mmap.c revision ecaf18c15aac8bb9bed7b7aa0e382fe252e275d5
1/* 2 * mm/mmap.c 3 * 4 * Written by obz. 5 * 6 * Address space accounting code <alan@redhat.com> 7 */ 8 9#include <linux/slab.h> 10#include <linux/backing-dev.h> 11#include <linux/mm.h> 12#include <linux/shm.h> 13#include <linux/mman.h> 14#include <linux/pagemap.h> 15#include <linux/swap.h> 16#include <linux/syscalls.h> 17#include <linux/capability.h> 18#include <linux/init.h> 19#include <linux/file.h> 20#include <linux/fs.h> 21#include <linux/personality.h> 22#include <linux/security.h> 23#include <linux/hugetlb.h> 24#include <linux/profile.h> 25#include <linux/module.h> 26#include <linux/mount.h> 27#include <linux/mempolicy.h> 28#include <linux/rmap.h> 29 30#include <asm/uaccess.h> 31#include <asm/cacheflush.h> 32#include <asm/tlb.h> 33#include <asm/mmu_context.h> 34 35#ifndef arch_mmap_check 36#define arch_mmap_check(addr, len, flags) (0) 37#endif 38 39static void unmap_region(struct mm_struct *mm, 40 struct vm_area_struct *vma, struct vm_area_struct *prev, 41 unsigned long start, unsigned long end); 42 43/* 44 * WARNING: the debugging will use recursive algorithms so never enable this 45 * unless you know what you are doing. 46 */ 47#undef DEBUG_MM_RB 48 49/* description of effects of mapping type and prot in current implementation. 50 * this is due to the limited x86 page protection hardware. The expected 51 * behavior is in parens: 52 * 53 * map_type prot 54 * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC 55 * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes 56 * w: (no) no w: (no) no w: (yes) yes w: (no) no 57 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes 58 * 59 * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes 60 * w: (no) no w: (no) no w: (copy) copy w: (no) no 61 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes 62 * 63 */ 64pgprot_t protection_map[16] = { 65 __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111, 66 __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111 67}; 68 69pgprot_t vm_get_page_prot(unsigned long vm_flags) 70{ 71 return protection_map[vm_flags & 72 (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)]; 73} 74EXPORT_SYMBOL(vm_get_page_prot); 75 76int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */ 77int sysctl_overcommit_ratio = 50; /* default is 50% */ 78int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT; 79atomic_t vm_committed_space = ATOMIC_INIT(0); 80 81/* 82 * Check that a process has enough memory to allocate a new virtual 83 * mapping. 0 means there is enough memory for the allocation to 84 * succeed and -ENOMEM implies there is not. 85 * 86 * We currently support three overcommit policies, which are set via the 87 * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting 88 * 89 * Strict overcommit modes added 2002 Feb 26 by Alan Cox. 90 * Additional code 2002 Jul 20 by Robert Love. 91 * 92 * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise. 93 * 94 * Note this is a helper function intended to be used by LSMs which 95 * wish to use this logic. 96 */ 97int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin) 98{ 99 unsigned long free, allowed; 100 101 vm_acct_memory(pages); 102 103 /* 104 * Sometimes we want to use more memory than we have 105 */ 106 if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS) 107 return 0; 108 109 if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) { 110 unsigned long n; 111 112 free = global_page_state(NR_FILE_PAGES); 113 free += nr_swap_pages; 114 115 /* 116 * Any slabs which are created with the 117 * SLAB_RECLAIM_ACCOUNT flag claim to have contents 118 * which are reclaimable, under pressure. The dentry 119 * cache and most inode caches should fall into this 120 */ 121 free += global_page_state(NR_SLAB_RECLAIMABLE); 122 123 /* 124 * Leave the last 3% for root 125 */ 126 if (!cap_sys_admin) 127 free -= free / 32; 128 129 if (free > pages) 130 return 0; 131 132 /* 133 * nr_free_pages() is very expensive on large systems, 134 * only call if we're about to fail. 135 */ 136 n = nr_free_pages(); 137 138 /* 139 * Leave reserved pages. The pages are not for anonymous pages. 140 */ 141 if (n <= totalreserve_pages) 142 goto error; 143 else 144 n -= totalreserve_pages; 145 146 /* 147 * Leave the last 3% for root 148 */ 149 if (!cap_sys_admin) 150 n -= n / 32; 151 free += n; 152 153 if (free > pages) 154 return 0; 155 156 goto error; 157 } 158 159 allowed = (totalram_pages - hugetlb_total_pages()) 160 * sysctl_overcommit_ratio / 100; 161 /* 162 * Leave the last 3% for root 163 */ 164 if (!cap_sys_admin) 165 allowed -= allowed / 32; 166 allowed += total_swap_pages; 167 168 /* Don't let a single process grow too big: 169 leave 3% of the size of this process for other processes */ 170 allowed -= mm->total_vm / 32; 171 172 /* 173 * cast `allowed' as a signed long because vm_committed_space 174 * sometimes has a negative value 175 */ 176 if (atomic_read(&vm_committed_space) < (long)allowed) 177 return 0; 178error: 179 vm_unacct_memory(pages); 180 181 return -ENOMEM; 182} 183 184/* 185 * Requires inode->i_mapping->i_mmap_lock 186 */ 187static void __remove_shared_vm_struct(struct vm_area_struct *vma, 188 struct file *file, struct address_space *mapping) 189{ 190 if (vma->vm_flags & VM_DENYWRITE) 191 atomic_inc(&file->f_path.dentry->d_inode->i_writecount); 192 if (vma->vm_flags & VM_SHARED) 193 mapping->i_mmap_writable--; 194 195 flush_dcache_mmap_lock(mapping); 196 if (unlikely(vma->vm_flags & VM_NONLINEAR)) 197 list_del_init(&vma->shared.vm_set.list); 198 else 199 vma_prio_tree_remove(vma, &mapping->i_mmap); 200 flush_dcache_mmap_unlock(mapping); 201} 202 203/* 204 * Unlink a file-based vm structure from its prio_tree, to hide 205 * vma from rmap and vmtruncate before freeing its page tables. 206 */ 207void unlink_file_vma(struct vm_area_struct *vma) 208{ 209 struct file *file = vma->vm_file; 210 211 if (file) { 212 struct address_space *mapping = file->f_mapping; 213 spin_lock(&mapping->i_mmap_lock); 214 __remove_shared_vm_struct(vma, file, mapping); 215 spin_unlock(&mapping->i_mmap_lock); 216 } 217} 218 219/* 220 * Close a vm structure and free it, returning the next. 221 */ 222static struct vm_area_struct *remove_vma(struct vm_area_struct *vma) 223{ 224 struct vm_area_struct *next = vma->vm_next; 225 226 might_sleep(); 227 if (vma->vm_ops && vma->vm_ops->close) 228 vma->vm_ops->close(vma); 229 if (vma->vm_file) 230 fput(vma->vm_file); 231 mpol_free(vma_policy(vma)); 232 kmem_cache_free(vm_area_cachep, vma); 233 return next; 234} 235 236asmlinkage unsigned long sys_brk(unsigned long brk) 237{ 238 unsigned long rlim, retval; 239 unsigned long newbrk, oldbrk; 240 struct mm_struct *mm = current->mm; 241 242 down_write(&mm->mmap_sem); 243 244 if (brk < mm->end_code) 245 goto out; 246 247 /* 248 * Check against rlimit here. If this check is done later after the test 249 * of oldbrk with newbrk then it can escape the test and let the data 250 * segment grow beyond its set limit the in case where the limit is 251 * not page aligned -Ram Gupta 252 */ 253 rlim = current->signal->rlim[RLIMIT_DATA].rlim_cur; 254 if (rlim < RLIM_INFINITY && brk - mm->start_data > rlim) 255 goto out; 256 257 newbrk = PAGE_ALIGN(brk); 258 oldbrk = PAGE_ALIGN(mm->brk); 259 if (oldbrk == newbrk) 260 goto set_brk; 261 262 /* Always allow shrinking brk. */ 263 if (brk <= mm->brk) { 264 if (!do_munmap(mm, newbrk, oldbrk-newbrk)) 265 goto set_brk; 266 goto out; 267 } 268 269 /* Check against existing mmap mappings. */ 270 if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE)) 271 goto out; 272 273 /* Ok, looks good - let it rip. */ 274 if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk) 275 goto out; 276set_brk: 277 mm->brk = brk; 278out: 279 retval = mm->brk; 280 up_write(&mm->mmap_sem); 281 return retval; 282} 283 284#ifdef DEBUG_MM_RB 285static int browse_rb(struct rb_root *root) 286{ 287 int i = 0, j; 288 struct rb_node *nd, *pn = NULL; 289 unsigned long prev = 0, pend = 0; 290 291 for (nd = rb_first(root); nd; nd = rb_next(nd)) { 292 struct vm_area_struct *vma; 293 vma = rb_entry(nd, struct vm_area_struct, vm_rb); 