Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/drivers/char/mem.c | |
3 | * | |
4 | * Copyright (C) 1991, 1992 Linus Torvalds | |
5 | * | |
6 | * Added devfs support. | |
7 | * Jan-11-1998, C. Scott Ananian <cananian@alumni.princeton.edu> | |
af901ca1 | 8 | * Shared /dev/zero mmapping support, Feb 2000, Kanoj Sarcar <kanoj@sgi.com> |
1da177e4 LT |
9 | */ |
10 | ||
1da177e4 LT |
11 | #include <linux/mm.h> |
12 | #include <linux/miscdevice.h> | |
13 | #include <linux/slab.h> | |
14 | #include <linux/vmalloc.h> | |
15 | #include <linux/mman.h> | |
16 | #include <linux/random.h> | |
17 | #include <linux/init.h> | |
18 | #include <linux/raw.h> | |
19 | #include <linux/tty.h> | |
20 | #include <linux/capability.h> | |
1da177e4 LT |
21 | #include <linux/ptrace.h> |
22 | #include <linux/device.h> | |
50b1fdbd VG |
23 | #include <linux/highmem.h> |
24 | #include <linux/crash_dump.h> | |
1da177e4 | 25 | #include <linux/backing-dev.h> |
315c215c | 26 | #include <linux/bootmem.h> |
d6b29d7c | 27 | #include <linux/splice.h> |
b8a3ad5b | 28 | #include <linux/pfn.h> |
1da177e4 LT |
29 | |
30 | #include <asm/uaccess.h> | |
31 | #include <asm/io.h> | |
32 | ||
33 | #ifdef CONFIG_IA64 | |
34 | # include <linux/efi.h> | |
35 | #endif | |
36 | ||
1da177e4 LT |
37 | /* |
38 | * Architectures vary in how they handle caching for addresses | |
39 | * outside of main memory. | |
40 | * | |
41 | */ | |
42 | static inline int uncached_access(struct file *file, unsigned long addr) | |
43 | { | |
f0970c13 | 44 | #if defined(CONFIG_IA64) |
1da177e4 | 45 | /* |
6b2f3d1f | 46 | * On ia64, we ignore O_DSYNC because we cannot tolerate memory attribute aliases. |
1da177e4 LT |
47 | */ |
48 | return !(efi_mem_attributes(addr) & EFI_MEMORY_WB); | |
24e9d0b9 RB |
49 | #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 | } | |
1da177e4 LT |
56 | #else |
57 | /* | |
58 | * Accessing memory above the top the kernel knows about or through a file pointer | |
6b2f3d1f | 59 | * that was marked O_DSYNC will be done non-cached. |
1da177e4 | 60 | */ |
6b2f3d1f | 61 | if (file->f_flags & O_DSYNC) |
1da177e4 LT |
62 | return 1; |
63 | return addr >= __pa(high_memory); | |
64 | #endif | |
65 | } | |
66 | ||
67 | #ifndef ARCH_HAS_VALID_PHYS_ADDR_RANGE | |
136939a2 | 68 | static inline int valid_phys_addr_range(unsigned long addr, size_t count) |
1da177e4 | 69 | { |
136939a2 | 70 | if (addr + count > __pa(high_memory)) |
1da177e4 LT |
71 | return 0; |
72 | ||
1da177e4 LT |
73 | return 1; |
74 | } | |
80851ef2 | 75 | |
06c67bef | 76 | static inline int valid_mmap_phys_addr_range(unsigned long pfn, size_t size) |
80851ef2 BH |
77 | { |
78 | return 1; | |
79 | } | |
1da177e4 LT |
80 | #endif |
81 | ||
d092633b | 82 | #ifdef CONFIG_STRICT_DEVMEM |
e2beb3ea | 83 | static inline int range_is_allowed(unsigned long pfn, unsigned long size) |
ae531c26 | 84 | { |
e2beb3ea VP |
85 | u64 from = ((u64)pfn) << PAGE_SHIFT; |
86 | u64 to = from + size; | |
87 | u64 cursor = from; | |
88 | ||
89 | 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", | |
ae531c26 AV |
93 | current->comm, from, to); |
94 | return 0; | |
95 | } | |
e2beb3ea VP |
96 | cursor += PAGE_SIZE; |
97 | pfn++; | |
ae531c26 AV |
98 | } |
99 | return 1; | |
100 | } | |
101 | #else | |
e2beb3ea | 102 | static inline int range_is_allowed(unsigned long pfn, unsigned long size) |
ae531c26 AV |
103 | { |
104 | return 1; | |
105 | } | |
106 | #endif | |
