Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * linux/drivers/char/mem.c
4 *
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 *
7 * Added devfs support.
8 * Jan-11-1998, C. Scott Ananian <cananian@alumni.princeton.edu>
9 * Shared /dev/zero mmapping support, Feb 2000, Kanoj Sarcar <kanoj@sgi.com>
10 */
11
12#include <linux/mm.h>
13#include <linux/miscdevice.h>
14#include <linux/slab.h>
15#include <linux/vmalloc.h>
16#include <linux/mman.h>
17#include <linux/random.h>
18#include <linux/init.h>
19#include <linux/tty.h>
20#include <linux/capability.h>
21#include <linux/ptrace.h>
22#include <linux/device.h>
23#include <linux/highmem.h>
24#include <linux/backing-dev.h>
25#include <linux/shmem_fs.h>
26#include <linux/splice.h>
27#include <linux/pfn.h>
28#include <linux/export.h>
29#include <linux/io.h>
30#include <linux/uio.h>
31#include <linux/uaccess.h>
32#include <linux/security.h>
33
34#ifdef CONFIG_IA64
35# include <linux/efi.h>
36#endif
37
38#define DEVMEM_MINOR 1
39#define DEVPORT_MINOR 4
40
41static inline unsigned long size_inside_page(unsigned long start,
42 unsigned long size)
43{
44 unsigned long sz;
45
46 sz = PAGE_SIZE - (start & (PAGE_SIZE - 1));
47
48 return min(sz, size);
49}
50
51#ifndef ARCH_HAS_VALID_PHYS_ADDR_RANGE
52static inline int valid_phys_addr_range(phys_addr_t addr, size_t count)
53{
54 return addr + count <= __pa(high_memory);
55}
56
57static inline int valid_mmap_phys_addr_range(unsigned long pfn, size_t size)
58{
59 return 1;
60}
61#endif
62
63#ifdef CONFIG_STRICT_DEVMEM
64static inline int page_is_allowed(unsigned long pfn)
65{
66 return devmem_is_allowed(pfn);
67}
68static inline int range_is_allowed(unsigned long pfn, unsigned long size)
69{
70 u64 from = ((u64)pfn) << PAGE_SHIFT;
71 u64 to = from + size;
72 u64 cursor = from;
73
74 while (cursor < to) {
75 if (!devmem_is_allowed(pfn))
76 return 0;
77 cursor += PAGE_SIZE;
78 pfn++;
79 }
80 return 1;
81}
82#else
83static inline int page_is_allowed(unsigned long pfn)
84{
85 return 1;
86}
87static inline int range_is_allowed(unsigned long pfn, unsigned long size)
88{
89 return 1;
90}
91#endif
92
93#ifndef unxlate_dev_mem_ptr
94#define unxlate_dev_mem_ptr unxlate_dev_mem_ptr
95void __weak unxlate_dev_mem_ptr(phys_addr_t phys, void *addr)
96{
97}
98#endif
99
100static inline bool should_stop_iteration(void)
101{
102 if (need_resched())
103 cond_resched();
104 return signal_pending(current);
105}
106
107/*
108 * This funcion reads the *physical* memory. The f_pos points directly to the
109 * memory location.
110 */
111static ssize_t read_mem(struct file *file, char __user *buf,
112 size_t count, loff_t *ppos)
113{
114 phys_addr_t p = *ppos;
115 ssize_t read, sz;
116 void *ptr;
117 char *bounce;
118 int err;
119
120 if (p != *ppos)
121 return 0;
122
123 if (!valid_phys_addr_range(p, count))
124 return -EFAULT;
125 read = 0;
126#ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
127 /* we don't have page 0 mapped on sparc and m68k.. */
128 if (p < PAGE_SIZE) {
129 sz = size_inside_page(p, count);
130 if (sz > 0) {
131 if (clear_user(buf, sz))
132 return -EFAULT;
133 buf += sz;
134 p += sz;
135 count -= sz;
136 read += sz;
137 }
138 }
139#endif
140
141 bounce = kmalloc(PAGE_SIZE, GFP_KERNEL);
142 if (!bounce)
143 return -ENOMEM;
144
145 while (count > 0) {
146 unsigned long remaining;
147 int allowed, probe;
148
149 sz = size_inside_page(p, count);
150
151 err = -EPERM;
152 allowed = page_is_allowed(p >> PAGE_SHIFT);
153 if (!allowed)
154 goto failed;
155
156 err = -EFAULT;
157 if (allowed == 2) {
158 /* Show zeros for restricted memory. */
159 remaining = clear_user(buf, sz);
160 } else {
161 /*
162 * On ia64 if a page has been mapped somewhere as
163 * uncached, then it must also be accessed uncached
164 * by the kernel or data corruption may occur.
165 */
166 ptr = xlate_dev_mem_ptr(p);
167 if (!ptr)
168 goto failed;
169
170 probe = copy_from_kernel_nofault(bounce, ptr, sz);
171 unxlate_dev_mem_ptr(p, ptr);
172 if (probe)
173 goto failed;
174
175 remaining = copy_to_user(buf, bounce, sz);
176 }
177
178 if (remaining)
179 goto failed;
180
181 buf += sz;
182 p += sz;
183 count -= sz;
184 read += sz;
185 if (should_stop_iteration())
186 break;
187 }
188 kfree(bounce);
189
190 *ppos += read;
191 return read;
192
193failed:
194 kfree(bounce);
195 return err;
196}
197
198static ssize_t write_mem(struct file *file, const char __user *buf,
199 size_t count, loff_t *ppos)
200{
201 phys_addr_t p = *ppos;
202 ssize_t written, sz;
203 unsigned long copied;
204 void *ptr;
205
206 if (p != *ppos)
207 return -EFBIG;
208
209 if (!valid_phys_addr_range(p, count))
210 return -EFAULT;
211
212 written = 0;
213
214#ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
215 /* we don't have page 0 mapped on sparc and m68k.. */
216 if (p < PAGE_SIZE) {
217 sz = size_inside_page(p, count);
218 /* Hmm. Do something? */
219 buf += sz;
220 p += sz;
221 count -= sz;
222 written += sz;
223 }
224#endif
225
226 while (count > 0) {
227 int allowed;
228
229 sz = size_inside_page(p, count);
230
231 allowed = page_is_allowed(p >> PAGE_SHIFT);
232 if (!allowed)
233 return -EPERM;
234
235 /* Skip actual writing when a page is marked as restricted. */
236 if (allowed == 1) {
237 /*
238 * On ia64 if a page has been mapped somewhere as
239 * uncached, then it must also be accessed uncached
240 * by the kernel or data corruption may occur.
