blob: 7e082bceed570b1e42673337d5bfcb3e641f094b [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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070021#include <linux/ptrace.h>
22#include <linux/device.h>
Vivek Goyal50b1fdb2005-06-25 14:58:23 -070023#include <linux/highmem.h>
24#include <linux/crash_dump.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/backing-dev.h>
Vivek Goyal315c2152005-06-25 14:58:24 -070026#include <linux/bootmem.h>
Jens Axboe1ebd32f2006-04-26 14:40:08 +020027#include <linux/pipe_fs_i.h>
Linus Torvaldsb8a3ad52006-10-13 08:42:10 -070028#include <linux/pfn.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);
Ralf Baechle24e9d0b2007-07-10 17:32:56 +010078#elif defined(CONFIG_MIPS)
79 {
80 extern int __uncached_access(struct file *file,
81 unsigned long addr);
82
83 return __uncached_access(file, addr);
84 }
Linus Torvalds1da177e2005-04-16 15:20:36 -070085#else
86 /*
87 * Accessing memory above the top the kernel knows about or through a file pointer
88 * that was marked O_SYNC will be done non-cached.
89 */
90 if (file->f_flags & O_SYNC)
91 return 1;
92 return addr >= __pa(high_memory);
93#endif
94}
95
96#ifndef ARCH_HAS_VALID_PHYS_ADDR_RANGE
Bjorn Helgaas136939a2006-03-26 01:37:05 -080097static inline int valid_phys_addr_range(unsigned long addr, size_t count)
Linus Torvalds1da177e2005-04-16 15:20:36 -070098{
Bjorn Helgaas136939a2006-03-26 01:37:05 -080099 if (addr + count > __pa(high_memory))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100 return 0;
101
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102 return 1;
103}
Bjorn Helgaas80851ef2006-01-08 01:04:13 -0800104
Lennert Buytenhek06c67be2006-07-10 04:45:27 -0700105static inline int valid_mmap_phys_addr_range(unsigned long pfn, size_t size)
Bjorn Helgaas80851ef2006-01-08 01:04:13 -0800106{
107 return 1;
108}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109#endif
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 unsigned long p = *ppos;
119 ssize_t read, sz;
120 char *ptr;
121
Bjorn Helgaas136939a2006-03-26 01:37:05 -0800122 if (!valid_phys_addr_range(p, count))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123 return -EFAULT;
124 read = 0;
125#ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
126 /* we don't have page 0 mapped on sparc and m68k.. */
127 if (p < PAGE_SIZE) {
128 sz = PAGE_SIZE - p;
129 if (sz > count)
130 sz = count;
131 if (sz > 0) {
132 if (clear_user(buf, sz))
133 return -EFAULT;
134 buf += sz;
135 p += sz;
136 count -= sz;
137 read += sz;
138 }
139 }
140#endif
141
142 while (count > 0) {
143 /*
144 * Handle first page in case it's not aligned
145 */
146 if (-p & (PAGE_SIZE - 1))
147 sz = -p & (PAGE_SIZE - 1);
148 else
149 sz = PAGE_SIZE;
150
151 sz = min_t(unsigned long, sz, count);
152
153 /*
154 * On ia64 if a page has been mapped somewhere as
155 * uncached, then it must also be accessed uncached
156 * by the kernel or data corruption may occur
157 */
158 ptr = xlate_dev_mem_ptr(p);
159
160 if (copy_to_user(buf, ptr, sz))
161 return -EFAULT;
162 buf += sz;
163 p += sz;
164 count -= sz;
165 read += sz;
166 }
167
168 *ppos += read;
169 return read;
170}
171
172static ssize_t write_mem(struct file * file, const char __user * buf,
173 size_t count, loff_t *ppos)
174{
175 unsigned long p = *ppos;
176 ssize_t written, sz;
177 unsigned long copied;
178 void *ptr;
179
Bjorn Helgaas136939a2006-03-26 01:37:05 -0800180 if (!valid_phys_addr_range(p, count))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181 return -EFAULT;
182
183 written = 0;
184
185#ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
186 /* we don't have page 0 mapped on sparc and m68k.. */
187 if (p < PAGE_SIZE) {
188 unsigned long sz = PAGE_SIZE - p;
189 if (sz > count)
190 sz = count;
191 /* Hmm. Do something? */
