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
  34 void signalfd_cleanup(struct sighand_struct *sighand)
  35 {
  36         wait_queue_head_t *wqh = &sighand->signalfd_wqh;
  37         /*
  38          * The lockless check can race with remove_wait_queue() in progress,
  39          * but in this case its caller should run under rcu_read_lock() and
  40          * sighand_cachep is SLAB_DESTROY_BY_RCU, we can safely return.
  41          */
  42         if (likely(!waitqueue_active(wqh)))
  43                 return;
  44
  45         /* wait_queue_t->func(POLLFREE) should do remove_wait_queue() */
  46         wake_up_poll(wqh, POLLHUP | POLLFREE);
  47 }
  48
  49 struct signalfd_ctx {
  50         sigset_t sigmask;
  51 };
  52
  53 static int signalfd_release(struct inode *inode, struct file *file)
  54 {
  55         kfree(file->private_data);
  56         return 0;
  57 }
  58
  59 static unsigned int signalfd_poll(struct file *file, poll_table *wait)
  60 {
  61         struct signalfd_ctx *ctx = file->private_data;
  62         unsigned int events = 0;
  63
  64         poll_wait(file, &current->sighand->signalfd_wqh, wait);
  65
  66         spin_lock_irq(&current->sighand->siglock);
  67         if (next_signal(&current->pending, &ctx->sigmask) ||
  68             next_signal(&current->signal->shared_pending,
  69                         &ctx->sigmask))
  70                 events |= POLLIN;
  71         spin_unlock_irq(&current->sighand->siglock);
  72
  73         return events;
  74 }
  75
  76 /*
  77  * Copied from copy_siginfo_to_user() in kernel/signal.c
  78  */
  79 static int signalfd_copyinfo(struct signalfd_siginfo __user *uinfo,
  80                              siginfo_t const *kinfo)
  81 {
  82         long err;
  83
  84         BUILD_BUG_ON(sizeof(struct signalfd_siginfo) != 128);
  85
  86         /*
  87          * Unused members should be zero ...
  88          */
  89         err = __clear_user(uinfo, sizeof(*uinfo));
  90
  91         /*
  92          * If you change siginfo_t structure, please be sure
  93          * this code is fixed accordingly.
  94          */
  95         err |= __put_user(kinfo->si_signo, &uinfo->ssi_signo);
  96         err |= __put_user(kinfo->si_errno, &uinfo->ssi_errno);
  97         err |= __put_user((short) kinfo->si_code, &uinfo->ssi_code);
  98         switch (kinfo->si_code & __SI_MASK) {
  99         case __SI_KILL:
 100                 err |= __put_user(kinfo->si_pid, &uinfo->ssi_pid);
 101                 err |= __put_user(kinfo->si_uid, &uinfo->ssi_uid);
 102                 break;
 103         case __SI_TIMER:
 104                  err |= __put_user(kinfo->si_tid, &uinfo->ssi_tid);
 105                  err |= __put_user(kinfo->si_overrun, &uinfo->ssi_overrun);
 106                  err |= __put_user((long) kinfo->si_ptr, &uinfo->ssi_ptr);
 107                  err |= __put_user(kinfo->si_int, &uinfo->ssi_int);
 108                 break;
 109         case __SI_POLL:
 110                 err |= __put_user(kinfo->si_band, &uinfo->ssi_band);
 111                 err |= __put_user(kinfo->si_fd, &uinfo->ssi_fd);
 112                 break;
 113         case __SI_FAULT:
 114                 err |= __put_user((long) kinfo->si_addr, &uinfo->ssi_addr);
 115 #ifdef __ARCH_SI_TRAPNO
 116                 err |= __put_user(kinfo->si_trapno, &uinfo->ssi_trapno);
 117 #endif
 118 #ifdef BUS_MCEERR_AO
 119                 /*
 120                  * Other callers might not initialize the si_lsb field,
 121                  * so check explicitly for the right codes here.
