Commit | Line | Data |
---|---|---|
457c8996 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
1da177e4 LT |
2 | /* |
3 | * linux/mm/nommu.c | |
4 | * | |
5 | * Replacement code for mm functions to support CPU's that don't | |
6 | * have any form of memory management unit (thus no virtual memory). | |
7 | * | |
dd19d293 | 8 | * See Documentation/admin-guide/mm/nommu-mmap.rst |
1da177e4 | 9 | * |
8feae131 | 10 | * Copyright (c) 2004-2008 David Howells <dhowells@redhat.com> |
1da177e4 LT |
11 | * Copyright (c) 2000-2003 David McCullough <davidm@snapgear.com> |
12 | * Copyright (c) 2000-2001 D Jeff Dionne <jeff@uClinux.org> | |
13 | * Copyright (c) 2002 Greg Ungerer <gerg@snapgear.com> | |
29c185e5 | 14 | * Copyright (c) 2007-2010 Paul Mundt <lethal@linux-sh.org> |
1da177e4 LT |
15 | */ |
16 | ||
b1de0d13 MH |
17 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
18 | ||
b95f1b31 | 19 | #include <linux/export.h> |
1da177e4 | 20 | #include <linux/mm.h> |
6e84f315 | 21 | #include <linux/sched/mm.h> |
615d6e87 | 22 | #include <linux/vmacache.h> |
1da177e4 LT |
23 | #include <linux/mman.h> |
24 | #include <linux/swap.h> | |
25 | #include <linux/file.h> | |
26 | #include <linux/highmem.h> | |
27 | #include <linux/pagemap.h> | |
28 | #include <linux/slab.h> | |
29 | #include <linux/vmalloc.h> | |
1da177e4 | 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 | */ |
a50b854e | 110 | return page_size(page); |
1da177e4 LT |
111 | } |
112 | ||
dfc2f91a PM |
113 | /** |
114 | * follow_pfn - look up PFN at a user virtual address | |
115 | * @vma: memory mapping | |
116 | * @address: user virtual address | |
117 | * @pfn: location to store found PFN | |
118 | * | |
119 | * Only IO mappings and raw PFN mappings are allowed. | |
120 | * | |
121 | * Returns zero and the pfn at @pfn on success, -ve otherwise. | |
122 | */ | |
123 | int follow_pfn(struct vm_area_struct *vma, unsigned long address, | |
124 | unsigned long *pfn) | |
125 | { | |
126 | if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) | |
127 | return -EINVAL; | |
128 | ||
129 | *pfn = address >> PAGE_SHIFT; | |
130 | return 0; | |
131 | } | |
132 | EXPORT_SYMBOL(follow_pfn); | |
133 | ||
f1c4069e | 134 | LIST_HEAD(vmap_area_list); |
1da177e4 | 135 | |
b3bdda02 | 136 | void vfree(const void *addr) |
1da177e4 LT |
137 | { |
138 | kfree(addr); | |
139 | } | |
b5073173 | 140 | EXPORT_SYMBOL(vfree); |
1da177e4 | 141 | |
88dca4ca | 142 | void *__vmalloc(unsigned long size, gfp_t gfp_mask) |
1da177e4 LT |
143 | { |
144 | /* | |
8518609d RD |
145 | * You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc() |
146 | * returns only a logical address. | |
1da177e4 | 147 | */ |
84097518 | 148 | return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM); |
1da177e4 | 149 | } |
b5073173 | 150 | EXPORT_SYMBOL(__vmalloc); |
1da177e4 | 151 | |
041de93f CH |
152 | void *__vmalloc_node_range(unsigned long size, unsigned long align, |
153 | unsigned long start, unsigned long end, gfp_t gfp_mask, | |
154 | pgprot_t prot, unsigned long vm_flags, int node, | |
155 | const void *caller) | |
156 | { | |
157 | return __vmalloc(size, gfp_mask); | |
158 | } | |
159 | ||
2b905948 CH |
160 | void *__vmalloc_node(unsigned long size, unsigned long align, gfp_t gfp_mask, |
161 | int node, const void *caller) | |
a7c3e901 | 162 | { |
2b905948 | 163 | return __vmalloc(size, gfp_mask); |
a7c3e901 MH |
164 | } |
165 | ||
ed81745a | 166 | static void *__vmalloc_user_flags(unsigned long size, gfp_t flags) |
f905bc44 PM |
167 | { |
168 | void *ret; | |
169 | ||
88dca4ca | 170 | ret = __vmalloc(size, flags); |
f905bc44 PM |
171 | if (ret) { |
172 | struct vm_area_struct *vma; | |
173 | ||
d8ed45c5 | 174 | mmap_write_lock(current->mm); |
f905bc44 PM |
175 | vma = find_vma(current->mm, (unsigned long)ret); |
176 | if (vma) | |
177 | vma->vm_flags |= VM_USERMAP; | |
d8ed45c5 | 178 | mmap_write_unlock(current->mm); |
f905bc44 PM |
179 | } |
180 | ||
181 | return ret; | |
182 | } | |
ed81745a AN |
183 | |
184 | void *vmalloc_user(unsigned long size) | |
185 | { | |
186 | return __vmalloc_user_flags(size, GFP_KERNEL | __GFP_ZERO); | |
187 | } | |
f905bc44 PM |
188 | EXPORT_SYMBOL(vmalloc_user); |
189 | ||
b3bdda02 | 190 | struct page *vmalloc_to_page(const void *addr) |
1da177e4 LT |
191 | { |
192 | return virt_to_page(addr); | |
193 | } | |
b5073173 | 194 | EXPORT_SYMBOL(vmalloc_to_page); |
1da177e4 | 195 | |
b3bdda02 | 196 | unsigned long vmalloc_to_pfn(const void *addr) |
1da177e4 LT |
197 | { |
198 | return page_to_pfn(virt_to_page(addr)); | |
199 | } | |
b5073173 | 200 | EXPORT_SYMBOL(vmalloc_to_pfn); |
1da177e4 LT |
201 | |
202 | long vread(char *buf, char *addr, unsigned long count) | |
203 | { | |
9bde916b CG |
204 | /* Don't allow overflow */ |
205 | if ((unsigned long) buf + count < count) | |