241 */
242 ptr = xlate_dev_mem_ptr(p);
243 if (!ptr) {
244 if (written)
245 break;
246 return -EFAULT;
247 }
248
249 copied = copy_from_user(ptr, buf, sz);
250 unxlate_dev_mem_ptr(p, ptr);
251 if (copied) {
252 written += sz - copied;
253 if (written)
254 break;
255 return -EFAULT;
256 }
257 }
258
259 buf += sz;
260 p += sz;
261 count -= sz;
262 written += sz;
263 if (should_stop_iteration())
264 break;
265 }
266
267 *ppos += written;
268 return written;
269}
270
271int __weak phys_mem_access_prot_allowed(struct file *file,
272 unsigned long pfn, unsigned long size, pgprot_t *vma_prot)
273{
274 return 1;
275}
276
277#ifndef __HAVE_PHYS_MEM_ACCESS_PROT
278
279/*
280 * Architectures vary in how they handle caching for addresses
281 * outside of main memory.
282 *
283 */
284#ifdef pgprot_noncached
285static int uncached_access(struct file *file, phys_addr_t addr)
286{
287#if defined(CONFIG_IA64)
288 /*
289 * On ia64, we ignore O_DSYNC because we cannot tolerate memory
290 * attribute aliases.
291 */
292 return !(efi_mem_attributes(addr) & EFI_MEMORY_WB);
293#else
294 /*
295 * Accessing memory above the top the kernel knows about or through a
296 * file pointer
297 * that was marked O_DSYNC will be done non-cached.
298 */
299 if (file->f_flags & O_DSYNC)
300 return 1;
301 return addr >= __pa(high_memory);
302#endif
303}
304#endif
305
306static pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
307 unsigned long size, pgprot_t vma_prot)
308{
309#ifdef pgprot_noncached
310 phys_addr_t offset = pfn << PAGE_SHIFT;
311
312 if (uncached_access(file, offset))
313 return pgprot_noncached(vma_prot);
314#endif
315 return vma_prot;
316}
317#endif
318
319#ifndef CONFIG_MMU
320static unsigned long get_unmapped_area_mem(struct file *file,
321 unsigned long addr,
322 unsigned long len,
323 unsigned long pgoff,
324 unsigned long flags)
325{
326 if (!valid_mmap_phys_addr_range(pgoff, len))
327 return (unsigned long) -EINVAL;
328 return pgoff << PAGE_SHIFT;
329}
330
331/* permit direct mmap, for read, write or exec */
332static unsigned memory_mmap_capabilities(struct file *file)
333{
334 return NOMMU_MAP_DIRECT |
335 NOMMU_MAP_READ | NOMMU_MAP_WRITE | NOMMU_MAP_EXEC;
336}
337
338static unsigned zero_mmap_capabilities(struct file *file)
339{
340 return NOMMU_MAP_COPY;
341}
342
343/* can't do an in-place private mapping if there's no MMU */
344static inline int private_mapping_ok(struct vm_area_struct *vma)
345{
346 return vma->vm_flags & VM_MAYSHARE;
347}
348#else
349
350static inline int private_mapping_ok(struct vm_area_struct *vma)
351{
352 return 1;
353}
354#endif
355
356static const struct vm_operations_struct mmap_mem_ops = {
357#ifdef CONFIG_HAVE_IOREMAP_PROT
358 .access = generic_access_phys
359#endif
360};
361
362static int mmap_mem(struct file *file, struct vm_area_struct *vma)
363{
364 size_t size = vma->vm_end - vma->vm_start;
365 phys_addr_t offset = (phys_addr_t)vma->vm_pgoff << PAGE_SHIFT;
366
367 /* Does it even fit in phys_addr_t? */
368 if (offset >> PAGE_SHIFT != vma->vm_pgoff)
369 return -EINVAL;
370
371 /* It's illegal to wrap around the end of the physical address space. */
372 if (offset + (phys_addr_t)size - 1 < offset)
373 return -EINVAL;
374
375 if (!valid_mmap_phys_addr_range(vma->vm_pgoff, size))
376 return -EINVAL;
377
378 if (!private_mapping_ok(vma))
379 return -ENOSYS;
380
381 if (!range_is_allowed(vma->vm_pgoff, size))
382 return -EPERM;
383
384 if (!phys_mem_access_prot_allowed(file, vma->vm_pgoff, size,
385 &vma->vm_page_prot))
386 return -EINVAL;
387
388 vma->vm_page_prot = phys_mem_access_prot(file, vma->vm_pgoff,
389 size,
390 vma->vm_page_prot);
391
392 vma->vm_ops = &mmap_mem_ops;
393
394 /* Remap-pfn-range will mark the range VM_IO */
395 if (remap_pfn_range(vma,
396 vma->vm_start,
397 vma->vm_pgoff,
398 size,
399 vma->vm_page_prot)) {
400 return -EAGAIN;
401 }
402 return 0;
403}
404
405static ssize_t read_port(struct file *file, char __user *buf,
406 size_t count, loff_t *ppos)
407{
408 unsigned long i = *ppos;
409 char __user *tmp = buf;
410
411 if (!access_ok(buf, count))
412 return -EFAULT;
413 while (count-- > 0 && i < 65536) {
414 if (__put_user(inb(i), tmp) < 0)
415 return -EFAULT;
416 i++;
417 tmp++;
418 }
419 *ppos = i;
420 return tmp-buf;
421}
422
423static ssize_t write_port(struct file *file, const char __user *buf,
424 size_t count, loff_t *ppos)
425{
426 unsigned long i = *ppos;
427 const char __user *tmp = buf;
428
429 if (!access_ok(buf, count))
430 return -EFAULT;
431 while (count-- > 0 && i < 65536) {
432 char c;
433
434 if (__get_user(c, tmp)) {
435 if (tmp > buf)
436 break;
437 return -EFAULT;
438 }
439 outb(c, i);
440 i++;
441 tmp++;
442 }
443 *ppos = i;
444 return tmp-buf;
445}
446
447static ssize_t read_null(struct file *file, char __user *buf,
448 size_t count, loff_t *ppos)
449{
450 return 0;
451}
452
