drivers/char: remove /dev/kmem for good
[linux-2.6-microblaze.git] / drivers / char / mem.c
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  *  linux/drivers/char/mem.c
4  *
5  *  Copyright (C) 1991, 1992  Linus Torvalds
6  *
7  *  Added devfs support.
8  *    Jan-11-1998, C. Scott Ananian <cananian@alumni.princeton.edu>
9  *  Shared /dev/zero mmapping support, Feb 2000, Kanoj Sarcar <kanoj@sgi.com>
10  */
11
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/ptrace.h>
23 #include <linux/device.h>
24 #include <linux/highmem.h>
25 #include <linux/backing-dev.h>
26 #include <linux/shmem_fs.h>
27 #include <linux/splice.h>
28 #include <linux/pfn.h>
29 #include <linux/export.h>
30 #include <linux/io.h>
31 #include <linux/uio.h>
32 #include <linux/uaccess.h>
33 #include <linux/security.h>
34
35 #ifdef CONFIG_IA64
36 # include <linux/efi.h>
37 #endif
38
39 #define DEVMEM_MINOR    1
40 #define DEVPORT_MINOR   4
41
42 static inline unsigned long size_inside_page(unsigned long start,
43                                              unsigned long size)
44 {
45         unsigned long sz;
46
47         sz = PAGE_SIZE - (start & (PAGE_SIZE - 1));
48
49         return min(sz, size);
50 }
51
52 #ifndef ARCH_HAS_VALID_PHYS_ADDR_RANGE
53 static inline int valid_phys_addr_range(phys_addr_t addr, size_t count)
54 {
55         return addr + count <= __pa(high_memory);
56 }
57
58 static inline int valid_mmap_phys_addr_range(unsigned long pfn, size_t size)
59 {
60         return 1;
61 }
62 #endif
63
64 #ifdef CONFIG_STRICT_DEVMEM
65 static inline int page_is_allowed(unsigned long pfn)
66 {
67         return devmem_is_allowed(pfn);
68 }
69 static inline int range_is_allowed(unsigned long pfn, unsigned long size)
70 {
71         u64 from = ((u64)pfn) << PAGE_SHIFT;
72         u64 to = from + size;
73         u64 cursor = from;
74
75         while (cursor < to) {
76                 if (!devmem_is_allowed(pfn))
77                         return 0;
78                 cursor += PAGE_SIZE;
79                 pfn++;
80         }
81         return 1;
82 }
83 #else
84 static inline int page_is_allowed(unsigned long pfn)
85 {
86         return 1;
87 }
88 static inline int range_is_allowed(unsigned long pfn, unsigned long size)
89 {
90         return 1;
91 }
92 #endif
93
94 #ifndef unxlate_dev_mem_ptr
95 #define unxlate_dev_mem_ptr unxlate_dev_mem_ptr
96 void __weak unxlate_dev_mem_ptr(phys_addr_t phys, void *addr)
97 {
98 }
99 #endif
100
101 static inline bool should_stop_iteration(void)
102 {
103         if (need_resched())
104                 cond_resched();
105         return fatal_signal_pending(current);
106 }
107
108 /*
109  * This funcion reads the *physical* memory. The f_pos points directly to the
110  * memory location.
