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 | ||
f222318e WF |
37 | static inline unsigned long size_inside_page(unsigned long start, |
38 | unsigned long size) | |
39 | { | |
40 | unsigned long sz; | |
41 | ||
7fabaddd | 42 | sz = PAGE_SIZE - (start & (PAGE_SIZE - 1)); |
f222318e | 43 | |
7fabaddd | 44 | return min(sz, size); |
f222318e WF |
45 | } |
46 | ||
1da177e4 LT |
47 | /* |
48 | * Architectures vary in how they handle caching for addresses | |
49 | * outside of main memory. | |
50 | * | |
51 | */ | |
52 | static inline int uncached_access(struct file *file, unsigned long addr) | |
53 | { | |
f0970c13 | 54 | #if defined(CONFIG_IA64) |
1da177e4 | 55 | /* |
6b2f3d1f | 56 | * On ia64, we ignore O_DSYNC because we cannot tolerate memory attribute aliases. |
1da177e4 LT |
57 | */ |
58 | return !(efi_mem_attributes(addr) & EFI_MEMORY_WB); | |
24e9d0b9 RB |
59 | #elif defined(CONFIG_MIPS) |
60 | { | |
61 | extern int __uncached_access(struct file *file, | |
62 | unsigned long addr); | |
63 | ||
64 | return __uncached_access(file, addr); | |
65 | } | |
1da177e4 LT |
66 | #else |
67 | /* | |
68 | * Accessing memory above the top the kernel knows about or through a file pointer | |
6b2f3d1f | 69 | * that was marked O_DSYNC will be done non-cached. |
1da177e4 | 70 | */ |
6b2f3d1f | 71 | if (file->f_flags & O_DSYNC) |
1da177e4 LT |
72 | return 1; |
73 | return addr >= __pa(high_memory); | |
74 | #endif | |
75 | } | |
76 | ||
77 | #ifndef ARCH_HAS_VALID_PHYS_ADDR_RANGE | |
136939a2 | 78 | static inline int valid_phys_addr_range(unsigned long addr, size_t count) |
1da177e4 | 79 | { |
136939a2 | 80 | if (addr + count > __pa(high_memory)) |
1da177e4 LT |
81 | return 0; |
82 | ||
1da177e4 LT |
83 | return 1; |
84 | } | |
80851ef2 | 85 | |
06c67bef | 86 | static inline int valid_mmap_phys_addr_range(unsigned long pfn, size_t size) |
80851ef2 BH |
87 | { |
88 | return 1; | |
89 | } | |
1da177e4 LT |
90 | #endif |
91 | ||
d092633b | 92 | #ifdef CONFIG_STRICT_DEVMEM |
e2beb3ea | 93 | static inline int range_is_allowed(unsigned long pfn, unsigned long size) |
ae531c26 | 94 | { |
e2beb3ea VP |
95 | u64 from = ((u64)pfn) << PAGE_SHIFT; |
96 | u64 to = from + size; | |
97 | u64 cursor = from; | |
98 | ||
99 | while (cursor < to) { | |
100 | if (!devmem_is_allowed(pfn)) { | |
101 | printk(KERN_INFO | |
102 | "Program %s tried to access /dev/mem between %Lx->%Lx.\n", | |
ae531c26 AV |
103 | current->comm, from, to); |
104 | return 0; | |
105 | } | |
e2beb3ea VP |
106 | cursor += PAGE_SIZE; |
107 | pfn++; | |
ae531c26 AV |
108 | } |
109 | return 1; | |
110 | } | |
111 | #else | |
e2beb3ea | 112 | static inline int range_is_allowed(unsigned long pfn, unsigned long size) |
ae531c26 AV |
113 | { |
114 | return 1; | |
115 | } | |
116 | #endif | |
117 | ||
e045fb2a | 118 | void __attribute__((weak)) unxlate_dev_mem_ptr(unsigned long phys, void *addr) |
119 | { | |
120 | } | |
121 | ||
1da177e4 LT |
122 | /* |
123 | * This funcion reads the *physical* memory. The f_pos points directly to the | |
124 | * memory location. | |
125 | */ | |
126 | static ssize_t read_mem(struct file * file, char __user * buf, | |
127 | size_t count, loff_t *ppos) | |
128 | { | |
129 | unsigned long p = *ppos; | |
130 | ssize_t read, sz; | |
131 | char *ptr; | |
132 | ||
136939a2 | 133 | if (!valid_phys_addr_range(p, count)) |
1da177e4 LT |
134 | return -EFAULT; |
135 | read = 0; | |
136 | #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED | |
137 | /* we don't have page 0 mapped on sparc and m68k.. */ | |
138 | if (p < PAGE_SIZE) { | |
7fabaddd | 139 | sz = size_inside_page(p, count); |
