fs/signalfd.c

   1 /*
   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.
  14  *  Sat May 19, 2007: Davi E. M. Arnaut <davi@haxent.com.br>
  15  *      Retrieve multiple signals with one read() call
  16  *  Sun Jul 15, 2007: Davide Libenzi <davidel@xmailserver.org>
  17  *      Attach to the sighand only during read() and poll().
  18  */
  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>
  25 #include <linux/slab.h>
  26 #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>
  31 #include <linux/syscalls.h>
  32 #include <linux/proc_fs.h>
  33 #include <linux/compat.h>
  34
  35 void signalfd_cleanup(struct sighand_struct *sighand)
  36 {
  37         wait_queue_head_t *wqh = &sighand->signalfd_wqh;
  38         /*
  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_TYPESAFE_BY_RCU, we can safely return.
  42          */
  43         if (likely(!waitqueue_active(wqh)))
  44                 return;
  45
  46         /* wait_queue_entry_t->func(POLLFREE) should do remove_wait_queue() */
  47         wake_up_poll(wqh, POLLHUP | POLLFREE);
  48 }
  49
  50 struct signalfd_ctx {
  51         sigset_t sigmask;
  52 };
  53
  54 static int signalfd_release(struct inode *inode, struct file *file)
  55 {
  56         kfree(file->private_data);
  57         return 0;
  58 }
  59
  60 static unsigned int signalfd_poll(struct file *file, poll_table *wait)
  61 {
  62         struct signalfd_ctx *ctx = file->private_data;
  63         unsigned int events = 0;
  64
  65         poll_wait(file, &current->sighand->signalfd_wqh, wait);
  66
  67         spin_lock_irq(&current->sighand->siglock);
  68         if (next_signal(&current->pending, &ctx->sigmask) ||
  69             next_signal(&current->signal->shared_pending,
  70                         &ctx->sigmask))
  71                 events |= POLLIN;
  72         spin_unlock_irq(&current->sighand->siglock);
  73
  74         return events;
  75 }
  76
  77 /*
  78  * Copied from copy_siginfo_to_user() in kernel/signal.c
  79  */
  80 static 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         /*
  88          * Unused members should be zero ...
  89          */
  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          */
  96         err |= __put_user(kinfo->si_signo, &uinfo->ssi_signo);
  97         err |= __put_user(kinfo->si_errno, &uinfo->ssi_errno);
  98         err |= __put_user(kinfo->si_code, &uinfo->ssi_code);
  99         switch (siginfo_layout(kinfo->si_signo, kinfo->si_code)) {
 100         case SIL_KILL:
 101                 err |= __put_user(kinfo->si_pid, &uinfo->ssi_pid);
 102                 err |= __put_user(kinfo->si_uid, &uinfo->ssi_uid);
 103                 break;
 104         case SIL_TIMER:
 105                  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);
 108                  err |= __put_user(kinfo->si_int, &uinfo->ssi_int);
 109                 break;
 110         case SIL_POLL:
 111                 err |= __put_user(kinfo->si_band, &uinfo->ssi_band);
 112                 err |= __put_user(kinfo->si_fd, &uinfo->ssi_fd);
 113                 break;
 114         case SIL_FAULT:
 115                 err |= __put_user((long) kinfo->si_addr, &uinfo->ssi_addr);
 116 #ifdef __ARCH_SI_TRAPNO
 117                 err |= __put_user(kinfo->si_trapno, &uinfo->ssi_trapno);
 118 #endif
 119 #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_signo == SIGBUS &&
 125                     (kinfo->si_code == BUS_MCEERR_AR ||
 126                      kinfo->si_code == BUS_MCEERR_AO))
 127                         err |= __put_user((short) kinfo->si_addr_lsb,
 128                                           &uinfo->ssi_addr_lsb);
 129 #endif
 130                 break;
 131         case SIL_CHLD:
 132                 err |= __put_user(kinfo->si_pid, &uinfo->ssi_pid);
 133                 err |= __put_user(kinfo->si_uid, &uinfo->ssi_uid);
 134                 err |= __put_user(kinfo->si_status, &uinfo->ssi_status);
 135                 err |= __put_user(kinfo->si_utime, &uinfo->ssi_utime);
 136                 err |= __put_user(kinfo->si_stime, &uinfo->ssi_stime);
 137                 break;
 138         case SIL_RT:
 139         default:
 140                 /*
 141                  * This case catches also the signals queued by sigqueue().
