Linux Audio

Check our new training course

Loading...
v3.1
 
   1/*
   2 * mm/mmap.c
   3 *
   4 * Written by obz.
   5 *
   6 * Address space accounting code	<alan@lxorguk.ukuu.org.uk>
   7 */
   8
 
 
 
   9#include <linux/slab.h>
  10#include <linux/backing-dev.h>
  11#include <linux/mm.h>
 
  12#include <linux/shm.h>
  13#include <linux/mman.h>
  14#include <linux/pagemap.h>
  15#include <linux/swap.h>
  16#include <linux/syscalls.h>
  17#include <linux/capability.h>
  18#include <linux/init.h>
  19#include <linux/file.h>
  20#include <linux/fs.h>
  21#include <linux/personality.h>
  22#include <linux/security.h>
  23#include <linux/hugetlb.h>
 
  24#include <linux/profile.h>
  25#include <linux/module.h>
  26#include <linux/mount.h>
  27#include <linux/mempolicy.h>
  28#include <linux/rmap.h>
  29#include <linux/mmu_notifier.h>
 
  30#include <linux/perf_event.h>
  31#include <linux/audit.h>
  32#include <linux/khugepaged.h>
 
 
 
 
 
 
 
 
 
 
 
  33
  34#include <asm/uaccess.h>
  35#include <asm/cacheflush.h>
  36#include <asm/tlb.h>
  37#include <asm/mmu_context.h>
  38
 
 
 
  39#include "internal.h"
  40
  41#ifndef arch_mmap_check
  42#define arch_mmap_check(addr, len, flags)	(0)
  43#endif
  44
  45#ifndef arch_rebalance_pgtables
  46#define arch_rebalance_pgtables(addr, len)		(addr)
 
 
 
 
 
 
 
  47#endif
  48
  49static void unmap_region(struct mm_struct *mm,
  50		struct vm_area_struct *vma, struct vm_area_struct *prev,
  51		unsigned long start, unsigned long end);
  52
  53/*
  54 * WARNING: the debugging will use recursive algorithms so never enable this
  55 * unless you know what you are doing.
  56 */
  57#undef DEBUG_MM_RB
  58
  59/* description of effects of mapping type and prot in current implementation.
  60 * this is due to the limited x86 page protection hardware.  The expected
  61 * behavior is in parens:
  62 *
  63 * map_type	prot
  64 *		PROT_NONE	PROT_READ	PROT_WRITE	PROT_EXEC
  65 * MAP_SHARED	r: (no) no	r: (yes) yes	r: (no) yes	r: (no) yes
  66 *		w: (no) no	w: (no) no	w: (yes) yes	w: (no) no
  67 *		x: (no) no	x: (no) yes	x: (no) yes	x: (yes) yes
  68 *		
  69 * MAP_PRIVATE	r: (no) no	r: (yes) yes	r: (no) yes	r: (no) yes
  70 *		w: (no) no	w: (no) no	w: (copy) copy	w: (no) no
  71 *		x: (no) no	x: (no) yes	x: (no) yes	x: (yes) yes
  72 *
  73 */
  74pgprot_t protection_map[16] = {
  75	__P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111,
  76	__S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111
  77};
  78
  79pgprot_t vm_get_page_prot(unsigned long vm_flags)
 
  80{
  81	return __pgprot(pgprot_val(protection_map[vm_flags &
  82				(VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)]) |
  83			pgprot_val(arch_vm_get_page_prot(vm_flags)));
  84}
  85EXPORT_SYMBOL(vm_get_page_prot);
  86
  87int sysctl_overcommit_memory __read_mostly = OVERCOMMIT_GUESS;  /* heuristic overcommit */
  88int sysctl_overcommit_ratio __read_mostly = 50;	/* default is 50% */
  89int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
  90/*
  91 * Make sure vm_committed_as in one cacheline and not cacheline shared with
  92 * other variables. It can be updated by several CPUs frequently.
  93 */
  94struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp;
  95
  96/*
  97 * Check that a process has enough memory to allocate a new virtual
  98 * mapping. 0 means there is enough memory for the allocation to
  99 * succeed and -ENOMEM implies there is not.
 100 *
 101 * We currently support three overcommit policies, which are set via the
 102 * vm.overcommit_memory sysctl.  See Documentation/vm/overcommit-accounting
 103 *
 104 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
 105 * Additional code 2002 Jul 20 by Robert Love.
 106 *
 107 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
 108 *
 109 * Note this is a helper function intended to be used by LSMs which
 110 * wish to use this logic.
 111 */
 112int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
 113{
 114	unsigned long free, allowed;
 115
 116	vm_acct_memory(pages);
 117
 118	/*
 119	 * Sometimes we want to use more memory than we have
 120	 */
 121	if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
 122		return 0;
 123
 124	if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
 125		free = global_page_state(NR_FREE_PAGES);
 126		free += global_page_state(NR_FILE_PAGES);
 127
 128		/*
 129		 * shmem pages shouldn't be counted as free in this
 130		 * case, they can't be purged, only swapped out, and
 131		 * that won't affect the overall amount of available
 132		 * memory in the system.
 133		 */
 134		free -= global_page_state(NR_SHMEM);
 135
 136		free += nr_swap_pages;
 137
 138		/*
 139		 * Any slabs which are created with the
 140		 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
 141		 * which are reclaimable, under pressure.  The dentry
 142		 * cache and most inode caches should fall into this
 143		 */
 144		free += global_page_state(NR_SLAB_RECLAIMABLE);
 145
 146		/*
 147		 * Leave reserved pages. The pages are not for anonymous pages.
 148		 */
 149		if (free <= totalreserve_pages)
 150			goto error;
 151		else
 152			free -= totalreserve_pages;
 153
 154		/*
 155		 * Leave the last 3% for root
 156		 */
 157		if (!cap_sys_admin)
 158			free -= free / 32;
 159
 160		if (free > pages)
 161			return 0;
 162
 163		goto error;
 164	}
 165
 166	allowed = (totalram_pages - hugetlb_total_pages())
 167	       	* sysctl_overcommit_ratio / 100;
 168	/*
 169	 * Leave the last 3% for root
 170	 */
 171	if (!cap_sys_admin)
 172		allowed -= allowed / 32;
 173	allowed += total_swap_pages;
 174
 175	/* Don't let a single process grow too big:
 176	   leave 3% of the size of this process for other processes */
 177	if (mm)
 178		allowed -= mm->total_vm / 32;
 179
 180	if (percpu_counter_read_positive(&vm_committed_as) < allowed)
 181		return 0;
 182error:
 183	vm_unacct_memory(pages);
 184
 185	return -ENOMEM;
 186}
 187
 188/*
 189 * Requires inode->i_mapping->i_mmap_mutex
 
 
 
 
 
 190 */
 191static void __remove_shared_vm_struct(struct vm_area_struct *vma,
 192		struct file *file, struct address_space *mapping)
 193{
 194	if (vma->vm_flags & VM_DENYWRITE)
 195		atomic_inc(&file->f_path.dentry->d_inode->i_writecount);
 196	if (vma->vm_flags & VM_SHARED)
 197		mapping->i_mmap_writable--;
 198
 199	flush_dcache_mmap_lock(mapping);
 200	if (unlikely(vma->vm_flags & VM_NONLINEAR))
 201		list_del_init(&vma->shared.vm_set.list);
 202	else
 203		vma_prio_tree_remove(vma, &mapping->i_mmap);
 204	flush_dcache_mmap_unlock(mapping);
 205}
 206
 207/*
 208 * Unlink a file-based vm structure from its prio_tree, to hide
 209 * vma from rmap and vmtruncate before freeing its page tables.
 210 */
 211void unlink_file_vma(struct vm_area_struct *vma)
 212{
 213	struct file *file = vma->vm_file;
 214
 215	if (file) {
 216		struct address_space *mapping = file->f_mapping;
 217		mutex_lock(&mapping->i_mmap_mutex);
 218		__remove_shared_vm_struct(vma, file, mapping);
 219		mutex_unlock(&mapping->i_mmap_mutex);
 220	}
 221}
 222
 223/*
 224 * Close a vm structure and free it, returning the next.
 225 */
 226static struct vm_area_struct *remove_vma(struct vm_area_struct *vma)
 227{
 228	struct vm_area_struct *next = vma->vm_next;
 229
 230	might_sleep();
 231	if (vma->vm_ops && vma->vm_ops->close)
 232		vma->vm_ops->close(vma);
 233	if (vma->vm_file) {
 234		fput(vma->vm_file);
 235		if (vma->vm_flags & VM_EXECUTABLE)
 236			removed_exe_file_vma(vma->vm_mm);
 237	}
 238	mpol_put(vma_policy(vma));
 239	kmem_cache_free(vm_area_cachep, vma);
 240	return next;
 241}
 242
 243SYSCALL_DEFINE1(brk, unsigned long, brk)
 244{
 245	unsigned long rlim, retval;
 246	unsigned long newbrk, oldbrk;
 247	struct mm_struct *mm = current->mm;
 
 248	unsigned long min_brk;
 
 
 
 249
 250	down_write(&mm->mmap_sem);
 
 
 
 251
 252#ifdef CONFIG_COMPAT_BRK
 253	/*
 254	 * CONFIG_COMPAT_BRK can still be overridden by setting
 255	 * randomize_va_space to 2, which will still cause mm->start_brk
 256	 * to be arbitrarily shifted
 257	 */
 258	if (current->brk_randomized)
 259		min_brk = mm->start_brk;
 260	else
 261		min_brk = mm->end_data;
 262#else
 263	min_brk = mm->start_brk;
 264#endif
 265	if (brk < min_brk)
 266		goto out;
 267
 268	/*
 269	 * Check against rlimit here. If this check is done later after the test
 270	 * of oldbrk with newbrk then it can escape the test and let the data
 271	 * segment grow beyond its set limit the in case where the limit is
 272	 * not page aligned -Ram Gupta
 273	 */
 274	rlim = rlimit(RLIMIT_DATA);
 275	if (rlim < RLIM_INFINITY && (brk - mm->start_brk) +
 276			(mm->end_data - mm->start_data) > rlim)
 277		goto out;
 278
 279	newbrk = PAGE_ALIGN(brk);
 280	oldbrk = PAGE_ALIGN(mm->brk);
 281	if (oldbrk == newbrk)
 282		goto set_brk;
 
 
 283
 284	/* Always allow shrinking brk. */
 285	if (brk <= mm->brk) {
 286		if (!do_munmap(mm, newbrk, oldbrk-newbrk))
 287			goto set_brk;
 288		goto out;
 289	}
 290
 291	/* Check against existing mmap mappings. */
 292	if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE))
 293		goto out;
 294
 295	/* Ok, looks good - let it rip. */
 296	if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk)
 297		goto out;
 298set_brk:
 299	mm->brk = brk;
 300out:
 301	retval = mm->brk;
 302	up_write(&mm->mmap_sem);
 303	return retval;
 304}
 305
 306#ifdef DEBUG_MM_RB
 307static int browse_rb(struct rb_root *root)
 308{
 309	int i = 0, j;
 310	struct rb_node *nd, *pn = NULL;
 311	unsigned long prev = 0, pend = 0;
 312
 313	for (nd = rb_first(root); nd; nd = rb_next(nd)) {
 314		struct vm_area_struct *vma;
 315		vma = rb_entry(nd, struct vm_area_struct, vm_rb);
 316		if (vma->vm_start < prev)
 317			printk("vm_start %lx prev %lx\n", vma->vm_start, prev), i = -1;
 318		if (vma->vm_start < pend)
 319			printk("vm_start %lx pend %lx\n", vma->vm_start, pend);
 320		if (vma->vm_start > vma->vm_end)
 321			printk("vm_end %lx < vm_start %lx\n", vma->vm_end, vma->vm_start);
 322		i++;
 323		pn = nd;
 324		prev = vma->vm_start;
 325		pend = vma->vm_end;
 326	}
 327	j = 0;
 328	for (nd = pn; nd; nd = rb_prev(nd)) {
 329		j++;
 330	}
 331	if (i != j)
 332		printk("backwards %d, forwards %d\n", j, i), i = 0;
 333	return i;
 334}
 335
 336void validate_mm(struct mm_struct *mm)
 337{
 338	int bug = 0;
 339	int i = 0;
 340	struct vm_area_struct *tmp = mm->mmap;
 341	while (tmp) {
 342		tmp = tmp->vm_next;
 343		i++;
 344	}
 345	if (i != mm->map_count)
 346		printk("map_count %d vm_next %d\n", mm->map_count, i), bug = 1;
 347	i = browse_rb(&mm->mm_rb);
 348	if (i != mm->map_count)
 349		printk("map_count %d rb %d\n", mm->map_count, i), bug = 1;
 350	BUG_ON(bug);
 351}
 352#else
 353#define validate_mm(mm) do { } while (0)
 354#endif
 355
 356static struct vm_area_struct *
 357find_vma_prepare(struct mm_struct *mm, unsigned long addr,
 358		struct vm_area_struct **pprev, struct rb_node ***rb_link,
 359		struct rb_node ** rb_parent)
 360{
 361	struct vm_area_struct * vma;
 362	struct rb_node ** __rb_link, * __rb_parent, * rb_prev;
 363
 364	__rb_link = &mm->mm_rb.rb_node;
 365	rb_prev = __rb_parent = NULL;
 366	vma = NULL;
 367
 368	while (*__rb_link) {
 369		struct vm_area_struct *vma_tmp;
 370
 371		__rb_parent = *__rb_link;
 372		vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb);
 373
 374		if (vma_tmp->vm_end > addr) {
 375			vma = vma_tmp;
 376			if (vma_tmp->vm_start <= addr)
 377				break;
 378			__rb_link = &__rb_parent->rb_left;
 379		} else {
 380			rb_prev = __rb_parent;
 381			__rb_link = &__rb_parent->rb_right;
 382		}
 383	}
 384
 385	*pprev = NULL;
 386	if (rb_prev)
 387		*pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
 388	*rb_link = __rb_link;
 389	*rb_parent = __rb_parent;
 390	return vma;
 391}
 392
 393void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma,
 394		struct rb_node **rb_link, struct rb_node *rb_parent)
 395{
 396	rb_link_node(&vma->vm_rb, rb_parent, rb_link);
 397	rb_insert_color(&vma->vm_rb, &mm->mm_rb);
 398}
 399
 400static void __vma_link_file(struct vm_area_struct *vma)
 401{
 402	struct file *file;
 403
 404	file = vma->vm_file;
 405	if (file) {
 406		struct address_space *mapping = file->f_mapping;
 407
 408		if (vma->vm_flags & VM_DENYWRITE)
 409			atomic_dec(&file->f_path.dentry->d_inode->i_writecount);
 410		if (vma->vm_flags & VM_SHARED)
 411			mapping->i_mmap_writable++;
 412
 413		flush_dcache_mmap_lock(mapping);
 414		if (unlikely(vma->vm_flags & VM_NONLINEAR))
 415			vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear);
 416		else
 417			vma_prio_tree_insert(vma, &mapping->i_mmap);
 418		flush_dcache_mmap_unlock(mapping);
 419	}
 420}
 421
 422static void
 423__vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
 424	struct vm_area_struct *prev, struct rb_node **rb_link,
 425	struct rb_node *rb_parent)
 426{
 427	__vma_link_list(mm, vma, prev, rb_parent);
 428	__vma_link_rb(mm, vma, rb_link, rb_parent);
 429}
 430
 431static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
 432			struct vm_area_struct *prev, struct rb_node **rb_link,
 433			struct rb_node *rb_parent)
 434{
 435	struct address_space *mapping = NULL;
 436
 437	if (vma->vm_file)
 438		mapping = vma->vm_file->f_mapping;
 439
 440	if (mapping)
 441		mutex_lock(&mapping->i_mmap_mutex);
 442
 443	__vma_link(mm, vma, prev, rb_link, rb_parent);
 444	__vma_link_file(vma);
 445
 446	if (mapping)
 447		mutex_unlock(&mapping->i_mmap_mutex);
 448
 449	mm->map_count++;
 450	validate_mm(mm);
 451}
 452
 453/*
 454 * Helper for vma_adjust in the split_vma insert case:
 455 * insert vm structure into list and rbtree and anon_vma,
 456 * but it has already been inserted into prio_tree earlier.
 457 */
 458static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
 459{
 460	struct vm_area_struct *__vma, *prev;
 461	struct rb_node **rb_link, *rb_parent;
 462
 463	__vma = find_vma_prepare(mm, vma->vm_start,&prev, &rb_link, &rb_parent);
 464	BUG_ON(__vma && __vma->vm_start < vma->vm_end);
 465	__vma_link(mm, vma, prev, rb_link, rb_parent);
 466	mm->map_count++;
 467}
 468
 469static inline void
 470__vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma,
 471		struct vm_area_struct *prev)
 472{
 473	struct vm_area_struct *next = vma->vm_next;
 474
 475	prev->vm_next = next;
 476	if (next)
 477		next->vm_prev = prev;
 478	rb_erase(&vma->vm_rb, &mm->mm_rb);
 479	if (mm->mmap_cache == vma)
 480		mm->mmap_cache = prev;
 481}
 482
 483/*
 484 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
 485 * is already present in an i_mmap tree without adjusting the tree.
 486 * The following helper function should be used when such adjustments
 487 * are necessary.  The "insert" vma (if any) is to be inserted
 488 * before we drop the necessary locks.
 489 */
 490int vma_adjust(struct vm_area_struct *vma, unsigned long start,
 491	unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert)
 492{
 493	struct mm_struct *mm = vma->vm_mm;
 494	struct vm_area_struct *next = vma->vm_next;
 495	struct vm_area_struct *importer = NULL;
 496	struct address_space *mapping = NULL;
 497	struct prio_tree_root *root = NULL;
 498	struct anon_vma *anon_vma = NULL;
 499	struct file *file = vma->vm_file;
 500	long adjust_next = 0;
 501	int remove_next = 0;
 502
 503	if (next && !insert) {
 504		struct vm_area_struct *exporter = NULL;
 505
 506		if (end >= next->vm_end) {
 507			/*
 508			 * vma expands, overlapping all the next, and
 509			 * perhaps the one after too (mprotect case 6).
 510			 */
 511again:			remove_next = 1 + (end > next->vm_end);
 512			end = next->vm_end;
 513			exporter = next;
 514			importer = vma;
 515		} else if (end > next->vm_start) {
 516			/*
 517			 * vma expands, overlapping part of the next:
 518			 * mprotect case 5 shifting the boundary up.
 519			 */
 520			adjust_next = (end - next->vm_start) >> PAGE_SHIFT;
 521			exporter = next;
 522			importer = vma;
 523		} else if (end < vma->vm_end) {
 524			/*
 525			 * vma shrinks, and !insert tells it's not
 526			 * split_vma inserting another: so it must be
 527			 * mprotect case 4 shifting the boundary down.
 528			 */
 529			adjust_next = - ((vma->vm_end - end) >> PAGE_SHIFT);
 530			exporter = vma;
 531			importer = next;
 532		}
 533
 534		/*
 535		 * Easily overlooked: when mprotect shifts the boundary,
 536		 * make sure the expanding vma has anon_vma set if the
 537		 * shrinking vma had, to cover any anon pages imported.
 538		 */
 539		if (exporter && exporter->anon_vma && !importer->anon_vma) {
 540			if (anon_vma_clone(importer, exporter))
 541				return -ENOMEM;
 542			importer->anon_vma = exporter->anon_vma;
 543		}
 544	}
 545
 546	if (file) {
 547		mapping = file->f_mapping;
 548		if (!(vma->vm_flags & VM_NONLINEAR))
 549			root = &mapping->i_mmap;
 550		mutex_lock(&mapping->i_mmap_mutex);
 551		if (insert) {
 552			/*
 553			 * Put into prio_tree now, so instantiated pages
 554			 * are visible to arm/parisc __flush_dcache_page
 555			 * throughout; but we cannot insert into address
 556			 * space until vma start or end is updated.
 557			 */
 558			__vma_link_file(insert);
 559		}
 560	}
 561
 562	vma_adjust_trans_huge(vma, start, end, adjust_next);
 
 563
 564	/*
 565	 * When changing only vma->vm_end, we don't really need anon_vma
 566	 * lock. This is a fairly rare case by itself, but the anon_vma
 567	 * lock may be shared between many sibling processes.  Skipping
 568	 * the lock for brk adjustments makes a difference sometimes.
 569	 */
 570	if (vma->anon_vma && (importer || start != vma->vm_start)) {
 571		anon_vma = vma->anon_vma;
 572		anon_vma_lock(anon_vma);
 573	}
 574
 575	if (root) {
 576		flush_dcache_mmap_lock(mapping);
 577		vma_prio_tree_remove(vma, root);
 578		if (adjust_next)
 579			vma_prio_tree_remove(next, root);
 580	}
 581
 582	vma->vm_start = start;
 583	vma->vm_end = end;
 584	vma->vm_pgoff = pgoff;
 585	if (adjust_next) {
 586		next->vm_start += adjust_next << PAGE_SHIFT;
 587		next->vm_pgoff += adjust_next;
 588	}
 589
 590	if (root) {
 591		if (adjust_next)
 592			vma_prio_tree_insert(next, root);
 593		vma_prio_tree_insert(vma, root);
 594		flush_dcache_mmap_unlock(mapping);
 595	}
 596
 597	if (remove_next) {
 598		/*
 599		 * vma_merge has merged next into vma, and needs
 600		 * us to remove next before dropping the locks.
 601		 */
 602		__vma_unlink(mm, next, vma);
 603		if (file)
 604			__remove_shared_vm_struct(next, file, mapping);
 605	} else if (insert) {
 606		/*
 607		 * split_vma has split insert from vma, and needs
 608		 * us to insert it before dropping the locks
 609		 * (it may either follow vma or precede it).
 610		 */
 611		__insert_vm_struct(mm, insert);
 612	}
 613
 614	if (anon_vma)
 615		anon_vma_unlock(anon_vma);
 616	if (mapping)
 617		mutex_unlock(&mapping->i_mmap_mutex);
 618
 619	if (remove_next) {
 620		if (file) {
 621			fput(file);
 622			if (next->vm_flags & VM_EXECUTABLE)
 623				removed_exe_file_vma(mm);
 624		}
 625		if (next->anon_vma)
 626			anon_vma_merge(vma, next);
 627		mm->map_count--;
 628		mpol_put(vma_policy(next));
 629		kmem_cache_free(vm_area_cachep, next);
 630		/*
 631		 * In mprotect's case 6 (see comments on vma_merge),
 632		 * we must remove another next too. It would clutter
 633		 * up the code too much to do both in one go.
 634		 */
 635		if (remove_next == 2) {
 636			next = vma->vm_next;
 637			goto again;
 638		}
 639	}
 640
 641	validate_mm(mm);
 
