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 | ||
fc9b52cd | 29 | void set_close_on_exec(unsigned int fd, int flag) |
1da177e4 LT |
30 | { |
31 | struct files_struct *files = current->files; | |
badf1662 | 32 | struct fdtable *fdt; |
1da177e4 | 33 | spin_lock(&files->file_lock); |
badf1662 | 34 | fdt = files_fdtable(files); |
1da177e4 | 35 | if (flag) |
1dce27c5 | 36 | __set_close_on_exec(fd, fdt); |
1da177e4 | 37 | else |
1dce27c5 | 38 | __clear_close_on_exec(fd, fdt); |
1da177e4 LT |
39 | spin_unlock(&files->file_lock); |
40 | } | |
41 | ||
1dce27c5 | 42 | static bool get_close_on_exec(unsigned int fd) |
1da177e4 LT |
43 | { |
44 | struct files_struct *files = current->files; | |
badf1662 | 45 | struct fdtable *fdt; |
1dce27c5 | 46 | bool res; |
b835996f | 47 | rcu_read_lock(); |
badf1662 | 48 | fdt = files_fdtable(files); |
1dce27c5 | 49 | res = close_on_exec(fd, fdt); |
b835996f | 50 | rcu_read_unlock(); |
1da177e4 LT |
51 | return res; |
52 | } | |
53 | ||
a26eab24 | 54 | SYSCALL_DEFINE3(dup3, unsigned int, oldfd, unsigned int, newfd, int, flags) |
1da177e4 LT |
55 | { |
56 | int err = -EBADF; | |
57 | struct file * file, *tofree; | |
58 | struct files_struct * files = current->files; | |
badf1662 | 59 | struct fdtable *fdt; |
1da177e4 | 60 | |
336dd1f7 UD |
61 | if ((flags & ~O_CLOEXEC) != 0) |
62 | return -EINVAL; | |
63 | ||
6c5d0512 AV |
64 | if (unlikely(oldfd == newfd)) |
65 | return -EINVAL; | |
f33ff992 AV |
66 | |
67 | if (newfd >= rlimit(RLIMIT_NOFILE)) | |
68 | return -EMFILE; | |
6c5d0512 | 69 | |
1da177e4 | 70 | spin_lock(&files->file_lock); |
1da177e4 | 71 | err = expand_files(files, newfd); |
1b7e190b AV |
72 | file = fcheck(oldfd); |
73 | if (unlikely(!file)) | |
74 | goto Ebadf; | |
4e1e018e AV |
75 | if (unlikely(err < 0)) { |
76 | if (err == -EMFILE) | |
1b7e190b AV |
77 | goto Ebadf; |
78 | goto out_unlock; | |
4e1e018e | 79 | } |
1b7e190b AV |
80 | /* |
81 | * We need to detect attempts to do dup2() over allocated but still | |
82 | * not finished descriptor. NB: OpenBSD avoids that at the price of | |
83 | * extra work in their equivalent of fget() - they insert struct | |
84 | * file immediately after grabbing descriptor, mark it larval if | |
85 | * more work (e.g. actual opening) is needed and make sure that | |
86 | * fget() treats larval files as absent. Potentially interesting, | |
87 | * but while extra work in fget() is trivial, locking implications | |
88 | * and amount of surgery on open()-related paths in VFS are not. | |
89 | * FreeBSD fails with -EBADF in the same situation, NetBSD "solution" | |
90 | * deadlocks in rather amusing ways, AFAICS. All of that is out of | |
91 | * scope of POSIX or SUS, since neither considers shared descriptor | |
92 | * tables and this condition does not arise without those. | |
93 | */ | |
1da177e4 | 94 | err = -EBUSY; |
badf1662 DS |
95 | fdt = files_fdtable(files); |
96 | tofree = fdt->fd[newfd]; | |
1dce27c5 | 97 | if (!tofree && fd_is_open(newfd, fdt)) |
1b7e190b AV |
98 | goto out_unlock; |
99 | get_file(file); | |
ab2af1f5 | 100 | rcu_assign_pointer(fdt->fd[newfd], file); |
1dce27c5 | 101 | __set_open_fd(newfd, fdt); |
336dd1f7 | 102 | if (flags & O_CLOEXEC) |
1dce27c5 | 103 | __set_close_on_exec(newfd, fdt); |
336dd1f7 | 104 | else |
1dce27c5 | 105 | __clear_close_on_exec(newfd, fdt); |
1da177e4 LT |
106 | spin_unlock(&files->file_lock); |
107 | ||
108 | if (tofree) | |
109 | filp_close(tofree, files); | |
1da177e4 | 110 | |
1b7e190b AV |
111 | return newfd; |
112 | ||
113 | Ebadf: | |
114 | err = -EBADF; | |
115 | out_unlock: | |
1da177e4 | 116 | spin_unlock(&files->file_lock); |
1b7e190b | 117 | return err; |
1da177e4 | 118 | } |
336dd1f7 | 119 | |
