]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/fs/fcntl.c | |
3 | * | |
4 | * Copyright (C) 1991, 1992 Linus Torvalds | |
5 | */ | |
6 | ||
7 | #include <linux/syscalls.h> | |
8 | #include <linux/init.h> | |
9 | #include <linux/mm.h> | |
10 | #include <linux/fs.h> | |
11 | #include <linux/file.h> | |
9f3acc31 | 12 | #include <linux/fdtable.h> |
16f7e0fe | 13 | #include <linux/capability.h> |
1da177e4 | 14 | #include <linux/dnotify.h> |
1da177e4 LT |
15 | #include <linux/slab.h> |
16 | #include <linux/module.h> | |
35f3d14d | 17 | #include <linux/pipe_fs_i.h> |
1da177e4 LT |
18 | #include <linux/security.h> |
19 | #include <linux/ptrace.h> | |
7ed20e1a | 20 | #include <linux/signal.h> |
ab2af1f5 | 21 | #include <linux/rcupdate.h> |
b488893a | 22 | #include <linux/pid_namespace.h> |
1d151c33 | 23 | #include <linux/user_namespace.h> |
1da177e4 LT |
24 | |
25 | #include <asm/poll.h> | |
26 | #include <asm/siginfo.h> | |
27 | #include <asm/uaccess.h> | |
28 | ||
76398425 | 29 | #define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | O_DIRECT | O_NOATIME) |
1da177e4 LT |
30 | |
31 | static int setfl(int fd, struct file * filp, unsigned long arg) | |
32 | { | |
0f7fc9e4 | 33 | struct inode * inode = filp->f_path.dentry->d_inode; |
1da177e4 LT |
34 | int error = 0; |
35 | ||
7d95c8f2 | 36 | /* |
37 | * O_APPEND cannot be cleared if the file is marked as append-only | |
38 | * and the file is open for write. | |
39 | */ | |
40 | if (((arg ^ filp->f_flags) & O_APPEND) && IS_APPEND(inode)) | |
1da177e4 LT |
41 | return -EPERM; |
42 | ||
43 | /* O_NOATIME can only be set by the owner or superuser */ | |
44 | if ((arg & O_NOATIME) && !(filp->f_flags & O_NOATIME)) | |
2e149670 | 45 | if (!inode_owner_or_capable(inode)) |
1da177e4 LT |
46 | return -EPERM; |
47 | ||
48 | /* required for strict SunOS emulation */ | |
49 | if (O_NONBLOCK != O_NDELAY) | |
50 | if (arg & O_NDELAY) | |
51 | arg |= O_NONBLOCK; | |
52 | ||
53 | if (arg & O_DIRECT) { | |
54 | if (!filp->f_mapping || !filp->f_mapping->a_ops || | |
55 | !filp->f_mapping->a_ops->direct_IO) | |
56 | return -EINVAL; | |
57 | } | |
58 | ||
59 | if (filp->f_op && filp->f_op->check_flags) | |
60 | error = filp->f_op->check_flags(arg); | |
61 | if (error) | |
62 | return error; | |
63 | ||
218d11a8 | 64 | /* |
76398425 | 65 | * ->fasync() is responsible for setting the FASYNC bit. |
218d11a8 | 66 | */ |
76398425 JC |
67 | if (((arg ^ filp->f_flags) & FASYNC) && filp->f_op && |
68 | filp->f_op->fasync) { | |
69 | error = filp->f_op->fasync(fd, filp, (arg & FASYNC) != 0); | |
70 | if (error < 0) | |
71 | goto out; | |
60aa4924 JC |
72 | if (error > 0) |
73 | error = 0; | |
1da177e4 | 74 | } |
db1dd4d3 | 75 | spin_lock(&filp->f_lock); |
1da177e4 | 76 | filp->f_flags = (arg & SETFL_MASK) | (filp->f_flags & ~SETFL_MASK); |
db1dd4d3 | 77 | spin_unlock(&filp->f_lock); |
76398425 | 78 | |
1da177e4 | 79 | out: |
1da177e4 LT |
80 | return error; |
81 | } | |
82 | ||
609d7fa9 | 83 | static void f_modown(struct file *filp, struct pid *pid, enum pid_type type, |
2f38d70f | 84 | int force) |
1da177e4 | 85 | { |
80e1e823 | 86 | write_lock_irq(&filp->f_owner.lock); |
1da177e4 | 87 | if (force || !filp->f_owner.pid) { |
