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/backing-dev.h>
25 #include <linux/splice.h>
26 #include <linux/pfn.h>
27 #include <linux/export.h>
29 #include <linux/aio.h>
31 #include <asm/uaccess.h>
34 # include <linux/efi.h>
37 #define DEVPORT_MINOR 4
39 static inline unsigned long size_inside_page(unsigned long start
,
44 sz
= PAGE_SIZE
- (start
& (PAGE_SIZE
- 1));
49 #ifndef ARCH_HAS_VALID_PHYS_ADDR_RANGE
50 static inline int valid_phys_addr_range(phys_addr_t addr
, size_t count
)
52 return addr
+ count
<= __pa(high_memory
);
55 static inline int valid_mmap_phys_addr_range(unsigned long pfn
, size_t size
)
61 #ifdef CONFIG_STRICT_DEVMEM
62 static inline int range_is_allowed(unsigned long pfn
, unsigned long size
)
64 u64 from
= ((u64
)pfn
) << PAGE_SHIFT
;
69 if (!devmem_is_allowed(pfn
)) {
71 "Program %s tried to access /dev/mem between %Lx->%Lx.\n",
72 current
->comm
, from
, to
);
81 static inline int range_is_allowed(unsigned long pfn
, unsigned long size
)
87 void __weak
unxlate_dev_mem_ptr(unsigned long phys
, void *addr
)
92 * This funcion reads the *physical* memory. The f_pos points directly to the
95 static ssize_t
read_mem(struct file
*file
, char __user
*buf
,
96 size_t count
, loff_t
*ppos
)
98 phys_addr_t p
= *ppos
;
105 if (!valid_phys_addr_range(p
, count
))
108 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
109 /* we don't have page 0 mapped on sparc and m68k.. */
111 sz
= size_inside_page(p
, count
);
113 if (clear_user(buf
, sz
))
124 unsigned long remaining
;
126 sz
= size_inside_page(p
, count
);
128 if (!range_is_allowed(p
>> PAGE_SHIFT
, count
))
132 * On ia64 if a page has been mapped somewhere as uncached, then
133 * it must also be accessed uncached by the kernel or data
134 * corruption may occur.
136 ptr
= xlate_dev_mem_ptr(p
);
140 remaining
= copy_to_user(buf
, ptr
, sz
);
141 unxlate_dev_mem_ptr(p
, ptr
);
155 static ssize_t
write_mem(struct file
*file
, const char __user
*buf
,
156 size_t count
, loff_t
*ppos
)
158 phys_addr_t p
= *ppos
;
160 unsigned long copied
;
166 if (!valid_phys_addr_range(p
, count
))
171 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
172 /* we don't have page 0 mapped on sparc and m68k.. */
174 sz
= size_inside_page(p
, count
);
175 /* Hmm. Do something? */
184 sz
= size_inside_page(p
, count
);
186 if (!range_is_allowed(p
>> PAGE_SHIFT
, sz
))
190 * On ia64 if a page has been mapped somewhere as uncached, then
191 * it must also be accessed uncached by the kernel or data
192 * corruption may occur.
194 ptr
= xlate_dev_mem_ptr(p
);
201 copied
= copy_from_user(ptr
, buf
, sz
);
202 unxlate_dev_mem_ptr(p
, ptr
);
204 written
+= sz
- copied
;
220 int __weak
phys_mem_access_prot_allowed(struct file
*file
,
221 unsigned long pfn
, unsigned long size
, pgprot_t
*vma_prot
)
226 #ifndef __HAVE_PHYS_MEM_ACCESS_PROT
229 * Architectures vary in how they handle caching for addresses
230 * outside of main memory.
233 #ifdef pgprot_noncached
234 static int uncached_access(struct file
*file
, phys_addr_t addr
)
236 #if defined(CONFIG_IA64)
238 * On ia64, we ignore O_DSYNC because we cannot tolerate memory
241 return !(efi_mem_attributes(addr
) & EFI_MEMORY_WB
);
242 #elif defined(CONFIG_MIPS)
244 extern int __uncached_access(struct file
*file
,
247 return __uncached_access(file
, addr
);
251 * Accessing memory above the top the kernel knows about or through a
253 * that was marked O_DSYNC will be done non-cached.