a26eab24 | 120 | SYSCALL_DEFINE2(dup2, unsigned int, oldfd, unsigned int, newfd) |
336dd1f7 | 121 | { |
6c5d0512 AV |
122 | if (unlikely(newfd == oldfd)) { /* corner case */ |
123 | struct files_struct *files = current->files; | |
2b79bc4f JM |
124 | int retval = oldfd; |
125 | ||
6c5d0512 AV |
126 | rcu_read_lock(); |
127 | if (!fcheck_files(files, oldfd)) | |
2b79bc4f | 128 | retval = -EBADF; |
6c5d0512 | 129 | rcu_read_unlock(); |
2b79bc4f | 130 | return retval; |
6c5d0512 | 131 | } |
336dd1f7 UD |
132 | return sys_dup3(oldfd, newfd, 0); |
133 | } | |
1da177e4 | 134 | |
a26eab24 | 135 | SYSCALL_DEFINE1(dup, unsigned int, fildes) |
1da177e4 LT |
136 | { |
137 | int ret = -EBADF; | |
1abf0c71 | 138 | struct file *file = fget_raw(fildes); |
1027abe8 AV |
139 | |
140 | if (file) { | |
141 | ret = get_unused_fd(); | |
142 | if (ret >= 0) | |
143 | fd_install(ret, file); | |
144 | else | |
145 | fput(file); | |
146 | } | |
1da177e4 LT |
147 | return ret; |
148 | } | |
149 | ||
76398425 | 150 | #define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | O_DIRECT | O_NOATIME) |
1da177e4 LT |
151 | |
152 | static int setfl(int fd, struct file * filp, unsigned long arg) | |
153 | { | |
0f7fc9e4 | 154 | struct inode * inode = filp->f_path.dentry->d_inode; |
1da177e4 LT |
155 | int error = 0; |
156 | ||
7d95c8f2 | 157 | /* |
158 | * O_APPEND cannot be cleared if the file is marked as append-only | |
159 | * and the file is open for write. | |
160 | */ | |
161 | if (((arg ^ filp->f_flags) & O_APPEND) && IS_APPEND(inode)) | |
1da177e4 LT |
162 | return -EPERM; |
163 | ||
164 | /* O_NOATIME can only be set by the owner or superuser */ | |
165 | if ((arg & O_NOATIME) && !(filp->f_flags & O_NOATIME)) | |
2e149670 | 166 | if (!inode_owner_or_capable(inode)) |
1da177e4 LT |
167 | return -EPERM; |
168 | ||
169 | /* required for strict SunOS emulation */ | |
170 | if (O_NONBLOCK != O_NDELAY) | |
171 | if (arg & O_NDELAY) | |
172 | arg |= O_NONBLOCK; | |
173 | ||
174 | if (arg & O_DIRECT) { | |
175 | if (!filp->f_mapping || !filp->f_mapping->a_ops || | |
176 | !filp->f_mapping->a_ops->direct_IO) | |
177 | return -EINVAL; | |
178 | } | |
179 | ||
180 | if (filp->f_op && filp->f_op->check_flags) | |
181 | error = filp->f_op->check_flags(arg); | |
182 | if (error) | |
183 | return error; | |
184 | ||
218d11a8 | 185 | /* |
76398425 | 186 | * ->fasync() is responsible for setting the FASYNC bit. |
218d11a8 | 187 | */ |
76398425 JC |
188 | if (((arg ^ filp->f_flags) & FASYNC) && filp->f_op && |
189 | filp->f_op->fasync) { | |
190 | error = filp->f_op->fasync(fd, filp, (arg & FASYNC) != 0); | |
191 | if (error < 0) | |
192 | goto out; | |
60aa4924 JC |
193 | if (error > 0) |
194 | error = 0; | |
1da177e4 | 195 | } |
db1dd4d3 | 196 | spin_lock(&filp->f_lock); |
1da177e4 | 197 | filp->f_flags = (arg & SETFL_MASK) | (filp->f_flags & ~SETFL_MASK); |
db1dd4d3 | 198 | spin_unlock(&filp->f_lock); |
76398425 | 199 | |
1da177e4 | 200 | out: |
1da177e4 LT |
201 | return error; |
202 | } | |
203 | ||
609d7fa9 | 204 | static void f_modown(struct file *filp, struct pid *pid, enum pid_type type, |
2f38d70f | 205 | int force) |
1da177e4 | 206 | { |
80e1e823 | 207 | write_lock_irq(&filp->f_owner.lock); |
1da177e4 | 208 | if (force || !filp->f_owner.pid) { |
609d7fa9 EB |
209 | put_pid(filp->f_owner.pid); |
210 | filp->f_owner.pid = get_pid(pid); | |
211 | filp->f_owner.pid_type = type; | |
2f38d70f ON |
212 | |
213 | if (pid) { | |
214 | const struct cred *cred = current_cred(); | |
215 | filp->f_owner.uid = cred->uid; | |
216 | filp->f_owner.euid = cred->euid; | |
217 | } | |
1da177e4 | 218 | } |
80e1e823 | 219 | write_unlock_irq(&filp->f_owner.lock); |
1da177e4 LT |
220 | } |
