aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorJonathan Anderson <jonathan@FreeBSD.org>2011-08-18 22:51:30 +0000
committerJonathan Anderson <jonathan@FreeBSD.org>2011-08-18 22:51:30 +0000
commitcfb5f76865887cb85825ed970b8bbc4ab2448b5d (patch)
treeec80efcada771dd68fe28d71eaf850289af0e772
parentf55d3fbe8438d73a393729763a4abca5b72233a2 (diff)
downloadsrc-cfb5f76865887cb85825ed970b8bbc4ab2448b5d.tar.gz
src-cfb5f76865887cb85825ed970b8bbc4ab2448b5d.zip
Add experimental support for process descriptors
A "process descriptor" file descriptor is used to manage processes without using the PID namespace. This is required for Capsicum's Capability Mode, where the PID namespace is unavailable. New system calls pdfork(2) and pdkill(2) offer the functional equivalents of fork(2) and kill(2). pdgetpid(2) allows querying the PID of the remote process for debugging purposes. The currently-unimplemented pdwait(2) will, in the future, allow querying rusage/exit status. In the interim, poll(2) may be used to check (and wait for) process termination. When a process is referenced by a process descriptor, it does not issue SIGCHLD to the parent, making it suitable for use in libraries---a common scenario when using library compartmentalisation from within large applications (such as web browsers). Some observers may note a similarity to Mach task ports; process descriptors provide a subset of this behaviour, but in a UNIX style. This feature is enabled by "options PROCDESC", but as with several other Capsicum kernel features, is not enabled by default in GENERIC 9.0. Reviewed by: jhb, kib Approved by: re (kib), mentor (rwatson) Sponsored by: Google Inc
Notes
Notes: svn path=/head/; revision=224987
-rw-r--r--lib/libc/sys/Makefile.inc5
-rw-r--r--lib/libc/sys/Symbol.map3
-rw-r--r--lib/libc/sys/cap_new.24
-rw-r--r--lib/libc/sys/pdfork.2182
-rw-r--r--sys/compat/linux/linux_fork.c8
-rw-r--r--sys/conf/NOTES3
-rw-r--r--sys/conf/files1
-rw-r--r--sys/conf/options1
-rw-r--r--sys/kern/capabilities.conf2
-rw-r--r--sys/kern/init_main.c3
-rw-r--r--sys/kern/kern_descrip.c30
-rw-r--r--sys/kern/kern_exit.c82
-rw-r--r--sys/kern/kern_fork.c96
-rw-r--r--sys/kern/kern_kthread.c2
-rw-r--r--sys/kern/kern_sig.c31
-rw-r--r--sys/kern/sys_procdesc.c524
-rw-r--r--sys/kern/syscalls.master8
-rw-r--r--sys/sys/capability.h7
-rw-r--r--sys/sys/file.h1
-rw-r--r--sys/sys/proc.h6
-rw-r--r--sys/sys/procdesc.h119
-rw-r--r--sys/sys/unistd.h5
-rw-r--r--sys/sys/user.h4
23 files changed, 1074 insertions, 53 deletions
diff --git a/lib/libc/sys/Makefile.inc b/lib/libc/sys/Makefile.inc
index ddc157ee1b57..fe5061d97f3c 100644
--- a/lib/libc/sys/Makefile.inc
+++ b/lib/libc/sys/Makefile.inc
@@ -96,7 +96,7 @@ MAN+= abort2.2 accept.2 access.2 acct.2 adjtime.2 \
mq_setattr.2 \
msgctl.2 msgget.2 msgrcv.2 msgsnd.2 \
msync.2 munmap.2 nanosleep.2 nfssvc.2 ntp_adjtime.2 open.2 \
- pathconf.2 pipe.2 poll.2 posix_fallocate.2 posix_openpt.2 profil.2 \
+ pathconf.2 pdfork.2 pipe.2 poll.2 posix_fallocate.2 posix_openpt.2 profil.2 \
pselect.2 ptrace.2 quotactl.2 \
read.2 readlink.2 reboot.2 recv.2 rename.2 revoke.2 rfork.2 rmdir.2 \
rtprio.2
@@ -178,6 +178,9 @@ MLINKS+=ntp_adjtime.2 ntp_gettime.2
MLINKS+=open.2 openat.2
MLINKS+=pathconf.2 fpathconf.2
MLINKS+=pathconf.2 lpathconf.2
+MLINKS+=pdfork.2 pdgetpid.2\
+ pdfork.2 pdkill.2 \
+ pdfork.2 pdwait4.2
MLINKS+=read.2 pread.2 read.2 preadv.2 read.2 readv.2
MLINKS+=readlink.2 readlinkat.2
MLINKS+=recv.2 recvfrom.2 recv.2 recvmsg.2
diff --git a/lib/libc/sys/Symbol.map b/lib/libc/sys/Symbol.map
index 547a2cff9720..095751a441cd 100644
--- a/lib/libc/sys/Symbol.map
+++ b/lib/libc/sys/Symbol.map
@@ -366,6 +366,9 @@ FBSD_1.2 {
cap_new;
cap_getrights;
getloginclass;
+ pdfork;
+ pdgetpid;
+ pdkill;
posix_fallocate;
rctl_get_racct;
rctl_get_rules;
diff --git a/lib/libc/sys/cap_new.2 b/lib/libc/sys/cap_new.2
index 7710e12abdfa..206715e84b4e 100644
--- a/lib/libc/sys/cap_new.2
+++ b/lib/libc/sys/cap_new.2
@@ -260,7 +260,7 @@ Permit
.Xr pdkill 2 .
.It Dv CAP_PDWAIT
Permit
-.Xr pdwait 2 .
+.Xr pdwait4 2 .
.It Dv CAP_PEELOFF
Permit
.Xr sctp_peeloff 2 .
@@ -429,7 +429,7 @@ argument is not a capability.
.Xr openat 2 ,
.Xr pdgetpid 2 ,
.Xr pdkill 2 ,
-.Xr pdwait 2 ,
+.Xr pdwait4 2 ,
.Xr pipe 2 ,
.Xr poll 2 ,
.Xr pread 2 ,
diff --git a/lib/libc/sys/pdfork.2 b/lib/libc/sys/pdfork.2
new file mode 100644
index 000000000000..3f36e881ec9e
--- /dev/null
+++ b/lib/libc/sys/pdfork.2
@@ -0,0 +1,182 @@
+.\"
+.\" Copyright (c) 2009-2010 Robert N. M. Watson
+.\" All rights reserved.
+.\"
+.\" This software was developed at the University of Cambridge Computer
+.\" Laboratory with support from a grant from Google, Inc.
+.\"
+.\" Redistribution and use in source and binary forms, with or without
+.\" modification, are permitted provided that the following conditions
+.\" are met:
+.\" 1. Redistributions of source code must retain the above copyright
+.\" notice, this list of conditions and the following disclaimer.
+.\" 2. Redistributions in binary form must reproduce the above copyright
+.\" notice, this list of conditions and the following disclaimer in the
+.\" documentation and/or other materials provided with the distribution.
+.\"
+.\" THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
+.\" ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+.\" IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+.\" ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
+.\" FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+.\" DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
+.\" OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+.\" HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
+.\" LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
+.\" OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+.\" SUCH DAMAGE.
