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> | |
8 | * Shared /dev/zero mmaping support, Feb 2000, Kanoj Sarcar <kanoj@sgi.com> | |
9 | */ | |
10 | ||
11 | #include <linux/config.h> | |
12 | #include <linux/mm.h> | |
13 | #include <linux/miscdevice.h> | |
14 | #include <linux/slab.h> | |
15 | #include <linux/vmalloc.h> | |
16 | #include <linux/mman.h> | |
17 | #include <linux/random.h> | |
18 | #include <linux/init.h> | |
19 | #include <linux/raw.h> | |
20 | #include <linux/tty.h> | |
21 | #include <linux/capability.h> | |
22 | #include <linux/smp_lock.h> | |
23 | #include <linux/devfs_fs_kernel.h> | |
24 | #include <linux/ptrace.h> | |
25 | #include <linux/device.h> | |
50b1fdbd VG |
26 | #include <linux/highmem.h> |
27 | #include <linux/crash_dump.h> | |
1da177e4 | 28 | #include <linux/backing-dev.h> |
315c215c | 29 | #include <linux/bootmem.h> |
1da177e4 LT |
30 | |
31 | #include <asm/uaccess.h> | |
32 | #include <asm/io.h> | |
33 | ||
34 | #ifdef CONFIG_IA64 | |
35 | # include <linux/efi.h> | |
36 | #endif | |
37 | ||
38 | #if defined(CONFIG_S390_TAPE) && defined(CONFIG_S390_TAPE_CHAR) | |
39 | extern void tapechar_init(void); | |
40 | #endif | |
41 | ||
42 | /* | |
43 | * Architectures vary in how they handle caching for addresses | |
44 | * outside of main memory. | |
45 | * | |
46 | */ | |
47 | static inline int uncached_access(struct file *file, unsigned long addr) | |
48 | { | |
49 | #if defined(__i386__) | |
50 | /* | |
51 | * On the PPro and successors, the MTRRs are used to set | |
52 | * memory types for physical addresses outside main memory, | |
53 | * so blindly setting PCD or PWT on those pages is wrong. | |
54 | * For Pentiums and earlier, the surround logic should disable | |
55 | * caching for the high addresses through the KEN pin, but | |
56 | * we maintain the tradition of paranoia in this code. | |
57 | */ | |
58 | if (file->f_flags & O_SYNC) | |
59 | return 1; | |
60 | return !( test_bit(X86_FEATURE_MTRR, boot_cpu_data.x86_capability) || | |
61 | test_bit(X86_FEATURE_K6_MTRR, boot_cpu_data.x86_capability) || | |
62 | test_bit(X86_FEATURE_CYRIX_ARR, boot_cpu_data.x86_capability) || | |
63 | test_bit(X86_FEATURE_CENTAUR_MCR, boot_cpu_data.x86_capability) ) | |
64 | && addr >= __pa(high_memory); | |
65 | #elif defined(__x86_64__) | |
66 | /* | |
67 | * This is broken because it can generate memory type aliases, | |
68 | * which can cause cache corruptions | |
69 | * But it is only available for root and we have to be bug-to-bug | |
70 | * compatible with i386. | |
71 | */ | |
72 | if (file->f_flags & O_SYNC) | |
73 | return 1; | |
74 | /* same behaviour as i386. PAT always set to cached and MTRRs control the | |
75 | caching behaviour. | |
76 | Hopefully a full PAT implementation will fix that soon. */ | |
77 | return 0; | |
78 | #elif defined(CONFIG_IA64) | |
79 | /* | |
80 | * On ia64, we ignore O_SYNC because we cannot tolerate memory attribute aliases. | |
81 | */ | |
82 | return !(efi_mem_attributes(addr) & EFI_MEMORY_WB); | |
83 | #else | |
84 | /* | |
85 | * Accessing memory above the top the kernel knows about or through a file pointer | |
86 | * that was marked O_SYNC will be done non-cached. | |
87 | */ | |
88 | if (file->f_flags & O_SYNC) | |
89 | return 1; | |
90 | return addr >= __pa(high_memory); | |
91 | #endif | |
92 | } | |
93 | ||
94 | #ifndef ARCH_HAS_VALID_PHYS_ADDR_RANGE | |
95 | static inline int valid_phys_addr_range(unsigned long addr, size_t *count) | |
96 | { | |
97 | unsigned long end_mem; | |
98 | ||
99 | end_mem = __pa(high_memory); | |
100 | if (addr >= end_mem) | |
101 | return 0; | |
102 | ||
