Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * linux/drivers/char/mem.c
4 *
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 *
7 * Added devfs support.
8 * Jan-11-1998, C. Scott Ananian <cananian@alumni.princeton.edu>
9 * Shared /dev/zero mmapping support, Feb 2000, Kanoj Sarcar <kanoj@sgi.com>
10 */
11
12#include <linux/mm.h>
13#include <linux/miscdevice.h>
14#include <linux/slab.h>
15#include <linux/vmalloc.h>
16#include <linux/mman.h>
17#include <linux/random.h>
18#include <linux/init.h>
19#include <linux/raw.h>
20#include <linux/tty.h>
21#include <linux/capability.h>
22#include <linux/ptrace.h>
23#include <linux/device.h>
24#include <linux/highmem.h>
25#include <linux/backing-dev.h>
26#include <linux/shmem_fs.h>
27#include <linux/splice.h>
28#include <linux/pfn.h>
29#include <linux/export.h>
30#include <linux/io.h>
31#include <linux/uio.h>
32#include <linux/uaccess.h>
33#include <linux/security.h>
34#include <linux/pseudo_fs.h>
35#include <uapi/linux/magic.h>
36#include <linux/mount.h>
37
38#ifdef CONFIG_IA64
39# include <linux/efi.h>
40#endif
41
42#define DEVMEM_MINOR 1
43#define DEVPORT_MINOR 4
44
45static inline unsigned long size_inside_page(unsigned long start,
46 unsigned long size)
47{
48 unsigned long sz;
49
50 sz = PAGE_SIZE - (start & (PAGE_SIZE - 1));
51
52 return min(sz, size);
53}
54
55#ifndef ARCH_HAS_VALID_PHYS_ADDR_RANGE
56static inline int valid_phys_addr_range(phys_addr_t addr, size_t count)
57{
58 return addr + count <= __pa(high_memory);
59}
60
61static inline int valid_mmap_phys_addr_range(unsigned long pfn, size_t size)
62{
63 return 1;
64}
65#endif
66
67#ifdef CONFIG_STRICT_DEVMEM
68static inline int page_is_allowed(unsigned long pfn)
69{
70 return devmem_is_allowed(pfn);
71}
72static inline int range_is_allowed(unsigned long pfn, unsigned long size)
73{
74 u64 from = ((u64)pfn) << PAGE_SHIFT;
75 u64 to = from + size;
76 u64 cursor = from;
77
78 while (cursor < to) {
79 if (!devmem_is_allowed(pfn))
80 return 0;
81 cursor += PAGE_SIZE;
82 pfn++;
83 }
84 return 1;
85}
86#else
87static inline int page_is_allowed(unsigned long pfn)
88{
89 return 1;
90}
91static inline int range_is_allowed(unsigned long pfn, unsigned long size)
92{
93 return 1;
94}
95#endif
96
97#ifndef unxlate_dev_mem_ptr
98#define unxlate_dev_mem_ptr unxlate_dev_mem_ptr
99void __weak unxlate_dev_mem_ptr(phys_addr_t phys, void *addr)
100{
101}
102#endif
103
104static inline bool should_stop_iteration(void)
105{
106 if (need_resched())
107 cond_resched();
108 return fatal_signal_pending(current);
109}
110
111/*
112 * This funcion reads the *physical* memory. The f_pos points directly to the
113 * memory location.
114 */
115static ssize_t read_mem(struct file *file, char __user *buf,
116 size_t count, loff_t *ppos)
117{
118 phys_addr_t p = *ppos;
119 ssize_t read, sz;
120 void *ptr;
121 char *bounce;
122 int err;
123
124 if (p != *ppos)
125 return 0;
126
127 if (!valid_phys_addr_range(p, count))
128 return -EFAULT;
129 read = 0;
130#ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
131 /* we don't have page 0 mapped on sparc and m68k.. */
132 if (p < PAGE_SIZE) {
133 sz = size_inside_page(p, count);
134 if (sz > 0) {
135 if (clear_user(buf, sz))
136 return -EFAULT;
137 buf += sz;
138 p += sz;
139 count -= sz;
140 read += sz;
141 }
142 }
143#endif
144
145 bounce = kmalloc(PAGE_SIZE, GFP_KERNEL);
146 if (!bounce)
147 return -ENOMEM;
148
149 while (count > 0) {
150 unsigned long remaining;
151 int allowed, probe;
152
153 sz = size_inside_page(p, count);
154
155 err = -EPERM;
156 allowed = page_is_allowed(p >> PAGE_SHIFT);
157 if (!allowed)
158 goto failed;
159
160 err = -EFAULT;
161 if (allowed == 2) {
162 /* Show zeros for restricted memory. */
163 remaining = clear_user(buf, sz);
164 } else {
165 /*
166 * On ia64 if a page has been mapped somewhere as
167 * uncached, then it must also be accessed uncached
168 * by the kernel or data corruption may occur.
169 */
170 ptr = xlate_dev_mem_ptr(p);
171 if (!ptr)
172 goto failed;
173
174 probe = copy_from_kernel_nofault(bounce, ptr, sz);
175 unxlate_dev_mem_ptr(p, ptr);
176 if (probe)
177 goto failed;
178
179 remaining = copy_to_user(buf, bounce, sz);
180 }
181
182 if (remaining)
183 goto failed;
184
185 buf += sz;
186 p += sz;
187 count -= sz;
188 read += sz;
189 if (should_stop_iteration())
190 break;
191 }
192 kfree(bounce);
193
194 *ppos += read;
195 return read;
196
197failed:
198 kfree(bounce);
199 return err;
200}
201
202static ssize_t write_mem(struct file *file, const char __user *buf,
203 size_t count, loff_t *ppos)
204{
205 phys_addr_t p = *ppos;
206 ssize_t written, sz;
207 unsigned long copied;
208 void *ptr;
209
210 if (p != *ppos)
211 return -EFBIG;
212
213 if (!valid_phys_addr_range(p, count))
214 return -EFAULT;
215
216 written = 0;
217
218#ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
219 /* we don't have page 0 mapped on sparc and m68k.. */
220 if (p < PAGE_SIZE) {
221 sz = size_inside_page(p, count);
222 /* Hmm. Do something? */
223 buf += sz;
224 p += sz;
225 count -= sz;
226 written += sz;
227 }
228#endif
229
230 while (count > 0) {
231 int allowed;
232
233 sz = size_inside_page(p, count);
234
235 allowed = page_is_allowed(p >> PAGE_SHIFT);
236 if (!allowed)
237 return -EPERM;
238
239 /* Skip actual writing when a page is marked as restricted. */
240 if (allowed == 1) {
241 /*
242 * On ia64 if a page has been mapped somewhere as
243 * uncached, then it must also be accessed uncached
244 * by the kernel or data corruption may occur.
