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
  33 void signalfd_cleanup(struct sighand_struct *sighand)
  34 {
  35         wait_queue_head_t *wqh = &sighand->signalfd_wqh;
  36         /*
  37          * The lockless check can race with remove_wait_queue() in progress,
  38          * but in this case its caller should run under rcu_read_lock() and
  39          * sighand_cachep is SLAB_DESTROY_BY_RCU, we can safely return.
  40          */
  41         if (likely(!waitqueue_active(wqh)))
  42                 return;
  43
  44         /* wait_queue_t->func(POLLFREE) should do remove_wait_queue() */
  45         wake_up_poll(wqh, POLLHUP | POLLFREE);
  46 }
  47
  48 struct signalfd_ctx {
  49         sigset_t sigmask;
  50 };
  51
  52 static int signalfd_release(struct inode *inode, struct file *file)
  53 {
  54         kfree(file->private_data);
  55         return 0;
  56 }
  57
  58 static unsigned int signalfd_poll(struct file *file, poll_table *wait)
  59 {
  60         struct signalfd_ctx *ctx = file->private_data;
  61         unsigned int events = 0;
  62
  63         poll_wait(file, &current->sighand->signalfd_wqh, wait);
  64
  65         spin_lock_irq(&current->sighand->siglock);
  66         if (next_signal(&current->pending, &ctx->sigmask) ||
  67             next_signal(&current->signal->shared_pending,
  68                         &ctx->sigmask))
  69                 events |= POLLIN;
  70         spin_unlock_irq(&current->sighand->siglock);
  71
  72         return events;
  73 }
  74
  75 /*
  76  * Copied from copy_siginfo_to_user() in kernel/signal.c
  77  */
  78 static int signalfd_copyinfo(struct signalfd_siginfo __user *uinfo,
  79                              siginfo_t const *kinfo)
  80 {
  81         long err;
  82
  83         BUILD_BUG_ON(sizeof(struct signalfd_siginfo) != 128);
  84
  85         /*
  86          * Unused members should be zero ...
  87          */
  88         err = __clear_user(uinfo, sizeof(*uinfo));
  89
  90         /*
  91          * If you change siginfo_t structure, please be sure
  92          * this code is fixed accordingly.
  93          */
  94         err |= __put_user(kinfo->si_signo, &uinfo->ssi_signo);
  95         err |= __put_user(kinfo->si_errno, &uinfo->ssi_errno);
  96         err |= __put_user((short) kinfo->si_code, &uinfo->ssi_code);
  97         switch (kinfo->si_code & __SI_MASK) {
  98         case __SI_KILL:
  99                 err |= __put_user(kinfo->si_pid, &uinfo->ssi_pid);
 100                 err |= __put_user(kinfo->si_uid, &uinfo->ssi_uid);
 101                 break;
 102         case __SI_TIMER:
 103                  err |= __put_user(kinfo->si_tid, &uinfo->ssi_tid);
 104                  err |= __put_user(kinfo->si_overrun, &uinfo->ssi_overrun);
 105                  err |= __put_user((long) kinfo->si_ptr, &uinfo->ssi_ptr);
 106                  err |= __put_user(kinfo->si_int, &uinfo->ssi_int);
 107                 break;
 108         case __SI_POLL:
 109                 err |= __put_user(kinfo->si_band, &uinfo->ssi_band);
 110                 err |= __put_user(kinfo->si_fd, &uinfo->ssi_fd);
 111                 break;
 112         case __SI_FAULT:
 113                 err |= __put_user((long) kinfo->si_addr, &uinfo->ssi_addr);
 114 #ifdef __ARCH_SI_TRAPNO
 115                 err |= __put_user(kinfo->si_trapno, &uinfo->ssi_trapno);
 116 #endif
 117 #ifdef BUS_MCEERR_AO
 118                 /*
 119                  * Other callers might not initialize the si_lsb field,
 120                  * so check explicitly for the right codes here.
 121                  */
 122                 if (kinfo->si_code == BUS_MCEERR_AR ||
 123                     kinfo->si_code == BUS_MCEERR_AO)
 124                         err |= __put_user((short) kinfo->si_addr_lsb,
 125                                           &uinfo->ssi_addr_lsb);
 126 #endif
 127                 break;
 128         case __SI_CHLD:
 129                 err |= __put_user(kinfo->si_pid, &uinfo->ssi_pid);
 130                 err |= __put_user(kinfo->si_uid, &uinfo->ssi_uid);
 131                 err |= __put_user(kinfo->si_status, &uinfo->ssi_status);
 132                 err |= __put_user(kinfo->si_utime, &uinfo->ssi_utime);
 133                 err |= __put_user(kinfo->si_stime, &uinfo->ssi_stime);
 134                 break;
 135         case __SI_RT: /* This is not generated by the kernel as of now. */
 136         case __SI_MESGQ: /* But this is */
 137                 err |= __put_user(kinfo->si_pid, &uinfo->ssi_pid);
 138                 err |= __put_user(kinfo->si_uid, &uinfo->ssi_uid);
 139                 err |= __put_user((long) kinfo->si_ptr, &uinfo->ssi_ptr);
 140                 err |= __put_user(kinfo->si_int, &uinfo->ssi_int);
 141                 break;
 142         default:
 143                 /*
 144                  * This case catches also the signals queued by sigqueue().
