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