255 if (file
->f_flags
& O_DSYNC
)
257 return addr
>= __pa(high_memory
);
262 static pgprot_t
phys_mem_access_prot(struct file
*file
, unsigned long pfn
,
263 unsigned long size
, pgprot_t vma_prot
)
265 #ifdef pgprot_noncached
266 phys_addr_t offset
= pfn
<< PAGE_SHIFT
;
268 if (uncached_access(file
, offset
))
269 return pgprot_noncached(vma_prot
);
276 static unsigned long get_unmapped_area_mem(struct file
*file
,
282 if (!valid_mmap_phys_addr_range(pgoff
, len
))
283 return (unsigned long) -EINVAL
;
284 return pgoff
<< PAGE_SHIFT
;
287 /* can't do an in-place private mapping if there's no MMU */
288 static inline int private_mapping_ok(struct vm_area_struct
*vma
)
290 return vma
->vm_flags
& VM_MAYSHARE
;
293 #define get_unmapped_area_mem NULL
295 static inline int private_mapping_ok(struct vm_area_struct
*vma
)
301 static const struct vm_operations_struct mmap_mem_ops
= {
302 #ifdef CONFIG_HAVE_IOREMAP_PROT
303 .access
= generic_access_phys
307 static int mmap_mem(struct file
*file
, struct vm_area_struct
*vma
)
309 size_t size
= vma
->vm_end
- vma
->vm_start
;
311 if (!valid_mmap_phys_addr_range(vma
->vm_pgoff
, size
))
314 if (!private_mapping_ok(vma
))
317 if (!range_is_allowed(vma
->vm_pgoff
, size
))
320 if (!phys_mem_access_prot_allowed(file
, vma
->vm_pgoff
, size
,
324 vma
->vm_page_prot
= phys_mem_access_prot(file
, vma
->vm_pgoff
,
328 vma
->vm_ops
= &mmap_mem_ops
;
330 /* Remap-pfn-range will mark the range VM_IO */
331 if (remap_pfn_range(vma
,
335 vma
->vm_page_prot
)) {
341 #ifdef CONFIG_DEVKMEM
342 static int mmap_kmem(struct file
*file
, struct vm_area_struct
*vma
)
346 /* Turn a kernel-virtual address into a physical page frame */
347 pfn
= __pa((u64
)vma
->vm_pgoff
<< PAGE_SHIFT
) >> PAGE_SHIFT
;
350 * RED-PEN: on some architectures there is more mapped memory than
351 * available in mem_map which pfn_valid checks for. Perhaps should add a
354 * RED-PEN: vmalloc is not supported right now.
360 return mmap_mem(file
, vma
);
364 #ifdef CONFIG_DEVKMEM
366 * This function reads the *virtual* memory as seen by the kernel.
368 static ssize_t
read_kmem(struct file
*file
, char __user
*buf
,
369 size_t count
, loff_t
*ppos
)
371 unsigned long p
= *ppos
;
372 ssize_t low_count
, read
, sz
;
373 char *kbuf
; /* k-addr because vread() takes vmlist_lock rwlock */
377 if (p
< (unsigned long) high_memory
) {
379 if (count
> (unsigned long)high_memory
- p
)
380 low_count
= (unsigned long)high_memory
- p
;
382 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
383 /* we don't have page 0 mapped on sparc and m68k.. */
384 if (p
< PAGE_SIZE
&& low_count
> 0) {
385 sz
= size_inside_page(p
, low_count
);
386 if (clear_user(buf
, sz
))
395 while (low_count
> 0) {
396 sz
= size_inside_page(p
, low_count
);
399 * On ia64 if a page has been mapped somewhere as
400 * uncached, then it must also be accessed uncached
401 * by the kernel or data corruption may occur
403 kbuf
= xlate_dev_kmem_ptr((char *)p
);
405 if (copy_to_user(buf
, kbuf
, sz
))
416 kbuf
= (char *)__get_free_page(GFP_KERNEL
);
420 sz
= size_inside_page(p
, count
);
421 if (!is_vmalloc_or_module_addr((void *)p
)) {
425 sz
= vread(kbuf
, (char *)p
, sz
);
428 if (copy_to_user(buf
, kbuf
, sz
)) {
437 free_page((unsigned long)kbuf
);
440 return read
? read
: err
;
444 static ssize_t
do_write_kmem(unsigned long p
, const char __user
*buf
,
445 size_t count
, loff_t
*ppos
)
448 unsigned long copied
;
451 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
452 /* we don't have page 0 mapped on sparc and m68k.. */
454 sz
= size_inside_page(p
, count
);
455 /* Hmm. Do something? */
466 sz
= size_inside_page(p
, count
);
469 * On ia64 if a page has been mapped somewhere as uncached, then
470 * it must also be accessed uncached by the kernel or data
471 * corruption may occur.
473 ptr
= xlate_dev_kmem_ptr((char *)p
);
475 copied
= copy_from_user(ptr
, buf
, sz
);
477 written
+= sz
- copied
;
493 * This function writes to the *virtual* memory as seen by the kernel.
