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)
99 static inline bool should_stop_iteration(void)
103 return fatal_signal_pending(current);
107 * This funcion reads the *physical* memory. The f_pos points directly to the
110 static ssize_t read_mem(struct file *file, char __user *buf,
111 size_t count, loff_t *ppos)
113 phys_addr_t p = *ppos;
120 if (!valid_phys_addr_range(p, count))
123 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
124 /* we don't have page 0 mapped on sparc and m68k.. */
126 sz = size_inside_page(p, count);
128 if (clear_user(buf, sz))
139 unsigned long remaining;
142 sz = size_inside_page(p, count);
144 allowed = page_is_allowed(p >> PAGE_SHIFT);
148 /* Show zeros for restricted memory. */
149 remaining = clear_user(buf, sz);
152 * On ia64 if a page has been mapped somewhere as
153 * uncached, then it must also be accessed uncached
154 * by the kernel or data corruption may occur.
156 ptr = xlate_dev_mem_ptr(p);
160 remaining = copy_to_user(buf, ptr, sz);
162 unxlate_dev_mem_ptr(p, ptr);
172 if (should_stop_iteration())
180 static ssize_t write_mem(struct file *file, const char __user *buf,
181 size_t count, loff_t *ppos)
183 phys_addr_t p = *ppos;
185 unsigned long copied;
191 if (!valid_phys_addr_range(p, count))
196 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
197 /* we don't have page 0 mapped on sparc and m68k.. */
199 sz = size_inside_page(p, count);
200 /* Hmm. Do something? */
211 sz = size_inside_page(p, count);
213 allowed = page_is_allowed(p >> PAGE_SHIFT);
217 /* Skip actual writing when a page is marked as restricted. */
220 * On ia64 if a page has been mapped somewhere as
221 * uncached, then it must also be accessed uncached
222 * by the kernel or data corruption may occur.
224 ptr = xlate_dev_mem_ptr(p);
231 copied = copy_from_user(ptr, buf, sz);
232 unxlate_dev_mem_ptr(p, ptr);
234 written += sz - copied;
245 if (should_stop_iteration())
253 int __weak phys_mem_access_prot_allowed(struct file *file,
254 unsigned long pfn, unsigned long size, pgprot_t *vma_prot)
259 #ifndef __HAVE_PHYS_MEM_ACCESS_PROT
262 * Architectures vary in how they handle caching for addresses
263 * outside of main memory.
266 #ifdef pgprot_noncached
267 static int uncached_access(struct file *file, phys_addr_t addr)
269 #if defined(CONFIG_IA64)
271 * On ia64, we ignore O_DSYNC because we cannot tolerate memory
274 return !(efi_mem_attributes(addr) & EFI_MEMORY_WB);
275 #elif defined(CONFIG_MIPS)
277 extern int __uncached_access(struct file *file,
280 return __uncached_access(file, addr);
284 * Accessing memory above the top the kernel knows about or through a
286 * that was marked O_DSYNC will be done non-cached.
288 if (file->f_flags & O_DSYNC)
290 return addr >= __pa(high_memory);
295 static pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
296 unsigned long size, pgprot_t vma_prot)
298 #ifdef pgprot_noncached
299 phys_addr_t offset = pfn << PAGE_SHIFT;
301 if (uncached_access(file, offset))
302 return pgprot_noncached(vma_prot);
309 static unsigned long get_unmapped_area_mem(struct file *file,
315 if (!valid_mmap_phys_addr_range(pgoff, len))
316 return (unsigned long) -EINVAL;
317 return pgoff << PAGE_SHIFT;
320 /* permit direct mmap, for read, write or exec */
321 static unsigned memory_mmap_capabilities(struct file *file)
323 return NOMMU_MAP_DIRECT |
324 NOMMU_MAP_READ | NOMMU_MAP_WRITE | NOMMU_MAP_EXEC;
327 static unsigned zero_mmap_capabilities(struct file *file)
329 return NOMMU_MAP_COPY;
332 /* can't do an in-place private mapping if there's no MMU */
