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> |
1da177e4 LT |
23 | |
24 | #include <asm/poll.h> | |
25 | #include <asm/siginfo.h> | |
26 | #include <asm/uaccess.h> | |
27 | ||
fc9b52cd | 28 | void set_close_on_exec(unsigned int fd, int flag) |
1da177e4 LT |
29 | { |
30 | struct files_struct *files = current->files; | |
badf1662 | 31 | struct fdtable *fdt; |
1da177e4 | 32 | spin_lock(&files->file_lock); |
badf1662 | 33 | fdt = files_fdtable(files); |
1da177e4 | 34 | if (flag) |
1dce27c5 | 35 | __set_close_on_exec(fd, fdt); |
1da177e4 | 36 | else |
1dce27c5 | 37 | __clear_close_on_exec(fd, fdt); |
1da177e4 LT |
38 | spin_unlock(&files->file_lock); |
39 | } | |
40 | ||
1dce27c5 | 41 | static bool get_close_on_exec(unsigned int fd) |
1da177e4 LT |
42 | { |
43 | struct files_struct *files = current->files; | |
badf1662 | 44 | struct fdtable *fdt; |
1dce27c5 | 45 | bool res; |
b835996f | 46 | rcu_read_lock(); |
badf1662 | 47 | fdt = files_fdtable(files); |
1dce27c5 | 48 | res = close_on_exec(fd, fdt); |
b835996f | 49 | rcu_read_unlock(); |
1da177e4 LT |
50 | return res; |
51 | } | |
52 | ||
a26eab24 | 53 | SYSCALL_DEFINE3(dup3, unsigned int, oldfd, unsigned int, newfd, int, flags) |
1da177e4 LT |
54 | { |
55 | int err = -EBADF; | |
56 | struct file * file, *tofree; | |
57 | struct files_struct * files = current->files; | |
badf1662 | 58 | struct fdtable *fdt; |
1da177e4 | 59 | |
336dd1f7 UD |
60 | if ((flags & ~O_CLOEXEC) != 0) |
61 | return -EINVAL; | |
62 | ||
6c5d0512 AV |
63 | if (unlikely(oldfd == newfd)) |
64 | return -EINVAL; | |
65 | ||
1da177e4 | 66 | spin_lock(&files->file_lock); |
1da177e4 | 67 | err = expand_files(files, newfd); |
1b7e190b AV |
68 | file = fcheck(oldfd); |
69 | if (unlikely(!file)) | |
70 | goto Ebadf; | |
4e1e018e AV |
71 | if (unlikely(err < 0)) { |
72 | if (err == -EMFILE) | |
1b7e190b AV |
73 | goto Ebadf; |
74 | goto out_unlock; | |
4e1e018e | 75 | } |
1b7e190b AV |
76 | /* |
77 | * We need to detect attempts to do dup2() over allocated but still | |
78 | * not finished descriptor. NB: OpenBSD avoids that at the price of | |
79 | * extra work in their equivalent of fget() - they insert struct | |
80 | * file immediately after grabbing descriptor, mark it larval if | |
81 | * more work (e.g. actual opening) is needed and make sure that | |
82 | * fget() treats larval files as absent. Potentially interesting, | |
83 | * but while extra work in fget() is trivial, locking implications | |
84 | * and amount of surgery on open()-related paths in VFS are not. | |
85 | * FreeBSD fails with -EBADF in the same situation, NetBSD "solution" | |
86 | * deadlocks in rather amusing ways, AFAICS. All of that is out of | |
87 | * scope of POSIX or SUS, since neither considers shared descriptor | |
88 | * tables and this condition does not arise without those. | |
89 | */ | |
1da177e4 | 90 | err = -EBUSY; |
badf1662 DS |
91 | fdt = files_fdtable(files); |
92 | tofree = fdt->fd[newfd]; | |
1dce27c5 | 93 | if (!tofree && fd_is_open(newfd, fdt)) |
1b7e190b AV |
94 | goto out_unlock; |
95 | get_file(file); | |
ab2af1f5 | 96 | rcu_assign_pointer(fdt->fd[newfd], file); |
1dce27c5 | 97 | __set_open_fd(newfd, fdt); |
336dd1f7 | 98 | if (flags & O_CLOEXEC) |
1dce27c5 | 99 | __set_close_on_exec(newfd, fdt); |
336dd1f7 | 100 | else |
1dce27c5 | 101 | __clear_close_on_exec(newfd, fdt); |
1da177e4 LT |
102 | spin_unlock(&files->file_lock); |
103 | ||
104 | if (tofree) | |
105 | filp_close(tofree, files); | |
1da177e4 | 106 | |
1b7e190b AV |
107 | return newfd; |
108 | ||
109 | Ebadf: | |
110 | err = -EBADF; | |
