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/shmem_fs.h>
26 #include <linux/splice.h>
27 #include <linux/pfn.h>
28 #include <linux/export.h>
30 #include <linux/uio.h>
32 #include <linux/uaccess.h>
35 # include <linux/efi.h>
38 #define DEVPORT_MINOR 4
40 static inline unsigned long size_inside_page(unsigned long start,
45 sz = PAGE_SIZE - (start & (PAGE_SIZE - 1));
50 #ifndef ARCH_HAS_VALID_PHYS_ADDR_RANGE
51 static inline int valid_phys_addr_range(phys_addr_t addr, size_t count)
53 return 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 page_is_allowed(unsigned long pfn)
65 return devmem_is_allowed(pfn);
67 static inline int range_is_allowed(unsigned long pfn, unsigned long size)
69 u64 from = ((u64)pfn) << PAGE_SHIFT;
74 if (!devmem_is_allowed(pfn))
82 static inline int page_is_allowed(unsigned long pfn)
86 static inline int range_is_allowed(unsigned long pfn, unsigned long size)
92 #ifndef unxlate_dev_mem_ptr
93 #define unxlate_dev_mem_ptr unxlate_dev_mem_ptr
94 void __weak unxlate_dev_mem_ptr(phys_addr_t phys, void *addr)
100 * This funcion reads the *physical* memory. The f_pos points directly to the
103 static ssize_t read_mem(struct file *file, char __user *buf,
104 size_t count, loff_t *ppos)
106 phys_addr_t p = *ppos;
113 if (!valid_phys_addr_range(p, count))
116 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
117 /* we don't have page 0 mapped on sparc and m68k.. */
119 sz = size_inside_page(p, count);
121 if (clear_user(buf, sz))
132 unsigned long remaining;
135 sz = size_inside_page(p, count);
137 allowed = page_is_allowed(p >> PAGE_SHIFT);
141 /* Show zeros for restricted memory. */
142 remaining = clear_user(buf, sz);
145 * On ia64 if a page has been mapped somewhere as
146 * uncached, then it must also be accessed uncached
147 * by the kernel or data corruption may occur.
149 ptr = xlate_dev_mem_ptr(p);
153 remaining = copy_to_user(buf, ptr, sz);
155 unxlate_dev_mem_ptr(p, ptr);
171 static ssize_t write_mem(struct file *file, const char __user *buf,
172 size_t count, loff_t *ppos)
174 phys_addr_t p = *ppos;
176 unsigned long copied;
182 if (!valid_phys_addr_range(p, count))
187 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
188 /* we don't have page 0 mapped on sparc and m68k.. */
190 sz = size_inside_page(p, count);
191 /* Hmm. Do something? */
202 sz = size_inside_page(p, count);
204 allowed = page_is_allowed(p >> PAGE_SHIFT);
208 /* Skip actual writing when a page is marked as restricted. */
211 * On ia64 if a page has been mapped somewhere as
212 * uncached, then it must also be accessed uncached
213 * by the kernel or data corruption may occur.
215 ptr = xlate_dev_mem_ptr(p);
222 copied = copy_from_user(ptr, buf, sz);
223 unxlate_dev_mem_ptr(p, ptr);
225 written += sz - copied;
242 int __weak phys_mem_access_prot_allowed(struct file *file,
243 unsigned long pfn, unsigned long size, pgprot_t *vma_prot)
248 #ifndef __HAVE_PHYS_MEM_ACCESS_PROT
251 * Architectures vary in how they handle caching for addresses
252 * outside of main memory.
255 #ifdef pgprot_noncached
256 static int uncached_access(struct file *file, phys_addr_t addr)
258 #if defined(CONFIG_IA64)
260 * On ia64, we ignore O_DSYNC because we cannot tolerate memory
263 return !(efi_mem_attributes(addr) & EFI_MEMORY_WB);
264 #elif defined(CONFIG_MIPS)
266 extern int __uncached_access(struct file *file,
269 return __uncached_access(file, addr);
273 * Accessing memory above the top the kernel knows about or through a
275 * that was marked O_DSYNC will be done non-cached.
