blob: 4ac70ec697f07f7b6b15c52d231709b7dd64577b [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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 mmaping support, Feb 2000, Kanoj Sarcar <kanoj@sgi.com>
9 */
10
Linus Torvalds1da177e2005-04-16 15:20:36 -070011#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/smp_lock.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include <linux/ptrace.h>
23#include <linux/device.h>
Vivek Goyal50b1fdb2005-06-25 14:58:23 -070024#include <linux/highmem.h>
25#include <linux/crash_dump.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include <linux/backing-dev.h>
Vivek Goyal315c2152005-06-25 14:58:24 -070027#include <linux/bootmem.h>
Jens Axboe1ebd32f2006-04-26 14:40:08 +020028#include <linux/pipe_fs_i.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070029
30#include <asm/uaccess.h>
31#include <asm/io.h>
32
33#ifdef CONFIG_IA64
34# include <linux/efi.h>
35#endif
36
Linus Torvalds1da177e2005-04-16 15:20:36 -070037/*
38 * Architectures vary in how they handle caching for addresses
39 * outside of main memory.
40 *
41 */
42static inline int uncached_access(struct file *file, unsigned long addr)
43{
44#if defined(__i386__)
45 /*
46 * On the PPro and successors, the MTRRs are used to set
47 * memory types for physical addresses outside main memory,
48 * so blindly setting PCD or PWT on those pages is wrong.
49 * For Pentiums and earlier, the surround logic should disable
50 * caching for the high addresses through the KEN pin, but
51 * we maintain the tradition of paranoia in this code.
52 */
53 if (file->f_flags & O_SYNC)
54 return 1;
55 return !( test_bit(X86_FEATURE_MTRR, boot_cpu_data.x86_capability) ||
56 test_bit(X86_FEATURE_K6_MTRR, boot_cpu_data.x86_capability) ||
57 test_bit(X86_FEATURE_CYRIX_ARR, boot_cpu_data.x86_capability) ||
58 test_bit(X86_FEATURE_CENTAUR_MCR, boot_cpu_data.x86_capability) )
59 && addr >= __pa(high_memory);
60#elif defined(__x86_64__)
61 /*
62 * This is broken because it can generate memory type aliases,
63 * which can cause cache corruptions
64 * But it is only available for root and we have to be bug-to-bug
65 * compatible with i386.
66 */
67 if (file->f_flags & O_SYNC)
68 return 1;
69 /* same behaviour as i386. PAT always set to cached and MTRRs control the
70 caching behaviour.
71 Hopefully a full PAT implementation will fix that soon. */
72 return 0;
73#elif defined(CONFIG_IA64)
74 /*
75 * On ia64, we ignore O_SYNC because we cannot tolerate memory attribute aliases.
76 */
77 return !(efi_mem_attributes(addr) & EFI_MEMORY_WB);
78#else
79 /*
80 * Accessing memory above the top the kernel knows about or through a file pointer
81 * that was marked O_SYNC will be done non-cached.
82 */
83 if (file->f_flags & O_SYNC)
84 return 1;
85 return addr >= __pa(high_memory);
86#endif
87}
88
89#ifndef ARCH_HAS_VALID_PHYS_ADDR_RANGE
Bjorn Helgaas136939a2006-03-26 01:37:05 -080090static inline int valid_phys_addr_range(unsigned long addr, size_t count)
Linus Torvalds1da177e2005-04-16 15:20:36 -070091{
Bjorn Helgaas136939a2006-03-26 01:37:05 -080092 if (addr + count > __pa(high_memory))
Linus Torvalds1da177e2005-04-16 15:20:36 -070093 return 0;
94
Linus Torvalds1da177e2005-04-16 15:20:36 -070095 return 1;
96}
Bjorn Helgaas80851ef2006-01-08 01:04:13 -080097
Lennert Buytenhek06c67be2006-07-10 04:45:27 -070098static inline int valid_mmap_phys_addr_range(unsigned long pfn, size_t size)
Bjorn Helgaas80851ef2006-01-08 01:04:13 -080099{
100 return 1;
101}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102#endif
103
104/*
105 * This funcion reads the *physical* memory. The f_pos points directly to the
106 * memory location.
