sched/wait: Fix the signal handling fix
Subsystems (4)
Peter Zijlstra
maintainerPATCH 1/1 sched/wait: Fix the signal handling fix
Jan Stancek reported that I wrecked things for him by fixing things for Vladimir :/
His report was due to an UNINTERRUPTIBLE wait getting -EINTR, which should not be possible, however my previous patch made this possible by unconditionally checking signal_pending().
We cannot use current->state as was done previously, because the instruction after the store to that variable it can be changed. We must instead pass the initial state along and use that.
--- fs/cifs/inode.c | 6 +++--- fs/nfs/inode.c | 6 +++--- fs/nfs/internal.h | 2 +- fs/nfs/pagelist.c | 2 +- fs/nfs/pnfs.c | 4 ++-- include/linux/wait.h | 10 +++++----- kernel/sched/wait.c | 20 ++++++++++---------- net/sunrpc/sched.c | 6 +++--- 8 files changed, 28 insertions(+), 28 deletions(-)
fs/cifs/inode.c +3 -3
@@ -1831,11 +1831,11 @@ cifs_invalidate_mapping(struct inode *in* @word: long word containing the bit lock */ static int -cifs_wait_bit_killable (struct wait_bit_key *key) +cifs_wait_bit_killable (struct wait_bit_key *key,int mode) { -if (fatal_signal_pending (current)) -return -ERESTARTSYS ;freezable_schedule_unsafe (); +if (signal_pending_state (mode, current)) +return -ERESTARTSYS ;return 0 ; }
fs/nfs/inode.c +3 -3
@@ -75,11 +75,11 @@ nfs_fattr_to_ino_t(struct nfs_fattr *fat* nfs_wait_bit_killable - helper for functions that are sleeping on bit locks * @word: long word containing the bit lock */ -int nfs_wait_bit_killable (struct wait_bit_key *key) +int nfs_wait_bit_killable (struct wait_bit_key *key,int mode) { -if (fatal_signal_pending (current)) -return -ERESTARTSYS ;freezable_schedule_unsafe (); +if (signal_pending_state (mode, current)) +return -ERESTARTSYS ;return 0 ; }EXPORT_SYMBOL_GPL (nfs_wait_bit_killable);
fs/nfs/internal.h +1 -1
@@ -379,7 +379,7 @@ extern int nfs_drop_inode(struct inode *extern void nfs_clear_inode (struct inode *);extern void nfs_evict_inode (struct inode *);void nfs_zap_acl_cache (struct inode *inode); -extern int nfs_wait_bit_killable (struct wait_bit_key *key); +extern int nfs_wait_bit_killable (struct wait_bit_key *key,int mode);/* super.c */ extern const struct super_operations nfs_sops;
fs/nfs/pagelist.c +1 -1
@@ -129,7 +129,7 @@ __nfs_iocounter_wait(struct nfs_io_countset_bit (NFS_IO_INPROGRESS , &c->flags);if (atomic_read (&c->io_count) ==0 )break ; - ret =nfs_wait_bit_killable (&q.key); + ret =nfs_wait_bit_killable (&q.key,TASK_KILLABLE ); }while (atomic_read (&c->io_count) !=0 && !ret);finish_wait (wq, &q.wait);return ret;
fs/nfs/pnfs.c +2 -2
@@ -1466,11 +1466,11 @@ static bool pnfs_within_mdsthreshold(str }/* stop waiting if someone clears NFS_LAYOUT_RETRY_LAYOUTGET bit. */ -static int pnfs_layoutget_retry_bit_wait (struct wait_bit_key *key) +static int pnfs_layoutget_retry_bit_wait (struct wait_bit_key *key,int mode) {if (!test_bit (NFS_LAYOUT_RETRY_LAYOUTGET , key->flags))return 1 ; -return nfs_wait_bit_killable (key); +return nfs_wait_bit_killable (key, mode); }static bool pnfs_prepare_to_retry_layoutget (struct pnfs_layout_hdr *lo)
include/linux/wait.h +5 -5
@@ -145,7 +145,7 @@ __remove_wait_queue(wait_queue_head_t *hlist_del (&old->task_list); } -typedef int wait_bit_action_f (struct wait_bit_key *); +typedef int wait_bit_action_f (struct wait_bit_key *,int mode);void __wake_up (wait_queue_head_t *q,unsigned int mode,int nr,void *key);void __wake_up_locked_key (wait_queue_head_t *q,unsigned int mode,void *key);void __wake_up_sync_key (wait_queue_head_t *q,unsigned int mode,int nr,void *key);
@@ -960,10 +960,10 @@ int wake_bit_function(wait_queue_t *wait }while (0 ) -extern int bit_wait(struct wait_bit_key *); -extern int bit_wait_io (struct wait_bit_key *); -extern int bit_wait_timeout (struct wait_bit_key *); -extern int bit_wait_io_timeout (struct wait_bit_key *); +extern int bit_wait(struct wait_bit_key *, int); +extern int bit_wait_io (struct wait_bit_key *,int ); +extern int bit_wait_timeout (struct wait_bit_key *,int ); +extern int bit_wait_io_timeout (struct wait_bit_key *,int );/** * wait_on_bit - wait for a bit to be cleared
kernel/sched/wait.c +10 -10