+.\"
+.\" $FreeBSD$
+.\"
+.Dd August 16, 2011
+.Dt PDFORK 2
+.Os
+.Sh NAME
+.Nm pdfork ,
+.Nm pdgetpid ,
+.Nm pdkill ,
+.Nm pdwait4
+.Nd System calls to manage process descriptors
+.Sh LIBRARY
+.Lb libc
+.Sh SYNOPSIS
+.In sys/procdesc.h
+.Ft int
+.Fn pdfork "int *fdp" "int flags"
+.Ft int
+.Fn pdgetpid "int fd" "pid_t *pidp"
+.Ft int
+.Fn pdkill "int fd" "int signum"
+.Ft int
+.Fn pdwait4 "int fd" "int *status" "int options" "struct rusage *rusage"
+.Sh DESCRIPTION
+Process descriptors are special file descriptors that represent processes,
+and are created using
+.Fn pdfork ,
+a variant of
+.Xr fork 2 ,
+which, if successful, returns a process descriptor in the integer pointed to
+by
+.Fa pidp .
+Processes created via
+.Fn pdfork
+will not cause
+.Dv SIGCHLD
+on termination.
+.Fn pdfork
+can accept the flags:
+.Bl -tag -width ".Dv PD_DAEMON"
+.It Dv PD_DAEMON
+Instead of the default terminate-on-close behaviour, allow the process to
+live until it is explicitly killed with
+.Xr kill 2 .
+.Pp
+This option is not permitted in Capsicum capability mode (see
+.Xr cap_enter 2 ) .
+.El
+.Pp
+.Fn pdgetpid
+queries the process ID (PID) if the process descriptor
+.Fa fd .
+.Pp
+.Fn pdkill
+is functionally identical to
+.Xr kill 2 ,
+except that it accepts a process descriptor,
+.Fa fd ,
+rather than a PID.
+.Pp
+.Fn pdwait4
+behaves identially to
+.Xr wait4 2 ,
+but operates with respect to a process descriptor argument rather than a PID.
+.Pp
+The following system calls also have effects specific to process descriptors:
+.Pp
+.Xr fstat 2
+queries status of a process descriptor; currently only the
+.Fa st_mode ,
+.Fa st_birthtime ,
+.Fa st_atime ,
+.Fa st_ctime
+and
+.Fa st_mtime
+fields are defined. If the owner read, write, and execute bits are set then the
+process represented by the process descriptor is still alive.
+.Pp
+.Xr poll 2
+and
+.Xr select 2
+allow waiting for process state transitions; currently only
+.Dv POLLHUP
+is defined, and will be raised when the process dies.
+.Pp
+.Xr close 2
+will close the process descriptor unless
+.Dv PD_DAEMON
+is set; if the process is still alive and this is
+the last reference to the process descriptor, the process will be terminated
+with the signal
+.Dv SIGKILL .
+.Sh RETURN VALUES
+.Fn pdfork
+returns a PID, 0 or -1, as
+.Xr fork 2
+does.
+.Pp
+.Fn pdgetpid
+and
+.Fn pdkill
+return 0 on success and -1 on failure.
+.Pp
+.Fn pdwait4
+returns a PID on success and -1 on failure.
+.Sh ERRORS
+These functions may return the same error numbers as their PID-based equivalents
+(e.g.
+.Fn pdfork
+may return the same error numbers as
+.Xr fork 2 ) ,
+with the following additions:
+.Bl -tag -width Er
+.It Bq Er EINVAL
+The signal number given to
+.Fn pdkill
+is invalid.
+.It Bq Er ENOTCAPABLE
+The process descriptor being operated on has insufficient rights (e.g.
+.Dv CAP_PDKILL
+for
+.Fn pdkill ) .
+.El
+.Sh SEE ALSO
+.Xr close 2 ,
+.Xr fork 2 ,
+.Xr fstat 2 ,
+.Xr kill 2 ,
+.Xr poll 2 ,
+.Xr wait4 2
+.Sh HISTORY
+The
+.Fn pdfork ,
+.Fn pdgetpid ,
+.Fn pdkill
+and
+.Fn pdwait4
+system calls first appeared in
+.Fx 9.0 .
+.Pp
+Support for process descriptors mode was developed as part of the
+.Tn TrustedBSD
+Project.
+.Sh AUTHORS
+.An -nosplit
+These functions and the capability facility were created by
+.An "Robert N. M. Watson" Aq rwatson@FreeBSD.org
+and
+.An "Jonathan Anderson" Aq jonathan@FreeBSD.org
+at the University of Cambridge Computer Laboratory with support from a grant
+from Google, Inc.
+.Sh BUGS
+.Fn pdwait4
+has not yet been implemented.
diff --git a/sys/compat/linux/linux_fork.c b/sys/compat/linux/linux_fork.c
index bf1d45c9ea34..5d2ce5bdb0cb 100644
--- a/sys/compat/linux/linux_fork.c
+++ b/sys/compat/linux/linux_fork.c
@@ -64,7 +64,8 @@ linux_fork(struct thread *td, struct linux_fork_args *args)
printf(ARGS(fork, ""));
#endif
- if ((error = fork1(td, RFFDG | RFPROC | RFSTOPPED, 0, &p2)) != 0)
+ if ((error = fork1(td, RFFDG | RFPROC | RFSTOPPED, 0, &p2, NULL, 0))
+ != 0)
return (error);
td->td_retval[0] = p2->p_pid;
@@ -100,7 +101,8 @@ linux_vfork(struct thread *td, struct linux_vfork_args *args)
#endif
/* Exclude RFPPWAIT */
- if ((error = fork1(td, RFFDG | RFPROC | RFMEM | RFSTOPPED, 0, &p2)) != 0)
+ if ((error = fork1(td, RFFDG | RFPROC | RFMEM | RFSTOPPED, 0, &p2,
+ NULL, 0)) != 0)
return (error);
td->td_retval[0] = p2->p_pid;
@@ -190,7 +192,7 @@ linux_clone(struct thread *td, struct linux_clone_args *args)
if (args->parent_tidptr == NULL)
return (EINVAL);
- error = fork1(td, ff, 0, &p2);
+ error = fork1(td, ff, 0, &p2, NULL, 0);
if (error)
return (error);
