Linux Audio

Check our new training course

Loading...
v6.2
   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
  50#include <linux/uaccess.h>
  51#include <asm/cacheflush.h>
  52#include <asm/tlb.h>
  53#include <asm/mmu_context.h>
  54
  55#define CREATE_TRACE_POINTS
  56#include <trace/events/mmap.h>
  57
  58#include "internal.h"
  59
  60#ifndef arch_mmap_check
  61#define arch_mmap_check(addr, len, flags)	(0)
  62#endif
  63
  64#ifdef CONFIG_HAVE_ARCH_MMAP_RND_BITS
  65const int mmap_rnd_bits_min = CONFIG_ARCH_MMAP_RND_BITS_MIN;
  66const int mmap_rnd_bits_max = CONFIG_ARCH_MMAP_RND_BITS_MAX;
  67int mmap_rnd_bits __read_mostly = CONFIG_ARCH_MMAP_RND_BITS;
  68#endif
  69#ifdef CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS
  70const int mmap_rnd_compat_bits_min = CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MIN;
  71const int mmap_rnd_compat_bits_max = CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MAX;
  72int mmap_rnd_compat_bits __read_mostly = CONFIG_ARCH_MMAP_RND_COMPAT_BITS;
  73#endif
  74
  75static bool ignore_rlimit_data;
  76core_param(ignore_rlimit_data, ignore_rlimit_data, bool, 0644);
  77
  78static void unmap_region(struct mm_struct *mm, struct maple_tree *mt,
  79		struct vm_area_struct *vma, struct vm_area_struct *prev,
  80		struct vm_area_struct *next, unsigned long start,
  81		unsigned long end);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  82
  83static pgprot_t vm_pgprot_modify(pgprot_t oldprot, unsigned long vm_flags)
  84{
  85	return pgprot_modify(oldprot, vm_get_page_prot(vm_flags));
 
 
  86}
 
  87
  88/* Update vma->vm_page_prot to reflect vma->vm_flags. */
  89void vma_set_page_prot(struct vm_area_struct *vma)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  90{
  91	unsigned long vm_flags = vma->vm_flags;
  92	pgprot_t vm_page_prot;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  93
  94	vm_page_prot = vm_pgprot_modify(vma->vm_page_prot, vm_flags);
  95	if (vma_wants_writenotify(vma, vm_page_prot)) {
  96		vm_flags &= ~VM_SHARED;
  97		vm_page_prot = vm_pgprot_modify(vm_page_prot, vm_flags);
 
 
 
 
 
 
  98	}
  99	/* remove_protection_ptes reads vma->vm_page_prot without mmap_lock */
 100	WRITE_ONCE(vma->vm_page_prot, vm_page_prot);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 101}
 102
 103/*
 104 * Requires inode->i_mapping->i_mmap_rwsem
 105 */
 106static void __remove_shared_vm_struct(struct vm_area_struct *vma,
 107		struct file *file, struct address_space *mapping)
 108{
 
 
 109	if (vma->vm_flags & VM_SHARED)
 110		mapping_unmap_writable(mapping);
 111
 112	flush_dcache_mmap_lock(mapping);
 113	vma_interval_tree_remove(vma, &mapping->i_mmap);
 
 
 
 114	flush_dcache_mmap_unlock(mapping);
 115}
 116
 117/*
 118 * Unlink a file-based vm structure from its interval tree, to hide
 119 * vma from rmap and vmtruncate before freeing its page tables.
 120 */
 121void unlink_file_vma(struct vm_area_struct *vma)
 122{
 123	struct file *file = vma->vm_file;
 124
 125	if (file) {
 126		struct address_space *mapping = file->f_mapping;
 127		i_mmap_lock_write(mapping);
 128		__remove_shared_vm_struct(vma, file, mapping);
 129		i_mmap_unlock_write(mapping);
 130	}
 131}
 132
 133/*
 134 * Close a vm structure and free it.
 135 */
 136static void remove_vma(struct vm_area_struct *vma)
 137{
 
 
 138	might_sleep();
 139	if (vma->vm_ops && vma->vm_ops->close)
 140		vma->vm_ops->close(vma);
 141	if (vma->vm_file)
 142		fput(vma->vm_file);
 
 
 
 143	mpol_put(vma_policy(vma));
 144	vm_area_free(vma);
 
 145}
 146
 147/*
 148 * check_brk_limits() - Use platform specific check of range & verify mlock
 149 * limits.
 150 * @addr: The address to check
 151 * @len: The size of increase.
 152 *
 153 * Return: 0 on success.
 154 */
 155static int check_brk_limits(unsigned long addr, unsigned long len)
 156{
 157	unsigned long mapped_addr;
 158
 159	mapped_addr = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED);
 160	if (IS_ERR_VALUE(mapped_addr))
 161		return mapped_addr;
 162
 163	return mlock_future_check(current->mm, current->mm->def_flags, len);
 164}
 165static int do_brk_munmap(struct ma_state *mas, struct vm_area_struct *vma,
 166			 unsigned long newbrk, unsigned long oldbrk,
 167			 struct list_head *uf);
 168static int do_brk_flags(struct ma_state *mas, struct vm_area_struct *brkvma,
 169		unsigned long addr, unsigned long request, unsigned long flags);
 170SYSCALL_DEFINE1(brk, unsigned long, brk)
 171{
 172	unsigned long newbrk, oldbrk, origbrk;
 
 173	struct mm_struct *mm = current->mm;
 174	struct vm_area_struct *brkvma, *next = NULL;
 175	unsigned long min_brk;
 176	bool populate;
 177	bool downgraded = false;
 178	LIST_HEAD(uf);
 179	MA_STATE(mas, &mm->mm_mt, 0, 0);
 180
 181	if (mmap_write_lock_killable(mm))
 182		return -EINTR;
 183
 184	origbrk = mm->brk;
 185
 186#ifdef CONFIG_COMPAT_BRK
 187	/*
 188	 * CONFIG_COMPAT_BRK can still be overridden by setting
 189	 * randomize_va_space to 2, which will still cause mm->start_brk
 190	 * to be arbitrarily shifted
 191	 */
 192	if (current->brk_randomized)
 193		min_brk = mm->start_brk;
 194	else
 195		min_brk = mm->end_data;
 196#else
 197	min_brk = mm->start_brk;
 198#endif
 199	if (brk < min_brk)
 200		goto out;
 201
 202	/*
 203	 * Check against rlimit here. If this check is done later after the test
 204	 * of oldbrk with newbrk then it can escape the test and let the data
 205	 * segment grow beyond its set limit the in case where the limit is
 206	 * not page aligned -Ram Gupta
 207	 */
 208	if (check_data_rlimit(rlimit(RLIMIT_DATA), brk, mm->start_brk,
 209			      mm->end_data, mm->start_data))
 
 210		goto out;
 211
 212	newbrk = PAGE_ALIGN(brk);
 213	oldbrk = PAGE_ALIGN(mm->brk);
 214	if (oldbrk == newbrk) {
 215		mm->brk = brk;
 216		goto success;
 217	}
 218
 219	/*
 220	 * Always allow shrinking brk.
 221	 * do_brk_munmap() may downgrade mmap_lock to read.
 222	 */
 223	if (brk <= mm->brk) {
 224		int ret;
 225
 226		/* Search one past newbrk */
 227		mas_set(&mas, newbrk);
 228		brkvma = mas_find(&mas, oldbrk);
 229		if (!brkvma || brkvma->vm_start >= oldbrk)
 230			goto out; /* mapping intersects with an existing non-brk vma. */
 231		/*
 232		 * mm->brk must be protected by write mmap_lock.
 233		 * do_brk_munmap() may downgrade the lock,  so update it
 234		 * before calling do_brk_munmap().
 235		 */
 236		mm->brk = brk;
 237		ret = do_brk_munmap(&mas, brkvma, newbrk, oldbrk, &uf);
 238		if (ret == 1)  {
 239			downgraded = true;
 240			goto success;
 241		} else if (!ret)
 242			goto success;
 243
 244		mm->brk = origbrk;
 245		goto out;
 246	}
 247
 248	if (check_brk_limits(oldbrk, newbrk - oldbrk))
 249		goto out;
 250
 251	/*
 252	 * Only check if the next VMA is within the stack_guard_gap of the
 253	 * expansion area
 254	 */
 255	mas_set(&mas, oldbrk);
 256	next = mas_find(&mas, newbrk - 1 + PAGE_SIZE + stack_guard_gap);
 257	if (next && newbrk + PAGE_SIZE > vm_start_gap(next))
 258		goto out;
 259
 260	brkvma = mas_prev(&mas, mm->start_brk);
 261	/* Ok, looks good - let it rip. */
 262	if (do_brk_flags(&mas, brkvma, oldbrk, newbrk - oldbrk, 0) < 0)
 263		goto out;
 264
 265	mm->brk = brk;
 266
 267success:
 268	populate = newbrk > oldbrk && (mm->def_flags & VM_LOCKED) != 0;
 269	if (downgraded)
 270		mmap_read_unlock(mm);
 271	else
 272		mmap_write_unlock(mm);
 273	userfaultfd_unmap_complete(mm, &uf);
 274	if (populate)
 275		mm_populate(oldbrk, newbrk - oldbrk);
 276	return brk;
 277
 278out:
 279	mmap_write_unlock(mm);
 280	return origbrk;
 
 281}
 282
 283#if defined(CONFIG_DEBUG_VM_MAPLE_TREE)
 284extern void mt_validate(struct maple_tree *mt);
 285extern void mt_dump(const struct maple_tree *mt);
 286
 287/* Validate the maple tree */
 288static void validate_mm_mt(struct mm_struct *mm)
 289{
 290	struct maple_tree *mt = &mm->mm_mt;
 291	struct vm_area_struct *vma_mt;
 292
 293	MA_STATE(mas, mt, 0, 0);
 294
 295	mt_validate(&mm->mm_mt);
 296	mas_for_each(&mas, vma_mt, ULONG_MAX) {
 297		if ((vma_mt->vm_start != mas.index) ||
 298		    (vma_mt->vm_end - 1 != mas.last)) {
 299			pr_emerg("issue in %s\n", current->comm);
 300			dump_stack();
 301			dump_vma(vma_mt);
 302			pr_emerg("mt piv: %p %lu - %lu\n", vma_mt,
 303				 mas.index, mas.last);
 304			pr_emerg("mt vma: %p %lu - %lu\n", vma_mt,
 305				 vma_mt->vm_start, vma_mt->vm_end);
 306
 307			mt_dump(mas.tree);
 308			if (vma_mt->vm_end != mas.last + 1) {
 309				pr_err("vma: %p vma_mt %lu-%lu\tmt %lu-%lu\n",
 310						mm, vma_mt->vm_start, vma_mt->vm_end,
 311						mas.index, mas.last);
 312				mt_dump(mas.tree);
 313			}
 314			VM_BUG_ON_MM(vma_mt->vm_end != mas.last + 1, mm);
 315			if (vma_mt->vm_start != mas.index) {
 316				pr_err("vma: %p vma_mt %p %lu - %lu doesn't match\n",
 317						mm, vma_mt, vma_mt->vm_start, vma_mt->vm_end);
 318				mt_dump(mas.tree);
 319			}
 320			VM_BUG_ON_MM(vma_mt->vm_start != mas.index, mm);
 321		}
 322	}
 
 
 
 323}
 324
 325static void validate_mm(struct mm_struct *mm)
 326{
 327	int bug = 0;
 328	int i = 0;
 329	struct vm_area_struct *vma;
 330	MA_STATE(mas, &mm->mm_mt, 0, 0);
 331
 332	validate_mm_mt(mm);
 333
 334	mas_for_each(&mas, vma, ULONG_MAX) {
 335#ifdef CONFIG_DEBUG_VM_RB
 336		struct anon_vma *anon_vma = vma->anon_vma;
 337		struct anon_vma_chain *avc;
 338
 339		if (anon_vma) {
 340			anon_vma_lock_read(anon_vma);
 341			list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
 342				anon_vma_interval_tree_verify(avc);
 343			anon_vma_unlock_read(anon_vma);
 344		}
 345#endif
 346		i++;
 347	}
 348	if (i != mm->map_count) {
 349		pr_emerg("map_count %d mas_for_each %d\n", mm->map_count, i);
 350		bug = 1;
 351	}
 352	VM_BUG_ON_MM(bug, mm);
 
 353}
 354
 355#else /* !CONFIG_DEBUG_VM_MAPLE_TREE */
 356#define validate_mm_mt(root) do { } while (0)
 357#define validate_mm(mm) do { } while (0)
 358#endif /* CONFIG_DEBUG_VM_MAPLE_TREE */
 359
 360/*
 361 * vma has some anon_vma assigned, and is already inserted on that
 362 * anon_vma's interval trees.
 363 *
 364 * Before updating the vma's vm_start / vm_end / vm_pgoff fields, the
 365 * vma must be removed from the anon_vma's interval trees using
 366 * anon_vma_interval_tree_pre_update_vma().
 367 *
 368 * After the update, the vma will be reinserted using
 369 * anon_vma_interval_tree_post_update_vma().
 370 *
 371 * The entire update must be protected by exclusive mmap_lock and by
 372 * the root anon_vma's mutex.
 373 */
 374static inline void
 375anon_vma_interval_tree_pre_update_vma(struct vm_area_struct *vma)
 376{
 377	struct anon_vma_chain *avc;
 378
 379	list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
 380		anon_vma_interval_tree_remove(avc, &avc->anon_vma->rb_root);
 381}
 382
 383static inline void
 384anon_vma_interval_tree_post_update_vma(struct vm_area_struct *vma)
 385{
 386	struct anon_vma_chain *avc;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 387
 388	list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
 389		anon_vma_interval_tree_insert(avc, &avc->anon_vma->rb_root);
 
 
 
 
 390}
 391
 392static unsigned long count_vma_pages_range(struct mm_struct *mm,
 393		unsigned long addr, unsigned long end)
 394{
 395	VMA_ITERATOR(vmi, mm, addr);
 396	struct vm_area_struct *vma;
 397	unsigned long nr_pages = 0;
 398
 399	for_each_vma_range(vmi, vma, end) {
 400		unsigned long vm_start = max(addr, vma->vm_start);
 401		unsigned long vm_end = min(end, vma->vm_end);
 402
 403		nr_pages += PHYS_PFN(vm_end - vm_start);
 404	}
 405
 406	return nr_pages;
 407}
 408
 409static void __vma_link_file(struct vm_area_struct *vma,
 410			    struct address_space *mapping)
 411{
 412	if (vma->vm_flags & VM_SHARED)
 413		mapping_allow_writable(mapping);
 414
 415	flush_dcache_mmap_lock(mapping);
 416	vma_interval_tree_insert(vma, &mapping->i_mmap);
 417	flush_dcache_mmap_unlock(mapping);
 418}
 419
 420/*
 421 * vma_mas_store() - Store a VMA in the maple tree.
 422 * @vma: The vm_area_struct
 423 * @mas: The maple state
 424 *
 425 * Efficient way to store a VMA in the maple tree when the @mas has already
 426 * walked to the correct location.
 427 *
 428 * Note: the end address is inclusive in the maple tree.
 429 */
 430void vma_mas_store(struct vm_area_struct *vma, struct ma_state *mas)
 431{
 432	trace_vma_store(mas->tree, vma);
 433	mas_set_range(mas, vma->vm_start, vma->vm_end - 1);
 434	mas_store_prealloc(mas, vma);
 435}
 436
 437/*
 438 * vma_mas_remove() - Remove a VMA from the maple tree.
 439 * @vma: The vm_area_struct
 440 * @mas: The maple state
 441 *
 442 * Efficient way to remove a VMA from the maple tree when the @mas has already
 443 * been established and points to the correct location.
 444 * Note: the end address is inclusive in the maple tree.
 445 */
 446void vma_mas_remove(struct vm_area_struct *vma, struct ma_state *mas)
 447{
 448	trace_vma_mas_szero(mas->tree, vma->vm_start, vma->vm_end - 1);
 449	mas->index = vma->vm_start;
 450	mas->last = vma->vm_end - 1;
 451	mas_store_prealloc(mas, NULL);
 452}
 453
 454/*
 455 * vma_mas_szero() - Set a given range to zero.  Used when modifying a
 456 * vm_area_struct start or end.
 457 *
 458 * @mas: The maple tree ma_state
 459 * @start: The start address to zero
 460 * @end: The end address to zero.
 461 */
 462static inline void vma_mas_szero(struct ma_state *mas, unsigned long start,
 463				unsigned long end)
 464{
 465	trace_vma_mas_szero(mas->tree, start, end - 1);
 466	mas_set_range(mas, start, end - 1);
 467	mas_store_prealloc(mas, NULL);
 468}
 469
 470static int vma_link(struct mm_struct *mm, struct vm_area_struct *vma)
 
 
 471{
 472	MA_STATE(mas, &mm->mm_mt, 0, 0);
 473	struct address_space *mapping = NULL;
 474
 475	if (mas_preallocate(&mas, vma, GFP_KERNEL))
 476		return -ENOMEM;
 477
 478	if (vma->vm_file) {
 479		mapping = vma->vm_file->f_mapping;
 480		i_mmap_lock_write(mapping);
 481	}
 482
 483	vma_mas_store(vma, &mas);
 
 484
 485	if (mapping) {
 486		__vma_link_file(vma, mapping);
 487		i_mmap_unlock_write(mapping);
 488	}
 
 489
 490	mm->map_count++;
 491	validate_mm(mm);
 492	return 0;
 493}
 494
 495/*
 496 * vma_expand - Expand an existing VMA
 497 *
 498 * @mas: The maple state
 499 * @vma: The vma to expand
 500 * @start: The start of the vma
 501 * @end: The exclusive end of the vma
 502 * @pgoff: The page offset of vma
 503 * @next: The current of next vma.
 504 *
 505 * Expand @vma to @start and @end.  Can expand off the start and end.  Will
 506 * expand over @next if it's different from @vma and @end == @next->vm_end.
 507 * Checking if the @vma can expand and merge with @next needs to be handled by
 508 * the caller.
 509 *
 510 * Returns: 0 on success
 511 */
 512inline int vma_expand(struct ma_state *mas, struct vm_area_struct *vma,
 513		      unsigned long start, unsigned long end, pgoff_t pgoff,
 514		      struct vm_area_struct *next)
 515{
 516	struct mm_struct *mm = vma->vm_mm;
 517	struct address_space *mapping = NULL;
 518	struct rb_root_cached *root = NULL;
 519	struct anon_vma *anon_vma = vma->anon_vma;
 520	struct file *file = vma->vm_file;
 521	bool remove_next = false;
 522
 523	if (next && (vma != next) && (end == next->vm_end)) {
 524		remove_next = true;
 525		if (next->anon_vma && !vma->anon_vma) {
 526			int error;
 527
 528			anon_vma = next->anon_vma;
 529			vma->anon_vma = anon_vma;
 530			error = anon_vma_clone(vma, next);
 531			if (error)
 532				return error;
 533		}
 534	}
 535
 536	/* Not merging but overwriting any part of next is not handled. */
 537	VM_BUG_ON(next && !remove_next && next != vma && end > next->vm_start);
 538	/* Only handles expanding */
 539	VM_BUG_ON(vma->vm_start < start || vma->vm_end > end);
 540
 541	if (mas_preallocate(mas, vma, GFP_KERNEL))
 542		goto nomem;
 543
 544	vma_adjust_trans_huge(vma, start, end, 0);
 545
 546	if (file) {
 547		mapping = file->f_mapping;
 548		root = &mapping->i_mmap;
 549		uprobe_munmap(vma, vma->vm_start, vma->vm_end);
 550		i_mmap_lock_write(mapping);
 551	}
 552
 553	if (anon_vma) {
 554		anon_vma_lock_write(anon_vma);
 555		anon_vma_interval_tree_pre_update_vma(vma);
 556	}
 557
 558	if (file) {
 559		flush_dcache_mmap_lock(mapping);
 560		vma_interval_tree_remove(vma, root);
 561	}
 562
 563	vma->vm_start = start;
 564	vma->vm_end = end;
 565	vma->vm_pgoff = pgoff;
 566	/* Note: mas must be pointing to the expanding VMA */
 567	vma_mas_store(vma, mas);
 568
 569	if (file) {
 570		vma_interval_tree_insert(vma, root);
 571		flush_dcache_mmap_unlock(mapping);
 572	}
 573
 574	/* Expanding over the next vma */
 575	if (remove_next && file) {
 576		__remove_shared_vm_struct(next, file, mapping);
 577	}
 578
 579	if (anon_vma) {
 580		anon_vma_interval_tree_post_update_vma(vma);
 581		anon_vma_unlock_write(anon_vma);
 582	}
 583
 584	if (file) {
 585		i_mmap_unlock_write(mapping);
 586		uprobe_mmap(vma);
 587	}
 588
 589	if (remove_next) {
 590		if (file) {
 591			uprobe_munmap(next, next->vm_start, next->vm_end);
 592			fput(file);
 593		}
 594		if (next->anon_vma)
 595			anon_vma_merge(vma, next);
 596		mm->map_count--;
 597		mpol_put(vma_policy(next));
 598		vm_area_free(next);
 599	}
 600
 601	validate_mm(mm);
 602	return 0;
 
 
 
 603
 604nomem:
 605	return -ENOMEM;
 
 
 
 
 606}
 607
 608/*
 609 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
 610 * is already present in an i_mmap tree without adjusting the tree.
 611 * The following helper function should be used when such adjustments
 612 * are necessary.  The "insert" vma (if any) is to be inserted
 613 * before we drop the necessary locks.
 614 */
 615int __vma_adjust(struct vm_area_struct *vma, unsigned long start,
 616	unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert,
 617	struct vm_area_struct *expand)
 618{
 619	struct mm_struct *mm = vma->vm_mm;
 620	struct vm_area_struct *next_next = NULL;	/* uninit var warning */
 621	struct vm_area_struct *next = find_vma(mm, vma->vm_end);
 622	struct vm_area_struct *orig_vma = vma;
 623	struct address_space *mapping = NULL;
 624	struct rb_root_cached *root = NULL;
 625	struct anon_vma *anon_vma = NULL;
 626	struct file *file = vma->vm_file;
 627	bool vma_changed = false;
 628	long adjust_next = 0;
 629	int remove_next = 0;
 630	MA_STATE(mas, &mm->mm_mt, 0, 0);
 631	struct vm_area_struct *exporter = NULL, *importer = NULL;
 632
 633	if (next && !insert) {
 
 
 634		if (end >= next->vm_end) {
 635			/*
 636			 * vma expands, overlapping all the next, and
 637			 * perhaps the one after too (mprotect case 6).
 638			 * The only other cases that gets here are
 639			 * case 1, case 7 and case 8.
 640			 */
 641			if (next == expand) {
 642				/*
 643				 * The only case where we don't expand "vma"
 644				 * and we expand "next" instead is case 8.
 645				 */
 646				VM_WARN_ON(end != next->vm_end);
 647				/*
 648				 * remove_next == 3 means we're
 649				 * removing "vma" and that to do so we
 650				 * swapped "vma" and "next".
 651				 */
 652				remove_next = 3;
 653				VM_WARN_ON(file != next->vm_file);
 654				swap(vma, next);
 655			} else {
 656				VM_WARN_ON(expand != vma);
 657				/*
 658				 * case 1, 6, 7, remove_next == 2 is case 6,
 659				 * remove_next == 1 is case 1 or 7.
 660				 */
 661				remove_next = 1 + (end > next->vm_end);
 662				if (remove_next == 2)
 663					next_next = find_vma(mm, next->vm_end);
 664
 665				VM_WARN_ON(remove_next == 2 &&
 666					   end != next_next->vm_end);
 667			}
 668
 669			exporter = next;
 670			importer = vma;
 671
 672			/*
 673			 * If next doesn't have anon_vma, import from vma after
 674			 * next, if the vma overlaps with it.
 675			 */
 676			if (remove_next == 2 && !next->anon_vma)
 677				exporter = next_next;
 678
 679		} else if (end > next->vm_start) {
 680			/*
 681			 * vma expands, overlapping part of the next:
 682			 * mprotect case 5 shifting the boundary up.
 683			 */
 684			adjust_next = (end - next->vm_start);
 685			exporter = next;
 686			importer = vma;
 687			VM_WARN_ON(expand != importer);
 688		} else if (end < vma->vm_end) {
 689			/*
 690			 * vma shrinks, and !insert tells it's not
 691			 * split_vma inserting another: so it must be
 692			 * mprotect case 4 shifting the boundary down.
 693			 */
 694			adjust_next = -(vma->vm_end - end);
 695			exporter = vma;
 696			importer = next;
 697			VM_WARN_ON(expand != importer);
 698		}
 699
 700		/*
 701		 * Easily overlooked: when mprotect shifts the boundary,
 702		 * make sure the expanding vma has anon_vma set if the
 703		 * shrinking vma had, to cover any anon pages imported.
 704		 */
 705		if (exporter && exporter->anon_vma && !importer->anon_vma) {
 706			int error;
 707
 708			importer->anon_vma = exporter->anon_vma;
 709			error = anon_vma_clone(importer, exporter);
 710			if (error)
 711				return error;
 712		}
 713	}
 714
 715	if (mas_preallocate(&mas, vma, GFP_KERNEL))
 716		return -ENOMEM;
 717
 718	vma_adjust_trans_huge(orig_vma, start, end, adjust_next);
 719	if (file) {
 720		mapping = file->f_mapping;
 721		root = &mapping->i_mmap;
 722		uprobe_munmap(vma, vma->vm_start, vma->vm_end);
 
 723
 724		if (adjust_next)
 725			uprobe_munmap(next, next->vm_start, next->vm_end);
 
 
 726
 727		i_mmap_lock_write(mapping);
 728		if (insert && insert->vm_file) {
 729			/*
 730			 * Put into interval tree now, so instantiated pages
 731			 * are visible to arm/parisc __flush_dcache_page
 732			 * throughout; but we cannot insert into address
 733			 * space until vma start or end is updated.
 734			 */
 735			__vma_link_file(insert, insert->vm_file->f_mapping);
 736		}
 737	}
 738
 739	anon_vma = vma->anon_vma;
 740	if (!anon_vma && adjust_next)
 741		anon_vma = next->anon_vma;
 742	if (anon_vma) {
 743		VM_WARN_ON(adjust_next && next->anon_vma &&
 744			   anon_vma != next->anon_vma);
 745		anon_vma_lock_write(anon_vma);
 746		anon_vma_interval_tree_pre_update_vma(vma);
 747		if (adjust_next)
 748			anon_vma_interval_tree_pre_update_vma(next);
 
 749	}
 750
 751	if (file) {
 752		flush_dcache_mmap_lock(mapping);
 753		vma_interval_tree_remove(vma, root);
 754		if (adjust_next)
 755			vma_interval_tree_remove(next, root);
 756	}
 757
 758	if (start != vma->vm_start) {
 759		if ((vma->vm_start < start) &&
 760		    (!insert || (insert->vm_end != start))) {
 761			vma_mas_szero(&mas, vma->vm_start, start);
 762			VM_WARN_ON(insert && insert->vm_start > vma->vm_start);
 763		} else {
 764			vma_changed = true;
 765		}
 766		vma->vm_start = start;
 767	}
 768	if (end != vma->vm_end) {
 769		if (vma->vm_end > end) {
 770			if (!insert || (insert->vm_start != end)) {
 771				vma_mas_szero(&mas, end, vma->vm_end);
 772				mas_reset(&mas);
 773				VM_WARN_ON(insert &&
 774					   insert->vm_end < vma->vm_end);
 775			}
 776		} else {
 777			vma_changed = true;
 778		}
 779		vma->vm_end = end;
 780	}
 781
 782	if (vma_changed)
 783		vma_mas_store(vma, &mas);
 784
 785	vma->vm_pgoff = pgoff;
 786	if (adjust_next) {
 787		next->vm_start += adjust_next;
 788		next->vm_pgoff += adjust_next >> PAGE_SHIFT;
 789		vma_mas_store(next, &mas);
 790	}
 791
 792	if (file) {
 793		if (adjust_next)
 794			vma_interval_tree_insert(next, root);
 795		vma_interval_tree_insert(vma, root);
 796		flush_dcache_mmap_unlock(mapping);
 797	}
 798
 799	if (remove_next && file) {
 800		__remove_shared_vm_struct(next, file, mapping);
 801		if (remove_next == 2)
 802			__remove_shared_vm_struct(next_next, file, mapping);
 
 
 
 
 803	} else if (insert) {
 804		/*
 805		 * split_vma has split insert from vma, and needs
 806		 * us to insert it before dropping the locks
 807		 * (it may either follow vma or precede it).
 808		 */
 809		mas_reset(&mas);
 810		vma_mas_store(insert, &mas);
 811		mm->map_count++;
 812	}
 813
 814	if (anon_vma) {
 815		anon_vma_interval_tree_post_update_vma(vma);
 816		if (adjust_next)
 817			anon_vma_interval_tree_post_update_vma(next);
 818		anon_vma_unlock_write(anon_vma);
 819	}
 820
 821	if (file) {
 822		i_mmap_unlock_write(mapping);
 823		uprobe_mmap(vma);
 824
 825		if (adjust_next)
 826			uprobe_mmap(next);
 827	}
 828
 829	if (remove_next) {
 830again:
 831		if (file) {
 832			uprobe_munmap(next, next->vm_start, next->vm_end);
 833			fput(file);
 
 
 834		}
 835		if (next->anon_vma)
 836			anon_vma_merge(vma, next);
 837		mm->map_count--;
 838		mpol_put(vma_policy(next));
 839		if (remove_next != 2)
 840			BUG_ON(vma->vm_end < next->vm_end);
 841		vm_area_free(next);
 842
 843		/*
 844		 * In mprotect's case 6 (see comments on vma_merge),
 845		 * we must remove next_next too.
 