103 | if (*count > end_mem - addr) | |
104 | *count = end_mem - addr; | |
105 | ||
106 | return 1; | |
107 | } | |
108 | #endif | |
109 | ||
110 | /* | |
111 | * This funcion reads the *physical* memory. The f_pos points directly to the | |
112 | * memory location. | |
113 | */ | |
114 | static ssize_t read_mem(struct file * file, char __user * buf, | |
115 | size_t count, loff_t *ppos) | |
116 | { | |
117 | unsigned long p = *ppos; | |
118 | ssize_t read, sz; | |
119 | char *ptr; | |
120 | ||
121 | if (!valid_phys_addr_range(p, &count)) | |
122 | return -EFAULT; | |
123 | read = 0; | |
124 | #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED | |
125 | /* we don't have page 0 mapped on sparc and m68k.. */ | |
126 | if (p < PAGE_SIZE) { | |
127 | sz = PAGE_SIZE - p; | |
128 | if (sz > count) | |
129 | sz = count; | |
130 | if (sz > 0) { | |
131 | if (clear_user(buf, sz)) | |
132 | return -EFAULT; | |
133 | buf += sz; | |
134 | p += sz; | |
135 | count -= sz; | |
136 | read += sz; | |
137 | } | |
138 | } | |
139 | #endif | |
140 | ||
141 | while (count > 0) { | |
142 | /* | |
143 | * Handle first page in case it's not aligned | |
144 | */ | |
145 | if (-p & (PAGE_SIZE - 1)) | |
146 | sz = -p & (PAGE_SIZE - 1); | |
147 | else | |
148 | sz = PAGE_SIZE; | |
149 | ||
150 | sz = min_t(unsigned long, sz, count); | |
151 | ||
152 | /* | |
153 | * On ia64 if a page has been mapped somewhere as | |
154 | * uncached, then it must also be accessed uncached | |
155 | * by the kernel or data corruption may occur | |
156 | */ | |
157 | ptr = xlate_dev_mem_ptr(p); | |
158 | ||
159 | if (copy_to_user(buf, ptr, sz)) | |
160 | return -EFAULT; | |
161 | buf += sz; | |
162 | p += sz; | |
163 | count -= sz; | |
164 | read += sz; | |
165 | } | |
166 | ||
167 | *ppos += read; | |
168 | return read; | |
169 | } | |
170 | ||
171 | static ssize_t write_mem(struct file * file, const char __user * buf, | |
172 | size_t count, loff_t *ppos) | |
173 | { | |
174 | unsigned long p = *ppos; | |
175 | ssize_t written, sz; | |
176 | unsigned long copied; | |
177 | void *ptr; | |
178 | ||
179 | if (!valid_phys_addr_range(p, &count)) | |
180 | return -EFAULT; | |
181 | ||
182 | written = 0; | |
183 | ||
184 | #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED | |
185 | /* we don't have page 0 mapped on sparc and m68k.. */ | |
186 | if (p < PAGE_SIZE) { | |
187 | unsigned long sz = PAGE_SIZE - p; | |
188 | if (sz > count) | |
189 | sz = count; | |
190 | /* Hmm. Do something? */ | |
191 | buf += sz; | |
192 | p += sz; | |
193 | count -= sz; | |
194 | written += sz; | |
195 | } | |
196 | #endif | |
197 | ||
198 | while (count > 0) { | |
199 | /* | |
200 | * Handle first page in case it's not aligned | |
201 | */ | |
202 | if (-p & (PAGE_SIZE - 1)) | |
203 | sz = -p & (PAGE_SIZE - 1); | |
204 | else | |
205 | sz = PAGE_SIZE; | |
206 | ||
207 | sz = min_t(unsigned long, sz, count); | |
208 | ||
209 | /* | |
210 | * On ia64 if a page has been mapped somewhere as | |
211 | * uncached, then it must also be accessed uncached | |
212 | * by the kernel or data corruption may occur | |
213 | */ | |
214 | ptr = xlate_dev_mem_ptr(p); | |
215 | ||
216 | copied = copy_from_user(ptr, buf, sz); | |
217 | if (copied) { | |
218 | ssize_t ret; | |
219 | ||
220 | ret = written + (sz - copied); | |
221 | if (ret) | |
222 | return ret; | |
223 | return -EFAULT; | |
224 | } | |
225 | buf += sz; | |
226 | p += sz; | |
227 | count -= sz; | |
228 | written += sz; | |
229 | } | |
230 | ||
231 | *ppos += written; | |
232 | return written; | |
233 | } | |
234 | ||
235 | static int mmap_mem(struct file * file, struct vm_area_struct * vma) | |
