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_mount.h" |
Nathan Scott | a844f45 | 2005-11-02 14:38:42 +1100 | [diff] [blame] | 28 | #include "xfs_buf_item.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 29 | #include "xfs_trans_priv.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 30 | #include "xfs_error.h" |
Christoph Hellwig | 0b1b213 | 2009-12-14 23:14:59 +0000 | [diff] [blame] | 31 | #include "xfs_trace.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 32 | |
| 33 | |
| 34 | kmem_zone_t *xfs_buf_item_zone; |
| 35 | |
Christoph Hellwig | 7bfa31d | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 36 | static inline struct xfs_buf_log_item *BUF_ITEM(struct xfs_log_item *lip) |
| 37 | { |
| 38 | return container_of(lip, struct xfs_buf_log_item, bli_item); |
| 39 | } |
| 40 | |
| 41 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 42 | #ifdef XFS_TRANS_DEBUG |
| 43 | /* |
| 44 | * This function uses an alternate strategy for tracking the bytes |
| 45 | * that the user requests to be logged. This can then be used |
| 46 | * in conjunction with the bli_orig array in the buf log item to |
| 47 | * catch bugs in our callers' code. |
| 48 | * |
| 49 | * We also double check the bits set in xfs_buf_item_log using a |
| 50 | * simple algorithm to check that every byte is accounted for. |
| 51 | */ |
| 52 | STATIC void |
| 53 | xfs_buf_item_log_debug( |
| 54 | xfs_buf_log_item_t *bip, |
| 55 | uint first, |
| 56 | uint last) |
| 57 | { |
| 58 | uint x; |
| 59 | uint byte; |
| 60 | uint nbytes; |
| 61 | uint chunk_num; |
| 62 | uint word_num; |
| 63 | uint bit_num; |
| 64 | uint bit_set; |
| 65 | uint *wordp; |
| 66 | |
| 67 | ASSERT(bip->bli_logged != NULL); |
| 68 | byte = first; |
| 69 | nbytes = last - first + 1; |
| 70 | bfset(bip->bli_logged, first, nbytes); |
| 71 | for (x = 0; x < nbytes; x++) { |
Dave Chinner | c115541 | 2010-05-07 11:05:19 +1000 | [diff] [blame] | 72 | chunk_num = byte >> XFS_BLF_SHIFT; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 73 | word_num = chunk_num >> BIT_TO_WORD_SHIFT; |
| 74 | bit_num = chunk_num & (NBWORD - 1); |
| 75 | wordp = &(bip->bli_format.blf_data_map[word_num]); |
| 76 | bit_set = *wordp & (1 << bit_num); |
| 77 | ASSERT(bit_set); |
| 78 | byte++; |
| 79 | } |
| 80 | } |
| 81 | |
| 82 | /* |
| 83 | * This function is called when we flush something into a buffer without |
| 84 | * logging it. This happens for things like inodes which are logged |
| 85 | * separately from the buffer. |
| 86 | */ |
| 87 | void |
| 88 | xfs_buf_item_flush_log_debug( |
| 89 | xfs_buf_t *bp, |
| 90 | uint first, |
| 91 | uint last) |
| 92 | { |
Christoph Hellwig | adadbee | 2011-07-13 13:43:49 +0200 | [diff] [blame] | 93 | xfs_buf_log_item_t *bip = bp->b_fspriv; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 94 | uint nbytes; |
| 95 | |
Christoph Hellwig | adadbee | 2011-07-13 13:43:49 +0200 | [diff] [blame] | 96 | if (bip == NULL || (bip->bli_item.li_type != XFS_LI_BUF)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 97 | return; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 98 | |
| 99 | ASSERT(bip->bli_logged != NULL); |
| 100 | nbytes = last - first + 1; |
| 101 | bfset(bip->bli_logged, first, nbytes); |
| 102 | } |
| 103 | |
| 104 | /* |
Nathan Scott | c41564b | 2006-03-29 08:55:14 +1000 | [diff] [blame] | 105 | * This function is called to verify that our callers have logged |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 106 | * all the bytes that they changed. |
| 107 | * |
| 108 | * It does this by comparing the original copy of the buffer stored in |
| 109 | * 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] | 110 | * 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] | 111 | * array of the buf log item. |
| 112 | */ |
| 113 | STATIC void |
| 114 | xfs_buf_item_log_check( |
| 115 | xfs_buf_log_item_t *bip) |
| 116 | { |
| 117 | char *orig; |
| 118 | char *buffer; |
| 119 | int x; |
| 120 | xfs_buf_t *bp; |
| 121 | |
| 122 | ASSERT(bip->bli_orig != NULL); |
| 123 | ASSERT(bip->bli_logged != NULL); |
| 124 | |
| 125 | bp = bip->bli_buf; |
| 126 | ASSERT(XFS_BUF_COUNT(bp) > 0); |
Chandra Seetharaman | 6292604 | 2011-07-22 23:40:15 +0000 | [diff] [blame] | 127 | ASSERT(bp->b_addr != NULL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 128 | orig = bip->bli_orig; |
Chandra Seetharaman | 6292604 | 2011-07-22 23:40:15 +0000 | [diff] [blame] | 129 | buffer = bp->b_addr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 130 | for (x = 0; x < XFS_BUF_COUNT(bp); x++) { |
Dave Chinner | 0b932cc | 2011-03-07 10:08:35 +1100 | [diff] [blame] | 131 | if (orig[x] != buffer[x] && !btst(bip->bli_logged, x)) { |
| 132 | xfs_emerg(bp->b_mount, |
| 133 | "%s: bip %x buffer %x orig %x index %d", |
| 134 | __func__, bip, bp, orig, x); |
| 135 | ASSERT(0); |
| 136 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 137 | } |
| 138 | } |
| 139 | #else |
| 140 | #define xfs_buf_item_log_debug(x,y,z) |
| 141 | #define xfs_buf_item_log_check(x) |
| 142 | #endif |
| 143 | |
Dave Chinner | c90821a | 2010-12-03 17:00:52 +1100 | [diff] [blame] | 144 | STATIC void xfs_buf_do_callbacks(struct xfs_buf *bp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 145 | |
| 146 | /* |
| 147 | * This returns the number of log iovecs needed to log the |
| 148 | * given buf log item. |
| 149 | * |
| 150 | * It calculates this as 1 iovec for the buf log format structure |
| 151 | * and 1 for each stretch of non-contiguous chunks to be logged. |
| 152 | * Contiguous chunks are logged in a single iovec. |
| 153 | * |
| 154 | * If the XFS_BLI_STALE flag has been set, then log nothing. |
| 155 | */ |
Christoph Hellwig | ba0f32d | 2005-06-21 15:36:52 +1000 | [diff] [blame] | 156 | STATIC uint |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 157 | xfs_buf_item_size( |
Christoph Hellwig | 7bfa31d | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 158 | struct xfs_log_item *lip) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 159 | { |
Christoph Hellwig | 7bfa31d | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 160 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); |
| 161 | struct xfs_buf *bp = bip->bli_buf; |
| 162 | uint nvecs; |
| 163 | int next_bit; |
| 164 | int last_bit; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 165 | |
| 166 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
| 167 | if (bip->bli_flags & XFS_BLI_STALE) { |
| 168 | /* |
| 169 | * The buffer is stale, so all we need to log |
| 170 | * is the buf log format structure with the |
| 171 | * cancel flag in it. |
| 172 | */ |
Christoph Hellwig | 0b1b213 | 2009-12-14 23:14:59 +0000 | [diff] [blame] | 173 | trace_xfs_buf_item_size_stale(bip); |
Dave Chinner | c115541 | 2010-05-07 11:05:19 +1000 | [diff] [blame] | 174 | ASSERT(bip->bli_format.blf_flags & XFS_BLF_CANCEL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 175 | return 1; |
| 176 | } |
| 177 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 178 | ASSERT(bip->bli_flags & XFS_BLI_LOGGED); |
| 179 | nvecs = 1; |
| 180 | last_bit = xfs_next_bit(bip->bli_format.blf_data_map, |
| 181 | bip->bli_format.blf_map_size, 0); |
| 182 | ASSERT(last_bit != -1); |
| 183 | nvecs++; |