 142                  */
 143                 err |= __put_user(kinfo->si_pid, &uinfo->ssi_pid);
 144                 err |= __put_user(kinfo->si_uid, &uinfo->ssi_uid);
 145                 err |= __put_user((long) kinfo->si_ptr, &uinfo->ssi_ptr);
 146                 err |= __put_user(kinfo->si_int, &uinfo->ssi_int);
 147                 break;
 148         }
 149
 150         return err ? -EFAULT: sizeof(*uinfo);
 151 }
 152
 153 static ssize_t signalfd_dequeue(struct signalfd_ctx *ctx, siginfo_t *info,
 154                                 int nonblock)
 155 {
 156         ssize_t ret;
 157         DECLARE_WAITQUEUE(wait, current);
 158
 159         spin_lock_irq(&current->sighand->siglock);
 160         ret = dequeue_signal(current, &ctx->sigmask, info);
 161         switch (ret) {
 162         case 0:
 163                 if (!nonblock)
 164                         break;
 165                 ret = -EAGAIN;
 166         default:
 167                 spin_unlock_irq(&current->sighand->siglock);
 168                 return ret;
 169         }
 170
 171         add_wait_queue(&current->sighand->signalfd_wqh, &wait);
 172         for (;;) {
 173                 set_current_state(TASK_INTERRUPTIBLE);
 174                 ret = dequeue_signal(current, &ctx->sigmask, info);
 175                 if (ret != 0)
 176                         break;
 177                 if (signal_pending(current)) {
 178                         ret = -ERESTARTSYS;
 179                         break;
 180                 }
 181                 spin_unlock_irq(&current->sighand->siglock);
 182                 schedule();
 183                 spin_lock_irq(&current->sighand->siglock);
 184         }
 185         spin_unlock_irq(&current->sighand->siglock);
 186
 187         remove_wait_queue(&current->sighand->signalfd_wqh, &wait);
 188         __set_current_state(TASK_RUNNING);
 189
 190         return ret;
 191 }
 192
 193 /*
 194  * Returns a multiple of the size of a "struct signalfd_siginfo", or a negative
 195  * error code. The "count" parameter must be at least the size of a
 196  * "struct signalfd_siginfo".
 197  */
 198 static ssize_t signalfd_read(struct file *file, char __user *buf, size_t count,
 199                              loff_t *ppos)
 200 {
 201         struct signalfd_ctx *ctx = file->private_data;
 202         struct signalfd_siginfo __user *siginfo;
 203         int nonblock = file->f_flags & O_NONBLOCK;
 204         ssize_t ret, total = 0;
 205         siginfo_t info;
 206
 207         count /= sizeof(struct signalfd_siginfo);
 208         if (!count)
 209                 return -EINVAL;
 210
 211         siginfo = (struct signalfd_siginfo __user *) buf;
 212         do {
 213                 ret = signalfd_dequeue(ctx, &info, nonblock);
 214                 if (unlikely(ret <= 0))
 215                         break;
 216                 ret = signalfd_copyinfo(siginfo, &info);
 217                 if (ret < 0)
 218                         break;
 219                 siginfo++;
 220                 total += ret;
 221                 nonblock = 1;
 222         } while (--count);
 223
 224         return total ? total: ret;
 225 }
 226
 227 #ifdef CONFIG_PROC_FS
 228 static void signalfd_show_fdinfo(struct seq_file *m, struct file *f)
 229 {
 230         struct signalfd_ctx *ctx = f->private_data;
 231         sigset_t sigmask;
 232
 233         sigmask = ctx->sigmask;
 234         signotset(&sigmask);
 235         render_sigset_t(m, "sigmask:\t", &sigmask);
 236 }
