| 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 | wake_up_pollfree(wq_head: &sighand->signalfd_wqh); |
| 39 | } |
| 40 | |
| 41 | struct signalfd_ctx { |
| 42 | sigset_t sigmask; |
| 43 | }; |
| 44 | |
| 45 | static int signalfd_release(struct inode *inode, struct file *file) |
| 46 | { |
| 47 | kfree(objp: file->private_data); |
| 48 | return 0; |
| 49 | } |
| 50 | |
| 51 | static __poll_t signalfd_poll(struct file *file, poll_table *wait) |
| 52 | { |
| 53 | struct signalfd_ctx *ctx = file->private_data; |
| 54 | __poll_t events = 0; |
| 55 | |
| 56 | poll_wait(filp: file, wait_address: ¤t->sighand->signalfd_wqh, p: wait); |
| 57 | |
| 58 | spin_lock_irq(lock: ¤t->sighand->siglock); |
| 59 | if (next_signal(pending: ¤t->pending, mask: &ctx->sigmask) || |
| 60 | next_signal(pending: ¤t->signal->shared_pending, |
| 61 | mask: &ctx->sigmask)) |
| 62 | events |= EPOLLIN; |
| 63 | spin_unlock_irq(lock: ¤t->sighand->siglock); |
| 64 | |
| 65 | return events; |
| 66 | } |
| 67 | |
| 68 | /* |
| 69 | * Copied from copy_siginfo_to_user() in kernel/signal.c |
| 70 | */ |
| 71 | static int signalfd_copyinfo(struct iov_iter *to, kernel_siginfo_t const *kinfo) |
| 72 | { |
| 73 | struct signalfd_siginfo new; |
| 74 | |
| 75 | BUILD_BUG_ON(sizeof(struct signalfd_siginfo) != 128); |
| 76 | |
| 77 | /* |
| 78 | * Unused members should be zero ... |
| 79 | */ |
| 80 | memset(s: &new, c: 0, n: sizeof(new)); |
| 81 | |
| 82 | /* |
| 83 | * If you change siginfo_t structure, please be sure |
| 84 | * this code is fixed accordingly. |
| 85 | */ |
| 86 | new.ssi_signo = kinfo->si_signo; |
| 87 | new.ssi_errno = kinfo->si_errno; |
| 88 | new.ssi_code = kinfo->si_code; |
| 89 | switch (siginfo_layout(sig: kinfo->si_signo, si_code: kinfo->si_code)) { |
| 90 | case SIL_KILL: |
| 91 | new.ssi_pid = kinfo->si_pid; |
| 92 | new.ssi_uid = kinfo->si_uid; |
| 93 | break; |
| 94 | case SIL_TIMER: |
| 95 | new.ssi_tid = kinfo->si_tid; |
| 96 | new.ssi_overrun = kinfo->si_overrun; |
| 97 | new.ssi_ptr = (long) kinfo->si_ptr; |
| 98 | new.ssi_int = kinfo->si_int; |
| 99 | break; |
| 100 | case SIL_POLL: |
| 101 | new.ssi_band = kinfo->si_band; |
| 102 | new.ssi_fd = kinfo->si_fd; |
| 103 | break; |
| 104 | case SIL_FAULT_BNDERR: |
| 105 | case SIL_FAULT_PKUERR: |
| 106 | case SIL_FAULT_PERF_EVENT: |
| 107 | /* |
| 108 | * Fall through to the SIL_FAULT case. SIL_FAULT_BNDERR, |
| 109 | * SIL_FAULT_PKUERR, and SIL_FAULT_PERF_EVENT are only |
| 110 | * generated by faults that deliver them synchronously to |
| 111 | * userspace. In case someone injects one of these signals |
| 112 | * and signalfd catches it treat it as SIL_FAULT. |
| 113 | */ |
| 114 | case SIL_FAULT: |