107 | ||
e045fb2a | 108 | void __attribute__((weak)) unxlate_dev_mem_ptr(unsigned long phys, void *addr) |
109 | { | |
110 | } | |
111 | ||
1da177e4 LT |
112 | /* |
113 | * This funcion reads the *physical* memory. The f_pos points directly to the | |
114 | * memory location. | |
115 | */ | |
116 | static 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 | ||
136939a2 | 123 | if (!valid_phys_addr_range(p, count)) |
1da177e4 LT |
124 | 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 | ||
e045fb2a | 154 | if (!range_is_allowed(p >> PAGE_SHIFT, count)) |
155 | return -EPERM; | |
156 | ||
1da177e4 LT |
157 | /* |
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); | |
e045fb2a | 163 | if (!ptr) |
164 | return -EFAULT; | |
1da177e4 | 165 | |
e045fb2a | 166 | if (copy_to_user(buf, ptr, sz)) { |
167 | unxlate_dev_mem_ptr(p, ptr); | |
1da177e4 | 168 | return -EFAULT; |
e045fb2a | 169 | } |
170 | ||
171 | unxlate_dev_mem_ptr(p, ptr); | |
172 | ||
1da177e4 LT |
173 | buf += sz; |
174 | p += sz; | |
175 | count -= sz; | |
176 | read += sz; | |
177 | } | |
178 | ||
179 | *ppos += read; | |
180 | return read; | |
181 | } | |
182 | ||
183 | static 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 | ||
136939a2 | 191 | if (!valid_phys_addr_range(p, count)) |
1da177e4 LT |
192 | 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 | ||
e045fb2a | 221 | if (!range_is_allowed(p >> PAGE_SHIFT, sz)) |
222 | return -EPERM; | |
223 | ||
1da177e4 LT |
224 | /* |
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); | |
e045fb2a | 230 | if (!ptr) { |
231 | if (written) | |
232 | break; | |
233 | return -EFAULT; | |
234 | } | |
1da177e4 LT |
235 | |
236 | copied = copy_from_user(ptr, buf, sz); | |
237 | if (copied) { | |
c654d60e | 238 | written += sz - copied; |
e045fb2a | 239 | unxlate_dev_mem_ptr(p, ptr); |
c654d60e JB |
240 | if (written) |
241 | break; | |
1da177e4 LT |
242 | return -EFAULT; |
243 | } | |
e045fb2a | 244 | |
245 | unxlate_dev_mem_ptr(p, ptr); | |
246 | ||
1da177e4 LT |
247 | buf += sz; |
248 | p += sz; | |
249 | count -= sz; | |
250 | written += sz; | |
251 | } | |
252 | ||
253 | *ppos += written; | |
254 | return written; | |
255 | } | |
256 | ||
f0970c13 | 257 | int __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 | ||
44ac8413 BH |
263 | #ifndef __HAVE_PHYS_MEM_ACCESS_PROT |
264 | static 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 | ||
5da6185b DH |
277 | #ifndef CONFIG_MMU |
278 | static 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; | |
8a93258c | 286 | return pgoff << PAGE_SHIFT; |
5da6185b DH |
287 | } |
288 | ||
289 | /* can't do an in-place private mapping if there's no MMU */ | |
290 | static 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 | ||
297 | static inline int private_mapping_ok(struct vm_area_struct *vma) | |
298 | { | |
299 | return 1; | |
300 | } | |
301 | #endif | |
302 | ||
f0f37e2f | 303 | static const struct vm_operations_struct mmap_mem_ops = { |
7ae8ed50 RR |
304 | #ifdef CONFIG_HAVE_IOREMAP_PROT |
305 | .access = generic_access_phys | |
306 | #endif | |
e7f260a2 | 307 | }; |
308 | ||
1da177e4 LT |
309 | static int mmap_mem(struct file * file, struct vm_area_struct * vma) |
310 | { | |
80851ef2 BH |
311 | size_t size = vma->vm_end - vma->vm_start; |
312 | ||
06c67bef | 313 | if (!valid_mmap_phys_addr_range(vma->vm_pgoff, size)) |
80851ef2 BH |
314 | return -EINVAL; |
315 | ||
5da6185b DH |
