2 * linux/drivers/char/mem.c
4 * Copyright (C) 1991, 1992 Linus Torvalds
7 * Jan-11-1998, C. Scott Ananian <cananian@alumni.princeton.edu>
8 * Shared /dev/zero mmaping support, Feb 2000, Kanoj Sarcar <kanoj@sgi.com>
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>
30 #include <asm/uaccess.h>
34 # include <linux/efi.h>
38 * Architectures vary in how they handle caching for addresses
39 * outside of main memory.
42 static inline int uncached_access(struct file
*file
, unsigned long addr
)
44 #if defined(__i386__) && !defined(__arch_um__)
46 * On the PPro and successors, the MTRRs are used to set
47 * memory types for physical addresses outside main memory,
48 * so blindly setting PCD or PWT on those pages is wrong.
49 * For Pentiums and earlier, the surround logic should disable
50 * caching for the high addresses through the KEN pin, but
51 * we maintain the tradition of paranoia in this code.
53 if (file
->f_flags
& O_SYNC
)
55 return !( test_bit(X86_FEATURE_MTRR
, boot_cpu_data
.x86_capability
) ||
56 test_bit(X86_FEATURE_K6_MTRR
, boot_cpu_data
.x86_capability
) ||
57 test_bit(X86_FEATURE_CYRIX_ARR
, boot_cpu_data
.x86_capability
) ||
58 test_bit(X86_FEATURE_CENTAUR_MCR
, boot_cpu_data
.x86_capability
) )
59 && addr
>= __pa(high_memory
);
60 #elif defined(__x86_64__) && !defined(__arch_um__)
62 * This is broken because it can generate memory type aliases,
63 * which can cause cache corruptions
64 * But it is only available for root and we have to be bug-to-bug
65 * compatible with i386.
67 if (file
->f_flags
& O_SYNC
)
69 /* same behaviour as i386. PAT always set to cached and MTRRs control the
71 Hopefully a full PAT implementation will fix that soon. */
73 #elif defined(CONFIG_IA64)
75 * On ia64, we ignore O_SYNC because we cannot tolerate memory attribute aliases.
77 return !(efi_mem_attributes(addr
) & EFI_MEMORY_WB
);
78 #elif defined(CONFIG_MIPS)
80 extern int __uncached_access(struct file
*file
,
83 return __uncached_access(file
, addr
);
87 * Accessing memory above the top the kernel knows about or through a file pointer
88 * that was marked O_SYNC will be done non-cached.
90 if (file
->f_flags
& O_SYNC
)
92 return addr
>= __pa(high_memory
);
96 #ifndef ARCH_HAS_VALID_PHYS_ADDR_RANGE
97 static inline int valid_phys_addr_range(unsigned long addr
, size_t count
)
99 if (addr
+ count
> __pa(high_memory
))
105 static inline int valid_mmap_phys_addr_range(unsigned long pfn
, size_t size
)
111 #ifdef CONFIG_NONPROMISC_DEVMEM
112 static inline int range_is_allowed(unsigned long from
, unsigned long to
)
114 unsigned long cursor
;
116 cursor
= from
>> PAGE_SHIFT
;
117 while ((cursor
<< PAGE_SHIFT
) < to
) {
118 if (!devmem_is_allowed(cursor
)) {
119 printk(KERN_INFO
"Program %s tried to read /dev/mem "
120 "between %lx->%lx.\n",
121 current
->comm
, from
, to
);
129 static inline int range_is_allowed(unsigned long from
, unsigned long to
)
136 * This funcion reads the *physical* memory. The f_pos points directly to the
139 static ssize_t
read_mem(struct file
* file
, char __user
* buf
,
140 size_t count
, loff_t
*ppos
)
142 unsigned long p
= *ppos
;
146 if (!valid_phys_addr_range(p
, count
))
149 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
150 /* we don't have page 0 mapped on sparc and m68k.. */
156 if (clear_user(buf
, sz
))
168 * Handle first page in case it's not aligned
170 if (-p
& (PAGE_SIZE
- 1))
171 sz
= -p
& (PAGE_SIZE
- 1);
175 sz
= min_t(unsigned long, sz
, count
);