245 */
246 ptr = xlate_dev_mem_ptr(p);
247 if (!ptr) {
248 if (written)
249 break;
250 return -EFAULT;
251 }
252
253 copied = copy_from_user(ptr, buf, sz);
254 unxlate_dev_mem_ptr(p, ptr);
255 if (copied) {
256 written += sz - copied;
257 if (written)
258 break;
259 return -EFAULT;
260 }
261 }
262
263 buf += sz;
264 p += sz;
265 count -= sz;
266 written += sz;
267 if (should_stop_iteration())
268 break;
269 }
270
271 *ppos += written;
272 return written;
273}
274
275int __weak phys_mem_access_prot_allowed(struct file *file,
276 unsigned long pfn, unsigned long size, pgprot_t *vma_prot)
277{
278 return 1;
279}
280
281#ifndef __HAVE_PHYS_MEM_ACCESS_PROT
282
283/*
284 * Architectures vary in how they handle caching for addresses
285 * outside of main memory.
286 *
287 */
288#ifdef pgprot_noncached
289static int uncached_access(struct file *file, phys_addr_t addr)
290{
291#if defined(CONFIG_IA64)
292 /*
293 * On ia64, we ignore O_DSYNC because we cannot tolerate memory
294 * attribute aliases.
295 */
296 return !(efi_mem_attributes(addr) & EFI_MEMORY_WB);
297#elif defined(CONFIG_MIPS)
298 {
299 extern int __uncached_access(struct file *file,
300 unsigned long addr);
301
302 return __uncached_access(file, addr);
303 }
304#else
305 /*
306 * Accessing memory above the top the kernel knows about or through a
307 * file pointer
308 * that was marked O_DSYNC will be done non-cached.
309 */
310 if (file->f_flags & O_DSYNC)
311 return 1;
312 return addr >= __pa(high_memory);
313#endif
314}
315#endif
316
317static pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
318 unsigned long size, pgprot_t vma_prot)
319{
320#ifdef pgprot_noncached
321 phys_addr_t offset = pfn << PAGE_SHIFT;
322
323 if (uncached_access(file, offset))
324 return pgprot_noncached(vma_prot);
325#endif
326 return vma_prot;
327}
328#endif
329
330#ifndef CONFIG_MMU
331static unsigned long get_unmapped_area_mem(struct file *file,
332 unsigned long addr,
333 unsigned long len,
334 unsigned long pgoff,
335 unsigned long flags)
336{
337 if (!valid_mmap_phys_addr_range(pgoff, len))
338 return (unsigned long) -EINVAL;
339 return pgoff << PAGE_SHIFT;
340}
341
342/* permit direct mmap, for read, write or exec */
343static unsigned memory_mmap_capabilities(struct file *file)
344{
345 return NOMMU_MAP_DIRECT |
346 NOMMU_MAP_READ | NOMMU_MAP_WRITE | NOMMU_MAP_EXEC;
347}
348
349static unsigned zero_mmap_capabilities(struct file *file)
350{
351 return NOMMU_MAP_COPY;
352}
353
354/* can't do an in-place private mapping if there's no MMU */
355static inline int private_mapping_ok(struct vm_area_struct *vma)
356{
357 return vma->vm_flags & VM_MAYSHARE;
358}
359#else
360
361static inline int private_mapping_ok(struct vm_area_struct *vma)
362{
363 return 1;
364}
365#endif
366
367static const struct vm_operations_struct mmap_mem_ops = {
368#ifdef CONFIG_HAVE_IOREMAP_PROT
369 .access = generic_access_phys
370#endif
371};
372
373static int mmap_mem(struct file *file, struct vm_area_struct *vma)
374{
375 size_t size = vma->vm_end - vma->vm_start;
376 phys_addr_t offset = (phys_addr_t)vma->vm_pgoff << PAGE_SHIFT;
377
378 /* Does it even fit in phys_addr_t? */
379 if (offset >> PAGE_SHIFT != vma->vm_pgoff)
380 return -EINVAL;
381
382 /* It's illegal to wrap around the end of the physical address space. */
383 if (offset + (phys_addr_t)size - 1 < offset)
384 return -EINVAL;
385
386 if (!valid_mmap_phys_addr_range(vma->vm_pgoff, size))
387 return -EINVAL;
388
389 if (!private_mapping_ok(vma))
390 return -ENOSYS;
391
392 if (!range_is_allowed(vma->vm_pgoff, size))
393 return -EPERM;
394
395 if (!phys_mem_access_prot_allowed(file, vma->vm_pgoff, size,
396 &vma->vm_page_prot))
397 return -EINVAL;
398
399 vma->vm_page_prot = phys_mem_access_prot(file, vma->vm_pgoff,
400 size,
401 vma->vm_page_prot);
402
403 vma->vm_ops = &mmap_mem_ops;
404
405 /* Remap-pfn-range will mark the range VM_IO */
406 if (remap_pfn_range(vma,
407 vma->vm_start,
408 vma->vm_pgoff,
409 size,
410 vma->vm_page_prot)) {
411 return -EAGAIN;
412 }
413 return 0;
414}
415
416static int mmap_kmem(struct file *file, struct vm_area_struct *vma)
417{
418 unsigned long pfn;
419
420 /* Turn a kernel-virtual address into a physical page frame */
421 pfn = __pa((u64)vma->vm_pgoff << PAGE_SHIFT) >> PAGE_SHIFT;
422
423 /*
424 * RED-PEN: on some architectures there is more mapped memory than
425 * available in mem_map which pfn_valid checks for. Perhaps should add a
426 * new macro here.
427 *
428 * RED-PEN: vmalloc is not supported right now.
429 */
430 if (!pfn_valid(pfn))
431 return -EIO;
432
433 vma->vm_pgoff = pfn;
434 return mmap_mem(file, vma);
435}
436
437/*
438 * This function reads the *virtual* memory as seen by the kernel.
439 */
440static ssize_t read_kmem(struct file *file, char __user *buf,
441 size_t count, loff_t *ppos)
442{
443 unsigned long p = *ppos;
444 ssize_t low_count, read, sz;
445 char *kbuf; /* k-addr because vread() takes vmlist_lock rwlock */
446 int err = 0;
447
448 read = 0;
449 if (p < (unsigned long) high_memory) {
450 low_count = count;
451 if (count > (unsigned long)high_memory - p)
452 low_count = (unsigned long)high_memory - p;
453
454#ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
455 /* we don't have page 0 mapped on sparc and m68k.. */
456 if (p < PAGE_SIZE && low_count > 0) {
457 sz = size_inside_page(p, low_count);
458 if (clear_user(buf, sz))
459 return -EFAULT;
460 buf += sz;
461 p += sz;
462 read += sz;
463 low_count -= sz;
464 count -= sz;
465 }
466#endif
467 while (low_count > 0) {
468 sz = size_inside_page(p, low_count);
469
470 /*
471 * On ia64 if a page has been mapped somewhere as
472 * uncached, then it must also be accessed uncached
473 * by the kernel or data corruption may occur
474 */
475 kbuf = xlate_dev_kmem_ptr((void *)p);
476 if (!virt_addr_valid(kbuf))
477 return -ENXIO;
478
479 if (copy_to_user(buf, kbuf, sz))
480 return -EFAULT;
481 buf += sz;
482 p += sz;
483 read += sz;
484 low_count -= sz;
485 count -= sz;
486 if (should_stop_iteration()) {
487 count = 0;
488 break;
489 }
490 }
491 }
492
493 if (count > 0) {
494 kbuf = (char *)__get_free_page(GFP_KERNEL);
495 if (!kbuf)
496 return -ENOMEM;
497 while (count > 0) {
498 sz = size_inside_page(p, count);
499 if (!is_vmalloc_or_module_addr((void *)p)) {
500 err = -ENXIO;
501 break;
502 }
503 sz = vread(kbuf, (char *)p, sz);
504 if (!sz)
505 break;
506 if (copy_to_user(buf, kbuf, sz)) {
507 err = -EFAULT;
508 break;
509 }
510 count -= sz;
511 buf += sz;
512 read += sz;
513 p += sz;
514 if (should_stop_iteration())
515 break;
516 }
517 free_page((unsigned long)kbuf);
518 }
519 *ppos = p;
520 return read ? read : err;
521}
522
523
524static ssize_t do_write_kmem(unsigned long p, const char __user *buf,
525 size_t count, loff_t *ppos)
526{
527 ssize_t written, sz;
528 unsigned long copied;
529
530 written = 0;
531#ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
532 /* we don't have page 0 mapped on sparc and m68k.. */
533 if (p < PAGE_SIZE) {
534 sz = size_inside_page(p, count);
535 /* Hmm. Do something? */
536 buf += sz;
537 p += sz;
538 count -= sz;
539 written += sz;
540 }
541#endif
542
543 while (count > 0) {
544 void *ptr;
545
546 sz = size_inside_page(p, count);
547
548 /*
549 * On ia64 if a page has been mapped somewhere as uncached, then
550 * it must also be accessed uncached by the kernel or data
551 * corruption may occur.
