linux/fs/xfs
Zach Brown 8459d86aff [PATCH] dio: only call aio_complete() after returning -EIOCBQUEUED
The only time it is safe to call aio_complete() is when the ->ki_retry
function returns -EIOCBQUEUED to the AIO core.  direct_io_worker() has
historically done this by relying on its caller to translate positive return
codes into -EIOCBQUEUED for the aio case.  It did this by trying to keep
conditionals in sync.  direct_io_worker() knew when finished_one_bio() was
going to call aio_complete().  It would reverse the test and wait and free the
dio in the cases it thought that finished_one_bio() wasn't going to.

Not surprisingly, it ended up getting it wrong.  'ret' could be a negative
errno from the submission path but it failed to communicate this to
finished_one_bio().  direct_io_worker() would return < 0, it's callers
wouldn't raise -EIOCBQUEUED, and aio_complete() would be called.  In the
future finished_one_bio()'s tests wouldn't reflect this and aio_complete()
would be called for a second time which can manifest as an oops.

The previous cleanups have whittled the sync and async completion paths down
to the point where we can collapse them and clearly reassert the invariant
that we must only call aio_complete() after returning -EIOCBQUEUED.
direct_io_worker() will only return -EIOCBQUEUED when it is not the last to
drop the dio refcount and the aio bio completion path will only call
aio_complete() when it is the last to drop the dio refcount.
direct_io_worker() can ensure that it is the last to drop the reference count
by waiting for bios to drain.  It does this for sync ops, of course, and for
partial dio writes that must fall back to buffered and for aio ops that saw
errors during submission.

This means that operations that end up waiting, even if they were issued as
aio ops, will not call aio_complete() from dio.  Instead we return the return
code of the operation and let the aio core call aio_complete().  This is
purposely done to fix a bug where AIO DIO file extensions would call
aio_complete() before their callers have a chance to update i_size.

Now that direct_io_worker() is explicitly returning -EIOCBQUEUED its callers
no longer have to translate for it.  XFS needs to be careful not to free
resources that will be used during AIO completion if -EIOCBQUEUED is returned.
 We maintain the previous behaviour of trying to write fs metadata for O_SYNC
aio+dio writes.

Signed-off-by: Zach Brown <zach.brown@oracle.com>
Cc: Badari Pulavarty <pbadari@us.ibm.com>
Cc: Suparna Bhattacharya <suparna@in.ibm.com>
Acked-by: Jeff Moyer <jmoyer@redhat.com>
Cc: <xfs-masters@oss.sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-10 09:57:21 -08:00
..
linux-2.6 [PATCH] dio: only call aio_complete() after returning -EIOCBQUEUED 2006-12-10 09:57:21 -08:00
quota [XFS] Fix kmem_zalloc_greedy warnings on 64 bit platforms. 2006-09-28 11:04:43 +10:00
support [XFS] rename uio_read() to xfs_uio_read() 2006-11-11 18:04:41 +11:00
Kbuild kbuild/xfs: introduce fs/xfs/Kbuild 2006-01-09 20:48:03 +01:00
Kconfig [PATCH] BLOCK: Make it possible to disable the block layer [try #6] 2006-09-30 20:52:31 +02:00
Makefile [XFS] Sort out some cosmetic differences between XFS trees. 2005-09-05 11:47:01 +10:00
Makefile-linux-2.6 [XFS] 956618: Linux crashes on boot with XFS-DMAPI filesystem when 2006-11-11 18:03:49 +11:00
xfs.h [XFS] 956618: Linux crashes on boot with XFS-DMAPI filesystem when 2006-11-11 18:03:49 +11:00
xfs_acl.c [XFS] Remove version 1 directory code. Never functioned on Linux, just 2006-06-20 13:04:51 +10:00
