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>
35 # include <linux/efi.h>
38 #define DEVMEM_MINOR 1
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)
100 static inline bool should_stop_iteration(void)
104 return signal_pending(current);
108 * This funcion reads the *physical* memory. The f_pos points directly to the
111 static ssize_t read_mem(struct file *file, char __user *buf,
112 size_t count, loff_t *ppos)
114 phys_addr_t p = *ppos;
123 if (!valid_phys_addr_range(p, count))
126 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
127 /* we don't have page 0 mapped on sparc and m68k.. */
129 sz = size_inside_page(p, count);
131 if (clear_user(buf, sz))
141 bounce = kmalloc(PAGE_SIZE, GFP_KERNEL);
146 unsigned long remaining;
149 sz = size_inside_page(p, count);
152 allowed = page_is_allowed(p >> PAGE_SHIFT);
158 /* Show zeros for restricted memory. */
159 remaining = clear_user(buf, sz);
162 * On ia64 if a page has been mapped somewhere as
163 * uncached, then it must also be accessed uncached
164 * by the kernel or data corruption may occur.
166 ptr = xlate_dev_mem_ptr(p);
170 probe = copy_from_kernel_nofault(bounce, ptr, sz);
171 unxlate_dev_mem_ptr(p, ptr);
175 remaining = copy_to_user(buf, bounce, sz);
185 if (should_stop_iteration())
198 static ssize_t write_mem(struct file *file, const char __user *buf,
199 size_t count, loff_t *ppos)
201 phys_addr_t p = *ppos;
203 unsigned long copied;
209 if (!valid_phys_addr_range(p, count))
214 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
215 /* we don't have page 0 mapped on sparc and m68k.. */
217 sz = size_inside_page(p, count);
218 /* Hmm. Do something? */
229 sz = size_inside_page(p, count);
231 allowed = page_is_allowed(p >> PAGE_SHIFT);
235 /* Skip actual writing when a page is marked as restricted. */
238 * On ia64 if a page has been mapped somewhere as
239 * uncached, then it must also be accessed uncached
240 * by the kernel or data corruption may occur.
242 ptr = xlate_dev_mem_ptr(p);
249 copied = copy_from_user(ptr, buf, sz);
250 unxlate_dev_mem_ptr(p, ptr);
252 written += sz - copied;
263 if (should_stop_iteration())
271 int __weak phys_mem_access_prot_allowed(struct file *file,
272 unsigned long pfn, unsigned long size, pgprot_t *vma_prot)
277 #ifndef __HAVE_PHYS_MEM_ACCESS_PROT
280 * Architectures vary in how they handle caching for addresses
281 * outside of main memory.
284 #ifdef pgprot_noncached
285 static int uncached_access(struct file *file, phys_addr_t addr)
287 #if defined(CONFIG_IA64)
289 * On ia64, we ignore O_DSYNC because we cannot tolerate memory
292 return !(efi_mem_attributes(addr) & EFI_MEMORY_WB);
295 * Accessing memory above the top the kernel knows about or through a
297 * that was marked O_DSYNC will be done non-cached.
299 if (file->f_flags & O_DSYNC)
301 return addr >= __pa(high_memory);
306 static pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
307 unsigned long size, pgprot_t vma_prot)
309 #ifdef pgprot_noncached
310 phys_addr_t offset = pfn << PAGE_SHIFT;
312 if (uncached_access(file, offset))
313 return pgprot_noncached(vma_prot);
320 static unsigned long get_unmapped_area_mem(struct file *file,
326 if (!valid_mmap_phys_addr_range(pgoff, len))
327 return (unsigned long) -EINVAL;
328 return pgoff << PAGE_SHIFT;
331 /* permit direct mmap, for read, write or exec */
332 static unsigned memory_mmap_capabilities(struct file *file)
334 return NOMMU_MAP_DIRECT |
335 NOMMU_MAP_READ | NOMMU_MAP_WRITE | NOMMU_MAP_EXEC;
338 static unsigned zero_mmap_capabilities(struct file *file)
340 return NOMMU_MAP_COPY;
343 /* can't do an in-place private mapping if there's no MMU */
344 static inline int private_mapping_ok(struct vm_area_struct *vma)
346 return vma->vm_flags & VM_MAYSHARE;
350 static inline int private_mapping_ok(struct vm_area_struct *vma)
356 static const struct vm_operations_struct mmap_mem_ops = {
357 #ifdef CONFIG_HAVE_IOREMAP_PROT
358 .access = generic_access_phys
362 static int mmap_mem(struct file *file, struct vm_area_struct *vma)
