| 1 | /* |
| 2 | * linux/mm/nommu.c |
| 3 | * |
| 4 | * Replacement code for mm functions to support CPU's that don't |
| 5 | * have any form of memory management unit (thus no virtual memory). |
| 6 | * |
| 7 | * See Documentation/nommu-mmap.txt |
| 8 | * |
| 9 | * Copyright (c) 2004-2008 David Howells <dhowells@redhat.com> |
| 10 | * Copyright (c) 2000-2003 David McCullough <davidm@snapgear.com> |
| 11 | * Copyright (c) 2000-2001 D Jeff Dionne <jeff@uClinux.org> |
| 12 | * Copyright (c) 2002 Greg Ungerer <gerg@snapgear.com> |
| 13 | * Copyright (c) 2007-2010 Paul Mundt <lethal@linux-sh.org> |
| 14 | */ |
| 15 | |
| 16 | #include <linux/export.h> |
| 17 | #include <linux/mm.h> |
| 18 | #include <linux/vmacache.h> |
| 19 | #include <linux/mman.h> |
| 20 | #include <linux/swap.h> |
| 21 | #include <linux/file.h> |
| 22 | #include <linux/highmem.h> |
| 23 | #include <linux/pagemap.h> |
| 24 | #include <linux/slab.h> |
| 25 | #include <linux/vmalloc.h> |
| 26 | #include <linux/blkdev.h> |
| 27 | #include <linux/backing-dev.h> |
| 28 | #include <linux/compiler.h> |
| 29 | #include <linux/mount.h> |
| 30 | #include <linux/personality.h> |
| 31 | #include <linux/security.h> |
| 32 | #include <linux/syscalls.h> |
| 33 | #include <linux/audit.h> |
| 34 | #include <linux/sched/sysctl.h> |
| 35 | |
| 36 | #include <asm/uaccess.h> |
| 37 | #include <asm/tlb.h> |
| 38 | #include <asm/tlbflush.h> |
| 39 | #include <asm/mmu_context.h> |
| 40 | #include "internal.h" |
| 41 | |
| 42 | #if 0 |
| 43 | #define kenter(FMT, ...) \ |
| 44 | printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__) |
| 45 | #define kleave(FMT, ...) \ |
| 46 | printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__) |
| 47 | #define kdebug(FMT, ...) \ |
| 48 | printk(KERN_DEBUG "xxx" FMT"yyy\n", ##__VA_ARGS__) |
| 49 | #else |
| 50 | #define kenter(FMT, ...) \ |
| 51 | no_printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__) |
| 52 | #define kleave(FMT, ...) \ |
| 53 | no_printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__) |
| 54 | #define kdebug(FMT, ...) \ |
| 55 | no_printk(KERN_DEBUG FMT"\n", ##__VA_ARGS__) |
| 56 | #endif |
| 57 | |
| 58 | void *high_memory; |
| 59 | struct page *mem_map; |
| 60 | unsigned long max_mapnr; |
| 61 | unsigned long highest_memmap_pfn; |
| 62 | struct percpu_counter vm_committed_as; |
| 63 | int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */ |
| 64 | int sysctl_overcommit_ratio = 50; /* default is 50% */ |
| 65 | unsigned long sysctl_overcommit_kbytes __read_mostly; |
| 66 | int sysctl_max_map_count = DEFAULT_MAX_MAP_COUNT; |
| 67 | int sysctl_nr_trim_pages = CONFIG_NOMMU_INITIAL_TRIM_EXCESS; |
| 68 | unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */ |
| 69 | unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */ |
| 70 | int heap_stack_gap = 0; |
| 71 | |
| 72 | atomic_long_t mmap_pages_allocated; |
| 73 | |
| 74 | /* |
| 75 | * The global memory commitment made in the system can be a metric |
| 76 | * that can be used to drive ballooning decisions when Linux is hosted |
| 77 | * as a guest. On Hyper-V, the host implements a policy engine for dynamically |
| 78 | * balancing memory across competing virtual machines that are hosted. |
| 79 | * Several metrics drive this policy engine including the guest reported |
| 80 | * memory commitment. |
| 81 | */ |
| 82 | unsigned long vm_memory_committed(void) |
| 83 | { |
| 84 | return percpu_counter_read_positive(&vm_committed_as); |
| 85 | } |
| 86 | |
| 87 | EXPORT_SYMBOL_GPL(vm_memory_committed); |
| 88 | |
| 89 | EXPORT_SYMBOL(mem_map); |
| 90 | |
| 91 | /* list of mapped, potentially shareable regions */ |
| 92 | static struct kmem_cache *vm_region_jar; |
| 93 | struct rb_root nommu_region_tree = RB_ROOT; |
| 94 | DECLARE_RWSEM(nommu_region_sem); |
| 95 | |
| 96 | const struct vm_operations_struct generic_file_vm_ops = { |
| 97 | }; |
| 98 | |
| 99 | /* |
| 100 | * Return the total memory allocated for this pointer, not |
| 101 | * just what the caller asked for. |
| 102 | * |
| 103 | * Doesn't have to be accurate, i.e. may have races. |
| 104 | */ |
| 105 | unsigned int kobjsize(const void *objp) |
| 106 | { |
| 107 | struct page *page; |
| 108 | |
| 109 | /* |
| 110 | * If the object we have should not have ksize performed on it, |
| 111 | * return size of 0 |
| 112 | */ |
| 113 | if (!objp || !virt_addr_valid(objp)) |
| 114 | return 0; |
| 115 | |
| 116 | page = virt_to_head_page(objp); |
| 117 | |
| 118 | /* |
| 119 | * If the allocator sets PageSlab, we know the pointer came from |
| 120 | * kmalloc(). |
| 121 | */ |
| 122 | if (PageSlab(page)) |
| 123 | return ksize(objp); |
| 124 | |
| 125 | /* |
| 126 | * If it's not a compound page, see if we have a matching VMA |
| 127 | * region. This test is intentionally done in reverse order, |
| 128 | * so if there's no VMA, we still fall through and hand back |
| 129 | * PAGE_SIZE for 0-order pages. |
| 130 | */ |
| 131 | if (!PageCompound(page)) { |
| 132 | struct vm_area_struct *vma; |
| 133 | |
| 134 | vma = find_vma(current->mm, (unsigned long)objp); |
| 135 | if (vma) |
| 136 | return vma->vm_end - vma->vm_start; |
| 137 | } |
| 138 | |
| 139 | /* |
| 140 | * The ksize() function is only guaranteed to work for pointers |
| 141 | * returned by kmalloc(). So handle arbitrary pointers here. |
| 142 | */ |
| 143 | return PAGE_SIZE << compound_order(page); |
| 144 | } |
| 145 | |
| 146 | long __get_user_pages(struct task_struct *tsk, struct mm_struct *mm, |
| 147 | unsigned long start, unsigned long nr_pages, |
| 148 | unsigned int foll_flags, struct page **pages, |
| 149 | struct vm_area_struct **vmas, int *nonblocking) |
| 150 | { |
| 151 | struct vm_area_struct *vma; |
| 152 | unsigned long vm_flags; |
| 153 | int i; |
| 154 | |
| 155 | /* calculate required read or write permissions. |
| 156 | * If FOLL_FORCE is set, we only require the "MAY" flags. |
| 157 | */ |
| 158 | vm_flags = (foll_flags & FOLL_WRITE) ? |
| 159 | (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD); |
| 160 | vm_flags &= (foll_flags & FOLL_FORCE) ? |
| 161 | (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE); |
| 162 | |
| 163 | for (i = 0; i < nr_pages; i++) { |
| 164 | vma = find_vma(mm, start); |
| 165 | if (!vma) |
| 166 | goto finish_or_fault; |
| 167 | |
| 168 | /* protect what we can, including chardevs */ |
| 169 | if ((vma->vm_flags & (VM_IO | VM_PFNMAP)) || |
| 170 | !(vm_flags & vma->vm_flags)) |
| 171 | goto finish_or_fault; |
| 172 | |
| 173 | if (pages) { |
| 174 | pages[i] = virt_to_page(start); |
| 175 | if (pages[i]) |
| 176 | page_cache_get(pages[i]); |
| 177 | } |
| 178 | if (vmas) |
| 179 | vmas[i] = vma; |
| 180 | start = (start + PAGE_SIZE) & PAGE_MASK; |
| 181 | } |
| 182 | |
| 183 | return i; |
| 184 | |
| 185 | finish_or_fault: |
| 186 | return i ? : -EFAULT; |
| 187 | } |
| 188 | |
| 189 | /* |
| 190 | * get a list of pages in an address range belonging to the specified process |
| 191 | * and indicate the VMA that covers each page |
| 192 | * - this is potentially dodgy as we may end incrementing the page count of a |
| 193 | * slab page or a secondary page from a compound page |
| 194 | * - don't permit access to VMAs that don't support it, such as I/O mappings |
| 195 | */ |
| 196 | long get_user_pages(struct task_struct *tsk, struct mm_struct *mm, |
| 197 | unsigned long start, unsigned long nr_pages, |
| 198 | int write, int force, struct page **pages, |
| 199 | struct vm_area_struct **vmas) |
| 200 | { |
| 201 | int flags = 0; |
| 202 | |
| 203 | if (write) |
| 204 | flags |= FOLL_WRITE; |
| 205 | if (force) |
| 206 | flags |= FOLL_FORCE; |
| 207 | |
| 208 | return __get_user_pages(tsk, mm, start, nr_pages, flags, pages, vmas, |
| 209 | NULL); |
| 210 | } |
| 211 | EXPORT_SYMBOL(get_user_pages); |
| 212 | |
| 213 | /** |
| 214 | * follow_pfn - look up PFN at a user virtual address |
| 215 | * @vma: memory mapping |
| 216 | * @address: user virtual address |
| 217 | * @pfn: location to store found PFN |
| 218 | * |
| 219 | * Only IO mappings and raw PFN mappings are allowed. |
| 220 | * |
| 221 | * Returns zero and the pfn at @pfn on success, -ve otherwise. |
| 222 | */ |
| 223 | int follow_pfn(struct vm_area_struct *vma, unsigned long address, |
| 224 | unsigned long *pfn) |
| 225 | { |
| 226 | if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) |
| 227 | return -EINVAL; |
| 228 | |
| 229 | *pfn = address >> PAGE_SHIFT; |
| 230 | return 0; |
| 231 | } |
| 232 | EXPORT_SYMBOL(follow_pfn); |
| 233 | |
| 234 | LIST_HEAD(vmap_area_list); |
| 235 | |
| 236 | void vfree(const void *addr) |
| 237 | { |
| 238 | kfree(addr); |
| 239 | } |
| 240 | EXPORT_SYMBOL(vfree); |
| 241 | |
| 242 | void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot) |
| 243 | { |
| 244 | /* |
| 245 | * You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc() |
| 246 | * returns only a logical address. |
| 247 | */ |
| 248 | return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM); |
| 249 | } |
| 250 | EXPORT_SYMBOL(__vmalloc); |
| 251 | |
| 252 | void *vmalloc_user(unsigned long size) |
| 253 | { |
| 254 | void *ret; |
| 255 | |
| 256 | ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, |
| 257 | PAGE_KERNEL); |
| 258 | if (ret) { |