diff --git a/sys/conf/NOTES b/sys/conf/NOTES
index 4a9ec35d0380..59f02c812a13 100644
--- a/sys/conf/NOTES
+++ b/sys/conf/NOTES
@@ -1159,6 +1159,9 @@ options MAC_TEST
options CAPABILITIES # fine-grained rights on file descriptors
options CAPABILITY_MODE # sandboxes with no global namespace access
+# Support for process descriptors
+options PROCDESC
+
#####################################################################
# CLOCK OPTIONS
diff --git a/sys/conf/files b/sys/conf/files
index 0dc814e0934a..5c5d92d6b163 100644
--- a/sys/conf/files
+++ b/sys/conf/files
@@ -2412,6 +2412,7 @@ kern/subr_witness.c optional witness
kern/sys_capability.c standard
kern/sys_generic.c standard
kern/sys_pipe.c standard
+kern/sys_procdesc.c standard
kern/sys_process.c standard
kern/sys_socket.c standard
kern/syscalls.c standard
diff --git a/sys/conf/options b/sys/conf/options
index f7026c134177..27fdbedacb60 100644
--- a/sys/conf/options
+++ b/sys/conf/options
@@ -149,6 +149,7 @@ PPC_DEBUG opt_ppc.h
PPC_PROBE_CHIPSET opt_ppc.h
PPS_SYNC opt_ntp.h
PREEMPTION opt_sched.h
+PROCDESC opt_procdesc.h
QUOTA
SCHED_4BSD opt_sched.h
SCHED_STATS opt_sched.h
diff --git a/sys/kern/capabilities.conf b/sys/kern/capabilities.conf
index 004c2ddf2077..4a62643623eb 100644
--- a/sys/kern/capabilities.conf
+++ b/sys/kern/capabilities.conf
@@ -475,7 +475,7 @@ openbsd_poll
pdfork
pdgetpid
pdkill
-pdwait4
+#pdwait4 # not yet implemented
##
## Allow pipe(2).
diff --git a/sys/kern/init_main.c b/sys/kern/init_main.c
index be5c26fec029..fc072457c191 100644
--- a/sys/kern/init_main.c
+++ b/sys/kern/init_main.c
@@ -790,7 +790,8 @@ create_init(const void *udata __unused)
struct ucred *newcred, *oldcred;
int error;
- error = fork1(&thread0, RFFDG | RFPROC | RFSTOPPED, 0, &initproc);
+ error = fork1(&thread0, RFFDG | RFPROC | RFSTOPPED, 0, &initproc,
+ NULL, 0);
if (error)
panic("cannot fork init: %d\n", error);
KASSERT(initproc->p_pid == 1, ("create_init: initproc->p_pid != 1"));
diff --git a/sys/kern/kern_descrip.c b/sys/kern/kern_descrip.c
index 85f866c39c0a..4aaed1f040e8 100644
--- a/sys/kern/kern_descrip.c
+++ b/sys/kern/kern_descrip.c
@@ -41,6 +41,7 @@ __FBSDID("$FreeBSD$");
#include "opt_compat.h"
#include "opt_ddb.h"
#include "opt_ktrace.h"
+#include "opt_procdesc.h"
#include <sys/param.h>
#include <sys/systm.h>
@@ -65,6 +66,7 @@ __FBSDID("$FreeBSD$");
#include <sys/pipe.h>
#include <sys/priv.h>
#include <sys/proc.h>
+#include <sys/procdesc.h>
#include <sys/protosw.h>
#include <sys/racct.h>
#include <sys/resourcevar.h>
@@ -120,6 +122,8 @@ static int fill_vnode_info(struct vnode *vp, struct kinfo_file *kif);
static int fill_socket_info(struct socket *so, struct kinfo_file *kif);
static int fill_pts_info(struct tty *tp, struct kinfo_file *kif);
static int fill_pipe_info(struct pipe *pi, struct kinfo_file *kif);
+static int fill_procdesc_info(struct procdesc *pdp,
+ struct kinfo_file *kif);
/*
* A process is initially started out with NDFILE descriptors stored within
@@ -3056,6 +3060,12 @@ sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS)
tp = fp->f_data;
break;
+#ifdef PROCDESC
+ case DTYPE_PROCDESC:
+ kif->kf_type = KF_TYPE_PROCDESC;
+ break;
+#endif
+
default:
kif->kf_type = KF_TYPE_UNKNOWN;
break;
@@ -3218,6 +3228,9 @@ export_fd_for_sysctl(void *data, int type, int fd, int fflags, int refcnt,
case KF_TYPE_PTS:
error = fill_pts_info((struct tty *)data, kif);
break;
+ case KF_TYPE_PROCDESC:
+ error = fill_procdesc_info((struct procdesc *)data, kif);
+ break;
default:
error = 0;
}
@@ -3391,6 +3404,13 @@ sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS)
data = fp->f_data;
break;
+#ifdef PROCDESC
+ case DTYPE_PROCDESC:
+ type = KF_TYPE_PROCDESC;
+ data = fp->f_data;
+ break;
+#endif
+
default:
type = KF_TYPE_UNKNOWN;
break;
@@ -3586,6 +3606,16 @@ fill_pipe_info(struct pipe *pi, struct kinfo_file *kif)
return (0);
}
+static int
+fill_procdesc_info(struct procdesc *pdp, struct kinfo_file *kif)
+{
+
+ if (pdp == NULL)
+ return (1);
+ kif->kf_un.kf_proc.kf_pid = pdp->pd_pid;
+ return (0);
+}
+
static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc, CTLFLAG_RD,
sysctl_kern_proc_filedesc, "Process filedesc entries");
diff --git a/sys/kern/kern_exit.c b/sys/kern/kern_exit.c
index 30b94b6a28a4..e5d60942f7c0 100644
--- a/sys/kern/kern_exit.c
+++ b/sys/kern/kern_exit.c
@@ -40,16 +40,19 @@ __FBSDID("$FreeBSD$");
#include "opt_compat.h"
#include "opt_kdtrace.h"
#include "opt_ktrace.h"
+#include "opt_procdesc.h"
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/sysproto.h>
+#include <sys/capability.h>
#include <sys/eventhandler.h>
#include <sys/kernel.h>
#include <sys/malloc.h>
#include <sys/lock.h>
#include <sys/mutex.h>
#include <sys/proc.h>
+#include <sys/procdesc.h>
#include <sys/pioctl.h>
#include <sys/jail.h>
#include <sys/tty.h>
@@ -461,39 +464,54 @@ exit1(struct thread *td, int rv)
knlist_clear(&p->p_klist, 1);
/*
- * Notify parent that we're gone. If parent has the PS_NOCLDWAIT
- * flag set, or if the handler is set to SIG_IGN, notify process
- * 1 instead (and hope it will handle this situation).
+ * If this is a process with a descriptor, we may not need to deliver
+ * a signal to the parent. proctree_lock is held over
+ * procdesc_exit() to serialize concurrent calls to close() and
+ * exit().
*/
- PROC_LOCK(p->p_pptr);
- mtx_lock(&p->p_pptr->p_sigacts->ps_mtx);
- if (p->p_pptr->p_sigacts->ps_flag & (PS_NOCLDWAIT | PS_CLDSIGIGN)) {
- struct proc *pp;
-
- mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
- pp = p->p_pptr;
- PROC_UNLOCK(pp);
- proc_reparent(p, initproc);
- p->p_sigparent = SIGCHLD;
- PROC_LOCK(p->p_pptr);
-
+#ifdef PROCDESC
+ if (p->p_procdesc == NULL || procdesc_exit(p)) {
+#endif
/*
- * Notify parent, so in case he was wait(2)ing or
- * executing waitpid(2) with our pid, he will
- * continue.
+ * Notify parent that we're gone. If parent has the
+ * PS_NOCLDWAIT flag set, or if the handler is set to SIG_IGN,
+ * notify process 1 instead (and hope it will handle this
+ * situation).
*/
- wakeup(pp);
+ PROC_LOCK(p->p_pptr);
+ mtx_lock(&p->p_pptr->p_sigacts->ps_mtx);
+ if (p->p_pptr->p_sigacts->ps_flag &
+ (PS_NOCLDWAIT | PS_CLDSIGIGN)) {
+ struct proc *pp;
+
+ mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
+ pp = p->p_pptr;
+ PROC_UNLOCK(pp);
+ proc_reparent(p, initproc);
+ p->p_sigparent = SIGCHLD;
+ PROC_LOCK(p->p_pptr);
+
+ /*
+ * Notify parent, so in case he was wait(2)ing or
+ * executing waitpid(2) with our pid, he will
+ * continue.
+ */
+ wakeup(pp);
+ } else
+ mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
+
+ if (p->p_pptr == initproc)
+ psignal(p->p_pptr, SIGCHLD);
+ else if (p->p_sigparent != 0) {
+ if (p->p_sigparent == SIGCHLD)
+ childproc_exited(p);
+ else /* LINUX thread */
+ psignal(p->p_pptr, p->p_sigparent);
+ }
+#ifdef PROCDESC
} else
- mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
-
- if (p->p_pptr == initproc)
- psignal(p->p_pptr, SIGCHLD);
- else if (p->p_sigparent != 0) {
- if (p->p_sigparent == SIGCHLD)
- childproc_exited(p);
- else /* LINUX thread */
- psignal(p->p_pptr, p->p_sigparent);
- }
+ PROC_LOCK(p->p_pptr);
+#endif
sx_xunlock(&proctree_lock);
/*
@@ -660,7 +678,7 @@ wait4(struct thread *td, struct wait_args *uap)
* rusage. Asserts and will release both the proctree_lock and the process
* lock as part of its work.
