1 // SPDX-License-Identifier: GPL-2.0
3 * linux/drivers/char/mem.c
5 * Copyright (C) 1991, 1992 Linus Torvalds
8 * Jan-11-1998, C. Scott Ananian <cananian@alumni.princeton.edu>
9 * Shared /dev/zero mmapping support, Feb 2000, Kanoj Sarcar <kanoj@sgi.com>
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/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/shmem_fs.h>
26 #include <linux/splice.h>
27 #include <linux/pfn.h>
28 #include <linux/export.h>
30 #include <linux/uio.h>
31 #include <linux/uaccess.h>
32 #include <linux/security.h>
34 #define DEVMEM_MINOR 1
35 #define DEVPORT_MINOR 4
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(phys_addr_t addr
, size_t count
)
50 return addr
+ count
<= __pa(high_memory
);
53 static inline int valid_mmap_phys_addr_range(unsigned long pfn
, size_t size
)
59 #ifdef CONFIG_STRICT_DEVMEM
60 static inline int page_is_allowed(unsigned long pfn
)
62 return devmem_is_allowed(pfn
);
64 static inline int range_is_allowed(unsigned long pfn
, unsigned long size
)
66 u64 from
= ((u64
)pfn
) << PAGE_SHIFT
;
71 if (!devmem_is_allowed(pfn
))
79 static inline int page_is_allowed(unsigned long pfn
)
83 static inline int range_is_allowed(unsigned long pfn
, unsigned long size
)
89 static inline bool should_stop_iteration(void)
93 return signal_pending(current
);
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
;
112 if (!valid_phys_addr_range(p
, count
))
115 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
116 /* we don't have page 0 mapped on sparc and m68k.. */
118 sz
= size_inside_page(p
, count
);
120 if (clear_user(buf
, sz
))
130 bounce
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
135 unsigned long remaining
;
138 sz
= size_inside_page(p
, count
);
141 allowed
= page_is_allowed(p
>> PAGE_SHIFT
);
147 /* Show zeros for restricted memory. */
148 remaining
= clear_user(buf
, sz
);
151 * On ia64 if a page has been mapped somewhere as
152 * uncached, then it must also be accessed uncached
153 * by the kernel or data corruption may occur.
155 ptr
= xlate_dev_mem_ptr(p
);
159 probe
= copy_from_kernel_nofault(bounce
, ptr
, sz
);
160 unxlate_dev_mem_ptr(p
, ptr
);
164 remaining
= copy_to_user(buf
, bounce
, sz
);
174 if (should_stop_iteration())
187 static ssize_t
write_mem(struct file
*file
, const char __user
*buf
,
188 size_t count
, loff_t
*ppos
)
190 phys_addr_t p
= *ppos
;
192 unsigned long copied
;
198 if (!valid_phys_addr_range(p
, count
))
203 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
204 /* we don't have page 0 mapped on sparc and m68k.. */
206 sz
= size_inside_page(p
, count
);
207 /* Hmm. Do something? */
218 sz
= size_inside_page(p
, count
);
220 allowed
= page_is_allowed(p
>> PAGE_SHIFT
);
224 /* Skip actual writing when a page is marked as restricted. */
227 * On ia64 if a page has been mapped somewhere as
228 * uncached, then it must also be accessed uncached
229 * by the kernel or data corruption may occur.
231 ptr
= xlate_dev_mem_ptr(p
);
238 copied
= copy_from_user(ptr
, buf
, sz
);
239 unxlate_dev_mem_ptr(p
, ptr
);
241 written
+= sz
- copied
;
252 if (should_stop_iteration())
260 int __weak
phys_mem_access_prot_allowed(struct file
*file
,
261 unsigned long pfn
, unsigned long size
, pgprot_t
*vma_prot
)
266 #ifndef __HAVE_PHYS_MEM_ACCESS_PROT
269 * Architectures vary in how they handle caching for addresses
270 * outside of main memory.
273 #ifdef pgprot_noncached
274 static int uncached_access(struct file
*file
, phys_addr_t addr
)
277 * Accessing memory above the top the kernel knows about or through a
279 * that was marked O_DSYNC will be done non-cached.
