Loading...
1/*
2 * linux/drivers/char/mem.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 *
6 * Added devfs support.
7 * Jan-11-1998, C. Scott Ananian <cananian@alumni.princeton.edu>
8 * Shared /dev/zero mmapping support, Feb 2000, Kanoj Sarcar <kanoj@sgi.com>
9 */
10
11#include <linux/mm.h>
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>
29
30#include <asm/uaccess.h>
31#include <asm/io.h>
32
33#ifdef CONFIG_IA64
34# include <linux/efi.h>
35#endif
36
37static inline unsigned long size_inside_page(unsigned long start,
38 unsigned long size)
39{
40 unsigned long sz;
41
42 sz = PAGE_SIZE - (start & (PAGE_SIZE - 1));
43
44 return min(sz, size);
45}
46
47#ifndef ARCH_HAS_VALID_PHYS_ADDR_RANGE
48static inline int valid_phys_addr_range(unsigned long addr, size_t count)
49{
50 return addr + count <= __pa(high_memory);
51}
52
53static inline int valid_mmap_phys_addr_range(unsigned long pfn, size_t size)
54{
55 return 1;
56}
57#endif
58
59#ifdef CONFIG_STRICT_DEVMEM
60static inline int range_is_allowed(unsigned long pfn, unsigned long size)
61{
62 u64 from = ((u64)pfn) << PAGE_SHIFT;
63 u64 to = from + size;
64 u64 cursor = from;
65
66 while (cursor < to) {
67 if (!devmem_is_allowed(pfn)) {
68 printk(KERN_INFO
69 "Program %s tried to access /dev/mem between %Lx->%Lx.\n",
70 current->comm, from, to);
71 return 0;
72 }
73 cursor += PAGE_SIZE;
74 pfn++;
75 }
76 return 1;
77}
78#else
79static inline int range_is_allowed(unsigned long pfn, unsigned long size)
80{
81 return 1;
82}
83#endif
84
85void __weak unxlate_dev_mem_ptr(unsigned long phys, void *addr)
86{
87}
88
89/*
90 * This funcion reads the *physical* memory. The f_pos points directly to the
91 * memory location.
92 */
93static ssize_t read_mem(struct file *file, char __user *buf,
94 size_t count, loff_t *ppos)
95{
96 unsigned long p = *ppos;
97 ssize_t read, sz;
98 char *ptr;
99
100 if (!valid_phys_addr_range(p, count))
101 return -EFAULT;
102 read = 0;
103#ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
104 /* we don't have page 0 mapped on sparc and m68k.. */
105 if (p < PAGE_SIZE) {
106 sz = size_inside_page(p, count);
107 if (sz > 0) {
108 if (clear_user(buf, sz))
109 return -EFAULT;
110 buf += sz;
111 p += sz;
112 count -= sz;
113 read += sz;
114 }
115 }
116#endif
117
118 while (count > 0) {
119 unsigned long remaining;
120
121 sz = size_inside_page(p, count);
122
123 if (!range_is_allowed(p >> PAGE_SHIFT, count))
124 return -EPERM;
125
126 /*
127 * On ia64 if a page has been mapped somewhere as uncached, then
128 * it must also be accessed uncached by the kernel or data
129 * corruption may occur.
130 */
131 ptr = xlate_dev_mem_ptr(p);
132 if (!ptr)
133 return -EFAULT;
134
135 remaining = copy_to_user(buf, ptr, sz);
136 unxlate_dev_mem_ptr(p, ptr);
137 if (remaining)
138 return -EFAULT;
139
140 buf += sz;
141 p += sz;
142 count -= sz;
143 read += sz;
144 }
145
146 *ppos += read;
147 return read;
148}
149
150static ssize_t write_mem(struct file *file, const char __user *buf,
151 size_t count, loff_t *ppos)
152{
153 unsigned long p = *ppos;
154 ssize_t written, sz;
155 unsigned long copied;
156 void *ptr;
157
158 if (!valid_phys_addr_range(p, count))
159 return -EFAULT;
160
161 written = 0;
162
163#ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
164 /* we don't have page 0 mapped on sparc and m68k.. */
165 if (p < PAGE_SIZE) {
166 sz = size_inside_page(p, count);
167 /* Hmm. Do something? */
168 buf += sz;
169 p += sz;
170 count -= sz;
171 written += sz;
172 }
173#endif
174
175 while (count > 0) {
176 sz = size_inside_page(p, count);
177
178 if (!range_is_allowed(p >> PAGE_SHIFT, sz))
179 return -EPERM;
180
181 /*
182 * On ia64 if a page has been mapped somewhere as uncached, then
183 * it must also be accessed uncached by the kernel or data
184 * corruption may occur.
185 */
186 ptr = xlate_dev_mem_ptr(p);
187 if (!ptr) {
188 if (written)
189 break;
190 return -EFAULT;
191 }
192
193 copied = copy_from_user(ptr, buf, sz);
194 unxlate_dev_mem_ptr(p, ptr);
195 if (copied) {
196 written += sz - copied;
197 if (written)
198 break;
199 return -EFAULT;
200 }
201
202 buf += sz;
203 p += sz;
204 count -= sz;
205 written += sz;
206 }
207
208 *ppos += written;
209 return written;
210}
211
212int __weak phys_mem_access_prot_allowed(struct file *file,
213 unsigned long pfn, unsigned long size, pgprot_t *vma_prot)
214{
215 return 1;
216}
217
218#ifndef __HAVE_PHYS_MEM_ACCESS_PROT
219
220/*
221 * Architectures vary in how they handle caching for addresses
222 * outside of main memory.
223 *
224 */
225#ifdef pgprot_noncached
226static int uncached_access(struct file *file, unsigned long addr)
227{
228#if defined(CONFIG_IA64)
229 /*
230 * On ia64, we ignore O_DSYNC because we cannot tolerate memory
231 * attribute aliases.
232 */
233 return !(efi_mem_attributes(addr) & EFI_MEMORY_WB);
234#elif defined(CONFIG_MIPS)
235 {
236 extern int __uncached_access(struct file *file,
237 unsigned long addr);
238
239 return __uncached_access(file, addr);
240 }
241#else
242 /*
243 * Accessing memory above the top the kernel knows about or through a
244 * file pointer
245 * that was marked O_DSYNC will be done non-cached.
