* Re: Linux 2.4.11-pre4, extremely long umount times
2001-10-07 3:25 ` Andrew Morton
@ 2001-10-07 4:02 ` Bob McElrath
0 siblings, 0 replies; 3+ messages in thread
From: Bob McElrath @ 2001-10-07 4:02 UTC (permalink / raw)
To: Andrew Morton; +Cc: Kernel Mailing List
[-- Attachment #1.1: Type: text/plain, Size: 1384 bytes --]
Andrew Morton [akpm@zip.com.au] wrote:
> Bob McElrath wrote:
> >
> > I'm running 2.4.11-pre4 with the ext3 patch and Andrew Morton's low-latency
> > patch on an alpha LX164.
> >
> > umount times are extremely long (> 30 minutes) for both ext2 and ext3
> > filesystems, though they eventually succeed.
> >
> > Is this a known problem?
> >
>
> Nope. It's possible to get swapoff durations of many minutes,
> but I don't think similar problems with unmount have been reported.
> Is there any disk activity? ps and top output? Any theories?
umount is using 100% of the cpu. I've been through four or five mount/umounts
and all of them hung like this.
> BTW: I'm faintly surprised to hear that ext3 actually works in
> 2.4.11-pre4. Quite a lot of things with which ext3 has an intimate
> relationship were changed....
I had to do some non-trival patch merging to fs/buffer.c and fs/super.c. It's
possible I did something wrong. Attached are the diffs between 2.4.11-pre4 for
those two files.
There's some low-latency business in buffer.c too. This may be a case of
overpatchitis.
Aside from the umount though, all the filesystems (ext2, ext3, and reiserfs)
seem to work fine. Machine has been up for about a day now.
Cheers,
-- Bob
Bob McElrath (rsmcelrath@students.wisc.edu)
Univ. of Wisconsin at Madison, Department of Physics
[-- Attachment #1.2: buffer.c.diff --]
[-- Type: text/plain, Size: 13681 bytes --]
*** linux-2.4.11-pre4/fs/buffer.c Sat Oct 6 01:24:53 2001
--- linux-2.4.11-pre4-ext3/fs/buffer.c Sat Oct 6 02:02:58 2001
***************
*** 45,50 ****
--- 45,52 ----
#include <linux/quotaops.h>
#include <linux/iobuf.h>
#include <linux/highmem.h>
+ #include <linux/jbd.h>
+ #include <linux/module.h>
#include <linux/completion.h>
#include <asm/uaccess.h>
***************
*** 604,611 ****
information that was supposed to be just stored on the physical layer
by the user.
! Thus invalidate_buffers in general usage is not allwowed to trash dirty
! buffers. For example ioctl(FLSBLKBUF) expects dirty data to be preserved.
NOTE: In the case where the user removed a removable-media-disk even if
there's still dirty data not synced on disk (due a bug in the device driver
--- 606,617 ----
information that was supposed to be just stored on the physical layer
by the user.
! Thus invalidate_buffers in general usage is not allwowed to trash
! dirty buffers. For example ioctl(FLSBLKBUF) expects dirty data to
! be preserved. These buffers are simply skipped.
!
! We also skip buffers which are still in use. For example this can
! happen if a userspace program is reading the block device.
