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>
29 #include <linux/export.h>
31 #include <linux/aio.h>
33 #include <asm/uaccess.h>
36 # include <linux/efi.h>
39 #define DEVPORT_MINOR 4
41 static inline unsigned long size_inside_page(unsigned long start
,
46 sz
= PAGE_SIZE
- (start
& (PAGE_SIZE
- 1));
51 #ifndef ARCH_HAS_VALID_PHYS_ADDR_RANGE
52 static inline int valid_phys_addr_range(phys_addr_t addr
, size_t count
)
54 return addr
+ count
<= __pa(high_memory
);
57 static inline int valid_mmap_phys_addr_range(unsigned long pfn
, size_t size
)
63 #if defined(CONFIG_DEVMEM) || defined(CONFIG_DEVKMEM)
64 #ifdef CONFIG_STRICT_DEVMEM
65 static inline int range_is_allowed(unsigned long pfn
, unsigned long size
)
67 u64 from
= ((u64
)pfn
) << PAGE_SHIFT
;
72 if (!devmem_is_allowed(pfn
)) {
74 "Program %s tried to access /dev/mem between %Lx->%Lx.\n",
75 current
->comm
, from
, to
);
84 static inline int range_is_allowed(unsigned long pfn
, unsigned long size
)
92 void __weak
unxlate_dev_mem_ptr(unsigned long phys
, void *addr
)
97 * This funcion reads the *physical* memory. The f_pos points directly to the
100 static ssize_t
read_mem(struct file
*file
, char __user
*buf
,
101 size_t count
, loff_t
*ppos
)
103 phys_addr_t p
= *ppos
;
107 if (!valid_phys_addr_range(p
, count
))
110 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
111 /* we don't have page 0 mapped on sparc and m68k.. */
113 sz
= size_inside_page(p
, count
);
115 if (clear_user(buf
, sz
))
126 unsigned long remaining
;
128 sz
= size_inside_page(p
, count
);
130 if (!range_is_allowed(p
>> PAGE_SHIFT
, count
))
134 * On ia64 if a page has been mapped somewhere as uncached, then
135 * it must also be accessed uncached by the kernel or data
136 * corruption may occur.
138 ptr
= xlate_dev_mem_ptr(p
);
142 remaining
= copy_to_user(buf
, ptr
, sz
);
143 unxlate_dev_mem_ptr(p
, ptr
);
157 static ssize_t
write_mem(struct file
*file
, const char __user
*buf
,
158 size_t count
, loff_t
*ppos
)
160 phys_addr_t p
= *ppos
;
162 unsigned long copied
;
165 if (!valid_phys_addr_range(p
, count
))
170 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
171 /* we don't have page 0 mapped on sparc and m68k.. */
173 sz
= size_inside_page(p
, count
);
174 /* Hmm. Do something? */
183 sz
= size_inside_page(p
, count
);
185 if (!range_is_allowed(p
>> PAGE_SHIFT
, sz
))
189 * On ia64 if a page has been mapped somewhere as uncached, then
190 * it must also be accessed uncached by the kernel or data
191 * corruption may occur.
193 ptr
= xlate_dev_mem_ptr(p
);
200 copied
= copy_from_user(ptr
, buf
, sz
);
201 unxlate_dev_mem_ptr(p
, ptr
);
203 written
+= sz
- copied
;
218 #endif /* CONFIG_DEVMEM */
220 #if defined(CONFIG_DEVMEM) || defined(CONFIG_DEVKMEM)
222 int __weak
phys_mem_access_prot_allowed(struct file
*file
,
223 unsigned long pfn
, unsigned long size
, pgprot_t
*vma_prot
)
228 #ifndef __HAVE_PHYS_MEM_ACCESS_PROT
231 * Architectures vary in how they handle caching for addresses
232 * outside of main memory.
235 #ifdef pgprot_noncached
236 static int uncached_access(struct file
*file
, phys_addr_t addr
)
238 #if defined(CONFIG_IA64)
240 * On ia64, we ignore O_DSYNC because we cannot tolerate memory
243 return !(efi_mem_attributes(addr
) & EFI_MEMORY_WB
);
244 #elif defined(CONFIG_MIPS)
246 extern int __uncached_access(struct file
*file
,
249 return __uncached_access(file
, addr
);
253 * Accessing memory above the top the kernel knows about or through a
255 * that was marked O_DSYNC will be done non-cached.
