]>
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 | ||
b1de0d13 MH |
16 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
17 | ||
b95f1b31 | 18 | #include <linux/export.h> |
1da177e4 | 19 | #include <linux/mm.h> |
6e84f315 | 20 | #include <linux/sched/mm.h> |
615d6e87 | 21 | #include <linux/vmacache.h> |
1da177e4 LT |
22 | #include <linux/mman.h> |
23 | #include <linux/swap.h> | |
24 | #include <linux/file.h> | |
25 | #include <linux/highmem.h> | |
26 | #include <linux/pagemap.h> | |
27 | #include <linux/slab.h> | |
28 | #include <linux/vmalloc.h> | |
1da177e4 LT |
29 | #include <linux/blkdev.h> |
30 | #include <linux/backing-dev.h> | |
3b32123d | 31 | #include <linux/compiler.h> |
1da177e4 LT |
32 | #include <linux/mount.h> |
33 | #include <linux/personality.h> | |
34 | #include <linux/security.h> | |
35 | #include <linux/syscalls.h> | |
120a795d | 36 | #include <linux/audit.h> |
b1de0d13 | 37 | #include <linux/printk.h> |
1da177e4 | 38 | |
7c0f6ba6 | 39 | #include <linux/uaccess.h> |
1da177e4 LT |
40 | #include <asm/tlb.h> |
41 | #include <asm/tlbflush.h> | |
eb8cdec4 | 42 | #include <asm/mmu_context.h> |
8feae131 DH |
43 | #include "internal.h" |
44 | ||
1da177e4 | 45 | void *high_memory; |
944b6874 | 46 | EXPORT_SYMBOL(high_memory); |
1da177e4 LT |
47 | struct page *mem_map; |
48 | unsigned long max_mapnr; | |
5b8bf307 | 49 | EXPORT_SYMBOL(max_mapnr); |
4266c97a | 50 | unsigned long highest_memmap_pfn; |
fc4d5c29 | 51 | int sysctl_nr_trim_pages = CONFIG_NOMMU_INITIAL_TRIM_EXCESS; |
1da177e4 LT |
52 | int heap_stack_gap = 0; |
53 | ||
33e5d769 | 54 | atomic_long_t mmap_pages_allocated; |
8feae131 | 55 | |
1da177e4 | 56 | EXPORT_SYMBOL(mem_map); |
1da177e4 | 57 | |
8feae131 DH |
58 | /* list of mapped, potentially shareable regions */ |
59 | static struct kmem_cache *vm_region_jar; | |
60 | struct rb_root nommu_region_tree = RB_ROOT; | |
61 | DECLARE_RWSEM(nommu_region_sem); | |
1da177e4 | 62 | |
f0f37e2f | 63 | const struct vm_operations_struct generic_file_vm_ops = { |
1da177e4 LT |
64 | }; |
65 | ||
1da177e4 LT |
66 | /* |
67 | * Return the total memory allocated for this pointer, not | |
68 | * just what the caller asked for. | |
69 | * | |
70 | * Doesn't have to be accurate, i.e. may have races. | |
71 | */ | |
72 | unsigned int kobjsize(const void *objp) | |
73 | { | |
74 | struct page *page; | |
75 | ||
4016a139 MH |
76 | /* |
77 | * If the object we have should not have ksize performed on it, | |
78 | * return size of 0 | |
79 | */ | |
5a1603be | 80 | if (!objp || !virt_addr_valid(objp)) |
6cfd53fc PM |
81 | return 0; |
82 | ||
83 | page = virt_to_head_page(objp); | |
6cfd53fc PM |
84 | |
85 | /* | |
86 | * If the allocator sets PageSlab, we know the pointer came from | |
87 | * kmalloc(). | |
88 | */ | |
1da177e4 LT |
89 | if (PageSlab(page)) |
90 | return ksize(objp); | |
91 | ||
ab2e83ea PM |
92 | /* |
93 | * If it's not a compound page, see if we have a matching VMA | |
94 | * region. This test is intentionally done in reverse order, | |
95 | * so if there's no VMA, we still fall through and hand back | |
96 | * PAGE_SIZE for 0-order pages. | |
97 | */ | |
98 | if (!PageCompound(page)) { | |
99 | struct vm_area_struct *vma; | |
100 | ||
101 | vma = find_vma(current->mm, (unsigned long)objp); | |
102 | if (vma) | |
103 | return vma->vm_end - vma->vm_start; | |
104 | } | |
105 | ||
6cfd53fc PM |
106 | /* |
107 | * The ksize() function is only guaranteed to work for pointers | |
5a1603be | 108 | * returned by kmalloc(). So handle arbitrary pointers here. |
6cfd53fc | 109 | */ |
5a1603be | 110 | return PAGE_SIZE << compound_order(page); |
1da177e4 LT |
111 | } |
112 | ||
0d731759 | 113 | static long __get_user_pages(struct task_struct *tsk, struct mm_struct *mm, |
28a35716 ML |
114 | unsigned long start, unsigned long nr_pages, |
115 | unsigned int foll_flags, struct page **pages, | |
116 | struct vm_area_struct **vmas, int *nonblocking) | |
1da177e4 | 117 | { |
910e46da | 118 | struct vm_area_struct *vma; |
7b4d5b8b DH |
119 | unsigned long vm_flags; |
120 | int i; | |
121 | ||
122 | /* calculate required read or write permissions. | |
58fa879e | 123 | * If FOLL_FORCE is set, we only require the "MAY" flags. |
7b4d5b8b | 124 | */ |
58fa879e HD |
125 | vm_flags = (foll_flags & FOLL_WRITE) ? |
126 | (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD); | |
127 | vm_flags &= (foll_flags & FOLL_FORCE) ? | |
128 | (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE); | |
1da177e4 | 129 | |
9d73777e | 130 | for (i = 0; i < nr_pages; i++) { |
7561e8ca | 131 | vma = find_vma(mm, start); |
7b4d5b8b DH |
132 | if (!vma) |
133 | goto finish_or_fault; | |
134 | ||
135 | /* protect what we can, including chardevs */ | |
1c3aff1c HD |
136 | if ((vma->vm_flags & (VM_IO | VM_PFNMAP)) || |
137 | !(vm_flags & vma->vm_flags)) | |
7b4d5b8b | 138 | goto finish_or_fault; |
910e46da | 139 | |
1da177e4 LT |
140 | if (pages) { |
141 | pages[i] = virt_to_page(start); | |
142 | if (pages[i]) | |
09cbfeaf | 143 | get_page(pages[i]); |
1da177e4 LT |
144 | } |
145 | if (vmas) | |
910e46da | 146 | vmas[i] = vma; |
e1ee65d8 | 147 | start = (start + PAGE_SIZE) & PAGE_MASK; |
1da177e4 | 148 | } |
7b4d5b8b DH |
149 | |
150 | return i; | |
151 | ||
152 | finish_or_fault: | |
153 | return i ? : -EFAULT; | |
1da177e4 | 154 | } |
b291f000 | 155 | |
b291f000 NP |
156 | /* |
157 | * get a list of pages in an address range belonging to the specified process | |
158 | * and indicate the VMA that covers each page | |
159 | * - this is potentially dodgy as we may end incrementing the page count of a | |
160 | * slab page or a secondary page from a compound page | |
161 | * - don't permit access to VMAs that don't support it, such as I/O mappings | |
162 | */ | |
c12d2da5 | 163 | long get_user_pages(unsigned long start, unsigned long nr_pages, |
768ae309 | 164 | unsigned int gup_flags, struct page **pages, |
28a35716 | 165 | struct vm_area_struct **vmas) |
b291f000 | 166 | { |
768ae309 LS |
167 | return __get_user_pages(current, current->mm, start, nr_pages, |
168 | gup_flags, pages, vmas, NULL); | |
b291f000 | 169 | } |
c12d2da5 | 170 | EXPORT_SYMBOL(get_user_pages); |
66aa2b4b | 171 | |
c12d2da5 | 172 | long get_user_pages_locked(unsigned long start, unsigned long nr_pages, |
3b913179 | 173 | unsigned int gup_flags, struct page **pages, |
cde70140 | 174 | int *locked) |
f0818f47 | 175 | { |
768ae309 | 176 | return get_user_pages(start, nr_pages, gup_flags, pages, NULL); |
f0818f47 | 177 | } |
c12d2da5 | 178 | EXPORT_SYMBOL(get_user_pages_locked); |
f0818f47 | 179 | |
8b7457ef LS |
180 | static long __get_user_pages_unlocked(struct task_struct *tsk, |
181 | struct mm_struct *mm, unsigned long start, | |
182 | unsigned long nr_pages, struct page **pages, | |
183 | unsigned int gup_flags) | |
f0818f47 AA |
184 | { |
185 | long ret; | |
186 | down_read(&mm->mmap_sem); | |
cde70140 DH |
187 | ret = __get_user_pages(tsk, mm, start, nr_pages, gup_flags, pages, |
188 | NULL, NULL); | |
f0818f47 AA |
189 | up_read(&mm->mmap_sem); |
190 | return ret; | |
191 | } | |
0fd71a56 | 192 | |
c12d2da5 | 193 | long get_user_pages_unlocked(unsigned long start, unsigned long nr_pages, |
c164154f | 194 | struct page **pages, unsigned int gup_flags) |
0fd71a56 | 195 | { |
cde70140 | 196 | return __get_user_pages_unlocked(current, current->mm, start, nr_pages, |
c164154f | 197 | pages, gup_flags); |
0fd71a56 | 198 | } |
c12d2da5 | 199 | EXPORT_SYMBOL(get_user_pages_unlocked); |
f0818f47 | 200 | |
dfc2f91a PM |
201 | /** |
202 | * follow_pfn - look up PFN at a user virtual address | |
203 | * @vma: memory mapping | |
204 | * @address: user virtual address | |
205 | * @pfn: location to store found PFN | |
206 | * | |
207 | * Only IO mappings and raw PFN mappings are allowed. | |
208 | * | |
209 | * Returns zero and the pfn at @pfn on success, -ve otherwise. | |
210 | */ | |
211 | int follow_pfn(struct vm_area_struct *vma, unsigned long address, | |
212 | unsigned long *pfn) | |
213 | { | |
214 | if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) | |
215 | return -EINVAL; | |
216 | ||
217 | *pfn = address >> PAGE_SHIFT; | |
218 | return 0; | |
219 | } | |
220 | EXPORT_SYMBOL(follow_pfn); | |
221 | ||
f1c4069e | 222 | LIST_HEAD(vmap_area_list); |
1da177e4 | 223 | |
b3bdda02 | 224 | void vfree(const void *addr) |
1da177e4 LT |
225 | { |
226 | kfree(addr); | |
227 | } | |
b5073173 | 228 | EXPORT_SYMBOL(vfree); |
1da177e4 | 229 | |
dd0fc66f | 230 | void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot) |
1da177e4 LT |
231 | { |
232 | /* | |
8518609d RD |
233 | * You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc() |
234 | * returns only a logical address. | |
1da177e4 | 235 | */ |
84097518 | 236 | return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM); |
1da177e4 | 237 | } |
b5073173 | 238 | EXPORT_SYMBOL(__vmalloc); |
1da177e4 | 239 | |
f905bc44 PM |
240 | void *vmalloc_user(unsigned long size) |
241 | { | |
242 | void *ret; | |
243 | ||
244 | ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, | |
245 | PAGE_KERNEL); | |
246 | if (ret) { | |
247 | struct vm_area_struct *vma; | |
248 | ||
249 | down_write(¤t->mm->mmap_sem); | |
250 | vma = find_vma(current->mm, (unsigned long)ret); | |
251 | if (vma) | |
252 | vma->vm_flags |= VM_USERMAP; | |
253 | up_write(¤t->mm->mmap_sem); | |
