[RFC PATCH v2] waitfd

[Casey Dahlin <cdahlin@redhat.com>]

Linux now exposes signals, timers, and events via file descriptors 
through signalfd, timerfd, and eventfd. This means programmers can use a 
single select/[e]poll call to monitor all change in their program. This 
patch aims to expose child death via the same mechanism.

waitfd provides a file descriptor out of which may be read a series of 
siginfo_t objects describing child death. A child process is reaped as 
soon as its information is read. This means child monitoring too can be 
performed with that same poll call.

Patch is against v2.6.28

--CJD

diff --git a/arch/x86/include/asm/unistd_32.h 
b/arch/x86/include/asm/unistd_32.h
index f2bba78..134d83c 100644
--- a/arch/x86/include/asm/unistd_32.h
+++ b/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_waitfd        333
 
 #ifdef __KERNEL__
 
diff --git a/arch/x86/include/asm/unistd_64.h 
b/arch/x86/include/asm/unistd_64.h
index d2e415e..b28eb07 100644
--- a/arch/x86/include/asm/unistd_64.h
+++ b/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_waitfd                295
+__SYSCALL(__NR_waitfd, sys_waitfd)
 
 
 #ifndef __NO_STUBS
diff --git a/arch/x86/kernel/syscall_table_32.S 
b/arch/x86/kernel/syscall_table_32.S
index d44395f..c796a8b 100644
--- a/arch/x86/kernel/syscall_table_32.S
+++ b/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_waitfd
diff --git a/fs/Makefile b/fs/Makefile
index d9f8afe..74c31fb 100644
--- a/fs/Makefile
+++ b/fs/Makefile
@@ -25,6 +25,7 @@ obj-$(CONFIG_INOTIFY_USER)    += inotify_user.o
 obj-$(CONFIG_EPOLL)        += eventpoll.o
 obj-$(CONFIG_ANON_INODES)    += anon_inodes.o
 obj-$(CONFIG_SIGNALFD)        += signalfd.o
+obj-$(CONFIG_WAITFD)        += waitfd.o
 obj-$(CONFIG_TIMERFD)        += timerfd.o
 obj-$(CONFIG_EVENTFD)        += eventfd.o
 obj-$(CONFIG_AIO)               += aio.o
diff --git a/fs/waitfd.c b/fs/waitfd.c
new file mode 100644
index 0000000..0155a83
--- /dev/null
+++ b/fs/waitfd.c
@@ -0,0 +1,117 @@
+/*
+ *  fs/waitfd.c
+ *
+ *  Copyright (C) 2008  Red Hat, Casey Dahlin <cdahlin@redhat.com>
+ *
+ *  Largely derived from fs/signalfd.c
+ */
+
+#include <linux/file.h>
+#include <linux/poll.h>
+#include <linux/init.h>
+#include <linux/fs.h>
+#include <linux/sched.h>
+#include <linux/kernel.h>
+#include <linux/signal.h>
+#include <linux/list.h>
+#include <linux/anon_inodes.h>
+#include <linux/syscalls.h>
+
+long do_waitid(int which, pid_t upid,
+           struct siginfo __user *infop, int options,
+           struct rusage __user *ru);
+
+struct waitfd_ctx {
+    int ops;
+    int which;
+    pid_t upid;
+};
+
+static int waitfd_release(struct inode *inode, struct file *file)
+{
+    kfree(file->private_data);
+    return 0;
+}
+
+static unsigned int waitfd_poll(struct file *file, poll_table *wait)
+{
+    struct waitfd_ctx *ctx = file->private_data;
+    long value;
+
+    poll_wait(file, &current->signal->wait_chldexit, wait);
+
+    value = do_waitid(ctx->which, ctx->upid, NULL,
+               ctx->ops | WNOHANG | WNOWAIT, NULL);
+    if (value > 0 || value == -ECHILD)
+        return POLLIN;
+
+    return 0;
+}
+
+/*
+ * Returns a multiple of the size of a struct siginfo, or a negative
+ * error code. The "count" parameter must be at least sizeof(struct 
siginfo)
+ */
+static ssize_t waitfd_read(struct file *file, char __user *buf, size_t 
count,
+                 loff_t *ppos)
+{
+    struct waitfd_ctx *ctx = file->private_data;
+    struct siginfo __user *info_addr = (struct siginfo *)buf;
+    int flags = ctx->ops;
+    ssize_t ret, total = 0;
+
+    count /= sizeof(struct siginfo);
+    if (!count)
+        return -EINVAL;
+
+    do {
+        ret = do_waitid(ctx->which, ctx->upid, info_addr, flags, NULL);
+        if (ret == 0)
+            ret = -EAGAIN;
+        if (ret == -ECHILD)
+            ret = 0;
+        if (ret <= 0)
+            break;
+
+        info_addr++;
+        total += sizeof(struct siginfo);
+        flags |= WNOHANG;
+    } while (--count);
+
+    return total ? total: ret;
+}
+
+static const struct file_operations waitfd_fops = {
+    .release    = waitfd_release,
+    .poll        = waitfd_poll,
+    .read        = waitfd_read,
+};
+
+asmlinkage long sys_waitfd(int which, pid_t upid, int options, int unused)
+{
+    int ufd;
+    struct waitfd_ctx *ctx;
+
+    /* Just to make sure we don't end up with a sys_waitfd4 */
+    (void)unused;
+
+    if (options & ~(WNOHANG|WEXITED|WSTOPPED|WCONTINUED))
+        return -EINVAL;
+    if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
+        return -EINVAL;
+
+    ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
+    if (!ctx)
+        return -ENOMEM;
+
+    ctx->ops = options;
+    ctx->upid = upid;
+    ctx->which = which;
+
+    ufd = anon_inode_getfd("[waitfd]", &waitfd_fops, ctx,
+                   (options & WNOHANG) ? O_NONBLOCK : 0);
+    if (ufd < 0)
+        kfree(ctx);
+
+    return ufd;
+}
diff --git a/init/Kconfig b/init/Kconfig
index f763762..bc34871 100644
--- a/init/Kconfig
+++ b/init/Kconfig
@@ -683,6 +683,16 @@ config EPOLL
       Disabling this option will cause the kernel to be built without
       support for epoll family of system calls.
 