xfs_acl.h [XFS] Resolve a namespace collision on remaining vtypes for FreeBSD 2006-06-09 17:07:12 +10:00
xfs_ag.h [XFS] endianess annotation for xfs_agfl_t. Trivial, xfs_agfl_t is always 2006-09-28 10:56:51 +10:00
xfs_alloc.c [XFS] Minor code rearranging and cleanup to prevent some coverity false 2006-09-28 11:03:44 +10:00
xfs_alloc.h [XFS] Prevent free space oversubscription and xfssyncd looping. 2006-09-07 14:26:50 +10:00
xfs_alloc_btree.c [XFS] Reduce endian flipping in alloc_btree, same as was done for 2006-09-28 11:05:40 +10:00
xfs_alloc_btree.h [XFS] Endianess annotations for various allocator data structures 2005-11-02 15:11:25 +11:00
xfs_arch.h [XFS] Merge in trivial changes, sync up headers with userspace 2006-01-12 10:29:53 +11:00
xfs_attr.c [XFS] Add EA list callbacks for xfs kernel use. Cleanup some namespace 2006-09-28 11:01:37 +10:00
xfs_attr.h [XFS] Add EA list callbacks for xfs kernel use. Cleanup some namespace 2006-09-28 11:01:37 +10:00
xfs_attr_leaf.c [XFS] Add EA list callbacks for xfs kernel use. Cleanup some namespace 2006-09-28 11:01:37 +10:00
xfs_attr_leaf.h [XFS] Add EA list callbacks for xfs kernel use. Cleanup some namespace 2006-09-28 11:01:37 +10:00
xfs_attr_sf.h [XFS] endianess annotations for xfs_attr_shortform_t 2006-03-17 17:29:25 +11:00
xfs_behavior.c [XFS] remove bhv_lookup, _range version works aswell and has more useful 2006-09-28 10:58:52 +10:00
xfs_behavior.h [XFS] remove bhv_lookup, _range version works aswell and has more useful 2006-09-28 10:58:52 +10:00
xfs_bit.c [XFS] Update license/copyright notices to match the prefered SGI 2005-11-02 14:58:39 +11:00
xfs_bit.h [XFS] Update license/copyright notices to match the prefered SGI 2005-11-02 14:58:39 +11:00
xfs_bmap.c [XFS] Fix uninitialized br_state and br_startoff in 2006-11-21 18:55:16 +11:00
xfs_bmap.h [XFS] Add parameters to xfs_bmapi() and xfs_bunmapi() to have them report 2006-06-09 14:48:12 +10:00
xfs_bmap_btree.c [XFS] use NULL for pointer initialisation instead of zero-cast-to-ptr 2006-09-28 10:58:40 +10:00
xfs_bmap_btree.h [XFS] endianess annotations for xfs_bmbt_key Trivial as there are no 2006-09-28 10:58:17 +10:00
xfs_btree.c [XFS] endianess annotations for xfs_bmbt_key Trivial as there are no 2006-09-28 10:58:17 +10:00
xfs_btree.h [XFS] endianess annotate XFS_BMAP_BROOT_PTR_ADDR Make sure it returns a 2006-09-28 10:58:06 +10:00
xfs_buf_item.c [XFS] Remove unused iop_abort log item operation 2006-09-28 11:02:44 +10:00
xfs_buf_item.h [XFS] Update license/copyright notices to match the prefered SGI 2005-11-02 14:58:39 +11:00
xfs_cap.h [XFS] Resolve a namespace collision on vnode/vnodeops for FreeBSD porters. 2006-06-09 17:00:52 +10:00
xfs_clnt.h [XFS] Implement the silent parameter to fill_super, previously ignored. 2006-03-31 13:04:17 +10:00
xfs_da_btree.c [XFS] Minor code rearranging and cleanup to prevent some coverity false 2006-09-28 11:03:44 +10:00
xfs_da_btree.h [XFS] Remove version 1 directory code. Never functioned on Linux, just 2006-06-20 13:04:51 +10:00
xfs_dfrag.c [PATCH] xfs: change uses of f_{dentry,vfsmnt} to use f_path 2006-12-08 08:28:43 -08:00
xfs_dfrag.h [XFS] Add parameters to xfs_bmapi() and xfs_bunmapi() to have them report 2006-06-09 14:48:12 +10:00
xfs_dinode.h [XFS] Remove version 1 directory code. Never functioned on Linux, just 2006-06-20 13:04:51 +10:00
xfs_dir2.c [XFS] rename uio_read() to xfs_uio_read() 2006-11-11 18:04:41 +11:00
xfs_dir2.h [XFS] Remove version 1 directory code. Never functioned on Linux, just 2006-06-20 13:04:51 +10:00