236 | { | |
237 | #if defined(__HAVE_PHYS_MEM_ACCESS_PROT) | |
238 | unsigned long offset = vma->vm_pgoff << PAGE_SHIFT; | |
239 | ||
240 | vma->vm_page_prot = phys_mem_access_prot(file, offset, | |
241 | vma->vm_end - vma->vm_start, | |
242 | vma->vm_page_prot); | |
243 | #elif defined(pgprot_noncached) | |
244 | unsigned long offset = vma->vm_pgoff << PAGE_SHIFT; | |
245 | int uncached; | |
246 | ||
247 | uncached = uncached_access(file, offset); | |
248 | if (uncached) | |
249 | vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot); | |
250 | #endif | |
251 | ||
252 | /* Remap-pfn-range will mark the range VM_IO and VM_RESERVED */ | |
253 | if (remap_pfn_range(vma, | |
254 | vma->vm_start, | |
255 | vma->vm_pgoff, | |
256 | vma->vm_end-vma->vm_start, | |
257 | vma->vm_page_prot)) | |
258 | return -EAGAIN; | |
259 | return 0; | |
260 | } | |
261 | ||
262 | static int mmap_kmem(struct file * file, struct vm_area_struct * vma) | |
263 | { | |
264 | unsigned long long val; | |
265 | /* | |
266 | * RED-PEN: on some architectures there is more mapped memory | |
267 | * than available in mem_map which pfn_valid checks | |
268 | * for. Perhaps should add a new macro here. | |
269 | * | |
270 | * RED-PEN: vmalloc is not supported right now. | |
271 | */ | |
272 | if (!pfn_valid(vma->vm_pgoff)) | |
273 | return -EIO; | |
274 | val = (u64)vma->vm_pgoff << PAGE_SHIFT; | |
275 | vma->vm_pgoff = __pa(val) >> PAGE_SHIFT; | |
276 | return mmap_mem(file, vma); | |
277 | } | |
278 | ||
50b1fdbd VG |
279 | #ifdef CONFIG_CRASH_DUMP |
280 | /* | |
281 | * Read memory corresponding to the old kernel. | |
50b1fdbd | 282 | */ |
315c215c | 283 | static ssize_t read_oldmem(struct file *file, char __user *buf, |
50b1fdbd VG |
284 | size_t count, loff_t *ppos) |
285 | { | |
315c215c VG |
286 | unsigned long pfn, offset; |
287 | size_t read = 0, csize; | |
288 | int rc = 0; | |
50b1fdbd | 289 | |
72414d3f | 290 | while (count) { |
50b1fdbd | 291 | pfn = *ppos / PAGE_SIZE; |
315c215c VG |
292 | if (pfn > saved_max_pfn) |
293 | return read; | |
50b1fdbd | 294 | |
315c215c VG |
295 | offset = (unsigned long)(*ppos % PAGE_SIZE); |
296 | if (count > PAGE_SIZE - offset) | |
297 | csize = PAGE_SIZE - offset; | |
298 | else | |
299 | csize = count; | |
50b1fdbd | 300 | |
315c215c VG |
301 | rc = copy_oldmem_page(pfn, buf, csize, offset, 1); |
302 | if (rc < 0) | |
303 | return rc; | |
50b1fdbd VG |
304 | buf += csize; |
305 | *ppos += csize; | |
306 | read += csize; | |
307 | count -= csize; | |
308 | } | |
50b1fdbd VG |
309 | return read; |
310 | } | |
311 | #endif | |
312 | ||
1da177e4 LT |
313 | extern long vread(char *buf, char *addr, unsigned long count); |
314 | extern long vwrite(char *buf, char *addr, unsigned long count); | |
315 | ||
316 | /* | |
317 | * This function reads the *virtual* memory as seen by the kernel. | |
318 | */ | |
319 | static ssize_t read_kmem(struct file *file, char __user *buf, | |
320 | size_t count, loff_t *ppos) | |
321 | { | |
322 | unsigned long p = *ppos; | |
323 | ssize_t low_count, read, sz; | |
324 | char * kbuf; /* k-addr because vread() takes vmlist_lock rwlock */ | |
325 | ||
326 | read = 0; | |
327 | if (p < (unsigned long) high_memory) { | |
328 | low_count = count; | |
329 | if (count > (unsigned long) high_memory - p) | |
330 | low_count = (unsigned long) high_memory - p; | |
331 | ||
332 | #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED | |
333 | /* we don't have page 0 mapped on sparc and m68k.. */ | |
334 | if (p < PAGE_SIZE && low_count > 0) { | |
335 | size_t tmp = PAGE_SIZE - p; | |
336 | if (tmp > low_count) tmp = low_count; | |
