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/splice.h>
26 #include <linux/pfn.h>
27 #include <linux/export.h>
29 #include <linux/aio.h>
31 #include <asm/uaccess.h>
34 # include <linux/efi.h>
37 #define DEVPORT_MINOR 4
39 static inline unsigned long size_inside_page(unsigned long start,
44 sz = PAGE_SIZE - (start & (PAGE_SIZE - 1));
49 #ifndef ARCH_HAS_VALID_PHYS_ADDR_RANGE
50 static inline int valid_phys_addr_range(phys_addr_t addr, size_t count)
52 return addr + count <= __pa(high_memory);
55 static inline int valid_mmap_phys_addr_range(unsigned long pfn, size_t size)
61 #ifdef CONFIG_STRICT_DEVMEM
62 static inline int range_is_allowed(unsigned long pfn, unsigned long size)
64 u64 from = ((u64)pfn) << PAGE_SHIFT;
69 if (!devmem_is_allowed(pfn)) {
71 "Program %s tried to access /dev/mem between %Lx->%Lx.\n",
72 current->comm, from, to);
81 static inline int range_is_allowed(unsigned long pfn, unsigned long size)
87 #ifndef unxlate_dev_mem_ptr
88 #define unxlate_dev_mem_ptr unxlate_dev_mem_ptr
89 void __weak unxlate_dev_mem_ptr(phys_addr_t phys, void *addr)
95 * This funcion reads the *physical* memory. The f_pos points directly to the
98 static ssize_t read_mem(struct file *file, char __user *buf,
99 size_t count, loff_t *ppos)
101 phys_addr_t p = *ppos;
108 if (!valid_phys_addr_range(p, count))
111 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
112 /* we don't have page 0 mapped on sparc and m68k.. */
114 sz = size_inside_page(p, count);
116 if (clear_user(buf, sz))
127 unsigned long remaining;
129 sz = size_inside_page(p, count);
131 if (!range_is_allowed(p >> PAGE_SHIFT, count))
135 * On ia64 if a page has been mapped somewhere as uncached, then
136 * it must also be accessed uncached by the kernel or data
137 * corruption may occur.
139 ptr = xlate_dev_mem_ptr(p);
143 remaining = copy_to_user(buf, ptr, sz);
144 unxlate_dev_mem_ptr(p, ptr);
158 static ssize_t write_mem(struct file *file, const char __user *buf,
159 size_t count, loff_t *ppos)
161 phys_addr_t p = *ppos;
163 unsigned long copied;
169 if (!valid_phys_addr_range(p, count))
174 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
175 /* we don't have page 0 mapped on sparc and m68k.. */
177 sz = size_inside_page(p, count);
178 /* Hmm. Do something? */
187 sz = size_inside_page(p, count);
189 if (!range_is_allowed(p >> PAGE_SHIFT, sz))
193 * On ia64 if a page has been mapped somewhere as uncached, then
194 * it must also be accessed uncached by the kernel or data
195 * corruption may occur.
197 ptr = xlate_dev_mem_ptr(p);
204 copied = copy_from_user(ptr, buf, sz);
205 unxlate_dev_mem_ptr(p, ptr);
207 written += sz - copied;
223 int __weak phys_mem_access_prot_allowed(struct file *file,
224 unsigned long pfn, unsigned long size, pgprot_t *vma_prot)
229 #ifndef __HAVE_PHYS_MEM_ACCESS_PROT
232 * Architectures vary in how they handle caching for addresses
233 * outside of main memory.
236 #ifdef pgprot_noncached
237 static int uncached_access(struct file *file, phys_addr_t addr)
239 #if defined(CONFIG_IA64)
241 * On ia64, we ignore O_DSYNC because we cannot tolerate memory
244 return !(efi_mem_attributes(addr) & EFI_MEMORY_WB);
245 #elif defined(CONFIG_MIPS)
247 extern int __uncached_access(struct file *file,
250 return __uncached_access(file, addr);
254 * Accessing memory above the top the kernel knows about or through a
256 * that was marked O_DSYNC will be done non-cached.