178 * On ia64 if a page has been mapped somewhere as
179 * uncached, then it must also be accessed uncached
180 * by the kernel or data corruption may occur
182 ptr
= xlate_dev_mem_ptr(p
);
184 if (!range_is_allowed(p
, p
+count
))
186 if (copy_to_user(buf
, ptr
, sz
))
198 static ssize_t
write_mem(struct file
* file
, const char __user
* buf
,
199 size_t count
, loff_t
*ppos
)
201 unsigned long p
= *ppos
;
203 unsigned long copied
;
206 if (!valid_phys_addr_range(p
, count
))
211 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
212 /* we don't have page 0 mapped on sparc and m68k.. */
214 unsigned long sz
= PAGE_SIZE
- p
;
217 /* Hmm. Do something? */
227 * Handle first page in case it's not aligned
229 if (-p
& (PAGE_SIZE
- 1))
230 sz
= -p
& (PAGE_SIZE
- 1);
234 sz
= min_t(unsigned long, sz
, count
);
237 * On ia64 if a page has been mapped somewhere as
238 * uncached, then it must also be accessed uncached
239 * by the kernel or data corruption may occur
241 ptr
= xlate_dev_mem_ptr(p
);
243 if (!range_is_allowed(p
, p
+sz
))
245 copied
= copy_from_user(ptr
, buf
, sz
);
247 written
+= sz
- copied
;
262 #ifndef __HAVE_PHYS_MEM_ACCESS_PROT
263 static pgprot_t
phys_mem_access_prot(struct file
*file
, unsigned long pfn
,
264 unsigned long size
, pgprot_t vma_prot
)
266 #ifdef pgprot_noncached
267 unsigned long offset
= pfn
<< PAGE_SHIFT
;
269 if (uncached_access(file
, offset
))
270 return pgprot_noncached(vma_prot
);
277 static unsigned long get_unmapped_area_mem(struct file
*file
,
283 if (!valid_mmap_phys_addr_range(pgoff
, len
))
284 return (unsigned long) -EINVAL
;
285 return pgoff
<< PAGE_SHIFT
;
288 /* can't do an in-place private mapping if there's no MMU */
289 static inline int private_mapping_ok(struct vm_area_struct
*vma
)
291 return vma
->vm_flags
& VM_MAYSHARE
;
294 #define get_unmapped_area_mem NULL
296 static inline int private_mapping_ok(struct vm_area_struct
*vma
)
302 static int mmap_mem(struct file
* file
, struct vm_area_struct
* vma
)
304 size_t size
= vma
->vm_end
- vma
->vm_start
;
306 if (!valid_mmap_phys_addr_range(vma
->vm_pgoff
, size
))
309 if (!private_mapping_ok(vma
))
312 vma
->vm_page_prot
= phys_mem_access_prot(file
, vma
->vm_pgoff
,
316 /* Remap-pfn-range will mark the range VM_IO and VM_RESERVED */
317 if (remap_pfn_range(vma
,
326 static int mmap_kmem(struct file
* file
, struct vm_area_struct
* vma
)
330 /* Turn a kernel-virtual address into a physical page frame */
331 pfn
= __pa((u64
)vma
->vm_pgoff
<< PAGE_SHIFT
) >> PAGE_SHIFT
;
334 * RED-PEN: on some architectures there is more mapped memory
335 * than available in mem_map which pfn_valid checks
336 * for. Perhaps should add a new macro here.
338 * RED-PEN: vmalloc is not supported right now.
344 return mmap_mem(file
, vma
);
347 #ifdef CONFIG_CRASH_DUMP
349 * Read memory corresponding to the old kernel.
351 static ssize_t
read_oldmem(struct file
*file
, char __user
*buf
,
352 size_t count
, loff_t
*ppos
)
354 unsigned long pfn
, offset
;
355 size_t read
= 0, csize
;
359 pfn
= *ppos
/ PAGE_SIZE
;
360 if (pfn
> saved_max_pfn
)
363 offset
= (unsigned long)(*ppos
% PAGE_SIZE
);
364 if (count
> PAGE_SIZE
- offset
)
365 csize
= PAGE_SIZE
- offset
;
369 rc
= copy_oldmem_page(pfn
, buf
, csize
, offset
, 1);
381 extern long vread(char *buf
, char *addr
, unsigned long count
);
382 extern long vwrite(char *buf
, char *addr
, unsigned long count
);
385 * This function reads the *virtual* memory as seen by the kernel.