552 */
553 ptr = xlate_dev_kmem_ptr((void *)p);
554 if (!virt_addr_valid(ptr))
555 return -ENXIO;
556
557 copied = copy_from_user(ptr, buf, sz);
558 if (copied) {
559 written += sz - copied;
560 if (written)
561 break;
562 return -EFAULT;
563 }
564 buf += sz;
565 p += sz;
566 count -= sz;
567 written += sz;
568 if (should_stop_iteration())
569 break;
570 }
571
572 *ppos += written;
573 return written;
574}
575
576/*
577 * This function writes to the *virtual* memory as seen by the kernel.
578 */
579static ssize_t write_kmem(struct file *file, const char __user *buf,
580 size_t count, loff_t *ppos)
581{
582 unsigned long p = *ppos;
583 ssize_t wrote = 0;
584 ssize_t virtr = 0;
585 char *kbuf; /* k-addr because vwrite() takes vmlist_lock rwlock */
586 int err = 0;
587
588 if (p < (unsigned long) high_memory) {
589 unsigned long to_write = min_t(unsigned long, count,
590 (unsigned long)high_memory - p);
591 wrote = do_write_kmem(p, buf, to_write, ppos);
592 if (wrote != to_write)
593 return wrote;
594 p += wrote;
595 buf += wrote;
596 count -= wrote;
597 }
598
599 if (count > 0) {
600 kbuf = (char *)__get_free_page(GFP_KERNEL);
601 if (!kbuf)
602 return wrote ? wrote : -ENOMEM;
603 while (count > 0) {
604 unsigned long sz = size_inside_page(p, count);
605 unsigned long n;
606
607 if (!is_vmalloc_or_module_addr((void *)p)) {
608 err = -ENXIO;
609 break;
610 }
611 n = copy_from_user(kbuf, buf, sz);
612 if (n) {
613 err = -EFAULT;
614 break;
615 }
616 vwrite(kbuf, (char *)p, sz);
617 count -= sz;
618 buf += sz;
619 virtr += sz;
620 p += sz;
621 if (should_stop_iteration())
622 break;
623 }
624 free_page((unsigned long)kbuf);
625 }
626
627 *ppos = p;
628 return virtr + wrote ? : err;
629}
630
631static ssize_t read_port(struct file *file, char __user *buf,
632 size_t count, loff_t *ppos)
633{
634 unsigned long i = *ppos;
635 char __user *tmp = buf;
636
637 if (!access_ok(buf, count))
638 return -EFAULT;
639 while (count-- > 0 && i < 65536) {
640 if (__put_user(inb(i), tmp) < 0)
641 return -EFAULT;
642 i++;
643 tmp++;
644 }
645 *ppos = i;
646 return tmp-buf;
647}
648
649static ssize_t write_port(struct file *file, const char __user *buf,
650 size_t count, loff_t *ppos)
651{
652 unsigned long i = *ppos;
653 const char __user *tmp = buf;
654
655 if (!access_ok(buf, count))
656 return -EFAULT;
657 while (count-- > 0 && i < 65536) {
658 char c;
659
660 if (__get_user(c, tmp)) {
661 if (tmp > buf)
662 break;
663 return -EFAULT;
664 }
665 outb(c, i);
666 i++;
667 tmp++;
668 }
669 *ppos = i;
670 return tmp-buf;
671}
672
673static ssize_t read_null(struct file *file, char __user *buf,
674 size_t count, loff_t *ppos)
675{
676 return 0;
677}
678
679static ssize_t write_null(struct file *file, const char __user *buf,
680 size_t count, loff_t *ppos)
681{
682 return count;
683}
684
685static ssize_t read_iter_null(struct kiocb *iocb, struct iov_iter *to)
686{
687 return 0;
688}
689
690static ssize_t write_iter_null(struct kiocb *iocb, struct iov_iter *from)
691{
692 size_t count = iov_iter_count(from);
693 iov_iter_advance(from, count);
694 return count;
695}
696
697static int pipe_to_null(struct pipe_inode_info *info, struct pipe_buffer *buf,
698 struct splice_desc *sd)
699{
700 return sd->len;
701}
702
703static ssize_t splice_write_null(struct pipe_inode_info *pipe, struct file *out,
704 loff_t *ppos, size_t len, unsigned int flags)
705{
706 return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_null);
707}
708
709static ssize_t read_iter_zero(struct kiocb *iocb, struct iov_iter *iter)
710{
711 size_t written = 0;
712
713 while (iov_iter_count(iter)) {
714 size_t chunk = iov_iter_count(iter), n;
715
716 if (chunk > PAGE_SIZE)
717 chunk = PAGE_SIZE; /* Just for latency reasons */
718 n = iov_iter_zero(chunk, iter);
719 if (!n && iov_iter_count(iter))
720 return written ? written : -EFAULT;
721 written += n;
722 if (signal_pending(current))
723 return written ? written : -ERESTARTSYS;
724 cond_resched();
725 }
726 return written;
727}
728
729static int mmap_zero(struct file *file, struct vm_area_struct *vma)
730{
731#ifndef CONFIG_MMU
732 return -ENOSYS;
733#endif
734 if (vma->vm_flags & VM_SHARED)
735 return shmem_zero_setup(vma);
736 vma_set_anonymous(vma);
737 return 0;
738}
739
740static unsigned long get_unmapped_area_zero(struct file *file,
741 unsigned long addr, unsigned long len,
742 unsigned long pgoff, unsigned long flags)
743{
744#ifdef CONFIG_MMU
745 if (flags & MAP_SHARED) {
746 /*
747 * mmap_zero() will call shmem_zero_setup() to create a file,
748 * so use shmem's get_unmapped_area in case it can be huge;
749 * and pass NULL for file as in mmap.c's get_unmapped_area(),
750 * so as not to confuse shmem with our handle on "/dev/zero".