107 */
108static ssize_t read_mem(struct file * file, char __user * buf,
109 size_t count, loff_t *ppos)
110{
111 unsigned long p = *ppos;
112 ssize_t read, sz;
113 char *ptr;
114
Bjorn Helgaas136939a2006-03-26 01:37:05 -0800115 if (!valid_phys_addr_range(p, count))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116 return -EFAULT;
117 read = 0;
118#ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
119 /* we don't have page 0 mapped on sparc and m68k.. */
120 if (p < PAGE_SIZE) {
121 sz = PAGE_SIZE - p;
122 if (sz > count)
123 sz = count;
124 if (sz > 0) {
125 if (clear_user(buf, sz))
126 return -EFAULT;
127 buf += sz;
128 p += sz;
129 count -= sz;
130 read += sz;
131 }
132 }
133#endif
134
135 while (count > 0) {
136 /*
137 * Handle first page in case it's not aligned
138 */
139 if (-p & (PAGE_SIZE - 1))
140 sz = -p & (PAGE_SIZE - 1);
141 else
142 sz = PAGE_SIZE;
143
144 sz = min_t(unsigned long, sz, count);
145
146 /*
147 * On ia64 if a page has been mapped somewhere as
148 * uncached, then it must also be accessed uncached
149 * by the kernel or data corruption may occur
150 */
151 ptr = xlate_dev_mem_ptr(p);
152
153 if (copy_to_user(buf, ptr, sz))
154 return -EFAULT;
155 buf += sz;
156 p += sz;
157 count -= sz;
158 read += sz;
159 }
160
161 *ppos += read;
162 return read;
163}
164
165static ssize_t write_mem(struct file * file, const char __user * buf,
166 size_t count, loff_t *ppos)
167{
168 unsigned long p = *ppos;
169 ssize_t written, sz;
170 unsigned long copied;
171 void *ptr;
172
Bjorn Helgaas136939a2006-03-26 01:37:05 -0800173 if (!valid_phys_addr_range(p, count))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174 return -EFAULT;
175
176 written = 0;
177
178#ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
179 /* we don't have page 0 mapped on sparc and m68k.. */
180 if (p < PAGE_SIZE) {
181 unsigned long sz = PAGE_SIZE - p;
182 if (sz > count)
183 sz = count;
184 /* Hmm. Do something? */
185 buf += sz;
186 p += sz;
187 count -= sz;
188 written += sz;
189 }
190#endif
191
192 while (count > 0) {
193 /*
194 * Handle first page in case it's not aligned
195 */
196 if (-p & (PAGE_SIZE - 1))
197 sz = -p & (PAGE_SIZE - 1);
198 else
199 sz = PAGE_SIZE;
200
201 sz = min_t(unsigned long, sz, count);
202
203 /*
204 * On ia64 if a page has been mapped somewhere as
205 * uncached, then it must also be accessed uncached
206 * by the kernel or data corruption may occur
207 */
208 ptr = xlate_dev_mem_ptr(p);
209
210 copied = copy_from_user(ptr, buf, sz);
211 if (copied) {
Jan Beulichc654d602006-03-25 03:07:31 -0800212 written += sz - copied;
213 if (written)
214 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215 return -EFAULT;
216 }
217 buf += sz;
218 p += sz;
219 count -= sz;
220 written += sz;
221 }
222
223 *ppos += written;
224 return written;
225}
226
Bjorn Helgaas44ac8412006-01-08 01:04:10 -0800227#ifndef __HAVE_PHYS_MEM_ACCESS_PROT
228static pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
229 unsigned long size, pgprot_t vma_prot)
230{
231#ifdef pgprot_noncached
232 unsigned long offset = pfn << PAGE_SHIFT;
233
234 if (uncached_access(file, offset))
235 return pgprot_noncached(vma_prot);
236#endif
237 return vma_prot;
238}
239#endif
240
David Howells5da61852006-09-27 01:50:16 -0700241#ifndef CONFIG_MMU
242static unsigned long get_unmapped_area_mem(struct file *file,
243 unsigned long addr,
244 unsigned long len,
245 unsigned long pgoff,
246 unsigned long flags)
247{
248 if (!valid_mmap_phys_addr_range(pgoff, len))
249 return (unsigned long) -EINVAL;
250 return pgoff;
251}
252
253/* can't do an in-place private mapping if there's no MMU */
254static inline int private_mapping_ok(struct vm_area_struct *vma)
255{
256 return vma->vm_flags & VM_MAYSHARE;
257}
258#else