387 static ssize_t
read_kmem(struct file
*file
, char __user
*buf
,
388 size_t count
, loff_t
*ppos
)
390 unsigned long p
= *ppos
;
391 ssize_t low_count
, read
, sz
;
392 char * kbuf
; /* k-addr because vread() takes vmlist_lock rwlock */
395 if (p
< (unsigned long) high_memory
) {
397 if (count
> (unsigned long) high_memory
- p
)
398 low_count
= (unsigned long) high_memory
- p
;
400 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
401 /* we don't have page 0 mapped on sparc and m68k.. */
402 if (p
< PAGE_SIZE
&& low_count
> 0) {
403 size_t tmp
= PAGE_SIZE
- p
;
404 if (tmp
> low_count
) tmp
= low_count
;
405 if (clear_user(buf
, tmp
))
414 while (low_count
> 0) {
416 * Handle first page in case it's not aligned
418 if (-p
& (PAGE_SIZE
- 1))
419 sz
= -p
& (PAGE_SIZE
- 1);
423 sz
= min_t(unsigned long, sz
, low_count
);
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
430 kbuf
= xlate_dev_kmem_ptr((char *)p
);
432 if (copy_to_user(buf
, kbuf
, sz
))
443 kbuf
= (char *)__get_free_page(GFP_KERNEL
);
451 len
= vread(kbuf
, (char *)p
, len
);
454 if (copy_to_user(buf
, kbuf
, len
)) {
455 free_page((unsigned long)kbuf
);
463 free_page((unsigned long)kbuf
);
470 static inline ssize_t
471 do_write_kmem(void *p
, unsigned long realp
, const char __user
* buf
,
472 size_t count
, loff_t
*ppos
)
475 unsigned long copied
;
478 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
479 /* we don't have page 0 mapped on sparc and m68k.. */
480 if (realp
< PAGE_SIZE
) {
481 unsigned long sz
= PAGE_SIZE
- realp
;
484 /* Hmm. Do something? */
496 * Handle first page in case it's not aligned
498 if (-realp
& (PAGE_SIZE
- 1))
499 sz
= -realp
& (PAGE_SIZE
- 1);
503 sz
= min_t(unsigned long, sz
, count
);
506 * On ia64 if a page has been mapped somewhere as
507 * uncached, then it must also be accessed uncached
508 * by the kernel or data corruption may occur
510 ptr
= xlate_dev_kmem_ptr(p
);
512 copied
= copy_from_user(ptr
, buf
, sz
);
514 written
+= sz
- copied
;
532 * This function writes to the *virtual* memory as seen by the kernel.
