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_PERF_EVENT: 138 new.ssi_addr = (long) kinfo->si_addr; 139 new.ssi_perf = kinfo->si_perf; 140 break; 141 case SIL_CHLD: 142 new.ssi_pid = kinfo->si_pid; 143 new.ssi_uid = kinfo->si_uid; 144 new.ssi_status = kinfo->si_status; 145 new.ssi_utime = kinfo->si_utime; 146 new.ssi_stime = kinfo->si_stime; 147 break; 148 case SIL_RT: 149 /* 150 * This case catches also the signals queued by sigqueue(). 151 */ 152 new.ssi_pid = kinfo->si_pid; 153 new.ssi_uid = kinfo->si_uid; 154 new.ssi_ptr = (long) kinfo->si_ptr; 155 new.ssi_int = kinfo->si_int; 156 break; 157 case SIL_SYS: 158 new.ssi_call_addr = (long) kinfo->si_call_addr; 159 new.ssi_syscall = kinfo->si_syscall; 160 new.ssi_arch = kinfo->si_arch; 161 break; 162 } 163 164 if (copy_to_user(uinfo, &new, sizeof(struct signalfd_siginfo))) 165 return -EFAULT; 166 167 return sizeof(*uinfo); 168 } 169 170 static ssize_t signalfd_dequeue(struct signalfd_ctx *ctx, kernel_siginfo_t *info, 171 int nonblock) 172 { 173 ssize_t ret; 174 DECLARE_WAITQUEUE(wait, current); 175 176 spin_lock_irq(¤t->sighand->siglock); 177 ret = dequeue_signal(current, &ctx->sigmask, info); 178 switch (ret) { 179 case 0: 180 if (!nonblock) 181 break; 182 ret = -EAGAIN; 183 fallthrough; 184 default: 185 spin_unlock_irq(¤t->sighand->siglock); 186 return ret; 187 } 188 189 add_wait_queue(¤t->sighand->signalfd_wqh, &wait); 190 for (;;) { 191 set_current_state(TASK_INTERRUPTIBLE); 192 ret = dequeue_signal(current, &ctx->sigmask, info); 193 if (ret != 0) 194 break; 195 if (signal_pending(current)) { 196 ret = -ERESTARTSYS; 197 break; 198 } 199 spin_unlock_irq(¤t->sighand->siglock); 200 schedule(); 201 spin_lock_irq(¤t->sighand->siglock); 202 } 203 spin_unlock_irq(¤t->sighand->siglock); 204 205 remove_wait_queue(¤t->sighand->signalfd_wqh, &wait); 206 __set_current_state(TASK_RUNNING); 207 208 return ret; 209 } 210 211 /* 212 * Returns a multiple of the size of a "struct signalfd_siginfo", or a negative 213 * error code. The "count" parameter must be at least the size of a 214 * "struct signalfd_siginfo". 215 */ 216 static ssize_t signalfd_read(struct file *file, char __user *buf, size_t count, 217 loff_t *ppos) 218 { 219 struct signalfd_ctx *ctx = file->private_data; 220 struct signalfd_siginfo __user *siginfo; 221 int nonblock = file->f_flags & O_NONBLOCK; 222 ssize_t ret, total = 0; 223 kernel_siginfo_t info; 224 225 count /= sizeof(struct signalfd_siginfo); 226 if (!count) 227 return -EINVAL; 228 229 siginfo = (struct signalfd_siginfo __user *) buf; 230 do { 231 ret = signalfd_dequeue(ctx, &info, nonblock); 232 if (unlikely(ret <= 0)) 233 break; 234 ret = signalfd_copyinfo(siginfo, &info); 235 if (ret < 0) 236 break; 237 siginfo++; 238 total += ret; 239 nonblock = 1; 240 } while (--count); 241 242 return total ? total: ret; 243 } 244 245 #ifdef CONFIG_PROC_FS 246 static void signalfd_show_fdinfo(struct seq_file *m, struct file *f) 247 { 248 struct signalfd_ctx *ctx = f->private_data; 249 sigset_t sigmask; 250 251 sigmask = ctx->sigmask; 252 signotset(&sigmask); 253 render_sigset_t(m, "sigmask:\t", &sigmask); 254 } 255 #endif 256 257 static const struct file_operations signalfd_fops = { 258 #ifdef CONFIG_PROC_FS 259 .show_fdinfo = signalfd_show_fdinfo, 260 #endif 261 .release = signalfd_release, 262 .poll = signalfd_poll, 263 .read = signalfd_read, 264 .llseek = noop_llseek, 265 }; 266 267 static int do_signalfd4(int ufd, sigset_t *mask, int flags) 268 { 269 struct signalfd_ctx *ctx; 270 271 /* Check the SFD_* constants for consistency. */ 272 BUILD_BUG_ON(SFD_CLOEXEC != O_CLOEXEC); 273 BUILD_BUG_ON(SFD_NONBLOCK != O_NONBLOCK); 274 275 if (flags & ~(SFD_CLOEXEC | SFD_NONBLOCK)) 276 return -EINVAL; 277 278 sigdelsetmask(mask, sigmask(SIGKILL) | sigmask(SIGSTOP)); 279 signotset(mask); 280 281 if (ufd == -1) { 282 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL); 283 if (!ctx) 284 return -ENOMEM; 285 286 ctx->sigmask = *mask; 287 288 /* 289 * When we call this, the initialization must be complete, since 290 * anon_inode_getfd() will install the fd. 291 */ 292 ufd = anon_inode_getfd("[signalfd]", &signalfd_fops, ctx, 293 O_RDWR | (flags & (O_CLOEXEC | O_NONBLOCK))); 294 if (ufd < 0) 295 kfree(ctx); 296 } else { 297 struct fd f = fdget(ufd); 298 if (!f.file) 299 return -EBADF; 300 ctx = f.file->private_data; 301 if (f.file->f_op != &signalfd_fops) { 302 fdput(f); 303 return -EINVAL; 304 } 305 spin_lock_irq(¤t->sighand->siglock); 306 ctx->sigmask = *mask; 307 spin_unlock_irq(¤t->sighand->siglock); 308 309 wake_up(¤t->sighand->signalfd_wqh); 310 fdput(f); 311 } 312 313 return ufd; 314 } 315 316 SYSCALL_DEFINE4(signalfd4, int, ufd, sigset_t __user *, user_mask, 317 size_t, sizemask, int, flags) 318 { 319 sigset_t mask; 320 321 if (sizemask != sizeof(sigset_t)) 322 return -EINVAL; 323 if (copy_from_user(&mask, user_mask, sizeof(mask))) 324 return -EFAULT; 325 return do_signalfd4(ufd, &mask, flags); 326 } 327 328 SYSCALL_DEFINE3(signalfd, int, ufd, sigset_t __user *, user_mask, 329 size_t, sizemask) 330 { 331 sigset_t mask; 332 333 if (sizemask != sizeof(sigset_t)) 334 return -EINVAL; 335 if (copy_from_user(&mask, user_mask, sizeof(mask))) 336 return -EFAULT; 337 return do_signalfd4(ufd, &mask, 0); 338 } 339 340 #ifdef CONFIG_COMPAT 341 static long do_compat_signalfd4(int ufd, 342 const compat_sigset_t __user *user_mask, 343 compat_size_t sigsetsize, int flags) 344 { 345 sigset_t mask; 346 347 if (sigsetsize != sizeof(compat_sigset_t)) 348 return -EINVAL; 349 if (get_compat_sigset(&mask, user_mask)) 350 return -EFAULT; 351 return do_signalfd4(ufd, &mask, flags); 352 } 353 354 COMPAT_SYSCALL_DEFINE4(signalfd4, int, ufd, 355 const compat_sigset_t __user *, user_mask, 356 compat_size_t, sigsetsize, 357 int, flags) 358 { 359 return do_compat_signalfd4(ufd, user_mask, sigsetsize, flags); 360 } 361 362 COMPAT_SYSCALL_DEFINE3(signalfd, int, ufd, 363 const compat_sigset_t __user *, user_mask, 364 compat_size_t, sigsetsize) 365 { 366 return do_compat_signalfd4(ufd, user_mask, sigsetsize, 0); 367 } 368 #endif 369