fs/eventfd.c

   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/list.h>
  15 #include <linux/spinlock.h>
  16 #include <linux/anon_inodes.h>
  17 #include <linux/syscalls.h>
  18 #include <linux/module.h>
  19 #include <linux/kref.h>
  20 #include <linux/eventfd.h>
  21
  22 struct eventfd_ctx {
  23         struct kref kref;
  24         wait_queue_head_t wqh;
  25         /*
  26          * Every time that a write(2) is performed on an eventfd, the
  27          * value of the __u64 being written is added to "count" and a
  28          * wakeup is performed on "wqh". A read(2) will return the "count"
  29          * value to userspace, and will reset "count" to zero. The kernel
  30          * side eventfd_signal() also, adds to the "count" counter and
  31          * issue a wakeup.
  32          */
  33         __u64 count;
  34         unsigned int flags;
  35 };
  36
  37 /**
  38  * eventfd_signal - Adds @n to the eventfd counter.
  39  * @ctx: [in] Pointer to the eventfd context.
  40  * @n: [in] Value of the counter to be added to the eventfd internal counter.
  41  *          The value cannot be negative.
  42  *
  43  * This function is supposed to be called by the kernel in paths that do not
  44  * allow sleeping. In this function we allow the counter to reach the ULLONG_MAX
  45  * value, and we signal this as overflow condition by returining a POLLERR
  46  * to poll(2).
  47  *
  48  * Returns @n in case of success, a non-negative number lower than @n in case
  49  * of overflow, or the following error codes:
  50  *
  51  * -EINVAL    : The value of @n is negative.
  52  */
  53 int eventfd_signal(struct eventfd_ctx *ctx, int n)
  54 {
  55         unsigned long flags;
  56
  57         if (n < 0)
  58                 return -EINVAL;
  59         spin_lock_irqsave(&ctx->wqh.lock, flags);
  60         if (ULLONG_MAX - ctx->count < n)
  61                 n = (int) (ULLONG_MAX - ctx->count);
  62         ctx->count += n;
  63         if (waitqueue_active(&ctx->wqh))
  64                 wake_up_locked_poll(&ctx->wqh, POLLIN);
  65         spin_unlock_irqrestore(&ctx->wqh.lock, flags);
  66
  67         return n;
  68 }
  69 EXPORT_SYMBOL_GPL(eventfd_signal);
  70
  71 static void eventfd_free(struct kref *kref)
  72 {
  73         struct eventfd_ctx *ctx = container_of(kref, struct eventfd_ctx, kref);
  74
  75         kfree(ctx);
  76 }
  77
  78 /**
  79  * eventfd_ctx_get - Acquires a reference to the internal eventfd context.
  80  * @ctx: [in] Pointer to the eventfd context.
  81  *
  82  * Returns: In case of success, returns a pointer to the eventfd context.
  83  */
  84 struct eventfd_ctx *eventfd_ctx_get(struct eventfd_ctx *ctx)
  85 {
  86         kref_get(&ctx->kref);
  87         return ctx;
  88 }
  89 EXPORT_SYMBOL_GPL(eventfd_ctx_get);
  90
  91 /**
  92  * eventfd_ctx_put - Releases a reference to the internal eventfd context.
  93  * @ctx: [in] Pointer to eventfd context.
  94  *
  95  * The eventfd context reference must have been previously acquired either
  96  * with eventfd_ctx_get() or eventfd_ctx_fdget()).
