Davide Libenzi | e1ad746 | 2007-05-10 22:23:19 -0700 | [diff] [blame] | 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> |
Tejun Heo | 5a0e3ad | 2010-03-24 17:04:11 +0900 | [diff] [blame] | 14 | #include <linux/slab.h> |
Davide Libenzi | e1ad746 | 2007-05-10 22:23:19 -0700 | [diff] [blame] | 15 | #include <linux/list.h> |
| 16 | #include <linux/spinlock.h> |
| 17 | #include <linux/anon_inodes.h> |
Adrian Bunk | 7747cdb | 2008-02-06 01:36:49 -0800 | [diff] [blame] | 18 | #include <linux/syscalls.h> |
Paul Gortmaker | 630d9c4 | 2011-11-16 23:57:37 -0500 | [diff] [blame] | 19 | #include <linux/export.h> |
Davide Libenzi | 1338901 | 2009-06-30 11:41:11 -0700 | [diff] [blame] | 20 | #include <linux/kref.h> |
| 21 | #include <linux/eventfd.h> |
Cyrill Gorcunov | cbac554 | 2012-12-17 16:04:57 -0800 | [diff] [blame] | 22 | #include <linux/proc_fs.h> |
| 23 | #include <linux/seq_file.h> |
Davide Libenzi | e1ad746 | 2007-05-10 22:23:19 -0700 | [diff] [blame] | 24 | |
| 25 | struct eventfd_ctx { |
Davide Libenzi | 1338901 | 2009-06-30 11:41:11 -0700 | [diff] [blame] | 26 | struct kref kref; |
Davide Libenzi | e1ad746 | 2007-05-10 22:23:19 -0700 | [diff] [blame] | 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 |
Davide Libenzi | 1338901 | 2009-06-30 11:41:11 -0700 | [diff] [blame] | 33 | * side eventfd_signal() also, adds to the "count" counter and |
Davide Libenzi | e1ad746 | 2007-05-10 22:23:19 -0700 | [diff] [blame] | 34 | * issue a wakeup. |
| 35 | */ |
| 36 | __u64 count; |
Davide Libenzi | bcd0b23 | 2009-03-31 15:24:18 -0700 | [diff] [blame] | 37 | unsigned int flags; |
Davide Libenzi | e1ad746 | 2007-05-10 22:23:19 -0700 | [diff] [blame] | 38 | }; |
| 39 | |
Davide Libenzi | 1338901 | 2009-06-30 11:41:11 -0700 | [diff] [blame] | 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 |
Masanari Iida | 20d5a86 | 2015-09-22 12:04:17 +0900 | [diff] [blame] | 48 | * value, and we signal this as overflow condition by returning a POLLERR |
Davide Libenzi | 1338901 | 2009-06-30 11:41:11 -0700 | [diff] [blame] | 49 | * to poll(2). |
| 50 | * |
Masanari Iida | 20d5a86 | 2015-09-22 12:04:17 +0900 | [diff] [blame] | 51 | * Returns the amount by which the counter was incremented. This will be less |
Sha Zhengju | ee62c6b | 2012-05-31 16:26:41 -0700 | [diff] [blame] | 52 | * than @n if the counter has overflowed. |
Davide Libenzi | e1ad746 | 2007-05-10 22:23:19 -0700 | [diff] [blame] | 53 | */ |
Sha Zhengju | ee62c6b | 2012-05-31 16:26:41 -0700 | [diff] [blame] | 54 | __u64 eventfd_signal(struct eventfd_ctx *ctx, __u64 n) |
Davide Libenzi | e1ad746 | 2007-05-10 22:23:19 -0700 | [diff] [blame] | 55 | { |
Davide Libenzi | e1ad746 | 2007-05-10 22:23:19 -0700 | [diff] [blame] | 56 | unsigned long flags; |
| 57 | |
Davide Libenzi | d48eb23 | 2007-05-18 12:02:33 -0700 | [diff] [blame] | 58 | spin_lock_irqsave(&ctx->wqh.lock, flags); |
Davide Libenzi | e1ad746 | 2007-05-10 22:23:19 -0700 | [diff] [blame] | 59 | if (ULLONG_MAX - ctx->count < n) |
Sha Zhengju | ee62c6b | 2012-05-31 16:26:41 -0700 | [diff] [blame] | 60 | n = ULLONG_MAX - ctx->count; |
