Commit | Line | Data |
---|---|---|
1da177e4 LT |
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 | * | |
8feae131 | 9 | * Copyright (c) 2004-2008 David Howells <dhowells@redhat.com> |
1da177e4 LT |
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> | |
29c185e5 | 13 | * Copyright (c) 2007-2010 Paul Mundt <lethal@linux-sh.org> |
1da177e4 LT |
14 | */ |
15 | ||
f2b8544f | 16 | #include <linux/module.h> |
1da177e4 LT |
17 | #include <linux/mm.h> |
18 | #include <linux/mman.h> | |
19 | #include <linux/swap.h> | |
20 | #include <linux/file.h> | |
21 | #include <linux/highmem.h> | |
22 | #include <linux/pagemap.h> | |
23 | #include <linux/slab.h> | |
24 | #include <linux/vmalloc.h> | |
fa8e26cc | 25 | #include <linux/tracehook.h> |
1da177e4 LT |
26 | #include <linux/blkdev.h> |
27 | #include <linux/backing-dev.h> | |
28 | #include <linux/mount.h> | |
29 | #include <linux/personality.h> | |
30 | #include <linux/security.h> | |
31 | #include <linux/syscalls.h> | |
120a795d | 32 | #include <linux/audit.h> |
1da177e4 LT |
33 | |
34 | #include <asm/uaccess.h> | |
35 | #include <asm/tlb.h> | |
36 | #include <asm/tlbflush.h> | |
eb8cdec4 | 37 | #include <asm/mmu_context.h> |
8feae131 DH |
38 | #include "internal.h" |
39 | ||
8feae131 DH |
40 | #if 0 |
41 | #define kenter(FMT, ...) \ | |
42 | printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__) | |
43 | #define kleave(FMT, ...) \ | |
44 | printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__) | |
45 | #define kdebug(FMT, ...) \ | |
46 | printk(KERN_DEBUG "xxx" FMT"yyy\n", ##__VA_ARGS__) | |
47 | #else | |
48 | #define kenter(FMT, ...) \ | |
49 | no_printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__) | |
50 | #define kleave(FMT, ...) \ | |
51 | no_printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__) | |
52 | #define kdebug(FMT, ...) \ | |
53 | no_printk(KERN_DEBUG FMT"\n", ##__VA_ARGS__) | |
54 | #endif | |
1da177e4 LT |
55 | |
56 | void *high_memory; | |
57 | struct page *mem_map; | |
58 | unsigned long max_mapnr; | |
59 | unsigned long num_physpages; | |
4266c97a | 60 | unsigned long highest_memmap_pfn; |
00a62ce9 | 61 | struct percpu_counter vm_committed_as; |
1da177e4 LT |
62 | int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */ |
63 | int sysctl_overcommit_ratio = 50; /* default is 50% */ | |
64 | int sysctl_max_map_count = DEFAULT_MAX_MAP_COUNT; | |
fc4d5c29 | 65 | int sysctl_nr_trim_pages = CONFIG_NOMMU_INITIAL_TRIM_EXCESS; |
1da177e4 LT |
66 | int heap_stack_gap = 0; |
67 | ||
33e5d769 | 68 | atomic_long_t mmap_pages_allocated; |
8feae131 | 69 | |
1da177e4 | 70 | EXPORT_SYMBOL(mem_map); |
6a04de6d | 71 | EXPORT_SYMBOL(num_physpages); |
1da177e4 | 72 | |
8feae131 DH |
73 | /* list of mapped, potentially shareable regions */ |
74 | static struct kmem_cache *vm_region_jar; | |
75 | struct rb_root nommu_region_tree = RB_ROOT; | |
76 | DECLARE_RWSEM(nommu_region_sem); | |
1da177e4 | 77 | |
f0f37e2f | 78 | const struct vm_operations_struct generic_file_vm_ops = { |
1da177e4 LT |
79 | }; |
80 | ||
1da177e4 LT |
81 | /* |
82 | * Return the total memory allocated for this pointer, not | |
83 | * just what the caller asked for. | |
84 | * | |
85 | * Doesn't have to be accurate, i.e. may have races. | |
86 | */ | |
87 | unsigned int kobjsize(const void *objp) | |
88 | { | |
89 | struct page *page; | |
90 | ||
4016a139 MH |
91 | /* |
92 | * If the object we have should not have ksize performed on it, | |
93 | * return size of 0 | |
94 | */ | |
5a1603be | 95 | if (!objp || !virt_addr_valid(objp)) |
6cfd53fc PM |
96 | return 0; |
97 | ||
98 | page = virt_to_head_page(objp); | |
6cfd53fc PM |
99 | |
100 | /* | |
101 | * If the allocator sets PageSlab, we know the pointer came from | |
102 | * kmalloc(). | |
103 | */ | |
1da177e4 LT |
104 | if (PageSlab(page)) |
105 | return ksize(objp); | |
106 | ||
ab2e83ea PM |
107 | /* |
108 | * If it's not a compound page, see if we have a matching VMA | |
109 | * region. This test is intentionally done in reverse order, | |
110 | * so if there's no VMA, we still fall through and hand back | |
111 | * PAGE_SIZE for 0-order pages. | |
112 | */ | |
113 | if (!PageCompound(page)) { | |
114 | struct vm_area_struct *vma; | |
115 | ||
116 | vma = find_vma(current->mm, (unsigned long)objp); | |
117 | if (vma) | |
118 | return vma->vm_end - vma->vm_start; | |
119 | } | |
120 | ||
6cfd53fc PM |
121 | /* |
122 | * The ksize() function is only guaranteed to work for pointers | |
5a1603be | 123 | * returned by kmalloc(). So handle arbitrary pointers here. |
6cfd53fc | 124 | */ |
5a1603be | 125 | return PAGE_SIZE << compound_order(page); |
1da177e4 LT |
126 | } |
127 | ||
b291f000 | 128 | int __get_user_pages(struct task_struct *tsk, struct mm_struct *mm, |
4266c97a | 129 | unsigned long start, int nr_pages, unsigned int foll_flags, |
53a7706d ML |
130 | struct page **pages, struct vm_area_struct **vmas, |
131 | int *retry) | |
1da177e4 | 132 | { |
910e46da | 133 | struct vm_area_struct *vma; |
7b4d5b8b DH |
134 | unsigned long vm_flags; |
135 | int i; | |
136 | ||
137 | /* calculate required read or write permissions. | |
58fa879e | 138 | * If FOLL_FORCE is set, we only require the "MAY" flags. |
7b4d5b8b | 139 | */ |
58fa879e HD |
140 | vm_flags = (foll_flags & FOLL_WRITE) ? |
141 | (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD); | |
142 | vm_flags &= (foll_flags & FOLL_FORCE) ? | |
143 | (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE); | |
1da177e4 | 144 | |
9d73777e | 145 | for (i = 0; i < nr_pages; i++) { |
7561e8ca | 146 | vma = find_vma(mm, start); |
7b4d5b8b DH |
147 | if (!vma) |
148 | goto finish_or_fault; | |
149 | ||
150 | /* protect what we can, including chardevs */ | |
1c3aff1c HD |
151 | if ((vma->vm_flags & (VM_IO | VM_PFNMAP)) || |
152 | !(vm_flags & vma->vm_flags)) | |
7b4d5b8b | 153 | goto finish_or_fault; |
910e46da | 154 | |
1da177e4 LT |
155 | if (pages) { |
156 | pages[i] = virt_to_page(start); | |
157 | if (pages[i]) | |
158 | page_cache_get(pages[i]); | |
159 | } | |
160 | if (vmas) | |
910e46da | 161 | vmas[i] = vma; |
e1ee65d8 | 162 | start = (start + PAGE_SIZE) & PAGE_MASK; |
1da177e4 | 163 | } |
7b4d5b8b DH |
164 | |
165 | return i; | |
166 | ||
167 | finish_or_fault: | |
168 | return i ? : -EFAULT; | |
1da177e4 | 169 | } |
b291f000 | 170 | |
b291f000 NP |
171 | /* |
172 | * get a list of pages in an address range belonging to the specified process | |
173 | * and indicate the VMA that covers each page | |
174 | * - this is potentially dodgy as we may end incrementing the page count of a | |
175 | * slab page or a secondary page from a compound page | |
176 | * - don't permit access to VMAs that don't support it, such as I/O mappings | |
177 | */ | |
178 | int get_user_pages(struct task_struct *tsk, struct mm_struct *mm, | |
9d73777e | 179 | unsigned long start, int nr_pages, int write, int force, |
b291f000 NP |
180 | struct page **pages, struct vm_area_struct **vmas) |
181 | { | |
182 | int flags = 0; | |
183 | ||
184 | if (write) | |
58fa879e | 185 | flags |= FOLL_WRITE; |
b291f000 | 186 | if (force) |
58fa879e | 187 | flags |= FOLL_FORCE; |
b291f000 | 188 | |
53a7706d ML |
189 | return __get_user_pages(tsk, mm, start, nr_pages, flags, pages, vmas, |
190 | NULL); | |
b291f000 | 191 | } |
66aa2b4b GU |
192 | EXPORT_SYMBOL(get_user_pages); |
193 | ||
dfc2f91a PM |
194 | /** |
195 | * follow_pfn - look up PFN at a user virtual address | |
196 | * @vma: memory mapping | |
197 | * @address: user virtual address | |
198 | * @pfn: location to store found PFN | |
199 | * | |
200 | * Only IO mappings and raw PFN mappings are allowed. | |
201 | * | |
202 | * Returns zero and the pfn at @pfn on success, -ve otherwise. | |
203 | */ | |
204 | int follow_pfn(struct vm_area_struct *vma, unsigned long address, | |
205 | unsigned long *pfn) | |
206 | { | |
207 | if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) | |
208 | return -EINVAL; | |
209 | ||
210 | *pfn = address >> PAGE_SHIFT; | |
211 | return 0; | |
212 | } | |
213 | EXPORT_SYMBOL(follow_pfn); | |
214 | ||
1da177e4 LT |
215 | DEFINE_RWLOCK(vmlist_lock); |
216 | struct vm_struct *vmlist; | |
217 | ||
b3bdda02 | 218 | void vfree(const void *addr) |
1da177e4 LT |
219 | { |
220 | kfree(addr); | |
221 | } | |
b5073173 | 222 | EXPORT_SYMBOL(vfree); |
1da177e4 | 223 | |
dd0fc66f | 224 | void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot) |
1da177e4 LT |
225 | { |
226 | /* | |
8518609d RD |
227 | * You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc() |
228 | * returns only a logical address. | |
1da177e4 | 229 | */ |
84097518 | 230 | return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM); |
1da177e4 | 231 | } |
b5073173 | 232 | EXPORT_SYMBOL(__vmalloc); |
1da177e4 | 233 | |
f905bc44 PM |
234 | void *vmalloc_user(unsigned long size) |
235 | { | |
236 | void *ret; | |
237 | ||
238 | ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, | |
239 | PAGE_KERNEL); | |
240 | if (ret) { | |
241 | struct vm_area_struct *vma; | |
242 | ||
243 | down_write(¤t->mm->mmap_sem); | |
244 | vma = find_vma(current->mm, (unsigned long)ret); | |
245 | if (vma) | |
246 | vma->vm_flags |= VM_USERMAP; | |
247 | up_write(¤t->mm->mmap_sem); | |
248 | } | |
249 | ||
250 | return ret; | |
251 | } | |
252 | EXPORT_SYMBOL(vmalloc_user); | |
253 | ||
b3bdda02 | 254 | struct page *vmalloc_to_page(const void *addr) |
1da177e4 LT |
255 | { |
256 | return virt_to_page(addr); | |
257 | } | |
b5073173 | 258 | EXPORT_SYMBOL(vmalloc_to_page); |
1da177e4 | 259 | |
b3bdda02 | 260 | unsigned long vmalloc_to_pfn(const void *addr) |
