1// SPDX-License-Identifier: GPL-2.0
2/*
3 * fs/signalfd.c
4 *
5 * Copyright (C) 2003 Linus Torvalds
6 *
7 * Mon Mar 5, 2007: Davide Libenzi <davidel@xmailserver.org>
8 * Changed ->read() to return a siginfo strcture instead of signal number.
9 * Fixed locking in ->poll().
10 * Added sighand-detach notification.
11 * Added fd re-use in sys_signalfd() syscall.
12 * Now using anonymous inode source.
13 * Thanks to Oleg Nesterov for useful code review and suggestions.
14 * More comments and suggestions from Arnd Bergmann.
15 * Sat May 19, 2007: Davi E. M. Arnaut <davi@haxent.com.br>
16 * Retrieve multiple signals with one read() call
17 * Sun Jul 15, 2007: Davide Libenzi <davidel@xmailserver.org>
18 * Attach to the sighand only during read() and poll().
19 */
20
21#include <linux/file.h>
22#include <linux/poll.h>
23#include <linux/init.h>
24#include <linux/fs.h>
25#include <linux/sched.h>
26#include <linux/slab.h>
27#include <linux/kernel.h>
28#include <linux/signal.h>
29#include <linux/list.h>
30#include <linux/anon_inodes.h>
31#include <linux/signalfd.h>
32#include <linux/syscalls.h>
33#include <linux/proc_fs.h>
34#include <linux/compat.h>
35
36void signalfd_cleanup(struct sighand_struct *sighand)
37{
38 wait_queue_head_t *wqh = &sighand->signalfd_wqh;
39 /*
40 * The lockless check can race with remove_wait_queue() in progress,
41 * but in this case its caller should run under rcu_read_lock() and
42 * sighand_cachep is SLAB_TYPESAFE_BY_RCU, we can safely return.
43 */
44 if (likely(!waitqueue_active(wqh)))
45 return;
46
47 /* wait_queue_entry_t->func(POLLFREE) should do remove_wait_queue() */
48 wake_up_poll(wqh, EPOLLHUP | POLLFREE);
49}
50
51struct signalfd_ctx {
52 sigset_t sigmask;
53};
54
55static int signalfd_release(struct inode *inode, struct file *file)
56{
57 kfree(file->private_data);
58 return 0;
59}
60
61static __poll_t signalfd_poll(struct file *file, poll_table *wait)
62{
63 struct signalfd_ctx *ctx = file->private_data;
64 __poll_t events = 0;
65
66 poll_wait(file, &current->sighand->signalfd_wqh, wait);
67
68 spin_lock_irq(&current->sighand->siglock);
69 if (next_signal(&current->pending, &ctx->sigmask) ||
70 next_signal(&current->signal->shared_pending,
71 &ctx->sigmask))
72 events |= EPOLLIN;
73 spin_unlock_irq(&current->sighand->siglock);
74
75 return events;
76}
77
78/*
79 * Copied from copy_siginfo_to_user() in kernel/signal.c
80 */
81static int signalfd_copyinfo(struct signalfd_siginfo __user *uinfo,
82 kernel_siginfo_t const *kinfo)
83{
84 struct signalfd_siginfo new;
85
86 BUILD_BUG_ON(sizeof(struct signalfd_siginfo) != 128);
87
88 /*
89 * Unused members should be zero ...
90 */
91 memset(&new, 0, sizeof(new));
92
93 /*
94 * If you change siginfo_t structure, please be sure
95 * this code is fixed accordingly.
96 */
97 new.ssi_signo = kinfo->si_signo;
98 new.ssi_errno = kinfo->si_errno;
99 new.ssi_code = kinfo->si_code;
100 switch (siginfo_layout(kinfo->si_signo, kinfo->si_code)) {
101 case SIL_KILL:
102 new.ssi_pid = kinfo->si_pid;
103 new.ssi_uid = kinfo->si_uid;
104 break;
105 case SIL_TIMER:
106 new.ssi_tid = kinfo->si_tid;
107 new.ssi_overrun = kinfo->si_overrun;
108 new.ssi_ptr = (long) kinfo->si_ptr;
109 new.ssi_int = kinfo->si_int;
110 break;
111 case SIL_POLL:
112 new.ssi_band = kinfo->si_band;
113 new.ssi_fd = kinfo->si_fd;
114 break;
115 case SIL_FAULT_BNDERR:
116 case SIL_FAULT_PKUERR:
117 /*
118 * Fall through to the SIL_FAULT case. Both SIL_FAULT_BNDERR
119 * and SIL_FAULT_PKUERR are only generated by faults that
120 * deliver them synchronously to userspace. In case someone
121 * injects one of these signals and signalfd catches it treat
122 * it as SIL_FAULT.
