2018-06-06 05:42:14 +03:00
// SPDX-License-Identifier: GPL-2.0+
2016-08-03 05:04:45 +03:00
/*
* Copyright ( C ) 2016 Oracle . All Rights Reserved .
* Author : Darrick J . Wong < darrick . wong @ oracle . com >
*/
# include "xfs.h"
# include "xfs_fs.h"
# include "xfs_format.h"
# include "xfs_log_format.h"
# include "xfs_trans_resv.h"
2016-08-03 05:09:48 +03:00
# include "xfs_bit.h"
2018-02-23 01:41:25 +03:00
# include "xfs_shared.h"
2016-08-03 05:04:45 +03:00
# include "xfs_mount.h"
xfs: propagate bmap updates to rmapbt
When we map, unmap, or convert an extent in a file's data or attr
fork, schedule a respective update in the rmapbt. Previous versions
of this patch required a 1:1 correspondence between bmap and rmap,
but this is no longer true as we now have ability to make interval
queries against the rmapbt.
We use the deferred operations code to handle redo operations
atomically and deadlock free. This plumbs in all five rmap actions
(map, unmap, convert extent, alloc, free); we'll use the first three
now for file data, and reflink will want the last two. We also add
an error injection site to test log recovery.
Finally, we need to fix the bmap shift extent code to adjust the
rmaps correctly.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-08-03 05:16:05 +03:00
# include "xfs_defer.h"
2016-08-03 05:04:45 +03:00
# include "xfs_trans.h"
# include "xfs_trans_priv.h"
# include "xfs_rmap_item.h"
# include "xfs_log.h"
xfs: propagate bmap updates to rmapbt
When we map, unmap, or convert an extent in a file's data or attr
fork, schedule a respective update in the rmapbt. Previous versions
of this patch required a 1:1 correspondence between bmap and rmap,
but this is no longer true as we now have ability to make interval
queries against the rmapbt.
We use the deferred operations code to handle redo operations
atomically and deadlock free. This plumbs in all five rmap actions
(map, unmap, convert extent, alloc, free); we'll use the first three
now for file data, and reflink will want the last two. We also add
an error injection site to test log recovery.
Finally, we need to fix the bmap shift extent code to adjust the
rmaps correctly.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-08-03 05:16:05 +03:00
# include "xfs_rmap.h"
2019-11-02 19:40:53 +03:00
# include "xfs_error.h"
2020-05-02 02:00:49 +03:00
# include "xfs_log_priv.h"
2020-05-02 02:00:45 +03:00
# include "xfs_log_recover.h"
2016-08-03 05:04:45 +03:00
kmem_zone_t * xfs_rui_zone ;
kmem_zone_t * xfs_rud_zone ;
2020-05-02 02:00:51 +03:00
static const struct xfs_item_ops xfs_rui_item_ops ;
2016-08-03 05:04:45 +03:00
static inline struct xfs_rui_log_item * RUI_ITEM ( struct xfs_log_item * lip )
{
return container_of ( lip , struct xfs_rui_log_item , rui_item ) ;
}
2020-05-02 02:00:49 +03:00
STATIC void
2016-08-03 05:04:45 +03:00
xfs_rui_item_free (
struct xfs_rui_log_item * ruip )
{
if ( ruip - > rui_format . rui_nextents > XFS_RUI_MAX_FAST_EXTENTS )
kmem_free ( ruip ) ;
else
2019-11-14 23:43:04 +03:00
kmem_cache_free ( xfs_rui_zone , ruip ) ;
2016-08-03 05:04:45 +03:00
}
2018-04-03 06:08:27 +03:00
/*
* Freeing the RUI requires that we remove it from the AIL if it has already
* been placed there . However , the RUI may not yet have been placed in the AIL
* when called by xfs_rui_release ( ) from RUD processing due to the ordering of
* committed vs unpin operations in bulk insert operations . Hence the reference
* count to ensure only the last caller frees the RUI .
*/
2020-05-02 02:00:51 +03:00
STATIC void
2018-04-03 06:08:27 +03:00
xfs_rui_release (
struct xfs_rui_log_item * ruip )
{
ASSERT ( atomic_read ( & ruip - > rui_refcount ) > 0 ) ;
if ( atomic_dec_and_test ( & ruip - > rui_refcount ) ) {
2020-05-06 23:25:23 +03:00
xfs_trans_ail_delete ( & ruip - > rui_item , SHUTDOWN_LOG_IO_ERROR ) ;
2018-04-03 06:08:27 +03:00
xfs_rui_item_free ( ruip ) ;
}
}
2016-08-03 05:04:45 +03:00
STATIC void
xfs_rui_item_size (
struct xfs_log_item * lip ,
int * nvecs ,
int * nbytes )
{
2016-09-19 03:24:27 +03:00
struct xfs_rui_log_item * ruip = RUI_ITEM ( lip ) ;
2016-08-03 05:04:45 +03:00
* nvecs + = 1 ;
2016-09-19 03:24:27 +03:00
* nbytes + = xfs_rui_log_format_sizeof ( ruip - > rui_format . rui_nextents ) ;
2016-08-03 05:04:45 +03:00
}
/*
* This is called to fill in the vector of log iovecs for the
* given rui log item . We use only 1 iovec , and we point that
* at the rui_log_format structure embedded in the rui item .
* It is at this point that we assert that all of the extent
* slots in the rui item have been filled .