Davide Libenzi | e1ad746 | 2007-05-10 22:23:19 -0700 | [diff] [blame] | 61 | ctx->count += n; |
| 62 | if (waitqueue_active(&ctx->wqh)) |
Davide Libenzi | 3951088 | 2009-03-31 15:24:23 -0700 | [diff] [blame] | 63 | wake_up_locked_poll(&ctx->wqh, POLLIN); |
Davide Libenzi | d48eb23 | 2007-05-18 12:02:33 -0700 | [diff] [blame] | 64 | spin_unlock_irqrestore(&ctx->wqh.lock, flags); |
Davide Libenzi | e1ad746 | 2007-05-10 22:23:19 -0700 | [diff] [blame] | 65 | |
| 66 | return n; |
| 67 | } |
Rusty Russell | 5718607 | 2009-06-12 22:27:09 -0600 | [diff] [blame] | 68 | EXPORT_SYMBOL_GPL(eventfd_signal); |
Davide Libenzi | e1ad746 | 2007-05-10 22:23:19 -0700 | [diff] [blame] | 69 | |
Davide Libenzi | 562787a | 2009-09-22 16:43:57 -0700 | [diff] [blame] | 70 | static void eventfd_free_ctx(struct eventfd_ctx *ctx) |
| 71 | { |
| 72 | kfree(ctx); |
| 73 | } |
| 74 | |
Davide Libenzi | 1338901 | 2009-06-30 11:41:11 -0700 | [diff] [blame] | 75 | static void eventfd_free(struct kref *kref) |
| 76 | { |
| 77 | struct eventfd_ctx *ctx = container_of(kref, struct eventfd_ctx, kref); |
| 78 | |
Davide Libenzi | 562787a | 2009-09-22 16:43:57 -0700 | [diff] [blame] | 79 | eventfd_free_ctx(ctx); |
Davide Libenzi | 1338901 | 2009-06-30 11:41:11 -0700 | [diff] [blame] | 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 | */ |
| 88 | struct eventfd_ctx *eventfd_ctx_get(struct eventfd_ctx *ctx) |
| 89 | { |
| 90 | kref_get(&ctx->kref); |
| 91 | return ctx; |
| 92 | } |
| 93 | EXPORT_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 |
Randy Dunlap | 3618218 | 2011-02-20 20:08:35 -0800 | [diff] [blame] | 100 | * with eventfd_ctx_get() or eventfd_ctx_fdget(). |
Davide Libenzi | 1338901 | 2009-06-30 11:41:11 -0700 | [diff] [blame] | 101 | */ |
| 102 | void eventfd_ctx_put(struct eventfd_ctx *ctx) |
| 103 | { |
| 104 | kref_put(&ctx->kref, eventfd_free); |
| 105 | } |
| 106 | EXPORT_SYMBOL_GPL(eventfd_ctx_put); |
| 107 | |
Davide Libenzi | e1ad746 | 2007-05-10 22:23:19 -0700 | [diff] [blame] | 108 | static int eventfd_release(struct inode *inode, struct file *file) |
| 109 | { |
Davide Libenzi | 1338901 | 2009-06-30 11:41:11 -0700 | [diff] [blame] | 110 | struct eventfd_ctx *ctx = file->private_data; |
| 111 | |
| 112 | wake_up_poll(&ctx->wqh, POLLHUP); |
| 113 | eventfd_ctx_put(ctx); |
Davide Libenzi | e1ad746 | 2007-05-10 22:23:19 -0700 | [diff] [blame] | 114 | return 0; |
| 115 | } |
| 116 | |
| 117 | static unsigned int eventfd_poll(struct file *file, poll_table *wait) |
| 118 | { |
| 119 | struct eventfd_ctx *ctx = file->private_data; |
| 120 | unsigned int events = 0; |
Chris Mason | e22553e | 2015-02-17 13:46:07 -0800 | [diff] [blame] | 121 | u64 count; |
Davide Libenzi | e1ad746 | 2007-05-10 22:23:19 -0700 | [diff] [blame] | 122 | |
| 123 | poll_wait(file, &ctx->wqh, wait); |
Chris Mason | e22553e | 2015-02-17 13:46:07 -0800 | [diff] [blame] | 124 | smp_rmb(); |
| 125 | count = ctx->count; |
Davide Libenzi | e1ad746 | 2007-05-10 22:23:19 -0700 | [diff] [blame] | 126 | |
Chris Mason | e22553e | 2015-02-17 13:46:07 -0800 | [diff] [blame] | 127 | if (count > 0) |
