xref: /openbmc/linux/fs/fcntl.c (revision 0e01d176d5788f66dc64a7e61119edb56eb08339)
1  // SPDX-License-Identifier: GPL-2.0
2  /*
3   *  linux/fs/fcntl.c
4   *
5   *  Copyright (C) 1991, 1992  Linus Torvalds
6   */
7  
8  #include <linux/syscalls.h>
9  #include <linux/init.h>
10  #include <linux/mm.h>
11  #include <linux/sched/task.h>
12  #include <linux/fs.h>
13  #include <linux/file.h>
14  #include <linux/fdtable.h>
15  #include <linux/capability.h>
16  #include <linux/dnotify.h>
17  #include <linux/slab.h>
18  #include <linux/module.h>
19  #include <linux/pipe_fs_i.h>
20  #include <linux/security.h>
21  #include <linux/ptrace.h>
22  #include <linux/signal.h>
23  #include <linux/rcupdate.h>
24  #include <linux/pid_namespace.h>
25  #include <linux/user_namespace.h>
26  #include <linux/memfd.h>
27  #include <linux/compat.h>
28  #include <linux/mount.h>
29  
30  #include <linux/poll.h>
31  #include <asm/siginfo.h>
32  #include <linux/uaccess.h>
33  
34  #define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | O_DIRECT | O_NOATIME)
35  
36  static int setfl(int fd, struct file * filp, unsigned long arg)
37  {
38  	struct inode * inode = file_inode(filp);
39  	int error = 0;
40  
41  	/*
42  	 * O_APPEND cannot be cleared if the file is marked as append-only
43  	 * and the file is open for write.
44  	 */
45  	if (((arg ^ filp->f_flags) & O_APPEND) && IS_APPEND(inode))
46  		return -EPERM;
47  
48  	/* O_NOATIME can only be set by the owner or superuser */
49  	if ((arg & O_NOATIME) && !(filp->f_flags & O_NOATIME))
50  		if (!inode_owner_or_capable(file_mnt_user_ns(filp), inode))
51  			return -EPERM;
52  
53  	/* required for strict SunOS emulation */
54  	if (O_NONBLOCK != O_NDELAY)
55  	       if (arg & O_NDELAY)
56  		   arg |= O_NONBLOCK;
57  
58  	/* Pipe packetized mode is controlled by O_DIRECT flag */
59  	if (!S_ISFIFO(inode->i_mode) && (arg & O_DIRECT)) {
60  		if (!filp->f_mapping || !filp->f_mapping->a_ops ||
61  			!filp->f_mapping->a_ops->direct_IO)
62  				return -EINVAL;
63  	}
64  
65  	if (filp->f_op->check_flags)
66  		error = filp->f_op->check_flags(arg);
67  	if (error)
68  		return error;
69  
70  	/*
71  	 * ->fasync() is responsible for setting the FASYNC bit.
72  	 */
73  	if (((arg ^ filp->f_flags) & FASYNC) && filp->f_op->fasync) {
74  		error = filp->f_op->fasync(fd, filp, (arg & FASYNC) != 0);
75  		if (error < 0)
76  			goto out;
77  		if (error > 0)
78  			error = 0;
79  	}
80  	spin_lock(&filp->f_lock);
81  	filp->f_flags = (arg & SETFL_MASK) | (filp->f_flags & ~SETFL_MASK);
82  	spin_unlock(&filp->f_lock);
83  
84   out:
85  	return error;
86  }
87  
88  static void f_modown(struct file *filp, struct pid *pid, enum pid_type type,
89                       int force)
90  {
91  	write_lock_irq(&filp->f_owner.lock);
92  	if (force || !filp->f_owner.pid) {
93  		put_pid(filp->f_owner.pid);
94  		filp->f_owner.pid = get_pid(pid);
95  		filp->f_owner.pid_type = type;
96  
97  		if (pid) {
98  			const struct cred *cred = current_cred();
99  			filp->f_owner.uid = cred->uid;
100  			filp->f_owner.euid = cred->euid;
101  		}
102  	}
103  	write_unlock_irq(&filp->f_owner.lock);
104  }
105  
106  void __f_setown(struct file *filp, struct pid *pid, enum pid_type type,
107  		int force)
108  {
