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 void signalfd_cleanup(struct sighand_struct *sighand)
  33 {
  34         wait_queue_head_t *wqh = &sighand->signalfd_wqh;
  35         /*
  36          * The lockless check can race with remove_wait_queue() in progress,
  37          * but in this case its caller should run under rcu_read_lock() and
  38          * sighand_cachep is SLAB_DESTROY_BY_RCU, we can safely return.
  39          */
  40         if (likely(!waitqueue_active(wqh)))
  41                 return;
  42
  43         /* wait_queue_t->func(POLLFREE) should do remove_wait_queue() */
  44         wake_up_poll(wqh, POLLHUP | POLLFREE);
  45 }
  46
  47 struct signalfd_ctx {
  48         sigset_t sigmask;
  49 };
  50
  51 static int signalfd_release(struct inode *inode, struct file *file)
  52 {
  53         kfree(file->private_data);
  54         return 0;
  55 }
  56
  57 static unsigned int signalfd_poll(struct file *file, poll_table *wait)
  58 {
  59         struct signalfd_ctx *ctx = file->private_data;
  60         unsigned int events = 0;
  61
  62         poll_wait(file, &current->sighand->signalfd_wqh, wait);
  63
  64         spin_lock_irq(&current->sighand->siglock);
  65         if (next_signal(&current->pending, &ctx->sigmask) ||
  66             next_signal(&current->signal->shared_pending,
  67                         &ctx->sigmask))
  68                 events |= POLLIN;
  69         spin_unlock_irq(&current->sighand->siglock);
  70
  71         return events;
  72 }
  73
  74 /*
  75  * Copied from copy_siginfo_to_user() in kernel/signal.c
  76  */
  77 static int signalfd_copyinfo(struct signalfd_siginfo __user *uinfo,
  78                              siginfo_t const *kinfo)
  79 {
  80         long err;
  81
  82         BUILD_BUG_ON(sizeof(struct signalfd_siginfo) != 128);
  83
  84         /*
  85          * Unused members should be zero ...
  86          */
  87         err = __clear_user(uinfo, sizeof(*uinfo));
  88
  89         /*
  90          * If you change siginfo_t structure, please be sure
  91          * this code is fixed accordingly.
  92          */
  93         err |= __put_user(kinfo->si_signo, &uinfo->ssi_signo);
  94         err |= __put_user(kinfo->si_errno, &uinfo->ssi_errno);
  95         err |= __put_user((short) kinfo->si_code, &uinfo->ssi_code);
  96         switch (kinfo->si_code & __SI_MASK) {
  97         case __SI_KILL:
  98                 err |= __put_user(kinfo->si_pid, &uinfo->ssi_pid);
  99                 err |= __put_user(kinfo->si_uid, &uinfo->ssi_uid);
 100                 break;
 101         case __SI_TIMER:
 102                  err |= __put_user(kinfo->si_tid, &uinfo->ssi_tid);
 103                  err |= __put_user(kinfo->si_overrun, &uinfo->ssi_overrun);
 104                  err |= __put_user((long) kinfo->si_ptr, &uinfo->ssi_ptr);
 105                  err |= __put_user(kinfo->si_int, &uinfo->ssi_int);
 106                 break;
 107         case __SI_POLL:
 108                 err |= __put_user(kinfo->si_band, &uinfo->ssi_band);
 109                 err |= __put_user(kinfo->si_fd, &uinfo->ssi_fd);
 110                 break;
 111         case __SI_FAULT:
 112                 err |= __put_user((long) kinfo->si_addr, &uinfo->ssi_addr);
 113 #ifdef __ARCH_SI_TRAPNO
 114                 err |= __put_user(kinfo->si_trapno, &uinfo->ssi_trapno);
 115 #endif
 116                 break;
 117         case __SI_CHLD:
 118                 err |= __put_user(kinfo->si_pid, &uinfo->ssi_pid);
 119                 err |= __put_user(kinfo->si_uid, &uinfo->ssi_uid);
 120                 err |= __put_user(kinfo->si_status, &uinfo->ssi_status);
 121                 err |= __put_user(kinfo->si_utime, &uinfo->ssi_utime);
 122                 err |= __put_user(kinfo->si_stime, &uinfo->ssi_stime);
 123                 break;
 124         case __SI_RT: /* This is not generated by the kernel as of now. */
 125         case __SI_MESGQ: /* But this is */
 126                 err |= __put_user(kinfo->si_pid, &uinfo->ssi_pid);
 127                 err |= __put_user(kinfo->si_uid, &uinfo->ssi_uid);
 128                 err |= __put_user((long) kinfo->si_ptr, &uinfo->ssi_ptr);
 129                 err |= __put_user(kinfo->si_int, &uinfo->ssi_int);
 130                 break;
 131         default:
 132                 /*
 133                  * This case catches also the signals queued by sigqueue().
