blob: 2a701d593d35f36d2511c25df916a5d61098b019 [file] [log] [blame]
Davide Libenzie1ad7462007-05-10 22:23:19 -07001/*
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/eventfd.h>
Adrian Bunk7747cdb2008-02-06 01:36:49 -080018#include <linux/syscalls.h>
Davide Libenzie1ad7462007-05-10 22:23:19 -070019
20struct eventfd_ctx {
Davide Libenzie1ad7462007-05-10 22:23:19 -070021 wait_queue_head_t wqh;
22 /*
23 * Every time that a write(2) is performed on an eventfd, the
24 * value of the __u64 being written is added to "count" and a
25 * wakeup is performed on "wqh". A read(2) will return the "count"
26 * value to userspace, and will reset "count" to zero. The kernel
27 * size eventfd_signal() also, adds to the "count" counter and
28 * issue a wakeup.
29 */
30 __u64 count;
Davide Libenzibcd0b232009-03-31 15:24:18 -070031 unsigned int flags;
Davide Libenzie1ad7462007-05-10 22:23:19 -070032};
33
34/*
35 * Adds "n" to the eventfd counter "count". Returns "n" in case of
36 * success, or a value lower then "n" in case of coutner overflow.
37 * This function is supposed to be called by the kernel in paths
38 * that do not allow sleeping. In this function we allow the counter
39 * to reach the ULLONG_MAX value, and we signal this as overflow
40 * condition by returining a POLLERR to poll(2).
41 */
42int eventfd_signal(struct file *file, int n)
43{
44 struct eventfd_ctx *ctx = file->private_data;
45 unsigned long flags;
46
47 if (n < 0)
48 return -EINVAL;
Davide Libenzid48eb232007-05-18 12:02:33 -070049 spin_lock_irqsave(&ctx->wqh.lock, flags);
Davide Libenzie1ad7462007-05-10 22:23:19 -070050 if (ULLONG_MAX - ctx->count < n)
51 n = (int) (ULLONG_MAX - ctx->count);
52 ctx->count += n;
53 if (waitqueue_active(&ctx->wqh))
Davide Libenzi39510882009-03-31 15:24:23 -070054 wake_up_locked_poll(&ctx->wqh, POLLIN);
Davide Libenzid48eb232007-05-18 12:02:33 -070055 spin_unlock_irqrestore(&ctx->wqh.lock, flags);
Davide Libenzie1ad7462007-05-10 22:23:19 -070056
57 return n;
58}
59
60static int eventfd_release(struct inode *inode, struct file *file)
61{
62 kfree(file->private_data);
63 return 0;
64}
65
66static unsigned int eventfd_poll(struct file *file, poll_table *wait)
67{
68 struct eventfd_ctx *ctx = file->private_data;
69 unsigned int events = 0;
70 unsigned long flags;
71
72 poll_wait(file, &ctx->wqh, wait);
73
Davide Libenzid48eb232007-05-18 12:02:33 -070074 spin_lock_irqsave(&ctx->wqh.lock, flags);
Davide Libenzie1ad7462007-05-10 22:23:19 -070075 if (ctx->count > 0)
76 events |= POLLIN;
77 if (ctx->count == ULLONG_MAX)
78 events |= POLLERR;
79 if (ULLONG_MAX - 1 > ctx->count)
80 events |= POLLOUT;
Davide Libenzid48eb232007-05-18 12:02:33 -070081 spin_unlock_irqrestore(&ctx->wqh.lock, flags);
Davide Libenzie1ad7462007-05-10 22:23:19 -070082
83 return events;
84}
85
86static ssize_t eventfd_read(struct file *file, char __user *buf, size_t count,
87 loff_t *ppos)
88{
