vserver 1.9.5.x5
[linux-2.6.git] / drivers / char / mem.c
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
27 #include <asm/uaccess.h>
28 #include <asm/io.h>
29
30 #ifdef CONFIG_IA64
31 # include <linux/efi.h>
32 #endif
33
34 #if defined(CONFIG_S390_TAPE) && defined(CONFIG_S390_TAPE_CHAR)
35 extern void tapechar_init(void);
36 #endif
37
38 /*
39  * Architectures vary in how they handle caching for addresses
40  * outside of main memory.
41  *
42  */
43 static inline int uncached_access(struct file *file, unsigned long addr)
44 {
45 #if defined(__i386__)
46         /*
47          * On the PPro and successors, the MTRRs are used to set
48          * memory types for physical addresses outside main memory,
49          * so blindly setting PCD or PWT on those pages is wrong.
50          * For Pentiums and earlier, the surround logic should disable
51          * caching for the high addresses through the KEN pin, but
52          * we maintain the tradition of paranoia in this code.
53          */
54         if (file->f_flags & O_SYNC)
55                 return 1;
56         return !( test_bit(X86_FEATURE_MTRR, boot_cpu_data.x86_capability) ||
57                   test_bit(X86_FEATURE_K6_MTRR, boot_cpu_data.x86_capability) ||
58                   test_bit(X86_FEATURE_CYRIX_ARR, boot_cpu_data.x86_capability) ||
59                   test_bit(X86_FEATURE_CENTAUR_MCR, boot_cpu_data.x86_capability) )
60           && addr >= __pa(high_memory);
61 #elif defined(__x86_64__)
62         /* 
63          * This is broken because it can generate memory type aliases,
64          * which can cause cache corruptions
65          * But it is only available for root and we have to be bug-to-bug
66          * compatible with i386.
67          */
68         if (file->f_flags & O_SYNC)
69                 return 1;
70         /* same behaviour as i386. PAT always set to cached and MTRRs control the
71            caching behaviour. 
72            Hopefully a full PAT implementation will fix that soon. */      
73         return 0;
74 #elif defined(CONFIG_IA64)
75         /*
76          * On ia64, we ignore O_SYNC because we cannot tolerate memory attribute aliases.
77          */
78         return !(efi_mem_attributes(addr) & EFI_MEMORY_WB);
79 #elif defined(CONFIG_PPC64)
80         /* On PPC64, we always do non-cacheable access to the IO hole and
81          * cacheable elsewhere. Cache paradox can checkstop the CPU and
82          * the high_memory heuristic below is wrong on machines with memory
83          * above the IO hole... Ah, and of course, XFree86 doesn't pass
84          * O_SYNC when mapping us to tap IO space. Surprised ?
85          */
86         return !page_is_ram(addr >> PAGE_SHIFT);
87 #else
88         /*
89          * Accessing memory above the top the kernel knows about or through a file pointer
90          * that was marked O_SYNC will be done non-cached.
91          */
92         if (file->f_flags & O_SYNC)
93                 return 1;
94         return addr >= __pa(high_memory);
95 #endif
96 }
97
98 #ifndef ARCH_HAS_VALID_PHYS_ADDR_RANGE
99 static inline int valid_phys_addr_range(unsigned long addr, size_t *count)
100 {
101         unsigned long end_mem;
102
103         end_mem = __pa(high_memory);
104         if (addr >= end_mem)
105                 return 0;
106
107         if (*count > end_mem - addr)
108                 *count = end_mem - addr;
109
110         return 1;
111 }
112 #endif
113
114 static ssize_t do_write_mem(void *p, unsigned long realp,
115                             const char __user * buf, size_t count, loff_t *ppos)
116 {
117         ssize_t written;
118         unsigned long copied;
119
120         written = 0;
121 #if defined(__sparc__) || (defined(__mc68000__) && defined(CONFIG_MMU))
122         /* we don't have page 0 mapped on sparc and m68k.. */
123         if (realp < PAGE_SIZE) {
124                 unsigned long sz = PAGE_SIZE-realp;
125                 if (sz > count) sz = count; 
126                 /* Hmm. Do something? */
127                 buf+=sz;
128                 p+=sz;
129                 count-=sz;
130                 written+=sz;
131         }
132 #endif
133         copied = copy_from_user(p, buf, count);
134         if (copied) {
135                 ssize_t ret = written + (count - copied);
136
137                 if (ret)
138                         return ret;
139                 return -EFAULT;
140         }
141         written += count;
142         *ppos += written;
143         return written;
144 }
145
146
147 /*
148  * This funcion reads the *physical* memory. The f_pos points directly to the 
149  * memory location. 