109  	security_file_set_fowner(filp);
110  	f_modown(filp, pid, type, force);
111  }
112  EXPORT_SYMBOL(__f_setown);
113  
114  int f_setown(struct file *filp, unsigned long arg, int force)
115  {
116  	enum pid_type type;
117  	struct pid *pid = NULL;
118  	int who = arg, ret = 0;
119  
120  	type = PIDTYPE_TGID;
121  	if (who < 0) {
122  		/* avoid overflow below */
123  		if (who == INT_MIN)
124  			return -EINVAL;
125  
126  		type = PIDTYPE_PGID;
127  		who = -who;
128  	}
129  
130  	rcu_read_lock();
131  	if (who) {
132  		pid = find_vpid(who);
133  		if (!pid)
134  			ret = -ESRCH;
135  	}
136  
137  	if (!ret)
138  		__f_setown(filp, pid, type, force);
139  	rcu_read_unlock();
140  
141  	return ret;
142  }
143  EXPORT_SYMBOL(f_setown);
144  
145  void f_delown(struct file *filp)
146  {
147  	f_modown(filp, NULL, PIDTYPE_TGID, 1);
148  }
149  
150  pid_t f_getown(struct file *filp)
151  {
152  	pid_t pid = 0;
153  
154  	read_lock_irq(&filp->f_owner.lock);
155  	rcu_read_lock();
156  	if (pid_task(filp->f_owner.pid, filp->f_owner.pid_type)) {
157  		pid = pid_vnr(filp->f_owner.pid);
158  		if (filp->f_owner.pid_type == PIDTYPE_PGID)
159  			pid = -pid;
160  	}
161  	rcu_read_unlock();
162  	read_unlock_irq(&filp->f_owner.lock);
163  	return pid;
164  }
165  
166  static int f_setown_ex(struct file *filp, unsigned long arg)
167  {
168  	struct f_owner_ex __user *owner_p = (void __user *)arg;
169  	struct f_owner_ex owner;
170  	struct pid *pid;
171  	int type;
172  	int ret;
173  
174  	ret = copy_from_user(&owner, owner_p, sizeof(owner));
175  	if (ret)
176  		return -EFAULT;
177  
178  	switch (owner.type) {
179  	case F_OWNER_TID:
180  		type = PIDTYPE_PID;
181  		break;
182  
183  	case F_OWNER_PID:
184  		type = PIDTYPE_TGID;
185  		break;
186  
187  	case F_OWNER_PGRP:
188  		type = PIDTYPE_PGID;
189  		break;
190  
191  	default:
192  		return -EINVAL;
193  	}
194  
195  	rcu_read_lock();
196  	pid = find_vpid(owner.pid);
197  	if (owner.pid && !pid)
198  		ret = -ESRCH;
199  	else
200  		 __f_setown(filp, pid, type, 1);
201  	rcu_read_unlock();
202  
203  	return ret;
204  }
205  
206  static int f_getown_ex(struct file *filp, unsigned long arg)
207  {
208  	struct f_owner_ex __user *owner_p = (void __user *)arg;
209  	struct f_owner_ex owner = {};
210  	int ret = 0;
211  
212  	read_lock_irq(&filp->f_owner.lock);
213  	rcu_read_lock();
214  	if (pid_task(filp->f_owner.pid, filp->f_owner.pid_type))
215  		owner.pid = pid_vnr(filp->f_owner.pid);
216  	rcu_read_unlock();
217  	switch (filp->f_owner.pid_type) {
218  	case PIDTYPE_PID:
219  		owner.type = F_OWNER_TID;
220  		break;
221  
222  	case PIDTYPE_TGID:
223  		owner.type = F_OWNER_PID;
224  		break;
225  
226  	case PIDTYPE_PGID:
227  		owner.type = F_OWNER_PGRP;
228  		break;
229  
230  	default:
231  		WARN_ON(1);
232  		ret = -EINVAL;
233  		break;
234  	}
235  	read_unlock_irq(&filp->f_owner.lock);
236  
237  	if (!ret) {
238  		ret = copy_to_user(owner_p, &owner, sizeof(owner));
239  		if (ret)
240  			ret = -EFAULT;
241  	}
242  	return ret;
243  }
244  
245  #ifdef CONFIG_CHECKPOINT_RESTORE
246  static int f_getowner_uids(struct file *filp, unsigned long arg)
247  {
248  	struct user_namespace *user_ns = current_user_ns();
249  	uid_t __user *dst = (void __user *)arg;