111 | out_unlock: | |
1da177e4 | 112 | spin_unlock(&files->file_lock); |
1b7e190b | 113 | return err; |
1da177e4 | 114 | } |
336dd1f7 | 115 | |
a26eab24 | 116 | SYSCALL_DEFINE2(dup2, unsigned int, oldfd, unsigned int, newfd) |
336dd1f7 | 117 | { |
6c5d0512 AV |
118 | if (unlikely(newfd == oldfd)) { /* corner case */ |
119 | struct files_struct *files = current->files; | |
2b79bc4f JM |
120 | int retval = oldfd; |
121 | ||
6c5d0512 AV |
122 | rcu_read_lock(); |
123 | if (!fcheck_files(files, oldfd)) | |
2b79bc4f | 124 | retval = -EBADF; |
6c5d0512 | 125 | rcu_read_unlock(); |
2b79bc4f | 126 | return retval; |
6c5d0512 | 127 | } |
336dd1f7 UD |
128 | return sys_dup3(oldfd, newfd, 0); |
129 | } | |
1da177e4 | 130 | |
a26eab24 | 131 | SYSCALL_DEFINE1(dup, unsigned int, fildes) |
1da177e4 LT |
132 | { |
133 | int ret = -EBADF; | |
1abf0c71 | 134 | struct file *file = fget_raw(fildes); |
1027abe8 AV |
135 | |
136 | if (file) { | |
137 | ret = get_unused_fd(); | |
138 | if (ret >= 0) | |
139 | fd_install(ret, file); | |
140 | else | |
141 | fput(file); | |
142 | } | |
1da177e4 LT |
143 | return ret; |
144 | } | |
145 | ||
76398425 | 146 | #define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | O_DIRECT | O_NOATIME) |
1da177e4 LT |
147 | |
148 | static int setfl(int fd, struct file * filp, unsigned long arg) | |
149 | { | |
0f7fc9e4 | 150 | struct inode * inode = filp->f_path.dentry->d_inode; |
1da177e4 LT |
151 | int error = 0; |
152 | ||
7d95c8f2 | 153 | /* |
154 | * O_APPEND cannot be cleared if the file is marked as append-only | |
155 | * and the file is open for write. | |
156 | */ | |
157 | if (((arg ^ filp->f_flags) & O_APPEND) && IS_APPEND(inode)) | |
1da177e4 LT |
158 | return -EPERM; |
159 | ||
160 | /* O_NOATIME can only be set by the owner or superuser */ | |
161 | if ((arg & O_NOATIME) && !(filp->f_flags & O_NOATIME)) | |
2e149670 | 162 | if (!inode_owner_or_capable(inode)) |
1da177e4 LT |
163 | return -EPERM; |
164 | ||
165 | /* required for strict SunOS emulation */ | |
166 | if (O_NONBLOCK != O_NDELAY) | |
167 | if (arg & O_NDELAY) | |
168 | arg |= O_NONBLOCK; | |
169 | ||
170 | if (arg & O_DIRECT) { | |
171 | if (!filp->f_mapping || !filp->f_mapping->a_ops || | |
172 | !filp->f_mapping->a_ops->direct_IO) | |
173 | return -EINVAL; | |
174 | } | |
175 | ||
176 | if (filp->f_op && filp->f_op->check_flags) | |
177 | error = filp->f_op->check_flags(arg); | |
178 | if (error) | |
179 | return error; | |
180 | ||
218d11a8 | 181 | /* |
76398425 | 182 | * ->fasync() is responsible for setting the FASYNC bit. |
218d11a8 | 183 | */ |
76398425 JC |
184 | if (((arg ^ filp->f_flags) & FASYNC) && filp->f_op && |
185 | filp->f_op->fasync) { | |
186 | error = filp->f_op->fasync(fd, filp, (arg & FASYNC) != 0); | |
187 | if (error < 0) | |
188 | goto out; | |
60aa4924 JC |
189 | if (error > 0) |
190 | error = 0; | |
1da177e4 | 191 | } |
db1dd4d3 | 192 | spin_lock(&filp->f_lock); |
1da177e4 | 193 | filp->f_flags = (arg & SETFL_MASK) | (filp->f_flags & ~SETFL_MASK); |
db1dd4d3 | 194 | spin_unlock(&filp->f_lock); |
76398425 | 195 | |
1da177e4 | 196 | out: |
1da177e4 LT |
197 | return error; |
198 | } | |
199 | ||
609d7fa9 | 200 | static void f_modown(struct file *filp, struct pid *pid, enum pid_type type, |
2f38d70f | 201 | int force) |
1da177e4 | 202 | { |
80e1e823 | 203 | write_lock_irq(&filp->f_owner.lock); |
1da177e4 | 204 | if (force || !filp->f_owner.pid) { |
609d7fa9 EB |
205 | put_pid(filp->f_owner.pid); |
206 | filp->f_owner.pid = get_pid(pid); | |
207 | filp->f_owner.pid_type = type; | |
2f38d70f ON |
208 | |
209 | if (pid) { | |
