hrtimer: optimize the softirq time optimization
[linux-2.6/zen-sources.git] / fs / signalfd.c
blob8ead0db359339ca7f5986e3b89ff3b245938557e
1 /*
2 * fs/signalfd.c
4 * Copyright (C) 2003 Linus Torvalds
6 * Mon Mar 5, 2007: Davide Libenzi <davidel@xmailserver.org>
7 * Changed ->read() to return a siginfo strcture instead of signal number.
8 * Fixed locking in ->poll().
9 * Added sighand-detach notification.
10 * Added fd re-use in sys_signalfd() syscall.
11 * Now using anonymous inode source.
12 * Thanks to Oleg Nesterov for useful code review and suggestions.
13 * More comments and suggestions from Arnd Bergmann.
14 * Sat May 19, 2007: Davi E. M. Arnaut <davi@haxent.com.br>
15 * Retrieve multiple signals with one read() call
16 * Sun Jul 15, 2007: Davide Libenzi <davidel@xmailserver.org>
17 * Attach to the sighand only during read() and poll().
20 #include <linux/file.h>
21 #include <linux/poll.h>
22 #include <linux/init.h>
23 #include <linux/fs.h>
24 #include <linux/sched.h>
25 #include <linux/kernel.h>
26 #include <linux/signal.h>
27 #include <linux/list.h>
28 #include <linux/anon_inodes.h>
29 #include <linux/signalfd.h>
30 #include <linux/syscalls.h>
32 struct signalfd_ctx {
33 sigset_t sigmask;
36 static int signalfd_release(struct inode *inode, struct file *file)
38 kfree(file->private_data);
39 return 0;
42 static unsigned int signalfd_poll(struct file *file, poll_table *wait)
44 struct signalfd_ctx *ctx = file->private_data;
45 unsigned int events = 0;
47 poll_wait(file, &current->sighand->signalfd_wqh, wait);
49 spin_lock_irq(&current->sighand->siglock);
50 if (next_signal(&current->pending, &ctx->sigmask) ||
51 next_signal(&current->signal->shared_pending,
52 &ctx->sigmask))
53 events |= POLLIN;
54 spin_unlock_irq(&current->sighand->siglock);
56 return events;
60 * Copied from copy_siginfo_to_user() in kernel/signal.c
62 static int signalfd_copyinfo(struct signalfd_siginfo __user *uinfo,
63 siginfo_t const *kinfo)
65 long err;
67 BUILD_BUG_ON(sizeof(struct signalfd_siginfo) != 128);
70 * Unused members should be zero ...
72 err = __clear_user(uinfo, sizeof(*uinfo));
75 * If you change siginfo_t structure, please be sure
76 * this code is fixed accordingly.
78 err |= __put_user(kinfo->si_signo, &uinfo->ssi_signo);
79 err |= __put_user(kinfo->si_errno, &uinfo->ssi_errno);
80 err |= __put_user((short) kinfo->si_code, &uinfo->ssi_code);
81 switch (kinfo->si_code & __SI_MASK) {
82 case __SI_KILL:
83 err |= __put_user(kinfo->si_pid, &uinfo->ssi_pid);
84 err |= __put_user(kinfo->si_uid, &uinfo->ssi_uid);
85 break;
86 case __SI_TIMER:
87 err |= __put_user(kinfo->si_tid, &uinfo->ssi_tid);
88 err |= __put_user(kinfo->si_overrun, &uinfo->ssi_overrun);
89 err |= __put_user((long) kinfo->si_ptr, &uinfo->ssi_ptr);
90 break;
91 case __SI_POLL:
92 err |= __put_user(kinfo->si_band, &uinfo->ssi_band);
93 err |= __put_user(kinfo->si_fd, &uinfo->ssi_fd);
94 break;
95 case __SI_FAULT:
96 err |= __put_user((long) kinfo->si_addr, &uinfo->ssi_addr);
97 #ifdef __ARCH_SI_TRAPNO
98 err |= __put_user(kinfo->si_trapno, &uinfo->ssi_trapno);
99 #endif
100 break;
101 case __SI_CHLD:
102 err |= __put_user(kinfo->si_pid, &uinfo->ssi_pid);
103 err |= __put_user(kinfo->si_uid, &uinfo->ssi_uid);
104 err |= __put_user(kinfo->si_status, &uinfo->ssi_status);
105 err |= __put_user(kinfo->si_utime, &uinfo->ssi_utime);
106 err |= __put_user(kinfo->si_stime, &uinfo->ssi_stime);
107 break;
108 case __SI_RT: /* This is not generated by the kernel as of now. */
109 case __SI_MESGQ: /* But this is */
110 err |= __put_user(kinfo->si_pid, &uinfo->ssi_pid);
111 err |= __put_user(kinfo->si_uid, &uinfo->ssi_uid);
112 err |= __put_user((long) kinfo->si_ptr, &uinfo->ssi_ptr);
113 break;
114 default:
116 * This case catches also the signals queued by sigqueue().