Davide Libenzi | e1ad746 | 2007-05-10 22:23:19 -0700 | [diff] [blame] | 128 | events |= POLLIN; |
Chris Mason | e22553e | 2015-02-17 13:46:07 -0800 | [diff] [blame] | 129 | if (count == ULLONG_MAX) |
Davide Libenzi | e1ad746 | 2007-05-10 22:23:19 -0700 | [diff] [blame] | 130 | events |= POLLERR; |
Chris Mason | e22553e | 2015-02-17 13:46:07 -0800 | [diff] [blame] | 131 | if (ULLONG_MAX - 1 > count) |
Davide Libenzi | e1ad746 | 2007-05-10 22:23:19 -0700 | [diff] [blame] | 132 | events |= POLLOUT; |
Davide Libenzi | e1ad746 | 2007-05-10 22:23:19 -0700 | [diff] [blame] | 133 | |
| 134 | return events; |
| 135 | } |
| 136 | |
Davide Libenzi | cb289d6 | 2010-01-13 09:34:36 -0800 | [diff] [blame] | 137 | static void eventfd_ctx_do_read(struct eventfd_ctx *ctx, __u64 *cnt) |
Davide Libenzi | e1ad746 | 2007-05-10 22:23:19 -0700 | [diff] [blame] | 138 | { |
Davide Libenzi | cb289d6 | 2010-01-13 09:34:36 -0800 | [diff] [blame] | 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. |
Randy Dunlap | 3618218 | 2011-02-20 20:08:35 -0800 | [diff] [blame] | 147 | * @cnt: [out] Pointer to the 64-bit counter value. |
Davide Libenzi | cb289d6 | 2010-01-13 09:34:36 -0800 | [diff] [blame] | 148 | * |
Randy Dunlap | 3618218 | 2011-02-20 20:08:35 -0800 | [diff] [blame] | 149 | * Returns %0 if successful, or the following error codes: |
Davide Libenzi | cb289d6 | 2010-01-13 09:34:36 -0800 | [diff] [blame] | 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 | */ |
| 156 | int 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 | } |
| 170 | EXPORT_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. |
Randy Dunlap | 3618218 | 2011-02-20 20:08:35 -0800 | [diff] [blame] | 176 | * @cnt: [out] Pointer to the 64-bit counter value. |
Davide Libenzi | cb289d6 | 2010-01-13 09:34:36 -0800 | [diff] [blame] | 177 | * |
Randy Dunlap | 3618218 | 2011-02-20 20:08:35 -0800 | [diff] [blame] | 178 | * Returns %0 if successful, or the following error codes: |
Davide Libenzi | cb289d6 | 2010-01-13 09:34:36 -0800 | [diff] [blame] | 179 | * |
Randy Dunlap | 3618218 | 2011-02-20 20:08:35 -0800 | [diff] [blame] | 180 | * -EAGAIN : The operation would have blocked but @no_wait was non-zero. |
Davide Libenzi | cb289d6 | 2010-01-13 09:34:36 -0800 | [diff] [blame] | 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 | */ |
| 186 | ssize_t eventfd_ctx_read(struct eventfd_ctx *ctx, int no_wait, __u64 *cnt) |
| 187 | { |
Davide Libenzi | e1ad746 | 2007-05-10 22:23:19 -0700 | [diff] [blame] | 188 | ssize_t res; |
Davide Libenzi | e1ad746 | 2007-05-10 22:23:19 -0700 | [diff] [blame] | 189 | DECLARE_WAITQUEUE(wait, current); |
| 190 | |
Davide Libenzi | d48eb23 | 2007-05-18 12:02:33 -0700 | [diff] [blame] | 191 | spin_lock_irq(&ctx->wqh.lock); |
Davide Libenzi | cb289d6 | 2010-01-13 09:34:36 -0800 | [diff] [blame] | 192 | *cnt = 0; |
Davide Libenzi | e1ad746 | 2007-05-10 22:23:19 -0700 | [diff] [blame] | 193 | res = -EAGAIN; |
Davide Libenzi | bcd0b23 | 2009-03-31 15:24:18 -0700 | [diff] [blame] | 194 | if (ctx->count > 0) |
