blob: 424b7b65321fecbe2afff24cb3f3f69dfa81bc26 [file] [log] [blame]
Davide Libenzifba2afa2007-05-10 22:23:13 -07001/*
2 * fs/signalfd.c
3 *
4 * Copyright (C) 2003 Linus Torvalds
5 *
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.
Davide Libenzib8fceee2007-09-20 12:40:16 -070014 * Sat May 19, 2007: Davi E. M. Arnaut <davi@haxent.com.br>
Davi Arnautb3762bf2007-05-23 13:58:04 -070015 * Retrieve multiple signals with one read() call
Davide Libenzib8fceee2007-09-20 12:40:16 -070016 * Sun Jul 15, 2007: Davide Libenzi <davidel@xmailserver.org>
17 * Attach to the sighand only during read() and poll().
Davide Libenzifba2afa2007-05-10 22:23:13 -070018 */
19
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>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090025#include <linux/slab.h>
Davide Libenzifba2afa2007-05-10 22:23:13 -070026#include <linux/kernel.h>
27#include <linux/signal.h>
28#include <linux/list.h>
29#include <linux/anon_inodes.h>
30#include <linux/signalfd.h>
Adrian Bunk7ec37df2008-02-06 01:36:48 -080031#include <linux/syscalls.h>
Cyrill Gorcunov138d22b2012-12-17 16:05:02 -080032#include <linux/proc_fs.h>
Al Viro7d197ed2013-02-24 01:41:39 -050033#include <linux/compat.h>
Davide Libenzifba2afa2007-05-10 22:23:13 -070034
Oleg Nesterovd80e7312012-02-24 20:07:11 +010035void signalfd_cleanup(struct sighand_struct *sighand)
36{
37 wait_queue_head_t *wqh = &sighand->signalfd_wqh;
Oleg Nesterov971316f2012-02-24 20:07:29 +010038 /*
39 * The lockless check can race with remove_wait_queue() in progress,
40 * but in this case its caller should run under rcu_read_lock() and
41 * sighand_cachep is SLAB_DESTROY_BY_RCU, we can safely return.
42 */
Oleg Nesterovd80e7312012-02-24 20:07:11 +010043 if (likely(!waitqueue_active(wqh)))
44 return;
45
46 /* wait_queue_t->func(POLLFREE) should do remove_wait_queue() */
47 wake_up_poll(wqh, POLLHUP | POLLFREE);
48}
49
Davide Libenzifba2afa2007-05-10 22:23:13 -070050struct signalfd_ctx {
Davide Libenzifba2afa2007-05-10 22:23:13 -070051 sigset_t sigmask;
Davide Libenzifba2afa2007-05-10 22:23:13 -070052};
53
Davide Libenzifba2afa2007-05-10 22:23:13 -070054static int signalfd_release(struct inode *inode, struct file *file)
55{
Davide Libenzib8fceee2007-09-20 12:40:16 -070056 kfree(file->private_data);
Davide Libenzifba2afa2007-05-10 22:23:13 -070057 return 0;
58}
59
60static unsigned int signalfd_poll(struct file *file, poll_table *wait)
61{
62 struct signalfd_ctx *ctx = file->private_data;
63 unsigned int events = 0;
Davide Libenzifba2afa2007-05-10 22:23:13 -070064
Davide Libenzib8fceee2007-09-20 12:40:16 -070065 poll_wait(file, &current->sighand->signalfd_wqh, wait);
Davide Libenzifba2afa2007-05-10 22:23:13 -070066
Davide Libenzib8fceee2007-09-20 12:40:16 -070067 spin_lock_irq(&current->sighand->siglock);
68 if (next_signal(&current->pending, &ctx->sigmask) ||
69 next_signal(&current->signal->shared_pending,
70 &ctx->sigmask))
Davide Libenzifba2afa2007-05-10 22:23:13 -070071 events |= POLLIN;
Davide Libenzib8fceee2007-09-20 12:40:16 -070072 spin_unlock_irq(&current->sighand->siglock);
Davide Libenzifba2afa2007-05-10 22:23:13 -070073
74 return events;
75}
76
77/*
78 * Copied from copy_siginfo_to_user() in kernel/signal.c
79 */
80static int signalfd_copyinfo(struct signalfd_siginfo __user *uinfo,
81 siginfo_t const *kinfo)
82{
83 long err;
84
85 BUILD_BUG_ON(sizeof(struct signalfd_siginfo) != 128);
86
87 /*
Robert P. J. Day14e4a0f2008-02-03 15:12:15 +020088 * Unused members should be zero ...
