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/ptrace.h>
24 #include <linux/device.h>
25 #include <linux/highmem.h>
26 #include <linux/crash_dump.h>
27 #include <linux/backing-dev.h>
28 #include <linux/bootmem.h>
29 #include <linux/pipe_fs_i.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 if (addr
+ count
> __pa(high_memory
))
99 static inline int valid_mmap_phys_addr_range(unsigned long addr
, size_t size
)
106 * This funcion reads the *physical* memory. The f_pos points directly to the
109 static ssize_t
read_mem(struct file
* file
, char __user
* buf
,
110 size_t count
, loff_t
*ppos
)
112 unsigned long p
= *ppos
;
116 if (!valid_phys_addr_range(p
, count
))
119 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
120 /* we don't have page 0 mapped on sparc and m68k.. */
126 if (clear_user(buf
, sz
))
138 * Handle first page in case it's not aligned
140 if (-p
& (PAGE_SIZE
- 1))
141 sz
= -p
& (PAGE_SIZE
- 1);
145 sz
= min_t(unsigned long, sz
, count
);
148 * On ia64 if a page has been mapped somewhere as
149 * uncached, then it must also be accessed uncached
150 * by the kernel or data corruption may occur
152 ptr
= xlate_dev_mem_ptr(p
);
154 if (copy_to_user(buf
, ptr
, sz
))
166 static ssize_t
write_mem(struct file
* file
, const char __user
* buf
,
167 size_t count
, loff_t
*ppos
)
169 unsigned long p
= *ppos
;
171 unsigned long copied
;
174 if (!valid_phys_addr_range(p
, count
))
179 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
180 /* we don't have page 0 mapped on sparc and m68k.. */
182 unsigned long sz
= PAGE_SIZE
- p
;
185 /* Hmm. Do something? */
195 * Handle first page in case it's not aligned
197 if (-p
& (PAGE_SIZE
- 1))
198 sz
= -p
& (PAGE_SIZE
- 1);
202 sz
= min_t(unsigned long, sz
, count
);
205 * On ia64 if a page has been mapped somewhere as
206 * uncached, then it must also be accessed uncached
207 * by the kernel or data corruption may occur
209 ptr
= xlate_dev_mem_ptr(p
);
211 copied
= copy_from_user(ptr
, buf
, sz
);
213 written
+= sz
- copied
;
228 #ifndef __HAVE_PHYS_MEM_ACCESS_PROT
229 static pgprot_t
phys_mem_access_prot(struct file
*file
, unsigned long pfn
,
230 unsigned long size
, pgprot_t vma_prot
)
232 #ifdef pgprot_noncached
233 unsigned long offset
= pfn
<< PAGE_SHIFT
;
235 if (uncached_access(file
, offset
))
236 return pgprot_noncached(vma_prot
);
242 static int mmap_mem(struct file
* file
, struct vm_area_struct
* vma
)
244 size_t size
= vma
->vm_end
- vma
->vm_start
;
246 if (!valid_mmap_phys_addr_range(vma
->vm_pgoff
<< PAGE_SHIFT
, size
))
249 vma
->vm_page_prot
= phys_mem_access_prot(file
, vma
->vm_pgoff
,
253 /* Remap-pfn-range will mark the range VM_IO and VM_RESERVED */
254 if (remap_pfn_range(vma
,
263 static int mmap_kmem(struct file
* file
, struct vm_area_struct
* vma
)
267 /* Turn a kernel-virtual address into a physical page frame */
268 pfn
= __pa((u64
)vma
->vm_pgoff
<< PAGE_SHIFT
) >> PAGE_SHIFT
;
271 * RED-PEN: on some architectures there is more mapped memory
272 * than available in mem_map which pfn_valid checks
273 * for. Perhaps should add a new macro here.
275 * RED-PEN: vmalloc is not supported right now.
281 return mmap_mem(file
, vma
);
284 #ifdef CONFIG_CRASH_DUMP
286 * Read memory corresponding to the old kernel.
288 static ssize_t
read_oldmem(struct file
*file
, char __user
*buf
,
289 size_t count
, loff_t
*ppos
)
291 unsigned long pfn
, offset
;
292 size_t read
= 0, csize
;
296 pfn
= *ppos
/ PAGE_SIZE
;
297 if (pfn
> saved_max_pfn
)
300 offset
= (unsigned long)(*ppos
% PAGE_SIZE
);
301 if (count
> PAGE_SIZE
- offset
)
302 csize
= PAGE_SIZE
- offset
;
306 rc
= copy_oldmem_page(pfn
, buf
, csize
, offset
, 1);
318 extern long vread(char *buf
, char *addr
, unsigned long count
);
319 extern long vwrite(char *buf
, char *addr
, unsigned long count
);
322 * This function reads the *virtual* memory as seen by the kernel.
