blob: 934ffafedaea207131f7d193bed9e75c929f4db9 [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 Axboed6b29d72007-06-04 09:59:47 +020027#include <linux/splice.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{
venkatesh.pallipadi@intel.comf0970c12008-03-18 17:00:20 -070044#if defined(CONFIG_IA64)
Linus Torvalds1da177e2005-04-16 15:20:36 -070045 /*
46 * On ia64, we ignore O_SYNC because we cannot tolerate memory attribute aliases.
47 */
48 return !(efi_mem_attributes(addr) & EFI_MEMORY_WB);
Ralf Baechle24e9d0b962007-07-10 17:32:56 +010049#elif defined(CONFIG_MIPS)
50 {
51 extern int __uncached_access(struct file *file,
52 unsigned long addr);
53
54 return __uncached_access(file, addr);
55 }
Linus Torvalds1da177e2005-04-16 15:20:36 -070056#else
57 /*
58 * Accessing memory above the top the kernel knows about or through a file pointer
59 * that was marked O_SYNC will be done non-cached.
60 */
61 if (file->f_flags & O_SYNC)
62 return 1;
63 return addr >= __pa(high_memory);
64#endif
65}
66
67#ifndef ARCH_HAS_VALID_PHYS_ADDR_RANGE
Bjorn Helgaas136939a2006-03-26 01:37:05 -080068static inline int valid_phys_addr_range(unsigned long addr, size_t count)
Linus Torvalds1da177e2005-04-16 15:20:36 -070069{
Bjorn Helgaas136939a2006-03-26 01:37:05 -080070 if (addr + count > __pa(high_memory))
Linus Torvalds1da177e2005-04-16 15:20:36 -070071 return 0;
72
Linus Torvalds1da177e2005-04-16 15:20:36 -070073 return 1;
74}
Bjorn Helgaas80851ef2006-01-08 01:04:13 -080075
Lennert Buytenhek06c67be2006-07-10 04:45:27 -070076static inline int valid_mmap_phys_addr_range(unsigned long pfn, size_t size)
Bjorn Helgaas80851ef2006-01-08 01:04:13 -080077{
78 return 1;
79}
Linus Torvalds1da177e2005-04-16 15:20:36 -070080#endif
81
Arjan van de Venae531c22008-04-24 23:40:47 +020082#ifdef CONFIG_NONPROMISC_DEVMEM
Venki Pallipadie2beb3e2008-03-06 23:01:47 -080083static inline int range_is_allowed(unsigned long pfn, unsigned long size)
Arjan van de Venae531c22008-04-24 23:40:47 +020084{
Venki Pallipadie2beb3e2008-03-06 23:01:47 -080085 u64 from = ((u64)pfn) << PAGE_SHIFT;
86 u64 to = from + size;
87 u64 cursor = from;
Arjan van de Venae531c22008-04-24 23:40:47 +020088
Venki Pallipadie2beb3e2008-03-06 23:01:47 -080089 while (cursor < to) {
90 if (!devmem_is_allowed(pfn)) {
91 printk(KERN_INFO
92 "Program %s tried to access /dev/mem between %Lx->%Lx.\n",
Arjan van de Venae531c22008-04-24 23:40:47 +020093 current->comm, from, to);
94 return 0;
95 }
Venki Pallipadie2beb3e2008-03-06 23:01:47 -080096 cursor += PAGE_SIZE;
97 pfn++;
Arjan van de Venae531c22008-04-24 23:40:47 +020098 }
99 return 1;
100}
101#else
Venki Pallipadie2beb3e2008-03-06 23:01:47 -0800102static inline int range_is_allowed(unsigned long pfn, unsigned long size)
Arjan van de Venae531c22008-04-24 23:40:47 +0200103{
104 return 1;
105}
106#endif
107
venkatesh.pallipadi@intel.come045fb22008-03-18 17:00:15 -0700108void __attribute__((weak)) unxlate_dev_mem_ptr(unsigned long phys, void *addr)
109{
110}
111
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112/*
113 * This funcion reads the *physical* memory. The f_pos points directly to the
114 * memory location.