192 buf += sz;
193 p += sz;
194 count -= sz;
195 written += sz;
196 }
197#endif
198
199 while (count > 0) {
200 /*
201 * Handle first page in case it's not aligned
202 */
203 if (-p & (PAGE_SIZE - 1))
204 sz = -p & (PAGE_SIZE - 1);
205 else
206 sz = PAGE_SIZE;
207
208 sz = min_t(unsigned long, sz, count);
209
210 /*
211 * On ia64 if a page has been mapped somewhere as
212 * uncached, then it must also be accessed uncached
213 * by the kernel or data corruption may occur
214 */
215 ptr = xlate_dev_mem_ptr(p);
216
217 copied = copy_from_user(ptr, buf, sz);
218 if (copied) {
Jan Beulichc654d602006-03-25 03:07:31 -0800219 written += sz - copied;
220 if (written)
221 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222 return -EFAULT;
223 }
224 buf += sz;
225 p += sz;
226 count -= sz;
227 written += sz;
228 }
229
230 *ppos += written;
231 return written;
232}
233
Bjorn Helgaas44ac8412006-01-08 01:04:10 -0800234#ifndef __HAVE_PHYS_MEM_ACCESS_PROT
235static pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
236 unsigned long size, pgprot_t vma_prot)
237{
238#ifdef pgprot_noncached
239 unsigned long offset = pfn << PAGE_SHIFT;
240
241 if (uncached_access(file, offset))
242 return pgprot_noncached(vma_prot);
243#endif
244 return vma_prot;
245}
246#endif
247
David Howells5da61852006-09-27 01:50:16 -0700248#ifndef CONFIG_MMU
249static unsigned long get_unmapped_area_mem(struct file *file,
250 unsigned long addr,
251 unsigned long len,
252 unsigned long pgoff,
253 unsigned long flags)
254{
255 if (!valid_mmap_phys_addr_range(pgoff, len))
256 return (unsigned long) -EINVAL;
Benjamin Herrenschmidt8a932582007-04-16 22:53:16 -0700257 return pgoff << PAGE_SHIFT;
David Howells5da61852006-09-27 01:50:16 -0700258}
259
260/* can't do an in-place private mapping if there's no MMU */
261static inline int private_mapping_ok(struct vm_area_struct *vma)
262{
263 return vma->vm_flags & VM_MAYSHARE;
264}
265#else
266#define get_unmapped_area_mem NULL
267
268static inline int private_mapping_ok(struct vm_area_struct *vma)
269{
270 return 1;
271}
272#endif
273
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274static int mmap_mem(struct file * file, struct vm_area_struct * vma)
275{
Bjorn Helgaas80851ef2006-01-08 01:04:13 -0800276 size_t size = vma->vm_end - vma->vm_start;
277
Lennert Buytenhek06c67be2006-07-10 04:45:27 -0700278 if (!valid_mmap_phys_addr_range(vma->vm_pgoff, size))
Bjorn Helgaas80851ef2006-01-08 01:04:13 -0800279 return -EINVAL;
280
David Howells5da61852006-09-27 01:50:16 -0700281 if (!private_mapping_ok(vma))
282 return -ENOSYS;
283
Roland Dreier8b150472005-10-28 17:46:18 -0700284 vma->vm_page_prot = phys_mem_access_prot(file, vma->vm_pgoff,
Bjorn Helgaas80851ef2006-01-08 01:04:13 -0800285 size,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286 vma->vm_page_prot);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287
288 /* Remap-pfn-range will mark the range VM_IO and VM_RESERVED */
289 if (remap_pfn_range(vma,
290 vma->vm_start,
291 vma->vm_pgoff,
Bjorn Helgaas80851ef2006-01-08 01:04:13 -0800292 size,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293 vma->vm_page_prot))
294 return -EAGAIN;
295 return 0;
296}
297
298static int mmap_kmem(struct file * file, struct vm_area_struct * vma)
299{
Linus Torvalds4bb82552005-08-13 14:22:59 -0700300 unsigned long pfn;
301
Linus Torvalds6d3154c2007-01-22 08:53:24 -0800302 /* Turn a kernel-virtual address into a physical page frame */
303 pfn = __pa((u64)vma->vm_pgoff << PAGE_SHIFT) >> PAGE_SHIFT;
Linus Torvalds4bb82552005-08-13 14:22:59 -0700304
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305 /*
306 * RED-PEN: on some architectures there is more mapped memory
307 * than available in mem_map which pfn_valid checks
308 * for. Perhaps should add a new macro here.