111  */
112 static ssize_t read_mem(struct file *file, char __user *buf,
113                         size_t count, loff_t *ppos)
114 {
115         phys_addr_t p = *ppos;
116         ssize_t read, sz;
117         void *ptr;
118         char *bounce;
119         int err;
120
121         if (p != *ppos)
122                 return 0;
123
124         if (!valid_phys_addr_range(p, count))
125                 return -EFAULT;
126         read = 0;
127 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
128         /* we don't have page 0 mapped on sparc and m68k.. */
129         if (p < PAGE_SIZE) {
130                 sz = size_inside_page(p, count);
131                 if (sz > 0) {
132                         if (clear_user(buf, sz))
133                                 return -EFAULT;
134                         buf += sz;
135                         p += sz;
136                         count -= sz;
137                         read += sz;
138                 }
139         }
140 #endif
141
142         bounce = kmalloc(PAGE_SIZE, GFP_KERNEL);
143         if (!bounce)
144                 return -ENOMEM;
145
146         while (count > 0) {
147                 unsigned long remaining;
148                 int allowed, probe;
149
150                 sz = size_inside_page(p, count);
151
152                 err = -EPERM;
153                 allowed = page_is_allowed(p >> PAGE_SHIFT);
154                 if (!allowed)
155                         goto failed;
156
157                 err = -EFAULT;
158                 if (allowed == 2) {
159                         /* Show zeros for restricted memory. */
160                         remaining = clear_user(buf, sz);
161                 } else {
162                         /*
163                          * On ia64 if a page has been mapped somewhere as
164                          * uncached, then it must also be accessed uncached
165                          * by the kernel or data corruption may occur.
166                          */
167                         ptr = xlate_dev_mem_ptr(p);
168                         if (!ptr)
169                                 goto failed;
170
171                         probe = copy_from_kernel_nofault(bounce, ptr, sz);
172                         unxlate_dev_mem_ptr(p, ptr);
173                         if (probe)
174                                 goto failed;
175
176                         remaining = copy_to_user(buf, bounce, sz);
177                 }
178
179                 if (remaining)
180                         goto failed;
181
182                 buf += sz;
183                 p += sz;
184                 count -= sz;
185                 read += sz;
186                 if (should_stop_iteration())
187                         break;
188         }
189         kfree(bounce);
190
191         *ppos += read;
192         return read;
193
194 failed:
195         kfree(bounce);
196         return err;
197 }
198
199 static ssize_t write_mem(struct file *file, const char __user *buf,
200                          size_t count, loff_t *ppos)
201 {
202         phys_addr_t p = *ppos;
203         ssize_t written, sz;
204         unsigned long copied;
205         void *ptr;
206
207         if (p != *ppos)
208                 return -EFBIG;
209
210         if (!valid_phys_addr_range(p, count))
211                 return -EFAULT;
212
213         written = 0;
214
215 #ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
216         /* we don't have page 0 mapped on sparc and m68k.. */
217         if (p < PAGE_SIZE) {
218                 sz = size_inside_page(p, count);
219                 /* Hmm. Do something? */
220                 buf += sz;
221                 p += sz;
222                 count -= sz;
223                 written += sz;
224         }
225 #endif
226
227         while (count > 0) {
228                 int allowed;
229
230                 sz = size_inside_page(p, count);
231
232                 allowed = page_is_allowed(p >> PAGE_SHIFT);
233                 if (!allowed)
234                         return -EPERM;
235
236                 /* Skip actual writing when a page is marked as restricted. */
237                 if (allowed == 1) {
238                         /*
239                          * On ia64 if a page has been mapped somewhere as
240                          * uncached, then it must also be accessed uncached
241                          * by the kernel or data corruption may occur.
242                          */
243                         ptr = xlate_dev_mem_ptr(p);
244                         if (!ptr) {
245                                 if (written)
246                                         break;
247                                 return -EFAULT;
248                         }
249
250                         copied = copy_from_user(ptr, buf, sz);
251                         unxlate_dev_mem_ptr(p, ptr);
252                         if (copied) {
253                                 written += sz - copied;
254                                 if (written)
255                                         break;
256                                 return -EFAULT;
257                         }
258                 }
259
260                 buf += sz;
261                 p += sz;
262                 count -= sz;
263                 written += sz;
264                 if (should_stop_iteration())
265                         break;
266         }
267
268         *ppos += written;
269         return written;
270 }
271
272 int __weak phys_mem_access_prot_allowed(struct file *file,
273         unsigned long pfn, unsigned long size, pgprot_t *vma_prot)
274 {
275         return 1;
276 }
277
278 #ifndef __HAVE_PHYS_MEM_ACCESS_PROT
279
280 /*
281  * Architectures vary in how they handle caching for addresses
282  * outside of main memory.