 
 
 
 
 
 642
 643	return 0;
 
 
 
 644}
 645
 646/*
 647 * If the vma has a ->close operation then the driver probably needs to release
 648 * per-vma resources, so we don't attempt to merge those.
 649 */
 650static inline int is_mergeable_vma(struct vm_area_struct *vma,
 651			struct file *file, unsigned long vm_flags)
 652{
 653	/* VM_CAN_NONLINEAR may get set later by f_op->mmap() */
 654	if ((vma->vm_flags ^ vm_flags) & ~VM_CAN_NONLINEAR)
 655		return 0;
 656	if (vma->vm_file != file)
 657		return 0;
 658	if (vma->vm_ops && vma->vm_ops->close)
 659		return 0;
 660	return 1;
 661}
 662
 663static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
 664					struct anon_vma *anon_vma2,
 665					struct vm_area_struct *vma)
 666{
 667	/*
 668	 * The list_is_singular() test is to avoid merging VMA cloned from
 669	 * parents. This can improve scalability caused by anon_vma lock.
 670	 */
 671	if ((!anon_vma1 || !anon_vma2) && (!vma ||
 672		list_is_singular(&vma->anon_vma_chain)))
 673		return 1;
 674	return anon_vma1 == anon_vma2;
 675}
 676
 677/*
 678 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
 679 * in front of (at a lower virtual address and file offset than) the vma.
 680 *
 681 * We cannot merge two vmas if they have differently assigned (non-NULL)
 682 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
 683 *
 684 * We don't check here for the merged mmap wrapping around the end of pagecache
 685 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
 686 * wrap, nor mmaps which cover the final page at index -1UL.
 687 */
 688static int
 689can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
 690	struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
 691{
 692	if (is_mergeable_vma(vma, file, vm_flags) &&
 693	    is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
 694		if (vma->vm_pgoff == vm_pgoff)
 695			return 1;
 696	}
 697	return 0;
 698}
 699
 700/*
 701 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
 702 * beyond (at a higher virtual address and file offset than) the vma.
 703 *
 704 * We cannot merge two vmas if they have differently assigned (non-NULL)
 705 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
 706 */
 707static int
 708can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
 709	struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
 710{
 711	if (is_mergeable_vma(vma, file, vm_flags) &&
 712	    is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
 713		pgoff_t vm_pglen;
 714		vm_pglen = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
 715		if (vma->vm_pgoff + vm_pglen == vm_pgoff)
 716			return 1;
 717	}
 718	return 0;
 719}
 720
 721/*
 722 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
 723 * whether that can be merged with its predecessor or its successor.
 724 * Or both (it neatly fills a hole).
 725 *
 726 * In most cases - when called for mmap, brk or mremap - [addr,end) is
 727 * certain not to be mapped by the time vma_merge is called; but when
 728 * called for mprotect, it is certain to be already mapped (either at
 729 * an offset within prev, or at the start of next), and the flags of
 730 * this area are about to be changed to vm_flags - and the no-change
 731 * case has already been eliminated.
 732 *
 733 * The following mprotect cases have to be considered, where AAAA is
 734 * the area passed down from mprotect_fixup, never extending beyond one
 735 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
 736 *
 737 *     AAAA             AAAA                AAAA          AAAA
 738 *    PPPPPPNNNNNN    PPPPPPNNNNNN    PPPPPPNNNNNN    PPPPNNNNXXXX
 739 *    cannot merge    might become    might become    might become
 740 *                    PPNNNNNNNNNN    PPPPPPPPPPNN    PPPPPPPPPPPP 6 or
 741 *    mmap, brk or    case 4 below    case 5 below    PPPPPPPPXXXX 7 or
 742 *    mremap move:                                    PPPPNNNNNNNN 8
 743 *        AAAA
 744 *    PPPP    NNNN    PPPPPPPPPPPP    PPPPPPPPNNNN    PPPPNNNNNNNN
 745 *    might become    case 1 below    case 2 below    case 3 below
 746 *
 747 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
 748 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
 749 */
 750struct vm_area_struct *vma_merge(struct mm_struct *mm,
 751			struct vm_area_struct *prev, unsigned long addr,
 752			unsigned long end, unsigned long vm_flags,
 753		     	struct anon_vma *anon_vma, struct file *file,
 754			pgoff_t pgoff, struct mempolicy *policy)
 755{
 756	pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
 757	struct vm_area_struct *area, *next;
 758	int err;
 759
 760	/*
 761	 * We later require that vma->vm_flags == vm_flags,
 762	 * so this tests vma->vm_flags & VM_SPECIAL, too.
 763	 */
 764	if (vm_flags & VM_SPECIAL)
 765		return NULL;
 766
 767	if (prev)
 768		next = prev->vm_next;
 769	else
 770		next = mm->mmap;
 771	area = next;
 772	if (next && next->vm_end == end)		/* cases 6, 7, 8 */
 773		next = next->vm_next;
 774
 775	/*
 776	 * Can it merge with the predecessor?
 777	 */
 778	if (prev && prev->vm_end == addr &&
 779  			mpol_equal(vma_policy(prev), policy) &&
 780			can_vma_merge_after(prev, vm_flags,
 781						anon_vma, file, pgoff)) {
 782		/*
 783		 * OK, it can.  Can we now merge in the successor as well?
 784		 */
 785		if (next && end == next->vm_start &&
 786				mpol_equal(policy, vma_policy(next)) &&
 787				can_vma_merge_before(next, vm_flags,
 788					anon_vma, file, pgoff+pglen) &&
 789				is_mergeable_anon_vma(prev->anon_vma,
 790						      next->anon_vma, NULL)) {
 791							/* cases 1, 6 */
 792			err = vma_adjust(prev, prev->vm_start,
 793				next->vm_end, prev->vm_pgoff, NULL);
 794		} else					/* cases 2, 5, 7 */
 795			err = vma_adjust(prev, prev->vm_start,
 796				end, prev->vm_pgoff, NULL);
 797		if (err)
 798			return NULL;
 799		khugepaged_enter_vma_merge(prev);
 800		return prev;
 801	}
 802
 803	/*
 804	 * Can this new request be merged in front of next?
 805	 */
 806	if (next && end == next->vm_start &&
 807 			mpol_equal(policy, vma_policy(next)) &&
 808			can_vma_merge_before(next, vm_flags,
 809					anon_vma, file, pgoff+pglen)) {
 810		if (prev && addr < prev->vm_end)	/* case 4 */
 811			err = vma_adjust(prev, prev->vm_start,
 812				addr, prev->vm_pgoff, NULL);
 813		else					/* cases 3, 8 */
 814			err = vma_adjust(area, addr, next->vm_end,
 815				next->vm_pgoff - pglen, NULL);
 816		if (err)
 817			return NULL;
 818		khugepaged_enter_vma_merge(area);
 819		return area;
 820	}
 821
 822	return NULL;
 823}
 824
 825/*
 826 * Rough compatbility check to quickly see if it's even worth looking
 827 * at sharing an anon_vma.
 828 *
 829 * They need to have the same vm_file, and the flags can only differ
 830 * in things that mprotect may change.
 831 *
 832 * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that
 833 * we can merge the two vma's. For example, we refuse to merge a vma if
 834 * there is a vm_ops->close() function, because that indicates that the
 835 * driver is doing some kind of reference counting. But that doesn't
 836 * really matter for the anon_vma sharing case.
 837 */
 838static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b)
 839{
 840	return a->vm_end == b->vm_start &&
 841		mpol_equal(vma_policy(a), vma_policy(b)) &&
 842		a->vm_file == b->vm_file &&
 843		!((a->vm_flags ^ b->vm_flags) & ~(VM_READ|VM_WRITE|VM_EXEC)) &&
 844		b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT);
 845}
 846
 847/*
 848 * Do some basic sanity checking to see if we can re-use the anon_vma
 849 * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be
 850 * the same as 'old', the other will be the new one that is trying
 851 * to share the anon_vma.
 852 *
 853 * NOTE! This runs with mm_sem held for reading, so it is possible that
 854 * the anon_vma of 'old' is concurrently in the process of being set up
 855 * by another page fault trying to merge _that_. But that's ok: if it
 856 * is being set up, that automatically means that it will be a singleton
 857 * acceptable for merging, so we can do all of this optimistically. But
 858 * we do that ACCESS_ONCE() to make sure that we never re-load the pointer.
 859 *
 860 * IOW: that the "list_is_singular()" test on the anon_vma_chain only
 861 * matters for the 'stable anon_vma' case (ie the thing we want to avoid
 862 * is to return an anon_vma that is "complex" due to having gone through
 863 * a fork).
 864 *
 865 * We also make sure that the two vma's are compatible (adjacent,
 866 * and with the same memory policies). That's all stable, even with just
 867 * a read lock on the mm_sem.
 868 */
 869static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b)
 870{
 871	if (anon_vma_compatible(a, b)) {
 872		struct anon_vma *anon_vma = ACCESS_ONCE(old->anon_vma);
 873
 874		if (anon_vma && list_is_singular(&old->anon_vma_chain))
 875			return anon_vma;
 876	}
 877	return NULL;
 878}
 879
 880/*
 881 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
 882 * neighbouring vmas for a suitable anon_vma, before it goes off
 883 * to allocate a new anon_vma.  It checks because a repetitive
 884 * sequence of mprotects and faults may otherwise lead to distinct
 885 * anon_vmas being allocated, preventing vma merge in subsequent
 886 * mprotect.
 887 */
 888struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
 889{
 890	struct anon_vma *anon_vma;
 891	struct vm_area_struct *near;
 892
 893	near = vma->vm_next;
 894	if (!near)
 895		goto try_prev;
 896
 897	anon_vma = reusable_anon_vma(near, vma, near);
 898	if (anon_vma)
 899		return anon_vma;
 900try_prev:
 901	near = vma->vm_prev;
 902	if (!near)
 903		goto none;
 904
 905	anon_vma = reusable_anon_vma(near, near, vma);
 906	if (anon_vma)
 907		return anon_vma;
 908none:
 909	/*
 910	 * There's no absolute need to look only at touching neighbours:
 911	 * we could search further afield for "compatible" anon_vmas.
 912	 * But it would probably just be a waste of time searching,
 913	 * or lead to too many vmas hanging off the same anon_vma.
 914	 * We're trying to allow mprotect remerging later on,
 915	 * not trying to minimize memory used for anon_vmas.
 916	 */
 917	return NULL;
 918}
 919
 920#ifdef CONFIG_PROC_FS
 921void vm_stat_account(struct mm_struct *mm, unsigned long flags,
 922						struct file *file, long pages)
 923{
 924	const unsigned long stack_flags
 925		= VM_STACK_FLAGS & (VM_GROWSUP|VM_GROWSDOWN);
 926
 927	if (file) {
 928		mm->shared_vm += pages;
 929		if ((flags & (VM_EXEC|VM_WRITE)) == VM_EXEC)
 930			mm->exec_vm += pages;
 931	} else if (flags & stack_flags)
 932		mm->stack_vm += pages;
 933	if (flags & (VM_RESERVED|VM_IO))
 934		mm->reserved_vm += pages;
 935}
 936#endif /* CONFIG_PROC_FS */
 937
 938/*
 939 * The caller must hold down_write(&current->mm->mmap_sem).
 940 */
 941
 942unsigned long do_mmap_pgoff(struct file *file, unsigned long addr,
 943			unsigned long len, unsigned long prot,
 944			unsigned long flags, unsigned long pgoff)
 
 
 945{
 946	struct mm_struct * mm = current->mm;
 947	struct inode *inode;
 948	vm_flags_t vm_flags;
 949	int error;
 950	unsigned long reqprot = prot;
 
 
 951
 952	/*
 953	 * Does the application expect PROT_READ to imply PROT_EXEC?
 954	 *
 955	 * (the exception is when the underlying filesystem is noexec
 956	 *  mounted, in which case we dont add PROT_EXEC.)
 957	 */
 958	if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
 959		if (!(file && (file->f_path.mnt->mnt_flags & MNT_NOEXEC)))
 960			prot |= PROT_EXEC;
 961
 962	if (!len)
 963		return -EINVAL;
 
 964
 965	if (!(flags & MAP_FIXED))
 966		addr = round_hint_to_min(addr);
 967
 968	/* Careful about overflows.. */
 969	len = PAGE_ALIGN(len);
 970	if (!len)
 971		return -ENOMEM;
 972
 973	/* offset overflow? */
 974	if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
 975               return -EOVERFLOW;
 976
 977	/* Too many mappings? */
 978	if (mm->map_count > sysctl_max_map_count)
 979		return -ENOMEM;
 980
 981	/* Obtain the address to map to. we verify (or select) it and ensure
 982	 * that it represents a valid section of the address space.
 
 
 
 
 
 
 983	 */
 984	addr = get_unmapped_area(file, addr, len, pgoff, flags);
 985	if (addr & ~PAGE_MASK)
 986		return addr;
 
 
 
 987
 988	/* Do simple checking here so the lower-level routines won't have
 989	 * to. we assume access permissions have been handled by the open
 990	 * of the memory object, so we don't do any here.
 991	 */
 992	vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) |
 993			mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
 994
 
 
 
 
 
 
 
 
 
 
 
 
 995	if (flags & MAP_LOCKED)
 996		if (!can_do_mlock())
 997			return -EPERM;
 998
 999	/* mlock MCL_FUTURE? */
1000	if (vm_flags & VM_LOCKED) {
1001		unsigned long locked, lock_limit;
1002		locked = len >> PAGE_SHIFT;
1003		locked += mm->locked_vm;
1004		lock_limit = rlimit(RLIMIT_MEMLOCK);
1005		lock_limit >>= PAGE_SHIFT;
1006		if (locked > lock_limit && !capable(CAP_IPC_LOCK))
1007			return -EAGAIN;
1008	}
1009
1010	inode = file ? file->f_path.dentry->d_inode : NULL;
1011
1012	if (file) {
 
 
 
 
 
 
 
 
 
 
 
1013		switch (flags & MAP_TYPE) {
1014		case MAP_SHARED:
1015			if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
1016				return -EACCES;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1017
1018			/*
1019			 * Make sure we don't allow writing to an append-only
1020			 * file..
1021			 */
1022			if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
1023				return -EACCES;
1024
1025			/*
1026			 * Make sure there are no mandatory locks on the file.
1027			 */
1028			if (locks_verify_locked(inode))
1029				return -EAGAIN;
1030
1031			vm_flags |= VM_SHARED | VM_MAYSHARE;
1032			if (!(file->f_mode & FMODE_WRITE))
1033				vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
1034
1035			/* fall through */
 
1036		case MAP_PRIVATE:
1037			if (!(file->f_mode & FMODE_READ))
1038				return -EACCES;
1039			if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
1040				if (vm_flags & VM_EXEC)
1041					return -EPERM;
1042				vm_flags &= ~VM_MAYEXEC;
1043			}
1044
1045			if (!file->f_op || !file->f_op->mmap)
1046				return -ENODEV;
 
 
1047			break;
1048
1049		default:
1050			return -EINVAL;
1051		}
1052	} else {
1053		switch (flags & MAP_TYPE) {
1054		case MAP_SHARED:
 
 
1055			/*
1056			 * Ignore pgoff.
1057			 */
1058			pgoff = 0;
1059			vm_flags |= VM_SHARED | VM_MAYSHARE;
1060			break;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1061		case MAP_PRIVATE:
1062			/*
1063			 * Set pgoff according to addr for anon_vma.
1064			 */
1065			pgoff = addr >> PAGE_SHIFT;
1066			break;
1067		default:
1068			return -EINVAL;
1069		}
1070	}
1071
1072	error = security_file_mmap(file, reqprot, prot, flags, addr, 0);
1073	if (error)
1074		return error;
 
 
 
 
 
1075
1076	return mmap_region(file, addr, len, flags, vm_flags, pgoff);
 
 
 
 
 
 
 
 
 
 
1077}
1078EXPORT_SYMBOL(do_mmap_pgoff);
1079
1080SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1081		unsigned long, prot, unsigned long, flags,
1082		unsigned long, fd, unsigned long, pgoff)
1083{
1084	struct file *file = NULL;
1085	unsigned long retval = -EBADF;
1086
1087	if (!(flags & MAP_ANONYMOUS)) {
1088		audit_mmap_fd(fd, flags);
1089		if (unlikely(flags & MAP_HUGETLB))
1090			return -EINVAL;
1091		file = fget(fd);
1092		if (!file)
1093			goto out;
 
 
 
 
 
 
1094	} else if (flags & MAP_HUGETLB) {
1095		struct user_struct *user = NULL;
 
 
 
 
 
 
1096		/*
1097		 * VM_NORESERVE is used because the reservations will be
1098		 * taken when vm_ops->mmap() is called
1099		 * A dummy user value is used because we are not locking
1100		 * memory so no accounting is necessary
1101		 */
1102		len = ALIGN(len, huge_page_size(&default_hstate));
1103		file = hugetlb_file_setup(HUGETLB_ANON_FILE, len, VM_NORESERVE,
1104						&user, HUGETLB_ANONHUGE_INODE);
 
1105		if (IS_ERR(file))
1106			return PTR_ERR(file);
1107	}
1108
1109	flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1110
1111	down_write(&current->mm->mmap_sem);
1112	retval = do_mmap_pgoff(file, addr, len, prot, flags, pgoff);
1113	up_write(&current->mm->mmap_sem);
1114
1115	if (file)
1116		fput(file);
1117out:
1118	return retval;
1119}
1120
 
 
 
 
 
 
 
1121#ifdef __ARCH_WANT_SYS_OLD_MMAP
1122struct mmap_arg_struct {
1123	unsigned long addr;
1124	unsigned long len;
1125	unsigned long prot;
1126	unsigned long flags;
1127	unsigned long fd;
1128	unsigned long offset;
1129};
1130
1131SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1132{
1133	struct mmap_arg_struct a;
1134
1135	if (copy_from_user(&a, arg, sizeof(a)))
1136		return -EFAULT;
1137	if (a.offset & ~PAGE_MASK)
1138		return -EINVAL;
1139
1140	return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1141			      a.offset >> PAGE_SHIFT);
1142}
1143#endif /* __ARCH_WANT_SYS_OLD_MMAP */
1144
1145/*
1146 * Some shared mappigns will want the pages marked read-only
1147 * to track write events. If so, we'll downgrade vm_page_prot
1148 * to the private version (using protection_map[] without the
1149 * VM_SHARED bit).
1150 */
1151int vma_wants_writenotify(struct vm_area_struct *vma)
1152{
1153	vm_flags_t vm_flags = vma->vm_flags;
1154
1155	/* If it was private or non-writable, the write bit is already clear */
1156	if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED)))
1157		return 0;
1158
1159	/* The backer wishes to know when pages are first written to? */
1160	if (vma->vm_ops && vma->vm_ops->page_mkwrite)
1161		return 1;
1162
1163	/* The open routine did something to the protections already? */
1164	if (pgprot_val(vma->vm_page_prot) !=
1165	    pgprot_val(vm_get_page_prot(vm_flags)))
1166		return 0;
1167
1168	/* Specialty mapping? */
1169	if (vm_flags & (VM_PFNMAP|VM_INSERTPAGE))
1170		return 0;
1171
1172	/* Can the mapping track the dirty pages? */
1173	return vma->vm_file && vma->vm_file->f_mapping &&
1174		mapping_cap_account_dirty(vma->vm_file->f_mapping);
1175}
1176
1177/*
1178 * We account for memory if it's a private writeable mapping,
1179 * not hugepages and VM_NORESERVE wasn't set.
1180 */
1181static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags)
1182{
1183	/*
1184	 * hugetlb has its own accounting separate from the core VM
1185	 * VM_HUGETLB may not be set yet so we cannot check for that flag.
1186	 */
1187	if (file && is_file_hugepages(file))
1188		return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1189
1190	return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE;
1191}
1192
1193unsigned long mmap_region(struct file *file, unsigned long addr,
1194			  unsigned long len, unsigned long flags,
1195			  vm_flags_t vm_flags, unsigned long pgoff)
 
 
 
 
 
 
 
 
1196{
1197	struct mm_struct *mm = current->mm;
1198	struct vm_area_struct *vma, *prev;
1199	int correct_wcount = 0;
1200	int error;
1201	struct rb_node **rb_link, *rb_parent;
1202	unsigned long charged = 0;
1203	struct inode *inode =  file ? file->f_path.dentry->d_inode : NULL;
1204
1205	/* Clear old maps */
1206	error = -ENOMEM;
1207munmap_back:
1208	vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
1209	if (vma && vma->vm_start < addr + len) {
1210		if (do_munmap(mm, addr, len))
1211			return -ENOMEM;
1212		goto munmap_back;
1213	}
1214
1215	/* Check against address space limit. */
1216	if (!may_expand_vm(mm, len >> PAGE_SHIFT))
 