309 *
310 * RED-PEN: vmalloc is not supported right now.
311 */
Linus Torvalds4bb82552005-08-13 14:22:59 -0700312 if (!pfn_valid(pfn))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313 return -EIO;
Linus Torvalds4bb82552005-08-13 14:22:59 -0700314
315 vma->vm_pgoff = pfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316 return mmap_mem(file, vma);
317}
318
Vivek Goyal50b1fdb2005-06-25 14:58:23 -0700319#ifdef CONFIG_CRASH_DUMP
320/*
321 * Read memory corresponding to the old kernel.
Vivek Goyal50b1fdb2005-06-25 14:58:23 -0700322 */
Vivek Goyal315c2152005-06-25 14:58:24 -0700323static ssize_t read_oldmem(struct file *file, char __user *buf,
Vivek Goyal50b1fdb2005-06-25 14:58:23 -0700324 size_t count, loff_t *ppos)
325{
Vivek Goyal315c2152005-06-25 14:58:24 -0700326 unsigned long pfn, offset;
327 size_t read = 0, csize;
328 int rc = 0;
Vivek Goyal50b1fdb2005-06-25 14:58:23 -0700329
Maneesh Soni72414d32005-06-25 14:58:28 -0700330 while (count) {
Vivek Goyal50b1fdb2005-06-25 14:58:23 -0700331 pfn = *ppos / PAGE_SIZE;
Vivek Goyal315c2152005-06-25 14:58:24 -0700332 if (pfn > saved_max_pfn)
333 return read;
Vivek Goyal50b1fdb2005-06-25 14:58:23 -0700334
Vivek Goyal315c2152005-06-25 14:58:24 -0700335 offset = (unsigned long)(*ppos % PAGE_SIZE);
336 if (count > PAGE_SIZE - offset)
337 csize = PAGE_SIZE - offset;
338 else
339 csize = count;
Vivek Goyal50b1fdb2005-06-25 14:58:23 -0700340
Vivek Goyal315c2152005-06-25 14:58:24 -0700341 rc = copy_oldmem_page(pfn, buf, csize, offset, 1);
342 if (rc < 0)
343 return rc;
Vivek Goyal50b1fdb2005-06-25 14:58:23 -0700344 buf += csize;
345 *ppos += csize;
346 read += csize;
347 count -= csize;
348 }
Vivek Goyal50b1fdb2005-06-25 14:58:23 -0700349 return read;
350}
351#endif
352
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353extern long vread(char *buf, char *addr, unsigned long count);
354extern long vwrite(char *buf, char *addr, unsigned long count);
355
356/*
357 * This function reads the *virtual* memory as seen by the kernel.
