For some reason which remains mysterious to me the shared library
which calls pthread_create() must be linked with -pthread. If this
is not done on 32-bit system the default ulimit stack size is used.
Surprisingly, on a 64-bit system the stack limit specified by the
pthread_attr is honored even when -pthread is not passed when linking
the shared library.
The following files were being left out the tarball which
is used by the 'make rpm' target. This results in a build
failure when building packages which use the tarball.
lib/libspl/include/sys/tsol/*.h
lib/libspl/include/util/*.h
It turns out the gcc option -Wframe-larger-than=<size> which I recently
added to the build system is not supported in older versions of gcc.
Since this is just a flag to ensure I keep stack usage under control
I've added a configure check to detect if gcc supports it. If it's
available we use it in the proper places, if it's not we don't.
For all module/library functions ensure so stack frame exceeds 1024
bytes. Ideally this should be set lower to say 512 bytes but there
are still numerous functions which exceed even this limit. For now
this is set to 1024 to ensure we catch the worst offenders.
Additionally, set the limit for ztest to 1024 bytes since the idea
here is to catch stack issues in user space before we find them by
overrunning a kernel stack. This should also be reduced to 512
bytes as soon as all the trouble makes are fixed.
Finally, add -fstack-check to gcc build options when --enable-debug
is specified at configure time. This ensures that each page on the
stack will be touched and we will generate a segfault on stack
overflow.
Over time we can gradually fix the following functions:
536 zfs:dsl_deadlist_regenerate
536 zfs:dsl_load_sets
536 zfs:zil_parse
544 zfs:zfs_ioc_recv
552 zfs:dsl_deadlist_insert_bpobj
552 zfs:vdev_dtl_sync
584 zfs:copy_create_perms
608 zfs:ddt_class_contains
608 zfs:ddt_prefetch
608 zfs:__dprintf
616 zfs:ddt_lookup
648 zfs:dsl_scan_ddt
696 zfs:dsl_deadlist_merge
736 zfs:ddt_zap_walk
744 zfs:dsl_prop_get_all_impl
872 zfs:dnode_evict_dbufs
There are 3 fixes in thie commit. First, update ztest_run() to store
the thread id and not the address of the kthread_t. This will be freed
on thread exit and is not safe to use. This is pretty close to how
things were done in the original ztest code before I got there.
Second, for extra paranoia update thread_exit() to return a special
TS_MAGIC value via pthread_exit(). This value is then verified in
pthread_join() to ensure the thread exited cleanly. This can be
done cleanly because the kthread doesn't provide a return code
mechanism we need to worry about.
Third, replace the ztest deadman thread with a signal handler. We
cannot use the previous approach because the correct behavior for
pthreads is to wait for all threads to exit before terminating the
process. Since the deadman thread won't call exit by design we
end up hanging in kernel_exit(). To avoid this we just setup a
SIGALRM signal handle and register a deadman alarm. IMHO this
is simpler and cleaner anyway.
On a Linux system simply use the native aprintf and vasprintf
functions respectively. Also update the call points to correctly
use va_copy() or va_start() as appropriate.
There was previous discussion of a race with joinable threads but to
be honest I can neither exactly remember the race, or recrease the
issue. I believe it may have had to do with pthread_create() returning
without having set kt->tid since this was done in the created thread.
If that was the race then I've 'fixed' it by ensuring the thread id
is set in the thread AND as the first pthread_create() argument. Why
this wasn't done originally I'm not sure, with luck Ricardo remembers.
Additionally, explicitly set a PAGESIZE guard frame at the end of the
stack to aid in detecting stack overflow. And add some conditional
logic to set STACK_SIZE correctly for Solaris.
Certain function must never be automatically inlined by gcc because
they are stack heavy or called recursively. This patch flags all
such functions I have found as 'noinline' to prevent gcc from making
the optimization.
Solaris recently introduced the idea of drive topology because
where a drive is located does matter. I have already handled
this with udev/blkid integration under Linux so I'm hopeful
this case can simply be removed but for now I've just stubbed
out what is needed in libspl and commented out the rest here.
The upstream ZFS code has correctly moved to a faster native sha2
implementation. Unfortunately, under Linux that's going to be a little
problematic so we revert the code to the more portable version contained
in earlier ZFS releases. Using the native sha2 implementation in Linux
is possible but the API is slightly different in kernel version user
space depending on which libraries are used. Ideally, we need a fast
implementation of SHA256 which builds as part of ZFS this shouldn't be
that hard to do but it will take some effort.
This is a portability change which removes the dependence of the Solaris
thread library. All locations where Solaris thread API was used before
have been replaced with equivilant Solaris kernel style thread calls.
In user space the kernel style threading API is implemented in term of
the portable pthreads library. This includes all threads, mutexs,
condition variables, reader/writer locks, and taskqs.
The major change is removing the thread pool when importing devices.
This may be reintroduced at some point if needed, but it is added
complexity which has already been handled by blkid on modern Linux
systems. We only need to fallback to probing everything is /dev/
if you config file is toast and even then it only takes a few seconds.
Brian Behlendorf
Ned Bass