Linux Audio

Check our new training course

Loading...
v5.9
   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#include <linux/pseudo_fs.h>
  35#include <uapi/linux/magic.h>
  36#include <linux/mount.h>
  37
  38#ifdef CONFIG_IA64
  39# include <linux/efi.h>
  40#endif
  41
  42#define DEVMEM_MINOR	1
  43#define DEVPORT_MINOR	4
  44
  45static inline unsigned long size_inside_page(unsigned long start,
  46					     unsigned long size)
  47{
  48	unsigned long sz;
  49
  50	sz = PAGE_SIZE - (start & (PAGE_SIZE - 1));
  51
  52	return min(sz, size);
  53}
  54
  55#ifndef ARCH_HAS_VALID_PHYS_ADDR_RANGE
  56static inline int valid_phys_addr_range(phys_addr_t addr, size_t count)
  57{
  58	return addr + count <= __pa(high_memory);
  59}
  60
  61static inline int valid_mmap_phys_addr_range(unsigned long pfn, size_t size)
  62{
  63	return 1;
  64}
  65#endif
  66
  67#ifdef CONFIG_STRICT_DEVMEM
  68static inline int page_is_allowed(unsigned long pfn)
  69{
  70	return devmem_is_allowed(pfn);
  71}
  72static inline int range_is_allowed(unsigned long pfn, unsigned long size)
  73{
  74	u64 from = ((u64)pfn) << PAGE_SHIFT;
  75	u64 to = from + size;
  76	u64 cursor = from;
  77
  78	while (cursor < to) {
  79		if (!devmem_is_allowed(pfn))
  80			return 0;
  81		cursor += PAGE_SIZE;
  82		pfn++;
  83	}
  84	return 1;
  85}
  86#else
  87static inline int page_is_allowed(unsigned long pfn)
  88{
  89	return 1;
  90}
  91static inline int range_is_allowed(unsigned long pfn, unsigned long size)
  92{
  93	return 1;
  94}
  95#endif
  96
  97#ifndef unxlate_dev_mem_ptr
  98#define unxlate_dev_mem_ptr unxlate_dev_mem_ptr
  99void __weak unxlate_dev_mem_ptr(phys_addr_t phys, void *addr)
 100{
 101}
 102#endif
 103
 104static inline bool should_stop_iteration(void)
 105{
 106	if (need_resched())
 107		cond_resched();
 108	return fatal_signal_pending(current);
 109}
 110
 111/*
 112 * This funcion reads the *physical* memory. The f_pos points directly to the
 113 * memory location.
 114 */
 115static ssize_t read_mem(struct file *file, char __user *buf,
 116			size_t count, loff_t *ppos)
 117{
 118	phys_addr_t p = *ppos;
 119	ssize_t read, sz;
 120	void *ptr;
 121	char *bounce;
 122	int err;
 123
 124	if (p != *ppos)
 125		return 0;
 126
 127	if (!valid_phys_addr_range(p, count))
 128		return -EFAULT;
 129	read = 0;
 130#ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 131	/* we don't have page 0 mapped on sparc and m68k.. */
 132	if (p < PAGE_SIZE) {
 133		sz = size_inside_page(p, count);
 134		if (sz > 0) {
 135			if (clear_user(buf, sz))
 136				return -EFAULT;
 137			buf += sz;
 138			p += sz;
 139			count -= sz;
 140			read += sz;
 141		}
 142	}
 143#endif
 144
 145	bounce = kmalloc(PAGE_SIZE, GFP_KERNEL);
 146	if (!bounce)
 147		return -ENOMEM;
 148
 149	while (count > 0) {
 150		unsigned long remaining;
 151		int allowed, probe;
 152
 153		sz = size_inside_page(p, count);
 154
 155		err = -EPERM;
 156		allowed = page_is_allowed(p >> PAGE_SHIFT);
 157		if (!allowed)
 158			goto failed;
 159
 160		err = -EFAULT;
 161		if (allowed == 2) {
 162			/* Show zeros for restricted memory. */
 163			remaining = clear_user(buf, sz);
 164		} else {
 165			/*
 166			 * On ia64 if a page has been mapped somewhere as
 167			 * uncached, then it must also be accessed uncached
 168			 * by the kernel or data corruption may occur.
 169			 */
 170			ptr = xlate_dev_mem_ptr(p);
 171			if (!ptr)
 172				goto failed;
 173
 174			probe = copy_from_kernel_nofault(bounce, ptr, sz);
 175			unxlate_dev_mem_ptr(p, ptr);
 176			if (probe)
 177				goto failed;
 178
 179			remaining = copy_to_user(buf, bounce, sz);
 180		}
 181
 182		if (remaining)
 183			goto failed;
 184
 185		buf += sz;
 186		p += sz;
 187		count -= sz;
 188		read += sz;
 189		if (should_stop_iteration())
 190			break;
 191	}
 192	kfree(bounce);
 193
 194	*ppos += read;
 195	return read;
 196
 197failed:
 198	kfree(bounce);
 199	return err;
 200}
 201
 202static ssize_t write_mem(struct file *file, const char __user *buf,
 203			 size_t count, loff_t *ppos)
 204{
 205	phys_addr_t p = *ppos;
 206	ssize_t written, sz;
 207	unsigned long copied;
 208	void *ptr;
 209
 210	if (p != *ppos)
 211		return -EFBIG;
 212
 213	if (!valid_phys_addr_range(p, count))
 214		return -EFAULT;
 215
 216	written = 0;
 217
 218#ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 219	/* we don't have page 0 mapped on sparc and m68k.. */
 220	if (p < PAGE_SIZE) {
 221		sz = size_inside_page(p, count);
 222		/* Hmm. Do something? */
 223		buf += sz;
 224		p += sz;
 225		count -= sz;
 226		written += sz;
 227	}
 228#endif
 229
 230	while (count > 0) {
 231		int allowed;
 232
 233		sz = size_inside_page(p, count);
 234
 235		allowed = page_is_allowed(p >> PAGE_SHIFT);
 236		if (!allowed)
 237			return -EPERM;
 238
 239		/* Skip actual writing when a page is marked as restricted. */
 240		if (allowed == 1) {
 241			/*
 242			 * On ia64 if a page has been mapped somewhere as
 243			 * uncached, then it must also be accessed uncached
 244			 * by the kernel or data corruption may occur.
 245			 */
 246			ptr = xlate_dev_mem_ptr(p);
 247			if (!ptr) {
 248				if (written)
 249					break;
 250				return -EFAULT;
 251			}
 252
 253			copied = copy_from_user(ptr, buf, sz);
 254			unxlate_dev_mem_ptr(p, ptr);
 255			if (copied) {
 256				written += sz - copied;
 257				if (written)
 258					break;
 259				return -EFAULT;
 260			}
 261		}
 262
 263		buf += sz;
 264		p += sz;
 265		count -= sz;
 266		written += sz;
 267		if (should_stop_iteration())
 268			break;
 269	}
 270
 271	*ppos += written;