453static ssize_t write_null(struct file *file, const char __user *buf,
454 size_t count, loff_t *ppos)
455{
456 return count;
457}
458
459static ssize_t read_iter_null(struct kiocb *iocb, struct iov_iter *to)
460{
461 return 0;
462}
463
464static ssize_t write_iter_null(struct kiocb *iocb, struct iov_iter *from)
465{
466 size_t count = iov_iter_count(from);
467 iov_iter_advance(from, count);
468 return count;
469}
470
471static int pipe_to_null(struct pipe_inode_info *info, struct pipe_buffer *buf,
472 struct splice_desc *sd)
473{
474 return sd->len;
475}
476
477static ssize_t splice_write_null(struct pipe_inode_info *pipe, struct file *out,
478 loff_t *ppos, size_t len, unsigned int flags)
479{
480 return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_null);
481}
482
483static int uring_cmd_null(struct io_uring_cmd *ioucmd, unsigned int issue_flags)
484{
485 return 0;
486}
487
488static ssize_t read_iter_zero(struct kiocb *iocb, struct iov_iter *iter)
489{
490 size_t written = 0;
491
492 while (iov_iter_count(iter)) {
493 size_t chunk = iov_iter_count(iter), n;
494
495 if (chunk > PAGE_SIZE)
496 chunk = PAGE_SIZE; /* Just for latency reasons */
497 n = iov_iter_zero(chunk, iter);
498 if (!n && iov_iter_count(iter))
499 return written ? written : -EFAULT;
500 written += n;
501 if (signal_pending(current))
502 return written ? written : -ERESTARTSYS;
503 if (!need_resched())
504 continue;
505 if (iocb->ki_flags & IOCB_NOWAIT)
506 return written ? written : -EAGAIN;
507 cond_resched();
508 }
509 return written;
510}
511
512static ssize_t read_zero(struct file *file, char __user *buf,
513 size_t count, loff_t *ppos)
514{
515 size_t cleared = 0;
516
517 while (count) {
518 size_t chunk = min_t(size_t, count, PAGE_SIZE);
519 size_t left;
520
521 left = clear_user(buf + cleared, chunk);
522 if (unlikely(left)) {
523 cleared += (chunk - left);
524 if (!cleared)
525 return -EFAULT;
526 break;
527 }
528 cleared += chunk;
529 count -= chunk;
530
531 if (signal_pending(current))
532 break;
533 cond_resched();
534 }
535
536 return cleared;
537}
538
539static int mmap_zero(struct file *file, struct vm_area_struct *vma)
540{
541#ifndef CONFIG_MMU
542 return -ENOSYS;
543#endif
544 if (vma->vm_flags & VM_SHARED)
545 return shmem_zero_setup(vma);
546 vma_set_anonymous(vma);
547 return 0;
548}
549
550static unsigned long get_unmapped_area_zero(struct file *file,
551 unsigned long addr, unsigned long len,
552 unsigned long pgoff, unsigned long flags)
553{
554#ifdef CONFIG_MMU
555 if (flags & MAP_SHARED) {
556 /*
557 * mmap_zero() will call shmem_zero_setup() to create a file,
558 * so use shmem's get_unmapped_area in case it can be huge;
559 * and pass NULL for file as in mmap.c's get_unmapped_area(),
560 * so as not to confuse shmem with our handle on "/dev/zero".
561 */
562 return shmem_get_unmapped_area(NULL, addr, len, pgoff, flags);
563 }
564
565 /* Otherwise flags & MAP_PRIVATE: with no shmem object beneath it */
566 return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
567#else
568 return -ENOSYS;
569#endif
570}
571
572static ssize_t write_full(struct file *file, const char __user *buf,
573 size_t count, loff_t *ppos)
574{
575 return -ENOSPC;
576}
577
578/*
579 * Special lseek() function for /dev/null and /dev/zero. Most notably, you
580 * can fopen() both devices with "a" now. This was previously impossible.
581 * -- SRB.
582 */
583static loff_t null_lseek(struct file *file, loff_t offset, int orig)
584{
585 return file->f_pos = 0;
586}
587
588/*
589 * The memory devices use the full 32/64 bits of the offset, and so we cannot
590 * check against negative addresses: they are ok. The return value is weird,
591 * though, in that case (0).
592 *
593 * also note that seeking relative to the "end of file" isn't supported:
594 * it has no meaning, so it returns -EINVAL.
595 */
596static loff_t memory_lseek(struct file *file, loff_t offset, int orig)
597{
598 loff_t ret;
599
600 inode_lock(file_inode(file));
601 switch (orig) {
602 case SEEK_CUR:
603 offset += file->f_pos;
604 fallthrough;
605 case SEEK_SET:
606 /* to avoid userland mistaking f_pos=-9 as -EBADF=-9 */
607 if ((unsigned long long)offset >= -MAX_ERRNO) {
608 ret = -EOVERFLOW;
609 break;
610 }
611 file->f_pos = offset;
612 ret = file->f_pos;
613 force_successful_syscall_return();
614 break;
615 default:
616 ret = -EINVAL;
617 }
618 inode_unlock(file_inode(file));
619 return ret;
620}
621
622static int open_port(struct inode *inode, struct file *filp)
623{
624 int rc;
625
626 if (!capable(CAP_SYS_RAWIO))
627 return -EPERM;
628
629 rc = security_locked_down(LOCKDOWN_DEV_MEM);
630 if (rc)
631 return rc;
632
633 if (iminor(inode) != DEVMEM_MINOR)
634 return 0;
635
636 /*
637 * Use a unified address space to have a single point to manage
638 * revocations when drivers want to take over a /dev/mem mapped
639 * range.
