fs/timerfd.c

   1 /*
   2  *  fs/timerfd.c
   3  *
   4  *  Copyright (C) 2007  Davide Libenzi <davidel@xmailserver.org>
   5  *
   6  *
   7  *  Thanks to Thomas Gleixner for code reviews and useful comments.
   8  *
   9  */
  10
  11 #include <linux/file.h>
  12 #include <linux/poll.h>
  13 #include <linux/init.h>
  14 #include <linux/fs.h>
  15 #include <linux/sched.h>
  16 #include <linux/kernel.h>
  17 #include <linux/list.h>
  18 #include <linux/spinlock.h>
  19 #include <linux/time.h>
  20 #include <linux/hrtimer.h>
  21 #include <linux/anon_inodes.h>
  22 #include <linux/timerfd.h>
  23 #include <linux/syscalls.h>
  24
  25 struct timerfd_ctx {
  26         struct hrtimer tmr;
  27         ktime_t tintv;
  28         wait_queue_head_t wqh;
  29         u64 ticks;
  30         int expired;
  31         int clockid;
  32 };
  33
  34 /*
  35  * This gets called when the timer event triggers. We set the "expired"
  36  * flag, but we do not re-arm the timer (in case it's necessary,
  37  * tintv.tv64 != 0) until the timer is accessed.
  38  */
  39 static enum hrtimer_restart timerfd_tmrproc(struct hrtimer *htmr)
  40 {
  41         struct timerfd_ctx *ctx = container_of(htmr, struct timerfd_ctx, tmr);
  42         unsigned long flags;
  43
  44         spin_lock_irqsave(&ctx->wqh.lock, flags);
  45         ctx->expired = 1;
  46         ctx->ticks++;
  47         wake_up_locked(&ctx->wqh);
  48         spin_unlock_irqrestore(&ctx->wqh.lock, flags);
  49
  50         return HRTIMER_NORESTART;
  51 }
  52
  53 static ktime_t timerfd_get_remaining(struct timerfd_ctx *ctx)
  54 {
  55         ktime_t remaining;
  56
  57         remaining = hrtimer_expires_remaining(&ctx->tmr);
  58         return remaining.tv64 < 0 ? ktime_set(0, 0): remaining;
  59 }
  60
  61 static void timerfd_setup(struct timerfd_ctx *ctx, int flags,
  62                           const struct itimerspec *ktmr)
  63 {
  64         enum hrtimer_mode htmode;
  65         ktime_t texp;
  66
  67         htmode = (flags & TFD_TIMER_ABSTIME) ?
  68                 HRTIMER_MODE_ABS: HRTIMER_MODE_REL;
  69
  70         texp = timespec_to_ktime(ktmr->it_value);
  71         ctx->expired = 0;
  72         ctx->ticks = 0;
  73         ctx->tintv = timespec_to_ktime(ktmr->it_interval);
  74         hrtimer_init(&ctx->tmr, ctx->clockid, htmode);
  75         hrtimer_set_expires(&ctx->tmr, texp);
  76         ctx->tmr.function = timerfd_tmrproc;
  77         if (texp.tv64 != 0)
  78                 hrtimer_start(&ctx->tmr, texp, htmode);
  79 }
  80
  81 static int timerfd_release(struct inode *inode, struct file *file)
  82 {
  83         struct timerfd_ctx *ctx = file->private_data;
  84
  85         hrtimer_cancel(&ctx->tmr);
  86         kfree(ctx);
  87         return 0;
  88 }
  89
  90 static unsigned int timerfd_poll(struct file *file, poll_table *wait)
  91 {
  92         struct timerfd_ctx *ctx = file->private_data;
  93         unsigned int events = 0;
  94         unsigned long flags;
  95
  96         poll_wait(file, &ctx->wqh, wait);
  97
  98         spin_lock_irqsave(&ctx->wqh.lock, flags);
  99         if (ctx->ticks)
 100                 events |= POLLIN;
 101         spin_unlock_irqrestore(&ctx->wqh.lock, flags);
 102
 103         return events;
 104 }
 105
 106 static ssize_t timerfd_read(struct file *file, char __user *buf, size_t count,
 107                             loff_t *ppos)
 108 {
 109         struct timerfd_ctx *ctx = file->private_data;
