linux/fs/eventfd.c
<<
>>
Prefs
   1/*
   2 *  fs/eventfd.c
   3 *
   4 *  Copyright (C) 2007  Davide Libenzi <davidel@xmailserver.org>
   5 *
   6 */
   7
   8#include <linux/file.h>
   9#include <linux/poll.h>
  10#include <linux/init.h>
  11#include <linux/fs.h>
  12#include <linux/sched.h>
  13#include <linux/kernel.h>
  14#include <linux/slab.h>
  15#include <linux/list.h>
  16#include <linux/spinlock.h>
  17#include <linux/anon_inodes.h>
  18#include <linux/syscalls.h>
  19#include <linux/export.h>
  20#include <linux/kref.h>
  21#include <linux/eventfd.h>
  22#include <linux/proc_fs.h>
  23#include <linux/seq_file.h>
  24
  25struct eventfd_ctx {
  26        struct kref kref;
  27        wait_queue_head_t wqh;
  28        /*
  29         * Every time that a write(2) is performed on an eventfd, the
  30         * value of the __u64 being written is added to "count" and a
  31         * wakeup is performed on "wqh". A read(2) will return the "count"
  32         * value to userspace, and will reset "count" to zero. The kernel
  33         * side eventfd_signal() also, adds to the "count" counter and
  34         * issue a wakeup.
  35         */
  36        __u64 count;
  37        unsigned int flags;
  38};
  39
  40/**
  41 * eventfd_signal - Adds @n to the eventfd counter.
  42 * @ctx: [in] Pointer to the eventfd context.
  43 * @n: [in] Value of the counter to be added to the eventfd internal counter.
  44 *          The value cannot be negative.
  45 *
  46 * This function is supposed to be called by the kernel in paths that do not
  47 * allow sleeping. In this function we allow the counter to reach the ULLONG_MAX
  48 * value, and we signal this as overflow condition by returining a POLLERR
  49 * to poll(2).
  50 *
  51 * Returns the amount by which the counter was incrememnted.  This will be less
  52 * than @n if the counter has overflowed.
  53 */
  54__u64 eventfd_signal(struct eventfd_ctx *ctx, __u64 n)
  55{
  56        unsigned long flags;
  57
  58        spin_lock_irqsave(&ctx->wqh.lock, flags);
  59        if (ULLONG_MAX - ctx->count < n)
  60                n = ULLONG_MAX - ctx->count;
  61        ctx->count += n;
  62        if (waitqueue_active(&ctx->wqh))
  63                wake_up_locked_poll(&ctx->wqh, POLLIN);
  64        spin_unlock_irqrestore(&ctx->wqh.lock, flags);
  65
  66        return n;
  67}
  68EXPORT_SYMBOL_GPL(eventfd_signal);
  69
  70static void eventfd_free_ctx(struct eventfd_ctx *ctx)
  71{
  72        kfree(ctx);
  73}
  74
  75static void eventfd_free(struct kref *kref)
  76{
  77        struct eventfd_ctx *ctx = container_of(kref, struct eventfd_ctx, kref);
  78
  79        eventfd_free_ctx(ctx);
  80}
  81
  82/**
  83 * eventfd_ctx_get - Acquires a reference to the internal eventfd context.
  84 * @ctx: [in] Pointer to the eventfd context.
  85 *
  86 * Returns: In case of success, returns a pointer to the eventfd context.
  87 */
  88struct eventfd_ctx *eventfd_ctx_get(struct eventfd_ctx *ctx)
  89{
  90        kref_get(&ctx->kref);
  91        return ctx;
  92}
  93EXPORT_SYMBOL_GPL(eventfd_ctx_get);
  94
  95/**
  96 * eventfd_ctx_put - Releases a reference to the internal eventfd context.
  97 * @ctx: [in] Pointer to eventfd context.
  98 *
  99 * The eventfd context reference must have been previously acquired either
 100 * with eventfd_ctx_get() or eventfd_ctx_fdget().