  97  */
  98 void eventfd_ctx_put(struct eventfd_ctx *ctx)
  99 {
 100         kref_put(&ctx->kref, eventfd_free);
 101 }
 102 EXPORT_SYMBOL_GPL(eventfd_ctx_put);
 103
 104 static int eventfd_release(struct inode *inode, struct file *file)
 105 {
 106         struct eventfd_ctx *ctx = file->private_data;
 107
 108         wake_up_poll(&ctx->wqh, POLLHUP);
 109         eventfd_ctx_put(ctx);
 110         return 0;
 111 }
 112
 113 static unsigned int eventfd_poll(struct file *file, poll_table *wait)
 114 {
 115         struct eventfd_ctx *ctx = file->private_data;
 116         unsigned int events = 0;
 117         unsigned long flags;
 118
 119         poll_wait(file, &ctx->wqh, wait);
 120
 121         spin_lock_irqsave(&ctx->wqh.lock, flags);
 122         if (ctx->count > 0)
 123                 events |= POLLIN;
 124         if (ctx->count == ULLONG_MAX)
 125                 events |= POLLERR;
 126         if (ULLONG_MAX - 1 > ctx->count)
 127                 events |= POLLOUT;
 128         spin_unlock_irqrestore(&ctx->wqh.lock, flags);
 129
 130         return events;
 131 }
 132
 133 static ssize_t eventfd_read(struct file *file, char __user *buf, size_t count,
 134                             loff_t *ppos)
 135 {
 136         struct eventfd_ctx *ctx = file->private_data;
 137         ssize_t res;
 138         __u64 ucnt = 0;
 139         DECLARE_WAITQUEUE(wait, current);
 140
 141         if (count < sizeof(ucnt))
 142                 return -EINVAL;
 143         spin_lock_irq(&ctx->wqh.lock);
 144         res = -EAGAIN;
 145         if (ctx->count > 0)
 146                 res = sizeof(ucnt);
 147         else if (!(file->f_flags & O_NONBLOCK)) {
 148                 __add_wait_queue(&ctx->wqh, &wait);
 149                 for (res = 0;;) {
 150                         set_current_state(TASK_INTERRUPTIBLE);
 151                         if (ctx->count > 0) {
 152                                 res = sizeof(ucnt);
 153                                 break;
 154                         }
 155                         if (signal_pending(current)) {
 156                                 res = -ERESTARTSYS;
 157                                 break;
 158                         }
 159                         spin_unlock_irq(&ctx->wqh.lock);
 160                         schedule();
 161                         spin_lock_irq(&ctx->wqh.lock);
 162                 }
 163                 __remove_wait_queue(&ctx->wqh, &wait);
 164                 __set_current_state(TASK_RUNNING);
 165         }
 166         if (likely(res > 0)) {
 167                 ucnt = (ctx->flags & EFD_SEMAPHORE) ? 1 : ctx->count;
 168                 ctx->count -= ucnt;
 169                 if (waitqueue_active(&ctx->wqh))
 170                         wake_up_locked_poll(&ctx->wqh, POLLOUT);
 171         }
 172         spin_unlock_irq(&ctx->wqh.lock);
 173         if (res > 0 && put_user(ucnt, (__u64 __user *) buf))
 174                 return -EFAULT;
 175
 176         return res;
 177 }
 178
 179 static ssize_t eventfd_write(struct file *file, const char __user *buf, size_t count,
 180                              loff_t *ppos)
 181 {
 182         struct eventfd_ctx *ctx = file->private_data;
 183         ssize_t res;
 184         __u64 ucnt;
 185         DECLARE_WAITQUEUE(wait, current);
 186
 187         if (count < sizeof(ucnt))
 188                 return -EINVAL;
 189         if (copy_from_user(&ucnt, buf, sizeof(ucnt)))
 190                 return -EFAULT;
 191         if (ucnt == ULLONG_MAX)
 192                 return -EINVAL;
 193         spin_lock_irq(&ctx->wqh.lock);
 194         res = -EAGAIN;
 195         if (ULLONG_MAX - ctx->count > ucnt)
 196                 res = sizeof(ucnt);
 197         else if (!(file->f_flags & O_NONBLOCK)) {
 198                 __add_wait_queue(&ctx->wqh, &wait);
 199                 for (res = 0;;) {
 200                         set_current_state(TASK_INTERRUPTIBLE);
 201                         if (ULLONG_MAX - ctx->count > ucnt) {
 202                                 res = sizeof(ucnt);
 203                                 break;
 204                         }
 205                         if (signal_pending(current)) {
 206                                 res = -ERESTARTSYS;
 207                                 break;
 208                         }
 209                         spin_unlock_irq(&ctx->wqh.lock);
 210                         schedule();
 211                         spin_lock_irq(&ctx->wqh.lock);
 212                 }
 213                 __remove_wait_queue(&ctx->wqh, &wait);
 214                 __set_current_state(TASK_RUNNING);
 215         }
 216         if (likely(res > 0)) {
 217                 ctx->count += ucnt;
 218                 if (waitqueue_active(&ctx->wqh))
 219                         wake_up_locked_poll(&ctx->wqh, POLLIN);
 220         }
 221         spin_unlock_irq(&ctx->wqh.lock);
 222
 223         return res;
 224 }
 225
 226 static const struct file_operations eventfd_fops = {
 227         .release        = eventfd_release,
 228         .poll           = eventfd_poll,
 229         .read           = eventfd_read,
 230         .write          = eventfd_write,
 231 };
 232
 233 /**
 234  * eventfd_fget - Acquire a reference of an eventfd file descriptor.