364 size_t size = vma->vm_end - vma->vm_start;
365 phys_addr_t offset = (phys_addr_t)vma->vm_pgoff << PAGE_SHIFT;
367 /* Does it even fit in phys_addr_t? */
368 if (offset >> PAGE_SHIFT != vma->vm_pgoff)
371 /* It's illegal to wrap around the end of the physical address space. */
372 if (offset + (phys_addr_t)size - 1 < offset)
375 if (!valid_mmap_phys_addr_range(vma->vm_pgoff, size))
378 if (!private_mapping_ok(vma))
381 if (!range_is_allowed(vma->vm_pgoff, size))
384 if (!phys_mem_access_prot_allowed(file, vma->vm_pgoff, size,
388 vma->vm_page_prot = phys_mem_access_prot(file, vma->vm_pgoff,
392 vma->vm_ops = &mmap_mem_ops;
394 /* Remap-pfn-range will mark the range VM_IO */
395 if (remap_pfn_range(vma,
399 vma->vm_page_prot)) {
405 static ssize_t read_port(struct file *file, char __user *buf,
406 size_t count, loff_t *ppos)
408 unsigned long i = *ppos;
409 char __user *tmp = buf;
411 if (!access_ok(buf, count))
413 while (count-- > 0 && i < 65536) {
414 if (__put_user(inb(i), tmp) < 0)
423 static ssize_t write_port(struct file *file, const char __user *buf,
424 size_t count, loff_t *ppos)
426 unsigned long i = *ppos;
427 const char __user *tmp = buf;
429 if (!access_ok(buf, count))
431 while (count-- > 0 && i < 65536) {
434 if (__get_user(c, tmp)) {
447 static ssize_t read_null(struct file *file, char __user *buf,
448 size_t count, loff_t *ppos)
453 static ssize_t write_null(struct file *file, const char __user *buf,
454 size_t count, loff_t *ppos)
459 static ssize_t read_iter_null(struct kiocb *iocb, struct iov_iter *to)
464 static ssize_t write_iter_null(struct kiocb *iocb, struct iov_iter *from)
466 size_t count = iov_iter_count(from);
467 iov_iter_advance(from, count);
471 static int pipe_to_null(struct pipe_inode_info *info, struct pipe_buffer *buf,
472 struct splice_desc *sd)
477 static ssize_t splice_write_null(struct pipe_inode_info *pipe, struct file *out,
478 loff_t *ppos, size_t len, unsigned int flags)
480 return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_null);
483 static int uring_cmd_null(struct io_uring_cmd *ioucmd, unsigned int issue_flags)
488 static ssize_t read_iter_zero(struct kiocb *iocb, struct iov_iter *iter)
492 while (iov_iter_count(iter)) {
493 size_t chunk = iov_iter_count(iter), n;
495 if (chunk > PAGE_SIZE)
496 chunk = PAGE_SIZE; /* Just for latency reasons */
497 n = iov_iter_zero(chunk, iter);
498 if (!n && iov_iter_count(iter))
499 return written ? written : -EFAULT;
501 if (signal_pending(current))
502 return written ? written : -ERESTARTSYS;
505 if (iocb->ki_flags & IOCB_NOWAIT)
506 return written ? written : -EAGAIN;
512 static ssize_t read_zero(struct file *file, char __user *buf,
513 size_t count, loff_t *ppos)
518 size_t chunk = min_t(size_t, count, PAGE_SIZE);
521 left = clear_user(buf + cleared, chunk);
522 if (unlikely(left)) {
523 cleared += (chunk - left);
531 if (signal_pending(current))
539 static int mmap_zero(struct file *file, struct vm_area_struct *vma)
544 if (vma->vm_flags & VM_SHARED)
545 return shmem_zero_setup(vma);
546 vma_set_anonymous(vma);
550 static unsigned long get_unmapped_area_zero(struct file *file,
551 unsigned long addr, unsigned long len,
552 unsigned long pgoff, unsigned long flags)
555 if (flags & MAP_SHARED) {
557 * mmap_zero() will call shmem_zero_setup() to create a file,
558 * so use shmem's get_unmapped_area in case it can be huge;
559 * and pass NULL for file as in mmap.c's get_unmapped_area(),
560 * so as not to confuse shmem with our handle on "/dev/zero".
562 return shmem_get_unmapped_area(NULL, addr, len, pgoff, flags);
565 /* Otherwise flags & MAP_PRIVATE: with no shmem object beneath it */
566 return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
572 static ssize_t write_full(struct file *file, const char __user *buf,
573 size_t count, loff_t *ppos)
579 * Special lseek() function for /dev/null and /dev/zero. Most notably, you
580 * can fopen() both devices with "a" now. This was previously impossible.