*/
-static void
+void
proc_reap(struct thread *td, struct proc *p, int *status, int options,
struct rusage *rusage)
{
@@ -722,6 +740,10 @@ proc_reap(struct thread *td, struct proc *p, int *status, int options,
sx_xunlock(&allproc_lock);
LIST_REMOVE(p, p_sibling);
leavepgrp(p);
+#ifdef PROCDESC
+ if (p->p_procdesc != NULL)
+ procdesc_reap(p);
+#endif
sx_xunlock(&proctree_lock);
/*
diff --git a/sys/kern/kern_fork.c b/sys/kern/kern_fork.c
index 9d3e22d224cf..32d00550a815 100644
--- a/sys/kern/kern_fork.c
+++ b/sys/kern/kern_fork.c
@@ -40,11 +40,13 @@ __FBSDID("$FreeBSD$");
#include "opt_kdtrace.h"
#include "opt_ktrace.h"
#include "opt_kstack_pages.h"
+#include "opt_procdesc.h"
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/sysproto.h>
#include <sys/eventhandler.h>
+#include <sys/fcntl.h>
#include <sys/filedesc.h>
#include <sys/jail.h>
#include <sys/kernel.h>
@@ -55,6 +57,7 @@ __FBSDID("$FreeBSD$");
#include <sys/mutex.h>
#include <sys/priv.h>
#include <sys/proc.h>
+#include <sys/procdesc.h>
#include <sys/pioctl.h>
#include <sys/racct.h>
#include <sys/resourcevar.h>
@@ -104,7 +107,7 @@ fork(struct thread *td, struct fork_args *uap)
int error;
struct proc *p2;
- error = fork1(td, RFFDG | RFPROC, 0, &p2);
+ error = fork1(td, RFFDG | RFPROC, 0, &p2, NULL, 0);
if (error == 0) {
td->td_retval[0] = p2->p_pid;
td->td_retval[1] = 0;
@@ -112,6 +115,34 @@ fork(struct thread *td, struct fork_args *uap)
return (error);
}
+/* ARGUSED */
+int
+pdfork(td, uap)
+ struct thread *td;
+ struct pdfork_args *uap;
+{
+#ifdef PROCDESC
+ int error, fd;
+ struct proc *p2;
+
+ /*
+ * It is necessary to return fd by reference because 0 is a valid file
+ * descriptor number, and the child needs to be able to distinguish
+ * itself from the parent using the return value.
+ */
+ error = fork1(td, RFFDG | RFPROC | RFPROCDESC, 0, &p2,
+ &fd, uap->flags);
+ if (error == 0) {
+ td->td_retval[0] = p2->p_pid;
+ td->td_retval[1] = 0;
+ error = copyout(&fd, uap->fdp, sizeof(fd));
+ }
+ return (error);
+#else
+ return (ENOSYS);
+#endif
+}
+
/* ARGSUSED */
int
vfork(struct thread *td, struct vfork_args *uap)
@@ -124,7 +155,7 @@ vfork(struct thread *td, struct vfork_args *uap)
#else
flags = RFFDG | RFPROC | RFPPWAIT | RFMEM;
#endif
- error = fork1(td, flags, 0, &p2);
+ error = fork1(td, flags, 0, &p2, NULL, 0);
if (error == 0) {
td->td_retval[0] = p2->p_pid;
td->td_retval[1] = 0;
@@ -143,7 +174,7 @@ rfork(struct thread *td, struct rfork_args *uap)
return (EINVAL);
AUDIT_ARG_FFLAGS(uap->flags);
- error = fork1(td, uap->flags, 0, &p2);
+ error = fork1(td, uap->flags, 0, &p2, NULL, 0);
if (error == 0) {
td->td_retval[0] = p2 ? p2->p_pid : 0;
td->td_retval[1] = 0;
@@ -337,7 +368,7 @@ fail:
static void
do_fork(struct thread *td, int flags, struct proc *p2, struct thread *td2,
- struct vmspace *vm2)
+ struct vmspace *vm2, int pdflags)
{
struct proc *p1, *pptr;
int p2_held, trypid;
@@ -625,6 +656,16 @@ do_fork(struct thread *td, int flags, struct proc *p2, struct thread *td2,
p2->p_vmspace->vm_ssize);
}
+#ifdef PROCDESC
+ /*
+ * Associate the process descriptor with the process before anything
+ * can happen that might cause that process to need the descriptor.
+ * However, don't do this until after fork(2) can no longer fail.
+ */
+ if (flags & RFPROCDESC)
+ procdesc_new(p2, pdflags);
+#endif
+
/*
* Both processes are set up, now check if any loadable modules want
* to adjust anything.
@@ -710,7 +751,8 @@ do_fork(struct thread *td, int flags, struct proc *p2, struct thread *td2,
}
int
-fork1(struct thread *td, int flags, int pages, struct proc **procp)
+fork1(struct thread *td, int flags, int pages, struct proc **procp,
+ int *procdescp, int pdflags)
{
struct proc *p1;
struct proc *newproc;
@@ -721,6 +763,9 @@ fork1(struct thread *td, int flags, int pages, struct proc **procp)
int error;
static int curfail;
static struct timeval lastfail;
+#ifdef PROCDESC
+ struct file *fp_procdesc = NULL;
+#endif
/* Check for the undefined or unimplemented flags. */
if ((flags & ~(RFFLAGS | RFTSIGFLAGS(RFTSIGMASK))) != 0)
@@ -738,6 +783,18 @@ fork1(struct thread *td, int flags, int pages, struct proc **procp)
if ((flags & RFTSIGZMB) != 0 && (u_int)RFTSIGNUM(flags) > _SIG_MAXSIG)
return (EINVAL);
+#ifdef PROCDESC
+ if ((flags & RFPROCDESC) != 0) {
+ /* Can't not create a process yet get a process descriptor. */
+ if ((flags & RFPROC) == 0)
+ return (EINVAL);
+
+ /* Must provide a place to put a procdesc if creating one. */
+ if (procdescp == NULL)
+ return (EINVAL);
+ }
+#endif
+
p1 = td->td_proc;
/*
@@ -757,6 +814,25 @@ fork1(struct thread *td, int flags, int pages, struct proc **procp)
return (EAGAIN);
#endif
+#ifdef PROCDESC
+ /*
+ * If required, create a process descriptor in the parent first; we
+ * will abandon it if something goes wrong. We don't finit() until
+ * later.
+ */
+ if (flags & RFPROCDESC) {
+ error = falloc(td, &fp_procdesc, procdescp, 0);
+ if (error != 0) {
+#ifdef RACCT
+ PROC_LOCK(p1);
+ racct_sub(p1, RACCT_NPROC, 1);
+ PROC_UNLOCK(p1);
+#endif
+ return (error);
+ }
+ }
+#endif
+
mem_charged = 0;
vm2 = NULL;
if (pages == 0)
@@ -868,12 +944,16 @@ fork1(struct thread *td, int flags, int pages, struct proc **procp)
PROC_UNLOCK(p1);
}
if (ok) {
- do_fork(td, flags, newproc, td2, vm2);
+ do_fork(td, flags, newproc, td2, vm2, pdflags);
/*
* Return child proc pointer to parent.