358 */
359static ssize_t read_kmem(struct file *file, char __user *buf,
360 size_t count, loff_t *ppos)
361{
362 unsigned long p = *ppos;
363 ssize_t low_count, read, sz;
364 char * kbuf; /* k-addr because vread() takes vmlist_lock rwlock */
365
366 read = 0;
367 if (p < (unsigned long) high_memory) {
368 low_count = count;
369 if (count > (unsigned long) high_memory - p)
370 low_count = (unsigned long) high_memory - p;
371
372#ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
373 /* we don't have page 0 mapped on sparc and m68k.. */
374 if (p < PAGE_SIZE && low_count > 0) {
375 size_t tmp = PAGE_SIZE - p;
376 if (tmp > low_count) tmp = low_count;
377 if (clear_user(buf, tmp))
378 return -EFAULT;
379 buf += tmp;
380 p += tmp;
381 read += tmp;
382 low_count -= tmp;
383 count -= tmp;
384 }
385#endif
386 while (low_count > 0) {
387 /*
388 * Handle first page in case it's not aligned
389 */
390 if (-p & (PAGE_SIZE - 1))
391 sz = -p & (PAGE_SIZE - 1);
392 else
393 sz = PAGE_SIZE;
394
395 sz = min_t(unsigned long, sz, low_count);
396
397 /*
398 * On ia64 if a page has been mapped somewhere as
399 * uncached, then it must also be accessed uncached
400 * by the kernel or data corruption may occur
401 */
402 kbuf = xlate_dev_kmem_ptr((char *)p);
403
404 if (copy_to_user(buf, kbuf, sz))
405 return -EFAULT;
406 buf += sz;
407 p += sz;
408 read += sz;
409 low_count -= sz;
410 count -= sz;
411 }
412 }
413
414 if (count > 0) {
415 kbuf = (char *)__get_free_page(GFP_KERNEL);
416 if (!kbuf)
417 return -ENOMEM;
418 while (count > 0) {
419 int len = count;
420
421 if (len > PAGE_SIZE)
422 len = PAGE_SIZE;
423 len = vread(kbuf, (char *)p, len);
424 if (!len)
425 break;
426 if (copy_to_user(buf, kbuf, len)) {
427 free_page((unsigned long)kbuf);
428 return -EFAULT;
429 }
430 count -= len;
431 buf += len;
432 read += len;
433 p += len;
434 }
435 free_page((unsigned long)kbuf);
436 }
437 *ppos = p;
438 return read;
439}
440
441
442static inline ssize_t
443do_write_kmem(void *p, unsigned long realp, const char __user * buf,
444 size_t count, loff_t *ppos)
445{
446 ssize_t written, sz;
447 unsigned long copied;
448
449 written = 0;
450#ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
451 /* we don't have page 0 mapped on sparc and m68k.. */
452 if (realp < PAGE_SIZE) {
453 unsigned long sz = PAGE_SIZE - realp;
454 if (sz > count)
455 sz = count;
456 /* Hmm. Do something? */
457 buf += sz;
458 p += sz;
459 realp += sz;
460 count -= sz;
461 written += sz;
462 }
463#endif
464
465 while (count > 0) {
466 char *ptr;
467 /*
468 * Handle first page in case it's not aligned
469 */
470 if (-realp & (PAGE_SIZE - 1))
471 sz = -realp & (PAGE_SIZE - 1);
472 else
473 sz = PAGE_SIZE;
474
475 sz = min_t(unsigned long, sz, count);
476
477 /*
478 * On ia64 if a page has been mapped somewhere as
479 * uncached, then it must also be accessed uncached
480 * by the kernel or data corruption may occur
481 */
482 ptr = xlate_dev_kmem_ptr(p);
483
484 copied = copy_from_user(ptr, buf, sz);
485 if (copied) {
Jan Beulichc654d602006-03-25 03:07:31 -0800486 written += sz - copied;
487 if (written)
488 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489 return -EFAULT;
490 }
491 buf += sz;
492 p += sz;
493 realp += sz;
494 count -= sz;
495 written += sz;
496 }
497
498 *ppos += written;
499 return written;
500}
501
502
503/*
504 * This function writes to the *virtual* memory as seen by the kernel.