Davide Libenzifba2afa2007-05-10 22:23:13 -070089 */
90 err = __clear_user(uinfo, sizeof(*uinfo));
91
92 /*
93 * If you change siginfo_t structure, please be sure
94 * this code is fixed accordingly.
95 */
Davide Libenzi96358de2007-10-16 23:30:23 -070096 err |= __put_user(kinfo->si_signo, &uinfo->ssi_signo);
97 err |= __put_user(kinfo->si_errno, &uinfo->ssi_errno);
98 err |= __put_user((short) kinfo->si_code, &uinfo->ssi_code);
Davide Libenzifba2afa2007-05-10 22:23:13 -070099 switch (kinfo->si_code & __SI_MASK) {
100 case __SI_KILL:
Davide Libenzi96358de2007-10-16 23:30:23 -0700101 err |= __put_user(kinfo->si_pid, &uinfo->ssi_pid);
102 err |= __put_user(kinfo->si_uid, &uinfo->ssi_uid);
Davide Libenzifba2afa2007-05-10 22:23:13 -0700103 break;
104 case __SI_TIMER:
Davide Libenzi96358de2007-10-16 23:30:23 -0700105 err |= __put_user(kinfo->si_tid, &uinfo->ssi_tid);
106 err |= __put_user(kinfo->si_overrun, &uinfo->ssi_overrun);
107 err |= __put_user((long) kinfo->si_ptr, &uinfo->ssi_ptr);
Nathan Lyncha2a20c42010-08-10 18:03:08 -0700108 err |= __put_user(kinfo->si_int, &uinfo->ssi_int);
Davide Libenzifba2afa2007-05-10 22:23:13 -0700109 break;
110 case __SI_POLL:
Davide Libenzi96358de2007-10-16 23:30:23 -0700111 err |= __put_user(kinfo->si_band, &uinfo->ssi_band);
112 err |= __put_user(kinfo->si_fd, &uinfo->ssi_fd);
Davide Libenzifba2afa2007-05-10 22:23:13 -0700113 break;
114 case __SI_FAULT:
Davide Libenzi96358de2007-10-16 23:30:23 -0700115 err |= __put_user((long) kinfo->si_addr, &uinfo->ssi_addr);
Davide Libenzifba2afa2007-05-10 22:23:13 -0700116#ifdef __ARCH_SI_TRAPNO
Davide Libenzi96358de2007-10-16 23:30:23 -0700117 err |= __put_user(kinfo->si_trapno, &uinfo->ssi_trapno);
Davide Libenzifba2afa2007-05-10 22:23:13 -0700118#endif
Hidetoshi Setob8aeec32010-10-07 15:31:31 +0900119#ifdef BUS_MCEERR_AO
120 /*
121 * Other callers might not initialize the si_lsb field,
122 * so check explicitly for the right codes here.
123 */
124 if (kinfo->si_code == BUS_MCEERR_AR ||
125 kinfo->si_code == BUS_MCEERR_AO)
126 err |= __put_user((short) kinfo->si_addr_lsb,
127 &uinfo->ssi_addr_lsb);
128#endif
Davide Libenzifba2afa2007-05-10 22:23:13 -0700129 break;
130 case __SI_CHLD:
Davide Libenzi96358de2007-10-16 23:30:23 -0700131 err |= __put_user(kinfo->si_pid, &uinfo->ssi_pid);
132 err |= __put_user(kinfo->si_uid, &uinfo->ssi_uid);
133 err |= __put_user(kinfo->si_status, &uinfo->ssi_status);
134 err |= __put_user(kinfo->si_utime, &uinfo->ssi_utime);
135 err |= __put_user(kinfo->si_stime, &uinfo->ssi_stime);
Davide Libenzifba2afa2007-05-10 22:23:13 -0700136 break;
137 case __SI_RT: /* This is not generated by the kernel as of now. */
138 case __SI_MESGQ: /* But this is */
Davide Libenzi96358de2007-10-16 23:30:23 -0700139 err |= __put_user(kinfo->si_pid, &uinfo->ssi_pid);
140 err |= __put_user(kinfo->si_uid, &uinfo->ssi_uid);
141 err |= __put_user((long) kinfo->si_ptr, &uinfo->ssi_ptr);
Nathan Lyncha2a20c42010-08-10 18:03:08 -0700142 err |= __put_user(kinfo->si_int, &uinfo->ssi_int);
Davide Libenzifba2afa2007-05-10 22:23:13 -0700143 break;
Davide Libenzi0859ab52008-04-10 21:29:29 -0700144 default:
145 /*
146 * This case catches also the signals queued by sigqueue().
