“6b2fb3c65844452bb9e8b449d50863d1b36c5dc0”上不存在“README.md”
timerfd.c 8.6 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
/*
 *  fs/timerfd.c
 *
 *  Copyright (C) 2007  Davide Libenzi <davidel@xmailserver.org>
 *
 *
 *  Thanks to Thomas Gleixner for code reviews and useful comments.
 *
 */

#include <linux/file.h>
#include <linux/poll.h>
#include <linux/init.h>
#include <linux/fs.h>
#include <linux/sched.h>
#include <linux/kernel.h>
17
#include <linux/slab.h>
18 19 20 21 22 23
#include <linux/list.h>
#include <linux/spinlock.h>
#include <linux/time.h>
#include <linux/hrtimer.h>
#include <linux/anon_inodes.h>
#include <linux/timerfd.h>
24
#include <linux/syscalls.h>
25
#include <linux/rcupdate.h>
26 27 28 29

struct timerfd_ctx {
	struct hrtimer tmr;
	ktime_t tintv;
30
	ktime_t moffs;
31
	wait_queue_head_t wqh;
D
Davide Libenzi 已提交
32
	u64 ticks;
33
	int expired;
D
Davide Libenzi 已提交
34
	int clockid;
35 36
	struct rcu_head rcu;
	struct list_head clist;
37
	bool might_cancel;
38 39
};

40 41 42
static LIST_HEAD(cancel_list);
static DEFINE_SPINLOCK(cancel_lock);

43 44 45
/*
 * This gets called when the timer event triggers. We set the "expired"
 * flag, but we do not re-arm the timer (in case it's necessary,
D
Davide Libenzi 已提交
46
 * tintv.tv64 != 0) until the timer is accessed.
47 48 49 50 51 52
 */
static enum hrtimer_restart timerfd_tmrproc(struct hrtimer *htmr)
{
	struct timerfd_ctx *ctx = container_of(htmr, struct timerfd_ctx, tmr);
	unsigned long flags;

53
	spin_lock_irqsave(&ctx->wqh.lock, flags);
54
	ctx->expired = 1;
D
Davide Libenzi 已提交
55
	ctx->ticks++;
56
	wake_up_locked(&ctx->wqh);
57
	spin_unlock_irqrestore(&ctx->wqh.lock, flags);
58 59 60 61

	return HRTIMER_NORESTART;
}

62 63
/*
 * Called when the clock was set to cancel the timers in the cancel
64 65 66
 * list. This will wake up processes waiting on these timers. The
 * wake-up requires ctx->ticks to be non zero, therefore we increment
 * it before calling wake_up_locked().
67 68
 */
void timerfd_clock_was_set(void)
D
Davide Libenzi 已提交
69
{
70 71 72
	ktime_t moffs = ktime_get_monotonic_offset();
	struct timerfd_ctx *ctx;
	unsigned long flags;
D
Davide Libenzi 已提交
73

74 75 76 77 78 79 80
	rcu_read_lock();
	list_for_each_entry_rcu(ctx, &cancel_list, clist) {
		if (!ctx->might_cancel)
			continue;
		spin_lock_irqsave(&ctx->wqh.lock, flags);
		if (ctx->moffs.tv64 != moffs.tv64) {
			ctx->moffs.tv64 = KTIME_MAX;
81
			ctx->ticks++;
82 83 84 85 86
			wake_up_locked(&ctx->wqh);
		}
		spin_unlock_irqrestore(&ctx->wqh.lock, flags);
	}
	rcu_read_unlock();
D
Davide Libenzi 已提交
87 88
}

89
static void timerfd_remove_cancel(struct timerfd_ctx *ctx)
90
{
91 92 93 94 95 96 97
	if (ctx->might_cancel) {
		ctx->might_cancel = false;
		spin_lock(&cancel_lock);
		list_del_rcu(&ctx->clist);
		spin_unlock(&cancel_lock);
	}
}
98

99 100 101
static bool timerfd_canceled(struct timerfd_ctx *ctx)
{
	if (!ctx->might_cancel || ctx->moffs.tv64 != KTIME_MAX)
102
		return false;
103 104 105
	ctx->moffs = ktime_get_monotonic_offset();
	return true;
}
106

107 108 109 110 111 112 113 114 115 116 117 118 119 120
static void timerfd_setup_cancel(struct timerfd_ctx *ctx, int flags)
{
	if (ctx->clockid == CLOCK_REALTIME && (flags & TFD_TIMER_ABSTIME) &&
	    (flags & TFD_TIMER_CANCEL_ON_SET)) {
		if (!ctx->might_cancel) {
			ctx->might_cancel = true;
			spin_lock(&cancel_lock);
			list_add_rcu(&ctx->clist, &cancel_list);
			spin_unlock(&cancel_lock);
		}
	} else if (ctx->might_cancel) {
		timerfd_remove_cancel(ctx);
	}
}
121

122 123 124
static ktime_t timerfd_get_remaining(struct timerfd_ctx *ctx)
{
	ktime_t remaining;
125

