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 static inline bool should_stop_iteration(void)
97 return signal_pending(current);
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;
116 if (!valid_phys_addr_range(p, count))
119 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
120 /* we don't have page 0 mapped on sparc and m68k.. */
122 sz = size_inside_page(p, count);
124 if (clear_user(buf, sz))
134 bounce = kmalloc(PAGE_SIZE, GFP_KERNEL);
139 unsigned long remaining;
142 sz = size_inside_page(p, count);
145 allowed = page_is_allowed(p >> PAGE_SHIFT);
151 /* Show zeros for restricted memory. */
152 remaining = clear_user(buf, sz);
155 * On ia64 if a page has been mapped somewhere as
156 * uncached, then it must also be accessed uncached
157 * by the kernel or data corruption may occur.
159 ptr = xlate_dev_mem_ptr(p);
163 probe = copy_from_kernel_nofault(bounce, ptr, sz);
164 unxlate_dev_mem_ptr(p, ptr);
168 remaining = copy_to_user(buf, bounce, sz);
178 if (should_stop_iteration())
191 static ssize_t write_mem(struct file *file, const char __user *buf,
192 size_t count, loff_t *ppos)
194 phys_addr_t p = *ppos;
196 unsigned long copied;
202 if (!valid_phys_addr_range(p, count))
207 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
208 /* we don't have page 0 mapped on sparc and m68k.. */
210 sz = size_inside_page(p, count);
211 /* Hmm. Do something? */
222 sz = size_inside_page(p, count);
224 allowed = page_is_allowed(p >> PAGE_SHIFT);
228 /* Skip actual writing when a page is marked as restricted. */
231 * On ia64 if a page has been mapped somewhere as
232 * uncached, then it must also be accessed uncached
233 * by the kernel or data corruption may occur.
235 ptr = xlate_dev_mem_ptr(p);
242 copied = copy_from_user(ptr, buf, sz);
243 unxlate_dev_mem_ptr(p, ptr);
245 written += sz - copied;
256 if (should_stop_iteration())
264 int __weak phys_mem_access_prot_allowed(struct file *file,
265 unsigned long pfn, unsigned long size, pgprot_t *vma_prot)
270 #ifndef __HAVE_PHYS_MEM_ACCESS_PROT
273 * Architectures vary in how they handle caching for addresses
274 * outside of main memory.
277 #ifdef pgprot_noncached
278 static int uncached_access(struct file *file, phys_addr_t addr)
280 #if defined(CONFIG_IA64)
282 * On ia64, we ignore O_DSYNC because we cannot tolerate memory
285 return !(efi_mem_attributes(addr) & EFI_MEMORY_WB);
288 * Accessing memory above the top the kernel knows about or through a
290 * that was marked O_DSYNC will be done non-cached.
292 if (file->f_flags & O_DSYNC)
294 return addr >= __pa(high_memory);
299 static pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
300 unsigned long size, pgprot_t vma_prot)
302 #ifdef pgprot_noncached
303 phys_addr_t offset = pfn << PAGE_SHIFT;
305 if (uncached_access(file, offset))
306 return pgprot_noncached(vma_prot);
313 static unsigned long get_unmapped_area_mem(struct file *file,
319 if (!valid_mmap_phys_addr_range(pgoff, len))
320 return (unsigned long) -EINVAL;
321 return pgoff << PAGE_SHIFT;
324 /* permit direct mmap, for read, write or exec */
325 static unsigned memory_mmap_capabilities(struct file *file)
327 return NOMMU_MAP_DIRECT |
328 NOMMU_MAP_READ | NOMMU_MAP_WRITE | NOMMU_MAP_EXEC;
331 static unsigned zero_mmap_capabilities(struct file *file)
333 return NOMMU_MAP_COPY;
336 /* can't do an in-place private mapping if there's no MMU */
337 static inline int private_mapping_ok(struct vm_area_struct *vma)
339 return is_nommu_shared_mapping(vma->vm_flags);
343 static inline int private_mapping_ok(struct vm_area_struct *vma)
349 static const struct vm_operations_struct mmap_mem_ops = {
350 #ifdef CONFIG_HAVE_IOREMAP_PROT
351 .access = generic_access_phys
355 static int mmap_mem(struct file *file, struct vm_area_struct *vma)
357 size_t size = vma->vm_end - vma->vm_start;
358 phys_addr_t offset = (phys_addr_t)vma->vm_pgoff << PAGE_SHIFT;
