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" |
Dave Chinner | 239880e | 2013-10-23 10:50:10 +1100 | [diff] [blame] | 20 | #include "xfs_log_format.h" |
| 21 | #include "xfs_trans_resv.h" |
Nathan Scott | a844f45 | 2005-11-02 14:38:42 +1100 | [diff] [blame] | 22 | #include "xfs_bit.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 23 | #include "xfs_sb.h" |
David Chinner | da353b0 | 2007-08-28 14:00:13 +1000 | [diff] [blame] | 24 | #include "xfs_ag.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 25 | #include "xfs_mount.h" |
Dave Chinner | 239880e | 2013-10-23 10:50:10 +1100 | [diff] [blame] | 26 | #include "xfs_trans.h" |
Nathan Scott | a844f45 | 2005-11-02 14:38:42 +1100 | [diff] [blame] | 27 | #include "xfs_buf_item.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 28 | #include "xfs_trans_priv.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 29 | #include "xfs_error.h" |
Christoph Hellwig | 0b1b213 | 2009-12-14 23:14:59 +0000 | [diff] [blame] | 30 | #include "xfs_trace.h" |
Dave Chinner | 239880e | 2013-10-23 10:50:10 +1100 | [diff] [blame] | 31 | #include "xfs_log.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 | |
Dave Chinner | c90821a | 2010-12-03 17:00:52 +1100 | [diff] [blame] | 41 | STATIC void xfs_buf_do_callbacks(struct xfs_buf *bp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 42 | |
Dave Chinner | 166d136 | 2013-08-12 20:50:04 +1000 | [diff] [blame] | 43 | static inline int |
| 44 | xfs_buf_log_format_size( |
| 45 | struct xfs_buf_log_format *blfp) |
| 46 | { |
| 47 | return offsetof(struct xfs_buf_log_format, blf_data_map) + |
| 48 | (blfp->blf_map_size * sizeof(blfp->blf_data_map[0])); |
| 49 | } |
| 50 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 51 | /* |
| 52 | * This returns the number of log iovecs needed to log the |
| 53 | * given buf log item. |
| 54 | * |
| 55 | * It calculates this as 1 iovec for the buf log format structure |
| 56 | * and 1 for each stretch of non-contiguous chunks to be logged. |
| 57 | * Contiguous chunks are logged in a single iovec. |
| 58 | * |
| 59 | * If the XFS_BLI_STALE flag has been set, then log nothing. |
| 60 | */ |
Dave Chinner | 166d136 | 2013-08-12 20:50:04 +1000 | [diff] [blame] | 61 | STATIC void |
Dave Chinner | 372cc85e | 2012-06-22 18:50:12 +1000 | [diff] [blame] | 62 | xfs_buf_item_size_segment( |
| 63 | struct xfs_buf_log_item *bip, |
Dave Chinner | 166d136 | 2013-08-12 20:50:04 +1000 | [diff] [blame] | 64 | struct xfs_buf_log_format *blfp, |
| 65 | int *nvecs, |
| 66 | int *nbytes) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 67 | { |
Christoph Hellwig | 7bfa31d | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 68 | struct xfs_buf *bp = bip->bli_buf; |
Christoph Hellwig | 7bfa31d | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 69 | int next_bit; |
| 70 | int last_bit; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 71 | |
Dave Chinner | 372cc85e | 2012-06-22 18:50:12 +1000 | [diff] [blame] | 72 | last_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size, 0); |
| 73 | if (last_bit == -1) |
Dave Chinner | 166d136 | 2013-08-12 20:50:04 +1000 | [diff] [blame] | 74 | return; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 75 | |
Dave Chinner | 372cc85e | 2012-06-22 18:50:12 +1000 | [diff] [blame] | 76 | /* |
| 77 | * initial count for a dirty buffer is 2 vectors - the format structure |
| 78 | * and the first dirty region. |
| 79 | */ |
Dave Chinner | 166d136 | 2013-08-12 20:50:04 +1000 | [diff] [blame] | 80 | *nvecs += 2; |
| 81 | *nbytes += xfs_buf_log_format_size(blfp) + XFS_BLF_CHUNK; |
Dave Chinner | 372cc85e | 2012-06-22 18:50:12 +1000 | [diff] [blame] | 82 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 83 | while (last_bit != -1) { |
| 84 | /* |
| 85 | * This takes the bit number to start looking from and |
| 86 | * returns the next set bit from there. It returns -1 |
| 87 | * if there are no more bits set or the start bit is |
| 88 | * beyond the end of the bitmap. |
| 89 | */ |
Dave Chinner | 372cc85e | 2012-06-22 18:50:12 +1000 | [diff] [blame] | 90 | next_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size, |
| 91 | last_bit + 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 92 | /* |
| 93 | * If we run out of bits, leave the loop, |
| 94 | * else if we find a new set of bits bump the number of vecs, |
| 95 | * else keep scanning the current set of bits. |
| 96 | */ |
| 97 | if (next_bit == -1) { |
Dave Chinner | 372cc85e | 2012-06-22 18:50:12 +1000 | [diff] [blame] | 98 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 99 | } else if (next_bit != last_bit + 1) { |
| 100 | last_bit = next_bit; |
Dave Chinner | 166d136 | 2013-08-12 20:50:04 +1000 | [diff] [blame] | 101 | (*nvecs)++; |
Dave Chinner | c115541 | 2010-05-07 11:05:19 +1000 | [diff] [blame] | 102 | } else if (xfs_buf_offset(bp, next_bit * XFS_BLF_CHUNK) != |
| 103 | (xfs_buf_offset(bp, last_bit * XFS_BLF_CHUNK) + |
| 104 | XFS_BLF_CHUNK)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 105 | last_bit = next_bit; |
Dave Chinner | 166d136 | 2013-08-12 20:50:04 +1000 | [diff] [blame] | 106 | (*nvecs)++; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 107 | } else { |
| 108 | last_bit++; |
| 109 | } |
Dave Chinner | 166d136 | 2013-08-12 20:50:04 +1000 | [diff] [blame] | 110 | *nbytes += XFS_BLF_CHUNK; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 111 | } |
Dave Chinner | 372cc85e | 2012-06-22 18:50:12 +1000 | [diff] [blame] | 112 | } |
| 113 | |
| 114 | /* |
| 115 | * This returns the number of log iovecs needed to log the given buf log item. |
| 116 | * |
| 117 | * It calculates this as 1 iovec for the buf log format structure and 1 for each |
| 118 | * stretch of non-contiguous chunks to be logged. Contiguous chunks are logged |
| 119 | * in a single iovec. |
| 120 | * |
| 121 | * Discontiguous buffers need a format structure per region that that is being |
| 122 | * logged. This makes the changes in the buffer appear to log recovery as though |
| 123 | * they came from separate buffers, just like would occur if multiple buffers |
| 124 | * were used instead of a single discontiguous buffer. This enables |
| 125 | * discontiguous buffers to be in-memory constructs, completely transparent to |
| 126 | * what ends up on disk. |
| 127 | * |
| 128 | * If the XFS_BLI_STALE flag has been set, then log nothing but the buf log |
| 129 | * format structures. |
| 130 | */ |
Dave Chinner | 166d136 | 2013-08-12 20:50:04 +1000 | [diff] [blame] | 131 | STATIC void |
Dave Chinner | 372cc85e | 2012-06-22 18:50:12 +1000 | [diff] [blame] | 132 | xfs_buf_item_size( |
Dave Chinner | 166d136 | 2013-08-12 20:50:04 +1000 | [diff] [blame] | 133 | struct xfs_log_item *lip, |
| 134 | int *nvecs, |
| 135 | int *nbytes) |
Dave Chinner | 372cc85e | 2012-06-22 18:50:12 +1000 | [diff] [blame] | 136 | { |
| 137 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); |
Dave Chinner | 372cc85e | 2012-06-22 18:50:12 +1000 | [diff] [blame] | 138 | int i; |
| 139 | |
| 140 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
| 141 | if (bip->bli_flags & XFS_BLI_STALE) { |
| 142 | /* |
| 143 | * The buffer is stale, so all we need to log |
| 144 | * is the buf log format structure with the |
| 145 | * cancel flag in it. |
| 146 | */ |
| 147 | trace_xfs_buf_item_size_stale(bip); |
Mark Tinguely | b943817 | 2012-12-04 17:18:03 -0600 | [diff] [blame] | 148 | ASSERT(bip->__bli_format.blf_flags & XFS_BLF_CANCEL); |
Dave Chinner | 166d136 | 2013-08-12 20:50:04 +1000 | [diff] [blame] | 149 | *nvecs += bip->bli_format_count; |
| 150 | for (i = 0; i < bip->bli_format_count; i++) { |
| 151 | *nbytes += xfs_buf_log_format_size(&bip->bli_formats[i]); |
| 152 | } |
| 153 | return; |
Dave Chinner | 372cc85e | 2012-06-22 18:50:12 +1000 | [diff] [blame] | 154 | } |
| 155 | |
| 156 | ASSERT(bip->bli_flags & XFS_BLI_LOGGED); |
| 157 | |
Dave Chinner | 5f6bed7 | 2013-06-27 16:04:52 +1000 | [diff] [blame] | 158 | if (bip->bli_flags & XFS_BLI_ORDERED) { |
| 159 | /* |
| 160 | * The buffer has been logged just to order it. |
| 161 | * It is not being included in the transaction |
| 162 | * commit, so no vectors are used at all. |
| 163 | */ |
| 164 | trace_xfs_buf_item_size_ordered(bip); |
Dave Chinner | 166d136 | 2013-08-12 20:50:04 +1000 | [diff] [blame] | 165 | *nvecs = XFS_LOG_VEC_ORDERED; |
| 166 | return; |
Dave Chinner | 5f6bed7 | 2013-06-27 16:04:52 +1000 | [diff] [blame] | 167 | } |
| 168 | |
