[lustre-devel] [PATCH 06/19] staging: lustre: introduce and use l_wait_event_abortable()

NeilBrown neilb at suse.com
Mon Feb 12 15:58:09 PST 2018


On Mon, Feb 12 2018, Patrick Farrell wrote:

> It's worth noting that the change from -EINTR to -ERESTARTSYS will
> modify the behavior of userspace slightly.  Specifically, when a
> signal handler is setup with retry set (SA_RESTART flag set), the
> syscall will be restarted rather than aborted.

Thanks for the review.
This state is true if the error status ever gets all the way up to user
space.  I don't think it does, though I haven't carefully audited every
l_wait_event_abortable() call site.  I did look at a few and the return
value is only used locally in the same function.
I suspect I would have checked for error codes escaping when I wrote
the patch, but I don't have a clear memory.

Some sort of audit wouldn't hurt of course.

Thanks,
NeilBrown


>
> This should be fine.  It may eventually shake out some stuble bugs in Lustre when we abort an operation and then restart it from a point where we didn't in the past - past instances where we changed from -EINTR to -ERESTARTSYS certainly have - but it's for the best.  As I understand it from past conversations with Andreas and others, Lustre is not really intending to claim it's not restartable, it's just an artifact of people copying older code that was written without awareness of the difference in EINTR vs ERESTARTSYS.
>
> Ideally someone should go through and audit the remaining uses of -EINTR and replace most of them with -ERESTARTSYS.
>
> James, maybe you want to add that to the TODO list?
>
> On 2/12/18, 3:24 PM, "lustre-devel on behalf of NeilBrown" <lustre-devel-bounces at lists.lustre.org on behalf of neilb at suse.com> wrote:
>
>     lustre sometimes wants to wait for an event, but abort if
>     one of a specific list of signals arrives.  This is a little
>     bit like wait_event_killable(), except that the signals are
>     identified a different way.
>     
>     So introduce l_wait_event_abortable() which provides this
>     functionality.
>     Having separate functions for separate needs is more in line
>     with the pattern set by include/linux/wait.h, than having a
>     single function which tries to include all possible needs.
>     
>     Also introduce l_wait_event_abortable_exclusive().
>     
>     Note that l_wait_event() return -EINTR on a signal, while
>     Linux wait_event functions return -ERESTARTSYS.
>     l_wait_event_{abortable_,}exclusive follow the Linux pattern.
>     
>     Reviewed-by: James Simmons <jsimmons at infradead.org>
>     Signed-off-by: NeilBrown <neilb at suse.com>
>     ---
>      drivers/staging/lustre/lustre/include/lustre_lib.h |   24 ++++++++++++++++++++
>      drivers/staging/lustre/lustre/ldlm/ldlm_resource.c |   12 +++++-----
>      drivers/staging/lustre/lustre/llite/llite_lib.c    |   12 +++-------
>      drivers/staging/lustre/lustre/obdclass/genops.c    |    9 +++-----
>      drivers/staging/lustre/lustre/obdclass/llog_obd.c  |    5 ++--
>      drivers/staging/lustre/lustre/osc/osc_page.c       |    6 ++---
>      drivers/staging/lustre/lustre/osc/osc_request.c    |    6 ++---
>      7 files changed, 43 insertions(+), 31 deletions(-)
>     
>     diff --git a/drivers/staging/lustre/lustre/include/lustre_lib.h b/drivers/staging/lustre/lustre/include/lustre_lib.h
>     index 7d950c53e962..b2a64d0e682c 100644
>     --- a/drivers/staging/lustre/lustre/include/lustre_lib.h
>     +++ b/drivers/staging/lustre/lustre/include/lustre_lib.h
>     @@ -336,4 +336,28 @@ do {									   \
>      /** @} lib */
>      
>      
>     +
>     +/* l_wait_event_abortable() is a bit like wait_event_killable()
>     + * except there is a fixed set of signals which will abort:
>     + * LUSTRE_FATAL_SIGS
>     + */
>     +#define l_wait_event_abortable(wq, condition)				\
>     +({									\
>     +	sigset_t __blocked;						\