337 | if (clear_user(buf, tmp)) | |
338 | return -EFAULT; | |
339 | buf += tmp; | |
340 | p += tmp; | |
341 | read += tmp; | |
342 | low_count -= tmp; | |
343 | count -= tmp; | |
344 | } | |
345 | #endif | |
346 | while (low_count > 0) { | |
347 | /* | |
348 | * Handle first page in case it's not aligned | |
349 | */ | |
350 | if (-p & (PAGE_SIZE - 1)) | |
351 | sz = -p & (PAGE_SIZE - 1); | |
352 | else | |
353 | sz = PAGE_SIZE; | |
354 | ||
355 | sz = min_t(unsigned long, sz, low_count); | |
356 | ||
357 | /* | |
358 | * On ia64 if a page has been mapped somewhere as | |
359 | * uncached, then it must also be accessed uncached | |
360 | * by the kernel or data corruption may occur | |
361 | */ | |
362 | kbuf = xlate_dev_kmem_ptr((char *)p); | |
363 | ||
364 | if (copy_to_user(buf, kbuf, sz)) | |
365 | return -EFAULT; | |
366 | buf += sz; | |
367 | p += sz; | |
368 | read += sz; | |
369 | low_count -= sz; | |
370 | count -= sz; | |
371 | } | |
372 | } | |
373 | ||
374 | if (count > 0) { | |
375 | kbuf = (char *)__get_free_page(GFP_KERNEL); | |
376 | if (!kbuf) | |
377 | return -ENOMEM; | |
378 | while (count > 0) { | |
379 | int len = count; | |
380 | ||
381 | if (len > PAGE_SIZE) | |
382 | len = PAGE_SIZE; | |
383 | len = vread(kbuf, (char *)p, len); | |
384 | if (!len) | |
385 | break; | |
386 | if (copy_to_user(buf, kbuf, len)) { | |
387 | free_page((unsigned long)kbuf); | |
388 | return -EFAULT; | |
389 | } | |
390 | count -= len; | |
391 | buf += len; | |
392 | read += len; | |
393 | p += len; | |
394 | } | |
395 | free_page((unsigned long)kbuf); | |
396 | } | |
397 | *ppos = p; | |
398 | return read; | |
399 | } | |
400 | ||
401 | ||
402 | static inline ssize_t | |
403 | do_write_kmem(void *p, unsigned long realp, const char __user * buf, | |
404 | size_t count, loff_t *ppos) | |
405 | { | |
406 | ssize_t written, sz; | |
407 | unsigned long copied; | |
408 | ||
409 | written = 0; | |
410 | #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED | |
411 | /* we don't have page 0 mapped on sparc and m68k.. */ | |
412 | if (realp < PAGE_SIZE) { | |
413 | unsigned long sz = PAGE_SIZE - realp; | |
414 | if (sz > count) | |
415 | sz = count; | |
416 | /* Hmm. Do something? */ | |
417 | buf += sz; | |
418 | p += sz; | |
419 | realp += sz; | |
420 | count -= sz; | |
421 | written += sz; | |
422 | } | |
423 | #endif | |
424 | ||
425 | while (count > 0) { | |
426 | char *ptr; | |
427 | /* | |
428 | * Handle first page in case it's not aligned | |
429 | */ | |
430 | if (-realp & (PAGE_SIZE - 1)) | |
431 | sz = -realp & (PAGE_SIZE - 1); | |
432 | else | |
433 | sz = PAGE_SIZE; | |
434 | ||
435 | sz = min_t(unsigned long, sz, count); | |
436 | ||
437 | /* | |
438 | * On ia64 if a page has been mapped somewhere as | |
439 | * uncached, then it must also be accessed uncached | |
440 | * by the kernel or data corruption may occur | |
441 | */ | |
442 | ptr = xlate_dev_kmem_ptr(p); | |
443 | ||
444 | copied = copy_from_user(ptr, buf, sz); | |
445 | if (copied) { | |
446 | ssize_t ret; | |
447 | ||
448 | ret = written + (sz - copied); | |
449 | if (ret) | |
450 | return ret; | |
451 | return -EFAULT; | |
452 | } | |
453 | buf += sz; | |
454 | p += sz; | |
455 | realp += sz; | |
456 | count -= sz; | |
457 | written += sz; | |
458 | } | |
459 | ||
460 | *ppos += written; | |
461 | return written; | |
462 | } | |
463 | ||
464 | ||
465 | /* | |
466 | * This function writes to the *virtual* memory as seen by the kernel. | |
467 | */ | |
468 | static ssize_t write_kmem(struct file * file, const char __user * buf, | |
469 | size_t count, loff_t *ppos) | |