xfs_dir2_block.c [XFS] Remove version 1 directory code. Never functioned on Linux, just 2006-06-20 13:04:51 +10:00
xfs_dir2_block.h [XFS] endianess annotations for xfs_dir2_block_tail_t 2006-03-17 17:27:56 +11:00
xfs_dir2_data.c [XFS] Remove version 1 directory code. Never functioned on Linux, just 2006-06-20 13:04:51 +10:00
xfs_dir2_data.h [XFS] endianess annotations for xfs_dir2_data_entry_t 2006-06-09 14:48:37 +10:00
xfs_dir2_leaf.c [XFS] Remove version 1 directory code. Never functioned on Linux, just 2006-06-20 13:04:51 +10:00
xfs_dir2_leaf.h [XFS] endianess annotations for xfs_dir2_leaf_entry_t 2006-03-17 17:28:18 +11:00
xfs_dir2_node.c [XFS] Remove version 1 directory code. Never functioned on Linux, just 2006-06-20 13:04:51 +10:00
xfs_dir2_node.h [XFS] endianess annotations for xfs_dir2_free_hdr_t 2006-03-17 17:27:07 +11:00
xfs_dir2_sf.c [XFS] Remove version 1 directory code. Never functioned on Linux, just 2006-06-20 13:04:51 +10:00
xfs_dir2_sf.h [XFS] Update license/copyright notices to match the prefered SGI 2005-11-02 14:58:39 +11:00
xfs_dir2_trace.c [XFS] Remove version 1 directory code. Never functioned on Linux, just 2006-06-20 13:04:51 +10:00
xfs_dir2_trace.h [XFS] Update license/copyright notices to match the prefered SGI 2005-11-02 14:58:39 +11:00
xfs_dmapi.h [XFS] Remove KERNEL_VERSION macros from xfs_dmapi.h 2006-11-11 18:05:06 +11:00
xfs_dmops.c [XFS] Remove version 1 directory code. Never functioned on Linux, just 2006-06-20 13:04:51 +10:00
xfs_error.c [XFS] Remove version 1 directory code. Never functioned on Linux, just 2006-06-20 13:04:51 +10:00
xfs_error.h [XFS] Improve error handling for the zero-fsblock extent detection code. 2006-09-28 11:03:20 +10:00
xfs_extfree_item.c [XFS] Remove unused iop_abort log item operation 2006-09-28 11:02:44 +10:00
xfs_extfree_item.h [XFS] cleanup the field types of some item format structures 2006-09-28 10:55:43 +10:00
xfs_fs.h [XFS] Remove several macros that are no longer used anywhere 2006-09-28 11:02:57 +10:00
xfs_fsops.c [XFS] Prevent free space oversubscription and xfssyncd looping. 2006-09-07 14:26:50 +10:00
xfs_fsops.h [XFS] Write log dummy record when freezing filesystem 2006-01-11 15:30:08 +11:00
xfs_ialloc.c [XFS] Remove last bulkstat false-positives with debug kernels. 2006-09-28 11:02:23 +10:00
xfs_ialloc.h [XFS] Update license/copyright notices to match the prefered SGI 2005-11-02 14:58:39 +11:00
xfs_ialloc_btree.c [XFS] use NULL for pointer initialisation instead of zero-cast-to-ptr 2006-09-28 10:58:40 +10:00
xfs_ialloc_btree.h [XFS] endianess annotations for xfs_inobt_rec_t / xfs_inobt_key_t 2006-09-28 10:57:04 +10:00
xfs_iget.c [XFS] Prevent a deadlock when xfslogd unpins inodes. 2006-11-11 18:05:00 +11:00
xfs_imap.h [XFS] Update license/copyright notices to match the prefered SGI 2005-11-02 14:58:39 +11:00
xfs_inode.c [XFS] Stale the correct inode when freeing clusters. 2006-11-21 18:55:33 +11:00
xfs_inode.h [XFS] Clean up i_flags and i_flags_lock handling. 2006-11-11 18:04:54 +11:00
xfs_inode_item.c [XFS] Remove unused iop_abort log item operation 2006-09-28 11:02:44 +10:00
xfs_inode_item.h [XFS] cleanup the field types of some item format structures 2006-09-28 10:55:43 +10:00
xfs_inum.h [XFS] Update license/copyright notices to match the prefered SGI 2005-11-02 14:58:39 +11:00
xfs_iocore.c [XFS] Remove version 1 directory code. Never functioned on Linux, just 2006-06-20 13:04:51 +10:00
xfs_iomap.c [XFS] Improve error handling for the zero-fsblock extent detection code. 2006-09-28 11:03:20 +10:00