Dave Chinner | 372cc85e | 2012-06-22 18:50:12 +1000 | [diff] [blame] | 169 | /* |
| 170 | * the vector count is based on the number of buffer vectors we have |
| 171 | * dirty bits in. This will only be greater than one when we have a |
| 172 | * compound buffer with more than one segment dirty. Hence for compound |
| 173 | * buffers we need to track which segment the dirty bits correspond to, |
| 174 | * and when we move from one segment to the next increment the vector |
| 175 | * count for the extra buf log format structure that will need to be |
| 176 | * written. |
| 177 | */ |
Dave Chinner | 372cc85e | 2012-06-22 18:50:12 +1000 | [diff] [blame] | 178 | for (i = 0; i < bip->bli_format_count; i++) { |
Dave Chinner | 166d136 | 2013-08-12 20:50:04 +1000 | [diff] [blame] | 179 | xfs_buf_item_size_segment(bip, &bip->bli_formats[i], |
| 180 | nvecs, nbytes); |
Dave Chinner | 372cc85e | 2012-06-22 18:50:12 +1000 | [diff] [blame] | 181 | } |
Christoph Hellwig | 0b1b213 | 2009-12-14 23:14:59 +0000 | [diff] [blame] | 182 | trace_xfs_buf_item_size(bip); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 183 | } |
| 184 | |
Christoph Hellwig | 1234351 | 2013-12-13 11:00:43 +1100 | [diff] [blame] | 185 | static inline void |
Christoph Hellwig | 7aeb722 | 2013-12-13 11:00:43 +1100 | [diff] [blame] | 186 | xfs_buf_item_copy_iovec( |
Christoph Hellwig | bde7cff | 2013-12-13 11:34:02 +1100 | [diff] [blame] | 187 | struct xfs_log_vec *lv, |
Christoph Hellwig | 1234351 | 2013-12-13 11:00:43 +1100 | [diff] [blame] | 188 | struct xfs_log_iovec **vecp, |
Christoph Hellwig | 7aeb722 | 2013-12-13 11:00:43 +1100 | [diff] [blame] | 189 | struct xfs_buf *bp, |
| 190 | uint offset, |
| 191 | int first_bit, |
| 192 | uint nbits) |
| 193 | { |
| 194 | offset += first_bit * XFS_BLF_CHUNK; |
Christoph Hellwig | bde7cff | 2013-12-13 11:34:02 +1100 | [diff] [blame] | 195 | xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_BCHUNK, |
Christoph Hellwig | 1234351 | 2013-12-13 11:00:43 +1100 | [diff] [blame] | 196 | xfs_buf_offset(bp, offset), |
| 197 | nbits * XFS_BLF_CHUNK); |
Christoph Hellwig | 7aeb722 | 2013-12-13 11:00:43 +1100 | [diff] [blame] | 198 | } |
| 199 | |
| 200 | static inline bool |
| 201 | xfs_buf_item_straddle( |
| 202 | struct xfs_buf *bp, |
| 203 | uint offset, |
| 204 | int next_bit, |
| 205 | int last_bit) |
| 206 | { |
| 207 | return xfs_buf_offset(bp, offset + (next_bit << XFS_BLF_SHIFT)) != |
| 208 | (xfs_buf_offset(bp, offset + (last_bit << XFS_BLF_SHIFT)) + |
| 209 | XFS_BLF_CHUNK); |
| 210 | } |
| 211 | |
Christoph Hellwig | 1234351 | 2013-12-13 11:00:43 +1100 | [diff] [blame] | 212 | static void |
Dave Chinner | 372cc85e | 2012-06-22 18:50:12 +1000 | [diff] [blame] | 213 | xfs_buf_item_format_segment( |
| 214 | struct xfs_buf_log_item *bip, |
Christoph Hellwig | bde7cff | 2013-12-13 11:34:02 +1100 | [diff] [blame] | 215 | struct xfs_log_vec *lv, |
Christoph Hellwig | 1234351 | 2013-12-13 11:00:43 +1100 | [diff] [blame] | 216 | struct xfs_log_iovec **vecp, |
Dave Chinner | 372cc85e | 2012-06-22 18:50:12 +1000 | [diff] [blame] | 217 | uint offset, |
| 218 | struct xfs_buf_log_format *blfp) |
| 219 | { |
| 220 | struct xfs_buf *bp = bip->bli_buf; |
| 221 | uint base_size; |
Dave Chinner | 372cc85e | 2012-06-22 18:50:12 +1000 | [diff] [blame] | 222 | int first_bit; |
| 223 | int last_bit; |
| 224 | int next_bit; |
| 225 | uint nbits; |
Dave Chinner | 372cc85e | 2012-06-22 18:50:12 +1000 | [diff] [blame] | 226 | |
| 227 | /* copy the flags across from the base format item */ |
Mark Tinguely | b943817 | 2012-12-04 17:18:03 -0600 | [diff] [blame] | 228 | blfp->blf_flags = bip->__bli_format.blf_flags; |
Dave Chinner | 372cc85e | 2012-06-22 18:50:12 +1000 | [diff] [blame] | 229 | |
| 230 | /* |
| 231 | * Base size is the actual size of the ondisk structure - it reflects |
| 232 | * the actual size of the dirty bitmap rather than the size of the in |
| 233 | * memory structure. |
| 234 | */ |
Dave Chinner | 166d136 | 2013-08-12 20:50:04 +1000 | [diff] [blame] | 235 | base_size = xfs_buf_log_format_size(blfp); |
Mark Tinguely | 820a554 | 2012-12-04 17:18:04 -0600 | [diff] [blame] | 236 | |
Mark Tinguely | 820a554 | 2012-12-04 17:18:04 -0600 | [diff] [blame] | 237 | first_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size, 0); |
| 238 | if (!(bip->bli_flags & XFS_BLI_STALE) && first_bit == -1) { |
| 239 | /* |
| 240 | * If the map is not be dirty in the transaction, mark |
| 241 | * the size as zero and do not advance the vector pointer. |
| 242 | */ |
Christoph Hellwig | bde7cff | 2013-12-13 11:34:02 +1100 | [diff] [blame] | 243 | return; |
Mark Tinguely | 820a554 | 2012-12-04 17:18:04 -0600 | [diff] [blame] | 244 | } |
| 245 | |
Christoph Hellwig | bde7cff | 2013-12-13 11:34:02 +1100 | [diff] [blame] | 246 | blfp = xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_BFORMAT, blfp, base_size); |
| 247 | blfp->blf_size = 1; |
Dave Chinner | 372cc85e | 2012-06-22 18:50:12 +1000 | [diff] [blame] | 248 | |
| 249 | if (bip->bli_flags & XFS_BLI_STALE) { |
| 250 | /* |
| 251 | * The buffer is stale, so all we need to log |
| 252 | * is the buf log format structure with the |
| 253 | * cancel flag in it. |
| 254 | */ |
| 255 | trace_xfs_buf_item_format_stale(bip); |
| 256 | ASSERT(blfp->blf_flags & XFS_BLF_CANCEL); |
Christoph Hellwig | bde7cff | 2013-12-13 11:34:02 +1100 | [diff] [blame] | 257 | return; |
Dave Chinner | 372cc85e | 2012-06-22 18:50:12 +1000 | [diff] [blame] | 258 | } |
| 259 | |
Dave Chinner | 5f6bed7 | 2013-06-27 16:04:52 +1000 | [diff] [blame] | 260 | |
Dave Chinner | 372cc85e | 2012-06-22 18:50:12 +1000 | [diff] [blame] | 261 | /* |
| 262 | * Fill in an iovec for each set of contiguous chunks. |
| 263 | */ |
Dave Chinner | 372cc85e | 2012-06-22 18:50:12 +1000 | [diff] [blame] | 264 | last_bit = first_bit; |
| 265 | nbits = 1; |
| 266 | for (;;) { |
| 267 | /* |
| 268 | * This takes the bit number to start looking from and |
| 269 | * returns the next set bit from there. It returns -1 |
| 270 | * if there are no more bits set or the start bit is |
| 271 | * beyond the end of the bitmap. |
| 272 | */ |
| 273 | next_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size, |
| 274 | (uint)last_bit + 1); |
| 275 | /* |
Christoph Hellwig | 7aeb722 | 2013-12-13 11:00:43 +1100 | [diff] [blame] | 276 | * If we run out of bits fill in the last iovec and get out of |
| 277 | * the loop. Else if we start a new set of bits then fill in |
| 278 | * the iovec for the series we were looking at and start |
| 279 | * counting the bits in the new one. Else we're still in the |
| 280 | * same set of bits so just keep counting and scanning. |
Dave Chinner | 372cc85e | 2012-06-22 18:50:12 +1000 | [diff] [blame] | 281 | */ |
| 282 | if (next_bit == -1) { |
Christoph Hellwig | bde7cff | 2013-12-13 11:34:02 +1100 | [diff] [blame] | 283 | xfs_buf_item_copy_iovec(lv, vecp, bp, offset, |
Christoph Hellwig | 7aeb722 | 2013-12-13 11:00:43 +1100 | [diff] [blame] | 284 | first_bit, nbits); |
Christoph Hellwig | bde7cff | 2013-12-13 11:34:02 +1100 | [diff] [blame] | 285 | blfp->blf_size++; |
Dave Chinner | 372cc85e | 2012-06-22 18:50:12 +1000 | [diff] [blame] | 286 | break; |
Christoph Hellwig | 7aeb722 | 2013-12-13 11:00:43 +1100 | [diff] [blame] | 287 | } else if (next_bit != last_bit + 1 || |
| 288 | xfs_buf_item_straddle(bp, offset, next_bit, last_bit)) { |
Christoph Hellwig | bde7cff | 2013-12-13 11:34:02 +1100 | [diff] [blame] | 289 | xfs_buf_item_copy_iovec(lv, vecp, bp, offset, |
Christoph Hellwig | 1234351 | 2013-12-13 11:00:43 +1100 | [diff] [blame] | 290 | first_bit, nbits); |
Christoph Hellwig | bde7cff | 2013-12-13 11:34:02 +1100 | [diff] [blame] | 291 | blfp->blf_size++; |
Dave Chinner | 372cc85e | 2012-06-22 18:50:12 +1000 | [diff] [blame] | 292 | first_bit = next_bit; |
| 293 | last_bit = next_bit; |
| 294 | nbits = 1; |
| 295 | } else { |
| 296 | last_bit++; |
| 297 | nbits++; |
| 298 | } |
| 299 | } |
Dave Chinner | 372cc85e | 2012-06-22 18:50:12 +1000 | [diff] [blame] | 300 | } |
| 301 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 302 | /* |
| 303 | * This is called to fill in the vector of log iovecs for the |
| 304 | * given log buf item. It fills the first entry with a buf log |
| 305 | * format structure, and the rest point to contiguous chunks |
| 306 | * within the buffer. |
| 307 | */ |
Christoph Hellwig | ba0f32d | 2005-06-21 15:36:52 +1000 | [diff] [blame] | 308 | STATIC void |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 309 | xfs_buf_item_format( |