| 115 | new.ssi_addr = (long) kinfo->si_addr; |
| 116 | break; |
| 117 | case SIL_FAULT_TRAPNO: |
| 118 | new.ssi_addr = (long) kinfo->si_addr; |
| 119 | new.ssi_trapno = kinfo->si_trapno; |
| 120 | break; |
| 121 | case SIL_FAULT_MCEERR: |
| 122 | new.ssi_addr = (long) kinfo->si_addr; |
| 123 | new.ssi_addr_lsb = (short) kinfo->si_addr_lsb; |
| 124 | break; |
| 125 | case SIL_CHLD: |
| 126 | new.ssi_pid = kinfo->si_pid; |
| 127 | new.ssi_uid = kinfo->si_uid; |
| 128 | new.ssi_status = kinfo->si_status; |
| 129 | new.ssi_utime = kinfo->si_utime; |
| 130 | new.ssi_stime = kinfo->si_stime; |
| 131 | break; |
| 132 | case SIL_RT: |
| 133 | /* |
| 134 | * This case catches also the signals queued by sigqueue(). |
| 135 | */ |
| 136 | new.ssi_pid = kinfo->si_pid; |
| 137 | new.ssi_uid = kinfo->si_uid; |
| 138 | new.ssi_ptr = (long) kinfo->si_ptr; |
| 139 | new.ssi_int = kinfo->si_int; |
| 140 | break; |
| 141 | case SIL_SYS: |
| 142 | new.ssi_call_addr = (long) kinfo->si_call_addr; |
| 143 | new.ssi_syscall = kinfo->si_syscall; |
| 144 | new.ssi_arch = kinfo->si_arch; |
| 145 | break; |
| 146 | } |
| 147 | |
| 148 | if (!copy_to_iter_full(addr: &new, bytes: sizeof(struct signalfd_siginfo), i: to)) |
| 149 | return -EFAULT; |
| 150 | |
| 151 | return sizeof(struct signalfd_siginfo); |
| 152 | } |
| 153 | |
| 154 | static ssize_t signalfd_dequeue(struct signalfd_ctx *ctx, kernel_siginfo_t *info, |
| 155 | int nonblock) |
| 156 | { |
| 157 | enum pid_type type; |
| 158 | ssize_t ret; |
| 159 | DECLARE_WAITQUEUE(wait, current); |
| 160 | |
| 161 | spin_lock_irq(lock: ¤t->sighand->siglock); |
| 162 | ret = dequeue_signal(mask: &ctx->sigmask, info, type: &type); |
| 163 | switch (ret) { |
| 164 | case 0: |
| 165 | if (!nonblock) |
| 166 | break; |
| 167 | ret = -EAGAIN; |
| 168 | fallthrough; |
| 169 | default: |
| 170 | spin_unlock_irq(lock: ¤t->sighand->siglock); |
| 171 | return ret; |
| 172 | } |
| 173 | |
| 174 | add_wait_queue(wq_head: ¤t->sighand->signalfd_wqh, wq_entry: &wait); |
| 175 | for (;;) { |
| 176 | set_current_state(TASK_INTERRUPTIBLE); |
| 177 | ret = dequeue_signal(mask: &ctx->sigmask, info, type: &type); |
| 178 | if (ret != 0) |
| 179 | break; |
| 180 | if (signal_pending(current)) { |
| 181 | ret = -ERESTARTSYS; |
| 182 | break; |
| 183 | } |
| 184 | spin_unlock_irq(lock: ¤t->sighand->siglock); |
| 185 | schedule(); |
| 186 | spin_lock_irq(lock: ¤t->sighand->siglock); |
| 187 | } |
| 188 | spin_unlock_irq(lock: ¤t->sighand->siglock); |
| 189 | |
| 190 | remove_wait_queue(wq_head: ¤t->sighand->signalfd_wqh, wq_entry: &wait); |
| 191 | __set_current_state(TASK_RUNNING); |
| 192 | |
| 193 | return ret; |
| 194 | } |
| 195 | |
