diff options
author | Mark Fasheh <mark.fasheh@oracle.com> | 2006-05-05 22:04:03 -0400 |
---|---|---|
committer | Mark Fasheh <mark.fasheh@oracle.com> | 2006-05-17 17:38:47 -0400 |
commit | 53013cba4118a5cfe8f7c7ea5e5bc1c48b160f76 (patch) | |
tree | 5170ed12fbe07b5e8557e61952aa27c25034bd7a /fs/ocfs2/file.c | |
parent | 0c056c50a6218e0e577817c16ba8851af593d742 (diff) |
ocfs2: take data locks around extend
We need to take a data lock around extends to protect the pages that
ocfs2_zero_extend is going to be pulling into the page cache. Otherwise an
extend on one node might populate the page cache with data pages that have
no lock coverage.
Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>
Diffstat (limited to 'fs/ocfs2/file.c')
-rw-r--r-- | fs/ocfs2/file.c | 70 |
1 files changed, 46 insertions, 24 deletions
diff --git a/fs/ocfs2/file.c b/fs/ocfs2/file.c index 581eb451a41..20fffeed630 100644 --- a/fs/ocfs2/file.c +++ b/fs/ocfs2/file.c | |||
@@ -613,7 +613,8 @@ leave: | |||
613 | 613 | ||
614 | /* Some parts of this taken from generic_cont_expand, which turned out | 614 | /* Some parts of this taken from generic_cont_expand, which turned out |
615 | * to be too fragile to do exactly what we need without us having to | 615 | * to be too fragile to do exactly what we need without us having to |
616 | * worry about recursive locking in ->commit_write(). */ | 616 | * worry about recursive locking in ->prepare_write() and |
617 | * ->commit_write(). */ | ||
617 | static int ocfs2_write_zero_page(struct inode *inode, | 618 | static int ocfs2_write_zero_page(struct inode *inode, |
618 | u64 size) | 619 | u64 size) |
619 | { | 620 | { |
@@ -641,7 +642,7 @@ static int ocfs2_write_zero_page(struct inode *inode, | |||
641 | goto out; | 642 | goto out; |
642 | } | 643 | } |
643 | 644 | ||
644 | ret = ocfs2_prepare_write(NULL, page, offset, offset); | 645 | ret = ocfs2_prepare_write_nolock(inode, page, offset, offset); |
645 | if (ret < 0) { | 646 | if (ret < 0) { |
646 | mlog_errno(ret); | 647 | mlog_errno(ret); |
647 | goto out_unlock; | 648 | goto out_unlock; |
@@ -695,13 +696,26 @@ out: | |||
695 | return ret; | 696 | return ret; |
696 | } | 697 | } |
697 | 698 | ||
699 | /* | ||
700 | * A tail_to_skip value > 0 indicates that we're being called from | ||
701 | * ocfs2_file_aio_write(). This has the following implications: | ||
702 | * | ||
703 | * - we don't want to update i_size | ||
704 | * - di_bh will be NULL, which is fine because it's only used in the | ||
705 | * case where we want to update i_size. | ||
706 | * - ocfs2_zero_extend() will then only be filling the hole created | ||
707 | * between i_size and the start of the write. | ||
708 | */ | ||
698 | static int ocfs2_extend_file(struct inode *inode, | 709 | static int ocfs2_extend_file(struct inode *inode, |
699 | struct buffer_head *di_bh, | 710 | struct buffer_head *di_bh, |
700 | u64 new_i_size) | 711 | u64 new_i_size, |
712 | size_t tail_to_skip) | ||
701 | { | 713 | { |
702 | int ret = 0; | 714 | int ret = 0; |
703 | u32 clusters_to_add; | 715 | u32 clusters_to_add; |
704 | 716 | ||
717 | BUG_ON(!tail_to_skip && !di_bh); | ||
718 | |||
705 | /* setattr sometimes calls us like this. */ | 719 | /* setattr sometimes calls us like this. */ |
706 | if (new_i_size == 0) | 720 | if (new_i_size == 0) |
707 | goto out; | 721 | goto out; |
@@ -714,27 +728,44 @@ static int ocfs2_extend_file(struct inode *inode, | |||
714 | OCFS2_I(inode)->ip_clusters; | 728 | OCFS2_I(inode)->ip_clusters; |