505 */
506static ssize_t write_kmem(struct file * file, const char __user * buf,
507 size_t count, loff_t *ppos)
508{
509 unsigned long p = *ppos;
510 ssize_t wrote = 0;
511 ssize_t virtr = 0;
512 ssize_t written;
513 char * kbuf; /* k-addr because vwrite() takes vmlist_lock rwlock */
514
515 if (p < (unsigned long) high_memory) {
516
517 wrote = count;
518 if (count > (unsigned long) high_memory - p)
519 wrote = (unsigned long) high_memory - p;
520
521 written = do_write_kmem((void*)p, p, buf, wrote, ppos);
522 if (written != wrote)
523 return written;
524 wrote = written;
525 p += wrote;
526 buf += wrote;
527 count -= wrote;
528 }
529
530 if (count > 0) {
531 kbuf = (char *)__get_free_page(GFP_KERNEL);
532 if (!kbuf)
533 return wrote ? wrote : -ENOMEM;
534 while (count > 0) {
535 int len = count;
536
537 if (len > PAGE_SIZE)
538 len = PAGE_SIZE;
539 if (len) {
540 written = copy_from_user(kbuf, buf, len);
541 if (written) {
Jan Beulichc654d602006-03-25 03:07:31 -0800542 if (wrote + virtr)
543 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544 free_page((unsigned long)kbuf);
Jan Beulichc654d602006-03-25 03:07:31 -0800545 return -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546 }
547 }
548 len = vwrite(kbuf, (char *)p, len);
549 count -= len;
550 buf += len;
551 virtr += len;
552 p += len;
553 }
554 free_page((unsigned long)kbuf);
555 }
556
557 *ppos = p;
558 return virtr + wrote;
559}
560
Russell King4f911d62007-05-08 00:28:17 -0700561#ifdef CONFIG_DEVPORT
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562static ssize_t read_port(struct file * file, char __user * buf,
563 size_t count, loff_t *ppos)
564{
565 unsigned long i = *ppos;
566 char __user *tmp = buf;
567
568 if (!access_ok(VERIFY_WRITE, buf, count))
569 return -EFAULT;
570 while (count-- > 0 && i < 65536) {
571 if (__put_user(inb(i),tmp) < 0)
572 return -EFAULT;
573 i++;
574 tmp++;
575 }
576 *ppos = i;
577 return tmp-buf;
578}
579
580static ssize_t write_port(struct file * file, const char __user * buf,
581 size_t count, loff_t *ppos)
582{
583 unsigned long i = *ppos;
584 const char __user * tmp = buf;
585
586 if (!access_ok(VERIFY_READ,buf,count))
587 return -EFAULT;
588 while (count-- > 0 && i < 65536) {
589 char c;
Jan Beulichc654d602006-03-25 03:07:31 -0800590 if (__get_user(c, tmp)) {
591 if (tmp > buf)
592 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593 return -EFAULT;
Jan Beulichc654d602006-03-25 03:07:31 -0800594 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595 outb(c,i);
596 i++;
597 tmp++;
598 }
599 *ppos = i;
600 return tmp-buf;
601}
602#endif
603
604static ssize_t read_null(struct file * file, char __user * buf,
605 size_t count, loff_t *ppos)
606{
607 return 0;
608}
609
610static ssize_t write_null(struct file * file, const char __user * buf,
611 size_t count, loff_t *ppos)
612{
613 return count;
614}
615
Jens Axboe1ebd32f2006-04-26 14:40:08 +0200616static int pipe_to_null(struct pipe_inode_info *info, struct pipe_buffer *buf,
617 struct splice_desc *sd)
618{
619 return sd->len;
620}
621
622static ssize_t splice_write_null(struct pipe_inode_info *pipe,struct file *out,
623 loff_t *ppos, size_t len, unsigned int flags)
624{
625 return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_null);
626}
627
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628#ifdef CONFIG_MMU
629/*
630 * For fun, we are using the MMU for this.
631 */
632static inline size_t read_zero_pagealigned(char __user * buf, size_t size)
633{
634 struct mm_struct *mm;
635 struct vm_area_struct * vma;
636 unsigned long addr=(unsigned long)buf;
637
638 mm = current->mm;
639 /* Oops, this was forgotten before. -ben */
640 down_read(&mm->mmap_sem);
641
642 /* For private mappings, just map in zero pages. */
643 for (vma = find_vma(mm, addr); vma; vma = vma->vm_next) {