246 */
247 if (file->f_flags & O_DSYNC)
248 return 1;
249 return addr >= __pa(high_memory);
250#endif
251}
252#endif
253
254static pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
255 unsigned long size, pgprot_t vma_prot)
256{
257#ifdef pgprot_noncached
258 unsigned long offset = pfn << PAGE_SHIFT;
259
260 if (uncached_access(file, offset))
261 return pgprot_noncached(vma_prot);
262#endif
263 return vma_prot;
264}
265#endif
266
267#ifndef CONFIG_MMU
268static unsigned long get_unmapped_area_mem(struct file *file,
269 unsigned long addr,
270 unsigned long len,
271 unsigned long pgoff,
272 unsigned long flags)
273{
274 if (!valid_mmap_phys_addr_range(pgoff, len))
275 return (unsigned long) -EINVAL;
276 return pgoff << PAGE_SHIFT;
277}
278
279/* can't do an in-place private mapping if there's no MMU */
280static inline int private_mapping_ok(struct vm_area_struct *vma)
281{
282 return vma->vm_flags & VM_MAYSHARE;
283}
284#else
285#define get_unmapped_area_mem NULL
286
287static inline int private_mapping_ok(struct vm_area_struct *vma)
288{
289 return 1;
290}
291#endif
292
293static const struct vm_operations_struct mmap_mem_ops = {
294#ifdef CONFIG_HAVE_IOREMAP_PROT
295 .access = generic_access_phys
296#endif
297};
298
299static int mmap_mem(struct file *file, struct vm_area_struct *vma)
300{
301 size_t size = vma->vm_end - vma->vm_start;
302
303 if (!valid_mmap_phys_addr_range(vma->vm_pgoff, size))
304 return -EINVAL;
305
306 if (!private_mapping_ok(vma))
307 return -ENOSYS;
308
309 if (!range_is_allowed(vma->vm_pgoff, size))
310 return -EPERM;
311
312 if (!phys_mem_access_prot_allowed(file, vma->vm_pgoff, size,
313 &vma->vm_page_prot))
314 return -EINVAL;
315
316 vma->vm_page_prot = phys_mem_access_prot(file, vma->vm_pgoff,
317 size,
318 vma->vm_page_prot);
319
320 vma->vm_ops = &mmap_mem_ops;
321
322 /* Remap-pfn-range will mark the range VM_IO and VM_RESERVED */
323 if (remap_pfn_range(vma,
324 vma->vm_start,
325 vma->vm_pgoff,
326 size,
327 vma->vm_page_prot)) {
328 return -EAGAIN;
329 }
330 return 0;
331}
332
333#ifdef CONFIG_DEVKMEM
334static int mmap_kmem(struct file *file, struct vm_area_struct *vma)
335{
336 unsigned long pfn;
337
338 /* Turn a kernel-virtual address into a physical page frame */
339 pfn = __pa((u64)vma->vm_pgoff << PAGE_SHIFT) >> PAGE_SHIFT;
340
341 /*
342 * RED-PEN: on some architectures there is more mapped memory than
343 * available in mem_map which pfn_valid checks for. Perhaps should add a
344 * new macro here.
345 *
346 * RED-PEN: vmalloc is not supported right now.
347 */
348 if (!pfn_valid(pfn))
349 return -EIO;
350
351 vma->vm_pgoff = pfn;
352 return mmap_mem(file, vma);
353}
354#endif
355
356#ifdef CONFIG_CRASH_DUMP
357/*
358 * Read memory corresponding to the old kernel.
359 */
360static ssize_t read_oldmem(struct file *file, char __user *buf,
361 size_t count, loff_t *ppos)
362{
363 unsigned long pfn, offset;
364 size_t read = 0, csize;
365 int rc = 0;
366
367 while (count) {
368 pfn = *ppos / PAGE_SIZE;
369 if (pfn > saved_max_pfn)
370 return read;
371
372 offset = (unsigned long)(*ppos % PAGE_SIZE);
373 if (count > PAGE_SIZE - offset)
374 csize = PAGE_SIZE - offset;
375 else
376 csize = count;
377
378 rc = copy_oldmem_page(pfn, buf, csize, offset, 1);
379 if (rc < 0)
380 return rc;
381 buf += csize;
382 *ppos += csize;
383 read += csize;
384 count -= csize;
385 }
386 return read;
387}
388#endif
389
390#ifdef CONFIG_DEVKMEM
391/*
392 * This function reads the *virtual* memory as seen by the kernel.
393 */
394static ssize_t read_kmem(struct file *file, char __user *buf,
395 size_t count, loff_t *ppos)
396{
397 unsigned long p = *ppos;
398 ssize_t low_count, read, sz;
399 char * kbuf; /* k-addr because vread() takes vmlist_lock rwlock */
400 int err = 0;
401
402 read = 0;
403 if (p < (unsigned long) high_memory) {
404 low_count = count;
405 if (count > (unsigned long)high_memory - p)
406 low_count = (unsigned long)high_memory - p;
407
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 sz = size_inside_page(p, low_count);
412 if (clear_user(buf, sz))
413 return -EFAULT;
414 buf += sz;
415 p += sz;
416 read += sz;
417 low_count -= sz;
418 count -= sz;
419 }
420#endif
421 while (low_count > 0) {
422 sz = size_inside_page(p, low_count);
423
424 /*
425 * On ia64 if a page has been mapped somewhere as
426 * uncached, then it must also be accessed uncached
427 * by the kernel or data corruption may occur
428 */
429 kbuf = xlate_dev_kmem_ptr((char *)p);
430
431 if (copy_to_user(buf, kbuf, sz))
432 return -EFAULT;
433 buf += sz;
434 p += sz;
435 read += sz;
436 low_count -= sz;
437 count -= sz;
438 }
439 }
440
441 if (count > 0) {
442 kbuf = (char *)__get_free_page(GFP_KERNEL);
443 if (!kbuf)
444 return -ENOMEM;
445 while (count > 0) {
446 sz = size_inside_page(p, count);
447 if (!is_vmalloc_or_module_addr((void *)p)) {
448 err = -ENXIO;
449 break;
450 }
451 sz = vread(kbuf, (char *)p, sz);
452 if (!sz)
453 break;
454 if (copy_to_user(buf, kbuf, sz)) {
455 err = -EFAULT;
456 break;
457 }
458 count -= sz;
459 buf += sz;
460 read += sz;
461 p += sz;
462 }
463 free_page((unsigned long)kbuf);
464 }
465 *ppos = p;
466 return read ? read : err;
467}
468
469
470static ssize_t do_write_kmem(unsigned long p, const char __user *buf,
471 size_t count, loff_t *ppos)
472{
473 ssize_t written, sz;
474 unsigned long copied;
475
476 written = 0;
477#ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
478 /* we don't have page 0 mapped on sparc and m68k.. */
479 if (p < PAGE_SIZE) {
480 sz = size_inside_page(p, count);
481 /* Hmm. Do something? */
482 buf += sz;
483 p += sz;
484 count -= sz;
485 written += sz;
486 }
487#endif
488
489 while (count > 0) {
490 char *ptr;
491
492 sz = size_inside_page(p, count);
493
494 /*
495 * On ia64 if a page has been mapped somewhere as uncached, then
496 * it must also be accessed uncached by the kernel or data
497 * corruption may occur.