Christoph Hellwig | 7bfa31d | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 310 | struct xfs_log_item *lip, |
Christoph Hellwig | bde7cff | 2013-12-13 11:34:02 +1100 | [diff] [blame] | 311 | struct xfs_log_vec *lv) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 312 | { |
Christoph Hellwig | 7bfa31d | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 313 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); |
Dave Chinner | 372cc85e | 2012-06-22 18:50:12 +1000 | [diff] [blame] | 314 | struct xfs_buf *bp = bip->bli_buf; |
Christoph Hellwig | bde7cff | 2013-12-13 11:34:02 +1100 | [diff] [blame] | 315 | struct xfs_log_iovec *vecp = NULL; |
Dave Chinner | 372cc85e | 2012-06-22 18:50:12 +1000 | [diff] [blame] | 316 | uint offset = 0; |
| 317 | int i; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 318 | |
| 319 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
| 320 | ASSERT((bip->bli_flags & XFS_BLI_LOGGED) || |
| 321 | (bip->bli_flags & XFS_BLI_STALE)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 322 | |
| 323 | /* |
Dave Chinner | ccf7c23 | 2010-05-20 23:19:42 +1000 | [diff] [blame] | 324 | * If it is an inode buffer, transfer the in-memory state to the |
Dave Chinner | ddf6ad0 | 2013-06-27 16:04:56 +1000 | [diff] [blame] | 325 | * format flags and clear the in-memory state. |
| 326 | * |
| 327 | * For buffer based inode allocation, we do not transfer |
Dave Chinner | ccf7c23 | 2010-05-20 23:19:42 +1000 | [diff] [blame] | 328 | * this state if the inode buffer allocation has not yet been committed |
| 329 | * to the log as setting the XFS_BLI_INODE_BUF flag will prevent |
| 330 | * correct replay of the inode allocation. |
Dave Chinner | ddf6ad0 | 2013-06-27 16:04:56 +1000 | [diff] [blame] | 331 | * |
| 332 | * For icreate item based inode allocation, the buffers aren't written |
| 333 | * to the journal during allocation, and hence we should always tag the |
| 334 | * buffer as an inode buffer so that the correct unlinked list replay |
| 335 | * occurs during recovery. |
Dave Chinner | ccf7c23 | 2010-05-20 23:19:42 +1000 | [diff] [blame] | 336 | */ |
| 337 | if (bip->bli_flags & XFS_BLI_INODE_BUF) { |
Dave Chinner | ddf6ad0 | 2013-06-27 16:04:56 +1000 | [diff] [blame] | 338 | if (xfs_sb_version_hascrc(&lip->li_mountp->m_sb) || |
| 339 | !((bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF) && |
Christoph Hellwig | 7bfa31d | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 340 | xfs_log_item_in_current_chkpt(lip))) |
Mark Tinguely | b943817 | 2012-12-04 17:18:03 -0600 | [diff] [blame] | 341 | bip->__bli_format.blf_flags |= XFS_BLF_INODE_BUF; |
Dave Chinner | ccf7c23 | 2010-05-20 23:19:42 +1000 | [diff] [blame] | 342 | bip->bli_flags &= ~XFS_BLI_INODE_BUF; |
| 343 | } |
| 344 | |
Dave Chinner | 5f6bed7 | 2013-06-27 16:04:52 +1000 | [diff] [blame] | 345 | if ((bip->bli_flags & (XFS_BLI_ORDERED|XFS_BLI_STALE)) == |
| 346 | XFS_BLI_ORDERED) { |
| 347 | /* |
| 348 | * The buffer has been logged just to order it. It is not being |
| 349 | * included in the transaction commit, so don't format it. |
| 350 | */ |
| 351 | trace_xfs_buf_item_format_ordered(bip); |
| 352 | return; |
| 353 | } |
| 354 | |
Dave Chinner | 372cc85e | 2012-06-22 18:50:12 +1000 | [diff] [blame] | 355 | for (i = 0; i < bip->bli_format_count; i++) { |
Christoph Hellwig | bde7cff | 2013-12-13 11:34:02 +1100 | [diff] [blame] | 356 | xfs_buf_item_format_segment(bip, lv, &vecp, offset, |
Christoph Hellwig | 1234351 | 2013-12-13 11:00:43 +1100 | [diff] [blame] | 357 | &bip->bli_formats[i]); |
Dave Chinner | 372cc85e | 2012-06-22 18:50:12 +1000 | [diff] [blame] | 358 | offset += bp->b_maps[i].bm_len; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 359 | } |
| 360 | |
| 361 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 362 | * Check to make sure everything is consistent. |
| 363 | */ |
Christoph Hellwig | 0b1b213 | 2009-12-14 23:14:59 +0000 | [diff] [blame] | 364 | trace_xfs_buf_item_format(bip); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 365 | } |
| 366 | |
| 367 | /* |
Dave Chinner | 64fc35d | 2010-05-07 11:04:34 +1000 | [diff] [blame] | 368 | * 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] | 369 | * so it cannot be written out. |
Dave Chinner | 64fc35d | 2010-05-07 11:04:34 +1000 | [diff] [blame] | 370 | * |
| 371 | * We also always take a reference to the buffer log item here so that the bli |
| 372 | * is held while the item is pinned in memory. This means that we can |
| 373 | * unconditionally drop the reference count a transaction holds when the |
| 374 | * transaction is completed. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 375 | */ |
Christoph Hellwig | ba0f32d | 2005-06-21 15:36:52 +1000 | [diff] [blame] | 376 | STATIC void |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 377 | xfs_buf_item_pin( |
Christoph Hellwig | 7bfa31d | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 378 | struct xfs_log_item *lip) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 379 | { |
Christoph Hellwig | 7bfa31d | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 380 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 381 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 382 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
| 383 | ASSERT((bip->bli_flags & XFS_BLI_LOGGED) || |
Dave Chinner | 5f6bed7 | 2013-06-27 16:04:52 +1000 | [diff] [blame] | 384 | (bip->bli_flags & XFS_BLI_ORDERED) || |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 385 | (bip->bli_flags & XFS_BLI_STALE)); |
Christoph Hellwig | 7bfa31d | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 386 | |
Christoph Hellwig | 0b1b213 | 2009-12-14 23:14:59 +0000 | [diff] [blame] | 387 | trace_xfs_buf_item_pin(bip); |
Christoph Hellwig | 4d16e92 | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 388 | |
| 389 | atomic_inc(&bip->bli_refcount); |
| 390 | atomic_inc(&bip->bli_buf->b_pin_count); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 391 | } |
| 392 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 393 | /* |
| 394 | * This is called to unpin the buffer associated with the buf log |
| 395 | * 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] | 396 | * |
| 397 | * Also drop the reference to the buf item for the current transaction. |
| 398 | * If the XFS_BLI_STALE flag is set and we are the last reference, |
| 399 | * then free up the buf log item and unlock the buffer. |
Christoph Hellwig | 9412e31 | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 400 | * |
| 401 | * If the remove flag is set we are called from uncommit in the |
| 402 | * forced-shutdown path. If that is true and the reference count on |
| 403 | * the log item is going to drop to zero we need to free the item's |
| 404 | * descriptor in the transaction. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 405 | */ |
Christoph Hellwig | ba0f32d | 2005-06-21 15:36:52 +1000 | [diff] [blame] | 406 | STATIC void |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 407 | xfs_buf_item_unpin( |
Christoph Hellwig | 7bfa31d | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 408 | struct xfs_log_item *lip, |
Christoph Hellwig | 9412e31 | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 409 | int remove) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 410 | { |
Christoph Hellwig | 7bfa31d | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 411 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); |
Christoph Hellwig | 9412e31 | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 412 | xfs_buf_t *bp = bip->bli_buf; |
Christoph Hellwig | 7bfa31d | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 413 | struct xfs_ail *ailp = lip->li_ailp; |
Dave Chinner | 8e12385 | 2010-03-08 11:26:03 +1100 | [diff] [blame] | 414 | int stale = bip->bli_flags & XFS_BLI_STALE; |
Christoph Hellwig | 7bfa31d | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 415 | int freed; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 416 | |
Christoph Hellwig | adadbee | 2011-07-13 13:43:49 +0200 | [diff] [blame] | 417 | ASSERT(bp->b_fspriv == bip); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 418 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
Christoph Hellwig | 9412e31 | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 419 | |
