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>
11 #include <linux/config.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/raw.h>
20 #include <linux/tty.h>
21 #include <linux/capability.h>
22 #include <linux/smp_lock.h>
23 #include <linux/devfs_fs_kernel.h>
24 #include <linux/ptrace.h>
25 #include <linux/device.h>
26 #include <linux/highmem.h>
27 #include <linux/crash_dump.h>
28 #include <linux/backing-dev.h>
29 #include <linux/bootmem.h>
31 #include <asm/uaccess.h>
35 # include <linux/efi.h>
39 * Architectures vary in how they handle caching for addresses
40 * outside of main memory.
43 static inline int uncached_access(struct file
*file
, unsigned long addr
)
47 * On the PPro and successors, the MTRRs are used to set
48 * memory types for physical addresses outside main memory,
49 * so blindly setting PCD or PWT on those pages is wrong.
50 * For Pentiums and earlier, the surround logic should disable
51 * caching for the high addresses through the KEN pin, but
52 * we maintain the tradition of paranoia in this code.
54 if (file
->f_flags
& O_SYNC
)
56 return !( test_bit(X86_FEATURE_MTRR
, boot_cpu_data
.x86_capability
) ||
57 test_bit(X86_FEATURE_K6_MTRR
, boot_cpu_data
.x86_capability
) ||
58 test_bit(X86_FEATURE_CYRIX_ARR
, boot_cpu_data
.x86_capability
) ||
59 test_bit(X86_FEATURE_CENTAUR_MCR
, boot_cpu_data
.x86_capability
) )
60 && addr
>= __pa(high_memory
);
61 #elif defined(__x86_64__)
63 * This is broken because it can generate memory type aliases,
64 * which can cause cache corruptions
65 * But it is only available for root and we have to be bug-to-bug
66 * compatible with i386.
68 if (file
->f_flags
& O_SYNC
)
70 /* same behaviour as i386. PAT always set to cached and MTRRs control the
72 Hopefully a full PAT implementation will fix that soon. */
74 #elif defined(CONFIG_IA64)
76 * On ia64, we ignore O_SYNC because we cannot tolerate memory attribute aliases.
78 return !(efi_mem_attributes(addr
) & EFI_MEMORY_WB
);
81 * Accessing memory above the top the kernel knows about or through a file pointer
82 * that was marked O_SYNC will be done non-cached.
84 if (file
->f_flags
& O_SYNC
)
86 return addr
>= __pa(high_memory
);
90 #ifndef ARCH_HAS_VALID_PHYS_ADDR_RANGE
91 static inline int valid_phys_addr_range(unsigned long addr
, size_t *count
)
93 unsigned long end_mem
;
95 end_mem
= __pa(high_memory
);
99 if (*count
> end_mem
- addr
)
100 *count
= end_mem
- addr
;
105 static inline int valid_mmap_phys_addr_range(unsigned long addr
, size_t *size
)
112 * This funcion reads the *physical* memory. The f_pos points directly to the
115 static ssize_t
read_mem(struct file
* file
, char __user
* buf
,
116 size_t count
, loff_t
*ppos
)
118 unsigned long p
= *ppos
;
122 if (!valid_phys_addr_range(p
, &count
))
125 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
126 /* we don't have page 0 mapped on sparc and m68k.. */
132 if (clear_user(buf
, sz
))
144 * Handle first page in case it's not aligned
146 if (-p
& (PAGE_SIZE