316 | if (!private_mapping_ok(vma)) |
317 | return -ENOSYS; | |
318 | ||
e2beb3ea VP |
319 | if (!range_is_allowed(vma->vm_pgoff, size)) |
320 | return -EPERM; | |
321 | ||
f0970c13 | 322 | if (!phys_mem_access_prot_allowed(file, vma->vm_pgoff, size, |
323 | &vma->vm_page_prot)) | |
324 | return -EINVAL; | |
325 | ||
8b150478 | 326 | vma->vm_page_prot = phys_mem_access_prot(file, vma->vm_pgoff, |
80851ef2 | 327 | size, |
1da177e4 | 328 | vma->vm_page_prot); |
1da177e4 | 329 | |
e7f260a2 | 330 | vma->vm_ops = &mmap_mem_ops; |
331 | ||
1da177e4 LT |
332 | /* Remap-pfn-range will mark the range VM_IO and VM_RESERVED */ |
333 | if (remap_pfn_range(vma, | |
334 | vma->vm_start, | |
335 | vma->vm_pgoff, | |
80851ef2 | 336 | size, |
e7f260a2 | 337 | vma->vm_page_prot)) { |
1da177e4 | 338 | return -EAGAIN; |
e7f260a2 | 339 | } |
1da177e4 LT |
340 | return 0; |
341 | } | |
342 | ||
b781ecb6 | 343 | #ifdef CONFIG_DEVKMEM |
1da177e4 LT |
344 | static int mmap_kmem(struct file * file, struct vm_area_struct * vma) |
345 | { | |
4bb82551 LT |
346 | unsigned long pfn; |
347 | ||
6d3154cc LT |
348 | /* Turn a kernel-virtual address into a physical page frame */ |
349 | pfn = __pa((u64)vma->vm_pgoff << PAGE_SHIFT) >> PAGE_SHIFT; | |
4bb82551 | 350 | |
1da177e4 LT |
351 | /* |
352 | * RED-PEN: on some architectures there is more mapped memory | |
353 | * than available in mem_map which pfn_valid checks | |
354 | * for. Perhaps should add a new macro here. | |
355 | * | |
356 | * RED-PEN: vmalloc is not supported right now. | |
357 | */ | |
4bb82551 | 358 | if (!pfn_valid(pfn)) |
1da177e4 | 359 | return -EIO; |
4bb82551 LT |
360 | |
361 | vma->vm_pgoff = pfn; | |
1da177e4 LT |
362 | return mmap_mem(file, vma); |
363 | } | |
b781ecb6 | 364 | #endif |
1da177e4 | 365 | |
50b1fdbd VG |
366 | #ifdef CONFIG_CRASH_DUMP |
367 | /* | |
368 | * Read memory corresponding to the old kernel. | |
50b1fdbd | 369 | */ |
315c215c | 370 | static ssize_t read_oldmem(struct file *file, char __user *buf, |
50b1fdbd VG |
371 | size_t count, loff_t *ppos) |
372 | { | |
315c215c VG |
373 | unsigned long pfn, offset; |
374 | size_t read = 0, csize; | |
375 | int rc = 0; | |
50b1fdbd | 376 | |
72414d3f | 377 | while (count) { |
50b1fdbd | 378 | pfn = *ppos / PAGE_SIZE; |
315c215c VG |
379 | if (pfn > saved_max_pfn) |
380 | return read; | |
50b1fdbd | 381 | |
315c215c VG |
382 | offset = (unsigned long)(*ppos % PAGE_SIZE); |
383 | if (count > PAGE_SIZE - offset) | |
384 | csize = PAGE_SIZE - offset; | |
385 | else | |
386 | csize = count; | |
50b1fdbd | 387 | |
315c215c VG |
388 | rc = copy_oldmem_page(pfn, buf, csize, offset, 1); |
389 | if (rc < 0) | |
390 | return rc; | |
50b1fdbd VG |
391 | buf += csize; |
392 | *ppos += csize; | |
393 | read += csize; | |
394 | count -= csize; | |
395 | } | |
50b1fdbd VG |
396 | return read; |
397 | } | |
398 | #endif | |
1da177e4 | 399 | |
b781ecb6 | 400 | #ifdef CONFIG_DEVKMEM |
1da177e4 LT |
401 | /* |
402 | * This function reads the *virtual* memory as seen by the kernel. | |
403 | */ | |
404 | static ssize_t read_kmem(struct file *file, char __user *buf, | |
405 | size_t count, loff_t *ppos) | |
406 | { | |
407 | unsigned long p = *ppos; | |
408 | ssize_t low_count, read, sz; | |
409 | char * kbuf; /* k-addr because vread() takes vmlist_lock rwlock */ | |
410 | ||
411 | read = 0; | |
412 | if (p < (unsigned long) high_memory) { | |
413 | low_count = count; | |