Davide Libenzi | cb289d6 | 2010-01-13 09:34:36 -0800 | [diff] [blame] | 195 | res = 0; |
| 196 | else if (!no_wait) { |
Davide Libenzi | e1ad746 | 2007-05-10 22:23:19 -0700 | [diff] [blame] | 197 | __add_wait_queue(&ctx->wqh, &wait); |
Davide Libenzi | cb289d6 | 2010-01-13 09:34:36 -0800 | [diff] [blame] | 198 | for (;;) { |
Davide Libenzi | e1ad746 | 2007-05-10 22:23:19 -0700 | [diff] [blame] | 199 | set_current_state(TASK_INTERRUPTIBLE); |
| 200 | if (ctx->count > 0) { |
Davide Libenzi | cb289d6 | 2010-01-13 09:34:36 -0800 | [diff] [blame] | 201 | res = 0; |
Davide Libenzi | e1ad746 | 2007-05-10 22:23:19 -0700 | [diff] [blame] | 202 | break; |
| 203 | } |
| 204 | if (signal_pending(current)) { |
| 205 | res = -ERESTARTSYS; |
| 206 | break; |
| 207 | } |
Davide Libenzi | d48eb23 | 2007-05-18 12:02:33 -0700 | [diff] [blame] | 208 | spin_unlock_irq(&ctx->wqh.lock); |
Davide Libenzi | e1ad746 | 2007-05-10 22:23:19 -0700 | [diff] [blame] | 209 | schedule(); |
Davide Libenzi | d48eb23 | 2007-05-18 12:02:33 -0700 | [diff] [blame] | 210 | spin_lock_irq(&ctx->wqh.lock); |
Davide Libenzi | e1ad746 | 2007-05-10 22:23:19 -0700 | [diff] [blame] | 211 | } |
| 212 | __remove_wait_queue(&ctx->wqh, &wait); |
| 213 | __set_current_state(TASK_RUNNING); |
| 214 | } |
Davide Libenzi | cb289d6 | 2010-01-13 09:34:36 -0800 | [diff] [blame] | 215 | if (likely(res == 0)) { |
| 216 | eventfd_ctx_do_read(ctx, cnt); |
Davide Libenzi | e1ad746 | 2007-05-10 22:23:19 -0700 | [diff] [blame] | 217 | if (waitqueue_active(&ctx->wqh)) |
Davide Libenzi | 3951088 | 2009-03-31 15:24:23 -0700 | [diff] [blame] | 218 | wake_up_locked_poll(&ctx->wqh, POLLOUT); |
Davide Libenzi | e1ad746 | 2007-05-10 22:23:19 -0700 | [diff] [blame] | 219 | } |
Davide Libenzi | d48eb23 | 2007-05-18 12:02:33 -0700 | [diff] [blame] | 220 | spin_unlock_irq(&ctx->wqh.lock); |
Davide Libenzi | e1ad746 | 2007-05-10 22:23:19 -0700 | [diff] [blame] | 221 | |
| 222 | return res; |
| 223 | } |
Davide Libenzi | cb289d6 | 2010-01-13 09:34:36 -0800 | [diff] [blame] | 224 | EXPORT_SYMBOL_GPL(eventfd_ctx_read); |
| 225 | |
| 226 | static 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 | } |
Davide Libenzi | e1ad746 | 2007-05-10 22:23:19 -0700 | [diff] [blame] | 241 | |
| 242 | static 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; |
Davide Libenzi | d48eb23 | 2007-05-18 12:02:33 -0700 | [diff] [blame] | 256 | spin_lock_irq(&ctx->wqh.lock); |
Davide Libenzi | e1ad746 | 2007-05-10 22:23:19 -0700 | [diff] [blame] | 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 | } |
Davide Libenzi | d48eb23 | 2007-05-18 12:02:33 -0700 | [diff] [blame] | 272 | spin_unlock_irq(&ctx->wqh.lock); |
Davide Libenzi | e1ad746 | 2007-05-10 22:23:19 -0700 | [diff] [blame] | 273 | schedule(); |
Davide Libenzi | d48eb23 | 2007-05-18 12:02:33 -0700 | [diff] [blame] | 274 | spin_lock_irq(&ctx->wqh.lock); |
Davide Libenzi | e1ad746 | 2007-05-10 22:23:19 -0700 | [diff] [blame] | 275 | } |
| 276 | __remove_wait_queue(&ctx->wqh, &wait); |
| 277 | __set_current_state(TASK_RUNNING); |
| 278 | } |
