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 mmaping 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/smp_lock.h>
22 #include <linux/ptrace.h>
23 #include <linux/device.h>
24 #include <linux/highmem.h>
25 #include <linux/crash_dump.h>
26 #include <linux/backing-dev.h>
27 #include <linux/bootmem.h>
28 #include <linux/pipe_fs_i.h>
29 #include <linux/pfn.h>
31 #include <asm/uaccess.h>
35 # include <linux/efi.h>
38 static inline int range_is_allowed(unsigned long from, unsigned long to)
42 cursor = from >> PAGE_SHIFT;
43 while ((cursor << PAGE_SHIFT) < to) {
44 if (!devmem_is_allowed(cursor)) {
45 printk ("Program %s tried to read /dev/mem between %lx->%lx.\n",
46 current->comm, from, to);
55 * Architectures vary in how they handle caching for addresses
56 * outside of main memory.
59 static inline int uncached_access(struct file *file, unsigned long addr)
63 * On the PPro and successors, the MTRRs are used to set
64 * memory types for physical addresses outside main memory,
65 * so blindly setting PCD or PWT on those pages is wrong.
66 * For Pentiums and earlier, the surround logic should disable
67 * caching for the high addresses through the KEN pin, but
68 * we maintain the tradition of paranoia in this code.
70 if (file->f_flags & O_SYNC)
72 return !( test_bit(X86_FEATURE_MTRR, boot_cpu_data.x86_capability) ||
73 test_bit(X86_FEATURE_K6_MTRR, boot_cpu_data.x86_capability) ||
74 test_bit(X86_FEATURE_CYRIX_ARR, boot_cpu_data.x86_capability) ||
75 test_bit(X86_FEATURE_CENTAUR_MCR, boot_cpu_data.x86_capability) )
76 && addr >= __pa(high_memory);
77 #elif defined(__x86_64__)
79 * This is broken because it can generate memory type aliases,
80 * which can cause cache corruptions
81 * But it is only available for root and we have to be bug-to-bug
82 * compatible with i386.
84 if (file->f_flags & O_SYNC)
86 /* same behaviour as i386. PAT always set to cached and MTRRs control the
88 Hopefully a full PAT implementation will fix that soon. */
90 #elif defined(CONFIG_IA64)
92 * On ia64, we ignore O_SYNC because we cannot tolerate memory attribute aliases.
94 return !(efi_mem_attributes(addr) & EFI_MEMORY_WB);
97 * Accessing memory above the top the kernel knows about or through a file pointer
98 * that was marked O_SYNC will be done non-cached.
100 if (file->f_flags & O_SYNC)
102 return addr >= __pa(high_memory);
106 #ifndef ARCH_HAS_VALID_PHYS_ADDR_RANGE
107 static inline int valid_phys_addr_range(unsigned long addr, size_t count)
109 if (addr + count > __pa(high_memory))
115 static inline int valid_mmap_phys_addr_range(unsigned long pfn, size_t size)
121 #ifndef ARCH_HAS_DEV_MEM
123 * This funcion reads the *physical* memory. The f_pos points directly to the
126 static ssize_t read_mem(struct file * file, char __user * buf,
127 size_t count, loff_t *ppos)
129 unsigned long p = *ppos;
133 if (!valid_phys_addr_range(p, count))
136 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
137 /* we don't have page 0 mapped on sparc and m68k.. */
143 if (clear_user(buf, sz))
155 * Handle first page in case it's not aligned
157 if (-p & (PAGE_SIZE - 1))
158 sz = -p & (PAGE_SIZE - 1);
162 sz = min_t(unsigned long, sz, count);
165 * On ia64 if a page has been mapped somewhere as
166 * uncached, then it must also be accessed uncached
167 * by the kernel or data corruption may occur
169 ptr = xlate_dev_mem_ptr(p);
171 if (!range_is_allowed(p, p+count))
173 if (copy_to_user(buf, ptr, sz))
185 static ssize_t write_mem(struct file * file, const char __user * buf,
186 size_t count, loff_t *ppos)
188 unsigned long p = *ppos;
190 unsigned long copied;
193 if (!valid_phys_addr_range(p, count))
198 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
199 /* we don't have page 0 mapped on sparc and m68k.. */
201 unsigned long sz = PAGE_SIZE - p;
204 /* Hmm. Do something? */
214 * Handle first page in case it's not aligned
216 if (-p & (PAGE_SIZE - 1))
217 sz = -p & (PAGE_SIZE - 1);
221 sz = min_t(unsigned long, sz, count);
224 * On ia64 if a page has been mapped somewhere as
225 * uncached, then it must also be accessed uncached
226 * by the kernel or data corruption may occur
228 ptr = xlate_dev_mem_ptr(p);
230 if (!range_is_allowed(ptr, ptr+sz))
232 copied = copy_from_user(ptr, buf, sz);
234 written += sz - copied;
250 #ifndef __HAVE_PHYS_MEM_ACCESS_PROT
251 static pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
252 unsigned long size, pgprot_t vma_prot)
254 #ifdef pgprot_noncached
255 unsigned long offset = pfn << PAGE_SHIFT;