294 if (vma->vm_start < prev) 295 printk("vm_start %lx prev %lx\n", vma->vm_start, prev), i = -1; 296 if (vma->vm_start < pend) 297 printk("vm_start %lx pend %lx\n", vma->vm_start, pend); 298 if (vma->vm_start > vma->vm_end) 299 printk("vm_end %lx < vm_start %lx\n", vma->vm_end, vma->vm_start); 300 i++; 301 pn = nd; 302 prev = vma->vm_start; 303 pend = vma->vm_end; 304 } 305 j = 0; 306 for (nd = pn; nd; nd = rb_prev(nd)) { 307 j++; 308 } 309 if (i != j) 310 printk("backwards %d, forwards %d\n", j, i), i = 0; 311 return i; 312} 313 314void validate_mm(struct mm_struct *mm) 315{ 316 int bug = 0; 317 int i = 0; 318 struct vm_area_struct *tmp = mm->mmap; 319 while (tmp) { 320 tmp = tmp->vm_next; 321 i++; 322 } 323 if (i != mm->map_count) 324 printk("map_count %d vm_next %d\n", mm->map_count, i), bug = 1; 325 i = browse_rb(&mm->mm_rb); 326 if (i != mm->map_count) 327 printk("map_count %d rb %d\n", mm->map_count, i), bug = 1; 328 BUG_ON(bug); 329} 330#else 331#define validate_mm(mm) do { } while (0) 332#endif 333 334static struct vm_area_struct * 335find_vma_prepare(struct mm_struct *mm, unsigned long addr, 336 struct vm_area_struct **pprev, struct rb_node ***rb_link, 337 struct rb_node ** rb_parent) 338{ 339 struct vm_area_struct * vma; 340 struct rb_node ** __rb_link, * __rb_parent, * rb_prev; 341 342 __rb_link = &mm->mm_rb.rb_node; 343 rb_prev = __rb_parent = NULL; 344 vma = NULL; 345 346 while (*__rb_link) { 347 struct vm_area_struct *vma_tmp; 348 349 __rb_parent = *__rb_link; 350 vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb); 351 352 if (vma_tmp->vm_end > addr) { 353 vma = vma_tmp; 354 if (vma_tmp->vm_start <= addr) 355 return vma; 356 __rb_link = &__rb_parent->rb_left; 357 } else { 358 rb_prev = __rb_parent; 359 __rb_link = &__rb_parent->rb_right; 360 } 361 } 362 363 *pprev = NULL; 364 if (rb_prev) 365 *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb); 366 *rb_link = __rb_link; 367 *rb_parent = __rb_parent; 368 return vma; 369} 370 371static inline void 372__vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma, 373 struct vm_area_struct *prev, struct rb_node *rb_parent) 374{ 375 if (prev) { 376 vma->vm_next = prev->vm_next; 377 prev->vm_next = vma; 378 } else { 379 mm->mmap = vma; 380 if (rb_parent) 381 vma->vm_next = rb_entry(rb_parent, 382 struct vm_area_struct, vm_rb); 383 else 384 vma->vm_next = NULL; 385 } 386} 387 388void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma, 389 struct rb_node **rb_link, struct rb_node *rb_parent) 390{ 391 rb_link_node(&vma->vm_rb, rb_parent, rb_link); 392 rb_insert_color(&vma->vm_rb, &mm->mm_rb); 393} 394 395static inline void __vma_link_file(struct vm_area_struct *vma) 396{ 397 struct file * file; 398 399 file = vma->vm_file; 400 if (file) { 401 struct address_space *mapping = file->f_mapping; 402 403 if (vma->vm_flags & VM_DENYWRITE) 404 atomic_dec(&file->f_path.dentry->d_inode->i_writecount); 405 if (vma->vm_flags & VM_SHARED) 406 mapping->i_mmap_writable++; 407 408 flush_dcache_mmap_lock(mapping); 409 if (unlikely(vma->vm_flags & VM_NONLINEAR)) 410 vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear); 411 else 412 vma_prio_tree_insert(vma, &mapping->i_mmap); 413 flush_dcache_mmap_unlock(mapping); 414 } 415} 416 417static void 418__vma_link(struct mm_struct *mm, struct vm_area_struct *vma, 419 struct vm_area_struct *prev, struct rb_node **rb_link, 420 struct rb_node *rb_parent) 421{ 422 __vma_link_list(mm, vma, prev, rb_parent); 423 __vma_link_rb(mm, vma, rb_link, rb_parent); 424 __anon_vma_link(vma); 425} 426 427static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma, 428 struct vm_area_struct *prev, struct rb_node **rb_link, 429 struct rb_node *rb_parent) 430{ 431 struct address_space *mapping = NULL; 432 433 if (vma->vm_file) 434 mapping = vma->vm_file->f_mapping; 435 436 if (mapping) { 437 spin_lock(&mapping->i_mmap_lock); 438 vma->vm_truncate_count = mapping->truncate_count; 439 } 440 anon_vma_lock(vma); 441 442 __vma_link(mm, vma, prev, rb_link, rb_parent); 443 __vma_link_file(vma); 444 445 anon_vma_unlock(vma); 446 if (mapping) 447 spin_unlock(&mapping->i_mmap_lock); 448 449 mm->map_count++; 450 validate_mm(mm); 451} 452 453/* 454 * Helper for vma_adjust in the split_vma insert case: 455 * insert vm structure into list and rbtree and anon_vma, 456 * but it has already been inserted into prio_tree earlier. 457 */ 458static void 459__insert_vm_struct(struct mm_struct * mm, struct vm_area_struct * vma) 460{ 461 struct vm_area_struct * __vma, * prev; 462 struct rb_node ** rb_link, * rb_parent; 463 464 __vma = find_vma_prepare(mm, vma->vm_start,&prev, &rb_link, &rb_parent); 465 BUG_ON(__vma && __vma->vm_start < vma->vm_end); 466 __vma_link(mm, vma, prev, rb_link, rb_parent); 467 mm->map_count++; 468} 469 470static inline void 471__vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma, 472 struct vm_area_struct *prev) 473{ 474 prev->vm_next = vma->vm_next; 475 rb_erase(&vma->vm_rb, &mm->mm_rb); 476 if (mm->mmap_cache == vma) 477 mm->mmap_cache = prev; 478} 479 480/* 481 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that 482 * is already present in an i_mmap tree without adjusting the tree. 483 * The following helper function should be used when such adjustments 484 * are necessary. The "insert" vma (if any) is to be inserted 485 * before we drop the necessary locks. 486 */ 487void vma_adjust(struct vm_area_struct *vma, unsigned long start, 488 unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert) 489{ 490 struct mm_struct *mm = vma->vm_mm; 491 struct vm_area_struct *next = vma->vm_next; 492 struct vm_area_struct *importer = NULL; 493 struct address_space *mapping = NULL; 494 struct prio_tree_root *root = NULL; 495 struct file *file = vma->vm_file; 496 struct anon_vma *anon_vma = NULL; 497 long adjust_next = 0; 498 int remove_next = 0; 499 500 if (next && !insert) { 501 if (end >= next->vm_end) { 502 /* 503 * vma expands, overlapping all the next, and 504 * perhaps the one after too (mprotect case 6). 505 */ 506again: remove_next = 1 + (end > next->vm_end); 507 end = next->vm_end; 508 anon_vma = next->anon_vma; 509 importer = vma; 510 } else if (end > next->vm_start) { 511 /* 512 * vma expands, overlapping part of the next: 513 * mprotect case 5 shifting the boundary up. 514 */ 515 adjust_next = (end - next->vm_start) >> PAGE_SHIFT; 516 anon_vma = next->anon_vma; 517 importer = vma; 518 } else if (end < vma->vm_end) { 519 /* 520 * vma shrinks, and !insert tells it's not 521 * split_vma inserting another: so it must be 522 * mprotect case 4 shifting the boundary down. 523 */ 524 adjust_next = - ((vma->vm_end - end) >> PAGE_SHIFT); 525 anon_vma = next->anon_vma; 526 importer = next; 527 } 528 } 529 530 if (file) { 531 mapping = file->f_mapping; 532 if (!(vma->vm_flags & VM_NONLINEAR)) 533 root = &mapping->i_mmap; 534 spin_lock(&mapping->i_mmap_lock); 535 if (importer && 536 vma->vm_truncate_count != next->vm_truncate_count) { 537 /* 538 * unmap_mapping_range might be in progress: 539 * ensure that the expanding vma is rescanned. 540 */ 541 importer->vm_truncate_count = 0; 542 } 543 if (insert) { 544 insert->vm_truncate_count = vma->vm_truncate_count; 545 /* 546 * Put into prio_tree now, so instantiated pages 547 * are visible to arm/parisc __flush_dcache_page 548 * throughout; but we cannot insert into address 549 * space until vma start or end is updated. 550 */ 551 __vma_link_file(insert); 552 } 553 } 554 555 /* 556 * When changing only vma->vm_end, we don't really need 557 * anon_vma lock: but is that case worth optimizing out? 558 */ 559 if (vma->anon_vma) 560 anon_vma = vma->anon_vma; 561 if (anon_vma) { 562 spin_lock(&anon_vma->lock); 563 /* 564 * Easily overlooked: when mprotect shifts the boundary, 565 * make sure the expanding vma has anon_vma set if the 566 * shrinking vma had, to cover any anon pages imported. 