609d7fa9 EB |
88 | put_pid(filp->f_owner.pid); |
89 | filp->f_owner.pid = get_pid(pid); | |
90 | filp->f_owner.pid_type = type; | |
2f38d70f ON |
91 | |
92 | if (pid) { | |
93 | const struct cred *cred = current_cred(); | |
94 | filp->f_owner.uid = cred->uid; | |
95 | filp->f_owner.euid = cred->euid; | |
96 | } | |
1da177e4 | 97 | } |
80e1e823 | 98 | write_unlock_irq(&filp->f_owner.lock); |
1da177e4 LT |
99 | } |
100 | ||
609d7fa9 EB |
101 | int __f_setown(struct file *filp, struct pid *pid, enum pid_type type, |
102 | int force) | |
1da177e4 LT |
103 | { |
104 | int err; | |
2f38d70f | 105 | |
1da177e4 LT |
106 | err = security_file_set_fowner(filp); |
107 | if (err) | |
108 | return err; | |
109 | ||
2f38d70f | 110 | f_modown(filp, pid, type, force); |
1da177e4 LT |
111 | return 0; |
112 | } | |
609d7fa9 | 113 | EXPORT_SYMBOL(__f_setown); |
1da177e4 | 114 | |
609d7fa9 EB |
115 | int f_setown(struct file *filp, unsigned long arg, int force) |
116 | { | |
117 | enum pid_type type; | |
118 | struct pid *pid; | |
119 | int who = arg; | |
120 | int result; | |
121 | type = PIDTYPE_PID; | |
122 | if (who < 0) { | |
123 | type = PIDTYPE_PGID; | |
124 | who = -who; | |
125 | } | |
126 | rcu_read_lock(); | |
b488893a | 127 | pid = find_vpid(who); |
609d7fa9 EB |
128 | result = __f_setown(filp, pid, type, force); |
129 | rcu_read_unlock(); | |
130 | return result; | |
131 | } | |
1da177e4 LT |
132 | EXPORT_SYMBOL(f_setown); |
133 | ||
134 | void f_delown(struct file *filp) | |
135 | { | |
2f38d70f | 136 | f_modown(filp, NULL, PIDTYPE_PID, 1); |
609d7fa9 EB |
137 | } |
138 | ||
139 | pid_t f_getown(struct file *filp) | |
140 | { | |
141 | pid_t pid; | |
43fa1adb | 142 | read_lock(&filp->f_owner.lock); |
6c5f3e7b | 143 | pid = pid_vnr(filp->f_owner.pid); |
609d7fa9 EB |
144 | if (filp->f_owner.pid_type == PIDTYPE_PGID) |
145 | pid = -pid; | |
43fa1adb | 146 | read_unlock(&filp->f_owner.lock); |
609d7fa9 | 147 | return pid; |
1da177e4 LT |
148 | } |
149 | ||
ba0a6c9f PZ |
150 | static int f_setown_ex(struct file *filp, unsigned long arg) |
151 | { | |
152 | struct f_owner_ex * __user owner_p = (void * __user)arg; | |
153 | struct f_owner_ex owner; | |
154 | struct pid *pid; | |
155 | int type; | |
156 | int ret; | |
157 | ||
158 | ret = copy_from_user(&owner, owner_p, sizeof(owner)); | |
159 | if (ret) | |
5b54470d | 160 | return -EFAULT; |
ba0a6c9f PZ |
161 | |
162 | switch (owner.type) { | |
163 | case F_OWNER_TID: | |
164 | type = PIDTYPE_MAX; | |
165 | break; | |
166 | ||
167 | case F_OWNER_PID: | |
168 | type = PIDTYPE_PID; | |
169 | break; | |
170 | ||
978b4053 | 171 | case F_OWNER_PGRP: |
ba0a6c9f PZ |
172 | type = PIDTYPE_PGID; |
173 | break; | |
174 | ||
175 | default: | |
176 | return -EINVAL; | |
177 | } | |
178 | ||
179 | rcu_read_lock(); | |
180 | pid = find_vpid(owner.pid); | |
181 | if (owner.pid && !pid) | |
182 | ret = -ESRCH; | |
183 | else | |
184 | ret = __f_setown(filp, pid, type, 1); | |
185 | rcu_read_unlock(); | |
186 | ||
187 | return ret; | |
188 | } | |
189 | ||
190 | static int f_getown_ex(struct file *filp, unsigned long arg) | |
191 | { | |
192 | struct f_owner_ex * __user owner_p = (void * __user)arg; | |
193 | struct f_owner_ex owner; | |