 272	return written;
 273}
 274
 275int __weak phys_mem_access_prot_allowed(struct file *file,
 276	unsigned long pfn, unsigned long size, pgprot_t *vma_prot)
 277{
 278	return 1;
 279}
 280
 281#ifndef __HAVE_PHYS_MEM_ACCESS_PROT
 282
 283/*
 284 * Architectures vary in how they handle caching for addresses
 285 * outside of main memory.
 286 *
 287 */
 288#ifdef pgprot_noncached
 289static int uncached_access(struct file *file, phys_addr_t addr)
 290{
 291#if defined(CONFIG_IA64)
 292	/*
 293	 * On ia64, we ignore O_DSYNC because we cannot tolerate memory
 294	 * attribute aliases.
 295	 */
 296	return !(efi_mem_attributes(addr) & EFI_MEMORY_WB);
 297#elif defined(CONFIG_MIPS)
 298	{
 299		extern int __uncached_access(struct file *file,
 300					     unsigned long addr);
 301
 302		return __uncached_access(file, addr);
 303	}
 304#else
 305	/*
 306	 * Accessing memory above the top the kernel knows about or through a
 307	 * file pointer
 308	 * that was marked O_DSYNC will be done non-cached.
 309	 */
 310	if (file->f_flags & O_DSYNC)
 311		return 1;
 312	return addr >= __pa(high_memory);
 313#endif
 314}
 315#endif
 316
 317static pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
 318				     unsigned long size, pgprot_t vma_prot)
 319{
 320#ifdef pgprot_noncached
 321	phys_addr_t offset = pfn << PAGE_SHIFT;
 322
 323	if (uncached_access(file, offset))
 324		return pgprot_noncached(vma_prot);
 325#endif
 326	return vma_prot;
 327}
 328#endif
 329
 330#ifndef CONFIG_MMU
 331static unsigned long get_unmapped_area_mem(struct file *file,
 332					   unsigned long addr,
 333					   unsigned long len,
 334					   unsigned long pgoff,
 335					   unsigned long flags)
 336{
 337	if (!valid_mmap_phys_addr_range(pgoff, len))
 338		return (unsigned long) -EINVAL;
 339	return pgoff << PAGE_SHIFT;
 340}
 341
 342/* permit direct mmap, for read, write or exec */
 343static unsigned memory_mmap_capabilities(struct file *file)
 344{
 345	return NOMMU_MAP_DIRECT |
 346		NOMMU_MAP_READ | NOMMU_MAP_WRITE | NOMMU_MAP_EXEC;
 347}
 348
 349static unsigned zero_mmap_capabilities(struct file *file)
 350{
 351	return NOMMU_MAP_COPY;
 352}
 353
 354/* can't do an in-place private mapping if there's no MMU */
 355static inline int private_mapping_ok(struct vm_area_struct *vma)
 356{
 357	return vma->vm_flags & VM_MAYSHARE;
 358}
 359#else
 360
 361static inline int private_mapping_ok(struct vm_area_struct *vma)
 362{
 363	return 1;
 364}
 365#endif
 366
 367static const struct vm_operations_struct mmap_mem_ops = {
 368#ifdef CONFIG_HAVE_IOREMAP_PROT
 369	.access = generic_access_phys
 370#endif
 371};
 372
 373static int mmap_mem(struct file *file, struct vm_area_struct *vma)
 374{
 375	size_t size = vma->vm_end - vma->vm_start;
 376	phys_addr_t offset = (phys_addr_t)vma->vm_pgoff << PAGE_SHIFT;
 377
 378	/* Does it even fit in phys_addr_t? */
 379	if (offset >> PAGE_SHIFT != vma->vm_pgoff)
 380		return -EINVAL;
 381
 382	/* It's illegal to wrap around the end of the physical address space. */
 383	if (offset + (phys_addr_t)size - 1 < offset)
 384		return -EINVAL;
 385
 386	if (!valid_mmap_phys_addr_range(vma->vm_pgoff, size))
 387		return -EINVAL;
 388
 389	if (!private_mapping_ok(vma))
 390		return -ENOSYS;
 391
 392	if (!range_is_allowed(vma->vm_pgoff, size))
 393		return -EPERM;
 394
 395	if (!phys_mem_access_prot_allowed(file, vma->vm_pgoff, size,
 396						&vma->vm_page_prot))
 397		return -EINVAL;
 398
 399	vma->vm_page_prot = phys_mem_access_prot(file, vma->vm_pgoff,
 400						 size,
 401						 vma->vm_page_prot);
 402
 403	vma->vm_ops = &mmap_mem_ops;
 404
 405	/* Remap-pfn-range will mark the range VM_IO */
 406	if (remap_pfn_range(vma,
 407			    vma->vm_start,
 408			    vma->vm_pgoff,
 409			    size,
 410			    vma->vm_page_prot)) {
 411		return -EAGAIN;
 412	}
 413	return 0;
 414}
 415
 416static int mmap_kmem(struct file *file, struct vm_area_struct *vma)
 417{
 418	unsigned long pfn;
 419
 420	/* Turn a kernel-virtual address into a physical page frame */
 421	pfn = __pa((u64)vma->vm_pgoff << PAGE_SHIFT) >> PAGE_SHIFT;
 422
 423	/*
 424	 * RED-PEN: on some architectures there is more mapped memory than
 425	 * available in mem_map which pfn_valid checks for. Perhaps should add a
 426	 * new macro here.
 427	 *
 428	 * RED-PEN: vmalloc is not supported right now.
 429	 */
 430	if (!pfn_valid(pfn))
 431		return -EIO;
 432
 433	vma->vm_pgoff = pfn;
 434	return mmap_mem(file, vma);
 435}
 436
 437/*
 438 * This function reads the *virtual* memory as seen by the kernel.
 439 */
 440static ssize_t read_kmem(struct file *file, char __user *buf,
 441			 size_t count, loff_t *ppos)
 442{
 443	unsigned long p = *ppos;
 444	ssize_t low_count, read, sz;
 445	char *kbuf; /* k-addr because vread() takes vmlist_lock rwlock */
 446	int err = 0;
 447
 448	read = 0;
 449	if (p < (unsigned long) high_memory) {
 450		low_count = count;
 451		if (count > (unsigned long)high_memory - p)
 452			low_count = (unsigned long)high_memory - p;
 453
 454#ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 455		/* we don't have page 0 mapped on sparc and m68k.. */
 456		if (p < PAGE_SIZE && low_count > 0) {
 457			sz = size_inside_page(p, low_count);
 458			if (clear_user(buf, sz))
 459				return -EFAULT;
 460			buf += sz;
 461			p += sz;
 462			read += sz;
 463			low_count -= sz;
 464			count -= sz;
 465		}
 466#endif
 467		while (low_count > 0) {
 468			sz = size_inside_page(p, low_count);
 469
 470			/*
 471			 * On ia64 if a page has been mapped somewhere as
 472			 * uncached, then it must also be accessed uncached