640 */
641 filp->f_mapping = iomem_get_mapping();
642
643 return 0;
644}
645
646#define zero_lseek null_lseek
647#define full_lseek null_lseek
648#define write_zero write_null
649#define write_iter_zero write_iter_null
650#define open_mem open_port
651
652static const struct file_operations __maybe_unused mem_fops = {
653 .llseek = memory_lseek,
654 .read = read_mem,
655 .write = write_mem,
656 .mmap = mmap_mem,
657 .open = open_mem,
658#ifndef CONFIG_MMU
659 .get_unmapped_area = get_unmapped_area_mem,
660 .mmap_capabilities = memory_mmap_capabilities,
661#endif
662};
663
664static const struct file_operations null_fops = {
665 .llseek = null_lseek,
666 .read = read_null,
667 .write = write_null,
668 .read_iter = read_iter_null,
669 .write_iter = write_iter_null,
670 .splice_write = splice_write_null,
671 .uring_cmd = uring_cmd_null,
672};
673
674static const struct file_operations __maybe_unused port_fops = {
675 .llseek = memory_lseek,
676 .read = read_port,
677 .write = write_port,
678 .open = open_port,
679};
680
681static const struct file_operations zero_fops = {
682 .llseek = zero_lseek,
683 .write = write_zero,
684 .read_iter = read_iter_zero,
685 .read = read_zero,
686 .write_iter = write_iter_zero,
687 .mmap = mmap_zero,
688 .get_unmapped_area = get_unmapped_area_zero,
689#ifndef CONFIG_MMU
690 .mmap_capabilities = zero_mmap_capabilities,
691#endif
692};
693
694static const struct file_operations full_fops = {
695 .llseek = full_lseek,
696 .read_iter = read_iter_zero,
697 .write = write_full,
698};
699
700static const struct memdev {
701 const char *name;
702 umode_t mode;
703 const struct file_operations *fops;
704 fmode_t fmode;
705} devlist[] = {
706#ifdef CONFIG_DEVMEM
707 [DEVMEM_MINOR] = { "mem", 0, &mem_fops, FMODE_UNSIGNED_OFFSET },
708#endif
709 [3] = { "null", 0666, &null_fops, FMODE_NOWAIT },
710#ifdef CONFIG_DEVPORT
711 [4] = { "port", 0, &port_fops, 0 },
712#endif
713 [5] = { "zero", 0666, &zero_fops, FMODE_NOWAIT },
714 [7] = { "full", 0666, &full_fops, 0 },
715 [8] = { "random", 0666, &random_fops, FMODE_NOWAIT },
716 [9] = { "urandom", 0666, &urandom_fops, FMODE_NOWAIT },
717#ifdef CONFIG_PRINTK
718 [11] = { "kmsg", 0644, &kmsg_fops, 0 },
719#endif
720};
721
722static int memory_open(struct inode *inode, struct file *filp)
723{
724 int minor;
725 const struct memdev *dev;
726
727 minor = iminor(inode);
728 if (minor >= ARRAY_SIZE(devlist))
729 return -ENXIO;
730
731 dev = &devlist[minor];
732 if (!dev->fops)
733 return -ENXIO;
734
735 filp->f_op = dev->fops;
736 filp->f_mode |= dev->fmode;
737
738 if (dev->fops->open)
739 return dev->fops->open(inode, filp);
740
741 return 0;
742}
743
744static const struct file_operations memory_fops = {
745 .open = memory_open,
746 .llseek = noop_llseek,
747};
748
749static char *mem_devnode(const struct device *dev, umode_t *mode)
750{
751 if (mode && devlist[MINOR(dev->devt)].mode)
752 *mode = devlist[MINOR(dev->devt)].mode;
753 return NULL;
754}
755
756static struct class *mem_class;
757
758static int __init chr_dev_init(void)
759{
760 int minor;
761
762 if (register_chrdev(MEM_MAJOR, "mem", &memory_fops))
763 printk("unable to get major %d for memory devs\n", MEM_MAJOR);
764
765 mem_class = class_create(THIS_MODULE, "mem");
766 if (IS_ERR(mem_class))
767 return PTR_ERR(mem_class);
768
769 mem_class->devnode = mem_devnode;
770 for (minor = 1; minor < ARRAY_SIZE(devlist); minor++) {
771 if (!devlist[minor].name)
772 continue;
773
774 /*
775 * Create /dev/port?
776 */
777 if ((minor == DEVPORT_MINOR) && !arch_has_dev_port())
778 continue;
779
780 device_create(mem_class, NULL, MKDEV(MEM_MAJOR, minor),
781 NULL, devlist[minor].name);
782 }
783
784 return tty_init();
785}
786
787fs_initcall(chr_dev_init);
1/*
2 * linux/drivers/char/mem.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 *
6 * Added devfs support.
7 * Jan-11-1998, C. Scott Ananian <cananian@alumni.princeton.edu>
8 * Shared /dev/zero mmapping support, Feb 2000, Kanoj Sarcar <kanoj@sgi.com>
9 */
10
11#include <linux/mm.h>
12#include <linux/miscdevice.h>
13#include <linux/slab.h>
14#include <linux/vmalloc.h>
15#include <linux/mman.h>
16#include <linux/random.h>
17#include <linux/init.h>
18#include <linux/raw.h>
19#include <linux/tty.h>
20#include <linux/capability.h>
21#include <linux/ptrace.h>
22#include <linux/device.h>
23#include <linux/highmem.h>
24#include <linux/backing-dev.h>
25#include <linux/splice.h>
26#include <linux/pfn.h>
27#include <linux/export.h>
28#include <linux/io.h>
29#include <linux/aio.h>
30
31#include <asm/uaccess.h>
32
33#ifdef CONFIG_IA64
34# include <linux/efi.h>
35#endif
36
37#define DEVPORT_MINOR 4
38
39static inline unsigned long size_inside_page(unsigned long start,
40 unsigned long size)
41{
42 unsigned long sz;
43
44 sz = PAGE_SIZE - (start & (PAGE_SIZE - 1));
45
46 return min(sz, size);
47}
48
49#ifndef ARCH_HAS_VALID_PHYS_ADDR_RANGE
50static inline int valid_phys_addr_range(phys_addr_t addr, size_t count)
51{
52 return addr + count <= __pa(high_memory);
53}
54
55static inline int valid_mmap_phys_addr_range(unsigned long pfn, size_t size)
56{
57 return 1;
58}
59#endif
60
61#ifdef CONFIG_STRICT_DEVMEM
62static inline int range_is_allowed(unsigned long pfn, unsigned long size)
63{
64 u64 from = ((u64)pfn) << PAGE_SHIFT;
65 u64 to = from + size;
66 u64 cursor = from;
67
68 while (cursor < to) {
69 if (!devmem_is_allowed(pfn)) {
70 printk(KERN_INFO
71 "Program %s tried to access /dev/mem between %Lx->%Lx.\n",
72 current->comm, from, to);
73 return 0;
74 }
75 cursor += PAGE_SIZE;
76 pfn++;
77 }
78 return 1;
79}
80#else
81static inline int range_is_allowed(unsigned long pfn, unsigned long size)
82{
83 return 1;
84}
85#endif
86
87void __weak unxlate_dev_mem_ptr(unsigned long phys, void *addr)
88{
89}
90
91/*
92 * This funcion reads the *physical* memory. The f_pos points directly to the
93 * memory location.