1da177e4 LT |
261 | { |
262 | return page_to_pfn(virt_to_page(addr)); | |
263 | } | |
b5073173 | 264 | EXPORT_SYMBOL(vmalloc_to_pfn); |
1da177e4 LT |
265 | |
266 | long vread(char *buf, char *addr, unsigned long count) | |
267 | { | |
268 | memcpy(buf, addr, count); | |
269 | return count; | |
270 | } | |
271 | ||
272 | long vwrite(char *buf, char *addr, unsigned long count) | |
273 | { | |
274 | /* Don't allow overflow */ | |
275 | if ((unsigned long) addr + count < count) | |
276 | count = -(unsigned long) addr; | |
277 | ||
278 | memcpy(addr, buf, count); | |
279 | return(count); | |
280 | } | |
281 | ||
282 | /* | |
283 | * vmalloc - allocate virtually continguos memory | |
284 | * | |
285 | * @size: allocation size | |
286 | * | |
287 | * Allocate enough pages to cover @size from the page level | |
288 | * allocator and map them into continguos kernel virtual space. | |
289 | * | |
c1c8897f | 290 | * For tight control over page level allocator and protection flags |
1da177e4 LT |
291 | * use __vmalloc() instead. |
292 | */ | |
293 | void *vmalloc(unsigned long size) | |
294 | { | |
295 | return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL); | |
296 | } | |
f6138882 AM |
297 | EXPORT_SYMBOL(vmalloc); |
298 | ||
e1ca7788 DY |
299 | /* |
300 | * vzalloc - allocate virtually continguos memory with zero fill | |
301 | * | |
302 | * @size: allocation size | |
303 | * | |
304 | * Allocate enough pages to cover @size from the page level | |
305 | * allocator and map them into continguos kernel virtual space. | |
306 | * The memory allocated is set to zero. | |
307 | * | |
308 | * For tight control over page level allocator and protection flags | |
309 | * use __vmalloc() instead. | |
310 | */ | |
311 | void *vzalloc(unsigned long size) | |
312 | { | |
313 | return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, | |
314 | PAGE_KERNEL); | |
315 | } | |
316 | EXPORT_SYMBOL(vzalloc); | |
317 | ||
318 | /** | |
319 | * vmalloc_node - allocate memory on a specific node | |
320 | * @size: allocation size | |
321 | * @node: numa node | |
322 | * | |
323 | * Allocate enough pages to cover @size from the page level | |
324 | * allocator and map them into contiguous kernel virtual space. | |
325 | * | |
326 | * For tight control over page level allocator and protection flags | |
327 | * use __vmalloc() instead. | |
328 | */ | |
f6138882 AM |
329 | void *vmalloc_node(unsigned long size, int node) |
330 | { | |
331 | return vmalloc(size); | |
332 | } | |
9a14f653 | 333 | EXPORT_SYMBOL(vmalloc_node); |
e1ca7788 DY |
334 | |
335 | /** | |
336 | * vzalloc_node - allocate memory on a specific node with zero fill | |
337 | * @size: allocation size | |
338 | * @node: numa node | |
339 | * | |
340 | * Allocate enough pages to cover @size from the page level | |
341 | * allocator and map them into contiguous kernel virtual space. | |
342 | * The memory allocated is set to zero. | |
343 | * | |
344 | * For tight control over page level allocator and protection flags | |
345 | * use __vmalloc() instead. | |
346 | */ | |
347 | void *vzalloc_node(unsigned long size, int node) | |
348 | { | |
349 | return vzalloc(size); | |
350 | } | |
351 | EXPORT_SYMBOL(vzalloc_node); | |
1da177e4 | 352 | |
1af446ed PM |
353 | #ifndef PAGE_KERNEL_EXEC |
354 | # define PAGE_KERNEL_EXEC PAGE_KERNEL | |
355 | #endif | |
356 | ||
357 | /** | |
358 | * vmalloc_exec - allocate virtually contiguous, executable memory | |
359 | * @size: allocation size | |
360 | * | |
361 | * Kernel-internal function to allocate enough pages to cover @size | |
362 | * the page level allocator and map them into contiguous and | |
363 | * executable kernel virtual space. | |
364 | * | |
365 | * For tight control over page level allocator and protection flags | |
366 | * use __vmalloc() instead. | |
367 | */ | |
368 | ||
369 | void *vmalloc_exec(unsigned long size) | |
370 | { | |
371 | return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC); | |
372 | } | |
373 | ||
b5073173 PM |
374 | /** |
375 | * vmalloc_32 - allocate virtually contiguous memory (32bit addressable) | |
1da177e4 LT |
376 | * @size: allocation size |
377 | * | |
378 | * Allocate enough 32bit PA addressable pages to cover @size from the | |
379 | * page level allocator and map them into continguos kernel virtual space. | |
380 | */ | |
381 | void *vmalloc_32(unsigned long size) | |
382 | { | |
383 | return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL); | |
384 | } | |
b5073173 PM |
385 | EXPORT_SYMBOL(vmalloc_32); |
386 | ||
387 | /** | |
388 | * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory | |
389 | * @size: allocation size | |
390 | * | |
391 | * The resulting memory area is 32bit addressable and zeroed so it can be | |
392 | * mapped to userspace without leaking data. | |
f905bc44 PM |
393 | * |
394 | * VM_USERMAP is set on the corresponding VMA so that subsequent calls to | |
395 | * remap_vmalloc_range() are permissible. | |
b5073173 PM |
396 | */ |
397 | void *vmalloc_32_user(unsigned long size) | |
398 | { | |
f905bc44 PM |
399 | /* |
400 | * We'll have to sort out the ZONE_DMA bits for 64-bit, | |
401 | * but for now this can simply use vmalloc_user() directly. | |
402 | */ | |
403 | return vmalloc_user(size); | |
b5073173 PM |
404 | } |
405 | EXPORT_SYMBOL(vmalloc_32_user); | |
1da177e4 LT |
406 | |
407 | void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot) | |
408 | { | |
409 | BUG(); | |
410 | return NULL; | |
411 | } | |
b5073173 | 412 | EXPORT_SYMBOL(vmap); |
1da177e4 | 413 | |
b3bdda02 | 414 | void vunmap(const void *addr) |
1da177e4 LT |
415 | { |
416 | BUG(); | |
417 | } | |
b5073173 | 418 | EXPORT_SYMBOL(vunmap); |
1da177e4 | 419 | |
eb6434d9 PM |
420 | void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot) |
421 | { | |
422 | BUG(); | |
423 | return NULL; | |
424 | } | |
425 | EXPORT_SYMBOL(vm_map_ram); | |
426 | ||
427 | void vm_unmap_ram(const void *mem, unsigned int count) | |
428 | { | |
429 | BUG(); | |
430 | } | |
431 | EXPORT_SYMBOL(vm_unmap_ram); | |
432 | ||
433 | void vm_unmap_aliases(void) | |
434 | { | |
435 | } | |
436 | EXPORT_SYMBOL_GPL(vm_unmap_aliases); | |
437 | ||
1eeb66a1 CH |
438 | /* |
439 | * Implement a stub for vmalloc_sync_all() if the architecture chose not to | |
440 | * have one. | |
441 | */ | |
442 | void __attribute__((weak)) vmalloc_sync_all(void) | |
443 | { | |
444 | } | |
445 | ||
29c185e5 PM |
446 | /** |
447 | * alloc_vm_area - allocate a range of kernel address space | |
448 | * @size: size of the area | |
449 | * | |
450 | * Returns: NULL on failure, vm_struct on success | |
451 | * | |
452 | * This function reserves a range of kernel address space, and | |
453 | * allocates pagetables to map that range. No actual mappings | |
454 | * are created. If the kernel address space is not shared | |
455 | * between processes, it syncs the pagetable across all | |
456 | * processes. | |
457 | */ | |
458 | struct vm_struct *alloc_vm_area(size_t size) | |
459 | { | |
460 | BUG(); | |
461 | return NULL; | |
462 | } | |
463 | EXPORT_SYMBOL_GPL(alloc_vm_area); | |
464 | ||
465 | void free_vm_area(struct vm_struct *area) | |
466 | { | |
467 | BUG(); | |
468 | } | |
469 | EXPORT_SYMBOL_GPL(free_vm_area); | |
470 | ||
b5073173 PM |
471 | int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, |
472 | struct page *page) | |
473 | { | |
474 | return -EINVAL; | |
475 | } | |
476 | EXPORT_SYMBOL(vm_insert_page); | |
477 | ||
1da177e4 LT |
478 | /* |
479 | * sys_brk() for the most part doesn't need the global kernel | |
480 | * lock, except when an application is doing something nasty | |
481 | * like trying to un-brk an area that has already been mapped | |
482 | * to a regular file. in this case, the unmapping will need | |
483 | * to invoke file system routines that need the global lock. | |
484 | */ | |
6a6160a7 | 485 | SYSCALL_DEFINE1(brk, unsigned long, brk) |
1da177e4 LT |
486 | { |
487 | struct mm_struct *mm = current->mm; | |
488 | ||
489 | if (brk < mm->start_brk || brk > mm->context.end_brk) | |
490 | return mm->brk; | |
491 | ||
492 | if (mm->brk == brk) | |
493 | return mm->brk; | |
494 | ||
495 | /* | |
496 | * Always allow shrinking brk | |
497 | */ | |
498 | if (brk <= mm->brk) { | |
499 | mm->brk = brk; | |
500 | return brk; | |
501 | } | |
502 | ||
503 | /* | |
504 | * Ok, looks good - let it rip. | |
505 | */ | |
cfe79c00 | 506 | flush_icache_range(mm->brk, brk); |
1da177e4 LT |
507 | return mm->brk = brk; |
508 | } | |
509 | ||
8feae131 DH |
510 | /* |
511 | * initialise the VMA and region record slabs | |
512 | */ | |
513 | void __init mmap_init(void) | |
1da177e4 | 514 | { |
00a62ce9 KM |
515 | int ret; |
516 | ||
517 | ret = percpu_counter_init(&vm_committed_as, 0); | |
518 | VM_BUG_ON(ret); | |
33e5d769 | 519 | vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC); |
1da177e4 | 520 | } |
1da177e4 | 521 | |
3034097a | 522 | /* |
8feae131 DH |
523 | * validate the region tree |
524 | * - the caller must hold the region lock | |
3034097a | 525 | */ |
8feae131 DH |
526 | #ifdef CONFIG_DEBUG_NOMMU_REGIONS |
527 | static noinline void validate_nommu_regions(void) | |
3034097a | 528 | { |
8feae131 DH |
529 | struct vm_region *region, *last; |
530 | struct rb_node *p, *lastp; | |
3034097a | 531 | |
8feae131 DH |
532 | lastp = rb_first(&nommu_region_tree); |
533 | if (!lastp) | |
534 | return; | |
535 | ||
536 | last = rb_entry(lastp, struct vm_region, vm_rb); | |
33e5d769 DH |
537 | BUG_ON(unlikely(last->vm_end <= last->vm_start)); |
538 | BUG_ON(unlikely(last->vm_top < last->vm_end)); | |
