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/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_DSYNC 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_DSYNC will be done non-cached.
61 if (file->f_flags & O_DSYNC)
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_STRICT_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 const struct vm_operations_struct mmap_mem_ops = {
304 #ifdef CONFIG_HAVE_IOREMAP_PROT
305 .access = generic_access_phys
309 static int mmap_mem(struct file * file, struct vm_area_struct * vma)
311 size_t size = vma->vm_end - vma->vm_start;
313 if (!valid_mmap_phys_addr_range(vma->vm_pgoff, size))
316 if (!private_mapping_ok(vma))
319 if (!range_is_allowed(vma->vm_pgoff, size))
322 if (!phys_mem_access_prot_allowed(file, vma->vm_pgoff, size,
326 vma->vm_page_prot = phys_mem_access_prot(file, vma->vm_pgoff,
330 vma->vm_ops = &mmap_mem_ops;
332 /* Remap-pfn-range will mark the range VM_IO and VM_RESERVED */
333 if (remap_pfn_range(vma,
337 vma->vm_page_prot)) {
343 #ifdef CONFIG_DEVKMEM
344 static int mmap_kmem(struct file * file, struct vm_area_struct * vma)
348 /* Turn a kernel-virtual address into a physical page frame */
349 pfn = __pa((u64)vma->vm_pgoff << PAGE_SHIFT) >> PAGE_SHIFT;
352 * RED-PEN: on some architectures there is more mapped memory
353 * than available in mem_map which pfn_valid checks
354 * for. Perhaps should add a new macro here.
356 * RED-PEN: vmalloc is not supported right now.
362 return mmap_mem(file, vma);
366 #ifdef CONFIG_CRASH_DUMP
368 * Read memory corresponding to the old kernel.
370 static ssize_t read_oldmem(struct file *file, char __user *buf,
371 size_t count, loff_t *ppos)
373 unsigned long pfn, offset;
374 size_t read = 0, csize;
378 pfn = *ppos / PAGE_SIZE;
379 if (pfn > saved_max_pfn)
382 offset = (unsigned long)(*ppos % PAGE_SIZE);
383 if (count > PAGE_SIZE - offset)
384 csize = PAGE_SIZE - offset;
388 rc = copy_oldmem_page(pfn, buf, csize, offset, 1);
400 #ifdef CONFIG_DEVKMEM
402 * This function reads the *virtual* memory as seen by the kernel.
404 static ssize_t read_kmem(struct file *file, char __user *buf,
405 size_t count, loff_t *ppos)
407 unsigned long p = *ppos;
408 ssize_t low_count, read, sz;
409 char * kbuf; /* k-addr because vread() takes vmlist_lock rwlock */
412 if (p < (unsigned long) high_memory) {
414 if (count > (unsigned long) high_memory - p)
415 low_count = (unsigned long) high_memory - p;
417 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
418 /* we don't have page 0 mapped on sparc and m68k.. */
419 if (p < PAGE_SIZE && low_count > 0) {
420 size_t tmp = PAGE_SIZE - p;
421 if (tmp > low_count) tmp = low_count;
422 if (clear_user(buf, tmp))
431 while (low_count > 0) {
433 * Handle first page in case it's not aligned
435 if (-p & (PAGE_SIZE - 1))
436 sz = -p & (PAGE_SIZE - 1);
440 sz = min_t(unsigned long, sz, low_count);
443 * On ia64 if a page has been mapped somewhere as
444 * uncached, then it must also be accessed uncached
445 * by the kernel or data corruption may occur
447 kbuf = xlate_dev_kmem_ptr((char *)p);
449 if (copy_to_user(buf, kbuf, sz))
460 kbuf = (char *)__get_free_page(GFP_KERNEL);
468 len = vread(kbuf, (char *)p, len);
471 if (copy_to_user(buf, kbuf, len)) {
472 free_page((unsigned long)kbuf);
480 free_page((unsigned long)kbuf);
487 static inline ssize_t
488 do_write_kmem(void *p, unsigned long realp, const char __user * buf,
489 size_t count, loff_t *ppos)
492 unsigned long copied;
495 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
496 /* we don't have page 0 mapped on sparc and m68k.. */
497 if (realp < PAGE_SIZE) {
498 unsigned long sz = PAGE_SIZE - realp;
501 /* Hmm. Do something? */
513 * Handle first page in case it's not aligned
515 if (-realp & (PAGE_SIZE - 1))
516 sz = -realp & (PAGE_SIZE - 1);
520 sz = min_t(unsigned long, sz, count);
523 * On ia64 if a page has been mapped somewhere as
524 * uncached, then it must also be accessed uncached
525 * by the kernel or data corruption may occur
527 ptr = xlate_dev_kmem_ptr(p);
529 copied = copy_from_user(ptr, buf, sz);
531 written += sz - copied;
549 * This function writes to the *virtual* memory as seen by the kernel.