| 196 | /* |
| 197 | * Returns a multiple of the size of a "struct signalfd_siginfo", or a negative |
| 198 | * error code. The "count" parameter must be at least the size of a |
| 199 | * "struct signalfd_siginfo". |
| 200 | */ |
| 201 | static ssize_t signalfd_read_iter(struct kiocb *iocb, struct iov_iter *to) |
| 202 | { |
| 203 | struct file *file = iocb->ki_filp; |
| 204 | struct signalfd_ctx *ctx = file->private_data; |
| 205 | size_t count = iov_iter_count(i: to); |
| 206 | ssize_t ret, total = 0; |
| 207 | kernel_siginfo_t info; |
| 208 | bool nonblock; |
| 209 | |
| 210 | count /= sizeof(struct signalfd_siginfo); |
| 211 | if (!count) |
| 212 | return -EINVAL; |
| 213 | |
| 214 | nonblock = file->f_flags & O_NONBLOCK || iocb->ki_flags & IOCB_NOWAIT; |
| 215 | do { |
| 216 | ret = signalfd_dequeue(ctx, info: &info, nonblock); |
| 217 | if (unlikely(ret <= 0)) |
| 218 | break; |
| 219 | ret = signalfd_copyinfo(to, kinfo: &info); |
| 220 | if (ret < 0) |
| 221 | break; |
| 222 | total += ret; |
| 223 | nonblock = 1; |
| 224 | } while (--count); |
| 225 | |
| 226 | return total ? total: ret; |
| 227 | } |
| 228 | |
| 229 | #ifdef CONFIG_PROC_FS |
| 230 | static void signalfd_show_fdinfo(struct seq_file *m, struct file *f) |
| 231 | { |
| 232 | struct signalfd_ctx *ctx = f->private_data; |
| 233 | sigset_t sigmask; |
| 234 | |
| 235 | sigmask = ctx->sigmask; |
| 236 | signotset(set: &sigmask); |
| 237 | render_sigset_t(m, "sigmask:\t" , &sigmask); |
| 238 | } |
| 239 | #endif |
| 240 | |
| 241 | static const struct file_operations signalfd_fops = { |
| 242 | #ifdef CONFIG_PROC_FS |
| 243 | .show_fdinfo = signalfd_show_fdinfo, |
| 244 | #endif |
| 245 | .release = signalfd_release, |
| 246 | .poll = signalfd_poll, |
| 247 | .read_iter = signalfd_read_iter, |
| 248 | .llseek = noop_llseek, |
| 249 | }; |
| 250 | |
| 251 | static int do_signalfd4(int ufd, sigset_t *mask, int flags) |
| 252 | { |
| 253 | struct signalfd_ctx *ctx; |
| 254 | |
| 255 | /* Check the SFD_* constants for consistency. */ |
| 256 | BUILD_BUG_ON(SFD_CLOEXEC != O_CLOEXEC); |
| 257 | BUILD_BUG_ON(SFD_NONBLOCK != O_NONBLOCK); |
| 258 | |
| 259 | if (flags & ~(SFD_CLOEXEC | SFD_NONBLOCK)) |
| 260 | return -EINVAL; |
| 261 | |
| 262 | sigdelsetmask(set: mask, sigmask(SIGKILL) | sigmask(SIGSTOP)); |
| 263 | signotset(set: mask); |
| 264 | |
| 265 | if (ufd == -1) { |
| 266 | struct file *file; |
| 267 | |
| 268 | ctx = kmalloc(sizeof(*ctx), GFP_KERNEL); |
| 269 | if (!ctx) |
| 270 | return -ENOMEM; |
| 271 | |
| 272 | ctx->sigmask = *mask; |
| 273 | |
| 274 | ufd = get_unused_fd_flags(flags: flags & O_CLOEXEC); |
| 275 | if (ufd < 0) { |
| 276 | kfree(objp: ctx); |
| 277 | return ufd; |