8feae131 DH |
539 | |
540 | while ((p = rb_next(lastp))) { | |
541 | region = rb_entry(p, struct vm_region, vm_rb); | |
542 | last = rb_entry(lastp, struct vm_region, vm_rb); | |
543 | ||
33e5d769 DH |
544 | BUG_ON(unlikely(region->vm_end <= region->vm_start)); |
545 | BUG_ON(unlikely(region->vm_top < region->vm_end)); | |
546 | BUG_ON(unlikely(region->vm_start < last->vm_top)); | |
3034097a | 547 | |
8feae131 DH |
548 | lastp = p; |
549 | } | |
3034097a | 550 | } |
8feae131 | 551 | #else |
33e5d769 DH |
552 | static void validate_nommu_regions(void) |
553 | { | |
554 | } | |
8feae131 | 555 | #endif |
3034097a DH |
556 | |
557 | /* | |
8feae131 | 558 | * add a region into the global tree |
3034097a | 559 | */ |
8feae131 | 560 | static void add_nommu_region(struct vm_region *region) |
3034097a | 561 | { |
8feae131 DH |
562 | struct vm_region *pregion; |
563 | struct rb_node **p, *parent; | |
3034097a | 564 | |
8feae131 DH |
565 | validate_nommu_regions(); |
566 | ||
8feae131 DH |
567 | parent = NULL; |
568 | p = &nommu_region_tree.rb_node; | |
569 | while (*p) { | |
570 | parent = *p; | |
571 | pregion = rb_entry(parent, struct vm_region, vm_rb); | |
572 | if (region->vm_start < pregion->vm_start) | |
573 | p = &(*p)->rb_left; | |
574 | else if (region->vm_start > pregion->vm_start) | |
575 | p = &(*p)->rb_right; | |
576 | else if (pregion == region) | |
577 | return; | |
578 | else | |
579 | BUG(); | |
3034097a DH |
580 | } |
581 | ||
8feae131 DH |
582 | rb_link_node(®ion->vm_rb, parent, p); |
583 | rb_insert_color(®ion->vm_rb, &nommu_region_tree); | |
3034097a | 584 | |
8feae131 | 585 | validate_nommu_regions(); |
3034097a | 586 | } |
3034097a | 587 | |
930e652a | 588 | /* |
8feae131 | 589 | * delete a region from the global tree |
930e652a | 590 | */ |
8feae131 | 591 | static void delete_nommu_region(struct vm_region *region) |
930e652a | 592 | { |
8feae131 | 593 | BUG_ON(!nommu_region_tree.rb_node); |
930e652a | 594 | |
8feae131 DH |
595 | validate_nommu_regions(); |
596 | rb_erase(®ion->vm_rb, &nommu_region_tree); | |
597 | validate_nommu_regions(); | |
57c8f63e GU |
598 | } |
599 | ||
6fa5f80b | 600 | /* |
8feae131 | 601 | * free a contiguous series of pages |
6fa5f80b | 602 | */ |
8feae131 | 603 | static void free_page_series(unsigned long from, unsigned long to) |
6fa5f80b | 604 | { |
8feae131 DH |
605 | for (; from < to; from += PAGE_SIZE) { |
606 | struct page *page = virt_to_page(from); | |
607 | ||
608 | kdebug("- free %lx", from); | |
33e5d769 | 609 | atomic_long_dec(&mmap_pages_allocated); |
8feae131 | 610 | if (page_count(page) != 1) |
33e5d769 DH |
611 | kdebug("free page %p: refcount not one: %d", |
612 | page, page_count(page)); | |
8feae131 | 613 | put_page(page); |
6fa5f80b | 614 | } |
6fa5f80b DH |
615 | } |
616 | ||
3034097a | 617 | /* |
8feae131 | 618 | * release a reference to a region |
33e5d769 | 619 | * - the caller must hold the region semaphore for writing, which this releases |
dd8632a1 | 620 | * - the region may not have been added to the tree yet, in which case vm_top |
8feae131 | 621 | * will equal vm_start |
3034097a | 622 | */ |
8feae131 DH |
623 | static void __put_nommu_region(struct vm_region *region) |
624 | __releases(nommu_region_sem) | |
1da177e4 | 625 | { |
1e2ae599 | 626 | kenter("%p{%d}", region, region->vm_usage); |
1da177e4 | 627 | |
8feae131 | 628 | BUG_ON(!nommu_region_tree.rb_node); |
1da177e4 | 629 | |
1e2ae599 | 630 | if (--region->vm_usage == 0) { |
dd8632a1 | 631 | if (region->vm_top > region->vm_start) |
8feae131 DH |
632 | delete_nommu_region(region); |
633 | up_write(&nommu_region_sem); | |
634 | ||
635 | if (region->vm_file) | |
636 | fput(region->vm_file); | |
637 | ||
638 | /* IO memory and memory shared directly out of the pagecache | |
639 | * from ramfs/tmpfs mustn't be released here */ | |
640 | if (region->vm_flags & VM_MAPPED_COPY) { | |
641 | kdebug("free series"); | |
dd8632a1 | 642 | free_page_series(region->vm_start, region->vm_top); |
8feae131 DH |
643 | } |
644 | kmem_cache_free(vm_region_jar, region); | |
645 | } else { | |
646 | up_write(&nommu_region_sem); | |
1da177e4 | 647 | } |
8feae131 | 648 | } |
1da177e4 | 649 | |
8feae131 DH |
650 | /* |
651 | * release a reference to a region | |
652 | */ | |
653 | static void put_nommu_region(struct vm_region *region) | |
654 | { | |
655 | down_write(&nommu_region_sem); | |
656 | __put_nommu_region(region); | |
1da177e4 LT |
657 | } |
658 | ||
eb8cdec4 BS |
659 | /* |
660 | * update protection on a vma | |
661 | */ | |
662 | static void protect_vma(struct vm_area_struct *vma, unsigned long flags) | |
663 | { | |
664 | #ifdef CONFIG_MPU | |
665 | struct mm_struct *mm = vma->vm_mm; | |
666 | long start = vma->vm_start & PAGE_MASK; | |
667 | while (start < vma->vm_end) { | |
668 | protect_page(mm, start, flags); | |
669 | start += PAGE_SIZE; | |
670 | } | |
671 | update_protections(mm); | |
672 | #endif | |
673 | } | |
674 | ||
3034097a | 675 | /* |
8feae131 DH |
676 | * add a VMA into a process's mm_struct in the appropriate place in the list |
677 | * and tree and add to the address space's page tree also if not an anonymous | |
678 | * page | |
679 | * - should be called with mm->mmap_sem held writelocked | |
3034097a | 680 | */ |
8feae131 | 681 | static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma) |
1da177e4 | 682 | { |
6038def0 | 683 | struct vm_area_struct *pvma, *prev; |
1da177e4 | 684 | struct address_space *mapping; |
6038def0 | 685 | struct rb_node **p, *parent, *rb_prev; |
8feae131 DH |
686 | |
687 | kenter(",%p", vma); | |
688 | ||
689 | BUG_ON(!vma->vm_region); | |
690 | ||
691 | mm->map_count++; | |
692 | vma->vm_mm = mm; | |
1da177e4 | 693 | |
eb8cdec4 BS |
694 | protect_vma(vma, vma->vm_flags); |
695 | ||
1da177e4 LT |
696 | /* add the VMA to the mapping */ |
697 | if (vma->vm_file) { | |
698 | mapping = vma->vm_file->f_mapping; | |
699 | ||
700 | flush_dcache_mmap_lock(mapping); | |
701 | vma_prio_tree_insert(vma, &mapping->i_mmap); | |
702 | flush_dcache_mmap_unlock(mapping); | |
703 | } | |
704 | ||
8feae131 | 705 | /* add the VMA to the tree */ |
6038def0 | 706 | parent = rb_prev = NULL; |
8feae131 | 707 | p = &mm->mm_rb.rb_node; |
1da177e4 LT |
708 | while (*p) { |
709 | parent = *p; | |
710 | pvma = rb_entry(parent, struct vm_area_struct, vm_rb); | |
711 | ||
8feae131 DH |
712 | /* sort by: start addr, end addr, VMA struct addr in that order |
713 | * (the latter is necessary as we may get identical VMAs) */ | |
714 | if (vma->vm_start < pvma->vm_start) | |
1da177e4 | 715 | p = &(*p)->rb_left; |
6038def0 NK |
716 | else if (vma->vm_start > pvma->vm_start) { |
717 | rb_prev = parent; | |
1da177e4 | 718 | p = &(*p)->rb_right; |
6038def0 | 719 | } else if (vma->vm_end < pvma->vm_end) |
8feae131 | 720 | p = &(*p)->rb_left; |
6038def0 NK |
721 | else if (vma->vm_end > pvma->vm_end) { |
722 | rb_prev = parent; | |
8feae131 | 723 | p = &(*p)->rb_right; |
6038def0 | 724 | } else if (vma < pvma) |
8feae131 | 725 | p = &(*p)->rb_left; |
6038def0 NK |
726 | else if (vma > pvma) { |
727 | rb_prev = parent; | |
8feae131 | 728 | p = &(*p)->rb_right; |
6038def0 | 729 | } else |
8feae131 | 730 | BUG(); |
1da177e4 LT |
731 | } |
732 | ||
733 | rb_link_node(&vma->vm_rb, parent, p); | |
8feae131 DH |
734 | rb_insert_color(&vma->vm_rb, &mm->mm_rb); |
735 | ||
736 | /* add VMA to the VMA list also */ | |
6038def0 NK |
737 | prev = NULL; |
738 | if (rb_prev) | |
739 | prev = rb_entry(rb_prev, struct vm_area_struct, vm_rb); | |
8feae131 | 740 | |
6038def0 | 741 | __vma_link_list(mm, vma, prev, parent); |
1da177e4 LT |
742 | } |
743 | ||
3034097a | 744 | /* |
8feae131 | 745 | * delete a VMA from its owning mm_struct and address space |
3034097a | 746 | */ |
8feae131 | 747 | static void delete_vma_from_mm(struct vm_area_struct *vma) |
1da177e4 LT |
748 | { |
749 | struct address_space *mapping; | |
8feae131 DH |
750 | struct mm_struct *mm = vma->vm_mm; |
751 | ||
752 | kenter("%p", vma); | |
753 | ||
eb8cdec4 BS |
754 | protect_vma(vma, 0); |
755 | ||
8feae131 DH |
756 | mm->map_count--; |
757 | if (mm->mmap_cache == vma) | |
758 | mm->mmap_cache = NULL; | |
1da177e4 LT |
759 | |
760 | /* remove the VMA from the mapping */ | |
761 | if (vma->vm_file) { | |
762 | mapping = vma->vm_file->f_mapping; | |
763 | ||
764 | flush_dcache_mmap_lock(mapping); | |
765 | vma_prio_tree_remove(vma, &mapping->i_mmap); | |
766 | flush_dcache_mmap_unlock(mapping); | |
767 | } | |
768 | ||
8feae131 DH |
769 | /* remove from the MM's tree and list */ |
770 | rb_erase(&vma->vm_rb, &mm->mm_rb); | |
b951bf2c NK |
771 | |
772 | if (vma->vm_prev) | |
773 | vma->vm_prev->vm_next = vma->vm_next; | |
774 | else | |
775 | mm->mmap = vma->vm_next; | |
776 | ||
777 | if (vma->vm_next) | |
778 | vma->vm_next->vm_prev = vma->vm_prev; | |
8feae131 DH |
779 | |
780 | vma->vm_mm = NULL; | |
781 | } | |
782 | ||
783 | /* | |
784 | * destroy a VMA record | |
785 | */ | |
786 | static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma) | |
787 | { | |
788 | kenter("%p", vma); | |
789 | if (vma->vm_ops && vma->vm_ops->close) | |
790 | vma->vm_ops->close(vma); | |
791 | if (vma->vm_file) { | |
792 | fput(vma->vm_file); | |
793 | if (vma->vm_flags & VM_EXECUTABLE) | |
794 | removed_exe_file_vma(mm); | |
795 | } | |
796 | put_nommu_region(vma->vm_region); | |
797 | kmem_cache_free(vm_area_cachep, vma); | |
798 | } | |
799 | ||
800 | /* | |
801 | * look up the first VMA in which addr resides, NULL if none | |
802 | * - should be called with mm->mmap_sem at least held readlocked | |
803 | */ | |
