2005-04-17 00:20:36 +02:00
|
|
|
/*
|
2005-11-02 04:58:39 +01:00
|
|
|
* Copyright (c) 2000-2005 Silicon Graphics, Inc.
|
|
|
|
* All Rights Reserved.
|
2005-04-17 00:20:36 +02:00
|
|
|
*
|
2005-11-02 04:58:39 +01:00
|
|
|
* This program is free software; you can redistribute it and/or
|
|
|
|
* modify it under the terms of the GNU General Public License as
|
2005-04-17 00:20:36 +02:00
|
|
|
* published by the Free Software Foundation.
|
|
|
|
*
|
2005-11-02 04:58:39 +01:00
|
|
|
* This program is distributed in the hope that it would be useful,
|
|
|
|
* but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
|
|
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
|
|
|
* GNU General Public License for more details.
|
2005-04-17 00:20:36 +02:00
|
|
|
*
|
2005-11-02 04:58:39 +01:00
|
|
|
* You should have received a copy of the GNU General Public License
|
|
|
|
* along with this program; if not, write the Free Software Foundation,
|
|
|
|
* Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
|
2005-04-17 00:20:36 +02:00
|
|
|
*/
|
|
|
|
#include "xfs.h"
|
2005-11-02 04:38:42 +01:00
|
|
|
#include "xfs_fs.h"
|
2005-04-17 00:20:36 +02:00
|
|
|
#include "xfs_types.h"
|
2005-11-02 04:38:42 +01:00
|
|
|
#include "xfs_bit.h"
|
2005-04-17 00:20:36 +02:00
|
|
|
#include "xfs_log.h"
|
|
|
|
#include "xfs_trans.h"
|
|
|
|
#include "xfs_sb.h"
|
2007-08-28 06:00:13 +02:00
|
|
|
#include "xfs_ag.h"
|
2005-04-17 00:20:36 +02:00
|
|
|
#include "xfs_mount.h"
|
2005-11-02 04:38:42 +01:00
|
|
|
#include "xfs_buf_item.h"
|
2005-04-17 00:20:36 +02:00
|
|
|
#include "xfs_trans_priv.h"
|
|
|
|
#include "xfs_error.h"
|
2009-12-15 00:14:59 +01:00
|
|
|
#include "xfs_trace.h"
|
2005-04-17 00:20:36 +02:00
|
|
|
|
|
|
|
|
|
|
|
kmem_zone_t *xfs_buf_item_zone;
|
|
|
|
|
2010-06-23 10:11:15 +02:00
|
|
|
static inline struct xfs_buf_log_item *BUF_ITEM(struct xfs_log_item *lip)
|
|
|
|
{
|
|
|
|
return container_of(lip, struct xfs_buf_log_item, bli_item);
|
|
|
|
}
|
|
|
|
|
2010-12-03 07:00:52 +01:00
|
|
|
STATIC void xfs_buf_do_callbacks(struct xfs_buf *bp);
|
2005-04-17 00:20:36 +02:00
|
|
|
|
|
|
|
/*
|
|
|
|
* This returns the number of log iovecs needed to log the
|
|
|
|
* given buf log item.
|
|
|
|
*
|
|
|
|
* It calculates this as 1 iovec for the buf log format structure
|
|
|
|
* and 1 for each stretch of non-contiguous chunks to be logged.
|
|
|
|
* Contiguous chunks are logged in a single iovec.
|
|
|
|
*
|
|
|
|
* If the XFS_BLI_STALE flag has been set, then log nothing.
|
|
|
|
*/
|
2005-06-21 07:36:52 +02:00
|
|
|
STATIC uint
|
2012-06-22 10:50:12 +02:00
|
|
|
xfs_buf_item_size_segment(
|
|
|
|
struct xfs_buf_log_item *bip,
|
|
|
|
struct xfs_buf_log_format *blfp)
|
2005-04-17 00:20:36 +02:00
|
|
|
{
|
2010-06-23 10:11:15 +02:00
|
|
|
struct xfs_buf *bp = bip->bli_buf;
|
|
|
|
uint nvecs;
|
|
|
|
int next_bit;
|
|
|
|
int last_bit;
|
2005-04-17 00:20:36 +02:00
|
|
|
|
2012-06-22 10:50:12 +02:00
|
|
|
last_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size, 0);
|
|
|
|
if (last_bit == -1)
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* initial count for a dirty buffer is 2 vectors - the format structure
|
|
|
|
* and the first dirty region.
|
|
|
|
*/
|
|
|
|
nvecs = 2;
|
2005-04-17 00:20:36 +02:00
|
|
|
|
|
|
|
while (last_bit != -1) {
|
|
|
|
/*
|
|
|
|
* This takes the bit number to start looking from and
|
|
|
|
* returns the next set bit from there. It returns -1
|
|
|
|
* if there are no more bits set or the start bit is
|
|
|
|
* beyond the end of the bitmap.
|
|
|
|
*/
|
2012-06-22 10:50:12 +02:00
|
|
|
next_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size,
|
|
|
|
last_bit + 1);
|
2005-04-17 00:20:36 +02:00
|
|
|
/*
|
|
|
|
* If we run out of bits, leave the loop,
|
|
|
|
* else if we find a new set of bits bump the number of vecs,
|
|
|
|
* else keep scanning the current set of bits.
|
|
|
|
*/
|
|
|
|
if (next_bit == -1) {
|
2012-06-22 10:50:12 +02:00
|
|
|
break;
|
2005-04-17 00:20:36 +02:00
|
|
|
} else if (next_bit != last_bit + 1) {
|
|
|
|
last_bit = next_bit;
|
|
|
|
nvecs++;
|
2010-05-07 03:05:19 +02:00
|
|
|
} else if (xfs_buf_offset(bp, next_bit * XFS_BLF_CHUNK) !=
|
|
|
|
(xfs_buf_offset(bp, last_bit * XFS_BLF_CHUNK) +
|
|
|
|
XFS_BLF_CHUNK)) {
|
2005-04-17 00:20:36 +02:00
|
|
|
last_bit = next_bit;
|
|
|
|
nvecs++;
|
|
|
|
} else {
|
|
|
|
last_bit++;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
return nvecs;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
2012-06-22 10:50:12 +02:00
|
|
|
* This returns the number of log iovecs needed to log the given buf log item.
|
|
|
|
*
|
|
|
|
* It calculates this as 1 iovec for the buf log format structure and 1 for each
|
|
|
|
* stretch of non-contiguous chunks to be logged. Contiguous chunks are logged
|
|
|
|
* in a single iovec.
|
|
|
|
*
|
|
|
|
* Discontiguous buffers need a format structure per region that that is being
|
|
|
|
* logged. This makes the changes in the buffer appear to log recovery as though
|
|
|
|
* they came from separate buffers, just like would occur if multiple buffers
|
|
|
|
* were used instead of a single discontiguous buffer. This enables
|
|
|
|
* discontiguous buffers to be in-memory constructs, completely transparent to
|
|
|
|
* what ends up on disk.
|
|
|
|
*
|
|
|
|
* If the XFS_BLI_STALE flag has been set, then log nothing but the buf log
|
|
|
|
* format structures.
|
2005-04-17 00:20:36 +02:00
|
|
|
*/
|
2012-06-22 10:50:12 +02:00
|
|
|
STATIC uint
|
|
|
|
xfs_buf_item_size(
|
|
|
|
struct xfs_log_item *lip)
|
2005-04-17 00:20:36 +02:00
|
|
|
{
|
2010-06-23 10:11:15 +02:00
|
|
|
struct xfs_buf_log_item *bip = BUF_ITEM(lip);
|
2012-06-22 10:50:12 +02:00
|
|
|
uint nvecs;
|
|
|
|
int i;
|
|
|
|
|
|
|
|
ASSERT(atomic_read(&bip->bli_refcount) > 0);
|
|
|
|
if (bip->bli_flags & XFS_BLI_STALE) {
|
|
|
|
/*
|
|
|
|
* The buffer is stale, so all we need to log
|
|
|
|
* is the buf log format structure with the
|
|
|
|
* cancel flag in it.
|
|
|
|
*/
|
|
|
|
trace_xfs_buf_item_size_stale(bip);
|
2012-12-05 00:18:03 +01:00
|
|
|
ASSERT(bip->__bli_format.blf_flags & XFS_BLF_CANCEL);
|
2012-06-22 10:50:12 +02:00
|
|
|
return bip->bli_format_count;
|
|
|
|
}
|
|
|
|
|
|
|
|
ASSERT(bip->bli_flags & XFS_BLI_LOGGED);
|
|
|
|
|
2013-06-27 08:04:52 +02:00
|
|
|
if (bip->bli_flags & XFS_BLI_ORDERED) {
|
|
|
|
/*
|
|
|
|
* The buffer has been logged just to order it.
|
|
|
|
* It is not being included in the transaction
|
|
|
|
* commit, so no vectors are used at all.
|
|
|
|
*/
|
|
|
|
trace_xfs_buf_item_size_ordered(bip);
|
|
|
|
return XFS_LOG_VEC_ORDERED;
|
|
|
|
}
|
|
|
|
|
2012-06-22 10:50:12 +02:00
|
|
|
/*
|
|
|
|
* the vector count is based on the number of buffer vectors we have
|
|
|
|
* dirty bits in. This will only be greater than one when we have a
|
|
|
|
* compound buffer with more than one segment dirty. Hence for compound
|
|
|
|
* buffers we need to track which segment the dirty bits correspond to,
|
|
|
|
* and when we move from one segment to the next increment the vector
|
|
|
|
* count for the extra buf log format structure that will need to be
|
|
|
|
* written.
|
|
|
|
*/
|
|
|
|
nvecs = 0;
|
|
|
|
for (i = 0; i < bip->bli_format_count; i++) {
|
|
|
|
nvecs += xfs_buf_item_size_segment(bip, &bip->bli_formats[i]);
|
|
|
|
}
|
|
|
|
|
|
|
|
trace_xfs_buf_item_size(bip);
|
|
|
|
return nvecs;
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct xfs_log_iovec *
|
|
|
|
xfs_buf_item_format_segment(
|
|
|
|
struct xfs_buf_log_item *bip,
|
|
|
|
struct xfs_log_iovec *vecp,
|
|
|
|
uint offset,
|
|
|
|
struct xfs_buf_log_format *blfp)
|
|
|
|
{
|
2010-06-23 10:11:15 +02:00
|
|
|
struct xfs_buf *bp = bip->bli_buf;
|
2005-04-17 00:20:36 +02:00
|
|
|
uint base_size;
|
|
|
|
uint nvecs;
|
|
|
|
int first_bit;
|
|
|
|
int last_bit;
|
|
|
|
int next_bit;
|
|
|
|
uint nbits;
|
|
|
|
uint buffer_offset;
|
|
|
|
|
2012-06-22 10:50:12 +02:00
|
|
|
/* copy the flags across from the base format item */
|
2012-12-05 00:18:03 +01:00
|
|
|
blfp->blf_flags = bip->__bli_format.blf_flags;
|
2005-04-17 00:20:36 +02:00
|
|
|
|
|
|
|
/*
|
2012-06-22 10:50:07 +02:00
|
|
|
* Base size is the actual size of the ondisk structure - it reflects
|
|
|
|
* the actual size of the dirty bitmap rather than the size of the in
|
|
|
|
* memory structure.
|
2005-04-17 00:20:36 +02:00
|
|
|
*/
|
2012-06-22 10:50:07 +02:00
|
|
|
base_size = offsetof(struct xfs_buf_log_format, blf_data_map) +
|
2012-06-22 10:50:12 +02:00
|
|
|
(blfp->blf_map_size * sizeof(blfp->blf_data_map[0]));
|
2012-12-05 00:18:04 +01:00
|
|
|
|
|
|
|
nvecs = 0;
|
|
|
|
first_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size, 0);
|
|
|
|
if (!(bip->bli_flags & XFS_BLI_STALE) && first_bit == -1) {
|
|
|
|
/*
|
|
|
|
* If the map is not be dirty in the transaction, mark
|
|
|
|
* the size as zero and do not advance the vector pointer.
|
|
|
|
*/
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
2012-06-22 10:50:12 +02:00
|
|
|
vecp->i_addr = blfp;
|
2005-04-17 00:20:36 +02:00
|
|
|
vecp->i_len = base_size;
|
2010-01-19 10:56:45 +01:00
|
|
|
vecp->i_type = XLOG_REG_TYPE_BFORMAT;
|
2005-04-17 00:20:36 +02:00
|
|
|
vecp++;
|
|
|
|
nvecs = 1;
|
|
|
|
|
|
|
|
if (bip->bli_flags & XFS_BLI_STALE) {
|
|
|
|
/*
|
|
|
|
* The buffer is stale, so all we need to log
|
|
|
|
* is the buf log format structure with the
|
|
|
|
* cancel flag in it.
|
|
|
|
*/
|
2009-12-15 00:14:59 +01:00
|
|
|
trace_xfs_buf_item_format_stale(bip);
|
2012-06-22 10:50:12 +02:00
|
|
|
ASSERT(blfp->blf_flags & XFS_BLF_CANCEL);
|
2012-12-05 00:18:04 +01:00
|
|
|
goto out;
|
2005-04-17 00:20:36 +02:00
|
|
|
}
|
|
|
|
|
2013-06-27 08:04:52 +02:00
|
|
|
|
2005-04-17 00:20:36 +02:00
|
|
|
/*
|
|
|
|
* Fill in an iovec for each set of contiguous chunks.
|
|
|
|
*/
|
2012-12-05 00:18:04 +01:00
|
|
|
|
2005-04-17 00:20:36 +02:00
|
|
|
last_bit = first_bit;
|
|
|
|
nbits = 1;
|
|
|
|
for (;;) {
|
|
|
|
/*
|
|
|
|
* This takes the bit number to start looking from and
|
|
|
|
* returns the next set bit from there. It returns -1
|
|
|
|
* if there are no more bits set or the start bit is
|
|
|
|
* beyond the end of the bitmap.
|
|
|
|
*/
|
2012-06-22 10:50:12 +02:00
|
|
|
next_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size,
|
|
|
|
(uint)last_bit + 1);
|
2005-04-17 00:20:36 +02:00
|
|
|
/*
|
|
|
|
* If we run out of bits fill in the last iovec and get
|
|
|
|
* out of the loop.
|
|
|
|
* Else if we start a new set of bits then fill in the
|
|
|
|
* iovec for the series we were looking at and start
|
|
|
|
* counting the bits in the new one.
|
|
|
|
* Else we're still in the same set of bits so just
|
|
|
|
* keep counting and scanning.
|
|
|
|
*/
|
|
|
|
if (next_bit == -1) {
|
2012-06-22 10:50:12 +02:00
|
|
|
buffer_offset = offset + first_bit * XFS_BLF_CHUNK;
|
2005-04-17 00:20:36 +02:00
|
|
|
vecp->i_addr = xfs_buf_offset(bp, buffer_offset);
|
2010-05-07 03:05:19 +02:00
|
|
|
vecp->i_len = nbits * XFS_BLF_CHUNK;
|
2010-01-19 10:56:45 +01:00
|
|
|
vecp->i_type = XLOG_REG_TYPE_BCHUNK;
|
2005-04-17 00:20:36 +02:00
|
|
|
nvecs++;
|
|
|
|
break;
|
|
|
|
} else if (next_bit != last_bit + 1) {
|
2012-06-22 10:50:12 +02:00
|
|
|
buffer_offset = offset + first_bit * XFS_BLF_CHUNK;
|
2005-04-17 00:20:36 +02:00
|
|
|
vecp->i_addr = xfs_buf_offset(bp, buffer_offset);
|
2010-05-07 03:05:19 +02:00
|
|
|
vecp->i_len = nbits * XFS_BLF_CHUNK;
|
2010-01-19 10:56:45 +01:00
|
|
|
vecp->i_type = XLOG_REG_TYPE_BCHUNK;
|
2005-04-17 00:20:36 +02:00
|
|
|
nvecs++;
|
|
|
|
vecp++;
|
|
|
|
first_bit = next_bit;
|
|
|
|
last_bit = next_bit;
|
|
|
|
nbits = 1;
|
2012-06-22 10:50:12 +02:00
|
|
|
} else if (xfs_buf_offset(bp, offset +
|
|
|
|
(next_bit << XFS_BLF_SHIFT)) !=
|
|
|
|
(xfs_buf_offset(bp, offset +
|
|
|
|
(last_bit << XFS_BLF_SHIFT)) +
|
2010-05-07 03:05:19 +02:00
|
|
|
XFS_BLF_CHUNK)) {
|
2012-06-22 10:50:12 +02:00
|
|
|
buffer_offset = offset + first_bit * XFS_BLF_CHUNK;
|
2005-04-17 00:20:36 +02:00
|
|
|
vecp->i_addr = xfs_buf_offset(bp, buffer_offset);
|
2010-05-07 03:05:19 +02:00
|
|
|
vecp->i_len = nbits * XFS_BLF_CHUNK;
|
2010-01-19 10:56:45 +01:00
|
|
|
vecp->i_type = XLOG_REG_TYPE_BCHUNK;
|
xfs: fix split buffer vector log recovery support
A long time ago in a galaxy far away....
