> On Fri, Nov 04, 2016 at 04:44:10PM +0100, J. Hannken-Illjes wrote: >> - This change results in "panic: ffs_blkfree_common: freeing free block" >> if I put a file system under stress (*1). >> >> - I suppose not zeroing the blocks to be freed before freeing them >> makes the life of fsck harder. >> >> - Running "brelse(bp, BC_INVAL)" doesn't look OK.
The brelse(bp, BC_INVAL) call was there before as well, but the condition changed and is wrong. I can repeat the problem with your script and the packaged fsx (thanks Thomas). Whipped up a patch to what looked wrong there, and it no longer panics for me. Patch is attached. I'll test further and commit it tomorrow. Jaromir
Index: ffs_inode.c =================================================================== RCS file: /cvsroot/src/sys/ufs/ffs/ffs_inode.c,v retrieving revision 1.118 diff -u -p -r1.118 ffs_inode.c --- ffs_inode.c 28 Oct 2016 20:38:12 -0000 1.118 +++ ffs_inode.c 6 Nov 2016 23:09:15 -0000 @@ -675,18 +675,18 @@ ffs_indirtrunc(struct inode *ip, daddr_t * Recursively free blocks on the now last partial indirect block. */ if (level > SINGLE && lastbn >= 0) { - nb = RBAP(ip, last); + last = lastbn % factor; + nb = RBAP(ip, i); if (nb != 0) { error = ffs_indirtrunc(ip, nlbn, FFS_FSBTODB(fs, nb), - lastbn % factor, level - 1, - countp); + last, level - 1, countp); if (error) goto out; } } out: - if (RBAP(ip, 0) == 0) { + if (lastbn < 0 && error == 0) { /* all freed, release without writing back */ brelse(bp, BC_INVAL); } else {