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 mmapping 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>
37 static inline unsigned long size_inside_page(unsigned long start
,
42 sz
= PAGE_SIZE
- (start
& (PAGE_SIZE
- 1));
47 #ifndef ARCH_HAS_VALID_PHYS_ADDR_RANGE
48 static inline int valid_phys_addr_range(unsigned long addr
, size_t count
)
50 if (addr
+ count
> __pa(high_memory
))
56 static inline int valid_mmap_phys_addr_range(unsigned long pfn
, size_t size
)
62 #ifdef CONFIG_STRICT_DEVMEM
63 static inline int range_is_allowed(unsigned long pfn
, unsigned long size
)
65 u64 from
= ((u64
)pfn
) << PAGE_SHIFT
;
70 if (!devmem_is_allowed(pfn
)) {
72 "Program %s tried to access /dev/mem between %Lx->%Lx.\n",
73 current
->comm
, from
, to
);
82 static inline int range_is_allowed(unsigned long pfn
, unsigned long size
)
88 void __weak
unxlate_dev_mem_ptr(unsigned long phys
, void *addr
)
93 * This funcion reads the *physical* memory. The f_pos points directly to the
96 static ssize_t
read_mem(struct file
*file
, char __user
*buf
,
97 size_t count
, loff_t
*ppos
)
99 unsigned long p
= *ppos
;
103 if (!valid_phys_addr_range(p
, count
))
106 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
107 /* we don't have page 0 mapped on sparc and m68k.. */
109 sz
= size_inside_page(p
, count
);
111 if (clear_user(buf
, sz
))
122 unsigned long remaining
;
124 sz
= size_inside_page(p
, count
);
126 if (!range_is_allowed(p
>> PAGE_SHIFT
, count
))
130 * On ia64 if a page has been mapped somewhere as uncached, then
131 * it must also be accessed uncached by the kernel or data
132 * corruption may occur.
134 ptr
= xlate_dev_mem_ptr(p
);
138 remaining
= copy_to_user(buf
, ptr
, sz
);
139 unxlate_dev_mem_ptr(p
, ptr
);
153 static ssize_t
write_mem(struct file
*file
, const char __user
*buf
,
154 size_t count
, loff_t
*ppos
)
156 unsigned long p
= *ppos
;
158 unsigned long copied
;
161 if (!valid_phys_addr_range(p
, count
))
166 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
167 /* we don't have page 0 mapped on sparc and m68k.. */
169 sz
= size_inside_page(p
, count
);
170 /* Hmm. Do something? */
179 sz
= size_inside_page(p
, count
);
181 if (!range_is_allowed(p
>> PAGE_SHIFT
, sz
))
185 * On ia64 if a page has been mapped somewhere as uncached, then
186 * it must also be accessed uncached by the kernel or data
187 * corruption may occur.
189 ptr
= xlate_dev_mem_ptr(p
);
196 copied
= copy_from_user(ptr
, buf
, sz
);
197 unxlate_dev_mem_ptr(p
, ptr
);
199 written
+= sz
- copied
;
215 int __weak
phys_mem_access_prot_allowed(struct file
*file
,
216 unsigned long pfn
, unsigned long size
, pgprot_t
*vma_prot
)
221 #ifndef __HAVE_PHYS_MEM_ACCESS_PROT
224 * Architectures vary in how they handle caching for addresses
225 * outside of main memory.
228 static int uncached_access(struct file
*file
, unsigned long addr
)
230 #if defined(CONFIG_IA64)
232 * On ia64, we ignore O_DSYNC because we cannot tolerate memory
235 return !(efi_mem_attributes(addr
) & EFI_MEMORY_WB
);
236 #elif defined(CONFIG_MIPS)
238 extern int __uncached_access(struct file
*file
,
241 return __uncached_access(file
, addr
);
245 * Accessing memory above the top the kernel knows about or through a
247 * that was marked O_DSYNC will be done non-cached.