 846		 */
 847		if (remove_next == 2) {
 848			remove_next = 1;
 849			next = next_next;
 850			goto again;
 851		}
 852	}
 853	if (insert && file)
 854		uprobe_mmap(insert);
 855
 856	mas_destroy(&mas);
 857	validate_mm(mm);
 858
 859	return 0;
 860}
 861
 862/*
 863 * If the vma has a ->close operation then the driver probably needs to release
 864 * per-vma resources, so we don't attempt to merge those.
 865 */
 866static inline int is_mergeable_vma(struct vm_area_struct *vma,
 867				struct file *file, unsigned long vm_flags,
 868				struct vm_userfaultfd_ctx vm_userfaultfd_ctx,
 869				struct anon_vma_name *anon_name)
 870{
 871	/*
 872	 * VM_SOFTDIRTY should not prevent from VMA merging, if we
 873	 * match the flags but dirty bit -- the caller should mark
 874	 * merged VMA as dirty. If dirty bit won't be excluded from
 875	 * comparison, we increase pressure on the memory system forcing
 876	 * the kernel to generate new VMAs when old one could be
 877	 * extended instead.
 878	 */
 879	if ((vma->vm_flags ^ vm_flags) & ~VM_SOFTDIRTY)
 880		return 0;
 881	if (vma->vm_file != file)
 882		return 0;
 883	if (vma->vm_ops && vma->vm_ops->close)
 884		return 0;
 885	if (!is_mergeable_vm_userfaultfd_ctx(vma, vm_userfaultfd_ctx))
 886		return 0;
 887	if (!anon_vma_name_eq(anon_vma_name(vma), anon_name))
 888		return 0;
 889	return 1;
 890}
 891
 892static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
 893					struct anon_vma *anon_vma2,
 894					struct vm_area_struct *vma)
 895{
 896	/*
 897	 * The list_is_singular() test is to avoid merging VMA cloned from
 898	 * parents. This can improve scalability caused by anon_vma lock.
 899	 */
 900	if ((!anon_vma1 || !anon_vma2) && (!vma ||
 901		list_is_singular(&vma->anon_vma_chain)))
 902		return 1;
 903	return anon_vma1 == anon_vma2;
 904}
 905
 906/*
 907 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
 908 * in front of (at a lower virtual address and file offset than) the vma.
 909 *
 910 * We cannot merge two vmas if they have differently assigned (non-NULL)
 911 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
 912 *
 913 * We don't check here for the merged mmap wrapping around the end of pagecache
 914 * indices (16TB on ia32) because do_mmap() does not permit mmap's which
 915 * wrap, nor mmaps which cover the final page at index -1UL.
 916 */
 917static int
 918can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
 919		     struct anon_vma *anon_vma, struct file *file,
 920		     pgoff_t vm_pgoff,
 921		     struct vm_userfaultfd_ctx vm_userfaultfd_ctx,
 922		     struct anon_vma_name *anon_name)
 923{
 924	if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx, anon_name) &&
 925	    is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
 926		if (vma->vm_pgoff == vm_pgoff)
 927			return 1;
 928	}
 929	return 0;
 930}
 931
 932/*
 933 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
 934 * beyond (at a higher virtual address and file offset than) the vma.
 935 *
 936 * We cannot merge two vmas if they have differently assigned (non-NULL)
 937 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
 938 */
 939static int
 940can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
 941		    struct anon_vma *anon_vma, struct file *file,
 942		    pgoff_t vm_pgoff,
 943		    struct vm_userfaultfd_ctx vm_userfaultfd_ctx,
 944		    struct anon_vma_name *anon_name)
 945{
 946	if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx, anon_name) &&
 947	    is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
 948		pgoff_t vm_pglen;
 949		vm_pglen = vma_pages(vma);
 950		if (vma->vm_pgoff + vm_pglen == vm_pgoff)
 951			return 1;
 952	}
 953	return 0;
 954}
 955
 956/*
 957 * Given a mapping request (addr,end,vm_flags,file,pgoff,anon_name),
 958 * figure out whether that can be merged with its predecessor or its
 959 * successor.  Or both (it neatly fills a hole).
 960 *
 961 * In most cases - when called for mmap, brk or mremap - [addr,end) is
 962 * certain not to be mapped by the time vma_merge is called; but when
 963 * called for mprotect, it is certain to be already mapped (either at
 964 * an offset within prev, or at the start of next), and the flags of
 965 * this area are about to be changed to vm_flags - and the no-change
 966 * case has already been eliminated.
 967 *
 968 * The following mprotect cases have to be considered, where AAAA is
 969 * the area passed down from mprotect_fixup, never extending beyond one
 970 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
 971 *
 972 *     AAAA             AAAA                   AAAA
 973 *    PPPPPPNNNNNN    PPPPPPNNNNNN       PPPPPPNNNNNN
 974 *    cannot merge    might become       might become
 975 *                    PPNNNNNNNNNN       PPPPPPPPPPNN
 976 *    mmap, brk or    case 4 below       case 5 below
 977 *    mremap move:
 978 *                        AAAA               AAAA
 979 *                    PPPP    NNNN       PPPPNNNNXXXX
 980 *                    might become       might become
 981 *                    PPPPPPPPPPPP 1 or  PPPPPPPPPPPP 6 or
 982 *                    PPPPPPPPNNNN 2 or  PPPPPPPPXXXX 7 or
 983 *                    PPPPNNNNNNNN 3     PPPPXXXXXXXX 8
 984 *
 985 * It is important for case 8 that the vma NNNN overlapping the
 986 * region AAAA is never going to extended over XXXX. Instead XXXX must
 987 * be extended in region AAAA and NNNN must be removed. This way in
 988 * all cases where vma_merge succeeds, the moment vma_adjust drops the
 989 * rmap_locks, the properties of the merged vma will be already
 990 * correct for the whole merged range. Some of those properties like
 991 * vm_page_prot/vm_flags may be accessed by rmap_walks and they must
 992 * be correct for the whole merged range immediately after the
 993 * rmap_locks are released. Otherwise if XXXX would be removed and
 994 * NNNN would be extended over the XXXX range, remove_migration_ptes
 995 * or other rmap walkers (if working on addresses beyond the "end"
 996 * parameter) may establish ptes with the wrong permissions of NNNN
 997 * instead of the right permissions of XXXX.
 998 */
 999struct vm_area_struct *vma_merge(struct mm_struct *mm,
1000			struct vm_area_struct *prev, unsigned long addr,
1001			unsigned long end, unsigned long vm_flags,
1002			struct anon_vma *anon_vma, struct file *file,
1003			pgoff_t pgoff, struct mempolicy *policy,
1004			struct vm_userfaultfd_ctx vm_userfaultfd_ctx,
1005			struct anon_vma_name *anon_name)
1006{
1007	pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
1008	struct vm_area_struct *mid, *next, *res;
1009	int err = -1;
1010	bool merge_prev = false;
1011	bool merge_next = false;
1012
1013	/*
1014	 * We later require that vma->vm_flags == vm_flags,
1015	 * so this tests vma->vm_flags & VM_SPECIAL, too.
1016	 */
1017	if (vm_flags & VM_SPECIAL)
1018		return NULL;
1019
1020	next = find_vma(mm, prev ? prev->vm_end : 0);
1021	mid = next;
 
 
 
1022	if (next && next->vm_end == end)		/* cases 6, 7, 8 */
1023		next = find_vma(mm, next->vm_end);
1024
1025	/* verify some invariant that must be enforced by the caller */
1026	VM_WARN_ON(prev && addr <= prev->vm_start);
1027	VM_WARN_ON(mid && end > mid->vm_end);
1028	VM_WARN_ON(addr >= end);
1029
1030	/* Can we merge the predecessor? */
1031	if (prev && prev->vm_end == addr &&
1032			mpol_equal(vma_policy(prev), policy) &&
1033			can_vma_merge_after(prev, vm_flags,
1034					    anon_vma, file, pgoff,
1035					    vm_userfaultfd_ctx, anon_name)) {
1036		merge_prev = true;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1037	}
1038	/* Can we merge the successor? */
 
 
 
1039	if (next && end == next->vm_start &&
1040			mpol_equal(policy, vma_policy(next)) &&
1041			can_vma_merge_before(next, vm_flags,
1042					     anon_vma, file, pgoff+pglen,
1043					     vm_userfaultfd_ctx, anon_name)) {
1044		merge_next = true;
1045	}
1046	/* Can we merge both the predecessor and the successor? */
1047	if (merge_prev && merge_next &&
1048			is_mergeable_anon_vma(prev->anon_vma,
1049				next->anon_vma, NULL)) {	 /* cases 1, 6 */
1050		err = __vma_adjust(prev, prev->vm_start,
1051					next->vm_end, prev->vm_pgoff, NULL,
1052					prev);
1053		res = prev;
1054	} else if (merge_prev) {			/* cases 2, 5, 7 */
1055		err = __vma_adjust(prev, prev->vm_start,
1056					end, prev->vm_pgoff, NULL, prev);
1057		res = prev;
1058	} else if (merge_next) {
1059		if (prev && addr < prev->vm_end)	/* case 4 */
1060			err = __vma_adjust(prev, prev->vm_start,
1061					addr, prev->vm_pgoff, NULL, next);
1062		else					/* cases 3, 8 */
1063			err = __vma_adjust(mid, addr, next->vm_end,
1064					next->vm_pgoff - pglen, NULL, next);
1065		res = next;
 
 
 
1066	}
1067
1068	/*
1069	 * Cannot merge with predecessor or successor or error in __vma_adjust?
1070	 */
1071	if (err)
1072		return NULL;
1073	khugepaged_enter_vma(res, vm_flags);
1074	return res;
1075}
1076
1077/*
1078 * Rough compatibility check to quickly see if it's even worth looking
1079 * at sharing an anon_vma.
1080 *
1081 * They need to have the same vm_file, and the flags can only differ
1082 * in things that mprotect may change.
1083 *
1084 * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that
1085 * we can merge the two vma's. For example, we refuse to merge a vma if
1086 * there is a vm_ops->close() function, because that indicates that the
1087 * driver is doing some kind of reference counting. But that doesn't
1088 * really matter for the anon_vma sharing case.
1089 */
1090static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b)
1091{
1092	return a->vm_end == b->vm_start &&
1093		mpol_equal(vma_policy(a), vma_policy(b)) &&
1094		a->vm_file == b->vm_file &&
1095		!((a->vm_flags ^ b->vm_flags) & ~(VM_ACCESS_FLAGS | VM_SOFTDIRTY)) &&
1096		b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT);
1097}
1098
1099/*
1100 * Do some basic sanity checking to see if we can re-use the anon_vma
1101 * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be
1102 * the same as 'old', the other will be the new one that is trying
1103 * to share the anon_vma.
1104 *
1105 * NOTE! This runs with mmap_lock held for reading, so it is possible that
1106 * the anon_vma of 'old' is concurrently in the process of being set up
1107 * by another page fault trying to merge _that_. But that's ok: if it
1108 * is being set up, that automatically means that it will be a singleton
1109 * acceptable for merging, so we can do all of this optimistically. But
1110 * we do that READ_ONCE() to make sure that we never re-load the pointer.
1111 *
1112 * IOW: that the "list_is_singular()" test on the anon_vma_chain only
1113 * matters for the 'stable anon_vma' case (ie the thing we want to avoid
1114 * is to return an anon_vma that is "complex" due to having gone through
1115 * a fork).
1116 *
1117 * We also make sure that the two vma's are compatible (adjacent,
1118 * and with the same memory policies). That's all stable, even with just
1119 * a read lock on the mmap_lock.
1120 */
1121static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b)
1122{
1123	if (anon_vma_compatible(a, b)) {
1124		struct anon_vma *anon_vma = READ_ONCE(old->anon_vma);
1125
1126		if (anon_vma && list_is_singular(&old->anon_vma_chain))
1127			return anon_vma;
1128	}
1129	return NULL;
1130}
1131
1132/*
1133 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
1134 * neighbouring vmas for a suitable anon_vma, before it goes off
1135 * to allocate a new anon_vma.  It checks because a repetitive
1136 * sequence of mprotects and faults may otherwise lead to distinct
1137 * anon_vmas being allocated, preventing vma merge in subsequent
1138 * mprotect.
1139 */
1140struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
1141{
1142	MA_STATE(mas, &vma->vm_mm->mm_mt, vma->vm_end, vma->vm_end);
1143	struct anon_vma *anon_vma = NULL;
1144	struct vm_area_struct *prev, *next;
1145
1146	/* Try next first. */
1147	next = mas_walk(&mas);
1148	if (next) {
1149		anon_vma = reusable_anon_vma(next, vma, next);
1150		if (anon_vma)
1151			return anon_vma;
1152	}
1153
1154	prev = mas_prev(&mas, 0);
1155	VM_BUG_ON_VMA(prev != vma, vma);
1156	prev = mas_prev(&mas, 0);
1157	/* Try prev next. */
1158	if (prev)
1159		anon_vma = reusable_anon_vma(prev, prev, vma);
1160
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1161	/*
1162	 * We might reach here with anon_vma == NULL if we can't find
1163	 * any reusable anon_vma.
1164	 * There's no absolute need to look only at touching neighbours:
1165	 * we could search further afield for "compatible" anon_vmas.
1166	 * But it would probably just be a waste of time searching,
1167	 * or lead to too many vmas hanging off the same anon_vma.
1168	 * We're trying to allow mprotect remerging later on,
1169	 * not trying to minimize memory used for anon_vmas.
1170	 */
1171	return anon_vma;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1172}
 
1173
1174/*
1175 * If a hint addr is less than mmap_min_addr change hint to be as
1176 * low as possible but still greater than mmap_min_addr
1177 */
1178static inline unsigned long round_hint_to_min(unsigned long hint)
1179{
1180	hint &= PAGE_MASK;
1181	if (((void *)hint != NULL) &&
1182	    (hint < mmap_min_addr))
1183		return PAGE_ALIGN(mmap_min_addr);
1184	return hint;
1185}
1186
1187int mlock_future_check(struct mm_struct *mm, unsigned long flags,
1188		       unsigned long len)
1189{
1190	unsigned long locked, lock_limit;
1191
1192	/*  mlock MCL_FUTURE? */
1193	if (flags & VM_LOCKED) {
1194		locked = len >> PAGE_SHIFT;
1195		locked += mm->locked_vm;
1196		lock_limit = rlimit(RLIMIT_MEMLOCK);
1197		lock_limit >>= PAGE_SHIFT;
1198		if (locked > lock_limit && !capable(CAP_IPC_LOCK))
1199			return -EAGAIN;
1200	}
1201	return 0;
1202}
1203
1204static inline u64 file_mmap_size_max(struct file *file, struct inode *inode)
1205{
1206	if (S_ISREG(inode->i_mode))
1207		return MAX_LFS_FILESIZE;
1208
1209	if (S_ISBLK(inode->i_mode))
1210		return MAX_LFS_FILESIZE;
1211
1212	if (S_ISSOCK(inode->i_mode))
1213		return MAX_LFS_FILESIZE;
1214
1215	/* Special "we do even unsigned file positions" case */
1216	if (file->f_mode & FMODE_UNSIGNED_OFFSET)
1217		return 0;
1218
1219	/* Yes, random drivers might want more. But I'm tired of buggy drivers */
1220	return ULONG_MAX;
1221}
1222
1223static inline bool file_mmap_ok(struct file *file, struct inode *inode,
1224				unsigned long pgoff, unsigned long len)
1225{
1226	u64 maxsize = file_mmap_size_max(file, inode);
1227
1228	if (maxsize && len > maxsize)
1229		return false;
1230	maxsize -= len;
1231	if (pgoff > maxsize >> PAGE_SHIFT)
1232		return false;
1233	return true;
1234}
1235
1236/*
1237 * The caller must write-lock current->mm->mmap_lock.
1238 */
1239unsigned long do_mmap(struct file *file, unsigned long addr,
 
1240			unsigned long len, unsigned long prot,
1241			unsigned long flags, unsigned long pgoff,
1242			unsigned long *populate, struct list_head *uf)
1243{
1244	struct mm_struct *mm = current->mm;
 
1245	vm_flags_t vm_flags;
1246	int pkey = 0;
1247
1248	validate_mm(mm);
1249	*populate = 0;
1250
1251	if (!len)
1252		return -EINVAL;
1253
1254	/*
1255	 * Does the application expect PROT_READ to imply PROT_EXEC?
1256	 *
1257	 * (the exception is when the underlying filesystem is noexec
1258	 *  mounted, in which case we dont add PROT_EXEC.)
1259	 */
1260	if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
1261		if (!(file && path_noexec(&file->f_path)))
1262			prot |= PROT_EXEC;
1263
1264	/* force arch specific MAP_FIXED handling in get_unmapped_area */
1265	if (flags & MAP_FIXED_NOREPLACE)
1266		flags |= MAP_FIXED;
1267
1268	if (!(flags & MAP_FIXED))
1269		addr = round_hint_to_min(addr);
1270
1271	/* Careful about overflows.. */
1272	len = PAGE_ALIGN(len);
1273	if (!len)
1274		return -ENOMEM;
1275
1276	/* offset overflow? */
1277	if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
1278		return -EOVERFLOW;
1279
1280	/* Too many mappings? */
1281	if (mm->map_count > sysctl_max_map_count)
1282		return -ENOMEM;
1283
1284	/* Obtain the address to map to. we verify (or select) it and ensure
1285	 * that it represents a valid section of the address space.
1286	 */
1287	addr = get_unmapped_area(file, addr, len, pgoff, flags);
1288	if (IS_ERR_VALUE(addr))
1289		return addr;
1290
1291	if (flags & MAP_FIXED_NOREPLACE) {
1292		if (find_vma_intersection(mm, addr, addr + len))
1293			return -EEXIST;
1294	}
1295
1296	if (prot == PROT_EXEC) {
1297		pkey = execute_only_pkey(mm);
1298		if (pkey < 0)
1299			pkey = 0;
1300	}
1301
1302	/* Do simple checking here so the lower-level routines won't have
1303	 * to. we assume access permissions have been handled by the open
1304	 * of the memory object, so we don't do any here.
1305	 */
1306	vm_flags = calc_vm_prot_bits(prot, pkey) | calc_vm_flag_bits(flags) |
1307			mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1308
1309	if (flags & MAP_LOCKED)
1310		if (!can_do_mlock())
1311			return -EPERM;
1312
1313	if (mlock_future_check(mm, vm_flags, len))
1314		return -EAGAIN;
1315
1316	if (file) {
1317		struct inode *inode = file_inode(file);
1318		unsigned long flags_mask;
1319
1320		if (!file_mmap_ok(file, inode, pgoff, len))
1321			return -EOVERFLOW;
 
1322
1323		flags_mask = LEGACY_MAP_MASK | file->f_op->mmap_supported_flags;
1324
 
1325		switch (flags & MAP_TYPE) {
1326		case MAP_SHARED:
1327			/*
1328			 * Force use of MAP_SHARED_VALIDATE with non-legacy
1329			 * flags. E.g. MAP_SYNC is dangerous to use with
1330			 * MAP_SHARED as you don't know which consistency model
1331			 * you will get. We silently ignore unsupported flags
1332			 * with MAP_SHARED to preserve backward compatibility.
1333			 */
1334			flags &= LEGACY_MAP_MASK;
1335			fallthrough;
1336		case MAP_SHARED_VALIDATE:
1337			if (flags & ~flags_mask)
1338				return -EOPNOTSUPP;
1339			if (prot & PROT_WRITE) {
1340				if (!(file->f_mode & FMODE_WRITE))
1341					return -EACCES;
1342				if (IS_SWAPFILE(file->f_mapping->host))
1343					return -ETXTBSY;
1344			}
1345
1346			/*
1347			 * Make sure we don't allow writing to an append-only
1348			 * file..
1349			 */
1350			if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
1351				return -EACCES;
1352
 
 
 
 
 
 
1353			vm_flags |= VM_SHARED | VM_MAYSHARE;
1354			if (!(file->f_mode & FMODE_WRITE))
1355				vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
1356			fallthrough;
 
1357		case MAP_PRIVATE:
1358			if (!(file->f_mode & FMODE_READ))
1359				return -EACCES;
1360			if (path_noexec(&file->f_path)) {
1361				if (vm_flags & VM_EXEC)
1362					return -EPERM;
1363				vm_flags &= ~VM_MAYEXEC;
1364			}
1365
1366			if (!file->f_op->mmap)
1367				return -ENODEV;
1368			if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1369				return -EINVAL;
1370			break;
1371
1372		default:
1373			return -EINVAL;
1374		}
1375	} else {
1376		switch (flags & MAP_TYPE) {
1377		case MAP_SHARED:
1378			if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1379				return -EINVAL;
1380			/*
1381			 * Ignore pgoff.
1382			 */
1383			pgoff = 0;
1384			vm_flags |= VM_SHARED | VM_MAYSHARE;
1385			break;
1386		case MAP_PRIVATE:
1387			/*
1388			 * Set pgoff according to addr for anon_vma.
1389			 */
1390			pgoff = addr >> PAGE_SHIFT;
1391			break;
1392		default:
1393			return -EINVAL;
1394		}
1395	}
1396
1397	/*
1398	 * Set 'VM_NORESERVE' if we should not account for the
1399	 * memory use of this mapping.
1400	 */
1401	if (flags & MAP_NORESERVE) {
1402		/* We honor MAP_NORESERVE if allowed to overcommit */
1403		if (sysctl_overcommit_memory != OVERCOMMIT_NEVER)
1404			vm_flags |= VM_NORESERVE;
1405
1406		/* hugetlb applies strict overcommit unless MAP_NORESERVE */
1407		if (file && is_file_hugepages(file))
1408			vm_flags |= VM_NORESERVE;
1409	}
1410
1411	addr = mmap_region(file, addr, len, vm_flags, pgoff, uf);
1412	if (!IS_ERR_VALUE(addr) &&
1413	    ((vm_flags & VM_LOCKED) ||
1414	     (flags & (MAP_POPULATE | MAP_NONBLOCK)) == MAP_POPULATE))
1415		*populate = len;
1416	return addr;
1417}
1418
1419unsigned long ksys_mmap_pgoff(unsigned long addr, unsigned long len,
1420			      unsigned long prot, unsigned long flags,
1421			      unsigned long fd, unsigned long pgoff)
1422{
1423	struct file *file = NULL;
1424	unsigned long retval;
1425
1426	if (!(flags & MAP_ANONYMOUS)) {
1427		audit_mmap_fd(fd, flags);
 
 
1428		file = fget(fd);
1429		if (!file)
1430			return -EBADF;
1431		if (is_file_hugepages(file)) {
1432			len = ALIGN(len, huge_page_size(hstate_file(file)));
1433		} else if (unlikely(flags & MAP_HUGETLB)) {
1434			retval = -EINVAL;
1435			goto out_fput;
1436		}
1437	} else if (flags & MAP_HUGETLB) {
1438		struct hstate *hs;
1439
1440		hs = hstate_sizelog((flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK);
1441		if (!hs)
1442			return -EINVAL;
1443
1444		len = ALIGN(len, huge_page_size(hs));
1445		/*
1446		 * VM_NORESERVE is used because the reservations will be
1447		 * taken when vm_ops->mmap() is called
 
 
1448		 */
1449		file = hugetlb_file_setup(HUGETLB_ANON_FILE, len,
1450				VM_NORESERVE,
1451				HUGETLB_ANONHUGE_INODE,
1452				(flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK);
1453		if (IS_ERR(file))
1454			return PTR_ERR(file);
1455	}
1456
 
 
1457	retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
1458out_fput:
1459	if (file)
1460		fput(file);
 
1461	return retval;
1462}
1463
1464SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1465		unsigned long, prot, unsigned long, flags,
1466		unsigned long, fd, unsigned long, pgoff)
1467{
1468	return ksys_mmap_pgoff(addr, len, prot, flags, fd, pgoff);
1469}
1470
1471#ifdef __ARCH_WANT_SYS_OLD_MMAP
1472struct mmap_arg_struct {
1473	unsigned long addr;
1474	unsigned long len;
1475	unsigned long prot;
1476	unsigned long flags;
1477	unsigned long fd;
1478	unsigned long offset;
1479};
1480
1481SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1482{
1483	struct mmap_arg_struct a;
1484
1485	if (copy_from_user(&a, arg, sizeof(a)))
1486		return -EFAULT;
1487	if (offset_in_page(a.offset))
1488		return -EINVAL;
1489
1490	return ksys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1491			       a.offset >> PAGE_SHIFT);
1492}
1493#endif /* __ARCH_WANT_SYS_OLD_MMAP */
1494
1495/*
1496 * Some shared mappings will want the pages marked read-only
1497 * to track write events. If so, we'll downgrade vm_page_prot
1498 * to the private version (using protection_map[] without the
1499 * VM_SHARED bit).
1500 */
1501int vma_wants_writenotify(struct vm_area_struct *vma, pgprot_t vm_page_prot)
1502{
1503	vm_flags_t vm_flags = vma->vm_flags;
1504	const struct vm_operations_struct *vm_ops = vma->vm_ops;
1505
1506	/* If it was private or non-writable, the write bit is already clear */
1507	if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED)))
1508		return 0;
1509
1510	/* The backer wishes to know when pages are first written to? */
1511	if (vm_ops && (vm_ops->page_mkwrite || vm_ops->pfn_mkwrite))
1512		return 1;
1513
1514	/* The open routine did something to the protections that pgprot_modify
1515	 * won't preserve? */
1516	if (pgprot_val(vm_page_prot) !=
1517	    pgprot_val(vm_pgprot_modify(vm_page_prot, vm_flags)))
1518		return 0;
1519
1520	/*
1521	 * Do we need to track softdirty? hugetlb does not support softdirty
1522	 * tracking yet.
1523	 */
1524	if (vma_soft_dirty_enabled(vma) && !is_vm_hugetlb_page(vma))
1525		return 1;
1526
1527	/* Do we need write faults for uffd-wp tracking? */
1528	if (userfaultfd_wp(vma))
1529		return 1;
1530
1531	/* Specialty mapping? */
1532	if (vm_flags & VM_PFNMAP)
1533		return 0;
1534
1535	/* Can the mapping track the dirty pages? */
1536	return vma->vm_file && vma->vm_file->f_mapping &&
1537		mapping_can_writeback(vma->vm_file->f_mapping);
1538}
1539
1540/*
1541 * We account for memory if it's a private writeable mapping,
1542 * not hugepages and VM_NORESERVE wasn't set.
1543 */
1544static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags)
1545{
1546	/*
1547	 * hugetlb has its own accounting separate from the core VM
1548	 * VM_HUGETLB may not be set yet so we cannot check for that flag.
1549	 */
1550	if (file && is_file_hugepages(file))
1551		return 0;
1552
1553	return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE;
1554}
1555
1556/**
1557 * unmapped_area() - Find an area between the low_limit and the high_limit with
1558 * the correct alignment and offset, all from @info. Note: current->mm is used
1559 * for the search.
1560 *
1561 * @info: The unmapped area information including the range (low_limit -
1562 * hight_limit), the alignment offset and mask.
1563 *
1564 * Return: A memory address or -ENOMEM.
1565 */
1566static unsigned long unmapped_area(struct vm_unmapped_area_info *info)
1567{
1568	unsigned long length, gap;
 
 
 
 
 
 
1569
1570	MA_STATE(mas, &current->mm->mm_mt, 0, 0);
 
 
 
 
 
 
 
 
1571
1572	/* Adjust search length to account for worst case alignment overhead */
1573	length = info->length + info->align_mask;
1574	if (length < info->length)
1575		return -ENOMEM;
1576
1577	if (mas_empty_area(&mas, info->low_limit, info->high_limit - 1,
1578				  length))
1579		return -ENOMEM;
 
 
 
 
 
1580
1581	gap = mas.index;
1582	gap += (info->align_offset - gap) & info->align_mask;
1583	return gap;
1584}
1585
1586/**
1587 * unmapped_area_topdown() - Find an area between the low_limit and the
1588 * high_limit with * the correct alignment and offset at the highest available
1589 * address, all from @info. Note: current->mm is used for the search.
1590 *
1591 * @info: The unmapped area information including the range (low_limit -
1592 * hight_limit), the alignment offset and mask.
1593 *
1594 * Return: A memory address or -ENOMEM.
1595 */
1596static unsigned long unmapped_area_topdown(struct vm_unmapped_area_info *info)
1597{
1598	unsigned long length, gap;
1599
1600	MA_STATE(mas, &current->mm->mm_mt, 0, 0);
1601	/* Adjust search length to account for worst case alignment overhead */
1602	length = info->length + info->align_mask;
1603	if (length < info->length)
1604		return -ENOMEM;
 