283  *
284  */
285 #ifdef pgprot_noncached
286 static int uncached_access(struct file *file, phys_addr_t addr)
287 {
288 #if defined(CONFIG_IA64)
289         /*
290          * On ia64, we ignore O_DSYNC because we cannot tolerate memory
291          * attribute aliases.
292          */
293         return !(efi_mem_attributes(addr) & EFI_MEMORY_WB);
294 #else
295         /*
296          * Accessing memory above the top the kernel knows about or through a
297          * file pointer
298          * that was marked O_DSYNC will be done non-cached.
299          */
300         if (file->f_flags & O_DSYNC)
301                 return 1;
302         return addr >= __pa(high_memory);
303 #endif
304 }
305 #endif
306
307 static pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
308                                      unsigned long size, pgprot_t vma_prot)
309 {
310 #ifdef pgprot_noncached
311         phys_addr_t offset = pfn << PAGE_SHIFT;
312
313         if (uncached_access(file, offset))
314                 return pgprot_noncached(vma_prot);
315 #endif
316         return vma_prot;
317 }
318 #endif
319
320 #ifndef CONFIG_MMU
321 static unsigned long get_unmapped_area_mem(struct file *file,
322                                            unsigned long addr,
323                                            unsigned long len,
324                                            unsigned long pgoff,
325                                            unsigned long flags)
326 {
327         if (!valid_mmap_phys_addr_range(pgoff, len))
328                 return (unsigned long) -EINVAL;
329         return pgoff << PAGE_SHIFT;
330 }
331
332 /* permit direct mmap, for read, write or exec */
333 static unsigned memory_mmap_capabilities(struct file *file)
334 {
335         return NOMMU_MAP_DIRECT |
336                 NOMMU_MAP_READ | NOMMU_MAP_WRITE | NOMMU_MAP_EXEC;
337 }
338
339 static unsigned zero_mmap_capabilities(struct file *file)
340 {
341         return NOMMU_MAP_COPY;
342 }
343
344 /* can't do an in-place private mapping if there's no MMU */
345 static inline int private_mapping_ok(struct vm_area_struct *vma)
346 {
347         return vma->vm_flags & VM_MAYSHARE;
348 }
349 #else
350
351 static inline int private_mapping_ok(struct vm_area_struct *vma)
352 {
353         return 1;
354 }
355 #endif
356
357 static const struct vm_operations_struct mmap_mem_ops = {
358 #ifdef CONFIG_HAVE_IOREMAP_PROT
359         .access = generic_access_phys
360 #endif
361 };
362
363 static int mmap_mem(struct file *file, struct vm_area_struct *vma)
364 {
365         size_t size = vma->vm_end - vma->vm_start;
366         phys_addr_t offset = (phys_addr_t)vma->vm_pgoff << PAGE_SHIFT;
367
368         /* Does it even fit in phys_addr_t? */
369         if (offset >> PAGE_SHIFT != vma->vm_pgoff)
370                 return -EINVAL;
371
372         /* It's illegal to wrap around the end of the physical address space. */
373         if (offset + (phys_addr_t)size - 1 < offset)
374                 return -EINVAL;
375
376         if (!valid_mmap_phys_addr_range(vma->vm_pgoff, size))
377                 return -EINVAL;
378
379         if (!private_mapping_ok(vma))
380                 return -ENOSYS;
381
382         if (!range_is_allowed(vma->vm_pgoff, size))
383                 return -EPERM;
384
385         if (!phys_mem_access_prot_allowed(file, vma->vm_pgoff, size,
386                                                 &vma->vm_page_prot))
387                 return -EINVAL;
388
389         vma->vm_page_prot = phys_mem_access_prot(file, vma->vm_pgoff,
390                                                  size,
391                                                  vma->vm_page_prot);
392
393         vma->vm_ops = &mmap_mem_ops;
394
395         /* Remap-pfn-range will mark the range VM_IO */
396         if (remap_pfn_range(vma,
397                             vma->vm_start,
398                             vma->vm_pgoff,
399                             size,
400                             vma->vm_page_prot)) {
401                 return -EAGAIN;
402         }
403         return 0;
404 }
405
406 static ssize_t read_port(struct file *file, char __user *buf,
407                          size_t count, loff_t *ppos)
408 {
409         unsigned long i = *ppos;
410         char __user *tmp = buf;
411
412         if (!access_ok(buf, count))
413                 return -EFAULT;
414         while (count-- > 0 && i < 65536) {
415                 if (__put_user(inb(i), tmp) < 0)
416                         return -EFAULT;
