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_DESTROY_BY_RCU, we can safely return.
  42          */
  43         if (likely(!waitqueue_active(wqh)))
  44                 return;
  45
  46         /* wait_queue_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((short) kinfo->si_code, &uinfo->ssi_code);
  99         switch (kinfo->si_code & __SI_MASK) {
 100         case __SI_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 __SI_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 __SI_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 __SI_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_code == BUS_MCEERR_AR ||
 125                     kinfo->si_code == BUS_MCEERR_AO)
 126                         err |= __put_user((short) kinfo->si_addr_lsb,
 127                                           &uinfo->ssi_addr_lsb);
 128 #endif
 129                 break;
 130         case __SI_CHLD:
 131                 err |= __put_user(kinfo->si_pid, &uinfo->ssi_pid);
 132                 err |= __put_user(kinfo->si_uid, &uinfo->ssi_uid);
 133                 err |= __put_user(kinfo->si_status, &uinfo->ssi_status);
 134                 err |= __put_user(kinfo->si_utime, &uinfo->ssi_utime);
 135                 err |= __put_user(kinfo->si_stime, &uinfo->ssi_stime);
 136                 break;
 137         case __SI_RT: /* This is not generated by the kernel as of now. */
 138         case __SI_MESGQ: /* But this is */
 139                 err |= __put_user(kinfo->si_pid, &uinfo->ssi_pid);
 140                 err |= __put_user(kinfo->si_uid, &uinfo->ssi_uid);
 141                 err |= __put_user((long) kinfo->si_ptr, &uinfo->ssi_ptr);
 142                 err |= __put_user(kinfo->si_int, &uinfo->ssi_int);
 143                 break;
 144         default:
 145                 /*
 146                  * This case catches also the signals queued by sigqueue().
 147                  */
 148                 err |= __put_user(kinfo->si_pid, &uinfo->ssi_pid);
 149                 err |= __put_user(kinfo->si_uid, &uinfo->ssi_uid);
 150                 err |= __put_user((long) kinfo->si_ptr, &uinfo->ssi_ptr);
 151                 err |= __put_user(kinfo->si_int, &uinfo->ssi_int);
 152                 break;
 153         }
 154
 155         return err ? -EFAULT: sizeof(*uinfo);
 156 }
 157
 158 static ssize_t signalfd_dequeue(struct signalfd_ctx *ctx, siginfo_t *info,
 159                                 int nonblock)
 160 {
 161         ssize_t ret;
 162         DECLARE_WAITQUEUE(wait, current);
 163
 164         spin_lock_irq(&current->sighand->siglock);
 165         ret = dequeue_signal(current, &ctx->sigmask, info);
 166         switch (ret) {
 167         case 0:
 168                 if (!nonblock)
 169                         break;
 170                 ret = -EAGAIN;
 171         default:
 172                 spin_unlock_irq(&current->sighand->siglock);
 173                 return ret;
 174         }
 175
 176         add_wait_queue(&current->sighand->signalfd_wqh, &wait);
 177         for (;;) {
 178                 set_current_state(TASK_INTERRUPTIBLE);
 179                 ret = dequeue_signal(current, &ctx->sigmask, info);
 180                 if (ret != 0)
 181                         break;
 182                 if (signal_pending(current)) {
 183                         ret = -ERESTARTSYS;
 184                         break;
 185                 }
 186                 spin_unlock_irq(&current->sighand->siglock);
 187                 schedule();
 188                 spin_lock_irq(&current->sighand->siglock);
 189         }
 190         spin_unlock_irq(&current->sighand->siglock);
 191
 192         remove_wait_queue(&current->sighand->signalfd_wqh, &wait);
 193         __set_current_state(TASK_RUNNING);
 194
 195         return ret;
 196 }
 197
 198 /*
 199  * Returns a multiple of the size of a "struct signalfd_siginfo", or a negative
 200  * error code. The "count" parameter must be at least the size of a
 201  * "struct signalfd_siginfo".