551 static ssize_t write_kmem(struct file * file, const char __user * buf,
552 size_t count, loff_t *ppos)
554 unsigned long p = *ppos;
558 char * kbuf; /* k-addr because vwrite() takes vmlist_lock rwlock */
560 if (p < (unsigned long) high_memory) {
563 if (count > (unsigned long) high_memory - p)
564 wrote = (unsigned long) high_memory - p;
566 written = do_write_kmem((void*)p, p, buf, wrote, ppos);
567 if (written != wrote)
576 kbuf = (char *)__get_free_page(GFP_KERNEL);
578 return wrote ? wrote : -ENOMEM;
585 written = copy_from_user(kbuf, buf, len);
589 free_page((unsigned long)kbuf);
593 len = vwrite(kbuf, (char *)p, len);
599 free_page((unsigned long)kbuf);
603 return virtr + wrote;
607 #ifdef CONFIG_DEVPORT
608 static ssize_t read_port(struct file * file, char __user * buf,
609 size_t count, loff_t *ppos)
611 unsigned long i = *ppos;
612 char __user *tmp = buf;
614 if (!access_ok(VERIFY_WRITE, buf, count))
616 while (count-- > 0 && i < 65536) {
617 if (__put_user(inb(i),tmp) < 0)
626 static ssize_t write_port(struct file * file, const char __user * buf,
627 size_t count, loff_t *ppos)
629 unsigned long i = *ppos;
630 const char __user * tmp = buf;
632 if (!access_ok(VERIFY_READ,buf,count))
634 while (count-- > 0 && i < 65536) {
636 if (__get_user(c, tmp)) {
650 static ssize_t read_null(struct file * file, char __user * buf,
651 size_t count, loff_t *ppos)
656 static ssize_t write_null(struct file * file, const char __user * buf,
657 size_t count, loff_t *ppos)
662 static int pipe_to_null(struct pipe_inode_info *info, struct pipe_buffer *buf,
663 struct splice_desc *sd)
668 static ssize_t splice_write_null(struct pipe_inode_info *pipe,struct file *out,
669 loff_t *ppos, size_t len, unsigned int flags)
671 return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_null);
674 static ssize_t read_zero(struct file * file, char __user * buf,
675 size_t count, loff_t *ppos)
682 if (!access_ok(VERIFY_WRITE, buf, count))
687 unsigned long unwritten;
688 size_t chunk = count;
690 if (chunk > PAGE_SIZE)
691 chunk = PAGE_SIZE; /* Just for latency reasons */
692 unwritten = __clear_user(buf, chunk);
693 written += chunk - unwritten;
696 if (signal_pending(current))
697 return written ? written : -ERESTARTSYS;
702 return written ? written : -EFAULT;
705 static int mmap_zero(struct file * file, struct vm_area_struct * vma)
710 if (vma->vm_flags & VM_SHARED)
711 return shmem_zero_setup(vma);
715 static ssize_t write_full(struct file * file, const char __user * buf,
716 size_t count, loff_t *ppos)
722 * Special lseek() function for /dev/null and /dev/zero. Most notably, you
723 * can fopen() both devices with "a" now. This was previously impossible.
727 static loff_t null_lseek(struct file * file, loff_t offset, int orig)
729 return file->f_pos = 0;
733 * The memory devices use the full 32/64 bits of the offset, and so we cannot
734 * check against negative addresses: they are ok. The return value is weird,
735 * though, in that case (0).
737 * also note that seeking relative to the "end of file" isn't supported:
738 * it has no meaning, so it returns -EINVAL.