257 if (uncached_access(file, offset))
258 return pgprot_noncached(vma_prot);
265 static unsigned long get_unmapped_area_mem(struct file *file,
271 if (!valid_mmap_phys_addr_range(pgoff, len))
272 return (unsigned long) -EINVAL;
273 return pgoff << PAGE_SHIFT;
276 /* can't do an in-place private mapping if there's no MMU */
277 static inline int private_mapping_ok(struct vm_area_struct *vma)
279 return vma->vm_flags & VM_MAYSHARE;
282 #define get_unmapped_area_mem NULL
284 static inline int private_mapping_ok(struct vm_area_struct *vma)
290 static int mmap_mem(struct file * file, struct vm_area_struct * vma)
292 size_t size = vma->vm_end - vma->vm_start;
294 if (!valid_mmap_phys_addr_range(vma->vm_pgoff, size))
297 if (!private_mapping_ok(vma))
300 vma->vm_page_prot = phys_mem_access_prot(file, vma->vm_pgoff,
304 /* Remap-pfn-range will mark the range VM_IO and VM_RESERVED */
305 if (remap_pfn_range(vma,
314 static int mmap_kmem(struct file * file, struct vm_area_struct * vma)
318 /* Turn a kernel-virtual address into a physical page frame */
319 pfn = __pa((u64)vma->vm_pgoff << PAGE_SHIFT) >> PAGE_SHIFT;
322 * RED-PEN: on some architectures there is more mapped memory
323 * than available in mem_map which pfn_valid checks
324 * for. Perhaps should add a new macro here.
326 * RED-PEN: vmalloc is not supported right now.
332 return mmap_mem(file, vma);
335 #ifdef CONFIG_CRASH_DUMP
337 * Read memory corresponding to the old kernel.
339 static ssize_t read_oldmem(struct file *file, char __user *buf,
340 size_t count, loff_t *ppos)
342 unsigned long pfn, offset;
343 size_t read = 0, csize;
347 pfn = *ppos / PAGE_SIZE;
348 if (pfn > saved_max_pfn)
351 offset = (unsigned long)(*ppos % PAGE_SIZE);
352 if (count > PAGE_SIZE - offset)
353 csize = PAGE_SIZE - offset;
357 rc = copy_oldmem_page(pfn, buf, csize, offset, 1);
369 extern long vread(char *buf, char *addr, unsigned long count);
370 extern long vwrite(char *buf, char *addr, unsigned long count);
373 * This function reads the *virtual* memory as seen by the kernel.
375 static ssize_t read_kmem(struct file *file, char __user *buf,
376 size_t count, loff_t *ppos)
378 unsigned long p = *ppos;
379 ssize_t low_count, read, sz;
380 char * kbuf; /* k-addr because vread() takes vmlist_lock rwlock */
385 if (p < (unsigned long) high_memory) {
387 if (count > (unsigned long) high_memory - p)
388 low_count = (unsigned long) high_memory - p;
390 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
391 /* we don't have page 0 mapped on sparc and m68k.. */
392 if (p < PAGE_SIZE && low_count > 0) {
393 size_t tmp = PAGE_SIZE - p;
394 if (tmp > low_count) tmp = low_count;
395 if (clear_user(buf, tmp))
404 while (low_count > 0) {
406 * Handle first page in case it's not aligned
408 if (-p & (PAGE_SIZE - 1))
409 sz = -p & (PAGE_SIZE - 1);
413 sz = min_t(unsigned long, sz, low_count);
416 * On ia64 if a page has been mapped somewhere as
417 * uncached, then it must also be accessed uncached
418 * by the kernel or data corruption may occur
420 kbuf = xlate_dev_kmem_ptr((char *)p);
422 if (copy_to_user(buf, kbuf, sz))
433 kbuf = (char *)__get_free_page(GFP_KERNEL);
441 len = vread(kbuf, (char *)p, len);
444 if (copy_to_user(buf, kbuf, len)) {
445 free_page((unsigned long)kbuf);
453 free_page((unsigned long)kbuf);
459 #if (defined(CONFIG_ISA) || defined(CONFIG_PCI)) && !defined(__mc68000__)
460 static ssize_t read_port(struct file * file, char __user * buf,
461 size_t count, loff_t *ppos)
463 unsigned long i = *ppos;
464 char __user *tmp = buf;
466 if (!access_ok(VERIFY_WRITE, buf, count))
468 while (count-- > 0 && i < 65536) {
469 if (__put_user(inb(i),tmp) < 0)
478 static ssize_t write_port(struct file * file, const char __user * buf,
479 size_t count, loff_t *ppos)
481 unsigned long i = *ppos;
482 const char __user * tmp = buf;
484 if (!access_ok(VERIFY_READ,buf,count))
486 while (count-- > 0 && i < 65536) {
488 if (__get_user(c, tmp)) {
502 static ssize_t read_null(struct file * file, char __user * buf,
503 size_t count, loff_t *ppos)
508 static ssize_t write_null(struct file * file, const char __user * buf,
509 size_t count, loff_t *ppos)
514 static int pipe_to_null(struct pipe_inode_info *info, struct pipe_buffer *buf,
515 struct splice_desc *sd)
520 static ssize_t splice_write_null(struct pipe_inode_info *pipe,struct file *out,
521 loff_t *ppos, size_t len, unsigned int flags)
523 return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_null);
528 * For fun, we are using the MMU for this.
