aboutsummaryrefslogtreecommitdiffstats
path: root/fs
diff options
context:
space:
mode:
authorLinus Torvalds <torvalds@linux-foundation.org>2010-05-25 11:17:01 -0400
committerLinus Torvalds <torvalds@linux-foundation.org>2010-05-25 11:17:01 -0400
commit110b93842e36b17598cf24874e90d0401431cda2 (patch)
treed95591d129ea8ed8d5b8e489e2d5ae68e79586c4 /fs
parent4961ab934a1254b1ad9420cea0ded617b57f022b (diff)
parent88e88374ee4958786397a57f684de6f1fc5e0242 (diff)
Merge branch 'for-linus' of git://oss.sgi.com/xfs/xfs
* 'for-linus' of git://oss.sgi.com/xfs/xfs: xfs: Ensure inode allocation buffers are fully replayed xfs: enable background pushing of the CIL xfs: forced unmounts need to push the CIL xfs: Introduce delayed logging core code xfs: Delayed logging design documentation xfs: Improve scalability of busy extent tracking xfs: make the log ticket ID available outside the log infrastructure xfs: clean up log ticket overrun debug output xfs: Clean up XFS_BLI_* flag namespace xfs: modify buffer item reference counting xfs: allow log ticket allocation to take allocation flags xfs: Don't reuse the same transaction ID for duplicated transactions.
Diffstat (limited to 'fs')
-rw-r--r--fs/xfs/Makefile1
-rw-r--r--fs/xfs/linux-2.6/xfs_buf.c9
-rw-r--r--fs/xfs/linux-2.6/xfs_quotaops.c1
-rw-r--r--fs/xfs/linux-2.6/xfs_super.c12
-rw-r--r--fs/xfs/linux-2.6/xfs_trace.h83
-rw-r--r--fs/xfs/quota/xfs_dquot.c6
-rw-r--r--fs/xfs/xfs_ag.h24
-rw-r--r--fs/xfs/xfs_alloc.c357
-rw-r--r--fs/xfs/xfs_alloc.h7
-rw-r--r--fs/xfs/xfs_alloc_btree.c2
-rw-r--r--fs/xfs/xfs_buf_item.c166
-rw-r--r--fs/xfs/xfs_buf_item.h18
-rw-r--r--fs/xfs/xfs_error.c2
-rw-r--r--fs/xfs/xfs_log.c120
-rw-r--r--fs/xfs/xfs_log.h14
-rw-r--r--fs/xfs/xfs_log_cil.c725
-rw-r--r--fs/xfs/xfs_log_priv.h118
-rw-r--r--fs/xfs/xfs_log_recover.c46
-rw-r--r--fs/xfs/xfs_log_recover.h2
-rw-r--r--fs/xfs/xfs_mount.h1
-rw-r--r--fs/xfs/xfs_trans.c144
-rw-r--r--fs/xfs/xfs_trans.h44
-rw-r--r--fs/xfs/xfs_trans_buf.c46
-rw-r--r--fs/xfs/xfs_trans_item.c114
-rw-r--r--fs/xfs/xfs_trans_priv.h15
-rw-r--r--fs/xfs/xfs_types.h2
26 files changed, 1566 insertions, 513 deletions
diff --git a/fs/xfs/Makefile b/fs/xfs/Makefile
index b4769e40e8bc..c8fb13f83b3f 100644
--- a/fs/xfs/Makefile
+++ b/fs/xfs/Makefile
@@ -77,6 +77,7 @@ xfs-y += xfs_alloc.o \
77 xfs_itable.o \ 77 xfs_itable.o \
78 xfs_dfrag.o \ 78 xfs_dfrag.o \
79 xfs_log.o \ 79 xfs_log.o \
80 xfs_log_cil.o \
80 xfs_log_recover.o \ 81 xfs_log_recover.o \
81 xfs_mount.o \ 82 xfs_mount.o \
82 xfs_mru_cache.o \ 83 xfs_mru_cache.o \
diff --git a/fs/xfs/linux-2.6/xfs_buf.c b/fs/xfs/linux-2.6/xfs_buf.c
index f01de3c55c43..649ade8ef598 100644
--- a/fs/xfs/linux-2.6/xfs_buf.c
+++ b/fs/xfs/linux-2.6/xfs_buf.c
@@ -37,6 +37,7 @@
37 37
38#include "xfs_sb.h" 38#include "xfs_sb.h"
39#include "xfs_inum.h" 39#include "xfs_inum.h"
40#include "xfs_log.h"
40#include "xfs_ag.h" 41#include "xfs_ag.h"
41#include "xfs_dmapi.h" 42#include "xfs_dmapi.h"
42#include "xfs_mount.h" 43#include "xfs_mount.h"
@@ -850,6 +851,12 @@ xfs_buf_lock_value(
850 * Note that this in no way locks the underlying pages, so it is only 851 * Note that this in no way locks the underlying pages, so it is only
851 * useful for synchronizing concurrent use of buffer objects, not for 852 * useful for synchronizing concurrent use of buffer objects, not for
852 * synchronizing independent access to the underlying pages. 853 * synchronizing independent access to the underlying pages.
854 *
855 * If we come across a stale, pinned, locked buffer, we know that we
856 * are being asked to lock a buffer that has been reallocated. Because
857 * it is pinned, we know that the log has not been pushed to disk and
858 * hence it will still be locked. Rather than sleeping until someone
859 * else pushes the log, push it ourselves before trying to get the lock.
853 */ 860 */
854void 861void
855xfs_buf_lock( 862xfs_buf_lock(
@@ -857,6 +864,8 @@ xfs_buf_lock(
857{ 864{
858 trace_xfs_buf_lock(bp, _RET_IP_); 865 trace_xfs_buf_lock(bp, _RET_IP_);
859 866
867 if (atomic_read(&bp->b_pin_count) && (bp->b_flags & XBF_STALE))
868 xfs_log_force(bp->b_mount, 0);
860 if (atomic_read(&bp->b_io_remaining)) 869 if (atomic_read(&bp->b_io_remaining))
861 blk_run_address_space(bp->b_target->bt_mapping); 870 blk_run_address_space(bp->b_target->bt_mapping);
862 down(&bp->b_sema); 871 down(&bp->b_sema);
diff --git a/fs/xfs/linux-2.6/xfs_quotaops.c b/fs/xfs/linux-2.6/xfs_quotaops.c
index e31bf21fe5d3..9ac8aea91529 100644
--- a/fs/xfs/linux-2.6/xfs_quotaops.c
+++ b/fs/xfs/linux-2.6/xfs_quotaops.c
@@ -19,6 +19,7 @@
19#include "xfs_dmapi.h" 19#include "xfs_dmapi.h"
20#include "xfs_sb.h" 20#include "xfs_sb.h"
21#include "xfs_inum.h" 21#include "xfs_inum.h"
22#include "xfs_log.h"
22#include "xfs_ag.h" 23#include "xfs_ag.h"
23#include "xfs_mount.h" 24#include "xfs_mount.h"
24#include "xfs_quota.h" 25#include "xfs_quota.h"
diff --git a/fs/xfs/linux-2.6/xfs_super.c b/fs/xfs/linux-2.6/xfs_super.c
index f24dbe5efde3..f2d1718c9165 100644
--- a/fs/xfs/linux-2.6/xfs_super.c
+++ b/fs/xfs/linux-2.6/xfs_super.c
@@ -119,6 +119,8 @@ mempool_t *xfs_ioend_pool;
119#define MNTOPT_DMAPI "dmapi" /* DMI enabled (DMAPI / XDSM) */ 119#define MNTOPT_DMAPI "dmapi" /* DMI enabled (DMAPI / XDSM) */
120#define MNTOPT_XDSM "xdsm" /* DMI enabled (DMAPI / XDSM) */ 120#define MNTOPT_XDSM "xdsm" /* DMI enabled (DMAPI / XDSM) */
121#define MNTOPT_DMI "dmi" /* DMI enabled (DMAPI / XDSM) */ 121#define MNTOPT_DMI "dmi" /* DMI enabled (DMAPI / XDSM) */
122#define MNTOPT_DELAYLOG "delaylog" /* Delayed loging enabled */
123#define MNTOPT_NODELAYLOG "nodelaylog" /* Delayed loging disabled */
122 124
123/* 125/*
124 * Table driven mount option parser. 126 * Table driven mount option parser.
@@ -374,6 +376,13 @@ xfs_parseargs(
374 mp->m_flags |= XFS_MOUNT_DMAPI; 376 mp->m_flags |= XFS_MOUNT_DMAPI;
375 } else if (!strcmp(this_char, MNTOPT_DMI)) { 377 } else if (!strcmp(this_char, MNTOPT_DMI)) {
376 mp->m_flags |= XFS_MOUNT_DMAPI; 378 mp->m_flags |= XFS_MOUNT_DMAPI;
379 } else if (!strcmp(this_char, MNTOPT_DELAYLOG)) {
380 mp->m_flags |= XFS_MOUNT_DELAYLOG;
381 cmn_err(CE_WARN,
382 "Enabling EXPERIMENTAL delayed logging feature "
383 "- use at your own risk.\n");
384 } else if (!strcmp(this_char, MNTOPT_NODELAYLOG)) {
385 mp->m_flags &= ~XFS_MOUNT_DELAYLOG;
377 } else if (!strcmp(this_char, "ihashsize")) { 386 } else if (!strcmp(this_char, "ihashsize")) {
378 cmn_err(CE_WARN, 387 cmn_err(CE_WARN,
379 "XFS: ihashsize no longer used, option is deprecated."); 388 "XFS: ihashsize no longer used, option is deprecated.");
@@ -535,6 +544,7 @@ xfs_showargs(
535 { XFS_MOUNT_FILESTREAMS, "," MNTOPT_FILESTREAM }, 544 { XFS_MOUNT_FILESTREAMS, "," MNTOPT_FILESTREAM },
536 { XFS_MOUNT_DMAPI, "," MNTOPT_DMAPI }, 545 { XFS_MOUNT_DMAPI, "," MNTOPT_DMAPI },
537 { XFS_MOUNT_GRPID, "," MNTOPT_GRPID }, 546 { XFS_MOUNT_GRPID, "," MNTOPT_GRPID },
547 { XFS_MOUNT_DELAYLOG, "," MNTOPT_DELAYLOG },
538 { 0, NULL } 548 { 0, NULL }
539 }; 549 };
540 static struct proc_xfs_info xfs_info_unset[] = { 550 static struct proc_xfs_info xfs_info_unset[] = {
@@ -1755,7 +1765,7 @@ xfs_init_zones(void)
1755 * but it is much faster. 1765 * but it is much faster.
1756 */ 1766 */
1757 xfs_buf_item_zone = kmem_zone_init((sizeof(xfs_buf_log_item_t) + 1767 xfs_buf_item_zone = kmem_zone_init((sizeof(xfs_buf_log_item_t) +
1758 (((XFS_MAX_BLOCKSIZE / XFS_BLI_CHUNK) / 1768 (((XFS_MAX_BLOCKSIZE / XFS_BLF_CHUNK) /
1759 NBWORD) * sizeof(int))), "xfs_buf_item"); 1769 NBWORD) * sizeof(int))), "xfs_buf_item");
1760 if (!xfs_buf_item_zone) 1770 if (!xfs_buf_item_zone)
1761 goto out_destroy_trans_zone; 1771 goto out_destroy_trans_zone;
diff --git a/fs/xfs/linux-2.6/xfs_trace.h b/fs/xfs/linux-2.6/xfs_trace.h
index 8a319cfd2901..ff6bc797baf2 100644
--- a/fs/xfs/linux-2.6/xfs_trace.h
+++ b/fs/xfs/linux-2.6/xfs_trace.h
@@ -1059,83 +1059,112 @@ TRACE_EVENT(xfs_bunmap,
1059 1059
1060); 1060);
1061 1061
1062#define XFS_BUSY_SYNC \
1063 { 0, "async" }, \
1064 { 1, "sync" }
1065
1062TRACE_EVENT(xfs_alloc_busy, 1066TRACE_EVENT(xfs_alloc_busy,
1063 TP_PROTO(struct xfs_mount *mp, xfs_agnumber_t agno, xfs_agblock_t agbno, 1067 TP_PROTO(struct xfs_trans *trans, xfs_agnumber_t agno,
1064 xfs_extlen_t len, int slot), 1068 xfs_agblock_t agbno, xfs_extlen_t len, int sync),
1065 TP_ARGS(mp, agno, agbno, len, slot), 1069 TP_ARGS(trans, agno, agbno, len, sync),
1066 TP_STRUCT__entry( 1070 TP_STRUCT__entry(
1067 __field(dev_t, dev) 1071 __field(dev_t, dev)
1072 __field(struct xfs_trans *, tp)
1073 __field(int, tid)
1068 __field(xfs_agnumber_t, agno) 1074 __field(xfs_agnumber_t, agno)
1069 __field(xfs_agblock_t, agbno) 1075 __field(xfs_agblock_t, agbno)
1070 __field(xfs_extlen_t, len) 1076 __field(xfs_extlen_t, len)
1071 __field(int, slot) 1077 __field(int, sync)
1072 ), 1078 ),
1073 TP_fast_assign( 1079 TP_fast_assign(
1074 __entry->dev = mp->m_super->s_dev; 1080 __entry->dev = trans->t_mountp->m_super->s_dev;
1081 __entry->tp = trans;
1082 __entry->tid = trans->t_ticket->t_tid;
1075 __entry->agno = agno; 1083 __entry->agno = agno;
1076 __entry->agbno = agbno; 1084 __entry->agbno = agbno;
1077 __entry->len = len; 1085 __entry->len = len;
1078 __entry->slot = slot; 1086 __entry->sync = sync;
1079 ), 1087 ),
1080 TP_printk("dev %d:%d agno %u agbno %u len %u slot %d", 1088 TP_printk("dev %d:%d trans 0x%p tid 0x%x agno %u agbno %u len %u %s",
1081 MAJOR(__entry->dev), MINOR(__entry->dev), 1089 MAJOR(__entry->dev), MINOR(__entry->dev),
1090 __entry->tp,
1091 __entry->tid,
1082 __entry->agno, 1092 __entry->agno,
1083 __entry->agbno, 1093 __entry->agbno,
1084 __entry->len, 1094 __entry->len,
1085 __entry->slot) 1095 __print_symbolic(__entry->sync, XFS_BUSY_SYNC))
1086 1096
1087); 1097);
1088 1098
1089#define XFS_BUSY_STATES \
1090 { 0, "found" }, \
1091 { 1, "missing" }
1092
1093TRACE_EVENT(xfs_alloc_unbusy, 1099TRACE_EVENT(xfs_alloc_unbusy,
1094 TP_PROTO(struct xfs_mount *mp, xfs_agnumber_t agno, 1100 TP_PROTO(struct xfs_mount *mp, xfs_agnumber_t agno,
1095 int slot, int found), 1101 xfs_agblock_t agbno, xfs_extlen_t len),
1096 TP_ARGS(mp, agno, slot, found), 1102 TP_ARGS(mp, agno, agbno, len),
1097 TP_STRUCT__entry( 1103 TP_STRUCT__entry(
1098 __field(dev_t, dev) 1104 __field(dev_t, dev)
1099 __field(xfs_agnumber_t, agno) 1105 __field(xfs_agnumber_t, agno)
1100 __field(int, slot) 1106 __field(xfs_agblock_t, agbno)
1101 __field(int, found) 1107 __field(xfs_extlen_t, len)
1102 ), 1108 ),
1103 TP_fast_assign( 1109 TP_fast_assign(
1104 __entry->dev = mp->m_super->s_dev; 1110 __entry->dev = mp->m_super->s_dev;
1105 __entry->agno = agno; 1111 __entry->agno = agno;
1106 __entry->slot = slot; 1112 __entry->agbno = agbno;
1107 __entry->found = found; 1113 __entry->len = len;
1108 ), 1114 ),
1109 TP_printk("dev %d:%d agno %u slot %d %s", 1115 TP_printk("dev %d:%d agno %u agbno %u len %u",
1110 MAJOR(__entry->dev), MINOR(__entry->dev), 1116 MAJOR(__entry->dev), MINOR(__entry->dev),
1111 __entry->agno, 1117 __entry->agno,
1112 __entry->slot, 1118 __entry->agbno,
1113 __print_symbolic(__entry->found, XFS_BUSY_STATES)) 1119 __entry->len)
1114); 1120);
1115 1121
1122#define XFS_BUSY_STATES \
1123 { 0, "missing" }, \
1124 { 1, "found" }
1125
1116TRACE_EVENT(xfs_alloc_busysearch, 1126TRACE_EVENT(xfs_alloc_busysearch,
1117 TP_PROTO(struct xfs_mount *mp, xfs_agnumber_t agno, xfs_agblock_t agbno, 1127 TP_PROTO(struct xfs_mount *mp, xfs_agnumber_t agno,
1118 xfs_extlen_t len, xfs_lsn_t lsn), 1128 xfs_agblock_t agbno, xfs_extlen_t len, int found),
1119 TP_ARGS(mp, agno, agbno, len, lsn), 1129 TP_ARGS(mp, agno, agbno, len, found),
1120 TP_STRUCT__entry( 1130 TP_STRUCT__entry(
1121 __field(dev_t, dev) 1131 __field(dev_t, dev)
1122 __field(xfs_agnumber_t, agno) 1132 __field(xfs_agnumber_t, agno)
1123 __field(xfs_agblock_t, agbno) 1133 __field(xfs_agblock_t, agbno)
1124 __field(xfs_extlen_t, len) 1134 __field(xfs_extlen_t, len)
1125 __field(xfs_lsn_t, lsn) 1135 __field(int, found)
1126 ), 1136 ),
1127 TP_fast_assign( 1137 TP_fast_assign(
1128 __entry->dev = mp->m_super->s_dev; 1138 __entry->dev = mp->m_super->s_dev;
1129 __entry->agno = agno; 1139 __entry->agno = agno;
1130 __entry->agbno = agbno; 1140 __entry->agbno = agbno;
1131 __entry->len = len; 1141 __entry->len = len;
1132 __entry->lsn = lsn; 1142 __entry->found = found;
1133 ), 1143 ),
1134 TP_printk("dev %d:%d agno %u agbno %u len %u force lsn 0x%llx", 1144 TP_printk("dev %d:%d agno %u agbno %u len %u %s",
1135 MAJOR(__entry->dev), MINOR(__entry->dev), 1145 MAJOR(__entry->dev), MINOR(__entry->dev),
1136 __entry->agno, 1146 __entry->agno,
1137 __entry->agbno, 1147 __entry->agbno,
1138 __entry->len, 1148 __entry->len,
1149 __print_symbolic(__entry->found, XFS_BUSY_STATES))
1150);
1151
1152TRACE_EVENT(xfs_trans_commit_lsn,
1153 TP_PROTO(struct xfs_trans *trans),
1154 TP_ARGS(trans),
1155 TP_STRUCT__entry(
1156 __field(dev_t, dev)
1157 __field(struct xfs_trans *, tp)
1158 __field(xfs_lsn_t, lsn)
1159 ),
1160 TP_fast_assign(
1161 __entry->dev = trans->t_mountp->m_super->s_dev;
1162 __entry->tp = trans;
1163 __entry->lsn = trans->t_commit_lsn;
1164 ),
1165 TP_printk("dev %d:%d trans 0x%p commit_lsn 0x%llx",
1166 MAJOR(__entry->dev), MINOR(__entry->dev),
1167 __entry->tp,
1139 __entry->lsn) 1168 __entry->lsn)
1140); 1169);
1141 1170
diff --git a/fs/xfs/quota/xfs_dquot.c b/fs/xfs/quota/xfs_dquot.c
index b89ec5df0129..585e7633dfc7 100644
--- a/fs/xfs/quota/xfs_dquot.c
+++ b/fs/xfs/quota/xfs_dquot.c
@@ -344,9 +344,9 @@ xfs_qm_init_dquot_blk(
344 for (i = 0; i < q->qi_dqperchunk; i++, d++, curid++) 344 for (i = 0; i < q->qi_dqperchunk; i++, d++, curid++)
345 xfs_qm_dqinit_core(curid, type, d); 345 xfs_qm_dqinit_core(curid, type, d);
346 xfs_trans_dquot_buf(tp, bp, 346 xfs_trans_dquot_buf(tp, bp,
347 (type & XFS_DQ_USER ? XFS_BLI_UDQUOT_BUF : 347 (type & XFS_DQ_USER ? XFS_BLF_UDQUOT_BUF :
348 ((type & XFS_DQ_PROJ) ? XFS_BLI_PDQUOT_BUF : 348 ((type & XFS_DQ_PROJ) ? XFS_BLF_PDQUOT_BUF :
349 XFS_BLI_GDQUOT_BUF))); 349 XFS_BLF_GDQUOT_BUF)));
350 xfs_trans_log_buf(tp, bp, 0, BBTOB(q->qi_dqchunklen) - 1); 350 xfs_trans_log_buf(tp, bp, 0, BBTOB(q->qi_dqchunklen) - 1);
351} 351}
352 352
diff --git a/fs/xfs/xfs_ag.h b/fs/xfs/xfs_ag.h
index abb8222b88c9..401f364ad36c 100644
--- a/fs/xfs/xfs_ag.h
+++ b/fs/xfs/xfs_ag.h
@@ -175,14 +175,20 @@ typedef struct xfs_agfl {
175} xfs_agfl_t; 175} xfs_agfl_t;
176 176
177/* 177/*
178 * Busy block/extent entry. Used in perag to mark blocks that have been freed 178 * Busy block/extent entry. Indexed by a rbtree in perag to mark blocks that
179 * but whose transactions aren't committed to disk yet. 179 * have been freed but whose transactions aren't committed to disk yet.
180 *
181 * Note that we use the transaction ID to record the transaction, not the
182 * transaction structure itself. See xfs_alloc_busy_insert() for details.