1217		return -ENOMEM;
1218
1219	/*
1220	 * Set 'VM_NORESERVE' if we should not account for the
1221	 * memory use of this mapping.
1222	 */
1223	if ((flags & MAP_NORESERVE)) {
1224		/* We honor MAP_NORESERVE if allowed to overcommit */
1225		if (sysctl_overcommit_memory != OVERCOMMIT_NEVER)
1226			vm_flags |= VM_NORESERVE;
1227
1228		/* hugetlb applies strict overcommit unless MAP_NORESERVE */
1229		if (file && is_file_hugepages(file))
1230			vm_flags |= VM_NORESERVE;
1231	}
1232
1233	/*
1234	 * Private writable mapping: check memory availability
1235	 */
1236	if (accountable_mapping(file, vm_flags)) {
1237		charged = len >> PAGE_SHIFT;
1238		if (security_vm_enough_memory(charged))
1239			return -ENOMEM;
1240		vm_flags |= VM_ACCOUNT;
1241	}
1242
1243	/*
1244	 * Can we just expand an old mapping?
1245	 */
1246	vma = vma_merge(mm, prev, addr, addr + len, vm_flags, NULL, file, pgoff, NULL);
1247	if (vma)
1248		goto out;
1249
1250	/*
1251	 * Determine the object being mapped and call the appropriate
1252	 * specific mapper. the address has already been validated, but
1253	 * not unmapped, but the maps are removed from the list.
1254	 */
1255	vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1256	if (!vma) {
1257		error = -ENOMEM;
1258		goto unacct_error;
1259	}
1260
1261	vma->vm_mm = mm;
1262	vma->vm_start = addr;
1263	vma->vm_end = addr + len;
1264	vma->vm_flags = vm_flags;
1265	vma->vm_page_prot = vm_get_page_prot(vm_flags);
1266	vma->vm_pgoff = pgoff;
1267	INIT_LIST_HEAD(&vma->anon_vma_chain);
1268
1269	if (file) {
1270		error = -EINVAL;
1271		if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1272			goto free_vma;
1273		if (vm_flags & VM_DENYWRITE) {
1274			error = deny_write_access(file);
1275			if (error)
1276				goto free_vma;
1277			correct_wcount = 1;
 
 
 
 
 
 
 
1278		}
1279		vma->vm_file = file;
1280		get_file(file);
1281		error = file->f_op->mmap(file, vma);
1282		if (error)
1283			goto unmap_and_free_vma;
1284		if (vm_flags & VM_EXECUTABLE)
1285			added_exe_file_vma(mm);
1286
1287		/* Can addr have changed??
1288		 *
1289		 * Answer: Yes, several device drivers can do it in their
1290		 *         f_op->mmap method. -DaveM
1291		 */
1292		addr = vma->vm_start;
1293		pgoff = vma->vm_pgoff;
1294		vm_flags = vma->vm_flags;
1295	} else if (vm_flags & VM_SHARED) {
1296		error = shmem_zero_setup(vma);
1297		if (error)
1298			goto free_vma;
1299	}
1300
1301	if (vma_wants_writenotify(vma)) {
1302		pgprot_t pprot = vma->vm_page_prot;
1303
1304		/* Can vma->vm_page_prot have changed??
1305		 *
1306		 * Answer: Yes, drivers may have changed it in their
1307		 *         f_op->mmap method.
1308		 *
1309		 * Ensures that vmas marked as uncached stay that way.
1310		 */
1311		vma->vm_page_prot = vm_get_page_prot(vm_flags & ~VM_SHARED);
1312		if (pgprot_val(pprot) == pgprot_val(pgprot_noncached(pprot)))
1313			vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1314	}
1315
1316	vma_link(mm, vma, prev, rb_link, rb_parent);
1317	file = vma->vm_file;
 
 
 
 
 
 
1318
1319	/* Once vma denies write, undo our temporary denial count */
1320	if (correct_wcount)
1321		atomic_inc(&inode->i_writecount);
1322out:
1323	perf_event_mmap(vma);
1324
1325	mm->total_vm += len >> PAGE_SHIFT;
1326	vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT);
1327	if (vm_flags & VM_LOCKED) {
1328		if (!mlock_vma_pages_range(vma, addr, addr + len))
1329			mm->locked_vm += (len >> PAGE_SHIFT);
1330	} else if ((flags & MAP_POPULATE) && !(flags & MAP_NONBLOCK))
1331		make_pages_present(addr, addr + len);
1332	return addr;
 
 
 
 
1333
1334unmap_and_free_vma:
1335	if (correct_wcount)
1336		atomic_inc(&inode->i_writecount);
1337	vma->vm_file = NULL;
1338	fput(file);
1339
1340	/* Undo any partial mapping done by a device driver. */
1341	unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
1342	charged = 0;
1343free_vma:
1344	kmem_cache_free(vm_area_cachep, vma);
1345unacct_error:
1346	if (charged)
1347		vm_unacct_memory(charged);
1348	return error;
1349}
1350
1351/* Get an address range which is currently unmapped.
1352 * For shmat() with addr=0.
1353 *
1354 * Ugly calling convention alert:
1355 * Return value with the low bits set means error value,
1356 * ie
1357 *	if (ret & ~PAGE_MASK)
1358 *		error = ret;
1359 *
1360 * This function "knows" that -ENOMEM has the bits set.
1361 */
1362#ifndef HAVE_ARCH_UNMAPPED_AREA
1363unsigned long
1364arch_get_unmapped_area(struct file *filp, unsigned long addr,
1365		unsigned long len, unsigned long pgoff, unsigned long flags)
 
1366{
1367	struct mm_struct *mm = current->mm;
1368	struct vm_area_struct *vma;
1369	unsigned long start_addr;
 
1370
1371	if (len > TASK_SIZE)
1372		return -ENOMEM;
1373
1374	if (flags & MAP_FIXED)
1375		return addr;
1376
1377	if (addr) {
1378		addr = PAGE_ALIGN(addr);
1379		vma = find_vma(mm, addr);
1380		if (TASK_SIZE - len >= addr &&
1381		    (!vma || addr + len <= vma->vm_start))
 
1382			return addr;
1383	}
1384	if (len > mm->cached_hole_size) {
1385	        start_addr = addr = mm->free_area_cache;
1386	} else {
1387	        start_addr = addr = TASK_UNMAPPED_BASE;
1388	        mm->cached_hole_size = 0;
1389	}
1390
1391full_search:
1392	for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
1393		/* At this point:  (!vma || addr < vma->vm_end). */
1394		if (TASK_SIZE - len < addr) {
1395			/*
1396			 * Start a new search - just in case we missed
1397			 * some holes.
1398			 */
1399			if (start_addr != TASK_UNMAPPED_BASE) {
1400				addr = TASK_UNMAPPED_BASE;
1401			        start_addr = addr;
1402				mm->cached_hole_size = 0;
1403				goto full_search;
1404			}
1405			return -ENOMEM;
1406		}
1407		if (!vma || addr + len <= vma->vm_start) {
1408			/*
1409			 * Remember the place where we stopped the search:
1410			 */
1411			mm->free_area_cache = addr + len;
1412			return addr;
1413		}
1414		if (addr + mm->cached_hole_size < vma->vm_start)
1415		        mm->cached_hole_size = vma->vm_start - addr;
1416		addr = vma->vm_end;
1417	}
1418}
1419#endif	
1420
1421void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
 
 
 
 
1422{
1423	/*
1424	 * Is this a new hole at the lowest possible address?
1425	 */
1426	if (addr >= TASK_UNMAPPED_BASE && addr < mm->free_area_cache) {
1427		mm->free_area_cache = addr;
1428		mm->cached_hole_size = ~0UL;
1429	}
1430}
 
1431
1432/*
1433 * This mmap-allocator allocates new areas top-down from below the
1434 * stack's low limit (the base):
1435 */
1436#ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1437unsigned long
1438arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
1439			  const unsigned long len, const unsigned long pgoff,
1440			  const unsigned long flags)
1441{
1442	struct vm_area_struct *vma;
1443	struct mm_struct *mm = current->mm;
1444	unsigned long addr = addr0;
 
1445
1446	/* requested length too big for entire address space */
1447	if (len > TASK_SIZE)
1448		return -ENOMEM;
1449
1450	if (flags & MAP_FIXED)
1451		return addr;
1452
1453	/* requesting a specific address */
1454	if (addr) {
1455		addr = PAGE_ALIGN(addr);
1456		vma = find_vma(mm, addr);
1457		if (TASK_SIZE - len >= addr &&
1458				(!vma || addr + len <= vma->vm_start))
 
1459			return addr;
1460	}
1461
1462	/* check if free_area_cache is useful for us */
1463	if (len <= mm->cached_hole_size) {
1464 	        mm->cached_hole_size = 0;
1465 		mm->free_area_cache = mm->mmap_base;
1466 	}
1467
1468	/* either no address requested or can't fit in requested address hole */
1469	addr = mm->free_area_cache;
1470
1471	/* make sure it can fit in the remaining address space */
1472	if (addr > len) {
1473		vma = find_vma(mm, addr-len);
1474		if (!vma || addr <= vma->vm_start)
1475			/* remember the address as a hint for next time */
1476			return (mm->free_area_cache = addr-len);
1477	}
1478
1479	if (mm->mmap_base < len)
1480		goto bottomup;
1481
1482	addr = mm->mmap_base-len;
1483
1484	do {
1485		/*
1486		 * Lookup failure means no vma is above this address,
1487		 * else if new region fits below vma->vm_start,
1488		 * return with success:
1489		 */
1490		vma = find_vma(mm, addr);
1491		if (!vma || addr+len <= vma->vm_start)
1492			/* remember the address as a hint for next time */
1493			return (mm->free_area_cache = addr);
1494
1495 		/* remember the largest hole we saw so far */
1496 		if (addr + mm->cached_hole_size < vma->vm_start)
1497 		        mm->cached_hole_size = vma->vm_start - addr;
1498
1499		/* try just below the current vma->vm_start */
1500		addr = vma->vm_start-len;
1501	} while (len < vma->vm_start);
1502
1503bottomup:
1504	/*
1505	 * A failed mmap() very likely causes application failure,
1506	 * so fall back to the bottom-up function here. This scenario
1507	 * can happen with large stack limits and large mmap()
1508	 * allocations.
1509	 */
1510	mm->cached_hole_size = ~0UL;
1511  	mm->free_area_cache = TASK_UNMAPPED_BASE;
1512	addr = arch_get_unmapped_area(filp, addr0, len, pgoff, flags);
1513	/*
1514	 * Restore the topdown base:
1515	 */
1516	mm->free_area_cache = mm->mmap_base;
1517	mm->cached_hole_size = ~0UL;
1518
1519	return addr;
1520}
1521#endif
1522
1523void arch_unmap_area_topdown(struct mm_struct *mm, unsigned long addr)
 
 
 
 
1524{
1525	/*
1526	 * Is this a new hole at the highest possible address?
1527	 */
1528	if (addr > mm->free_area_cache)
1529		mm->free_area_cache = addr;
1530
1531	/* dont allow allocations above current base */
1532	if (mm->free_area_cache > mm->mmap_base)
1533		mm->free_area_cache = mm->mmap_base;
 
 
 
 
 
 
1534}
1535
1536unsigned long
1537get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
1538		unsigned long pgoff, unsigned long flags)
1539{
1540	unsigned long (*get_area)(struct file *, unsigned long,
1541				  unsigned long, unsigned long, unsigned long);
 
1542
1543	unsigned long error = arch_mmap_check(addr, len, flags);
1544	if (error)
1545		return error;
1546
1547	/* Careful about overflows.. */
1548	if (len > TASK_SIZE)
1549		return -ENOMEM;
1550
1551	get_area = current->mm->get_unmapped_area;
1552	if (file && file->f_op && file->f_op->get_unmapped_area)
1553		get_area = file->f_op->get_unmapped_area;
1554	addr = get_area(file, addr, len, pgoff, flags);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1555	if (IS_ERR_VALUE(addr))
1556		return addr;
1557
1558	if (addr > TASK_SIZE - len)
1559		return -ENOMEM;
1560	if (addr & ~PAGE_MASK)
1561		return -EINVAL;
1562
1563	return arch_rebalance_pgtables(addr, len);
 
1564}
1565
1566EXPORT_SYMBOL(get_unmapped_area);
 
 
 
 
 
 
 
 
 
1567
1568/* Look up the first VMA which satisfies  addr < vm_end,  NULL if none. */
1569struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
 
 
 
 
 
 
 
 
 
 
1570{
1571	struct vm_area_struct *vma = NULL;
1572
1573	if (mm) {
1574		/* Check the cache first. */
1575		/* (Cache hit rate is typically around 35%.) */
1576		vma = mm->mmap_cache;
1577		if (!(vma && vma->vm_end > addr && vma->vm_start <= addr)) {
1578			struct rb_node * rb_node;
1579
1580			rb_node = mm->mm_rb.rb_node;
1581			vma = NULL;
1582
1583			while (rb_node) {
1584				struct vm_area_struct * vma_tmp;
1585
1586				vma_tmp = rb_entry(rb_node,
1587						struct vm_area_struct, vm_rb);
1588
1589				if (vma_tmp->vm_end > addr) {
1590					vma = vma_tmp;
1591					if (vma_tmp->vm_start <= addr)
1592						break;
1593					rb_node = rb_node->rb_left;
1594				} else
1595					rb_node = rb_node->rb_right;
1596			}
1597			if (vma)
1598				mm->mmap_cache = vma;
1599		}
1600	}
1601	return vma;
1602}
 
 
 
 
 
 
 
 
 
 
 
 
 
1603
 
 
 
1604EXPORT_SYMBOL(find_vma);
1605
1606/* Same as find_vma, but also return a pointer to the previous VMA in *pprev. */
 
 
 
 
 
 
 
 
 
 
 
 
1607struct vm_area_struct *
1608find_vma_prev(struct mm_struct *mm, unsigned long addr,
1609			struct vm_area_struct **pprev)
1610{
1611	struct vm_area_struct *vma = NULL, *prev = NULL;
1612	struct rb_node *rb_node;
1613	if (!mm)
1614		goto out;
1615
1616	/* Guard against addr being lower than the first VMA */
1617	vma = mm->mmap;
1618
1619	/* Go through the RB tree quickly. */
1620	rb_node = mm->mm_rb.rb_node;
1621
1622	while (rb_node) {
1623		struct vm_area_struct *vma_tmp;
1624		vma_tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb);
1625
1626		if (addr < vma_tmp->vm_end) {
1627			rb_node = rb_node->rb_left;
1628		} else {
1629			prev = vma_tmp;
1630			if (!prev->vm_next || (addr < prev->vm_next->vm_end))
1631				break;
1632			rb_node = rb_node->rb_right;
1633		}
1634	}
1635
1636out:
1637	*pprev = prev;
1638	return prev ? prev->vm_next : vma;
 
 
1639}
1640
1641/*
1642 * Verify that the stack growth is acceptable and
1643 * update accounting. This is shared with both the
1644 * grow-up and grow-down cases.
1645 */
1646static int acct_stack_growth(struct vm_area_struct *vma, unsigned long size, unsigned long grow)
 
1647{
1648	struct mm_struct *mm = vma->vm_mm;
1649	struct rlimit *rlim = current->signal->rlim;
1650	unsigned long new_start;
1651
1652	/* address space limit tests */
1653	if (!may_expand_vm(mm, grow))
1654		return -ENOMEM;
1655
1656	/* Stack limit test */
1657	if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur))
1658		return -ENOMEM;
1659
1660	/* mlock limit tests */
1661	if (vma->vm_flags & VM_LOCKED) {
1662		unsigned long locked;
1663		unsigned long limit;
1664		locked = mm->locked_vm + grow;
1665		limit = ACCESS_ONCE(rlim[RLIMIT_MEMLOCK].rlim_cur);
1666		limit >>= PAGE_SHIFT;
1667		if (locked > limit && !capable(CAP_IPC_LOCK))
1668			return -ENOMEM;
1669	}
1670
1671	/* Check to ensure the stack will not grow into a hugetlb-only region */
1672	new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start :
1673			vma->vm_end - size;
1674	if (is_hugepage_only_range(vma->vm_mm, new_start, size))
1675		return -EFAULT;
1676
1677	/*
1678	 * Overcommit..  This must be the final test, as it will
1679	 * update security statistics.
1680	 */
1681	if (security_vm_enough_memory_mm(mm, grow))
1682		return -ENOMEM;
1683
1684	/* Ok, everything looks good - let it rip */
1685	mm->total_vm += grow;
1686	if (vma->vm_flags & VM_LOCKED)
1687		mm->locked_vm += grow;
1688	vm_stat_account(mm, vma->vm_flags, vma->vm_file, grow);
1689	return 0;
1690}
1691
1692#if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
1693/*
1694 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
1695 * vma is the last one with address > vma->vm_end.  Have to extend vma.
1696 */
1697int expand_upwards(struct vm_area_struct *vma, unsigned long address)
1698{
1699	int error;
 
 
 
 
1700
1701	if (!(vma->vm_flags & VM_GROWSUP))
1702		return -EFAULT;
1703
1704	/*
1705	 * We must make sure the anon_vma is allocated
1706	 * so that the anon_vma locking is not a noop.
1707	 */
1708	if (unlikely(anon_vma_prepare(vma)))
1709		return -ENOMEM;
1710	vma_lock_anon_vma(vma);
1711
1712	/*
1713	 * vma->vm_start/vm_end cannot change under us because the caller
1714	 * is required to hold the mmap_sem in read mode.  We need the
1715	 * anon_vma lock to serialize against concurrent expand_stacks.
1716	 * Also guard against wrapping around to address 0.
1717	 */
1718	if (address < PAGE_ALIGN(address+4))
1719		address = PAGE_ALIGN(address+4);
1720	else {
1721		vma_unlock_anon_vma(vma);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1722		return -ENOMEM;
1723	}
1724	error = 0;
 
 
 
 
1725
1726	/* Somebody else might have raced and expanded it already */
1727	if (address > vma->vm_end) {
1728		unsigned long size, grow;
1729
1730		size = address - vma->vm_start;
1731		grow = (address - vma->vm_end) >> PAGE_SHIFT;
1732
1733		error = -ENOMEM;
1734		if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) {
1735			error = acct_stack_growth(vma, size, grow);
1736			if (!error) {
 
 
 
 
1737				vma->vm_end = address;
 
 
 
 
1738				perf_event_mmap(vma);
1739			}
1740		}
1741	}
1742	vma_unlock_anon_vma(vma);
1743	khugepaged_enter_vma_merge(vma);
 
1744	return error;
1745}
1746#endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
1747
1748/*
1749 * vma is the first one with address < vma->vm_start.  Have to extend vma.
 
1750 */
1751int expand_downwards(struct vm_area_struct *vma,
1752				   unsigned long address)
1753{
1754	int error;
 
 
 
1755
1756	/*
1757	 * We must make sure the anon_vma is allocated
1758	 * so that the anon_vma locking is not a noop.
1759	 */
1760	if (unlikely(anon_vma_prepare(vma)))
1761		return -ENOMEM;
1762
1763	address &= PAGE_MASK;
1764	error = security_file_mmap(NULL, 0, 0, 0, address, 1);
1765	if (error)
1766		return error;
1767
1768	vma_lock_anon_vma(vma);
 
 
 
 
 
 
 
 
1769
1770	/*
1771	 * vma->vm_start/vm_end cannot change under us because the caller
1772	 * is required to hold the mmap_sem in read mode.  We need the
1773	 * anon_vma lock to serialize against concurrent expand_stacks.
1774	 */
 
 
 
 
 
 
 
 
 
 
 
 
1775
1776	/* Somebody else might have raced and expanded it already */
1777	if (address < vma->vm_start) {
1778		unsigned long size, grow;
1779
1780		size = vma->vm_end - address;
1781		grow = (vma->vm_start - address) >> PAGE_SHIFT;
1782
1783		error = -ENOMEM;
1784		if (grow <= vma->vm_pgoff) {
1785			error = acct_stack_growth(vma, size, grow);
1786			if (!error) {
 
 
 
 
1787				vma->vm_start = address;
1788				vma->vm_pgoff -= grow;
 
 
 
 
1789				perf_event_mmap(vma);
1790			}
1791		}
1792	}
1793	vma_unlock_anon_vma(vma);
1794	khugepaged_enter_vma_merge(vma);
 
1795	return error;
1796}
1797
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1798#ifdef CONFIG_STACK_GROWSUP
1799int expand_stack(struct vm_area_struct *vma, unsigned long address)
1800{
1801	return expand_upwards(vma, address);
1802}
1803
1804struct vm_area_struct *
1805find_extend_vma(struct mm_struct *mm, unsigned long addr)
1806{
1807	struct vm_area_struct *vma, *prev;
1808
1809	addr &= PAGE_MASK;
1810	vma = find_vma_prev(mm, addr, &prev);
1811	if (vma && (vma->vm_start <= addr))
1812		return vma;
1813	if (!prev || expand_stack(prev, addr))
1814		return NULL;
1815	if (prev->vm_flags & VM_LOCKED) {
1816		mlock_vma_pages_range(prev, addr, prev->vm_end);
1817	}
 
1818	return prev;
1819}
1820#else
1821int expand_stack(struct vm_area_struct *vma, unsigned long address)
1822{
1823	return expand_downwards(vma, address);
1824}
1825
1826struct vm_area_struct *
1827find_extend_vma(struct mm_struct * mm, unsigned long addr)
1828{
1829	struct vm_area_struct * vma;
1830	unsigned long start;
1831
1832	addr &= PAGE_MASK;
1833	vma = find_vma(mm,addr);
1834	if (!vma)
1835		return NULL;
1836	if (vma->vm_start <= addr)
1837		return vma;
1838	if (!(vma->vm_flags & VM_GROWSDOWN))
1839		return NULL;
1840	start = vma->vm_start;
1841	if (expand_stack(vma, addr))
1842		return NULL;
1843	if (vma->vm_flags & VM_LOCKED) {
1844		mlock_vma_pages_range(vma, addr, start);
1845	}
1846	return vma;
1847}
1848#endif
1849
1850/*
1851 * Ok - we have the memory areas we should free on the vma list,
1852 * so release them, and do the vma updates.
1853 *
1854 * Called with the mm semaphore held.
1855 */
1856static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma)
1857{
1858	/* Update high watermark before we lower total_vm */
1859	update_hiwater_vm(mm);
1860	do {
1861		long nrpages = vma_pages(vma);
1862
1863		mm->total_vm -= nrpages;
1864		vm_stat_account(mm, vma->vm_flags, vma->vm_file, -nrpages);
1865		vma = remove_vma(vma);
1866	} while (vma);
1867	validate_mm(mm);
1868}
 
 
 
1869
1870/*
1871 * Get rid of page table information in the indicated region.
 