257 if (file
->f_flags
& O_DSYNC
)
259 return addr
>= __pa(high_memory
);
264 static pgprot_t
phys_mem_access_prot(struct file
*file
, unsigned long pfn
,
265 unsigned long size
, pgprot_t vma_prot
)
267 #ifdef pgprot_noncached
268 phys_addr_t offset
= pfn
<< PAGE_SHIFT
;
270 if (uncached_access(file
, offset
))
271 return pgprot_noncached(vma_prot
);
278 static unsigned long get_unmapped_area_mem(struct file
*file
,
284 if (!valid_mmap_phys_addr_range(pgoff
, len
))
285 return (unsigned long) -EINVAL
;
286 return pgoff
<< PAGE_SHIFT
;
289 /* can't do an in-place private mapping if there's no MMU */
290 static inline int private_mapping_ok(struct vm_area_struct
*vma
)
292 return vma
->vm_flags
& VM_MAYSHARE
;
295 #define get_unmapped_area_mem NULL
297 static inline int private_mapping_ok(struct vm_area_struct
*vma
)
303 static const struct vm_operations_struct mmap_mem_ops
= {
304 #ifdef CONFIG_HAVE_IOREMAP_PROT
305 .access
= generic_access_phys
309 static int mmap_mem(struct file
*file
, struct vm_area_struct
*vma
)
311 size_t size
= vma
->vm_end
- vma
->vm_start
;
313 if (!valid_mmap_phys_addr_range(vma
->vm_pgoff
, size
))
316 if (!private_mapping_ok(vma
))
319 if (!range_is_allowed(vma
->vm_pgoff
, size
))
322 if (!phys_mem_access_prot_allowed(file
, vma
->vm_pgoff
, size
,
326 vma
->vm_page_prot
= phys_mem_access_prot(file
, vma
->vm_pgoff
,
330 vma
->vm_ops
= &mmap_mem_ops
;
332 /* Remap-pfn-range will mark the range VM_IO */
333 if (remap_pfn_range(vma
,
337 vma
->vm_page_prot
)) {
342 #endif /* CONFIG_DEVMEM */
344 #ifdef CONFIG_DEVKMEM
345 static int mmap_kmem(struct file
*file
, struct vm_area_struct
*vma
)
349 /* Turn a kernel-virtual address into a physical page frame */
350 pfn
= __pa((u64
)vma
->vm_pgoff
<< PAGE_SHIFT
) >> PAGE_SHIFT
;
353 * RED-PEN: on some architectures there is more mapped memory than
354 * available in mem_map which pfn_valid checks for. Perhaps should add a
357 * RED-PEN: vmalloc is not supported right now.
363 return mmap_mem(file
, vma
);
367 #ifdef CONFIG_CRASH_DUMP
369 * Read memory corresponding to the old kernel.
371 static ssize_t
read_oldmem(struct file
*file
, char __user
*buf
,
372 size_t count
, loff_t
*ppos
)
374 unsigned long pfn
, offset
;
375 size_t read
= 0, csize
;
379 pfn
= *ppos
/ PAGE_SIZE
;
380 if (pfn
> saved_max_pfn
)
383 offset
= (unsigned long)(*ppos
% PAGE_SIZE
);
384 if (count
> PAGE_SIZE
- offset
)
385 csize
= PAGE_SIZE
- offset
;
389 rc
= copy_oldmem_page(pfn
, buf
, csize
, offset
, 1);
401 #ifdef CONFIG_DEVKMEM
403 * This function reads the *virtual* memory as seen by the kernel.