259#define get_unmapped_area_mem NULL
260
261static inline int private_mapping_ok(struct vm_area_struct *vma)
262{
263 return 1;
264}
265#endif
266
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267static int mmap_mem(struct file * file, struct vm_area_struct * vma)
268{
Bjorn Helgaas80851ef2006-01-08 01:04:13 -0800269 size_t size = vma->vm_end - vma->vm_start;
270
Lennert Buytenhek06c67be2006-07-10 04:45:27 -0700271 if (!valid_mmap_phys_addr_range(vma->vm_pgoff, size))
Bjorn Helgaas80851ef2006-01-08 01:04:13 -0800272 return -EINVAL;
273
David Howells5da61852006-09-27 01:50:16 -0700274 if (!private_mapping_ok(vma))
275 return -ENOSYS;
276
Roland Dreier8b150472005-10-28 17:46:18 -0700277 vma->vm_page_prot = phys_mem_access_prot(file, vma->vm_pgoff,
Bjorn Helgaas80851ef2006-01-08 01:04:13 -0800278 size,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279 vma->vm_page_prot);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280
281 /* Remap-pfn-range will mark the range VM_IO and VM_RESERVED */
282 if (remap_pfn_range(vma,
283 vma->vm_start,
284 vma->vm_pgoff,
Bjorn Helgaas80851ef2006-01-08 01:04:13 -0800285 size,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286 vma->vm_page_prot))
287 return -EAGAIN;
288 return 0;
289}
290
291static int mmap_kmem(struct file * file, struct vm_area_struct * vma)
292{
Linus Torvalds4bb82552005-08-13 14:22:59 -0700293 unsigned long pfn;
294
295 /* Turn a kernel-virtual address into a physical page frame */
296 pfn = __pa((u64)vma->vm_pgoff << PAGE_SHIFT) >> PAGE_SHIFT;
297
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298 /*
299 * RED-PEN: on some architectures there is more mapped memory
300 * than available in mem_map which pfn_valid checks
301 * for. Perhaps should add a new macro here.
302 *
303 * RED-PEN: vmalloc is not supported right now.
304 */
Linus Torvalds4bb82552005-08-13 14:22:59 -0700305 if (!pfn_valid(pfn))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306 return -EIO;
Linus Torvalds4bb82552005-08-13 14:22:59 -0700307
308 vma->vm_pgoff = pfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309 return mmap_mem(file, vma);
310}
311
Vivek Goyal50b1fdb2005-06-25 14:58:23 -0700312#ifdef CONFIG_CRASH_DUMP
313/*
314 * Read memory corresponding to the old kernel.
Vivek Goyal50b1fdb2005-06-25 14:58:23 -0700315 */
Vivek Goyal315c2152005-06-25 14:58:24 -0700316static ssize_t read_oldmem(struct file *file, char __user *buf,
Vivek Goyal50b1fdb2005-06-25 14:58:23 -0700317 size_t count, loff_t *ppos)
318{
Vivek Goyal315c2152005-06-25 14:58:24 -0700319 unsigned long pfn, offset;
320 size_t read = 0, csize;
321 int rc = 0;
Vivek Goyal50b1fdb2005-06-25 14:58:23 -0700322
Maneesh Soni72414d32005-06-25 14:58:28 -0700323 while (count) {
Vivek Goyal50b1fdb2005-06-25 14:58:23 -0700324 pfn = *ppos / PAGE_SIZE;
Vivek Goyal315c2152005-06-25 14:58:24 -0700325 if (pfn > saved_max_pfn)
326 return read;
Vivek Goyal50b1fdb2005-06-25 14:58:23 -0700327
Vivek Goyal315c2152005-06-25 14:58:24 -0700328 offset = (unsigned long)(*ppos % PAGE_SIZE);
329 if (count > PAGE_SIZE - offset)
330 csize = PAGE_SIZE - offset;
331 else
332 csize = count;
Vivek Goyal50b1fdb2005-06-25 14:58:23 -0700333
Vivek Goyal315c2152005-06-25 14:58:24 -0700334 rc = copy_oldmem_page(pfn, buf, csize, offset, 1);
335 if (rc < 0)
336 return rc;
Vivek Goyal50b1fdb2005-06-25 14:58:23 -0700337 buf += csize;
338 *ppos += csize;
339 read += csize;
340 count -= csize;
341 }
Vivek Goyal50b1fdb2005-06-25 14:58:23 -0700342 return read;
343}
344#endif
345
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346extern long vread(char *buf, char *addr, unsigned long count);
347extern long vwrite(char *buf, char *addr, unsigned long count);
348
349/*
350 * This function reads the *virtual* memory as seen by the kernel.