498 */
499 ptr = xlate_dev_kmem_ptr((char *)p);
500
501 copied = copy_from_user(ptr, buf, sz);
502 if (copied) {
503 written += sz - copied;
504 if (written)
505 break;
506 return -EFAULT;
507 }
508 buf += sz;
509 p += sz;
510 count -= sz;
511 written += sz;
512 }
513
514 *ppos += written;
515 return written;
516}
517
518/*
519 * This function writes to the *virtual* memory as seen by the kernel.
520 */
521static ssize_t write_kmem(struct file *file, const char __user *buf,
522 size_t count, loff_t *ppos)
523{
524 unsigned long p = *ppos;
525 ssize_t wrote = 0;
526 ssize_t virtr = 0;
527 char * kbuf; /* k-addr because vwrite() takes vmlist_lock rwlock */
528 int err = 0;
529
530 if (p < (unsigned long) high_memory) {
531 unsigned long to_write = min_t(unsigned long, count,
532 (unsigned long)high_memory - p);
533 wrote = do_write_kmem(p, buf, to_write, ppos);
534 if (wrote != to_write)
535 return wrote;
536 p += wrote;
537 buf += wrote;
538 count -= wrote;
539 }
540
541 if (count > 0) {
542 kbuf = (char *)__get_free_page(GFP_KERNEL);
543 if (!kbuf)
544 return wrote ? wrote : -ENOMEM;
545 while (count > 0) {
546 unsigned long sz = size_inside_page(p, count);
547 unsigned long n;
548
549 if (!is_vmalloc_or_module_addr((void *)p)) {
550 err = -ENXIO;
551 break;
552 }
553 n = copy_from_user(kbuf, buf, sz);
554 if (n) {
555 err = -EFAULT;
556 break;
557 }
558 vwrite(kbuf, (char *)p, sz);
559 count -= sz;
560 buf += sz;
561 virtr += sz;
562 p += sz;
563 }
564 free_page((unsigned long)kbuf);
565 }
566
567 *ppos = p;
568 return virtr + wrote ? : err;
569}
570#endif
571
572#ifdef CONFIG_DEVPORT
573static ssize_t read_port(struct file *file, char __user *buf,
574 size_t count, loff_t *ppos)
575{
576 unsigned long i = *ppos;
577 char __user *tmp = buf;
578
579 if (!access_ok(VERIFY_WRITE, buf, count))
580 return -EFAULT;
581 while (count-- > 0 && i < 65536) {
582 if (__put_user(inb(i), tmp) < 0)
583 return -EFAULT;
584 i++;
585 tmp++;
586 }
587 *ppos = i;
588 return tmp-buf;
589}
590
591static ssize_t write_port(struct file *file, const char __user *buf,
592 size_t count, loff_t *ppos)
593{
594 unsigned long i = *ppos;
595 const char __user * tmp = buf;
596
597 if (!access_ok(VERIFY_READ, buf, count))
598 return -EFAULT;
599 while (count-- > 0 && i < 65536) {
600 char c;
601 if (__get_user(c, tmp)) {
602 if (tmp > buf)
603 break;
604 return -EFAULT;
605 }
606 outb(c, i);
607 i++;
608 tmp++;
609 }
610 *ppos = i;
611 return tmp-buf;
612}
613#endif
614
615static ssize_t read_null(struct file *file, char __user *buf,
616 size_t count, loff_t *ppos)
617{
618 return 0;
619}
620
621static ssize_t write_null(struct file *file, const char __user *buf,
622 size_t count, loff_t *ppos)
623{
624 return count;
625}
626
627static int pipe_to_null(struct pipe_inode_info *info, struct pipe_buffer *buf,
628 struct splice_desc *sd)
629{
630 return sd->len;
631}
632
633static ssize_t splice_write_null(struct pipe_inode_info *pipe, struct file *out,
634 loff_t *ppos, size_t len, unsigned int flags)
635{
636 return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_null);
637}
638
639static ssize_t read_zero(struct file *file, char __user *buf,
640 size_t count, loff_t *ppos)
641{
642 size_t written;
643
644 if (!count)
645 return 0;
646
647 if (!access_ok(VERIFY_WRITE, buf, count))
648 return -EFAULT;
649
650 written = 0;
651 while (count) {
652 unsigned long unwritten;
653 size_t chunk = count;
654
655 if (chunk > PAGE_SIZE)
656 chunk = PAGE_SIZE; /* Just for latency reasons */
657 unwritten = __clear_user(buf, chunk);
658 written += chunk - unwritten;
659 if (unwritten)
660 break;
661 if (signal_pending(current))
662 return written ? written : -ERESTARTSYS;
663 buf += chunk;
664 count -= chunk;
665 cond_resched();
666 }
667 return written ? written : -EFAULT;
668}
669
670static int mmap_zero(struct file *file, struct vm_area_struct *vma)
671{
672#ifndef CONFIG_MMU
673 return -ENOSYS;
674#endif
675 if (vma->vm_flags & VM_SHARED)
676 return shmem_zero_setup(vma);
677 return 0;
678}
679
680static ssize_t write_full(struct file *file, const char __user *buf,
681 size_t count, loff_t *ppos)
682{
683 return -ENOSPC;
684}
685
686/*
687 * Special lseek() function for /dev/null and /dev/zero. Most notably, you
688 * can fopen() both devices with "a" now. This was previously impossible.
689 * -- SRB.
690 */
691static loff_t null_lseek(struct file *file, loff_t offset, int orig)
692{
693 return file->f_pos = 0;
694}
695
696/*
697 * The memory devices use the full 32/64 bits of the offset, and so we cannot
698 * check against negative addresses: they are ok. The return value is weird,
699 * though, in that case (0).
700 *
701 * also note that seeking relative to the "end of file" isn't supported:
702 * it has no meaning, so it returns -EINVAL.