1872 *
1873 * Called with the mm semaphore held.
 
 
 
 
 
1874 */
1875static void unmap_region(struct mm_struct *mm,
1876		struct vm_area_struct *vma, struct vm_area_struct *prev,
1877		unsigned long start, unsigned long end)
1878{
1879	struct vm_area_struct *next = prev? prev->vm_next: mm->mmap;
1880	struct mmu_gather tlb;
1881	unsigned long nr_accounted = 0;
1882
1883	lru_add_drain();
1884	tlb_gather_mmu(&tlb, mm, 0);
1885	update_hiwater_rss(mm);
1886	unmap_vmas(&tlb, vma, start, end, &nr_accounted, NULL);
1887	vm_unacct_memory(nr_accounted);
1888	free_pgtables(&tlb, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS,
1889				 next ? next->vm_start : 0);
1890	tlb_finish_mmu(&tlb, start, end);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1891}
1892
1893/*
1894 * Create a list of vma's touched by the unmap, removing them from the mm's
1895 * vma list as we go..
 
 
 
 
1896 */
1897static void
1898detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
1899	struct vm_area_struct *prev, unsigned long end)
1900{
1901	struct vm_area_struct **insertion_point;
1902	struct vm_area_struct *tail_vma = NULL;
1903	unsigned long addr;
1904
1905	insertion_point = (prev ? &prev->vm_next : &mm->mmap);
1906	vma->vm_prev = NULL;
1907	do {
1908		rb_erase(&vma->vm_rb, &mm->mm_rb);
1909		mm->map_count--;
1910		tail_vma = vma;
1911		vma = vma->vm_next;
1912	} while (vma && vma->vm_start < end);
1913	*insertion_point = vma;
1914	if (vma)
1915		vma->vm_prev = prev;
1916	tail_vma->vm_next = NULL;
1917	if (mm->unmap_area == arch_unmap_area)
1918		addr = prev ? prev->vm_end : mm->mmap_base;
1919	else
1920		addr = vma ?  vma->vm_start : mm->mmap_base;
1921	mm->unmap_area(mm, addr);
1922	mm->mmap_cache = NULL;		/* Kill the cache. */
1923}
1924
1925/*
1926 * __split_vma() bypasses sysctl_max_map_count checking.  We use this on the
1927 * munmap path where it doesn't make sense to fail.
1928 */
1929static int __split_vma(struct mm_struct * mm, struct vm_area_struct * vma,
1930	      unsigned long addr, int new_below)
1931{
1932	struct mempolicy *pol;
1933	struct vm_area_struct *new;
1934	int err = -ENOMEM;
1935
1936	if (is_vm_hugetlb_page(vma) && (addr &
1937					~(huge_page_mask(hstate_vma(vma)))))
1938		return -EINVAL;
1939
1940	new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
1941	if (!new)
1942		goto out_err;
1943
1944	/* most fields are the same, copy all, and then fixup */
1945	*new = *vma;
 
1946
1947	INIT_LIST_HEAD(&new->anon_vma_chain);
 
 
1948
1949	if (new_below)
1950		new->vm_end = addr;
1951	else {
1952		new->vm_start = addr;
1953		new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
1954	}
1955
1956	pol = mpol_dup(vma_policy(vma));
1957	if (IS_ERR(pol)) {
1958		err = PTR_ERR(pol);
1959		goto out_free_vma;
1960	}
1961	vma_set_policy(new, pol);
1962
1963	if (anon_vma_clone(new, vma))
1964		goto out_free_mpol;
 
1965
1966	if (new->vm_file) {
1967		get_file(new->vm_file);
1968		if (vma->vm_flags & VM_EXECUTABLE)
1969			added_exe_file_vma(mm);
1970	}
1971
1972	if (new->vm_ops && new->vm_ops->open)
1973		new->vm_ops->open(new);
 
 
 
 
1974
1975	if (new_below)
1976		err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
1977			((addr - new->vm_start) >> PAGE_SHIFT), new);
1978	else
1979		err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
1980
1981	/* Success. */
1982	if (!err)
1983		return 0;
1984
1985	/* Clean everything up if vma_adjust failed. */
1986	if (new->vm_ops && new->vm_ops->close)
1987		new->vm_ops->close(new);
1988	if (new->vm_file) {
1989		if (vma->vm_flags & VM_EXECUTABLE)
1990			removed_exe_file_vma(mm);
1991		fput(new->vm_file);
1992	}
1993	unlink_anon_vmas(new);
1994 out_free_mpol:
1995	mpol_put(pol);
1996 out_free_vma:
1997	kmem_cache_free(vm_area_cachep, new);
1998 out_err:
1999	return err;
2000}
2001
2002/*
2003 * Split a vma into two pieces at address 'addr', a new vma is allocated
2004 * either for the first part or the tail.
2005 */
2006int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
2007	      unsigned long addr, int new_below)
2008{
2009	if (mm->map_count >= sysctl_max_map_count)
2010		return -ENOMEM;
 
2011
2012	return __split_vma(mm, vma, addr, new_below);
 
 
 
2013}
2014
2015/* Munmap is split into 2 main parts -- this part which finds
2016 * what needs doing, and the areas themselves, which do the
2017 * work.  This now handles partial unmappings.
2018 * Jeremy Fitzhardinge <jeremy@goop.org>
2019 */
2020int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
 
2021{
2022	unsigned long end;
2023	struct vm_area_struct *vma, *prev, *last;
2024
2025	if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start)
2026		return -EINVAL;
 
 
 
 
2027
2028	if ((len = PAGE_ALIGN(len)) == 0)
2029		return -EINVAL;
2030
2031	/* Find the first overlapping VMA */
2032	vma = find_vma(mm, start);
2033	if (!vma)
2034		return 0;
2035	prev = vma->vm_prev;
2036	/* we have  start < vma->vm_end  */
2037
2038	/* if it doesn't overlap, we have nothing.. */
2039	end = start + len;
2040	if (vma->vm_start >= end)
2041		return 0;
 
 
 
 
 
2042
2043	/*
2044	 * If we need to split any vma, do it now to save pain later.
2045	 *
2046	 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2047	 * unmapped vm_area_struct will remain in use: so lower split_vma
2048	 * places tmp vma above, and higher split_vma places tmp vma below.
2049	 */
2050	if (start > vma->vm_start) {
2051		int error;
2052
2053		/*
2054		 * Make sure that map_count on return from munmap() will
2055		 * not exceed its limit; but let map_count go just above
2056		 * its limit temporarily, to help free resources as expected.
2057		 */
2058		if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count)
2059			return -ENOMEM;
2060
2061		error = __split_vma(mm, vma, start, 0);
2062		if (error)
2063			return error;
2064		prev = vma;
2065	}
2066
2067	/* Does it split the last one? */
2068	last = find_vma(mm, end);
2069	if (last && end > last->vm_start) {
2070		int error = __split_vma(mm, last, end, 1);
2071		if (error)
2072			return error;
2073	}
2074	vma = prev? prev->vm_next: mm->mmap;
2075
2076	/*
2077	 * unlock any mlock()ed ranges before detaching vmas
2078	 */
2079	if (mm->locked_vm) {
2080		struct vm_area_struct *tmp = vma;
2081		while (tmp && tmp->vm_start < end) {
2082			if (tmp->vm_flags & VM_LOCKED) {
2083				mm->locked_vm -= vma_pages(tmp);
2084				munlock_vma_pages_all(tmp);
2085			}
2086			tmp = tmp->vm_next;
2087		}
2088	}
2089
2090	/*
2091	 * Remove the vma's, and unmap the actual pages
2092	 */
2093	detach_vmas_to_be_unmapped(mm, vma, prev, end);
2094	unmap_region(mm, vma, prev, start, end);
2095
2096	/* Fix up all other VM information */
2097	remove_vma_list(mm, vma);
2098
2099	return 0;
2100}
 
 
 
 
2101
2102EXPORT_SYMBOL(do_munmap);
2103
2104SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
2105{
2106	int ret;
2107	struct mm_struct *mm = current->mm;
 
2108
2109	profile_munmap(addr);
2110
2111	down_write(&mm->mmap_sem);
2112	ret = do_munmap(mm, addr, len);
2113	up_write(&mm->mmap_sem);
2114	return ret;
2115}
2116
2117static inline void verify_mm_writelocked(struct mm_struct *mm)
2118{
2119#ifdef CONFIG_DEBUG_VM
2120	if (unlikely(down_read_trylock(&mm->mmap_sem))) {
2121		WARN_ON(1);
2122		up_read(&mm->mmap_sem);
2123	}
2124#endif
2125}
2126
2127/*
2128 *  this is really a simplified "do_mmap".  it only handles
2129 *  anonymous maps.  eventually we may be able to do some
2130 *  brk-specific accounting here.
2131 */
2132unsigned long do_brk(unsigned long addr, unsigned long len)
2133{
2134	struct mm_struct * mm = current->mm;
2135	struct vm_area_struct * vma, * prev;
2136	unsigned long flags;
2137	struct rb_node ** rb_link, * rb_parent;
2138	pgoff_t pgoff = addr >> PAGE_SHIFT;
2139	int error;
2140
2141	len = PAGE_ALIGN(len);
2142	if (!len)
2143		return addr;
2144
2145	error = security_file_mmap(NULL, 0, 0, 0, addr, 1);
2146	if (error)
2147		return error;
2148
2149	flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
 
2150
2151	error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED);
2152	if (error & ~PAGE_MASK)
2153		return error;
2154
2155	/*
2156	 * mlock MCL_FUTURE?
2157	 */
2158	if (mm->def_flags & VM_LOCKED) {
2159		unsigned long locked, lock_limit;
2160		locked = len >> PAGE_SHIFT;
2161		locked += mm->locked_vm;
2162		lock_limit = rlimit(RLIMIT_MEMLOCK);
2163		lock_limit >>= PAGE_SHIFT;
2164		if (locked > lock_limit && !capable(CAP_IPC_LOCK))
2165			return -EAGAIN;
2166	}
2167
2168	/*
2169	 * mm->mmap_sem is required to protect against another thread
2170	 * changing the mappings in case we sleep.
2171	 */
2172	verify_mm_writelocked(mm);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2173
2174	/*
2175	 * Clear old maps.  this also does some error checking for us
 
2176	 */
2177 munmap_back:
2178	vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
2179	if (vma && vma->vm_start < addr + len) {
2180		if (do_munmap(mm, addr, len))
2181			return -ENOMEM;
2182		goto munmap_back;
2183	}
2184
2185	/* Check against address space limits *after* clearing old maps... */
2186	if (!may_expand_vm(mm, len >> PAGE_SHIFT))
2187		return -ENOMEM;
2188
2189	if (mm->map_count > sysctl_max_map_count)
2190		return -ENOMEM;
2191
2192	if (security_vm_enough_memory(len >> PAGE_SHIFT))
2193		return -ENOMEM;
2194
2195	/* Can we just expand an old private anonymous mapping? */
2196	vma = vma_merge(mm, prev, addr, addr + len, flags,
2197					NULL, NULL, pgoff, NULL);
2198	if (vma)
2199		goto out;
2200
2201	/*
2202	 * create a vma struct for an anonymous mapping
 
2203	 */
2204	vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
2205	if (!vma) {
2206		vm_unacct_memory(len >> PAGE_SHIFT);
2207		return -ENOMEM;
 
 
 
 
 
 
 
2208	}
2209
2210	INIT_LIST_HEAD(&vma->anon_vma_chain);
2211	vma->vm_mm = mm;
2212	vma->vm_start = addr;
2213	vma->vm_end = addr + len;
2214	vma->vm_pgoff = pgoff;
2215	vma->vm_flags = flags;
 
 
 
 
2216	vma->vm_page_prot = vm_get_page_prot(flags);
2217	vma_link(mm, vma, prev, rb_link, rb_parent);
 
 
 
 
 
 
2218out:
2219	perf_event_mmap(vma);
2220	mm->total_vm += len >> PAGE_SHIFT;
2221	if (flags & VM_LOCKED) {
2222		if (!mlock_vma_pages_range(vma, addr, addr + len))
2223			mm->locked_vm += (len >> PAGE_SHIFT);
2224	}
2225	return addr;
 
 
 
 
 
 
2226}
2227
2228EXPORT_SYMBOL(do_brk);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2229
2230/* Release all mmaps. */
2231void exit_mmap(struct mm_struct *mm)
2232{
2233	struct mmu_gather tlb;
2234	struct vm_area_struct *vma;
2235	unsigned long nr_accounted = 0;
2236	unsigned long end;
 
2237
2238	/* mm's last user has gone, and its about to be pulled down */
2239	mmu_notifier_release(mm);
2240
2241	if (mm->locked_vm) {
2242		vma = mm->mmap;
2243		while (vma) {
2244			if (vma->vm_flags & VM_LOCKED)
2245				munlock_vma_pages_all(vma);
2246			vma = vma->vm_next;
2247		}
2248	}
2249
2250	arch_exit_mmap(mm);
2251
2252	vma = mm->mmap;
2253	if (!vma)	/* Can happen if dup_mmap() received an OOM */
2254		return;
 
 
 
 
2255
2256	lru_add_drain();
2257	flush_cache_mm(mm);
2258	tlb_gather_mmu(&tlb, mm, 1);
2259	/* update_hiwater_rss(mm) here? but nobody should be looking */
2260	/* Use -1 here to ensure all VMAs in the mm are unmapped */
2261	end = unmap_vmas(&tlb, vma, 0, -1, &nr_accounted, NULL);
2262	vm_unacct_memory(nr_accounted);
2263
2264	free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, 0);
2265	tlb_finish_mmu(&tlb, 0, end);
2266
2267	/*
2268	 * Walk the list again, actually closing and freeing it,
2269	 * with preemption enabled, without holding any MM locks.
2270	 */
2271	while (vma)
2272		vma = remove_vma(vma);
 
 
 
 
 
2273
2274	BUG_ON(mm->nr_ptes > (FIRST_USER_ADDRESS+PMD_SIZE-1)>>PMD_SHIFT);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2275}
2276
2277/* Insert vm structure into process list sorted by address
2278 * and into the inode's i_mmap tree.  If vm_file is non-NULL
2279 * then i_mmap_mutex is taken here.
2280 */
2281int insert_vm_struct(struct mm_struct * mm, struct vm_area_struct * vma)
2282{
2283	struct vm_area_struct * __vma, * prev;
2284	struct rb_node ** rb_link, * rb_parent;
 
 
 
 
 
 
 
2285
2286	/*
2287	 * The vm_pgoff of a purely anonymous vma should be irrelevant
2288	 * until its first write fault, when page's anon_vma and index
2289	 * are set.  But now set the vm_pgoff it will almost certainly
2290	 * end up with (unless mremap moves it elsewhere before that
2291	 * first wfault), so /proc/pid/maps tells a consistent story.
2292	 *
2293	 * By setting it to reflect the virtual start address of the
2294	 * vma, merges and splits can happen in a seamless way, just
2295	 * using the existing file pgoff checks and manipulations.
2296	 * Similarly in do_mmap_pgoff and in do_brk.
2297	 */
2298	if (!vma->vm_file) {
2299		BUG_ON(vma->anon_vma);
2300		vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
2301	}
2302	__vma = find_vma_prepare(mm,vma->vm_start,&prev,&rb_link,&rb_parent);
2303	if (__vma && __vma->vm_start < vma->vm_end)
2304		return -ENOMEM;
2305	if ((vma->vm_flags & VM_ACCOUNT) &&
2306	     security_vm_enough_memory_mm(mm, vma_pages(vma)))
2307		return -ENOMEM;
2308	vma_link(mm, vma, prev, rb_link, rb_parent);
 
2309	return 0;
2310}
2311
2312/*
2313 * Copy the vma structure to a new location in the same mm,
2314 * prior to moving page table entries, to effect an mremap move.
2315 */
2316struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
2317	unsigned long addr, unsigned long len, pgoff_t pgoff)
2318{
2319	struct vm_area_struct *vma = *vmap;
2320	unsigned long vma_start = vma->vm_start;
2321	struct mm_struct *mm = vma->vm_mm;
2322	struct vm_area_struct *new_vma, *prev;
2323	struct rb_node **rb_link, *rb_parent;
2324	struct mempolicy *pol;
2325
2326	/*
2327	 * If anonymous vma has not yet been faulted, update new pgoff
2328	 * to match new location, to increase its chance of merging.
2329	 */
2330	if (!vma->vm_file && !vma->anon_vma)
2331		pgoff = addr >> PAGE_SHIFT;
2332
2333	find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
2334	new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
2335			vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
2336	if (new_vma) {
2337		/*
2338		 * Source vma may have been merged into new_vma
2339		 */
2340		if (vma_start >= new_vma->vm_start &&
2341		    vma_start < new_vma->vm_end)
2342			*vmap = new_vma;
2343	} else {
2344		new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
2345		if (new_vma) {
2346			*new_vma = *vma;
2347			pol = mpol_dup(vma_policy(vma));
2348			if (IS_ERR(pol))
2349				goto out_free_vma;
2350			INIT_LIST_HEAD(&new_vma->anon_vma_chain);
2351			if (anon_vma_clone(new_vma, vma))
2352				goto out_free_mempol;
2353			vma_set_policy(new_vma, pol);
2354			new_vma->vm_start = addr;
2355			new_vma->vm_end = addr + len;
2356			new_vma->vm_pgoff = pgoff;
2357			if (new_vma->vm_file) {
2358				get_file(new_vma->vm_file);
2359				if (vma->vm_flags & VM_EXECUTABLE)
2360					added_exe_file_vma(mm);
2361			}
2362			if (new_vma->vm_ops && new_vma->vm_ops->open)
2363				new_vma->vm_ops->open(new_vma);
2364			vma_link(mm, new_vma, prev, rb_link, rb_parent);
2365		}
2366	}
2367	return new_vma;
2368
2369 out_free_mempol:
2370	mpol_put(pol);
2371 out_free_vma:
2372	kmem_cache_free(vm_area_cachep, new_vma);
2373	return NULL;
 
 
 
 
 
 
 
 
2374}
2375
 
 
2376/*
2377 * Return true if the calling process may expand its vm space by the passed
2378 * number of pages
 
2379 */
2380int may_expand_vm(struct mm_struct *mm, unsigned long npages)
2381{
2382	unsigned long cur = mm->total_vm;	/* pages */
2383	unsigned long lim;
2384
2385	lim = rlimit(RLIMIT_AS) >> PAGE_SHIFT;
 
 
2386
2387	if (cur + npages > lim)
2388		return 0;
2389	return 1;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2390}
2391
 
 
 
 
 
 
 
 
 
2392
2393static int special_mapping_fault(struct vm_area_struct *vma,
2394				struct vm_fault *vmf)
2395{
 
2396	pgoff_t pgoff;
2397	struct page **pages;
 
2398
2399	/*
2400	 * special mappings have no vm_file, and in that case, the mm
2401	 * uses vm_pgoff internally. So we have to subtract it from here.
2402	 * We are allowed to do this because we are the mm; do not copy
2403	 * this code into drivers!
2404	 */
2405	pgoff = vmf->pgoff - vma->vm_pgoff;
2406
2407	for (pages = vma->vm_private_data; pgoff && *pages; ++pages)
2408		pgoff--;
2409
2410	if (*pages) {
2411		struct page *page = *pages;
2412		get_page(page);
2413		vmf->page = page;
2414		return 0;
2415	}
2416
2417	return VM_FAULT_SIGBUS;
2418}
2419
2420/*
2421 * Having a close hook prevents vma merging regardless of flags.
2422 */
2423static void special_mapping_close(struct vm_area_struct *vma)
2424{
2425}
2426
2427static const struct vm_operations_struct special_mapping_vmops = {
2428	.close = special_mapping_close,
2429	.fault = special_mapping_fault,
2430};
2431
2432/*
2433 * Called with mm->mmap_sem held for writing.
2434 * Insert a new vma covering the given region, with the given flags.
2435 * Its pages are supplied by the given array of struct page *.
2436 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
2437 * The region past the last page supplied will always produce SIGBUS.
2438 * The array pointer and the pages it points to are assumed to stay alive
2439 * for as long as this mapping might exist.
2440 */
2441int install_special_mapping(struct mm_struct *mm,
2442			    unsigned long addr, unsigned long len,
2443			    unsigned long vm_flags, struct page **pages)
2444{
2445	int ret;
2446	struct vm_area_struct *vma;
2447
2448	vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
2449	if (unlikely(vma == NULL))
2450		return -ENOMEM;
2451
2452	INIT_LIST_HEAD(&vma->anon_vma_chain);
2453	vma->vm_mm = mm;
2454	vma->vm_start = addr;
2455	vma->vm_end = addr + len;
2456
2457	vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND;
2458	vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
2459
2460	vma->vm_ops = &special_mapping_vmops;
2461	vma->vm_private_data = pages;
2462
2463	ret = security_file_mmap(NULL, 0, 0, 0, vma->vm_start, 1);
2464	if (ret)
2465		goto out;
2466
2467	ret = insert_vm_struct(mm, vma);
2468	if (ret)
2469		goto out;
2470
2471	mm->total_vm += len >> PAGE_SHIFT;
2472
2473	perf_event_mmap(vma);
2474
2475	return 0;
2476
2477out:
2478	kmem_cache_free(vm_area_cachep, vma);
2479	return ret;
2480}
2481
2482static DEFINE_MUTEX(mm_all_locks_mutex);
 