210 | const struct cred *cred = current_cred(); | |
211 | filp->f_owner.uid = cred->uid; | |
212 | filp->f_owner.euid = cred->euid; | |
213 | } | |
1da177e4 | 214 | } |
80e1e823 | 215 | write_unlock_irq(&filp->f_owner.lock); |
1da177e4 LT |
216 | } |
217 | ||
609d7fa9 EB |
218 | int __f_setown(struct file *filp, struct pid *pid, enum pid_type type, |
219 | int force) | |
1da177e4 LT |
220 | { |
221 | int err; | |
2f38d70f | 222 | |
1da177e4 LT |
223 | err = security_file_set_fowner(filp); |
224 | if (err) | |
225 | return err; | |
226 | ||
2f38d70f | 227 | f_modown(filp, pid, type, force); |
1da177e4 LT |
228 | return 0; |
229 | } | |
609d7fa9 | 230 | EXPORT_SYMBOL(__f_setown); |
1da177e4 | 231 | |
609d7fa9 EB |
232 | int f_setown(struct file *filp, unsigned long arg, int force) |
233 | { | |
234 | enum pid_type type; | |
235 | struct pid *pid; | |
236 | int who = arg; | |
237 | int result; | |
238 | type = PIDTYPE_PID; | |
239 | if (who < 0) { | |
240 | type = PIDTYPE_PGID; | |
241 | who = -who; | |
242 | } | |
243 | rcu_read_lock(); | |
b488893a | 244 | pid = find_vpid(who); |
609d7fa9 EB |
245 | result = __f_setown(filp, pid, type, force); |
246 | rcu_read_unlock(); | |
247 | return result; | |
248 | } | |
1da177e4 LT |
249 | EXPORT_SYMBOL(f_setown); |
250 | ||
251 | void f_delown(struct file *filp) | |
252 | { | |
2f38d70f | 253 | f_modown(filp, NULL, PIDTYPE_PID, 1); |
609d7fa9 EB |
254 | } |
255 | ||
256 | pid_t f_getown(struct file *filp) | |
257 | { | |
258 | pid_t pid; | |
43fa1adb | 259 | read_lock(&filp->f_owner.lock); |
6c5f3e7b | 260 | pid = pid_vnr(filp->f_owner.pid); |
609d7fa9 EB |
261 | if (filp->f_owner.pid_type == PIDTYPE_PGID) |
262 | pid = -pid; | |
43fa1adb | 263 | read_unlock(&filp->f_owner.lock); |
609d7fa9 | 264 | return pid; |
1da177e4 LT |
265 | } |
266 | ||
ba0a6c9f PZ |
267 | static int f_setown_ex(struct file *filp, unsigned long arg) |
268 | { | |
269 | struct f_owner_ex * __user owner_p = (void * __user)arg; | |
270 | struct f_owner_ex owner; | |
271 | struct pid *pid; | |
272 | int type; | |
273 | int ret; | |
274 | ||
275 | ret = copy_from_user(&owner, owner_p, sizeof(owner)); | |
276 | if (ret) | |
5b54470d | 277 | return -EFAULT; |
ba0a6c9f PZ |
278 | |
279 | switch (owner.type) { | |
280 | case F_OWNER_TID: | |
281 | type = PIDTYPE_MAX; | |
282 | break; | |
283 | ||
284 | case F_OWNER_PID: | |
285 | type = PIDTYPE_PID; | |
286 | break; | |
287 | ||
978b4053 | 288 | case F_OWNER_PGRP: |
ba0a6c9f PZ |
289 | type = PIDTYPE_PGID; |
290 | break; | |
291 | ||
292 | default: | |
293 | return -EINVAL; | |
294 | } | |
295 | ||
296 | rcu_read_lock(); | |
297 | pid = find_vpid(owner.pid); | |
298 | if (owner.pid && !pid) | |
299 | ret = -ESRCH; | |
300 | else | |
301 | ret = __f_setown(filp, pid, type, 1); | |
302 | rcu_read_unlock(); | |
303 | ||
304 | return ret; | |
305 | } | |
306 | ||
307 | static int f_getown_ex(struct file *filp, unsigned long arg) | |
308 | { | |
309 | struct f_owner_ex * __user owner_p = (void * __user)arg; | |
310 | struct f_owner_ex owner; | |
311 | int ret = 0; | |
312 | ||
313 | read_lock(&filp->f_owner.lock); | |
314 | owner.pid = pid_vnr(filp->f_owner.pid); | |
315 | switch (filp->f_owner.pid_type) { | |
316 | case PIDTYPE_MAX: | |
317 | owner.type = F_OWNER_TID; | |
318 | break; | |
319 | ||
320 | case PIDTYPE_PID: | |
321 | owner.type = F_OWNER_PID; | |
322 | break; | |
323 | ||
324 | case PIDTYPE_PGID: | |
978b4053 | 325 | owner.type = F_OWNER_PGRP; |
ba0a6c9f PZ |
326 | break; |
327 | ||
328 | default: | |
329 | WARN_ON(1); | |
330 | ret = -EINVAL; | |
331 | break; | |