703 */
704static loff_t memory_lseek(struct file *file, loff_t offset, int orig)
705{
706 loff_t ret;
707
708 mutex_lock(&file->f_path.dentry->d_inode->i_mutex);
709 switch (orig) {
710 case SEEK_CUR:
711 offset += file->f_pos;
712 case SEEK_SET:
713 /* to avoid userland mistaking f_pos=-9 as -EBADF=-9 */
714 if ((unsigned long long)offset >= ~0xFFFULL) {
715 ret = -EOVERFLOW;
716 break;
717 }
718 file->f_pos = offset;
719 ret = file->f_pos;
720 force_successful_syscall_return();
721 break;
722 default:
723 ret = -EINVAL;
724 }
725 mutex_unlock(&file->f_path.dentry->d_inode->i_mutex);
726 return ret;
727}
728
729static int open_port(struct inode * inode, struct file * filp)
730{
731 return capable(CAP_SYS_RAWIO) ? 0 : -EPERM;
732}
733
734#define zero_lseek null_lseek
735#define full_lseek null_lseek
736#define write_zero write_null
737#define read_full read_zero
738#define open_mem open_port
739#define open_kmem open_mem
740#define open_oldmem open_mem
741
742static const struct file_operations mem_fops = {
743 .llseek = memory_lseek,
744 .read = read_mem,
745 .write = write_mem,
746 .mmap = mmap_mem,
747 .open = open_mem,
748 .get_unmapped_area = get_unmapped_area_mem,
749};
750
751#ifdef CONFIG_DEVKMEM
752static const struct file_operations kmem_fops = {
753 .llseek = memory_lseek,
754 .read = read_kmem,
755 .write = write_kmem,
756 .mmap = mmap_kmem,
757 .open = open_kmem,
758 .get_unmapped_area = get_unmapped_area_mem,
759};
760#endif
761
762static const struct file_operations null_fops = {
763 .llseek = null_lseek,
764 .read = read_null,
765 .write = write_null,
766 .splice_write = splice_write_null,
767};
768
769#ifdef CONFIG_DEVPORT
770static const struct file_operations port_fops = {
771 .llseek = memory_lseek,
772 .read = read_port,
773 .write = write_port,
774 .open = open_port,
775};
776#endif
777
778static const struct file_operations zero_fops = {
779 .llseek = zero_lseek,
780 .read = read_zero,
781 .write = write_zero,
782 .mmap = mmap_zero,
783};
784
785/*
786 * capabilities for /dev/zero
787 * - permits private mappings, "copies" are taken of the source of zeros
788 * - no writeback happens
789 */
790static struct backing_dev_info zero_bdi = {
791 .name = "char/mem",
792 .capabilities = BDI_CAP_MAP_COPY | BDI_CAP_NO_ACCT_AND_WRITEBACK,
793};
794
795static const struct file_operations full_fops = {
796 .llseek = full_lseek,
797 .read = read_full,
798 .write = write_full,
799};
800
801#ifdef CONFIG_CRASH_DUMP
802static const struct file_operations oldmem_fops = {
803 .read = read_oldmem,
804 .open = open_oldmem,
805 .llseek = default_llseek,
806};
807#endif
808
809static ssize_t kmsg_writev(struct kiocb *iocb, const struct iovec *iv,
810 unsigned long count, loff_t pos)
811{
812 char *line, *p;
813 int i;
814 ssize_t ret = -EFAULT;
815 size_t len = iov_length(iv, count);
816
817 line = kmalloc(len + 1, GFP_KERNEL);
818 if (line == NULL)
819 return -ENOMEM;
820
821 /*
822 * copy all vectors into a single string, to ensure we do
823 * not interleave our log line with other printk calls
824 */
825 p = line;
826 for (i = 0; i < count; i++) {
827 if (copy_from_user(p, iv[i].iov_base, iv[i].iov_len))
828 goto out;
829 p += iv[i].iov_len;
830 }
831 p[0] = '\0';
832
833 ret = printk("%s", line);
834 /* printk can add a prefix */
835 if (ret > len)
836 ret = len;
837out:
838 kfree(line);
839 return ret;
840}
841
842static const struct file_operations kmsg_fops = {
843 .aio_write = kmsg_writev,
844 .llseek = noop_llseek,
845};
846
847static const struct memdev {
848 const char *name;
849 mode_t mode;
850 const struct file_operations *fops;
851 struct backing_dev_info *dev_info;
852} devlist[] = {
853 [1] = { "mem", 0, &mem_fops, &directly_mappable_cdev_bdi },
854#ifdef CONFIG_DEVKMEM
855 [2] = { "kmem", 0, &kmem_fops, &directly_mappable_cdev_bdi },
856#endif
857 [3] = { "null", 0666, &null_fops, NULL },
858#ifdef CONFIG_DEVPORT
859 [4] = { "port", 0, &port_fops, NULL },
860#endif
861 [5] = { "zero", 0666, &zero_fops, &zero_bdi },
862 [7] = { "full", 0666, &full_fops, NULL },
863 [8] = { "random", 0666, &random_fops, NULL },
864 [9] = { "urandom", 0666, &urandom_fops, NULL },
865 [11] = { "kmsg", 0, &kmsg_fops, NULL },
866#ifdef CONFIG_CRASH_DUMP
867 [12] = { "oldmem", 0, &oldmem_fops, NULL },
868#endif
869};
870
871static int memory_open(struct inode *inode, struct file *filp)
872{
873 int minor;
874 const struct memdev *dev;
875
876 minor = iminor(inode);
877 if (minor >= ARRAY_SIZE(devlist))
878 return -ENXIO;
879
880 dev = &devlist[minor];
881 if (!dev->fops)
882 return -ENXIO;
883
884 filp->f_op = dev->fops;
885 if (dev->dev_info)
886 filp->f_mapping->backing_dev_info = dev->dev_info;
887
888 /* Is /dev/mem or /dev/kmem ? */
889 if (dev->dev_info == &directly_mappable_cdev_bdi)
890 filp->f_mode |= FMODE_UNSIGNED_OFFSET;
891
892 if (dev->fops->open)
893 return dev->fops->open(inode, filp);
894
895 return 0;
896}
897
898static const struct file_operations memory_fops = {
899 .open = memory_open,
900 .llseek = noop_llseek,
901};
902
903static char *mem_devnode(struct device *dev, mode_t *mode)
904{
905 if (mode && devlist[MINOR(dev->devt)].mode)
906 *mode = devlist[MINOR(dev->devt)].mode;
907 return NULL;
908}
909
910static struct class *mem_class;
911
912static int __init chr_dev_init(void)
913{
914 int minor;
915 int err;
916
917 err = bdi_init(&zero_bdi);
918 if (err)
919 return err;
920
921 if (register_chrdev(MEM_MAJOR, "mem", &memory_fops))
922 printk("unable to get major %d for memory devs\n", MEM_MAJOR);
923
924 mem_class = class_create(THIS_MODULE, "mem");
925 if (IS_ERR(mem_class))
926 return PTR_ERR(mem_class);
927
928 mem_class->devnode = mem_devnode;
929 for (minor = 1; minor < ARRAY_SIZE(devlist); minor++) {
930 if (!devlist[minor].name)
931 continue;
932 device_create(mem_class, NULL, MKDEV(MEM_MAJOR, minor),
933 NULL, devlist[minor].name);
934 }
935
936 return tty_init();
937}
938
939fs_initcall(chr_dev_init);
1/*
2 * linux/drivers/char/mem.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 *
6 * Added devfs support.
