diff options
Diffstat (limited to 'fs/xfs/xfs_trans.c')
| -rw-r--r-- | fs/xfs/xfs_trans.c | 1135 |
1 files changed, 367 insertions, 768 deletions
diff --git a/fs/xfs/xfs_trans.c b/fs/xfs/xfs_trans.c index 237badcbac3..d03932564cc 100644 --- a/fs/xfs/xfs_trans.c +++ b/fs/xfs/xfs_trans.c @@ -1,5 +1,6 @@ /* * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc. + * Copyright (C) 2010 Red Hat, Inc. * All Rights Reserved. * * This program is free software; you can redistribute it and/or @@ -17,176 +18,24 @@ */ #include "xfs.h" #include "xfs_fs.h" -#include "xfs_types.h" -#include "xfs_bit.h" -#include "xfs_log.h" -#include "xfs_inum.h" -#include "xfs_trans.h" +#include "xfs_shared.h" +#include "xfs_format.h" +#include "xfs_log_format.h" +#include "xfs_trans_resv.h" #include "xfs_sb.h" #include "xfs_ag.h" -#include "xfs_dir2.h" -#include "xfs_dmapi.h" #include "xfs_mount.h" -#include "xfs_error.h" -#include "xfs_da_btree.h" -#include "xfs_bmap_btree.h" -#include "xfs_alloc_btree.h" -#include "xfs_ialloc_btree.h" -#include "xfs_dir2_sf.h" -#include "xfs_attr_sf.h" -#include "xfs_dinode.h" #include "xfs_inode.h" -#include "xfs_btree.h" -#include "xfs_ialloc.h" -#include "xfs_alloc.h" -#include "xfs_bmap.h" +#include "xfs_extent_busy.h" #include "xfs_quota.h" +#include "xfs_trans.h" #include "xfs_trans_priv.h" -#include "xfs_trans_space.h" -#include "xfs_inode_item.h" - - -STATIC void xfs_trans_apply_sb_deltas(xfs_trans_t *); -STATIC uint xfs_trans_count_vecs(xfs_trans_t *); -STATIC void xfs_trans_fill_vecs(xfs_trans_t *, xfs_log_iovec_t *); -STATIC void xfs_trans_uncommit(xfs_trans_t *, uint); -STATIC void xfs_trans_committed(xfs_trans_t *, int); -STATIC void xfs_trans_chunk_committed(xfs_log_item_chunk_t *, xfs_lsn_t, int); -STATIC void xfs_trans_free(xfs_trans_t *); +#include "xfs_log.h" +#include "xfs_trace.h" +#include "xfs_error.h" kmem_zone_t *xfs_trans_zone; - - -/* - * Reservation functions here avoid a huge stack in xfs_trans_init - * due to register overflow from temporaries in the calculations. - */ - -STATIC uint -xfs_calc_write_reservation(xfs_mount_t *mp) -{ - return XFS_CALC_WRITE_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp); -} - -STATIC uint -xfs_calc_itruncate_reservation(xfs_mount_t *mp) -{ - return XFS_CALC_ITRUNCATE_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp); -} - -STATIC uint -xfs_calc_rename_reservation(xfs_mount_t *mp) -{ - return XFS_CALC_RENAME_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp); -} - -STATIC uint -xfs_calc_link_reservation(xfs_mount_t *mp) -{ - return XFS_CALC_LINK_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp); -} - -STATIC uint -xfs_calc_remove_reservation(xfs_mount_t *mp) -{ - return XFS_CALC_REMOVE_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp); -} - -STATIC uint -xfs_calc_symlink_reservation(xfs_mount_t *mp) -{ - return XFS_CALC_SYMLINK_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp); -} - -STATIC uint -xfs_calc_create_reservation(xfs_mount_t *mp) -{ - return XFS_CALC_CREATE_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp); -} - -STATIC uint -xfs_calc_mkdir_reservation(xfs_mount_t *mp) -{ - return XFS_CALC_MKDIR_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp); -} - -STATIC uint -xfs_calc_ifree_reservation(xfs_mount_t *mp) -{ - return XFS_CALC_IFREE_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp); -} - -STATIC uint -xfs_calc_ichange_reservation(xfs_mount_t *mp) -{ - return XFS_CALC_ICHANGE_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp); -} - -STATIC uint -xfs_calc_growdata_reservation(xfs_mount_t *mp) -{ - return XFS_CALC_GROWDATA_LOG_RES(mp); -} - -STATIC uint -xfs_calc_growrtalloc_reservation(xfs_mount_t *mp) -{ - return XFS_CALC_GROWRTALLOC_LOG_RES(mp); -} - -STATIC uint -xfs_calc_growrtzero_reservation(xfs_mount_t *mp) -{ - return XFS_CALC_GROWRTZERO_LOG_RES(mp); -} - -STATIC uint -xfs_calc_growrtfree_reservation(xfs_mount_t *mp) -{ - return XFS_CALC_GROWRTFREE_LOG_RES(mp); -} - -STATIC uint -xfs_calc_swrite_reservation(xfs_mount_t *mp) -{ - return XFS_CALC_SWRITE_LOG_RES(mp); -} - -STATIC uint -xfs_calc_writeid_reservation(xfs_mount_t *mp) -{ - return XFS_CALC_WRITEID_LOG_RES(mp); -} - -STATIC uint -xfs_calc_addafork_reservation(xfs_mount_t *mp) -{ - return XFS_CALC_ADDAFORK_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp); -} - -STATIC uint -xfs_calc_attrinval_reservation(xfs_mount_t *mp) -{ - return XFS_CALC_ATTRINVAL_LOG_RES(mp); -} - -STATIC uint -xfs_calc_attrset_reservation(xfs_mount_t *mp) -{ - return XFS_CALC_ATTRSET_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp); -} - -STATIC uint -xfs_calc_attrrm_reservation(xfs_mount_t *mp) -{ - return XFS_CALC_ATTRRM_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp); -} - -STATIC uint -xfs_calc_clear_agi_bucket_reservation(xfs_mount_t *mp) -{ - return XFS_CALC_CLEAR_AGI_BUCKET_LOG_RES(mp); -} +kmem_zone_t *xfs_log_item_desc_zone; /* * Initialize the precomputed transaction reservation values @@ -194,32 +43,9 @@ xfs_calc_clear_agi_bucket_reservation(xfs_mount_t *mp) */ void xfs_trans_init( - xfs_mount_t *mp) + struct xfs_mount *mp) { - xfs_trans_reservations_t *resp; - - resp = &(mp->m_reservations); - resp->tr_write = xfs_calc_write_reservation(mp); - resp->tr_itruncate = xfs_calc_itruncate_reservation(mp); - resp->tr_rename = xfs_calc_rename_reservation(mp); - resp->tr_link = xfs_calc_link_reservation(mp); - resp->tr_remove = xfs_calc_remove_reservation(mp); - resp->tr_symlink = xfs_calc_symlink_reservation(mp); - resp->tr_create = xfs_calc_create_reservation(mp); - resp->tr_mkdir = xfs_calc_mkdir_reservation(mp); - resp->tr_ifree = xfs_calc_ifree_reservation(mp); - resp->tr_ichange = xfs_calc_ichange_reservation(mp); - resp->tr_growdata = xfs_calc_growdata_reservation(mp); - resp->tr_swrite = xfs_calc_swrite_reservation(mp); - resp->tr_writeid = xfs_calc_writeid_reservation(mp); - resp->tr_addafork = xfs_calc_addafork_reservation(mp); - resp->tr_attrinval = xfs_calc_attrinval_reservation(mp); - resp->tr_attrset = xfs_calc_attrset_reservation(mp); - resp->tr_attrrm = xfs_calc_attrrm_reservation(mp); - resp->tr_clearagi = xfs_calc_clear_agi_bucket_reservation(mp); - resp->tr_growrtalloc = xfs_calc_growrtalloc_reservation(mp); - resp->tr_growrtzero = xfs_calc_growrtzero_reservation(mp); - resp->tr_growrtfree = xfs_calc_growrtfree_reservation(mp); + xfs_trans_resv_calc(mp, M_RES(mp)); } /* @@ -235,32 +61,53 @@ xfs_trans_alloc( xfs_mount_t *mp, uint type) { - xfs_wait_for_freeze(mp, SB_FREEZE_TRANS); - return _xfs_trans_alloc(mp, type, KM_SLEEP); + xfs_trans_t *tp; + + sb_start_intwrite(mp->m_super); + tp = _xfs_trans_alloc(mp, type, KM_SLEEP); + tp->t_flags |= XFS_TRANS_FREEZE_PROT; + return tp; } xfs_trans_t * _xfs_trans_alloc( xfs_mount_t *mp, uint type, - uint memflags) + xfs_km_flags_t memflags) { xfs_trans_t *tp; + WARN_ON(mp->m_super->s_writers.frozen == SB_FREEZE_COMPLETE); atomic_inc(&mp->m_active_trans); tp = kmem_zone_zalloc(xfs_trans_zone, memflags); - tp->t_magic = XFS_TRANS_MAGIC; + tp->t_magic = XFS_TRANS_HEADER_MAGIC; tp->t_type = type; tp->t_mountp = mp; - tp->t_items_free = XFS_LIC_NUM_SLOTS; - tp->t_busy_free = XFS_LBC_NUM_SLOTS; - xfs_lic_init(&(tp->t_items)); - XFS_LBC_INIT(&(tp->t_busy)); + INIT_LIST_HEAD(&tp->t_items); + INIT_LIST_HEAD(&tp->t_busy); return tp; } /* + * Free the transaction structure. If there is more clean up + * to do when the structure is freed, add it here. + */ +STATIC void +xfs_trans_free( + struct xfs_trans *tp) +{ + xfs_extent_busy_sort(&tp->t_busy); + xfs_extent_busy_clear(tp->t_mountp, &tp->t_busy, false); + + atomic_dec(&tp->t_mountp->m_active_trans); + if (tp->t_flags & XFS_TRANS_FREEZE_PROT) + sb_end_intwrite(tp->t_mountp->m_super); + xfs_trans_free_dqinfo(tp); + kmem_zone_free(xfs_trans_zone, tp); +} + +/* * This is called to create a new transaction which will share the * permanent log reservation of the given transaction. The remaining * unused block and rt extent reservations are also inherited. This @@ -279,18 +126,20 @@ xfs_trans_dup( /* * Initialize the new transaction structure. */ - ntp->t_magic = XFS_TRANS_MAGIC; + ntp->t_magic = XFS_TRANS_HEADER_MAGIC; ntp->t_type = tp->t_type; ntp->t_mountp = tp->t_mountp; - ntp->t_items_free = XFS_LIC_NUM_SLOTS; - ntp->t_busy_free = XFS_LBC_NUM_SLOTS; - xfs_lic_init(&(ntp->t_items)); - XFS_LBC_INIT(&(ntp->t_busy)); + INIT_LIST_HEAD(&ntp->t_items); + INIT_LIST_HEAD(&ntp->t_busy); ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES); ASSERT(tp->t_ticket != NULL); - ntp->t_flags = XFS_TRANS_PERM_LOG_RES | (tp->t_flags & XFS_TRANS_RESERVE); + ntp->t_flags = XFS_TRANS_PERM_LOG_RES | + (tp->t_flags & XFS_TRANS_RESERVE) | + (tp->t_flags & XFS_TRANS_FREEZE_PROT); + /* We gave our writer reference to the new transaction */ + tp->t_flags &= ~XFS_TRANS_FREEZE_PROT; ntp->t_ticket = xfs_log_ticket_get(tp->t_ticket); ntp->t_blk_res = tp->t_blk_res - tp->t_blk_res_used; tp->t_blk_res = tp->t_blk_res_used; @@ -320,14 +169,11 @@ xfs_trans_dup( */ int xfs_trans_reserve( - xfs_trans_t *tp, - uint blocks, - uint logspace, - uint rtextents, - uint flags, - uint logcount) + struct xfs_trans *tp, + struct xfs_trans_res *resp, + uint blocks, + uint rtextents) { - int log_flags; int error = 0; int rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0; @@ -340,7 +186,7 @@ xfs_trans_reserve( * fail if the count would go below zero. */ if (blocks > 0) { - error = xfs_mod_incore_sb(tp->t_mountp, XFS_SBS_FDBLOCKS, + error = xfs_icsb_modify_counters(tp->t_mountp, XFS_SBS_FDBLOCKS, -((int64_t)blocks), rsvd); if (error != 