- 1))
147 sz
= -p
& (PAGE_SIZE
- 1);
151 sz
= min_t(unsigned long, sz
, count
);
154 * On ia64 if a page has been mapped somewhere as
155 * uncached, then it must also be accessed uncached
156 * by the kernel or data corruption may occur
158 ptr
= xlate_dev_mem_ptr(p
);
160 if (copy_to_user(buf
, ptr
, sz
))
172 static ssize_t
write_mem(struct file
* file
, const char __user
* buf
,
173 size_t count
, loff_t
*ppos
)
175 unsigned long p
= *ppos
;
177 unsigned long copied
;
180 if (!valid_phys_addr_range(p
, &count
))
185 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
186 /* we don't have page 0 mapped on sparc and m68k.. */
188 unsigned long sz
= PAGE_SIZE
- p
;
191 /* Hmm. Do something? */
201 * Handle first page in case it's not aligned
203 if (-p
& (PAGE_SIZE
- 1))
204 sz
= -p
& (PAGE_SIZE
- 1);
208 sz
= min_t(unsigned long, sz
, count
);
211 * On ia64 if a page has been mapped somewhere as
212 * uncached, then it must also be accessed uncached
213 * by the kernel or data corruption may occur
215 ptr
= xlate_dev_mem_ptr(p
);
217 copied
= copy_from_user(ptr
, buf
, sz
);
221 ret
= written
+ (sz
- copied
);
236 #ifndef __HAVE_PHYS_MEM_ACCESS_PROT
237 static pgprot_t
phys_mem_access_prot(struct file
*file
, unsigned long pfn
,
238 unsigned long size
, pgprot_t vma_prot
)
240 #ifdef pgprot_noncached
241 unsigned long offset
= pfn
<< PAGE_SHIFT
;
243 if (uncached_access(file
, offset
))
244 return pgprot_noncached(vma_prot
);
250 static int mmap_mem(struct file
* file
, struct vm_area_struct
* vma
)
252 size_t size
= vma
->vm_end
- vma
->vm_start
;
254 if (!valid_mmap_phys_addr_range(vma
->vm_pgoff
<< PAGE_SHIFT
, &size
))
257 vma
->vm_page_prot
= phys_mem_access_prot(file
, vma
->vm_pgoff
,
261 /* Remap-pfn-range will mark the range VM_IO and VM_RESERVED */
262 if (remap_pfn_range(vma
,
271 static int mmap_kmem(struct file
* file
, struct vm_area_struct
* vma
)
275 /* Turn a kernel-virtual address into a physical page frame */
276 pfn
= __pa((u64
)vma
->vm_pgoff
<< PAGE_SHIFT
) >> PAGE_SHIFT
;
279 * RED-PEN: on some architectures there is more mapped memory
280 * than available in mem_map which pfn_valid checks
281 * for. Perhaps should add a new macro here.
283 * RED-PEN: vmalloc is not supported right now.
289 return mmap_mem(file
, vma
);
292 #ifdef CONFIG_CRASH_DUMP
294 * Read memory corresponding to the old kernel.
296 static ssize_t
read_oldmem(struct file
*file
, char __user
*buf
,
297 size_t count
, loff_t
*ppos
)
299 unsigned long pfn
, offset
;
300 size_t read
= 0, csize
;
304 pfn
= *ppos
/ PAGE_SIZE
;
305 if (pfn
> saved_max_pfn
)
308 offset
= (unsigned long)(*ppos
% PAGE_SIZE
);
309 if (count
> PAGE_SIZE
- offset
)
310 csize
= PAGE_SIZE
- offset
;
314 rc
= copy_oldmem_page(pfn
, buf
, csize
, offset
, 1);
326 extern long vread(char *buf
, char *addr
, unsigned long count
);
327 extern long vwrite(char *buf
, char *addr
, unsigned long count
);
330 * This function reads the *virtual* memory as seen by the kernel.