Christoph Hellwig | 0b1b213 | 2009-12-14 23:14:59 +0000 | [diff] [blame] | 420 | trace_xfs_buf_item_unpin(bip); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 421 | |
| 422 | freed = atomic_dec_and_test(&bip->bli_refcount); |
Christoph Hellwig | 4d16e92 | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 423 | |
| 424 | if (atomic_dec_and_test(&bp->b_pin_count)) |
| 425 | wake_up_all(&bp->b_waiters); |
Christoph Hellwig | 7bfa31d | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 426 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 427 | if (freed && stale) { |
| 428 | ASSERT(bip->bli_flags & XFS_BLI_STALE); |
Christoph Hellwig | 0c842ad | 2011-07-08 14:36:19 +0200 | [diff] [blame] | 429 | ASSERT(xfs_buf_islocked(bp)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 430 | ASSERT(XFS_BUF_ISSTALE(bp)); |
Mark Tinguely | b943817 | 2012-12-04 17:18:03 -0600 | [diff] [blame] | 431 | ASSERT(bip->__bli_format.blf_flags & XFS_BLF_CANCEL); |
Christoph Hellwig | 9412e31 | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 432 | |
Christoph Hellwig | 0b1b213 | 2009-12-14 23:14:59 +0000 | [diff] [blame] | 433 | trace_xfs_buf_item_unpin_stale(bip); |
| 434 | |
Christoph Hellwig | 9412e31 | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 435 | if (remove) { |
| 436 | /* |
Dave Chinner | e34a314 | 2011-01-27 12:13:35 +1100 | [diff] [blame] | 437 | * If we are in a transaction context, we have to |
| 438 | * remove the log item from the transaction as we are |
| 439 | * about to release our reference to the buffer. If we |
| 440 | * don't, the unlock that occurs later in |
| 441 | * xfs_trans_uncommit() will try to reference the |
Christoph Hellwig | 9412e31 | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 442 | * buffer which we no longer have a hold on. |
| 443 | */ |
Dave Chinner | e34a314 | 2011-01-27 12:13:35 +1100 | [diff] [blame] | 444 | if (lip->li_desc) |
| 445 | xfs_trans_del_item(lip); |
Christoph Hellwig | 9412e31 | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 446 | |
| 447 | /* |
| 448 | * Since the transaction no longer refers to the buffer, |
| 449 | * the buffer should no longer refer to the transaction. |
| 450 | */ |
Christoph Hellwig | bf9d901 | 2011-07-13 13:43:49 +0200 | [diff] [blame] | 451 | bp->b_transp = NULL; |
Christoph Hellwig | 9412e31 | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 452 | } |
| 453 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 454 | /* |
| 455 | * If we get called here because of an IO error, we may |
David Chinner | 783a2f6 | 2008-10-30 17:39:58 +1100 | [diff] [blame] | 456 | * 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] | 457 | * will take care of that situation. |
David Chinner | 783a2f6 | 2008-10-30 17:39:58 +1100 | [diff] [blame] | 458 | * xfs_trans_ail_delete() drops the AIL lock. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 459 | */ |
| 460 | if (bip->bli_flags & XFS_BLI_STALE_INODE) { |
Dave Chinner | c90821a | 2010-12-03 17:00:52 +1100 | [diff] [blame] | 461 | xfs_buf_do_callbacks(bp); |
Christoph Hellwig | adadbee | 2011-07-13 13:43:49 +0200 | [diff] [blame] | 462 | bp->b_fspriv = NULL; |
Christoph Hellwig | cb669ca | 2011-07-13 13:43:49 +0200 | [diff] [blame] | 463 | bp->b_iodone = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 464 | } else { |
David Chinner | 783a2f6 | 2008-10-30 17:39:58 +1100 | [diff] [blame] | 465 | spin_lock(&ailp->xa_lock); |
Dave Chinner | 04913fd | 2012-04-23 15:58:41 +1000 | [diff] [blame] | 466 | xfs_trans_ail_delete(ailp, lip, SHUTDOWN_LOG_IO_ERROR); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 467 | xfs_buf_item_relse(bp); |
Christoph Hellwig | adadbee | 2011-07-13 13:43:49 +0200 | [diff] [blame] | 468 | ASSERT(bp->b_fspriv == NULL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 469 | } |
| 470 | xfs_buf_relse(bp); |
Christoph Hellwig | 960c60a | 2012-04-23 15:58:38 +1000 | [diff] [blame] | 471 | } else if (freed && remove) { |
Dave Chinner | 137fff0 | 2012-11-02 14:23:12 +1100 | [diff] [blame] | 472 | /* |
| 473 | * There are currently two references to the buffer - the active |
| 474 | * LRU reference and the buf log item. What we are about to do |
| 475 | * here - simulate a failed IO completion - requires 3 |
| 476 | * references. |
| 477 | * |
| 478 | * The LRU reference is removed by the xfs_buf_stale() call. The |
| 479 | * buf item reference is removed by the xfs_buf_iodone() |
| 480 | * callback that is run by xfs_buf_do_callbacks() during ioend |
| 481 | * processing (via the bp->b_iodone callback), and then finally |
| 482 | * the ioend processing will drop the IO reference if the buffer |
| 483 | * is marked XBF_ASYNC. |
| 484 | * |
| 485 | * Hence we need to take an additional reference here so that IO |
| 486 | * completion processing doesn't free the buffer prematurely. |
| 487 | */ |
Christoph Hellwig | 960c60a | 2012-04-23 15:58:38 +1000 | [diff] [blame] | 488 | xfs_buf_lock(bp); |
Dave Chinner | 137fff0 | 2012-11-02 14:23:12 +1100 | [diff] [blame] | 489 | xfs_buf_hold(bp); |
| 490 | bp->b_flags |= XBF_ASYNC; |
Christoph Hellwig | 960c60a | 2012-04-23 15:58:38 +1000 | [diff] [blame] | 491 | xfs_buf_ioerror(bp, EIO); |
| 492 | XFS_BUF_UNDONE(bp); |
| 493 | xfs_buf_stale(bp); |
| 494 | xfs_buf_ioend(bp, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 495 | } |
| 496 | } |
| 497 | |
Dave Chinner | ac8809f | 2013-12-12 16:34:38 +1100 | [diff] [blame] | 498 | /* |
| 499 | * Buffer IO error rate limiting. Limit it to no more than 10 messages per 30 |
| 500 | * seconds so as to not spam logs too much on repeated detection of the same |
| 501 | * buffer being bad.. |
| 502 | */ |
| 503 | |
| 504 | DEFINE_RATELIMIT_STATE(xfs_buf_write_fail_rl_state, 30 * HZ, 10); |
| 505 | |
Christoph Hellwig | ba0f32d | 2005-06-21 15:36:52 +1000 | [diff] [blame] | 506 | STATIC uint |
Christoph Hellwig | 43ff212 | 2012-04-23 15:58:39 +1000 | [diff] [blame] | 507 | xfs_buf_item_push( |
| 508 | struct xfs_log_item *lip, |
| 509 | struct list_head *buffer_list) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 510 | { |
Christoph Hellwig | 7bfa31d | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 511 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); |
| 512 | struct xfs_buf *bp = bip->bli_buf; |
Christoph Hellwig | 43ff212 | 2012-04-23 15:58:39 +1000 | [diff] [blame] | 513 | uint rval = XFS_ITEM_SUCCESS; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 514 | |
Chandra Seetharaman | 811e64c | 2011-07-22 23:40:27 +0000 | [diff] [blame] | 515 | if (xfs_buf_ispinned(bp)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 516 | return XFS_ITEM_PINNED; |
Brian Foster | 5337fe9 | 2013-02-11 10:08:21 -0500 | [diff] [blame] | 517 | if (!xfs_buf_trylock(bp)) { |
| 518 | /* |
| 519 | * If we have just raced with a buffer being pinned and it has |
| 520 | * been marked stale, we could end up stalling until someone else |
| 521 | * issues a log force to unpin the stale buffer. Check for the |
| 522 | * race condition here so xfsaild recognizes the buffer is pinned |
| 523 | * and queues a log force to move it along. |
| 524 | */ |
| 525 | if (xfs_buf_ispinned(bp)) |
| 526 | return XFS_ITEM_PINNED; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 527 | return XFS_ITEM_LOCKED; |
Brian Foster | 5337fe9 | 2013-02-11 10:08:21 -0500 | [diff] [blame] | 528 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 529 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 530 | ASSERT(!(bip->bli_flags & XFS_BLI_STALE)); |
Christoph Hellwig | 43ff212 | 2012-04-23 15:58:39 +1000 | [diff] [blame] | 531 | |
| 532 | trace_xfs_buf_item_push(bip); |
| 533 | |
Dave Chinner | ac8809f | 2013-12-12 16:34:38 +1100 | [diff] [blame] | 534 | /* has a previous flush failed due to IO errors? */ |
| 535 | if ((bp->b_flags & XBF_WRITE_FAIL) && |
| 536 | ___ratelimit(&xfs_buf_write_fail_rl_state, "XFS:")) { |
| 537 | xfs_warn(bp->b_target->bt_mount, |
| 538 | "Detected failing async write on buffer block 0x%llx. Retrying async write.\n", |
| 539 | (long long)bp->b_bn); |
| 540 | } |
| 541 | |
Christoph Hellwig | 43ff212 | 2012-04-23 15:58:39 +1000 | [diff] [blame] | 542 | if (!xfs_buf_delwri_queue(bp, buffer_list)) |
| 543 | rval = XFS_ITEM_FLUSHING; |
| 544 | xfs_buf_unlock(bp); |
| 545 | return rval; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 546 | } |
| 547 | |
| 548 | /* |
Dave Chinner | 64fc35d | 2010-05-07 11:04:34 +1000 | [diff] [blame] | 549 | * Release the buffer associated with the buf log item. If there is no dirty |
