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/ptrace.h>
22 #include <linux/device.h>
23 #include <linux/highmem.h>
24 #include <linux/crash_dump.h>
25 #include <linux/backing-dev.h>
26 #include <linux/bootmem.h>
27 #include <linux/splice.h>
28 #include <linux/pfn.h>
30 #include <asm/uaccess.h>
34 # include <linux/efi.h>
38 * Architectures vary in how they handle caching for addresses
39 * outside of main memory.
42 static inline int uncached_access(struct file *file, unsigned long addr)
44 #if defined(CONFIG_IA64)
46 * On ia64, we ignore O_SYNC because we cannot tolerate memory attribute aliases.
48 return !(efi_mem_attributes(addr) & EFI_MEMORY_WB);
49 #elif defined(CONFIG_MIPS)
51 extern int __uncached_access(struct file *file,
54 return __uncached_access(file, addr);
58 * Accessing memory above the top the kernel knows about or through a file pointer
59 * that was marked O_SYNC will be done non-cached.
61 if (file->f_flags & O_SYNC)
63 return addr >= __pa(high_memory);
67 #ifndef ARCH_HAS_VALID_PHYS_ADDR_RANGE
68 static inline int valid_phys_addr_range(unsigned long addr, size_t count)
70 if (addr + count > __pa(high_memory))
76 static inline int valid_mmap_phys_addr_range(unsigned long pfn, size_t size)
82 #ifdef CONFIG_NONPROMISC_DEVMEM
83 static inline int range_is_allowed(unsigned long pfn, unsigned long size)
85 u64 from = ((u64)pfn) << PAGE_SHIFT;
90 if (!devmem_is_allowed(pfn)) {
92 "Program %s tried to access /dev/mem between %Lx->%Lx.\n",
93 current->comm, from, to);
102 static inline int range_is_allowed(unsigned long pfn, unsigned long size)
108 void __attribute__((weak)) unxlate_dev_mem_ptr(unsigned long phys, void *addr)
113 * This funcion reads the *physical* memory. The f_pos points directly to the
116 static ssize_t read_mem(struct file * file, char __user * buf,
117 size_t count, loff_t *ppos)
119 unsigned long 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.. */
133 if (clear_user(buf, sz))
145 * Handle first page in case it's not aligned
147 if (-p & (PAGE_SIZE - 1))
148 sz = -p & (PAGE_SIZE - 1);
152 sz = min_t(unsigned long, sz, count);
154 if (!range_is_allowed(p >> PAGE_SHIFT, count))
158 * On ia64 if a page has been mapped somewhere as
159 * uncached, then it must also be accessed uncached
160 * by the kernel or data corruption may occur
162 ptr = xlate_dev_mem_ptr(p);
166 if (copy_to_user(buf, ptr, sz)) {
167 unxlate_dev_mem_ptr(p, ptr);
171 unxlate_dev_mem_ptr(p, ptr);
183 static ssize_t write_mem(struct file * file, const char __user * buf,
184 size_t count, loff_t *ppos)
186 unsigned long p = *ppos;
188 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 unsigned long sz = PAGE_SIZE - p;
202 /* Hmm. Do something? */
212 * Handle first page in case it's not aligned
214 if (-p & (PAGE_SIZE - 1))
215 sz = -p & (PAGE_SIZE - 1);
219 sz = min_t(unsigned long, sz, count);
221 if (!range_is_allowed(p >> PAGE_SHIFT, sz))
225 * On ia64 if a page has been mapped somewhere as
226 * uncached, then it must also be accessed uncached
227 * by the kernel or data corruption may occur
229 ptr = xlate_dev_mem_ptr(p);
236 copied = copy_from_user(ptr, buf, sz);
238 written += sz - copied;
239 unxlate_dev_mem_ptr(p, ptr);
245 unxlate_dev_mem_ptr(p, ptr);
257 int __attribute__((weak)) phys_mem_access_prot_allowed(struct file *file,
258 unsigned long pfn, unsigned long size, pgprot_t *vma_prot)
263 #ifndef __HAVE_PHYS_MEM_ACCESS_PROT
264 static pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
265 unsigned long size, pgprot_t vma_prot)
267 #ifdef pgprot_noncached
268 unsigned long offset = pfn << PAGE_SHIFT;
270 if (uncached_access(file, offset))
271 return pgprot_noncached(vma_prot);
278 static unsigned long get_unmapped_area_mem(struct file *file,
284 if (!valid_mmap_phys_addr_range(pgoff, len))
285 return (unsigned long) -EINVAL;
286 return pgoff << PAGE_SHIFT;
289 /* can't do an in-place private mapping if there's no MMU */