221 | ||
609d7fa9 EB |
222 | int __f_setown(struct file *filp, struct pid *pid, enum pid_type type, |
223 | int force) | |
1da177e4 LT |
224 | { |
225 | int err; | |
2f38d70f | 226 | |
1da177e4 LT |
227 | err = security_file_set_fowner(filp); |
228 | if (err) | |
229 | return err; | |
230 | ||
2f38d70f | 231 | f_modown(filp, pid, type, force); |
1da177e4 LT |
232 | return 0; |
233 | } | |
609d7fa9 | 234 | EXPORT_SYMBOL(__f_setown); |
1da177e4 | 235 | |
609d7fa9 EB |
236 | int f_setown(struct file *filp, unsigned long arg, int force) |
237 | { | |
238 | enum pid_type type; | |
239 | struct pid *pid; | |
240 | int who = arg; | |
241 | int result; | |
242 | type = PIDTYPE_PID; | |
243 | if (who < 0) { | |
244 | type = PIDTYPE_PGID; | |
245 | who = -who; | |
246 | } | |
247 | rcu_read_lock(); | |
b488893a | 248 | pid = find_vpid(who); |
609d7fa9 EB |
249 | result = __f_setown(filp, pid, type, force); |
250 | rcu_read_unlock(); | |
251 | return result; | |
252 | } | |
1da177e4 LT |
253 | EXPORT_SYMBOL(f_setown); |
254 | ||
255 | void f_delown(struct file *filp) | |
256 | { | |
2f38d70f | 257 | f_modown(filp, NULL, PIDTYPE_PID, 1); |
609d7fa9 EB |
258 | } |
259 | ||
260 | pid_t f_getown(struct file *filp) | |
261 | { | |
262 | pid_t pid; | |
43fa1adb | 263 | read_lock(&filp->f_owner.lock); |
6c5f3e7b | 264 | pid = pid_vnr(filp->f_owner.pid); |
609d7fa9 EB |
265 | if (filp->f_owner.pid_type == PIDTYPE_PGID) |
266 | pid = -pid; | |
43fa1adb | 267 | read_unlock(&filp->f_owner.lock); |
609d7fa9 | 268 | return pid; |
1da177e4 LT |
269 | } |
270 | ||
ba0a6c9f PZ |
271 | static int f_setown_ex(struct file *filp, unsigned long arg) |
272 | { | |
273 | struct f_owner_ex * __user owner_p = (void * __user)arg; | |
274 | struct f_owner_ex owner; | |
275 | struct pid *pid; | |
276 | int type; | |
277 | int ret; | |
278 | ||
279 | ret = copy_from_user(&owner, owner_p, sizeof(owner)); | |
280 | if (ret) | |
5b54470d | 281 | return -EFAULT; |
ba0a6c9f PZ |
282 | |
283 | switch (owner.type) { | |
284 | case F_OWNER_TID: | |
285 | type = PIDTYPE_MAX; | |
286 | break; | |
287 | ||
288 | case F_OWNER_PID: | |
289 | type = PIDTYPE_PID; | |
290 | break; | |
291 | ||
978b4053 | 292 | case F_OWNER_PGRP: |
ba0a6c9f PZ |
293 | type = PIDTYPE_PGID; |
294 | break; | |
295 | ||
296 | default: | |
297 | return -EINVAL; | |
298 | } | |
299 | ||
300 | rcu_read_lock(); | |
301 | pid = find_vpid(owner.pid); | |
302 | if (owner.pid && !pid) | |
303 | ret = -ESRCH; | |
304 | else | |
305 | ret = __f_setown(filp, pid, type, 1); | |
306 | rcu_read_unlock(); | |
307 | ||
308 | return ret; | |
309 | } | |
310 | ||
311 | static int f_getown_ex(struct file *filp, unsigned long arg) | |
312 | { | |
313 | struct f_owner_ex * __user owner_p = (void * __user)arg; | |
314 | struct f_owner_ex owner; | |
315 | int ret = 0; | |
316 | ||
317 | read_lock(&filp->f_owner.lock); | |
318 | owner.pid = pid_vnr(filp->f_owner.pid); | |
319 | switch (filp->f_owner.pid_type) { | |
320 | case PIDTYPE_MAX: | |
321 | owner.type = F_OWNER_TID; | |
322 | break; | |
323 | ||
324 | case PIDTYPE_PID: | |
325 | owner.type = F_OWNER_PID; | |
326 | break; | |
327 | ||
328 | case PIDTYPE_PGID: | |
978b4053 | 329 | owner.type = F_OWNER_PGRP; |
ba0a6c9f PZ |
330 | break; |
331 | ||
332 | default: | |
333 | WARN_ON(1); | |
334 | ret = -EINVAL; | |
335 | break; | |
336 | } | |
337 | read_unlock(&filp->f_owner.lock); | |
338 | ||
5b54470d | 339 | if (!ret) { |
ba0a6c9f | 340 | ret = copy_to_user(owner_p, &owner, sizeof(owner)); |
5b54470d DC |
341 | if (ret) |
342 | ret = -EFAULT; | |
343 | } | |
ba0a6c9f PZ |
344 | return ret; |
345 | } | |
346 | ||
1d151c33 CG |
347 | #ifdef CONFIG_CHECKPOINT_RESTORE |