470 | { | |
471 | unsigned long p = *ppos; | |
472 | ssize_t wrote = 0; | |
473 | ssize_t virtr = 0; | |
474 | ssize_t written; | |
475 | char * kbuf; /* k-addr because vwrite() takes vmlist_lock rwlock */ | |
476 | ||
477 | if (p < (unsigned long) high_memory) { | |
478 | ||
479 | wrote = count; | |
480 | if (count > (unsigned long) high_memory - p) | |
481 | wrote = (unsigned long) high_memory - p; | |
482 | ||
483 | written = do_write_kmem((void*)p, p, buf, wrote, ppos); | |
484 | if (written != wrote) | |
485 | return written; | |
486 | wrote = written; | |
487 | p += wrote; | |
488 | buf += wrote; | |
489 | count -= wrote; | |
490 | } | |
491 | ||
492 | if (count > 0) { | |
493 | kbuf = (char *)__get_free_page(GFP_KERNEL); | |
494 | if (!kbuf) | |
495 | return wrote ? wrote : -ENOMEM; | |
496 | while (count > 0) { | |
497 | int len = count; | |
498 | ||
499 | if (len > PAGE_SIZE) | |
500 | len = PAGE_SIZE; | |
501 | if (len) { | |
502 | written = copy_from_user(kbuf, buf, len); | |
503 | if (written) { | |
504 | ssize_t ret; | |
505 | ||
506 | free_page((unsigned long)kbuf); | |
507 | ret = wrote + virtr + (len - written); | |
508 | return ret ? ret : -EFAULT; | |
509 | } | |
510 | } | |
511 | len = vwrite(kbuf, (char *)p, len); | |
512 | count -= len; | |
513 | buf += len; | |
514 | virtr += len; | |
515 | p += len; | |
516 | } | |
517 | free_page((unsigned long)kbuf); | |
518 | } | |
519 | ||
520 | *ppos = p; | |
521 | return virtr + wrote; | |
522 | } | |
523 | ||
145d01e4 | 524 | #if (defined(CONFIG_ISA) || !defined(__mc68000__)) && (!defined(CONFIG_PPC_ISERIES) || defined(CONFIG_PCI)) |
1da177e4 LT |
525 | static ssize_t read_port(struct file * file, char __user * buf, |
526 | size_t count, loff_t *ppos) | |
527 | { | |
528 | unsigned long i = *ppos; | |
529 | char __user *tmp = buf; | |
530 | ||
531 | if (!access_ok(VERIFY_WRITE, buf, count)) | |
532 | return -EFAULT; | |
533 | while (count-- > 0 && i < 65536) { | |
534 | if (__put_user(inb(i),tmp) < 0) | |
535 | return -EFAULT; | |
536 | i++; | |
537 | tmp++; | |
538 | } | |
539 | *ppos = i; | |
540 | return tmp-buf; | |
541 | } | |
542 | ||
543 | static ssize_t write_port(struct file * file, const char __user * buf, | |
544 | size_t count, loff_t *ppos) | |
545 | { | |
546 | unsigned long i = *ppos; | |
547 | const char __user * tmp = buf; | |
548 | ||
549 | if (!access_ok(VERIFY_READ,buf,count)) | |
550 | return -EFAULT; | |
551 | while (count-- > 0 && i < 65536) { | |
552 | char c; | |
553 | if (__get_user(c, tmp)) | |
554 | return -EFAULT; | |
555 | outb(c,i); | |
556 | i++; | |
557 | tmp++; | |
558 | } | |
559 | *ppos = i; | |
560 | return tmp-buf; | |
561 | } | |
562 | #endif | |
563 | ||
564 | static ssize_t read_null(struct file * file, char __user * buf, | |
565 | size_t count, loff_t *ppos) | |
566 | { | |
567 | return 0; | |
568 | } | |
569 | ||
570 | static ssize_t write_null(struct file * file, const char __user * buf, | |
571 | size_t count, loff_t *ppos) | |
572 | { | |
573 | return count; | |
574 | } | |
575 | ||
576 | #ifdef CONFIG_MMU | |
577 | /* | |
578 | * For fun, we are using the MMU for this. | |
579 | */ | |
580 | static inline size_t read_zero_pagealigned(char __user * buf, size_t size) | |
581 | { | |
582 | struct mm_struct *mm; | |
583 | struct vm_area_struct * vma; | |
584 | unsigned long addr=(unsigned long)buf; | |
585 | ||
586 | mm = current->mm; | |
587 | /* Oops, this was forgotten before. -ben */ | |
588 | down_read(&mm->mmap_sem); | |
589 | ||
590 | /* For private mappings, just map in zero pages. */ | |
591 | for (vma = find_vma(mm, addr); vma; vma = vma->vm_next) { | |
