diff options
author | Linus Torvalds <torvalds@ppc970.osdl.org> | 2005-04-16 18:20:36 -0400 |
---|---|---|
committer | Linus Torvalds <torvalds@ppc970.osdl.org> | 2005-04-16 18:20:36 -0400 |
commit | 1da177e4c3f41524e886b7f1b8a0c1fc7321cac2 (patch) | |
tree | 0bba044c4ce775e45a88a51686b5d9f90697ea9d /drivers/char/mem.c |
Linux-2.6.12-rc2v2.6.12-rc2
Initial git repository build. I'm not bothering with the full history,
even though we have it. We can create a separate "historical" git
archive of that later if we want to, and in the meantime it's about
3.2GB when imported into git - space that would just make the early
git days unnecessarily complicated, when we don't have a lot of good
infrastructure for it.
Let it rip!
Diffstat (limited to 'drivers/char/mem.c')
-rw-r--r-- | drivers/char/mem.c | 880 |
1 files changed, 880 insertions, 0 deletions
diff --git a/drivers/char/mem.c b/drivers/char/mem.c new file mode 100644 index 000000000000..947cb3cef816 --- /dev/null +++ b/drivers/char/mem.c | |||
@@ -0,0 +1,880 @@ | |||
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> | ||
26 | #include <linux/backing-dev.h> | ||
27 | |||
28 | #include <asm/uaccess.h> | ||
29 | #include <asm/io.h> | ||
30 | |||
31 | #ifdef CONFIG_IA64 | ||
32 | # include <linux/efi.h> | ||
33 | #endif | ||
34 | |||
35 | #if defined(CONFIG_S390_TAPE) && defined(CONFIG_S390_TAPE_CHAR) | ||
36 | extern void tapechar_init(void); | ||
37 | #endif | ||
38 | |||
39 | /* | ||
40 | * Architectures vary in how they handle caching for addresses | ||
41 | * outside of main memory. | ||
42 | * | ||
43 | */ | ||
44 | static inline int uncached_access(struct file *file, unsigned long addr) | ||
45 | { | ||
46 | #if defined(__i386__) | ||
47 | /* | ||
48 | * On the PPro and successors, the MTRRs are used to set | ||
49 | * memory types for physical addresses outside main memory, | ||
50 | * so blindly setting PCD or PWT on those pages is wrong. | ||
51 | * For Pentiums and earlier, the surround logic should disable | ||
52 | * caching for the high addresses through the KEN pin, but | ||
53 | * we maintain the tradition of paranoia in this code. | ||
54 | */ | ||
55 | if (file->f_flags & O_SYNC) | ||
56 | return 1; | ||
57 | return !( test_bit(X86_FEATURE_MTRR, boot_cpu_data.x86_capability) || | ||
58 | test_bit(X86_FEATURE_K6_MTRR, boot_cpu_data.x86_capability) || | ||
59 | test_bit(X86_FEATURE_CYRIX_ARR, boot_cpu_data.x86_capability) || | ||
60 | test_bit(X86_FEATURE_CENTAUR_MCR, boot_cpu_data.x86_capability) ) | ||
61 | && addr >= __pa(high_memory); | ||
62 | #elif defined(__x86_64__) | ||
63 | /* | ||
64 | * This is broken because it can generate memory type aliases, | ||
65 | * which can cause cache corruptions | ||
66 | * But it is only available for root and we have to be bug-to-bug | ||
67 | * compatible with i386. | ||
68 | */ | ||
69 | if (file->f_flags & O_SYNC) | ||
70 | return 1; | ||
71 | /* same behaviour as i386. PAT always set to cached and MTRRs control the | ||
72 | caching behaviour. | ||
73 | Hopefully a full PAT implementation will fix that soon. */ | ||
74 | return 0; | ||
75 | #elif defined(CONFIG_IA64) | ||
76 | /* | ||
77 | * On ia64, we ignore O_SYNC because we cannot tolerate memory attribute aliases. | ||
78 | */ | ||
79 | return !(efi_mem_attributes(addr) & EFI_MEMORY_WB); | ||
80 | #else | ||
81 | /* | ||
82 | * Accessing memory above the top the kernel knows about or through a file pointer | ||
83 | * that was marked O_SYNC will be done non-cached. | ||
84 | */ | ||
