Skip to contentSkip to conversation

add rt_tgsigqueueinfo syscall [RESEND]

PATCH In review Thomas Gleixner 3 participants 12 messages Feb 26, 2009 – Mar 24, 2009

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

cover add rt_tgsigqueueinfo syscall [RESEND]

sys_kill has a counterpart sys_tgkill which allows to send signals to a particular thread. sys_rt_sigqueueinfo is lacking such a counterpart.

Aside of the asymetry it is a show stopper for migrating applications from other unix-alike RTOSes.

The following patch series implements rt_tgsigqueueinfo and hooks it up for x86.

Find below the raw documentation.

Thanks,

tglx
----

NAME
rt_tgsigqueueinfo - Send signal information to a signal to a thread

SYNOPSIS

       long sys_rt_tgsigqueueinfo (int tgid, int tid, int sig, siginfo_t *uinfo);

DESCRIPTION

rt_tgsigqueueinfo sends signal sig information uinfo to the thread with the thread ID tid in the thread group tgid. (By contrast, rt_sigqueueinfo(2) can only be used to send a signal info to a process (i.e., thread group) as a whole, and the signal will be delivered to an arbitrary thread within that process.)

RETURN VALUE

rt_tgsigqueueinfo returns 0 on success; otherwise, rt_sigqueueinfo returns one of the errors listed in the "Errors" section.

ERRORS
-EFAULT
An invalid value for uinfo was specified.

-EINVAL
An invalid TID, TGID or signal was specified.

-EPERM
Permission denied. For the required permissions, see rt_sigqueueinfo(2).

-ESRCH
No process with the specified thread ID and thread group ID exists.

#1 lore raw

PATCH 1/3 signals: split do_tkill

Split out the code from do_tkill to make it reusable by the follow up patch which implements sys_rt_tgsigqueueinfo

---
 kernel/signal.c |   29 ++++++++++++++++++-----------
 1 file changed, 18 insertions(+), 11 deletions(-)
kernel/signal.c +18 -11
Index: linux-2.6-tip/kernel/signal.c
===================================================================
@@ -2235,24 +2235,20 @@ SYSCALL_DEFINE2(kill, pid_t, pid, int, s
 	return kill_something_info(sig, &info, pid);
 }
 
-static int do_tkill(pid_t tgid, pid_t pid, int sig)
+static int
+do_send_specific(pid_t tgid, pid_t pid, int sig, struct siginfo *info)
 {
-	int error;
-	struct siginfo info;
 	struct task_struct *p;
 	unsigned long flags;
+	int error = -ESRCH;
 
-	error = -ESRCH;
-	info.si_signo = sig;
-	info.si_errno = 0;
-	info.si_code = SI_TKILL;
-	info.si_pid = task_tgid_vnr(current);
-	info.si_uid = current_uid();
+	info->si_pid = task_tgid_vnr(current);
+	info->si_uid = current_uid();
 
 	rcu_read_lock();
 	p = find_task_by_vpid(pid);
 	if (p && (tgid <= 0 || task_tgid_vnr(p) == tgid)) {
-		error = check_kill_permission(sig, &info, p);
+		error = check_kill_permission(sig, info, p);
 		/*
 		 * The null signal is a permissions and process existence
 		 * probe.  No signal is actually delivered.
@@ -2262,7 +2258,7 @@ static int do_tkill(pid_t tgid, pid_t pi
 		 * signal is private anyway.
 		 */
 		if (!error && sig && lock_task_sighand(p, &flags)) {
-			error = specific_send_sig_info(sig, &info, p);
+			error = specific_send_sig_info(sig, info, p);
 			unlock_task_sighand(p, &flags);
 		}
 	}
@@ -2271,6 +2267,17 @@ static int do_tkill(pid_t tgid, pid_t pi
 	return error;
 }
 
+static int do_tkill(pid_t tgid, pid_t pid, int sig)
+{
+	struct siginfo info;
+
+	info.si_signo = sig;
+	info.si_errno = 0;
+	info.si_code = SI_TKILL;
+
+	return do_send_specific(tgid, pid, sig, &info);
+}
+
 /**
  *  sys_tgkill - send signal to one specific thread
  *  @tgid: the thread group ID of the thread
#2 lore raw

PATCH 2/3 signals: implement sys_rt_tgsigqueueinfo

In reply to Thomas Gleixner

sys_kill has the per thread counterpart sys_tgkill. sigqueueinfo is missing a thread directed counterpart. Such an interface is important for migrating applications from other OSes which have the per thread delivery implemented.