804 | struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr) | |
805 | { | |
806 | struct vm_area_struct *vma; | |
807 | struct rb_node *n = mm->mm_rb.rb_node; | |
808 | ||
809 | /* check the cache first */ | |
810 | vma = mm->mmap_cache; | |
811 | if (vma && vma->vm_start <= addr && vma->vm_end > addr) | |
812 | return vma; | |
813 | ||
814 | /* trawl the tree (there may be multiple mappings in which addr | |
815 | * resides) */ | |
816 | for (n = rb_first(&mm->mm_rb); n; n = rb_next(n)) { | |
817 | vma = rb_entry(n, struct vm_area_struct, vm_rb); | |
818 | if (vma->vm_start > addr) | |
819 | return NULL; | |
820 | if (vma->vm_end > addr) { | |
821 | mm->mmap_cache = vma; | |
822 | return vma; | |
823 | } | |
824 | } | |
825 | ||
826 | return NULL; | |
827 | } | |
828 | EXPORT_SYMBOL(find_vma); | |
829 | ||
830 | /* | |
831 | * find a VMA | |
832 | * - we don't extend stack VMAs under NOMMU conditions | |
833 | */ | |
834 | struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr) | |
835 | { | |
7561e8ca | 836 | return find_vma(mm, addr); |
8feae131 DH |
837 | } |
838 | ||
839 | /* | |
840 | * expand a stack to a given address | |
841 | * - not supported under NOMMU conditions | |
842 | */ | |
843 | int expand_stack(struct vm_area_struct *vma, unsigned long address) | |
844 | { | |
845 | return -ENOMEM; | |
846 | } | |
847 | ||
848 | /* | |
849 | * look up the first VMA exactly that exactly matches addr | |
850 | * - should be called with mm->mmap_sem at least held readlocked | |
851 | */ | |
852 | static struct vm_area_struct *find_vma_exact(struct mm_struct *mm, | |
853 | unsigned long addr, | |
854 | unsigned long len) | |
855 | { | |
856 | struct vm_area_struct *vma; | |
857 | struct rb_node *n = mm->mm_rb.rb_node; | |
858 | unsigned long end = addr + len; | |
859 | ||
860 | /* check the cache first */ | |
861 | vma = mm->mmap_cache; | |
862 | if (vma && vma->vm_start == addr && vma->vm_end == end) | |
863 | return vma; | |
864 | ||
865 | /* trawl the tree (there may be multiple mappings in which addr | |
866 | * resides) */ | |
867 | for (n = rb_first(&mm->mm_rb); n; n = rb_next(n)) { | |
868 | vma = rb_entry(n, struct vm_area_struct, vm_rb); | |
869 | if (vma->vm_start < addr) | |
870 | continue; | |
871 | if (vma->vm_start > addr) | |
872 | return NULL; | |
873 | if (vma->vm_end == end) { | |
874 | mm->mmap_cache = vma; | |
875 | return vma; | |
876 | } | |
877 | } | |
878 | ||
879 | return NULL; | |
1da177e4 LT |
880 | } |
881 | ||
882 | /* | |
883 | * determine whether a mapping should be permitted and, if so, what sort of | |
884 | * mapping we're capable of supporting | |
885 | */ | |
886 | static int validate_mmap_request(struct file *file, | |
887 | unsigned long addr, | |
888 | unsigned long len, | |
889 | unsigned long prot, | |
890 | unsigned long flags, | |
891 | unsigned long pgoff, | |
892 | unsigned long *_capabilities) | |
893 | { | |
8feae131 | 894 | unsigned long capabilities, rlen; |
1da177e4 LT |
895 | unsigned long reqprot = prot; |
896 | int ret; | |
897 | ||
898 | /* do the simple checks first */ | |
06aab5a3 | 899 | if (flags & MAP_FIXED) { |
1da177e4 LT |
900 | printk(KERN_DEBUG |
901 | "%d: Can't do fixed-address/overlay mmap of RAM\n", | |
902 | current->pid); | |
903 | return -EINVAL; | |
904 | } | |
905 | ||
906 | if ((flags & MAP_TYPE) != MAP_PRIVATE && | |
907 | (flags & MAP_TYPE) != MAP_SHARED) | |
908 | return -EINVAL; | |
909 | ||
f81cff0d | 910 | if (!len) |
1da177e4 LT |
911 | return -EINVAL; |
912 | ||
f81cff0d | 913 | /* Careful about overflows.. */ |
8feae131 DH |
914 | rlen = PAGE_ALIGN(len); |
915 | if (!rlen || rlen > TASK_SIZE) | |
f81cff0d MF |
916 | return -ENOMEM; |
917 | ||
1da177e4 | 918 | /* offset overflow? */ |
8feae131 | 919 | if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff) |
f81cff0d | 920 | return -EOVERFLOW; |
1da177e4 LT |
921 | |
922 | if (file) { | |
923 | /* validate file mapping requests */ | |
924 | struct address_space *mapping; | |
925 | ||
926 | /* files must support mmap */ | |
927 | if (!file->f_op || !file->f_op->mmap) | |
928 | return -ENODEV; | |
929 | ||
930 | /* work out if what we've got could possibly be shared | |
931 | * - we support chardevs that provide their own "memory" | |
932 | * - we support files/blockdevs that are memory backed | |
933 | */ | |
934 | mapping = file->f_mapping; | |
935 | if (!mapping) | |
e9536ae7 | 936 | mapping = file->f_path.dentry->d_inode->i_mapping; |
1da177e4 LT |
937 | |
938 | capabilities = 0; | |
939 | if (mapping && mapping->backing_dev_info) | |
940 | capabilities = mapping->backing_dev_info->capabilities; | |
941 | ||
942 | if (!capabilities) { | |
943 | /* no explicit capabilities set, so assume some | |
944 | * defaults */ | |
e9536ae7 | 945 | switch (file->f_path.dentry->d_inode->i_mode & S_IFMT) { |
1da177e4 LT |
946 | case S_IFREG: |
947 | case S_IFBLK: | |
948 | capabilities = BDI_CAP_MAP_COPY; | |
949 | break; | |
950 | ||
951 | case S_IFCHR: | |
952 | capabilities = | |
953 | BDI_CAP_MAP_DIRECT | | |
954 | BDI_CAP_READ_MAP | | |
955 | BDI_CAP_WRITE_MAP; | |
956 | break; | |
957 | ||
958 | default: | |
959 | return -EINVAL; | |
960 | } | |
961 | } | |
962 | ||
963 | /* eliminate any capabilities that we can't support on this | |
964 | * device */ | |
965 | if (!file->f_op->get_unmapped_area) | |
966 | capabilities &= ~BDI_CAP_MAP_DIRECT; | |
967 | if (!file->f_op->read) | |
968 | capabilities &= ~BDI_CAP_MAP_COPY; | |
969 | ||
28d7a6ae GY |
970 | /* The file shall have been opened with read permission. */ |
971 | if (!(file->f_mode & FMODE_READ)) | |
972 | return -EACCES; | |
973 | ||
1da177e4 LT |
974 | if (flags & MAP_SHARED) { |
975 | /* do checks for writing, appending and locking */ | |
976 | if ((prot & PROT_WRITE) && | |
977 | !(file->f_mode & FMODE_WRITE)) | |
978 | return -EACCES; | |
979 | ||
e9536ae7 | 980 | if (IS_APPEND(file->f_path.dentry->d_inode) && |
1da177e4 LT |
981 | (file->f_mode & FMODE_WRITE)) |
982 | return -EACCES; | |
983 | ||
e9536ae7 | 984 | if (locks_verify_locked(file->f_path.dentry->d_inode)) |
1da177e4 LT |
985 | return -EAGAIN; |
986 | ||
987 | if (!(capabilities & BDI_CAP_MAP_DIRECT)) | |
988 | return -ENODEV; | |
989 | ||
1da177e4 LT |
990 | /* we mustn't privatise shared mappings */ |
991 | capabilities &= ~BDI_CAP_MAP_COPY; | |
992 | } | |
993 | else { | |
994 | /* we're going to read the file into private memory we | |
995 | * allocate */ | |
996 | if (!(capabilities & BDI_CAP_MAP_COPY)) | |
997 | return -ENODEV; | |
998 | ||
999 | /* we don't permit a private writable mapping to be | |
1000 | * shared with the backing device */ | |
1001 | if (prot & PROT_WRITE) | |
1002 | capabilities &= ~BDI_CAP_MAP_DIRECT; | |
1003 | } | |
1004 | ||
3c7b2045 BS |
1005 | if (capabilities & BDI_CAP_MAP_DIRECT) { |
1006 | if (((prot & PROT_READ) && !(capabilities & BDI_CAP_READ_MAP)) || | |
1007 | ((prot & PROT_WRITE) && !(capabilities & BDI_CAP_WRITE_MAP)) || | |
1008 | ((prot & PROT_EXEC) && !(capabilities & BDI_CAP_EXEC_MAP)) | |
1009 | ) { | |
1010 | capabilities &= ~BDI_CAP_MAP_DIRECT; | |
1011 | if (flags & MAP_SHARED) { | |
1012 | printk(KERN_WARNING | |
1013 | "MAP_SHARED not completely supported on !MMU\n"); | |
1014 | return -EINVAL; | |
1015 | } | |
1016 | } | |
1017 | } | |
1018 | ||
1da177e4 LT |
1019 | /* handle executable mappings and implied executable |
1020 | * mappings */ | |
e9536ae7 | 1021 | if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) { |
1da177e4 LT |
1022 | if (prot & PROT_EXEC) |
1023 | return -EPERM; | |
1024 | } | |
1025 | else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) { | |
1026 | /* handle implication of PROT_EXEC by PROT_READ */ | |
1027 | if (current->personality & READ_IMPLIES_EXEC) { | |
1028 | if (capabilities & BDI_CAP_EXEC_MAP) | |
1029 | prot |= PROT_EXEC; | |
1030 | } | |
1031 | } | |
1032 | else if ((prot & PROT_READ) && | |
1033 | (prot & PROT_EXEC) && | |
1034 | !(capabilities & BDI_CAP_EXEC_MAP) | |
1035 | ) { | |
1036 | /* backing file is not executable, try to copy */ | |
1037 | capabilities &= ~BDI_CAP_MAP_DIRECT; | |
1038 | } | |
1039 | } | |
1040 | else { | |
1041 | /* anonymous mappings are always memory backed and can be | |
1042 | * privately mapped | |
1043 | */ | |
1044 | capabilities = BDI_CAP_MAP_COPY; | |
1045 | ||
1046 | /* handle PROT_EXEC implication by PROT_READ */ | |
1047 | if ((prot & PROT_READ) && | |
1048 | (current->personality & READ_IMPLIES_EXEC)) | |
1049 | prot |= PROT_EXEC; | |
1050 | } | |
1051 | ||
1052 | /* allow the security API to have its say */ | |
ed032189 | 1053 | ret = security_file_mmap(file, reqprot, prot, flags, addr, 0); |
1da177e4 LT |
1054 | if (ret < 0) |
1055 | return ret; | |
1056 | ||
1057 | /* looks okay */ | |
1058 | *_capabilities = capabilities; | |
1059 | return 0; | |
1060 | } | |
1061 | ||
1062 | /* | |
1063 | * we've determined that we can make the mapping, now translate what we | |
1064 | * now know into VMA flags | |
1065 | */ | |
1066 | static unsigned long determine_vm_flags(struct file *file, | |
1067 | unsigned long prot, | |
1068 | unsigned long flags, | |
1069 | unsigned long capabilities) | |
1070 | { | |
1071 | unsigned long vm_flags; | |
1072 | ||
1073 | vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags); | |
1da177e4 LT |
1074 | /* vm_flags |= mm->def_flags; */ |
1075 | ||
1076 | if (!(capabilities & BDI_CAP_MAP_DIRECT)) { | |
1077 | /* attempt to share read-only copies of mapped file chunks */ | |
3c7b2045 | 1078 | vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC; |
1da177e4 LT |
1079 | if (file && !(prot & PROT_WRITE)) |