*/
STATIC void
xfs_rui_item_format (
struct xfs_log_item * lip ,
struct xfs_log_vec * lv )
{
struct xfs_rui_log_item * ruip = RUI_ITEM ( lip ) ;
struct xfs_log_iovec * vecp = NULL ;
ASSERT ( atomic_read ( & ruip - > rui_next_extent ) = =
ruip - > rui_format . rui_nextents ) ;
ruip - > rui_format . rui_type = XFS_LI_RUI ;
ruip - > rui_format . rui_size = 1 ;
xlog_copy_iovec ( lv , & vecp , XLOG_REG_TYPE_RUI_FORMAT , & ruip - > rui_format ,
2016-09-19 03:24:27 +03:00
xfs_rui_log_format_sizeof ( ruip - > rui_format . rui_nextents ) ) ;
2016-08-03 05:04:45 +03:00
}
/*
* The unpin operation is the last place an RUI is manipulated in the log . It is
* either inserted in the AIL or aborted in the event of a log I / O error . In
* either case , the RUI transaction has been successfully committed to make it
* this far . Therefore , we expect whoever committed the RUI to either construct
* and commit the RUD or drop the RUD ' s reference in the event of error . Simply
* drop the log ' s RUI reference now that the log is done with it .
*/
STATIC void
xfs_rui_item_unpin (
struct xfs_log_item * lip ,
int remove )
{
struct xfs_rui_log_item * ruip = RUI_ITEM ( lip ) ;
xfs_rui_release ( ruip ) ;
}
/*
* The RUI has been either committed or aborted if the transaction has been
* cancelled . If the transaction was cancelled , an RUD isn ' t going to be
* constructed and thus we free the RUI here directly .
*/
STATIC void
2019-06-29 05:27:32 +03:00
xfs_rui_item_release (
2016-08-03 05:04:45 +03:00
struct xfs_log_item * lip )
{
2019-06-29 05:27:32 +03:00
xfs_rui_release ( RUI_ITEM ( lip ) ) ;
2016-08-03 05:04:45 +03:00
}
/*
* Allocate and initialize an rui item with the given number of extents .
*/
2020-05-02 02:00:49 +03:00
STATIC struct xfs_rui_log_item *
2016-08-03 05:04:45 +03:00
xfs_rui_init (
struct xfs_mount * mp ,
uint nextents )
{
struct xfs_rui_log_item * ruip ;
ASSERT ( nextents > 0 ) ;
2016-09-19 03:24:27 +03:00
if ( nextents > XFS_RUI_MAX_FAST_EXTENTS )
2019-08-26 22:06:22 +03:00
ruip = kmem_zalloc ( xfs_rui_log_item_sizeof ( nextents ) , 0 ) ;
2016-09-19 03:24:27 +03:00
else
2020-07-22 19:23:10 +03:00
ruip = kmem_cache_zalloc ( xfs_rui_zone ,
GFP_KERNEL | __GFP_NOFAIL ) ;
2016-08-03 05:04:45 +03:00
xfs_log_item_init ( mp , & ruip - > rui_item , XFS_LI_RUI , & xfs_rui_item_ops ) ;
ruip - > rui_format . rui_nextents = nextents ;
ruip - > rui_format . rui_id = ( uintptr_t ) ( void * ) ruip ;
atomic_set ( & ruip - > rui_next_extent , 0 ) ;
atomic_set ( & ruip - > rui_refcount , 2 ) ;
return ruip ;
}
/*
* Copy an RUI format buffer from the given buf , and into the destination
* RUI format structure . The RUI / RUD items were designed not to need any
* special alignment handling .
*/
2020-05-02 02:00:49 +03:00
STATIC int
2016-08-03 05:04:45 +03:00
xfs_rui_copy_format (
struct xfs_log_iovec * buf ,
struct xfs_rui_log_format * dst_rui_fmt )
{
struct xfs_rui_log_format * src_rui_fmt ;
uint len ;
src_rui_fmt = buf - > i_addr ;
2016-09-19 03:24:27 +03:00
len = xfs_rui_log_format_sizeof ( src_rui_fmt - > rui_nextents ) ;
2016-08-03 05:04:45 +03:00
2019-11-02 19:40:53 +03:00
if ( buf - > i_len ! = len ) {
XFS_ERROR_REPORT ( __func__ , XFS_ERRLEVEL_LOW , NULL ) ;
2016-08-03 05:04:45 +03:00
return - EFSCORRUPTED ;
2019-11-02 19:40:53 +03:00
}
2016-08-03 05:04:45 +03:00
2016-09-19 03:24:27 +03:00
memcpy ( dst_rui_fmt , src_rui_fmt , len ) ;
2016-08-03 05:04:45 +03:00
return 0 ;
}
static inline struct xfs_rud_log_item * RUD_ITEM ( struct xfs_log_item * lip )
{
return container_of ( lip , struct xfs_rud_log_item , rud_item ) ;
}
STATIC void
xfs_rud_item_size (
struct xfs_log_item * lip ,
int * nvecs ,
int * nbytes )
{
* nvecs + = 1 ;
2016-08-03 05:28:43 +03:00
* nbytes + = sizeof ( struct xfs_rud_log_format ) ;
2016-08-03 05:04:45 +03:00
}
/*
* This is called to fill in the vector of log iovecs for the
* given rud log item . We use only 1 iovec , and we point that
* at the rud_log_format structure embedded in the rud item .
* It is at this point that we assert that all of the extent
* slots in the rud item have been filled .
*/
STATIC void
xfs_rud_item_format (
struct xfs_log_item * lip ,
struct xfs_log_vec * lv )
{
struct xfs_rud_log_item * rudp = RUD_ITEM ( lip ) ;
struct xfs_log_iovec * vecp = NULL ;
rudp - > rud_format . rud_type = XFS_LI_RUD ;
rudp - > rud_format . rud_size = 1 ;
xlog_copy_iovec ( lv , & vecp , XLOG_REG_TYPE_RUD_FORMAT , & rudp - > rud_format ,
2016-08-03 05:28:43 +03:00
sizeof ( struct xfs_rud_log_format ) ) ;
2016-08-03 05:04:45 +03:00
}
/*
* The RUD is either committed or aborted if the transaction is cancelled . If
* the transaction is cancelled , drop our reference to the RUI and free the
* RUD .