333 static inline int private_mapping_ok(struct vm_area_struct *vma)
335 return vma->vm_flags & VM_MAYSHARE;
339 static inline int private_mapping_ok(struct vm_area_struct *vma)
345 static const struct vm_operations_struct mmap_mem_ops = {
346 #ifdef CONFIG_HAVE_IOREMAP_PROT
347 .access = generic_access_phys
351 static int mmap_mem(struct file *file, struct vm_area_struct *vma)
353 size_t size = vma->vm_end - vma->vm_start;
354 phys_addr_t offset = (phys_addr_t)vma->vm_pgoff << PAGE_SHIFT;
356 /* It's illegal to wrap around the end of the physical address space. */
357 if (offset + (phys_addr_t)size - 1 < offset)
360 if (!valid_mmap_phys_addr_range(vma->vm_pgoff, size))
363 if (!private_mapping_ok(vma))
366 if (!range_is_allowed(vma->vm_pgoff, size))
369 if (!phys_mem_access_prot_allowed(file, vma->vm_pgoff, size,
373 vma->vm_page_prot = phys_mem_access_prot(file, vma->vm_pgoff,
377 vma->vm_ops = &mmap_mem_ops;
379 /* Remap-pfn-range will mark the range VM_IO */
380 if (remap_pfn_range(vma,
384 vma->vm_page_prot)) {
390 static int mmap_kmem(struct file *file, struct vm_area_struct *vma)
394 /* Turn a kernel-virtual address into a physical page frame */
395 pfn = __pa((u64)vma->vm_pgoff << PAGE_SHIFT) >> PAGE_SHIFT;
398 * RED-PEN: on some architectures there is more mapped memory than
399 * available in mem_map which pfn_valid checks for. Perhaps should add a
402 * RED-PEN: vmalloc is not supported right now.
408 return mmap_mem(file, vma);
412 * This function reads the *virtual* memory as seen by the kernel.
414 static ssize_t read_kmem(struct file *file, char __user *buf,
415 size_t count, loff_t *ppos)
417 unsigned long p = *ppos;
418 ssize_t low_count, read, sz;
419 char *kbuf; /* k-addr because vread() takes vmlist_lock rwlock */
423 if (p < (unsigned long) high_memory) {
425 if (count > (unsigned long)high_memory - p)
426 low_count = (unsigned long)high_memory - p;
428 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
429 /* we don't have page 0 mapped on sparc and m68k.. */
430 if (p < PAGE_SIZE && low_count > 0) {
431 sz = size_inside_page(p, low_count);
432 if (clear_user(buf, sz))
441 while (low_count > 0) {
442 sz = size_inside_page(p, low_count);
445 * On ia64 if a page has been mapped somewhere as
446 * uncached, then it must also be accessed uncached
447 * by the kernel or data corruption may occur
449 kbuf = xlate_dev_kmem_ptr((void *)p);
450 if (!virt_addr_valid(kbuf))
453 if (copy_to_user(buf, kbuf, sz))
460 if (should_stop_iteration()) {
468 kbuf = (char *)__get_free_page(GFP_KERNEL);
472 sz = size_inside_page(p, count);
473 if (!is_vmalloc_or_module_addr((void *)p)) {
477 sz = vread(kbuf, (char *)p, sz);
480 if (copy_to_user(buf, kbuf, sz)) {
488 if (should_stop_iteration())
491 free_page((unsigned long)kbuf);
494 return read ? read : err;
498 static ssize_t do_write_kmem(unsigned long p, const char __user *buf,
499 size_t count, loff_t *ppos)
502 unsigned long copied;
505 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
506 /* we don't have page 0 mapped on sparc and m68k.. */
508 sz = size_inside_page(p, count);
509 /* Hmm. Do something? */
520 sz = size_inside_page(p, count);
523 * On ia64 if a page has been mapped somewhere as uncached, then
524 * it must also be accessed uncached by the kernel or data
525 * corruption may occur.
527 ptr = xlate_dev_kmem_ptr((void *)p);
528 if (!virt_addr_valid(ptr))
531 copied = copy_from_user(ptr, buf, sz);
533 written += sz - copied;
542 if (should_stop_iteration())
551 * This function writes to the *virtual* memory as seen by the kernel.