---
 include/linux/compat.h |    2 ++
 include/linux/signal.h |    2 ++
 kernel/compat.c        |   11 +++++++++++
 kernel/signal.c        |   26 ++++++++++++++++++++++++++
 4 files changed, 41 insertions(+)
include/linux/compat.h +2 -0
Index: linux-2.6-tip/include/linux/compat.h
===================================================================
@@ -208,6 +208,8 @@ int copy_siginfo_from_user32(siginfo_t *
 int copy_siginfo_to_user32(struct compat_siginfo __user *to, siginfo_t *from);
 int get_compat_sigevent(struct sigevent *event,
 		const struct compat_sigevent __user *u_event);
+long compat_sys_rt_tgsigqueueinfo(compat_pid_t tgid, compat_pid_t pid, int sig,
+				  struct compat_siginfo __user *uinfo);
 
 static inline int compat_timeval_compare(struct compat_timeval *lhs,
 					struct compat_timeval *rhs)
include/linux/signal.h +2 -0
Index: linux-2.6-tip/include/linux/signal.h
===================================================================
@@ -235,6 +235,8 @@ static inline int valid_signal(unsigned 
 extern int next_signal(struct sigpending *pending, sigset_t *mask);
 extern int group_send_sig_info(int sig, struct siginfo *info, struct task_struct *p);
 extern int __group_send_sig_info(int, struct siginfo *, struct task_struct *);
+extern long do_rt_tgsigqueueinfo(pid_t tgid, pid_t pid, int sig,
+				 siginfo_t *info);
 extern long do_sigpending(void __user *, unsigned long);
 extern int sigprocmask(int, sigset_t *, sigset_t *);
 extern int show_unhandled_signals;
kernel/compat.c +11 -0
Index: linux-2.6-tip/kernel/compat.c
===================================================================
@@ -882,6 +882,17 @@ compat_sys_rt_sigtimedwait (compat_sigse
 
 }
 
+asmlinkage long
+compat_sys_rt_tgsigqueueinfo(compat_pid_t tgid, compat_pid_t pid, int sig,
+			     struct compat_siginfo __user *uinfo)
+{
+	siginfo_t info;
+
+	if (copy_siginfo_from_user32(&info, uinfo))
+		return -EFAULT;
+	return do_rt_tgsigqueueinfo(tgid, pid, sig, &info);
+}
+
 #ifdef __ARCH_WANT_COMPAT_SYS_TIME
 
 /* compat_time_t is a 32 bit "long" and needs to get converted. */
kernel/signal.c +26 -0
Index: linux-2.6-tip/kernel/signal.c
===================================================================
@@ -2327,6 +2327,32 @@ SYSCALL_DEFINE3(rt_sigqueueinfo, pid_t, 
 	return kill_proc_info(sig, &info, pid);
 }
 
+long do_rt_tgsigqueueinfo(pid_t tgid, pid_t pid, int sig, siginfo_t *info)
+{
+	/* This is only valid for single tasks */
+	if (pid <= 0 || tgid <= 0)
+		return -EINVAL;
+
+	/* Not even root can pretend to send signals from the kernel.
+	   Nor can they impersonate a kill(), which adds source info.  */
+	if (info->si_code >= 0)
+		return -EPERM;
+	info->si_signo = sig;
+
+	return do_send_specific(tgid, pid, sig, info);
+}
+
+SYSCALL_DEFINE4(rt_tgsigqueueinfo, pid_t, tgid, pid_t, pid, int, sig,
+		siginfo_t __user *, uinfo)
+{
+	siginfo_t info;
+
+	if (copy_from_user(&info, uinfo, sizeof(siginfo_t)))
+		return -EFAULT;
+
+	return do_rt_tgsigqueueinfo(tgid, pid, sig, &info);
+}
+
 int do_sigaction(int sig, struct k_sigaction *act, struct k_sigaction *oact)
 {
 	struct task_struct *t = current;
#3 lore raw

PATCH 3/3 x86: hookup sys_rt_tgsigqueueinfo

In reply to Thomas Gleixner

Make the new sys_rt_tgsigqueueinfo available for x86.