7 * Jan-11-1998, C. Scott Ananian <cananian@alumni.princeton.edu>
8 * Shared /dev/zero mmapping support, Feb 2000, Kanoj Sarcar <kanoj@sgi.com>
9 */
10
11#include <linux/mm.h>
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>
29#include <linux/export.h>
30
31#include <asm/uaccess.h>
32#include <asm/io.h>
33
34#ifdef CONFIG_IA64
35# include <linux/efi.h>
36#endif
37
38static inline unsigned long size_inside_page(unsigned long start,
39 unsigned long size)
40{
41 unsigned long sz;
42
43 sz = PAGE_SIZE - (start & (PAGE_SIZE - 1));
44
45 return min(sz, size);
46}
47
48#ifndef ARCH_HAS_VALID_PHYS_ADDR_RANGE
49static inline int valid_phys_addr_range(unsigned long addr, size_t count)
50{
51 return addr + count <= __pa(high_memory);
52}
53
54static inline int valid_mmap_phys_addr_range(unsigned long pfn, size_t size)
55{
56 return 1;
57}
58#endif
59
60#ifdef CONFIG_STRICT_DEVMEM
61static inline int range_is_allowed(unsigned long pfn, unsigned long size)
62{
63 u64 from = ((u64)pfn) << PAGE_SHIFT;
64 u64 to = from + size;
65 u64 cursor = from;
66
67 while (cursor < to) {
68 if (!devmem_is_allowed(pfn)) {
69 printk(KERN_INFO
70 "Program %s tried to access /dev/mem between %Lx->%Lx.\n",
71 current->comm, from, to);
72 return 0;
73 }
74 cursor += PAGE_SIZE;
75 pfn++;
76 }
77 return 1;
78}
79#else
80static inline int range_is_allowed(unsigned long pfn, unsigned long size)
81{
82 return 1;
83}
84#endif
85
86void __weak unxlate_dev_mem_ptr(unsigned long phys, void *addr)
87{
88}
89
90/*
91 * This funcion reads the *physical* memory. The f_pos points directly to the
92 * memory location.
93 */
94static ssize_t read_mem(struct file *file, char __user *buf,
95 size_t count, loff_t *ppos)
96{
97 unsigned long p = *ppos;
98 ssize_t read, sz;
99 char *ptr;
100
101 if (!valid_phys_addr_range(p, count))
102 return -EFAULT;
103 read = 0;
104#ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
105 /* we don't have page 0 mapped on sparc and m68k.. */
106 if (p < PAGE_SIZE) {
107 sz = size_inside_page(p, count);
108 if (sz > 0) {
109 if (clear_user(buf, sz))
110 return -EFAULT;
111 buf += sz;
112 p += sz;
113 count -= sz;
114 read += sz;
115 }
116 }
117#endif
118
119 while (count > 0) {
120 unsigned long remaining;
121
122 sz = size_inside_page(p, count);
123
124 if (!range_is_allowed(p >> PAGE_SHIFT, count))
125 return -EPERM;
126
127 /*
128 * On ia64 if a page has been mapped somewhere as uncached, then
129 * it must also be accessed uncached by the kernel or data
130 * corruption may occur.
131 */
132 ptr = xlate_dev_mem_ptr(p);
133 if (!ptr)
134 return -EFAULT;
135
136 remaining = copy_to_user(buf, ptr, sz);
137 unxlate_dev_mem_ptr(p, ptr);
138 if (remaining)
139 return -EFAULT;
140
141 buf += sz;
142 p += sz;
143 count -= sz;
144 read += sz;
145 }
146
147 *ppos += read;
148 return read;
149}
150
151static ssize_t write_mem(struct file *file, const char __user *buf,
152 size_t count, loff_t *ppos)
153{
154 unsigned long p = *ppos;
155 ssize_t written, sz;
156 unsigned long copied;
157 void *ptr;
158
159 if (!valid_phys_addr_range(p, count))
160 return -EFAULT;
161
162 written = 0;
163
164#ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
165 /* we don't have page 0 mapped on sparc and m68k.. */
166 if (p < PAGE_SIZE) {
167 sz = size_inside_page(p, count);
168 /* Hmm. Do something? */
169 buf += sz;
170 p += sz;
171 count -= sz;
172 written += sz;
173 }
174#endif
175
176 while (count > 0) {
177 sz = size_inside_page(p, count);
178
179 if (!range_is_allowed(p >> PAGE_SHIFT, sz))
180 return -EPERM;
181
182 /*
183 * On ia64 if a page has been mapped somewhere as uncached, then
184 * it must also be accessed uncached by the kernel or data
185 * corruption may occur.
186 */
187 ptr = xlate_dev_mem_ptr(p);
188 if (!ptr) {
189 if (written)
190 break;
191 return -EFAULT;
192 }
193
194 copied = copy_from_user(ptr, buf, sz);
195 unxlate_dev_mem_ptr(p, ptr);
196 if (copied) {
197 written += sz - copied;
198 if (written)
199 break;
200 return -EFAULT;
201 }
202
203 buf += sz;
204 p += sz;
205 count -= sz;
206 written += sz;
207 }
208
209 *ppos += written;
210 return written;
211}
212
213int __weak phys_mem_access_prot_allowed(struct file *file,
214 unsigned long pfn, unsigned long size, pgprot_t *vma_prot)
215{
216 return 1;
217}
218
219#ifndef __HAVE_PHYS_MEM_ACCESS_PROT
220
221/*
222 * Architectures vary in how they handle caching for addresses
223 * outside of main memory.
224 *
225 */
226#ifdef pgprot_noncached
227static int uncached_access(struct file *file, unsigned long addr)
228{
229#if defined(CONFIG_IA64)
230 /*
231 * On ia64, we ignore O_DSYNC because we cannot tolerate memory
232 * attribute aliases.
233 */
234 return !(efi_mem_attributes(addr) & EFI_MEMORY_WB);
235#elif defined(CONFIG_MIPS)
236 {
237 extern int __uncached_access(struct file *file,
238 unsigned long addr);
239
240 return __uncached_access(file, addr);
241 }
242#else
243 /*
244 * Accessing memory above the top the kernel knows about or through a
245 * file pointer
246 * that was marked O_DSYNC will be done non-cached.
