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