644 unsigned long count;
645
646 if (vma->vm_start > addr || (vma->vm_flags & VM_WRITE) == 0)
647 goto out_up;
Linus Torvalds6aab3412005-11-28 14:34:23 -0800648 if (vma->vm_flags & (VM_SHARED | VM_HUGETLB))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649 break;
650 count = vma->vm_end - addr;
651 if (count > size)
652 count = size;
653
654 zap_page_range(vma, addr, count, NULL);
Hugh Dickins5fcf7bb2006-12-10 02:18:43 -0800655 if (zeromap_page_range(vma, addr, count, PAGE_COPY))
656 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657
658 size -= count;
659 buf += count;
660 addr += count;
661 if (size == 0)
662 goto out_up;
663 }
664
665 up_read(&mm->mmap_sem);
666
667 /* The shared case is hard. Let's do the conventional zeroing. */
668 do {
669 unsigned long unwritten = clear_user(buf, PAGE_SIZE);
670 if (unwritten)
671 return size + unwritten - PAGE_SIZE;
672 cond_resched();
673 buf += PAGE_SIZE;
674 size -= PAGE_SIZE;
675 } while (size);
676
677 return size;
678out_up:
679 up_read(&mm->mmap_sem);
680 return size;
681}
682
683static ssize_t read_zero(struct file * file, char __user * buf,
684 size_t count, loff_t *ppos)
685{
686 unsigned long left, unwritten, written = 0;
687
688 if (!count)
689 return 0;
690
691 if (!access_ok(VERIFY_WRITE, buf, count))
692 return -EFAULT;
693
694 left = count;
695
696 /* do we want to be clever? Arbitrary cut-off */
697 if (count >= PAGE_SIZE*4) {
698 unsigned long partial;
699
700 /* How much left of the page? */
701 partial = (PAGE_SIZE-1) & -(unsigned long) buf;
702 unwritten = clear_user(buf, partial);
703 written = partial - unwritten;
704 if (unwritten)
705 goto out;
706 left -= partial;
707 buf += partial;
708 unwritten = read_zero_pagealigned(buf, left & PAGE_MASK);
709 written += (left & PAGE_MASK) - unwritten;
710 if (unwritten)
711 goto out;
712 buf += left & PAGE_MASK;
713 left &= ~PAGE_MASK;
714 }
715 unwritten = clear_user(buf, left);
716 written += left - unwritten;
717out:
718 return written ? written : -EFAULT;
719}
720
721static int mmap_zero(struct file * file, struct vm_area_struct * vma)
722{
Hugh Dickins5fcf7bb2006-12-10 02:18:43 -0800723 int err;
724
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725 if (vma->vm_flags & VM_SHARED)
726 return shmem_zero_setup(vma);
Hugh Dickins5fcf7bb2006-12-10 02:18:43 -0800727 err = zeromap_page_range(vma, vma->vm_start,
728 vma->vm_end - vma->vm_start, vma->vm_page_prot);
729 BUG_ON(err == -EEXIST);
730 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731}
732#else /* CONFIG_MMU */
733static ssize_t read_zero(struct file * file, char * buf,
734 size_t count, loff_t *ppos)
735{
736 size_t todo = count;
737
738 while (todo) {
739 size_t chunk = todo;
740
741 if (chunk > 4096)
742 chunk = 4096; /* Just for latency reasons */
743 if (clear_user(buf, chunk))
744 return -EFAULT;
745 buf += chunk;
746 todo -= chunk;
747 cond_resched();
748 }
749 return count;
750}
751
752static int mmap_zero(struct file * file, struct vm_area_struct * vma)
753{
754 return -ENOSYS;
755}
756#endif /* CONFIG_MMU */
757
758static ssize_t write_full(struct file * file, const char __user * buf,
759 size_t count, loff_t *ppos)
760{
761 return -ENOSPC;
762}
763
764/*
765 * Special lseek() function for /dev/null and /dev/zero. Most notably, you
766 * can fopen() both devices with "a" now. This was previously impossible.
767 * -- SRB.
768 */
769
770static loff_t null_lseek(struct file * file, loff_t offset, int orig)
771{
772 return file->f_pos = 0;
773}
774
775/*
776 * The memory devices use the full 32/64 bits of the offset, and so we cannot
777 * check against negative addresses: they are ok. The return value is weird,
778 * though, in that case (0).
779 *
780 * also note that seeking relative to the "end of file" isn't supported:
781 * it has no meaning, so it returns -EINVAL.