247 */
248 if (file->f_flags & O_DSYNC)
249 return 1;
250 return addr >= __pa(high_memory);
251#endif
252}
253#endif
254
255static pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
256 unsigned long size, pgprot_t vma_prot)
257{
258#ifdef pgprot_noncached
259 unsigned long offset = pfn << PAGE_SHIFT;
260
261 if (uncached_access(file, offset))
262 return pgprot_noncached(vma_prot);
263#endif
264 return vma_prot;
265}
266#endif
267
268#ifndef CONFIG_MMU
269static unsigned long get_unmapped_area_mem(struct file *file,
270 unsigned long addr,
271 unsigned long len,
272 unsigned long pgoff,
273 unsigned long flags)
274{
275 if (!valid_mmap_phys_addr_range(pgoff, len))
276 return (unsigned long) -EINVAL;
277 return pgoff << PAGE_SHIFT;
278}
279
280/* can't do an in-place private mapping if there's no MMU */
281static inline int private_mapping_ok(struct vm_area_struct *vma)
282{
283 return vma->vm_flags & VM_MAYSHARE;
284}
285#else
286#define get_unmapped_area_mem NULL
287
288static inline int private_mapping_ok(struct vm_area_struct *vma)
289{
290 return 1;
291}
292#endif
293
294static const struct vm_operations_struct mmap_mem_ops = {
295#ifdef CONFIG_HAVE_IOREMAP_PROT
296 .access = generic_access_phys
297#endif
298};
299
300static int mmap_mem(struct file *file, struct vm_area_struct *vma)
301{
302 size_t size = vma->vm_end - vma->vm_start;
303
304 if (!valid_mmap_phys_addr_range(vma->vm_pgoff, size))
305 return -EINVAL;
306
307 if (!private_mapping_ok(vma))
308 return -ENOSYS;
309
310 if (!range_is_allowed(vma->vm_pgoff, size))
311 return -EPERM;
312
313 if (!phys_mem_access_prot_allowed(file, vma->vm_pgoff, size,
314 &vma->vm_page_prot))
315 return -EINVAL;
316
317 vma->vm_page_prot = phys_mem_access_prot(file, vma->vm_pgoff,
318 size,
319 vma->vm_page_prot);
320
321 vma->vm_ops = &mmap_mem_ops;
322
323 /* Remap-pfn-range will mark the range VM_IO and VM_RESERVED */
324 if (remap_pfn_range(vma,
325 vma->vm_start,
326 vma->vm_pgoff,
327 size,
328 vma->vm_page_prot)) {
329 return -EAGAIN;
330 }
331 return 0;
332}
333
334#ifdef CONFIG_DEVKMEM
335static int mmap_kmem(struct file *file, struct vm_area_struct *vma)
336{
337 unsigned long pfn;
338
339 /* Turn a kernel-virtual address into a physical page frame */
340 pfn = __pa((u64)vma->vm_pgoff << PAGE_SHIFT) >> PAGE_SHIFT;
341
342 /*
343 * RED-PEN: on some architectures there is more mapped memory than
344 * available in mem_map which pfn_valid checks for. Perhaps should add a
345 * new macro here.
346 *
347 * RED-PEN: vmalloc is not supported right now.
348 */
349 if (!pfn_valid(pfn))
350 return -EIO;
351
352 vma->vm_pgoff = pfn;
353 return mmap_mem(file, vma);
354}
355#endif
356
357#ifdef CONFIG_CRASH_DUMP
358/*
359 * Read memory corresponding to the old kernel.
360 */
361static ssize_t read_oldmem(struct file *file, char __user *buf,
362 size_t count, loff_t *ppos)
363{
364 unsigned long pfn, offset;
365 size_t read = 0, csize;
366 int rc = 0;
367
368 while (count) {
369 pfn = *ppos / PAGE_SIZE;
370 if (pfn > saved_max_pfn)
371 return read;
372
373 offset = (unsigned long)(*ppos % PAGE_SIZE);
374 if (count > PAGE_SIZE - offset)
375 csize = PAGE_SIZE - offset;
376 else
377 csize = count;
378
379 rc = copy_oldmem_page(pfn, buf, csize, offset, 1);
380 if (rc < 0)
381 return rc;
382 buf += csize;
383 *ppos += csize;
384 read += csize;
385 count -= csize;
386 }
387 return read;
388}
389#endif
390
391#ifdef CONFIG_DEVKMEM
392/*
393 * This function reads the *virtual* memory as seen by the kernel.
394 */
395static ssize_t read_kmem(struct file *file, char __user *buf,
396 size_t count, loff_t *ppos)
397{
398 unsigned long p = *ppos;
399 ssize_t low_count, read, sz;
400 char * kbuf; /* k-addr because vread() takes vmlist_lock rwlock */
401 int err = 0;
402
403 read = 0;
404 if (p < (unsigned long) high_memory) {
405 low_count = count;
406 if (count > (unsigned long)high_memory - p)
407 low_count = (unsigned long)high_memory - p;
408
409#ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
410 /* we don't have page 0 mapped on sparc and m68k.. */
411 if (p < PAGE_SIZE && low_count > 0) {
412 sz = size_inside_page(p, low_count);
413 if (clear_user(buf, sz))
414 return -EFAULT;
415 buf += sz;
416 p += sz;
417 read += sz;
418 low_count -= sz;
419 count -= sz;
420 }
421#endif
422 while (low_count > 0) {
423 sz = size_inside_page(p, low_count);
424
425 /*
426 * On ia64 if a page has been mapped somewhere as
427 * uncached, then it must also be accessed uncached
428 * by the kernel or data corruption may occur
429 */
430 kbuf = xlate_dev_kmem_ptr((char *)p);
431
432 if (copy_to_user(buf, kbuf, sz))
433 return -EFAULT;
434 buf += sz;
435 p += sz;
436 read += sz;
437 low_count -= sz;
438 count -= sz;
439 }
440 }
441
442 if (count > 0) {
443 kbuf = (char *)__get_free_page(GFP_KERNEL);
444 if (!kbuf)
445 return -ENOMEM;
446 while (count > 0) {
447 sz = size_inside_page(p, count);
448 if (!is_vmalloc_or_module_addr((void *)p)) {
449 err = -ENXIO;
450 break;
451 }
452 sz = vread(kbuf, (char *)p, sz);
453 if (!sz)
454 break;
455 if (copy_to_user(buf, kbuf, sz)) {
456 err = -EFAULT;
457 break;
458 }
459 count -= sz;
460 buf += sz;
461 read += sz;
462 p += sz;
463 }
464 free_page((unsigned long)kbuf);
465 }
466 *ppos = p;
467 return read ? read : err;
468}
469
470
471static ssize_t do_write_kmem(unsigned long p, const char __user *buf,
472 size_t count, loff_t *ppos)
473{
474 ssize_t written, sz;
475 unsigned long copied;
476
477 written = 0;
478#ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
479 /* we don't have page 0 mapped on sparc and m68k.. */
480 if (p < PAGE_SIZE) {
481 sz = size_inside_page(p, count);
482 /* Hmm. Do something? */
483 buf += sz;
484 p += sz;
485 count -= sz;
486 written += sz;
487 }
488#endif
489
490 while (count > 0) {
491 char *ptr;
492
493 sz = size_inside_page(p, count);
494
495 /*
496 * On ia64 if a page has been mapped somewhere as uncached, then
497 * it must also be accessed uncached by the kernel or data
498 * corruption may occur.