1605
1606	if (mas_empty_area_rev(&mas, info->low_limit, info->high_limit - 1,
1607				length))
1608		return -ENOMEM;
 
 
 
 
 
 
 
1609
1610	gap = mas.last + 1 - info->length;
1611	gap -= (gap - info->align_offset) & info->align_mask;
1612	return gap;
1613}
 
 
 
1614
1615/*
1616 * Search for an unmapped address range.
1617 *
1618 * We are looking for a range that:
1619 * - does not intersect with any VMA;
1620 * - is contained within the [low_limit, high_limit) interval;
1621 * - is at least the desired size.
1622 * - satisfies (begin_addr & align_mask) == (align_offset & align_mask)
1623 */
1624unsigned long vm_unmapped_area(struct vm_unmapped_area_info *info)
1625{
1626	unsigned long addr;
1627
1628	if (info->flags & VM_UNMAPPED_AREA_TOPDOWN)
1629		addr = unmapped_area_topdown(info);
1630	else
1631		addr = unmapped_area(info);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1632
1633	trace_vm_unmapped_area(addr, info);
1634	return addr;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1635}
1636
1637/* Get an address range which is currently unmapped.
1638 * For shmat() with addr=0.
1639 *
1640 * Ugly calling convention alert:
1641 * Return value with the low bits set means error value,
1642 * ie
1643 *	if (ret & ~PAGE_MASK)
1644 *		error = ret;
1645 *
1646 * This function "knows" that -ENOMEM has the bits set.
1647 */
 
1648unsigned long
1649generic_get_unmapped_area(struct file *filp, unsigned long addr,
1650			  unsigned long len, unsigned long pgoff,
1651			  unsigned long flags)
1652{
1653	struct mm_struct *mm = current->mm;
1654	struct vm_area_struct *vma, *prev;
1655	struct vm_unmapped_area_info info;
1656	const unsigned long mmap_end = arch_get_mmap_end(addr, len, flags);
1657
1658	if (len > mmap_end - mmap_min_addr)
1659		return -ENOMEM;
1660
1661	if (flags & MAP_FIXED)
1662		return addr;
1663
1664	if (addr) {
1665		addr = PAGE_ALIGN(addr);
1666		vma = find_vma_prev(mm, addr, &prev);
1667		if (mmap_end - len >= addr && addr >= mmap_min_addr &&
1668		    (!vma || addr + len <= vm_start_gap(vma)) &&
1669		    (!prev || addr >= vm_end_gap(prev)))
1670			return addr;
1671	}
 
 
 
 
 
 
1672
1673	info.flags = 0;
1674	info.length = len;
1675	info.low_limit = mm->mmap_base;
1676	info.high_limit = mmap_end;
1677	info.align_mask = 0;
1678	info.align_offset = 0;
1679	return vm_unmapped_area(&info);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1680}
 
1681
1682#ifndef HAVE_ARCH_UNMAPPED_AREA
1683unsigned long
1684arch_get_unmapped_area(struct file *filp, unsigned long addr,
1685		       unsigned long len, unsigned long pgoff,
1686		       unsigned long flags)
1687{
1688	return generic_get_unmapped_area(filp, addr, len, pgoff, flags);
 
 
 
 
1689}
1690#endif
1691
1692/*
1693 * This mmap-allocator allocates new areas top-down from below the
1694 * stack's low limit (the base):
1695 */
 
1696unsigned long
1697generic_get_unmapped_area_topdown(struct file *filp, unsigned long addr,
1698				  unsigned long len, unsigned long pgoff,
1699				  unsigned long flags)
1700{
1701	struct vm_area_struct *vma, *prev;
1702	struct mm_struct *mm = current->mm;
1703	struct vm_unmapped_area_info info;
1704	const unsigned long mmap_end = arch_get_mmap_end(addr, len, flags);
1705
1706	/* requested length too big for entire address space */
1707	if (len > mmap_end - mmap_min_addr)
1708		return -ENOMEM;
1709
1710	if (flags & MAP_FIXED)
1711		return addr;
1712
1713	/* requesting a specific address */
1714	if (addr) {
1715		addr = PAGE_ALIGN(addr);
1716		vma = find_vma_prev(mm, addr, &prev);
1717		if (mmap_end - len >= addr && addr >= mmap_min_addr &&
1718				(!vma || addr + len <= vm_start_gap(vma)) &&
1719				(!prev || addr >= vm_end_gap(prev)))
1720			return addr;
1721	}
1722
1723	info.flags = VM_UNMAPPED_AREA_TOPDOWN;
1724	info.length = len;
1725	info.low_limit = max(PAGE_SIZE, mmap_min_addr);
1726	info.high_limit = arch_get_mmap_base(addr, mm->mmap_base);
1727	info.align_mask = 0;
1728	info.align_offset = 0;
1729	addr = vm_unmapped_area(&info);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1730
1731	/*
1732	 * A failed mmap() very likely causes application failure,
1733	 * so fall back to the bottom-up function here. This scenario
1734	 * can happen with large stack limits and large mmap()
1735	 * allocations.
1736	 */
1737	if (offset_in_page(addr)) {
1738		VM_BUG_ON(addr != -ENOMEM);
1739		info.flags = 0;
1740		info.low_limit = TASK_UNMAPPED_BASE;
1741		info.high_limit = mmap_end;
1742		addr = vm_unmapped_area(&info);
1743	}
 
1744
1745	return addr;
1746}
 
1747
1748#ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1749unsigned long
1750arch_get_unmapped_area_topdown(struct file *filp, unsigned long addr,
1751			       unsigned long len, unsigned long pgoff,
1752			       unsigned long flags)
1753{
1754	return generic_get_unmapped_area_topdown(filp, addr, len, pgoff, flags);
 
 
 
 
 
 
 
 
1755}
1756#endif
1757
1758unsigned long
1759get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
1760		unsigned long pgoff, unsigned long flags)
1761{
1762	unsigned long (*get_area)(struct file *, unsigned long,
1763				  unsigned long, unsigned long, unsigned long);
1764
1765	unsigned long error = arch_mmap_check(addr, len, flags);
1766	if (error)
1767		return error;
1768
1769	/* Careful about overflows.. */
1770	if (len > TASK_SIZE)
1771		return -ENOMEM;
1772
1773	get_area = current->mm->get_unmapped_area;
1774	if (file) {
1775		if (file->f_op->get_unmapped_area)
1776			get_area = file->f_op->get_unmapped_area;
1777	} else if (flags & MAP_SHARED) {
1778		/*
1779		 * mmap_region() will call shmem_zero_setup() to create a file,
1780		 * so use shmem's get_unmapped_area in case it can be huge.
1781		 * do_mmap() will clear pgoff, so match alignment.
1782		 */
1783		pgoff = 0;
1784		get_area = shmem_get_unmapped_area;
1785	}
1786
1787	addr = get_area(file, addr, len, pgoff, flags);
1788	if (IS_ERR_VALUE(addr))
1789		return addr;
1790
1791	if (addr > TASK_SIZE - len)
1792		return -ENOMEM;
1793	if (offset_in_page(addr))
1794		return -EINVAL;
1795
 
1796	error = security_mmap_addr(addr);
1797	return error ? error : addr;
1798}
1799
1800EXPORT_SYMBOL(get_unmapped_area);
1801
1802/**
1803 * find_vma_intersection() - Look up the first VMA which intersects the interval
1804 * @mm: The process address space.
1805 * @start_addr: The inclusive start user address.
1806 * @end_addr: The exclusive end user address.
1807 *
1808 * Returns: The first VMA within the provided range, %NULL otherwise.  Assumes
1809 * start_addr < end_addr.
1810 */
1811struct vm_area_struct *find_vma_intersection(struct mm_struct *mm,
1812					     unsigned long start_addr,
1813					     unsigned long end_addr)
1814{
1815	unsigned long index = start_addr;
1816
1817	mmap_assert_locked(mm);
1818	return mt_find(&mm->mm_mt, &index, end_addr - 1);
1819}
1820EXPORT_SYMBOL(find_vma_intersection);
1821
1822/**
1823 * find_vma() - Find the VMA for a given address, or the next VMA.
1824 * @mm: The mm_struct to check
1825 * @addr: The address
1826 *
1827 * Returns: The VMA associated with addr, or the next VMA.
1828 * May return %NULL in the case of no VMA at addr or above.
1829 */
1830struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
1831{
1832	unsigned long index = addr;
 
 
 
1833
1834	mmap_assert_locked(mm);
1835	return mt_find(&mm->mm_mt, &index, ULONG_MAX);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1836}
 
1837EXPORT_SYMBOL(find_vma);
1838
1839/**
1840 * find_vma_prev() - Find the VMA for a given address, or the next vma and
1841 * set %pprev to the previous VMA, if any.
1842 * @mm: The mm_struct to check
1843 * @addr: The address
1844 * @pprev: The pointer to set to the previous VMA
1845 *
1846 * Note that RCU lock is missing here since the external mmap_lock() is used
1847 * instead.
1848 *
1849 * Returns: The VMA associated with @addr, or the next vma.
1850 * May return %NULL in the case of no vma at addr or above.
1851 */
1852struct vm_area_struct *
1853find_vma_prev(struct mm_struct *mm, unsigned long addr,
1854			struct vm_area_struct **pprev)
1855{
1856	struct vm_area_struct *vma;
1857	MA_STATE(mas, &mm->mm_mt, addr, addr);
1858
1859	vma = mas_walk(&mas);
1860	*pprev = mas_prev(&mas, 0);
1861	if (!vma)
1862		vma = mas_next(&mas, ULONG_MAX);
 
 
 
 
 
 
 
1863	return vma;
1864}
1865
1866/*
1867 * Verify that the stack growth is acceptable and
1868 * update accounting. This is shared with both the
1869 * grow-up and grow-down cases.
1870 */
1871static int acct_stack_growth(struct vm_area_struct *vma,
1872			     unsigned long size, unsigned long grow)
1873{
1874	struct mm_struct *mm = vma->vm_mm;
 
1875	unsigned long new_start;
1876
1877	/* address space limit tests */
1878	if (!may_expand_vm(mm, vma->vm_flags, grow))
1879		return -ENOMEM;
1880
1881	/* Stack limit test */
1882	if (size > rlimit(RLIMIT_STACK))
1883		return -ENOMEM;
1884
1885	/* mlock limit tests */
1886	if (mlock_future_check(mm, vma->vm_flags, grow << PAGE_SHIFT))
1887		return -ENOMEM;
 
 
 
 
 
 
 
1888
1889	/* Check to ensure the stack will not grow into a hugetlb-only region */
1890	new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start :
1891			vma->vm_end - size;
1892	if (is_hugepage_only_range(vma->vm_mm, new_start, size))
1893		return -EFAULT;
1894
1895	/*
1896	 * Overcommit..  This must be the final test, as it will
1897	 * update security statistics.
1898	 */
1899	if (security_vm_enough_memory_mm(mm, grow))
1900		return -ENOMEM;
1901
 
 
 
 
 
1902	return 0;
1903}
1904
1905#if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
1906/*
1907 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
1908 * vma is the last one with address > vma->vm_end.  Have to extend vma.
1909 */
1910int expand_upwards(struct vm_area_struct *vma, unsigned long address)
1911{
1912	struct mm_struct *mm = vma->vm_mm;
1913	struct vm_area_struct *next;
1914	unsigned long gap_addr;
1915	int error = 0;
1916	MA_STATE(mas, &mm->mm_mt, 0, 0);
1917
1918	if (!(vma->vm_flags & VM_GROWSUP))
1919		return -EFAULT;
1920
1921	/* Guard against exceeding limits of the address space. */
1922	address &= PAGE_MASK;
1923	if (address >= (TASK_SIZE & PAGE_MASK))
1924		return -ENOMEM;
1925	address += PAGE_SIZE;
1926
1927	/* Enforce stack_guard_gap */
1928	gap_addr = address + stack_guard_gap;
1929
1930	/* Guard against overflow */
1931	if (gap_addr < address || gap_addr > TASK_SIZE)
1932		gap_addr = TASK_SIZE;
1933
1934	next = find_vma_intersection(mm, vma->vm_end, gap_addr);
1935	if (next && vma_is_accessible(next)) {
1936		if (!(next->vm_flags & VM_GROWSUP))
1937			return -ENOMEM;
1938		/* Check that both stack segments have the same anon_vma? */
1939	}
1940
1941	if (mas_preallocate(&mas, vma, GFP_KERNEL))
1942		return -ENOMEM;
1943
1944	/* We must make sure the anon_vma is allocated. */
1945	if (unlikely(anon_vma_prepare(vma))) {
1946		mas_destroy(&mas);
1947		return -ENOMEM;
1948	}
1949
1950	/*
1951	 * vma->vm_start/vm_end cannot change under us because the caller
1952	 * is required to hold the mmap_lock in read mode.  We need the
1953	 * anon_vma lock to serialize against concurrent expand_stacks.
 
1954	 */
1955	anon_vma_lock_write(vma->anon_vma);
 
 
 
 
 
 
1956
1957	/* Somebody else might have raced and expanded it already */
1958	if (address > vma->vm_end) {
1959		unsigned long size, grow;
1960
1961		size = address - vma->vm_start;
1962		grow = (address - vma->vm_end) >> PAGE_SHIFT;
1963
1964		error = -ENOMEM;
1965		if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) {
1966			error = acct_stack_growth(vma, size, grow);
1967			if (!error) {
1968				/*
1969				 * We only hold a shared mmap_lock lock here, so
1970				 * we need to protect against concurrent vma
1971				 * expansions.  anon_vma_lock_write() doesn't
1972				 * help here, as we don't guarantee that all
1973				 * growable vmas in a mm share the same root
1974				 * anon vma.  So, we reuse mm->page_table_lock
1975				 * to guard against concurrent vma expansions.
1976				 */
1977				spin_lock(&mm->page_table_lock);
1978				if (vma->vm_flags & VM_LOCKED)
1979					mm->locked_vm += grow;
1980				vm_stat_account(mm, vma->vm_flags, grow);
1981				anon_vma_interval_tree_pre_update_vma(vma);
1982				vma->vm_end = address;
1983				/* Overwrite old entry in mtree. */
1984				vma_mas_store(vma, &mas);
1985				anon_vma_interval_tree_post_update_vma(vma);
1986				spin_unlock(&mm->page_table_lock);
1987
1988				perf_event_mmap(vma);
1989			}
1990		}
1991	}
1992	anon_vma_unlock_write(vma->anon_vma);
1993	khugepaged_enter_vma(vma, vma->vm_flags);
1994	mas_destroy(&mas);
1995	return error;
1996}
1997#endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
1998
1999/*
2000 * vma is the first one with address < vma->vm_start.  Have to extend vma.
2001 */
2002int expand_downwards(struct vm_area_struct *vma, unsigned long address)
 
2003{
2004	struct mm_struct *mm = vma->vm_mm;
2005	MA_STATE(mas, &mm->mm_mt, vma->vm_start, vma->vm_start);
2006	struct vm_area_struct *prev;
2007	int error = 0;
2008
2009	address &= PAGE_MASK;
2010	if (address < mmap_min_addr)
2011		return -EPERM;
2012
2013	/* Enforce stack_guard_gap */
2014	prev = mas_prev(&mas, 0);
2015	/* Check that both stack segments have the same anon_vma? */
2016	if (prev && !(prev->vm_flags & VM_GROWSDOWN) &&
2017			vma_is_accessible(prev)) {
2018		if (address - prev->vm_end < stack_guard_gap)
2019			return -ENOMEM;
2020	}
2021
2022	if (mas_preallocate(&mas, vma, GFP_KERNEL))
 
 
 
 
2023		return -ENOMEM;
2024
2025	/* We must make sure the anon_vma is allocated. */
2026	if (unlikely(anon_vma_prepare(vma))) {
2027		mas_destroy(&mas);
2028		return -ENOMEM;
2029	}
 
2030
2031	/*
2032	 * vma->vm_start/vm_end cannot change under us because the caller
2033	 * is required to hold the mmap_lock in read mode.  We need the
2034	 * anon_vma lock to serialize against concurrent expand_stacks.
2035	 */
2036	anon_vma_lock_write(vma->anon_vma);
2037
2038	/* Somebody else might have raced and expanded it already */
2039	if (address < vma->vm_start) {
2040		unsigned long size, grow;
2041
2042		size = vma->vm_end - address;
2043		grow = (vma->vm_start - address) >> PAGE_SHIFT;
2044
2045		error = -ENOMEM;
2046		if (grow <= vma->vm_pgoff) {
2047			error = acct_stack_growth(vma, size, grow);
2048			if (!error) {
2049				/*
2050				 * We only hold a shared mmap_lock lock here, so
2051				 * we need to protect against concurrent vma
2052				 * expansions.  anon_vma_lock_write() doesn't
2053				 * help here, as we don't guarantee that all
2054				 * growable vmas in a mm share the same root
2055				 * anon vma.  So, we reuse mm->page_table_lock
2056				 * to guard against concurrent vma expansions.
2057				 */
2058				spin_lock(&mm->page_table_lock);
2059				if (vma->vm_flags & VM_LOCKED)
2060					mm->locked_vm += grow;
2061				vm_stat_account(mm, vma->vm_flags, grow);
2062				anon_vma_interval_tree_pre_update_vma(vma);
2063				vma->vm_start = address;
2064				vma->vm_pgoff -= grow;
2065				/* Overwrite old entry in mtree. */
2066				vma_mas_store(vma, &mas);
2067				anon_vma_interval_tree_post_update_vma(vma);
2068				spin_unlock(&mm->page_table_lock);
2069
2070				perf_event_mmap(vma);
2071			}
2072		}
2073	}
2074	anon_vma_unlock_write(vma->anon_vma);
2075	khugepaged_enter_vma(vma, vma->vm_flags);
2076	mas_destroy(&mas);
2077	return error;
2078}
2079
2080/* enforced gap between the expanding stack and other mappings. */
2081unsigned long stack_guard_gap = 256UL<<PAGE_SHIFT;
2082
2083static int __init cmdline_parse_stack_guard_gap(char *p)
2084{
2085	unsigned long val;
2086	char *endptr;
2087
2088	val = simple_strtoul(p, &endptr, 10);
2089	if (!*endptr)
2090		stack_guard_gap = val << PAGE_SHIFT;
2091
2092	return 1;
2093}
2094__setup("stack_guard_gap=", cmdline_parse_stack_guard_gap);
2095
2096#ifdef CONFIG_STACK_GROWSUP
2097int expand_stack(struct vm_area_struct *vma, unsigned long address)
2098{
2099	return expand_upwards(vma, address);
2100}
2101
2102struct vm_area_struct *
2103find_extend_vma(struct mm_struct *mm, unsigned long addr)
2104{
2105	struct vm_area_struct *vma, *prev;
2106
2107	addr &= PAGE_MASK;
2108	vma = find_vma_prev(mm, addr, &prev);
2109	if (vma && (vma->vm_start <= addr))
2110		return vma;
2111	if (!prev || expand_stack(prev, addr))
2112		return NULL;
2113	if (prev->vm_flags & VM_LOCKED)
2114		populate_vma_page_range(prev, addr, prev->vm_end, NULL);
 
2115	return prev;
2116}
2117#else
2118int expand_stack(struct vm_area_struct *vma, unsigned long address)
2119{
2120	return expand_downwards(vma, address);
2121}
2122
2123struct vm_area_struct *
2124find_extend_vma(struct mm_struct *mm, unsigned long addr)
2125{
2126	struct vm_area_struct *vma;
2127	unsigned long start;
2128
2129	addr &= PAGE_MASK;
2130	vma = find_vma(mm, addr);
2131	if (!vma)
2132		return NULL;
2133	if (vma->vm_start <= addr)
2134		return vma;
2135	if (!(vma->vm_flags & VM_GROWSDOWN))
2136		return NULL;
2137	start = vma->vm_start;
2138	if (expand_stack(vma, addr))
2139		return NULL;
2140	if (vma->vm_flags & VM_LOCKED)
2141		populate_vma_page_range(vma, addr, start, NULL);
 
2142	return vma;
2143}
2144#endif
2145
2146EXPORT_SYMBOL_GPL(find_extend_vma);
2147
2148/*
2149 * Ok - we have the memory areas we should free on a maple tree so release them,
2150 * and do the vma updates.
2151 *
2152 * Called with the mm semaphore held.
2153 */
2154static inline void remove_mt(struct mm_struct *mm, struct ma_state *mas)
2155{
2156	unsigned long nr_accounted = 0;
2157	struct vm_area_struct *vma;
2158
2159	/* Update high watermark before we lower total_vm */
2160	update_hiwater_vm(mm);
2161	mas_for_each(mas, vma, ULONG_MAX) {
2162		long nrpages = vma_pages(vma);
2163
2164		if (vma->vm_flags & VM_ACCOUNT)
2165			nr_accounted += nrpages;
2166		vm_stat_account(mm, vma->vm_flags, -nrpages);
2167		remove_vma(vma);
2168	}
 
2169	vm_unacct_memory(nr_accounted);
2170	validate_mm(mm);
2171}
2172
2173/*
2174 * Get rid of page table information in the indicated region.
2175 *
2176 * Called with the mm semaphore held.
2177 */
2178static void unmap_region(struct mm_struct *mm, struct maple_tree *mt,
2179		struct vm_area_struct *vma, struct vm_area_struct *prev,
2180		struct vm_area_struct *next,
2181		unsigned long start, unsigned long end)
2182{
 
2183	struct mmu_gather tlb;
2184
2185	lru_add_drain();
2186	tlb_gather_mmu(&tlb, mm);
2187	update_hiwater_rss(mm);
2188	unmap_vmas(&tlb, mt, vma, start, end);
2189	free_pgtables(&tlb, mt, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS,
2190				 next ? next->vm_start : USER_PGTABLES_CEILING);
2191	tlb_finish_mmu(&tlb);
2192}
2193
2194/*
2195 * __split_vma() bypasses sysctl_max_map_count checking.  We use this where it
2196 * has already been checked or doesn't make sense to fail.
2197 */
2198int __split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
2199		unsigned long addr, int new_below)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2200{
 
2201	struct vm_area_struct *new;
2202	int err;
2203	validate_mm_mt(mm);
2204
2205	if (vma->vm_ops && vma->vm_ops->may_split) {
2206		err = vma->vm_ops->may_split(vma, addr);
2207		if (err)
2208			return err;
2209	}
2210
2211	new = vm_area_dup(vma);
2212	if (!new)
2213		return -ENOMEM;
 
 
 
 
 
2214
2215	if (new_below)
2216		new->vm_end = addr;
2217	else {
2218		new->vm_start = addr;
2219		new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
2220	}
2221
2222	err = vma_dup_policy(vma, new);
2223	if (err)
 
2224		goto out_free_vma;
 
 
2225
2226	err = anon_vma_clone(new, vma);
2227	if (err)
2228		goto out_free_mpol;
2229
2230	if (new->vm_file)
2231		get_file(new->vm_file);
 
 
 
2232
2233	if (new->vm_ops && new->vm_ops->open)
2234		new->vm_ops->open(new);
2235
2236	if (new_below)
2237		err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
2238			((addr - new->vm_start) >> PAGE_SHIFT), new);
2239	else
2240		err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
2241
2242	/* Success. */
2243	if (!err)
2244		return 0;
2245
2246	/* Avoid vm accounting in close() operation */
2247	new->vm_start = new->vm_end;
2248	new->vm_pgoff = 0;
2249	/* Clean everything up if vma_adjust failed. */
2250	if (new->vm_ops && new->vm_ops->close)
2251		new->vm_ops->close(new);
2252	if (new->vm_file)
 
 
2253		fput(new->vm_file);
 
2254	unlink_anon_vmas(new);
2255 out_free_mpol:
2256	mpol_put(vma_policy(new));
2257 out_free_vma:
2258	vm_area_free(new);
2259	validate_mm_mt(mm);
2260	return err;
2261}
2262
2263/*
2264 * Split a vma into two pieces at address 'addr', a new vma is allocated
2265 * either for the first part or the tail.
2266 */
2267int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
2268	      unsigned long addr, int new_below)
2269{
2270	if (mm->map_count >= sysctl_max_map_count)
2271		return -ENOMEM;
2272
2273	return __split_vma(mm, vma, addr, new_below);
2274}
2275
2276static inline int munmap_sidetree(struct vm_area_struct *vma,
2277				   struct ma_state *mas_detach)
 
 
 
 
2278{
2279	mas_set_range(mas_detach, vma->vm_start, vma->vm_end - 1);
2280	if (mas_store_gfp(mas_detach, vma, GFP_KERNEL))
2281		return -ENOMEM;
2282
2283	if (vma->vm_flags & VM_LOCKED)
2284		vma->vm_mm->locked_vm -= vma_pages(vma);
2285
2286	return 0;
2287}
2288
2289/*
2290 * do_mas_align_munmap() - munmap the aligned region from @start to @end.
2291 * @mas: The maple_state, ideally set up to alter the correct tree location.
2292 * @vma: The starting vm_area_struct
2293 * @mm: The mm_struct
2294 * @start: The aligned start address to munmap.
2295 * @end: The aligned end address to munmap.
2296 * @uf: The userfaultfd list_head
2297 * @downgrade: Set to true to attempt a write downgrade of the mmap_lock
2298 *
2299 * If @downgrade is true, check return code for potential release of the lock.
2300 */
2301static int
2302do_mas_align_munmap(struct ma_state *mas, struct vm_area_struct *vma,
2303		    struct mm_struct *mm, unsigned long start,
2304		    unsigned long end, struct list_head *uf, bool downgrade)
2305{
2306	struct vm_area_struct *prev, *next = NULL;
2307	struct maple_tree mt_detach;
2308	int count = 0;
2309	int error = -ENOMEM;
2310	MA_STATE(mas_detach, &mt_detach, 0, 0);
2311	mt_init_flags(&mt_detach, MT_FLAGS_LOCK_EXTERN);
2312	mt_set_external_lock(&mt_detach, &mm->mmap_lock);
2313
2314	if (mas_preallocate(mas, vma, GFP_KERNEL))
2315		return -ENOMEM;
 
 
2316
2317	mas->last = end - 1;
2318	/*
2319	 * If we need to split any vma, do it now to save pain later.
2320	 *
2321	 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2322	 * unmapped vm_area_struct will remain in use: so lower split_vma
2323	 * places tmp vma above, and higher split_vma places tmp vma below.
2324	 */
2325
2326	/* Does it split the first one? */
2327	if (start > vma->vm_start) {
 