206 | count = -(unsigned long) buf; | |
207 | ||
1da177e4 LT |
208 | memcpy(buf, addr, count); |
209 | return count; | |
210 | } | |
211 | ||
1da177e4 | 212 | /* |
e1c05067 | 213 | * vmalloc - allocate virtually contiguous memory |
1da177e4 LT |
214 | * |
215 | * @size: allocation size | |
216 | * | |
217 | * Allocate enough pages to cover @size from the page level | |
e1c05067 | 218 | * allocator and map them into contiguous kernel virtual space. |
1da177e4 | 219 | * |
c1c8897f | 220 | * For tight control over page level allocator and protection flags |
1da177e4 LT |
221 | * use __vmalloc() instead. |
222 | */ | |
223 | void *vmalloc(unsigned long size) | |
224 | { | |
176056fd | 225 | return __vmalloc(size, GFP_KERNEL); |
1da177e4 | 226 | } |
f6138882 AM |
227 | EXPORT_SYMBOL(vmalloc); |
228 | ||
0fc74d82 LT |
229 | void *vmalloc_huge(unsigned long size, gfp_t gfp_mask) __weak __alias(__vmalloc); |
230 | ||
e1ca7788 | 231 | /* |
e1c05067 | 232 | * vzalloc - allocate virtually contiguous memory with zero fill |
e1ca7788 DY |
233 | * |
234 | * @size: allocation size | |
235 | * | |
236 | * Allocate enough pages to cover @size from the page level | |
e1c05067 | 237 | * allocator and map them into contiguous kernel virtual space. |
e1ca7788 DY |
238 | * The memory allocated is set to zero. |
239 | * | |
240 | * For tight control over page level allocator and protection flags | |
241 | * use __vmalloc() instead. | |
242 | */ | |
243 | void *vzalloc(unsigned long size) | |
244 | { | |
176056fd | 245 | return __vmalloc(size, GFP_KERNEL | __GFP_ZERO); |
e1ca7788 DY |
246 | } |
247 | EXPORT_SYMBOL(vzalloc); | |
248 | ||
249 | /** | |
250 | * vmalloc_node - allocate memory on a specific node | |
251 | * @size: allocation size | |
252 | * @node: numa node | |
253 | * | |
254 | * Allocate enough pages to cover @size from the page level | |
255 | * allocator and map them into contiguous kernel virtual space. | |
256 | * | |
257 | * For tight control over page level allocator and protection flags | |
258 | * use __vmalloc() instead. | |
259 | */ | |
f6138882 AM |
260 | void *vmalloc_node(unsigned long size, int node) |
261 | { | |
262 | return vmalloc(size); | |
263 | } | |
9a14f653 | 264 | EXPORT_SYMBOL(vmalloc_node); |
e1ca7788 DY |
265 | |
266 | /** | |
267 | * vzalloc_node - allocate memory on a specific node with zero fill | |
268 | * @size: allocation size | |
269 | * @node: numa node | |
270 | * | |
271 | * Allocate enough pages to cover @size from the page level | |
272 | * allocator and map them into contiguous kernel virtual space. | |
273 | * The memory allocated is set to zero. | |
274 | * | |
275 | * For tight control over page level allocator and protection flags | |
276 | * use __vmalloc() instead. | |
277 | */ | |
278 | void *vzalloc_node(unsigned long size, int node) | |
279 | { | |
280 | return vzalloc(size); | |
281 | } | |
282 | EXPORT_SYMBOL(vzalloc_node); | |
1da177e4 | 283 | |
b5073173 PM |
284 | /** |
285 | * vmalloc_32 - allocate virtually contiguous memory (32bit addressable) | |
1da177e4 LT |
286 | * @size: allocation size |
287 | * | |
288 | * Allocate enough 32bit PA addressable pages to cover @size from the | |
e1c05067 | 289 | * page level allocator and map them into contiguous kernel virtual space. |
1da177e4 LT |
290 | */ |
291 | void *vmalloc_32(unsigned long size) | |
292 | { | |
88dca4ca | 293 | return __vmalloc(size, GFP_KERNEL); |
1da177e4 | 294 | } |
b5073173 PM |
295 | EXPORT_SYMBOL(vmalloc_32); |
296 | ||
297 | /** | |
298 | * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory | |
299 | * @size: allocation size | |
300 | * | |
301 | * The resulting memory area is 32bit addressable and zeroed so it can be | |
302 | * mapped to userspace without leaking data. | |
f905bc44 PM |
303 | * |
304 | * VM_USERMAP is set on the corresponding VMA so that subsequent calls to | |
305 | * remap_vmalloc_range() are permissible. | |
b5073173 PM |
306 | */ |
307 | void *vmalloc_32_user(unsigned long size) | |
308 | { | |
f905bc44 PM |
309 | /* |
310 | * We'll have to sort out the ZONE_DMA bits for 64-bit, | |
311 | * but for now this can simply use vmalloc_user() directly. | |
312 | */ | |
313 | return vmalloc_user(size); | |
b5073173 PM |
314 | } |
315 | EXPORT_SYMBOL(vmalloc_32_user); | |
1da177e4 LT |
316 | |
317 | void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot) | |
318 | { | |
319 | BUG(); | |
320 | return NULL; | |
321 | } | |
b5073173 | 322 | EXPORT_SYMBOL(vmap); |
1da177e4 | 323 | |
b3bdda02 | 324 | void vunmap(const void *addr) |
1da177e4 LT |
325 | { |
326 | BUG(); | |
327 | } | |
b5073173 | 328 | EXPORT_SYMBOL(vunmap); |
1da177e4 | 329 | |
d4efd79a | 330 | void *vm_map_ram(struct page **pages, unsigned int count, int node) |
eb6434d9 PM |
331 | { |
332 | BUG(); | |
333 | return NULL; | |