89 struct eventfd_ctx *ctx = file->private_data;
90 ssize_t res;
Davide Libenzibcd0b232009-03-31 15:24:18 -070091 __u64 ucnt = 0;
Davide Libenzie1ad7462007-05-10 22:23:19 -070092 DECLARE_WAITQUEUE(wait, current);
93
94 if (count < sizeof(ucnt))
95 return -EINVAL;
Davide Libenzid48eb232007-05-18 12:02:33 -070096 spin_lock_irq(&ctx->wqh.lock);
Davide Libenzie1ad7462007-05-10 22:23:19 -070097 res = -EAGAIN;
Davide Libenzibcd0b232009-03-31 15:24:18 -070098 if (ctx->count > 0)
Davide Libenzie1ad7462007-05-10 22:23:19 -070099 res = sizeof(ucnt);
100 else if (!(file->f_flags & O_NONBLOCK)) {
101 __add_wait_queue(&ctx->wqh, &wait);
102 for (res = 0;;) {
103 set_current_state(TASK_INTERRUPTIBLE);
104 if (ctx->count > 0) {
Davide Libenzie1ad7462007-05-10 22:23:19 -0700105 res = sizeof(ucnt);
106 break;
107 }
108 if (signal_pending(current)) {
109 res = -ERESTARTSYS;
110 break;
111 }
Davide Libenzid48eb232007-05-18 12:02:33 -0700112 spin_unlock_irq(&ctx->wqh.lock);
Davide Libenzie1ad7462007-05-10 22:23:19 -0700113 schedule();
Davide Libenzid48eb232007-05-18 12:02:33 -0700114 spin_lock_irq(&ctx->wqh.lock);
Davide Libenzie1ad7462007-05-10 22:23:19 -0700115 }
116 __remove_wait_queue(&ctx->wqh, &wait);
117 __set_current_state(TASK_RUNNING);
118 }
Davide Libenzibcd0b232009-03-31 15:24:18 -0700119 if (likely(res > 0)) {
120 ucnt = (ctx->flags & EFD_SEMAPHORE) ? 1 : ctx->count;
121 ctx->count -= ucnt;
Davide Libenzie1ad7462007-05-10 22:23:19 -0700122 if (waitqueue_active(&ctx->wqh))
Davide Libenzi39510882009-03-31 15:24:23 -0700123 wake_up_locked_poll(&ctx->wqh, POLLOUT);
Davide Libenzie1ad7462007-05-10 22:23:19 -0700124 }
Davide Libenzid48eb232007-05-18 12:02:33 -0700125 spin_unlock_irq(&ctx->wqh.lock);
Davide Libenzie1ad7462007-05-10 22:23:19 -0700126 if (res > 0 && put_user(ucnt, (__u64 __user *) buf))
127 return -EFAULT;
128
129 return res;
130}
131
132static ssize_t eventfd_write(struct file *file, const char __user *buf, size_t count,
133 loff_t *ppos)
134{
135 struct eventfd_ctx *ctx = file->private_data;
136 ssize_t res;
137 __u64 ucnt;
138 DECLARE_WAITQUEUE(wait, current);
139
140 if (count < sizeof(ucnt))
141 return -EINVAL;
142 if (copy_from_user(&ucnt, buf, sizeof(ucnt)))
143 return -EFAULT;
144 if (ucnt == ULLONG_MAX)
145 return -EINVAL;
Davide Libenzid48eb232007-05-18 12:02:33 -0700146 spin_lock_irq(&ctx->wqh.lock);
Davide Libenzie1ad7462007-05-10 22:23:19 -0700147 res = -EAGAIN;
148 if (ULLONG_MAX - ctx->count > ucnt)
149 res = sizeof(ucnt);
150 else if (!(file->f_flags & O_NONBLOCK)) {
151 __add_wait_queue(&ctx->wqh, &wait);
152 for (res = 0;;) {
153 set_current_state(TASK_INTERRUPTIBLE);
154 if (ULLONG_MAX - ctx->count > ucnt) {
155 res = sizeof(ucnt);
156 break;
157 }
158 if (signal_pending(current)) {
159 res = -ERESTARTSYS;
160 break;
161 }
Davide Libenzid48eb232007-05-18 12:02:33 -0700162 spin_unlock_irq(&ctx->wqh.lock);