.. the was a commit made to fix some ilinux specific "fragmented
buffer" log recovery problem:
http://oss.sgi.com/cgi-bin/gitweb.cgi?p=archive/xfs-import.git;a=commitdiff;h=b29c0bece51da72fb3ff3b61391a391ea54e1603
That problem occurred when a contiguous dirty region of a buffer was
split across across two pages of an unmapped buffer. It's been a
long time since that has been done in XFS, and the changes to log
the entire inode buffers for CRC enabled filesystems has
re-introduced that corner case.
And, of course, it turns out that the above commit didn't actually
fix anything - it just ensured that log recovery is guaranteed to
fail when this situation occurs. And now for the gory details.
xfstest xfs/085 is failing with this assert:
XFS (vdb): bad number of regions (0) in inode log format
XFS: Assertion failed: 0, file: fs/xfs/xfs_log_recover.c, line: 1583
Largely undocumented factoid #1: Log recovery depends on all log
buffer format items starting with this format:
struct foo_log_format {
__uint16_t type;
__uint16_t size;
....
As recoery uses the size field and assumptions about 32 bit
alignment in decoding format items. So don't pay much attention to
the fact log recovery thinks that it decoding an inode log format
item - it just uses them to determine what the size of the item is.
But why would it see a log format item with a zero size? Well,
luckily enough xfs_logprint uses the same code and gives the same
error, so with a bit of gdb magic, it turns out that it isn't a log
format that is being decoded. What logprint tells us is this:
Oper (130): tid: a0375e1a len: 28 clientid: TRANS flags: none
BUF: #regs: 2 start blkno: 144 (0x90) len: 16 bmap size: 2 flags: 0x4000
Oper (131): tid: a0375e1a len: 4096 clientid: TRANS flags: none
BUF DATA
----------------------------------------------------------------------------
Oper (132): tid: a0375e1a len: 4096 clientid: TRANS flags: none
xfs_logprint: unknown log operation type (4e49)
**********************************************************************
* ERROR: data block=2 *
**********************************************************************
That we've got a buffer format item (oper 130) that has two regions;
the format item itself and one dirty region. The subsequent region
after the buffer format item and it's data is them what we are
tripping over, and the first bytes of it at an inode magic number.
Not a log opheader like there is supposed to be.
That means there's a problem with the buffer format item. It's dirty
data region is 4096 bytes, and it contains - you guessed it -
initialised inodes. But inode buffers are 8k, not 4k, and we log
them in their entirety. So something is wrong here. The buffer
format item contains:
(gdb) p /x *(struct xfs_buf_log_format *)in_f
$22 = {blf_type = 0x123c, blf_size = 0x2, blf_flags = 0x4000,
blf_len = 0x10, blf_blkno = 0x90, blf_map_size = 0x2,
blf_data_map = {0xffffffff, 0xffffffff, .... }}
Two regions, and a signle dirty contiguous region of 64 bits. 64 *
128 = 8k, so this should be followed by a single 8k region of data.
And the blf_flags tell us that the type of buffer is a
XFS_BLFT_DINO_BUF. It contains inodes. And because it doesn't have
the XFS_BLF_INODE_BUF flag set, that means it's an inode allocation
buffer. So, it should be followed by 8k of inode data.
But we know that the next region has a header of:
(gdb) p /x *ohead
$25 = {oh_tid = 0x1a5e37a0, oh_len = 0x100000, oh_clientid = 0x69,
oh_flags = 0x0, oh_res2 = 0x0}
and so be32_to_cpu(oh_len) = 0x1000 = 4096 bytes. It's simply not
long enough to hold all the logged data. There must be another
region. There is - there's a following opheader for another 4k of
data that contains the other half of the inode cluster data - the
one we assert fail on because it's not a log format header.
So why is the second part of the data not being accounted to the
correct buffer log format structure? It took a little more work with
gdb to work out that the buffer log format structure was both
expecting it to be there but hadn't accounted for it. It was at that
point I went to the kernel code, as clearly this wasn't a bug in
xfs_logprint and the kernel was writing bad stuff to the log.
First port of call was the buffer item formatting code, and the
discontiguous memory/contiguous dirty region handling code
immediately stood out. I've wondered for a long time why the code
had this comment in it:
vecp->i_addr = xfs_buf_offset(bp, buffer_offset);
vecp->i_len = nbits * XFS_BLF_CHUNK;
vecp->i_type = XLOG_REG_TYPE_BCHUNK;
/*
* You would think we need to bump the nvecs here too, but we do not
* this number is used by recovery, and it gets confused by the boundary
* split here
* nvecs++;
*/
vecp++;
And it didn't account for the extra vector pointer. The case being
handled here is that a contiguous dirty region lies across a
boundary that cannot be memcpy()d across, and so has to be split
into two separate operations for xlog_write() to perform.
What this code assumes is that what is written to the log is two
consecutive blocks of data that are accounted in the buf log format
item as the same contiguous dirty region and so will get decoded as
such by the log recovery code.
The thing is, xlog_write() knows nothing about this, and so just
does it's normal thing of adding an opheader for each vector. That
means the 8k region gets written to the log as two separate regions
of 4k each, but because nvecs has not been incremented, the buf log
format item accounts for only one of them.
Hence when we come to log recovery, we process the first 4k region
and then expect to come across a new item that starts with a log
format structure of some kind that tells us whenteh next data is
going to be. Instead, we hit raw buffer data and things go bad real
quick.
So, the commit from 2002 that commented out nvecs++ is just plain
wrong. It breaks log recovery completely, and it would seem the only
reason this hasn't been since then is that we don't log large
contigous regions of multi-page unmapped buffers very often. Never
would be a closer estimate, at least until the CRC code came along....
So, lets fix that by restoring the nvecs accounting for the extra
region when we hit this case.....
.... and there's the problemin log recovery it is apparently working
around:
XFS: Assertion failed: i == item->ri_total, file: fs/xfs/xfs_log_recover.c, line: 2135
Yup, xlog_recover_do_reg_buffer() doesn't handle contigous dirty
regions being broken up into multiple regions by the log formatting
code. That's an easy fix, though - if the number of contiguous dirty
bits exceeds the length of the region being copied out of the log,
only account for the number of dirty bits that region covers, and
then loop again and copy more from the next region. It's a 2 line
fix.
Now xfstests xfs/085 passes, we have one less piece of mystery
code, and one more important piece of knowledge about how to
structure new log format items..
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Mark Tinguely <tinguely@sgi.com>
Signed-off-by: Ben Myers <bpm@sgi.com>
2013-05-27 08:38:23 +02:00
|
|
|
nvecs++;
|
2005-04-17 00:20:36 +02:00
|
|
|
vecp++;
|
|
|
|
first_bit = next_bit;
|
|
|
|
last_bit = next_bit;
|
|
|
|
nbits = 1;
|
|
|
|
} else {
|
|
|
|
last_bit++;
|
|
|
|
nbits++;
|
|
|
|
}
|
|
|
|
}
|
2012-12-05 00:18:04 +01:00
|
|
|
out:
|
|
|
|
blfp->blf_size = nvecs;
|
2012-06-22 10:50:12 +02:00
|
|
|
return vecp;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* This is called to fill in the vector of log iovecs for the
|
|
|
|
* given log buf item. It fills the first entry with a buf log
|
|
|
|
* format structure, and the rest point to contiguous chunks
|
|
|
|
* within the buffer.
|
|
|
|
*/
|
|
|
|
STATIC void
|
|
|
|
xfs_buf_item_format(
|
|
|
|
struct xfs_log_item *lip,
|
|
|
|
struct xfs_log_iovec *vecp)
|
|
|
|
{
|
|
|
|
struct xfs_buf_log_item *bip = BUF_ITEM(lip);
|
|
|
|
struct xfs_buf *bp = bip->bli_buf;
|
|
|
|
uint offset = 0;
|
|
|
|
int i;
|
|
|
|
|
|
|
|
ASSERT(atomic_read(&bip->bli_refcount) > 0);
|
|
|
|
ASSERT((bip->bli_flags & XFS_BLI_LOGGED) ||
|
|
|
|
(bip->bli_flags & XFS_BLI_STALE));
|
|
|
|
|
|
|
|
/*
|
|
|
|
* If it is an inode buffer, transfer the in-memory state to the
|
2013-06-27 08:04:56 +02:00
|
|
|
* format flags and clear the in-memory state.
|
|
|
|
*
|
|
|
|
* For buffer based inode allocation, we do not transfer
|
2012-06-22 10:50:12 +02:00
|
|
|
* this state if the inode buffer allocation has not yet been committed
|
|
|
|
* to the log as setting the XFS_BLI_INODE_BUF flag will prevent
|
|
|
|
* correct replay of the inode allocation.
|
2013-06-27 08:04:56 +02:00
|
|
|
*
|
|
|
|
* For icreate item based inode allocation, the buffers aren't written
|
|
|
|
* to the journal during allocation, and hence we should always tag the
|
|
|
|
* buffer as an inode buffer so that the correct unlinked list replay
|
|
|
|
* occurs during recovery.
|
2012-06-22 10:50:12 +02:00
|
|
|
*/
|
|
|
|
if (bip->bli_flags & XFS_BLI_INODE_BUF) {
|
2013-06-27 08:04:56 +02:00
|
|
|
if (xfs_sb_version_hascrc(&lip->li_mountp->m_sb) ||
|
|
|
|
!((bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF) &&
|
2012-06-22 10:50:12 +02:00
|
|
|
xfs_log_item_in_current_chkpt(lip)))
|
2012-12-05 00:18:03 +01:00
|
|
|
bip->__bli_format.blf_flags |= XFS_BLF_INODE_BUF;
|
2012-06-22 10:50:12 +02:00
|
|
|
bip->bli_flags &= ~XFS_BLI_INODE_BUF;
|
|
|
|
}
|
|
|
|
|
2013-06-27 08:04:52 +02:00
|
|
|
if ((bip->bli_flags & (XFS_BLI_ORDERED|XFS_BLI_STALE)) ==
|
|
|
|
XFS_BLI_ORDERED) {
|
|
|
|
/*
|
|
|
|
* The buffer has been logged just to order it. It is not being
|
|
|
|
* included in the transaction commit, so don't format it.
|
|
|
|
*/
|
|
|
|
trace_xfs_buf_item_format_ordered(bip);
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
2012-06-22 10:50:12 +02:00
|
|
|
for (i = 0; i < bip->bli_format_count; i++) {
|
|
|
|
vecp = xfs_buf_item_format_segment(bip, vecp, offset,
|
|
|
|
&bip->bli_formats[i]);
|
|
|
|
offset += bp->b_maps[i].bm_len;
|
|
|
|
}
|
2005-04-17 00:20:36 +02:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Check to make sure everything is consistent.
|
|
|
|
*/
|
2009-12-15 00:14:59 +01:00
|
|
|
trace_xfs_buf_item_format(bip);
|
2005-04-17 00:20:36 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
2010-05-07 03:04:34 +02:00
|
|
|
* This is called to pin the buffer associated with the buf log item in memory
|
2010-06-23 10:11:15 +02:00
|
|
|
* so it cannot be written out.
|
2010-05-07 03:04:34 +02:00
|
|
|
*
|
|
|
|
* We also always take a reference to the buffer log item here so that the bli
|
|
|
|
* is held while the item is pinned in memory. This means that we can
|
|
|
|
* unconditionally drop the reference count a transaction holds when the
|
|
|
|
* transaction is completed.
|
2005-04-17 00:20:36 +02:00
|
|
|
*/
|
2005-06-21 07:36:52 +02:00
|
|
|
STATIC void
|
2005-04-17 00:20:36 +02:00
|
|
|
xfs_buf_item_pin(
|
2010-06-23 10:11:15 +02:00
|
|
|
struct xfs_log_item *lip)
|
2005-04-17 00:20:36 +02:00
|
|
|
{
|
2010-06-23 10:11:15 +02:00
|
|
|
struct xfs_buf_log_item *bip = BUF_ITEM(lip);
|
2005-04-17 00:20:36 +02:00
|
|
|
|
|
|
|
ASSERT(atomic_read(&bip->bli_refcount) > 0);
|
|
|
|
ASSERT((bip->bli_flags & XFS_BLI_LOGGED) ||
|
2013-06-27 08:04:52 +02:00
|
|
|
(bip->bli_flags & XFS_BLI_ORDERED) ||
|
2005-04-17 00:20:36 +02:00
|
|
|
(bip->bli_flags & XFS_BLI_STALE));
|
2010-06-23 10:11:15 +02:00
|
|
|
|
2009-12-15 00:14:59 +01:00
|
|
|
trace_xfs_buf_item_pin(bip);
|
2010-06-23 10:11:15 +02:00
|
|
|
|
|
|
|
atomic_inc(&bip->bli_refcount);
|
|
|
|
atomic_inc(&bip->bli_buf->b_pin_count);
|
2005-04-17 00:20:36 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* This is called to unpin the buffer associated with the buf log
|
|
|
|
* item which was previously pinned with a call to xfs_buf_item_pin().
|
|
|
|
*
|
|
|
|
* Also drop the reference to the buf item for the current transaction.
|
|
|
|
* If the XFS_BLI_STALE flag is set and we are the last reference,
|
|
|
|
* then free up the buf log item and unlock the buffer.
|
2010-06-23 10:11:15 +02:00
|
|
|
*
|
|
|
|
* If the remove flag is set we are called from uncommit in the
|
|
|
|
* forced-shutdown path. If that is true and the reference count on
|
|
|
|
* the log item is going to drop to zero we need to free the item's
|
|
|
|
* descriptor in the transaction.