495 static ssize_t
write_kmem(struct file
*file
, const char __user
*buf
,
496 size_t count
, loff_t
*ppos
)
498 unsigned long p
= *ppos
;
501 char *kbuf
; /* k-addr because vwrite() takes vmlist_lock rwlock */
504 if (p
< (unsigned long) high_memory
) {
505 unsigned long to_write
= min_t(unsigned long, count
,
506 (unsigned long)high_memory
- p
);
507 wrote
= do_write_kmem(p
, buf
, to_write
, ppos
);
508 if (wrote
!= to_write
)
516 kbuf
= (char *)__get_free_page(GFP_KERNEL
);
518 return wrote
? wrote
: -ENOMEM
;
520 unsigned long sz
= size_inside_page(p
, count
);
523 if (!is_vmalloc_or_module_addr((void *)p
)) {
527 n
= copy_from_user(kbuf
, buf
, sz
);
532 vwrite(kbuf
, (char *)p
, sz
);
538 free_page((unsigned long)kbuf
);
542 return virtr
+ wrote
? : err
;
546 #ifdef CONFIG_DEVPORT
547 static ssize_t
read_port(struct file
*file
, char __user
*buf
,
548 size_t count
, loff_t
*ppos
)
550 unsigned long i
= *ppos
;
551 char __user
*tmp
= buf
;
553 if (!access_ok(VERIFY_WRITE
, buf
, count
))
555 while (count
-- > 0 && i
< 65536) {
556 if (__put_user(inb(i
), tmp
) < 0)
565 static ssize_t
write_port(struct file
*file
, const char __user
*buf
,
566 size_t count
, loff_t
*ppos
)
568 unsigned long i
= *ppos
;
569 const char __user
*tmp
= buf
;
571 if (!access_ok(VERIFY_READ
, buf
, count
))
573 while (count
-- > 0 && i
< 65536) {
575 if (__get_user(c
, tmp
)) {
589 static ssize_t
read_null(struct file
*file
, char __user
*buf
,
590 size_t count
, loff_t
*ppos
)
595 static ssize_t
write_null(struct file
*file
, const char __user
*buf
,
596 size_t count
, loff_t
*ppos
)
601 static ssize_t
aio_read_null(struct kiocb
*iocb
, const struct iovec
*iov
,
602 unsigned long nr_segs
, loff_t pos
)
607 static ssize_t
aio_write_null(struct kiocb
*iocb
, const struct iovec
*iov
,
608 unsigned long nr_segs
, loff_t pos
)
610 return iov_length(iov
, nr_segs
);
613 static int pipe_to_null(struct pipe_inode_info
*info
, struct pipe_buffer
*buf
,
614 struct splice_desc
*sd
)
619 static ssize_t
splice_write_null(struct pipe_inode_info
*pipe
, struct file
*out
,
620 loff_t
*ppos
, size_t len
, unsigned int flags
)
622 return splice_from_pipe(pipe
, out
, ppos
, len
, flags
, pipe_to_null
);
625 static ssize_t
read_zero(struct file
*file
, char __user
*buf
,
626 size_t count
, loff_t
*ppos
)
633 if (!access_ok(VERIFY_WRITE
, buf
, count
))
638 unsigned long unwritten
;
639 size_t chunk
= count
;
641 if (chunk
> PAGE_SIZE
)
642 chunk
= PAGE_SIZE
; /* Just for latency reasons */
643 unwritten
= __clear_user(buf
, chunk
);
644 written
+= chunk
- unwritten
;
647 if (signal_pending(current
))
648 return written
? written
: -ERESTARTSYS
;
653 return written
? written
: -EFAULT
;
656 static ssize_t
aio_read_zero(struct kiocb
*iocb
, const struct iovec
*iov
,
657 unsigned long nr_segs
, loff_t pos
)
663 for (i
= 0; i
< nr_segs
; i
++) {
664 ret
= read_zero(iocb
->ki_filp
, iov
[i
].iov_base
, iov
[i
].iov_len
,
671 return written
? written
: -EFAULT
;
674 static int mmap_zero(struct file
*file
, struct vm_area_struct
*vma
)
679 if (vma
->vm_flags
& VM_SHARED
)
680 return shmem_zero_setup(vma
);
684 static ssize_t
write_full(struct file
*file
, const char __user
*buf
,
685 size_t count
, loff_t
*ppos
)
691 * Special lseek() function for /dev/null and /dev/zero. Most notably, you
692 * can fopen() both devices with "a" now. This was previously impossible.
695 static loff_t
null_lseek(struct file
*file
, loff_t offset
, int orig
)
697 return file
->f_pos
= 0;
701 * The memory devices use the full 32/64 bits of the offset, and so we cannot
702 * check against negative addresses: they are ok. The return value is weird,
703 * though, in that case (0).
705 * also note that seeking relative to the "end of file" isn't supported:
706 * it has no meaning, so it returns -EINVAL.