250  	uid_t src[2];
251  	int err;
252  
253  	read_lock_irq(&filp->f_owner.lock);
254  	src[0] = from_kuid(user_ns, filp->f_owner.uid);
255  	src[1] = from_kuid(user_ns, filp->f_owner.euid);
256  	read_unlock_irq(&filp->f_owner.lock);
257  
258  	err  = put_user(src[0], &dst[0]);
259  	err |= put_user(src[1], &dst[1]);
260  
261  	return err;
262  }
263  #else
264  static int f_getowner_uids(struct file *filp, unsigned long arg)
265  {
266  	return -EINVAL;
267  }
268  #endif
269  
270  static bool rw_hint_valid(enum rw_hint hint)
271  {
272  	switch (hint) {
273  	case RWH_WRITE_LIFE_NOT_SET:
274  	case RWH_WRITE_LIFE_NONE:
275  	case RWH_WRITE_LIFE_SHORT:
276  	case RWH_WRITE_LIFE_MEDIUM:
277  	case RWH_WRITE_LIFE_LONG:
278  	case RWH_WRITE_LIFE_EXTREME:
279  		return true;
280  	default:
281  		return false;
282  	}
283  }
284  
285  static long fcntl_rw_hint(struct file *file, unsigned int cmd,
286  			  unsigned long arg)
287  {
288  	struct inode *inode = file_inode(file);
289  	u64 __user *argp = (u64 __user *)arg;
290  	enum rw_hint hint;
291  	u64 h;
292  
293  	switch (cmd) {
294  	case F_GET_RW_HINT:
295  		h = inode->i_write_hint;
296  		if (copy_to_user(argp, &h, sizeof(*argp)))
297  			return -EFAULT;
298  		return 0;
299  	case F_SET_RW_HINT:
300  		if (copy_from_user(&h, argp, sizeof(h)))
301  			return -EFAULT;
302  		hint = (enum rw_hint) h;
303  		if (!rw_hint_valid(hint))
304  			return -EINVAL;
305  
306  		inode_lock(inode);
307  		inode->i_write_hint = hint;
308  		inode_unlock(inode);
309  		return 0;
310  	default:
311  		return -EINVAL;
312  	}
313  }
314  
315  static long do_fcntl(int fd, unsigned int cmd, unsigned long arg,
316  		struct file *filp)
317  {
318  	void __user *argp = (void __user *)arg;
319  	struct flock flock;
320  	long err = -EINVAL;
321  
322  	switch (cmd) {
323  	case F_DUPFD:
324  		err = f_dupfd(arg, filp, 0);
325  		break;
326  	case F_DUPFD_CLOEXEC:
327  		err = f_dupfd(arg, filp, O_CLOEXEC);
328  		break;
329  	case F_GETFD:
330  		err = get_close_on_exec(fd) ? FD_CLOEXEC : 0;
331  		break;
332  	case F_SETFD:
333  		err = 0;
334  		set_close_on_exec(fd, arg & FD_CLOEXEC);
335  		break;
336  	case F_GETFL:
337  		err = filp->f_flags;
338  		break;
339  	case F_SETFL:
340  		err = setfl(fd, filp, arg);
341  		break;
342  #if BITS_PER_LONG != 32
343  	/* 32-bit arches must use fcntl64() */
344  	case F_OFD_GETLK:
345  #endif
346  	case F_GETLK:
347  		if (copy_from_user(&flock, argp, sizeof(flock)))
348  			return -EFAULT;
349  		err = fcntl_getlk(filp, cmd, &flock);
350  		if (!err && copy_to_user(argp, &flock, sizeof(flock)))
351  			return -EFAULT;
352  		break;
353  #if BITS_PER_LONG != 32
354  	/* 32-bit arches must use fcntl64() */
355  	case F_OFD_SETLK:
356  	case F_OFD_SETLKW:
357  		fallthrough;
358  #endif
359  	case F_SETLK:
360  	case F_SETLKW:
361  		if (copy_from_user(&flock, argp, sizeof(flock)))
362  			return -EFAULT;
363  		err = fcntl_setlk(fd, filp, cmd, &flock);
364  		break;
365  	case F_GETOWN:
366  		/*
367  		 * XXX If f_owner is a process group, the
368  		 * negative return value will get converted
369  		 * into an error.  Oops.  If we keep the
370  		 * current syscall conventions, the only way
371  		 * to fix this will be in libc.
