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/kernel.h>
  26 #include <linux/signal.h>
  27 #include <linux/list.h>
  28 #include <linux/anon_inodes.h>
  29 #include <linux/signalfd.h>
  30 #include <linux/syscalls.h>
  31
  32 struct signalfd_ctx {
  33         sigset_t sigmask;
  34 };
  35
  36 static int signalfd_release(struct inode *inode, struct file *file)
  37 {
  38         kfree(file->private_data);
  39         return 0;
  40 }
  41
  42 static unsigned int signalfd_poll(struct file *file, poll_table *wait)
  43 {
  44         struct signalfd_ctx *ctx = file->private_data;
  45         unsigned int events = 0;
  46
  47         poll_wait(file, &current->sighand->signalfd_wqh, wait);
  48
  49         spin_lock_irq(&current->sighand->siglock);
  50         if (next_signal(&current->pending, &ctx->sigmask) ||
  51             next_signal(&current->signal->shared_pending,
  52                         &ctx->sigmask))
  53                 events |= POLLIN;
  54         spin_unlock_irq(&current->sighand->siglock);
  55
  56         return events;
  57 }
  58
  59 /*
  60  * Copied from copy_siginfo_to_user() in kernel/signal.c
  61  */
  62 static int signalfd_copyinfo(struct signalfd_siginfo __user *uinfo,
  63                              siginfo_t const *kinfo)
  64 {
  65         long err;
  66
  67         BUILD_BUG_ON(sizeof(struct signalfd_siginfo) != 128);
  68
  69         /*
  70          * Unused members should be zero ...
  71          */
  72         err = __clear_user(uinfo, sizeof(*uinfo));
  73
  74         /*
  75          * If you change siginfo_t structure, please be sure
  76          * this code is fixed accordingly.
  77          */
  78         err |= __put_user(kinfo->si_signo, &uinfo->ssi_signo);
  79         err |= __put_user(kinfo->si_errno, &uinfo->ssi_errno);
  80         err |= __put_user((short) kinfo->si_code, &uinfo->ssi_code);
  81         switch (kinfo->si_code & __SI_MASK) {
  82         case __SI_KILL:
  83                 err |= __put_user(kinfo->si_pid, &uinfo->ssi_pid);
  84                 err |= __put_user(kinfo->si_uid, &uinfo->ssi_uid);
  85                 break;
  86         case __SI_TIMER:
  87                  err |= __put_user(kinfo->si_tid, &uinfo->ssi_tid);
  88                  err |= __put_user(kinfo->si_overrun, &uinfo->ssi_overrun);
  89                  err |= __put_user((long) kinfo->si_ptr, &uinfo->ssi_ptr);
  90                  err |= __put_user(kinfo->si_int, &uinfo->ssi_int);
  91                 break;
  92         case __SI_POLL:
  93                 err |= __put_user(kinfo->si_band, &uinfo->ssi_band);
  94                 err |= __put_user(kinfo->si_fd, &uinfo->ssi_fd);
  95                 break;
  96         case __SI_FAULT:
  97                 err |= __put_user((long) kinfo->si_addr, &uinfo->ssi_addr);
  98 #ifdef __ARCH_SI_TRAPNO
  99                 err |= __put_user(kinfo->si_trapno, &uinfo->ssi_trapno);
 100 #endif
 101                 break;
 102         case __SI_CHLD:
 103                 err |= __put_user(kinfo->si_pid, &uinfo->ssi_pid);
 104                 err |= __put_user(kinfo->si_uid, &uinfo->ssi_uid);
 105                 err |= __put_user(kinfo->si_status, &uinfo->ssi_status);
 106                 err |= __put_user(kinfo->si_utime, &uinfo->ssi_utime);
 107                 err |= __put_user(kinfo->si_stime, &uinfo->ssi_stime);
 108                 break;
 109         case __SI_RT: /* This is not generated by the kernel as of now. */
 110         case __SI_MESGQ: /* But this is */
 111                 err |= __put_user(kinfo->si_pid, &uinfo->ssi_pid);
 112                 err |= __put_user(kinfo->si_uid, &uinfo->ssi_uid);
 113                 err |= __put_user((long) kinfo->si_ptr, &uinfo->ssi_ptr);
 114                 err |= __put_user(kinfo->si_int, &uinfo->ssi_int);
 115                 break;
 116         default:
 117                 /*
 118                  * This case catches also the signals queued by sigqueue().