417                 i++;
418                 tmp++;
419         }
420         *ppos = i;
421         return tmp-buf;
422 }
423
424 static ssize_t write_port(struct file *file, const char __user *buf,
425                           size_t count, loff_t *ppos)
426 {
427         unsigned long i = *ppos;
428         const char __user *tmp = buf;
429
430         if (!access_ok(buf, count))
431                 return -EFAULT;
432         while (count-- > 0 && i < 65536) {
433                 char c;
434
435                 if (__get_user(c, tmp)) {
436                         if (tmp > buf)
437                                 break;
438                         return -EFAULT;
439                 }
440                 outb(c, i);
441                 i++;
442                 tmp++;
443         }
444         *ppos = i;
445         return tmp-buf;
446 }
447
448 static ssize_t read_null(struct file *file, char __user *buf,
449                          size_t count, loff_t *ppos)
450 {
451         return 0;
452 }
453
454 static ssize_t write_null(struct file *file, const char __user *buf,
455                           size_t count, loff_t *ppos)
456 {
457         return count;
458 }
459
460 static ssize_t read_iter_null(struct kiocb *iocb, struct iov_iter *to)
461 {
462         return 0;
463 }
464
465 static ssize_t write_iter_null(struct kiocb *iocb, struct iov_iter *from)
466 {
467         size_t count = iov_iter_count(from);
468         iov_iter_advance(from, count);
469         return count;
470 }
471
472 static int pipe_to_null(struct pipe_inode_info *info, struct pipe_buffer *buf,
473                         struct splice_desc *sd)
474 {
475         return sd->len;
476 }
477
478 static ssize_t splice_write_null(struct pipe_inode_info *pipe, struct file *out,
479                                  loff_t *ppos, size_t len, unsigned int flags)
480 {
481         return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_null);
482 }
483
484 static ssize_t read_iter_zero(struct kiocb *iocb, struct iov_iter *iter)
485 {
486         size_t written = 0;
487
488         while (iov_iter_count(iter)) {
489                 size_t chunk = iov_iter_count(iter), n;
490
491                 if (chunk > PAGE_SIZE)
492                         chunk = PAGE_SIZE;      /* Just for latency reasons */
493                 n = iov_iter_zero(chunk, iter);
494                 if (!n && iov_iter_count(iter))
495                         return written ? written : -EFAULT;
496                 written += n;
497                 if (signal_pending(current))
498                         return written ? written : -ERESTARTSYS;
499                 cond_resched();
500         }
501         return written;
502 }
503
504 static ssize_t read_zero(struct file *file, char __user *buf,
505                          size_t count, loff_t *ppos)
506 {
507         size_t cleared = 0;
508
509         while (count) {
510                 size_t chunk = min_t(size_t, count, PAGE_SIZE);
511                 size_t left;
512
513                 left = clear_user(buf + cleared, chunk);
514                 if (unlikely(left)) {
515                         cleared += (chunk - left);
516                         if (!cleared)
517                                 return -EFAULT;
518                         break;
519                 }
520                 cleared += chunk;
521                 count -= chunk;
522
523                 if (signal_pending(current))
524                         break;
525                 cond_resched();
526         }
527
528         return cleared;
529 }
530
531 static int mmap_zero(struct file *file, struct vm_area_struct *vma)
532 {
533 #ifndef CONFIG_MMU
534         return -ENOSYS;
535 #endif
536         if (vma->vm_flags & VM_SHARED)
537                 return shmem_zero_setup(vma);
538         vma_set_anonymous(vma);
539         return 0;
540 }
541
542 static unsigned long get_unmapped_area_zero(struct file *file,
543                                 unsigned long addr, unsigned long len,
544                                 unsigned long pgoff, unsigned long flags)
545 {
546 #ifdef CONFIG_MMU
547         if (flags & MAP_SHARED) {
548                 /*
549                  * mmap_zero() will call shmem_zero_setup() to create a file,
550                  * so use shmem's get_unmapped_area in case it can be huge;
551                  * and pass NULL for file as in mmap.c's get_unmapped_area(),
552                  * so as not to confuse shmem with our handle on "/dev/zero".