| 259 | struct vm_area_struct *vma; |
| 260 | |
| 261 | down_write(¤t->mm->mmap_sem); |
| 262 | vma = find_vma(current->mm, (unsigned long)ret); |
| 263 | if (vma) |
| 264 | vma->vm_flags |= VM_USERMAP; |
| 265 | up_write(¤t->mm->mmap_sem); |
| 266 | } |
| 267 | |
| 268 | return ret; |
| 269 | } |
| 270 | EXPORT_SYMBOL(vmalloc_user); |
| 271 | |
| 272 | struct page *vmalloc_to_page(const void *addr) |
| 273 | { |
| 274 | return virt_to_page(addr); |
| 275 | } |
| 276 | EXPORT_SYMBOL(vmalloc_to_page); |
| 277 | |
| 278 | unsigned long vmalloc_to_pfn(const void *addr) |
| 279 | { |
| 280 | return page_to_pfn(virt_to_page(addr)); |
| 281 | } |
| 282 | EXPORT_SYMBOL(vmalloc_to_pfn); |
| 283 | |
| 284 | long vread(char *buf, char *addr, unsigned long count) |
| 285 | { |
| 286 | /* Don't allow overflow */ |
| 287 | if ((unsigned long) buf + count < count) |
| 288 | count = -(unsigned long) buf; |
| 289 | |
| 290 | memcpy(buf, addr, count); |
| 291 | return count; |
| 292 | } |
| 293 | |
| 294 | long vwrite(char *buf, char *addr, unsigned long count) |
| 295 | { |
| 296 | /* Don't allow overflow */ |
| 297 | if ((unsigned long) addr + count < count) |
| 298 | count = -(unsigned long) addr; |
| 299 | |
| 300 | memcpy(addr, buf, count); |
| 301 | return(count); |
| 302 | } |
| 303 | |
| 304 | /* |
| 305 | * vmalloc - allocate virtually continguos memory |
| 306 | * |
| 307 | * @size: allocation size |
| 308 | * |
| 309 | * Allocate enough pages to cover @size from the page level |
| 310 | * allocator and map them into continguos kernel virtual space. |
| 311 | * |
| 312 | * For tight control over page level allocator and protection flags |
| 313 | * use __vmalloc() instead. |
| 314 | */ |
| 315 | void *vmalloc(unsigned long size) |
| 316 | { |
| 317 | return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL); |
| 318 | } |
| 319 | EXPORT_SYMBOL(vmalloc); |
| 320 | |
| 321 | /* |
| 322 | * vzalloc - allocate virtually continguos memory with zero fill |
| 323 | * |
| 324 | * @size: allocation size |
| 325 | * |
| 326 | * Allocate enough pages to cover @size from the page level |
| 327 | * allocator and map them into continguos kernel virtual space. |
| 328 | * The memory allocated is set to zero. |
| 329 | * |
| 330 | * For tight control over page level allocator and protection flags |
| 331 | * use __vmalloc() instead. |
| 332 | */ |
| 333 | void *vzalloc(unsigned long size) |
| 334 | { |
| 335 | return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, |
| 336 | PAGE_KERNEL); |
| 337 | } |
| 338 | EXPORT_SYMBOL(vzalloc); |
| 339 | |
| 340 | /** |
| 341 | * vmalloc_node - allocate memory on a specific node |
| 342 | * @size: allocation size |
| 343 | * @node: numa node |
| 344 | * |
| 345 | * Allocate enough pages to cover @size from the page level |
| 346 | * allocator and map them into contiguous kernel virtual space. |
| 347 | * |
| 348 | * For tight control over page level allocator and protection flags |
| 349 | * use __vmalloc() instead. |
| 350 | */ |
| 351 | void *vmalloc_node(unsigned long size, int node) |
| 352 | { |
| 353 | return vmalloc(size); |
| 354 | } |
| 355 | EXPORT_SYMBOL(vmalloc_node); |
| 356 | |
| 357 | /** |
| 358 | * vzalloc_node - allocate memory on a specific node with zero fill |
| 359 | * @size: allocation size |
| 360 | * @node: numa node |
| 361 | * |
| 362 | * Allocate enough pages to cover @size from the page level |
| 363 | * allocator and map them into contiguous kernel virtual space. |
| 364 | * The memory allocated is set to zero. |
| 365 | * |
| 366 | * For tight control over page level allocator and protection flags |
| 367 | * use __vmalloc() instead. |
| 368 | */ |
| 369 | void *vzalloc_node(unsigned long size, int node) |
| 370 | { |
| 371 | return vzalloc(size); |
| 372 | } |
| 373 | EXPORT_SYMBOL(vzalloc_node); |
| 374 | |
| 375 | #ifndef PAGE_KERNEL_EXEC |
| 376 | # define PAGE_KERNEL_EXEC PAGE_KERNEL |
| 377 | #endif |
| 378 | |
| 379 | /** |
| 380 | * vmalloc_exec - allocate virtually contiguous, executable memory |
| 381 | * @size: allocation size |
| 382 | * |
| 383 | * Kernel-internal function to allocate enough pages to cover @size |
| 384 | * the page level allocator and map them into contiguous and |
| 385 | * executable kernel virtual space. |
| 386 | * |
| 387 | * For tight control over page level allocator and protection flags |
| 388 | * use __vmalloc() instead. |
| 389 | */ |
| 390 | |
| 391 | void *vmalloc_exec(unsigned long size) |
| 392 | { |
| 393 | return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC); |
| 394 | } |
| 395 | |
| 396 | /** |
| 397 | * vmalloc_32 - allocate virtually contiguous memory (32bit addressable) |
| 398 | * @size: allocation size |
| 399 | * |
| 400 | * Allocate enough 32bit PA addressable pages to cover @size from the |
| 401 | * page level allocator and map them into continguos kernel virtual space. |
| 402 | */ |
| 403 | void *vmalloc_32(unsigned long size) |
| 404 | { |
| 405 | return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL); |
| 406 | } |
| 407 | EXPORT_SYMBOL(vmalloc_32); |
| 408 | |
| 409 | /** |
| 410 | * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory |
| 411 | * @size: allocation size |
| 412 | * |
| 413 | * The resulting memory area is 32bit addressable and zeroed so it can be |
| 414 | * mapped to userspace without leaking data. |
| 415 | * |
| 416 | * VM_USERMAP is set on the corresponding VMA so that subsequent calls to |
| 417 | * remap_vmalloc_range() are permissible. |
| 418 | */ |
| 419 | void *vmalloc_32_user(unsigned long size) |
| 420 | { |
| 421 | /* |
| 422 | * We'll have to sort out the ZONE_DMA bits for 64-bit, |
| 423 | * but for now this can simply use vmalloc_user() directly. |
| 424 | */ |
| 425 | return vmalloc_user(size); |
| 426 | } |
| 427 | EXPORT_SYMBOL(vmalloc_32_user); |
| 428 | |
| 429 | void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot) |
| 430 | { |
| 431 | BUG(); |
| 432 | return NULL; |
| 433 | } |
| 434 | EXPORT_SYMBOL(vmap); |
| 435 | |
| 436 | void vunmap(const void *addr) |
| 437 | { |
| 438 | BUG(); |
| 439 | } |
| 440 | EXPORT_SYMBOL(vunmap); |
| 441 | |
| 442 | void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot) |
| 443 | { |
| 444 | BUG(); |
| 445 | return NULL; |
| 446 | } |
| 447 | EXPORT_SYMBOL(vm_map_ram); |
| 448 | |
| 449 | void vm_unmap_ram(const void *mem, unsigned int count) |
| 450 | { |
| 451 | BUG(); |
| 452 | } |
| 453 | EXPORT_SYMBOL(vm_unmap_ram); |
| 454 | |
| 455 | void vm_unmap_aliases(void) |
| 456 | { |
| 457 | } |
| 458 | EXPORT_SYMBOL_GPL(vm_unmap_aliases); |
| 459 | |
| 460 | /* |
| 461 | * Implement a stub for vmalloc_sync_all() if the architecture chose not to |
| 462 | * have one. |
| 463 | */ |
| 464 | void __weak vmalloc_sync_all(void) |
| 465 | { |
| 466 | } |
| 467 | |
| 468 | /** |
| 469 | * alloc_vm_area - allocate a range of kernel address space |
| 470 | * @size: size of the area |
| 471 | * |
| 472 | * Returns: NULL on failure, vm_struct on success |
| 473 | * |
| 474 | * This function reserves a range of kernel address space, and |
| 475 | * allocates pagetables to map that range. No actual mappings |
| 476 | * are created. If the kernel address space is not shared |
| 477 | * between processes, it syncs the pagetable across all |
| 478 | * processes. |
| 479 | */ |
| 480 | struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes) |
| 481 | { |
| 482 | BUG(); |
| 483 | return NULL; |
| 484 | } |
| 485 | EXPORT_SYMBOL_GPL(alloc_vm_area); |
| 486 | |
| 487 | void free_vm_area(struct vm_struct *area) |
| 488 | { |
| 489 | BUG(); |
| 490 | } |
| 491 | EXPORT_SYMBOL_GPL(free_vm_area); |
| 492 | |
| 493 | int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, |
| 494 | struct page *page) |
| 495 | { |
| 496 | return -EINVAL; |
| 497 | } |
| 498 | EXPORT_SYMBOL(vm_insert_page); |
| 499 | |
| 500 | /* |
| 501 | * sys_brk() for the most part doesn't need the global kernel |
| 502 | * lock, except when an application is doing something nasty |
| 503 | * like trying to un-brk an area that has already been mapped |
| 504 | * to a regular file. in this case, the unmapping will need |
| 505 | * to invoke file system routines that need the global lock. |
| 506 | */ |
| 507 | SYSCALL_DEFINE1(brk, unsigned long, brk) |
| 508 | { |
| 509 | struct mm_struct *mm = current->mm; |
| 510 | |
| 511 | if (brk < mm->start_brk || brk > mm->context.end_brk) |
| 512 | return mm->brk; |
| 513 | |
| 514 | if (mm->brk == brk) |
| 515 | return mm->brk; |
| 516 | |
| 517 | /* |
| 518 | * Always allow shrinking brk |
| 519 | */ |
| 520 | if (brk <= mm->brk) { |
| 521 | mm->brk = brk; |
| 522 | return brk; |
| 523 | } |
| 524 | |
| 525 | /* |
| 526 | * Ok, looks good - let it rip. |
| 527 | */ |
| 528 | flush_icache_range(mm->brk, brk); |
| 529 | return mm->brk = brk; |
| 530 | } |
| 531 | |
| 532 | /* |
| 533 | * initialise the VMA and region record slabs |
| 534 | */ |
| 535 | void __init mmap_init(void) |
| 536 | { |
| 537 | int ret; |
| 538 | |
| 539 | ret = percpu_counter_init(&vm_committed_as, 0); |
| 540 | VM_BUG_ON(ret); |
| 541 | vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC); |
| 542 | } |
| 543 | |
| 544 | /* |
| 545 | * validate the region tree |
