From mboxrd@z Thu Jan 1 00:00:00 1970 From: Fabian Frederick Subject: [RFC] AFFS: Trying to fix fsx/O_DIRECT Date: Fri, 2 Jan 2015 17:08:11 +0100 Message-ID: <1420214891-3444-1-git-send-email-fabf@skynet.be> Cc: Andrew Morton , Jan Kara , Fabian Frederick To: linux-fsdevel@vger.kernel.org Return-path: Received: from mailrelay011.isp.belgacom.be ([195.238.6.178]:12431 "EHLO mailrelay011.isp.belgacom.be" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752114AbbABQIS (ORCPT ); Fri, 2 Jan 2015 11:08:18 -0500 Sender: linux-fsdevel-owner@vger.kernel.org List-ID: xfstests/ltp/fsx version (fsx file -d -Z -r 4096 -w 4096) always ends up with filesystem being remounted read-only due to a write operation done beyond truncate offset. affs debug: affs: error (device sdd1): get_block(): strange block request 136 affs: Remounting filesystem read-only Fsx report: " 129(129 mod 256): TRUNCATE DOWN from 0x3ff01 to 0xb3f6 130(130 mod 256): WRITE 0x22000 thru 0x2dfff (0xc000 bytes) HOLE " Error in affs_get_block: if (block >= AFFS_I(inode)->i_blkcnt) { if (block > AFFS_I(inode)->i_blkcnt || !create) goto err_big; } else create = 0; When I display values there, block is 136; i_blkcnt: 45 It seems operations are mixed up and especially truncate/get_block. I tried the patch below to add truncate mutex featuring in some other filesystems but problem remains the same. Is there something else I could do to avoid such problem ? Regards, Fabian --- fs/affs/affs.h | 1 + fs/affs/file.c | 10 +++++++++- fs/affs/super.c | 1 + 3 files changed, 11 insertions(+), 1 deletion(-) diff --git a/fs/affs/affs.h b/fs/affs/affs.h index ff44ff3..c7942d9 100644 --- a/fs/affs/affs.h +++ b/fs/affs/affs.h @@ -56,6 +56,7 @@ struct affs_inode_info { u32 i_protect; /* unused attribute bits */ u32 i_lastalloc; /* last allocated block */ int i_pa_cnt; /* number of preallocated blocks */ + struct mutex truncate_mutex; struct inode vfs_inode; }; diff --git a/fs/affs/file.c b/fs/affs/file.c index 8faa659..7e1ab3e 100644 --- a/fs/affs/file.c +++ b/fs/affs/file.c @@ -321,7 +321,10 @@ affs_get_block(struct inode *inode, sector_t block, struct buffer_head *bh_resul map_bh(bh_result, sb, (sector_t)be32_to_cpu(AFFS_BLOCK(sb, ext_bh, block))); if (create) { - u32 blocknr = affs_alloc_block(inode, ext_bh->b_blocknr); + u32 blocknr; + + mutex_lock(&AFFS_I(inode)->truncate_mutex); + blocknr = affs_alloc_block(inode, ext_bh->b_blocknr); if (!blocknr) goto err_alloc; set_buffer_new(bh_result); @@ -349,6 +352,7 @@ affs_get_block(struct inode *inode, sector_t block, struct buffer_head *bh_resul affs_brelse(ext_bh); //unlock cache + mutex_unlock(&AFFS_I(inode)->truncate_mutex); affs_unlock_ext(inode); return 0; @@ -365,6 +369,7 @@ err_alloc: clear_buffer_mapped(bh_result); bh_result->b_bdev = NULL; // unlock cache + mutex_unlock(&AFFS_I(inode)->truncate_mutex); affs_unlock_ext(inode); return -ENOSPC; } @@ -860,6 +865,7 @@ affs_truncate(struct inode *inode) // lock cache ext_bh = affs_get_extblock(inode, ext); + mutex_lock(&AFFS_I(inode)->truncate_mutex); if (IS_ERR(ext_bh)) { affs_warning(sb, "truncate", "unexpected read error for ext block %u (%ld)", @@ -912,6 +918,7 @@ affs_truncate(struct inode *inode) affs_warning(sb, "truncate", "unexpected read error for last block %u (%ld)", (unsigned int)ext, PTR_ERR(bh)); + mutex_unlock(&AFFS_I(inode)->truncate_mutex); return; } tmp = be32_to_cpu(AFFS_DATA_HEAD(bh)->next); @@ -938,6 +945,7 @@ affs_truncate(struct inode *inode) affs_brelse(ext_bh); } affs_free_prealloc(inode); + mutex_unlock(&AFFS_I(inode)->truncate_mutex); } int affs_file_fsync(struct file *filp, loff_t start, loff_t end, int datasync) diff --git a/fs/affs/super.c b/fs/affs/super.c index f754ab6..66c74df 100644 --- a/fs/affs/super.c +++ b/fs/affs/super.c @@ -123,6 +123,7 @@ static void init_once(void *foo) sema_init(&ei->i_link_lock, 1); sema_init(&ei->i_ext_lock, 1); + mutex_init(&ei->truncate_mutex); inode_init_once(&ei->vfs_inode); } -- 2.1.0