*/
*procp = newproc;
+#ifdef PROCDESC
+ if (flags & RFPROCDESC)
+ procdesc_finit(newproc->p_procdesc, fp_procdesc);
+#endif
return (0);
}
@@ -892,6 +972,10 @@ fail1:
if (vm2 != NULL)
vmspace_free(vm2);
uma_zfree(proc_zone, newproc);
+#ifdef PROCDESC
+ if (((flags & RFPROCDESC) != 0) && (fp_procdesc != NULL))
+ fdrop(fp_procdesc, td);
+#endif
pause("fork", hz / 2);
#ifdef RACCT
PROC_LOCK(p1);
diff --git a/sys/kern/kern_kthread.c b/sys/kern/kern_kthread.c
index 95f896fa5aa6..bb1246980d5b 100644
--- a/sys/kern/kern_kthread.c
+++ b/sys/kern/kern_kthread.c
@@ -88,7 +88,7 @@ kproc_create(void (*func)(void *), void *arg,
panic("kproc_create called too soon");
error = fork1(&thread0, RFMEM | RFFDG | RFPROC | RFSTOPPED | flags,
- pages, &p2);
+ pages, &p2, NULL, 0);
if (error)
return error;
diff --git a/sys/kern/kern_sig.c b/sys/kern/kern_sig.c
index e1861eb1e81d..26ef0d7f31d4 100644
--- a/sys/kern/kern_sig.c
+++ b/sys/kern/kern_sig.c
@@ -41,12 +41,14 @@ __FBSDID("$FreeBSD$");
#include "opt_kdtrace.h"
#include "opt_ktrace.h"
#include "opt_core.h"
+#include "opt_procdesc.h"
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/signalvar.h>
#include <sys/vnode.h>
#include <sys/acct.h>
+#include <sys/capability.h>
#include <sys/condvar.h>
#include <sys/event.h>
#include <sys/fcntl.h>
@@ -59,6 +61,7 @@ __FBSDID("$FreeBSD$");
#include <sys/mutex.h>
#include <sys/namei.h>
#include <sys/proc.h>
+#include <sys/procdesc.h>
#include <sys/posix4.h>
#include <sys/pioctl.h>
#include <sys/racct.h>
@@ -1698,6 +1701,34 @@ kill(struct thread *td, struct kill_args *uap)
/* NOTREACHED */
}
+int
+pdkill(td, uap)
+ struct thread *td;
+ struct pdkill_args *uap;
+{
+#ifdef PROCDESC
+ struct proc *p;
+ int error;
+
+ AUDIT_ARG_SIGNUM(uap->signum);
+ AUDIT_ARG_FD(uap->fd);
+ if ((u_int)uap->signum > _SIG_MAXSIG)
+ return (EINVAL);
+
+ error = procdesc_find(td, uap->fd, CAP_PDKILL, &p);
+ if (error)
+ return (error);
+ AUDIT_ARG_PROCESS(p);
+ error = p_cansignal(td, p, uap->signum);
+ if (error == 0 && uap->signum)
+ psignal(p, uap->signum);
+ PROC_UNLOCK(p);
+ return (error);
+#else
+ return (ENOSYS);
+#endif
+}
+
#if defined(COMPAT_43)
#ifndef _SYS_SYSPROTO_H_
struct okillpg_args {
diff --git a/sys/kern/sys_procdesc.c b/sys/kern/sys_procdesc.c
new file mode 100644
index 000000000000..9993732527cd
--- /dev/null
+++ b/sys/kern/sys_procdesc.c
@@ -0,0 +1,524 @@
+/*-
+ * Copyright (c) 2009 Robert N. M. Watson
+ * All rights reserved.
+ *
+ * This software was developed at the University of Cambridge Computer
+ * Laboratory with support from a grant from Google, Inc.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ * notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ * notice, this list of conditions and the following disclaimer in the
+ * documentation and/or other materials provided with the distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
+ * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
+ * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+ * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
+ * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
+ * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
+ * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ */
+
+/*-
+ * FreeBSD process descriptor facility.
+ *
+ * Some processes are represented by a file descriptor, which will be used in
+ * preference to signaling and pids for the purposes of process management,
+ * and is, in effect, a form of capability. When a process descriptor is
+ * used with a process, it ceases to be visible to certain traditional UNIX
+ * process facilities, such as waitpid(2).
+ *
+ * Some semantics:
+ *
+ * - At most one process descriptor will exist for any process, although
+ * references to that descriptor may be held from many processes (or even
+ * be in flight between processes over a local domain socket).
+ * - Last close on the process descriptor will terminate the process using
+ * SIGKILL and reparent it to init so that there's a process to reap it
+ * when it's done exiting.
+ * - If the process exits before the descriptor is closed, it will not
+ * generate SIGCHLD on termination, or be picked up by waitpid().
+ * - The pdkill(2) system call may be used to deliver a signal to the process
+ * using its process descriptor.
+ * - The pdwait4(2) system call may be used to block (or not) on a process
+ * descriptor to collect termination information.
+ *
+ * Open questions:
+ *
+ * - How to handle ptrace(2)?
+ * - Will we want to add a pidtoprocdesc(2) system call to allow process
+ * descriptors to be created for processes without pfork(2)?
+ */
+
+#include <sys/cdefs.h>
+__FBSDID("$FreeBSD$");
+
+#include "opt_procdesc.h"
+
+#include <sys/param.h>
+#include <sys/capability.h>
+#include <sys/fcntl.h>
+#include <sys/file.h>
+#include <sys/filedesc.h>
+#include <sys/kernel.h>
+#include <sys/lock.h>
+#include <sys/mutex.h>
+#include <sys/poll.h>
+#include <sys/proc.h>
+#include <sys/procdesc.h>
+#include <sys/resourcevar.h>
+#include <sys/stat.h>
+#include <sys/sysproto.h>
+#include <sys/sysctl.h>
+#include <sys/systm.h>
+#include <sys/ucred.h>
+
+#include <security/audit/audit.h>
+
+#include <vm/uma.h>
+
+#ifdef PROCDESC
+
+FEATURE(process_descriptors, "Process Descriptors");
+
+static uma_zone_t procdesc_zone;
+
+static fo_rdwr_t procdesc_read;
+static fo_rdwr_t procdesc_write;
+static fo_truncate_t procdesc_truncate;
+static fo_ioctl_t procdesc_ioctl;
+static fo_poll_t procdesc_poll;
+static fo_kqfilter_t procdesc_kqfilter;
+static fo_stat_t procdesc_stat;
+static fo_close_t procdesc_close;
+static fo_chmod_t procdesc_chmod;
+static fo_chown_t procdesc_chown;
+
+static struct fileops procdesc_ops = {
+ .fo_read = procdesc_read,
+ .fo_write = procdesc_write,
+ .fo_truncate = procdesc_truncate,
+ .fo_ioctl = procdesc_ioctl,
+ .fo_poll = procdesc_poll,
+ .fo_kqfilter = procdesc_kqfilter,
+ .fo_stat = procdesc_stat,
+ .fo_close = procdesc_close,
+ .fo_chmod = procdesc_chmod,
+ .fo_chown = procdesc_chown,
+ .fo_flags = DFLAG_PASSABLE,
+};
+
+/*
+ * Initialize with VFS so that process descriptors are available along with
+ * other file descriptor types. As long as it runs before init(8) starts,
+ * there shouldn't be a problem.
+ */
+static void
+procdesc_init(void *dummy __unused)
+{
+
+ procdesc_zone = uma_zcreate("procdesc", sizeof(struct procdesc),
+ NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
+ if (procdesc_zone == NULL)
+ panic("procdesc_init: procdesc_zone not initialized");
+}
+SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_ANY, procdesc_init, NULL);
+
+/*
+ * Return a locked process given a process descriptor, or ESRCH if it has
+ * died.