1da177e4 LT |
140 | if (sz > 0) { |
141 | if (clear_user(buf, sz)) | |
142 | return -EFAULT; | |
143 | buf += sz; | |
144 | p += sz; | |
145 | count -= sz; | |
146 | read += sz; | |
147 | } | |
148 | } | |
149 | #endif | |
150 | ||
151 | while (count > 0) { | |
fa29e97b WF |
152 | unsigned long remaining; |
153 | ||
f222318e | 154 | sz = size_inside_page(p, count); |
1da177e4 | 155 | |
e045fb2a | 156 | if (!range_is_allowed(p >> PAGE_SHIFT, count)) |
157 | return -EPERM; | |
158 | ||
1da177e4 LT |
159 | /* |
160 | * On ia64 if a page has been mapped somewhere as | |
161 | * uncached, then it must also be accessed uncached | |
162 | * by the kernel or data corruption may occur | |
163 | */ | |
164 | ptr = xlate_dev_mem_ptr(p); | |
e045fb2a | 165 | if (!ptr) |
166 | return -EFAULT; | |
1da177e4 | 167 | |
fa29e97b | 168 | remaining = copy_to_user(buf, ptr, sz); |
e045fb2a | 169 | unxlate_dev_mem_ptr(p, ptr); |
fa29e97b WF |
170 | if (remaining) |
171 | return -EFAULT; | |
e045fb2a | 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) { | |
7fabaddd | 199 | sz = size_inside_page(p, count); |
1da177e4 LT |
200 | /* Hmm. Do something? */ |
201 | buf += sz; | |
202 | p += sz; | |
203 | count -= sz; | |
204 | written += sz; | |
205 | } | |
206 | #endif | |
207 | ||
208 | while (count > 0) { | |
f222318e | 209 | sz = size_inside_page(p, count); |
1da177e4 | 210 | |
e045fb2a | 211 | if (!range_is_allowed(p >> PAGE_SHIFT, sz)) |
212 | return -EPERM; | |
213 | ||
1da177e4 LT |
214 | /* |
215 | * On ia64 if a page has been mapped somewhere as | |
216 | * uncached, then it must also be accessed uncached | |
217 | * by the kernel or data corruption may occur | |
218 | */ | |
219 | ptr = xlate_dev_mem_ptr(p); | |
e045fb2a | 220 | if (!ptr) { |
221 | if (written) | |
222 | break; | |
223 | return -EFAULT; | |
224 | } | |
1da177e4 LT |
225 | |
226 | copied = copy_from_user(ptr, buf, sz); | |
fa29e97b | 227 | unxlate_dev_mem_ptr(p, ptr); |
1da177e4 | 228 | if (copied) { |
c654d60e JB |
229 | written += sz - copied; |
230 | if (written) | |
231 | break; | |
1da177e4 LT |
232 | return -EFAULT; |
233 | } | |
e045fb2a | 234 | |
1da177e4 LT |
235 | buf += sz; |
236 | p += sz; | |
237 | count -= sz; | |
238 | written += sz; | |
239 | } | |
240 | ||
241 | *ppos += written; | |
242 | return written; | |
243 | } | |
244 | ||
f0970c13 | 245 | int __attribute__((weak)) phys_mem_access_prot_allowed(struct file *file, |
246 | unsigned long pfn, unsigned long size, pgprot_t *vma_prot) | |
247 | { | |
248 | return 1; | |
249 | } | |
250 | ||
44ac8413 BH |
251 | #ifndef __HAVE_PHYS_MEM_ACCESS_PROT |
252 | static pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn, | |
253 | unsigned long size, pgprot_t vma_prot) | |
254 | { | |
255 | #ifdef pgprot_noncached | |
256 | unsigned long offset = pfn << PAGE_SHIFT; | |
257 | ||
258 | if (uncached_access(file, offset)) | |
259 | return pgprot_noncached(vma_prot); | |
260 | #endif | |
261 | return vma_prot; | |
262 | } | |
263 | #endif | |
264 | ||
5da6185b DH |
265 | #ifndef CONFIG_MMU |
266 | static unsigned long get_unmapped_area_mem(struct file *file, | |
267 | unsigned long addr, | |
268 | unsigned long len, | |
269 | unsigned long pgoff, | |
270 | unsigned long flags) | |
271 | { | |
272 | if (!valid_mmap_phys_addr_range(pgoff, len)) | |
273 | return (unsigned long) -EINVAL; | |
8a93258c | 274 | return pgoff << PAGE_SHIFT; |
5da6185b DH |
275 | } |
276 | ||
277 | /* can't do an in-place private mapping if there's no MMU */ | |
278 | static inline int private_mapping_ok(struct vm_area_struct *vma) | |