334 | } | |
335 | EXPORT_SYMBOL(vm_map_ram); | |
336 | ||
337 | void vm_unmap_ram(const void *mem, unsigned int count) | |
338 | { | |
339 | BUG(); | |
340 | } | |
341 | EXPORT_SYMBOL(vm_unmap_ram); | |
342 | ||
343 | void vm_unmap_aliases(void) | |
344 | { | |
345 | } | |
346 | EXPORT_SYMBOL_GPL(vm_unmap_aliases); | |
347 | ||
29c185e5 PM |
348 | void free_vm_area(struct vm_struct *area) |
349 | { | |
350 | BUG(); | |
351 | } | |
352 | EXPORT_SYMBOL_GPL(free_vm_area); | |
353 | ||
b5073173 PM |
354 | int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, |
355 | struct page *page) | |
356 | { | |
357 | return -EINVAL; | |
358 | } | |
359 | EXPORT_SYMBOL(vm_insert_page); | |
360 | ||
a667d745 SJ |
361 | int vm_map_pages(struct vm_area_struct *vma, struct page **pages, |
362 | unsigned long num) | |
363 | { | |
364 | return -EINVAL; | |
365 | } | |
366 | EXPORT_SYMBOL(vm_map_pages); | |
367 | ||
368 | int vm_map_pages_zero(struct vm_area_struct *vma, struct page **pages, | |
369 | unsigned long num) | |
370 | { | |
371 | return -EINVAL; | |
372 | } | |
373 | EXPORT_SYMBOL(vm_map_pages_zero); | |
374 | ||
1da177e4 LT |
375 | /* |
376 | * sys_brk() for the most part doesn't need the global kernel | |
377 | * lock, except when an application is doing something nasty | |
378 | * like trying to un-brk an area that has already been mapped | |
379 | * to a regular file. in this case, the unmapping will need | |
380 | * to invoke file system routines that need the global lock. | |
381 | */ | |
6a6160a7 | 382 | SYSCALL_DEFINE1(brk, unsigned long, brk) |
1da177e4 LT |
383 | { |
384 | struct mm_struct *mm = current->mm; | |
385 | ||
386 | if (brk < mm->start_brk || brk > mm->context.end_brk) | |
387 | return mm->brk; | |
388 | ||
389 | if (mm->brk == brk) | |
390 | return mm->brk; | |
391 | ||
392 | /* | |
393 | * Always allow shrinking brk | |
394 | */ | |
395 | if (brk <= mm->brk) { | |
396 | mm->brk = brk; | |
397 | return brk; | |
398 | } | |
399 | ||
400 | /* | |
401 | * Ok, looks good - let it rip. | |
402 | */ | |
a75a2df6 | 403 | flush_icache_user_range(mm->brk, brk); |
1da177e4 LT |
404 | return mm->brk = brk; |
405 | } | |
406 | ||
8feae131 | 407 | /* |
3edf41d8 | 408 | * initialise the percpu counter for VM and region record slabs |
8feae131 DH |
409 | */ |
410 | void __init mmap_init(void) | |
1da177e4 | 411 | { |
00a62ce9 KM |
412 | int ret; |
413 | ||
908c7f19 | 414 | ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL); |
00a62ce9 | 415 | VM_BUG_ON(ret); |
5d097056 | 416 | vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC|SLAB_ACCOUNT); |
1da177e4 | 417 | } |
1da177e4 | 418 | |
3034097a | 419 | /* |
8feae131 DH |
420 | * validate the region tree |
421 | * - the caller must hold the region lock | |
3034097a | 422 | */ |
8feae131 DH |
423 | #ifdef CONFIG_DEBUG_NOMMU_REGIONS |
424 | static noinline void validate_nommu_regions(void) | |
3034097a | 425 | { |
8feae131 DH |
426 | struct vm_region *region, *last; |
427 | struct rb_node *p, *lastp; | |
3034097a | 428 | |
8feae131 DH |
429 | lastp = rb_first(&nommu_region_tree); |
430 | if (!lastp) | |
431 | return; | |
432 | ||
433 | last = rb_entry(lastp, struct vm_region, vm_rb); | |
c9427bc0 GT |
434 | BUG_ON(last->vm_end <= last->vm_start); |
435 | BUG_ON(last->vm_top < last->vm_end); | |
8feae131 DH |
436 | |
437 | while ((p = rb_next(lastp))) { | |
438 | region = rb_entry(p, struct vm_region, vm_rb); | |
439 | last = rb_entry(lastp, struct vm_region, vm_rb); | |
440 | ||
c9427bc0 GT |
441 | BUG_ON(region->vm_end <= region->vm_start); |
442 | BUG_ON(region->vm_top < region->vm_end); | |
443 | BUG_ON(region->vm_start < last->vm_top); | |
3034097a | 444 | |
8feae131 DH |
445 | lastp = p; |
446 | } | |
3034097a | 447 | } |
8feae131 | 448 | #else |
33e5d769 DH |
449 | static void validate_nommu_regions(void) |
450 | { | |
451 | } | |
8feae131 | 452 | #endif |
3034097a DH |
453 | |
454 | /* | |
8feae131 | 455 | * add a region into the global tree |
3034097a | 456 | */ |
8feae131 | 457 | static void add_nommu_region(struct vm_region *region) |
3034097a | 458 | { |
8feae131 DH |
459 | struct vm_region *pregion; |
460 | struct rb_node **p, *parent; | |
3034097a | 461 | |
8feae131 DH |
462 | validate_nommu_regions(); |
463 | ||
8feae131 DH |
464 | parent = NULL; |
465 | p = &nommu_region_tree.rb_node; | |
466 | while (*p) { | |
467 | parent = *p; | |
468 | pregion = rb_entry(parent, struct vm_region, vm_rb); | |
469 | if (region->vm_start < pregion->vm_start) | |
470 | p = &(*p)->rb_left; | |
471 | else if (region->vm_start > pregion->vm_start) | |
472 | p = &(*p)->rb_right; | |
473 | else if (pregion == region) | |
474 | return; | |
475 | else | |
476 | BUG(); | |
3034097a DH |
477 | } |
478 | ||
8feae131 DH |
479 | rb_link_node(®ion->vm_rb, parent, p); |