+ */
+int
+procdesc_find(struct thread *td, int fd, cap_rights_t rights,
+ struct proc **p)
+{
+ struct procdesc *pd;
+ struct file *fp;
+ int error;
+
+ error = fget(td, fd, rights, &fp);
+ if (error)
+ return (error);
+ if (fp->f_type != DTYPE_PROCDESC) {
+ error = EBADF;
+ goto out;
+ }
+ pd = fp->f_data;
+ sx_slock(&proctree_lock);
+ if (pd->pd_proc != NULL) {
+ *p = pd->pd_proc;
+ PROC_LOCK(*p);
+ } else
+ error = ESRCH;
+ sx_sunlock(&proctree_lock);
+out:
+ fdrop(fp, td);
+ return (error);
+}
+
+/*
+ * Function to be used by procstat(1) sysctls when returning procdesc
+ * information.
+ */
+pid_t
+procdesc_pid(struct file *fp_procdesc)
+{
+ struct procdesc *pd;
+
+ KASSERT(fp_procdesc->f_type == DTYPE_PROCDESC,
+ ("procdesc_pid: !procdesc"));
+
+ pd = fp_procdesc->f_data;
+ return (pd->pd_pid);
+}
+
+/*
+ * Retrieve the PID associated with a process descriptor.
+ */
+int
+kern_pdgetpid(struct thread *td, int fd, cap_rights_t rights, pid_t *pidp)
+{
+ struct file *fp;
+ int error;
+
+ error = fget(td, fd, rights, &fp);
+ if (error)
+ return (error);
+ if (fp->f_type != DTYPE_PROCDESC) {
+ error = EBADF;
+ goto out;
+ }
+ *pidp = procdesc_pid(fp);
+out:
+ fdrop(fp, td);
+ return (error);
+}
+
+/*
+ * System call to return the pid of a process given its process descriptor.
+ */
+int
+pdgetpid(struct thread *td, struct pdgetpid_args *uap)
+{
+ pid_t pid;
+ int error;
+
+ AUDIT_ARG_FD(uap->fd);
+ error = kern_pdgetpid(td, uap->fd, CAP_PDGETPID, &pid);
+ if (error == 0)
+ error = copyout(&pid, uap->pidp, sizeof(pid));
+ return (error);
+}
+
+/*
+ * When a new process is forked by pdfork(), a file descriptor is allocated
+ * by the fork code first, then the process is forked, and then we get a
+ * chance to set up the process descriptor. Failure is not permitted at this
+ * point, so procdesc_new() must succeed.
+ */
+void
+procdesc_new(struct proc *p, int flags)
+{
+ struct procdesc *pd;
+
+ pd = uma_zalloc(procdesc_zone, M_WAITOK | M_ZERO);
+ pd->pd_proc = p;
+ pd->pd_pid = p->p_pid;
+ p->p_procdesc = pd;
+ pd->pd_flags = 0;
+ if (flags & PD_DAEMON)
+ pd->pd_flags |= PDF_DAEMON;
+ PROCDESC_LOCK_INIT(pd);
+
+ /*
+ * Process descriptors start out with two references: one from their
+ * struct file, and the other from their struct proc.
+ */
+ refcount_init(&pd->pd_refcount, 2);
+}
+
+/*
+ * Initialize a file with a process descriptor.
+ */
+void
+procdesc_finit(struct procdesc *pdp, struct file *fp)
+{
+
+ finit(fp, FREAD | FWRITE, DTYPE_PROCDESC, pdp, &procdesc_ops);
+}
+
+static void
+procdesc_free(struct procdesc *pd)
+{
+
+ /*
+ * When the last reference is released, we assert that the descriptor
+ * has been closed, but not that the process has exited, as we will
+ * detach the descriptor before the process dies if the descript is
+ * closed, as we can't wait synchronously.
+ */
+ if (refcount_release(&pd->pd_refcount)) {
+ KASSERT(pd->pd_proc == NULL,
+ ("procdesc_free: pd_proc != NULL"));
+ KASSERT((pd->pd_flags & PDF_CLOSED),
+ ("procdesc_free: !PDF_CLOSED"));
+
+ PROCDESC_LOCK_DESTROY(pd);
+ uma_zfree(procdesc_zone, pd);
+ }
+}
+
+/*
+ * procdesc_exit() - notify a process descriptor that its process is exiting.
+ * We use the proctree_lock to ensure that process exit either happens
+ * strictly before or strictly after a concurrent call to procdesc_close().
+ */
+int
+procdesc_exit(struct proc *p)
+{
+ struct procdesc *pd;
+
+ sx_assert(&proctree_lock, SA_XLOCKED);
+ PROC_LOCK_ASSERT(p, MA_OWNED);
+ KASSERT(p->p_procdesc != NULL, ("procdesc_exit: p_procdesc NULL"));
+
+ pd = p->p_procdesc;
+
+ PROCDESC_LOCK(pd);
+ KASSERT((pd->pd_flags & PDF_CLOSED) == 0 || p->p_pptr == initproc,
+ ("procdesc_exit: closed && parent not init"));
+
+ pd->pd_flags |= PDF_EXITED;
+
+ /*
+ * If the process descriptor has been closed, then we have nothing
+ * to do; return 1 so that init will get SIGCHLD and do the reaping.
+ * Clean up the procdesc now rather than letting it happen during
+ * that reap.
+ */
+ if (pd->pd_flags & PDF_CLOSED) {
+ PROCDESC_UNLOCK(pd);
+ pd->pd_proc = NULL;
+ p->p_procdesc = NULL;
+ procdesc_free(pd);
+ return (1);
+ }
+ if (pd->pd_flags & PDF_SELECTED) {
+ pd->pd_flags &= ~PDF_SELECTED;
+ selwakeup(&pd->pd_selinfo);
+ }
+ PROCDESC_UNLOCK(pd);
+ return (0);
+}
+
+/*
+ * When a process descriptor is reaped, perhaps as a result of close() or
+ * pdwait4(), release the process's reference on the process descriptor.
+ */
+void
+procdesc_reap(struct proc *p)
+{
+ struct procdesc *pd;
+
+ sx_assert(&proctree_lock, SA_XLOCKED);
+ KASSERT(p->p_procdesc != NULL, ("procdesc_reap: p_procdesc == NULL"));
+
+ pd = p->p_procdesc;
+ pd->pd_proc = NULL;
+ procdesc_free(pd);
+}
+
+/*
+ * procdesc_close() - last close on a process descriptor. If the process is
+ * still running, terminate with SIGKILL (unless PD_DAEMON is set) and let
+ * init(8) clean up the mess; if not, we have to clean up the zombie ourselves.
+ */
+static int
+procdesc_close(struct file *fp, struct thread *td)
+{
+ struct procdesc *pd;
+ struct proc *p;
+
+ KASSERT(fp->f_type == DTYPE_PROCDESC, ("procdesc_close: !procdesc"));
+
+ pd = fp->f_data;
+ fp->f_ops = &badfileops;
+ fp->f_data = NULL;
+
+ sx_xlock(&proctree_lock);
+ PROCDESC_LOCK(pd);
+ pd->pd_flags |= PDF_CLOSED;
+ PROCDESC_UNLOCK(pd);
+ p = pd->pd_proc;
+ PROC_LOCK(p);
+ if (p->p_state == PRS_ZOMBIE) {
+ /*
+ * If the process is already dead and just awaiting reaping,
+ * do that now. This will release the process's reference to
+ * the process descriptor when it calls back into
+ * procdesc_reap().