Davide Libenzie1ad7462007-05-10 22:23:19 -0700163 schedule();
Davide Libenzid48eb232007-05-18 12:02:33 -0700164 spin_lock_irq(&ctx->wqh.lock);
Davide Libenzie1ad7462007-05-10 22:23:19 -0700165 }
166 __remove_wait_queue(&ctx->wqh, &wait);
167 __set_current_state(TASK_RUNNING);
168 }
Davide Libenzibcd0b232009-03-31 15:24:18 -0700169 if (likely(res > 0)) {
Davide Libenzie1ad7462007-05-10 22:23:19 -0700170 ctx->count += ucnt;
171 if (waitqueue_active(&ctx->wqh))
Davide Libenzi39510882009-03-31 15:24:23 -0700172 wake_up_locked_poll(&ctx->wqh, POLLIN);
Davide Libenzie1ad7462007-05-10 22:23:19 -0700173 }
Davide Libenzid48eb232007-05-18 12:02:33 -0700174 spin_unlock_irq(&ctx->wqh.lock);
Davide Libenzie1ad7462007-05-10 22:23:19 -0700175
176 return res;
177}
178
179static const struct file_operations eventfd_fops = {
180 .release = eventfd_release,
181 .poll = eventfd_poll,
182 .read = eventfd_read,
183 .write = eventfd_write,
184};
185
186struct file *eventfd_fget(int fd)
187{
188 struct file *file;
189
190 file = fget(fd);
191 if (!file)
192 return ERR_PTR(-EBADF);
193 if (file->f_op != &eventfd_fops) {
194 fput(file);
195 return ERR_PTR(-EINVAL);
196 }
197
198 return file;
199}
200
Heiko Carstensd4e82042009-01-14 14:14:34 +0100201SYSCALL_DEFINE2(eventfd2, unsigned int, count, int, flags)
Davide Libenzie1ad7462007-05-10 22:23:19 -0700202{
Al Viro2030a422008-02-23 06:46:49 -0500203 int fd;
Davide Libenzie1ad7462007-05-10 22:23:19 -0700204 struct eventfd_ctx *ctx;
Davide Libenzie1ad7462007-05-10 22:23:19 -0700205
Ulrich Dreppere38b36f2008-07-23 21:29:42 -0700206 /* Check the EFD_* constants for consistency. */
207 BUILD_BUG_ON(EFD_CLOEXEC != O_CLOEXEC);
208 BUILD_BUG_ON(EFD_NONBLOCK != O_NONBLOCK);
209
Davide Libenzibcd0b232009-03-31 15:24:18 -0700210 if (flags & ~EFD_FLAGS_SET)
Ulrich Drepperb087498e2008-07-23 21:29:25 -0700211 return -EINVAL;
212
Davide Libenzie1ad7462007-05-10 22:23:19 -0700213 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
214 if (!ctx)
215 return -ENOMEM;
216
217 init_waitqueue_head(&ctx->wqh);
Davide Libenzie1ad7462007-05-10 22:23:19 -0700218 ctx->count = count;
Davide Libenzibcd0b232009-03-31 15:24:18 -0700219 ctx->flags = flags;
Davide Libenzie1ad7462007-05-10 22:23:19 -0700220
221 /*
222 * When we call this, the initialization must be complete, since
223 * anon_inode_getfd() will install the fd.
224 */
Ulrich Drepperb087498e2008-07-23 21:29:25 -0700225 fd = anon_inode_getfd("[eventfd]", &eventfd_fops, ctx,
Davide Libenzibcd0b232009-03-31 15:24:18 -0700226 flags & EFD_SHARED_FCNTL_FLAGS);
Al Viro2030a422008-02-23 06:46:49 -0500227 if (fd < 0)
228 kfree(ctx);
229 return fd;
Davide Libenzie1ad7462007-05-10 22:23:19 -0700230}
231
Heiko Carstensd4e82042009-01-14 14:14:34 +0100232SYSCALL_DEFINE1(eventfd, unsigned int, count)
Ulrich Drepperb087498e2008-07-23 21:29:25 -0700233{
234 return sys_eventfd2(count, 0);
235}
Davide Libenzibcd0b232009-03-31 15:24:18 -0700236