| 184 | while (last_bit != -1) { |
| 185 | /* |
| 186 | * This takes the bit number to start looking from and |
| 187 | * returns the next set bit from there. It returns -1 |
| 188 | * if there are no more bits set or the start bit is |
| 189 | * beyond the end of the bitmap. |
| 190 | */ |
| 191 | next_bit = xfs_next_bit(bip->bli_format.blf_data_map, |
| 192 | bip->bli_format.blf_map_size, |
| 193 | last_bit + 1); |
| 194 | /* |
| 195 | * If we run out of bits, leave the loop, |
| 196 | * else if we find a new set of bits bump the number of vecs, |
| 197 | * else keep scanning the current set of bits. |
| 198 | */ |
| 199 | if (next_bit == -1) { |
| 200 | last_bit = -1; |
| 201 | } else if (next_bit != last_bit + 1) { |
| 202 | last_bit = next_bit; |
| 203 | nvecs++; |
Dave Chinner | c115541 | 2010-05-07 11:05:19 +1000 | [diff] [blame] | 204 | } else if (xfs_buf_offset(bp, next_bit * XFS_BLF_CHUNK) != |
| 205 | (xfs_buf_offset(bp, last_bit * XFS_BLF_CHUNK) + |
| 206 | XFS_BLF_CHUNK)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 207 | last_bit = next_bit; |
| 208 | nvecs++; |
| 209 | } else { |
| 210 | last_bit++; |
| 211 | } |
| 212 | } |
| 213 | |
Christoph Hellwig | 0b1b213 | 2009-12-14 23:14:59 +0000 | [diff] [blame] | 214 | trace_xfs_buf_item_size(bip); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 215 | return nvecs; |
| 216 | } |
| 217 | |
| 218 | /* |
| 219 | * This is called to fill in the vector of log iovecs for the |
| 220 | * given log buf item. It fills the first entry with a buf log |
| 221 | * format structure, and the rest point to contiguous chunks |
| 222 | * within the buffer. |
| 223 | */ |
Christoph Hellwig | ba0f32d | 2005-06-21 15:36:52 +1000 | [diff] [blame] | 224 | STATIC void |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 225 | xfs_buf_item_format( |
Christoph Hellwig | 7bfa31d | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 226 | struct xfs_log_item *lip, |
| 227 | struct xfs_log_iovec *vecp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 228 | { |
Christoph Hellwig | 7bfa31d | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 229 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); |
| 230 | struct xfs_buf *bp = bip->bli_buf; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 231 | uint base_size; |
| 232 | uint nvecs; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 233 | int first_bit; |
| 234 | int last_bit; |
| 235 | int next_bit; |
| 236 | uint nbits; |
| 237 | uint buffer_offset; |
| 238 | |
| 239 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
| 240 | ASSERT((bip->bli_flags & XFS_BLI_LOGGED) || |
| 241 | (bip->bli_flags & XFS_BLI_STALE)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 242 | |
| 243 | /* |
| 244 | * The size of the base structure is the size of the |
| 245 | * declared structure plus the space for the extra words |
| 246 | * of the bitmap. We subtract one from the map size, because |
| 247 | * the first element of the bitmap is accounted for in the |
| 248 | * size of the base structure. |
| 249 | */ |
| 250 | base_size = |
| 251 | (uint)(sizeof(xfs_buf_log_format_t) + |
| 252 | ((bip->bli_format.blf_map_size - 1) * sizeof(uint))); |
Christoph Hellwig | 4e0d5f9 | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 253 | vecp->i_addr = &bip->bli_format; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 254 | vecp->i_len = base_size; |
Christoph Hellwig | 4139b3b | 2010-01-19 09:56:45 +0000 | [diff] [blame] | 255 | vecp->i_type = XLOG_REG_TYPE_BFORMAT; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 256 | vecp++; |
| 257 | nvecs = 1; |
| 258 | |
Dave Chinner | ccf7c23 | 2010-05-20 23:19:42 +1000 | [diff] [blame] | 259 | /* |
| 260 | * If it is an inode buffer, transfer the in-memory state to the |
| 261 | * format flags and clear the in-memory state. We do not transfer |
| 262 | * this state if the inode buffer allocation has not yet been committed |
| 263 | * to the log as setting the XFS_BLI_INODE_BUF flag will prevent |
| 264 | * correct replay of the inode allocation. |
| 265 | */ |
| 266 | if (bip->bli_flags & XFS_BLI_INODE_BUF) { |
| 267 | if (!((bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF) && |
Christoph Hellwig | 7bfa31d | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 268 | xfs_log_item_in_current_chkpt(lip))) |
Dave Chinner | ccf7c23 | 2010-05-20 23:19:42 +1000 | [diff] [blame] | 269 | bip->bli_format.blf_flags |= XFS_BLF_INODE_BUF; |
| 270 | bip->bli_flags &= ~XFS_BLI_INODE_BUF; |
| 271 | } |
| 272 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 273 | if (bip->bli_flags & XFS_BLI_STALE) { |
| 274 | /* |
| 275 | * The buffer is stale, so all we need to log |
| 276 | * is the buf log format structure with the |
| 277 | * cancel flag in it. |
| 278 | */ |
Christoph Hellwig | 0b1b213 | 2009-12-14 23:14:59 +0000 | [diff] [blame] | 279 | trace_xfs_buf_item_format_stale(bip); |
Dave Chinner | c115541 | 2010-05-07 11:05:19 +1000 | [diff] [blame] | 280 | ASSERT(bip->bli_format.blf_flags & XFS_BLF_CANCEL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 281 | bip->bli_format.blf_size = nvecs; |
| 282 | return; |
| 283 | } |
| 284 | |
| 285 | /* |
| 286 | * Fill in an iovec for each set of contiguous chunks. |
| 287 | */ |
| 288 | first_bit = xfs_next_bit(bip->bli_format.blf_data_map, |
| 289 | bip->bli_format.blf_map_size, 0); |
| 290 | ASSERT(first_bit != -1); |
| 291 | last_bit = first_bit; |
| 292 | nbits = 1; |
| 293 | for (;;) { |
| 294 | /* |
| 295 | * This takes the bit number to start looking from and |
| 296 | * returns the next set bit from there. It returns -1 |
| 297 | * if there are no more bits set or the start bit is |
| 298 | * beyond the end of the bitmap. |
| 299 | */ |
| 300 | next_bit = xfs_next_bit(bip->bli_format.blf_data_map, |
| 301 | bip->bli_format.blf_map_size, |
| 302 | (uint)last_bit + 1); |
| 303 | /* |
| 304 | * If we run out of bits fill in the last iovec and get |
| 305 | * out of the loop. |
| 306 | * Else if we start a new set of bits then fill in the |
| 307 | * iovec for the series we were looking at and start |
| 308 | * counting the bits in the new one. |
| 309 | * Else we're still in the same set of bits so just |
| 310 | * keep counting and scanning. |
| 311 | */ |
| 312 | if (next_bit == -1) { |
Dave Chinner | c115541 | 2010-05-07 11:05:19 +1000 | [diff] [blame] | 313 | buffer_offset = first_bit * XFS_BLF_CHUNK; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 314 | vecp->i_addr = xfs_buf_offset(bp, buffer_offset); |
Dave Chinner | c115541 | 2010-05-07 11:05:19 +1000 | [diff] [blame] | 315 | vecp->i_len = nbits * XFS_BLF_CHUNK; |
Christoph Hellwig | 4139b3b | 2010-01-19 09:56:45 +0000 | [diff] [blame] | 316 | vecp->i_type = XLOG_REG_TYPE_BCHUNK; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 317 | nvecs++; |
| 318 | break; |
| 319 | } else if (next_bit != last_bit + 1) { |
Dave Chinner | c115541 | 2010-05-07 11:05:19 +1000 | [diff] [blame] | 320 | buffer_offset = first_bit * XFS_BLF_CHUNK; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 321 | vecp->i_addr = xfs_buf_offset(bp, buffer_offset); |
Dave Chinner | c115541 | 2010-05-07 11:05:19 +1000 | [diff] [blame] | 322 | vecp->i_len = nbits * XFS_BLF_CHUNK; |