279 | { | |
280 | return vma->vm_flags & VM_MAYSHARE; | |
281 | } | |
282 | #else | |
283 | #define get_unmapped_area_mem NULL | |
284 | ||
285 | static inline int private_mapping_ok(struct vm_area_struct *vma) | |
286 | { | |
287 | return 1; | |
288 | } | |
289 | #endif | |
290 | ||
f0f37e2f | 291 | static const struct vm_operations_struct mmap_mem_ops = { |
7ae8ed50 RR |
292 | #ifdef CONFIG_HAVE_IOREMAP_PROT |
293 | .access = generic_access_phys | |
294 | #endif | |
e7f260a2 | 295 | }; |
296 | ||
1da177e4 LT |
297 | static int mmap_mem(struct file * file, struct vm_area_struct * vma) |
298 | { | |
80851ef2 BH |
299 | size_t size = vma->vm_end - vma->vm_start; |
300 | ||
06c67bef | 301 | if (!valid_mmap_phys_addr_range(vma->vm_pgoff, size)) |
80851ef2 BH |
302 | return -EINVAL; |
303 | ||
5da6185b DH |
304 | if (!private_mapping_ok(vma)) |
305 | return -ENOSYS; | |
306 | ||
e2beb3ea VP |
307 | if (!range_is_allowed(vma->vm_pgoff, size)) |
308 | return -EPERM; | |
309 | ||
f0970c13 | 310 | if (!phys_mem_access_prot_allowed(file, vma->vm_pgoff, size, |
311 | &vma->vm_page_prot)) | |
312 | return -EINVAL; | |
313 | ||
8b150478 | 314 | vma->vm_page_prot = phys_mem_access_prot(file, vma->vm_pgoff, |
80851ef2 | 315 | size, |
1da177e4 | 316 | vma->vm_page_prot); |
1da177e4 | 317 | |
e7f260a2 | 318 | vma->vm_ops = &mmap_mem_ops; |
319 | ||
1da177e4 LT |
320 | /* Remap-pfn-range will mark the range VM_IO and VM_RESERVED */ |
321 | if (remap_pfn_range(vma, | |
322 | vma->vm_start, | |
323 | vma->vm_pgoff, | |
80851ef2 | 324 | size, |
e7f260a2 | 325 | vma->vm_page_prot)) { |
1da177e4 | 326 | return -EAGAIN; |
e7f260a2 | 327 | } |
1da177e4 LT |
328 | return 0; |
329 | } | |
330 | ||
b781ecb6 | 331 | #ifdef CONFIG_DEVKMEM |
1da177e4 LT |
332 | static int mmap_kmem(struct file * file, struct vm_area_struct * vma) |
333 | { | |
4bb82551 LT |
334 | unsigned long pfn; |
335 | ||
6d3154cc LT |
336 | /* Turn a kernel-virtual address into a physical page frame */ |
337 | pfn = __pa((u64)vma->vm_pgoff << PAGE_SHIFT) >> PAGE_SHIFT; | |
4bb82551 | 338 | |
1da177e4 LT |
339 | /* |
340 | * RED-PEN: on some architectures there is more mapped memory | |
341 | * than available in mem_map which pfn_valid checks | |
342 | * for. Perhaps should add a new macro here. | |
343 | * | |
344 | * RED-PEN: vmalloc is not supported right now. | |
345 | */ | |
4bb82551 | 346 | if (!pfn_valid(pfn)) |
1da177e4 | 347 | return -EIO; |
4bb82551 LT |
348 | |
349 | vma->vm_pgoff = pfn; | |
1da177e4 LT |
350 | return mmap_mem(file, vma); |
351 | } | |
b781ecb6 | 352 | #endif |
1da177e4 | 353 | |
50b1fdbd VG |
354 | #ifdef CONFIG_CRASH_DUMP |
355 | /* | |
356 | * Read memory corresponding to the old kernel. | |
50b1fdbd | 357 | */ |
315c215c | 358 | static ssize_t read_oldmem(struct file *file, char __user *buf, |
50b1fdbd VG |
359 | size_t count, loff_t *ppos) |
360 | { | |
315c215c VG |
361 | unsigned long pfn, offset; |
362 | size_t read = 0, csize; | |
363 | int rc = 0; | |
50b1fdbd | 364 | |
72414d3f | 365 | while (count) { |
50b1fdbd | 366 | pfn = *ppos / PAGE_SIZE; |
315c215c VG |
367 | if (pfn > saved_max_pfn) |
368 | return read; | |
50b1fdbd | 369 | |
315c215c VG |
370 | offset = (unsigned long)(*ppos % PAGE_SIZE); |
371 | if (count > PAGE_SIZE - offset) | |
372 | csize = PAGE_SIZE - offset; | |
373 | else | |
374 | csize = count; | |
50b1fdbd | 375 | |
315c215c VG |
376 | rc = copy_oldmem_page(pfn, buf, csize, offset, 1); |
377 | if (rc < 0) | |
378 | return rc; | |
50b1fdbd VG |
379 | buf += csize; |
380 | *ppos += csize; | |
381 | read += csize; | |
382 | count -= csize; | |