150  */
151 static ssize_t read_mem(struct file * file, char __user * buf,
152                         size_t count, loff_t *ppos)
153 {
154         unsigned long p = *ppos;
155         ssize_t read;
156
157         if (!valid_phys_addr_range(p, &count))
158                 return -EFAULT;
159         read = 0;
160 #if defined(__sparc__) || (defined(__mc68000__) && defined(CONFIG_MMU))
161         /* we don't have page 0 mapped on sparc and m68k.. */
162         if (p < PAGE_SIZE) {
163                 unsigned long sz = PAGE_SIZE-p;
164                 if (sz > count) 
165                         sz = count; 
166                 if (sz > 0) {
167                         if (clear_user(buf, sz))
168                                 return -EFAULT;
169                         buf += sz; 
170                         p += sz; 
171                         count -= sz; 
172                         read += sz; 
173                 }
174         }
175 #endif
176         if (copy_to_user(buf, __va(p), count))
177                 return -EFAULT;
178         read += count;
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
188         if (!valid_phys_addr_range(p, &count))
189                 return -EFAULT;
190         return do_write_mem(__va(p), p, buf, count, ppos);
191 }
192
193 static int mmap_mem(struct file * file, struct vm_area_struct * vma)
194 {
195 #ifdef pgprot_noncached
196         unsigned long offset = vma->vm_pgoff << PAGE_SHIFT;
197         int uncached;
198
199         uncached = uncached_access(file, offset);
200         if (uncached)
201                 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
202 #endif
203
204         /* Remap-pfn-range will mark the range VM_IO and VM_RESERVED */
205         if (remap_pfn_range(vma,
206                             vma->vm_start,
207                             vma->vm_pgoff,
208                             vma->vm_end-vma->vm_start,
209                             vma->vm_page_prot))
210                 return -EAGAIN;
211         return 0;
212 }
213
214 extern long vread(char *buf, char *addr, unsigned long count);
215 extern long vwrite(char *buf, char *addr, unsigned long count);
216
217 /*
218  * This function reads the *virtual* memory as seen by the kernel.