281 if (file
->f_flags
& O_DSYNC
)
283 return addr
>= __pa(high_memory
);
287 static pgprot_t
phys_mem_access_prot(struct file
*file
, unsigned long pfn
,
288 unsigned long size
, pgprot_t vma_prot
)
290 #ifdef pgprot_noncached
291 phys_addr_t offset
= pfn
<< PAGE_SHIFT
;
293 if (uncached_access(file
, offset
))
294 return pgprot_noncached(vma_prot
);
301 static unsigned long get_unmapped_area_mem(struct file
*file
,
307 if (!valid_mmap_phys_addr_range(pgoff
, len
))
308 return (unsigned long) -EINVAL
;
309 return pgoff
<< PAGE_SHIFT
;
312 /* permit direct mmap, for read, write or exec */
313 static unsigned memory_mmap_capabilities(struct file
*file
)
315 return NOMMU_MAP_DIRECT
|
316 NOMMU_MAP_READ
| NOMMU_MAP_WRITE
| NOMMU_MAP_EXEC
;
319 static unsigned zero_mmap_capabilities(struct file
*file
)
321 return NOMMU_MAP_COPY
;
324 /* can't do an in-place private mapping if there's no MMU */
325 static inline int private_mapping_ok(struct vm_area_struct
*vma
)
327 return is_nommu_shared_mapping(vma
->vm_flags
);
331 static inline int private_mapping_ok(struct vm_area_struct
*vma
)
337 static const struct vm_operations_struct mmap_mem_ops
= {
338 #ifdef CONFIG_HAVE_IOREMAP_PROT
339 .access
= generic_access_phys
343 static int mmap_mem(struct file
*file
, struct vm_area_struct
*vma
)
345 size_t size
= vma
->vm_end
- vma
->vm_start
;
346 phys_addr_t offset
= (phys_addr_t
)vma
->vm_pgoff
<< PAGE_SHIFT
;
348 /* Does it even fit in phys_addr_t? */
349 if (offset
>> PAGE_SHIFT
!= vma
->vm_pgoff
)
352 /* It's illegal to wrap around the end of the physical address space. */
353 if (offset
+ (phys_addr_t
)size
- 1 < offset
)
356 if (!valid_mmap_phys_addr_range(vma
->vm_pgoff
, size
))
359 if (!private_mapping_ok(vma
))
362 if (!range_is_allowed(vma
->vm_pgoff
, size
))
365 if (!phys_mem_access_prot_allowed(file
, vma
->vm_pgoff
, size
,
369 vma
->vm_page_prot
= phys_mem_access_prot(file
, vma
->vm_pgoff
,
373 vma
->vm_ops
= &mmap_mem_ops
;
375 /* Remap-pfn-range will mark the range VM_IO */
376 if (remap_pfn_range(vma
,
380 vma
->vm_page_prot
)) {
386 #ifdef CONFIG_DEVPORT
387 static ssize_t
read_port(struct file
*file
, char __user
*buf
,
388 size_t count
, loff_t
*ppos
)
390 unsigned long i
= *ppos
;
391 char __user
*tmp
= buf
;
393 if (!access_ok(buf
, count
))
395 while (count
-- > 0 && i
< 65536) {
396 if (__put_user(inb(i
), tmp
) < 0)
405 static ssize_t
write_port(struct file
*file
, const char __user
*buf
,
406 size_t count
, loff_t
*ppos
)
408 unsigned long i
= *ppos
;
409 const char __user
*tmp
= buf
;
411 if (!access_ok(buf
, count
))
413 while (count
-- > 0 && i
< 65536) {
416 if (__get_user(c
, tmp
)) {
430 static ssize_t
read_null(struct file
*file
, char __user
*buf
,
431 size_t count
, loff_t
*ppos
)
436 static ssize_t
write_null(struct file
*file
, const char __user
*buf
,
437 size_t count
, loff_t
*ppos
)
442 static ssize_t
read_iter_null(struct kiocb
*iocb
, struct iov_iter
*to