194 | int ret = 0; | |
195 | ||
196 | read_lock(&filp->f_owner.lock); | |
197 | owner.pid = pid_vnr(filp->f_owner.pid); | |
198 | switch (filp->f_owner.pid_type) { | |
199 | case PIDTYPE_MAX: | |
200 | owner.type = F_OWNER_TID; | |
201 | break; | |
202 | ||
203 | case PIDTYPE_PID: | |
204 | owner.type = F_OWNER_PID; | |
205 | break; | |
206 | ||
207 | case PIDTYPE_PGID: | |
978b4053 | 208 | owner.type = F_OWNER_PGRP; |
ba0a6c9f PZ |
209 | break; |
210 | ||
211 | default: | |
212 | WARN_ON(1); | |
213 | ret = -EINVAL; | |
214 | break; | |
215 | } | |
216 | read_unlock(&filp->f_owner.lock); | |
217 | ||
5b54470d | 218 | if (!ret) { |
ba0a6c9f | 219 | ret = copy_to_user(owner_p, &owner, sizeof(owner)); |
5b54470d DC |
220 | if (ret) |
221 | ret = -EFAULT; | |
222 | } | |
ba0a6c9f PZ |
223 | return ret; |
224 | } | |
225 | ||
1d151c33 CG |
226 | #ifdef CONFIG_CHECKPOINT_RESTORE |
227 | static int f_getowner_uids(struct file *filp, unsigned long arg) | |
228 | { | |
229 | struct user_namespace *user_ns = current_user_ns(); | |
230 | uid_t * __user dst = (void * __user)arg; | |
231 | uid_t src[2]; | |
232 | int err; | |
233 | ||
234 | read_lock(&filp->f_owner.lock); | |
235 | src[0] = from_kuid(user_ns, filp->f_owner.uid); | |
236 | src[1] = from_kuid(user_ns, filp->f_owner.euid); | |
237 | read_unlock(&filp->f_owner.lock); | |
238 | ||
239 | err = put_user(src[0], &dst[0]); | |
240 | err |= put_user(src[1], &dst[1]); | |
241 | ||
242 | return err; | |
243 | } | |
244 | #else | |
245 | static int f_getowner_uids(struct file *filp, unsigned long arg) | |
246 | { | |
247 | return -EINVAL; | |
248 | } | |
249 | #endif | |
250 | ||
1da177e4 LT |
251 | static long do_fcntl(int fd, unsigned int cmd, unsigned long arg, |
252 | struct file *filp) | |
253 | { | |
254 | long err = -EINVAL; | |
255 | ||
256 | switch (cmd) { | |
257 | case F_DUPFD: | |
fe17f22d AV |
258 | err = f_dupfd(arg, filp, 0); |
259 | break; | |
22d2b35b | 260 | case F_DUPFD_CLOEXEC: |
fe17f22d | 261 | err = f_dupfd(arg, filp, FD_CLOEXEC); |
1da177e4 LT |
262 | break; |
263 | case F_GETFD: | |
264 | err = get_close_on_exec(fd) ? FD_CLOEXEC : 0; | |
265 | break; | |
266 | case F_SETFD: | |
267 | err = 0; | |
268 | set_close_on_exec(fd, arg & FD_CLOEXEC); | |
269 | break; | |
270 | case F_GETFL: | |
271 | err = filp->f_flags; | |
272 | break; | |
273 | case F_SETFL: | |
274 | err = setfl(fd, filp, arg); | |
275 | break; | |
276 | case F_GETLK: | |
277 | err = fcntl_getlk(filp, (struct flock __user *) arg); | |
278 | break; | |
279 | case F_SETLK: | |
280 | case F_SETLKW: | |
c293621b | 281 | err = fcntl_setlk(fd, filp, cmd, (struct flock __user *) arg); |
1da177e4 LT |
282 | break; |
283 | case F_GETOWN: | |
284 | /* | |
285 | * XXX If f_owner is a process group, the | |
286 | * negative return value will get converted | |
287 | * into an error. Oops. If we keep the | |
288 | * current syscall conventions, the only way | |
289 | * to fix this will be in libc. | |
290 | */ | |
609d7fa9 | 291 | err = f_getown(filp); |
1da177e4 LT |
292 | force_successful_syscall_return(); |
293 | break; | |
294 | case F_SETOWN: | |
295 | err = f_setown(filp, arg, 1); | |
296 | break; | |
ba0a6c9f PZ |
297 | case F_GETOWN_EX: |