290 static inline int private_mapping_ok(struct vm_area_struct *vma)
292 return vma->vm_flags & VM_MAYSHARE;
295 #define get_unmapped_area_mem NULL
297 static inline int private_mapping_ok(struct vm_area_struct *vma)
303 static int mmap_mem(struct file * file, struct vm_area_struct * vma)
305 size_t size = vma->vm_end - vma->vm_start;
307 if (!valid_mmap_phys_addr_range(vma->vm_pgoff, size))
310 if (!private_mapping_ok(vma))
313 if (!range_is_allowed(vma->vm_pgoff, size))
316 if (!phys_mem_access_prot_allowed(file, vma->vm_pgoff, size,
320 vma->vm_page_prot = phys_mem_access_prot(file, vma->vm_pgoff,
324 /* Remap-pfn-range will mark the range VM_IO and VM_RESERVED */
325 if (remap_pfn_range(vma,
334 static int mmap_kmem(struct file * file, struct vm_area_struct * vma)
338 /* Turn a kernel-virtual address into a physical page frame */
339 pfn = __pa((u64)vma->vm_pgoff << PAGE_SHIFT) >> PAGE_SHIFT;
342 * RED-PEN: on some architectures there is more mapped memory
343 * than available in mem_map which pfn_valid checks
344 * for. Perhaps should add a new macro here.
346 * RED-PEN: vmalloc is not supported right now.
352 return mmap_mem(file, vma);
355 #ifdef CONFIG_CRASH_DUMP
357 * Read memory corresponding to the old kernel.
359 static ssize_t read_oldmem(struct file *file, char __user *buf,
360 size_t count, loff_t *ppos)
362 unsigned long pfn, offset;
363 size_t read = 0, csize;
367 pfn = *ppos / PAGE_SIZE;
368 if (pfn > saved_max_pfn)
371 offset = (unsigned long)(*ppos % PAGE_SIZE);
372 if (count > PAGE_SIZE - offset)
373 csize = PAGE_SIZE - offset;
377 rc = copy_oldmem_page(pfn, buf, csize, offset, 1);
389 extern long vread(char *buf, char *addr, unsigned long count);
390 extern long vwrite(char *buf, char *addr, unsigned long count);
393 * This function reads the *virtual* memory as seen by the kernel.
395 static ssize_t read_kmem(struct file *file, char __user *buf,
396 size_t count, loff_t *ppos)
398 unsigned long p = *ppos;
399 ssize_t low_count, read, sz;
400 char * kbuf; /* k-addr because vread() takes vmlist_lock rwlock */
403 if (p < (unsigned long) high_memory) {
405 if (count > (unsigned long) high_memory - p)
406 low_count = (unsigned long) high_memory - p;
408 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
409 /* we don't have page 0 mapped on sparc and m68k.. */
410 if (p < PAGE_SIZE && low_count > 0) {
411 size_t tmp = PAGE_SIZE - p;
412 if (tmp > low_count) tmp = low_count;
413 if (clear_user(buf, tmp))
422 while (low_count > 0) {
424 * Handle first page in case it's not aligned
426 if (-p & (PAGE_SIZE - 1))
427 sz = -p & (PAGE_SIZE - 1);
431 sz = min_t(unsigned long, sz, low_count);
434 * On ia64 if a page has been mapped somewhere as
435 * uncached, then it must also be accessed uncached
436 * by the kernel or data corruption may occur
438 kbuf = xlate_dev_kmem_ptr((char *)p);
440 if (copy_to_user(buf, kbuf, sz))
451 kbuf = (char *)__get_free_page(GFP_KERNEL);
459 len = vread(kbuf, (char *)p, len);
462 if (copy_to_user(buf, kbuf, len)) {
463 free_page((unsigned long)kbuf);
471 free_page((unsigned long)kbuf);
478 static inline ssize_t
479 do_write_kmem(void *p, unsigned long realp, const char __user * buf,
480 size_t count, loff_t *ppos)
483 unsigned long copied;
486 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
487 /* we don't have page 0 mapped on sparc and m68k.. */
488 if (realp < PAGE_SIZE) {
489 unsigned long sz = PAGE_SIZE - realp;
492 /* Hmm. Do something? */
504 * Handle first page in case it's not aligned
506 if (-realp & (PAGE_SIZE - 1))
507 sz = -realp & (PAGE_SIZE - 1);
511 sz = min_t(unsigned long, sz, count);
514 * On ia64 if a page has been mapped somewhere as
515 * uncached, then it must also be accessed uncached
516 * by the kernel or data corruption may occur
518 ptr = xlate_dev_kmem_ptr(p);
520 copied = copy_from_user(ptr, buf, sz);
522 written += sz - copied;
540 * This function writes to the *virtual* memory as seen by the kernel.