 110         ssize_t res;
 111         u64 ticks = 0;
 112         DECLARE_WAITQUEUE(wait, current);
 113
 114         if (count < sizeof(ticks))
 115                 return -EINVAL;
 116         spin_lock_irq(&ctx->wqh.lock);
 117         res = -EAGAIN;
 118         if (!ctx->ticks && !(file->f_flags & O_NONBLOCK)) {
 119                 __add_wait_queue(&ctx->wqh, &wait);
 120                 for (res = 0;;) {
 121                         set_current_state(TASK_INTERRUPTIBLE);
 122                         if (ctx->ticks) {
 123                                 res = 0;
 124                                 break;
 125                         }
 126                         if (signal_pending(current)) {
 127                                 res = -ERESTARTSYS;
 128                                 break;
 129                         }
 130                         spin_unlock_irq(&ctx->wqh.lock);
 131                         schedule();
 132                         spin_lock_irq(&ctx->wqh.lock);
 133                 }
 134                 __remove_wait_queue(&ctx->wqh, &wait);
 135                 __set_current_state(TASK_RUNNING);
 136         }
 137         if (ctx->ticks) {
 138                 ticks = ctx->ticks;
 139                 if (ctx->expired && ctx->tintv.tv64) {
 140                         /*
 141                          * If tintv.tv64 != 0, this is a periodic timer that
 142                          * needs to be re-armed. We avoid doing it in the timer
 143                          * callback to avoid DoS attacks specifying a very
 144                          * short timer period.
 145                          */
 146                         ticks += hrtimer_forward_now(&ctx->tmr,
 147                                                      ctx->tintv) - 1;
 148                         hrtimer_restart(&ctx->tmr);
 149                 }
 150                 ctx->expired = 0;
 151                 ctx->ticks = 0;
 152         }
 153         spin_unlock_irq(&ctx->wqh.lock);
 154         if (ticks)
 155                 res = put_user(ticks, (u64 __user *) buf) ? -EFAULT: sizeof(ticks);
 156         return res;
 157 }
 158
 159 static const struct file_operations timerfd_fops = {
 160         .release        = timerfd_release,
 161         .poll           = timerfd_poll,
 162         .read           = timerfd_read,
 163 };
 164
 165 static struct file *timerfd_fget(int fd)
 166 {
 167         struct file *file;
 168
 169         file = fget(fd);
 170         if (!file)
 171                 return ERR_PTR(-EBADF);
 172         if (file->f_op != &timerfd_fops) {
 173                 fput(file);
 174                 return ERR_PTR(-EINVAL);
 175         }
 176
 177         return file;
 178 }
 179
 180 SYSCALL_DEFINE2(timerfd_create, int, clockid, int, flags)
 181 {
 182         int ufd;
 183         struct timerfd_ctx *ctx;
 184
 185         /* Check the TFD_* constants for consistency.  */
 186         BUILD_BUG_ON(TFD_CLOEXEC != O_CLOEXEC);
 187         BUILD_BUG_ON(TFD_NONBLOCK != O_NONBLOCK);
 188
 189         if ((flags & ~TFD_CREATE_FLAGS) ||
 190             (clockid != CLOCK_MONOTONIC &&
 191              clockid != CLOCK_REALTIME))
 192                 return -EINVAL;
 193
 194         ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
 195         if (!ctx)
 196                 return -ENOMEM;
 197
 198         init_waitqueue_head(&ctx->wqh);
 199         ctx->clockid = clockid;