 473			 * by the kernel or data corruption may occur
 474			 */
 475			kbuf = xlate_dev_kmem_ptr((void *)p);
 476			if (!virt_addr_valid(kbuf))
 477				return -ENXIO;
 478
 479			if (copy_to_user(buf, kbuf, sz))
 480				return -EFAULT;
 481			buf += sz;
 482			p += sz;
 483			read += sz;
 484			low_count -= sz;
 485			count -= sz;
 486			if (should_stop_iteration()) {
 487				count = 0;
 488				break;
 489			}
 490		}
 491	}
 492
 493	if (count > 0) {
 494		kbuf = (char *)__get_free_page(GFP_KERNEL);
 495		if (!kbuf)
 496			return -ENOMEM;
 497		while (count > 0) {
 498			sz = size_inside_page(p, count);
 499			if (!is_vmalloc_or_module_addr((void *)p)) {
 500				err = -ENXIO;
 501				break;
 502			}
 503			sz = vread(kbuf, (char *)p, sz);
 504			if (!sz)
 505				break;
 506			if (copy_to_user(buf, kbuf, sz)) {
 507				err = -EFAULT;
 508				break;
 509			}
 510			count -= sz;
 511			buf += sz;
 512			read += sz;
 513			p += sz;
 514			if (should_stop_iteration())
 515				break;
 516		}
 517		free_page((unsigned long)kbuf);
 518	}
 519	*ppos = p;
 520	return read ? read : err;
 521}
 522
 523
 524static ssize_t do_write_kmem(unsigned long p, const char __user *buf,
 525				size_t count, loff_t *ppos)
 526{
 527	ssize_t written, sz;
 528	unsigned long copied;
 529
 530	written = 0;
 531#ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
 532	/* we don't have page 0 mapped on sparc and m68k.. */
 533	if (p < PAGE_SIZE) {
 534		sz = size_inside_page(p, count);
 535		/* Hmm. Do something? */
 536		buf += sz;
 537		p += sz;
 538		count -= sz;
 539		written += sz;
 540	}
 541#endif
 542
 543	while (count > 0) {
 544		void *ptr;
 545
 546		sz = size_inside_page(p, count);
 547
 548		/*
 549		 * On ia64 if a page has been mapped somewhere as uncached, then
 550		 * it must also be accessed uncached by the kernel or data
 551		 * corruption may occur.
 552		 */
 553		ptr = xlate_dev_kmem_ptr((void *)p);
 554		if (!virt_addr_valid(ptr))
 555			return -ENXIO;
 556
 557		copied = copy_from_user(ptr, buf, sz);
 558		if (copied) {
 559			written += sz - copied;
 560			if (written)
 561				break;
 562			return -EFAULT;
 563		}
 564		buf += sz;
 565		p += sz;
 566		count -= sz;
 567		written += sz;
 568		if (should_stop_iteration())
 569			break;
 570	}
 571
 572	*ppos += written;
 573	return written;
 574}
 575
 576/*
 577 * This function writes to the *virtual* memory as seen by the kernel.
 578 */
 579static ssize_t write_kmem(struct file *file, const char __user *buf,
 580			  size_t count, loff_t *ppos)
 581{
 582	unsigned long p = *ppos;
 583	ssize_t wrote = 0;
 584	ssize_t virtr = 0;
 585	char *kbuf; /* k-addr because vwrite() takes vmlist_lock rwlock */
 586	int err = 0;
 587
 588	if (p < (unsigned long) high_memory) {
 589		unsigned long to_write = min_t(unsigned long, count,
 590					       (unsigned long)high_memory - p);
 591		wrote = do_write_kmem(p, buf, to_write, ppos);
 592		if (wrote != to_write)
 593			return wrote;
 594		p += wrote;
 595		buf += wrote;
 596		count -= wrote;
 597	}
 598
 599	if (count > 0) {
 600		kbuf = (char *)__get_free_page(GFP_KERNEL);
 601		if (!kbuf)
 602			return wrote ? wrote : -ENOMEM;
 603		while (count > 0) {
 604			unsigned long sz = size_inside_page(p, count);
 605			unsigned long n;
 606
 607			if (!is_vmalloc_or_module_addr((void *)p)) {
 608				err = -ENXIO;
 609				break;
 610			}
 611			n = copy_from_user(kbuf, buf, sz);
 612			if (n) {
 613				err = -EFAULT;
 614				break;
 615			}
 616			vwrite(kbuf, (char *)p, sz);
 617			count -= sz;
 618			buf += sz;
 619			virtr += sz;
 620			p += sz;
 621			if (should_stop_iteration())
 622				break;
 623		}
 624		free_page((unsigned long)kbuf);
 625	}
 626
 627	*ppos = p;
 628	return virtr + wrote ? : err;
 629}
 630
 631static ssize_t read_port(struct file *file, char __user *buf,
 632			 size_t count, loff_t *ppos)
 633{
 634	unsigned long i = *ppos;
 635	char __user *tmp = buf;
 636
 637	if (!access_ok(buf, count))
 638		return -EFAULT;
 639	while (count-- > 0 && i < 65536) {
 640		if (__put_user(inb(i), tmp) < 0)
 641			return -EFAULT;
 642		i++;
 643		tmp++;
 644	}
 645	*ppos = i;
 646	return tmp-buf;
 647}
 648
 649static ssize_t write_port(struct file *file, const char __user *buf,
 650			  size_t count, loff_t *ppos)
 651{
 652	unsigned long i = *ppos;
 653	const char __user *tmp = buf;
 654
 655	if (!access_ok(buf, count))
 656		return -EFAULT;
 657	while (count-- > 0 && i < 65536) {
 658		char c;
 659
 660		if (__get_user(c, tmp)) {
 661			if (tmp > buf)
 662				break;
 663			return -EFAULT;
 664		}
 665		outb(c, i);
 666		i++;
 667		tmp++;
 668	}
 669	*ppos = i;
 670	return tmp-buf;
 671}
 672
 673static ssize_t read_null(struct file *file, char __user *buf,
 674			 size_t count, loff_t *ppos)
 675{
 676	return 0;
 677}
 678
 679static ssize_t write_null(struct file *file, const char __user *buf,
 680			  size_t count, loff_t *ppos)
 681{
 682	return count;
 683}
 684
 685static ssize_t read_iter_null(struct kiocb *iocb, struct iov_iter *to)
 686{
 687	return 0;
 688}
 689
 690static ssize_t write_iter_null(struct kiocb *iocb, struct iov_iter *from)
 691{
 692	size_t count = iov_iter_count(from);
 693	iov_iter_advance(from, count);
 694	return count;
 695}
 696
 697static int pipe_to_null(struct pipe_inode_info *info, struct pipe_buffer *buf,
 698			struct splice_desc *sd)
 699{
 700	return sd->len;
 701}
 702
 703static ssize_t splice_write_null(struct pipe_inode_info *pipe, struct file *out,
 704				 loff_t *ppos, size_t len, unsigned int flags)
 705{
 706	return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_null);
 707}
 708
 