 119                  */
 120                 err |= __put_user(kinfo->si_pid, &uinfo->ssi_pid);
 121                 err |= __put_user(kinfo->si_uid, &uinfo->ssi_uid);
 122                 err |= __put_user((long) kinfo->si_ptr, &uinfo->ssi_ptr);
 123                 err |= __put_user(kinfo->si_int, &uinfo->ssi_int);
 124                 break;
 125         }
 126
 127         return err ? -EFAULT: sizeof(*uinfo);
 128 }
 129
 130 static ssize_t signalfd_dequeue(struct signalfd_ctx *ctx, siginfo_t *info,
 131                                 int nonblock)
 132 {
 133         ssize_t ret;
 134         DECLARE_WAITQUEUE(wait, current);
 135
 136         spin_lock_irq(&current->sighand->siglock);
 137         ret = dequeue_signal(current, &ctx->sigmask, info);
 138         switch (ret) {
 139         case 0:
 140                 if (!nonblock)
 141                         break;
 142                 ret = -EAGAIN;
 143         default:
 144                 spin_unlock_irq(&current->sighand->siglock);
 145                 return ret;
 146         }
 147
 148         add_wait_queue(&current->sighand->signalfd_wqh, &wait);
 149         for (;;) {
 150                 set_current_state(TASK_INTERRUPTIBLE);
 151                 ret = dequeue_signal(current, &ctx->sigmask, info);
 152                 if (ret != 0)
 153                         break;
 154                 if (signal_pending(current)) {
 155                         ret = -ERESTARTSYS;
 156                         break;
 157                 }
 158                 spin_unlock_irq(&current->sighand->siglock);
 159                 schedule();
 160                 spin_lock_irq(&current->sighand->siglock);
 161         }
 162         spin_unlock_irq(&current->sighand->siglock);
 163
 164         remove_wait_queue(&current->sighand->signalfd_wqh, &wait);
 165         __set_current_state(TASK_RUNNING);
 166
 167         return ret;
 168 }
 169
 170 /*
 171  * Returns a multiple of the size of a "struct signalfd_siginfo", or a negative
 172  * error code. The "count" parameter must be at least the size of a
 173  * "struct signalfd_siginfo".
 174  */
 175 static ssize_t signalfd_read(struct file *file, char __user *buf, size_t count,
 176                              loff_t *ppos)
 177 {
 178         struct signalfd_ctx *ctx = file->private_data;
 179         struct signalfd_siginfo __user *siginfo;
 180         int nonblock = file->f_flags & O_NONBLOCK;
 181         ssize_t ret, total = 0;
 182         siginfo_t info;
 183
 184         count /= sizeof(struct signalfd_siginfo);
 185         if (!count)
 186                 return -EINVAL;
 187
 188         siginfo = (struct signalfd_siginfo __user *) buf;
 189         do {
 190                 ret = signalfd_dequeue(ctx, &info, nonblock);
 191                 if (unlikely(ret <= 0))
 192                         break;
 193                 ret = signalfd_copyinfo(siginfo, &info);
 194                 if (ret < 0)
 195                         break;
 196                 siginfo++;
 197                 total += ret;
 198                 nonblock = 1;
 199         } while (--count);
 200
 201         return total ? total: ret;
 202 }
 203
 204 static const struct file_operations signalfd_fops = {
 205         .release        = signalfd_release,
 206         .poll           = signalfd_poll,
 207         .read           = signalfd_read,
 208 };
 209
 210 SYSCALL_DEFINE4(signalfd4, int, ufd, sigset_t __user *, user_mask,
 211                 size_t, sizemask, int, flags)
 212 {
 213         sigset_t sigmask;
 214         struct signalfd_ctx *ctx;
 215
 216         /* Check the SFD_* constants for consistency.  */
 217         BUILD_BUG_ON(SFD_CLOEXEC != O_CLOEXEC);
 218         BUILD_BUG_ON(SFD_NONBLOCK != O_NONBLOCK);
 219
 220         if (flags & ~(SFD_CLOEXEC | SFD_NONBLOCK))
 221                 return -EINVAL;
 222
 223         if (sizemask != sizeof(sigset_t) ||
 224             copy_from_user(&sigmask, user_mask, sizeof(sigmask)))
 225                 return -EINVAL;
 226         sigdelsetmask(&sigmask, sigmask(SIGKILL) | sigmask(SIGSTOP));
 227         signotset(&sigmask);
 228
 229         if (ufd == -1) {
 230                 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
 231                 if (!ctx)
 232                         return -ENOMEM;
 233
 234                 ctx->sigmask = sigmask;
 235
 236                 /*
 237                  * When we call this, the initialization must be complete, since
 238                  * anon_inode_getfd() will install the fd.
 239                  */
 240                 ufd = anon_inode_getfd("[signalfd]", &signalfd_fops, ctx,
 241                                        flags & (O_CLOEXEC | O_NONBLOCK));
 242                 if (ufd < 0)
 243                         kfree(ctx);
 244         } else {
 245                 struct file *file = fget(ufd);
 246                 if (!file)
 247                         return -EBADF;
 248                 ctx = file->private_data;
 249                 if (file->f_op != &signalfd_fops) {
 250                         fput(file);
 251                         return -EINVAL;
 252                 }
 253                 spin_lock_irq(&current->sighand->siglock);
 254                 ctx->sigmask = sigmask;
 255                 spin_unlock_irq(&current->sighand->siglock);
 256
 257                 wake_up(&current->sighand->signalfd_wqh);
 258                 fput(file);
 259         }
 260
 261         return ufd;
 262 }
 263
 264 SYSCALL_DEFINE3(signalfd, int, ufd, sigset_t __user *, user_mask,
 265                 size_t, sizemask)
 266 {
 267         return sys_signalfd4(ufd, user_mask, sizemask, 0);
 268 }