 101 */
 102void eventfd_ctx_put(struct eventfd_ctx *ctx)
 103{
 104        kref_put(&ctx->kref, eventfd_free);
 105}
 106EXPORT_SYMBOL_GPL(eventfd_ctx_put);
 107
 108static int eventfd_release(struct inode *inode, struct file *file)
 109{
 110        struct eventfd_ctx *ctx = file->private_data;
 111
 112        wake_up_poll(&ctx->wqh, POLLHUP);
 113        eventfd_ctx_put(ctx);
 114        return 0;
 115}
 116
 117static unsigned int eventfd_poll(struct file *file, poll_table *wait)
 118{
 119        struct eventfd_ctx *ctx = file->private_data;
 120        unsigned int events = 0;
 121        unsigned long flags;
 122
 123        poll_wait(file, &ctx->wqh, wait);
 124
 125        spin_lock_irqsave(&ctx->wqh.lock, flags);
 126        if (ctx->count > 0)
 127                events |= POLLIN;
 128        if (ctx->count == ULLONG_MAX)
 129                events |= POLLERR;
 130        if (ULLONG_MAX - 1 > ctx->count)
 131                events |= POLLOUT;
 132        spin_unlock_irqrestore(&ctx->wqh.lock, flags);
 133
 134        return events;
 135}
 136
 137static void eventfd_ctx_do_read(struct eventfd_ctx *ctx, __u64 *cnt)
 138{
 139        *cnt = (ctx->flags & EFD_SEMAPHORE) ? 1 : ctx->count;
 140        ctx->count -= *cnt;
 141}
 142
 143/**
 144 * eventfd_ctx_remove_wait_queue - Read the current counter and removes wait queue.
 145 * @ctx: [in] Pointer to eventfd context.
 146 * @wait: [in] Wait queue to be removed.
 147 * @cnt: [out] Pointer to the 64-bit counter value.
 148 *
 149 * Returns %0 if successful, or the following error codes:
 150 *
 151 * -EAGAIN      : The operation would have blocked.
 152 *
 153 * This is used to atomically remove a wait queue entry from the eventfd wait
 154 * queue head, and read/reset the counter value.
 155 */
 156int eventfd_ctx_remove_wait_queue(struct eventfd_ctx *ctx, wait_queue_t *wait,
 157                                  __u64 *cnt)
 158{
 159        unsigned long flags;
 160
 161        spin_lock_irqsave(&ctx->wqh.lock, flags);
 162        eventfd_ctx_do_read(ctx, cnt);
 163        __remove_wait_queue(&ctx->wqh, wait);
 164        if (*cnt != 0 && waitqueue_active(&ctx->wqh))
 165                wake_up_locked_poll(&ctx->wqh, POLLOUT);
 166        spin_unlock_irqrestore(&ctx->wqh.lock, flags);
 167
 168        return *cnt != 0 ? 0 : -EAGAIN;
 169}
 170EXPORT_SYMBOL_GPL(eventfd_ctx_remove_wait_queue);
 171
 172/**
 173 * eventfd_ctx_read - Reads the eventfd counter or wait if it is zero.
 174 * @ctx: [in] Pointer to eventfd context.
 175 * @no_wait: [in] Different from zero if the operation should not block.
 176 * @cnt: [out] Pointer to the 64-bit counter value.
 177 *
 178 * Returns %0 if successful, or the following error codes:
 179 *
 180 * -EAGAIN      : The operation would have blocked but @no_wait was non-zero.
 181 * -ERESTARTSYS : A signal interrupted the wait operation.
 182 *
 183 * If @no_wait is zero, the function might sleep until the eventfd internal
 184 * counter becomes greater than zero.