 
 
 
 
 709static ssize_t read_iter_zero(struct kiocb *iocb, struct iov_iter *iter)
 710{
 711	size_t written = 0;
 712
 713	while (iov_iter_count(iter)) {
 714		size_t chunk = iov_iter_count(iter), n;
 715
 716		if (chunk > PAGE_SIZE)
 717			chunk = PAGE_SIZE;	/* Just for latency reasons */
 718		n = iov_iter_zero(chunk, iter);
 719		if (!n && iov_iter_count(iter))
 720			return written ? written : -EFAULT;
 721		written += n;
 722		if (signal_pending(current))
 723			return written ? written : -ERESTARTSYS;
 
 
 
 
 724		cond_resched();
 725	}
 726	return written;
 727}
 728
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 729static int mmap_zero(struct file *file, struct vm_area_struct *vma)
 730{
 731#ifndef CONFIG_MMU
 732	return -ENOSYS;
 733#endif
 734	if (vma->vm_flags & VM_SHARED)
 735		return shmem_zero_setup(vma);
 736	vma_set_anonymous(vma);
 737	return 0;
 738}
 739
 740static unsigned long get_unmapped_area_zero(struct file *file,
 741				unsigned long addr, unsigned long len,
 742				unsigned long pgoff, unsigned long flags)
 743{
 744#ifdef CONFIG_MMU
 745	if (flags & MAP_SHARED) {
 746		/*
 747		 * mmap_zero() will call shmem_zero_setup() to create a file,
 748		 * so use shmem's get_unmapped_area in case it can be huge;
 749		 * and pass NULL for file as in mmap.c's get_unmapped_area(),
 750		 * so as not to confuse shmem with our handle on "/dev/zero".
 751		 */
 752		return shmem_get_unmapped_area(NULL, addr, len, pgoff, flags);
 753	}
 754
 755	/* Otherwise flags & MAP_PRIVATE: with no shmem object beneath it */
 756	return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
 757#else
 758	return -ENOSYS;
 759#endif
 760}
 761
 762static ssize_t write_full(struct file *file, const char __user *buf,
 763			  size_t count, loff_t *ppos)
 764{
 765	return -ENOSPC;
 766}
 767
 768/*
 769 * Special lseek() function for /dev/null and /dev/zero.  Most notably, you
 770 * can fopen() both devices with "a" now.  This was previously impossible.
 771 * -- SRB.
 772 */
 773static loff_t null_lseek(struct file *file, loff_t offset, int orig)
 774{
 775	return file->f_pos = 0;
 776}
 777
 778/*
 779 * The memory devices use the full 32/64 bits of the offset, and so we cannot
 780 * check against negative addresses: they are ok. The return value is weird,
 781 * though, in that case (0).
 782 *
 783 * also note that seeking relative to the "end of file" isn't supported:
 784 * it has no meaning, so it returns -EINVAL.
 785 */
 786static loff_t memory_lseek(struct file *file, loff_t offset, int orig)
 787{
 788	loff_t ret;
 789
 790	inode_lock(file_inode(file));
 791	switch (orig) {
 792	case SEEK_CUR:
 793		offset += file->f_pos;
 794		fallthrough;
 795	case SEEK_SET:
 796		/* to avoid userland mistaking f_pos=-9 as -EBADF=-9 */
 797		if ((unsigned long long)offset >= -MAX_ERRNO) {
 798			ret = -EOVERFLOW;
 799			break;
 800		}
 801		file->f_pos = offset;
 802		ret = file->f_pos;
 803		force_successful_syscall_return();
 804		break;
 805	default:
 806		ret = -EINVAL;
 807	}
 808	inode_unlock(file_inode(file));
 809	return ret;
 810}
 811
 812static struct inode *devmem_inode;
 813
 814#ifdef CONFIG_IO_STRICT_DEVMEM
 815void revoke_devmem(struct resource *res)
 816{
 817	/* pairs with smp_store_release() in devmem_init_inode() */
 818	struct inode *inode = smp_load_acquire(&devmem_inode);
 819
 820	/*
 821	 * Check that the initialization has completed. Losing the race
 822	 * is ok because it means drivers are claiming resources before
 823	 * the fs_initcall level of init and prevent /dev/mem from
 824	 * establishing mappings.
 825	 */
 826	if (!inode)
 827		return;
 828
 829	/*
 830	 * The expectation is that the driver has successfully marked
 831	 * the resource busy by this point, so devmem_is_allowed()
 832	 * should start returning false, however for performance this
 833	 * does not iterate the entire resource range.
 834	 */
 835	if (devmem_is_allowed(PHYS_PFN(res->start)) &&
 836	    devmem_is_allowed(PHYS_PFN(res->end))) {
 837		/*
 838		 * *cringe* iomem=relaxed says "go ahead, what's the
 839		 * worst that can happen?"
 840		 */
 841		return;
 842	}
 843
 844	unmap_mapping_range(inode->i_mapping, res->start, resource_size(res), 1);
 845}
 846#endif
 847
 848static int open_port(struct inode *inode, struct file *filp)
 849{
 850	int rc;
 851
 852	if (!capable(CAP_SYS_RAWIO))
 853		return -EPERM;
 854
 855	rc = security_locked_down(LOCKDOWN_DEV_MEM);
 856	if (rc)
 857		return rc;
 858
 859	if (iminor(inode) != DEVMEM_MINOR)
 860		return 0;
 861
 862	/*
 863	 * Use a unified address space to have a single point to manage
 864	 * revocations when drivers want to take over a /dev/mem mapped
 865	 * range.
 866	 */
 867	inode->i_mapping = devmem_inode->i_mapping;
 868	filp->f_mapping = inode->i_mapping;
 869
 870	return 0;
 871}
 872
 873#define zero_lseek	null_lseek
 874#define full_lseek      null_lseek
 875#define write_zero	write_null
 876#define write_iter_zero	write_iter_null
 877#define open_mem	open_port
 878#define open_kmem	open_mem
 879
 880static const struct file_operations __maybe_unused mem_fops = {
 881	.llseek		= memory_lseek,
 882	.read		= read_mem,
 883	.write		= write_mem,
 884	.mmap		= mmap_mem,
 885	.open		= open_mem,
 886#ifndef CONFIG_MMU
 887	.get_unmapped_area = get_unmapped_area_mem,
 888	.mmap_capabilities = memory_mmap_capabilities,
 889#endif
 890};
 891
 892static const struct file_operations __maybe_unused kmem_fops = {
 893	.llseek		= memory_lseek,
 894	.read		= read_kmem,
 895	.write		= write_kmem,
 896	.mmap		= mmap_kmem,
 897	.open		= open_kmem,
 898#ifndef CONFIG_MMU
 899	.get_unmapped_area = get_unmapped_area_mem,
 900	.mmap_capabilities = memory_mmap_capabilities,
 901#endif
 902};
 903
 904static const struct file_operations null_fops = {
 905	.llseek		= null_lseek,
 906	.read		= read_null,
 907	.write		= write_null,
 908	.read_iter	= read_iter_null,
 909	.write_iter	= write_iter_null,
 910	.splice_write	= splice_write_null,
 
 911};
 912
 913static const struct file_operations __maybe_unused port_fops = {
 914	.llseek		= memory_lseek,
 915	.read		= read_port,
 916	.write		= write_port,
 917	.open		= open_port,
 918};
 919
 920static const struct file_operations zero_fops = {
 921	.llseek		= zero_lseek,
 922	.write		= write_zero,
 923	.read_iter	= read_iter_zero,
 