0) { current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS); @@ -352,27 +198,38 @@ xfs_trans_reserve( /* * Reserve the log space needed for this transaction. */ - if (logspace > 0) { - ASSERT((tp->t_log_res == 0) || (tp->t_log_res == logspace)); - ASSERT((tp->t_log_count == 0) || - (tp->t_log_count == logcount)); - if (flags & XFS_TRANS_PERM_LOG_RES) { - log_flags = XFS_LOG_PERM_RESERV; + if (resp->tr_logres > 0) { + bool permanent = false; + + ASSERT(tp->t_log_res == 0 || + tp->t_log_res == resp->tr_logres); + ASSERT(tp->t_log_count == 0 || + tp->t_log_count == resp->tr_logcount); + + if (resp->tr_logflags & XFS_TRANS_PERM_LOG_RES) { tp->t_flags |= XFS_TRANS_PERM_LOG_RES; + permanent = true; } else { ASSERT(tp->t_ticket == NULL); ASSERT(!(tp->t_flags & XFS_TRANS_PERM_LOG_RES)); - log_flags = 0; } - error = xfs_log_reserve(tp->t_mountp, logspace, logcount, - &tp->t_ticket, - XFS_TRANSACTION, log_flags, tp->t_type); - if (error) { - goto undo_blocks; + if (tp->t_ticket != NULL) { + ASSERT(resp->tr_logflags & XFS_TRANS_PERM_LOG_RES); + error = xfs_log_regrant(tp->t_mountp, tp->t_ticket); + } else { + error = xfs_log_reserve(tp->t_mountp, + resp->tr_logres, + resp->tr_logcount, + &tp->t_ticket, XFS_TRANSACTION, + permanent, tp->t_type); } - tp->t_log_res = logspace; - tp->t_log_count = logcount; + + if (error) + goto undo_blocks; + + tp->t_log_res = resp->tr_logres; + tp->t_log_count = resp->tr_logcount; } /* @@ -397,8 +254,10 @@ xfs_trans_reserve( * reservations which have already been performed. */ undo_log: - if (logspace > 0) { - if (flags & XFS_TRANS_PERM_LOG_RES) { + if (resp->tr_logres > 0) { + int log_flags; + + if (resp->tr_logflags & XFS_TRANS_PERM_LOG_RES) { log_flags = XFS_LOG_REL_PERM_RESERV; } else { log_flags = 0; @@ -411,7 +270,7 @@ undo_log: undo_blocks: if (blocks > 0) { - (void) xfs_mod_incore_sb(tp->t_mountp, XFS_SBS_FDBLOCKS, + xfs_icsb_modify_counters(tp->t_mountp, XFS_SBS_FDBLOCKS, (int64_t)blocks, rsvd); tp->t_blk_res = 0; } @@ -421,7 +280,6 @@ undo_blocks: return error; } - /* * Record the indicated change to the given field for application * to the file system's superblock when the transaction commits. @@ -650,11 +508,11 @@ xfs_trans_apply_sb_deltas( * XFS_TRANS_SB_DIRTY will not be set when the transaction is updated but we * still need to update the incore superblock with the changes. */ -STATIC void +void xfs_trans_unreserve_and_mod_sb( xfs_trans_t *tp) { - xfs_mod_sb_t msb[14]; /* If you add cases, add entries */ + xfs_mod_sb_t msb[9]; /* If you add cases, add entries */ xfs_mod_sb_t *msbp; xfs_mount_t *mp = tp->t_mountp; /* REFERENCED */ @@ -662,55 +520,61 @@ xfs_trans_unreserve_and_mod_sb( int rsvd; int64_t blkdelta = 0; int64_t rtxdelta = 0; + int64_t idelta = 0; + int64_t ifreedelta = 0; msbp = msb; rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0; - /* calculate free blocks delta */ + /* calculate deltas */ if (tp->t_blk_res > 0) blkdelta = tp->t_blk_res; - if ((tp->t_fdblocks_delta != 0) && (xfs_sb_version_haslazysbcount(&mp->m_sb) || (tp->t_flags & XFS_TRANS_SB_DIRTY))) blkdelta += tp->t_fdblocks_delta; - if (blkdelta != 0) { - msbp->msb_field = XFS_SBS_FDBLOCKS; - msbp->msb_delta = blkdelta; - msbp++; - } - - /* calculate free realtime extents delta */ if (tp->t_rtx_res > 0) rtxdelta = tp->t_rtx_res; - if ((tp->t_frextents_delta != 0) && (tp->t_flags & XFS_TRANS_SB_DIRTY)) rtxdelta += tp->t_frextents_delta; + if (xfs_sb_version_haslazysbcount(&mp->m_sb) || + (tp->t_flags & XFS_TRANS_SB_DIRTY)) { + idelta = tp->t_icount_delta; + ifreedelta = tp->t_ifree_delta; + } + + /* apply the per-cpu counters */ + if (blkdelta) { + error = xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS, + blkdelta, rsvd); + if (error) + goto out; + } + + if (idelta) { + error = xfs_icsb_modify_counters(mp, XFS_SBS_ICOUNT, + idelta, rsvd); + if (error) + goto out_undo_fdblocks; + } + + if (ifreedelta) { + error = xfs_icsb_modify_counters(mp, XFS_SBS_IFREE, + ifreedelta, rsvd); + if (error) + goto out_undo_icount; + } + + /* apply remaining deltas */ if (rtxdelta != 0) { msbp->msb_field = XFS_SBS_FREXTENTS; msbp->msb_delta = rtxdelta; msbp++; } - /* apply remaining deltas */ - - if (xfs_sb_version_haslazysbcount(&mp->m_sb) || - (tp->t_flags & XFS_TRANS_SB_DIRTY)) { - if (tp->t_icount_delta != 0) { - msbp->msb_field = XFS_SBS_ICOUNT; - msbp->msb_delta = tp->t_icount_delta; - msbp++; - } - if (tp->t_ifree_delta != 0) { - msbp->msb_field = XFS_SBS_IFREE; - msbp->msb_delta = tp->t_ifree_delta; - msbp++; - } - } - if (tp->t_flags & XFS_TRANS_SB_DIRTY) { if (tp->t_dblocks_delta != 0) { msbp->msb_field = XFS_SBS_DBLOCKS; @@ -760,15 +624,215 @@ xfs_trans_unreserve_and_mod_sb( if (msbp > msb) { error = xfs_mod_incore_sb_batch(tp->t_mountp, msb, (uint)(msbp - msb), rsvd); - ASSERT(error == 0); + if (error) + goto out_undo_ifreecount; + } + + return; + +out_undo_ifreecount: + if (ifreedelta) + xfs_icsb_modify_counters(mp, XFS_SBS_IFREE, -ifreedelta, rsvd); +out_undo_icount: + if (idelta) + xfs_icsb_modify_counters(mp, XFS_SBS_ICOUNT, -idelta, rsvd); +out_undo_fdblocks: + if (blkdelta) + xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS, -blkdelta, rsvd); +out: + ASSERT(error == 0); + return; +} + +/* + * Add the given log item to the transaction's list of log items. + * + * The log item will now point to its new descriptor with its li_desc field. + */ +void +xfs_trans_add_item( + struct xfs_trans *tp, + struct xfs_log_item *lip) +{ + struct xfs_log_item_desc *lidp; + + ASSERT(lip->li_mountp == tp->t_mountp); + ASSERT(lip->li_ailp == tp->t_mountp->m_ail); + + lidp = kmem_zone_zalloc(xfs_log_item_desc_zone, KM_SLEEP | KM_NOFS); + + lidp->lid_item = lip; + lidp->lid_flags = 0; + list_add_tail(&lidp->lid_trans, &tp->t_items); + + lip->li_desc = lidp; +} + +STATIC void +xfs_trans_free_item_desc( + struct xfs_log_item_desc *lidp) +{ + list_del_init(&lidp->lid_trans); + kmem_zone_free(xfs_log_item_desc_zone, lidp); +} + +/* + * Unlink and free the given descriptor. + */ +void +xfs_trans_del_item( + struct xfs_log_item *lip) +{ + xfs_trans_free_item_desc(lip->li_desc); + lip->li_desc = NULL; +} + +/* + * Unlock all of the items of a transaction and free all the descriptors + * of that transaction. + */ +void +xfs_trans_free_items( + struct xfs_trans *tp, + xfs_lsn_t commit_lsn, + int flags) +{ + struct xfs_log_item_desc *lidp, *next; + + list_for_each_entry_safe(lidp, next, &tp->t_items, lid_trans) { + struct xfs_log_item *lip = lidp->lid_item; + + lip->li_desc = NULL; + + if (commit_lsn != NULLCOMMITLSN) + lip->li_ops->iop_committing(lip, commit_lsn); + if (flags & XFS_TRANS_ABORT) + lip->li_flags |= XFS_LI_ABORTED; + lip->li_ops->iop_unlock(lip); + + xfs_trans_free_item_desc(lidp); } } +static inline void +xfs_log_item_batch_insert( + struct xfs_ail *ailp, + struct xfs_ail_cursor *cur, + struct xfs_log_item **log_items, + int nr_items, + xfs_lsn_t commit_lsn) +{ + int i; + + spin_lock(&ailp->xa_lock); + /* xfs_trans_ail_update_bulk drops ailp->xa_lock */ + xfs_trans_ail_update_bulk(ailp, cur, log_items, nr_items, commit_lsn); + + for (i = 0; i < nr_items; i++) { + struct xfs_log_item *lip = log_items[i]; + + lip->li_ops->iop_unpin(lip, 0); + } +} /* - * xfs_trans_commit + * Bulk operation version of xfs_trans_committed that takes a log vector of + * items to insert into the AIL. This uses bulk AIL insertion techniques to + * minimise lock traffic. + * + * If we are called with the aborted flag set, it is because a log write during + * a CIL checkpoint commit has failed. In this case, all the items in the + * checkpoint have already gone through iop_commited and iop_unlock, which + * means that checkpoint commit abort handling is treated exactly the same + * as an iclog write error even though we haven't started any IO yet. Hence in + * this case all we need to do is iop_committed processing, followed by an + * iop_unpin(aborted) call. * - * Commit the given transaction to the log a/synchronously. + * The AIL cursor is used to optimise the insert process. If commit_lsn is not + * at the end of the AIL, the insert cursor avoids the need to walk + * the AIL to find the insertion point on every xfs_log_item_batch_insert() + * call. This saves a lot of needless list walking and is a net win, even + * though it slightly increases that amount of AIL lock traffic to set it up + * and tear it down. + */ +void +xfs_trans_committed_bulk( + struct xfs_ail *ailp, + struct xfs_log_vec *log_vector, + xfs_lsn_t commit_lsn, + int aborted) +{ +#define LOG_ITEM_BATCH_SIZE 32 + struct xfs_log_item *log_items[LOG_ITEM_BATCH_SIZE]; + struct xfs_log_vec *lv; + struct xfs_ail_cursor cur; + int i = 0; + + spin_lock(&ailp->xa_lock); + xfs_trans_ail_cursor_last(ailp, &cur, commit_lsn); + spin_unlock(&ailp->xa_lock); + + /* unpin all the log items */ + for (lv = log_vector; lv; lv = lv->lv_next ) { + struct xfs_log_item *lip = lv->lv_item; + xfs_lsn_t item_lsn; + + if (aborted) + lip->li_flags |= XFS_LI_ABORTED; + item_lsn = lip->li_ops->iop_committed(lip, commit_lsn); + + /* item_lsn of -1 means the item needs no further processing */ + if (XFS_LSN_CMP(item_lsn, (xfs_lsn_t)-1) == 0) + continue; + + /* + * if we are aborting the operation, no point in inserting the + * object into the AIL as we are in a shutdown situation. + */ + if (aborted) { + ASSERT(XFS_FORCED_SHUTDOWN(ailp->xa_mount)); + lip->li_ops->iop_unpin(lip, 1); + continue; + } + + if (item_lsn != commit_lsn) { + + /* + * Not a bulk update option due to unusual item_lsn. + * Push into AIL immediately, rechecking the lsn once + * we have the ail