|
2005-04-17 00:20:36 +02:00
|
|
|
*/
|
2005-06-21 07:36:52 +02:00
|
|
|
STATIC void
|
2005-04-17 00:20:36 +02:00
|
|
|
xfs_buf_item_unpin(
|
2010-06-23 10:11:15 +02:00
|
|
|
struct xfs_log_item *lip,
|
2010-06-23 10:11:15 +02:00
|
|
|
int remove)
|
2005-04-17 00:20:36 +02:00
|
|
|
{
|
2010-06-23 10:11:15 +02:00
|
|
|
struct xfs_buf_log_item *bip = BUF_ITEM(lip);
|
2010-06-23 10:11:15 +02:00
|
|
|
xfs_buf_t *bp = bip->bli_buf;
|
2010-06-23 10:11:15 +02:00
|
|
|
struct xfs_ail *ailp = lip->li_ailp;
|
2010-03-08 01:26:03 +01:00
|
|
|
int stale = bip->bli_flags & XFS_BLI_STALE;
|
2010-06-23 10:11:15 +02:00
|
|
|
int freed;
|
2005-04-17 00:20:36 +02:00
|
|
|
|
2011-07-13 13:43:49 +02:00
|
|
|
ASSERT(bp->b_fspriv == bip);
|
2005-04-17 00:20:36 +02:00
|
|
|
ASSERT(atomic_read(&bip->bli_refcount) > 0);
|
2010-06-23 10:11:15 +02:00
|
|
|
|
2009-12-15 00:14:59 +01:00
|
|
|
trace_xfs_buf_item_unpin(bip);
|
2005-04-17 00:20:36 +02:00
|
|
|
|
|
|
|
freed = atomic_dec_and_test(&bip->bli_refcount);
|
2010-06-23 10:11:15 +02:00
|
|
|
|
|
|
|
if (atomic_dec_and_test(&bp->b_pin_count))
|
|
|
|
wake_up_all(&bp->b_waiters);
|
2010-06-23 10:11:15 +02:00
|
|
|
|
2005-04-17 00:20:36 +02:00
|
|
|
if (freed && stale) {
|
|
|
|
ASSERT(bip->bli_flags & XFS_BLI_STALE);
|
2011-07-08 14:36:19 +02:00
|
|
|
ASSERT(xfs_buf_islocked(bp));
|
2005-04-17 00:20:36 +02:00
|
|
|
ASSERT(XFS_BUF_ISSTALE(bp));
|
2012-12-05 00:18:03 +01:00
|
|
|
ASSERT(bip->__bli_format.blf_flags & XFS_BLF_CANCEL);
|
2010-06-23 10:11:15 +02:00
|
|
|
|
2009-12-15 00:14:59 +01:00
|
|
|
trace_xfs_buf_item_unpin_stale(bip);
|
|
|
|
|
2010-06-23 10:11:15 +02:00
|
|
|
if (remove) {
|
|
|
|
/*
|
2011-01-27 02:13:35 +01:00
|
|
|
* If we are in a transaction context, we have to
|
|
|
|
* remove the log item from the transaction as we are
|
|
|
|
* about to release our reference to the buffer. If we
|
|
|
|
* don't, the unlock that occurs later in
|
|
|
|
* xfs_trans_uncommit() will try to reference the
|
2010-06-23 10:11:15 +02:00
|
|
|
* buffer which we no longer have a hold on.
|
|
|
|
*/
|
2011-01-27 02:13:35 +01:00
|
|
|
if (lip->li_desc)
|
|
|
|
xfs_trans_del_item(lip);
|
2010-06-23 10:11:15 +02:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Since the transaction no longer refers to the buffer,
|
|
|
|
* the buffer should no longer refer to the transaction.
|
|
|
|
*/
|
2011-07-13 13:43:49 +02:00
|
|
|
bp->b_transp = NULL;
|
2010-06-23 10:11:15 +02:00
|
|
|
}
|
|
|
|
|
2005-04-17 00:20:36 +02:00
|
|
|
/*
|
|
|
|
* If we get called here because of an IO error, we may
|
2008-10-30 07:39:58 +01:00
|
|
|
* or may not have the item on the AIL. xfs_trans_ail_delete()
|
2005-04-17 00:20:36 +02:00
|
|
|
* will take care of that situation.
|
2008-10-30 07:39:58 +01:00
|
|
|
* xfs_trans_ail_delete() drops the AIL lock.
|
2005-04-17 00:20:36 +02:00
|
|
|
*/
|
|
|
|
if (bip->bli_flags & XFS_BLI_STALE_INODE) {
|
2010-12-03 07:00:52 +01:00
|
|
|
xfs_buf_do_callbacks(bp);
|
2011-07-13 13:43:49 +02:00
|
|
|
bp->b_fspriv = NULL;
|
2011-07-13 13:43:49 +02:00
|
|
|
bp->b_iodone = NULL;
|
2005-04-17 00:20:36 +02:00
|
|
|
} else {
|
2008-10-30 07:39:58 +01:00
|
|
|
spin_lock(&ailp->xa_lock);
|
2012-04-23 07:58:41 +02:00
|
|
|
xfs_trans_ail_delete(ailp, lip, SHUTDOWN_LOG_IO_ERROR);
|
2005-04-17 00:20:36 +02:00
|
|
|
xfs_buf_item_relse(bp);
|
2011-07-13 13:43:49 +02:00
|
|
|
ASSERT(bp->b_fspriv == NULL);
|
2005-04-17 00:20:36 +02:00
|
|
|
}
|
|
|
|
xfs_buf_relse(bp);
|
2012-04-23 07:58:38 +02:00
|
|
|
} else if (freed && remove) {
|
2012-11-02 04:23:12 +01:00
|
|
|
/*
|
|
|
|
* There are currently two references to the buffer - the active
|
|
|
|
* LRU reference and the buf log item. What we are about to do
|
|
|
|
* here - simulate a failed IO completion - requires 3
|
|
|
|
* references.
|
|
|
|
*
|
|
|
|
* The LRU reference is removed by the xfs_buf_stale() call. The
|
|
|
|
* buf item reference is removed by the xfs_buf_iodone()
|
|
|
|
* callback that is run by xfs_buf_do_callbacks() during ioend
|
|
|
|
* processing (via the bp->b_iodone callback), and then finally
|
|
|
|
* the ioend processing will drop the IO reference if the buffer
|
|
|
|
* is marked XBF_ASYNC.
|
|
|
|
*
|
|
|
|
* Hence we need to take an additional reference here so that IO
|
|
|
|
* completion processing doesn't free the buffer prematurely.
|
|
|
|
*/
|
2012-04-23 07:58:38 +02:00
|
|
|
xfs_buf_lock(bp);
|
2012-11-02 04:23:12 +01:00
|
|
|
xfs_buf_hold(bp);
|
|
|
|
bp->b_flags |= XBF_ASYNC;
|
2012-04-23 07:58:38 +02:00
|
|
|
xfs_buf_ioerror(bp, EIO);
|
|
|
|
XFS_BUF_UNDONE(bp);
|
|
|
|
xfs_buf_stale(bp);
|
|
|
|
xfs_buf_ioend(bp, 0);
|
2005-04-17 00:20:36 +02:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2005-06-21 07:36:52 +02:00
|
|
|
STATIC uint
|
xfs: on-stack delayed write buffer lists
Queue delwri buffers on a local on-stack list instead of a per-buftarg one,
and write back the buffers per-process instead of by waking up xfsbufd.
This is now easily doable given that we have very few places left that write
delwri buffers:
- log recovery:
Only done at mount time, and already forcing out the buffers
synchronously using xfs_flush_buftarg
- quotacheck:
Same story.
- dquot reclaim:
Writes out dirty dquots on the LRU under memory pressure. We might
want to look into doing more of this via xfsaild, but it's already
more optimal than the synchronous inode reclaim that writes each
buffer synchronously.
- xfsaild:
This is the main beneficiary of the change. By keeping a local list
of buffers to write we reduce latency of writing out buffers, and
more importably we can remove all the delwri list promotions which
were hitting the buffer cache hard under sustained metadata loads.
The implementation is very straight forward - xfs_buf_delwri_queue now gets
a new list_head pointer that it adds the delwri buffers to, and all callers
need to eventually submit the list using xfs_buf_delwi_submit or
xfs_buf_delwi_submit_nowait. Buffers that already are on a delwri list are
skipped in xfs_buf_delwri_queue, assuming they already are on another delwri
list. The biggest change to pass down the buffer list was done to the AIL
pushing. Now that we operate on buffers the trylock, push and pushbuf log
item methods are merged into a single push routine, which tries to lock the
item, and if possible add the buffer that needs writeback to the buffer list.
This leads to much simpler code than the previous split but requires the
individual IOP_PUSH instances to unlock and reacquire the AIL around calls
to blocking routines.
Given that xfsailds now also handle writing out buffers, the conditions for
log forcing and the sleep times needed some small changes. The most
important one is that we consider an AIL busy as long we still have buffers
to push, and the other one is that we do increment the pushed LSN for
buffers that are under flushing at this moment, but still count them towards
the stuck items for restart purposes. Without this we could hammer on stuck
items without ever forcing the log and not make progress under heavy random
delete workloads on fast flash storage devices.
[ Dave Chinner:
- rebase on previous patches.
- improved comments for XBF_DELWRI_Q handling
- fix XBF_ASYNC handling in queue submission (test 106 failure)
- rename delwri submit function buffer list parameters for clarity
- xfs_efd_item_push() should return XFS_ITEM_PINNED ]
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Mark Tinguely <tinguely@sgi.com>
Signed-off-by: Ben Myers <bpm@sgi.com>
2012-04-23 07:58:39 +02:00
|
|
|
xfs_buf_item_push(
|
|
|
|
struct xfs_log_item *lip,
|
|
|
|
struct list_head *buffer_list)
|
2005-04-17 00:20:36 +02:00
|
|
|
{
|
2010-06-23 10:11:15 +02:00
|
|
|
struct xfs_buf_log_item *bip = BUF_ITEM(lip);
|
|
|
|
struct xfs_buf *bp = bip->bli_buf;
|
xfs: on-stack delayed write buffer lists
Queue delwri buffers on a local on-stack list instead of a per-buftarg one,
and write back the buffers per-process instead of by waking up xfsbufd.
This is now easily doable given that we have very few places left that write
delwri buffers:
- log recovery:
Only done at mount time, and already forcing out the buffers
synchronously using xfs_flush_buftarg
- quotacheck:
Same story.
- dquot reclaim:
Writes out dirty dquots on the LRU under memory pressure. We might
want to look into doing more of this via xfsaild, but it's already
more optimal than the synchronous inode reclaim that writes each
buffer synchronously.
- xfsaild:
This is the main beneficiary of the change. By keeping a local list
of buffers to write we reduce latency of writing out buffers, and
more importably we can remove all the delwri list promotions which
were hitting the buffer cache hard under sustained metadata loads.
The implementation is very straight forward - xfs_buf_delwri_queue now gets
a new list_head pointer that it adds the delwri buffers to, and all callers
need to eventually submit the list using xfs_buf_delwi_submit or
xfs_buf_delwi_submit_nowait. Buffers that already are on a delwri list are
skipped in xfs_buf_delwri_queue, assuming they already are on another delwri
list. The biggest change to pass down the buffer list was done to the AIL
pushing. Now that we operate on buffers the trylock, push and pushbuf log
item methods are merged into a single push routine, which tries to lock the
item, and if possible add the buffer that needs writeback to the buffer list.
This leads to much simpler code than the previous split but requires the
individual IOP_PUSH instances to unlock and reacquire the AIL around calls
to blocking routines.
Given that xfsailds now also handle writing out buffers, the conditions for
log forcing and the sleep times needed some small changes. The most
important one is that we consider an AIL busy as long we still have buffers
to push, and the other one is that we do increment the pushed LSN for
buffers that are under flushing at this moment, but still count them towards
the stuck items for restart purposes. Without this we could hammer on stuck
items without ever forcing the log and not make progress under heavy random
delete workloads on fast flash storage devices.
[ Dave Chinner:
- rebase on previous patches.
- improved comments for XBF_DELWRI_Q handling
- fix XBF_ASYNC handling in queue submission (test 106 failure)
- rename delwri submit function buffer list parameters for clarity
- xfs_efd_item_push() should return XFS_ITEM_PINNED ]
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Mark Tinguely <tinguely@sgi.com>
Signed-off-by: Ben Myers <bpm@sgi.com>
2012-04-23 07:58:39 +02:00
|
|
|
uint rval = XFS_ITEM_SUCCESS;
|
2005-04-17 00:20:36 +02:00
|
|
|
|
2011-07-23 01:40:27 +02:00
|
|
|
if (xfs_buf_ispinned(bp))
|
2005-04-17 00:20:36 +02:00
|
|
|
return XFS_ITEM_PINNED;
|
2013-02-11 16:08:21 +01:00
|
|
|
if (!xfs_buf_trylock(bp)) {
|
|
|
|
/*
|
|
|
|
* If we have just raced with a buffer being pinned and it has
|
|
|
|
* been marked stale, we could end up stalling until someone else
|
|
|
|
* issues a log force to unpin the stale buffer. Check for the
|
|
|
|
* race condition here so xfsaild recognizes the buffer is pinned
|
|
|
|
* and queues a log force to move it along.
|
|
|
|
*/
|
|
|
|
if (xfs_buf_ispinned(bp))
|
|
|
|
return XFS_ITEM_PINNED;
|
2005-04-17 00:20:36 +02:00
|
|
|
return XFS_ITEM_LOCKED;
|
2013-02-11 16:08:21 +01:00
|
|
|
}
|
2005-04-17 00:20:36 +02:00
|
|
|
|
|
|
|
ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
|
xfs: on-stack delayed write buffer lists
Queue delwri buffers on a local on-stack list instead of a per-buftarg one,
and write back the buffers per-process instead of by waking up xfsbufd.
This is now easily doable given that we have very few places left that write
delwri buffers:
- log recovery:
Only done at mount time, and already forcing out the buffers
synchronously using xfs_flush_buftarg
- quotacheck:
Same story.
- dquot reclaim:
Writes out dirty dquots on the LRU under memory pressure. We might
want to look into doing more of this via xfsaild, but it's already
more optimal than the synchronous inode reclaim that writes each
buffer synchronously.
- xfsaild:
This is the main beneficiary of the change. By keeping a local list
of buffers to write we reduce latency of writing out buffers, and
more importably we can remove all the delwri list promotions which
were hitting the buffer cache hard under sustained metadata loads.
The implementation is very straight forward - xfs_buf_delwri_queue now gets
a new list_head pointer that it adds the delwri buffers to, and all callers
need to eventually submit the list using xfs_buf_delwi_submit or
xfs_buf_delwi_submit_nowait. Buffers that already are on a delwri list are
skipped in xfs_buf_delwri_queue, assuming they already are on another delwri
list. The biggest change to pass down the buffer list was done to the AIL
pushing. Now that we operate on buffers the trylock, push and pushbuf log
item methods are merged into a single push routine, which tries to lock the
item, and if possible add the buffer that needs writeback to the buffer list.
This leads to much simpler code than the previous split but requires the
individual IOP_PUSH instances to unlock and reacquire the AIL around calls
to blocking routines.