219  */
220 static ssize_t read_kmem(struct file *file, char __user *buf, 
221                          size_t count, loff_t *ppos)
222 {
223         unsigned long p = *ppos;
224         ssize_t read = 0;
225         ssize_t virtr = 0;
226         char * kbuf; /* k-addr because vread() takes vmlist_lock rwlock */
227                 
228         if (p < (unsigned long) high_memory) {
229                 read = count;
230                 if (count > (unsigned long) high_memory - p)
231                         read = (unsigned long) high_memory - p;
232
233 #if defined(__sparc__) || (defined(__mc68000__) && defined(CONFIG_MMU))
234                 /* we don't have page 0 mapped on sparc and m68k.. */
235                 if (p < PAGE_SIZE && read > 0) {
236                         size_t tmp = PAGE_SIZE - p;
237                         if (tmp > read) tmp = read;
238                         if (clear_user(buf, tmp))
239                                 return -EFAULT;
240                         buf += tmp;
241                         p += tmp;
242                         read -= tmp;
243                         count -= tmp;
244                 }
245 #endif
246                 if (copy_to_user(buf, (char *)p, read))
247                         return -EFAULT;
248                 p += read;
249                 buf += read;
250                 count -= read;
251         }
252
253         if (count > 0) {
254                 kbuf = (char *)__get_free_page(GFP_KERNEL);
255                 if (!kbuf)
256                         return -ENOMEM;
257                 while (count > 0) {
258                         int len = count;
259
260                         if (len > PAGE_SIZE)
261                                 len = PAGE_SIZE;
262                         len = vread(kbuf, (char *)p, len);
263                         if (!len)
264                                 break;
265                         if (copy_to_user(buf, kbuf, len)) {
266                                 free_page((unsigned long)kbuf);
267                                 return -EFAULT;
268                         }
269                         count -= len;
270                         buf += len;
271                         virtr += len;
272                         p += len;
273                 }
274                 free_page((unsigned long)kbuf);
275         }
276         *ppos = p;
277         return virtr + read;
278 }
279
280 /*
281  * This function writes to the *virtual* memory as seen by the kernel.
282  */
283 static ssize_t write_kmem(struct file * file, const char __user * buf, 
284                           size_t count, loff_t *ppos)
285 {
286         unsigned long p = *ppos;
287         ssize_t wrote = 0;
288         ssize_t virtr = 0;
289         ssize_t written;
290         char * kbuf; /* k-addr because vwrite() takes vmlist_lock rwlock */
291
292         if (p < (unsigned long) high_memory) {
293
294                 wrote = count;
295                 if (count > (unsigned long) high_memory - p)
296                         wrote = (unsigned long) high_memory - p;
297
298                 written = do_write_mem((void*)p, p, buf, wrote, ppos);
299                 if (written != wrote)
300                         return written;
301                 wrote = written;
302                 p += wrote;
303                 buf += wrote;
304                 count -= wrote;
305         }
306
307         if (count > 0) {
308                 kbuf = (char *)__get_free_page(GFP_KERNEL);
309                 if (!kbuf)
310                         return wrote ? wrote : -ENOMEM;
311                 while (count > 0) {
312                         int len = count;
313
314                         if (len > PAGE_SIZE)
315                                 len = PAGE_SIZE;
316                         if (len) {
317                                 written = copy_from_user(kbuf, buf, len);
318                                 if (written) {
319                                         ssize_t ret;
320
321                                         free_page((unsigned long)kbuf);
322                                         ret = wrote + virtr + (len - written);
323                                         return ret ? ret : -EFAULT;
324                                 }
325                         }
326                         len = vwrite(kbuf, (char *)p, len);
327                         count -= len;
328                         buf += len;
329                         virtr += len;
330                         p += len;
331                 }
332                 free_page((unsigned long)kbuf);
333         }
334
335         *ppos = p;
336         return virtr + wrote;
337 }
338
339 #if defined(CONFIG_ISA) || !defined(__mc68000__)
340 static ssize_t read_port(struct file * file, char __user * buf,
341                          size_t count, loff_t *ppos)
342 {
343         unsigned long i = *ppos;
344         char __user *tmp = buf;
345
346         if (verify_area(VERIFY_WRITE,buf,count))
347                 return -EFAULT; 
348         while (count-- > 0 && i < 65536) {
349                 if (__put_user(inb(i),tmp) < 0) 
350                         return -EFAULT;  
351                 i++;
352                 tmp++;
353         }
354         *ppos = i;
355         return tmp-buf;
356 }
357
358 static ssize_t write_port(struct file * file, const char __user * buf,
359                           size_t count, loff_t *ppos)
360 {
361         unsigned long i = *ppos;
362         const char __user * tmp = buf;
363
364         if (verify_area(VERIFY_READ,buf,count))
365                 return -EFAULT;
366         while (count-- > 0 && i < 65536) {
367                 char c;
368                 if (__get_user(c, tmp)) 
369                         return -EFAULT; 
370                 outb(c,i);
371                 i++;
372                 tmp++;
373         }
374         *ppos = i;
375         return tmp-buf;
376 }
377 #endif
378
379 static ssize_t read_null(struct file * file, char __user * buf,
380                          size_t count, loff_t *ppos)
381 {
382         return 0;
383 }
384
385 static ssize_t write_null(struct file * file, const char __user * buf,
386                           size_t count, loff_t *ppos)
387 {
388         return count;
389 }
390
391 #ifdef CONFIG_MMU
392 /*
393  * For fun, we are using the MMU for this.