383 | } | |
50b1fdbd VG |
384 | return read; |
385 | } | |
386 | #endif | |
1da177e4 | 387 | |
b781ecb6 | 388 | #ifdef CONFIG_DEVKMEM |
1da177e4 LT |
389 | /* |
390 | * This function reads the *virtual* memory as seen by the kernel. | |
391 | */ | |
392 | static ssize_t read_kmem(struct file *file, char __user *buf, | |
393 | size_t count, loff_t *ppos) | |
394 | { | |
395 | unsigned long p = *ppos; | |
396 | ssize_t low_count, read, sz; | |
397 | char * kbuf; /* k-addr because vread() takes vmlist_lock rwlock */ | |
325fda71 | 398 | int err = 0; |
1da177e4 LT |
399 | |
400 | read = 0; | |
401 | if (p < (unsigned long) high_memory) { | |
402 | low_count = count; | |
403 | if (count > (unsigned long) high_memory - p) | |
404 | low_count = (unsigned long) high_memory - p; | |
405 | ||
406 | #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED | |
407 | /* we don't have page 0 mapped on sparc and m68k.. */ | |
408 | if (p < PAGE_SIZE && low_count > 0) { | |
7fabaddd WF |
409 | sz = size_inside_page(p, low_count); |
410 | if (clear_user(buf, sz)) | |
1da177e4 | 411 | return -EFAULT; |
7fabaddd WF |
412 | buf += sz; |
413 | p += sz; | |
414 | read += sz; | |
415 | low_count -= sz; | |
416 | count -= sz; | |
1da177e4 LT |
417 | } |
418 | #endif | |
419 | while (low_count > 0) { | |
f222318e | 420 | sz = size_inside_page(p, low_count); |
1da177e4 LT |
421 | |
422 | /* | |
423 | * On ia64 if a page has been mapped somewhere as | |
424 | * uncached, then it must also be accessed uncached | |
425 | * by the kernel or data corruption may occur | |
426 | */ | |
427 | kbuf = xlate_dev_kmem_ptr((char *)p); | |
428 | ||
429 | if (copy_to_user(buf, kbuf, sz)) | |
430 | return -EFAULT; | |
431 | buf += sz; | |
432 | p += sz; | |
433 | read += sz; | |
434 | low_count -= sz; | |
435 | count -= sz; | |
436 | } | |
437 | } | |
438 | ||
439 | if (count > 0) { | |
440 | kbuf = (char *)__get_free_page(GFP_KERNEL); | |
441 | if (!kbuf) | |
442 | return -ENOMEM; | |
443 | while (count > 0) { | |
80ad89a0 | 444 | sz = size_inside_page(p, count); |
325fda71 KH |
445 | if (!is_vmalloc_or_module_addr((void *)p)) { |
446 | err = -ENXIO; | |
447 | break; | |
448 | } | |
80ad89a0 WF |
449 | sz = vread(kbuf, (char *)p, sz); |
450 | if (!sz) | |
1da177e4 | 451 | break; |
80ad89a0 | 452 | if (copy_to_user(buf, kbuf, sz)) { |
325fda71 KH |
453 | err = -EFAULT; |
454 | break; | |
1da177e4 | 455 | } |
80ad89a0 WF |
456 | count -= sz; |
457 | buf += sz; | |
458 | read += sz; | |
459 | p += sz; | |
1da177e4 LT |
460 | } |
461 | free_page((unsigned long)kbuf); | |
462 | } | |
325fda71 KH |
463 | *ppos = p; |
464 | return read ? read : err; | |
1da177e4 LT |
465 | } |
466 | ||
467 | ||
468 | static inline ssize_t | |
ee32398f | 469 | do_write_kmem(unsigned long p, const char __user *buf, |
1da177e4 LT |
470 | size_t count, loff_t *ppos) |
471 | { | |
472 | ssize_t written, sz; | |
473 | unsigned long copied; | |
474 | ||
475 | written = 0; | |
476 | #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED | |
477 | /* we don't have page 0 mapped on sparc and m68k.. */ | |
ee32398f WF |
478 | if (p < PAGE_SIZE) { |
479 | sz = size_inside_page(p, count); | |
1da177e4 LT |
480 | /* Hmm. Do something? */ |
481 | buf += sz; | |
482 | p += sz; | |
1da177e4 LT |
483 | count -= sz; |
484 | written += sz; | |
485 | } | |
486 | #endif | |
487 | ||
488 | while (count > 0) { | |
489 | char *ptr; | |
1da177e4 | 490 | |
ee32398f | 491 | sz = size_inside_page(p, count); |
1da177e4 LT |
492 | |
493 | /* | |
494 | * On ia64 if a page has been mapped somewhere as | |