Given that xfsailds now also handle writing out buffers, the conditions for
log forcing and the sleep times needed some small changes. The most
important one is that we consider an AIL busy as long we still have buffers
to push, and the other one is that we do increment the pushed LSN for
buffers that are under flushing at this moment, but still count them towards
the stuck items for restart purposes. Without this we could hammer on stuck
items without ever forcing the log and not make progress under heavy random
delete workloads on fast flash storage devices.
[ Dave Chinner:
- rebase on previous patches.
- improved comments for XBF_DELWRI_Q handling
- fix XBF_ASYNC handling in queue submission (test 106 failure)
- rename delwri submit function buffer list parameters for clarity
- xfs_efd_item_push() should return XFS_ITEM_PINNED ]
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Mark Tinguely <tinguely@sgi.com>
Signed-off-by: Ben Myers <bpm@sgi.com>
2012-04-23 07:58:39 +02:00
|
|
|
|
|
|
|
trace_xfs_buf_item_push(bip);
|
|
|
|
|
|
|
|
if (!xfs_buf_delwri_queue(bp, buffer_list))
|
|
|
|
rval = XFS_ITEM_FLUSHING;
|
|
|
|
xfs_buf_unlock(bp);
|
|
|
|
return rval;
|
2005-04-17 00:20:36 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
2010-05-07 03:04:34 +02:00
|
|
|
* Release the buffer associated with the buf log item. If there is no dirty
|
|
|
|
* logged data associated with the buffer recorded in the buf log item, then
|
|
|
|
* free the buf log item and remove the reference to it in the buffer.
|
2005-04-17 00:20:36 +02:00
|
|
|
*
|
2010-05-07 03:04:34 +02:00
|
|
|
* This call ignores the recursion count. It is only called when the buffer
|
|
|
|
* should REALLY be unlocked, regardless of the recursion count.
|
2005-04-17 00:20:36 +02:00
|
|
|
*
|
2010-05-07 03:04:34 +02:00
|
|
|
* We unconditionally drop the transaction's reference to the log item. If the
|
|
|
|
* item was logged, then another reference was taken when it was pinned, so we
|
|
|
|
* can safely drop the transaction reference now. This also allows us to avoid
|
|
|
|
* potential races with the unpin code freeing the bli by not referencing the
|
|
|
|
* bli after we've dropped the reference count.
|
|
|
|
*
|
|
|
|
* If the XFS_BLI_HOLD flag is set in the buf log item, then free the log item
|
|
|
|
* if necessary but do not unlock the buffer. This is for support of
|
|
|
|
* xfs_trans_bhold(). Make sure the XFS_BLI_HOLD field is cleared if we don't
|
|
|
|
* free the item.
|
2005-04-17 00:20:36 +02:00
|
|
|
*/
|
2005-06-21 07:36:52 +02:00
|
|
|
STATIC void
|
2005-04-17 00:20:36 +02:00
|
|
|
xfs_buf_item_unlock(
|
2010-06-23 10:11:15 +02:00
|
|
|
struct xfs_log_item *lip)
|
2005-04-17 00:20:36 +02:00
|
|
|
{
|
2010-06-23 10:11:15 +02:00
|
|
|
struct xfs_buf_log_item *bip = BUF_ITEM(lip);
|
|
|
|
struct xfs_buf *bp = bip->bli_buf;
|
2013-06-27 08:04:52 +02:00
|
|
|
bool clean;
|
|
|
|
bool aborted;
|
|
|
|
int flags;
|
2005-04-17 00:20:36 +02:00
|
|
|
|
2010-05-07 03:04:34 +02:00
|
|
|
/* Clear the buffer's association with this transaction. */
|
2011-07-13 13:43:49 +02:00
|
|
|
bp->b_transp = NULL;
|
2005-04-17 00:20:36 +02:00
|
|
|
|
|
|
|
/*
|
2010-05-07 03:04:34 +02:00
|
|
|
* If this is a transaction abort, don't return early. Instead, allow
|
|
|
|
* the brelse to happen. Normally it would be done for stale
|
|
|
|
* (cancelled) buffers at unpin time, but we'll never go through the
|
|
|
|
* pin/unpin cycle if we abort inside commit.
|
2005-04-17 00:20:36 +02:00
|
|
|
*/
|
2013-06-27 08:04:52 +02:00
|
|
|
aborted = (lip->li_flags & XFS_LI_ABORTED) ? true : false;
|
2005-04-17 00:20:36 +02:00
|
|
|
/*
|
2013-06-27 08:04:52 +02:00
|
|
|
* Before possibly freeing the buf item, copy the per-transaction state
|
|
|
|
* so we can reference it safely later after clearing it from the
|
|
|
|
* buffer log item.
|
2010-05-07 03:04:34 +02:00
|
|
|
*/
|
2013-06-27 08:04:52 +02:00
|
|
|
flags = bip->bli_flags;
|
|
|
|
bip->bli_flags &= ~(XFS_BLI_LOGGED | XFS_BLI_HOLD | XFS_BLI_ORDERED);
|
2010-05-07 03:04:34 +02:00
|
|
|
|
|
|
|
/*
|
|
|
|
* If the buf item is marked stale, then don't do anything. We'll
|
|
|
|
* unlock the buffer and free the buf item when the buffer is unpinned
|
|
|
|
* for the last time.
|
2005-04-17 00:20:36 +02:00
|
|
|
*/
|
2013-06-27 08:04:52 +02:00
|
|
|
if (flags & XFS_BLI_STALE) {
|
2009-12-15 00:14:59 +01:00
|
|
|
trace_xfs_buf_item_unlock_stale(bip);
|
2012-12-05 00:18:03 +01:00
|
|
|
ASSERT(bip->__bli_format.blf_flags & XFS_BLF_CANCEL);
|
2010-05-07 03:04:34 +02:00
|
|
|
if (!aborted) {
|
|
|
|
atomic_dec(&bip->bli_refcount);
|
2005-04-17 00:20:36 +02:00
|
|
|
return;
|
2010-05-07 03:04:34 +02:00
|
|
|
}
|
2005-04-17 00:20:36 +02:00
|
|
|
}
|
|
|
|
|
2009-12-15 00:14:59 +01:00
|
|
|
trace_xfs_buf_item_unlock(bip);
|
2005-04-17 00:20:36 +02:00
|
|
|
|
|
|
|
/*
|
2010-05-07 03:04:34 +02:00
|
|
|
* If the buf item isn't tracking any data, free it, otherwise drop the
|
2013-01-21 13:53:55 +01:00
|
|
|
* reference we hold to it. If we are aborting the transaction, this may
|
|
|
|
* be the only reference to the buf item, so we free it anyway
|
|
|
|
* regardless of whether it is dirty or not. A dirty abort implies a
|
|
|
|
* shutdown, anyway.
|
2013-06-27 08:04:52 +02:00
|
|
|
*
|
|
|
|
* Ordered buffers are dirty but may have no recorded changes, so ensure
|
|
|
|
* we only release clean items here.
|
2005-04-17 00:20:36 +02:00
|
|
|
*/
|
2013-06-27 08:04:52 +02:00
|
|
|
clean = (flags & XFS_BLI_DIRTY) ? false : true;
|
|
|
|
if (clean) {
|
|
|
|
int i;
|
|
|
|
for (i = 0; i < bip->bli_format_count; i++) {
|
|
|
|
if (!xfs_bitmap_empty(bip->bli_formats[i].blf_data_map,
|
|
|
|
bip->bli_formats[i].blf_map_size)) {
|
|
|
|
clean = false;
|
|
|
|
break;
|
|
|
|
}
|
2012-12-05 00:18:05 +01:00
|
|
|
}
|
|
|
|
}
|
|
|
|
if (clean)
|
2005-04-17 00:20:36 +02:00
|
|
|
xfs_buf_item_relse(bp);
|
2013-01-21 13:53:55 +01:00
|
|
|
else if (aborted) {
|
|
|
|
if (atomic_dec_and_test(&bip->bli_refcount)) {
|
|
|
|
ASSERT(XFS_FORCED_SHUTDOWN(lip->li_mountp));
|
|
|
|
xfs_buf_item_relse(bp);
|
|
|
|
}
|
|
|
|
} else
|
2010-05-07 03:04:34 +02:00
|
|
|
atomic_dec(&bip->bli_refcount);
|
2005-04-17 00:20:36 +02:00
|
|
|
|
2013-06-27 08:04:52 +02:00
|
|
|
if (!(flags & XFS_BLI_HOLD))
|
2005-04-17 00:20:36 +02:00
|
|
|
xfs_buf_relse(bp);
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* This is called to find out where the oldest active copy of the
|
|
|
|
* buf log item in the on disk log resides now that the last log
|
|
|
|
* write of it completed at the given lsn.
|
|
|
|
* We always re-log all the dirty data in a buffer, so usually the
|
|
|
|
* latest copy in the on disk log is the only one that matters. For
|
|
|
|
* those cases we simply return the given lsn.
|
|
|
|
*
|
|
|
|
* The one exception to this is for buffers full of newly allocated
|
|
|
|
* inodes. These buffers are only relogged with the XFS_BLI_INODE_BUF
|
|
|
|
* flag set, indicating that only the di_next_unlinked fields from the
|
|
|
|
* inodes in the buffers will be replayed during recovery. If the
|
|
|
|
* original newly allocated inode images have not yet been flushed
|
|
|
|
* when the buffer is so relogged, then we need to make sure that we
|
|
|
|
* keep the old images in the 'active' portion of the log. We do this
|
|
|
|
* by returning the original lsn of that transaction here rather than
|
|
|
|
* the current one.
|
|
|
|
*/
|
2005-06-21 07:36:52 +02:00
|
|
|
STATIC xfs_lsn_t
|
2005-04-17 00:20:36 +02:00
|
|
|
xfs_buf_item_committed(
|
2010-06-23 10:11:15 +02:00
|
|
|
struct xfs_log_item *lip,
|
2005-04-17 00:20:36 +02:00
|
|
|
xfs_lsn_t lsn)
|
|
|
|
{
|
2010-06-23 10:11:15 +02:00
|
|
|
struct xfs_buf_log_item *bip = BUF_ITEM(lip);
|
|
|
|
|
2009-12-15 00:14:59 +01:00
|
|
|
trace_xfs_buf_item_committed(bip);
|
|
|
|
|
2010-06-23 10:11:15 +02:00
|
|
|
if ((bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF) && lip->li_lsn != 0)
|
|
|
|
return lip->li_lsn;
|
|
|
|
return lsn;
|
2005-04-17 00:20:36 +02:00
|
|
|
}
|
|
|
|
|
2005-06-21 07:36:52 +02:00
|
|
|
STATIC void
|
2010-06-23 10:11:15 +02:00
|
|
|
xfs_buf_item_committing(
|
|
|
|
struct xfs_log_item *lip,
|
|
|
|
xfs_lsn_t commit_lsn)
|
2005-04-17 00:20:36 +02:00
|
|
|
{
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* This is the ops vector shared by all buf log items.
|
|
|
|
*/
|
2011-10-28 11:54:24 +02:00
|
|
|
static const struct xfs_item_ops xfs_buf_item_ops = {
|
2010-06-23 10:11:15 +02:00
|
|
|
.iop_size = xfs_buf_item_size,
|
|
|
|
.iop_format = xfs_buf_item_format,
|
|
|
|
.iop_pin = xfs_buf_item_pin,
|
|
|
|
.iop_unpin = xfs_buf_item_unpin,
|
|
|
|
.iop_unlock = xfs_buf_item_unlock,
|
|
|
|
.iop_committed = xfs_buf_item_committed,
|
|
|
|
.iop_push = xfs_buf_item_push,
|
|
|
|
.iop_committing = xfs_buf_item_committing
|
2005-04-17 00:20:36 +02:00
|
|
|
};
|
|
|
|
|
2012-06-22 10:50:12 +02:00
|
|
|
STATIC int
|
|
|
|
xfs_buf_item_get_format(
|
|
|
|
struct xfs_buf_log_item *bip,
|
|
|
|
int count)
|
|
|
|
{
|
|
|
|
ASSERT(bip->bli_formats == NULL);
|
|
|
|
bip->bli_format_count = count;
|
|
|
|
|
|
|
|
if (count == 1) {
|
2012-12-05 00:18:03 +01:00
|
|
|
bip->bli_formats = &bip->__bli_format;
|
2012-06-22 10:50:12 +02:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
bip->bli_formats = kmem_zalloc(count * sizeof(struct xfs_buf_log_format),
|
|
|
|
KM_SLEEP);
|
|
|
|
if (!bip->bli_formats)
|
|
|
|
return ENOMEM;
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
STATIC void
|
|
|
|
xfs_buf_item_free_format(
|
|
|
|
struct xfs_buf_log_item *bip)
|
|
|
|
{
|
2012-12-05 00:18:03 +01:00
|
|
|
if (bip->bli_formats != &bip->__bli_format) {
|
2012-06-22 10:50:12 +02:00
|
|
|
kmem_free(bip->bli_formats);
|
|
|
|
bip->bli_formats = NULL;
|
|
|
|
}
|
|
|
|
}
|
2005-04-17 00:20:36 +02:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Allocate a new buf log item to go with the given buffer.
|
|
|
|
* Set the buffer's b_fsprivate field to point to the new
|
|
|
|
* buf log item. If there are other item's attached to the
|
|
|
|
* buffer (see xfs_buf_attach_iodone() below), then put the
|
|
|
|
* buf log item at the front.
|
|
|
|
*/
|
|
|
|
void
|
|
|
|
xfs_buf_item_init(
|
|
|
|
xfs_buf_t *bp,
|
|
|
|
xfs_mount_t *mp)
|
|
|
|
{
|
2011-07-13 13:43:49 +02:00
|
|
|
xfs_log_item_t *lip = bp->b_fspriv;
|
2005-04-17 00:20:36 +02:00
|
|
|
xfs_buf_log_item_t *bip;
|
|
|
|
int chunks;
|
|
|
|
int map_size;
|
2012-06-22 10:50:12 +02:00
|
|
|
int error;
|
|
|
|
int i;
|
2005-04-17 00:20:36 +02:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Check to see if there is already a buf log item for
|
|
|
|
* this buffer. If there is, it is guaranteed to be
|
|
|
|
* the first. If we do already have one, there is
|
|
|
|
* nothing to do here so return.
|
|
|
|
*/
|
2010-09-22 02:47:20 +02:00
|
|
|
ASSERT(bp->b_target->bt_mount == mp);
|
2011-07-13 13:43:49 +02:00
|
|
|
if (lip != NULL && lip->li_type == XFS_LI_BUF)
|
|
|
|
return;
|
2005-04-17 00:20:36 +02:00
|
|
|
|
2012-06-22 10:50:12 +02:00
|
|
|
bip = kmem_zone_zalloc(xfs_buf_item_zone, KM_SLEEP);
|
2010-03-23 00:10:00 +01:00
|
|
|
xfs_log_item_init(mp, &bip->bli_item, XFS_LI_BUF, &xfs_buf_item_ops);
|
2005-04-17 00:20:36 +02:00
|
|
|
bip->bli_buf = bp;
|
2008-09-17 08:52:13 +02:00
|
|
|
xfs_buf_hold(bp);
|
2012-06-22 10:50:12 +02:00
|
|
|
|
|
|
|
/*
|
|
|
|
* chunks is the number of XFS_BLF_CHUNK size pieces the buffer
|
|
|
|
* can be divided into. Make sure not to truncate any pieces.