258 if (file->f_flags & O_DSYNC)
260 return addr >= __pa(high_memory);
265 static pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
266 unsigned long size, pgprot_t vma_prot)
268 #ifdef pgprot_noncached
269 phys_addr_t offset = pfn << PAGE_SHIFT;
271 if (uncached_access(file, offset))
272 return pgprot_noncached(vma_prot);
279 static unsigned long get_unmapped_area_mem(struct file *file,
285 if (!valid_mmap_phys_addr_range(pgoff, len))
286 return (unsigned long) -EINVAL;
287 return pgoff << PAGE_SHIFT;
290 /* can't do an in-place private mapping if there's no MMU */
291 static inline int private_mapping_ok(struct vm_area_struct *vma)
293 return vma->vm_flags & VM_MAYSHARE;
296 #define get_unmapped_area_mem NULL
298 static inline int private_mapping_ok(struct vm_area_struct *vma)
304 static const struct vm_operations_struct mmap_mem_ops = {
305 #ifdef CONFIG_HAVE_IOREMAP_PROT
306 .access = generic_access_phys
310 static int mmap_mem(struct file *file, struct vm_area_struct *vma)
312 size_t size = vma->vm_end - vma->vm_start;
314 if (!valid_mmap_phys_addr_range(vma->vm_pgoff, size))
317 if (!private_mapping_ok(vma))
320 if (!range_is_allowed(vma->vm_pgoff, size))
323 if (!phys_mem_access_prot_allowed(file, vma->vm_pgoff, size,
327 vma->vm_page_prot = phys_mem_access_prot(file, vma->vm_pgoff,
331 vma->vm_ops = &mmap_mem_ops;
333 /* Remap-pfn-range will mark the range VM_IO */
334 if (remap_pfn_range(vma,
338 vma->vm_page_prot)) {
344 #ifdef CONFIG_DEVKMEM
345 static int mmap_kmem(struct file *file, struct vm_area_struct *vma)
349 /* Turn a kernel-virtual address into a physical page frame */
350 pfn = __pa((u64)vma->vm_pgoff << PAGE_SHIFT) >> PAGE_SHIFT;
353 * RED-PEN: on some architectures there is more mapped memory than
354 * available in mem_map which pfn_valid checks for. Perhaps should add a
357 * RED-PEN: vmalloc is not supported right now.
363 return mmap_mem(file, vma);
367 #ifdef CONFIG_DEVKMEM
369 * This function reads the *virtual* memory as seen by the kernel.
371 static ssize_t read_kmem(struct file *file, char __user *buf,
372 size_t count, loff_t *ppos)
374 unsigned long p = *ppos;
375 ssize_t low_count, read, sz;
376 char *kbuf; /* k-addr because vread() takes vmlist_lock rwlock */
380 if (p < (unsigned long) high_memory) {
382 if (count > (unsigned long)high_memory - p)
383 low_count = (unsigned long)high_memory - p;
385 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
386 /* we don't have page 0 mapped on sparc and m68k.. */
387 if (p < PAGE_SIZE && low_count > 0) {
388 sz = size_inside_page(p, low_count);
389 if (clear_user(buf, sz))
398 while (low_count > 0) {
399 sz = size_inside_page(p, low_count);
402 * On ia64 if a page has been mapped somewhere as
403 * uncached, then it must also be accessed uncached
404 * by the kernel or data corruption may occur
406 kbuf = xlate_dev_kmem_ptr((void *)p);
408 if (copy_to_user(buf, kbuf, sz))
419 kbuf = (char *)__get_free_page(GFP_KERNEL);
423 sz = size_inside_page(p, count);
424 if (!is_vmalloc_or_module_addr((void *)p)) {
428 sz = vread(kbuf, (char *)p, sz);
431 if (copy_to_user(buf, kbuf, sz)) {
440 free_page((unsigned long)kbuf);
443 return read ? read : err;
447 static ssize_t do_write_kmem(unsigned long p, const char __user *buf,
448 size_t count, loff_t *ppos)
451 unsigned long copied;
454 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
455 /* we don't have page 0 mapped on sparc and m68k.. */
457 sz = size_inside_page(p, count);
458 /* Hmm. Do something? */
469 sz = size_inside_page(p, count);
472 * On ia64 if a page has been mapped somewhere as uncached, then
473 * it must also be accessed uncached by the kernel or data
474 * corruption may occur.
476 ptr = xlate_dev_kmem_ptr((void *)p);
478 copied = copy_from_user(ptr, buf, sz);
480 written += sz - copied;
496 * This function writes to the *virtual* memory as seen by the kernel.