348 | static int f_getowner_uids(struct file *filp, unsigned long arg) | |
349 | { | |
350 | struct user_namespace *user_ns = current_user_ns(); | |
351 | uid_t * __user dst = (void * __user)arg; | |
352 | uid_t src[2]; | |
353 | int err; | |
354 | ||
355 | read_lock(&filp->f_owner.lock); | |
356 | src[0] = from_kuid(user_ns, filp->f_owner.uid); | |
357 | src[1] = from_kuid(user_ns, filp->f_owner.euid); | |
358 | read_unlock(&filp->f_owner.lock); | |
359 | ||
360 | err = put_user(src[0], &dst[0]); | |
361 | err |= put_user(src[1], &dst[1]); | |
362 | ||
363 | return err; | |
364 | } | |
365 | #else | |
366 | static int f_getowner_uids(struct file *filp, unsigned long arg) | |
367 | { | |
368 | return -EINVAL; | |
369 | } | |
370 | #endif | |
371 | ||
1da177e4 LT |
372 | static long do_fcntl(int fd, unsigned int cmd, unsigned long arg, |
373 | struct file *filp) | |
374 | { | |
375 | long err = -EINVAL; | |
376 | ||
377 | switch (cmd) { | |
378 | case F_DUPFD: | |
22d2b35b | 379 | case F_DUPFD_CLOEXEC: |
d554ed89 | 380 | if (arg >= rlimit(RLIMIT_NOFILE)) |
4e1e018e | 381 | break; |
1027abe8 AV |
382 | err = alloc_fd(arg, cmd == F_DUPFD_CLOEXEC ? O_CLOEXEC : 0); |
383 | if (err >= 0) { | |
384 | get_file(filp); | |
385 | fd_install(err, filp); | |
386 | } | |
1da177e4 LT |
387 | break; |
388 | case F_GETFD: | |
389 | err = get_close_on_exec(fd) ? FD_CLOEXEC : 0; | |
390 | break; | |
391 | case F_SETFD: | |
392 | err = 0; | |
393 | set_close_on_exec(fd, arg & FD_CLOEXEC); | |
394 | break; | |
395 | case F_GETFL: | |
396 | err = filp->f_flags; | |
397 | break; | |
398 | case F_SETFL: | |
399 | err = setfl(fd, filp, arg); | |
400 | break; | |
401 | case F_GETLK: | |
402 | err = fcntl_getlk(filp, (struct flock __user *) arg); | |
403 | break; | |
404 | case F_SETLK: | |
405 | case F_SETLKW: | |
c293621b | 406 | err = fcntl_setlk(fd, filp, cmd, (struct flock __user *) arg); |
1da177e4 LT |
407 | break; |
408 | case F_GETOWN: | |
409 | /* | |
410 | * XXX If f_owner is a process group, the | |
411 | * negative return value will get converted | |
412 | * into an error. Oops. If we keep the | |
413 | * current syscall conventions, the only way | |
414 | * to fix this will be in libc. | |
415 | */ | |
609d7fa9 | 416 | err = f_getown(filp); |
1da177e4 LT |
417 | force_successful_syscall_return(); |
418 | break; | |
419 | case F_SETOWN: | |
420 | err = f_setown(filp, arg, 1); | |
421 | break; | |
ba0a6c9f PZ |
422 | case F_GETOWN_EX: |
423 | err = f_getown_ex(filp, arg); | |
424 | break; | |
425 | case F_SETOWN_EX: | |
426 | err = f_setown_ex(filp, arg); | |
427 | break; | |
1d151c33 CG |
428 | case F_GETOWNER_UIDS: |
429 | err = f_getowner_uids(filp, arg); | |
430 | break; | |
1da177e4 LT |
431 | case F_GETSIG: |
432 | err = filp->f_owner.signum; | |
433 | break; | |
434 | case F_SETSIG: | |
435 | /* arg == 0 restores default behaviour. */ | |
7ed20e1a | 436 | if (!valid_signal(arg)) { |
1da177e4 LT |
437 | break; |
438 | } | |
439 | err = 0; | |
440 | filp->f_owner.signum = arg; | |
441 | break; | |
442 | case F_GETLEASE: | |
443 | err = fcntl_getlease(filp); | |
444 | break; | |
445 | case F_SETLEASE: | |
446 | err = fcntl_setlease(fd, filp, arg); | |
447 | break; | |
448 | case F_NOTIFY: | |
449 | err = fcntl_dirnotify(fd, filp, arg); | |
450 | break; | |
35f3d14d JA |
451 | case F_SETPIPE_SZ: |
452 | case F_GETPIPE_SZ: | |
453 | err = pipe_fcntl(filp, cmd, arg); | |
454 | break; | |
1da177e4 LT |
455 | default: |
456 | break; | |
457 | } | |
458 | return err; | |
459 | } | |
460 | ||
1abf0c71 AV |
461 | static int check_fcntl_cmd(unsigned cmd) |
462 | { | |
463 | switch (cmd) { | |
464 | case F_DUPFD: | |
465 | case F_DUPFD_CLOEXEC: | |
466 | case F_GETFD: | |