115 */
116static ssize_t read_mem(struct file * file, char __user * buf,
117 size_t count, loff_t *ppos)
118{
119 unsigned long p = *ppos;
120 ssize_t read, sz;
121 char *ptr;
122
Bjorn Helgaas136939a2006-03-26 01:37:05 -0800123 if (!valid_phys_addr_range(p, count))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124 return -EFAULT;
125 read = 0;
126#ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
127 /* we don't have page 0 mapped on sparc and m68k.. */
128 if (p < PAGE_SIZE) {
129 sz = PAGE_SIZE - p;
130 if (sz > count)
131 sz = count;
132 if (sz > 0) {
133 if (clear_user(buf, sz))
134 return -EFAULT;
135 buf += sz;
136 p += sz;
137 count -= sz;
138 read += sz;
139 }
140 }
141#endif
142
143 while (count > 0) {
144 /*
145 * Handle first page in case it's not aligned
146 */
147 if (-p & (PAGE_SIZE - 1))
148 sz = -p & (PAGE_SIZE - 1);
149 else
150 sz = PAGE_SIZE;
151
152 sz = min_t(unsigned long, sz, count);
153
venkatesh.pallipadi@intel.come045fb22008-03-18 17:00:15 -0700154 if (!range_is_allowed(p >> PAGE_SHIFT, count))
155 return -EPERM;
156
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157 /*
158 * On ia64 if a page has been mapped somewhere as
159 * uncached, then it must also be accessed uncached
160 * by the kernel or data corruption may occur
161 */
162 ptr = xlate_dev_mem_ptr(p);
venkatesh.pallipadi@intel.come045fb22008-03-18 17:00:15 -0700163 if (!ptr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164 return -EFAULT;
venkatesh.pallipadi@intel.come045fb22008-03-18 17:00:15 -0700165
166 if (copy_to_user(buf, ptr, sz)) {
167 unxlate_dev_mem_ptr(p, ptr);
168 return -EFAULT;
169 }
170
171 unxlate_dev_mem_ptr(p, ptr);
172
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173 buf += sz;
174 p += sz;
175 count -= sz;
176 read += sz;
177 }
178
179 *ppos += read;
180 return read;
181}
182
183static ssize_t write_mem(struct file * file, const char __user * buf,
184 size_t count, loff_t *ppos)
185{
186 unsigned long p = *ppos;
187 ssize_t written, sz;
188 unsigned long copied;
189 void *ptr;
190
Bjorn Helgaas136939a2006-03-26 01:37:05 -0800191 if (!valid_phys_addr_range(p, count))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192 return -EFAULT;
193
194 written = 0;
195
196#ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
197 /* we don't have page 0 mapped on sparc and m68k.. */
198 if (p < PAGE_SIZE) {
199 unsigned long sz = PAGE_SIZE - p;
200 if (sz > count)
201 sz = count;
202 /* Hmm. Do something? */
203 buf += sz;
204 p += sz;
205 count -= sz;
206 written += sz;
207 }
208#endif
209
210 while (count > 0) {
211 /*
212 * Handle first page in case it's not aligned
213 */
214 if (-p & (PAGE_SIZE - 1))
215 sz = -p & (PAGE_SIZE - 1);
216 else
217 sz = PAGE_SIZE;
218
219 sz = min_t(unsigned long, sz, count);
220
venkatesh.pallipadi@intel.come045fb22008-03-18 17:00:15 -0700221 if (!range_is_allowed(p >> PAGE_SHIFT, sz))
222 return -EPERM;
223
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224 /*
225 * On ia64 if a page has been mapped somewhere as
226 * uncached, then it must also be accessed uncached
227 * by the kernel or data corruption may occur
228 */
229 ptr = xlate_dev_mem_ptr(p);
venkatesh.pallipadi@intel.come045fb22008-03-18 17:00:15 -0700230 if (!ptr) {
Jan Beulichc654d602006-03-25 03:07:31 -0800231 if (written)
232 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233 return -EFAULT;
234 }
venkatesh.pallipadi@intel.come045fb22008-03-18 17:00:15 -0700235
236 copied = copy_from_user(ptr, buf, sz);
237 if (copied) {
238 written += sz - copied;
239 unxlate_dev_mem_ptr(p, ptr);
240 if (written)
241 break;
242 return -EFAULT;
243 }
244
245 unxlate_dev_mem_ptr(p, ptr);
246
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247 buf += sz;
248 p += sz;
249 count -= sz;
250 written += sz;
251 }
252
253 *ppos += written;
254 return written;
255}
256
venkatesh.pallipadi@intel.comf0970c12008-03-18 17:00:20 -0700257int __attribute__((weak)) phys_mem_access_prot_allowed(struct file *file,
258 unsigned long pfn, unsigned long size, pgprot_t *vma_prot)
259{
260 return 1;
261}
262
Bjorn Helgaas44ac8412006-01-08 01:04:10 -0800263#ifndef __HAVE_PHYS_MEM_ACCESS_PROT
264static pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
265 unsigned long size, pgprot_t vma_prot)
266{
267#ifdef pgprot_noncached