*/
STATIC void
2019-06-29 05:27:32 +03:00
xfs_rud_item_release (
2016-08-03 05:04:45 +03:00
struct xfs_log_item * lip )
{
struct xfs_rud_log_item * rudp = RUD_ITEM ( lip ) ;
2019-06-29 05:27:32 +03:00
xfs_rui_release ( rudp - > rud_ruip ) ;
2019-11-14 23:43:04 +03:00
kmem_cache_free ( xfs_rud_zone , rudp ) ;
2016-08-03 05:04:45 +03:00
}
static const struct xfs_item_ops xfs_rud_item_ops = {
2019-06-29 05:27:32 +03:00
. flags = XFS_ITEM_RELEASE_WHEN_COMMITTED ,
2016-08-03 05:04:45 +03:00
. iop_size = xfs_rud_item_size ,
. iop_format = xfs_rud_item_format ,
2019-06-29 05:27:32 +03:00
. iop_release = xfs_rud_item_release ,
2016-08-03 05:04:45 +03:00
} ;
2019-06-29 05:29:41 +03:00
static struct xfs_rud_log_item *
2019-06-29 05:27:36 +03:00
xfs_trans_get_rud (
struct xfs_trans * tp ,
2016-08-03 05:28:43 +03:00
struct xfs_rui_log_item * ruip )
2016-08-03 05:04:45 +03:00
{
2019-06-29 05:27:36 +03:00
struct xfs_rud_log_item * rudp ;
2016-08-03 05:04:45 +03:00
2020-07-22 19:23:10 +03:00
rudp = kmem_cache_zalloc ( xfs_rud_zone , GFP_KERNEL | __GFP_NOFAIL ) ;
2019-06-29 05:27:36 +03:00
xfs_log_item_init ( tp - > t_mountp , & rudp - > rud_item , XFS_LI_RUD ,
& xfs_rud_item_ops ) ;
2016-08-03 05:04:45 +03:00
rudp - > rud_ruip = ruip ;
rudp - > rud_format . rud_rui_id = ruip - > rui_format . rui_id ;
2019-06-29 05:27:36 +03:00
xfs_trans_add_item ( tp , & rudp - > rud_item ) ;
2016-08-03 05:04:45 +03:00
return rudp ;
}
2016-08-03 05:09:48 +03:00
2019-06-29 05:29:41 +03:00
/* Set the map extent flags for this reverse mapping. */
static void
xfs_trans_set_rmap_flags (
struct xfs_map_extent * rmap ,
enum xfs_rmap_intent_type type ,
int whichfork ,
xfs_exntst_t state )
{
rmap - > me_flags = 0 ;
if ( state = = XFS_EXT_UNWRITTEN )
rmap - > me_flags | = XFS_RMAP_EXTENT_UNWRITTEN ;
if ( whichfork = = XFS_ATTR_FORK )
rmap - > me_flags | = XFS_RMAP_EXTENT_ATTR_FORK ;
switch ( type ) {
case XFS_RMAP_MAP :
rmap - > me_flags | = XFS_RMAP_EXTENT_MAP ;
break ;
case XFS_RMAP_MAP_SHARED :
rmap - > me_flags | = XFS_RMAP_EXTENT_MAP_SHARED ;
break ;
case XFS_RMAP_UNMAP :
rmap - > me_flags | = XFS_RMAP_EXTENT_UNMAP ;
break ;
case XFS_RMAP_UNMAP_SHARED :
rmap - > me_flags | = XFS_RMAP_EXTENT_UNMAP_SHARED ;
break ;
case XFS_RMAP_CONVERT :
rmap - > me_flags | = XFS_RMAP_EXTENT_CONVERT ;
break ;
case XFS_RMAP_CONVERT_SHARED :
rmap - > me_flags | = XFS_RMAP_EXTENT_CONVERT_SHARED ;
break ;
case XFS_RMAP_ALLOC :
rmap - > me_flags | = XFS_RMAP_EXTENT_ALLOC ;
break ;
case XFS_RMAP_FREE :
rmap - > me_flags | = XFS_RMAP_EXTENT_FREE ;
break ;
default :
ASSERT ( 0 ) ;
}
}
/*
* Finish an rmap update and log it to the RUD . Note that the transaction is
* marked dirty regardless of whether the rmap update succeeds or fails to
* support the RUI / RUD lifecycle rules .
*/
static int
xfs_trans_log_finish_rmap_update (
struct xfs_trans * tp ,
struct xfs_rud_log_item * rudp ,
enum xfs_rmap_intent_type type ,
uint64_t owner ,
int whichfork ,
xfs_fileoff_t startoff ,
xfs_fsblock_t startblock ,
xfs_filblks_t blockcount ,
xfs_exntst_t state ,
struct xfs_btree_cur * * pcur )
{
int error ;
error = xfs_rmap_finish_one ( tp , type , owner , whichfork , startoff ,
startblock , blockcount , state , pcur ) ;
/*
* Mark the transaction dirty , even on error . This ensures the
* transaction is aborted , which :
*
* 1. ) releases the RUI and frees the RUD
* 2. ) shuts down the filesystem
*/
tp - > t_flags | = XFS_TRANS_DIRTY ;
set_bit ( XFS_LI_DIRTY , & rudp - > rud_item . li_flags ) ;
return error ;
}
/* Sort rmap intents by AG. */
static int
xfs_rmap_update_diff_items (
void * priv ,
struct list_head * a ,
struct list_head * b )
{
struct xfs_mount * mp = priv ;
struct xfs_rmap_intent * ra ;
struct xfs_rmap_intent * rb ;
ra = container_of ( a , struct xfs_rmap_intent , ri_list ) ;
rb = container_of ( b , struct xfs_rmap_intent , ri_list ) ;
return XFS_FSB_TO_AGNO ( mp , ra - > ri_bmap . br_startblock ) -
XFS_FSB_TO_AGNO ( mp , rb - > ri_bmap . br_startblock ) ;
}
/* Log rmap updates in the intent item. */
STATIC void
xfs_rmap_update_log_item (
struct xfs_trans * tp ,
2020-04-30 22:52:20 +03:00
struct xfs_rui_log_item * ruip ,
struct xfs_rmap_intent * rmap )
2019-06-29 05:29:41 +03:00
{
uint next_extent ;
struct xfs_map_extent * map ;
tp - > t_flags | = XFS_TRANS_DIRTY ;
set_bit ( XFS_LI_DIRTY , & ruip - > rui_item . li_flags ) ;
/*
* atomic_inc_return gives us the value after the increment ;
* we want to use it as an array index so we need to subtract 1 from
* it .