414 | if (count > (unsigned long) high_memory - p) | |
415 | low_count = (unsigned long) high_memory - p; | |
416 | ||
417 | #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED | |
418 | /* we don't have page 0 mapped on sparc and m68k.. */ | |
419 | if (p < PAGE_SIZE && low_count > 0) { | |
420 | size_t tmp = PAGE_SIZE - p; | |
421 | if (tmp > low_count) tmp = low_count; | |
422 | if (clear_user(buf, tmp)) | |
423 | return -EFAULT; | |
424 | buf += tmp; | |
425 | p += tmp; | |
426 | read += tmp; | |
427 | low_count -= tmp; | |
428 | count -= tmp; | |
429 | } | |
430 | #endif | |
431 | while (low_count > 0) { | |
432 | /* | |
433 | * Handle first page in case it's not aligned | |
434 | */ | |
435 | if (-p & (PAGE_SIZE - 1)) | |
436 | sz = -p & (PAGE_SIZE - 1); | |
437 | else | |
438 | sz = PAGE_SIZE; | |
439 | ||
440 | sz = min_t(unsigned long, sz, low_count); | |
441 | ||
442 | /* | |
443 | * On ia64 if a page has been mapped somewhere as | |
444 | * uncached, then it must also be accessed uncached | |
445 | * by the kernel or data corruption may occur | |
446 | */ | |
447 | kbuf = xlate_dev_kmem_ptr((char *)p); | |
448 | ||
449 | if (copy_to_user(buf, kbuf, sz)) | |
450 | return -EFAULT; | |
451 | buf += sz; | |
452 | p += sz; | |
453 | read += sz; | |
454 | low_count -= sz; | |
455 | count -= sz; | |
456 | } | |
457 | } | |
458 | ||
459 | if (count > 0) { | |
460 | kbuf = (char *)__get_free_page(GFP_KERNEL); | |
461 | if (!kbuf) | |
462 | return -ENOMEM; | |
463 | while (count > 0) { | |
464 | int len = count; | |
465 | ||
466 | if (len > PAGE_SIZE) | |
467 | len = PAGE_SIZE; | |
468 | len = vread(kbuf, (char *)p, len); | |
469 | if (!len) | |
470 | break; | |
471 | if (copy_to_user(buf, kbuf, len)) { | |
472 | free_page((unsigned long)kbuf); | |
473 | return -EFAULT; | |
474 | } | |
475 | count -= len; | |
476 | buf += len; | |
477 | read += len; | |
478 | p += len; | |
479 | } | |
480 | free_page((unsigned long)kbuf); | |
481 | } | |
482 | *ppos = p; | |
483 | return read; | |
484 | } | |
485 | ||
486 | ||
487 | static inline ssize_t | |
488 | do_write_kmem(void *p, unsigned long realp, const char __user * buf, | |
489 | size_t count, loff_t *ppos) | |
490 | { | |
491 | ssize_t written, sz; | |
492 | unsigned long copied; | |
493 | ||
494 | written = 0; | |
495 | #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED | |
496 | /* we don't have page 0 mapped on sparc and m68k.. */ | |
497 | if (realp < PAGE_SIZE) { | |
498 | unsigned long sz = PAGE_SIZE - realp; | |
499 | if (sz > count) | |
500 | sz = count; | |
501 | /* Hmm. Do something? */ | |
502 | buf += sz; | |
503 | p += sz; | |
504 | realp += sz; | |
505 | count -= sz; | |
506 | written += sz; | |
507 | } | |
508 | #endif | |
509 | ||
510 | while (count > 0) { | |
511 | char *ptr; | |
512 | /* | |
513 | * Handle first page in case it's not aligned | |
514 | */ | |
515 | if (-realp & (PAGE_SIZE - 1)) | |
516 | sz = -realp & (PAGE_SIZE - 1); | |
517 | else | |
518 | sz = PAGE_SIZE; | |
519 | ||
520 | sz = min_t(unsigned long, sz, count); | |
521 | ||
522 | /* | |
523 | * On ia64 if a page has been mapped somewhere as | |
524 | * uncached, then it must also be accessed uncached | |
525 | * by the kernel or data corruption may occur | |
526 | */ | |
527 | ptr = xlate_dev_kmem_ptr(p); | |
528 | ||
529 | copied = copy_from_user(ptr, buf, sz); | |
530 | if (copied) { | |
c654d60e JB |
531 | written += sz - copied; |