>     +	int __ret = 0;							\
>     +	__blocked = cfs_block_sigsinv(LUSTRE_FATAL_SIGS);		\
>     +	__ret = wait_event_interruptible(wq, condition);		\
>     +	cfs_restore_sigs(__blocked);					\
>     +	__ret;								\
>     +})
>     +
>     +#define l_wait_event_abortable_exclusive(wq, condition)			\
>     +({									\
>     +	sigset_t __blocked;						\
>     +	int __ret = 0;							\
>     +	__blocked = cfs_block_sigsinv(LUSTRE_FATAL_SIGS);		\
>     +	__ret = wait_event_interruptible_exclusive(wq, condition);	\
>     +	cfs_restore_sigs(__blocked);					\
>     +	__ret;								\
>     +})
>      #endif /* _LUSTRE_LIB_H */
>     diff --git a/drivers/staging/lustre/lustre/ldlm/ldlm_resource.c b/drivers/staging/lustre/lustre/ldlm/ldlm_resource.c
>     index 2e66825c8f4b..4c44603ab6f9 100644
>     --- a/drivers/staging/lustre/lustre/ldlm/ldlm_resource.c
>     +++ b/drivers/staging/lustre/lustre/ldlm/ldlm_resource.c
>     @@ -879,7 +879,6 @@ static int __ldlm_namespace_free(struct ldlm_namespace *ns, int force)
>      	ldlm_namespace_cleanup(ns, force ? LDLM_FL_LOCAL_ONLY : 0);
>      
>      	if (atomic_read(&ns->ns_bref) > 0) {
>     -		struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
>      		int rc;
>      
>      		CDEBUG(D_DLMTRACE,
>     @@ -887,11 +886,12 @@ static int __ldlm_namespace_free(struct ldlm_namespace *ns, int force)
>      		       ldlm_ns_name(ns), atomic_read(&ns->ns_bref));
>      force_wait:
>      		if (force)
>     -			lwi = LWI_TIMEOUT(msecs_to_jiffies(obd_timeout *
>     -					  MSEC_PER_SEC) / 4, NULL, NULL);
>     -
>     -		rc = l_wait_event(ns->ns_waitq,
>     -				  atomic_read(&ns->ns_bref) == 0, &lwi);
>     +			rc = wait_event_idle_timeout(ns->ns_waitq,
>     +						     atomic_read(&ns->ns_bref) == 0,
>     +						     obd_timeout * HZ / 4) ? 0 : -ETIMEDOUT;
>     +		else
>     +			rc = l_wait_event_abortable(ns->ns_waitq,
>     +						    atomic_read(&ns->ns_bref) == 0);
>      
>      		/* Forced cleanups should be able to reclaim all references,
>      		 * so it's safe to wait forever... we can't leak locks...
>     diff --git a/drivers/staging/lustre/lustre/llite/llite_lib.c b/drivers/staging/lustre/lustre/llite/llite_lib.c
>     index c820b201af71..ccb614bd7f53 100644
>     --- a/drivers/staging/lustre/lustre/llite/llite_lib.c
>     +++ b/drivers/staging/lustre/lustre/llite/llite_lib.c
>     @@ -986,16 +986,12 @@ void ll_put_super(struct super_block *sb)
>      	}
>      
>      	/* Wait for unstable pages to be committed to stable storage */
>     -	if (!force) {
>     -		struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
>     -
>     -		rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
>     -				  !atomic_long_read(&sbi->ll_cache->ccc_unstable_nr),
>     -				  &lwi);
>     -	}
>     +	if (!force)
>     +		rc = l_wait_event_abortable(sbi->ll_cache->ccc_unstable_waitq,
>     +					    !atomic_long_read(&sbi->ll_cache->ccc_unstable_nr));
>      
>      	ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
>     -	if (!force && rc != -EINTR)
>     +	if (!force && rc != -ERESTARTSYS)
>      		LASSERTF(!ccc_count, "count: %li\n", ccc_count);
>      
>      	/* We need to set force before the lov_disconnect in
>     diff --git a/drivers/staging/lustre/lustre/obdclass/genops.c b/drivers/staging/lustre/lustre/obdclass/genops.c
>     index 3ff25b8d3b48..8f776a4058a9 100644
>     --- a/drivers/staging/lustre/lustre/obdclass/genops.c
>     +++ b/drivers/staging/lustre/lustre/obdclass/genops.c
>     @@ -1332,7 +1332,6 @@ static bool obd_request_slot_avail(struct client_obd *cli,
>      int obd_get_request_slot(struct client_obd *cli)
>      {
>      	struct obd_request_slot_waiter orsw;
>     -	struct l_wait_info lwi;
>      	int rc;
>      
>      	spin_lock(&cli->cl_loi_list_lock);
>     @@ -1347,11 +1346,9 @@ int obd_get_request_slot(struct client_obd *cli)
>      	orsw.orsw_signaled = false;