---
 arch/x86/ia32/ia32entry.S          |    1 +
 arch/x86/include/asm/unistd_32.h   |    1 +
 arch/x86/include/asm/unistd_64.h   |    2 ++
 arch/x86/kernel/syscall_table_32.S |    1 +
 4 files changed, 5 insertions(+)
arch/x86/ia32/ia32entry.S +1 -0
Index: linux-2.6-tip/arch/x86/ia32/ia32entry.S
===================================================================
@@ -828,4 +828,5 @@ ia32_sys_call_table:
 	.quad sys_dup3			/* 330 */
 	.quad sys_pipe2
 	.quad sys_inotify_init1
+	.quad compat_sys_rt_tgsigqueueinfo
 ia32_syscall_end:
arch/x86/include/asm/unistd_32.h +1 -0
Index: linux-2.6-tip/arch/x86/include/asm/unistd_32.h
===================================================================
@@ -338,6 +338,7 @@
 #define __NR_dup3		330
 #define __NR_pipe2		331
 #define __NR_inotify_init1	332
+#define __NR_rt_tgsigqueueinfo	333
 
 #ifdef __KERNEL__
 
arch/x86/include/asm/unistd_64.h +2 -0
Index: linux-2.6-tip/arch/x86/include/asm/unistd_64.h
===================================================================
@@ -653,6 +653,8 @@ __SYSCALL(__NR_dup3, sys_dup3)
 __SYSCALL(__NR_pipe2, sys_pipe2)
 #define __NR_inotify_init1			294
 __SYSCALL(__NR_inotify_init1, sys_inotify_init1)
+#define __NR_rt_tgsigqueueinfo			295
+__SYSCALL(__NR_rt_tgsigqueueinfo, sys_rt_tgsigqueueinfo)
 
 
 #ifndef __NO_STUBS
arch/x86/kernel/syscall_table_32.S +1 -0
Index: linux-2.6-tip/arch/x86/kernel/syscall_table_32.S
===================================================================
@@ -332,3 +332,4 @@ ENTRY(sys_call_table)
 	.long sys_dup3			/* 330 */
 	.long sys_pipe2
 	.long sys_inotify_init1
+	.long sys_rt_tgsigqueueinfo
#4 lore raw
In reply to Thomas Gleixner

Thomas,

Please CC linux-api@ (http://thread.gmane.org/gmane.linux.ltp/5658/)
on patches that change the k-u interface...

Further comments below.

Quoted 10 lines · Thomas Gleixner

sys_kill has a counterpart sys_tgkill which allows to send signals to a particular thread. sys_rt_sigqueueinfo is lacking such a counterpart.

Aside of the asymetry it is a show stopper for migrating applications from other unix-alike RTOSes.

The following patch series implements rt_tgsigqueueinfo and hooks it up for x86.

Find below the raw documentation.

I haven't looked at this iteration of the patch, to check if it is different from the version you posted a few months ago. I assume it isn't different, since you didn't mention any differences. In that case, there are still the questions I raised back then, so I'll just repeat them now.

Back in October, I did some testing of this interface. Two potential issues that I saw, both of which relate to inconsistencies with rt_sigqueueinfo():

1) With rt_siqueueinfo(), we can get the PID (TGID) of the sender, which enables the receiver of the signal to know who the sender was, and perhaps use that information to (for example) send a signal in the other direction.

With rt_tgsigqueueinfo(), we don't quite have that ability: all that the receiver gets is the TGID of the sender, not the TID. This means that we can't (for example) send a signal back to the precise thread that signaled us. I'm not sure if this matters or not (but perhaps it might when sender and receiver are in the same process?). I'm also not sure whether we want to do anything about this (i.e., extending siginfo_t to include a si_tid field), but I wanted to point out this assymetry w.r.t. to rt_sigqueueinfo(), in case you had not considered it.