*/
next_extent = atomic_inc_return ( & ruip - > rui_next_extent ) - 1 ;
ASSERT ( next_extent < ruip - > rui_format . rui_nextents ) ;
map = & ruip - > rui_format . rui_extents [ next_extent ] ;
map - > me_owner = rmap - > ri_owner ;
map - > me_startblock = rmap - > ri_bmap . br_startblock ;
map - > me_startoff = rmap - > ri_bmap . br_startoff ;
map - > me_len = rmap - > ri_bmap . br_blockcount ;
xfs_trans_set_rmap_flags ( map , rmap - > ri_type , rmap - > ri_whichfork ,
rmap - > ri_bmap . br_state ) ;
}
2020-04-30 22:52:21 +03:00
static struct xfs_log_item *
2020-04-30 22:52:20 +03:00
xfs_rmap_update_create_intent (
struct xfs_trans * tp ,
struct list_head * items ,
2020-04-30 22:52:20 +03:00
unsigned int count ,
bool sort )
2020-04-30 22:52:20 +03:00
{
struct xfs_mount * mp = tp - > t_mountp ;
struct xfs_rui_log_item * ruip = xfs_rui_init ( mp , count ) ;
struct xfs_rmap_intent * rmap ;
ASSERT ( count > 0 ) ;
xfs_trans_add_item ( tp , & ruip - > rui_item ) ;
2020-04-30 22:52:20 +03:00
if ( sort )
list_sort ( mp , items , xfs_rmap_update_diff_items ) ;
2020-04-30 22:52:20 +03:00
list_for_each_entry ( rmap , items , ri_list )
xfs_rmap_update_log_item ( tp , ruip , rmap ) ;
2020-04-30 22:52:21 +03:00
return & ruip - > rui_item ;
2020-04-30 22:52:20 +03:00
}
2019-06-29 05:29:41 +03:00
/* Get an RUD so we can process all the deferred rmap updates. */
2020-04-30 22:52:22 +03:00
static struct xfs_log_item *
2019-06-29 05:29:41 +03:00
xfs_rmap_update_create_done (
struct xfs_trans * tp ,
2020-04-30 22:52:21 +03:00
struct xfs_log_item * intent ,
2019-06-29 05:29:41 +03:00
unsigned int count )
{
2020-04-30 22:52:22 +03:00
return & xfs_trans_get_rud ( tp , RUI_ITEM ( intent ) ) - > rud_item ;
2019-06-29 05:29:41 +03:00
}
/* Process a deferred rmap update. */
STATIC int
xfs_rmap_update_finish_item (
struct xfs_trans * tp ,
2020-04-30 22:52:22 +03:00
struct xfs_log_item * done ,
2019-06-29 05:29:41 +03:00
struct list_head * item ,
2020-04-30 22:52:22 +03:00
struct xfs_btree_cur * * state )
2019-06-29 05:29:41 +03:00
{
struct xfs_rmap_intent * rmap ;
int error ;
rmap = container_of ( item , struct xfs_rmap_intent , ri_list ) ;
2020-04-30 22:52:22 +03:00
error = xfs_trans_log_finish_rmap_update ( tp , RUD_ITEM ( done ) ,
2020-04-30 22:52:22 +03:00
rmap - > ri_type , rmap - > ri_owner , rmap - > ri_whichfork ,
rmap - > ri_bmap . br_startoff , rmap - > ri_bmap . br_startblock ,
rmap - > ri_bmap . br_blockcount , rmap - > ri_bmap . br_state ,
state ) ;
2019-06-29 05:29:41 +03:00
kmem_free ( rmap ) ;
return error ;
}
/* Abort all pending RUIs. */
STATIC void
xfs_rmap_update_abort_intent (
2020-04-30 22:52:21 +03:00
struct xfs_log_item * intent )
2019-06-29 05:29:41 +03:00
{
2020-04-30 22:52:21 +03:00
xfs_rui_release ( RUI_ITEM ( intent ) ) ;
2019-06-29 05:29:41 +03:00
}
/* Cancel a deferred rmap update. */
STATIC void
xfs_rmap_update_cancel_item (
struct list_head * item )
{
struct xfs_rmap_intent * rmap ;
rmap = container_of ( item , struct xfs_rmap_intent , ri_list ) ;
kmem_free ( rmap ) ;
}
const struct xfs_defer_op_type xfs_rmap_update_defer_type = {
. max_items = XFS_RUI_MAX_FAST_EXTENTS ,
. create_intent = xfs_rmap_update_create_intent ,
. abort_intent = xfs_rmap_update_abort_intent ,
. create_done = xfs_rmap_update_create_done ,
. finish_item = xfs_rmap_update_finish_item ,
2020-04-30 22:52:22 +03:00
. finish_cleanup = xfs_rmap_finish_one_cleanup ,
2019-06-29 05:29:41 +03:00
. cancel_item = xfs_rmap_update_cancel_item ,
} ;
2016-08-03 05:09:48 +03:00
/*
* Process an rmap update intent item that was recovered from the log .
* We need to update the rmapbt .