NOTE: In the case where the user removed a removable-media-disk even if
there's still dirty data not synced on disk (due a bug in the device driver
***************
*** 649,654 ****
--- 655,670 ----
/* Not hashed? */
if (!bh->b_pprev)
continue;
+
+ if (conditional_schedule_needed()) {
+ atomic_inc(&bh->b_count);
+ spin_unlock(&lru_list_lock);
+ unconditional_schedule();
+ spin_lock(&lru_list_lock);
+ atomic_dec(&bh->b_count);
+ slept = 1;
+ }
+
if (buffer_locked(bh)) {
get_bh(bh);
spin_unlock(&lru_list_lock);
***************
*** 708,713 ****
--- 724,730 ----
bh->b_list = BUF_CLEAN;
bh->b_end_io = handler;
bh->b_private = private;
+ buffer_trace_init(&bh->b_history);
}
static void end_buffer_io_async(struct buffer_head * bh, int uptodate)
***************
*** 717,722 ****
--- 734,740 ----
struct buffer_head *tmp;
struct page *page;
+ BUFFER_TRACE(bh, "enter");
mark_buffer_uptodate(bh, uptodate);
/* This is a temporary buffer used for page I/O. */
***************
*** 802,807 ****
--- 820,826 ----
struct buffer_head *bh;
struct inode tmp;
int err = 0, err2;
+ DEFINE_RESCHED_COUNT;
INIT_LIST_HEAD(&tmp.i_dirty_buffers);
***************
*** 823,830 ****
--- 842,859 ----
spin_lock(&lru_list_lock);
}
}
+ if (TEST_RESCHED_COUNT(32)) {
+ RESET_RESCHED_COUNT();
+ if (conditional_schedule_needed()) {
+ spin_unlock(&lru_list_lock);
+ unconditional_schedule(); /* Syncing many dirty buffers */
+ spin_lock(&lru_list_lock);
+ }
+ }
}
+ RESET_RESCHED_COUNT();
+
while (!list_empty(&tmp.i_dirty_buffers)) {
bh = BH_ENTRY(tmp.i_dirty_buffers.prev);
remove_inode_queue(bh);
***************
*** 852,857 ****
--- 881,887 ----
struct inode tmp;
int err = 0, err2;
+ DEFINE_RESCHED_COUNT;
INIT_LIST_HEAD(&tmp.i_dirty_data_buffers);
spin_lock(&lru_list_lock);
***************
*** 883,888 ****
--- 913,922 ----
if (!buffer_uptodate(bh))
err = -EIO;
brelse(bh);
+ if (TEST_RESCHED_COUNT(32)) {
+ RESET_RESCHED_COUNT();
+ conditional_schedule();
+ }
spin_lock(&lru_list_lock);
}
***************
*** 911,924 ****
struct buffer_head *bh;
struct list_head *list;
int err = 0;
!
! spin_lock(&lru_list_lock);
repeat:
for (list = inode->i_dirty_buffers.prev;
bh = BH_ENTRY(list), list != &inode->i_dirty_buffers;
list = bh->b_inode_buffers.prev) {
if (buffer_locked(bh)) {
get_bh(bh);
spin_unlock(&lru_list_lock);
--- 945,967 ----
struct buffer_head *bh;
struct list_head *list;
int err = 0;
! DEFINE_RESCHED_COUNT;
repeat:
+ conditional_schedule();
+ spin_lock(&lru_list_lock);
for (list = inode->i_dirty_buffers.prev;
bh = BH_ENTRY(list), list != &inode->i_dirty_buffers;
list = bh->b_inode_buffers.prev) {
+ if (TEST_RESCHED_COUNT(32)) {
+ RESET_RESCHED_COUNT();
+ if (conditional_schedule_needed()) {
+ spin_unlock(&lru_list_lock);
+ goto repeat;
+ }
+ }
+
if (buffer_locked(bh)) {
get_bh(bh);
spin_unlock(&lru_list_lock);
***************
*** 955,961 ****
if (!buffer_uptodate(bh))
err = -EIO;
brelse(bh);
- spin_lock(&lru_list_lock);
goto repeat;
}
}
--- 998,1003 ----
***************
*** 1083,1088 ****
--- 1125,1136 ----
}
}
+ void set_buffer_flushtime(struct buffer_head *bh)
+ {
+ bh->b_flushtime = jiffies + bdf_prm.b_un.age_buffer;
+ }
+ EXPORT_SYMBOL(set_buffer_flushtime);
+
/*
* A buffer may need to be moved from one buffer list to another
* (e.g. in case it is not shared any more). Handle this.