)
447 static ssize_t
write_iter_null(struct kiocb
*iocb
, struct iov_iter
*from
)
449 size_t count
= iov_iter_count(from
);
450 iov_iter_advance(from
, count
);
454 static int pipe_to_null(struct pipe_inode_info
*info
, struct pipe_buffer
*buf
,
455 struct splice_desc
*sd
)
460 static ssize_t
splice_write_null(struct pipe_inode_info
*pipe
, struct file
*out
,
461 loff_t
*ppos
, size_t len
, unsigned int flags
)
463 return splice_from_pipe(pipe
, out
, ppos
, len
, flags
, pipe_to_null
);
466 static int uring_cmd_null(struct io_uring_cmd
*ioucmd
, unsigned int issue_flags
)
471 static ssize_t
read_iter_zero(struct kiocb
*iocb
, struct iov_iter
*iter
)
475 while (iov_iter_count(iter
)) {
476 size_t chunk
= iov_iter_count(iter
), n
;
478 if (chunk
> PAGE_SIZE
)
479 chunk
= PAGE_SIZE
; /* Just for latency reasons */
480 n
= iov_iter_zero(chunk
, iter
);
481 if (!n
&& iov_iter_count(iter
))
482 return written
? written
: -EFAULT
;
484 if (signal_pending(current
))
485 return written
? written
: -ERESTARTSYS
;
488 if (iocb
->ki_flags
& IOCB_NOWAIT
)
489 return written
? written
: -EAGAIN
;
495 static ssize_t
read_zero(struct file
*file
, char __user
*buf
,
496 size_t count
, loff_t
*ppos
)
501 size_t chunk
= min_t(size_t, count
, PAGE_SIZE
);
504 left
= clear_user(buf
+ cleared
, chunk
);
505 if (unlikely(left
)) {
506 cleared
+= (chunk
- left
);
514 if (signal_pending(current
))
522 static int mmap_zero(struct file
*file
, struct vm_area_struct
*vma
)
527 if (vma
->vm_flags
& VM_SHARED
)
528 return shmem_zero_setup(vma
);
529 vma_set_anonymous(vma
);
533 static unsigned long get_unmapped_area_zero(struct file
*file
,
534 unsigned long addr
, unsigned long len
,
535 unsigned long pgoff
, unsigned long flags
)
538 if (flags
& MAP_SHARED
) {
540 * mmap_zero() will call shmem_zero_setup() to create a file,
541 * so use shmem's get_unmapped_area in case it can be huge;
542 * and pass NULL for file as in mmap.c's get_unmapped_area(),
543 * so as not to confuse shmem with our handle on "/dev/zero".
545 return shmem_get_unmapped_area(NULL
, addr
, len
, pgoff
, flags
);
548 /* Otherwise flags & MAP_PRIVATE: with no shmem object beneath it */
549 return mm_get_unmapped_area(current
->mm
, file
, addr
, len
, pgoff
, flags
);
555 static ssize_t
write_full(struct file
*file
, const char __user
*buf
,
556 size_t count
, loff_t
*ppos
)
562 * Special lseek() function for /dev/null and /dev/zero. Most notably, you
563 * can fopen() both devices with "a" now. This was previously impossible.
566 static loff_t
null_lseek(struct file
*file
, loff_t offset
, int orig
)
568 return file
->f_pos
= 0;
572 * The memory devices use the full 32/64 bits of the offset, and so we cannot
573 * check against negative addresses: they are ok. The return value is weird,
574 * though, in that case (0).
576 * also note that seeking relative to the "end of file" isn't supported:
577 * it has no meaning, so it returns -EINVAL.
