| 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 | |
| 36 | void 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 | |
| 51 | struct signalfd_ctx { |
| 52 | sigset_t sigmask; |
| 53 | }; |
| 54 | |
| 55 | static int signalfd_release(struct inode *inode, struct file *file) |
| 56 | { |
| 57 | kfree(file->private_data); |
| 58 | return 0; |
| 59 | } |
| 60 | |
| 61 | static __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, ¤t->sighand->signalfd_wqh, wait); |
| 67 | |
| 68 | spin_lock_irq(¤t->sighand->siglock); |
| 69 | if (next_signal(¤t->pending, &ctx->sigmask) || |
| 70 | next_signal(¤t->signal->shared_pending, |
| 71 | &ctx->sigmask)) |
| 72 | events |= EPOLLIN; |
| 73 | spin_unlock_irq(¤t->sighand->siglock); |
| 74 | |
| 75 | return events; |
| 76 | } |
| 77 | |
| 78 | /* |
| 79 | * Copied from copy_siginfo_to_user() in kernel/signal.c |
| 80 | */ |
| 81 | static 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 | |
| 166 | static 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(¤t->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 | /* fall through */ |
| 180 | default: |
| 181 | spin_unlock_irq(¤t->sighand->siglock); |
| 182 | return ret; |
| 183 | } |
| 184 | |
| 185 | add_wait_queue(¤t->sighand->signalfd_wqh, &wait); |
| 186 | for (;;) { |
| 187 | set_current_state(TASK_INTERRUPTIBLE); |
| 188 | ret = dequeue_signal(current, &ctx->sigmask, info); |
| 189 | if (ret != 0) |
| 190 | break; |
| 191 | if (signal_pending(current)) { |
| 192 | ret = -ERESTARTSYS; |
| 193 | break; |
| 194 | } |
| 195 | spin_unlock_irq(¤t->sighand->siglock); |
| 196 | schedule(); |
| 197 | spin_lock_irq(¤t->sighand->siglock); |
| 198 | } |
| 199 | spin_unlock_irq(¤t->sighand->siglock); |
| 200 | |
| 201 | remove_wait_queue(¤t->sighand->signalfd_wqh, &wait); |
| 202 | __set_current_state(TASK_RUNNING); |
| 203 | |
| 204 | return ret; |
| 205 | } |
| 206 | |
| 207 | /* |
| 208 | * Returns a multiple of the size of a "struct signalfd_siginfo", or a negative |
| 209 | * error code. The "count" parameter must be at least the size of a |
| 210 | * "struct signalfd_siginfo". |
| 211 | */ |
| 212 | static ssize_t signalfd_read(struct file *file, char __user *buf, size_t count, |
| 213 | loff_t *ppos) |
| 214 | { |
| 215 | struct signalfd_ctx *ctx = file->private_data; |
| 216 | struct signalfd_siginfo __user *siginfo; |
| 217 | int nonblock = file->f_flags & O_NONBLOCK; |
| 218 | ssize_t ret, total = 0; |
| 219 | kernel_siginfo_t info; |
| 220 | |
| 221 | count /= sizeof(struct signalfd_siginfo); |
| 222 | if (!count) |
| 223 | return -EINVAL; |
| 224 | |
| 225 | siginfo = (struct signalfd_siginfo __user *) buf; |
| 226 | do { |
| 227 | ret = signalfd_dequeue(ctx, &info, nonblock); |
| 228 | if (unlikely(ret <= 0)) |
| 229 | break; |
| 230 | ret = signalfd_copyinfo(siginfo, &info); |
| 231 | if (ret < 0) |
| 232 | break; |
| 233 | siginfo++; |
| 234 | total += ret; |
| 235 | nonblock = 1; |
| 236 | } while (--count); |
| 237 | |
| 238 | return total ? total: ret; |
| 239 | } |
| 240 | |
| 241 | #ifdef CONFIG_PROC_FS |
| 242 | static void signalfd_show_fdinfo(struct seq_file *m, struct file *f) |
| 243 | { |
| 244 | struct signalfd_ctx *ctx = f->private_data; |
| 245 | sigset_t sigmask; |
| 246 | |
| 247 | sigmask = ctx->sigmask; |
| 248 | signotset(&sigmask); |
| 249 | render_sigset_t(m, "sigmask:\t", &sigmask); |
| 250 | } |
| 251 | #endif |
| 252 | |
| 253 | static const struct file_operations signalfd_fops = { |