lock. Then unpin the item. This does + * not affect the AIL cursor the bulk insert path is + * using. + */ + spin_lock(&ailp->xa_lock); + if (XFS_LSN_CMP(item_lsn, lip->li_lsn) > 0) + xfs_trans_ail_update(ailp, lip, item_lsn); + else + spin_unlock(&ailp->xa_lock); + lip->li_ops->iop_unpin(lip, 0); + continue; + } + + /* Item is a candidate for bulk AIL insert. */ + log_items[i++] = lv->lv_item; + if (i >= LOG_ITEM_BATCH_SIZE) { + xfs_log_item_batch_insert(ailp, &cur, log_items, + LOG_ITEM_BATCH_SIZE, commit_lsn); + i = 0; + } + } + + /* make sure we insert the remainder! */ + if (i) + xfs_log_item_batch_insert(ailp, &cur, log_items, i, commit_lsn); + + spin_lock(&ailp->xa_lock); + xfs_trans_ail_cursor_done(&cur); + spin_unlock(&ailp->xa_lock); +} + +/* + * Commit the given transaction to the log. * * XFS disk error handling mechanism is not based on a typical * transaction abort mechanism. Logically after the filesystem @@ -779,27 +843,16 @@ xfs_trans_unreserve_and_mod_sb( * have already been unlocked as if the commit had succeeded. * Do not reference the transaction structure after this call. */ - /*ARGSUSED*/ int -_xfs_trans_commit( - xfs_trans_t *tp, - uint flags, - int *log_flushed) +xfs_trans_commit( + struct xfs_trans *tp, + uint flags) { - xfs_log_iovec_t *log_vector; - int nvec; - xfs_mount_t *mp; - xfs_lsn_t commit_lsn; - /* REFERENCED */ - int error; - int log_flags; - int sync; -#define XFS_TRANS_LOGVEC_COUNT 16 - xfs_log_iovec_t log_vector_fast[XFS_TRANS_LOGVEC_COUNT]; - void *commit_iclog; - int shutdown; - - commit_lsn = -1; + struct xfs_mount *mp = tp->t_mountp; + xfs_lsn_t commit_lsn = -1; + int error = 0; + int log_flags = 0; + int sync = tp->t_flags & XFS_TRANS_SYNC; /* * Determine whether this commit is releasing a permanent @@ -808,10 +861,7 @@ _xfs_trans_commit( if (flags & XFS_TRANS_RELEASE_LOG_RES) { ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES); log_flags = XFS_LOG_REL_PERM_RESERV; - } else { - log_flags = 0; } - mp = tp->t_mountp; /* * If there is nothing to be logged by the transaction, @@ -820,30 +870,14 @@ _xfs_trans_commit( * Also make sure to return any reserved blocks to * the free pool. */ -shut_us_down: - shutdown = XFS_FORCED_SHUTDOWN(mp) ? EIO : 0; - if (!(tp->t_flags & XFS_TRANS_DIRTY) || shutdown) { - xfs_trans_unreserve_and_mod_sb(tp); - /* - * It is indeed possible for the transaction to be - * not dirty but the dqinfo portion to be. All that - * means is that we have some (non-persistent) quota - * reservations that need to be unreserved. - */ - xfs_trans_unreserve_and_mod_dquots(tp); - if (tp->t_ticket) { - commit_lsn = xfs_log_done(mp, tp->t_ticket, - NULL, log_flags); - if (commit_lsn == -1 && !shutdown) - shutdown = XFS_ERROR(EIO); - } - current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS); - xfs_trans_free_items(tp, shutdown? XFS_TRANS_ABORT : 0); - xfs_trans_free_busy(tp); - xfs_trans_free(tp); - XFS_STATS_INC(xs_trans_empty); - return (shutdown); + if (!(tp->t_flags & XFS_TRANS_DIRTY)) + goto out_unreserve; + + if (XFS_FORCED_SHUTDOWN(mp)) { + error = XFS_ERROR(EIO); + goto out_unreserve; } + ASSERT(tp->t_ticket != NULL); /* @@ -853,278 +887,46 @@ shut_us_down: xfs_trans_apply_sb_deltas(tp); xfs_trans_apply_dquot_deltas(tp); - /* - * Ask each log item how many log_vector entries it will - * need so we can figure out how many to allocate. - * Try to avoid the kmem_alloc() call in the common case - * by using a vector from the stack when it fits. - */ - nvec = xfs_trans_count_vecs(tp); - if (nvec == 0) { - xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR); - goto shut_us_down; - } else if (nvec <= XFS_TRANS_LOGVEC_COUNT) { - log_vector = log_vector_fast; - } else { - log_vector = (xfs_log_iovec_t *)kmem_alloc(nvec * - sizeof(xfs_log_iovec_t), - KM_SLEEP); - } - - /* - * Fill in the log_vector and pin the logged items, and - * then write the transaction to the log. - */ - xfs_trans_fill_vecs(tp, log_vector); - - error = xfs_log_write(mp, log_vector, nvec, tp->t_ticket, &(tp->t_lsn)); - - /* - * The transaction is committed incore here, and can go out to disk - * at any time after this call. However, all the items associated - * with the transaction are still locked and pinned in memory. - */ - commit_lsn = xfs_log_done(mp, tp->t_ticket, &commit_iclog, log_flags); - - tp->t_commit_lsn = commit_lsn; - if (nvec > XFS_TRANS_LOGVEC_COUNT) { - kmem_free(log_vector); - } - - /* - * If we got a log write error. Unpin the logitems that we - * had pinned, clean up, free trans structure, and return error. - */ - if (error || commit_lsn == -1) { - current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS); - xfs_trans_uncommit(tp, flags|XFS_TRANS_ABORT); - return XFS_ERROR(EIO); - } - - /* - * Once the transaction has committed, unused - * reservations need to be released and changes to - * the superblock need to be reflected in the in-core - * version. Do that now. - */ - xfs_trans_unreserve_and_mod_sb(tp); - - sync = tp->t_flags & XFS_TRANS_SYNC; + xfs_log_commit_cil(mp, tp, &commit_lsn, flags); - /* - * Tell the LM to call the transaction completion routine - * when the log write with LSN commit_lsn completes (e.g. - * when the transaction commit really hits the on-disk log). - * After this call we cannot reference tp, because the call - * can happen at any time and the call will free the transaction - * structure pointed to by tp. The only case where we call - * the completion routine (xfs_trans_committed) directly is - * if the log is turned off on a debug kernel or we're - * running in simulation mode (the log is explicitly turned - * off). - */ - tp->t_logcb.cb_func = (void(*)(void*, int))xfs_trans_committed; - tp->t_logcb.cb_arg = tp; - - /* - * We need to pass the iclog buffer which was used for the - * transaction commit record into this function, and attach - * the callback to it. The callback must be attached before - * the items are unlocked to avoid racing with other threads - * waiting for an item to unlock. - */ - shutdown = xfs_log_notify(mp, commit_iclog, &(tp->t_logcb)); - - /* - * Mark this thread as no longer being in a transaction - */ current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS); - - /* - * Once all the items of the transaction have been copied - * to the in core log and the callback is attached, the - * items can be unlocked. - * - * This will free descriptors pointing to items which were - * not logged since there is nothing more to do with them. - * For items which were logged, we will keep pointers to them - * so they can be unpinned after the transaction commits to disk. - * This will also stamp each modified meta-data item with - * the commit lsn of this transaction for dependency tracking - * purposes. - */ - xfs_trans_unlock_items(tp, commit_lsn); - - /* - * If we detected a log error earlier, finish committing - * the transaction now (unpin log items, etc). - * - * Order is critical here, to avoid using the transaction - * pointer after its been freed (by xfs_trans_committed - * either here now, or as a callback). We cannot do this - * step inside xfs_log_notify as was done earlier because - * of this issue. - */ - if (shutdown) - xfs_trans_committed(tp, XFS_LI_ABORTED); - - /* - * Now that the xfs_trans_committed callback has been attached, - * and the items are released we can finally allow the iclog to - * go to disk. - */ - error = xfs_log_release_iclog(mp, commit_iclog); + xfs_trans_free(tp); /* * If the transaction needs to be synchronous, then force the * log out now and wait for it. */ if (sync) { - if (!error) { - error = _xfs_log_force(mp, commit_lsn, - XFS_LOG_FORCE | XFS_LOG_SYNC, - log_flushed); - } + error = _xfs_log_force_lsn(mp, commit_lsn, XFS_LOG_SYNC, NULL); XFS_STATS_INC(xs_trans_sync); } else { XFS_STATS_INC(xs_trans_async); } - return (error); -} - - -/* - * Total up the number of log iovecs needed to commit this - * transaction. The transaction itself needs one for the - * transaction header. Ask each dirty item in turn how many - * it needs to get the total. - */ -STATIC uint -xfs_trans_count_vecs( - xfs_trans_t *tp) -{ - int nvecs; - xfs_log_item_desc_t *lidp; - - nvecs = 1; - lidp = xfs_trans_first_item(tp); - ASSERT(lidp != NULL); - - /* In the non-debug case we need to start bailing out if we - * didn't find a log_item here, return zero and let trans_commit - * deal with it. - */ - if (lidp == NULL) - return 0; - - while (lidp != NULL) { - /* - * Skip items which aren't dirty in this transaction. - */ - if (!(lidp->lid_flags & XFS_LID_DIRTY)) { - lidp = xfs_trans_next_item(tp, lidp); - continue; - } - lidp->lid_size = IOP_SIZE(lidp->lid_item); - nvecs += lidp->lid_size; - lidp = xfs_trans_next_item(tp, lidp); - } - - return nvecs; -} - -/* - * Called from the trans_commit code when we notice that - * the filesystem is in the middle of a forced shutdown. - */ -STATIC void -xfs_trans_uncommit( - xfs_trans_t *tp, - uint flags) -{ - xfs_log_item_desc_t *lidp; - - for (lidp = xfs_trans_first_item(tp); - lidp != NULL; - lidp = xfs_trans_next_item(tp, lidp)) { - /* - * Unpin all but those that aren't dirty. - */ - if (lidp->lid_flags & XFS_LID_DIRTY) - IOP_UNPIN_REMOVE(lidp->lid_item, tp); - } + return error; +out_unreserve: xfs_trans_unreserve_and_mod_sb(tp); - xfs_trans_unreserve_and_mod_dquots(tp); - - xfs_trans_free_items(tp, flags); - xfs_trans_free_busy(tp); - xfs_trans_free(tp); -} - -/* - * Fill in the vector with pointers to data to be logged - * by this transaction. The transaction header takes - * the first vector, and then each dirty item takes