 202  */
 203 static ssize_t signalfd_read(struct file *file, char __user *buf, size_t count,
 204                              loff_t *ppos)
 205 {
 206         struct signalfd_ctx *ctx = file->private_data;
 207         struct signalfd_siginfo __user *siginfo;
 208         int nonblock = file->f_flags & O_NONBLOCK;
 209         ssize_t ret, total = 0;
 210         siginfo_t info;
 211
 212         count /= sizeof(struct signalfd_siginfo);
 213         if (!count)
 214                 return -EINVAL;
 215
 216         siginfo = (struct signalfd_siginfo __user *) buf;
 217         do {
 218                 ret = signalfd_dequeue(ctx, &info, nonblock);
 219                 if (unlikely(ret <= 0))
 220                         break;
 221                 ret = signalfd_copyinfo(siginfo, &info);
 222                 if (ret < 0)
 223                         break;
 224                 siginfo++;
 225                 total += ret;
 226                 nonblock = 1;
 227         } while (--count);
 228
 229         return total ? total: ret;
 230 }
 231
 232 #ifdef CONFIG_PROC_FS
 233 static void signalfd_show_fdinfo(struct seq_file *m, struct file *f)
 234 {
 235         struct signalfd_ctx *ctx = f->private_data;
 236         sigset_t sigmask;
 237
 238         sigmask = ctx->sigmask;
 239         signotset(&sigmask);
 240         render_sigset_t(m, "sigmask:\t", &sigmask);
 241 }
 242 #endif
 243
 244 static const struct file_operations signalfd_fops = {
 245 #ifdef CONFIG_PROC_FS
 246         .show_fdinfo    = signalfd_show_fdinfo,
 247 #endif
 248         .release        = signalfd_release,
 249         .poll           = signalfd_poll,
 250         .read           = signalfd_read,
 251         .llseek         = noop_llseek,
 252 };
 253
 254 SYSCALL_DEFINE4(signalfd4, int, ufd, sigset_t __user *, user_mask,
 255                 size_t, sizemask, int, flags)
 256 {
 257         sigset_t sigmask;
 258         struct signalfd_ctx *ctx;
 259
 260         /* Check the SFD_* constants for consistency.  */
 261         BUILD_BUG_ON(SFD_CLOEXEC != O_CLOEXEC);
 262         BUILD_BUG_ON(SFD_NONBLOCK != O_NONBLOCK);
 263
 264         if (flags & ~(SFD_CLOEXEC | SFD_NONBLOCK))
 265                 return -EINVAL;
 266
 267         if (sizemask != sizeof(sigset_t) ||
 268             copy_from_user(&sigmask, user_mask, sizeof(sigmask)))
 269                 return -EINVAL;
 270         sigdelsetmask(&sigmask, sigmask(SIGKILL) | sigmask(SIGSTOP));
 271         signotset(&sigmask);
 272
 273         if (ufd == -1) {
 274                 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
 275                 if (!ctx)
 276                         return -ENOMEM;
 277
 278                 ctx->sigmask = sigmask;
 279
 280                 /*
 281                  * When we call this, the initialization must be complete, since
 282                  * anon_inode_getfd() will install the fd.
 283                  */
 284                 ufd = anon_inode_getfd("[signalfd]", &signalfd_fops, ctx,
 285                                        O_RDWR | (flags & (O_CLOEXEC | O_NONBLOCK)));
 286                 if (ufd < 0)
 287                         kfree(ctx);
 288         } else {
 289                 struct fd f = fdget(ufd);
 290                 if (!f.file)
 291                         return -EBADF;
 292                 ctx = f.file->private_data;
 293                 if (f.file->f_op != &signalfd_fops) {
 294                         fdput(f);
 295                         return -EINVAL;
 296                 }
 297                 spin_lock_irq(&current->sighand->siglock);
 298                 ctx->sigmask = sigmask;
 299                 spin_unlock_irq(&current->sighand->siglock);
 300
 301                 wake_up(&current->sighand->signalfd_wqh);
 302                 fdput(f);
 303         }
 304
 305         return ufd;
 306 }
 307
 308 SYSCALL_DEFINE3(signalfd, int, ufd, sigset_t __user *, user_mask,
 309                 size_t, sizemask)
 310 {
 311         return sys_signalfd4(ufd, user_mask, sizemask, 0);
 312 }
 313
 314 #ifdef CONFIG_COMPAT
 315 COMPAT_SYSCALL_DEFINE4(signalfd4, int, ufd,
 316                      const compat_sigset_t __user *,sigmask,
 317                      compat_size_t, sigsetsize,
 318                      int, flags)
 319 {
 320         compat_sigset_t ss32;
 321         sigset_t tmp;
 322         sigset_t __user *ksigmask;
 323
 324         if (sigsetsize != sizeof(compat_sigset_t))
 325                 return -EINVAL;
 326         if (copy_from_user(&ss32, sigmask, sizeof(ss32)))
 327                 return -EFAULT;
 328         sigset_from_compat(&tmp, &ss32);
 329         ksigmask = compat_alloc_user_space(sizeof(sigset_t));
 330         if (copy_to_user(ksigmask, &tmp, sizeof(sigset_t)))
 331                 return -EFAULT;
 332
 333         return sys_signalfd4(ufd, ksigmask, sizeof(sigset_t), flags);
 334 }
 335
 336 COMPAT_SYSCALL_DEFINE3(signalfd, int, ufd,
 337                      const compat_sigset_t __user *,sigmask,
 338                      compat_size_t, sigsetsize)
 339 {
 340         return compat_sys_signalfd4(ufd, sigmask, sigsetsize, 0);
 341 }
 342 #endif