298 | err = f_getown_ex(filp, arg); | |
299 | break; | |
300 | case F_SETOWN_EX: | |
301 | err = f_setown_ex(filp, arg); | |
302 | break; | |
1d151c33 CG |
303 | case F_GETOWNER_UIDS: |
304 | err = f_getowner_uids(filp, arg); | |
305 | break; | |
1da177e4 LT |
306 | case F_GETSIG: |
307 | err = filp->f_owner.signum; | |
308 | break; | |
309 | case F_SETSIG: | |
310 | /* arg == 0 restores default behaviour. */ | |
7ed20e1a | 311 | if (!valid_signal(arg)) { |
1da177e4 LT |
312 | break; |
313 | } | |
314 | err = 0; | |
315 | filp->f_owner.signum = arg; | |
316 | break; | |
317 | case F_GETLEASE: | |
318 | err = fcntl_getlease(filp); | |
319 | break; | |
320 | case F_SETLEASE: | |
321 | err = fcntl_setlease(fd, filp, arg); | |
322 | break; | |
323 | case F_NOTIFY: | |
324 | err = fcntl_dirnotify(fd, filp, arg); | |
325 | break; | |
35f3d14d JA |
326 | case F_SETPIPE_SZ: |
327 | case F_GETPIPE_SZ: | |
328 | err = pipe_fcntl(filp, cmd, arg); | |
329 | break; | |
1da177e4 LT |
330 | default: |
331 | break; | |
332 | } | |
333 | return err; | |
334 | } | |
335 | ||
1abf0c71 AV |
336 | static int check_fcntl_cmd(unsigned cmd) |
337 | { | |
338 | switch (cmd) { | |
339 | case F_DUPFD: | |
340 | case F_DUPFD_CLOEXEC: | |
341 | case F_GETFD: | |
342 | case F_SETFD: | |
343 | case F_GETFL: | |
344 | return 1; | |
345 | } | |
346 | return 0; | |
347 | } | |
348 | ||
a26eab24 | 349 | SYSCALL_DEFINE3(fcntl, unsigned int, fd, unsigned int, cmd, unsigned long, arg) |
1da177e4 LT |
350 | { |
351 | struct file *filp; | |
545ec2c7 | 352 | int fput_needed; |
1da177e4 LT |
353 | long err = -EBADF; |
354 | ||
545ec2c7 | 355 | filp = fget_raw_light(fd, &fput_needed); |
1da177e4 LT |
356 | if (!filp) |
357 | goto out; | |
358 | ||
1abf0c71 | 359 | if (unlikely(filp->f_mode & FMODE_PATH)) { |
545ec2c7 AV |
360 | if (!check_fcntl_cmd(cmd)) |
361 | goto out1; | |
1abf0c71 AV |
362 | } |
363 | ||
1da177e4 | 364 | err = security_file_fcntl(filp, cmd, arg); |
545ec2c7 AV |
365 | if (!err) |
366 | err = do_fcntl(fd, cmd, arg, filp); | |
1da177e4 | 367 | |
545ec2c7 AV |
368 | out1: |
369 | fput_light(filp, fput_needed); | |
1da177e4 LT |
370 | out: |
371 | return err; | |
372 | } | |
373 | ||
374 | #if BITS_PER_LONG == 32 | |
a26eab24 HC |
375 | SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd, |
376 | unsigned long, arg) | |
1da177e4 LT |
377 | { |
378 | struct file * filp; | |
545ec2c7 AV |
379 | long err = -EBADF; |
380 | int fput_needed; | |
1da177e4 | 381 | |
545ec2c7 | 382 | filp = fget_raw_light(fd, &fput_needed); |
1da177e4 LT |
383 | if (!filp) |
384 | goto out; | |
385 | ||
1abf0c71 | 386 | if (unlikely(filp->f_mode & FMODE_PATH)) { |
545ec2c7 AV |
387 | if (!check_fcntl_cmd(cmd)) |
388 | goto out1; | |
1abf0c71 AV |
389 | } |
390 | ||
1da177e4 | 391 | err = security_file_fcntl(filp, cmd, arg); |
545ec2c7 AV |
392 | if (err) |
393 | goto out1; | |
1da177e4 LT |
394 | |
395 | switch (cmd) { | |
396 | case F_GETLK64: | |
397 | err = fcntl_getlk64(filp, (struct flock64 __user *) arg); | |
398 | break; | |
399 | case F_SETLK64: | |
400 | case F_SETLKW64: | |
c293621b PS |
401 | err = fcntl_setlk64(fd, filp, cmd, |
402 | (struct flock64 __user *) arg); | |
1da177e4 LT |