782 */
783static loff_t memory_lseek(struct file * file, loff_t offset, int orig)
784{
785 loff_t ret;
786
Josef Sipeka7113a92006-12-08 02:36:55 -0800787 mutex_lock(&file->f_path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788 switch (orig) {
789 case 0:
790 file->f_pos = offset;
791 ret = file->f_pos;
792 force_successful_syscall_return();
793 break;
794 case 1:
795 file->f_pos += offset;
796 ret = file->f_pos;
797 force_successful_syscall_return();
798 break;
799 default:
800 ret = -EINVAL;
801 }
Josef Sipeka7113a92006-12-08 02:36:55 -0800802 mutex_unlock(&file->f_path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803 return ret;
804}
805
806static int open_port(struct inode * inode, struct file * filp)
807{
808 return capable(CAP_SYS_RAWIO) ? 0 : -EPERM;
809}
810
811#define zero_lseek null_lseek
812#define full_lseek null_lseek
813#define write_zero write_null
814#define read_full read_zero
815#define open_mem open_port
816#define open_kmem open_mem
Vivek Goyal50b1fdb2005-06-25 14:58:23 -0700817#define open_oldmem open_mem
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818
Arjan van de Ven62322d22006-07-03 00:24:21 -0700819static const struct file_operations mem_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820 .llseek = memory_lseek,
821 .read = read_mem,
822 .write = write_mem,
823 .mmap = mmap_mem,
824 .open = open_mem,
David Howells5da61852006-09-27 01:50:16 -0700825 .get_unmapped_area = get_unmapped_area_mem,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826};
827
Arjan van de Ven62322d22006-07-03 00:24:21 -0700828static const struct file_operations kmem_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829 .llseek = memory_lseek,
830 .read = read_kmem,
831 .write = write_kmem,
832 .mmap = mmap_kmem,
833 .open = open_kmem,
David Howells5da61852006-09-27 01:50:16 -0700834 .get_unmapped_area = get_unmapped_area_mem,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835};
836
Arjan van de Ven62322d22006-07-03 00:24:21 -0700837static const struct file_operations null_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838 .llseek = null_lseek,
839 .read = read_null,
840 .write = write_null,
Jens Axboe1ebd32f2006-04-26 14:40:08 +0200841 .splice_write = splice_write_null,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842};
843
Russell King4f911d62007-05-08 00:28:17 -0700844#ifdef CONFIG_DEVPORT
Arjan van de Ven62322d22006-07-03 00:24:21 -0700845static const struct file_operations port_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846 .llseek = memory_lseek,
847 .read = read_port,
848 .write = write_port,
849 .open = open_port,
850};
851#endif
852
Arjan van de Ven62322d22006-07-03 00:24:21 -0700853static const struct file_operations zero_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854 .llseek = zero_lseek,
855 .read = read_zero,
856 .write = write_zero,
857 .mmap = mmap_zero,
858};
859
David Howells5da61852006-09-27 01:50:16 -0700860/*
861 * capabilities for /dev/zero
862 * - permits private mappings, "copies" are taken of the source of zeros
863 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864static struct backing_dev_info zero_bdi = {
865 .capabilities = BDI_CAP_MAP_COPY,
866};
867
Arjan van de Ven62322d22006-07-03 00:24:21 -0700868static const struct file_operations full_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869 .llseek = full_lseek,
870 .read = read_full,
871 .write = write_full,
872};
873
Vivek Goyal50b1fdb2005-06-25 14:58:23 -0700874#ifdef CONFIG_CRASH_DUMP
Arjan van de Ven62322d22006-07-03 00:24:21 -0700875static const struct file_operations oldmem_fops = {
Vivek Goyal50b1fdb2005-06-25 14:58:23 -0700876 .read = read_oldmem,
877 .open = open_oldmem,
878};
879#endif
880
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881static ssize_t kmsg_write(struct file * file, const char __user * buf,
882 size_t count, loff_t *ppos)
883{
884 char *tmp;
Guillaume Chazaraincd140a52006-01-08 01:02:43 -0800885 ssize_t ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886
887 tmp = kmalloc(count + 1, GFP_KERNEL);