467 | case F_SETFD: | |
468 | case F_GETFL: | |
469 | return 1; | |
470 | } | |
471 | return 0; | |
472 | } | |
473 | ||
a26eab24 | 474 | SYSCALL_DEFINE3(fcntl, unsigned int, fd, unsigned int, cmd, unsigned long, arg) |
1da177e4 LT |
475 | { |
476 | struct file *filp; | |
545ec2c7 | 477 | int fput_needed; |
1da177e4 LT |
478 | long err = -EBADF; |
479 | ||
545ec2c7 | 480 | filp = fget_raw_light(fd, &fput_needed); |
1da177e4 LT |
481 | if (!filp) |
482 | goto out; | |
483 | ||
1abf0c71 | 484 | if (unlikely(filp->f_mode & FMODE_PATH)) { |
545ec2c7 AV |
485 | if (!check_fcntl_cmd(cmd)) |
486 | goto out1; | |
1abf0c71 AV |
487 | } |
488 | ||
1da177e4 | 489 | err = security_file_fcntl(filp, cmd, arg); |
545ec2c7 AV |
490 | if (!err) |
491 | err = do_fcntl(fd, cmd, arg, filp); | |
1da177e4 | 492 | |
545ec2c7 AV |
493 | out1: |
494 | fput_light(filp, fput_needed); | |
1da177e4 LT |
495 | out: |
496 | return err; | |
497 | } | |
498 | ||
499 | #if BITS_PER_LONG == 32 | |
a26eab24 HC |
500 | SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd, |
501 | unsigned long, arg) | |
1da177e4 LT |
502 | { |
503 | struct file * filp; | |
545ec2c7 AV |
504 | long err = -EBADF; |
505 | int fput_needed; | |
1da177e4 | 506 | |
545ec2c7 | 507 | filp = fget_raw_light(fd, &fput_needed); |
1da177e4 LT |
508 | if (!filp) |
509 | goto out; | |
510 | ||
1abf0c71 | 511 | if (unlikely(filp->f_mode & FMODE_PATH)) { |
545ec2c7 AV |
512 | if (!check_fcntl_cmd(cmd)) |
513 | goto out1; | |
1abf0c71 AV |
514 | } |
515 | ||
1da177e4 | 516 | err = security_file_fcntl(filp, cmd, arg); |
545ec2c7 AV |
517 | if (err) |
518 | goto out1; | |
1da177e4 LT |
519 | |
520 | switch (cmd) { | |
521 | case F_GETLK64: | |
522 | err = fcntl_getlk64(filp, (struct flock64 __user *) arg); | |
523 | break; | |
524 | case F_SETLK64: | |
525 | case F_SETLKW64: | |
c293621b PS |
526 | err = fcntl_setlk64(fd, filp, cmd, |
527 | (struct flock64 __user *) arg); | |
1da177e4 LT |
528 | break; |
529 | default: | |
530 | err = do_fcntl(fd, cmd, arg, filp); | |
531 | break; | |
532 | } | |
545ec2c7 AV |
533 | out1: |
534 | fput_light(filp, fput_needed); | |
1da177e4 LT |
535 | out: |
536 | return err; | |
537 | } | |
538 | #endif | |
539 | ||
540 | /* Table to convert sigio signal codes into poll band bitmaps */ | |
541 | ||
fa3536cc | 542 | static const long band_table[NSIGPOLL] = { |
1da177e4 LT |
543 | POLLIN | POLLRDNORM, /* POLL_IN */ |
544 | POLLOUT | POLLWRNORM | POLLWRBAND, /* POLL_OUT */ | |
545 | POLLIN | POLLRDNORM | POLLMSG, /* POLL_MSG */ | |
546 | POLLERR, /* POLL_ERR */ | |
547 | POLLPRI | POLLRDBAND, /* POLL_PRI */ | |
548 | POLLHUP | POLLERR /* POLL_HUP */ | |
549 | }; | |
550 | ||
551 | static inline int sigio_perm(struct task_struct *p, | |
552 | struct fown_struct *fown, int sig) | |
553 | { | |
c69e8d9c DH |
554 | const struct cred *cred; |
555 | int ret; | |
556 | ||
557 | rcu_read_lock(); | |
558 | cred = __task_cred(p); | |
8e96e3b7 EB |
559 | ret = ((uid_eq(fown->euid, GLOBAL_ROOT_UID) || |
560 | uid_eq(fown->euid, cred->suid) || uid_eq(fown->euid, cred->uid) || | |
561 | uid_eq(fown->uid, cred->suid) || uid_eq(fown->uid, cred->uid)) && | |
c69e8d9c DH |
562 | !security_file_send_sigiotask(p, fown, sig)); |
563 | rcu_read_unlock(); | |
564 | return ret; | |
1da177e4 LT |
565 | } |
566 | ||
567 | static void send_sigio_to_task(struct task_struct *p, | |
8eeee4e2 | 568 | struct fown_struct *fown, |
ba0a6c9f | 569 | int fd, int reason, int group) |
1da177e4 | 570 | { |
8eeee4e2 ON |
571 | /* |
572 | * F_SETSIG can change ->signum lockless in parallel, make | |
573 | * sure we read it once and use the same value throughout. | |
574 | */ | |
575 | int signum = ACCESS_ONCE(fown->signum); | |