360 /* Does it even fit in phys_addr_t? */
361 if (offset >> PAGE_SHIFT != vma->vm_pgoff)
364 /* It's illegal to wrap around the end of the physical address space. */
365 if (offset + (phys_addr_t)size - 1 < offset)
368 if (!valid_mmap_phys_addr_range(vma->vm_pgoff, size))
371 if (!private_mapping_ok(vma))
374 if (!range_is_allowed(vma->vm_pgoff, size))
377 if (!phys_mem_access_prot_allowed(file, vma->vm_pgoff, size,
381 vma->vm_page_prot = phys_mem_access_prot(file, vma->vm_pgoff,
385 vma->vm_ops = &mmap_mem_ops;
387 /* Remap-pfn-range will mark the range VM_IO */
388 if (remap_pfn_range(vma,
392 vma->vm_page_prot)) {
398 static ssize_t read_port(struct file *file, char __user *buf,
399 size_t count, loff_t *ppos)
401 unsigned long i = *ppos;
402 char __user *tmp = buf;
404 if (!access_ok(buf, count))
406 while (count-- > 0 && i < 65536) {
407 if (__put_user(inb(i), tmp) < 0)
416 static ssize_t write_port(struct file *file, const char __user *buf,
417 size_t count, loff_t *ppos)
419 unsigned long i = *ppos;
420 const char __user *tmp = buf;
422 if (!access_ok(buf, count))
424 while (count-- > 0 && i < 65536) {
427 if (__get_user(c, tmp)) {
440 static ssize_t read_null(struct file *file, char __user *buf,
441 size_t count, loff_t *ppos)
446 static ssize_t write_null(struct file *file, const char __user *buf,
447 size_t count, loff_t *ppos)
452 static ssize_t read_iter_null(struct kiocb *iocb, struct iov_iter *to)
457 static ssize_t write_iter_null(struct kiocb *iocb, struct iov_iter *from)
459 size_t count = iov_iter_count(from);
460 iov_iter_advance(from, count);
464 static int pipe_to_null(struct pipe_inode_info *info, struct pipe_buffer *buf,
465 struct splice_desc *sd)
470 static ssize_t splice_write_null(struct pipe_inode_info *pipe, struct file *out,
471 loff_t *ppos, size_t len, unsigned int flags)
473 return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_null);
476 static int uring_cmd_null(struct io_uring_cmd *ioucmd, unsigned int issue_flags)
481 static ssize_t read_iter_zero(struct kiocb *iocb, struct iov_iter *iter)
485 while (iov_iter_count(iter)) {
486 size_t chunk = iov_iter_count(iter), n;
488 if (chunk > PAGE_SIZE)
489 chunk = PAGE_SIZE; /* Just for latency reasons */
490 n = iov_iter_zero(chunk, iter);
491 if (!n && iov_iter_count(iter))
492 return written ? written : -EFAULT;
494 if (signal_pending(current))
495 return written ? written : -ERESTARTSYS;
498 if (iocb->ki_flags & IOCB_NOWAIT)
499 return written ? written : -EAGAIN;
505 static ssize_t read_zero(struct file *file, char __user *buf,
506 size_t count, loff_t *ppos)
511 size_t chunk = min_t(size_t, count, PAGE_SIZE);
514 left = clear_user(buf + cleared, chunk);
515 if (unlikely(left)) {
516 cleared += (chunk - left);
524 if (signal_pending(current))
532 static int mmap_zero(struct file *file, struct vm_area_struct *vma)
537 if (vma->vm_flags & VM_SHARED)
538 return shmem_zero_setup(vma);
539 vma_set_anonymous(vma);
543 static unsigned long get_unmapped_area_zero(struct file *file,
544 unsigned long addr, unsigned long len,
545 unsigned long pgoff, unsigned long flags)
548 if (flags & MAP_SHARED) {
550 * mmap_zero() will call shmem_zero_setup() to create a file,
551 * so use shmem's get_unmapped_area in case it can be huge;
552 * and pass NULL for file as in mmap.c's get_unmapped_area(),
553 * so as not to confuse shmem with our handle on "/dev/zero".
555 return shmem_get_unmapped_area(NULL, addr, len, pgoff, flags);
558 /* Otherwise flags & MAP_PRIVATE: with no shmem object beneath it */
559 return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
565 static ssize_t write_full(struct file *file, const char __user *buf,
566 size_t count, loff_t *ppos)
572 * Special lseek() function for /dev/null and /dev/zero. Most notably, you
573 * can fopen() both devices with "a" now. This was previously impossible.
