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#define DEVMEM_MINOR 1
35#define DEVPORT_MINOR 4
36
37static inline unsigned long size_inside_page(unsigned long start,
38 unsigned long size)
39{
40 unsigned long sz;
41
42 sz = PAGE_SIZE - (start & (PAGE_SIZE - 1));
43
44 return min(sz, size);
45}
46
47#ifndef ARCH_HAS_VALID_PHYS_ADDR_RANGE
48static inline int valid_phys_addr_range(phys_addr_t addr, size_t count)
49{
50 return addr + count <= __pa(high_memory);
51}
52
53static inline int valid_mmap_phys_addr_range(unsigned long pfn, size_t size)
54{
55 return 1;
56}
57#endif
58
59#ifdef CONFIG_STRICT_DEVMEM
60static inline int page_is_allowed(unsigned long pfn)
61{
62 return devmem_is_allowed(pfn);
63}
64static inline int range_is_allowed(unsigned long pfn, unsigned long size)
65{
66 u64 from = ((u64)pfn) << PAGE_SHIFT;
67 u64 to = from + size;
68 u64 cursor = from;
69
70 while (cursor < to) {
71 if (!devmem_is_allowed(pfn))
72 return 0;
73 cursor += PAGE_SIZE;
74 pfn++;
75 }
76 return 1;
77}
78#else
79static inline int page_is_allowed(unsigned long pfn)
80{
81 return 1;
82}
83static inline int range_is_allowed(unsigned long pfn, unsigned long size)
84{
85 return 1;
86}
87#endif
88
89static inline bool should_stop_iteration(void)
90{
91 if (need_resched())
92 cond_resched();
93 return signal_pending(current);
94}
95
96/*
97 * This funcion reads the *physical* memory. The f_pos points directly to the
98 * memory location.
99 */
100static ssize_t read_mem(struct file *file, char __user *buf,
101 size_t count, loff_t *ppos)
102{
103 phys_addr_t p = *ppos;
104 ssize_t read, sz;
105 void *ptr;
106 char *bounce;
107 int err;
108
109 if (p != *ppos)
110 return 0;
111
112 if (!valid_phys_addr_range(p, count))
113 return -EFAULT;
114 read = 0;
115#ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
116 /* we don't have page 0 mapped on sparc and m68k.. */
117 if (p < PAGE_SIZE) {
118 sz = size_inside_page(p, count);
119 if (sz > 0) {
120 if (clear_user(buf, sz))
121 return -EFAULT;
122 buf += sz;
123 p += sz;
124 count -= sz;
125 read += sz;
126 }
127 }
128#endif
129
130 bounce = kmalloc(PAGE_SIZE, GFP_KERNEL);
131 if (!bounce)
132 return -ENOMEM;
133
134 while (count > 0) {
135 unsigned long remaining;
136 int allowed, probe;
137
138 sz = size_inside_page(p, count);
139
140 err = -EPERM;
141 allowed = page_is_allowed(p >> PAGE_SHIFT);
142 if (!allowed)
143 goto failed;
144
145 err = -EFAULT;
146 if (allowed == 2) {
147 /* Show zeros for restricted memory. */
148 remaining = clear_user(buf, sz);
149 } else {
150 /*
151 * On ia64 if a page has been mapped somewhere as
152 * uncached, then it must also be accessed uncached
153 * by the kernel or data corruption may occur.
154 */
155 ptr = xlate_dev_mem_ptr(p);
156 if (!ptr)
157 goto failed;
158
159 probe = copy_from_kernel_nofault(bounce, ptr, sz);
160 unxlate_dev_mem_ptr(p, ptr);
161 if (probe)
162 goto failed;
163
164 remaining = copy_to_user(buf, bounce, sz);
165 }
166
167 if (remaining)
168 goto failed;
169
170 buf += sz;
171 p += sz;
172 count -= sz;
173 read += sz;
174 if (should_stop_iteration())
175 break;
176 }
177 kfree(bounce);
178
179 *ppos += read;
180 return read;
181
182failed:
183 kfree(bounce);
184 return err;
185}
186
187static ssize_t write_mem(struct file *file, const char __user *buf,
188 size_t count, loff_t *ppos)
189{
190 phys_addr_t p = *ppos;
191 ssize_t written, sz;
192 unsigned long copied;
193 void *ptr;
194
195 if (p != *ppos)
196 return -EFBIG;
197
198 if (!valid_phys_addr_range(p, count))
199 return -EFAULT;
200
201 written = 0;
202
203#ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
204 /* we don't have page 0 mapped on sparc and m68k.. */
205 if (p < PAGE_SIZE) {
206 sz = size_inside_page(p, count);
207 /* Hmm. Do something? */
208 buf += sz;
209 p += sz;
210 count -= sz;
211 written += sz;
212 }
213#endif
214
215 while (count > 0) {
216 int allowed;
217
218 sz = size_inside_page(p, count);
219
220 allowed = page_is_allowed(p >> PAGE_SHIFT);
221 if (!allowed)
222 return -EPERM;
223
224 /* Skip actual writing when a page is marked as restricted. */
225 if (allowed == 1) {
226 /*
227 * On ia64 if a page has been mapped somewhere as
228 * uncached, then it must also be accessed uncached
229 * by the kernel or data corruption may occur.