|
|
|
|
* map_size is the size of the bitmap needed to describe the
|
|
|
|
* chunks of the buffer.
|
|
|
|
*
|
|
|
|
* Discontiguous buffer support follows the layout of the underlying
|
|
|
|
* buffer. This makes the implementation as simple as possible.
|
|
|
|
*/
|
|
|
|
error = xfs_buf_item_get_format(bip, bp->b_map_count);
|
|
|
|
ASSERT(error == 0);
|
|
|
|
|
|
|
|
for (i = 0; i < bip->bli_format_count; i++) {
|
|
|
|
chunks = DIV_ROUND_UP(BBTOB(bp->b_maps[i].bm_len),
|
|
|
|
XFS_BLF_CHUNK);
|
|
|
|
map_size = DIV_ROUND_UP(chunks, NBWORD);
|
|
|
|
|
|
|
|
bip->bli_formats[i].blf_type = XFS_LI_BUF;
|
|
|
|
bip->bli_formats[i].blf_blkno = bp->b_maps[i].bm_bn;
|
|
|
|
bip->bli_formats[i].blf_len = bp->b_maps[i].bm_len;
|
|
|
|
bip->bli_formats[i].blf_map_size = map_size;
|
|
|
|
}
|
2005-04-17 00:20:36 +02:00
|
|
|
|
|
|
|
#ifdef XFS_TRANS_DEBUG
|
|
|
|
/*
|
|
|
|
* Allocate the arrays for tracking what needs to be logged
|
|
|
|
* and what our callers request to be logged. bli_orig
|
|
|
|
* holds a copy of the original, clean buffer for comparison
|
|
|
|
* against, and bli_logged keeps a 1 bit flag per byte in
|
|
|
|
* the buffer to indicate which bytes the callers have asked
|
|
|
|
* to have logged.
|
|
|
|
*/
|
2012-04-23 07:58:52 +02:00
|
|
|
bip->bli_orig = kmem_alloc(BBTOB(bp->b_length), KM_SLEEP);
|
|
|
|
memcpy(bip->bli_orig, bp->b_addr, BBTOB(bp->b_length));
|
|
|
|
bip->bli_logged = kmem_zalloc(BBTOB(bp->b_length) / NBBY, KM_SLEEP);
|
2005-04-17 00:20:36 +02:00
|
|
|
#endif
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Put the buf item into the list of items attached to the
|
|
|
|
* buffer at the front.
|
|
|
|
*/
|
2011-07-13 13:43:49 +02:00
|
|
|
if (bp->b_fspriv)
|
|
|
|
bip->bli_item.li_bio_list = bp->b_fspriv;
|
|
|
|
bp->b_fspriv = bip;
|
2005-04-17 00:20:36 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Mark bytes first through last inclusive as dirty in the buf
|
|
|
|
* item's bitmap.
|
|
|
|
*/
|
|
|
|
void
|
2012-06-22 10:50:12 +02:00
|
|
|
xfs_buf_item_log_segment(
|
|
|
|
struct xfs_buf_log_item *bip,
|
2005-04-17 00:20:36 +02:00
|
|
|
uint first,
|
2012-06-22 10:50:12 +02:00
|
|
|
uint last,
|
|
|
|
uint *map)
|
2005-04-17 00:20:36 +02:00
|
|
|
{
|
|
|
|
uint first_bit;
|
|
|
|
uint last_bit;
|
|
|
|
uint bits_to_set;
|
|
|
|
uint bits_set;
|
|
|
|
uint word_num;
|
|
|
|
uint *wordp;
|
|
|
|
uint bit;
|
|
|
|
uint end_bit;
|
|
|
|
uint mask;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Convert byte offsets to bit numbers.
|
|
|
|
*/
|
2010-05-07 03:05:19 +02:00
|
|
|
first_bit = first >> XFS_BLF_SHIFT;
|
|
|
|
last_bit = last >> XFS_BLF_SHIFT;
|
2005-04-17 00:20:36 +02:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Calculate the total number of bits to be set.
|
|
|
|
*/
|
|
|
|
bits_to_set = last_bit - first_bit + 1;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Get a pointer to the first word in the bitmap
|
|
|
|
* to set a bit in.
|
|
|
|
*/
|
|
|
|
word_num = first_bit >> BIT_TO_WORD_SHIFT;
|
2012-06-22 10:50:12 +02:00
|
|
|
wordp = &map[word_num];
|
2005-04-17 00:20:36 +02:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Calculate the starting bit in the first word.
|
|
|
|
*/
|
|
|
|
bit = first_bit & (uint)(NBWORD - 1);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* First set any bits in the first word of our range.
|
|
|
|
* If it starts at bit 0 of the word, it will be
|
|
|
|
* set below rather than here. That is what the variable
|
|
|
|
* bit tells us. The variable bits_set tracks the number
|
|
|
|
* of bits that have been set so far. End_bit is the number
|
|
|
|
* of the last bit to be set in this word plus one.
|
|
|
|
*/
|
|
|
|
if (bit) {
|
|
|
|
end_bit = MIN(bit + bits_to_set, (uint)NBWORD);
|
|
|
|
mask = ((1 << (end_bit - bit)) - 1) << bit;
|
|
|
|
*wordp |= mask;
|
|
|
|
wordp++;
|
|
|
|
bits_set = end_bit - bit;
|
|
|
|
} else {
|
|
|
|
bits_set = 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Now set bits a whole word at a time that are between
|
|
|
|
* first_bit and last_bit.
|
|
|
|
*/
|
|
|
|
while ((bits_to_set - bits_set) >= NBWORD) {
|
|
|
|
*wordp |= 0xffffffff;
|
|
|
|
bits_set += NBWORD;
|
|
|
|
wordp++;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Finally, set any bits left to be set in one last partial word.
|
|
|
|
*/
|
|
|
|
end_bit = bits_to_set - bits_set;
|
|
|
|
if (end_bit) {
|
|
|
|
mask = (1 << end_bit) - 1;
|
|
|
|
*wordp |= mask;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2012-06-22 10:50:12 +02:00
|
|
|
/*
|
|
|
|
* Mark bytes first through last inclusive as dirty in the buf
|
|
|
|
* item's bitmap.
|
|
|
|
*/
|
|
|
|
void
|
|
|
|
xfs_buf_item_log(
|
|
|
|
xfs_buf_log_item_t *bip,
|
|
|
|
uint first,
|
|
|
|
uint last)
|
|
|
|
{
|
|
|
|
int i;
|
|
|
|
uint start;
|
|
|
|
uint end;
|
|
|
|
struct xfs_buf *bp = bip->bli_buf;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* walk each buffer segment and mark them dirty appropriately.
|
|
|
|
*/
|
|
|
|
start = 0;
|
|
|
|
for (i = 0; i < bip->bli_format_count; i++) {
|
|
|
|
if (start > last)
|
|
|
|
break;
|
|
|
|
end = start + BBTOB(bp->b_maps[i].bm_len);
|
|
|
|
if (first > end) {
|
|
|
|
start += BBTOB(bp->b_maps[i].bm_len);
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
if (first < start)
|
|
|
|
first = start;
|
|
|
|
if (end > last)
|
|
|
|
end = last;
|
|
|
|
|
|
|
|
xfs_buf_item_log_segment(bip, first, end,
|
|
|
|
&bip->bli_formats[i].blf_data_map[0]);
|
|
|
|
|
|
|
|
start += bp->b_maps[i].bm_len;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2005-04-17 00:20:36 +02:00
|
|
|
|
|
|
|
/*
|
2013-06-27 08:04:52 +02:00
|
|
|
* Return 1 if the buffer has been logged or ordered in a transaction (at any
|
2005-04-17 00:20:36 +02:00
|
|
|
* point, not just the current transaction) and 0 if not.
|
|
|
|
*/
|
|
|
|
uint
|
|
|
|
xfs_buf_item_dirty(
|
|
|
|
xfs_buf_log_item_t *bip)
|
|
|
|
{
|
|
|
|
return (bip->bli_flags & XFS_BLI_DIRTY);
|
|
|
|
}
|
|
|
|
|
2008-09-17 08:52:13 +02:00
|
|
|
STATIC void
|
|
|
|
xfs_buf_item_free(
|
|
|
|
xfs_buf_log_item_t *bip)
|
|
|
|
{
|
|
|
|
#ifdef XFS_TRANS_DEBUG
|
|
|
|
kmem_free(bip->bli_orig);
|
|
|
|
kmem_free(bip->bli_logged);
|
|
|
|
#endif /* XFS_TRANS_DEBUG */
|
|
|
|
|
2012-06-22 10:50:12 +02:00
|
|
|
xfs_buf_item_free_format(bip);
|
2008-09-17 08:52:13 +02:00
|
|
|
kmem_zone_free(xfs_buf_item_zone, bip);
|
|
|
|
}
|
|
|
|
|
2005-04-17 00:20:36 +02:00
|
|
|
/*
|
|
|
|
* This is called when the buf log item is no longer needed. It should
|
|
|
|
* free the buf log item associated with the given buffer and clear
|
|
|
|
* the buffer's pointer to the buf log item. If there are no more
|
|
|
|
* items in the list, clear the b_iodone field of the buffer (see
|
|
|
|
* xfs_buf_attach_iodone() below).
|
|
|
|
*/
|
|
|
|
void
|
|
|
|
xfs_buf_item_relse(
|
|
|
|
xfs_buf_t *bp)
|
|
|
|
{
|
2013-06-27 08:04:52 +02:00
|
|
|
xfs_buf_log_item_t *bip = bp->b_fspriv;
|
2005-04-17 00:20:36 +02:00
|
|
|
|
2009-12-15 00:14:59 +01:00
|
|
|
trace_xfs_buf_item_relse(bp, _RET_IP_);
|
2013-06-27 08:04:52 +02:00
|
|
|
ASSERT(!(bip->bli_item.li_flags & XFS_LI_IN_AIL));
|
2009-12-15 00:14:59 +01:00
|
|
|
|
2011-07-13 13:43:49 +02:00
|
|
|
bp->b_fspriv = bip->bli_item.li_bio_list;
|
2011-07-13 13:43:49 +02:00
|
|
|
if (bp->b_fspriv == NULL)
|
|
|
|
bp->b_iodone = NULL;
|
2011-07-13 13:43:49 +02:00
|
|
|
|
2008-09-17 08:52:13 +02:00
|
|
|
xfs_buf_rele(bp);
|
|
|
|
xfs_buf_item_free(bip);
|
2005-04-17 00:20:36 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Add the given log item with its callback to the list of callbacks
|
|
|
|
* to be called when the buffer's I/O completes. If it is not set
|
|
|
|
* already, set the buffer's b_iodone() routine to be
|
|
|
|
* xfs_buf_iodone_callbacks() and link the log item into the list of
|
|
|
|
* items rooted at b_fsprivate. Items are always added as the second
|
|
|
|
* entry in the list if there is a first, because the buf item code
|
|
|
|
* assumes that the buf log item is first.
|
|
|
|
*/
|
|
|
|
void
|
|
|
|
xfs_buf_attach_iodone(
|
|
|
|
xfs_buf_t *bp,
|
|
|
|
void (*cb)(xfs_buf_t *, xfs_log_item_t *),
|
|
|
|
xfs_log_item_t *lip)
|
|
|
|
{
|
|
|
|
xfs_log_item_t *head_lip;
|
|
|
|
|
2011-07-08 14:36:19 +02:00
|
|
|
ASSERT(xfs_buf_islocked(bp));
|
2005-04-17 00:20:36 +02:00
|
|
|
|
|
|
|
lip->li_cb = cb;
|
2011-07-13 13:43:49 +02:00
|
|
|
head_lip = bp->b_fspriv;
|
|
|
|
if (head_lip) {
|
2005-04-17 00:20:36 +02:00
|
|
|
lip->li_bio_list = head_lip->li_bio_list;
|
|
|
|
head_lip->li_bio_list = lip;
|
|
|
|
} else {
|
2011-07-13 13:43:49 +02:00
|
|
|
bp->b_fspriv = lip;
|
2005-04-17 00:20:36 +02:00
|
|
|
}
|
|
|
|
|
2011-07-13 13:43:49 +02:00
|
|
|
ASSERT(bp->b_iodone == NULL ||
|
|
|
|
bp->b_iodone == xfs_buf_iodone_callbacks);
|
|
|
|
bp->b_iodone = xfs_buf_iodone_callbacks;
|
2005-04-17 00:20:36 +02:00
|
|
|
}
|
|
|
|
|
2010-12-03 07:00:52 +01:00
|
|
|
/*
|
|
|
|
* We can have many callbacks on a buffer. Running the callbacks individually
|
|
|
|
* can cause a lot of contention on the AIL lock, so we allow for a single
|
|
|
|
* callback to be able to scan the remaining lip->li_bio_list for other items
|
|
|
|
* of the same type and callback to be processed in the first call.
|
|
|
|
*
|
|
|
|
* As a result, the loop walking the callback list below will also modify the
|
|
|
|
* list. it removes the first item from the list and then runs the callback.
|
|
|
|
* The loop then restarts from the new head of the list. This allows the
|
|
|
|
* callback to scan and modify the list attached to the buffer and we don't
|
|
|
|
* have to care about maintaining a next item pointer.
|
|
|
|
*/
|
2005-04-17 00:20:36 +02:00
|
|
|
STATIC void
|
|
|
|
xfs_buf_do_callbacks(
|
2010-12-03 07:00:52 +01:00
|
|
|
struct xfs_buf *bp)
|
2005-04-17 00:20:36 +02:00
|
|
|
{
|
2010-12-03 07:00:52 +01:00
|
|
|
struct xfs_log_item *lip;
|
2005-04-17 00:20:36 +02:00
|
|
|
|
2011-07-13 13:43:49 +02:00
|
|
|
while ((lip = bp->b_fspriv) != NULL) {
|
|
|
|
bp->b_fspriv = lip->li_bio_list;
|
2005-04-17 00:20:36 +02:00
|
|
|
ASSERT(lip->li_cb != NULL);
|
|
|
|
/*
|
|
|
|
* Clear the next pointer so we don't have any
|
|
|
|
* confusion if the item is added to another buf.
|
|
|
|
* Don't touch the log item after calling its
|
|
|
|
* callback, because it could have freed itself.
|
|
|
|
*/
|
|
|
|
lip->li_bio_list = NULL;
|
|
|
|
lip->li_cb(bp, lip);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* This is the iodone() function for buffers which have had callbacks
|
|
|
|
* attached to them by xfs_buf_attach_iodone(). It should remove each
|
|
|
|
* log item from the buffer's list and call the callback of each in turn.
|
|
|
|
* When done, the buffer's fsprivate field is set to NULL and the buffer
|
|
|
|
* is unlocked with a call to iodone().
|
|
|
|
*/
|
|
|
|
void
|
|
|
|
xfs_buf_iodone_callbacks(
|
xfs: fix error handling for synchronous writes
If we get an IO error on a synchronous superblock write, we attach an
error release function to it so that when the last reference goes away
the release function is called and the buffer is invalidated and
unlocked. The buffer is left locked until the release function is
called so that other concurrent users of the buffer will be locked out
until the buffer error is fully processed.