751 */
752 return shmem_get_unmapped_area(NULL, addr, len, pgoff, flags);
753 }
754
755 /* Otherwise flags & MAP_PRIVATE: with no shmem object beneath it */
756 return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
757#else
758 return -ENOSYS;
759#endif
760}
761
762static ssize_t write_full(struct file *file, const char __user *buf,
763 size_t count, loff_t *ppos)
764{
765 return -ENOSPC;
766}
767
768/*
769 * Special lseek() function for /dev/null and /dev/zero. Most notably, you
770 * can fopen() both devices with "a" now. This was previously impossible.
771 * -- SRB.
772 */
773static loff_t null_lseek(struct file *file, loff_t offset, int orig)
774{
775 return file->f_pos = 0;
776}
777
778/*
779 * The memory devices use the full 32/64 bits of the offset, and so we cannot
780 * check against negative addresses: they are ok. The return value is weird,
781 * though, in that case (0).
782 *
783 * also note that seeking relative to the "end of file" isn't supported:
784 * it has no meaning, so it returns -EINVAL.
785 */
786static loff_t memory_lseek(struct file *file, loff_t offset, int orig)
787{
788 loff_t ret;
789
790 inode_lock(file_inode(file));
791 switch (orig) {
792 case SEEK_CUR:
793 offset += file->f_pos;
794 fallthrough;
795 case SEEK_SET:
796 /* to avoid userland mistaking f_pos=-9 as -EBADF=-9 */
797 if ((unsigned long long)offset >= -MAX_ERRNO) {
798 ret = -EOVERFLOW;
799 break;
800 }
801 file->f_pos = offset;
802 ret = file->f_pos;
803 force_successful_syscall_return();
804 break;
805 default:
806 ret = -EINVAL;
807 }
808 inode_unlock(file_inode(file));
809 return ret;
810}
811
812static struct inode *devmem_inode;
813
814#ifdef CONFIG_IO_STRICT_DEVMEM
815void revoke_devmem(struct resource *res)
816{
817 /* pairs with smp_store_release() in devmem_init_inode() */
818 struct inode *inode = smp_load_acquire(&devmem_inode);
819
820 /*
821 * Check that the initialization has completed. Losing the race
822 * is ok because it means drivers are claiming resources before
823 * the fs_initcall level of init and prevent /dev/mem from
824 * establishing mappings.
825 */
826 if (!inode)
827 return;
828
829 /*
830 * The expectation is that the driver has successfully marked
831 * the resource busy by this point, so devmem_is_allowed()
832 * should start returning false, however for performance this
833 * does not iterate the entire resource range.
834 */
835 if (devmem_is_allowed(PHYS_PFN(res->start)) &&
836 devmem_is_allowed(PHYS_PFN(res->end))) {
837 /*
838 * *cringe* iomem=relaxed says "go ahead, what's the
839 * worst that can happen?"
840 */
841 return;
842 }
843
844 unmap_mapping_range(inode->i_mapping, res->start, resource_size(res), 1);
845}
846#endif
847
848static int open_port(struct inode *inode, struct file *filp)
849{
850 int rc;
851
852 if (!capable(CAP_SYS_RAWIO))
853 return -EPERM;
854
855 rc = security_locked_down(LOCKDOWN_DEV_MEM);
856 if (rc)
857 return rc;
858
859 if (iminor(inode) != DEVMEM_MINOR)
860 return 0;
861
862 /*
863 * Use a unified address space to have a single point to manage
864 * revocations when drivers want to take over a /dev/mem mapped
865 * range.
866 */
867 inode->i_mapping = devmem_inode->i_mapping;
868 filp->f_mapping = inode->i_mapping;
869
870 return 0;
871}
872
873#define zero_lseek null_lseek
874#define full_lseek null_lseek
875#define write_zero write_null
876#define write_iter_zero write_iter_null
877#define open_mem open_port
878#define open_kmem open_mem
879
880static const struct file_operations __maybe_unused mem_fops = {
881 .llseek = memory_lseek,
882 .read = read_mem,
883 .write = write_mem,
884 .mmap = mmap_mem,
885 .open = open_mem,
886#ifndef CONFIG_MMU
887 .get_unmapped_area = get_unmapped_area_mem,
888 .mmap_capabilities = memory_mmap_capabilities,
889#endif
890};
891
892static const struct file_operations __maybe_unused kmem_fops = {
893 .llseek = memory_lseek,
894 .read = read_kmem,
895 .write = write_kmem,
896 .mmap = mmap_kmem,
897 .open = open_kmem,
898#ifndef CONFIG_MMU
899 .get_unmapped_area = get_unmapped_area_mem,
900 .mmap_capabilities = memory_mmap_capabilities,
901#endif
902};
903
904static const struct file_operations null_fops = {
905 .llseek = null_lseek,
906 .read = read_null,
907 .write = write_null,
908 .read_iter = read_iter_null,
909 .write_iter = write_iter_null,
910 .splice_write = splice_write_null,
911};
912
913static const struct file_operations __maybe_unused port_fops = {
914 .llseek = memory_lseek,
915 .read = read_port,
916 .write = write_port,
917 .open = open_port,
918};
919
920static const struct file_operations zero_fops = {
921 .llseek = zero_lseek,
922 .write = write_zero,
923 .read_iter = read_iter_zero,
924 .write_iter = write_iter_zero,
925 .mmap = mmap_zero,
926 .get_unmapped_area = get_unmapped_area_zero,
927#ifndef CONFIG_MMU
928 .mmap_capabilities = zero_mmap_capabilities,
929#endif
930};
931
932static const struct file_operations full_fops = {
933 .llseek = full_lseek,
934 .read_iter = read_iter_zero,
935 .write = write_full,
936};
937
938static const struct memdev {
939 const char *name;
940 umode_t mode;
941 const struct file_operations *fops;
942 fmode_t fmode;
943} devlist[] = {
944#ifdef CONFIG_DEVMEM
945 [DEVMEM_MINOR] = { "mem", 0, &mem_fops, FMODE_UNSIGNED_OFFSET },
946#endif
947#ifdef CONFIG_DEVKMEM
948 [2] = { "kmem", 0, &kmem_fops, FMODE_UNSIGNED_OFFSET },
949#endif
950 [3] = { "null", 0666, &null_fops, 0 },
951#ifdef CONFIG_DEVPORT
952 [4] = { "port", 0, &port_fops, 0 },
953#endif
954 [5] = { "zero", 0666, &zero_fops, 0 },
955 [7] = { "full", 0666, &full_fops, 0 },
956 [8] = { "random", 0666, &random_fops, 0 },
957 [9] = { "urandom", 0666, &urandom_fops, 0 },
958#ifdef CONFIG_PRINTK
959 [11] = { "kmsg", 0644, &kmsg_fops, 0 },
960#endif
961};
962
963static int memory_open(struct inode *inode, struct file *filp)
964{
965 int minor;
966 const struct memdev *dev;
967
968 minor = iminor(inode);
969 if (minor >= ARRAY_SIZE(devlist))
970 return -ENXIO;
971
972 dev = &devlist[minor];
973 if (!dev->fops)
974 return -ENXIO;
975
976 filp->f_op = dev->fops;
977 filp->f_mode |= dev->fmode;
978
979 if (dev->fops->open)
980 return dev->fops->open(inode, filp);
981
982 return 0;
983}
984
985static const struct file_operations memory_fops = {
986 .open = memory_open,
987 .llseek = noop_llseek,
988};
989
990static char *mem_devnode(struct device *dev, umode_t *mode)
991{
992 if (mode && devlist[MINOR(dev->devt)].mode)
993 *mode = devlist[MINOR(dev->devt)].mode;
994 return NULL;
995}
996
997static struct class *mem_class;
998
999static int devmem_fs_init_fs_context(struct fs_context *fc)
1000{
1001 return init_pseudo(fc, DEVMEM_MAGIC) ? 0 : -ENOMEM;
1002}
1003
1004static struct file_system_type devmem_fs_type = {
1005 .name = "devmem",
1006 .owner = THIS_MODULE,
1007 .init_fs_context = devmem_fs_init_fs_context,
1008 .kill_sb = kill_anon_super,
1009};
1010
1011static int devmem_init_inode(void)
1012{
1013 static struct vfsmount *devmem_vfs_mount;
1014 static int devmem_fs_cnt;
1015 struct inode *inode;
1016 int rc;
1017
1018 rc = simple_pin_fs(&devmem_fs_type, &devmem_vfs_mount, &devmem_fs_cnt);
1019 if (rc < 0) {
1020 pr_err("Cannot mount /dev/mem pseudo filesystem: %d\n", rc);
1021 return rc;
1022 }
1023
1024 inode = alloc_anon_inode(devmem_vfs_mount->mnt_sb);
1025 if (IS_ERR(inode)) {
1026 rc = PTR_ERR(inode);
1027 pr_err("Cannot allocate inode for /dev/mem: %d\n", rc);
1028 simple_release_fs(&devmem_vfs_mount, &devmem_fs_cnt);
1029 return rc;
1030 }
1031
1032 /*
1033 * Publish /dev/mem initialized.