147 */
Davide Libenzi96358de2007-10-16 23:30:23 -0700148 err |= __put_user(kinfo->si_pid, &uinfo->ssi_pid);
149 err |= __put_user(kinfo->si_uid, &uinfo->ssi_uid);
Davide Libenzi0859ab52008-04-10 21:29:29 -0700150 err |= __put_user((long) kinfo->si_ptr, &uinfo->ssi_ptr);
151 err |= __put_user(kinfo->si_int, &uinfo->ssi_int);
Davide Libenzifba2afa2007-05-10 22:23:13 -0700152 break;
153 }
154
155 return err ? -EFAULT: sizeof(*uinfo);
156}
157
Davi Arnautb3762bf2007-05-23 13:58:04 -0700158static ssize_t signalfd_dequeue(struct signalfd_ctx *ctx, siginfo_t *info,
159 int nonblock)
160{
161 ssize_t ret;
Davi Arnautb3762bf2007-05-23 13:58:04 -0700162 DECLARE_WAITQUEUE(wait, current);
163
Davide Libenzib8fceee2007-09-20 12:40:16 -0700164 spin_lock_irq(&current->sighand->siglock);
165 ret = dequeue_signal(current, &ctx->sigmask, info);
Davi Arnautb3762bf2007-05-23 13:58:04 -0700166 switch (ret) {
167 case 0:
168 if (!nonblock)
169 break;
170 ret = -EAGAIN;
171 default:
Davide Libenzib8fceee2007-09-20 12:40:16 -0700172 spin_unlock_irq(&current->sighand->siglock);
Davi Arnautb3762bf2007-05-23 13:58:04 -0700173 return ret;
174 }
175
Davide Libenzib8fceee2007-09-20 12:40:16 -0700176 add_wait_queue(&current->sighand->signalfd_wqh, &wait);
Davi Arnautb3762bf2007-05-23 13:58:04 -0700177 for (;;) {
178 set_current_state(TASK_INTERRUPTIBLE);
Davide Libenzib8fceee2007-09-20 12:40:16 -0700179 ret = dequeue_signal(current, &ctx->sigmask, info);
Davi Arnautb3762bf2007-05-23 13:58:04 -0700180 if (ret != 0)
181 break;
182 if (signal_pending(current)) {
183 ret = -ERESTARTSYS;
184 break;
185 }
Davide Libenzib8fceee2007-09-20 12:40:16 -0700186 spin_unlock_irq(&current->sighand->siglock);
Davi Arnautb3762bf2007-05-23 13:58:04 -0700187 schedule();
Davide Libenzib8fceee2007-09-20 12:40:16 -0700188 spin_lock_irq(&current->sighand->siglock);
Davi Arnautb3762bf2007-05-23 13:58:04 -0700189 }
Davide Libenzib8fceee2007-09-20 12:40:16 -0700190 spin_unlock_irq(&current->sighand->siglock);
Davi Arnautb3762bf2007-05-23 13:58:04 -0700191
Davide Libenzib8fceee2007-09-20 12:40:16 -0700192 remove_wait_queue(&current->sighand->signalfd_wqh, &wait);
Davi Arnautb3762bf2007-05-23 13:58:04 -0700193 __set_current_state(TASK_RUNNING);
194
195 return ret;
196}
197
Davide Libenzifba2afa2007-05-10 22:23:13 -0700198/*
Davide Libenzib8fceee2007-09-20 12:40:16 -0700199 * Returns a multiple of the size of a "struct signalfd_siginfo", or a negative
200 * error code. The "count" parameter must be at least the size of a
201 * "struct signalfd_siginfo".