Unfortunately, for the superblock buffer the filesyetm itself holds a
reference to the buffer which prevents the reference count from
dropping to zero and the release function being called. As a result,
once an IO error occurs on a sync write, the buffer will never be
unlocked and all future attempts to lock the buffer will hang.
To make matters worse, this problems is not unique to such buffers;
if there is a concurrent _xfs_buf_find() running, the lookup will grab
a reference to the buffer and then wait on the buffer lock, preventing
the reference count from ever falling to zero and hence unlocking the
buffer.
As such, the whole b_relse function implementation is broken because it
cannot rely on the buffer reference count falling to zero to unlock the
errored buffer. The synchronous write error path is the only path that
uses this callback - it is used to ensure that the synchronous waiter
gets the buffer error before the error state is cleared from the buffer
by the release function.
Given that the only sychronous buffer writes now go through xfs_bwrite
and the error path in question can only occur for a write of a dirty,
logged buffer, we can move most of the b_relse processing to happen
inline in xfs_buf_iodone_callbacks, just like a normal I/O completion.
In addition to that we make sure the error is not cleared in
xfs_buf_iodone_callbacks, so that xfs_bwrite can reliably check it.
Given that xfs_bwrite keeps the buffer locked until it has waited for
it and checked the error this allows to reliably propagate the error
to the caller, and make sure that the buffer is reliably unlocked.
Given that xfs_buf_iodone_callbacks was the only instance of the
b_relse callback we can remove it entirely.
Based on earlier patches by Dave Chinner and Ajeet Yadav.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reported-by: Ajeet Yadav <ajeet.yadav.77@gmail.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Alex Elder <aelder@sgi.com>
2011-01-07 14:02:23 +01:00
|
|
|
struct xfs_buf *bp)
|
2005-04-17 00:20:36 +02:00
|
|
|
{
|
xfs: fix error handling for synchronous writes
If we get an IO error on a synchronous superblock write, we attach an
error release function to it so that when the last reference goes away
the release function is called and the buffer is invalidated and
unlocked. The buffer is left locked until the release function is
called so that other concurrent users of the buffer will be locked out
until the buffer error is fully processed.
Unfortunately, for the superblock buffer the filesyetm itself holds a
reference to the buffer which prevents the reference count from
dropping to zero and the release function being called. As a result,
once an IO error occurs on a sync write, the buffer will never be
unlocked and all future attempts to lock the buffer will hang.
To make matters worse, this problems is not unique to such buffers;
if there is a concurrent _xfs_buf_find() running, the lookup will grab
a reference to the buffer and then wait on the buffer lock, preventing
the reference count from ever falling to zero and hence unlocking the
buffer.
As such, the whole b_relse function implementation is broken because it
cannot rely on the buffer reference count falling to zero to unlock the
errored buffer. The synchronous write error path is the only path that
uses this callback - it is used to ensure that the synchronous waiter
gets the buffer error before the error state is cleared from the buffer
by the release function.
Given that the only sychronous buffer writes now go through xfs_bwrite
and the error path in question can only occur for a write of a dirty,
logged buffer, we can move most of the b_relse processing to happen
inline in xfs_buf_iodone_callbacks, just like a normal I/O completion.
In addition to that we make sure the error is not cleared in
xfs_buf_iodone_callbacks, so that xfs_bwrite can reliably check it.
Given that xfs_bwrite keeps the buffer locked until it has waited for
it and checked the error this allows to reliably propagate the error
to the caller, and make sure that the buffer is reliably unlocked.
Given that xfs_buf_iodone_callbacks was the only instance of the
b_relse callback we can remove it entirely.
Based on earlier patches by Dave Chinner and Ajeet Yadav.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reported-by: Ajeet Yadav <ajeet.yadav.77@gmail.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Alex Elder <aelder@sgi.com>
2011-01-07 14:02:23 +01:00
|
|
|
struct xfs_log_item *lip = bp->b_fspriv;
|
|
|
|
struct xfs_mount *mp = lip->li_mountp;
|
|
|
|
static ulong lasttime;
|
|
|
|
static xfs_buftarg_t *lasttarg;
|
2005-04-17 00:20:36 +02:00
|
|
|
|
2011-07-23 01:39:51 +02:00
|
|
|
if (likely(!xfs_buf_geterror(bp)))
|
xfs: fix error handling for synchronous writes
If we get an IO error on a synchronous superblock write, we attach an
error release function to it so that when the last reference goes away
the release function is called and the buffer is invalidated and
unlocked. The buffer is left locked until the release function is
called so that other concurrent users of the buffer will be locked out
until the buffer error is fully processed.
Unfortunately, for the superblock buffer the filesyetm itself holds a
reference to the buffer which prevents the reference count from
dropping to zero and the release function being called. As a result,
once an IO error occurs on a sync write, the buffer will never be
unlocked and all future attempts to lock the buffer will hang.
To make matters worse, this problems is not unique to such buffers;
if there is a concurrent _xfs_buf_find() running, the lookup will grab
a reference to the buffer and then wait on the buffer lock, preventing
the reference count from ever falling to zero and hence unlocking the
buffer.
As such, the whole b_relse function implementation is broken because it
cannot rely on the buffer reference count falling to zero to unlock the
errored buffer. The synchronous write error path is the only path that
uses this callback - it is used to ensure that the synchronous waiter
gets the buffer error before the error state is cleared from the buffer
by the release function.
Given that the only sychronous buffer writes now go through xfs_bwrite
and the error path in question can only occur for a write of a dirty,
logged buffer, we can move most of the b_relse processing to happen
inline in xfs_buf_iodone_callbacks, just like a normal I/O completion.
In addition to that we make sure the error is not cleared in
xfs_buf_iodone_callbacks, so that xfs_bwrite can reliably check it.
Given that xfs_bwrite keeps the buffer locked until it has waited for
it and checked the error this allows to reliably propagate the error
to the caller, and make sure that the buffer is reliably unlocked.
Given that xfs_buf_iodone_callbacks was the only instance of the
b_relse callback we can remove it entirely.
Based on earlier patches by Dave Chinner and Ajeet Yadav.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reported-by: Ajeet Yadav <ajeet.yadav.77@gmail.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Alex Elder <aelder@sgi.com>
2011-01-07 14:02:23 +01:00
|
|
|
goto do_callbacks;
|
2005-04-17 00:20:36 +02:00
|
|
|
|
xfs: fix error handling for synchronous writes
If we get an IO error on a synchronous superblock write, we attach an
error release function to it so that when the last reference goes away
the release function is called and the buffer is invalidated and
unlocked. The buffer is left locked until the release function is
called so that other concurrent users of the buffer will be locked out
until the buffer error is fully processed.
Unfortunately, for the superblock buffer the filesyetm itself holds a
reference to the buffer which prevents the reference count from
dropping to zero and the release function being called. As a result,
once an IO error occurs on a sync write, the buffer will never be
unlocked and all future attempts to lock the buffer will hang.
To make matters worse, this problems is not unique to such buffers;
if there is a concurrent _xfs_buf_find() running, the lookup will grab
a reference to the buffer and then wait on the buffer lock, preventing
the reference count from ever falling to zero and hence unlocking the
buffer.
As such, the whole b_relse function implementation is broken because it
cannot rely on the buffer reference count falling to zero to unlock the
errored buffer. The synchronous write error path is the only path that
uses this callback - it is used to ensure that the synchronous waiter
gets the buffer error before the error state is cleared from the buffer
by the release function.
Given that the only sychronous buffer writes now go through xfs_bwrite
and the error path in question can only occur for a write of a dirty,
logged buffer, we can move most of the b_relse processing to happen
inline in xfs_buf_iodone_callbacks, just like a normal I/O completion.
In addition to that we make sure the error is not cleared in
xfs_buf_iodone_callbacks, so that xfs_bwrite can reliably check it.
Given that xfs_bwrite keeps the buffer locked until it has waited for
it and checked the error this allows to reliably propagate the error
to the caller, and make sure that the buffer is reliably unlocked.
Given that xfs_buf_iodone_callbacks was the only instance of the
b_relse callback we can remove it entirely.
Based on earlier patches by Dave Chinner and Ajeet Yadav.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reported-by: Ajeet Yadav <ajeet.yadav.77@gmail.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Alex Elder <aelder@sgi.com>
2011-01-07 14:02:23 +01:00
|
|
|
/*
|
|
|
|
* If we've already decided to shutdown the filesystem because of
|
|
|
|
* I/O errors, there's no point in giving this a retry.
|
|
|
|
*/
|
|
|
|
if (XFS_FORCED_SHUTDOWN(mp)) {
|
2011-10-10 18:52:46 +02:00
|
|
|
xfs_buf_stale(bp);
|
|
|
|
XFS_BUF_DONE(bp);
|
xfs: fix error handling for synchronous writes
If we get an IO error on a synchronous superblock write, we attach an
error release function to it so that when the last reference goes away
the release function is called and the buffer is invalidated and
unlocked. The buffer is left locked until the release function is
called so that other concurrent users of the buffer will be locked out
until the buffer error is fully processed.
Unfortunately, for the superblock buffer the filesyetm itself holds a
reference to the buffer which prevents the reference count from
dropping to zero and the release function being called. As a result,
once an IO error occurs on a sync write, the buffer will never be
unlocked and all future attempts to lock the buffer will hang.
To make matters worse, this problems is not unique to such buffers;
if there is a concurrent _xfs_buf_find() running, the lookup will grab
a reference to the buffer and then wait on the buffer lock, preventing
the reference count from ever falling to zero and hence unlocking the
buffer.
As such, the whole b_relse function implementation is broken because it
cannot rely on the buffer reference count falling to zero to unlock the
errored buffer. The synchronous write error path is the only path that
uses this callback - it is used to ensure that the synchronous waiter
gets the buffer error before the error state is cleared from the buffer
by the release function.
Given that the only sychronous buffer writes now go through xfs_bwrite
and the error path in question can only occur for a write of a dirty,
logged buffer, we can move most of the b_relse processing to happen
inline in xfs_buf_iodone_callbacks, just like a normal I/O completion.
In addition to that we make sure the error is not cleared in
xfs_buf_iodone_callbacks, so that xfs_bwrite can reliably check it.
Given that xfs_bwrite keeps the buffer locked until it has waited for
it and checked the error this allows to reliably propagate the error
to the caller, and make sure that the buffer is reliably unlocked.
Given that xfs_buf_iodone_callbacks was the only instance of the
b_relse callback we can remove it entirely.
Based on earlier patches by Dave Chinner and Ajeet Yadav.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reported-by: Ajeet Yadav <ajeet.yadav.77@gmail.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Alex Elder <aelder@sgi.com>
2011-01-07 14:02:23 +01:00
|
|
|
trace_xfs_buf_item_iodone(bp, _RET_IP_);
|
|
|
|
goto do_callbacks;
|
|
|
|
}
|
2005-04-17 00:20:36 +02:00
|
|
|
|
2011-07-23 01:40:40 +02:00
|
|
|
if (bp->b_target != lasttarg ||
|
xfs: fix error handling for synchronous writes
If we get an IO error on a synchronous superblock write, we attach an
error release function to it so that when the last reference goes away
the release function is called and the buffer is invalidated and
unlocked. The buffer is left locked until the release function is
called so that other concurrent users of the buffer will be locked out
until the buffer error is fully processed.
Unfortunately, for the superblock buffer the filesyetm itself holds a
reference to the buffer which prevents the reference count from
dropping to zero and the release function being called. As a result,
once an IO error occurs on a sync write, the buffer will never be
unlocked and all future attempts to lock the buffer will hang.
To make matters worse, this problems is not unique to such buffers;
if there is a concurrent _xfs_buf_find() running, the lookup will grab
a reference to the buffer and then wait on the buffer lock, preventing
the reference count from ever falling to zero and hence unlocking the
buffer.
As such, the whole b_relse function implementation is broken because it
cannot rely on the buffer reference count falling to zero to unlock the
errored buffer. The synchronous write error path is the only path that
uses this callback - it is used to ensure that the synchronous waiter
gets the buffer error before the error state is cleared from the buffer
by the release function.
Given that the only sychronous buffer writes now go through xfs_bwrite
and the error path in question can only occur for a write of a dirty,
logged buffer, we can move most of the b_relse processing to happen
inline in xfs_buf_iodone_callbacks, just like a normal I/O completion.
In addition to that we make sure the error is not cleared in
xfs_buf_iodone_callbacks, so that xfs_bwrite can reliably check it.
Given that xfs_bwrite keeps the buffer locked until it has waited for
it and checked the error this allows to reliably propagate the error
to the caller, and make sure that the buffer is reliably unlocked.
Given that xfs_buf_iodone_callbacks was the only instance of the
b_relse callback we can remove it entirely.
Based on earlier patches by Dave Chinner and Ajeet Yadav.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reported-by: Ajeet Yadav <ajeet.yadav.77@gmail.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Alex Elder <aelder@sgi.com>
2011-01-07 14:02:23 +01:00
|
|
|
time_after(jiffies, (lasttime + 5*HZ))) {
|
|
|
|
lasttime = jiffies;
|
2011-10-10 18:52:50 +02:00
|
|
|
xfs_buf_ioerror_alert(bp, __func__);
|
xfs: fix error handling for synchronous writes
If we get an IO error on a synchronous superblock write, we attach an
error release function to it so that when the last reference goes away
the release function is called and the buffer is invalidated and
unlocked. The buffer is left locked until the release function is
called so that other concurrent users of the buffer will be locked out
until the buffer error is fully processed.
Unfortunately, for the superblock buffer the filesyetm itself holds a
reference to the buffer which prevents the reference count from
dropping to zero and the release function being called. As a result,
once an IO error occurs on a sync write, the buffer will never be
unlocked and all future attempts to lock the buffer will hang.
To make matters worse, this problems is not unique to such buffers;
if there is a concurrent _xfs_buf_find() running, the lookup will grab
a reference to the buffer and then wait on the buffer lock, preventing
the reference count from ever falling to zero and hence unlocking the
buffer.
As such, the whole b_relse function implementation is broken because it
cannot rely on the buffer reference count falling to zero to unlock the
errored buffer. The synchronous write error path is the only path that
uses this callback - it is used to ensure that the synchronous waiter
gets the buffer error before the error state is cleared from the buffer
by the release function.
Given that the only sychronous buffer writes now go through xfs_bwrite
and the error path in question can only occur for a write of a dirty,
logged buffer, we can move most of the b_relse processing to happen
inline in xfs_buf_iodone_callbacks, just like a normal I/O completion.
In addition to that we make sure the error is not cleared in
xfs_buf_iodone_callbacks, so that xfs_bwrite can reliably check it.
Given that xfs_bwrite keeps the buffer locked until it has waited for
it and checked the error this allows to reliably propagate the error
to the caller, and make sure that the buffer is reliably unlocked.
Given that xfs_buf_iodone_callbacks was the only instance of the
b_relse callback we can remove it entirely.
Based on earlier patches by Dave Chinner and Ajeet Yadav.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reported-by: Ajeet Yadav <ajeet.yadav.77@gmail.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Alex Elder <aelder@sgi.com>
2011-01-07 14:02:23 +01:00
|
|
|
}
|
2011-07-23 01:40:40 +02:00
|
|
|
lasttarg = bp->b_target;
|
2005-04-17 00:20:36 +02:00
|
|
|
|
xfs: fix error handling for synchronous writes
If we get an IO error on a synchronous superblock write, we attach an
error release function to it so that when the last reference goes away
the release function is called and the buffer is invalidated and
unlocked. The buffer is left locked until the release function is
called so that other concurrent users of the buffer will be locked out
until the buffer error is fully processed.