the - * number of vectors it indicated it needed in xfs_trans_count_vecs(). - * - * As each item fills in the entries it needs, also pin the item - * so that it cannot be flushed out until the log write completes. - */ -STATIC void -xfs_trans_fill_vecs( - xfs_trans_t *tp, - xfs_log_iovec_t *log_vector) -{ - xfs_log_item_desc_t *lidp; - xfs_log_iovec_t *vecp; - uint nitems; /* - * Skip over the entry for the transaction header, we'll - * fill that in at the end. + * It is indeed possible for the transaction to be not dirty but + * the dqinfo portion to be. All that means is that we have some + * (non-persistent) quota reservations that need to be unreserved. */ - vecp = log_vector + 1; /* pointer arithmetic */ - - nitems = 0; - lidp = xfs_trans_first_item(tp); - ASSERT(lidp != NULL); - while (lidp != NULL) { - /* - * Skip items which aren't dirty in this transaction. - */ - if (!(lidp->lid_flags & XFS_LID_DIRTY)) { - lidp = xfs_trans_next_item(tp, lidp); - continue; - } - /* - * The item may be marked dirty but not log anything. - * This can be used to get called when a transaction - * is committed. - */ - if (lidp->lid_size) { - nitems++; - } - IOP_FORMAT(lidp->lid_item, vecp); - vecp += lidp->lid_size; /* pointer arithmetic */ - IOP_PIN(lidp->lid_item); - lidp = xfs_trans_next_item(tp, lidp); + xfs_trans_unreserve_and_mod_dquots(tp); + if (tp->t_ticket) { + commit_lsn = xfs_log_done(mp, tp->t_ticket, NULL, log_flags); + if (commit_lsn == -1 && !error) + error = XFS_ERROR(EIO); } + current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS); + xfs_trans_free_items(tp, NULLCOMMITLSN, error ? XFS_TRANS_ABORT : 0); + xfs_trans_free(tp); - /* - * Now that we've counted the number of items in this - * transaction, fill in the transaction header. - */ - tp->t_header.th_magic = XFS_TRANS_HEADER_MAGIC; - tp->t_header.th_type = tp->t_type; - tp->t_header.th_num_items = nitems; - log_vector->i_addr = (xfs_caddr_t)&tp->t_header; - log_vector->i_len = sizeof(xfs_trans_header_t); - XLOG_VEC_SET_TYPE(log_vector, XLOG_REG_TYPE_TRANSHDR); + XFS_STATS_INC(xs_trans_empty); + return error; } - /* * Unlock all of the transaction's items and free the transaction. * The transaction must not have modified any of its items, because @@ -1139,12 +941,6 @@ xfs_trans_cancel( int flags) { int log_flags; -#ifdef DEBUG - xfs_log_item_chunk_t *licp; - xfs_log_item_desc_t *lidp; - xfs_log_item_t *lip; - int i; -#endif xfs_mount_t *mp = tp->t_mountp; /* @@ -1163,21 +959,11 @@ xfs_trans_cancel( xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); } #ifdef DEBUG - if (!(flags & XFS_TRANS_ABORT)) { - licp = &(tp->t_items); - while (licp != NULL) { - lidp = licp->lic_descs; - for (i = 0; i < licp->lic_unused; i++, lidp++) { - if (xfs_lic_isfree(licp, i)) { - continue; - } - - lip = lidp->lid_item; - if (!XFS_FORCED_SHUTDOWN(mp)) - ASSERT(!(lip->li_type == XFS_LI_EFD)); - } - licp = licp->lic_next; - } + if (!(flags & XFS_TRANS_ABORT) && !XFS_FORCED_SHUTDOWN(mp)) { + struct xfs_log_item_desc *lidp; + + list_for_each_entry(lidp, &tp->t_items, lid_trans) + ASSERT(!(lidp->lid_item->li_type == XFS_LI_EFD)); } #endif xfs_trans_unreserve_and_mod_sb(tp); @@ -1196,25 +982,10 @@ xfs_trans_cancel( /* mark this thread as no longer being in a transaction */ current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS); - xfs_trans_free_items(tp, flags); - xfs_trans_free_busy(tp); + xfs_trans_free_items(tp, NULLCOMMITLSN, flags); xfs_trans_free(tp); } - -/* - * Free the transaction structure. If there is more clean up - * to do when the structure is freed, add it here. - */ -STATIC void -xfs_trans_free( - xfs_trans_t *tp) -{ - atomic_dec(&tp->t_mountp->m_active_trans); - xfs_trans_free_dqinfo(tp); - kmem_zone_free(xfs_trans_zone, tp); -} - /* * Roll from one trans in the sequence of PERMANENT transactions to * the next: permanent transactions are only flushed out when @@ -1228,7 +999,7 @@ xfs_trans_roll( struct xfs_inode *dp) { struct xfs_trans *trans; - unsigned int logres, count; + struct xfs_trans_res tres; int error; /* @@ -1240,8 +1011,8 @@ xfs_trans_roll( /* * Copy the critical parameters from one trans to the next. */ - logres = trans->t_log_res; - count = trans->t_log_count; + tres.tr_logres = trans->t_log_res; + tres.tr_logcount = trans->t_log_count; *tpp = xfs_trans_dup(trans); /* @@ -1272,186 +1043,14 @@ xfs_trans_roll( * across this call, or that anything that is locked be logged in * the prior and the next transactions. */ - error = xfs_trans_reserve(trans, 0, logres, 0, - XFS_TRANS_PERM_LOG_RES, count); + tres.tr_logflags = XFS_TRANS_PERM_LOG_RES; + error = xfs_trans_reserve(trans, &tres, 0, 0); /* * Ensure that the inode is in the new transaction and locked. */ if (error) return