553                  */
554                 return shmem_get_unmapped_area(NULL, addr, len, pgoff, flags);
555         }
556
557         /* Otherwise flags & MAP_PRIVATE: with no shmem object beneath it */
558         return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
559 #else
560         return -ENOSYS;
561 #endif
562 }
563
564 static ssize_t write_full(struct file *file, const char __user *buf,
565                           size_t count, loff_t *ppos)
566 {
567         return -ENOSPC;
568 }
569
570 /*
571  * Special lseek() function for /dev/null and /dev/zero.  Most notably, you
572  * can fopen() both devices with "a" now.  This was previously impossible.
573  * -- SRB.
574  */
575 static loff_t null_lseek(struct file *file, loff_t offset, int orig)
576 {
577         return file->f_pos = 0;
578 }
579
580 /*
581  * The memory devices use the full 32/64 bits of the offset, and so we cannot
582  * check against negative addresses: they are ok. The return value is weird,
583  * though, in that case (0).
584  *
585  * also note that seeking relative to the "end of file" isn't supported:
586  * it has no meaning, so it returns -EINVAL.
587  */
588 static loff_t memory_lseek(struct file *file, loff_t offset, int orig)
589 {
590         loff_t ret;
591
592         inode_lock(file_inode(file));
593         switch (orig) {
594         case SEEK_CUR:
595                 offset += file->f_pos;
596                 fallthrough;
597         case SEEK_SET:
598                 /* to avoid userland mistaking f_pos=-9 as -EBADF=-9 */
599                 if ((unsigned long long)offset >= -MAX_ERRNO) {
600                         ret = -EOVERFLOW;
601                         break;
602                 }
603                 file->f_pos = offset;
604                 ret = file->f_pos;
605                 force_successful_syscall_return();
606                 break;
607         default:
608                 ret = -EINVAL;
609         }
610         inode_unlock(file_inode(file));
611         return ret;
612 }
613
614 static int open_port(struct inode *inode, struct file *filp)
615 {
616         int rc;
617
618         if (!capable(CAP_SYS_RAWIO))
619                 return -EPERM;
620
621         rc = security_locked_down(LOCKDOWN_DEV_MEM);
622         if (rc)
623                 return rc;
624
625         if (iminor(inode) != DEVMEM_MINOR)
626                 return 0;
627
628         /*
629          * Use a unified address space to have a single point to manage
630          * revocations when drivers want to take over a /dev/mem mapped
631          * range.
