Merge git://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux-2.6
[wrt350n-kernel.git] / fs / eventfd.c
bloba9f130cd50ac2be6009ed2091fce0ec21d134fdd
1 /*
2 * fs/eventfd.c
4 * Copyright (C) 2007 Davide Libenzi <davidel@xmailserver.org>
6 */
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>
18 #include <linux/syscalls.h>
20 struct eventfd_ctx {
21 wait_queue_head_t wqh;
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.
30 __u64 count;
34 * Adds "n" to the eventfd counter "count". Returns "n" in case of
35 * success, or a value lower then "n" in case of coutner overflow.
36 * This function is supposed to be called by the kernel in paths
37 * that do not allow sleeping. In this function we allow the counter
38 * to reach the ULLONG_MAX value, and we signal this as overflow
39 * condition by returining a POLLERR to poll(2).
41 int eventfd_signal(struct file *file, int n)
43 struct eventfd_ctx *ctx = file->private_data;
44 unsigned long flags;
46 if (n < 0)
47 return -EINVAL;
48 spin_lock_irqsave(&ctx->wqh.lock, flags);
49 if (ULLONG_MAX - ctx->count < n)
50 n = (int) (ULLONG_MAX - ctx->count);
51 ctx->count += n;
52 if (waitqueue_active(&ctx->wqh))
53 wake_up_locked(&ctx->wqh);
54 spin_unlock_irqrestore(&ctx->wqh.lock, flags);
56 return n;
59 static int eventfd_release(struct inode *inode, struct file *file)
61 kfree(file->private_data);
62 return 0;
65 static unsigned int eventfd_poll(struct file *file, poll_table *wait)
67 struct eventfd_ctx *ctx = file->private_data;
68 unsigned int events = 0;
69 unsigned long flags;
71 poll_wait(file, &ctx->wqh, wait);
73 spin_lock_irqsave(&ctx->wqh.lock, flags);
74 if (ctx->count > 0)
75 events |= POLLIN;
76 if (ctx->count == ULLONG_MAX)
77 events |= POLLERR;
78 if (ULLONG_MAX - 1 > ctx->count)
79 events |= POLLOUT;
80 spin_unlock_irqrestore(&ctx->wqh.lock, flags);
82 return events;
85 static ssize_t eventfd_read(struct file *file, char __user *buf, size_t count,
86 loff_t *ppos)
88 struct eventfd_ctx *ctx = file->private_data;
89 ssize_t res;
90 __u64 ucnt;
91 DECLARE_WAITQUEUE(wait, current);
93 if (count < sizeof(ucnt))
94 return -EINVAL;
95 spin_lock_irq(&ctx->wqh.lock);
96 res = -EAGAIN;
97 ucnt = ctx->count;
98 if (ucnt > 0)
99 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) {
105 ucnt = ctx->count;
106 res = sizeof(ucnt);
107 break;
109 if (signal_pending(current)) {
110 res = -ERESTARTSYS;
111 break;
113 spin_unlock_irq(&ctx->wqh.lock);
114 schedule();
115 spin_lock_irq(&ctx->wqh.lock);
117 __remove_wait_queue(&ctx->wqh, &wait);
118 __set_current_state(TASK_RUNNING);
120 if (res > 0) {
121 ctx->count = 0;
122 if (waitqueue_active(&ctx->wqh))
123 wake_up_locked(&ctx->wqh);
125 spin_unlock_irq(&ctx->wqh.lock);
126 if (res > 0 && put_user(ucnt, (__u64 __user *) buf))
127 return -EFAULT;
129 return res;
132 static ssize_t eventfd_write(struct file *file, const char __user *buf, size_t count,
133 loff_t *ppos)
135 struct eventfd_ctx *ctx = file->private_data;
136 ssize_t res;
137 __u64 ucnt;
138 DECLARE_WAITQUEUE(wait, current);
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;
146 spin_lock_irq(&ctx->wqh.lock);
147 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;
158 if (signal_pending(current)) {
159 res = -ERESTARTSYS;
160 break;
162 spin_unlock_irq(&ctx->wqh.lock);
163 schedule();
164 spin_lock_irq(&ctx->wqh.lock);
166 __remove_wait_queue(&ctx->wqh, &wait);
167 __set_current_state(TASK_RUNNING);
169 if (res > 0) {
170 ctx->count += ucnt;
171 if (waitqueue_active(&ctx->wqh))
172 wake_up_locked(&ctx->wqh);
174 spin_unlock_irq(&ctx->wqh.lock);
176 return res;
179 static const struct file_operations eventfd_fops = {
180 .release = eventfd_release,
181 .poll = eventfd_poll,
182 .read = eventfd_read,
183 .write = eventfd_write,
186 struct file *eventfd_fget(int fd)
188 struct file *file;
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);
198 return file;
201 asmlinkage long sys_eventfd(unsigned int count)
203 int error, fd;
204 struct eventfd_ctx *ctx;
205 struct file *file;
206 struct inode *inode;
208 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
209 if (!ctx)
210 return -ENOMEM;
212 init_waitqueue_head(&ctx->wqh);
213 ctx->count = count;
216 * When we call this, the initialization must be complete, since
217 * anon_inode_getfd() will install the fd.
219 error = anon_inode_getfd(&fd, &inode, &file, "[eventfd]",
220 &eventfd_fops, ctx);
221 if (!error)
222 return fd;
224 kfree(ctx);
225 return error;