[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: shutdown umount hangs
Utz Lehmann wrote:
>
> xfs_unmountfs_writesb pre strat flags 0xa08017e pb_iodonesema cnt(0,0)sl
> xfs_unmountfs_writesb prot strat flags 0xa08017e pb_iodonesema cnt(0,0)sl
yup the flags are wrong... ok I have another patch for you to try.
I think this will fix the problem.
I haven't tested it yet but you can give it a try if you want.
>
>
> (hangs)
>
> utz
--
Russell Cattelan
--
Digital Elves inc. -- Currently on loan to SGI
Linux XFS core developer.
===========================================================================
linux/fs/pagebuf/page_buf.c
===========================================================================
--- /usr/tmp/TmpDir.17129-0/linux/fs/pagebuf/page_buf.c_1.72 Fri Apr 6 11:51:48 2001
+++ linux/fs/pagebuf/page_buf.c Fri Apr 6 11:25:53 2001
@@ -2136,6 +2136,7 @@
struct list_head *head, *curr;
unsigned long save;
pagebuf_marker_t *pb_marker_ptr;
+ int lr=1;
pb_marker_ptr = kmalloc(sizeof(pagebuf_marker_t), GFP_KERNEL);
@@ -2152,73 +2153,58 @@
while (curr != head) {
pb = list_entry(curr, page_buf_t, pb_list);
- /*
- * Skip other markers.
- */
- if (pb->pb_flags == 0 ) {
- curr = curr->next;
- continue;
- }
PB_TRACE(pb, PB_TRACE_REC(walkq2), pagebuf_ispin(pb));
- if (pb->pb_target == target) {
- if (pb->pb_flags & PBF_DELWRI) {
+
+ if ((pb->pb_flags == 0) || (pb->pb_target != target) ||
+ !(pb->pb_flags & PBF_DELWRI)) {
+ curr = curr->next;
+ continue;
+ }
- if ((flags & PBDF_PINCOUNT) && pagebuf_ispin(pb)){
- /* just want a count of the number of pinned buffer */
- (*pincount)++;
- curr = curr->next;
- continue;
- }
-
- pb->pb_flags &= ~PBF_DELWRI;
- pb->pb_flags |= PBF_WRITE;
- if (flags & PBDF_WAIT)
- pb->pb_flags &= ~PBF_ASYNC;
-
- /* insert a place holder */
- list_add(&pb_marker_ptr->pb_list, curr);
-
- spin_unlock_irqrestore(
- &pb_daemon->pb_delwrite_lock,
- save);
-
- if (pb->pb_flags & _PBF_LOCKABLE){
- if (flags & PBDF_PINCOUNT){
- /* if we are doing this pin count thing and we can't get the lock just skip it for now */
- }if (!pagebuf_cond_lock(pb)){
- __pagebuf_iorequest(pb);
- } else {
- (*pincount)++;
- /* make it async again... don't want to wait on in below */
- pb->pb_flags |= PBF_ASYNC;
- }
- }else {
- pagebuf_lock(pb);
- __pagebuf_iorequest(pb);
- }
-
- spin_lock_irqsave(
- &pb_daemon->pb_delwrite_lock,
- save);
- /*
- * ok got the lock back; pick up the place
- * holder and continue on
- */
- curr = pb_marker_ptr->pb_list.next;
- list_del(&pb_marker_ptr->pb_list);
- } else {
- /* not doing anything with current...
- * move on to the next one */
- curr = curr->next;
+ if (!pagebuf_ispin(pb)){
+ /* insert a place holder */
+ list_add(&pb_marker_ptr->pb_list, curr);
+
+ spin_unlock_irqrestore(&pb_daemon->pb_delwrite_lock, save);
+
+ /* meta data pagebufs better be lockable pagebuf's */
+ assert(pb->pb_flags & _PBF_LOCKABLE);
+
+ if ((flags & (PBDF_WAIT | PBDF_PINCOUNT)) == PBDF_WAIT)
+ lr = pagebuf_lock(pb);
+ else if (flags & PBDF_PINCOUNT)
+ lr = pagebuf_cond_lock(pb);
+
+ if (lr == 0){ /* got the lock */
+ /* check the pin count one more time just make sure */
+ if (pagebuf_ispin(pb)){
+ /* opps somebody bumped the pin count since the last time we checked it */
+ pagebuf_unlock(pb);
+ (*pincount)++;
+ goto out;
}
- } else {
- /* not doing anything with current...
- * move on to the next one */
- curr = curr->next;
+ pb->pb_flags &= ~PBF_DELWRI;
+ pb->pb_flags |= PBF_WRITE;
+ if (flags & PBDF_WAIT)
+ pb->pb_flags &= ~PBF_ASYNC;
+ __pagebuf_iorequest(pb);
+ } else /* didn't get the lock just bump pin count return val*/
+ (*pincount)++;
+ out:
+ spin_lock_irqsave(&pb_daemon->pb_delwrite_lock, save);
+ /*
+ * ok got the list lock back; pick up the place
+ * holder and continue on
+ */
+ curr = pb_marker_ptr->pb_list.next;
+ list_del(&pb_marker_ptr->pb_list);
+ } else { /* still pinned */
+ (*pincount)++;
+ curr = curr->next;
}
- }
+ } /* while */
spin_unlock_irqrestore(&pb_daemon->pb_delwrite_lock, save);
@@ -2238,63 +2224,45 @@
* order to synchronize with it.