249 if (file
->f_flags
& O_DSYNC
)
251 return addr
>= __pa(high_memory
);
255 static pgprot_t
phys_mem_access_prot(struct file
*file
, unsigned long pfn
,
256 unsigned long size
, pgprot_t vma_prot
)
258 #ifdef pgprot_noncached
259 unsigned long offset
= pfn
<< PAGE_SHIFT
;
261 if (uncached_access(file
, offset
))
262 return pgprot_noncached(vma_prot
);
269 static unsigned long get_unmapped_area_mem(struct file
*file
,
275 if (!valid_mmap_phys_addr_range(pgoff
, len
))
276 return (unsigned long) -EINVAL
;
277 return pgoff
<< PAGE_SHIFT
;
280 /* can't do an in-place private mapping if there's no MMU */
281 static inline int private_mapping_ok(struct vm_area_struct
*vma
)
283 return vma
->vm_flags
& VM_MAYSHARE
;
286 #define get_unmapped_area_mem NULL
288 static inline int private_mapping_ok(struct vm_area_struct
*vma
)
294 static const struct vm_operations_struct mmap_mem_ops
= {
295 #ifdef CONFIG_HAVE_IOREMAP_PROT
296 .access
= generic_access_phys
300 static int mmap_mem(struct file
*file
, struct vm_area_struct
*vma
)
302 size_t size
= vma
->vm_end
- vma
->vm_start
;
304 if (!valid_mmap_phys_addr_range(vma
->vm_pgoff
, size
))
307 if (!private_mapping_ok(vma
))
310 if (!range_is_allowed(vma
->vm_pgoff
, size
))
313 if (!phys_mem_access_prot_allowed(file
, vma
->vm_pgoff
, size
,
317 vma
->vm_page_prot
= phys_mem_access_prot(file
, vma
->vm_pgoff
,
321 vma
->vm_ops
= &mmap_mem_ops
;
323 /* Remap-pfn-range will mark the range VM_IO and VM_RESERVED */
324 if (remap_pfn_range(vma
,
328 vma
->vm_page_prot
)) {
334 #ifdef CONFIG_DEVKMEM
335 static int mmap_kmem(struct file
*file
, struct vm_area_struct
*vma
)
339 /* Turn a kernel-virtual address into a physical page frame */
340 pfn
= __pa((u64
)vma
->vm_pgoff
<< PAGE_SHIFT
) >> PAGE_SHIFT
;
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
347 * RED-PEN: vmalloc is not supported right now.
353 return mmap_mem(file
, vma
);
357 #ifdef CONFIG_CRASH_DUMP
359 * Read memory corresponding to the old kernel.
361 static ssize_t
read_oldmem(struct file
*file
, char __user
*buf
,
362 size_t count
, loff_t
*ppos
)
364 unsigned long pfn
, offset
;
365 size_t read
= 0, csize
;
369 pfn
= *ppos
/ PAGE_SIZE
;
370 if (pfn
> saved_max_pfn
)
373 offset
= (unsigned long)(*ppos
% PAGE_SIZE
);
374 if (count
> PAGE_SIZE
- offset
)
375 csize
= PAGE_SIZE
- offset
;
379 rc
= copy_oldmem_page(pfn
, buf
, csize
, offset
, 1);
391 #ifdef CONFIG_DEVKMEM
393 * This function reads the *virtual* memory as seen by the kernel.
395 static ssize_t
read_kmem(struct file
*file
, char __user
*buf
,
396 size_t count
, loff_t
*ppos
)
398 unsigned long p
= *ppos
;
399 ssize_t low_count
, read
, sz
;
400 char * kbuf
; /* k-addr because vread() takes vmlist_lock rwlock */
404 if (p
< (unsigned long) high_memory
) {
406 if (count
> (unsigned long)high_memory
- p
)
407 low_count
= (unsigned long)high_memory
- p
;
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
))
422 while (low_count
> 0) {
423 sz
= size_inside_page(p
, 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
);
447 sz
= size_inside_page(p
, count
);
448 if (!is_vmalloc_or_module_addr((void *)p
)) {
452 sz
= vread(kbuf
, (char *)p
, sz
);
455 if (copy_to_user(buf
, kbuf
, sz
)) {
464 free_page((unsigned long)kbuf
);
467 return read
? read
: err
;
471 static ssize_t
do_write_kmem(unsigned long p
, 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.. */
481 sz
= size_inside_page(p
, count
);
482 /* Hmm. Do something? */
493 sz
= size_inside_page(p
, count
);
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.
500 ptr
= xlate_dev_kmem_ptr((char *)p
);
502 copied
= copy_from_user(ptr
, buf
, sz
);
504 written
+= sz
- copied
;
520 * This function writes to the *virtual* memory as seen by the kernel.