 
 
 
 
2483
2484static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma)
 
 
 
 
 
 
 
 
 
 
 
 
2485{
2486	if (!test_bit(0, (unsigned long *) &anon_vma->root->head.next)) {
2487		/*
2488		 * The LSB of head.next can't change from under us
2489		 * because we hold the mm_all_locks_mutex.
2490		 */
2491		mutex_lock_nest_lock(&anon_vma->root->mutex, &mm->mmap_sem);
2492		/*
2493		 * We can safely modify head.next after taking the
2494		 * anon_vma->root->mutex. If some other vma in this mm shares
2495		 * the same anon_vma we won't take it again.
2496		 *
2497		 * No need of atomic instructions here, head.next
2498		 * can't change from under us thanks to the
2499		 * anon_vma->root->mutex.
2500		 */
2501		if (__test_and_set_bit(0, (unsigned long *)
2502				       &anon_vma->root->head.next))
2503			BUG();
2504	}
2505}
2506
2507static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping)
 
 
 
2508{
2509	if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
2510		/*
2511		 * AS_MM_ALL_LOCKS can't change from under us because
2512		 * we hold the mm_all_locks_mutex.
2513		 *
2514		 * Operations on ->flags have to be atomic because
2515		 * even if AS_MM_ALL_LOCKS is stable thanks to the
2516		 * mm_all_locks_mutex, there may be other cpus
2517		 * changing other bitflags in parallel to us.
2518		 */
2519		if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags))
2520			BUG();
2521		mutex_lock_nest_lock(&mapping->i_mmap_mutex, &mm->mmap_sem);
2522	}
2523}
2524
2525/*
2526 * This operation locks against the VM for all pte/vma/mm related
2527 * operations that could ever happen on a certain mm. This includes
2528 * vmtruncate, try_to_unmap, and all page faults.
2529 *
2530 * The caller must take the mmap_sem in write mode before calling
2531 * mm_take_all_locks(). The caller isn't allowed to release the
2532 * mmap_sem until mm_drop_all_locks() returns.
2533 *
2534 * mmap_sem in write mode is required in order to block all operations
2535 * that could modify pagetables and free pages without need of
2536 * altering the vma layout (for example populate_range() with
2537 * nonlinear vmas). It's also needed in write mode to avoid new
2538 * anon_vmas to be associated with existing vmas.
2539 *
2540 * A single task can't take more than one mm_take_all_locks() in a row
2541 * or it would deadlock.
2542 *
2543 * The LSB in anon_vma->head.next and the AS_MM_ALL_LOCKS bitflag in
2544 * mapping->flags avoid to take the same lock twice, if more than one
2545 * vma in this mm is backed by the same anon_vma or address_space.
2546 *
2547 * We can take all the locks in random order because the VM code
2548 * taking i_mmap_mutex or anon_vma->mutex outside the mmap_sem never
2549 * takes more than one of them in a row. Secondly we're protected
2550 * against a concurrent mm_take_all_locks() by the mm_all_locks_mutex.
 
 
 
 
 
 
 
 
 
 
 
 
2551 *
2552 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
2553 * that may have to take thousand of locks.
2554 *
2555 * mm_take_all_locks() can fail if it's interrupted by signals.
 
 
2556 */
2557int mm_take_all_locks(struct mm_struct *mm)
2558{
2559	struct vm_area_struct *vma;
2560	struct anon_vma_chain *avc;
2561	int ret = -EINTR;
2562
2563	BUG_ON(down_read_trylock(&mm->mmap_sem));
2564
2565	mutex_lock(&mm_all_locks_mutex);
 
 
 
2566
2567	for (vma = mm->mmap; vma; vma = vma->vm_next) {
2568		if (signal_pending(current))
2569			goto out_unlock;
2570		if (vma->vm_file && vma->vm_file->f_mapping)
2571			vm_lock_mapping(mm, vma->vm_file->f_mapping);
2572	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2573
2574	for (vma = mm->mmap; vma; vma = vma->vm_next) {
2575		if (signal_pending(current))
2576			goto out_unlock;
2577		if (vma->anon_vma)
2578			list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
2579				vm_lock_anon_vma(mm, avc->anon_vma);
2580	}
2581
2582	ret = 0;
 
 
 
2583
2584out_unlock:
2585	if (ret)
2586		mm_drop_all_locks(mm);
2587
2588	return ret;
2589}
 
 
 
2590
2591static void vm_unlock_anon_vma(struct anon_vma *anon_vma)
2592{
2593	if (test_bit(0, (unsigned long *) &anon_vma->root->head.next)) {
2594		/*
2595		 * The LSB of head.next can't change to 0 from under
2596		 * us because we hold the mm_all_locks_mutex.
2597		 *
2598		 * We must however clear the bitflag before unlocking
2599		 * the vma so the users using the anon_vma->head will
2600		 * never see our bitflag.
2601		 *
2602		 * No need of atomic instructions here, head.next
2603		 * can't change from under us until we release the
2604		 * anon_vma->root->mutex.
2605		 */
2606		if (!__test_and_clear_bit(0, (unsigned long *)
2607					  &anon_vma->root->head.next))
2608			BUG();
2609		anon_vma_unlock(anon_vma);
2610	}
 
2611}
2612
2613static void vm_unlock_mapping(struct address_space *mapping)
2614{
2615	if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
2616		/*
2617		 * AS_MM_ALL_LOCKS can't change to 0 from under us
2618		 * because we hold the mm_all_locks_mutex.
2619		 */
2620		mutex_unlock(&mapping->i_mmap_mutex);
2621		if (!test_and_clear_bit(AS_MM_ALL_LOCKS,
2622					&mapping->flags))
2623			BUG();
2624	}
2625}
 
2626
2627/*
2628 * The mmap_sem cannot be released by the caller until
2629 * mm_drop_all_locks() returns.
 
 
 
 
2630 */
2631void mm_drop_all_locks(struct mm_struct *mm)
2632{
2633	struct vm_area_struct *vma;
2634	struct anon_vma_chain *avc;
 
 
 
 
 
 
 
 
2635
2636	BUG_ON(down_read_trylock(&mm->mmap_sem));
2637	BUG_ON(!mutex_is_locked(&mm_all_locks_mutex));
 
 
 
 
 
 
 
 
2638
2639	for (vma = mm->mmap; vma; vma = vma->vm_next) {
2640		if (vma->anon_vma)
2641			list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
2642				vm_unlock_anon_vma(avc->anon_vma);
2643		if (vma->vm_file && vma->vm_file->f_mapping)
2644			vm_unlock_mapping(vma->vm_file->f_mapping);
2645	}
2646
2647	mutex_unlock(&mm_all_locks_mutex);
2648}
 
 
 
 
2649
2650/*
2651 * initialise the VMA slab
2652 */
2653void __init mmap_init(void)
2654{
2655	int ret;
 
2656
2657	ret = percpu_counter_init(&vm_committed_as, 0);
2658	VM_BUG_ON(ret);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2659}
v6.13.7
   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 * mm/mmap.c
   4 *
   5 * Written by obz.
   6 *
   7 * Address space accounting code	<alan@lxorguk.ukuu.org.uk>
   8 */
   9
  10#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  11
  12#include <linux/kernel.h>
  13#include <linux/slab.h>
  14#include <linux/backing-dev.h>
  15#include <linux/mm.h>
  16#include <linux/mm_inline.h>
  17#include <linux/shm.h>
  18#include <linux/mman.h>
  19#include <linux/pagemap.h>
  20#include <linux/swap.h>
  21#include <linux/syscalls.h>
  22#include <linux/capability.h>
  23#include <linux/init.h>
  24#include <linux/file.h>
  25#include <linux/fs.h>
  26#include <linux/personality.h>
  27#include <linux/security.h>
  28#include <linux/hugetlb.h>
  29#include <linux/shmem_fs.h>
  30#include <linux/profile.h>
  31#include <linux/export.h>
  32#include <linux/mount.h>
  33#include <linux/mempolicy.h>
  34#include <linux/rmap.h>
  35#include <linux/mmu_notifier.h>
  36#include <linux/mmdebug.h>
  37#include <linux/perf_event.h>
  38#include <linux/audit.h>
  39#include <linux/khugepaged.h>
  40#include <linux/uprobes.h>
  41#include <linux/notifier.h>
  42#include <linux/memory.h>
  43#include <linux/printk.h>
  44#include <linux/userfaultfd_k.h>
  45#include <linux/moduleparam.h>
  46#include <linux/pkeys.h>
  47#include <linux/oom.h>
  48#include <linux/sched/mm.h>
  49#include <linux/ksm.h>
  50#include <linux/memfd.h>
  51
  52#include <linux/uaccess.h>
  53#include <asm/cacheflush.h>
  54#include <asm/tlb.h>
  55#include <asm/mmu_context.h>
  56
  57#define CREATE_TRACE_POINTS
  58#include <trace/events/mmap.h>
  59
  60#include "internal.h"
  61
  62#ifndef arch_mmap_check
  63#define arch_mmap_check(addr, len, flags)	(0)
  64#endif
  65
  66#ifdef CONFIG_HAVE_ARCH_MMAP_RND_BITS
  67const int mmap_rnd_bits_min = CONFIG_ARCH_MMAP_RND_BITS_MIN;
  68int mmap_rnd_bits_max __ro_after_init = CONFIG_ARCH_MMAP_RND_BITS_MAX;
  69int mmap_rnd_bits __read_mostly = CONFIG_ARCH_MMAP_RND_BITS;
  70#endif
  71#ifdef CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS
  72const int mmap_rnd_compat_bits_min = CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MIN;
  73const int mmap_rnd_compat_bits_max = CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MAX;
  74int mmap_rnd_compat_bits __read_mostly = CONFIG_ARCH_MMAP_RND_COMPAT_BITS;
  75#endif
  76
  77static bool ignore_rlimit_data;
  78core_param(ignore_rlimit_data, ignore_rlimit_data, bool, 0644);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  79
  80/* Update vma->vm_page_prot to reflect vma->vm_flags. */
  81void vma_set_page_prot(struct vm_area_struct *vma)
  82{
  83	unsigned long vm_flags = vma->vm_flags;
  84	pgprot_t vm_page_prot;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  85
  86	vm_page_prot = vm_pgprot_modify(vma->vm_page_prot, vm_flags);
  87	if (vma_wants_writenotify(vma, vm_page_prot)) {
  88		vm_flags &= ~VM_SHARED;
  89		vm_page_prot = vm_pgprot_modify(vm_page_prot, vm_flags);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  90	}
  91	/* remove_protection_ptes reads vma->vm_page_prot without mmap_lock */
  92	WRITE_ONCE(vma->vm_page_prot, vm_page_prot);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  93}
  94
  95/*
  96 * check_brk_limits() - Use platform specific check of range & verify mlock
  97 * limits.
  98 * @addr: The address to check
  99 * @len: The size of increase.
 100 *
 101 * Return: 0 on success.
 102 */
 103static int check_brk_limits(unsigned long addr, unsigned long len)
 
 104{
 105	unsigned long mapped_addr;
 
 
 
 
 
 
 
 
 
 
 
 106
 107	mapped_addr = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED);
 108	if (IS_ERR_VALUE(mapped_addr))
 109		return mapped_addr;
 
 
 
 
 110
 111	return mlock_future_ok(current->mm, current->mm->def_flags, len)
 112		? 0 : -EAGAIN;
 
 
 
 
 113}
 114static int do_brk_flags(struct vma_iterator *vmi, struct vm_area_struct *brkvma,
 115		unsigned long addr, unsigned long request, unsigned long flags);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 116SYSCALL_DEFINE1(brk, unsigned long, brk)
 117{
 118	unsigned long newbrk, oldbrk, origbrk;
 
 119	struct mm_struct *mm = current->mm;
 120	struct vm_area_struct *brkvma, *next = NULL;
 121	unsigned long min_brk;
 122	bool populate = false;
 123	LIST_HEAD(uf);
 124	struct vma_iterator vmi;
 125
 126	if (mmap_write_lock_killable(mm))
 127		return -EINTR;
 128
 129	origbrk = mm->brk;
 130
 131#ifdef CONFIG_COMPAT_BRK
 132	/*
 133	 * CONFIG_COMPAT_BRK can still be overridden by setting
 134	 * randomize_va_space to 2, which will still cause mm->start_brk
 135	 * to be arbitrarily shifted
 136	 */
 137	if (current->brk_randomized)
 138		min_brk = mm->start_brk;
 139	else
 140		min_brk = mm->end_data;
 141#else
 142	min_brk = mm->start_brk;
 143#endif
 144	if (brk < min_brk)
 145		goto out;
 146
 147	/*
 148	 * Check against rlimit here. If this check is done later after the test
 149	 * of oldbrk with newbrk then it can escape the test and let the data
 150	 * segment grow beyond its set limit the in case where the limit is
 151	 * not page aligned -Ram Gupta
 152	 */
 153	if (check_data_rlimit(rlimit(RLIMIT_DATA), brk, mm->start_brk,
 154			      mm->end_data, mm->start_data))
 
 155		goto out;
 156
 157	newbrk = PAGE_ALIGN(brk);
 158	oldbrk = PAGE_ALIGN(mm->brk);
 159	if (oldbrk == newbrk) {
 160		mm->brk = brk;
 161		goto success;
 162	}
 163
 164	/* Always allow shrinking brk. */
 165	if (brk <= mm->brk) {
 166		/* Search one past newbrk */
 167		vma_iter_init(&vmi, mm, newbrk);
 168		brkvma = vma_find(&vmi, oldbrk);
 169		if (!brkvma || brkvma->vm_start >= oldbrk)
 170			goto out; /* mapping intersects with an existing non-brk vma. */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 171		/*
 172		 * mm->brk must be protected by write mmap_lock.
 173		 * do_vmi_align_munmap() will drop the lock on success,  so
 174		 * update it before calling do_vma_munmap().
 175		 */
 176		mm->brk = brk;
 177		if (do_vmi_align_munmap(&vmi, brkvma, mm, newbrk, oldbrk, &uf,
 178					/* unlock = */ true))
 179			goto out;
 
 
 180
 181		goto success_unlocked;
 
 
 
 
 
 
 
 
 
 
 
 
 
 182	}
 183
 184	if (check_brk_limits(oldbrk, newbrk - oldbrk))
 185		goto out;
 186
 187	/*
 188	 * Only check if the next VMA is within the stack_guard_gap of the
 189	 * expansion area
 
 
 190	 */
 191	vma_iter_init(&vmi, mm, oldbrk);
 192	next = vma_find(&vmi, newbrk + PAGE_SIZE + stack_guard_gap);
 193	if (next && newbrk + PAGE_SIZE > vm_start_gap(next))
 194		goto out;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 195
 196	brkvma = vma_prev_limit(&vmi, mm->start_brk);
 197	/* Ok, looks good - let it rip. */
 198	if (do_brk_flags(&vmi, brkvma, oldbrk, newbrk - oldbrk, 0) < 0)
 199		goto out;
 
 
 
 
 
 
 
 
 
 
 
 
 200
 201	mm->brk = brk;
 202	if (mm->def_flags & VM_LOCKED)
 203		populate = true;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 204
 205success:
 206	mmap_write_unlock(mm);
 207success_unlocked:
 208	userfaultfd_unmap_complete(mm, &uf);
 209	if (populate)
 210		mm_populate(oldbrk, newbrk - oldbrk);
 211	return brk;
 212
 213out:
 214	mm->brk = origbrk;
 215	mmap_write_unlock(mm);
 216	return origbrk;
 217}
 218
 219/*
 220 * If a hint addr is less than mmap_min_addr change hint to be as
 221 * low as possible but still greater than mmap_min_addr
 222 */
 223static inline unsigned long round_hint_to_min(unsigned long hint)
 
 224{
 225	hint &= PAGE_MASK;
 226	if (((void *)hint != NULL) &&
 227	    (hint < mmap_min_addr))
 228		return PAGE_ALIGN(mmap_min_addr);
 229	return hint;
 
 
 
 230}
 231
 232bool mlock_future_ok(struct mm_struct *mm, unsigned long flags,
 233			unsigned long bytes)
 
 234{
 235	unsigned long locked_pages, limit_pages;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 236
 237	if (!(flags & VM_LOCKED) || capable(CAP_IPC_LOCK))
 238		return true;
 
 
 
 
 239
 240	locked_pages = bytes >> PAGE_SHIFT;
 241	locked_pages += mm->locked_vm;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 242
 243	limit_pages = rlimit(RLIMIT_MEMLOCK);
 244	limit_pages >>= PAGE_SHIFT;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 245
 246	return locked_pages <= limit_pages;
 247}
 248
 249static inline u64 file_mmap_size_max(struct file *file, struct inode *inode)
 
 
 
 
 
 
 
 
 
 
 
 
 
 250{
 251	if (S_ISREG(inode->i_mode))
 252		return MAX_LFS_FILESIZE;
 
 
 
 
 253
 254	if (S_ISBLK(inode->i_mode))
 255		return MAX_LFS_FILESIZE;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 256
 257	if (S_ISSOCK(inode->i_mode))
 258		return MAX_LFS_FILESIZE;
 
 
 
 259
 260	/* Special "we do even unsigned file positions" case */
 261	if (file->f_op->fop_flags & FOP_UNSIGNED_OFFSET)
 262		return 0;
 263
 264	/* Yes, random drivers might want more. But I'm tired of buggy drivers */
 265	return ULONG_MAX;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 266}
 267
 268static inline bool file_mmap_ok(struct file *file, struct inode *inode,
 269				unsigned long pgoff, unsigned long len)
 
 270{
 271	u64 maxsize = file_mmap_size_max(file, inode);
 
 272
 273	if (maxsize && len > maxsize)
 274		return false;
 275	maxsize -= len;
 276	if (pgoff > maxsize >> PAGE_SHIFT)
 277		return false;
 278	return true;
 
 
 279}
 
 280
 281/*
 282 * The caller must write-lock current->mm->mmap_lock.
 283 */
 284unsigned long do_mmap(struct file *file, unsigned long addr,
 
 285			unsigned long len, unsigned long prot,
 286			unsigned long flags, vm_flags_t vm_flags,
 287			unsigned long pgoff, unsigned long *populate,
 288			struct list_head *uf)
 289{
 290	struct mm_struct *mm = current->mm;
 291	int pkey = 0;
 292
 293	*populate = 0;
 294
 295	if (!len)
 296		return -EINVAL;
 297
 298	/*
 299	 * Does the application expect PROT_READ to imply PROT_EXEC?
 300	 *
 301	 * (the exception is when the underlying filesystem is noexec
 302	 *  mounted, in which case we don't add PROT_EXEC.)
 303	 */
 304	if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
 305		if (!(file && path_noexec(&file->f_path)))
 306			prot |= PROT_EXEC;
 307
 308	/* force arch specific MAP_FIXED handling in get_unmapped_area */
 309	if (flags & MAP_FIXED_NOREPLACE)
 310		flags |= MAP_FIXED;
 311
 312	if (!(flags & MAP_FIXED))
 313		addr = round_hint_to_min(addr);
 314
 315	/* Careful about overflows.. */
 316	len = PAGE_ALIGN(len);
 317	if (!len)
 318		return -ENOMEM;
 319
 320	/* offset overflow? */
 321	if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
 322		return -EOVERFLOW;
 323
 324	/* Too many mappings? */
 325	if (mm->map_count > sysctl_max_map_count)
 326		return -ENOMEM;
 327
 328	/*
 329	 * addr is returned from get_unmapped_area,
 330	 * There are two cases:
 331	 * 1> MAP_FIXED == false
 332	 *	unallocated memory, no need to check sealing.
 333	 * 1> MAP_FIXED == true
 334	 *	sealing is checked inside mmap_region when
 335	 *	do_vmi_munmap is called.
 336	 */
 337
 338	if (prot == PROT_EXEC) {
 339		pkey = execute_only_pkey(mm);
 340		if (pkey < 0)
 341			pkey = 0;
 342	}
 343
 344	/* Do simple checking here so the lower-level routines won't have
 345	 * to. we assume access permissions have been handled by the open
 346	 * of the memory object, so we don't do any here.
 347	 */
 348	vm_flags |= calc_vm_prot_bits(prot, pkey) | calc_vm_flag_bits(file, flags) |
 349			mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
 350
 351	/* Obtain the address to map to. we verify (or select) it and ensure
 352	 * that it represents a valid section of the address space.
 353	 */
 354	addr = __get_unmapped_area(file, addr, len, pgoff, flags, vm_flags);
 355	if (IS_ERR_VALUE(addr))
 356		return addr;
 357
 358	if (flags & MAP_FIXED_NOREPLACE) {
 359		if (find_vma_intersection(mm, addr, addr + len))
 360			return -EEXIST;
 361	}
 362
 363	if (flags & MAP_LOCKED)
 364		if (!can_do_mlock())
 365			return -EPERM;
 366
 367	if (!mlock_future_ok(mm, vm_flags, len))
 368		return -EAGAIN;
 
 
 
 
 
 
 
 
 