 122                  */
 123                 if (kinfo->si_code == BUS_MCEERR_AR ||
 124                     kinfo->si_code == BUS_MCEERR_AO)
 125                         err |= __put_user((short) kinfo->si_addr_lsb,
 126                                           &uinfo->ssi_addr_lsb);
 127 #endif
 128                 break;
 129         case __SI_CHLD:
 130                 err |= __put_user(kinfo->si_pid, &uinfo->ssi_pid);
 131                 err |= __put_user(kinfo->si_uid, &uinfo->ssi_uid);
 132                 err |= __put_user(kinfo->si_status, &uinfo->ssi_status);
 133                 err |= __put_user(kinfo->si_utime, &uinfo->ssi_utime);
 134                 err |= __put_user(kinfo->si_stime, &uinfo->ssi_stime);
 135                 break;
 136         case __SI_RT: /* This is not generated by the kernel as of now. */
 137         case __SI_MESGQ: /* But this is */
 138                 err |= __put_user(kinfo->si_pid, &uinfo->ssi_pid);
 139                 err |= __put_user(kinfo->si_uid, &uinfo->ssi_uid);
 140                 err |= __put_user((long) kinfo->si_ptr, &uinfo->ssi_ptr);
 141                 err |= __put_user(kinfo->si_int, &uinfo->ssi_int);
 142                 break;
 143         default:
 144                 /*
 145                  * This case catches also the signals queued by sigqueue().
 146                  */
 147                 err |= __put_user(kinfo->si_pid, &uinfo->ssi_pid);
 148                 err |= __put_user(kinfo->si_uid, &uinfo->ssi_uid);
 149                 err |= __put_user((long) kinfo->si_ptr, &uinfo->ssi_ptr);
 150                 err |= __put_user(kinfo->si_int, &uinfo->ssi_int);
 151                 break;
 152         }
 153
 154         return err ? -EFAULT: sizeof(*uinfo);
 155 }
 156
 157 static ssize_t signalfd_dequeue(struct signalfd_ctx *ctx, siginfo_t *info,
 158                                 int nonblock)
 159 {
 160         ssize_t ret;
 161         DECLARE_WAITQUEUE(wait, current);
 162
 163         spin_lock_irq(&current->sighand->siglock);
 164         ret = dequeue_signal(current, &ctx->sigmask, info);
 165         switch (ret) {
 166         case 0:
 167                 if (!nonblock)
 168                         break;
 169                 ret = -EAGAIN;
 170         default:
 171                 spin_unlock_irq(&current->sighand->siglock);
 172                 return ret;
 173         }
 174
 175         add_wait_queue(&current->sighand->signalfd_wqh, &wait);
 176         for (;;) {
 177                 set_current_state(TASK_INTERRUPTIBLE);
 178                 ret = dequeue_signal(current, &ctx->sigmask, info);
 179                 if (ret != 0)
 180                         break;
 181                 if (signal_pending(current)) {
 182                         ret = -ERESTARTSYS;
 183                         break;
 184                 }
 185                 spin_unlock_irq(&current->sighand->siglock);
 186                 schedule();
 187                 spin_lock_irq(&current->sighand->siglock);
 188         }
 189         spin_unlock_irq(&current->sighand->siglock);
 190
 191         remove_wait_queue(&current->sighand->signalfd_wqh, &wait);
 192         __set_current_state(TASK_RUNNING);
 193
 194         return ret;
 195 }
 196
 197 /*
 198  * Returns a multiple of the size of a "struct signalfd_siginfo", or a negative
 199  * error code. The "count" parameter must be at least the size of a
 200  * "struct signalfd_siginfo".