1080 | vm_flags |= VM_MAYSHARE; | |
3c7b2045 | 1081 | } else { |
1da177e4 LT |
1082 | /* overlay a shareable mapping on the backing device or inode |
1083 | * if possible - used for chardevs, ramfs/tmpfs/shmfs and | |
1084 | * romfs/cramfs */ | |
3c7b2045 | 1085 | vm_flags |= VM_MAYSHARE | (capabilities & BDI_CAP_VMFLAGS); |
1da177e4 | 1086 | if (flags & MAP_SHARED) |
3c7b2045 | 1087 | vm_flags |= VM_SHARED; |
1da177e4 LT |
1088 | } |
1089 | ||
1090 | /* refuse to let anyone share private mappings with this process if | |
1091 | * it's being traced - otherwise breakpoints set in it may interfere | |
1092 | * with another untraced process | |
1093 | */ | |
fa8e26cc | 1094 | if ((flags & MAP_PRIVATE) && tracehook_expect_breakpoints(current)) |
1da177e4 LT |
1095 | vm_flags &= ~VM_MAYSHARE; |
1096 | ||
1097 | return vm_flags; | |
1098 | } | |
1099 | ||
1100 | /* | |
8feae131 DH |
1101 | * set up a shared mapping on a file (the driver or filesystem provides and |
1102 | * pins the storage) | |
1da177e4 | 1103 | */ |
8feae131 | 1104 | static int do_mmap_shared_file(struct vm_area_struct *vma) |
1da177e4 LT |
1105 | { |
1106 | int ret; | |
1107 | ||
1108 | ret = vma->vm_file->f_op->mmap(vma->vm_file, vma); | |
dd8632a1 PM |
1109 | if (ret == 0) { |
1110 | vma->vm_region->vm_top = vma->vm_region->vm_end; | |
645d83c5 | 1111 | return 0; |
dd8632a1 | 1112 | } |
1da177e4 LT |
1113 | if (ret != -ENOSYS) |
1114 | return ret; | |
1115 | ||
3fa30460 DH |
1116 | /* getting -ENOSYS indicates that direct mmap isn't possible (as |
1117 | * opposed to tried but failed) so we can only give a suitable error as | |
1118 | * it's not possible to make a private copy if MAP_SHARED was given */ | |
1da177e4 LT |
1119 | return -ENODEV; |
1120 | } | |
1121 | ||
1122 | /* | |
1123 | * set up a private mapping or an anonymous shared mapping | |
1124 | */ | |
8feae131 DH |
1125 | static int do_mmap_private(struct vm_area_struct *vma, |
1126 | struct vm_region *region, | |
645d83c5 DH |
1127 | unsigned long len, |
1128 | unsigned long capabilities) | |
1da177e4 | 1129 | { |
8feae131 DH |
1130 | struct page *pages; |
1131 | unsigned long total, point, n, rlen; | |
1da177e4 | 1132 | void *base; |
8feae131 | 1133 | int ret, order; |
1da177e4 LT |
1134 | |
1135 | /* invoke the file's mapping function so that it can keep track of | |
1136 | * shared mappings on devices or memory | |
1137 | * - VM_MAYSHARE will be set if it may attempt to share | |
1138 | */ | |
645d83c5 | 1139 | if (capabilities & BDI_CAP_MAP_DIRECT) { |
1da177e4 | 1140 | ret = vma->vm_file->f_op->mmap(vma->vm_file, vma); |
dd8632a1 | 1141 | if (ret == 0) { |
1da177e4 | 1142 | /* shouldn't return success if we're not sharing */ |
dd8632a1 PM |
1143 | BUG_ON(!(vma->vm_flags & VM_MAYSHARE)); |
1144 | vma->vm_region->vm_top = vma->vm_region->vm_end; | |
645d83c5 | 1145 | return 0; |
1da177e4 | 1146 | } |
dd8632a1 PM |
1147 | if (ret != -ENOSYS) |
1148 | return ret; | |
1da177e4 LT |
1149 | |
1150 | /* getting an ENOSYS error indicates that direct mmap isn't | |
1151 | * possible (as opposed to tried but failed) so we'll try to | |
1152 | * make a private copy of the data and map that instead */ | |
1153 | } | |
1154 | ||
8feae131 DH |
1155 | rlen = PAGE_ALIGN(len); |
1156 | ||
1da177e4 LT |
1157 | /* allocate some memory to hold the mapping |
1158 | * - note that this may not return a page-aligned address if the object | |
1159 | * we're allocating is smaller than a page | |
1160 | */ | |
8feae131 DH |
1161 | order = get_order(rlen); |
1162 | kdebug("alloc order %d for %lx", order, len); | |
1163 | ||
1164 | pages = alloc_pages(GFP_KERNEL, order); | |
1165 | if (!pages) | |
1da177e4 LT |
1166 | goto enomem; |
1167 | ||
8feae131 | 1168 | total = 1 << order; |
33e5d769 | 1169 | atomic_long_add(total, &mmap_pages_allocated); |
8feae131 DH |
1170 | |
1171 | point = rlen >> PAGE_SHIFT; | |
dd8632a1 PM |
1172 | |
1173 | /* we allocated a power-of-2 sized page set, so we may want to trim off | |
1174 | * the excess */ | |
1175 | if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages) { | |
1176 | while (total > point) { | |
1177 | order = ilog2(total - point); | |
1178 | n = 1 << order; | |
1179 | kdebug("shave %lu/%lu @%lu", n, total - point, total); | |
33e5d769 | 1180 | atomic_long_sub(n, &mmap_pages_allocated); |
dd8632a1 PM |
1181 | total -= n; |
1182 | set_page_refcounted(pages + total); | |
1183 | __free_pages(pages + total, order); | |
1184 | } | |
8feae131 DH |
1185 | } |
1186 | ||
8feae131 DH |
1187 | for (point = 1; point < total; point++) |
1188 | set_page_refcounted(&pages[point]); | |
1da177e4 | 1189 | |
8feae131 DH |
1190 | base = page_address(pages); |
1191 | region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY; | |
1192 | region->vm_start = (unsigned long) base; | |
1193 | region->vm_end = region->vm_start + rlen; | |
dd8632a1 | 1194 | region->vm_top = region->vm_start + (total << PAGE_SHIFT); |
8feae131 DH |
1195 | |
1196 | vma->vm_start = region->vm_start; | |
1197 | vma->vm_end = region->vm_start + len; | |
1da177e4 LT |
1198 | |
1199 | if (vma->vm_file) { | |
1200 | /* read the contents of a file into the copy */ | |
1201 | mm_segment_t old_fs; | |
1202 | loff_t fpos; | |
1203 | ||
1204 | fpos = vma->vm_pgoff; | |
1205 | fpos <<= PAGE_SHIFT; | |
1206 | ||
1207 | old_fs = get_fs(); | |
1208 | set_fs(KERNEL_DS); | |
8feae131 | 1209 | ret = vma->vm_file->f_op->read(vma->vm_file, base, rlen, &fpos); |
1da177e4 LT |
1210 | set_fs(old_fs); |
1211 | ||
1212 | if (ret < 0) | |
1213 | goto error_free; | |
1214 | ||
1215 | /* clear the last little bit */ | |
8feae131 DH |
1216 | if (ret < rlen) |
1217 | memset(base + ret, 0, rlen - ret); | |
1da177e4 | 1218 | |
1da177e4 LT |
1219 | } |
1220 | ||
1221 | return 0; | |
1222 | ||
1223 | error_free: | |
8feae131 DH |
1224 | free_page_series(region->vm_start, region->vm_end); |
1225 | region->vm_start = vma->vm_start = 0; | |
1226 | region->vm_end = vma->vm_end = 0; | |
dd8632a1 | 1227 | region->vm_top = 0; |
1da177e4 LT |
1228 | return ret; |
1229 | ||
1230 | enomem: | |
05ae6fa3 GU |
1231 | printk("Allocation of length %lu from process %d (%s) failed\n", |
1232 | len, current->pid, current->comm); | |
7bf02ea2 | 1233 | show_free_areas(0); |
1da177e4 LT |
1234 | return -ENOMEM; |
1235 | } | |
1236 | ||
1237 | /* | |
1238 | * handle mapping creation for uClinux | |
1239 | */ | |
1240 | unsigned long do_mmap_pgoff(struct file *file, | |
1241 | unsigned long addr, | |
1242 | unsigned long len, | |
1243 | unsigned long prot, | |
1244 | unsigned long flags, | |
1245 | unsigned long pgoff) | |
1246 | { | |
8feae131 DH |
1247 | struct vm_area_struct *vma; |
1248 | struct vm_region *region; | |
1da177e4 | 1249 | struct rb_node *rb; |
8feae131 | 1250 | unsigned long capabilities, vm_flags, result; |
1da177e4 LT |
1251 | int ret; |
1252 | ||
8feae131 DH |
1253 | kenter(",%lx,%lx,%lx,%lx,%lx", addr, len, prot, flags, pgoff); |
1254 | ||
1da177e4 LT |
1255 | /* decide whether we should attempt the mapping, and if so what sort of |
1256 | * mapping */ | |
1257 | ret = validate_mmap_request(file, addr, len, prot, flags, pgoff, | |
1258 | &capabilities); | |
8feae131 DH |
1259 | if (ret < 0) { |
1260 | kleave(" = %d [val]", ret); | |
1da177e4 | 1261 | return ret; |
8feae131 | 1262 | } |
1da177e4 | 1263 | |
06aab5a3 DH |
1264 | /* we ignore the address hint */ |
1265 | addr = 0; | |
1266 | ||
1da177e4 LT |
1267 | /* we've determined that we can make the mapping, now translate what we |
1268 | * now know into VMA flags */ | |
1269 | vm_flags = determine_vm_flags(file, prot, flags, capabilities); | |
1270 | ||
8feae131 DH |
1271 | /* we're going to need to record the mapping */ |
1272 | region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL); | |
1273 | if (!region) | |
1274 | goto error_getting_region; | |
1275 | ||
1276 | vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); | |
1277 | if (!vma) | |
1278 | goto error_getting_vma; | |
1da177e4 | 1279 | |
1e2ae599 | 1280 | region->vm_usage = 1; |
8feae131 DH |
1281 | region->vm_flags = vm_flags; |
1282 | region->vm_pgoff = pgoff; | |
1283 | ||
5beb4930 | 1284 | INIT_LIST_HEAD(&vma->anon_vma_chain); |
8feae131 DH |
1285 | vma->vm_flags = vm_flags; |
1286 | vma->vm_pgoff = pgoff; | |
1da177e4 | 1287 | |
8feae131 DH |
1288 | if (file) { |
1289 | region->vm_file = file; | |
1290 | get_file(file); | |
1291 | vma->vm_file = file; | |
1292 | get_file(file); | |
1293 | if (vm_flags & VM_EXECUTABLE) { | |
1294 | added_exe_file_vma(current->mm); | |
1295 | vma->vm_mm = current->mm; | |
1296 | } | |
1297 | } | |
1298 | ||
1299 | down_write(&nommu_region_sem); | |
1300 | ||
1301 | /* if we want to share, we need to check for regions created by other | |
1da177e4 | 1302 | * mmap() calls that overlap with our proposed mapping |
8feae131 | 1303 | * - we can only share with a superset match on most regular files |
1da177e4 LT |
1304 | * - shared mappings on character devices and memory backed files are |
1305 | * permitted to overlap inexactly as far as we are concerned for in | |
1306 | * these cases, sharing is handled in the driver or filesystem rather | |
1307 | * than here | |
1308 | */ | |
1309 | if (vm_flags & VM_MAYSHARE) { | |
8feae131 DH |
1310 | struct vm_region *pregion; |
1311 | unsigned long pglen, rpglen, pgend, rpgend, start; | |
1da177e4 | 1312 | |
8feae131 DH |
1313 | pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; |
1314 | pgend = pgoff + pglen; | |
165b2392 | 1315 | |
8feae131 DH |
1316 | for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) { |
