linux/fs/gfs2/log.c

896 lines
23 KiB
C
Raw Normal View History

/*
* Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
* Copyright (C) 2004-2007 Red Hat, Inc. All rights reserved.
*
* This copyrighted material is made available to anyone wishing to use,
* modify, copy, or redistribute it subject to the terms and conditions
* of the GNU General Public License version 2.
*/
#include <linux/sched.h>
#include <linux/slab.h>
#include <linux/spinlock.h>
#include <linux/completion.h>
#include <linux/buffer_head.h>
#include <linux/gfs2_ondisk.h>
#include <linux/crc32.h>
#include <linux/delay.h>
#include <linux/kthread.h>
#include <linux/freezer.h>
#include <linux/bio.h>
#include <linux/writeback.h>
#include <linux/list_sort.h>
#include "gfs2.h"
#include "incore.h"
#include "bmap.h"
#include "glock.h"
#include "log.h"
#include "lops.h"
#include "meta_io.h"
#include "util.h"
#include "dir.h"
#include "trace_gfs2.h"
#define PULL 1
/**
* gfs2_struct2blk - compute stuff
* @sdp: the filesystem
* @nstruct: the number of structures
* @ssize: the size of the structures
*
* Compute the number of log descriptor blocks needed to hold a certain number
* of structures of a certain size.
*
* Returns: the number of blocks needed (minimum is always 1)
*/
unsigned int gfs2_struct2blk(struct gfs2_sbd *sdp, unsigned int nstruct,
unsigned int ssize)
{
unsigned int blks;
unsigned int first, second;
blks = 1;
first = (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_log_descriptor)) / ssize;
if (nstruct > first) {
second = (sdp->sd_sb.sb_bsize -
sizeof(struct gfs2_meta_header)) / ssize;
blks += DIV_ROUND_UP(nstruct - first, second);
}
return blks;
}
/**
* gfs2_remove_from_ail - Remove an entry from the ail lists, updating counters
* @mapping: The associated mapping (maybe NULL)
* @bd: The gfs2_bufdata to remove
*
* The ail lock _must_ be held when calling this function
*
*/
void gfs2_remove_from_ail(struct gfs2_bufdata *bd)
{
bd->bd_ail = NULL;
list_del_init(&bd->bd_ail_st_list);
list_del_init(&bd->bd_ail_gl_list);
atomic_dec(&bd->bd_gl->gl_ail_count);
brelse(bd->bd_bh);
}
/**
* gfs2_ail1_start_one - Start I/O on a part of the AIL
* @sdp: the filesystem
* @wbc: The writeback control structure
* @ai: The ail structure
*
*/
static int gfs2_ail1_start_one(struct gfs2_sbd *sdp,
struct writeback_control *wbc,
struct gfs2_ail *ai)
__releases(&sdp->sd_ail_lock)
__acquires(&sdp->sd_ail_lock)
{
struct gfs2_glock *gl = NULL;
struct address_space *mapping;
struct gfs2_bufdata *bd, *s;
struct buffer_head *bh;
list_for_each_entry_safe_reverse(bd, s, &ai->ai_ail1_list, bd_ail_st_list) {
bh = bd->bd_bh;
gfs2_assert(sdp, bd->bd_ail == ai);
if (!buffer_busy(bh)) {
if (!buffer_uptodate(bh))
gfs2_io_error_bh(sdp, bh);
list_move(&bd->bd_ail_st_list, &ai->ai_ail2_list);
continue;
}
if (!buffer_dirty(bh))
continue;
if (gl == bd->bd_gl)
continue;
gl = bd->bd_gl;
list_move(&bd->bd_ail_st_list, &ai->ai_ail1_list);
mapping = bh->b_page->mapping;
if (!mapping)
continue;
spin_unlock(&sdp->sd_ail_lock);
generic_writepages(mapping, wbc);
spin_lock(&sdp->sd_ail_lock);
if (wbc->nr_to_write <= 0)
break;
return 1;
}
return 0;
}
/**
* gfs2_ail1_flush - start writeback of some ail1 entries
* @sdp: The super block
* @wbc: The writeback control structure
*
* Writes back some ail1 entries, according to the limits in the
* writeback control structure
*/
void gfs2_ail1_flush(struct gfs2_sbd *sdp, struct writeback_control *wbc)
{
struct list_head *head = &sdp->sd_ail1_list;
struct gfs2_ail *ai;
trace_gfs2_ail_flush(sdp, wbc, 1);
spin_lock(&sdp->sd_ail_lock);
restart:
list_for_each_entry_reverse(ai, head, ai_list) {
if (wbc->nr_to_write <= 0)
break;
if (gfs2_ail1_start_one(sdp, wbc, ai))
goto restart;
}
spin_unlock(&sdp->sd_ail_lock);
trace_gfs2_ail_flush(sdp, wbc, 0);
}
/**
* gfs2_ail1_start - start writeback of all ail1 entries
* @sdp: The superblock
*/
static void gfs2_ail1_start(struct gfs2_sbd *sdp)
{
struct writeback_control wbc = {
.sync_mode = WB_SYNC_NONE,
.nr_to_write = LONG_MAX,
.range_start = 0,
.range_end = LLONG_MAX,
};
return gfs2_ail1_flush(sdp, &wbc);
}
/**
* gfs2_ail1_empty_one - Check whether or not a trans in the AIL has been synced
* @sdp: the filesystem
* @ai: the AIL entry
*
*/
static void gfs2_ail1_empty_one(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
{
struct gfs2_bufdata *bd, *s;
struct buffer_head *bh;
list_for_each_entry_safe_reverse(bd, s, &ai->ai_ail1_list,
bd_ail_st_list) {
bh = bd->bd_bh;
gfs2_assert(sdp, bd->bd_ail == ai);
if (buffer_busy(bh))
continue;
if (!buffer_uptodate(bh))
gfs2_io_error_bh(sdp, bh);
list_move(&bd->bd_ail_st_list, &ai->ai_ail2_list);
}
}
/**
* gfs2_ail1_empty - Try to empty the ail1 lists
* @sdp: The superblock
*
* Tries to empty the ail1 lists, starting with the oldest first
*/
static int gfs2_ail1_empty(struct gfs2_sbd *sdp)
{
struct gfs2_ail *ai, *s;
int ret;
spin_lock(&sdp->sd_ail_lock);
list_for_each_entry_safe_reverse(ai, s, &sdp->sd_ail1_list, ai_list) {
gfs2_ail1_empty_one(sdp, ai);
if (list_empty(&ai->ai_ail1_list))
list_move(&ai->ai_list, &sdp->sd_ail2_list);
else
break;
}
ret = list_empty(&sdp->sd_ail1_list);
spin_unlock(&sdp->sd_ail_lock);
return ret;
}
static void gfs2_ail1_wait(struct gfs2_sbd *sdp)
{
struct gfs2_ail *ai;
struct gfs2_bufdata *bd;
struct buffer_head *bh;
spin_lock(&sdp->sd_ail_lock);
list_for_each_entry_reverse(ai, &sdp->sd_ail1_list, ai_list) {
list_for_each_entry(bd, &ai->ai_ail1_list, bd_ail_st_list) {
bh = bd->bd_bh;
if (!buffer_locked(bh))
continue;
get_bh(bh);
spin_unlock(&sdp->sd_ail_lock);
wait_on_buffer(bh);
brelse(bh);
return;
}
}
spin_unlock(&sdp->sd_ail_lock);
}
/**
* gfs2_ail2_empty_one - Check whether or not a trans in the AIL has been synced
* @sdp: the filesystem
* @ai: the AIL entry
*
*/
static void gfs2_ail2_empty_one(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
{
struct list_head *head = &ai->ai_ail2_list;
struct gfs2_bufdata *bd;
while (!list_empty(head)) {
bd = list_entry(head->prev, struct gfs2_bufdata,
bd_ail_st_list);
gfs2_assert(sdp, bd->bd_ail == ai);
gfs2_remove_from_ail(bd);
}
}
static void ail2_empty(struct gfs2_sbd *sdp, unsigned int new_tail)
{
struct gfs2_ail *ai, *safe;
unsigned int old_tail = sdp->sd_log_tail;
int wrap = (new_tail < old_tail);
int a, b, rm;
spin_lock(&sdp->sd_ail_lock);
list_for_each_entry_safe(ai, safe, &sdp->sd_ail2_list, ai_list) {
a = (old_tail <= ai->ai_first);
b = (ai->ai_first < new_tail);
rm = (wrap) ? (a || b) : (a && b);
if (!rm)
continue;
gfs2_ail2_empty_one(sdp, ai);
list_del(&ai->ai_list);
gfs2_assert_warn(sdp, list_empty(&ai->ai_ail1_list));
gfs2_assert_warn(sdp, list_empty(&ai->ai_ail2_list));
kfree(ai);
}
spin_unlock(&sdp->sd_ail_lock);
}
/**
* gfs2_log_reserve - Make a log reservation
* @sdp: The GFS2 superblock
* @blks: The number of blocks to reserve
*
* Note that we never give out the last few blocks of the journal. Thats
[GFS2] assertion failure after writing to journaled file, umount This patch passes all my nasty tests that were causing the code to fail under one circumstance or another. Here is a complete summary of all changes from today's git tree, in order of appearance: 1. There are now separate variables for metadata buffer accounting. 2. Variable sd_log_num_hdrs is no longer needed, since the header accounting is taken care of by the reserve/refund sequence. 3. Fixed a tiny grammatical problem in a comment. 4. Added a new function "calc_reserved" to calculate the reserved log space. This isn't entirely necessary, but it has two benefits: First, it simplifies the gfs2_log_refund function greatly. Second, it allows for easier debugging because I could sprinkle the code with calls to this function to make sure the accounting is proper (by adding asserts and printks) at strategic point of the code. 5. In log_pull_tail there apparently was a kludge to fix up the accounting based on a "pull" parameter. The buffer accounting is now done properly, so the kludge was removed. 6. File sync operations were making a call to gfs2_log_flush that writes another journal header. Since that header was unplanned for (reserved) by the reserve/refund sequence, the free space had to be decremented so that when log_pull_tail gets called, the free space is be adjusted properly. (Did I hear you call that a kludge? well, maybe, but a lot more justifiable than the one I removed). 