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_ctx(struct eventfd_ctx *ctx)
  72 {
  73         kfree(ctx);
  74 }
  75
  76 static void eventfd_free(struct kref *kref)
  77 {
  78         struct eventfd_ctx *ctx = container_of(kref, struct eventfd_ctx, kref);
  79
  80         eventfd_free_ctx(ctx);
  81 }
  82
  83 /**
  84  * eventfd_ctx_get - Acquires a reference to the internal eventfd context.
  85  * @ctx: [in] Pointer to the eventfd context.
  86  *
  87  * Returns: In case of success, returns a pointer to the eventfd context.
  88  */
  89 struct eventfd_ctx *eventfd_ctx_get(struct eventfd_ctx *ctx)
  90 {
  91         kref_get(&ctx->kref);
  92         return ctx;
  93 }
  94 EXPORT_SYMBOL_GPL(eventfd_ctx_get);
  95
  96 /**
  97  * eventfd_ctx_put - Releases a reference to the internal eventfd context.
  98  * @ctx: [in] Pointer to eventfd context.
  99  *
 100  * The eventfd context reference must have been previously acquired either
 101  * with eventfd_ctx_get() or eventfd_ctx_fdget()).
 102  */
 103 void eventfd_ctx_put(struct eventfd_ctx *ctx)
 104 {
 105         kref_put(&ctx->kref, eventfd_free);
 106 }
 107 EXPORT_SYMBOL_GPL(eventfd_ctx_put);
 108
 109 static int eventfd_release(struct inode *inode, struct file *file)
 110 {
 111         struct eventfd_ctx *ctx = file->private_data;
 112
 113         wake_up_poll(&ctx->wqh, POLLHUP);
 114         eventfd_ctx_put(ctx);
 115         return 0;
 116 }
 117
 118 static unsigned int eventfd_poll(struct file *file, poll_table *wait)
 119 {
 120         struct eventfd_ctx *ctx = file->private_data;
 121         unsigned int events = 0;
 122         unsigned long flags;
 123
 124         poll_wait(file, &ctx->wqh, wait);
 125
 126         spin_lock_irqsave(&ctx->wqh.lock, flags);
 127         if (ctx->count > 0)
 128                 events |= POLLIN;
 129         if (ctx->count == ULLONG_MAX)
 130                 events |= POLLERR;
 131         if (ULLONG_MAX - 1 > ctx->count)
 132                 events |= POLLOUT;
 133         spin_unlock_irqrestore(&ctx->wqh.lock, flags);
 134
 135         return events;
 136 }
 137
 138 static ssize_t eventfd_read(struct file *file, char __user *buf, size_t count,
 139                             loff_t *ppos)
 140 {
 141         struct eventfd_ctx *ctx = file->private_data;
 142         ssize_t res;
 143         __u64 ucnt = 0;
 144         DECLARE_WAITQUEUE(wait, current);
 145
 146         if (count < sizeof(ucnt))
 147                 return -EINVAL;
 148         spin_lock_irq(&ctx->wqh.lock);
 149         res = -EAGAIN;
 150         if (ctx->count > 0)
 151                 res = sizeof(ucnt);
 152         else if (!(file->f_flags & O_NONBLOCK)) {
 153                 __add_wait_queue(&ctx->wqh, &wait);
 154                 for (res = 0;;) {
 155                         set_current_state(TASK_INTERRUPTIBLE);
 156                         if (ctx->count > 0) {
 157                                 res = sizeof(ucnt);
 158                                 break;
 159                         }
 160                         if (signal_pending(current)) {
 161                                 res = -ERESTARTSYS;
 162                                 break;
 163                         }
 164                         spin_unlock_irq(&ctx->wqh.lock);
 165                         schedule();
 166                         spin_lock_irq(&ctx->wqh.lock);
 167                 }
 168                 __remove_wait_queue(&ctx->wqh, &wait);
 169                 __set_current_state(TASK_RUNNING);
 170         }
 171         if (likely(res > 0)) {