Christoph Hellwig | 4139b3b | 2010-01-19 09:56:45 +0000 | [diff] [blame] | 323 | vecp->i_type = XLOG_REG_TYPE_BCHUNK; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 324 | nvecs++; |
| 325 | vecp++; |
| 326 | first_bit = next_bit; |
| 327 | last_bit = next_bit; |
| 328 | nbits = 1; |
Dave Chinner | c115541 | 2010-05-07 11:05:19 +1000 | [diff] [blame] | 329 | } else if (xfs_buf_offset(bp, next_bit << XFS_BLF_SHIFT) != |
| 330 | (xfs_buf_offset(bp, last_bit << XFS_BLF_SHIFT) + |
| 331 | XFS_BLF_CHUNK)) { |
| 332 | buffer_offset = first_bit * XFS_BLF_CHUNK; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 333 | vecp->i_addr = xfs_buf_offset(bp, buffer_offset); |
Dave Chinner | c115541 | 2010-05-07 11:05:19 +1000 | [diff] [blame] | 334 | vecp->i_len = nbits * XFS_BLF_CHUNK; |
Christoph Hellwig | 4139b3b | 2010-01-19 09:56:45 +0000 | [diff] [blame] | 335 | vecp->i_type = XLOG_REG_TYPE_BCHUNK; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 336 | /* You would think we need to bump the nvecs here too, but we do not |
| 337 | * this number is used by recovery, and it gets confused by the boundary |
| 338 | * split here |
| 339 | * nvecs++; |
| 340 | */ |
| 341 | vecp++; |
| 342 | first_bit = next_bit; |
| 343 | last_bit = next_bit; |
| 344 | nbits = 1; |
| 345 | } else { |
| 346 | last_bit++; |
| 347 | nbits++; |
| 348 | } |
| 349 | } |
| 350 | bip->bli_format.blf_size = nvecs; |
| 351 | |
| 352 | /* |
| 353 | * Check to make sure everything is consistent. |
| 354 | */ |
Christoph Hellwig | 0b1b213 | 2009-12-14 23:14:59 +0000 | [diff] [blame] | 355 | trace_xfs_buf_item_format(bip); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 356 | xfs_buf_item_log_check(bip); |
| 357 | } |
| 358 | |
| 359 | /* |
Dave Chinner | 64fc35d | 2010-05-07 11:04:34 +1000 | [diff] [blame] | 360 | * This is called to pin the buffer associated with the buf log item in memory |
Christoph Hellwig | 4d16e92 | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 361 | * so it cannot be written out. |
Dave Chinner | 64fc35d | 2010-05-07 11:04:34 +1000 | [diff] [blame] | 362 | * |
| 363 | * We also always take a reference to the buffer log item here so that the bli |
| 364 | * is held while the item is pinned in memory. This means that we can |
| 365 | * unconditionally drop the reference count a transaction holds when the |
| 366 | * transaction is completed. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 367 | */ |
Christoph Hellwig | ba0f32d | 2005-06-21 15:36:52 +1000 | [diff] [blame] | 368 | STATIC void |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 369 | xfs_buf_item_pin( |
Christoph Hellwig | 7bfa31d | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 370 | struct xfs_log_item *lip) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 371 | { |
Christoph Hellwig | 7bfa31d | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 372 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 373 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 374 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
| 375 | ASSERT((bip->bli_flags & XFS_BLI_LOGGED) || |
| 376 | (bip->bli_flags & XFS_BLI_STALE)); |
Christoph Hellwig | 7bfa31d | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 377 | |
Christoph Hellwig | 0b1b213 | 2009-12-14 23:14:59 +0000 | [diff] [blame] | 378 | trace_xfs_buf_item_pin(bip); |
Christoph Hellwig | 4d16e92 | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 379 | |
| 380 | atomic_inc(&bip->bli_refcount); |
| 381 | atomic_inc(&bip->bli_buf->b_pin_count); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 382 | } |
| 383 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 384 | /* |
| 385 | * This is called to unpin the buffer associated with the buf log |
| 386 | * item which was previously pinned with a call to xfs_buf_item_pin(). |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 387 | * |
| 388 | * Also drop the reference to the buf item for the current transaction. |
| 389 | * If the XFS_BLI_STALE flag is set and we are the last reference, |
| 390 | * then free up the buf log item and unlock the buffer. |
Christoph Hellwig | 9412e31 | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 391 | * |
| 392 | * If the remove flag is set we are called from uncommit in the |
| 393 | * forced-shutdown path. If that is true and the reference count on |
| 394 | * the log item is going to drop to zero we need to free the item's |
| 395 | * descriptor in the transaction. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 396 | */ |
Christoph Hellwig | ba0f32d | 2005-06-21 15:36:52 +1000 | [diff] [blame] | 397 | STATIC void |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 398 | xfs_buf_item_unpin( |
Christoph Hellwig | 7bfa31d | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 399 | struct xfs_log_item *lip, |
Christoph Hellwig | 9412e31 | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 400 | int remove) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 401 | { |
Christoph Hellwig | 7bfa31d | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 402 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); |
Christoph Hellwig | 9412e31 | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 403 | xfs_buf_t *bp = bip->bli_buf; |
Christoph Hellwig | 7bfa31d | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 404 | struct xfs_ail *ailp = lip->li_ailp; |
Dave Chinner | 8e12385 | 2010-03-08 11:26:03 +1100 | [diff] [blame] | 405 | int stale = bip->bli_flags & XFS_BLI_STALE; |
Christoph Hellwig | 7bfa31d | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 406 | int freed; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 407 | |
Christoph Hellwig | adadbee | 2011-07-13 13:43:49 +0200 | [diff] [blame] | 408 | ASSERT(bp->b_fspriv == bip); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 409 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
Christoph Hellwig | 9412e31 | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 410 | |
Christoph Hellwig | 0b1b213 | 2009-12-14 23:14:59 +0000 | [diff] [blame] | 411 | trace_xfs_buf_item_unpin(bip); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 412 | |
| 413 | freed = atomic_dec_and_test(&bip->bli_refcount); |
Christoph Hellwig | 4d16e92 | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 414 | |
| 415 | if (atomic_dec_and_test(&bp->b_pin_count)) |
| 416 | wake_up_all(&bp->b_waiters); |
Christoph Hellwig | 7bfa31d | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 417 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 418 | if (freed && stale) { |
| 419 | ASSERT(bip->bli_flags & XFS_BLI_STALE); |
Christoph Hellwig | 0c842ad | 2011-07-08 14:36:19 +0200 | [diff] [blame] | 420 | ASSERT(xfs_buf_islocked(bp)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 421 | ASSERT(!(XFS_BUF_ISDELAYWRITE(bp))); |
| 422 | ASSERT(XFS_BUF_ISSTALE(bp)); |
Dave Chinner | c115541 | 2010-05-07 11:05:19 +1000 | [diff] [blame] | 423 | ASSERT(bip->bli_format.blf_flags & XFS_BLF_CANCEL); |
Christoph Hellwig | 9412e31 | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 424 | |
Christoph Hellwig | 0b1b213 | 2009-12-14 23:14:59 +0000 | [diff] [blame] | 425 | trace_xfs_buf_item_unpin_stale(bip); |
| 426 | |