405 static ssize_t
read_kmem(struct file
*file
, char __user
*buf
,
406 size_t count
, loff_t
*ppos
)
408 unsigned long p
= *ppos
;
409 ssize_t low_count
, read
, sz
;
410 char *kbuf
; /* k-addr because vread() takes vmlist_lock rwlock */
414 if (p
< (unsigned long) high_memory
) {
416 if (count
> (unsigned long)high_memory
- p
)
417 low_count
= (unsigned long)high_memory
- p
;
419 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
420 /* we don't have page 0 mapped on sparc and m68k.. */
421 if (p
< PAGE_SIZE
&& low_count
> 0) {
422 sz
= size_inside_page(p
, low_count
);
423 if (clear_user(buf
, sz
))
432 while (low_count
> 0) {
433 sz
= size_inside_page(p
, low_count
);
436 * On ia64 if a page has been mapped somewhere as
437 * uncached, then it must also be accessed uncached
438 * by the kernel or data corruption may occur
440 kbuf
= xlate_dev_kmem_ptr((char *)p
);
442 if (copy_to_user(buf
, kbuf
, sz
))
453 kbuf
= (char *)__get_free_page(GFP_KERNEL
);
457 sz
= size_inside_page(p
, count
);
458 if (!is_vmalloc_or_module_addr((void *)p
)) {
462 sz
= vread(kbuf
, (char *)p
, sz
);
465 if (copy_to_user(buf
, kbuf
, sz
)) {
474 free_page((unsigned long)kbuf
);
477 return read
? read
: err
;
481 static ssize_t
do_write_kmem(unsigned long p
, const char __user
*buf
,
482 size_t count
, loff_t
*ppos
)
485 unsigned long copied
;
488 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
489 /* we don't have page 0 mapped on sparc and m68k.. */
491 sz
= size_inside_page(p
, count
);
492 /* Hmm. Do something? */
503 sz
= size_inside_page(p
, count
);
506 * On ia64 if a page has been mapped somewhere as uncached, then
507 * it must also be accessed uncached by the kernel or data
508 * corruption may occur.
510 ptr
= xlate_dev_kmem_ptr((char *)p
);
512 copied
= copy_from_user(ptr
, buf
, sz
);
514 written
+= sz
- copied
;
530 * This function writes to the *virtual* memory as seen by the kernel.
532 static ssize_t
write_kmem(struct file
*file
, const char __user
*buf
,
533 size_t count
, loff_t
*ppos
)
535 unsigned long p
= *ppos
;
538 char *kbuf
; /* k-addr because vwrite() takes vmlist_lock rwlock */
541 if (p
< (unsigned long) high_memory
) {
542 unsigned long to_write
= min_t(unsigned long, count
,
543 (unsigned long)high_memory
- p
);
544 wrote
= do_write_kmem(p
, buf
, to_write
, ppos
);
545 if (wrote
!= to_write
)
553 kbuf
= (char *)__get_free_page(GFP_KERNEL
);
555 return wrote
? wrote
: -ENOMEM
;
557 unsigned long sz
= size_inside_page(p
, count
);
560 if (!is_vmalloc_or_module_addr((void *)p
)) {
564 n
= copy_from_user(kbuf
, buf
, sz
);
569 vwrite(kbuf
, (char *)p
, sz
);
575 free_page((unsigned long)kbuf
);
579 return virtr
+ wrote
? : err
;
583 #ifdef CONFIG_DEVPORT
584 static ssize_t
read_port(struct file
*file
, char __user
*buf
,
585 size_t count
, loff_t
*ppos
)
587 unsigned long i
= *ppos
;
588 char __user
*tmp
= buf
;
590 if (!access_ok(VERIFY_WRITE
, buf
, count
))
592 while (count
-- > 0 && i
< 65536) {
593 if (__put_user(inb(i
), tmp
) < 0)
602 static ssize_t
write_port(struct file
*file
, const char __user
*buf
,
603 size_t count
, loff_t
*ppos
)
605 unsigned long i
= *ppos
;
606 const char __user
*tmp
= buf
;
608 if (!access_ok(VERIFY_READ
, buf
, count
))
610 while (count
-- > 0 && i
< 65536) {
612 if (__get_user(c
, tmp
)) {
626 static ssize_t