567 */ 568 if (importer && !importer->anon_vma) { 569 importer->anon_vma = anon_vma; 570 __anon_vma_link(importer); 571 } 572 } 573 574 if (root) { 575 flush_dcache_mmap_lock(mapping); 576 vma_prio_tree_remove(vma, root); 577 if (adjust_next) 578 vma_prio_tree_remove(next, root); 579 } 580 581 vma->vm_start = start; 582 vma->vm_end = end; 583 vma->vm_pgoff = pgoff; 584 if (adjust_next) { 585 next->vm_start += adjust_next << PAGE_SHIFT; 586 next->vm_pgoff += adjust_next; 587 } 588 589 if (root) { 590 if (adjust_next) 591 vma_prio_tree_insert(next, root); 592 vma_prio_tree_insert(vma, root); 593 flush_dcache_mmap_unlock(mapping); 594 } 595 596 if (remove_next) { 597 /* 598 * vma_merge has merged next into vma, and needs 599 * us to remove next before dropping the locks. 600 */ 601 __vma_unlink(mm, next, vma); 602 if (file) 603 __remove_shared_vm_struct(next, file, mapping); 604 if (next->anon_vma) 605 __anon_vma_merge(vma, next); 606 } else if (insert) { 607 /* 608 * split_vma has split insert from vma, and needs 609 * us to insert it before dropping the locks 610 * (it may either follow vma or precede it). 611 */ 612 __insert_vm_struct(mm, insert); 613 } 614 615 if (anon_vma) 616 spin_unlock(&anon_vma->lock); 617 if (mapping) 618 spin_unlock(&mapping->i_mmap_lock); 619 620 if (remove_next) { 621 if (file) 622 fput(file); 623 mm->map_count--; 624 mpol_free(vma_policy(next)); 625 kmem_cache_free(vm_area_cachep, next); 626 /* 627 * In mprotect's case 6 (see comments on vma_merge), 628 * we must remove another next too. It would clutter 629 * up the code too much to do both in one go. 630 */ 631 if (remove_next == 2) { 632 next = vma->vm_next; 633 goto again; 634 } 635 } 636 637 validate_mm(mm); 638} 639 640/* 641 * If the vma has a ->close operation then the driver probably needs to release 642 * per-vma resources, so we don't attempt to merge those. 643 */ 644#define VM_SPECIAL (VM_IO | VM_DONTEXPAND | VM_RESERVED | VM_PFNMAP) 645 646static inline int is_mergeable_vma(struct vm_area_struct *vma, 647 struct file *file, unsigned long vm_flags) 648{ 649 if (vma->vm_flags != vm_flags) 650 return 0; 651 if (vma->vm_file != file) 652 return 0; 653 if (vma->vm_ops && vma->vm_ops->close) 654 return 0; 655 return 1; 656} 657 658static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1, 659 struct anon_vma *anon_vma2) 660{ 661 return !anon_vma1 || !anon_vma2 || (anon_vma1 == anon_vma2); 662} 663 664/* 665 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) 666 * in front of (at a lower virtual address and file offset than) the vma. 667 * 668 * We cannot merge two vmas if they have differently assigned (non-NULL) 669 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. 670 * 671 * We don't check here for the merged mmap wrapping around the end of pagecache 672 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which 673 * wrap, nor mmaps which cover the final page at index -1UL. 674 */ 675static int 676can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags, 677 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff) 678{ 679 if (is_mergeable_vma(vma, file, vm_flags) && 680 is_mergeable_anon_vma(anon_vma, vma->anon_vma)) { 681 if (vma->vm_pgoff == vm_pgoff) 682 return 1; 683 } 684 return 0; 685} 686 687/* 688 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) 689 * beyond (at a higher virtual address and file offset than) the vma. 690 * 691 * We cannot merge two vmas if they have differently assigned (non-NULL) 692 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. 693 */ 694static int 695can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags, 696 struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff) 697{ 698 if (is_mergeable_vma(vma, file, vm_flags) && 699 is_mergeable_anon_vma(anon_vma, vma->anon_vma)) { 700 pgoff_t vm_pglen; 701 vm_pglen = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT; 702 if (vma->vm_pgoff + vm_pglen == vm_pgoff) 703 return 1; 704 } 705 return 0; 706} 707 708/* 709 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out 710 * whether that can be merged with its predecessor or its successor. 711 * Or both (it neatly fills a hole). 712 * 713 * In most cases - when called for mmap, brk or mremap - [addr,end) is 714 * certain not to be mapped by the time vma_merge is called; but when 715 * called for mprotect, it is certain to be already mapped (either at 716 * an offset within prev, or at the start of next), and the flags of 717 * this area are about to be changed to vm_flags - and the no-change 718 * case has already been eliminated. 719 * 720 * The following mprotect cases have to be considered, where AAAA is 721 * the area passed down from mprotect_fixup, never extending beyond one 722 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after: 723 * 724 * AAAA AAAA AAAA AAAA 725 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX 726 * cannot merge might become might become might become 727 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or 728 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or 729 * mremap move: PPPPNNNNNNNN 8 730 * AAAA 731 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN 732 * might become case 1 below case 2 below case 3 below 733 * 734 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX: 735 * mprotect_fixup updates vm_flags & vm_page_prot on successful return. 736 */ 737struct vm_area_struct *vma_merge(struct mm_struct *mm, 738 struct vm_area_struct *prev, unsigned long addr, 739 unsigned long end, unsigned long vm_flags, 740 struct anon_vma *anon_vma, struct file *file, 741 pgoff_t pgoff, struct mempolicy *policy) 742{ 743 pgoff_t pglen = (end - addr) >> PAGE_SHIFT; 744 struct vm_area_struct *area, *next; 745 746 /* 747 * We later require that vma->vm_flags == vm_flags, 748 * so this tests vma->vm_flags & VM_SPECIAL, too. 749 */ 750 if (vm_flags & VM_SPECIAL) 751 return NULL; 752 753 if (prev) 754 next = prev->vm_next; 755 else 756 next = mm->mmap; 757 area = next; 758 if (next && next->vm_end == end) /* cases 6, 7, 8 */ 759 next = next->vm_next; 760 761 /* 762 * Can it merge with the predecessor? 763 */ 764 if (prev && prev->vm_end == addr && 765 mpol_equal(vma_policy(prev), policy) && 766 can_vma_merge_after(prev, vm_flags, 767 anon_vma, file, pgoff)) { 768 /* 769 * OK, it can. Can we now merge in the successor as well? 770 */ 771 if (next && end == next->vm_start && 772 mpol_equal(policy, vma_policy(next)) && 773 can_vma_merge_before(next, vm_flags, 774 anon_vma, file, pgoff+pglen) && 775 is_mergeable_anon_vma(prev->anon_vma, 776 next->anon_vma)) { 777 /* cases 1, 6 */ 778 vma_adjust(prev, prev->vm_start, 779 next->vm_end, prev->vm_pgoff, NULL); 780 } else /* cases 2, 5, 7 */ 781 vma_adjust(prev, prev->vm_start, 782 end, prev->vm_pgoff, NULL); 783 return prev; 784 } 785 786 /* 787 * Can this new request be merged in front of next? 788 */ 789 if (next && end == next->vm_start && 790 mpol_equal(policy, vma_policy(next)) && 791 can_vma_merge_before(next, vm_flags, 792 anon_vma, file, pgoff+pglen)) { 793 if (prev && addr < prev->vm_end) /* case 4 */ 794 vma_adjust(prev, prev->vm_start, 795 addr, prev->vm_pgoff, NULL); 796 else /* cases 3, 8 */ 797 vma_adjust(area, addr, next->vm_end, 798 next->vm_pgoff - pglen, NULL); 799 return area; 800 } 801 802 return NULL; 803} 804 805/* 806 * find_mergeable_anon_vma is used by anon_vma_prepare, to check 807 * neighbouring vmas for a suitable anon_vma, before it goes off 808 * to allocate a new anon_vma. It checks because a repetitive 809 * sequence of mprotects and faults may otherwise lead to distinct 810 * anon_vmas being allocated, preventing vma merge in subsequent 811 * mprotect. 812 */ 813struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma) 814{ 815 struct vm_area_struct *near; 816 unsigned long vm_flags; 817 818 near = vma->vm_next; 819 if (!near) 820 goto try_prev; 821 822 /* 823 * Since only mprotect tries to remerge vmas, match flags 824 * which might be mprotected into each other later on. 825 * Neither mlock nor madvise tries to remerge at present, 826 * so leave their flags as obstructing a merge. 