Christoph Hellwig | 9412e31 | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 427 | if (remove) { |
| 428 | /* |
Dave Chinner | e34a314 | 2011-01-27 12:13:35 +1100 | [diff] [blame] | 429 | * If we are in a transaction context, we have to |
| 430 | * remove the log item from the transaction as we are |
| 431 | * about to release our reference to the buffer. If we |
| 432 | * don't, the unlock that occurs later in |
| 433 | * xfs_trans_uncommit() will try to reference the |
Christoph Hellwig | 9412e31 | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 434 | * buffer which we no longer have a hold on. |
| 435 | */ |
Dave Chinner | e34a314 | 2011-01-27 12:13:35 +1100 | [diff] [blame] | 436 | if (lip->li_desc) |
| 437 | xfs_trans_del_item(lip); |
Christoph Hellwig | 9412e31 | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 438 | |
| 439 | /* |
| 440 | * Since the transaction no longer refers to the buffer, |
| 441 | * the buffer should no longer refer to the transaction. |
| 442 | */ |
Christoph Hellwig | bf9d901 | 2011-07-13 13:43:49 +0200 | [diff] [blame] | 443 | bp->b_transp = NULL; |
Christoph Hellwig | 9412e31 | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 444 | } |
| 445 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 446 | /* |
| 447 | * If we get called here because of an IO error, we may |
David Chinner | 783a2f6 | 2008-10-30 17:39:58 +1100 | [diff] [blame] | 448 | * 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] | 449 | * will take care of that situation. |
David Chinner | 783a2f6 | 2008-10-30 17:39:58 +1100 | [diff] [blame] | 450 | * xfs_trans_ail_delete() drops the AIL lock. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 451 | */ |
| 452 | if (bip->bli_flags & XFS_BLI_STALE_INODE) { |
Dave Chinner | c90821a | 2010-12-03 17:00:52 +1100 | [diff] [blame] | 453 | xfs_buf_do_callbacks(bp); |
Christoph Hellwig | adadbee | 2011-07-13 13:43:49 +0200 | [diff] [blame] | 454 | bp->b_fspriv = NULL; |
Christoph Hellwig | cb669ca | 2011-07-13 13:43:49 +0200 | [diff] [blame] | 455 | bp->b_iodone = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 456 | } else { |
David Chinner | 783a2f6 | 2008-10-30 17:39:58 +1100 | [diff] [blame] | 457 | spin_lock(&ailp->xa_lock); |
| 458 | xfs_trans_ail_delete(ailp, (xfs_log_item_t *)bip); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 459 | xfs_buf_item_relse(bp); |
Christoph Hellwig | adadbee | 2011-07-13 13:43:49 +0200 | [diff] [blame] | 460 | ASSERT(bp->b_fspriv == NULL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 461 | } |
| 462 | xfs_buf_relse(bp); |
| 463 | } |
| 464 | } |
| 465 | |
| 466 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 467 | * This is called to attempt to lock the buffer associated with this |
| 468 | * 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] | 469 | * the lock right away, return 0. If we can get the lock, take a |
| 470 | * reference to the buffer. If this is a delayed write buffer that |
| 471 | * needs AIL help to be written back, invoke the pushbuf routine |
| 472 | * rather than the normal success path. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 473 | */ |
Christoph Hellwig | ba0f32d | 2005-06-21 15:36:52 +1000 | [diff] [blame] | 474 | STATIC uint |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 475 | xfs_buf_item_trylock( |
Christoph Hellwig | 7bfa31d | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 476 | struct xfs_log_item *lip) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 477 | { |
Christoph Hellwig | 7bfa31d | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 478 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); |
| 479 | struct xfs_buf *bp = bip->bli_buf; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 480 | |
Chandra Seetharaman | 811e64c | 2011-07-22 23:40:27 +0000 | [diff] [blame] | 481 | if (xfs_buf_ispinned(bp)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 482 | return XFS_ITEM_PINNED; |
Christoph Hellwig | 0c842ad | 2011-07-08 14:36:19 +0200 | [diff] [blame] | 483 | if (!xfs_buf_trylock(bp)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 484 | return XFS_ITEM_LOCKED; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 485 | |
Dave Chinner | d808f61 | 2010-02-02 10:13:42 +1100 | [diff] [blame] | 486 | /* take a reference to the buffer. */ |
Chandra Seetharaman | 72790aa | 2011-07-22 23:40:04 +0000 | [diff] [blame] | 487 | xfs_buf_hold(bp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 488 | |
| 489 | ASSERT(!(bip->bli_flags & XFS_BLI_STALE)); |
Christoph Hellwig | 0b1b213 | 2009-12-14 23:14:59 +0000 | [diff] [blame] | 490 | trace_xfs_buf_item_trylock(bip); |
Dave Chinner | d808f61 | 2010-02-02 10:13:42 +1100 | [diff] [blame] | 491 | if (XFS_BUF_ISDELAYWRITE(bp)) |
| 492 | return XFS_ITEM_PUSHBUF; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 493 | return XFS_ITEM_SUCCESS; |
| 494 | } |
| 495 | |
| 496 | /* |
Dave Chinner | 64fc35d | 2010-05-07 11:04:34 +1000 | [diff] [blame] | 497 | * Release the buffer associated with the buf log item. If there is no dirty |
| 498 | * logged data associated with the buffer recorded in the buf log item, then |
| 499 | * free the buf log item and remove the reference to it in the buffer. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 500 | * |
Dave Chinner | 64fc35d | 2010-05-07 11:04:34 +1000 | [diff] [blame] | 501 | * This call ignores the recursion count. It is only called when the buffer |
| 502 | * should REALLY be unlocked, regardless of the recursion count. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 503 | * |
Dave Chinner | 64fc35d | 2010-05-07 11:04:34 +1000 | [diff] [blame] | 504 | * We unconditionally drop the transaction's reference to the log item. If the |
| 505 | * item was logged, then another reference was taken when it was pinned, so we |
| 506 | * can safely drop the transaction reference now. This also allows us to avoid |
| 507 | * potential races with the unpin code freeing the bli by not referencing the |
| 508 | * bli after we've dropped the reference count. |
| 509 | * |
| 510 | * If the XFS_BLI_HOLD flag is set in the buf log item, then free the log item |
| 511 | * if necessary but do not unlock the buffer. This is for support of |
| 512 | * xfs_trans_bhold(). Make sure the XFS_BLI_HOLD field is cleared if we don't |
| 513 | * free the item. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 514 | */ |
Christoph Hellwig | ba0f32d | 2005-06-21 15:36:52 +1000 | [diff] [blame] | 515 | STATIC void |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 516 | xfs_buf_item_unlock( |
Christoph Hellwig | 7bfa31d | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 517 | struct xfs_log_item *lip) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 518 | { |
Christoph Hellwig | 7bfa31d | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 519 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); |
| 520 | struct xfs_buf *bp = bip->bli_buf; |
| 521 | int aborted; |
| 522 | uint hold; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 523 | |
Dave Chinner | 64fc35d | 2010-05-07 11:04:34 +1000 | [diff] [blame] | 524 | /* Clear the buffer's association with this transaction. */ |
Christoph Hellwig | bf9d901 | 2011-07-13 13:43:49 +0200 | [diff] [blame] | 525 | bp->b_transp = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 526 | |