372  		 */
373  		err = f_getown(filp);
374  		force_successful_syscall_return();
375  		break;
376  	case F_SETOWN:
377  		err = f_setown(filp, arg, 1);
378  		break;
379  	case F_GETOWN_EX:
380  		err = f_getown_ex(filp, arg);
381  		break;
382  	case F_SETOWN_EX:
383  		err = f_setown_ex(filp, arg);
384  		break;
385  	case F_GETOWNER_UIDS:
386  		err = f_getowner_uids(filp, arg);
387  		break;
388  	case F_GETSIG:
389  		err = filp->f_owner.signum;
390  		break;
391  	case F_SETSIG:
392  		/* arg == 0 restores default behaviour. */
393  		if (!valid_signal(arg)) {
394  			break;
395  		}
396  		err = 0;
397  		filp->f_owner.signum = arg;
398  		break;
399  	case F_GETLEASE:
400  		err = fcntl_getlease(filp);
401  		break;
402  	case F_SETLEASE:
403  		err = fcntl_setlease(fd, filp, arg);
404  		break;
405  	case F_NOTIFY:
406  		err = fcntl_dirnotify(fd, filp, arg);
407  		break;
408  	case F_SETPIPE_SZ:
409  	case F_GETPIPE_SZ:
410  		err = pipe_fcntl(filp, cmd, arg);
411  		break;
412  	case F_ADD_SEALS:
413  	case F_GET_SEALS:
414  		err = memfd_fcntl(filp, cmd, arg);
415  		break;
416  	case F_GET_RW_HINT:
417  	case F_SET_RW_HINT:
418  		err = fcntl_rw_hint(filp, cmd, arg);
419  		break;
420  	default:
421  		break;
422  	}
423  	return err;
424  }
425  
426  static int check_fcntl_cmd(unsigned cmd)
427  {
428  	switch (cmd) {
429  	case F_DUPFD:
430  	case F_DUPFD_CLOEXEC:
431  	case F_GETFD:
432  	case F_SETFD:
433  	case F_GETFL:
434  		return 1;
435  	}
436  	return 0;
437  }
438  
439  SYSCALL_DEFINE3(fcntl, unsigned int, fd, unsigned int, cmd, unsigned long, arg)
440  {
441  	struct fd f = fdget_raw(fd);
442  	long err = -EBADF;
443  
444  	if (!f.file)
445  		goto out;
446  
447  	if (unlikely(f.file->f_mode & FMODE_PATH)) {
448  		if (!check_fcntl_cmd(cmd))
449  			goto out1;
450  	}
451  
452  	err = security_file_fcntl(f.file, cmd, arg);
453  	if (!err)
454  		err = do_fcntl(fd, cmd, arg, f.file);
455  
456  out1:
457   	fdput(f);
458  out:
459  	return err;
460  }
461  
462  #if BITS_PER_LONG == 32
463  SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd,
464  		unsigned long, arg)
465  {
466  	void __user *argp = (void __user *)arg;
467  	struct fd f = fdget_raw(fd);
468  	struct flock64 flock;
469  	long err = -EBADF;
470  
471  	if (!f.file)
472  		goto out;
473  
474  	if (unlikely(f.file->f_mode & FMODE_PATH)) {
475  		if (!check_fcntl_cmd(cmd))
476  			goto out1;
477  	}
478  
479  	err = security_file_fcntl(f.file, cmd, arg);
480  	if (err)
481  		goto out1;
482  
483  	switch (cmd) {
484  	case F_GETLK64:
485  	case F_OFD_GETLK:
486  		err = -EFAULT;
487  		if (copy_from_user(&flock, argp, sizeof(flock)))
488  			break;
489  		err = fcntl_getlk64(f.file, cmd, &flock);
490  		if (!err && copy_to_user(argp, &flock, sizeof(flock)))
491  			err = -EFAULT;
492  		break;
493  	case F_SETLK64:
494  	case F_SETLKW64:
495  	case F_OFD_SETLK:
496  	case F_OFD_SETLKW:
497  		err = -EFAULT;
498  		if (copy_from_user(&flock, argp, sizeof(flock)))
499  			break;
500  		err = fcntl_setlk64(fd, f.file, cmd, &flock);
501  		break;
502  	default:
503  		err = do_fcntl(fd, cmd, arg, f.file);
504  		break;
505  	}
506  out1:
507  	fdput(f);
508  out:
509  	return err;
510  }
511  #endif
512  
513  #ifdef CONFIG_COMPAT
514  /* careful - don't use anywhere else */
515  #define copy_flock_fields(dst, src)		\
516  	(dst)->l_type = (src)->l_type;		\
517  	(dst)->l_whence = (src)->l_whence;	\
518  	(dst)->l_start = (src)->l_start;	\
519  	(dst)->l_len = (src)->l_len;		\