 185 */
 186ssize_t eventfd_ctx_read(struct eventfd_ctx *ctx, int no_wait, __u64 *cnt)
 187{
 188        ssize_t res;
 189        DECLARE_WAITQUEUE(wait, current);
 190
 191        spin_lock_irq(&ctx->wqh.lock);
 192        *cnt = 0;
 193        res = -EAGAIN;
 194        if (ctx->count > 0)
 195                res = 0;
 196        else if (!no_wait) {
 197                __add_wait_queue(&ctx->wqh, &wait);
 198                for (;;) {
 199                        set_current_state(TASK_INTERRUPTIBLE);
 200                        if (ctx->count > 0) {
 201                                res = 0;
 202                                break;
 203                        }
 204                        if (signal_pending(current)) {
 205                                res = -ERESTARTSYS;
 206                                break;
 207                        }
 208                        spin_unlock_irq(&ctx->wqh.lock);
 209                        schedule();
 210                        spin_lock_irq(&ctx->wqh.lock);
 211                }
 212                __remove_wait_queue(&ctx->wqh, &wait);
 213                __set_current_state(TASK_RUNNING);
 214        }
 215        if (likely(res == 0)) {
 216                eventfd_ctx_do_read(ctx, cnt);
 217                if (waitqueue_active(&ctx->wqh))
 218                        wake_up_locked_poll(&ctx->wqh, POLLOUT);
 219        }
 220        spin_unlock_irq(&ctx->wqh.lock);
 221
 222        return res;
 223}
 224EXPORT_SYMBOL_GPL(eventfd_ctx_read);
 225
 226static ssize_t eventfd_read(struct file *file, char __user *buf, size_t count,
 227                            loff_t *ppos)
 228{
 229        struct eventfd_ctx *ctx = file->private_data;
 230        ssize_t res;
 231        __u64 cnt;
 232
 233        if (count < sizeof(cnt))
 234                return -EINVAL;
 235        res = eventfd_ctx_read(ctx, file->f_flags & O_NONBLOCK, &cnt);
 236        if (res < 0)
 237                return res;
 238
 239        return put_user(cnt, (__u64 __user *) buf) ? -EFAULT : sizeof(cnt);
 240}
 241
 242static ssize_t eventfd_write(struct file *file, const char __user *buf, size_t count,
 243                             loff_t *ppos)
 244{
 245        struct eventfd_ctx *ctx = file->private_data;
 246        ssize_t res;
 247        __u64 ucnt;
 248        DECLARE_WAITQUEUE(wait, current);
 249
 250        if (count < sizeof(ucnt))
 251                return -EINVAL;
 252        if (copy_from_user(&ucnt, buf, sizeof(ucnt)))
 253                return -EFAULT;
 254        if (ucnt == ULLONG_MAX)
 255                return -EINVAL;
 256        spin_lock_irq(&ctx->wqh.lock);
 257        res = -EAGAIN;
 258        if (ULLONG_MAX - ctx->count > ucnt)
 259                res = sizeof(ucnt);
 260        else if (!(file->f_flags & O_NONBLOCK)) {
 261                __add_wait_queue(&ctx->wqh, &wait);
 262                for (res = 0;;) {
 263                        set_current_state(TASK_INTERRUPTIBLE);
 264                        if (ULLONG_MAX - ctx->count > ucnt) {
 265                                res = sizeof(ucnt);
 266                                break;
 267                        }
 268                        if (signal_pending(current)) {
 269                                res = -ERESTARTSYS;
 270                                break;
 271                        }
 272                        spin_unlock_irq(&ctx->wqh.lock);
 273                        schedule();
 274                        spin_lock_irq(&ctx->wqh.lock);
 275                }
 276                __remove_wait_queue(&ctx->wqh, &wait);
 277                __set_current_state(TASK_RUNNING);
 278        }
 279        if (likely(res > 0)) {
 280                ctx->count += ucnt;
 281                if (waitqueue_active(&ctx->wqh))
 282                        wake_up_locked_poll(&ctx->wqh, POLLIN);
 283        }
 284        spin_unlock_irq(&ctx->wqh.lock);
 285
 286        return res;
 287}
 288
 289#ifdef CONFIG_PROC_FS
 290static int eventfd_show_fdinfo(struct seq_file *m, struct file *f)
 291{
 292        struct eventfd_ctx *ctx = f->private_data;
 293        int ret;
 294
 295        spin_lock_irq(&ctx->wqh.lock);
 296        ret = seq_printf(m, "eventfd-count: %16llx\n",
 297                         (unsigned long long)ctx->count);
 298        spin_unlock_irq(&ctx->wqh.lock);
 299
 300        return ret;
 301}
 302#endif
 303
 304static const struct file_operations eventfd_fops = {
 305#ifdef CONFIG_PROC_FS
 306        .show_fdinfo    = eventfd_show_fdinfo,
 307#endif
 308        .release        = eventfd_release,
 309        .poll           = eventfd_poll,
 310        .read           = eventfd_read,
 311        .write          = eventfd_write,
 312        .llseek         = noop_llseek,
 313};
 314
 315/**
 316 * eventfd_fget - Acquire a reference of an eventfd file descriptor.
 317 * @fd: [in] Eventfd file descriptor.
 318 *
 319 * Returns a pointer to the eventfd file structure in case of success, or the
 320 * following error pointer:
 321 *
 322 * -EBADF    : Invalid @fd file descriptor.
 323 * -EINVAL   : The @fd file descriptor is not an eventfd file.