827 */ 828 vm_flags = vma->vm_flags & ~(VM_READ|VM_WRITE|VM_EXEC); 829 vm_flags |= near->vm_flags & (VM_READ|VM_WRITE|VM_EXEC); 830 831 if (near->anon_vma && vma->vm_end == near->vm_start && 832 mpol_equal(vma_policy(vma), vma_policy(near)) && 833 can_vma_merge_before(near, vm_flags, 834 NULL, vma->vm_file, vma->vm_pgoff + 835 ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT))) 836 return near->anon_vma; 837try_prev: 838 /* 839 * It is potentially slow to have to call find_vma_prev here. 840 * But it's only on the first write fault on the vma, not 841 * every time, and we could devise a way to avoid it later 842 * (e.g. stash info in next's anon_vma_node when assigning 843 * an anon_vma, or when trying vma_merge). Another time. 844 */ 845 BUG_ON(find_vma_prev(vma->vm_mm, vma->vm_start, &near) != vma); 846 if (!near) 847 goto none; 848 849 vm_flags = vma->vm_flags & ~(VM_READ|VM_WRITE|VM_EXEC); 850 vm_flags |= near->vm_flags & (VM_READ|VM_WRITE|VM_EXEC); 851 852 if (near->anon_vma && near->vm_end == vma->vm_start && 853 mpol_equal(vma_policy(near), vma_policy(vma)) && 854 can_vma_merge_after(near, vm_flags, 855 NULL, vma->vm_file, vma->vm_pgoff)) 856 return near->anon_vma; 857none: 858 /* 859 * There's no absolute need to look only at touching neighbours: 860 * we could search further afield for "compatible" anon_vmas. 861 * But it would probably just be a waste of time searching, 862 * or lead to too many vmas hanging off the same anon_vma. 863 * We're trying to allow mprotect remerging later on, 864 * not trying to minimize memory used for anon_vmas. 865 */ 866 return NULL; 867} 868 869#ifdef CONFIG_PROC_FS 870void vm_stat_account(struct mm_struct *mm, unsigned long flags, 871 struct file *file, long pages) 872{ 873 const unsigned long stack_flags 874 = VM_STACK_FLAGS & (VM_GROWSUP|VM_GROWSDOWN); 875 876 if (file) { 877 mm->shared_vm += pages; 878 if ((flags & (VM_EXEC|VM_WRITE)) == VM_EXEC) 879 mm->exec_vm += pages; 880 } else if (flags & stack_flags) 881 mm->stack_vm += pages; 882 if (flags & (VM_RESERVED|VM_IO)) 883 mm->reserved_vm += pages; 884} 885#endif /* CONFIG_PROC_FS */ 886 887/* 888 * The caller must hold down_write(current->mm->mmap_sem). 889 */ 890 891unsigned long do_mmap_pgoff(struct file * file, unsigned long addr, 892 unsigned long len, unsigned long prot, 893 unsigned long flags, unsigned long pgoff) 894{ 895 struct mm_struct * mm = current->mm; 896 struct inode *inode; 897 unsigned int vm_flags; 898 int error; 899 int accountable = 1; 900 unsigned long reqprot = prot; 901 902 /* 903 * Does the application expect PROT_READ to imply PROT_EXEC? 904 * 905 * (the exception is when the underlying filesystem is noexec 906 * mounted, in which case we dont add PROT_EXEC.) 907 */ 908 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC)) 909 if (!(file && (file->f_path.mnt->mnt_flags & MNT_NOEXEC))) 910 prot |= PROT_EXEC; 911 912 if (!len) 913 return -EINVAL; 914 915 error = arch_mmap_check(addr, len, flags); 916 if (error) 917 return error; 918 919 /* Careful about overflows.. */ 920 len = PAGE_ALIGN(len); 921 if (!len || len > TASK_SIZE) 922 return -ENOMEM; 923 924 /* offset overflow? */ 925 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff) 926 return -EOVERFLOW; 927 928 /* Too many mappings? */ 929 if (mm->map_count > sysctl_max_map_count) 930 return -ENOMEM; 931 932 /* Obtain the address to map to. we verify (or select) it and ensure 933 * that it represents a valid section of the address space. 934 */ 935 addr = get_unmapped_area(file, addr, len, pgoff, flags); 936 if (addr & ~PAGE_MASK) 937 return addr; 938 939 /* Do simple checking here so the lower-level routines won't have 940 * to. we assume access permissions have been handled by the open 941 * of the memory object, so we don't do any here. 942 */ 943 vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) | 944 mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC; 945 946 if (flags & MAP_LOCKED) { 947 if (!can_do_mlock()) 948 return -EPERM; 949 vm_flags |= VM_LOCKED; 950 } 951 /* mlock MCL_FUTURE? */ 952 if (vm_flags & VM_LOCKED) { 953 unsigned long locked, lock_limit; 954 locked = len >> PAGE_SHIFT; 955 locked += mm->locked_vm; 956 lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur; 957 lock_limit >>= PAGE_SHIFT; 958 if (locked > lock_limit && !capable(CAP_IPC_LOCK)) 959 return -EAGAIN; 960 } 961 962 inode = file ? file->f_path.dentry->d_inode : NULL; 963 964 if (file) { 965 switch (flags & MAP_TYPE) { 966 case MAP_SHARED: 967 if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE)) 968 return -EACCES; 969 970 /* 971 * Make sure we don't allow writing to an append-only 972 * file.. 973 */ 974 if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE)) 975 return -EACCES; 976 977 /* 978 * Make sure there are no mandatory locks on the file. 979 */ 980 if (locks_verify_locked(inode)) 981 return -EAGAIN; 982 983 vm_flags |= VM_SHARED | VM_MAYSHARE; 984 if (!(file->f_mode & FMODE_WRITE)) 985 vm_flags &= ~(VM_MAYWRITE | VM_SHARED); 986 987 /* fall through */ 988 case MAP_PRIVATE: 989 if (!(file->f_mode & FMODE_READ)) 990 return -EACCES; 991 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) { 992 if (vm_flags & VM_EXEC) 993 return -EPERM; 994 vm_flags &= ~VM_MAYEXEC; 995 } 996 if (is_file_hugepages(file)) 997 accountable = 0; 998 999 if (!file->f_op || !file->f_op->mmap) 1000 return -ENODEV; 1001 break; 1002 1003 default: 1004 return -EINVAL; 1005 } 1006 } else { 1007 switch (flags & MAP_TYPE) { 1008 case MAP_SHARED: 1009 vm_flags |= VM_SHARED | VM_MAYSHARE; 1010 break; 1011 case MAP_PRIVATE: 1012 /* 1013 * Set pgoff according to addr for anon_vma. 1014 */ 1015 pgoff = addr >> PAGE_SHIFT; 1016 break; 1017 default: 1018 return -EINVAL; 1019 } 1020 } 1021 1022 error = security_file_mmap(file, reqprot, prot, flags, addr, 0); 1023 if (error) 1024 return error; 1025 1026 return mmap_region(file, addr, len, flags, vm_flags, pgoff, 1027 accountable); 1028} 1029EXPORT_SYMBOL(do_mmap_pgoff); 1030 1031/* 1032 * Some shared mappigns will want the pages marked read-only 1033 * to track write events. If so, we'll downgrade vm_page_prot 1034 * to the private version (using protection_map[] without the 1035 * VM_SHARED bit). 1036 */ 1037int vma_wants_writenotify(struct vm_area_struct *vma) 1038{ 1039 unsigned int vm_flags = vma->vm_flags; 1040 1041 /* If it was private or non-writable, the write bit is already clear */ 1042 if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED))) 1043 return 0; 1044 1045 /* The backer wishes to know when pages are first written to? */ 1046 if (vma->vm_ops && vma->vm_ops->page_mkwrite) 1047 return 1; 1048 1049 /* The open routine did something to the protections already? */ 1050 if (pgprot_val(vma->vm_page_prot) != 1051 pgprot_val(vm_get_page_prot(vm_flags))) 1052 return 0; 1053 1054 /* Specialty mapping? */ 1055 if (vm_flags & (VM_PFNMAP|VM_INSERTPAGE)) 1056 return 0; 1057 1058 /* Can the mapping track the dirty pages? */ 1059 return vma->vm_file && vma->vm_file->f_mapping && 1060 mapping_cap_account_dirty(vma->vm_file->f_mapping); 1061} 1062 1063 1064unsigned long mmap_region(struct file *file, unsigned long addr, 1065 unsigned long len, unsigned long flags, 1066 unsigned int vm_flags, unsigned long pgoff, 1067 int accountable) 1068{ 1069 struct mm_struct *mm = current->mm; 1070 struct vm_area_struct *vma, *prev; 1071 int correct_wcount = 0; 1072 int error; 1073 struct rb_node **rb_link, *rb_parent; 1074 unsigned long charged = 0; 1075 struct inode *inode = file ? file->f_path.dentry->d_inode : NULL; 1076 1077 /* Clear old maps */ 1078 error = -ENOMEM; 1079munmap_back: 1080 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent); 1081 if (vma && vma->vm_start < addr + len) { 1082 if (do_munmap(mm, addr, len)) 1083 return -ENOMEM; 1084 goto munmap_back; 1085 } 1086 1087 /* Check against address space limit. */ 1088 if (!may_expand_vm(mm, len >> PAGE_SHIFT)) 1089 return -ENOMEM; 1090 1091 if (accountable && (!(flags & MAP_NORESERVE) || 1092 sysctl_overcommit_memory == OVERCOMMIT_NEVER)) { 1093 if (vm_flags & VM_SHARED) { 1094 /* Check memory availability in shmem_file_setup? */ 1095 vm_flags |= VM_ACCOUNT; 1096 } else if (vm_flags & VM_WRITE) { 1097 /* 1098 * Private writable mapping: check memory availability 1099 */ 1100 charged = len >> PAGE_SHIFT; 1101 if (security_vm_enough_memory(charged)) 1102 return -ENOMEM; 1103 vm_flags |= VM_ACCOUNT; 1104 } 1105 } 1106 1107 /* 1108 * Can we just expand an old private anonymous mapping? 