180 */ 183 */
181typedef struct xfs_perag_busy { 184struct xfs_busy_extent {
182 xfs_agblock_t busy_start; 185 struct rb_node rb_node; /* ag by-bno indexed search tree */
183 xfs_extlen_t busy_length; 186 struct list_head list; /* transaction busy extent list */
184 struct xfs_trans *busy_tp; /* transaction that did the free */ 187 xfs_agnumber_t agno;
185} xfs_perag_busy_t; 188 xfs_agblock_t bno;
189 xfs_extlen_t length;
190 xlog_tid_t tid; /* transaction that created this */
191};
186 192
187/* 193/*
188 * Per-ag incore structure, copies of information in agf and agi, 194 * Per-ag incore structure, copies of information in agf and agi,
@@ -216,7 +222,8 @@ typedef struct xfs_perag {
216 xfs_agino_t pagl_leftrec; 222 xfs_agino_t pagl_leftrec;
217 xfs_agino_t pagl_rightrec; 223 xfs_agino_t pagl_rightrec;
218#ifdef __KERNEL__ 224#ifdef __KERNEL__
219 spinlock_t pagb_lock; /* lock for pagb_list */ 225 spinlock_t pagb_lock; /* lock for pagb_tree */
226 struct rb_root pagb_tree; /* ordered tree of busy extents */
220 227
221 atomic_t pagf_fstrms; /* # of filestreams active in this AG */ 228 atomic_t pagf_fstrms; /* # of filestreams active in this AG */
222 229
@@ -226,7 +233,6 @@ typedef struct xfs_perag {
226 int pag_ici_reclaimable; /* reclaimable inodes */ 233 int pag_ici_reclaimable; /* reclaimable inodes */
227#endif 234#endif
228 int pagb_count; /* pagb slots in use */ 235 int pagb_count; /* pagb slots in use */
229 xfs_perag_busy_t pagb_list[XFS_PAGB_NUM_SLOTS]; /* unstable blocks */
230} xfs_perag_t; 236} xfs_perag_t;
231 237
232/* 238/*
diff --git a/fs/xfs/xfs_alloc.c b/fs/xfs/xfs_alloc.c
index 94cddbfb2560..a7fbe8a99b12 100644
--- a/fs/xfs/xfs_alloc.c
+++ b/fs/xfs/xfs_alloc.c
@@ -46,11 +46,9 @@
46#define XFSA_FIXUP_BNO_OK 1 46#define XFSA_FIXUP_BNO_OK 1
47#define XFSA_FIXUP_CNT_OK 2 47#define XFSA_FIXUP_CNT_OK 2
48 48
49STATIC void 49static int
50xfs_alloc_search_busy(xfs_trans_t *tp, 50xfs_alloc_busy_search(struct xfs_mount *mp, xfs_agnumber_t agno,
51 xfs_agnumber_t agno, 51 xfs_agblock_t bno, xfs_extlen_t len);
52 xfs_agblock_t bno,
53 xfs_extlen_t len);
54 52
55/* 53/*
56 * Prototypes for per-ag allocation routines 54 * Prototypes for per-ag allocation routines
@@ -540,9 +538,16 @@ xfs_alloc_ag_vextent(
540 be32_to_cpu(agf->agf_length)); 538 be32_to_cpu(agf->agf_length));
541 xfs_alloc_log_agf(args->tp, args->agbp, 539 xfs_alloc_log_agf(args->tp, args->agbp,
542 XFS_AGF_FREEBLKS); 540 XFS_AGF_FREEBLKS);
543 /* search the busylist for these blocks */ 541 /*
544 xfs_alloc_search_busy(args->tp, args->agno, 542 * Search the busylist for these blocks and mark the
545 args->agbno, args->len); 543 * transaction as synchronous if blocks are found. This
544 * avoids the need to block due to a synchronous log
545 * force to ensure correct ordering as the synchronous
546 * transaction will guarantee that for us.
547 */
548 if (xfs_alloc_busy_search(args->mp, args->agno,
549 args->agbno, args->len))
550 xfs_trans_set_sync(args->tp);
546 } 551 }
547 if (!args->isfl) 552 if (!args->isfl)
548 xfs_trans_mod_sb(args->tp, 553 xfs_trans_mod_sb(args->tp,
@@ -1693,7 +1698,7 @@ xfs_free_ag_extent(
1693 * when the iclog commits to disk. If a busy block is allocated, 1698 * when the iclog commits to disk. If a busy block is allocated,
1694 * the iclog is pushed up to the LSN that freed the block. 1699 * the iclog is pushed up to the LSN that freed the block.
1695 */ 1700 */
1696 xfs_alloc_mark_busy(tp, agno, bno, len); 1701 xfs_alloc_busy_insert(tp, agno, bno, len);
1697 return 0; 1702 return 0;
1698 1703
1699 error0: 1704 error0:
@@ -1989,14 +1994,20 @@ xfs_alloc_get_freelist(
1989 *bnop = bno; 1994 *bnop = bno;
1990 1995
1991 /* 1996 /*
1992 * As blocks are freed, they are added to the per-ag busy list 1997 * As blocks are freed, they are added to the per-ag busy list and
1993 * and remain there until the freeing transaction is committed to 1998 * remain there until the freeing transaction is committed to disk.
1994 * disk. Now that we have allocated blocks, this list must be 1999 * Now that we have allocated blocks, this list must be searched to see
1995 * searched to see if a block is being reused. If one is, then 2000 * if a block is being reused. If one is, then the freeing transaction
1996 * the freeing transaction must be pushed to disk NOW by forcing 2001 * must be pushed to disk before this transaction.
1997 * to disk all iclogs up that transaction's LSN. 2002 *
2003 * We do this by setting the current transaction to a sync transaction
2004 * which guarantees that the freeing transaction is on disk before this
2005 * transaction. This is done instead of a synchronous log force here so
2006 * that we don't sit and wait with the AGF locked in the transaction
2007 * during the log force.
1998 */ 2008 */
1999 xfs_alloc_search_busy(tp, be32_to_cpu(agf->agf_seqno), bno, 1); 2009 if (xfs_alloc_busy_search(mp, be32_to_cpu(agf->agf_seqno), bno, 1))
2010 xfs_trans_set_sync(tp);
2000 return 0; 2011 return 0;
2001} 2012}
2002 2013
@@ -2201,7 +2212,7 @@ xfs_alloc_read_agf(
2201 be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNTi]); 2212 be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNTi]);
2202 spin_lock_init(&pag->pagb_lock); 2213 spin_lock_init(&pag->pagb_lock);
2203 pag->pagb_count = 0; 2214 pag->pagb_count = 0;
2204 memset(pag->pagb_list, 0, sizeof(pag->pagb_list)); 2215 pag->pagb_tree = RB_ROOT;
2205 pag->pagf_init = 1; 2216 pag->pagf_init = 1;
2206 } 2217 }
2207#ifdef DEBUG 2218#ifdef DEBUG
@@ -2479,127 +2490,263 @@ error0:
2479 * list is reused, the transaction that freed it must be forced to disk 2490 * list is reused, the transaction that freed it must be forced to disk
2480 * before continuing to use the block. 2491 * before continuing to use the block.
2481 * 2492 *
2482 * xfs_alloc_mark_busy - add to the per-ag busy list 2493 * xfs_alloc_busy_insert - add to the per-ag busy list
2483 * xfs_alloc_clear_busy - remove an item from the per-ag busy list 2494 * xfs_alloc_busy_clear - remove an item from the per-ag busy list
2495 * xfs_alloc_busy_search - search for a busy extent
2496 */
2497
2498/*
2499 * Insert a new extent into the busy tree.
2500 *
2501 * The busy extent tree is indexed by the start block of the busy extent.
2502 * there can be multiple overlapping ranges in the busy extent tree but only
2503 * ever one entry at a given start block. The reason for this is that
2504 * multi-block extents can be freed, then smaller chunks of that extent
2505 * allocated and freed again before the first transaction commit is on disk.
2506 * If the exact same start block is freed a second time, we have to wait for
2507 * that busy extent to pass out of the tree before the new extent is inserted.
2508 * There are two main cases we have to handle here.
2509 *
2510 * The first case is a transaction that triggers a "free - allocate - free"
2511 * cycle. This can occur during btree manipulations as a btree block is freed
2512 * to the freelist, then allocated from the free list, then freed again. In
2513 * this case, the second extxpnet free is what triggers the duplicate and as
2514 * such the transaction IDs should match. Because the extent was allocated in
2515 * this transaction, the transaction must be marked as synchronous. This is
2516 * true for all cases where the free/alloc/free occurs in the one transaction,
2517 * hence the addition of the ASSERT(tp->t_flags & XFS_TRANS_SYNC) to this case.
2518 * This serves to catch violations of the second case quite effectively.
2519 *
2520 * The second case is where the free/alloc/free occur in different
2521 * transactions. In this case, the thread freeing the extent the second time
2522 * can't mark the extent busy immediately because it is already tracked in a
2523 * transaction that may be committing. When the log commit for the existing
2524 * busy extent completes, the busy extent will be removed from the tree. If we
2525 * allow the second busy insert to continue using that busy extent structure,
2526 * it can be freed before this transaction is safely in the log. Hence our
2527 * only option in this case is to force the log to remove the existing busy
2528 * extent from the list before we insert the new one with the current
2529 * transaction ID.
2530 *
2531 * The problem we are trying to avoid in the free-alloc-free in separate
2532 * transactions is most easily described with a timeline:
2533 *
2534 * Thread 1 Thread 2 Thread 3 xfslogd
2535 * xact alloc
2536 * free X
2537 * mark busy
2538 * commit xact
2539 * free xact
2540 * xact alloc
2541 * alloc X
2542 * busy search
2543 * mark xact sync
2544 * commit xact
2545 * free xact
2546 * force log
2547 * checkpoint starts
2548 * ....
2549 * xact alloc
2550 * free X
2551 * mark busy
2552 * finds match
2553 * *** KABOOM! ***
2554 * ....
2555 * log IO completes
2556 * unbusy X
2557 * checkpoint completes
2558 *
2559 * By issuing a log force in thread 3 @ "KABOOM", the thread will block until
2560 * the checkpoint completes, and the busy extent it matched will have been
2561 * removed from the tree when it is woken. Hence it can then continue safely.
2562 *
2563 * However, to ensure this matching process is robust, we need to use the
2564 * transaction ID for identifying transaction, as delayed logging results in
2565 * the busy extent and transaction lifecycles being different. i.e. the busy
2566 * extent is active for a lot longer than the transaction. Hence the
2567 * transaction structure can be freed and reallocated, then mark the same
2568 * extent busy again in the new transaction. In this case the new transaction
2569 * will have a different tid but can have the same address, and hence we need
2570 * to check against the tid.
2571 *
2572 * Future: for delayed logging, we could avoid the log force if the extent was
2573 * first freed in the current checkpoint sequence. This, however, requires the
2574 * ability to pin the current checkpoint in memory until this transaction
2575 * commits to ensure that both the original free and the current one combine
2576 * logically into the one checkpoint. If the checkpoint sequences are
2577 * different, however, we still need to wait on a log force.
2484 */ 2578 */
2485void 2579void
2486xfs_alloc_mark_busy(xfs_trans_t *tp, 2580xfs_alloc_busy_insert(
2487 xfs_agnumber_t agno, 2581 struct xfs_trans *tp,
2488 xfs_agblock_t bno, 2582 xfs_agnumber_t agno,
2489 xfs_extlen_t len) 2583 xfs_agblock_t bno,
2584 xfs_extlen_t len)
2490{ 2585{
2491 xfs_perag_busy_t *bsy; 2586 struct xfs_busy_extent *new;
2587 struct xfs_busy_extent *busyp;
2492 struct xfs_perag *pag; 2588 struct xfs_perag *pag;
2493 int n; 2589 struct rb_node **rbp;
2590 struct rb_node *parent;
2591 int match;
2494 2592
2495 pag = xfs_perag_get(tp->t_mountp, agno);
2496 spin_lock(&pag->pagb_lock);
2497 2593
2498 /* search pagb_list for an open slot */ 2594 new = kmem_zalloc(sizeof(struct xfs_busy_extent), KM_MAYFAIL);
2499 for (bsy = pag->pagb_list, n = 0; 2595 if (!new) {
2500 n < XFS_PAGB_NUM_SLOTS; 2596 /*
2501 bsy++, n++) { 2597 * No Memory! Since it is now not possible to track the free
2502 if (bsy->busy_tp == NULL) { 2598 * block, make this a synchronous transaction to insure that
2503 break; 2599 * the block is not reused before this transaction commits.
2504 } 2600 */
2601 trace_xfs_alloc_busy(tp, agno, bno, len, 1);
2602 xfs_trans_set_sync(tp);
2603 return;
2505 } 2604 }
2506 2605
2507 trace_xfs_alloc_busy(tp->t_mountp, agno, bno, len, n); 2606 new->agno = agno;
2607 new->bno = bno;
2608 new->length = len;
2609 new->tid = xfs_log_get_trans_ident(tp);
2508 2610
2509 if (n < XFS_PAGB_NUM_SLOTS) { 2611 INIT_LIST_HEAD(&new->list);
2510 bsy = &pag->pagb_list[n]; 2612
2511 pag->pagb_count++; 2613 /* trace before insert to be able to see failed inserts */
2512 bsy->busy_start = bno; 2614 trace_xfs_alloc_busy(tp, agno, bno, len, 0);
2513 bsy->busy_length = len; 2615
2514 bsy->busy_tp = tp; 2616 pag = xfs_perag_get(tp->t_mountp, new->agno);
2515 xfs_trans_add_busy(tp, agno, n); 2617restart:
2516 } else { 2618 spin_lock(&pag->pagb_lock);
2619 rbp = &pag->pagb_tree.rb_node;
2620 parent = NULL;
2621 busyp = NULL;
2622 match = 0;
2623 while (*rbp && match >= 0) {
2624 parent = *rbp;
2625 busyp = rb_entry(parent, struct xfs_busy_extent, rb_node);
2626
2627 if (new->bno < busyp->bno) {
2628 /* may overlap, but exact start block is lower */
2629 rbp = &(*rbp)->rb_left;
2630 if (new->bno + new->length > busyp->bno)
2631 match = busyp->tid == new->tid ? 1 : -1;
2632 } else if (new->bno > busyp->bno) {
2633 /* may overlap, but exact start block is higher */
2634 rbp = &(*rbp)->rb_right;
2635 if (bno < busyp->bno + busyp->length)
2636 match = busyp->tid == new->tid ? 1 : -1;
2637 } else {
2638 match = busyp->tid == new->tid ? 1 : -1;
2639 break;
2640 }
2641 }
2642 if (match < 0) {
2643 /* overlap marked busy in different transaction */
2644 spin_unlock(&pag->pagb_lock);
2645 xfs_log_force(tp->t_mountp, XFS_LOG_SYNC);
2646 goto restart;
2647 }
2648 if (match > 0) {
2517 /* 2649 /*
2518 * The busy list is full! Since it is now not possible to 2650 * overlap marked busy in same transaction. Update if exact
2519 * track the free block, make this a synchronous transaction 2651 * start block match, otherwise combine the busy extents into
2520 * to insure that the block is not reused before this 2652 * a single range.
2521 * transaction commits.
2522 */ 2653 */
2523 xfs_trans_set_sync(tp); 2654 if (busyp->bno == new->bno) {
2524 } 2655 busyp->length = max(busyp->length, new->length);
2656 spin_unlock(&pag->pagb_lock);
2657 ASSERT(tp->t_flags & XFS_TRANS_SYNC);
2658 xfs_perag_put(pag);
2659 kmem_free(new);
2660 return;
2661 }
2662 rb_erase(&busyp->rb_node, &pag->pagb_tree);
2663 new->length = max(busyp->bno + busyp->length,
2664 new->bno + new->length) -
2665 min(busyp->bno, new->bno);
2666 new->bno = min(busyp->bno, new->bno);
2667 } else
2668 busyp = NULL;
2525 2669
2670 rb_link_node(&new->rb_node, parent, rbp);
2671 rb_insert_color(&new->rb_node, &pag->pagb_tree);
2672
2673 list_add(&new->list, &tp->t_busy);
2526 spin_unlock(&pag->pagb_lock); 2674 spin_unlock(&pag->pagb_lock);
2527 xfs_perag_put(pag); 2675 xfs_perag_put(pag);
2676 kmem_free(busyp);
2528} 2677}
2529 2678
2530void 2679/*
2531xfs_alloc_clear_busy(xfs_trans_t *tp, 2680 * Search for a busy extent within the range of the extent we are about to
2532 xfs_agnumber_t agno, 2681 * allocate. You need to be holding the busy extent tree lock when calling
2533 int idx) 2682 * xfs_alloc_busy_search(). This function returns 0 for no overlapping busy
2683 * extent, -1 for an overlapping but not exact busy extent, and 1 for an exact
2684 * match. This is done so that a non-zero return indicates an overlap that
2685 * will require a synchronous transaction, but it can still be
2686 * used to distinguish between a partial or exact match.
2687 */
2688static int
2689xfs_alloc_busy_search(
2690 struct xfs_mount *mp,
2691 xfs_agnumber_t agno,
2692 xfs_agblock_t bno,
2693 xfs_extlen_t len)
2534{ 2694{
2535 struct xfs_perag *pag; 2695 struct xfs_perag *pag;
2536 xfs_perag_busy_t *list; 2696 struct rb_node *rbp;
2697 struct xfs_busy_extent *busyp;
2698 int match = 0;
2537 2699
2538 ASSERT(idx < XFS_PAGB_NUM_SLOTS); 2700 pag = xfs_perag_get(mp, agno);
2539 pag = xfs_perag_get(tp->t_mountp, agno);
2540 spin_lock(&pag->pagb_lock); 2701 spin_lock(&pag->pagb_lock);
2541 list = pag->pagb_list;
2542 2702
2543 trace_xfs_alloc_unbusy(tp->t_mountp, agno, idx, list[idx].busy_tp == tp); 2703 rbp = pag->pagb_tree.rb_node;
2544 2704
2545 if (list[idx].busy_tp == tp) { 2705 /* find closest start bno overlap */
2546 list[idx].busy_tp = NULL; 2706 while (rbp) {
2547 pag->pagb_count--; 2707 busyp = rb_entry(rbp, struct xfs_busy_extent, rb_node);
2708 if (bno < busyp->bno) {
2709 /* may overlap, but exact start block is lower */
2710 if (bno + len > busyp->bno)
2711 match = -1;
2712 rbp = rbp->rb_left;
2713 } else if (bno > busyp->bno) {
2714 /* may overlap, but exact start block is higher */
2715 if (bno < busyp->bno + busyp->length)
2716 match = -1;
2717 rbp = rbp->rb_right;
2718 } else {
2719 /* bno matches busyp, length determines exact match */
2720 match = (busyp->length == len) ? 1 : -1;
2721 break;
2722 }
2548 } 2723 }
2549
2550 spin_unlock(&pag->pagb_lock); 2724 spin_unlock(&pag->pagb_lock);
2725 trace_xfs_alloc_busysearch(mp, agno, bno, len, !!match);
2551 xfs_perag_put(pag); 2726 xfs_perag_put(pag);
2727 return match;
2552} 2728}
2553 2729
2554 2730void
2555/* 2731xfs_alloc_busy_clear(
2556 * If we find the extent in the busy list, force the log out to get the 2732 struct xfs_mount *mp,
2557 * extent out of the busy list so the caller can use it straight away. 2733 struct xfs_busy_extent *busyp)
2558 */
2559STATIC void
2560xfs_alloc_search_busy(xfs_trans_t *tp,
2561 xfs_agnumber_t agno,
2562 xfs_agblock_t bno,
2563 xfs_extlen_t len)
2564{ 2734{
2565 struct xfs_perag *pag; 2735 struct xfs_perag *pag;
2566 xfs_perag_busy_t *bsy;
2567 xfs_agblock_t uend, bend;
2568 xfs_lsn_t lsn = 0;
2569 int cnt;
2570 2736
2571 pag = xfs_perag_get(tp->t_mountp, agno); 2737 trace_xfs_alloc_unbusy(mp, busyp->agno, busyp->bno,
2572 spin_lock(&pag->pagb_lock); 2738 busyp->length);
2573 cnt = pag->pagb_count;
2574 2739
2575 /* 2740 ASSERT(xfs_alloc_busy_search(mp, busyp->agno, busyp->bno,
2576 * search pagb_list for this slot, skipping open slots. We have to 2741 busyp->length) == 1);
2577 * search the entire array as there may be multiple overlaps and
2578 * we have to get the most recent LSN for the log force to push out
2579 * all the transactions that span the range.
2580 */
2581 uend = bno + len - 1;
2582 for (cnt = 0; cnt < pag->pagb_count; cnt++) {
2583 bsy = &pag->pagb_list[cnt];
2584 if (!bsy->busy_tp)
2585 continue;
2586 2742
2587 bend = bsy->busy_start + bsy->busy_length - 1; 2743 list_del_init(&busyp->list);
2588 if (bno > bend || uend < bsy->busy_start)
2589 continue;
2590 2744
2591 /* (start1,length1) within (start2, length2) */ 2745 pag = xfs_perag_get(mp, busyp->agno);
2592 if (XFS_LSN_CMP(bsy->busy_tp->t_commit_lsn, lsn) > 0) 2746 spin_lock(&pag->pagb_lock);
2593 lsn = bsy->busy_tp->t_commit_lsn; 2747 rb_erase(&busyp->rb_node, &pag->pagb_tree);
2594 }
2595 spin_unlock(&pag->pagb_lock); 2748 spin_unlock(&pag->pagb_lock);
2596 xfs_perag_put(pag); 2749 xfs_perag_put(pag);
2597 trace_xfs_alloc_busysearch(tp->t_mountp, agno, bno, len, lsn);
2598 2750
2599 /* 2751 kmem_free(busyp);
2600 * If a block was found, force the log through the LSN of the
2601 * transaction that freed the block
2602 */
2603 if (lsn)
2604 xfs_log_force_lsn(tp->t_mountp, lsn, XFS_LOG_SYNC);
2605} 2752}
diff --git a/fs/xfs/xfs_alloc.h b/fs/xfs/xfs_alloc.h
index 599bffa39784..6d05199b667c 100644
--- a/fs/xfs/xfs_alloc.h
+++ b/fs/xfs/xfs_alloc.h
@@ -22,6 +22,7 @@ struct xfs_buf;
22struct xfs_mount; 22struct xfs_mount;
23struct xfs_perag; 23struct xfs_perag;
24struct xfs_trans; 24struct xfs_trans;
25struct xfs_busy_extent;
25 26
26/* 27/*
27 * Freespace allocation types. Argument to xfs_alloc_[v]extent. 28 * Freespace allocation types. Argument to xfs_alloc_[v]extent.