254 | } | |
255 | ||
256 | return ret; | |
257 | } | |
258 | EXPORT_SYMBOL(vmalloc_user); | |
259 | ||
b3bdda02 | 260 | struct page *vmalloc_to_page(const void *addr) |
1da177e4 LT |
261 | { |
262 | return virt_to_page(addr); | |
263 | } | |
b5073173 | 264 | EXPORT_SYMBOL(vmalloc_to_page); |
1da177e4 | 265 | |
b3bdda02 | 266 | unsigned long vmalloc_to_pfn(const void *addr) |
1da177e4 LT |
267 | { |
268 | return page_to_pfn(virt_to_page(addr)); | |
269 | } | |
b5073173 | 270 | EXPORT_SYMBOL(vmalloc_to_pfn); |
1da177e4 LT |
271 | |
272 | long vread(char *buf, char *addr, unsigned long count) | |
273 | { | |
9bde916b CG |
274 | /* Don't allow overflow */ |
275 | if ((unsigned long) buf + count < count) | |
276 | count = -(unsigned long) buf; | |
277 | ||
1da177e4 LT |
278 | memcpy(buf, addr, count); |
279 | return count; | |
280 | } | |
281 | ||
282 | long vwrite(char *buf, char *addr, unsigned long count) | |
283 | { | |
284 | /* Don't allow overflow */ | |
285 | if ((unsigned long) addr + count < count) | |
286 | count = -(unsigned long) addr; | |
287 | ||
288 | memcpy(addr, buf, count); | |
ac714904 | 289 | return count; |
1da177e4 LT |
290 | } |
291 | ||
292 | /* | |
e1c05067 | 293 | * vmalloc - allocate virtually contiguous memory |
1da177e4 LT |
294 | * |
295 | * @size: allocation size | |
296 | * | |
297 | * Allocate enough pages to cover @size from the page level | |
e1c05067 | 298 | * allocator and map them into contiguous kernel virtual space. |
1da177e4 | 299 | * |
c1c8897f | 300 | * For tight control over page level allocator and protection flags |
1da177e4 LT |
301 | * use __vmalloc() instead. |
302 | */ | |
303 | void *vmalloc(unsigned long size) | |
304 | { | |
305 | return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL); | |
306 | } | |
f6138882 AM |
307 | EXPORT_SYMBOL(vmalloc); |
308 | ||
e1ca7788 | 309 | /* |
e1c05067 | 310 | * vzalloc - allocate virtually contiguous memory with zero fill |
e1ca7788 DY |
311 | * |
312 | * @size: allocation size | |
313 | * | |
314 | * Allocate enough pages to cover @size from the page level | |
e1c05067 | 315 | * allocator and map them into contiguous kernel virtual space. |
e1ca7788 DY |
316 | * The memory allocated is set to zero. |
317 | * | |
318 | * For tight control over page level allocator and protection flags | |
319 | * use __vmalloc() instead. | |
320 | */ | |
321 | void *vzalloc(unsigned long size) | |
322 | { | |
323 | return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, | |
324 | PAGE_KERNEL); | |
325 | } | |
326 | EXPORT_SYMBOL(vzalloc); | |
327 | ||
328 | /** | |
329 | * vmalloc_node - allocate memory on a specific node | |
330 | * @size: allocation size | |
331 | * @node: numa node | |
332 | * | |
333 | * Allocate enough pages to cover @size from the page level | |
334 | * allocator and map them into contiguous kernel virtual space. | |
335 | * | |
336 | * For tight control over page level allocator and protection flags | |
337 | * use __vmalloc() instead. | |
338 | */ | |
f6138882 AM |
339 | void *vmalloc_node(unsigned long size, int node) |
340 | { | |
341 | return vmalloc(size); | |
342 | } | |
9a14f653 | 343 | EXPORT_SYMBOL(vmalloc_node); |
e1ca7788 DY |
344 | |
345 | /** | |
346 | * vzalloc_node - allocate memory on a specific node with zero fill | |
347 | * @size: allocation size | |
348 | * @node: numa node | |
349 | * | |
350 | * Allocate enough pages to cover @size from the page level | |
351 | * allocator and map them into contiguous kernel virtual space. | |
352 | * The memory allocated is set to zero. | |
353 | * | |
354 | * For tight control over page level allocator and protection flags | |
355 | * use __vmalloc() instead. | |
356 | */ | |
357 | void *vzalloc_node(unsigned long size, int node) | |
358 | { | |
359 | return vzalloc(size); | |
360 | } | |
361 | EXPORT_SYMBOL(vzalloc_node); | |
1da177e4 | 362 | |
1af446ed PM |
363 | #ifndef PAGE_KERNEL_EXEC |
364 | # define PAGE_KERNEL_EXEC PAGE_KERNEL | |
365 | #endif | |
366 | ||
367 | /** | |
368 | * vmalloc_exec - allocate virtually contiguous, executable memory | |
369 | * @size: allocation size | |
370 | * | |
371 | * Kernel-internal function to allocate enough pages to cover @size | |
372 | * the page level allocator and map them into contiguous and | |
373 | * executable kernel virtual space. | |
374 | * | |
375 | * For tight control over page level allocator and protection flags | |
376 | * use __vmalloc() instead. | |
377 | */ | |
378 | ||
379 | void *vmalloc_exec(unsigned long size) | |
380 | { | |
381 | return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC); | |
382 | } | |
383 | ||
b5073173 PM |
384 | /** |
385 | * vmalloc_32 - allocate virtually contiguous memory (32bit addressable) | |
1da177e4 LT |
386 | * @size: allocation size |
387 | * | |
388 | * Allocate enough 32bit PA addressable pages to cover @size from the | |
e1c05067 | 389 | * page level allocator and map them into contiguous kernel virtual space. |
1da177e4 LT |
390 | */ |
391 | void *vmalloc_32(unsigned long size) | |
392 | { | |
393 | return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL); | |
394 | } | |
b5073173 PM |
395 | EXPORT_SYMBOL(vmalloc_32); |
396 | ||
397 | /** | |
398 | * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory | |
399 | * @size: allocation size | |
400 | * | |
401 | * The resulting memory area is 32bit addressable and zeroed so it can be | |
402 | * mapped to userspace without leaking data. | |
f905bc44 PM |
403 | * |
404 | * VM_USERMAP is set on the corresponding VMA so that subsequent calls to | |
405 | * remap_vmalloc_range() are permissible. | |
b5073173 PM |
406 | */ |
407 | void *vmalloc_32_user(unsigned long size) | |
408 | { | |
f905bc44 PM |
409 | /* |
410 | * We'll have to sort out the ZONE_DMA bits for 64-bit, | |
411 | * but for now this can simply use vmalloc_user() directly. | |
412 | */ | |
413 | return vmalloc_user(size); | |
b5073173 PM |
414 | } |
415 | EXPORT_SYMBOL(vmalloc_32_user); | |
1da177e4 LT |
416 | |
417 | void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot) | |
418 | { | |
419 | BUG(); | |
420 | return NULL; | |
421 | } | |
b5073173 | 422 | EXPORT_SYMBOL(vmap); |
1da177e4 | 423 | |
b3bdda02 | 424 | void vunmap(const void *addr) |
1da177e4 LT |
425 | { |
426 | BUG(); | |
427 | } | |
b5073173 | 428 | EXPORT_SYMBOL(vunmap); |
1da177e4 | 429 | |
eb6434d9 PM |
430 | void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot) |
431 | { | |
432 | BUG(); | |
433 | return NULL; | |
434 | } | |
435 | EXPORT_SYMBOL(vm_map_ram); | |
436 | ||
437 | void vm_unmap_ram(const void *mem, unsigned int count) | |
438 | { | |
439 | BUG(); | |
440 | } | |
441 | EXPORT_SYMBOL(vm_unmap_ram); | |
442 | ||
443 | void vm_unmap_aliases(void) | |
444 | { | |
445 | } | |
446 | EXPORT_SYMBOL_GPL(vm_unmap_aliases); | |
447 | ||
1eeb66a1 CH |
448 | /* |
449 | * Implement a stub for vmalloc_sync_all() if the architecture chose not to | |
450 | * have one. | |
451 | */ | |
3b32123d | 452 | void __weak vmalloc_sync_all(void) |
1eeb66a1 CH |
453 | { |
454 | } | |
455 | ||
29c185e5 PM |
456 | /** |
457 | * alloc_vm_area - allocate a range of kernel address space | |
458 | * @size: size of the area | |
459 | * | |
460 | * Returns: NULL on failure, vm_struct on success | |
461 | * | |
462 | * This function reserves a range of kernel address space, and | |
463 | * allocates pagetables to map that range. No actual mappings | |
464 | * are created. If the kernel address space is not shared | |
465 | * between processes, it syncs the pagetable across all | |
466 | * processes. | |
467 | */ | |
cd12909c | 468 | struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes) |
29c185e5 PM |
469 | { |
470 | BUG(); | |
471 | return NULL; | |
472 | } | |
473 | EXPORT_SYMBOL_GPL(alloc_vm_area); | |
474 | ||
475 | void free_vm_area(struct vm_struct *area) | |
476 | { | |
477 | BUG(); | |
478 | } | |
479 | EXPORT_SYMBOL_GPL(free_vm_area); | |
480 | ||
b5073173 PM |
481 | int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, |
482 | struct page *page) | |
483 | { | |
484 | return -EINVAL; | |
485 | } | |
486 | EXPORT_SYMBOL(vm_insert_page); | |
487 | ||
1da177e4 LT |
488 | /* |
489 | * sys_brk() for the most part doesn't need the global kernel | |
490 | * lock, except when an application is doing something nasty | |
491 | * like trying to un-brk an area that has already been mapped | |
492 | * to a regular file. in this case, the unmapping will need | |
493 | * to invoke file system routines that need the global lock. | |
494 | */ | |
6a6160a7 | 495 | SYSCALL_DEFINE1(brk, unsigned long, brk) |
1da177e4 LT |
496 | { |
497 | struct mm_struct *mm = current->mm; | |
498 | ||
499 | if (brk < mm->start_brk || brk > mm->context.end_brk) | |
500 | return mm->brk; | |
501 | ||
502 | if (mm->brk == brk) | |
503 | return mm->brk; | |
504 | ||
505 | /* | |
506 | * Always allow shrinking brk | |
507 | */ | |
508 | if (brk <= mm->brk) { | |
509 | mm->brk = brk; | |
510 | return brk; | |
511 | } | |
512 | ||
513 | /* | |
514 | * Ok, looks good - let it rip. | |
515 | */ | |
cfe79c00 | 516 | flush_icache_range(mm->brk, brk); |
1da177e4 LT |
517 | return mm->brk = brk; |
518 | } | |
519 | ||
8feae131 | 520 | /* |
3edf41d8 | 521 | * initialise the percpu counter for VM and region record slabs |
8feae131 DH |
522 | */ |