522 static ssize_t
write_kmem(struct file
*file
, const char __user
*buf
,
523 size_t count
, loff_t
*ppos
)
525 unsigned long p
= *ppos
;
528 char * kbuf
; /* k-addr because vwrite() takes vmlist_lock rwlock */
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
)
543 kbuf
= (char *)__get_free_page(GFP_KERNEL
);
545 return wrote
? wrote
: -ENOMEM
;
547 unsigned long sz
= size_inside_page(p
, count
);
550 if (!is_vmalloc_or_module_addr((void *)p
)) {
554 n
= copy_from_user(kbuf
, buf
, sz
);
559 vwrite(kbuf
, (char *)p
, sz
);
565 free_page((unsigned long)kbuf
);
569 return virtr
+ wrote
? : err
;
573 #ifdef CONFIG_DEVPORT
574 static ssize_t
read_port(struct file
*file
, char __user
*buf
,
575 size_t count
, loff_t
*ppos
)
577 unsigned long i
= *ppos
;
578 char __user
*tmp
= buf
;
580 if (!access_ok(VERIFY_WRITE
, buf
, count
))
582 while (count
-- > 0 && i
< 65536) {
583 if (__put_user(inb(i
), tmp
) < 0)
592 static ssize_t
write_port(struct file
*file
, const char __user
*buf
,
593 size_t count
, loff_t
*ppos
)
595 unsigned long i
= *ppos
;
596 const char __user
* tmp
= buf
;
598 if (!access_ok(VERIFY_READ
, buf
, count
))
600 while (count
-- > 0 && i
< 65536) {
602 if (__get_user(c
, tmp
)) {
616 static ssize_t
read_null(struct file
*file
, char __user
*buf
,
617 size_t count
, loff_t
*ppos
)
622 static ssize_t
write_null(struct file
*file
, const char __user
*buf
,
623 size_t count
, loff_t
*ppos
)
628 static int pipe_to_null(struct pipe_inode_info
*info
, struct pipe_buffer
*buf
,
629 struct splice_desc
*sd
)
634 static ssize_t
splice_write_null(struct pipe_inode_info
*pipe
, struct file
*out
,
635 loff_t
*ppos
, size_t len
, unsigned int flags
)
637 return splice_from_pipe(pipe
, out
, ppos
, len
, flags
, pipe_to_null
);
640 static ssize_t
read_zero(struct file
*file
, char __user
*buf
,
641 size_t count
, loff_t
*ppos
)
648 if (!access_ok(VERIFY_WRITE
, buf
, count
))
653 unsigned long unwritten
;
654 size_t chunk
= count
;
656 if (chunk
> PAGE_SIZE
)
657 chunk
= PAGE_SIZE
; /* Just for latency reasons */
658 unwritten
= __clear_user(buf
, chunk
);
659 written
+= chunk
- unwritten
;
662 if (signal_pending(current
))
663 return written
? written
: -ERESTARTSYS
;
668 return written
? written
: -EFAULT
;
671 static int mmap_zero(struct file
*file
, struct vm_area_struct
*vma
)
676 if (vma
->vm_flags
& VM_SHARED
)
677 return shmem_zero_setup(vma
);
681 static ssize_t
write_full(struct file
*file
, const char __user
*buf
,
682 size_t count
, loff_t
*ppos
)
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.
692 static loff_t
null_lseek(struct file
*file
, loff_t offset
, int orig
)
694 return file
->f_pos
= 0;
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).
702 * also note that seeking relative to the "end of file" isn't supported:
703 * it has no meaning, so it returns -EINVAL.