| 527 | /* |
Dave Chinner | 64fc35d | 2010-05-07 11:04:34 +1000 | [diff] [blame] | 528 | * If this is a transaction abort, don't return early. Instead, allow |
| 529 | * the brelse to happen. Normally it would be done for stale |
| 530 | * (cancelled) buffers at unpin time, but we'll never go through the |
| 531 | * pin/unpin cycle if we abort inside commit. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 532 | */ |
Christoph Hellwig | 7bfa31d | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 533 | aborted = (lip->li_flags & XFS_LI_ABORTED) != 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 534 | |
| 535 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 536 | * Before possibly freeing the buf item, determine if we should |
| 537 | * release the buffer at the end of this routine. |
| 538 | */ |
| 539 | hold = bip->bli_flags & XFS_BLI_HOLD; |
Dave Chinner | 64fc35d | 2010-05-07 11:04:34 +1000 | [diff] [blame] | 540 | |
| 541 | /* Clear the per transaction state. */ |
| 542 | bip->bli_flags &= ~(XFS_BLI_LOGGED | XFS_BLI_HOLD); |
| 543 | |
| 544 | /* |
| 545 | * If the buf item is marked stale, then don't do anything. We'll |
| 546 | * unlock the buffer and free the buf item when the buffer is unpinned |
| 547 | * for the last time. |
| 548 | */ |
| 549 | if (bip->bli_flags & XFS_BLI_STALE) { |
| 550 | trace_xfs_buf_item_unlock_stale(bip); |
Dave Chinner | c115541 | 2010-05-07 11:05:19 +1000 | [diff] [blame] | 551 | ASSERT(bip->bli_format.blf_flags & XFS_BLF_CANCEL); |
Dave Chinner | 64fc35d | 2010-05-07 11:04:34 +1000 | [diff] [blame] | 552 | if (!aborted) { |
| 553 | atomic_dec(&bip->bli_refcount); |
| 554 | return; |
| 555 | } |
| 556 | } |
| 557 | |
Christoph Hellwig | 0b1b213 | 2009-12-14 23:14:59 +0000 | [diff] [blame] | 558 | trace_xfs_buf_item_unlock(bip); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 559 | |
| 560 | /* |
Dave Chinner | 64fc35d | 2010-05-07 11:04:34 +1000 | [diff] [blame] | 561 | * If the buf item isn't tracking any data, free it, otherwise drop the |
| 562 | * reference we hold to it. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 563 | */ |
Eric Sandeen | 24ad33f | 2007-06-28 16:43:30 +1000 | [diff] [blame] | 564 | if (xfs_bitmap_empty(bip->bli_format.blf_data_map, |
Dave Chinner | 64fc35d | 2010-05-07 11:04:34 +1000 | [diff] [blame] | 565 | bip->bli_format.blf_map_size)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 566 | xfs_buf_item_relse(bp); |
Dave Chinner | 64fc35d | 2010-05-07 11:04:34 +1000 | [diff] [blame] | 567 | else |
| 568 | atomic_dec(&bip->bli_refcount); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 569 | |
Dave Chinner | 64fc35d | 2010-05-07 11:04:34 +1000 | [diff] [blame] | 570 | if (!hold) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 571 | xfs_buf_relse(bp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 572 | } |
| 573 | |
| 574 | /* |
| 575 | * This is called to find out where the oldest active copy of the |
| 576 | * buf log item in the on disk log resides now that the last log |
| 577 | * write of it completed at the given lsn. |
| 578 | * We always re-log all the dirty data in a buffer, so usually the |
| 579 | * latest copy in the on disk log is the only one that matters. For |
| 580 | * those cases we simply return the given lsn. |
| 581 | * |
| 582 | * The one exception to this is for buffers full of newly allocated |
| 583 | * inodes. These buffers are only relogged with the XFS_BLI_INODE_BUF |
| 584 | * flag set, indicating that only the di_next_unlinked fields from the |
| 585 | * inodes in the buffers will be replayed during recovery. If the |
| 586 | * original newly allocated inode images have not yet been flushed |
| 587 | * when the buffer is so relogged, then we need to make sure that we |
| 588 | * keep the old images in the 'active' portion of the log. We do this |
| 589 | * by returning the original lsn of that transaction here rather than |
| 590 | * the current one. |
| 591 | */ |
Christoph Hellwig | ba0f32d | 2005-06-21 15:36:52 +1000 | [diff] [blame] | 592 | STATIC xfs_lsn_t |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 593 | xfs_buf_item_committed( |
Christoph Hellwig | 7bfa31d | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 594 | struct xfs_log_item *lip, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 595 | xfs_lsn_t lsn) |
| 596 | { |
Christoph Hellwig | 7bfa31d | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 597 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); |
| 598 | |
Christoph Hellwig | 0b1b213 | 2009-12-14 23:14:59 +0000 | [diff] [blame] | 599 | trace_xfs_buf_item_committed(bip); |
| 600 | |
Christoph Hellwig | 7bfa31d | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 601 | if ((bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF) && lip->li_lsn != 0) |
| 602 | return lip->li_lsn; |
| 603 | return lsn; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 604 | } |
| 605 | |
| 606 | /* |
Dave Chinner | d808f61 | 2010-02-02 10:13:42 +1100 | [diff] [blame] | 607 | * The buffer is locked, but is not a delayed write buffer. This happens |
| 608 | * if we race with IO completion and hence we don't want to try to write it |
| 609 | * again. Just release the buffer. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 610 | */ |
Christoph Hellwig | ba0f32d | 2005-06-21 15:36:52 +1000 | [diff] [blame] | 611 | STATIC void |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 612 | xfs_buf_item_push( |
Christoph Hellwig | 7bfa31d | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 613 | struct xfs_log_item *lip) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 614 | { |
Christoph Hellwig | 7bfa31d | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 615 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); |
| 616 | struct xfs_buf *bp = bip->bli_buf; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 617 | |
| 618 | ASSERT(!(bip->bli_flags & XFS_BLI_STALE)); |
Christoph Hellwig | 7bfa31d | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 619 | ASSERT(!XFS_BUF_ISDELAYWRITE(bp)); |
| 620 | |
Christoph Hellwig | 0b1b213 | 2009-12-14 23:14:59 +0000 | [diff] [blame] | 621 | trace_xfs_buf_item_push(bip); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 622 | |
Dave Chinner | d808f61 | 2010-02-02 10:13:42 +1100 | [diff] [blame] | 623 | xfs_buf_relse(bp); |
| 624 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 625 | |
Dave Chinner | d808f61 | 2010-02-02 10:13:42 +1100 | [diff] [blame] | 626 | /* |
| 627 | * The buffer is locked and is a delayed write buffer. Promote the buffer |
| 628 | * in the delayed write queue as the caller knows that they must invoke |
| 629 | * the xfsbufd to get this buffer written. We have to unlock the buffer |
| 630 | * to allow the xfsbufd to write it, too. |
| 631 | */ |
| 632 | STATIC void |
| 633 | xfs_buf_item_pushbuf( |
Christoph Hellwig | 7bfa31d | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 634 | struct xfs_log_item *lip) |
Dave Chinner | d808f61 | 2010-02-02 10:13:42 +1100 | [diff] [blame] | 635 | { |
Christoph Hellwig | 7bfa31d | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 636 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); |