85 | if (file->f_flags & O_SYNC) | ||
86 | return 1; | ||
87 | return addr >= __pa(high_memory); | ||
88 | #endif | ||
89 | } | ||
90 | |||
91 | #ifndef ARCH_HAS_VALID_PHYS_ADDR_RANGE | ||
92 | static inline int valid_phys_addr_range(unsigned long addr, size_t *count) | ||
93 | { | ||
94 | unsigned long end_mem; | ||
95 | |||
96 | end_mem = __pa(high_memory); | ||
97 | if (addr >= end_mem) | ||
98 | return 0; | ||
99 | |||
100 | if (*count > end_mem - addr) | ||
101 | *count = end_mem - addr; | ||
102 | |||
103 | return 1; | ||
104 | } | ||
105 | #endif | ||
106 | |||
107 | /* | ||
108 | * This funcion reads the *physical* memory. The f_pos points directly to the | ||
109 | * memory location. | ||
110 | */ | ||
111 | static ssize_t read_mem(struct file * file, char __user * buf, | ||
112 | size_t count, loff_t *ppos) | ||
113 | { | ||
114 | unsigned long p = *ppos; | ||
115 | ssize_t read, sz; | ||
116 | char *ptr; | ||
117 | |||
118 | if (!valid_phys_addr_range(p, &count)) | ||
119 | return -EFAULT; | ||
120 | read = 0; | ||
121 | #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED | ||
122 | /* we don't have page 0 mapped on sparc and m68k.. */ | ||
123 | if (p < PAGE_SIZE) { | ||
124 | sz = PAGE_SIZE - p; | ||
125 | if (sz > count) | ||
126 | sz = count; | ||
127 | if (sz > 0) { | ||
128 | if (clear_user(buf, sz)) | ||
129 | return -EFAULT; | ||
130 | buf += sz; | ||
131 | p += sz; | ||
132 | count -= sz; | ||
133 | read += sz; | ||
134 | } | ||
135 | } | ||
136 | #endif | ||
137 | |||
138 | while (count > 0) { | ||
139 | /* | ||
140 | * Handle first page in case it's not aligned | ||
141 | */ | ||
142 | if (-p & (PAGE_SIZE - 1)) | ||
143 | sz = -p & (PAGE_SIZE - 1); | ||
144 | else | ||
145 | sz = PAGE_SIZE; | ||
146 | |||
147 | sz = min_t(unsigned long, sz, count); | ||
148 | |||
149 | /* | ||
150 | * On ia64 if a page has been mapped somewhere as | ||
151 | * uncached, then it must also be accessed uncached | ||
152 | * by the kernel or data corruption may occur | ||
153 | */ | ||
154 | ptr = xlate_dev_mem_ptr(p); | ||
155 | |||
156 | if (copy_to_user(buf, ptr, sz)) | ||
157 | return -EFAULT; | ||
158 | buf += sz; | ||
159 | p += sz; | ||
160 | count -= sz; | ||
161 | read += sz; | ||
162 | } | ||
163 | |||
164 | *ppos += read; | ||
165 | return read; | ||
166 | } | ||
167 | |||
168 | static ssize_t write_mem(struct file * file, const char __user * buf, | ||
169 | size_t count, loff_t *ppos) | ||
170 | { | ||
171 | unsigned long p = *ppos; | ||
172 | ssize_t written, sz; | ||
173 | unsigned long copied; | ||
174 | void *ptr; | ||
175 | |||
176 | if (!valid_phys_addr_range(p, &count)) | ||
177 | return -EFAULT; | ||
178 | |||
179 | written = 0; | ||
180 | |||
181 | #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED | ||
182 | /* we don't have page 0 mapped on sparc and m68k.. */ | ||
183 | if (p < PAGE_SIZE) { | ||
184 | unsigned long sz = PAGE_SIZE - p; | ||
185 | if (sz > count) | ||
186 | sz = count; | ||
187 | /* Hmm. Do something? */ | ||
188 | buf += sz; | ||
189 | p += sz; | ||
190 | count -= sz; | ||
191 | written += sz; | ||
192 | } | ||
193 | #endif | ||
194 | |||
195 | while (count > 0) { | ||
196 | /* | ||
197 | * Handle first page in case it's not aligned | ||
198 | */ | ||
199 | if (-p & (PAGE_SIZE - 1)) | ||
200 | sz = -p & (PAGE_SIZE - 1); | ||
201 | else | ||
202 | sz = PAGE_SIZE; | ||
203 | |||
204 | sz = min_t(unsigned long, sz, count); | ||
205 | |||
206 | /* | ||
207 | * On ia64 if a page has been mapped somewhere as | ||
