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/raw.h>
20 #include <linux/tty.h>
21 #include <linux/capability.h>
22 #include <linux/ptrace.h>
23 #include <linux/device.h>
24 #include <linux/highmem.h>
25 #include <linux/backing-dev.h>
26 #include <linux/shmem_fs.h>
27 #include <linux/splice.h>
28 #include <linux/pfn.h>
29 #include <linux/export.h>
31 #include <linux/uio.h>
33 #include <linux/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 #ifdef CONFIG_STRICT_DEVMEM
64 static inline int page_is_allowed(unsigned long pfn)
66 return devmem_is_allowed(pfn);
68 static inline int range_is_allowed(unsigned long pfn, unsigned long size)
70 u64 from = ((u64)pfn) << PAGE_SHIFT;
75 if (!devmem_is_allowed(pfn))
83 static inline int page_is_allowed(unsigned long pfn)
87 static inline int range_is_allowed(unsigned long pfn, unsigned long size)
93 #ifndef unxlate_dev_mem_ptr
94 #define unxlate_dev_mem_ptr unxlate_dev_mem_ptr
95 void __weak unxlate_dev_mem_ptr(phys_addr_t phys, void *addr)
101 * This funcion reads the *physical* memory. The f_pos points directly to the
104 static ssize_t read_mem(struct file *file, char __user *buf,
105 size_t count, loff_t *ppos)
107 phys_addr_t p = *ppos;
114 if (!valid_phys_addr_range(p, count))
117 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
118 /* we don't have page 0 mapped on sparc and m68k.. */
120 sz = size_inside_page(p, count);
122 if (clear_user(buf, sz))
133 unsigned long remaining;
136 sz = size_inside_page(p, count);
138 allowed = page_is_allowed(p >> PAGE_SHIFT);
142 /* Show zeros for restricted memory. */
143 remaining = clear_user(buf, sz);
146 * On ia64 if a page has been mapped somewhere as
147 * uncached, then it must also be accessed uncached
148 * by the kernel or data corruption may occur.
150 ptr = xlate_dev_mem_ptr(p);
154 remaining = copy_to_user(buf, ptr, sz);
156 unxlate_dev_mem_ptr(p, ptr);
172 static ssize_t write_mem(struct file *file, const char __user *buf,
173 size_t count, loff_t *ppos)
175 phys_addr_t p = *ppos;
177 unsigned long copied;
183 if (!valid_phys_addr_range(p, count))
188 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
189 /* we don't have page 0 mapped on sparc and m68k.. */
191 sz = size_inside_page(p, count);
192 /* Hmm. Do something? */
203 sz = size_inside_page(p, count);
205 allowed = page_is_allowed(p >> PAGE_SHIFT);
209 /* Skip actual writing when a page is marked as restricted. */
212 * On ia64 if a page has been mapped somewhere as
213 * uncached, then it must also be accessed uncached
214 * by the kernel or data corruption may occur.
216 ptr = xlate_dev_mem_ptr(p);
223 copied = copy_from_user(ptr, buf, sz);
224 unxlate_dev_mem_ptr(p, ptr);
226 written += sz - copied;
243 int __weak phys_mem_access_prot_allowed(struct file *file,
244 unsigned long pfn, unsigned long size, pgprot_t *vma_prot)
249 #ifndef __HAVE_PHYS_MEM_ACCESS_PROT
252 * Architectures vary in how they handle caching for addresses
253 * outside of main memory.
256 #ifdef pgprot_noncached
257 static int uncached_access(struct file *file, phys_addr_t addr)
259 #if defined(CONFIG_IA64)
261 * On ia64, we ignore O_DSYNC because we cannot tolerate memory
264 return !(efi_mem_attributes(addr) & EFI_MEMORY_WB);
265 #elif defined(CONFIG_MIPS)
267 extern int __uncached_access(struct file *file,
270 return __uncached_access(file, addr);
274 * Accessing memory above the top the kernel knows about or through a
276 * that was marked O_DSYNC will be done non-cached.