1317 | pregion = rb_entry(rb, struct vm_region, vm_rb); | |
1da177e4 | 1318 | |
8feae131 | 1319 | if (!(pregion->vm_flags & VM_MAYSHARE)) |
1da177e4 LT |
1320 | continue; |
1321 | ||
1322 | /* search for overlapping mappings on the same file */ | |
8feae131 DH |
1323 | if (pregion->vm_file->f_path.dentry->d_inode != |
1324 | file->f_path.dentry->d_inode) | |
1da177e4 LT |
1325 | continue; |
1326 | ||
8feae131 | 1327 | if (pregion->vm_pgoff >= pgend) |
1da177e4 LT |
1328 | continue; |
1329 | ||
8feae131 DH |
1330 | rpglen = pregion->vm_end - pregion->vm_start; |
1331 | rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
1332 | rpgend = pregion->vm_pgoff + rpglen; | |
1333 | if (pgoff >= rpgend) | |
1da177e4 LT |
1334 | continue; |
1335 | ||
8feae131 DH |
1336 | /* handle inexactly overlapping matches between |
1337 | * mappings */ | |
1338 | if ((pregion->vm_pgoff != pgoff || rpglen != pglen) && | |
1339 | !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) { | |
1340 | /* new mapping is not a subset of the region */ | |
1da177e4 LT |
1341 | if (!(capabilities & BDI_CAP_MAP_DIRECT)) |
1342 | goto sharing_violation; | |
1343 | continue; | |
1344 | } | |
1345 | ||
8feae131 | 1346 | /* we've found a region we can share */ |
1e2ae599 | 1347 | pregion->vm_usage++; |
8feae131 DH |
1348 | vma->vm_region = pregion; |
1349 | start = pregion->vm_start; | |
1350 | start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT; | |
1351 | vma->vm_start = start; | |
1352 | vma->vm_end = start + len; | |
1353 | ||
1354 | if (pregion->vm_flags & VM_MAPPED_COPY) { | |
1355 | kdebug("share copy"); | |
1356 | vma->vm_flags |= VM_MAPPED_COPY; | |
1357 | } else { | |
1358 | kdebug("share mmap"); | |
1359 | ret = do_mmap_shared_file(vma); | |
1360 | if (ret < 0) { | |
1361 | vma->vm_region = NULL; | |
1362 | vma->vm_start = 0; | |
1363 | vma->vm_end = 0; | |
1e2ae599 | 1364 | pregion->vm_usage--; |
8feae131 DH |
1365 | pregion = NULL; |
1366 | goto error_just_free; | |
1367 | } | |
1368 | } | |
1369 | fput(region->vm_file); | |
1370 | kmem_cache_free(vm_region_jar, region); | |
1371 | region = pregion; | |
1372 | result = start; | |
1373 | goto share; | |
1da177e4 LT |
1374 | } |
1375 | ||
1da177e4 LT |
1376 | /* obtain the address at which to make a shared mapping |
1377 | * - this is the hook for quasi-memory character devices to | |
1378 | * tell us the location of a shared mapping | |
1379 | */ | |
645d83c5 | 1380 | if (capabilities & BDI_CAP_MAP_DIRECT) { |
1da177e4 LT |
1381 | addr = file->f_op->get_unmapped_area(file, addr, len, |
1382 | pgoff, flags); | |
1383 | if (IS_ERR((void *) addr)) { | |
1384 | ret = addr; | |
1385 | if (ret != (unsigned long) -ENOSYS) | |
8feae131 | 1386 | goto error_just_free; |
1da177e4 LT |
1387 | |
1388 | /* the driver refused to tell us where to site | |
1389 | * the mapping so we'll have to attempt to copy | |
1390 | * it */ | |
1391 | ret = (unsigned long) -ENODEV; | |
1392 | if (!(capabilities & BDI_CAP_MAP_COPY)) | |
8feae131 | 1393 | goto error_just_free; |
1da177e4 LT |
1394 | |
1395 | capabilities &= ~BDI_CAP_MAP_DIRECT; | |
8feae131 DH |
1396 | } else { |
1397 | vma->vm_start = region->vm_start = addr; | |
1398 | vma->vm_end = region->vm_end = addr + len; | |
1da177e4 LT |
1399 | } |
1400 | } | |
1401 | } | |
1402 | ||
8feae131 | 1403 | vma->vm_region = region; |
1da177e4 | 1404 | |
645d83c5 DH |
1405 | /* set up the mapping |
1406 | * - the region is filled in if BDI_CAP_MAP_DIRECT is still set | |
1407 | */ | |
1da177e4 | 1408 | if (file && vma->vm_flags & VM_SHARED) |
8feae131 | 1409 | ret = do_mmap_shared_file(vma); |
1da177e4 | 1410 | else |
645d83c5 | 1411 | ret = do_mmap_private(vma, region, len, capabilities); |
1da177e4 | 1412 | if (ret < 0) |
645d83c5 DH |
1413 | goto error_just_free; |
1414 | add_nommu_region(region); | |
8feae131 | 1415 | |
ea637639 JZ |
1416 | /* clear anonymous mappings that don't ask for uninitialized data */ |
1417 | if (!vma->vm_file && !(flags & MAP_UNINITIALIZED)) | |
1418 | memset((void *)region->vm_start, 0, | |
1419 | region->vm_end - region->vm_start); | |
1420 | ||
1da177e4 | 1421 | /* okay... we have a mapping; now we have to register it */ |
8feae131 | 1422 | result = vma->vm_start; |
1da177e4 | 1423 | |
1da177e4 LT |
1424 | current->mm->total_vm += len >> PAGE_SHIFT; |
1425 | ||
8feae131 DH |
1426 | share: |
1427 | add_vma_to_mm(current->mm, vma); | |
1da177e4 | 1428 | |
cfe79c00 MF |
1429 | /* we flush the region from the icache only when the first executable |
1430 | * mapping of it is made */ | |
1431 | if (vma->vm_flags & VM_EXEC && !region->vm_icache_flushed) { | |
1432 | flush_icache_range(region->vm_start, region->vm_end); | |
1433 | region->vm_icache_flushed = true; | |
1434 | } | |
1da177e4 | 1435 | |
cfe79c00 | 1436 | up_write(&nommu_region_sem); |
1da177e4 | 1437 | |
8feae131 DH |
1438 | kleave(" = %lx", result); |
1439 | return result; | |
1da177e4 | 1440 | |
8feae131 DH |
1441 | error_just_free: |
1442 | up_write(&nommu_region_sem); | |
1443 | error: | |
89a86402 DH |
1444 | if (region->vm_file) |
1445 | fput(region->vm_file); | |
8feae131 | 1446 | kmem_cache_free(vm_region_jar, region); |
89a86402 DH |
1447 | if (vma->vm_file) |
1448 | fput(vma->vm_file); | |
8feae131 DH |
1449 | if (vma->vm_flags & VM_EXECUTABLE) |
1450 | removed_exe_file_vma(vma->vm_mm); | |
1451 | kmem_cache_free(vm_area_cachep, vma); | |
1452 | kleave(" = %d", ret); | |
1453 | return ret; | |
1454 | ||
1455 | sharing_violation: | |
1456 | up_write(&nommu_region_sem); | |
1457 | printk(KERN_WARNING "Attempt to share mismatched mappings\n"); | |
1458 | ret = -EINVAL; | |
1459 | goto error; | |
1da177e4 | 1460 | |
8feae131 DH |
1461 | error_getting_vma: |
1462 | kmem_cache_free(vm_region_jar, region); | |
1463 | printk(KERN_WARNING "Allocation of vma for %lu byte allocation" | |
1464 | " from process %d failed\n", | |
1da177e4 | 1465 | len, current->pid); |
7bf02ea2 | 1466 | show_free_areas(0); |
1da177e4 LT |
1467 | return -ENOMEM; |
1468 | ||
8feae131 DH |
1469 | error_getting_region: |
1470 | printk(KERN_WARNING "Allocation of vm region for %lu byte allocation" | |
1471 | " from process %d failed\n", | |
1da177e4 | 1472 | len, current->pid); |
7bf02ea2 | 1473 | show_free_areas(0); |
1da177e4 LT |
1474 | return -ENOMEM; |
1475 | } | |
b5073173 | 1476 | EXPORT_SYMBOL(do_mmap_pgoff); |
1da177e4 | 1477 | |
66f0dc48 HD |
1478 | SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len, |
1479 | unsigned long, prot, unsigned long, flags, | |
1480 | unsigned long, fd, unsigned long, pgoff) | |
1481 | { | |
1482 | struct file *file = NULL; | |
1483 | unsigned long retval = -EBADF; | |
1484 | ||
120a795d | 1485 | audit_mmap_fd(fd, flags); |
66f0dc48 HD |
1486 | if (!(flags & MAP_ANONYMOUS)) { |
1487 | file = fget(fd); | |
1488 | if (!file) | |
1489 | goto out; | |
1490 | } | |
1491 | ||
1492 | flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE); | |
1493 | ||
1494 | down_write(¤t->mm->mmap_sem); | |
1495 | retval = do_mmap_pgoff(file, addr, len, prot, flags, pgoff); | |
1496 | up_write(¤t->mm->mmap_sem); | |
1497 | ||
1498 | if (file) | |
1499 | fput(file); | |
1500 | out: | |
1501 | return retval; | |
1502 | } | |
1503 | ||
a4679373 CH |
1504 | #ifdef __ARCH_WANT_SYS_OLD_MMAP |
1505 | struct mmap_arg_struct { | |
1506 | unsigned long addr; | |
1507 | unsigned long len; | |
1508 | unsigned long prot; | |
1509 | unsigned long flags; | |
1510 | unsigned long fd; | |
1511 | unsigned long offset; | |
1512 | }; | |
1513 | ||
1514 | SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg) | |
1515 | { | |
1516 | struct mmap_arg_struct a; | |
1517 | ||
1518 | if (copy_from_user(&a, arg, sizeof(a))) | |
1519 | return -EFAULT; | |
1520 | if (a.offset & ~PAGE_MASK) | |
1521 | return -EINVAL; | |
1522 | ||
1523 | return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd, | |
1524 | a.offset >> PAGE_SHIFT); | |
1525 | } | |
1526 | #endif /* __ARCH_WANT_SYS_OLD_MMAP */ | |
1527 | ||
1da177e4 | 1528 | /* |
8feae131 DH |
1529 | * split a vma into two pieces at address 'addr', a new vma is allocated either |
1530 | * for the first part or the tail. | |
1da177e4 | 1531 | */ |
8feae131 DH |
1532 | int split_vma(struct mm_struct *mm, struct vm_area_struct *vma, |
1533 | unsigned long addr, int new_below) | |
1da177e4 | 1534 | { |
8feae131 DH |
1535 | struct vm_area_struct *new; |
1536 | struct vm_region *region; | |
1537 | unsigned long npages; | |
1da177e4 | 1538 | |
8feae131 | 1539 | kenter(""); |
1da177e4 | 1540 | |
779c1023 DH |
1541 | /* we're only permitted to split anonymous regions (these should have |
1542 | * only a single usage on the region) */ | |
1543 | if (vma->vm_file) | |
8feae131 | 1544 | return -ENOMEM; |
1da177e4 | 1545 | |
8feae131 DH |
1546 | if (mm->map_count >= sysctl_max_map_count) |
1547 | return -ENOMEM; | |
1da177e4 | 1548 | |
8feae131 DH |
1549 | region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL); |
1550 | if (!region) | |
1551 | return -ENOMEM; | |
1da177e4 | 1552 | |
8feae131 DH |
1553 | new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); |
1554 | if (!new) { | |
1555 | kmem_cache_free(vm_region_jar, region); | |
1556 | return -ENOMEM; | |
1557 | } | |
1558 | ||
1559 | /* most fields are the same, copy all, and then fixup */ | |
1560 | *new = *vma; | |
1561 | *region = *vma->vm_region; | |
1562 | new->vm_region = region; | |
1563 | ||
1564 | npages = (addr - vma->vm_start) >> PAGE_SHIFT; | |
1565 | ||
1566 | if (new_below) { | |
dd8632a1 | 1567 | region->vm_top = region->vm_end = new->vm_end = addr; |
8feae131 DH |