576 static loff_t null_lseek(struct file *file, loff_t offset, int orig)
578 return file->f_pos = 0;
582 * The memory devices use the full 32/64 bits of the offset, and so we cannot
583 * check against negative addresses: they are ok. The return value is weird,
584 * though, in that case (0).
586 * also note that seeking relative to the "end of file" isn't supported:
587 * it has no meaning, so it returns -EINVAL.
589 static loff_t memory_lseek(struct file *file, loff_t offset, int orig)
593 inode_lock(file_inode(file));
596 offset += file->f_pos;
599 /* to avoid userland mistaking f_pos=-9 as -EBADF=-9 */
600 if ((unsigned long long)offset >= -MAX_ERRNO) {
604 file->f_pos = offset;
606 force_successful_syscall_return();
611 inode_unlock(file_inode(file));
615 static int open_port(struct inode *inode, struct file *filp)
619 if (!capable(CAP_SYS_RAWIO))
622 rc = security_locked_down(LOCKDOWN_DEV_MEM);
626 if (iminor(inode) != DEVMEM_MINOR)
630 * Use a unified address space to have a single point to manage
631 * revocations when drivers want to take over a /dev/mem mapped
634 filp->f_mapping = iomem_get_mapping();
639 #define zero_lseek null_lseek
640 #define full_lseek null_lseek
641 #define write_zero write_null
642 #define write_iter_zero write_iter_null
643 #define open_mem open_port
645 static const struct file_operations __maybe_unused mem_fops = {
646 .llseek = memory_lseek,
652 .get_unmapped_area = get_unmapped_area_mem,
653 .mmap_capabilities = memory_mmap_capabilities,
657 static const struct file_operations null_fops = {
658 .llseek = null_lseek,
661 .read_iter = read_iter_null,
662 .write_iter = write_iter_null,
663 .splice_write = splice_write_null,
664 .uring_cmd = uring_cmd_null,
667 static const struct file_operations __maybe_unused port_fops = {
668 .llseek = memory_lseek,
674 static const struct file_operations zero_fops = {
675 .llseek = zero_lseek,
677 .read_iter = read_iter_zero,
679 .write_iter = write_iter_zero,
681 .get_unmapped_area = get_unmapped_area_zero,
683 .mmap_capabilities = zero_mmap_capabilities,
687 static const struct file_operations full_fops = {
688 .llseek = full_lseek,
689 .read_iter = read_iter_zero,
693 static const struct memdev {
695 const struct file_operations *fops;
700 [DEVMEM_MINOR] = { "mem", &mem_fops, FMODE_UNSIGNED_OFFSET, 0 },
702 [3] = { "null", &null_fops, FMODE_NOWAIT, 0666 },
703 #ifdef CONFIG_DEVPORT
704 [4] = { "port", &port_fops, 0, 0 },
706 [5] = { "zero", &zero_fops, FMODE_NOWAIT, 0666 },
707 [7] = { "full", &full_fops, 0, 0666 },
708 [8] = { "random", &random_fops, FMODE_NOWAIT, 0666 },
709 [9] = { "urandom", &urandom_fops, FMODE_NOWAIT, 0666 },
711 [11] = { "kmsg", &kmsg_fops, 0, 0644 },
715 static int memory_open(struct inode *inode, struct file *filp)
718 const struct memdev *dev;
720 minor = iminor(inode);
721 if (minor >= ARRAY_SIZE(devlist))
724 dev = &devlist[minor];
728 filp->f_op = dev->fops;
729 filp->f_mode |= dev->fmode;
732 return dev->fops->open(inode, filp);
737 static const struct file_operations memory_fops = {
739 .llseek = noop_llseek,
742 static char *mem_devnode(const struct device *dev, umode_t *mode)
744 if (mode && devlist[MINOR(dev->devt)].mode)
745 *mode = devlist[MINOR(dev->devt)].mode;
749 static const struct class mem_class = {
751 .devnode = mem_devnode,
754 static int __init chr_dev_init(void)
759 if (register_chrdev(MEM_MAJOR, "mem", &memory_fops))
760 printk("unable to get major %d for memory devs\n", MEM_MAJOR);
762 retval = class_register(&mem_class);
766 for (minor = 1; minor < ARRAY_SIZE(devlist); minor++) {
767 if (!devlist[minor].name)
773 if ((minor == DEVPORT_MINOR) && !arch_has_dev_port())
776 device_create(&mem_class, NULL, MKDEV(MEM_MAJOR, minor),
777 NULL, devlist[minor].name);
783 fs_initcall(chr_dev_init);