740 static loff_t memory_lseek(struct file * file, loff_t offset, int orig)
744 mutex_lock(&file->f_path.dentry->d_inode->i_mutex);
747 file->f_pos = offset;
749 force_successful_syscall_return();
752 file->f_pos += offset;
754 force_successful_syscall_return();
759 mutex_unlock(&file->f_path.dentry->d_inode->i_mutex);
763 static int open_port(struct inode * inode, struct file * filp)
765 return capable(CAP_SYS_RAWIO) ? 0 : -EPERM;
768 #define zero_lseek null_lseek
769 #define full_lseek null_lseek
770 #define write_zero write_null
771 #define read_full read_zero
772 #define open_mem open_port
773 #define open_kmem open_mem
774 #define open_oldmem open_mem
776 static const struct file_operations mem_fops = {
777 .llseek = memory_lseek,
782 .get_unmapped_area = get_unmapped_area_mem,
785 #ifdef CONFIG_DEVKMEM
786 static const struct file_operations kmem_fops = {
787 .llseek = memory_lseek,
792 .get_unmapped_area = get_unmapped_area_mem,
796 static const struct file_operations null_fops = {
797 .llseek = null_lseek,
800 .splice_write = splice_write_null,
803 #ifdef CONFIG_DEVPORT
804 static const struct file_operations port_fops = {
805 .llseek = memory_lseek,
812 static const struct file_operations zero_fops = {
813 .llseek = zero_lseek,
820 * capabilities for /dev/zero
821 * - permits private mappings, "copies" are taken of the source of zeros
823 static struct backing_dev_info zero_bdi = {
825 .capabilities = BDI_CAP_MAP_COPY,
828 static const struct file_operations full_fops = {
829 .llseek = full_lseek,
834 #ifdef CONFIG_CRASH_DUMP
835 static const struct file_operations oldmem_fops = {
841 static ssize_t kmsg_write(struct file * file, const char __user * buf,
842 size_t count, loff_t *ppos)
847 tmp = kmalloc(count + 1, GFP_KERNEL);
851 if (!copy_from_user(tmp, buf, count)) {
853 ret = printk("%s", tmp);
855 /* printk can add a prefix */
862 static const struct file_operations kmsg_fops = {
866 static const struct memdev {
869 const struct file_operations *fops;
870 struct backing_dev_info *dev_info;
872 [1] = { "mem", 0, &mem_fops, &directly_mappable_cdev_bdi },
873 #ifdef CONFIG_DEVKMEM
874 [2] = { "kmem", 0, &kmem_fops, &directly_mappable_cdev_bdi },
876 [3] = { "null", 0666, &null_fops, NULL },
877 #ifdef CONFIG_DEVPORT
878 [4] = { "port", 0, &port_fops, NULL },
880 [5] = { "zero", 0666, &zero_fops, &zero_bdi },
881 [7] = { "full", 0666, &full_fops, NULL },
882 [8] = { "random", 0666, &random_fops, NULL },
883 [9] = { "urandom", 0666, &urandom_fops, NULL },
884 [11] = { "kmsg", 0, &kmsg_fops, NULL },
885 #ifdef CONFIG_CRASH_DUMP
886 [12] = { "oldmem", 0, &oldmem_fops, NULL },
890 static int memory_open(struct inode *inode, struct file *filp)
893 const struct memdev *dev;
895 minor = iminor(inode);
896 if (minor >= ARRAY_SIZE(devlist))
899 dev = &devlist[minor];
903 filp->f_op = dev->fops;
905 filp->f_mapping->backing_dev_info = dev->dev_info;
908 return dev->fops->open(inode, filp);
913 static const struct file_operations memory_fops = {
917 static char *mem_devnode(struct device *dev, mode_t *mode)
919 if (mode && devlist[MINOR(dev->devt)].mode)
920 *mode = devlist[MINOR(dev->devt)].mode;
924 static struct class *mem_class;
926 static int __init chr_dev_init(void)
931 err = bdi_init(&zero_bdi);
935 if (register_chrdev(MEM_MAJOR,"mem",&memory_fops))
936 printk("unable to get major %d for memory devs\n", MEM_MAJOR);
938 mem_class = class_create(THIS_MODULE, "mem");
939 mem_class->devnode = mem_devnode;
940 for (minor = 1; minor < ARRAY_SIZE(devlist); minor++) {
941 if (!devlist[minor].name)
943 device_create(mem_class, NULL, MKDEV(MEM_MAJOR, minor),
944 NULL, devlist[minor].name);
950 fs_initcall(chr_dev_init);