123 */
124 case SIL_FAULT:
125 new.ssi_addr = (long) kinfo->si_addr;
126#ifdef __ARCH_SI_TRAPNO
127 new.ssi_trapno = kinfo->si_trapno;
128#endif
129 break;
130 case SIL_FAULT_MCEERR:
131 new.ssi_addr = (long) kinfo->si_addr;
132#ifdef __ARCH_SI_TRAPNO
133 new.ssi_trapno = kinfo->si_trapno;
134#endif
135 new.ssi_addr_lsb = (short) kinfo->si_addr_lsb;
136 break;
137 case SIL_CHLD:
138 new.ssi_pid = kinfo->si_pid;
139 new.ssi_uid = kinfo->si_uid;
140 new.ssi_status = kinfo->si_status;
141 new.ssi_utime = kinfo->si_utime;
142 new.ssi_stime = kinfo->si_stime;
143 break;
144 case SIL_RT:
145 /*
146 * This case catches also the signals queued by sigqueue().
147 */
148 new.ssi_pid = kinfo->si_pid;
149 new.ssi_uid = kinfo->si_uid;
150 new.ssi_ptr = (long) kinfo->si_ptr;
151 new.ssi_int = kinfo->si_int;
152 break;
153 case SIL_SYS:
154 new.ssi_call_addr = (long) kinfo->si_call_addr;
155 new.ssi_syscall = kinfo->si_syscall;
156 new.ssi_arch = kinfo->si_arch;
157 break;
158 }
159
160 if (copy_to_user(uinfo, &new, sizeof(struct signalfd_siginfo)))
161 return -EFAULT;
162
163 return sizeof(*uinfo);
164}
165
166static ssize_t signalfd_dequeue(struct signalfd_ctx *ctx, kernel_siginfo_t *info,
167 int nonblock)
168{
169 ssize_t ret;
170 DECLARE_WAITQUEUE(wait, current);
171
172 spin_lock_irq(&current->sighand->siglock);
173 ret = dequeue_signal(current, &ctx->sigmask, info);
174 switch (ret) {
175 case 0:
176 if (!nonblock)
177 break;
178 ret = -EAGAIN;
179 default:
180 spin_unlock_irq(&current->sighand->siglock);
181 return ret;
182 }
183
184 add_wait_queue(&current->sighand->signalfd_wqh, &wait);
185 for (;;) {
186 set_current_state(TASK_INTERRUPTIBLE);
187 ret = dequeue_signal(current, &ctx->sigmask, info);
188 if (ret != 0)
189 break;
190 if (signal_pending(current)) {
191 ret = -ERESTARTSYS;
192 break;
193 }
194 spin_unlock_irq(&current->sighand->siglock);
195 schedule();
196 spin_lock_irq(&current->sighand->siglock);
197 }
198 spin_unlock_irq(&current->sighand->siglock);
199
200 remove_wait_queue(&current->sighand->signalfd_wqh, &wait);
201 __set_current_state(TASK_RUNNING);
202
203 return ret;
204}
205
206/*
207 * Returns a multiple of the size of a "struct signalfd_siginfo", or a negative
208 * error code. The "count" parameter must be at least the size of a
209 * "struct signalfd_siginfo".