 235  * @fd: [in] Eventfd file descriptor.
 236  *
 237  * Returns a pointer to the eventfd file structure in case of success, or the
 238  * following error pointer:
 239  *
 240  * -EBADF    : Invalid @fd file descriptor.
 241  * -EINVAL   : The @fd file descriptor is not an eventfd file.
 242  */
 243 struct file *eventfd_fget(int fd)
 244 {
 245         struct file *file;
 246
 247         file = fget(fd);
 248         if (!file)
 249                 return ERR_PTR(-EBADF);
 250         if (file->f_op != &eventfd_fops) {
 251                 fput(file);
 252                 return ERR_PTR(-EINVAL);
 253         }
 254
 255         return file;
 256 }
 257 EXPORT_SYMBOL_GPL(eventfd_fget);
 258
 259 /**
 260  * eventfd_ctx_fdget - Acquires a reference to the internal eventfd context.
 261  * @fd: [in] Eventfd file descriptor.
 262  *
 263  * Returns a pointer to the internal eventfd context, otherwise the error
 264  * pointers returned by the following functions:
 265  *
 266  * eventfd_fget
 267  */
 268 struct eventfd_ctx *eventfd_ctx_fdget(int fd)
 269 {
 270         struct file *file;
 271         struct eventfd_ctx *ctx;
 272
 273         file = eventfd_fget(fd);
 274         if (IS_ERR(file))
 275                 return (struct eventfd_ctx *) file;
 276         ctx = eventfd_ctx_get(file->private_data);
 277         fput(file);
 278
 279         return ctx;
 280 }
 281 EXPORT_SYMBOL_GPL(eventfd_ctx_fdget);
 282
 283 /**
 284  * eventfd_ctx_fileget - Acquires a reference to the internal eventfd context.
 285  * @file: [in] Eventfd file pointer.
 286  *
 287  * Returns a pointer to the internal eventfd context, otherwise the error
 288  * pointer:
 289  *
 290  * -EINVAL   : The @fd file descriptor is not an eventfd file.
 291  */
 292 struct eventfd_ctx *eventfd_ctx_fileget(struct file *file)
 293 {
 294         if (file->f_op != &eventfd_fops)
 295                 return ERR_PTR(-EINVAL);
 296
 297         return eventfd_ctx_get(file->private_data);
 298 }
 299 EXPORT_SYMBOL_GPL(eventfd_ctx_fileget);
 300
 301 SYSCALL_DEFINE2(eventfd2, unsigned int, count, int, flags)
 302 {
 303         int fd;
 304         struct eventfd_ctx *ctx;
 305
 306         /* Check the EFD_* constants for consistency.  */
 307         BUILD_BUG_ON(EFD_CLOEXEC != O_CLOEXEC);
 308         BUILD_BUG_ON(EFD_NONBLOCK != O_NONBLOCK);
 309
 310         if (flags & ~EFD_FLAGS_SET)
 311                 return -EINVAL;
 312
 313         ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
 314         if (!ctx)
 315                 return -ENOMEM;
 316
 317         kref_init(&ctx->kref);
 318         init_waitqueue_head(&ctx->wqh);
 319         ctx->count = count;
 320         ctx->flags = flags;
 321
 322         /*
 323          * When we call this, the initialization must be complete, since
 324          * anon_inode_getfd() will install the fd.
 325          */
 326         fd = anon_inode_getfd("[eventfd]", &eventfd_fops, ctx,
 327                               flags & EFD_SHARED_FCNTL_FLAGS);
 328         if (fd < 0)
 329                 kfree(ctx);
 330         return fd;
 331 }
 332
 333 SYSCALL_DEFINE1(eventfd, unsigned int, count)
 334 {
 335         return sys_eventfd2(count, 0);
 336 }
 337