498 static ssize_t write_kmem(struct file *file, const char __user *buf,
499 size_t count, loff_t *ppos)
501 unsigned long p = *ppos;
504 char *kbuf; /* k-addr because vwrite() takes vmlist_lock rwlock */
507 if (p < (unsigned long) high_memory) {
508 unsigned long to_write = min_t(unsigned long, count,
509 (unsigned long)high_memory - p);
510 wrote = do_write_kmem(p, buf, to_write, ppos);
511 if (wrote != to_write)
519 kbuf = (char *)__get_free_page(GFP_KERNEL);
521 return wrote ? wrote : -ENOMEM;
523 unsigned long sz = size_inside_page(p, count);
526 if (!is_vmalloc_or_module_addr((void *)p)) {
530 n = copy_from_user(kbuf, buf, sz);
535 vwrite(kbuf, (char *)p, sz);
541 free_page((unsigned long)kbuf);
545 return virtr + wrote ? : err;
549 #ifdef CONFIG_DEVPORT
550 static ssize_t read_port(struct file *file, char __user *buf,
551 size_t count, loff_t *ppos)
553 unsigned long i = *ppos;
554 char __user *tmp = buf;
556 if (!access_ok(VERIFY_WRITE, buf, count))
558 while (count-- > 0 && i < 65536) {
559 if (__put_user(inb(i), tmp) < 0)
568 static ssize_t write_port(struct file *file, const char __user *buf,
569 size_t count, loff_t *ppos)
571 unsigned long i = *ppos;
572 const char __user *tmp = buf;
574 if (!access_ok(VERIFY_READ, buf, count))
576 while (count-- > 0 && i < 65536) {
578 if (__get_user(c, tmp)) {
592 static ssize_t read_null(struct file *file, char __user *buf,
593 size_t count, loff_t *ppos)
598 static ssize_t write_null(struct file *file, const char __user *buf,
599 size_t count, loff_t *ppos)
604 static ssize_t aio_read_null(struct kiocb *iocb, const struct iovec *iov,
605 unsigned long nr_segs, loff_t pos)
610 static ssize_t aio_write_null(struct kiocb *iocb, const struct iovec *iov,
611 unsigned long nr_segs, loff_t pos)
613 return iov_length(iov, nr_segs);
616 static int pipe_to_null(struct pipe_inode_info *info, struct pipe_buffer *buf,
617 struct splice_desc *sd)
622 static ssize_t splice_write_null(struct pipe_inode_info *pipe, struct file *out,
623 loff_t *ppos, size_t len, unsigned int flags)
625 return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_null);
628 static ssize_t read_iter_zero(struct kiocb *iocb, struct iov_iter *iter)
632 while (iov_iter_count(iter)) {
633 size_t chunk = iov_iter_count(iter), n;
634 if (chunk > PAGE_SIZE)
635 chunk = PAGE_SIZE; /* Just for latency reasons */
636 n = iov_iter_zero(chunk, iter);
637 if (!n && iov_iter_count(iter))
638 return written ? written : -EFAULT;
640 if (signal_pending(current))
641 return written ? written : -ERESTARTSYS;
647 static int mmap_zero(struct file *file, struct vm_area_struct *vma)
652 if (vma->vm_flags & VM_SHARED)
653 return shmem_zero_setup(vma);
657 static ssize_t write_full(struct file *file, const char __user *buf,
658 size_t count, loff_t *ppos)
664 * Special lseek() function for /dev/null and /dev/zero. Most notably, you
665 * can fopen() both devices with "a" now. This was previously impossible.
668 static loff_t null_lseek(struct file *file, loff_t offset, int orig)
670 return file->f_pos = 0;
674 * The memory devices use the full 32/64 bits of the offset, and so we cannot
675 * check against negative addresses: they are ok. The return value is weird,
676 * though, in that case (0).
678 * also note that seeking relative to the "end of file" isn't supported:
679 * it has no meaning, so it returns -EINVAL.