324 static ssize_t
read_kmem(struct file
*file
, char __user
*buf
,
325 size_t count
, loff_t
*ppos
)
327 unsigned long p
= *ppos
;
328 ssize_t low_count
, read
, sz
;
329 char * kbuf
; /* k-addr because vread() takes vmlist_lock rwlock */
332 if (p
< (unsigned long) high_memory
) {
334 if (count
> (unsigned long) high_memory
- p
)
335 low_count
= (unsigned long) high_memory
- p
;
337 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
338 /* we don't have page 0 mapped on sparc and m68k.. */
339 if (p
< PAGE_SIZE
&& low_count
> 0) {
340 size_t tmp
= PAGE_SIZE
- p
;
341 if (tmp
> low_count
) tmp
= low_count
;
342 if (clear_user(buf
, tmp
))
351 while (low_count
> 0) {
353 * Handle first page in case it's not aligned
355 if (-p
& (PAGE_SIZE
- 1))
356 sz
= -p
& (PAGE_SIZE
- 1);
360 sz
= min_t(unsigned long, sz
, low_count
);
363 * On ia64 if a page has been mapped somewhere as
364 * uncached, then it must also be accessed uncached
365 * by the kernel or data corruption may occur
367 kbuf
= xlate_dev_kmem_ptr((char *)p
);
369 if (copy_to_user(buf
, kbuf
, sz
))
380 kbuf
= (char *)__get_free_page(GFP_KERNEL
);
388 len
= vread(kbuf
, (char *)p
, len
);
391 if (copy_to_user(buf
, kbuf
, len
)) {
392 free_page((unsigned long)kbuf
);
400 free_page((unsigned long)kbuf
);
407 static inline ssize_t
408 do_write_kmem(void *p
, unsigned long realp
, const char __user
* buf
,
409 size_t count
, loff_t
*ppos
)
412 unsigned long copied
;
415 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
416 /* we don't have page 0 mapped on sparc and m68k.. */
417 if (realp
< PAGE_SIZE
) {
418 unsigned long sz
= PAGE_SIZE
- realp
;
421 /* Hmm. Do something? */
433 * Handle first page in case it's not aligned
435 if (-realp
& (PAGE_SIZE
- 1))
436 sz
= -realp
& (PAGE_SIZE
- 1);
440 sz
= min_t(unsigned long, sz
, count
);
443 * On ia64 if a page has been mapped somewhere as
444 * uncached, then it must also be accessed uncached
445 * by the kernel or data corruption may occur
447 ptr
= xlate_dev_kmem_ptr(p
);
449 copied
= copy_from_user(ptr
, buf
, sz
);
451 written
+= sz
- copied
;
469 * This function writes to the *virtual* memory as seen by the kernel.
471 static ssize_t
write_kmem(struct file
* file
, const char __user
* buf
,
472 size_t count
, loff_t
*ppos
)
474 unsigned long p
= *ppos
;
478 char * kbuf
; /* k-addr because vwrite() takes vmlist_lock rwlock */
480 if (p
< (unsigned long) high_memory
) {
483 if (count
> (unsigned long) high_memory
- p
)
484 wrote
= (unsigned long) high_memory
- p
;
486 written
= do_write_kmem((void*)p
, p
, buf
, wrote
, ppos
);
487 if (written
!= wrote
)
496 kbuf
= (char *)__get_free_page(GFP_KERNEL
);
498 return wrote
? wrote
: -ENOMEM
;
505 written
= copy_from_user(kbuf
, buf
, len
);
509 free_page((unsigned long)kbuf
);
513 len
= vwrite(kbuf
, (char *)p
, len
);
519 free_page((unsigned long)kbuf
);
523 return virtr
+ wrote
;
526 #if defined(CONFIG_ISA) || !defined(__mc68000__)
527 static ssize_t
read_port(struct file
* file
, char __user
* buf
,
528 size_t count
, loff_t
*ppos
)
530 unsigned long i
= *ppos
;
531 char __user
*tmp
= buf
;
533 if (!access_ok(VERIFY_WRITE
, buf
, count
))
535 while (count
-- > 0 && i
< 65536) {
536 if (__put_user(inb(i
),tmp
) < 0)
545 static ssize_t
write_port(struct file
* file
, const char __user
* buf
,
546 size_t count
, loff_t
*ppos
)
548 unsigned long i
= *ppos
;
549 const char __user
* tmp
= buf
;
551 if (!access_ok(VERIFY_READ
,buf
,count
))
553 while (count
-- > 0 && i
< 65536) {
555 if (__get_user(c
, tmp
)) {
569 static ssize_t
read_null(struct file
* file
, char __user
* buf
,
570 size_t count
, loff_t
*ppos
)
575 static ssize_t
write_null(struct file
* file
, const char __user
* buf
,
576 size_t count
, loff_t
*ppos
)
581 static int pipe_to_null(struct pipe_inode_info
*info
, struct pipe_buffer
*buf
,
582 struct splice_desc
*sd
)
587 static ssize_t
splice_write_null(struct pipe_inode_info
*pipe
,struct file
*out
,
588 loff_t
*ppos
, size_t len
, unsigned int flags
)
590 return splice_from_pipe(pipe
, out
, ppos
, len
, flags
, pipe_to_null
);
595 * For fun, we are using the MMU for this.