480 | rb_insert_color(®ion->vm_rb, &nommu_region_tree); | |
3034097a | 481 | |
8feae131 | 482 | validate_nommu_regions(); |
3034097a | 483 | } |
3034097a | 484 | |
930e652a | 485 | /* |
8feae131 | 486 | * delete a region from the global tree |
930e652a | 487 | */ |
8feae131 | 488 | static void delete_nommu_region(struct vm_region *region) |
930e652a | 489 | { |
8feae131 | 490 | BUG_ON(!nommu_region_tree.rb_node); |
930e652a | 491 | |
8feae131 DH |
492 | validate_nommu_regions(); |
493 | rb_erase(®ion->vm_rb, &nommu_region_tree); | |
494 | validate_nommu_regions(); | |
57c8f63e GU |
495 | } |
496 | ||
6fa5f80b | 497 | /* |
8feae131 | 498 | * free a contiguous series of pages |
6fa5f80b | 499 | */ |
8feae131 | 500 | static void free_page_series(unsigned long from, unsigned long to) |
6fa5f80b | 501 | { |
8feae131 | 502 | for (; from < to; from += PAGE_SIZE) { |
9330723c | 503 | struct page *page = virt_to_page((void *)from); |
8feae131 | 504 | |
33e5d769 | 505 | atomic_long_dec(&mmap_pages_allocated); |
8feae131 | 506 | put_page(page); |
6fa5f80b | 507 | } |
6fa5f80b DH |
508 | } |
509 | ||
3034097a | 510 | /* |
8feae131 | 511 | * release a reference to a region |
33e5d769 | 512 | * - the caller must hold the region semaphore for writing, which this releases |
dd8632a1 | 513 | * - the region may not have been added to the tree yet, in which case vm_top |
8feae131 | 514 | * will equal vm_start |
3034097a | 515 | */ |
8feae131 DH |
516 | static void __put_nommu_region(struct vm_region *region) |
517 | __releases(nommu_region_sem) | |
1da177e4 | 518 | { |
8feae131 | 519 | BUG_ON(!nommu_region_tree.rb_node); |
1da177e4 | 520 | |
1e2ae599 | 521 | if (--region->vm_usage == 0) { |
dd8632a1 | 522 | if (region->vm_top > region->vm_start) |
8feae131 DH |
523 | delete_nommu_region(region); |
524 | up_write(&nommu_region_sem); | |
525 | ||
526 | if (region->vm_file) | |
527 | fput(region->vm_file); | |
528 | ||
529 | /* IO memory and memory shared directly out of the pagecache | |
530 | * from ramfs/tmpfs mustn't be released here */ | |
22cc877b | 531 | if (region->vm_flags & VM_MAPPED_COPY) |
dd8632a1 | 532 | free_page_series(region->vm_start, region->vm_top); |
8feae131 DH |
533 | kmem_cache_free(vm_region_jar, region); |
534 | } else { | |
535 | up_write(&nommu_region_sem); | |
1da177e4 | 536 | } |
8feae131 | 537 | } |
1da177e4 | 538 | |
8feae131 DH |
539 | /* |
540 | * release a reference to a region | |
541 | */ | |
542 | static void put_nommu_region(struct vm_region *region) | |
543 | { | |
544 | down_write(&nommu_region_sem); | |
545 | __put_nommu_region(region); | |
1da177e4 LT |
546 | } |
547 | ||
d4af56c5 LH |
548 | void vma_mas_store(struct vm_area_struct *vma, struct ma_state *mas) |
549 | { | |
550 | mas_set_range(mas, vma->vm_start, vma->vm_end - 1); | |
551 | mas_store_prealloc(mas, vma); | |
552 | } | |
553 | ||
554 | void vma_mas_remove(struct vm_area_struct *vma, struct ma_state *mas) | |
555 | { | |
556 | mas->index = vma->vm_start; | |
557 | mas->last = vma->vm_end - 1; | |
558 | mas_store_prealloc(mas, NULL); | |
559 | } | |
560 | ||
3034097a | 561 | /* |
8feae131 DH |
562 | * add a VMA into a process's mm_struct in the appropriate place in the list |
563 | * and tree and add to the address space's page tree also if not an anonymous | |
564 | * page | |
c1e8d7c6 | 565 | * - should be called with mm->mmap_lock held writelocked |
3034097a | 566 | */ |
8feae131 | 567 | static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma) |
1da177e4 | 568 | { |
1da177e4 | 569 | struct address_space *mapping; |
524e00b3 LH |
570 | struct vm_area_struct *prev; |
571 | MA_STATE(mas, &mm->mm_mt, vma->vm_start, vma->vm_end); | |
8feae131 | 572 | |
8feae131 DH |
573 | BUG_ON(!vma->vm_region); |
574 | ||
575 | mm->map_count++; | |
576 | vma->vm_mm = mm; | |
1da177e4 LT |
577 | |
578 | /* add the VMA to the mapping */ | |
579 | if (vma->vm_file) { | |
580 | mapping = vma->vm_file->f_mapping; | |
581 | ||
83cde9e8 | 582 | i_mmap_lock_write(mapping); |
1da177e4 | 583 | flush_dcache_mmap_lock(mapping); |
6b2dbba8 | 584 | vma_interval_tree_insert(vma, &mapping->i_mmap); |
1da177e4 | 585 | flush_dcache_mmap_unlock(mapping); |
83cde9e8 | 586 | i_mmap_unlock_write(mapping); |
1da177e4 LT |
587 | } |
588 | ||
524e00b3 LH |
589 | prev = mas_prev(&mas, 0); |
590 | mas_reset(&mas); | |
8feae131 | 591 | /* add the VMA to the tree */ |
524e00b3 | 592 | vma_mas_store(vma, &mas); |
aba6dfb7 | 593 | __vma_link_list(mm, vma, prev); |
1da177e4 LT |
594 | } |
595 | ||
3034097a | 596 | /* |
8feae131 | 597 | * delete a VMA from its owning mm_struct and address space |
3034097a | 598 | */ |
8feae131 | 599 | static void delete_vma_from_mm(struct vm_area_struct *vma) |
1da177e4 | 600 | { |
615d6e87 | 601 | int i; |
1da177e4 | 602 | struct address_space *mapping; |