7. In the gfs2_log_shutdown code, it optionally syncs the log by specifying the PULL parameter to log_write_header. I'm not sure this is necessary anymore. It just seems to me there could be cases where shutdown is called while there are outstanding log buffers. 8. In the (data)buf_lo_before_commit functions, I changed some offset values from being calculated on the fly to being constants. That simplified some code and we might as well let the compiler do the calculation once rather than redoing those cycles at run time. 9. This version has my rewritten databuf_lo_add function. This version is much more like its predecessor, buf_lo_add, which makes it easier to understand. Again, this might not be necessary, but it seems as if this one works as well as the previous one, maybe even better, so I decided to leave it in. 10. In databuf_lo_before_commit, a previous data corruption problem was caused by going off the end of the buffer. The proper solution is to have the proper limit in place, rather than stopping earlier. (Thus my previous attempt to fix it is wrong). If you don't wrap the buffer, you're stopping too early and that causes more log buffer accounting problems. 11. In lops.h there are two new (previously mentioned) constants for figuring out the data offset for the journal buffers. 12. There are also two new functions, buf_limit and databuf_limit to calculate how many entries will fit in the buffer. 13. In function gfs2_meta_wipe, it needs to distinguish between pinned metadata buffers and journaled data buffers for proper journal buffer accounting. It can't use the JDATA gfs2_inode flag because it's sometimes passed the "real" inode and sometimes the "metadata inode" and the inode flags will be random bits in a metadata gfs2_inode. It needs to base its decision on which was passed in. Signed-off-by: Bob Peterson <rpeterso@redhat.com> Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
2007-06-18 19:50:20 +00:00
* due to the fact that there is a small number of header blocks
* associated with each log flush. The exact number can't be known until
* flush time, so we ensure that we have just enough free blocks at all
* times to avoid running out during a log flush.
*
* We no longer flush the log here, instead we wake up logd to do that
* for us. To avoid the thundering herd and to ensure that we deal fairly
* with queued waiters, we use an exclusive wait. This means that when we
* get woken with enough journal space to get our reservation, we need to
* wake the next waiter on the list.
*
* Returns: errno
*/
int gfs2_log_reserve(struct gfs2_sbd *sdp, unsigned int blks)
{
unsigned reserved_blks = 6 * (4096 / sdp->sd_vfs->s_blocksize);
unsigned wanted = blks + reserved_blks;
DEFINE_WAIT(wait);
int did_wait = 0;
unsigned int free_blocks;
if (gfs2_assert_warn(sdp, blks) ||
gfs2_assert_warn(sdp, blks <= sdp->sd_jdesc->jd_blocks))
return -EINVAL;
retry:
free_blocks = atomic_read(&sdp->sd_log_blks_free);
if (unlikely(free_blocks <= wanted)) {
do {
prepare_to_wait_exclusive(&sdp->sd_log_waitq, &wait,
TASK_UNINTERRUPTIBLE);
wake_up(&sdp->sd_logd_waitq);
did_wait = 1;
if (atomic_read(&sdp->sd_log_blks_free) <= wanted)
io_schedule();
free_blocks = atomic_read(&sdp->sd_log_blks_free);
} while(free_blocks <= wanted);
finish_wait(&sdp->sd_log_waitq, &wait);
}
if (atomic_cmpxchg(&sdp->sd_log_blks_free, free_blocks,
free_blocks - blks) != free_blocks)
goto retry;
trace_gfs2_log_blocks(sdp, -blks);
/*
* If we waited, then so might others, wake them up _after_ we get
* our share of the log.
*/
if (unlikely(did_wait))
wake_up(&sdp->sd_log_waitq);
down_read(&sdp->sd_log_flush_lock);
return 0;
}
u64 gfs2_log_bmap(struct gfs2_sbd *sdp, unsigned int lbn)
{
struct gfs2_journal_extent *je;
list_for_each_entry(je, &sdp->sd_jdesc->extent_list, extent_list) {
if (lbn >= je->lblock && lbn < je->lblock + je->blocks)
return je->dblock + lbn - je->lblock;
}
return -1;
}
/**
* log_distance - Compute distance between two journal blocks
* @sdp: The GFS2 superblock
* @newer: The most recent journal block of the pair
* @older: The older journal block of the pair
*
* Compute the distance (in the journal direction) between two
* blocks in the journal
*
* Returns: the distance in blocks
*/
static inline unsigned int log_distance(struct gfs2_sbd *sdp, unsigned int newer,
unsigned int older)
{
int dist;
dist = newer - older;
if (dist < 0)
dist += sdp->sd_jdesc->jd_blocks;
return dist;
}
[GFS2] assertion failure after writing to journaled file, umount This patch passes all my nasty tests that were causing the code to fail under one circumstance or another. Here is a complete summary of all changes from today's git tree, in order of appearance: 1. There are now separate variables for metadata buffer accounting. 2. Variable sd_log_num_hdrs is no longer needed, since the header accounting is taken care of by the reserve/refund sequence. 3. Fixed a tiny grammatical problem in a comment. 4. Added a new function "calc_reserved" to calculate the reserved log space. This isn't entirely necessary, but it has two benefits: First, it simplifies the gfs2_log_refund function greatly. Second, it allows for easier debugging because I could sprinkle the code with calls to this function to make sure the accounting is proper (by adding asserts and printks) at strategic point of the code. 5. In log_pull_tail there apparently was a kludge to fix up the accounting based on a "pull" parameter. The buffer accounting is now done properly, so the kludge was removed. 6. File sync operations were making a call to gfs2_log_flush that writes another journal header. Since that header was unplanned for (reserved) by the reserve/refund sequence, the free space had to be decremented so that when log_pull_tail gets called, the free space is be adjusted properly. (Did I hear you call that a kludge? well, maybe, but a lot more justifiable than the one I removed). 7. In the gfs2_log_shutdown code, it optionally syncs the log by specifying the PULL parameter to log_write_header. I'm not sure this is necessary anymore. It just seems to me there could be cases where shutdown is called while there are outstanding log buffers. 8. In the (data)buf_lo_before_commit functions, I changed some offset values from being calculated on the fly to being constants. That simplified some code and we might as well let the compiler do the calculation once rather than redoing those cycles at run time. 9. This version has my rewritten databuf_lo_add function. This version is much more like its predecessor, buf_lo_add, which makes it easier to understand. Again, this might not be necessary, but it seems as if this one works as well as the previous one, maybe even better, so I decided to leave it in. 10. In databuf_lo_before_commit, a previous data corruption problem was caused by going off the end of the buffer. The proper solution is to have the proper limit in place, rather than stopping earlier. (Thus my previous attempt to fix it is wrong). If you don't wrap the buffer, you're stopping too early and that causes more log buffer accounting problems. 11. In lops.h there are two new (previously mentioned) constants for figuring out the data offset for the journal buffers. 12. There are also two new functions, buf_limit and databuf_limit to calculate how many entries will fit in the buffer. 13. In function gfs2_meta_wipe, it needs to distinguish between pinned metadata buffers and journaled data buffers for proper journal buffer accounting. It can't use the JDATA gfs2_inode flag because it's sometimes passed the "real" inode and sometimes the "metadata inode" and the inode flags will be random bits in a metadata gfs2_inode. It needs to base its decision on which was passed in. Signed-off-by: Bob Peterson <rpeterso@redhat.com> Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
2007-06-18 19:50:20 +00:00
/**
* calc_reserved - Calculate the number of blocks to reserve when
* refunding a transaction's unused buffers.
* @sdp: The GFS2 superblock
*
* This is complex. We need to reserve room for all our currently used
* metadata buffers (e.g. normal file I/O rewriting file time stamps) and
* all our journaled data buffers for journaled files (e.g. files in the
* meta_fs like rindex, or files for which chattr +j was done.)
