diff options
Diffstat (limited to 'fs/signalfd.c')
-rw-r--r-- | fs/signalfd.c | 190 |
1 files changed, 29 insertions, 161 deletions
diff --git a/fs/signalfd.c b/fs/signalfd.c index a8e293d30034..aefb0be07942 100644 --- a/fs/signalfd.c +++ b/fs/signalfd.c | |||
@@ -11,8 +11,10 @@ | |||
11 | * Now using anonymous inode source. | 11 | * Now using anonymous inode source. |
12 | * Thanks to Oleg Nesterov for useful code review and suggestions. | 12 | * Thanks to Oleg Nesterov for useful code review and suggestions. |
13 | * More comments and suggestions from Arnd Bergmann. | 13 | * More comments and suggestions from Arnd Bergmann. |
14 | * Sat May 19, 2007: Davi E. M. Arnaut <davi@haxent.com.br> | 14 | * Sat May 19, 2007: Davi E. M. Arnaut <davi@haxent.com.br> |
15 | * Retrieve multiple signals with one read() call | 15 | * Retrieve multiple signals with one read() call |
16 | * Sun Jul 15, 2007: Davide Libenzi <davidel@xmailserver.org> | ||
17 | * Attach to the sighand only during read() and poll(). | ||
16 | */ | 18 | */ |
17 | 19 | ||
18 | #include <linux/file.h> | 20 | #include <linux/file.h> |
@@ -27,102 +29,12 @@ | |||
27 | #include <linux/signalfd.h> | 29 | #include <linux/signalfd.h> |
28 | 30 | ||
29 | struct signalfd_ctx { | 31 | struct signalfd_ctx { |
30 | struct list_head lnk; | ||
31 | wait_queue_head_t wqh; | ||
32 | sigset_t sigmask; | 32 | sigset_t sigmask; |
33 | struct task_struct *tsk; | ||
34 | }; | 33 | }; |
35 | 34 | ||
36 | struct signalfd_lockctx { | ||
37 | struct task_struct *tsk; | ||
38 | unsigned long flags; | ||
39 | }; | ||
40 | |||
41 | /* | ||
42 | * Tries to acquire the sighand lock. We do not increment the sighand | ||
43 | * use count, and we do not even pin the task struct, so we need to | ||
44 | * do it inside an RCU read lock, and we must be prepared for the | ||
45 | * ctx->tsk going to NULL (in signalfd_deliver()), and for the sighand | ||
46 | * being detached. We return 0 if the sighand has been detached, or | ||
47 | * 1 if we were able to pin the sighand lock. | ||
48 | */ | ||
49 | static int signalfd_lock(struct signalfd_ctx *ctx, struct signalfd_lockctx *lk) | ||
50 | { | ||
51 | struct sighand_struct *sighand = NULL; | ||
52 | |||
53 | rcu_read_lock(); | ||
54 | lk->tsk = rcu_dereference(ctx->tsk); | ||
55 | if (likely(lk->tsk != NULL)) | ||
56 | sighand = lock_task_sighand(lk->tsk, &lk->flags); | ||
57 | rcu_read_unlock(); | ||
58 | |||
59 | if (!sighand) | ||
60 | return 0; | ||
61 | |||
62 | if (!ctx->tsk) { | ||
63 | unlock_task_sighand(lk->tsk, &lk->flags); | ||
64 | return 0; | ||
65 | } | ||
66 | |||
67 | if (lk->tsk->tgid == current->tgid) | ||
68 | lk->tsk = current; | ||
69 | |||
70 | return 1; | ||
71 | } | ||
72 | |||
73 | static void signalfd_unlock(struct signalfd_lockctx *lk) | ||
74 | { | ||
75 | unlock_task_sighand(lk->tsk, &lk->flags); | ||
76 | } | ||
77 | |||
78 | /* | ||
79 | * This must be called with the sighand lock held. | ||
80 | */ | ||
81 | void signalfd_deliver(struct task_struct *tsk, int sig) | ||
82 | { | ||
83 | struct sighand_struct *sighand = tsk->sighand; | ||
84 | struct signalfd_ctx *ctx, *tmp; | ||
85 | |||
86 | BUG_ON(!sig); | ||
87 | list_for_each_entry_safe(ctx, tmp, &sighand->signalfd_list, lnk) { | ||
88 | /* | ||
89 | * We use a negative signal value as a way to broadcast that the | ||