1109 * The VM_SHARED test is necessary because shmem_zero_setup 1110 * will create the file object for a shared anonymous map below. 1111 */ 1112 if (!file && !(vm_flags & VM_SHARED) && 1113 vma_merge(mm, prev, addr, addr + len, vm_flags, 1114 NULL, NULL, pgoff, NULL)) 1115 goto out; 1116 1117 /* 1118 * Determine the object being mapped and call the appropriate 1119 * specific mapper. the address has already been validated, but 1120 * not unmapped, but the maps are removed from the list. 1121 */ 1122 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); 1123 if (!vma) { 1124 error = -ENOMEM; 1125 goto unacct_error; 1126 } 1127 1128 vma->vm_mm = mm; 1129 vma->vm_start = addr; 1130 vma->vm_end = addr + len; 1131 vma->vm_flags = vm_flags; 1132 vma->vm_page_prot = vm_get_page_prot(vm_flags); 1133 vma->vm_pgoff = pgoff; 1134 1135 if (file) { 1136 error = -EINVAL; 1137 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP)) 1138 goto free_vma; 1139 if (vm_flags & VM_DENYWRITE) { 1140 error = deny_write_access(file); 1141 if (error) 1142 goto free_vma; 1143 correct_wcount = 1; 1144 } 1145 vma->vm_file = file; 1146 get_file(file); 1147 error = file->f_op->mmap(file, vma); 1148 if (error) 1149 goto unmap_and_free_vma; 1150 } else if (vm_flags & VM_SHARED) { 1151 error = shmem_zero_setup(vma); 1152 if (error) 1153 goto free_vma; 1154 } 1155 1156 /* We set VM_ACCOUNT in a shared mapping's vm_flags, to inform 1157 * shmem_zero_setup (perhaps called through /dev/zero's ->mmap) 1158 * that memory reservation must be checked; but that reservation 1159 * belongs to shared memory object, not to vma: so now clear it. 1160 */ 1161 if ((vm_flags & (VM_SHARED|VM_ACCOUNT)) == (VM_SHARED|VM_ACCOUNT)) 1162 vma->vm_flags &= ~VM_ACCOUNT; 1163 1164 /* Can addr have changed?? 1165 * 1166 * Answer: Yes, several device drivers can do it in their 1167 * f_op->mmap method. -DaveM 1168 */ 1169 addr = vma->vm_start; 1170 pgoff = vma->vm_pgoff; 1171 vm_flags = vma->vm_flags; 1172 1173 if (vma_wants_writenotify(vma)) 1174 vma->vm_page_prot = vm_get_page_prot(vm_flags & ~VM_SHARED); 1175 1176 if (!file || !vma_merge(mm, prev, addr, vma->vm_end, 1177 vma->vm_flags, NULL, file, pgoff, vma_policy(vma))) { 1178 file = vma->vm_file; 1179 vma_link(mm, vma, prev, rb_link, rb_parent); 1180 if (correct_wcount) 1181 atomic_inc(&inode->i_writecount); 1182 } else { 1183 if (file) { 1184 if (correct_wcount) 1185 atomic_inc(&inode->i_writecount); 1186 fput(file); 1187 } 1188 mpol_free(vma_policy(vma)); 1189 kmem_cache_free(vm_area_cachep, vma); 1190 } 1191out: 1192 mm->total_vm += len >> PAGE_SHIFT; 1193 vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT); 1194 if (vm_flags & VM_LOCKED) { 1195 mm->locked_vm += len >> PAGE_SHIFT; 1196 make_pages_present(addr, addr + len); 1197 } 1198 if ((flags & MAP_POPULATE) && !(flags & MAP_NONBLOCK)) 1199 make_pages_present(addr, addr + len); 1200 return addr; 1201 1202unmap_and_free_vma: 1203 if (correct_wcount) 1204 atomic_inc(&inode->i_writecount); 1205 vma->vm_file = NULL; 1206 fput(file); 1207 1208 /* Undo any partial mapping done by a device driver. */ 1209 unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end); 1210 charged = 0; 1211free_vma: 1212 kmem_cache_free(vm_area_cachep, vma); 1213unacct_error: 1214 if (charged) 1215 vm_unacct_memory(charged); 1216 return error; 1217} 1218 1219/* Get an address range which is currently unmapped. 1220 * For shmat() with addr=0. 1221 * 1222 * Ugly calling convention alert: 1223 * Return value with the low bits set means error value, 1224 * ie 1225 * if (ret & ~PAGE_MASK) 1226 * error = ret; 1227 * 1228 * This function "knows" that -ENOMEM has the bits set. 1229 */ 1230#ifndef HAVE_ARCH_UNMAPPED_AREA 1231unsigned long 1232arch_get_unmapped_area(struct file *filp, unsigned long addr, 1233 unsigned long len, unsigned long pgoff, unsigned long flags) 1234{ 1235 struct mm_struct *mm = current->mm; 1236 struct vm_area_struct *vma; 1237 unsigned long start_addr; 1238 1239 if (len > TASK_SIZE) 1240 return -ENOMEM; 1241 1242 if (flags & MAP_FIXED) 1243 return addr; 1244 1245 if (addr) { 1246 addr = PAGE_ALIGN(addr); 1247 vma = find_vma(mm, addr); 1248 if (TASK_SIZE - len >= addr && 1249 (!vma || addr + len <= vma->vm_start)) 1250 return addr; 1251 } 1252 if (len > mm->cached_hole_size) { 1253 start_addr = addr = mm->free_area_cache; 1254 } else { 1255 start_addr = addr = TASK_UNMAPPED_BASE; 1256 mm->cached_hole_size = 0; 1257 } 1258 1259full_search: 1260 for (vma = find_vma(mm, addr); ; vma = vma->vm_next) { 1261 /* At this point: (!vma || addr < vma->vm_end). */ 1262 if (TASK_SIZE - len < addr) { 1263 /* 1264 * Start a new search - just in case we missed 1265 * some holes. 1266 */ 1267 if (start_addr != TASK_UNMAPPED_BASE) { 1268 addr = TASK_UNMAPPED_BASE; 1269 start_addr = addr; 1270 mm->cached_hole_size = 0; 1271 goto full_search; 1272 } 1273 return -ENOMEM; 1274 } 1275 if (!vma || addr + len <= vma->vm_start) { 1276 /* 1277 * Remember the place where we stopped the search: 1278 */ 1279 mm->free_area_cache = addr + len; 1280 return addr; 1281 } 1282 if (addr + mm->cached_hole_size < vma->vm_start) 1283 mm->cached_hole_size = vma->vm_start - addr; 1284 addr = vma->vm_end; 1285 } 1286} 1287#endif 1288 1289void arch_unmap_area(struct mm_struct *mm, unsigned long addr) 1290{ 1291 /* 1292 * Is this a new hole at the lowest possible address? 1293 */ 1294 if (addr >= TASK_UNMAPPED_BASE && addr < mm->free_area_cache) { 1295 mm->free_area_cache = addr; 1296 mm->cached_hole_size = ~0UL; 1297 } 1298} 1299 1300/* 1301 * This mmap-allocator allocates new areas top-down from below the 1302 * stack's low limit (the base): 1303 */ 1304#ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN 1305unsigned long 1306arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0, 1307 const unsigned long len, const unsigned long pgoff, 1308 const unsigned long flags) 1309{ 1310 struct vm_area_struct *vma; 1311 struct mm_struct *mm = current->mm; 1312 unsigned long addr = addr0; 1313 1314 /* requested length too big for entire address space */ 1315 if (len > TASK_SIZE) 1316 return -ENOMEM; 1317 1318 if (flags & MAP_FIXED) 1319 return addr; 1320 1321 /* requesting a specific address */ 1322 if (addr) { 1323 addr = PAGE_ALIGN(addr); 1324 vma = find_vma(mm, addr); 1325 if (TASK_SIZE - len >= addr && 1326 (!vma || addr + len <= vma->vm_start)) 1327 return addr; 1328 } 1329 1330 /* check if free_area_cache is useful for us */ 1331 if (len <= mm->cached_hole_size) { 1332 mm->cached_hole_size = 0; 1333 mm->free_area_cache = mm->mmap_base; 1334 } 1335 1336 /* either no address requested or can't fit in requested address hole */ 1337 addr = mm->free_area_cache; 1338 1339 /* make sure it can fit in the remaining address space */ 1340 if (addr > len) { 1341 vma = find_vma(mm, addr-len); 1342 if (!vma || addr <= vma->vm_start) 1343 /* remember the address as a hint for next time */ 1344 return (mm->free_area_cache = addr-len); 1345 } 1346 1347 if (mm->mmap_base < len) 1348 goto bottomup; 1349 1350 addr = mm->mmap_base-len; 1351 1352 do { 1353 /* 1354 * Lookup failure means no vma is above this address, 1355 * else if new region fits below vma->vm_start, 1356 * return with success: 1357 */ 1358 vma = find_vma(mm, addr); 1359 if (!vma || addr+len <= vma->vm_start) 1360 /* remember the address as a hint for next time */ 1361 return (mm->free_area_cache = addr); 1362 1363 /* remember the largest hole we saw so far */ 1364 if (addr + mm->cached_hole_size < vma->vm_start) 1365 mm->cached_hole_size = vma->vm_start - addr; 1366 1367 /* try just below the current vma->vm_start */ 1368 addr = vma->vm_start-len; 1369 } while (len < vma->vm_start); 1370 1371bottomup: 1372 /* 1373 * A failed mmap() very likely causes application failure, 1374 * so fall back to the bottom-up function here. This scenario 1375 * can happen with large stack limits and large mmap() 1376 * allocations. 