read_null(struct file
*file
, char __user
*buf
,
627 size_t count
, loff_t
*ppos
)
632 static ssize_t
write_null(struct file
*file
, const char __user
*buf
,
633 size_t count
, loff_t
*ppos
)
638 static ssize_t
aio_read_null(struct kiocb
*iocb
, const struct iovec
*iov
,
639 unsigned long nr_segs
, loff_t pos
)
644 static ssize_t
aio_write_null(struct kiocb
*iocb
, const struct iovec
*iov
,
645 unsigned long nr_segs
, loff_t pos
)
647 return iov_length(iov
, nr_segs
);
650 static int pipe_to_null(struct pipe_inode_info
*info
, struct pipe_buffer
*buf
,
651 struct splice_desc
*sd
)
656 static ssize_t
splice_write_null(struct pipe_inode_info
*pipe
, struct file
*out
,
657 loff_t
*ppos
, size_t len
, unsigned int flags
)
659 return splice_from_pipe(pipe
, out
, ppos
, len
, flags
, pipe_to_null
);
662 static ssize_t
read_zero(struct file
*file
, char __user
*buf
,
663 size_t count
, loff_t
*ppos
)
670 if (!access_ok(VERIFY_WRITE
, buf
, count
))
675 unsigned long unwritten
;
676 size_t chunk
= count
;
678 if (chunk
> PAGE_SIZE
)
679 chunk
= PAGE_SIZE
; /* Just for latency reasons */
680 unwritten
= __clear_user(buf
, chunk
);
681 written
+= chunk
- unwritten
;
684 if (signal_pending(current
))
685 return written
? written
: -ERESTARTSYS
;
690 return written
? written
: -EFAULT
;
693 static ssize_t
aio_read_zero(struct kiocb
*iocb
, const struct iovec
*iov
,
694 unsigned long nr_segs
, loff_t pos
)
700 for (i
= 0; i
< nr_segs
; i
++) {
701 ret
= read_zero(iocb
->ki_filp
, iov
[i
].iov_base
, iov
[i
].iov_len
,
708 return written
? written
: -EFAULT
;
711 static int mmap_zero(struct file
*file
, struct vm_area_struct
*vma
)
716 if (vma
->vm_flags
& VM_SHARED
)
717 return shmem_zero_setup(vma
);
721 static ssize_t
write_full(struct file
*file
, const char __user
*buf
,
722 size_t count
, loff_t
*ppos
)
728 * Special lseek() function for /dev/null and /dev/zero. Most notably, you
729 * can fopen() both devices with "a" now. This was previously impossible.
732 static loff_t
null_lseek(struct file
*file
, loff_t offset
, int orig
)
734 return file
->f_pos
= 0;
737 #if defined(CONFIG_DEVMEM) || defined(CONFIG_DEVKMEM) || defined(CONFIG_DEVPORT)
740 * The memory devices use the full 32/64 bits of the offset, and so we cannot
741 * check against negative addresses: they are ok. The return value is weird,
742 * though, in that case (0).
744 * also note that seeking relative to the "end of file" isn't supported:
745 * it has no meaning, so it returns -EINVAL.
747 static loff_t
memory_lseek(struct file
*file
, loff_t offset
, int orig
)
751 mutex_lock(&file_inode(file
)->i_mutex
);
754 offset
+= file
->f_pos
;
756 /* to avoid userland mistaking f_pos=-9 as -EBADF=-9 */
757 if ((unsigned long long)offset
>= ~0xFFFULL
) {
761 file
->f_pos
= offset
;
763 force_successful_syscall_return();
768 mutex_unlock(&file_inode(file
)->i_mutex
);
774 #if defined(CONFIG_DEVMEM) || defined(CONFIG_DEVKMEM) || defined(CONFIG_DEVPORT)
775 static int open_port(struct inode
*inode
, struct file
*filp
)
777 return capable(CAP_SYS_RAWIO
) ? 0 : -EPERM
;
781 #define zero_lseek null_lseek
782 #define full_lseek null_lseek
783 #define write_zero write_null
784 #define read_full read_zero
785 #define aio_write_zero aio_write_null
786 #define open_mem open_port
787 #define open_kmem open_mem