+ */
+ PROC_SLOCK(p);
+ proc_reap(curthread, p, NULL, 0, NULL);
+ } else {
+ /*
+ * If the process is not yet dead, we need to kill it, but we
+ * can't wait around synchronously for it to go away, as that
+ * path leads to madness (and deadlocks). First, detach the
+ * process from its descriptor so that its exit status will
+ * be reported normally.
+ */
+ pd->pd_proc = NULL;
+ p->p_procdesc = NULL;
+ procdesc_free(pd);
+
+ /*
+ * Next, reparent it to init(8) so that there's someone to
+ * pick up the pieces; finally, terminate with prejudice.
+ */
+ p->p_sigparent = SIGCHLD;
+ proc_reparent(p, initproc);
+ if ((pd->pd_flags & PD_DAEMON) == 0)
+ psignal(p, SIGKILL);
+ PROC_UNLOCK(p);
+ sx_xunlock(&proctree_lock);
+ }
+
+ /*
+ * Release the file descriptor's reference on the process descriptor.
+ */
+ procdesc_free(pd);
+ return (0);
+}
+
+static int
+procdesc_read(struct file *fp, struct uio *uio, struct ucred *active_cred,
+ int flags, struct thread *td)
+{
+
+ return (EOPNOTSUPP);
+}
+
+static int
+procdesc_write(struct file *fp, struct uio *uio, struct ucred *active_cred,
+ int flags, struct thread *td)
+{
+
+ return (EOPNOTSUPP);
+}
+
+static int
+procdesc_truncate(struct file *fp, off_t length, struct ucred *active_cred,
+ struct thread *td)
+{
+
+ return (EOPNOTSUPP);
+}
+
+static int
+procdesc_ioctl(struct file *fp, u_long com, void *data,
+ struct ucred *active_cred, struct thread *td)
+{
+
+ return (EOPNOTSUPP);
+}
+
+static int
+procdesc_poll(struct file *fp, int events, struct ucred *active_cred,
+ struct thread *td)
+{
+ struct procdesc *pd;
+ int revents;
+
+ revents = 0;
+ pd = fp->f_data;
+ PROCDESC_LOCK(pd);
+ if (pd->pd_flags & PDF_EXITED)
+ revents |= POLLHUP;
+ if (revents == 0) {
+ selrecord(td, &pd->pd_selinfo);
+ pd->pd_flags |= PDF_SELECTED;
+ }
+ PROCDESC_UNLOCK(pd);
+ return (revents);
+}
+
+static int
+procdesc_kqfilter(struct file *fp, struct knote *kn)
+{
+
+ return (EOPNOTSUPP);
+}
+
+static int
+procdesc_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
+ struct thread *td)
+{
+ struct procdesc *pd;
+ struct timeval pstart;
+
+ /*
+ * XXXRW: Perhaps we should cache some more information from the
+ * process so that we can return it reliably here even after it has
+ * died. For example, caching its credential data.
+ */
+ bzero(sb, sizeof(*sb));
+ pd = fp->f_data;
+ sx_slock(&proctree_lock);
+ if (pd->pd_proc != NULL) {
+ PROC_LOCK(pd->pd_proc);
+
+ /* Set birth and [acm] times to process start time. */
+ pstart = pd->pd_proc->p_stats->p_start;
+ timevaladd(&pstart, &boottime);
+ TIMEVAL_TO_TIMESPEC(&pstart, &sb->st_birthtim);
+ sb->st_atim = sb->st_birthtim;
+ sb->st_ctim = sb->st_birthtim;
+ sb->st_mtim = sb->st_birthtim;
+ if (pd->pd_proc->p_state != PRS_ZOMBIE)
+ sb->st_mode = S_IFREG | S_IRWXU;
+ else
+ sb->st_mode = S_IFREG;
+ sb->st_uid = pd->pd_proc->p_ucred->cr_ruid;
+ sb->st_gid = pd->pd_proc->p_ucred->cr_rgid;
+ PROC_UNLOCK(pd->pd_proc);
+ } else
+ sb->st_mode = S_IFREG;
+ sx_sunlock(&proctree_lock);
+ return (0);
+}
+
+static int
+procdesc_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
+ struct thread *td)
+{
+
+ return (EOPNOTSUPP);
+}
+
+static int
+procdesc_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
+ struct thread *td)
+{
+
+ return (EOPNOTSUPP);
+}
+
+#else /* !PROCDESC */
+
+int
+pdgetpid(struct thread *td, struct pdgetpid_args *uap)
+{
+
+ return (ENOSYS);
+}
+
+#endif /* PROCDESC */
diff --git a/sys/kern/syscalls.master b/sys/kern/syscalls.master
index 0b249a5b55ad..b79c6c7109cc 100644
--- a/sys/kern/syscalls.master
+++ b/sys/kern/syscalls.master
@@ -919,10 +919,10 @@
u_int64_t *rightsp); }
516 AUE_CAP_ENTER STD { int cap_enter(void); }
517 AUE_CAP_GETMODE STD { int cap_getmode(u_int *modep); }
-518 AUE_PDFORK UNIMPL pdfork
-519 AUE_PDKILL UNIMPL pdkill
-520 AUE_PDGETPID UNIMPL pdgetpid
-521 AUE_PDWAIT UNIMPL pdwait
+518 AUE_PDFORK STD { int pdfork(int *fdp, int flags); }
+519 AUE_PDKILL STD { int pdkill(int fd, int signum); }
+520 AUE_PDGETPID STD { int pdgetpid(int fd, pid_t *pidp); }
+521 AUE_PDWAIT UNIMPL pdwait4
522 AUE_SELECT STD { int pselect(int nd, fd_set *in, \
fd_set *ou, fd_set *ex, \
const struct timespec *ts, \
diff --git a/sys/sys/capability.h b/sys/sys/capability.h
index d67dc179b0de..81446a281904 100644
--- a/sys/sys/capability.h
+++ b/sys/sys/capability.h
@@ -131,8 +131,13 @@
#define CAP_IOCTL 0x0004000000000000ULL
#define CAP_TTYHOOK 0x0008000000000000ULL
+/* Process management via process descriptors. */
+#define CAP_PDGETPID 0x0010000000000000ULL
+#define CAP_PDWAIT 0x0020000000000000ULL
+#define CAP_PDKILL 0x0040000000000000ULL
+
/* The mask of all valid method rights. */
-#define CAP_MASK_VALID 0x000fffffffffffffULL
+#define CAP_MASK_VALID 0x007fffffffffffffULL
#ifdef _KERNEL
diff --git a/sys/sys/file.h b/sys/sys/file.h
index 5a4af332ebb9..57e7047e8b32 100644
--- a/sys/sys/file.h
+++ b/sys/sys/file.h
@@ -65,6 +65,7 @@ struct socket;
#define DTYPE_PTS 10 /* pseudo teletype master device */
#define DTYPE_DEV 11 /* Device specific fd type */
#define DTYPE_CAPABILITY 12 /* capability */
+#define DTYPE_PROCDESC 13 /* process descriptor */
#ifdef _KERNEL
diff --git a/sys/sys/proc.h b/sys/sys/proc.h
index 233efe985640..67adbe5b5899 100644
--- a/sys/sys/proc.h
+++ b/sys/sys/proc.h
@@ -166,6 +166,7 @@ struct mqueue_notifier;
struct nlminfo;
struct p_sched;
struct proc;
+struct procdesc;
struct racct;
struct sleepqueue;
struct td_sched;
@@ -534,6 +535,7 @@ struct proc {
int p_boundary_count;/* (c) Num threads at user boundary */
int p_pendingcnt; /* how many signals are pending */
struct itimers *p_itimers; /* (c) POSIX interval timers. */
+ struct procdesc *p_procdesc; /* (e) Process descriptor, if any. */
/* End area that is zeroed on creation. */
#define p_endzero p_magic
@@ -822,7 +824,7 @@ int enterpgrp(struct proc *p, pid_t pgid, struct pgrp *pgrp,
int enterthispgrp(struct proc *p, struct pgrp *pgrp);
void faultin(struct proc *p);
void fixjobc(struct proc *p, struct pgrp *pgrp, int entering);
-int fork1(struct thread *, int, int, struct proc **);
+int fork1(struct thread *, int, int, struct proc **, int *, int);
void fork_exit(void (*)(void *, struct trapframe *), void *,
struct trapframe *);
void fork_return(struct thread *, struct trapframe *);
@@ -844,6 +846,8 @@ void pargs_hold(struct pargs *pa);
void procinit(void);
void proc_linkup0(struct proc *p, struct thread *td);
void proc_linkup(struct proc *p, struct thread *td);
+void proc_reap(struct thread *td, struct proc *p, int *status, int options,
+ struct rusage *rusage);
void proc_reparent(struct proc *child, struct proc *newparent);
struct pstats *pstats_alloc(void);
void pstats_fork(struct pstats *src, struct pstats *dst);
diff --git a/sys/sys/procdesc.h b/sys/sys/procdesc.h
new file mode 100644
index 000000000000..cc8b7166f639
--- /dev/null
+++ b/sys/sys/procdesc.h
@@ -0,0 +1,119 @@
+/*-
+ * Copyright (c) 2009 Robert N. M. Watson
+ * All rights reserved.