403 | break; |
404 | default: | |
405 | err = do_fcntl(fd, cmd, arg, filp); | |
406 | break; | |
407 | } | |
545ec2c7 AV |
408 | out1: |
409 | fput_light(filp, fput_needed); | |
1da177e4 LT |
410 | out: |
411 | return err; | |
412 | } | |
413 | #endif | |
414 | ||
415 | /* Table to convert sigio signal codes into poll band bitmaps */ | |
416 | ||
fa3536cc | 417 | static const long band_table[NSIGPOLL] = { |
1da177e4 LT |
418 | POLLIN | POLLRDNORM, /* POLL_IN */ |
419 | POLLOUT | POLLWRNORM | POLLWRBAND, /* POLL_OUT */ | |
420 | POLLIN | POLLRDNORM | POLLMSG, /* POLL_MSG */ | |
421 | POLLERR, /* POLL_ERR */ | |
422 | POLLPRI | POLLRDBAND, /* POLL_PRI */ | |
423 | POLLHUP | POLLERR /* POLL_HUP */ | |
424 | }; | |
425 | ||
426 | static inline int sigio_perm(struct task_struct *p, | |
427 | struct fown_struct *fown, int sig) | |
428 | { | |
c69e8d9c DH |
429 | const struct cred *cred; |
430 | int ret; | |
431 | ||
432 | rcu_read_lock(); | |
433 | cred = __task_cred(p); | |
8e96e3b7 EB |
434 | ret = ((uid_eq(fown->euid, GLOBAL_ROOT_UID) || |
435 | uid_eq(fown->euid, cred->suid) || uid_eq(fown->euid, cred->uid) || | |
436 | uid_eq(fown->uid, cred->suid) || uid_eq(fown->uid, cred->uid)) && | |
c69e8d9c DH |
437 | !security_file_send_sigiotask(p, fown, sig)); |
438 | rcu_read_unlock(); | |
439 | return ret; | |
1da177e4 LT |
440 | } |
441 | ||
442 | static void send_sigio_to_task(struct task_struct *p, | |
8eeee4e2 | 443 | struct fown_struct *fown, |
ba0a6c9f | 444 | int fd, int reason, int group) |
1da177e4 | 445 | { |
8eeee4e2 ON |
446 | /* |
447 | * F_SETSIG can change ->signum lockless in parallel, make | |
448 | * sure we read it once and use the same value throughout. | |
449 | */ | |
450 | int signum = ACCESS_ONCE(fown->signum); | |
451 | ||
452 | if (!sigio_perm(p, fown, signum)) | |
1da177e4 LT |
453 | return; |
454 | ||
8eeee4e2 | 455 | switch (signum) { |
1da177e4 LT |
456 | siginfo_t si; |
457 | default: | |
458 | /* Queue a rt signal with the appropriate fd as its | |
459 | value. We use SI_SIGIO as the source, not | |
460 | SI_KERNEL, since kernel signals always get | |
461 | delivered even if we can't queue. Failure to | |
462 | queue in this case _should_ be reported; we fall | |
463 | back to SIGIO in that case. --sct */ | |
8eeee4e2 | 464 | si.si_signo = signum; |
1da177e4 LT |
465 | si.si_errno = 0; |
466 | si.si_code = reason; | |
467 | /* Make sure we are called with one of the POLL_* | |
468 | reasons, otherwise we could leak kernel stack into | |
469 | userspace. */ | |
f6298aab | 470 | BUG_ON((reason & __SI_MASK) != __SI_POLL); |
1da177e4 LT |
471 | if (reason - POLL_IN >= NSIGPOLL) |
472 | si.si_band = ~0L; | |
473 | else | |
474 | si.si_band = band_table[reason - POLL_IN]; | |
475 | si.si_fd = fd; | |
ba0a6c9f | 476 | if (!do_send_sig_info(signum, &si, p, group)) |
1da177e4 LT |
477 | break; |
478 | /* fall-through: fall back on the old plain SIGIO signal */ | |
479 | case 0: | |
ba0a6c9f | 480 | do_send_sig_info(SIGIO, SEND_SIG_PRIV, p, group); |
1da177e4 LT |
481 | } |
482 | } | |
483 | ||
484 | void send_sigio(struct fown_struct *fown, int fd, int band) | |
485 | { | |
486 | struct task_struct *p; | |
609d7fa9 EB |
487 | enum pid_type type; |