 
 369
 370	if (file) {
 371		struct inode *inode = file_inode(file);
 372		unsigned int seals = memfd_file_seals(file);
 373		unsigned long flags_mask;
 374
 375		if (!file_mmap_ok(file, inode, pgoff, len))
 376			return -EOVERFLOW;
 377
 378		flags_mask = LEGACY_MAP_MASK;
 379		if (file->f_op->fop_flags & FOP_MMAP_SYNC)
 380			flags_mask |= MAP_SYNC;
 381
 382		switch (flags & MAP_TYPE) {
 383		case MAP_SHARED:
 384			/*
 385			 * Force use of MAP_SHARED_VALIDATE with non-legacy
 386			 * flags. E.g. MAP_SYNC is dangerous to use with
 387			 * MAP_SHARED as you don't know which consistency model
 388			 * you will get. We silently ignore unsupported flags
 389			 * with MAP_SHARED to preserve backward compatibility.
 390			 */
 391			flags &= LEGACY_MAP_MASK;
 392			fallthrough;
 393		case MAP_SHARED_VALIDATE:
 394			if (flags & ~flags_mask)
 395				return -EOPNOTSUPP;
 396			if (prot & PROT_WRITE) {
 397				if (!(file->f_mode & FMODE_WRITE))
 398					return -EACCES;
 399				if (IS_SWAPFILE(file->f_mapping->host))
 400					return -ETXTBSY;
 401			}
 402
 403			/*
 404			 * Make sure we don't allow writing to an append-only
 405			 * file..
 406			 */
 407			if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
 408				return -EACCES;
 409
 
 
 
 
 
 
 410			vm_flags |= VM_SHARED | VM_MAYSHARE;
 411			if (!(file->f_mode & FMODE_WRITE))
 412				vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
 413			else if (is_readonly_sealed(seals, vm_flags))
 414				vm_flags &= ~VM_MAYWRITE;
 415			fallthrough;
 416		case MAP_PRIVATE:
 417			if (!(file->f_mode & FMODE_READ))
 418				return -EACCES;
 419			if (path_noexec(&file->f_path)) {
 420				if (vm_flags & VM_EXEC)
 421					return -EPERM;
 422				vm_flags &= ~VM_MAYEXEC;
 423			}
 424
 425			if (!file->f_op->mmap)
 426				return -ENODEV;
 427			if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
 428				return -EINVAL;
 429			break;
 430
 431		default:
 432			return -EINVAL;
 433		}
 434	} else {
 435		switch (flags & MAP_TYPE) {
 436		case MAP_SHARED:
 437			if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
 438				return -EINVAL;
 439			/*
 440			 * Ignore pgoff.
 441			 */
 442			pgoff = 0;
 443			vm_flags |= VM_SHARED | VM_MAYSHARE;
 444			break;
 445		case MAP_DROPPABLE:
 446			if (VM_DROPPABLE == VM_NONE)
 447				return -ENOTSUPP;
 448			/*
 449			 * A locked or stack area makes no sense to be droppable.
 450			 *
 451			 * Also, since droppable pages can just go away at any time
 452			 * it makes no sense to copy them on fork or dump them.
 453			 *
 454			 * And don't attempt to combine with hugetlb for now.
 455			 */
 456			if (flags & (MAP_LOCKED | MAP_HUGETLB))
 457			        return -EINVAL;
 458			if (vm_flags & (VM_GROWSDOWN | VM_GROWSUP))
 459			        return -EINVAL;
 460
 461			vm_flags |= VM_DROPPABLE;
 462
 463			/*
 464			 * If the pages can be dropped, then it doesn't make
 465			 * sense to reserve them.
 466			 */
 467			vm_flags |= VM_NORESERVE;
 468
 469			/*
 470			 * Likewise, they're volatile enough that they
 471			 * shouldn't survive forks or coredumps.
 472			 */
 473			vm_flags |= VM_WIPEONFORK | VM_DONTDUMP;
 474			fallthrough;
 475		case MAP_PRIVATE:
 476			/*
 477			 * Set pgoff according to addr for anon_vma.
 478			 */
 479			pgoff = addr >> PAGE_SHIFT;
 480			break;
 481		default:
 482			return -EINVAL;
 483		}
 484	}
 485
 486	/*
 487	 * Set 'VM_NORESERVE' if we should not account for the
 488	 * memory use of this mapping.
 489	 */
 490	if (flags & MAP_NORESERVE) {
 491		/* We honor MAP_NORESERVE if allowed to overcommit */
 492		if (sysctl_overcommit_memory != OVERCOMMIT_NEVER)
 493			vm_flags |= VM_NORESERVE;
 494
 495		/* hugetlb applies strict overcommit unless MAP_NORESERVE */
 496		if (file && is_file_hugepages(file))
 497			vm_flags |= VM_NORESERVE;
 498	}
 499
 500	addr = mmap_region(file, addr, len, vm_flags, pgoff, uf);
 501	if (!IS_ERR_VALUE(addr) &&
 502	    ((vm_flags & VM_LOCKED) ||
 503	     (flags & (MAP_POPULATE | MAP_NONBLOCK)) == MAP_POPULATE))
 504		*populate = len;
 505	return addr;
 506}
 
 507
 508unsigned long ksys_mmap_pgoff(unsigned long addr, unsigned long len,
 509			      unsigned long prot, unsigned long flags,
 510			      unsigned long fd, unsigned long pgoff)
 511{
 512	struct file *file = NULL;
 513	unsigned long retval;
 514
 515	if (!(flags & MAP_ANONYMOUS)) {
 516		audit_mmap_fd(fd, flags);
 
 
 517		file = fget(fd);
 518		if (!file)
 519			return -EBADF;
 520		if (is_file_hugepages(file)) {
 521			len = ALIGN(len, huge_page_size(hstate_file(file)));
 522		} else if (unlikely(flags & MAP_HUGETLB)) {
 523			retval = -EINVAL;
 524			goto out_fput;
 525		}
 526	} else if (flags & MAP_HUGETLB) {
 527		struct hstate *hs;
 528
 529		hs = hstate_sizelog((flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK);
 530		if (!hs)
 531			return -EINVAL;
 532
 533		len = ALIGN(len, huge_page_size(hs));
 534		/*
 535		 * VM_NORESERVE is used because the reservations will be
 536		 * taken when vm_ops->mmap() is called
 
 
 537		 */
 538		file = hugetlb_file_setup(HUGETLB_ANON_FILE, len,
 539				VM_NORESERVE,
 540				HUGETLB_ANONHUGE_INODE,
 541				(flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK);
 542		if (IS_ERR(file))
 543			return PTR_ERR(file);
 544	}
 545
 546	retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
 547out_fput:
 
 
 
 
 548	if (file)
 549		fput(file);
 
 550	return retval;
 551}
 552
 553SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
 554		unsigned long, prot, unsigned long, flags,
 555		unsigned long, fd, unsigned long, pgoff)
 556{
 557	return ksys_mmap_pgoff(addr, len, prot, flags, fd, pgoff);
 558}
 559
 560#ifdef __ARCH_WANT_SYS_OLD_MMAP
 561struct mmap_arg_struct {
 562	unsigned long addr;
 563	unsigned long len;
 564	unsigned long prot;
 565	unsigned long flags;
 566	unsigned long fd;
 567	unsigned long offset;
 568};
 569
 570SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
 571{
 572	struct mmap_arg_struct a;
 573
 574	if (copy_from_user(&a, arg, sizeof(a)))
 575		return -EFAULT;
 576	if (offset_in_page(a.offset))
 577		return -EINVAL;
 578
 579	return ksys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
 580			       a.offset >> PAGE_SHIFT);
 581}
 582#endif /* __ARCH_WANT_SYS_OLD_MMAP */
 583
 584/**
 585 * unmapped_area() - Find an area between the low_limit and the high_limit with
 586 * the correct alignment and offset, all from @info. Note: current->mm is used
 587 * for the search.
 588 *
 589 * @info: The unmapped area information including the range [low_limit -
 590 * high_limit), the alignment offset and mask.
 591 *
 592 * Return: A memory address or -ENOMEM.
 593 */
 594static unsigned long unmapped_area(struct vm_unmapped_area_info *info)
 595{
 596	unsigned long length, gap;
 597	unsigned long low_limit, high_limit;
 598	struct vm_area_struct *tmp;
 599	VMA_ITERATOR(vmi, current->mm, 0);
 600
 601	/* Adjust search length to account for worst case alignment overhead */
 602	length = info->length + info->align_mask + info->start_gap;
 603	if (length < info->length)
 604		return -ENOMEM;
 605
 606	low_limit = info->low_limit;
 607	if (low_limit < mmap_min_addr)
 608		low_limit = mmap_min_addr;
 609	high_limit = info->high_limit;
 610retry:
 611	if (vma_iter_area_lowest(&vmi, low_limit, high_limit, length))
 612		return -ENOMEM;
 
 
 613
 
 
 
 
 
 
 614	/*
 615	 * Adjust for the gap first so it doesn't interfere with the
 616	 * later alignment. The first step is the minimum needed to
 617	 * fulill the start gap, the next steps is the minimum to align
 618	 * that. It is the minimum needed to fulill both.
 619	 */
 620	gap = vma_iter_addr(&vmi) + info->start_gap;
 621	gap += (info->align_offset - gap) & info->align_mask;
 622	tmp = vma_next(&vmi);
 623	if (tmp && (tmp->vm_flags & VM_STARTGAP_FLAGS)) { /* Avoid prev check if possible */
 624		if (vm_start_gap(tmp) < gap + length - 1) {
 625			low_limit = tmp->vm_end;
 626			vma_iter_reset(&vmi);
 627			goto retry;
 628		}
 629	} else {
 630		tmp = vma_prev(&vmi);
 631		if (tmp && vm_end_gap(tmp) > gap) {
 632			low_limit = vm_end_gap(tmp);
 633			vma_iter_reset(&vmi);
 634			goto retry;
 635		}
 636	}
 637
 638	return gap;
 639}
 640
 641/**
 642 * unmapped_area_topdown() - Find an area between the low_limit and the
 643 * high_limit with the correct alignment and offset at the highest available
 644 * address, all from @info. Note: current->mm is used for the search.
 645 *
 646 * @info: The unmapped area information including the range [low_limit -
 647 * high_limit), the alignment offset and mask.
 648 *
 649 * Return: A memory address or -ENOMEM.
 650 */
 651static unsigned long unmapped_area_topdown(struct vm_unmapped_area_info *info)
 652{
 653	unsigned long length, gap, gap_end;
 654	unsigned long low_limit, high_limit;
 655	struct vm_area_struct *tmp;
 656	VMA_ITERATOR(vmi, current->mm, 0);
 
 
 
 
 
 
 
 
 
 
 
 
 
 657
 658	/* Adjust search length to account for worst case alignment overhead */
 659	length = info->length + info->align_mask + info->start_gap;
 660	if (length < info->length)
 661		return -ENOMEM;
 662
 663	low_limit = info->low_limit;
 664	if (low_limit < mmap_min_addr)
 665		low_limit = mmap_min_addr;
 666	high_limit = info->high_limit;
 667retry:
 668	if (vma_iter_area_highest(&vmi, low_limit, high_limit, length))
 669		return -ENOMEM;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 670
 671	gap = vma_iter_end(&vmi) - info->length;
 672	gap -= (gap - info->align_offset) & info->align_mask;
 673	gap_end = vma_iter_end(&vmi);
 674	tmp = vma_next(&vmi);
 675	if (tmp && (tmp->vm_flags & VM_STARTGAP_FLAGS)) { /* Avoid prev check if possible */
 676		if (vm_start_gap(tmp) < gap_end) {
 677			high_limit = vm_start_gap(tmp);
 678			vma_iter_reset(&vmi);
 679			goto retry;
 680		}
 681	} else {
 682		tmp = vma_prev(&vmi);
 683		if (tmp && vm_end_gap(tmp) > gap) {
 684			high_limit = tmp->vm_start;
 685			vma_iter_reset(&vmi);
 686			goto retry;
 687		}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 688	}
 689
 690	return gap;
 691}
 
 
 
 
 
 
 
 
 
 
 
 
 692
 693/*
 694 * Determine if the allocation needs to ensure that there is no
 695 * existing mapping within it's guard gaps, for use as start_gap.
 696 */
 697static inline unsigned long stack_guard_placement(vm_flags_t vm_flags)
 698{
 699	if (vm_flags & VM_SHADOW_STACK)
 700		return PAGE_SIZE;
 701
 702	return 0;
 703}
 
 
 
 704
 705/*
 706 * Search for an unmapped address range.
 707 *
 708 * We are looking for a range that:
 709 * - does not intersect with any VMA;
 710 * - is contained within the [low_limit, high_limit) interval;
 711 * - is at least the desired size.
 712 * - satisfies (begin_addr & align_mask) == (align_offset & align_mask)
 713 */
 714unsigned long vm_unmapped_area(struct vm_unmapped_area_info *info)
 715{
 716	unsigned long addr;
 717
 718	if (info->flags & VM_UNMAPPED_AREA_TOPDOWN)
 719		addr = unmapped_area_topdown(info);
 720	else
 721		addr = unmapped_area(info);
 
 722
 723	trace_vm_unmapped_area(addr, info);
 724	return addr;
 
 
 
 
 
 
 
 725}
 726
 727/* Get an address range which is currently unmapped.
 728 * For shmat() with addr=0.
 729 *
 730 * Ugly calling convention alert:
 731 * Return value with the low bits set means error value,
 732 * ie
 733 *	if (ret & ~PAGE_MASK)
 734 *		error = ret;
 735 *
 736 * This function "knows" that -ENOMEM has the bits set.
 737 */
 
 738unsigned long
 739generic_get_unmapped_area(struct file *filp, unsigned long addr,
 740			  unsigned long len, unsigned long pgoff,
 741			  unsigned long flags, vm_flags_t vm_flags)
 742{
 743	struct mm_struct *mm = current->mm;
 744	struct vm_area_struct *vma, *prev;
 745	struct vm_unmapped_area_info info = {};
 746	const unsigned long mmap_end = arch_get_mmap_end(addr, len, flags);
 747
 748	if (len > mmap_end - mmap_min_addr)
 749		return -ENOMEM;
 750
 751	if (flags & MAP_FIXED)
 752		return addr;
 753
 754	if (addr) {
 755		addr = PAGE_ALIGN(addr);
 756		vma = find_vma_prev(mm, addr, &prev);
 757		if (mmap_end - len >= addr && addr >= mmap_min_addr &&
 758		    (!vma || addr + len <= vm_start_gap(vma)) &&
 759		    (!prev || addr >= vm_end_gap(prev)))
 760			return addr;
 761	}
 
 
 
 
 
 
 762
 763	info.length = len;
 764	info.low_limit = mm->mmap_base;
 765	info.high_limit = mmap_end;
 766	info.start_gap = stack_guard_placement(vm_flags);
 767	if (filp && is_file_hugepages(filp))
 768		info.align_mask = huge_page_mask_align(filp);
 769	return vm_unmapped_area(&info);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 770}
 
 771
 772#ifndef HAVE_ARCH_UNMAPPED_AREA
 773unsigned long
 774arch_get_unmapped_area(struct file *filp, unsigned long addr,
 775		       unsigned long len, unsigned long pgoff,
 776		       unsigned long flags, vm_flags_t vm_flags)
 777{
 778	return generic_get_unmapped_area(filp, addr, len, pgoff, flags,
 779					 vm_flags);
 
 
 
 
 
 780}
 781#endif
 782
 783/*
 784 * This mmap-allocator allocates new areas top-down from below the
 785 * stack's low limit (the base):
 786 */
 
 787unsigned long
 788generic_get_unmapped_area_topdown(struct file *filp, unsigned long addr,
 789				  unsigned long len, unsigned long pgoff,
 790				  unsigned long flags, vm_flags_t vm_flags)
 791{
 792	struct vm_area_struct *vma, *prev;
 793	struct mm_struct *mm = current->mm;
 794	struct vm_unmapped_area_info info = {};
 795	const unsigned long mmap_end = arch_get_mmap_end(addr, len, flags);
 796
 797	/* requested length too big for entire address space */
 798	if (len > mmap_end - mmap_min_addr)
 799		return -ENOMEM;
 800
 801	if (flags & MAP_FIXED)
 802		return addr;
 803
 804	/* requesting a specific address */
 805	if (addr) {
 806		addr = PAGE_ALIGN(addr);
 807		vma = find_vma_prev(mm, addr, &prev);
 808		if (mmap_end - len >= addr && addr >= mmap_min_addr &&
 809				(!vma || addr + len <= vm_start_gap(vma)) &&
 810				(!prev || addr >= vm_end_gap(prev)))
 811			return addr;
 812	}
 813
 814	info.flags = VM_UNMAPPED_AREA_TOPDOWN;
 815	info.length = len;
 816	info.low_limit = PAGE_SIZE;
 817	info.high_limit = arch_get_mmap_base(addr, mm->mmap_base);
 818	info.start_gap = stack_guard_placement(vm_flags);
 819	if (filp && is_file_hugepages(filp))
 820		info.align_mask = huge_page_mask_align(filp);
 821	addr = vm_unmapped_area(&info);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 822
 
 823	/*
 824	 * A failed mmap() very likely causes application failure,
 825	 * so fall back to the bottom-up function here. This scenario
 826	 * can happen with large stack limits and large mmap()
 827	 * allocations.
 828	 */
 829	if (offset_in_page(addr)) {
 830		VM_BUG_ON(addr != -ENOMEM);
 831		info.flags = 0;
 832		info.low_limit = TASK_UNMAPPED_BASE;
 833		info.high_limit = mmap_end;
 834		addr = vm_unmapped_area(&info);
 835	}
 
 836
 837	return addr;
 838}
 
 839
 840#ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
 841unsigned long
 842arch_get_unmapped_area_topdown(struct file *filp, unsigned long addr,
 843			       unsigned long len, unsigned long pgoff,
 844			       unsigned long flags, vm_flags_t vm_flags)
 845{
 846	return generic_get_unmapped_area_topdown(filp, addr, len, pgoff, flags,
 847						 vm_flags);
 848}
 849#endif
 
 850
 851unsigned long mm_get_unmapped_area_vmflags(struct mm_struct *mm, struct file *filp,
 852					   unsigned long addr, unsigned long len,
 853					   unsigned long pgoff, unsigned long flags,
 854					   vm_flags_t vm_flags)
 855{
 856	if (test_bit(MMF_TOPDOWN, &mm->flags))
 857		return arch_get_unmapped_area_topdown(filp, addr, len, pgoff,
 858						      flags, vm_flags);
 859	return arch_get_unmapped_area(filp, addr, len, pgoff, flags, vm_flags);
 860}
 861
 862unsigned long
 863__get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
 864		unsigned long pgoff, unsigned long flags, vm_flags_t vm_flags)
 865{
 866	unsigned long (*get_area)(struct file *, unsigned long,
 867				  unsigned long, unsigned long, unsigned long)
 868				  = NULL;
 869
 870	unsigned long error = arch_mmap_check(addr, len, flags);
 871	if (error)
 872		return error;
 873
 874	/* Careful about overflows.. */
 875	if (len > TASK_SIZE)
 876		return -ENOMEM;
 877
 878	if (file) {
 879		if (file->f_op->get_unmapped_area)
 880			get_area = file->f_op->get_unmapped_area;
 881	} else if (flags & MAP_SHARED) {
 882		/*
 883		 * mmap_region() will call shmem_zero_setup() to create a file,
 884		 * so use shmem's get_unmapped_area in case it can be huge.
 885		 */
 886		get_area = shmem_get_unmapped_area;
 887	}
 888
 889	/* Always treat pgoff as zero for anonymous memory. */
 890	if (!file)
 891		pgoff = 0;
 892
 893	if (get_area) {
 894		addr = get_area(file, addr, len, pgoff, flags);
 895	} else if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) && !file
 896		   && !addr /* no hint */
 897		   && IS_ALIGNED(len, PMD_SIZE)) {
 898		/* Ensures that larger anonymous mappings are THP aligned. */
 899		addr = thp_get_unmapped_area_vmflags(file, addr, len,
 900						     pgoff, flags, vm_flags);
 901	} else {
 902		addr = mm_get_unmapped_area_vmflags(current->mm, file, addr, len,
 903						    pgoff, flags, vm_flags);
 904	}
 905	if (IS_ERR_VALUE(addr))
 906		return addr;
 907
 908	if (addr > TASK_SIZE - len)
 909		return -ENOMEM;
 910	if (offset_in_page(addr))
 911		return -EINVAL;
 912
 913	error = security_mmap_addr(addr);
 914	return error ? error : addr;
 915}
 916
 917unsigned long
 918mm_get_unmapped_area(struct mm_struct *mm, struct file *file,
 919		     unsigned long addr, unsigned long len,
 920		     unsigned long pgoff, unsigned long flags)
 921{
 922	if (test_bit(MMF_TOPDOWN, &mm->flags))
 923		return arch_get_unmapped_area_topdown(file, addr, len, pgoff, flags, 0);
 924	return arch_get_unmapped_area(file, addr, len, pgoff, flags, 0);
 925}
 926EXPORT_SYMBOL(mm_get_unmapped_area);
 927
 928/**
 929 * find_vma_intersection() - Look up the first VMA which intersects the interval
 930 * @mm: The process address space.
 931 * @start_addr: The inclusive start user address.
 932 * @end_addr: The exclusive end user address.
 933 *
 934 * Returns: The first VMA within the provided range, %NULL otherwise.  Assumes
 935 * start_addr < end_addr.
 936 */
 937struct vm_area_struct *find_vma_intersection(struct mm_struct *mm,
 938					     unsigned long start_addr,
 939					     unsigned long end_addr)
 940{
 941	unsigned long index = start_addr;
 942
 943	mmap_assert_locked(mm);
 944	return mt_find(&mm->mm_mt, &index, end_addr - 1);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 945}
 946EXPORT_SYMBOL(find_vma_intersection);
 947
 948/**
 949 * find_vma() - Find the VMA for a given address, or the next VMA.
 950 * @mm: The mm_struct to check
 951 * @addr: The address
 952 *
 953 * Returns: The VMA associated with addr, or the next VMA.
 954 * May return %NULL in the case of no VMA at addr or above.
 955 */
 956struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
 957{
 958	unsigned long index = addr;
 959
 960	mmap_assert_locked(mm);
 961	return mt_find(&mm->mm_mt, &index, ULONG_MAX);
 962}
 963EXPORT_SYMBOL(find_vma);
 964
 965/**
 966 * find_vma_prev() - Find the VMA for a given address, or the next vma and
 967 * set %pprev to the previous VMA, if any.
 968 * @mm: The mm_struct to check
 969 * @addr: The address
 970 * @pprev: The pointer to set to the previous VMA
 971 *
 972 * Note that RCU lock is missing here since the external mmap_lock() is used
 973 * instead.
 974 *
 975 * Returns: The VMA associated with @addr, or the next vma.
 976 * May return %NULL in the case of no vma at addr or above.
 977 */
 978struct vm_area_struct *
 979find_vma_prev(struct mm_struct *mm, unsigned long addr,
 980			struct vm_area_struct **pprev)
 981{
 982	struct vm_area_struct *vma;
 983	VMA_ITERATOR(vmi, mm, addr);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 984
 985	vma = vma_iter_load(&vmi);
 986	*pprev = vma_prev(&vmi);
 987	if (!vma)
 988		vma = vma_next(&vmi);
 989	return vma;
 990}
 991
 992/*
 993 * Verify that the stack growth is acceptable and
 994 * update accounting. This is shared with both the
 995 * grow-up and grow-down cases.
 996 */
 997static int acct_stack_growth(struct vm_area_struct *vma,
 998			     unsigned long size, unsigned long grow)
 999{
1000	struct mm_struct *mm = vma->vm_mm;
 