error; - xfs_trans_ijoin(trans, dp, XFS_ILOCK_EXCL); - xfs_trans_ihold(trans, dp); + xfs_trans_ijoin(trans, dp, 0); return 0; } - -/* - * THIS SHOULD BE REWRITTEN TO USE xfs_trans_next_item(). - * - * This is typically called by the LM when a transaction has been fully - * committed to disk. It needs to unpin the items which have - * been logged by the transaction and update their positions - * in the AIL if necessary. - * This also gets called when the transactions didn't get written out - * because of an I/O error. Abortflag & XFS_LI_ABORTED is set then. - * - * Call xfs_trans_chunk_committed() to process the items in - * each chunk. - */ -STATIC void -xfs_trans_committed( - xfs_trans_t *tp, - int abortflag) -{ - xfs_log_item_chunk_t *licp; - xfs_log_item_chunk_t *next_licp; - xfs_log_busy_chunk_t *lbcp; - xfs_log_busy_slot_t *lbsp; - int i; - - /* - * Call the transaction's completion callback if there - * is one. - */ - if (tp->t_callback != NULL) { - tp->t_callback(tp, tp->t_callarg); - } - - /* - * Special case the chunk embedded in the transaction. - */ - licp = &(tp->t_items); - if (!(xfs_lic_are_all_free(licp))) { - xfs_trans_chunk_committed(licp, tp->t_lsn, abortflag); - } - - /* - * Process the items in each chunk in turn. - */ - licp = licp->lic_next; - while (licp != NULL) { - ASSERT(!xfs_lic_are_all_free(licp)); - xfs_trans_chunk_committed(licp, tp->t_lsn, abortflag); - next_licp = licp->lic_next; - kmem_free(licp); - licp = next_licp; - } - - /* - * Clear all the per-AG busy list items listed in this transaction - */ - lbcp = &tp->t_busy; - while (lbcp != NULL) { - for (i = 0, lbsp = lbcp->lbc_busy; i < lbcp->lbc_unused; i++, lbsp++) { - if (!XFS_LBC_ISFREE(lbcp, i)) { - xfs_alloc_clear_busy(tp, lbsp->lbc_ag, - lbsp->lbc_idx); - } - } - lbcp = lbcp->lbc_next; - } - xfs_trans_free_busy(tp); - - /* - * That's it for the transaction structure. Free it. - */ - xfs_trans_free(tp); -} - -/* - * This is called to perform the commit processing for each - * item described by the given chunk. - * - * The commit processing consists of unlocking items which were - * held locked with the SYNC_UNLOCK attribute, calling the committed - * routine of each logged item, updating the item's position in the AIL - * if necessary, and unpinning each item. If the committed routine - * returns -1, then do nothing further with the item because it - * may have been freed. - * - * Since items are unlocked when they are copied to the incore - * log, it is possible for two transactions to be completing - * and manipulating the same item simultaneously. The AIL lock - * will protect the lsn field of each item. The value of this - * field can never go backwards. - * - * We unpin the items after repositioning them in the AIL, because - * otherwise they could be immediately flushed and we'd have to race - * with the flusher trying to pull the item from the AIL as we add it. - */ -STATIC void -xfs_trans_chunk_committed( - xfs_log_item_chunk_t *licp, - xfs_lsn_t lsn, - int aborted) -{ - xfs_log_item_desc_t *lidp; - xfs_log_item_t *lip; - xfs_lsn_t item_lsn; - int i; - - lidp = licp->lic_descs; - for (i = 0; i < licp->lic_unused; i++, lidp++) { - struct xfs_ail *ailp; - - if (xfs_lic_isfree(licp, i)) { - continue; - } - - lip = lidp->lid_item; - if (aborted) - lip->li_flags |= XFS_LI_ABORTED; - - /* - * Send in the ABORTED flag to the COMMITTED routine - * so that it knows whether the transaction was aborted - * or not. - */ - item_lsn = IOP_COMMITTED(lip, lsn); - - /* - * If the committed routine returns -1, make - * no more references to the item. - */ - if (XFS_LSN_CMP(item_lsn, (xfs_lsn_t)-1) == 0) { - continue; - } - - /* - * If the returned lsn is greater than what it - * contained before, update the location of the - * item in the AIL. If it is not, then do nothing. - * Items can never move backwards in the AIL. - * - * While the new lsn should usually be greater, it - * is possible that a later transaction completing - * simultaneously with an earlier one using the - * same item could complete first with a higher lsn. - * This would cause the earlier transaction to fail - * the test below. - */ - ailp = lip->li_ailp; - spin_lock(&ailp->xa_lock); - if (XFS_LSN_CMP(item_lsn, lip->li_lsn) > 0) { - /* - * This will set the item's lsn to item_lsn - * and update the position of the item in - * the AIL. - * - * xfs_trans_ail_update() drops the AIL lock. - */ - xfs_trans_ail_update(ailp, lip, item_lsn); - } else { - spin_unlock(&ailp->xa_lock); - } - - /* - * Now that we've repositioned the item in the AIL, - * unpin it so it can be flushed. Pass information - * about buffer stale state down from the log item - * flags, if anyone else stales the buffer we do not - * want to pay any attention to it. - */ - IOP_UNPIN(lip, lidp->lid_flags & XFS_LID_BUF_STALE); - } -} |