@@ -119,15 +120,13 @@ xfs_alloc_longest_free_extent(struct xfs_mount *mp,
119#ifdef __KERNEL__ 120#ifdef __KERNEL__
120 121
121void 122void
122xfs_alloc_mark_busy(xfs_trans_t *tp, 123xfs_alloc_busy_insert(xfs_trans_t *tp,
123 xfs_agnumber_t agno, 124 xfs_agnumber_t agno,
124 xfs_agblock_t bno, 125 xfs_agblock_t bno,
125 xfs_extlen_t len); 126 xfs_extlen_t len);
126 127
127void 128void
128xfs_alloc_clear_busy(xfs_trans_t *tp, 129xfs_alloc_busy_clear(struct xfs_mount *mp, struct xfs_busy_extent *busyp);
129 xfs_agnumber_t ag,
130 int idx);
131 130
132#endif /* __KERNEL__ */ 131#endif /* __KERNEL__ */
133 132
diff --git a/fs/xfs/xfs_alloc_btree.c b/fs/xfs/xfs_alloc_btree.c
index b726e10d2c1c..83f494218759 100644
--- a/fs/xfs/xfs_alloc_btree.c
+++ b/fs/xfs/xfs_alloc_btree.c
@@ -134,7 +134,7 @@ xfs_allocbt_free_block(
134 * disk. If a busy block is allocated, the iclog is pushed up to the 134 * disk. If a busy block is allocated, the iclog is pushed up to the
135 * LSN that freed the block. 135 * LSN that freed the block.
136 */ 136 */
137 xfs_alloc_mark_busy(cur->bc_tp, be32_to_cpu(agf->agf_seqno), bno, 1); 137 xfs_alloc_busy_insert(cur->bc_tp, be32_to_cpu(agf->agf_seqno), bno, 1);
138 xfs_trans_agbtree_delta(cur->bc_tp, -1); 138 xfs_trans_agbtree_delta(cur->bc_tp, -1);
139 return 0; 139 return 0;
140} 140}
diff --git a/fs/xfs/xfs_buf_item.c b/fs/xfs/xfs_buf_item.c
index 240340a4727b..02a80984aa05 100644
--- a/fs/xfs/xfs_buf_item.c
+++ b/fs/xfs/xfs_buf_item.c
@@ -64,7 +64,7 @@ xfs_buf_item_log_debug(
64 nbytes = last - first + 1; 64 nbytes = last - first + 1;
65 bfset(bip->bli_logged, first, nbytes); 65 bfset(bip->bli_logged, first, nbytes);
66 for (x = 0; x < nbytes; x++) { 66 for (x = 0; x < nbytes; x++) {
67 chunk_num = byte >> XFS_BLI_SHIFT; 67 chunk_num = byte >> XFS_BLF_SHIFT;
68 word_num = chunk_num >> BIT_TO_WORD_SHIFT; 68 word_num = chunk_num >> BIT_TO_WORD_SHIFT;
69 bit_num = chunk_num & (NBWORD - 1); 69 bit_num = chunk_num & (NBWORD - 1);
70 wordp = &(bip->bli_format.blf_data_map[word_num]); 70 wordp = &(bip->bli_format.blf_data_map[word_num]);
@@ -166,7 +166,7 @@ xfs_buf_item_size(
166 * cancel flag in it. 166 * cancel flag in it.
167 */ 167 */
168 trace_xfs_buf_item_size_stale(bip); 168 trace_xfs_buf_item_size_stale(bip);
169 ASSERT(bip->bli_format.blf_flags & XFS_BLI_CANCEL); 169 ASSERT(bip->bli_format.blf_flags & XFS_BLF_CANCEL);
170 return 1; 170 return 1;
171 } 171 }
172 172
@@ -197,9 +197,9 @@ xfs_buf_item_size(
197 } else if (next_bit != last_bit + 1) { 197 } else if (next_bit != last_bit + 1) {
198 last_bit = next_bit; 198 last_bit = next_bit;
199 nvecs++; 199 nvecs++;
200 } else if (xfs_buf_offset(bp, next_bit * XFS_BLI_CHUNK) != 200 } else if (xfs_buf_offset(bp, next_bit * XFS_BLF_CHUNK) !=
201 (xfs_buf_offset(bp, last_bit * XFS_BLI_CHUNK) + 201 (xfs_buf_offset(bp, last_bit * XFS_BLF_CHUNK) +
202 XFS_BLI_CHUNK)) { 202 XFS_BLF_CHUNK)) {
203 last_bit = next_bit; 203 last_bit = next_bit;
204 nvecs++; 204 nvecs++;
205 } else { 205 } else {
@@ -254,6 +254,20 @@ xfs_buf_item_format(
254 vecp++; 254 vecp++;
255 nvecs = 1; 255 nvecs = 1;
256 256
257 /*
258 * If it is an inode buffer, transfer the in-memory state to the
259 * format flags and clear the in-memory state. We do not transfer
260 * this state if the inode buffer allocation has not yet been committed
261 * to the log as setting the XFS_BLI_INODE_BUF flag will prevent
262 * correct replay of the inode allocation.
263 */
264 if (bip->bli_flags & XFS_BLI_INODE_BUF) {
265 if (!((bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF) &&
266 xfs_log_item_in_current_chkpt(&bip->bli_item)))
267 bip->bli_format.blf_flags |= XFS_BLF_INODE_BUF;
268 bip->bli_flags &= ~XFS_BLI_INODE_BUF;
269 }
270
257 if (bip->bli_flags & XFS_BLI_STALE) { 271 if (bip->bli_flags & XFS_BLI_STALE) {
258 /* 272 /*
259 * The buffer is stale, so all we need to log 273 * The buffer is stale, so all we need to log
@@ -261,7 +275,7 @@ xfs_buf_item_format(
261 * cancel flag in it. 275 * cancel flag in it.
262 */ 276 */
263 trace_xfs_buf_item_format_stale(bip); 277 trace_xfs_buf_item_format_stale(bip);
264 ASSERT(bip->bli_format.blf_flags & XFS_BLI_CANCEL); 278 ASSERT(bip->bli_format.blf_flags & XFS_BLF_CANCEL);
265 bip->bli_format.blf_size = nvecs; 279 bip->bli_format.blf_size = nvecs;
266 return; 280 return;
267 } 281 }
@@ -294,28 +308,28 @@ xfs_buf_item_format(
294 * keep counting and scanning. 308 * keep counting and scanning.
295 */ 309 */
296 if (next_bit == -1) { 310 if (next_bit == -1) {
297 buffer_offset = first_bit * XFS_BLI_CHUNK; 311 buffer_offset = first_bit * XFS_BLF_CHUNK;
298 vecp->i_addr = xfs_buf_offset(bp, buffer_offset); 312 vecp->i_addr = xfs_buf_offset(bp, buffer_offset);
299 vecp->i_len = nbits * XFS_BLI_CHUNK; 313 vecp->i_len = nbits * XFS_BLF_CHUNK;
300 vecp->i_type = XLOG_REG_TYPE_BCHUNK; 314 vecp->i_type = XLOG_REG_TYPE_BCHUNK;
301 nvecs++; 315 nvecs++;
302 break; 316 break;
303 } else if (next_bit != last_bit + 1) { 317 } else if (next_bit != last_bit + 1) {
304 buffer_offset = first_bit * XFS_BLI_CHUNK; 318 buffer_offset = first_bit * XFS_BLF_CHUNK;
305 vecp->i_addr = xfs_buf_offset(bp, buffer_offset); 319 vecp->i_addr = xfs_buf_offset(bp, buffer_offset);
306 vecp->i_len = nbits * XFS_BLI_CHUNK; 320 vecp->i_len = nbits * XFS_BLF_CHUNK;
307 vecp->i_type = XLOG_REG_TYPE_BCHUNK; 321 vecp->i_type = XLOG_REG_TYPE_BCHUNK;
308 nvecs++; 322 nvecs++;
309 vecp++; 323 vecp++;
310 first_bit = next_bit; 324 first_bit = next_bit;
311 last_bit = next_bit; 325 last_bit = next_bit;
312 nbits = 1; 326 nbits = 1;
313 } else if (xfs_buf_offset(bp, next_bit << XFS_BLI_SHIFT) != 327 } else if (xfs_buf_offset(bp, next_bit << XFS_BLF_SHIFT) !=
314 (xfs_buf_offset(bp, last_bit << XFS_BLI_SHIFT) + 328 (xfs_buf_offset(bp, last_bit << XFS_BLF_SHIFT) +
315 XFS_BLI_CHUNK)) { 329 XFS_BLF_CHUNK)) {
316 buffer_offset = first_bit * XFS_BLI_CHUNK; 330 buffer_offset = first_bit * XFS_BLF_CHUNK;
317 vecp->i_addr = xfs_buf_offset(bp, buffer_offset); 331 vecp->i_addr = xfs_buf_offset(bp, buffer_offset);
318 vecp->i_len = nbits * XFS_BLI_CHUNK; 332 vecp->i_len = nbits * XFS_BLF_CHUNK;
319 vecp->i_type = XLOG_REG_TYPE_BCHUNK; 333 vecp->i_type = XLOG_REG_TYPE_BCHUNK;
320/* You would think we need to bump the nvecs here too, but we do not 334/* You would think we need to bump the nvecs here too, but we do not
321 * this number is used by recovery, and it gets confused by the boundary 335 * this number is used by recovery, and it gets confused by the boundary
@@ -341,10 +355,15 @@ xfs_buf_item_format(
341} 355}
342 356
343/* 357/*
344 * This is called to pin the buffer associated with the buf log 358 * This is called to pin the buffer associated with the buf log item in memory
345 * item in memory so it cannot be written out. Simply call bpin() 359 * so it cannot be written out. Simply call bpin() on the buffer to do this.
346 * on the buffer to do this. 360 *
361 * We also always take a reference to the buffer log item here so that the bli
362 * is held while the item is pinned in memory. This means that we can
363 * unconditionally drop the reference count a transaction holds when the
364 * transaction is completed.
347 */ 365 */
366
348STATIC void 367STATIC void
349xfs_buf_item_pin( 368xfs_buf_item_pin(
350 xfs_buf_log_item_t *bip) 369 xfs_buf_log_item_t *bip)
@@ -356,6 +375,7 @@ xfs_buf_item_pin(
356 ASSERT(atomic_read(&bip->bli_refcount) > 0); 375 ASSERT(atomic_read(&bip->bli_refcount) > 0);
357 ASSERT((bip->bli_flags & XFS_BLI_LOGGED) || 376 ASSERT((bip->bli_flags & XFS_BLI_LOGGED) ||
358 (bip->bli_flags & XFS_BLI_STALE)); 377 (bip->bli_flags & XFS_BLI_STALE));
378 atomic_inc(&bip->bli_refcount);
359 trace_xfs_buf_item_pin(bip); 379 trace_xfs_buf_item_pin(bip);
360 xfs_bpin(bp); 380 xfs_bpin(bp);
361} 381}
@@ -393,7 +413,7 @@ xfs_buf_item_unpin(
393 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0); 413 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
394 ASSERT(!(XFS_BUF_ISDELAYWRITE(bp))); 414 ASSERT(!(XFS_BUF_ISDELAYWRITE(bp)));
395 ASSERT(XFS_BUF_ISSTALE(bp)); 415 ASSERT(XFS_BUF_ISSTALE(bp));
396 ASSERT(bip->bli_format.blf_flags & XFS_BLI_CANCEL); 416 ASSERT(bip->bli_format.blf_flags & XFS_BLF_CANCEL);
397 trace_xfs_buf_item_unpin_stale(bip); 417 trace_xfs_buf_item_unpin_stale(bip);
398 418
399 /* 419 /*
@@ -489,20 +509,23 @@ xfs_buf_item_trylock(
489} 509}
490 510
491/* 511/*
492 * Release the buffer associated with the buf log item. 512 * Release the buffer associated with the buf log item. If there is no dirty
493 * If there is no dirty logged data associated with the 513 * logged data associated with the buffer recorded in the buf log item, then
494 * buffer recorded in the buf log item, then free the 514 * free the buf log item and remove the reference to it in the buffer.
495 * buf log item and remove the reference to it in the 515 *
496 * buffer. 516 * This call ignores the recursion count. It is only called when the buffer
517 * should REALLY be unlocked, regardless of the recursion count.
497 * 518 *
498 * This call ignores the recursion count. It is only called 519 * We unconditionally drop the transaction's reference to the log item. If the
499 * when the buffer should REALLY be unlocked, regardless 520 * item was logged, then another reference was taken when it was pinned, so we
500 * of the recursion count. 521 * can safely drop the transaction reference now. This also allows us to avoid
522 * potential races with the unpin code freeing the bli by not referencing the
523 * bli after we've dropped the reference count.
501 * 524 *
502 * If the XFS_BLI_HOLD flag is set in the buf log item, then 525 * If the XFS_BLI_HOLD flag is set in the buf log item, then free the log item
503 * free the log item if necessary but do not unlock the buffer. 526 * if necessary but do not unlock the buffer. This is for support of
504 * This is for support of xfs_trans_bhold(). Make sure the 527 * xfs_trans_bhold(). Make sure the XFS_BLI_HOLD field is cleared if we don't
505 * XFS_BLI_HOLD field is cleared if we don't free the item. 528 * free the item.
506 */ 529 */
507STATIC void 530STATIC void
508xfs_buf_item_unlock( 531xfs_buf_item_unlock(
@@ -514,73 +537,54 @@ xfs_buf_item_unlock(
514 537
515 bp = bip->bli_buf; 538 bp = bip->bli_buf;
516 539
517 /* 540 /* Clear the buffer's association with this transaction. */
518 * Clear the buffer's association with this transaction.
519 */
520 XFS_BUF_SET_FSPRIVATE2(bp, NULL); 541 XFS_BUF_SET_FSPRIVATE2(bp, NULL);
521 542
522 /* 543 /*
523 * If this is a transaction abort, don't return early. 544 * If this is a transaction abort, don't return early. Instead, allow
524 * Instead, allow the brelse to happen. 545 * the brelse to happen. Normally it would be done for stale
525 * Normally it would be done for stale (cancelled) buffers 546 * (cancelled) buffers at unpin time, but we'll never go through the
526 * at unpin time, but we'll never go through the pin/unpin 547 * pin/unpin cycle if we abort inside commit.
527 * cycle if we abort inside commit.
528 */ 548 */
529 aborted = (bip->bli_item.li_flags & XFS_LI_ABORTED) != 0; 549 aborted = (bip->bli_item.li_flags & XFS_LI_ABORTED) != 0;
530 550
531 /* 551 /*
532 * If the buf item is marked stale, then don't do anything. 552 * Before possibly freeing the buf item, determine if we should
533 * We'll unlock the buffer and free the buf item when the 553 * release the buffer at the end of this routine.
534 * buffer is unpinned for the last time.
535 */ 554 */
536 if (bip->bli_flags & XFS_BLI_STALE) { 555 hold = bip->bli_flags & XFS_BLI_HOLD;
537 bip->bli_flags &= ~XFS_BLI_LOGGED; 556
538 trace_xfs_buf_item_unlock_stale(bip); 557 /* Clear the per transaction state. */
539 ASSERT(bip->bli_format.blf_flags & XFS_BLI_CANCEL); 558 bip->bli_flags &= ~(XFS_BLI_LOGGED | XFS_BLI_HOLD);
540 if (!aborted)
541 return;
542 }
543 559
544 /* 560 /*
545 * Drop the transaction's reference to the log item if 561 * If the buf item is marked stale, then don't do anything. We'll
546 * it was not logged as part of the transaction. Otherwise 562 * unlock the buffer and free the buf item when the buffer is unpinned
547 * we'll drop the reference in xfs_buf_item_unpin() when 563 * for the last time.
548 * the transaction is really through with the buffer.
549 */ 564 */
550 if (!(bip->bli_flags & XFS_BLI_LOGGED)) { 565 if (bip->bli_flags & XFS_BLI_STALE) {
551 atomic_dec(&bip->bli_refcount); 566 trace_xfs_buf_item_unlock_stale(bip);
552 } else { 567 ASSERT(bip->bli_format.blf_flags & XFS_BLF_CANCEL);
553 /* 568 if (!aborted) {
554 * Clear the logged flag since this is per 569 atomic_dec(&bip->bli_refcount);
555 * transaction state. 570 return;
556 */ 571 }
557 bip->bli_flags &= ~XFS_BLI_LOGGED;
558 } 572 }
559 573
560 /*
561 * Before possibly freeing the buf item, determine if we should
562 * release the buffer at the end of this routine.
563 */
564 hold = bip->bli_flags & XFS_BLI_HOLD;
565 trace_xfs_buf_item_unlock(bip); 574 trace_xfs_buf_item_unlock(bip);
566 575
567 /* 576 /*
568 * If the buf item isn't tracking any data, free it. 577 * If the buf item isn't tracking any data, free it, otherwise drop the
569 * Otherwise, if XFS_BLI_HOLD is set clear it. 578 * reference we hold to it.
570 */ 579 */
571 if (xfs_bitmap_empty(bip->bli_format.blf_data_map, 580 if (xfs_bitmap_empty(bip->bli_format.blf_data_map,
572 bip->bli_format.blf_map_size)) { 581 bip->bli_format.blf_map_size))
573 xfs_buf_item_relse(bp); 582 xfs_buf_item_relse(bp);
574 } else if (hold) { 583 else
575 bip->bli_flags &= ~XFS_BLI_HOLD; 584 atomic_dec(&bip->bli_refcount);
576 }
577 585
578 /* 586 if (!hold)
579 * Release the buffer if XFS_BLI_HOLD was not set.
580 */
581 if (!hold) {
582 xfs_buf_relse(bp); 587 xfs_buf_relse(bp);
583 }
584} 588}
585 589
586/* 590/*
@@ -717,12 +721,12 @@ xfs_buf_item_init(
717 } 721 }
718 722
719 /* 723 /*
720 * chunks is the number of XFS_BLI_CHUNK size pieces 724 * chunks is the number of XFS_BLF_CHUNK size pieces
721 * the buffer can be divided into. Make sure not to 725 * the buffer can be divided into. Make sure not to
722 * truncate any pieces. map_size is the size of the 726 * truncate any pieces. map_size is the size of the
723 * bitmap needed to describe the chunks of the buffer. 727 * bitmap needed to describe the chunks of the buffer.
724 */ 728 */
725 chunks = (int)((XFS_BUF_COUNT(bp) + (XFS_BLI_CHUNK - 1)) >> XFS_BLI_SHIFT); 729 chunks = (int)((XFS_BUF_COUNT(bp) + (XFS_BLF_CHUNK - 1)) >> XFS_BLF_SHIFT);
726 map_size = (int)((chunks + NBWORD) >> BIT_TO_WORD_SHIFT); 730 map_size = (int)((chunks + NBWORD) >> BIT_TO_WORD_SHIFT);
727 731
728 bip = (xfs_buf_log_item_t*)kmem_zone_zalloc(xfs_buf_item_zone, 732 bip = (xfs_buf_log_item_t*)kmem_zone_zalloc(xfs_buf_item_zone,
@@ -790,8 +794,8 @@ xfs_buf_item_log(
790 /* 794 /*
791 * Convert byte offsets to bit numbers. 795 * Convert byte offsets to bit numbers.
792 */ 796 */
793 first_bit = first >> XFS_BLI_SHIFT; 797 first_bit = first >> XFS_BLF_SHIFT;
794 last_bit = last >> XFS_BLI_SHIFT; 798 last_bit = last >> XFS_BLF_SHIFT;
795 799
796 /* 800 /*
797 * Calculate the total number of bits to be set. 801 * Calculate the total number of bits to be set.
diff --git a/fs/xfs/xfs_buf_item.h b/fs/xfs/xfs_buf_item.h
index df4454511f73..f20bb472d582 100644
--- a/fs/xfs/xfs_buf_item.h
+++ b/fs/xfs/xfs_buf_item.h
@@ -41,22 +41,22 @@ typedef struct xfs_buf_log_format {
41 * This flag indicates that the buffer contains on disk inodes 41 * This flag indicates that the buffer contains on disk inodes
42 * and requires special recovery handling. 42 * and requires special recovery handling.
43 */ 43 */
44#define XFS_BLI_INODE_BUF 0x1 44#define XFS_BLF_INODE_BUF 0x1
45/* 45/*
46 * This flag indicates that the buffer should not be replayed 46 * This flag indicates that the buffer should not be replayed
47 * during recovery because its blocks are being freed. 47 * during recovery because its blocks are being freed.
48 */ 48 */
49#define XFS_BLI_CANCEL 0x2 49#define XFS_BLF_CANCEL 0x2
50/* 50/*
51 * This flag indicates that the buffer contains on disk 51 * This flag indicates that the buffer contains on disk
52 * user or group dquots and may require special recovery handling. 52 * user or group dquots and may require special recovery handling.