| 278 | } |
| 279 | |
| 280 | file = anon_inode_getfile_fmode(name: "[signalfd]" , fops: &signalfd_fops, |
| 281 | priv: ctx, O_RDWR | (flags & O_NONBLOCK), |
| 282 | FMODE_NOWAIT); |
| 283 | if (IS_ERR(ptr: file)) { |
| 284 | put_unused_fd(fd: ufd); |
| 285 | kfree(objp: ctx); |
| 286 | return PTR_ERR(ptr: file); |
| 287 | } |
| 288 | fd_install(fd: ufd, file); |
| 289 | } else { |
| 290 | CLASS(fd, f)(fd: ufd); |
| 291 | if (fd_empty(f)) |
| 292 | return -EBADF; |
| 293 | ctx = fd_file(f)->private_data; |
| 294 | if (fd_file(f)->f_op != &signalfd_fops) |
| 295 | return -EINVAL; |
| 296 | spin_lock_irq(lock: ¤t->sighand->siglock); |
| 297 | ctx->sigmask = *mask; |
| 298 | spin_unlock_irq(lock: ¤t->sighand->siglock); |
| 299 | |
| 300 | wake_up(¤t->sighand->signalfd_wqh); |
| 301 | } |
| 302 | |
| 303 | return ufd; |
| 304 | } |
| 305 | |
| 306 | SYSCALL_DEFINE4(signalfd4, int, ufd, sigset_t __user *, user_mask, |
| 307 | size_t, sizemask, int, flags) |
| 308 | { |
| 309 | sigset_t mask; |
| 310 | |
| 311 | if (sizemask != sizeof(sigset_t)) |
| 312 | return -EINVAL; |
| 313 | if (copy_from_user(to: &mask, from: user_mask, n: sizeof(mask))) |
| 314 | return -EFAULT; |
| 315 | return do_signalfd4(ufd, mask: &mask, flags); |
| 316 | } |
| 317 | |
| 318 | SYSCALL_DEFINE3(signalfd, int, ufd, sigset_t __user *, user_mask, |
| 319 | size_t, sizemask) |
| 320 | { |
| 321 | sigset_t mask; |
| 322 | |
| 323 | if (sizemask != sizeof(sigset_t)) |
| 324 | return -EINVAL; |
| 325 | if (copy_from_user(to: &mask, from: user_mask, n: sizeof(mask))) |
| 326 | return -EFAULT; |
| 327 | return do_signalfd4(ufd, mask: &mask, flags: 0); |
| 328 | } |
| 329 | |
| 330 | #ifdef CONFIG_COMPAT |
| 331 | static long do_compat_signalfd4(int ufd, |
| 332 | const compat_sigset_t __user *user_mask, |
| 333 | compat_size_t sigsetsize, int flags) |
| 334 | { |
| 335 | sigset_t mask; |
| 336 | |
| 337 | if (sigsetsize != sizeof(compat_sigset_t)) |
| 338 | return -EINVAL; |
| 339 | if (get_compat_sigset(set: &mask, compat: user_mask)) |
| 340 | return -EFAULT; |
| 341 | return do_signalfd4(ufd, mask: &mask, flags); |
| 342 | } |
| 343 | |
| 344 | COMPAT_SYSCALL_DEFINE4(signalfd4, int, ufd, |
| 345 | const compat_sigset_t __user *, user_mask, |
| 346 | compat_size_t, sigsetsize, |
| 347 | int, flags) |
| 348 | { |
| 349 | return do_compat_signalfd4(ufd, user_mask, sigsetsize, flags); |
| 350 | } |
| 351 | |
| 352 | COMPAT_SYSCALL_DEFINE3(signalfd, int, ufd, |
| 353 | const compat_sigset_t __user *, user_mask, |
| 354 | compat_size_t, sigsetsize) |
| 355 | { |
| 356 | return do_compat_signalfd4(ufd, user_mask, sigsetsize, flags: 0); |
| 357 | } |
| 358 | #endif |
| 359 | |