Davide Libenzi | bcd0b23 | 2009-03-31 15:24:18 -0700 | [diff] [blame] | 279 | if (likely(res > 0)) { |
Davide Libenzi | e1ad746 | 2007-05-10 22:23:19 -0700 | [diff] [blame] | 280 | ctx->count += ucnt; |
| 281 | if (waitqueue_active(&ctx->wqh)) |
Davide Libenzi | 3951088 | 2009-03-31 15:24:23 -0700 | [diff] [blame] | 282 | wake_up_locked_poll(&ctx->wqh, POLLIN); |
Davide Libenzi | e1ad746 | 2007-05-10 22:23:19 -0700 | [diff] [blame] | 283 | } |
Davide Libenzi | d48eb23 | 2007-05-18 12:02:33 -0700 | [diff] [blame] | 284 | spin_unlock_irq(&ctx->wqh.lock); |
Davide Libenzi | e1ad746 | 2007-05-10 22:23:19 -0700 | [diff] [blame] | 285 | |
| 286 | return res; |
| 287 | } |
| 288 | |
Cyrill Gorcunov | cbac554 | 2012-12-17 16:04:57 -0800 | [diff] [blame] | 289 | #ifdef CONFIG_PROC_FS |
Joe Perches | a3816ab | 2014-09-29 16:08:25 -0700 | [diff] [blame] | 290 | static void eventfd_show_fdinfo(struct seq_file *m, struct file *f) |
Cyrill Gorcunov | cbac554 | 2012-12-17 16:04:57 -0800 | [diff] [blame] | 291 | { |
| 292 | struct eventfd_ctx *ctx = f->private_data; |
Cyrill Gorcunov | cbac554 | 2012-12-17 16:04:57 -0800 | [diff] [blame] | 293 | |
| 294 | spin_lock_irq(&ctx->wqh.lock); |
Joe Perches | a3816ab | 2014-09-29 16:08:25 -0700 | [diff] [blame] | 295 | seq_printf(m, "eventfd-count: %16llx\n", |
| 296 | (unsigned long long)ctx->count); |
Cyrill Gorcunov | cbac554 | 2012-12-17 16:04:57 -0800 | [diff] [blame] | 297 | spin_unlock_irq(&ctx->wqh.lock); |
Cyrill Gorcunov | cbac554 | 2012-12-17 16:04:57 -0800 | [diff] [blame] | 298 | } |
| 299 | #endif |
| 300 | |
Davide Libenzi | e1ad746 | 2007-05-10 22:23:19 -0700 | [diff] [blame] | 301 | static const struct file_operations eventfd_fops = { |
Cyrill Gorcunov | cbac554 | 2012-12-17 16:04:57 -0800 | [diff] [blame] | 302 | #ifdef CONFIG_PROC_FS |
| 303 | .show_fdinfo = eventfd_show_fdinfo, |
| 304 | #endif |
Davide Libenzi | e1ad746 | 2007-05-10 22:23:19 -0700 | [diff] [blame] | 305 | .release = eventfd_release, |
| 306 | .poll = eventfd_poll, |
| 307 | .read = eventfd_read, |
| 308 | .write = eventfd_write, |
Arnd Bergmann | 6038f37 | 2010-08-15 18:52:59 +0200 | [diff] [blame] | 309 | .llseek = noop_llseek, |
Davide Libenzi | e1ad746 | 2007-05-10 22:23:19 -0700 | [diff] [blame] | 310 | }; |
| 311 | |
Davide Libenzi | 1338901 | 2009-06-30 11:41:11 -0700 | [diff] [blame] | 312 | /** |
| 313 | * eventfd_fget - Acquire a reference of an eventfd file descriptor. |
| 314 | * @fd: [in] Eventfd file descriptor. |
| 315 | * |
| 316 | * Returns a pointer to the eventfd file structure in case of success, or the |
| 317 | * following error pointer: |
| 318 | * |
| 319 | * -EBADF : Invalid @fd file descriptor. |
| 320 | * -EINVAL : The @fd file descriptor is not an eventfd file. |
| 321 | */ |
Davide Libenzi | e1ad746 | 2007-05-10 22:23:19 -0700 | [diff] [blame] | 322 | struct file *eventfd_fget(int fd) |
| 323 | { |
| 324 | struct file *file; |
| 325 | |
| 326 | file = fget(fd); |
| 327 | if (!file) |
| 328 | return ERR_PTR(-EBADF); |
| 329 | if (file->f_op != &eventfd_fops) { |
| 330 | fput(file); |
| 331 | return ERR_PTR(-EINVAL); |
| 332 | } |
