Dave Chinner | 71e330b | 2010-05-21 14:37:18 +1000 | [diff] [blame] | 1 | /* |
| 2 | * Copyright (c) 2010 Red Hat, Inc. All Rights Reserved. |
| 3 | * |
| 4 | * This program is free software; you can redistribute it and/or |
| 5 | * modify it under the terms of the GNU General Public License as |
| 6 | * published by the Free Software Foundation. |
| 7 | * |
| 8 | * This program is distributed in the hope that it would be useful, |
| 9 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 11 | * GNU General Public License for more details. |
| 12 | * |
| 13 | * You should have received a copy of the GNU General Public License |
| 14 | * along with this program; if not, write the Free Software Foundation, |
| 15 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA |
| 16 | */ |
| 17 | |
| 18 | #include "xfs.h" |
| 19 | #include "xfs_fs.h" |
| 20 | #include "xfs_types.h" |
Dave Chinner | 71e330b | 2010-05-21 14:37:18 +1000 | [diff] [blame] | 21 | #include "xfs_log.h" |
Dave Chinner | 71e330b | 2010-05-21 14:37:18 +1000 | [diff] [blame] | 22 | #include "xfs_trans.h" |
| 23 | #include "xfs_trans_priv.h" |
| 24 | #include "xfs_log_priv.h" |
| 25 | #include "xfs_sb.h" |
| 26 | #include "xfs_ag.h" |
Dave Chinner | 71e330b | 2010-05-21 14:37:18 +1000 | [diff] [blame] | 27 | #include "xfs_mount.h" |
| 28 | #include "xfs_error.h" |
| 29 | #include "xfs_alloc.h" |
Dave Chinner | efc27b5 | 2012-04-29 10:39:43 +0000 | [diff] [blame] | 30 | #include "xfs_extent_busy.h" |
Christoph Hellwig | e84661a | 2011-05-20 13:45:32 +0000 | [diff] [blame] | 31 | #include "xfs_discard.h" |
Dave Chinner | 71e330b | 2010-05-21 14:37:18 +1000 | [diff] [blame] | 32 | |
| 33 | /* |
Dave Chinner | 71e330b | 2010-05-21 14:37:18 +1000 | [diff] [blame] | 34 | * Allocate a new ticket. Failing to get a new ticket makes it really hard to |
| 35 | * recover, so we don't allow failure here. Also, we allocate in a context that |
| 36 | * we don't want to be issuing transactions from, so we need to tell the |
| 37 | * allocation code this as well. |
| 38 | * |
| 39 | * We don't reserve any space for the ticket - we are going to steal whatever |
| 40 | * space we require from transactions as they commit. To ensure we reserve all |
| 41 | * the space required, we need to set the current reservation of the ticket to |
| 42 | * zero so that we know to steal the initial transaction overhead from the |
| 43 | * first transaction commit. |
| 44 | */ |
| 45 | static struct xlog_ticket * |
| 46 | xlog_cil_ticket_alloc( |
Mark Tinguely | f7bdf03 | 2012-06-14 09:22:15 -0500 | [diff] [blame] | 47 | struct xlog *log) |
Dave Chinner | 71e330b | 2010-05-21 14:37:18 +1000 | [diff] [blame] | 48 | { |
| 49 | struct xlog_ticket *tic; |
| 50 | |
| 51 | tic = xlog_ticket_alloc(log, 0, 1, XFS_TRANSACTION, 0, |
| 52 | KM_SLEEP|KM_NOFS); |
| 53 | tic->t_trans_type = XFS_TRANS_CHECKPOINT; |
| 54 | |
| 55 | /* |
| 56 | * set the current reservation to zero so we know to steal the basic |
| 57 | * transaction overhead reservation from the first transaction commit. |
| 58 | */ |
| 59 | tic->t_curr_res = 0; |
| 60 | return tic; |
| 61 | } |
| 62 | |
| 63 | /* |
| 64 | * After the first stage of log recovery is done, we know where the head and |
| 65 | * tail of the log are. We need this log initialisation done before we can |
| 66 | * initialise the first CIL checkpoint context. |
| 67 | * |
| 68 | * Here we allocate a log ticket to track space usage during a CIL push. This |
| 69 | * ticket is passed to xlog_write() directly so that we don't slowly leak log |
| 70 | * space by failing to account for space used by log headers and additional |
| 71 | * region headers for split regions. |
| 72 | */ |
| 73 | void |
| 74 | xlog_cil_init_post_recovery( |
Mark Tinguely | f7bdf03 | 2012-06-14 09:22:15 -0500 | [diff] [blame] | 75 | struct xlog *log) |
Dave Chinner | 71e330b | 2010-05-21 14:37:18 +1000 | [diff] [blame] | 76 | { |
Dave Chinner | 71e330b | 2010-05-21 14:37:18 +1000 | [diff] [blame] | 77 | log->l_cilp->xc_ctx->ticket = xlog_cil_ticket_alloc(log); |
| 78 | log->l_cilp->xc_ctx->sequence = 1; |
| 79 | log->l_cilp->xc_ctx->commit_lsn = xlog_assign_lsn(log->l_curr_cycle, |
| 80 | log->l_curr_block); |
| 81 | } |
| 82 | |
Dave Chinner | 7492c5b | 2013-08-12 20:50:05 +1000 | [diff] [blame^] | 83 | STATIC int |
| 84 | xlog_cil_lv_item_format( |
| 85 | struct xfs_log_item *lip, |
| 86 | struct xfs_log_vec *lv) |
| 87 | { |
| 88 | int index; |
| 89 | char *ptr; |
| 90 | |
| 91 | /* format new vectors into array */ |
| 92 | lip->li_ops->iop_format(lip, lv->lv_iovecp); |
| 93 | |
| 94 | /* copy data into existing array */ |
| 95 | ptr = lv->lv_buf; |
| 96 | for (index = 0; index < lv->lv_niovecs; index++) { |
| 97 | struct xfs_log_iovec *vec = &lv->lv_iovecp[index]; |
| 98 | |
| 99 | memcpy(ptr, vec->i_addr, vec->i_len); |
| 100 | vec->i_addr = ptr; |
| 101 | ptr += vec->i_len; |
| 102 | } |
| 103 | |
| 104 | /* |
| 105 | * some size calculations for log vectors over-estimate, so the caller |
| 106 | * doesn't know the amount of space actually used by the item. Return |
| 107 | * the byte count to the caller so they can check and store it |
| 108 | * appropriately. |
| 109 | */ |
| 110 | return ptr - lv->lv_buf; |
| 111 | } |
| 112 | |
Dave Chinner | 71e330b | 2010-05-21 14:37:18 +1000 | [diff] [blame] | 113 | /* |
Dave Chinner | 71e330b | 2010-05-21 14:37:18 +1000 | [diff] [blame] | 114 | * Format log item into a flat buffers |
| 115 | * |
| 116 | * For delayed logging, we need to hold a formatted buffer containing all the |
| 117 | * changes on the log item. This enables us to relog the item in memory and |
| 118 | * write it out asynchronously without needing to relock the object that was |
| 119 | * modified at the time it gets written into the iclog. |
| 120 | * |
| 121 | * This function builds a vector for the changes in each log item in the |
| 122 | * transaction. It then works out the length of the buffer needed for each log |
| 123 | * item, allocates them and formats the vector for the item into the buffer. |
| 124 | * The buffer is then attached to the log item are then inserted into the |