520  	(dst)->l_pid = (src)->l_pid;
521  
522  static int get_compat_flock(struct flock *kfl, const struct compat_flock __user *ufl)
523  {
524  	struct compat_flock fl;
525  
526  	if (copy_from_user(&fl, ufl, sizeof(struct compat_flock)))
527  		return -EFAULT;
528  	copy_flock_fields(kfl, &fl);
529  	return 0;
530  }
531  
532  static int get_compat_flock64(struct flock *kfl, const struct compat_flock64 __user *ufl)
533  {
534  	struct compat_flock64 fl;
535  
536  	if (copy_from_user(&fl, ufl, sizeof(struct compat_flock64)))
537  		return -EFAULT;
538  	copy_flock_fields(kfl, &fl);
539  	return 0;
540  }
541  
542  static int put_compat_flock(const struct flock *kfl, struct compat_flock __user *ufl)
543  {
544  	struct compat_flock fl;
545  
546  	memset(&fl, 0, sizeof(struct compat_flock));
547  	copy_flock_fields(&fl, kfl);
548  	if (copy_to_user(ufl, &fl, sizeof(struct compat_flock)))
549  		return -EFAULT;
550  	return 0;
551  }
552  
553  static int put_compat_flock64(const struct flock *kfl, struct compat_flock64 __user *ufl)
554  {
555  	struct compat_flock64 fl;
556  
557  	BUILD_BUG_ON(sizeof(kfl->l_start) > sizeof(ufl->l_start));
558  	BUILD_BUG_ON(sizeof(kfl->l_len) > sizeof(ufl->l_len));
559  
560  	memset(&fl, 0, sizeof(struct compat_flock64));
561  	copy_flock_fields(&fl, kfl);
562  	if (copy_to_user(ufl, &fl, sizeof(struct compat_flock64)))
563  		return -EFAULT;
564  	return 0;
565  }
566  #undef copy_flock_fields
567  
568  static unsigned int
569  convert_fcntl_cmd(unsigned int cmd)
570  {
571  	switch (cmd) {
572  	case F_GETLK64:
573  		return F_GETLK;
574  	case F_SETLK64:
575  		return F_SETLK;
576  	case F_SETLKW64:
577  		return F_SETLKW;
578  	}
579  
580  	return cmd;
581  }
582  
583  /*
584   * GETLK was successful and we need to return the data, but it needs to fit in
585   * the compat structure.
586   * l_start shouldn't be too big, unless the original start + end is greater than
587   * COMPAT_OFF_T_MAX, in which case the app was asking for trouble, so we return
588   * -EOVERFLOW in that case.  l_len could be too big, in which case we just
589   * truncate it, and only allow the app to see that part of the conflicting lock
590   * that might make sense to it anyway
591   */
592  static int fixup_compat_flock(struct flock *flock)
593  {
594  	if (flock->l_start > COMPAT_OFF_T_MAX)
595  		return -EOVERFLOW;
596  	if (flock->l_len > COMPAT_OFF_T_MAX)
597  		flock->l_len = COMPAT_OFF_T_MAX;
598  	return 0;
599  }
600  
601  static long do_compat_fcntl64(unsigned int fd, unsigned int cmd,
602  			     compat_ulong_t arg)
603  {
604  	struct fd f = fdget_raw(fd);
605  	struct flock flock;
606  	long err = -EBADF;
607  
608  	if (!f.file)
609  		return err;
610  
611  	if (unlikely(f.file->f_mode & FMODE_PATH)) {
612  		if (!check_fcntl_cmd(cmd))
613  			goto out_put;
614  	}
615  
616  	err = security_file_fcntl(f.file, cmd, arg);
617  	if (err)
618  		goto out_put;
619  
620  	switch (cmd) {
621  	case F_GETLK:
622  		err = get_compat_flock(&flock, compat_ptr(arg));
623  		if (err)
624  			break;
625  		err = fcntl_getlk(f.file, convert_fcntl_cmd(cmd), &flock);
626  		if (err)
627  			break;
628  		err = fixup_compat_flock(&flock);
629  		if (!err)
630  			err = put_compat_flock(&flock, compat_ptr(arg));
631  		break;
632  	case F_GETLK64:
633  	case F_OFD_GETLK:
634  		err = get_compat_flock64(&flock, compat_ptr(arg));
635  		if (err)
636  			break;
637  		err = fcntl_getlk(f.file, convert_fcntl_cmd(cmd), &flock);
638  		if (!err)