 172                 ucnt = (ctx->flags & EFD_SEMAPHORE) ? 1 : ctx->count;
 173                 ctx->count -= ucnt;
 174                 if (waitqueue_active(&ctx->wqh))
 175                         wake_up_locked_poll(&ctx->wqh, POLLOUT);
 176         }
 177         spin_unlock_irq(&ctx->wqh.lock);
 178         if (res > 0 && put_user(ucnt, (__u64 __user *) buf))
 179                 return -EFAULT;
 180
 181         return res;
 182 }
 183
 184 static ssize_t eventfd_write(struct file *file, const char __user *buf, size_t count,
 185                              loff_t *ppos)
 186 {
 187         struct eventfd_ctx *ctx = file->private_data;
 188         ssize_t res;
 189         __u64 ucnt;
 190         DECLARE_WAITQUEUE(wait, current);
 191
 192         if (count < sizeof(ucnt))
 193                 return -EINVAL;
 194         if (copy_from_user(&ucnt, buf, sizeof(ucnt)))
 195                 return -EFAULT;
 196         if (ucnt == ULLONG_MAX)
 197                 return -EINVAL;
 198         spin_lock_irq(&ctx->wqh.lock);
 199         res = -EAGAIN;
 200         if (ULLONG_MAX - ctx->count > ucnt)
 201                 res = sizeof(ucnt);
 202         else if (!(file->f_flags & O_NONBLOCK)) {
 203                 __add_wait_queue(&ctx->wqh, &wait);
 204                 for (res = 0;;) {
 205                         set_current_state(TASK_INTERRUPTIBLE);
 206                         if (ULLONG_MAX - ctx->count > ucnt) {
 207                                 res = sizeof(ucnt);
 208                                 break;
 209                         }
 210                         if (signal_pending(current)) {
 211                                 res = -ERESTARTSYS;
 212                                 break;
 213                         }
 214                         spin_unlock_irq(&ctx->wqh.lock);
 215                         schedule();
 216                         spin_lock_irq(&ctx->wqh.lock);
 217                 }
 218                 __remove_wait_queue(&ctx->wqh, &wait);
 219                 __set_current_state(TASK_RUNNING);
 220         }
 221         if (likely(res > 0)) {
 222                 ctx->count += ucnt;
 223                 if (waitqueue_active(&ctx->wqh))
 224                         wake_up_locked_poll(&ctx->wqh, POLLIN);
 225         }
 226         spin_unlock_irq(&ctx->wqh.lock);
 227
 228         return res;
 229 }
 230
 231 static const struct file_operations eventfd_fops = {
 232         .release        = eventfd_release,
 233         .poll           = eventfd_poll,
 234         .read           = eventfd_read,
 235         .write          = eventfd_write,
 236 };
 237
 238 /**
 239  * eventfd_fget - Acquire a reference of an eventfd file descriptor.
 240  * @fd: [in] Eventfd file descriptor.
 241  *
 242  * Returns a pointer to the eventfd file structure in case of success, or the
 243  * following error pointer:
 244  *
 245  * -EBADF    : Invalid @fd file descriptor.
 246  * -EINVAL   : The @fd file descriptor is not an eventfd file.
 247  */
 248 struct file *eventfd_fget(int fd)
 249 {
 250         struct file *file;
 251
 252         file = fget(fd);
 253         if (!file)
 254                 return ERR_PTR(-EBADF);
 255         if (file->f_op != &eventfd_fops) {
 256                 fput(file);
 257                 return ERR_PTR(-EINVAL);
 258         }
 259
 260         return file;
 261 }
 262 EXPORT_SYMBOL_GPL(eventfd_fget);
 263
 264 /**
 265  * eventfd_ctx_fdget - Acquires a reference to the internal eventfd context.
 266  * @fd: [in] Eventfd file descriptor.