Unfortunately, for the superblock buffer the filesyetm itself holds a
reference to the buffer which prevents the reference count from
dropping to zero and the release function being called. As a result,
once an IO error occurs on a sync write, the buffer will never be
unlocked and all future attempts to lock the buffer will hang.
To make matters worse, this problems is not unique to such buffers;
if there is a concurrent _xfs_buf_find() running, the lookup will grab
a reference to the buffer and then wait on the buffer lock, preventing
the reference count from ever falling to zero and hence unlocking the
buffer.
As such, the whole b_relse function implementation is broken because it
cannot rely on the buffer reference count falling to zero to unlock the
errored buffer. The synchronous write error path is the only path that
uses this callback - it is used to ensure that the synchronous waiter
gets the buffer error before the error state is cleared from the buffer
by the release function.
Given that the only sychronous buffer writes now go through xfs_bwrite
and the error path in question can only occur for a write of a dirty,
logged buffer, we can move most of the b_relse processing to happen
inline in xfs_buf_iodone_callbacks, just like a normal I/O completion.
In addition to that we make sure the error is not cleared in
xfs_buf_iodone_callbacks, so that xfs_bwrite can reliably check it.
Given that xfs_bwrite keeps the buffer locked until it has waited for
it and checked the error this allows to reliably propagate the error
to the caller, and make sure that the buffer is reliably unlocked.
Given that xfs_buf_iodone_callbacks was the only instance of the
b_relse callback we can remove it entirely.
Based on earlier patches by Dave Chinner and Ajeet Yadav.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reported-by: Ajeet Yadav <ajeet.yadav.77@gmail.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Alex Elder <aelder@sgi.com>
2011-01-07 14:02:23 +01:00
|
|
|
/*
|
2011-03-31 03:57:33 +02:00
|
|
|
* If the write was asynchronous then no one will be looking for the
|
xfs: fix error handling for synchronous writes
If we get an IO error on a synchronous superblock write, we attach an
error release function to it so that when the last reference goes away
the release function is called and the buffer is invalidated and
unlocked. The buffer is left locked until the release function is
called so that other concurrent users of the buffer will be locked out
until the buffer error is fully processed.
Unfortunately, for the superblock buffer the filesyetm itself holds a
reference to the buffer which prevents the reference count from
dropping to zero and the release function being called. As a result,
once an IO error occurs on a sync write, the buffer will never be
unlocked and all future attempts to lock the buffer will hang.
To make matters worse, this problems is not unique to such buffers;
if there is a concurrent _xfs_buf_find() running, the lookup will grab
a reference to the buffer and then wait on the buffer lock, preventing
the reference count from ever falling to zero and hence unlocking the
buffer.
As such, the whole b_relse function implementation is broken because it
cannot rely on the buffer reference count falling to zero to unlock the
errored buffer. The synchronous write error path is the only path that
uses this callback - it is used to ensure that the synchronous waiter
gets the buffer error before the error state is cleared from the buffer
by the release function.
Given that the only sychronous buffer writes now go through xfs_bwrite
and the error path in question can only occur for a write of a dirty,
logged buffer, we can move most of the b_relse processing to happen
inline in xfs_buf_iodone_callbacks, just like a normal I/O completion.
In addition to that we make sure the error is not cleared in
xfs_buf_iodone_callbacks, so that xfs_bwrite can reliably check it.
Given that xfs_bwrite keeps the buffer locked until it has waited for
it and checked the error this allows to reliably propagate the error
to the caller, and make sure that the buffer is reliably unlocked.
Given that xfs_buf_iodone_callbacks was the only instance of the
b_relse callback we can remove it entirely.
Based on earlier patches by Dave Chinner and Ajeet Yadav.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reported-by: Ajeet Yadav <ajeet.yadav.77@gmail.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Alex Elder <aelder@sgi.com>
2011-01-07 14:02:23 +01:00
|
|
|
* error. Clear the error state and write the buffer out again.
|
|
|
|
*
|
xfs: on-stack delayed write buffer lists
Queue delwri buffers on a local on-stack list instead of a per-buftarg one,
and write back the buffers per-process instead of by waking up xfsbufd.
This is now easily doable given that we have very few places left that write
delwri buffers:
- log recovery:
Only done at mount time, and already forcing out the buffers
synchronously using xfs_flush_buftarg
- quotacheck:
Same story.
- dquot reclaim:
Writes out dirty dquots on the LRU under memory pressure. We might
want to look into doing more of this via xfsaild, but it's already
more optimal than the synchronous inode reclaim that writes each
buffer synchronously.
- xfsaild:
This is the main beneficiary of the change. By keeping a local list
of buffers to write we reduce latency of writing out buffers, and
more importably we can remove all the delwri list promotions which
were hitting the buffer cache hard under sustained metadata loads.
The implementation is very straight forward - xfs_buf_delwri_queue now gets
a new list_head pointer that it adds the delwri buffers to, and all callers
need to eventually submit the list using xfs_buf_delwi_submit or
xfs_buf_delwi_submit_nowait. Buffers that already are on a delwri list are
skipped in xfs_buf_delwri_queue, assuming they already are on another delwri
list. The biggest change to pass down the buffer list was done to the AIL
pushing. Now that we operate on buffers the trylock, push and pushbuf log
item methods are merged into a single push routine, which tries to lock the
item, and if possible add the buffer that needs writeback to the buffer list.
This leads to much simpler code than the previous split but requires the
individual IOP_PUSH instances to unlock and reacquire the AIL around calls
to blocking routines.
Given that xfsailds now also handle writing out buffers, the conditions for
log forcing and the sleep times needed some small changes. The most
important one is that we consider an AIL busy as long we still have buffers
to push, and the other one is that we do increment the pushed LSN for
buffers that are under flushing at this moment, but still count them towards
the stuck items for restart purposes. Without this we could hammer on stuck
items without ever forcing the log and not make progress under heavy random
delete workloads on fast flash storage devices.
[ Dave Chinner:
- rebase on previous patches.
- improved comments for XBF_DELWRI_Q handling
- fix XBF_ASYNC handling in queue submission (test 106 failure)
- rename delwri submit function buffer list parameters for clarity
- xfs_efd_item_push() should return XFS_ITEM_PINNED ]
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Mark Tinguely <tinguely@sgi.com>
Signed-off-by: Ben Myers <bpm@sgi.com>
2012-04-23 07:58:39 +02:00
|
|
|
* XXX: This helps against transient write errors, but we need to find
|
|
|
|
* a way to shut the filesystem down if the writes keep failing.
|
|
|
|
*
|
|
|
|
* In practice we'll shut the filesystem down soon as non-transient
|
|
|
|
* erorrs tend to affect the whole device and a failing log write
|
|
|
|
* will make us give up. But we really ought to do better here.
|
xfs: fix error handling for synchronous writes
If we get an IO error on a synchronous superblock write, we attach an
error release function to it so that when the last reference goes away
the release function is called and the buffer is invalidated and
unlocked. The buffer is left locked until the release function is
called so that other concurrent users of the buffer will be locked out
until the buffer error is fully processed.
Unfortunately, for the superblock buffer the filesyetm itself holds a
reference to the buffer which prevents the reference count from
dropping to zero and the release function being called. As a result,
once an IO error occurs on a sync write, the buffer will never be
unlocked and all future attempts to lock the buffer will hang.
To make matters worse, this problems is not unique to such buffers;
if there is a concurrent _xfs_buf_find() running, the lookup will grab
a reference to the buffer and then wait on the buffer lock, preventing
the reference count from ever falling to zero and hence unlocking the
buffer.
As such, the whole b_relse function implementation is broken because it
cannot rely on the buffer reference count falling to zero to unlock the
errored buffer. The synchronous write error path is the only path that
uses this callback - it is used to ensure that the synchronous waiter
gets the buffer error before the error state is cleared from the buffer
by the release function.
Given that the only sychronous buffer writes now go through xfs_bwrite
and the error path in question can only occur for a write of a dirty,
logged buffer, we can move most of the b_relse processing to happen
inline in xfs_buf_iodone_callbacks, just like a normal I/O completion.
In addition to that we make sure the error is not cleared in
xfs_buf_iodone_callbacks, so that xfs_bwrite can reliably check it.
Given that xfs_bwrite keeps the buffer locked until it has waited for
it and checked the error this allows to reliably propagate the error
to the caller, and make sure that the buffer is reliably unlocked.
Given that xfs_buf_iodone_callbacks was the only instance of the
b_relse callback we can remove it entirely.
Based on earlier patches by Dave Chinner and Ajeet Yadav.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reported-by: Ajeet Yadav <ajeet.yadav.77@gmail.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Alex Elder <aelder@sgi.com>
2011-01-07 14:02:23 +01:00
|
|
|
*/
|
|
|
|
if (XFS_BUF_ISASYNC(bp)) {
|
xfs: on-stack delayed write buffer lists
Queue delwri buffers on a local on-stack list instead of a per-buftarg one,
and write back the buffers per-process instead of by waking up xfsbufd.
This is now easily doable given that we have very few places left that write
delwri buffers:
- log recovery:
Only done at mount time, and already forcing out the buffers
synchronously using xfs_flush_buftarg
- quotacheck:
Same story.
- dquot reclaim:
Writes out dirty dquots on the LRU under memory pressure. We might
want to look into doing more of this via xfsaild, but it's already
more optimal than the synchronous inode reclaim that writes each
buffer synchronously.
- xfsaild:
This is the main beneficiary of the change. By keeping a local list
of buffers to write we reduce latency of writing out buffers, and
more importably we can remove all the delwri list promotions which
were hitting the buffer cache hard under sustained metadata loads.
The implementation is very straight forward - xfs_buf_delwri_queue now gets
a new list_head pointer that it adds the delwri buffers to, and all callers
need to eventually submit the list using xfs_buf_delwi_submit or
xfs_buf_delwi_submit_nowait. Buffers that already are on a delwri list are
skipped in xfs_buf_delwri_queue, assuming they already are on another delwri
list. The biggest change to pass down the buffer list was done to the AIL
pushing. Now that we operate on buffers the trylock, push and pushbuf log
item methods are merged into a single push routine, which tries to lock the
item, and if possible add the buffer that needs writeback to the buffer list.
This leads to much simpler code than the previous split but requires the
individual IOP_PUSH instances to unlock and reacquire the AIL around calls
to blocking routines.
Given that xfsailds now also handle writing out buffers, the conditions for
log forcing and the sleep times needed some small changes. The most
important one is that we consider an AIL busy as long we still have buffers
to push, and the other one is that we do increment the pushed LSN for
buffers that are under flushing at this moment, but still count them towards
the stuck items for restart purposes. Without this we could hammer on stuck
items without ever forcing the log and not make progress under heavy random
delete workloads on fast flash storage devices.
[ Dave Chinner:
- rebase on previous patches.
- improved comments for XBF_DELWRI_Q handling
- fix XBF_ASYNC handling in queue submission (test 106 failure)
- rename delwri submit function buffer list parameters for clarity
- xfs_efd_item_push() should return XFS_ITEM_PINNED ]
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Mark Tinguely <tinguely@sgi.com>
Signed-off-by: Ben Myers <bpm@sgi.com>
2012-04-23 07:58:39 +02:00
|
|
|
ASSERT(bp->b_iodone != NULL);
|
|
|
|
|
|
|
|
trace_xfs_buf_item_iodone_async(bp, _RET_IP_);
|
|
|
|
|
2011-07-23 01:39:51 +02:00
|
|
|
xfs_buf_ioerror(bp, 0); /* errno of 0 unsets the flag */
|
xfs: fix error handling for synchronous writes
If we get an IO error on a synchronous superblock write, we attach an
error release function to it so that when the last reference goes away
the release function is called and the buffer is invalidated and
unlocked. The buffer is left locked until the release function is
called so that other concurrent users of the buffer will be locked out
until the buffer error is fully processed.
Unfortunately, for the superblock buffer the filesyetm itself holds a
reference to the buffer which prevents the reference count from
dropping to zero and the release function being called. As a result,
once an IO error occurs on a sync write, the buffer will never be
unlocked and all future attempts to lock the buffer will hang.
To make matters worse, this problems is not unique to such buffers;
if there is a concurrent _xfs_buf_find() running, the lookup will grab
a reference to the buffer and then wait on the buffer lock, preventing
the reference count from ever falling to zero and hence unlocking the
buffer.
As such, the whole b_relse function implementation is broken because it
cannot rely on the buffer reference count falling to zero to unlock the
errored buffer. The synchronous write error path is the only path that
uses this callback - it is used to ensure that the synchronous waiter
gets the buffer error before the error state is cleared from the buffer
by the release function.
Given that the only sychronous buffer writes now go through xfs_bwrite
and the error path in question can only occur for a write of a dirty,
logged buffer, we can move most of the b_relse processing to happen
inline in xfs_buf_iodone_callbacks, just like a normal I/O completion.
In addition to that we make sure the error is not cleared in
xfs_buf_iodone_callbacks, so that xfs_bwrite can reliably check it.
Given that xfs_bwrite keeps the buffer locked until it has waited for
it and checked the error this allows to reliably propagate the error
to the caller, and make sure that the buffer is reliably unlocked.
Given that xfs_buf_iodone_callbacks was the only instance of the
b_relse callback we can remove it entirely.
Based on earlier patches by Dave Chinner and Ajeet Yadav.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reported-by: Ajeet Yadav <ajeet.yadav.77@gmail.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Alex Elder <aelder@sgi.com>
2011-01-07 14:02:23 +01:00
|
|
|
|
|
|
|
if (!XFS_BUF_ISSTALE(bp)) {
|
xfs: on-stack delayed write buffer lists
Queue delwri buffers on a local on-stack list instead of a per-buftarg one,
and write back the buffers per-process instead of by waking up xfsbufd.
This is now easily doable given that we have very few places left that write
delwri buffers:
- log recovery:
Only done at mount time, and already forcing out the buffers
synchronously using xfs_flush_buftarg
- quotacheck:
Same story.
- dquot reclaim:
Writes out dirty dquots on the LRU under memory pressure. We might
want to look into doing more of this via xfsaild, but it's already
more optimal than the synchronous inode reclaim that writes each
buffer synchronously.
- xfsaild:
This is the main beneficiary of the change. By keeping a local list
of buffers to write we reduce latency of writing out buffers, and
more importably we can remove all the delwri list promotions which
were hitting the buffer cache hard under sustained metadata loads.
The implementation is very straight forward - xfs_buf_delwri_queue now gets
a new list_head pointer that it adds the delwri buffers to, and all callers
need to eventually submit the list using xfs_buf_delwi_submit or
xfs_buf_delwi_submit_nowait. Buffers that already are on a delwri list are
skipped in xfs_buf_delwri_queue, assuming they already are on another delwri
list. The biggest change to pass down the buffer list was done to the AIL
pushing. Now that we operate on buffers the trylock, push and pushbuf log
item methods are merged into a single push routine, which tries to lock the
item, and if possible add the buffer that needs writeback to the buffer list.
This leads to much simpler code than the previous split but requires the
individual IOP_PUSH instances to unlock and reacquire the AIL around calls
to blocking routines.