1568 | } else { |
1569 | region->vm_start = new->vm_start = addr; | |
1570 | region->vm_pgoff = new->vm_pgoff += npages; | |
1da177e4 | 1571 | } |
8feae131 DH |
1572 | |
1573 | if (new->vm_ops && new->vm_ops->open) | |
1574 | new->vm_ops->open(new); | |
1575 | ||
1576 | delete_vma_from_mm(vma); | |
1577 | down_write(&nommu_region_sem); | |
1578 | delete_nommu_region(vma->vm_region); | |
1579 | if (new_below) { | |
1580 | vma->vm_region->vm_start = vma->vm_start = addr; | |
1581 | vma->vm_region->vm_pgoff = vma->vm_pgoff += npages; | |
1582 | } else { | |
1583 | vma->vm_region->vm_end = vma->vm_end = addr; | |
dd8632a1 | 1584 | vma->vm_region->vm_top = addr; |
8feae131 DH |
1585 | } |
1586 | add_nommu_region(vma->vm_region); | |
1587 | add_nommu_region(new->vm_region); | |
1588 | up_write(&nommu_region_sem); | |
1589 | add_vma_to_mm(mm, vma); | |
1590 | add_vma_to_mm(mm, new); | |
1591 | return 0; | |
1da177e4 LT |
1592 | } |
1593 | ||
3034097a | 1594 | /* |
8feae131 DH |
1595 | * shrink a VMA by removing the specified chunk from either the beginning or |
1596 | * the end | |
3034097a | 1597 | */ |
8feae131 DH |
1598 | static int shrink_vma(struct mm_struct *mm, |
1599 | struct vm_area_struct *vma, | |
1600 | unsigned long from, unsigned long to) | |
1da177e4 | 1601 | { |
8feae131 | 1602 | struct vm_region *region; |
1da177e4 | 1603 | |
8feae131 | 1604 | kenter(""); |
1da177e4 | 1605 | |
8feae131 DH |
1606 | /* adjust the VMA's pointers, which may reposition it in the MM's tree |
1607 | * and list */ | |
1608 | delete_vma_from_mm(vma); | |
1609 | if (from > vma->vm_start) | |
1610 | vma->vm_end = from; | |
1611 | else | |
1612 | vma->vm_start = to; | |
1613 | add_vma_to_mm(mm, vma); | |
1da177e4 | 1614 | |
8feae131 DH |
1615 | /* cut the backing region down to size */ |
1616 | region = vma->vm_region; | |
1e2ae599 | 1617 | BUG_ON(region->vm_usage != 1); |
8feae131 DH |
1618 | |
1619 | down_write(&nommu_region_sem); | |
1620 | delete_nommu_region(region); | |
dd8632a1 PM |
1621 | if (from > region->vm_start) { |
1622 | to = region->vm_top; | |
1623 | region->vm_top = region->vm_end = from; | |
1624 | } else { | |
8feae131 | 1625 | region->vm_start = to; |
dd8632a1 | 1626 | } |
8feae131 DH |
1627 | add_nommu_region(region); |
1628 | up_write(&nommu_region_sem); | |
1629 | ||
1630 | free_page_series(from, to); | |
1631 | return 0; | |
1632 | } | |
1da177e4 | 1633 | |
8feae131 DH |
1634 | /* |
1635 | * release a mapping | |
1636 | * - under NOMMU conditions the chunk to be unmapped must be backed by a single | |
1637 | * VMA, though it need not cover the whole VMA | |
1638 | */ | |
1639 | int do_munmap(struct mm_struct *mm, unsigned long start, size_t len) | |
1640 | { | |
1641 | struct vm_area_struct *vma; | |
1642 | struct rb_node *rb; | |
1643 | unsigned long end = start + len; | |
1644 | int ret; | |
1da177e4 | 1645 | |
8feae131 | 1646 | kenter(",%lx,%zx", start, len); |
1da177e4 | 1647 | |
8feae131 DH |
1648 | if (len == 0) |
1649 | return -EINVAL; | |
365e9c87 | 1650 | |
8feae131 DH |
1651 | /* find the first potentially overlapping VMA */ |
1652 | vma = find_vma(mm, start); | |
1653 | if (!vma) { | |
33e5d769 DH |
1654 | static int limit = 0; |
1655 | if (limit < 5) { | |
1656 | printk(KERN_WARNING | |
1657 | "munmap of memory not mmapped by process %d" | |
1658 | " (%s): 0x%lx-0x%lx\n", | |
1659 | current->pid, current->comm, | |
1660 | start, start + len - 1); | |
1661 | limit++; | |
1662 | } | |
8feae131 DH |
1663 | return -EINVAL; |
1664 | } | |
1da177e4 | 1665 | |
8feae131 DH |
1666 | /* we're allowed to split an anonymous VMA but not a file-backed one */ |
1667 | if (vma->vm_file) { | |
1668 | do { | |
1669 | if (start > vma->vm_start) { | |
1670 | kleave(" = -EINVAL [miss]"); | |
1671 | return -EINVAL; | |
1672 | } | |
1673 | if (end == vma->vm_end) | |
1674 | goto erase_whole_vma; | |
1675 | rb = rb_next(&vma->vm_rb); | |
1676 | vma = rb_entry(rb, struct vm_area_struct, vm_rb); | |
1677 | } while (rb); | |
1678 | kleave(" = -EINVAL [split file]"); | |
1679 | return -EINVAL; | |
1680 | } else { | |
1681 | /* the chunk must be a subset of the VMA found */ | |
1682 | if (start == vma->vm_start && end == vma->vm_end) | |
1683 | goto erase_whole_vma; | |
1684 | if (start < vma->vm_start || end > vma->vm_end) { | |
1685 | kleave(" = -EINVAL [superset]"); | |
1686 | return -EINVAL; | |
1687 | } | |
1688 | if (start & ~PAGE_MASK) { | |
1689 | kleave(" = -EINVAL [unaligned start]"); | |
1690 | return -EINVAL; | |
1691 | } | |
1692 | if (end != vma->vm_end && end & ~PAGE_MASK) { | |
1693 | kleave(" = -EINVAL [unaligned split]"); | |
1694 | return -EINVAL; | |
1695 | } | |
1696 | if (start != vma->vm_start && end != vma->vm_end) { | |
1697 | ret = split_vma(mm, vma, start, 1); | |
1698 | if (ret < 0) { | |
1699 | kleave(" = %d [split]", ret); | |
1700 | return ret; | |
1701 | } | |
1702 | } | |
1703 | return shrink_vma(mm, vma, start, end); | |
1704 | } | |
1da177e4 | 1705 | |
8feae131 DH |
1706 | erase_whole_vma: |
1707 | delete_vma_from_mm(vma); | |
1708 | delete_vma(mm, vma); | |
1709 | kleave(" = 0"); | |
1da177e4 LT |
1710 | return 0; |
1711 | } | |
b5073173 | 1712 | EXPORT_SYMBOL(do_munmap); |
1da177e4 | 1713 | |
6a6160a7 | 1714 | SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len) |
3034097a DH |
1715 | { |
1716 | int ret; | |
1717 | struct mm_struct *mm = current->mm; | |
1718 | ||
1719 | down_write(&mm->mmap_sem); | |
1720 | ret = do_munmap(mm, addr, len); | |
1721 | up_write(&mm->mmap_sem); | |
1722 | return ret; | |
1723 | } | |
1724 | ||
1725 | /* | |
8feae131 | 1726 | * release all the mappings made in a process's VM space |
3034097a | 1727 | */ |
8feae131 | 1728 | void exit_mmap(struct mm_struct *mm) |
1da177e4 | 1729 | { |
8feae131 | 1730 | struct vm_area_struct *vma; |
1da177e4 | 1731 | |
8feae131 DH |
1732 | if (!mm) |
1733 | return; | |
1da177e4 | 1734 | |
8feae131 | 1735 | kenter(""); |
1da177e4 | 1736 | |
8feae131 | 1737 | mm->total_vm = 0; |
1da177e4 | 1738 | |
8feae131 DH |
1739 | while ((vma = mm->mmap)) { |
1740 | mm->mmap = vma->vm_next; | |
1741 | delete_vma_from_mm(vma); | |
1742 | delete_vma(mm, vma); | |
04c34961 | 1743 | cond_resched(); |
1da177e4 | 1744 | } |
8feae131 DH |
1745 | |
1746 | kleave(""); | |
1da177e4 LT |
1747 | } |
1748 | ||
1da177e4 LT |
1749 | unsigned long do_brk(unsigned long addr, unsigned long len) |
1750 | { | |
1751 | return -ENOMEM; | |
1752 | } | |
1753 | ||
1754 | /* | |
6fa5f80b DH |
1755 | * expand (or shrink) an existing mapping, potentially moving it at the same |
1756 | * time (controlled by the MREMAP_MAYMOVE flag and available VM space) | |
1da177e4 | 1757 | * |
6fa5f80b | 1758 | * under NOMMU conditions, we only permit changing a mapping's size, and only |
8feae131 DH |
1759 | * as long as it stays within the region allocated by do_mmap_private() and the |
1760 | * block is not shareable | |
1da177e4 | 1761 | * |
6fa5f80b | 1762 | * MREMAP_FIXED is not supported under NOMMU conditions |
1da177e4 LT |
1763 | */ |
1764 | unsigned long do_mremap(unsigned long addr, | |
1765 | unsigned long old_len, unsigned long new_len, | |
1766 | unsigned long flags, unsigned long new_addr) | |
1767 | { | |
6fa5f80b | 1768 | struct vm_area_struct *vma; |
1da177e4 LT |
1769 | |
1770 | /* insanity checks first */ | |
8feae131 | 1771 | if (old_len == 0 || new_len == 0) |
1da177e4 LT |
1772 | return (unsigned long) -EINVAL; |
1773 | ||
8feae131 DH |
1774 | if (addr & ~PAGE_MASK) |
1775 | return -EINVAL; | |
1776 | ||
1da177e4 LT |
1777 | if (flags & MREMAP_FIXED && new_addr != addr) |
1778 | return (unsigned long) -EINVAL; | |
1779 | ||
8feae131 | 1780 | vma = find_vma_exact(current->mm, addr, old_len); |
6fa5f80b DH |
1781 | if (!vma) |
1782 | return (unsigned long) -EINVAL; | |
1da177e4 | 1783 | |
6fa5f80b | 1784 | if (vma->vm_end != vma->vm_start + old_len) |
1da177e4 LT |
1785 | return (unsigned long) -EFAULT; |
1786 | ||
6fa5f80b | 1787 | if (vma->vm_flags & VM_MAYSHARE) |
1da177e4 LT |
1788 | return (unsigned long) -EPERM; |
1789 | ||
8feae131 | 1790 | if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start) |
1da177e4 LT |
1791 | return (unsigned long) -ENOMEM; |
1792 | ||
1793 | /* all checks complete - do it */ | |
6fa5f80b | 1794 | vma->vm_end = vma->vm_start + new_len; |
6fa5f80b DH |
1795 | return vma->vm_start; |
1796 | } | |
b5073173 | 1797 | EXPORT_SYMBOL(do_mremap); |
6fa5f80b | 1798 | |
6a6160a7 HC |
1799 | SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len, |
1800 | unsigned long, new_len, unsigned long, flags, | |
1801 | unsigned long, new_addr) | |
6fa5f80b DH |
1802 | { |
1803 | unsigned long ret; | |
1804 | ||
1805 | down_write(¤t->mm->mmap_sem); | |
1806 | ret = do_mremap(addr, old_len, new_len, flags, new_addr); | |
1807 | up_write(¤t->mm->mmap_sem); | |
1808 | return ret; | |
1da177e4 LT |
1809 | } |
1810 | ||
6aab341e | 1811 | struct page *follow_page(struct vm_area_struct *vma, unsigned long address, |
deceb6cd | 1812 | unsigned int foll_flags) |
1da177e4 LT |
1813 | { |
1814 | return NULL; | |
1815 | } | |
1816 | ||
1da177e4 LT |
1817 | int remap_pfn_range(struct vm_area_struct *vma, unsigned long from, |
1818 | unsigned long to, unsigned long size, pgprot_t prot) | |
1819 | { | |
66aa2b4b GU |
1820 | vma->vm_start = vma->vm_pgoff << PAGE_SHIFT; |
1821 | return 0; | |
1da177e4 | 1822 | } |
22c4af40 | 1823 | EXPORT_SYMBOL(remap_pfn_range); |
1da177e4 | 1824 | |
f905bc44 PM |
1825 | int remap_vmalloc_range(struct vm_area_struct *vma, void *addr, |
1826 | unsigned long pgoff) | |
1827 | { | |