2) With rt_sigqueueinfo(), we can specify the si_pid and and si_uid fields that should be sent to the receiver. This is not possible with rt_tgsigqueueinfo(), which always supplied the caller';s PID and UID in the si_pid and si_uid fields sent to the receiver. See the following, created using my test programs below (the 111 & 222 arguments to t_*sigqueueinfo set the si_pid and si_uid fields in the siginfo_t given to the *sigqueueinfo() syscall):

$ ./multithread_sig_receiver 0
Main: TGID = 11711; TID = 11711
Main: blocked signals
Thread 1: PID = 11711; TID = 11712
Thread 1: about to wait for signals
Thread 0: about to wait for signals
                        $ ./t_rt_sigqueueinfo 11711 44 1 -1 111 222
                        My PID is 11714
Thread 0: got signal: si_signo=44 si_code=-1 si_pid=111 si_uid=222
si_value.sival_int=1
                        $ ./t_rt_tgsigqueueinfo 11711 11712 44 1 -1 111 222
                        My PID is 11715
Thread 1: got signal: si_signo=44 si_code=-1 si_pid=11715 si_uid=1000
si_value.sival_int=1

The problem is the different code paths taken by rt_tgsigqueueinfo()/tgkill() versus rt_sigqueueinfo():

sys_rt_sigqueueinfo() sys_tgkill() sys_rt_sigqueueinfo()

        |                   |                              |
        v                   v                              v
do_rt_sigqueueinfo()   do_tkill()                  kill_proc_info()
        |                   |                              |
        |                   v                              v
        +------------> do_send_specific()          kill_pid_info()
                            |                              |
                            v                              v
                       specific_send_sig_info()    group_send_sig_info()
                            |                              |
                            |                              v
                            |                      __group_send_sig_info()
                            |                              |
                            |                              v
                            +--------------------> send_signal()

And in do_send_specifc() there are the lines:

        info->si_pid = task_tgid_vnr(current);
        info->si_uid = current->uid;

I suspect this should be fixed to be consistent with sigqueueinfo().

Cheers,

Michael

=====

/*#* multithread_sig_receiver.c

Copyright 2008, Linux Foundation;
written by Michael Kerrisk <[email protected]>
*/
#define _GNU_SOURCE
#include <sys/syscall.h>
#include <ctype.h>
#include <unistd.h>
#include <signal.h>
#include <pthread.h>
#include <sys/types.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <errno.h>

#define errExit(msg)    do { perror(msg); exit(EXIT_FAILURE); \
                        } while (0)
#define errExitEN(en, msg) \
                        do { errno = en; perror(msg); \
                             exit(EXIT_FAILURE); } while (0)

static sigset_t test_sigs;

struct thread_arg {
    int   thread_num;
    char *argv_string;
};

static pid_t
gettid(void)

{
    return syscall(SYS_gettid);
}

#if 0
static void
handler(int sig, siginfo_t *si, void *uc)

{
    printf("PID = %ld; TID = %ld; caught signal %d: \n",
            (long) getpid(), (long) gettid(), sig);
    printf("%p\n", si);
    printf("si_signo=%d ", si->si_signo);
    printf("si_code=%d ", si->si_code);
    printf("si_pid=%ld ", (long) si->si_pid);
    printf("si_uid=%ld ", (long) si->si_uid);
    printf("si_value.sival_int=%d\n", si->si_value.sival_int);
}

#endif

static void
wait_for_signals(int tnum)

{
    siginfo_t si;

    printf("Thread %d: about to wait for signals\n", tnum);

    for (;;) {
        if (sigwaitinfo(&test_sigs, &si) == -1)
            errExit("sigwaitinfo");
        printf("Thread %d: got signal: ", tnum);
        printf("si_signo=%d ", si.si_signo);
        printf("si_code=%d ", si.si_code);
        printf("si_pid=%ld ", (long) si.si_pid);
        printf("si_uid=%ld ", (long) si.si_uid);
        printf("si_value.sival_int=%d\n", si.si_value.sival_int);
    }
} /* wait_for_signals */

static void *
threadFunc(void *arg)

{
    struct thread_arg *ta = (struct thread_arg *) arg;
    int nsecs;

    printf("Thread %d: PID = %ld; TID = %ld\n", ta->thread_num,
            (long) getpid(), (long) gettid());

    nsecs = atoi(ta->argv_string);

    if (nsecs > 0) {
        printf("Thread %d: sleeping\n", ta->thread_num);
        sleep(nsecs);
        printf("Thread %d: finished sleeping\n", ta->thread_num);
    }

    wait_for_signals(ta->thread_num);
    return NULL;        /* NOTREACHED */
} /* threadFunc */