 924	.write_iter	= write_iter_zero,
 925	.mmap		= mmap_zero,
 926	.get_unmapped_area = get_unmapped_area_zero,
 927#ifndef CONFIG_MMU
 928	.mmap_capabilities = zero_mmap_capabilities,
 929#endif
 930};
 931
 932static const struct file_operations full_fops = {
 933	.llseek		= full_lseek,
 934	.read_iter	= read_iter_zero,
 935	.write		= write_full,
 936};
 937
 938static const struct memdev {
 939	const char *name;
 940	umode_t mode;
 941	const struct file_operations *fops;
 942	fmode_t fmode;
 943} devlist[] = {
 944#ifdef CONFIG_DEVMEM
 945	 [DEVMEM_MINOR] = { "mem", 0, &mem_fops, FMODE_UNSIGNED_OFFSET },
 946#endif
 947#ifdef CONFIG_DEVKMEM
 948	 [2] = { "kmem", 0, &kmem_fops, FMODE_UNSIGNED_OFFSET },
 949#endif
 950	 [3] = { "null", 0666, &null_fops, 0 },
 951#ifdef CONFIG_DEVPORT
 952	 [4] = { "port", 0, &port_fops, 0 },
 953#endif
 954	 [5] = { "zero", 0666, &zero_fops, 0 },
 955	 [7] = { "full", 0666, &full_fops, 0 },
 956	 [8] = { "random", 0666, &random_fops, 0 },
 957	 [9] = { "urandom", 0666, &urandom_fops, 0 },
 958#ifdef CONFIG_PRINTK
 959	[11] = { "kmsg", 0644, &kmsg_fops, 0 },
 960#endif
 961};
 962
 963static int memory_open(struct inode *inode, struct file *filp)
 964{
 965	int minor;
 966	const struct memdev *dev;
 967
 968	minor = iminor(inode);
 969	if (minor >= ARRAY_SIZE(devlist))
 970		return -ENXIO;
 971
 972	dev = &devlist[minor];
 973	if (!dev->fops)
 974		return -ENXIO;
 975
 976	filp->f_op = dev->fops;
 977	filp->f_mode |= dev->fmode;
 978
 979	if (dev->fops->open)
 980		return dev->fops->open(inode, filp);
 981
 982	return 0;
 983}
 984
 985static const struct file_operations memory_fops = {
 986	.open = memory_open,
 987	.llseek = noop_llseek,
 988};
 989
 990static char *mem_devnode(struct device *dev, umode_t *mode)
 991{
 992	if (mode && devlist[MINOR(dev->devt)].mode)
 993		*mode = devlist[MINOR(dev->devt)].mode;
 994	return NULL;
 995}
 996
 997static struct class *mem_class;
 998
 999static int devmem_fs_init_fs_context(struct fs_context *fc)
1000{
1001	return init_pseudo(fc, DEVMEM_MAGIC) ? 0 : -ENOMEM;
1002}
1003
1004static struct file_system_type devmem_fs_type = {
1005	.name		= "devmem",
1006	.owner		= THIS_MODULE,
1007	.init_fs_context = devmem_fs_init_fs_context,
1008	.kill_sb	= kill_anon_super,
1009};
1010
1011static int devmem_init_inode(void)
1012{
1013	static struct vfsmount *devmem_vfs_mount;
1014	static int devmem_fs_cnt;
1015	struct inode *inode;
1016	int rc;
1017
1018	rc = simple_pin_fs(&devmem_fs_type, &devmem_vfs_mount, &devmem_fs_cnt);
1019	if (rc < 0) {
1020		pr_err("Cannot mount /dev/mem pseudo filesystem: %d\n", rc);
1021		return rc;
1022	}
1023
1024	inode = alloc_anon_inode(devmem_vfs_mount->mnt_sb);
1025	if (IS_ERR(inode)) {
1026		rc = PTR_ERR(inode);
1027		pr_err("Cannot allocate inode for /dev/mem: %d\n", rc);
1028		simple_release_fs(&devmem_vfs_mount, &devmem_fs_cnt);
1029		return rc;
1030	}
1031
1032	/*
1033	 * Publish /dev/mem initialized.
1034	 * Pairs with smp_load_acquire() in revoke_devmem().
1035	 */
1036	smp_store_release(&devmem_inode, inode);
1037
1038	return 0;
1039}
1040
1041static int __init chr_dev_init(void)
1042{
1043	int minor;
1044
1045	if (register_chrdev(MEM_MAJOR, "mem", &memory_fops))
1046		printk("unable to get major %d for memory devs\n", MEM_MAJOR);
1047
1048	mem_class = class_create(THIS_MODULE, "mem");
1049	if (IS_ERR(mem_class))
1050		return PTR_ERR(mem_class);
1051
1052	mem_class->devnode = mem_devnode;
1053	for (minor = 1; minor < ARRAY_SIZE(devlist); minor++) {
1054		if (!devlist[minor].name)
1055			continue;
1056
1057		/*
1058		 * Create /dev/port?
1059		 */
1060		if ((minor == DEVPORT_MINOR) && !arch_has_dev_port())
1061			continue;
1062		if ((minor == DEVMEM_MINOR) && devmem_init_inode() != 0)
1063			continue;
1064
1065		device_create(mem_class, NULL, MKDEV(MEM_MAJOR, minor),
1066			      NULL, devlist[minor].name);
1067	}
1068
1069	return tty_init();
1070}
1071
1072fs_initcall(chr_dev_init);
v6.2
  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/tty.h>
 20#include <linux/capability.h>
 21#include <linux/ptrace.h>
 22#include <linux/device.h>
 23#include <linux/highmem.h>
 24#include <linux/backing-dev.h>
 25#include <linux/shmem_fs.h>
 26#include <linux/splice.h>
 27#include <linux/pfn.h>
 28#include <linux/export.h>
 29#include <linux/io.h>
 30#include <linux/uio.h>
 31#include <linux/uaccess.h>
 32#include <linux/security.h>
 
 
 