| 546 | * - the caller must hold the region lock |
| 547 | */ |
| 548 | #ifdef CONFIG_DEBUG_NOMMU_REGIONS |
| 549 | static noinline void validate_nommu_regions(void) |
| 550 | { |
| 551 | struct vm_region *region, *last; |
| 552 | struct rb_node *p, *lastp; |
| 553 | |
| 554 | lastp = rb_first(&nommu_region_tree); |
| 555 | if (!lastp) |
| 556 | return; |
| 557 | |
| 558 | last = rb_entry(lastp, struct vm_region, vm_rb); |
| 559 | BUG_ON(unlikely(last->vm_end <= last->vm_start)); |
| 560 | BUG_ON(unlikely(last->vm_top < last->vm_end)); |
| 561 | |
| 562 | while ((p = rb_next(lastp))) { |
| 563 | region = rb_entry(p, struct vm_region, vm_rb); |
| 564 | last = rb_entry(lastp, struct vm_region, vm_rb); |
| 565 | |
| 566 | BUG_ON(unlikely(region->vm_end <= region->vm_start)); |
| 567 | BUG_ON(unlikely(region->vm_top < region->vm_end)); |
| 568 | BUG_ON(unlikely(region->vm_start < last->vm_top)); |
| 569 | |
| 570 | lastp = p; |
| 571 | } |
| 572 | } |
| 573 | #else |
| 574 | static void validate_nommu_regions(void) |
| 575 | { |
| 576 | } |
| 577 | #endif |
| 578 | |
| 579 | /* |
| 580 | * add a region into the global tree |
| 581 | */ |
| 582 | static void add_nommu_region(struct vm_region *region) |
| 583 | { |
| 584 | struct vm_region *pregion; |
| 585 | struct rb_node **p, *parent; |
| 586 | |
| 587 | validate_nommu_regions(); |
| 588 | |
| 589 | parent = NULL; |
| 590 | p = &nommu_region_tree.rb_node; |
| 591 | while (*p) { |
| 592 | parent = *p; |
| 593 | pregion = rb_entry(parent, struct vm_region, vm_rb); |
| 594 | if (region->vm_start < pregion->vm_start) |
| 595 | p = &(*p)->rb_left; |
| 596 | else if (region->vm_start > pregion->vm_start) |
| 597 | p = &(*p)->rb_right; |
| 598 | else if (pregion == region) |
| 599 | return; |
| 600 | else |
| 601 | BUG(); |
| 602 | } |
| 603 | |
| 604 | rb_link_node(®ion->vm_rb, parent, p); |
| 605 | rb_insert_color(®ion->vm_rb, &nommu_region_tree); |
| 606 | |
| 607 | validate_nommu_regions(); |
| 608 | } |
| 609 | |
| 610 | /* |
| 611 | * delete a region from the global tree |
| 612 | */ |
| 613 | static void delete_nommu_region(struct vm_region *region) |
| 614 | { |
| 615 | BUG_ON(!nommu_region_tree.rb_node); |
| 616 | |
| 617 | validate_nommu_regions(); |
| 618 | rb_erase(®ion->vm_rb, &nommu_region_tree); |
| 619 | validate_nommu_regions(); |
| 620 | } |
| 621 | |
| 622 | /* |
| 623 | * free a contiguous series of pages |
| 624 | */ |
| 625 | static void free_page_series(unsigned long from, unsigned long to) |
| 626 | { |
| 627 | for (; from < to; from += PAGE_SIZE) { |
| 628 | struct page *page = virt_to_page(from); |
| 629 | |
| 630 | kdebug("- free %lx", from); |
| 631 | atomic_long_dec(&mmap_pages_allocated); |
| 632 | if (page_count(page) != 1) |
| 633 | kdebug("free page %p: refcount not one: %d", |
| 634 | page, page_count(page)); |
| 635 | put_page(page); |
| 636 | } |
| 637 | } |
| 638 | |
| 639 | /* |
| 640 | * release a reference to a region |
| 641 | * - the caller must hold the region semaphore for writing, which this releases |
| 642 | * - the region may not have been added to the tree yet, in which case vm_top |
| 643 | * will equal vm_start |
| 644 | */ |
| 645 | static void __put_nommu_region(struct vm_region *region) |
| 646 | __releases(nommu_region_sem) |
| 647 | { |
| 648 | kenter("%p{%d}", region, region->vm_usage); |
| 649 | |
| 650 | BUG_ON(!nommu_region_tree.rb_node); |
| 651 | |
| 652 | if (--region->vm_usage == 0) { |
| 653 | if (region->vm_top > region->vm_start) |
| 654 | delete_nommu_region(region); |
| 655 | up_write(&nommu_region_sem); |
| 656 | |
| 657 | if (region->vm_file) |
| 658 | fput(region->vm_file); |
| 659 | |
| 660 | /* IO memory and memory shared directly out of the pagecache |
| 661 | * from ramfs/tmpfs mustn't be released here */ |
| 662 | if (region->vm_flags & VM_MAPPED_COPY) { |
| 663 | kdebug("free series"); |
| 664 | free_page_series(region->vm_start, region->vm_top); |
| 665 | } |
| 666 | kmem_cache_free(vm_region_jar, region); |
| 667 | } else { |
| 668 | up_write(&nommu_region_sem); |
| 669 | } |
| 670 | } |
| 671 | |
| 672 | /* |
| 673 | * release a reference to a region |
| 674 | */ |
| 675 | static void put_nommu_region(struct vm_region *region) |
| 676 | { |
| 677 | down_write(&nommu_region_sem); |
| 678 | __put_nommu_region(region); |
| 679 | } |
| 680 | |
| 681 | /* |
| 682 | * update protection on a vma |
| 683 | */ |
| 684 | static void protect_vma(struct vm_area_struct *vma, unsigned long flags) |
| 685 | { |
| 686 | #ifdef CONFIG_MPU |
| 687 | struct mm_struct *mm = vma->vm_mm; |
| 688 | long start = vma->vm_start & PAGE_MASK; |
| 689 | while (start < vma->vm_end) { |
| 690 | protect_page(mm, start, flags); |
| 691 | start += PAGE_SIZE; |
| 692 | } |
| 693 | update_protections(mm); |
| 694 | #endif |
| 695 | } |
| 696 | |
| 697 | /* |
| 698 | * add a VMA into a process's mm_struct in the appropriate place in the list |
| 699 | * and tree and add to the address space's page tree also if not an anonymous |
| 700 | * page |
| 701 | * - should be called with mm->mmap_sem held writelocked |
| 702 | */ |
| 703 | static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma) |
| 704 | { |
| 705 | struct vm_area_struct *pvma, *prev; |
| 706 | struct address_space *mapping; |
| 707 | struct rb_node **p, *parent, *rb_prev; |
| 708 | |
| 709 | kenter(",%p", vma); |
| 710 | |
| 711 | BUG_ON(!vma->vm_region); |
| 712 | |
| 713 | mm->map_count++; |
| 714 | vma->vm_mm = mm; |
| 715 | |
| 716 | protect_vma(vma, vma->vm_flags); |
| 717 | |
| 718 | /* add the VMA to the mapping */ |
| 719 | if (vma->vm_file) { |
| 720 | mapping = vma->vm_file->f_mapping; |
| 721 | |
| 722 | mutex_lock(&mapping->i_mmap_mutex); |
| 723 | flush_dcache_mmap_lock(mapping); |
| 724 | vma_interval_tree_insert(vma, &mapping->i_mmap); |
| 725 | flush_dcache_mmap_unlock(mapping); |
| 726 | mutex_unlock(&mapping->i_mmap_mutex); |
| 727 | } |
| 728 | |
| 729 | /* add the VMA to the tree */ |
| 730 | parent = rb_prev = NULL; |
| 731 | p = &mm->mm_rb.rb_node; |
| 732 | while (*p) { |
| 733 | parent = *p; |
| 734 | pvma = rb_entry(parent, struct vm_area_struct, vm_rb); |
| 735 | |
| 736 | /* sort by: start addr, end addr, VMA struct addr in that order |
| 737 | * (the latter is necessary as we may get identical VMAs) */ |
| 738 | if (vma->vm_start < pvma->vm_start) |
| 739 | p = &(*p)->rb_left; |
| 740 | else if (vma->vm_start > pvma->vm_start) { |
| 741 | rb_prev = parent; |
| 742 | p = &(*p)->rb_right; |
| 743 | } else if (vma->vm_end < pvma->vm_end) |
| 744 | p = &(*p)->rb_left; |
| 745 | else if (vma->vm_end > pvma->vm_end) { |
| 746 | rb_prev = parent; |
| 747 | p = &(*p)->rb_right; |
| 748 | } else if (vma < pvma) |
| 749 | p = &(*p)->rb_left; |
| 750 | else if (vma > pvma) { |
| 751 | rb_prev = parent; |
| 752 | p = &(*p)->rb_right; |
| 753 | } else |
| 754 | BUG(); |
| 755 | } |
| 756 | |
| 757 | rb_link_node(&vma->vm_rb, parent, p); |
| 758 | rb_insert_color(&vma->vm_rb, &mm->mm_rb); |
| 759 | |
| 760 | /* add VMA to the VMA list also */ |
| 761 | prev = NULL; |
| 762 | if (rb_prev) |
| 763 | prev = rb_entry(rb_prev, struct vm_area_struct, vm_rb); |
| 764 | |
| 765 | __vma_link_list(mm, vma, prev, parent); |
| 766 | } |
| 767 | |
| 768 | /* |
| 769 | * delete a VMA from its owning mm_struct and address space |
| 770 | */ |
| 771 | static void delete_vma_from_mm(struct vm_area_struct *vma) |
| 772 | { |
| 773 | int i; |
| 774 | struct address_space *mapping; |
| 775 | struct mm_struct *mm = vma->vm_mm; |
| 776 | struct task_struct *curr = current; |
| 777 | |
| 778 | kenter("%p", vma); |
| 779 | |
| 780 | protect_vma(vma, 0); |
| 781 | |
| 782 | mm->map_count--; |
| 783 | for (i = 0; i < VMACACHE_SIZE; i++) { |
| 784 | /* if the vma is cached, invalidate the entire cache */ |
| 785 | if (curr->vmacache[i] == vma) { |
| 786 | vmacache_invalidate(curr->mm); |
| 787 | break; |
| 788 | } |
| 789 | } |
| 790 | |
| 791 | /* remove the VMA from the mapping */ |
| 792 | if (vma->vm_file) { |
| 793 | mapping = vma->vm_file->f_mapping; |
| 794 | |
| 795 | mutex_lock(&mapping->i_mmap_mutex); |
| 796 | flush_dcache_mmap_lock(mapping); |
| 797 | vma_interval_tree_remove(vma, &mapping->i_mmap); |
| 798 | flush_dcache_mmap_unlock(mapping); |
| 799 | mutex_unlock(&mapping->i_mmap_mutex); |
| 800 | } |
| 801 | |
| 802 | /* remove from the MM's tree and list */ |
| 803 | rb_erase(&vma->vm_rb, &mm->mm_rb); |
| 804 | |
| 805 | if (vma->vm_prev) |
| 806 | vma->vm_prev->vm_next = vma->vm_next; |
| 807 | else |
| 808 | mm->mmap = vma->vm_next; |
| 809 | |
| 810 | if (vma->vm_next) |
| 811 | vma->vm_next->vm_prev = vma->vm_prev; |
| 812 | } |
| 813 | |
| 814 | /* |
| 815 | * destroy a VMA record |
| 816 | */ |
| 817 | static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma) |
| 818 | { |
| 819 | kenter("%p", vma); |
| 820 | if (vma->vm_ops && vma->vm_ops->close) |
| 821 | vma->vm_ops->close(vma); |
| 822 | if (vma->vm_file) |
| 823 | fput(vma->vm_file); |
| 824 | put_nommu_region(vma->vm_region); |
| 825 | kmem_cache_free(vm_area_cachep, vma); |
| 826 | } |
| 827 | |
| 828 | /* |
| 829 | * look up the first VMA in which addr resides, NULL if none |
| 830 | * - should be called with mm->mmap_sem at least held readlocked |
| 831 | */ |