* If we don't reserve enough space, gfs2_log_refund and gfs2_log_flush
* will count it as free space (sd_log_blks_free) and corruption will follow.
*
* We can have metadata bufs and jdata bufs in the same journal. So each
* type gets its own log header, for which we need to reserve a block.
* In fact, each type has the potential for needing more than one header
* in cases where we have more buffers than will fit on a journal page.
* Metadata journal entries take up half the space of journaled buffer entries.
* Thus, metadata entries have buf_limit (502) and journaled buffers have
* databuf_limit (251) before they cause a wrap around.
*
* Also, we need to reserve blocks for revoke journal entries and one for an
* overall header for the lot.
*
* Returns: the number of blocks reserved
*/
static unsigned int calc_reserved(struct gfs2_sbd *sdp)
{
unsigned int reserved = 0;
unsigned int mbuf_limit, metabufhdrs_needed;
unsigned int dbuf_limit, databufhdrs_needed;
unsigned int revokes = 0;
mbuf_limit = buf_limit(sdp);
metabufhdrs_needed = (sdp->sd_log_commited_buf +
(mbuf_limit - 1)) / mbuf_limit;
dbuf_limit = databuf_limit(sdp);
databufhdrs_needed = (sdp->sd_log_commited_databuf +
(dbuf_limit - 1)) / dbuf_limit;
if (sdp->sd_log_commited_revoke > 0)
[GFS2] assertion failure after writing to journaled file, umount This patch passes all my nasty tests that were causing the code to fail under one circumstance or another. Here is a complete summary of all changes from today's git tree, in order of appearance: 1. There are now separate variables for metadata buffer accounting. 2. Variable sd_log_num_hdrs is no longer needed, since the header accounting is taken care of by the reserve/refund sequence. 3. Fixed a tiny grammatical problem in a comment. 4. Added a new function "calc_reserved" to calculate the reserved log space. This isn't entirely necessary, but it has two benefits: First, it simplifies the gfs2_log_refund function greatly. Second, it allows for easier debugging because I could sprinkle the code with calls to this function to make sure the accounting is proper (by adding asserts and printks) at strategic point of the code. 5. In log_pull_tail there apparently was a kludge to fix up the accounting based on a "pull" parameter. The buffer accounting is now done properly, so the kludge was removed. 6. File sync operations were making a call to gfs2_log_flush that writes another journal header. Since that header was unplanned for (reserved) by the reserve/refund sequence, the free space had to be decremented so that when log_pull_tail gets called, the free space is be adjusted properly. (Did I hear you call that a kludge? well, maybe, but a lot more justifiable than the one I removed). 7. In the gfs2_log_shutdown code, it optionally syncs the log by specifying the PULL parameter to log_write_header. I'm not sure this is necessary anymore. It just seems to me there could be cases where shutdown is called while there are outstanding log buffers. 8. In the (data)buf_lo_before_commit functions, I changed some offset values from being calculated on the fly to being constants. That simplified some code and we might as well let the compiler do the calculation once rather than redoing those cycles at run time. 9. This version has my rewritten databuf_lo_add function. This version is much more like its predecessor, buf_lo_add, which makes it easier to understand. Again, this might not be necessary, but it seems as if this one works as well as the previous one, maybe even better, so I decided to leave it in. 10. In databuf_lo_before_commit, a previous data corruption problem was caused by going off the end of the buffer. The proper solution is to have the proper limit in place, rather than stopping earlier. (Thus my previous attempt to fix it is wrong). If you don't wrap the buffer, you're stopping too early and that causes more log buffer accounting problems. 11. In lops.h there are two new (previously mentioned) constants for figuring out the data offset for the journal buffers. 12. There are also two new functions, buf_limit and databuf_limit to calculate how many entries will fit in the buffer. 13. In function gfs2_meta_wipe, it needs to distinguish between pinned metadata buffers and journaled data buffers for proper journal buffer accounting. It can't use the JDATA gfs2_inode flag because it's sometimes passed the "real" inode and sometimes the "metadata inode" and the inode flags will be random bits in a metadata gfs2_inode. It needs to base its decision on which was passed in. Signed-off-by: Bob Peterson <rpeterso@redhat.com> Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
2007-06-18 19:50:20 +00:00
revokes = gfs2_struct2blk(sdp, sdp->sd_log_commited_revoke,
sizeof(u64));
reserved = sdp->sd_log_commited_buf + metabufhdrs_needed +
sdp->sd_log_commited_databuf + databufhdrs_needed +
revokes;
/* One for the overall header */
if (reserved)
reserved++;
return reserved;
}
static unsigned int current_tail(struct gfs2_sbd *sdp)
{
struct gfs2_ail *ai;
unsigned int tail;
spin_lock(&sdp->sd_ail_lock);
if (list_empty(&sdp->sd_ail1_list)) {
tail = sdp->sd_log_head;
} else {
ai = list_entry(sdp->sd_ail1_list.prev, struct gfs2_ail, ai_list);
tail = ai->ai_first;
}
spin_unlock(&sdp->sd_ail_lock);
return tail;
}
void gfs2_log_incr_head(struct gfs2_sbd *sdp)
{
BUG_ON((sdp->sd_log_flush_head == sdp->sd_log_tail) &&
(sdp->sd_log_flush_head != sdp->sd_log_head));
if (++sdp->sd_log_flush_head == sdp->sd_jdesc->jd_blocks) {
sdp->sd_log_flush_head = 0;
sdp->sd_log_flush_wrapped = 1;
}
}
[GFS2] assertion failure after writing to journaled file, umount This patch passes all my nasty tests that were causing the code to fail under one circumstance or another. Here is a complete summary of all changes from today's git tree, in order of appearance: 1. There are now separate variables for metadata buffer accounting. 2. Variable sd_log_num_hdrs is no longer needed, since the header accounting is taken care of by the reserve/refund sequence. 3. Fixed a tiny grammatical problem in a comment. 4. Added a new function "calc_reserved" to calculate the reserved log space. This isn't entirely necessary, but it has two benefits: First, it simplifies the gfs2_log_refund function greatly. Second, it allows for easier debugging because I could sprinkle the code with calls to this function to make sure the accounting is proper (by adding asserts and printks) at strategic point of the code. 5. In log_pull_tail there apparently was a kludge to fix up the accounting based on a "pull" parameter. The buffer accounting is now done properly, so the kludge was removed. 6. File sync operations were making a call to gfs2_log_flush that writes another journal header. Since that header was unplanned for (reserved) by the reserve/refund sequence, the free space had to be decremented so that when log_pull_tail gets called, the free space is be adjusted properly. (Did I hear you call that a kludge? well, maybe, but a lot more justifiable than the one I removed). 7. In the gfs2_log_shutdown code, it optionally syncs the log by specifying the PULL parameter to log_write_header. I'm not sure this is necessary anymore. It just seems to me there could be cases where shutdown is called while there are outstanding log buffers. 8. In the (data)buf_lo_before_commit functions, I changed some offset values from being calculated on the fly to being constants. That simplified some code and we might as well let the compiler do the calculation once rather than redoing those cycles at run time. 9. This version has my rewritten databuf_lo_add function. This version is much more like its predecessor, buf_lo_add, which makes it easier to understand. Again, this might not be necessary, but it seems as if this one works as well as the previous one, maybe even better, so I decided to leave it in. 10. In databuf_lo_before_commit, a previous data corruption problem was caused by going off the end of the buffer. The proper solution is to have the proper limit in place, rather than stopping earlier. (Thus my previous attempt to fix it is wrong). If you don't wrap the buffer, you're stopping too early and that causes more log buffer accounting problems. 11. In lops.h there are two new (previously mentioned) constants for figuring out the data offset for the journal buffers. 12. There are also two new functions, buf_limit and databuf_limit to calculate how many entries will fit in the buffer. 13. In function gfs2_meta_wipe, it needs to distinguish between pinned metadata buffers and journaled data buffers for proper journal buffer accounting. It can't use the JDATA gfs2_inode flag because it's sometimes passed the "real" inode and sometimes the "metadata inode" and the inode flags will be random bits in a metadata gfs2_inode. It needs to base its decision on which was passed in. Signed-off-by: Bob Peterson <rpeterso@redhat.com> Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
2007-06-18 19:50:20 +00:00
static void log_pull_tail(struct gfs2_sbd *sdp, unsigned int new_tail)
{
unsigned int dist = log_distance(sdp, new_tail, sdp->sd_log_tail);
ail2_empty(sdp, new_tail);
atomic_add(dist, &sdp->sd_log_blks_free);
trace_gfs2_log_blocks(sdp, dist);
gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <=
sdp->sd_jdesc->jd_blocks);
sdp->sd_log_tail = new_tail;
}
/**
* log_write_header - Get and initialize a journal header buffer
* @sdp: The GFS2 superblock
*
* Returns: the initialized log buffer descriptor
*/
static void log_write_header(struct gfs2_sbd *sdp, u32 flags, int pull)
{
u64 blkno = gfs2_log_bmap(sdp, sdp->sd_log_flush_head);
struct buffer_head *bh;
struct gfs2_log_header *lh;
unsigned int tail;
u32 hash;
bh = sb_getblk(sdp->sd_vfs, blkno);
lock_buffer(bh);
memset(bh->b_data, 0, bh->b_size);
set_buffer_uptodate(bh);
clear_buffer_dirty(bh);
gfs2_ail1_empty(sdp);
tail = current_tail(sdp);
lh = (struct gfs2_log_header *)bh->b_data;
memset(lh, 0, sizeof(struct gfs2_log_header));
lh->lh_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
lh->lh_header.mh_type = cpu_to_be32(GFS2_METATYPE_LH);