#define MAX_ARGS 1000

int
main(int argc, char *argv[])

{
    pthread_t t;
    int s, j;
//    struct sigaction sa;
    struct thread_arg ta[MAX_ARGS];
    sigset_t new;

    if (argc < 2) {
        fprintf(stderr, "Usage: %s <nsecs>...\n", argv[0]);
        fprintf(stderr, "Each <nsecs> argument causes a separate "
                "thread to be created;\n");
        fprintf(stderr, "The thread sleeps <nsecs> seconds before "
                "accepting signals with sigwaitinfo()\n");

        fprintf(stderr, "\n");
        exit(EXIT_SUCCESS);
    }

    if (argc > MAX_ARGS) {
        fprintf(stderr, "Too many arguments\n");
        exit(EXIT_SUCCESS);
    }

    sigemptyset(&test_sigs);
    for (s = 1; s < NSIG; s++)
        if (s != SIGINT && s != SIGQUIT && s != SIGTERM)
            sigaddset(&test_sigs, s);

    sigemptyset(&new);
    if (sigprocmask(SIG_SETMASK, &new, NULL) == -1)
        errExit("sigprocmask");
#if 0
    for (s = 1; s < NSIG; s++) {
        if (sigismember(&test_sigs, s)) {
            sa.sa_flags = SA_SIGINFO;
            sa.sa_sigaction = handler;
            sigemptyset(&sa.sa_mask);
            if (sigaction(s, &sa, NULL) == -1 && errno != EINVAL)
                errExit("sigaction");
        }
    }
    printf("Established handler for signals\n");

#endif

    printf("Main: TGID = %ld; TID = %ld\n",
            (long) getpid(), (long) gettid());

    s = pthread_sigmask(SIG_SETMASK, &test_sigs, NULL);
    if (s != 0)
        errExit("pthread_sigmask");
    printf("Main: blocked signals\n");

    for (j = 0; j + 1 < argc; j++) {
        ta[j].thread_num = j + 1;
        ta[j].argv_string = argv[j + 1];

        s = pthread_create(&t, NULL, threadFunc, &ta[j]);
        if (s != 0)
            errExitEN(s, "pthread_create");
    }

    wait_for_signals(0);
} /* main */

=====

/*#* t_rt_tgsigqueueinfo.c

Copyright 2008, Linux Foundation;
written by Michael Kerrisk <[email protected]>
*/
#define _GNU_SOURCE
#include <unistd.h>
#include <sys/syscall.h>
#include <stdlib.h>
#include <string.h>
#include <stdio.h>
#include <signal.h>
#include <sys/types.h>

#define errExit(msg)    do { perror(msg); exit(EXIT_FAILURE); \
                        } while (0)

#if defined(__i386__)
#define SYS_rt_tgsigqueueinfo 333
#endif

static int
rt_tgsigqueueinfo(pid_t tgid, pid_t tid, int sig, siginfo_t *si)

{
    return syscall(SYS_rt_tgsigqueueinfo, tgid, tid, sig, si);
}

int
main(int argc, char *argv[])

{
    pid_t tgid, tid;
    siginfo_t si;
    int sig, val;

    if (argc < 5) {
        fprintf(stderr, "Usage: %s <tgid> <tid> <sig> <val> "
                "[<si_code> [<si_pid [<si_uid>]]]\n", argv[0]);
        exit(EXIT_SUCCESS);
    }

    printf("My PID is %ld\n", (long) getpid());

    tgid = atoi(argv[1]);
    tid = atoi(argv[2]);
    sig = atoi(argv[3]);
    val = atoi(argv[4]);

    si.si_signo = sig + 1;
    si.si_code = (argc > 5) ? atoi(argv[5]) : SI_QUEUE;
    si.si_pid = (argc > 6) ? atoi(argv[6]) : getpid();
    si.si_uid = (argc > 7) ? atoi(argv[7]) : getuid();
    si.si_value.sival_int = val;

