From owner-freebsd-current@freebsd.org Tue Jun 7 16:30:19 2016 Return-Path: Delivered-To: freebsd-current@mailman.ysv.freebsd.org Received: from mx1.freebsd.org (mx1.freebsd.org [IPv6:2001:1900:2254:206a::19:1]) by mailman.ysv.freebsd.org (Postfix) with ESMTP id 6B928B6E65D for ; Tue, 7 Jun 2016 16:30:19 +0000 (UTC) (envelope-from kostikbel@gmail.com) Received: from kib.kiev.ua (kib.kiev.ua [IPv6:2001:470:d5e7:1::1]) (using TLSv1 with cipher DHE-RSA-CAMELLIA256-SHA (256/256 bits)) (Client did not present a certificate) by mx1.freebsd.org (Postfix) with ESMTPS id F06A11ACC; Tue, 7 Jun 2016 16:30:18 +0000 (UTC) (envelope-from kostikbel@gmail.com) Received: from tom.home (kib@localhost [127.0.0.1]) by kib.kiev.ua (8.15.2/8.15.2) with ESMTPS id u57GUC0m018461 (version=TLSv1 cipher=DHE-RSA-CAMELLIA256-SHA bits=256 verify=NO); Tue, 7 Jun 2016 19:30:12 +0300 (EEST) (envelope-from kostikbel@gmail.com) DKIM-Filter: OpenDKIM Filter v2.10.3 kib.kiev.ua u57GUC0m018461 Received: (from kostik@localhost) by tom.home (8.15.2/8.15.2/Submit) id u57GUB0I018446; Tue, 7 Jun 2016 19:30:11 +0300 (EEST) (envelope-from kostikbel@gmail.com) X-Authentication-Warning: tom.home: kostik set sender to kostikbel@gmail.com using -f Date: Tue, 7 Jun 2016 19:30:11 +0300 From: Konstantin Belousov To: Jilles Tjoelker Cc: Mark Johnston , freebsd-current@FreeBSD.org, cem@FreeBSD.org Subject: Re: thread suspension when dumping core Message-ID: <20160607163011.GR38613@kib.kiev.ua> References: <20160604022347.GA1096@wkstn-mjohnston.west.isilon.com> <20160604093236.GA38613@kib.kiev.ua> <20160606171311.GC10101@wkstn-mjohnston.west.isilon.com> <20160607024610.GI38613@kib.kiev.ua> <20160607041741.GA29017@wkstn-mjohnston.west.isilon.com> <20160607042956.GM38613@kib.kiev.ua> <20160607142452.GA48251@stack.nl> <20160607160155.GP38613@kib.kiev.ua> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <20160607160155.GP38613@kib.kiev.ua> User-Agent: Mutt/1.6.1 (2016-04-27) X-Spam-Status: No, score=-2.0 required=5.0 tests=ALL_TRUSTED,BAYES_00, DKIM_ADSP_CUSTOM_MED,FREEMAIL_FROM,NML_ADSP_CUSTOM_MED autolearn=no autolearn_force=no version=3.4.1 X-Spam-Checker-Version: SpamAssassin 3.4.1 (2015-04-28) on tom.home X-BeenThere: freebsd-current@freebsd.org X-Mailman-Version: 2.1.22 Precedence: list List-Id: Discussions about the use of FreeBSD-current List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Tue, 07 Jun 2016 16:30:19 -0000 On Tue, Jun 07, 2016 at 07:01:55PM +0300, Konstantin Belousov wrote: > On Tue, Jun 07, 2016 at 04:24:53PM +0200, Jilles Tjoelker wrote: > > On Tue, Jun 07, 2016 at 07:29:56AM +0300, Konstantin Belousov wrote: > > > This looks as if we should not ignore suspension requests in > > > thread_suspend_check() completely in TDF_SBDRY case, but return either > > > EINTR or ERESTART (most likely ERESTART). Note that the goal of > > > TDF_SBDRY is to avoid suspending in the protected region, not to make an > > > impression that the suspension does not occur at all. > > > > This looks like it would revert r246417 and re-introduce the bug fixed > > by it (unexpected [EINTR] and short reads/writes after stop signals). > Well, the patch returns ERESTART and not EINTR, so the syscall should > be retried after all the unwinding. > > > > > After r246417, TDF_SBDRY is intended for sleeps that occur while holding > > resources such as vnode locks and are normally short but should be > > interruptible by fatal signals because they may occasionally be > > indefinitely long (such as a non-responsive NFS server). > > > > It looks like yet another kind of sleep may be required, since advisory > > locks still hold some filesystem resources across the sleep (though not > > vnode locks). > I do not think that adv locks enter sleep with any resource held which > would block other threads. But I agree with the statement because the > lock might be granted and then the stopped thread would appear to own > the blocking resource. > > > > > We then have four kinds: > > > > * uninterruptible by signals, ignores stops (default) > > * interruptible by