332 static ssize_t
read_kmem(struct file
*file
, char __user
*buf
,
333 size_t count
, loff_t
*ppos
)
335 unsigned long p
= *ppos
;
336 ssize_t low_count
, read
, sz
;
337 char * kbuf
; /* k-addr because vread() takes vmlist_lock rwlock */
340 if (p
< (unsigned long) high_memory
) {
342 if (count
> (unsigned long) high_memory
- p
)
343 low_count
= (unsigned long) high_memory
- p
;
345 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
346 /* we don't have page 0 mapped on sparc and m68k.. */
347 if (p
< PAGE_SIZE
&& low_count
> 0) {
348 size_t tmp
= PAGE_SIZE
- p
;
349 if (tmp
> low_count
) tmp
= low_count
;
350 if (clear_user(buf
, tmp
))
359 while (low_count
> 0) {
361 * Handle first page in case it's not aligned
363 if (-p
& (PAGE_SIZE
- 1))
364 sz
= -p
& (PAGE_SIZE
- 1);
368 sz
= min_t(unsigned long, sz
, low_count
);
371 * On ia64 if a page has been mapped somewhere as
372 * uncached, then it must also be accessed uncached
373 * by the kernel or data corruption may occur
375 kbuf
= xlate_dev_kmem_ptr((char *)p
);
377 if (copy_to_user(buf
, kbuf
, sz
))
388 kbuf
= (char *)__get_free_page(GFP_KERNEL
);
396 len
= vread(kbuf
, (char *)p
, len
);
399 if (copy_to_user(buf
, kbuf
, len
)) {
400 free_page((unsigned long)kbuf
);
408 free_page((unsigned long)kbuf
);
415 static inline ssize_t
416 do_write_kmem(void *p
, unsigned long realp
, const char __user
* buf
,
417 size_t count
, loff_t
*ppos
)
420 unsigned long copied
;
423 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
424 /* we don't have page 0 mapped on sparc and m68k.. */
425 if (realp
< PAGE_SIZE
) {
426 unsigned long sz
= PAGE_SIZE
- realp
;
429 /* Hmm. Do something? */
441 * Handle first page in case it's not aligned
443 if (-realp
& (PAGE_SIZE
- 1))
444 sz
= -realp
& (PAGE_SIZE
- 1);
448 sz
= min_t(unsigned long, sz
, count
);
451 * On ia64 if a page has been mapped somewhere as
452 * uncached, then it must also be accessed uncached
453 * by the kernel or data corruption may occur
455 ptr
= xlate_dev_kmem_ptr(p
);
457 copied
= copy_from_user(ptr
, buf
, sz
);
461 ret
= written
+ (sz
- copied
);
479 * This function writes to the *virtual* memory as seen by the kernel.
481 static ssize_t
write_kmem(struct file
* file
, const char __user
* buf
,
482 size_t count
, loff_t
*ppos
)
484 unsigned long p
= *ppos
;
488 char * kbuf
; /* k-addr because vwrite() takes vmlist_lock rwlock */
490 if (p
< (unsigned long) high_memory
) {
493 if (count
> (unsigned long) high_memory
- p
)
494 wrote
= (unsigned long) high_memory
- p
;
496 written
= do_write_kmem((void*)p
, p
, buf
, wrote
, ppos
);
497 if (written
!= wrote
)
506 kbuf
= (char *)__get_free_page(GFP_KERNEL
);
508 return wrote
? wrote
: -ENOMEM
;
515 written
= copy_from_user(kbuf
, buf
, len
);
519 free_page((unsigned long)kbuf
);
520 ret
= wrote
+ virtr
+ (len
- written
);
521 return ret
? ret
: -EFAULT
;
524 len
= vwrite(kbuf
, (char *)p
, len
);
530 free_page((unsigned long)kbuf
);
534 return virtr
+ wrote
;
537 #if defined(CONFIG_ISA) || !defined(__mc68000__)
538 static ssize_t
read_port(struct file
* file
, char __user
* buf
,
539 size_t count
, loff_t
*ppos
)
541 unsigned long i
= *ppos
;
542 char __user
*tmp
= buf
;
544 if (!access_ok(VERIFY_WRITE
, buf
, count
))
546 while (count
-- > 0 && i
< 65536) {
547 if (__put_user(inb(i
),tmp
) < 0)
556 static ssize_t
write_port(struct file
* file
, const char __user
* buf
,
557 size_t count
, loff_t
*ppos
)
559 unsigned long i
= *ppos
;
560 const char __user
* tmp
= buf
;
562 if (!access_ok(VERIFY_READ
,buf
,count
))
564 while (count
-- > 0 && i
< 65536) {
566 if (__get_user(c
, tmp
))
577 static ssize_t
read_null(struct file
* file
, char __user
* buf
,
578 size_t count
, loff_t
*ppos
)
583 static ssize_t
write_null(struct file
* file
, const char __user
* buf
,
584 size_t count
, loff_t
*ppos
)
591 * For fun, we are using the MMU for this.