GFS2: Tag all metadata with jid There are two spare field in the header common to all GFS2 metadata. One is just the right size to fit a journal id in it, and this patch updates the journal code so that each time a metadata block is modified, we tag it with the journal id of the node which is performing the modification. The reason for this is that it should make it much easier to debug issues which arise if we can tell which node was the last to modify a particular metadata block. Since the field is updated before the block is written into the journal, each journal should only contain metadata which is tagged with its own journal id. The one exception to this is the journal header block, which might have a different node's id in it, if that journal was recovered by another node in the cluster. Thus each journal will contain a record of which nodes recovered it, via the journal header. The other field in the metadata header could potentially be used to hold information about what kind of operation was performed, but for the time being we just zero it on each transaction so that if we use it for that in future, we'll know that the information (where it exists) is reliable. I did consider using the other field to hold the journal sequence number, however since in GFS2's journaling we write the modified data into the journal and not the original data, this gives no information as to what action caused the modification, so I think we can probably come up with a better use for those 64 bits in the future. Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
2009-11-06 16:20:51 +00:00
lh->lh_header.__pad0 = cpu_to_be64(0);
lh->lh_header.mh_format = cpu_to_be32(GFS2_FORMAT_LH);
GFS2: Tag all metadata with jid There are two spare field in the header common to all GFS2 metadata. One is just the right size to fit a journal id in it, and this patch updates the journal code so that each time a metadata block is modified, we tag it with the journal id of the node which is performing the modification. The reason for this is that it should make it much easier to debug issues which arise if we can tell which node was the last to modify a particular metadata block. Since the field is updated before the block is written into the journal, each journal should only contain metadata which is tagged with its own journal id. The one exception to this is the journal header block, which might have a different node's id in it, if that journal was recovered by another node in the cluster. Thus each journal will contain a record of which nodes recovered it, via the journal header. The other field in the metadata header could potentially be used to hold information about what kind of operation was performed, but for the time being we just zero it on each transaction so that if we use it for that in future, we'll know that the information (where it exists) is reliable. I did consider using the other field to hold the journal sequence number, however since in GFS2's journaling we write the modified data into the journal and not the original data, this gives no information as to what action caused the modification, so I think we can probably come up with a better use for those 64 bits in the future. Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
2009-11-06 16:20:51 +00:00
lh->lh_header.mh_jid = cpu_to_be32(sdp->sd_jdesc->jd_jid);
lh->lh_sequence = cpu_to_be64(sdp->sd_log_sequence++);
lh->lh_flags = cpu_to_be32(flags);
lh->lh_tail = cpu_to_be32(tail);
lh->lh_blkno = cpu_to_be32(sdp->sd_log_flush_head);
hash = gfs2_disk_hash(bh->b_data, sizeof(struct gfs2_log_header));
lh->lh_hash = cpu_to_be32(hash);
bh->b_end_io = end_buffer_write_sync;
get_bh(bh);
if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags))
submit_bh(WRITE_SYNC | REQ_META | REQ_PRIO, bh);
else
submit_bh(WRITE_FLUSH_FUA | REQ_META, bh);
wait_on_buffer(bh);
if (!buffer_uptodate(bh))
gfs2_io_error_bh(sdp, bh);
brelse(bh);
if (sdp->sd_log_tail != tail)
[GFS2] assertion failure after writing to journaled file, umount This patch passes all my nasty tests that were causing the code to fail under one circumstance or another. Here is a complete summary of all changes from today's git tree, in order of appearance: 1. There are now separate variables for metadata buffer accounting. 2. Variable sd_log_num_hdrs is no longer needed, since the header accounting is taken care of by the reserve/refund sequence. 3. Fixed a tiny grammatical problem in a comment. 4. Added a new function "calc_reserved" to calculate the reserved log space. This isn't entirely necessary, but it has two benefits: First, it simplifies the gfs2_log_refund function greatly. Second, it allows for easier debugging because I could sprinkle the code with calls to this function to make sure the accounting is proper (by adding asserts and printks) at strategic point of the code. 5. In log_pull_tail there apparently was a kludge to fix up the accounting based on a "pull" parameter. The buffer accounting is now done properly, so the kludge was removed. 6. File sync operations were making a call to gfs2_log_flush that writes another journal header. Since that header was unplanned for (reserved) by the reserve/refund sequence, the free space had to be decremented so that when log_pull_tail gets called, the free space is be adjusted properly. (Did I hear you call that a kludge? well, maybe, but a lot more justifiable than the one I removed). 7. In the gfs2_log_shutdown code, it optionally syncs the log by specifying the PULL parameter to log_write_header. I'm not sure this is necessary anymore. It just seems to me there could be cases where shutdown is called while there are outstanding log buffers. 8. In the (data)buf_lo_before_commit functions, I changed some offset values from being calculated on the fly to being constants. That simplified some code and we might as well let the compiler do the calculation once rather than redoing those cycles at run time. 9. This version has my rewritten databuf_lo_add function. This version is much more like its predecessor, buf_lo_add, which makes it easier to understand. Again, this might not be necessary, but it seems as if this one works as well as the previous one, maybe even better, so I decided to leave it in. 10. In databuf_lo_before_commit, a previous data corruption problem was caused by going off the end of the buffer. The proper solution is to have the proper limit in place, rather than stopping earlier. (Thus my previous attempt to fix it is wrong). If you don't wrap the buffer, you're stopping too early and that causes more log buffer accounting problems. 11. In lops.h there are two new (previously mentioned) constants for figuring out the data offset for the journal buffers. 12. There are also two new functions, buf_limit and databuf_limit to calculate how many entries will fit in the buffer. 13. In function gfs2_meta_wipe, it needs to distinguish between pinned metadata buffers and journaled data buffers for proper journal buffer accounting. It can't use the JDATA gfs2_inode flag because it's sometimes passed the "real" inode and sometimes the "metadata inode" and the inode flags will be random bits in a metadata gfs2_inode. It needs to base its decision on which was passed in. Signed-off-by: Bob Peterson <rpeterso@redhat.com> Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
2007-06-18 19:50:20 +00:00
log_pull_tail(sdp, tail);
else
gfs2_assert_withdraw(sdp, !pull);
sdp->sd_log_idle = (tail == sdp->sd_log_flush_head);
gfs2_log_incr_head(sdp);
}
static void log_flush_commit(struct gfs2_sbd *sdp)
{
DEFINE_WAIT(wait);
if (atomic_read(&sdp->sd_log_in_flight)) {
do {
prepare_to_wait(&sdp->sd_log_flush_wait, &wait,
TASK_UNINTERRUPTIBLE);
if (atomic_read(&sdp->sd_log_in_flight))
io_schedule();
} while(atomic_read(&sdp->sd_log_in_flight));
finish_wait(&sdp->sd_log_flush_wait, &wait);
}
log_write_header(sdp, 0, 0);
}
int bd_cmp(void *priv, struct list_head *a, struct list_head *b)
{
struct gfs2_bufdata *bda, *bdb;
bda = list_entry(a, struct gfs2_bufdata, bd_le.le_list);
bdb = list_entry(b, struct gfs2_bufdata, bd_le.le_list);
if (bda->bd_bh->b_blocknr < bdb->bd_bh->b_blocknr)
return -1;
if (bda->bd_bh->b_blocknr > bdb->bd_bh->b_blocknr)
return 1;
return 0;
}
static void gfs2_ordered_write(struct gfs2_sbd *sdp)
{
struct gfs2_bufdata *bd;
struct buffer_head *bh;
LIST_HEAD(written);
gfs2_log_lock(sdp);
list_sort(NULL, &sdp->sd_log_le_ordered, &bd_cmp);
while (!list_empty(&sdp->sd_log_le_ordered)) {
bd = list_entry(sdp->sd_log_le_ordered.next, struct gfs2_bufdata, bd_le.le_list);
list_move(&bd->bd_le.le_list, &written);
bh = bd->bd_bh;
if (!buffer_dirty(bh))
continue;
get_bh(bh);
gfs2_log_unlock(sdp);
lock_buffer(bh);
if (buffer_mapped(bh) && test_clear_buffer_dirty(bh)) {
bh->b_end_io = end_buffer_write_sync;
submit_bh(WRITE_SYNC, bh);
} else {
unlock_buffer(bh);
brelse(bh);
}
gfs2_log_lock(sdp);
}
list_splice(&written, &sdp->sd_log_le_ordered);
gfs2_log_unlock(sdp);
}
static void gfs2_ordered_wait(struct gfs2_sbd *sdp)
{
struct gfs2_bufdata *bd;
struct buffer_head *bh;
gfs2_log_lock(sdp);
while (!list_empty(&sdp->sd_log_le_ordered)) {
bd = list_entry(sdp->sd_log_le_ordered.prev, struct gfs2_bufdata, bd_le.le_list);
bh = bd->bd_bh;
if (buffer_locked(bh)) {
get_bh(bh);
gfs2_log_unlock(sdp);
wait_on_buffer(bh);
brelse(bh);
gfs2_log_lock(sdp);
continue;
}
list_del_init(&bd->bd_le.le_list);
}
gfs2_log_unlock(sdp);
}
/**
* gfs2_log_flush - flush incore transaction(s)
* @sdp: the filesystem
* @gl: The glock structure to flush. If NULL, flush the whole incore log
*
*/
void gfs2_log_flush(struct gfs2_sbd *sdp, struct gfs2_glock *gl)
{
struct gfs2_ail *ai;
down_write(&sdp->sd_log_flush_lock);
/* Log might have been flushed while we waited for the flush lock */
if (gl && !test_bit(GLF_LFLUSH, &gl->gl_flags)) {
up_write(&sdp->sd_log_flush_lock);
return;
}
trace_gfs2_log_flush(sdp, 1);
ai = kzalloc(sizeof(struct gfs2_ail), GFP_NOFS | __GFP_NOFAIL);
INIT_LIST_HEAD(&ai->ai_ail1_list);
INIT_LIST_HEAD(&ai->ai_ail2_list);
if (sdp->sd_log_num_buf != sdp->sd_log_commited_buf) {