532 | if (written) | |
533 | break; | |
1da177e4 LT |
534 | return -EFAULT; |
535 | } | |
536 | buf += sz; | |
537 | p += sz; | |
538 | realp += sz; | |
539 | count -= sz; | |
540 | written += sz; | |
541 | } | |
542 | ||
543 | *ppos += written; | |
544 | return written; | |
545 | } | |
546 | ||
547 | ||
548 | /* | |
549 | * This function writes to the *virtual* memory as seen by the kernel. | |
550 | */ | |
551 | static ssize_t write_kmem(struct file * file, const char __user * buf, | |
552 | size_t count, loff_t *ppos) | |
553 | { | |
554 | unsigned long p = *ppos; | |
555 | ssize_t wrote = 0; | |
556 | ssize_t virtr = 0; | |
557 | ssize_t written; | |
558 | char * kbuf; /* k-addr because vwrite() takes vmlist_lock rwlock */ | |
559 | ||
560 | if (p < (unsigned long) high_memory) { | |
561 | ||
562 | wrote = count; | |
563 | if (count > (unsigned long) high_memory - p) | |
564 | wrote = (unsigned long) high_memory - p; | |
565 | ||
566 | written = do_write_kmem((void*)p, p, buf, wrote, ppos); | |
567 | if (written != wrote) | |
568 | return written; | |
569 | wrote = written; | |
570 | p += wrote; | |
571 | buf += wrote; | |
572 | count -= wrote; | |
573 | } | |
574 | ||
575 | if (count > 0) { | |
576 | kbuf = (char *)__get_free_page(GFP_KERNEL); | |
577 | if (!kbuf) | |
578 | return wrote ? wrote : -ENOMEM; | |
579 | while (count > 0) { | |
580 | int len = count; | |
581 | ||
582 | if (len > PAGE_SIZE) | |
583 | len = PAGE_SIZE; | |
4ea2f43f WF |
584 | written = copy_from_user(kbuf, buf, len); |
585 | if (written) { | |
586 | if (wrote + virtr) | |
587 | break; | |
588 | free_page((unsigned long)kbuf); | |
589 | return -EFAULT; | |
1da177e4 LT |
590 | } |
591 | len = vwrite(kbuf, (char *)p, len); | |
592 | count -= len; | |
593 | buf += len; | |
594 | virtr += len; | |
595 | p += len; | |
596 | } | |
597 | free_page((unsigned long)kbuf); | |
598 | } | |
599 | ||
600 | *ppos = p; | |
601 | return virtr + wrote; | |
602 | } | |
b781ecb6 | 603 | #endif |
1da177e4 | 604 | |
4f911d64 | 605 | #ifdef CONFIG_DEVPORT |
1da177e4 LT |
606 | static ssize_t read_port(struct file * file, char __user * buf, |
607 | size_t count, loff_t *ppos) | |
608 | { | |
609 | unsigned long i = *ppos; | |
610 | char __user *tmp = buf; | |
611 | ||
612 | if (!access_ok(VERIFY_WRITE, buf, count)) | |
613 | return -EFAULT; | |
614 | while (count-- > 0 && i < 65536) { | |
615 | if (__put_user(inb(i),tmp) < 0) | |
616 | return -EFAULT; | |
617 | i++; | |
618 | tmp++; | |
619 | } | |
620 | *ppos = i; | |
621 | return tmp-buf; | |
622 | } | |
623 | ||
624 | static ssize_t write_port(struct file * file, const char __user * buf, | |
625 | size_t count, loff_t *ppos) | |
626 | { | |
627 | unsigned long i = *ppos; | |
628 | const char __user * tmp = buf; | |
629 | ||
630 | if (!access_ok(VERIFY_READ,buf,count)) | |
631 | return -EFAULT; | |
632 | while (count-- > 0 && i < 65536) { | |
633 | char c; | |
c654d60e JB |
634 | if (__get_user(c, tmp)) { |
635 | if (tmp > buf) | |
636 | break; | |
1da177e4 | 637 | return -EFAULT; |
c654d60e | 638 | } |
1da177e4 LT |
639 | outb(c,i); |
640 | i++; | |
641 | tmp++; | |
642 | } | |
643 | *ppos = i; | |
644 | return tmp-buf; | |
645 | } | |
646 | #endif | |
647 | ||
648 | static ssize_t read_null(struct file * file, char __user * buf, | |
649 | size_t count, loff_t *ppos) | |
650 | { | |
651 | return 0; | |
652 | } | |
653 | ||