1377 */ 1378 mm->cached_hole_size = ~0UL; 1379 mm->free_area_cache = TASK_UNMAPPED_BASE; 1380 addr = arch_get_unmapped_area(filp, addr0, len, pgoff, flags); 1381 /* 1382 * Restore the topdown base: 1383 */ 1384 mm->free_area_cache = mm->mmap_base; 1385 mm->cached_hole_size = ~0UL; 1386 1387 return addr; 1388} 1389#endif 1390 1391void arch_unmap_area_topdown(struct mm_struct *mm, unsigned long addr) 1392{ 1393 /* 1394 * Is this a new hole at the highest possible address? 1395 */ 1396 if (addr > mm->free_area_cache) 1397 mm->free_area_cache = addr; 1398 1399 /* dont allow allocations above current base */ 1400 if (mm->free_area_cache > mm->mmap_base) 1401 mm->free_area_cache = mm->mmap_base; 1402} 1403 1404unsigned long 1405get_unmapped_area(struct file *file, unsigned long addr, unsigned long len, 1406 unsigned long pgoff, unsigned long flags) 1407{ 1408 unsigned long (*get_area)(struct file *, unsigned long, 1409 unsigned long, unsigned long, unsigned long); 1410 1411 get_area = current->mm->get_unmapped_area; 1412 if (file && file->f_op && file->f_op->get_unmapped_area) 1413 get_area = file->f_op->get_unmapped_area; 1414 addr = get_area(file, addr, len, pgoff, flags); 1415 if (IS_ERR_VALUE(addr)) 1416 return addr; 1417 1418 if (addr > TASK_SIZE - len) 1419 return -ENOMEM; 1420 if (addr & ~PAGE_MASK) 1421 return -EINVAL; 1422 1423 return addr; 1424} 1425 1426EXPORT_SYMBOL(get_unmapped_area); 1427 1428/* Look up the first VMA which satisfies addr < vm_end, NULL if none. */ 1429struct vm_area_struct * find_vma(struct mm_struct * mm, unsigned long addr) 1430{ 1431 struct vm_area_struct *vma = NULL; 1432 1433 if (mm) { 1434 /* Check the cache first. */ 1435 /* (Cache hit rate is typically around 35%.) */ 1436 vma = mm->mmap_cache; 1437 if (!(vma && vma->vm_end > addr && vma->vm_start <= addr)) { 1438 struct rb_node * rb_node; 1439 1440 rb_node = mm->mm_rb.rb_node; 1441 vma = NULL; 1442 1443 while (rb_node) { 1444 struct vm_area_struct * vma_tmp; 1445 1446 vma_tmp = rb_entry(rb_node, 1447 struct vm_area_struct, vm_rb); 1448 1449 if (vma_tmp->vm_end > addr) { 1450 vma = vma_tmp; 1451 if (vma_tmp->vm_start <= addr) 1452 break; 1453 rb_node = rb_node->rb_left; 1454 } else 1455 rb_node = rb_node->rb_right; 1456 } 1457 if (vma) 1458 mm->mmap_cache = vma; 1459 } 1460 } 1461 return vma; 1462} 1463 1464EXPORT_SYMBOL(find_vma); 1465 1466/* Same as find_vma, but also return a pointer to the previous VMA in *pprev. */ 1467struct vm_area_struct * 1468find_vma_prev(struct mm_struct *mm, unsigned long addr, 1469 struct vm_area_struct **pprev) 1470{ 1471 struct vm_area_struct *vma = NULL, *prev = NULL; 1472 struct rb_node * rb_node; 1473 if (!mm) 1474 goto out; 1475 1476 /* Guard against addr being lower than the first VMA */ 1477 vma = mm->mmap; 1478 1479 /* Go through the RB tree quickly. */ 1480 rb_node = mm->mm_rb.rb_node; 1481 1482 while (rb_node) { 1483 struct vm_area_struct *vma_tmp; 1484 vma_tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb); 1485 1486 if (addr < vma_tmp->vm_end) { 1487 rb_node = rb_node->rb_left; 1488 } else { 1489 prev = vma_tmp; 1490 if (!prev->vm_next || (addr < prev->vm_next->vm_end)) 1491 break; 1492 rb_node = rb_node->rb_right; 1493 } 1494 } 1495 1496out: 1497 *pprev = prev; 1498 return prev ? prev->vm_next : vma; 1499} 1500 1501/* 1502 * Verify that the stack growth is acceptable and 1503 * update accounting. This is shared with both the 1504 * grow-up and grow-down cases. 1505 */ 1506static int acct_stack_growth(struct vm_area_struct * vma, unsigned long size, unsigned long grow) 1507{ 1508 struct mm_struct *mm = vma->vm_mm; 1509 struct rlimit *rlim = current->signal->rlim; 1510 unsigned long new_start; 1511 1512 /* address space limit tests */ 1513 if (!may_expand_vm(mm, grow)) 1514 return -ENOMEM; 1515 1516 /* Stack limit test */ 1517 if (size > rlim[RLIMIT_STACK].rlim_cur) 1518 return -ENOMEM; 1519 1520 /* mlock limit tests */ 1521 if (vma->vm_flags & VM_LOCKED) { 1522 unsigned long locked; 1523 unsigned long limit; 1524 locked = mm->locked_vm + grow; 1525 limit = rlim[RLIMIT_MEMLOCK].rlim_cur >> PAGE_SHIFT; 1526 if (locked > limit && !capable(CAP_IPC_LOCK)) 1527 return -ENOMEM; 1528 } 1529 1530 /* Check to ensure the stack will not grow into a hugetlb-only region */ 1531 new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start : 1532 vma->vm_end - size; 1533 if (is_hugepage_only_range(vma->vm_mm, new_start, size)) 1534 return -EFAULT; 1535 1536 /* 1537 * Overcommit.. This must be the final test, as it will 1538 * update security statistics. 1539 */ 1540 if (security_vm_enough_memory(grow)) 1541 return -ENOMEM; 1542 1543 /* Ok, everything looks good - let it rip */ 1544 mm->total_vm += grow; 1545 if (vma->vm_flags & VM_LOCKED) 1546 mm->locked_vm += grow; 1547 vm_stat_account(mm, vma->vm_flags, vma->vm_file, grow); 1548 return 0; 1549} 1550 1551#if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64) 1552/* 1553 * PA-RISC uses this for its stack; IA64 for its Register Backing Store. 1554 * vma is the last one with address > vma->vm_end. Have to extend vma. 1555 */ 1556#ifndef CONFIG_IA64 1557static inline 1558#endif 1559int expand_upwards(struct vm_area_struct *vma, unsigned long address) 1560{ 1561 int error; 1562 1563 if (!(vma->vm_flags & VM_GROWSUP)) 1564 return -EFAULT; 1565 1566 /* 1567 * We must make sure the anon_vma is allocated 1568 * so that the anon_vma locking is not a noop. 1569 */ 1570 if (unlikely(anon_vma_prepare(vma))) 1571 return -ENOMEM; 1572 anon_vma_lock(vma); 1573 1574 /* 1575 * vma->vm_start/vm_end cannot change under us because the caller 1576 * is required to hold the mmap_sem in read mode. We need the 1577 * anon_vma lock to serialize against concurrent expand_stacks. 1578 * Also guard against wrapping around to address 0. 1579 */ 1580 if (address < PAGE_ALIGN(address+4)) 1581 address = PAGE_ALIGN(address+4); 1582 else { 1583 anon_vma_unlock(vma); 1584 return -ENOMEM; 1585 } 1586 error = 0; 1587 1588 /* Somebody else might have raced and expanded it already */ 1589 if (address > vma->vm_end) { 1590 unsigned long size, grow; 1591 1592 size = address - vma->vm_start; 1593 grow = (address - vma->vm_end) >> PAGE_SHIFT; 1594 1595 error = acct_stack_growth(vma, size, grow); 1596 if (!error) 1597 vma->vm_end = address; 1598 } 1599 anon_vma_unlock(vma); 1600 return error; 1601} 1602#endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */ 1603 1604/* 1605 * vma is the first one with address < vma->vm_start. Have to extend vma. 1606 */ 1607static inline int expand_downwards(struct vm_area_struct *vma, 1608 unsigned long address) 1609{ 1610 int error; 1611 1612 /* 1613 * We must make sure the anon_vma is allocated 1614 * so that the anon_vma locking is not a noop. 1615 */ 1616 if (unlikely(anon_vma_prepare(vma))) 1617 return -ENOMEM; 1618 anon_vma_lock(vma); 1619 1620 /* 1621 * vma->vm_start/vm_end cannot change under us because the caller 1622 * is required to hold the mmap_sem in read mode. We need the 1623 * anon_vma lock to serialize against concurrent expand_stacks. 1624 */ 1625 address &= PAGE_MASK; 1626 error = 0; 1627 1628 /* Somebody else might have raced and expanded it already */ 1629 if (address < vma->vm_start) { 1630 unsigned long size, grow; 1631 1632 size = vma->vm_end - address; 1633 grow = (vma->vm_start - address) >> PAGE_SHIFT; 1634 1635 error = acct_stack_growth(vma, size, grow); 1636 if (!error) { 1637 vma->vm_start = address; 1638 vma->vm_pgoff -= grow; 1639 } 1640 } 1641 anon_vma_unlock(vma); 1642 return error; 1643} 1644 1645int expand_stack_downwards(struct vm_area_struct *vma, unsigned long address) 1646{ 1647 return expand_downwards(vma, address); 1648} 1649 1650#ifdef CONFIG_STACK_GROWSUP 1651int expand_stack(struct vm_area_struct *vma, unsigned long address) 1652{ 1653 return expand_upwards(vma, address); 1654} 1655 1656struct vm_area_struct * 1657find_extend_vma(struct mm_struct *mm, unsigned long addr) 1658{ 1659 struct vm_area_struct *vma, *prev; 1660 1661 addr &= PAGE_MASK; 1662 vma = find_vma_prev(mm, addr, &prev); 1663 if (vma && (vma->vm_start <= addr)) 1664 return vma; 1665 if (!prev || expand_stack(prev, addr)) 1666 return NULL; 1667 if (prev->vm_flags & VM_LOCKED) 1668 make_pages_present(addr, prev->vm_end); 1669 return prev; 1670} 1671#else 1672int expand_stack(struct vm_area_struct *vma, unsigned long address) 1673{ 1674 return expand_downwards(vma, address); 1675} 1676 1677struct vm_area_struct * 1678find_extend_vma(struct mm_struct * mm, unsigned long addr) 1679{ 1680 struct vm_area_struct * vma; 1681 unsigned long start; 1682 1683 addr &= PAGE_MASK; 1684 vma = find_vma(mm,addr); 1685 if (!vma) 1686 return NULL; 1687 if (vma->vm_start <= addr) 1688 return vma; 1689 if (!