1001	unsigned long new_start;
1002
1003	/* address space limit tests */
1004	if (!may_expand_vm(mm, vma->vm_flags, grow))
1005		return -ENOMEM;
1006
1007	/* Stack limit test */
1008	if (size > rlimit(RLIMIT_STACK))
1009		return -ENOMEM;
1010
1011	/* mlock limit tests */
1012	if (!mlock_future_ok(mm, vma->vm_flags, grow << PAGE_SHIFT))
1013		return -ENOMEM;
 
 
 
 
 
 
 
1014
1015	/* Check to ensure the stack will not grow into a hugetlb-only region */
1016	new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start :
1017			vma->vm_end - size;
1018	if (is_hugepage_only_range(vma->vm_mm, new_start, size))
1019		return -EFAULT;
1020
1021	/*
1022	 * Overcommit..  This must be the final test, as it will
1023	 * update security statistics.
1024	 */
1025	if (security_vm_enough_memory_mm(mm, grow))
1026		return -ENOMEM;
1027
 
 
 
 
 
1028	return 0;
1029}
1030
1031#if defined(CONFIG_STACK_GROWSUP)
1032/*
1033 * PA-RISC uses this for its stack.
1034 * vma is the last one with address > vma->vm_end.  Have to extend vma.
1035 */
1036static int expand_upwards(struct vm_area_struct *vma, unsigned long address)
1037{
1038	struct mm_struct *mm = vma->vm_mm;
1039	struct vm_area_struct *next;
1040	unsigned long gap_addr;
1041	int error = 0;
1042	VMA_ITERATOR(vmi, mm, vma->vm_start);
1043
1044	if (!(vma->vm_flags & VM_GROWSUP))
1045		return -EFAULT;
1046
1047	mmap_assert_write_locked(mm);
1048
1049	/* Guard against exceeding limits of the address space. */
1050	address &= PAGE_MASK;
1051	if (address >= (TASK_SIZE & PAGE_MASK))
1052		return -ENOMEM;
1053	address += PAGE_SIZE;
1054
1055	/* Enforce stack_guard_gap */
1056	gap_addr = address + stack_guard_gap;
1057
1058	/* Guard against overflow */
1059	if (gap_addr < address || gap_addr > TASK_SIZE)
1060		gap_addr = TASK_SIZE;
1061
1062	next = find_vma_intersection(mm, vma->vm_end, gap_addr);
1063	if (next && vma_is_accessible(next)) {
1064		if (!(next->vm_flags & VM_GROWSUP))
1065			return -ENOMEM;
1066		/* Check that both stack segments have the same anon_vma? */
1067	}
1068
1069	if (next)
1070		vma_iter_prev_range_limit(&vmi, address);
1071
1072	vma_iter_config(&vmi, vma->vm_start, address);
1073	if (vma_iter_prealloc(&vmi, vma))
1074		return -ENOMEM;
1075
1076	/* We must make sure the anon_vma is allocated. */
1077	if (unlikely(anon_vma_prepare(vma))) {
1078		vma_iter_free(&vmi);
1079		return -ENOMEM;
1080	}
1081
1082	/* Lock the VMA before expanding to prevent concurrent page faults */
1083	vma_start_write(vma);
1084	/* We update the anon VMA tree. */
1085	anon_vma_lock_write(vma->anon_vma);
1086
1087	/* Somebody else might have raced and expanded it already */
1088	if (address > vma->vm_end) {
1089		unsigned long size, grow;
1090
1091		size = address - vma->vm_start;
1092		grow = (address - vma->vm_end) >> PAGE_SHIFT;
1093
1094		error = -ENOMEM;
1095		if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) {
1096			error = acct_stack_growth(vma, size, grow);
1097			if (!error) {
1098				if (vma->vm_flags & VM_LOCKED)
1099					mm->locked_vm += grow;
1100				vm_stat_account(mm, vma->vm_flags, grow);
1101				anon_vma_interval_tree_pre_update_vma(vma);
1102				vma->vm_end = address;
1103				/* Overwrite old entry in mtree. */
1104				vma_iter_store(&vmi, vma);
1105				anon_vma_interval_tree_post_update_vma(vma);
1106
1107				perf_event_mmap(vma);
1108			}
1109		}
1110	}
1111	anon_vma_unlock_write(vma->anon_vma);
1112	vma_iter_free(&vmi);
1113	validate_mm(mm);
1114	return error;
1115}
1116#endif /* CONFIG_STACK_GROWSUP */
1117
1118/*
1119 * vma is the first one with address < vma->vm_start.  Have to extend vma.
1120 * mmap_lock held for writing.
1121 */
1122int expand_downwards(struct vm_area_struct *vma, unsigned long address)
 
1123{
1124	struct mm_struct *mm = vma->vm_mm;
1125	struct vm_area_struct *prev;
1126	int error = 0;
1127	VMA_ITERATOR(vmi, mm, vma->vm_start);
1128
1129	if (!(vma->vm_flags & VM_GROWSDOWN))
1130		return -EFAULT;
1131
1132	mmap_assert_write_locked(mm);
 
 
1133
1134	address &= PAGE_MASK;
1135	if (address < mmap_min_addr || address < FIRST_USER_ADDRESS)
1136		return -EPERM;
 
1137
1138	/* Enforce stack_guard_gap */
1139	prev = vma_prev(&vmi);
1140	/* Check that both stack segments have the same anon_vma? */
1141	if (prev) {
1142		if (!(prev->vm_flags & VM_GROWSDOWN) &&
1143		    vma_is_accessible(prev) &&
1144		    (address - prev->vm_end < stack_guard_gap))
1145			return -ENOMEM;
1146	}
1147
1148	if (prev)
1149		vma_iter_next_range_limit(&vmi, vma->vm_start);
1150
1151	vma_iter_config(&vmi, address, vma->vm_end);
1152	if (vma_iter_prealloc(&vmi, vma))
1153		return -ENOMEM;
1154
1155	/* We must make sure the anon_vma is allocated. */
1156	if (unlikely(anon_vma_prepare(vma))) {
1157		vma_iter_free(&vmi);
1158		return -ENOMEM;
1159	}
1160
1161	/* Lock the VMA before expanding to prevent concurrent page faults */
1162	vma_start_write(vma);
1163	/* We update the anon VMA tree. */
1164	anon_vma_lock_write(vma->anon_vma);
1165
1166	/* Somebody else might have raced and expanded it already */
1167	if (address < vma->vm_start) {
1168		unsigned long size, grow;
1169
1170		size = vma->vm_end - address;
1171		grow = (vma->vm_start - address) >> PAGE_SHIFT;
1172
1173		error = -ENOMEM;
1174		if (grow <= vma->vm_pgoff) {
1175			error = acct_stack_growth(vma, size, grow);
1176			if (!error) {
1177				if (vma->vm_flags & VM_LOCKED)
1178					mm->locked_vm += grow;
1179				vm_stat_account(mm, vma->vm_flags, grow);
1180				anon_vma_interval_tree_pre_update_vma(vma);
1181				vma->vm_start = address;
1182				vma->vm_pgoff -= grow;
1183				/* Overwrite old entry in mtree. */
1184				vma_iter_store(&vmi, vma);
1185				anon_vma_interval_tree_post_update_vma(vma);
1186
1187				perf_event_mmap(vma);
1188			}
1189		}
1190	}
1191	anon_vma_unlock_write(vma->anon_vma);
1192	vma_iter_free(&vmi);
1193	validate_mm(mm);
1194	return error;
1195}
1196
1197/* enforced gap between the expanding stack and other mappings. */
1198unsigned long stack_guard_gap = 256UL<<PAGE_SHIFT;
1199
1200static int __init cmdline_parse_stack_guard_gap(char *p)
1201{
1202	unsigned long val;
1203	char *endptr;
1204
1205	val = simple_strtoul(p, &endptr, 10);
1206	if (!*endptr)
1207		stack_guard_gap = val << PAGE_SHIFT;
1208
1209	return 1;
1210}
1211__setup("stack_guard_gap=", cmdline_parse_stack_guard_gap);
1212
1213#ifdef CONFIG_STACK_GROWSUP
1214int expand_stack_locked(struct vm_area_struct *vma, unsigned long address)
1215{
1216	return expand_upwards(vma, address);
1217}
1218
1219struct vm_area_struct *find_extend_vma_locked(struct mm_struct *mm, unsigned long addr)
 
1220{
1221	struct vm_area_struct *vma, *prev;
1222
1223	addr &= PAGE_MASK;
1224	vma = find_vma_prev(mm, addr, &prev);
1225	if (vma && (vma->vm_start <= addr))
1226		return vma;
1227	if (!prev)
1228		return NULL;
1229	if (expand_stack_locked(prev, addr))
1230		return NULL;
1231	if (prev->vm_flags & VM_LOCKED)
1232		populate_vma_page_range(prev, addr, prev->vm_end, NULL);
1233	return prev;
1234}
1235#else
1236int expand_stack_locked(struct vm_area_struct *vma, unsigned long address)
1237{
1238	return expand_downwards(vma, address);
1239}
1240
1241struct vm_area_struct *find_extend_vma_locked(struct mm_struct *mm, unsigned long addr)
 
1242{
1243	struct vm_area_struct *vma;
1244	unsigned long start;
1245
1246	addr &= PAGE_MASK;
1247	vma = find_vma(mm, addr);
1248	if (!vma)
1249		return NULL;
1250	if (vma->vm_start <= addr)
1251		return vma;
 
 
1252	start = vma->vm_start;
1253	if (expand_stack_locked(vma, addr))
1254		return NULL;
1255	if (vma->vm_flags & VM_LOCKED)
1256		populate_vma_page_range(vma, addr, start, NULL);
 
1257	return vma;
1258}
1259#endif
1260
1261#if defined(CONFIG_STACK_GROWSUP)
 
 
 
 
 
 
 
 
 
 
 
1262
1263#define vma_expand_up(vma,addr) expand_upwards(vma, addr)
1264#define vma_expand_down(vma, addr) (-EFAULT)
1265
1266#else
1267
1268#define vma_expand_up(vma,addr) (-EFAULT)
1269#define vma_expand_down(vma, addr) expand_downwards(vma, addr)
1270
1271#endif
1272
1273/*
1274 * expand_stack(): legacy interface for page faulting. Don't use unless
1275 * you have to.
1276 *
1277 * This is called with the mm locked for reading, drops the lock, takes
1278 * the lock for writing, tries to look up a vma again, expands it if
1279 * necessary, and downgrades the lock to reading again.
1280 *
1281 * If no vma is found or it can't be expanded, it returns NULL and has
1282 * dropped the lock.
1283 */
1284struct vm_area_struct *expand_stack(struct mm_struct *mm, unsigned long addr)
 
 
1285{
1286	struct vm_area_struct *vma, *prev;
 
 
1287
1288	mmap_read_unlock(mm);
1289	if (mmap_write_lock_killable(mm))
1290		return NULL;
1291
1292	vma = find_vma_prev(mm, addr, &prev);
1293	if (vma && vma->vm_start <= addr)
1294		goto success;
1295
1296	if (prev && !vma_expand_up(prev, addr)) {
1297		vma = prev;
1298		goto success;
1299	}
1300
1301	if (vma && !vma_expand_down(vma, addr))
1302		goto success;
1303
1304	mmap_write_unlock(mm);
1305	return NULL;
1306
1307success:
1308	mmap_write_downgrade(mm);
1309	return vma;
1310}
1311
1312/* do_munmap() - Wrapper function for non-maple tree aware do_munmap() calls.
1313 * @mm: The mm_struct
1314 * @start: The start address to munmap
1315 * @len: The length to be munmapped.
1316 * @uf: The userfaultfd list_head
1317 *
1318 * Return: 0 on success, error otherwise.
1319 */
1320int do_munmap(struct mm_struct *mm, unsigned long start, size_t len,
1321	      struct list_head *uf)
 
1322{
1323	VMA_ITERATOR(vmi, mm, start);
 
 
1324
1325	return do_vmi_munmap(&vmi, mm, start, len, uf, false);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1326}
1327
1328unsigned long mmap_region(struct file *file, unsigned long addr,
1329			  unsigned long len, vm_flags_t vm_flags, unsigned long pgoff,
1330			  struct list_head *uf)
 
 
 
1331{
1332	unsigned long ret;
1333	bool writable_file_mapping = false;
 
 
 
 
 
1334
1335	/* Check to see if MDWE is applicable. */
1336	if (map_deny_write_exec(vm_flags, vm_flags))
1337		return -EACCES;
1338
1339	/* Allow architectures to sanity-check the vm_flags. */
1340	if (!arch_validate_flags(vm_flags))
1341		return -EINVAL;
1342
1343	/* Map writable and ensure this isn't a sealed memfd. */
1344	if (file && is_shared_maywrite(vm_flags)) {
1345		int error = mapping_map_writable(file->f_mapping);
1346
1347		if (error)
1348			return error;
1349		writable_file_mapping = true;
 
 
1350	}
1351
1352	ret = __mmap_region(file, addr, len, vm_flags, pgoff, uf);
 
 
 
 
 
1353
1354	/* Clear our write mapping regardless of error. */
1355	if (writable_file_mapping)
1356		mapping_unmap_writable(file->f_mapping);
1357
1358	validate_mm(current->mm);
1359	return ret;
1360}
 
 
1361
1362static int __vm_munmap(unsigned long start, size_t len, bool unlock)
1363{
1364	int ret;
1365	struct mm_struct *mm = current->mm;
1366	LIST_HEAD(uf);
1367	VMA_ITERATOR(vmi, mm, start);
1368
1369	if (mmap_write_lock_killable(mm))
1370		return -EINTR;
 
 
 
1371
1372	ret = do_vmi_munmap(&vmi, mm, start, len, &uf, unlock);
1373	if (ret || !unlock)
1374		mmap_write_unlock(mm);
1375
1376	userfaultfd_unmap_complete(mm, &uf);
1377	return ret;
 
 
 
 
 
 
 
 
 
 
 
 
 
1378}
1379
1380int vm_munmap(unsigned long start, size_t len)
 
 
 
 
 
1381{
1382	return __vm_munmap(start, len, false);
1383}
1384EXPORT_SYMBOL(vm_munmap);
1385
1386SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
1387{
1388	addr = untagged_addr(addr);
1389	return __vm_munmap(addr, len, true);
1390}
1391
1392
1393/*
1394 * Emulation of deprecated remap_file_pages() syscall.
 
1395 */
1396SYSCALL_DEFINE5(remap_file_pages, unsigned long, start, unsigned long, size,
1397		unsigned long, prot, unsigned long, pgoff, unsigned long, flags)
1398{
 
 
1399
1400	struct mm_struct *mm = current->mm;
1401	struct vm_area_struct *vma;
1402	unsigned long populate = 0;
1403	unsigned long ret = -EINVAL;
1404	struct file *file;
1405	vm_flags_t vm_flags;
1406
1407	pr_warn_once("%s (%d) uses deprecated remap_file_pages() syscall. See Documentation/mm/remap_file_pages.rst.\n",
1408		     current->comm, current->pid);
1409
1410	if (prot)
1411		return ret;
1412	start = start & PAGE_MASK;
1413	size = size & PAGE_MASK;
 
 
1414
1415	if (start + size <= start)
1416		return ret;
1417
1418	/* Does pgoff wrap? */
1419	if (pgoff + (size >> PAGE_SHIFT) < pgoff)
1420		return ret;
1421
1422	if (mmap_read_lock_killable(mm))
1423		return -EINTR;
1424
1425	/*
1426	 * Look up VMA under read lock first so we can perform the security
1427	 * without holding locks (which can be problematic). We reacquire a
1428	 * write lock later and check nothing changed underneath us.
 
 
1429	 */
1430	vma = vma_lookup(mm, start);
 
 
 
 
 
 
 
 
 
1431
1432	if (!vma || !(vma->vm_flags & VM_SHARED)) {
1433		mmap_read_unlock(mm);
1434		return -EINVAL;
 
1435	}
1436
1437	prot |= vma->vm_flags & VM_READ ? PROT_READ : 0;
1438	prot |= vma->vm_flags & VM_WRITE ? PROT_WRITE : 0;
1439	prot |= vma->vm_flags & VM_EXEC ? PROT_EXEC : 0;
 
 
 
 
 
1440
1441	flags &= MAP_NONBLOCK;
1442	flags |= MAP_SHARED | MAP_FIXED | MAP_POPULATE;
1443	if (vma->vm_flags & VM_LOCKED)
1444		flags |= MAP_LOCKED;
 
 
 
 
 
 
 
 
 
1445
1446	/* Save vm_flags used to calculate prot and flags, and recheck later. */
1447	vm_flags = vma->vm_flags;
1448	file = get_file(vma->vm_file);
 
 
1449
1450	mmap_read_unlock(mm);
 
1451
1452	/* Call outside mmap_lock to be consistent with other callers. */
1453	ret = security_mmap_file(file, prot, flags);
1454	if (ret) {
1455		fput(file);
1456		return ret;
1457	}
1458
1459	ret = -EINVAL;
1460
1461	/* OK security check passed, take write lock + let it rip. */
1462	if (mmap_write_lock_killable(mm)) {
1463		fput(file);
1464		return -EINTR;
1465	}
1466
1467	vma = vma_lookup(mm, start);
1468
1469	if (!vma)
1470		goto out;
 
 
 
1471
1472	/* Make sure things didn't change under us. */
1473	if (vma->vm_flags != vm_flags)
1474		goto out;
1475	if (vma->vm_file != file)
1476		goto out;
 
 
 
 
1477
1478	if (start + size > vma->vm_end) {
1479		VMA_ITERATOR(vmi, mm, vma->vm_end);
1480		struct vm_area_struct *next, *prev = vma;
1481
1482		for_each_vma_range(vmi, next, start + size) {
1483			/* hole between vmas ? */
1484			if (next->vm_start != prev->vm_end)
1485				goto out;
 
 
 
 
 
1486
1487			if (next->vm_file != vma->vm_file)
1488				goto out;
 
1489
1490			if (next->vm_flags != vma->vm_flags)
1491				goto out;
 
1492
1493			if (start + size <= next->vm_end)
1494				break;
1495
1496			prev = next;
1497		}
 
1498
1499		if (!next)
1500			goto out;
 
 
 
 
 
 
 
 
 
1501	}
1502
1503	ret = do_mmap(vma->vm_file, start, size,
1504			prot, flags, 0, pgoff, &populate, NULL);
1505out:
1506	mmap_write_unlock(mm);
1507	fput(file);
1508	if (populate)
1509		mm_populate(ret, populate);
1510	if (!IS_ERR_VALUE(ret))
1511		ret = 0;
1512	return ret;
1513}
1514
1515/*
1516 * do_brk_flags() - Increase the brk vma if the flags match.
1517 * @vmi: The vma iterator
1518 * @addr: The start address
1519 * @len: The length of the increase
1520 * @vma: The vma,
1521 * @flags: The VMA Flags
1522 *
1523 * Extend the brk VMA from addr to addr + len.  If the VMA is NULL or the flags
1524 * do not match then create a new anonymous VMA.  Eventually we may be able to
1525 * do some brk-specific accounting here.
1526 */
1527static int do_brk_flags(struct vma_iterator *vmi, struct vm_area_struct *vma,
1528		unsigned long addr, unsigned long len, unsigned long flags)
1529{
1530	struct mm_struct *mm = current->mm;
1531
1532	/*
1533	 * Check against address space limits by the changed size
1534	 * Note: This happens *after* clearing old mappings in some code paths.
1535	 */
1536	flags |= VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
1537	if (!may_expand_vm(mm, flags, len >> PAGE_SHIFT))
 
 
 
 
 
 
 
 
1538		return -ENOMEM;
1539
1540	if (mm->map_count > sysctl_max_map_count)
1541		return -ENOMEM;
1542
1543	if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT))
1544		return -ENOMEM;
1545
 
 
 
 
 