499 */
500 ptr = xlate_dev_kmem_ptr((char *)p);
501
502 copied = copy_from_user(ptr, buf, sz);
503 if (copied) {
504 written += sz - copied;
505 if (written)
506 break;
507 return -EFAULT;
508 }
509 buf += sz;
510 p += sz;
511 count -= sz;
512 written += sz;
513 }
514
515 *ppos += written;
516 return written;
517}
518
519/*
520 * This function writes to the *virtual* memory as seen by the kernel.
521 */
522static ssize_t write_kmem(struct file *file, const char __user *buf,
523 size_t count, loff_t *ppos)
524{
525 unsigned long p = *ppos;
526 ssize_t wrote = 0;
527 ssize_t virtr = 0;
528 char * kbuf; /* k-addr because vwrite() takes vmlist_lock rwlock */
529 int err = 0;
530
531 if (p < (unsigned long) high_memory) {
532 unsigned long to_write = min_t(unsigned long, count,
533 (unsigned long)high_memory - p);
534 wrote = do_write_kmem(p, buf, to_write, ppos);
535 if (wrote != to_write)
536 return wrote;
537 p += wrote;
538 buf += wrote;
539 count -= wrote;
540 }
541
542 if (count > 0) {
543 kbuf = (char *)__get_free_page(GFP_KERNEL);
544 if (!kbuf)
545 return wrote ? wrote : -ENOMEM;
546 while (count > 0) {
547 unsigned long sz = size_inside_page(p, count);
548 unsigned long n;
549
550 if (!is_vmalloc_or_module_addr((void *)p)) {
551 err = -ENXIO;
552 break;
553 }
554 n = copy_from_user(kbuf, buf, sz);
555 if (n) {
556 err = -EFAULT;
557 break;
558 }
559 vwrite(kbuf, (char *)p, sz);
560 count -= sz;
561 buf += sz;
562 virtr += sz;
563 p += sz;
564 }
565 free_page((unsigned long)kbuf);
566 }
567
568 *ppos = p;
569 return virtr + wrote ? : err;
570}
571#endif
572
573#ifdef CONFIG_DEVPORT
574static ssize_t read_port(struct file *file, char __user *buf,
575 size_t count, loff_t *ppos)
576{
577 unsigned long i = *ppos;
578 char __user *tmp = buf;
579
580 if (!access_ok(VERIFY_WRITE, buf, count))
581 return -EFAULT;
582 while (count-- > 0 && i < 65536) {
583 if (__put_user(inb(i), tmp) < 0)
584 return -EFAULT;
585 i++;
586 tmp++;
587 }
588 *ppos = i;
589 return tmp-buf;
590}
591
592static ssize_t write_port(struct file *file, const char __user *buf,
593 size_t count, loff_t *ppos)
594{
595 unsigned long i = *ppos;
596 const char __user * tmp = buf;
597
598 if (!access_ok(VERIFY_READ, buf, count))
599 return -EFAULT;
600 while (count-- > 0 && i < 65536) {
601 char c;
602 if (__get_user(c, tmp)) {
603 if (tmp > buf)
604 break;
605 return -EFAULT;
606 }
607 outb(c, i);
608 i++;
609 tmp++;
610 }
611 *ppos = i;
612 return tmp-buf;
613}
614#endif
615
616static ssize_t read_null(struct file *file, char __user *buf,
617 size_t count, loff_t *ppos)
618{
619 return 0;
620}
621
622static ssize_t write_null(struct file *file, const char __user *buf,
623 size_t count, loff_t *ppos)
624{
625 return count;
626}
627
628static int pipe_to_null(struct pipe_inode_info *info, struct pipe_buffer *buf,
629 struct splice_desc *sd)
630{
631 return sd->len;
632}
633
634static ssize_t splice_write_null(struct pipe_inode_info *pipe, struct file *out,
635 loff_t *ppos, size_t len, unsigned int flags)
636{
637 return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_null);
638}
639
640static ssize_t read_zero(struct file *file, char __user *buf,
641 size_t count, loff_t *ppos)
642{
643 size_t written;
644
645 if (!count)
646 return 0;
647
648 if (!access_ok(VERIFY_WRITE, buf, count))
649 return -EFAULT;
650
651 written = 0;
652 while (count) {
653 unsigned long unwritten;
654 size_t chunk = count;
655
656 if (chunk > PAGE_SIZE)
657 chunk = PAGE_SIZE; /* Just for latency reasons */
658 unwritten = __clear_user(buf, chunk);
659 written += chunk - unwritten;
660 if (unwritten)
661 break;
662 if (signal_pending(current))
663 return written ? written : -ERESTARTSYS;
664 buf += chunk;
665 count -= chunk;
666 cond_resched();
667 }
668 return written ? written : -EFAULT;
669}
670
671static int mmap_zero(struct file *file, struct vm_area_struct *vma)
672{
673#ifndef CONFIG_MMU
674 return -ENOSYS;
675#endif
676 if (vma->vm_flags & VM_SHARED)
677 return shmem_zero_setup(vma);
678 return 0;
679}
680
681static ssize_t write_full(struct file *file, const char __user *buf,
682 size_t count, loff_t *ppos)
683{
684 return -ENOSPC;
685}
686
687/*
688 * Special lseek() function for /dev/null and /dev/zero. Most notably, you
689 * can fopen() both devices with "a" now. This was previously impossible.
690 * -- SRB.
691 */
692static loff_t null_lseek(struct file *file, loff_t offset, int orig)
693{
694 return file->f_pos = 0;
695}
696
697/*
698 * The memory devices use the full 32/64 bits of the offset, and so we cannot
699 * check against negative addresses: they are ok. The return value is weird,
700 * though, in that case (0).