632          */
633         filp->f_mapping = iomem_get_mapping();
634
635         return 0;
636 }
637
638 #define zero_lseek      null_lseek
639 #define full_lseek      null_lseek
640 #define write_zero      write_null
641 #define write_iter_zero write_iter_null
642 #define open_mem        open_port
643
644 static const struct file_operations __maybe_unused mem_fops = {
645         .llseek         = memory_lseek,
646         .read           = read_mem,
647         .write          = write_mem,
648         .mmap           = mmap_mem,
649         .open           = open_mem,
650 #ifndef CONFIG_MMU
651         .get_unmapped_area = get_unmapped_area_mem,
652         .mmap_capabilities = memory_mmap_capabilities,
653 #endif
654 };
655
656 static const struct file_operations null_fops = {
657         .llseek         = null_lseek,
658         .read           = read_null,
659         .write          = write_null,
660         .read_iter      = read_iter_null,
661         .write_iter     = write_iter_null,
662         .splice_write   = splice_write_null,
663 };
664
665 static const struct file_operations __maybe_unused port_fops = {
666         .llseek         = memory_lseek,
667         .read           = read_port,
668         .write          = write_port,
669         .open           = open_port,
670 };
671
672 static const struct file_operations zero_fops = {
673         .llseek         = zero_lseek,
674         .write          = write_zero,
675         .read_iter      = read_iter_zero,
676         .read           = read_zero,
677         .write_iter     = write_iter_zero,
678         .mmap           = mmap_zero,
679         .get_unmapped_area = get_unmapped_area_zero,
680 #ifndef CONFIG_MMU
681         .mmap_capabilities = zero_mmap_capabilities,
682 #endif
683 };
684
685 static const struct file_operations full_fops = {
686         .llseek         = full_lseek,
687         .read_iter      = read_iter_zero,
688         .write          = write_full,
689 };
690
691 static const struct memdev {
692         const char *name;
693         umode_t mode;
694         const struct file_operations *fops;
695         fmode_t fmode;
696 } devlist[] = {
697 #ifdef CONFIG_DEVMEM
698          [DEVMEM_MINOR] = { "mem", 0, &mem_fops, FMODE_UNSIGNED_OFFSET },
699 #endif
700          [3] = { "null", 0666, &null_fops, 0 },
701 #ifdef CONFIG_DEVPORT
702          [4] = { "port", 0, &port_fops, 0 },
703 #endif
704          [5] = { "zero", 0666, &zero_fops, 0 },
705          [7] = { "full", 0666, &full_fops, 0 },
706          [8] = { "random", 0666, &random_fops, 0 },
707          [9] = { "urandom", 0666, &urandom_fops, 0 },
708 #ifdef CONFIG_PRINTK
709         [11] = { "kmsg", 0644, &kmsg_fops, 0 },
710 #endif
711 };
712
713 static int memory_open(struct inode *inode, struct file *filp)
714 {
715         int minor;
716         const struct memdev *dev;
717
718         minor = iminor(inode);
719         if (minor >= ARRAY_SIZE(devlist))
720                 return -ENXIO;
721
722         dev = &devlist[minor];
723         if (!dev->fops)
724                 return -ENXIO;
725
726         filp->f_op = dev->fops;
727         filp->f_mode |= dev->fmode;
728
729         if (dev->fops->open)
730                 return dev->fops->open(inode, filp);
731
732         return 0;
733 }
734
735 static const struct file_operations memory_fops = {
736         .open = memory_open,
737         .llseek = noop_llseek,
738 };
739
740 static char *mem_devnode(struct device *dev, umode_t *mode)
741 {
742         if (mode && devlist[MINOR(dev->devt)].mode)
743                 *mode = devlist[MINOR(dev->devt)].mode;
744         return NULL;
745 }
746
747 static struct class *mem_class;
748
749 static int __init chr_dev_init(void)
750 {
751         int minor;
752
753         if (register_chrdev(MEM_MAJOR, "mem", &memory_fops))
754                 printk("unable to get major %d for memory devs\n", MEM_MAJOR);
755
756         mem_class = class_create(THIS_MODULE, "mem");
757         if (IS_ERR(mem_class))
758                 return PTR_ERR(mem_class);
759
760         mem_class->devnode = mem_devnode;
761         for (minor = 1; minor < ARRAY_SIZE(devlist); minor++) {
762                 if (!devlist[minor].name)
763                         continue;
764
765                 /*
766                  * Create /dev/port?
767                  */
768                 if ((minor == DEVPORT_MINOR) && !arch_has_dev_port())
769                         continue;
770
771                 device_create(mem_class, NULL, MKDEV(MEM_MAJOR, minor),
772                               NULL, devlist[minor].name);
773         }
774
775         return tty_init();
776 }
777
778 fs_initcall(chr_dev_init);