 324 */
 325struct file *eventfd_fget(int fd)
 326{
 327        struct file *file;
 328
 329        file = fget(fd);
 330        if (!file)
 331                return ERR_PTR(-EBADF);
 332        if (file->f_op != &eventfd_fops) {
 333                fput(file);
 334                return ERR_PTR(-EINVAL);
 335        }
 336
 337        return file;
 338}
 339EXPORT_SYMBOL_GPL(eventfd_fget);
 340
 341/**
 342 * eventfd_ctx_fdget - Acquires a reference to the internal eventfd context.
 343 * @fd: [in] Eventfd file descriptor.
 344 *
 345 * Returns a pointer to the internal eventfd context, otherwise the error
 346 * pointers returned by the following functions:
 347 *
 348 * eventfd_fget
 349 */
 350struct eventfd_ctx *eventfd_ctx_fdget(int fd)
 351{
 352        struct file *file;
 353        struct eventfd_ctx *ctx;
 354
 355        file = eventfd_fget(fd);
 356        if (IS_ERR(file))
 357                return (struct eventfd_ctx *) file;
 358        ctx = eventfd_ctx_get(file->private_data);
 359        fput(file);
 360
 361        return ctx;
 362}
 363EXPORT_SYMBOL_GPL(eventfd_ctx_fdget);
 364
 365/**
 366 * eventfd_ctx_fileget - Acquires a reference to the internal eventfd context.
 367 * @file: [in] Eventfd file pointer.
 368 *
 369 * Returns a pointer to the internal eventfd context, otherwise the error
 370 * pointer:
 371 *
 372 * -EINVAL   : The @fd file descriptor is not an eventfd file.
 373 */
 374struct eventfd_ctx *eventfd_ctx_fileget(struct file *file)
 375{
 376        if (file->f_op != &eventfd_fops)
 377                return ERR_PTR(-EINVAL);
 378
 379        return eventfd_ctx_get(file->private_data);
 380}
 381EXPORT_SYMBOL_GPL(eventfd_ctx_fileget);
 382
 383/**
 384 * eventfd_file_create - Creates an eventfd file pointer.
 385 * @count: Initial eventfd counter value.
 386 * @flags: Flags for the eventfd file.
 387 *
 388 * This function creates an eventfd file pointer, w/out installing it into
 389 * the fd table. This is useful when the eventfd file is used during the
 390 * initialization of data structures that require extra setup after the eventfd
 391 * creation. So the eventfd creation is split into the file pointer creation
 392 * phase, and the file descriptor installation phase.
 393 * In this way races with userspace closing the newly installed file descriptor
 394 * can be avoided.
 395 * Returns an eventfd file pointer, or a proper error pointer.
 396 */
 397struct file *eventfd_file_create(unsigned int count, int flags)
 398{
 399        struct file *file;
 400        struct eventfd_ctx *ctx;
 401
 402        /* Check the EFD_* constants for consistency.  */
 403        BUILD_BUG_ON(EFD_CLOEXEC != O_CLOEXEC);
 404        BUILD_BUG_ON(EFD_NONBLOCK != O_NONBLOCK);
 405
 406        if (flags & ~EFD_FLAGS_SET)
 407                return ERR_PTR(-EINVAL);
 408
 409        ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
 410        if (!ctx)
 411                return ERR_PTR(-ENOMEM);
 412
 413        kref_init(&ctx->kref);
 414        init_waitqueue_head(&ctx->wqh);
 415        ctx->count = count;
 416        ctx->flags = flags;
 417
 418        file = anon_inode_getfile("[eventfd]", &eventfd_fops, ctx,
 419                                  O_RDWR | (flags & EFD_SHARED_FCNTL_FLAGS));
 420        if (IS_ERR(file))
 421                eventfd_free_ctx(ctx);
 422
 423        return file;
 424}
 425
 426SYSCALL_DEFINE2(eventfd2, unsigned int, count, int, flags)
 427{
 428        int fd, error;
 429        struct file *file;
 430
 431        error = get_unused_fd_flags(flags & EFD_SHARED_FCNTL_FLAGS);
 432        if (error < 0)
 433                return error;
 434        fd = error;
 435
 436        file = eventfd_file_create(count, flags);
 437        if (IS_ERR(file)) {
 438                error = PTR_ERR(file);
 439                goto err_put_unused_fd;
 440        }
 441        fd_install(fd, file);
 442
 443        return fd;
 444
 445err_put_unused_fd:
 446        put_unused_fd(fd);
 447
 448        return error;
 449}
 450
 451SYSCALL_DEFINE1(eventfd, unsigned int, count)
 452{
 453        return sys_eventfd2(count, 0);
 454}
 455
 456
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.