1034 * Pairs with smp_load_acquire() in revoke_devmem().
1035 */
1036 smp_store_release(&devmem_inode, inode);
1037
1038 return 0;
1039}
1040
1041static int __init chr_dev_init(void)
1042{
1043 int minor;
1044
1045 if (register_chrdev(MEM_MAJOR, "mem", &memory_fops))
1046 printk("unable to get major %d for memory devs\n", MEM_MAJOR);
1047
1048 mem_class = class_create(THIS_MODULE, "mem");
1049 if (IS_ERR(mem_class))
1050 return PTR_ERR(mem_class);
1051
1052 mem_class->devnode = mem_devnode;
1053 for (minor = 1; minor < ARRAY_SIZE(devlist); minor++) {
1054 if (!devlist[minor].name)
1055 continue;
1056
1057 /*
1058 * Create /dev/port?
1059 */
1060 if ((minor == DEVPORT_MINOR) && !arch_has_dev_port())
1061 continue;
1062 if ((minor == DEVMEM_MINOR) && devmem_init_inode() != 0)
1063 continue;
1064
1065 device_create(mem_class, NULL, MKDEV(MEM_MAJOR, minor),
1066 NULL, devlist[minor].name);
1067 }
1068
1069 return tty_init();
1070}
1071
1072fs_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/backing-dev.h>
25#include <linux/splice.h>
26#include <linux/pfn.h>
27#include <linux/export.h>
28#include <linux/io.h>
29#include <linux/aio.h>
30
31#include <asm/uaccess.h>
32
33#ifdef CONFIG_IA64
34# include <linux/efi.h>
35#endif
36
37#define DEVPORT_MINOR 4
38
39static inline unsigned long size_inside_page(unsigned long start,
40 unsigned long size)
41{
42 unsigned long sz;
43
44 sz = PAGE_SIZE - (start & (PAGE_SIZE - 1));
45
46 return min(sz, size);
47}
48
49#ifndef ARCH_HAS_VALID_PHYS_ADDR_RANGE
50static inline int valid_phys_addr_range(phys_addr_t addr, size_t count)
51{
52 return addr + count <= __pa(high_memory);
53}
54
55static inline int valid_mmap_phys_addr_range(unsigned long pfn, size_t size)
56{
57 return 1;
58}
59#endif
60
61#ifdef CONFIG_STRICT_DEVMEM
62static inline int range_is_allowed(unsigned long pfn, unsigned long size)
63{
64 u64 from = ((u64)pfn) << PAGE_SHIFT;
65 u64 to = from + size;
66 u64 cursor = from;
67
68 while (cursor < to) {
69 if (!devmem_is_allowed(pfn)) {
70 printk(KERN_INFO
71 "Program %s tried to access /dev/mem between %Lx->%Lx.\n",
72 current->comm, from, to);
73 return 0;
74 }
75 cursor += PAGE_SIZE;
76 pfn++;
77 }
78 return 1;
79}
80#else
81static inline int range_is_allowed(unsigned long pfn, unsigned long size)
82{
83 return 1;
84}
85#endif
86
87void __weak unxlate_dev_mem_ptr(unsigned long phys, void *addr)
88{
89}
90
91/*
92 * This funcion reads the *physical* memory. The f_pos points directly to the
93 * memory location.
94 */
95static ssize_t read_mem(struct file *file, char __user *buf,
96 size_t count, loff_t *ppos)
97{
98 phys_addr_t p = *ppos;
99 ssize_t read, sz;
100 char *ptr;
101
102 if (p != *ppos)
103 return 0;
104
105 if (!valid_phys_addr_range(p, count))
106 return -EFAULT;
107 read = 0;
108#ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
109 /* we don't have page 0 mapped on sparc and m68k.. */
110 if (p < PAGE_SIZE) {
111 sz = size_inside_page(p, count);
112 if (sz > 0) {
113 if (clear_user(buf, sz))
114 return -EFAULT;
115 buf += sz;
116 p += sz;
117 count -= sz;
118 read += sz;
119 }
120 }
121#endif
122
123 while (count > 0) {
124 unsigned long remaining;
125
126 sz = size_inside_page(p, count);
127
128 if (!range_is_allowed(p >> PAGE_SHIFT, count))
129 return -EPERM;
130
131 /*
132 * On ia64 if a page has been mapped somewhere as uncached, then
133 * it must also be accessed uncached by the kernel or data
134 * corruption may occur.
135 */
136 ptr = xlate_dev_mem_ptr(p);
137 if (!ptr)
138 return -EFAULT;
139
140 remaining = copy_to_user(buf, ptr, sz);
141 unxlate_dev_mem_ptr(p, ptr);
142 if (remaining)
143 return -EFAULT;
144
145 buf += sz;
146 p += sz;
147 count -= sz;
148 read += sz;
149 }
150
151 *ppos += read;
152 return read;
153}
154
155static ssize_t write_mem(struct file *file, const char __user *buf,
156 size_t count, loff_t *ppos)
157{
158 phys_addr_t p = *ppos;
159 ssize_t written, sz;
160 unsigned long copied;
161 void *ptr;
162
163 if (p != *ppos)
164 return -EFBIG;
165
166 if (!valid_phys_addr_range(p, count))
167 return -EFAULT;
168
169 written = 0;
170
171#ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
172 /* we don't have page 0 mapped on sparc and m68k.. */
173 if (p < PAGE_SIZE) {
174 sz = size_inside_page(p, count);
175 /* Hmm. Do something? */
176 buf += sz;
177 p += sz;
178 count -= sz;
179 written += sz;
180 }
181#endif
182
183 while (count > 0) {
184 sz = size_inside_page(p, count);
185
186 if (!range_is_allowed(p >> PAGE_SHIFT, sz))
187 return -EPERM;
188
189 /*
190 * On ia64 if a page has been mapped somewhere as uncached, then
191 * it must also be accessed uncached by the kernel or data
192 * corruption may occur.