90 | * sighand has been orphaned, so that we can notify all the | ||
91 | * listeners about this. Remember the ctx->sigmask is inverted, | ||
92 | * so if the user is interested in a signal, that corresponding | ||
93 | * bit will be zero. | ||
94 | */ | ||
95 | if (sig < 0) { | ||
96 | if (ctx->tsk == tsk) { | ||
97 | ctx->tsk = NULL; | ||
98 | list_del_init(&ctx->lnk); | ||
99 | wake_up(&ctx->wqh); | ||
100 | } | ||
101 | } else { | ||
102 | if (!sigismember(&ctx->sigmask, sig)) | ||
103 | wake_up(&ctx->wqh); | ||
104 | } | ||
105 | } | ||
106 | } | ||
107 | |||
108 | static void signalfd_cleanup(struct signalfd_ctx *ctx) | ||
109 | { | ||
110 | struct signalfd_lockctx lk; | ||
111 | |||
112 | /* | ||
113 | * This is tricky. If the sighand is gone, we do not need to remove | ||
114 | * context from the list, the list itself won't be there anymore. | ||
115 | */ | ||
116 | if (signalfd_lock(ctx, &lk)) { | ||
117 | list_del(&ctx->lnk); | ||
118 | signalfd_unlock(&lk); | ||
119 | } | ||
120 | kfree(ctx); | ||
121 | } | ||
122 | |||
123 | static int signalfd_release(struct inode *inode, struct file *file) | 35 | static int signalfd_release(struct inode *inode, struct file *file) |
124 | { | 36 | { |
125 | signalfd_cleanup(file->private_data); | 37 | kfree(file->private_data); |
126 | return 0; | 38 | return 0; |
127 | } | 39 | } |
128 | 40 | ||
@@ -130,23 +42,15 @@ static unsigned int signalfd_poll(struct file *file, poll_table *wait) | |||
130 | { | 42 | { |
131 | struct signalfd_ctx *ctx = file->private_data; | 43 | struct signalfd_ctx *ctx = file->private_data; |
132 | unsigned int events = 0; | 44 | unsigned int events = 0; |
133 | struct signalfd_lockctx lk; | ||
134 | 45 | ||
135 | poll_wait(file, &ctx->wqh, wait); | 46 | poll_wait(file, ¤t->sighand->signalfd_wqh, wait); |
136 | 47 | ||
137 | /* | 48 | spin_lock_irq(¤t->sighand->siglock); |
138 | * Let the caller get a POLLIN in this case, ala socket recv() when | 49 | if (next_signal(¤t->pending, &ctx->sigmask) || |
139 | * the peer disconnects. | 50 | next_signal(¤t->signal->shared_pending, |
140 | */ | 51 | &ctx->sigmask)) |
141 | if (signalfd_lock(ctx, &lk)) { | ||
142 | if ((lk.tsk == current && | ||
143 | next_signal(&lk.tsk->pending, &ctx->sigmask) > 0) || | ||
144 | next_signal(&lk.tsk->signal->shared_pending, | ||
145 | &ctx->sigmask) > 0) | ||
146 | events |= POLLIN; | ||
147 | signalfd_unlock(&lk); | ||
148 | } else | ||
149 | events |= POLLIN; | 52 | events |= POLLIN; |
53 | spin_unlock_irq(¤t->sighand->siglock); | ||
150 | 54 | ||
151 | return events; | 55 | return events; |
152 | } | 56 | } |
@@ -219,59 +123,46 @@ static ssize_t signalfd_dequeue(struct signalfd_ctx *ctx, siginfo_t *info, | |||
219 | int nonblock) | 123 | int nonblock) |
220 | { | 124 | { |
221 | ssize_t ret; | 125 | ssize_t ret; |
222 | struct signalfd_lockctx lk; | ||
223 | DECLARE_WAITQUEUE(wait, current); | 126 | DECLARE_WAITQUEUE(wait, current); |
224 | 127 | ||
225 | if (!signalfd_lock(ctx, &lk)) | 128 | spin_lock_irq(¤t->sighand->siglock); |
226 | return 0; | 129 | ret = dequeue_signal(current, &ctx->sigmask, info); |
227 | |||
228 | ret = dequeue_signal(lk.tsk, &ctx->sigmask, info); | ||
229 | switch (ret) { | 130 | switch (ret) { |
230 | case 0: | 131 | case 0: |