394  */
395 static inline size_t read_zero_pagealigned(char __user * buf, size_t size)
396 {
397         struct mm_struct *mm;
398         struct vm_area_struct * vma;
399         unsigned long addr=(unsigned long)buf;
400
401         mm = current->mm;
402         /* Oops, this was forgotten before. -ben */
403         down_read(&mm->mmap_sem);
404
405         /* For private mappings, just map in zero pages. */
406         for (vma = find_vma(mm, addr); vma; vma = vma->vm_next) {
407                 unsigned long count;
408
409                 if (vma->vm_start > addr || (vma->vm_flags & VM_WRITE) == 0)
410                         goto out_up;
411                 if (vma->vm_flags & (VM_SHARED | VM_HUGETLB))
412                         break;
413                 count = vma->vm_end - addr;
414                 if (count > size)
415                         count = size;
416
417                 zap_page_range(vma, addr, count, NULL);
418                 zeromap_page_range(vma, addr, count, PAGE_COPY);
419
420                 size -= count;
421                 buf += count;
422                 addr += count;
423                 if (size == 0)
424                         goto out_up;
425         }
426
427         up_read(&mm->mmap_sem);
428         
429         /* The shared case is hard. Let's do the conventional zeroing. */ 
430         do {
431                 unsigned long unwritten = clear_user(buf, PAGE_SIZE);
432                 if (unwritten)
433                         return size + unwritten - PAGE_SIZE;
434                 cond_resched();
435                 buf += PAGE_SIZE;
436                 size -= PAGE_SIZE;
437         } while (size);
438
439         return size;
440 out_up:
441         up_read(&mm->mmap_sem);
442         return size;
443 }
444
445 static ssize_t read_zero(struct file * file, char __user * buf, 
446                          size_t count, loff_t *ppos)
447 {
448         unsigned long left, unwritten, written = 0;
449
450         if (!count)
451                 return 0;
452
453         if (!access_ok(VERIFY_WRITE, buf, count))
454                 return -EFAULT;
455
456         left = count;
457
458         /* do we want to be clever? Arbitrary cut-off */
459         if (count >= PAGE_SIZE*4) {
460                 unsigned long partial;
461
462                 /* How much left of the page? */
463                 partial = (PAGE_SIZE-1) & -(unsigned long) buf;
464                 unwritten = clear_user(buf, partial);
465                 written = partial - unwritten;
466                 if (unwritten)
467                         goto out;
468                 left -= partial;
469                 buf += partial;
470                 unwritten = read_zero_pagealigned(buf, left & PAGE_MASK);
471                 written += (left & PAGE_MASK) - unwritten;
472                 if (unwritten)
473                         goto out;
474                 buf += left & PAGE_MASK;
475                 left &= ~PAGE_MASK;
476         }
477         unwritten = clear_user(buf, left);
478         written += left - unwritten;
479 out:
480         return written ? written : -EFAULT;
481 }
482
483 static int mmap_zero(struct file * file, struct vm_area_struct * vma)
484 {
485         if (vma->vm_flags & VM_SHARED)
486                 return shmem_zero_setup(vma);
487         if (zeromap_page_range(vma, vma->vm_start, vma->vm_end - vma->vm_start, vma->vm_page_prot))
488                 return -EAGAIN;
489         return 0;
490 }
491 #else /* CONFIG_MMU */
492 static ssize_t read_zero(struct file * file, char * buf, 
493                          size_t count, loff_t *ppos)
494 {
495         size_t todo = count;
496
497         while (todo) {
498                 size_t chunk = todo;
499
500                 if (chunk > 4096)
501                         chunk = 4096;   /* Just for latency reasons */
502                 if (clear_user(buf, chunk))
503                         return -EFAULT;
504                 buf += chunk;
505                 todo -= chunk;
506                 cond_resched();
507         }
508         return count;
509 }
510
511 static int mmap_zero(struct file * file, struct vm_area_struct * vma)
512 {
513         return -ENOSYS;
514 }
515 #endif /* CONFIG_MMU */
516
517 static ssize_t write_full(struct file * file, const char __user * buf,
518                           size_t count, loff_t *ppos)
519 {
520         return -ENOSPC;
521 }
522
523 /*
524  * Special lseek() function for /dev/null and /dev/zero.  Most notably, you
525  * can fopen() both devices with "a" now.  This was previously impossible.