 134                  */
 135                 err |= __put_user(kinfo->si_pid, &uinfo->ssi_pid);
 136                 err |= __put_user(kinfo->si_uid, &uinfo->ssi_uid);
 137                 err |= __put_user((long) kinfo->si_ptr, &uinfo->ssi_ptr);
 138                 err |= __put_user(kinfo->si_int, &uinfo->ssi_int);
 139                 break;
 140         }
 141
 142         return err ? -EFAULT: sizeof(*uinfo);
 143 }
 144
 145 static ssize_t signalfd_dequeue(struct signalfd_ctx *ctx, siginfo_t *info,
 146                                 int nonblock)
 147 {
 148         ssize_t ret;
 149         DECLARE_WAITQUEUE(wait, current);
 150
 151         spin_lock_irq(&current->sighand->siglock);
 152         ret = dequeue_signal(current, &ctx->sigmask, info);
 153         switch (ret) {
 154         case 0:
 155                 if (!nonblock)
 156                         break;
 157                 ret = -EAGAIN;
 158         default:
 159                 spin_unlock_irq(&current->sighand->siglock);
 160                 return ret;
 161         }
 162
 163         add_wait_queue(&current->sighand->signalfd_wqh, &wait);
 164         for (;;) {
 165                 set_current_state(TASK_INTERRUPTIBLE);
 166                 ret = dequeue_signal(current, &ctx->sigmask, info);
 167                 if (ret != 0)
 168                         break;
 169                 if (signal_pending(current)) {
 170                         ret = -ERESTARTSYS;
 171                         break;
 172                 }
 173                 spin_unlock_irq(&current->sighand->siglock);
 174                 schedule();
 175                 spin_lock_irq(&current->sighand->siglock);
 176         }
 177         spin_unlock_irq(&current->sighand->siglock);
 178
 179         remove_wait_queue(&current->sighand->signalfd_wqh, &wait);
 180         __set_current_state(TASK_RUNNING);
 181
 182         return ret;
 183 }
 184
 185 /*
 186  * Returns a multiple of the size of a "struct signalfd_siginfo", or a negative
 187  * error code. The "count" parameter must be at least the size of a
 188  * "struct signalfd_siginfo".
 189  */
 190 static ssize_t signalfd_read(struct file *file, char __user *buf, size_t count,
 191                              loff_t *ppos)
 192 {
 193         struct signalfd_ctx *ctx = file->private_data;
 194         struct signalfd_siginfo __user *siginfo;
 195         int nonblock = file->f_flags & O_NONBLOCK;
 196         ssize_t ret, total = 0;
 197         siginfo_t info;
 198
 199         count /= sizeof(struct signalfd_siginfo);
 200         if (!count)
 201                 return -EINVAL;
 202
 203         siginfo = (struct signalfd_siginfo __user *) buf;
 204         do {
 205                 ret = signalfd_dequeue(ctx, &info, nonblock);
 206                 if (unlikely(ret <= 0))
 207                         break;
 208                 ret = signalfd_copyinfo(siginfo, &info);
 209                 if (ret < 0)
 210                         break;
 211                 siginfo++;
 212                 total += ret;
 213                 nonblock = 1;
 214         } while (--count);
 215
 216         return total ? total: ret;
 217 }
 218
 219 static const struct file_operations signalfd_fops = {
 220         .release        = signalfd_release,
 221         .poll           = signalfd_poll,
 222         .read           = signalfd_read,
 223 };
 224
 225 SYSCALL_DEFINE4(signalfd4, int, ufd, sigset_t __user *, user_mask,
 226                 size_t, sizemask, int, flags)
 227 {
 228         sigset_t sigmask;
 229         struct signalfd_ctx *ctx;
 230
 231         /* Check the SFD_* constants for consistency.  */
 232         BUILD_BUG_ON(SFD_CLOEXEC != O_CLOEXEC);
 233         BUILD_BUG_ON(SFD_NONBLOCK != O_NONBLOCK);
 234
 235         if (flags & ~(SFD_CLOEXEC | SFD_NONBLOCK))
 236                 return -EINVAL;
 237
 238         if (sizemask != sizeof(sigset_t) ||
 239             copy_from_user(&sigmask, user_mask, sizeof(sigmask)))
 240                 return -EINVAL;
 241         sigdelsetmask(&sigmask, sigmask(SIGKILL) | sigmask(SIGSTOP));
 242         signotset(&sigmask);
 243
 244         if (ufd == -1) {
 245                 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
 246                 if (!ctx)
 247                         return -ENOMEM;
 248
 249                 ctx->sigmask = sigmask;
 250
 251                 /*
 252                  * When we call this, the initialization must be complete, since
 253                  * anon_inode_getfd() will install the fd.
 254                  */
 255                 ufd = anon_inode_getfd("[signalfd]", &signalfd_fops, ctx,
 256                                        flags & (O_CLOEXEC | O_NONBLOCK));
 257                 if (ufd < 0)
 258                         kfree(ctx);
 259         } else {
 260                 struct file *file = fget(ufd);
 261                 if (!file)
 262                         return -EBADF;
 263                 ctx = file->private_data;
 264                 if (file->f_op != &signalfd_fops) {
 265                         fput(file);
 266                         return -EINVAL;
 267                 }
 268                 spin_lock_irq(&current->sighand->siglock);
 269                 ctx->sigmask = sigmask;
 270                 spin_unlock_irq(&current->sighand->siglock);
 271
 272                 wake_up(&current->sighand->signalfd_wqh);
 273                 fput(file);
 274         }
 275
 276         return ufd;
 277 }
 278
 279 SYSCALL_DEFINE3(signalfd, int, ufd, sigset_t __user *, user_mask,
 280                 size_t, sizemask)
 281 {
 282         return sys_signalfd4(ufd, user_mask, sizemask, 0);
 283 }