Davide Libenzifba2afa2007-05-10 22:23:13 -0700202 */
203static ssize_t signalfd_read(struct file *file, char __user *buf, size_t count,
204 loff_t *ppos)
205{
206 struct signalfd_ctx *ctx = file->private_data;
Davi Arnautb3762bf2007-05-23 13:58:04 -0700207 struct signalfd_siginfo __user *siginfo;
208 int nonblock = file->f_flags & O_NONBLOCK;
209 ssize_t ret, total = 0;
Davide Libenzifba2afa2007-05-10 22:23:13 -0700210 siginfo_t info;
Davide Libenzifba2afa2007-05-10 22:23:13 -0700211
Davi Arnautb3762bf2007-05-23 13:58:04 -0700212 count /= sizeof(struct signalfd_siginfo);
213 if (!count)
Davide Libenzifba2afa2007-05-10 22:23:13 -0700214 return -EINVAL;
Davide Libenzifba2afa2007-05-10 22:23:13 -0700215
Davi Arnautb3762bf2007-05-23 13:58:04 -0700216 siginfo = (struct signalfd_siginfo __user *) buf;
Davi Arnautb3762bf2007-05-23 13:58:04 -0700217 do {
218 ret = signalfd_dequeue(ctx, &info, nonblock);
219 if (unlikely(ret <= 0))
220 break;
221 ret = signalfd_copyinfo(siginfo, &info);
222 if (ret < 0)
223 break;
224 siginfo++;
225 total += ret;
226 nonblock = 1;
227 } while (--count);
228
Davide Libenzib8fceee2007-09-20 12:40:16 -0700229 return total ? total: ret;
Davide Libenzifba2afa2007-05-10 22:23:13 -0700230}
231
Cyrill Gorcunov138d22b2012-12-17 16:05:02 -0800232#ifdef CONFIG_PROC_FS
233static int signalfd_show_fdinfo(struct seq_file *m, struct file *f)
234{
235 struct signalfd_ctx *ctx = f->private_data;
236 sigset_t sigmask;
237
238 sigmask = ctx->sigmask;
239 signotset(&sigmask);
240 render_sigset_t(m, "sigmask:\t", &sigmask);
241
242 return 0;
243}
244#endif
245
Davide Libenzifba2afa2007-05-10 22:23:13 -0700246static const struct file_operations signalfd_fops = {
Cyrill Gorcunov138d22b2012-12-17 16:05:02 -0800247#ifdef CONFIG_PROC_FS
248 .show_fdinfo = signalfd_show_fdinfo,
249#endif
Davide Libenzifba2afa2007-05-10 22:23:13 -0700250 .release = signalfd_release,
251 .poll = signalfd_poll,
252 .read = signalfd_read,
Arnd Bergmann6038f372010-08-15 18:52:59 +0200253 .llseek = noop_llseek,
Davide Libenzifba2afa2007-05-10 22:23:13 -0700254};
255
Heiko Carstens836f92a2009-01-14 14:14:33 +0100256SYSCALL_DEFINE4(signalfd4, int, ufd, sigset_t __user *, user_mask,
257 size_t, sizemask, int, flags)
Davide Libenzifba2afa2007-05-10 22:23:13 -0700258{
Davide Libenzifba2afa2007-05-10 22:23:13 -0700259 sigset_t sigmask;
260 struct signalfd_ctx *ctx;
Davide Libenzifba2afa2007-05-10 22:23:13 -0700261
Ulrich Dreppere38b36f2008-07-23 21:29:42 -0700262 /* Check the SFD_* constants for consistency. */
263 BUILD_BUG_ON(SFD_CLOEXEC != O_CLOEXEC);
264 BUILD_BUG_ON(SFD_NONBLOCK != O_NONBLOCK);
265
Ulrich Drepper5fb5e042008-07-23 21:29:37 -0700266 if (flags & ~(SFD_CLOEXEC | SFD_NONBLOCK))
Ulrich Drepper9deb27b2008-07-23 21:29:24 -0700267 return -EINVAL;
268
Davide Libenzifba2afa2007-05-10 22:23:13 -0700269 if (sizemask != sizeof(sigset_t) ||
270 copy_from_user(&sigmask, user_mask, sizeof(sigmask)))
Ulrich Drepperf50cadaa2007-07-26 10:40:55 -0700271 return -EINVAL;
Davide Libenzifba2afa2007-05-10 22:23:13 -0700272 sigdelsetmask(&sigmask, sigmask(SIGKILL) | sigmask(SIGSTOP));
273 signotset(&sigmask);
274
275 if (ufd == -1) {
276 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
277 if (!ctx)
278 return -ENOMEM;
279
Davide Libenzifba2afa2007-05-10 22:23:13 -0700280 ctx->sigmask = sigmask;
Davide Libenzifba2afa2007-05-10 22:23:13 -0700281
282 /*
283 * When we call this, the initialization must be complete, since
284 * anon_inode_getfd() will install the fd.