| 637 | struct xfs_buf *bp = bip->bli_buf; |
Dave Chinner | d808f61 | 2010-02-02 10:13:42 +1100 | [diff] [blame] | 638 | |
| 639 | ASSERT(!(bip->bli_flags & XFS_BLI_STALE)); |
Christoph Hellwig | 7bfa31d | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 640 | ASSERT(XFS_BUF_ISDELAYWRITE(bp)); |
| 641 | |
Dave Chinner | d808f61 | 2010-02-02 10:13:42 +1100 | [diff] [blame] | 642 | trace_xfs_buf_item_pushbuf(bip); |
| 643 | |
Dave Chinner | d808f61 | 2010-02-02 10:13:42 +1100 | [diff] [blame] | 644 | xfs_buf_delwri_promote(bp); |
| 645 | xfs_buf_relse(bp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 646 | } |
| 647 | |
Christoph Hellwig | ba0f32d | 2005-06-21 15:36:52 +1000 | [diff] [blame] | 648 | STATIC void |
Christoph Hellwig | 7bfa31d | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 649 | xfs_buf_item_committing( |
| 650 | struct xfs_log_item *lip, |
| 651 | xfs_lsn_t commit_lsn) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 652 | { |
| 653 | } |
| 654 | |
| 655 | /* |
| 656 | * This is the ops vector shared by all buf log items. |
| 657 | */ |
David Chinner | 7989cb8 | 2007-02-10 18:34:56 +1100 | [diff] [blame] | 658 | static struct xfs_item_ops xfs_buf_item_ops = { |
Christoph Hellwig | 7bfa31d | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 659 | .iop_size = xfs_buf_item_size, |
| 660 | .iop_format = xfs_buf_item_format, |
| 661 | .iop_pin = xfs_buf_item_pin, |
| 662 | .iop_unpin = xfs_buf_item_unpin, |
| 663 | .iop_trylock = xfs_buf_item_trylock, |
| 664 | .iop_unlock = xfs_buf_item_unlock, |
| 665 | .iop_committed = xfs_buf_item_committed, |
| 666 | .iop_push = xfs_buf_item_push, |
| 667 | .iop_pushbuf = xfs_buf_item_pushbuf, |
| 668 | .iop_committing = xfs_buf_item_committing |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 669 | }; |
| 670 | |
| 671 | |
| 672 | /* |
| 673 | * Allocate a new buf log item to go with the given buffer. |
| 674 | * Set the buffer's b_fsprivate field to point to the new |
| 675 | * buf log item. If there are other item's attached to the |
| 676 | * buffer (see xfs_buf_attach_iodone() below), then put the |
| 677 | * buf log item at the front. |
| 678 | */ |
| 679 | void |
| 680 | xfs_buf_item_init( |
| 681 | xfs_buf_t *bp, |
| 682 | xfs_mount_t *mp) |
| 683 | { |
Christoph Hellwig | adadbee | 2011-07-13 13:43:49 +0200 | [diff] [blame] | 684 | xfs_log_item_t *lip = bp->b_fspriv; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 685 | xfs_buf_log_item_t *bip; |
| 686 | int chunks; |
| 687 | int map_size; |
| 688 | |
| 689 | /* |
| 690 | * Check to see if there is already a buf log item for |
| 691 | * this buffer. If there is, it is guaranteed to be |
| 692 | * the first. If we do already have one, there is |
| 693 | * nothing to do here so return. |
| 694 | */ |
Dave Chinner | ebad861 | 2010-09-22 10:47:20 +1000 | [diff] [blame] | 695 | ASSERT(bp->b_target->bt_mount == mp); |
Christoph Hellwig | adadbee | 2011-07-13 13:43:49 +0200 | [diff] [blame] | 696 | if (lip != NULL && lip->li_type == XFS_LI_BUF) |
| 697 | return; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 698 | |
| 699 | /* |
Dave Chinner | c115541 | 2010-05-07 11:05:19 +1000 | [diff] [blame] | 700 | * chunks is the number of XFS_BLF_CHUNK size pieces |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 701 | * the buffer can be divided into. Make sure not to |
| 702 | * truncate any pieces. map_size is the size of the |
| 703 | * bitmap needed to describe the chunks of the buffer. |
| 704 | */ |
Dave Chinner | c115541 | 2010-05-07 11:05:19 +1000 | [diff] [blame] | 705 | chunks = (int)((XFS_BUF_COUNT(bp) + (XFS_BLF_CHUNK - 1)) >> XFS_BLF_SHIFT); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 706 | map_size = (int)((chunks + NBWORD) >> BIT_TO_WORD_SHIFT); |
| 707 | |
| 708 | bip = (xfs_buf_log_item_t*)kmem_zone_zalloc(xfs_buf_item_zone, |
| 709 | KM_SLEEP); |
Dave Chinner | 43f5efc | 2010-03-23 10:10:00 +1100 | [diff] [blame] | 710 | xfs_log_item_init(mp, &bip->bli_item, XFS_LI_BUF, &xfs_buf_item_ops); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 711 | bip->bli_buf = bp; |
Lachlan McIlroy | e1f5dbd | 2008-09-17 16:52:13 +1000 | [diff] [blame] | 712 | xfs_buf_hold(bp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 713 | bip->bli_format.blf_type = XFS_LI_BUF; |
| 714 | bip->bli_format.blf_blkno = (__int64_t)XFS_BUF_ADDR(bp); |
| 715 | bip->bli_format.blf_len = (ushort)BTOBB(XFS_BUF_COUNT(bp)); |
| 716 | bip->bli_format.blf_map_size = map_size; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 717 | |
| 718 | #ifdef XFS_TRANS_DEBUG |
| 719 | /* |
| 720 | * Allocate the arrays for tracking what needs to be logged |
| 721 | * and what our callers request to be logged. bli_orig |
| 722 | * holds a copy of the original, clean buffer for comparison |
| 723 | * against, and bli_logged keeps a 1 bit flag per byte in |
| 724 | * the buffer to indicate which bytes the callers have asked |
| 725 | * to have logged. |
| 726 | */ |
| 727 | bip->bli_orig = (char *)kmem_alloc(XFS_BUF_COUNT(bp), KM_SLEEP); |
Chandra Seetharaman | 6292604 | 2011-07-22 23:40:15 +0000 | [diff] [blame] | 728 | memcpy(bip->bli_orig, bp->b_addr, XFS_BUF_COUNT(bp)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 729 | bip->bli_logged = (char *)kmem_zalloc(XFS_BUF_COUNT(bp) / NBBY, KM_SLEEP); |
| 730 | #endif |
| 731 | |
| 732 | /* |
| 733 | * Put the buf item into the list of items attached to the |
| 734 | * buffer at the front. |
| 735 | */ |
Christoph Hellwig | adadbee | 2011-07-13 13:43:49 +0200 | [diff] [blame] | 736 | if (bp->b_fspriv) |
| 737 | bip->bli_item.li_bio_list = bp->b_fspriv; |
| 738 | bp->b_fspriv = bip; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 739 | } |
| 740 | |
| 741 | |
| 742 | /* |
| 743 | * Mark bytes first through last inclusive as dirty in the buf |
| 744 | * item's bitmap. |
| 745 | */ |
| 746 | void |
| 747 | xfs_buf_item_log( |
| 748 | xfs_buf_log_item_t *bip, |
| 749 | uint first, |
| 750 | uint last) |
| 751 | { |
| 752 | uint first_bit; |
| 753 | uint last_bit; |
| 754 | uint bits_to_set; |
| 755 | uint bits_set; |
| 756 | uint word_num; |
| 757 | uint *wordp; |
| 758 | uint bit; |
| 759 | uint end_bit; |
| 760 | uint mask; |
| 761 | |
| 762 | /* |
| 763 | * Mark the item as having some dirty data for |
| 764 | * quick reference in xfs_buf_item_dirty. |
| 765 | */ |
| 766 | bip->bli_flags |= XFS_BLI_DIRTY; |
| 767 | |
| 768 | /* |
| 769 | * Convert byte offsets to bit numbers. |
| 770 | */ |
Dave Chinner | c115541 | 2010-05-07 11:05:19 +1000 | [diff] [blame] | 771 | first_bit = first >> XFS_BLF_SHIFT; |
| 772 | last_bit = last >> XFS_BLF_SHIFT; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 773 | |
| 774 | /* |
| 775 | * Calculate the total number of bits to be set. |
| 776 | */ |
| 777 | bits_to_set = last_bit - first_bit + 1; |
| 778 | |
| 779 | /* |
| 780 | * Get a pointer to the first word in the bitmap |
| 781 | * to set a bit in. |
| 782 | */ |
| 783 | word_num = first_bit >> BIT_TO_WORD_SHIFT; |
| 784 | wordp = &(bip->bli_format.blf_data_map[word_num]); |
| 785 | |
| 786 | /* |
| 787 | * Calculate the starting bit in the first word. |
| 788 | */ |
| 789 | bit = first_bit & (uint)(NBWORD - 1); |