542 static ssize_t write_kmem(struct file * file, const char __user * buf,
543 size_t count, loff_t *ppos)
545 unsigned long p = *ppos;
549 char * kbuf; /* k-addr because vwrite() takes vmlist_lock rwlock */
551 if (p < (unsigned long) high_memory) {
554 if (count > (unsigned long) high_memory - p)
555 wrote = (unsigned long) high_memory - p;
557 written = do_write_kmem((void*)p, p, buf, wrote, ppos);
558 if (written != wrote)
567 kbuf = (char *)__get_free_page(GFP_KERNEL);
569 return wrote ? wrote : -ENOMEM;
576 written = copy_from_user(kbuf, buf, len);
580 free_page((unsigned long)kbuf);
584 len = vwrite(kbuf, (char *)p, len);
590 free_page((unsigned long)kbuf);
594 return virtr + wrote;
597 #ifdef CONFIG_DEVPORT
598 static ssize_t read_port(struct file * file, char __user * buf,
599 size_t count, loff_t *ppos)
601 unsigned long i = *ppos;
602 char __user *tmp = buf;
604 if (!access_ok(VERIFY_WRITE, buf, count))
606 while (count-- > 0 && i < 65536) {
607 if (__put_user(inb(i),tmp) < 0)
616 static ssize_t write_port(struct file * file, const char __user * buf,
617 size_t count, loff_t *ppos)
619 unsigned long i = *ppos;
620 const char __user * tmp = buf;
622 if (!access_ok(VERIFY_READ,buf,count))
624 while (count-- > 0 && i < 65536) {
626 if (__get_user(c, tmp)) {
640 static ssize_t read_null(struct file * file, char __user * buf,
641 size_t count, loff_t *ppos)
646 static ssize_t write_null(struct file * file, const char __user * buf,
647 size_t count, loff_t *ppos)
652 static int pipe_to_null(struct pipe_inode_info *info, struct pipe_buffer *buf,
653 struct splice_desc *sd)
658 static ssize_t splice_write_null(struct pipe_inode_info *pipe,struct file *out,
659 loff_t *ppos, size_t len, unsigned int flags)
661 return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_null);
664 static ssize_t read_zero(struct file * file, char __user * buf,
665 size_t count, loff_t *ppos)
672 if (!access_ok(VERIFY_WRITE, buf, count))
677 unsigned long unwritten;
678 size_t chunk = count;
680 if (chunk > PAGE_SIZE)
681 chunk = PAGE_SIZE; /* Just for latency reasons */
682 unwritten = clear_user(buf, chunk);
683 written += chunk - unwritten;
690 return written ? written : -EFAULT;
693 static int mmap_zero(struct file * file, struct vm_area_struct * vma)
698 if (vma->vm_flags & VM_SHARED)
699 return shmem_zero_setup(vma);
703 static ssize_t write_full(struct file * file, const char __user * buf,
704 size_t count, loff_t *ppos)
710 * Special lseek() function for /dev/null and /dev/zero. Most notably, you
711 * can fopen() both devices with "a" now. This was previously impossible.
715 static loff_t null_lseek(struct file * file, loff_t offset, int orig)
717 return file->f_pos = 0;
721 * The memory devices use the full 32/64 bits of the offset, and so we cannot
722 * check against negative addresses: they are ok. The return value is weird,
723 * though, in that case (0).
725 * also note that seeking relative to the "end of file" isn't supported:
726 * it has no meaning, so it returns -EINVAL.