593 static inline size_t read_zero_pagealigned(char __user
* buf
, size_t size
)
595 struct mm_struct
*mm
;
596 struct vm_area_struct
* vma
;
597 unsigned long addr
=(unsigned long)buf
;
600 /* Oops, this was forgotten before. -ben */
601 down_read(&mm
->mmap_sem
);
603 /* For private mappings, just map in zero pages. */
604 for (vma
= find_vma(mm
, addr
); vma
; vma
= vma
->vm_next
) {
607 if (vma
->vm_start
> addr
|| (vma
->vm_flags
& VM_WRITE
) == 0)
609 if (vma
->vm_flags
& (VM_SHARED
| VM_HUGETLB
))
611 count
= vma
->vm_end
- addr
;
615 zap_page_range(vma
, addr
, count
, NULL
);
616 zeromap_page_range(vma
, addr
, count
, PAGE_COPY
);
625 up_read(&mm
->mmap_sem
);
627 /* The shared case is hard. Let's do the conventional zeroing. */
629 unsigned long unwritten
= clear_user(buf
, PAGE_SIZE
);
631 return size
+ unwritten
- PAGE_SIZE
;
639 up_read(&mm
->mmap_sem
);
643 static ssize_t
read_zero(struct file
* file
, char __user
* buf
,
644 size_t count
, loff_t
*ppos
)
646 unsigned long left
, unwritten
, written
= 0;
651 if (!access_ok(VERIFY_WRITE
, buf
, count
))
656 /* do we want to be clever? Arbitrary cut-off */
657 if (count
>= PAGE_SIZE
*4) {
658 unsigned long partial
;
660 /* How much left of the page? */
661 partial
= (PAGE_SIZE
-1) & -(unsigned long) buf
;
662 unwritten
= clear_user(buf
, partial
);
663 written
= partial
- unwritten
;
668 unwritten
= read_zero_pagealigned(buf
, left
& PAGE_MASK
);
669 written
+= (left
& PAGE_MASK
) - unwritten
;
672 buf
+= left
& PAGE_MASK
;
675 unwritten
= clear_user(buf
, left
);
676 written
+= left
- unwritten
;
678 return written
? written
: -EFAULT
;
681 static int mmap_zero(struct file
* file
, struct vm_area_struct
* vma
)
683 if (vma
->vm_flags
& VM_SHARED
)
684 return shmem_zero_setup(vma
);
685 if (zeromap_page_range(vma
, vma
->vm_start
, vma
->vm_end
- vma
->vm_start
, vma
->vm_page_prot
))
689 #else /* CONFIG_MMU */
690 static ssize_t
read_zero(struct file
* file
, char * buf
,
691 size_t count
, loff_t
*ppos
)
699 chunk
= 4096; /* Just for latency reasons */
700 if (clear_user(buf
, chunk
))
709 static int mmap_zero(struct file
* file
, struct vm_area_struct
* vma
)
713 #endif /* CONFIG_MMU */
715 static ssize_t
write_full(struct file
* file
, const char __user
* buf
,
716 size_t count
, loff_t
*ppos
)
722 * Special lseek() function for /dev/null and /dev/zero. Most notably, you
723 * can fopen() both devices with "a" now. This was previously impossible.
727 static loff_t
null_lseek(struct file
* file
, loff_t offset
, int orig
)
729 return file
->f_pos
= 0;
733 * The memory devices use the full 32/64 bits of the offset, and so we cannot
734 * check against negative addresses: they are ok. The return value is weird,
735 * though, in that case (0).
737 * also note that seeking relative to the "end of file" isn't supported:
738 * it has no meaning, so it returns -EINVAL.