332 | } | |
333 | read_unlock(&filp->f_owner.lock); | |
334 | ||
5b54470d | 335 | if (!ret) { |
ba0a6c9f | 336 | ret = copy_to_user(owner_p, &owner, sizeof(owner)); |
5b54470d DC |
337 | if (ret) |
338 | ret = -EFAULT; | |
339 | } | |
ba0a6c9f PZ |
340 | return ret; |
341 | } | |
342 | ||
1da177e4 LT |
343 | static long do_fcntl(int fd, unsigned int cmd, unsigned long arg, |
344 | struct file *filp) | |
345 | { | |
346 | long err = -EINVAL; | |
347 | ||
348 | switch (cmd) { | |
349 | case F_DUPFD: | |
22d2b35b | 350 | case F_DUPFD_CLOEXEC: |
d554ed89 | 351 | if (arg >= rlimit(RLIMIT_NOFILE)) |
4e1e018e | 352 | break; |
1027abe8 AV |
353 | err = alloc_fd(arg, cmd == F_DUPFD_CLOEXEC ? O_CLOEXEC : 0); |
354 | if (err >= 0) { | |
355 | get_file(filp); | |
356 | fd_install(err, filp); | |
357 | } | |
1da177e4 LT |
358 | break; |
359 | case F_GETFD: | |
360 | err = get_close_on_exec(fd) ? FD_CLOEXEC : 0; | |
361 | break; | |
362 | case F_SETFD: | |
363 | err = 0; | |
364 | set_close_on_exec(fd, arg & FD_CLOEXEC); | |
365 | break; | |
366 | case F_GETFL: | |
367 | err = filp->f_flags; | |
368 | break; | |
369 | case F_SETFL: | |
370 | err = setfl(fd, filp, arg); | |
371 | break; | |
372 | case F_GETLK: | |
373 | err = fcntl_getlk(filp, (struct flock __user *) arg); | |
374 | break; | |
375 | case F_SETLK: | |
376 | case F_SETLKW: | |
c293621b | 377 | err = fcntl_setlk(fd, filp, cmd, (struct flock __user *) arg); |
1da177e4 LT |
378 | break; |
379 | case F_GETOWN: | |
380 | /* | |
381 | * XXX If f_owner is a process group, the | |
382 | * negative return value will get converted | |
383 | * into an error. Oops. If we keep the | |
384 | * current syscall conventions, the only way | |
385 | * to fix this will be in libc. | |
386 | */ | |
609d7fa9 | 387 | err = f_getown(filp); |
1da177e4 LT |
388 | force_successful_syscall_return(); |
389 | break; | |
390 | case F_SETOWN: | |
391 | err = f_setown(filp, arg, 1); | |
392 | break; | |
ba0a6c9f PZ |
393 | case F_GETOWN_EX: |
394 | err = f_getown_ex(filp, arg); | |
395 | break; | |
396 | case F_SETOWN_EX: | |
397 | err = f_setown_ex(filp, arg); | |
398 | break; | |
1da177e4 LT |
399 | case F_GETSIG: |
400 | err = filp->f_owner.signum; | |
401 | break; | |
402 | case F_SETSIG: | |
403 | /* arg == 0 restores default behaviour. */ | |
7ed20e1a | 404 | if (!valid_signal(arg)) { |
1da177e4 LT |
405 | break; |
406 | } | |
407 | err = 0; | |
408 | filp->f_owner.signum = arg; | |
409 | break; | |
410 | case F_GETLEASE: | |
411 | err = fcntl_getlease(filp); | |
412 | break; | |
413 | case F_SETLEASE: | |
414 | err = fcntl_setlease(fd, filp, arg); | |
415 | break; | |
416 | case F_NOTIFY: | |
417 | err = fcntl_dirnotify(fd, filp, arg); | |
418 | break; | |
35f3d14d JA |
419 | case F_SETPIPE_SZ: |
420 | case F_GETPIPE_SZ: | |
421 | err = pipe_fcntl(filp, cmd, arg); | |
422 | break; | |
1da177e4 LT |
423 | default: |
424 | break; | |
425 | } | |
426 | return err; | |
427 | } | |
428 | ||
1abf0c71 AV |
429 | static int check_fcntl_cmd(unsigned cmd) |
430 | { | |
431 | switch (cmd) { | |
432 | case F_DUPFD: | |
433 | case F_DUPFD_CLOEXEC: | |
434 | case F_GETFD: | |
435 | case F_SETFD: | |
436 | case F_GETFL: | |
437 | return 1; | |
438 | } | |
439 | return 0; | |
440 | } | |
441 | ||
a26eab24 | 442 | SYSCALL_DEFINE3(fcntl, unsigned int, fd, unsigned int, cmd, unsigned long, arg) |
1da177e4 LT |
443 | { |
444 | struct file *filp; | |
545ec2c7 | 445 | int fput_needed; |
1da177e4 LT |
446 | long err = -EBADF; |
447 | ||
545ec2c7 | 448 | filp = fget_raw_light(fd, &fput_needed); |
1da177e4 LT |
449 | if (!filp) |