523 | void __init mmap_init(void) | |
1da177e4 | 524 | { |
00a62ce9 KM |
525 | int ret; |
526 | ||
908c7f19 | 527 | ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL); |
00a62ce9 | 528 | VM_BUG_ON(ret); |
5d097056 | 529 | vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC|SLAB_ACCOUNT); |
1da177e4 | 530 | } |
1da177e4 | 531 | |
3034097a | 532 | /* |
8feae131 DH |
533 | * validate the region tree |
534 | * - the caller must hold the region lock | |
3034097a | 535 | */ |
8feae131 DH |
536 | #ifdef CONFIG_DEBUG_NOMMU_REGIONS |
537 | static noinline void validate_nommu_regions(void) | |
3034097a | 538 | { |
8feae131 DH |
539 | struct vm_region *region, *last; |
540 | struct rb_node *p, *lastp; | |
3034097a | 541 | |
8feae131 DH |
542 | lastp = rb_first(&nommu_region_tree); |
543 | if (!lastp) | |
544 | return; | |
545 | ||
546 | last = rb_entry(lastp, struct vm_region, vm_rb); | |
c9427bc0 GT |
547 | BUG_ON(last->vm_end <= last->vm_start); |
548 | BUG_ON(last->vm_top < last->vm_end); | |
8feae131 DH |
549 | |
550 | while ((p = rb_next(lastp))) { | |
551 | region = rb_entry(p, struct vm_region, vm_rb); | |
552 | last = rb_entry(lastp, struct vm_region, vm_rb); | |
553 | ||
c9427bc0 GT |
554 | BUG_ON(region->vm_end <= region->vm_start); |
555 | BUG_ON(region->vm_top < region->vm_end); | |
556 | BUG_ON(region->vm_start < last->vm_top); | |
3034097a | 557 | |
8feae131 DH |
558 | lastp = p; |
559 | } | |
3034097a | 560 | } |
8feae131 | 561 | #else |
33e5d769 DH |
562 | static void validate_nommu_regions(void) |
563 | { | |
564 | } | |
8feae131 | 565 | #endif |
3034097a DH |
566 | |
567 | /* | |
8feae131 | 568 | * add a region into the global tree |
3034097a | 569 | */ |
8feae131 | 570 | static void add_nommu_region(struct vm_region *region) |
3034097a | 571 | { |
8feae131 DH |
572 | struct vm_region *pregion; |
573 | struct rb_node **p, *parent; | |
3034097a | 574 | |
8feae131 DH |
575 | validate_nommu_regions(); |
576 | ||
8feae131 DH |
577 | parent = NULL; |
578 | p = &nommu_region_tree.rb_node; | |
579 | while (*p) { | |
580 | parent = *p; | |
581 | pregion = rb_entry(parent, struct vm_region, vm_rb); | |
582 | if (region->vm_start < pregion->vm_start) | |
583 | p = &(*p)->rb_left; | |
584 | else if (region->vm_start > pregion->vm_start) | |
585 | p = &(*p)->rb_right; | |
586 | else if (pregion == region) | |
587 | return; | |
588 | else | |
589 | BUG(); | |
3034097a DH |
590 | } |
591 | ||
8feae131 DH |
592 | rb_link_node(®ion->vm_rb, parent, p); |
593 | rb_insert_color(®ion->vm_rb, &nommu_region_tree); | |
3034097a | 594 | |
8feae131 | 595 | validate_nommu_regions(); |
3034097a | 596 | } |
3034097a | 597 | |
930e652a | 598 | /* |
8feae131 | 599 | * delete a region from the global tree |
930e652a | 600 | */ |
8feae131 | 601 | static void delete_nommu_region(struct vm_region *region) |
930e652a | 602 | { |
8feae131 | 603 | BUG_ON(!nommu_region_tree.rb_node); |
930e652a | 604 | |
8feae131 DH |
605 | validate_nommu_regions(); |
606 | rb_erase(®ion->vm_rb, &nommu_region_tree); | |
607 | validate_nommu_regions(); | |
57c8f63e GU |
608 | } |
609 | ||
6fa5f80b | 610 | /* |
8feae131 | 611 | * free a contiguous series of pages |
6fa5f80b | 612 | */ |
8feae131 | 613 | static void free_page_series(unsigned long from, unsigned long to) |
6fa5f80b | 614 | { |
8feae131 DH |
615 | for (; from < to; from += PAGE_SIZE) { |
616 | struct page *page = virt_to_page(from); | |
617 | ||
33e5d769 | 618 | atomic_long_dec(&mmap_pages_allocated); |
8feae131 | 619 | put_page(page); |
6fa5f80b | 620 | } |
6fa5f80b DH |
621 | } |
622 | ||
3034097a | 623 | /* |
8feae131 | 624 | * release a reference to a region |
33e5d769 | 625 | * - the caller must hold the region semaphore for writing, which this releases |
dd8632a1 | 626 | * - the region may not have been added to the tree yet, in which case vm_top |
8feae131 | 627 | * will equal vm_start |
3034097a | 628 | */ |
8feae131 DH |
629 | static void __put_nommu_region(struct vm_region *region) |
630 | __releases(nommu_region_sem) | |
1da177e4 | 631 | { |
8feae131 | 632 | BUG_ON(!nommu_region_tree.rb_node); |
1da177e4 | 633 | |
1e2ae599 | 634 | if (--region->vm_usage == 0) { |
dd8632a1 | 635 | if (region->vm_top > region->vm_start) |
8feae131 DH |
636 | delete_nommu_region(region); |
637 | up_write(&nommu_region_sem); | |
638 | ||
639 | if (region->vm_file) | |
640 | fput(region->vm_file); | |
641 | ||
642 | /* IO memory and memory shared directly out of the pagecache | |
643 | * from ramfs/tmpfs mustn't be released here */ | |
22cc877b | 644 | if (region->vm_flags & VM_MAPPED_COPY) |
dd8632a1 | 645 | free_page_series(region->vm_start, region->vm_top); |
8feae131 DH |
646 | kmem_cache_free(vm_region_jar, region); |
647 | } else { | |
648 | up_write(&nommu_region_sem); | |
1da177e4 | 649 | } |
8feae131 | 650 | } |
1da177e4 | 651 | |
8feae131 DH |
652 | /* |
653 | * release a reference to a region | |
654 | */ | |
655 | static void put_nommu_region(struct vm_region *region) | |
656 | { | |
657 | down_write(&nommu_region_sem); | |
658 | __put_nommu_region(region); | |
1da177e4 LT |
659 | } |
660 | ||
eb8cdec4 BS |
661 | /* |
662 | * update protection on a vma | |
663 | */ | |
664 | static void protect_vma(struct vm_area_struct *vma, unsigned long flags) | |
665 | { | |
666 | #ifdef CONFIG_MPU | |
667 | struct mm_struct *mm = vma->vm_mm; | |
668 | long start = vma->vm_start & PAGE_MASK; | |
669 | while (start < vma->vm_end) { | |
670 | protect_page(mm, start, flags); | |
671 | start += PAGE_SIZE; | |
672 | } | |
673 | update_protections(mm); | |
674 | #endif | |
675 | } | |
676 | ||
3034097a | 677 | /* |
8feae131 DH |
678 | * add a VMA into a process's mm_struct in the appropriate place in the list |
679 | * and tree and add to the address space's page tree also if not an anonymous | |
680 | * page | |
681 | * - should be called with mm->mmap_sem held writelocked | |
3034097a | 682 | */ |
8feae131 | 683 | static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma) |
1da177e4 | 684 | { |
6038def0 | 685 | struct vm_area_struct *pvma, *prev; |
1da177e4 | 686 | struct address_space *mapping; |
6038def0 | 687 | struct rb_node **p, *parent, *rb_prev; |
8feae131 | 688 | |
8feae131 DH |
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 | ||
83cde9e8 | 700 | i_mmap_lock_write(mapping); |
1da177e4 | 701 | flush_dcache_mmap_lock(mapping); |
6b2dbba8 | 702 | vma_interval_tree_insert(vma, &mapping->i_mmap); |
1da177e4 | 703 | flush_dcache_mmap_unlock(mapping); |
83cde9e8 | 704 | i_mmap_unlock_write(mapping); |
1da177e4 LT |
705 | } |
706 | ||
8feae131 | 707 | /* add the VMA to the tree */ |
6038def0 | 708 | parent = rb_prev = NULL; |
8feae131 | 709 | p = &mm->mm_rb.rb_node; |
1da177e4 LT |
710 | while (*p) { |
711 | parent = *p; | |
712 | pvma = rb_entry(parent, struct vm_area_struct, vm_rb); | |
713 | ||
8feae131 DH |
714 | /* sort by: start addr, end addr, VMA struct addr in that order |
715 | * (the latter is necessary as we may get identical VMAs) */ | |
716 | if (vma->vm_start < pvma->vm_start) | |
1da177e4 | 717 | p = &(*p)->rb_left; |
6038def0 NK |
718 | else if (vma->vm_start > pvma->vm_start) { |
719 | rb_prev = parent; | |
1da177e4 | 720 | p = &(*p)->rb_right; |
6038def0 | 721 | } else if (vma->vm_end < pvma->vm_end) |
8feae131 | 722 | p = &(*p)->rb_left; |
6038def0 NK |
723 | else if (vma->vm_end > pvma->vm_end) { |
724 | rb_prev = parent; | |
8feae131 | 725 | p = &(*p)->rb_right; |
6038def0 | 726 | } else if (vma < pvma) |
8feae131 | 727 | p = &(*p)->rb_left; |
6038def0 NK |
728 | else if (vma > pvma) { |
729 | rb_prev = parent; | |
8feae131 | 730 | p = &(*p)->rb_right; |
6038def0 | 731 | } else |
8feae131 | 732 | BUG(); |
1da177e4 LT |
733 | } |
734 | ||
735 | rb_link_node(&vma->vm_rb, parent, p); | |
8feae131 DH |
736 | rb_insert_color(&vma->vm_rb, &mm->mm_rb); |
737 | ||
738 | /* add VMA to the VMA list also */ | |
6038def0 NK |
739 | prev = NULL; |
740 | if (rb_prev) | |
741 | prev = rb_entry(rb_prev, struct vm_area_struct, vm_rb); | |
8feae131 | 742 | |
6038def0 | 743 | __vma_link_list(mm, vma, prev, parent); |
1da177e4 LT |
744 | } |
745 | ||
3034097a | 746 | /* |
8feae131 | 747 | * delete a VMA from its owning mm_struct and address space |
3034097a | 748 | */ |
8feae131 | 749 | static void delete_vma_from_mm(struct vm_area_struct *vma) |
1da177e4 | 750 | { |
615d6e87 | 751 | int i; |
1da177e4 | 752 | struct address_space *mapping; |
8feae131 | 753 | struct mm_struct *mm = vma->vm_mm; |
615d6e87 | 754 | struct task_struct *curr = current; |
8feae131 | 755 | |
eb8cdec4 BS |
756 | protect_vma(vma, 0); |
757 | ||
8feae131 | 758 | mm->map_count--; |
615d6e87 DB |
759 | for (i = 0; i < VMACACHE_SIZE; i++) { |
760 | /* if the vma is cached, invalidate the entire cache */ | |
314ff785 | 761 | if (curr->vmacache.vmas[i] == vma) { |
e020d5bd | 762 | vmacache_invalidate(mm); |
615d6e87 DB |
763 | break; |
764 | } | |
765 | } | |
1da177e4 LT |
766 | |
767 | /* remove the VMA from the mapping */ | |
768 | if (vma->vm_file) { | |
769 | mapping = vma->vm_file->f_mapping; | |
770 | ||
83cde9e8 | 771 | i_mmap_lock_write(mapping); |
1da177e4 | 772 | flush_dcache_mmap_lock(mapping); |
6b2dbba8 | 773 | vma_interval_tree_remove(vma, &mapping->i_mmap); |
1da177e4 | 774 | flush_dcache_mmap_unlock(mapping); |
83cde9e8 | 775 | i_mmap_unlock_write(mapping); |
