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