351 */
352static ssize_t read_kmem(struct file *file, char __user *buf,
353 size_t count, loff_t *ppos)
354{
355 unsigned long p = *ppos;
356 ssize_t low_count, read, sz;
357 char * kbuf; /* k-addr because vread() takes vmlist_lock rwlock */
358
359 read = 0;
360 if (p < (unsigned long) high_memory) {
361 low_count = count;
362 if (count > (unsigned long) high_memory - p)
363 low_count = (unsigned long) high_memory - p;
364
365#ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
366 /* we don't have page 0 mapped on sparc and m68k.. */
367 if (p < PAGE_SIZE && low_count > 0) {
368 size_t tmp = PAGE_SIZE - p;
369 if (tmp > low_count) tmp = low_count;
370 if (clear_user(buf, tmp))
371 return -EFAULT;
372 buf += tmp;
373 p += tmp;
374 read += tmp;
375 low_count -= tmp;
376 count -= tmp;
377 }
378#endif
379 while (low_count > 0) {
380 /*
381 * Handle first page in case it's not aligned
382 */
383 if (-p & (PAGE_SIZE - 1))
384 sz = -p & (PAGE_SIZE - 1);
385 else
386 sz = PAGE_SIZE;
387
388 sz = min_t(unsigned long, sz, low_count);
389
390 /*
391 * On ia64 if a page has been mapped somewhere as
392 * uncached, then it must also be accessed uncached
393 * by the kernel or data corruption may occur
394 */
395 kbuf = xlate_dev_kmem_ptr((char *)p);
396
397 if (copy_to_user(buf, kbuf, sz))
398 return -EFAULT;
399 buf += sz;
400 p += sz;
401 read += sz;
402 low_count -= sz;
403 count -= sz;
404 }
405 }
406
407 if (count > 0) {
408 kbuf = (char *)__get_free_page(GFP_KERNEL);
409 if (!kbuf)
410 return -ENOMEM;
411 while (count > 0) {
412 int len = count;
413
414 if (len > PAGE_SIZE)
415 len = PAGE_SIZE;
416 len = vread(kbuf, (char *)p, len);
417 if (!len)
418 break;
419 if (copy_to_user(buf, kbuf, len)) {
420 free_page((unsigned long)kbuf);
421 return -EFAULT;
422 }
423 count -= len;
424 buf += len;
425 read += len;
426 p += len;
427 }
428 free_page((unsigned long)kbuf);
429 }
430 *ppos = p;
431 return read;
432}
433
434
435static inline ssize_t
436do_write_kmem(void *p, unsigned long realp, const char __user * buf,
437 size_t count, loff_t *ppos)
438{
439 ssize_t written, sz;
440 unsigned long copied;
441
442 written = 0;
443#ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
444 /* we don't have page 0 mapped on sparc and m68k.. */
445 if (realp < PAGE_SIZE) {
446 unsigned long sz = PAGE_SIZE - realp;
447 if (sz > count)
448 sz = count;
449 /* Hmm. Do something? */
450 buf += sz;
451 p += sz;
452 realp += sz;
453 count -= sz;
454 written += sz;
455 }
456#endif
457
458 while (count > 0) {
459 char *ptr;
460 /*
461 * Handle first page in case it's not aligned
462 */
463 if (-realp & (PAGE_SIZE - 1))
464 sz = -realp & (PAGE_SIZE - 1);
465 else
466 sz = PAGE_SIZE;
467
468 sz = min_t(unsigned long, sz, 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 ptr = xlate_dev_kmem_ptr(p);
476
477 copied = copy_from_user(ptr, buf, sz);
478 if (copied) {
Jan Beulichc654d602006-03-25 03:07:31 -0800479 written += sz - copied;
480 if (written)
481 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482 return -EFAULT;
483 }
484 buf += sz;
485 p += sz;
486 realp += sz;
487 count -= sz;
488 written += sz;
489 }
490
491 *ppos += written;
492 return written;
493}
494
495
496/*
497 * This function writes to the *virtual* memory as seen by the kernel.