    /*
    printf("Sending: si.si_signo = %d, si.si_code = %d, "
            "si.si_pid = %ld, si.si_uid = %ld\n",
            si.si_signo, si.si_code, (long) si.si_pid, (long) si.si_uid);
    */
    if (rt_tgsigqueueinfo(tgid, tid, sig, &si) == -1)
        errExit("rt_tgsigqueueinfo");

    exit(EXIT_SUCCESS);
} /* main */

=====

/*#* t_rt_sigqueueinfo.c

Copyright 2008, Linux Foundation;
written by Michael Kerrisk <[email protected]>
*/
#define _GNU_SOURCE
#include <unistd.h>
#include <sys/syscall.h>
#include <stdlib.h>
#include <string.h>
#include <stdio.h>
#include <signal.h>
#include <sys/types.h>

#define errExit(msg)    do { perror(msg); exit(EXIT_FAILURE); \
                        } while (0)

static int
rt_sigqueueinfo(pid_t tgid, int sig, siginfo_t *si)

{
    return syscall(SYS_rt_sigqueueinfo, tgid, sig, si);
}

int
main(int argc, char *argv[])

{
    pid_t tgid;
    siginfo_t si;
    int sig, val;

    if (argc < 4) {
        fprintf(stderr, "Usage: %s <tgid> <sig> <val> "
                "[<si_code> [<si_pid [<si_uid>]]]\n", argv[0]);
        exit(EXIT_SUCCESS);
    }

    printf("My PID is %ld\n", (long) getpid());

    tgid = atoi(argv[1]);
    sig = atoi(argv[2]);
    val = atoi(argv[3]);

    si.si_signo = sig + 1;
    si.si_code = (argc > 4) ? atoi(argv[4]) : SI_QUEUE;
    si.si_pid = (argc > 5) ? atoi(argv[5]) : getpid();
    si.si_uid = (argc > 6) ? atoi(argv[6]) : getuid();
    si.si_value.sival_int = val;

    /*
    printf("Sending: si.si_signo = %d, si.si_code = %d, "
            "si.si_pid = %ld, si.si_uid = %ld\n",
            si.si_signo, si.si_code, (long) si.si_pid, (long) si.si_uid);
    */
    if (rt_sigqueueinfo(tgid, sig, &si) == -1)
        errExit("rt_sigqueueinfo");

    exit(EXIT_SUCCESS);
} /* main */

=====

Signature · 9 lines
Michael Kerrisk
Linux man-pages maintainer; http://www.kernel.org/doc/man-pages/
git://git.kernel.org/pub/scm/docs/man-pages/man-pages.git
man-pages online: http://www.kernel.org/doc/man-pages/online_pages.html
Found a bug? http://www.kernel.org/doc/man-pages/reporting_bugs.html
--
To unsubscribe from this list: send the line "unsubscribe linux-api" in
the body of a message to [email protected]
More majordomo info at  http://vger.kernel.org/majordomo-info.html
#5 lore raw

Michael,

Quoted 5 lines · Michael Kerrisk

I haven't looked at this iteration of the patch, to check if it is different from the version you posted a few months ago. I assume it isn't different, since you didn't mention any differences. In that case, there are still the questions I raised back then, so I'll just repeat them now.

Right, no fundamental changes.

Quoted 19 lines

Back in October, I did some testing of this interface. Two potential issues that I saw, both of which relate to inconsistencies with rt_sigqueueinfo():

1) With rt_siqueueinfo(), we can get the PID (TGID) of the sender, which enables the receiver of the signal to know who the sender was, and perhaps use that information to (for example) send a signal in the other direction.

With rt_tgsigqueueinfo(), we don't quite have that ability: all that the receiver gets is the TGID of the sender, not the TID. This means that we can't (for example) send a signal back to the precise thread that signaled us. I'm not sure if this matters or not (but perhaps it might when sender and receiver are in the same process?). I'm also not sure whether we want to do anything about this (i.e., extending siginfo_t to include a si_tid field), but I wanted to point out this assymetry w.r.t. to rt_sigqueueinfo(), in case you had not considered it.

Quoted 7 lines

2) With rt_sigqueueinfo(), we can specify the si_pid and and si_uid fields that should be sent to the receiver. This is not possible with rt_tgsigqueueinfo(), which always supplied the caller';s PID and UID in the si_pid and si_uid fields sent to the receiver. See the following, created using my test programs below (the 111 & 222 arguments to t_*sigqueueinfo set the si_pid and si_uid fields in the siginfo_t given to the *sigqueueinfo() syscall):

I can see your concern, but I have no strong opinion in either direction.

