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 now, remaining;
  56
  57         now = ctx->tmr.base->get_time();
  58         remaining = ktime_sub(ctx->tmr.expires, now);
  59
  60         return remaining.tv64 < 0 ? ktime_set(0, 0): remaining;
  61 }
  62
  63 static void timerfd_setup(struct timerfd_ctx *ctx, int flags,
  64                           const struct itimerspec *ktmr)
  65 {
  66         enum hrtimer_mode htmode;
  67         ktime_t texp;
  68
  69         htmode = (flags & TFD_TIMER_ABSTIME) ?
  70                 HRTIMER_MODE_ABS: HRTIMER_MODE_REL;
  71
  72         texp = timespec_to_ktime(ktmr->it_value);
  73         ctx->expired = 0;
  74         ctx->ticks = 0;
  75         ctx->tintv = timespec_to_ktime(ktmr->it_interval);
  76         hrtimer_init(&ctx->tmr, ctx->clockid, htmode);
  77         ctx->tmr.expires = texp;
  78         ctx->tmr.function = timerfd_tmrproc;
  79         if (texp.tv64 != 0)
  80                 hrtimer_start(&ctx->tmr, texp, htmode);
  81 }
  82
  83 static int timerfd_release(struct inode *inode, struct file *file)
  84 {
  85         struct timerfd_ctx *ctx = file->private_data;
  86
  87         hrtimer_cancel(&ctx->tmr);
  88         kfree(ctx);
  89         return 0;
  90 }
  91
  92 static unsigned int timerfd_poll(struct file *file, poll_table *wait)
  93 {
  94         struct timerfd_ctx *ctx = file->private_data;
  95         unsigned int events = 0;
  96         unsigned long flags;
  97
  98         poll_wait(file, &ctx->wqh, wait);
  99
 100         spin_lock_irqsave(&ctx->wqh.lock, flags);
 101         if (ctx->ticks)
 102                 events |= POLLIN;
 103         spin_unlock_irqrestore(&ctx->wqh.lock, flags);
 104
 105         return events;
 106 }
 107
 108 static ssize_t timerfd_read(struct file *file, char __user *buf, size_t count,
 109                             loff_t *ppos)
 110 {
 111         struct timerfd_ctx *ctx = file->private_data;
 112         ssize_t res;
 113         u64 ticks = 0;
 114         DECLARE_WAITQUEUE(wait, current);
 115
 116         if (count < sizeof(ticks))
 117                 return -EINVAL;
 118         spin_lock_irq(&ctx->wqh.lock);
 119         res = -EAGAIN;
 120         if (!ctx->ticks && !(file->f_flags & O_NONBLOCK)) {
 121                 __add_wait_queue(&ctx->wqh, &wait);
 122                 for (res = 0;;) {
 123                         set_current_state(TASK_INTERRUPTIBLE);
 124                         if (ctx->ticks) {
 125                                 res = 0;
 126                                 break;
 127                         }
 128                         if (signal_pending(current)) {
 129                                 res = -ERESTARTSYS;
 130                                 break;
 131                         }
 132                         spin_unlock_irq(&ctx->wqh.lock);
 133                         schedule();
 134                         spin_lock_irq(&ctx->wqh.lock);
 135                 }
 136                 __remove_wait_queue(&ctx->wqh, &wait);
 137                 __set_current_state(TASK_RUNNING);
 138         }
 139         if (ctx->ticks) {
 140                 ticks = ctx->ticks;
 141                 if (ctx->expired && ctx->tintv.tv64) {
 142                         /*
 143                          * If tintv.tv64 != 0, this is a periodic timer that
 144                          * needs to be re-armed. We avoid doing it in the timer
 145                          * callback to avoid DoS attacks specifying a very
 146                          * short timer period.