***************
*** 1090,1095 ****
--- 1138,1146 ----
static void __refile_buffer(struct buffer_head *bh)
{
int dispose = BUF_CLEAN;
+
+ BUFFER_TRACE(bh, "enter");
+
if (buffer_locked(bh))
dispose = BUF_LOCKED;
if (buffer_dirty(bh))
***************
*** 1101,1106 ****
--- 1152,1158 ----
remove_inode_queue(bh);
__insert_into_lru_list(bh, dispose);
}
+ BUFFER_TRACE(bh, "exit");
}
void refile_buffer(struct buffer_head *bh)
***************
*** 1115,1120 ****
--- 1167,1173 ----
*/
void __brelse(struct buffer_head * buf)
{
+ BUFFER_TRACE(buf, "entry");
if (atomic_read(&buf->b_count)) {
put_bh(buf);
return;
***************
*** 1159,1168 ****
/*
* Note: the caller should wake up the buffer_wait list if needed.
*/
! static __inline__ void __put_unused_buffer_head(struct buffer_head * bh)
{
if (bh->b_inode)
BUG();
if (nr_unused_buffer_heads >= MAX_UNUSED_BUFFERS) {
kmem_cache_free(bh_cachep, bh);
} else {
--- 1212,1233 ----
/*
* Note: the caller should wake up the buffer_wait list if needed.
*/
! static void __put_unused_buffer_head(struct buffer_head * bh)
{
if (bh->b_inode)
BUG();
+
+ J_ASSERT_BH(bh, bh->b_prev_free == 0);
+ #if defined(CONFIG_JBD) || defined(CONFIG_JBD_MODULE)
+ if (buffer_jbd(bh)) {
+ J_ASSERT_BH(bh, bh2jh(bh)->b_transaction == 0);
+ J_ASSERT_BH(bh, bh2jh(bh)->b_next_transaction == 0);
+ J_ASSERT_BH(bh, bh2jh(bh)->b_frozen_data == 0);
+ J_ASSERT_BH(bh, bh2jh(bh)->b_committed_data == 0);
+ }
+ #endif
+ buffer_trace_init(&bh->b_history);
+
if (nr_unused_buffer_heads >= MAX_UNUSED_BUFFERS) {
kmem_cache_free(bh_cachep, bh);
} else {
***************
*** 1176,1187 ****
}
}
/*
* Reserve NR_RESERVED buffer heads for async IO requests to avoid
* no-buffer-head deadlock. Return NULL on failure; waiting for
* buffer heads is now handled in create_buffers().
*/
! static struct buffer_head * get_unused_buffer_head(int async)
{
struct buffer_head * bh;
--- 1241,1260 ----
}
}
+ void put_unused_buffer_head(struct buffer_head *bh)
+ {
+ spin_lock(&unused_list_lock);
+ __put_unused_buffer_head(bh);
+ spin_unlock(&unused_list_lock);
+ }
+ EXPORT_SYMBOL(put_unused_buffer_head);
+
/*
* Reserve NR_RESERVED buffer heads for async IO requests to avoid
* no-buffer-head deadlock. Return NULL on failure; waiting for
* buffer heads is now handled in create_buffers().