1da177e4 LT |
776 | } |
777 | ||
8feae131 DH |
778 | /* remove from the MM's tree and list */ |
779 | rb_erase(&vma->vm_rb, &mm->mm_rb); | |
b951bf2c NK |
780 | |
781 | if (vma->vm_prev) | |
782 | vma->vm_prev->vm_next = vma->vm_next; | |
783 | else | |
784 | mm->mmap = vma->vm_next; | |
785 | ||
786 | if (vma->vm_next) | |
787 | vma->vm_next->vm_prev = vma->vm_prev; | |
8feae131 DH |
788 | } |
789 | ||
790 | /* | |
791 | * destroy a VMA record | |
792 | */ | |
793 | static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma) | |
794 | { | |
8feae131 DH |
795 | if (vma->vm_ops && vma->vm_ops->close) |
796 | vma->vm_ops->close(vma); | |
e9714acf | 797 | if (vma->vm_file) |
8feae131 | 798 | fput(vma->vm_file); |
8feae131 DH |
799 | put_nommu_region(vma->vm_region); |
800 | kmem_cache_free(vm_area_cachep, vma); | |
801 | } | |
802 | ||
803 | /* | |
804 | * look up the first VMA in which addr resides, NULL if none | |
805 | * - should be called with mm->mmap_sem at least held readlocked | |
806 | */ | |
807 | struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr) | |
808 | { | |
809 | struct vm_area_struct *vma; | |
8feae131 DH |
810 | |
811 | /* check the cache first */ | |
615d6e87 DB |
812 | vma = vmacache_find(mm, addr); |
813 | if (likely(vma)) | |
8feae131 DH |
814 | return vma; |
815 | ||
e922c4c5 | 816 | /* trawl the list (there may be multiple mappings in which addr |
8feae131 | 817 | * resides) */ |
e922c4c5 | 818 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
8feae131 DH |
819 | if (vma->vm_start > addr) |
820 | return NULL; | |
821 | if (vma->vm_end > addr) { | |
615d6e87 | 822 | vmacache_update(addr, vma); |
8feae131 DH |
823 | return vma; |
824 | } | |
825 | } | |
826 | ||
827 | return NULL; | |
828 | } | |
829 | EXPORT_SYMBOL(find_vma); | |
830 | ||
831 | /* | |
832 | * find a VMA | |
833 | * - we don't extend stack VMAs under NOMMU conditions | |
834 | */ | |
835 | struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr) | |
836 | { | |
7561e8ca | 837 | return find_vma(mm, addr); |
8feae131 DH |
838 | } |
839 | ||
840 | /* | |
841 | * expand a stack to a given address | |
842 | * - not supported under NOMMU conditions | |
843 | */ | |
844 | int expand_stack(struct vm_area_struct *vma, unsigned long address) | |
845 | { | |
846 | return -ENOMEM; | |
847 | } | |
848 | ||
849 | /* | |
850 | * look up the first VMA exactly that exactly matches addr | |
851 | * - should be called with mm->mmap_sem at least held readlocked | |
852 | */ | |
853 | static struct vm_area_struct *find_vma_exact(struct mm_struct *mm, | |
854 | unsigned long addr, | |
855 | unsigned long len) | |
856 | { | |
857 | struct vm_area_struct *vma; | |
8feae131 DH |
858 | unsigned long end = addr + len; |
859 | ||
860 | /* check the cache first */ | |
615d6e87 DB |
861 | vma = vmacache_find_exact(mm, addr, end); |
862 | if (vma) | |
8feae131 DH |
863 | return vma; |
864 | ||
e922c4c5 | 865 | /* trawl the list (there may be multiple mappings in which addr |
8feae131 | 866 | * resides) */ |
e922c4c5 | 867 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
8feae131 DH |
868 | if (vma->vm_start < addr) |
869 | continue; | |
870 | if (vma->vm_start > addr) | |
871 | return NULL; | |
872 | if (vma->vm_end == end) { | |
615d6e87 | 873 | vmacache_update(addr, vma); |
8feae131 DH |
874 | return vma; |
875 | } | |
876 | } | |
877 | ||
878 | return NULL; | |
1da177e4 LT |
879 | } |
880 | ||
881 | /* | |
882 | * determine whether a mapping should be permitted and, if so, what sort of | |
883 | * mapping we're capable of supporting | |
884 | */ | |
885 | static int validate_mmap_request(struct file *file, | |
886 | unsigned long addr, | |
887 | unsigned long len, | |
888 | unsigned long prot, | |
889 | unsigned long flags, | |
890 | unsigned long pgoff, | |
891 | unsigned long *_capabilities) | |
892 | { | |
8feae131 | 893 | unsigned long capabilities, rlen; |
1da177e4 LT |
894 | int ret; |
895 | ||
896 | /* do the simple checks first */ | |
22cc877b | 897 | if (flags & MAP_FIXED) |
1da177e4 | 898 | return -EINVAL; |
1da177e4 LT |
899 | |
900 | if ((flags & MAP_TYPE) != MAP_PRIVATE && | |
901 | (flags & MAP_TYPE) != MAP_SHARED) | |
902 | return -EINVAL; | |
903 | ||
f81cff0d | 904 | if (!len) |
1da177e4 LT |
905 | return -EINVAL; |
906 | ||
f81cff0d | 907 | /* Careful about overflows.. */ |
8feae131 DH |
908 | rlen = PAGE_ALIGN(len); |
909 | if (!rlen || rlen > TASK_SIZE) | |
f81cff0d MF |
910 | return -ENOMEM; |
911 | ||
1da177e4 | 912 | /* offset overflow? */ |
8feae131 | 913 | if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff) |
f81cff0d | 914 | return -EOVERFLOW; |
1da177e4 LT |
915 | |
916 | if (file) { | |
1da177e4 | 917 | /* files must support mmap */ |
72c2d531 | 918 | if (!file->f_op->mmap) |
1da177e4 LT |
919 | return -ENODEV; |
920 | ||
921 | /* work out if what we've got could possibly be shared | |
922 | * - we support chardevs that provide their own "memory" | |
923 | * - we support files/blockdevs that are memory backed | |
924 | */ | |
b4caecd4 CH |
925 | if (file->f_op->mmap_capabilities) { |
926 | capabilities = file->f_op->mmap_capabilities(file); | |
927 | } else { | |
1da177e4 LT |
928 | /* no explicit capabilities set, so assume some |
929 | * defaults */ | |
496ad9aa | 930 | switch (file_inode(file)->i_mode & S_IFMT) { |
1da177e4 LT |
931 | case S_IFREG: |
932 | case S_IFBLK: | |
b4caecd4 | 933 | capabilities = NOMMU_MAP_COPY; |
1da177e4 LT |
934 | break; |
935 | ||
936 | case S_IFCHR: | |
937 | capabilities = | |
b4caecd4 CH |
938 | NOMMU_MAP_DIRECT | |
939 | NOMMU_MAP_READ | | |
940 | NOMMU_MAP_WRITE; | |
1da177e4 LT |
941 | break; |
942 | ||
943 | default: | |
944 | return -EINVAL; | |
945 | } | |
946 | } | |
947 | ||
948 | /* eliminate any capabilities that we can't support on this | |
949 | * device */ | |
950 | if (!file->f_op->get_unmapped_area) | |
b4caecd4 | 951 | capabilities &= ~NOMMU_MAP_DIRECT; |
6e242a1c | 952 | if (!(file->f_mode & FMODE_CAN_READ)) |
b4caecd4 | 953 | capabilities &= ~NOMMU_MAP_COPY; |
1da177e4 | 954 | |
28d7a6ae GY |
955 | /* The file shall have been opened with read permission. */ |
956 | if (!(file->f_mode & FMODE_READ)) | |
957 | return -EACCES; | |
958 | ||
1da177e4 LT |
959 | if (flags & MAP_SHARED) { |
960 | /* do checks for writing, appending and locking */ | |
961 | if ((prot & PROT_WRITE) && | |
962 | !(file->f_mode & FMODE_WRITE)) | |
963 | return -EACCES; | |
964 | ||
496ad9aa | 965 | if (IS_APPEND(file_inode(file)) && |
1da177e4 LT |
966 | (file->f_mode & FMODE_WRITE)) |
967 | return -EACCES; | |
968 | ||
d7a06983 | 969 | if (locks_verify_locked(file)) |
1da177e4 LT |
970 | return -EAGAIN; |
971 | ||
b4caecd4 | 972 | if (!(capabilities & NOMMU_MAP_DIRECT)) |
1da177e4 LT |
973 | return -ENODEV; |
974 | ||
1da177e4 | 975 | /* we mustn't privatise shared mappings */ |
b4caecd4 | 976 | capabilities &= ~NOMMU_MAP_COPY; |
ac714904 | 977 | } else { |
1da177e4 LT |
978 | /* we're going to read the file into private memory we |
979 | * allocate */ | |
b4caecd4 | 980 | if (!(capabilities & NOMMU_MAP_COPY)) |
1da177e4 LT |
981 | return -ENODEV; |
982 | ||
983 | /* we don't permit a private writable mapping to be | |
984 | * shared with the backing device */ | |
985 | if (prot & PROT_WRITE) | |
b4caecd4 | 986 | capabilities &= ~NOMMU_MAP_DIRECT; |
1da177e4 LT |
987 | } |
988 | ||
b4caecd4 CH |
989 | if (capabilities & NOMMU_MAP_DIRECT) { |
990 | if (((prot & PROT_READ) && !(capabilities & NOMMU_MAP_READ)) || | |
991 | ((prot & PROT_WRITE) && !(capabilities & NOMMU_MAP_WRITE)) || | |
992 | ((prot & PROT_EXEC) && !(capabilities & NOMMU_MAP_EXEC)) | |
3c7b2045 | 993 | ) { |
b4caecd4 | 994 | capabilities &= ~NOMMU_MAP_DIRECT; |
3c7b2045 | 995 | if (flags & MAP_SHARED) { |
22cc877b | 996 | pr_warn("MAP_SHARED not completely supported on !MMU\n"); |
3c7b2045 BS |
997 | return -EINVAL; |
998 | } | |
999 | } | |
1000 | } | |
1001 | ||
1da177e4 LT |
1002 | /* handle executable mappings and implied executable |
1003 | * mappings */ | |
90f8572b | 1004 | if (path_noexec(&file->f_path)) { |
1da177e4 LT |
1005 | if (prot & PROT_EXEC) |
1006 | return -EPERM; | |
ac714904 | 1007 | } else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) { |
1da177e4 LT |
1008 | /* handle implication of PROT_EXEC by PROT_READ */ |
1009 | if (current->personality & READ_IMPLIES_EXEC) { | |
b4caecd4 | 1010 | if (capabilities & NOMMU_MAP_EXEC) |
1da177e4 LT |
1011 | prot |= PROT_EXEC; |
1012 | } | |
ac714904 | 1013 | } else if ((prot & PROT_READ) && |
1da177e4 | 1014 | (prot & PROT_EXEC) && |
b4caecd4 | 1015 | !(capabilities & NOMMU_MAP_EXEC) |
1da177e4 LT |
1016 | ) { |
1017 | /* backing file is not executable, try to copy */ | |
b4caecd4 | 1018 | capabilities &= ~NOMMU_MAP_DIRECT; |
1da177e4 | 1019 | } |
ac714904 | 1020 | } else { |
1da177e4 LT |
1021 | /* anonymous mappings are always memory backed and can be |
1022 | * privately mapped | |
1023 | */ | |
b4caecd4 | 1024 | capabilities = NOMMU_MAP_COPY; |
1da177e4 LT |
1025 | |
1026 | /* handle PROT_EXEC implication by PROT_READ */ | |
1027 | if ((prot & PROT_READ) && | |
1028 | (current->personality & READ_IMPLIES_EXEC)) | |
1029 | prot |= PROT_EXEC; | |
1030 | } | |
1031 | ||
1032 | /* allow the security API to have its say */ | |
e5467859 | 1033 | ret = security_mmap_addr(addr); |