278 if (file->f_flags & O_DSYNC)
280 return addr >= __pa(high_memory);
285 static pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
286 unsigned long size, pgprot_t vma_prot)
288 #ifdef pgprot_noncached
289 phys_addr_t offset = pfn << PAGE_SHIFT;
291 if (uncached_access(file, offset))
292 return pgprot_noncached(vma_prot);
299 static unsigned long get_unmapped_area_mem(struct file *file,
305 if (!valid_mmap_phys_addr_range(pgoff, len))
306 return (unsigned long) -EINVAL;
307 return pgoff << PAGE_SHIFT;
310 /* permit direct mmap, for read, write or exec */
311 static unsigned memory_mmap_capabilities(struct file *file)
313 return NOMMU_MAP_DIRECT |
314 NOMMU_MAP_READ | NOMMU_MAP_WRITE | NOMMU_MAP_EXEC;
317 static unsigned zero_mmap_capabilities(struct file *file)
319 return NOMMU_MAP_COPY;
322 /* can't do an in-place private mapping if there's no MMU */
323 static inline int private_mapping_ok(struct vm_area_struct *vma)
325 return vma->vm_flags & VM_MAYSHARE;
329 static inline int private_mapping_ok(struct vm_area_struct *vma)
335 static const struct vm_operations_struct mmap_mem_ops = {
336 #ifdef CONFIG_HAVE_IOREMAP_PROT
337 .access = generic_access_phys
341 static int mmap_mem(struct file *file, struct vm_area_struct *vma)
343 size_t size = vma->vm_end - vma->vm_start;
344 phys_addr_t offset = (phys_addr_t)vma->vm_pgoff << PAGE_SHIFT;
346 /* It's illegal to wrap around the end of the physical address space. */
347 if (offset + (phys_addr_t)size - 1 < offset)
350 if (!valid_mmap_phys_addr_range(vma->vm_pgoff, size))
353 if (!private_mapping_ok(vma))
356 if (!range_is_allowed(vma->vm_pgoff, size))
359 if (!phys_mem_access_prot_allowed(file, vma->vm_pgoff, size,
363 vma->vm_page_prot = phys_mem_access_prot(file, vma->vm_pgoff,
367 vma->vm_ops = &mmap_mem_ops;
369 /* Remap-pfn-range will mark the range VM_IO */
370 if (remap_pfn_range(vma,
374 vma->vm_page_prot)) {
380 static int mmap_kmem(struct file *file, struct vm_area_struct *vma)
384 /* Turn a kernel-virtual address into a physical page frame */
385 pfn = __pa((u64)vma->vm_pgoff << PAGE_SHIFT) >> PAGE_SHIFT;
388 * RED-PEN: on some architectures there is more mapped memory than
389 * available in mem_map which pfn_valid checks for. Perhaps should add a
392 * RED-PEN: vmalloc is not supported right now.
398 return mmap_mem(file, vma);
402 * This function reads the *virtual* memory as seen by the kernel.
404 static ssize_t read_kmem(struct file *file, char __user *buf,
405 size_t count, loff_t *ppos)
407 unsigned long p = *ppos;
408 ssize_t low_count, read, sz;
409 char *kbuf; /* k-addr because vread() takes vmlist_lock rwlock */
413 if (p < (unsigned long) high_memory) {
415 if (count > (unsigned long)high_memory - p)
416 low_count = (unsigned long)high_memory - p;
418 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
419 /* we don't have page 0 mapped on sparc and m68k.. */
420 if (p < PAGE_SIZE && low_count > 0) {
421 sz = size_inside_page(p, low_count);
422 if (clear_user(buf, sz))
431 while (low_count > 0) {
432 sz = size_inside_page(p, low_count);
435 * On ia64 if a page has been mapped somewhere as
436 * uncached, then it must also be accessed uncached
437 * by the kernel or data corruption may occur
439 kbuf = xlate_dev_kmem_ptr((void *)p);
440 if (!virt_addr_valid(kbuf))
443 if (copy_to_user(buf, kbuf, sz))
454 kbuf = (char *)__get_free_page(GFP_KERNEL);
458 sz = size_inside_page(p, count);
459 if (!is_vmalloc_or_module_addr((void *)p)) {
463 sz = vread(kbuf, (char *)p, sz);
466 if (copy_to_user(buf, kbuf, sz)) {
475 free_page((unsigned long)kbuf);
478 return read ? read : err;
482 static ssize_t do_write_kmem(unsigned long p, const char __user *buf,
483 size_t count, loff_t *ppos)
486 unsigned long copied;
489 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
490 /* we don't have page 0 mapped on sparc and m68k.. */
492 sz = size_inside_page(p, count);
493 /* Hmm. Do something? */
504 sz = size_inside_page(p, count);
507 * On ia64 if a page has been mapped somewhere as uncached, then
508 * it must also be accessed uncached by the kernel or data
509 * corruption may occur.