| 832 | struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr) |
| 833 | { |
| 834 | struct vm_area_struct *vma; |
| 835 | |
| 836 | /* check the cache first */ |
| 837 | vma = vmacache_find(mm, addr); |
| 838 | if (likely(vma)) |
| 839 | return vma; |
| 840 | |
| 841 | /* trawl the list (there may be multiple mappings in which addr |
| 842 | * resides) */ |
| 843 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
| 844 | if (vma->vm_start > addr) |
| 845 | return NULL; |
| 846 | if (vma->vm_end > addr) { |
| 847 | vmacache_update(addr, vma); |
| 848 | return vma; |
| 849 | } |
| 850 | } |
| 851 | |
| 852 | return NULL; |
| 853 | } |
| 854 | EXPORT_SYMBOL(find_vma); |
| 855 | |
| 856 | /* |
| 857 | * find a VMA |
| 858 | * - we don't extend stack VMAs under NOMMU conditions |
| 859 | */ |
| 860 | struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr) |
| 861 | { |
| 862 | return find_vma(mm, addr); |
| 863 | } |
| 864 | |
| 865 | /* |
| 866 | * expand a stack to a given address |
| 867 | * - not supported under NOMMU conditions |
| 868 | */ |
| 869 | int expand_stack(struct vm_area_struct *vma, unsigned long address) |
| 870 | { |
| 871 | return -ENOMEM; |
| 872 | } |
| 873 | |
| 874 | /* |
| 875 | * look up the first VMA exactly that exactly matches addr |
| 876 | * - should be called with mm->mmap_sem at least held readlocked |
| 877 | */ |
| 878 | static struct vm_area_struct *find_vma_exact(struct mm_struct *mm, |
| 879 | unsigned long addr, |
| 880 | unsigned long len) |
| 881 | { |
| 882 | struct vm_area_struct *vma; |
| 883 | unsigned long end = addr + len; |
| 884 | |
| 885 | /* check the cache first */ |
| 886 | vma = vmacache_find_exact(mm, addr, end); |
| 887 | if (vma) |
| 888 | return vma; |
| 889 | |
| 890 | /* trawl the list (there may be multiple mappings in which addr |
| 891 | * resides) */ |
| 892 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
| 893 | if (vma->vm_start < addr) |
| 894 | continue; |
| 895 | if (vma->vm_start > addr) |
| 896 | return NULL; |
| 897 | if (vma->vm_end == end) { |
| 898 | vmacache_update(addr, vma); |
| 899 | return vma; |
| 900 | } |
| 901 | } |
| 902 | |
| 903 | return NULL; |
| 904 | } |
| 905 | |
| 906 | /* |
| 907 | * determine whether a mapping should be permitted and, if so, what sort of |
| 908 | * mapping we're capable of supporting |
| 909 | */ |
| 910 | static int validate_mmap_request(struct file *file, |
| 911 | unsigned long addr, |
| 912 | unsigned long len, |
| 913 | unsigned long prot, |
| 914 | unsigned long flags, |
| 915 | unsigned long pgoff, |
| 916 | unsigned long *_capabilities) |
| 917 | { |
| 918 | unsigned long capabilities, rlen; |
| 919 | int ret; |
| 920 | |
| 921 | /* do the simple checks first */ |
| 922 | if (flags & MAP_FIXED) { |
| 923 | printk(KERN_DEBUG |
| 924 | "%d: Can't do fixed-address/overlay mmap of RAM\n", |
| 925 | current->pid); |
| 926 | return -EINVAL; |
| 927 | } |
| 928 | |
| 929 | if ((flags & MAP_TYPE) != MAP_PRIVATE && |
| 930 | (flags & MAP_TYPE) != MAP_SHARED) |
| 931 | return -EINVAL; |
| 932 | |
| 933 | if (!len) |
| 934 | return -EINVAL; |
| 935 | |
| 936 | /* Careful about overflows.. */ |
| 937 | rlen = PAGE_ALIGN(len); |
| 938 | if (!rlen || rlen > TASK_SIZE) |
| 939 | return -ENOMEM; |
| 940 | |
| 941 | /* offset overflow? */ |
| 942 | if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff) |
| 943 | return -EOVERFLOW; |
| 944 | |
| 945 | if (file) { |
| 946 | /* validate file mapping requests */ |
| 947 | struct address_space *mapping; |
| 948 | |
| 949 | /* files must support mmap */ |
| 950 | if (!file->f_op->mmap) |
| 951 | return -ENODEV; |
| 952 | |
| 953 | /* work out if what we've got could possibly be shared |
| 954 | * - we support chardevs that provide their own "memory" |
| 955 | * - we support files/blockdevs that are memory backed |
| 956 | */ |
| 957 | mapping = file->f_mapping; |
| 958 | if (!mapping) |
| 959 | mapping = file_inode(file)->i_mapping; |
| 960 | |
| 961 | capabilities = 0; |
| 962 | if (mapping && mapping->backing_dev_info) |
| 963 | capabilities = mapping->backing_dev_info->capabilities; |
| 964 | |
| 965 | if (!capabilities) { |
| 966 | /* no explicit capabilities set, so assume some |
| 967 | * defaults */ |
| 968 | switch (file_inode(file)->i_mode & S_IFMT) { |
| 969 | case S_IFREG: |
| 970 | case S_IFBLK: |
| 971 | capabilities = BDI_CAP_MAP_COPY; |
| 972 | break; |
| 973 | |
| 974 | case S_IFCHR: |
| 975 | capabilities = |
| 976 | BDI_CAP_MAP_DIRECT | |
| 977 | BDI_CAP_READ_MAP | |
| 978 | BDI_CAP_WRITE_MAP; |
| 979 | break; |
| 980 | |
| 981 | default: |
| 982 | return -EINVAL; |
| 983 | } |
| 984 | } |
| 985 | |
| 986 | /* eliminate any capabilities that we can't support on this |
| 987 | * device */ |
| 988 | if (!file->f_op->get_unmapped_area) |
| 989 | capabilities &= ~BDI_CAP_MAP_DIRECT; |
| 990 | if (!file->f_op->read) |
| 991 | capabilities &= ~BDI_CAP_MAP_COPY; |
| 992 | |
| 993 | /* The file shall have been opened with read permission. */ |
| 994 | if (!(file->f_mode & FMODE_READ)) |
| 995 | return -EACCES; |
| 996 | |
| 997 | if (flags & MAP_SHARED) { |
| 998 | /* do checks for writing, appending and locking */ |
| 999 | if ((prot & PROT_WRITE) && |
| 1000 | !(file->f_mode & FMODE_WRITE)) |
| 1001 | return -EACCES; |
| 1002 | |
| 1003 | if (IS_APPEND(file_inode(file)) && |
| 1004 | (file->f_mode & FMODE_WRITE)) |
| 1005 | return -EACCES; |
| 1006 | |
| 1007 | if (locks_verify_locked(file)) |
| 1008 | return -EAGAIN; |
| 1009 | |
| 1010 | if (!(capabilities & BDI_CAP_MAP_DIRECT)) |
| 1011 | return -ENODEV; |
| 1012 | |
| 1013 | /* we mustn't privatise shared mappings */ |
| 1014 | capabilities &= ~BDI_CAP_MAP_COPY; |
| 1015 | } |
| 1016 | else { |
| 1017 | /* we're going to read the file into private memory we |
| 1018 | * allocate */ |
| 1019 | if (!(capabilities & BDI_CAP_MAP_COPY)) |
| 1020 | return -ENODEV; |
| 1021 | |
| 1022 | /* we don't permit a private writable mapping to be |
| 1023 | * shared with the backing device */ |
| 1024 | if (prot & PROT_WRITE) |
| 1025 | capabilities &= ~BDI_CAP_MAP_DIRECT; |
| 1026 | } |
| 1027 | |
| 1028 | if (capabilities & BDI_CAP_MAP_DIRECT) { |
| 1029 | if (((prot & PROT_READ) && !(capabilities & BDI_CAP_READ_MAP)) || |
| 1030 | ((prot & PROT_WRITE) && !(capabilities & BDI_CAP_WRITE_MAP)) || |
| 1031 | ((prot & PROT_EXEC) && !(capabilities & BDI_CAP_EXEC_MAP)) |
| 1032 | ) { |
| 1033 | capabilities &= ~BDI_CAP_MAP_DIRECT; |
| 1034 | if (flags & MAP_SHARED) { |
| 1035 | printk(KERN_WARNING |
| 1036 | "MAP_SHARED not completely supported on !MMU\n"); |
| 1037 | return -EINVAL; |
| 1038 | } |
| 1039 | } |
| 1040 | } |
| 1041 | |
| 1042 | /* handle executable mappings and implied executable |
| 1043 | * mappings */ |
| 1044 | if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) { |
| 1045 | if (prot & PROT_EXEC) |
| 1046 | return -EPERM; |
| 1047 | } |
| 1048 | else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) { |
| 1049 | /* handle implication of PROT_EXEC by PROT_READ */ |
| 1050 | if (current->personality & READ_IMPLIES_EXEC) { |
| 1051 | if (capabilities & BDI_CAP_EXEC_MAP) |
| 1052 | prot |= PROT_EXEC; |
| 1053 | } |
| 1054 | } |
| 1055 | else if ((prot & PROT_READ) && |
| 1056 | (prot & PROT_EXEC) && |
| 1057 | !(capabilities & BDI_CAP_EXEC_MAP) |
| 1058 | ) { |
| 1059 | /* backing file is not executable, try to copy */ |
| 1060 | capabilities &= ~BDI_CAP_MAP_DIRECT; |
| 1061 | } |
| 1062 | } |
| 1063 | else { |
| 1064 | /* anonymous mappings are always memory backed and can be |
| 1065 | * privately mapped |
| 1066 | */ |
| 1067 | capabilities = BDI_CAP_MAP_COPY; |
| 1068 | |
| 1069 | /* handle PROT_EXEC implication by PROT_READ */ |
| 1070 | if ((prot & PROT_READ) && |
| 1071 | (current->personality & READ_IMPLIES_EXEC)) |
| 1072 | prot |= PROT_EXEC; |
| 1073 | } |
| 1074 | |
| 1075 | /* allow the security API to have its say */ |
| 1076 | ret = security_mmap_addr(addr); |
| 1077 | if (ret < 0) |
| 1078 | return ret; |
| 1079 | |
| 1080 | /* looks okay */ |
| 1081 | *_capabilities = capabilities; |
| 1082 | return 0; |
| 1083 | } |
| 1084 | |
| 1085 | /* |
| 1086 | * we've determined that we can make the mapping, now translate what we |
| 1087 | * now know into VMA flags |
| 1088 | */ |
| 1089 | static unsigned long determine_vm_flags(struct file *file, |
| 1090 | unsigned long prot, |
| 1091 | unsigned long flags, |
| 1092 | unsigned long capabilities) |
| 1093 | { |
| 1094 | unsigned long vm_flags; |
| 1095 | |
| 1096 | vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags); |
| 1097 | /* vm_flags |= mm->def_flags; */ |
| 1098 | |
| 1099 | if (!(capabilities & BDI_CAP_MAP_DIRECT)) { |
| 1100 | /* attempt to share read-only copies of mapped file chunks */ |
| 1101 | vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC; |
| 1102 | if (file && !(prot & PROT_WRITE)) |
| 1103 | vm_flags |= VM_MAYSHARE; |
| 1104 | } else { |
| 1105 | /* overlay a shareable mapping on the backing device or inode |
| 1106 | * if possible - used for chardevs, ramfs/tmpfs/shmfs and |
| 1107 | * romfs/cramfs */ |
| 1108 | vm_flags |= VM_MAYSHARE | (capabilities & BDI_CAP_VMFLAGS); |
| 1109 | if (flags & MAP_SHARED) |
| 1110 | vm_flags |= VM_SHARED; |
| 1111 | } |
| 1112 | |