592 | unsigned long count; | |
593 | ||
594 | if (vma->vm_start > addr || (vma->vm_flags & VM_WRITE) == 0) | |
595 | goto out_up; | |
596 | if (vma->vm_flags & (VM_SHARED | VM_HUGETLB)) | |
597 | break; | |
598 | count = vma->vm_end - addr; | |
599 | if (count > size) | |
600 | count = size; | |
601 | ||
602 | zap_page_range(vma, addr, count, NULL); | |
603 | zeromap_page_range(vma, addr, count, PAGE_COPY); | |
604 | ||
605 | size -= count; | |
606 | buf += count; | |
607 | addr += count; | |
608 | if (size == 0) | |
609 | goto out_up; | |
610 | } | |
611 | ||
612 | up_read(&mm->mmap_sem); | |
613 | ||
614 | /* The shared case is hard. Let's do the conventional zeroing. */ | |
615 | do { | |
616 | unsigned long unwritten = clear_user(buf, PAGE_SIZE); | |
617 | if (unwritten) | |
618 | return size + unwritten - PAGE_SIZE; | |
619 | cond_resched(); | |
620 | buf += PAGE_SIZE; | |
621 | size -= PAGE_SIZE; | |
622 | } while (size); | |
623 | ||
624 | return size; | |
625 | out_up: | |
626 | up_read(&mm->mmap_sem); | |
627 | return size; | |
628 | } | |
629 | ||
630 | static ssize_t read_zero(struct file * file, char __user * buf, | |
631 | size_t count, loff_t *ppos) | |
632 | { | |
633 | unsigned long left, unwritten, written = 0; | |
634 | ||
635 | if (!count) | |
636 | return 0; | |
637 | ||
638 | if (!access_ok(VERIFY_WRITE, buf, count)) | |
639 | return -EFAULT; | |
640 | ||
641 | left = count; | |
642 | ||
643 | /* do we want to be clever? Arbitrary cut-off */ | |
644 | if (count >= PAGE_SIZE*4) { | |
645 | unsigned long partial; | |
646 | ||
647 | /* How much left of the page? */ | |
648 | partial = (PAGE_SIZE-1) & -(unsigned long) buf; | |
649 | unwritten = clear_user(buf, partial); | |
650 | written = partial - unwritten; | |
651 | if (unwritten) | |
652 | goto out; | |
653 | left -= partial; | |
654 | buf += partial; | |
655 | unwritten = read_zero_pagealigned(buf, left & PAGE_MASK); | |
656 | written += (left & PAGE_MASK) - unwritten; | |
657 | if (unwritten) | |
658 | goto out; | |
659 | buf += left & PAGE_MASK; | |
660 | left &= ~PAGE_MASK; | |
661 | } | |
662 | unwritten = clear_user(buf, left); | |
663 | written += left - unwritten; | |
664 | out: | |
665 | return written ? written : -EFAULT; | |
666 | } | |
667 | ||
668 | static int mmap_zero(struct file * file, struct vm_area_struct * vma) | |
669 | { | |
670 | if (vma->vm_flags & VM_SHARED) | |
671 | return shmem_zero_setup(vma); | |
672 | if (zeromap_page_range(vma, vma->vm_start, vma->vm_end - vma->vm_start, vma->vm_page_prot)) | |
673 | return -EAGAIN; | |
674 | return 0; | |
675 | } | |
676 | #else /* CONFIG_MMU */ | |
677 | static ssize_t read_zero(struct file * file, char * buf, | |
678 | size_t count, loff_t *ppos) | |
679 | { | |
680 | size_t todo = count; | |
681 | ||
682 | while (todo) { | |
683 | size_t chunk = todo; | |
684 | ||
685 | if (chunk > 4096) | |
686 | chunk = 4096; /* Just for latency reasons */ | |
687 | if (clear_user(buf, chunk)) | |
688 | return -EFAULT; | |
689 | buf += chunk; | |
690 | todo -= chunk; | |
691 | cond_resched(); | |
692 | } | |
693 | return count; | |
694 | } | |
695 | ||
696 | static int mmap_zero(struct file * file, struct vm_area_struct * vma) | |
697 | { | |
698 | return -ENOSYS; | |
699 | } | |
700 | #endif /* CONFIG_MMU */ | |
701 | ||
702 | static ssize_t write_full(struct file * file, const char __user * buf, | |
703 | size_t count, loff_t *ppos) | |
704 | { | |
705 | return -ENOSPC; | |
706 | } | |
707 | ||
708 | /* | |
709 | * Special lseek() function for /dev/null and /dev/zero. Most notably, you | |
710 | * can fopen() both devices with "a" now. This was previously impossible. | |