printk(KERN_INFO "GFS2: log buf %u %u\n", sdp->sd_log_num_buf,
sdp->sd_log_commited_buf);
gfs2_assert_withdraw(sdp, 0);
}
if (sdp->sd_log_num_databuf != sdp->sd_log_commited_databuf) {
printk(KERN_INFO "GFS2: log databuf %u %u\n",
sdp->sd_log_num_databuf, sdp->sd_log_commited_databuf);
gfs2_assert_withdraw(sdp, 0);
}
gfs2_assert_withdraw(sdp,
sdp->sd_log_num_revoke == sdp->sd_log_commited_revoke);
sdp->sd_log_flush_head = sdp->sd_log_head;
sdp->sd_log_flush_wrapped = 0;
ai->ai_first = sdp->sd_log_flush_head;
gfs2_ordered_write(sdp);
lops_before_commit(sdp);
gfs2_ordered_wait(sdp);
if (sdp->sd_log_head != sdp->sd_log_flush_head)
log_flush_commit(sdp);
[GFS2] assertion failure after writing to journaled file, umount This patch passes all my nasty tests that were causing the code to fail under one circumstance or another. Here is a complete summary of all changes from today's git tree, in order of appearance: 1. There are now separate variables for metadata buffer accounting. 2. Variable sd_log_num_hdrs is no longer needed, since the header accounting is taken care of by the reserve/refund sequence. 3. Fixed a tiny grammatical problem in a comment. 4. Added a new function "calc_reserved" to calculate the reserved log space. This isn't entirely necessary, but it has two benefits: First, it simplifies the gfs2_log_refund function greatly. Second, it allows for easier debugging because I could sprinkle the code with calls to this function to make sure the accounting is proper (by adding asserts and printks) at strategic point of the code. 5. In log_pull_tail there apparently was a kludge to fix up the accounting based on a "pull" parameter. The buffer accounting is now done properly, so the kludge was removed. 6. File sync operations were making a call to gfs2_log_flush that writes another journal header. Since that header was unplanned for (reserved) by the reserve/refund sequence, the free space had to be decremented so that when log_pull_tail gets called, the free space is be adjusted properly. (Did I hear you call that a kludge? well, maybe, but a lot more justifiable than the one I removed). 7. In the gfs2_log_shutdown code, it optionally syncs the log by specifying the PULL parameter to log_write_header. I'm not sure this is necessary anymore. It just seems to me there could be cases where shutdown is called while there are outstanding log buffers. 8. In the (data)buf_lo_before_commit functions, I changed some offset values from being calculated on the fly to being constants. That simplified some code and we might as well let the compiler do the calculation once rather than redoing those cycles at run time. 9. This version has my rewritten databuf_lo_add function. This version is much more like its predecessor, buf_lo_add, which makes it easier to understand. Again, this might not be necessary, but it seems as if this one works as well as the previous one, maybe even better, so I decided to leave it in. 10. In databuf_lo_before_commit, a previous data corruption problem was caused by going off the end of the buffer. The proper solution is to have the proper limit in place, rather than stopping earlier. (Thus my previous attempt to fix it is wrong). If you don't wrap the buffer, you're stopping too early and that causes more log buffer accounting problems. 11. In lops.h there are two new (previously mentioned) constants for figuring out the data offset for the journal buffers. 12. There are also two new functions, buf_limit and databuf_limit to calculate how many entries will fit in the buffer. 13. In function gfs2_meta_wipe, it needs to distinguish between pinned metadata buffers and journaled data buffers for proper journal buffer accounting. It can't use the JDATA gfs2_inode flag because it's sometimes passed the "real" inode and sometimes the "metadata inode" and the inode flags will be random bits in a metadata gfs2_inode. It needs to base its decision on which was passed in. Signed-off-by: Bob Peterson <rpeterso@redhat.com> Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
2007-06-18 19:50:20 +00:00
else if (sdp->sd_log_tail != current_tail(sdp) && !sdp->sd_log_idle){
gfs2_log_lock(sdp);
atomic_dec(&sdp->sd_log_blks_free); /* Adjust for unreserved buffer */
trace_gfs2_log_blocks(sdp, -1);
[GFS2] assertion failure after writing to journaled file, umount This patch passes all my nasty tests that were causing the code to fail under one circumstance or another. Here is a complete summary of all changes from today's git tree, in order of appearance: 1. There are now separate variables for metadata buffer accounting. 2. Variable sd_log_num_hdrs is no longer needed, since the header accounting is taken care of by the reserve/refund sequence. 3. Fixed a tiny grammatical problem in a comment. 4. Added a new function "calc_reserved" to calculate the reserved log space. This isn't entirely necessary, but it has two benefits: First, it simplifies the gfs2_log_refund function greatly. Second, it allows for easier debugging because I could sprinkle the code with calls to this function to make sure the accounting is proper (by adding asserts and printks) at strategic point of the code. 5. In log_pull_tail there apparently was a kludge to fix up the accounting based on a "pull" parameter. The buffer accounting is now done properly, so the kludge was removed. 6. File sync operations were making a call to gfs2_log_flush that writes another journal header. Since that header was unplanned for (reserved) by the reserve/refund sequence, the free space had to be decremented so that when log_pull_tail gets called, the free space is be adjusted properly. (Did I hear you call that a kludge? well, maybe, but a lot more justifiable than the one I removed). 7. In the gfs2_log_shutdown code, it optionally syncs the log by specifying the PULL parameter to log_write_header. I'm not sure this is necessary anymore. It just seems to me there could be cases where shutdown is called while there are outstanding log buffers. 8. In the (data)buf_lo_before_commit functions, I changed some offset values from being calculated on the fly to being constants. That simplified some code and we might as well let the compiler do the calculation once rather than redoing those cycles at run time. 9. This version has my rewritten databuf_lo_add function. This version is much more like its predecessor, buf_lo_add, which makes it easier to understand. Again, this might not be necessary, but it seems as if this one works as well as the previous one, maybe even better, so I decided to leave it in. 10. In databuf_lo_before_commit, a previous data corruption problem was caused by going off the end of the buffer. The proper solution is to have the proper limit in place, rather than stopping earlier. (Thus my previous attempt to fix it is wrong). If you don't wrap the buffer, you're stopping too early and that causes more log buffer accounting problems. 11. In lops.h there are two new (previously mentioned) constants for figuring out the data offset for the journal buffers. 12. There are also two new functions, buf_limit and databuf_limit to calculate how many entries will fit in the buffer. 13. In function gfs2_meta_wipe, it needs to distinguish between pinned metadata buffers and journaled data buffers for proper journal buffer accounting. It can't use the JDATA gfs2_inode flag because it's sometimes passed the "real" inode and sometimes the "metadata inode" and the inode flags will be random bits in a metadata gfs2_inode. It needs to base its decision on which was passed in. Signed-off-by: Bob Peterson <rpeterso@redhat.com> Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
2007-06-18 19:50:20 +00:00
gfs2_log_unlock(sdp);
log_write_header(sdp, 0, PULL);
[GFS2] assertion failure after writing to journaled file, umount This patch passes all my nasty tests that were causing the code to fail under one circumstance or another. Here is a complete summary of all changes from today's git tree, in order of appearance: 1. There are now separate variables for metadata buffer accounting. 2. Variable sd_log_num_hdrs is no longer needed, since the header accounting is taken care of by the reserve/refund sequence. 3. Fixed a tiny grammatical problem in a comment. 4. Added a new function "calc_reserved" to calculate the reserved log space. This isn't entirely necessary, but it has two benefits: First, it simplifies the gfs2_log_refund function greatly. Second, it allows for easier debugging because I could sprinkle the code with calls to this function to make sure the accounting is proper (by adding asserts and printks) at strategic point of the code. 5. In log_pull_tail there apparently was a kludge to fix up the accounting based on a "pull" parameter. The buffer accounting is now done properly, so the kludge was removed. 6. File sync operations were making a call to gfs2_log_flush that writes another journal header. Since that header was unplanned for (reserved) by the reserve/refund sequence, the free space had to be decremented so that when log_pull_tail gets called, the free space is be adjusted properly. (Did I hear you call that a kludge? well, maybe, but a lot more justifiable than the one I removed). 7. In the gfs2_log_shutdown code, it optionally syncs the log by specifying the PULL parameter to log_write_header. I'm not sure this is necessary anymore. It just seems to me there could be cases where shutdown is called while there are outstanding log buffers. 8. In the (data)buf_lo_before_commit functions, I changed some offset values from being calculated on the fly to being constants. That simplified some code and we might as well let the compiler do the calculation once rather than redoing those cycles at run time. 9. This version has my rewritten databuf_lo_add function. This version is much more like its predecessor, buf_lo_add, which makes it easier to understand. Again, this might not be necessary, but it seems as if this one works as well as the previous one, maybe even better, so I decided to leave it in. 10. In databuf_lo_before_commit, a previous data corruption problem was caused by going off the end of the buffer. The proper solution is to have the proper limit in place, rather than stopping earlier. (Thus my previous attempt to fix it is wrong). If you don't wrap the buffer, you're stopping too early and that causes more log buffer accounting problems. 11. In lops.h there are two new (previously mentioned) constants for figuring out the data offset for the journal buffers. 12. There are also two new functions, buf_limit and databuf_limit to calculate how many entries will fit in the buffer. 13. In function gfs2_meta_wipe, it needs to distinguish between pinned metadata buffers and journaled data buffers for proper journal buffer accounting. It can't use the JDATA gfs2_inode flag because it's sometimes passed the "real" inode and sometimes the "metadata inode" and the inode flags will be random bits in a metadata gfs2_inode. It needs to base its decision on which was passed in. Signed-off-by: Bob Peterson <rpeterso@redhat.com> Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