450 | goto out; | |
451 | ||
1abf0c71 | 452 | if (unlikely(filp->f_mode & FMODE_PATH)) { |
545ec2c7 AV |
453 | if (!check_fcntl_cmd(cmd)) |
454 | goto out1; | |
1abf0c71 AV |
455 | } |
456 | ||
1da177e4 | 457 | err = security_file_fcntl(filp, cmd, arg); |
545ec2c7 AV |
458 | if (!err) |
459 | err = do_fcntl(fd, cmd, arg, filp); | |
1da177e4 | 460 | |
545ec2c7 AV |
461 | out1: |
462 | fput_light(filp, fput_needed); | |
1da177e4 LT |
463 | out: |
464 | return err; | |
465 | } | |
466 | ||
467 | #if BITS_PER_LONG == 32 | |
a26eab24 HC |
468 | SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd, |
469 | unsigned long, arg) | |
1da177e4 LT |
470 | { |
471 | struct file * filp; | |
545ec2c7 AV |
472 | long err = -EBADF; |
473 | int fput_needed; | |
1da177e4 | 474 | |
545ec2c7 | 475 | filp = fget_raw_light(fd, &fput_needed); |
1da177e4 LT |
476 | if (!filp) |
477 | goto out; | |
478 | ||
1abf0c71 | 479 | if (unlikely(filp->f_mode & FMODE_PATH)) { |
545ec2c7 AV |
480 | if (!check_fcntl_cmd(cmd)) |
481 | goto out1; | |
1abf0c71 AV |
482 | } |
483 | ||
1da177e4 | 484 | err = security_file_fcntl(filp, cmd, arg); |
545ec2c7 AV |
485 | if (err) |
486 | goto out1; | |
1da177e4 LT |
487 | |
488 | switch (cmd) { | |
489 | case F_GETLK64: | |
490 | err = fcntl_getlk64(filp, (struct flock64 __user *) arg); | |
491 | break; | |
492 | case F_SETLK64: | |
493 | case F_SETLKW64: | |
c293621b PS |
494 | err = fcntl_setlk64(fd, filp, cmd, |
495 | (struct flock64 __user *) arg); | |
1da177e4 LT |
496 | break; |
497 | default: | |
498 | err = do_fcntl(fd, cmd, arg, filp); | |
499 | break; | |
500 | } | |
545ec2c7 AV |
501 | out1: |
502 | fput_light(filp, fput_needed); | |
1da177e4 LT |
503 | out: |
504 | return err; | |
505 | } | |
506 | #endif | |
507 | ||
508 | /* Table to convert sigio signal codes into poll band bitmaps */ | |
509 | ||
fa3536cc | 510 | static const long band_table[NSIGPOLL] = { |
1da177e4 LT |
511 | POLLIN | POLLRDNORM, /* POLL_IN */ |
512 | POLLOUT | POLLWRNORM | POLLWRBAND, /* POLL_OUT */ | |
513 | POLLIN | POLLRDNORM | POLLMSG, /* POLL_MSG */ | |
514 | POLLERR, /* POLL_ERR */ | |
515 | POLLPRI | POLLRDBAND, /* POLL_PRI */ | |
516 | POLLHUP | POLLERR /* POLL_HUP */ | |
517 | }; | |
518 | ||
519 | static inline int sigio_perm(struct task_struct *p, | |
520 | struct fown_struct *fown, int sig) | |
521 | { | |
c69e8d9c DH |
522 | const struct cred *cred; |
523 | int ret; | |
524 | ||
525 | rcu_read_lock(); | |
526 | cred = __task_cred(p); | |
8e96e3b7 EB |
527 | ret = ((uid_eq(fown->euid, GLOBAL_ROOT_UID) || |
528 | uid_eq(fown->euid, cred->suid) || uid_eq(fown->euid, cred->uid) || | |
529 | uid_eq(fown->uid, cred->suid) || uid_eq(fown->uid, cred->uid)) && | |
c69e8d9c DH |
530 | !security_file_send_sigiotask(p, fown, sig)); |
531 | rcu_read_unlock(); | |
532 | return ret; | |
1da177e4 LT |
533 | } |
534 | ||
535 | static void send_sigio_to_task(struct task_struct *p, | |
8eeee4e2 | 536 | struct fown_struct *fown, |
ba0a6c9f | 537 | int fd, int reason, int group) |
1da177e4 | 538 | { |
8eeee4e2 ON |
539 | /* |
540 | * F_SETSIG can change ->signum lockless in parallel, make | |
541 | * sure we read it once and use the same value throughout. | |
542 | */ | |
543 | int signum = ACCESS_ONCE(fown->signum); | |
544 | ||
545 | if (!sigio_perm(p, fown, signum)) | |
1da177e4 LT |
546 | return; |
547 | ||
8eeee4e2 | 548 | switch (signum) { |
1da177e4 LT |
549 | siginfo_t si; |
550 | default: | |
551 | /* Queue a rt signal with the appropriate fd as its | |
552 | value. We use SI_SIGIO as the source, not | |