1da177e4 LT |
1034 | if (ret < 0) |
1035 | return ret; | |
1036 | ||
1037 | /* looks okay */ | |
1038 | *_capabilities = capabilities; | |
1039 | return 0; | |
1040 | } | |
1041 | ||
1042 | /* | |
1043 | * we've determined that we can make the mapping, now translate what we | |
1044 | * now know into VMA flags | |
1045 | */ | |
1046 | static unsigned long determine_vm_flags(struct file *file, | |
1047 | unsigned long prot, | |
1048 | unsigned long flags, | |
1049 | unsigned long capabilities) | |
1050 | { | |
1051 | unsigned long vm_flags; | |
1052 | ||
e6bfb709 | 1053 | vm_flags = calc_vm_prot_bits(prot, 0) | calc_vm_flag_bits(flags); |
1da177e4 LT |
1054 | /* vm_flags |= mm->def_flags; */ |
1055 | ||
b4caecd4 | 1056 | if (!(capabilities & NOMMU_MAP_DIRECT)) { |
1da177e4 | 1057 | /* attempt to share read-only copies of mapped file chunks */ |
3c7b2045 | 1058 | vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC; |
1da177e4 LT |
1059 | if (file && !(prot & PROT_WRITE)) |
1060 | vm_flags |= VM_MAYSHARE; | |
3c7b2045 | 1061 | } else { |
1da177e4 LT |
1062 | /* overlay a shareable mapping on the backing device or inode |
1063 | * if possible - used for chardevs, ramfs/tmpfs/shmfs and | |
1064 | * romfs/cramfs */ | |
b4caecd4 | 1065 | vm_flags |= VM_MAYSHARE | (capabilities & NOMMU_VMFLAGS); |
1da177e4 | 1066 | if (flags & MAP_SHARED) |
3c7b2045 | 1067 | vm_flags |= VM_SHARED; |
1da177e4 LT |
1068 | } |
1069 | ||
1070 | /* refuse to let anyone share private mappings with this process if | |
1071 | * it's being traced - otherwise breakpoints set in it may interfere | |
1072 | * with another untraced process | |
1073 | */ | |
a288eecc | 1074 | if ((flags & MAP_PRIVATE) && current->ptrace) |
1da177e4 LT |
1075 | vm_flags &= ~VM_MAYSHARE; |
1076 | ||
1077 | return vm_flags; | |
1078 | } | |
1079 | ||
1080 | /* | |
8feae131 DH |
1081 | * set up a shared mapping on a file (the driver or filesystem provides and |
1082 | * pins the storage) | |
1da177e4 | 1083 | */ |
8feae131 | 1084 | static int do_mmap_shared_file(struct vm_area_struct *vma) |
1da177e4 LT |
1085 | { |
1086 | int ret; | |
1087 | ||
f74ac015 | 1088 | ret = call_mmap(vma->vm_file, vma); |
dd8632a1 PM |
1089 | if (ret == 0) { |
1090 | vma->vm_region->vm_top = vma->vm_region->vm_end; | |
645d83c5 | 1091 | return 0; |
dd8632a1 | 1092 | } |
1da177e4 LT |
1093 | if (ret != -ENOSYS) |
1094 | return ret; | |
1095 | ||
3fa30460 DH |
1096 | /* getting -ENOSYS indicates that direct mmap isn't possible (as |
1097 | * opposed to tried but failed) so we can only give a suitable error as | |
1098 | * it's not possible to make a private copy if MAP_SHARED was given */ | |
1da177e4 LT |
1099 | return -ENODEV; |
1100 | } | |
1101 | ||
1102 | /* | |
1103 | * set up a private mapping or an anonymous shared mapping | |
1104 | */ | |
8feae131 DH |
1105 | static int do_mmap_private(struct vm_area_struct *vma, |
1106 | struct vm_region *region, | |
645d83c5 DH |
1107 | unsigned long len, |
1108 | unsigned long capabilities) | |
1da177e4 | 1109 | { |
dbc8358c | 1110 | unsigned long total, point; |
1da177e4 | 1111 | void *base; |
8feae131 | 1112 | int ret, order; |
1da177e4 LT |
1113 | |
1114 | /* invoke the file's mapping function so that it can keep track of | |
1115 | * shared mappings on devices or memory | |
1116 | * - VM_MAYSHARE will be set if it may attempt to share | |
1117 | */ | |
b4caecd4 | 1118 | if (capabilities & NOMMU_MAP_DIRECT) { |
f74ac015 | 1119 | ret = call_mmap(vma->vm_file, vma); |
dd8632a1 | 1120 | if (ret == 0) { |
1da177e4 | 1121 | /* shouldn't return success if we're not sharing */ |
dd8632a1 PM |
1122 | BUG_ON(!(vma->vm_flags & VM_MAYSHARE)); |
1123 | vma->vm_region->vm_top = vma->vm_region->vm_end; | |
645d83c5 | 1124 | return 0; |
1da177e4 | 1125 | } |
dd8632a1 PM |
1126 | if (ret != -ENOSYS) |
1127 | return ret; | |
1da177e4 LT |
1128 | |
1129 | /* getting an ENOSYS error indicates that direct mmap isn't | |
1130 | * possible (as opposed to tried but failed) so we'll try to | |
1131 | * make a private copy of the data and map that instead */ | |
1132 | } | |
1133 | ||
8feae131 | 1134 | |
1da177e4 LT |
1135 | /* allocate some memory to hold the mapping |
1136 | * - note that this may not return a page-aligned address if the object | |
1137 | * we're allocating is smaller than a page | |
1138 | */ | |
f67d9b15 | 1139 | order = get_order(len); |
8feae131 | 1140 | total = 1 << order; |
f67d9b15 | 1141 | point = len >> PAGE_SHIFT; |
dd8632a1 | 1142 | |
dbc8358c | 1143 | /* we don't want to allocate a power-of-2 sized page set */ |
22cc877b | 1144 | if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages) |
dbc8358c | 1145 | total = point; |
8feae131 | 1146 | |
da616534 | 1147 | base = alloc_pages_exact(total << PAGE_SHIFT, GFP_KERNEL); |
dbc8358c JK |
1148 | if (!base) |
1149 | goto enomem; | |
1150 | ||
1151 | atomic_long_add(total, &mmap_pages_allocated); | |
1da177e4 | 1152 | |
8feae131 DH |
1153 | region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY; |
1154 | region->vm_start = (unsigned long) base; | |
f67d9b15 | 1155 | region->vm_end = region->vm_start + len; |
dd8632a1 | 1156 | region->vm_top = region->vm_start + (total << PAGE_SHIFT); |
8feae131 DH |
1157 | |
1158 | vma->vm_start = region->vm_start; | |
1159 | vma->vm_end = region->vm_start + len; | |
1da177e4 LT |
1160 | |
1161 | if (vma->vm_file) { | |
1162 | /* read the contents of a file into the copy */ | |
1163 | mm_segment_t old_fs; | |
1164 | loff_t fpos; | |
1165 | ||
1166 | fpos = vma->vm_pgoff; | |
1167 | fpos <<= PAGE_SHIFT; | |
1168 | ||
1169 | old_fs = get_fs(); | |
1170 | set_fs(KERNEL_DS); | |
6e242a1c | 1171 | ret = __vfs_read(vma->vm_file, base, len, &fpos); |
1da177e4 LT |
1172 | set_fs(old_fs); |
1173 | ||
1174 | if (ret < 0) | |
1175 | goto error_free; | |
1176 | ||
1177 | /* clear the last little bit */ | |
f67d9b15 BL |
1178 | if (ret < len) |
1179 | memset(base + ret, 0, len - ret); | |
1da177e4 | 1180 | |
1da177e4 LT |
1181 | } |
1182 | ||
1183 | return 0; | |
1184 | ||
1185 | error_free: | |
7223bb4a | 1186 | free_page_series(region->vm_start, region->vm_top); |
8feae131 DH |
1187 | region->vm_start = vma->vm_start = 0; |
1188 | region->vm_end = vma->vm_end = 0; | |
dd8632a1 | 1189 | region->vm_top = 0; |
1da177e4 LT |
1190 | return ret; |
1191 | ||
1192 | enomem: | |
b1de0d13 | 1193 | pr_err("Allocation of length %lu from process %d (%s) failed\n", |
05ae6fa3 | 1194 | len, current->pid, current->comm); |
9af744d7 | 1195 | show_free_areas(0, NULL); |
1da177e4 LT |
1196 | return -ENOMEM; |
1197 | } | |
1198 | ||
1199 | /* | |
1200 | * handle mapping creation for uClinux | |
1201 | */ | |
1fcfd8db ON |
1202 | unsigned long do_mmap(struct file *file, |
1203 | unsigned long addr, | |
1204 | unsigned long len, | |
1205 | unsigned long prot, | |
1206 | unsigned long flags, | |
1207 | vm_flags_t vm_flags, | |
1208 | unsigned long pgoff, | |
897ab3e0 MR |
1209 | unsigned long *populate, |
1210 | struct list_head *uf) | |
1da177e4 | 1211 | { |
8feae131 DH |
1212 | struct vm_area_struct *vma; |
1213 | struct vm_region *region; | |
1da177e4 | 1214 | struct rb_node *rb; |
1fcfd8db | 1215 | unsigned long capabilities, result; |
1da177e4 LT |
1216 | int ret; |
1217 | ||
41badc15 | 1218 | *populate = 0; |
bebeb3d6 | 1219 | |
1da177e4 LT |
1220 | /* decide whether we should attempt the mapping, and if so what sort of |
1221 | * mapping */ | |
1222 | ret = validate_mmap_request(file, addr, len, prot, flags, pgoff, | |
1223 | &capabilities); | |
22cc877b | 1224 | if (ret < 0) |
1da177e4 LT |
1225 | return ret; |
1226 | ||
06aab5a3 DH |
1227 | /* we ignore the address hint */ |
1228 | addr = 0; | |
f67d9b15 | 1229 | len = PAGE_ALIGN(len); |
06aab5a3 | 1230 | |
1da177e4 LT |
1231 | /* we've determined that we can make the mapping, now translate what we |
1232 | * now know into VMA flags */ | |
1fcfd8db | 1233 | vm_flags |= determine_vm_flags(file, prot, flags, capabilities); |
1da177e4 | 1234 | |
8feae131 DH |
1235 | /* we're going to need to record the mapping */ |
1236 | region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL); | |
1237 | if (!region) | |
1238 | goto error_getting_region; | |
1239 | ||
1240 | vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); | |
1241 | if (!vma) | |
1242 | goto error_getting_vma; | |
1da177e4 | 1243 | |
1e2ae599 | 1244 | region->vm_usage = 1; |
8feae131 DH |
1245 | region->vm_flags = vm_flags; |
1246 | region->vm_pgoff = pgoff; | |
1247 | ||
5beb4930 | 1248 | INIT_LIST_HEAD(&vma->anon_vma_chain); |
8feae131 DH |
1249 | vma->vm_flags = vm_flags; |
1250 | vma->vm_pgoff = pgoff; | |
1da177e4 | 1251 | |
8feae131 | 1252 | if (file) { |
cb0942b8 AV |
1253 | region->vm_file = get_file(file); |
1254 | vma->vm_file = get_file(file); | |
8feae131 DH |
1255 | } |
1256 | ||
1257 | down_write(&nommu_region_sem); | |
1258 | ||
1259 | /* if we want to share, we need to check for regions created by other | |
1da177e4 | 1260 | * mmap() calls that overlap with our proposed mapping |
8feae131 | 1261 | * - we can only share with a superset match on most regular files |
1da177e4 LT |
1262 | * - shared mappings on character devices and memory backed files are |
1263 | * permitted to overlap inexactly as far as we are concerned for in | |
1264 | * these cases, sharing is handled in the driver or filesystem rather | |