53 */ 53 */
54#define XFS_BLI_UDQUOT_BUF 0x4 54#define XFS_BLF_UDQUOT_BUF 0x4
55#define XFS_BLI_PDQUOT_BUF 0x8 55#define XFS_BLF_PDQUOT_BUF 0x8
56#define XFS_BLI_GDQUOT_BUF 0x10 56#define XFS_BLF_GDQUOT_BUF 0x10
57 57
58#define XFS_BLI_CHUNK 128 58#define XFS_BLF_CHUNK 128
59#define XFS_BLI_SHIFT 7 59#define XFS_BLF_SHIFT 7
60#define BIT_TO_WORD_SHIFT 5 60#define BIT_TO_WORD_SHIFT 5
61#define NBWORD (NBBY * sizeof(unsigned int)) 61#define NBWORD (NBBY * sizeof(unsigned int))
62 62
@@ -69,6 +69,7 @@ typedef struct xfs_buf_log_format {
69#define XFS_BLI_LOGGED 0x08 69#define XFS_BLI_LOGGED 0x08
70#define XFS_BLI_INODE_ALLOC_BUF 0x10 70#define XFS_BLI_INODE_ALLOC_BUF 0x10
71#define XFS_BLI_STALE_INODE 0x20 71#define XFS_BLI_STALE_INODE 0x20
72#define XFS_BLI_INODE_BUF 0x40
72 73
73#define XFS_BLI_FLAGS \ 74#define XFS_BLI_FLAGS \
74 { XFS_BLI_HOLD, "HOLD" }, \ 75 { XFS_BLI_HOLD, "HOLD" }, \
@@ -76,7 +77,8 @@ typedef struct xfs_buf_log_format {
76 { XFS_BLI_STALE, "STALE" }, \ 77 { XFS_BLI_STALE, "STALE" }, \
77 { XFS_BLI_LOGGED, "LOGGED" }, \ 78 { XFS_BLI_LOGGED, "LOGGED" }, \
78 { XFS_BLI_INODE_ALLOC_BUF, "INODE_ALLOC" }, \ 79 { XFS_BLI_INODE_ALLOC_BUF, "INODE_ALLOC" }, \
79 { XFS_BLI_STALE_INODE, "STALE_INODE" } 80 { XFS_BLI_STALE_INODE, "STALE_INODE" }, \
81 { XFS_BLI_INODE_BUF, "INODE_BUF" }
80 82
81 83
82#ifdef __KERNEL__ 84#ifdef __KERNEL__
diff --git a/fs/xfs/xfs_error.c b/fs/xfs/xfs_error.c
index ef96175c0744..047b8a8e5c29 100644
--- a/fs/xfs/xfs_error.c
+++ b/fs/xfs/xfs_error.c
@@ -170,7 +170,7 @@ xfs_cmn_err(int panic_tag, int level, xfs_mount_t *mp, char *fmt, ...)
170 va_list ap; 170 va_list ap;
171 171
172#ifdef DEBUG 172#ifdef DEBUG
173 xfs_panic_mask |= XFS_PTAG_SHUTDOWN_CORRUPT; 173 xfs_panic_mask |= (XFS_PTAG_SHUTDOWN_CORRUPT | XFS_PTAG_LOGRES);
174#endif 174#endif
175 175
176 if (xfs_panic_mask && (xfs_panic_mask & panic_tag) 176 if (xfs_panic_mask && (xfs_panic_mask & panic_tag)
diff --git a/fs/xfs/xfs_log.c b/fs/xfs/xfs_log.c
index 3038dd52c72a..5215abc8023a 100644
--- a/fs/xfs/xfs_log.c
+++ b/fs/xfs/xfs_log.c
@@ -54,9 +54,6 @@ STATIC xlog_t * xlog_alloc_log(xfs_mount_t *mp,
54STATIC int xlog_space_left(xlog_t *log, int cycle, int bytes); 54STATIC int xlog_space_left(xlog_t *log, int cycle, int bytes);
55STATIC int xlog_sync(xlog_t *log, xlog_in_core_t *iclog); 55STATIC int xlog_sync(xlog_t *log, xlog_in_core_t *iclog);
56STATIC void xlog_dealloc_log(xlog_t *log); 56STATIC void xlog_dealloc_log(xlog_t *log);
57STATIC int xlog_write(struct log *log, struct xfs_log_vec *log_vector,
58 struct xlog_ticket *tic, xfs_lsn_t *start_lsn,
59 xlog_in_core_t **commit_iclog, uint flags);
60 57
61/* local state machine functions */ 58/* local state machine functions */
62STATIC void xlog_state_done_syncing(xlog_in_core_t *iclog, int); 59STATIC void xlog_state_done_syncing(xlog_in_core_t *iclog, int);
@@ -86,14 +83,6 @@ STATIC int xlog_regrant_write_log_space(xlog_t *log,
86STATIC void xlog_ungrant_log_space(xlog_t *log, 83STATIC void xlog_ungrant_log_space(xlog_t *log,
87 xlog_ticket_t *ticket); 84 xlog_ticket_t *ticket);
88 85
89
90/* local ticket functions */
91STATIC xlog_ticket_t *xlog_ticket_alloc(xlog_t *log,
92 int unit_bytes,
93 int count,
94 char clientid,
95 uint flags);
96
97#if defined(DEBUG) 86#if defined(DEBUG)
98STATIC void xlog_verify_dest_ptr(xlog_t *log, char *ptr); 87STATIC void xlog_verify_dest_ptr(xlog_t *log, char *ptr);
99STATIC void xlog_verify_grant_head(xlog_t *log, int equals); 88STATIC void xlog_verify_grant_head(xlog_t *log, int equals);
@@ -360,6 +349,15 @@ xfs_log_reserve(
360 ASSERT(flags & XFS_LOG_PERM_RESERV); 349 ASSERT(flags & XFS_LOG_PERM_RESERV);
361 internal_ticket = *ticket; 350 internal_ticket = *ticket;
362 351
352 /*
353 * this is a new transaction on the ticket, so we need to
354 * change the transaction ID so that the next transaction has a
355 * different TID in the log. Just add one to the existing tid
356 * so that we can see chains of rolling transactions in the log
357 * easily.
358 */
359 internal_ticket->t_tid++;
360
363 trace_xfs_log_reserve(log, internal_ticket); 361 trace_xfs_log_reserve(log, internal_ticket);
364 362
365 xlog_grant_push_ail(mp, internal_ticket->t_unit_res); 363 xlog_grant_push_ail(mp, internal_ticket->t_unit_res);
@@ -367,7 +365,8 @@ xfs_log_reserve(
367 } else { 365 } else {
368 /* may sleep if need to allocate more tickets */ 366 /* may sleep if need to allocate more tickets */
369 internal_ticket = xlog_ticket_alloc(log, unit_bytes, cnt, 367 internal_ticket = xlog_ticket_alloc(log, unit_bytes, cnt,
370 client, flags); 368 client, flags,
369 KM_SLEEP|KM_MAYFAIL);
371 if (!internal_ticket) 370 if (!internal_ticket)
372 return XFS_ERROR(ENOMEM); 371 return XFS_ERROR(ENOMEM);
373 internal_ticket->t_trans_type = t_type; 372 internal_ticket->t_trans_type = t_type;
@@ -452,6 +451,13 @@ xfs_log_mount(
452 /* Normal transactions can now occur */ 451 /* Normal transactions can now occur */
453 mp->m_log->l_flags &= ~XLOG_ACTIVE_RECOVERY; 452 mp->m_log->l_flags &= ~XLOG_ACTIVE_RECOVERY;
454 453
454 /*
455 * Now the log has been fully initialised and we know were our
456 * space grant counters are, we can initialise the permanent ticket
457 * needed for delayed logging to work.
458 */
459 xlog_cil_init_post_recovery(mp->m_log);
460
455 return 0; 461 return 0;
456 462
457out_destroy_ail: 463out_destroy_ail:
@@ -658,6 +664,10 @@ xfs_log_item_init(
658 item->li_ailp = mp->m_ail; 664 item->li_ailp = mp->m_ail;
659 item->li_type = type; 665 item->li_type = type;
660 item->li_ops = ops; 666 item->li_ops = ops;
667 item->li_lv = NULL;
668
669 INIT_LIST_HEAD(&item->li_ail);
670 INIT_LIST_HEAD(&item->li_cil);
661} 671}
662 672
663/* 673/*
@@ -1168,6 +1178,9 @@ xlog_alloc_log(xfs_mount_t *mp,
1168 *iclogp = log->l_iclog; /* complete ring */ 1178 *iclogp = log->l_iclog; /* complete ring */
1169 log->l_iclog->ic_prev = prev_iclog; /* re-write 1st prev ptr */ 1179 log->l_iclog->ic_prev = prev_iclog; /* re-write 1st prev ptr */
1170 1180
1181 error = xlog_cil_init(log);
1182 if (error)
1183 goto out_free_iclog;
1171 return log; 1184 return log;
1172 1185
1173out_free_iclog: 1186out_free_iclog:
@@ -1494,6 +1507,8 @@ xlog_dealloc_log(xlog_t *log)
1494 xlog_in_core_t *iclog, *next_iclog; 1507 xlog_in_core_t *iclog, *next_iclog;
1495 int i; 1508 int i;
1496 1509
1510 xlog_cil_destroy(log);
1511
1497 iclog = log->l_iclog; 1512 iclog = log->l_iclog;
1498 for (i=0; i<log->l_iclog_bufs; i++) { 1513 for (i=0; i<log->l_iclog_bufs; i++) {
1499 sv_destroy(&iclog->ic_force_wait); 1514 sv_destroy(&iclog->ic_force_wait);
@@ -1536,8 +1551,10 @@ xlog_state_finish_copy(xlog_t *log,
1536 * print out info relating to regions written which consume 1551 * print out info relating to regions written which consume
1537 * the reservation 1552 * the reservation
1538 */ 1553 */
1539STATIC void 1554void
1540xlog_print_tic_res(xfs_mount_t *mp, xlog_ticket_t *ticket) 1555xlog_print_tic_res(
1556 struct xfs_mount *mp,
1557 struct xlog_ticket *ticket)
1541{ 1558{
1542 uint i; 1559 uint i;
1543 uint ophdr_spc = ticket->t_res_num_ophdrs * (uint)sizeof(xlog_op_header_t); 1560 uint ophdr_spc = ticket->t_res_num_ophdrs * (uint)sizeof(xlog_op_header_t);
@@ -1637,6 +1654,10 @@ xlog_print_tic_res(xfs_mount_t *mp, xlog_ticket_t *ticket)
1637 "bad-rtype" : res_type_str[r_type-1]), 1654 "bad-rtype" : res_type_str[r_type-1]),
1638 ticket->t_res_arr[i].r_len); 1655 ticket->t_res_arr[i].r_len);
1639 } 1656 }
1657
1658 xfs_cmn_err(XFS_PTAG_LOGRES, CE_ALERT, mp,
1659 "xfs_log_write: reservation ran out. Need to up reservation");
1660 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1640} 1661}
1641 1662
1642/* 1663/*
@@ -1865,7 +1886,7 @@ xlog_write_copy_finish(
1865 * we don't update ic_offset until the end when we know exactly how many 1886 * we don't update ic_offset until the end when we know exactly how many
1866 * bytes have been written out. 1887 * bytes have been written out.
1867 */ 1888 */
1868STATIC int 1889int
1869xlog_write( 1890xlog_write(
1870 struct log *log, 1891 struct log *log,
1871 struct xfs_log_vec *log_vector, 1892 struct xfs_log_vec *log_vector,
@@ -1889,22 +1910,26 @@ xlog_write(
1889 *start_lsn = 0; 1910 *start_lsn = 0;
1890 1911
1891 len = xlog_write_calc_vec_length(ticket, log_vector); 1912 len = xlog_write_calc_vec_length(ticket, log_vector);
1892 if (ticket->t_curr_res < len) { 1913 if (log->l_cilp) {
1893 xlog_print_tic_res(log->l_mp, ticket); 1914 /*
1894#ifdef DEBUG 1915 * Region headers and bytes are already accounted for.
1895 xlog_panic( 1916 * We only need to take into account start records and
1896 "xfs_log_write: reservation ran out. Need to up reservation"); 1917 * split regions in this function.
1897#else 1918 */
1898 /* Customer configurable panic */ 1919 if (ticket->t_flags & XLOG_TIC_INITED)
1899 xfs_cmn_err(XFS_PTAG_LOGRES, CE_ALERT, log->l_mp, 1920 ticket->t_curr_res -= sizeof(xlog_op_header_t);
1900 "xfs_log_write: reservation ran out. Need to up reservation");
1901 1921
1902 /* If we did not panic, shutdown the filesystem */ 1922 /*
1903 xfs_force_shutdown(log->l_mp, SHUTDOWN_CORRUPT_INCORE); 1923 * Commit record headers need to be accounted for. These
1904#endif 1924 * come in as separate writes so are easy to detect.
1905 } 1925 */
1926 if (flags & (XLOG_COMMIT_TRANS | XLOG_UNMOUNT_TRANS))
1927 ticket->t_curr_res -= sizeof(xlog_op_header_t);
1928 } else
1929 ticket->t_curr_res -= len;
1906 1930
1907 ticket->t_curr_res -= len; 1931 if (ticket->t_curr_res < 0)
1932 xlog_print_tic_res(log->l_mp, ticket);
1908 1933
1909 index = 0; 1934 index = 0;
1910 lv = log_vector; 1935 lv = log_vector;
@@ -3000,6 +3025,8 @@ _xfs_log_force(
3000 3025
3001 XFS_STATS_INC(xs_log_force); 3026 XFS_STATS_INC(xs_log_force);
3002 3027
3028 xlog_cil_push(log, 1);
3029
3003 spin_lock(&log->l_icloglock); 3030 spin_lock(&log->l_icloglock);
3004 3031
3005 iclog = log->l_iclog; 3032 iclog = log->l_iclog;
@@ -3149,6 +3176,12 @@ _xfs_log_force_lsn(
3149 3176
3150 XFS_STATS_INC(xs_log_force); 3177 XFS_STATS_INC(xs_log_force);
3151 3178
3179 if (log->l_cilp) {
3180 lsn = xlog_cil_push_lsn(log, lsn);
3181 if (lsn == NULLCOMMITLSN)
3182 return 0;
3183 }
3184
3152try_again: 3185try_again:
3153 spin_lock(&log->l_icloglock); 3186 spin_lock(&log->l_icloglock);
3154 iclog = log->l_iclog; 3187 iclog = log->l_iclog;
@@ -3313,22 +3346,30 @@ xfs_log_ticket_get(
3313 return ticket; 3346 return ticket;
3314} 3347}
3315 3348
3349xlog_tid_t
3350xfs_log_get_trans_ident(
3351 struct xfs_trans *tp)
3352{
3353 return tp->t_ticket->t_tid;
3354}
3355
3316/* 3356/*
3317 * Allocate and initialise a new log ticket. 3357 * Allocate and initialise a new log ticket.
3318 */ 3358 */
3319STATIC xlog_ticket_t * 3359xlog_ticket_t *
3320xlog_ticket_alloc( 3360xlog_ticket_alloc(
3321 struct log *log, 3361 struct log *log,
3322 int unit_bytes, 3362 int unit_bytes,
3323 int cnt, 3363 int cnt,
3324 char client, 3364 char client,
3325 uint xflags) 3365 uint xflags,
3366 int alloc_flags)
3326{ 3367{
3327 struct xlog_ticket *tic; 3368 struct xlog_ticket *tic;
3328 uint num_headers; 3369 uint num_headers;
3329 int iclog_space; 3370 int iclog_space;
3330 3371
3331 tic = kmem_zone_zalloc(xfs_log_ticket_zone, KM_SLEEP|KM_MAYFAIL); 3372 tic = kmem_zone_zalloc(xfs_log_ticket_zone, alloc_flags);
3332 if (!tic) 3373 if (!tic)
3333 return NULL; 3374 return NULL;
3334 3375
@@ -3647,6 +3688,11 @@ xlog_state_ioerror(
3647 * c. nothing new gets queued up after (a) and (b) are done. 3688 * c. nothing new gets queued up after (a) and (b) are done.
3648 * d. if !logerror, flush the iclogs to disk, then seal them off 3689 * d. if !logerror, flush the iclogs to disk, then seal them off
3649 * for business. 3690 * for business.
3691 *
3692 * Note: for delayed logging the !logerror case needs to flush the regions
3693 * held in memory out to the iclogs before flushing them to disk. This needs
3694 * to be done before the log is marked as shutdown, otherwise the flush to the
3695 * iclogs will fail.
3650 */ 3696 */
3651int 3697int
3652xfs_log_force_umount( 3698xfs_log_force_umount(
@@ -3680,6 +3726,16 @@ xfs_log_force_umount(
3680 return 1; 3726 return 1;
3681 } 3727 }
3682 retval = 0; 3728 retval = 0;
3729
3730 /*
3731 * Flush the in memory commit item list before marking the log as
3732 * being shut down. We need to do it in this order to ensure all the
3733 * completed transactions are flushed to disk with the xfs_log_force()
3734 * call below.
3735 */
3736 if (!logerror && (mp->m_flags & XFS_MOUNT_DELAYLOG))
3737 xlog_cil_push(log, 1);
3738
3683 /* 3739 /*
3684 * We must hold both the GRANT lock and the LOG lock, 3740 * We must hold both the GRANT lock and the LOG lock,
3685 * before we mark the filesystem SHUTDOWN and wake 3741 * before we mark the filesystem SHUTDOWN and wake
diff --git a/fs/xfs/xfs_log.h b/fs/xfs/xfs_log.h
index 229d1f36ba9a..04c78e642cc8 100644
--- a/fs/xfs/xfs_log.h
+++ b/fs/xfs/xfs_log.h
@@ -19,7 +19,6 @@
19#define __XFS_LOG_H__ 19#define __XFS_LOG_H__
20 20
21/* get lsn fields */ 21/* get lsn fields */
22
23#define CYCLE_LSN(lsn) ((uint)((lsn)>>32)) 22#define CYCLE_LSN(lsn) ((uint)((lsn)>>32))
24#define BLOCK_LSN(lsn) ((uint)(lsn)) 23#define BLOCK_LSN(lsn) ((uint)(lsn))
25 24
@@ -114,6 +113,9 @@ struct xfs_log_vec {
114 struct xfs_log_vec *lv_next; /* next lv in build list */ 113 struct xfs_log_vec *lv_next; /* next lv in build list */
115 int lv_niovecs; /* number of iovecs in lv */ 114 int lv_niovecs; /* number of iovecs in lv */
116 struct xfs_log_iovec *lv_iovecp; /* iovec array */ 115 struct xfs_log_iovec *lv_iovecp; /* iovec array */
116 struct xfs_log_item *lv_item; /* owner */
117 char *lv_buf; /* formatted buffer */
118 int lv_buf_len; /* size of formatted buffer */
117}; 119};
118 120
119/* 121/*
@@ -134,6 +136,7 @@ struct xlog_in_core;
134struct xlog_ticket; 136struct xlog_ticket;
135struct xfs_log_item; 137struct xfs_log_item;
136struct xfs_item_ops; 138struct xfs_item_ops;
139struct xfs_trans;
137 140
138void xfs_log_item_init(struct xfs_mount *mp, 141void xfs_log_item_init(struct xfs_mount *mp,
139 struct xfs_log_item *item, 142 struct xfs_log_item *item,
@@ -187,9 +190,16 @@ int xfs_log_need_covered(struct xfs_mount *mp);
187 190
188void xlog_iodone(struct xfs_buf *); 191void xlog_iodone(struct xfs_buf *);
189 192
190struct xlog_ticket * xfs_log_ticket_get(struct xlog_ticket *ticket); 193struct xlog_ticket *xfs_log_ticket_get(struct xlog_ticket *ticket);
191void xfs_log_ticket_put(struct xlog_ticket *ticket); 194void xfs_log_ticket_put(struct xlog_ticket *ticket);
192 195
196xlog_tid_t xfs_log_get_trans_ident(struct xfs_trans *tp);
197
198int xfs_log_commit_cil(struct xfs_mount *mp, struct xfs_trans *tp,
199 struct xfs_log_vec *log_vector,
200 xfs_lsn_t *commit_lsn, int flags);
201bool xfs_log_item_in_current_chkpt(struct xfs_log_item *lip);
202
193#endif 203#endif
194 204
195 205
diff --git a/fs/xfs/xfs_log_cil.c b/fs/xfs/xfs_log_cil.c
new file mode 100644
index 000000000000..bb17cc044bf3
--- /dev/null
+++ b/fs/xfs/xfs_log_cil.c
@@ -0,0 +1,725 @@
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"
21#include "xfs_bit.h"
22#include "xfs_log.h"
23#include "xfs_inum.h"
24#include "xfs_trans.h"
25#include "xfs_trans_priv.h"
26#include "xfs_log_priv.h"
27#include "xfs_sb.h"
28#include "xfs_ag.h"
29#include "xfs_dir2.h"
30#include "xfs_dmapi.h"
31#include "xfs_mount.h"
32#include "xfs_error.h"
33#include "xfs_alloc.h"
34
35/*
36 * Perform initial CIL structure initialisation. If the CIL is not
37 * enabled in this filesystem, ensure the log->l_cilp is null so
38 * we can check this conditional to determine if we are doing delayed
39 * logging or not.
40 */
41int
42xlog_cil_init(
43 struct log *log)
44{
45 struct xfs_cil *cil;
46 struct xfs_cil_ctx *ctx;
47
48 log->l_cilp = NULL;
49 if (!(log->l_mp->m_flags & XFS_MOUNT_DELAYLOG))
50 return 0;
51
52 cil = kmem_zalloc(sizeof(*cil), KM_SLEEP|KM_MAYFAIL);
53 if (!cil)
54 return ENOMEM;
55
56 ctx = kmem_zalloc(sizeof(*ctx), KM_SLEEP|KM_MAYFAIL);
57 if (!ctx) {
58 kmem_free(cil);
59 return ENOMEM;
60 }
61
62 INIT_LIST_HEAD(&cil->xc_cil);
63 INIT_LIST_HEAD(&cil->xc_committing);
64 spin_lock_init(&cil->xc_cil_lock);
65 init_rwsem(&cil->xc_ctx_lock);
66 sv_init(&cil->xc_commit_wait, SV_DEFAULT, "cilwait");
67
68 INIT_LIST_HEAD(&ctx->committing);
69 INIT_LIST_HEAD(&ctx->busy_extents);
70 ctx->sequence = 1;
71 ctx->cil = cil;
72 cil->xc_ctx = ctx;
73
74 cil->xc_log = log;
75 log->l_cilp = cil;
76 return 0;
77}
78
79void
80xlog_cil_destroy(
81 struct log *log)
82{
83 if (!log->l_cilp)
84 return;
85
86 if (log->l_cilp->xc_ctx) {
87 if (log->l_cilp->xc_ctx->ticket)
88 xfs_log_ticket_put(log->l_cilp->xc_ctx->ticket);
89 kmem_free(log->l_cilp->xc_ctx);
90 }
91
92 ASSERT(list_empty(&log->l_cilp->xc_cil));
93 kmem_free(log->l_cilp);
94}
95
96/*
97 * Allocate a new ticket. Failing to get a new ticket makes it really hard to
98 * recover, so we don't allow failure here. Also, we allocate in a context that
99 * we don't want to be issuing transactions from, so we need to tell the
100 * allocation code this as well.