2328
2329		/*
2330		 * Make sure that map_count on return from munmap() will
2331		 * not exceed its limit; but let map_count go just above
2332		 * its limit temporarily, to help free resources as expected.
2333		 */
2334		if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count)
2335			goto map_count_exceeded;
2336
2337		/*
2338		 * mas_pause() is not needed since mas->index needs to be set
2339		 * differently than vma->vm_end anyways.
2340		 */
2341		error = __split_vma(mm, vma, start, 0);
2342		if (error)
2343			goto start_split_failed;
2344
2345		mas_set(mas, start);
2346		vma = mas_walk(mas);
2347	}
2348
2349	prev = mas_prev(mas, 0);
2350	if (unlikely((!prev)))
2351		mas_set(mas, start);
2352
2353	/*
2354	 * Detach a range of VMAs from the mm. Using next as a temp variable as
2355	 * it is always overwritten.
2356	 */
2357	mas_for_each(mas, next, end - 1) {
2358		/* Does it split the end? */
2359		if (next->vm_end > end) {
2360			struct vm_area_struct *split;
2361
2362			error = __split_vma(mm, next, end, 1);
2363			if (error)
2364				goto end_split_failed;
2365
2366			mas_set(mas, end);
2367			split = mas_prev(mas, 0);
2368			error = munmap_sidetree(split, &mas_detach);
2369			if (error)
2370				goto munmap_sidetree_failed;
2371
2372			count++;
2373			if (vma == next)
2374				vma = split;
2375			break;
2376		}
2377		error = munmap_sidetree(next, &mas_detach);
2378		if (error)
2379			goto munmap_sidetree_failed;
2380
2381		count++;
2382#ifdef CONFIG_DEBUG_VM_MAPLE_TREE
2383		BUG_ON(next->vm_start < start);
2384		BUG_ON(next->vm_start > end);
2385#endif
2386	}
2387
2388	if (!next)
2389		next = mas_next(mas, ULONG_MAX);
2390
2391	if (unlikely(uf)) {
2392		/*
2393		 * If userfaultfd_unmap_prep returns an error the vmas
2394		 * will remain split, but userland will get a
2395		 * highly unexpected error anyway. This is no
2396		 * different than the case where the first of the two
2397		 * __split_vma fails, but we don't undo the first
2398		 * split, despite we could. This is unlikely enough
2399		 * failure that it's not worth optimizing it for.
2400		 */
2401		error = userfaultfd_unmap_prep(mm, start, end, uf);
2402
2403		if (error)
2404			goto userfaultfd_error;
2405	}
2406
2407	/* Point of no return */
2408	mas_set_range(mas, start, end - 1);
2409#if defined(CONFIG_DEBUG_VM_MAPLE_TREE)
2410	/* Make sure no VMAs are about to be lost. */
2411	{
2412		MA_STATE(test, &mt_detach, start, end - 1);
2413		struct vm_area_struct *vma_mas, *vma_test;
2414		int test_count = 0;
2415
2416		rcu_read_lock();
2417		vma_test = mas_find(&test, end - 1);
2418		mas_for_each(mas, vma_mas, end - 1) {
2419			BUG_ON(vma_mas != vma_test);
2420			test_count++;
2421			vma_test = mas_next(&test, end - 1);
2422		}
2423		rcu_read_unlock();
2424		BUG_ON(count != test_count);
2425		mas_set_range(mas, start, end - 1);
2426	}
2427#endif
2428	mas_store_prealloc(mas, NULL);
2429	mm->map_count -= count;
2430	/*
2431	 * Do not downgrade mmap_lock if we are next to VM_GROWSDOWN or
2432	 * VM_GROWSUP VMA. Such VMAs can change their size under
2433	 * down_read(mmap_lock) and collide with the VMA we are about to unmap.
2434	 */
2435	if (downgrade) {
2436		if (next && (next->vm_flags & VM_GROWSDOWN))
2437			downgrade = false;
2438		else if (prev && (prev->vm_flags & VM_GROWSUP))
2439			downgrade = false;
2440		else
2441			mmap_write_downgrade(mm);
2442	}
2443
2444	unmap_region(mm, &mt_detach, vma, prev, next, start, end);
2445	/* Statistics and freeing VMAs */
2446	mas_set(&mas_detach, start);
2447	remove_mt(mm, &mas_detach);
2448	__mt_destroy(&mt_detach);
2449
2450
2451	validate_mm(mm);
2452	return downgrade ? 1 : 0;
2453
2454userfaultfd_error:
2455munmap_sidetree_failed:
2456end_split_failed:
2457	__mt_destroy(&mt_detach);
2458start_split_failed:
2459map_count_exceeded:
2460	mas_destroy(mas);
2461	return error;
2462}
2463
2464/*
2465 * do_mas_munmap() - munmap a given range.
2466 * @mas: The maple state
2467 * @mm: The mm_struct
2468 * @start: The start address to munmap
2469 * @len: The length of the range to munmap
2470 * @uf: The userfaultfd list_head
2471 * @downgrade: set to true if the user wants to attempt to write_downgrade the
2472 * mmap_lock
2473 *
2474 * This function takes a @mas that is either pointing to the previous VMA or set
2475 * to MA_START and sets it up to remove the mapping(s).  The @len will be
2476 * aligned and any arch_unmap work will be preformed.
2477 *
2478 * Returns: -EINVAL on failure, 1 on success and unlock, 0 otherwise.
2479 */
2480int do_mas_munmap(struct ma_state *mas, struct mm_struct *mm,
2481		  unsigned long start, size_t len, struct list_head *uf,
2482		  bool downgrade)
2483{
2484	unsigned long end;
2485	struct vm_area_struct *vma;
2486
2487	if ((offset_in_page(start)) || start > TASK_SIZE || len > TASK_SIZE-start)
2488		return -EINVAL;
2489
2490	end = start + PAGE_ALIGN(len);
2491	if (end == start)
2492		return -EINVAL;
2493
2494	 /* arch_unmap() might do unmaps itself.  */
2495	arch_unmap(mm, start, end);
2496
2497	/* Find the first overlapping VMA */
2498	vma = mas_find(mas, end - 1);
2499	if (!vma)
2500		return 0;
2501
2502	return do_mas_align_munmap(mas, vma, mm, start, end, uf, downgrade);
2503}
2504
2505/* do_munmap() - Wrapper function for non-maple tree aware do_munmap() calls.
2506 * @mm: The mm_struct
2507 * @start: The start address to munmap
2508 * @len: The length to be munmapped.
2509 * @uf: The userfaultfd list_head
2510 */
2511int do_munmap(struct mm_struct *mm, unsigned long start, size_t len,
2512	      struct list_head *uf)
2513{
2514	MA_STATE(mas, &mm->mm_mt, start, start);
2515
2516	return do_mas_munmap(&mas, mm, start, len, uf, false);
2517}
2518
2519unsigned long mmap_region(struct file *file, unsigned long addr,
2520		unsigned long len, vm_flags_t vm_flags, unsigned long pgoff,
2521		struct list_head *uf)
2522{
2523	struct mm_struct *mm = current->mm;
2524	struct vm_area_struct *vma = NULL;
2525	struct vm_area_struct *next, *prev, *merge;
2526	pgoff_t pglen = len >> PAGE_SHIFT;
2527	unsigned long charged = 0;
2528	unsigned long end = addr + len;
2529	unsigned long merge_start = addr, merge_end = end;
2530	pgoff_t vm_pgoff;
2531	int error;
2532	MA_STATE(mas, &mm->mm_mt, addr, end - 1);
2533
2534	/* Check against address space limit. */
2535	if (!may_expand_vm(mm, vm_flags, len >> PAGE_SHIFT)) {
2536		unsigned long nr_pages;
2537
2538		/*
2539		 * MAP_FIXED may remove pages of mappings that intersects with
2540		 * requested mapping. Account for the pages it would unmap.
2541		 */
2542		nr_pages = count_vma_pages_range(mm, addr, end);
2543
2544		if (!may_expand_vm(mm, vm_flags,
2545					(len >> PAGE_SHIFT) - nr_pages))
2546			return -ENOMEM;
2547	}
2548
2549	/* Unmap any existing mapping in the area */
2550	if (do_mas_munmap(&mas, mm, addr, len, uf, false))
2551		return -ENOMEM;
2552
2553	/*
2554	 * Private writable mapping: check memory availability
2555	 */
2556	if (accountable_mapping(file, vm_flags)) {
2557		charged = len >> PAGE_SHIFT;
2558		if (security_vm_enough_memory_mm(mm, charged))
2559			return -ENOMEM;
2560		vm_flags |= VM_ACCOUNT;
2561	}
2562
2563	next = mas_next(&mas, ULONG_MAX);
2564	prev = mas_prev(&mas, 0);
2565	if (vm_flags & VM_SPECIAL)
2566		goto cannot_expand;
2567
2568	/* Attempt to expand an old mapping */
2569	/* Check next */
2570	if (next && next->vm_start == end && !vma_policy(next) &&
2571	    can_vma_merge_before(next, vm_flags, NULL, file, pgoff+pglen,
2572				 NULL_VM_UFFD_CTX, NULL)) {
2573		merge_end = next->vm_end;
2574		vma = next;
2575		vm_pgoff = next->vm_pgoff - pglen;
2576	}
2577
2578	/* Check prev */
2579	if (prev && prev->vm_end == addr && !vma_policy(prev) &&
2580	    (vma ? can_vma_merge_after(prev, vm_flags, vma->anon_vma, file,
2581				       pgoff, vma->vm_userfaultfd_ctx, NULL) :
2582		   can_vma_merge_after(prev, vm_flags, NULL, file, pgoff,
2583				       NULL_VM_UFFD_CTX, NULL))) {
2584		merge_start = prev->vm_start;
2585		vma = prev;
2586		vm_pgoff = prev->vm_pgoff;
2587	}
 
2588
2589
2590	/* Actually expand, if possible */
2591	if (vma &&
2592	    !vma_expand(&mas, vma, merge_start, merge_end, vm_pgoff, next)) {
2593		khugepaged_enter_vma(vma, vm_flags);
2594		goto expanded;
2595	}
2596
2597	mas.index = addr;
2598	mas.last = end - 1;
2599cannot_expand:
2600	/*
2601	 * Determine the object being mapped and call the appropriate
2602	 * specific mapper. the address has already been validated, but
2603	 * not unmapped, but the maps are removed from the list.
2604	 */
2605	vma = vm_area_alloc(mm);
2606	if (!vma) {
2607		error = -ENOMEM;
2608		goto unacct_error;
2609	}
2610
2611	vma->vm_start = addr;
2612	vma->vm_end = end;
2613	vma->vm_flags = vm_flags;
2614	vma->vm_page_prot = vm_get_page_prot(vm_flags);
2615	vma->vm_pgoff = pgoff;
2616
2617	if (file) {
2618		if (vm_flags & VM_SHARED) {
2619			error = mapping_map_writable(file->f_mapping);
2620			if (error)
2621				goto free_vma;
2622		}
2623
2624		vma->vm_file = get_file(file);
2625		error = call_mmap(file, vma);
2626		if (error)
2627			goto unmap_and_free_vma;
2628
2629		/*
2630		 * Expansion is handled above, merging is handled below.
2631		 * Drivers should not alter the address of the VMA.
2632		 */
2633		if (WARN_ON((addr != vma->vm_start))) {
2634			error = -EINVAL;
2635			goto close_and_free_vma;
2636		}
2637		mas_reset(&mas);
2638
2639		/*
2640		 * If vm_flags changed after call_mmap(), we should try merge
2641		 * vma again as we may succeed this time.
2642		 */
2643		if (unlikely(vm_flags != vma->vm_flags && prev)) {
2644			merge = vma_merge(mm, prev, vma->vm_start, vma->vm_end, vma->vm_flags,
2645				NULL, vma->vm_file, vma->vm_pgoff, NULL, NULL_VM_UFFD_CTX, NULL);
2646			if (merge) {
2647				/*
2648				 * ->mmap() can change vma->vm_file and fput
2649				 * the original file. So fput the vma->vm_file
2650				 * here or we would add an extra fput for file
2651				 * and cause general protection fault
2652				 * ultimately.
2653				 */
2654				fput(vma->vm_file);
2655				vm_area_free(vma);
2656				vma = merge;
2657				/* Update vm_flags to pick up the change. */
2658				vm_flags = vma->vm_flags;
2659				goto unmap_writable;
2660			}
 
2661		}
2662
2663		vm_flags = vma->vm_flags;
2664	} else if (vm_flags & VM_SHARED) {
2665		error = shmem_zero_setup(vma);
2666		if (error)
2667			goto free_vma;
2668	} else {
2669		vma_set_anonymous(vma);
2670	}
2671
2672	/* Allow architectures to sanity-check the vm_flags */
2673	if (!arch_validate_flags(vma->vm_flags)) {
2674		error = -EINVAL;
2675		if (file)
2676			goto close_and_free_vma;
2677		else if (vma->vm_file)
2678			goto unmap_and_free_vma;
2679		else
2680			goto free_vma;
2681	}
2682
2683	if (mas_preallocate(&mas, vma, GFP_KERNEL)) {
2684		error = -ENOMEM;
2685		if (file)
2686			goto close_and_free_vma;
2687		else if (vma->vm_file)
2688			goto unmap_and_free_vma;
2689		else
2690			goto free_vma;
2691	}
2692
2693	if (vma->vm_file)
2694		i_mmap_lock_write(vma->vm_file->f_mapping);
2695
2696	vma_mas_store(vma, &mas);
2697	mm->map_count++;
2698	if (vma->vm_file) {
2699		if (vma->vm_flags & VM_SHARED)
2700			mapping_allow_writable(vma->vm_file->f_mapping);
2701
2702		flush_dcache_mmap_lock(vma->vm_file->f_mapping);
2703		vma_interval_tree_insert(vma, &vma->vm_file->f_mapping->i_mmap);
2704		flush_dcache_mmap_unlock(vma->vm_file->f_mapping);
2705		i_mmap_unlock_write(vma->vm_file->f_mapping);
2706	}
2707
2708	/*
2709	 * vma_merge() calls khugepaged_enter_vma() either, the below
2710	 * call covers the non-merge case.
2711	 */
2712	khugepaged_enter_vma(vma, vma->vm_flags);
2713
2714	/* Once vma denies write, undo our temporary denial count */
2715unmap_writable:
2716	if (file && vm_flags & VM_SHARED)
2717		mapping_unmap_writable(file->f_mapping);
2718	file = vma->vm_file;
2719expanded:
2720	perf_event_mmap(vma);
2721
2722	vm_stat_account(mm, vm_flags, len >> PAGE_SHIFT);
2723	if (vm_flags & VM_LOCKED) {
2724		if ((vm_flags & VM_SPECIAL) || vma_is_dax(vma) ||
2725					is_vm_hugetlb_page(vma) ||
2726					vma == get_gate_vma(current->mm))
2727			vma->vm_flags &= VM_LOCKED_CLEAR_MASK;
2728		else
2729			mm->locked_vm += (len >> PAGE_SHIFT);
2730	}
2731
2732	if (file)
2733		uprobe_mmap(vma);
2734
2735	/*
2736	 * New (or expanded) vma always get soft dirty status.
2737	 * Otherwise user-space soft-dirty page tracker won't
2738	 * be able to distinguish situation when vma area unmapped,
2739	 * then new mapped in-place (which must be aimed as
2740	 * a completely new data area).
2741	 */
2742	vma->vm_flags |= VM_SOFTDIRTY;
 
2743
2744	vma_set_page_prot(vma);
 
2745
2746	validate_mm(mm);
2747	return addr;
2748
2749close_and_free_vma:
2750	if (vma->vm_ops && vma->vm_ops->close)
2751		vma->vm_ops->close(vma);
2752unmap_and_free_vma:
2753	fput(vma->vm_file);
2754	vma->vm_file = NULL;
2755
2756	/* Undo any partial mapping done by a device driver. */
2757	unmap_region(mm, mas.tree, vma, prev, next, vma->vm_start, vma->vm_end);
2758	if (file && (vm_flags & VM_SHARED))
2759		mapping_unmap_writable(file->f_mapping);
2760free_vma:
2761	vm_area_free(vma);
2762unacct_error:
2763	if (charged)
2764		vm_unacct_memory(charged);
2765	validate_mm(mm);
2766	return error;
2767}
2768
2769static int __vm_munmap(unsigned long start, size_t len, bool downgrade)
2770{
2771	int ret;
2772	struct mm_struct *mm = current->mm;
2773	LIST_HEAD(uf);
2774	MA_STATE(mas, &mm->mm_mt, start, start);
2775
2776	if (mmap_write_lock_killable(mm))
2777		return -EINTR;
2778
2779	ret = do_mas_munmap(&mas, mm, start, len, &uf, downgrade);
2780	/*
2781	 * Returning 1 indicates mmap_lock is downgraded.
2782	 * But 1 is not legal return value of vm_munmap() and munmap(), reset
2783	 * it to 0 before return.
2784	 */
2785	if (ret == 1) {
2786		mmap_read_unlock(mm);
2787		ret = 0;
2788	} else
2789		mmap_write_unlock(mm);
2790
2791	userfaultfd_unmap_complete(mm, &uf);
 
 
2792	return ret;
2793}
2794
2795int vm_munmap(unsigned long start, size_t len)
2796{
2797	return __vm_munmap(start, len, false);
2798}
2799EXPORT_SYMBOL(vm_munmap);
2800
2801SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
2802{
2803	addr = untagged_addr(addr);
2804	return __vm_munmap(addr, len, true);
2805}
2806
 
 
 
 
 
 
 
 
 
2807
2808/*
2809 * Emulation of deprecated remap_file_pages() syscall.
 
 
2810 */
2811SYSCALL_DEFINE5(remap_file_pages, unsigned long, start, unsigned long, size,
2812		unsigned long, prot, unsigned long, pgoff, unsigned long, flags)
2813{
 
 
 
 
 
 
2814
2815	struct mm_struct *mm = current->mm;
2816	struct vm_area_struct *vma;
2817	unsigned long populate = 0;
2818	unsigned long ret = -EINVAL;
2819	struct file *file;
2820
2821	pr_warn_once("%s (%d) uses deprecated remap_file_pages() syscall. See Documentation/mm/remap_file_pages.rst.\n",
2822		     current->comm, current->pid);
2823
2824	if (prot)
2825		return ret;
2826	start = start & PAGE_MASK;
2827	size = size & PAGE_MASK;
2828
2829	if (start + size <= start)
2830		return ret;
2831
2832	/* Does pgoff wrap? */
2833	if (pgoff + (size >> PAGE_SHIFT) < pgoff)
2834		return ret;
2835
2836	if (mmap_write_lock_killable(mm))
2837		return -EINTR;
2838
2839	vma = vma_lookup(mm, start);
2840
2841	if (!vma || !(vma->vm_flags & VM_SHARED))
2842		goto out;
2843
2844	if (start + size > vma->vm_end) {
2845		VMA_ITERATOR(vmi, mm, vma->vm_end);
2846		struct vm_area_struct *next, *prev = vma;
2847
2848		for_each_vma_range(vmi, next, start + size) {
2849			/* hole between vmas ? */
2850			if (next->vm_start != prev->vm_end)
2851				goto out;
2852
2853			if (next->vm_file != vma->vm_file)
2854				goto out;
2855
2856			if (next->vm_flags != vma->vm_flags)
2857				goto out;
2858
2859			if (start + size <= next->vm_end)
2860				break;
2861
2862			prev = next;
2863		}
2864
2865		if (!next)
2866			goto out;
 
 
 
 
 
 
 
 
 
2867	}
2868
2869	prot |= vma->vm_flags & VM_READ ? PROT_READ : 0;
2870	prot |= vma->vm_flags & VM_WRITE ? PROT_WRITE : 0;
2871	prot |= vma->vm_flags & VM_EXEC ? PROT_EXEC : 0;
2872
2873	flags &= MAP_NONBLOCK;
2874	flags |= MAP_SHARED | MAP_FIXED | MAP_POPULATE;
2875	if (vma->vm_flags & VM_LOCKED)
2876		flags |= MAP_LOCKED;
2877
2878	file = get_file(vma->vm_file);
2879	ret = do_mmap(vma->vm_file, start, size,
2880			prot, flags, pgoff, &populate, NULL);
2881	fput(file);
2882out:
2883	mmap_write_unlock(mm);
2884	if (populate)
2885		mm_populate(ret, populate);
2886	if (!IS_ERR_VALUE(ret))
2887		ret = 0;
2888	return ret;
2889}
2890
2891/*
2892 * brk_munmap() - Unmap a parital vma.
2893 * @mas: The maple tree state.
2894 * @vma: The vma to be modified
2895 * @newbrk: the start of the address to unmap
2896 * @oldbrk: The end of the address to unmap
2897 * @uf: The userfaultfd list_head
2898 *
2899 * Returns: 1 on success.
2900 * unmaps a partial VMA mapping.  Does not handle alignment, downgrades lock if
2901 * possible.
2902 */
2903static int do_brk_munmap(struct ma_state *mas, struct vm_area_struct *vma,
2904			 unsigned long newbrk, unsigned long oldbrk,
2905			 struct list_head *uf)
2906{
2907	struct mm_struct *mm = vma->vm_mm;
2908	int ret;
2909
2910	arch_unmap(mm, newbrk, oldbrk);
2911	ret = do_mas_align_munmap(mas, vma, mm, newbrk, oldbrk, uf, true);
2912	validate_mm_mt(mm);
2913	return ret;
2914}
2915
2916/*
2917 * do_brk_flags() - Increase the brk vma if the flags match.
2918 * @mas: The maple tree state.
2919 * @addr: The start address
2920 * @len: The length of the increase
2921 * @vma: The vma,
2922 * @flags: The VMA Flags
2923 *
2924 * Extend the brk VMA from addr to addr + len.  If the VMA is NULL or the flags
2925 * do not match then create a new anonymous VMA.  Eventually we may be able to
2926 * do some brk-specific accounting here.
2927 */
2928static int do_brk_flags(struct ma_state *mas, struct vm_area_struct *vma,
2929		unsigned long addr, unsigned long len, unsigned long flags)
2930{
2931	struct mm_struct *mm = current->mm;
2932
2933	validate_mm_mt(mm);
2934	/*
2935	 * Check against address space limits by the changed size
2936	 * Note: This happens *after* clearing old mappings in some code paths.
2937	 */
2938	flags |= VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
2939	if (!may_expand_vm(mm, flags, len >> PAGE_SHIFT))
 
 
 
 
 
 
 
 
2940		return -ENOMEM;
2941
2942	if (mm->map_count > sysctl_max_map_count)
2943		return -ENOMEM;
2944
2945	if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT))
2946		return -ENOMEM;
2947
 
 
 
 
 
 
2948	/*
2949	 * Expand the existing vma if possible; Note that singular lists do not
2950	 * occur after forking, so the expand will only happen on new VMAs.
2951	 */
2952	if (vma && vma->vm_end == addr && !vma_policy(vma) &&
2953	    can_vma_merge_after(vma, flags, NULL, NULL,
2954				addr >> PAGE_SHIFT, NULL_VM_UFFD_CTX, NULL)) {
2955		mas_set_range(mas, vma->vm_start, addr + len - 1);
2956		if (mas_preallocate(mas, vma, GFP_KERNEL))
2957			goto unacct_fail;
2958
2959		vma_adjust_trans_huge(vma, vma->vm_start, addr + len, 0);
2960		if (vma->anon_vma) {
2961			anon_vma_lock_write(vma->anon_vma);
2962			anon_vma_interval_tree_pre_update_vma(vma);
2963		}
2964		vma->vm_end = addr + len;
2965		vma->vm_flags |= VM_SOFTDIRTY;
2966		mas_store_prealloc(mas, vma);
2967
2968		if (vma->anon_vma) {
2969			anon_vma_interval_tree_post_update_vma(vma);
2970			anon_vma_unlock_write(vma->anon_vma);
2971		}
2972		khugepaged_enter_vma(vma, flags);
2973		goto out;
2974	}
2975
2976	/* create a vma struct for an anonymous mapping */
2977	vma = vm_area_alloc(mm);
2978	if (!vma)
2979		goto unacct_fail;
2980
2981	vma_set_anonymous(vma);
2982	vma->vm_start = addr;
2983	vma->vm_end = addr + len;
2984	vma->vm_pgoff = addr >> PAGE_SHIFT;
2985	vma->vm_flags = flags;
2986	vma->vm_page_prot = vm_get_page_prot(flags);
2987	mas_set_range(mas, vma->vm_start, addr + len - 1);
2988	if (mas_store_gfp(mas, vma, GFP_KERNEL))
2989		goto mas_store_fail;
2990
2991	mm->map_count++;
2992out:
2993	perf_event_mmap(vma);
2994	mm->total_vm += len >> PAGE_SHIFT;
2995	mm->data_vm += len >> PAGE_SHIFT;
2996	if (flags & VM_LOCKED)
2997		mm->locked_vm += (len >> PAGE_SHIFT);
2998	vma->vm_flags |= VM_SOFTDIRTY;
2999	validate_mm(mm);
3000	return 0;
3001
3002mas_store_fail:
3003	vm_area_free(vma);
3004unacct_fail:
3005	vm_unacct_memory(len >> PAGE_SHIFT);
3006	return -ENOMEM;
3007}
3008
3009int vm_brk_flags(unsigned long addr, unsigned long request, unsigned long flags)
3010{
3011	struct mm_struct *mm = current->mm;
3012	struct vm_area_struct *vma = NULL;
3013	unsigned long len;
3014	int ret;
3015	bool populate;
3016	LIST_HEAD(uf);
3017	MA_STATE(mas, &mm->mm_mt, addr, addr);
3018
3019	len = PAGE_ALIGN(request);
3020	if (len < request)
3021		return -ENOMEM;
3022	if (!len)
3023		return 0;
3024
3025	if (mmap_write_lock_killable(mm))
3026		return -EINTR;
3027
3028	/* Until we need other flags, refuse anything except VM_EXEC. */
3029	if ((flags & (~VM_EXEC)) != 0)
3030		return -EINVAL;
3031
3032	ret = check_brk_limits(addr, len);
3033	if (ret)
3034		goto limits_failed;
3035
3036	ret = do_mas_munmap(&mas, mm, addr, len, &uf, 0);
3037	if (ret)
3038		goto munmap_failed;
3039
3040	vma = mas_prev(&mas, 0);
3041	ret = do_brk_flags(&mas, vma, addr, len, flags);
3042	populate = ((mm->def_flags & VM_LOCKED) != 0);
3043	mmap_write_unlock(mm);
3044	userfaultfd_unmap_complete(mm, &uf);
3045	if (populate && !ret)
3046		mm_populate(addr, len);
3047	return ret;
3048
3049munmap_failed:
3050limits_failed:
3051	mmap_write_unlock(mm);
3052	return ret;
3053}
3054EXPORT_SYMBOL(vm_brk_flags);
3055
3056int vm_brk(unsigned long addr, unsigned long len)
3057{
3058	return vm_brk_flags(addr, len, 0);
3059}
3060EXPORT_SYMBOL(vm_brk);
3061
3062/* Release all mmaps. */
3063void exit_mmap(struct mm_struct *mm)
3064{
3065	struct mmu_gather tlb;
3066	struct vm_area_struct *vma;
3067	unsigned long nr_accounted = 0;
3068	MA_STATE(mas, &mm->mm_mt, 0, 0);
3069	int count = 0;
3070
3071	/* mm's last user has gone, and its about to be pulled down */
3072	mmu_notifier_release(mm);
3073
3074	mmap_read_lock(mm);
 
 
 
 
 
 
 
 
3075	arch_exit_mmap(mm);
3076
3077	vma = mas_find(&mas, ULONG_MAX);
3078	if (!vma) {
3079		/* Can happen if dup_mmap() received an OOM */
3080		mmap_read_unlock(mm);
3081		return;
3082	}
3083
3084	lru_add_drain();
3085	flush_cache_mm(mm);
3086	tlb_gather_mmu_fullmm(&tlb, mm);
3087	/* update_hiwater_rss(mm) here? but nobody should be looking */
3088	/* Use ULONG_MAX here to ensure all VMAs in the mm are unmapped */
3089	unmap_vmas(&tlb, &mm->mm_mt, vma, 0, ULONG_MAX);
3090	mmap_read_unlock(mm);
3091
3092	/*
3093	 * Set MMF_OOM_SKIP to hide this task from the oom killer/reaper
3094	 * because the memory has been already freed.
3095	 */
3096	set_bit(MMF_OOM_SKIP, &mm->flags);
3097	mmap_write_lock(mm);
3098	free_pgtables(&tlb, &mm->mm_mt, vma, FIRST_USER_ADDRESS,
3099		      USER_PGTABLES_CEILING);
3100	tlb_finish_mmu(&tlb);
3101
3102	/*
3103	 * Walk the list again, actually closing and freeing it, with preemption
3104	 * enabled, without holding any MM locks besides the unreachable
3105	 * mmap_write_lock.
3106	 */
3107	do {
3108		if (vma->vm_flags & VM_ACCOUNT)
3109			nr_accounted += vma_pages(vma);
3110		remove_vma(vma);
3111		count++;
3112		cond_resched();
3113	} while ((vma = mas_find(&mas, ULONG_MAX)) != NULL);
3114
3115	BUG_ON(count != mm->map_count);
3116
3117	trace_exit_mmap(mm);
3118	__mt_destroy(&mm->mm_mt);
3119	mmap_write_unlock(mm);
3120	vm_unacct_memory(nr_accounted);
 
 
3121}
3122
3123/* Insert vm structure into process list sorted by address
3124 * and into the inode's i_mmap tree.  If vm_file is non-NULL
3125 * then i_mmap_rwsem is taken here.
3126 */
3127int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
3128{
3129	unsigned long charged = vma_pages(vma);
3130
3131
3132	if (find_vma_intersection(mm, vma->vm_start, vma->vm_end))
3133		return -ENOMEM;
3134
3135	if ((vma->vm_flags & VM_ACCOUNT) &&
3136	     security_vm_enough_memory_mm(mm, charged))
3137		return -ENOMEM;
3138
3139	/*
3140	 * The vm_pgoff of a purely anonymous vma should be irrelevant
3141	 * until its first write fault, when page's anon_vma and index
3142	 * are set.  But now set the vm_pgoff it will almost certainly
3143	 * end up with (unless mremap moves it elsewhere before that
3144	 * first wfault), so /proc/pid/maps tells a consistent story.
3145	 *
3146	 * By setting it to reflect the virtual start address of the
3147	 * vma, merges and splits can happen in a seamless way, just
3148	 * using the existing file pgoff checks and manipulations.
3149	 * Similarly in do_mmap and in do_brk_flags.
3150	 */
3151	if (vma_is_anonymous(vma)) {
3152		BUG_ON(vma->anon_vma);
3153		vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
3154	}
3155
3156	if (vma_link(mm, vma)) {
3157		vm_unacct_memory(charged);
3158		return -ENOMEM;
3159	}
 
 
 
 
 
3160
 
3161	return 0;
3162}
3163
3164/*
3165 * Copy the vma structure to a new location in the same mm,
3166 * prior to moving page table entries, to effect an mremap move.
3167 */
3168struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
3169	unsigned long addr, unsigned long len, pgoff_t pgoff,
3170	bool *need_rmap_locks)
3171{
3172	struct vm_area_struct *vma = *vmap;
3173	unsigned long vma_start = vma->vm_start;
3174	struct mm_struct *mm = vma->vm_mm;
3175	struct vm_area_struct *new_vma, *prev;
 