193 */
194 ptr = xlate_dev_mem_ptr(p);
195 if (!ptr) {
196 if (written)
197 break;
198 return -EFAULT;
199 }
200
201 copied = copy_from_user(ptr, buf, sz);
202 unxlate_dev_mem_ptr(p, ptr);
203 if (copied) {
204 written += sz - copied;
205 if (written)
206 break;
207 return -EFAULT;
208 }
209
210 buf += sz;
211 p += sz;
212 count -= sz;
213 written += sz;
214 }
215
216 *ppos += written;
217 return written;
218}
219
220int __weak phys_mem_access_prot_allowed(struct file *file,
221 unsigned long pfn, unsigned long size, pgprot_t *vma_prot)
222{
223 return 1;
224}
225
226#ifndef __HAVE_PHYS_MEM_ACCESS_PROT
227
228/*
229 * Architectures vary in how they handle caching for addresses
230 * outside of main memory.
231 *
232 */
233#ifdef pgprot_noncached
234static int uncached_access(struct file *file, phys_addr_t addr)
235{
236#if defined(CONFIG_IA64)
237 /*
238 * On ia64, we ignore O_DSYNC because we cannot tolerate memory
239 * attribute aliases.
240 */
241 return !(efi_mem_attributes(addr) & EFI_MEMORY_WB);
242#elif defined(CONFIG_MIPS)
243 {
244 extern int __uncached_access(struct file *file,
245 unsigned long addr);
246
247 return __uncached_access(file, addr);
248 }
249#else
250 /*
251 * Accessing memory above the top the kernel knows about or through a
252 * file pointer
253 * that was marked O_DSYNC will be done non-cached.
254 */
255 if (file->f_flags & O_DSYNC)
256 return 1;
257 return addr >= __pa(high_memory);
258#endif
259}
260#endif
261
262static pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
263 unsigned long size, pgprot_t vma_prot)
264{
265#ifdef pgprot_noncached
266 phys_addr_t offset = pfn << PAGE_SHIFT;
267
268 if (uncached_access(file, offset))
269 return pgprot_noncached(vma_prot);
270#endif
271 return vma_prot;
272}
273#endif
274
275#ifndef CONFIG_MMU
276static unsigned long get_unmapped_area_mem(struct file *file,
277 unsigned long addr,
278 unsigned long len,
279 unsigned long pgoff,
280 unsigned long flags)
281{
282 if (!valid_mmap_phys_addr_range(pgoff, len))
283 return (unsigned long) -EINVAL;
284 return pgoff << PAGE_SHIFT;
285}
286
287/* can't do an in-place private mapping if there's no MMU */
288static inline int private_mapping_ok(struct vm_area_struct *vma)
289{
290 return vma->vm_flags & VM_MAYSHARE;
291}
292#else
293#define get_unmapped_area_mem NULL
294
295static inline int private_mapping_ok(struct vm_area_struct *vma)
296{
297 return 1;
298}
299#endif
300
301static const struct vm_operations_struct mmap_mem_ops = {
302#ifdef CONFIG_HAVE_IOREMAP_PROT
303 .access = generic_access_phys
304#endif
305};
306
307static int mmap_mem(struct file *file, struct vm_area_struct *vma)
308{
309 size_t size = vma->vm_end - vma->vm_start;
310
311 if (!valid_mmap_phys_addr_range(vma->vm_pgoff, size))
312 return -EINVAL;
313
314 if (!private_mapping_ok(vma))
315 return -ENOSYS;
316
317 if (!range_is_allowed(vma->vm_pgoff, size))
318 return -EPERM;
319
320 if (!phys_mem_access_prot_allowed(file, vma->vm_pgoff, size,
321 &vma->vm_page_prot))
322 return -EINVAL;
323
324 vma->vm_page_prot = phys_mem_access_prot(file, vma->vm_pgoff,
325 size,
326 vma->vm_page_prot);
327
328 vma->vm_ops = &mmap_mem_ops;
329
330 /* Remap-pfn-range will mark the range VM_IO */
331 if (remap_pfn_range(vma,
332 vma->vm_start,
333 vma->vm_pgoff,
334 size,
335 vma->vm_page_prot)) {
336 return -EAGAIN;
337 }
338 return 0;
339}
340
341#ifdef CONFIG_DEVKMEM
342static int mmap_kmem(struct file *file, struct vm_area_struct *vma)
343{
344 unsigned long pfn;
345
346 /* Turn a kernel-virtual address into a physical page frame */
347 pfn = __pa((u64)vma->vm_pgoff << PAGE_SHIFT) >> PAGE_SHIFT;
348
349 /*
350 * RED-PEN: on some architectures there is more mapped memory than
351 * available in mem_map which pfn_valid checks for. Perhaps should add a
352 * new macro here.
353 *
354 * RED-PEN: vmalloc is not supported right now.
355 */
356 if (!pfn_valid(pfn))
357 return -EIO;
358
359 vma->vm_pgoff = pfn;
360 return mmap_mem(file, vma);
361}
362#endif
363
364#ifdef CONFIG_DEVKMEM
365/*
366 * This function reads the *virtual* memory as seen by the kernel.
367 */
368static ssize_t read_kmem(struct file *file, char __user *buf,
369 size_t count, loff_t *ppos)
370{
371 unsigned long p = *ppos;
372 ssize_t low_count, read, sz;
373 char *kbuf; /* k-addr because vread() takes vmlist_lock rwlock */
374 int err = 0;
375
376 read = 0;
377 if (p < (unsigned long) high_memory) {
378 low_count = count;
379 if (count > (unsigned long)high_memory - p)
380 low_count = (unsigned long)high_memory - p;
381
382#ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
383 /* we don't have page 0 mapped on sparc and m68k.. */
384 if (p < PAGE_SIZE && low_count > 0) {
385 sz = size_inside_page(p, low_count);
386 if (clear_user(buf, sz))
387 return -EFAULT;
388 buf += sz;
389 p += sz;
390 read += sz;
391 low_count -= sz;
392 count -= sz;
393 }
394#endif
395 while (low_count > 0) {
396 sz = size_inside_page(p, low_count);
397
398 /*
399 * On ia64 if a page has been mapped somewhere as
400 * uncached, then it must also be accessed uncached
401 * by the kernel or data corruption may occur
402 */
403 kbuf = xlate_dev_kmem_ptr((char *)p);
404
405 if (copy_to_user(buf, kbuf, sz))
406 return -EFAULT;
407 buf += sz;
408 p += sz;
409 read += sz;
410 low_count -= sz;
411 count -= sz;
412 }
413 }
414
415 if (count > 0) {
416 kbuf = (char *)__get_free_page(GFP_KERNEL);
417 if (!kbuf)
418 return -ENOMEM;
419 while (count > 0) {
420 sz = size_inside_page(p, count);
421 if (!is_vmalloc_or_module_addr((void *)p)) {
422 err = -ENXIO;
423 break;
424 }
425 sz = vread(kbuf, (char *)p, sz);
426 if (!sz)
427 break;
428 if (copy_to_user(buf, kbuf, sz)) {
429 err = -EFAULT;
430 break;
431 }
432 count -= sz;
433 buf += sz;
434 read += sz;
435 p += sz;
436 }
437 free_page((unsigned long)kbuf);
438 }
439 *ppos = p;
440 return read ? read : err;
441}
442
443
444static ssize_t do_write_kmem(unsigned long p, const char __user *buf,
445 size_t count, loff_t *ppos)
446{
447 ssize_t written, sz;
448 unsigned long copied;
449
450 written = 0;
451#ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
452 /* we don't have page 0 mapped on sparc and m68k.. */
453 if (p < PAGE_SIZE) {
454 sz = size_inside_page(p, count);
455 /* Hmm. Do something? */
456 buf += sz;
457 p += sz;
458 count -= sz;
459 written += sz;
460 }
461#endif
462
463 while (count > 0) {
464 char *ptr;
465
466 sz = size_inside_page(p, count);
467
468 /*
469 * On ia64 if a page has been mapped somewhere as uncached, then
470 * it must also be accessed uncached by the kernel or data
471 * corruption may occur.