| 1113 | /* refuse to let anyone share private mappings with this process if |
| 1114 | * it's being traced - otherwise breakpoints set in it may interfere |
| 1115 | * with another untraced process |
| 1116 | */ |
| 1117 | if ((flags & MAP_PRIVATE) && current->ptrace) |
| 1118 | vm_flags &= ~VM_MAYSHARE; |
| 1119 | |
| 1120 | return vm_flags; |
| 1121 | } |
| 1122 | |
| 1123 | /* |
| 1124 | * set up a shared mapping on a file (the driver or filesystem provides and |
| 1125 | * pins the storage) |
| 1126 | */ |
| 1127 | static int do_mmap_shared_file(struct vm_area_struct *vma) |
| 1128 | { |
| 1129 | int ret; |
| 1130 | |
| 1131 | ret = vma->vm_file->f_op->mmap(vma->vm_file, vma); |
| 1132 | if (ret == 0) { |
| 1133 | vma->vm_region->vm_top = vma->vm_region->vm_end; |
| 1134 | return 0; |
| 1135 | } |
| 1136 | if (ret != -ENOSYS) |
| 1137 | return ret; |
| 1138 | |
| 1139 | /* getting -ENOSYS indicates that direct mmap isn't possible (as |
| 1140 | * opposed to tried but failed) so we can only give a suitable error as |
| 1141 | * it's not possible to make a private copy if MAP_SHARED was given */ |
| 1142 | return -ENODEV; |
| 1143 | } |
| 1144 | |
| 1145 | /* |
| 1146 | * set up a private mapping or an anonymous shared mapping |
| 1147 | */ |
| 1148 | static int do_mmap_private(struct vm_area_struct *vma, |
| 1149 | struct vm_region *region, |
| 1150 | unsigned long len, |
| 1151 | unsigned long capabilities) |
| 1152 | { |
| 1153 | struct page *pages; |
| 1154 | unsigned long total, point, n; |
| 1155 | void *base; |
| 1156 | int ret, order; |
| 1157 | |
| 1158 | /* invoke the file's mapping function so that it can keep track of |
| 1159 | * shared mappings on devices or memory |
| 1160 | * - VM_MAYSHARE will be set if it may attempt to share |
| 1161 | */ |
| 1162 | if (capabilities & BDI_CAP_MAP_DIRECT) { |
| 1163 | ret = vma->vm_file->f_op->mmap(vma->vm_file, vma); |
| 1164 | if (ret == 0) { |
| 1165 | /* shouldn't return success if we're not sharing */ |
| 1166 | BUG_ON(!(vma->vm_flags & VM_MAYSHARE)); |
| 1167 | vma->vm_region->vm_top = vma->vm_region->vm_end; |
| 1168 | return 0; |
| 1169 | } |
| 1170 | if (ret != -ENOSYS) |
| 1171 | return ret; |
| 1172 | |
| 1173 | /* getting an ENOSYS error indicates that direct mmap isn't |
| 1174 | * possible (as opposed to tried but failed) so we'll try to |
| 1175 | * make a private copy of the data and map that instead */ |
| 1176 | } |
| 1177 | |
| 1178 | |
| 1179 | /* allocate some memory to hold the mapping |
| 1180 | * - note that this may not return a page-aligned address if the object |
| 1181 | * we're allocating is smaller than a page |
| 1182 | */ |
| 1183 | order = get_order(len); |
| 1184 | kdebug("alloc order %d for %lx", order, len); |
| 1185 | |
| 1186 | pages = alloc_pages(GFP_KERNEL, order); |
| 1187 | if (!pages) |
| 1188 | goto enomem; |
| 1189 | |
| 1190 | total = 1 << order; |
| 1191 | atomic_long_add(total, &mmap_pages_allocated); |
| 1192 | |
| 1193 | point = len >> PAGE_SHIFT; |
| 1194 | |
| 1195 | /* we allocated a power-of-2 sized page set, so we may want to trim off |
| 1196 | * the excess */ |
| 1197 | if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages) { |
| 1198 | while (total > point) { |
| 1199 | order = ilog2(total - point); |
| 1200 | n = 1 << order; |
| 1201 | kdebug("shave %lu/%lu @%lu", n, total - point, total); |
| 1202 | atomic_long_sub(n, &mmap_pages_allocated); |
| 1203 | total -= n; |
| 1204 | set_page_refcounted(pages + total); |
| 1205 | __free_pages(pages + total, order); |
| 1206 | } |
| 1207 | } |
| 1208 | |
| 1209 | for (point = 1; point < total; point++) |
| 1210 | set_page_refcounted(&pages[point]); |
| 1211 | |
| 1212 | base = page_address(pages); |
| 1213 | region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY; |
| 1214 | region->vm_start = (unsigned long) base; |
| 1215 | region->vm_end = region->vm_start + len; |
| 1216 | region->vm_top = region->vm_start + (total << PAGE_SHIFT); |
| 1217 | |
| 1218 | vma->vm_start = region->vm_start; |
| 1219 | vma->vm_end = region->vm_start + len; |
| 1220 | |
| 1221 | if (vma->vm_file) { |
| 1222 | /* read the contents of a file into the copy */ |
| 1223 | mm_segment_t old_fs; |
| 1224 | loff_t fpos; |
| 1225 | |
| 1226 | fpos = vma->vm_pgoff; |
| 1227 | fpos <<= PAGE_SHIFT; |
| 1228 | |
| 1229 | old_fs = get_fs(); |
| 1230 | set_fs(KERNEL_DS); |
| 1231 | ret = vma->vm_file->f_op->read(vma->vm_file, base, len, &fpos); |
| 1232 | set_fs(old_fs); |
| 1233 | |
| 1234 | if (ret < 0) |
| 1235 | goto error_free; |
| 1236 | |
| 1237 | /* clear the last little bit */ |
| 1238 | if (ret < len) |
| 1239 | memset(base + ret, 0, len - ret); |
| 1240 | |
| 1241 | } |
| 1242 | |
| 1243 | return 0; |
| 1244 | |
| 1245 | error_free: |
| 1246 | free_page_series(region->vm_start, region->vm_top); |
| 1247 | region->vm_start = vma->vm_start = 0; |
| 1248 | region->vm_end = vma->vm_end = 0; |
| 1249 | region->vm_top = 0; |
| 1250 | return ret; |
| 1251 | |
| 1252 | enomem: |
| 1253 | printk("Allocation of length %lu from process %d (%s) failed\n", |
| 1254 | len, current->pid, current->comm); |
| 1255 | show_free_areas(0); |
| 1256 | return -ENOMEM; |
| 1257 | } |
| 1258 | |
| 1259 | /* |
| 1260 | * handle mapping creation for uClinux |
| 1261 | */ |
| 1262 | unsigned long do_mmap_pgoff(struct file *file, |
| 1263 | unsigned long addr, |
| 1264 | unsigned long len, |
| 1265 | unsigned long prot, |
| 1266 | unsigned long flags, |
| 1267 | unsigned long pgoff, |
| 1268 | unsigned long *populate) |
| 1269 | { |
| 1270 | struct vm_area_struct *vma; |
| 1271 | struct vm_region *region; |
| 1272 | struct rb_node *rb; |
| 1273 | unsigned long capabilities, vm_flags, result; |
| 1274 | int ret; |
| 1275 | |
| 1276 | kenter(",%lx,%lx,%lx,%lx,%lx", addr, len, prot, flags, pgoff); |
| 1277 | |
| 1278 | *populate = 0; |
| 1279 | |
| 1280 | /* decide whether we should attempt the mapping, and if so what sort of |
| 1281 | * mapping */ |
| 1282 | ret = validate_mmap_request(file, addr, len, prot, flags, pgoff, |
| 1283 | &capabilities); |
| 1284 | if (ret < 0) { |
| 1285 | kleave(" = %d [val]", ret); |
| 1286 | return ret; |
| 1287 | } |
| 1288 | |
| 1289 | /* we ignore the address hint */ |
| 1290 | addr = 0; |
| 1291 | len = PAGE_ALIGN(len); |
| 1292 | |
| 1293 | /* we've determined that we can make the mapping, now translate what we |
| 1294 | * now know into VMA flags */ |
| 1295 | vm_flags = determine_vm_flags(file, prot, flags, capabilities); |
| 1296 | |
| 1297 | /* we're going to need to record the mapping */ |
| 1298 | region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL); |
| 1299 | if (!region) |
| 1300 | goto error_getting_region; |
| 1301 | |
| 1302 | vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); |
| 1303 | if (!vma) |
| 1304 | goto error_getting_vma; |
| 1305 | |
| 1306 | region->vm_usage = 1; |
| 1307 | region->vm_flags = vm_flags; |
| 1308 | region->vm_pgoff = pgoff; |
| 1309 | |
| 1310 | INIT_LIST_HEAD(&vma->anon_vma_chain); |
| 1311 | vma->vm_flags = vm_flags; |
| 1312 | vma->vm_pgoff = pgoff; |
| 1313 | |
| 1314 | if (file) { |
| 1315 | region->vm_file = get_file(file); |
| 1316 | vma->vm_file = get_file(file); |
| 1317 | } |
| 1318 | |
| 1319 | down_write(&nommu_region_sem); |
| 1320 | |
| 1321 | /* if we want to share, we need to check for regions created by other |
| 1322 | * mmap() calls that overlap with our proposed mapping |
| 1323 | * - we can only share with a superset match on most regular files |
| 1324 | * - shared mappings on character devices and memory backed files are |
| 1325 | * permitted to overlap inexactly as far as we are concerned for in |
| 1326 | * these cases, sharing is handled in the driver or filesystem rather |
| 1327 | * than here |
| 1328 | */ |
| 1329 | if (vm_flags & VM_MAYSHARE) { |
| 1330 | struct vm_region *pregion; |
| 1331 | unsigned long pglen, rpglen, pgend, rpgend, start; |
| 1332 | |
| 1333 | pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; |
| 1334 | pgend = pgoff + pglen; |
| 1335 | |
| 1336 | for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) { |
| 1337 | pregion = rb_entry(rb, struct vm_region, vm_rb); |
| 1338 | |
| 1339 | if (!(pregion->vm_flags & VM_MAYSHARE)) |
| 1340 | continue; |
| 1341 | |
| 1342 | /* search for overlapping mappings on the same file */ |
| 1343 | if (file_inode(pregion->vm_file) != |
| 1344 | file_inode(file)) |
| 1345 | continue; |
| 1346 | |
| 1347 | if (pregion->vm_pgoff >= pgend) |
| 1348 | continue; |
| 1349 | |
| 1350 | rpglen = pregion->vm_end - pregion->vm_start; |
| 1351 | rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT; |
| 1352 | rpgend = pregion->vm_pgoff + rpglen; |
| 1353 | if (pgoff >= rpgend) |
| 1354 | continue; |
| 1355 | |
| 1356 | /* handle inexactly overlapping matches between |
| 1357 | * mappings */ |
| 1358 | if ((pregion->vm_pgoff != pgoff || rpglen != pglen) && |
| 1359 | !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) { |
| 1360 | /* new mapping is not a subset of the region */ |
| 1361 | if (!(capabilities & BDI_CAP_MAP_DIRECT)) |
| 1362 | goto sharing_violation; |
| 1363 | continue; |
| 1364 | } |
| 1365 | |
| 1366 | /* we've found a region we can share */ |
| 1367 | pregion->vm_usage++; |