 200         hrtimer_init(&ctx->tmr, clockid, HRTIMER_MODE_ABS);
 201
 202         ufd = anon_inode_getfd("[timerfd]", &timerfd_fops, ctx,
 203                                O_RDWR | (flags & TFD_SHARED_FCNTL_FLAGS));
 204         if (ufd < 0)
 205                 kfree(ctx);
 206
 207         return ufd;
 208 }
 209
 210 SYSCALL_DEFINE4(timerfd_settime, int, ufd, int, flags,
 211                 const struct itimerspec __user *, utmr,
 212                 struct itimerspec __user *, otmr)
 213 {
 214         struct file *file;
 215         struct timerfd_ctx *ctx;
 216         struct itimerspec ktmr, kotmr;
 217
 218         if (copy_from_user(&ktmr, utmr, sizeof(ktmr)))
 219                 return -EFAULT;
 220
 221         if ((flags & ~TFD_SETTIME_FLAGS) ||
 222             !timespec_valid(&ktmr.it_value) ||
 223             !timespec_valid(&ktmr.it_interval))
 224                 return -EINVAL;
 225
 226         file = timerfd_fget(ufd);
 227         if (IS_ERR(file))
 228                 return PTR_ERR(file);
 229         ctx = file->private_data;
 230
 231         /*
 232          * We need to stop the existing timer before reprogramming
 233          * it to the new values.
 234          */
 235         for (;;) {
 236                 spin_lock_irq(&ctx->wqh.lock);
 237                 if (hrtimer_try_to_cancel(&ctx->tmr) >= 0)
 238                         break;
 239                 spin_unlock_irq(&ctx->wqh.lock);
 240                 cpu_relax();
 241         }
 242
 243         /*
 244          * If the timer is expired and it's periodic, we need to advance it
 245          * because the caller may want to know the previous expiration time.
 246          * We do not update "ticks" and "expired" since the timer will be
 247          * re-programmed again in the following timerfd_setup() call.
 248          */
 249         if (ctx->expired && ctx->tintv.tv64)
 250                 hrtimer_forward_now(&ctx->tmr, ctx->tintv);
 251
 252         kotmr.it_value = ktime_to_timespec(timerfd_get_remaining(ctx));
 253         kotmr.it_interval = ktime_to_timespec(ctx->tintv);
 254
 255         /*
 256          * Re-program the timer to the new value ...
 257          */
 258         timerfd_setup(ctx, flags, &ktmr);
 259
 260         spin_unlock_irq(&ctx->wqh.lock);
 261         fput(file);
 262         if (otmr && copy_to_user(otmr, &kotmr, sizeof(kotmr)))
 263                 return -EFAULT;
 264
 265         return 0;
 266 }
 267
 268 SYSCALL_DEFINE2(timerfd_gettime, int, ufd, struct itimerspec __user *, otmr)
 269 {
 270         struct file *file;
 271         struct timerfd_ctx *ctx;
 272         struct itimerspec kotmr;
 273
 274         file = timerfd_fget(ufd);
 275         if (IS_ERR(file))
 276                 return PTR_ERR(file);
 277         ctx = file->private_data;
 278
 279         spin_lock_irq(&ctx->wqh.lock);
 280         if (ctx->expired && ctx->tintv.tv64) {
 281                 ctx->expired = 0;
 282                 ctx->ticks +=
 283                         hrtimer_forward_now(&ctx->tmr, ctx->tintv) - 1;
 284                 hrtimer_restart(&ctx->tmr);
 285         }
 286         kotmr.it_value = ktime_to_timespec(timerfd_get_remaining(ctx));
 287         kotmr.it_interval = ktime_to_timespec(ctx->tintv);
 288         spin_unlock_irq(&ctx->wqh.lock);
 289         fput(file);
 290
 291         return copy_to_user(otmr, &kotmr, sizeof(kotmr)) ? -EFAULT: 0;
 292 }
 293