530 static inline size_t read_zero_pagealigned(char __user * buf, size_t size)
532 struct mm_struct *mm;
533 struct vm_area_struct * vma;
534 unsigned long addr=(unsigned long)buf;
537 /* Oops, this was forgotten before. -ben */
538 down_read(&mm->mmap_sem);
540 /* For private mappings, just map in zero pages. */
541 for (vma = find_vma(mm, addr); vma; vma = vma->vm_next) {
544 if (vma->vm_start > addr || (vma->vm_flags & VM_WRITE) == 0)
546 if (vma->vm_flags & (VM_SHARED | VM_HUGETLB))
548 count = vma->vm_end - addr;
552 zap_page_range(vma, addr, count, NULL);
553 if (zeromap_page_range(vma, addr, count, PAGE_COPY))
563 up_read(&mm->mmap_sem);
565 /* The shared case is hard. Let's do the conventional zeroing. */
567 unsigned long unwritten = clear_user(buf, PAGE_SIZE);
569 return size + unwritten - PAGE_SIZE;
577 up_read(&mm->mmap_sem);
581 static ssize_t read_zero(struct file * file, char __user * buf,
582 size_t count, loff_t *ppos)
584 unsigned long left, unwritten, written = 0;
589 if (!access_ok(VERIFY_WRITE, buf, count))
594 /* do we want to be clever? Arbitrary cut-off */
595 if (count >= PAGE_SIZE*4) {
596 unsigned long partial;
598 /* How much left of the page? */
599 partial = (PAGE_SIZE-1) & -(unsigned long) buf;
600 unwritten = clear_user(buf, partial);
601 written = partial - unwritten;
606 unwritten = read_zero_pagealigned(buf, left & PAGE_MASK);
607 written += (left & PAGE_MASK) - unwritten;
610 buf += left & PAGE_MASK;
613 unwritten = clear_user(buf, left);
614 written += left - unwritten;
616 return written ? written : -EFAULT;
619 static int mmap_zero(struct file * file, struct vm_area_struct * vma)
623 if (vma->vm_flags & VM_SHARED)
624 return shmem_zero_setup(vma);
625 err = zeromap_page_range(vma, vma->vm_start,
626 vma->vm_end - vma->vm_start, vma->vm_page_prot);
627 BUG_ON(err == -EEXIST);
630 #else /* CONFIG_MMU */
631 static ssize_t read_zero(struct file * file, char * buf,
632 size_t count, loff_t *ppos)
640 chunk = 4096; /* Just for latency reasons */
641 if (clear_user(buf, chunk))
650 static int mmap_zero(struct file * file, struct vm_area_struct * vma)
654 #endif /* CONFIG_MMU */
656 static ssize_t write_full(struct file * file, const char __user * buf,
657 size_t count, loff_t *ppos)
663 * Special lseek() function for /dev/null and /dev/zero. Most notably, you
664 * 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->f_path.dentry->d_inode->i_mutex);
688 file->f_pos = offset;
690 force_successful_syscall_return();
693 file->f_pos += offset;
695 force_successful_syscall_return();
700 mutex_unlock(&file->f_path.dentry->d_inode->i_mutex);
704 static int open_port(struct inode * inode, struct file * filp)
706 return capable(CAP_SYS_RAWIO) ? 0 : -EPERM;
709 #define zero_lseek null_lseek
710 #define full_lseek null_lseek
711 #define write_zero write_null
712 #define read_full read_zero
713 #define open_mem open_port
714 #define open_kmem open_mem
715 #define open_oldmem open_mem
717 #ifndef ARCH_HAS_DEV_MEM
718 static const struct file_operations mem_fops = {
719 .llseek = memory_lseek,
724 .get_unmapped_area = get_unmapped_area_mem,
727 extern struct file_operations mem_fops;