553 | SI_KERNEL, since kernel signals always get | |
554 | delivered even if we can't queue. Failure to | |
555 | queue in this case _should_ be reported; we fall | |
556 | back to SIGIO in that case. --sct */ | |
8eeee4e2 | 557 | si.si_signo = signum; |
1da177e4 LT |
558 | si.si_errno = 0; |
559 | si.si_code = reason; | |
560 | /* Make sure we are called with one of the POLL_* | |
561 | reasons, otherwise we could leak kernel stack into | |
562 | userspace. */ | |
f6298aab | 563 | BUG_ON((reason & __SI_MASK) != __SI_POLL); |
1da177e4 LT |
564 | if (reason - POLL_IN >= NSIGPOLL) |
565 | si.si_band = ~0L; | |
566 | else | |
567 | si.si_band = band_table[reason - POLL_IN]; | |
568 | si.si_fd = fd; | |
ba0a6c9f | 569 | if (!do_send_sig_info(signum, &si, p, group)) |
1da177e4 LT |
570 | break; |
571 | /* fall-through: fall back on the old plain SIGIO signal */ | |
572 | case 0: | |
ba0a6c9f | 573 | do_send_sig_info(SIGIO, SEND_SIG_PRIV, p, group); |
1da177e4 LT |
574 | } |
575 | } | |
576 | ||
577 | void send_sigio(struct fown_struct *fown, int fd, int band) | |
578 | { | |
579 | struct task_struct *p; | |
609d7fa9 EB |
580 | enum pid_type type; |
581 | struct pid *pid; | |
ba0a6c9f | 582 | int group = 1; |
1da177e4 LT |
583 | |
584 | read_lock(&fown->lock); | |
ba0a6c9f | 585 | |
609d7fa9 | 586 | type = fown->pid_type; |
ba0a6c9f PZ |
587 | if (type == PIDTYPE_MAX) { |
588 | group = 0; | |
589 | type = PIDTYPE_PID; | |
590 | } | |
591 | ||
1da177e4 LT |
592 | pid = fown->pid; |
593 | if (!pid) | |
594 | goto out_unlock_fown; | |
595 | ||
596 | read_lock(&tasklist_lock); | |
609d7fa9 | 597 | do_each_pid_task(pid, type, p) { |
ba0a6c9f | 598 | send_sigio_to_task(p, fown, fd, band, group); |
609d7fa9 | 599 | } while_each_pid_task(pid, type, p); |
1da177e4 LT |
600 | read_unlock(&tasklist_lock); |
601 | out_unlock_fown: | |
602 | read_unlock(&fown->lock); | |
603 | } | |
604 | ||
605 | static void send_sigurg_to_task(struct task_struct *p, | |
ba0a6c9f | 606 | struct fown_struct *fown, int group) |
1da177e4 LT |
607 | { |
608 | if (sigio_perm(p, fown, SIGURG)) | |
ba0a6c9f | 609 | do_send_sig_info(SIGURG, SEND_SIG_PRIV, p, group); |
1da177e4 LT |
610 | } |
611 | ||
612 | int send_sigurg(struct fown_struct *fown) | |
613 | { | |
614 | struct task_struct *p; | |
609d7fa9 EB |
615 | enum pid_type type; |
616 | struct pid *pid; | |
ba0a6c9f | 617 | int group = 1; |
609d7fa9 | 618 | int ret = 0; |
1da177e4 LT |
619 | |
620 | read_lock(&fown->lock); | |
ba0a6c9f | 621 | |
609d7fa9 | 622 | type = fown->pid_type; |
ba0a6c9f PZ |
623 | if (type == PIDTYPE_MAX) { |
624 | group = 0; | |
625 | type = PIDTYPE_PID; | |
626 | } | |
627 | ||
1da177e4 LT |
628 | pid = fown->pid; |
629 | if (!pid) | |
630 | goto out_unlock_fown; | |
631 | ||
632 | ret = 1; | |
633 | ||
634 | read_lock(&tasklist_lock); | |
609d7fa9 | 635 | do_each_pid_task(pid, type, p) { |
ba0a6c9f | 636 | send_sigurg_to_task(p, fown, group); |
609d7fa9 | 637 | } while_each_pid_task(pid, type, p); |
1da177e4 LT |
638 | read_unlock(&tasklist_lock); |
639 | out_unlock_fown: | |
640 | read_unlock(&fown->lock); | |
641 | return ret; | |
642 | } | |
643 | ||
989a2979 | 644 | static DEFINE_SPINLOCK(fasync_lock); |
e18b890b | 645 | static struct kmem_cache *fasync_cache __read_mostly; |
1da177e4 | 646 | |
989a2979 ED |
647 | static void fasync_free_rcu(struct rcu_head *head) |
648 | { | |
649 | kmem_cache_free(fasync_cache, | |
650 | container_of(head, struct fasync_struct, fa_rcu)); | |
651 | } | |
652 | ||
1da177e4 | 653 | /* |
53281b6d LT |
654 | * Remove a fasync entry. If successfully removed, return |