488 | struct pid *pid; | |
ba0a6c9f | 489 | int group = 1; |
1da177e4 LT |
490 | |
491 | read_lock(&fown->lock); | |
ba0a6c9f | 492 | |
609d7fa9 | 493 | type = fown->pid_type; |
ba0a6c9f PZ |
494 | if (type == PIDTYPE_MAX) { |
495 | group = 0; | |
496 | type = PIDTYPE_PID; | |
497 | } | |
498 | ||
1da177e4 LT |
499 | pid = fown->pid; |
500 | if (!pid) | |
501 | goto out_unlock_fown; | |
502 | ||
503 | read_lock(&tasklist_lock); | |
609d7fa9 | 504 | do_each_pid_task(pid, type, p) { |
ba0a6c9f | 505 | send_sigio_to_task(p, fown, fd, band, group); |
609d7fa9 | 506 | } while_each_pid_task(pid, type, p); |
1da177e4 LT |
507 | read_unlock(&tasklist_lock); |
508 | out_unlock_fown: | |
509 | read_unlock(&fown->lock); | |
510 | } | |
511 | ||
512 | static void send_sigurg_to_task(struct task_struct *p, | |
ba0a6c9f | 513 | struct fown_struct *fown, int group) |
1da177e4 LT |
514 | { |
515 | if (sigio_perm(p, fown, SIGURG)) | |
ba0a6c9f | 516 | do_send_sig_info(SIGURG, SEND_SIG_PRIV, p, group); |
1da177e4 LT |
517 | } |
518 | ||
519 | int send_sigurg(struct fown_struct *fown) | |
520 | { | |
521 | struct task_struct *p; | |
609d7fa9 EB |
522 | enum pid_type type; |
523 | struct pid *pid; | |
ba0a6c9f | 524 | int group = 1; |
609d7fa9 | 525 | int ret = 0; |
1da177e4 LT |
526 | |
527 | read_lock(&fown->lock); | |
ba0a6c9f | 528 | |
609d7fa9 | 529 | type = fown->pid_type; |
ba0a6c9f PZ |
530 | if (type == PIDTYPE_MAX) { |
531 | group = 0; | |
532 | type = PIDTYPE_PID; | |
533 | } | |
534 | ||
1da177e4 LT |
535 | pid = fown->pid; |
536 | if (!pid) | |
537 | goto out_unlock_fown; | |
538 | ||
539 | ret = 1; | |
540 | ||
541 | read_lock(&tasklist_lock); | |
609d7fa9 | 542 | do_each_pid_task(pid, type, p) { |
ba0a6c9f | 543 | send_sigurg_to_task(p, fown, group); |
609d7fa9 | 544 | } while_each_pid_task(pid, type, p); |
1da177e4 LT |
545 | read_unlock(&tasklist_lock); |
546 | out_unlock_fown: | |
547 | read_unlock(&fown->lock); | |
548 | return ret; | |
549 | } | |
550 | ||
989a2979 | 551 | static DEFINE_SPINLOCK(fasync_lock); |
e18b890b | 552 | static struct kmem_cache *fasync_cache __read_mostly; |
1da177e4 | 553 | |
989a2979 ED |
554 | static void fasync_free_rcu(struct rcu_head *head) |
555 | { | |
556 | kmem_cache_free(fasync_cache, | |
557 | container_of(head, struct fasync_struct, fa_rcu)); | |
558 | } | |
559 | ||
1da177e4 | 560 | /* |
53281b6d LT |
561 | * Remove a fasync entry. If successfully removed, return |
562 | * positive and clear the FASYNC flag. If no entry exists, | |
563 | * do nothing and return 0. | |
564 | * | |
565 | * NOTE! It is very important that the FASYNC flag always | |
566 | * match the state "is the filp on a fasync list". | |
567 | * | |
1da177e4 | 568 | */ |
f7347ce4 | 569 | int fasync_remove_entry(struct file *filp, struct fasync_struct **fapp) |
1da177e4 LT |
570 | { |
571 | struct fasync_struct *fa, **fp; | |
1da177e4 LT |
572 | int result = 0; |
573 | ||
53281b6d | 574 | spin_lock(&filp->f_lock); |
989a2979 | 575 | spin_lock(&fasync_lock); |
53281b6d LT |
576 | for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) { |
577 | if (fa->fa_file != filp) | |
578 | continue; | |
989a2979 ED |
579 | |
580 | spin_lock_irq(&fa->fa_lock); | |