231 | if (!nonblock) | 132 | if (!nonblock) |
232 | break; | 133 | break; |
233 | ret = -EAGAIN; | 134 | ret = -EAGAIN; |
234 | default: | 135 | default: |
235 | signalfd_unlock(&lk); | 136 | spin_unlock_irq(¤t->sighand->siglock); |
236 | return ret; | 137 | return ret; |
237 | } | 138 | } |
238 | 139 | ||
239 | add_wait_queue(&ctx->wqh, &wait); | 140 | add_wait_queue(¤t->sighand->signalfd_wqh, &wait); |
240 | for (;;) { | 141 | for (;;) { |
241 | set_current_state(TASK_INTERRUPTIBLE); | 142 | set_current_state(TASK_INTERRUPTIBLE); |
242 | ret = dequeue_signal(lk.tsk, &ctx->sigmask, info); | 143 | ret = dequeue_signal(current, &ctx->sigmask, info); |
243 | signalfd_unlock(&lk); | ||
244 | if (ret != 0) | 144 | if (ret != 0) |
245 | break; | 145 | break; |
246 | if (signal_pending(current)) { | 146 | if (signal_pending(current)) { |
247 | ret = -ERESTARTSYS; | 147 | ret = -ERESTARTSYS; |
248 | break; | 148 | break; |
249 | } | 149 | } |
150 | spin_unlock_irq(¤t->sighand->siglock); | ||
250 | schedule(); | 151 | schedule(); |
251 | ret = signalfd_lock(ctx, &lk); | 152 | spin_lock_irq(¤t->sighand->siglock); |
252 | if (unlikely(!ret)) { | ||
253 | /* | ||
254 | * Let the caller read zero byte, ala socket | ||
255 | * recv() when the peer disconnect. This test | ||
256 | * must be done before doing a dequeue_signal(), | ||
257 | * because if the sighand has been orphaned, | ||
258 | * the dequeue_signal() call is going to crash | ||
259 | * because ->sighand will be long gone. | ||
260 | */ | ||
261 | break; | ||
262 | } | ||
263 | } | 153 | } |
154 | spin_unlock_irq(¤t->sighand->siglock); | ||
264 | 155 | ||
265 | remove_wait_queue(&ctx->wqh, &wait); | 156 | remove_wait_queue(¤t->sighand->signalfd_wqh, &wait); |
266 | __set_current_state(TASK_RUNNING); | 157 | __set_current_state(TASK_RUNNING); |
267 | 158 | ||
268 | return ret; | 159 | return ret; |
269 | } | 160 | } |
270 | 161 | ||
271 | /* | 162 | /* |
272 | * Returns either the size of a "struct signalfd_siginfo", or zero if the | 163 | * Returns a multiple of the size of a "struct signalfd_siginfo", or a negative |
273 | * sighand we are attached to, has been orphaned. The "count" parameter | 164 | * error code. The "count" parameter must be at least the size of a |
274 | * must be at least the size of a "struct signalfd_siginfo". | 165 | * "struct signalfd_siginfo". |
275 | */ | 166 | */ |
276 | static ssize_t signalfd_read(struct file *file, char __user *buf, size_t count, | 167 | static ssize_t signalfd_read(struct file *file, char __user *buf, size_t count, |
277 | loff_t *ppos) | 168 | loff_t *ppos) |
@@ -287,7 +178,6 @@ static ssize_t signalfd_read(struct file *file, char __user *buf, size_t count, | |||
287 | return -EINVAL; | 178 | return -EINVAL; |
288 | 179 | ||
289 | siginfo = (struct signalfd_siginfo __user *) buf; | 180 | siginfo = (struct signalfd_siginfo __user *) buf; |
290 | |||
291 | do { | 181 | do { |
292 | ret = signalfd_dequeue(ctx, &info, nonblock); | 182 | ret = signalfd_dequeue(ctx, &info, nonblock); |
293 | if (unlikely(ret <= 0)) | 183 | if (unlikely(ret <= 0)) |
@@ -300,7 +190,7 @@ static ssize_t signalfd_read(struct file *file, char __user *buf, size_t count, | |||
300 | nonblock = 1; | 190 | nonblock = 1; |
301 | } while (--count); | 191 | } while (--count); |
302 | 192 | ||
