Date: Sun, 4 May 2003 12:48:24 -0400 (EDT) From: Robert Watson <rwatson@FreeBSD.org> To: Philippe Charnier <charnier@xp11.frmug.org> Cc: jhb@FreeBSD.org Subject: Re: panic: mutex process lock not owned at ../../../kern/sys_process.c:97 Message-ID: <Pine.NEB.3.96L.1030504124255.24722E-100000@fledge.watson.org> In-Reply-To: <200305041541.h44Ffb7n001131@xp11.frmug.org>
next in thread | previous in thread | raw e-mail | index | archive | help
On Sun, 4 May 2003, Philippe Charnier wrote: > On Wed, 30 Apr 2003, Radko Keves <rado@studnet.sk> got this panic. The > panic is easy to reproduce using `truss ls'. John Baldwin asked for a > stack trace. Here is one: > 79 PROC_UNLOCK(p); > 80 if (kl < 0) > 81 error = EINVAL; > 82 else > 83 /* XXXKSE: */ > 84 error = proc_read_regs(FIRST_THREAD_IN_PROC(p), &r); > 85 if (error == 0) > 86 error = uiomove(kv, kl, uio); > 87 PROC_LOCK(p); > 88 if (error == 0 && uio->uio_rw == UIO_WRITE) { Try moving the PROC_UNLOCK() call from line 79 to just after line 84 (i.e., before the error check and possible uiomove()). It looks like some similar bugs might exist in other bits of procfs. I've attached a patch that tries to more generally handle use of the proc lock more properly with uiomove(), but might also not be perfect. Robert N M Watson FreeBSD Core Team, TrustedBSD Projects robert@fledge.watson.org Network Associates Laboratories Index: procfs_dbregs.c =================================================================== RCS file: /home/ncvs/src/sys/fs/procfs/procfs_dbregs.c,v retrieving revision 1.21 diff -u -r1.21 procfs_dbregs.c --- procfs_dbregs.c 29 Jun 2002 17:26:15 -0000 1.21 +++ procfs_dbregs.c 4 May 2003 16:45:28 -0000 @@ -87,8 +87,11 @@ else /* XXXKSE: */ error = proc_read_dbregs(FIRST_THREAD_IN_PROC(p), &r); - if (error == 0) + if (error == 0) { + PROC_UNLOCK(p); error = uiomove(kv, kl, uio); + PROC_LOCK(p); + } if (error == 0 && uio->uio_rw == UIO_WRITE) { if (!P_SHOULDSTOP(p)) /* XXXKSE should be P_TRACED? */ error = EBUSY; Index: procfs_fpregs.c =================================================================== RCS file: /home/ncvs/src/sys/fs/procfs/procfs_fpregs.c,v retrieving revision 1.27 diff -u -r1.27 procfs_fpregs.c --- procfs_fpregs.c 29 Jun 2002 17:26:15 -0000 1.27 +++ procfs_fpregs.c 4 May 2003 16:44:43 -0000 @@ -81,8 +81,11 @@ else /* XXXKSE: */ error = proc_read_fpregs(FIRST_THREAD_IN_PROC(p), &r); - if (error == 0) + if (error == 0) { + PROC_UNLOCK(p); error = uiomove(kv, kl, uio); + PROC_LOCK(p); + } if (error == 0 && uio->uio_rw == UIO_WRITE) { if (!P_SHOULDSTOP(p)) error = EBUSY; Index: procfs_ioctl.c =================================================================== RCS file: /home/ncvs/src/sys/fs/procfs/procfs_ioctl.c,v retrieving revision 1.9 diff -u -r1.9 procfs_ioctl.c --- procfs_ioctl.c 17 Apr 2003 22:13:46 -0000 1.9 +++ procfs_ioctl.c 4 May 2003 16:46:16 -0000 @@ -67,6 +67,9 @@ *(unsigned int *)data = p->p_pfsflags; break; case PIOCWAIT: + /* + * Should PHOLD() and relase proc lock here? + */ while (p->p_step == 0) { /* sleep until p stops */ error = msleep(&p->p_stype, &p->p_mtx, Index: procfs_regs.c =================================================================== RCS file: /home/ncvs/src/sys/fs/procfs/procfs_regs.c,v retrieving revision 1.26 diff -u -r1.26 procfs_regs.c --- procfs_regs.c 29 Jun 2002 17:26:15 -0000 1.26 +++ procfs_regs.c 4 May 2003 16:44:57 -0000 @@ -76,15 +76,16 @@ kl = uio->uio_resid; _PHOLD(p); - PROC_UNLOCK(p); if (kl < 0) error = EINVAL; else /* XXXKSE: */ error = proc_read_regs(FIRST_THREAD_IN_PROC(p), &r); - if (error == 0) + if (error == 0) { + PROC_UNLOCK(p); error = uiomove(kv, kl, uio); - PROC_LOCK(p); + PROC_LOCK(p); + } if (error == 0 && uio->uio_rw == UIO_WRITE) { if (!P_SHOULDSTOP(p)) error = EBUSY;
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?Pine.NEB.3.96L.1030504124255.24722E-100000>