597 static inline size_t read_zero_pagealigned(char __user
* buf
, size_t size
)
599 struct mm_struct
*mm
;
600 struct vm_area_struct
* vma
;
601 unsigned long addr
=(unsigned long)buf
;
604 /* Oops, this was forgotten before. -ben */
605 down_read(&mm
->mmap_sem
);
607 /* For private mappings, just map in zero pages. */
608 for (vma
= find_vma(mm
, addr
); vma
; vma
= vma
->vm_next
) {
611 if (vma
->vm_start
> addr
|| (vma
->vm_flags
& VM_WRITE
) == 0)
613 if (vma
->vm_flags
& (VM_SHARED
| VM_HUGETLB
))
615 count
= vma
->vm_end
- addr
;
619 zap_page_range(vma
, addr
, count
, NULL
);
620 zeromap_page_range(vma
, addr
, count
, PAGE_COPY
);
629 up_read(&mm
->mmap_sem
);
631 /* The shared case is hard. Let's do the conventional zeroing. */
633 unsigned long unwritten
= clear_user(buf
, PAGE_SIZE
);
635 return size
+ unwritten
- PAGE_SIZE
;
643 up_read(&mm
->mmap_sem
);
647 static ssize_t
read_zero(struct file
* file
, char __user
* buf
,
648 size_t count
, loff_t
*ppos
)
650 unsigned long left
, unwritten
, written
= 0;
655 if (!access_ok(VERIFY_WRITE
, buf
, count
))
660 /* do we want to be clever? Arbitrary cut-off */
661 if (count
>= PAGE_SIZE
*4) {
662 unsigned long partial
;
664 /* How much left of the page? */
665 partial
= (PAGE_SIZE
-1) & -(unsigned long) buf
;
666 unwritten
= clear_user(buf
, partial
);
667 written
= partial
- unwritten
;
672 unwritten
= read_zero_pagealigned(buf
, left
& PAGE_MASK
);
673 written
+= (left
& PAGE_MASK
) - unwritten
;
676 buf
+= left
& PAGE_MASK
;
679 unwritten
= clear_user(buf
, left
);
680 written
+= left
- unwritten
;
682 return written
? written
: -EFAULT
;
685 static int mmap_zero(struct file
* file
, struct vm_area_struct
* vma
)
687 if (vma
->vm_flags
& VM_SHARED
)
688 return shmem_zero_setup(vma
);
689 if (zeromap_page_range(vma
, vma
->vm_start
, vma
->vm_end
- vma
->vm_start
, vma
->vm_page_prot
))
693 #else /* CONFIG_MMU */
694 static ssize_t
read_zero(struct file
* file
, char * buf
,
695 size_t count
, loff_t
*ppos
)
703 chunk
= 4096; /* Just for latency reasons */
704 if (clear_user(buf
, chunk
))
713 static int mmap_zero(struct file
* file
, struct vm_area_struct
* vma
)
717 #endif /* CONFIG_MMU */
719 static ssize_t
write_full(struct file
* file
, const char __user
* buf
,
720 size_t count
, loff_t
*ppos
)
726 * Special lseek() function for /dev/null and /dev/zero. Most notably, you
727 * can fopen() both devices with "a" now. This was previously impossible.
731 static loff_t
null_lseek(struct file
* file
, loff_t offset
, int orig
)
733 return file
->f_pos
= 0;
737 * The memory devices use the full 32/64 bits of the offset, and so we cannot
738 * check against negative addresses: they are ok. The return value is weird,
739 * though, in that case (0).
741 * also note that seeking relative to the "end of file" isn't supported:
742 * it has no meaning, so it returns -EINVAL.