| 125 | * Committed Item List for tracking until the next checkpoint is written out. |
| 126 | * |
| 127 | * We don't set up region headers during this process; we simply copy the |
| 128 | * regions into the flat buffer. We can do this because we still have to do a |
| 129 | * formatting step to write the regions into the iclog buffer. Writing the |
| 130 | * ophdrs during the iclog write means that we can support splitting large |
| 131 | * regions across iclog boundares without needing a change in the format of the |
| 132 | * item/region encapsulation. |
| 133 | * |
| 134 | * Hence what we need to do now is change the rewrite the vector array to point |
| 135 | * to the copied region inside the buffer we just allocated. This allows us to |
| 136 | * format the regions into the iclog as though they are being formatted |
| 137 | * directly out of the objects themselves. |
| 138 | */ |
Christoph Hellwig | 0244b96 | 2011-12-06 21:58:08 +0000 | [diff] [blame] | 139 | static struct xfs_log_vec * |
| 140 | xlog_cil_prepare_log_vecs( |
| 141 | struct xfs_trans *tp) |
Dave Chinner | 71e330b | 2010-05-21 14:37:18 +1000 | [diff] [blame] | 142 | { |
Christoph Hellwig | 0244b96 | 2011-12-06 21:58:08 +0000 | [diff] [blame] | 143 | struct xfs_log_item_desc *lidp; |
Dave Chinner | 7492c5b | 2013-08-12 20:50:05 +1000 | [diff] [blame^] | 144 | struct xfs_log_vec *prev_lv = NULL; |
Christoph Hellwig | 0244b96 | 2011-12-06 21:58:08 +0000 | [diff] [blame] | 145 | struct xfs_log_vec *ret_lv = NULL; |
Dave Chinner | 71e330b | 2010-05-21 14:37:18 +1000 | [diff] [blame] | 146 | |
Christoph Hellwig | 0244b96 | 2011-12-06 21:58:08 +0000 | [diff] [blame] | 147 | |
| 148 | /* Bail out if we didn't find a log item. */ |
| 149 | if (list_empty(&tp->t_items)) { |
| 150 | ASSERT(0); |
| 151 | return NULL; |
| 152 | } |
| 153 | |
| 154 | list_for_each_entry(lidp, &tp->t_items, lid_trans) { |
Dave Chinner | 166d136 | 2013-08-12 20:50:04 +1000 | [diff] [blame] | 155 | struct xfs_log_item *lip = lidp->lid_item; |
Dave Chinner | 7492c5b | 2013-08-12 20:50:05 +1000 | [diff] [blame^] | 156 | struct xfs_log_vec *lv; |
| 157 | int niovecs = 0; |
| 158 | int nbytes = 0; |
| 159 | int buf_size; |
Dave Chinner | fd63875 | 2013-06-27 16:04:51 +1000 | [diff] [blame] | 160 | bool ordered = false; |
Dave Chinner | 71e330b | 2010-05-21 14:37:18 +1000 | [diff] [blame] | 161 | |
Christoph Hellwig | 0244b96 | 2011-12-06 21:58:08 +0000 | [diff] [blame] | 162 | /* Skip items which aren't dirty in this transaction. */ |
| 163 | if (!(lidp->lid_flags & XFS_LID_DIRTY)) |
| 164 | continue; |
| 165 | |
Dave Chinner | 166d136 | 2013-08-12 20:50:04 +1000 | [diff] [blame] | 166 | /* get number of vecs and size of data to be stored */ |
| 167 | lip->li_ops->iop_size(lip, &niovecs, &nbytes); |
| 168 | |
Christoph Hellwig | 0244b96 | 2011-12-06 21:58:08 +0000 | [diff] [blame] | 169 | /* Skip items that do not have any vectors for writing */ |
Christoph Hellwig | b393421 | 2011-12-06 21:58:09 +0000 | [diff] [blame] | 170 | if (!niovecs) |
Christoph Hellwig | 0244b96 | 2011-12-06 21:58:08 +0000 | [diff] [blame] | 171 | continue; |
| 172 | |
Dave Chinner | fd63875 | 2013-06-27 16:04:51 +1000 | [diff] [blame] | 173 | /* |
| 174 | * Ordered items need to be tracked but we do not wish to write |
| 175 | * them. We need a logvec to track the object, but we do not |
| 176 | * need an iovec or buffer to be allocated for copying data. |
| 177 | */ |
| 178 | if (niovecs == XFS_LOG_VEC_ORDERED) { |
| 179 | ordered = true; |
| 180 | niovecs = 0; |
Dave Chinner | 7492c5b | 2013-08-12 20:50:05 +1000 | [diff] [blame^] | 181 | nbytes = 0; |
Dave Chinner | fd63875 | 2013-06-27 16:04:51 +1000 | [diff] [blame] | 182 | } |
| 183 | |
Dave Chinner | 7492c5b | 2013-08-12 20:50:05 +1000 | [diff] [blame^] | 184 | /* calc buffer size */ |
| 185 | buf_size = sizeof(struct xfs_log_vec) + nbytes + |
| 186 | niovecs * sizeof(struct xfs_log_iovec); |
Christoph Hellwig | 0244b96 | 2011-12-06 21:58:08 +0000 | [diff] [blame] | 187 | |
Dave Chinner | 7492c5b | 2013-08-12 20:50:05 +1000 | [diff] [blame^] | 188 | /* allocate new data chunk */ |
| 189 | lv = kmem_zalloc(buf_size, KM_SLEEP|KM_NOFS); |
| 190 | lv->lv_item = lip; |
| 191 | lv->lv_size = buf_size; |
| 192 | lv->lv_niovecs = niovecs; |
Dave Chinner | fd63875 | 2013-06-27 16:04:51 +1000 | [diff] [blame] | 193 | if (ordered) { |
| 194 | /* track as an ordered logvec */ |
Dave Chinner | 7492c5b | 2013-08-12 20:50:05 +1000 | [diff] [blame^] | 195 | ASSERT(lip->li_lv == NULL); |
| 196 | lv->lv_buf_len = XFS_LOG_VEC_ORDERED; |
| 197 | goto insert; |
Dave Chinner | fd63875 | 2013-06-27 16:04:51 +1000 | [diff] [blame] | 198 | } |
| 199 | |
Christoph Hellwig | 0244b96 | 2011-12-06 21:58:08 +0000 | [diff] [blame] | 200 | /* The allocated iovec region lies beyond the log vector. */ |
Dave Chinner | 7492c5b | 2013-08-12 20:50:05 +1000 | [diff] [blame^] | 201 | lv->lv_iovecp = (struct xfs_log_iovec *)&lv[1]; |
Christoph Hellwig | 0244b96 | 2011-12-06 21:58:08 +0000 | [diff] [blame] | 202 | |
Dave Chinner | 7492c5b | 2013-08-12 20:50:05 +1000 | [diff] [blame^] | 203 | /* The allocated data region lies beyond the iovec region */ |
| 204 | lv->lv_buf = (char *)lv + buf_size - nbytes; |
Dave Chinner | 71e330b | 2010-05-21 14:37:18 +1000 | [diff] [blame] | 205 | |
Dave Chinner | 7492c5b | 2013-08-12 20:50:05 +1000 | [diff] [blame^] | 206 | lv->lv_buf_len = xlog_cil_lv_item_format(lip, lv); |
| 207 | ASSERT(lv->lv_buf_len <= nbytes); |
| 208 | insert: |
Christoph Hellwig | 0244b96 | 2011-12-06 21:58:08 +0000 | [diff] [blame] | 209 | if (!ret_lv) |
Dave Chinner | 7492c5b | 2013-08-12 20:50:05 +1000 | [diff] [blame^] | 210 | ret_lv = lv; |
Christoph Hellwig | 0244b96 | 2011-12-06 21:58:08 +0000 | [diff] [blame] | 211 | else |
Dave Chinner | 7492c5b | 2013-08-12 20:50:05 +1000 | [diff] [blame^] | 212 | prev_lv->lv_next = lv; |
| 213 | prev_lv = lv; |
Dave Chinner | 71e330b | 2010-05-21 14:37:18 +1000 | [diff] [blame] | 214 | } |
Christoph Hellwig | 0244b96 | 2011-12-06 21:58:08 +0000 | [diff] [blame] | 215 | |
| 216 | return ret_lv; |