210 */
211static ssize_t signalfd_read(struct file *file, char __user *buf, size_t count,
212 loff_t *ppos)
213{
214 struct signalfd_ctx *ctx = file->private_data;
215 struct signalfd_siginfo __user *siginfo;
216 int nonblock = file->f_flags & O_NONBLOCK;
217 ssize_t ret, total = 0;
218 kernel_siginfo_t info;
219
220 count /= sizeof(struct signalfd_siginfo);
221 if (!count)
222 return -EINVAL;
223
224 siginfo = (struct signalfd_siginfo __user *) buf;
225 do {
226 ret = signalfd_dequeue(ctx, &info, nonblock);
227 if (unlikely(ret <= 0))
228 break;
229 ret = signalfd_copyinfo(siginfo, &info);
230 if (ret < 0)
231 break;
232 siginfo++;
233 total += ret;
234 nonblock = 1;
235 } while (--count);
236
237 return total ? total: ret;
238}
239
240#ifdef CONFIG_PROC_FS
241static void signalfd_show_fdinfo(struct seq_file *m, struct file *f)
242{
243 struct signalfd_ctx *ctx = f->private_data;
244 sigset_t sigmask;
245
246 sigmask = ctx->sigmask;
247 signotset(&sigmask);
248 render_sigset_t(m, "sigmask:\t", &sigmask);
249}
250#endif
251
252static const struct file_operations signalfd_fops = {
253#ifdef CONFIG_PROC_FS
254 .show_fdinfo = signalfd_show_fdinfo,
255#endif
256 .release = signalfd_release,
257 .poll = signalfd_poll,
258 .read = signalfd_read,
259 .llseek = noop_llseek,
260};
261
262static int do_signalfd4(int ufd, sigset_t *mask, int flags)
263{
264 struct signalfd_ctx *ctx;
265
266 /* Check the SFD_* constants for consistency. */
267 BUILD_BUG_ON(SFD_CLOEXEC != O_CLOEXEC);
268 BUILD_BUG_ON(SFD_NONBLOCK != O_NONBLOCK);
269
270 if (flags & ~(SFD_CLOEXEC | SFD_NONBLOCK))
271 return -EINVAL;
272
273 sigdelsetmask(mask, sigmask(SIGKILL) | sigmask(SIGSTOP));
274 signotset(mask);
275
276 if (ufd == -1) {
277 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
278 if (!ctx)
279 return -ENOMEM;
280
281 ctx->sigmask = *mask;
282
283 /*
284 * When we call this, the initialization must be complete, since
285 * anon_inode_getfd() will install the fd.
286 */
287 ufd = anon_inode_getfd("[signalfd]", &signalfd_fops, ctx,
288 O_RDWR | (flags & (O_CLOEXEC | O_NONBLOCK)));
289 if (ufd < 0)
290 kfree(ctx);
291 } else {
292 struct fd f = fdget(ufd);
293 if (!f.file)
294 return -EBADF;
295 ctx = f.file->private_data;
296 if (f.file->f_op != &signalfd_fops) {
297 fdput(f);
298 return -EINVAL;
299 }
300 spin_lock_irq(&current->sighand->siglock);
301 ctx->sigmask = *mask;
302 spin_unlock_irq(&current->sighand->siglock);
303
304 wake_up(&current->sighand->signalfd_wqh);
305 fdput(f);
306 }
307
308 return ufd;
309}
310
311SYSCALL_DEFINE4(signalfd4, int, ufd, sigset_t __user *, user_mask,
312 size_t, sizemask, int, flags)
313{
314 sigset_t mask;
315
316 if (sizemask != sizeof(sigset_t) ||
317 copy_from_user(&mask, user_mask, sizeof(mask)))
318 return -EINVAL;
319 return do_signalfd4(ufd, &mask, flags);
320}
321
322SYSCALL_DEFINE3(signalfd, int, ufd, sigset_t __user *, user_mask,
323 size_t, sizemask)
324{
325 sigset_t mask;
326
327 if (sizemask != sizeof(sigset_t) ||
328 copy_from_user(&mask, user_mask, sizeof(mask)))
329 return -EINVAL;
330 return do_signalfd4(ufd, &mask, 0);
331}
332
333#ifdef CONFIG_COMPAT
334static long do_compat_signalfd4(int ufd,
335 const compat_sigset_t __user *user_mask,
336 compat_size_t sigsetsize, int flags)
337{
338 sigset_t mask;
339
340 if (sigsetsize != sizeof(compat_sigset_t))
341 return -EINVAL;
342 if (get_compat_sigset(&mask, user_mask))
343 return -EFAULT;
344 return do_signalfd4(ufd, &mask, flags);
345}
346
347COMPAT_SYSCALL_DEFINE4(signalfd4, int, ufd,
348 const compat_sigset_t __user *, user_mask,
349 compat_size_t, sigsetsize,
350 int, flags)
351{
352 return do_compat_signalfd4(ufd, user_mask, sigsetsize, flags);
353}
354
355COMPAT_SYSCALL_DEFINE3(signalfd, int, ufd,
356 const compat_sigset_t __user *, user_mask,
357 compat_size_t, sigsetsize)
358{
359 return do_compat_signalfd4(ufd, user_mask, sigsetsize, 0);
360}
361#endif
362