639  			err = put_compat_flock64(&flock, compat_ptr(arg));
640  		break;
641  	case F_SETLK:
642  	case F_SETLKW:
643  		err = get_compat_flock(&flock, compat_ptr(arg));
644  		if (err)
645  			break;
646  		err = fcntl_setlk(fd, f.file, convert_fcntl_cmd(cmd), &flock);
647  		break;
648  	case F_SETLK64:
649  	case F_SETLKW64:
650  	case F_OFD_SETLK:
651  	case F_OFD_SETLKW:
652  		err = get_compat_flock64(&flock, compat_ptr(arg));
653  		if (err)
654  			break;
655  		err = fcntl_setlk(fd, f.file, convert_fcntl_cmd(cmd), &flock);
656  		break;
657  	default:
658  		err = do_fcntl(fd, cmd, arg, f.file);
659  		break;
660  	}
661  out_put:
662  	fdput(f);
663  	return err;
664  }
665  
666  COMPAT_SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd,
667  		       compat_ulong_t, arg)
668  {
669  	return do_compat_fcntl64(fd, cmd, arg);
670  }
671  
672  COMPAT_SYSCALL_DEFINE3(fcntl, unsigned int, fd, unsigned int, cmd,
673  		       compat_ulong_t, arg)
674  {
675  	switch (cmd) {
676  	case F_GETLK64:
677  	case F_SETLK64:
678  	case F_SETLKW64:
679  	case F_OFD_GETLK:
680  	case F_OFD_SETLK:
681  	case F_OFD_SETLKW:
682  		return -EINVAL;
683  	}
684  	return do_compat_fcntl64(fd, cmd, arg);
685  }
686  #endif
687  
688  /* Table to convert sigio signal codes into poll band bitmaps */
689  
690  static const __poll_t band_table[NSIGPOLL] = {
691  	EPOLLIN | EPOLLRDNORM,			/* POLL_IN */
692  	EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND,	/* POLL_OUT */
693  	EPOLLIN | EPOLLRDNORM | EPOLLMSG,		/* POLL_MSG */
694  	EPOLLERR,				/* POLL_ERR */
695  	EPOLLPRI | EPOLLRDBAND,			/* POLL_PRI */
696  	EPOLLHUP | EPOLLERR			/* POLL_HUP */
697  };
698  
699  static inline int sigio_perm(struct task_struct *p,
700                               struct fown_struct *fown, int sig)
701  {
702  	const struct cred *cred;
703  	int ret;
704  
705  	rcu_read_lock();
706  	cred = __task_cred(p);
707  	ret = ((uid_eq(fown->euid, GLOBAL_ROOT_UID) ||
708  		uid_eq(fown->euid, cred->suid) || uid_eq(fown->euid, cred->uid) ||
709  		uid_eq(fown->uid,  cred->suid) || uid_eq(fown->uid,  cred->uid)) &&
710  	       !security_file_send_sigiotask(p, fown, sig));
711  	rcu_read_unlock();
712  	return ret;
713  }
714  
715  static void send_sigio_to_task(struct task_struct *p,
716  			       struct fown_struct *fown,
717  			       int fd, int reason, enum pid_type type)
718  {
719  	/*
720  	 * F_SETSIG can change ->signum lockless in parallel, make
721  	 * sure we read it once and use the same value throughout.
722  	 */
723  	int signum = READ_ONCE(fown->signum);
724  
725  	if (!sigio_perm(p, fown, signum))
726  		return;
727  
728  	switch (signum) {
729  		default: {
730  			kernel_siginfo_t si;
731  
732  			/* Queue a rt signal with the appropriate fd as its
733  			   value.  We use SI_SIGIO as the source, not
734  			   SI_KERNEL, since kernel signals always get
735  			   delivered even if we can't queue.  Failure to
736  			   queue in this case _should_ be reported; we fall
737  			   back to SIGIO in that case. --sct */
738  			clear_siginfo(&si);
739  			si.si_signo = signum;
740  			si.si_errno = 0;
741  		        si.si_code  = reason;
742  			/*
743  			 * Posix definies POLL_IN and friends to be signal
744  			 * specific si_codes for SIG_POLL.  Linux extended
745  			 * these si_codes to other signals in a way that is
746  			 * ambiguous if other signals also have signal
747  			 * specific si_codes.  In that case use SI_SIGIO instead
748  			 * to remove the ambiguity.