94 */
95static ssize_t read_mem(struct file *file, char __user *buf,
96 size_t count, loff_t *ppos)
97{
98 phys_addr_t p = *ppos;
99 ssize_t read, sz;
100 char *ptr;
101
102 if (p != *ppos)
103 return 0;
104
105 if (!valid_phys_addr_range(p, count))
106 return -EFAULT;
107 read = 0;
108#ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
109 /* we don't have page 0 mapped on sparc and m68k.. */
110 if (p < PAGE_SIZE) {
111 sz = size_inside_page(p, count);
112 if (sz > 0) {
113 if (clear_user(buf, sz))
114 return -EFAULT;
115 buf += sz;
116 p += sz;
117 count -= sz;
118 read += sz;
119 }
120 }
121#endif
122
123 while (count > 0) {
124 unsigned long remaining;
125
126 sz = size_inside_page(p, count);
127
128 if (!range_is_allowed(p >> PAGE_SHIFT, count))
129 return -EPERM;
130
131 /*
132 * On ia64 if a page has been mapped somewhere as uncached, then
133 * it must also be accessed uncached by the kernel or data
134 * corruption may occur.
135 */
136 ptr = xlate_dev_mem_ptr(p);
137 if (!ptr)
138 return -EFAULT;
139
140 remaining = copy_to_user(buf, ptr, sz);
141 unxlate_dev_mem_ptr(p, ptr);
142 if (remaining)
143 return -EFAULT;
144
145 buf += sz;
146 p += sz;
147 count -= sz;
148 read += sz;
149 }
150
151 *ppos += read;
152 return read;
153}
154
155static ssize_t write_mem(struct file *file, const char __user *buf,
156 size_t count, loff_t *ppos)
157{
158 phys_addr_t p = *ppos;
159 ssize_t written, sz;
160 unsigned long copied;
161 void *ptr;
162
163 if (p != *ppos)
164 return -EFBIG;
165
166 if (!valid_phys_addr_range(p, count))
167 return -EFAULT;
168
169 written = 0;
170
171#ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
172 /* we don't have page 0 mapped on sparc and m68k.. */
173 if (p < PAGE_SIZE) {
174 sz = size_inside_page(p, count);
175 /* Hmm. Do something? */
176 buf += sz;
177 p += sz;
178 count -= sz;
179 written += sz;
180 }
181#endif
182
183 while (count > 0) {
184 sz = size_inside_page(p, count);
185
186 if (!range_is_allowed(p >> PAGE_SHIFT, sz))
187 return -EPERM;
188
189 /*
190 * On ia64 if a page has been mapped somewhere as uncached, then
191 * it must also be accessed uncached by the kernel or data
192 * corruption may occur.
193 */
194 ptr = xlate_dev_mem_ptr(p);
195 if (!ptr) {
196 if (written)
197 break;
198 return -EFAULT;
199 }
200
201 copied = copy_from_user(ptr, buf, sz);
202 unxlate_dev_mem_ptr(p, ptr);
203 if (copied) {
204 written += sz - copied;
205 if (written)
206 break;
207 return -EFAULT;
208 }
209
210 buf += sz;
211 p += sz;
212 count -= sz;
213 written += sz;
214 }
215
216 *ppos += written;
217 return written;
218}
219
220int __weak phys_mem_access_prot_allowed(struct file *file,
221 unsigned long pfn, unsigned long size, pgprot_t *vma_prot)
222{
223 return 1;
224}
225
226#ifndef __HAVE_PHYS_MEM_ACCESS_PROT
227
228/*
229 * Architectures vary in how they handle caching for addresses
230 * outside of main memory.
231 *
232 */
233#ifdef pgprot_noncached
234static int uncached_access(struct file *file, phys_addr_t addr)
235{
236#if defined(CONFIG_IA64)
237 /*
238 * On ia64, we ignore O_DSYNC because we cannot tolerate memory
239 * attribute aliases.
240 */
241 return !(efi_mem_attributes(addr) & EFI_MEMORY_WB);
242#elif defined(CONFIG_MIPS)
243 {
244 extern int __uncached_access(struct file *file,
245 unsigned long addr);
246
247 return __uncached_access(file, addr);
248 }
249#else
250 /*
251 * Accessing memory above the top the kernel knows about or through a
252 * file pointer
253 * that was marked O_DSYNC will be done non-cached.