285 */
Ulrich Drepper7d9dbca2008-07-23 21:29:22 -0700286 ufd = anon_inode_getfd("[signalfd]", &signalfd_fops, ctx,
Roland Dreier628ff7c2009-12-18 09:41:24 -0800287 O_RDWR | (flags & (O_CLOEXEC | O_NONBLOCK)));
Al Viro2030a422008-02-23 06:46:49 -0500288 if (ufd < 0)
289 kfree(ctx);
Davide Libenzifba2afa2007-05-10 22:23:13 -0700290 } else {
Al Viro2903ff02012-08-28 12:52:22 -0400291 struct fd f = fdget(ufd);
292 if (!f.file)
Davide Libenzifba2afa2007-05-10 22:23:13 -0700293 return -EBADF;
Al Viro2903ff02012-08-28 12:52:22 -0400294 ctx = f.file->private_data;
295 if (f.file->f_op != &signalfd_fops) {
296 fdput(f);
Davide Libenzifba2afa2007-05-10 22:23:13 -0700297 return -EINVAL;
298 }
Davide Libenzib8fceee2007-09-20 12:40:16 -0700299 spin_lock_irq(&current->sighand->siglock);
300 ctx->sigmask = sigmask;
301 spin_unlock_irq(&current->sighand->siglock);
302
303 wake_up(&current->sighand->signalfd_wqh);
Al Viro2903ff02012-08-28 12:52:22 -0400304 fdput(f);
Davide Libenzifba2afa2007-05-10 22:23:13 -0700305 }
306
307 return ufd;
Davide Libenzifba2afa2007-05-10 22:23:13 -0700308}
Ulrich Drepper9deb27b2008-07-23 21:29:24 -0700309
Heiko Carstens836f92a2009-01-14 14:14:33 +0100310SYSCALL_DEFINE3(signalfd, int, ufd, sigset_t __user *, user_mask,
311 size_t, sizemask)
Ulrich Drepper9deb27b2008-07-23 21:29:24 -0700312{
313 return sys_signalfd4(ufd, user_mask, sizemask, 0);
314}
Al Viro7d197ed2013-02-24 01:41:39 -0500315
316#ifdef CONFIG_COMPAT
317COMPAT_SYSCALL_DEFINE4(signalfd4, int, ufd,
318 const compat_sigset_t __user *,sigmask,
319 compat_size_t, sigsetsize,
320 int, flags)
321{
322 compat_sigset_t ss32;
323 sigset_t tmp;
324 sigset_t __user *ksigmask;
325
326 if (sigsetsize != sizeof(compat_sigset_t))
327 return -EINVAL;
328 if (copy_from_user(&ss32, sigmask, sizeof(ss32)))
329 return -EFAULT;
330 sigset_from_compat(&tmp, &ss32);
331 ksigmask = compat_alloc_user_space(sizeof(sigset_t));
332 if (copy_to_user(ksigmask, &tmp, sizeof(sigset_t)))
333 return -EFAULT;
334
335 return sys_signalfd4(ufd, ksigmask, sizeof(sigset_t), flags);
336}
337
338COMPAT_SYSCALL_DEFINE3(signalfd, int, ufd,
339 const compat_sigset_t __user *,sigmask,
340 compat_size_t, sigsetsize)
341{
342 return compat_sys_signalfd4(ufd, sigmask, sigsetsize, 0);
343}
344#endif