 145                  */
 146                 err |= __put_user(kinfo->si_pid, &uinfo->ssi_pid);
 147                 err |= __put_user(kinfo->si_uid, &uinfo->ssi_uid);
 148                 err |= __put_user((long) kinfo->si_ptr, &uinfo->ssi_ptr);
 149                 err |= __put_user(kinfo->si_int, &uinfo->ssi_int);
 150                 break;
 151         }
 152
 153         return err ? -EFAULT: sizeof(*uinfo);
 154 }
 155
 156 static ssize_t signalfd_dequeue(struct signalfd_ctx *ctx, siginfo_t *info,
 157                                 int nonblock)
 158 {
 159         ssize_t ret;
 160         DECLARE_WAITQUEUE(wait, current);
 161
 162         spin_lock_irq(&current->sighand->siglock);
 163         ret = dequeue_signal(current, &ctx->sigmask, info);
 164         switch (ret) {
 165         case 0:
 166                 if (!nonblock)
 167                         break;
 168                 ret = -EAGAIN;
 169         default:
 170                 spin_unlock_irq(&current->sighand->siglock);
 171                 return ret;
 172         }
 173
 174         add_wait_queue(&current->sighand->signalfd_wqh, &wait);
 175         for (;;) {
 176                 set_current_state(TASK_INTERRUPTIBLE);
 177                 ret = dequeue_signal(current, &ctx->sigmask, info);
 178                 if (ret != 0)
 179                         break;
 180                 if (signal_pending(current)) {
 181                         ret = -ERESTARTSYS;
 182                         break;
 183                 }
 184                 spin_unlock_irq(&current->sighand->siglock);
 185                 schedule();
 186                 spin_lock_irq(&current->sighand->siglock);
 187         }
 188         spin_unlock_irq(&current->sighand->siglock);
 189
 190         remove_wait_queue(&current->sighand->signalfd_wqh, &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(struct file *file, char __user *buf, size_t count,
 202                              loff_t *ppos)
 203 {
 204         struct signalfd_ctx *ctx = file->private_data;
 205         struct signalfd_siginfo __user *siginfo;
 206         int nonblock = file->f_flags & O_NONBLOCK;
 207         ssize_t ret, total = 0;
 208         siginfo_t info;
 209
 210         count /= sizeof(struct signalfd_siginfo);
 211         if (!count)
 212                 return -EINVAL;
 213
 214         siginfo = (struct signalfd_siginfo __user *) buf;
 215         do {
 216                 ret = signalfd_dequeue(ctx, &info, nonblock);
 217                 if (unlikely(ret <= 0))
 218                         break;
 219                 ret = signalfd_copyinfo(siginfo, &info);
 220                 if (ret < 0)
 221                         break;
 222                 siginfo++;
 223                 total += ret;
 224                 nonblock = 1;
 225         } while (--count);
 226
 227         return total ? total: ret;
 228 }
 229
 230 static const struct file_operations signalfd_fops = {
 231         .release        = signalfd_release,
 232         .poll           = signalfd_poll,
 233         .read           = signalfd_read,
 234         .llseek         = noop_llseek,
 235 };
 236
 237 SYSCALL_DEFINE4(signalfd4, int, ufd, sigset_t __user *, user_mask,
 238                 size_t, sizemask, int, flags)
 239 {
 240         sigset_t sigmask;
 241         struct signalfd_ctx *ctx;
 242
 243         /* Check the SFD_* constants for consistency.  */
 244         BUILD_BUG_ON(SFD_CLOEXEC != O_CLOEXEC);
 245         BUILD_BUG_ON(SFD_NONBLOCK != O_NONBLOCK);
 246
 247         if (flags & ~(SFD_CLOEXEC | SFD_NONBLOCK))
 248                 return -EINVAL;
 249
 250         if (sizemask != sizeof(sigset_t) ||
 251             copy_from_user(&sigmask, user_mask, sizeof(sigmask)))
 252                 return -EINVAL;
 253         sigdelsetmask(&sigmask, sigmask(SIGKILL) | sigmask(SIGSTOP));
 254         signotset(&sigmask);
 255
 256         if (ufd == -1) {
 257                 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
 258                 if (!ctx)
 259                         return -ENOMEM;
 260
 261                 ctx->sigmask = sigmask;
 262
 263                 /*
 264                  * When we call this, the initialization must be complete, since
 265                  * anon_inode_getfd() will install the fd.
 266                  */
 267                 ufd = anon_inode_getfd("[signalfd]", &signalfd_fops, ctx,
 268                                        O_RDWR | (flags & (O_CLOEXEC | O_NONBLOCK)));
 269                 if (ufd < 0)
 270                         kfree(ctx);
 271         } else {
 272                 int fput_needed;
 273                 struct file *file = fget_light(ufd, &fput_needed);
 274                 if (!file)
 275                         return -EBADF;
 276                 ctx = file->private_data;
 277                 if (file->f_op != &signalfd_fops) {
 278                         fput_light(file, fput_needed);
 279                         return -EINVAL;
 280                 }
 281                 spin_lock_irq(&current->sighand->siglock);
 282                 ctx->sigmask = sigmask;
 283                 spin_unlock_irq(&current->sighand->siglock);
 284
 285                 wake_up(&current->sighand->signalfd_wqh);
 286                 fput_light(file, fput_needed);
 287         }
 288
 289         return ufd;
 290 }
 291
 292 SYSCALL_DEFINE3(signalfd, int, ufd, sigset_t __user *, user_mask,
 293                 size_t, sizemask)
 294 {
 295         return sys_signalfd4(ufd, user_mask, sizemask, 0);
 296 }