268 unsigned long offset = pfn << PAGE_SHIFT;
269
270 if (uncached_access(file, offset))
271 return pgprot_noncached(vma_prot);
272#endif
273 return vma_prot;
274}
275#endif
276
David Howells5da61852006-09-27 01:50:16 -0700277#ifndef CONFIG_MMU
278static unsigned long get_unmapped_area_mem(struct file *file,
279 unsigned long addr,
280 unsigned long len,
281 unsigned long pgoff,
282 unsigned long flags)
283{
284 if (!valid_mmap_phys_addr_range(pgoff, len))
285 return (unsigned long) -EINVAL;
Benjamin Herrenschmidt8a932582007-04-16 22:53:16 -0700286 return pgoff << PAGE_SHIFT;
David Howells5da61852006-09-27 01:50:16 -0700287}
288
289/* can't do an in-place private mapping if there's no MMU */
290static inline int private_mapping_ok(struct vm_area_struct *vma)
291{
292 return vma->vm_flags & VM_MAYSHARE;
293}
294#else
295#define get_unmapped_area_mem NULL
296
297static inline int private_mapping_ok(struct vm_area_struct *vma)
298{
299 return 1;
300}
301#endif
302
venkatesh.pallipadi@intel.come7f260a2008-03-18 17:00:21 -0700303void __attribute__((weak))
304map_devmem(unsigned long pfn, unsigned long len, pgprot_t prot)
305{
306 /* nothing. architectures can override. */
307}
308
309void __attribute__((weak))
310unmap_devmem(unsigned long pfn, unsigned long len, pgprot_t prot)
311{
312 /* nothing. architectures can override. */
313}
314
315static void mmap_mem_open(struct vm_area_struct *vma)
316{
317 map_devmem(vma->vm_pgoff, vma->vm_end - vma->vm_start,
318 vma->vm_page_prot);
319}
320
321static void mmap_mem_close(struct vm_area_struct *vma)
322{
323 unmap_devmem(vma->vm_pgoff, vma->vm_end - vma->vm_start,
324 vma->vm_page_prot);
325}
326
327static struct vm_operations_struct mmap_mem_ops = {
328 .open = mmap_mem_open,
329 .close = mmap_mem_close
330};
331
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332static int mmap_mem(struct file * file, struct vm_area_struct * vma)
333{
Bjorn Helgaas80851ef2006-01-08 01:04:13 -0800334 size_t size = vma->vm_end - vma->vm_start;
335
Lennert Buytenhek06c67be2006-07-10 04:45:27 -0700336 if (!valid_mmap_phys_addr_range(vma->vm_pgoff, size))
Bjorn Helgaas80851ef2006-01-08 01:04:13 -0800337 return -EINVAL;
338
David Howells5da61852006-09-27 01:50:16 -0700339 if (!private_mapping_ok(vma))
340 return -ENOSYS;
341
Venki Pallipadie2beb3e2008-03-06 23:01:47 -0800342 if (!range_is_allowed(vma->vm_pgoff, size))
343 return -EPERM;
344
venkatesh.pallipadi@intel.comf0970c12008-03-18 17:00:20 -0700345 if (!phys_mem_access_prot_allowed(file, vma->vm_pgoff, size,
346 &vma->vm_page_prot))
347 return -EINVAL;
348
Roland Dreier8b150472005-10-28 17:46:18 -0700349 vma->vm_page_prot = phys_mem_access_prot(file, vma->vm_pgoff,
Bjorn Helgaas80851ef2006-01-08 01:04:13 -0800350 size,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351 vma->vm_page_prot);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352
venkatesh.pallipadi@intel.come7f260a2008-03-18 17:00:21 -0700353 vma->vm_ops = &mmap_mem_ops;
354
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355 /* Remap-pfn-range will mark the range VM_IO and VM_RESERVED */
356 if (remap_pfn_range(vma,
357 vma->vm_start,
358 vma->vm_pgoff,
Bjorn Helgaas80851ef2006-01-08 01:04:13 -0800359 size,
venkatesh.pallipadi@intel.come7f260a2008-03-18 17:00:21 -0700360 vma->vm_page_prot)) {
361 unmap_devmem(vma->vm_pgoff, size, vma->vm_page_prot);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362 return -EAGAIN;
venkatesh.pallipadi@intel.come7f260a2008-03-18 17:00:21 -0700363 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364 return 0;
365}
366
Arjan van de Venb781ecb2008-04-29 00:58:34 -0700367#ifdef CONFIG_DEVKMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368static int mmap_kmem(struct file * file, struct vm_area_struct * vma)
369{
Linus Torvalds4bb82552005-08-13 14:22:59 -0700370 unsigned long pfn;
371
Linus Torvalds6d3154c2007-01-22 08:53:24 -0800372 /* Turn a kernel-virtual address into a physical page frame */
373 pfn = __pa((u64)vma->vm_pgoff << PAGE_SHIFT) >> PAGE_SHIFT;
Linus Torvalds4bb82552005-08-13 14:22:59 -0700374
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375 /*
376 * RED-PEN: on some architectures there is more mapped memory
377 * than available in mem_map which pfn_valid checks
378 * for. Perhaps should add a new macro here.