654 | static ssize_t write_null(struct file * file, const char __user * buf, | |
655 | size_t count, loff_t *ppos) | |
656 | { | |
657 | return count; | |
658 | } | |
659 | ||
1ebd32fc JA |
660 | static int pipe_to_null(struct pipe_inode_info *info, struct pipe_buffer *buf, |
661 | struct splice_desc *sd) | |
662 | { | |
663 | return sd->len; | |
664 | } | |
665 | ||
666 | static ssize_t splice_write_null(struct pipe_inode_info *pipe,struct file *out, | |
667 | loff_t *ppos, size_t len, unsigned int flags) | |
668 | { | |
669 | return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_null); | |
670 | } | |
671 | ||
1da177e4 LT |
672 | static ssize_t read_zero(struct file * file, char __user * buf, |
673 | size_t count, loff_t *ppos) | |
674 | { | |
557ed1fa | 675 | size_t written; |
1da177e4 LT |
676 | |
677 | if (!count) | |
678 | return 0; | |
679 | ||
680 | if (!access_ok(VERIFY_WRITE, buf, count)) | |
681 | return -EFAULT; | |
682 | ||
557ed1fa NP |
683 | written = 0; |
684 | while (count) { | |
685 | unsigned long unwritten; | |
686 | size_t chunk = count; | |
1da177e4 | 687 | |
557ed1fa NP |
688 | if (chunk > PAGE_SIZE) |
689 | chunk = PAGE_SIZE; /* Just for latency reasons */ | |
bb521c5d | 690 | unwritten = __clear_user(buf, chunk); |
557ed1fa | 691 | written += chunk - unwritten; |
1da177e4 | 692 | if (unwritten) |
557ed1fa | 693 | break; |
2b838687 LT |
694 | if (signal_pending(current)) |
695 | return written ? written : -ERESTARTSYS; | |
1da177e4 | 696 | buf += chunk; |
557ed1fa | 697 | count -= chunk; |
1da177e4 LT |
698 | cond_resched(); |
699 | } | |
557ed1fa | 700 | return written ? written : -EFAULT; |
1da177e4 LT |
701 | } |
702 | ||
703 | static int mmap_zero(struct file * file, struct vm_area_struct * vma) | |
704 | { | |
557ed1fa | 705 | #ifndef CONFIG_MMU |
1da177e4 | 706 | return -ENOSYS; |
557ed1fa NP |
707 | #endif |
708 | if (vma->vm_flags & VM_SHARED) | |
709 | return shmem_zero_setup(vma); | |
710 | return 0; | |
1da177e4 | 711 | } |
1da177e4 LT |
712 | |
713 | static ssize_t write_full(struct file * file, const char __user * buf, | |
714 | size_t count, loff_t *ppos) | |
715 | { | |
716 | return -ENOSPC; | |
717 | } | |
718 | ||
719 | /* | |
720 | * Special lseek() function for /dev/null and /dev/zero. Most notably, you | |
721 | * can fopen() both devices with "a" now. This was previously impossible. | |
722 | * -- SRB. | |
723 | */ | |
724 | ||
725 | static loff_t null_lseek(struct file * file, loff_t offset, int orig) | |
726 | { | |
727 | return file->f_pos = 0; | |
728 | } | |
729 | ||
730 | /* | |
731 | * The memory devices use the full 32/64 bits of the offset, and so we cannot | |
732 | * check against negative addresses: they are ok. The return value is weird, | |
733 | * though, in that case (0). | |
734 | * | |
735 | * also note that seeking relative to the "end of file" isn't supported: | |
736 | * it has no meaning, so it returns -EINVAL. | |
737 | */ | |
738 | static loff_t memory_lseek(struct file * file, loff_t offset, int orig) | |
739 | { | |
740 | loff_t ret; | |
741 | ||
a7113a96 | 742 | mutex_lock(&file->f_path.dentry->d_inode->i_mutex); |
1da177e4 LT |
743 | switch (orig) { |
744 | case 0: | |
745 | file->f_pos = offset; | |
746 | ret = file->f_pos; | |
747 | force_successful_syscall_return(); | |
748 | break; | |
749 | case 1: | |
750 | file->f_pos += offset; | |
751 | ret = file->f_pos; | |
752 | force_successful_syscall_return(); | |
753 | break; | |