8feae131 | 603 | struct mm_struct *mm = vma->vm_mm; |
615d6e87 | 604 | struct task_struct *curr = current; |
524e00b3 | 605 | MA_STATE(mas, &vma->vm_mm->mm_mt, 0, 0); |
8feae131 | 606 | |
8feae131 | 607 | mm->map_count--; |
615d6e87 DB |
608 | for (i = 0; i < VMACACHE_SIZE; i++) { |
609 | /* if the vma is cached, invalidate the entire cache */ | |
314ff785 | 610 | if (curr->vmacache.vmas[i] == vma) { |
e020d5bd | 611 | vmacache_invalidate(mm); |
615d6e87 DB |
612 | break; |
613 | } | |
614 | } | |
1da177e4 LT |
615 | |
616 | /* remove the VMA from the mapping */ | |
617 | if (vma->vm_file) { | |
618 | mapping = vma->vm_file->f_mapping; | |
619 | ||
83cde9e8 | 620 | i_mmap_lock_write(mapping); |
1da177e4 | 621 | flush_dcache_mmap_lock(mapping); |
6b2dbba8 | 622 | vma_interval_tree_remove(vma, &mapping->i_mmap); |
1da177e4 | 623 | flush_dcache_mmap_unlock(mapping); |
83cde9e8 | 624 | i_mmap_unlock_write(mapping); |
1da177e4 LT |
625 | } |
626 | ||
8feae131 | 627 | /* remove from the MM's tree and list */ |
524e00b3 | 628 | vma_mas_remove(vma, &mas); |
1b9fc5b2 | 629 | __vma_unlink_list(mm, vma); |
8feae131 DH |
630 | } |
631 | ||
632 | /* | |
633 | * destroy a VMA record | |
634 | */ | |
635 | static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma) | |
636 | { | |
8feae131 DH |
637 | if (vma->vm_ops && vma->vm_ops->close) |
638 | vma->vm_ops->close(vma); | |
e9714acf | 639 | if (vma->vm_file) |
8feae131 | 640 | fput(vma->vm_file); |
8feae131 | 641 | put_nommu_region(vma->vm_region); |
3928d4f5 | 642 | vm_area_free(vma); |
8feae131 DH |
643 | } |
644 | ||
645 | /* | |
646 | * look up the first VMA in which addr resides, NULL if none | |
c1e8d7c6 | 647 | * - should be called with mm->mmap_lock at least held readlocked |
8feae131 DH |
648 | */ |
649 | struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr) | |
650 | { | |
651 | struct vm_area_struct *vma; | |
524e00b3 | 652 | MA_STATE(mas, &mm->mm_mt, addr, addr); |
8feae131 DH |
653 | |
654 | /* check the cache first */ | |
615d6e87 DB |
655 | vma = vmacache_find(mm, addr); |
656 | if (likely(vma)) | |
8feae131 DH |
657 | return vma; |
658 | ||
524e00b3 | 659 | vma = mas_walk(&mas); |
8feae131 | 660 | |
524e00b3 LH |
661 | if (vma) |
662 | vmacache_update(addr, vma); | |
663 | ||
664 | return vma; | |
8feae131 DH |
665 | } |
666 | EXPORT_SYMBOL(find_vma); | |
667 | ||
668 | /* | |
669 | * find a VMA | |
670 | * - we don't extend stack VMAs under NOMMU conditions | |
671 | */ | |
672 | struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr) | |
673 | { | |
7561e8ca | 674 | return find_vma(mm, addr); |
8feae131 DH |
675 | } |
676 | ||
677 | /* | |
678 | * expand a stack to a given address | |
679 | * - not supported under NOMMU conditions | |
680 | */ | |
681 | int expand_stack(struct vm_area_struct *vma, unsigned long address) | |
682 | { | |
683 | return -ENOMEM; | |
684 | } | |
685 | ||
686 | /* | |
687 | * look up the first VMA exactly that exactly matches addr | |
c1e8d7c6 | 688 | * - should be called with mm->mmap_lock at least held readlocked |
8feae131 DH |
689 | */ |
690 | static struct vm_area_struct *find_vma_exact(struct mm_struct *mm, | |
691 | unsigned long addr, | |
692 | unsigned long len) | |
693 | { | |
694 | struct vm_area_struct *vma; | |
8feae131 | 695 | unsigned long end = addr + len; |
524e00b3 | 696 | MA_STATE(mas, &mm->mm_mt, addr, addr); |
8feae131 DH |
697 | |
698 | /* check the cache first */ | |
615d6e87 DB |
699 | vma = vmacache_find_exact(mm, addr, end); |
700 | if (vma) | |
8feae131 DH |
701 | return vma; |
702 | ||
524e00b3 LH |
703 | vma = mas_walk(&mas); |
704 | if (!vma) | |
705 | return NULL; | |
706 | if (vma->vm_start != addr) | |
707 | return NULL; | |
708 | if (vma->vm_end != end) | |
709 | return NULL; | |
710 | ||
711 | vmacache_update(addr, vma); | |
712 | return vma; | |
1da177e4 LT |
713 | } |
714 | ||
715 | /* | |
716 | * determine whether a mapping should be permitted and, if so, what sort of | |
717 | * mapping we're capable of supporting | |
718 | */ | |
719 | static int validate_mmap_request(struct file *file, | |
720 | unsigned long addr, | |
721 | unsigned long len, | |
722 | unsigned long prot, | |
723 | unsigned long flags, | |
724 | unsigned long pgoff, | |
725 | unsigned long *_capabilities) | |
726 | { | |
8feae131 | 727 | unsigned long capabilities, rlen; |
1da177e4 LT |
728 | int ret; |
729 | ||
730 | /* do the simple checks first */ | |
22cc877b | 731 | if (flags & MAP_FIXED) |
1da177e4 | 732 | return -EINVAL; |
1da177e4 LT |
733 | |
734 | if ((flags & MAP_TYPE) != MAP_PRIVATE && | |
735 | (flags & MAP_TYPE) != MAP_SHARED) | |
736 | return -EINVAL; | |
737 | ||
f81cff0d | 738 | if (!len) |
1da177e4 LT |
739 | return -EINVAL; |
740 | ||
f81cff0d | 741 | /* Careful about overflows.. */ |
8feae131 DH |
742 | rlen = PAGE_ALIGN(len); |