708 static loff_t
memory_lseek(struct file
*file
, loff_t offset
, int orig
)
712 mutex_lock(&file_inode(file
)->i_mutex
);
715 offset
+= file
->f_pos
;
717 /* to avoid userland mistaking f_pos=-9 as -EBADF=-9 */
718 if (IS_ERR_VALUE((unsigned long long)offset
)) {
722 file
->f_pos
= offset
;
724 force_successful_syscall_return();
729 mutex_unlock(&file_inode(file
)->i_mutex
);
733 static int open_port(struct inode
*inode
, struct file
*filp
)
735 return capable(CAP_SYS_RAWIO
) ? 0 : -EPERM
;
738 #define zero_lseek null_lseek
739 #define full_lseek null_lseek
740 #define write_zero write_null
741 #define read_full read_zero
742 #define aio_write_zero aio_write_null
743 #define open_mem open_port
744 #define open_kmem open_mem
746 static const struct file_operations mem_fops
= {
747 .llseek
= memory_lseek
,
752 .get_unmapped_area
= get_unmapped_area_mem
,
755 #ifdef CONFIG_DEVKMEM
756 static const struct file_operations kmem_fops
= {
757 .llseek
= memory_lseek
,
762 .get_unmapped_area
= get_unmapped_area_mem
,
766 static const struct file_operations null_fops
= {
767 .llseek
= null_lseek
,
770 .aio_read
= aio_read_null
,
771 .aio_write
= aio_write_null
,
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
,
788 .aio_read
= aio_read_zero
,
789 .aio_write
= aio_write_zero
,
794 * capabilities for /dev/zero
795 * - permits private mappings, "copies" are taken of the source of zeros
796 * - no writeback happens
798 static struct backing_dev_info zero_bdi
= {
800 .capabilities
= BDI_CAP_MAP_COPY
| BDI_CAP_NO_ACCT_AND_WRITEBACK
,
803 static const struct file_operations full_fops
= {
804 .llseek
= full_lseek
,
809 static const struct memdev
{
812 const struct file_operations
*fops
;
813 struct backing_dev_info
*dev_info
;
815 [1] = { "mem", 0, &mem_fops
, &directly_mappable_cdev_bdi
},
816 #ifdef CONFIG_DEVKMEM
817 [2] = { "kmem", 0, &kmem_fops
, &directly_mappable_cdev_bdi
},
819 [3] = { "null", 0666, &null_fops
, NULL
},
820 #ifdef CONFIG_DEVPORT
821 [4] = { "port", 0, &port_fops
, NULL
},
823 [5] = { "zero", 0666, &zero_fops
, &zero_bdi
},
824 [7] = { "full", 0666, &full_fops
, NULL
},
825 [8] = { "random", 0666, &random_fops
, NULL
},
826 [9] = { "urandom", 0666, &urandom_fops
, NULL
},
828 [11] = { "kmsg", 0644, &kmsg_fops
, NULL
},
832 static int memory_open(struct inode
*inode
, struct file
*filp
)
835 const struct memdev
*dev
;
837 minor
= iminor(inode
);
838 if (minor
>= ARRAY_SIZE(devlist
))
841 dev
= &devlist
[minor
];
845 filp
->f_op
= dev
->fops
;
847 filp
->f_mapping
->backing_dev_info
= dev
->dev_info
;
849 /* Is /dev/mem or /dev/kmem ? */
850 if (dev
->dev_info
== &directly_mappable_cdev_bdi
)
851 filp
->f_mode
|= FMODE_UNSIGNED_OFFSET
;
854 return dev
->fops
->open(inode
, filp
);
859 static const struct file_operations memory_fops
= {
861 .llseek
= noop_llseek
,
864 static char *mem_devnode(struct device
*dev
, umode_t
*mode
)
866 if (mode
&& devlist
[MINOR(dev
->devt
)].mode
)
867 *mode
= devlist
[MINOR(dev
->devt
)].mode
;
871 static struct class *mem_class
;
873 static int __init
chr_dev_init(void)
878 err
= bdi_init(&zero_bdi
);
882 if (register_chrdev(MEM_MAJOR
, "mem", &memory_fops
))
883 printk("unable to get major %d for memory devs\n", MEM_MAJOR
);
885 mem_class
= class_create(THIS_MODULE
, "mem");
886 if (IS_ERR(mem_class
))
887 return PTR_ERR(mem_class
);
889 mem_class
->devnode
= mem_devnode
;
890 for (minor
= 1; minor
< ARRAY_SIZE(devlist
); minor
++) {
891 if (!devlist
[minor
].name
)
897 if ((minor
== DEVPORT_MINOR
) && !arch_has_dev_port())
900 device_create(mem_class
, NULL
, MKDEV(MEM_MAJOR
, minor
),
901 NULL
, devlist
[minor
].name
);
907 fs_initcall(chr_dev_init
);