 
3176	bool faulted_in_anon_vma = true;
3177
3178	validate_mm_mt(mm);
3179	/*
3180	 * If anonymous vma has not yet been faulted, update new pgoff
3181	 * to match new location, to increase its chance of merging.
3182	 */
3183	if (unlikely(vma_is_anonymous(vma) && !vma->anon_vma)) {
3184		pgoff = addr >> PAGE_SHIFT;
3185		faulted_in_anon_vma = false;
3186	}
3187
3188	new_vma = find_vma_prev(mm, addr, &prev);
3189	if (new_vma && new_vma->vm_start < addr + len)
3190		return NULL;	/* should never get here */
3191
3192	new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
3193			    vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma),
3194			    vma->vm_userfaultfd_ctx, anon_vma_name(vma));
3195	if (new_vma) {
3196		/*
3197		 * Source vma may have been merged into new_vma
3198		 */
3199		if (unlikely(vma_start >= new_vma->vm_start &&
3200			     vma_start < new_vma->vm_end)) {
3201			/*
3202			 * The only way we can get a vma_merge with
3203			 * self during an mremap is if the vma hasn't
3204			 * been faulted in yet and we were allowed to
3205			 * reset the dst vma->vm_pgoff to the
3206			 * destination address of the mremap to allow
3207			 * the merge to happen. mremap must change the
3208			 * vm_pgoff linearity between src and dst vmas
3209			 * (in turn preventing a vma_merge) to be
3210			 * safe. It is only safe to keep the vm_pgoff
3211			 * linear if there are no pages mapped yet.
3212			 */
3213			VM_BUG_ON_VMA(faulted_in_anon_vma, new_vma);
3214			*vmap = vma = new_vma;
3215		}
3216		*need_rmap_locks = (new_vma->vm_pgoff <= vma->vm_pgoff);
3217	} else {
3218		new_vma = vm_area_dup(vma);
3219		if (!new_vma)
3220			goto out;
3221		new_vma->vm_start = addr;
3222		new_vma->vm_end = addr + len;
3223		new_vma->vm_pgoff = pgoff;
3224		if (vma_dup_policy(vma, new_vma))
3225			goto out_free_vma;
3226		if (anon_vma_clone(new_vma, vma))
3227			goto out_free_mempol;
3228		if (new_vma->vm_file)
3229			get_file(new_vma->vm_file);
3230		if (new_vma->vm_ops && new_vma->vm_ops->open)
3231			new_vma->vm_ops->open(new_vma);
3232		if (vma_link(mm, new_vma))
3233			goto out_vma_link;
3234		*need_rmap_locks = false;
 
 
 
 
 
 
 
 
 
3235	}
3236	validate_mm_mt(mm);
3237	return new_vma;
3238
3239out_vma_link:
3240	if (new_vma->vm_ops && new_vma->vm_ops->close)
3241		new_vma->vm_ops->close(new_vma);
3242
3243	if (new_vma->vm_file)
3244		fput(new_vma->vm_file);
3245
3246	unlink_anon_vmas(new_vma);
3247out_free_mempol:
3248	mpol_put(vma_policy(new_vma));
3249out_free_vma:
3250	vm_area_free(new_vma);
3251out:
3252	validate_mm_mt(mm);
3253	return NULL;
3254}
3255
3256/*
3257 * Return true if the calling process may expand its vm space by the passed
3258 * number of pages
3259 */
3260bool may_expand_vm(struct mm_struct *mm, vm_flags_t flags, unsigned long npages)
3261{
3262	if (mm->total_vm + npages > rlimit(RLIMIT_AS) >> PAGE_SHIFT)
3263		return false;
3264
3265	if (is_data_mapping(flags) &&
3266	    mm->data_vm + npages > rlimit(RLIMIT_DATA) >> PAGE_SHIFT) {
3267		/* Workaround for Valgrind */
3268		if (rlimit(RLIMIT_DATA) == 0 &&
3269		    mm->data_vm + npages <= rlimit_max(RLIMIT_DATA) >> PAGE_SHIFT)
3270			return true;
3271
3272		pr_warn_once("%s (%d): VmData %lu exceed data ulimit %lu. Update limits%s.\n",
3273			     current->comm, current->pid,
3274			     (mm->data_vm + npages) << PAGE_SHIFT,
3275			     rlimit(RLIMIT_DATA),
3276			     ignore_rlimit_data ? "" : " or use boot option ignore_rlimit_data");
3277
3278		if (!ignore_rlimit_data)
3279			return false;
3280	}
3281
3282	return true;
3283}
3284
3285void vm_stat_account(struct mm_struct *mm, vm_flags_t flags, long npages)
3286{
3287	WRITE_ONCE(mm->total_vm, READ_ONCE(mm->total_vm)+npages);
3288
3289	if (is_exec_mapping(flags))
3290		mm->exec_vm += npages;
3291	else if (is_stack_mapping(flags))
3292		mm->stack_vm += npages;
3293	else if (is_data_mapping(flags))
3294		mm->data_vm += npages;
3295}
3296
3297static vm_fault_t special_mapping_fault(struct vm_fault *vmf);
3298
3299/*
3300 * Having a close hook prevents vma merging regardless of flags.
3301 */
3302static void special_mapping_close(struct vm_area_struct *vma)
3303{
3304}
3305
3306static const char *special_mapping_name(struct vm_area_struct *vma)
3307{
3308	return ((struct vm_special_mapping *)vma->vm_private_data)->name;
3309}
3310
3311static int special_mapping_mremap(struct vm_area_struct *new_vma)
3312{
3313	struct vm_special_mapping *sm = new_vma->vm_private_data;
3314
3315	if (WARN_ON_ONCE(current->mm != new_vma->vm_mm))
3316		return -EFAULT;
3317
3318	if (sm->mremap)
3319		return sm->mremap(sm, new_vma);
3320
3321	return 0;
3322}
3323
3324static int special_mapping_split(struct vm_area_struct *vma, unsigned long addr)
3325{
3326	/*
3327	 * Forbid splitting special mappings - kernel has expectations over
3328	 * the number of pages in mapping. Together with VM_DONTEXPAND
3329	 * the size of vma should stay the same over the special mapping's
3330	 * lifetime.
3331	 */
3332	return -EINVAL;
3333}
3334
3335static const struct vm_operations_struct special_mapping_vmops = {
3336	.close = special_mapping_close,
3337	.fault = special_mapping_fault,
3338	.mremap = special_mapping_mremap,
3339	.name = special_mapping_name,
3340	/* vDSO code relies that VVAR can't be accessed remotely */
3341	.access = NULL,
3342	.may_split = special_mapping_split,
3343};
3344
3345static const struct vm_operations_struct legacy_special_mapping_vmops = {
3346	.close = special_mapping_close,
3347	.fault = special_mapping_fault,
3348};
3349
3350static vm_fault_t special_mapping_fault(struct vm_fault *vmf)
 
3351{
3352	struct vm_area_struct *vma = vmf->vma;
3353	pgoff_t pgoff;
3354	struct page **pages;
3355
3356	if (vma->vm_ops == &legacy_special_mapping_vmops) {
3357		pages = vma->vm_private_data;
3358	} else {
3359		struct vm_special_mapping *sm = vma->vm_private_data;
3360
3361		if (sm->fault)
3362			return sm->fault(sm, vmf->vma, vmf);
3363
3364		pages = sm->pages;
3365	}
3366
3367	for (pgoff = vmf->pgoff; pgoff && *pages; ++pages)
3368		pgoff--;
3369
3370	if (*pages) {
3371		struct page *page = *pages;
3372		get_page(page);
3373		vmf->page = page;
3374		return 0;
3375	}
3376
3377	return VM_FAULT_SIGBUS;
3378}
3379
3380static struct vm_area_struct *__install_special_mapping(
3381	struct mm_struct *mm,
3382	unsigned long addr, unsigned long len,
3383	unsigned long vm_flags, void *priv,
3384	const struct vm_operations_struct *ops)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3385{
3386	int ret;
3387	struct vm_area_struct *vma;
3388
3389	validate_mm_mt(mm);
3390	vma = vm_area_alloc(mm);
3391	if (unlikely(vma == NULL))
3392		return ERR_PTR(-ENOMEM);
3393
 
 
3394	vma->vm_start = addr;
3395	vma->vm_end = addr + len;
3396
3397	vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND | VM_SOFTDIRTY;
3398	vma->vm_flags &= VM_LOCKED_CLEAR_MASK;
3399	vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
3400
3401	vma->vm_ops = ops;
3402	vma->vm_private_data = priv;
3403
3404	ret = insert_vm_struct(mm, vma);
3405	if (ret)
3406		goto out;
3407
3408	vm_stat_account(mm, vma->vm_flags, len >> PAGE_SHIFT);
3409
3410	perf_event_mmap(vma);
3411
3412	validate_mm_mt(mm);
3413	return vma;
3414
3415out:
3416	vm_area_free(vma);
3417	validate_mm_mt(mm);
3418	return ERR_PTR(ret);
3419}
3420
3421bool vma_is_special_mapping(const struct vm_area_struct *vma,
3422	const struct vm_special_mapping *sm)
3423{
3424	return vma->vm_private_data == sm &&
3425		(vma->vm_ops == &special_mapping_vmops ||
3426		 vma->vm_ops == &legacy_special_mapping_vmops);
3427}
3428
3429/*
3430 * Called with mm->mmap_lock held for writing.
3431 * Insert a new vma covering the given region, with the given flags.
3432 * Its pages are supplied by the given array of struct page *.
3433 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
3434 * The region past the last page supplied will always produce SIGBUS.
3435 * The array pointer and the pages it points to are assumed to stay alive
3436 * for as long as this mapping might exist.
3437 */
3438struct vm_area_struct *_install_special_mapping(
3439	struct mm_struct *mm,
3440	unsigned long addr, unsigned long len,
3441	unsigned long vm_flags, const struct vm_special_mapping *spec)
3442{
3443	return __install_special_mapping(mm, addr, len, vm_flags, (void *)spec,
3444					&special_mapping_vmops);
3445}
3446
3447int install_special_mapping(struct mm_struct *mm,
3448			    unsigned long addr, unsigned long len,
3449			    unsigned long vm_flags, struct page **pages)
3450{
3451	struct vm_area_struct *vma = __install_special_mapping(
3452		mm, addr, len, vm_flags, (void *)pages,
3453		&legacy_special_mapping_vmops);
3454
3455	return PTR_ERR_OR_ZERO(vma);
3456}
3457
3458static DEFINE_MUTEX(mm_all_locks_mutex);
3459
3460static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma)
3461{
3462	if (!test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_root.rb_node)) {
3463		/*
3464		 * The LSB of head.next can't change from under us
3465		 * because we hold the mm_all_locks_mutex.
3466		 */
3467		down_write_nest_lock(&anon_vma->root->rwsem, &mm->mmap_lock);
3468		/*
3469		 * We can safely modify head.next after taking the
3470		 * anon_vma->root->rwsem. If some other vma in this mm shares
3471		 * the same anon_vma we won't take it again.
3472		 *
3473		 * No need of atomic instructions here, head.next
3474		 * can't change from under us thanks to the
3475		 * anon_vma->root->rwsem.
3476		 */
3477		if (__test_and_set_bit(0, (unsigned long *)
3478				       &anon_vma->root->rb_root.rb_root.rb_node))
3479			BUG();
3480	}
3481}
3482
3483static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping)
3484{
3485	if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
3486		/*
3487		 * AS_MM_ALL_LOCKS can't change from under us because
3488		 * we hold the mm_all_locks_mutex.
3489		 *
3490		 * Operations on ->flags have to be atomic because
3491		 * even if AS_MM_ALL_LOCKS is stable thanks to the
3492		 * mm_all_locks_mutex, there may be other cpus
3493		 * changing other bitflags in parallel to us.
3494		 */
3495		if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags))
3496			BUG();
3497		down_write_nest_lock(&mapping->i_mmap_rwsem, &mm->mmap_lock);
3498	}
3499}
3500
3501/*
3502 * This operation locks against the VM for all pte/vma/mm related
3503 * operations that could ever happen on a certain mm. This includes
3504 * vmtruncate, try_to_unmap, and all page faults.
3505 *
3506 * The caller must take the mmap_lock in write mode before calling
3507 * mm_take_all_locks(). The caller isn't allowed to release the
3508 * mmap_lock until mm_drop_all_locks() returns.
3509 *
3510 * mmap_lock in write mode is required in order to block all operations
3511 * that could modify pagetables and free pages without need of
3512 * altering the vma layout. It's also needed in write mode to avoid new
 
3513 * anon_vmas to be associated with existing vmas.
3514 *
3515 * A single task can't take more than one mm_take_all_locks() in a row
3516 * or it would deadlock.
3517 *
3518 * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in
3519 * mapping->flags avoid to take the same lock twice, if more than one
3520 * vma in this mm is backed by the same anon_vma or address_space.
3521 *
3522 * We take locks in following order, accordingly to comment at beginning
3523 * of mm/rmap.c:
3524 *   - all hugetlbfs_i_mmap_rwsem_key locks (aka mapping->i_mmap_rwsem for
3525 *     hugetlb mapping);
3526 *   - all i_mmap_rwsem locks;
3527 *   - all anon_vma->rwseml
3528 *
3529 * We can take all locks within these types randomly because the VM code
3530 * doesn't nest them and we protected from parallel mm_take_all_locks() by
3531 * mm_all_locks_mutex.
3532 *
3533 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
3534 * that may have to take thousand of locks.
3535 *
3536 * mm_take_all_locks() can fail if it's interrupted by signals.
3537 */
3538int mm_take_all_locks(struct mm_struct *mm)
3539{
3540	struct vm_area_struct *vma;
3541	struct anon_vma_chain *avc;
3542	MA_STATE(mas, &mm->mm_mt, 0, 0);
3543
3544	mmap_assert_write_locked(mm);
3545
3546	mutex_lock(&mm_all_locks_mutex);
3547
3548	mas_for_each(&mas, vma, ULONG_MAX) {
3549		if (signal_pending(current))
3550			goto out_unlock;
3551		if (vma->vm_file && vma->vm_file->f_mapping &&
3552				is_vm_hugetlb_page(vma))
3553			vm_lock_mapping(mm, vma->vm_file->f_mapping);
3554	}
3555
3556	mas_set(&mas, 0);
3557	mas_for_each(&mas, vma, ULONG_MAX) {
3558		if (signal_pending(current))
3559			goto out_unlock;
3560		if (vma->vm_file && vma->vm_file->f_mapping &&
3561				!is_vm_hugetlb_page(vma))
3562			vm_lock_mapping(mm, vma->vm_file->f_mapping);
3563	}
3564
3565	mas_set(&mas, 0);
3566	mas_for_each(&mas, vma, ULONG_MAX) {
3567		if (signal_pending(current))
3568			goto out_unlock;
3569		if (vma->anon_vma)
3570			list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
3571				vm_lock_anon_vma(mm, avc->anon_vma);
3572	}
3573
3574	return 0;
3575
3576out_unlock:
3577	mm_drop_all_locks(mm);
3578	return -EINTR;
3579}
3580
3581static void vm_unlock_anon_vma(struct anon_vma *anon_vma)
3582{
3583	if (test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_root.rb_node)) {
3584		/*
3585		 * The LSB of head.next can't change to 0 from under
3586		 * us because we hold the mm_all_locks_mutex.
3587		 *
3588		 * We must however clear the bitflag before unlocking
3589		 * the vma so the users using the anon_vma->rb_root will
3590		 * never see our bitflag.
3591		 *
3592		 * No need of atomic instructions here, head.next
3593		 * can't change from under us until we release the
3594		 * anon_vma->root->rwsem.
3595		 */
3596		if (!__test_and_clear_bit(0, (unsigned long *)
3597					  &anon_vma->root->rb_root.rb_root.rb_node))
3598			BUG();
3599		anon_vma_unlock_write(anon_vma);
3600	}
3601}
3602
3603static void vm_unlock_mapping(struct address_space *mapping)
3604{
3605	if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
3606		/*
3607		 * AS_MM_ALL_LOCKS can't change to 0 from under us
3608		 * because we hold the mm_all_locks_mutex.
3609		 */
3610		i_mmap_unlock_write(mapping);
3611		if (!test_and_clear_bit(AS_MM_ALL_LOCKS,
3612					&mapping->flags))
3613			BUG();
3614	}
3615}
3616
3617/*
3618 * The mmap_lock cannot be released by the caller until
3619 * mm_drop_all_locks() returns.
3620 */
3621void mm_drop_all_locks(struct mm_struct *mm)
3622{
3623	struct vm_area_struct *vma;
3624	struct anon_vma_chain *avc;
3625	MA_STATE(mas, &mm->mm_mt, 0, 0);
3626
3627	mmap_assert_write_locked(mm);
3628	BUG_ON(!mutex_is_locked(&mm_all_locks_mutex));
3629
3630	mas_for_each(&mas, vma, ULONG_MAX) {
3631		if (vma->anon_vma)
3632			list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
3633				vm_unlock_anon_vma(avc->anon_vma);
3634		if (vma->vm_file && vma->vm_file->f_mapping)
3635			vm_unlock_mapping(vma->vm_file->f_mapping);
3636	}
3637
3638	mutex_unlock(&mm_all_locks_mutex);
3639}
3640
3641/*
3642 * initialise the percpu counter for VM
3643 */
3644void __init mmap_init(void)
3645{
3646	int ret;
3647
3648	ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL);
3649	VM_BUG_ON(ret);
3650}
3651
3652/*
3653 * Initialise sysctl_user_reserve_kbytes.
3654 *
3655 * This is intended to prevent a user from starting a single memory hogging
3656 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
3657 * mode.
3658 *
3659 * The default value is min(3% of free memory, 128MB)
3660 * 128MB is enough to recover with sshd/login, bash, and top/kill.
3661 */
3662static int init_user_reserve(void)
3663{
3664	unsigned long free_kbytes;
3665
3666	free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3667
3668	sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17);
3669	return 0;
3670}
3671subsys_initcall(init_user_reserve);
3672
3673/*
3674 * Initialise sysctl_admin_reserve_kbytes.
3675 *
3676 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
3677 * to log in and kill a memory hogging process.
3678 *
3679 * Systems with more than 256MB will reserve 8MB, enough to recover
3680 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
3681 * only reserve 3% of free pages by default.
3682 */
3683static int init_admin_reserve(void)
3684{
3685	unsigned long free_kbytes;
3686
3687	free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3688
3689	sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13);
3690	return 0;
3691}
3692subsys_initcall(init_admin_reserve);
3693
3694/*
3695 * Reinititalise user and admin reserves if memory is added or removed.
3696 *
3697 * The default user reserve max is 128MB, and the default max for the
3698 * admin reserve is 8MB. These are usually, but not always, enough to
3699 * enable recovery from a memory hogging process using login/sshd, a shell,
3700 * and tools like top. It may make sense to increase or even disable the
3701 * reserve depending on the existence of swap or variations in the recovery
3702 * tools. So, the admin may have changed them.
3703 *
3704 * If memory is added and the reserves have been eliminated or increased above
3705 * the default max, then we'll trust the admin.
3706 *
3707 * If memory is removed and there isn't enough free memory, then we
3708 * need to reset the reserves.
3709 *
3710 * Otherwise keep the reserve set by the admin.
3711 */
3712static int reserve_mem_notifier(struct notifier_block *nb,
3713			     unsigned long action, void *data)
3714{
3715	unsigned long tmp, free_kbytes;
3716
3717	switch (action) {
3718	case MEM_ONLINE:
3719		/* Default max is 128MB. Leave alone if modified by operator. */
3720		tmp = sysctl_user_reserve_kbytes;
3721		if (0 < tmp && tmp < (1UL << 17))
3722			init_user_reserve();
3723
3724		/* Default max is 8MB.  Leave alone if modified by operator. */
3725		tmp = sysctl_admin_reserve_kbytes;
3726		if (0 < tmp && tmp < (1UL << 13))
3727			init_admin_reserve();
3728
3729		break;
3730	case MEM_OFFLINE:
3731		free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3732
3733		if (sysctl_user_reserve_kbytes > free_kbytes) {
3734			init_user_reserve();
3735			pr_info("vm.user_reserve_kbytes reset to %lu\n",
3736				sysctl_user_reserve_kbytes);
3737		}
3738
3739		if (sysctl_admin_reserve_kbytes > free_kbytes) {
3740			init_admin_reserve();
3741			pr_info("vm.admin_reserve_kbytes reset to %lu\n",
3742				sysctl_admin_reserve_kbytes);
3743		}
3744		break;
3745	default:
3746		break;
3747	}
3748	return NOTIFY_OK;
3749}
3750
3751static int __meminit init_reserve_notifier(void)
3752{
3753	if (hotplug_memory_notifier(reserve_mem_notifier, DEFAULT_CALLBACK_PRI))
3754		pr_err("Failed registering memory add/remove notifier for admin reserve\n");
3755
3756	return 0;
3757}
3758subsys_initcall(init_reserve_notifier);
v3.5.6
 
   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/export.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#include <linux/uprobes.h>
 
 
 
 
 
 
 
 
  34
  35#include <asm/uaccess.h>
  36#include <asm/cacheflush.h>
  37#include <asm/tlb.h>
  38#include <asm/mmu_context.h>
  39
 
 
 
  40#include "internal.h"
  41
  42#ifndef arch_mmap_check
  43#define arch_mmap_check(addr, len, flags)	(0)
  44#endif
  45
  46#ifndef arch_rebalance_pgtables
  47#define arch_rebalance_pgtables(addr, len)		(addr)
 
 
 
 
 
 
 
  48#endif
  49
  50static void unmap_region(struct mm_struct *mm,
 
 
 
  51		struct vm_area_struct *vma, struct vm_area_struct *prev,
  52		unsigned long start, unsigned long end);
  53
  54/*
  55 * WARNING: the debugging will use recursive algorithms so never enable this
  56 * unless you know what you are doing.
  57 */
  58#undef DEBUG_MM_RB
  59
  60/* description of effects of mapping type and prot in current implementation.
  61 * this is due to the limited x86 page protection hardware.  The expected
  62 * behavior is in parens:
  63 *
  64 * map_type	prot
  65 *		PROT_NONE	PROT_READ	PROT_WRITE	PROT_EXEC
  66 * MAP_SHARED	r: (no) no	r: (yes) yes	r: (no) yes	r: (no) yes
  67 *		w: (no) no	w: (no) no	w: (yes) yes	w: (no) no
  68 *		x: (no) no	x: (no) yes	x: (no) yes	x: (yes) yes
  69 *		
  70 * MAP_PRIVATE	r: (no) no	r: (yes) yes	r: (no) yes	r: (no) yes
  71 *		w: (no) no	w: (no) no	w: (copy) copy	w: (no) no
  72 *		x: (no) no	x: (no) yes	x: (no) yes	x: (yes) yes
  73 *
  74 */
  75pgprot_t protection_map[16] = {
  76	__P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111,
  77	__S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111
  78};
  79
  80pgprot_t vm_get_page_prot(unsigned long vm_flags)
  81{
  82	return __pgprot(pgprot_val(protection_map[vm_flags &
  83				(VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)]) |
  84			pgprot_val(arch_vm_get_page_prot(vm_flags)));
  85}
  86EXPORT_SYMBOL(vm_get_page_prot);
  87
  88int sysctl_overcommit_memory __read_mostly = OVERCOMMIT_GUESS;  /* heuristic overcommit */
  89int sysctl_overcommit_ratio __read_mostly = 50;	/* default is 50% */
  90int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
  91/*
  92 * Make sure vm_committed_as in one cacheline and not cacheline shared with
  93 * other variables. It can be updated by several CPUs frequently.
  94 */
  95struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp;
  96
  97/*
  98 * Check that a process has enough memory to allocate a new virtual
  99 * mapping. 0 means there is enough memory for the allocation to
 100 * succeed and -ENOMEM implies there is not.
 101 *
 102 * We currently support three overcommit policies, which are set via the
 103 * vm.overcommit_memory sysctl.  See Documentation/vm/overcommit-accounting
 104 *
 105 * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
 106 * Additional code 2002 Jul 20 by Robert Love.
 107 *
 108 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
 109 *
 110 * Note this is a helper function intended to be used by LSMs which
 111 * wish to use this logic.
 112 */
 113int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
 114{
 115	unsigned long free, allowed;
 116
 117	vm_acct_memory(pages);
 118
 119	/*
 120	 * Sometimes we want to use more memory than we have
 121	 */
 122	if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
 123		return 0;
 124
 125	if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
 126		free = global_page_state(NR_FREE_PAGES);
 127		free += global_page_state(NR_FILE_PAGES);
 128
 129		/*
 130		 * shmem pages shouldn't be counted as free in this
 131		 * case, they can't be purged, only swapped out, and
 132		 * that won't affect the overall amount of available
 133		 * memory in the system.
 134		 */
 135		free -= global_page_state(NR_SHMEM);
 136
 137		free += nr_swap_pages;
 138
 139		/*
 140		 * Any slabs which are created with the
 141		 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
 142		 * which are reclaimable, under pressure.  The dentry
 143		 * cache and most inode caches should fall into this
 144		 */
 145		free += global_page_state(NR_SLAB_RECLAIMABLE);
 146
 147		/*
 148		 * Leave reserved pages. The pages are not for anonymous pages.
 149		 */
 150		if (free <= totalreserve_pages)
 151			goto error;
 152		else
 153			free -= totalreserve_pages;
 154
 155		/*
 156		 * Leave the last 3% for root
 157		 */
 158		if (!cap_sys_admin)
 159			free -= free / 32;
 160
 161		if (free > pages)
 162			return 0;
 163
 164		goto error;
 165	}
 166
 167	allowed = (totalram_pages - hugetlb_total_pages())
 168	       	* sysctl_overcommit_ratio / 100;
 169	/*
 170	 * Leave the last 3% for root
 171	 */
 172	if (!cap_sys_admin)
 173		allowed -= allowed / 32;
 174	allowed += total_swap_pages;
 175
 176	/* Don't let a single process grow too big:
 177	   leave 3% of the size of this process for other processes */
 178	if (mm)
 179		allowed -= mm->total_vm / 32;
 180
 181	if (percpu_counter_read_positive(&vm_committed_as) < allowed)
 182		return 0;
 183error:
 184	vm_unacct_memory(pages);
 185
 186	return -ENOMEM;
 187}
 188
 189/*
 190 * Requires inode->i_mapping->i_mmap_mutex
 191 */
 192static void __remove_shared_vm_struct(struct vm_area_struct *vma,
 193		struct file *file, struct address_space *mapping)
 194{
 195	if (vma->vm_flags & VM_DENYWRITE)
 196		atomic_inc(&file->f_path.dentry->d_inode->i_writecount);
 197	if (vma->vm_flags & VM_SHARED)
 198		mapping->i_mmap_writable--;
 199
 200	flush_dcache_mmap_lock(mapping);
 201	if (unlikely(vma->vm_flags & VM_NONLINEAR))
 202		list_del_init(&vma->shared.vm_set.list);
 203	else
 204		vma_prio_tree_remove(vma, &mapping->i_mmap);
 205	flush_dcache_mmap_unlock(mapping);
 206}
 207
 208/*
 209 * Unlink a file-based vm structure from its prio_tree, to hide
 210 * vma from rmap and vmtruncate before freeing its page tables.
 211 */
 212void unlink_file_vma(struct vm_area_struct *vma)
 213{
 214	struct file *file = vma->vm_file;
 215
 216	if (file) {
 217		struct address_space *mapping = file->f_mapping;
 218		mutex_lock(&mapping->i_mmap_mutex);
 219		__remove_shared_vm_struct(vma, file, mapping);
 220		mutex_unlock(&mapping->i_mmap_mutex);
 221	}
 222}
 223
 224/*
 225 * Close a vm structure and free it, returning the next.
 226 */
 227static struct vm_area_struct *remove_vma(struct vm_area_struct *vma)
 228{
 229	struct vm_area_struct *next = vma->vm_next;
 230
 231	might_sleep();
 232	if (vma->vm_ops && vma->vm_ops->close)
 233		vma->vm_ops->close(vma);
 234	if (vma->vm_file) {
 235		fput(vma->vm_file);
 236		if (vma->vm_flags & VM_EXECUTABLE)
 237			removed_exe_file_vma(vma->vm_mm);
 238	}
 239	mpol_put(vma_policy(vma));
 240	kmem_cache_free(vm_area_cachep, vma);
 241	return next;
 242}
 243
 244static unsigned long do_brk(unsigned long addr, unsigned long len);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 245
 
 
 
 
 
 
 
 246SYSCALL_DEFINE1(brk, unsigned long, brk)
 247{
 248	unsigned long rlim, retval;
 249	unsigned long newbrk, oldbrk;
 250	struct mm_struct *mm = current->mm;
 
 251	unsigned long min_brk;
 
 
 
 
 
 
 