744 static loff_t
memory_lseek(struct file
* file
, loff_t offset
, int orig
)
748 mutex_lock(&file
->f_dentry
->d_inode
->i_mutex
);
751 file
->f_pos
= offset
;
753 force_successful_syscall_return();
756 file
->f_pos
+= offset
;
758 force_successful_syscall_return();
763 mutex_unlock(&file
->f_dentry
->d_inode
->i_mutex
);
767 static int open_port(struct inode
* inode
, struct file
* filp
)
769 return capable(CAP_SYS_RAWIO
) ? 0 : -EPERM
;
772 #define zero_lseek null_lseek
773 #define full_lseek null_lseek
774 #define write_zero write_null
775 #define read_full read_zero
776 #define open_mem open_port
777 #define open_kmem open_mem
778 #define open_oldmem open_mem
780 static struct file_operations mem_fops
= {
781 .llseek
= memory_lseek
,
788 static struct file_operations kmem_fops
= {
789 .llseek
= memory_lseek
,
796 static struct file_operations null_fops
= {
797 .llseek
= null_lseek
,
800 .splice_write
= splice_write_null
,
803 #if defined(CONFIG_ISA) || !defined(__mc68000__)
804 static struct file_operations port_fops
= {
805 .llseek
= memory_lseek
,
812 static struct file_operations zero_fops
= {
813 .llseek
= zero_lseek
,
819 static struct backing_dev_info zero_bdi
= {
820 .capabilities
= BDI_CAP_MAP_COPY
,
823 static struct file_operations full_fops
= {
824 .llseek
= full_lseek
,
829 #ifdef CONFIG_CRASH_DUMP
830 static struct file_operations oldmem_fops
= {
836 static ssize_t
kmsg_write(struct file
* file
, const char __user
* buf
,
837 size_t count
, loff_t
*ppos
)
842 tmp
= kmalloc(count
+ 1, GFP_KERNEL
);
846 if (!copy_from_user(tmp
, buf
, count
)) {
848 ret
= printk("%s", tmp
);
850 /* printk can add a prefix */
857 static struct file_operations kmsg_fops
= {
861 static int memory_open(struct inode
* inode
, struct file
* filp
)
863 switch (iminor(inode
)) {
865 filp
->f_op
= &mem_fops
;
868 filp
->f_op
= &kmem_fops
;
871 filp
->f_op
= &null_fops
;
873 #if defined(CONFIG_ISA) || !defined(__mc68000__)
875 filp
->f_op
= &port_fops
;
879 filp
->f_mapping
->backing_dev_info
= &zero_bdi
;
880 filp
->f_op
= &zero_fops
;
883 filp
->f_op
= &full_fops
;
886 filp
->f_op
= &random_fops
;
889 filp
->f_op
= &urandom_fops
;
892 filp
->f_op
= &kmsg_fops
;
894 #ifdef CONFIG_CRASH_DUMP
896 filp
->f_op
= &oldmem_fops
;
902 if (filp
->f_op
&& filp
->f_op
->open
)
903 return filp
->f_op
->open(inode
,filp
);
907 static struct file_operations memory_fops
= {
908 .open
= memory_open
, /* just a selector for the real open */
911 static const struct {
915 const struct file_operations
*fops
;
916 } devlist
[] = { /* list of minor devices */
917 {1, "mem", S_IRUSR
| S_IWUSR
| S_IRGRP
, &mem_fops
},
918 {2, "kmem", S_IRUSR
| S_IWUSR
| S_IRGRP
, &kmem_fops
},
919 {3, "null", S_IRUGO
| S_IWUGO
, &null_fops
},
920 #if defined(CONFIG_ISA) || !defined(__mc68000__)
921 {4, "port", S_IRUSR
| S_IWUSR
| S_IRGRP
, &port_fops
},
923 {5, "zero", S_IRUGO
| S_IWUGO
, &zero_fops
},
924 {7, "full", S_IRUGO
| S_IWUGO
, &full_fops
},
925 {8, "random", S_IRUGO
| S_IWUSR
, &random_fops
},
926 {9, "urandom", S_IRUGO
| S_IWUSR
, &urandom_fops
},
927 {11,"kmsg", S_IRUGO
| S_IWUSR
, &kmsg_fops
},
928 #ifdef CONFIG_CRASH_DUMP
929 {12,"oldmem", S_IRUSR
| S_IWUSR
| S_IRGRP
, &oldmem_fops
},
933 static struct class *mem_class
;
935 static int __init
chr_dev_init(void)
939 if (register_chrdev(MEM_MAJOR
,"mem",&memory_fops
))
940 printk("unable to get major %d for memory devs\n", MEM_MAJOR
);
942 mem_class
= class_create(THIS_MODULE
, "mem");
943 for (i
= 0; i
< ARRAY_SIZE(devlist
); i
++)
944 class_device_create(mem_class
, NULL
,
945 MKDEV(MEM_MAJOR
, devlist
[i
].minor
),
946 NULL
, devlist
[i
].name
);
951 fs_initcall(chr_dev_init
);