(vma->vm_flags & VM_GROWSDOWN)) 1690 return NULL; 1691 start = vma->vm_start; 1692 if (expand_stack(vma, addr)) 1693 return NULL; 1694 if (vma->vm_flags & VM_LOCKED) 1695 make_pages_present(addr, start); 1696 return vma; 1697} 1698#endif 1699 1700/* 1701 * Ok - we have the memory areas we should free on the vma list, 1702 * so release them, and do the vma updates. 1703 * 1704 * Called with the mm semaphore held. 1705 */ 1706static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma) 1707{ 1708 /* Update high watermark before we lower total_vm */ 1709 update_hiwater_vm(mm); 1710 do { 1711 long nrpages = vma_pages(vma); 1712 1713 mm->total_vm -= nrpages; 1714 if (vma->vm_flags & VM_LOCKED) 1715 mm->locked_vm -= nrpages; 1716 vm_stat_account(mm, vma->vm_flags, vma->vm_file, -nrpages); 1717 vma = remove_vma(vma); 1718 } while (vma); 1719 validate_mm(mm); 1720} 1721 1722/* 1723 * Get rid of page table information in the indicated region. 1724 * 1725 * Called with the mm semaphore held. 1726 */ 1727static void unmap_region(struct mm_struct *mm, 1728 struct vm_area_struct *vma, struct vm_area_struct *prev, 1729 unsigned long start, unsigned long end) 1730{ 1731 struct vm_area_struct *next = prev? prev->vm_next: mm->mmap; 1732 struct mmu_gather *tlb; 1733 unsigned long nr_accounted = 0; 1734 1735 lru_add_drain(); 1736 tlb = tlb_gather_mmu(mm, 0); 1737 update_hiwater_rss(mm); 1738 unmap_vmas(&tlb, vma, start, end, &nr_accounted, NULL); 1739 vm_unacct_memory(nr_accounted); 1740 free_pgtables(&tlb, vma, prev? prev->vm_end: FIRST_USER_ADDRESS, 1741 next? next->vm_start: 0); 1742 tlb_finish_mmu(tlb, start, end); 1743} 1744 1745/* 1746 * Create a list of vma's touched by the unmap, removing them from the mm's 1747 * vma list as we go.. 1748 */ 1749static void 1750detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma, 1751 struct vm_area_struct *prev, unsigned long end) 1752{ 1753 struct vm_area_struct **insertion_point; 1754 struct vm_area_struct *tail_vma = NULL; 1755 unsigned long addr; 1756 1757 insertion_point = (prev ? &prev->vm_next : &mm->mmap); 1758 do { 1759 rb_erase(&vma->vm_rb, &mm->mm_rb); 1760 mm->map_count--; 1761 tail_vma = vma; 1762 vma = vma->vm_next; 1763 } while (vma && vma->vm_start < end); 1764 *insertion_point = vma; 1765 tail_vma->vm_next = NULL; 1766 if (mm->unmap_area == arch_unmap_area) 1767 addr = prev ? prev->vm_end : mm->mmap_base; 1768 else 1769 addr = vma ? vma->vm_start : mm->mmap_base; 1770 mm->unmap_area(mm, addr); 1771 mm->mmap_cache = NULL; /* Kill the cache. */ 1772} 1773 1774/* 1775 * Split a vma into two pieces at address 'addr', a new vma is allocated 1776 * either for the first part or the tail. 1777 */ 1778int split_vma(struct mm_struct * mm, struct vm_area_struct * vma, 1779 unsigned long addr, int new_below) 1780{ 1781 struct mempolicy *pol; 1782 struct vm_area_struct *new; 1783 1784 if (is_vm_hugetlb_page(vma) && (addr & ~HPAGE_MASK)) 1785 return -EINVAL; 1786 1787 if (mm->map_count >= sysctl_max_map_count) 1788 return -ENOMEM; 1789 1790 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); 1791 if (!new) 1792 return -ENOMEM; 1793 1794 /* most fields are the same, copy all, and then fixup */ 1795 *new = *vma; 1796 1797 if (new_below) 1798 new->vm_end = addr; 1799 else { 1800 new->vm_start = addr; 1801 new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT); 1802 } 1803 1804 pol = mpol_copy(vma_policy(vma)); 1805 if (IS_ERR(pol)) { 1806 kmem_cache_free(vm_area_cachep, new); 1807 return PTR_ERR(pol); 1808 } 1809 vma_set_policy(new, pol); 1810 1811 if (new->vm_file) 1812 get_file(new->vm_file); 1813 1814 if (new->vm_ops && new->vm_ops->open) 1815 new->vm_ops->open(new); 1816 1817 if (new_below) 1818 vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff + 1819 ((addr - new->vm_start) >> PAGE_SHIFT), new); 1820 else 1821 vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new); 1822 1823 return 0; 1824} 1825 1826/* Munmap is split into 2 main parts -- this part which finds 1827 * what needs doing, and the areas themselves, which do the 1828 * work. This now handles partial unmappings. 1829 * Jeremy Fitzhardinge <jeremy@goop.org> 1830 */ 1831int do_munmap(struct mm_struct *mm, unsigned long start, size_t len) 1832{ 1833 unsigned long end; 1834 struct vm_area_struct *vma, *prev, *last; 1835 1836 if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start) 1837 return -EINVAL; 1838 1839 if ((len = PAGE_ALIGN(len)) == 0) 1840 return -EINVAL; 1841 1842 /* Find the first overlapping VMA */ 1843 vma = find_vma_prev(mm, start, &prev); 1844 if (!vma) 1845 return 0; 1846 /* we have start < vma->vm_end */ 1847 1848 /* if it doesn't overlap, we have nothing.. */ 1849 end = start + len; 1850 if (vma->vm_start >= end) 1851 return 0; 1852 1853 /* 1854 * If we need to split any vma, do it now to save pain later. 1855 * 1856 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially 1857 * unmapped vm_area_struct will remain in use: so lower split_vma 1858 * places tmp vma above, and higher split_vma places tmp vma below. 1859 */ 1860 if (start > vma->vm_start) { 1861 int error = split_vma(mm, vma, start, 0); 1862 if (error) 1863 return error; 1864 prev = vma; 1865 } 1866 1867 /* Does it split the last one? */ 1868 last = find_vma(mm, end); 1869 if (last && end > last->vm_start) { 1870 int error = split_vma(mm, last, end, 1); 1871 if (error) 1872 return error; 1873 } 1874 vma = prev? prev->vm_next: mm->mmap; 1875 1876 /* 1877 * Remove the vma's, and unmap the actual pages 1878 */ 1879 detach_vmas_to_be_unmapped(mm, vma, prev, end); 1880 unmap_region(mm, vma, prev, start, end); 1881 1882 /* Fix up all other VM information */ 1883 remove_vma_list(mm, vma); 1884 1885 return 0; 1886} 1887 1888EXPORT_SYMBOL(do_munmap); 1889 1890asmlinkage long sys_munmap(unsigned long addr, size_t len) 1891{ 1892 int ret; 1893 struct mm_struct *mm = current->mm; 1894 1895 profile_munmap(addr); 1896 1897 down_write(&mm->mmap_sem); 1898 ret = do_munmap(mm, addr, len); 1899 up_write(&mm->mmap_sem); 1900 return ret; 1901} 1902 1903static inline void verify_mm_writelocked(struct mm_struct *mm) 1904{ 1905#ifdef CONFIG_DEBUG_VM 1906 if (unlikely(down_read_trylock(&mm->mmap_sem))) { 1907 WARN_ON(1); 1908 up_read(&mm->mmap_sem); 1909 } 1910#endif 1911} 1912 1913/* 1914 * this is really a simplified "do_mmap". it only handles 1915 * anonymous maps. eventually we may be able to do some 1916 * brk-specific accounting here. 1917 */ 1918unsigned long do_brk(unsigned long addr, unsigned long len) 1919{ 1920 struct mm_struct * mm = current->mm; 1921 struct vm_area_struct * vma, * prev; 1922 unsigned long flags; 1923 struct rb_node ** rb_link, * rb_parent; 1924 pgoff_t pgoff = addr >> PAGE_SHIFT; 1925 int error; 1926 1927 len = PAGE_ALIGN(len); 1928 if (!len) 1929 return addr; 1930 1931 if ((addr + len) > TASK_SIZE || (addr + len) < addr) 1932 return -EINVAL; 1933 1934 if (is_hugepage_only_range(mm, addr, len)) 1935 return -EINVAL; 1936 1937 error = security_file_mmap(0, 0, 0, 0, addr, 1); 1938 if (error) 1939 return error; 1940 1941 flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags; 1942 1943 error = arch_mmap_check(addr, len, flags); 1944 if (error) 1945 return error; 1946 1947 /* 1948 * mlock MCL_FUTURE? 1949 */ 1950 if (mm->def_flags & VM_LOCKED) { 1951 unsigned long locked, lock_limit; 1952 locked = len >> PAGE_SHIFT; 1953 locked += mm->locked_vm; 1954 lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur; 1955 lock_limit >>= PAGE_SHIFT; 1956 if (locked > lock_limit && !capable(CAP_IPC_LOCK)) 1957 return -EAGAIN; 1958 } 1959 1960 /* 1961 * mm->mmap_sem is required to protect against another thread 1962 * changing the mappings in case we sleep. 