888 if (tmp == NULL)
889 return -ENOMEM;
890 ret = -EFAULT;
891 if (!copy_from_user(tmp, buf, count)) {
892 tmp[count] = 0;
893 ret = printk("%s", tmp);
Guillaume Chazaraincd140a52006-01-08 01:02:43 -0800894 if (ret > count)
895 /* printk can add a prefix */
896 ret = count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897 }
898 kfree(tmp);
899 return ret;
900}
901
Arjan van de Ven62322d22006-07-03 00:24:21 -0700902static const struct file_operations kmsg_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903 .write = kmsg_write,
904};
905
906static int memory_open(struct inode * inode, struct file * filp)
907{
908 switch (iminor(inode)) {
909 case 1:
910 filp->f_op = &mem_fops;
David Howells5da61852006-09-27 01:50:16 -0700911 filp->f_mapping->backing_dev_info =
912 &directly_mappable_cdev_bdi;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913 break;
914 case 2:
915 filp->f_op = &kmem_fops;
David Howells5da61852006-09-27 01:50:16 -0700916 filp->f_mapping->backing_dev_info =
917 &directly_mappable_cdev_bdi;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918 break;
919 case 3:
920 filp->f_op = &null_fops;
921 break;
Russell King4f911d62007-05-08 00:28:17 -0700922#ifdef CONFIG_DEVPORT
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923 case 4:
924 filp->f_op = &port_fops;
925 break;
926#endif
927 case 5:
928 filp->f_mapping->backing_dev_info = &zero_bdi;
929 filp->f_op = &zero_fops;
930 break;
931 case 7:
932 filp->f_op = &full_fops;
933 break;
934 case 8:
935 filp->f_op = &random_fops;
936 break;
937 case 9:
938 filp->f_op = &urandom_fops;
939 break;
940 case 11:
941 filp->f_op = &kmsg_fops;
942 break;
Vivek Goyal50b1fdb2005-06-25 14:58:23 -0700943#ifdef CONFIG_CRASH_DUMP
944 case 12:
945 filp->f_op = &oldmem_fops;
946 break;
947#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948 default:
949 return -ENXIO;
950 }
951 if (filp->f_op && filp->f_op->open)
952 return filp->f_op->open(inode,filp);
953 return 0;
954}
955
Arjan van de Ven62322d22006-07-03 00:24:21 -0700956static const struct file_operations memory_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957 .open = memory_open, /* just a selector for the real open */
958};
959
960static const struct {
961 unsigned int minor;
962 char *name;
963 umode_t mode;
Arjan van de Ven99ac48f2006-03-28 01:56:41 -0800964 const struct file_operations *fops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965} devlist[] = { /* list of minor devices */
966 {1, "mem", S_IRUSR | S_IWUSR | S_IRGRP, &mem_fops},
967 {2, "kmem", S_IRUSR | S_IWUSR | S_IRGRP, &kmem_fops},
968 {3, "null", S_IRUGO | S_IWUGO, &null_fops},
Russell King4f911d62007-05-08 00:28:17 -0700969#ifdef CONFIG_DEVPORT
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970 {4, "port", S_IRUSR | S_IWUSR | S_IRGRP, &port_fops},
971#endif
972 {5, "zero", S_IRUGO | S_IWUGO, &zero_fops},
973 {7, "full", S_IRUGO | S_IWUGO, &full_fops},
974 {8, "random", S_IRUGO | S_IWUSR, &random_fops},
975 {9, "urandom", S_IRUGO | S_IWUSR, &urandom_fops},
976 {11,"kmsg", S_IRUGO | S_IWUSR, &kmsg_fops},
Vivek Goyal50b1fdb2005-06-25 14:58:23 -0700977#ifdef CONFIG_CRASH_DUMP
978 {12,"oldmem", S_IRUSR | S_IWUSR | S_IRGRP, &oldmem_fops},
979#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980};
981
gregkh@suse.deca8eca62005-03-23 09:53:09 -0800982static struct class *mem_class;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983
984static int __init chr_dev_init(void)
985{
986 int i;
987
988 if (register_chrdev(MEM_MAJOR,"mem",&memory_fops))
989 printk("unable to get major %d for memory devs\n", MEM_MAJOR);
990
gregkh@suse.deca8eca62005-03-23 09:53:09 -0800991 mem_class = class_create(THIS_MODULE, "mem");
Greg Kroah-Hartman7c69ef72005-06-20 21:15:16 -0700992 for (i = 0; i < ARRAY_SIZE(devlist); i++)
Greg Kroah-Hartmanebf644c2006-07-25 17:13:31 -0700993 device_create(mem_class, NULL,
994 MKDEV(MEM_MAJOR, devlist[i].minor),
995 devlist[i].name);
996
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997 return 0;
998}
999
1000fs_initcall(chr_dev_init);