472 */
473 ptr = xlate_dev_kmem_ptr((char *)p);
474
475 copied = copy_from_user(ptr, buf, sz);
476 if (copied) {
477 written += sz - copied;
478 if (written)
479 break;
480 return -EFAULT;
481 }
482 buf += sz;
483 p += sz;
484 count -= sz;
485 written += sz;
486 }
487
488 *ppos += written;
489 return written;
490}
491
492/*
493 * This function writes to the *virtual* memory as seen by the kernel.
494 */
495static ssize_t write_kmem(struct file *file, const char __user *buf,
496 size_t count, loff_t *ppos)
497{
498 unsigned long p = *ppos;
499 ssize_t wrote = 0;
500 ssize_t virtr = 0;
501 char *kbuf; /* k-addr because vwrite() takes vmlist_lock rwlock */
502 int err = 0;
503
504 if (p < (unsigned long) high_memory) {
505 unsigned long to_write = min_t(unsigned long, count,
506 (unsigned long)high_memory - p);
507 wrote = do_write_kmem(p, buf, to_write, ppos);
508 if (wrote != to_write)
509 return wrote;
510 p += wrote;
511 buf += wrote;
512 count -= wrote;
513 }
514
515 if (count > 0) {
516 kbuf = (char *)__get_free_page(GFP_KERNEL);
517 if (!kbuf)
518 return wrote ? wrote : -ENOMEM;
519 while (count > 0) {
520 unsigned long sz = size_inside_page(p, count);
521 unsigned long n;
522
523 if (!is_vmalloc_or_module_addr((void *)p)) {
524 err = -ENXIO;
525 break;
526 }
527 n = copy_from_user(kbuf, buf, sz);
528 if (n) {
529 err = -EFAULT;
530 break;
531 }
532 vwrite(kbuf, (char *)p, sz);
533 count -= sz;
534 buf += sz;
535 virtr += sz;
536 p += sz;
537 }
538 free_page((unsigned long)kbuf);
539 }
540
541 *ppos = p;
542 return virtr + wrote ? : err;
543}
544#endif
545
546#ifdef CONFIG_DEVPORT
547static ssize_t read_port(struct file *file, char __user *buf,
548 size_t count, loff_t *ppos)
549{
550 unsigned long i = *ppos;
551 char __user *tmp = buf;
552
553 if (!access_ok(VERIFY_WRITE, buf, count))
554 return -EFAULT;
555 while (count-- > 0 && i < 65536) {
556 if (__put_user(inb(i), tmp) < 0)
557 return -EFAULT;
558 i++;
559 tmp++;
560 }
561 *ppos = i;
562 return tmp-buf;
563}
564
565static ssize_t write_port(struct file *file, const char __user *buf,
566 size_t count, loff_t *ppos)
567{
568 unsigned long i = *ppos;
569 const char __user *tmp = buf;
570
571 if (!access_ok(VERIFY_READ, buf, count))
572 return -EFAULT;
573 while (count-- > 0 && i < 65536) {
574 char c;
575 if (__get_user(c, tmp)) {
576 if (tmp > buf)
577 break;
578 return -EFAULT;
579 }
580 outb(c, i);
581 i++;
582 tmp++;
583 }
584 *ppos = i;
585 return tmp-buf;
586}
587#endif
588
589static ssize_t read_null(struct file *file, char __user *buf,
590 size_t count, loff_t *ppos)
591{
592 return 0;
593}
594
595static ssize_t write_null(struct file *file, const char __user *buf,
596 size_t count, loff_t *ppos)
597{
598 return count;
599}
600
601static ssize_t aio_read_null(struct kiocb *iocb, const struct iovec *iov,
602 unsigned long nr_segs, loff_t pos)
603{
604 return 0;
605}
606
607static ssize_t aio_write_null(struct kiocb *iocb, const struct iovec *iov,
608 unsigned long nr_segs, loff_t pos)
609{
610 return iov_length(iov, nr_segs);
611}
612
613static int pipe_to_null(struct pipe_inode_info *info, struct pipe_buffer *buf,
614 struct splice_desc *sd)
615{
616 return sd->len;
617}
618
619static ssize_t splice_write_null(struct pipe_inode_info *pipe, struct file *out,
620 loff_t *ppos, size_t len, unsigned int flags)
621{
622 return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_null);
623}
624
625static ssize_t read_zero(struct file *file, char __user *buf,
626 size_t count, loff_t *ppos)
627{
628 size_t written;
629
630 if (!count)
631 return 0;
632
633 if (!access_ok(VERIFY_WRITE, buf, count))
634 return -EFAULT;
635
636 written = 0;
637 while (count) {
638 unsigned long unwritten;
639 size_t chunk = count;
640
641 if (chunk > PAGE_SIZE)
642 chunk = PAGE_SIZE; /* Just for latency reasons */
643 unwritten = __clear_user(buf, chunk);
644 written += chunk - unwritten;
645 if (unwritten)
646 break;
647 if (signal_pending(current))
648 return written ? written : -ERESTARTSYS;
649 buf += chunk;
650 count -= chunk;
651 cond_resched();
652 }
653 return written ? written : -EFAULT;
654}
655
656static ssize_t aio_read_zero(struct kiocb *iocb, const struct iovec *iov,
657 unsigned long nr_segs, loff_t pos)
658{
659 size_t written = 0;
660 unsigned long i;
661 ssize_t ret;
662
663 for (i = 0; i < nr_segs; i++) {
664 ret = read_zero(iocb->ki_filp, iov[i].iov_base, iov[i].iov_len,
665 &pos);
666 if (ret < 0)
667 break;
668 written += ret;
669 }
670
671 return written ? written : -EFAULT;
672}
673
674static int mmap_zero(struct file *file, struct vm_area_struct *vma)
675{
676#ifndef CONFIG_MMU
677 return -ENOSYS;
678#endif
679 if (vma->vm_flags & VM_SHARED)
680 return shmem_zero_setup(vma);
681 return 0;
682}
683
684static ssize_t write_full(struct file *file, const char __user *buf,
685 size_t count, loff_t *ppos)
686{
687 return -ENOSPC;
688}
689
690/*
691 * Special lseek() function for /dev/null and /dev/zero. Most notably, you
692 * can fopen() both devices with "a" now. This was previously impossible.