| 1368 | vma->vm_region = pregion; |
| 1369 | start = pregion->vm_start; |
| 1370 | start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT; |
| 1371 | vma->vm_start = start; |
| 1372 | vma->vm_end = start + len; |
| 1373 | |
| 1374 | if (pregion->vm_flags & VM_MAPPED_COPY) { |
| 1375 | kdebug("share copy"); |
| 1376 | vma->vm_flags |= VM_MAPPED_COPY; |
| 1377 | } else { |
| 1378 | kdebug("share mmap"); |
| 1379 | ret = do_mmap_shared_file(vma); |
| 1380 | if (ret < 0) { |
| 1381 | vma->vm_region = NULL; |
| 1382 | vma->vm_start = 0; |
| 1383 | vma->vm_end = 0; |
| 1384 | pregion->vm_usage--; |
| 1385 | pregion = NULL; |
| 1386 | goto error_just_free; |
| 1387 | } |
| 1388 | } |
| 1389 | fput(region->vm_file); |
| 1390 | kmem_cache_free(vm_region_jar, region); |
| 1391 | region = pregion; |
| 1392 | result = start; |
| 1393 | goto share; |
| 1394 | } |
| 1395 | |
| 1396 | /* obtain the address at which to make a shared mapping |
| 1397 | * - this is the hook for quasi-memory character devices to |
| 1398 | * tell us the location of a shared mapping |
| 1399 | */ |
| 1400 | if (capabilities & BDI_CAP_MAP_DIRECT) { |
| 1401 | addr = file->f_op->get_unmapped_area(file, addr, len, |
| 1402 | pgoff, flags); |
| 1403 | if (IS_ERR_VALUE(addr)) { |
| 1404 | ret = addr; |
| 1405 | if (ret != -ENOSYS) |
| 1406 | goto error_just_free; |
| 1407 | |
| 1408 | /* the driver refused to tell us where to site |
| 1409 | * the mapping so we'll have to attempt to copy |
| 1410 | * it */ |
| 1411 | ret = -ENODEV; |
| 1412 | if (!(capabilities & BDI_CAP_MAP_COPY)) |
| 1413 | goto error_just_free; |
| 1414 | |
| 1415 | capabilities &= ~BDI_CAP_MAP_DIRECT; |
| 1416 | } else { |
| 1417 | vma->vm_start = region->vm_start = addr; |
| 1418 | vma->vm_end = region->vm_end = addr + len; |
| 1419 | } |
| 1420 | } |
| 1421 | } |
| 1422 | |
| 1423 | vma->vm_region = region; |
| 1424 | |
| 1425 | /* set up the mapping |
| 1426 | * - the region is filled in if BDI_CAP_MAP_DIRECT is still set |
| 1427 | */ |
| 1428 | if (file && vma->vm_flags & VM_SHARED) |
| 1429 | ret = do_mmap_shared_file(vma); |
| 1430 | else |
| 1431 | ret = do_mmap_private(vma, region, len, capabilities); |
| 1432 | if (ret < 0) |
| 1433 | goto error_just_free; |
| 1434 | add_nommu_region(region); |
| 1435 | |
| 1436 | /* clear anonymous mappings that don't ask for uninitialized data */ |
| 1437 | if (!vma->vm_file && !(flags & MAP_UNINITIALIZED)) |
| 1438 | memset((void *)region->vm_start, 0, |
| 1439 | region->vm_end - region->vm_start); |
| 1440 | |
| 1441 | /* okay... we have a mapping; now we have to register it */ |
| 1442 | result = vma->vm_start; |
| 1443 | |
| 1444 | current->mm->total_vm += len >> PAGE_SHIFT; |
| 1445 | |
| 1446 | share: |
| 1447 | add_vma_to_mm(current->mm, vma); |
| 1448 | |
| 1449 | /* we flush the region from the icache only when the first executable |
| 1450 | * mapping of it is made */ |
| 1451 | if (vma->vm_flags & VM_EXEC && !region->vm_icache_flushed) { |
| 1452 | flush_icache_range(region->vm_start, region->vm_end); |
| 1453 | region->vm_icache_flushed = true; |
| 1454 | } |
| 1455 | |
| 1456 | up_write(&nommu_region_sem); |
| 1457 | |
| 1458 | kleave(" = %lx", result); |
| 1459 | return result; |
| 1460 | |
| 1461 | error_just_free: |
| 1462 | up_write(&nommu_region_sem); |
| 1463 | error: |
| 1464 | if (region->vm_file) |
| 1465 | fput(region->vm_file); |
| 1466 | kmem_cache_free(vm_region_jar, region); |
| 1467 | if (vma->vm_file) |
| 1468 | fput(vma->vm_file); |
| 1469 | kmem_cache_free(vm_area_cachep, vma); |
| 1470 | kleave(" = %d", ret); |
| 1471 | return ret; |
| 1472 | |
| 1473 | sharing_violation: |
| 1474 | up_write(&nommu_region_sem); |
| 1475 | printk(KERN_WARNING "Attempt to share mismatched mappings\n"); |
| 1476 | ret = -EINVAL; |
| 1477 | goto error; |
| 1478 | |
| 1479 | error_getting_vma: |
| 1480 | kmem_cache_free(vm_region_jar, region); |
| 1481 | printk(KERN_WARNING "Allocation of vma for %lu byte allocation" |
| 1482 | " from process %d failed\n", |
| 1483 | len, current->pid); |
| 1484 | show_free_areas(0); |
| 1485 | return -ENOMEM; |
| 1486 | |
| 1487 | error_getting_region: |
| 1488 | printk(KERN_WARNING "Allocation of vm region for %lu byte allocation" |
| 1489 | " from process %d failed\n", |
| 1490 | len, current->pid); |
| 1491 | show_free_areas(0); |
| 1492 | return -ENOMEM; |
| 1493 | } |
| 1494 | |
| 1495 | SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len, |
| 1496 | unsigned long, prot, unsigned long, flags, |
| 1497 | unsigned long, fd, unsigned long, pgoff) |
| 1498 | { |
| 1499 | struct file *file = NULL; |
| 1500 | unsigned long retval = -EBADF; |
| 1501 | |
| 1502 | audit_mmap_fd(fd, flags); |
| 1503 | if (!(flags & MAP_ANONYMOUS)) { |
| 1504 | file = fget(fd); |
| 1505 | if (!file) |
| 1506 | goto out; |
| 1507 | } |
| 1508 | |
| 1509 | flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE); |
| 1510 | |
| 1511 | retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff); |
| 1512 | |
| 1513 | if (file) |
| 1514 | fput(file); |
| 1515 | out: |
| 1516 | return retval; |
| 1517 | } |
| 1518 | |
| 1519 | #ifdef __ARCH_WANT_SYS_OLD_MMAP |
| 1520 | struct mmap_arg_struct { |
| 1521 | unsigned long addr; |
| 1522 | unsigned long len; |
| 1523 | unsigned long prot; |
| 1524 | unsigned long flags; |
| 1525 | unsigned long fd; |
| 1526 | unsigned long offset; |
| 1527 | }; |
| 1528 | |
| 1529 | SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg) |
| 1530 | { |
| 1531 | struct mmap_arg_struct a; |
| 1532 | |
| 1533 | if (copy_from_user(&a, arg, sizeof(a))) |
| 1534 | return -EFAULT; |
| 1535 | if (a.offset & ~PAGE_MASK) |
| 1536 | return -EINVAL; |
| 1537 | |
| 1538 | return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd, |
| 1539 | a.offset >> PAGE_SHIFT); |
| 1540 | } |
| 1541 | #endif /* __ARCH_WANT_SYS_OLD_MMAP */ |
| 1542 | |
| 1543 | /* |
| 1544 | * split a vma into two pieces at address 'addr', a new vma is allocated either |
| 1545 | * for the first part or the tail. |
| 1546 | */ |
| 1547 | int split_vma(struct mm_struct *mm, struct vm_area_struct *vma, |
| 1548 | unsigned long addr, int new_below) |
| 1549 | { |
| 1550 | struct vm_area_struct *new; |
| 1551 | struct vm_region *region; |
| 1552 | unsigned long npages; |
| 1553 | |
| 1554 | kenter(""); |
| 1555 | |
| 1556 | /* we're only permitted to split anonymous regions (these should have |
| 1557 | * only a single usage on the region) */ |
| 1558 | if (vma->vm_file) |
| 1559 | return -ENOMEM; |
| 1560 | |
| 1561 | if (mm->map_count >= sysctl_max_map_count) |
| 1562 | return -ENOMEM; |
| 1563 | |
| 1564 | region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL); |
| 1565 | if (!region) |
| 1566 | return -ENOMEM; |
| 1567 | |
| 1568 | new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); |
| 1569 | if (!new) { |
| 1570 | kmem_cache_free(vm_region_jar, region); |
| 1571 | return -ENOMEM; |
| 1572 | } |
| 1573 | |
| 1574 | /* most fields are the same, copy all, and then fixup */ |
| 1575 | *new = *vma; |
| 1576 | *region = *vma->vm_region; |
| 1577 | new->vm_region = region; |
| 1578 | |
| 1579 | npages = (addr - vma->vm_start) >> PAGE_SHIFT; |
| 1580 | |
| 1581 | if (new_below) { |
| 1582 | region->vm_top = region->vm_end = new->vm_end = addr; |
| 1583 | } else { |
| 1584 | region->vm_start = new->vm_start = addr; |
| 1585 | region->vm_pgoff = new->vm_pgoff += npages; |
| 1586 | } |
| 1587 | |
| 1588 | if (new->vm_ops && new->vm_ops->open) |
| 1589 | new->vm_ops->open(new); |
| 1590 | |
| 1591 | delete_vma_from_mm(vma); |
| 1592 | down_write(&nommu_region_sem); |
| 1593 | delete_nommu_region(vma->vm_region); |
| 1594 | if (new_below) { |
| 1595 | vma->vm_region->vm_start = vma->vm_start = addr; |
| 1596 | vma->vm_region->vm_pgoff = vma->vm_pgoff += npages; |
| 1597 | } else { |
| 1598 | vma->vm_region->vm_end = vma->vm_end = addr; |
| 1599 | vma->vm_region->vm_top = addr; |
| 1600 | } |
| 1601 | add_nommu_region(vma->vm_region); |
| 1602 | add_nommu_region(new->vm_region); |
| 1603 | up_write(&nommu_region_sem); |
| 1604 | add_vma_to_mm(mm, vma); |
| 1605 | add_vma_to_mm(mm, new); |
| 1606 | return 0; |
| 1607 | } |
| 1608 | |
| 1609 | /* |
| 1610 | * shrink a VMA by removing the specified chunk from either the beginning or |
| 1611 | * the end |
| 1612 | */ |
| 1613 | static int shrink_vma(struct mm_struct *mm, |
| 1614 | struct vm_area_struct *vma, |
| 1615 | unsigned long from, unsigned long to) |
| 1616 | { |
| 1617 | struct vm_region *region; |
| 1618 | |
| 1619 | kenter(""); |
| 1620 | |
| 1621 | /* adjust the VMA's pointers, which may reposition it in the MM's tree |
| 1622 | * and list */ |
| 1623 | delete_vma_from_mm(vma); |
| 1624 | if (from > vma->vm_start) |
| 1625 | vma->vm_end = from; |
| 1626 | else |
| 1627 | vma->vm_start = to; |
| 1628 | add_vma_to_mm(mm, vma); |
| 1629 | |
| 1630 | /* cut the backing region down to size */ |
| 1631 | region = vma->vm_region; |
| 1632 | BUG_ON(region->vm_usage != 1); |
| 1633 | |
| 1634 | down_write(&nommu_region_sem); |
| 1635 | delete_nommu_region(region); |
| 1636 | if (from > region->vm_start) { |
| 1637 | to = region->vm_top; |
| 1638 | region->vm_top = region->vm_end = from; |
| 1639 | } else { |
| 1640 | region->vm_start = to; |