553 static ssize_t write_kmem(struct file *file, const char __user *buf,
554 size_t count, loff_t *ppos)
556 unsigned long p = *ppos;
559 char *kbuf; /* k-addr because vwrite() takes vmlist_lock rwlock */
562 if (p < (unsigned long) high_memory) {
563 unsigned long to_write = min_t(unsigned long, count,
564 (unsigned long)high_memory - p);
565 wrote = do_write_kmem(p, buf, to_write, ppos);
566 if (wrote != to_write)
574 kbuf = (char *)__get_free_page(GFP_KERNEL);
576 return wrote ? wrote : -ENOMEM;
578 unsigned long sz = size_inside_page(p, count);
581 if (!is_vmalloc_or_module_addr((void *)p)) {
585 n = copy_from_user(kbuf, buf, sz);
590 vwrite(kbuf, (char *)p, sz);
595 if (should_stop_iteration())
598 free_page((unsigned long)kbuf);
602 return virtr + wrote ? : err;
605 static ssize_t read_port(struct file *file, char __user *buf,
606 size_t count, loff_t *ppos)
608 unsigned long i = *ppos;
609 char __user *tmp = buf;
611 if (!access_ok(VERIFY_WRITE, buf, count))
613 while (count-- > 0 && i < 65536) {
614 if (__put_user(inb(i), tmp) < 0)
623 static ssize_t write_port(struct file *file, const char __user *buf,
624 size_t count, loff_t *ppos)
626 unsigned long i = *ppos;
627 const char __user *tmp = buf;
629 if (!access_ok(VERIFY_READ, buf, count))
631 while (count-- > 0 && i < 65536) {
634 if (__get_user(c, tmp)) {
647 static ssize_t read_null(struct file *file, char __user *buf,
648 size_t count, loff_t *ppos)
653 static ssize_t write_null(struct file *file, const char __user *buf,
654 size_t count, loff_t *ppos)
659 static ssize_t read_iter_null(struct kiocb *iocb, struct iov_iter *to)
664 static ssize_t write_iter_null(struct kiocb *iocb, struct iov_iter *from)
666 size_t count = iov_iter_count(from);
667 iov_iter_advance(from, count);
671 static int pipe_to_null(struct pipe_inode_info *info, struct pipe_buffer *buf,
672 struct splice_desc *sd)
677 static ssize_t splice_write_null(struct pipe_inode_info *pipe, struct file *out,
678 loff_t *ppos, size_t len, unsigned int flags)
680 return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_null);
683 static ssize_t read_iter_zero(struct kiocb *iocb, struct iov_iter *iter)
687 while (iov_iter_count(iter)) {
688 size_t chunk = iov_iter_count(iter), n;
690 if (chunk > PAGE_SIZE)
691 chunk = PAGE_SIZE; /* Just for latency reasons */
692 n = iov_iter_zero(chunk, iter);
693 if (!n && iov_iter_count(iter))
694 return written ? written : -EFAULT;
696 if (signal_pending(current))
697 return written ? written : -ERESTARTSYS;
703 static int mmap_zero(struct file *file, struct vm_area_struct *vma)
708 if (vma->vm_flags & VM_SHARED)
709 return shmem_zero_setup(vma);
713 static unsigned long get_unmapped_area_zero(struct file *file,
714 unsigned long addr, unsigned long len,
715 unsigned long pgoff, unsigned long flags)
718 if (flags & MAP_SHARED) {
720 * mmap_zero() will call shmem_zero_setup() to create a file,
721 * so use shmem's get_unmapped_area in case it can be huge;
722 * and pass NULL for file as in mmap.c's get_unmapped_area(),
723 * so as not to confuse shmem with our handle on "/dev/zero".
725 return shmem_get_unmapped_area(NULL, addr, len, pgoff, flags);
728 /* Otherwise flags & MAP_PRIVATE: with no shmem object beneath it */
729 return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
735 static ssize_t write_full(struct file *file, const char __user *buf,
736 size_t count, loff_t *ppos)
742 * Special lseek() function for /dev/null and /dev/zero. Most notably, you
743 * can fopen() both devices with "a" now. This was previously impossible.
746 static loff_t null_lseek(struct file *file, loff_t offset, int orig)
748 return file->f_pos = 0;
752 * The memory devices use the full 32/64 bits of the offset, and so we cannot
753 * check against negative addresses: they are ok. The return value is weird,
754 * though, in that case (0).
756 * also note that seeking relative to the "end of file" isn't supported:
757 * it has no meaning, so it returns -EINVAL.