498 */
499static ssize_t write_kmem(struct file * file, const char __user * buf,
500 size_t count, loff_t *ppos)
501{
502 unsigned long p = *ppos;
503 ssize_t wrote = 0;
504 ssize_t virtr = 0;
505 ssize_t written;
506 char * kbuf; /* k-addr because vwrite() takes vmlist_lock rwlock */
507
508 if (p < (unsigned long) high_memory) {
509
510 wrote = count;
511 if (count > (unsigned long) high_memory - p)
512 wrote = (unsigned long) high_memory - p;
513
514 written = do_write_kmem((void*)p, p, buf, wrote, ppos);
515 if (written != wrote)
516 return written;
517 wrote = written;
518 p += wrote;
519 buf += wrote;
520 count -= wrote;
521 }
522
523 if (count > 0) {
524 kbuf = (char *)__get_free_page(GFP_KERNEL);
525 if (!kbuf)
526 return wrote ? wrote : -ENOMEM;
527 while (count > 0) {
528 int len = count;
529
530 if (len > PAGE_SIZE)
531 len = PAGE_SIZE;
532 if (len) {
533 written = copy_from_user(kbuf, buf, len);
534 if (written) {
Jan Beulichc654d602006-03-25 03:07:31 -0800535 if (wrote + virtr)
536 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537 free_page((unsigned long)kbuf);
Jan Beulichc654d602006-03-25 03:07:31 -0800538 return -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539 }
540 }
541 len = vwrite(kbuf, (char *)p, len);
542 count -= len;
543 buf += len;
544 virtr += len;
545 p += len;
546 }
547 free_page((unsigned long)kbuf);
548 }
549
550 *ppos = p;
551 return virtr + wrote;
552}
553
Stephen Rothwellee2cdec2006-01-12 13:54:20 +1100554#if defined(CONFIG_ISA) || !defined(__mc68000__)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555static ssize_t read_port(struct file * file, char __user * buf,
556 size_t count, loff_t *ppos)
557{
558 unsigned long i = *ppos;
559 char __user *tmp = buf;
560
561 if (!access_ok(VERIFY_WRITE, buf, count))
562 return -EFAULT;
563 while (count-- > 0 && i < 65536) {
564 if (__put_user(inb(i),tmp) < 0)
565 return -EFAULT;
566 i++;
567 tmp++;
568 }
569 *ppos = i;
570 return tmp-buf;
571}
572
573static ssize_t write_port(struct file * file, const char __user * buf,
574 size_t count, loff_t *ppos)
575{
576 unsigned long i = *ppos;
577 const char __user * tmp = buf;
578
579 if (!access_ok(VERIFY_READ,buf,count))
580 return -EFAULT;
581 while (count-- > 0 && i < 65536) {
582 char c;
Jan Beulichc654d602006-03-25 03:07:31 -0800583 if (__get_user(c, tmp)) {
584 if (tmp > buf)
585 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586 return -EFAULT;
Jan Beulichc654d602006-03-25 03:07:31 -0800587 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588 outb(c,i);
589 i++;
590 tmp++;
591 }
592 *ppos = i;
593 return tmp-buf;
594}
595#endif
596
597static ssize_t read_null(struct file * file, char __user * buf,
598 size_t count, loff_t *ppos)
599{
600 return 0;
601}
602
603static ssize_t write_null(struct file * file, const char __user * buf,
604 size_t count, loff_t *ppos)
605{
606 return count;
607}
608
Jens Axboe1ebd32f2006-04-26 14:40:08 +0200609static int pipe_to_null(struct pipe_inode_info *info, struct pipe_buffer *buf,
610 struct splice_desc *sd)
611{
612 return sd->len;
613}
614
615static ssize_t splice_write_null(struct pipe_inode_info *pipe,struct file *out,
616 loff_t *ppos, size_t len, unsigned int flags)
617{
618 return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_null);
619}
620
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621#ifdef CONFIG_MMU
622/*
623 * For fun, we are using the MMU for this.