*/
2020-05-02 02:00:51 +03:00
STATIC int
2020-05-02 02:00:55 +03:00
xfs_rui_item_recover (
struct xfs_log_item * lip ,
xfs: proper replay of deferred ops queued during log recovery
When we replay unfinished intent items that have been recovered from the
log, it's possible that the replay will cause the creation of more
deferred work items. As outlined in commit 509955823cc9c ("xfs: log
recovery should replay deferred ops in order"), later work items have an
implicit ordering dependency on earlier work items. Therefore, recovery
must replay the items (both recovered and created) in the same order
that they would have been during normal operation.
For log recovery, we enforce this ordering by using an empty transaction
to collect deferred ops that get created in the process of recovering a
log intent item to prevent them from being committed before the rest of
the recovered intent items. After we finish committing all the
recovered log items, we allocate a transaction with an enormous block
reservation, splice our huge list of created deferred ops into that
transaction, and commit it, thereby finishing all those ops.
This is /really/ hokey -- it's the one place in XFS where we allow
nested transactions; the splicing of the defer ops list is is inelegant
and has to be done twice per recovery function; and the broken way we
handle inode pointers and block reservations cause subtle use-after-free
and allocator problems that will be fixed by this patch and the two
patches after it.
Therefore, replace the hokey empty transaction with a structure designed
to capture each chain of deferred ops that are created as part of
recovering a single unfinished log intent. Finally, refactor the loop
that replays those chains to do so using one transaction per chain.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
2020-09-26 03:39:37 +03:00
struct list_head * capture_list )
2016-08-03 05:09:48 +03:00
{
2020-05-02 02:00:55 +03:00
struct xfs_rui_log_item * ruip = RUI_ITEM ( lip ) ;
2016-08-03 05:09:48 +03:00
struct xfs_map_extent * rmap ;
xfs: propagate bmap updates to rmapbt
When we map, unmap, or convert an extent in a file's data or attr
fork, schedule a respective update in the rmapbt. Previous versions
of this patch required a 1:1 correspondence between bmap and rmap,
but this is no longer true as we now have ability to make interval
queries against the rmapbt.
We use the deferred operations code to handle redo operations
atomically and deadlock free. This plumbs in all five rmap actions
(map, unmap, convert extent, alloc, free); we'll use the first three
now for file data, and reflink will want the last two. We also add
an error injection site to test log recovery.
Finally, we need to fix the bmap shift extent code to adjust the
rmaps correctly.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-08-03 05:16:05 +03:00
struct xfs_rud_log_item * rudp ;
struct xfs_trans * tp ;
struct xfs_btree_cur * rcur = NULL ;
xfs: proper replay of deferred ops queued during log recovery
When we replay unfinished intent items that have been recovered from the
log, it's possible that the replay will cause the creation of more
deferred work items. As outlined in commit 509955823cc9c ("xfs: log
recovery should replay deferred ops in order"), later work items have an
implicit ordering dependency on earlier work items. Therefore, recovery
must replay the items (both recovered and created) in the same order
that they would have been during normal operation.
For log recovery, we enforce this ordering by using an empty transaction
to collect deferred ops that get created in the process of recovering a
log intent item to prevent them from being committed before the rest of
the recovered intent items. After we finish committing all the
recovered log items, we allocate a transaction with an enormous block
reservation, splice our huge list of created deferred ops into that
transaction, and commit it, thereby finishing all those ops.
This is /really/ hokey -- it's the one place in XFS where we allow
nested transactions; the splicing of the defer ops list is is inelegant
and has to be done twice per recovery function; and the broken way we
handle inode pointers and block reservations cause subtle use-after-free
and allocator problems that will be fixed by this patch and the two
patches after it.
Therefore, replace the hokey empty transaction with a structure designed
to capture each chain of deferred ops that are created as part of
recovering a single unfinished log intent. Finally, refactor the loop
that replays those chains to do so using one transaction per chain.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
2020-09-26 03:39:37 +03:00
struct xfs_mount * mp = lip - > li_mountp ;
2020-05-02 02:00:55 +03:00
xfs_fsblock_t startblock_fsb ;
enum xfs_rmap_intent_type type ;
xfs_exntst_t state ;
bool op_ok ;
int i ;
int whichfork ;
int error = 0 ;
2016-08-03 05:09:48 +03:00
/*
* First check the validity of the extents described by the
* RUI . If any are bad , then assume that all are bad and
* just toss the RUI .
*/
for ( i = 0 ; i < ruip - > rui_format . rui_nextents ; i + + ) {
2016-08-03 05:29:32 +03:00
rmap = & ruip - > rui_format . rui_extents [ i ] ;
2016-08-03 05:09:48 +03:00
startblock_fsb = XFS_BB_TO_FSB ( mp ,
XFS_FSB_TO_DADDR ( mp , rmap - > me_startblock ) ) ;
switch ( rmap - > me_flags & XFS_RMAP_EXTENT_TYPE_MASK ) {
case XFS_RMAP_EXTENT_MAP :
2016-10-03 19:11:47 +03:00
case XFS_RMAP_EXTENT_MAP_SHARED :
2016-08-03 05:09:48 +03:00
case XFS_RMAP_EXTENT_UNMAP :
2016-10-03 19:11:47 +03:00
case XFS_RMAP_EXTENT_UNMAP_SHARED :
2016-08-03 05:09:48 +03:00
case XFS_RMAP_EXTENT_CONVERT :
2016-10-03 19:11:47 +03:00
case XFS_RMAP_EXTENT_CONVERT_SHARED :
2016-08-03 05:09:48 +03:00
case XFS_RMAP_EXTENT_ALLOC :
case XFS_RMAP_EXTENT_FREE :
op_ok = true ;
break ;
default :
op_ok = false ;
break ;
}
2016-08-03 05:29:32 +03:00
if ( ! op_ok | | startblock_fsb = = 0 | |
rmap - > me_len = = 0 | |
startblock_fsb > = mp - > m_sb . sb_dblocks | |
rmap - > me_len > = mp - > m_sb . sb_agblocks | |
2020-09-21 19:15:10 +03:00
( rmap - > me_flags & ~ XFS_RMAP_EXTENT_FLAGS ) )
2019-11-06 20:17:43 +03:00
return - EFSCORRUPTED ;
2016-08-03 05:09:48 +03:00
}
2018-02-23 01:41:25 +03:00
error = xfs_trans_alloc ( mp , & M_RES ( mp ) - > tr_itruncate ,
mp - > m_rmap_maxlevels , 0 , XFS_TRANS_RESERVE , & tp ) ;
xfs: propagate bmap updates to rmapbt
When we map, unmap, or convert an extent in a file's data or attr
fork, schedule a respective update in the rmapbt. Previous versions
of this patch required a 1:1 correspondence between bmap and rmap,
but this is no longer true as we now have ability to make interval
queries against the rmapbt.