715 | 729 | ||
716 | if (clusters_to_add) { | 730 | if (clusters_to_add) { |
717 | ret = ocfs2_extend_allocation(inode, clusters_to_add); | 731 | /* |
732 | * protect the pages that ocfs2_zero_extend is going to | ||
733 | * be pulling into the page cache.. we do this before the | ||
734 | * metadata extend so that we don't get into the situation | ||
735 | * where we've extended the metadata but can't get the data | ||
736 | * lock to zero. | ||
737 | */ | ||
738 | ret = ocfs2_data_lock(inode, 1); | ||
718 | if (ret < 0) { | 739 | if (ret < 0) { |
719 | mlog_errno(ret); | 740 | mlog_errno(ret); |
720 | goto out; | 741 | goto out; |
721 | } | 742 | } |
722 | 743 | ||
723 | ret = ocfs2_zero_extend(inode, new_i_size); | 744 | ret = ocfs2_extend_allocation(inode, clusters_to_add); |
724 | if (ret < 0) { | 745 | if (ret < 0) { |
725 | mlog_errno(ret); | 746 | mlog_errno(ret); |
726 | goto out; | 747 | goto out_unlock; |
727 | } | 748 | } |
728 | } | ||
729 | 749 | ||
730 | /* No allocation required, we just use this helper to | 750 | ret = ocfs2_zero_extend(inode, (u64)new_i_size - tail_to_skip); |
731 | * do a trivial update of i_size. */ | 751 | if (ret < 0) { |
732 | ret = ocfs2_simple_size_update(inode, di_bh, new_i_size); | 752 | mlog_errno(ret); |
733 | if (ret < 0) { | 753 | goto out_unlock; |
734 | mlog_errno(ret); | 754 | } |
735 | goto out; | 755 | } |
756 | |||
757 | if (!tail_to_skip) { | ||
758 | /* We're being called from ocfs2_setattr() which wants | ||
759 | * us to update i_size */ | ||
760 | ret = ocfs2_simple_size_update(inode, di_bh, new_i_size); | ||
761 | if (ret < 0) | ||
762 | mlog_errno(ret); | ||
736 | } | 763 | } |
737 | 764 | ||
765 | out_unlock: | ||
766 | if (clusters_to_add) /* this is the only case in which we lock */ | ||
767 | ocfs2_data_unlock(inode, 1); | ||
768 | |||
738 | out: | 769 | out: |
739 | return ret; | 770 | return ret; |
740 | } | 771 | } |
@@ -793,7 +824,7 @@ int ocfs2_setattr(struct dentry *dentry, struct iattr *attr) | |||
793 | if (i_size_read(inode) > attr->ia_size) | 824 | if (i_size_read(inode) > attr->ia_size) |
794 | status = ocfs2_truncate_file(inode, bh, attr->ia_size); | 825 | status = ocfs2_truncate_file(inode, bh, attr->ia_size); |
795 | else | 826 | else |
796 | status = ocfs2_extend_file(inode, bh, attr->ia_size); | 827 | status = ocfs2_extend_file(inode, bh, attr->ia_size, 0); |
797 | if (status < 0) { | 828 | if (status < 0) { |
798 | if (status != -ENOSPC) | 829 | if (status != -ENOSPC) |
799 | mlog_errno(status); | 830 | mlog_errno(status); |
@@ -1049,21 +1080,12 @@ static ssize_t ocfs2_file_aio_write(struct kiocb *iocb, | |||
1049 | if (!clusters) | 1080 | if (!clusters) |
1050 | break; | 1081 | break; |
1051 | 1082 | ||
1052 | ret = ocfs2_extend_allocation(inode, clusters); | 1083 | ret = ocfs2_extend_file(inode, NULL, newsize, count); |
1053 | if (ret < 0) { | 1084 | if (ret < 0) { |
1054 | if (ret != -ENOSPC) | 1085 | if (ret != -ENOSPC) |
1055 | mlog_errno(ret); | 1086 | mlog_errno(ret); |
1056 | goto out; | 1087 | goto out; |
1057 | } | 1088 | } |
1058 | |||
1059 | /* Fill any holes which would've been created by this | ||
1060 | * write. If we're O_APPEND, this will wind up | ||
1061 | * (correctly) being a noop. */ | ||
1062 | ret = ocfs2_zero_extend(inode, (u64) newsize - count); | ||
1063 | if (ret < 0) { | ||
1064 | mlog_errno(ret); | ||
1065 | goto out; | ||
1066 | } | ||
1067 | break; | 1089 | break; |
1068 | } | 1090 | } |
1069 | 1091 | ||