254 */
255 if (file->f_flags & O_DSYNC)
256 return 1;
257 return addr >= __pa(high_memory);
258#endif
259}
260#endif
261
262static pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
263 unsigned long size, pgprot_t vma_prot)
264{
265#ifdef pgprot_noncached
266 phys_addr_t offset = pfn << PAGE_SHIFT;
267
268 if (uncached_access(file, offset))
269 return pgprot_noncached(vma_prot);
270#endif
271 return vma_prot;
272}
273#endif
274
275#ifndef CONFIG_MMU
276static unsigned long get_unmapped_area_mem(struct file *file,
277 unsigned long addr,
278 unsigned long len,
279 unsigned long pgoff,
280 unsigned long flags)
281{
282 if (!valid_mmap_phys_addr_range(pgoff, len))
283 return (unsigned long) -EINVAL;
284 return pgoff << PAGE_SHIFT;
285}
286
287/* can't do an in-place private mapping if there's no MMU */
288static inline int private_mapping_ok(struct vm_area_struct *vma)
289{
290 return vma->vm_flags & VM_MAYSHARE;
291}
292#else
293#define get_unmapped_area_mem NULL
294
295static inline int private_mapping_ok(struct vm_area_struct *vma)
296{
297 return 1;
298}
299#endif
300
301static const struct vm_operations_struct mmap_mem_ops = {
302#ifdef CONFIG_HAVE_IOREMAP_PROT
303 .access = generic_access_phys
304#endif
305};
306
307static int mmap_mem(struct file *file, struct vm_area_struct *vma)
308{
309 size_t size = vma->vm_end - vma->vm_start;
310
311 if (!valid_mmap_phys_addr_range(vma->vm_pgoff, size))
312 return -EINVAL;
313
314 if (!private_mapping_ok(vma))
315 return -ENOSYS;
316
317 if (!range_is_allowed(vma->vm_pgoff, size))
318 return -EPERM;
319
320 if (!phys_mem_access_prot_allowed(file, vma->vm_pgoff, size,
321 &vma->vm_page_prot))
322 return -EINVAL;
323
324 vma->vm_page_prot = phys_mem_access_prot(file, vma->vm_pgoff,
325 size,
326 vma->vm_page_prot);
327
328 vma->vm_ops = &mmap_mem_ops;
329
330 /* Remap-pfn-range will mark the range VM_IO */
331 if (remap_pfn_range(vma,
332 vma->vm_start,
333 vma->vm_pgoff,
334 size,
335 vma->vm_page_prot)) {
336 return -EAGAIN;
337 }
338 return 0;
339}
340
341#ifdef CONFIG_DEVKMEM
342static int mmap_kmem(struct file *file, struct vm_area_struct *vma)
343{
344 unsigned long pfn;
345
346 /* Turn a kernel-virtual address into a physical page frame */
347 pfn = __pa((u64)vma->vm_pgoff << PAGE_SHIFT) >> PAGE_SHIFT;
348
349 /*
350 * RED-PEN: on some architectures there is more mapped memory than
351 * available in mem_map which pfn_valid checks for. Perhaps should add a
352 * new macro here.
353 *
354 * RED-PEN: vmalloc is not supported right now.
355 */
356 if (!pfn_valid(pfn))
357 return -EIO;
358
359 vma->vm_pgoff = pfn;
360 return mmap_mem(file, vma);
361}
362#endif
363
364#ifdef CONFIG_DEVKMEM
365/*
366 * This function reads the *virtual* memory as seen by the kernel.
367 */
368static ssize_t read_kmem(struct file *file, char __user *buf,
369 size_t count, loff_t *ppos)
370{
371 unsigned long p = *ppos;
372 ssize_t low_count, read, sz;
373 char *kbuf; /* k-addr because vread() takes vmlist_lock rwlock */
374 int err = 0;
375
376 read = 0;
377 if (p < (unsigned long) high_memory) {
378 low_count = count;
379 if (count > (unsigned long)high_memory - p)
380 low_count = (unsigned long)high_memory - p;
381
382#ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
383 /* we don't have page 0 mapped on sparc and m68k.. */
384 if (p < PAGE_SIZE && low_count > 0) {
385 sz = size_inside_page(p, low_count);
386 if (clear_user(buf, sz))
387 return -EFAULT;
388 buf += sz;
389 p += sz;
390 read += sz;
391 low_count -= sz;
392 count -= sz;
393 }
394#endif
395 while (low_count > 0) {
396 sz = size_inside_page(p, low_count);
397
398 /*
399 * On ia64 if a page has been mapped somewhere as
400 * uncached, then it must also be accessed uncached
401 * by the kernel or data corruption may occur
402 */
403 kbuf = xlate_dev_kmem_ptr((char *)p);
404
405 if (copy_to_user(buf, kbuf, sz))
406 return -EFAULT;
407 buf += sz;
408 p += sz;
409 read += sz;
410 low_count -= sz;
411 count -= sz;
412 }
413 }
414
415 if (count > 0) {
416 kbuf = (char *)__get_free_page(GFP_KERNEL);
417 if (!kbuf)
418 return -ENOMEM;
419 while (count > 0) {
420 sz = size_inside_page(p, count);
421 if (!is_vmalloc_or_module_addr((void *)p)) {
422 err = -ENXIO;
423 break;
424 }
425 sz = vread(kbuf, (char *)p, sz);
426 if (!sz)
427 break;
428 if (copy_to_user(buf, kbuf, sz)) {
429 err = -EFAULT;
430 break;
431 }
432 count -= sz;
433 buf += sz;
434 read += sz;
435 p += sz;
436 }
437 free_page((unsigned long)kbuf);
438 }
439 *ppos = p;
440 return read ? read : err;
441}
442
443
444static ssize_t do_write_kmem(unsigned long p, const char __user *buf,
445 size_t count, loff_t *ppos)
446{
447 ssize_t written, sz;
448 unsigned long copied;
449
450 written = 0;
451#ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
452 /* we don't have page 0 mapped on sparc and m68k.. */
453 if (p < PAGE_SIZE) {
454 sz = size_inside_page(p, count);
455 /* Hmm. Do something? */
456 buf += sz;
457 p += sz;
458 count -= sz;
459 written += sz;
460 }
461#endif
462
463 while (count > 0) {
464 char *ptr;
465
466 sz = size_inside_page(p, count);
467
468 /*
469 * On ia64 if a page has been mapped somewhere as uncached, then
470 * it must also be accessed uncached by the kernel or data
471 * corruption may occur.
472 */
473 ptr = xlate_dev_kmem_ptr((char *)p);
474
475 copied = copy_from_user(ptr, buf, sz);
476 if (copied) {
477 written += sz - copied;
478 if (written)
479 break;
480 return -EFAULT;
481 }
482 buf += sz;
483 p += sz;
484 count -= sz;
485 written += sz;
486 }
487
488 *ppos += written;
489 return written;
490}
491
492/*
493 * This function writes to the *virtual* memory as seen by the kernel.