526  * -- SRB.
527  */
528
529 static loff_t null_lseek(struct file * file, loff_t offset, int orig)
530 {
531         return file->f_pos = 0;
532 }
533
534 /*
535  * The memory devices use the full 32/64 bits of the offset, and so we cannot
536  * check against negative addresses: they are ok. The return value is weird,
537  * though, in that case (0).
538  *
539  * also note that seeking relative to the "end of file" isn't supported:
540  * it has no meaning, so it returns -EINVAL.
541  */
542 static loff_t memory_lseek(struct file * file, loff_t offset, int orig)
543 {
544         loff_t ret;
545
546         down(&file->f_dentry->d_inode->i_sem);
547         switch (orig) {
548                 case 0:
549                         file->f_pos = offset;
550                         ret = file->f_pos;
551                         force_successful_syscall_return();
552                         break;
553                 case 1:
554                         file->f_pos += offset;
555                         ret = file->f_pos;
556                         force_successful_syscall_return();
557                         break;
558                 default:
559                         ret = -EINVAL;
560         }
561         up(&file->f_dentry->d_inode->i_sem);
562         return ret;
563 }
564
565 static int open_port(struct inode * inode, struct file * filp)
566 {
567         return capable(CAP_SYS_RAWIO) ? 0 : -EPERM;
568 }
569
570 #define mmap_kmem       mmap_mem
571 #define zero_lseek      null_lseek
572 #define full_lseek      null_lseek
573 #define write_zero      write_null
574 #define read_full       read_zero
575 #define open_mem        open_port
576 #define open_kmem       open_mem
577
578 static struct file_operations mem_fops = {
579         .llseek         = memory_lseek,
580         .read           = read_mem,
581         .write          = write_mem,
582         .mmap           = mmap_mem,
583         .open           = open_mem,
584 };
585
586 static struct file_operations kmem_fops = {
587         .llseek         = memory_lseek,
588         .read           = read_kmem,
589         .write          = write_kmem,
590         .mmap           = mmap_kmem,
591         .open           = open_kmem,
592 };
593
594 static struct file_operations null_fops = {
595         .llseek         = null_lseek,
596         .read           = read_null,
597         .write          = write_null,
598 };
599
600 #if defined(CONFIG_ISA) || !defined(__mc68000__)
601 static struct file_operations port_fops = {
602         .llseek         = memory_lseek,
603         .read           = read_port,
604         .write          = write_port,
605         .open           = open_port,
606 };
607 #endif
608
609 static struct file_operations zero_fops = {
610         .llseek         = zero_lseek,
611         .read           = read_zero,
612         .write          = write_zero,
613         .mmap           = mmap_zero,
614 };
615
616 static struct file_operations full_fops = {
617         .llseek         = full_lseek,
618         .read           = read_full,
619         .write          = write_full,
620 };
621
622 static ssize_t kmsg_write(struct file * file, const char __user * buf,
623                           size_t count, loff_t *ppos)
624 {
625         char *tmp;
626         int ret;
627
628         tmp = kmalloc(count + 1, GFP_KERNEL);
629         if (tmp == NULL)
630                 return -ENOMEM;
631         ret = -EFAULT;
632         if (!copy_from_user(tmp, buf, count)) {
633                 tmp[count] = 0;