583 static loff_t null_lseek(struct file *file, loff_t offset, int orig)
585 return file->f_pos = 0;
589 * The memory devices use the full 32/64 bits of the offset, and so we cannot
590 * check against negative addresses: they are ok. The return value is weird,
591 * though, in that case (0).
593 * also note that seeking relative to the "end of file" isn't supported:
594 * it has no meaning, so it returns -EINVAL.
596 static loff_t memory_lseek(struct file *file, loff_t offset, int orig)
600 inode_lock(file_inode(file));
603 offset += file->f_pos;
606 /* to avoid userland mistaking f_pos=-9 as -EBADF=-9 */
607 if ((unsigned long long)offset >= -MAX_ERRNO) {
611 file->f_pos = offset;
613 force_successful_syscall_return();
618 inode_unlock(file_inode(file));
622 static int open_port(struct inode *inode, struct file *filp)
626 if (!capable(CAP_SYS_RAWIO))
629 rc = security_locked_down(LOCKDOWN_DEV_MEM);
633 if (iminor(inode) != DEVMEM_MINOR)
637 * Use a unified address space to have a single point to manage
638 * revocations when drivers want to take over a /dev/mem mapped
641 filp->f_mapping = iomem_get_mapping();
646 #define zero_lseek null_lseek
647 #define full_lseek null_lseek
648 #define write_zero write_null
649 #define write_iter_zero write_iter_null
650 #define open_mem open_port
652 static const struct file_operations __maybe_unused mem_fops = {
653 .llseek = memory_lseek,
659 .get_unmapped_area = get_unmapped_area_mem,
660 .mmap_capabilities = memory_mmap_capabilities,
664 static const struct file_operations null_fops = {
665 .llseek = null_lseek,
668 .read_iter = read_iter_null,
669 .write_iter = write_iter_null,
670 .splice_write = splice_write_null,
671 .uring_cmd = uring_cmd_null,
674 static const struct file_operations __maybe_unused port_fops = {
675 .llseek = memory_lseek,
681 static const struct file_operations zero_fops = {
682 .llseek = zero_lseek,
684 .read_iter = read_iter_zero,
686 .write_iter = write_iter_zero,
688 .get_unmapped_area = get_unmapped_area_zero,
690 .mmap_capabilities = zero_mmap_capabilities,
694 static const struct file_operations full_fops = {
695 .llseek = full_lseek,
696 .read_iter = read_iter_zero,
700 static const struct memdev {
703 const struct file_operations *fops;
707 [DEVMEM_MINOR] = { "mem", 0, &mem_fops, FMODE_UNSIGNED_OFFSET },
709 [3] = { "null", 0666, &null_fops, FMODE_NOWAIT },
710 #ifdef CONFIG_DEVPORT
711 [4] = { "port", 0, &port_fops, 0 },
713 [5] = { "zero", 0666, &zero_fops, FMODE_NOWAIT },
714 [7] = { "full", 0666, &full_fops, 0 },
715 [8] = { "random", 0666, &random_fops, FMODE_NOWAIT },
716 [9] = { "urandom", 0666, &urandom_fops, FMODE_NOWAIT },
718 [11] = { "kmsg", 0644, &kmsg_fops, 0 },
722 static int memory_open(struct inode *inode, struct file *filp)
725 const struct memdev *dev;
727 minor = iminor(inode);
728 if (minor >= ARRAY_SIZE(devlist))
731 dev = &devlist[minor];
735 filp->f_op = dev->fops;
736 filp->f_mode |= dev->fmode;
739 return dev->fops->open(inode, filp);
744 static const struct file_operations memory_fops = {
746 .llseek = noop_llseek,
749 static char *mem_devnode(struct device *dev, umode_t *mode)
751 if (mode && devlist[MINOR(dev->devt)].mode)
752 *mode = devlist[MINOR(dev->devt)].mode;
756 static struct class *mem_class;
758 static int __init chr_dev_init(void)
762 if (register_chrdev(MEM_MAJOR, "mem", &memory_fops))
763 printk("unable to get major %d for memory devs\n", MEM_MAJOR);
765 mem_class = class_create(THIS_MODULE, "mem");
766 if (IS_ERR(mem_class))
767 return PTR_ERR(mem_class);
769 mem_class->devnode = mem_devnode;
770 for (minor = 1; minor < ARRAY_SIZE(devlist); minor++) {
771 if (!devlist[minor].name)
777 if ((minor == DEVPORT_MINOR) && !arch_has_dev_port())
780 device_create(mem_class, NULL, MKDEV(MEM_MAJOR, minor),
781 NULL, devlist[minor].name);
787 fs_initcall(chr_dev_init);