101 *
102 * We don't reserve any space for the ticket - we are going to steal whatever
103 * space we require from transactions as they commit. To ensure we reserve all
104 * the space required, we need to set the current reservation of the ticket to
105 * zero so that we know to steal the initial transaction overhead from the
106 * first transaction commit.
107 */
108static struct xlog_ticket *
109xlog_cil_ticket_alloc(
110 struct log *log)
111{
112 struct xlog_ticket *tic;
113
114 tic = xlog_ticket_alloc(log, 0, 1, XFS_TRANSACTION, 0,
115 KM_SLEEP|KM_NOFS);
116 tic->t_trans_type = XFS_TRANS_CHECKPOINT;
117
118 /*
119 * set the current reservation to zero so we know to steal the basic
120 * transaction overhead reservation from the first transaction commit.
121 */
122 tic->t_curr_res = 0;
123 return tic;
124}
125
126/*
127 * After the first stage of log recovery is done, we know where the head and
128 * tail of the log are. We need this log initialisation done before we can
129 * initialise the first CIL checkpoint context.
130 *
131 * Here we allocate a log ticket to track space usage during a CIL push. This
132 * ticket is passed to xlog_write() directly so that we don't slowly leak log
133 * space by failing to account for space used by log headers and additional
134 * region headers for split regions.
135 */
136void
137xlog_cil_init_post_recovery(
138 struct log *log)
139{
140 if (!log->l_cilp)
141 return;
142
143 log->l_cilp->xc_ctx->ticket = xlog_cil_ticket_alloc(log);
144 log->l_cilp->xc_ctx->sequence = 1;
145 log->l_cilp->xc_ctx->commit_lsn = xlog_assign_lsn(log->l_curr_cycle,
146 log->l_curr_block);
147}
148
149/*
150 * Insert the log item into the CIL and calculate the difference in space
151 * consumed by the item. Add the space to the checkpoint ticket and calculate
152 * if the change requires additional log metadata. If it does, take that space
153 * as well. Remove the amount of space we addded to the checkpoint ticket from
154 * the current transaction ticket so that the accounting works out correctly.
155 *
156 * If this is the first time the item is being placed into the CIL in this
157 * context, pin it so it can't be written to disk until the CIL is flushed to
158 * the iclog and the iclog written to disk.
159 */
160static void
161xlog_cil_insert(
162 struct log *log,
163 struct xlog_ticket *ticket,
164 struct xfs_log_item *item,
165 struct xfs_log_vec *lv)
166{
167 struct xfs_cil *cil = log->l_cilp;
168 struct xfs_log_vec *old = lv->lv_item->li_lv;
169 struct xfs_cil_ctx *ctx = cil->xc_ctx;
170 int len;
171 int diff_iovecs;
172 int iclog_space;
173
174 if (old) {
175 /* existing lv on log item, space used is a delta */
176 ASSERT(!list_empty(&item->li_cil));
177 ASSERT(old->lv_buf && old->lv_buf_len && old->lv_niovecs);
178
179 len = lv->lv_buf_len - old->lv_buf_len;
180 diff_iovecs = lv->lv_niovecs - old->lv_niovecs;
181 kmem_free(old->lv_buf);
182 kmem_free(old);
183 } else {
184 /* new lv, must pin the log item */
185 ASSERT(!lv->lv_item->li_lv);
186 ASSERT(list_empty(&item->li_cil));
187
188 len = lv->lv_buf_len;
189 diff_iovecs = lv->lv_niovecs;
190 IOP_PIN(lv->lv_item);
191
192 }
193 len += diff_iovecs * sizeof(xlog_op_header_t);
194
195 /* attach new log vector to log item */
196 lv->lv_item->li_lv = lv;
197
198 spin_lock(&cil->xc_cil_lock);
199 list_move_tail(&item->li_cil, &cil->xc_cil);
200 ctx->nvecs += diff_iovecs;
201
202 /*
203 * If this is the first time the item is being committed to the CIL,
204 * store the sequence number on the log item so we can tell
205 * in future commits whether this is the first checkpoint the item is
206 * being committed into.
207 */
208 if (!item->li_seq)
209 item->li_seq = ctx->sequence;
210
211 /*
212 * Now transfer enough transaction reservation to the context ticket
213 * for the checkpoint. The context ticket is special - the unit
214 * reservation has to grow as well as the current reservation as we
215 * steal from tickets so we can correctly determine the space used
216 * during the transaction commit.
217 */
218 if (ctx->ticket->t_curr_res == 0) {
219 /* first commit in checkpoint, steal the header reservation */
220 ASSERT(ticket->t_curr_res >= ctx->ticket->t_unit_res + len);
221 ctx->ticket->t_curr_res = ctx->ticket->t_unit_res;
222 ticket->t_curr_res -= ctx->ticket->t_unit_res;
223 }
224
225 /* do we need space for more log record headers? */
226 iclog_space = log->l_iclog_size - log->l_iclog_hsize;
227 if (len > 0 && (ctx->space_used / iclog_space !=
228 (ctx->space_used + len) / iclog_space)) {
229 int hdrs;
230
231 hdrs = (len + iclog_space - 1) / iclog_space;
232 /* need to take into account split region headers, too */
233 hdrs *= log->l_iclog_hsize + sizeof(struct xlog_op_header);
234 ctx->ticket->t_unit_res += hdrs;
235 ctx->ticket->t_curr_res += hdrs;
236 ticket->t_curr_res -= hdrs;
237 ASSERT(ticket->t_curr_res >= len);
238 }
239 ticket->t_curr_res -= len;
240 ctx->space_used += len;
241
242 spin_unlock(&cil->xc_cil_lock);
243}
244
245/*
246 * Format log item into a flat buffers
247 *
248 * For delayed logging, we need to hold a formatted buffer containing all the
249 * changes on the log item. This enables us to relog the item in memory and
250 * write it out asynchronously without needing to relock the object that was
251 * modified at the time it gets written into the iclog.
252 *
253 * This function builds a vector for the changes in each log item in the
254 * transaction. It then works out the length of the buffer needed for each log
255 * item, allocates them and formats the vector for the item into the buffer.
256 * The buffer is then attached to the log item are then inserted into the
257 * Committed Item List for tracking until the next checkpoint is written out.
258 *
259 * We don't set up region headers during this process; we simply copy the
260 * regions into the flat buffer. We can do this because we still have to do a
261 * formatting step to write the regions into the iclog buffer. Writing the
262 * ophdrs during the iclog write means that we can support splitting large
263 * regions across iclog boundares without needing a change in the format of the
264 * item/region encapsulation.
265 *
266 * Hence what we need to do now is change the rewrite the vector array to point
267 * to the copied region inside the buffer we just allocated. This allows us to
268 * format the regions into the iclog as though they are being formatted
269 * directly out of the objects themselves.
270 */
271static void
272xlog_cil_format_items(
273 struct log *log,
274 struct xfs_log_vec *log_vector,
275 struct xlog_ticket *ticket,
276 xfs_lsn_t *start_lsn)
277{
278 struct xfs_log_vec *lv;
279
280 if (start_lsn)
281 *start_lsn = log->l_cilp->xc_ctx->sequence;
282
283 ASSERT(log_vector);
284 for (lv = log_vector; lv; lv = lv->lv_next) {
285 void *ptr;
286 int index;
287 int len = 0;
288
289 /* build the vector array and calculate it's length */
290 IOP_FORMAT(lv->lv_item, lv->lv_iovecp);
291 for (index = 0; index < lv->lv_niovecs; index++)
292 len += lv->lv_iovecp[index].i_len;
293
294 lv->lv_buf_len = len;
295 lv->lv_buf = kmem_zalloc(lv->lv_buf_len, KM_SLEEP|KM_NOFS);
296 ptr = lv->lv_buf;
297
298 for (index = 0; index < lv->lv_niovecs; index++) {
299 struct xfs_log_iovec *vec = &lv->lv_iovecp[index];
300
301 memcpy(ptr, vec->i_addr, vec->i_len);
302 vec->i_addr = ptr;
303 ptr += vec->i_len;
304 }
305 ASSERT(ptr == lv->lv_buf + lv->lv_buf_len);
306
307 xlog_cil_insert(log, ticket, lv->lv_item, lv);
308 }
309}
310
311static void
312xlog_cil_free_logvec(
313 struct xfs_log_vec *log_vector)
314{
315 struct xfs_log_vec *lv;
316
317 for (lv = log_vector; lv; ) {
318 struct xfs_log_vec *next = lv->lv_next;
319 kmem_free(lv->lv_buf);
320 kmem_free(lv);
321 lv = next;
322 }
323}
324
325/*
326 * Commit a transaction with the given vector to the Committed Item List.
327 *
328 * To do this, we need to format the item, pin it in memory if required and
329 * account for the space used by the transaction. Once we have done that we
330 * need to release the unused reservation for the transaction, attach the
331 * transaction to the checkpoint context so we carry the busy extents through
332 * to checkpoint completion, and then unlock all the items in the transaction.
333 *
334 * For more specific information about the order of operations in
335 * xfs_log_commit_cil() please refer to the comments in
336 * xfs_trans_commit_iclog().
337 *
338 * Called with the context lock already held in read mode to lock out
339 * background commit, returns without it held once background commits are
340 * allowed again.
341 */
342int
343xfs_log_commit_cil(
344 struct xfs_mount *mp,
345 struct xfs_trans *tp,
346 struct xfs_log_vec *log_vector,
347 xfs_lsn_t *commit_lsn,
348 int flags)
349{
350 struct log *log = mp->m_log;
351 int log_flags = 0;
352 int push = 0;
353
354 if (flags & XFS_TRANS_RELEASE_LOG_RES)
355 log_flags = XFS_LOG_REL_PERM_RESERV;
356
357 if (XLOG_FORCED_SHUTDOWN(log)) {
358 xlog_cil_free_logvec(log_vector);
359 return XFS_ERROR(EIO);
360 }
361
362 /* lock out background commit */
363 down_read(&log->l_cilp->xc_ctx_lock);
364 xlog_cil_format_items(log, log_vector, tp->t_ticket, commit_lsn);
365
366 /* check we didn't blow the reservation */
367 if (tp->t_ticket->t_curr_res < 0)
368 xlog_print_tic_res(log->l_mp, tp->t_ticket);
369
370 /* attach the transaction to the CIL if it has any busy extents */
371 if (!list_empty(&tp->t_busy)) {
372 spin_lock(&log->l_cilp->xc_cil_lock);
373 list_splice_init(&tp->t_busy,
374 &log->l_cilp->xc_ctx->busy_extents);
375 spin_unlock(&log->l_cilp->xc_cil_lock);
376 }
377
378 tp->t_commit_lsn = *commit_lsn;
379 xfs_log_done(mp, tp->t_ticket, NULL, log_flags);
380 xfs_trans_unreserve_and_mod_sb(tp);
381
382 /* check for background commit before unlock */
383 if (log->l_cilp->xc_ctx->space_used > XLOG_CIL_SPACE_LIMIT(log))
384 push = 1;
385 up_read(&log->l_cilp->xc_ctx_lock);
386
387 /*
388 * We need to push CIL every so often so we don't cache more than we
389 * can fit in the log. The limit really is that a checkpoint can't be
390 * more than half the log (the current checkpoint is not allowed to
391 * overwrite the previous checkpoint), but commit latency and memory
392 * usage limit this to a smaller size in most cases.
393 */
394 if (push)
395 xlog_cil_push(log, 0);
396 return 0;
397}
398
399/*
400 * Mark all items committed and clear busy extents. We free the log vector
401 * chains in a separate pass so that we unpin the log items as quickly as
402 * possible.
403 */
404static void
405xlog_cil_committed(
406 void *args,
407 int abort)
408{
409 struct xfs_cil_ctx *ctx = args;
410 struct xfs_log_vec *lv;
411 int abortflag = abort ? XFS_LI_ABORTED : 0;
412 struct xfs_busy_extent *busyp, *n;
413
414 /* unpin all the log items */
415 for (lv = ctx->lv_chain; lv; lv = lv->lv_next ) {
416 xfs_trans_item_committed(lv->lv_item, ctx->start_lsn,
417 abortflag);
418 }
419
420 list_for_each_entry_safe(busyp, n, &ctx->busy_extents, list)
421 xfs_alloc_busy_clear(ctx->cil->xc_log->l_mp, busyp);
422
423 spin_lock(&ctx->cil->xc_cil_lock);
424 list_del(&ctx->committing);
425 spin_unlock(&ctx->cil->xc_cil_lock);
426
427 xlog_cil_free_logvec(ctx->lv_chain);
428 kmem_free(ctx);
429}
430
431/*
432 * Push the Committed Item List to the log. If the push_now flag is not set,
433 * then it is a background flush and so we can chose to ignore it.
434 */
435int
436xlog_cil_push(
437 struct log *log,
438 int push_now)
439{
440 struct xfs_cil *cil = log->l_cilp;
441 struct xfs_log_vec *lv;
442 struct xfs_cil_ctx *ctx;
443 struct xfs_cil_ctx *new_ctx;
444 struct xlog_in_core *commit_iclog;
445 struct xlog_ticket *tic;
446 int num_lv;
447 int num_iovecs;
448 int len;
449 int error = 0;
450 struct xfs_trans_header thdr;
451 struct xfs_log_iovec lhdr;
452 struct xfs_log_vec lvhdr = { NULL };
453 xfs_lsn_t commit_lsn;
454
455 if (!cil)
456 return 0;
457
458 new_ctx = kmem_zalloc(sizeof(*new_ctx), KM_SLEEP|KM_NOFS);
459 new_ctx->ticket = xlog_cil_ticket_alloc(log);
460
461 /* lock out transaction commit, but don't block on background push */
462 if (!down_write_trylock(&cil->xc_ctx_lock)) {
463 if (!push_now)
464 goto out_free_ticket;
465 down_write(&cil->xc_ctx_lock);
466 }
467 ctx = cil->xc_ctx;
468
469 /* check if we've anything to push */
470 if (list_empty(&cil->xc_cil))
471 goto out_skip;
472
473 /* check for spurious background flush */
474 if (!push_now && cil->xc_ctx->space_used < XLOG_CIL_SPACE_LIMIT(log))
475 goto out_skip;
476
477 /*
478 * pull all the log vectors off the items in the CIL, and
479 * remove the items from the CIL. We don't need the CIL lock
480 * here because it's only needed on the transaction commit
481 * side which is currently locked out by the flush lock.
482 */
483 lv = NULL;
484 num_lv = 0;
485 num_iovecs = 0;
486 len = 0;
487 while (!list_empty(&cil->xc_cil)) {
488 struct xfs_log_item *item;
489 int i;
490
491 item = list_first_entry(&cil->xc_cil,
492 struct xfs_log_item, li_cil);
493 list_del_init(&item->li_cil);
494 if (!ctx->lv_chain)
495 ctx->lv_chain = item->li_lv;
496 else
497 lv->lv_next = item->li_lv;
498 lv = item->li_lv;
499 item->li_lv = NULL;
500
501 num_lv++;
502 num_iovecs += lv->lv_niovecs;
503 for (i = 0; i < lv->lv_niovecs; i++)
504 len += lv->lv_iovecp[i].i_len;
505 }
506
507 /*
508 * initialise the new context and attach it to the CIL. Then attach
509 * the current context to the CIL committing lsit so it can be found
510 * during log forces to extract the commit lsn of the sequence that
511 * needs to be forced.
512 */
513 INIT_LIST_HEAD(&new_ctx->committing);
514 INIT_LIST_HEAD(&new_ctx->busy_extents);
515 new_ctx->sequence = ctx->sequence + 1;
516 new_ctx->cil = cil;
517 cil->xc_ctx = new_ctx;
518
519 /*
520 * The switch is now done, so we can drop the context lock and move out
521 * of a shared context. We can't just go straight to the commit record,
522 * though - we need to synchronise with previous and future commits so
523 * that the commit records are correctly ordered in the log to ensure
524 * that we process items during log IO completion in the correct order.
525 *
526 * For example, if we get an EFI in one checkpoint and the EFD in the
527 * next (e.g. due to log forces), we do not want the checkpoint with
528 * the EFD to be committed before the checkpoint with the EFI. Hence
529 * we must strictly order the commit records of the checkpoints so
530 * that: a) the checkpoint callbacks are attached to the iclogs in the
531 * correct order; and b) the checkpoints are replayed in correct order
532 * in log recovery.
533 *
534 * Hence we need to add this context to the committing context list so
535 * that higher sequences will wait for us to write out a commit record
536 * before they do.
537 */
538 spin_lock(&cil->xc_cil_lock);
539 list_add(&ctx->committing, &cil->xc_committing);
540 spin_unlock(&cil->xc_cil_lock);
541 up_write(&cil->xc_ctx_lock);
542
543 /*
544 * Build a checkpoint transaction header and write it to the log to
545 * begin the transaction. We need to account for the space used by the
546 * transaction header here as it is not accounted for in xlog_write().
547 *
548 * The LSN we need to pass to the log items on transaction commit is
549 * the LSN reported by the first log vector write. If we use the commit
550 * record lsn then we can move the tail beyond the grant write head.
551 */
552 tic = ctx->ticket;
553 thdr.th_magic = XFS_TRANS_HEADER_MAGIC;
554 thdr.th_type = XFS_TRANS_CHECKPOINT;
555 thdr.th_tid = tic->t_tid;
556 thdr.th_num_items = num_iovecs;
557 lhdr.i_addr = (xfs_caddr_t)&thdr;
558 lhdr.i_len = sizeof(xfs_trans_header_t);
559 lhdr.i_type = XLOG_REG_TYPE_TRANSHDR;
560 tic->t_curr_res -= lhdr.i_len + sizeof(xlog_op_header_t);
561
562 lvhdr.lv_niovecs = 1;
563 lvhdr.lv_iovecp = &lhdr;
564 lvhdr.lv_next = ctx->lv_chain;
565
566 error = xlog_write(log, &lvhdr, tic, &ctx->start_lsn, NULL, 0);
567 if (error)
568 goto out_abort;
569
570 /*
571 * now that we've written the checkpoint into the log, strictly
572 * order the commit records so replay will get them in the right order.
573 */
574restart:
575 spin_lock(&cil->xc_cil_lock);
576 list_for_each_entry(new_ctx, &cil->xc_committing, committing) {
577 /*
578 * Higher sequences will wait for this one so skip them.
579 * Don't wait for own own sequence, either.
580 */
581 if (new_ctx->sequence >= ctx->sequence)
582 continue;
583 if (!new_ctx->commit_lsn) {
584 /*
585 * It is still being pushed! Wait for the push to
586 * complete, then start again from the beginning.
587 */
588 sv_wait(&cil->xc_commit_wait, 0, &cil->xc_cil_lock, 0);
589 goto restart;
590 }
591 }
592 spin_unlock(&cil->xc_cil_lock);
593
594 commit_lsn = xfs_log_done(log->l_mp, tic, &commit_iclog, 0);
595 if (error || commit_lsn == -1)
596 goto out_abort;
597
598 /* attach all the transactions w/ busy extents to iclog */
599 ctx->log_cb.cb_func = xlog_cil_committed;
600 ctx->log_cb.cb_arg = ctx;
601 error = xfs_log_notify(log->l_mp, commit_iclog, &ctx->log_cb);
602 if (error)
603 goto out_abort;
604
605 /*
606 * now the checkpoint commit is complete and we've attached the
607 * callbacks to the iclog we can assign the commit LSN to the context
608 * and wake up anyone who is waiting for the commit to complete.
609 */
610 spin_lock(&cil->xc_cil_lock);
611 ctx->commit_lsn = commit_lsn;
612 sv_broadcast(&cil->xc_commit_wait);
613 spin_unlock(&cil->xc_cil_lock);
614
615 /* release the hounds! */
616 return xfs_log_release_iclog(log->l_mp, commit_iclog);
617
618out_skip:
619 up_write(&cil->xc_ctx_lock);
620out_free_ticket:
621 xfs_log_ticket_put(new_ctx->ticket);
622 kmem_free(new_ctx);
623 return 0;
624
625out_abort:
626 xlog_cil_committed(ctx, XFS_LI_ABORTED);
627 return XFS_ERROR(EIO);
628}
629
630/*
631 * Conditionally push the CIL based on the sequence passed in.
632 *
633 * We only need to push if we haven't already pushed the sequence
634 * number given. Hence the only time we will trigger a push here is
635 * if the push sequence is the same as the current context.
636 *
637 * We return the current commit lsn to allow the callers to determine if a
638 * iclog flush is necessary following this call.
639 *
640 * XXX: Initially, just push the CIL unconditionally and return whatever
641 * commit lsn is there. It'll be empty, so this is broken for now.