379 *
380 * RED-PEN: vmalloc is not supported right now.
381 */
Linus Torvalds4bb82552005-08-13 14:22:59 -0700382 if (!pfn_valid(pfn))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383 return -EIO;
Linus Torvalds4bb82552005-08-13 14:22:59 -0700384
385 vma->vm_pgoff = pfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386 return mmap_mem(file, vma);
387}
Arjan van de Venb781ecb2008-04-29 00:58:34 -0700388#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389
Vivek Goyal50b1fdb2005-06-25 14:58:23 -0700390#ifdef CONFIG_CRASH_DUMP
391/*
392 * Read memory corresponding to the old kernel.
Vivek Goyal50b1fdb2005-06-25 14:58:23 -0700393 */
Vivek Goyal315c2152005-06-25 14:58:24 -0700394static ssize_t read_oldmem(struct file *file, char __user *buf,
Vivek Goyal50b1fdb2005-06-25 14:58:23 -0700395 size_t count, loff_t *ppos)
396{
Vivek Goyal315c2152005-06-25 14:58:24 -0700397 unsigned long pfn, offset;
398 size_t read = 0, csize;
399 int rc = 0;
Vivek Goyal50b1fdb2005-06-25 14:58:23 -0700400
Maneesh Soni72414d32005-06-25 14:58:28 -0700401 while (count) {
Vivek Goyal50b1fdb2005-06-25 14:58:23 -0700402 pfn = *ppos / PAGE_SIZE;
Vivek Goyal315c2152005-06-25 14:58:24 -0700403 if (pfn > saved_max_pfn)
404 return read;
Vivek Goyal50b1fdb2005-06-25 14:58:23 -0700405
Vivek Goyal315c2152005-06-25 14:58:24 -0700406 offset = (unsigned long)(*ppos % PAGE_SIZE);
407 if (count > PAGE_SIZE - offset)
408 csize = PAGE_SIZE - offset;
409 else
410 csize = count;
Vivek Goyal50b1fdb2005-06-25 14:58:23 -0700411
Vivek Goyal315c2152005-06-25 14:58:24 -0700412 rc = copy_oldmem_page(pfn, buf, csize, offset, 1);
413 if (rc < 0)
414 return rc;
Vivek Goyal50b1fdb2005-06-25 14:58:23 -0700415 buf += csize;
416 *ppos += csize;
417 read += csize;
418 count -= csize;
419 }
Vivek Goyal50b1fdb2005-06-25 14:58:23 -0700420 return read;
421}
422#endif
423
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424extern long vread(char *buf, char *addr, unsigned long count);
425extern long vwrite(char *buf, char *addr, unsigned long count);
426
Arjan van de Venb781ecb2008-04-29 00:58:34 -0700427#ifdef CONFIG_DEVKMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428/*
429 * This function reads the *virtual* memory as seen by the kernel.