728 static loff_t memory_lseek(struct file * file, loff_t offset, int orig)
732 mutex_lock(&file->f_path.dentry->d_inode->i_mutex);
735 file->f_pos = offset;
737 force_successful_syscall_return();
740 file->f_pos += offset;
742 force_successful_syscall_return();
747 mutex_unlock(&file->f_path.dentry->d_inode->i_mutex);
751 static int open_port(struct inode * inode, struct file * filp)
753 return capable(CAP_SYS_RAWIO) ? 0 : -EPERM;
756 #define zero_lseek null_lseek
757 #define full_lseek null_lseek
758 #define write_zero write_null
759 #define read_full read_zero
760 #define open_mem open_port
761 #define open_kmem open_mem
762 #define open_oldmem open_mem
764 static const struct file_operations mem_fops = {
765 .llseek = memory_lseek,
770 .get_unmapped_area = get_unmapped_area_mem,
773 static const struct file_operations kmem_fops = {
774 .llseek = memory_lseek,
779 .get_unmapped_area = get_unmapped_area_mem,
782 static const struct file_operations null_fops = {
783 .llseek = null_lseek,
786 .splice_write = splice_write_null,
789 #ifdef CONFIG_DEVPORT
790 static const struct file_operations port_fops = {
791 .llseek = memory_lseek,
798 static const struct file_operations zero_fops = {
799 .llseek = zero_lseek,
806 * capabilities for /dev/zero
807 * - permits private mappings, "copies" are taken of the source of zeros
809 static struct backing_dev_info zero_bdi = {
810 .capabilities = BDI_CAP_MAP_COPY,
813 static const struct file_operations full_fops = {
814 .llseek = full_lseek,
819 #ifdef CONFIG_CRASH_DUMP
820 static const struct file_operations oldmem_fops = {
826 static ssize_t kmsg_write(struct file * file, const char __user * buf,
827 size_t count, loff_t *ppos)
832 tmp = kmalloc(count + 1, GFP_KERNEL);
836 if (!copy_from_user(tmp, buf, count)) {
838 ret = printk("%s", tmp);
840 /* printk can add a prefix */
847 static const struct file_operations kmsg_fops = {
851 static int memory_open(struct inode * inode, struct file * filp)
853 switch (iminor(inode)) {
855 filp->f_op = &mem_fops;
856 filp->f_mapping->backing_dev_info =
857 &directly_mappable_cdev_bdi;
860 filp->f_op = &kmem_fops;
861 filp->f_mapping->backing_dev_info =
862 &directly_mappable_cdev_bdi;
865 filp->f_op = &null_fops;
867 #ifdef CONFIG_DEVPORT
869 filp->f_op = &port_fops;
873 filp->f_mapping->backing_dev_info = &zero_bdi;
874 filp->f_op = &zero_fops;
877 filp->f_op = &full_fops;
880 filp->f_op = &random_fops;
883 filp->f_op = &urandom_fops;
886 filp->f_op = &kmsg_fops;
888 #ifdef CONFIG_CRASH_DUMP
890 filp->f_op = &oldmem_fops;
896 if (filp->f_op && filp->f_op->open)
897 return filp->f_op->open(inode,filp);
901 static const struct file_operations memory_fops = {
902 .open = memory_open, /* just a selector for the real open */
905 static const struct {
909 const struct file_operations *fops;
910 } devlist[] = { /* list of minor devices */
911 {1, "mem", S_IRUSR | S_IWUSR | S_IRGRP, &mem_fops},
912 {2, "kmem", S_IRUSR | S_IWUSR | S_IRGRP, &kmem_fops},
913 {3, "null", S_IRUGO | S_IWUGO, &null_fops},
914 #ifdef CONFIG_DEVPORT
915 {4, "port", S_IRUSR | S_IWUSR | S_IRGRP, &port_fops},
917 {5, "zero", S_IRUGO | S_IWUGO, &zero_fops},
918 {7, "full", S_IRUGO | S_IWUGO, &full_fops},
919 {8, "random", S_IRUGO | S_IWUSR, &random_fops},
920 {9, "urandom", S_IRUGO | S_IWUSR, &urandom_fops},
921 {11,"kmsg", S_IRUGO | S_IWUSR, &kmsg_fops},
922 #ifdef CONFIG_CRASH_DUMP
923 {12,"oldmem", S_IRUSR | S_IWUSR | S_IRGRP, &oldmem_fops},
927 static struct class *mem_class;
929 static int __init chr_dev_init(void)
934 err = bdi_init(&zero_bdi);
938 if (register_chrdev(MEM_MAJOR,"mem",&memory_fops))
939 printk("unable to get major %d for memory devs\n", MEM_MAJOR);
941 mem_class = class_create(THIS_MODULE, "mem");
942 for (i = 0; i < ARRAY_SIZE(devlist); i++)
943 device_create(mem_class, NULL,
944 MKDEV(MEM_MAJOR, devlist[i].minor),
950 fs_initcall(chr_dev_init);