495 | * uncached, then it must also be accessed uncached | |
496 | * by the kernel or data corruption may occur | |
497 | */ | |
ee32398f | 498 | ptr = xlate_dev_kmem_ptr((char *)p); |
1da177e4 LT |
499 | |
500 | copied = copy_from_user(ptr, buf, sz); | |
501 | if (copied) { | |
c654d60e JB |
502 | written += sz - copied; |
503 | if (written) | |
504 | break; | |
1da177e4 LT |
505 | return -EFAULT; |
506 | } | |
507 | buf += sz; | |
508 | p += sz; | |
1da177e4 LT |
509 | count -= sz; |
510 | written += sz; | |
511 | } | |
512 | ||
513 | *ppos += written; | |
514 | return written; | |
515 | } | |
516 | ||
517 | ||
518 | /* | |
519 | * This function writes to the *virtual* memory as seen by the kernel. | |
520 | */ | |
521 | static ssize_t write_kmem(struct file * file, const char __user * buf, | |
522 | size_t count, loff_t *ppos) | |
523 | { | |
524 | unsigned long p = *ppos; | |
525 | ssize_t wrote = 0; | |
526 | ssize_t virtr = 0; | |
1da177e4 | 527 | char * kbuf; /* k-addr because vwrite() takes vmlist_lock rwlock */ |
325fda71 | 528 | int err = 0; |
1da177e4 LT |
529 | |
530 | if (p < (unsigned long) high_memory) { | |
80ad89a0 WF |
531 | unsigned long to_write = min_t(unsigned long, count, |
532 | (unsigned long)high_memory - p); | |
ee32398f | 533 | wrote = do_write_kmem(p, buf, to_write, ppos); |
80ad89a0 WF |
534 | if (wrote != to_write) |
535 | return wrote; | |
1da177e4 LT |
536 | p += wrote; |
537 | buf += wrote; | |
538 | count -= wrote; | |
539 | } | |
540 | ||
541 | if (count > 0) { | |
542 | kbuf = (char *)__get_free_page(GFP_KERNEL); | |
543 | if (!kbuf) | |
544 | return wrote ? wrote : -ENOMEM; | |
545 | while (count > 0) { | |
80ad89a0 WF |
546 | unsigned long sz = size_inside_page(p, count); |
547 | unsigned long n; | |
1da177e4 | 548 | |
325fda71 KH |
549 | if (!is_vmalloc_or_module_addr((void *)p)) { |
550 | err = -ENXIO; | |
551 | break; | |
552 | } | |
80ad89a0 WF |
553 | n = copy_from_user(kbuf, buf, sz); |
554 | if (n) { | |
325fda71 KH |
555 | err = -EFAULT; |
556 | break; | |
1da177e4 | 557 | } |
c85e9a97 | 558 | vwrite(kbuf, (char *)p, sz); |
80ad89a0 WF |
559 | count -= sz; |
560 | buf += sz; | |
561 | virtr += sz; | |
562 | p += sz; | |
1da177e4 LT |
563 | } |
564 | free_page((unsigned long)kbuf); | |
565 | } | |
566 | ||
325fda71 KH |
567 | *ppos = p; |
568 | return virtr + wrote ? : err; | |
1da177e4 | 569 | } |
b781ecb6 | 570 | #endif |
1da177e4 | 571 | |
4f911d64 | 572 | #ifdef CONFIG_DEVPORT |
1da177e4 LT |
573 | static ssize_t read_port(struct file * file, char __user * buf, |
574 | size_t count, loff_t *ppos) | |
575 | { | |
576 | unsigned long i = *ppos; | |
577 | char __user *tmp = buf; | |
578 | ||
579 | if (!access_ok(VERIFY_WRITE, buf, count)) | |
580 | return -EFAULT; | |
581 | while (count-- > 0 && i < 65536) { | |
582 | if (__put_user(inb(i),tmp) < 0) | |
583 | return -EFAULT; | |
584 | i++; | |
585 | tmp++; | |
586 | } | |
587 | *ppos = i; | |
588 | return tmp-buf; | |
589 | } | |
590 | ||
591 | static ssize_t write_port(struct file * file, const char __user * buf, | |
592 | size_t count, loff_t *ppos) | |
593 | { | |
594 | unsigned long i = *ppos; | |
595 | const char __user * tmp = buf; | |
596 | ||
597 | if (!access_ok(VERIFY_READ,buf,count)) | |
598 | return -EFAULT; | |
599 | while (count-- > 0 && i < 65536) { | |
600 | char c; | |
c654d60e JB |
601 | if (__get_user(c, tmp)) { |
602 | if (tmp > buf) | |
603 | break; | |
1da177e4 | 604 | return -EFAULT; |
c654d60e | 605 | } |
1da177e4 LT |
606 | outb(c,i); |
607 | i++; | |
608 | tmp++; | |
609 | } | |
610 | *ppos = i; | |
611 | return tmp-buf; | |