576 | ||
577 | if (!sigio_perm(p, fown, signum)) | |
1da177e4 LT |
578 | return; |
579 | ||
8eeee4e2 | 580 | switch (signum) { |
1da177e4 LT |
581 | siginfo_t si; |
582 | default: | |
583 | /* Queue a rt signal with the appropriate fd as its | |
584 | value. We use SI_SIGIO as the source, not | |
585 | SI_KERNEL, since kernel signals always get | |
586 | delivered even if we can't queue. Failure to | |
587 | queue in this case _should_ be reported; we fall | |
588 | back to SIGIO in that case. --sct */ | |
8eeee4e2 | 589 | si.si_signo = signum; |
1da177e4 LT |
590 | si.si_errno = 0; |
591 | si.si_code = reason; | |
592 | /* Make sure we are called with one of the POLL_* | |
593 | reasons, otherwise we could leak kernel stack into | |
594 | userspace. */ | |
f6298aab | 595 | BUG_ON((reason & __SI_MASK) != __SI_POLL); |
1da177e4 LT |
596 | if (reason - POLL_IN >= NSIGPOLL) |
597 | si.si_band = ~0L; | |
598 | else | |
599 | si.si_band = band_table[reason - POLL_IN]; | |
600 | si.si_fd = fd; | |
ba0a6c9f | 601 | if (!do_send_sig_info(signum, &si, p, group)) |
1da177e4 LT |
602 | break; |
603 | /* fall-through: fall back on the old plain SIGIO signal */ | |
604 | case 0: | |
ba0a6c9f | 605 | do_send_sig_info(SIGIO, SEND_SIG_PRIV, p, group); |
1da177e4 LT |
606 | } |
607 | } | |
608 | ||
609 | void send_sigio(struct fown_struct *fown, int fd, int band) | |
610 | { | |
611 | struct task_struct *p; | |
609d7fa9 EB |
612 | enum pid_type type; |
613 | struct pid *pid; | |
ba0a6c9f | 614 | int group = 1; |
1da177e4 LT |
615 | |
616 | read_lock(&fown->lock); | |
ba0a6c9f | 617 | |
609d7fa9 | 618 | type = fown->pid_type; |
ba0a6c9f PZ |
619 | if (type == PIDTYPE_MAX) { |
620 | group = 0; | |
621 | type = PIDTYPE_PID; | |
622 | } | |
623 | ||
1da177e4 LT |
624 | pid = fown->pid; |
625 | if (!pid) | |
626 | goto out_unlock_fown; | |
627 | ||
628 | read_lock(&tasklist_lock); | |
609d7fa9 | 629 | do_each_pid_task(pid, type, p) { |
ba0a6c9f | 630 | send_sigio_to_task(p, fown, fd, band, group); |
609d7fa9 | 631 | } while_each_pid_task(pid, type, p); |
1da177e4 LT |
632 | read_unlock(&tasklist_lock); |
633 | out_unlock_fown: | |
634 | read_unlock(&fown->lock); | |
635 | } | |
636 | ||
637 | static void send_sigurg_to_task(struct task_struct *p, | |
ba0a6c9f | 638 | struct fown_struct *fown, int group) |
1da177e4 LT |
639 | { |
640 | if (sigio_perm(p, fown, SIGURG)) | |
ba0a6c9f | 641 | do_send_sig_info(SIGURG, SEND_SIG_PRIV, p, group); |
1da177e4 LT |
642 | } |
643 | ||
644 | int send_sigurg(struct fown_struct *fown) | |
645 | { | |
646 | struct task_struct *p; | |
609d7fa9 EB |
647 | enum pid_type type; |
648 | struct pid *pid; | |
ba0a6c9f | 649 | int group = 1; |
609d7fa9 | 650 | int ret = 0; |
1da177e4 LT |
651 | |
652 | read_lock(&fown->lock); | |
ba0a6c9f | 653 | |
609d7fa9 | 654 | type = fown->pid_type; |
ba0a6c9f PZ |
655 | if (type == PIDTYPE_MAX) { |
656 | group = 0; | |
657 | type = PIDTYPE_PID; | |
658 | } | |
659 | ||
1da177e4 LT |
660 | pid = fown->pid; |
661 | if (!pid) | |
662 | goto out_unlock_fown; | |
663 | ||
664 | ret = 1; | |
665 | ||
666 | read_lock(&tasklist_lock); | |
609d7fa9 | 667 | do_each_pid_task(pid, type, p) { |
ba0a6c9f | 668 | send_sigurg_to_task(p, fown, group); |
609d7fa9 | 669 | } while_each_pid_task(pid, type, p); |
1da177e4 LT |
670 | read_unlock(&tasklist_lock); |
671 | out_unlock_fown: | |
672 | read_unlock(&fown->lock); | |
673 | return ret; | |
674 | } | |
675 | ||
989a2979 | 676 | static DEFINE_SPINLOCK(fasync_lock); |
e18b890b | 677 | static struct kmem_cache *fasync_cache __read_mostly; |
1da177e4 | 678 | |
989a2979 ED |
679 | static void fasync_free_rcu(struct rcu_head *head) |
680 | { | |