2007-06-18 19:50:20 +00:00
}
lops_after_commit(sdp, ai);
gfs2_log_lock(sdp);
sdp->sd_log_head = sdp->sd_log_flush_head;
sdp->sd_log_blks_reserved = 0;
sdp->sd_log_commited_buf = 0;
[GFS2] assertion failure after writing to journaled file, umount This patch passes all my nasty tests that were causing the code to fail under one circumstance or another. Here is a complete summary of all changes from today's git tree, in order of appearance: 1. There are now separate variables for metadata buffer accounting. 2. Variable sd_log_num_hdrs is no longer needed, since the header accounting is taken care of by the reserve/refund sequence. 3. Fixed a tiny grammatical problem in a comment. 4. Added a new function "calc_reserved" to calculate the reserved log space. This isn't entirely necessary, but it has two benefits: First, it simplifies the gfs2_log_refund function greatly. Second, it allows for easier debugging because I could sprinkle the code with calls to this function to make sure the accounting is proper (by adding asserts and printks) at strategic point of the code. 5. In log_pull_tail there apparently was a kludge to fix up the accounting based on a "pull" parameter. The buffer accounting is now done properly, so the kludge was removed. 6. File sync operations were making a call to gfs2_log_flush that writes another journal header. Since that header was unplanned for (reserved) by the reserve/refund sequence, the free space had to be decremented so that when log_pull_tail gets called, the free space is be adjusted properly. (Did I hear you call that a kludge? well, maybe, but a lot more justifiable than the one I removed). 7. In the gfs2_log_shutdown code, it optionally syncs the log by specifying the PULL parameter to log_write_header. I'm not sure this is necessary anymore. It just seems to me there could be cases where shutdown is called while there are outstanding log buffers. 8. In the (data)buf_lo_before_commit functions, I changed some offset values from being calculated on the fly to being constants. That simplified some code and we might as well let the compiler do the calculation once rather than redoing those cycles at run time. 9. This version has my rewritten databuf_lo_add function. This version is much more like its predecessor, buf_lo_add, which makes it easier to understand. Again, this might not be necessary, but it seems as if this one works as well as the previous one, maybe even better, so I decided to leave it in. 10. In databuf_lo_before_commit, a previous data corruption problem was caused by going off the end of the buffer. The proper solution is to have the proper limit in place, rather than stopping earlier. (Thus my previous attempt to fix it is wrong). If you don't wrap the buffer, you're stopping too early and that causes more log buffer accounting problems. 11. In lops.h there are two new (previously mentioned) constants for figuring out the data offset for the journal buffers. 12. There are also two new functions, buf_limit and databuf_limit to calculate how many entries will fit in the buffer. 13. In function gfs2_meta_wipe, it needs to distinguish between pinned metadata buffers and journaled data buffers for proper journal buffer accounting. It can't use the JDATA gfs2_inode flag because it's sometimes passed the "real" inode and sometimes the "metadata inode" and the inode flags will be random bits in a metadata gfs2_inode. It needs to base its decision on which was passed in. Signed-off-by: Bob Peterson <rpeterso@redhat.com> Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
2007-06-18 19:50:20 +00:00
sdp->sd_log_commited_databuf = 0;
sdp->sd_log_commited_revoke = 0;
spin_lock(&sdp->sd_ail_lock);
if (!list_empty(&ai->ai_ail1_list)) {
list_add(&ai->ai_list, &sdp->sd_ail1_list);
ai = NULL;
}
spin_unlock(&sdp->sd_ail_lock);
gfs2_log_unlock(sdp);
trace_gfs2_log_flush(sdp, 0);
up_write(&sdp->sd_log_flush_lock);
kfree(ai);
}
static void log_refund(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
{
[GFS2] assertion failure after writing to journaled file, umount This patch passes all my nasty tests that were causing the code to fail under one circumstance or another. Here is a complete summary of all changes from today's git tree, in order of appearance: 1. There are now separate variables for metadata buffer accounting. 2. Variable sd_log_num_hdrs is no longer needed, since the header accounting is taken care of by the reserve/refund sequence. 3. Fixed a tiny grammatical problem in a comment. 4. Added a new function "calc_reserved" to calculate the reserved log space. This isn't entirely necessary, but it has two benefits: First, it simplifies the gfs2_log_refund function greatly. Second, it allows for easier debugging because I could sprinkle the code with calls to this function to make sure the accounting is proper (by adding asserts and printks) at strategic point of the code. 5. In log_pull_tail there apparently was a kludge to fix up the accounting based on a "pull" parameter. The buffer accounting is now done properly, so the kludge was removed. 6. File sync operations were making a call to gfs2_log_flush that writes another journal header. Since that header was unplanned for (reserved) by the reserve/refund sequence, the free space had to be decremented so that when log_pull_tail gets called, the free space is be adjusted properly. (Did I hear you call that a kludge? well, maybe, but a lot more justifiable than the one I removed). 7. In the gfs2_log_shutdown code, it optionally syncs the log by specifying the PULL parameter to log_write_header. I'm not sure this is necessary anymore. It just seems to me there could be cases where shutdown is called while there are outstanding log buffers. 8. In the (data)buf_lo_before_commit functions, I changed some offset values from being calculated on the fly to being constants. That simplified some code and we might as well let the compiler do the calculation once rather than redoing those cycles at run time. 9. This version has my rewritten databuf_lo_add function. This version is much more like its predecessor, buf_lo_add, which makes it easier to understand. Again, this might not be necessary, but it seems as if this one works as well as the previous one, maybe even better, so I decided to leave it in. 10. In databuf_lo_before_commit, a previous data corruption problem was caused by going off the end of the buffer. The proper solution is to have the proper limit in place, rather than stopping earlier. (Thus my previous attempt to fix it is wrong). If you don't wrap the buffer, you're stopping too early and that causes more log buffer accounting problems. 11. In lops.h there are two new (previously mentioned) constants for figuring out the data offset for the journal buffers. 12. There are also two new functions, buf_limit and databuf_limit to calculate how many entries will fit in the buffer. 13. In function gfs2_meta_wipe, it needs to distinguish between pinned metadata buffers and journaled data buffers for proper journal buffer accounting. It can't use the JDATA gfs2_inode flag because it's sometimes passed the "real" inode and sometimes the "metadata inode" and the inode flags will be random bits in a metadata gfs2_inode. It needs to base its decision on which was passed in. Signed-off-by: Bob Peterson <rpeterso@redhat.com> Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
2007-06-18 19:50:20 +00:00
unsigned int reserved;
unsigned int unused;
gfs2_log_lock(sdp);
sdp->sd_log_commited_buf += tr->tr_num_buf_new - tr->tr_num_buf_rm;