208 | * uncached, then it must also be accessed uncached | ||
209 | * by the kernel or data corruption may occur | ||
210 | */ | ||
211 | ptr = xlate_dev_mem_ptr(p); | ||
212 | |||
213 | copied = copy_from_user(ptr, buf, sz); | ||
214 | if (copied) { | ||
215 | ssize_t ret; | ||
216 | |||
217 | ret = written + (sz - copied); | ||
218 | if (ret) | ||
219 | return ret; | ||
220 | return -EFAULT; | ||
221 | } | ||
222 | buf += sz; | ||
223 | p += sz; | ||
224 | count -= sz; | ||
225 | written += sz; | ||
226 | } | ||
227 | |||
228 | *ppos += written; | ||
229 | return written; | ||
230 | } | ||
231 | |||
232 | static int mmap_mem(struct file * file, struct vm_area_struct * vma) | ||
233 | { | ||
234 | #if defined(__HAVE_PHYS_MEM_ACCESS_PROT) | ||
235 | unsigned long offset = vma->vm_pgoff << PAGE_SHIFT; | ||
236 | |||
237 | vma->vm_page_prot = phys_mem_access_prot(file, offset, | ||
238 | vma->vm_end - vma->vm_start, | ||
239 | vma->vm_page_prot); | ||
240 | #elif defined(pgprot_noncached) | ||
241 | unsigned long offset = vma->vm_pgoff << PAGE_SHIFT; | ||
242 | int uncached; | ||
243 | |||
244 | uncached = uncached_access(file, offset); | ||
245 | if (uncached) | ||
246 | vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot); | ||
247 | #endif | ||
248 | |||
249 | /* Remap-pfn-range will mark the range VM_IO and VM_RESERVED */ | ||
250 | if (remap_pfn_range(vma, | ||
251 | vma->vm_start, | ||
252 | vma->vm_pgoff, | ||
253 | vma->vm_end-vma->vm_start, | ||
254 | vma->vm_page_prot)) | ||
255 | return -EAGAIN; | ||
256 | return 0; | ||
257 | } | ||
258 | |||
259 | static int mmap_kmem(struct file * file, struct vm_area_struct * vma) | ||
260 | { | ||
261 | unsigned long long val; | ||
262 | /* | ||
263 | * RED-PEN: on some architectures there is more mapped memory | ||
264 | * than available in mem_map which pfn_valid checks | ||
265 | * for. Perhaps should add a new macro here. | ||
266 | * | ||
267 | * RED-PEN: vmalloc is not supported right now. | ||
268 | */ | ||
269 | if (!pfn_valid(vma->vm_pgoff)) | ||
270 | return -EIO; | ||
271 | val = (u64)vma->vm_pgoff << PAGE_SHIFT; | ||
272 | vma->vm_pgoff = __pa(val) >> PAGE_SHIFT; | ||
273 | return mmap_mem(file, vma); | ||
274 | } | ||
275 | |||
276 | extern long vread(char *buf, char *addr, unsigned long count); | ||
277 | extern long vwrite(char *buf, char *addr, unsigned long count); | ||
278 | |||
279 | /* | ||
280 | * This function reads the *virtual* memory as seen by the kernel. | ||
281 | */ | ||
282 | static ssize_t read_kmem(struct file *file, char __user *buf, | ||
283 | size_t count, loff_t *ppos) | ||
284 | { | ||
285 | unsigned long p = *ppos; | ||
286 | ssize_t low_count, read, sz; | ||
287 | char * kbuf; /* k-addr because vread() takes vmlist_lock rwlock */ | ||
288 | |||
289 | read = 0; | ||
290 | if (p < (unsigned long) high_memory) { | ||
291 | low_count = count; | ||
292 | if (count > (unsigned long) high_memory - p) | ||
293 | low_count = (unsigned long) high_memory - p; | ||
294 | |||
295 | #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED | ||
296 | /* we don't have page 0 mapped on sparc and m68k.. */ | ||
297 | if (p < PAGE_SIZE && low_count > 0) { | ||
298 | size_t tmp = PAGE_SIZE - p; | ||
299 | if (tmp > low_count) tmp = low_count; | ||
300 | if (clear_user(buf, tmp)) | ||
301 | return -EFAULT; | ||
302 | buf += tmp; | ||
303 | p += tmp; | ||
304 | read += tmp; | ||
305 | low_count -= tmp; | ||
306 | count -= tmp; | ||
307 | } | ||
308 | #endif | ||