788 #define open_oldmem open_mem
791 static const struct file_operations mem_fops
= {
792 .llseek
= memory_lseek
,
797 .get_unmapped_area
= get_unmapped_area_mem
,
801 #ifdef CONFIG_DEVKMEM
802 static const struct file_operations kmem_fops
= {
803 .llseek
= memory_lseek
,
808 .get_unmapped_area
= get_unmapped_area_mem
,
812 static const struct file_operations null_fops
= {
813 .llseek
= null_lseek
,
816 .aio_read
= aio_read_null
,
817 .aio_write
= aio_write_null
,
818 .splice_write
= splice_write_null
,
821 #ifdef CONFIG_DEVPORT
822 static const struct file_operations port_fops
= {
823 .llseek
= memory_lseek
,
830 static const struct file_operations zero_fops
= {
831 .llseek
= zero_lseek
,
834 .aio_read
= aio_read_zero
,
835 .aio_write
= aio_write_zero
,
840 * capabilities for /dev/zero
841 * - permits private mappings, "copies" are taken of the source of zeros
842 * - no writeback happens
844 static struct backing_dev_info zero_bdi
= {
846 .capabilities
= BDI_CAP_MAP_COPY
| BDI_CAP_NO_ACCT_AND_WRITEBACK
,
849 static const struct file_operations full_fops
= {
850 .llseek
= full_lseek
,
855 #ifdef CONFIG_CRASH_DUMP
856 static const struct file_operations oldmem_fops
= {
859 .llseek
= default_llseek
,
863 static const struct memdev
{
866 const struct file_operations
*fops
;
867 struct backing_dev_info
*dev_info
;
870 [1] = { "mem", 0, &mem_fops
, &directly_mappable_cdev_bdi
},
872 #ifdef CONFIG_DEVKMEM
873 [2] = { "kmem", 0, &kmem_fops
, &directly_mappable_cdev_bdi
},
875 [3] = { "null", 0666, &null_fops
, NULL
},
876 #ifdef CONFIG_DEVPORT
877 [4] = { "port", 0, &port_fops
, NULL
},
879 [5] = { "zero", 0666, &zero_fops
, &zero_bdi
},
880 [7] = { "full", 0666, &full_fops
, NULL
},
881 [8] = { "random", 0666, &random_fops
, NULL
},
882 [9] = { "urandom", 0666, &urandom_fops
, NULL
},
884 [11] = { "kmsg", 0644, &kmsg_fops
, NULL
},
886 #ifdef CONFIG_CRASH_DUMP
887 [12] = { "oldmem", 0, &oldmem_fops
, NULL
},
891 static int memory_open(struct inode
*inode
, struct file
*filp
)
894 const struct memdev
*dev
;
896 minor
= iminor(inode
);
897 if (minor
>= ARRAY_SIZE(devlist
))
900 dev
= &devlist
[minor
];
904 filp
->f_op
= dev
->fops
;
906 filp
->f_mapping
->backing_dev_info
= dev
->dev_info
;
908 /* Is /dev/mem or /dev/kmem ? */
909 if (dev
->dev_info
== &directly_mappable_cdev_bdi
)
910 filp
->f_mode
|= FMODE_UNSIGNED_OFFSET
;
913 return dev
->fops
->open(inode
, filp
);
918 static const struct file_operations memory_fops
= {
920 .llseek
= noop_llseek
,
923 static char *mem_devnode(struct device
*dev
, umode_t
*mode
)
925 if (mode
&& devlist
[MINOR(dev
->devt
)].mode
)
926 *mode
= devlist
[MINOR(dev
->devt
)].mode
;
930 static struct class *mem_class
;
932 static int __init
chr_dev_init(void)
937 err
= bdi_init(&zero_bdi
);
941 if (register_chrdev(MEM_MAJOR
, "mem", &memory_fops
))
942 printk("unable to get major %d for memory devs\n", MEM_MAJOR
);
944 mem_class
= class_create(THIS_MODULE
, "mem");
945 if (IS_ERR(mem_class
))
946 return PTR_ERR(mem_class
);
948 mem_class
->devnode
= mem_devnode
;
949 for (minor
= 1; minor
< ARRAY_SIZE(devlist
); minor
++) {
950 if (!devlist
[minor
].name
)
956 if ((minor
== DEVPORT_MINOR
) && !arch_has_dev_port())
959 device_create(mem_class
, NULL
, MKDEV(MEM_MAJOR
, minor
),
960 NULL
, devlist
[minor
].name
);
966 fs_initcall(chr_dev_init
);