711 | * -- SRB. | |
712 | */ | |
713 | ||
714 | static loff_t null_lseek(struct file * file, loff_t offset, int orig) | |
715 | { | |
716 | return file->f_pos = 0; | |
717 | } | |
718 | ||
719 | /* | |
720 | * The memory devices use the full 32/64 bits of the offset, and so we cannot | |
721 | * check against negative addresses: they are ok. The return value is weird, | |
722 | * though, in that case (0). | |
723 | * | |
724 | * also note that seeking relative to the "end of file" isn't supported: | |
725 | * it has no meaning, so it returns -EINVAL. | |
726 | */ | |
727 | static loff_t memory_lseek(struct file * file, loff_t offset, int orig) | |
728 | { | |
729 | loff_t ret; | |
730 | ||
731 | down(&file->f_dentry->d_inode->i_sem); | |
732 | switch (orig) { | |
733 | case 0: | |
734 | file->f_pos = offset; | |
735 | ret = file->f_pos; | |
736 | force_successful_syscall_return(); | |
737 | break; | |
738 | case 1: | |
739 | file->f_pos += offset; | |
740 | ret = file->f_pos; | |
741 | force_successful_syscall_return(); | |
742 | break; | |
743 | default: | |
744 | ret = -EINVAL; | |
745 | } | |
746 | up(&file->f_dentry->d_inode->i_sem); | |
747 | return ret; | |
748 | } | |
749 | ||
750 | static int open_port(struct inode * inode, struct file * filp) | |
751 | { | |
752 | return capable(CAP_SYS_RAWIO) ? 0 : -EPERM; | |
753 | } | |
754 | ||
755 | #define zero_lseek null_lseek | |
756 | #define full_lseek null_lseek | |
757 | #define write_zero write_null | |
758 | #define read_full read_zero | |
759 | #define open_mem open_port | |
760 | #define open_kmem open_mem | |
50b1fdbd | 761 | #define open_oldmem open_mem |
1da177e4 LT |
762 | |
763 | static struct file_operations mem_fops = { | |
764 | .llseek = memory_lseek, | |
765 | .read = read_mem, | |
766 | .write = write_mem, | |
767 | .mmap = mmap_mem, | |
768 | .open = open_mem, | |
769 | }; | |
770 | ||
771 | static struct file_operations kmem_fops = { | |
772 | .llseek = memory_lseek, | |
773 | .read = read_kmem, | |
774 | .write = write_kmem, | |
775 | .mmap = mmap_kmem, | |
776 | .open = open_kmem, | |
777 | }; | |
778 | ||
779 | static struct file_operations null_fops = { | |
780 | .llseek = null_lseek, | |
781 | .read = read_null, | |
782 | .write = write_null, | |
783 | }; | |
784 | ||
145d01e4 | 785 | #if (defined(CONFIG_ISA) || !defined(__mc68000__)) && (!defined(CONFIG_PPC_ISERIES) || defined(CONFIG_PCI)) |
1da177e4 LT |
786 | static struct file_operations port_fops = { |
787 | .llseek = memory_lseek, | |
788 | .read = read_port, | |
789 | .write = write_port, | |
790 | .open = open_port, | |
791 | }; | |
792 | #endif | |
793 | ||
794 | static struct file_operations zero_fops = { | |
795 | .llseek = zero_lseek, | |
796 | .read = read_zero, | |
797 | .write = write_zero, | |
798 | .mmap = mmap_zero, | |
799 | }; | |
800 | ||
801 | static struct backing_dev_info zero_bdi = { | |
802 | .capabilities = BDI_CAP_MAP_COPY, | |
803 | }; | |
804 | ||
805 | static struct file_operations full_fops = { | |
806 | .llseek = full_lseek, | |
807 | .read = read_full, | |
808 | .write = write_full, | |
809 | }; | |
810 | ||
50b1fdbd VG |
811 | #ifdef CONFIG_CRASH_DUMP |
812 | static struct file_operations oldmem_fops = { | |
813 | .read = read_oldmem, | |
814 | .open = open_oldmem, | |
815 | }; | |
816 | #endif | |
817 | ||
1da177e4 LT |
818 | static ssize_t kmsg_write(struct file * file, const char __user * buf, |
819 | size_t count, loff_t *ppos) | |
820 | { | |
821 | char *tmp; | |
822 | int ret; | |
823 | ||
824 | tmp = kmalloc(count + 1, GFP_KERNEL); | |
825 | if (tmp == NULL) | |