Dave Chinner | 71e330b | 2010-05-21 14:37:18 +1000 | [diff] [blame] | 217 | } |
| 218 | |
Dave Chinner | d1583a3 | 2010-09-24 18:14:13 +1000 | [diff] [blame] | 219 | /* |
| 220 | * Prepare the log item for insertion into the CIL. Calculate the difference in |
| 221 | * log space and vectors it will consume, and if it is a new item pin it as |
| 222 | * well. |
| 223 | */ |
| 224 | STATIC void |
| 225 | xfs_cil_prepare_item( |
Mark Tinguely | f7bdf03 | 2012-06-14 09:22:15 -0500 | [diff] [blame] | 226 | struct xlog *log, |
Dave Chinner | d1583a3 | 2010-09-24 18:14:13 +1000 | [diff] [blame] | 227 | struct xfs_log_vec *lv, |
| 228 | int *len, |
| 229 | int *diff_iovecs) |
| 230 | { |
| 231 | struct xfs_log_vec *old = lv->lv_item->li_lv; |
| 232 | |
| 233 | if (old) { |
| 234 | /* existing lv on log item, space used is a delta */ |
Dave Chinner | fd63875 | 2013-06-27 16:04:51 +1000 | [diff] [blame] | 235 | ASSERT((old->lv_buf && old->lv_buf_len && old->lv_niovecs) || |
| 236 | old->lv_buf_len == XFS_LOG_VEC_ORDERED); |
| 237 | |
| 238 | /* |
| 239 | * If the new item is ordered, keep the old one that is already |
| 240 | * tracking dirty or ordered regions |
| 241 | */ |
| 242 | if (lv->lv_buf_len == XFS_LOG_VEC_ORDERED) { |
| 243 | ASSERT(!lv->lv_buf); |
| 244 | kmem_free(lv); |
| 245 | return; |
| 246 | } |
Dave Chinner | d1583a3 | 2010-09-24 18:14:13 +1000 | [diff] [blame] | 247 | |
| 248 | *len += lv->lv_buf_len - old->lv_buf_len; |
| 249 | *diff_iovecs += lv->lv_niovecs - old->lv_niovecs; |
Dave Chinner | d1583a3 | 2010-09-24 18:14:13 +1000 | [diff] [blame] | 250 | kmem_free(old); |
| 251 | } else { |
| 252 | /* new lv, must pin the log item */ |
| 253 | ASSERT(!lv->lv_item->li_lv); |
Dave Chinner | d1583a3 | 2010-09-24 18:14:13 +1000 | [diff] [blame] | 254 | |
Dave Chinner | fd63875 | 2013-06-27 16:04:51 +1000 | [diff] [blame] | 255 | if (lv->lv_buf_len != XFS_LOG_VEC_ORDERED) { |
| 256 | *len += lv->lv_buf_len; |
| 257 | *diff_iovecs += lv->lv_niovecs; |
| 258 | } |
Dave Chinner | d1583a3 | 2010-09-24 18:14:13 +1000 | [diff] [blame] | 259 | IOP_PIN(lv->lv_item); |
| 260 | |
| 261 | } |
| 262 | |
| 263 | /* attach new log vector to log item */ |
| 264 | lv->lv_item->li_lv = lv; |
| 265 | |
| 266 | /* |
| 267 | * If this is the first time the item is being committed to the |
| 268 | * CIL, store the sequence number on the log item so we can |
| 269 | * tell in future commits whether this is the first checkpoint |
| 270 | * the item is being committed into. |
| 271 | */ |
| 272 | if (!lv->lv_item->li_seq) |
| 273 | lv->lv_item->li_seq = log->l_cilp->xc_ctx->sequence; |
| 274 | } |
| 275 | |
| 276 | /* |
| 277 | * Insert the log items into the CIL and calculate the difference in space |
| 278 | * consumed by the item. Add the space to the checkpoint ticket and calculate |
| 279 | * if the change requires additional log metadata. If it does, take that space |
Justin P. Mattock | 42b2aa8 | 2011-11-28 20:31:00 -0800 | [diff] [blame] | 280 | * as well. Remove the amount of space we added to the checkpoint ticket from |
Dave Chinner | d1583a3 | 2010-09-24 18:14:13 +1000 | [diff] [blame] | 281 | * the current transaction ticket so that the accounting works out correctly. |
| 282 | */ |
Dave Chinner | 71e330b | 2010-05-21 14:37:18 +1000 | [diff] [blame] | 283 | static void |
Dave Chinner | 3b93c7a | 2010-08-24 11:45:53 +1000 | [diff] [blame] | 284 | xlog_cil_insert_items( |
Mark Tinguely | f7bdf03 | 2012-06-14 09:22:15 -0500 | [diff] [blame] | 285 | struct xlog *log, |
Dave Chinner | 3b93c7a | 2010-08-24 11:45:53 +1000 | [diff] [blame] | 286 | struct xfs_log_vec *log_vector, |
Dave Chinner | d1583a3 | 2010-09-24 18:14:13 +1000 | [diff] [blame] | 287 | struct xlog_ticket *ticket) |
Dave Chinner | 3b93c7a | 2010-08-24 11:45:53 +1000 | [diff] [blame] | 288 | { |
Dave Chinner | d1583a3 | 2010-09-24 18:14:13 +1000 | [diff] [blame] | 289 | struct xfs_cil *cil = log->l_cilp; |
| 290 | struct xfs_cil_ctx *ctx = cil->xc_ctx; |
| 291 | struct xfs_log_vec *lv; |
| 292 | int len = 0; |
| 293 | int diff_iovecs = 0; |
| 294 | int iclog_space; |
Dave Chinner | 3b93c7a | 2010-08-24 11:45:53 +1000 | [diff] [blame] | 295 | |
| 296 | ASSERT(log_vector); |
Dave Chinner | d1583a3 | 2010-09-24 18:14:13 +1000 | [diff] [blame] | 297 | |
| 298 | /* |
| 299 | * Do all the accounting aggregation and switching of log vectors |
| 300 | * around in a separate loop to the insertion of items into the CIL. |
| 301 | * Then we can do a separate loop to update the CIL within a single |
| 302 | * lock/unlock pair. This reduces the number of round trips on the CIL |
| 303 | * lock from O(nr_logvectors) to O(1) and greatly reduces the overall |
| 304 | * hold time for the transaction commit. |
| 305 | * |
| 306 | * If this is the first time the item is being placed into the CIL in |
| 307 | * this context, pin it so it can't be written to disk until the CIL is |
| 308 | * flushed to the iclog and the iclog written to disk. |
| 309 | * |
| 310 | * We can do this safely because the context can't checkpoint until we |
| 311 | * are done so it doesn't matter exactly how we update the CIL. |
| 312 | */ |
Dave Chinner | fd63875 | 2013-06-27 16:04:51 +1000 | [diff] [blame] | 313 | spin_lock(&cil->xc_cil_lock); |
| 314 | for (lv = log_vector; lv; ) { |
| 315 | struct xfs_log_vec *next = lv->lv_next; |
| 316 | |
| 317 | ASSERT(lv->lv_item->li_lv || list_empty(&lv->lv_item->li_cil)); |
| 318 | lv->lv_next = NULL; |
| 319 | |
| 320 | /* |
| 321 | * xfs_cil_prepare_item() may free the lv, so move the item on |
| 322 | * the CIL first. |
| 323 | */ |
| 324 | list_move_tail(&lv->lv_item->li_cil, &cil->xc_cil); |
Dave Chinner | d1583a3 | 2010-09-24 18:14:13 +1000 | [diff] [blame] | 325 | xfs_cil_prepare_item(log, lv, &len, &diff_iovecs); |
Dave Chinner | fd63875 | 2013-06-27 16:04:51 +1000 | [diff] [blame] | 326 | lv = next; |
| 327 | } |
Dave Chinner | d1583a3 | 2010-09-24 18:14:13 +1000 | [diff] [blame] | 328 | |
| 329 | /* account for space used by new iovec headers */ |
| 330 | len += diff_iovecs * sizeof(xlog_op_header_t); |