309 | while (low_count > 0) { | ||
310 | /* | ||
311 | * Handle first page in case it's not aligned | ||
312 | */ | ||
313 | if (-p & (PAGE_SIZE - 1)) | ||
314 | sz = -p & (PAGE_SIZE - 1); | ||
315 | else | ||
316 | sz = PAGE_SIZE; | ||
317 | |||
318 | sz = min_t(unsigned long, sz, low_count); | ||
319 | |||
320 | /* | ||
321 | * On ia64 if a page has been mapped somewhere as | ||
322 | * uncached, then it must also be accessed uncached | ||
323 | * by the kernel or data corruption may occur | ||
324 | */ | ||
325 | kbuf = xlate_dev_kmem_ptr((char *)p); | ||
326 | |||
327 | if (copy_to_user(buf, kbuf, sz)) | ||
328 | return -EFAULT; | ||
329 | buf += sz; | ||
330 | p += sz; | ||
331 | read += sz; | ||
332 | low_count -= sz; | ||
333 | count -= sz; | ||
334 | } | ||
335 | } | ||
336 | |||
337 | if (count > 0) { | ||
338 | kbuf = (char *)__get_free_page(GFP_KERNEL); | ||
339 | if (!kbuf) | ||
340 | return -ENOMEM; | ||
341 | while (count > 0) { | ||
342 | int len = count; | ||
343 | |||
344 | if (len > PAGE_SIZE) | ||
345 | len = PAGE_SIZE; | ||
346 | len = vread(kbuf, (char *)p, len); | ||
347 | if (!len) | ||
348 | break; | ||
349 | if (copy_to_user(buf, kbuf, len)) { | ||
350 | free_page((unsigned long)kbuf); | ||
351 | return -EFAULT; | ||
352 | } | ||
353 | count -= len; | ||
354 | buf += len; | ||
355 | read += len; | ||
356 | p += len; | ||
357 | } | ||
358 | free_page((unsigned long)kbuf); | ||
359 | } | ||
360 | *ppos = p; | ||
361 | return read; | ||
362 | } | ||
363 | |||
364 | |||
365 | static inline ssize_t | ||
366 | do_write_kmem(void *p, unsigned long realp, const char __user * buf, | ||
367 | size_t count, loff_t *ppos) | ||
368 | { | ||
369 | ssize_t written, sz; | ||
370 | unsigned long copied; | ||
371 | |||
372 | written = 0; | ||
373 | #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED | ||
374 | /* we don't have page 0 mapped on sparc and m68k.. */ | ||
375 | if (realp < PAGE_SIZE) { | ||
376 | unsigned long sz = PAGE_SIZE - realp; | ||
377 | if (sz > count) | ||
378 | sz = count; | ||
379 | /* Hmm. Do something? */ | ||
380 | buf += sz; | ||
381 | p += sz; | ||
382 | realp += sz; | ||
383 | count -= sz; | ||
384 | written += sz; | ||
385 | } | ||
386 | #endif | ||
387 | |||
388 | while (count > 0) { | ||
389 | char *ptr; | ||
390 | /* | ||
391 | * Handle first page in case it's not aligned | ||
392 | */ | ||
393 | if (-realp & (PAGE_SIZE - 1)) | ||
394 | sz = -realp & (PAGE_SIZE - 1); | ||
395 | else | ||
396 | sz = PAGE_SIZE; | ||
397 | |||
398 | sz = min_t(unsigned long, sz, count); | ||
399 | |||
400 | /* | ||
401 | * On ia64 if a page has been mapped somewhere as | ||
402 | * uncached, then it must also be accessed uncached | ||
403 | * by the kernel or data corruption may occur | ||
404 | */ | ||
405 | ptr = xlate_dev_kmem_ptr(p); | ||
406 | |||
407 | copied = copy_from_user(ptr, buf, sz); | ||
408 | if (copied) { | ||
409 | ssize_t ret; | ||
410 | |||
411 | ret = written + (sz - copied); | ||
412 | if (ret) | ||
413 | return ret; | ||
414 | return -EFAULT; | ||
415 | } | ||
416 | buf += sz; | ||
417 | p += sz; | ||
418 | realp += sz; | ||
419 | count -= sz; | ||
420 | written += sz; | ||
421 | } | ||
422 | |||
423 | *ppos += written; | ||
424 | return written; | ||
425 | } | ||
426 | |||
427 | |||
428 | /* | ||
429 | * This function writes to the *virtual* memory as seen by the kernel. | ||
430 | */ | ||
431 | static ssize_t write_kmem(struct file * file, const char __user * buf, | ||
432 | size_t count, loff_t *ppos) | ||