*/
-
/* Now do that again, just waiting for the lock */
spin_lock_irqsave(&pb_daemon->pb_delwrite_lock, flags);
head = curr = &pb_daemon->pb_delwrite_l;
curr = curr->next;
-
while (curr != head) {
pb = list_entry(curr, page_buf_t, pb_list);
/*
* Skip other markers.
- */
- if (pb->pb_flags == 0 ) {
- curr = curr->next;
- continue;
- }
-
- /*
* Skip pagebuf's that may have been added
* after the previous loop initiated I/O.
*/
- if (pb->pb_flags & PBF_DELWRI) {
+ if ((pb->pb_flags == 0) ||(pb->pb_flags & PBF_DELWRI)
+ || !(pb->pb_flags & PBF_ASYNC)
+ || !(pb->pb_target == target)) {
curr = curr->next;
continue;
}
PB_TRACE(pb, PB_TRACE_REC(walkq3), pagebuf_ispin(pb));
- if (pb->pb_target == target) {
- int sync = (pb->pb_flags & PBF_ASYNC) == 0;
- list_add(&pb_marker_ptr->pb_list, curr);
-
- spin_unlock_irqrestore(
- &pb_daemon->pb_delwrite_lock,
- flags);
-
- if (sync) {
- pagebuf_iowait(pb);
- pb->pb_flags |= PBF_ASYNC;
- pagebuf_delwri_dequeue(pb);
- if (!pb->pb_relse)
- pagebuf_unlock(pb);
- pagebuf_rele(pb);
- }
+ list_add(&pb_marker_ptr->pb_list, curr);
+
+ spin_unlock_irqrestore(&pb_daemon->pb_delwrite_lock, flags);
+
+ pagebuf_iowait(pb);
+ pb->pb_flags |= PBF_ASYNC;
+ pagebuf_delwri_dequeue(pb);
+ if (!pb->pb_relse)
+ pagebuf_unlock(pb);
+ pagebuf_rele(pb);
+
+ spin_lock_irqsave( &pb_daemon->pb_delwrite_lock, flags);
- spin_lock_irqsave(
- &pb_daemon->pb_delwrite_lock,
- flags);
-
- curr = pb_marker_ptr->pb_list.next;
- list_del(&pb_marker_ptr->pb_list);
- } else {
- curr = curr->next;
- }
+ curr = pb_marker_ptr->pb_list.next;
+ list_del(&pb_marker_ptr->pb_list);
}
spin_unlock_irqrestore(&pb_daemon->pb_delwrite_lock, flags);
===========================================================================
linux/fs/xfs/xfs_buf.h
===========================================================================
--- /usr/tmp/TmpDir.17129-0/linux/fs/xfs/xfs_buf.h_1.70 Fri Apr 6 11:51:48 2001
+++ linux/fs/xfs/xfs_buf.h Fri Apr 6 11:15:28 2001
@@ -317,7 +317,7 @@
#define xfs_iowait(pb) \
pagebuf_iowait(pb)
-#define xfs_binval(buftarg) /* NOT used with pagebufs... do nothing */
+#define xfs_binval(buftarg) XFS_bflush(buftarg)
extern void XFS_bflush(buftarg_t);
===========================================================================
linux/fs/xfs/xfs_mount.c
===========================================================================
--- /usr/tmp/TmpDir.17129-0/linux/fs/xfs/xfs_mount.c_1.249 Fri Apr 6 11:51:48 2001
+++ linux/fs/xfs/xfs_mount.c Fri Apr 6 11:20:20 2001
@@ -29,7 +29,7 @@
*
* http://oss.sgi.com/projects/GenInfo/SGIGPLNoticeExplan/
*/
-
+#define _PAGE_BUF_INTERNAL_
#include <xfs.h>
@@ -1075,12 +1075,13 @@
xfs_buf_t *sbp;
xfs_sb_t *sb;
int error = 0;
-
+ page_buf_private_t *pb;
/*
* skip superblock write if fs is read-only, or
* if we are doing a forced umount.
*/
sbp = xfs_getsb(mp, 0);
+ pb = (page_buf_private_t *)sbp;
if (!(XFS_MTOVFS(mp)->vfs_flag & VFS_RDONLY ||
XFS_FORCED_SHUTDOWN(mp))) {
/*
@@ -1099,8 +1100,19 @@
XFS_BUF_UNREAD(sbp);
XFS_BUF_WRITE(sbp);
xfs_bwait_unpin(sbp);
+ /* the next two lines added for linux port
+ * make sure the superblock buffer won't simply
+ * get requeued. If delwri or async is set iowait will never see
+ * the iodone up(pb_iodonesema)
+ */
+ XFS_BUF_UNDELAYWRITE(sbp);
+ XFS_BUF_UNASYNC(sbp);
ASSERT(XFS_BUF_TARGET(sbp) == mp->m_dev);
+ printk ("xfs_unmountfs_writesb pre strat flags 0x%x pb_iodonesema cnt(%d,%d)sl\n",
+ sbp->pb_flags,pb->pb_iodonesema.count.counter, pb->pb_iodonesema.sleepers);
xfsbdstrat(mp, sbp);
+ printk ("xfs_unmountfs_writesb prot strat flags 0x%x pb_iodonesema cnt(%d,%d)sl\n",
+ sbp->pb_flags,pb->pb_iodonesema.count.counter, pb->pb_iodonesema.sleepers);
/* Nevermind errors we might get here. */
error = xfs_iowait(sbp);
if (error && mp->m_mk_sharedro)