126 127
	remaining = hrtimer_expires_remaining(&ctx->tmr);
	return remaining.tv64 < 0 ? ktime_set(0, 0): remaining;
128 129 130 131
}

static int timerfd_setup(struct timerfd_ctx *ctx, int flags,
			 const struct itimerspec *ktmr)
132 133 134
{
	enum hrtimer_mode htmode;
	ktime_t texp;
135
	int clockid = ctx->clockid;
136 137 138 139 140 141

	htmode = (flags & TFD_TIMER_ABSTIME) ?
		HRTIMER_MODE_ABS: HRTIMER_MODE_REL;

	texp = timespec_to_ktime(ktmr->it_value);
	ctx->expired = 0;
D
Davide Libenzi 已提交
142
	ctx->ticks = 0;
143
	ctx->tintv = timespec_to_ktime(ktmr->it_interval);
144
	hrtimer_init(&ctx->tmr, clockid, htmode);
145
	hrtimer_set_expires(&ctx->tmr, texp);
146
	ctx->tmr.function = timerfd_tmrproc;
147
	if (texp.tv64 != 0) {
148
		hrtimer_start(&ctx->tmr, texp, htmode);
149 150 151 152
		if (timerfd_canceled(ctx))
			return -ECANCELED;
	}
	return 0;
153 154 155 156 157 158
}

static int timerfd_release(struct inode *inode, struct file *file)
{
	struct timerfd_ctx *ctx = file->private_data;

159
	timerfd_remove_cancel(ctx);
160
	hrtimer_cancel(&ctx->tmr);
161
	kfree_rcu(ctx, rcu);
162 163 164 165 166 167 168 169 170 171 172
	return 0;
}

static unsigned int timerfd_poll(struct file *file, poll_table *wait)
{
	struct timerfd_ctx *ctx = file->private_data;
	unsigned int events = 0;
	unsigned long flags;

	poll_wait(file, &ctx->wqh, wait);

173
	spin_lock_irqsave(&ctx->wqh.lock, flags);
D
Davide Libenzi 已提交
174
	if (ctx->ticks)
175
		events |= POLLIN;
176
	spin_unlock_irqrestore(&ctx->wqh.lock, flags);
177 178 179 180 181 182 183 184 185

	return events;
}

static ssize_t timerfd_read(struct file *file, char __user *buf, size_t count,
			    loff_t *ppos)
{
	struct timerfd_ctx *ctx = file->private_data;
	ssize_t res;
186
	u64 ticks = 0;
187 188 189

	if (count < sizeof(ticks))
		return -EINVAL;
190
	spin_lock_irq(&ctx->wqh.lock);
191 192 193 194
	if (file->f_flags & O_NONBLOCK)
		res = -EAGAIN;
	else
		res = wait_event_interruptible_locked_irq(ctx->wqh, ctx->ticks);
195

196 197 198 199 200 201 202 203 204 205 206
	/*
	 * If clock has changed, we do not care about the
	 * ticks and we do not rearm the timer. Userspace must
	 * reevaluate anyway.
	 */
	if (timerfd_canceled(ctx)) {
		ctx->ticks = 0;
		ctx->expired = 0;
		res = -ECANCELED;
	}

D
Davide Libenzi 已提交
207 208
	if (ctx->ticks) {
		ticks = ctx->ticks;
209

D
Davide Libenzi 已提交
210
		if (ctx->expired && ctx->tintv.tv64) {
211 212 213 214 215 216
			/*
			 * If tintv.tv64 != 0, this is a periodic timer that
			 * needs to be re-armed. We avoid doing it in the timer
			 * callback to avoid DoS attacks specifying a very
			 * short timer period.
			 */
D
Davide Libenzi 已提交
217 218
			ticks += hrtimer_forward_now(&ctx->tmr,
						     ctx->tintv) - 1;
219
			hrtimer_restart(&ctx->tmr);
D
Davide Libenzi 已提交
220 221 222
		}
		ctx->expired = 0;
		ctx->ticks = 0;
223
	}
224
	spin_unlock_irq(&ctx->wqh.lock);
225
	if (ticks)
226
		res = put_user(ticks, (u64 __user *) buf) ? -EFAULT: sizeof(ticks);
227 228 229 230 231 232 233
	return res;
}

static const struct file_operations timerfd_fops = {
	.release	= timerfd_release,
	.poll		= timerfd_poll,
	.read		= timerfd_read,
234
	.llseek		= noop_llseek,
235 236
};

D
Davide Libenzi 已提交
237 238 239 240 241 242 243 244 245 246 247 248 249 250 251
static struct file *timerfd_fget(int fd)
{
	struct file *file;

	file = fget(fd);
	if (!file)
		return ERR_PTR(-EBADF);
	if (file->f_op != &timerfd_fops) {
		fput(file);
		return ERR_PTR(-EINVAL);
	}

	return file;
}

252
SYSCALL_DEFINE2(timerfd_create, int, clockid, int, flags)
253
{
254
	int ufd;
255 256
	struct timerfd_ctx *ctx;

257 258 259 260
	/* Check the TFD_* constants for consistency.  */
	BUILD_BUG_ON(TFD_CLOEXEC != O_CLOEXEC);
	BUILD_BUG_ON(TFD_NONBLOCK != O_NONBLOCK);