581 | fa->fa_file = NULL; | |
582 | spin_unlock_irq(&fa->fa_lock); | |
583 | ||
53281b6d | 584 | *fp = fa->fa_next; |
989a2979 | 585 | call_rcu(&fa->fa_rcu, fasync_free_rcu); |
53281b6d LT |
586 | filp->f_flags &= ~FASYNC; |
587 | result = 1; | |
588 | break; | |
1da177e4 | 589 | } |
989a2979 | 590 | spin_unlock(&fasync_lock); |
53281b6d LT |
591 | spin_unlock(&filp->f_lock); |
592 | return result; | |
593 | } | |
594 | ||
f7347ce4 LT |
595 | struct fasync_struct *fasync_alloc(void) |
596 | { | |
597 | return kmem_cache_alloc(fasync_cache, GFP_KERNEL); | |
598 | } | |
599 | ||
53281b6d | 600 | /* |
f7347ce4 LT |
601 | * NOTE! This can be used only for unused fasync entries: |
602 | * entries that actually got inserted on the fasync list | |
603 | * need to be released by rcu - see fasync_remove_entry. | |
53281b6d | 604 | */ |
f7347ce4 | 605 | void fasync_free(struct fasync_struct *new) |
53281b6d | 606 | { |
f7347ce4 LT |
607 | kmem_cache_free(fasync_cache, new); |
608 | } | |
53281b6d | 609 | |
f7347ce4 LT |
610 | /* |
611 | * Insert a new entry into the fasync list. Return the pointer to the | |
612 | * old one if we didn't use the new one. | |
55f335a8 LT |
613 | * |
614 | * NOTE! It is very important that the FASYNC flag always | |
615 | * match the state "is the filp on a fasync list". | |
f7347ce4 LT |
616 | */ |
617 | struct fasync_struct *fasync_insert_entry(int fd, struct file *filp, struct fasync_struct **fapp, struct fasync_struct *new) | |
618 | { | |
619 | struct fasync_struct *fa, **fp; | |
4a6a4499 | 620 | |
4a6a4499 | 621 | spin_lock(&filp->f_lock); |
989a2979 | 622 | spin_lock(&fasync_lock); |
1da177e4 | 623 | for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) { |
53281b6d LT |
624 | if (fa->fa_file != filp) |
625 | continue; | |
989a2979 ED |
626 | |
627 | spin_lock_irq(&fa->fa_lock); | |
53281b6d | 628 | fa->fa_fd = fd; |
989a2979 | 629 | spin_unlock_irq(&fa->fa_lock); |
53281b6d | 630 | goto out; |
1da177e4 LT |
631 | } |
632 | ||
989a2979 | 633 | spin_lock_init(&new->fa_lock); |
53281b6d LT |
634 | new->magic = FASYNC_MAGIC; |
635 | new->fa_file = filp; | |
636 | new->fa_fd = fd; | |
637 | new->fa_next = *fapp; | |
989a2979 | 638 | rcu_assign_pointer(*fapp, new); |
53281b6d LT |
639 | filp->f_flags |= FASYNC; |
640 | ||
1da177e4 | 641 | out: |
989a2979 | 642 | spin_unlock(&fasync_lock); |
4a6a4499 | 643 | spin_unlock(&filp->f_lock); |
f7347ce4 LT |
644 | return fa; |
645 | } | |
646 | ||
647 | /* | |
648 | * Add a fasync entry. Return negative on error, positive if | |
649 | * added, and zero if did nothing but change an existing one. | |
f7347ce4 LT |
650 | */ |
651 | static int fasync_add_entry(int fd, struct file *filp, struct fasync_struct **fapp) | |
652 | { | |
653 | struct fasync_struct *new; | |
654 | ||
655 | new = fasync_alloc(); | |
656 | if (!new) | |
657 | return -ENOMEM; | |
658 | ||
659 | /* | |
660 | * fasync_insert_entry() returns the old (update) entry if | |
661 | * it existed. | |
662 | * | |
663 | * So free the (unused) new entry and return 0 to let the | |
664 | * caller know that we didn't add any new fasync entries. | |
665 | */ | |
666 | if (fasync_insert_entry(fd, filp, fapp, new)) { | |