| 254 | #ifdef CONFIG_PROC_FS |
| 255 | .show_fdinfo = signalfd_show_fdinfo, |
| 256 | #endif |
| 257 | .release = signalfd_release, |
| 258 | .poll = signalfd_poll, |
| 259 | .read = signalfd_read, |
| 260 | .llseek = noop_llseek, |
| 261 | }; |
| 262 | |
| 263 | static int do_signalfd4(int ufd, sigset_t *mask, int flags) |
| 264 | { |
| 265 | struct signalfd_ctx *ctx; |
| 266 | |
| 267 | /* Check the SFD_* constants for consistency. */ |
| 268 | BUILD_BUG_ON(SFD_CLOEXEC != O_CLOEXEC); |
| 269 | BUILD_BUG_ON(SFD_NONBLOCK != O_NONBLOCK); |
| 270 | |
| 271 | if (flags & ~(SFD_CLOEXEC | SFD_NONBLOCK)) |
| 272 | return -EINVAL; |
| 273 | |
| 274 | sigdelsetmask(mask, sigmask(SIGKILL) | sigmask(SIGSTOP)); |
| 275 | signotset(mask); |
| 276 | |
| 277 | if (ufd == -1) { |
| 278 | ctx = kmalloc(sizeof(*ctx), GFP_KERNEL); |
| 279 | if (!ctx) |
| 280 | return -ENOMEM; |
| 281 | |
| 282 | ctx->sigmask = *mask; |
| 283 | |
| 284 | /* |
| 285 | * When we call this, the initialization must be complete, since |
| 286 | * anon_inode_getfd() will install the fd. |
| 287 | */ |
| 288 | ufd = anon_inode_getfd("[signalfd]", &signalfd_fops, ctx, |
| 289 | O_RDWR | (flags & (O_CLOEXEC | O_NONBLOCK))); |
| 290 | if (ufd < 0) |
| 291 | kfree(ctx); |
| 292 | } else { |
| 293 | struct fd f = fdget(ufd); |
| 294 | if (!f.file) |
| 295 | return -EBADF; |
| 296 | ctx = f.file->private_data; |
| 297 | if (f.file->f_op != &signalfd_fops) { |
| 298 | fdput(f); |
| 299 | return -EINVAL; |
| 300 | } |
| 301 | spin_lock_irq(¤t->sighand->siglock); |
| 302 | ctx->sigmask = *mask; |
| 303 | spin_unlock_irq(¤t->sighand->siglock); |
| 304 | |
| 305 | wake_up(¤t->sighand->signalfd_wqh); |
| 306 | fdput(f); |
| 307 | } |
| 308 | |
| 309 | return ufd; |
| 310 | } |
| 311 | |
| 312 | SYSCALL_DEFINE4(signalfd4, int, ufd, sigset_t __user *, user_mask, |
| 313 | size_t, sizemask, int, flags) |
| 314 | { |
| 315 | sigset_t mask; |
| 316 | |
| 317 | if (sizemask != sizeof(sigset_t) || |
| 318 | copy_from_user(&mask, user_mask, sizeof(mask))) |
| 319 | return -EINVAL; |
| 320 | return do_signalfd4(ufd, &mask, flags); |
| 321 | } |
| 322 | |
| 323 | SYSCALL_DEFINE3(signalfd, int, ufd, sigset_t __user *, user_mask, |
| 324 | size_t, sizemask) |
| 325 | { |
| 326 | sigset_t mask; |
| 327 | |
| 328 | if (sizemask != sizeof(sigset_t) || |
| 329 | copy_from_user(&mask, user_mask, sizeof(mask))) |
| 330 | return -EINVAL; |
| 331 | return do_signalfd4(ufd, &mask, 0); |
| 332 | } |
| 333 | |
| 334 | #ifdef CONFIG_COMPAT |
| 335 | static long do_compat_signalfd4(int ufd, |
| 336 | const compat_sigset_t __user *user_mask, |
| 337 | compat_size_t sigsetsize, int flags) |
| 338 | { |
| 339 | sigset_t mask; |
| 340 | |
| 341 | if (sigsetsize != sizeof(compat_sigset_t)) |
| 342 | return -EINVAL; |
| 343 | if (get_compat_sigset(&mask, user_mask)) |
| 344 | return -EFAULT; |
| 345 | return do_signalfd4(ufd, &mask, flags); |
| 346 | } |
| 347 | |
| 348 | COMPAT_SYSCALL_DEFINE4(signalfd4, int, ufd, |
| 349 | const compat_sigset_t __user *, user_mask, |
| 350 | compat_size_t, sigsetsize, |
| 351 | int, flags) |
| 352 | { |
| 353 | return do_compat_signalfd4(ufd, user_mask, sigsetsize, flags); |
| 354 | } |
| 355 | |
| 356 | COMPAT_SYSCALL_DEFINE3(signalfd, int, ufd, |
| 357 | const compat_sigset_t __user *, user_mask, |
| 358 | compat_size_t, sigsetsize) |
| 359 | { |
| 360 | return do_compat_signalfd4(ufd, user_mask, sigsetsize, 0); |
| 361 | } |
| 362 | #endif |