D
Davide Libenzi 已提交
261 262 263
	if ((flags & ~TFD_CREATE_FLAGS) ||
	    (clockid != CLOCK_MONOTONIC &&
	     clockid != CLOCK_REALTIME))
264
		return -EINVAL;
D
Davide Libenzi 已提交
265 266 267 268 269 270 271 272

	ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
	if (!ctx)
		return -ENOMEM;

	init_waitqueue_head(&ctx->wqh);
	ctx->clockid = clockid;
	hrtimer_init(&ctx->tmr, clockid, HRTIMER_MODE_ABS);
273
	ctx->moffs = ktime_get_monotonic_offset();
D
Davide Libenzi 已提交
274

275
	ufd = anon_inode_getfd("[timerfd]", &timerfd_fops, ctx,
276
			       O_RDWR | (flags & TFD_SHARED_FCNTL_FLAGS));
277
	if (ufd < 0)
D
Davide Libenzi 已提交
278 279 280 281 282
		kfree(ctx);

	return ufd;
}

283 284 285
SYSCALL_DEFINE4(timerfd_settime, int, ufd, int, flags,
		const struct itimerspec __user *, utmr,
		struct itimerspec __user *, otmr)
D
Davide Libenzi 已提交
286 287 288 289
{
	struct file *file;
	struct timerfd_ctx *ctx;
	struct itimerspec ktmr, kotmr;
290
	int ret;
D
Davide Libenzi 已提交
291 292 293 294

	if (copy_from_user(&ktmr, utmr, sizeof(ktmr)))
		return -EFAULT;

D
Davide Libenzi 已提交
295 296
	if ((flags & ~TFD_SETTIME_FLAGS) ||
	    !timespec_valid(&ktmr.it_value) ||
297 298 299
	    !timespec_valid(&ktmr.it_interval))
		return -EINVAL;

D
Davide Libenzi 已提交
300 301 302 303
	file = timerfd_fget(ufd);
	if (IS_ERR(file))
		return PTR_ERR(file);
	ctx = file->private_data;
304

305 306
	timerfd_setup_cancel(ctx, flags);

D
Davide Libenzi 已提交
307 308 309 310 311 312 313 314
	/*
	 * We need to stop the existing timer before reprogramming
	 * it to the new values.
	 */
	for (;;) {
		spin_lock_irq(&ctx->wqh.lock);
		if (hrtimer_try_to_cancel(&ctx->tmr) >= 0)
			break;
315
		spin_unlock_irq(&ctx->wqh.lock);
D
Davide Libenzi 已提交
316
		cpu_relax();
317 318
	}

D
Davide Libenzi 已提交
319 320 321 322 323 324 325 326
	/*
	 * If the timer is expired and it's periodic, we need to advance it
	 * because the caller may want to know the previous expiration time.
	 * We do not update "ticks" and "expired" since the timer will be
	 * re-programmed again in the following timerfd_setup() call.
	 */
	if (ctx->expired && ctx->tintv.tv64)
		hrtimer_forward_now(&ctx->tmr, ctx->tintv);
327

D
Davide Libenzi 已提交
328 329 330 331 332 333
	kotmr.it_value = ktime_to_timespec(timerfd_get_remaining(ctx));
	kotmr.it_interval = ktime_to_timespec(ctx->tintv);

	/*
	 * Re-program the timer to the new value ...
	 */
334
	ret = timerfd_setup(ctx, flags, &ktmr);
D
Davide Libenzi 已提交
335 336 337 338 339 340

	spin_unlock_irq(&ctx->wqh.lock);
	fput(file);
	if (otmr && copy_to_user(otmr, &kotmr, sizeof(kotmr)))
		return -EFAULT;

341
	return ret;
D
Davide Libenzi 已提交
342 343
}

344
SYSCALL_DEFINE2(timerfd_gettime, int, ufd, struct itimerspec __user *, otmr)
D
Davide Libenzi 已提交
345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367
{
	struct file *file;
	struct timerfd_ctx *ctx;
	struct itimerspec kotmr;

	file = timerfd_fget(ufd);
	if (IS_ERR(file))
		return PTR_ERR(file);
	ctx = file->private_data;

	spin_lock_irq(&ctx->wqh.lock);
	if (ctx->expired && ctx->tintv.tv64) {
		ctx->expired = 0;
		ctx->ticks +=
			hrtimer_forward_now(&ctx->tmr, ctx->tintv) - 1;
		hrtimer_restart(&ctx->tmr);
	}
	kotmr.it_value = ktime_to_timespec(timerfd_get_remaining(ctx));
	kotmr.it_interval = ktime_to_timespec(ctx->tintv);
	spin_unlock_irq(&ctx->wqh.lock);
	fput(file);

	return copy_to_user(otmr, &kotmr, sizeof(kotmr)) ? -EFAULT: 0;
368 369
}
新手
引导
客服 返回
顶部