749  			 */
750  			if ((signum != SIGPOLL) && sig_specific_sicodes(signum))
751  				si.si_code = SI_SIGIO;
752  
753  			/* Make sure we are called with one of the POLL_*
754  			   reasons, otherwise we could leak kernel stack into
755  			   userspace.  */
756  			BUG_ON((reason < POLL_IN) || ((reason - POLL_IN) >= NSIGPOLL));
757  			if (reason - POLL_IN >= NSIGPOLL)
758  				si.si_band  = ~0L;
759  			else
760  				si.si_band = mangle_poll(band_table[reason - POLL_IN]);
761  			si.si_fd    = fd;
762  			if (!do_send_sig_info(signum, &si, p, type))
763  				break;
764  		}
765  			fallthrough;	/* fall back on the old plain SIGIO signal */
766  		case 0:
767  			do_send_sig_info(SIGIO, SEND_SIG_PRIV, p, type);
768  	}
769  }
770  
771  void send_sigio(struct fown_struct *fown, int fd, int band)
772  {
773  	struct task_struct *p;
774  	enum pid_type type;
775  	unsigned long flags;
776  	struct pid *pid;
777  
778  	read_lock_irqsave(&fown->lock, flags);
779  
780  	type = fown->pid_type;
781  	pid = fown->pid;
782  	if (!pid)
783  		goto out_unlock_fown;
784  
785  	if (type <= PIDTYPE_TGID) {
786  		rcu_read_lock();
787  		p = pid_task(pid, PIDTYPE_PID);
788  		if (p)
789  			send_sigio_to_task(p, fown, fd, band, type);
790  		rcu_read_unlock();
791  	} else {
792  		read_lock(&tasklist_lock);
793  		do_each_pid_task(pid, type, p) {
794  			send_sigio_to_task(p, fown, fd, band, type);
795  		} while_each_pid_task(pid, type, p);
796  		read_unlock(&tasklist_lock);
797  	}
798   out_unlock_fown:
799  	read_unlock_irqrestore(&fown->lock, flags);
800  }
801  
802  static void send_sigurg_to_task(struct task_struct *p,
803  				struct fown_struct *fown, enum pid_type type)
804  {
805  	if (sigio_perm(p, fown, SIGURG))
806  		do_send_sig_info(SIGURG, SEND_SIG_PRIV, p, type);
807  }
808  
809  int send_sigurg(struct fown_struct *fown)
810  {
811  	struct task_struct *p;
812  	enum pid_type type;
813  	struct pid *pid;
814  	unsigned long flags;
815  	int ret = 0;
816  
817  	read_lock_irqsave(&fown->lock, flags);
818  
819  	type = fown->pid_type;
820  	pid = fown->pid;
821  	if (!pid)
822  		goto out_unlock_fown;
823  
824  	ret = 1;
825  
826  	if (type <= PIDTYPE_TGID) {
827  		rcu_read_lock();
828  		p = pid_task(pid, PIDTYPE_PID);
829  		if (p)
830  			send_sigurg_to_task(p, fown, type);
831  		rcu_read_unlock();
832  	} else {
833  		read_lock(&tasklist_lock);
834  		do_each_pid_task(pid, type, p) {
835  			send_sigurg_to_task(p, fown, type);
836  		} while_each_pid_task(pid, type, p);
837  		read_unlock(&tasklist_lock);
838  	}
839   out_unlock_fown:
840  	read_unlock_irqrestore(&fown->lock, flags);
841  	return ret;
842  }
843  
844  static DEFINE_SPINLOCK(fasync_lock);
845  static struct kmem_cache *fasync_cache __read_mostly;
846  
847  static void fasync_free_rcu(struct rcu_head *head)
848  {
849  	kmem_cache_free(fasync_cache,
850  			container_of(head, struct fasync_struct, fa_rcu));
851  }
852  
853  /*
854   * Remove a fasync entry. If successfully removed, return
855   * positive and clear the FASYNC flag. If no entry exists,
856   * do nothing and return 0.
857   *
858   * NOTE! It is very important that the FASYNC flag always
859   * match the state "is the filp on a fasync list".
860   *
861   */
862  int fasync_remove_entry(struct file *filp, struct fasync_struct **fapp)
863  {
864  	struct fasync_struct *fa, **fp;
865  	int result = 0;
866  
867  	spin_lock(&filp->f_lock);
868  	spin_lock(&fasync_lock);
869  	for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
870  		if (fa->fa_file != filp)
871  			continue;
872  
873  		write_lock_irq(&fa->fa_lock);
874  		fa->fa_file = NULL;
875  		write_unlock_irq(&fa->fa_lock);
876  
877  		*fp = fa->fa_next;
878  		call_rcu(&fa->fa_rcu, fasync_free_rcu);
879  		filp->f_flags &= ~FASYNC;
880  		result = 1;
881  		break;
882  	}
883  	spin_unlock(&fasync_lock);
884  	spin_unlock(&filp->f_lock);
885  	return result;
886  }
887  
888  struct fasync_struct *fasync_alloc(void)
889  {
890  	return kmem_cache_alloc(fasync_cache, GFP_KERNEL);
891  }
892  
893  /*
894   * NOTE! This can be used only for unused fasync entries:
895   * entries that actually got inserted on the fasync list
896   * need to be released by rcu - see fasync_remove_entry.
897   */
898  void fasync_free(struct fasync_struct *new)
899  {
900  	kmem_cache_free(fasync_cache, new);
901  }
902  
903  /*
904   * Insert a new entry into the fasync list.  Return the pointer to the
905   * old one if we didn't use the new one.
906   *
907   * NOTE! It is very important that the FASYNC flag always
908   * match the state "is the filp on a fasync list".