+ *
+ * This software was developed at the University of Cambridge Computer
+ * Laboratory with support from a grant from Google, Inc.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ * notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ * notice, this list of conditions and the following disclaimer in the
+ * documentation and/or other materials provided with the distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
+ * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
+ * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+ * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
+ * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
+ * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
+ * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ *
+ * $FreeBSD$
+ */
+
+#ifndef _SYS_PROCDESC_H_
+#define _SYS_PROCDESC_H_
+
+#ifdef _KERNEL
+#include <sys/selinfo.h> /* struct selinfo */
+#include <sys/_lock.h>
+#include <sys/_mutex.h>
+
+/*-
+ * struct procdesc describes a process descriptor, and essentially consists
+ * of two pointers -- one to the file descriptor, and one to the process.
+ * When both become NULL, the process descriptor will be freed. An important
+ * invariant is that there is only ever one process descriptor for a process,
+ * so a single file pointer will suffice.
+ *
+ * Locking key:
+ * (c) - Constant after initial setup.
+ * (p) - Protected by the process descriptor mutex.
+ * (r) - Atomic eference count.
+ * (s) - Protected by selinfo.
+ * (t) - Protected by the proctree_lock
+ */
+struct proc;
+struct sigio;
+struct procdesc {
+ /*
+ * Basic process descriptor state: the process, a cache of its pid to
+ * satisfy queries after the process exits, and process descriptor
+ * refcount.
+ */
+ struct proc *pd_proc; /* (t) Process. */
+ pid_t pd_pid; /* (c) Cached pid. */
+ u_int pd_refcount; /* (r) Reference count. */
+
+ /*
+ * In-flight data and notification of events.
+ */
+ int pd_flags; /* (p) PD_ flags. */
+ struct selinfo pd_selinfo; /* (p) Event notification. */
+ struct mtx pd_lock; /* Protect data + events. */
+};
+
+/*
+ * Locking macros for the procdesc itself.
+ */
+#define PROCDESC_LOCK_DESTROY(pd) mtx_destroy(&(pd)->pd_lock)
+#define PROCDESC_LOCK_INIT(pd) mtx_init(&(pd)->pd_lock, "procdesc", NULL, \
+ MTX_DEF)
+#define PROCDESC_LOCK(pd) mtx_lock(&(pd)->pd_lock)
+#define PROCDESC_UNLOCK(pd) mtx_unlock(&(pd)->pd_lock)
+
+/*
+ * Flags for the pd_flags field.
+ */
+#define PDF_CLOSED 0x00000001 /* Descriptor has closed. */
+#define PDF_SELECTED 0x00000002 /* Issue selwakeup(). */
+#define PDF_EXITED 0x00000004 /* Process exited. */
+#define PDF_DAEMON 0x00000008 /* Don't exit when procdesc closes. */
+
+/*
+ * In-kernel interfaces to process descriptors.
+ */
+int procdesc_exit(struct proc *);
+int procdesc_find(struct thread *, int fd, cap_rights_t, struct proc **);
+int kern_pdgetpid(struct thread *, int fd, cap_rights_t, pid_t *pidp);
+void procdesc_new(struct proc *, int);
+void procdesc_finit(struct procdesc *, struct file *);
+pid_t procdesc_pid(struct file *);
+void procdesc_reap(struct proc *);
+
+#else /* !_KERNEL */
+
+/*
+ * Process descriptor system calls.
+ */
+struct rusage;
+int pdfork(int *, int);
+int pdkill(int, int);
+int pdgetpid(int, pid_t *);
+
+#endif /* _KERNEL */
+
+/*
+ * Flags which can be passed to pdfork(2).
+ */
+#define PD_DAEMON 0x00000001 /* Don't exit when procdesc closes. */
+
+#endif /* !_SYS_PROCDESC_H_ */
diff --git a/sys/sys/unistd.h b/sys/sys/unistd.h
index 9d56a3a42730..9e7f7e6d8342 100644
--- a/sys/sys/unistd.h
+++ b/sys/sys/unistd.h
@@ -185,11 +185,12 @@
#define RFTSIGMASK 0xFF
#define RFTSIGNUM(flags) (((flags) >> RFTSIGSHIFT) & RFTSIGMASK)
#define RFTSIGFLAGS(signum) ((signum) << RFTSIGSHIFT)
+#define RFPROCDESC (1<<28) /* return a process descriptor */
#define RFPPWAIT (1<<31) /* parent sleeps until child exits (vfork) */
-#define RFKERNELONLY (RFSTOPPED | RFHIGHPID | RFPPWAIT)
#define RFFLAGS (RFFDG | RFPROC | RFMEM | RFNOWAIT | RFCFDG | \
RFTHREAD | RFSIGSHARE | RFLINUXTHPN | RFSTOPPED | RFHIGHPID | RFTSIGZMB | \
- RFPPWAIT)
+ RFPROCDESC | RFPPWAIT)
+#define RFKERNELONLY (RFSTOPPED | RFHIGHPID | RFPPWAIT | RFPROCDESC)
#endif /* __BSD_VISIBLE */
diff --git a/sys/sys/user.h b/sys/sys/user.h
index ecf4ea940138..a139d4fdfabe 100644
--- a/sys/sys/user.h
+++ b/sys/sys/user.h
@@ -252,6 +252,7 @@ struct user {
#define KF_TYPE_SEM 9
#define KF_TYPE_PTS 10
/* no KF_TYPE_CAPABILITY (11), since capabilities wrap other file objects */
+#define KF_TYPE_PROCDESC 12
#define KF_TYPE_UNKNOWN 255
#define KF_VTYPE_VNON 0
@@ -377,6 +378,9 @@ struct kinfo_file {
/* Round to 64 bit alignment. */
uint32_t kf_pts_pad0[7];
} kf_pts;
+ struct {
+ pid_t kf_pid;
+ } kf_proc;
} kf_un;
uint16_t kf_status; /* Status flags. */
uint16_t kf_pad1; /* Round to 32 bit alignment. */