| 790 | |
| 791 | /* |
| 792 | * First set any bits in the first word of our range. |
| 793 | * If it starts at bit 0 of the word, it will be |
| 794 | * set below rather than here. That is what the variable |
| 795 | * bit tells us. The variable bits_set tracks the number |
| 796 | * of bits that have been set so far. End_bit is the number |
| 797 | * of the last bit to be set in this word plus one. |
| 798 | */ |
| 799 | if (bit) { |
| 800 | end_bit = MIN(bit + bits_to_set, (uint)NBWORD); |
| 801 | mask = ((1 << (end_bit - bit)) - 1) << bit; |
| 802 | *wordp |= mask; |
| 803 | wordp++; |
| 804 | bits_set = end_bit - bit; |
| 805 | } else { |
| 806 | bits_set = 0; |
| 807 | } |
| 808 | |
| 809 | /* |
| 810 | * Now set bits a whole word at a time that are between |
| 811 | * first_bit and last_bit. |
| 812 | */ |
| 813 | while ((bits_to_set - bits_set) >= NBWORD) { |
| 814 | *wordp |= 0xffffffff; |
| 815 | bits_set += NBWORD; |
| 816 | wordp++; |
| 817 | } |
| 818 | |
| 819 | /* |
| 820 | * Finally, set any bits left to be set in one last partial word. |
| 821 | */ |
| 822 | end_bit = bits_to_set - bits_set; |
| 823 | if (end_bit) { |
| 824 | mask = (1 << end_bit) - 1; |
| 825 | *wordp |= mask; |
| 826 | } |
| 827 | |
| 828 | xfs_buf_item_log_debug(bip, first, last); |
| 829 | } |
| 830 | |
| 831 | |
| 832 | /* |
| 833 | * Return 1 if the buffer has some data that has been logged (at any |
| 834 | * point, not just the current transaction) and 0 if not. |
| 835 | */ |
| 836 | uint |
| 837 | xfs_buf_item_dirty( |
| 838 | xfs_buf_log_item_t *bip) |
| 839 | { |
| 840 | return (bip->bli_flags & XFS_BLI_DIRTY); |
| 841 | } |
| 842 | |
Lachlan McIlroy | e1f5dbd | 2008-09-17 16:52:13 +1000 | [diff] [blame] | 843 | STATIC void |
| 844 | xfs_buf_item_free( |
| 845 | xfs_buf_log_item_t *bip) |
| 846 | { |
| 847 | #ifdef XFS_TRANS_DEBUG |
| 848 | kmem_free(bip->bli_orig); |
| 849 | kmem_free(bip->bli_logged); |
| 850 | #endif /* XFS_TRANS_DEBUG */ |
| 851 | |
Lachlan McIlroy | e1f5dbd | 2008-09-17 16:52:13 +1000 | [diff] [blame] | 852 | kmem_zone_free(xfs_buf_item_zone, bip); |
| 853 | } |
| 854 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 855 | /* |
| 856 | * This is called when the buf log item is no longer needed. It should |
| 857 | * free the buf log item associated with the given buffer and clear |
| 858 | * the buffer's pointer to the buf log item. If there are no more |
| 859 | * items in the list, clear the b_iodone field of the buffer (see |
| 860 | * xfs_buf_attach_iodone() below). |
| 861 | */ |
| 862 | void |
| 863 | xfs_buf_item_relse( |
| 864 | xfs_buf_t *bp) |
| 865 | { |
| 866 | xfs_buf_log_item_t *bip; |
| 867 | |
Christoph Hellwig | 0b1b213 | 2009-12-14 23:14:59 +0000 | [diff] [blame] | 868 | trace_xfs_buf_item_relse(bp, _RET_IP_); |
| 869 | |
Christoph Hellwig | adadbee | 2011-07-13 13:43:49 +0200 | [diff] [blame] | 870 | bip = bp->b_fspriv; |
| 871 | bp->b_fspriv = bip->bli_item.li_bio_list; |
Christoph Hellwig | cb669ca | 2011-07-13 13:43:49 +0200 | [diff] [blame] | 872 | if (bp->b_fspriv == NULL) |
| 873 | bp->b_iodone = NULL; |
Christoph Hellwig | adadbee | 2011-07-13 13:43:49 +0200 | [diff] [blame] | 874 | |
Lachlan McIlroy | e1f5dbd | 2008-09-17 16:52:13 +1000 | [diff] [blame] | 875 | xfs_buf_rele(bp); |
| 876 | xfs_buf_item_free(bip); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 877 | } |
| 878 | |
| 879 | |
| 880 | /* |
| 881 | * Add the given log item with its callback to the list of callbacks |
| 882 | * to be called when the buffer's I/O completes. If it is not set |
| 883 | * already, set the buffer's b_iodone() routine to be |
| 884 | * xfs_buf_iodone_callbacks() and link the log item into the list of |
| 885 | * items rooted at b_fsprivate. Items are always added as the second |
| 886 | * entry in the list if there is a first, because the buf item code |
| 887 | * assumes that the buf log item is first. |
| 888 | */ |
| 889 | void |
| 890 | xfs_buf_attach_iodone( |
| 891 | xfs_buf_t *bp, |
| 892 | void (*cb)(xfs_buf_t *, xfs_log_item_t *), |
| 893 | xfs_log_item_t *lip) |
| 894 | { |
| 895 | xfs_log_item_t *head_lip; |
| 896 | |
Christoph Hellwig | 0c842ad | 2011-07-08 14:36:19 +0200 | [diff] [blame] | 897 | ASSERT(xfs_buf_islocked(bp)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 898 | |
| 899 | lip->li_cb = cb; |
Christoph Hellwig | adadbee | 2011-07-13 13:43:49 +0200 | [diff] [blame] | 900 | head_lip = bp->b_fspriv; |
| 901 | if (head_lip) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 902 | lip->li_bio_list = head_lip->li_bio_list; |
| 903 | head_lip->li_bio_list = lip; |
| 904 | } else { |
Christoph Hellwig | adadbee | 2011-07-13 13:43:49 +0200 | [diff] [blame] | 905 | bp->b_fspriv = lip; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 906 | } |
| 907 | |
Christoph Hellwig | cb669ca | 2011-07-13 13:43:49 +0200 | [diff] [blame] | 908 | ASSERT(bp->b_iodone == NULL || |
| 909 | bp->b_iodone == xfs_buf_iodone_callbacks); |
| 910 | bp->b_iodone = xfs_buf_iodone_callbacks; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 911 | } |
| 912 | |
Dave Chinner | c90821a | 2010-12-03 17:00:52 +1100 | [diff] [blame] | 913 | /* |
| 914 | * We can have many callbacks on a buffer. Running the callbacks individually |
| 915 | * can cause a lot of contention on the AIL lock, so we allow for a single |
| 916 | * callback to be able to scan the remaining lip->li_bio_list for other items |
| 917 | * of the same type and callback to be processed in the first call. |
| 918 | * |
| 919 | * As a result, the loop walking the callback list below will also modify the |
| 920 | * list. it removes the first item from the list and then runs the callback. |
| 921 | * The loop then restarts from the new head of the list. This allows the |
| 922 | * callback to scan and modify the list attached to the buffer and we don't |
| 923 | * have to care about maintaining a next item pointer. |
| 924 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 925 | STATIC void |
| 926 | xfs_buf_do_callbacks( |
Dave Chinner | c90821a | 2010-12-03 17:00:52 +1100 | [diff] [blame] | 927 | struct xfs_buf *bp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 928 | { |
Dave Chinner | c90821a | 2010-12-03 17:00:52 +1100 | [diff] [blame] | 929 | struct xfs_log_item *lip; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 930 | |
Christoph Hellwig | adadbee | 2011-07-13 13:43:49 +0200 | [diff] [blame] | 931 | while ((lip = bp->b_fspriv) != NULL) { |
| 932 | bp->b_fspriv = lip->li_bio_list; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 933 | ASSERT(lip->li_cb != NULL); |
| 934 | /* |
| 935 | * Clear the next pointer so we don't have any |
| 936 | * confusion if the item is added to another buf. |
| 937 | * Don't touch the log item after calling its |
| 938 | * callback, because it could have freed itself. |
| 939 | */ |
| 940 | lip->li_bio_list = NULL; |
| 941 | lip->li_cb(bp, lip); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 942 | } |
| 943 | } |
| 944 | |
| 945 | /* |
| 946 | * This is the iodone() function for buffers which have had callbacks |
| 947 | * attached to them by xfs_buf_attach_iodone(). It should remove each |
| 948 | * log item from the buffer's list and call the callback of each in turn. |