433 | { | ||
434 | unsigned long p = *ppos; | ||
435 | ssize_t wrote = 0; | ||
436 | ssize_t virtr = 0; | ||
437 | ssize_t written; | ||
438 | char * kbuf; /* k-addr because vwrite() takes vmlist_lock rwlock */ | ||
439 | |||
440 | if (p < (unsigned long) high_memory) { | ||
441 | |||
442 | wrote = count; | ||
443 | if (count > (unsigned long) high_memory - p) | ||
444 | wrote = (unsigned long) high_memory - p; | ||
445 | |||
446 | written = do_write_kmem((void*)p, p, buf, wrote, ppos); | ||
447 | if (written != wrote) | ||
448 | return written; | ||
449 | wrote = written; | ||
450 | p += wrote; | ||
451 | buf += wrote; | ||
452 | count -= wrote; | ||
453 | } | ||
454 | |||
455 | if (count > 0) { | ||
456 | kbuf = (char *)__get_free_page(GFP_KERNEL); | ||
457 | if (!kbuf) | ||
458 | return wrote ? wrote : -ENOMEM; | ||
459 | while (count > 0) { | ||
460 | int len = count; | ||
461 | |||
462 | if (len > PAGE_SIZE) | ||
463 | len = PAGE_SIZE; | ||
464 | if (len) { | ||
465 | written = copy_from_user(kbuf, buf, len); | ||
466 | if (written) { | ||
467 | ssize_t ret; | ||
468 | |||
469 | free_page((unsigned long)kbuf); | ||
470 | ret = wrote + virtr + (len - written); | ||
471 | return ret ? ret : -EFAULT; | ||
472 | } | ||
473 | } | ||
474 | len = vwrite(kbuf, (char *)p, len); | ||
475 | count -= len; | ||
476 | buf += len; | ||
477 | virtr += len; | ||
478 | p += len; | ||
479 | } | ||
480 | free_page((unsigned long)kbuf); | ||
481 | } | ||
482 | |||
483 | *ppos = p; | ||
484 | return virtr + wrote; | ||
485 | } | ||
486 | |||
487 | #if defined(CONFIG_ISA) || !defined(__mc68000__) | ||
488 | static ssize_t read_port(struct file * file, char __user * buf, | ||
489 | size_t count, loff_t *ppos) | ||
490 | { | ||
491 | unsigned long i = *ppos; | ||
492 | char __user *tmp = buf; | ||
493 | |||
494 | if (!access_ok(VERIFY_WRITE, buf, count)) | ||
495 | return -EFAULT; | ||
496 | while (count-- > 0 && i < 65536) { | ||
497 | if (__put_user(inb(i),tmp) < 0) | ||
498 | return -EFAULT; | ||
499 | i++; | ||
500 | tmp++; | ||
501 | } | ||
502 | *ppos = i; | ||
503 | return tmp-buf; | ||
504 | } | ||
505 | |||
506 | static ssize_t write_port(struct file * file, const char __user * buf, | ||
507 | size_t count, loff_t *ppos) | ||
508 | { | ||
509 | unsigned long i = *ppos; | ||
510 | const char __user * tmp = buf; | ||
511 | |||
512 | if (!access_ok(VERIFY_READ,buf,count)) | ||
513 | return -EFAULT; | ||
514 | while (count-- > 0 && i < 65536) { | ||
515 | char c; | ||
516 | if (__get_user(c, tmp)) | ||
517 | return -EFAULT; | ||
518 | outb(c,i); | ||
519 | i++; | ||
520 | tmp++; | ||
521 | } | ||
522 | *ppos = i; | ||
523 | return tmp-buf; | ||
524 | } | ||
525 | #endif | ||
526 | |||
527 | static ssize_t read_null(struct file * file, char __user * buf, | ||
528 | size_t count, loff_t *ppos) | ||
529 | { | ||
530 | return 0; | ||
531 | } | ||
532 | |||
533 | static ssize_t write_null(struct file * file, const char __user * buf, | ||
534 | size_t count, loff_t *ppos) | ||
535 | { | ||
536 | return count; | ||
537 | } | ||
538 | |||
539 | #ifdef CONFIG_MMU | ||
540 | /* | ||
541 | * For fun, we are using the MMU for this. | ||
542 | */ | ||
543 | static inline size_t read_zero_pagealigned(char __user * buf, size_t size) | ||
544 | { | ||
545 | struct mm_struct *mm; | ||
546 | struct vm_area_struct * vma; | ||
547 | unsigned long addr=(unsigned long)buf; | ||
548 | |||
549 | mm = current->mm; | ||
550 | /* Oops, this was forgotten before. -ben */ | ||
