blame] | 1008 | cmn_err(CE_ALERT, "Device %s, XFS metadata write error" |
| 1009 | " block 0x%llx in %s", |
| 1010 | XFS_BUFTARG_NAME(XFS_BUF_TARGET(bp)), |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1011 | (__uint64_t)XFS_BUF_ADDR(bp), mp->m_fsname); |
| 1012 | } |
| 1013 | lasttarg = XFS_BUF_TARGET(bp); |
| 1014 | |
| 1015 | if (XFS_BUF_ISASYNC(bp)) { |
| 1016 | /* |
| 1017 | * If the write was asynchronous then noone will be |
| 1018 | * looking for the error. Clear the error state |
| 1019 | * and write the buffer out again delayed write. |
| 1020 | * |
| 1021 | * XXXsup This is OK, so long as we catch these |
| 1022 | * before we start the umount; we don't want these |
| 1023 | * DELWRI metadata bufs to be hanging around. |
| 1024 | */ |
| 1025 | XFS_BUF_ERROR(bp,0); /* errno of 0 unsets the flag */ |
| 1026 | |
| 1027 | if (!(XFS_BUF_ISSTALE(bp))) { |
| 1028 | XFS_BUF_DELAYWRITE(bp); |
| 1029 | XFS_BUF_DONE(bp); |
| 1030 | XFS_BUF_SET_START(bp); |
| 1031 | } |
| 1032 | ASSERT(XFS_BUF_IODONE_FUNC(bp)); |
Christoph Hellwig | 0b1b213 | 2009-12-14 23:14:59 +0000 | [diff] [blame] | 1033 | trace_xfs_buf_item_iodone_async(bp, _RET_IP_); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1034 | xfs_buf_relse(bp); |
| 1035 | } else { |
| 1036 | /* |
| 1037 | * If the write of the buffer was not asynchronous, |
| 1038 | * then we want to make sure to return the error |
| 1039 | * to the caller of bwrite(). Because of this we |
| 1040 | * cannot clear the B_ERROR state at this point. |
| 1041 | * Instead we install a callback function that |
| 1042 | * will be called when the buffer is released, and |
| 1043 | * that routine will clear the error state and |
| 1044 | * set the buffer to be written out again after |
| 1045 | * some delay. |
| 1046 | */ |
| 1047 | /* We actually overwrite the existing b-relse |
| 1048 | function at times, but we're gonna be shutting down |
| 1049 | anyway. */ |
| 1050 | XFS_BUF_SET_BRELSE_FUNC(bp,xfs_buf_error_relse); |
| 1051 | XFS_BUF_DONE(bp); |
David Chinner | b4dd330 | 2008-08-13 16:36:11 +1000 | [diff] [blame] | 1052 | XFS_BUF_FINISH_IOWAIT(bp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1053 | } |
| 1054 | return; |
| 1055 | } |
Christoph Hellwig | 0b1b213 | 2009-12-14 23:14:59 +0000 | [diff] [blame] | 1056 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1057 | xfs_buf_do_callbacks(bp, lip); |
| 1058 | XFS_BUF_SET_FSPRIVATE(bp, NULL); |
| 1059 | XFS_BUF_CLR_IODONE_FUNC(bp); |
| 1060 | xfs_biodone(bp); |
| 1061 | } |
| 1062 | |
| 1063 | /* |
| 1064 | * This is a callback routine attached to a buffer which gets an error |
| 1065 | * when being written out synchronously. |
| 1066 | */ |
| 1067 | STATIC void |
| 1068 | xfs_buf_error_relse( |
| 1069 | xfs_buf_t *bp) |
| 1070 | { |
| 1071 | xfs_log_item_t *lip; |
| 1072 | xfs_mount_t *mp; |
| 1073 | |
| 1074 | lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *); |
| 1075 | mp = (xfs_mount_t *)lip->li_mountp; |
| 1076 | ASSERT(XFS_BUF_TARGET(bp) == mp->m_ddev_targp); |
| 1077 | |
| 1078 | XFS_BUF_STALE(bp); |
| 1079 | XFS_BUF_DONE(bp); |
| 1080 | XFS_BUF_UNDELAYWRITE(bp); |
| 1081 | XFS_BUF_ERROR(bp,0); |
Christoph Hellwig | 0b1b213 | 2009-12-14 23:14:59 +0000 | [diff] [blame] | 1082 | |
| 1083 | trace_xfs_buf_error_relse(bp, _RET_IP_); |
| 1084 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1085 | if (! XFS_FORCED_SHUTDOWN(mp)) |
Nathan Scott | 7d04a33 | 2006-06-09 14:58:38 +1000 | [diff] [blame] | 1086 | xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1087 | /* |
| 1088 | * We have to unpin the pinned buffers so do the |
| 1089 | * callbacks. |
| 1090 | */ |
| 1091 | xfs_buf_do_callbacks(bp, lip); |
| 1092 | XFS_BUF_SET_FSPRIVATE(bp, NULL); |
| 1093 | XFS_BUF_CLR_IODONE_FUNC(bp); |
| 1094 | XFS_BUF_SET_BRELSE_FUNC(bp,NULL); |
| 1095 | xfs_buf_relse(bp); |
| 1096 | } |
| 1097 | |
| 1098 | |
| 1099 | /* |
| 1100 | * This is the iodone() function for buffers which have been |
| 1101 | * logged. It is called when they are eventually flushed out. |
| 1102 | * It should remove the buf item from the AIL, and free the buf item. |
| 1103 | * It is called by xfs_buf_iodone_callbacks() above which will take |
| 1104 | * care of cleaning up the buffer itself. |
| 1105 | */ |
| 1106 | /* ARGSUSED */ |
| 1107 | void |
| 1108 | xfs_buf_iodone( |
| 1109 | xfs_buf_t *bp, |
| 1110 | xfs_buf_log_item_t *bip) |
| 1111 | { |
David Chinner | 783a2f6 | 2008-10-30 17:39:58 +1100 | [diff] [blame] | 1112 | struct xfs_ail *ailp = bip->bli_item.li_ailp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1113 | |
| 1114 | ASSERT(bip->bli_buf == bp); |
| 1115 | |
Lachlan McIlroy | e1f5dbd | 2008-09-17 16:52:13 +1000 | [diff] [blame] | 1116 | xfs_buf_rele(bp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1117 | |
| 1118 | /* |
| 1119 | * If we are forcibly shutting down, this may well be |
| 1120 | * off the AIL already. That's because we simulate the |
| 1121 | * log-committed callbacks to unpin these buffers. Or we may never |
| 1122 | * have put this item on AIL because of the transaction was |
David Chinner | 783a2f6 | 2008-10-30 17:39:58 +1100 | [diff] [blame] | 1123 | * aborted forcibly. xfs_trans_ail_delete() takes care of these. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1124 | * |
| 1125 | * Either way, AIL is useless if we're forcing a shutdown. |
| 1126 | */ |
David Chinner | fc1829f | 2008-10-30 17:39:46 +1100 | [diff] [blame] | 1127 | spin_lock(&ailp->xa_lock); |
David Chinner | 783a2f6 | 2008-10-30 17:39:58 +1100 | [diff] [blame] | 1128 | xfs_trans_ail_delete(ailp, (xfs_log_item_t *)bip); |
Lachlan McIlroy | e1f5dbd | 2008-09-17 16:52:13 +1000 | [diff] [blame] | 1129 | xfs_buf_item_free(bip); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1130 | } |