624 */
625static inline size_t read_zero_pagealigned(char __user * buf, size_t size)
626{
627 struct mm_struct *mm;
628 struct vm_area_struct * vma;
629 unsigned long addr=(unsigned long)buf;
630
631 mm = current->mm;
632 /* Oops, this was forgotten before. -ben */
633 down_read(&mm->mmap_sem);
634
635 /* For private mappings, just map in zero pages. */
636 for (vma = find_vma(mm, addr); vma; vma = vma->vm_next) {
637 unsigned long count;
638
639 if (vma->vm_start > addr || (vma->vm_flags & VM_WRITE) == 0)
640 goto out_up;
Linus Torvalds6aab3412005-11-28 14:34:23 -0800641 if (vma->vm_flags & (VM_SHARED | VM_HUGETLB))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642 break;
643 count = vma->vm_end - addr;
644 if (count > size)
645 count = size;
646
647 zap_page_range(vma, addr, count, NULL);
648 zeromap_page_range(vma, addr, count, PAGE_COPY);
649
650 size -= count;
651 buf += count;
652 addr += count;
653 if (size == 0)
654 goto out_up;
655 }
656
657 up_read(&mm->mmap_sem);
658
659 /* The shared case is hard. Let's do the conventional zeroing. */
660 do {
661 unsigned long unwritten = clear_user(buf, PAGE_SIZE);
662 if (unwritten)
663 return size + unwritten - PAGE_SIZE;
664 cond_resched();
665 buf += PAGE_SIZE;
666 size -= PAGE_SIZE;
667 } while (size);
668
669 return size;
670out_up:
671 up_read(&mm->mmap_sem);
672 return size;
673}
674
675static ssize_t read_zero(struct file * file, char __user * buf,
676 size_t count, loff_t *ppos)
677{
678 unsigned long left, unwritten, written = 0;
679
680 if (!count)
681 return 0;
682
683 if (!access_ok(VERIFY_WRITE, buf, count))
684 return -EFAULT;
685
686 left = count;
687
688 /* do we want to be clever? Arbitrary cut-off */
689 if (count >= PAGE_SIZE*4) {
690 unsigned long partial;
691
692 /* How much left of the page? */
693 partial = (PAGE_SIZE-1) & -(unsigned long) buf;
694 unwritten = clear_user(buf, partial);
695 written = partial - unwritten;
696 if (unwritten)
697 goto out;
698 left -= partial;
699 buf += partial;
700 unwritten = read_zero_pagealigned(buf, left & PAGE_MASK);
701 written += (left & PAGE_MASK) - unwritten;
702 if (unwritten)
703 goto out;
704 buf += left & PAGE_MASK;
705 left &= ~PAGE_MASK;
706 }
707 unwritten = clear_user(buf, left);
708 written += left - unwritten;
709out:
710 return written ? written : -EFAULT;
711}
712
713static int mmap_zero(struct file * file, struct vm_area_struct * vma)
714{
715 if (vma->vm_flags & VM_SHARED)
716 return shmem_zero_setup(vma);
717 if (zeromap_page_range(vma, vma->vm_start, vma->vm_end - vma->vm_start, vma->vm_page_prot))
718 return -EAGAIN;
719 return 0;
720}
721#else /* CONFIG_MMU */
722static ssize_t read_zero(struct file * file, char * buf,
723 size_t count, loff_t *ppos)
724{
725 size_t todo = count;
726
727 while (todo) {
728 size_t chunk = todo;
729
730 if (chunk > 4096)
731 chunk = 4096; /* Just for latency reasons */
732 if (clear_user(buf, chunk))
733 return -EFAULT;
734 buf += chunk;
735 todo -= chunk;
736 cond_resched();
737 }
738 return count;
739}
740
741static int mmap_zero(struct file * file, struct vm_area_struct * vma)
742{
743 return -ENOSYS;
744}
745#endif /* CONFIG_MMU */
746
747static ssize_t write_full(struct file * file, const char __user * buf,
748 size_t count, loff_t *ppos)
749{
750 return -ENOSPC;
751}
752
753/*
754 * Special lseek() function for /dev/null and /dev/zero. Most notably, you
755 * can fopen() both devices with "a" now. This was previously impossible.
756 * -- SRB.
757 */
758
759static loff_t null_lseek(struct file * file, loff_t offset, int orig)
760{
761 return file->f_pos = 0;
762}
763
764/*
765 * The memory devices use the full 32/64 bits of the offset, and so we cannot
766 * check against negative addresses: they are ok. The return value is weird,
767 * though, in that case (0).
768 *
769 * also note that seeking relative to the "end of file" isn't supported:
770 * it has no meaning, so it returns -EINVAL.