551 | down_read(&mm->mmap_sem); | ||
552 | |||
553 | /* For private mappings, just map in zero pages. */ | ||
554 | for (vma = find_vma(mm, addr); vma; vma = vma->vm_next) { | ||
555 | unsigned long count; | ||
556 | |||
557 | if (vma->vm_start > addr || (vma->vm_flags & VM_WRITE) == 0) | ||
558 | goto out_up; | ||
559 | if (vma->vm_flags & (VM_SHARED | VM_HUGETLB)) | ||
560 | break; | ||
561 | count = vma->vm_end - addr; | ||
562 | if (count > size) | ||
563 | count = size; | ||
564 | |||
565 | zap_page_range(vma, addr, count, NULL); | ||
566 | zeromap_page_range(vma, addr, count, PAGE_COPY); | ||
567 | |||
568 | size -= count; | ||
569 | buf += count; | ||
570 | addr += count; | ||
571 | if (size == 0) | ||
572 | goto out_up; | ||
573 | } | ||
574 | |||
575 | up_read(&mm->mmap_sem); | ||
576 | |||
577 | /* The shared case is hard. Let's do the conventional zeroing. */ | ||
578 | do { | ||
579 | unsigned long unwritten = clear_user(buf, PAGE_SIZE); | ||
580 | if (unwritten) | ||
581 | return size + unwritten - PAGE_SIZE; | ||
582 | cond_resched(); | ||
583 | buf += PAGE_SIZE; | ||
584 | size -= PAGE_SIZE; | ||
585 | } while (size); | ||
586 | |||
587 | return size; | ||
588 | out_up: | ||
589 | up_read(&mm->mmap_sem); | ||
590 | return size; | ||
591 | } | ||
592 | |||
593 | static ssize_t read_zero(struct file * file, char __user * buf, | ||
594 | size_t count, loff_t *ppos) | ||
595 | { | ||
596 | unsigned long left, unwritten, written = 0; | ||
597 | |||
598 | if (!count) | ||
599 | return 0; | ||
600 | |||
601 | if (!access_ok(VERIFY_WRITE, buf, count)) | ||
602 | return -EFAULT; | ||
603 | |||
604 | left = count; | ||
605 | |||
606 | /* do we want to be clever? Arbitrary cut-off */ | ||
607 | if (count >= PAGE_SIZE*4) { | ||
608 | unsigned long partial; | ||
609 | |||
610 | /* How much left of the page? */ | ||
611 | partial = (PAGE_SIZE-1) & -(unsigned long) buf; | ||
612 | unwritten = clear_user(buf, partial); | ||
613 | written = partial - unwritten; | ||
614 | if (unwritten) | ||
615 | goto out; | ||
616 | left -= partial; | ||
617 | buf += partial; | ||
618 | unwritten = read_zero_pagealigned(buf, left & PAGE_MASK); | ||
619 | written += (left & PAGE_MASK) - unwritten; | ||
620 | if (unwritten) | ||
621 | goto out; | ||
622 | buf += left & PAGE_MASK; | ||
623 | left &= ~PAGE_MASK; | ||
624 | } | ||
625 | unwritten = clear_user(buf, left); | ||
626 | written += left - unwritten; | ||
627 | out: | ||
628 | return written ? written : -EFAULT; | ||
629 | } | ||
630 | |||
631 | static int mmap_zero(struct file * file, struct vm_area_struct * vma) | ||
632 | { | ||
633 | if (vma->vm_flags & VM_SHARED) | ||
634 | return shmem_zero_setup(vma); | ||
635 | if (zeromap_page_range(vma, vma->vm_start, vma->vm_end - vma->vm_start, vma->vm_page_prot)) | ||
636 | return -EAGAIN; | ||
637 | return 0; | ||
638 | } | ||
639 | #else /* CONFIG_MMU */ | ||
640 | static ssize_t read_zero(struct file * file, char * buf, | ||
641 | size_t count, loff_t *ppos) | ||
642 | { | ||
643 | size_t todo = count; | ||
644 | |||
645 | while (todo) { | ||
646 | size_t chunk = todo; | ||
647 | |||
648 | if (chunk > 4096) | ||
649 | chunk = 4096; /* Just for latency reasons */ | ||
650 | if (clear_user(buf, chunk)) | ||
651 | return -EFAULT; | ||
652 | buf += chunk; | ||
653 | todo -= chunk; | ||
654 | cond_resched(); | ||
655 | } | ||
656 | return count; | ||
657 | } | ||
658 | |||
659 | static int mmap_zero(struct file * file, struct vm_area_struct * vma) | ||
660 | { | ||