612 | } | |
613 | #endif | |
614 | ||
615 | static ssize_t read_null(struct file * file, char __user * buf, | |
616 | size_t count, loff_t *ppos) | |
617 | { | |
618 | return 0; | |
619 | } | |
620 | ||
621 | static ssize_t write_null(struct file * file, const char __user * buf, | |
622 | size_t count, loff_t *ppos) | |
623 | { | |
624 | return count; | |
625 | } | |
626 | ||
1ebd32fc JA |
627 | static int pipe_to_null(struct pipe_inode_info *info, struct pipe_buffer *buf, |
628 | struct splice_desc *sd) | |
629 | { | |
630 | return sd->len; | |
631 | } | |
632 | ||
633 | static ssize_t splice_write_null(struct pipe_inode_info *pipe,struct file *out, | |
634 | loff_t *ppos, size_t len, unsigned int flags) | |
635 | { | |
636 | return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_null); | |
637 | } | |
638 | ||
1da177e4 LT |
639 | static ssize_t read_zero(struct file * file, char __user * buf, |
640 | size_t count, loff_t *ppos) | |
641 | { | |
557ed1fa | 642 | size_t written; |
1da177e4 LT |
643 | |
644 | if (!count) | |
645 | return 0; | |
646 | ||
647 | if (!access_ok(VERIFY_WRITE, buf, count)) | |
648 | return -EFAULT; | |
649 | ||
557ed1fa NP |
650 | written = 0; |
651 | while (count) { | |
652 | unsigned long unwritten; | |
653 | size_t chunk = count; | |
1da177e4 | 654 | |
557ed1fa NP |
655 | if (chunk > PAGE_SIZE) |
656 | chunk = PAGE_SIZE; /* Just for latency reasons */ | |
bb521c5d | 657 | unwritten = __clear_user(buf, chunk); |
557ed1fa | 658 | written += chunk - unwritten; |
1da177e4 | 659 | if (unwritten) |
557ed1fa | 660 | break; |
2b838687 LT |
661 | if (signal_pending(current)) |
662 | return written ? written : -ERESTARTSYS; | |
1da177e4 | 663 | buf += chunk; |
557ed1fa | 664 | count -= chunk; |
1da177e4 LT |
665 | cond_resched(); |
666 | } | |
557ed1fa | 667 | return written ? written : -EFAULT; |
1da177e4 LT |
668 | } |
669 | ||
670 | static int mmap_zero(struct file * file, struct vm_area_struct * vma) | |
671 | { | |
557ed1fa | 672 | #ifndef CONFIG_MMU |
1da177e4 | 673 | return -ENOSYS; |
557ed1fa NP |
674 | #endif |
675 | if (vma->vm_flags & VM_SHARED) | |
676 | return shmem_zero_setup(vma); | |
677 | return 0; | |
1da177e4 | 678 | } |
1da177e4 LT |
679 | |
680 | static ssize_t write_full(struct file * file, const char __user * buf, | |
681 | size_t count, loff_t *ppos) | |
682 | { | |
683 | return -ENOSPC; | |
684 | } | |
685 | ||
686 | /* | |
687 | * Special lseek() function for /dev/null and /dev/zero. Most notably, you | |
688 | * can fopen() both devices with "a" now. This was previously impossible. | |
689 | * -- SRB. | |
690 | */ | |
691 | ||
692 | static loff_t null_lseek(struct file * file, loff_t offset, int orig) | |
693 | { | |
694 | return file->f_pos = 0; | |
695 | } | |
696 | ||
697 | /* | |
698 | * The memory devices use the full 32/64 bits of the offset, and so we cannot | |
699 | * check against negative addresses: they are ok. The return value is weird, | |
700 | * though, in that case (0). | |
701 | * | |
702 | * also note that seeking relative to the "end of file" isn't supported: | |
703 | * it has no meaning, so it returns -EINVAL. | |
704 | */ | |
705 | static loff_t memory_lseek(struct file * file, loff_t offset, int orig) | |
706 | { | |
707 | loff_t ret; | |
708 | ||
a7113a96 | 709 | mutex_lock(&file->f_path.dentry->d_inode->i_mutex); |
1da177e4 | 710 | switch (orig) { |
dcefafb6 WF |
711 | case SEEK_CUR: |
712 | offset += file->f_pos; | |
713 | if ((unsigned long long)offset < | |
714 | (unsigned long long)file->f_pos) { | |
715 | ret = -EOVERFLOW; | |
716 | break; | |
717 | } | |