681 static loff_t memory_lseek(struct file *file, loff_t offset, int orig)
685 mutex_lock(&file_inode(file)->i_mutex);
688 offset += file->f_pos;
690 /* to avoid userland mistaking f_pos=-9 as -EBADF=-9 */
691 if (IS_ERR_VALUE((unsigned long long)offset)) {
695 file->f_pos = offset;
697 force_successful_syscall_return();
702 mutex_unlock(&file_inode(file)->i_mutex);
706 static int open_port(struct inode *inode, struct file *filp)
708 return capable(CAP_SYS_RAWIO) ? 0 : -EPERM;
711 #define zero_lseek null_lseek
712 #define full_lseek null_lseek
713 #define write_zero write_null
714 #define aio_write_zero aio_write_null
715 #define open_mem open_port
716 #define open_kmem open_mem
718 static const struct file_operations mem_fops = {
719 .llseek = memory_lseek,
724 .get_unmapped_area = get_unmapped_area_mem,
727 #ifdef CONFIG_DEVKMEM
728 static const struct file_operations kmem_fops = {
729 .llseek = memory_lseek,
734 .get_unmapped_area = get_unmapped_area_mem,
738 static const struct file_operations null_fops = {
739 .llseek = null_lseek,
742 .aio_read = aio_read_null,
743 .aio_write = aio_write_null,
744 .splice_write = splice_write_null,
747 #ifdef CONFIG_DEVPORT
748 static const struct file_operations port_fops = {
749 .llseek = memory_lseek,
756 static const struct file_operations zero_fops = {
757 .llseek = zero_lseek,
758 .read = new_sync_read,
760 .read_iter = read_iter_zero,
761 .aio_write = aio_write_zero,
766 * capabilities for /dev/zero
767 * - permits private mappings, "copies" are taken of the source of zeros
768 * - no writeback happens
770 static struct backing_dev_info zero_bdi = {
772 .capabilities = BDI_CAP_MAP_COPY | BDI_CAP_NO_ACCT_AND_WRITEBACK,
775 static const struct file_operations full_fops = {
776 .llseek = full_lseek,
777 .read = new_sync_read,
778 .read_iter = read_iter_zero,
782 static const struct memdev {
785 const struct file_operations *fops;
786 struct backing_dev_info *dev_info;
788 [1] = { "mem", 0, &mem_fops, &directly_mappable_cdev_bdi },
789 #ifdef CONFIG_DEVKMEM
790 [2] = { "kmem", 0, &kmem_fops, &directly_mappable_cdev_bdi },
792 [3] = { "null", 0666, &null_fops, NULL },
793 #ifdef CONFIG_DEVPORT
794 [4] = { "port", 0, &port_fops, NULL },
796 [5] = { "zero", 0666, &zero_fops, &zero_bdi },
797 [7] = { "full", 0666, &full_fops, NULL },
798 [8] = { "random", 0666, &random_fops, NULL },
799 [9] = { "urandom", 0666, &urandom_fops, NULL },
801 [11] = { "kmsg", 0644, &kmsg_fops, NULL },
805 static int memory_open(struct inode *inode, struct file *filp)
808 const struct memdev *dev;
810 minor = iminor(inode);
811 if (minor >= ARRAY_SIZE(devlist))
814 dev = &devlist[minor];
818 filp->f_op = dev->fops;
820 filp->f_mapping->backing_dev_info = dev->dev_info;
822 /* Is /dev/mem or /dev/kmem ? */
823 if (dev->dev_info == &directly_mappable_cdev_bdi)
824 filp->f_mode |= FMODE_UNSIGNED_OFFSET;
827 return dev->fops->open(inode, filp);
832 static const struct file_operations memory_fops = {
834 .llseek = noop_llseek,
837 static char *mem_devnode(struct device *dev, umode_t *mode)
839 if (mode && devlist[MINOR(dev->devt)].mode)
840 *mode = devlist[MINOR(dev->devt)].mode;
844 static struct class *mem_class;
846 static int __init chr_dev_init(void)
851 err = bdi_init(&zero_bdi);
855 if (register_chrdev(MEM_MAJOR, "mem", &memory_fops))
856 printk("unable to get major %d for memory devs\n", MEM_MAJOR);
858 mem_class = class_create(THIS_MODULE, "mem");
859 if (IS_ERR(mem_class))
860 return PTR_ERR(mem_class);
862 mem_class->devnode = mem_devnode;
863 for (minor = 1; minor < ARRAY_SIZE(devlist); minor++) {
864 if (!devlist[minor].name)
870 if ((minor == DEVPORT_MINOR) && !arch_has_dev_port())
873 device_create(mem_class, NULL, MKDEV(MEM_MAJOR, minor),
874 NULL, devlist[minor].name);
880 fs_initcall(chr_dev_init);