| 949 | * When done, the buffer's fsprivate field is set to NULL and the buffer |
| 950 | * is unlocked with a call to iodone(). |
| 951 | */ |
| 952 | void |
| 953 | xfs_buf_iodone_callbacks( |
Christoph Hellwig | bfc6017 | 2011-01-07 13:02:23 +0000 | [diff] [blame] | 954 | struct xfs_buf *bp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 955 | { |
Christoph Hellwig | bfc6017 | 2011-01-07 13:02:23 +0000 | [diff] [blame] | 956 | struct xfs_log_item *lip = bp->b_fspriv; |
| 957 | struct xfs_mount *mp = lip->li_mountp; |
| 958 | static ulong lasttime; |
| 959 | static xfs_buftarg_t *lasttarg; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 960 | |
Chandra Seetharaman | 5a52c2a58 | 2011-07-22 23:39:51 +0000 | [diff] [blame] | 961 | if (likely(!xfs_buf_geterror(bp))) |
Christoph Hellwig | bfc6017 | 2011-01-07 13:02:23 +0000 | [diff] [blame] | 962 | goto do_callbacks; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 963 | |
Christoph Hellwig | bfc6017 | 2011-01-07 13:02:23 +0000 | [diff] [blame] | 964 | /* |
| 965 | * If we've already decided to shutdown the filesystem because of |
| 966 | * I/O errors, there's no point in giving this a retry. |
| 967 | */ |
| 968 | if (XFS_FORCED_SHUTDOWN(mp)) { |
| 969 | XFS_BUF_SUPER_STALE(bp); |
| 970 | trace_xfs_buf_item_iodone(bp, _RET_IP_); |
| 971 | goto do_callbacks; |
| 972 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 973 | |
Chandra Seetharaman | 49074c0 | 2011-07-22 23:40:40 +0000 | [diff] [blame] | 974 | if (bp->b_target != lasttarg || |
Christoph Hellwig | bfc6017 | 2011-01-07 13:02:23 +0000 | [diff] [blame] | 975 | time_after(jiffies, (lasttime + 5*HZ))) { |
| 976 | lasttime = jiffies; |
Dave Chinner | 0b932cc | 2011-03-07 10:08:35 +1100 | [diff] [blame] | 977 | xfs_alert(mp, "Device %s: metadata write error block 0x%llx", |
Chandra Seetharaman | c35a549 | 2011-07-22 23:40:46 +0000 | [diff] [blame] | 978 | xfs_buf_target_name(bp->b_target), |
Dave Chinner | 0b932cc | 2011-03-07 10:08:35 +1100 | [diff] [blame] | 979 | (__uint64_t)XFS_BUF_ADDR(bp)); |
Christoph Hellwig | bfc6017 | 2011-01-07 13:02:23 +0000 | [diff] [blame] | 980 | } |
Chandra Seetharaman | 49074c0 | 2011-07-22 23:40:40 +0000 | [diff] [blame] | 981 | lasttarg = bp->b_target; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 982 | |
Christoph Hellwig | bfc6017 | 2011-01-07 13:02:23 +0000 | [diff] [blame] | 983 | /* |
Lucas De Marchi | 25985ed | 2011-03-30 22:57:33 -0300 | [diff] [blame] | 984 | * If the write was asynchronous then no one will be looking for the |
Christoph Hellwig | bfc6017 | 2011-01-07 13:02:23 +0000 | [diff] [blame] | 985 | * error. Clear the error state and write the buffer out again. |
| 986 | * |
| 987 | * During sync or umount we'll write all pending buffers again |
| 988 | * synchronous, which will catch these errors if they keep hanging |
| 989 | * around. |
| 990 | */ |
| 991 | if (XFS_BUF_ISASYNC(bp)) { |
Chandra Seetharaman | 5a52c2a58 | 2011-07-22 23:39:51 +0000 | [diff] [blame] | 992 | xfs_buf_ioerror(bp, 0); /* errno of 0 unsets the flag */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 993 | |
Christoph Hellwig | bfc6017 | 2011-01-07 13:02:23 +0000 | [diff] [blame] | 994 | if (!XFS_BUF_ISSTALE(bp)) { |
Christoph Hellwig | 61551f1 | 2011-08-23 08:28:06 +0000 | [diff] [blame] | 995 | xfs_buf_delwri_queue(bp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 996 | XFS_BUF_DONE(bp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 997 | } |
Christoph Hellwig | cb669ca | 2011-07-13 13:43:49 +0200 | [diff] [blame] | 998 | ASSERT(bp->b_iodone != NULL); |
Christoph Hellwig | bfc6017 | 2011-01-07 13:02:23 +0000 | [diff] [blame] | 999 | trace_xfs_buf_item_iodone_async(bp, _RET_IP_); |
| 1000 | xfs_buf_relse(bp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1001 | return; |
| 1002 | } |
Christoph Hellwig | 0b1b213 | 2009-12-14 23:14:59 +0000 | [diff] [blame] | 1003 | |
Christoph Hellwig | bfc6017 | 2011-01-07 13:02:23 +0000 | [diff] [blame] | 1004 | /* |
| 1005 | * If the write of the buffer was synchronous, we want to make |
| 1006 | * sure to return the error to the caller of xfs_bwrite(). |
| 1007 | */ |
| 1008 | XFS_BUF_STALE(bp); |
| 1009 | XFS_BUF_DONE(bp); |
Christoph Hellwig | 61551f1 | 2011-08-23 08:28:06 +0000 | [diff] [blame] | 1010 | xfs_buf_delwri_dequeue(bp); |
Christoph Hellwig | bfc6017 | 2011-01-07 13:02:23 +0000 | [diff] [blame] | 1011 | |
| 1012 | trace_xfs_buf_error_relse(bp, _RET_IP_); |
Christoph Hellwig | bfc6017 | 2011-01-07 13:02:23 +0000 | [diff] [blame] | 1013 | |
| 1014 | do_callbacks: |
Dave Chinner | c90821a | 2010-12-03 17:00:52 +1100 | [diff] [blame] | 1015 | xfs_buf_do_callbacks(bp); |
Christoph Hellwig | adadbee | 2011-07-13 13:43:49 +0200 | [diff] [blame] | 1016 | bp->b_fspriv = NULL; |
Christoph Hellwig | cb669ca | 2011-07-13 13:43:49 +0200 | [diff] [blame] | 1017 | bp->b_iodone = NULL; |
Christoph Hellwig | 1a1a3e9 | 2010-10-06 18:41:18 +0000 | [diff] [blame] | 1018 | xfs_buf_ioend(bp, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1019 | } |
| 1020 | |
| 1021 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1022 | * This is the iodone() function for buffers which have been |
| 1023 | * logged. It is called when they are eventually flushed out. |
| 1024 | * It should remove the buf item from the AIL, and free the buf item. |
| 1025 | * It is called by xfs_buf_iodone_callbacks() above which will take |
| 1026 | * care of cleaning up the buffer itself. |
| 1027 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1028 | void |
| 1029 | xfs_buf_iodone( |
Christoph Hellwig | ca30b2a | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 1030 | struct xfs_buf *bp, |
| 1031 | struct xfs_log_item *lip) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1032 | { |
Christoph Hellwig | ca30b2a | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 1033 | struct xfs_ail *ailp = lip->li_ailp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1034 | |
Christoph Hellwig | ca30b2a | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 1035 | ASSERT(BUF_ITEM(lip)->bli_buf == bp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1036 | |
Lachlan McIlroy | e1f5dbd | 2008-09-17 16:52:13 +1000 | [diff] [blame] | 1037 | xfs_buf_rele(bp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1038 | |
| 1039 | /* |
| 1040 | * If we are forcibly shutting down, this may well be |
| 1041 | * off the AIL already. That's because we simulate the |
| 1042 | * log-committed callbacks to unpin these buffers. Or we may never |
| 1043 | * have put this item on AIL because of the transaction was |
David Chinner | 783a2f6 | 2008-10-30 17:39:58 +1100 | [diff] [blame] | 1044 | * aborted forcibly. xfs_trans_ail_delete() takes care of these. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1045 | * |
| 1046 | * Either way, AIL is useless if we're forcing a shutdown. |
| 1047 | */ |
David Chinner | fc1829f | 2008-10-30 17:39:46 +1100 | [diff] [blame] | 1048 | spin_lock(&ailp->xa_lock); |
Christoph Hellwig | ca30b2a | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 1049 | xfs_trans_ail_delete(ailp, lip); |
| 1050 | xfs_buf_item_free(BUF_ITEM(lip)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1051 | } |