430 */
431static ssize_t read_kmem(struct file *file, char __user *buf,
432 size_t count, loff_t *ppos)
433{
434 unsigned long p = *ppos;
435 ssize_t low_count, read, sz;
436 char * kbuf; /* k-addr because vread() takes vmlist_lock rwlock */
437
438 read = 0;
439 if (p < (unsigned long) high_memory) {
440 low_count = count;
441 if (count > (unsigned long) high_memory - p)
442 low_count = (unsigned long) high_memory - p;
443
444#ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
445 /* we don't have page 0 mapped on sparc and m68k.. */
446 if (p < PAGE_SIZE && low_count > 0) {
447 size_t tmp = PAGE_SIZE - p;
448 if (tmp > low_count) tmp = low_count;
449 if (clear_user(buf, tmp))
450 return -EFAULT;
451 buf += tmp;
452 p += tmp;
453 read += tmp;
454 low_count -= tmp;
455 count -= tmp;
456 }
457#endif
458 while (low_count > 0) {
459 /*
460 * Handle first page in case it's not aligned
461 */
462 if (-p & (PAGE_SIZE - 1))
463 sz = -p & (PAGE_SIZE - 1);
464 else
465 sz = PAGE_SIZE;
466
467 sz = min_t(unsigned long, sz, low_count);
468
469 /*
470 * On ia64 if a page has been mapped somewhere as
471 * uncached, then it must also be accessed uncached
472 * by the kernel or data corruption may occur
473 */
474 kbuf = xlate_dev_kmem_ptr((char *)p);
475
476 if (copy_to_user(buf, kbuf, sz))
477 return -EFAULT;
478 buf += sz;
479 p += sz;
480 read += sz;
481 low_count -= sz;
482 count -= sz;
483 }
484 }
485
486 if (count > 0) {
487 kbuf = (char *)__get_free_page(GFP_KERNEL);
488 if (!kbuf)
489 return -ENOMEM;
490 while (count > 0) {
491 int len = count;
492
493 if (len > PAGE_SIZE)
494 len = PAGE_SIZE;
495 len = vread(kbuf, (char *)p, len);
496 if (!len)
497 break;
498 if (copy_to_user(buf, kbuf, len)) {
499 free_page((unsigned long)kbuf);
500 return -EFAULT;
501 }
502 count -= len;
503 buf += len;
504 read += len;
505 p += len;
506 }
507 free_page((unsigned long)kbuf);
508 }
509 *ppos = p;
510 return read;
511}
512
513
514static inline ssize_t
515do_write_kmem(void *p, unsigned long realp, const char __user * buf,
516 size_t count, loff_t *ppos)
517{
518 ssize_t written, sz;
519 unsigned long copied;
520
521 written = 0;
522#ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
523 /* we don't have page 0 mapped on sparc and m68k.. */
524 if (realp < PAGE_SIZE) {
525 unsigned long sz = PAGE_SIZE - realp;
526 if (sz > count)
527 sz = count;
528 /* Hmm. Do something? */
529 buf += sz;
530 p += sz;
531 realp += sz;
532 count -= sz;
533 written += sz;
534 }
535#endif
536
537 while (count > 0) {
538 char *ptr;
539 /*
540 * Handle first page in case it's not aligned
541 */
542 if (-realp & (PAGE_SIZE - 1))
543 sz = -realp & (PAGE_SIZE - 1);
544 else
545 sz = PAGE_SIZE;
546
547 sz = min_t(unsigned long, sz, count);
548
549 /*
550 * On ia64 if a page has been mapped somewhere as
551 * uncached, then it must also be accessed uncached
552 * by the kernel or data corruption may occur
553 */
554 ptr = xlate_dev_kmem_ptr(p);
555
556 copied = copy_from_user(ptr, buf, sz);
557 if (copied) {
Jan Beulichc654d602006-03-25 03:07:31 -0800558 written += sz - copied;
559 if (written)
560 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561 return -EFAULT;
562 }
563 buf += sz;
564 p += sz;
565 realp += sz;
566 count -= sz;
567 written += sz;
568 }
569
570 *ppos += written;
571 return written;
572}
573
574
575/*
576 * This function writes to the *virtual* memory as seen by the kernel.
577 */
578static ssize_t write_kmem(struct file * file, const char __user * buf,
579 size_t count, loff_t *ppos)
580{
581 unsigned long p = *ppos;
582 ssize_t wrote = 0;
583 ssize_t virtr = 0;
584 ssize_t written;
585 char * kbuf; /* k-addr because vwrite() takes vmlist_lock rwlock */
586
587 if (p < (unsigned long) high_memory) {
588
589 wrote = count;
590 if (count > (unsigned long) high_memory - p)
591 wrote = (unsigned long) high_memory - p;
592
593 written = do_write_kmem((void*)p, p, buf, wrote, ppos);
594 if (written != wrote)
595 return written;
596 wrote = written;
597 p += wrote;
598 buf += wrote;
599 count -= wrote;
600 }
601
602 if (count > 0) {
603 kbuf = (char *)__get_free_page(GFP_KERNEL);
604 if (!kbuf)
605 return wrote ? wrote : -ENOMEM;
606 while (count > 0) {
607 int len = count;
608
609 if (len > PAGE_SIZE)
610 len = PAGE_SIZE;
611 if (len) {
612 written = copy_from_user(kbuf, buf, len);
613 if (written) {
Jan Beulichc654d602006-03-25 03:07:31 -0800614 if (wrote + virtr)
615 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616 free_page((unsigned long)kbuf);