642 */
643xfs_lsn_t
644xlog_cil_push_lsn(
645 struct log *log,
646 xfs_lsn_t push_seq)
647{
648 struct xfs_cil *cil = log->l_cilp;
649 struct xfs_cil_ctx *ctx;
650 xfs_lsn_t commit_lsn = NULLCOMMITLSN;
651
652restart:
653 down_write(&cil->xc_ctx_lock);
654 ASSERT(push_seq <= cil->xc_ctx->sequence);
655
656 /* check to see if we need to force out the current context */
657 if (push_seq == cil->xc_ctx->sequence) {
658 up_write(&cil->xc_ctx_lock);
659 xlog_cil_push(log, 1);
660 goto restart;
661 }
662
663 /*
664 * See if we can find a previous sequence still committing.
665 * We can drop the flush lock as soon as we have the cil lock
666 * because we are now only comparing contexts protected by
667 * the cil lock.
668 *
669 * We need to wait for all previous sequence commits to complete
670 * before allowing the force of push_seq to go ahead. Hence block
671 * on commits for those as well.
672 */
673 spin_lock(&cil->xc_cil_lock);
674 up_write(&cil->xc_ctx_lock);
675 list_for_each_entry(ctx, &cil->xc_committing, committing) {
676 if (ctx->sequence > push_seq)
677 continue;
678 if (!ctx->commit_lsn) {
679 /*
680 * It is still being pushed! Wait for the push to
681 * complete, then start again from the beginning.
682 */
683 sv_wait(&cil->xc_commit_wait, 0, &cil->xc_cil_lock, 0);
684 goto restart;
685 }
686 if (ctx->sequence != push_seq)
687 continue;
688 /* found it! */
689 commit_lsn = ctx->commit_lsn;
690 }
691 spin_unlock(&cil->xc_cil_lock);
692 return commit_lsn;
693}
694
695/*
696 * Check if the current log item was first committed in this sequence.
697 * We can't rely on just the log item being in the CIL, we have to check
698 * the recorded commit sequence number.
699 *
700 * Note: for this to be used in a non-racy manner, it has to be called with
701 * CIL flushing locked out. As a result, it should only be used during the
702 * transaction commit process when deciding what to format into the item.
703 */
704bool
705xfs_log_item_in_current_chkpt(
706 struct xfs_log_item *lip)
707{
708 struct xfs_cil_ctx *ctx;
709
710 if (!(lip->li_mountp->m_flags & XFS_MOUNT_DELAYLOG))
711 return false;
712 if (list_empty(&lip->li_cil))
713 return false;
714
715 ctx = lip->li_mountp->m_log->l_cilp->xc_ctx;
716
717 /*
718 * li_seq is written on the first commit of a log item to record the
719 * first checkpoint it is written to. Hence if it is different to the
720 * current sequence, we're in a new checkpoint.
721 */
722 if (XFS_LSN_CMP(lip->li_seq, ctx->sequence) != 0)
723 return false;
724 return true;
725}
diff --git a/fs/xfs/xfs_log_priv.h b/fs/xfs/xfs_log_priv.h
index 9cf695154451..8c072618965c 100644
--- a/fs/xfs/xfs_log_priv.h
+++ b/fs/xfs/xfs_log_priv.h
@@ -152,8 +152,6 @@ static inline uint xlog_get_client_id(__be32 i)
152#define XLOG_RECOVERY_NEEDED 0x4 /* log was recovered */ 152#define XLOG_RECOVERY_NEEDED 0x4 /* log was recovered */
153#define XLOG_IO_ERROR 0x8 /* log hit an I/O error, and being 153#define XLOG_IO_ERROR 0x8 /* log hit an I/O error, and being
154 shutdown */ 154 shutdown */
155typedef __uint32_t xlog_tid_t;
156
157 155
158#ifdef __KERNEL__ 156#ifdef __KERNEL__
159/* 157/*
@@ -379,6 +377,99 @@ typedef struct xlog_in_core {
379} xlog_in_core_t; 377} xlog_in_core_t;
380 378
381/* 379/*
380 * The CIL context is used to aggregate per-transaction details as well be
381 * passed to the iclog for checkpoint post-commit processing. After being
382 * passed to the iclog, another context needs to be allocated for tracking the
383 * next set of transactions to be aggregated into a checkpoint.
384 */
385struct xfs_cil;
386
387struct xfs_cil_ctx {
388 struct xfs_cil *cil;
389 xfs_lsn_t sequence; /* chkpt sequence # */
390 xfs_lsn_t start_lsn; /* first LSN of chkpt commit */
391 xfs_lsn_t commit_lsn; /* chkpt commit record lsn */
392 struct xlog_ticket *ticket; /* chkpt ticket */
393 int nvecs; /* number of regions */
394 int space_used; /* aggregate size of regions */
395 struct list_head busy_extents; /* busy extents in chkpt */
396 struct xfs_log_vec *lv_chain; /* logvecs being pushed */
397 xfs_log_callback_t log_cb; /* completion callback hook. */
398 struct list_head committing; /* ctx committing list */
399};
400
401/*
402 * Committed Item List structure
403 *
404 * This structure is used to track log items that have been committed but not
405 * yet written into the log. It is used only when the delayed logging mount
406 * option is enabled.
407 *
408 * This structure tracks the list of committing checkpoint contexts so
409 * we can avoid the problem of having to hold out new transactions during a
410 * flush until we have a the commit record LSN of the checkpoint. We can
411 * traverse the list of committing contexts in xlog_cil_push_lsn() to find a
412 * sequence match and extract the commit LSN directly from there. If the
413 * checkpoint is still in the process of committing, we can block waiting for
414 * the commit LSN to be determined as well. This should make synchronous
415 * operations almost as efficient as the old logging methods.
416 */
417struct xfs_cil {
418 struct log *xc_log;
419 struct list_head xc_cil;
420 spinlock_t xc_cil_lock;
421 struct xfs_cil_ctx *xc_ctx;
422 struct rw_semaphore xc_ctx_lock;
423 struct list_head xc_committing;
424 sv_t xc_commit_wait;
425};
426
427/*
428 * The amount of log space we should the CIL to aggregate is difficult to size.
429 * Whatever we chose we have to make we can get a reservation for the log space
430 * effectively, that it is large enough to capture sufficient relogging to
431 * reduce log buffer IO significantly, but it is not too large for the log or
432 * induces too much latency when writing out through the iclogs. We track both
433 * space consumed and the number of vectors in the checkpoint context, so we
434 * need to decide which to use for limiting.
435 *
436 * Every log buffer we write out during a push needs a header reserved, which
437 * is at least one sector and more for v2 logs. Hence we need a reservation of
438 * at least 512 bytes per 32k of log space just for the LR headers. That means
439 * 16KB of reservation per megabyte of delayed logging space we will consume,
440 * plus various headers. The number of headers will vary based on the num of
441 * io vectors, so limiting on a specific number of vectors is going to result
442 * in transactions of varying size. IOWs, it is more consistent to track and
443 * limit space consumed in the log rather than by the number of objects being
444 * logged in order to prevent checkpoint ticket overruns.
445 *
446 * Further, use of static reservations through the log grant mechanism is
447 * problematic. It introduces a lot of complexity (e.g. reserve grant vs write
448 * grant) and a significant deadlock potential because regranting write space
449 * can block on log pushes. Hence if we have to regrant log space during a log
450 * push, we can deadlock.
451 *
452 * However, we can avoid this by use of a dynamic "reservation stealing"
453 * technique during transaction commit whereby unused reservation space in the
454 * transaction ticket is transferred to the CIL ctx commit ticket to cover the
455 * space needed by the checkpoint transaction. This means that we never need to
456 * specifically reserve space for the CIL checkpoint transaction, nor do we
457 * need to regrant space once the checkpoint completes. This also means the
458 * checkpoint transaction ticket is specific to the checkpoint context, rather
459 * than the CIL itself.
460 *
461 * With dynamic reservations, we can basically make up arbitrary limits for the
462 * checkpoint size so long as they don't violate any other size rules. Hence
463 * the initial maximum size for the checkpoint transaction will be set to a
464 * quarter of the log or 8MB, which ever is smaller. 8MB is an arbitrary limit
465 * right now based on the latency of writing out a large amount of data through
466 * the circular iclog buffers.
467 */
468
469#define XLOG_CIL_SPACE_LIMIT(log) \
470 (min((log->l_logsize >> 2), (8 * 1024 * 1024)))
471
472/*
382 * The reservation head lsn is not made up of a cycle number and block number. 473 * The reservation head lsn is not made up of a cycle number and block number.
383 * Instead, it uses a cycle number and byte number. Logs don't expect to 474 * Instead, it uses a cycle number and byte number. Logs don't expect to
384 * overflow 31 bits worth of byte offset, so using a byte number will mean 475 * overflow 31 bits worth of byte offset, so using a byte number will mean
@@ -388,6 +479,7 @@ typedef struct log {
388 /* The following fields don't need locking */ 479 /* The following fields don't need locking */
389 struct xfs_mount *l_mp; /* mount point */ 480 struct xfs_mount *l_mp; /* mount point */
390 struct xfs_ail *l_ailp; /* AIL log is working with */ 481 struct xfs_ail *l_ailp; /* AIL log is working with */
482 struct xfs_cil *l_cilp; /* CIL log is working with */
391 struct xfs_buf *l_xbuf; /* extra buffer for log 483 struct xfs_buf *l_xbuf; /* extra buffer for log
392 * wrapping */ 484 * wrapping */
393 struct xfs_buftarg *l_targ; /* buftarg of log */ 485 struct xfs_buftarg *l_targ; /* buftarg of log */
@@ -438,14 +530,17 @@ typedef struct log {
438 530
439#define XLOG_FORCED_SHUTDOWN(log) ((log)->l_flags & XLOG_IO_ERROR) 531#define XLOG_FORCED_SHUTDOWN(log) ((log)->l_flags & XLOG_IO_ERROR)
440 532
441
442/* common routines */ 533/* common routines */
443extern xfs_lsn_t xlog_assign_tail_lsn(struct xfs_mount *mp); 534extern xfs_lsn_t xlog_assign_tail_lsn(struct xfs_mount *mp);
444extern int xlog_recover(xlog_t *log); 535extern int xlog_recover(xlog_t *log);
445extern int xlog_recover_finish(xlog_t *log); 536extern int xlog_recover_finish(xlog_t *log);
446extern void xlog_pack_data(xlog_t *log, xlog_in_core_t *iclog, int); 537extern void xlog_pack_data(xlog_t *log, xlog_in_core_t *iclog, int);
447 538
448extern kmem_zone_t *xfs_log_ticket_zone; 539extern kmem_zone_t *xfs_log_ticket_zone;
540struct xlog_ticket *xlog_ticket_alloc(struct log *log, int unit_bytes,
541 int count, char client, uint xflags,
542 int alloc_flags);
543
449 544
450static inline void 545static inline void
451xlog_write_adv_cnt(void **ptr, int *len, int *off, size_t bytes) 546xlog_write_adv_cnt(void **ptr, int *len, int *off, size_t bytes)
@@ -455,6 +550,21 @@ xlog_write_adv_cnt(void **ptr, int *len, int *off, size_t bytes)
455 *off += bytes; 550 *off += bytes;
456} 551}
457 552
553void xlog_print_tic_res(struct xfs_mount *mp, struct xlog_ticket *ticket);
554int xlog_write(struct log *log, struct xfs_log_vec *log_vector,
555 struct xlog_ticket *tic, xfs_lsn_t *start_lsn,
556 xlog_in_core_t **commit_iclog, uint flags);
557
558/*
559 * Committed Item List interfaces
560 */
561int xlog_cil_init(struct log *log);
562void xlog_cil_init_post_recovery(struct log *log);
563void xlog_cil_destroy(struct log *log);
564
565int xlog_cil_push(struct log *log, int push_now);
566xfs_lsn_t xlog_cil_push_lsn(struct log *log, xfs_lsn_t push_sequence);
567
458/* 568/*
459 * Unmount record type is used as a pseudo transaction type for the ticket. 569 * Unmount record type is used as a pseudo transaction type for the ticket.
460 * It's value must be outside the range of XFS_TRANS_* values. 570 * It's value must be outside the range of XFS_TRANS_* values.
diff --git a/fs/xfs/xfs_log_recover.c b/fs/xfs/xfs_log_recover.c
index 0de08e366315..14a69aec2c0b 100644
--- a/fs/xfs/xfs_log_recover.c
+++ b/fs/xfs/xfs_log_recover.c
@@ -1576,7 +1576,7 @@ xlog_recover_reorder_trans(
1576 1576
1577 switch (ITEM_TYPE(item)) { 1577 switch (ITEM_TYPE(item)) {
1578 case XFS_LI_BUF: 1578 case XFS_LI_BUF:
1579 if (!(buf_f->blf_flags & XFS_BLI_CANCEL)) { 1579 if (!(buf_f->blf_flags & XFS_BLF_CANCEL)) {
1580 trace_xfs_log_recover_item_reorder_head(log, 1580 trace_xfs_log_recover_item_reorder_head(log,
1581 trans, item, pass); 1581 trans, item, pass);
1582 list_move(&item->ri_list, &trans->r_itemq); 1582 list_move(&item->ri_list, &trans->r_itemq);
@@ -1638,7 +1638,7 @@ xlog_recover_do_buffer_pass1(
1638 /* 1638 /*
1639 * If this isn't a cancel buffer item, then just return. 1639 * If this isn't a cancel buffer item, then just return.
1640 */ 1640 */
1641 if (!(flags & XFS_BLI_CANCEL)) { 1641 if (!(flags & XFS_BLF_CANCEL)) {
1642 trace_xfs_log_recover_buf_not_cancel(log, buf_f); 1642 trace_xfs_log_recover_buf_not_cancel(log, buf_f);
1643 return; 1643 return;
1644 } 1644 }
@@ -1696,7 +1696,7 @@ xlog_recover_do_buffer_pass1(
1696 * Check to see whether the buffer being recovered has a corresponding 1696 * Check to see whether the buffer being recovered has a corresponding
1697 * entry in the buffer cancel record table. If it does then return 1 1697 * entry in the buffer cancel record table. If it does then return 1
1698 * so that it will be cancelled, otherwise return 0. If the buffer is 1698 * so that it will be cancelled, otherwise return 0. If the buffer is
1699 * actually a buffer cancel item (XFS_BLI_CANCEL is set), then decrement 1699 * actually a buffer cancel item (XFS_BLF_CANCEL is set), then decrement
1700 * the refcount on the entry in the table and remove it from the table 1700 * the refcount on the entry in the table and remove it from the table
1701 * if this is the last reference. 1701 * if this is the last reference.
1702 * 1702 *
@@ -1721,7 +1721,7 @@ xlog_check_buffer_cancelled(
1721 * There is nothing in the table built in pass one, 1721 * There is nothing in the table built in pass one,
1722 * so this buffer must not be cancelled. 1722 * so this buffer must not be cancelled.
1723 */ 1723 */
1724 ASSERT(!(flags & XFS_BLI_CANCEL)); 1724 ASSERT(!(flags & XFS_BLF_CANCEL));
1725 return 0; 1725 return 0;
1726 } 1726 }
1727 1727
@@ -1733,7 +1733,7 @@ xlog_check_buffer_cancelled(
1733 * There is no corresponding entry in the table built 1733 * There is no corresponding entry in the table built
1734 * in pass one, so this buffer has not been cancelled. 1734 * in pass one, so this buffer has not been cancelled.
1735 */ 1735 */
1736 ASSERT(!(flags & XFS_BLI_CANCEL)); 1736 ASSERT(!(flags & XFS_BLF_CANCEL));
1737 return 0; 1737 return 0;
1738 } 1738 }
1739 1739
@@ -1752,7 +1752,7 @@ xlog_check_buffer_cancelled(
1752 * one in the table and remove it if this is the 1752 * one in the table and remove it if this is the
1753 * last reference. 1753 * last reference.
1754 */ 1754 */
1755 if (flags & XFS_BLI_CANCEL) { 1755 if (flags & XFS_BLF_CANCEL) {
1756 bcp->bc_refcount--; 1756 bcp->bc_refcount--;
1757 if (bcp->bc_refcount == 0) { 1757 if (bcp->bc_refcount == 0) {
1758 if (prevp == NULL) { 1758 if (prevp == NULL) {
@@ -1772,7 +1772,7 @@ xlog_check_buffer_cancelled(
1772 * We didn't find a corresponding entry in the table, so 1772 * We didn't find a corresponding entry in the table, so
1773 * return 0 so that the buffer is NOT cancelled. 1773 * return 0 so that the buffer is NOT cancelled.
1774 */ 1774 */
1775 ASSERT(!(flags & XFS_BLI_CANCEL)); 1775 ASSERT(!(flags & XFS_BLF_CANCEL));
1776 return 0; 1776 return 0;
1777} 1777}
1778 1778
@@ -1874,8 +1874,8 @@ xlog_recover_do_inode_buffer(
1874 nbits = xfs_contig_bits(data_map, map_size, 1874 nbits = xfs_contig_bits(data_map, map_size,
1875 bit); 1875 bit);
1876 ASSERT(nbits > 0); 1876 ASSERT(nbits > 0);
1877 reg_buf_offset = bit << XFS_BLI_SHIFT; 1877 reg_buf_offset = bit << XFS_BLF_SHIFT;
1878 reg_buf_bytes = nbits << XFS_BLI_SHIFT; 1878 reg_buf_bytes = nbits << XFS_BLF_SHIFT;
1879 item_index++; 1879 item_index++;
1880 } 1880 }
1881 1881
@@ -1889,7 +1889,7 @@ xlog_recover_do_inode_buffer(
1889 } 1889 }
1890 1890
1891 ASSERT(item->ri_buf[item_index].i_addr != NULL); 1891 ASSERT(item->ri_buf[item_index].i_addr != NULL);
1892 ASSERT((item->ri_buf[item_index].i_len % XFS_BLI_CHUNK) == 0); 1892 ASSERT((item->ri_buf[item_index].i_len % XFS_BLF_CHUNK) == 0);
1893 ASSERT((reg_buf_offset + reg_buf_bytes) <= XFS_BUF_COUNT(bp)); 1893 ASSERT((reg_buf_offset + reg_buf_bytes) <= XFS_BUF_COUNT(bp));
1894 1894
1895 /* 1895 /*
@@ -1955,9 +1955,9 @@ xlog_recover_do_reg_buffer(
1955 nbits = xfs_contig_bits(data_map, map_size, bit); 1955 nbits = xfs_contig_bits(data_map, map_size, bit);
1956 ASSERT(nbits > 0); 1956 ASSERT(nbits > 0);
1957 ASSERT(item->ri_buf[i].i_addr != NULL); 1957 ASSERT(item->ri_buf[i].i_addr != NULL);
1958 ASSERT(item->ri_buf[i].i_len % XFS_BLI_CHUNK == 0); 1958 ASSERT(item->ri_buf[i].i_len % XFS_BLF_CHUNK == 0);
1959 ASSERT(XFS_BUF_COUNT(bp) >= 1959 ASSERT(XFS_BUF_COUNT(bp) >=
1960 ((uint)bit << XFS_BLI_SHIFT)+(nbits<<XFS_BLI_SHIFT)); 1960 ((uint)bit << XFS_BLF_SHIFT)+(nbits<<XFS_BLF_SHIFT));
1961 1961
1962 /* 1962 /*
1963 * Do a sanity check if this is a dquot buffer. Just checking 1963 * Do a sanity check if this is a dquot buffer. Just checking
@@ -1966,7 +1966,7 @@ xlog_recover_do_reg_buffer(
1966 */ 1966 */
1967 error = 0; 1967 error = 0;
1968 if (buf_f->blf_flags & 1968 if (buf_f->blf_flags &
1969 (XFS_BLI_UDQUOT_BUF|XFS_BLI_PDQUOT_BUF|XFS_BLI_GDQUOT_BUF)) { 1969 (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) {
1970 if (item->ri_buf[i].i_addr == NULL) { 1970 if (item->ri_buf[i].i_addr == NULL) {
1971 cmn_err(CE_ALERT, 1971 cmn_err(CE_ALERT,
1972 "XFS: NULL dquot in %s.", __func__); 1972 "XFS: NULL dquot in %s.", __func__);
@@ -1987,9 +1987,9 @@ xlog_recover_do_reg_buffer(
1987 } 1987 }
1988 1988
1989 memcpy(xfs_buf_offset(bp, 1989 memcpy(xfs_buf_offset(bp,
1990 (uint)bit << XFS_BLI_SHIFT), /* dest */ 1990 (uint)bit << XFS_BLF_SHIFT), /* dest */
1991 item->ri_buf[i].i_addr, /* source */ 1991 item->ri_buf[i].i_addr, /* source */
1992 nbits<<XFS_BLI_SHIFT); /* length */ 1992 nbits<<XFS_BLF_SHIFT); /* length */
1993 next: 1993 next:
1994 i++; 1994 i++;
1995 bit += nbits; 1995 bit += nbits;
@@ -2148,11 +2148,11 @@ xlog_recover_do_dquot_buffer(
2148 } 2148 }
2149 2149
2150 type = 0; 2150 type = 0;
2151 if (buf_f->blf_flags & XFS_BLI_UDQUOT_BUF) 2151 if (buf_f->blf_flags & XFS_BLF_UDQUOT_BUF)
2152 type |= XFS_DQ_USER; 2152 type |= XFS_DQ_USER;
2153 if (buf_f->blf_flags & XFS_BLI_PDQUOT_BUF) 2153 if (buf_f->blf_flags & XFS_BLF_PDQUOT_BUF)
2154 type |= XFS_DQ_PROJ; 2154 type |= XFS_DQ_PROJ;
2155 if (buf_f->blf_flags & XFS_BLI_GDQUOT_BUF) 2155 if (buf_f->blf_flags & XFS_BLF_GDQUOT_BUF)
2156 type |= XFS_DQ_GROUP; 2156 type |= XFS_DQ_GROUP;
2157 /* 2157 /*
2158 * This type of quotas was turned off, so ignore this buffer 2158 * This type of quotas was turned off, so ignore this buffer
@@ -2173,7 +2173,7 @@ xlog_recover_do_dquot_buffer(
2173 * here which overlaps that may be stale. 2173 * here which overlaps that may be stale.