signals, ignores stops (current TDF_SBDRY with > > PCATCH) > > * interruptible by signals, freezes in place on stops (avoids > > unexpected short I/O) (current PCATCH, otherwise) > > * interruptible by signals, fails with [ERESTART] on stops (avoids > > holding resources across a stop) (new) > > > > The new kind of sleep would fail with [ERESTART] only for stops, since > > [EINTR] should only be returned if a signal handler was called. There > > cannot be a signal handler since a SIGTSTP/SIGTTIN/SIGTTOU signal with a > > handler does not stop the process. > > > And where would this new kind of sleep used ? The advlock sleep is the one > place. Does fifo sleep for reader or writer on open require this kind > of handling (IMO no) ? > > I think this can be relatively easily implemented with either a flag > for XXXsleep(9) (my older style of PBDRY) or using only the thread flag > (jhb' newer TDF_SBDRY approach). Probably the later should be used, for > consistency and easier marking of larger blocks of code. Like this. diff --git a/sys/kern/kern_lockf.c b/sys/kern/kern_lockf.c index a0a3789..ee26596 100644 --- a/sys/kern/kern_lockf.c +++ b/sys/kern/kern_lockf.c @@ -1378,7 +1378,7 @@ lf_setlock(struct lockf *state, struct lockf_entry *lock, struct vnode *vp, void **cookiep) { static char lockstr[] = "lockf"; - int priority, error; + int error, priority, stoprestart; #ifdef LOCKF_DEBUG if (lockf_debug & 1) @@ -1466,7 +1466,10 @@ lf_setlock(struct lockf *state, struct lockf_entry *lock, struct vnode *vp, } lock->lf_refs++; + stoprestart = sigstoprestart(); error = sx_sleep(lock, &state->ls_lock, priority, lockstr, 0); + if (stoprestart) + sigstopnormal(); if (lf_free_lock(lock)) { error = EDOOFUS; goto out; diff --git a/sys/kern/kern_sig.c b/sys/kern/kern_sig.c index 75a1259..1d7036d 100644 --- a/sys/kern/kern_sig.c +++ b/sys/kern/kern_sig.c @@ -2633,6 +2633,35 @@ sigallowstop(void) return (prev); } +int +sigstoprestart(void) +{ + struct thread *td; + + td = curthread; + if ((td->td_flags & TDF_SBDRY) == 0 || + (td->td_flags & TDF_SRESTART) != 0) + return (0); + thread_lock(td); + td->td_flags |= TDF_SRESTART; + thread_unlock(td); + return (1); +} + +int +sigstopnormal(void) +{ + struct thread *td; + int prev; + + td = curthread; + thread_lock(td); + prev = (td->td_flags & TDF_SRESTART) != 0; + td->td_flags &= ~TDF_SRESTART; + thread_unlock(td); + return (prev); +} + /* * If the current process has received a signal (should be caught or cause * termination, should interrupt current syscall), return the signal number. diff --git a/sys/kern/kern_thread.c b/sys/kern/kern_thread.c index 9af377e..6460ae9 100644 --- a/sys/kern/kern_thread.c +++ b/sys/kern/kern_thread.c @@ -932,7 +932,8 @@ thread_suspend_check(int return_instead) if ((td->td_flags & TDF_SBDRY) != 0) { KASSERT(return_instead, ("TDF_SBDRY set for unsafe thread_suspend_check")); - return (0); + return ((td->td_flags & TDF_SRESTART) != 0 ? + ERESTART : 0); } /* diff --git a/sys/sys/proc.h b/sys/sys/proc.h index 629f7e8..1e986a9 100644 --- a/sys/sys/proc.h +++ b/sys/sys/proc.h @@ -395,7 +395,7 @@ do { \ #define TDF_NEEDRESCHED 0x00010000 /* Thread needs to yield. */ #define TDF_NEEDSIGCHK 0x00020000 /* Thread may need signal delivery. */ #define TDF_NOLOAD 0x00040000 /* Ignore during load avg calculations. */ -#define TDF_UNUSED19 0x00080000 /* --available-- */ +#define TDF_SRESTART 0x00080000 /* ERESTART on stop attempts. */ #define TDF_THRWAKEUP 0x00100000 /* Libthr thread must not suspend itself. */ #define TDF_UNUSED21 0x00200000 /* --available-- */ #define TDF_SWAPINREQ 0x00400000 /* Swapin request due to wakeup. */ diff --git a/sys/sys/signalvar.h b/sys/sys/signalvar.h index e574ec3..3d4c4a5 100644 --- a/sys/sys/signalvar.h +++ b/sys/sys/signalvar.h @@ -328,6 +328,8 @@ extern struct mtx sigio_lock; int cursig(struct thread *td); int sigdeferstop(void); int sigallowstop(void); +int sigstoprestart(void); +int sigstopnormal(void); void execsigs(struct proc *p); void gsignal(int pgid, int sig, ksiginfo_t *ksi); void killproc(struct proc *p, char *why);