 33
 34#ifdef CONFIG_IA64
 35# include <linux/efi.h>
 36#endif
 37
 38#define DEVMEM_MINOR	1
 39#define DEVPORT_MINOR	4
 40
 41static inline unsigned long size_inside_page(unsigned long start,
 42					     unsigned long size)
 43{
 44	unsigned long sz;
 45
 46	sz = PAGE_SIZE - (start & (PAGE_SIZE - 1));
 47
 48	return min(sz, size);
 49}
 50
 51#ifndef ARCH_HAS_VALID_PHYS_ADDR_RANGE
 52static inline int valid_phys_addr_range(phys_addr_t addr, size_t count)
 53{
 54	return addr + count <= __pa(high_memory);
 55}
 56
 57static inline int valid_mmap_phys_addr_range(unsigned long pfn, size_t size)
 58{
 59	return 1;
 60}
 61#endif
 62
 63#ifdef CONFIG_STRICT_DEVMEM
 64static inline int page_is_allowed(unsigned long pfn)
 65{
 66	return devmem_is_allowed(pfn);
 67}
 68static inline int range_is_allowed(unsigned long pfn, unsigned long size)
 69{
 70	u64 from = ((u64)pfn) << PAGE_SHIFT;
 71	u64 to = from + size;
 72	u64 cursor = from;
 73
 74	while (cursor < to) {
 75		if (!devmem_is_allowed(pfn))
 76			return 0;
 77		cursor += PAGE_SIZE;
 78		pfn++;
 79	}
 80	return 1;
 81}
 82#else
 83static inline int page_is_allowed(unsigned long pfn)
 84{
 85	return 1;
 86}
 87static inline int range_is_allowed(unsigned long pfn, unsigned long size)
 88{
 89	return 1;
 90}
 91#endif
 92
 93#ifndef unxlate_dev_mem_ptr
 94#define unxlate_dev_mem_ptr unxlate_dev_mem_ptr
 95void __weak unxlate_dev_mem_ptr(phys_addr_t phys, void *addr)
 96{
 97}
 98#endif
 99
100static inline bool should_stop_iteration(void)
101{
102	if (need_resched())
103		cond_resched();
104	return signal_pending(current);
105}
106
107/*
108 * This funcion reads the *physical* memory. The f_pos points directly to the
109 * memory location.
110 */
111static ssize_t read_mem(struct file *file, char __user *buf,
112			size_t count, loff_t *ppos)
113{
114	phys_addr_t p = *ppos;
115	ssize_t read, sz;
116	void *ptr;
117	char *bounce;
118	int err;
119
120	if (p != *ppos)
121		return 0;
122
123	if (!valid_phys_addr_range(p, count))
124		return -EFAULT;
125	read = 0;
126#ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
127	/* we don't have page 0 mapped on sparc and m68k.. */
128	if (p < PAGE_SIZE) {
129		sz = size_inside_page(p, 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	bounce = kmalloc(PAGE_SIZE, GFP_KERNEL);
142	if (!bounce)
143		return -ENOMEM;
144
145	while (count > 0) {
146		unsigned long remaining;
147		int allowed, probe;
148
149		sz = size_inside_page(p, count);
150
151		err = -EPERM;
152		allowed = page_is_allowed(p >> PAGE_SHIFT);
153		if (!allowed)
154			goto failed;
155
156		err = -EFAULT;
157		if (allowed == 2) {
158			/* Show zeros for restricted memory. */
159			remaining = clear_user(buf, sz);
160		} else {
161			/*
162			 * On ia64 if a page has been mapped somewhere as
163			 * uncached, then it must also be accessed uncached
164			 * by the kernel or data corruption may occur.
165			 */
166			ptr = xlate_dev_mem_ptr(p);
167			if (!ptr)
168				goto failed;
169
170			probe = copy_from_kernel_nofault(bounce, ptr, sz);
171			unxlate_dev_mem_ptr(p, ptr);
172			if (probe)
173				goto failed;
174
175			remaining = copy_to_user(buf, bounce, sz);
176		}
177
178		if (remaining)
179			goto failed;
180
181		buf += sz;
182		p += sz;
183		count -= sz;
184		read += sz;
185		if (should_stop_iteration())
186			break;
187	}
188	kfree(bounce);
189
190	*ppos += read;
191	return read;
192
193failed:
194	kfree(bounce);
195	return err;
196}
197
198static ssize_t write_mem(struct file *file, const char __user *buf,
199			 size_t count, loff_t *ppos)
200{
201	phys_addr_t p = *ppos;
202	ssize_t written, sz;
203	unsigned long copied;
204	void *ptr;
205
206	if (p != *ppos)
207		return -EFBIG;
208
209	if (!valid_phys_addr_range(p, count))
210		return -EFAULT;
211
212	written = 0;
213
214#ifdef __ARCH_HAS_NO_PAGE_ZERO_MAPPED
215	/* we don't have page 0 mapped on sparc and m68k.. */
216	if (p < PAGE_SIZE) {
217		sz = size_inside_page(p, count);
218		/* Hmm. Do something? */
219		buf += sz;
220		p += sz;
221		count -= sz;
222		written += sz;
223	}
224#endif
225
226	while (count > 0) {
227		int allowed;
228
229		sz = size_inside_page(p, count);
230
231		allowed = page_is_allowed(p >> PAGE_SHIFT);
232		if (!allowed)
233			return -EPERM;
234
235		/* Skip actual writing when a page is marked as restricted. */
236		if (allowed == 1) {
237			/*
238			 * On ia64 if a page has been mapped somewhere as
239			 * uncached, then it must also be accessed uncached
240			 * by the kernel or data corruption may occur.
241			 */
242			ptr = xlate_dev_mem_ptr(p);
243			if (!ptr) {
244				if (written)
245					break;
246				return -EFAULT;
247			}
248
249			copied = copy_from_user(ptr, buf, sz);
250			unxlate_dev_mem_ptr(p, ptr);
251			if (copied) {
252				written += sz - copied;
253				if (written)
254					break;
255				return -EFAULT;
256			}
257		}
258
259		buf += sz;
260		p += sz;
261		count -= sz;
262		written += sz;
263		if (should_stop_iteration())
264			break;
265	}
266
267	*ppos += written;
268	return written;
269}
270
271int __weak phys_mem_access_prot_allowed(struct file *file,
272	unsigned long pfn, unsigned long size, pgprot_t *vma_prot)
273{
274	return 1;
275}
276
277#ifndef __HAVE_PHYS_MEM_ACCESS_PROT
278
279/*
280 * Architectures vary in how they handle caching for addresses
281 * outside of main memory.
282 *
283 */
284#ifdef pgprot_noncached
285static int uncached_access(struct file *file, phys_addr_t addr)
286{
287#if defined(CONFIG_IA64)
288	/*
289	 * On ia64, we ignore O_DSYNC because we cannot tolerate memory
290	 * attribute aliases.
291	 */
292	return !(efi_mem_attributes(addr) & EFI_MEMORY_WB);
 
 
 
 
 
 
 