634                 ret = printk("%s", tmp);
635         }
636         kfree(tmp);
637         return ret;
638 }
639
640 static struct file_operations kmsg_fops = {
641         .write =        kmsg_write,
642 };
643
644 static int memory_open(struct inode * inode, struct file * filp)
645 {
646         switch (iminor(inode)) {
647                 case 1:
648                         filp->f_op = &mem_fops;
649                         break;
650                 case 2:
651                         filp->f_op = &kmem_fops;
652                         break;
653                 case 3:
654                         filp->f_op = &null_fops;
655                         break;
656 #if defined(CONFIG_ISA) || !defined(__mc68000__)
657                 case 4:
658                         filp->f_op = &port_fops;
659                         break;
660 #endif
661                 case 5:
662                         filp->f_op = &zero_fops;
663                         break;
664                 case 7:
665                         filp->f_op = &full_fops;
666                         break;
667                 case 8:
668                         filp->f_op = &random_fops;
669                         break;
670                 case 9:
671                         filp->f_op = &urandom_fops;
672                         break;
673                 case 11:
674                         filp->f_op = &kmsg_fops;
675                         break;
676                 default:
677                         return -ENXIO;
678         }
679         if (filp->f_op && filp->f_op->open)
680                 return filp->f_op->open(inode,filp);
681         return 0;
682 }
683
684 static struct file_operations memory_fops = {
685         .open           = memory_open,  /* just a selector for the real open */
686 };
687
688 static const struct {
689         unsigned int            minor;
690         char                    *name;
691         umode_t                 mode;
692         struct file_operations  *fops;
693 } devlist[] = { /* list of minor devices */
694         {1, "mem",     S_IRUSR | S_IWUSR | S_IRGRP, &mem_fops},
695         {2, "kmem",    S_IRUSR | S_IWUSR | S_IRGRP, &kmem_fops},
696         {3, "null",    S_IRUGO | S_IWUGO,           &null_fops},
697 #if defined(CONFIG_ISA) || !defined(__mc68000__)
698         {4, "port",    S_IRUSR | S_IWUSR | S_IRGRP, &port_fops},
699 #endif
700         {5, "zero",    S_IRUGO | S_IWUGO,           &zero_fops},
701         {7, "full",    S_IRUGO | S_IWUGO,           &full_fops},
702         {8, "random",  S_IRUGO | S_IWUSR,           &random_fops},
703         {9, "urandom", S_IRUGO | S_IWUSR,           &urandom_fops},
704         {11,"kmsg",    S_IRUGO | S_IWUSR,           &kmsg_fops},
705 };
706
707 static struct class_simple *mem_class;
708
709 static int __init chr_dev_init(void)
710 {
711         int i;
712
713         if (register_chrdev(MEM_MAJOR,"mem",&memory_fops))
714                 printk("unable to get major %d for memory devs\n", MEM_MAJOR);
715
716         mem_class = class_simple_create(THIS_MODULE, "mem");
717         for (i = 0; i < ARRAY_SIZE(devlist); i++) {
718                 class_simple_device_add(mem_class,
719                                         MKDEV(MEM_MAJOR, devlist[i].minor),
720                                         NULL, devlist[i].name);
721                 devfs_mk_cdev(MKDEV(MEM_MAJOR, devlist[i].minor),
722                                 S_IFCHR | devlist[i].mode, devlist[i].name);
723         }
724         
725         return 0;
726 }
727
728 fs_initcall(chr_dev_init);