| 333 | |
| 334 | return file; |
| 335 | } |
Rusty Russell | 5718607 | 2009-06-12 22:27:09 -0600 | [diff] [blame] | 336 | EXPORT_SYMBOL_GPL(eventfd_fget); |
Davide Libenzi | e1ad746 | 2007-05-10 22:23:19 -0700 | [diff] [blame] | 337 | |
Davide Libenzi | 1338901 | 2009-06-30 11:41:11 -0700 | [diff] [blame] | 338 | /** |
| 339 | * eventfd_ctx_fdget - Acquires a reference to the internal eventfd context. |
| 340 | * @fd: [in] Eventfd file descriptor. |
| 341 | * |
| 342 | * Returns a pointer to the internal eventfd context, otherwise the error |
| 343 | * pointers returned by the following functions: |
| 344 | * |
| 345 | * eventfd_fget |
| 346 | */ |
| 347 | struct eventfd_ctx *eventfd_ctx_fdget(int fd) |
| 348 | { |
Davide Libenzi | 1338901 | 2009-06-30 11:41:11 -0700 | [diff] [blame] | 349 | struct eventfd_ctx *ctx; |
Al Viro | 36a7411 | 2013-12-23 16:51:33 -0500 | [diff] [blame] | 350 | struct fd f = fdget(fd); |
| 351 | if (!f.file) |
| 352 | return ERR_PTR(-EBADF); |
| 353 | ctx = eventfd_ctx_fileget(f.file); |
| 354 | fdput(f); |
Davide Libenzi | 1338901 | 2009-06-30 11:41:11 -0700 | [diff] [blame] | 355 | return ctx; |
| 356 | } |
| 357 | EXPORT_SYMBOL_GPL(eventfd_ctx_fdget); |
| 358 | |
| 359 | /** |
| 360 | * eventfd_ctx_fileget - Acquires a reference to the internal eventfd context. |
| 361 | * @file: [in] Eventfd file pointer. |
| 362 | * |
| 363 | * Returns a pointer to the internal eventfd context, otherwise the error |
| 364 | * pointer: |
| 365 | * |
| 366 | * -EINVAL : The @fd file descriptor is not an eventfd file. |
| 367 | */ |
| 368 | struct eventfd_ctx *eventfd_ctx_fileget(struct file *file) |
| 369 | { |
| 370 | if (file->f_op != &eventfd_fops) |
| 371 | return ERR_PTR(-EINVAL); |
| 372 | |
| 373 | return eventfd_ctx_get(file->private_data); |
| 374 | } |
| 375 | EXPORT_SYMBOL_GPL(eventfd_ctx_fileget); |
| 376 | |
Davide Libenzi | 562787a | 2009-09-22 16:43:57 -0700 | [diff] [blame] | 377 | /** |
| 378 | * eventfd_file_create - Creates an eventfd file pointer. |
| 379 | * @count: Initial eventfd counter value. |
| 380 | * @flags: Flags for the eventfd file. |
| 381 | * |
| 382 | * This function creates an eventfd file pointer, w/out installing it into |
| 383 | * the fd table. This is useful when the eventfd file is used during the |
| 384 | * initialization of data structures that require extra setup after the eventfd |
| 385 | * creation. So the eventfd creation is split into the file pointer creation |
| 386 | * phase, and the file descriptor installation phase. |
| 387 | * In this way races with userspace closing the newly installed file descriptor |
| 388 | * can be avoided. |
| 389 | * Returns an eventfd file pointer, or a proper error pointer. |
| 390 | */ |
| 391 | struct file *eventfd_file_create(unsigned int count, int flags) |
Davide Libenzi | e1ad746 | 2007-05-10 22:23:19 -0700 | [diff] [blame] | 392 | { |
Davide Libenzi | 562787a | 2009-09-22 16:43:57 -0700 | [diff] [blame] | 393 | struct file *file; |
Davide Libenzi | e1ad746 | 2007-05-10 22:23:19 -0700 | [diff] [blame] | 394 | struct eventfd_ctx *ctx; |