230 */
231 ptr = xlate_dev_mem_ptr(p);
232 if (!ptr) {
233 if (written)
234 break;
235 return -EFAULT;
236 }
237
238 copied = copy_from_user(ptr, buf, sz);
239 unxlate_dev_mem_ptr(p, ptr);
240 if (copied) {
241 written += sz - copied;
242 if (written)
243 break;
244 return -EFAULT;
245 }
246 }
247
248 buf += sz;
249 p += sz;
250 count -= sz;
251 written += sz;
252 if (should_stop_iteration())
253 break;
254 }
255
256 *ppos += written;
257 return written;
258}
259
260int __weak phys_mem_access_prot_allowed(struct file *file,
261 unsigned long pfn, unsigned long size, pgprot_t *vma_prot)
262{
263 return 1;
264}
265
266#ifndef __HAVE_PHYS_MEM_ACCESS_PROT
267
268/*
269 * Architectures vary in how they handle caching for addresses
270 * outside of main memory.
271 *
272 */
273#ifdef pgprot_noncached
274static int uncached_access(struct file *file, phys_addr_t addr)
275{
276 /*
277 * Accessing memory above the top the kernel knows about or through a
278 * file pointer
279 * that was marked O_DSYNC will be done non-cached.
280 */
281 if (file->f_flags & O_DSYNC)
282 return 1;
283 return addr >= __pa(high_memory);
284}
285#endif
286
287static pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
288 unsigned long size, pgprot_t vma_prot)
289{
290#ifdef pgprot_noncached
291 phys_addr_t offset = pfn << PAGE_SHIFT;
292
293 if (uncached_access(file, offset))
294 return pgprot_noncached(vma_prot);
295#endif
296 return vma_prot;
297}
298#endif
299
300#ifndef CONFIG_MMU
301static unsigned long get_unmapped_area_mem(struct file *file,
302 unsigned long addr,
303 unsigned long len,
304 unsigned long pgoff,
305 unsigned long flags)
306{
307 if (!valid_mmap_phys_addr_range(pgoff, len))
308 return (unsigned long) -EINVAL;
309 return pgoff << PAGE_SHIFT;
310}
311
312/* permit direct mmap, for read, write or exec */
313static unsigned memory_mmap_capabilities(struct file *file)
314{
315 return NOMMU_MAP_DIRECT |
316 NOMMU_MAP_READ | NOMMU_MAP_WRITE | NOMMU_MAP_EXEC;
317}
318
319static unsigned zero_mmap_capabilities(struct file *file)
320{
321 return NOMMU_MAP_COPY;
322}
323
324/* can't do an in-place private mapping if there's no MMU */
325static inline int private_mapping_ok(struct vm_area_struct *vma)
326{
327 return is_nommu_shared_mapping(vma->vm_flags);
328}
329#else
330
331static inline int private_mapping_ok(struct vm_area_struct *vma)
332{
333 return 1;
334}
335#endif
336
337static const struct vm_operations_struct mmap_mem_ops = {
338#ifdef CONFIG_HAVE_IOREMAP_PROT
339 .access = generic_access_phys
340#endif
341};
342
343static int mmap_mem(struct file *file, struct vm_area_struct *vma)
344{
345 size_t size = vma->vm_end - vma->vm_start;
346 phys_addr_t offset = (phys_addr_t)vma->vm_pgoff << PAGE_SHIFT;
347
348 /* Does it even fit in phys_addr_t? */
349 if (offset >> PAGE_SHIFT != vma->vm_pgoff)
350 return -EINVAL;
351
352 /* It's illegal to wrap around the end of the physical address space. */
353 if (offset + (phys_addr_t)size - 1 < offset)
354 return -EINVAL;
355
356 if (!valid_mmap_phys_addr_range(vma->vm_pgoff, size))
357 return -EINVAL;
358
359 if (!private_mapping_ok(vma))
360 return -ENOSYS;
361