661 | return -ENOSYS; | ||
662 | } | ||
663 | #endif /* CONFIG_MMU */ | ||
664 | |||
665 | static ssize_t write_full(struct file * file, const char __user * buf, | ||
666 | size_t count, loff_t *ppos) | ||
667 | { | ||
668 | return -ENOSPC; | ||
669 | } | ||
670 | |||
671 | /* | ||
672 | * Special lseek() function for /dev/null and /dev/zero. Most notably, you | ||
673 | * can fopen() both devices with "a" now. This was previously impossible. | ||
674 | * -- SRB. | ||
675 | */ | ||
676 | |||
677 | static loff_t null_lseek(struct file * file, loff_t offset, int orig) | ||
678 | { | ||
679 | return file->f_pos = 0; | ||
680 | } | ||
681 | |||
682 | /* | ||
683 | * The memory devices use the full 32/64 bits of the offset, and so we cannot | ||
684 | * check against negative addresses: they are ok. The return value is weird, | ||
685 | * though, in that case (0). | ||
686 | * | ||
687 | * also note that seeking relative to the "end of file" isn't supported: | ||
688 | * it has no meaning, so it returns -EINVAL. | ||
689 | */ | ||
690 | static loff_t memory_lseek(struct file * file, loff_t offset, int orig) | ||
691 | { | ||
692 | loff_t ret; | ||
693 | |||
694 | down(&file->f_dentry->d_inode->i_sem); | ||
695 | switch (orig) { | ||
696 | case 0: | ||
697 | file->f_pos = offset; | ||
698 | ret = file->f_pos; | ||
699 | force_successful_syscall_return(); | ||
700 | break; | ||
701 | case 1: | ||
702 | file->f_pos += offset; | ||
703 | ret = file->f_pos; | ||
704 | force_successful_syscall_return(); | ||
705 | break; | ||
706 | default: | ||
707 | ret = -EINVAL; | ||
708 | } | ||
709 | up(&file->f_dentry->d_inode->i_sem); | ||
710 | return ret; | ||
711 | } | ||
712 | |||
713 | static int open_port(struct inode * inode, struct file * filp) | ||
714 | { | ||
715 | return capable(CAP_SYS_RAWIO) ? 0 : -EPERM; | ||
716 | } | ||
717 | |||
718 | #define zero_lseek null_lseek | ||
719 | #define full_lseek null_lseek | ||
720 | #define write_zero write_null | ||
721 | #define read_full read_zero | ||
722 | #define open_mem open_port | ||
723 | #define open_kmem open_mem | ||
724 | |||
725 | static struct file_operations mem_fops = { | ||
726 | .llseek = memory_lseek, | ||
727 | .read = read_mem, | ||
728 | .write = write_mem, | ||
729 | .mmap = mmap_mem, | ||
730 | .open = open_mem, | ||
731 | }; | ||
732 | |||
733 | static struct file_operations kmem_fops = { | ||
734 | .llseek = memory_lseek, | ||
735 | .read = read_kmem, | ||
736 | .write = write_kmem, | ||
737 | .mmap = mmap_kmem, | ||
738 | .open = open_kmem, | ||
739 | }; | ||
740 | |||
741 | static struct file_operations null_fops = { | ||
742 | .llseek = null_lseek, | ||
743 | .read = read_null, | ||
744 | .write = write_null, | ||
745 | }; | ||
746 | |||
747 | #if defined(CONFIG_ISA) || !defined(__mc68000__) | ||
748 | static struct file_operations port_fops = { | ||
749 | .llseek = memory_lseek, | ||
750 | .read = read_port, | ||
751 | .write = write_port, | ||
752 | .open = open_port, | ||
753 | }; | ||
754 | #endif | ||
755 | |||
756 | static struct file_operations zero_fops = { | ||
757 | .llseek = zero_lseek, | ||
758 | .read = read_zero, | ||
759 | .write = write_zero, | ||
760 | .mmap = mmap_zero, | ||
761 | }; | ||
762 | |||
763 | static struct backing_dev_info zero_bdi = { | ||
764 | .capabilities = BDI_CAP_MAP_COPY, | ||
765 | }; | ||
766 | |||
767 | static struct file_operations full_fops = { | ||
768 | .llseek = full_lseek, | ||
769 | .read = read_full, | ||
770 | .write = write_full, | ||
771 | }; | ||