Davide Libenzi | e1ad746 | 2007-05-10 22:23:19 -0700 | [diff] [blame] | 395 | |
Ulrich Drepper | e38b36f | 2008-07-23 21:29:42 -0700 | [diff] [blame] | 396 | /* Check the EFD_* constants for consistency. */ |
| 397 | BUILD_BUG_ON(EFD_CLOEXEC != O_CLOEXEC); |
| 398 | BUILD_BUG_ON(EFD_NONBLOCK != O_NONBLOCK); |
| 399 | |
Davide Libenzi | bcd0b23 | 2009-03-31 15:24:18 -0700 | [diff] [blame] | 400 | if (flags & ~EFD_FLAGS_SET) |
Davide Libenzi | 562787a | 2009-09-22 16:43:57 -0700 | [diff] [blame] | 401 | return ERR_PTR(-EINVAL); |
Ulrich Drepper | b087498e | 2008-07-23 21:29:25 -0700 | [diff] [blame] | 402 | |
Davide Libenzi | e1ad746 | 2007-05-10 22:23:19 -0700 | [diff] [blame] | 403 | ctx = kmalloc(sizeof(*ctx), GFP_KERNEL); |
| 404 | if (!ctx) |
Davide Libenzi | 562787a | 2009-09-22 16:43:57 -0700 | [diff] [blame] | 405 | return ERR_PTR(-ENOMEM); |
Davide Libenzi | e1ad746 | 2007-05-10 22:23:19 -0700 | [diff] [blame] | 406 | |
Davide Libenzi | 1338901 | 2009-06-30 11:41:11 -0700 | [diff] [blame] | 407 | kref_init(&ctx->kref); |
Davide Libenzi | e1ad746 | 2007-05-10 22:23:19 -0700 | [diff] [blame] | 408 | init_waitqueue_head(&ctx->wqh); |
Davide Libenzi | e1ad746 | 2007-05-10 22:23:19 -0700 | [diff] [blame] | 409 | ctx->count = count; |
Davide Libenzi | bcd0b23 | 2009-03-31 15:24:18 -0700 | [diff] [blame] | 410 | ctx->flags = flags; |
Davide Libenzi | e1ad746 | 2007-05-10 22:23:19 -0700 | [diff] [blame] | 411 | |
Davide Libenzi | 562787a | 2009-09-22 16:43:57 -0700 | [diff] [blame] | 412 | file = anon_inode_getfile("[eventfd]", &eventfd_fops, ctx, |
Roland Dreier | 628ff7c | 2009-12-18 09:41:24 -0800 | [diff] [blame] | 413 | O_RDWR | (flags & EFD_SHARED_FCNTL_FLAGS)); |
Davide Libenzi | 562787a | 2009-09-22 16:43:57 -0700 | [diff] [blame] | 414 | if (IS_ERR(file)) |
| 415 | eventfd_free_ctx(ctx); |
| 416 | |
| 417 | return file; |
| 418 | } |
| 419 | |
| 420 | SYSCALL_DEFINE2(eventfd2, unsigned int, count, int, flags) |
| 421 | { |
| 422 | int fd, error; |
| 423 | struct file *file; |
| 424 | |
| 425 | error = get_unused_fd_flags(flags & EFD_SHARED_FCNTL_FLAGS); |
| 426 | if (error < 0) |
| 427 | return error; |
| 428 | fd = error; |
| 429 | |
| 430 | file = eventfd_file_create(count, flags); |
| 431 | if (IS_ERR(file)) { |
| 432 | error = PTR_ERR(file); |
| 433 | goto err_put_unused_fd; |
| 434 | } |
| 435 | fd_install(fd, file); |
| 436 | |
Al Viro | 2030a42 | 2008-02-23 06:46:49 -0500 | [diff] [blame] | 437 | return fd; |
Davide Libenzi | 562787a | 2009-09-22 16:43:57 -0700 | [diff] [blame] | 438 | |
| 439 | err_put_unused_fd: |
| 440 | put_unused_fd(fd); |
| 441 | |
| 442 | return error; |
Davide Libenzi | e1ad746 | 2007-05-10 22:23:19 -0700 | [diff] [blame] | 443 | } |
| 444 | |
Heiko Carstens | d4e8204 | 2009-01-14 14:14:34 +0100 | [diff] [blame] | 445 | SYSCALL_DEFINE1(eventfd, unsigned int, count) |
Ulrich Drepper | b087498e | 2008-07-23 21:29:25 -0700 | [diff] [blame] | 446 | { |
| 447 | return sys_eventfd2(count, 0); |
| 448 | } |
Davide Libenzi | bcd0b23 | 2009-03-31 15:24:18 -0700 | [diff] [blame] | 449 | |