754 | default: | |
755 | ret = -EINVAL; | |
756 | } | |
a7113a96 | 757 | mutex_unlock(&file->f_path.dentry->d_inode->i_mutex); |
1da177e4 LT |
758 | return ret; |
759 | } | |
760 | ||
761 | static int open_port(struct inode * inode, struct file * filp) | |
762 | { | |
763 | return capable(CAP_SYS_RAWIO) ? 0 : -EPERM; | |
764 | } | |
765 | ||
766 | #define zero_lseek null_lseek | |
767 | #define full_lseek null_lseek | |
768 | #define write_zero write_null | |
769 | #define read_full read_zero | |
770 | #define open_mem open_port | |
771 | #define open_kmem open_mem | |
50b1fdbd | 772 | #define open_oldmem open_mem |
1da177e4 | 773 | |
62322d25 | 774 | static const struct file_operations mem_fops = { |
1da177e4 LT |
775 | .llseek = memory_lseek, |
776 | .read = read_mem, | |
777 | .write = write_mem, | |
778 | .mmap = mmap_mem, | |
779 | .open = open_mem, | |
5da6185b | 780 | .get_unmapped_area = get_unmapped_area_mem, |
1da177e4 LT |
781 | }; |
782 | ||
b781ecb6 | 783 | #ifdef CONFIG_DEVKMEM |
62322d25 | 784 | static const struct file_operations kmem_fops = { |
1da177e4 LT |
785 | .llseek = memory_lseek, |
786 | .read = read_kmem, | |
787 | .write = write_kmem, | |
788 | .mmap = mmap_kmem, | |
789 | .open = open_kmem, | |
5da6185b | 790 | .get_unmapped_area = get_unmapped_area_mem, |
1da177e4 | 791 | }; |
b781ecb6 | 792 | #endif |
1da177e4 | 793 | |
62322d25 | 794 | static const struct file_operations null_fops = { |
1da177e4 LT |
795 | .llseek = null_lseek, |
796 | .read = read_null, | |
797 | .write = write_null, | |
1ebd32fc | 798 | .splice_write = splice_write_null, |
1da177e4 LT |
799 | }; |
800 | ||
4f911d64 | 801 | #ifdef CONFIG_DEVPORT |
62322d25 | 802 | static const struct file_operations port_fops = { |
1da177e4 LT |
803 | .llseek = memory_lseek, |
804 | .read = read_port, | |
805 | .write = write_port, | |
806 | .open = open_port, | |
807 | }; | |
808 | #endif | |
809 | ||
62322d25 | 810 | static const struct file_operations zero_fops = { |
1da177e4 LT |
811 | .llseek = zero_lseek, |
812 | .read = read_zero, | |
813 | .write = write_zero, | |
814 | .mmap = mmap_zero, | |
815 | }; | |
816 | ||
5da6185b DH |
817 | /* |
818 | * capabilities for /dev/zero | |
819 | * - permits private mappings, "copies" are taken of the source of zeros | |
820 | */ | |
1da177e4 | 821 | static struct backing_dev_info zero_bdi = { |
d993831f | 822 | .name = "char/mem", |
1da177e4 LT |
823 | .capabilities = BDI_CAP_MAP_COPY, |
824 | }; | |
825 | ||
62322d25 | 826 | static const struct file_operations full_fops = { |
1da177e4 LT |
827 | .llseek = full_lseek, |
828 | .read = read_full, | |
829 | .write = write_full, | |
830 | }; | |
831 | ||
50b1fdbd | 832 | #ifdef CONFIG_CRASH_DUMP |
62322d25 | 833 | static const struct file_operations oldmem_fops = { |
50b1fdbd VG |
834 | .read = read_oldmem, |
835 | .open = open_oldmem, | |
836 | }; | |
837 | #endif | |
838 | ||
1da177e4 LT |
839 | static ssize_t kmsg_write(struct file * file, const char __user * buf, |
840 | size_t count, loff_t *ppos) | |
841 | { | |
842 | char *tmp; | |
cd140a5c | 843 | ssize_t ret; |
1da177e4 LT |
844 | |
845 | tmp = kmalloc(count + 1, GFP_KERNEL); | |
846 | if (tmp == NULL) | |
847 | return -ENOMEM; | |
848 | ret = -EFAULT; | |
849 | if (!copy_from_user(tmp, buf, count)) { | |
850 | tmp[count] = 0; | |
851 | ret = printk("%s", tmp); | |
cd140a5c GC |
852 | if (ret > count) |
853 | /* printk can add a prefix */ | |