494 */
495static ssize_t write_kmem(struct file *file, const char __user *buf,
496 size_t count, loff_t *ppos)
497{
498 unsigned long p = *ppos;
499 ssize_t wrote = 0;
500 ssize_t virtr = 0;
501 char *kbuf; /* k-addr because vwrite() takes vmlist_lock rwlock */
502 int err = 0;
503
504 if (p < (unsigned long) high_memory) {
505 unsigned long to_write = min_t(unsigned long, count,
506 (unsigned long)high_memory - p);
507 wrote = do_write_kmem(p, buf, to_write, ppos);
508 if (wrote != to_write)
509 return wrote;
510 p += wrote;
511 buf += wrote;
512 count -= wrote;
513 }
514
515 if (count > 0) {
516 kbuf = (char *)__get_free_page(GFP_KERNEL);
517 if (!kbuf)
518 return wrote ? wrote : -ENOMEM;
519 while (count > 0) {
520 unsigned long sz = size_inside_page(p, count);
521 unsigned long n;
522
523 if (!is_vmalloc_or_module_addr((void *)p)) {
524 err = -ENXIO;
525 break;
526 }
527 n = copy_from_user(kbuf, buf, sz);
528 if (n) {
529 err = -EFAULT;
530 break;
531 }
532 vwrite(kbuf, (char *)p, sz);
533 count -= sz;
534 buf += sz;
535 virtr += sz;
536 p += sz;
537 }
538 free_page((unsigned long)kbuf);
539 }
540
541 *ppos = p;
542 return virtr + wrote ? : err;
543}
544#endif
545
546#ifdef CONFIG_DEVPORT
547static ssize_t read_port(struct file *file, char __user *buf,
548 size_t count, loff_t *ppos)
549{
550 unsigned long i = *ppos;
551 char __user *tmp = buf;
552
553 if (!access_ok(VERIFY_WRITE, buf, count))
554 return -EFAULT;
555 while (count-- > 0 && i < 65536) {
556 if (__put_user(inb(i), tmp) < 0)
557 return -EFAULT;
558 i++;
559 tmp++;
560 }
561 *ppos = i;
562 return tmp-buf;
563}
564
565static ssize_t write_port(struct file *file, const char __user *buf,
566 size_t count, loff_t *ppos)
567{
568 unsigned long i = *ppos;
569 const char __user *tmp = buf;
570
571 if (!access_ok(VERIFY_READ, buf, count))
572 return -EFAULT;
573 while (count-- > 0 && i < 65536) {
574 char c;
575 if (__get_user(c, tmp)) {
576 if (tmp > buf)
577 break;
578 return -EFAULT;
579 }
580 outb(c, i);
581 i++;
582 tmp++;
583 }
584 *ppos = i;
585 return tmp-buf;
586}
587#endif
588
589static ssize_t read_null(struct file *file, char __user *buf,
590 size_t count, loff_t *ppos)
591{
592 return 0;
593}
594
595static ssize_t write_null(struct file *file, const char __user *buf,
596 size_t count, loff_t *ppos)
597{
598 return count;
599}
600
601static ssize_t aio_read_null(struct kiocb *iocb, const struct iovec *iov,
602 unsigned long nr_segs, loff_t pos)
603{
604 return 0;
605}
606
607static ssize_t aio_write_null(struct kiocb *iocb, const struct iovec *iov,
608 unsigned long nr_segs, loff_t pos)
609{
610 return iov_length(iov, nr_segs);
611}
612
613static int pipe_to_null(struct pipe_inode_info *info, struct pipe_buffer *buf,
614 struct splice_desc *sd)
615{
616 return sd->len;
617}
618
619static ssize_t splice_write_null(struct pipe_inode_info *pipe, struct file *out,
620 loff_t *ppos, size_t len, unsigned int flags)
621{
622 return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_null);
623}
624
625static ssize_t read_zero(struct file *file, char __user *buf,
626 size_t count, loff_t *ppos)
627{
628 size_t written;
629
630 if (!count)
631 return 0;
632
633 if (!access_ok(VERIFY_WRITE, buf, count))
634 return -EFAULT;
635
636 written = 0;
637 while (count) {
638 unsigned long unwritten;
639 size_t chunk = count;
640
641 if (chunk > PAGE_SIZE)
642 chunk = PAGE_SIZE; /* Just for latency reasons */
643 unwritten = __clear_user(buf, chunk);
644 written += chunk - unwritten;
645 if (unwritten)
646 break;
647 if (signal_pending(current))
648 return written ? written : -ERESTARTSYS;
649 buf += chunk;
650 count -= chunk;
651 cond_resched();
652 }
653 return written ? written : -EFAULT;
654}
655
656static ssize_t aio_read_zero(struct kiocb *iocb, const struct iovec *iov,
657 unsigned long nr_segs, loff_t pos)
658{
659 size_t written = 0;
660 unsigned long i;
661 ssize_t ret;
662
663 for (i = 0; i < nr_segs; i++) {
664 ret = read_zero(iocb->ki_filp, iov[i].iov_base, iov[i].iov_len,
665 &pos);
666 if (ret < 0)
667 break;
668 written += ret;
669 }
670
671 return written ? written : -EFAULT;
672}
673
674static int mmap_zero(struct file *file, struct vm_area_struct *vma)
675{
676#ifndef CONFIG_MMU
677 return -ENOSYS;
678#endif
679 if (vma->vm_flags & VM_SHARED)
680 return shmem_zero_setup(vma);
681 return 0;
682}
683
684static ssize_t write_full(struct file *file, const char __user *buf,
685 size_t count, loff_t *ppos)
686{
687 return -ENOSPC;
688}
689
690/*
691 * Special lseek() function for /dev/null and /dev/zero. Most notably, you
692 * can fopen() both devices with "a" now. This was previously impossible.
693 * -- SRB.
694 */
695static loff_t null_lseek(struct file *file, loff_t offset, int orig)
696{
697 return file->f_pos = 0;
698}
699
700/*
701 * The memory devices use the full 32/64 bits of the offset, and so we cannot
702 * check against negative addresses: they are ok. The return value is weird,
703 * though, in that case (0).
704 *
705 * also note that seeking relative to the "end of file" isn't supported:
706 * it has no meaning, so it returns -EINVAL.