 147                          */
 148                         ticks += hrtimer_forward_now(&ctx->tmr,
 149                                                      ctx->tintv) - 1;
 150                         hrtimer_restart(&ctx->tmr);
 151                 }
 152                 ctx->expired = 0;
 153                 ctx->ticks = 0;
 154         }
 155         spin_unlock_irq(&ctx->wqh.lock);
 156         if (ticks)
 157                 res = put_user(ticks, (u64 __user *) buf) ? -EFAULT: sizeof(ticks);
 158         return res;
 159 }
 160
 161 static const struct file_operations timerfd_fops = {
 162         .release        = timerfd_release,
 163         .poll           = timerfd_poll,
 164         .read           = timerfd_read,
 165 };
 166
 167 static struct file *timerfd_fget(int fd)
 168 {
 169         struct file *file;
 170
 171         file = fget(fd);
 172         if (!file)
 173                 return ERR_PTR(-EBADF);
 174         if (file->f_op != &timerfd_fops) {
 175                 fput(file);
 176                 return ERR_PTR(-EINVAL);
 177         }
 178
 179         return file;
 180 }
 181
 182 SYSCALL_DEFINE2(timerfd_create, int, clockid, int, flags)
 183 {
 184         int ufd;
 185         struct timerfd_ctx *ctx;
 186
 187         /* Check the TFD_* constants for consistency.  */
 188         BUILD_BUG_ON(TFD_CLOEXEC != O_CLOEXEC);
 189         BUILD_BUG_ON(TFD_NONBLOCK != O_NONBLOCK);
 190
 191         if (flags & ~(TFD_CLOEXEC | TFD_NONBLOCK))
 192                 return -EINVAL;
 193         if (clockid != CLOCK_MONOTONIC &&
 194             clockid != CLOCK_REALTIME)
 195                 return -EINVAL;
 196
 197         ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
 198         if (!ctx)
 199                 return -ENOMEM;
 200
 201         init_waitqueue_head(&ctx->wqh);
 202         ctx->clockid = clockid;
 203         hrtimer_init(&ctx->tmr, clockid, HRTIMER_MODE_ABS);
 204
 205         ufd = anon_inode_getfd("[timerfd]", &timerfd_fops, ctx,
 206                                flags & (O_CLOEXEC | O_NONBLOCK));
 207         if (ufd < 0)
 208                 kfree(ctx);
 209
 210         return ufd;
 211 }
 212
 213 SYSCALL_DEFINE4(timerfd_settime, int, ufd, int, flags,
 214                 const struct itimerspec __user *, utmr,
 215                 struct itimerspec __user *, otmr)
 216 {
 217         struct file *file;
 218         struct timerfd_ctx *ctx;
 219         struct itimerspec ktmr, kotmr;
 220
 221         if (copy_from_user(&ktmr, utmr, sizeof(ktmr)))
 222                 return -EFAULT;
 223
 224         if (!timespec_valid(&ktmr.it_value) ||
 225             !timespec_valid(&ktmr.it_interval))
 226                 return -EINVAL;
 227
 228         file = timerfd_fget(ufd);
 229         if (IS_ERR(file))
 230                 return PTR_ERR(file);
 231         ctx = file->private_data;
 232
 233         /*
 234          * We need to stop the existing timer before reprogramming
 235          * it to the new values.
 236          */
 237         for (;;) {
 238                 spin_lock_irq(&ctx->wqh.lock);
 239                 if (hrtimer_try_to_cancel(&ctx->tmr) >= 0)
 240                         break;
 241                 spin_unlock_irq(&ctx->wqh.lock);
 242                 cpu_relax();
 243         }
 244
 245         /*
 246          * If the timer is expired and it's periodic, we need to advance it
 247          * because the caller may want to know the previous expiration time.
 248          * We do not update "ticks" and "expired" since the timer will be
 249          * re-programmed again in the following timerfd_setup() call.
 250          */
 251         if (ctx->expired && ctx->tintv.tv64)
 252                 hrtimer_forward_now(&ctx->tmr, ctx->tintv);
 253
 254         kotmr.it_value = ktime_to_timespec(timerfd_get_remaining(ctx));
 255         kotmr.it_interval = ktime_to_timespec(ctx->tintv);
 256
 257         /*
 258          * Re-program the timer to the new value ...
 259          */
 260         timerfd_setup(ctx, flags, &ktmr);
 261
 262         spin_unlock_irq(&ctx->wqh.lock);
 263         fput(file);
 264         if (otmr && copy_to_user(otmr, &kotmr, sizeof(kotmr)))
 265                 return -EFAULT;
 266
 267         return 0;
 268 }
 269
 270 SYSCALL_DEFINE2(timerfd_gettime, int, ufd, struct itimerspec __user *, otmr)
 271 {
 272         struct file *file;
 273         struct timerfd_ctx *ctx;
 274         struct itimerspec kotmr;
 275
 276         file = timerfd_fget(ufd);
 277         if (IS_ERR(file))
 278                 return PTR_ERR(file);
 279         ctx = file->private_data;
 280
 281         spin_lock_irq(&ctx->wqh.lock);
 282         if (ctx->expired && ctx->tintv.tv64) {
 283                 ctx->expired = 0;
 284                 ctx->ticks +=
 285                         hrtimer_forward_now(&ctx->tmr, ctx->tintv) - 1;
 286                 hrtimer_restart(&ctx->tmr);
 287         }
 288         kotmr.it_value = ktime_to_timespec(timerfd_get_remaining(ctx));
 289         kotmr.it_interval = ktime_to_timespec(ctx->tintv);
 290         spin_unlock_irq(&ctx->wqh.lock);
 291         fput(file);
 292
 293         return copy_to_user(otmr, &kotmr, sizeof(kotmr)) ? -EFAULT: 0;
 294 }
 295