| 550 | * logged data associated with the buffer recorded in the buf log item, then |
| 551 | * 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] | 552 | * |
Dave Chinner | 64fc35d | 2010-05-07 11:04:34 +1000 | [diff] [blame] | 553 | * This call ignores the recursion count. It is only called when the buffer |
| 554 | * should REALLY be unlocked, regardless of the recursion count. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 555 | * |
Dave Chinner | 64fc35d | 2010-05-07 11:04:34 +1000 | [diff] [blame] | 556 | * We unconditionally drop the transaction's reference to the log item. If the |
| 557 | * item was logged, then another reference was taken when it was pinned, so we |
| 558 | * can safely drop the transaction reference now. This also allows us to avoid |
| 559 | * potential races with the unpin code freeing the bli by not referencing the |
| 560 | * bli after we've dropped the reference count. |
| 561 | * |
| 562 | * If the XFS_BLI_HOLD flag is set in the buf log item, then free the log item |
| 563 | * if necessary but do not unlock the buffer. This is for support of |
| 564 | * xfs_trans_bhold(). Make sure the XFS_BLI_HOLD field is cleared if we don't |
| 565 | * free the item. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 566 | */ |
Christoph Hellwig | ba0f32d | 2005-06-21 15:36:52 +1000 | [diff] [blame] | 567 | STATIC void |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 568 | xfs_buf_item_unlock( |
Christoph Hellwig | 7bfa31d | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 569 | struct xfs_log_item *lip) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 570 | { |
Christoph Hellwig | 7bfa31d | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 571 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); |
| 572 | struct xfs_buf *bp = bip->bli_buf; |
Dave Chinner | 5f6bed7 | 2013-06-27 16:04:52 +1000 | [diff] [blame] | 573 | bool clean; |
| 574 | bool aborted; |
| 575 | int flags; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 576 | |
Dave Chinner | 64fc35d | 2010-05-07 11:04:34 +1000 | [diff] [blame] | 577 | /* Clear the buffer's association with this transaction. */ |
Christoph Hellwig | bf9d901 | 2011-07-13 13:43:49 +0200 | [diff] [blame] | 578 | bp->b_transp = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 579 | |
| 580 | /* |
Dave Chinner | 64fc35d | 2010-05-07 11:04:34 +1000 | [diff] [blame] | 581 | * If this is a transaction abort, don't return early. Instead, allow |
| 582 | * the brelse to happen. Normally it would be done for stale |
| 583 | * (cancelled) buffers at unpin time, but we'll never go through the |
| 584 | * pin/unpin cycle if we abort inside commit. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 585 | */ |
Dave Chinner | 5f6bed7 | 2013-06-27 16:04:52 +1000 | [diff] [blame] | 586 | aborted = (lip->li_flags & XFS_LI_ABORTED) ? true : false; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 587 | /* |
Dave Chinner | 5f6bed7 | 2013-06-27 16:04:52 +1000 | [diff] [blame] | 588 | * Before possibly freeing the buf item, copy the per-transaction state |
| 589 | * so we can reference it safely later after clearing it from the |
| 590 | * buffer log item. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 591 | */ |
Dave Chinner | 5f6bed7 | 2013-06-27 16:04:52 +1000 | [diff] [blame] | 592 | flags = bip->bli_flags; |
| 593 | bip->bli_flags &= ~(XFS_BLI_LOGGED | XFS_BLI_HOLD | XFS_BLI_ORDERED); |
Dave Chinner | 64fc35d | 2010-05-07 11:04:34 +1000 | [diff] [blame] | 594 | |
| 595 | /* |
| 596 | * If the buf item is marked stale, then don't do anything. We'll |
| 597 | * unlock the buffer and free the buf item when the buffer is unpinned |
| 598 | * for the last time. |
| 599 | */ |
Dave Chinner | 5f6bed7 | 2013-06-27 16:04:52 +1000 | [diff] [blame] | 600 | if (flags & XFS_BLI_STALE) { |
Dave Chinner | 64fc35d | 2010-05-07 11:04:34 +1000 | [diff] [blame] | 601 | trace_xfs_buf_item_unlock_stale(bip); |
Mark Tinguely | b943817 | 2012-12-04 17:18:03 -0600 | [diff] [blame] | 602 | ASSERT(bip->__bli_format.blf_flags & XFS_BLF_CANCEL); |
Dave Chinner | 64fc35d | 2010-05-07 11:04:34 +1000 | [diff] [blame] | 603 | if (!aborted) { |
| 604 | atomic_dec(&bip->bli_refcount); |
| 605 | return; |
| 606 | } |
| 607 | } |
| 608 | |
Christoph Hellwig | 0b1b213 | 2009-12-14 23:14:59 +0000 | [diff] [blame] | 609 | trace_xfs_buf_item_unlock(bip); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 610 | |
| 611 | /* |
Dave Chinner | 64fc35d | 2010-05-07 11:04:34 +1000 | [diff] [blame] | 612 | * If the buf item isn't tracking any data, free it, otherwise drop the |
Dave Chinner | 3b19034 | 2013-01-21 23:53:55 +1100 | [diff] [blame] | 613 | * reference we hold to it. If we are aborting the transaction, this may |
| 614 | * be the only reference to the buf item, so we free it anyway |
| 615 | * regardless of whether it is dirty or not. A dirty abort implies a |
| 616 | * shutdown, anyway. |
Dave Chinner | 5f6bed7 | 2013-06-27 16:04:52 +1000 | [diff] [blame] | 617 | * |
| 618 | * Ordered buffers are dirty but may have no recorded changes, so ensure |
| 619 | * we only release clean items here. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 620 | */ |
Dave Chinner | 5f6bed7 | 2013-06-27 16:04:52 +1000 | [diff] [blame] | 621 | clean = (flags & XFS_BLI_DIRTY) ? false : true; |
| 622 | if (clean) { |
| 623 | int i; |
| 624 | for (i = 0; i < bip->bli_format_count; i++) { |
| 625 | if (!xfs_bitmap_empty(bip->bli_formats[i].blf_data_map, |
| 626 | bip->bli_formats[i].blf_map_size)) { |
| 627 | clean = false; |
| 628 | break; |
| 629 | } |
Mark Tinguely | c883d0c | 2012-12-04 17:18:05 -0600 | [diff] [blame] | 630 | } |
| 631 | } |
Dave Chinner | 46f9d2e | 2013-09-03 21:47:37 +1000 | [diff] [blame] | 632 | |
| 633 | /* |
| 634 | * Clean buffers, by definition, cannot be in the AIL. However, aborted |
| 635 | * buffers may be dirty and hence in the AIL. Therefore if we are |
| 636 | * aborting a buffer and we've just taken the last refernce away, we |
| 637 | * have to check if it is in the AIL before freeing it. We need to free |
| 638 | * it in this case, because an aborted transaction has already shut the |
| 639 | * filesystem down and this is the last chance we will have to do so. |
| 640 | */ |
| 641 | if (atomic_dec_and_test(&bip->bli_refcount)) { |
| 642 | if (clean) |
| 643 | xfs_buf_item_relse(bp); |
| 644 | else if (aborted) { |
| 645 | ASSERT(XFS_FORCED_SHUTDOWN(lip->li_mountp)); |
| 646 | if (lip->li_flags & XFS_LI_IN_AIL) { |
Dave Chinner | 4885235 | 2013-09-24 16:01:13 +1000 | [diff] [blame] | 647 | spin_lock(&lip->li_ailp->xa_lock); |
Dave Chinner | 46f9d2e | 2013-09-03 21:47:37 +1000 | [diff] [blame] | 648 | xfs_trans_ail_delete(lip->li_ailp, lip, |
| 649 | SHUTDOWN_LOG_IO_ERROR); |
| 650 | } |
Dave Chinner | 3b19034 | 2013-01-21 23:53:55 +1100 | [diff] [blame] | 651 | xfs_buf_item_relse(bp); |
| 652 | } |
Dave Chinner | 46f9d2e | 2013-09-03 21:47:37 +1000 | [diff] [blame] | 653 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 654 | |
Dave Chinner | 5f6bed7 | 2013-06-27 16:04:52 +1000 | [diff] [blame] | 655 | if (!(flags & XFS_BLI_HOLD)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 656 | xfs_buf_relse(bp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 657 | } |
| 658 | |
| 659 | /* |
| 660 | * This is called to find out where the oldest active copy of the |
| 661 | * buf log item in the on disk log resides now that the last log |
| 662 | * write of it completed at the given lsn. |
| 663 | * We always re-log all the dirty data in a buffer, so usually the |
| 664 | * latest copy in the on disk log is the only one that matters. For |
| 665 | * those cases we simply return the given lsn. |
| 666 | * |
| 667 | * The one exception to this is for buffers full of newly allocated |
| 668 | * inodes. These buffers are only relogged with the XFS_BLI_INODE_BUF |
| 669 | * flag set, indicating that only the di_next_unlinked fields from the |
| 670 | * inodes in the buffers will be replayed during recovery. If the |
| 671 | * original newly allocated inode images have not yet been flushed |
| 672 | * when the buffer is so relogged, then we need to make sure that we |
| 673 | * keep the old images in the 'active' portion of the log. We do this |
| 674 | * by returning the original lsn of that transaction here rather than |
| 675 | * the current one. |
| 676 | */ |