759 static loff_t memory_lseek(struct file *file, loff_t offset, int orig)
763 inode_lock(file_inode(file));
766 offset += file->f_pos;
768 /* to avoid userland mistaking f_pos=-9 as -EBADF=-9 */
769 if ((unsigned long long)offset >= -MAX_ERRNO) {
773 file->f_pos = offset;
775 force_successful_syscall_return();
780 inode_unlock(file_inode(file));
784 static int open_port(struct inode *inode, struct file *filp)
786 return capable(CAP_SYS_RAWIO) ? 0 : -EPERM;
789 #define zero_lseek null_lseek
790 #define full_lseek null_lseek
791 #define write_zero write_null
792 #define write_iter_zero write_iter_null
793 #define open_mem open_port
794 #define open_kmem open_mem
796 static const struct file_operations __maybe_unused mem_fops = {
797 .llseek = memory_lseek,
803 .get_unmapped_area = get_unmapped_area_mem,
804 .mmap_capabilities = memory_mmap_capabilities,
808 static const struct file_operations __maybe_unused kmem_fops = {
809 .llseek = memory_lseek,
815 .get_unmapped_area = get_unmapped_area_mem,
816 .mmap_capabilities = memory_mmap_capabilities,
820 static const struct file_operations null_fops = {
821 .llseek = null_lseek,
824 .read_iter = read_iter_null,
825 .write_iter = write_iter_null,
826 .splice_write = splice_write_null,
829 static const struct file_operations __maybe_unused port_fops = {
830 .llseek = memory_lseek,
836 static const struct file_operations zero_fops = {
837 .llseek = zero_lseek,
839 .read_iter = read_iter_zero,
840 .write_iter = write_iter_zero,
842 .get_unmapped_area = get_unmapped_area_zero,
844 .mmap_capabilities = zero_mmap_capabilities,
848 static const struct file_operations full_fops = {
849 .llseek = full_lseek,
850 .read_iter = read_iter_zero,
854 static const struct memdev {
857 const struct file_operations *fops;
861 [1] = { "mem", 0, &mem_fops, FMODE_UNSIGNED_OFFSET },
863 #ifdef CONFIG_DEVKMEM
864 [2] = { "kmem", 0, &kmem_fops, FMODE_UNSIGNED_OFFSET },
866 [3] = { "null", 0666, &null_fops, 0 },
867 #ifdef CONFIG_DEVPORT
868 [4] = { "port", 0, &port_fops, 0 },
870 [5] = { "zero", 0666, &zero_fops, 0 },
871 [7] = { "full", 0666, &full_fops, 0 },
872 [8] = { "random", 0666, &random_fops, 0 },
873 [9] = { "urandom", 0666, &urandom_fops, 0 },
875 [11] = { "kmsg", 0644, &kmsg_fops, 0 },
879 static int memory_open(struct inode *inode, struct file *filp)
882 const struct memdev *dev;
884 minor = iminor(inode);
885 if (minor >= ARRAY_SIZE(devlist))
888 dev = &devlist[minor];
892 filp->f_op = dev->fops;
893 filp->f_mode |= dev->fmode;
896 return dev->fops->open(inode, filp);
901 static const struct file_operations memory_fops = {
903 .llseek = noop_llseek,
906 static char *mem_devnode(struct device *dev, umode_t *mode)
908 if (mode && devlist[MINOR(dev->devt)].mode)
909 *mode = devlist[MINOR(dev->devt)].mode;
913 static struct class *mem_class;
915 static int __init chr_dev_init(void)
919 if (register_chrdev(MEM_MAJOR, "mem", &memory_fops))
920 printk("unable to get major %d for memory devs\n", MEM_MAJOR);
922 mem_class = class_create(THIS_MODULE, "mem");
923 if (IS_ERR(mem_class))
924 return PTR_ERR(mem_class);
926 mem_class->devnode = mem_devnode;
927 for (minor = 1; minor < ARRAY_SIZE(devlist); minor++) {
928 if (!devlist[minor].name)
934 if ((minor == DEVPORT_MINOR) && !arch_has_dev_port())
937 device_create(mem_class, NULL, MKDEV(MEM_MAJOR, minor),
938 NULL, devlist[minor].name);
944 fs_initcall(chr_dev_init);