| 1641 | } |
| 1642 | add_nommu_region(region); |
| 1643 | up_write(&nommu_region_sem); |
| 1644 | |
| 1645 | free_page_series(from, to); |
| 1646 | return 0; |
| 1647 | } |
| 1648 | |
| 1649 | /* |
| 1650 | * release a mapping |
| 1651 | * - under NOMMU conditions the chunk to be unmapped must be backed by a single |
| 1652 | * VMA, though it need not cover the whole VMA |
| 1653 | */ |
| 1654 | int do_munmap(struct mm_struct *mm, unsigned long start, size_t len) |
| 1655 | { |
| 1656 | struct vm_area_struct *vma; |
| 1657 | unsigned long end; |
| 1658 | int ret; |
| 1659 | |
| 1660 | kenter(",%lx,%zx", start, len); |
| 1661 | |
| 1662 | len = PAGE_ALIGN(len); |
| 1663 | if (len == 0) |
| 1664 | return -EINVAL; |
| 1665 | |
| 1666 | end = start + len; |
| 1667 | |
| 1668 | /* find the first potentially overlapping VMA */ |
| 1669 | vma = find_vma(mm, start); |
| 1670 | if (!vma) { |
| 1671 | static int limit = 0; |
| 1672 | if (limit < 5) { |
| 1673 | printk(KERN_WARNING |
| 1674 | "munmap of memory not mmapped by process %d" |
| 1675 | " (%s): 0x%lx-0x%lx\n", |
| 1676 | current->pid, current->comm, |
| 1677 | start, start + len - 1); |
| 1678 | limit++; |
| 1679 | } |
| 1680 | return -EINVAL; |
| 1681 | } |
| 1682 | |
| 1683 | /* we're allowed to split an anonymous VMA but not a file-backed one */ |
| 1684 | if (vma->vm_file) { |
| 1685 | do { |
| 1686 | if (start > vma->vm_start) { |
| 1687 | kleave(" = -EINVAL [miss]"); |
| 1688 | return -EINVAL; |
| 1689 | } |
| 1690 | if (end == vma->vm_end) |
| 1691 | goto erase_whole_vma; |
| 1692 | vma = vma->vm_next; |
| 1693 | } while (vma); |
| 1694 | kleave(" = -EINVAL [split file]"); |
| 1695 | return -EINVAL; |
| 1696 | } else { |
| 1697 | /* the chunk must be a subset of the VMA found */ |
| 1698 | if (start == vma->vm_start && end == vma->vm_end) |
| 1699 | goto erase_whole_vma; |
| 1700 | if (start < vma->vm_start || end > vma->vm_end) { |
| 1701 | kleave(" = -EINVAL [superset]"); |
| 1702 | return -EINVAL; |
| 1703 | } |
| 1704 | if (start & ~PAGE_MASK) { |
| 1705 | kleave(" = -EINVAL [unaligned start]"); |
| 1706 | return -EINVAL; |
| 1707 | } |
| 1708 | if (end != vma->vm_end && end & ~PAGE_MASK) { |
| 1709 | kleave(" = -EINVAL [unaligned split]"); |
| 1710 | return -EINVAL; |
| 1711 | } |
| 1712 | if (start != vma->vm_start && end != vma->vm_end) { |
| 1713 | ret = split_vma(mm, vma, start, 1); |
| 1714 | if (ret < 0) { |
| 1715 | kleave(" = %d [split]", ret); |
| 1716 | return ret; |
| 1717 | } |
| 1718 | } |
| 1719 | return shrink_vma(mm, vma, start, end); |
| 1720 | } |
| 1721 | |
| 1722 | erase_whole_vma: |
| 1723 | delete_vma_from_mm(vma); |
| 1724 | delete_vma(mm, vma); |
| 1725 | kleave(" = 0"); |
| 1726 | return 0; |
| 1727 | } |
| 1728 | EXPORT_SYMBOL(do_munmap); |
| 1729 | |
| 1730 | int vm_munmap(unsigned long addr, size_t len) |
| 1731 | { |
| 1732 | struct mm_struct *mm = current->mm; |
| 1733 | int ret; |
| 1734 | |
| 1735 | down_write(&mm->mmap_sem); |
| 1736 | ret = do_munmap(mm, addr, len); |
| 1737 | up_write(&mm->mmap_sem); |
| 1738 | return ret; |
| 1739 | } |
| 1740 | EXPORT_SYMBOL(vm_munmap); |
| 1741 | |
| 1742 | SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len) |
| 1743 | { |
| 1744 | return vm_munmap(addr, len); |
| 1745 | } |
| 1746 | |
| 1747 | /* |
| 1748 | * release all the mappings made in a process's VM space |
| 1749 | */ |
| 1750 | void exit_mmap(struct mm_struct *mm) |
| 1751 | { |
| 1752 | struct vm_area_struct *vma; |
| 1753 | |
| 1754 | if (!mm) |
| 1755 | return; |
| 1756 | |
| 1757 | kenter(""); |
| 1758 | |
| 1759 | mm->total_vm = 0; |
| 1760 | |
| 1761 | while ((vma = mm->mmap)) { |
| 1762 | mm->mmap = vma->vm_next; |
| 1763 | delete_vma_from_mm(vma); |
| 1764 | delete_vma(mm, vma); |
| 1765 | cond_resched(); |
| 1766 | } |
| 1767 | |
| 1768 | kleave(""); |
| 1769 | } |
| 1770 | |
| 1771 | unsigned long vm_brk(unsigned long addr, unsigned long len) |
| 1772 | { |
| 1773 | return -ENOMEM; |
| 1774 | } |
| 1775 | |
| 1776 | /* |
| 1777 | * expand (or shrink) an existing mapping, potentially moving it at the same |
| 1778 | * time (controlled by the MREMAP_MAYMOVE flag and available VM space) |
| 1779 | * |
| 1780 | * under NOMMU conditions, we only permit changing a mapping's size, and only |
| 1781 | * as long as it stays within the region allocated by do_mmap_private() and the |
| 1782 | * block is not shareable |
| 1783 | * |
| 1784 | * MREMAP_FIXED is not supported under NOMMU conditions |
| 1785 | */ |
| 1786 | static unsigned long do_mremap(unsigned long addr, |
| 1787 | unsigned long old_len, unsigned long new_len, |
| 1788 | unsigned long flags, unsigned long new_addr) |
| 1789 | { |
| 1790 | struct vm_area_struct *vma; |
| 1791 | |
| 1792 | /* insanity checks first */ |
| 1793 | old_len = PAGE_ALIGN(old_len); |
| 1794 | new_len = PAGE_ALIGN(new_len); |
| 1795 | if (old_len == 0 || new_len == 0) |
| 1796 | return (unsigned long) -EINVAL; |
| 1797 | |
| 1798 | if (addr & ~PAGE_MASK) |
| 1799 | return -EINVAL; |
| 1800 | |
| 1801 | if (flags & MREMAP_FIXED && new_addr != addr) |
| 1802 | return (unsigned long) -EINVAL; |
| 1803 | |
| 1804 | vma = find_vma_exact(current->mm, addr, old_len); |
| 1805 | if (!vma) |
| 1806 | return (unsigned long) -EINVAL; |
| 1807 | |
| 1808 | if (vma->vm_end != vma->vm_start + old_len) |
| 1809 | return (unsigned long) -EFAULT; |
| 1810 | |
| 1811 | if (vma->vm_flags & VM_MAYSHARE) |
| 1812 | return (unsigned long) -EPERM; |
| 1813 | |
| 1814 | if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start) |
| 1815 | return (unsigned long) -ENOMEM; |
| 1816 | |
| 1817 | /* all checks complete - do it */ |
| 1818 | vma->vm_end = vma->vm_start + new_len; |
| 1819 | return vma->vm_start; |
| 1820 | } |
| 1821 | |
| 1822 | SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len, |
| 1823 | unsigned long, new_len, unsigned long, flags, |
| 1824 | unsigned long, new_addr) |
| 1825 | { |
| 1826 | unsigned long ret; |
| 1827 | |
| 1828 | down_write(¤t->mm->mmap_sem); |
| 1829 | ret = do_mremap(addr, old_len, new_len, flags, new_addr); |
| 1830 | up_write(¤t->mm->mmap_sem); |
| 1831 | return ret; |
| 1832 | } |
| 1833 | |
| 1834 | struct page *follow_page_mask(struct vm_area_struct *vma, |
| 1835 | unsigned long address, unsigned int flags, |
| 1836 | unsigned int *page_mask) |
| 1837 | { |
| 1838 | *page_mask = 0; |
| 1839 | return NULL; |
| 1840 | } |
| 1841 | |
| 1842 | int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr, |
| 1843 | unsigned long pfn, unsigned long size, pgprot_t prot) |
| 1844 | { |
| 1845 | if (addr != (pfn << PAGE_SHIFT)) |
| 1846 | return -EINVAL; |
| 1847 | |
| 1848 | vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP; |
| 1849 | return 0; |
| 1850 | } |
| 1851 | EXPORT_SYMBOL(remap_pfn_range); |
| 1852 | |
| 1853 | int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len) |
| 1854 | { |
| 1855 | unsigned long pfn = start >> PAGE_SHIFT; |
| 1856 | unsigned long vm_len = vma->vm_end - vma->vm_start; |
| 1857 | |
| 1858 | pfn += vma->vm_pgoff; |
| 1859 | return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot); |
| 1860 | } |
| 1861 | EXPORT_SYMBOL(vm_iomap_memory); |
| 1862 | |
| 1863 | int remap_vmalloc_range(struct vm_area_struct *vma, void *addr, |
| 1864 | unsigned long pgoff) |
| 1865 | { |
| 1866 | unsigned int size = vma->vm_end - vma->vm_start; |
| 1867 | |
| 1868 | if (!(vma->vm_flags & VM_USERMAP)) |
| 1869 | return -EINVAL; |
| 1870 | |
| 1871 | vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT)); |
| 1872 | vma->vm_end = vma->vm_start + size; |
| 1873 | |
| 1874 | return 0; |
| 1875 | } |
| 1876 | EXPORT_SYMBOL(remap_vmalloc_range); |
| 1877 | |
| 1878 | unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr, |
| 1879 | unsigned long len, unsigned long pgoff, unsigned long flags) |
| 1880 | { |
| 1881 | return -ENOMEM; |
| 1882 | } |
| 1883 | |
| 1884 | void unmap_mapping_range(struct address_space *mapping, |
| 1885 | loff_t const holebegin, loff_t const holelen, |
| 1886 | int even_cows) |
| 1887 | { |
| 1888 | } |
| 1889 | EXPORT_SYMBOL(unmap_mapping_range); |
| 1890 | |
| 1891 | /* |
| 1892 | * Check that a process has enough memory to allocate a new virtual |
| 1893 | * mapping. 0 means there is enough memory for the allocation to |
| 1894 | * succeed and -ENOMEM implies there is not. |
| 1895 | * |
| 1896 | * We currently support three overcommit policies, which are set via the |
| 1897 | * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting |
| 1898 | * |
| 1899 | * Strict overcommit modes added 2002 Feb 26 by Alan Cox. |
| 1900 | * Additional code 2002 Jul 20 by Robert Love. |
| 1901 | * |
| 1902 | * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise. |
| 1903 | * |
| 1904 | * Note this is a helper function intended to be used by LSMs which |
| 1905 | * wish to use this logic. |
| 1906 | */ |
| 1907 | int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin) |
| 1908 | { |
| 1909 | unsigned long free, allowed, reserve; |
| 1910 | |
| 1911 | vm_acct_memory(pages); |
| 1912 | |
| 1913 | /* |
| 1914 | * Sometimes we want to use more memory than we have |
| 1915 | */ |
| 1916 | if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS) |
| 1917 | return 0; |