We use the deferred operations code to handle redo operations
atomically and deadlock free. This plumbs in all five rmap actions
(map, unmap, convert extent, alloc, free); we'll use the first three
now for file data, and reflink will want the last two. We also add
an error injection site to test log recovery.
Finally, we need to fix the bmap shift extent code to adjust the
rmaps correctly.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-08-03 05:16:05 +03:00
if ( error )
return error ;
2016-08-03 05:28:43 +03:00
rudp = xfs_trans_get_rud ( tp , ruip ) ;
xfs: propagate bmap updates to rmapbt
When we map, unmap, or convert an extent in a file's data or attr
fork, schedule a respective update in the rmapbt. Previous versions
of this patch required a 1:1 correspondence between bmap and rmap,
but this is no longer true as we now have ability to make interval
queries against the rmapbt.
We use the deferred operations code to handle redo operations
atomically and deadlock free. This plumbs in all five rmap actions
(map, unmap, convert extent, alloc, free); we'll use the first three
now for file data, and reflink will want the last two. We also add
an error injection site to test log recovery.
Finally, we need to fix the bmap shift extent code to adjust the
rmaps correctly.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-08-03 05:16:05 +03:00
for ( i = 0 ; i < ruip - > rui_format . rui_nextents ; i + + ) {
2016-08-03 05:29:32 +03:00
rmap = & ruip - > rui_format . rui_extents [ i ] ;
xfs: propagate bmap updates to rmapbt
When we map, unmap, or convert an extent in a file's data or attr
fork, schedule a respective update in the rmapbt. Previous versions
of this patch required a 1:1 correspondence between bmap and rmap,
but this is no longer true as we now have ability to make interval
queries against the rmapbt.
We use the deferred operations code to handle redo operations
atomically and deadlock free. This plumbs in all five rmap actions
(map, unmap, convert extent, alloc, free); we'll use the first three
now for file data, and reflink will want the last two. We also add
an error injection site to test log recovery.
Finally, we need to fix the bmap shift extent code to adjust the
rmaps correctly.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-08-03 05:16:05 +03:00
state = ( rmap - > me_flags & XFS_RMAP_EXTENT_UNWRITTEN ) ?
XFS_EXT_UNWRITTEN : XFS_EXT_NORM ;
whichfork = ( rmap - > me_flags & XFS_RMAP_EXTENT_ATTR_FORK ) ?
XFS_ATTR_FORK : XFS_DATA_FORK ;
switch ( rmap - > me_flags & XFS_RMAP_EXTENT_TYPE_MASK ) {
case XFS_RMAP_EXTENT_MAP :
type = XFS_RMAP_MAP ;
break ;
2016-10-03 19:11:48 +03:00
case XFS_RMAP_EXTENT_MAP_SHARED :
type = XFS_RMAP_MAP_SHARED ;
break ;
xfs: propagate bmap updates to rmapbt
When we map, unmap, or convert an extent in a file's data or attr
fork, schedule a respective update in the rmapbt. Previous versions
of this patch required a 1:1 correspondence between bmap and rmap,
but this is no longer true as we now have ability to make interval
queries against the rmapbt.
We use the deferred operations code to handle redo operations
atomically and deadlock free. This plumbs in all five rmap actions
(map, unmap, convert extent, alloc, free); we'll use the first three
now for file data, and reflink will want the last two. We also add
an error injection site to test log recovery.
Finally, we need to fix the bmap shift extent code to adjust the
rmaps correctly.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-08-03 05:16:05 +03:00
case XFS_RMAP_EXTENT_UNMAP :
type = XFS_RMAP_UNMAP ;
break ;
2016-10-03 19:11:48 +03:00
case XFS_RMAP_EXTENT_UNMAP_SHARED :
type = XFS_RMAP_UNMAP_SHARED ;
break ;
xfs: propagate bmap updates to rmapbt
When we map, unmap, or convert an extent in a file's data or attr
fork, schedule a respective update in the rmapbt. Previous versions
of this patch required a 1:1 correspondence between bmap and rmap,
but this is no longer true as we now have ability to make interval
queries against the rmapbt.
We use the deferred operations code to handle redo operations
atomically and deadlock free. This plumbs in all five rmap actions
(map, unmap, convert extent, alloc, free); we'll use the first three
now for file data, and reflink will want the last two. We also add
an error injection site to test log recovery.
Finally, we need to fix the bmap shift extent code to adjust the
rmaps correctly.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-08-03 05:16:05 +03:00
case XFS_RMAP_EXTENT_CONVERT :
type = XFS_RMAP_CONVERT ;
break ;
2016-10-03 19:11:48 +03:00
case XFS_RMAP_EXTENT_CONVERT_SHARED :
type = XFS_RMAP_CONVERT_SHARED ;
break ;
xfs: propagate bmap updates to rmapbt
When we map, unmap, or convert an extent in a file's data or attr
fork, schedule a respective update in the rmapbt. Previous versions
of this patch required a 1:1 correspondence between bmap and rmap,
but this is no longer true as we now have ability to make interval
queries against the rmapbt.