277 if (file->f_flags & O_DSYNC)
279 return addr >= __pa(high_memory);
284 static pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
285 unsigned long size, pgprot_t vma_prot)
287 #ifdef pgprot_noncached
288 phys_addr_t offset = pfn << PAGE_SHIFT;
290 if (uncached_access(file, offset))
291 return pgprot_noncached(vma_prot);
298 static unsigned long get_unmapped_area_mem(struct file *file,
304 if (!valid_mmap_phys_addr_range(pgoff, len))
305 return (unsigned long) -EINVAL;
306 return pgoff << PAGE_SHIFT;
309 /* permit direct mmap, for read, write or exec */
310 static unsigned memory_mmap_capabilities(struct file *file)
312 return NOMMU_MAP_DIRECT |
313 NOMMU_MAP_READ | NOMMU_MAP_WRITE | NOMMU_MAP_EXEC;
316 static unsigned zero_mmap_capabilities(struct file *file)
318 return NOMMU_MAP_COPY;
321 /* can't do an in-place private mapping if there's no MMU */
322 static inline int private_mapping_ok(struct vm_area_struct *vma)
324 return vma->vm_flags & VM_MAYSHARE;
328 static inline int private_mapping_ok(struct vm_area_struct *vma)
334 static const struct vm_operations_struct mmap_mem_ops = {
335 #ifdef CONFIG_HAVE_IOREMAP_PROT
336 .access = generic_access_phys
340 static int mmap_mem(struct file *file, struct vm_area_struct *vma)
342 size_t size = vma->vm_end - vma->vm_start;
344 if (!valid_mmap_phys_addr_range(vma->vm_pgoff, size))
347 if (!private_mapping_ok(vma))
350 if (!range_is_allowed(vma->vm_pgoff, size))
353 if (!phys_mem_access_prot_allowed(file, vma->vm_pgoff, size,
357 vma->vm_page_prot = phys_mem_access_prot(file, vma->vm_pgoff,
361 vma->vm_ops = &mmap_mem_ops;
363 /* Remap-pfn-range will mark the range VM_IO */
364 if (remap_pfn_range(vma,
368 vma->vm_page_prot)) {
374 static int mmap_kmem(struct file *file, struct vm_area_struct *vma)
378 /* Turn a kernel-virtual address into a physical page frame */
379 pfn = __pa((u64)vma->vm_pgoff << PAGE_SHIFT) >> PAGE_SHIFT;
382 * RED-PEN: on some architectures there is more mapped memory than
383 * available in mem_map which pfn_valid checks for. Perhaps should add a
386 * RED-PEN: vmalloc is not supported right now.
392 return mmap_mem(file, vma);
396 * This function reads the *virtual* memory as seen by the kernel.
398 static ssize_t read_kmem(struct file *file, char __user *buf,
399 size_t count, loff_t *ppos)
401 unsigned long p = *ppos;
402 ssize_t low_count, read, sz;
403 char *kbuf; /* k-addr because vread() takes vmlist_lock rwlock */
407 if (p < (unsigned long) high_memory) {
409 if (count > (unsigned long)high_memory - p)
410 low_count = (unsigned long)high_memory - p;
412 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
413 /* we don't have page 0 mapped on sparc and m68k.. */
414 if (p < PAGE_SIZE && low_count > 0) {
415 sz = size_inside_page(p, low_count);
416 if (clear_user(buf, sz))
425 while (low_count > 0) {
426 sz = size_inside_page(p, low_count);
429 * On ia64 if a page has been mapped somewhere as
430 * uncached, then it must also be accessed uncached
431 * by the kernel or data corruption may occur
433 kbuf = xlate_dev_kmem_ptr((void *)p);
434 if (!virt_addr_valid(kbuf))
437 if (copy_to_user(buf, kbuf, sz))
448 kbuf = (char *)__get_free_page(GFP_KERNEL);
452 sz = size_inside_page(p, count);
453 if (!is_vmalloc_or_module_addr((void *)p)) {
457 sz = vread(kbuf, (char *)p, sz);
460 if (copy_to_user(buf, kbuf, sz)) {
469 free_page((unsigned long)kbuf);
472 return read ? read : err;
476 static ssize_t do_write_kmem(unsigned long p, const char __user *buf,
477 size_t count, loff_t *ppos)
480 unsigned long copied;
483 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
484 /* we don't have page 0 mapped on sparc and m68k.. */
486 sz = size_inside_page(p, count);
487 /* Hmm. Do something? */
498 sz = size_inside_page(p, count);
501 * On ia64 if a page has been mapped somewhere as uncached, then
502 * it must also be accessed uncached by the kernel or data
503 * corruption may occur.