362 if (!range_is_allowed(vma->vm_pgoff, size))
363 return -EPERM;
364
365 if (!phys_mem_access_prot_allowed(file, vma->vm_pgoff, size,
366 &vma->vm_page_prot))
367 return -EINVAL;
368
369 vma->vm_page_prot = phys_mem_access_prot(file, vma->vm_pgoff,
370 size,
371 vma->vm_page_prot);
372
373 vma->vm_ops = &mmap_mem_ops;
374
375 /* Remap-pfn-range will mark the range VM_IO */
376 if (remap_pfn_range(vma,
377 vma->vm_start,
378 vma->vm_pgoff,
379 size,
380 vma->vm_page_prot)) {
381 return -EAGAIN;
382 }
383 return 0;
384}
385
386static ssize_t read_port(struct file *file, char __user *buf,
387 size_t count, loff_t *ppos)
388{
389 unsigned long i = *ppos;
390 char __user *tmp = buf;
391
392 if (!access_ok(buf, count))
393 return -EFAULT;
394 while (count-- > 0 && i < 65536) {
395 if (__put_user(inb(i), tmp) < 0)
396 return -EFAULT;
397 i++;
398 tmp++;
399 }
400 *ppos = i;
401 return tmp-buf;
402}
403
404static ssize_t write_port(struct file *file, const char __user *buf,
405 size_t count, loff_t *ppos)
406{
407 unsigned long i = *ppos;
408 const char __user *tmp = buf;
409
410 if (!access_ok(buf, count))
411 return -EFAULT;
412 while (count-- > 0 && i < 65536) {
413 char c;
414
415 if (__get_user(c, tmp)) {
416 if (tmp > buf)
417 break;
418 return -EFAULT;
419 }
420 outb(c, i);
421 i++;
422 tmp++;
423 }
424 *ppos = i;
425 return tmp-buf;
426}
427
428static ssize_t read_null(struct file *file, char __user *buf,
429 size_t count, loff_t *ppos)
430{
431 return 0;
432}
433
434static ssize_t write_null(struct file *file, const char __user *buf,
435 size_t count, loff_t *ppos)
436{
437 return count;
438}
439
440static ssize_t read_iter_null(struct kiocb *iocb, struct iov_iter *to)
441{
442 return 0;
443}
444
445static ssize_t write_iter_null(struct kiocb *iocb, struct iov_iter *from)
446{
447 size_t count = iov_iter_count(from);
448 iov_iter_advance(from, count);
449 return count;
450}
451
452static int pipe_to_null(struct pipe_inode_info *info, struct pipe_buffer *buf,
453 struct splice_desc *sd)
454{
455 return sd->len;
456}
457
458static ssize_t splice_write_null(struct pipe_inode_info *pipe, struct file *out,
459 loff_t *ppos, size_t len, unsigned int flags)
460{
461 return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_null);
462}
463
464static int uring_cmd_null(struct io_uring_cmd *ioucmd, unsigned int issue_flags)
465{
466 return 0;
467}
468
469static ssize_t read_iter_zero(struct kiocb *iocb, struct iov_iter *iter)
470{
471 size_t written = 0;
472
473 while (iov_iter_count(iter)) {
474 size_t chunk = iov_iter_count(iter), n;
475
476 if (chunk > PAGE_SIZE)
477 chunk = PAGE_SIZE; /* Just for latency reasons */
478 n = iov_iter_zero(chunk, iter);
479 if (!n && iov_iter_count(iter))
480 return written ? written : -EFAULT;
481 written += n;
482 if (signal_pending(current))
483 return written ? written : -ERESTARTSYS;
484 if (!need_resched())
485 continue;
486 if (iocb->ki_flags & IOCB_NOWAIT)
487 return written ? written : -EAGAIN;
488 cond_resched();
489 }
490 return written;
491}
492
493static ssize_t read_zero(struct file *file, char __user *buf,
494 size_t count, loff_t *ppos)
495{
496 size_t cleared = 0;
497
498 while (count) {