We use the deferred operations code to handle redo operations
atomically and deadlock free. This plumbs in all five rmap actions
(map, unmap, convert extent, alloc, free); we'll use the first three
now for file data, and reflink will want the last two. We also add
an error injection site to test log recovery.
Finally, we need to fix the bmap shift extent code to adjust the
rmaps correctly.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-08-03 05:16:05 +03:00
case XFS_RMAP_EXTENT_ALLOC :
type = XFS_RMAP_ALLOC ;
break ;
case XFS_RMAP_EXTENT_FREE :
type = XFS_RMAP_FREE ;
break ;
default :
2019-11-02 19:40:53 +03:00
XFS_ERROR_REPORT ( __func__ , XFS_ERRLEVEL_LOW , NULL ) ;
xfs: propagate bmap updates to rmapbt
When we map, unmap, or convert an extent in a file's data or attr
fork, schedule a respective update in the rmapbt. Previous versions
of this patch required a 1:1 correspondence between bmap and rmap,
but this is no longer true as we now have ability to make interval
queries against the rmapbt.
We use the deferred operations code to handle redo operations
atomically and deadlock free. This plumbs in all five rmap actions
(map, unmap, convert extent, alloc, free); we'll use the first three
now for file data, and reflink will want the last two. We also add
an error injection site to test log recovery.
Finally, we need to fix the bmap shift extent code to adjust the
rmaps correctly.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-08-03 05:16:05 +03:00
error = - EFSCORRUPTED ;
goto abort_error ;
}
error = xfs_trans_log_finish_rmap_update ( tp , rudp , type ,
rmap - > me_owner , whichfork ,
rmap - > me_startoff , rmap - > me_startblock ,
rmap - > me_len , state , & rcur ) ;
if ( error )
goto abort_error ;
}
xfs_rmap_finish_one_cleanup ( tp , rcur , error ) ;
2020-09-26 03:39:51 +03:00
return xfs_defer_ops_capture_and_commit ( tp , NULL , capture_list ) ;
xfs: propagate bmap updates to rmapbt
When we map, unmap, or convert an extent in a file's data or attr
fork, schedule a respective update in the rmapbt. Previous versions
of this patch required a 1:1 correspondence between bmap and rmap,
but this is no longer true as we now have ability to make interval
queries against the rmapbt.
We use the deferred operations code to handle redo operations
atomically and deadlock free. This plumbs in all five rmap actions
(map, unmap, convert extent, alloc, free); we'll use the first three
now for file data, and reflink will want the last two. We also add
an error injection site to test log recovery.
Finally, we need to fix the bmap shift extent code to adjust the
rmaps correctly.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
2016-08-03 05:16:05 +03:00
abort_error :
xfs_rmap_finish_one_cleanup ( tp , rcur , error ) ;
xfs_trans_cancel ( tp ) ;
2016-08-03 05:09:48 +03:00
return error ;
}
2020-05-02 02:00:45 +03:00
2020-05-02 02:00:54 +03:00
STATIC bool
xfs_rui_item_match (
struct xfs_log_item * lip ,
uint64_t intent_id )
{
return RUI_ITEM ( lip ) - > rui_format . rui_id = = intent_id ;
}
xfs: periodically relog deferred intent items
There's a subtle design flaw in the deferred log item code that can lead
to pinning the log tail. Taking up the defer ops chain examples from
the previous commit, we can get trapped in sequences like this:
Caller hands us a transaction t0 with D0-D3 attached. The defer ops
chain will look like the following if the transaction rolls succeed:
t1: D0(t0), D1(t0), D2(t0), D3(t0)
t2: d4(t1), d5(t1), D1(t0), D2(t0), D3(t0)
t3: d5(t1), D1(t0), D2(t0), D3(t0)
...
t9: d9(t7), D3(t0)
t10: D3(t0)
t11: d10(t10), d11(t10)
t12: d11(t10)
In transaction 9, we finish d9 and try to roll to t10 while holding onto
an intent item for D3 that we logged in t0.
The previous commit changed the order in which we place new defer ops in
the defer ops processing chain to reduce the maximum chain length. Now
make xfs_defer_finish_noroll capable of relogging the entire chain
periodically so that we can always move the log tail forward. Most
chains will never get relogged, except for operations that generate very
long chains (large extents containing many blocks with different sharing
levels) or are on filesystems with small logs and a lot of ongoing
metadata updates.
Callers are now required to ensure that the transaction reservation is
large enough to handle logging done items and new intent items for the
maximum possible chain length. Most callers are careful to keep the
chain lengths low, so the overhead should be minimal.