909   */
910  struct fasync_struct *fasync_insert_entry(int fd, struct file *filp, struct fasync_struct **fapp, struct fasync_struct *new)
911  {
912          struct fasync_struct *fa, **fp;
913  
914  	spin_lock(&filp->f_lock);
915  	spin_lock(&fasync_lock);
916  	for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
917  		if (fa->fa_file != filp)
918  			continue;
919  
920  		write_lock_irq(&fa->fa_lock);
921  		fa->fa_fd = fd;
922  		write_unlock_irq(&fa->fa_lock);
923  		goto out;
924  	}
925  
926  	rwlock_init(&new->fa_lock);
927  	new->magic = FASYNC_MAGIC;
928  	new->fa_file = filp;
929  	new->fa_fd = fd;
930  	new->fa_next = *fapp;
931  	rcu_assign_pointer(*fapp, new);
932  	filp->f_flags |= FASYNC;
933  
934  out:
935  	spin_unlock(&fasync_lock);
936  	spin_unlock(&filp->f_lock);
937  	return fa;
938  }
939  
940  /*
941   * Add a fasync entry. Return negative on error, positive if
942   * added, and zero if did nothing but change an existing one.
943   */
944  static int fasync_add_entry(int fd, struct file *filp, struct fasync_struct **fapp)
945  {
946  	struct fasync_struct *new;
947  
948  	new = fasync_alloc();
949  	if (!new)
950  		return -ENOMEM;
951  
952  	/*
953  	 * fasync_insert_entry() returns the old (update) entry if
954  	 * it existed.
955  	 *
956  	 * So free the (unused) new entry and return 0 to let the
957  	 * caller know that we didn't add any new fasync entries.
958  	 */
959  	if (fasync_insert_entry(fd, filp, fapp, new)) {
960  		fasync_free(new);
961  		return 0;
962  	}
963  
964  	return 1;
965  }
966  
967  /*
968   * fasync_helper() is used by almost all character device drivers
969   * to set up the fasync queue, and for regular files by the file
970   * lease code. It returns negative on error, 0 if it did no changes
971   * and positive if it added/deleted the entry.
972   */
973  int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
974  {
975  	if (!on)
976  		return fasync_remove_entry(filp, fapp);
977  	return fasync_add_entry(fd, filp, fapp);
978  }
979  
980  EXPORT_SYMBOL(fasync_helper);
981  
982  /*
983   * rcu_read_lock() is held
984   */
985  static void kill_fasync_rcu(struct fasync_struct *fa, int sig, int band)
986  {
987  	while (fa) {
988  		struct fown_struct *fown;
989  		unsigned long flags;
990  
991  		if (fa->magic != FASYNC_MAGIC) {
992  			printk(KERN_ERR "kill_fasync: bad magic number in "
993  			       "fasync_struct!\n");
994  			return;
995  		}
996  		read_lock_irqsave(&fa->fa_lock, flags);
997  		if (fa->fa_file) {
998  			fown = &fa->fa_file->f_owner;
999  			/* Don't send SIGURG to processes which have not set a
1000  			   queued signum: SIGURG has its own default signalling
1001  			   mechanism. */
1002  			if (!(sig == SIGURG && fown->signum == 0))
1003  				send_sigio(fown, fa->fa_fd, band);
1004  		}
1005  		read_unlock_irqrestore(&fa->fa_lock, flags);
1006  		fa = rcu_dereference(fa->fa_next);
1007  	}
1008  }
1009  
1010  void kill_fasync(struct fasync_struct **fp, int sig, int band)
1011  {
1012  	/* First a quick test without locking: usually
1013  	 * the list is empty.
1014  	 */
1015  	if (*fp) {
1016  		rcu_read_lock();
1017  		kill_fasync_rcu(rcu_dereference(*fp), sig, band);
1018  		rcu_read_unlock();
1019  	}
1020  }
1021  EXPORT_SYMBOL(kill_fasync);
1022  
1023  static int __init fcntl_init(void)
1024  {
1025  	/*
1026  	 * Please add new bits here to ensure allocation uniqueness.
1027  	 * Exceptions: O_NONBLOCK is a two bit define on parisc; O_NDELAY
1028  	 * is defined as O_NONBLOCK on some platforms and not on others.
1029  	 */
1030  	BUILD_BUG_ON(21 - 1 /* for O_RDONLY being 0 */ !=
1031  		HWEIGHT32(
1032  			(VALID_OPEN_FLAGS & ~(O_NONBLOCK | O_NDELAY)) |
1033  			__FMODE_EXEC | __FMODE_NONOTIFY));
1034  
1035  	fasync_cache = kmem_cache_create("fasync_cache",
1036  					 sizeof(struct fasync_struct), 0,
1037  					 SLAB_PANIC | SLAB_ACCOUNT, NULL);
1038  	return 0;
1039  }
1040  
1041  module_init(fcntl_init)
1042