511 ptr = xlate_dev_kmem_ptr((void *)p);
512 if (!virt_addr_valid(ptr))
515 copied = copy_from_user(ptr, buf, sz);
517 written += sz - copied;
533 * This function writes to the *virtual* memory as seen by the kernel.
535 static ssize_t write_kmem(struct file *file, const char __user *buf,
536 size_t count, loff_t *ppos)
538 unsigned long p = *ppos;
541 char *kbuf; /* k-addr because vwrite() takes vmlist_lock rwlock */
544 if (p < (unsigned long) high_memory) {
545 unsigned long to_write = min_t(unsigned long, count,
546 (unsigned long)high_memory - p);
547 wrote = do_write_kmem(p, buf, to_write, ppos);
548 if (wrote != to_write)
556 kbuf = (char *)__get_free_page(GFP_KERNEL);
558 return wrote ? wrote : -ENOMEM;
560 unsigned long sz = size_inside_page(p, count);
563 if (!is_vmalloc_or_module_addr((void *)p)) {
567 n = copy_from_user(kbuf, buf, sz);
572 vwrite(kbuf, (char *)p, sz);
578 free_page((unsigned long)kbuf);
582 return virtr + wrote ? : err;
585 static ssize_t read_port(struct file *file, char __user *buf,
586 size_t count, loff_t *ppos)
588 unsigned long i = *ppos;
589 char __user *tmp = buf;
591 if (!access_ok(VERIFY_WRITE, buf, count))
593 while (count-- > 0 && i < 65536) {
594 if (__put_user(inb(i), tmp) < 0)
603 static ssize_t write_port(struct file *file, const char __user *buf,
604 size_t count, loff_t *ppos)
606 unsigned long i = *ppos;
607 const char __user *tmp = buf;
609 if (!access_ok(VERIFY_READ, buf, count))
611 while (count-- > 0 && i < 65536) {
614 if (__get_user(c, tmp)) {
627 static ssize_t read_null(struct file *file, char __user *buf,
628 size_t count, loff_t *ppos)
633 static ssize_t write_null(struct file *file, const char __user *buf,
634 size_t count, loff_t *ppos)
639 static ssize_t read_iter_null(struct kiocb *iocb, struct iov_iter *to)
644 static ssize_t write_iter_null(struct kiocb *iocb, struct iov_iter *from)
646 size_t count = iov_iter_count(from);
647 iov_iter_advance(from, count);
651 static int pipe_to_null(struct pipe_inode_info *info, struct pipe_buffer *buf,
652 struct splice_desc *sd)
657 static ssize_t splice_write_null(struct pipe_inode_info *pipe, struct file *out,
658 loff_t *ppos, size_t len, unsigned int flags)
660 return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_null);
663 static ssize_t read_iter_zero(struct kiocb *iocb, struct iov_iter *iter)
667 while (iov_iter_count(iter)) {
668 size_t chunk = iov_iter_count(iter), n;
670 if (chunk > PAGE_SIZE)
671 chunk = PAGE_SIZE; /* Just for latency reasons */
672 n = iov_iter_zero(chunk, iter);
673 if (!n && iov_iter_count(iter))
674 return written ? written : -EFAULT;
676 if (signal_pending(current))
677 return written ? written : -ERESTARTSYS;
683 static int mmap_zero(struct file *file, struct vm_area_struct *vma)
688 if (vma->vm_flags & VM_SHARED)
689 return shmem_zero_setup(vma);
693 static unsigned long get_unmapped_area_zero(struct file *file,
694 unsigned long addr, unsigned long len,
695 unsigned long pgoff, unsigned long flags)
698 if (flags & MAP_SHARED) {
700 * mmap_zero() will call shmem_zero_setup() to create a file,
701 * so use shmem's get_unmapped_area in case it can be huge;
702 * and pass NULL for file as in mmap.c's get_unmapped_area(),
703 * so as not to confuse shmem with our handle on "/dev/zero".
705 return shmem_get_unmapped_area(NULL, addr, len, pgoff, flags);
708 /* Otherwise flags & MAP_PRIVATE: with no shmem object beneath it */
709 return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
715 static ssize_t write_full(struct file *file, const char __user *buf,
716 size_t count, loff_t *ppos)
722 * Special lseek() function for /dev/null and /dev/zero. Most notably, you
723 * can fopen() both devices with "a" now. This was previously impossible.
726 static loff_t null_lseek(struct file *file, loff_t offset, int orig)
728 return file->f_pos = 0;
732 * The memory devices use the full 32/64 bits of the offset, and so we cannot
733 * check against negative addresses: they are ok. The return value is weird,
734 * though, in that case (0).