667 | fasync_free(new); | |
668 | return 0; | |
669 | } | |
670 | ||
671 | return 1; | |
1da177e4 LT |
672 | } |
673 | ||
53281b6d LT |
674 | /* |
675 | * fasync_helper() is used by almost all character device drivers | |
676 | * to set up the fasync queue, and for regular files by the file | |
677 | * lease code. It returns negative on error, 0 if it did no changes | |
678 | * and positive if it added/deleted the entry. | |
679 | */ | |
680 | int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp) | |
681 | { | |
682 | if (!on) | |
683 | return fasync_remove_entry(filp, fapp); | |
684 | return fasync_add_entry(fd, filp, fapp); | |
685 | } | |
686 | ||
1da177e4 LT |
687 | EXPORT_SYMBOL(fasync_helper); |
688 | ||
989a2979 ED |
689 | /* |
690 | * rcu_read_lock() is held | |
691 | */ | |
692 | static void kill_fasync_rcu(struct fasync_struct *fa, int sig, int band) | |
1da177e4 LT |
693 | { |
694 | while (fa) { | |
989a2979 | 695 | struct fown_struct *fown; |
f4985dc7 AM |
696 | unsigned long flags; |
697 | ||
1da177e4 LT |
698 | if (fa->magic != FASYNC_MAGIC) { |
699 | printk(KERN_ERR "kill_fasync: bad magic number in " | |
700 | "fasync_struct!\n"); | |
701 | return; | |
702 | } | |
f4985dc7 | 703 | spin_lock_irqsave(&fa->fa_lock, flags); |
989a2979 ED |
704 | if (fa->fa_file) { |
705 | fown = &fa->fa_file->f_owner; | |
706 | /* Don't send SIGURG to processes which have not set a | |
707 | queued signum: SIGURG has its own default signalling | |
708 | mechanism. */ | |
709 | if (!(sig == SIGURG && fown->signum == 0)) | |
710 | send_sigio(fown, fa->fa_fd, band); | |
711 | } | |
f4985dc7 | 712 | spin_unlock_irqrestore(&fa->fa_lock, flags); |
989a2979 | 713 | fa = rcu_dereference(fa->fa_next); |
1da177e4 LT |
714 | } |
715 | } | |
716 | ||
1da177e4 LT |
717 | void kill_fasync(struct fasync_struct **fp, int sig, int band) |
718 | { | |
719 | /* First a quick test without locking: usually | |
720 | * the list is empty. | |
721 | */ | |
722 | if (*fp) { | |
989a2979 ED |
723 | rcu_read_lock(); |
724 | kill_fasync_rcu(rcu_dereference(*fp), sig, band); | |
725 | rcu_read_unlock(); | |
1da177e4 LT |
726 | } |
727 | } | |
728 | EXPORT_SYMBOL(kill_fasync); | |
729 | ||
454eedb8 | 730 | static int __init fcntl_init(void) |
1da177e4 | 731 | { |
3ab04d5c JB |
732 | /* |
733 | * Please add new bits here to ensure allocation uniqueness. | |
734 | * Exceptions: O_NONBLOCK is a two bit define on parisc; O_NDELAY | |
735 | * is defined as O_NONBLOCK on some platforms and not on others. | |
736 | */ | |
1abf0c71 | 737 | BUILD_BUG_ON(19 - 1 /* for O_RDONLY being 0 */ != HWEIGHT32( |
454eedb8 WF |
738 | O_RDONLY | O_WRONLY | O_RDWR | |
739 | O_CREAT | O_EXCL | O_NOCTTY | | |
3ab04d5c | 740 | O_TRUNC | O_APPEND | /* O_NONBLOCK | */ |
454eedb8 WF |
741 | __O_SYNC | O_DSYNC | FASYNC | |
742 | O_DIRECT | O_LARGEFILE | O_DIRECTORY | | |
743 | O_NOFOLLOW | O_NOATIME | O_CLOEXEC | | |
1abf0c71 | 744 | __FMODE_EXEC | O_PATH |
454eedb8 WF |
745 | )); |
746 | ||
1da177e4 | 747 | fasync_cache = kmem_cache_create("fasync_cache", |
20c2df83 | 748 | sizeof(struct fasync_struct), 0, SLAB_PANIC, NULL); |
1da177e4 LT |
749 | return 0; |
750 | } | |
751 | ||
454eedb8 | 752 | module_init(fcntl_init) |