2174 * 2174 *
2175 * When meta-data buffers are freed at run time we log a buffer item 2175 * When meta-data buffers are freed at run time we log a buffer item
2176 * with the XFS_BLI_CANCEL bit set to indicate that previous copies 2176 * with the XFS_BLF_CANCEL bit set to indicate that previous copies
2177 * of the buffer in the log should not be replayed at recovery time. 2177 * of the buffer in the log should not be replayed at recovery time.
2178 * This is so that if the blocks covered by the buffer are reused for 2178 * This is so that if the blocks covered by the buffer are reused for
2179 * file data before we crash we don't end up replaying old, freed 2179 * file data before we crash we don't end up replaying old, freed
@@ -2207,7 +2207,7 @@ xlog_recover_do_buffer_trans(
2207 if (pass == XLOG_RECOVER_PASS1) { 2207 if (pass == XLOG_RECOVER_PASS1) {
2208 /* 2208 /*
2209 * In this pass we're only looking for buf items 2209 * In this pass we're only looking for buf items
2210 * with the XFS_BLI_CANCEL bit set. 2210 * with the XFS_BLF_CANCEL bit set.
2211 */ 2211 */
2212 xlog_recover_do_buffer_pass1(log, buf_f); 2212 xlog_recover_do_buffer_pass1(log, buf_f);
2213 return 0; 2213 return 0;
@@ -2244,7 +2244,7 @@ xlog_recover_do_buffer_trans(
2244 2244
2245 mp = log->l_mp; 2245 mp = log->l_mp;
2246 buf_flags = XBF_LOCK; 2246 buf_flags = XBF_LOCK;
2247 if (!(flags & XFS_BLI_INODE_BUF)) 2247 if (!(flags & XFS_BLF_INODE_BUF))
2248 buf_flags |= XBF_MAPPED; 2248 buf_flags |= XBF_MAPPED;
2249 2249
2250 bp = xfs_buf_read(mp->m_ddev_targp, blkno, len, buf_flags); 2250 bp = xfs_buf_read(mp->m_ddev_targp, blkno, len, buf_flags);
@@ -2257,10 +2257,10 @@ xlog_recover_do_buffer_trans(
2257 } 2257 }
2258 2258
2259 error = 0; 2259 error = 0;
2260 if (flags & XFS_BLI_INODE_BUF) { 2260 if (flags & XFS_BLF_INODE_BUF) {
2261 error = xlog_recover_do_inode_buffer(mp, item, bp, buf_f); 2261 error = xlog_recover_do_inode_buffer(mp, item, bp, buf_f);
2262 } else if (flags & 2262 } else if (flags &
2263 (XFS_BLI_UDQUOT_BUF|XFS_BLI_PDQUOT_BUF|XFS_BLI_GDQUOT_BUF)) { 2263 (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) {
2264 xlog_recover_do_dquot_buffer(mp, log, item, bp, buf_f); 2264 xlog_recover_do_dquot_buffer(mp, log, item, bp, buf_f);
2265 } else { 2265 } else {
2266 xlog_recover_do_reg_buffer(mp, item, bp, buf_f); 2266 xlog_recover_do_reg_buffer(mp, item, bp, buf_f);
diff --git a/fs/xfs/xfs_log_recover.h b/fs/xfs/xfs_log_recover.h
index 75d749207258..1c55ccbb379d 100644
--- a/fs/xfs/xfs_log_recover.h
+++ b/fs/xfs/xfs_log_recover.h
@@ -28,7 +28,7 @@
28#define XLOG_RHASH(tid) \ 28#define XLOG_RHASH(tid) \
29 ((((__uint32_t)tid)>>XLOG_RHASH_SHIFT) & (XLOG_RHASH_SIZE-1)) 29 ((((__uint32_t)tid)>>XLOG_RHASH_SHIFT) & (XLOG_RHASH_SIZE-1))
30 30
31#define XLOG_MAX_REGIONS_IN_ITEM (XFS_MAX_BLOCKSIZE / XFS_BLI_CHUNK / 2 + 1) 31#define XLOG_MAX_REGIONS_IN_ITEM (XFS_MAX_BLOCKSIZE / XFS_BLF_CHUNK / 2 + 1)
32 32
33 33
34/* 34/*
diff --git a/fs/xfs/xfs_mount.h b/fs/xfs/xfs_mount.h
index 9ff48a16a7ee..1d2c7eed4eda 100644
--- a/fs/xfs/xfs_mount.h
+++ b/fs/xfs/xfs_mount.h
@@ -268,6 +268,7 @@ typedef struct xfs_mount {
268#define XFS_MOUNT_WSYNC (1ULL << 0) /* for nfs - all metadata ops 268#define XFS_MOUNT_WSYNC (1ULL << 0) /* for nfs - all metadata ops
269 must be synchronous except 269 must be synchronous except
270 for space allocations */ 270 for space allocations */
271#define XFS_MOUNT_DELAYLOG (1ULL << 1) /* delayed logging is enabled */
271#define XFS_MOUNT_DMAPI (1ULL << 2) /* dmapi is enabled */ 272#define XFS_MOUNT_DMAPI (1ULL << 2) /* dmapi is enabled */
272#define XFS_MOUNT_WAS_CLEAN (1ULL << 3) 273#define XFS_MOUNT_WAS_CLEAN (1ULL << 3)
273#define XFS_MOUNT_FS_SHUTDOWN (1ULL << 4) /* atomic stop of all filesystem 274#define XFS_MOUNT_FS_SHUTDOWN (1ULL << 4) /* atomic stop of all filesystem
diff --git a/fs/xfs/xfs_trans.c b/fs/xfs/xfs_trans.c
index be578ecb4af2..ce558efa2ea0 100644
--- a/fs/xfs/xfs_trans.c
+++ b/fs/xfs/xfs_trans.c
@@ -44,6 +44,7 @@
44#include "xfs_trans_priv.h" 44#include "xfs_trans_priv.h"
45#include "xfs_trans_space.h" 45#include "xfs_trans_space.h"
46#include "xfs_inode_item.h" 46#include "xfs_inode_item.h"
47#include "xfs_trace.h"
47 48
48kmem_zone_t *xfs_trans_zone; 49kmem_zone_t *xfs_trans_zone;
49 50
@@ -243,9 +244,8 @@ _xfs_trans_alloc(
243 tp->t_type = type; 244 tp->t_type = type;
244 tp->t_mountp = mp; 245 tp->t_mountp = mp;
245 tp->t_items_free = XFS_LIC_NUM_SLOTS; 246 tp->t_items_free = XFS_LIC_NUM_SLOTS;
246 tp->t_busy_free = XFS_LBC_NUM_SLOTS;
247 xfs_lic_init(&(tp->t_items)); 247 xfs_lic_init(&(tp->t_items));
248 XFS_LBC_INIT(&(tp->t_busy)); 248 INIT_LIST_HEAD(&tp->t_busy);
249 return tp; 249 return tp;
250} 250}
251 251
@@ -255,8 +255,13 @@ _xfs_trans_alloc(
255 */ 255 */
256STATIC void 256STATIC void
257xfs_trans_free( 257xfs_trans_free(
258 xfs_trans_t *tp) 258 struct xfs_trans *tp)
259{ 259{
260 struct xfs_busy_extent *busyp, *n;
261
262 list_for_each_entry_safe(busyp, n, &tp->t_busy, list)
263 xfs_alloc_busy_clear(tp->t_mountp, busyp);
264
260 atomic_dec(&tp->t_mountp->m_active_trans); 265 atomic_dec(&tp->t_mountp->m_active_trans);
261 xfs_trans_free_dqinfo(tp); 266 xfs_trans_free_dqinfo(tp);
262 kmem_zone_free(xfs_trans_zone, tp); 267 kmem_zone_free(xfs_trans_zone, tp);
@@ -285,9 +290,8 @@ xfs_trans_dup(
285 ntp->t_type = tp->t_type; 290 ntp->t_type = tp->t_type;
286 ntp->t_mountp = tp->t_mountp; 291 ntp->t_mountp = tp->t_mountp;
287 ntp->t_items_free = XFS_LIC_NUM_SLOTS; 292 ntp->t_items_free = XFS_LIC_NUM_SLOTS;
288 ntp->t_busy_free = XFS_LBC_NUM_SLOTS;
289 xfs_lic_init(&(ntp->t_items)); 293 xfs_lic_init(&(ntp->t_items));
290 XFS_LBC_INIT(&(ntp->t_busy)); 294 INIT_LIST_HEAD(&ntp->t_busy);
291 295
292 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES); 296 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
293 ASSERT(tp->t_ticket != NULL); 297 ASSERT(tp->t_ticket != NULL);
@@ -423,7 +427,6 @@ undo_blocks:
423 return error; 427 return error;
424} 428}
425 429
426
427/* 430/*
428 * Record the indicated change to the given field for application 431 * Record the indicated change to the given field for application
429 * to the file system's superblock when the transaction commits. 432 * to the file system's superblock when the transaction commits.
@@ -652,7 +655,7 @@ xfs_trans_apply_sb_deltas(
652 * XFS_TRANS_SB_DIRTY will not be set when the transaction is updated but we 655 * XFS_TRANS_SB_DIRTY will not be set when the transaction is updated but we
653 * still need to update the incore superblock with the changes. 656 * still need to update the incore superblock with the changes.
654 */ 657 */
655STATIC void 658void
656xfs_trans_unreserve_and_mod_sb( 659xfs_trans_unreserve_and_mod_sb(
657 xfs_trans_t *tp) 660 xfs_trans_t *tp)
658{ 661{
@@ -880,7 +883,7 @@ xfs_trans_fill_vecs(
880 * they could be immediately flushed and we'd have to race with the flusher 883 * they could be immediately flushed and we'd have to race with the flusher
881 * trying to pull the item from the AIL as we add it. 884 * trying to pull the item from the AIL as we add it.
882 */ 885 */
883static void 886void
884xfs_trans_item_committed( 887xfs_trans_item_committed(
885 struct xfs_log_item *lip, 888 struct xfs_log_item *lip,
886 xfs_lsn_t commit_lsn, 889 xfs_lsn_t commit_lsn,
@@ -930,26 +933,6 @@ xfs_trans_item_committed(
930 IOP_UNPIN(lip); 933 IOP_UNPIN(lip);
931} 934}
932 935
933/* Clear all the per-AG busy list items listed in this transaction */
934static void
935xfs_trans_clear_busy_extents(
936 struct xfs_trans *tp)
937{
938 xfs_log_busy_chunk_t *lbcp;
939 xfs_log_busy_slot_t *lbsp;
940 int i;
941
942 for (lbcp = &tp->t_busy; lbcp != NULL; lbcp = lbcp->lbc_next) {
943 i = 0;
944 for (lbsp = lbcp->lbc_busy; i < lbcp->lbc_unused; i++, lbsp++) {
945 if (XFS_LBC_ISFREE(lbcp, i))
946 continue;
947 xfs_alloc_clear_busy(tp, lbsp->lbc_ag, lbsp->lbc_idx);
948 }
949 }
950 xfs_trans_free_busy(tp);
951}
952
953/* 936/*
954 * This is typically called by the LM when a transaction has been fully 937 * This is typically called by the LM when a transaction has been fully
955 * committed to disk. It needs to unpin the items which have 938 * committed to disk. It needs to unpin the items which have
@@ -984,7 +967,6 @@ xfs_trans_committed(
984 kmem_free(licp); 967 kmem_free(licp);
985 } 968 }
986 969
987 xfs_trans_clear_busy_extents(tp);
988 xfs_trans_free(tp); 970 xfs_trans_free(tp);
989} 971}
990 972
@@ -1012,8 +994,7 @@ xfs_trans_uncommit(
1012 xfs_trans_unreserve_and_mod_sb(tp); 994 xfs_trans_unreserve_and_mod_sb(tp);
1013 xfs_trans_unreserve_and_mod_dquots(tp); 995 xfs_trans_unreserve_and_mod_dquots(tp);
1014 996
1015 xfs_trans_free_items(tp, flags); 997 xfs_trans_free_items(tp, NULLCOMMITLSN, flags);
1016 xfs_trans_free_busy(tp);
1017 xfs_trans_free(tp); 998 xfs_trans_free(tp);
1018} 999}
1019 1000
@@ -1075,6 +1056,8 @@ xfs_trans_commit_iclog(
1075 *commit_lsn = xfs_log_done(mp, tp->t_ticket, &commit_iclog, log_flags); 1056 *commit_lsn = xfs_log_done(mp, tp->t_ticket, &commit_iclog, log_flags);
1076 1057
1077 tp->t_commit_lsn = *commit_lsn; 1058 tp->t_commit_lsn = *commit_lsn;
1059 trace_xfs_trans_commit_lsn(tp);
1060
1078 if (nvec > XFS_TRANS_LOGVEC_COUNT) 1061 if (nvec > XFS_TRANS_LOGVEC_COUNT)
1079 kmem_free(log_vector); 1062 kmem_free(log_vector);
1080 1063
@@ -1161,6 +1144,93 @@ xfs_trans_commit_iclog(
1161 return xfs_log_release_iclog(mp, commit_iclog); 1144 return xfs_log_release_iclog(mp, commit_iclog);
1162} 1145}
1163 1146
1147/*
1148 * Walk the log items and allocate log vector structures for
1149 * each item large enough to fit all the vectors they require.
1150 * Note that this format differs from the old log vector format in
1151 * that there is no transaction header in these log vectors.
1152 */
1153STATIC struct xfs_log_vec *
1154xfs_trans_alloc_log_vecs(
1155 xfs_trans_t *tp)
1156{
1157 xfs_log_item_desc_t *lidp;
1158 struct xfs_log_vec *lv = NULL;
1159 struct xfs_log_vec *ret_lv = NULL;
1160
1161 lidp = xfs_trans_first_item(tp);
1162
1163 /* Bail out if we didn't find a log item. */
1164 if (!lidp) {
1165 ASSERT(0);
1166 return NULL;
1167 }
1168
1169 while (lidp != NULL) {
1170 struct xfs_log_vec *new_lv;
1171
1172 /* Skip items which aren't dirty in this transaction. */
1173 if (!(lidp->lid_flags & XFS_LID_DIRTY)) {
1174 lidp = xfs_trans_next_item(tp, lidp);
1175 continue;
1176 }
1177
1178 /* Skip items that do not have any vectors for writing */
1179 lidp->lid_size = IOP_SIZE(lidp->lid_item);
1180 if (!lidp->lid_size) {
1181 lidp = xfs_trans_next_item(tp, lidp);
1182 continue;
1183 }
1184
1185 new_lv = kmem_zalloc(sizeof(*new_lv) +
1186 lidp->lid_size * sizeof(struct xfs_log_iovec),
1187 KM_SLEEP);
1188
1189 /* The allocated iovec region lies beyond the log vector. */
1190 new_lv->lv_iovecp = (struct xfs_log_iovec *)&new_lv[1];
1191 new_lv->lv_niovecs = lidp->lid_size;
1192 new_lv->lv_item = lidp->lid_item;
1193 if (!ret_lv)
1194 ret_lv = new_lv;
1195 else
1196 lv->lv_next = new_lv;
1197 lv = new_lv;
1198 lidp = xfs_trans_next_item(tp, lidp);
1199 }
1200
1201 return ret_lv;
1202}
1203
1204static int
1205xfs_trans_commit_cil(
1206 struct xfs_mount *mp,
1207 struct xfs_trans *tp,
1208 xfs_lsn_t *commit_lsn,
1209 int flags)
1210{
1211 struct xfs_log_vec *log_vector;
1212 int error;
1213
1214 /*
1215 * Get each log item to allocate a vector structure for
1216 * the log item to to pass to the log write code. The
1217 * CIL commit code will format the vector and save it away.
1218 */
1219 log_vector = xfs_trans_alloc_log_vecs(tp);
1220 if (!log_vector)
1221 return ENOMEM;
1222
1223 error = xfs_log_commit_cil(mp, tp, log_vector, commit_lsn, flags);
1224 if (error)
1225 return error;
1226
1227 current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
1228
1229 /* xfs_trans_free_items() unlocks them first */
1230 xfs_trans_free_items(tp, *commit_lsn, 0);
1231 xfs_trans_free(tp);
1232 return 0;
1233}
1164 1234
1165/* 1235/*
1166 * xfs_trans_commit 1236 * xfs_trans_commit
@@ -1221,7 +1291,11 @@ _xfs_trans_commit(
1221 xfs_trans_apply_sb_deltas(tp); 1291 xfs_trans_apply_sb_deltas(tp);
1222 xfs_trans_apply_dquot_deltas(tp); 1292 xfs_trans_apply_dquot_deltas(tp);
1223 1293
1224 error = xfs_trans_commit_iclog(mp, tp, &commit_lsn, flags); 1294 if (mp->m_flags & XFS_MOUNT_DELAYLOG)
1295 error = xfs_trans_commit_cil(mp, tp, &commit_lsn, flags);
1296 else
1297 error = xfs_trans_commit_iclog(mp, tp, &commit_lsn, flags);
1298
1225 if (error == ENOMEM) { 1299 if (error == ENOMEM) {
1226 xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR); 1300 xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR);
1227 error = XFS_ERROR(EIO); 1301 error = XFS_ERROR(EIO);
@@ -1259,8 +1333,7 @@ out_unreserve:
1259 error = XFS_ERROR(EIO); 1333 error = XFS_ERROR(EIO);
1260 } 1334 }
1261 current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS); 1335 current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
1262 xfs_trans_free_items(tp, error ? XFS_TRANS_ABORT : 0); 1336 xfs_trans_free_items(tp, NULLCOMMITLSN, error ? XFS_TRANS_ABORT : 0);
1263 xfs_trans_free_busy(tp);
1264 xfs_trans_free(tp); 1337 xfs_trans_free(tp);
1265 1338
1266 XFS_STATS_INC(xs_trans_empty); 1339 XFS_STATS_INC(xs_trans_empty);
@@ -1338,8 +1411,7 @@ xfs_trans_cancel(
1338 /* mark this thread as no longer being in a transaction */ 1411 /* mark this thread as no longer being in a transaction */
1339 current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS); 1412 current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
1340 1413
1341 xfs_trans_free_items(tp, flags); 1414 xfs_trans_free_items(tp, NULLCOMMITLSN, flags);
1342 xfs_trans_free_busy(tp);
1343 xfs_trans_free(tp); 1415 xfs_trans_free(tp);
1344} 1416}
1345 1417
diff --git a/fs/xfs/xfs_trans.h b/fs/xfs/xfs_trans.h
index c62beee0921e..8c69e7824f68 100644
--- a/fs/xfs/xfs_trans.h
+++ b/fs/xfs/xfs_trans.h
@@ -106,7 +106,8 @@ typedef struct xfs_trans_header {
106#define XFS_TRANS_GROWFSRT_FREE 39 106#define XFS_TRANS_GROWFSRT_FREE 39
107#define XFS_TRANS_SWAPEXT 40 107#define XFS_TRANS_SWAPEXT 40
108#define XFS_TRANS_SB_COUNT 41 108#define XFS_TRANS_SB_COUNT 41
109#define XFS_TRANS_TYPE_MAX 41 109#define XFS_TRANS_CHECKPOINT 42
110#define XFS_TRANS_TYPE_MAX 42
110/* new transaction types need to be reflected in xfs_logprint(8) */ 111/* new transaction types need to be reflected in xfs_logprint(8) */
111 112
112#define XFS_TRANS_TYPES \ 113#define XFS_TRANS_TYPES \
@@ -148,6 +149,7 @@ typedef struct xfs_trans_header {
148 { XFS_TRANS_GROWFSRT_FREE, "GROWFSRT_FREE" }, \ 149 { XFS_TRANS_GROWFSRT_FREE, "GROWFSRT_FREE" }, \
149 { XFS_TRANS_SWAPEXT, "SWAPEXT" }, \ 150 { XFS_TRANS_SWAPEXT, "SWAPEXT" }, \
150 { XFS_TRANS_SB_COUNT, "SB_COUNT" }, \ 151 { XFS_TRANS_SB_COUNT, "SB_COUNT" }, \
152 { XFS_TRANS_CHECKPOINT, "CHECKPOINT" }, \
151 { XFS_TRANS_DUMMY1, "DUMMY1" }, \ 153 { XFS_TRANS_DUMMY1, "DUMMY1" }, \
152 { XFS_TRANS_DUMMY2, "DUMMY2" }, \ 154 { XFS_TRANS_DUMMY2, "DUMMY2" }, \
153 { XLOG_UNMOUNT_REC_TYPE, "UNMOUNT" } 155 { XLOG_UNMOUNT_REC_TYPE, "UNMOUNT" }
@@ -813,6 +815,7 @@ struct xfs_log_item_desc;
813struct xfs_mount; 815struct xfs_mount;
814struct xfs_trans; 816struct xfs_trans;
815struct xfs_dquot_acct; 817struct xfs_dquot_acct;
818struct xfs_busy_extent;
816 819
817typedef struct xfs_log_item { 820typedef struct xfs_log_item {
818 struct list_head li_ail; /* AIL pointers */ 821 struct list_head li_ail; /* AIL pointers */
@@ -828,6 +831,11 @@ typedef struct xfs_log_item {
828 /* buffer item iodone */ 831 /* buffer item iodone */
829 /* callback func */ 832 /* callback func */
830 struct xfs_item_ops *li_ops; /* function list */ 833 struct xfs_item_ops *li_ops; /* function list */
834
835 /* delayed logging */
836 struct list_head li_cil; /* CIL pointers */
837 struct xfs_log_vec *li_lv; /* active log vector */
838 xfs_lsn_t li_seq; /* CIL commit seq */
831} xfs_log_item_t; 839} xfs_log_item_t;
832 840
833#define XFS_LI_IN_AIL 0x1 841#define XFS_LI_IN_AIL 0x1
@@ -872,34 +880,6 @@ typedef struct xfs_item_ops {
872#define XFS_ITEM_PUSHBUF 3 880#define XFS_ITEM_PUSHBUF 3
873 881
874/* 882/*
875 * This structure is used to maintain a list of block ranges that have been
876 * freed in the transaction. The ranges are listed in the perag[] busy list
877 * between when they're freed and the transaction is committed to disk.