Jan Beulichc654d602006-03-25 03:07:31 -0800617 return -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618 }
619 }
620 len = vwrite(kbuf, (char *)p, len);
621 count -= len;
622 buf += len;
623 virtr += len;
624 p += len;
625 }
626 free_page((unsigned long)kbuf);
627 }
628
629 *ppos = p;
630 return virtr + wrote;
631}
Arjan van de Venb781ecb2008-04-29 00:58:34 -0700632#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633
Russell King4f911d62007-05-08 00:28:17 -0700634#ifdef CONFIG_DEVPORT
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635static ssize_t read_port(struct file * file, char __user * buf,
636 size_t count, loff_t *ppos)
637{
638 unsigned long i = *ppos;
639 char __user *tmp = buf;
640
641 if (!access_ok(VERIFY_WRITE, buf, count))
642 return -EFAULT;
643 while (count-- > 0 && i < 65536) {
644 if (__put_user(inb(i),tmp) < 0)
645 return -EFAULT;
646 i++;
647 tmp++;
648 }
649 *ppos = i;
650 return tmp-buf;
651}
652
653static ssize_t write_port(struct file * file, const char __user * buf,
654 size_t count, loff_t *ppos)
655{
656 unsigned long i = *ppos;
657 const char __user * tmp = buf;
658
659 if (!access_ok(VERIFY_READ,buf,count))
660 return -EFAULT;
661 while (count-- > 0 && i < 65536) {
662 char c;
Jan Beulichc654d602006-03-25 03:07:31 -0800663 if (__get_user(c, tmp)) {
664 if (tmp > buf)
665 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666 return -EFAULT;
Jan Beulichc654d602006-03-25 03:07:31 -0800667 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668 outb(c,i);
669 i++;
670 tmp++;
671 }
672 *ppos = i;
673 return tmp-buf;
674}
675#endif
676
677static ssize_t read_null(struct file * file, char __user * buf,
678 size_t count, loff_t *ppos)
679{
680 return 0;
681}
682
683static ssize_t write_null(struct file * file, const char __user * buf,
684 size_t count, loff_t *ppos)
685{
686 return count;
687}
688
Jens Axboe1ebd32f2006-04-26 14:40:08 +0200689static int pipe_to_null(struct pipe_inode_info *info, struct pipe_buffer *buf,
690 struct splice_desc *sd)
691{
692 return sd->len;
693}
694
695static ssize_t splice_write_null(struct pipe_inode_info *pipe,struct file *out,
696 loff_t *ppos, size_t len, unsigned int flags)
697{
698 return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_null);
699}
700
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701static ssize_t read_zero(struct file * file, char __user * buf,
702 size_t count, loff_t *ppos)
703{
Nick Piggin557ed1f2007-10-16 01:24:40 -0700704 size_t written;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705
706 if (!count)
707 return 0;
708
709 if (!access_ok(VERIFY_WRITE, buf, count))
710 return -EFAULT;
711
Nick Piggin557ed1f2007-10-16 01:24:40 -0700712 written = 0;
713 while (count) {
714 unsigned long unwritten;
715 size_t chunk = count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716
Nick Piggin557ed1f2007-10-16 01:24:40 -0700717 if (chunk > PAGE_SIZE)
718 chunk = PAGE_SIZE; /* Just for latency reasons */
719 unwritten = clear_user(buf, chunk);
720 written += chunk - unwritten;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721 if (unwritten)
Nick Piggin557ed1f2007-10-16 01:24:40 -0700722 break;
723 buf += chunk;
724 count -= chunk;
725 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727 return written ? written : -EFAULT;
728}
729
730static int mmap_zero(struct file * file, struct vm_area_struct * vma)
731{
Nick Piggin557ed1f2007-10-16 01:24:40 -0700732#ifndef CONFIG_MMU
733 return -ENOSYS;
734#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735 if (vma->vm_flags & VM_SHARED)
736 return shmem_zero_setup(vma);
Nick Piggin557ed1f2007-10-16 01:24:40 -0700737 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739
740static ssize_t write_full(struct file * file, const char __user * buf,
741 size_t count, loff_t *ppos)
742{
743 return -ENOSPC;
744}
745
746/*
747 * Special lseek() function for /dev/null and /dev/zero. Most notably, you
748 * can fopen() both devices with "a" now. This was previously impossible.
749 * -- SRB.
750 */
751
752static loff_t null_lseek(struct file * file, loff_t offset, int orig)
753{
754 return file->f_pos = 0;
755}
756
757/*
758 * The memory devices use the full 32/64 bits of the offset, and so we cannot
759 * check against negative addresses: they are ok. The return value is weird,
760 * though, in that case (0).
761 *
762 * also note that seeking relative to the "end of file" isn't supported:
763 * it has no meaning, so it returns -EINVAL.