Dave Chinner | d1583a3 | 2010-09-24 18:14:13 +1000 | [diff] [blame] | 331 | ctx->nvecs += diff_iovecs; |
| 332 | |
| 333 | /* |
| 334 | * Now transfer enough transaction reservation to the context ticket |
| 335 | * for the checkpoint. The context ticket is special - the unit |
| 336 | * reservation has to grow as well as the current reservation as we |
| 337 | * steal from tickets so we can correctly determine the space used |
| 338 | * during the transaction commit. |
| 339 | */ |
| 340 | if (ctx->ticket->t_curr_res == 0) { |
| 341 | /* first commit in checkpoint, steal the header reservation */ |
| 342 | ASSERT(ticket->t_curr_res >= ctx->ticket->t_unit_res + len); |
| 343 | ctx->ticket->t_curr_res = ctx->ticket->t_unit_res; |
| 344 | ticket->t_curr_res -= ctx->ticket->t_unit_res; |
| 345 | } |
| 346 | |
| 347 | /* do we need space for more log record headers? */ |
| 348 | iclog_space = log->l_iclog_size - log->l_iclog_hsize; |
| 349 | if (len > 0 && (ctx->space_used / iclog_space != |
| 350 | (ctx->space_used + len) / iclog_space)) { |
| 351 | int hdrs; |
| 352 | |
| 353 | hdrs = (len + iclog_space - 1) / iclog_space; |
| 354 | /* need to take into account split region headers, too */ |
| 355 | hdrs *= log->l_iclog_hsize + sizeof(struct xlog_op_header); |
| 356 | ctx->ticket->t_unit_res += hdrs; |
| 357 | ctx->ticket->t_curr_res += hdrs; |
| 358 | ticket->t_curr_res -= hdrs; |
| 359 | ASSERT(ticket->t_curr_res >= len); |
| 360 | } |
| 361 | ticket->t_curr_res -= len; |
| 362 | ctx->space_used += len; |
| 363 | |
| 364 | spin_unlock(&cil->xc_cil_lock); |
Dave Chinner | 3b93c7a | 2010-08-24 11:45:53 +1000 | [diff] [blame] | 365 | } |
| 366 | |
| 367 | static void |
Dave Chinner | 71e330b | 2010-05-21 14:37:18 +1000 | [diff] [blame] | 368 | xlog_cil_free_logvec( |
| 369 | struct xfs_log_vec *log_vector) |
| 370 | { |
| 371 | struct xfs_log_vec *lv; |
| 372 | |
| 373 | for (lv = log_vector; lv; ) { |
| 374 | struct xfs_log_vec *next = lv->lv_next; |
Dave Chinner | 71e330b | 2010-05-21 14:37:18 +1000 | [diff] [blame] | 375 | kmem_free(lv); |
| 376 | lv = next; |
| 377 | } |
| 378 | } |
| 379 | |
| 380 | /* |
Dave Chinner | 71e330b | 2010-05-21 14:37:18 +1000 | [diff] [blame] | 381 | * Mark all items committed and clear busy extents. We free the log vector |
| 382 | * chains in a separate pass so that we unpin the log items as quickly as |
| 383 | * possible. |
| 384 | */ |
| 385 | static void |
| 386 | xlog_cil_committed( |
| 387 | void *args, |
| 388 | int abort) |
| 389 | { |
| 390 | struct xfs_cil_ctx *ctx = args; |
Christoph Hellwig | e84661a | 2011-05-20 13:45:32 +0000 | [diff] [blame] | 391 | struct xfs_mount *mp = ctx->cil->xc_log->l_mp; |
Dave Chinner | 71e330b | 2010-05-21 14:37:18 +1000 | [diff] [blame] | 392 | |
Dave Chinner | 0e57f6a | 2010-12-20 12:02:19 +1100 | [diff] [blame] | 393 | xfs_trans_committed_bulk(ctx->cil->xc_log->l_ailp, ctx->lv_chain, |
| 394 | ctx->start_lsn, abort); |
Dave Chinner | 71e330b | 2010-05-21 14:37:18 +1000 | [diff] [blame] | 395 | |
Dave Chinner | 4ecbfe6 | 2012-04-29 10:41:10 +0000 | [diff] [blame] | 396 | xfs_extent_busy_sort(&ctx->busy_extents); |
| 397 | xfs_extent_busy_clear(mp, &ctx->busy_extents, |
Christoph Hellwig | e84661a | 2011-05-20 13:45:32 +0000 | [diff] [blame] | 398 | (mp->m_flags & XFS_MOUNT_DISCARD) && !abort); |
Dave Chinner | 71e330b | 2010-05-21 14:37:18 +1000 | [diff] [blame] | 399 | |
| 400 | spin_lock(&ctx->cil->xc_cil_lock); |
| 401 | list_del(&ctx->committing); |
| 402 | spin_unlock(&ctx->cil->xc_cil_lock); |
| 403 | |
| 404 | xlog_cil_free_logvec(ctx->lv_chain); |
Christoph Hellwig | e84661a | 2011-05-20 13:45:32 +0000 | [diff] [blame] | 405 | |
| 406 | if (!list_empty(&ctx->busy_extents)) { |
| 407 | ASSERT(mp->m_flags & XFS_MOUNT_DISCARD); |
| 408 | |
| 409 | xfs_discard_extents(mp, &ctx->busy_extents); |
Dave Chinner | 4ecbfe6 | 2012-04-29 10:41:10 +0000 | [diff] [blame] | 410 | xfs_extent_busy_clear(mp, &ctx->busy_extents, false); |
Christoph Hellwig | e84661a | 2011-05-20 13:45:32 +0000 | [diff] [blame] | 411 | } |
| 412 | |
Dave Chinner | 71e330b | 2010-05-21 14:37:18 +1000 | [diff] [blame] | 413 | kmem_free(ctx); |
| 414 | } |
| 415 | |
| 416 | /* |
Dave Chinner | a44f13e | 2010-08-24 11:40:03 +1000 | [diff] [blame] | 417 | * Push the Committed Item List to the log. If @push_seq flag is zero, then it |
| 418 | * is a background flush and so we can chose to ignore it. Otherwise, if the |
| 419 | * current sequence is the same as @push_seq we need to do a flush. If |
| 420 | * @push_seq is less than the current sequence, then it has already been |
| 421 | * flushed and we don't need to do anything - the caller will wait for it to |
| 422 | * complete if necessary. |
| 423 | * |
| 424 | * @push_seq is a value rather than a flag because that allows us to do an |
| 425 | * unlocked check of the sequence number for a match. Hence we can allows log |
| 426 | * forces to run racily and not issue pushes for the same sequence twice. If we |
| 427 | * get a race between multiple pushes for the same sequence they will block on |
| 428 | * the first one and then abort, hence avoiding needless pushes. |
Dave Chinner | 71e330b | 2010-05-21 14:37:18 +1000 | [diff] [blame] | 429 | */ |
Dave Chinner | a44f13e | 2010-08-24 11:40:03 +1000 | [diff] [blame] | 430 | STATIC int |
Dave Chinner | 71e330b | 2010-05-21 14:37:18 +1000 | [diff] [blame] | 431 | xlog_cil_push( |
Mark Tinguely | f7bdf03 | 2012-06-14 09:22:15 -0500 | [diff] [blame] | 432 | struct xlog *log) |
Dave Chinner | 71e330b | 2010-05-21 14:37:18 +1000 | [diff] [blame] | 433 | { |
| 434 | struct xfs_cil *cil = log->l_cilp; |
| 435 | struct xfs_log_vec *lv; |
| 436 | struct xfs_cil_ctx *ctx; |
| 437 | struct xfs_cil_ctx *new_ctx; |
| 438 | struct xlog_in_core *commit_iclog; |
| 439 | struct xlog_ticket *tic; |
Dave Chinner | 71e330b | 2010-05-21 14:37:18 +1000 | [diff] [blame] | 440 | int num_iovecs; |
Dave Chinner | 71e330b | 2010-05-21 14:37:18 +1000 | [diff] [blame] | 441 | int error = 0; |
| 442 | struct xfs_trans_header thdr; |
| 443 | struct xfs_log_iovec lhdr; |
| 444 | struct xfs_log_vec lvhdr = { NULL }; |