878 */
879
880typedef struct xfs_log_busy_slot {
881 xfs_agnumber_t lbc_ag;
882 ushort lbc_idx; /* index in perag.busy[] */
883} xfs_log_busy_slot_t;
884
885#define XFS_LBC_NUM_SLOTS 31
886typedef struct xfs_log_busy_chunk {
887 struct xfs_log_busy_chunk *lbc_next;
888 uint lbc_free; /* free slots bitmask */
889 ushort lbc_unused; /* first unused */
890 xfs_log_busy_slot_t lbc_busy[XFS_LBC_NUM_SLOTS];
891} xfs_log_busy_chunk_t;
892
893#define XFS_LBC_MAX_SLOT (XFS_LBC_NUM_SLOTS - 1)
894#define XFS_LBC_FREEMASK ((1U << XFS_LBC_NUM_SLOTS) - 1)
895
896#define XFS_LBC_INIT(cp) ((cp)->lbc_free = XFS_LBC_FREEMASK)
897#define XFS_LBC_CLAIM(cp, slot) ((cp)->lbc_free &= ~(1 << (slot)))
898#define XFS_LBC_SLOT(cp, slot) (&((cp)->lbc_busy[(slot)]))
899#define XFS_LBC_VACANCY(cp) (((cp)->lbc_free) & XFS_LBC_FREEMASK)
900#define XFS_LBC_ISFREE(cp, slot) ((cp)->lbc_free & (1 << (slot)))
901
902/*
903 * This is the type of function which can be given to xfs_trans_callback() 883 * This is the type of function which can be given to xfs_trans_callback()
904 * to be called upon the transaction's commit to disk. 884 * to be called upon the transaction's commit to disk.
905 */ 885 */
@@ -950,8 +930,7 @@ typedef struct xfs_trans {
950 unsigned int t_items_free; /* log item descs free */ 930 unsigned int t_items_free; /* log item descs free */
951 xfs_log_item_chunk_t t_items; /* first log item desc chunk */ 931 xfs_log_item_chunk_t t_items; /* first log item desc chunk */
952 xfs_trans_header_t t_header; /* header for in-log trans */ 932 xfs_trans_header_t t_header; /* header for in-log trans */
953 unsigned int t_busy_free; /* busy descs free */ 933 struct list_head t_busy; /* list of busy extents */
954 xfs_log_busy_chunk_t t_busy; /* busy/async free blocks */
955 unsigned long t_pflags; /* saved process flags state */ 934 unsigned long t_pflags; /* saved process flags state */
956} xfs_trans_t; 935} xfs_trans_t;
957 936
@@ -1025,9 +1004,6 @@ int _xfs_trans_commit(xfs_trans_t *,
1025void xfs_trans_cancel(xfs_trans_t *, int); 1004void xfs_trans_cancel(xfs_trans_t *, int);
1026int xfs_trans_ail_init(struct xfs_mount *); 1005int xfs_trans_ail_init(struct xfs_mount *);
1027void xfs_trans_ail_destroy(struct xfs_mount *); 1006void xfs_trans_ail_destroy(struct xfs_mount *);
1028xfs_log_busy_slot_t *xfs_trans_add_busy(xfs_trans_t *tp,
1029 xfs_agnumber_t ag,
1030 xfs_extlen_t idx);
1031 1007
1032extern kmem_zone_t *xfs_trans_zone; 1008extern kmem_zone_t *xfs_trans_zone;
1033 1009
diff --git a/fs/xfs/xfs_trans_buf.c b/fs/xfs/xfs_trans_buf.c
index 9cd809025f3a..63d81a22f4fd 100644
--- a/fs/xfs/xfs_trans_buf.c
+++ b/fs/xfs/xfs_trans_buf.c
@@ -114,7 +114,7 @@ _xfs_trans_bjoin(
114 xfs_buf_item_init(bp, tp->t_mountp); 114 xfs_buf_item_init(bp, tp->t_mountp);
115 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *); 115 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
116 ASSERT(!(bip->bli_flags & XFS_BLI_STALE)); 116 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
117 ASSERT(!(bip->bli_format.blf_flags & XFS_BLI_CANCEL)); 117 ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_CANCEL));
118 ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED)); 118 ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
119 if (reset_recur) 119 if (reset_recur)
120 bip->bli_recur = 0; 120 bip->bli_recur = 0;
@@ -511,7 +511,7 @@ xfs_trans_brelse(xfs_trans_t *tp,
511 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *); 511 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
512 ASSERT(bip->bli_item.li_type == XFS_LI_BUF); 512 ASSERT(bip->bli_item.li_type == XFS_LI_BUF);
513 ASSERT(!(bip->bli_flags & XFS_BLI_STALE)); 513 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
514 ASSERT(!(bip->bli_format.blf_flags & XFS_BLI_CANCEL)); 514 ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_CANCEL));
515 ASSERT(atomic_read(&bip->bli_refcount) > 0); 515 ASSERT(atomic_read(&bip->bli_refcount) > 0);
516 516
517 /* 517 /*
@@ -619,7 +619,7 @@ xfs_trans_bhold(xfs_trans_t *tp,
619 619
620 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *); 620 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
621 ASSERT(!(bip->bli_flags & XFS_BLI_STALE)); 621 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
622 ASSERT(!(bip->bli_format.blf_flags & XFS_BLI_CANCEL)); 622 ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_CANCEL));
623 ASSERT(atomic_read(&bip->bli_refcount) > 0); 623 ASSERT(atomic_read(&bip->bli_refcount) > 0);
624 bip->bli_flags |= XFS_BLI_HOLD; 624 bip->bli_flags |= XFS_BLI_HOLD;
625 trace_xfs_trans_bhold(bip); 625 trace_xfs_trans_bhold(bip);
@@ -641,7 +641,7 @@ xfs_trans_bhold_release(xfs_trans_t *tp,
641 641
642 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *); 642 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
643 ASSERT(!(bip->bli_flags & XFS_BLI_STALE)); 643 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
644 ASSERT(!(bip->bli_format.blf_flags & XFS_BLI_CANCEL)); 644 ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_CANCEL));
645 ASSERT(atomic_read(&bip->bli_refcount) > 0); 645 ASSERT(atomic_read(&bip->bli_refcount) > 0);
646 ASSERT(bip->bli_flags & XFS_BLI_HOLD); 646 ASSERT(bip->bli_flags & XFS_BLI_HOLD);
647 bip->bli_flags &= ~XFS_BLI_HOLD; 647 bip->bli_flags &= ~XFS_BLI_HOLD;
@@ -704,7 +704,7 @@ xfs_trans_log_buf(xfs_trans_t *tp,
704 bip->bli_flags &= ~XFS_BLI_STALE; 704 bip->bli_flags &= ~XFS_BLI_STALE;
705 ASSERT(XFS_BUF_ISSTALE(bp)); 705 ASSERT(XFS_BUF_ISSTALE(bp));
706 XFS_BUF_UNSTALE(bp); 706 XFS_BUF_UNSTALE(bp);
707 bip->bli_format.blf_flags &= ~XFS_BLI_CANCEL; 707 bip->bli_format.blf_flags &= ~XFS_BLF_CANCEL;
708 } 708 }
709 709
710 lidp = xfs_trans_find_item(tp, (xfs_log_item_t*)bip); 710 lidp = xfs_trans_find_item(tp, (xfs_log_item_t*)bip);
@@ -762,8 +762,8 @@ xfs_trans_binval(
762 ASSERT(!(XFS_BUF_ISDELAYWRITE(bp))); 762 ASSERT(!(XFS_BUF_ISDELAYWRITE(bp)));
763 ASSERT(XFS_BUF_ISSTALE(bp)); 763 ASSERT(XFS_BUF_ISSTALE(bp));
764 ASSERT(!(bip->bli_flags & (XFS_BLI_LOGGED | XFS_BLI_DIRTY))); 764 ASSERT(!(bip->bli_flags & (XFS_BLI_LOGGED | XFS_BLI_DIRTY)));
765 ASSERT(!(bip->bli_format.blf_flags & XFS_BLI_INODE_BUF)); 765 ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_INODE_BUF));
766 ASSERT(bip->bli_format.blf_flags & XFS_BLI_CANCEL); 766 ASSERT(bip->bli_format.blf_flags & XFS_BLF_CANCEL);
767 ASSERT(lidp->lid_flags & XFS_LID_DIRTY); 767 ASSERT(lidp->lid_flags & XFS_LID_DIRTY);
768 ASSERT(tp->t_flags & XFS_TRANS_DIRTY); 768 ASSERT(tp->t_flags & XFS_TRANS_DIRTY);
769 return; 769 return;
@@ -774,7 +774,7 @@ xfs_trans_binval(
774 * in the buf log item. The STALE flag will be used in 774 * in the buf log item. The STALE flag will be used in
775 * xfs_buf_item_unpin() to determine if it should clean up 775 * xfs_buf_item_unpin() to determine if it should clean up
776 * when the last reference to the buf item is given up. 776 * when the last reference to the buf item is given up.
777 * We set the XFS_BLI_CANCEL flag in the buf log format structure 777 * We set the XFS_BLF_CANCEL flag in the buf log format structure
778 * and log the buf item. This will be used at recovery time 778 * and log the buf item. This will be used at recovery time
779 * to determine that copies of the buffer in the log before 779 * to determine that copies of the buffer in the log before
780 * this should not be replayed. 780 * this should not be replayed.
@@ -792,9 +792,9 @@ xfs_trans_binval(
792 XFS_BUF_UNDELAYWRITE(bp); 792 XFS_BUF_UNDELAYWRITE(bp);
793 XFS_BUF_STALE(bp); 793 XFS_BUF_STALE(bp);
794 bip->bli_flags |= XFS_BLI_STALE; 794 bip->bli_flags |= XFS_BLI_STALE;
795 bip->bli_flags &= ~(XFS_BLI_LOGGED | XFS_BLI_DIRTY); 795 bip->bli_flags &= ~(XFS_BLI_INODE_BUF | XFS_BLI_LOGGED | XFS_BLI_DIRTY);
796 bip->bli_format.blf_flags &= ~XFS_BLI_INODE_BUF; 796 bip->bli_format.blf_flags &= ~XFS_BLF_INODE_BUF;
797 bip->bli_format.blf_flags |= XFS_BLI_CANCEL; 797 bip->bli_format.blf_flags |= XFS_BLF_CANCEL;
798 memset((char *)(bip->bli_format.blf_data_map), 0, 798 memset((char *)(bip->bli_format.blf_data_map), 0,
799 (bip->bli_format.blf_map_size * sizeof(uint))); 799 (bip->bli_format.blf_map_size * sizeof(uint)));
800 lidp->lid_flags |= XFS_LID_DIRTY; 800 lidp->lid_flags |= XFS_LID_DIRTY;
@@ -802,16 +802,16 @@ xfs_trans_binval(
802} 802}
803 803
804/* 804/*
805 * This call is used to indicate that the buffer contains on-disk 805 * This call is used to indicate that the buffer contains on-disk inodes which
806 * inodes which must be handled specially during recovery. They 806 * must be handled specially during recovery. They require special handling
807 * require special handling because only the di_next_unlinked from 807 * because only the di_next_unlinked from the inodes in the buffer should be
808 * the inodes in the buffer should be recovered. The rest of the 808 * recovered. The rest of the data in the buffer is logged via the inodes
809 * data in the buffer is logged via the inodes themselves. 809 * themselves.
810 * 810 *
811 * All we do is set the XFS_BLI_INODE_BUF flag in the buffer's log 811 * All we do is set the XFS_BLI_INODE_BUF flag in the items flags so it can be
812 * format structure so that we'll know what to do at recovery time. 812 * transferred to the buffer's log format structure so that we'll know what to
813 * do at recovery time.
813 */ 814 */
814/* ARGSUSED */
815void 815void
816xfs_trans_inode_buf( 816xfs_trans_inode_buf(
817 xfs_trans_t *tp, 817 xfs_trans_t *tp,
@@ -826,7 +826,7 @@ xfs_trans_inode_buf(
826 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *); 826 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
827 ASSERT(atomic_read(&bip->bli_refcount) > 0); 827 ASSERT(atomic_read(&bip->bli_refcount) > 0);
828 828
829 bip->bli_format.blf_flags |= XFS_BLI_INODE_BUF; 829 bip->bli_flags |= XFS_BLI_INODE_BUF;
830} 830}
831 831
832/* 832/*
@@ -908,9 +908,9 @@ xfs_trans_dquot_buf(
908 ASSERT(XFS_BUF_ISBUSY(bp)); 908 ASSERT(XFS_BUF_ISBUSY(bp));
909 ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp); 909 ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp);
910 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL); 910 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
911 ASSERT(type == XFS_BLI_UDQUOT_BUF || 911 ASSERT(type == XFS_BLF_UDQUOT_BUF ||
912 type == XFS_BLI_PDQUOT_BUF || 912 type == XFS_BLF_PDQUOT_BUF ||
913 type == XFS_BLI_GDQUOT_BUF); 913 type == XFS_BLF_GDQUOT_BUF);
914 914
915 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *); 915 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
916 ASSERT(atomic_read(&bip->bli_refcount) > 0); 916 ASSERT(atomic_read(&bip->bli_refcount) > 0);
diff --git a/fs/xfs/xfs_trans_item.c b/fs/xfs/xfs_trans_item.c
index eb3fc57f9eef..f11d37d06dcc 100644
--- a/fs/xfs/xfs_trans_item.c
+++ b/fs/xfs/xfs_trans_item.c
@@ -299,6 +299,7 @@ xfs_trans_next_item(xfs_trans_t *tp, xfs_log_item_desc_t *lidp)
299void 299void
300xfs_trans_free_items( 300xfs_trans_free_items(
301 xfs_trans_t *tp, 301 xfs_trans_t *tp,
302 xfs_lsn_t commit_lsn,
302 int flags) 303 int flags)
303{ 304{
304 xfs_log_item_chunk_t *licp; 305 xfs_log_item_chunk_t *licp;
@@ -311,7 +312,7 @@ xfs_trans_free_items(
311 * Special case the embedded chunk so we don't free it below. 312 * Special case the embedded chunk so we don't free it below.
312 */ 313 */
313 if (!xfs_lic_are_all_free(licp)) { 314 if (!xfs_lic_are_all_free(licp)) {
314 (void) xfs_trans_unlock_chunk(licp, 1, abort, NULLCOMMITLSN); 315 (void) xfs_trans_unlock_chunk(licp, 1, abort, commit_lsn);
315 xfs_lic_all_free(licp); 316 xfs_lic_all_free(licp);
316 licp->lic_unused = 0; 317 licp->lic_unused = 0;
317 } 318 }
@@ -322,7 +323,7 @@ xfs_trans_free_items(
322 */ 323 */
323 while (licp != NULL) { 324 while (licp != NULL) {
324 ASSERT(!xfs_lic_are_all_free(licp)); 325 ASSERT(!xfs_lic_are_all_free(licp));
325 (void) xfs_trans_unlock_chunk(licp, 1, abort, NULLCOMMITLSN); 326 (void) xfs_trans_unlock_chunk(licp, 1, abort, commit_lsn);
326 next_licp = licp->lic_next; 327 next_licp = licp->lic_next;
327 kmem_free(licp); 328 kmem_free(licp);
328 licp = next_licp; 329 licp = next_licp;
@@ -438,112 +439,3 @@ xfs_trans_unlock_chunk(
438 439
439 return freed; 440 return freed;
440} 441}
441
442
443/*
444 * This is called to add the given busy item to the transaction's
445 * list of busy items. It must find a free busy item descriptor
446 * or allocate a new one and add the item to that descriptor.
447 * The function returns a pointer to busy descriptor used to point
448 * to the new busy entry. The log busy entry will now point to its new
449 * descriptor with its ???? field.
450 */
451xfs_log_busy_slot_t *
452xfs_trans_add_busy(xfs_trans_t *tp, xfs_agnumber_t ag, xfs_extlen_t idx)
453{
454 xfs_log_busy_chunk_t *lbcp;
455 xfs_log_busy_slot_t *lbsp;
456 int i=0;
457
458 /*
459 * If there are no free descriptors, allocate a new chunk
460 * of them and put it at the front of the chunk list.
461 */
462 if (tp->t_busy_free == 0) {
463 lbcp = (xfs_log_busy_chunk_t*)
464 kmem_alloc(sizeof(xfs_log_busy_chunk_t), KM_SLEEP);
465 ASSERT(lbcp != NULL);
466 /*
467 * Initialize the chunk, and then
468 * claim the first slot in the newly allocated chunk.
469 */
470 XFS_LBC_INIT(lbcp);
471 XFS_LBC_CLAIM(lbcp, 0);
472 lbcp->lbc_unused = 1;
473 lbsp = XFS_LBC_SLOT(lbcp, 0);
474
475 /*
476 * Link in the new chunk and update the free count.
477 */
478 lbcp->lbc_next = tp->t_busy.lbc_next;
479 tp->t_busy.lbc_next = lbcp;
480 tp->t_busy_free = XFS_LIC_NUM_SLOTS - 1;
481
482 /*
483 * Initialize the descriptor and the generic portion
484 * of the log item.
485 *
486 * Point the new slot at this item and return it.
487 * Also point the log item at its currently active
488 * descriptor and set the item's mount pointer.
489 */
490 lbsp->lbc_ag = ag;
491 lbsp->lbc_idx = idx;
492 return lbsp;
493 }
494
495 /*
496 * Find the free descriptor. It is somewhere in the chunklist
497 * of descriptors.
498 */
499 lbcp = &tp->t_busy;
500 while (lbcp != NULL) {
501 if (XFS_LBC_VACANCY(lbcp)) {
502 if (lbcp->lbc_unused <= XFS_LBC_MAX_SLOT) {
503 i = lbcp->lbc_unused;
504 break;
505 } else {
506 /* out-of-order vacancy */
507 cmn_err(CE_DEBUG, "OOO vacancy lbcp 0x%p\n", lbcp);
508 ASSERT(0);
509 }
510 }
511 lbcp = lbcp->lbc_next;
512 }
513 ASSERT(lbcp != NULL);
514 /*
515 * If we find a free descriptor, claim it,
516 * initialize it, and return it.
517 */
518 XFS_LBC_CLAIM(lbcp, i);
519 if (lbcp->lbc_unused <= i) {
520 lbcp->lbc_unused = i + 1;
521 }
522 lbsp = XFS_LBC_SLOT(lbcp, i);
523 tp->t_busy_free--;
524 lbsp->lbc_ag = ag;
525 lbsp->lbc_idx = idx;
526 return lbsp;
527}
528
529
530/*
531 * xfs_trans_free_busy
532 * Free all of the busy lists from a transaction
533 */
534void
535xfs_trans_free_busy(xfs_trans_t *tp)
536{
537 xfs_log_busy_chunk_t *lbcp;
538 xfs_log_busy_chunk_t *lbcq;
539
540 lbcp = tp->t_busy.lbc_next;
541 while (lbcp != NULL) {
542 lbcq = lbcp->lbc_next;
543 kmem_free(lbcp);
544 lbcp = lbcq;
545 }
546
547 XFS_LBC_INIT(&tp->t_busy);
548 tp->t_busy.lbc_unused = 0;
549}
diff --git a/fs/xfs/xfs_trans_priv.h b/fs/xfs/xfs_trans_priv.h
index 73e2ad397432..c6e4f2c8de6e 100644
--- a/fs/xfs/xfs_trans_priv.h
+++ b/fs/xfs/xfs_trans_priv.h
@@ -35,13 +35,14 @@ struct xfs_log_item_desc *xfs_trans_find_item(struct xfs_trans *,
35struct xfs_log_item_desc *xfs_trans_first_item(struct xfs_trans *); 35struct xfs_log_item_desc *xfs_trans_first_item(struct xfs_trans *);
36struct xfs_log_item_desc *xfs_trans_next_item(struct xfs_trans *, 36struct xfs_log_item_desc *xfs_trans_next_item(struct xfs_trans *,
37 struct xfs_log_item_desc *); 37 struct xfs_log_item_desc *);
38void xfs_trans_free_items(struct xfs_trans *, int); 38
39void xfs_trans_unlock_items(struct xfs_trans *, 39void xfs_trans_unlock_items(struct xfs_trans *tp, xfs_lsn_t commit_lsn);
40 xfs_lsn_t); 40void xfs_trans_free_items(struct xfs_trans *tp, xfs_lsn_t commit_lsn,
41void xfs_trans_free_busy(xfs_trans_t *tp); 41 int flags);
42xfs_log_busy_slot_t *xfs_trans_add_busy(xfs_trans_t *tp, 42
43 xfs_agnumber_t ag, 43void xfs_trans_item_committed(struct xfs_log_item *lip,
44 xfs_extlen_t idx); 44 xfs_lsn_t commit_lsn, int aborted);
45void xfs_trans_unreserve_and_mod_sb(struct xfs_trans *tp);
45 46
46/* 47/*
47 * AIL traversal cursor. 48 * AIL traversal cursor.
diff --git a/fs/xfs/xfs_types.h b/fs/xfs/xfs_types.h
index b09904555d07..320775295e32 100644
--- a/fs/xfs/xfs_types.h
+++ b/fs/xfs/xfs_types.h
@@ -75,6 +75,8 @@ typedef __uint32_t xfs_dahash_t; /* dir/attr hash value */
75 75
76typedef __uint16_t xfs_prid_t; /* prid_t truncated to 16bits in XFS */ 76typedef __uint16_t xfs_prid_t; /* prid_t truncated to 16bits in XFS */
77 77
78typedef __uint32_t xlog_tid_t; /* transaction ID type */
79
78/* 80/*
79 * These types are 64 bits on disk but are either 32 or 64 bits in memory. 81 * These types are 64 bits on disk but are either 32 or 64 bits in memory.
80 * Disk based types: 82 * Disk based types:
generated by cgit v1.2.2 (git 2.25.0) at 2024-12-16 06:04:40 -0500