*/
! struct buffer_head * get_unused_buffer_head(int async)
{
struct buffer_head * bh;
***************
*** 1202,1207 ****
--- 1275,1281 ----
if((bh = kmem_cache_alloc(bh_cachep, SLAB_NOFS)) != NULL) {
bh->b_blocknr = -1;
bh->b_this_page = NULL;
+ buffer_trace_init(&bh->b_history);
return bh;
}
***************
*** 1215,1220 ****
--- 1289,1295 ----
unused_list = bh->b_next_free;
nr_unused_buffer_heads--;
spin_unlock(&unused_list_lock);
+ buffer_trace_init(&bh->b_history);
return bh;
}
spin_unlock(&unused_list_lock);
***************
*** 1222,1227 ****
--- 1297,1303 ----
return NULL;
}
+ EXPORT_SYMBOL(get_unused_buffer_head);
void set_bh_page (struct buffer_head *bh, struct page *page, unsigned long offset)
{
***************
*** 1236,1241 ****
--- 1312,1318 ----
else
bh->b_data = page_address(page) + offset;
}
+ EXPORT_SYMBOL(set_bh_page);
/*
* Create the appropriate buffers when given a page for data area and
***************
*** 1319,1324 ****
--- 1396,1402 ----
static void discard_buffer(struct buffer_head * bh)
{
if (buffer_mapped(bh)) {
+ BUFFER_TRACE(bh, "entry");
mark_buffer_clean(bh);
lock_buffer(bh);
clear_bit(BH_Uptodate, &bh->b_state);
***************
*** 1329,1334 ****
--- 1407,1437 ----
}
}
+ /**
+ * try_to_release_page - release old fs-specific metadata on a page
+ *
+ */
+
+ int try_to_release_page(struct page * page, int gfp_mask)
+ {
+ if (!PageLocked(page))
+ BUG();
+
+ if (!page->mapping)
+ goto try_to_free;
+ if (!page->mapping->a_ops->releasepage)
+ goto try_to_free;
+ if (page->mapping->a_ops->releasepage(page, gfp_mask))
+ goto try_to_free;
+ /*
+ * We couldn't release buffer metadata; don't even bother trying
+ * to release buffers.
+ */
+ return 0;
+ try_to_free:
+ return try_to_free_buffers(page, gfp_mask);
+ }
+
/*
* We don't have to release all buffers here, but
* we have to be sure that no dirty buffer is left
***************
*** 1400,1405 ****
--- 1503,1509 ----
page->buffers = head;
page_cache_get(page);
}
+ EXPORT_SYMBOL(create_empty_buffers);
/*
* We are taking a block for data and we don't want any output from any
***************
*** 1418,1424 ****
--- 1522,1531 ----
struct buffer_head *old_bh;
old_bh = get_hash_table(bh->b_dev, bh->b_blocknr, bh->b_size);
+ J_ASSERT_BH(bh, old_bh != bh);
if (old_bh) {
+ BUFFER_TRACE(old_bh, "old_bh - entry");
+ J_ASSERT_BH(old_bh, !buffer_jlist_eq(old_bh, BJ_Metadata));
mark_buffer_clean(old_bh);
wait_on_buffer(old_bh);
clear_bit(BH_Req, &old_bh->b_state);
***************
*** 1443,1450 ****
*/
/*
! * block_write_full_page() is SMP-safe - currently it's still
! * being called with the kernel lock held, but the code is ready.
*/
static int __block_write_full_page(struct inode *inode, struct page *page, get_block_t *get_block)
{
--- 1550,1556 ----
*/
/*
! * block_write_full_page() is SMP threaded - the kernel lock is not held.