 237 #endif
 238
 239 static const struct file_operations signalfd_fops = {
 240 #ifdef CONFIG_PROC_FS
 241         .show_fdinfo    = signalfd_show_fdinfo,
 242 #endif
 243         .release        = signalfd_release,
 244         .poll           = signalfd_poll,
 245         .read           = signalfd_read,
 246         .llseek         = noop_llseek,
 247 };
 248
 249 SYSCALL_DEFINE4(signalfd4, int, ufd, sigset_t __user *, user_mask,
 250                 size_t, sizemask, int, flags)
 251 {
 252         sigset_t sigmask;
 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         if (sizemask != sizeof(sigset_t) ||
 263             copy_from_user(&sigmask, user_mask, sizeof(sigmask)))
 264                 return -EINVAL;
 265         sigdelsetmask(&sigmask, sigmask(SIGKILL) | sigmask(SIGSTOP));
 266         signotset(&sigmask);
 267
 268         if (ufd == -1) {
 269                 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
 270                 if (!ctx)
 271                         return -ENOMEM;
 272
 273                 ctx->sigmask = sigmask;
 274
 275                 /*
 276                  * When we call this, the initialization must be complete, since
 277                  * anon_inode_getfd() will install the fd.
 278                  */
 279                 ufd = anon_inode_getfd("[signalfd]", &signalfd_fops, ctx,
 280                                        O_RDWR | (flags & (O_CLOEXEC | O_NONBLOCK)));
 281                 if (ufd < 0)
 282                         kfree(ctx);
 283         } else {
 284                 struct fd f = fdget(ufd);
 285                 if (!f.file)
 286                         return -EBADF;
 287                 ctx = f.file->private_data;
 288                 if (f.file->f_op != &signalfd_fops) {
 289                         fdput(f);
 290                         return -EINVAL;
 291                 }
 292                 spin_lock_irq(&current->sighand->siglock);
 293                 ctx->sigmask = sigmask;
 294                 spin_unlock_irq(&current->sighand->siglock);
 295
 296                 wake_up(&current->sighand->signalfd_wqh);
 297                 fdput(f);
 298         }
 299
 300         return ufd;
 301 }
 302
 303 SYSCALL_DEFINE3(signalfd, int, ufd, sigset_t __user *, user_mask,
 304                 size_t, sizemask)
 305 {
 306         return sys_signalfd4(ufd, user_mask, sizemask, 0);
 307 }
 308
 309 #ifdef CONFIG_COMPAT
 310 COMPAT_SYSCALL_DEFINE4(signalfd4, int, ufd,
 311                      const compat_sigset_t __user *,sigmask,
 312                      compat_size_t, sigsetsize,
 313                      int, flags)
 314 {
 315         compat_sigset_t ss32;
 316         sigset_t tmp;
 317         sigset_t __user *ksigmask;
 318
 319         if (sigsetsize != sizeof(compat_sigset_t))
 320                 return -EINVAL;
 321         if (copy_from_user(&ss32, sigmask, sizeof(ss32)))
 322                 return -EFAULT;
 323         sigset_from_compat(&tmp, &ss32);
 324         ksigmask = compat_alloc_user_space(sizeof(sigset_t));
 325         if (copy_to_user(ksigmask, &tmp, sizeof(sigset_t)))
 326                 return -EFAULT;
 327
 328         return sys_signalfd4(ufd, ksigmask, sizeof(sigset_t), flags);
 329 }
 330
 331 COMPAT_SYSCALL_DEFINE3(signalfd, int, ufd,
 332                      const compat_sigset_t __user *,sigmask,
 333                      compat_size_t, sigsetsize)
 334 {
 335         return compat_sys_signalfd4(ufd, sigmask, sigsetsize, 0);
 336 }
 337 #endif