826 | return -ENOMEM; | |
827 | ret = -EFAULT; | |
828 | if (!copy_from_user(tmp, buf, count)) { | |
829 | tmp[count] = 0; | |
830 | ret = printk("%s", tmp); | |
831 | } | |
832 | kfree(tmp); | |
833 | return ret; | |
834 | } | |
835 | ||
836 | static struct file_operations kmsg_fops = { | |
837 | .write = kmsg_write, | |
838 | }; | |
839 | ||
840 | static int memory_open(struct inode * inode, struct file * filp) | |
841 | { | |
842 | switch (iminor(inode)) { | |
843 | case 1: | |
844 | filp->f_op = &mem_fops; | |
845 | break; | |
846 | case 2: | |
847 | filp->f_op = &kmem_fops; | |
848 | break; | |
849 | case 3: | |
850 | filp->f_op = &null_fops; | |
851 | break; | |
145d01e4 | 852 | #if (defined(CONFIG_ISA) || !defined(__mc68000__)) && (!defined(CONFIG_PPC_ISERIES) || defined(CONFIG_PCI)) |
1da177e4 LT |
853 | case 4: |
854 | filp->f_op = &port_fops; | |
855 | break; | |
856 | #endif | |
857 | case 5: | |
858 | filp->f_mapping->backing_dev_info = &zero_bdi; | |
859 | filp->f_op = &zero_fops; | |
860 | break; | |
861 | case 7: | |
862 | filp->f_op = &full_fops; | |
863 | break; | |
864 | case 8: | |
865 | filp->f_op = &random_fops; | |
866 | break; | |
867 | case 9: | |
868 | filp->f_op = &urandom_fops; | |
869 | break; | |
870 | case 11: | |
871 | filp->f_op = &kmsg_fops; | |
872 | break; | |
50b1fdbd VG |
873 | #ifdef CONFIG_CRASH_DUMP |
874 | case 12: | |
875 | filp->f_op = &oldmem_fops; | |
876 | break; | |
877 | #endif | |
1da177e4 LT |
878 | default: |
879 | return -ENXIO; | |
880 | } | |
881 | if (filp->f_op && filp->f_op->open) | |
882 | return filp->f_op->open(inode,filp); | |
883 | return 0; | |
884 | } | |
885 | ||
886 | static struct file_operations memory_fops = { | |
887 | .open = memory_open, /* just a selector for the real open */ | |
888 | }; | |
889 | ||
890 | static const struct { | |
891 | unsigned int minor; | |
892 | char *name; | |
893 | umode_t mode; | |
894 | struct file_operations *fops; | |
895 | } devlist[] = { /* list of minor devices */ | |
896 | {1, "mem", S_IRUSR | S_IWUSR | S_IRGRP, &mem_fops}, | |
897 | {2, "kmem", S_IRUSR | S_IWUSR | S_IRGRP, &kmem_fops}, | |
898 | {3, "null", S_IRUGO | S_IWUGO, &null_fops}, | |
145d01e4 | 899 | #if (defined(CONFIG_ISA) || !defined(__mc68000__)) && (!defined(CONFIG_PPC_ISERIES) || defined(CONFIG_PCI)) |
1da177e4 LT |
900 | {4, "port", S_IRUSR | S_IWUSR | S_IRGRP, &port_fops}, |
901 | #endif | |
902 | {5, "zero", S_IRUGO | S_IWUGO, &zero_fops}, | |
903 | {7, "full", S_IRUGO | S_IWUGO, &full_fops}, | |
904 | {8, "random", S_IRUGO | S_IWUSR, &random_fops}, | |
905 | {9, "urandom", S_IRUGO | S_IWUSR, &urandom_fops}, | |
906 | {11,"kmsg", S_IRUGO | S_IWUSR, &kmsg_fops}, | |
50b1fdbd VG |
907 | #ifdef CONFIG_CRASH_DUMP |
908 | {12,"oldmem", S_IRUSR | S_IWUSR | S_IRGRP, &oldmem_fops}, | |
909 | #endif | |
1da177e4 LT |
910 | }; |
911 | ||
ca8eca68 | 912 | static struct class *mem_class; |
1da177e4 LT |
913 | |
914 | static int __init chr_dev_init(void) | |
915 | { | |
916 | int i; | |
917 | ||
918 | if (register_chrdev(MEM_MAJOR,"mem",&memory_fops)) | |
919 | printk("unable to get major %d for memory devs\n", MEM_MAJOR); | |
920 | ||
ca8eca68 | 921 | mem_class = class_create(THIS_MODULE, "mem"); |
1da177e4 | 922 | for (i = 0; i < ARRAY_SIZE(devlist); i++) { |
ca8eca68 | 923 | class_device_create(mem_class, MKDEV(MEM_MAJOR, devlist[i].minor), |
1da177e4 LT |
924 | NULL, devlist[i].name); |
925 | devfs_mk_cdev(MKDEV(MEM_MAJOR, devlist[i].minor), | |
926 | S_IFCHR | devlist[i].mode, devlist[i].name); | |
927 | } | |
928 | ||
929 | return 0; | |
930 | } | |
931 | ||
932 | fs_initcall(chr_dev_init); |