1828 | unsigned int size = vma->vm_end - vma->vm_start; | |
1829 | ||
1830 | if (!(vma->vm_flags & VM_USERMAP)) | |
1831 | return -EINVAL; | |
1832 | ||
1833 | vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT)); | |
1834 | vma->vm_end = vma->vm_start + size; | |
1835 | ||
1836 | return 0; | |
1837 | } | |
1838 | EXPORT_SYMBOL(remap_vmalloc_range); | |
1839 | ||
1da177e4 LT |
1840 | unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr, |
1841 | unsigned long len, unsigned long pgoff, unsigned long flags) | |
1842 | { | |
1843 | return -ENOMEM; | |
1844 | } | |
1845 | ||
1363c3cd | 1846 | void arch_unmap_area(struct mm_struct *mm, unsigned long addr) |
1da177e4 LT |
1847 | { |
1848 | } | |
1849 | ||
1da177e4 LT |
1850 | void unmap_mapping_range(struct address_space *mapping, |
1851 | loff_t const holebegin, loff_t const holelen, | |
1852 | int even_cows) | |
1853 | { | |
1854 | } | |
22c4af40 | 1855 | EXPORT_SYMBOL(unmap_mapping_range); |
1da177e4 LT |
1856 | |
1857 | /* | |
1858 | * Check that a process has enough memory to allocate a new virtual | |
1859 | * mapping. 0 means there is enough memory for the allocation to | |
1860 | * succeed and -ENOMEM implies there is not. | |
1861 | * | |
1862 | * We currently support three overcommit policies, which are set via the | |
1863 | * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting | |
1864 | * | |
1865 | * Strict overcommit modes added 2002 Feb 26 by Alan Cox. | |
1866 | * Additional code 2002 Jul 20 by Robert Love. | |
1867 | * | |
1868 | * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise. | |
1869 | * | |
1870 | * Note this is a helper function intended to be used by LSMs which | |
1871 | * wish to use this logic. | |
1872 | */ | |
34b4e4aa | 1873 | int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin) |
1da177e4 LT |
1874 | { |
1875 | unsigned long free, allowed; | |
1876 | ||
1877 | vm_acct_memory(pages); | |
1878 | ||
1879 | /* | |
1880 | * Sometimes we want to use more memory than we have | |
1881 | */ | |
1882 | if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS) | |
1883 | return 0; | |
1884 | ||
1885 | if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) { | |
1886 | unsigned long n; | |
1887 | ||
347ce434 | 1888 | free = global_page_state(NR_FILE_PAGES); |
1da177e4 LT |
1889 | free += nr_swap_pages; |
1890 | ||
1891 | /* | |
1892 | * Any slabs which are created with the | |
1893 | * SLAB_RECLAIM_ACCOUNT flag claim to have contents | |
1894 | * which are reclaimable, under pressure. The dentry | |
1895 | * cache and most inode caches should fall into this | |
1896 | */ | |
972d1a7b | 1897 | free += global_page_state(NR_SLAB_RECLAIMABLE); |
1da177e4 LT |
1898 | |
1899 | /* | |
1900 | * Leave the last 3% for root | |
1901 | */ | |
1902 | if (!cap_sys_admin) | |
1903 | free -= free / 32; | |
1904 | ||
1905 | if (free > pages) | |
1906 | return 0; | |
1907 | ||
1908 | /* | |
1909 | * nr_free_pages() is very expensive on large systems, | |
1910 | * only call if we're about to fail. | |
1911 | */ | |
1912 | n = nr_free_pages(); | |
d5ddc79b HA |
1913 | |
1914 | /* | |
1915 | * Leave reserved pages. The pages are not for anonymous pages. | |
1916 | */ | |
1917 | if (n <= totalreserve_pages) | |
1918 | goto error; | |
1919 | else | |
1920 | n -= totalreserve_pages; | |
1921 | ||
1922 | /* | |
1923 | * Leave the last 3% for root | |
1924 | */ | |
1da177e4 LT |
1925 | if (!cap_sys_admin) |
1926 | n -= n / 32; | |
1927 | free += n; | |
1928 | ||
1929 | if (free > pages) | |
1930 | return 0; | |
d5ddc79b HA |
1931 | |
1932 | goto error; | |
1da177e4 LT |
1933 | } |
1934 | ||
1935 | allowed = totalram_pages * sysctl_overcommit_ratio / 100; | |
1936 | /* | |
1937 | * Leave the last 3% for root | |
1938 | */ | |
1939 | if (!cap_sys_admin) | |
1940 | allowed -= allowed / 32; | |
1941 | allowed += total_swap_pages; | |
1942 | ||
1943 | /* Don't let a single process grow too big: | |
1944 | leave 3% of the size of this process for other processes */ | |
731572d3 AC |
1945 | if (mm) |
1946 | allowed -= mm->total_vm / 32; | |
1da177e4 | 1947 | |
00a62ce9 | 1948 | if (percpu_counter_read_positive(&vm_committed_as) < allowed) |
1da177e4 | 1949 | return 0; |
00a62ce9 | 1950 | |
d5ddc79b | 1951 | error: |
1da177e4 LT |
1952 | vm_unacct_memory(pages); |
1953 | ||
1954 | return -ENOMEM; | |
1955 | } | |
1956 | ||
cae5d390 | 1957 | int in_gate_area_no_mm(unsigned long addr) |
1da177e4 LT |
1958 | { |
1959 | return 0; | |
1960 | } | |
b0e15190 | 1961 | |
d0217ac0 | 1962 | int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf) |
b0e15190 DH |
1963 | { |
1964 | BUG(); | |
d0217ac0 | 1965 | return 0; |
b0e15190 | 1966 | } |
b5073173 | 1967 | EXPORT_SYMBOL(filemap_fault); |
0ec76a11 | 1968 | |
f55f199b MF |
1969 | static int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm, |
1970 | unsigned long addr, void *buf, int len, int write) | |
0ec76a11 | 1971 | { |
0ec76a11 | 1972 | struct vm_area_struct *vma; |
0ec76a11 DH |
1973 | |
1974 | down_read(&mm->mmap_sem); | |
1975 | ||
1976 | /* the access must start within one of the target process's mappings */ | |
0159b141 DH |
1977 | vma = find_vma(mm, addr); |
1978 | if (vma) { | |
0ec76a11 DH |
1979 | /* don't overrun this mapping */ |
1980 | if (addr + len >= vma->vm_end) | |
1981 | len = vma->vm_end - addr; | |
1982 | ||
1983 | /* only read or write mappings where it is permitted */ | |
d00c7b99 | 1984 | if (write && vma->vm_flags & VM_MAYWRITE) |
7959722b JZ |
1985 | copy_to_user_page(vma, NULL, addr, |
1986 | (void *) addr, buf, len); | |
d00c7b99 | 1987 | else if (!write && vma->vm_flags & VM_MAYREAD) |
7959722b JZ |
1988 | copy_from_user_page(vma, NULL, addr, |
1989 | buf, (void *) addr, len); | |
0ec76a11 DH |
1990 | else |
1991 | len = 0; | |
1992 | } else { | |
1993 | len = 0; | |
1994 | } | |
1995 | ||
1996 | up_read(&mm->mmap_sem); | |
f55f199b MF |
1997 | |
1998 | return len; | |
1999 | } | |
2000 | ||
2001 | /** | |
2002 | * @access_remote_vm - access another process' address space | |
2003 | * @mm: the mm_struct of the target address space | |
2004 | * @addr: start address to access | |
2005 | * @buf: source or destination buffer | |
2006 | * @len: number of bytes to transfer | |
2007 | * @write: whether the access is a write | |
2008 | * | |
2009 | * The caller must hold a reference on @mm. | |
2010 | */ | |
2011 | int access_remote_vm(struct mm_struct *mm, unsigned long addr, | |
2012 | void *buf, int len, int write) | |
2013 | { | |
2014 | return __access_remote_vm(NULL, mm, addr, buf, len, write); | |
2015 | } | |
2016 | ||
2017 | /* | |
2018 | * Access another process' address space. | |
2019 | * - source/target buffer must be kernel space | |
2020 | */ | |
2021 | int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write) | |
2022 | { | |
2023 | struct mm_struct *mm; | |
2024 | ||
2025 | if (addr + len < addr) | |
2026 | return 0; | |
2027 | ||
2028 | mm = get_task_mm(tsk); | |
2029 | if (!mm) | |
2030 | return 0; | |
2031 | ||
2032 | len = __access_remote_vm(tsk, mm, addr, buf, len, write); | |
2033 | ||
0ec76a11 DH |
2034 | mmput(mm); |
2035 | return len; | |
2036 | } | |
7e660872 DH |
2037 | |
2038 | /** | |
2039 | * nommu_shrink_inode_mappings - Shrink the shared mappings on an inode | |
2040 | * @inode: The inode to check | |
2041 | * @size: The current filesize of the inode | |
2042 | * @newsize: The proposed filesize of the inode | |
2043 | * | |
2044 | * Check the shared mappings on an inode on behalf of a shrinking truncate to | |
2045 | * make sure that that any outstanding VMAs aren't broken and then shrink the | |
2046 | * vm_regions that extend that beyond so that do_mmap_pgoff() doesn't | |
2047 | * automatically grant mappings that are too large. | |
2048 | */ | |
2049 | int nommu_shrink_inode_mappings(struct inode *inode, size_t size, | |
2050 | size_t newsize) | |
2051 | { | |
2052 | struct vm_area_struct *vma; | |
2053 | struct prio_tree_iter iter; | |
2054 | struct vm_region *region; | |
2055 | pgoff_t low, high; | |
2056 | size_t r_size, r_top; | |
2057 | ||
2058 | low = newsize >> PAGE_SHIFT; | |
2059 | high = (size + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
2060 | ||
2061 | down_write(&nommu_region_sem); | |
2062 | ||
2063 | /* search for VMAs that fall within the dead zone */ | |
2064 | vma_prio_tree_foreach(vma, &iter, &inode->i_mapping->i_mmap, | |
2065 | low, high) { | |
2066 | /* found one - only interested if it's shared out of the page | |
2067 | * cache */ | |
2068 | if (vma->vm_flags & VM_SHARED) { | |
2069 | up_write(&nommu_region_sem); | |
2070 | return -ETXTBSY; /* not quite true, but near enough */ | |
2071 | } | |
2072 | } | |
2073 | ||
2074 | /* reduce any regions that overlap the dead zone - if in existence, | |
2075 | * these will be pointed to by VMAs that don't overlap the dead zone | |
2076 | * | |
2077 | * we don't check for any regions that start beyond the EOF as there | |
2078 | * shouldn't be any | |
2079 | */ | |
2080 | vma_prio_tree_foreach(vma, &iter, &inode->i_mapping->i_mmap, | |
2081 | 0, ULONG_MAX) { | |
2082 | if (!(vma->vm_flags & VM_SHARED)) | |
2083 | continue; | |
2084 | ||
2085 | region = vma->vm_region; | |
2086 | r_size = region->vm_top - region->vm_start; | |
2087 | r_top = (region->vm_pgoff << PAGE_SHIFT) + r_size; | |
2088 | ||
2089 | if (r_top > newsize) { | |
2090 | region->vm_top -= r_top - newsize; | |
2091 | if (region->vm_end > region->vm_top) | |
2092 | region->vm_end = region->vm_top; | |
2093 | } | |
2094 | } | |
2095 | ||
2096 | up_write(&nommu_region_sem); | |
2097 | return 0; | |
2098 | } |