| 445 | xfs_lsn_t commit_lsn; |
Dave Chinner | 4c2d542 | 2012-04-23 17:54:32 +1000 | [diff] [blame] | 446 | xfs_lsn_t push_seq; |
Dave Chinner | 71e330b | 2010-05-21 14:37:18 +1000 | [diff] [blame] | 447 | |
| 448 | if (!cil) |
| 449 | return 0; |
| 450 | |
Dave Chinner | 71e330b | 2010-05-21 14:37:18 +1000 | [diff] [blame] | 451 | new_ctx = kmem_zalloc(sizeof(*new_ctx), KM_SLEEP|KM_NOFS); |
| 452 | new_ctx->ticket = xlog_cil_ticket_alloc(log); |
| 453 | |
Dave Chinner | 4c2d542 | 2012-04-23 17:54:32 +1000 | [diff] [blame] | 454 | down_write(&cil->xc_ctx_lock); |
Dave Chinner | 71e330b | 2010-05-21 14:37:18 +1000 | [diff] [blame] | 455 | ctx = cil->xc_ctx; |
| 456 | |
Dave Chinner | 4c2d542 | 2012-04-23 17:54:32 +1000 | [diff] [blame] | 457 | spin_lock(&cil->xc_cil_lock); |
| 458 | push_seq = cil->xc_push_seq; |
| 459 | ASSERT(push_seq <= ctx->sequence); |
Dave Chinner | 71e330b | 2010-05-21 14:37:18 +1000 | [diff] [blame] | 460 | |
Dave Chinner | 4c2d542 | 2012-04-23 17:54:32 +1000 | [diff] [blame] | 461 | /* |
| 462 | * Check if we've anything to push. If there is nothing, then we don't |
| 463 | * move on to a new sequence number and so we have to be able to push |
| 464 | * this sequence again later. |
| 465 | */ |
| 466 | if (list_empty(&cil->xc_cil)) { |
| 467 | cil->xc_push_seq = 0; |
| 468 | spin_unlock(&cil->xc_cil_lock); |
Dave Chinner | a44f13e | 2010-08-24 11:40:03 +1000 | [diff] [blame] | 469 | goto out_skip; |
Dave Chinner | 4c2d542 | 2012-04-23 17:54:32 +1000 | [diff] [blame] | 470 | } |
| 471 | spin_unlock(&cil->xc_cil_lock); |
| 472 | |
Dave Chinner | a44f13e | 2010-08-24 11:40:03 +1000 | [diff] [blame] | 473 | |
| 474 | /* check for a previously pushed seqeunce */ |
Dave Chinner | 4c2d542 | 2012-04-23 17:54:32 +1000 | [diff] [blame] | 475 | if (push_seq < cil->xc_ctx->sequence) |
Dave Chinner | df80615 | 2010-05-17 15:52:13 +1000 | [diff] [blame] | 476 | goto out_skip; |
| 477 | |
Dave Chinner | 71e330b | 2010-05-21 14:37:18 +1000 | [diff] [blame] | 478 | /* |
| 479 | * pull all the log vectors off the items in the CIL, and |
| 480 | * remove the items from the CIL. We don't need the CIL lock |
| 481 | * here because it's only needed on the transaction commit |
| 482 | * side which is currently locked out by the flush lock. |
| 483 | */ |
| 484 | lv = NULL; |
Dave Chinner | 71e330b | 2010-05-21 14:37:18 +1000 | [diff] [blame] | 485 | num_iovecs = 0; |
Dave Chinner | 71e330b | 2010-05-21 14:37:18 +1000 | [diff] [blame] | 486 | while (!list_empty(&cil->xc_cil)) { |
| 487 | struct xfs_log_item *item; |
Dave Chinner | 71e330b | 2010-05-21 14:37:18 +1000 | [diff] [blame] | 488 | |
| 489 | item = list_first_entry(&cil->xc_cil, |
| 490 | struct xfs_log_item, li_cil); |
| 491 | list_del_init(&item->li_cil); |
| 492 | if (!ctx->lv_chain) |
| 493 | ctx->lv_chain = item->li_lv; |
| 494 | else |
| 495 | lv->lv_next = item->li_lv; |
| 496 | lv = item->li_lv; |
| 497 | item->li_lv = NULL; |
Dave Chinner | 71e330b | 2010-05-21 14:37:18 +1000 | [diff] [blame] | 498 | num_iovecs += lv->lv_niovecs; |
Dave Chinner | 71e330b | 2010-05-21 14:37:18 +1000 | [diff] [blame] | 499 | } |
| 500 | |
| 501 | /* |
| 502 | * initialise the new context and attach it to the CIL. Then attach |
| 503 | * the current context to the CIL committing lsit so it can be found |
| 504 | * during log forces to extract the commit lsn of the sequence that |
| 505 | * needs to be forced. |
| 506 | */ |
| 507 | INIT_LIST_HEAD(&new_ctx->committing); |
| 508 | INIT_LIST_HEAD(&new_ctx->busy_extents); |
| 509 | new_ctx->sequence = ctx->sequence + 1; |
| 510 | new_ctx->cil = cil; |
| 511 | cil->xc_ctx = new_ctx; |
| 512 | |
| 513 | /* |
Dave Chinner | a44f13e | 2010-08-24 11:40:03 +1000 | [diff] [blame] | 514 | * mirror the new sequence into the cil structure so that we can do |
| 515 | * unlocked checks against the current sequence in log forces without |
| 516 | * risking deferencing a freed context pointer. |
| 517 | */ |
| 518 | cil->xc_current_sequence = new_ctx->sequence; |
| 519 | |
| 520 | /* |
Dave Chinner | 71e330b | 2010-05-21 14:37:18 +1000 | [diff] [blame] | 521 | * The switch is now done, so we can drop the context lock and move out |
| 522 | * of a shared context. We can't just go straight to the commit record, |
| 523 | * though - we need to synchronise with previous and future commits so |
| 524 | * that the commit records are correctly ordered in the log to ensure |
| 525 | * that we process items during log IO completion in the correct order. |
| 526 | * |
| 527 | * For example, if we get an EFI in one checkpoint and the EFD in the |
| 528 | * next (e.g. due to log forces), we do not want the checkpoint with |
| 529 | * the EFD to be committed before the checkpoint with the EFI. Hence |
| 530 | * we must strictly order the commit records of the checkpoints so |
| 531 | * that: a) the checkpoint callbacks are attached to the iclogs in the |
| 532 | * correct order; and b) the checkpoints are replayed in correct order |
| 533 | * in log recovery. |
| 534 | * |
| 535 | * Hence we need to add this context to the committing context list so |
| 536 | * that higher sequences will wait for us to write out a commit record |
| 537 | * before they do. |
| 538 | */ |
| 539 | spin_lock(&cil->xc_cil_lock); |
| 540 | list_add(&ctx->committing, &cil->xc_committing); |
| 541 | spin_unlock(&cil->xc_cil_lock); |
| 542 | up_write(&cil->xc_ctx_lock); |
| 543 | |
| 544 | /* |
| 545 | * Build a checkpoint transaction header and write it to the log to |
| 546 | * begin the transaction. We need to account for the space used by the |
| 547 | * transaction header here as it is not accounted for in xlog_write(). |
| 548 | * |
| 549 | * The LSN we need to pass to the log items on transaction commit is |
| 550 | * the LSN reported by the first log vector write. If we use the commit |
| 551 | * record lsn then we can move the tail beyond the grant write head. |
| 552 | */ |
| 553 | tic = ctx->ticket; |
| 554 | thdr.th_magic = XFS_TRANS_HEADER_MAGIC; |
| 555 | thdr.th_type = XFS_TRANS_CHECKPOINT; |
| 556 | thdr.th_tid = tic->t_tid; |