693 * -- SRB.
694 */
695static loff_t null_lseek(struct file *file, loff_t offset, int orig)
696{
697 return file->f_pos = 0;
698}
699
700/*
701 * The memory devices use the full 32/64 bits of the offset, and so we cannot
702 * check against negative addresses: they are ok. The return value is weird,
703 * though, in that case (0).
704 *
705 * also note that seeking relative to the "end of file" isn't supported:
706 * it has no meaning, so it returns -EINVAL.
707 */
708static loff_t memory_lseek(struct file *file, loff_t offset, int orig)
709{
710 loff_t ret;
711
712 mutex_lock(&file_inode(file)->i_mutex);
713 switch (orig) {
714 case SEEK_CUR:
715 offset += file->f_pos;
716 case SEEK_SET:
717 /* to avoid userland mistaking f_pos=-9 as -EBADF=-9 */
718 if (IS_ERR_VALUE((unsigned long long)offset)) {
719 ret = -EOVERFLOW;
720 break;
721 }
722 file->f_pos = offset;
723 ret = file->f_pos;
724 force_successful_syscall_return();
725 break;
726 default:
727 ret = -EINVAL;
728 }
729 mutex_unlock(&file_inode(file)->i_mutex);
730 return ret;
731}
732
733static int open_port(struct inode *inode, struct file *filp)
734{
735 return capable(CAP_SYS_RAWIO) ? 0 : -EPERM;
736}
737
738#define zero_lseek null_lseek
739#define full_lseek null_lseek
740#define write_zero write_null
741#define read_full read_zero
742#define aio_write_zero aio_write_null
743#define open_mem open_port
744#define open_kmem open_mem
745
746static const struct file_operations mem_fops = {
747 .llseek = memory_lseek,
748 .read = read_mem,
749 .write = write_mem,
750 .mmap = mmap_mem,
751 .open = open_mem,
752 .get_unmapped_area = get_unmapped_area_mem,
753};
754
755#ifdef CONFIG_DEVKMEM
756static const struct file_operations kmem_fops = {
757 .llseek = memory_lseek,
758 .read = read_kmem,
759 .write = write_kmem,
760 .mmap = mmap_kmem,
761 .open = open_kmem,
762 .get_unmapped_area = get_unmapped_area_mem,
763};
764#endif
765
766static const struct file_operations null_fops = {
767 .llseek = null_lseek,
768 .read = read_null,
769 .write = write_null,
770 .aio_read = aio_read_null,
771 .aio_write = aio_write_null,
772 .splice_write = splice_write_null,
773};
774
775#ifdef CONFIG_DEVPORT
776static const struct file_operations port_fops = {
777 .llseek = memory_lseek,
778 .read = read_port,
779 .write = write_port,
780 .open = open_port,
781};
782#endif
783
784static const struct file_operations zero_fops = {
785 .llseek = zero_lseek,
786 .read = read_zero,
787 .write = write_zero,
788 .aio_read = aio_read_zero,
789 .aio_write = aio_write_zero,
790 .mmap = mmap_zero,
791};
792
793/*
794 * capabilities for /dev/zero
795 * - permits private mappings, "copies" are taken of the source of zeros
796 * - no writeback happens
797 */
798static struct backing_dev_info zero_bdi = {
799 .name = "char/mem",
800 .capabilities = BDI_CAP_MAP_COPY | BDI_CAP_NO_ACCT_AND_WRITEBACK,
801};
802
803static const struct file_operations full_fops = {
804 .llseek = full_lseek,
805 .read = read_full,
806 .write = write_full,
807};
808
809static const struct memdev {
810 const char *name;
811 umode_t mode;
812 const struct file_operations *fops;
813 struct backing_dev_info *dev_info;
814} devlist[] = {
815 [1] = { "mem", 0, &mem_fops, &directly_mappable_cdev_bdi },
816#ifdef CONFIG_DEVKMEM
817 [2] = { "kmem", 0, &kmem_fops, &directly_mappable_cdev_bdi },
818#endif
819 [3] = { "null", 0666, &null_fops, NULL },
820#ifdef CONFIG_DEVPORT
821 [4] = { "port", 0, &port_fops, NULL },
822#endif
823 [5] = { "zero", 0666, &zero_fops, &zero_bdi },
824 [7] = { "full", 0666, &full_fops, NULL },
825 [8] = { "random", 0666, &random_fops, NULL },
826 [9] = { "urandom", 0666, &urandom_fops, NULL },
827#ifdef CONFIG_PRINTK
828 [11] = { "kmsg", 0644, &kmsg_fops, NULL },
829#endif
830};
831
832static int memory_open(struct inode *inode, struct file *filp)
833{
834 int minor;
835 const struct memdev *dev;
836
837 minor = iminor(inode);
838 if (minor >= ARRAY_SIZE(devlist))
839 return -ENXIO;
840
841 dev = &devlist[minor];
842 if (!dev->fops)
843 return -ENXIO;
844
845 filp->f_op = dev->fops;
846 if (dev->dev_info)
847 filp->f_mapping->backing_dev_info = dev->dev_info;
848
849 /* Is /dev/mem or /dev/kmem ? */
850 if (dev->dev_info == &directly_mappable_cdev_bdi)
851 filp->f_mode |= FMODE_UNSIGNED_OFFSET;
852
853 if (dev->fops->open)
854 return dev->fops->open(inode, filp);
855
856 return 0;
857}
858
859static const struct file_operations memory_fops = {
860 .open = memory_open,
861 .llseek = noop_llseek,
862};
863
864static char *mem_devnode(struct device *dev, umode_t *mode)
865{
866 if (mode && devlist[MINOR(dev->devt)].mode)
867 *mode = devlist[MINOR(dev->devt)].mode;
868 return NULL;
869}
870
871static struct class *mem_class;
872
873static int __init chr_dev_init(void)
874{
875 int minor;
876 int err;
877
878 err = bdi_init(&zero_bdi);
879 if (err)
880 return err;
881
882 if (register_chrdev(MEM_MAJOR, "mem", &memory_fops))
883 printk("unable to get major %d for memory devs\n", MEM_MAJOR);
884
885 mem_class = class_create(THIS_MODULE, "mem");
886 if (IS_ERR(mem_class))
887 return PTR_ERR(mem_class);
888
889 mem_class->devnode = mem_devnode;
890 for (minor = 1; minor < ARRAY_SIZE(devlist); minor++) {
891 if (!devlist[minor].name)
892 continue;
893
894 /*
895 * Create /dev/port?
896 */
897 if ((minor == DEVPORT_MINOR) && !arch_has_dev_port())
898 continue;
899
900 device_create(mem_class, NULL, MKDEV(MEM_MAJOR, minor),
901 NULL, devlist[minor].name);
902 }
903
904 return tty_init();
905}
906
907fs_initcall(chr_dev_init);