534 static ssize_t
write_kmem(struct file
* file
, const char __user
* buf
,
535 size_t count
, loff_t
*ppos
)
537 unsigned long p
= *ppos
;
541 char * kbuf
; /* k-addr because vwrite() takes vmlist_lock rwlock */
543 if (p
< (unsigned long) high_memory
) {
546 if (count
> (unsigned long) high_memory
- p
)
547 wrote
= (unsigned long) high_memory
- p
;
549 written
= do_write_kmem((void*)p
, p
, buf
, wrote
, ppos
);
550 if (written
!= wrote
)
559 kbuf
= (char *)__get_free_page(GFP_KERNEL
);
561 return wrote
? wrote
: -ENOMEM
;
568 written
= copy_from_user(kbuf
, buf
, len
);
572 free_page((unsigned long)kbuf
);
576 len
= vwrite(kbuf
, (char *)p
, len
);
582 free_page((unsigned long)kbuf
);
586 return virtr
+ wrote
;
589 #ifdef CONFIG_DEVPORT
590 static ssize_t
read_port(struct file
* file
, char __user
* buf
,
591 size_t count
, loff_t
*ppos
)
593 unsigned long i
= *ppos
;
594 char __user
*tmp
= buf
;
596 if (!access_ok(VERIFY_WRITE
, buf
, count
))
598 while (count
-- > 0 && i
< 65536) {
599 if (__put_user(inb(i
),tmp
) < 0)
608 static ssize_t
write_port(struct file
* file
, const char __user
* buf
,
609 size_t count
, loff_t
*ppos
)
611 unsigned long i
= *ppos
;
612 const char __user
* tmp
= buf
;
614 if (!access_ok(VERIFY_READ
,buf
,count
))
616 while (count
-- > 0 && i
< 65536) {
618 if (__get_user(c
, tmp
)) {
632 static ssize_t
read_null(struct file
* file
, char __user
* buf
,
633 size_t count
, loff_t
*ppos
)
638 static ssize_t
write_null(struct file
* file
, const char __user
* buf
,
639 size_t count
, loff_t
*ppos
)
644 static int pipe_to_null(struct pipe_inode_info
*info
, struct pipe_buffer
*buf
,
645 struct splice_desc
*sd
)
650 static ssize_t
splice_write_null(struct pipe_inode_info
*pipe
,struct file
*out
,
651 loff_t
*ppos
, size_t len
, unsigned int flags
)
653 return splice_from_pipe(pipe
, out
, ppos
, len
, flags
, pipe_to_null
);
656 static ssize_t
read_zero(struct file
* file
, char __user
* buf
,
657 size_t count
, loff_t
*ppos
)
664 if (!access_ok(VERIFY_WRITE
, buf
, count
))
669 unsigned long unwritten
;
670 size_t chunk
= count
;
672 if (chunk
> PAGE_SIZE
)
673 chunk
= PAGE_SIZE
; /* Just for latency reasons */
674 unwritten
= clear_user(buf
, chunk
);
675 written
+= chunk
- unwritten
;
682 return written
? written
: -EFAULT
;
685 static int mmap_zero(struct file
* file
, struct vm_area_struct
* vma
)
690 if (vma
->vm_flags
& VM_SHARED
)
691 return shmem_zero_setup(vma
);
695 static ssize_t
write_full(struct file
* file
, const char __user
* buf
,
696 size_t count
, loff_t
*ppos
)
702 * Special lseek() function for /dev/null and /dev/zero. Most notably, you
703 * can fopen() both devices with "a" now. This was previously impossible.
707 static loff_t
null_lseek(struct file
* file
, loff_t offset
, int orig
)
709 return file
->f_pos
= 0;
713 * The memory devices use the full 32/64 bits of the offset, and so we cannot
714 * check against negative addresses: they are ok. The return value is weird,
715 * though, in that case (0).
717 * also note that seeking relative to the "end of file" isn't supported:
718 * it has no meaning, so it returns -EINVAL.
720 static loff_t
memory_lseek(struct file
* file
, loff_t offset
, int orig
)
724 mutex_lock(&file
->f_path
.dentry
->d_inode
->i_mutex
);
727 file
->f_pos
= offset
;
729 force_successful_syscall_return();
732 file
->f_pos
+= offset
;
734 force_successful_syscall_return();
739 mutex_unlock(&file
->f_path
.dentry
->d_inode
->i_mutex
);
743 static int open_port(struct inode
* inode
, struct file
* filp
)
745 return capable(CAP_SYS_RAWIO