The decision to relog an intent item is made based on whether the intent
was logged in a previous checkpoint, since there's no point in relogging
an intent into the same checkpoint.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Brian Foster <bfoster@redhat.com>
2020-09-28 02:18:13 +03:00
/* Relog an intent item to push the log tail forward. */
static struct xfs_log_item *
xfs_rui_item_relog (
struct xfs_log_item * intent ,
struct xfs_trans * tp )
{
struct xfs_rud_log_item * rudp ;
struct xfs_rui_log_item * ruip ;
struct xfs_map_extent * extp ;
unsigned int count ;
count = RUI_ITEM ( intent ) - > rui_format . rui_nextents ;
extp = RUI_ITEM ( intent ) - > rui_format . rui_extents ;
tp - > t_flags | = XFS_TRANS_DIRTY ;
rudp = xfs_trans_get_rud ( tp , RUI_ITEM ( intent ) ) ;
set_bit ( XFS_LI_DIRTY , & rudp - > rud_item . li_flags ) ;
ruip = xfs_rui_init ( tp - > t_mountp , count ) ;
memcpy ( ruip - > rui_format . rui_extents , extp , count * sizeof ( * extp ) ) ;
atomic_set ( & ruip - > rui_next_extent , count ) ;
xfs_trans_add_item ( tp , & ruip - > rui_item ) ;
set_bit ( XFS_LI_DIRTY , & ruip - > rui_item . li_flags ) ;
return & ruip - > rui_item ;
}
2020-05-02 02:00:51 +03:00
static const struct xfs_item_ops xfs_rui_item_ops = {
. iop_size = xfs_rui_item_size ,
. iop_format = xfs_rui_item_format ,
. iop_unpin = xfs_rui_item_unpin ,
. iop_release = xfs_rui_item_release ,
. iop_recover = xfs_rui_item_recover ,
2020-05-02 02:00:54 +03:00
. iop_match = xfs_rui_item_match ,
xfs: periodically relog deferred intent items
There's a subtle design flaw in the deferred log item code that can lead
to pinning the log tail. Taking up the defer ops chain examples from
the previous commit, we can get trapped in sequences like this:
Caller hands us a transaction t0 with D0-D3 attached. The defer ops
chain will look like the following if the transaction rolls succeed:
t1: D0(t0), D1(t0), D2(t0), D3(t0)
t2: d4(t1), d5(t1), D1(t0), D2(t0), D3(t0)
t3: d5(t1), D1(t0), D2(t0), D3(t0)
...
t9: d9(t7), D3(t0)
t10: D3(t0)
t11: d10(t10), d11(t10)
t12: d11(t10)
In transaction 9, we finish d9 and try to roll to t10 while holding onto
an intent item for D3 that we logged in t0.
The previous commit changed the order in which we place new defer ops in
the defer ops processing chain to reduce the maximum chain length. Now
make xfs_defer_finish_noroll capable of relogging the entire chain
periodically so that we can always move the log tail forward. Most
chains will never get relogged, except for operations that generate very
long chains (large extents containing many blocks with different sharing
levels) or are on filesystems with small logs and a lot of ongoing
metadata updates.
Callers are now required to ensure that the transaction reservation is
large enough to handle logging done items and new intent items for the
maximum possible chain length. Most callers are careful to keep the
chain lengths low, so the overhead should be minimal.
The decision to relog an intent item is made based on whether the intent
was logged in a previous checkpoint, since there's no point in relogging
an intent into the same checkpoint.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Brian Foster <bfoster@redhat.com>
2020-09-28 02:18:13 +03:00
. iop_relog = xfs_rui_item_relog ,
2020-05-02 02:00:51 +03:00
} ;
2020-05-02 02:00:49 +03:00
/*
* This routine is called to create an in - core extent rmap update
* item from the rui format structure which was logged on disk .
* It allocates an in - core rui , copies the extents from the format
* structure into it , and adds the rui to the AIL with the given
* LSN .
*/
STATIC int
xlog_recover_rui_commit_pass2 (
struct xlog * log ,
struct list_head * buffer_list ,
struct xlog_recover_item * item ,
xfs_lsn_t lsn )
{
int error ;
struct xfs_mount * mp = log - > l_mp ;
struct xfs_rui_log_item * ruip ;
struct xfs_rui_log_format * rui_formatp ;
rui_formatp = item - > ri_buf [ 0 ] . i_addr ;
ruip = xfs_rui_init ( mp , rui_formatp - > rui_nextents ) ;
error = xfs_rui_copy_format ( & item - > ri_buf [ 0 ] , & ruip - > rui_format ) ;
if ( error ) {
xfs_rui_item_free ( ruip ) ;
return error ;
}
atomic_set ( & ruip - > rui_next_extent , rui_formatp - > rui_nextents ) ;
/*
2020-05-02 02:00:54 +03:00
* Insert the intent into the AIL directly and drop one reference so
* that finishing or canceling the work will drop the other .
2020-05-02 02:00:49 +03:00
*/
2020-05-02 02:00:54 +03:00
xfs_trans_ail_insert ( log - > l_ailp , & ruip - > rui_item , lsn ) ;
2020-05-02 02:00:49 +03:00
xfs_rui_release ( ruip ) ;
return 0 ;
}
2020-05-02 02:00:45 +03:00
const struct xlog_recover_item_ops xlog_rui_item_ops = {
. item_type = XFS_LI_RUI ,
2020-05-02 02:00:49 +03:00
. commit_pass2 = xlog_recover_rui_commit_pass2 ,
2020-05-02 02:00:45 +03:00
} ;
2020-05-02 02:00:49 +03:00
/*
* This routine is called when an RUD format structure is found in a committed
* transaction in the log . Its purpose is to cancel the corresponding RUI if it
* was still in the log . To do this it searches the AIL for the RUI with an id
* equal to that in the RUD format structure . If we find it we drop the RUD
* reference , which removes the RUI from the AIL and frees it .
*/
STATIC int
xlog_recover_rud_commit_pass2 (
struct xlog * log ,
struct list_head * buffer_list ,
struct xlog_recover_item * item ,
xfs_lsn_t lsn )
{
struct xfs_rud_log_format * rud_formatp ;
rud_formatp = item - > ri_buf [ 0 ] . i_addr ;
ASSERT ( item - > ri_buf [ 0 ] . i_len = = sizeof ( struct xfs_rud_log_format ) ) ;
2020-05-02 02:00:54 +03:00
xlog_recover_release_intent ( log , XFS_LI_RUI , rud_formatp - > rud_rui_id ) ;
2020-05-02 02:00:49 +03:00
return 0 ;
}
2020-05-02 02:00:45 +03:00
const struct xlog_recover_item_ops xlog_rud_item_ops = {
. item_type = XFS_LI_RUD ,
2020-05-02 02:00:49 +03:00
. commit_pass2 = xlog_recover_rud_commit_pass2 ,
2020-05-02 02:00:45 +03:00
} ;