1265 | * than here | |
1266 | */ | |
1267 | if (vm_flags & VM_MAYSHARE) { | |
8feae131 DH |
1268 | struct vm_region *pregion; |
1269 | unsigned long pglen, rpglen, pgend, rpgend, start; | |
1da177e4 | 1270 | |
8feae131 DH |
1271 | pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; |
1272 | pgend = pgoff + pglen; | |
165b2392 | 1273 | |
8feae131 DH |
1274 | for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) { |
1275 | pregion = rb_entry(rb, struct vm_region, vm_rb); | |
1da177e4 | 1276 | |
8feae131 | 1277 | if (!(pregion->vm_flags & VM_MAYSHARE)) |
1da177e4 LT |
1278 | continue; |
1279 | ||
1280 | /* search for overlapping mappings on the same file */ | |
496ad9aa AV |
1281 | if (file_inode(pregion->vm_file) != |
1282 | file_inode(file)) | |
1da177e4 LT |
1283 | continue; |
1284 | ||
8feae131 | 1285 | if (pregion->vm_pgoff >= pgend) |
1da177e4 LT |
1286 | continue; |
1287 | ||
8feae131 DH |
1288 | rpglen = pregion->vm_end - pregion->vm_start; |
1289 | rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
1290 | rpgend = pregion->vm_pgoff + rpglen; | |
1291 | if (pgoff >= rpgend) | |
1da177e4 LT |
1292 | continue; |
1293 | ||
8feae131 DH |
1294 | /* handle inexactly overlapping matches between |
1295 | * mappings */ | |
1296 | if ((pregion->vm_pgoff != pgoff || rpglen != pglen) && | |
1297 | !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) { | |
1298 | /* new mapping is not a subset of the region */ | |
b4caecd4 | 1299 | if (!(capabilities & NOMMU_MAP_DIRECT)) |
1da177e4 LT |
1300 | goto sharing_violation; |
1301 | continue; | |
1302 | } | |
1303 | ||
8feae131 | 1304 | /* we've found a region we can share */ |
1e2ae599 | 1305 | pregion->vm_usage++; |
8feae131 DH |
1306 | vma->vm_region = pregion; |
1307 | start = pregion->vm_start; | |
1308 | start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT; | |
1309 | vma->vm_start = start; | |
1310 | vma->vm_end = start + len; | |
1311 | ||
22cc877b | 1312 | if (pregion->vm_flags & VM_MAPPED_COPY) |
8feae131 | 1313 | vma->vm_flags |= VM_MAPPED_COPY; |
22cc877b | 1314 | else { |
8feae131 DH |
1315 | ret = do_mmap_shared_file(vma); |
1316 | if (ret < 0) { | |
1317 | vma->vm_region = NULL; | |
1318 | vma->vm_start = 0; | |
1319 | vma->vm_end = 0; | |
1e2ae599 | 1320 | pregion->vm_usage--; |
8feae131 DH |
1321 | pregion = NULL; |
1322 | goto error_just_free; | |
1323 | } | |
1324 | } | |
1325 | fput(region->vm_file); | |
1326 | kmem_cache_free(vm_region_jar, region); | |
1327 | region = pregion; | |
1328 | result = start; | |
1329 | goto share; | |
1da177e4 LT |
1330 | } |
1331 | ||
1da177e4 LT |
1332 | /* obtain the address at which to make a shared mapping |
1333 | * - this is the hook for quasi-memory character devices to | |
1334 | * tell us the location of a shared mapping | |
1335 | */ | |
b4caecd4 | 1336 | if (capabilities & NOMMU_MAP_DIRECT) { |
1da177e4 LT |
1337 | addr = file->f_op->get_unmapped_area(file, addr, len, |
1338 | pgoff, flags); | |
bb005a59 | 1339 | if (IS_ERR_VALUE(addr)) { |
1da177e4 | 1340 | ret = addr; |
bb005a59 | 1341 | if (ret != -ENOSYS) |
8feae131 | 1342 | goto error_just_free; |
1da177e4 LT |
1343 | |
1344 | /* the driver refused to tell us where to site | |
1345 | * the mapping so we'll have to attempt to copy | |
1346 | * it */ | |
bb005a59 | 1347 | ret = -ENODEV; |
b4caecd4 | 1348 | if (!(capabilities & NOMMU_MAP_COPY)) |
8feae131 | 1349 | goto error_just_free; |
1da177e4 | 1350 | |
b4caecd4 | 1351 | capabilities &= ~NOMMU_MAP_DIRECT; |
8feae131 DH |
1352 | } else { |
1353 | vma->vm_start = region->vm_start = addr; | |
1354 | vma->vm_end = region->vm_end = addr + len; | |
1da177e4 LT |
1355 | } |
1356 | } | |
1357 | } | |
1358 | ||
8feae131 | 1359 | vma->vm_region = region; |
1da177e4 | 1360 | |
645d83c5 | 1361 | /* set up the mapping |
b4caecd4 | 1362 | * - the region is filled in if NOMMU_MAP_DIRECT is still set |
645d83c5 | 1363 | */ |
1da177e4 | 1364 | if (file && vma->vm_flags & VM_SHARED) |
8feae131 | 1365 | ret = do_mmap_shared_file(vma); |
1da177e4 | 1366 | else |
645d83c5 | 1367 | ret = do_mmap_private(vma, region, len, capabilities); |
1da177e4 | 1368 | if (ret < 0) |
645d83c5 DH |
1369 | goto error_just_free; |
1370 | add_nommu_region(region); | |
8feae131 | 1371 | |
ea637639 JZ |
1372 | /* clear anonymous mappings that don't ask for uninitialized data */ |
1373 | if (!vma->vm_file && !(flags & MAP_UNINITIALIZED)) | |
1374 | memset((void *)region->vm_start, 0, | |
1375 | region->vm_end - region->vm_start); | |
1376 | ||
1da177e4 | 1377 | /* okay... we have a mapping; now we have to register it */ |
8feae131 | 1378 | result = vma->vm_start; |
1da177e4 | 1379 | |
1da177e4 LT |
1380 | current->mm->total_vm += len >> PAGE_SHIFT; |
1381 | ||
8feae131 DH |
1382 | share: |
1383 | add_vma_to_mm(current->mm, vma); | |
1da177e4 | 1384 | |
cfe79c00 MF |
1385 | /* we flush the region from the icache only when the first executable |
1386 | * mapping of it is made */ | |
1387 | if (vma->vm_flags & VM_EXEC && !region->vm_icache_flushed) { | |
1388 | flush_icache_range(region->vm_start, region->vm_end); | |
1389 | region->vm_icache_flushed = true; | |
1390 | } | |
1da177e4 | 1391 | |
cfe79c00 | 1392 | up_write(&nommu_region_sem); |
1da177e4 | 1393 | |
8feae131 | 1394 | return result; |
1da177e4 | 1395 | |
8feae131 DH |
1396 | error_just_free: |
1397 | up_write(&nommu_region_sem); | |
1398 | error: | |
89a86402 DH |
1399 | if (region->vm_file) |
1400 | fput(region->vm_file); | |
8feae131 | 1401 | kmem_cache_free(vm_region_jar, region); |
89a86402 DH |
1402 | if (vma->vm_file) |
1403 | fput(vma->vm_file); | |
8feae131 | 1404 | kmem_cache_free(vm_area_cachep, vma); |
8feae131 DH |
1405 | return ret; |
1406 | ||
1407 | sharing_violation: | |
1408 | up_write(&nommu_region_sem); | |
22cc877b | 1409 | pr_warn("Attempt to share mismatched mappings\n"); |
8feae131 DH |
1410 | ret = -EINVAL; |
1411 | goto error; | |
1da177e4 | 1412 | |
8feae131 DH |
1413 | error_getting_vma: |
1414 | kmem_cache_free(vm_region_jar, region); | |
22cc877b LR |
1415 | pr_warn("Allocation of vma for %lu byte allocation from process %d failed\n", |
1416 | len, current->pid); | |
9af744d7 | 1417 | show_free_areas(0, NULL); |
1da177e4 LT |
1418 | return -ENOMEM; |
1419 | ||
8feae131 | 1420 | error_getting_region: |
22cc877b LR |
1421 | pr_warn("Allocation of vm region for %lu byte allocation from process %d failed\n", |
1422 | len, current->pid); | |
9af744d7 | 1423 | show_free_areas(0, NULL); |
1da177e4 LT |
1424 | return -ENOMEM; |
1425 | } | |
6be5ceb0 | 1426 | |
66f0dc48 HD |
1427 | SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len, |
1428 | unsigned long, prot, unsigned long, flags, | |
1429 | unsigned long, fd, unsigned long, pgoff) | |
1430 | { | |
1431 | struct file *file = NULL; | |
1432 | unsigned long retval = -EBADF; | |
1433 | ||
120a795d | 1434 | audit_mmap_fd(fd, flags); |
66f0dc48 HD |
1435 | if (!(flags & MAP_ANONYMOUS)) { |
1436 | file = fget(fd); | |
1437 | if (!file) | |
1438 | goto out; | |
1439 | } | |
1440 | ||
1441 | flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE); | |
1442 | ||
ad1ed293 | 1443 | retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff); |
66f0dc48 HD |
1444 | |
1445 | if (file) | |
1446 | fput(file); | |
1447 | out: | |
1448 | return retval; | |
1449 | } | |
1450 | ||
a4679373 CH |
1451 | #ifdef __ARCH_WANT_SYS_OLD_MMAP |
1452 | struct mmap_arg_struct { | |
1453 | unsigned long addr; | |
1454 | unsigned long len; | |
1455 | unsigned long prot; | |
1456 | unsigned long flags; | |
1457 | unsigned long fd; | |
1458 | unsigned long offset; | |
1459 | }; | |
1460 | ||
1461 | SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg) | |
1462 | { | |
1463 | struct mmap_arg_struct a; | |
1464 | ||
1465 | if (copy_from_user(&a, arg, sizeof(a))) | |
1466 | return -EFAULT; | |
1824cb75 | 1467 | if (offset_in_page(a.offset)) |
a4679373 CH |
1468 | return -EINVAL; |
1469 | ||
1470 | return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd, | |
1471 | a.offset >> PAGE_SHIFT); | |
1472 | } | |
1473 | #endif /* __ARCH_WANT_SYS_OLD_MMAP */ | |
1474 | ||
1da177e4 | 1475 | /* |
8feae131 DH |
1476 | * split a vma into two pieces at address 'addr', a new vma is allocated either |
1477 | * for the first part or the tail. | |
1da177e4 | 1478 | */ |
8feae131 DH |
1479 | int split_vma(struct mm_struct *mm, struct vm_area_struct *vma, |
1480 | unsigned long addr, int new_below) | |
1da177e4 | 1481 | { |
8feae131 DH |
1482 | struct vm_area_struct *new; |
1483 | struct vm_region *region; | |
1484 | unsigned long npages; | |
1da177e4 | 1485 | |
779c1023 DH |
1486 | /* we're only permitted to split anonymous regions (these should have |
1487 | * only a single usage on the region) */ | |
1488 | if (vma->vm_file) | |
8feae131 | 1489 | return -ENOMEM; |
1da177e4 | 1490 | |
8feae131 DH |
1491 | if (mm->map_count >= sysctl_max_map_count) |
1492 | return -ENOMEM; | |
1da177e4 | 1493 | |
8feae131 DH |
1494 | region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL); |
1495 | if (!region) | |
1496 | return -ENOMEM; | |
1da177e4 | 1497 | |
8feae131 DH |
1498 | new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); |
1499 | if (!new) { | |
1500 | kmem_cache_free(vm_region_jar, region); | |
1501 | return -ENOMEM; | |
1502 | } | |
1503 | ||