| 1918 | |
| 1919 | if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) { |
| 1920 | free = global_page_state(NR_FREE_PAGES); |
| 1921 | free += global_page_state(NR_FILE_PAGES); |
| 1922 | |
| 1923 | /* |
| 1924 | * shmem pages shouldn't be counted as free in this |
| 1925 | * case, they can't be purged, only swapped out, and |
| 1926 | * that won't affect the overall amount of available |
| 1927 | * memory in the system. |
| 1928 | */ |
| 1929 | free -= global_page_state(NR_SHMEM); |
| 1930 | |
| 1931 | free += get_nr_swap_pages(); |
| 1932 | |
| 1933 | /* |
| 1934 | * Any slabs which are created with the |
| 1935 | * SLAB_RECLAIM_ACCOUNT flag claim to have contents |
| 1936 | * which are reclaimable, under pressure. The dentry |
| 1937 | * cache and most inode caches should fall into this |
| 1938 | */ |
| 1939 | free += global_page_state(NR_SLAB_RECLAIMABLE); |
| 1940 | |
| 1941 | /* |
| 1942 | * Leave reserved pages. The pages are not for anonymous pages. |
| 1943 | */ |
| 1944 | if (free <= totalreserve_pages) |
| 1945 | goto error; |
| 1946 | else |
| 1947 | free -= totalreserve_pages; |
| 1948 | |
| 1949 | /* |
| 1950 | * Reserve some for root |
| 1951 | */ |
| 1952 | if (!cap_sys_admin) |
| 1953 | free -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10); |
| 1954 | |
| 1955 | if (free > pages) |
| 1956 | return 0; |
| 1957 | |
| 1958 | goto error; |
| 1959 | } |
| 1960 | |
| 1961 | allowed = vm_commit_limit(); |
| 1962 | /* |
| 1963 | * Reserve some 3% for root |
| 1964 | */ |
| 1965 | if (!cap_sys_admin) |
| 1966 | allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10); |
| 1967 | |
| 1968 | /* |
| 1969 | * Don't let a single process grow so big a user can't recover |
| 1970 | */ |
| 1971 | if (mm) { |
| 1972 | reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10); |
| 1973 | allowed -= min(mm->total_vm / 32, reserve); |
| 1974 | } |
| 1975 | |
| 1976 | if (percpu_counter_read_positive(&vm_committed_as) < allowed) |
| 1977 | return 0; |
| 1978 | |
| 1979 | error: |
| 1980 | vm_unacct_memory(pages); |
| 1981 | |
| 1982 | return -ENOMEM; |
| 1983 | } |
| 1984 | |
| 1985 | int in_gate_area_no_mm(unsigned long addr) |
| 1986 | { |
| 1987 | return 0; |
| 1988 | } |
| 1989 | |
| 1990 | int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf) |
| 1991 | { |
| 1992 | BUG(); |
| 1993 | return 0; |
| 1994 | } |
| 1995 | EXPORT_SYMBOL(filemap_fault); |
| 1996 | |
| 1997 | void filemap_map_pages(struct vm_area_struct *vma, struct vm_fault *vmf) |
| 1998 | { |
| 1999 | BUG(); |
| 2000 | } |
| 2001 | EXPORT_SYMBOL(filemap_map_pages); |
| 2002 | |
| 2003 | int generic_file_remap_pages(struct vm_area_struct *vma, unsigned long addr, |
| 2004 | unsigned long size, pgoff_t pgoff) |
| 2005 | { |
| 2006 | BUG(); |
| 2007 | return 0; |
| 2008 | } |
| 2009 | EXPORT_SYMBOL(generic_file_remap_pages); |
| 2010 | |
| 2011 | static int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm, |
| 2012 | unsigned long addr, void *buf, int len, int write) |
| 2013 | { |
| 2014 | struct vm_area_struct *vma; |
| 2015 | |
| 2016 | down_read(&mm->mmap_sem); |
| 2017 | |
| 2018 | /* the access must start within one of the target process's mappings */ |
| 2019 | vma = find_vma(mm, addr); |
| 2020 | if (vma) { |
| 2021 | /* don't overrun this mapping */ |
| 2022 | if (addr + len >= vma->vm_end) |
| 2023 | len = vma->vm_end - addr; |
| 2024 | |
| 2025 | /* only read or write mappings where it is permitted */ |
| 2026 | if (write && vma->vm_flags & VM_MAYWRITE) |
| 2027 | copy_to_user_page(vma, NULL, addr, |
| 2028 | (void *) addr, buf, len); |
| 2029 | else if (!write && vma->vm_flags & VM_MAYREAD) |
| 2030 | copy_from_user_page(vma, NULL, addr, |
| 2031 | buf, (void *) addr, len); |
| 2032 | else |
| 2033 | len = 0; |
| 2034 | } else { |
| 2035 | len = 0; |
| 2036 | } |
| 2037 | |
| 2038 | up_read(&mm->mmap_sem); |
| 2039 | |
| 2040 | return len; |
| 2041 | } |
| 2042 | |
| 2043 | /** |
| 2044 | * @access_remote_vm - access another process' address space |
| 2045 | * @mm: the mm_struct of the target address space |
| 2046 | * @addr: start address to access |
| 2047 | * @buf: source or destination buffer |
| 2048 | * @len: number of bytes to transfer |
| 2049 | * @write: whether the access is a write |
| 2050 | * |
| 2051 | * The caller must hold a reference on @mm. |
| 2052 | */ |
| 2053 | int access_remote_vm(struct mm_struct *mm, unsigned long addr, |
| 2054 | void *buf, int len, int write) |
| 2055 | { |
| 2056 | return __access_remote_vm(NULL, mm, addr, buf, len, write); |
| 2057 | } |
| 2058 | |
| 2059 | /* |
| 2060 | * Access another process' address space. |
| 2061 | * - source/target buffer must be kernel space |
| 2062 | */ |
| 2063 | int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write) |
| 2064 | { |
| 2065 | struct mm_struct *mm; |
| 2066 | |
| 2067 | if (addr + len < addr) |
| 2068 | return 0; |
| 2069 | |
| 2070 | mm = get_task_mm(tsk); |
| 2071 | if (!mm) |
| 2072 | return 0; |
| 2073 | |
| 2074 | len = __access_remote_vm(tsk, mm, addr, buf, len, write); |
| 2075 | |
| 2076 | mmput(mm); |
| 2077 | return len; |
| 2078 | } |
| 2079 | |
| 2080 | /** |
| 2081 | * nommu_shrink_inode_mappings - Shrink the shared mappings on an inode |
| 2082 | * @inode: The inode to check |
| 2083 | * @size: The current filesize of the inode |
| 2084 | * @newsize: The proposed filesize of the inode |
| 2085 | * |
| 2086 | * Check the shared mappings on an inode on behalf of a shrinking truncate to |
| 2087 | * make sure that that any outstanding VMAs aren't broken and then shrink the |
| 2088 | * vm_regions that extend that beyond so that do_mmap_pgoff() doesn't |
| 2089 | * automatically grant mappings that are too large. |
| 2090 | */ |
| 2091 | int nommu_shrink_inode_mappings(struct inode *inode, size_t size, |
| 2092 | size_t newsize) |
| 2093 | { |
| 2094 | struct vm_area_struct *vma; |
| 2095 | struct vm_region *region; |
| 2096 | pgoff_t low, high; |
| 2097 | size_t r_size, r_top; |
| 2098 | |
| 2099 | low = newsize >> PAGE_SHIFT; |
| 2100 | high = (size + PAGE_SIZE - 1) >> PAGE_SHIFT; |
| 2101 | |
| 2102 | down_write(&nommu_region_sem); |
| 2103 | mutex_lock(&inode->i_mapping->i_mmap_mutex); |
| 2104 | |
| 2105 | /* search for VMAs that fall within the dead zone */ |
| 2106 | vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, low, high) { |
| 2107 | /* found one - only interested if it's shared out of the page |
| 2108 | * cache */ |
| 2109 | if (vma->vm_flags & VM_SHARED) { |
| 2110 | mutex_unlock(&inode->i_mapping->i_mmap_mutex); |
| 2111 | up_write(&nommu_region_sem); |
| 2112 | return -ETXTBSY; /* not quite true, but near enough */ |
| 2113 | } |
| 2114 | } |
| 2115 | |
| 2116 | /* reduce any regions that overlap the dead zone - if in existence, |
| 2117 | * these will be pointed to by VMAs that don't overlap the dead zone |
| 2118 | * |
| 2119 | * we don't check for any regions that start beyond the EOF as there |
| 2120 | * shouldn't be any |
| 2121 | */ |
| 2122 | vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, |
| 2123 | 0, ULONG_MAX) { |
| 2124 | if (!(vma->vm_flags & VM_SHARED)) |
| 2125 | continue; |
| 2126 | |
| 2127 | region = vma->vm_region; |
| 2128 | r_size = region->vm_top - region->vm_start; |
| 2129 | r_top = (region->vm_pgoff << PAGE_SHIFT) + r_size; |
| 2130 | |
| 2131 | if (r_top > newsize) { |
| 2132 | region->vm_top -= r_top - newsize; |
| 2133 | if (region->vm_end > region->vm_top) |
| 2134 | region->vm_end = region->vm_top; |
| 2135 | } |
| 2136 | } |
| 2137 | |
| 2138 | mutex_unlock(&inode->i_mapping->i_mmap_mutex); |
| 2139 | up_write(&nommu_region_sem); |
| 2140 | return 0; |
| 2141 | } |
| 2142 | |
| 2143 | /* |
| 2144 | * Initialise sysctl_user_reserve_kbytes. |
| 2145 | * |
| 2146 | * This is intended to prevent a user from starting a single memory hogging |
| 2147 | * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER |
| 2148 | * mode. |
| 2149 | * |
| 2150 | * The default value is min(3% of free memory, 128MB) |
| 2151 | * 128MB is enough to recover with sshd/login, bash, and top/kill. |
| 2152 | */ |
| 2153 | static int __meminit init_user_reserve(void) |
| 2154 | { |
| 2155 | unsigned long free_kbytes; |
| 2156 | |
| 2157 | free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); |
| 2158 | |
| 2159 | sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17); |
| 2160 | return 0; |
| 2161 | } |
| 2162 | module_init(init_user_reserve) |
| 2163 | |
| 2164 | /* |
| 2165 | * Initialise sysctl_admin_reserve_kbytes. |
| 2166 | * |
| 2167 | * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin |
| 2168 | * to log in and kill a memory hogging process. |
| 2169 | * |
| 2170 | * Systems with more than 256MB will reserve 8MB, enough to recover |
| 2171 | * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will |
| 2172 | * only reserve 3% of free pages by default. |
| 2173 | */ |
| 2174 | static int __meminit init_admin_reserve(void) |
| 2175 | { |
| 2176 | unsigned long free_kbytes; |
| 2177 | |
| 2178 | free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); |
| 2179 | |
| 2180 | sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13); |
| 2181 | return 0; |
| 2182 | } |
| 2183 | module_init(init_admin_reserve) |