 252
 253	down_write(&mm->mmap_sem);
 254
 255#ifdef CONFIG_COMPAT_BRK
 256	/*
 257	 * CONFIG_COMPAT_BRK can still be overridden by setting
 258	 * randomize_va_space to 2, which will still cause mm->start_brk
 259	 * to be arbitrarily shifted
 260	 */
 261	if (current->brk_randomized)
 262		min_brk = mm->start_brk;
 263	else
 264		min_brk = mm->end_data;
 265#else
 266	min_brk = mm->start_brk;
 267#endif
 268	if (brk < min_brk)
 269		goto out;
 270
 271	/*
 272	 * Check against rlimit here. If this check is done later after the test
 273	 * of oldbrk with newbrk then it can escape the test and let the data
 274	 * segment grow beyond its set limit the in case where the limit is
 275	 * not page aligned -Ram Gupta
 276	 */
 277	rlim = rlimit(RLIMIT_DATA);
 278	if (rlim < RLIM_INFINITY && (brk - mm->start_brk) +
 279			(mm->end_data - mm->start_data) > rlim)
 280		goto out;
 281
 282	newbrk = PAGE_ALIGN(brk);
 283	oldbrk = PAGE_ALIGN(mm->brk);
 284	if (oldbrk == newbrk)
 285		goto set_brk;
 
 
 286
 287	/* Always allow shrinking brk. */
 
 
 
 288	if (brk <= mm->brk) {
 289		if (!do_munmap(mm, newbrk, oldbrk-newbrk))
 290			goto set_brk;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 291		goto out;
 292	}
 293
 294	/* Check against existing mmap mappings. */
 295	if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE))
 
 
 
 
 
 
 
 
 296		goto out;
 297
 
 298	/* Ok, looks good - let it rip. */
 299	if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk)
 300		goto out;
 301set_brk:
 302	mm->brk = brk;
 
 
 
 
 
 
 
 
 
 
 
 
 303out:
 304	retval = mm->brk;
 305	up_write(&mm->mmap_sem);
 306	return retval;
 307}
 308
 309#ifdef DEBUG_MM_RB
 310static int browse_rb(struct rb_root *root)
 311{
 312	int i = 0, j;
 313	struct rb_node *nd, *pn = NULL;
 314	unsigned long prev = 0, pend = 0;
 315
 316	for (nd = rb_first(root); nd; nd = rb_next(nd)) {
 317		struct vm_area_struct *vma;
 318		vma = rb_entry(nd, struct vm_area_struct, vm_rb);
 319		if (vma->vm_start < prev)
 320			printk("vm_start %lx prev %lx\n", vma->vm_start, prev), i = -1;
 321		if (vma->vm_start < pend)
 322			printk("vm_start %lx pend %lx\n", vma->vm_start, pend);
 323		if (vma->vm_start > vma->vm_end)
 324			printk("vm_end %lx < vm_start %lx\n", vma->vm_end, vma->vm_start);
 325		i++;
 326		pn = nd;
 327		prev = vma->vm_start;
 328		pend = vma->vm_end;
 329	}
 330	j = 0;
 331	for (nd = pn; nd; nd = rb_prev(nd)) {
 332		j++;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 333	}
 334	if (i != j)
 335		printk("backwards %d, forwards %d\n", j, i), i = 0;
 336	return i;
 337}
 338
 339void validate_mm(struct mm_struct *mm)
 340{
 341	int bug = 0;
 342	int i = 0;
 343	struct vm_area_struct *tmp = mm->mmap;
 344	while (tmp) {
 345		tmp = tmp->vm_next;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 346		i++;
 347	}
 348	if (i != mm->map_count)
 349		printk("map_count %d vm_next %d\n", mm->map_count, i), bug = 1;
 350	i = browse_rb(&mm->mm_rb);
 351	if (i != mm->map_count)
 352		printk("map_count %d rb %d\n", mm->map_count, i), bug = 1;
 353	BUG_ON(bug);
 354}
 355#else
 
 
 356#define validate_mm(mm) do { } while (0)
 357#endif
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 358
 359static struct vm_area_struct *
 360find_vma_prepare(struct mm_struct *mm, unsigned long addr,
 361		struct vm_area_struct **pprev, struct rb_node ***rb_link,
 362		struct rb_node ** rb_parent)
 363{
 364	struct vm_area_struct * vma;
 365	struct rb_node ** __rb_link, * __rb_parent, * rb_prev;
 366
 367	__rb_link = &mm->mm_rb.rb_node;
 368	rb_prev = __rb_parent = NULL;
 369	vma = NULL;
 370
 371	while (*__rb_link) {
 372		struct vm_area_struct *vma_tmp;
 373
 374		__rb_parent = *__rb_link;
 375		vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb);
 376
 377		if (vma_tmp->vm_end > addr) {
 378			vma = vma_tmp;
 379			if (vma_tmp->vm_start <= addr)
 380				break;
 381			__rb_link = &__rb_parent->rb_left;
 382		} else {
 383			rb_prev = __rb_parent;
 384			__rb_link = &__rb_parent->rb_right;
 385		}
 386	}
 387
 388	*pprev = NULL;
 389	if (rb_prev)
 390		*pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
 391	*rb_link = __rb_link;
 392	*rb_parent = __rb_parent;
 393	return vma;
 394}
 395
 396void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma,
 397		struct rb_node **rb_link, struct rb_node *rb_parent)
 398{
 399	rb_link_node(&vma->vm_rb, rb_parent, rb_link);
 400	rb_insert_color(&vma->vm_rb, &mm->mm_rb);
 
 
 
 
 
 
 
 
 
 
 401}
 402
 403static void __vma_link_file(struct vm_area_struct *vma)
 
 404{
 405	struct file *file;
 
 406
 407	file = vma->vm_file;
 408	if (file) {
 409		struct address_space *mapping = file->f_mapping;
 
 410
 411		if (vma->vm_flags & VM_DENYWRITE)
 412			atomic_dec(&file->f_path.dentry->d_inode->i_writecount);
 413		if (vma->vm_flags & VM_SHARED)
 414			mapping->i_mmap_writable++;
 
 
 
 
 
 
 
 
 
 
 
 
 415
 416		flush_dcache_mmap_lock(mapping);
 417		if (unlikely(vma->vm_flags & VM_NONLINEAR))
 418			vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear);
 419		else
 420			vma_prio_tree_insert(vma, &mapping->i_mmap);
 421		flush_dcache_mmap_unlock(mapping);
 422	}
 
 
 
 
 
 
 
 
 423}
 424
 425static void
 426__vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
 427	struct vm_area_struct *prev, struct rb_node **rb_link,
 428	struct rb_node *rb_parent)
 429{
 430	__vma_link_list(mm, vma, prev, rb_parent);
 431	__vma_link_rb(mm, vma, rb_link, rb_parent);
 
 
 
 
 
 
 
 432}
 433
 434static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
 435			struct vm_area_struct *prev, struct rb_node **rb_link,
 436			struct rb_node *rb_parent)
 437{
 
 438	struct address_space *mapping = NULL;
 439
 440	if (vma->vm_file)
 
 
 
 441		mapping = vma->vm_file->f_mapping;
 
 
 442
 443	if (mapping)
 444		mutex_lock(&mapping->i_mmap_mutex);
 445
 446	__vma_link(mm, vma, prev, rb_link, rb_parent);
 447	__vma_link_file(vma);
 448
 449	if (mapping)
 450		mutex_unlock(&mapping->i_mmap_mutex);
 451
 452	mm->map_count++;
 453	validate_mm(mm);
 
 454}
 455
 456/*
 457 * Helper for vma_adjust() in the split_vma insert case: insert a vma into the
 458 * mm's list and rbtree.  It has already been inserted into the prio_tree.
 459 */
 460static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 461{
 462	struct vm_area_struct *__vma, *prev;
 463	struct rb_node **rb_link, *rb_parent;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 464
 465	__vma = find_vma_prepare(mm, vma->vm_start,&prev, &rb_link, &rb_parent);
 466	BUG_ON(__vma && __vma->vm_start < vma->vm_end);
 467	__vma_link(mm, vma, prev, rb_link, rb_parent);
 468	mm->map_count++;
 469}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 470
 471static inline void
 472__vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma,
 473		struct vm_area_struct *prev)
 474{
 475	struct vm_area_struct *next = vma->vm_next;
 476
 477	prev->vm_next = next;
 478	if (next)
 479		next->vm_prev = prev;
 480	rb_erase(&vma->vm_rb, &mm->mm_rb);
 481	if (mm->mmap_cache == vma)
 482		mm->mmap_cache = prev;
 483}
 484
 485/*
 486 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
 487 * is already present in an i_mmap tree without adjusting the tree.
 488 * The following helper function should be used when such adjustments
 489 * are necessary.  The "insert" vma (if any) is to be inserted
 490 * before we drop the necessary locks.
 491 */
 492int vma_adjust(struct vm_area_struct *vma, unsigned long start,
 493	unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert)
 
 494{
 495	struct mm_struct *mm = vma->vm_mm;
 496	struct vm_area_struct *next = vma->vm_next;
 497	struct vm_area_struct *importer = NULL;
 
 498	struct address_space *mapping = NULL;
 499	struct prio_tree_root *root = NULL;
 500	struct anon_vma *anon_vma = NULL;
 501	struct file *file = vma->vm_file;
 
 502	long adjust_next = 0;
 503	int remove_next = 0;
 
 
 504
 505	if (next && !insert) {
 506		struct vm_area_struct *exporter = NULL;
 507
 508		if (end >= next->vm_end) {
 509			/*
 510			 * vma expands, overlapping all the next, and
 511			 * perhaps the one after too (mprotect case 6).
 
 
 512			 */
 513again:			remove_next = 1 + (end > next->vm_end);
 514			end = next->vm_end;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 515			exporter = next;
 516			importer = vma;
 
 
 
 
 
 
 
 
 517		} else if (end > next->vm_start) {
 518			/*
 519			 * vma expands, overlapping part of the next:
 520			 * mprotect case 5 shifting the boundary up.
 521			 */
 522			adjust_next = (end - next->vm_start) >> PAGE_SHIFT;
 523			exporter = next;
 524			importer = vma;
 
 525		} else if (end < vma->vm_end) {
 526			/*
 527			 * vma shrinks, and !insert tells it's not
 528			 * split_vma inserting another: so it must be
 529			 * mprotect case 4 shifting the boundary down.
 530			 */
 531			adjust_next = - ((vma->vm_end - end) >> PAGE_SHIFT);
 532			exporter = vma;
 533			importer = next;
 
 534		}
 535
 536		/*
 537		 * Easily overlooked: when mprotect shifts the boundary,
 538		 * make sure the expanding vma has anon_vma set if the
 539		 * shrinking vma had, to cover any anon pages imported.
 540		 */
 541		if (exporter && exporter->anon_vma && !importer->anon_vma) {
 542			if (anon_vma_clone(importer, exporter))
 543				return -ENOMEM;
 544			importer->anon_vma = exporter->anon_vma;
 
 
 
 545		}
 546	}
 547
 
 
 
 
 548	if (file) {
 549		mapping = file->f_mapping;
 550		if (!(vma->vm_flags & VM_NONLINEAR)) {
 551			root = &mapping->i_mmap;
 552			uprobe_munmap(vma, vma->vm_start, vma->vm_end);
 553
 554			if (adjust_next)
 555				uprobe_munmap(next, next->vm_start,
 556							next->vm_end);
 557		}
 558
 559		mutex_lock(&mapping->i_mmap_mutex);
 560		if (insert) {
 561			/*
 562			 * Put into prio_tree now, so instantiated pages
 563			 * are visible to arm/parisc __flush_dcache_page
 564			 * throughout; but we cannot insert into address
 565			 * space until vma start or end is updated.
 566			 */
 567			__vma_link_file(insert);
 568		}
 569	}
 570
 571	vma_adjust_trans_huge(vma, start, end, adjust_next);
 572
 573	/*
 574	 * When changing only vma->vm_end, we don't really need anon_vma
 575	 * lock. This is a fairly rare case by itself, but the anon_vma
 576	 * lock may be shared between many sibling processes.  Skipping
 577	 * the lock for brk adjustments makes a difference sometimes.
 578	 */
 579	if (vma->anon_vma && (importer || start != vma->vm_start)) {
 580		anon_vma = vma->anon_vma;
 581		anon_vma_lock(anon_vma);
 582	}
 583
 584	if (root) {
 585		flush_dcache_mmap_lock(mapping);
 586		vma_prio_tree_remove(vma, root);
 587		if (adjust_next)
 588			vma_prio_tree_remove(next, root);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 589	}
 590
 591	vma->vm_start = start;
 592	vma->vm_end = end;
 
 593	vma->vm_pgoff = pgoff;
 594	if (adjust_next) {
 595		next->vm_start += adjust_next << PAGE_SHIFT;
 596		next->vm_pgoff += adjust_next;
 
 597	}
 598
 599	if (root) {
 600		if (adjust_next)
 601			vma_prio_tree_insert(next, root);
 602		vma_prio_tree_insert(vma, root);
 603		flush_dcache_mmap_unlock(mapping);
 604	}
 605
 606	if (remove_next) {
 607		/*
 608		 * vma_merge has merged next into vma, and needs
 609		 * us to remove next before dropping the locks.
 610		 */
 611		__vma_unlink(mm, next, vma);
 612		if (file)
 613			__remove_shared_vm_struct(next, file, mapping);
 614	} else if (insert) {
 615		/*
 616		 * split_vma has split insert from vma, and needs
 617		 * us to insert it before dropping the locks
 618		 * (it may either follow vma or precede it).
 619		 */
 620		__insert_vm_struct(mm, insert);
 
 
 621	}
 622
 623	if (anon_vma)
 624		anon_vma_unlock(anon_vma);
 625	if (mapping)
 626		mutex_unlock(&mapping->i_mmap_mutex);
 
 
 627
 628	if (root) {
 
 629		uprobe_mmap(vma);
 630
 631		if (adjust_next)
 632			uprobe_mmap(next);
 633	}
 634
 635	if (remove_next) {
 
 636		if (file) {
 637			uprobe_munmap(next, next->vm_start, next->vm_end);
 638			fput(file);
 639			if (next->vm_flags & VM_EXECUTABLE)
 640				removed_exe_file_vma(mm);
 641		}
 642		if (next->anon_vma)
 643			anon_vma_merge(vma, next);
 644		mm->map_count--;
 645		mpol_put(vma_policy(next));
 646		kmem_cache_free(vm_area_cachep, next);
 
 
 
 647		/*
 648		 * In mprotect's case 6 (see comments on vma_merge),
 649		 * we must remove another next too. It would clutter
 650		 * up the code too much to do both in one go.
 651		 */
 652		if (remove_next == 2) {
 653			next = vma->vm_next;
 
 654			goto again;
 655		}
 656	}
 657	if (insert && file)
 658		uprobe_mmap(insert);
 659
 
 660	validate_mm(mm);
 661
 662	return 0;
 663}
 664
 665/*
 666 * If the vma has a ->close operation then the driver probably needs to release
 667 * per-vma resources, so we don't attempt to merge those.
 668 */
 669static inline int is_mergeable_vma(struct vm_area_struct *vma,
 670			struct file *file, unsigned long vm_flags)
 
 
 671{
 672	/* VM_CAN_NONLINEAR may get set later by f_op->mmap() */
 673	if ((vma->vm_flags ^ vm_flags) & ~VM_CAN_NONLINEAR)
 
 
 
 
 
 
 
 674		return 0;
 675	if (vma->vm_file != file)
 676		return 0;
 677	if (vma->vm_ops && vma->vm_ops->close)
 678		return 0;
 
 
 
 
 679	return 1;
 680}
 681
 682static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
 683					struct anon_vma *anon_vma2,
 684					struct vm_area_struct *vma)
 685{
 686	/*
 687	 * The list_is_singular() test is to avoid merging VMA cloned from
 688	 * parents. This can improve scalability caused by anon_vma lock.
 689	 */
 690	if ((!anon_vma1 || !anon_vma2) && (!vma ||
 691		list_is_singular(&vma->anon_vma_chain)))
 692		return 1;
 693	return anon_vma1 == anon_vma2;
 694}
 695
 696/*
 697 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
 698 * in front of (at a lower virtual address and file offset than) the vma.
 699 *
 700 * We cannot merge two vmas if they have differently assigned (non-NULL)
 701 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
 702 *
 703 * We don't check here for the merged mmap wrapping around the end of pagecache
 704 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
 705 * wrap, nor mmaps which cover the final page at index -1UL.
 706 */
 707static int
 708can_vma_merge_before(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		if (vma->vm_pgoff == vm_pgoff)
 714			return 1;
 715	}
 716	return 0;
 717}
 718
 719/*
 720 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
 721 * beyond (at a higher virtual address and file offset than) the vma.
 722 *
 723 * We cannot merge two vmas if they have differently assigned (non-NULL)
 724 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
 725 */
 726static int
 727can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
 728	struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff)
 
 
 
 729{
 730	if (is_mergeable_vma(vma, file, vm_flags) &&
 731	    is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
 732		pgoff_t vm_pglen;
 733		vm_pglen = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
 734		if (vma->vm_pgoff + vm_pglen == vm_pgoff)
 735			return 1;
 736	}
 737	return 0;
 738}
 739
 740/*
 741 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
 742 * whether that can be merged with its predecessor or its successor.
 743 * Or both (it neatly fills a hole).
 744 *
 745 * In most cases - when called for mmap, brk or mremap - [addr,end) is
 746 * certain not to be mapped by the time vma_merge is called; but when
 747 * called for mprotect, it is certain to be already mapped (either at
 748 * an offset within prev, or at the start of next), and the flags of
 749 * this area are about to be changed to vm_flags - and the no-change
 750 * case has already been eliminated.
 751 *
 752 * The following mprotect cases have to be considered, where AAAA is
 753 * the area passed down from mprotect_fixup, never extending beyond one
 754 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
 755 *
 756 *     AAAA             AAAA                AAAA          AAAA
 757 *    PPPPPPNNNNNN    PPPPPPNNNNNN    PPPPPPNNNNNN    PPPPNNNNXXXX
 758 *    cannot merge    might become    might become    might become
 759 *                    PPNNNNNNNNNN    PPPPPPPPPPNN    PPPPPPPPPPPP 6 or
 760 *    mmap, brk or    case 4 below    case 5 below    PPPPPPPPXXXX 7 or
 761 *    mremap move:                                    PPPPNNNNNNNN 8
 762 *        AAAA
 763 *    PPPP    NNNN    PPPPPPPPPPPP    PPPPPPPPNNNN    PPPPNNNNNNNN
 764 *    might become    case 1 below    case 2 below    case 3 below
 765 *
 766 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
 767 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 768 */
 769struct vm_area_struct *vma_merge(struct mm_struct *mm,
 770			struct vm_area_struct *prev, unsigned long addr,
 771			unsigned long end, unsigned long vm_flags,
 772		     	struct anon_vma *anon_vma, struct file *file,
 773			pgoff_t pgoff, struct mempolicy *policy)
 
 
 774{
 775	pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
 776	struct vm_area_struct *area, *next;
 777	int err;
 
 
 778
 779	/*
 780	 * We later require that vma->vm_flags == vm_flags,
 781	 * so this tests vma->vm_flags & VM_SPECIAL, too.
 782	 */
 783	if (vm_flags & VM_SPECIAL)
 784		return NULL;
 785
 786	if (prev)
 787		next = prev->vm_next;
 788	else
 789		next = mm->mmap;
 790	area = next;
 791	if (next && next->vm_end == end)		/* cases 6, 7, 8 */
 792		next = next->vm_next;
 793
 794	/*
 795	 * Can it merge with the predecessor?
 796	 */
 
 
 
 797	if (prev && prev->vm_end == addr &&
 798  			mpol_equal(vma_policy(prev), policy) &&
 799			can_vma_merge_after(prev, vm_flags,
 800						anon_vma, file, pgoff)) {
 801		/*
 802		 * OK, it can.  Can we now merge in the successor as well?
 803		 */
 804		if (next && end == next->vm_start &&
 805				mpol_equal(policy, vma_policy(next)) &&
 806				can_vma_merge_before(next, vm_flags,
 807					anon_vma, file, pgoff+pglen) &&
 808				is_mergeable_anon_vma(prev->anon_vma,
 809						      next->anon_vma, NULL)) {
 810							/* cases 1, 6 */
 811			err = vma_adjust(prev, prev->vm_start,
 812				next->vm_end, prev->vm_pgoff, NULL);
 813		} else					/* cases 2, 5, 7 */
 814			err = vma_adjust(prev, prev->vm_start,
 815				end, prev->vm_pgoff, NULL);
 816		if (err)
 817			return NULL;
 818		khugepaged_enter_vma_merge(prev);
 819		return prev;
 820	}
 821
 822	/*
 823	 * Can this new request be merged in front of next?
 824	 */
 825	if (next && end == next->vm_start &&
 826 			mpol_equal(policy, vma_policy(next)) &&
 827			can_vma_merge_before(next, vm_flags,
 828					anon_vma, file, pgoff+pglen)) {
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 829		if (prev && addr < prev->vm_end)	/* case 4 */
 830			err = vma_adjust(prev, prev->vm_start,
 831				addr, prev->vm_pgoff, NULL);
 832		else					/* cases 3, 8 */
 833			err = vma_adjust(area, addr, next->vm_end,
 834				next->vm_pgoff - pglen, NULL);
 835		if (err)
 836			return NULL;
 837		khugepaged_enter_vma_merge(area);
 838		return area;
 839	}
 840
 841	return NULL;
 
 
 
 
 
 
 842}
 843
 844/*
 845 * Rough compatbility check to quickly see if it's even worth looking
 846 * at sharing an anon_vma.
 847 *
 848 * They need to have the same vm_file, and the flags can only differ
 849 * in things that mprotect may change.
 850 *
 851 * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that
 852 * we can merge the two vma's. For example, we refuse to merge a vma if
 853 * there is a vm_ops->close() function, because that indicates that the
 854 * driver is doing some kind of reference counting. But that doesn't
 855 * really matter for the anon_vma sharing case.
 856 */
 857static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b)
 858{
 859	return a->vm_end == b->vm_start &&
 860		mpol_equal(vma_policy(a), vma_policy(b)) &&
 861		a->vm_file == b->vm_file &&
 862		!((a->vm_flags ^ b->vm_flags) & ~(VM_READ|VM_WRITE|VM_EXEC)) &&
 863		b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT);
 864}
 865
 866/*
 867 * Do some basic sanity checking to see if we can re-use the anon_vma
 868 * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be
 869 * the same as 'old', the other will be the new one that is trying
 870 * to share the anon_vma.
 871 *
 872 * NOTE! This runs with mm_sem held for reading, so it is possible that
 873 * the anon_vma of 'old' is concurrently in the process of being set up
 874 * by another page fault trying to merge _that_. But that's ok: if it
 875 * is being set up, that automatically means that it will be a singleton
 876 * acceptable for merging, so we can do all of this optimistically. But
 877 * we do that ACCESS_ONCE() to make sure that we never re-load the pointer.
 878 *
 879 * IOW: that the "list_is_singular()" test on the anon_vma_chain only
 880 * matters for the 'stable anon_vma' case (ie the thing we want to avoid
 881 * is to return an anon_vma that is "complex" due to having gone through
 882 * a fork).
 883 *
 884 * We also make sure that the two vma's are compatible (adjacent,
 885 * and with the same memory policies). That's all stable, even with just
 886 * a read lock on the mm_sem.
 887 */
 888static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b)
 889{
 890	if (anon_vma_compatible(a, b)) {
 891		struct anon_vma *anon_vma = ACCESS_ONCE(old->anon_vma);
 892
 893		if (anon_vma && list_is_singular(&old->anon_vma_chain))
 894			return anon_vma;
 895	}
 896	return NULL;
 897}
 898
 899/*
 900 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
 901 * neighbouring vmas for a suitable anon_vma, before it goes off
 902 * to allocate a new anon_vma.  It checks because a repetitive
 903 * sequence of mprotects and faults may otherwise lead to distinct
 904 * anon_vmas being allocated, preventing vma merge in subsequent
 905 * mprotect.
 906 */
 907struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
 908{
 909	struct anon_vma *anon_vma;
 910	struct vm_area_struct *near;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 911
 912	near = vma->vm_next;
 913	if (!near)
 914		goto try_prev;
 915
 916	anon_vma = reusable_anon_vma(near, vma, near);
 917	if (anon_vma)
 918		return anon_vma;
 919try_prev:
 920	near = vma->vm_prev;
 921	if (!near)
 922		goto none;
 923
 924	anon_vma = reusable_anon_vma(near, near, vma);
 925	if (anon_vma)
 926		return anon_vma;
 927none:
 928	/*
 
 
 929	 * There's no absolute need to look only at touching neighbours:
 930	 * we could search further afield for "compatible" anon_vmas.
 931	 * But it would probably just be a waste of time searching,
 932	 * or lead to too many vmas hanging off the same anon_vma.
 933	 * We're trying to allow mprotect remerging later on,
 934	 * not trying to minimize memory used for anon_vmas.
 935	 */
 936	return NULL;
 937}
 938
 939#ifdef CONFIG_PROC_FS
 940void vm_stat_account(struct mm_struct *mm, unsigned long flags,
 941						struct file *file, long pages)
 942{
 943	const unsigned long stack_flags
 944		= VM_STACK_FLAGS & (VM_GROWSUP|VM_GROWSDOWN);
 945
 946	if (file) {
 947		mm->shared_vm += pages;
 948		if ((flags & (VM_EXEC|VM_WRITE)) == VM_EXEC)
 949			mm->exec_vm += pages;
 950	} else if (flags & stack_flags)
 951		mm->stack_vm += pages;
 952	if (flags & (VM_RESERVED|VM_IO))
 953		mm->reserved_vm += pages;
 954}
 955#endif /* CONFIG_PROC_FS */
 956
 957/*
 958 * If a hint addr is less than mmap_min_addr change hint to be as
 959 * low as possible but still greater than mmap_min_addr
 960 */
 961static inline unsigned long round_hint_to_min(unsigned long hint)
 962{
 963	hint &= PAGE_MASK;
 964	if (((void *)hint != NULL) &&
 965	    (hint < mmap_min_addr))
 966		return PAGE_ALIGN(mmap_min_addr);
 967	return hint;
 968}
 969
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 970/*
 971 * The caller must hold down_write(&current->mm->mmap_sem).
 972 */
 973
 974unsigned long do_mmap_pgoff(struct file *file, unsigned long addr,
 975			unsigned long len, unsigned long prot,
 976			unsigned long flags, unsigned long pgoff)
 
 977{
 978	struct mm_struct * mm = current->mm;
 979	struct inode *inode;
 980	vm_flags_t vm_flags;
 
 
 
 
 
 
 
 981
 982	/*
 983	 * Does the application expect PROT_READ to imply PROT_EXEC?
 984	 *
 985	 * (the exception is when the underlying filesystem is noexec
 986	 *  mounted, in which case we dont add PROT_EXEC.)
 987	 */
 988	if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
 989		if (!(file && (file->f_path.mnt->mnt_flags & MNT_NOEXEC)))
 990			prot |= PROT_EXEC;
 991
 992	if (!len)
 993		return -EINVAL;
 
 994
 995	if (!(flags & MAP_FIXED))
 996		addr = round_hint_to_min(addr);
 997
 998	/* Careful about overflows.. */
 999	len = PAGE_ALIGN(len);
1000	if (!len)
1001		return -ENOMEM;
1002
1003	/* offset overflow? */
1004	if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
1005               return -EOVERFLOW;
1006
1007	/* Too many mappings? */
1008	if (mm->map_count > sysctl_max_map_count)
1009		return -ENOMEM;
1010
1011	/* Obtain the address to map to. we verify (or select) it and ensure
1012	 * that it represents a valid section of the address space.
1013	 */
1014	addr = get_unmapped_area(file, addr, len, pgoff, flags);
1015	if (addr & ~PAGE_MASK)
1016		return addr;
1017
 
 
 
 
 
 
 
 
 
 
 
1018	/* Do simple checking here so the lower-level routines won't have
1019	 * to. we assume access permissions have been handled by the open
1020	 * of the memory object, so we don't do any here.
1021	 */
1022	vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) |
1023			mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1024
1025	if (flags & MAP_LOCKED)
1026		if (!can_do_mlock())
1027			return -EPERM;
1028
1029	/* mlock MCL_FUTURE? */
1030	if (vm_flags & VM_LOCKED) {
1031		unsigned long locked, lock_limit;
1032		locked = len >> PAGE_SHIFT;
1033		locked += mm->locked_vm;
1034		lock_limit = rlimit(RLIMIT_MEMLOCK);
1035		lock_limit >>= PAGE_SHIFT;
1036		if (locked > lock_limit && !capable(CAP_IPC_LOCK))
1037			return -EAGAIN;
1038	}
1039
1040	inode = file ? file->f_path.dentry->d_inode : NULL;
1041
1042	if (file) {
1043		switch (flags & MAP_TYPE) {
1044		case MAP_SHARED:
1045			if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
1046				return -EACCES;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1047
1048			/*
1049			 * Make sure we don't allow writing to an append-only
1050			 * file..
1051			 */
1052			if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
1053				return -EACCES;
1054
1055			/*
1056			 * Make sure there are no mandatory locks on the file.
1057			 */
1058			if (locks_verify_locked(inode))
1059				return -EAGAIN;
1060
1061			vm_flags |= VM_SHARED | VM_MAYSHARE;
1062			if (!(file->f_mode & FMODE_WRITE))
1063				vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
1064
1065			/* fall through */
1066		case MAP_PRIVATE:
1067			if (!(file->f_mode & FMODE_READ))
1068				return -EACCES;
1069			if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) {
1070				if (vm_flags & VM_EXEC)
1071					return -EPERM;
1072				vm_flags &= ~VM_MAYEXEC;
1073			}
1074
1075			if (!file->f_op || !file->f_op->mmap)
1076				return -ENODEV;
 