+config WAITFD
+    bool "Enable waitfd() system call" if EMBEDDED
+    select ANON_INODES
+    default y
+    help
+      Enable the waitfd() system call that allows receving child state
+      changes from a file descriptor.
+
+      If unsure, say Y.
+
 config SIGNALFD
     bool "Enable signalfd() system call" if EMBEDDED
     select ANON_INODES
diff --git a/kernel/exit.c b/kernel/exit.c
index 2d8be7e..b53e8ba 100644
--- a/kernel/exit.c
+++ b/kernel/exit.c
@@ -1233,18 +1233,20 @@ static int wait_noreap_copyout(struct 
task_struct *p, pid_t pid, uid_t uid,
     int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
 
     put_task_struct(p);
-    if (!retval)
-        retval = put_user(SIGCHLD, &infop->si_signo);
-    if (!retval)
-        retval = put_user(0, &infop->si_errno);
-    if (!retval)
-        retval = put_user((short)why, &infop->si_code);
-    if (!retval)
-        retval = put_user(pid, &infop->si_pid);
-    if (!retval)
-        retval = put_user(uid, &infop->si_uid);
-    if (!retval)
-        retval = put_user(status, &infop->si_status);
+    if (infop) {
+        if (!retval)
+            retval = put_user(SIGCHLD, &infop->si_signo);
+        if (!retval)
+            retval = put_user(0, &infop->si_errno);
+        if (!retval)
+            retval = put_user((short)why, &infop->si_code);
+        if (!retval)
+            retval = put_user(pid, &infop->si_pid);
+        if (!retval)
+            retval = put_user(uid, &infop->si_uid);
+        if (!retval)
+            retval = put_user(status, &infop->si_status);
+    }
     if (!retval)
         retval = pid;
     return retval;
@@ -1727,35 +1729,12 @@ repeat:
 end:
     current->state = TASK_RUNNING;
     remove_wait_queue(&current->signal->wait_chldexit,&wait);
-    if (infop) {
-        if (retval > 0)
-            retval = 0;
-        else {
-            /*
-             * For a WNOHANG return, clear out all the fields
-             * we would set so the user can easily tell the
-             * difference.
-             */
-            if (!retval)
-                retval = put_user(0, &infop->si_signo);
-            if (!retval)
-                retval = put_user(0, &infop->si_errno);
-            if (!retval)
-                retval = put_user(0, &infop->si_code);
-            if (!retval)
-                retval = put_user(0, &infop->si_pid);
-            if (!retval)
-                retval = put_user(0, &infop->si_uid);
-            if (!retval)
-                retval = put_user(0, &infop->si_status);
-        }
-    }
     return retval;
 }
 
-asmlinkage long sys_waitid(int which, pid_t upid,
-               struct siginfo __user *infop, int options,
-               struct rusage __user *ru)
+long do_waitid(int which, pid_t upid,
+           struct siginfo __user *infop, int options,
+           struct rusage __user *ru)
 {
     struct pid *pid = NULL;
     enum pid_type type;
@@ -1789,6 +1768,39 @@ asmlinkage long sys_waitid(int which, pid_t upid,
     ret = do_wait(type, pid, options, infop, NULL, ru);
     put_pid(pid);
 
+    return ret;
+}
+
+asmlinkage long sys_waitid(int which, pid_t upid,
+               struct siginfo __user *infop, int options,
+               struct rusage __user *ru)
+{
+    long ret;
+
+    ret = do_waitid(which, upid, infop, options, ru);
+
+    if (ret > 0)
+        ret = 0;
+    else {
+        /*
+         * For a WNOHANG return, clear out all the fields
+         * we would set so the user can easily tell the
+         * difference.
+         */
+        if (!ret)
+            ret = put_user(0, &infop->si_signo);
+        if (!ret)
+            ret = put_user(0, &infop->si_errno);
+        if (!ret)
+            ret = put_user(0, &infop->si_code);
+        if (!ret)
+            ret = put_user(0, &infop->si_pid);
+        if (!ret)
+            ret = put_user(0, &infop->si_uid);
+        if (!ret)
+            ret = put_user(0, &infop->si_status);
+    }
+
     /* avoid REGPARM breakage on x86: */
     asmlinkage_protect(5, ret, which, upid, infop, options, ru);
     return ret;
diff --git a/kernel/sys_ni.c b/kernel/sys_ni.c
index e14a232..e8d4da6 100644
--- a/kernel/sys_ni.c
+++ b/kernel/sys_ni.c
@@ -163,6 +163,7 @@ cond_syscall(sys_ioprio_set);
 cond_syscall(sys_ioprio_get);
 
 /* New file descriptors */
+cond_syscall(sys_waitfd);
 cond_syscall(sys_signalfd);
 cond_syscall(sys_signalfd4);
 cond_syscall(compat_sys_signalfd);