681 | kmem_cache_free(fasync_cache, | |
682 | container_of(head, struct fasync_struct, fa_rcu)); | |
683 | } | |
684 | ||
1da177e4 | 685 | /* |
53281b6d LT |
686 | * Remove a fasync entry. If successfully removed, return |
687 | * positive and clear the FASYNC flag. If no entry exists, | |
688 | * do nothing and return 0. | |
689 | * | |
690 | * NOTE! It is very important that the FASYNC flag always | |
691 | * match the state "is the filp on a fasync list". | |
692 | * | |
1da177e4 | 693 | */ |
f7347ce4 | 694 | int fasync_remove_entry(struct file *filp, struct fasync_struct **fapp) |
1da177e4 LT |
695 | { |
696 | struct fasync_struct *fa, **fp; | |
1da177e4 LT |
697 | int result = 0; |
698 | ||
53281b6d | 699 | spin_lock(&filp->f_lock); |
989a2979 | 700 | spin_lock(&fasync_lock); |
53281b6d LT |
701 | for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) { |
702 | if (fa->fa_file != filp) | |
703 | continue; | |
989a2979 ED |
704 | |
705 | spin_lock_irq(&fa->fa_lock); | |
706 | fa->fa_file = NULL; | |
707 | spin_unlock_irq(&fa->fa_lock); | |
708 | ||
53281b6d | 709 | *fp = fa->fa_next; |
989a2979 | 710 | call_rcu(&fa->fa_rcu, fasync_free_rcu); |
53281b6d LT |
711 | filp->f_flags &= ~FASYNC; |
712 | result = 1; | |
713 | break; | |
1da177e4 | 714 | } |
989a2979 | 715 | spin_unlock(&fasync_lock); |
53281b6d LT |
716 | spin_unlock(&filp->f_lock); |
717 | return result; | |
718 | } | |
719 | ||
f7347ce4 LT |
720 | struct fasync_struct *fasync_alloc(void) |
721 | { | |
722 | return kmem_cache_alloc(fasync_cache, GFP_KERNEL); | |
723 | } | |
724 | ||
53281b6d | 725 | /* |
f7347ce4 LT |
726 | * NOTE! This can be used only for unused fasync entries: |
727 | * entries that actually got inserted on the fasync list | |
728 | * need to be released by rcu - see fasync_remove_entry. | |
53281b6d | 729 | */ |
f7347ce4 | 730 | void fasync_free(struct fasync_struct *new) |
53281b6d | 731 | { |
f7347ce4 LT |
732 | kmem_cache_free(fasync_cache, new); |
733 | } | |
53281b6d | 734 | |
f7347ce4 LT |
735 | /* |
736 | * Insert a new entry into the fasync list. Return the pointer to the | |
737 | * old one if we didn't use the new one. | |
55f335a8 LT |
738 | * |
739 | * NOTE! It is very important that the FASYNC flag always | |
740 | * match the state "is the filp on a fasync list". | |
f7347ce4 LT |
741 | */ |
742 | struct fasync_struct *fasync_insert_entry(int fd, struct file *filp, struct fasync_struct **fapp, struct fasync_struct *new) | |
743 | { | |
744 | struct fasync_struct *fa, **fp; | |
4a6a4499 | 745 | |
4a6a4499 | 746 | spin_lock(&filp->f_lock); |
989a2979 | 747 | spin_lock(&fasync_lock); |
1da177e4 | 748 | for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) { |
53281b6d LT |
749 | if (fa->fa_file != filp) |
750 | continue; | |
989a2979 ED |
751 | |
752 | spin_lock_irq(&fa->fa_lock); | |
53281b6d | 753 | fa->fa_fd = fd; |
989a2979 | 754 | spin_unlock_irq(&fa->fa_lock); |
53281b6d | 755 | goto out; |
1da177e4 LT |
756 | } |
757 | ||
989a2979 | 758 | spin_lock_init(&new->fa_lock); |
53281b6d LT |
759 | new->magic = FASYNC_MAGIC; |
760 | new->fa_file = filp; | |
761 | new->fa_fd = fd; | |
762 | new->fa_next = *fapp; | |
989a2979 | 763 | rcu_assign_pointer(*fapp, new); |
53281b6d LT |
764 | filp->f_flags |= FASYNC; |
765 | ||
1da177e4 | 766 | out: |
989a2979 | 767 | spin_unlock(&fasync_lock); |
4a6a4499 | 768 | spin_unlock(&filp->f_lock); |
f7347ce4 LT |
769 | return fa; |
770 | } | |
771 | ||
772 | /* | |
773 | * Add a fasync entry. Return negative on error, positive if | |
774 | * added, and zero if did nothing but change an existing one. | |
f7347ce4 LT |
775 | */ |
776 | static int fasync_add_entry(int fd, struct file *filp, struct fasync_struct **fapp) | |