>      	spin_unlock(&cli->cl_loi_list_lock);
>      
>     -	lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
>     -	rc = l_wait_event(orsw.orsw_waitq,
>     -			  obd_request_slot_avail(cli, &orsw) ||
>     -			  orsw.orsw_signaled,
>     -			  &lwi);
>     +	rc = l_wait_event_abortable(orsw.orsw_waitq,
>     +				    obd_request_slot_avail(cli, &orsw) ||
>     +				    orsw.orsw_signaled);
>      
>      	/*
>      	 * Here, we must take the lock to avoid the on-stack 'orsw' to be
>     diff --git a/drivers/staging/lustre/lustre/obdclass/llog_obd.c b/drivers/staging/lustre/lustre/obdclass/llog_obd.c
>     index 28bbaa2136ac..26aea114a29b 100644
>     --- a/drivers/staging/lustre/lustre/obdclass/llog_obd.c
>     +++ b/drivers/staging/lustre/lustre/obdclass/llog_obd.c
>     @@ -104,7 +104,6 @@ EXPORT_SYMBOL(__llog_ctxt_put);
>      
>      int llog_cleanup(const struct lu_env *env, struct llog_ctxt *ctxt)
>      {
>     -	struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
>      	struct obd_llog_group *olg;
>      	int rc, idx;
>      
>     @@ -129,8 +128,8 @@ int llog_cleanup(const struct lu_env *env, struct llog_ctxt *ctxt)
>      		CERROR("Error %d while cleaning up ctxt %p\n",
>      		       rc, ctxt);
>      
>     -	l_wait_event(olg->olg_waitq,
>     -		     llog_group_ctxt_null(olg, idx), &lwi);
>     +	l_wait_event_abortable(olg->olg_waitq,
>     +			     llog_group_ctxt_null(olg, idx));
>      
>      	return rc;
>      }
>     diff --git a/drivers/staging/lustre/lustre/osc/osc_page.c b/drivers/staging/lustre/lustre/osc/osc_page.c
>     index 20094b6309f9..6fdd521feb21 100644
>     --- a/drivers/staging/lustre/lustre/osc/osc_page.c
>     +++ b/drivers/staging/lustre/lustre/osc/osc_page.c
>     @@ -759,7 +759,6 @@ static long osc_lru_reclaim(struct client_obd *cli, unsigned long npages)
>      static int osc_lru_alloc(const struct lu_env *env, struct client_obd *cli,
>      			 struct osc_page *opg)
>      {
>     -	struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
>      	struct osc_io *oio = osc_env_io(env);
>      	int rc = 0;
>      
>     @@ -782,9 +781,8 @@ static int osc_lru_alloc(const struct lu_env *env, struct client_obd *cli,
>      
>      		cond_resched();
>      
>     -		rc = l_wait_event(osc_lru_waitq,
>     -				  atomic_long_read(cli->cl_lru_left) > 0,
>     -				  &lwi);
>     +		rc = l_wait_event_abortable(osc_lru_waitq,
>     +					    atomic_long_read(cli->cl_lru_left) > 0);
>      
>      		if (rc < 0)
>      			break;
>     diff --git a/drivers/staging/lustre/lustre/osc/osc_request.c b/drivers/staging/lustre/lustre/osc/osc_request.c
>     index 45b1ebf33363..074b5ce6284c 100644
>     --- a/drivers/staging/lustre/lustre/osc/osc_request.c
>     +++ b/drivers/staging/lustre/lustre/osc/osc_request.c
>     @@ -552,14 +552,12 @@ static int osc_destroy(const struct lu_env *env, struct obd_export *exp,
>      
>      	req->rq_interpret_reply = osc_destroy_interpret;
>      	if (!osc_can_send_destroy(cli)) {
>     -		struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
>     -
>      		/*
>      		 * Wait until the number of on-going destroy RPCs drops
>      		 * under max_rpc_in_flight
>      		 */
>     -		l_wait_event_exclusive(cli->cl_destroy_waitq,
>     -				       osc_can_send_destroy(cli), &lwi);
>     +		l_wait_event_abortable_exclusive(cli->cl_destroy_waitq,
>     +					       osc_can_send_destroy(cli));
>      	}
>      
>      	/* Do not wait for response */
>     
>     
>     _______________________________________________
>     lustre-devel mailing list
>     lustre-devel at lists.lustre.org
>     http://lists.lustre.org/listinfo.cgi/lustre-devel-lustre.org
>     
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 832 bytes
Desc: not available
URL: <http://lists.lustre.org/pipermail/lustre-devel-lustre.org/attachments/20180213/71244c1c/attachment-0001.sig>


More information about the lustre-devel mailing list