771 */
772static loff_t memory_lseek(struct file * file, loff_t offset, int orig)
773{
774 loff_t ret;
775
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800776 mutex_lock(&file->f_dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777 switch (orig) {
778 case 0:
779 file->f_pos = offset;
780 ret = file->f_pos;
781 force_successful_syscall_return();
782 break;
783 case 1:
784 file->f_pos += offset;
785 ret = file->f_pos;
786 force_successful_syscall_return();
787 break;
788 default:
789 ret = -EINVAL;
790 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800791 mutex_unlock(&file->f_dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792 return ret;
793}
794
795static int open_port(struct inode * inode, struct file * filp)
796{
797 return capable(CAP_SYS_RAWIO) ? 0 : -EPERM;
798}
799
800#define zero_lseek null_lseek
801#define full_lseek null_lseek
802#define write_zero write_null
803#define read_full read_zero
804#define open_mem open_port
805#define open_kmem open_mem
Vivek Goyal50b1fdb2005-06-25 14:58:23 -0700806#define open_oldmem open_mem
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807
Arjan van de Ven62322d22006-07-03 00:24:21 -0700808static const struct file_operations mem_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809 .llseek = memory_lseek,
810 .read = read_mem,
811 .write = write_mem,
812 .mmap = mmap_mem,
813 .open = open_mem,
David Howells5da61852006-09-27 01:50:16 -0700814 .get_unmapped_area = get_unmapped_area_mem,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815};
816
Arjan van de Ven62322d22006-07-03 00:24:21 -0700817static const struct file_operations kmem_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818 .llseek = memory_lseek,
819 .read = read_kmem,
820 .write = write_kmem,
821 .mmap = mmap_kmem,
822 .open = open_kmem,
David Howells5da61852006-09-27 01:50:16 -0700823 .get_unmapped_area = get_unmapped_area_mem,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824};
825
Arjan van de Ven62322d22006-07-03 00:24:21 -0700826static const struct file_operations null_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827 .llseek = null_lseek,
828 .read = read_null,
829 .write = write_null,
Jens Axboe1ebd32f2006-04-26 14:40:08 +0200830 .splice_write = splice_write_null,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831};
832
Stephen Rothwellee2cdec2006-01-12 13:54:20 +1100833#if defined(CONFIG_ISA) || !defined(__mc68000__)
Arjan van de Ven62322d22006-07-03 00:24:21 -0700834static const struct file_operations port_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835 .llseek = memory_lseek,
836 .read = read_port,
837 .write = write_port,
838 .open = open_port,
839};
840#endif
841
Arjan van de Ven62322d22006-07-03 00:24:21 -0700842static const struct file_operations zero_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843 .llseek = zero_lseek,
844 .read = read_zero,
845 .write = write_zero,
846 .mmap = mmap_zero,
847};
848
David Howells5da61852006-09-27 01:50:16 -0700849/*
850 * capabilities for /dev/zero
851 * - permits private mappings, "copies" are taken of the source of zeros
852 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853static struct backing_dev_info zero_bdi = {
854 .capabilities = BDI_CAP_MAP_COPY,
855};
856
Arjan van de Ven62322d22006-07-03 00:24:21 -0700857static const struct file_operations full_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858 .llseek = full_lseek,
859 .read = read_full,
860 .write = write_full,
861};
862
Vivek Goyal50b1fdb2005-06-25 14:58:23 -0700863#ifdef CONFIG_CRASH_DUMP
Arjan van de Ven62322d22006-07-03 00:24:21 -0700864static const struct file_operations oldmem_fops = {
Vivek Goyal50b1fdb2005-06-25 14:58:23 -0700865 .read = read_oldmem,
866 .open = open_oldmem,
867};
868#endif
869
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870static ssize_t kmsg_write(struct file * file, const char __user * buf,
871 size_t count, loff_t *ppos)
872{
873 char *tmp;
Guillaume Chazaraincd140a52006-01-08 01:02:43 -0800874 ssize_t ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875
876 tmp = kmalloc(count + 1, GFP_KERNEL);
877 if (tmp == NULL)