499 size_t chunk = min_t(size_t, count, PAGE_SIZE);
500 size_t left;
501
502 left = clear_user(buf + cleared, chunk);
503 if (unlikely(left)) {
504 cleared += (chunk - left);
505 if (!cleared)
506 return -EFAULT;
507 break;
508 }
509 cleared += chunk;
510 count -= chunk;
511
512 if (signal_pending(current))
513 break;
514 cond_resched();
515 }
516
517 return cleared;
518}
519
520static int mmap_zero(struct file *file, struct vm_area_struct *vma)
521{
522#ifndef CONFIG_MMU
523 return -ENOSYS;
524#endif
525 if (vma->vm_flags & VM_SHARED)
526 return shmem_zero_setup(vma);
527 vma_set_anonymous(vma);
528 return 0;
529}
530
531static unsigned long get_unmapped_area_zero(struct file *file,
532 unsigned long addr, unsigned long len,
533 unsigned long pgoff, unsigned long flags)
534{
535#ifdef CONFIG_MMU
536 if (flags & MAP_SHARED) {
537 /*
538 * mmap_zero() will call shmem_zero_setup() to create a file,
539 * so use shmem's get_unmapped_area in case it can be huge;
540 * and pass NULL for file as in mmap.c's get_unmapped_area(),
541 * so as not to confuse shmem with our handle on "/dev/zero".
542 */
543 return shmem_get_unmapped_area(NULL, addr, len, pgoff, flags);
544 }
545
546 /* Otherwise flags & MAP_PRIVATE: with no shmem object beneath it */
547 return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
548#else
549 return -ENOSYS;
550#endif
551}
552
553static ssize_t write_full(struct file *file, const char __user *buf,
554 size_t count, loff_t *ppos)
555{
556 return -ENOSPC;
557}
558
559/*
560 * Special lseek() function for /dev/null and /dev/zero. Most notably, you
561 * can fopen() both devices with "a" now. This was previously impossible.
562 * -- SRB.
563 */
564static loff_t null_lseek(struct file *file, loff_t offset, int orig)
565{
566 return file->f_pos = 0;
567}
568
569/*
570 * The memory devices use the full 32/64 bits of the offset, and so we cannot
571 * check against negative addresses: they are ok. The return value is weird,
572 * though, in that case (0).
573 *
574 * also note that seeking relative to the "end of file" isn't supported:
575 * it has no meaning, so it returns -EINVAL.
576 */
577static loff_t memory_lseek(struct file *file, loff_t offset, int orig)
578{
579 loff_t ret;
580
581 inode_lock(file_inode(file));
582 switch (orig) {
583 case SEEK_CUR:
584 offset += file->f_pos;
585 fallthrough;
586 case SEEK_SET:
587 /* to avoid userland mistaking f_pos=-9 as -EBADF=-9 */
588 if ((unsigned long long)offset >= -MAX_ERRNO) {
589 ret = -EOVERFLOW;
590 break;
591 }
592 file->f_pos = offset;
593 ret = file->f_pos;
594 force_successful_syscall_return();
595 break;
596 default:
597 ret = -EINVAL;
598 }
599 inode_unlock(file_inode(file));
600 return ret;
601}
602
603static int open_port(struct inode *inode, struct file *filp)
604{
605 int rc;
606
607 if (!capable(CAP_SYS_RAWIO))
608 return -EPERM;
609
610 rc = security_locked_down(LOCKDOWN_DEV_MEM);
611 if (rc)
612 return rc;
613
614 if (iminor(inode) != DEVMEM_MINOR)
615 return 0;
616
617 /*
618 * Use a unified address space to have a single point to manage
619 * revocations when drivers want to take over a /dev/mem mapped
620 * range.
