On Sun, Dec 25, 2022 at 08:46:00PM +0800, Zorro Lang wrote: > > + # Reading the full file via mmap should fail. > > bash -c "trap '' SIGBUS; $XFS_IO_PROG -r $fsv_file \ > > -c 'mmap -r 0 $page_aligned_eof' \ > > - -c 'mread 0 $file_len'" |& filter_sigbus > > + -c 'mread 0 $file_len'" >/dev/null 2>$tmp.out > > + if ! grep -q 'Bus error' $tmp.out; then > > + echo "Didn't see SIGBUS when reading file via mmap" > > Hmm... will sigbus error really be output to stderr? From a testing, the > generic/574 fails as: > > # ./check -s simpledev generic/574 > SECTION -- simpledev > FSTYP -- ext4 > PLATFORM -- Linux/x86_64 xx-xxxxxx-xxx 6.1.0-rc3 #5 SMP PREEMPT_DYNAMIC Tue Nov 1 01:08:52 CST 2022 > MKFS_OPTIONS -- -F /dev/sdb > MOUNT_OPTIONS -- -o acl,user_xattr -o context=system_u:object_r:root_t:s0 /dev/sdb /mnt/scratch > > generic/574 - output mismatch (see /root/git/xfstests/results//simpledev/generic/574.out.bad) > --- tests/generic/574.out 2022-12-25 20:02:41.609104749 +0800 > +++ /root/git/xfstests/results//simpledev/generic/574.out.bad 2022-12-25 20:21:57.430719504 +0800 > @@ -1,6 +1,32 @@ > QA output created by 574 > > # Testing block_size=FSV_BLOCK_SIZE > +/root/git/xfstests/tests/generic/574: line 69: 1949533 Bus error (core dumped) bash -c "trap '' SIGBUS; $XFS_IO_PROG -r $fsv_file -c 'mmap -r 0 $page_aligned_eof' -c 'mread 0 $file_len'" > /dev/null 2> $tmp.out > +Didn't see SIGBUS when reading file via mmap This test passes for me both before and after this patch series. Both before and after, the way this is supposed to work is that in: bash -c "trap '' SIGBUS; command_that_exits_with_sigbus" ... bash should print "Bus error" to stderr due to 'command_that_exits_with_sigbus'. That "Bus error" is then redirected. Before it was redirected to a pipeline; after it is redirected to a file. I think what's happening is that the version of bash your system has is not forking before exec'ing 'command_that_exits_with_sigbus'. As a result, "Bus error" is printed by the *parent* bash process instead, skipping any redirection in the shell script. Apparently skipping fork is a real thing in bash, and different versions of bash have had subtly different conditions for enabling it. So this seems plausible. Adding an extra command after 'command_that_exits_with_sigbus' should fix this: bash -c "trap '' SIGBUS; command_that_exits_with_sigbus; true" The joy of working with a shell scripting system where everyone has different versions of everything installed... - Eric