579 static loff_t
memory_lseek(struct file
*file
, loff_t offset
, int orig
)
583 inode_lock(file_inode(file
));
586 offset
+= file
->f_pos
;
589 /* to avoid userland mistaking f_pos=-9 as -EBADF=-9 */
590 if ((unsigned long long)offset
>= -MAX_ERRNO
) {
594 file
->f_pos
= offset
;
596 force_successful_syscall_return();
601 inode_unlock(file_inode(file
));
605 static int open_port(struct inode
*inode
, struct file
*filp
)
609 if (!capable(CAP_SYS_RAWIO
))
612 rc
= security_locked_down(LOCKDOWN_DEV_MEM
);
616 if (iminor(inode
) != DEVMEM_MINOR
)
620 * Use a unified address space to have a single point to manage
621 * revocations when drivers want to take over a /dev/mem mapped
624 filp
->f_mapping
= iomem_get_mapping();
629 #define zero_lseek null_lseek
630 #define full_lseek null_lseek
631 #define write_zero write_null
632 #define write_iter_zero write_iter_null
633 #define splice_write_zero splice_write_null
634 #define open_mem open_port
636 static const struct file_operations __maybe_unused mem_fops
= {
637 .llseek
= memory_lseek
,
643 .get_unmapped_area
= get_unmapped_area_mem
,
644 .mmap_capabilities
= memory_mmap_capabilities
,
646 .fop_flags
= FOP_UNSIGNED_OFFSET
,
649 static const struct file_operations null_fops
= {
650 .llseek
= null_lseek
,
653 .read_iter
= read_iter_null
,
654 .write_iter
= write_iter_null
,
655 .splice_write
= splice_write_null
,
656 .uring_cmd
= uring_cmd_null
,
659 #ifdef CONFIG_DEVPORT
660 static const struct file_operations port_fops
= {
661 .llseek
= memory_lseek
,
668 static const struct file_operations zero_fops
= {
669 .llseek
= zero_lseek
,
671 .read_iter
= read_iter_zero
,
673 .write_iter
= write_iter_zero
,
674 .splice_read
= copy_splice_read
,
675 .splice_write
= splice_write_zero
,
677 .get_unmapped_area
= get_unmapped_area_zero
,
679 .mmap_capabilities
= zero_mmap_capabilities
,
683 static const struct file_operations full_fops
= {
684 .llseek
= full_lseek
,
685 .read_iter
= read_iter_zero
,
687 .splice_read
= copy_splice_read
,
690 static const struct memdev
{
692 const struct file_operations
*fops
;
697 [DEVMEM_MINOR
] = { "mem", &mem_fops
, 0, 0 },
699 [3] = { "null", &null_fops
, FMODE_NOWAIT
, 0666 },
700 #ifdef CONFIG_DEVPORT
701 [4] = { "port", &port_fops
, 0, 0 },
703 [5] = { "zero", &zero_fops
, FMODE_NOWAIT
, 0666 },
704 [7] = { "full", &full_fops
, 0, 0666 },
705 [8] = { "random", &random_fops
, FMODE_NOWAIT
, 0666 },
706 [9] = { "urandom", &urandom_fops
, FMODE_NOWAIT
, 0666 },
708 [11] = { "kmsg", &kmsg_fops
, 0, 0644 },
712 static int memory_open(struct inode
*inode
, struct file
*filp
)
715 const struct memdev
*dev
;
717 minor
= iminor(inode
);
718 if (minor
>= ARRAY_SIZE(devlist
))
721 dev
= &devlist
[minor
];
725 filp
->f_op
= dev
->fops
;
726 filp
->f_mode
|= dev
->fmode
;
729 return dev
->fops
->open(inode
, filp
);
734 static const struct file_operations memory_fops
= {
736 .llseek
= noop_llseek
,
739 static char *mem_devnode(const struct device
*dev
, umode_t
*mode
)
741 if (mode
&& devlist
[MINOR(dev
->devt
)].mode
)
742 *mode
= devlist
[MINOR(dev
->devt
)].mode
;
746 static const struct class mem_class
= {
748 .devnode
= mem_devnode
,
751 static int __init
chr_dev_init(void)
756 if (register_chrdev(MEM_MAJOR
, "mem", &memory_fops
))
757 printk("unable to get major %d for memory devs\n", MEM_MAJOR
);
759 retval
= class_register(&mem_class
);
763 for (minor
= 1; minor
< ARRAY_SIZE(devlist
); minor
++) {
764 if (!devlist
[minor
].name
)
770 if ((minor
== DEVPORT_MINOR
) && !arch_has_dev_port())
773 device_create(&mem_class
, NULL
, MKDEV(MEM_MAJOR
, minor
),
774 NULL
, devlist
[minor
].name
);
780 fs_initcall(chr_dev_init
);