505 ptr = xlate_dev_kmem_ptr((void *)p);
506 if (!virt_addr_valid(ptr))
509 copied = copy_from_user(ptr, buf, sz);
511 written += sz - copied;
527 * This function writes to the *virtual* memory as seen by the kernel.
529 static ssize_t write_kmem(struct file *file, const char __user *buf,
530 size_t count, loff_t *ppos)
532 unsigned long p = *ppos;
535 char *kbuf; /* k-addr because vwrite() takes vmlist_lock rwlock */
538 if (p < (unsigned long) high_memory) {
539 unsigned long to_write = min_t(unsigned long, count,
540 (unsigned long)high_memory - p);
541 wrote = do_write_kmem(p, buf, to_write, ppos);
542 if (wrote != to_write)
550 kbuf = (char *)__get_free_page(GFP_KERNEL);
552 return wrote ? wrote : -ENOMEM;
554 unsigned long sz = size_inside_page(p, count);
557 if (!is_vmalloc_or_module_addr((void *)p)) {
561 n = copy_from_user(kbuf, buf, sz);
566 vwrite(kbuf, (char *)p, sz);
572 free_page((unsigned long)kbuf);
576 return virtr + wrote ? : err;
579 static ssize_t read_port(struct file *file, char __user *buf,
580 size_t count, loff_t *ppos)
582 unsigned long i = *ppos;
583 char __user *tmp = buf;
585 if (!access_ok(VERIFY_WRITE, buf, count))
587 while (count-- > 0 && i < 65536) {
588 if (__put_user(inb(i), tmp) < 0)
597 static ssize_t write_port(struct file *file, const char __user *buf,
598 size_t count, loff_t *ppos)
600 unsigned long i = *ppos;
601 const char __user *tmp = buf;
603 if (!access_ok(VERIFY_READ, buf, count))
605 while (count-- > 0 && i < 65536) {
608 if (__get_user(c, tmp)) {
621 static ssize_t read_null(struct file *file, char __user *buf,
622 size_t count, loff_t *ppos)
627 static ssize_t write_null(struct file *file, const char __user *buf,
628 size_t count, loff_t *ppos)
633 static ssize_t read_iter_null(struct kiocb *iocb, struct iov_iter *to)
638 static ssize_t write_iter_null(struct kiocb *iocb, struct iov_iter *from)
640 size_t count = iov_iter_count(from);
641 iov_iter_advance(from, count);
645 static int pipe_to_null(struct pipe_inode_info *info, struct pipe_buffer *buf,
646 struct splice_desc *sd)
651 static ssize_t splice_write_null(struct pipe_inode_info *pipe, struct file *out,
652 loff_t *ppos, size_t len, unsigned int flags)
654 return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_null);
657 static ssize_t read_iter_zero(struct kiocb *iocb, struct iov_iter *iter)
661 while (iov_iter_count(iter)) {
662 size_t chunk = iov_iter_count(iter), n;
664 if (chunk > PAGE_SIZE)
665 chunk = PAGE_SIZE; /* Just for latency reasons */
666 n = iov_iter_zero(chunk, iter);
667 if (!n && iov_iter_count(iter))
668 return written ? written : -EFAULT;
670 if (signal_pending(current))
671 return written ? written : -ERESTARTSYS;
677 static int mmap_zero(struct file *file, struct vm_area_struct *vma)
682 if (vma->vm_flags & VM_SHARED)
683 return shmem_zero_setup(vma);
687 static unsigned long get_unmapped_area_zero(struct file *file,
688 unsigned long addr, unsigned long len,
689 unsigned long pgoff, unsigned long flags)
692 if (flags & MAP_SHARED) {
694 * mmap_zero() will call shmem_zero_setup() to create a file,
695 * so use shmem's get_unmapped_area in case it can be huge;
696 * and pass NULL for file as in mmap.c's get_unmapped_area(),
697 * so as not to confuse shmem with our handle on "/dev/zero".
699 return shmem_get_unmapped_area(NULL, addr, len, pgoff, flags);
702 /* Otherwise flags & MAP_PRIVATE: with no shmem object beneath it */
703 return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
709 static ssize_t write_full(struct file *file, const char __user *buf,
710 size_t count, loff_t *ppos)
716 * Special lseek() function for /dev/null and /dev/zero. Most notably, you
717 * can fopen() both devices with "a" now. This was previously impossible.
720 static loff_t null_lseek(struct file *file, loff_t offset, int orig)
722 return file->f_pos = 0;
726 * The memory devices use the full 32/64 bits of the offset, and so we cannot
727 * check against negative addresses: they are ok. The return value is weird,
728 * though, in that case (0).