655 | * positive and clear the FASYNC flag. If no entry exists, | |
656 | * do nothing and return 0. | |
657 | * | |
658 | * NOTE! It is very important that the FASYNC flag always | |
659 | * match the state "is the filp on a fasync list". | |
660 | * | |
1da177e4 | 661 | */ |
f7347ce4 | 662 | int fasync_remove_entry(struct file *filp, struct fasync_struct **fapp) |
1da177e4 LT |
663 | { |
664 | struct fasync_struct *fa, **fp; | |
1da177e4 LT |
665 | int result = 0; |
666 | ||
53281b6d | 667 | spin_lock(&filp->f_lock); |
989a2979 | 668 | spin_lock(&fasync_lock); |
53281b6d LT |
669 | for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) { |
670 | if (fa->fa_file != filp) | |
671 | continue; | |
989a2979 ED |
672 | |
673 | spin_lock_irq(&fa->fa_lock); | |
674 | fa->fa_file = NULL; | |
675 | spin_unlock_irq(&fa->fa_lock); | |
676 | ||
53281b6d | 677 | *fp = fa->fa_next; |
989a2979 | 678 | call_rcu(&fa->fa_rcu, fasync_free_rcu); |
53281b6d LT |
679 | filp->f_flags &= ~FASYNC; |
680 | result = 1; | |
681 | break; | |
1da177e4 | 682 | } |
989a2979 | 683 | spin_unlock(&fasync_lock); |
53281b6d LT |
684 | spin_unlock(&filp->f_lock); |
685 | return result; | |
686 | } | |
687 | ||
f7347ce4 LT |
688 | struct fasync_struct *fasync_alloc(void) |
689 | { | |
690 | return kmem_cache_alloc(fasync_cache, GFP_KERNEL); | |
691 | } | |
692 | ||
53281b6d | 693 | /* |
f7347ce4 LT |
694 | * NOTE! This can be used only for unused fasync entries: |
695 | * entries that actually got inserted on the fasync list | |
696 | * need to be released by rcu - see fasync_remove_entry. | |
53281b6d | 697 | */ |
f7347ce4 | 698 | void fasync_free(struct fasync_struct *new) |
53281b6d | 699 | { |
f7347ce4 LT |
700 | kmem_cache_free(fasync_cache, new); |
701 | } | |
53281b6d | 702 | |
f7347ce4 LT |
703 | /* |
704 | * Insert a new entry into the fasync list. Return the pointer to the | |
705 | * old one if we didn't use the new one. | |
55f335a8 LT |
706 | * |
707 | * NOTE! It is very important that the FASYNC flag always | |
708 | * match the state "is the filp on a fasync list". | |
f7347ce4 LT |
709 | */ |
710 | struct fasync_struct *fasync_insert_entry(int fd, struct file *filp, struct fasync_struct **fapp, struct fasync_struct *new) | |
711 | { | |
712 | struct fasync_struct *fa, **fp; | |
4a6a4499 | 713 | |
4a6a4499 | 714 | spin_lock(&filp->f_lock); |
989a2979 | 715 | spin_lock(&fasync_lock); |
1da177e4 | 716 | for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) { |
53281b6d LT |
717 | if (fa->fa_file != filp) |
718 | continue; | |
989a2979 ED |
719 | |
720 | spin_lock_irq(&fa->fa_lock); | |
53281b6d | 721 | fa->fa_fd = fd; |
989a2979 | 722 | spin_unlock_irq(&fa->fa_lock); |
53281b6d | 723 | goto out; |
1da177e4 LT |
724 | } |
725 | ||
989a2979 | 726 | spin_lock_init(&new->fa_lock); |
53281b6d LT |
727 | new->magic = FASYNC_MAGIC; |
728 | new->fa_file = filp; | |
729 | new->fa_fd = fd; | |
730 | new->fa_next = *fapp; | |
989a2979 | 731 | rcu_assign_pointer(*fapp, new); |
53281b6d LT |
732 | filp->f_flags |= FASYNC; |
733 | ||
1da177e4 | 734 | out: |
989a2979 | 735 | spin_unlock(&fasync_lock); |
4a6a4499 | 736 | spin_unlock(&filp->f_lock); |
f7347ce4 LT |
737 | return fa; |
738 | } | |
739 | ||
740 | /* | |
741 | * Add a fasync entry. Return negative on error, positive if | |
742 | * added, and zero if did nothing but change an existing one. | |
f7347ce4 LT |
743 | */ |
744 | static int fasync_add_entry(int fd, struct file *filp, struct fasync_struct **fapp) | |
745 | { | |
746 | struct fasync_struct *new; | |
747 | ||
748 | new = fasync_alloc(); | |