Christoph Hellwig | ba0f32d | 2005-06-21 15:36:52 +1000 | [diff] [blame] | 677 | STATIC xfs_lsn_t |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 678 | xfs_buf_item_committed( |
Christoph Hellwig | 7bfa31d | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 679 | struct xfs_log_item *lip, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 680 | xfs_lsn_t lsn) |
| 681 | { |
Christoph Hellwig | 7bfa31d | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 682 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); |
| 683 | |
Christoph Hellwig | 0b1b213 | 2009-12-14 23:14:59 +0000 | [diff] [blame] | 684 | trace_xfs_buf_item_committed(bip); |
| 685 | |
Christoph Hellwig | 7bfa31d | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 686 | if ((bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF) && lip->li_lsn != 0) |
| 687 | return lip->li_lsn; |
| 688 | return lsn; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 689 | } |
| 690 | |
Christoph Hellwig | ba0f32d | 2005-06-21 15:36:52 +1000 | [diff] [blame] | 691 | STATIC void |
Christoph Hellwig | 7bfa31d | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 692 | xfs_buf_item_committing( |
| 693 | struct xfs_log_item *lip, |
| 694 | xfs_lsn_t commit_lsn) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 695 | { |
| 696 | } |
| 697 | |
| 698 | /* |
| 699 | * This is the ops vector shared by all buf log items. |
| 700 | */ |
Christoph Hellwig | 272e42b | 2011-10-28 09:54:24 +0000 | [diff] [blame] | 701 | static const struct xfs_item_ops xfs_buf_item_ops = { |
Christoph Hellwig | 7bfa31d | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 702 | .iop_size = xfs_buf_item_size, |
| 703 | .iop_format = xfs_buf_item_format, |
| 704 | .iop_pin = xfs_buf_item_pin, |
| 705 | .iop_unpin = xfs_buf_item_unpin, |
Christoph Hellwig | 7bfa31d | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 706 | .iop_unlock = xfs_buf_item_unlock, |
| 707 | .iop_committed = xfs_buf_item_committed, |
| 708 | .iop_push = xfs_buf_item_push, |
Christoph Hellwig | 7bfa31d | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 709 | .iop_committing = xfs_buf_item_committing |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 710 | }; |
| 711 | |
Dave Chinner | 372cc85e | 2012-06-22 18:50:12 +1000 | [diff] [blame] | 712 | STATIC int |
| 713 | xfs_buf_item_get_format( |
| 714 | struct xfs_buf_log_item *bip, |
| 715 | int count) |
| 716 | { |
| 717 | ASSERT(bip->bli_formats == NULL); |
| 718 | bip->bli_format_count = count; |
| 719 | |
| 720 | if (count == 1) { |
Mark Tinguely | b943817 | 2012-12-04 17:18:03 -0600 | [diff] [blame] | 721 | bip->bli_formats = &bip->__bli_format; |
Dave Chinner | 372cc85e | 2012-06-22 18:50:12 +1000 | [diff] [blame] | 722 | return 0; |
| 723 | } |
| 724 | |
| 725 | bip->bli_formats = kmem_zalloc(count * sizeof(struct xfs_buf_log_format), |
| 726 | KM_SLEEP); |
| 727 | if (!bip->bli_formats) |
| 728 | return ENOMEM; |
| 729 | return 0; |
| 730 | } |
| 731 | |
| 732 | STATIC void |
| 733 | xfs_buf_item_free_format( |
| 734 | struct xfs_buf_log_item *bip) |
| 735 | { |
Mark Tinguely | b943817 | 2012-12-04 17:18:03 -0600 | [diff] [blame] | 736 | if (bip->bli_formats != &bip->__bli_format) { |
Dave Chinner | 372cc85e | 2012-06-22 18:50:12 +1000 | [diff] [blame] | 737 | kmem_free(bip->bli_formats); |
| 738 | bip->bli_formats = NULL; |
| 739 | } |
| 740 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 741 | |
| 742 | /* |
| 743 | * Allocate a new buf log item to go with the given buffer. |
| 744 | * Set the buffer's b_fsprivate field to point to the new |
| 745 | * buf log item. If there are other item's attached to the |
| 746 | * buffer (see xfs_buf_attach_iodone() below), then put the |
| 747 | * buf log item at the front. |
| 748 | */ |
| 749 | void |
| 750 | xfs_buf_item_init( |
| 751 | xfs_buf_t *bp, |
| 752 | xfs_mount_t *mp) |
| 753 | { |
Christoph Hellwig | adadbee | 2011-07-13 13:43:49 +0200 | [diff] [blame] | 754 | xfs_log_item_t *lip = bp->b_fspriv; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 755 | xfs_buf_log_item_t *bip; |
| 756 | int chunks; |
| 757 | int map_size; |
Dave Chinner | 372cc85e | 2012-06-22 18:50:12 +1000 | [diff] [blame] | 758 | int error; |
| 759 | int i; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 760 | |
| 761 | /* |
| 762 | * Check to see if there is already a buf log item for |
| 763 | * this buffer. If there is, it is guaranteed to be |
| 764 | * the first. If we do already have one, there is |
| 765 | * nothing to do here so return. |
| 766 | */ |
Dave Chinner | ebad861 | 2010-09-22 10:47:20 +1000 | [diff] [blame] | 767 | ASSERT(bp->b_target->bt_mount == mp); |
Christoph Hellwig | adadbee | 2011-07-13 13:43:49 +0200 | [diff] [blame] | 768 | if (lip != NULL && lip->li_type == XFS_LI_BUF) |
| 769 | return; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 770 | |
Dave Chinner | 372cc85e | 2012-06-22 18:50:12 +1000 | [diff] [blame] | 771 | bip = kmem_zone_zalloc(xfs_buf_item_zone, KM_SLEEP); |
Dave Chinner | 43f5efc | 2010-03-23 10:10:00 +1100 | [diff] [blame] | 772 | 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] | 773 | bip->bli_buf = bp; |
Lachlan McIlroy | e1f5dbd | 2008-09-17 16:52:13 +1000 | [diff] [blame] | 774 | xfs_buf_hold(bp); |
Dave Chinner | 372cc85e | 2012-06-22 18:50:12 +1000 | [diff] [blame] | 775 | |
| 776 | /* |
| 777 | * chunks is the number of XFS_BLF_CHUNK size pieces the buffer |
| 778 | * can be divided into. Make sure not to truncate any pieces. |
| 779 | * map_size is the size of the bitmap needed to describe the |
| 780 | * chunks of the buffer. |
| 781 | * |
| 782 | * Discontiguous buffer support follows the layout of the underlying |
| 783 | * buffer. This makes the implementation as simple as possible. |
| 784 | */ |
| 785 | error = xfs_buf_item_get_format(bip, bp->b_map_count); |
| 786 | ASSERT(error == 0); |
| 787 | |
| 788 | for (i = 0; i < bip->bli_format_count; i++) { |
| 789 | chunks = DIV_ROUND_UP(BBTOB(bp->b_maps[i].bm_len), |
| 790 | XFS_BLF_CHUNK); |
| 791 | map_size = DIV_ROUND_UP(chunks, NBWORD); |
| 792 | |
| 793 | bip->bli_formats[i].blf_type = XFS_LI_BUF; |
| 794 | bip->bli_formats[i].blf_blkno = bp->b_maps[i].bm_bn; |
| 795 | bip->bli_formats[i].blf_len = bp->b_maps[i].bm_len; |
| 796 | bip->bli_formats[i].blf_map_size = map_size; |
| 797 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 798 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 799 | /* |
| 800 | * Put the buf item into the list of items attached to the |
| 801 | * buffer at the front. |
| 802 | */ |
Christoph Hellwig | adadbee | 2011-07-13 13:43:49 +0200 | [diff] [blame] | 803 | if (bp->b_fspriv) |
| 804 | bip->bli_item.li_bio_list = bp->b_fspriv; |
| 805 | bp->b_fspriv = bip; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 806 | } |
| 807 | |
| 808 | |
| 809 | /* |
| 810 | * Mark bytes first through last inclusive as dirty in the buf |
| 811 | * item's bitmap. |
| 812 | */ |
Dave Chinner | 632b89e | 2013-10-29 22:11:58 +1100 | [diff] [blame] | 813 | static void |
Dave Chinner | 372cc85e | 2012-06-22 18:50:12 +1000 | [diff] [blame] | 814 | xfs_buf_item_log_segment( |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 815 | uint first, |
Dave Chinner | 372cc85e | 2012-06-22 18:50:12 +1000 | [diff] [blame] | 816 | uint last, |
| 817 | uint *map) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 818 | { |
| 819 | uint first_bit; |
| 820 | uint last_bit; |
| 821 | uint bits_to_set; |
| 822 | uint bits_set; |
| 823 | uint word_num; |
| 824 | uint *wordp; |
| 825 | uint bit; |
| 826 | uint end_bit; |
| 827 | uint mask; |
| 828 | |
| 829 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 830 | * Convert byte offsets to bit numbers. |
| 831 | */ |
Dave Chinner | c115541 | 2010-05-07 11:05:19 +1000 | [diff] [blame] | 832 | first_bit = first >> XFS_BLF_SHIFT; |
| 833 | last_bit = last >> XFS_BLF_SHIFT; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 834 | |
| 835 | /* |
| 836 | * Calculate the total number of bits to be set. |
| 837 | */ |
| 838 | bits_to_set = last_bit - first_bit + 1; |
| 839 | |
| 840 | /* |
| 841 | * Get a pointer to the first word in the bitmap |
| 842 | * to set a bit in. |
| 843 | */ |
| 844 | word_num = first_bit >> BIT_TO_WORD_SHIFT; |
Dave Chinner | 372cc85e | 2012-06-22 18:50:12 +1000 | [diff] [blame] | 845 | wordp = &map[word_num]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 846 | |
| 847 | /* |
| 848 | * Calculate the starting bit in the first word. |