730 static const struct file_operations kmem_fops = {
731 .llseek = memory_lseek,
735 .get_unmapped_area = get_unmapped_area_mem,
738 static const struct file_operations null_fops = {
739 .llseek = null_lseek,
742 .splice_write = splice_write_null,
745 #if (defined(CONFIG_ISA) || defined(CONFIG_PCI)) && !defined(__mc68000__)
746 static const struct file_operations port_fops = {
747 .llseek = memory_lseek,
754 static const struct file_operations zero_fops = {
755 .llseek = zero_lseek,
762 * capabilities for /dev/zero
763 * - permits private mappings, "copies" are taken of the source of zeros
765 static struct backing_dev_info zero_bdi = {
766 .capabilities = BDI_CAP_MAP_COPY,
769 static const struct file_operations full_fops = {
770 .llseek = full_lseek,
775 #ifdef CONFIG_CRASH_DUMP
776 static const struct file_operations oldmem_fops = {
782 static ssize_t kmsg_write(struct file * file, const char __user * buf,
783 size_t count, loff_t *ppos)
788 tmp = kmalloc(count + 1, GFP_KERNEL);
792 if (!copy_from_user(tmp, buf, count)) {
794 ret = printk("%s", tmp);
796 /* printk can add a prefix */
803 static const struct file_operations kmsg_fops = {
807 static int memory_open(struct inode * inode, struct file * filp)
809 switch (iminor(inode)) {
811 filp->f_op = &mem_fops;
812 filp->f_mapping->backing_dev_info =
813 &directly_mappable_cdev_bdi;
816 filp->f_op = &kmem_fops;
817 filp->f_mapping->backing_dev_info =
818 &directly_mappable_cdev_bdi;
821 filp->f_op = &null_fops;
823 #if (defined(CONFIG_ISA) || defined(CONFIG_PCI)) && !defined(__mc68000__)
825 filp->f_op = &port_fops;
829 filp->f_mapping->backing_dev_info = &zero_bdi;
830 filp->f_op = &zero_fops;
833 filp->f_op = &full_fops;
836 filp->f_op = &random_fops;
839 filp->f_op = &urandom_fops;
842 filp->f_op = &kmsg_fops;
844 #ifdef CONFIG_CRASH_DUMP
846 filp->f_op = &oldmem_fops;
852 if (filp->f_op && filp->f_op->open)
853 return filp->f_op->open(inode,filp);
857 static const struct file_operations memory_fops = {
858 .open = memory_open, /* just a selector for the real open */
861 static const struct {
865 const struct file_operations *fops;
866 } devlist[] = { /* list of minor devices */
867 {1, "mem", S_IRUSR | S_IWUSR | S_IRGRP, &mem_fops},
868 {3, "null", S_IRUGO | S_IWUGO, &null_fops},
869 #if (defined(CONFIG_ISA) || defined(CONFIG_PCI)) && !defined(__mc68000__)
870 {4, "port", S_IRUSR | S_IWUSR | S_IRGRP, &port_fops},
872 {5, "zero", S_IRUGO | S_IWUGO, &zero_fops},
873 {7, "full", S_IRUGO | S_IWUGO, &full_fops},
874 {8, "random", S_IRUGO | S_IWUSR, &random_fops},
875 {9, "urandom", S_IRUGO | S_IWUSR, &urandom_fops},
876 {11,"kmsg", S_IRUGO | S_IWUSR, &kmsg_fops},
877 #ifdef CONFIG_CRASH_DUMP
878 {12,"oldmem", S_IRUSR | S_IWUSR | S_IRGRP, &oldmem_fops},
882 static struct class *mem_class;
884 static int __init chr_dev_init(void)
888 if (register_chrdev(MEM_MAJOR,"mem",&memory_fops))
889 printk("unable to get major %d for memory devs\n", MEM_MAJOR);
891 mem_class = class_create(THIS_MODULE, "mem");
892 for (i = 0; i < ARRAY_SIZE(devlist); i++)
893 device_create(mem_class, NULL,
894 MKDEV(MEM_MAJOR, devlist[i].minor),
900 fs_initcall(chr_dev_init);