 
1077			break;
1078
1079		default:
1080			return -EINVAL;
1081		}
1082	} else {
1083		switch (flags & MAP_TYPE) {
1084		case MAP_SHARED:
 
 
1085			/*
1086			 * Ignore pgoff.
1087			 */
1088			pgoff = 0;
1089			vm_flags |= VM_SHARED | VM_MAYSHARE;
1090			break;
1091		case MAP_PRIVATE:
1092			/*
1093			 * Set pgoff according to addr for anon_vma.
1094			 */
1095			pgoff = addr >> PAGE_SHIFT;
1096			break;
1097		default:
1098			return -EINVAL;
1099		}
1100	}
1101
1102	return mmap_region(file, addr, len, flags, vm_flags, pgoff);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1103}
1104
1105SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1106		unsigned long, prot, unsigned long, flags,
1107		unsigned long, fd, unsigned long, pgoff)
1108{
1109	struct file *file = NULL;
1110	unsigned long retval = -EBADF;
1111
1112	if (!(flags & MAP_ANONYMOUS)) {
1113		audit_mmap_fd(fd, flags);
1114		if (unlikely(flags & MAP_HUGETLB))
1115			return -EINVAL;
1116		file = fget(fd);
1117		if (!file)
1118			goto out;
 
 
 
 
 
 
1119	} else if (flags & MAP_HUGETLB) {
1120		struct user_struct *user = NULL;
 
 
 
 
 
 
1121		/*
1122		 * VM_NORESERVE is used because the reservations will be
1123		 * taken when vm_ops->mmap() is called
1124		 * A dummy user value is used because we are not locking
1125		 * memory so no accounting is necessary
1126		 */
1127		file = hugetlb_file_setup(HUGETLB_ANON_FILE, addr, len,
1128						VM_NORESERVE, &user,
1129						HUGETLB_ANONHUGE_INODE);
 
1130		if (IS_ERR(file))
1131			return PTR_ERR(file);
1132	}
1133
1134	flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1135
1136	retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
 
1137	if (file)
1138		fput(file);
1139out:
1140	return retval;
1141}
1142
 
 
 
 
 
 
 
1143#ifdef __ARCH_WANT_SYS_OLD_MMAP
1144struct mmap_arg_struct {
1145	unsigned long addr;
1146	unsigned long len;
1147	unsigned long prot;
1148	unsigned long flags;
1149	unsigned long fd;
1150	unsigned long offset;
1151};
1152
1153SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1154{
1155	struct mmap_arg_struct a;
1156
1157	if (copy_from_user(&a, arg, sizeof(a)))
1158		return -EFAULT;
1159	if (a.offset & ~PAGE_MASK)
1160		return -EINVAL;
1161
1162	return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1163			      a.offset >> PAGE_SHIFT);
1164}
1165#endif /* __ARCH_WANT_SYS_OLD_MMAP */
1166
1167/*
1168 * Some shared mappigns will want the pages marked read-only
1169 * to track write events. If so, we'll downgrade vm_page_prot
1170 * to the private version (using protection_map[] without the
1171 * VM_SHARED bit).
1172 */
1173int vma_wants_writenotify(struct vm_area_struct *vma)
1174{
1175	vm_flags_t vm_flags = vma->vm_flags;
 
1176
1177	/* If it was private or non-writable, the write bit is already clear */
1178	if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED)))
1179		return 0;
1180
1181	/* The backer wishes to know when pages are first written to? */
1182	if (vma->vm_ops && vma->vm_ops->page_mkwrite)
1183		return 1;
1184
1185	/* The open routine did something to the protections already? */
1186	if (pgprot_val(vma->vm_page_prot) !=
1187	    pgprot_val(vm_get_page_prot(vm_flags)))
 
1188		return 0;
1189
 
 
 
 
 
 
 
 
 
 
 
1190	/* Specialty mapping? */
1191	if (vm_flags & (VM_PFNMAP|VM_INSERTPAGE))
1192		return 0;
1193
1194	/* Can the mapping track the dirty pages? */
1195	return vma->vm_file && vma->vm_file->f_mapping &&
1196		mapping_cap_account_dirty(vma->vm_file->f_mapping);
1197}
1198
1199/*
1200 * We account for memory if it's a private writeable mapping,
1201 * not hugepages and VM_NORESERVE wasn't set.
1202 */
1203static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags)
1204{
1205	/*
1206	 * hugetlb has its own accounting separate from the core VM
1207	 * VM_HUGETLB may not be set yet so we cannot check for that flag.
1208	 */
1209	if (file && is_file_hugepages(file))
1210		return 0;
1211
1212	return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE;
1213}
1214
1215unsigned long mmap_region(struct file *file, unsigned long addr,
1216			  unsigned long len, unsigned long flags,
1217			  vm_flags_t vm_flags, unsigned long pgoff)
 
 
 
 
 
 
 
 
1218{
1219	struct mm_struct *mm = current->mm;
1220	struct vm_area_struct *vma, *prev;
1221	int correct_wcount = 0;
1222	int error;
1223	struct rb_node **rb_link, *rb_parent;
1224	unsigned long charged = 0;
1225	struct inode *inode =  file ? file->f_path.dentry->d_inode : NULL;
1226
1227	/* Clear old maps */
1228	error = -ENOMEM;
1229munmap_back:
1230	vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
1231	if (vma && vma->vm_start < addr + len) {
1232		if (do_munmap(mm, addr, len))
1233			return -ENOMEM;
1234		goto munmap_back;
1235	}
1236
1237	/* Check against address space limit. */
1238	if (!may_expand_vm(mm, len >> PAGE_SHIFT))
 
1239		return -ENOMEM;
1240
1241	/*
1242	 * Set 'VM_NORESERVE' if we should not account for the
1243	 * memory use of this mapping.
1244	 */
1245	if ((flags & MAP_NORESERVE)) {
1246		/* We honor MAP_NORESERVE if allowed to overcommit */
1247		if (sysctl_overcommit_memory != OVERCOMMIT_NEVER)
1248			vm_flags |= VM_NORESERVE;
1249
1250		/* hugetlb applies strict overcommit unless MAP_NORESERVE */
1251		if (file && is_file_hugepages(file))
1252			vm_flags |= VM_NORESERVE;
1253	}
1254
1255	/*
1256	 * Private writable mapping: check memory availability
1257	 */
1258	if (accountable_mapping(file, vm_flags)) {
1259		charged = len >> PAGE_SHIFT;
1260		if (security_vm_enough_memory_mm(mm, charged))
1261			return -ENOMEM;
1262		vm_flags |= VM_ACCOUNT;
1263	}
 
 
 
 
1264
1265	/*
1266	 * Can we just expand an old mapping?
1267	 */
1268	vma = vma_merge(mm, prev, addr, addr + len, vm_flags, NULL, file, pgoff, NULL);
1269	if (vma)
1270		goto out;
1271
1272	/*
1273	 * Determine the object being mapped and call the appropriate
1274	 * specific mapper. the address has already been validated, but
1275	 * not unmapped, but the maps are removed from the list.
1276	 */
1277	vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1278	if (!vma) {
1279		error = -ENOMEM;
1280		goto unacct_error;
1281	}
1282
1283	vma->vm_mm = mm;
1284	vma->vm_start = addr;
1285	vma->vm_end = addr + len;
1286	vma->vm_flags = vm_flags;
1287	vma->vm_page_prot = vm_get_page_prot(vm_flags);
1288	vma->vm_pgoff = pgoff;
1289	INIT_LIST_HEAD(&vma->anon_vma_chain);
1290
1291	error = -EINVAL;	/* when rejecting VM_GROWSDOWN|VM_GROWSUP */
 
 
 
 
 
 
 
 
 
 
 
1292
1293	if (file) {
1294		if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1295			goto free_vma;
1296		if (vm_flags & VM_DENYWRITE) {
1297			error = deny_write_access(file);
1298			if (error)
1299				goto free_vma;
1300			correct_wcount = 1;
1301		}
1302		vma->vm_file = file;
1303		get_file(file);
1304		error = file->f_op->mmap(file, vma);
1305		if (error)
1306			goto unmap_and_free_vma;
1307		if (vm_flags & VM_EXECUTABLE)
1308			added_exe_file_vma(mm);
1309
1310		/* Can addr have changed??
1311		 *
1312		 * Answer: Yes, several device drivers can do it in their
1313		 *         f_op->mmap method. -DaveM
1314		 */
1315		addr = vma->vm_start;
1316		pgoff = vma->vm_pgoff;
1317		vm_flags = vma->vm_flags;
1318	} else if (vm_flags & VM_SHARED) {
1319		if (unlikely(vm_flags & (VM_GROWSDOWN|VM_GROWSUP)))
1320			goto free_vma;
1321		error = shmem_zero_setup(vma);
1322		if (error)
1323			goto free_vma;
1324	}
1325
1326	if (vma_wants_writenotify(vma)) {
1327		pgprot_t pprot = vma->vm_page_prot;
1328
1329		/* Can vma->vm_page_prot have changed??
1330		 *
1331		 * Answer: Yes, drivers may have changed it in their
1332		 *         f_op->mmap method.
1333		 *
1334		 * Ensures that vmas marked as uncached stay that way.
1335		 */
1336		vma->vm_page_prot = vm_get_page_prot(vm_flags & ~VM_SHARED);
1337		if (pgprot_val(pprot) == pgprot_val(pgprot_noncached(pprot)))
1338			vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1339	}
1340
1341	vma_link(mm, vma, prev, rb_link, rb_parent);
1342	file = vma->vm_file;
1343
1344	/* Once vma denies write, undo our temporary denial count */
1345	if (correct_wcount)
1346		atomic_inc(&inode->i_writecount);
1347out:
1348	perf_event_mmap(vma);
1349
1350	mm->total_vm += len >> PAGE_SHIFT;
1351	vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT);
1352	if (vm_flags & VM_LOCKED) {
1353		if (!mlock_vma_pages_range(vma, addr, addr + len))
1354			mm->locked_vm += (len >> PAGE_SHIFT);
1355	} else if ((flags & MAP_POPULATE) && !(flags & MAP_NONBLOCK))
1356		make_pages_present(addr, addr + len);
1357
1358	if (file)
1359		uprobe_mmap(vma);
1360
 
1361	return addr;
1362
1363unmap_and_free_vma:
1364	if (correct_wcount)
1365		atomic_inc(&inode->i_writecount);
1366	vma->vm_file = NULL;
1367	fput(file);
1368
1369	/* Undo any partial mapping done by a device driver. */
1370	unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
1371	charged = 0;
1372free_vma:
1373	kmem_cache_free(vm_area_cachep, vma);
1374unacct_error:
1375	if (charged)
1376		vm_unacct_memory(charged);
1377	return error;
1378}
1379
1380/* Get an address range which is currently unmapped.
1381 * For shmat() with addr=0.
1382 *
1383 * Ugly calling convention alert:
1384 * Return value with the low bits set means error value,
1385 * ie
1386 *	if (ret & ~PAGE_MASK)
1387 *		error = ret;
1388 *
1389 * This function "knows" that -ENOMEM has the bits set.
1390 */
1391#ifndef HAVE_ARCH_UNMAPPED_AREA
1392unsigned long
1393arch_get_unmapped_area(struct file *filp, unsigned long addr,
1394		unsigned long len, unsigned long pgoff, unsigned long flags)
 
1395{
1396	struct mm_struct *mm = current->mm;
1397	struct vm_area_struct *vma;
1398	unsigned long start_addr;
 
1399
1400	if (len > TASK_SIZE)
1401		return -ENOMEM;
1402
1403	if (flags & MAP_FIXED)
1404		return addr;
1405
1406	if (addr) {
1407		addr = PAGE_ALIGN(addr);
1408		vma = find_vma(mm, addr);
1409		if (TASK_SIZE - len >= addr &&
1410		    (!vma || addr + len <= vma->vm_start))
 
1411			return addr;
1412	}
1413	if (len > mm->cached_hole_size) {
1414	        start_addr = addr = mm->free_area_cache;
1415	} else {
1416	        start_addr = addr = TASK_UNMAPPED_BASE;
1417	        mm->cached_hole_size = 0;
1418	}
1419
1420full_search:
1421	for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
1422		/* At this point:  (!vma || addr < vma->vm_end). */
1423		if (TASK_SIZE - len < addr) {
1424			/*
1425			 * Start a new search - just in case we missed
1426			 * some holes.
1427			 */
1428			if (start_addr != TASK_UNMAPPED_BASE) {
1429				addr = TASK_UNMAPPED_BASE;
1430			        start_addr = addr;
1431				mm->cached_hole_size = 0;
1432				goto full_search;
1433			}
1434			return -ENOMEM;
1435		}
1436		if (!vma || addr + len <= vma->vm_start) {
1437			/*
1438			 * Remember the place where we stopped the search:
1439			 */
1440			mm->free_area_cache = addr + len;
1441			return addr;
1442		}
1443		if (addr + mm->cached_hole_size < vma->vm_start)
1444		        mm->cached_hole_size = vma->vm_start - addr;
1445		addr = vma->vm_end;
1446	}
1447}
1448#endif	
1449
1450void arch_unmap_area(struct mm_struct *mm, unsigned long addr)
 
 
 
 
1451{
1452	/*
1453	 * Is this a new hole at the lowest possible address?
1454	 */
1455	if (addr >= TASK_UNMAPPED_BASE && addr < mm->free_area_cache)
1456		mm->free_area_cache = addr;
1457}
 
1458
1459/*
1460 * This mmap-allocator allocates new areas top-down from below the
1461 * stack's low limit (the base):
1462 */
1463#ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1464unsigned long
1465arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
1466			  const unsigned long len, const unsigned long pgoff,
1467			  const unsigned long flags)
1468{
1469	struct vm_area_struct *vma;
1470	struct mm_struct *mm = current->mm;
1471	unsigned long addr = addr0, start_addr;
 
1472
1473	/* requested length too big for entire address space */
1474	if (len > TASK_SIZE)
1475		return -ENOMEM;
1476
1477	if (flags & MAP_FIXED)
1478		return addr;
1479
1480	/* requesting a specific address */
1481	if (addr) {
1482		addr = PAGE_ALIGN(addr);
1483		vma = find_vma(mm, addr);
1484		if (TASK_SIZE - len >= addr &&
1485				(!vma || addr + len <= vma->vm_start))
 
1486			return addr;
1487	}
1488
1489	/* check if free_area_cache is useful for us */
1490	if (len <= mm->cached_hole_size) {
1491 	        mm->cached_hole_size = 0;
1492 		mm->free_area_cache = mm->mmap_base;
1493 	}
1494
1495try_again:
1496	/* either no address requested or can't fit in requested address hole */
1497	start_addr = addr = mm->free_area_cache;
1498
1499	if (addr < len)
1500		goto fail;
1501
1502	addr -= len;
1503	do {
1504		/*
1505		 * Lookup failure means no vma is above this address,
1506		 * else if new region fits below vma->vm_start,
1507		 * return with success:
1508		 */
1509		vma = find_vma(mm, addr);
1510		if (!vma || addr+len <= vma->vm_start)
1511			/* remember the address as a hint for next time */
1512			return (mm->free_area_cache = addr);
1513
1514 		/* remember the largest hole we saw so far */
1515 		if (addr + mm->cached_hole_size < vma->vm_start)
1516 		        mm->cached_hole_size = vma->vm_start - addr;
1517
1518		/* try just below the current vma->vm_start */
1519		addr = vma->vm_start-len;
1520	} while (len < vma->vm_start);
1521
1522fail:
1523	/*
1524	 * if hint left us with no space for the requested
1525	 * mapping then try again:
1526	 *
1527	 * Note: this is different with the case of bottomup
1528	 * which does the fully line-search, but we use find_vma
1529	 * here that causes some holes skipped.
1530	 */
1531	if (start_addr != mm->mmap_base) {
1532		mm->free_area_cache = mm->mmap_base;
1533		mm->cached_hole_size = 0;
1534		goto try_again;
1535	}
1536
1537	/*
1538	 * A failed mmap() very likely causes application failure,
1539	 * so fall back to the bottom-up function here. This scenario
1540	 * can happen with large stack limits and large mmap()
1541	 * allocations.
1542	 */
1543	mm->cached_hole_size = ~0UL;
1544  	mm->free_area_cache = TASK_UNMAPPED_BASE;
1545	addr = arch_get_unmapped_area(filp, addr0, len, pgoff, flags);
1546	/*
1547	 * Restore the topdown base:
1548	 */
1549	mm->free_area_cache = mm->mmap_base;
1550	mm->cached_hole_size = ~0UL;
1551
1552	return addr;
1553}
1554#endif
1555
1556void arch_unmap_area_topdown(struct mm_struct *mm, unsigned long addr)
 
 
 
 
1557{
1558	/*
1559	 * Is this a new hole at the highest possible address?
1560	 */
1561	if (addr > mm->free_area_cache)
1562		mm->free_area_cache = addr;
1563
1564	/* dont allow allocations above current base */
1565	if (mm->free_area_cache > mm->mmap_base)
1566		mm->free_area_cache = mm->mmap_base;
1567}
 
1568
1569unsigned long
1570get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
1571		unsigned long pgoff, unsigned long flags)
1572{
1573	unsigned long (*get_area)(struct file *, unsigned long,
1574				  unsigned long, unsigned long, unsigned long);
1575
1576	unsigned long error = arch_mmap_check(addr, len, flags);
1577	if (error)
1578		return error;
1579
1580	/* Careful about overflows.. */
1581	if (len > TASK_SIZE)
1582		return -ENOMEM;
1583
1584	get_area = current->mm->get_unmapped_area;
1585	if (file && file->f_op && file->f_op->get_unmapped_area)
1586		get_area = file->f_op->get_unmapped_area;
 
 
 
 
 
 
 
 
 
 
 
1587	addr = get_area(file, addr, len, pgoff, flags);
1588	if (IS_ERR_VALUE(addr))
1589		return addr;
1590
1591	if (addr > TASK_SIZE - len)
1592		return -ENOMEM;
1593	if (addr & ~PAGE_MASK)
1594		return -EINVAL;
1595
1596	addr = arch_rebalance_pgtables(addr, len);
1597	error = security_mmap_addr(addr);
1598	return error ? error : addr;
1599}
1600
1601EXPORT_SYMBOL(get_unmapped_area);
1602
1603/* Look up the first VMA which satisfies  addr < vm_end,  NULL if none. */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1604struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
1605{
1606	struct vm_area_struct *vma = NULL;
1607
1608	if (WARN_ON_ONCE(!mm))		/* Remove this in linux-3.6 */
1609		return NULL;
1610
1611	/* Check the cache first. */
1612	/* (Cache hit rate is typically around 35%.) */
1613	vma = mm->mmap_cache;
1614	if (!(vma && vma->vm_end > addr && vma->vm_start <= addr)) {
1615		struct rb_node *rb_node;
1616
1617		rb_node = mm->mm_rb.rb_node;
1618		vma = NULL;
1619
1620		while (rb_node) {
1621			struct vm_area_struct *vma_tmp;
1622
1623			vma_tmp = rb_entry(rb_node,
1624					   struct vm_area_struct, vm_rb);
1625
1626			if (vma_tmp->vm_end > addr) {
1627				vma = vma_tmp;
1628				if (vma_tmp->vm_start <= addr)
1629					break;
1630				rb_node = rb_node->rb_left;
1631			} else
1632				rb_node = rb_node->rb_right;
1633		}
1634		if (vma)
1635			mm->mmap_cache = vma;
1636	}
1637	return vma;
1638}
1639
1640EXPORT_SYMBOL(find_vma);
1641
1642/*
1643 * Same as find_vma, but also return a pointer to the previous VMA in *pprev.
 
 
 
 
 
 
 
 
 
 
1644 */
1645struct vm_area_struct *
1646find_vma_prev(struct mm_struct *mm, unsigned long addr,
1647			struct vm_area_struct **pprev)
1648{
1649	struct vm_area_struct *vma;
 
1650
1651	vma = find_vma(mm, addr);
1652	if (vma) {
1653		*pprev = vma->vm_prev;
1654	} else {
1655		struct rb_node *rb_node = mm->mm_rb.rb_node;
1656		*pprev = NULL;
1657		while (rb_node) {
1658			*pprev = rb_entry(rb_node, struct vm_area_struct, vm_rb);
1659			rb_node = rb_node->rb_right;
1660		}
1661	}
1662	return vma;
1663}
1664
1665/*
1666 * Verify that the stack growth is acceptable and
1667 * update accounting. This is shared with both the
1668 * grow-up and grow-down cases.
1669 */
1670static int acct_stack_growth(struct vm_area_struct *vma, unsigned long size, unsigned long grow)
 
1671{
1672	struct mm_struct *mm = vma->vm_mm;
1673	struct rlimit *rlim = current->signal->rlim;
1674	unsigned long new_start;
1675
1676	/* address space limit tests */
1677	if (!may_expand_vm(mm, grow))
1678		return -ENOMEM;
1679
1680	/* Stack limit test */
1681	if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur))
1682		return -ENOMEM;
1683
1684	/* mlock limit tests */
1685	if (vma->vm_flags & VM_LOCKED) {
1686		unsigned long locked;
1687		unsigned long limit;
1688		locked = mm->locked_vm + grow;
1689		limit = ACCESS_ONCE(rlim[RLIMIT_MEMLOCK].rlim_cur);
1690		limit >>= PAGE_SHIFT;
1691		if (locked > limit && !capable(CAP_IPC_LOCK))
1692			return -ENOMEM;
1693	}
1694
1695	/* Check to ensure the stack will not grow into a hugetlb-only region */
1696	new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start :
1697			vma->vm_end - size;
1698	if (is_hugepage_only_range(vma->vm_mm, new_start, size))
1699		return -EFAULT;
1700
1701	/*
1702	 * Overcommit..  This must be the final test, as it will
1703	 * update security statistics.
1704	 */
1705	if (security_vm_enough_memory_mm(mm, grow))
1706		return -ENOMEM;
1707
1708	/* Ok, everything looks good - let it rip */
1709	mm->total_vm += grow;
1710	if (vma->vm_flags & VM_LOCKED)
1711		mm->locked_vm += grow;
1712	vm_stat_account(mm, vma->vm_flags, vma->vm_file, grow);
1713	return 0;
1714}
1715
1716#if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
1717/*
1718 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
1719 * vma is the last one with address > vma->vm_end.  Have to extend vma.
1720 */
1721int expand_upwards(struct vm_area_struct *vma, unsigned long address)
1722{
1723	int error;
 
 
 
 
1724
1725	if (!(vma->vm_flags & VM_GROWSUP))
1726		return -EFAULT;
1727
1728	/*
1729	 * We must make sure the anon_vma is allocated
1730	 * so that the anon_vma locking is not a noop.
1731	 */
1732	if (unlikely(anon_vma_prepare(vma)))
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1733		return -ENOMEM;
1734	vma_lock_anon_vma(vma);
1735
1736	/*
1737	 * vma->vm_start/vm_end cannot change under us because the caller
1738	 * is required to hold the mmap_sem in read mode.  We need the
1739	 * anon_vma lock to serialize against concurrent expand_stacks.
1740	 * Also guard against wrapping around to address 0.
1741	 */
1742	if (address < PAGE_ALIGN(address+4))
1743		address = PAGE_ALIGN(address+4);
1744	else {
1745		vma_unlock_anon_vma(vma);
1746		return -ENOMEM;
1747	}
1748	error = 0;
1749
1750	/* Somebody else might have raced and expanded it already */
1751	if (address > vma->vm_end) {
1752		unsigned long size, grow;
1753
1754		size = address - vma->vm_start;
1755		grow = (address - vma->vm_end) >> PAGE_SHIFT;
1756
1757		error = -ENOMEM;
1758		if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) {
1759			error = acct_stack_growth(vma, size, grow);
1760			if (!error) {
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1761				vma->vm_end = address;
 
 
 
 
 
1762				perf_event_mmap(vma);
1763			}
1764		}
1765	}
1766	vma_unlock_anon_vma(vma);
1767	khugepaged_enter_vma_merge(vma);
 
1768	return error;
1769}
1770#endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
1771
1772/*
1773 * vma is the first one with address < vma->vm_start.  Have to extend vma.
1774 */
1775int expand_downwards(struct vm_area_struct *vma,
1776				   unsigned long address)
1777{
1778	int error;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1779
1780	/*
1781	 * We must make sure the anon_vma is allocated
1782	 * so that the anon_vma locking is not a noop.
1783	 */
1784	if (unlikely(anon_vma_prepare(vma)))
1785		return -ENOMEM;
1786
1787	address &= PAGE_MASK;
1788	error = security_mmap_addr(address);
1789	if (error)
1790		return error;
1791
1792	vma_lock_anon_vma(vma);
1793
1794	/*
1795	 * vma->vm_start/vm_end cannot change under us because the caller
1796	 * is required to hold the mmap_sem in read mode.  We need the
1797	 * anon_vma lock to serialize against concurrent expand_stacks.
1798	 */
 
1799
1800	/* Somebody else might have raced and expanded it already */
1801	if (address < vma->vm_start) {
1802		unsigned long size, grow;
1803
1804		size = vma->vm_end - address;
1805		grow = (vma->vm_start - address) >> PAGE_SHIFT;
1806
1807		error = -ENOMEM;
1808		if (grow <= vma->vm_pgoff) {
1809			error = acct_stack_growth(vma, size, grow);
1810			if (!error) {
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1811				vma->vm_start = address;
1812				vma->vm_pgoff -= grow;
 
 
 
 
 
1813				perf_event_mmap(vma);
1814			}
1815		}
1816	}
1817	vma_unlock_anon_vma(vma);
1818	khugepaged_enter_vma_merge(vma);
 
1819	return error;
1820}
1821
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1822#ifdef CONFIG_STACK_GROWSUP
1823int expand_stack(struct vm_area_struct *vma, unsigned long address)
1824{
1825	return expand_upwards(vma, address);
1826}
1827
1828struct vm_area_struct *
1829find_extend_vma(struct mm_struct *mm, unsigned long addr)
1830{
1831	struct vm_area_struct *vma, *prev;
1832
1833	addr &= PAGE_MASK;
1834	vma = find_vma_prev(mm, addr, &prev);
1835	if (vma && (vma->vm_start <= addr))
1836		return vma;
1837	if (!prev || expand_stack(prev, addr))
1838		return NULL;
1839	if (prev->vm_flags & VM_LOCKED) {
1840		mlock_vma_pages_range(prev, addr, prev->vm_end);
1841	}
1842	return prev;
1843}
1844#else
1845int expand_stack(struct vm_area_struct *vma, unsigned long address)
1846{
1847	return expand_downwards(vma, address);
1848}
1849
1850struct vm_area_struct *
1851find_extend_vma(struct mm_struct * mm, unsigned long addr)
1852{
1853	struct vm_area_struct * vma;
1854	unsigned long start;
1855
1856	addr &= PAGE_MASK;
1857	vma = find_vma(mm,addr);
1858	if (!vma)
1859		return NULL;
1860	if (vma->vm_start <= addr)
1861		return vma;
1862	if (!(vma->vm_flags & VM_GROWSDOWN))
1863		return NULL;
1864	start = vma->vm_start;
1865	if (expand_stack(vma, addr))
1866		return NULL;
1867	if (vma->vm_flags & VM_LOCKED) {
1868		mlock_vma_pages_range(vma, addr, start);
1869	}
1870	return vma;
1871}
1872#endif
1873
 
 
1874/*
1875 * Ok - we have the memory areas we should free on the vma list,
1876 * so release them, and do the vma updates.
1877 *
1878 * Called with the mm semaphore held.
1879 */
1880static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma)
1881{
1882	unsigned long nr_accounted = 0;
 
1883
1884	/* Update high watermark before we lower total_vm */
1885	update_hiwater_vm(mm);
1886	do {
1887		long nrpages = vma_pages(vma);
1888
1889		if (vma->vm_flags & VM_ACCOUNT)
1890			nr_accounted += nrpages;
1891		mm->total_vm -= nrpages;
1892		vm_stat_account(mm, vma->vm_flags, vma->vm_file, -nrpages);
1893		vma = remove_vma(vma);
1894	} while (vma);
1895	vm_unacct_memory(nr_accounted);
1896	validate_mm(mm);
1897}
1898
1899/*
1900 * Get rid of page table information in the indicated region.
1901 *
1902 * Called with the mm semaphore held.
1903 */
1904static void unmap_region(struct mm_struct *mm,
1905		struct vm_area_struct *vma, struct vm_area_struct *prev,
 