| 849 | */ |
| 850 | bit = first_bit & (uint)(NBWORD - 1); |
| 851 | |
| 852 | /* |
| 853 | * First set any bits in the first word of our range. |
| 854 | * If it starts at bit 0 of the word, it will be |
| 855 | * set below rather than here. That is what the variable |
| 856 | * bit tells us. The variable bits_set tracks the number |
| 857 | * of bits that have been set so far. End_bit is the number |
| 858 | * of the last bit to be set in this word plus one. |
| 859 | */ |
| 860 | if (bit) { |
| 861 | end_bit = MIN(bit + bits_to_set, (uint)NBWORD); |
| 862 | mask = ((1 << (end_bit - bit)) - 1) << bit; |
| 863 | *wordp |= mask; |
| 864 | wordp++; |
| 865 | bits_set = end_bit - bit; |
| 866 | } else { |
| 867 | bits_set = 0; |
| 868 | } |
| 869 | |
| 870 | /* |
| 871 | * Now set bits a whole word at a time that are between |
| 872 | * first_bit and last_bit. |
| 873 | */ |
| 874 | while ((bits_to_set - bits_set) >= NBWORD) { |
| 875 | *wordp |= 0xffffffff; |
| 876 | bits_set += NBWORD; |
| 877 | wordp++; |
| 878 | } |
| 879 | |
| 880 | /* |
| 881 | * Finally, set any bits left to be set in one last partial word. |
| 882 | */ |
| 883 | end_bit = bits_to_set - bits_set; |
| 884 | if (end_bit) { |
| 885 | mask = (1 << end_bit) - 1; |
| 886 | *wordp |= mask; |
| 887 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 888 | } |
| 889 | |
Dave Chinner | 372cc85e | 2012-06-22 18:50:12 +1000 | [diff] [blame] | 890 | /* |
| 891 | * Mark bytes first through last inclusive as dirty in the buf |
| 892 | * item's bitmap. |
| 893 | */ |
| 894 | void |
| 895 | xfs_buf_item_log( |
| 896 | xfs_buf_log_item_t *bip, |
| 897 | uint first, |
| 898 | uint last) |
| 899 | { |
| 900 | int i; |
| 901 | uint start; |
| 902 | uint end; |
| 903 | struct xfs_buf *bp = bip->bli_buf; |
| 904 | |
| 905 | /* |
Dave Chinner | 372cc85e | 2012-06-22 18:50:12 +1000 | [diff] [blame] | 906 | * walk each buffer segment and mark them dirty appropriately. |
| 907 | */ |
| 908 | start = 0; |
| 909 | for (i = 0; i < bip->bli_format_count; i++) { |
| 910 | if (start > last) |
| 911 | break; |
| 912 | end = start + BBTOB(bp->b_maps[i].bm_len); |
| 913 | if (first > end) { |
| 914 | start += BBTOB(bp->b_maps[i].bm_len); |
| 915 | continue; |
| 916 | } |
| 917 | if (first < start) |
| 918 | first = start; |
| 919 | if (end > last) |
| 920 | end = last; |
| 921 | |
Eric Sandeen | 72b0636 | 2014-04-14 19:01:34 +1000 | [diff] [blame] | 922 | xfs_buf_item_log_segment(first, end, |
Dave Chinner | 372cc85e | 2012-06-22 18:50:12 +1000 | [diff] [blame] | 923 | &bip->bli_formats[i].blf_data_map[0]); |
| 924 | |
| 925 | start += bp->b_maps[i].bm_len; |
| 926 | } |
| 927 | } |
| 928 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 929 | |
| 930 | /* |
Dave Chinner | 5f6bed7 | 2013-06-27 16:04:52 +1000 | [diff] [blame] | 931 | * Return 1 if the buffer has been logged or ordered in a transaction (at any |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 932 | * point, not just the current transaction) and 0 if not. |
| 933 | */ |
| 934 | uint |
| 935 | xfs_buf_item_dirty( |
| 936 | xfs_buf_log_item_t *bip) |
| 937 | { |
| 938 | return (bip->bli_flags & XFS_BLI_DIRTY); |
| 939 | } |
| 940 | |
Lachlan McIlroy | e1f5dbd | 2008-09-17 16:52:13 +1000 | [diff] [blame] | 941 | STATIC void |
| 942 | xfs_buf_item_free( |
| 943 | xfs_buf_log_item_t *bip) |
| 944 | { |
Dave Chinner | 372cc85e | 2012-06-22 18:50:12 +1000 | [diff] [blame] | 945 | xfs_buf_item_free_format(bip); |
Lachlan McIlroy | e1f5dbd | 2008-09-17 16:52:13 +1000 | [diff] [blame] | 946 | kmem_zone_free(xfs_buf_item_zone, bip); |
| 947 | } |
| 948 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 949 | /* |
| 950 | * This is called when the buf log item is no longer needed. It should |
| 951 | * free the buf log item associated with the given buffer and clear |
| 952 | * the buffer's pointer to the buf log item. If there are no more |
| 953 | * items in the list, clear the b_iodone field of the buffer (see |
| 954 | * xfs_buf_attach_iodone() below). |
| 955 | */ |
| 956 | void |
| 957 | xfs_buf_item_relse( |
| 958 | xfs_buf_t *bp) |
| 959 | { |
Dave Chinner | 5f6bed7 | 2013-06-27 16:04:52 +1000 | [diff] [blame] | 960 | xfs_buf_log_item_t *bip = bp->b_fspriv; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 961 | |
Christoph Hellwig | 0b1b213 | 2009-12-14 23:14:59 +0000 | [diff] [blame] | 962 | trace_xfs_buf_item_relse(bp, _RET_IP_); |
Dave Chinner | 5f6bed7 | 2013-06-27 16:04:52 +1000 | [diff] [blame] | 963 | ASSERT(!(bip->bli_item.li_flags & XFS_LI_IN_AIL)); |
Christoph Hellwig | 0b1b213 | 2009-12-14 23:14:59 +0000 | [diff] [blame] | 964 | |
Christoph Hellwig | adadbee | 2011-07-13 13:43:49 +0200 | [diff] [blame] | 965 | bp->b_fspriv = bip->bli_item.li_bio_list; |
Christoph Hellwig | cb669ca | 2011-07-13 13:43:49 +0200 | [diff] [blame] | 966 | if (bp->b_fspriv == NULL) |
| 967 | bp->b_iodone = NULL; |
Christoph Hellwig | adadbee | 2011-07-13 13:43:49 +0200 | [diff] [blame] | 968 | |
Lachlan McIlroy | e1f5dbd | 2008-09-17 16:52:13 +1000 | [diff] [blame] | 969 | xfs_buf_rele(bp); |
| 970 | xfs_buf_item_free(bip); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 971 | } |
| 972 | |
| 973 | |
| 974 | /* |
| 975 | * Add the given log item with its callback to the list of callbacks |
| 976 | * to be called when the buffer's I/O completes. If it is not set |
| 977 | * already, set the buffer's b_iodone() routine to be |
| 978 | * xfs_buf_iodone_callbacks() and link the log item into the list of |
| 979 | * items rooted at b_fsprivate. Items are always added as the second |
| 980 | * entry in the list if there is a first, because the buf item code |
| 981 | * assumes that the buf log item is first. |
| 982 | */ |
| 983 | void |
| 984 | xfs_buf_attach_iodone( |
| 985 | xfs_buf_t *bp, |
| 986 | void (*cb)(xfs_buf_t *, xfs_log_item_t *), |
| 987 | xfs_log_item_t *lip) |
| 988 | { |
| 989 | xfs_log_item_t *head_lip; |
| 990 | |
Christoph Hellwig | 0c842ad | 2011-07-08 14:36:19 +0200 | [diff] [blame] | 991 | ASSERT(xfs_buf_islocked(bp)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 992 | |
| 993 | lip->li_cb = cb; |
Christoph Hellwig | adadbee | 2011-07-13 13:43:49 +0200 | [diff] [blame] | 994 | head_lip = bp->b_fspriv; |
| 995 | if (head_lip) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 996 | lip->li_bio_list = head_lip->li_bio_list; |
| 997 | head_lip->li_bio_list = lip; |
| 998 | } else { |
Christoph Hellwig | adadbee | 2011-07-13 13:43:49 +0200 | [diff] [blame] | 999 | bp->b_fspriv = lip; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1000 | } |
| 1001 | |
Christoph Hellwig | cb669ca | 2011-07-13 13:43:49 +0200 | [diff] [blame] | 1002 | ASSERT(bp->b_iodone == NULL || |
| 1003 | bp->b_iodone == xfs_buf_iodone_callbacks); |
| 1004 | bp->b_iodone = xfs_buf_iodone_callbacks; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1005 | } |
| 1006 | |
Dave Chinner | c90821a | 2010-12-03 17:00:52 +1100 | [diff] [blame] | 1007 | /* |
| 1008 | * We can have many callbacks on a buffer. Running the callbacks individually |
| 1009 | * can cause a lot of contention on the AIL lock, so we allow for a single |
| 1010 | * callback to be able to scan the remaining lip->li_bio_list for other items |
| 1011 | * of the same type and callback to be processed in the first call. |
| 1012 | * |
| 1013 | * As a result, the loop walking the callback list below will also modify the |
| 1014 | * list. it removes the first item from the list and then runs the callback. |
| 1015 | * The loop then restarts from the new head of the list. This allows the |
| 1016 | * callback to scan and modify the list attached to the buffer and we don't |
| 1017 | * have to care about maintaining a next item pointer. |
| 1018 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1019 | STATIC void |
| 1020 | xfs_buf_do_callbacks( |
Dave Chinner | c90821a | 2010-12-03 17:00:52 +1100 | [diff] [blame] | 1021 | struct xfs_buf *bp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1022 | { |
Dave Chinner | c90821a | 2010-12-03 17:00:52 +1100 | [diff] [blame] | 1023 | struct xfs_log_item *lip; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1024 | |
Christoph Hellwig | adadbee | 2011-07-13 13:43:49 +0200 | [diff] [blame] | 1025 | while ((lip = bp->b_fspriv) != NULL) { |
| 1026 | bp->b_fspriv = lip->li_bio_list; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1027 | ASSERT(lip->li_cb != NULL); |
| 1028 | /* |