621 */
622 filp->f_mapping = iomem_get_mapping();
623
624 return 0;
625}
626
627#define zero_lseek null_lseek
628#define full_lseek null_lseek
629#define write_zero write_null
630#define write_iter_zero write_iter_null
631#define splice_write_zero splice_write_null
632#define open_mem open_port
633
634static const struct file_operations __maybe_unused mem_fops = {
635 .llseek = memory_lseek,
636 .read = read_mem,
637 .write = write_mem,
638 .mmap = mmap_mem,
639 .open = open_mem,
640#ifndef CONFIG_MMU
641 .get_unmapped_area = get_unmapped_area_mem,
642 .mmap_capabilities = memory_mmap_capabilities,
643#endif
644};
645
646static const struct file_operations null_fops = {
647 .llseek = null_lseek,
648 .read = read_null,
649 .write = write_null,
650 .read_iter = read_iter_null,
651 .write_iter = write_iter_null,
652 .splice_write = splice_write_null,
653 .uring_cmd = uring_cmd_null,
654};
655
656static const struct file_operations __maybe_unused port_fops = {
657 .llseek = memory_lseek,
658 .read = read_port,
659 .write = write_port,
660 .open = open_port,
661};
662
663static const struct file_operations zero_fops = {
664 .llseek = zero_lseek,
665 .write = write_zero,
666 .read_iter = read_iter_zero,
667 .read = read_zero,
668 .write_iter = write_iter_zero,
669 .splice_read = copy_splice_read,
670 .splice_write = splice_write_zero,
671 .mmap = mmap_zero,
672 .get_unmapped_area = get_unmapped_area_zero,
673#ifndef CONFIG_MMU
674 .mmap_capabilities = zero_mmap_capabilities,
675#endif
676};
677
678static const struct file_operations full_fops = {
679 .llseek = full_lseek,
680 .read_iter = read_iter_zero,
681 .write = write_full,
682 .splice_read = copy_splice_read,
683};
684
685static const struct memdev {
686 const char *name;
687 const struct file_operations *fops;
688 fmode_t fmode;
689 umode_t mode;
690} devlist[] = {
691#ifdef CONFIG_DEVMEM
692 [DEVMEM_MINOR] = { "mem", &mem_fops, FMODE_UNSIGNED_OFFSET, 0 },
693#endif
694 [3] = { "null", &null_fops, FMODE_NOWAIT, 0666 },
695#ifdef CONFIG_DEVPORT
696 [4] = { "port", &port_fops, 0, 0 },
697#endif
698 [5] = { "zero", &zero_fops, FMODE_NOWAIT, 0666 },
699 [7] = { "full", &full_fops, 0, 0666 },
700 [8] = { "random", &random_fops, FMODE_NOWAIT, 0666 },
701 [9] = { "urandom", &urandom_fops, FMODE_NOWAIT, 0666 },
702#ifdef CONFIG_PRINTK
703 [11] = { "kmsg", &kmsg_fops, 0, 0644 },
704#endif
705};
706
707static int memory_open(struct inode *inode, struct file *filp)
708{
709 int minor;
710 const struct memdev *dev;
711
712 minor = iminor(inode);
713 if (minor >= ARRAY_SIZE(devlist))
714 return -ENXIO;
715
716 dev = &devlist[minor];
717 if (!dev->fops)
718 return -ENXIO;
719
720 filp->f_op = dev->fops;
721 filp->f_mode |= dev->fmode;
722
723 if (dev->fops->open)
724 return dev->fops->open(inode, filp);
725
726 return 0;
727}
728
729static const struct file_operations memory_fops = {
730 .open = memory_open,
731 .llseek = noop_llseek,
732};
733
734static char *mem_devnode(const struct device *dev, umode_t *mode)
735{
736 if (mode && devlist[MINOR(dev->devt)].mode)
737 *mode = devlist[MINOR(dev->devt)].mode;
738 return NULL;
739}
740
741static const struct class mem_class = {
742 .name = "mem",
743 .devnode = mem_devnode,
744};
745
746static int __init chr_dev_init(void)
747{
748 int retval;
749 int minor;
750
751 if (register_chrdev(MEM_MAJOR, "mem", &memory_fops))
752 printk("unable to get major %d for memory devs\n", MEM_MAJOR);
753
754 retval = class_register(&mem_class);
755 if (retval)
756 return retval;
757
758 for (minor = 1; minor < ARRAY_SIZE(devlist); minor++) {
759 if (!devlist[minor].name)
760 continue;
761
762 /*
763 * Create /dev/port?
764 */
765 if ((minor == DEVPORT_MINOR) && !arch_has_dev_port())
766 continue;
767
768 device_create(&mem_class, NULL, MKDEV(MEM_MAJOR, minor),
769 NULL, devlist[minor].name);
770 }
771
772 return tty_init();
773}
774
775fs_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);