707 */
708static loff_t memory_lseek(struct file *file, loff_t offset, int orig)
709{
710 loff_t ret;
711
712 mutex_lock(&file_inode(file)->i_mutex);
713 switch (orig) {
714 case SEEK_CUR:
715 offset += file->f_pos;
716 case SEEK_SET:
717 /* to avoid userland mistaking f_pos=-9 as -EBADF=-9 */
718 if (IS_ERR_VALUE((unsigned long long)offset)) {
719 ret = -EOVERFLOW;
720 break;
721 }
722 file->f_pos = offset;
723 ret = file->f_pos;
724 force_successful_syscall_return();
725 break;
726 default:
727 ret = -EINVAL;
728 }
729 mutex_unlock(&file_inode(file)->i_mutex);
730 return ret;
731}
732
733static int open_port(struct inode *inode, struct file *filp)
734{
735 return capable(CAP_SYS_RAWIO) ? 0 : -EPERM;
736}
737
738#define zero_lseek null_lseek
739#define full_lseek null_lseek
740#define write_zero write_null
741#define read_full read_zero
742#define aio_write_zero aio_write_null
743#define open_mem open_port
744#define open_kmem open_mem
745
746static const struct file_operations mem_fops = {
747 .llseek = memory_lseek,
748 .read = read_mem,
749 .write = write_mem,
750 .mmap = mmap_mem,
751 .open = open_mem,
752 .get_unmapped_area = get_unmapped_area_mem,
753};
754
755#ifdef CONFIG_DEVKMEM
756static const struct file_operations kmem_fops = {
757 .llseek = memory_lseek,
758 .read = read_kmem,
759 .write = write_kmem,
760 .mmap = mmap_kmem,
761 .open = open_kmem,
762 .get_unmapped_area = get_unmapped_area_mem,
763};
764#endif
765
766static const struct file_operations null_fops = {
767 .llseek = null_lseek,
768 .read = read_null,
769 .write = write_null,
770 .aio_read = aio_read_null,
771 .aio_write = aio_write_null,
772 .splice_write = splice_write_null,
773};
774
775#ifdef CONFIG_DEVPORT
776static const struct file_operations port_fops = {
777 .llseek = memory_lseek,
778 .read = read_port,
779 .write = write_port,
780 .open = open_port,
781};
782#endif
783
784static const struct file_operations zero_fops = {
785 .llseek = zero_lseek,
786 .read = read_zero,
787 .write = write_zero,
788 .aio_read = aio_read_zero,
789 .aio_write = aio_write_zero,
790 .mmap = mmap_zero,
791};
792
793/*
794 * capabilities for /dev/zero
795 * - permits private mappings, "copies" are taken of the source of zeros
796 * - no writeback happens
797 */
798static struct backing_dev_info zero_bdi = {
799 .name = "char/mem",
800 .capabilities = BDI_CAP_MAP_COPY | BDI_CAP_NO_ACCT_AND_WRITEBACK,
801};
802
803static const struct file_operations full_fops = {
804 .llseek = full_lseek,
805 .read = read_full,
806 .write = write_full,
807};
808
809static const struct memdev {
810 const char *name;
811 umode_t mode;
812 const struct file_operations *fops;
813 struct backing_dev_info *dev_info;
814} devlist[] = {
815 [1] = { "mem", 0, &mem_fops, &directly_mappable_cdev_bdi },
816#ifdef CONFIG_DEVKMEM
817 [2] = { "kmem", 0, &kmem_fops, &directly_mappable_cdev_bdi },
818#endif
819 [3] = { "null", 0666, &null_fops, NULL },
820#ifdef CONFIG_DEVPORT
821 [4] = { "port", 0, &port_fops, NULL },
822#endif
823 [5] = { "zero", 0666, &zero_fops, &zero_bdi },
824 [7] = { "full", 0666, &full_fops, NULL },
825 [8] = { "random", 0666, &random_fops, NULL },
826 [9] = { "urandom", 0666, &urandom_fops, NULL },
827#ifdef CONFIG_PRINTK
828 [11] = { "kmsg", 0644, &kmsg_fops, NULL },
829#endif
830};
831
832static int memory_open(struct inode *inode, struct file *filp)
833{
834 int minor;
835 const struct memdev *dev;
836
837 minor = iminor(inode);
838 if (minor >= ARRAY_SIZE(devlist))
839 return -ENXIO;
840
841 dev = &devlist[minor];
842 if (!dev->fops)
843 return -ENXIO;
844
845 filp->f_op = dev->fops;
846 if (dev->dev_info)
847 filp->f_mapping->backing_dev_info = dev->dev_info;
848
849 /* Is /dev/mem or /dev/kmem ? */
850 if (dev->dev_info == &directly_mappable_cdev_bdi)
851 filp->f_mode |= FMODE_UNSIGNED_OFFSET;
852
853 if (dev->fops->open)
854 return dev->fops->open(inode, filp);
855
856 return 0;
857}
858
859static const struct file_operations memory_fops = {
860 .open = memory_open,
861 .llseek = noop_llseek,
862};
863
864static char *mem_devnode(struct device *dev, umode_t *mode)
865{
866 if (mode && devlist[MINOR(dev->devt)].mode)
867 *mode = devlist[MINOR(dev->devt)].mode;
868 return NULL;
869}
870
871static struct class *mem_class;
872
873static int __init chr_dev_init(void)
874{
875 int minor;
876 int err;
877
878 err = bdi_init(&zero_bdi);
879 if (err)
880 return err;
881
882 if (register_chrdev(MEM_MAJOR, "mem", &memory_fops))
883 printk("unable to get major %d for memory devs\n", MEM_MAJOR);
884
885 mem_class = class_create(THIS_MODULE, "mem");
886 if (IS_ERR(mem_class))
887 return PTR_ERR(mem_class);
888
889 mem_class->devnode = mem_devnode;
890 for (minor = 1; minor < ARRAY_SIZE(devlist); minor++) {
891 if (!devlist[minor].name)
892 continue;
893
894 /*
895 * Create /dev/port?
896 */
897 if ((minor == DEVPORT_MINOR) && !arch_has_dev_port())
898 continue;
899
900 device_create(mem_class, NULL, MKDEV(MEM_MAJOR, minor),
901 NULL, devlist[minor].name);
902 }
903
904 return tty_init();
905}
906
907fs_initcall(chr_dev_init);