1504 | /* most fields are the same, copy all, and then fixup */ | |
1505 | *new = *vma; | |
1506 | *region = *vma->vm_region; | |
1507 | new->vm_region = region; | |
1508 | ||
1509 | npages = (addr - vma->vm_start) >> PAGE_SHIFT; | |
1510 | ||
1511 | if (new_below) { | |
dd8632a1 | 1512 | region->vm_top = region->vm_end = new->vm_end = addr; |
8feae131 DH |
1513 | } else { |
1514 | region->vm_start = new->vm_start = addr; | |
1515 | region->vm_pgoff = new->vm_pgoff += npages; | |
1da177e4 | 1516 | } |
8feae131 DH |
1517 | |
1518 | if (new->vm_ops && new->vm_ops->open) | |
1519 | new->vm_ops->open(new); | |
1520 | ||
1521 | delete_vma_from_mm(vma); | |
1522 | down_write(&nommu_region_sem); | |
1523 | delete_nommu_region(vma->vm_region); | |
1524 | if (new_below) { | |
1525 | vma->vm_region->vm_start = vma->vm_start = addr; | |
1526 | vma->vm_region->vm_pgoff = vma->vm_pgoff += npages; | |
1527 | } else { | |
1528 | vma->vm_region->vm_end = vma->vm_end = addr; | |
dd8632a1 | 1529 | vma->vm_region->vm_top = addr; |
8feae131 DH |
1530 | } |
1531 | add_nommu_region(vma->vm_region); | |
1532 | add_nommu_region(new->vm_region); | |
1533 | up_write(&nommu_region_sem); | |
1534 | add_vma_to_mm(mm, vma); | |
1535 | add_vma_to_mm(mm, new); | |
1536 | return 0; | |
1da177e4 LT |
1537 | } |
1538 | ||
3034097a | 1539 | /* |
8feae131 DH |
1540 | * shrink a VMA by removing the specified chunk from either the beginning or |
1541 | * the end | |
3034097a | 1542 | */ |
8feae131 DH |
1543 | static int shrink_vma(struct mm_struct *mm, |
1544 | struct vm_area_struct *vma, | |
1545 | unsigned long from, unsigned long to) | |
1da177e4 | 1546 | { |
8feae131 | 1547 | struct vm_region *region; |
1da177e4 | 1548 | |
8feae131 DH |
1549 | /* adjust the VMA's pointers, which may reposition it in the MM's tree |
1550 | * and list */ | |
1551 | delete_vma_from_mm(vma); | |
1552 | if (from > vma->vm_start) | |
1553 | vma->vm_end = from; | |
1554 | else | |
1555 | vma->vm_start = to; | |
1556 | add_vma_to_mm(mm, vma); | |
1da177e4 | 1557 | |
8feae131 DH |
1558 | /* cut the backing region down to size */ |
1559 | region = vma->vm_region; | |
1e2ae599 | 1560 | BUG_ON(region->vm_usage != 1); |
8feae131 DH |
1561 | |
1562 | down_write(&nommu_region_sem); | |
1563 | delete_nommu_region(region); | |
dd8632a1 PM |
1564 | if (from > region->vm_start) { |
1565 | to = region->vm_top; | |
1566 | region->vm_top = region->vm_end = from; | |
1567 | } else { | |
8feae131 | 1568 | region->vm_start = to; |
dd8632a1 | 1569 | } |
8feae131 DH |
1570 | add_nommu_region(region); |
1571 | up_write(&nommu_region_sem); | |
1572 | ||
1573 | free_page_series(from, to); | |
1574 | return 0; | |
1575 | } | |
1da177e4 | 1576 | |
8feae131 DH |
1577 | /* |
1578 | * release a mapping | |
1579 | * - under NOMMU conditions the chunk to be unmapped must be backed by a single | |
1580 | * VMA, though it need not cover the whole VMA | |
1581 | */ | |
897ab3e0 | 1582 | int do_munmap(struct mm_struct *mm, unsigned long start, size_t len, struct list_head *uf) |
8feae131 DH |
1583 | { |
1584 | struct vm_area_struct *vma; | |
f67d9b15 | 1585 | unsigned long end; |
8feae131 | 1586 | int ret; |
1da177e4 | 1587 | |
f67d9b15 | 1588 | len = PAGE_ALIGN(len); |
8feae131 DH |
1589 | if (len == 0) |
1590 | return -EINVAL; | |
365e9c87 | 1591 | |
f67d9b15 BL |
1592 | end = start + len; |
1593 | ||
8feae131 DH |
1594 | /* find the first potentially overlapping VMA */ |
1595 | vma = find_vma(mm, start); | |
1596 | if (!vma) { | |
ac714904 | 1597 | static int limit; |
33e5d769 | 1598 | if (limit < 5) { |
22cc877b LR |
1599 | pr_warn("munmap of memory not mmapped by process %d (%s): 0x%lx-0x%lx\n", |
1600 | current->pid, current->comm, | |
1601 | start, start + len - 1); | |
33e5d769 DH |
1602 | limit++; |
1603 | } | |
8feae131 DH |
1604 | return -EINVAL; |
1605 | } | |
1da177e4 | 1606 | |
8feae131 DH |
1607 | /* we're allowed to split an anonymous VMA but not a file-backed one */ |
1608 | if (vma->vm_file) { | |
1609 | do { | |
22cc877b | 1610 | if (start > vma->vm_start) |
8feae131 | 1611 | return -EINVAL; |
8feae131 DH |
1612 | if (end == vma->vm_end) |
1613 | goto erase_whole_vma; | |
d75a310c NK |
1614 | vma = vma->vm_next; |
1615 | } while (vma); | |
8feae131 DH |
1616 | return -EINVAL; |
1617 | } else { | |
1618 | /* the chunk must be a subset of the VMA found */ | |
1619 | if (start == vma->vm_start && end == vma->vm_end) | |
1620 | goto erase_whole_vma; | |
22cc877b | 1621 | if (start < vma->vm_start || end > vma->vm_end) |
8feae131 | 1622 | return -EINVAL; |
1824cb75 | 1623 | if (offset_in_page(start)) |
8feae131 | 1624 | return -EINVAL; |
1824cb75 | 1625 | if (end != vma->vm_end && offset_in_page(end)) |
8feae131 | 1626 | return -EINVAL; |
8feae131 DH |
1627 | if (start != vma->vm_start && end != vma->vm_end) { |
1628 | ret = split_vma(mm, vma, start, 1); | |
22cc877b | 1629 | if (ret < 0) |
8feae131 | 1630 | return ret; |
8feae131 DH |
1631 | } |
1632 | return shrink_vma(mm, vma, start, end); | |
1633 | } | |
1da177e4 | 1634 | |
8feae131 DH |
1635 | erase_whole_vma: |
1636 | delete_vma_from_mm(vma); | |
1637 | delete_vma(mm, vma); | |
1da177e4 LT |
1638 | return 0; |
1639 | } | |
b5073173 | 1640 | EXPORT_SYMBOL(do_munmap); |
1da177e4 | 1641 | |
bfce281c | 1642 | int vm_munmap(unsigned long addr, size_t len) |
3034097a | 1643 | { |
bfce281c | 1644 | struct mm_struct *mm = current->mm; |
3034097a | 1645 | int ret; |
3034097a DH |
1646 | |
1647 | down_write(&mm->mmap_sem); | |
897ab3e0 | 1648 | ret = do_munmap(mm, addr, len, NULL); |
3034097a DH |
1649 | up_write(&mm->mmap_sem); |
1650 | return ret; | |
1651 | } | |
a46ef99d LT |
1652 | EXPORT_SYMBOL(vm_munmap); |
1653 | ||
1654 | SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len) | |
1655 | { | |
bfce281c | 1656 | return vm_munmap(addr, len); |
a46ef99d | 1657 | } |
3034097a DH |
1658 | |
1659 | /* | |
8feae131 | 1660 | * release all the mappings made in a process's VM space |
3034097a | 1661 | */ |
8feae131 | 1662 | void exit_mmap(struct mm_struct *mm) |
1da177e4 | 1663 | { |
8feae131 | 1664 | struct vm_area_struct *vma; |
1da177e4 | 1665 | |
8feae131 DH |
1666 | if (!mm) |
1667 | return; | |
1da177e4 | 1668 | |
8feae131 | 1669 | mm->total_vm = 0; |
1da177e4 | 1670 | |
8feae131 DH |
1671 | while ((vma = mm->mmap)) { |
1672 | mm->mmap = vma->vm_next; | |
1673 | delete_vma_from_mm(vma); | |
1674 | delete_vma(mm, vma); | |
04c34961 | 1675 | cond_resched(); |
1da177e4 LT |
1676 | } |
1677 | } | |
1678 | ||
5d22fc25 | 1679 | int vm_brk(unsigned long addr, unsigned long len) |
1da177e4 LT |
1680 | { |
1681 | return -ENOMEM; | |
1682 | } | |
1683 | ||
1684 | /* | |
6fa5f80b DH |
1685 | * expand (or shrink) an existing mapping, potentially moving it at the same |
1686 | * time (controlled by the MREMAP_MAYMOVE flag and available VM space) | |
1da177e4 | 1687 | * |
6fa5f80b | 1688 | * under NOMMU conditions, we only permit changing a mapping's size, and only |
8feae131 DH |
1689 | * as long as it stays within the region allocated by do_mmap_private() and the |
1690 | * block is not shareable | |
1da177e4 | 1691 | * |
6fa5f80b | 1692 | * MREMAP_FIXED is not supported under NOMMU conditions |
1da177e4 | 1693 | */ |
4b377bab | 1694 | static unsigned long do_mremap(unsigned long addr, |
1da177e4 LT |
1695 | unsigned long old_len, unsigned long new_len, |
1696 | unsigned long flags, unsigned long new_addr) | |
1697 | { | |
6fa5f80b | 1698 | struct vm_area_struct *vma; |
1da177e4 LT |
1699 | |
1700 | /* insanity checks first */ | |
f67d9b15 BL |
1701 | old_len = PAGE_ALIGN(old_len); |
1702 | new_len = PAGE_ALIGN(new_len); | |
8feae131 | 1703 | if (old_len == 0 || new_len == 0) |
1da177e4 LT |
1704 | return (unsigned long) -EINVAL; |
1705 | ||
1824cb75 | 1706 | if (offset_in_page(addr)) |
8feae131 DH |
1707 | return -EINVAL; |
1708 | ||
1da177e4 LT |
1709 | if (flags & MREMAP_FIXED && new_addr != addr) |
1710 | return (unsigned long) -EINVAL; | |
1711 | ||
8feae131 | 1712 | vma = find_vma_exact(current->mm, addr, old_len); |
6fa5f80b DH |
1713 | if (!vma) |
1714 | return (unsigned long) -EINVAL; | |
1da177e4 | 1715 | |
6fa5f80b | 1716 | if (vma->vm_end != vma->vm_start + old_len) |
1da177e4 LT |
1717 | return (unsigned long) -EFAULT; |
1718 | ||
6fa5f80b | 1719 | if (vma->vm_flags & VM_MAYSHARE) |
1da177e4 LT |
1720 | return (unsigned long) -EPERM; |
1721 | ||
8feae131 | 1722 | if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start) |
1da177e4 LT |
1723 | return (unsigned long) -ENOMEM; |
1724 | ||
1725 | /* all checks complete - do it */ | |
6fa5f80b | 1726 | vma->vm_end = vma->vm_start + new_len; |
6fa5f80b DH |
1727 | return vma->vm_start; |
1728 | } | |
1729 | ||
6a6160a7 HC |
1730 | SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len, |
1731 | unsigned long, new_len, unsigned long, flags, | |
1732 | unsigned long, new_addr) | |
6fa5f80b DH |
1733 | { |
1734 | unsigned long ret; | |
1735 | ||
1736 | down_write(¤t->mm->mmap_sem); | |
1737 | ret = do_mremap(addr, old_len, new_len, flags, new_addr); | |
1738 | up_write(¤t->mm->mmap_sem); | |
1739 | return ret; | |
1da177e4 LT |
1740 | } |
1741 | ||
240aadee ML |
1742 | struct page *follow_page_mask(struct vm_area_struct *vma, |
1743 | unsigned long address, unsigned int flags, | |
1744 | unsigned int *page_mask) | |
1da177e4 | 1745 | { |