730 * also note that seeking relative to the "end of file" isn't supported:
731 * it has no meaning, so it returns -EINVAL.
733 static loff_t memory_lseek(struct file *file, loff_t offset, int orig)
737 inode_lock(file_inode(file));
740 offset += file->f_pos;
742 /* to avoid userland mistaking f_pos=-9 as -EBADF=-9 */
743 if ((unsigned long long)offset >= -MAX_ERRNO) {
747 file->f_pos = offset;
749 force_successful_syscall_return();
754 inode_unlock(file_inode(file));
758 static int open_port(struct inode *inode, struct file *filp)
760 return capable(CAP_SYS_RAWIO) ? 0 : -EPERM;
763 #define zero_lseek null_lseek
764 #define full_lseek null_lseek
765 #define write_zero write_null
766 #define write_iter_zero write_iter_null
767 #define open_mem open_port
768 #define open_kmem open_mem
770 static const struct file_operations __maybe_unused mem_fops = {
771 .llseek = memory_lseek,
777 .get_unmapped_area = get_unmapped_area_mem,
778 .mmap_capabilities = memory_mmap_capabilities,
782 static const struct file_operations __maybe_unused kmem_fops = {
783 .llseek = memory_lseek,
789 .get_unmapped_area = get_unmapped_area_mem,
790 .mmap_capabilities = memory_mmap_capabilities,
794 static const struct file_operations null_fops = {
795 .llseek = null_lseek,
798 .read_iter = read_iter_null,
799 .write_iter = write_iter_null,
800 .splice_write = splice_write_null,
803 static const struct file_operations __maybe_unused port_fops = {
804 .llseek = memory_lseek,
810 static const struct file_operations zero_fops = {
811 .llseek = zero_lseek,
813 .read_iter = read_iter_zero,
814 .write_iter = write_iter_zero,
816 .get_unmapped_area = get_unmapped_area_zero,
818 .mmap_capabilities = zero_mmap_capabilities,
822 static const struct file_operations full_fops = {
823 .llseek = full_lseek,
824 .read_iter = read_iter_zero,
828 static const struct memdev {
831 const struct file_operations *fops;
835 [1] = { "mem", 0, &mem_fops, FMODE_UNSIGNED_OFFSET },
837 #ifdef CONFIG_DEVKMEM
838 [2] = { "kmem", 0, &kmem_fops, FMODE_UNSIGNED_OFFSET },
840 [3] = { "null", 0666, &null_fops, 0 },
841 #ifdef CONFIG_DEVPORT
842 [4] = { "port", 0, &port_fops, 0 },
844 [5] = { "zero", 0666, &zero_fops, 0 },
845 [7] = { "full", 0666, &full_fops, 0 },
846 [8] = { "random", 0666, &random_fops, 0 },
847 [9] = { "urandom", 0666, &urandom_fops, 0 },
849 [11] = { "kmsg", 0644, &kmsg_fops, 0 },
853 static int memory_open(struct inode *inode, struct file *filp)
856 const struct memdev *dev;
858 minor = iminor(inode);
859 if (minor >= ARRAY_SIZE(devlist))
862 dev = &devlist[minor];
866 filp->f_op = dev->fops;
867 filp->f_mode |= dev->fmode;
870 return dev->fops->open(inode, filp);
875 static const struct file_operations memory_fops = {
877 .llseek = noop_llseek,
880 static char *mem_devnode(struct device *dev, umode_t *mode)
882 if (mode && devlist[MINOR(dev->devt)].mode)
883 *mode = devlist[MINOR(dev->devt)].mode;
887 static struct class *mem_class;
889 static int __init chr_dev_init(void)
893 if (register_chrdev(MEM_MAJOR, "mem", &memory_fops))
894 printk("unable to get major %d for memory devs\n", MEM_MAJOR);
896 mem_class = class_create(THIS_MODULE, "mem");
897 if (IS_ERR(mem_class))
898 return PTR_ERR(mem_class);
900 mem_class->devnode = mem_devnode;
901 for (minor = 1; minor < ARRAY_SIZE(devlist); minor++) {
902 if (!devlist[minor].name)
908 if ((minor == DEVPORT_MINOR) && !arch_has_dev_port())
911 device_create(mem_class, NULL, MKDEV(MEM_MAJOR, minor),
912 NULL, devlist[minor].name);
918 fs_initcall(chr_dev_init);