777 | { | |
778 | struct fasync_struct *new; | |
779 | ||
780 | new = fasync_alloc(); | |
781 | if (!new) | |
782 | return -ENOMEM; | |
783 | ||
784 | /* | |
785 | * fasync_insert_entry() returns the old (update) entry if | |
786 | * it existed. | |
787 | * | |
788 | * So free the (unused) new entry and return 0 to let the | |
789 | * caller know that we didn't add any new fasync entries. | |
790 | */ | |
791 | if (fasync_insert_entry(fd, filp, fapp, new)) { | |
792 | fasync_free(new); | |
793 | return 0; | |
794 | } | |
795 | ||
796 | return 1; | |
1da177e4 LT |
797 | } |
798 | ||
53281b6d LT |
799 | /* |
800 | * fasync_helper() is used by almost all character device drivers | |
801 | * to set up the fasync queue, and for regular files by the file | |
802 | * lease code. It returns negative on error, 0 if it did no changes | |
803 | * and positive if it added/deleted the entry. | |
804 | */ | |
805 | int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp) | |
806 | { | |
807 | if (!on) | |
808 | return fasync_remove_entry(filp, fapp); | |
809 | return fasync_add_entry(fd, filp, fapp); | |
810 | } | |
811 | ||
1da177e4 LT |
812 | EXPORT_SYMBOL(fasync_helper); |
813 | ||
989a2979 ED |
814 | /* |
815 | * rcu_read_lock() is held | |
816 | */ | |
817 | static void kill_fasync_rcu(struct fasync_struct *fa, int sig, int band) | |
1da177e4 LT |
818 | { |
819 | while (fa) { | |
989a2979 | 820 | struct fown_struct *fown; |
f4985dc7 AM |
821 | unsigned long flags; |
822 | ||
1da177e4 LT |
823 | if (fa->magic != FASYNC_MAGIC) { |
824 | printk(KERN_ERR "kill_fasync: bad magic number in " | |
825 | "fasync_struct!\n"); | |
826 | return; | |
827 | } | |
f4985dc7 | 828 | spin_lock_irqsave(&fa->fa_lock, flags); |
989a2979 ED |
829 | if (fa->fa_file) { |
830 | fown = &fa->fa_file->f_owner; | |
831 | /* Don't send SIGURG to processes which have not set a | |
832 | queued signum: SIGURG has its own default signalling | |
833 | mechanism. */ | |
834 | if (!(sig == SIGURG && fown->signum == 0)) | |
835 | send_sigio(fown, fa->fa_fd, band); | |
836 | } | |
f4985dc7 | 837 | spin_unlock_irqrestore(&fa->fa_lock, flags); |
989a2979 | 838 | fa = rcu_dereference(fa->fa_next); |
1da177e4 LT |
839 | } |
840 | } | |
841 | ||
1da177e4 LT |
842 | void kill_fasync(struct fasync_struct **fp, int sig, int band) |
843 | { | |
844 | /* First a quick test without locking: usually | |
845 | * the list is empty. | |
846 | */ | |
847 | if (*fp) { | |
989a2979 ED |
848 | rcu_read_lock(); |
849 | kill_fasync_rcu(rcu_dereference(*fp), sig, band); | |
850 | rcu_read_unlock(); | |
1da177e4 LT |
851 | } |
852 | } | |
853 | EXPORT_SYMBOL(kill_fasync); | |
854 | ||
454eedb8 | 855 | static int __init fcntl_init(void) |
1da177e4 | 856 | { |
3ab04d5c JB |
857 | /* |
858 | * Please add new bits here to ensure allocation uniqueness. | |
859 | * Exceptions: O_NONBLOCK is a two bit define on parisc; O_NDELAY | |
860 | * is defined as O_NONBLOCK on some platforms and not on others. | |
861 | */ | |
1abf0c71 | 862 | BUILD_BUG_ON(19 - 1 /* for O_RDONLY being 0 */ != HWEIGHT32( |
454eedb8 WF |
863 | O_RDONLY | O_WRONLY | O_RDWR | |
864 | O_CREAT | O_EXCL | O_NOCTTY | | |
3ab04d5c | 865 | O_TRUNC | O_APPEND | /* O_NONBLOCK | */ |
454eedb8 WF |
866 | __O_SYNC | O_DSYNC | FASYNC | |
867 | O_DIRECT | O_LARGEFILE | O_DIRECTORY | | |
868 | O_NOFOLLOW | O_NOATIME | O_CLOEXEC | | |
1abf0c71 | 869 | __FMODE_EXEC | O_PATH |
454eedb8 WF |
870 | )); |
871 | ||
1da177e4 | 872 | fasync_cache = kmem_cache_create("fasync_cache", |
20c2df83 | 873 | sizeof(struct fasync_struct), 0, SLAB_PANIC, NULL); |
1da177e4 LT |
874 | return 0; |
875 | } | |
876 | ||
454eedb8 | 877 | module_init(fcntl_init) |