743 | if (!rlen || rlen > TASK_SIZE) | |
f81cff0d MF |
744 | return -ENOMEM; |
745 | ||
1da177e4 | 746 | /* offset overflow? */ |
8feae131 | 747 | if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff) |
f81cff0d | 748 | return -EOVERFLOW; |
1da177e4 LT |
749 | |
750 | if (file) { | |
1da177e4 | 751 | /* files must support mmap */ |
72c2d531 | 752 | if (!file->f_op->mmap) |
1da177e4 LT |
753 | return -ENODEV; |
754 | ||
755 | /* work out if what we've got could possibly be shared | |
756 | * - we support chardevs that provide their own "memory" | |
757 | * - we support files/blockdevs that are memory backed | |
758 | */ | |
b4caecd4 CH |
759 | if (file->f_op->mmap_capabilities) { |
760 | capabilities = file->f_op->mmap_capabilities(file); | |
761 | } else { | |
1da177e4 LT |
762 | /* no explicit capabilities set, so assume some |
763 | * defaults */ | |
496ad9aa | 764 | switch (file_inode(file)->i_mode & S_IFMT) { |
1da177e4 LT |
765 | case S_IFREG: |
766 | case S_IFBLK: | |
b4caecd4 | 767 | capabilities = NOMMU_MAP_COPY; |
1da177e4 LT |
768 | break; |
769 | ||
770 | case S_IFCHR: | |
771 | capabilities = | |
b4caecd4 CH |
772 | NOMMU_MAP_DIRECT | |
773 | NOMMU_MAP_READ | | |
774 | NOMMU_MAP_WRITE; | |
1da177e4 LT |
775 | break; |
776 | ||
777 | default: | |
778 | return -EINVAL; | |
779 | } | |
780 | } | |
781 | ||
782 | /* eliminate any capabilities that we can't support on this | |
783 | * device */ | |
784 | if (!file->f_op->get_unmapped_area) | |
b4caecd4 | 785 | capabilities &= ~NOMMU_MAP_DIRECT; |
6e242a1c | 786 | if (!(file->f_mode & FMODE_CAN_READ)) |
b4caecd4 | 787 | capabilities &= ~NOMMU_MAP_COPY; |
1da177e4 | 788 | |
28d7a6ae GY |
789 | /* The file shall have been opened with read permission. */ |
790 | if (!(file->f_mode & FMODE_READ)) | |
791 | return -EACCES; | |
792 | ||
1da177e4 LT |
793 | if (flags & MAP_SHARED) { |
794 | /* do checks for writing, appending and locking */ | |
795 | if ((prot & PROT_WRITE) && | |
796 | !(file->f_mode & FMODE_WRITE)) | |
797 | return -EACCES; | |
798 | ||
496ad9aa | 799 | if (IS_APPEND(file_inode(file)) && |
1da177e4 LT |
800 | (file->f_mode & FMODE_WRITE)) |
801 | return -EACCES; | |
802 | ||
b4caecd4 | 803 | if (!(capabilities & NOMMU_MAP_DIRECT)) |
1da177e4 LT |
804 | return -ENODEV; |
805 | ||
1da177e4 | 806 | /* we mustn't privatise shared mappings */ |
b4caecd4 | 807 | capabilities &= ~NOMMU_MAP_COPY; |
ac714904 | 808 | } else { |
1da177e4 LT |
809 | /* we're going to read the file into private memory we |
810 | * allocate */ | |
b4caecd4 | 811 | if (!(capabilities & NOMMU_MAP_COPY)) |
1da177e4 LT |
812 | return -ENODEV; |
813 | ||
814 | /* we don't permit a private writable mapping to be | |
815 | * shared with the backing device */ | |
816 | if (prot & PROT_WRITE) | |
b4caecd4 | 817 | capabilities &= ~NOMMU_MAP_DIRECT; |
1da177e4 LT |
818 | } |
819 | ||
b4caecd4 CH |
820 | if (capabilities & NOMMU_MAP_DIRECT) { |
821 | if (((prot & PROT_READ) && !(capabilities & NOMMU_MAP_READ)) || | |
822 | ((prot & PROT_WRITE) && !(capabilities & NOMMU_MAP_WRITE)) || | |
823 | ((prot & PROT_EXEC) && !(capabilities & NOMMU_MAP_EXEC)) | |
3c7b2045 | 824 | ) { |
b4caecd4 | 825 | capabilities &= ~NOMMU_MAP_DIRECT; |
3c7b2045 | 826 | if (flags & MAP_SHARED) { |
22cc877b | 827 | pr_warn("MAP_SHARED not completely supported on !MMU\n"); |
3c7b2045 BS |
828 | return -EINVAL; |
829 | } | |
830 | } | |
831 | } | |
832 | ||
1da177e4 LT |
833 | /* handle executable mappings and implied executable |
834 | * mappings */ | |
90f8572b | 835 | if (path_noexec(&file->f_path)) { |
1da177e4 LT |
836 | if (prot & PROT_EXEC) |
837 | return -EPERM; | |
ac714904 | 838 | } else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) { |
1da177e4 LT |
839 | /* handle implication of PROT_EXEC by PROT_READ */ |
840 | if (current->personality & READ_IMPLIES_EXEC) { | |
b4caecd4 | 841 | if (capabilities & NOMMU_MAP_EXEC) |
1da177e4 LT |
842 | prot |= PROT_EXEC; |
843 | } | |
ac714904 | 844 | } else if ((prot & PROT_READ) && |
1da177e4 | 845 | (prot & PROT_EXEC) && |
b4caecd4 | 846 | !(capabilities & NOMMU_MAP_EXEC) |
1da177e4 LT |
847 | ) { |
848 | /* backing file is not executable, try to copy */ | |
b4caecd4 | 849 | capabilities &= ~NOMMU_MAP_DIRECT; |
1da177e4 | 850 | } |
ac714904 | 851 | } else { |
1da177e4 LT |
852 | /* anonymous mappings are always memory backed and can be |
853 | * privately mapped | |
854 | */ | |
b4caecd4 | 855 | capabilities = NOMMU_MAP_COPY; |
1da177e4 LT |
856 | |
857 | /* handle PROT_EXEC implication by PROT_READ */ | |
858 | if ((prot & PROT_READ) && | |
859 | (current->personality & READ_IMPLIES_EXEC)) | |
860 | prot |= PROT_EXEC; | |
861 | } | |
862 | ||
863 | /* allow the security API to have its say */ | |
e5467859 | 864 | ret = security_mmap_addr(addr); |
1da177e4 LT |