*/
static int __block_write_full_page(struct inode *inode, struct page *page, get_block_t *get_block)
{
***************
*** 1478,1492 ****
err = get_block(inode, block, bh, 1);
if (err)
goto out;
! if (buffer_new(bh))
unmap_underlying_metadata(bh);
}
bh = bh->b_this_page;
block++;
} while (bh != head);
/* Stage 2: lock the buffers, mark them clean */
do {
lock_buffer(bh);
set_buffer_async_io(bh);
set_bit(BH_Uptodate, &bh->b_state);
--- 1584,1601 ----
err = get_block(inode, block, bh, 1);
if (err)
goto out;
! if (buffer_new(bh)) {
! BUFFER_TRACE(bh, "new: call unmap_underlying_metadata");
unmap_underlying_metadata(bh);
}
+ }
bh = bh->b_this_page;
block++;
} while (bh != head);
/* Stage 2: lock the buffers, mark them clean */
do {
+ BUFFER_TRACE(bh, "lock it");
lock_buffer(bh);
set_buffer_async_io(bh);
set_bit(BH_Uptodate, &bh->b_state);
***************
*** 1543,1550 ****
--- 1652,1661 ----
if (err)
goto out;
if (buffer_new(bh)) {
+ BUFFER_TRACE(bh, "new: call unmap_underlying_metadata");
unmap_underlying_metadata(bh);
if (Page_Uptodate(page)) {
+ BUFFER_TRACE(bh, "setting uptodate");
set_bit(BH_Uptodate, &bh->b_state);
continue;
}
***************
*** 1558,1568 ****
--- 1669,1681 ----
}
}
if (Page_Uptodate(page)) {
+ BUFFER_TRACE(bh, "setting uptodate");
set_bit(BH_Uptodate, &bh->b_state);
continue;
}
if (!buffer_uptodate(bh) &&
(block_start < from || block_end > to)) {
+ BUFFER_TRACE(bh, "reading");
ll_rw_block(READ, 1, &bh);
*wait_bh++=bh;
}
***************
*** 1601,1606 ****
--- 1714,1720 ----
} else {
set_bit(BH_Uptodate, &bh->b_state);
if (!atomic_set_buffer_dirty(bh)) {
+ BUFFER_TRACE(bh, "mark dirty");
__mark_dirty(bh);
buffer_insert_inode_data_queue(bh, inode);
need_balance_dirty = 1;
***************
*** 1884,1889 ****
--- 1998,2004 ----
flush_dcache_page(page);
kunmap(page);
+ BUFFER_TRACE(bh, "zeroed end of block");
__mark_buffer_dirty(bh);
err = 0;
***************
*** 2430,2435 ****
--- 2545,2552 ----
wakeup_bdflush();
return 0;
}
+ EXPORT_SYMBOL(try_to_free_buffers);
+ EXPORT_SYMBOL(buffermem_pages);
/* ================== Debugging =================== */
***************
*** 2542,2548 ****
*/
DECLARE_WAIT_QUEUE_HEAD(bdflush_wait);
-
void wakeup_bdflush(void)
{
wake_up_interruptible(&bdflush_wait);
--- 2659,2664 ----
[-- Attachment #1.3: super.c.diff --]
[-- Type: text/plain, Size: 1126 bytes --]
*** linux-2.4.11-pre4/fs/super.c Sat Oct 6 01:24:54 2001
--- linux-2.4.11-pre4-ext3/fs/super.c Sat Oct 6 01:27:12 2001
***************
*** 883,888 ****
--- 883,899 ----
return do_kern_mount((char *)type->name, 0, (char *)type->name, NULL);
}
+ static char *root_mount_data;
+ static int __init root_data_setup(char *line)
+ {
+ static char buffer[128];
+
+ strcpy(buffer, line);
+ root_mount_data = buffer;
+ return 1;
+ }
+ __setup("rootflags=", root_data_setup);
+
void __init mount_root(void)
{
struct nameidata root_nd;
***************
*** 1017,1023 ****
if (!try_inc_mod_count(fs_type->owner))
continue;
read_unlock(&file_systems_lock);
! sb = read_super(ROOT_DEV,bdev,fs_type,root_mountflags,NULL,1);
if (sb)
goto mount_it;
read_lock(&file_systems_lock);
--- 1028,1035 ----
if (!try_inc_mod_count(fs_type->owner))
continue;
read_unlock(&file_systems_lock);
! sb = read_super(ROOT_DEV,bdev,fs_type,root_mountflags,
! root_mount_data,1);
if (sb)
goto mount_it;
read_lock(&file_systems_lock);
[-- Attachment #2: Type: application/pgp-signature, Size: 240 bytes --]
^ permalink raw reply [flat|nested] 3+ messages in thread