303 | return total ? total : ret; | 193 | return total ? total: ret; |
304 | } | 194 | } |
305 | 195 | ||
306 | static const struct file_operations signalfd_fops = { | 196 | static const struct file_operations signalfd_fops = { |
@@ -309,20 +199,13 @@ static const struct file_operations signalfd_fops = { | |||
309 | .read = signalfd_read, | 199 | .read = signalfd_read, |
310 | }; | 200 | }; |
311 | 201 | ||
312 | /* | ||
313 | * Create a file descriptor that is associated with our signal | ||
314 | * state. We can pass it around to others if we want to, but | ||
315 | * it will always be _our_ signal state. | ||
316 | */ | ||
317 | asmlinkage long sys_signalfd(int ufd, sigset_t __user *user_mask, size_t sizemask) | 202 | asmlinkage long sys_signalfd(int ufd, sigset_t __user *user_mask, size_t sizemask) |
318 | { | 203 | { |
319 | int error; | 204 | int error; |
320 | sigset_t sigmask; | 205 | sigset_t sigmask; |
321 | struct signalfd_ctx *ctx; | 206 | struct signalfd_ctx *ctx; |
322 | struct sighand_struct *sighand; | ||
323 | struct file *file; | 207 | struct file *file; |
324 | struct inode *inode; | 208 | struct inode *inode; |
325 | struct signalfd_lockctx lk; | ||
326 | 209 | ||
327 | if (sizemask != sizeof(sigset_t) || | 210 | if (sizemask != sizeof(sigset_t) || |
328 | copy_from_user(&sigmask, user_mask, sizeof(sigmask))) | 211 | copy_from_user(&sigmask, user_mask, sizeof(sigmask))) |
@@ -335,17 +218,7 @@ asmlinkage long sys_signalfd(int ufd, sigset_t __user *user_mask, size_t sizemas | |||
335 | if (!ctx) | 218 | if (!ctx) |
336 | return -ENOMEM; | 219 | return -ENOMEM; |
337 | 220 | ||
338 | init_waitqueue_head(&ctx->wqh); | ||
339 | ctx->sigmask = sigmask; | 221 | ctx->sigmask = sigmask; |
340 | ctx->tsk = current->group_leader; | ||
341 | |||
342 | sighand = current->sighand; | ||
343 | /* | ||
344 | * Add this fd to the list of signal listeners. | ||
345 | */ | ||
346 | spin_lock_irq(&sighand->siglock); | ||
347 | list_add_tail(&ctx->lnk, &sighand->signalfd_list); | ||
348 | spin_unlock_irq(&sighand->siglock); | ||
349 | 222 | ||
350 | /* | 223 | /* |
351 | * When we call this, the initialization must be complete, since | 224 | * When we call this, the initialization must be complete, since |
@@ -364,23 +237,18 @@ asmlinkage long sys_signalfd(int ufd, sigset_t __user *user_mask, size_t sizemas | |||
364 | fput(file); | 237 | fput(file); |
365 | return -EINVAL; | 238 | return -EINVAL; |
366 | } | 239 | } |
367 | /* | 240 | spin_lock_irq(¤t->sighand->siglock); |
368 | * We need to be prepared of the fact that the sighand this fd | 241 | ctx->sigmask = sigmask; |
369 | * is attached to, has been detched. In that case signalfd_lock() | 242 | spin_unlock_irq(¤t->sighand->siglock); |
370 | * will return 0, and we'll just skip setting the new mask. | 243 | |
371 | */ | 244 | wake_up(¤t->sighand->signalfd_wqh); |
372 | if (signalfd_lock(ctx, &lk)) { | ||
373 | ctx->sigmask = sigmask; | ||
374 | signalfd_unlock(&lk); | ||
375 | } | ||
376 | wake_up(&ctx->wqh); | ||
377 | fput(file); | 245 | fput(file); |
378 | } | 246 | } |
379 | 247 | ||
380 | return ufd; | 248 | return ufd; |
381 | 249 | ||
382 | err_fdalloc: | 250 | err_fdalloc: |
383 | signalfd_cleanup(ctx); | 251 | kfree(ctx); |
384 | return error; | 252 | return error; |
385 | } | 253 | } |
386 | 254 | ||