) ? 0 : -EPERM
;
748 #define zero_lseek null_lseek
749 #define full_lseek null_lseek
750 #define write_zero write_null
751 #define read_full read_zero
752 #define open_mem open_port
753 #define open_kmem open_mem
754 #define open_oldmem open_mem
756 static const struct file_operations mem_fops
= {
757 .llseek
= memory_lseek
,
762 .get_unmapped_area
= get_unmapped_area_mem
,
765 static const struct file_operations kmem_fops
= {
766 .llseek
= memory_lseek
,
771 .get_unmapped_area
= get_unmapped_area_mem
,
774 static const struct file_operations null_fops
= {
775 .llseek
= null_lseek
,
778 .splice_write
= splice_write_null
,
781 #ifdef CONFIG_DEVPORT
782 static const struct file_operations port_fops
= {
783 .llseek
= memory_lseek
,
790 static const struct file_operations zero_fops
= {
791 .llseek
= zero_lseek
,
798 * capabilities for /dev/zero
799 * - permits private mappings, "copies" are taken of the source of zeros
801 static struct backing_dev_info zero_bdi
= {
802 .capabilities
= BDI_CAP_MAP_COPY
,
805 static const struct file_operations full_fops
= {
806 .llseek
= full_lseek
,
811 #ifdef CONFIG_CRASH_DUMP
812 static const struct file_operations oldmem_fops
= {
818 static ssize_t
kmsg_write(struct file
* file
, const char __user
* buf
,
819 size_t count
, loff_t
*ppos
)
824 tmp
= kmalloc(count
+ 1, GFP_KERNEL
);
828 if (!copy_from_user(tmp
, buf
, count
)) {
830 ret
= printk("%s", tmp
);
832 /* printk can add a prefix */
839 static const struct file_operations kmsg_fops
= {
843 static int memory_open(struct inode
* inode
, struct file
* filp
)
845 switch (iminor(inode
)) {
847 filp
->f_op
= &mem_fops
;
848 filp
->f_mapping
->backing_dev_info
=
849 &directly_mappable_cdev_bdi
;
852 filp
->f_op
= &kmem_fops
;
853 filp
->f_mapping
->backing_dev_info
=
854 &directly_mappable_cdev_bdi
;
857 filp
->f_op
= &null_fops
;
859 #ifdef CONFIG_DEVPORT
861 filp
->f_op
= &port_fops
;
865 filp
->f_mapping
->backing_dev_info
= &zero_bdi
;
866 filp
->f_op
= &zero_fops
;
869 filp
->f_op
= &full_fops
;
872 filp
->f_op
= &random_fops
;
875 filp
->f_op
= &urandom_fops
;
878 filp
->f_op
= &kmsg_fops
;
880 #ifdef CONFIG_CRASH_DUMP
882 filp
->f_op
= &oldmem_fops
;
888 if (filp
->f_op
&& filp
->f_op
->open
)
889 return filp
->f_op
->open(inode
,filp
);
893 static const struct file_operations memory_fops
= {
894 .open
= memory_open
, /* just a selector for the real open */
897 static const struct {
901 const struct file_operations
*fops
;
902 } devlist
[] = { /* list of minor devices */
903 {1, "mem", S_IRUSR
| S_IWUSR
| S_IRGRP
, &mem_fops
},
904 {2, "kmem", S_IRUSR
| S_IWUSR
| S_IRGRP
, &kmem_fops
},
905 {3, "null", S_IRUGO
| S_IWUGO
, &null_fops
},
906 #ifdef CONFIG_DEVPORT
907 {4, "port", S_IRUSR
| S_IWUSR
| S_IRGRP
, &port_fops
},
909 {5, "zero", S_IRUGO
| S_IWUGO
, &zero_fops
},
910 {7, "full", S_IRUGO
| S_IWUGO
, &full_fops
},
911 {8, "random", S_IRUGO
| S_IWUSR
, &random_fops
},
912 {9, "urandom", S_IRUGO
| S_IWUSR
, &urandom_fops
},
913 {11,"kmsg", S_IRUGO
| S_IWUSR
, &kmsg_fops
},
914 #ifdef CONFIG_CRASH_DUMP
915 {12,"oldmem", S_IRUSR
| S_IWUSR
| S_IRGRP
, &oldmem_fops
},
919 static struct class *mem_class
;
921 static int __init
chr_dev_init(void)
926 err
= bdi_init(&zero_bdi
);
930 if (register_chrdev(MEM_MAJOR
,"mem",&memory_fops
))
931 printk("unable to get major %d for memory devs\n", MEM_MAJOR
);
933 mem_class
= class_create(THIS_MODULE
, "mem");
934 for (i
= 0; i
< ARRAY_SIZE(devlist
); i
++)
935 device_create(mem_class
, NULL
,
936 MKDEV(MEM_MAJOR
, devlist
[i
].minor
),
942 fs_initcall(chr_dev_init
);