718 | case SEEK_SET: | |
719 | /* to avoid userland mistaking f_pos=-9 as -EBADF=-9 */ | |
720 | if ((unsigned long long)offset >= ~0xFFFULL) { | |
721 | ret = -EOVERFLOW; | |
722 | break; | |
723 | } | |
1da177e4 LT |
724 | file->f_pos = offset; |
725 | ret = file->f_pos; | |
726 | force_successful_syscall_return(); | |
727 | break; | |
1da177e4 LT |
728 | default: |
729 | ret = -EINVAL; | |
730 | } | |
a7113a96 | 731 | mutex_unlock(&file->f_path.dentry->d_inode->i_mutex); |
1da177e4 LT |
732 | return ret; |
733 | } | |
734 | ||
735 | static int open_port(struct inode * inode, struct file * filp) | |
736 | { | |
737 | return capable(CAP_SYS_RAWIO) ? 0 : -EPERM; | |
738 | } | |
739 | ||
740 | #define zero_lseek null_lseek | |
741 | #define full_lseek null_lseek | |
742 | #define write_zero write_null | |
743 | #define read_full read_zero | |
744 | #define open_mem open_port | |
745 | #define open_kmem open_mem | |
50b1fdbd | 746 | #define open_oldmem open_mem |
1da177e4 | 747 | |
62322d25 | 748 | static const struct file_operations mem_fops = { |
1da177e4 LT |
749 | .llseek = memory_lseek, |
750 | .read = read_mem, | |
751 | .write = write_mem, | |
752 | .mmap = mmap_mem, | |
753 | .open = open_mem, | |
5da6185b | 754 | .get_unmapped_area = get_unmapped_area_mem, |
1da177e4 LT |
755 | }; |
756 | ||
b781ecb6 | 757 | #ifdef CONFIG_DEVKMEM |
62322d25 | 758 | static const struct file_operations kmem_fops = { |
1da177e4 LT |
759 | .llseek = memory_lseek, |
760 | .read = read_kmem, | |
761 | .write = write_kmem, | |
762 | .mmap = mmap_kmem, | |
763 | .open = open_kmem, | |
5da6185b | 764 | .get_unmapped_area = get_unmapped_area_mem, |
1da177e4 | 765 | }; |
b781ecb6 | 766 | #endif |
1da177e4 | 767 | |
62322d25 | 768 | static const struct file_operations null_fops = { |
1da177e4 LT |
769 | .llseek = null_lseek, |
770 | .read = read_null, | |
771 | .write = write_null, | |
1ebd32fc | 772 | .splice_write = splice_write_null, |
1da177e4 LT |
773 | }; |
774 | ||
4f911d64 | 775 | #ifdef CONFIG_DEVPORT |
62322d25 | 776 | static const struct file_operations port_fops = { |
1da177e4 LT |
777 | .llseek = memory_lseek, |
778 | .read = read_port, | |
779 | .write = write_port, | |
780 | .open = open_port, | |
781 | }; | |
782 | #endif | |
783 | ||
62322d25 | 784 | static const struct file_operations zero_fops = { |
1da177e4 LT |
785 | .llseek = zero_lseek, |
786 | .read = read_zero, | |
787 | .write = write_zero, | |
788 | .mmap = mmap_zero, | |
789 | }; | |
790 | ||
5da6185b DH |
791 | /* |
792 | * capabilities for /dev/zero | |
793 | * - permits private mappings, "copies" are taken of the source of zeros | |
794 | */ | |
1da177e4 | 795 | static struct backing_dev_info zero_bdi = { |
d993831f | 796 | .name = "char/mem", |
1da177e4 LT |
797 | .capabilities = BDI_CAP_MAP_COPY, |
798 | }; | |
799 | ||
62322d25 | 800 | static const struct file_operations full_fops = { |
1da177e4 LT |
801 | .llseek = full_lseek, |
802 | .read = read_full, | |
803 | .write = write_full, | |
804 | }; | |
805 | ||
50b1fdbd | 806 | #ifdef CONFIG_CRASH_DUMP |
62322d25 | 807 | static const struct file_operations oldmem_fops = { |
50b1fdbd VG |
808 | .read = read_oldmem, |
809 | .open = open_oldmem, | |
810 | }; | |
811 | #endif | |
812 | ||
1da177e4 LT |
813 | static ssize_t kmsg_write(struct file * file, const char __user * buf, |
814 | size_t count, loff_t *ppos) | |
815 | { | |
816 | char *tmp; | |
cd140a5c | 817 | ssize_t ret; |
1da177e4 LT |
818 | |
819 | tmp = kmalloc(count + 1, GFP_KERNEL); | |
820 | if (tmp == NULL) | |
821 | return -ENOMEM; | |
822 | ret = -EFAULT; | |