Roland ??

Thanks,

tglx
--
To unsubscribe from this list: send the line "unsubscribe linux-api" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html

#6 lore raw

Thomas,

Quoted 38 lines · Thomas Gleixner

Michael,

Quoted 5 lines · Michael Kerrisk

I haven't looked at this iteration of the patch, to check if it is different from the version you posted a few months ago. I assume it isn't different, since you didn't mention any differences. In that case, there are still the questions I raised back then, so I'll just repeat them now.

Right, no fundamental changes.

Quoted 19 lines

Back in October, I did some testing of this interface. Two potential issues that I saw, both of which relate to inconsistencies with rt_sigqueueinfo():

1) With rt_siqueueinfo(), we can get the PID (TGID) of the sender, which enables the receiver of the signal to know who the sender was, and perhaps use that information to (for example) send a signal in the other direction.

With rt_tgsigqueueinfo(), we don't quite have that ability: all that the receiver gets is the TGID of the sender, not the TID. This means that we can't (for example) send a signal back to the precise thread that signaled us. I'm not sure if this matters or not (but perhaps it might when sender and receiver are in the same process?). I'm also not sure whether we want to do anything about this (i.e., extending siginfo_t to include a si_tid field), but I wanted to point out this assymetry w.r.t. to rt_sigqueueinfo(), in case you had not considered it.

Quoted 7 lines

2) With rt_sigqueueinfo(), we can specify the si_pid and and si_uid fields that should be sent to the receiver. This is not possible with rt_tgsigqueueinfo(), which always supplied the caller';s PID and UID in the si_pid and si_uid fields sent to the receiver. See the following, created using my test programs below (the 111 & 222 arguments to t_*sigqueueinfo set the si_pid and si_uid fields in the siginfo_t given to the *sigqueueinfo() syscall):

Quoted 4 lines

I can see your concern, but I have no strong opinion in either direction.

Roland ??

The more I think about these two points, the more it seems to me they both should be fixed. Roland seems to concur on at least the second point. (Probably Ulrich should also be CCed for input -- done now.) With respect to the first point, it seems to me reasonably likely that there would be use cases where the receiving thread wants to know the thread ID of the sender -- especially when sender and receiver are in the same process.

Cheers,

Michael

Signature · 9 lines
Michael Kerrisk
Linux man-pages maintainer; http://www.kernel.org/doc/man-pages/
git://git.kernel.org/pub/scm/docs/man-pages/man-pages.git
man-pages online: http://www.kernel.org/doc/man-pages/online_pages.html
Found a bug? http://www.kernel.org/doc/man-pages/reporting_bugs.html
--
To unsubscribe from this list: send the line "unsubscribe linux-api" in
the body of a message to [email protected]
More majordomo info at  http://vger.kernel.org/majordomo-info.html
#7 lore raw
Quoted 4 lines

With respect to the first point, it seems to me reasonably likely that there would be use cases where the receiving thread wants to know the thread ID of the sender -- especially when sender and receiver are in the same process.

But expecting si_pid to play that role is bizarre. It's never what any POSIX-like program would do, both since POSIX says si_pid is a process ID, and because there are no POSIX-like interfaces at all that use Linux TIDs to refer to threads.

Wanting this only seems plausible within one process. In that case, sender and recipient know they share memory. The normal thing to do (and what POSIX applications will do) is to store that info somewhere pointed to by the sigval.

Thanks,
Roland
--
To unsubscribe from this list: send the line "unsubscribe linux-api" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html

#8 lore raw
Quoted 6 lines · Roland McGrath
Quoted 4 lines

With respect to the first point, it seems to me reasonably likely that there would be use cases where the receiving thread wants to know the thread ID of the sender -- especially when sender and receiver are in the same process.

But expecting si_pid to play that role is bizarre.

Oh -- I wasn't suggeting that si_pid do that task; I was instead wondering if we needed an additional si_threadid field (or some such).

Quoted 4 lines

It's never what any
POSIX-like program would do, both since POSIX says si_pid is a process ID, and because there are no POSIX-like interfaces at all that use Linux TIDs to refer to threads.