705 static loff_t
memory_lseek(struct file
*file
, loff_t offset
, int orig
)
709 mutex_lock(&file
->f_path
.dentry
->d_inode
->i_mutex
);
712 offset
+= file
->f_pos
;
713 if ((unsigned long long)offset
<
714 (unsigned long long)file
->f_pos
) {
719 /* to avoid userland mistaking f_pos=-9 as -EBADF=-9 */
720 if ((unsigned long long)offset
>= ~0xFFFULL
) {
724 file
->f_pos
= offset
;
726 force_successful_syscall_return();
731 mutex_unlock(&file
->f_path
.dentry
->d_inode
->i_mutex
);
735 static int open_port(struct inode
* inode
, struct file
* filp
)
737 return capable(CAP_SYS_RAWIO
) ? 0 : -EPERM
;
740 #define zero_lseek null_lseek
741 #define full_lseek null_lseek
742 #define write_zero write_null
743 #define read_full read_zero
744 #define open_mem open_port
745 #define open_kmem open_mem
746 #define open_oldmem open_mem
748 static const struct file_operations mem_fops
= {
749 .llseek
= memory_lseek
,
754 .get_unmapped_area
= get_unmapped_area_mem
,
757 #ifdef CONFIG_DEVKMEM
758 static const struct file_operations kmem_fops
= {
759 .llseek
= memory_lseek
,
764 .get_unmapped_area
= get_unmapped_area_mem
,
768 static const struct file_operations null_fops
= {
769 .llseek
= null_lseek
,
772 .splice_write
= splice_write_null
,
775 #ifdef CONFIG_DEVPORT
776 static const struct file_operations port_fops
= {
777 .llseek
= memory_lseek
,
784 static const struct file_operations zero_fops
= {
785 .llseek
= zero_lseek
,
792 * capabilities for /dev/zero
793 * - permits private mappings, "copies" are taken of the source of zeros
795 static struct backing_dev_info zero_bdi
= {
797 .capabilities
= BDI_CAP_MAP_COPY
,
800 static const struct file_operations full_fops
= {
801 .llseek
= full_lseek
,
806 #ifdef CONFIG_CRASH_DUMP
807 static const struct file_operations oldmem_fops
= {
813 static ssize_t
kmsg_write(struct file
*file
, const char __user
*buf
,
814 size_t count
, loff_t
*ppos
)
819 tmp
= kmalloc(count
+ 1, GFP_KERNEL
);
823 if (!copy_from_user(tmp
, buf
, count
)) {
825 ret
= printk("%s", tmp
);
827 /* printk can add a prefix */
834 static const struct file_operations kmsg_fops
= {
838 static const struct memdev
{
841 const struct file_operations
*fops
;
842 struct backing_dev_info
*dev_info
;
844 [1] = { "mem", 0, &mem_fops
, &directly_mappable_cdev_bdi
},
845 #ifdef CONFIG_DEVKMEM
846 [2] = { "kmem", 0, &kmem_fops
, &directly_mappable_cdev_bdi
},
848 [3] = { "null", 0666, &null_fops
, NULL
},
849 #ifdef CONFIG_DEVPORT
850 [4] = { "port", 0, &port_fops
, NULL
},
852 [5] = { "zero", 0666, &zero_fops
, &zero_bdi
},
853 [7] = { "full", 0666, &full_fops
, NULL
},
854 [8] = { "random", 0666, &random_fops
, NULL
},
855 [9] = { "urandom", 0666, &urandom_fops
, NULL
},
856 [11] = { "kmsg", 0, &kmsg_fops
, NULL
},
857 #ifdef CONFIG_CRASH_DUMP
858 [12] = { "oldmem", 0, &oldmem_fops
, NULL
},
862 static int memory_open(struct inode
*inode
, struct file
*filp
)
865 const struct memdev
*dev
;
867 minor
= iminor(inode
);
868 if (minor
>= ARRAY_SIZE(devlist
))
871 dev
= &devlist
[minor
];
875 filp
->f_op
= dev
->fops
;
877 filp
->f_mapping
->backing_dev_info
= dev
->dev_info
;
880 return dev
->fops
->open(inode
, filp
);
885 static const struct file_operations memory_fops
= {
889 static char *mem_devnode(struct device
*dev
, mode_t
*mode
)
891 if (mode
&& devlist
[MINOR(dev
->devt
)].mode
)
892 *mode
= devlist
[MINOR(dev
->devt
)].mode
;
896 static struct class *mem_class
;
898 static int __init
chr_dev_init(void)
903 err
= bdi_init(&zero_bdi
);
907 if (register_chrdev(MEM_MAJOR
, "mem", &memory_fops
))
908 printk("unable to get major %d for memory devs\n", MEM_MAJOR
);
910 mem_class
= class_create(THIS_MODULE
, "mem");
911 mem_class
->devnode
= mem_devnode
;
912 for (minor
= 1; minor
< ARRAY_SIZE(devlist
); minor
++) {
913 if (!devlist
[minor
].name
)
915 device_create(mem_class
, NULL
, MKDEV(MEM_MAJOR
, minor
),
916 NULL
, devlist
[minor
].name
);
922 fs_initcall(chr_dev_init
);