 
1546	/*
1547	 * Expand the existing vma if possible; Note that singular lists do not
1548	 * occur after forking, so the expand will only happen on new VMAs.
1549	 */
1550	if (vma && vma->vm_end == addr) {
1551		VMG_STATE(vmg, mm, vmi, addr, addr + len, flags, PHYS_PFN(addr));
1552
1553		vmg.prev = vma;
1554		/* vmi is positioned at prev, which this mode expects. */
1555		vmg.merge_flags = VMG_FLAG_JUST_EXPAND;
1556
1557		if (vma_merge_new_range(&vmg))
1558			goto out;
1559		else if (vmg_nomem(&vmg))
1560			goto unacct_fail;
1561	}
1562
1563	if (vma)
1564		vma_iter_next_range(vmi);
1565	/* create a vma struct for an anonymous mapping */
1566	vma = vm_area_alloc(mm);
1567	if (!vma)
1568		goto unacct_fail;
1569
1570	vma_set_anonymous(vma);
1571	vma_set_range(vma, addr, addr + len, addr >> PAGE_SHIFT);
1572	vm_flags_init(vma, flags);
1573	vma->vm_page_prot = vm_get_page_prot(flags);
1574	vma_start_write(vma);
1575	if (vma_iter_store_gfp(vmi, vma, GFP_KERNEL))
1576		goto mas_store_fail;
1577
1578	mm->map_count++;
1579	validate_mm(mm);
1580	ksm_add_vma(vma);
1581out:
1582	perf_event_mmap(vma);
1583	mm->total_vm += len >> PAGE_SHIFT;
1584	mm->data_vm += len >> PAGE_SHIFT;
1585	if (flags & VM_LOCKED)
1586		mm->locked_vm += (len >> PAGE_SHIFT);
1587	vm_flags_set(vma, VM_SOFTDIRTY);
1588	return 0;
1589
1590mas_store_fail:
1591	vm_area_free(vma);
1592unacct_fail:
1593	vm_unacct_memory(len >> PAGE_SHIFT);
1594	return -ENOMEM;
1595}
1596
1597int vm_brk_flags(unsigned long addr, unsigned long request, unsigned long flags)
1598{
1599	struct mm_struct *mm = current->mm;
1600	struct vm_area_struct *vma = NULL;
1601	unsigned long len;
1602	int ret;
1603	bool populate;
1604	LIST_HEAD(uf);
1605	VMA_ITERATOR(vmi, mm, addr);
1606
1607	len = PAGE_ALIGN(request);
1608	if (len < request)
1609		return -ENOMEM;
1610	if (!len)
1611		return 0;
1612
1613	/* Until we need other flags, refuse anything except VM_EXEC. */
1614	if ((flags & (~VM_EXEC)) != 0)
1615		return -EINVAL;
1616
1617	if (mmap_write_lock_killable(mm))
1618		return -EINTR;
1619
1620	ret = check_brk_limits(addr, len);
1621	if (ret)
1622		goto limits_failed;
1623
1624	ret = do_vmi_munmap(&vmi, mm, addr, len, &uf, 0);
1625	if (ret)
1626		goto munmap_failed;
1627
1628	vma = vma_prev(&vmi);
1629	ret = do_brk_flags(&vmi, vma, addr, len, flags);
1630	populate = ((mm->def_flags & VM_LOCKED) != 0);
1631	mmap_write_unlock(mm);
1632	userfaultfd_unmap_complete(mm, &uf);
1633	if (populate && !ret)
1634		mm_populate(addr, len);
1635	return ret;
1636
1637munmap_failed:
1638limits_failed:
1639	mmap_write_unlock(mm);
1640	return ret;
1641}
1642EXPORT_SYMBOL(vm_brk_flags);
1643
1644/* Release all mmaps. */
1645void exit_mmap(struct mm_struct *mm)
1646{
1647	struct mmu_gather tlb;
1648	struct vm_area_struct *vma;
1649	unsigned long nr_accounted = 0;
1650	VMA_ITERATOR(vmi, mm, 0);
1651	int count = 0;
1652
1653	/* mm's last user has gone, and its about to be pulled down */
1654	mmu_notifier_release(mm);
1655
1656	mmap_read_lock(mm);
 
 
 
 
 
 
 
 
1657	arch_exit_mmap(mm);
1658
1659	vma = vma_next(&vmi);
1660	if (!vma || unlikely(xa_is_zero(vma))) {
1661		/* Can happen if dup_mmap() received an OOM */
1662		mmap_read_unlock(mm);
1663		mmap_write_lock(mm);
1664		goto destroy;
1665	}
1666
1667	lru_add_drain();
1668	flush_cache_mm(mm);
1669	tlb_gather_mmu_fullmm(&tlb, mm);
1670	/* update_hiwater_rss(mm) here? but nobody should be looking */
1671	/* Use ULONG_MAX here to ensure all VMAs in the mm are unmapped */
1672	unmap_vmas(&tlb, &vmi.mas, vma, 0, ULONG_MAX, ULONG_MAX, false);
1673	mmap_read_unlock(mm);
 
 
 
1674
1675	/*
1676	 * Set MMF_OOM_SKIP to hide this task from the oom killer/reaper
1677	 * because the memory has been already freed.
1678	 */
1679	set_bit(MMF_OOM_SKIP, &mm->flags);
1680	mmap_write_lock(mm);
1681	mt_clear_in_rcu(&mm->mm_mt);
1682	vma_iter_set(&vmi, vma->vm_end);
1683	free_pgtables(&tlb, &vmi.mas, vma, FIRST_USER_ADDRESS,
1684		      USER_PGTABLES_CEILING, true);
1685	tlb_finish_mmu(&tlb);
1686
1687	/*
1688	 * Walk the list again, actually closing and freeing it, with preemption
1689	 * enabled, without holding any MM locks besides the unreachable
1690	 * mmap_write_lock.
1691	 */
1692	vma_iter_set(&vmi, vma->vm_end);
1693	do {
1694		if (vma->vm_flags & VM_ACCOUNT)
1695			nr_accounted += vma_pages(vma);
1696		remove_vma(vma, /* unreachable = */ true);
1697		count++;
1698		cond_resched();
1699		vma = vma_next(&vmi);
1700	} while (vma && likely(!xa_is_zero(vma)));
1701
1702	BUG_ON(count != mm->map_count);
1703
1704	trace_exit_mmap(mm);
1705destroy:
1706	__mt_destroy(&mm->mm_mt);
1707	mmap_write_unlock(mm);
1708	vm_unacct_memory(nr_accounted);
1709}
1710
1711/* Insert vm structure into process list sorted by address
1712 * and into the inode's i_mmap tree.  If vm_file is non-NULL
1713 * then i_mmap_rwsem is taken here.
1714 */
1715int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
1716{
1717	unsigned long charged = vma_pages(vma);
1718
1719
1720	if (find_vma_intersection(mm, vma->vm_start, vma->vm_end))
1721		return -ENOMEM;
1722
1723	if ((vma->vm_flags & VM_ACCOUNT) &&
1724	     security_vm_enough_memory_mm(mm, charged))
1725		return -ENOMEM;
1726
1727	/*
1728	 * The vm_pgoff of a purely anonymous vma should be irrelevant
1729	 * until its first write fault, when page's anon_vma and index
1730	 * are set.  But now set the vm_pgoff it will almost certainly
1731	 * end up with (unless mremap moves it elsewhere before that
1732	 * first wfault), so /proc/pid/maps tells a consistent story.
1733	 *
1734	 * By setting it to reflect the virtual start address of the
1735	 * vma, merges and splits can happen in a seamless way, just
1736	 * using the existing file pgoff checks and manipulations.
1737	 * Similarly in do_mmap and in do_brk_flags.
1738	 */
1739	if (vma_is_anonymous(vma)) {
1740		BUG_ON(vma->anon_vma);
1741		vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
1742	}
1743
1744	if (vma_link(mm, vma)) {
1745		if (vma->vm_flags & VM_ACCOUNT)
1746			vm_unacct_memory(charged);
 
1747		return -ENOMEM;
1748	}
1749
1750	return 0;
1751}
1752
1753/*
1754 * Return true if the calling process may expand its vm space by the passed
1755 * number of pages
1756 */
1757bool may_expand_vm(struct mm_struct *mm, vm_flags_t flags, unsigned long npages)
 
1758{
1759	if (mm->total_vm + npages > rlimit(RLIMIT_AS) >> PAGE_SHIFT)
1760		return false;
 
 
 
 
1761
1762	if (is_data_mapping(flags) &&
1763	    mm->data_vm + npages > rlimit(RLIMIT_DATA) >> PAGE_SHIFT) {
1764		/* Workaround for Valgrind */
1765		if (rlimit(RLIMIT_DATA) == 0 &&
1766		    mm->data_vm + npages <= rlimit_max(RLIMIT_DATA) >> PAGE_SHIFT)
1767			return true;
1768
1769		pr_warn_once("%s (%d): VmData %lu exceed data ulimit %lu. Update limits%s.\n",
1770			     current->comm, current->pid,
1771			     (mm->data_vm + npages) << PAGE_SHIFT,
1772			     rlimit(RLIMIT_DATA),
1773			     ignore_rlimit_data ? "" : " or use boot option ignore_rlimit_data");
1774
1775		if (!ignore_rlimit_data)
1776			return false;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1777	}
 
1778
1779	return true;
1780}
1781
1782void vm_stat_account(struct mm_struct *mm, vm_flags_t flags, long npages)
1783{
1784	WRITE_ONCE(mm->total_vm, READ_ONCE(mm->total_vm)+npages);
1785
1786	if (is_exec_mapping(flags))
1787		mm->exec_vm += npages;
1788	else if (is_stack_mapping(flags))
1789		mm->stack_vm += npages;
1790	else if (is_data_mapping(flags))
1791		mm->data_vm += npages;
1792}
1793
1794static vm_fault_t special_mapping_fault(struct vm_fault *vmf);
1795
1796/*
1797 * Close hook, called for unmap() and on the old vma for mremap().
1798 *
1799 * Having a close hook prevents vma merging regardless of flags.
1800 */
1801static void special_mapping_close(struct vm_area_struct *vma)
1802{
1803	const struct vm_special_mapping *sm = vma->vm_private_data;
 
1804
1805	if (sm->close)
1806		sm->close(sm, vma);
1807}
1808
1809static const char *special_mapping_name(struct vm_area_struct *vma)
1810{
1811	return ((struct vm_special_mapping *)vma->vm_private_data)->name;
1812}
1813
1814static int special_mapping_mremap(struct vm_area_struct *new_vma)
1815{
1816	struct vm_special_mapping *sm = new_vma->vm_private_data;
1817
1818	if (WARN_ON_ONCE(current->mm != new_vma->vm_mm))
1819		return -EFAULT;
1820
1821	if (sm->mremap)
1822		return sm->mremap(sm, new_vma);
1823
1824	return 0;
1825}
1826
1827static int special_mapping_split(struct vm_area_struct *vma, unsigned long addr)
1828{
1829	/*
1830	 * Forbid splitting special mappings - kernel has expectations over
1831	 * the number of pages in mapping. Together with VM_DONTEXPAND
1832	 * the size of vma should stay the same over the special mapping's
1833	 * lifetime.
1834	 */
1835	return -EINVAL;
1836}
1837
1838static const struct vm_operations_struct special_mapping_vmops = {
1839	.close = special_mapping_close,
1840	.fault = special_mapping_fault,
1841	.mremap = special_mapping_mremap,
1842	.name = special_mapping_name,
1843	/* vDSO code relies that VVAR can't be accessed remotely */
1844	.access = NULL,
1845	.may_split = special_mapping_split,
1846};
1847
1848static vm_fault_t special_mapping_fault(struct vm_fault *vmf)
 
1849{
1850	struct vm_area_struct *vma = vmf->vma;
1851	pgoff_t pgoff;
1852	struct page **pages;
1853	struct vm_special_mapping *sm = vma->vm_private_data;
1854
1855	if (sm->fault)
1856		return sm->fault(sm, vmf->vma, vmf);
1857
1858	pages = sm->pages;
 
 
 
1859
1860	for (pgoff = vmf->pgoff; pgoff && *pages; ++pages)
1861		pgoff--;
1862
1863	if (*pages) {
1864		struct page *page = *pages;
1865		get_page(page);
1866		vmf->page = page;
1867		return 0;
1868	}
1869
1870	return VM_FAULT_SIGBUS;
1871}
1872
1873static struct vm_area_struct *__install_special_mapping(
1874	struct mm_struct *mm,
1875	unsigned long addr, unsigned long len,
1876	unsigned long vm_flags, void *priv,
1877	const struct vm_operations_struct *ops)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1878{
1879	int ret;
1880	struct vm_area_struct *vma;
1881
1882	vma = vm_area_alloc(mm);
1883	if (unlikely(vma == NULL))
1884		return ERR_PTR(-ENOMEM);
1885
1886	vma_set_range(vma, addr, addr + len, 0);
1887	vm_flags_init(vma, (vm_flags | mm->def_flags |
1888		      VM_DONTEXPAND | VM_SOFTDIRTY) & ~VM_LOCKED_MASK);
 
 
 
1889	vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
1890
1891	vma->vm_ops = ops;
1892	vma->vm_private_data = priv;
 
 
 
 
1893
1894	ret = insert_vm_struct(mm, vma);
1895	if (ret)
1896		goto out;
1897
1898	vm_stat_account(mm, vma->vm_flags, len >> PAGE_SHIFT);
1899
1900	perf_event_mmap(vma);
1901
1902	return vma;
1903
1904out:
1905	vm_area_free(vma);
1906	return ERR_PTR(ret);
1907}
1908
1909bool vma_is_special_mapping(const struct vm_area_struct *vma,
1910	const struct vm_special_mapping *sm)
1911{
1912	return vma->vm_private_data == sm &&
1913		vma->vm_ops == &special_mapping_vmops;
1914}
1915
1916/*
1917 * Called with mm->mmap_lock held for writing.
1918 * Insert a new vma covering the given region, with the given flags.
1919 * Its pages are supplied by the given array of struct page *.
1920 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
1921 * The region past the last page supplied will always produce SIGBUS.
1922 * The array pointer and the pages it points to are assumed to stay alive
1923 * for as long as this mapping might exist.
1924 */
1925struct vm_area_struct *_install_special_mapping(
1926	struct mm_struct *mm,
1927	unsigned long addr, unsigned long len,
1928	unsigned long vm_flags, const struct vm_special_mapping *spec)
1929{
1930	return __install_special_mapping(mm, addr, len, vm_flags, (void *)spec,
1931					&special_mapping_vmops);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1932}
1933
1934/*
1935 * initialise the percpu counter for VM
1936 */
1937void __init mmap_init(void)
1938{
1939	int ret;
1940
1941	ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL);
1942	VM_BUG_ON(ret);
 
 
 
 
 
 
 
 
 
 
1943}
1944
1945/*
1946 * Initialise sysctl_user_reserve_kbytes.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1947 *
1948 * This is intended to prevent a user from starting a single memory hogging
1949 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
1950 * mode.
1951 *
1952 * The default value is min(3% of free memory, 128MB)
1953 * 128MB is enough to recover with sshd/login, bash, and top/kill.
1954 */
1955static int init_user_reserve(void)
1956{
1957	unsigned long free_kbytes;
1958
1959	free_kbytes = K(global_zone_page_state(NR_FREE_PAGES));
1960
1961	sysctl_user_reserve_kbytes = min(free_kbytes / 32, SZ_128K);
1962	return 0;
1963}
1964subsys_initcall(init_user_reserve);
1965
1966/*
1967 * Initialise sysctl_admin_reserve_kbytes.
1968 *
1969 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
1970 * to log in and kill a memory hogging process.
1971 *
1972 * Systems with more than 256MB will reserve 8MB, enough to recover
1973 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
1974 * only reserve 3% of free pages by default.
1975 */
1976static int init_admin_reserve(void)
1977{
1978	unsigned long free_kbytes;
 
 
1979
1980	free_kbytes = K(global_zone_page_state(NR_FREE_PAGES));
1981
1982	sysctl_admin_reserve_kbytes = min(free_kbytes / 32, SZ_8K);
1983	return 0;
1984}
1985subsys_initcall(init_admin_reserve);
1986
1987/*
1988 * Reinititalise user and admin reserves if memory is added or removed.
1989 *
1990 * The default user reserve max is 128MB, and the default max for the
1991 * admin reserve is 8MB. These are usually, but not always, enough to
1992 * enable recovery from a memory hogging process using login/sshd, a shell,
1993 * and tools like top. It may make sense to increase or even disable the
1994 * reserve depending on the existence of swap or variations in the recovery
1995 * tools. So, the admin may have changed them.
1996 *
1997 * If memory is added and the reserves have been eliminated or increased above
1998 * the default max, then we'll trust the admin.
1999 *
2000 * If memory is removed and there isn't enough free memory, then we
2001 * need to reset the reserves.
2002 *
2003 * Otherwise keep the reserve set by the admin.
2004 */
2005static int reserve_mem_notifier(struct notifier_block *nb,
2006			     unsigned long action, void *data)
2007{
2008	unsigned long tmp, free_kbytes;
2009
2010	switch (action) {
2011	case MEM_ONLINE:
2012		/* Default max is 128MB. Leave alone if modified by operator. */
2013		tmp = sysctl_user_reserve_kbytes;
2014		if (tmp > 0 && tmp < SZ_128K)
2015			init_user_reserve();
 
2016
2017		/* Default max is 8MB.  Leave alone if modified by operator. */
2018		tmp = sysctl_admin_reserve_kbytes;
2019		if (tmp > 0 && tmp < SZ_8K)
2020			init_admin_reserve();
2021
2022		break;
2023	case MEM_OFFLINE:
2024		free_kbytes = K(global_zone_page_state(NR_FREE_PAGES));
2025
2026		if (sysctl_user_reserve_kbytes > free_kbytes) {
2027			init_user_reserve();
2028			pr_info("vm.user_reserve_kbytes reset to %lu\n",
2029				sysctl_user_reserve_kbytes);
2030		}
2031
2032		if (sysctl_admin_reserve_kbytes > free_kbytes) {
2033			init_admin_reserve();
2034			pr_info("vm.admin_reserve_kbytes reset to %lu\n",
2035				sysctl_admin_reserve_kbytes);
2036		}
2037		break;
2038	default:
2039		break;
 
 
 
 
 
 
 
 
 
 
 
2040	}
2041	return NOTIFY_OK;
2042}
2043
2044static int __meminit init_reserve_notifier(void)
2045{
2046	if (hotplug_memory_notifier(reserve_mem_notifier, DEFAULT_CALLBACK_PRI))
2047		pr_err("Failed registering memory add/remove notifier for admin reserve\n");
2048
2049	return 0;
 
 
 
 
 
 
2050}
2051subsys_initcall(init_reserve_notifier);
2052
2053/*
2054 * Relocate a VMA downwards by shift bytes. There cannot be any VMAs between
2055 * this VMA and its relocated range, which will now reside at [vma->vm_start -
2056 * shift, vma->vm_end - shift).
2057 *
2058 * This function is almost certainly NOT what you want for anything other than
2059 * early executable temporary stack relocation.
2060 */
2061int relocate_vma_down(struct vm_area_struct *vma, unsigned long shift)
2062{
2063	/*
2064	 * The process proceeds as follows:
2065	 *
2066	 * 1) Use shift to calculate the new vma endpoints.
2067	 * 2) Extend vma to cover both the old and new ranges.  This ensures the
2068	 *    arguments passed to subsequent functions are consistent.
2069	 * 3) Move vma's page tables to the new range.
2070	 * 4) Free up any cleared pgd range.
2071	 * 5) Shrink the vma to cover only the new range.
2072	 */
2073
2074	struct mm_struct *mm = vma->vm_mm;
2075	unsigned long old_start = vma->vm_start;
2076	unsigned long old_end = vma->vm_end;
2077	unsigned long length = old_end - old_start;
2078	unsigned long new_start = old_start - shift;
2079	unsigned long new_end = old_end - shift;
2080	VMA_ITERATOR(vmi, mm, new_start);
2081	VMG_STATE(vmg, mm, &vmi, new_start, old_end, 0, vma->vm_pgoff);
2082	struct vm_area_struct *next;
2083	struct mmu_gather tlb;
2084
2085	BUG_ON(new_start > new_end);
 
 
 
 
 
 
2086
2087	/*
2088	 * ensure there are no vmas between where we want to go
2089	 * and where we are
2090	 */
2091	if (vma != vma_next(&vmi))
2092		return -EFAULT;
2093
2094	vma_iter_prev_range(&vmi);
2095	/*
2096	 * cover the whole range: [new_start, old_end)
2097	 */
2098	vmg.vma = vma;
2099	if (vma_expand(&vmg))
2100		return -ENOMEM;
2101
2102	/*
2103	 * move the page tables downwards, on failure we rely on
2104	 * process cleanup to remove whatever mess we made.
2105	 */
2106	if (length != move_page_tables(vma, old_start,
2107				       vma, new_start, length, false, true))
2108		return -ENOMEM;
2109
2110	lru_add_drain();
2111	tlb_gather_mmu(&tlb, mm);
2112	next = vma_next(&vmi);
2113	if (new_end > old_start) {
2114		/*
2115		 * when the old and new regions overlap clear from new_end.
2116		 */
2117		free_pgd_range(&tlb, new_end, old_end, new_end,
2118			next ? next->vm_start : USER_PGTABLES_CEILING);
2119	} else {
2120		/*
2121		 * otherwise, clean from old_start; this is done to not touch
2122		 * the address space in [new_end, old_start) some architectures
2123		 * have constraints on va-space that make this illegal (IA64) -
2124		 * for the others its just a little faster.
2125		 */
2126		free_pgd_range(&tlb, old_start, old_end, new_end,
2127			next ? next->vm_start : USER_PGTABLES_CEILING);
2128	}
2129	tlb_finish_mmu(&tlb);
2130
2131	vma_prev(&vmi);
2132	/* Shrink the vma to just the new range */
2133	return vma_shrink(&vmi, vma, new_start, new_end, vma->vm_pgoff);
2134}