(Nod.)

Quoted 1 line

Wanting this only seems plausible within one process.

We are in agreement.

Quoted 4 lines

In that case, sender
and recipient know they share memory. The normal thing to do (and what POSIX applications will do) is to store that info somewhere pointed to by the sigval.

Good point. That would be another of dealing with point I wondered about, and probably better than my idea.

Cheers,

Michael

Signature · 9 lines
Michael Kerrisk
Linux man-pages maintainer; http://www.kernel.org/doc/man-pages/
git://git.kernel.org/pub/scm/docs/man-pages/man-pages.git
man-pages online: http://www.kernel.org/doc/man-pages/online_pages.html
Found a bug? http://www.kernel.org/doc/man-pages/reporting_bugs.html
--
To unsubscribe from this list: send the line "unsubscribe linux-api" in
the body of a message to [email protected]
More majordomo info at  http://vger.kernel.org/majordomo-info.html
#9 lore raw
In reply to Michael Kerrisk

I don't see any rationale for rt_tgsigqueueinfo and rt_sigqueueinfo to differ in their treatment of si_pid/si_uid (whatever that is). It just seems like common sense that they would match.

Oleg and/or Sukadev have some patches floating around (maybe all in -mm?) that relate to setting those.

I notice that POSIX says that si_pid and si_uid have reliable values whenever si_code <= 0 (what we call SI_FROMUSER() in asm/siginfo.h). (POSIX only has sigqueue() and kill() et al to send these, so a POSIX application never explicitly supplies the values. libc/libpthread do. The {t,tg,}kill syscalls all set si_pid to tgid already.)
This means a POSIX-conformant application might check si_pid and si_uid and rely on them not being forged by some other process/user. Firstly this means that si_pid must be the POSIX PID, i.e. tgid, not the Linux TID (which is not a useful value in POSIX interfaces). Secondly it means the kernel should guarantee the correctness of these values (at least when crossing processes, might as well do always).

I don't recall if the pending changes already use tgid, I think they do.

I think what both calls should do is set si_pid to tgid and si_uid to uid whenever SI_FROMUSER(). This satisfies the POSIX and security trust concern, and makes them uniform. (In Sukadev's version, what si_pid value they fill in here depends on pid_ns details of sender and recipient.)

Vis a vis Sukadev's changes, I also notice that si_uid ought to be translated for the recipient user_ns.

Thanks,
Roland
--
To unsubscribe from this list: send the line "unsubscribe linux-api" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html

#10 lore raw

Roland,

Quoted 6 lines · Roland McGrath

I don't see any rationale for rt_tgsigqueueinfo and rt_sigqueueinfo to differ in their treatment of si_pid/si_uid (whatever that is). It just seems like common sense that they would match.

Oleg and/or Sukadev have some patches floating around (maybe all in -mm?) that relate to setting those.

You mean having the same logic as we have in {t,tg,}kill syscalls for rt_sigqueueinfo as well ?

Is there anything else which stands in the way of getting the rt_tgsigqueueinfo interface merged ?

Thanks,

tglx
--
To unsubscribe from this list: send the line "unsubscribe linux-api" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html

#11 lore raw
Quoted 9 lines
Quoted 6 lines

I don't see any rationale for rt_tgsigqueueinfo and rt_sigqueueinfo to differ in their treatment of si_pid/si_uid (whatever that is). It just seems like common sense that they would match.

Oleg and/or Sukadev have some patches floating around (maybe all in -mm?) that relate to setting those.

You mean having the same logic as we have in {t,tg,}kill syscalls for rt_sigqueueinfo as well ?

I mean just what I said: rt_tgsigqueueinfo's treatment of the siginfo_t should match rt_sigqueueinfo's. Make it match rt_sigqueueinfo today and make sure that if rt_sigqueueinfo changes, then rt_tgsigqueueinfo will change to match.

Quoted 2 lines

Is there anything else which stands in the way of getting the rt_tgsigqueueinfo interface merged ?

I don't see any problem. But you should at least make sure that Ulrich likes the syscall interface for implementing pthread_sigqueue or whatever it will be.

Thanks,
Roland
--
To unsubscribe from this list: send the line "unsubscribe linux-api" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html