865 | if (ret < 0) |
866 | return ret; | |
867 | ||
868 | /* looks okay */ | |
869 | *_capabilities = capabilities; | |
870 | return 0; | |
871 | } | |
872 | ||
873 | /* | |
874 | * we've determined that we can make the mapping, now translate what we | |
875 | * now know into VMA flags | |
876 | */ | |
877 | static unsigned long determine_vm_flags(struct file *file, | |
878 | unsigned long prot, | |
879 | unsigned long flags, | |
880 | unsigned long capabilities) | |
881 | { | |
882 | unsigned long vm_flags; | |
883 | ||
e6bfb709 | 884 | vm_flags = calc_vm_prot_bits(prot, 0) | calc_vm_flag_bits(flags); |
1da177e4 LT |
885 | /* vm_flags |= mm->def_flags; */ |
886 | ||
b4caecd4 | 887 | if (!(capabilities & NOMMU_MAP_DIRECT)) { |
1da177e4 | 888 | /* attempt to share read-only copies of mapped file chunks */ |
3c7b2045 | 889 | vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC; |
1da177e4 LT |
890 | if (file && !(prot & PROT_WRITE)) |
891 | vm_flags |= VM_MAYSHARE; | |
3c7b2045 | 892 | } else { |
1da177e4 LT |
893 | /* overlay a shareable mapping on the backing device or inode |
894 | * if possible - used for chardevs, ramfs/tmpfs/shmfs and | |
895 | * romfs/cramfs */ | |
b4caecd4 | 896 | vm_flags |= VM_MAYSHARE | (capabilities & NOMMU_VMFLAGS); |
1da177e4 | 897 | if (flags & MAP_SHARED) |
3c7b2045 | 898 | vm_flags |= VM_SHARED; |
1da177e4 LT |
899 | } |
900 | ||
901 | /* refuse to let anyone share private mappings with this process if | |
902 | * it's being traced - otherwise breakpoints set in it may interfere | |
903 | * with another untraced process | |
904 | */ | |
a288eecc | 905 | if ((flags & MAP_PRIVATE) && current->ptrace) |
1da177e4 LT |
906 | vm_flags &= ~VM_MAYSHARE; |
907 | ||
908 | return vm_flags; | |
909 | } | |
910 | ||
911 | /* | |
8feae131 DH |
912 | * set up a shared mapping on a file (the driver or filesystem provides and |
913 | * pins the storage) | |
1da177e4 | 914 | */ |
8feae131 | 915 | static int do_mmap_shared_file(struct vm_area_struct *vma) |
1da177e4 LT |
916 | { |
917 | int ret; | |
918 | ||
f74ac015 | 919 | ret = call_mmap(vma->vm_file, vma); |
dd8632a1 PM |
920 | if (ret == 0) { |
921 | vma->vm_region->vm_top = vma->vm_region->vm_end; | |
645d83c5 | 922 | return 0; |
dd8632a1 | 923 | } |
1da177e4 LT |
924 | if (ret != -ENOSYS) |
925 | return ret; | |
926 | ||
3fa30460 DH |
927 | /* getting -ENOSYS indicates that direct mmap isn't possible (as |
928 | * opposed to tried but failed) so we can only give a suitable error as | |
929 | * it's not possible to make a private copy if MAP_SHARED was given */ | |
1da177e4 LT |
930 | return -ENODEV; |
931 | } | |
932 | ||
933 | /* | |
934 | * set up a private mapping or an anonymous shared mapping | |
935 | */ | |
8feae131 DH |
936 | static int do_mmap_private(struct vm_area_struct *vma, |
937 | struct vm_region *region, | |
645d83c5 DH |
938 | unsigned long len, |
939 | unsigned long capabilities) | |
1da177e4 | 940 | { |
dbc8358c | 941 | unsigned long total, point; |
1da177e4 | 942 | void *base; |
8feae131 | 943 | int ret, order; |
1da177e4 LT |
944 | |
945 | /* invoke the file's mapping function so that it can keep track of | |
946 | * shared mappings on devices or memory | |
947 | * - VM_MAYSHARE will be set if it may attempt to share | |
948 | */ | |
b4caecd4 | 949 | if (capabilities & NOMMU_MAP_DIRECT) { |
f74ac015 | 950 | ret = call_mmap(vma->vm_file, vma); |
dd8632a1 | 951 | if (ret == 0) { |
1da177e4 | 952 | /* shouldn't return success if we're not sharing */ |
dd8632a1 PM |
953 | BUG_ON(!(vma->vm_flags & VM_MAYSHARE)); |
954 | vma->vm_region->vm_top = vma->vm_region->vm_end; | |
645d83c5 | 955 | return 0; |
1da177e4 | 956 | } |
dd8632a1 PM |
957 | if (ret != -ENOSYS) |
958 | return ret; | |
1da177e4 LT |
959 | |
960 | /* getting an ENOSYS error indicates that direct mmap isn't | |
961 | * possible (as opposed to tried but failed) so we'll try to | |
962 | * make a private copy of the data and map that instead */ | |
963 | } | |
964 | ||
8feae131 | 965 | |
1da177e4 LT |
966 | /* allocate some memory to hold the mapping |
967 | * - note that this may not return a page-aligned address if the object | |
968 | * we're allocating is smaller than a page | |
969 | */ | |
f67d9b15 | 970 | order = get_order(len); |
8feae131 | 971 | total = 1 << order; |
f67d9b15 | 972 | point = len >> PAGE_SHIFT; |
dd8632a1 | 973 | |
dbc8358c | 974 | /* we don't want to allocate a power-of-2 sized page set */ |
22cc877b | 975 | if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages) |
dbc8358c | 976 | total = point; |
8feae131 | 977 | |
da616534 | 978 | base = alloc_pages_exact(total << PAGE_SHIFT, GFP_KERNEL); |
dbc8358c JK |
979 | if (!base) |
980 | goto enomem; | |
981 | ||
982 | atomic_long_add(total, &mmap_pages_allocated); | |
1da177e4 | 983 | |
8feae131 DH |