293#else
294	/*
295	 * Accessing memory above the top the kernel knows about or through a
296	 * file pointer
297	 * that was marked O_DSYNC will be done non-cached.
298	 */
299	if (file->f_flags & O_DSYNC)
300		return 1;
301	return addr >= __pa(high_memory);
302#endif
303}
304#endif
305
306static pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
307				     unsigned long size, pgprot_t vma_prot)
308{
309#ifdef pgprot_noncached
310	phys_addr_t offset = pfn << PAGE_SHIFT;
311
312	if (uncached_access(file, offset))
313		return pgprot_noncached(vma_prot);
314#endif
315	return vma_prot;
316}
317#endif
318
319#ifndef CONFIG_MMU
320static unsigned long get_unmapped_area_mem(struct file *file,
321					   unsigned long addr,
322					   unsigned long len,
323					   unsigned long pgoff,
324					   unsigned long flags)
325{
326	if (!valid_mmap_phys_addr_range(pgoff, len))
327		return (unsigned long) -EINVAL;
328	return pgoff << PAGE_SHIFT;
329}
330
331/* permit direct mmap, for read, write or exec */
332static unsigned memory_mmap_capabilities(struct file *file)
333{
334	return NOMMU_MAP_DIRECT |
335		NOMMU_MAP_READ | NOMMU_MAP_WRITE | NOMMU_MAP_EXEC;
336}
337
338static unsigned zero_mmap_capabilities(struct file *file)
339{
340	return NOMMU_MAP_COPY;
341}
342
343/* can't do an in-place private mapping if there's no MMU */
344static inline int private_mapping_ok(struct vm_area_struct *vma)
345{
346	return vma->vm_flags & VM_MAYSHARE;
347}
348#else
349
350static inline int private_mapping_ok(struct vm_area_struct *vma)
351{
352	return 1;
353}
354#endif
355
356static const struct vm_operations_struct mmap_mem_ops = {
357#ifdef CONFIG_HAVE_IOREMAP_PROT
358	.access = generic_access_phys
359#endif
360};
361
362static int mmap_mem(struct file *file, struct vm_area_struct *vma)
363{
364	size_t size = vma->vm_end - vma->vm_start;
365	phys_addr_t offset = (phys_addr_t)vma->vm_pgoff << PAGE_SHIFT;
366
367	/* Does it even fit in phys_addr_t? */
368	if (offset >> PAGE_SHIFT != vma->vm_pgoff)
369		return -EINVAL;
370
371	/* It's illegal to wrap around the end of the physical address space. */
372	if (offset + (phys_addr_t)size - 1 < offset)
373		return -EINVAL;
374
375	if (!valid_mmap_phys_addr_range(vma->vm_pgoff, size))
376		return -EINVAL;
377
378	if (!private_mapping_ok(vma))
379		return -ENOSYS;
380
381	if (!range_is_allowed(vma->vm_pgoff, size))
382		return -EPERM;
383
384	if (!phys_mem_access_prot_allowed(file, vma->vm_pgoff, size,
385						&vma->vm_page_prot))
386		return -EINVAL;
387
388	vma->vm_page_prot = phys_mem_access_prot(file, vma->vm_pgoff,
389						 size,
390						 vma->vm_page_prot);
391
392	vma->vm_ops = &mmap_mem_ops;
393
394	/* Remap-pfn-range will mark the range VM_IO */
395	if (remap_pfn_range(vma,
396			    vma->vm_start,
397			    vma->vm_pgoff,
398			    size,
399			    vma->vm_page_prot)) {
400		return -EAGAIN;
401	}
402	return 0;
403}
404
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
405static ssize_t read_port(struct file *file, char __user *buf,
406			 size_t count, loff_t *ppos)
407{
408	unsigned long i = *ppos;
409	char __user *tmp = buf;
410
411	if (!access_ok(buf, count))
412		return -EFAULT;
413	while (count-- > 0 && i < 65536) {
414		if (__put_user(inb(i), tmp) < 0)
415			return -EFAULT;
416		i++;
417		tmp++;
418	}
419	*ppos = i;
420	return tmp-buf;
421}
422
423static ssize_t write_port(struct file *file, const char __user *buf,
424			  size_t count, loff_t *ppos)
425{
426	unsigned long i = *ppos;
427	const char __user *tmp = buf;
428
429	if (!access_ok(buf, count))
430		return -EFAULT;
431	while (count-- > 0 && i < 65536) {
432		char c;
433
434		if (__get_user(c, tmp)) {
435			if (tmp > buf)
436				break;
437			return -EFAULT;
438		}
439		outb(c, i);
440		i++;
441		tmp++;
442	}
443	*ppos = i;
444	return tmp-buf;
445}
446
447static ssize_t read_null(struct file *file, char __user *buf,
448			 size_t count, loff_t *ppos)
449{
450	return 0;
451}
452
453static ssize_t write_null(struct file *file, const char __user *buf,
454			  size_t count, loff_t *ppos)
455{
456	return count;
457}
458
459static ssize_t read_iter_null(struct kiocb *iocb, struct iov_iter *to)
460{
461	return 0;
462}
463
464static ssize_t write_iter_null(struct kiocb *iocb, struct iov_iter *from)
465{
466	size_t count = iov_iter_count(from);
467	iov_iter_advance(from, count);
468	return count;
469}
470
471static int pipe_to_null(struct pipe_inode_info *info, struct pipe_buffer *buf,
472			struct splice_desc *sd)
473{
474	return sd->len;
475}
476
477static ssize_t splice_write_null(struct pipe_inode_info *pipe, struct file *out,
478				 loff_t *ppos, size_t len, unsigned int flags)
479{
480	return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_null);
481}
482
483static int uring_cmd_null(struct io_uring_cmd *ioucmd, unsigned int issue_flags)
484{
485	return 0;
486}
487
488static ssize_t read_iter_zero(struct kiocb *iocb, struct iov_iter *iter)
489{
490	size_t written = 0;
491
492	while (iov_iter_count(iter)) {
493		size_t chunk = iov_iter_count(iter), n;
494
495		if (chunk > PAGE_SIZE)
496			chunk = PAGE_SIZE;	/* Just for latency reasons */
497		n = iov_iter_zero(chunk, iter);
498		if (!n && iov_iter_count(iter))
499			return written ? written : -EFAULT;
500		written += n;
501		if (signal_pending(current))
502			return written ? written : -ERESTARTSYS;
503		if (!need_resched())
504			continue;
505		if (iocb->ki_flags & IOCB_NOWAIT)
506			return written ? written : -EAGAIN;
507		cond_resched();
508	}
509	return written;
510}
511
512static ssize_t read_zero(struct file *file, char __user *buf,
513			 size_t count, loff_t *ppos)
514{
515	size_t cleared = 0;
516
517	while (count) {
518		size_t chunk = min_t(size_t, count, PAGE_SIZE);
519		size_t left;
520
521		left = clear_user(buf + cleared, chunk);
522		if (unlikely(left)) {
523			cleared += (chunk - left);
524			if (!cleared)
525				return -EFAULT;
526			break;
527		}
528		cleared += chunk;
529		count -= chunk;
530
531		if (signal_pending(current))
532			break;
533		cond_resched();
534	}
535
536	return cleared;
537}
538
539static int mmap_zero(struct file *file, struct vm_area_struct *vma)
540{
541#ifndef CONFIG_MMU
542	return -ENOSYS;
543#endif
544	if (vma->vm_flags & VM_SHARED)
545		return shmem_zero_setup(vma);
546	vma_set_anonymous(vma);
547	return 0;
548}
549
550static unsigned long get_unmapped_area_zero(struct file *file,
551				unsigned long addr, unsigned long len,
552				unsigned long pgoff, unsigned long flags)
553{
554#ifdef CONFIG_MMU
555	if (flags & MAP_SHARED) {
556		/*
557		 * mmap_zero() will call shmem_zero_setup() to create a file,
558		 * so use shmem's get_unmapped_area in case it can be huge;
559		 * and pass NULL for file as in mmap.c's get_unmapped_area(),
560		 * so as not to confuse shmem with our handle on "/dev/zero".
561		 */
562		return shmem_get_unmapped_area(NULL, addr, len, pgoff, flags);
563	}
564
565	/* Otherwise flags & MAP_PRIVATE: with no shmem object beneath it */
566	return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
567#else
568	return -ENOSYS;
569#endif
570}
571
572static ssize_t write_full(struct file *file, const char __user *buf,
573			  size_t count, loff_t *ppos)
574{
575	return -ENOSPC;
576}
577
578/*
579 * Special lseek() function for /dev/null and /dev/zero.  Most notably, you
580 * can fopen() both devices with "a" now.  This was previously impossible.
581 * -- SRB.
582 */
583static loff_t null_lseek(struct file *file, loff_t offset, int orig)
584{
585	return file->f_pos = 0;
586}
587
588/*
589 * The memory devices use the full 32/64 bits of the offset, and so we cannot
590 * check against negative addresses: they are ok. The return value is weird,
591 * though, in that case (0).
592 *
593 * also note that seeking relative to the "end of file" isn't supported:
594 * it has no meaning, so it returns -EINVAL.
595 */
596static loff_t memory_lseek(struct file *file, loff_t offset, int orig)
597{
598	loff_t ret;
599
600	inode_lock(file_inode(file));
601	switch (orig) {
602	case SEEK_CUR:
603		offset += file->f_pos;
604		fallthrough;
605	case SEEK_SET:
606		/* to avoid userland mistaking f_pos=-9 as -EBADF=-9 */
607		if ((unsigned long long)offset >= -MAX_ERRNO) {
608			ret = -EOVERFLOW;
609			break;
610		}
611		file->f_pos = offset;
612		ret = file->f_pos;
613		force_successful_syscall_return();
614		break;
615	default:
616		ret = -EINVAL;
617	}
618	inode_unlock(file_inode(file));
619	return ret;
620}
621
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
622static int open_port(struct inode *inode, struct file *filp)
623{
624	int rc;
625
626	if (!capable(CAP_SYS_RAWIO))
627		return -EPERM;
628
629	rc = security_locked_down(LOCKDOWN_DEV_MEM);
630	if (rc)
631		return rc;
632
633	if (iminor(inode) != DEVMEM_MINOR)
634		return 0;
635
636	/*
637	 * Use a unified address space to have a single point to manage
638	 * revocations when drivers want to take over a /dev/mem mapped
639	 * range.
640	 */
641	filp->f_mapping = iomem_get_mapping();
 
