The softdep_setup_freeblocks() adds worklist items before

deallocate_dependencies() is done. This opens a race between softdep
thread and the thread that does the truncation:
  A write of the indirect block causes the freeblks to become
  ALLCOMPLETE while softdep_setup_freeblocks() dropped softdep lock. And
  then, softdep_disk_write_complete() would reassign the workitem to the
  mount point worklist, causing premature processing of the workitem, or
  journal write exhaust the fb_jfreeblkhd and handle_written_jfreeblk does
  the same reassign.
indir_trunc() then would find the indirect block that is locked (with lock
owned by kernel) but without any dependencies, causing it to hang in
getblk() waiting for buffer lock.

Do not mark freeblks as DEPCOMPLETE until deallocate_dependencies()
finished.

Analyzed, suggested and reviewed by:	jeff
Tested by:	pho
This commit is contained in:
Konstantin Belousov 2010-11-11 11:54:01 +00:00
parent e2f6069c09
commit be913821af

View File

@ -5270,7 +5270,7 @@ softdep_setup_freeblocks(ip, length, flags)
if (delay)
WORKLIST_INSERT(&bp->b_dep, &freeblks->fb_list);
else if (needj)
freeblks->fb_state |= DEPCOMPLETE | COMPLETE;
freeblks->fb_state |= COMPLETE;
/*
* Because the file length has been truncated to zero, any
* pending block allocation dependency structures associated
@ -5332,8 +5332,9 @@ softdep_setup_freeblocks(ip, length, flags)
if (inodedep_lookup(mp, ip->i_number, 0, &inodedep) != 0)
(void) free_inodedep(inodedep);
if (delay) {
if (delay || needj)
freeblks->fb_state |= DEPCOMPLETE;
if (delay) {
/*
* If the inode with zeroed block pointers is now on disk
* we can start freeing blocks. Add freeblks to the worklist
@ -5344,6 +5345,8 @@ softdep_setup_freeblocks(ip, length, flags)
if ((freeblks->fb_state & ALLCOMPLETE) == ALLCOMPLETE)
add_to_worklist(&freeblks->fb_list, 1);
}
if (needj && LIST_EMPTY(&freeblks->fb_jfreeblkhd))
needj = 0;
FREE_LOCK(&lk);
/*