1906		unsigned long start, unsigned long end)
1907{
1908	struct vm_area_struct *next = prev? prev->vm_next: mm->mmap;
1909	struct mmu_gather tlb;
1910
1911	lru_add_drain();
1912	tlb_gather_mmu(&tlb, mm, 0);
1913	update_hiwater_rss(mm);
1914	unmap_vmas(&tlb, vma, start, end);
1915	free_pgtables(&tlb, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS,
1916				 next ? next->vm_start : 0);
1917	tlb_finish_mmu(&tlb, start, end);
1918}
1919
1920/*
1921 * Create a list of vma's touched by the unmap, removing them from the mm's
1922 * vma list as we go..
1923 */
1924static void
1925detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
1926	struct vm_area_struct *prev, unsigned long end)
1927{
1928	struct vm_area_struct **insertion_point;
1929	struct vm_area_struct *tail_vma = NULL;
1930	unsigned long addr;
1931
1932	insertion_point = (prev ? &prev->vm_next : &mm->mmap);
1933	vma->vm_prev = NULL;
1934	do {
1935		rb_erase(&vma->vm_rb, &mm->mm_rb);
1936		mm->map_count--;
1937		tail_vma = vma;
1938		vma = vma->vm_next;
1939	} while (vma && vma->vm_start < end);
1940	*insertion_point = vma;
1941	if (vma)
1942		vma->vm_prev = prev;
1943	tail_vma->vm_next = NULL;
1944	if (mm->unmap_area == arch_unmap_area)
1945		addr = prev ? prev->vm_end : mm->mmap_base;
1946	else
1947		addr = vma ?  vma->vm_start : mm->mmap_base;
1948	mm->unmap_area(mm, addr);
1949	mm->mmap_cache = NULL;		/* Kill the cache. */
1950}
1951
1952/*
1953 * __split_vma() bypasses sysctl_max_map_count checking.  We use this on the
1954 * munmap path where it doesn't make sense to fail.
1955 */
1956static int __split_vma(struct mm_struct * mm, struct vm_area_struct * vma,
1957	      unsigned long addr, int new_below)
1958{
1959	struct mempolicy *pol;
1960	struct vm_area_struct *new;
1961	int err = -ENOMEM;
 
1962
1963	if (is_vm_hugetlb_page(vma) && (addr &
1964					~(huge_page_mask(hstate_vma(vma)))))
1965		return -EINVAL;
 
 
1966
1967	new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
1968	if (!new)
1969		goto out_err;
1970
1971	/* most fields are the same, copy all, and then fixup */
1972	*new = *vma;
1973
1974	INIT_LIST_HEAD(&new->anon_vma_chain);
1975
1976	if (new_below)
1977		new->vm_end = addr;
1978	else {
1979		new->vm_start = addr;
1980		new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
1981	}
1982
1983	pol = mpol_dup(vma_policy(vma));
1984	if (IS_ERR(pol)) {
1985		err = PTR_ERR(pol);
1986		goto out_free_vma;
1987	}
1988	vma_set_policy(new, pol);
1989
1990	if (anon_vma_clone(new, vma))
 
1991		goto out_free_mpol;
1992
1993	if (new->vm_file) {
1994		get_file(new->vm_file);
1995		if (vma->vm_flags & VM_EXECUTABLE)
1996			added_exe_file_vma(mm);
1997	}
1998
1999	if (new->vm_ops && new->vm_ops->open)
2000		new->vm_ops->open(new);
2001
2002	if (new_below)
2003		err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
2004			((addr - new->vm_start) >> PAGE_SHIFT), new);
2005	else
2006		err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
2007
2008	/* Success. */
2009	if (!err)
2010		return 0;
2011
 
 
 
2012	/* Clean everything up if vma_adjust failed. */
2013	if (new->vm_ops && new->vm_ops->close)
2014		new->vm_ops->close(new);
2015	if (new->vm_file) {
2016		if (vma->vm_flags & VM_EXECUTABLE)
2017			removed_exe_file_vma(mm);
2018		fput(new->vm_file);
2019	}
2020	unlink_anon_vmas(new);
2021 out_free_mpol:
2022	mpol_put(pol);
2023 out_free_vma:
2024	kmem_cache_free(vm_area_cachep, new);
2025 out_err:
2026	return err;
2027}
2028
2029/*
2030 * Split a vma into two pieces at address 'addr', a new vma is allocated
2031 * either for the first part or the tail.
2032 */
2033int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
2034	      unsigned long addr, int new_below)
2035{
2036	if (mm->map_count >= sysctl_max_map_count)
2037		return -ENOMEM;
2038
2039	return __split_vma(mm, vma, addr, new_below);
2040}
2041
2042/* Munmap is split into 2 main parts -- this part which finds
2043 * what needs doing, and the areas themselves, which do the
2044 * work.  This now handles partial unmappings.
2045 * Jeremy Fitzhardinge <jeremy@goop.org>
2046 */
2047int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
2048{
2049	unsigned long end;
2050	struct vm_area_struct *vma, *prev, *last;
 
2051
2052	if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start)
2053		return -EINVAL;
2054
2055	if ((len = PAGE_ALIGN(len)) == 0)
2056		return -EINVAL;
2057
2058	/* Find the first overlapping VMA */
2059	vma = find_vma(mm, start);
2060	if (!vma)
2061		return 0;
2062	prev = vma->vm_prev;
2063	/* we have  start < vma->vm_end  */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2064
2065	/* if it doesn't overlap, we have nothing.. */
2066	end = start + len;
2067	if (vma->vm_start >= end)
2068		return 0;
2069
 
2070	/*
2071	 * If we need to split any vma, do it now to save pain later.
2072	 *
2073	 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2074	 * unmapped vm_area_struct will remain in use: so lower split_vma
2075	 * places tmp vma above, and higher split_vma places tmp vma below.
2076	 */
 
 
2077	if (start > vma->vm_start) {
2078		int error;
2079
2080		/*
2081		 * Make sure that map_count on return from munmap() will
2082		 * not exceed its limit; but let map_count go just above
2083		 * its limit temporarily, to help free resources as expected.
2084		 */
2085		if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count)
2086			return -ENOMEM;
2087
 
 
 
 
2088		error = __split_vma(mm, vma, start, 0);
2089		if (error)
2090			return error;
2091		prev = vma;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2092	}
2093
2094	/* Does it split the last one? */
2095	last = find_vma(mm, end);
2096	if (last && end > last->vm_start) {
2097		int error = __split_vma(mm, last, end, 1);
 
 
 
 
 
 
 
 
 
 
 
2098		if (error)
2099			return error;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2100	}
2101	vma = prev? prev->vm_next: mm->mmap;
2102
 
 
 
 
 
 
 
 
 
 
 
2103	/*
2104	 * unlock any mlock()ed ranges before detaching vmas
 
 
2105	 */
2106	if (mm->locked_vm) {
2107		struct vm_area_struct *tmp = vma;
2108		while (tmp && tmp->vm_start < end) {
2109			if (tmp->vm_flags & VM_LOCKED) {
2110				mm->locked_vm -= vma_pages(tmp);
2111				munlock_vma_pages_all(tmp);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2112			}
2113			tmp = tmp->vm_next;
2114		}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2115	}
2116
 
 
 
2117	/*
2118	 * Remove the vma's, and unmap the actual pages
 
 
 
 
2119	 */
2120	detach_vmas_to_be_unmapped(mm, vma, prev, end);
2121	unmap_region(mm, vma, prev, start, end);
2122
2123	/* Fix up all other VM information */
2124	remove_vma_list(mm, vma);
2125
2126	return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2127}
2128
2129int vm_munmap(unsigned long start, size_t len)
2130{
2131	int ret;
2132	struct mm_struct *mm = current->mm;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2133
2134	down_write(&mm->mmap_sem);
2135	ret = do_munmap(mm, start, len);
2136	up_write(&mm->mmap_sem);
2137	return ret;
2138}
 
 
 
 
 
2139EXPORT_SYMBOL(vm_munmap);
2140
2141SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
2142{
2143	profile_munmap(addr);
2144	return vm_munmap(addr, len);
2145}
2146
2147static inline void verify_mm_writelocked(struct mm_struct *mm)
2148{
2149#ifdef CONFIG_DEBUG_VM
2150	if (unlikely(down_read_trylock(&mm->mmap_sem))) {
2151		WARN_ON(1);
2152		up_read(&mm->mmap_sem);
2153	}
2154#endif
2155}
2156
2157/*
2158 *  this is really a simplified "do_mmap".  it only handles
2159 *  anonymous maps.  eventually we may be able to do some
2160 *  brk-specific accounting here.
2161 */
2162static unsigned long do_brk(unsigned long addr, unsigned long len)
 
2163{
2164	struct mm_struct * mm = current->mm;
2165	struct vm_area_struct * vma, * prev;
2166	unsigned long flags;
2167	struct rb_node ** rb_link, * rb_parent;
2168	pgoff_t pgoff = addr >> PAGE_SHIFT;
2169	int error;
2170
2171	len = PAGE_ALIGN(len);
2172	if (!len)
2173		return addr;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2174
2175	flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
 
2176
2177	error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED);
2178	if (error & ~PAGE_MASK)
2179		return error;
 
 
2180
2181	/*
2182	 * mlock MCL_FUTURE?
2183	 */
2184	if (mm->def_flags & VM_LOCKED) {
2185		unsigned long locked, lock_limit;
2186		locked = len >> PAGE_SHIFT;
2187		locked += mm->locked_vm;
2188		lock_limit = rlimit(RLIMIT_MEMLOCK);
2189		lock_limit >>= PAGE_SHIFT;
2190		if (locked > lock_limit && !capable(CAP_IPC_LOCK))
2191			return -EAGAIN;
2192	}
2193
2194	/*
2195	 * mm->mmap_sem is required to protect against another thread
2196	 * changing the mappings in case we sleep.
2197	 */
2198	verify_mm_writelocked(mm);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2199
 
2200	/*
2201	 * Clear old maps.  this also does some error checking for us
 
2202	 */
2203 munmap_back:
2204	vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
2205	if (vma && vma->vm_start < addr + len) {
2206		if (do_munmap(mm, addr, len))
2207			return -ENOMEM;
2208		goto munmap_back;
2209	}
2210
2211	/* Check against address space limits *after* clearing old maps... */
2212	if (!may_expand_vm(mm, len >> PAGE_SHIFT))
2213		return -ENOMEM;
2214
2215	if (mm->map_count > sysctl_max_map_count)
2216		return -ENOMEM;
2217
2218	if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT))
2219		return -ENOMEM;
2220
2221	/* Can we just expand an old private anonymous mapping? */
2222	vma = vma_merge(mm, prev, addr, addr + len, flags,
2223					NULL, NULL, pgoff, NULL);
2224	if (vma)
2225		goto out;
2226
2227	/*
2228	 * create a vma struct for an anonymous mapping
 
2229	 */
2230	vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
2231	if (!vma) {
2232		vm_unacct_memory(len >> PAGE_SHIFT);
2233		return -ENOMEM;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2234	}
2235
2236	INIT_LIST_HEAD(&vma->anon_vma_chain);
2237	vma->vm_mm = mm;
 
 
 
 
2238	vma->vm_start = addr;
2239	vma->vm_end = addr + len;
2240	vma->vm_pgoff = pgoff;
2241	vma->vm_flags = flags;
2242	vma->vm_page_prot = vm_get_page_prot(flags);
2243	vma_link(mm, vma, prev, rb_link, rb_parent);
 
 
 
 
2244out:
2245	perf_event_mmap(vma);
2246	mm->total_vm += len >> PAGE_SHIFT;
2247	if (flags & VM_LOCKED) {
2248		if (!mlock_vma_pages_range(vma, addr, addr + len))
2249			mm->locked_vm += (len >> PAGE_SHIFT);
2250	}
2251	return addr;
 
 
 
 
 
 
 
2252}
2253
2254unsigned long vm_brk(unsigned long addr, unsigned long len)
2255{
2256	struct mm_struct *mm = current->mm;
2257	unsigned long ret;
 
 
 
 
 
2258
2259	down_write(&mm->mmap_sem);
2260	ret = do_brk(addr, len);
2261	up_write(&mm->mmap_sem);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2262	return ret;
2263}
 
 
 
 
 
 
2264EXPORT_SYMBOL(vm_brk);
2265
2266/* Release all mmaps. */
2267void exit_mmap(struct mm_struct *mm)
2268{
2269	struct mmu_gather tlb;
2270	struct vm_area_struct *vma;
2271	unsigned long nr_accounted = 0;
 
 
2272
2273	/* mm's last user has gone, and its about to be pulled down */
2274	mmu_notifier_release(mm);
2275
2276	if (mm->locked_vm) {
2277		vma = mm->mmap;
2278		while (vma) {
2279			if (vma->vm_flags & VM_LOCKED)
2280				munlock_vma_pages_all(vma);
2281			vma = vma->vm_next;
2282		}
2283	}
2284
2285	arch_exit_mmap(mm);
2286
2287	vma = mm->mmap;
2288	if (!vma)	/* Can happen if dup_mmap() received an OOM */
 
 
2289		return;
 
2290
2291	lru_add_drain();
2292	flush_cache_mm(mm);
2293	tlb_gather_mmu(&tlb, mm, 1);
2294	/* update_hiwater_rss(mm) here? but nobody should be looking */
2295	/* Use -1 here to ensure all VMAs in the mm are unmapped */
2296	unmap_vmas(&tlb, vma, 0, -1);
 
2297
2298	free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, 0);
2299	tlb_finish_mmu(&tlb, 0, -1);
 
 
 
 
 
 
 
2300
2301	/*
2302	 * Walk the list again, actually closing and freeing it,
2303	 * with preemption enabled, without holding any MM locks.
 
2304	 */
2305	while (vma) {
2306		if (vma->vm_flags & VM_ACCOUNT)
2307			nr_accounted += vma_pages(vma);
2308		vma = remove_vma(vma);
2309	}
 
 
 
 
 
 
 
 
2310	vm_unacct_memory(nr_accounted);
2311
2312	BUG_ON(mm->nr_ptes > (FIRST_USER_ADDRESS+PMD_SIZE-1)>>PMD_SHIFT);
2313}
2314
2315/* Insert vm structure into process list sorted by address
2316 * and into the inode's i_mmap tree.  If vm_file is non-NULL
2317 * then i_mmap_mutex is taken here.
2318 */
2319int insert_vm_struct(struct mm_struct * mm, struct vm_area_struct * vma)
2320{
2321	struct vm_area_struct * __vma, * prev;
2322	struct rb_node ** rb_link, * rb_parent;
 
 
 
 
 
 
 
2323
2324	/*
2325	 * The vm_pgoff of a purely anonymous vma should be irrelevant
2326	 * until its first write fault, when page's anon_vma and index
2327	 * are set.  But now set the vm_pgoff it will almost certainly
2328	 * end up with (unless mremap moves it elsewhere before that
2329	 * first wfault), so /proc/pid/maps tells a consistent story.
2330	 *
2331	 * By setting it to reflect the virtual start address of the
2332	 * vma, merges and splits can happen in a seamless way, just
2333	 * using the existing file pgoff checks and manipulations.
2334	 * Similarly in do_mmap_pgoff and in do_brk.
2335	 */
2336	if (!vma->vm_file) {
2337		BUG_ON(vma->anon_vma);
2338		vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
2339	}
2340	__vma = find_vma_prepare(mm,vma->vm_start,&prev,&rb_link,&rb_parent);
2341	if (__vma && __vma->vm_start < vma->vm_end)
 
2342		return -ENOMEM;
2343	if ((vma->vm_flags & VM_ACCOUNT) &&
2344	     security_vm_enough_memory_mm(mm, vma_pages(vma)))
2345		return -ENOMEM;
2346
2347	if (vma->vm_file && uprobe_mmap(vma))
2348		return -EINVAL;
2349
2350	vma_link(mm, vma, prev, rb_link, rb_parent);
2351	return 0;
2352}
2353
2354/*
2355 * Copy the vma structure to a new location in the same mm,
2356 * prior to moving page table entries, to effect an mremap move.
2357 */
2358struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
2359	unsigned long addr, unsigned long len, pgoff_t pgoff)
 
2360{
2361	struct vm_area_struct *vma = *vmap;
2362	unsigned long vma_start = vma->vm_start;
2363	struct mm_struct *mm = vma->vm_mm;
2364	struct vm_area_struct *new_vma, *prev;
2365	struct rb_node **rb_link, *rb_parent;
2366	struct mempolicy *pol;
2367	bool faulted_in_anon_vma = true;
2368
 
2369	/*
2370	 * If anonymous vma has not yet been faulted, update new pgoff
2371	 * to match new location, to increase its chance of merging.
2372	 */
2373	if (unlikely(!vma->vm_file && !vma->anon_vma)) {
2374		pgoff = addr >> PAGE_SHIFT;
2375		faulted_in_anon_vma = false;
2376	}
2377
2378	find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent);
 
 
 
2379	new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
2380			vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma));
 
2381	if (new_vma) {
2382		/*
2383		 * Source vma may have been merged into new_vma
2384		 */
2385		if (unlikely(vma_start >= new_vma->vm_start &&
2386			     vma_start < new_vma->vm_end)) {
2387			/*
2388			 * The only way we can get a vma_merge with
2389			 * self during an mremap is if the vma hasn't
2390			 * been faulted in yet and we were allowed to
2391			 * reset the dst vma->vm_pgoff to the
2392			 * destination address of the mremap to allow
2393			 * the merge to happen. mremap must change the
2394			 * vm_pgoff linearity between src and dst vmas
2395			 * (in turn preventing a vma_merge) to be
2396			 * safe. It is only safe to keep the vm_pgoff
2397			 * linear if there are no pages mapped yet.
2398			 */
2399			VM_BUG_ON(faulted_in_anon_vma);
2400			*vmap = new_vma;
2401		} else
2402			anon_vma_moveto_tail(new_vma);
2403	} else {
2404		new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
2405		if (new_vma) {
2406			*new_vma = *vma;
2407			pol = mpol_dup(vma_policy(vma));
2408			if (IS_ERR(pol))
2409				goto out_free_vma;
2410			INIT_LIST_HEAD(&new_vma->anon_vma_chain);
2411			if (anon_vma_clone(new_vma, vma))
2412				goto out_free_mempol;
2413			vma_set_policy(new_vma, pol);
2414			new_vma->vm_start = addr;
2415			new_vma->vm_end = addr + len;
2416			new_vma->vm_pgoff = pgoff;
2417			if (new_vma->vm_file) {
2418				get_file(new_vma->vm_file);
2419
2420				if (uprobe_mmap(new_vma))
2421					goto out_free_mempol;
2422
2423				if (vma->vm_flags & VM_EXECUTABLE)
2424					added_exe_file_vma(mm);
2425			}
2426			if (new_vma->vm_ops && new_vma->vm_ops->open)
2427				new_vma->vm_ops->open(new_vma);
2428			vma_link(mm, new_vma, prev, rb_link, rb_parent);
2429		}
2430	}
 
2431	return new_vma;
2432
2433 out_free_mempol:
2434	mpol_put(pol);
2435 out_free_vma:
2436	kmem_cache_free(vm_area_cachep, new_vma);
 
 
 
 
 
 
 
 
 
 
2437	return NULL;
2438}
2439
2440/*
2441 * Return true if the calling process may expand its vm space by the passed
2442 * number of pages
2443 */
2444int may_expand_vm(struct mm_struct *mm, unsigned long npages)
2445{
2446	unsigned long cur = mm->total_vm;	/* pages */
2447	unsigned long lim;
 
 
 
 
 
 
 
2448
2449	lim = rlimit(RLIMIT_AS) >> PAGE_SHIFT;
 
 
 
 
 
 
 
 
 
 
 
2450
2451	if (cur + npages > lim)
2452		return 0;
2453	return 1;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2454}
2455
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2456
2457static int special_mapping_fault(struct vm_area_struct *vma,
2458				struct vm_fault *vmf)
2459{
 
2460	pgoff_t pgoff;
2461	struct page **pages;
2462
2463	/*
2464	 * special mappings have no vm_file, and in that case, the mm
2465	 * uses vm_pgoff internally. So we have to subtract it from here.
2466	 * We are allowed to do this because we are the mm; do not copy
2467	 * this code into drivers!
2468	 */
2469	pgoff = vmf->pgoff - vma->vm_pgoff;
 
 
 
2470
2471	for (pages = vma->vm_private_data; pgoff && *pages; ++pages)
2472		pgoff--;
2473
2474	if (*pages) {
2475		struct page *page = *pages;
2476		get_page(page);
2477		vmf->page = page;
2478		return 0;
2479	}
2480
2481	return VM_FAULT_SIGBUS;
2482}
2483
2484/*
2485 * Having a close hook prevents vma merging regardless of flags.
2486 */
2487static void special_mapping_close(struct vm_area_struct *vma)
2488{
2489}
2490
2491static const struct vm_operations_struct special_mapping_vmops = {
2492	.close = special_mapping_close,
2493	.fault = special_mapping_fault,
2494};
2495
2496/*
2497 * Called with mm->mmap_sem held for writing.
2498 * Insert a new vma covering the given region, with the given flags.
2499 * Its pages are supplied by the given array of struct page *.
2500 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
2501 * The region past the last page supplied will always produce SIGBUS.
2502 * The array pointer and the pages it points to are assumed to stay alive
2503 * for as long as this mapping might exist.
2504 */
2505int install_special_mapping(struct mm_struct *mm,
2506			    unsigned long addr, unsigned long len,
2507			    unsigned long vm_flags, struct page **pages)
2508{
2509	int ret;
2510	struct vm_area_struct *vma;
2511
2512	vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
 
2513	if (unlikely(vma == NULL))
2514		return -ENOMEM;
2515
2516	INIT_LIST_HEAD(&vma->anon_vma_chain);
2517	vma->vm_mm = mm;
2518	vma->vm_start = addr;
2519	vma->vm_end = addr + len;
2520
2521	vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND;
 
2522	vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
2523
2524	vma->vm_ops = &special_mapping_vmops;
2525	vma->vm_private_data = pages;
2526
2527	ret = insert_vm_struct(mm, vma);
2528	if (ret)
2529		goto out;
2530
2531	mm->total_vm += len >> PAGE_SHIFT;
2532
2533	perf_event_mmap(vma);
2534
2535	return 0;
 
2536
2537out:
2538	kmem_cache_free(vm_area_cachep, vma);
2539	return ret;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2540}
2541
2542static DEFINE_MUTEX(mm_all_locks_mutex);
2543
2544static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma)
2545{
2546	if (!test_bit(0, (unsigned long *) &anon_vma->root->head.next)) {
2547		/*
2548		 * The LSB of head.next can't change from under us
2549		 * because we hold the mm_all_locks_mutex.
2550		 */
2551		mutex_lock_nest_lock(&anon_vma->root->mutex, &mm->mmap_sem);
2552		/*
2553		 * We can safely modify head.next after taking the
2554		 * anon_vma->root->mutex. If some other vma in this mm shares
2555		 * the same anon_vma we won't take it again.
2556		 *
2557		 * No need of atomic instructions here, head.next
2558		 * can't change from under us thanks to the
2559		 * anon_vma->root->mutex.
2560		 */
2561		if (__test_and_set_bit(0, (unsigned long *)
2562				       &anon_vma->root->head.next))
2563			BUG();
2564	}
2565}
2566
2567static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping)
2568{
2569	if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
2570		/*
2571		 * AS_MM_ALL_LOCKS can't change from under us because
2572		 * we hold the mm_all_locks_mutex.
2573		 *
2574		 * Operations on ->flags have to be atomic because
2575		 * even if AS_MM_ALL_LOCKS is stable thanks to the
2576		 * mm_all_locks_mutex, there may be other cpus
2577		 * changing other bitflags in parallel to us.
2578		 */
2579		if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags))
2580			BUG();
2581		mutex_lock_nest_lock(&mapping->i_mmap_mutex, &mm->mmap_sem);
2582	}
2583}
2584
2585/*
2586 * This operation locks against the VM for all pte/vma/mm related
2587 * operations that could ever happen on a certain mm. This includes
2588 * vmtruncate, try_to_unmap, and all page faults.
2589 *
2590 * The caller must take the mmap_sem in write mode before calling
2591 * mm_take_all_locks(). The caller isn't allowed to release the
2592 * mmap_sem until mm_drop_all_locks() returns.
2593 *
2594 * mmap_sem in write mode is required in order to block all operations
2595 * that could modify pagetables and free pages without need of
2596 * altering the vma layout (for example populate_range() with
2597 * nonlinear vmas). It's also needed in write mode to avoid new
2598 * anon_vmas to be associated with existing vmas.
2599 *
2600 * A single task can't take more than one mm_take_all_locks() in a row
2601 * or it would deadlock.
2602 *
2603 * The LSB in anon_vma->head.next and the AS_MM_ALL_LOCKS bitflag in
2604 * mapping->flags avoid to take the same lock twice, if more than one
2605 * vma in this mm is backed by the same anon_vma or address_space.
2606 *
2607 * We can take all the locks in random order because the VM code
2608 * taking i_mmap_mutex or anon_vma->mutex outside the mmap_sem never
2609 * takes more than one of them in a row. Secondly we're protected
2610 * against a concurrent mm_take_all_locks() by the mm_all_locks_mutex.
 
 
 
 
 
 
2611 *
2612 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
2613 * that may have to take thousand of locks.
2614 *
2615 * mm_take_all_locks() can fail if it's interrupted by signals.
2616 */
2617int mm_take_all_locks(struct mm_struct *mm)
2618{
2619	struct vm_area_struct *vma;
2620	struct anon_vma_chain *avc;
 
2621
2622	BUG_ON(down_read_trylock(&mm->mmap_sem));
2623
2624	mutex_lock(&mm_all_locks_mutex);
2625
2626	for (vma = mm->mmap; vma; vma = vma->vm_next) {
2627		if (signal_pending(current))
2628			goto out_unlock;
2629		if (vma->vm_file && vma->vm_file->f_mapping)
 
 
 
 
 
 
 
 
 
 
2630			vm_lock_mapping(mm, vma->vm_file->f_mapping);
2631	}
2632
2633	for (vma = mm->mmap; vma; vma = vma->vm_next) {
 
2634		if (signal_pending(current))
2635			goto out_unlock;
2636		if (vma->anon_vma)
2637			list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
2638				vm_lock_anon_vma(mm, avc->anon_vma);
2639	}
2640
2641	return 0;
2642
2643out_unlock:
2644	mm_drop_all_locks(mm);
2645	return -EINTR;
2646}
2647
2648static void vm_unlock_anon_vma(struct anon_vma *anon_vma)
2649{
2650	if (test_bit(0, (unsigned long *) &anon_vma->root->head.next)) {
2651		/*
2652		 * The LSB of head.next can't change to 0 from under
2653		 * us because we hold the mm_all_locks_mutex.
2654		 *
2655		 * We must however clear the bitflag before unlocking
2656		 * the vma so the users using the anon_vma->head will
2657		 * never see our bitflag.
2658		 *
2659		 * No need of atomic instructions here, head.next
2660		 * can't change from under us until we release the
2661		 * anon_vma->root->mutex.
2662		 */
2663		if (!__test_and_clear_bit(0, (unsigned long *)
2664					  &anon_vma->root->head.next))
2665			BUG();
2666		anon_vma_unlock(anon_vma);
2667	}
2668}
2669
2670static void vm_unlock_mapping(struct address_space *mapping)
2671{
2672	if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
2673		/*
2674		 * AS_MM_ALL_LOCKS can't change to 0 from under us
2675		 * because we hold the mm_all_locks_mutex.
2676		 */
2677		mutex_unlock(&mapping->i_mmap_mutex);
2678		if (!test_and_clear_bit(AS_MM_ALL_LOCKS,
2679					&mapping->flags))
2680			BUG();
2681	}
2682}
2683
2684/*
2685 * The mmap_sem cannot be released by the caller until
2686 * mm_drop_all_locks() returns.
2687 */
2688void mm_drop_all_locks(struct mm_struct *mm)
2689{
2690	struct vm_area_struct *vma;
2691	struct anon_vma_chain *avc;
 
2692
2693	BUG_ON(down_read_trylock(&mm->mmap_sem));
2694	BUG_ON(!mutex_is_locked(&mm_all_locks_mutex));
2695
2696	for (vma = mm->mmap; vma; vma = vma->vm_next) {
2697		if (vma->anon_vma)
2698			list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
2699				vm_unlock_anon_vma(avc->anon_vma);
2700		if (vma->vm_file && vma->vm_file->f_mapping)
2701			vm_unlock_mapping(vma->vm_file->f_mapping);
2702	}
2703
2704	mutex_unlock(&mm_all_locks_mutex);
2705}
2706
2707/*
2708 * initialise the VMA slab
2709 */
2710void __init mmap_init(void)
2711{
2712	int ret;
2713
2714	ret = percpu_counter_init(&vm_committed_as, 0);
2715	VM_BUG_ON(ret);
2716}