 267  *
 268  * Returns a pointer to the internal eventfd context, otherwise the error
 269  * pointers returned by the following functions:
 270  *
 271  * eventfd_fget
 272  */
 273 struct eventfd_ctx *eventfd_ctx_fdget(int fd)
 274 {
 275         struct file *file;
 276         struct eventfd_ctx *ctx;
 277
 278         file = eventfd_fget(fd);
 279         if (IS_ERR(file))
 280                 return (struct eventfd_ctx *) file;
 281         ctx = eventfd_ctx_get(file->private_data);
 282         fput(file);
 283
 284         return ctx;
 285 }
 286 EXPORT_SYMBOL_GPL(eventfd_ctx_fdget);
 287
 288 /**
 289  * eventfd_ctx_fileget - Acquires a reference to the internal eventfd context.
 290  * @file: [in] Eventfd file pointer.
 291  *
 292  * Returns a pointer to the internal eventfd context, otherwise the error
 293  * pointer:
 294  *
 295  * -EINVAL   : The @fd file descriptor is not an eventfd file.
 296  */
 297 struct eventfd_ctx *eventfd_ctx_fileget(struct file *file)
 298 {
 299         if (file->f_op != &eventfd_fops)
 300                 return ERR_PTR(-EINVAL);
 301
 302         return eventfd_ctx_get(file->private_data);
 303 }
 304 EXPORT_SYMBOL_GPL(eventfd_ctx_fileget);
 305
 306 /**
 307  * eventfd_file_create - Creates an eventfd file pointer.
 308  * @count: Initial eventfd counter value.
 309  * @flags: Flags for the eventfd file.
 310  *
 311  * This function creates an eventfd file pointer, w/out installing it into
 312  * the fd table. This is useful when the eventfd file is used during the
 313  * initialization of data structures that require extra setup after the eventfd
 314  * creation. So the eventfd creation is split into the file pointer creation
 315  * phase, and the file descriptor installation phase.
 316  * In this way races with userspace closing the newly installed file descriptor
 317  * can be avoided.
 318  * Returns an eventfd file pointer, or a proper error pointer.
 319  */
 320 struct file *eventfd_file_create(unsigned int count, int flags)
 321 {
 322         struct file *file;
 323         struct eventfd_ctx *ctx;
 324
 325         /* Check the EFD_* constants for consistency.  */
 326         BUILD_BUG_ON(EFD_CLOEXEC != O_CLOEXEC);
 327         BUILD_BUG_ON(EFD_NONBLOCK != O_NONBLOCK);
 328
 329         if (flags & ~EFD_FLAGS_SET)
 330                 return ERR_PTR(-EINVAL);
 331
 332         ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
 333         if (!ctx)
 334                 return ERR_PTR(-ENOMEM);
 335
 336         kref_init(&ctx->kref);
 337         init_waitqueue_head(&ctx->wqh);
 338         ctx->count = count;
 339         ctx->flags = flags;
 340
 341         file = anon_inode_getfile("[eventfd]", &eventfd_fops, ctx,
 342                                   O_RDWR | (flags & EFD_SHARED_FCNTL_FLAGS));
 343         if (IS_ERR(file))
 344                 eventfd_free_ctx(ctx);
 345
 346         return file;
 347 }
 348
 349 SYSCALL_DEFINE2(eventfd2, unsigned int, count, int, flags)
 350 {
 351         int fd, error;
 352         struct file *file;
 353
 354         error = get_unused_fd_flags(flags & EFD_SHARED_FCNTL_FLAGS);
 355         if (error < 0)
 356                 return error;
 357         fd = error;
 358
 359         file = eventfd_file_create(count, flags);
 360         if (IS_ERR(file)) {
 361                 error = PTR_ERR(file);
 362                 goto err_put_unused_fd;
 363         }
 364         fd_install(fd, file);
 365
 366         return fd;
 367
 368 err_put_unused_fd:
 369         put_unused_fd(fd);
 370
 371         return error;
 372 }
 373
 374 SYSCALL_DEFINE1(eventfd, unsigned int, count)
 375 {
 376         return sys_eventfd2(count, 0);
 377 }
 378