823 | if (!copy_from_user(tmp, buf, count)) { | |
824 | tmp[count] = 0; | |
825 | ret = printk("%s", tmp); | |
cd140a5c GC |
826 | if (ret > count) |
827 | /* printk can add a prefix */ | |
828 | ret = count; | |
1da177e4 LT |
829 | } |
830 | kfree(tmp); | |
831 | return ret; | |
832 | } | |
833 | ||
62322d25 | 834 | static const struct file_operations kmsg_fops = { |
1da177e4 LT |
835 | .write = kmsg_write, |
836 | }; | |
837 | ||
389e0cb9 KS |
838 | static const struct memdev { |
839 | const char *name; | |
e454cea2 | 840 | mode_t mode; |
389e0cb9 KS |
841 | const struct file_operations *fops; |
842 | struct backing_dev_info *dev_info; | |
843 | } devlist[] = { | |
e454cea2 | 844 | [1] = { "mem", 0, &mem_fops, &directly_mappable_cdev_bdi }, |
b781ecb6 | 845 | #ifdef CONFIG_DEVKMEM |
e454cea2 | 846 | [2] = { "kmem", 0, &kmem_fops, &directly_mappable_cdev_bdi }, |
b781ecb6 | 847 | #endif |
e454cea2 | 848 | [3] = { "null", 0666, &null_fops, NULL }, |
4f911d64 | 849 | #ifdef CONFIG_DEVPORT |
e454cea2 | 850 | [4] = { "port", 0, &port_fops, NULL }, |
1da177e4 | 851 | #endif |
e454cea2 KS |
852 | [5] = { "zero", 0666, &zero_fops, &zero_bdi }, |
853 | [7] = { "full", 0666, &full_fops, NULL }, | |
854 | [8] = { "random", 0666, &random_fops, NULL }, | |
855 | [9] = { "urandom", 0666, &urandom_fops, NULL }, | |
856 | [11] = { "kmsg", 0, &kmsg_fops, NULL }, | |
50b1fdbd | 857 | #ifdef CONFIG_CRASH_DUMP |
e454cea2 | 858 | [12] = { "oldmem", 0, &oldmem_fops, NULL }, |
50b1fdbd | 859 | #endif |
d6f47bef ASF |
860 | }; |
861 | ||
862 | static int memory_open(struct inode *inode, struct file *filp) | |
863 | { | |
389e0cb9 KS |
864 | int minor; |
865 | const struct memdev *dev; | |
d6f47bef | 866 | |
389e0cb9 KS |
867 | minor = iminor(inode); |
868 | if (minor >= ARRAY_SIZE(devlist)) | |
205153aa | 869 | return -ENXIO; |
d6f47bef | 870 | |
389e0cb9 KS |
871 | dev = &devlist[minor]; |
872 | if (!dev->fops) | |
205153aa | 873 | return -ENXIO; |
d6f47bef | 874 | |
389e0cb9 KS |
875 | filp->f_op = dev->fops; |
876 | if (dev->dev_info) | |
877 | filp->f_mapping->backing_dev_info = dev->dev_info; | |
d6f47bef | 878 | |
389e0cb9 | 879 | if (dev->fops->open) |
205153aa FW |
880 | return dev->fops->open(inode, filp); |
881 | ||
882 | return 0; | |
1da177e4 LT |
883 | } |
884 | ||
62322d25 | 885 | static const struct file_operations memory_fops = { |
389e0cb9 | 886 | .open = memory_open, |
1da177e4 LT |
887 | }; |
888 | ||
e454cea2 KS |
889 | static char *mem_devnode(struct device *dev, mode_t *mode) |
890 | { | |
891 | if (mode && devlist[MINOR(dev->devt)].mode) | |
892 | *mode = devlist[MINOR(dev->devt)].mode; | |
893 | return NULL; | |
894 | } | |
895 | ||
ca8eca68 | 896 | static struct class *mem_class; |
1da177e4 LT |
897 | |
898 | static int __init chr_dev_init(void) | |
899 | { | |
389e0cb9 | 900 | int minor; |
e0bf68dd PZ |
901 | int err; |
902 | ||
903 | err = bdi_init(&zero_bdi); | |
904 | if (err) | |
905 | return err; | |
1da177e4 LT |
906 | |
907 | if (register_chrdev(MEM_MAJOR,"mem",&memory_fops)) | |
908 | printk("unable to get major %d for memory devs\n", MEM_MAJOR); | |
909 | ||
ca8eca68 | 910 | mem_class = class_create(THIS_MODULE, "mem"); |
e454cea2 | 911 | mem_class->devnode = mem_devnode; |
389e0cb9 KS |
912 | for (minor = 1; minor < ARRAY_SIZE(devlist); minor++) { |
913 | if (!devlist[minor].name) | |
914 | continue; | |
915 | device_create(mem_class, NULL, MKDEV(MEM_MAJOR, minor), | |
916 | NULL, devlist[minor].name); | |
917 | } | |
ebf644c4 | 918 | |
1da177e4 LT |
919 | return 0; |
920 | } | |
921 | ||
922 | fs_initcall(chr_dev_init); |