854 | ret = count; | |
1da177e4 LT |
855 | } |
856 | kfree(tmp); | |
857 | return ret; | |
858 | } | |
859 | ||
62322d25 | 860 | static const struct file_operations kmsg_fops = { |
1da177e4 LT |
861 | .write = kmsg_write, |
862 | }; | |
863 | ||
389e0cb9 KS |
864 | static const struct memdev { |
865 | const char *name; | |
e454cea2 | 866 | mode_t mode; |
389e0cb9 KS |
867 | const struct file_operations *fops; |
868 | struct backing_dev_info *dev_info; | |
869 | } devlist[] = { | |
e454cea2 | 870 | [1] = { "mem", 0, &mem_fops, &directly_mappable_cdev_bdi }, |
b781ecb6 | 871 | #ifdef CONFIG_DEVKMEM |
e454cea2 | 872 | [2] = { "kmem", 0, &kmem_fops, &directly_mappable_cdev_bdi }, |
b781ecb6 | 873 | #endif |
e454cea2 | 874 | [3] = { "null", 0666, &null_fops, NULL }, |
4f911d64 | 875 | #ifdef CONFIG_DEVPORT |
e454cea2 | 876 | [4] = { "port", 0, &port_fops, NULL }, |
1da177e4 | 877 | #endif |
e454cea2 KS |
878 | [5] = { "zero", 0666, &zero_fops, &zero_bdi }, |
879 | [7] = { "full", 0666, &full_fops, NULL }, | |
880 | [8] = { "random", 0666, &random_fops, NULL }, | |
881 | [9] = { "urandom", 0666, &urandom_fops, NULL }, | |
882 | [11] = { "kmsg", 0, &kmsg_fops, NULL }, | |
50b1fdbd | 883 | #ifdef CONFIG_CRASH_DUMP |
e454cea2 | 884 | [12] = { "oldmem", 0, &oldmem_fops, NULL }, |
50b1fdbd | 885 | #endif |
d6f47bef ASF |
886 | }; |
887 | ||
888 | static int memory_open(struct inode *inode, struct file *filp) | |
889 | { | |
389e0cb9 KS |
890 | int minor; |
891 | const struct memdev *dev; | |
d6f47bef | 892 | |
389e0cb9 KS |
893 | minor = iminor(inode); |
894 | if (minor >= ARRAY_SIZE(devlist)) | |
205153aa | 895 | return -ENXIO; |
d6f47bef | 896 | |
389e0cb9 KS |
897 | dev = &devlist[minor]; |
898 | if (!dev->fops) | |
205153aa | 899 | return -ENXIO; |
d6f47bef | 900 | |
389e0cb9 KS |
901 | filp->f_op = dev->fops; |
902 | if (dev->dev_info) | |
903 | filp->f_mapping->backing_dev_info = dev->dev_info; | |
d6f47bef | 904 | |
389e0cb9 | 905 | if (dev->fops->open) |
205153aa FW |
906 | return dev->fops->open(inode, filp); |
907 | ||
908 | return 0; | |
1da177e4 LT |
909 | } |
910 | ||
62322d25 | 911 | static const struct file_operations memory_fops = { |
389e0cb9 | 912 | .open = memory_open, |
1da177e4 LT |
913 | }; |
914 | ||
e454cea2 KS |
915 | static char *mem_devnode(struct device *dev, mode_t *mode) |
916 | { | |
917 | if (mode && devlist[MINOR(dev->devt)].mode) | |
918 | *mode = devlist[MINOR(dev->devt)].mode; | |
919 | return NULL; | |
920 | } | |
921 | ||
ca8eca68 | 922 | static struct class *mem_class; |
1da177e4 LT |
923 | |
924 | static int __init chr_dev_init(void) | |
925 | { | |
389e0cb9 | 926 | int minor; |
e0bf68dd PZ |
927 | int err; |
928 | ||
929 | err = bdi_init(&zero_bdi); | |
930 | if (err) | |
931 | return err; | |
1da177e4 LT |
932 | |
933 | if (register_chrdev(MEM_MAJOR,"mem",&memory_fops)) | |
934 | printk("unable to get major %d for memory devs\n", MEM_MAJOR); | |
935 | ||
ca8eca68 | 936 | mem_class = class_create(THIS_MODULE, "mem"); |
e454cea2 | 937 | mem_class->devnode = mem_devnode; |
389e0cb9 KS |
938 | for (minor = 1; minor < ARRAY_SIZE(devlist); minor++) { |
939 | if (!devlist[minor].name) | |
940 | continue; | |
941 | device_create(mem_class, NULL, MKDEV(MEM_MAJOR, minor), | |
942 | NULL, devlist[minor].name); | |
943 | } | |
ebf644c4 | 944 | |
1da177e4 LT |
945 | return 0; |
946 | } | |
947 | ||
948 | fs_initcall(chr_dev_init); |