 201  */
 202 static ssize_t signalfd_read(struct file *file, char __user *buf, size_t count,
 203                              loff_t *ppos)
 204 {
 205         struct signalfd_ctx *ctx = file->private_data;
 206         struct signalfd_siginfo __user *siginfo;
 207         int nonblock = file->f_flags & O_NONBLOCK;
 208         ssize_t ret, total = 0;
 209         siginfo_t info;
 210
 211         count /= sizeof(struct signalfd_siginfo);
 212         if (!count)
 213                 return -EINVAL;
 214
 215         siginfo = (struct signalfd_siginfo __user *) buf;
 216         do {
 217                 ret = signalfd_dequeue(ctx, &info, nonblock);
 218                 if (unlikely(ret <= 0))
 219                         break;
 220                 ret = signalfd_copyinfo(siginfo, &info);
 221                 if (ret < 0)
 222                         break;
 223                 siginfo++;
 224                 total += ret;
 225                 nonblock = 1;
 226         } while (--count);
 227
 228         return total ? total: ret;
 229 }
 230
 231 #ifdef CONFIG_PROC_FS
 232 static int signalfd_show_fdinfo(struct seq_file *m, struct file *f)
 233 {
 234         struct signalfd_ctx *ctx = f->private_data;
 235         sigset_t sigmask;
 236
 237         sigmask = ctx->sigmask;
 238         signotset(&sigmask);
 239         render_sigset_t(m, "sigmask:\t", &sigmask);
 240
 241         return 0;
 242 }
 243 #endif
 244
 245 static const struct file_operations signalfd_fops = {
 246 #ifdef CONFIG_PROC_FS
 247         .show_fdinfo    = signalfd_show_fdinfo,
 248 #endif
 249         .release        = signalfd_release,
 250         .poll           = signalfd_poll,
 251         .read           = signalfd_read,
 252         .llseek         = noop_llseek,
 253 };
 254
 255 SYSCALL_DEFINE4(signalfd4, int, ufd, sigset_t __user *, user_mask,
 256                 size_t, sizemask, int, flags)
 257 {
 258         sigset_t sigmask;
 259         struct signalfd_ctx *ctx;
 260
 261         /* Check the SFD_* constants for consistency.  */
 262         BUILD_BUG_ON(SFD_CLOEXEC != O_CLOEXEC);
 263         BUILD_BUG_ON(SFD_NONBLOCK != O_NONBLOCK);
 264
 265         if (flags & ~(SFD_CLOEXEC | SFD_NONBLOCK))
 266                 return -EINVAL;
 267
 268         if (sizemask != sizeof(sigset_t) ||
 269             copy_from_user(&sigmask, user_mask, sizeof(sigmask)))
 270                 return -EINVAL;
 271         sigdelsetmask(&sigmask, sigmask(SIGKILL) | sigmask(SIGSTOP));
 272         signotset(&sigmask);
 273
 274         if (ufd == -1) {
 275                 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
 276                 if (!ctx)
 277                         return -ENOMEM;
 278
 279                 ctx->sigmask = sigmask;
 280
 281                 /*
 282                  * When we call this, the initialization must be complete, since
 283                  * anon_inode_getfd() will install the fd.
 284                  */
 285                 ufd = anon_inode_getfd("[signalfd]", &signalfd_fops, ctx,
 286                                        O_RDWR | (flags & (O_CLOEXEC | O_NONBLOCK)));
 287                 if (ufd < 0)
 288                         kfree(ctx);
 289         } else {
 290                 struct fd f = fdget(ufd);
 291                 if (!f.file)
 292                         return -EBADF;
 293                 ctx = f.file->private_data;
 294                 if (f.file->f_op != &signalfd_fops) {
 295                         fdput(f);
 296                         return -EINVAL;
 297                 }
 298                 spin_lock_irq(&current->sighand->siglock);
 299                 ctx->sigmask = sigmask;
 300                 spin_unlock_irq(&current->sighand->siglock);
 301
 302                 wake_up(&current->sighand->signalfd_wqh);
 303                 fdput(f);
 304         }
 305
 306         return ufd;
 307 }
 308
 309 SYSCALL_DEFINE3(signalfd, int, ufd, sigset_t __user *, user_mask,
 310                 size_t, sizemask)
 311 {
 312         return sys_signalfd4(ufd, user_mask, sizemask, 0);
 313 }