@@ -392,7 +392,7 @@ __wait_on_bit(wait_queue_head_t *wq, strdo {prepare_to_wait (wq, &q->wait, mode);if (test_bit (q->key.bit_nr, q->key.flags)) - ret = (*action)(&q->key); + ret = (*action)(&q->key, mode); }while (test_bit (q->key.bit_nr, q->key.flags) && !ret);finish_wait (wq, &q->wait);return ret;
@@ -431,7 +431,7 @@ __wait_on_bit_lock(wait_queue_head_t *wqprepare_to_wait_exclusive (wq, &q->wait, mode);if (!test_bit (q->key.bit_nr, q->key.flags))continue ; - ret =action (&q->key); + ret =action (&q->key, mode);if (!ret)continue ;abort_exclusive_wait (wq, &q->wait, mode, &q->key);
@@ -581,43 +581,43 @@ void wake_up_atomic_t(atomic_t *p) }EXPORT_SYMBOL (wake_up_atomic_t); -__sched int bit_wait(struct wait_bit_key *word) +__sched int bit_wait(struct wait_bit_key *word, int mode) {schedule (); -if (signal_pending (current)) +if (signal_pending_state (mode, current))return -EINTR ;return 0 ; }EXPORT_SYMBOL (bit_wait); -__sched int bit_wait_io(struct wait_bit_key *word) +__sched int bit_wait_io(struct wait_bit_key *word, int mode) {io_schedule (); -if (signal_pending (current)) +if (signal_pending_state (mode, current))return -EINTR ;return 0 ; }EXPORT_SYMBOL (bit_wait_io); -__sched int bit_wait_timeout(struct wait_bit_key *word) +__sched int bit_wait_timeout(struct wait_bit_key *word, int mode) {unsigned long now =READ_ONCE (jiffies);if (time_after_eq (now, word->timeout))return -EAGAIN ;schedule_timeout (word->timeout - now); -if (signal_pending (current)) +if (signal_pending_state (mode, current))return -EINTR ;return 0 ; }EXPORT_SYMBOL_GPL (bit_wait_timeout); -__sched int bit_wait_io_timeout(struct wait_bit_key *word) +__sched int bit_wait_io_timeout(struct wait_bit_key *word, int mode) {unsigned long now =READ_ONCE (jiffies);if (time_after_eq (now, word->timeout))return -EAGAIN ;io_schedule_timeout (word->timeout - now); -if (signal_pending (current)) +if (signal_pending_state (mode, current))return -EINTR ;return 0 ; }
net/sunrpc/sched.c +3 -3
@@ -250,11 +250,11 @@ void rpc_destroy_wait_queue(struct rpc_w }EXPORT_SYMBOL_GPL (rpc_destroy_wait_queue); -static int rpc_wait_bit_killable (struct wait_bit_key *key) +static int rpc_wait_bit_killable (struct wait_bit_key *key,int mode) { -if (fatal_signal_pending (current)) -return -ERESTARTSYS ;freezable_schedule_unsafe (); +if (signal_pending_state (mode, current)) +return -ERESTARTSYS ;return 0 ; }
Peter, sorry. I didn't actually read this patch yet (and a lot of previous emails). Will try tomorrow, I am not even sure I understand the problem(s) correctly. But let me ask one question anyway,
kernel/sched/wait.c __wait_on_bit(wait_queue_head_t *wq, str · line 395
Quoted 9 lines · Peter Zijlstra
kernel/sched/wait.c +1
-1@@ -392,7 +392,7 @@ __wait_on_bit(wait_queue_head_t *wq, str do { prepare_to_wait(wq, &q->wait, mode); if (test_bit(q->key.bit_nr, q->key.flags)) - ret = (*action)(&q->key); + ret = (*action)(&q->key, mode);
And every action() should check signal_pending_state()...
So why we can't change __wait_on_bit/etc instead and remove all the signal- pending checks from the callbacks? It seems that we can just check signal_pending_state() before prepare_to_wait(). Or perhaps we can add another helper which acts like prepare_to_wait_event().
Yes, some callers want -EINTR, some -ERESTARTSYS, but this shouldn't be a problem.
And sorry if this was already discussed, another case when I am trying to return to lkml with a lot of unread emails.
Oleg.
Peter Zijlstra
maintainerQuoted 9 lines · Oleg Nesterov
kernel/sched/wait.c __wait_on_bit(wait_queue_head_t *wq, str · line 395
Quoted 1 line
kernel/sched/wait.c +1
-0+ ret = (*action)(&q->key, mode);And every
action()should checksignal_pending_state()...So why we can't change __wait_on_bit/etc instead and remove all the signal- pending checks from the callbacks? It seems that we can just check
signal_pending_state()beforeprepare_to_wait(). Or perhaps we can add another helper which acts likeprepare_to_wait_event().
Quoted 5 lines
Yes, some callers want
-EINTR, some-ERESTARTSYS, but this shouldn't be a problem.And sorry if this was already discussed, another case when I am trying to return to lkml with a lot of unread emails.
Yes that looks like a viable cleanup. But at least now we have a base that's working for everyone.
I'll try and do some patches tomorrow.