| 557 | thdr.th_num_items = num_iovecs; |
Christoph Hellwig | 4e0d5f9 | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 558 | lhdr.i_addr = &thdr; |
Dave Chinner | 71e330b | 2010-05-21 14:37:18 +1000 | [diff] [blame] | 559 | lhdr.i_len = sizeof(xfs_trans_header_t); |
| 560 | lhdr.i_type = XLOG_REG_TYPE_TRANSHDR; |
| 561 | tic->t_curr_res -= lhdr.i_len + sizeof(xlog_op_header_t); |
| 562 | |
| 563 | lvhdr.lv_niovecs = 1; |
| 564 | lvhdr.lv_iovecp = &lhdr; |
| 565 | lvhdr.lv_next = ctx->lv_chain; |
| 566 | |
| 567 | error = xlog_write(log, &lvhdr, tic, &ctx->start_lsn, NULL, 0); |
| 568 | if (error) |
Dave Chinner | 7db37c5 | 2011-01-27 12:02:00 +1100 | [diff] [blame] | 569 | goto out_abort_free_ticket; |
Dave Chinner | 71e330b | 2010-05-21 14:37:18 +1000 | [diff] [blame] | 570 | |
| 571 | /* |
| 572 | * now that we've written the checkpoint into the log, strictly |
| 573 | * order the commit records so replay will get them in the right order. |
| 574 | */ |
| 575 | restart: |
| 576 | spin_lock(&cil->xc_cil_lock); |
| 577 | list_for_each_entry(new_ctx, &cil->xc_committing, committing) { |
| 578 | /* |
| 579 | * Higher sequences will wait for this one so skip them. |
| 580 | * Don't wait for own own sequence, either. |
| 581 | */ |
| 582 | if (new_ctx->sequence >= ctx->sequence) |
| 583 | continue; |
| 584 | if (!new_ctx->commit_lsn) { |
| 585 | /* |
| 586 | * It is still being pushed! Wait for the push to |
| 587 | * complete, then start again from the beginning. |
| 588 | */ |
Dave Chinner | eb40a87 | 2010-12-21 12:09:01 +1100 | [diff] [blame] | 589 | xlog_wait(&cil->xc_commit_wait, &cil->xc_cil_lock); |
Dave Chinner | 71e330b | 2010-05-21 14:37:18 +1000 | [diff] [blame] | 590 | goto restart; |
| 591 | } |
| 592 | } |
| 593 | spin_unlock(&cil->xc_cil_lock); |
| 594 | |
Dave Chinner | 7db37c5 | 2011-01-27 12:02:00 +1100 | [diff] [blame] | 595 | /* xfs_log_done always frees the ticket on error. */ |
Dave Chinner | 71e330b | 2010-05-21 14:37:18 +1000 | [diff] [blame] | 596 | commit_lsn = xfs_log_done(log->l_mp, tic, &commit_iclog, 0); |
Dave Chinner | 7db37c5 | 2011-01-27 12:02:00 +1100 | [diff] [blame] | 597 | if (commit_lsn == -1) |
Dave Chinner | 71e330b | 2010-05-21 14:37:18 +1000 | [diff] [blame] | 598 | goto out_abort; |
| 599 | |
| 600 | /* attach all the transactions w/ busy extents to iclog */ |
| 601 | ctx->log_cb.cb_func = xlog_cil_committed; |
| 602 | ctx->log_cb.cb_arg = ctx; |
| 603 | error = xfs_log_notify(log->l_mp, commit_iclog, &ctx->log_cb); |
| 604 | if (error) |
| 605 | goto out_abort; |
| 606 | |
| 607 | /* |
| 608 | * now the checkpoint commit is complete and we've attached the |
| 609 | * callbacks to the iclog we can assign the commit LSN to the context |
| 610 | * and wake up anyone who is waiting for the commit to complete. |
| 611 | */ |
| 612 | spin_lock(&cil->xc_cil_lock); |
| 613 | ctx->commit_lsn = commit_lsn; |
Dave Chinner | eb40a87 | 2010-12-21 12:09:01 +1100 | [diff] [blame] | 614 | wake_up_all(&cil->xc_commit_wait); |
Dave Chinner | 71e330b | 2010-05-21 14:37:18 +1000 | [diff] [blame] | 615 | spin_unlock(&cil->xc_cil_lock); |
| 616 | |
| 617 | /* release the hounds! */ |
| 618 | return xfs_log_release_iclog(log->l_mp, commit_iclog); |
| 619 | |
| 620 | out_skip: |
| 621 | up_write(&cil->xc_ctx_lock); |
Dave Chinner | 71e330b | 2010-05-21 14:37:18 +1000 | [diff] [blame] | 622 | xfs_log_ticket_put(new_ctx->ticket); |
| 623 | kmem_free(new_ctx); |
| 624 | return 0; |
| 625 | |
Dave Chinner | 7db37c5 | 2011-01-27 12:02:00 +1100 | [diff] [blame] | 626 | out_abort_free_ticket: |
| 627 | xfs_log_ticket_put(tic); |
Dave Chinner | 71e330b | 2010-05-21 14:37:18 +1000 | [diff] [blame] | 628 | out_abort: |
| 629 | xlog_cil_committed(ctx, XFS_LI_ABORTED); |
| 630 | return XFS_ERROR(EIO); |
| 631 | } |
| 632 | |
Dave Chinner | 4c2d542 | 2012-04-23 17:54:32 +1000 | [diff] [blame] | 633 | static void |
| 634 | xlog_cil_push_work( |
| 635 | struct work_struct *work) |
| 636 | { |
| 637 | struct xfs_cil *cil = container_of(work, struct xfs_cil, |
| 638 | xc_push_work); |
| 639 | xlog_cil_push(cil->xc_log); |
| 640 | } |
| 641 | |
| 642 | /* |
| 643 | * We need to push CIL every so often so we don't cache more than we can fit in |
| 644 | * the log. The limit really is that a checkpoint can't be more than half the |
| 645 | * log (the current checkpoint is not allowed to overwrite the previous |
| 646 | * checkpoint), but commit latency and memory usage limit this to a smaller |
| 647 | * size. |
| 648 | */ |
| 649 | static void |
| 650 | xlog_cil_push_background( |
Mark Tinguely | f7bdf03 | 2012-06-14 09:22:15 -0500 | [diff] [blame] | 651 | struct xlog *log) |
Dave Chinner | 4c2d542 | 2012-04-23 17:54:32 +1000 | [diff] [blame] | 652 | { |
| 653 | struct xfs_cil *cil = log->l_cilp; |
| 654 | |
| 655 | /* |
| 656 | * The cil won't be empty because we are called while holding the |
| 657 | * context lock so whatever we added to the CIL will still be there |
| 658 | */ |
| 659 | ASSERT(!list_empty(&cil->xc_cil)); |
| 660 | |
| 661 | /* |
| 662 | * don't do a background push if we haven't used up all the |
| 663 | * space available yet. |
| 664 | */ |
| 665 | if (cil->xc_ctx->space_used < XLOG_CIL_SPACE_LIMIT(log)) |
| 666 | return; |
| 667 | |
| 668 | spin_lock(&cil->xc_cil_lock); |
| 669 | if (cil->xc_push_seq < cil->xc_current_sequence) { |
| 670 | cil->xc_push_seq = cil->xc_current_sequence; |
| 671 | queue_work(log->l_mp->m_cil_workqueue, &cil->xc_push_work); |
| 672 | } |
| 673 | spin_unlock(&cil->xc_cil_lock); |
| 674 | |
| 675 | } |
| 676 | |
| 677 | static void |
| 678 | xlog_cil_push_foreground( |
Mark Tinguely | f7bdf03 | 2012-06-14 09:22:15 -0500 | [diff] [blame] | 679 | struct xlog *log, |
Dave Chinner | 4c2d542 | 2012-04-23 17:54:32 +1000 | [diff] [blame] | 680 | xfs_lsn_t push_seq) |
| 681 | { |
| 682 | struct xfs_cil *cil = log->l_cilp; |
| 683 | |
| 684 | if (!cil) |
| 685 | return; |
| 686 | |
| 687 | ASSERT(push_seq && push_seq <= cil->xc_current_sequence); |
| 688 | |
| 689 | /* start on any pending background push to minimise wait time on it */ |