701 *
702 * also note that seeking relative to the "end of file" isn't supported:
703 * it has no meaning, so it returns -EINVAL.
704 */
705static loff_t memory_lseek(struct file *file, loff_t offset, int orig)
706{
707 loff_t ret;
708
709 mutex_lock(&file->f_path.dentry->d_inode->i_mutex);
710 switch (orig) {
711 case SEEK_CUR:
712 offset += file->f_pos;
713 case SEEK_SET:
714 /* to avoid userland mistaking f_pos=-9 as -EBADF=-9 */
715 if ((unsigned long long)offset >= ~0xFFFULL) {
716 ret = -EOVERFLOW;
717 break;
718 }
719 file->f_pos = offset;
720 ret = file->f_pos;
721 force_successful_syscall_return();
722 break;
723 default:
724 ret = -EINVAL;
725 }
726 mutex_unlock(&file->f_path.dentry->d_inode->i_mutex);
727 return ret;
728}
729
730static int open_port(struct inode * inode, struct file * filp)
731{
732 return capable(CAP_SYS_RAWIO) ? 0 : -EPERM;
733}
734
735#define zero_lseek null_lseek
736#define full_lseek null_lseek
737#define write_zero write_null
738#define read_full read_zero
739#define open_mem open_port
740#define open_kmem open_mem
741#define open_oldmem open_mem
742
743static const struct file_operations mem_fops = {
744 .llseek = memory_lseek,
745 .read = read_mem,
746 .write = write_mem,
747 .mmap = mmap_mem,
748 .open = open_mem,
749 .get_unmapped_area = get_unmapped_area_mem,
750};
751
752#ifdef CONFIG_DEVKMEM
753static const struct file_operations kmem_fops = {
754 .llseek = memory_lseek,
755 .read = read_kmem,
756 .write = write_kmem,
757 .mmap = mmap_kmem,
758 .open = open_kmem,
759 .get_unmapped_area = get_unmapped_area_mem,
760};
761#endif
762
763static const struct file_operations null_fops = {
764 .llseek = null_lseek,
765 .read = read_null,
766 .write = write_null,
767 .splice_write = splice_write_null,
768};
769
770#ifdef CONFIG_DEVPORT
771static const struct file_operations port_fops = {
772 .llseek = memory_lseek,
773 .read = read_port,
774 .write = write_port,
775 .open = open_port,
776};
777#endif
778
779static const struct file_operations zero_fops = {
780 .llseek = zero_lseek,
781 .read = read_zero,
782 .write = write_zero,
783 .mmap = mmap_zero,
784};
785
786/*
787 * capabilities for /dev/zero
788 * - permits private mappings, "copies" are taken of the source of zeros
789 * - no writeback happens
790 */
791static struct backing_dev_info zero_bdi = {
792 .name = "char/mem",
793 .capabilities = BDI_CAP_MAP_COPY | BDI_CAP_NO_ACCT_AND_WRITEBACK,
794};
795
796static const struct file_operations full_fops = {
797 .llseek = full_lseek,
798 .read = read_full,
799 .write = write_full,
800};
801
802#ifdef CONFIG_CRASH_DUMP
803static const struct file_operations oldmem_fops = {
804 .read = read_oldmem,
805 .open = open_oldmem,
806 .llseek = default_llseek,
807};
808#endif
809
810static const struct memdev {
811 const char *name;
812 umode_t mode;
813 const struct file_operations *fops;
814 struct backing_dev_info *dev_info;
815} devlist[] = {
816 [1] = { "mem", 0, &mem_fops, &directly_mappable_cdev_bdi },
817#ifdef CONFIG_DEVKMEM
818 [2] = { "kmem", 0, &kmem_fops, &directly_mappable_cdev_bdi },
819#endif
820 [3] = { "null", 0666, &null_fops, NULL },
821#ifdef CONFIG_DEVPORT
822 [4] = { "port", 0, &port_fops, NULL },
823#endif
824 [5] = { "zero", 0666, &zero_fops, &zero_bdi },
825 [7] = { "full", 0666, &full_fops, NULL },
826 [8] = { "random", 0666, &random_fops, NULL },
827 [9] = { "urandom", 0666, &urandom_fops, NULL },
828#ifdef CONFIG_PRINTK
829 [11] = { "kmsg", 0644, &kmsg_fops, NULL },
830#endif
831#ifdef CONFIG_CRASH_DUMP
832 [12] = { "oldmem", 0, &oldmem_fops, NULL },
833#endif
834};
835
836static int memory_open(struct inode *inode, struct file *filp)
837{
838 int minor;
839 const struct memdev *dev;
840
841 minor = iminor(inode);
842 if (minor >= ARRAY_SIZE(devlist))
843 return -ENXIO;
844
845 dev = &devlist[minor];
846 if (!dev->fops)
847 return -ENXIO;
848
849 filp->f_op = dev->fops;
850 if (dev->dev_info)
851 filp->f_mapping->backing_dev_info = dev->dev_info;
852
853 /* Is /dev/mem or /dev/kmem ? */
854 if (dev->dev_info == &directly_mappable_cdev_bdi)
855 filp->f_mode |= FMODE_UNSIGNED_OFFSET;
856
857 if (dev->fops->open)
858 return dev->fops->open(inode, filp);
859
860 return 0;
861}
862
863static const struct file_operations memory_fops = {
864 .open = memory_open,
865 .llseek = noop_llseek,
866};
867
868static char *mem_devnode(struct device *dev, umode_t *mode)
869{
870 if (mode && devlist[MINOR(dev->devt)].mode)
871 *mode = devlist[MINOR(dev->devt)].mode;
872 return NULL;
873}
874
875static struct class *mem_class;
876
877static int __init chr_dev_init(void)
878{
879 int minor;
880 int err;
881
882 err = bdi_init(&zero_bdi);
883 if (err)
884 return err;
885
886 if (register_chrdev(MEM_MAJOR, "mem", &memory_fops))
887 printk("unable to get major %d for memory devs\n", MEM_MAJOR);
888
889 mem_class = class_create(THIS_MODULE, "mem");
890 if (IS_ERR(mem_class))
891 return PTR_ERR(mem_class);
892
893 mem_class->devnode = mem_devnode;
894 for (minor = 1; minor < ARRAY_SIZE(devlist); minor++) {
895 if (!devlist[minor].name)
896 continue;
897 device_create(mem_class, NULL, MKDEV(MEM_MAJOR, minor),
898 NULL, devlist[minor].name);
899 }
900
901 return tty_init();
902}
903
904fs_initcall(chr_dev_init);