[GFS2] assertion failure after writing to journaled file, umount This patch passes all my nasty tests that were causing the code to fail under one circumstance or another. Here is a complete summary of all changes from today's git tree, in order of appearance: 1. There are now separate variables for metadata buffer accounting. 2. Variable sd_log_num_hdrs is no longer needed, since the header accounting is taken care of by the reserve/refund sequence. 3. Fixed a tiny grammatical problem in a comment. 4. Added a new function "calc_reserved" to calculate the reserved log space. This isn't entirely necessary, but it has two benefits: First, it simplifies the gfs2_log_refund function greatly. Second, it allows for easier debugging because I could sprinkle the code with calls to this function to make sure the accounting is proper (by adding asserts and printks) at strategic point of the code. 5. In log_pull_tail there apparently was a kludge to fix up the accounting based on a "pull" parameter. The buffer accounting is now done properly, so the kludge was removed. 6. File sync operations were making a call to gfs2_log_flush that writes another journal header. Since that header was unplanned for (reserved) by the reserve/refund sequence, the free space had to be decremented so that when log_pull_tail gets called, the free space is be adjusted properly. (Did I hear you call that a kludge? well, maybe, but a lot more justifiable than the one I removed). 7. In the gfs2_log_shutdown code, it optionally syncs the log by specifying the PULL parameter to log_write_header. I'm not sure this is necessary anymore. It just seems to me there could be cases where shutdown is called while there are outstanding log buffers. 8. In the (data)buf_lo_before_commit functions, I changed some offset values from being calculated on the fly to being constants. That simplified some code and we might as well let the compiler do the calculation once rather than redoing those cycles at run time. 9. This version has my rewritten databuf_lo_add function. This version is much more like its predecessor, buf_lo_add, which makes it easier to understand. Again, this might not be necessary, but it seems as if this one works as well as the previous one, maybe even better, so I decided to leave it in. 10. In databuf_lo_before_commit, a previous data corruption problem was caused by going off the end of the buffer. The proper solution is to have the proper limit in place, rather than stopping earlier. (Thus my previous attempt to fix it is wrong). If you don't wrap the buffer, you're stopping too early and that causes more log buffer accounting problems. 11. In lops.h there are two new (previously mentioned) constants for figuring out the data offset for the journal buffers. 12. There are also two new functions, buf_limit and databuf_limit to calculate how many entries will fit in the buffer. 13. In function gfs2_meta_wipe, it needs to distinguish between pinned metadata buffers and journaled data buffers for proper journal buffer accounting. It can't use the JDATA gfs2_inode flag because it's sometimes passed the "real" inode and sometimes the "metadata inode" and the inode flags will be random bits in a metadata gfs2_inode. It needs to base its decision on which was passed in. Signed-off-by: Bob Peterson <rpeterso@redhat.com> Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
2007-06-18 19:50:20 +00:00
sdp->sd_log_commited_databuf += tr->tr_num_databuf_new -
tr->tr_num_databuf_rm;
gfs2_assert_withdraw(sdp, (((int)sdp->sd_log_commited_buf) >= 0) ||
(((int)sdp->sd_log_commited_databuf) >= 0));
sdp->sd_log_commited_revoke += tr->tr_num_revoke - tr->tr_num_revoke_rm;
[GFS2] assertion failure after writing to journaled file, umount This patch passes all my nasty tests that were causing the code to fail under one circumstance or another. Here is a complete summary of all changes from today's git tree, in order of appearance: 1. There are now separate variables for metadata buffer accounting. 2. Variable sd_log_num_hdrs is no longer needed, since the header accounting is taken care of by the reserve/refund sequence. 3. Fixed a tiny grammatical problem in a comment. 4. Added a new function "calc_reserved" to calculate the reserved log space. This isn't entirely necessary, but it has two benefits: First, it simplifies the gfs2_log_refund function greatly. Second, it allows for easier debugging because I could sprinkle the code with calls to this function to make sure the accounting is proper (by adding asserts and printks) at strategic point of the code. 5. In log_pull_tail there apparently was a kludge to fix up the accounting based on a "pull" parameter. The buffer accounting is now done properly, so the kludge was removed. 6. File sync operations were making a call to gfs2_log_flush that writes another journal header. Since that header was unplanned for (reserved) by the reserve/refund sequence, the free space had to be decremented so that when log_pull_tail gets called, the free space is be adjusted properly. (Did I hear you call that a kludge? well, maybe, but a lot more justifiable than the one I removed). 7. In the gfs2_log_shutdown code, it optionally syncs the log by specifying the PULL parameter to log_write_header. I'm not sure this is necessary anymore. It just seems to me there could be cases where shutdown is called while there are outstanding log buffers. 8. In the (data)buf_lo_before_commit functions, I changed some offset values from being calculated on the fly to being constants. That simplified some code and we might as well let the compiler do the calculation once rather than redoing those cycles at run time. 9. This version has my rewritten databuf_lo_add function. This version is much more like its predecessor, buf_lo_add, which makes it easier to understand. Again, this might not be necessary, but it seems as if this one works as well as the previous one, maybe even better, so I decided to leave it in. 10. In databuf_lo_before_commit, a previous data corruption problem was caused by going off the end of the buffer. The proper solution is to have the proper limit in place, rather than stopping earlier. (Thus my previous attempt to fix it is wrong). If you don't wrap the buffer, you're stopping too early and that causes more log buffer accounting problems. 11. In lops.h there are two new (previously mentioned) constants for figuring out the data offset for the journal buffers. 12. There are also two new functions, buf_limit and databuf_limit to calculate how many entries will fit in the buffer. 13. In function gfs2_meta_wipe, it needs to distinguish between pinned metadata buffers and journaled data buffers for proper journal buffer accounting. It can't use the JDATA gfs2_inode flag because it's sometimes passed the "real" inode and sometimes the "metadata inode" and the inode flags will be random bits in a metadata gfs2_inode. It needs to base its decision on which was passed in. Signed-off-by: Bob Peterson <rpeterso@redhat.com> Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
2007-06-18 19:50:20 +00:00
reserved = calc_reserved(sdp);
gfs2_assert_withdraw(sdp, sdp->sd_log_blks_reserved + tr->tr_reserved >= reserved);
unused = sdp->sd_log_blks_reserved - reserved + tr->tr_reserved;
atomic_add(unused, &sdp->sd_log_blks_free);
trace_gfs2_log_blocks(sdp, unused);
gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <=
[GFS2] assertion failure after writing to journaled file, umount This patch passes all my nasty tests that were causing the code to fail under one circumstance or another. Here is a complete summary of all changes from today's git tree, in order of appearance: 1. There are now separate variables for metadata buffer accounting. 2. Variable sd_log_num_hdrs is no longer needed, since the header accounting is taken care of by the reserve/refund sequence. 3. Fixed a tiny grammatical problem in a comment. 4. Added a new function "calc_reserved" to calculate the reserved log space. This isn't entirely necessary, but it has two benefits: First, it simplifies the gfs2_log_refund function greatly. Second, it allows for easier debugging because I could sprinkle the code with calls to this function to make sure the accounting is proper (by adding asserts and printks) at strategic point of the code. 5. In log_pull_tail there apparently was a kludge to fix up the accounting based on a "pull" parameter. The buffer accounting is now done properly, so the kludge was removed. 6. File sync operations were making a call to gfs2_log_flush that writes another journal header. Since that header was unplanned for (reserved) by the reserve/refund sequence, the free space had to be decremented so that when log_pull_tail gets called, the free space is be adjusted properly. (Did I hear you call that a kludge? well, maybe, but a lot more justifiable than the one I removed). 7. In the gfs2_log_shutdown code, it optionally syncs the log by specifying the PULL parameter to log_write_header. I'm not sure this is necessary anymore. It just seems to me there could be cases where shutdown is called while there are outstanding log buffers. 8. In the (data)buf_lo_before_commit functions, I changed some offset values from being calculated on the fly to being constants. That simplified some code and we might as well let the compiler do the calculation once rather than redoing those cycles at run time. 9. This version has my rewritten databuf_lo_add function. This version is much more like its predecessor, buf_lo_add, which makes it easier to understand. Again, this might not be necessary, but it seems as if this one works as well as the previous one, maybe even better, so I decided to leave it in. 10. In databuf_lo_before_commit, a previous data corruption problem was caused by going off the end of the buffer. The proper solution is to have the proper limit in place, rather than stopping earlier. (Thus my previous attempt to fix it is wrong). If you don't wrap the buffer, you're stopping too early and that causes more log buffer accounting problems. 11. In lops.h there are two new (previously mentioned) constants for figuring out the data offset for the journal buffers. 12. There are also two new functions, buf_limit and databuf_limit to calculate how many entries will fit in the buffer. 13. In function gfs2_meta_wipe, it needs to distinguish between pinned metadata buffers and journaled data buffers for proper journal buffer accounting. It can't use the JDATA gfs2_inode flag because it's sometimes passed the "real" inode and sometimes the "metadata inode" and the inode flags will be random bits in a metadata gfs2_inode. It needs to base its decision on which was passed in. Signed-off-by: Bob Peterson <rpeterso@redhat.com> Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
2007-06-18 19:50:20 +00:00
sdp->sd_jdesc->jd_blocks);
sdp->sd_log_blks_reserved = reserved;
gfs2_log_unlock(sdp);
}
static void buf_lo_incore_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
{
struct list_head *head = &tr->tr_list_buf;
struct gfs2_bufdata *bd;
gfs2_log_lock(sdp);
while (!list_empty(head)) {
bd = list_entry(head->next, struct gfs2_bufdata, bd_list_tr);
list_del_init(&bd->bd_list_tr);
tr->tr_num_buf--;
}
gfs2_log_unlock(sdp);
gfs2_assert_warn(sdp, !tr->tr_num_buf);
}
/**
* gfs2_log_commit - Commit a transaction to the log
* @sdp: the filesystem
* @tr: the transaction
*
* We wake up gfs2_logd if the number of pinned blocks exceed thresh1
* or the total number of used blocks (pinned blocks plus AIL blocks)
* is greater than thresh2.