| 690 | flush_work(&cil->xc_push_work); |
| 691 | |
| 692 | /* |
| 693 | * If the CIL is empty or we've already pushed the sequence then |
| 694 | * there's no work we need to do. |
| 695 | */ |
| 696 | spin_lock(&cil->xc_cil_lock); |
| 697 | if (list_empty(&cil->xc_cil) || push_seq <= cil->xc_push_seq) { |
| 698 | spin_unlock(&cil->xc_cil_lock); |
| 699 | return; |
| 700 | } |
| 701 | |
| 702 | cil->xc_push_seq = push_seq; |
| 703 | spin_unlock(&cil->xc_cil_lock); |
| 704 | |
| 705 | /* do the push now */ |
| 706 | xlog_cil_push(log); |
| 707 | } |
| 708 | |
Dave Chinner | 71e330b | 2010-05-21 14:37:18 +1000 | [diff] [blame] | 709 | /* |
Dave Chinner | a44f13e | 2010-08-24 11:40:03 +1000 | [diff] [blame] | 710 | * Commit a transaction with the given vector to the Committed Item List. |
| 711 | * |
| 712 | * To do this, we need to format the item, pin it in memory if required and |
| 713 | * account for the space used by the transaction. Once we have done that we |
| 714 | * need to release the unused reservation for the transaction, attach the |
| 715 | * transaction to the checkpoint context so we carry the busy extents through |
| 716 | * to checkpoint completion, and then unlock all the items in the transaction. |
| 717 | * |
Dave Chinner | a44f13e | 2010-08-24 11:40:03 +1000 | [diff] [blame] | 718 | * Called with the context lock already held in read mode to lock out |
| 719 | * background commit, returns without it held once background commits are |
| 720 | * allowed again. |
| 721 | */ |
Christoph Hellwig | 0244b96 | 2011-12-06 21:58:08 +0000 | [diff] [blame] | 722 | int |
Dave Chinner | a44f13e | 2010-08-24 11:40:03 +1000 | [diff] [blame] | 723 | xfs_log_commit_cil( |
| 724 | struct xfs_mount *mp, |
| 725 | struct xfs_trans *tp, |
Dave Chinner | a44f13e | 2010-08-24 11:40:03 +1000 | [diff] [blame] | 726 | xfs_lsn_t *commit_lsn, |
| 727 | int flags) |
| 728 | { |
Mark Tinguely | f7bdf03 | 2012-06-14 09:22:15 -0500 | [diff] [blame] | 729 | struct xlog *log = mp->m_log; |
Dave Chinner | a44f13e | 2010-08-24 11:40:03 +1000 | [diff] [blame] | 730 | int log_flags = 0; |
Christoph Hellwig | 0244b96 | 2011-12-06 21:58:08 +0000 | [diff] [blame] | 731 | struct xfs_log_vec *log_vector; |
Dave Chinner | a44f13e | 2010-08-24 11:40:03 +1000 | [diff] [blame] | 732 | |
| 733 | if (flags & XFS_TRANS_RELEASE_LOG_RES) |
| 734 | log_flags = XFS_LOG_REL_PERM_RESERV; |
| 735 | |
Dave Chinner | 3b93c7a | 2010-08-24 11:45:53 +1000 | [diff] [blame] | 736 | /* |
Christoph Hellwig | 0244b96 | 2011-12-06 21:58:08 +0000 | [diff] [blame] | 737 | * Do all the hard work of formatting items (including memory |
Dave Chinner | 3b93c7a | 2010-08-24 11:45:53 +1000 | [diff] [blame] | 738 | * allocation) outside the CIL context lock. This prevents stalling CIL |
| 739 | * pushes when we are low on memory and a transaction commit spends a |
| 740 | * lot of time in memory reclaim. |
| 741 | */ |
Christoph Hellwig | 0244b96 | 2011-12-06 21:58:08 +0000 | [diff] [blame] | 742 | log_vector = xlog_cil_prepare_log_vecs(tp); |
| 743 | if (!log_vector) |
| 744 | return ENOMEM; |
Dave Chinner | 3b93c7a | 2010-08-24 11:45:53 +1000 | [diff] [blame] | 745 | |
Dave Chinner | a44f13e | 2010-08-24 11:40:03 +1000 | [diff] [blame] | 746 | /* lock out background commit */ |
| 747 | down_read(&log->l_cilp->xc_ctx_lock); |
Dave Chinner | d1583a3 | 2010-09-24 18:14:13 +1000 | [diff] [blame] | 748 | if (commit_lsn) |
| 749 | *commit_lsn = log->l_cilp->xc_ctx->sequence; |
| 750 | |
Dave Chinner | fd63875 | 2013-06-27 16:04:51 +1000 | [diff] [blame] | 751 | /* xlog_cil_insert_items() destroys log_vector list */ |
Dave Chinner | d1583a3 | 2010-09-24 18:14:13 +1000 | [diff] [blame] | 752 | xlog_cil_insert_items(log, log_vector, tp->t_ticket); |
Dave Chinner | a44f13e | 2010-08-24 11:40:03 +1000 | [diff] [blame] | 753 | |
| 754 | /* check we didn't blow the reservation */ |
| 755 | if (tp->t_ticket->t_curr_res < 0) |
| 756 | xlog_print_tic_res(log->l_mp, tp->t_ticket); |
| 757 | |
| 758 | /* attach the transaction to the CIL if it has any busy extents */ |
| 759 | if (!list_empty(&tp->t_busy)) { |
| 760 | spin_lock(&log->l_cilp->xc_cil_lock); |
| 761 | list_splice_init(&tp->t_busy, |
| 762 | &log->l_cilp->xc_ctx->busy_extents); |
| 763 | spin_unlock(&log->l_cilp->xc_cil_lock); |
| 764 | } |
| 765 | |
| 766 | tp->t_commit_lsn = *commit_lsn; |
| 767 | xfs_log_done(mp, tp->t_ticket, NULL, log_flags); |
| 768 | xfs_trans_unreserve_and_mod_sb(tp); |
| 769 | |
| 770 | /* |
| 771 | * Once all the items of the transaction have been copied to the CIL, |
| 772 | * the items can be unlocked and freed. |
| 773 | * |
| 774 | * This needs to be done before we drop the CIL context lock because we |
| 775 | * have to update state in the log items and unlock them before they go |
| 776 | * to disk. If we don't, then the CIL checkpoint can race with us and |
| 777 | * we can run checkpoint completion before we've updated and unlocked |
| 778 | * the log items. This affects (at least) processing of stale buffers, |
| 779 | * inodes and EFIs. |
| 780 | */ |
| 781 | xfs_trans_free_items(tp, *commit_lsn, 0); |
| 782 | |
Dave Chinner | 4c2d542 | 2012-04-23 17:54:32 +1000 | [diff] [blame] | 783 | xlog_cil_push_background(log); |
Dave Chinner | a44f13e | 2010-08-24 11:40:03 +1000 | [diff] [blame] | 784 | |
| 785 | up_read(&log->l_cilp->xc_ctx_lock); |
Christoph Hellwig | 0244b96 | 2011-12-06 21:58:08 +0000 | [diff] [blame] | 786 | return 0; |
Dave Chinner | a44f13e | 2010-08-24 11:40:03 +1000 | [diff] [blame] | 787 | } |
| 788 | |
| 789 | /* |
Dave Chinner | 71e330b | 2010-05-21 14:37:18 +1000 | [diff] [blame] | 790 | * Conditionally push the CIL based on the sequence passed in. |
| 791 | * |
| 792 | * We only need to push if we haven't already pushed the sequence |
| 793 | * number given. Hence the only time we will trigger a push here is |
| 794 | * if the push sequence is the same as the current context. |
| 795 | * |
| 796 | * We return the current commit lsn to allow the callers to determine if a |
| 797 | * iclog flush is necessary following this call. |