740 static loff_t
memory_lseek(struct file
* file
, loff_t offset
, int orig
)
744 mutex_lock(&file
->f_dentry
->d_inode
->i_mutex
);
747 file
->f_pos
= offset
;
749 force_successful_syscall_return();
752 file
->f_pos
+= offset
;
754 force_successful_syscall_return();
759 mutex_unlock(&file
->f_dentry
->d_inode
->i_mutex
);
763 static int open_port(struct inode
* inode
, struct file
* filp
)
765 return capable(CAP_SYS_RAWIO
) ? 0 : -EPERM
;
768 #define zero_lseek null_lseek
769 #define full_lseek null_lseek
770 #define write_zero write_null
771 #define read_full read_zero
772 #define open_mem open_port
773 #define open_kmem open_mem
774 #define open_oldmem open_mem
776 static struct file_operations mem_fops
= {
777 .llseek
= memory_lseek
,
784 static struct file_operations kmem_fops
= {
785 .llseek
= memory_lseek
,
792 static struct file_operations null_fops
= {
793 .llseek
= null_lseek
,
798 #if defined(CONFIG_ISA) || !defined(__mc68000__)
799 static struct file_operations port_fops
= {
800 .llseek
= memory_lseek
,
807 static struct file_operations zero_fops
= {
808 .llseek
= zero_lseek
,
814 static struct backing_dev_info zero_bdi
= {
815 .capabilities
= BDI_CAP_MAP_COPY
,
818 static struct file_operations full_fops
= {
819 .llseek
= full_lseek
,
824 #ifdef CONFIG_CRASH_DUMP
825 static struct file_operations oldmem_fops
= {
831 static ssize_t
kmsg_write(struct file
* file
, const char __user
* buf
,
832 size_t count
, loff_t
*ppos
)
837 tmp
= kmalloc(count
+ 1, GFP_KERNEL
);
841 if (!copy_from_user(tmp
, buf
, count
)) {
843 ret
= printk("%s", tmp
);
845 /* printk can add a prefix */
852 static struct file_operations kmsg_fops
= {
856 static int memory_open(struct inode
* inode
, struct file
* filp
)
858 switch (iminor(inode
)) {
860 filp
->f_op
= &mem_fops
;
863 filp
->f_op
= &kmem_fops
;
866 filp
->f_op
= &null_fops
;
868 #if defined(CONFIG_ISA) || !defined(__mc68000__)
870 filp
->f_op
= &port_fops
;
874 filp
->f_mapping
->backing_dev_info
= &zero_bdi
;
875 filp
->f_op
= &zero_fops
;
878 filp
->f_op
= &full_fops
;
881 filp
->f_op
= &random_fops
;
884 filp
->f_op
= &urandom_fops
;
887 filp
->f_op
= &kmsg_fops
;
889 #ifdef CONFIG_CRASH_DUMP
891 filp
->f_op
= &oldmem_fops
;
897 if (filp
->f_op
&& filp
->f_op
->open
)
898 return filp
->f_op
->open(inode
,filp
);
902 static struct file_operations memory_fops
= {
903 .open
= memory_open
, /* just a selector for the real open */
906 static const struct {
910 struct file_operations
*fops
;
911 } devlist
[] = { /* list of minor devices */
912 {1, "mem", S_IRUSR
| S_IWUSR
| S_IRGRP
, &mem_fops
},
913 {2, "kmem", S_IRUSR
| S_IWUSR
| S_IRGRP
, &kmem_fops
},
914 {3, "null", S_IRUGO
| S_IWUGO
, &null_fops
},
915 #if defined(CONFIG_ISA) || !defined(__mc68000__)
916 {4, "port", S_IRUSR
| S_IWUSR
| S_IRGRP
, &port_fops
},
918 {5, "zero", S_IRUGO
| S_IWUGO
, &zero_fops
},
919 {7, "full", S_IRUGO
| S_IWUGO
, &full_fops
},
920 {8, "random", S_IRUGO
| S_IWUSR
, &random_fops
},
921 {9, "urandom", S_IRUGO
| S_IWUSR
, &urandom_fops
},
922 {11,"kmsg", S_IRUGO
| S_IWUSR
, &kmsg_fops
},
923 #ifdef CONFIG_CRASH_DUMP
924 {12,"oldmem", S_IRUSR
| S_IWUSR
| S_IRGRP
, &oldmem_fops
},
928 static struct class *mem_class
;
930 static int __init
chr_dev_init(void)
934 if (register_chrdev(MEM_MAJOR
,"mem",&memory_fops
))
935 printk("unable to get major %d for memory devs\n", MEM_MAJOR
);
937 mem_class
= class_create(THIS_MODULE
, "mem");
938 for (i
= 0; i
< ARRAY_SIZE(devlist
); i
++) {
939 class_device_create(mem_class
, NULL
,
940 MKDEV(MEM_MAJOR
, devlist
[i
].minor
),
941 NULL
, devlist
[i
].name
);
942 devfs_mk_cdev(MKDEV(MEM_MAJOR
, devlist
[i
].minor
),
943 S_IFCHR
| devlist
[i
].mode
, devlist
[i
].name
);
949 fs_initcall(chr_dev_init
);