1963 */ 1964 verify_mm_writelocked(mm); 1965 1966 /* 1967 * Clear old maps. this also does some error checking for us 1968 */ 1969 munmap_back: 1970 vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent); 1971 if (vma && vma->vm_start < addr + len) { 1972 if (do_munmap(mm, addr, len)) 1973 return -ENOMEM; 1974 goto munmap_back; 1975 } 1976 1977 /* Check against address space limits *after* clearing old maps... */ 1978 if (!may_expand_vm(mm, len >> PAGE_SHIFT)) 1979 return -ENOMEM; 1980 1981 if (mm->map_count > sysctl_max_map_count) 1982 return -ENOMEM; 1983 1984 if (security_vm_enough_memory(len >> PAGE_SHIFT)) 1985 return -ENOMEM; 1986 1987 /* Can we just expand an old private anonymous mapping? */ 1988 if (vma_merge(mm, prev, addr, addr + len, flags, 1989 NULL, NULL, pgoff, NULL)) 1990 goto out; 1991 1992 /* 1993 * create a vma struct for an anonymous mapping 1994 */ 1995 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); 1996 if (!vma) { 1997 vm_unacct_memory(len >> PAGE_SHIFT); 1998 return -ENOMEM; 1999 } 2000 2001 vma->vm_mm = mm; 2002 vma->vm_start = addr; 2003 vma->vm_end = addr + len; 2004 vma->vm_pgoff = pgoff; 2005 vma->vm_flags = flags; 2006 vma->vm_page_prot = vm_get_page_prot(flags); 2007 vma_link(mm, vma, prev, rb_link, rb_parent); 2008out: 2009 mm->total_vm += len >> PAGE_SHIFT; 2010 if (flags & VM_LOCKED) { 2011 mm->locked_vm += len >> PAGE_SHIFT; 2012 make_pages_present(addr, addr + len); 2013 } 2014 return addr; 2015} 2016 2017EXPORT_SYMBOL(do_brk); 2018 2019/* Release all mmaps. */ 2020void exit_mmap(struct mm_struct *mm) 2021{ 2022 struct mmu_gather *tlb; 2023 struct vm_area_struct *vma = mm->mmap; 2024 unsigned long nr_accounted = 0; 2025 unsigned long end; 2026 2027 /* mm's last user has gone, and its about to be pulled down */ 2028 arch_exit_mmap(mm); 2029 2030 lru_add_drain(); 2031 flush_cache_mm(mm); 2032 tlb = tlb_gather_mmu(mm, 1); 2033 /* Don't update_hiwater_rss(mm) here, do_exit already did */ 2034 /* Use -1 here to ensure all VMAs in the mm are unmapped */ 2035 end = unmap_vmas(&tlb, vma, 0, -1, &nr_accounted, NULL); 2036 vm_unacct_memory(nr_accounted); 2037 free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, 0); 2038 tlb_finish_mmu(tlb, 0, end); 2039 2040 /* 2041 * Walk the list again, actually closing and freeing it, 2042 * with preemption enabled, without holding any MM locks. 2043 */ 2044 while (vma) 2045 vma = remove_vma(vma); 2046 2047 BUG_ON(mm->nr_ptes > (FIRST_USER_ADDRESS+PMD_SIZE-1)>>PMD_SHIFT); 2048} 2049 2050/* Insert vm structure into process list sorted by address 2051 * and into the inode's i_mmap tree. If vm_file is non-NULL 2052 * then i_mmap_lock is taken here. 2053 */ 2054int insert_vm_struct(struct mm_struct * mm, struct vm_area_struct * vma) 2055{ 2056 struct vm_area_struct * __vma, * prev; 2057 struct rb_node ** rb_link, * rb_parent; 2058 2059 /* 2060 * The vm_pgoff of a purely anonymous vma should be irrelevant 2061 * until its first write fault, when page's anon_vma and index 2062 * are set. But now set the vm_pgoff it will almost certainly 2063 * end up with (unless mremap moves it elsewhere before that 2064 * first wfault), so /proc/pid/maps tells a consistent story. 2065 * 2066 * By setting it to reflect the virtual start address of the 2067 * vma, merges and splits can happen in a seamless way, just 2068 * using the existing file pgoff checks and manipulations. 2069 * Similarly in do_mmap_pgoff and in do_brk. 2070 */ 2071 if (!vma->vm_file) { 2072 BUG_ON(vma->anon_vma); 2073 vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT; 2074 } 2075 __vma = find_vma_prepare(mm,vma->vm_start,&prev,&rb_link,&rb_parent); 2076 if (__vma && __vma->vm_start < vma->vm_end) 2077 return -ENOMEM; 2078 if ((vma->vm_flags & VM_ACCOUNT) && 2079 security_vm_enough_memory_mm(mm, vma_pages(vma))) 2080 return -ENOMEM; 2081 vma_link(mm, vma, prev, rb_link, rb_parent); 2082 return 0; 2083} 2084 2085/* 2086 * Copy the vma structure to a new location in the same mm, 2087 * prior to moving page table entries, to effect an mremap move. 2088 */ 2089struct vm_area_struct *copy_vma(struct vm_area_struct **vmap, 2090 unsigned long addr, unsigned long len, pgoff_t pgoff) 2091{ 2092 struct vm_area_struct *vma = *vmap; 2093 unsigned long vma_start = vma->vm_start; 2094 struct mm_struct *mm = vma->vm_mm; 2095 struct vm_area_struct *new_vma, *prev; 2096 struct rb_node **rb_link, *rb_parent; 2097 struct mempolicy *pol; 2098 2099 /* 2100 * If anonymous vma has not yet been faulted, update new pgoff 2101 * to match new location, to increase its chance of merging. 2102 */ 2103 if (!vma->vm_file && !vma->anon_vma) 2104 pgoff = addr >> PAGE_SHIFT; 2105 2106 find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent); 2107 new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags, 2108 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma)); 2109 if (new_vma) { 2110 /* 2111 * Source vma may have been merged into new_vma 2112 */ 2113 if (vma_start >= new_vma->vm_start && 2114 vma_start < new_vma->vm_end) 2115 *vmap = new_vma; 2116 } else { 2117 new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); 2118 if (new_vma) { 2119 *new_vma = *vma; 2120 pol = mpol_copy(vma_policy(vma)); 2121 if (IS_ERR(pol)) { 2122 kmem_cache_free(vm_area_cachep, new_vma); 2123 return NULL; 2124 } 2125 vma_set_policy(new_vma, pol); 2126 new_vma->vm_start = addr; 2127 new_vma->vm_end = addr + len; 2128 new_vma->vm_pgoff = pgoff; 2129 if (new_vma->vm_file) 2130 get_file(new_vma->vm_file); 2131 if (new_vma->vm_ops && new_vma->vm_ops->open) 2132 new_vma->vm_ops->open(new_vma); 2133 vma_link(mm, new_vma, prev, rb_link, rb_parent); 2134 } 2135 } 2136 return new_vma; 2137} 2138 2139/* 2140 * Return true if the calling process may expand its vm space by the passed 2141 * number of pages 2142 */ 2143int may_expand_vm(struct mm_struct *mm, unsigned long npages) 2144{ 2145 unsigned long cur = mm->total_vm; /* pages */ 2146 unsigned long lim; 2147 2148 lim = current->signal->rlim[RLIMIT_AS].rlim_cur >> PAGE_SHIFT; 2149 2150 if (cur + npages > lim) 2151 return 0; 2152 return 1; 2153} 2154 2155 2156static struct page *special_mapping_nopage(struct vm_area_struct *vma, 2157 unsigned long address, int *type) 2158{ 2159 struct page **pages; 2160 2161 BUG_ON(address < vma->vm_start || address >= vma->vm_end); 2162 2163 address -= vma->vm_start; 2164 for (pages = vma->vm_private_data; address > 0 && *pages; ++pages) 2165 address -= PAGE_SIZE; 2166 2167 if (*pages) { 2168 struct page *page = *pages; 2169 get_page(page); 2170 return page; 2171 } 2172 2173 return NOPAGE_SIGBUS; 2174} 2175 2176/* 2177 * Having a close hook prevents vma merging regardless of flags. 2178 */ 2179static void special_mapping_close(struct vm_area_struct *vma) 2180{ 2181} 2182 2183static struct vm_operations_struct special_mapping_vmops = { 2184 .close = special_mapping_close, 2185 .nopage = special_mapping_nopage, 2186}; 2187 2188/* 2189 * Called with mm->mmap_sem held for writing. 2190 * Insert a new vma covering the given region, with the given flags. 2191 * Its pages are supplied by the given array of struct page *. 2192 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated. 2193 * The region past the last page supplied will always produce SIGBUS. 2194 * The array pointer and the pages it points to are assumed to stay alive 2195 * for as long as this mapping might exist. 2196 */ 2197int install_special_mapping(struct mm_struct *mm, 2198 unsigned long addr, unsigned long len, 2199 unsigned long vm_flags, struct page **pages) 2200{ 2201 struct vm_area_struct *vma; 2202 2203 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); 2204 if (unlikely(vma == NULL)) 2205 return -ENOMEM; 2206 2207 vma->vm_mm = mm; 2208 vma->vm_start = addr; 2209 vma->vm_end = addr + len; 2210 2211 vma->vm_flags = vm_flags | mm->def_flags; 2212 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags); 2213 2214 vma->vm_ops = &special_mapping_vmops; 2215 vma->vm_private_data = pages; 2216 2217 if (unlikely(insert_vm_struct(mm, vma))) { 2218 kmem_cache_free(vm_area_cachep, vma); 2219 return -ENOMEM; 2220 } 2221 2222 mm->total_vm += len >> PAGE_SHIFT; 2223 2224 return 0; 2225} 2226