| 1029 | * Clear the next pointer so we don't have any |
| 1030 | * confusion if the item is added to another buf. |
| 1031 | * Don't touch the log item after calling its |
| 1032 | * callback, because it could have freed itself. |
| 1033 | */ |
| 1034 | lip->li_bio_list = NULL; |
| 1035 | lip->li_cb(bp, lip); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1036 | } |
| 1037 | } |
| 1038 | |
| 1039 | /* |
| 1040 | * This is the iodone() function for buffers which have had callbacks |
| 1041 | * attached to them by xfs_buf_attach_iodone(). It should remove each |
| 1042 | * log item from the buffer's list and call the callback of each in turn. |
| 1043 | * When done, the buffer's fsprivate field is set to NULL and the buffer |
| 1044 | * is unlocked with a call to iodone(). |
| 1045 | */ |
| 1046 | void |
| 1047 | xfs_buf_iodone_callbacks( |
Christoph Hellwig | bfc6017 | 2011-01-07 13:02:23 +0000 | [diff] [blame] | 1048 | struct xfs_buf *bp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1049 | { |
Christoph Hellwig | bfc6017 | 2011-01-07 13:02:23 +0000 | [diff] [blame] | 1050 | struct xfs_log_item *lip = bp->b_fspriv; |
| 1051 | struct xfs_mount *mp = lip->li_mountp; |
| 1052 | static ulong lasttime; |
| 1053 | static xfs_buftarg_t *lasttarg; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1054 | |
Dave Chinner | 36de955 | 2014-06-06 16:02:12 +1000 | [diff] [blame] | 1055 | if (likely(!bp->b_error)) |
Christoph Hellwig | bfc6017 | 2011-01-07 13:02:23 +0000 | [diff] [blame] | 1056 | goto do_callbacks; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1057 | |
Christoph Hellwig | bfc6017 | 2011-01-07 13:02:23 +0000 | [diff] [blame] | 1058 | /* |
| 1059 | * If we've already decided to shutdown the filesystem because of |
| 1060 | * I/O errors, there's no point in giving this a retry. |
| 1061 | */ |
| 1062 | if (XFS_FORCED_SHUTDOWN(mp)) { |
Christoph Hellwig | c867cb6 | 2011-10-10 16:52:46 +0000 | [diff] [blame] | 1063 | xfs_buf_stale(bp); |
Christoph Hellwig | c867cb6 | 2011-10-10 16:52:46 +0000 | [diff] [blame] | 1064 | XFS_BUF_DONE(bp); |
Christoph Hellwig | bfc6017 | 2011-01-07 13:02:23 +0000 | [diff] [blame] | 1065 | trace_xfs_buf_item_iodone(bp, _RET_IP_); |
| 1066 | goto do_callbacks; |
| 1067 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1068 | |
Chandra Seetharaman | 49074c0 | 2011-07-22 23:40:40 +0000 | [diff] [blame] | 1069 | if (bp->b_target != lasttarg || |
Christoph Hellwig | bfc6017 | 2011-01-07 13:02:23 +0000 | [diff] [blame] | 1070 | time_after(jiffies, (lasttime + 5*HZ))) { |
| 1071 | lasttime = jiffies; |
Christoph Hellwig | b38505b | 2011-10-10 16:52:50 +0000 | [diff] [blame] | 1072 | xfs_buf_ioerror_alert(bp, __func__); |
Christoph Hellwig | bfc6017 | 2011-01-07 13:02:23 +0000 | [diff] [blame] | 1073 | } |
Chandra Seetharaman | 49074c0 | 2011-07-22 23:40:40 +0000 | [diff] [blame] | 1074 | lasttarg = bp->b_target; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1075 | |
Christoph Hellwig | bfc6017 | 2011-01-07 13:02:23 +0000 | [diff] [blame] | 1076 | /* |
Lucas De Marchi | 25985ed | 2011-03-30 22:57:33 -0300 | [diff] [blame] | 1077 | * 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] | 1078 | * error. Clear the error state and write the buffer out again. |
| 1079 | * |
Christoph Hellwig | 43ff212 | 2012-04-23 15:58:39 +1000 | [diff] [blame] | 1080 | * XXX: This helps against transient write errors, but we need to find |
| 1081 | * a way to shut the filesystem down if the writes keep failing. |
| 1082 | * |
| 1083 | * In practice we'll shut the filesystem down soon as non-transient |
| 1084 | * erorrs tend to affect the whole device and a failing log write |
| 1085 | * will make us give up. But we really ought to do better here. |
Christoph Hellwig | bfc6017 | 2011-01-07 13:02:23 +0000 | [diff] [blame] | 1086 | */ |
| 1087 | if (XFS_BUF_ISASYNC(bp)) { |
Christoph Hellwig | 43ff212 | 2012-04-23 15:58:39 +1000 | [diff] [blame] | 1088 | ASSERT(bp->b_iodone != NULL); |
| 1089 | |
| 1090 | trace_xfs_buf_item_iodone_async(bp, _RET_IP_); |
| 1091 | |
Chandra Seetharaman | 5a52c2a58 | 2011-07-22 23:39:51 +0000 | [diff] [blame] | 1092 | xfs_buf_ioerror(bp, 0); /* errno of 0 unsets the flag */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1093 | |
Dave Chinner | ac8809f | 2013-12-12 16:34:38 +1100 | [diff] [blame] | 1094 | if (!(bp->b_flags & (XBF_STALE|XBF_WRITE_FAIL))) { |
| 1095 | bp->b_flags |= XBF_WRITE | XBF_ASYNC | |
| 1096 | XBF_DONE | XBF_WRITE_FAIL; |
Christoph Hellwig | a2dcf5d | 2012-07-13 02:24:10 -0400 | [diff] [blame] | 1097 | xfs_buf_iorequest(bp); |
Christoph Hellwig | 43ff212 | 2012-04-23 15:58:39 +1000 | [diff] [blame] | 1098 | } else { |
| 1099 | xfs_buf_relse(bp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1100 | } |
Christoph Hellwig | 43ff212 | 2012-04-23 15:58:39 +1000 | [diff] [blame] | 1101 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1102 | return; |
| 1103 | } |
Christoph Hellwig | 0b1b213 | 2009-12-14 23:14:59 +0000 | [diff] [blame] | 1104 | |
Christoph Hellwig | bfc6017 | 2011-01-07 13:02:23 +0000 | [diff] [blame] | 1105 | /* |
| 1106 | * If the write of the buffer was synchronous, we want to make |
| 1107 | * sure to return the error to the caller of xfs_bwrite(). |
| 1108 | */ |
Christoph Hellwig | c867cb6 | 2011-10-10 16:52:46 +0000 | [diff] [blame] | 1109 | xfs_buf_stale(bp); |
Christoph Hellwig | bfc6017 | 2011-01-07 13:02:23 +0000 | [diff] [blame] | 1110 | XFS_BUF_DONE(bp); |
Christoph Hellwig | bfc6017 | 2011-01-07 13:02:23 +0000 | [diff] [blame] | 1111 | |
| 1112 | trace_xfs_buf_error_relse(bp, _RET_IP_); |
Christoph Hellwig | bfc6017 | 2011-01-07 13:02:23 +0000 | [diff] [blame] | 1113 | |
| 1114 | do_callbacks: |
Dave Chinner | c90821a | 2010-12-03 17:00:52 +1100 | [diff] [blame] | 1115 | xfs_buf_do_callbacks(bp); |
Christoph Hellwig | adadbee | 2011-07-13 13:43:49 +0200 | [diff] [blame] | 1116 | bp->b_fspriv = NULL; |
Christoph Hellwig | cb669ca | 2011-07-13 13:43:49 +0200 | [diff] [blame] | 1117 | bp->b_iodone = NULL; |
Christoph Hellwig | 1a1a3e9 | 2010-10-06 18:41:18 +0000 | [diff] [blame] | 1118 | xfs_buf_ioend(bp, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1119 | } |
| 1120 | |
| 1121 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1122 | * This is the iodone() function for buffers which have been |
| 1123 | * logged. It is called when they are eventually flushed out. |
| 1124 | * It should remove the buf item from the AIL, and free the buf item. |
| 1125 | * It is called by xfs_buf_iodone_callbacks() above which will take |
| 1126 | * care of cleaning up the buffer itself. |
| 1127 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1128 | void |
| 1129 | xfs_buf_iodone( |
Christoph Hellwig | ca30b2a | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 1130 | struct xfs_buf *bp, |
| 1131 | struct xfs_log_item *lip) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1132 | { |
Christoph Hellwig | ca30b2a | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 1133 | struct xfs_ail *ailp = lip->li_ailp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1134 | |
Christoph Hellwig | ca30b2a | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 1135 | ASSERT(BUF_ITEM(lip)->bli_buf == bp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1136 | |
Lachlan McIlroy | e1f5dbd | 2008-09-17 16:52:13 +1000 | [diff] [blame] | 1137 | xfs_buf_rele(bp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1138 | |
| 1139 | /* |
| 1140 | * If we are forcibly shutting down, this may well be |
| 1141 | * off the AIL already. That's because we simulate the |
| 1142 | * log-committed callbacks to unpin these buffers. Or we may never |
| 1143 | * have put this item on AIL because of the transaction was |
David Chinner | 783a2f6 | 2008-10-30 17:39:58 +1100 | [diff] [blame] | 1144 | * aborted forcibly. xfs_trans_ail_delete() takes care of these. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1145 | * |
| 1146 | * Either way, AIL is useless if we're forcing a shutdown. |
| 1147 | */ |
David Chinner | fc1829f | 2008-10-30 17:39:46 +1100 | [diff] [blame] | 1148 | spin_lock(&ailp->xa_lock); |
Dave Chinner | 04913fd | 2012-04-23 15:58:41 +1000 | [diff] [blame] | 1149 | xfs_trans_ail_delete(ailp, lip, SHUTDOWN_CORRUPT_INCORE); |
Christoph Hellwig | ca30b2a | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 1150 | xfs_buf_item_free(BUF_ITEM(lip)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1151 | } |