118 err |= __put_user(kinfo->si_pid, &uinfo->ssi_pid);
119 err |= __put_user(kinfo->si_uid, &uinfo->ssi_uid);
120 err |= __put_user((long) kinfo->si_ptr, &uinfo->ssi_ptr);
121 err |= __put_user(kinfo->si_int, &uinfo->ssi_int);
122 break;
125 return err ? -EFAULT: sizeof(*uinfo);
128 static ssize_t signalfd_dequeue(struct signalfd_ctx *ctx, siginfo_t *info,
129 int nonblock)
131 ssize_t ret;
132 DECLARE_WAITQUEUE(wait, current);
134 spin_lock_irq(&current->sighand->siglock);
135 ret = dequeue_signal(current, &ctx->sigmask, info);
136 switch (ret) {
137 case 0:
138 if (!nonblock)
139 break;
140 ret = -EAGAIN;
141 default:
142 spin_unlock_irq(&current->sighand->siglock);
143 return ret;
146 add_wait_queue(&current->sighand->signalfd_wqh, &wait);
147 for (;;) {
148 set_current_state(TASK_INTERRUPTIBLE);
149 ret = dequeue_signal(current, &ctx->sigmask, info);
150 if (ret != 0)
151 break;
152 if (signal_pending(current)) {
153 ret = -ERESTARTSYS;
154 break;
156 spin_unlock_irq(&current->sighand->siglock);
157 schedule();
158 spin_lock_irq(&current->sighand->siglock);
160 spin_unlock_irq(&current->sighand->siglock);
162 remove_wait_queue(&current->sighand->signalfd_wqh, &wait);
163 __set_current_state(TASK_RUNNING);
165 return ret;
169 * Returns a multiple of the size of a "struct signalfd_siginfo", or a negative
170 * error code. The "count" parameter must be at least the size of a
171 * "struct signalfd_siginfo".
173 static ssize_t signalfd_read(struct file *file, char __user *buf, size_t count,
174 loff_t *ppos)
176 struct signalfd_ctx *ctx = file->private_data;
177 struct signalfd_siginfo __user *siginfo;
178 int nonblock = file->f_flags & O_NONBLOCK;
179 ssize_t ret, total = 0;
180 siginfo_t info;
182 count /= sizeof(struct signalfd_siginfo);
183 if (!count)
184 return -EINVAL;
186 siginfo = (struct signalfd_siginfo __user *) buf;
187 do {
188 ret = signalfd_dequeue(ctx, &info, nonblock);
189 if (unlikely(ret <= 0))
190 break;
191 ret = signalfd_copyinfo(siginfo, &info);
192 if (ret < 0)
193 break;
194 siginfo++;
195 total += ret;
196 nonblock = 1;
197 } while (--count);
199 return total ? total: ret;
202 static const struct file_operations signalfd_fops = {
203 .release = signalfd_release,
204 .poll = signalfd_poll,
205 .read = signalfd_read,
208 asmlinkage long sys_signalfd(int ufd, sigset_t __user *user_mask, size_t sizemask)
210 int error;
211 sigset_t sigmask;
212 struct signalfd_ctx *ctx;
213 struct file *file;
214 struct inode *inode;
216 if (sizemask != sizeof(sigset_t) ||
217 copy_from_user(&sigmask, user_mask, sizeof(sigmask)))
218 return -EINVAL;
219 sigdelsetmask(&sigmask, sigmask(SIGKILL) | sigmask(SIGSTOP));
220 signotset(&sigmask);
222 if (ufd == -1) {
223 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
224 if (!ctx)
225 return -ENOMEM;
227 ctx->sigmask = sigmask;
230 * When we call this, the initialization must be complete, since
231 * anon_inode_getfd() will install the fd.
233 error = anon_inode_getfd(&ufd, &inode, &file, "[signalfd]",
234 &signalfd_fops, ctx);
235 if (error)
236 goto err_fdalloc;
237 } else {
238 file = fget(ufd);
239 if (!file)
240 return -EBADF;
241 ctx = file->private_data;
242 if (file->f_op != &signalfd_fops) {
243 fput(file);
244 return -EINVAL;
246 spin_lock_irq(&current->sighand->siglock);
247 ctx->sigmask = sigmask;
248 spin_unlock_irq(&current->sighand->siglock);
250 wake_up(&current->sighand->signalfd_wqh);
251 fput(file);
254 return ufd;
256 err_fdalloc:
257 kfree(ctx);
258 return error;