240aadee | 1746 | *page_mask = 0; |
1da177e4 LT |
1747 | return NULL; |
1748 | } | |
1749 | ||
8f3b1327 BL |
1750 | int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr, |
1751 | unsigned long pfn, unsigned long size, pgprot_t prot) | |
1da177e4 | 1752 | { |
8f3b1327 BL |
1753 | if (addr != (pfn << PAGE_SHIFT)) |
1754 | return -EINVAL; | |
1755 | ||
314e51b9 | 1756 | vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP; |
66aa2b4b | 1757 | return 0; |
1da177e4 | 1758 | } |
22c4af40 | 1759 | EXPORT_SYMBOL(remap_pfn_range); |
1da177e4 | 1760 | |
3c0b9de6 LT |
1761 | int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len) |
1762 | { | |
1763 | unsigned long pfn = start >> PAGE_SHIFT; | |
1764 | unsigned long vm_len = vma->vm_end - vma->vm_start; | |
1765 | ||
1766 | pfn += vma->vm_pgoff; | |
1767 | return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot); | |
1768 | } | |
1769 | EXPORT_SYMBOL(vm_iomap_memory); | |
1770 | ||
f905bc44 PM |
1771 | int remap_vmalloc_range(struct vm_area_struct *vma, void *addr, |
1772 | unsigned long pgoff) | |
1773 | { | |
1774 | unsigned int size = vma->vm_end - vma->vm_start; | |
1775 | ||
1776 | if (!(vma->vm_flags & VM_USERMAP)) | |
1777 | return -EINVAL; | |
1778 | ||
1779 | vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT)); | |
1780 | vma->vm_end = vma->vm_start + size; | |
1781 | ||
1782 | return 0; | |
1783 | } | |
1784 | EXPORT_SYMBOL(remap_vmalloc_range); | |
1785 | ||
1da177e4 LT |
1786 | unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr, |
1787 | unsigned long len, unsigned long pgoff, unsigned long flags) | |
1788 | { | |
1789 | return -ENOMEM; | |
1790 | } | |
1791 | ||
1da177e4 LT |
1792 | void unmap_mapping_range(struct address_space *mapping, |
1793 | loff_t const holebegin, loff_t const holelen, | |
1794 | int even_cows) | |
1795 | { | |
1796 | } | |
22c4af40 | 1797 | EXPORT_SYMBOL(unmap_mapping_range); |
1da177e4 | 1798 | |
11bac800 | 1799 | int filemap_fault(struct vm_fault *vmf) |
b0e15190 DH |
1800 | { |
1801 | BUG(); | |
d0217ac0 | 1802 | return 0; |
b0e15190 | 1803 | } |
b5073173 | 1804 | EXPORT_SYMBOL(filemap_fault); |
0ec76a11 | 1805 | |
82b0f8c3 | 1806 | void filemap_map_pages(struct vm_fault *vmf, |
bae473a4 | 1807 | pgoff_t start_pgoff, pgoff_t end_pgoff) |
f1820361 KS |
1808 | { |
1809 | BUG(); | |
1810 | } | |
1811 | EXPORT_SYMBOL(filemap_map_pages); | |
1812 | ||
84d77d3f | 1813 | int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm, |
442486ec | 1814 | unsigned long addr, void *buf, int len, unsigned int gup_flags) |
0ec76a11 | 1815 | { |
0ec76a11 | 1816 | struct vm_area_struct *vma; |
442486ec | 1817 | int write = gup_flags & FOLL_WRITE; |
0ec76a11 DH |
1818 | |
1819 | down_read(&mm->mmap_sem); | |
1820 | ||
1821 | /* the access must start within one of the target process's mappings */ | |
0159b141 DH |
1822 | vma = find_vma(mm, addr); |
1823 | if (vma) { | |
0ec76a11 DH |
1824 | /* don't overrun this mapping */ |
1825 | if (addr + len >= vma->vm_end) | |
1826 | len = vma->vm_end - addr; | |
1827 | ||
1828 | /* only read or write mappings where it is permitted */ | |
d00c7b99 | 1829 | if (write && vma->vm_flags & VM_MAYWRITE) |
7959722b JZ |
1830 | copy_to_user_page(vma, NULL, addr, |
1831 | (void *) addr, buf, len); | |
d00c7b99 | 1832 | else if (!write && vma->vm_flags & VM_MAYREAD) |
7959722b JZ |
1833 | copy_from_user_page(vma, NULL, addr, |
1834 | buf, (void *) addr, len); | |
0ec76a11 DH |
1835 | else |
1836 | len = 0; | |
1837 | } else { | |
1838 | len = 0; | |
1839 | } | |
1840 | ||
1841 | up_read(&mm->mmap_sem); | |
f55f199b MF |
1842 | |
1843 | return len; | |
1844 | } | |
1845 | ||
1846 | /** | |
1847 | * @access_remote_vm - access another process' address space | |
1848 | * @mm: the mm_struct of the target address space | |
1849 | * @addr: start address to access | |
1850 | * @buf: source or destination buffer | |
1851 | * @len: number of bytes to transfer | |
6347e8d5 | 1852 | * @gup_flags: flags modifying lookup behaviour |
f55f199b MF |
1853 | * |
1854 | * The caller must hold a reference on @mm. | |
1855 | */ | |
1856 | int access_remote_vm(struct mm_struct *mm, unsigned long addr, | |
6347e8d5 | 1857 | void *buf, int len, unsigned int gup_flags) |
f55f199b | 1858 | { |
6347e8d5 | 1859 | return __access_remote_vm(NULL, mm, addr, buf, len, gup_flags); |
f55f199b MF |
1860 | } |
1861 | ||
1862 | /* | |
1863 | * Access another process' address space. | |
1864 | * - source/target buffer must be kernel space | |
1865 | */ | |
f307ab6d LS |
1866 | int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, |
1867 | unsigned int gup_flags) | |
f55f199b MF |
1868 | { |
1869 | struct mm_struct *mm; | |
1870 | ||
1871 | if (addr + len < addr) | |
1872 | return 0; | |
1873 | ||
1874 | mm = get_task_mm(tsk); | |
1875 | if (!mm) | |
1876 | return 0; | |
1877 | ||
f307ab6d | 1878 | len = __access_remote_vm(tsk, mm, addr, buf, len, gup_flags); |
f55f199b | 1879 | |
0ec76a11 DH |
1880 | mmput(mm); |
1881 | return len; | |
1882 | } | |
fcd35857 | 1883 | EXPORT_SYMBOL_GPL(access_process_vm); |
7e660872 DH |
1884 | |
1885 | /** | |
1886 | * nommu_shrink_inode_mappings - Shrink the shared mappings on an inode | |
1887 | * @inode: The inode to check | |
1888 | * @size: The current filesize of the inode | |
1889 | * @newsize: The proposed filesize of the inode | |
1890 | * | |
1891 | * Check the shared mappings on an inode on behalf of a shrinking truncate to | |
1892 | * make sure that that any outstanding VMAs aren't broken and then shrink the | |
1893 | * vm_regions that extend that beyond so that do_mmap_pgoff() doesn't | |
1894 | * automatically grant mappings that are too large. | |
1895 | */ | |
1896 | int nommu_shrink_inode_mappings(struct inode *inode, size_t size, | |
1897 | size_t newsize) | |
1898 | { | |
1899 | struct vm_area_struct *vma; | |
7e660872 DH |
1900 | struct vm_region *region; |
1901 | pgoff_t low, high; | |
1902 | size_t r_size, r_top; | |
1903 | ||
1904 | low = newsize >> PAGE_SHIFT; | |
1905 | high = (size + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
1906 | ||
1907 | down_write(&nommu_region_sem); | |
1acf2e04 | 1908 | i_mmap_lock_read(inode->i_mapping); |
7e660872 DH |
1909 | |
1910 | /* search for VMAs that fall within the dead zone */ | |
6b2dbba8 | 1911 | vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, low, high) { |
7e660872 DH |
1912 | /* found one - only interested if it's shared out of the page |
1913 | * cache */ | |
1914 | if (vma->vm_flags & VM_SHARED) { | |
1acf2e04 | 1915 | i_mmap_unlock_read(inode->i_mapping); |
7e660872 DH |
1916 | up_write(&nommu_region_sem); |
1917 | return -ETXTBSY; /* not quite true, but near enough */ | |
1918 | } | |
1919 | } | |
1920 | ||
1921 | /* reduce any regions that overlap the dead zone - if in existence, | |
1922 | * these will be pointed to by VMAs that don't overlap the dead zone | |
1923 | * | |
1924 | * we don't check for any regions that start beyond the EOF as there | |
1925 | * shouldn't be any | |
1926 | */ | |
1acf2e04 | 1927 | vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, 0, ULONG_MAX) { |
7e660872 DH |
1928 | if (!(vma->vm_flags & VM_SHARED)) |
1929 | continue; | |
1930 | ||
1931 | region = vma->vm_region; | |
1932 | r_size = region->vm_top - region->vm_start; | |
1933 | r_top = (region->vm_pgoff << PAGE_SHIFT) + r_size; | |
1934 | ||
1935 | if (r_top > newsize) { | |
1936 | region->vm_top -= r_top - newsize; | |
1937 | if (region->vm_end > region->vm_top) | |
1938 | region->vm_end = region->vm_top; | |
1939 | } | |
1940 | } | |
1941 | ||
1acf2e04 | 1942 | i_mmap_unlock_read(inode->i_mapping); |
7e660872 DH |
1943 | up_write(&nommu_region_sem); |
1944 | return 0; | |
1945 | } | |
c9b1d098 AS |
1946 | |
1947 | /* | |
1948 | * Initialise sysctl_user_reserve_kbytes. | |
1949 | * | |
1950 | * This is intended to prevent a user from starting a single memory hogging | |
1951 | * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER | |
1952 | * mode. | |
1953 | * | |
1954 | * The default value is min(3% of free memory, 128MB) | |
1955 | * 128MB is enough to recover with sshd/login, bash, and top/kill. | |
1956 | */ | |
1957 | static int __meminit init_user_reserve(void) | |
1958 | { | |
1959 | unsigned long free_kbytes; | |
1960 | ||
1961 | free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); | |
1962 | ||
1963 | sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17); | |
1964 | return 0; | |
1965 | } | |
a4bc6fc7 | 1966 | subsys_initcall(init_user_reserve); |
4eeab4f5 AS |
1967 | |
1968 | /* | |
1969 | * Initialise sysctl_admin_reserve_kbytes. | |
1970 | * | |
1971 | * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin | |
1972 | * to log in and kill a memory hogging process. | |
1973 | * | |
1974 | * Systems with more than 256MB will reserve 8MB, enough to recover | |
1975 | * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will | |
1976 | * only reserve 3% of free pages by default. | |
1977 | */ | |
1978 | static int __meminit init_admin_reserve(void) | |
1979 | { | |
1980 | unsigned long free_kbytes; | |
1981 | ||
1982 | free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); | |
1983 | ||
1984 | sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13); | |
1985 | return 0; | |
1986 | } | |
a4bc6fc7 | 1987 | subsys_initcall(init_admin_reserve); |