772 | |||
773 | static ssize_t kmsg_write(struct file * file, const char __user * buf, | ||
774 | size_t count, loff_t *ppos) | ||
775 | { | ||
776 | char *tmp; | ||
777 | int ret; | ||
778 | |||
779 | tmp = kmalloc(count + 1, GFP_KERNEL); | ||
780 | if (tmp == NULL) | ||
781 | return -ENOMEM; | ||
782 | ret = -EFAULT; | ||
783 | if (!copy_from_user(tmp, buf, count)) { | ||
784 | tmp[count] = 0; | ||
785 | ret = printk("%s", tmp); | ||
786 | } | ||
787 | kfree(tmp); | ||
788 | return ret; | ||
789 | } | ||
790 | |||
791 | static struct file_operations kmsg_fops = { | ||
792 | .write = kmsg_write, | ||
793 | }; | ||
794 | |||
795 | static int memory_open(struct inode * inode, struct file * filp) | ||
796 | { | ||
797 | switch (iminor(inode)) { | ||
798 | case 1: | ||
799 | filp->f_op = &mem_fops; | ||
800 | break; | ||
801 | case 2: | ||
802 | filp->f_op = &kmem_fops; | ||
803 | break; | ||
804 | case 3: | ||
805 | filp->f_op = &null_fops; | ||
806 | break; | ||
807 | #if defined(CONFIG_ISA) || !defined(__mc68000__) | ||
808 | case 4: | ||
809 | filp->f_op = &port_fops; | ||
810 | break; | ||
811 | #endif | ||
812 | case 5: | ||
813 | filp->f_mapping->backing_dev_info = &zero_bdi; | ||
814 | filp->f_op = &zero_fops; | ||
815 | break; | ||
816 | case 7: | ||
817 | filp->f_op = &full_fops; | ||
818 | break; | ||
819 | case 8: | ||
820 | filp->f_op = &random_fops; | ||
821 | break; | ||
822 | case 9: | ||
823 | filp->f_op = &urandom_fops; | ||
824 | break; | ||
825 | case 11: | ||
826 | filp->f_op = &kmsg_fops; | ||
827 | break; | ||
828 | default: | ||
829 | return -ENXIO; | ||
830 | } | ||
831 | if (filp->f_op && filp->f_op->open) | ||
832 | return filp->f_op->open(inode,filp); | ||
833 | return 0; | ||
834 | } | ||
835 | |||
836 | static struct file_operations memory_fops = { | ||
837 | .open = memory_open, /* just a selector for the real open */ | ||
838 | }; | ||
839 | |||
840 | static const struct { | ||
841 | unsigned int minor; | ||
842 | char *name; | ||
843 | umode_t mode; | ||
844 | struct file_operations *fops; | ||
845 | } devlist[] = { /* list of minor devices */ | ||
846 | {1, "mem", S_IRUSR | S_IWUSR | S_IRGRP, &mem_fops}, | ||
847 | {2, "kmem", S_IRUSR | S_IWUSR | S_IRGRP, &kmem_fops}, | ||
848 | {3, "null", S_IRUGO | S_IWUGO, &null_fops}, | ||
849 | #if defined(CONFIG_ISA) || !defined(__mc68000__) | ||
850 | {4, "port", S_IRUSR | S_IWUSR | S_IRGRP, &port_fops}, | ||
851 | #endif | ||
852 | {5, "zero", S_IRUGO | S_IWUGO, &zero_fops}, | ||
853 | {7, "full", S_IRUGO | S_IWUGO, &full_fops}, | ||
854 | {8, "random", S_IRUGO | S_IWUSR, &random_fops}, | ||
855 | {9, "urandom", S_IRUGO | S_IWUSR, &urandom_fops}, | ||
856 | {11,"kmsg", S_IRUGO | S_IWUSR, &kmsg_fops}, | ||
857 | }; | ||
858 | |||
859 | static struct class_simple *mem_class; | ||
860 | |||
861 | static int __init chr_dev_init(void) | ||
862 | { | ||
863 | int i; | ||
864 | |||
865 | if (register_chrdev(MEM_MAJOR,"mem",&memory_fops)) | ||
866 | printk("unable to get major %d for memory devs\n", MEM_MAJOR); | ||
867 | |||
868 | mem_class = class_simple_create(THIS_MODULE, "mem"); | ||
869 | for (i = 0; i < ARRAY_SIZE(devlist); i++) { | ||
870 | class_simple_device_add(mem_class, | ||
871 | MKDEV(MEM_MAJOR, devlist[i].minor), | ||
872 | NULL, devlist[i].name); | ||
873 | devfs_mk_cdev(MKDEV(MEM_MAJOR, devlist[i].minor), | ||
874 | S_IFCHR | devlist[i].mode, devlist[i].name); | ||
875 | } | ||
876 | |||
877 | return 0; | ||
878 | } | ||
879 | |||
880 | fs_initcall(chr_dev_init); | ||