Given that xfsailds now also handle writing out buffers, the conditions for
log forcing and the sleep times needed some small changes. The most
important one is that we consider an AIL busy as long we still have buffers
to push, and the other one is that we do increment the pushed LSN for
buffers that are under flushing at this moment, but still count them towards
the stuck items for restart purposes. Without this we could hammer on stuck
items without ever forcing the log and not make progress under heavy random
delete workloads on fast flash storage devices.
[ Dave Chinner:
- rebase on previous patches.
- improved comments for XBF_DELWRI_Q handling
- fix XBF_ASYNC handling in queue submission (test 106 failure)
- rename delwri submit function buffer list parameters for clarity
- xfs_efd_item_push() should return XFS_ITEM_PINNED ]
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Mark Tinguely <tinguely@sgi.com>
Signed-off-by: Ben Myers <bpm@sgi.com>
2012-04-23 07:58:39 +02:00
|
|
|
bp->b_flags |= XBF_WRITE | XBF_ASYNC | XBF_DONE;
|
2012-07-13 08:24:10 +02:00
|
|
|
xfs_buf_iorequest(bp);
|
xfs: on-stack delayed write buffer lists
Queue delwri buffers on a local on-stack list instead of a per-buftarg one,
and write back the buffers per-process instead of by waking up xfsbufd.
This is now easily doable given that we have very few places left that write
delwri buffers:
- log recovery:
Only done at mount time, and already forcing out the buffers
synchronously using xfs_flush_buftarg
- quotacheck:
Same story.
- dquot reclaim:
Writes out dirty dquots on the LRU under memory pressure. We might
want to look into doing more of this via xfsaild, but it's already
more optimal than the synchronous inode reclaim that writes each
buffer synchronously.
- xfsaild:
This is the main beneficiary of the change. By keeping a local list
of buffers to write we reduce latency of writing out buffers, and
more importably we can remove all the delwri list promotions which
were hitting the buffer cache hard under sustained metadata loads.
The implementation is very straight forward - xfs_buf_delwri_queue now gets
a new list_head pointer that it adds the delwri buffers to, and all callers
need to eventually submit the list using xfs_buf_delwi_submit or
xfs_buf_delwi_submit_nowait. Buffers that already are on a delwri list are
skipped in xfs_buf_delwri_queue, assuming they already are on another delwri
list. The biggest change to pass down the buffer list was done to the AIL
pushing. Now that we operate on buffers the trylock, push and pushbuf log
item methods are merged into a single push routine, which tries to lock the
item, and if possible add the buffer that needs writeback to the buffer list.
This leads to much simpler code than the previous split but requires the
individual IOP_PUSH instances to unlock and reacquire the AIL around calls
to blocking routines.
Given that xfsailds now also handle writing out buffers, the conditions for
log forcing and the sleep times needed some small changes. The most
important one is that we consider an AIL busy as long we still have buffers
to push, and the other one is that we do increment the pushed LSN for
buffers that are under flushing at this moment, but still count them towards
the stuck items for restart purposes. Without this we could hammer on stuck
items without ever forcing the log and not make progress under heavy random
delete workloads on fast flash storage devices.
[ Dave Chinner:
- rebase on previous patches.
- improved comments for XBF_DELWRI_Q handling
- fix XBF_ASYNC handling in queue submission (test 106 failure)
- rename delwri submit function buffer list parameters for clarity
- xfs_efd_item_push() should return XFS_ITEM_PINNED ]
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Mark Tinguely <tinguely@sgi.com>
Signed-off-by: Ben Myers <bpm@sgi.com>
2012-04-23 07:58:39 +02:00
|
|
|
} else {
|
|
|
|
xfs_buf_relse(bp);
|
2005-04-17 00:20:36 +02:00
|
|
|
}
|
xfs: on-stack delayed write buffer lists
Queue delwri buffers on a local on-stack list instead of a per-buftarg one,
and write back the buffers per-process instead of by waking up xfsbufd.
This is now easily doable given that we have very few places left that write
delwri buffers:
- log recovery:
Only done at mount time, and already forcing out the buffers
synchronously using xfs_flush_buftarg
- quotacheck:
Same story.
- dquot reclaim:
Writes out dirty dquots on the LRU under memory pressure. We might
want to look into doing more of this via xfsaild, but it's already
more optimal than the synchronous inode reclaim that writes each
buffer synchronously.
- xfsaild:
This is the main beneficiary of the change. By keeping a local list
of buffers to write we reduce latency of writing out buffers, and
more importably we can remove all the delwri list promotions which
were hitting the buffer cache hard under sustained metadata loads.
The implementation is very straight forward - xfs_buf_delwri_queue now gets
a new list_head pointer that it adds the delwri buffers to, and all callers
need to eventually submit the list using xfs_buf_delwi_submit or
xfs_buf_delwi_submit_nowait. Buffers that already are on a delwri list are
skipped in xfs_buf_delwri_queue, assuming they already are on another delwri
list. The biggest change to pass down the buffer list was done to the AIL
pushing. Now that we operate on buffers the trylock, push and pushbuf log
item methods are merged into a single push routine, which tries to lock the
item, and if possible add the buffer that needs writeback to the buffer list.
This leads to much simpler code than the previous split but requires the
individual IOP_PUSH instances to unlock and reacquire the AIL around calls
to blocking routines.
Given that xfsailds now also handle writing out buffers, the conditions for
log forcing and the sleep times needed some small changes. The most
important one is that we consider an AIL busy as long we still have buffers
to push, and the other one is that we do increment the pushed LSN for
buffers that are under flushing at this moment, but still count them towards
the stuck items for restart purposes. Without this we could hammer on stuck
items without ever forcing the log and not make progress under heavy random
delete workloads on fast flash storage devices.
[ Dave Chinner:
- rebase on previous patches.
- improved comments for XBF_DELWRI_Q handling
- fix XBF_ASYNC handling in queue submission (test 106 failure)
- rename delwri submit function buffer list parameters for clarity
- xfs_efd_item_push() should return XFS_ITEM_PINNED ]
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Mark Tinguely <tinguely@sgi.com>
Signed-off-by: Ben Myers <bpm@sgi.com>
2012-04-23 07:58:39 +02:00
|
|
|
|
2005-04-17 00:20:36 +02:00
|
|
|
return;
|
|
|
|
}
|
2009-12-15 00:14:59 +01:00
|
|
|
|
xfs: fix error handling for synchronous writes
If we get an IO error on a synchronous superblock write, we attach an
error release function to it so that when the last reference goes away
the release function is called and the buffer is invalidated and
unlocked. The buffer is left locked until the release function is
called so that other concurrent users of the buffer will be locked out
until the buffer error is fully processed.
Unfortunately, for the superblock buffer the filesyetm itself holds a
reference to the buffer which prevents the reference count from
dropping to zero and the release function being called. As a result,
once an IO error occurs on a sync write, the buffer will never be
unlocked and all future attempts to lock the buffer will hang.
To make matters worse, this problems is not unique to such buffers;
if there is a concurrent _xfs_buf_find() running, the lookup will grab
a reference to the buffer and then wait on the buffer lock, preventing
the reference count from ever falling to zero and hence unlocking the
buffer.
As such, the whole b_relse function implementation is broken because it
cannot rely on the buffer reference count falling to zero to unlock the
errored buffer. The synchronous write error path is the only path that
uses this callback - it is used to ensure that the synchronous waiter
gets the buffer error before the error state is cleared from the buffer
by the release function.
Given that the only sychronous buffer writes now go through xfs_bwrite
and the error path in question can only occur for a write of a dirty,
logged buffer, we can move most of the b_relse processing to happen
inline in xfs_buf_iodone_callbacks, just like a normal I/O completion.
In addition to that we make sure the error is not cleared in
xfs_buf_iodone_callbacks, so that xfs_bwrite can reliably check it.
Given that xfs_bwrite keeps the buffer locked until it has waited for
it and checked the error this allows to reliably propagate the error
to the caller, and make sure that the buffer is reliably unlocked.
Given that xfs_buf_iodone_callbacks was the only instance of the
b_relse callback we can remove it entirely.
Based on earlier patches by Dave Chinner and Ajeet Yadav.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reported-by: Ajeet Yadav <ajeet.yadav.77@gmail.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Alex Elder <aelder@sgi.com>
2011-01-07 14:02:23 +01:00
|
|
|
/*
|
|
|
|
* If the write of the buffer was synchronous, we want to make
|
|
|
|
* sure to return the error to the caller of xfs_bwrite().
|
|
|
|
*/
|
2011-10-10 18:52:46 +02:00
|
|
|
xfs_buf_stale(bp);
|
2005-04-17 00:20:36 +02:00
|
|
|
XFS_BUF_DONE(bp);
|
2009-12-15 00:14:59 +01:00
|
|
|
|
|
|
|
trace_xfs_buf_error_relse(bp, _RET_IP_);
|
|
|
|
|
xfs: fix error handling for synchronous writes
If we get an IO error on a synchronous superblock write, we attach an
error release function to it so that when the last reference goes away
the release function is called and the buffer is invalidated and
unlocked. The buffer is left locked until the release function is
called so that other concurrent users of the buffer will be locked out
until the buffer error is fully processed.
Unfortunately, for the superblock buffer the filesyetm itself holds a
reference to the buffer which prevents the reference count from
dropping to zero and the release function being called. As a result,
once an IO error occurs on a sync write, the buffer will never be
unlocked and all future attempts to lock the buffer will hang.
To make matters worse, this problems is not unique to such buffers;
if there is a concurrent _xfs_buf_find() running, the lookup will grab
a reference to the buffer and then wait on the buffer lock, preventing
the reference count from ever falling to zero and hence unlocking the
buffer.
As such, the whole b_relse function implementation is broken because it
cannot rely on the buffer reference count falling to zero to unlock the
errored buffer. The synchronous write error path is the only path that
uses this callback - it is used to ensure that the synchronous waiter
gets the buffer error before the error state is cleared from the buffer
by the release function.
Given that the only sychronous buffer writes now go through xfs_bwrite
and the error path in question can only occur for a write of a dirty,
logged buffer, we can move most of the b_relse processing to happen
inline in xfs_buf_iodone_callbacks, just like a normal I/O completion.
In addition to that we make sure the error is not cleared in
xfs_buf_iodone_callbacks, so that xfs_bwrite can reliably check it.
Given that xfs_bwrite keeps the buffer locked until it has waited for
it and checked the error this allows to reliably propagate the error
to the caller, and make sure that the buffer is reliably unlocked.
Given that xfs_buf_iodone_callbacks was the only instance of the
b_relse callback we can remove it entirely.
Based on earlier patches by Dave Chinner and Ajeet Yadav.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reported-by: Ajeet Yadav <ajeet.yadav.77@gmail.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Alex Elder <aelder@sgi.com>
2011-01-07 14:02:23 +01:00
|
|
|
do_callbacks:
|
2010-12-03 07:00:52 +01:00
|
|
|
xfs_buf_do_callbacks(bp);
|
2011-07-13 13:43:49 +02:00
|
|
|
bp->b_fspriv = NULL;
|
2011-07-13 13:43:49 +02:00
|
|
|
bp->b_iodone = NULL;
|
xfs: fix error handling for synchronous writes
If we get an IO error on a synchronous superblock write, we attach an
error release function to it so that when the last reference goes away
the release function is called and the buffer is invalidated and
unlocked. The buffer is left locked until the release function is
called so that other concurrent users of the buffer will be locked out
until the buffer error is fully processed.
Unfortunately, for the superblock buffer the filesyetm itself holds a
reference to the buffer which prevents the reference count from
dropping to zero and the release function being called. As a result,
once an IO error occurs on a sync write, the buffer will never be
unlocked and all future attempts to lock the buffer will hang.
To make matters worse, this problems is not unique to such buffers;
if there is a concurrent _xfs_buf_find() running, the lookup will grab
a reference to the buffer and then wait on the buffer lock, preventing
the reference count from ever falling to zero and hence unlocking the
buffer.
As such, the whole b_relse function implementation is broken because it
cannot rely on the buffer reference count falling to zero to unlock the
errored buffer. The synchronous write error path is the only path that
uses this callback - it is used to ensure that the synchronous waiter
gets the buffer error before the error state is cleared from the buffer
by the release function.
Given that the only sychronous buffer writes now go through xfs_bwrite
and the error path in question can only occur for a write of a dirty,
logged buffer, we can move most of the b_relse processing to happen
inline in xfs_buf_iodone_callbacks, just like a normal I/O completion.
In addition to that we make sure the error is not cleared in
xfs_buf_iodone_callbacks, so that xfs_bwrite can reliably check it.
Given that xfs_bwrite keeps the buffer locked until it has waited for
it and checked the error this allows to reliably propagate the error
to the caller, and make sure that the buffer is reliably unlocked.
Given that xfs_buf_iodone_callbacks was the only instance of the
b_relse callback we can remove it entirely.
Based on earlier patches by Dave Chinner and Ajeet Yadav.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reported-by: Ajeet Yadav <ajeet.yadav.77@gmail.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Alex Elder <aelder@sgi.com>
2011-01-07 14:02:23 +01:00
|
|
|
xfs_buf_ioend(bp, 0);
|
2005-04-17 00:20:36 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* This is the iodone() function for buffers which have been
|
|
|
|
* logged. It is called when they are eventually flushed out.
|
|
|
|
* It should remove the buf item from the AIL, and free the buf item.
|
|
|
|
* It is called by xfs_buf_iodone_callbacks() above which will take
|
|
|
|
* care of cleaning up the buffer itself.
|
|
|
|
*/
|
|
|
|
void
|
|
|
|
xfs_buf_iodone(
|
2010-06-23 10:11:15 +02:00
|
|
|
struct xfs_buf *bp,
|
|
|
|
struct xfs_log_item *lip)
|
2005-04-17 00:20:36 +02:00
|
|
|
{
|
2010-06-23 10:11:15 +02:00
|
|
|
struct xfs_ail *ailp = lip->li_ailp;
|
2005-04-17 00:20:36 +02:00
|
|
|
|
2010-06-23 10:11:15 +02:00
|
|
|
ASSERT(BUF_ITEM(lip)->bli_buf == bp);
|
2005-04-17 00:20:36 +02:00
|
|
|
|
2008-09-17 08:52:13 +02:00
|
|
|
xfs_buf_rele(bp);
|
2005-04-17 00:20:36 +02:00
|
|
|
|
|
|
|
/*
|
|
|
|
* If we are forcibly shutting down, this may well be
|
|
|
|
* off the AIL already. That's because we simulate the
|
|
|
|
* log-committed callbacks to unpin these buffers. Or we may never
|
|
|
|
* have put this item on AIL because of the transaction was
|
2008-10-30 07:39:58 +01:00
|
|
|
* aborted forcibly. xfs_trans_ail_delete() takes care of these.
|
2005-04-17 00:20:36 +02:00
|
|
|
*
|
|
|
|
* Either way, AIL is useless if we're forcing a shutdown.
|
|
|
|
*/
|
2008-10-30 07:39:46 +01:00
|
|
|
spin_lock(&ailp->xa_lock);
|
2012-04-23 07:58:41 +02:00
|
|
|
xfs_trans_ail_delete(ailp, lip, SHUTDOWN_CORRUPT_INCORE);
|
2010-06-23 10:11:15 +02:00
|
|
|
xfs_buf_item_free(BUF_ITEM(lip));
|
2005-04-17 00:20:36 +02:00
|
|
|
}
|