749 | if (!new) | |
750 | return -ENOMEM; | |
751 | ||
752 | /* | |
753 | * fasync_insert_entry() returns the old (update) entry if | |
754 | * it existed. | |
755 | * | |
756 | * So free the (unused) new entry and return 0 to let the | |
757 | * caller know that we didn't add any new fasync entries. | |
758 | */ | |
759 | if (fasync_insert_entry(fd, filp, fapp, new)) { | |
760 | fasync_free(new); | |
761 | return 0; | |
762 | } | |
763 | ||
764 | return 1; | |
1da177e4 LT |
765 | } |
766 | ||
53281b6d LT |
767 | /* |
768 | * fasync_helper() is used by almost all character device drivers | |
769 | * to set up the fasync queue, and for regular files by the file | |
770 | * lease code. It returns negative on error, 0 if it did no changes | |
771 | * and positive if it added/deleted the entry. | |
772 | */ | |
773 | int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp) | |
774 | { | |
775 | if (!on) | |
776 | return fasync_remove_entry(filp, fapp); | |
777 | return fasync_add_entry(fd, filp, fapp); | |
778 | } | |
779 | ||
1da177e4 LT |
780 | EXPORT_SYMBOL(fasync_helper); |
781 | ||
989a2979 ED |
782 | /* |
783 | * rcu_read_lock() is held | |
784 | */ | |
785 | static void kill_fasync_rcu(struct fasync_struct *fa, int sig, int band) | |
1da177e4 LT |
786 | { |
787 | while (fa) { | |
989a2979 | 788 | struct fown_struct *fown; |
f4985dc7 AM |
789 | unsigned long flags; |
790 | ||
1da177e4 LT |
791 | if (fa->magic != FASYNC_MAGIC) { |
792 | printk(KERN_ERR "kill_fasync: bad magic number in " | |
793 | "fasync_struct!\n"); | |
794 | return; | |
795 | } | |
f4985dc7 | 796 | spin_lock_irqsave(&fa->fa_lock, flags); |
989a2979 ED |
797 | if (fa->fa_file) { |
798 | fown = &fa->fa_file->f_owner; | |
799 | /* Don't send SIGURG to processes which have not set a | |
800 | queued signum: SIGURG has its own default signalling | |
801 | mechanism. */ | |
802 | if (!(sig == SIGURG && fown->signum == 0)) | |
803 | send_sigio(fown, fa->fa_fd, band); | |
804 | } | |
f4985dc7 | 805 | spin_unlock_irqrestore(&fa->fa_lock, flags); |
989a2979 | 806 | fa = rcu_dereference(fa->fa_next); |
1da177e4 LT |
807 | } |
808 | } | |
809 | ||
1da177e4 LT |
810 | void kill_fasync(struct fasync_struct **fp, int sig, int band) |
811 | { | |
812 | /* First a quick test without locking: usually | |
813 | * the list is empty. | |
814 | */ | |
815 | if (*fp) { | |
989a2979 ED |
816 | rcu_read_lock(); |
817 | kill_fasync_rcu(rcu_dereference(*fp), sig, band); | |
818 | rcu_read_unlock(); | |
1da177e4 LT |
819 | } |
820 | } | |
821 | EXPORT_SYMBOL(kill_fasync); | |
822 | ||
454eedb8 | 823 | static int __init fcntl_init(void) |
1da177e4 | 824 | { |
3ab04d5c JB |
825 | /* |
826 | * Please add new bits here to ensure allocation uniqueness. | |
827 | * Exceptions: O_NONBLOCK is a two bit define on parisc; O_NDELAY | |
828 | * is defined as O_NONBLOCK on some platforms and not on others. | |
829 | */ | |
1abf0c71 | 830 | BUILD_BUG_ON(19 - 1 /* for O_RDONLY being 0 */ != HWEIGHT32( |
454eedb8 WF |
831 | O_RDONLY | O_WRONLY | O_RDWR | |
832 | O_CREAT | O_EXCL | O_NOCTTY | | |
3ab04d5c | 833 | O_TRUNC | O_APPEND | /* O_NONBLOCK | */ |
454eedb8 WF |
834 | __O_SYNC | O_DSYNC | FASYNC | |
835 | O_DIRECT | O_LARGEFILE | O_DIRECTORY | | |
836 | O_NOFOLLOW | O_NOATIME | O_CLOEXEC | | |
1abf0c71 | 837 | __FMODE_EXEC | O_PATH |
454eedb8 WF |
838 | )); |
839 | ||
1da177e4 | 840 | fasync_cache = kmem_cache_create("fasync_cache", |
20c2df83 | 841 | sizeof(struct fasync_struct), 0, SLAB_PANIC, NULL); |
1da177e4 LT |
842 | return 0; |
843 | } | |
844 | ||
454eedb8 | 845 | module_init(fcntl_init) |