764 */
765static loff_t memory_lseek(struct file * file, loff_t offset, int orig)
766{
767 loff_t ret;
768
Josef Sipeka7113a92006-12-08 02:36:55 -0800769 mutex_lock(&file->f_path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770 switch (orig) {
771 case 0:
772 file->f_pos = offset;
773 ret = file->f_pos;
774 force_successful_syscall_return();
775 break;
776 case 1:
777 file->f_pos += offset;
778 ret = file->f_pos;
779 force_successful_syscall_return();
780 break;
781 default:
782 ret = -EINVAL;
783 }
Josef Sipeka7113a92006-12-08 02:36:55 -0800784 mutex_unlock(&file->f_path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785 return ret;
786}
787
788static int open_port(struct inode * inode, struct file * filp)
789{
790 return capable(CAP_SYS_RAWIO) ? 0 : -EPERM;
791}
792
793#define zero_lseek null_lseek
794#define full_lseek null_lseek
795#define write_zero write_null
796#define read_full read_zero
797#define open_mem open_port
798#define open_kmem open_mem
Vivek Goyal50b1fdb2005-06-25 14:58:23 -0700799#define open_oldmem open_mem
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800
Arjan van de Ven62322d22006-07-03 00:24:21 -0700801static const struct file_operations mem_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802 .llseek = memory_lseek,
803 .read = read_mem,
804 .write = write_mem,
805 .mmap = mmap_mem,
806 .open = open_mem,
David Howells5da61852006-09-27 01:50:16 -0700807 .get_unmapped_area = get_unmapped_area_mem,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808};
809
Arjan van de Venb781ecb2008-04-29 00:58:34 -0700810#ifdef CONFIG_DEVKMEM
Arjan van de Ven62322d22006-07-03 00:24:21 -0700811static const struct file_operations kmem_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812 .llseek = memory_lseek,
813 .read = read_kmem,
814 .write = write_kmem,
815 .mmap = mmap_kmem,
816 .open = open_kmem,
David Howells5da61852006-09-27 01:50:16 -0700817 .get_unmapped_area = get_unmapped_area_mem,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818};
Arjan van de Venb781ecb2008-04-29 00:58:34 -0700819#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820
Arjan van de Ven62322d22006-07-03 00:24:21 -0700821static const struct file_operations null_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822 .llseek = null_lseek,
823 .read = read_null,
824 .write = write_null,
Jens Axboe1ebd32f2006-04-26 14:40:08 +0200825 .splice_write = splice_write_null,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826};
827
Russell King4f911d62007-05-08 00:28:17 -0700828#ifdef CONFIG_DEVPORT
Arjan van de Ven62322d22006-07-03 00:24:21 -0700829static const struct file_operations port_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830 .llseek = memory_lseek,
831 .read = read_port,
832 .write = write_port,
833 .open = open_port,
834};
835#endif
836
Arjan van de Ven62322d22006-07-03 00:24:21 -0700837static const struct file_operations zero_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838 .llseek = zero_lseek,
839 .read = read_zero,
840 .write = write_zero,
841 .mmap = mmap_zero,
842};
843
David Howells5da61852006-09-27 01:50:16 -0700844/*
845 * capabilities for /dev/zero
846 * - permits private mappings, "copies" are taken of the source of zeros
847 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848static struct backing_dev_info zero_bdi = {
849 .capabilities = BDI_CAP_MAP_COPY,
850};
851
Arjan van de Ven62322d22006-07-03 00:24:21 -0700852static const struct file_operations full_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853 .llseek = full_lseek,
854 .read = read_full,
855 .write = write_full,
856};
857
Vivek Goyal50b1fdb2005-06-25 14:58:23 -0700858#ifdef CONFIG_CRASH_DUMP
Arjan van de Ven62322d22006-07-03 00:24:21 -0700859static const struct file_operations oldmem_fops = {
Vivek Goyal50b1fdb2005-06-25 14:58:23 -0700860 .read = read_oldmem,
861 .open = open_oldmem,
862};
863#endif
864
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865static ssize_t kmsg_write(struct file * file, const char __user * buf,
866 size_t count, loff_t *ppos)
867{
868 char *tmp;
Guillaume Chazaraincd140a52006-01-08 01:02:43 -0800869 ssize_t ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870
871 tmp = kmalloc(count + 1, GFP_KERNEL);
872 if (tmp == NULL)
873 return -ENOMEM;
874 ret = -EFAULT;
875 if (!copy_from_user(tmp, buf, count)) {
876 tmp[count] = 0;
877 ret = printk("%s", tmp);
Guillaume Chazaraincd140a52006-01-08 01:02:43 -0800878 if (ret > count)