736 * also note that seeking relative to the "end of file" isn't supported:
737 * it has no meaning, so it returns -EINVAL.
739 static loff_t memory_lseek(struct file *file, loff_t offset, int orig)
743 inode_lock(file_inode(file));
746 offset += file->f_pos;
748 /* to avoid userland mistaking f_pos=-9 as -EBADF=-9 */
749 if ((unsigned long long)offset >= -MAX_ERRNO) {
753 file->f_pos = offset;
755 force_successful_syscall_return();
760 inode_unlock(file_inode(file));
764 static int open_port(struct inode *inode, struct file *filp)
766 return capable(CAP_SYS_RAWIO) ? 0 : -EPERM;
769 #define zero_lseek null_lseek
770 #define full_lseek null_lseek
771 #define write_zero write_null
772 #define write_iter_zero write_iter_null
773 #define open_mem open_port
774 #define open_kmem open_mem
776 static const struct file_operations __maybe_unused mem_fops = {
777 .llseek = memory_lseek,
783 .get_unmapped_area = get_unmapped_area_mem,
784 .mmap_capabilities = memory_mmap_capabilities,
788 static const struct file_operations __maybe_unused kmem_fops = {
789 .llseek = memory_lseek,
795 .get_unmapped_area = get_unmapped_area_mem,
796 .mmap_capabilities = memory_mmap_capabilities,
800 static const struct file_operations null_fops = {
801 .llseek = null_lseek,
804 .read_iter = read_iter_null,
805 .write_iter = write_iter_null,
806 .splice_write = splice_write_null,
809 static const struct file_operations __maybe_unused port_fops = {
810 .llseek = memory_lseek,
816 static const struct file_operations zero_fops = {
817 .llseek = zero_lseek,
819 .read_iter = read_iter_zero,
820 .write_iter = write_iter_zero,
822 .get_unmapped_area = get_unmapped_area_zero,
824 .mmap_capabilities = zero_mmap_capabilities,
828 static const struct file_operations full_fops = {
829 .llseek = full_lseek,
830 .read_iter = read_iter_zero,
834 static const struct memdev {
837 const struct file_operations *fops;
841 [1] = { "mem", 0, &mem_fops, FMODE_UNSIGNED_OFFSET },
843 #ifdef CONFIG_DEVKMEM
844 [2] = { "kmem", 0, &kmem_fops, FMODE_UNSIGNED_OFFSET },
846 [3] = { "null", 0666, &null_fops, 0 },
847 #ifdef CONFIG_DEVPORT
848 [4] = { "port", 0, &port_fops, 0 },
850 [5] = { "zero", 0666, &zero_fops, 0 },
851 [7] = { "full", 0666, &full_fops, 0 },
852 [8] = { "random", 0666, &random_fops, 0 },
853 [9] = { "urandom", 0666, &urandom_fops, 0 },
855 [11] = { "kmsg", 0644, &kmsg_fops, 0 },
859 static int memory_open(struct inode *inode, struct file *filp)
862 const struct memdev *dev;
864 minor = iminor(inode);
865 if (minor >= ARRAY_SIZE(devlist))
868 dev = &devlist[minor];
872 filp->f_op = dev->fops;
873 filp->f_mode |= dev->fmode;
876 return dev->fops->open(inode, filp);
881 static const struct file_operations memory_fops = {
883 .llseek = noop_llseek,
886 static char *mem_devnode(struct device *dev, umode_t *mode)
888 if (mode && devlist[MINOR(dev->devt)].mode)
889 *mode = devlist[MINOR(dev->devt)].mode;
893 static struct class *mem_class;
895 static int __init chr_dev_init(void)
899 if (register_chrdev(MEM_MAJOR, "mem", &memory_fops))
900 printk("unable to get major %d for memory devs\n", MEM_MAJOR);
902 mem_class = class_create(THIS_MODULE, "mem");
903 if (IS_ERR(mem_class))
904 return PTR_ERR(mem_class);
906 mem_class->devnode = mem_devnode;
907 for (minor = 1; minor < ARRAY_SIZE(devlist); minor++) {
908 if (!devlist[minor].name)
914 if ((minor == DEVPORT_MINOR) && !arch_has_dev_port())
917 device_create(mem_class, NULL, MKDEV(MEM_MAJOR, minor),
918 NULL, devlist[minor].name);
924 fs_initcall(chr_dev_init);