Dave Chinner | 71e330b | 2010-05-21 14:37:18 +1000 | [diff] [blame] | 798 | */ |
| 799 | xfs_lsn_t |
Dave Chinner | a44f13e | 2010-08-24 11:40:03 +1000 | [diff] [blame] | 800 | xlog_cil_force_lsn( |
Mark Tinguely | f7bdf03 | 2012-06-14 09:22:15 -0500 | [diff] [blame] | 801 | struct xlog *log, |
Dave Chinner | a44f13e | 2010-08-24 11:40:03 +1000 | [diff] [blame] | 802 | xfs_lsn_t sequence) |
Dave Chinner | 71e330b | 2010-05-21 14:37:18 +1000 | [diff] [blame] | 803 | { |
| 804 | struct xfs_cil *cil = log->l_cilp; |
| 805 | struct xfs_cil_ctx *ctx; |
| 806 | xfs_lsn_t commit_lsn = NULLCOMMITLSN; |
| 807 | |
Dave Chinner | a44f13e | 2010-08-24 11:40:03 +1000 | [diff] [blame] | 808 | ASSERT(sequence <= cil->xc_current_sequence); |
Dave Chinner | 71e330b | 2010-05-21 14:37:18 +1000 | [diff] [blame] | 809 | |
Dave Chinner | a44f13e | 2010-08-24 11:40:03 +1000 | [diff] [blame] | 810 | /* |
| 811 | * check to see if we need to force out the current context. |
| 812 | * xlog_cil_push() handles racing pushes for the same sequence, |
| 813 | * so no need to deal with it here. |
| 814 | */ |
Dave Chinner | 4c2d542 | 2012-04-23 17:54:32 +1000 | [diff] [blame] | 815 | xlog_cil_push_foreground(log, sequence); |
Dave Chinner | 71e330b | 2010-05-21 14:37:18 +1000 | [diff] [blame] | 816 | |
| 817 | /* |
| 818 | * See if we can find a previous sequence still committing. |
Dave Chinner | 71e330b | 2010-05-21 14:37:18 +1000 | [diff] [blame] | 819 | * We need to wait for all previous sequence commits to complete |
| 820 | * before allowing the force of push_seq to go ahead. Hence block |
| 821 | * on commits for those as well. |
| 822 | */ |
Dave Chinner | a44f13e | 2010-08-24 11:40:03 +1000 | [diff] [blame] | 823 | restart: |
Dave Chinner | 71e330b | 2010-05-21 14:37:18 +1000 | [diff] [blame] | 824 | spin_lock(&cil->xc_cil_lock); |
Dave Chinner | 71e330b | 2010-05-21 14:37:18 +1000 | [diff] [blame] | 825 | list_for_each_entry(ctx, &cil->xc_committing, committing) { |
Dave Chinner | a44f13e | 2010-08-24 11:40:03 +1000 | [diff] [blame] | 826 | if (ctx->sequence > sequence) |
Dave Chinner | 71e330b | 2010-05-21 14:37:18 +1000 | [diff] [blame] | 827 | continue; |
| 828 | if (!ctx->commit_lsn) { |
| 829 | /* |
| 830 | * It is still being pushed! Wait for the push to |
| 831 | * complete, then start again from the beginning. |
| 832 | */ |
Dave Chinner | eb40a87 | 2010-12-21 12:09:01 +1100 | [diff] [blame] | 833 | xlog_wait(&cil->xc_commit_wait, &cil->xc_cil_lock); |
Dave Chinner | 71e330b | 2010-05-21 14:37:18 +1000 | [diff] [blame] | 834 | goto restart; |
| 835 | } |
Dave Chinner | a44f13e | 2010-08-24 11:40:03 +1000 | [diff] [blame] | 836 | if (ctx->sequence != sequence) |
Dave Chinner | 71e330b | 2010-05-21 14:37:18 +1000 | [diff] [blame] | 837 | continue; |
| 838 | /* found it! */ |
| 839 | commit_lsn = ctx->commit_lsn; |
| 840 | } |
| 841 | spin_unlock(&cil->xc_cil_lock); |
| 842 | return commit_lsn; |
| 843 | } |
Dave Chinner | ccf7c23 | 2010-05-20 23:19:42 +1000 | [diff] [blame] | 844 | |
| 845 | /* |
| 846 | * Check if the current log item was first committed in this sequence. |
| 847 | * We can't rely on just the log item being in the CIL, we have to check |
| 848 | * the recorded commit sequence number. |
| 849 | * |
| 850 | * Note: for this to be used in a non-racy manner, it has to be called with |
| 851 | * CIL flushing locked out. As a result, it should only be used during the |
| 852 | * transaction commit process when deciding what to format into the item. |
| 853 | */ |
| 854 | bool |
| 855 | xfs_log_item_in_current_chkpt( |
| 856 | struct xfs_log_item *lip) |
| 857 | { |
| 858 | struct xfs_cil_ctx *ctx; |
| 859 | |
Dave Chinner | ccf7c23 | 2010-05-20 23:19:42 +1000 | [diff] [blame] | 860 | if (list_empty(&lip->li_cil)) |
| 861 | return false; |
| 862 | |
| 863 | ctx = lip->li_mountp->m_log->l_cilp->xc_ctx; |
| 864 | |
| 865 | /* |
| 866 | * li_seq is written on the first commit of a log item to record the |
| 867 | * first checkpoint it is written to. Hence if it is different to the |
| 868 | * current sequence, we're in a new checkpoint. |
| 869 | */ |
| 870 | if (XFS_LSN_CMP(lip->li_seq, ctx->sequence) != 0) |
| 871 | return false; |
| 872 | return true; |
| 873 | } |
Dave Chinner | 4c2d542 | 2012-04-23 17:54:32 +1000 | [diff] [blame] | 874 | |
| 875 | /* |
| 876 | * Perform initial CIL structure initialisation. |
| 877 | */ |
| 878 | int |
| 879 | xlog_cil_init( |
Mark Tinguely | f7bdf03 | 2012-06-14 09:22:15 -0500 | [diff] [blame] | 880 | struct xlog *log) |
Dave Chinner | 4c2d542 | 2012-04-23 17:54:32 +1000 | [diff] [blame] | 881 | { |
| 882 | struct xfs_cil *cil; |
| 883 | struct xfs_cil_ctx *ctx; |
| 884 | |
| 885 | cil = kmem_zalloc(sizeof(*cil), KM_SLEEP|KM_MAYFAIL); |
| 886 | if (!cil) |
| 887 | return ENOMEM; |
| 888 | |
| 889 | ctx = kmem_zalloc(sizeof(*ctx), KM_SLEEP|KM_MAYFAIL); |
| 890 | if (!ctx) { |
| 891 | kmem_free(cil); |
| 892 | return ENOMEM; |
| 893 | } |
| 894 | |
| 895 | INIT_WORK(&cil->xc_push_work, xlog_cil_push_work); |
| 896 | INIT_LIST_HEAD(&cil->xc_cil); |
| 897 | INIT_LIST_HEAD(&cil->xc_committing); |
| 898 | spin_lock_init(&cil->xc_cil_lock); |
| 899 | init_rwsem(&cil->xc_ctx_lock); |
| 900 | init_waitqueue_head(&cil->xc_commit_wait); |
| 901 | |
| 902 | INIT_LIST_HEAD(&ctx->committing); |
| 903 | INIT_LIST_HEAD(&ctx->busy_extents); |
| 904 | ctx->sequence = 1; |
| 905 | ctx->cil = cil; |
| 906 | cil->xc_ctx = ctx; |
| 907 | cil->xc_current_sequence = ctx->sequence; |
| 908 | |
| 909 | cil->xc_log = log; |
| 910 | log->l_cilp = cil; |
| 911 | return 0; |
| 912 | } |
| 913 | |
| 914 | void |
| 915 | xlog_cil_destroy( |
Mark Tinguely | f7bdf03 | 2012-06-14 09:22:15 -0500 | [diff] [blame] | 916 | struct xlog *log) |
Dave Chinner | 4c2d542 | 2012-04-23 17:54:32 +1000 | [diff] [blame] | 917 | { |
| 918 | if (log->l_cilp->xc_ctx) { |
| 919 | if (log->l_cilp->xc_ctx->ticket) |
| 920 | xfs_log_ticket_put(log->l_cilp->xc_ctx->ticket); |
| 921 | kmem_free(log->l_cilp->xc_ctx); |
| 922 | } |
| 923 | |
| 924 | ASSERT(list_empty(&log->l_cilp->xc_cil)); |
| 925 | kmem_free(log->l_cilp); |
| 926 | } |
| 927 | |