879 /* printk can add a prefix */
880 ret = count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881 }
882 kfree(tmp);
883 return ret;
884}
885
Arjan van de Ven62322d22006-07-03 00:24:21 -0700886static const struct file_operations kmsg_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887 .write = kmsg_write,
888};
889
890static int memory_open(struct inode * inode, struct file * filp)
891{
892 switch (iminor(inode)) {
893 case 1:
894 filp->f_op = &mem_fops;
David Howells5da61852006-09-27 01:50:16 -0700895 filp->f_mapping->backing_dev_info =
896 &directly_mappable_cdev_bdi;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897 break;
Arjan van de Venb781ecb2008-04-29 00:58:34 -0700898#ifdef CONFIG_DEVKMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899 case 2:
900 filp->f_op = &kmem_fops;
David Howells5da61852006-09-27 01:50:16 -0700901 filp->f_mapping->backing_dev_info =
902 &directly_mappable_cdev_bdi;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903 break;
Arjan van de Venb781ecb2008-04-29 00:58:34 -0700904#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905 case 3:
906 filp->f_op = &null_fops;
907 break;
Russell King4f911d62007-05-08 00:28:17 -0700908#ifdef CONFIG_DEVPORT
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909 case 4:
910 filp->f_op = &port_fops;
911 break;
912#endif
913 case 5:
914 filp->f_mapping->backing_dev_info = &zero_bdi;
915 filp->f_op = &zero_fops;
916 break;
917 case 7:
918 filp->f_op = &full_fops;
919 break;
920 case 8:
921 filp->f_op = &random_fops;
922 break;
923 case 9:
924 filp->f_op = &urandom_fops;
925 break;
926 case 11:
927 filp->f_op = &kmsg_fops;
928 break;
Vivek Goyal50b1fdb2005-06-25 14:58:23 -0700929#ifdef CONFIG_CRASH_DUMP
930 case 12:
931 filp->f_op = &oldmem_fops;
932 break;
933#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934 default:
935 return -ENXIO;
936 }
937 if (filp->f_op && filp->f_op->open)
938 return filp->f_op->open(inode,filp);
939 return 0;
940}
941
Arjan van de Ven62322d22006-07-03 00:24:21 -0700942static const struct file_operations memory_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943 .open = memory_open, /* just a selector for the real open */
944};
945
946static const struct {
947 unsigned int minor;
948 char *name;
949 umode_t mode;
Arjan van de Ven99ac48f2006-03-28 01:56:41 -0800950 const struct file_operations *fops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951} devlist[] = { /* list of minor devices */
952 {1, "mem", S_IRUSR | S_IWUSR | S_IRGRP, &mem_fops},
Arjan van de Venb781ecb2008-04-29 00:58:34 -0700953#ifdef CONFIG_DEVKMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954 {2, "kmem", S_IRUSR | S_IWUSR | S_IRGRP, &kmem_fops},
Arjan van de Venb781ecb2008-04-29 00:58:34 -0700955#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956 {3, "null", S_IRUGO | S_IWUGO, &null_fops},
Russell King4f911d62007-05-08 00:28:17 -0700957#ifdef CONFIG_DEVPORT
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958 {4, "port", S_IRUSR | S_IWUSR | S_IRGRP, &port_fops},
959#endif
960 {5, "zero", S_IRUGO | S_IWUGO, &zero_fops},
961 {7, "full", S_IRUGO | S_IWUGO, &full_fops},
962 {8, "random", S_IRUGO | S_IWUSR, &random_fops},
963 {9, "urandom", S_IRUGO | S_IWUSR, &urandom_fops},
964 {11,"kmsg", S_IRUGO | S_IWUSR, &kmsg_fops},
Vivek Goyal50b1fdb2005-06-25 14:58:23 -0700965#ifdef CONFIG_CRASH_DUMP
966 {12,"oldmem", S_IRUSR | S_IWUSR | S_IRGRP, &oldmem_fops},
967#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968};
969
gregkh@suse.deca8eca62005-03-23 09:53:09 -0800970static struct class *mem_class;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971
972static int __init chr_dev_init(void)
973{
974 int i;
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700975 int err;
976
977 err = bdi_init(&zero_bdi);
978 if (err)
979 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980
981 if (register_chrdev(MEM_MAJOR,"mem",&memory_fops))
982 printk("unable to get major %d for memory devs\n", MEM_MAJOR);
983
gregkh@suse.deca8eca62005-03-23 09:53:09 -0800984 mem_class = class_create(THIS_MODULE, "mem");
Greg Kroah-Hartman7c69ef72005-06-20 21:15:16 -0700985 for (i = 0; i < ARRAY_SIZE(devlist); i++)
Greg Kroah-Hartmanebf644c2006-07-25 17:13:31 -0700986 device_create(mem_class, NULL,
987 MKDEV(MEM_MAJOR, devlist[i].minor),
988 devlist[i].name);
989
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990 return 0;
991}
992
993fs_initcall(chr_dev_init);