642
643	return 0;
644}
645
646#define zero_lseek	null_lseek
647#define full_lseek      null_lseek
648#define write_zero	write_null
649#define write_iter_zero	write_iter_null
650#define open_mem	open_port
 
651
652static const struct file_operations __maybe_unused mem_fops = {
653	.llseek		= memory_lseek,
654	.read		= read_mem,
655	.write		= write_mem,
656	.mmap		= mmap_mem,
657	.open		= open_mem,
658#ifndef CONFIG_MMU
659	.get_unmapped_area = get_unmapped_area_mem,
660	.mmap_capabilities = memory_mmap_capabilities,
661#endif
662};
663
 
 
 
 
 
 
 
 
 
 
 
 
664static const struct file_operations null_fops = {
665	.llseek		= null_lseek,
666	.read		= read_null,
667	.write		= write_null,
668	.read_iter	= read_iter_null,
669	.write_iter	= write_iter_null,
670	.splice_write	= splice_write_null,
671	.uring_cmd	= uring_cmd_null,
672};
673
674static const struct file_operations __maybe_unused port_fops = {
675	.llseek		= memory_lseek,
676	.read		= read_port,
677	.write		= write_port,
678	.open		= open_port,
679};
680
681static const struct file_operations zero_fops = {
682	.llseek		= zero_lseek,
683	.write		= write_zero,
684	.read_iter	= read_iter_zero,
685	.read		= read_zero,
686	.write_iter	= write_iter_zero,
687	.mmap		= mmap_zero,
688	.get_unmapped_area = get_unmapped_area_zero,
689#ifndef CONFIG_MMU
690	.mmap_capabilities = zero_mmap_capabilities,
691#endif
692};
693
694static const struct file_operations full_fops = {
695	.llseek		= full_lseek,
696	.read_iter	= read_iter_zero,
697	.write		= write_full,
698};
699
700static const struct memdev {
701	const char *name;
702	umode_t mode;
703	const struct file_operations *fops;
704	fmode_t fmode;
705} devlist[] = {
706#ifdef CONFIG_DEVMEM
707	 [DEVMEM_MINOR] = { "mem", 0, &mem_fops, FMODE_UNSIGNED_OFFSET },
708#endif
709	 [3] = { "null", 0666, &null_fops, FMODE_NOWAIT },
 
 
 
710#ifdef CONFIG_DEVPORT
711	 [4] = { "port", 0, &port_fops, 0 },
712#endif
713	 [5] = { "zero", 0666, &zero_fops, FMODE_NOWAIT },
714	 [7] = { "full", 0666, &full_fops, 0 },
715	 [8] = { "random", 0666, &random_fops, FMODE_NOWAIT },
716	 [9] = { "urandom", 0666, &urandom_fops, FMODE_NOWAIT },
717#ifdef CONFIG_PRINTK
718	[11] = { "kmsg", 0644, &kmsg_fops, 0 },
719#endif
720};
721
722static int memory_open(struct inode *inode, struct file *filp)
723{
724	int minor;
725	const struct memdev *dev;
726
727	minor = iminor(inode);
728	if (minor >= ARRAY_SIZE(devlist))
729		return -ENXIO;
730
731	dev = &devlist[minor];
732	if (!dev->fops)
733		return -ENXIO;
734
735	filp->f_op = dev->fops;
736	filp->f_mode |= dev->fmode;
737
738	if (dev->fops->open)
739		return dev->fops->open(inode, filp);
740
741	return 0;
742}
743
744static const struct file_operations memory_fops = {
745	.open = memory_open,
746	.llseek = noop_llseek,
747};
748
749static char *mem_devnode(const struct device *dev, umode_t *mode)
750{
751	if (mode && devlist[MINOR(dev->devt)].mode)
752		*mode = devlist[MINOR(dev->devt)].mode;
753	return NULL;
754}
755
756static struct class *mem_class;
757
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
758static int __init chr_dev_init(void)
759{
760	int minor;
761
762	if (register_chrdev(MEM_MAJOR, "mem", &memory_fops))
763		printk("unable to get major %d for memory devs\n", MEM_MAJOR);
764
765	mem_class = class_create(THIS_MODULE, "mem");
766	if (IS_ERR(mem_class))
767		return PTR_ERR(mem_class);
768
769	mem_class->devnode = mem_devnode;
770	for (minor = 1; minor < ARRAY_SIZE(devlist); minor++) {
771		if (!devlist[minor].name)
772			continue;
773
774		/*
775		 * Create /dev/port?
776		 */
777		if ((minor == DEVPORT_MINOR) && !arch_has_dev_port())
 
 
778			continue;
779
780		device_create(mem_class, NULL, MKDEV(MEM_MAJOR, minor),
781			      NULL, devlist[minor].name);
782	}
783
784	return tty_init();
785}
786
787fs_initcall(chr_dev_init);