xfs_iomap.h [XFS] Fix potential overflow in xfs_iomap_t delta for very large extents 2005-11-25 16:41:33 +11:00
xfs_itable.c [XFS] 955947: Infinite loop in xfs_bulkstat() on formatter() error 2006-09-28 11:06:21 +10:00
xfs_itable.h [XFS] pv 956241, author: nathans, rv: vapo - make ino validation checks 2006-09-28 11:06:15 +10:00
xfs_log.c [XFS] Fixes the leak in reservation space because we weren't ungranting 2006-09-28 11:04:16 +10:00
xfs_log.h [XFS] Remove several macros that are no longer used anywhere 2006-09-28 11:02:57 +10:00
xfs_log_priv.h [XFS] Fixes the leak in reservation space because we weren't ungranting 2006-09-28 11:04:16 +10:00
xfs_log_recover.c [XFS] Rework code snippets slightly to remove remaining recent-gcc 2006-06-28 10:13:52 +10:00
xfs_log_recover.h [XFS] Update license/copyright notices to match the prefered SGI 2005-11-02 14:58:39 +11:00
xfs_mac.h [XFS] Update license/copyright notices to match the prefered SGI 2005-11-02 14:58:39 +11:00
xfs_mount.c [XFS] Prevent free space oversubscription and xfssyncd looping. 2006-09-07 14:26:50 +10:00
xfs_mount.h [XFS] Fix kmem_zalloc_greedy warnings on 64 bit platforms. 2006-09-28 11:04:43 +10:00
xfs_qmops.c [XFS] Remove version 1 directory code. Never functioned on Linux, just 2006-06-20 13:04:51 +10:00
xfs_quota.h [XFS] Remove several macros that are no longer used anywhere 2006-09-28 11:02:57 +10:00
xfs_refcache.h [XFS] Cleanup cosmetic differences between source trees. 2005-11-03 16:14:31 +11:00
xfs_rename.c [XFS] Remove version 1 directory code. Never functioned on Linux, just 2006-06-20 13:04:51 +10:00
xfs_rtalloc.c [XFS] Fix a porting botch on the realtime subvol growfs code path. 2006-09-28 11:03:53 +10:00
xfs_rtalloc.h [XFS] Update license/copyright notices to match the prefered SGI 2005-11-02 14:58:39 +11:00
xfs_rw.c [XFS] Remove version 1 directory code. Never functioned on Linux, just 2006-06-20 13:04:51 +10:00
xfs_rw.h [XFS] Push some common code out of write path into core XFS code for 2006-06-19 08:39:53 +10:00
xfs_sb.h [XFS] Remove several macros that are no longer used anywhere 2006-09-28 11:02:57 +10:00
xfs_trans.c [XFS] Remove version 1 directory code. Never functioned on Linux, just 2006-06-20 13:04:51 +10:00
xfs_trans.h [XFS] Remove unused iop_abort log item operation 2006-09-28 11:02:44 +10:00
xfs_trans_ail.c [XFS] Add lock annotations to xfs_trans_update_ail and 2006-09-28 11:04:07 +10:00
xfs_trans_buf.c [XFS] Remove version 1 directory code. Never functioned on Linux, just 2006-06-20 13:04:51 +10:00
xfs_trans_extfree.c [XFS] Remove version 1 directory code. Never functioned on Linux, just 2006-06-20 13:04:51 +10:00
xfs_trans_inode.c [XFS] Remove version 1 directory code. Never functioned on Linux, just 2006-06-20 13:04:51 +10:00
xfs_trans_item.c [XFS] Portability changes: remove prdev, stick to one diagnostic 2006-06-09 15:29:40 +10:00
xfs_trans_priv.h [XFS] Add lock annotations to xfs_trans_update_ail and 2006-09-28 11:04:07 +10:00
xfs_trans_space.h [XFS] Remove version 1 directory code. Never functioned on Linux, just 2006-06-20 13:04:51 +10:00
xfs_types.h [XFS] Update license/copyright notices to match the prefered SGI 2005-11-02 14:58:39 +11:00
xfs_utils.c [XFS] Remove version 1 directory code. Never functioned on Linux, just 2006-06-20 13:04:51 +10:00
xfs_utils.h [XFS] Resolve a namespace collision on remaining vtypes for FreeBSD 2006-06-09 17:07:12 +10:00
xfs_vfsops.c [XFS] Fix kmem_zalloc_greedy warnings on 64 bit platforms. 2006-09-28 11:04:43 +10:00
xfs_vnodeops.c [XFS] Prevent a deadlock when xfslogd unpins inodes. 2006-11-11 18:05:00 +11:00