878 return -ENOMEM;
879 ret = -EFAULT;
880 if (!copy_from_user(tmp, buf, count)) {
881 tmp[count] = 0;
882 ret = printk("%s", tmp);
Guillaume Chazaraincd140a52006-01-08 01:02:43 -0800883 if (ret > count)
884 /* printk can add a prefix */
885 ret = count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886 }
887 kfree(tmp);
888 return ret;
889}
890
Arjan van de Ven62322d22006-07-03 00:24:21 -0700891static const struct file_operations kmsg_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892 .write = kmsg_write,
893};
894
895static int memory_open(struct inode * inode, struct file * filp)
896{
897 switch (iminor(inode)) {
898 case 1:
899 filp->f_op = &mem_fops;
David Howells5da61852006-09-27 01:50:16 -0700900 filp->f_mapping->backing_dev_info =
901 &directly_mappable_cdev_bdi;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902 break;
903 case 2:
904 filp->f_op = &kmem_fops;
David Howells5da61852006-09-27 01:50:16 -0700905 filp->f_mapping->backing_dev_info =
906 &directly_mappable_cdev_bdi;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907 break;
908 case 3:
909 filp->f_op = &null_fops;
910 break;
Stephen Rothwellee2cdec2006-01-12 13:54:20 +1100911#if defined(CONFIG_ISA) || !defined(__mc68000__)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912 case 4:
913 filp->f_op = &port_fops;
914 break;
915#endif
916 case 5:
917 filp->f_mapping->backing_dev_info = &zero_bdi;
918 filp->f_op = &zero_fops;
919 break;
920 case 7:
921 filp->f_op = &full_fops;
922 break;
923 case 8:
924 filp->f_op = &random_fops;
925 break;
926 case 9:
927 filp->f_op = &urandom_fops;
928 break;
929 case 11:
930 filp->f_op = &kmsg_fops;
931 break;
Vivek Goyal50b1fdb2005-06-25 14:58:23 -0700932#ifdef CONFIG_CRASH_DUMP
933 case 12:
934 filp->f_op = &oldmem_fops;
935 break;
936#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937 default:
938 return -ENXIO;
939 }
940 if (filp->f_op && filp->f_op->open)
941 return filp->f_op->open(inode,filp);
942 return 0;
943}
944
Arjan van de Ven62322d22006-07-03 00:24:21 -0700945static const struct file_operations memory_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946 .open = memory_open, /* just a selector for the real open */
947};
948
949static const struct {
950 unsigned int minor;
951 char *name;
952 umode_t mode;
Arjan van de Ven99ac48f2006-03-28 01:56:41 -0800953 const struct file_operations *fops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954} devlist[] = { /* list of minor devices */
955 {1, "mem", S_IRUSR | S_IWUSR | S_IRGRP, &mem_fops},
956 {2, "kmem", S_IRUSR | S_IWUSR | S_IRGRP, &kmem_fops},
957 {3, "null", S_IRUGO | S_IWUGO, &null_fops},
Stephen Rothwellee2cdec2006-01-12 13:54:20 +1100958#if defined(CONFIG_ISA) || !defined(__mc68000__)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959 {4, "port", S_IRUSR | S_IWUSR | S_IRGRP, &port_fops},
960#endif
961 {5, "zero", S_IRUGO | S_IWUGO, &zero_fops},
962 {7, "full", S_IRUGO | S_IWUGO, &full_fops},
963 {8, "random", S_IRUGO | S_IWUSR, &random_fops},
964 {9, "urandom", S_IRUGO | S_IWUSR, &urandom_fops},
965 {11,"kmsg", S_IRUGO | S_IWUSR, &kmsg_fops},
Vivek Goyal50b1fdb2005-06-25 14:58:23 -0700966#ifdef CONFIG_CRASH_DUMP
967 {12,"oldmem", S_IRUSR | S_IWUSR | S_IRGRP, &oldmem_fops},
968#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969};
970
gregkh@suse.deca8eca62005-03-23 09:53:09 -0800971static struct class *mem_class;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972
973static int __init chr_dev_init(void)
974{
975 int i;
976
977 if (register_chrdev(MEM_MAJOR,"mem",&memory_fops))
978 printk("unable to get major %d for memory devs\n", MEM_MAJOR);
979
gregkh@suse.deca8eca62005-03-23 09:53:09 -0800980 mem_class = class_create(THIS_MODULE, "mem");
Greg Kroah-Hartman7c69ef72005-06-20 21:15:16 -0700981 for (i = 0; i < ARRAY_SIZE(devlist); i++)
Greg Kroah-Hartman53f46542005-10-27 22:25:43 -0700982 class_device_create(mem_class, NULL,
983 MKDEV(MEM_MAJOR, devlist[i].minor),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984 NULL, devlist[i].name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985
986 return 0;
987}
988
989fs_initcall(chr_dev_init);