*
* At mount time thresh1 is 1/3rd of journal size, thresh2 is 2/3rd of
* journal size.
*
* Returns: errno
*/
void gfs2_log_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
{
log_refund(sdp, tr);
buf_lo_incore_commit(sdp, tr);
up_read(&sdp->sd_log_flush_lock);
if (atomic_read(&sdp->sd_log_pinned) > atomic_read(&sdp->sd_log_thresh1) ||
((sdp->sd_jdesc->jd_blocks - atomic_read(&sdp->sd_log_blks_free)) >
atomic_read(&sdp->sd_log_thresh2)))
wake_up(&sdp->sd_logd_waitq);
}
/**
* gfs2_log_shutdown - write a shutdown header into a journal
* @sdp: the filesystem
*
*/
void gfs2_log_shutdown(struct gfs2_sbd *sdp)
{
down_write(&sdp->sd_log_flush_lock);
gfs2_assert_withdraw(sdp, !sdp->sd_log_blks_reserved);
gfs2_assert_withdraw(sdp, !sdp->sd_log_num_buf);
gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
gfs2_assert_withdraw(sdp, !sdp->sd_log_num_rg);
gfs2_assert_withdraw(sdp, !sdp->sd_log_num_databuf);
gfs2_assert_withdraw(sdp, list_empty(&sdp->sd_ail1_list));
sdp->sd_log_flush_head = sdp->sd_log_head;
sdp->sd_log_flush_wrapped = 0;
[GFS2] assertion failure after writing to journaled file, umount This patch passes all my nasty tests that were causing the code to fail under one circumstance or another. Here is a complete summary of all changes from today's git tree, in order of appearance: 1. There are now separate variables for metadata buffer accounting. 2. Variable sd_log_num_hdrs is no longer needed, since the header accounting is taken care of by the reserve/refund sequence. 3. Fixed a tiny grammatical problem in a comment. 4. Added a new function "calc_reserved" to calculate the reserved log space. This isn't entirely necessary, but it has two benefits: First, it simplifies the gfs2_log_refund function greatly. Second, it allows for easier debugging because I could sprinkle the code with calls to this function to make sure the accounting is proper (by adding asserts and printks) at strategic point of the code. 5. In log_pull_tail there apparently was a kludge to fix up the accounting based on a "pull" parameter. The buffer accounting is now done properly, so the kludge was removed. 6. File sync operations were making a call to gfs2_log_flush that writes another journal header. Since that header was unplanned for (reserved) by the reserve/refund sequence, the free space had to be decremented so that when log_pull_tail gets called, the free space is be adjusted properly. (Did I hear you call that a kludge? well, maybe, but a lot more justifiable than the one I removed). 7. In the gfs2_log_shutdown code, it optionally syncs the log by specifying the PULL parameter to log_write_header. I'm not sure this is necessary anymore. It just seems to me there could be cases where shutdown is called while there are outstanding log buffers. 8. In the (data)buf_lo_before_commit functions, I changed some offset values from being calculated on the fly to being constants. That simplified some code and we might as well let the compiler do the calculation once rather than redoing those cycles at run time. 9. This version has my rewritten databuf_lo_add function. This version is much more like its predecessor, buf_lo_add, which makes it easier to understand. Again, this might not be necessary, but it seems as if this one works as well as the previous one, maybe even better, so I decided to leave it in. 10. In databuf_lo_before_commit, a previous data corruption problem was caused by going off the end of the buffer. The proper solution is to have the proper limit in place, rather than stopping earlier. (Thus my previous attempt to fix it is wrong). If you don't wrap the buffer, you're stopping too early and that causes more log buffer accounting problems. 11. In lops.h there are two new (previously mentioned) constants for figuring out the data offset for the journal buffers. 12. There are also two new functions, buf_limit and databuf_limit to calculate how many entries will fit in the buffer. 13. In function gfs2_meta_wipe, it needs to distinguish between pinned metadata buffers and journaled data buffers for proper journal buffer accounting. It can't use the JDATA gfs2_inode flag because it's sometimes passed the "real" inode and sometimes the "metadata inode" and the inode flags will be random bits in a metadata gfs2_inode. It needs to base its decision on which was passed in. Signed-off-by: Bob Peterson <rpeterso@redhat.com> Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
2007-06-18 19:50:20 +00:00
log_write_header(sdp, GFS2_LOG_HEAD_UNMOUNT,
(sdp->sd_log_tail == current_tail(sdp)) ? 0 : PULL);
gfs2_assert_warn(sdp, atomic_read(&sdp->sd_log_blks_free) == sdp->sd_jdesc->jd_blocks);
gfs2_assert_warn(sdp, sdp->sd_log_head == sdp->sd_log_tail);
gfs2_assert_warn(sdp, list_empty(&sdp->sd_ail2_list));
sdp->sd_log_head = sdp->sd_log_flush_head;
sdp->sd_log_tail = sdp->sd_log_head;
up_write(&sdp->sd_log_flush_lock);
}
/**
* gfs2_meta_syncfs - sync all the buffers in a filesystem
* @sdp: the filesystem
*
*/
void gfs2_meta_syncfs(struct gfs2_sbd *sdp)
{
gfs2_log_flush(sdp, NULL);
for (;;) {
gfs2_ail1_start(sdp);
gfs2_ail1_wait(sdp);
if (gfs2_ail1_empty(sdp))
break;
}
gfs2_log_flush(sdp, NULL);
}
static inline int gfs2_jrnl_flush_reqd(struct gfs2_sbd *sdp)
{
return (atomic_read(&sdp->sd_log_pinned) >= atomic_read(&sdp->sd_log_thresh1));
}
static inline int gfs2_ail_flush_reqd(struct gfs2_sbd *sdp)
{
unsigned int used_blocks = sdp->sd_jdesc->jd_blocks - atomic_read(&sdp->sd_log_blks_free);
return used_blocks >= atomic_read(&sdp->sd_log_thresh2);
}
/**
* gfs2_logd - Update log tail as Active Items get flushed to in-place blocks
* @sdp: Pointer to GFS2 superblock
*
* Also, periodically check to make sure that we're using the most recent
* journal index.
*/
int gfs2_logd(void *data)
{
struct gfs2_sbd *sdp = data;
unsigned long t = 1;
DEFINE_WAIT(wait);
unsigned preflush;
while (!kthread_should_stop()) {
preflush = atomic_read(&sdp->sd_log_pinned);
if (gfs2_jrnl_flush_reqd(sdp) || t == 0) {
gfs2_ail1_empty(sdp);
gfs2_log_flush(sdp, NULL);
}
if (gfs2_ail_flush_reqd(sdp)) {
gfs2_ail1_start(sdp);
gfs2_ail1_wait(sdp);
gfs2_ail1_empty(sdp);
gfs2_log_flush(sdp, NULL);
}
if (!gfs2_ail_flush_reqd(sdp))
wake_up(&sdp->sd_log_waitq);
t = gfs2_tune_get(sdp, gt_logd_secs) * HZ;
try_to_freeze();
do {
prepare_to_wait(&sdp->sd_logd_waitq, &wait,
TASK_INTERRUPTIBLE);
if (!gfs2_ail_flush_reqd(sdp) &&
!gfs2_jrnl_flush_reqd(sdp) &&
!kthread_should_stop())
t = schedule_timeout(t);
} while(t && !gfs2_ail_flush_reqd(sdp) &&
!gfs2_jrnl_flush_reqd(sdp) &&
!kthread_should_stop());
finish_wait(&sdp->sd_logd_waitq, &wait);
}
return 0;
}