984 | region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY; |
985 | region->vm_start = (unsigned long) base; | |
f67d9b15 | 986 | region->vm_end = region->vm_start + len; |
dd8632a1 | 987 | region->vm_top = region->vm_start + (total << PAGE_SHIFT); |
8feae131 DH |
988 | |
989 | vma->vm_start = region->vm_start; | |
990 | vma->vm_end = region->vm_start + len; | |
1da177e4 LT |
991 | |
992 | if (vma->vm_file) { | |
993 | /* read the contents of a file into the copy */ | |
1da177e4 LT |
994 | loff_t fpos; |
995 | ||
996 | fpos = vma->vm_pgoff; | |
997 | fpos <<= PAGE_SHIFT; | |
998 | ||
b4bf802a | 999 | ret = kernel_read(vma->vm_file, base, len, &fpos); |
1da177e4 LT |
1000 | if (ret < 0) |
1001 | goto error_free; | |
1002 | ||
1003 | /* clear the last little bit */ | |
f67d9b15 BL |
1004 | if (ret < len) |
1005 | memset(base + ret, 0, len - ret); | |
1da177e4 | 1006 | |
bfd40eaf KS |
1007 | } else { |
1008 | vma_set_anonymous(vma); | |
1da177e4 LT |
1009 | } |
1010 | ||
1011 | return 0; | |
1012 | ||
1013 | error_free: | |
7223bb4a | 1014 | free_page_series(region->vm_start, region->vm_top); |
8feae131 DH |
1015 | region->vm_start = vma->vm_start = 0; |
1016 | region->vm_end = vma->vm_end = 0; | |
dd8632a1 | 1017 | region->vm_top = 0; |
1da177e4 LT |
1018 | return ret; |
1019 | ||
1020 | enomem: | |
b1de0d13 | 1021 | pr_err("Allocation of length %lu from process %d (%s) failed\n", |
05ae6fa3 | 1022 | len, current->pid, current->comm); |
9af744d7 | 1023 | show_free_areas(0, NULL); |
1da177e4 LT |
1024 | return -ENOMEM; |
1025 | } | |
1026 | ||
1027 | /* | |
1028 | * handle mapping creation for uClinux | |
1029 | */ | |
1fcfd8db ON |
1030 | unsigned long do_mmap(struct file *file, |
1031 | unsigned long addr, | |
1032 | unsigned long len, | |
1033 | unsigned long prot, | |
1034 | unsigned long flags, | |
1fcfd8db | 1035 | unsigned long pgoff, |
897ab3e0 MR |
1036 | unsigned long *populate, |
1037 | struct list_head *uf) | |
1da177e4 | 1038 | { |
8feae131 DH |
1039 | struct vm_area_struct *vma; |
1040 | struct vm_region *region; | |
1da177e4 | 1041 | struct rb_node *rb; |
45e55300 | 1042 | vm_flags_t vm_flags; |
1fcfd8db | 1043 | unsigned long capabilities, result; |
1da177e4 LT |
1044 | int ret; |
1045 | ||
41badc15 | 1046 | *populate = 0; |
bebeb3d6 | 1047 | |
1da177e4 LT |
1048 | /* decide whether we should attempt the mapping, and if so what sort of |
1049 | * mapping */ | |
1050 | ret = validate_mmap_request(file, addr, len, prot, flags, pgoff, | |
1051 | &capabilities); | |
22cc877b | 1052 | if (ret < 0) |
1da177e4 LT |
1053 | return ret; |
1054 | ||
06aab5a3 DH |
1055 | /* we ignore the address hint */ |
1056 | addr = 0; | |
f67d9b15 | 1057 | len = PAGE_ALIGN(len); |
06aab5a3 | 1058 | |
1da177e4 LT |
1059 | /* we've determined that we can make the mapping, now translate what we |
1060 | * now know into VMA flags */ | |
45e55300 | 1061 | vm_flags = determine_vm_flags(file, prot, flags, capabilities); |
1da177e4 | 1062 | |
8feae131 DH |
1063 | /* we're going to need to record the mapping */ |
1064 | region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL); | |
1065 | if (!region) | |
1066 | goto error_getting_region; | |
1067 | ||
490fc053 | 1068 | vma = vm_area_alloc(current->mm); |
8feae131 DH |
1069 | if (!vma) |
1070 | goto error_getting_vma; | |
1da177e4 | 1071 | |
1e2ae599 | 1072 | region->vm_usage = 1; |
8feae131 DH |
1073 | region->vm_flags = vm_flags; |
1074 | region->vm_pgoff = pgoff; | |
1075 | ||
8feae131 DH |
1076 | vma->vm_flags = vm_flags; |
1077 | vma->vm_pgoff = pgoff; | |
1da177e4 | 1078 | |
8feae131 | 1079 | if (file) { |
cb0942b8 AV |
1080 | region->vm_file = get_file(file); |
1081 | vma->vm_file = get_file(file); | |
8feae131 DH |
1082 | } |
1083 | ||
1084 | down_write(&nommu_region_sem); | |
1085 | ||
1086 | /* if we want to share, we need to check for regions created by other | |
1da177e4 | 1087 | * mmap() calls that overlap with our proposed mapping |
8feae131 | 1088 | * - we can only share with a superset match on most regular files |
1da177e4 LT |
1089 | * - shared mappings on character devices and memory backed files are |
1090 | * permitted to overlap inexactly as far as we are concerned for in | |
1091 | * these cases, sharing is handled in the driver or filesystem rather | |
1092 | * than here | |
1093 | */ | |
1094 | if (vm_flags & VM_MAYSHARE) { | |
8feae131 DH |
1095 | struct vm_region *pregion; |
1096 | unsigned long pglen, rpglen, pgend, rpgend, start; | |
1da177e4 | 1097 | |
8feae131 DH |
1098 | pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; |
1099 | pgend = pgoff + pglen; | |
165b2392 | 1100 | |
8feae131 DH |
1101 | for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) { |
1102 | pregion = rb_entry(rb, struct vm_region, vm_rb); | |
1da177e4 | 1103 | |
8feae131 | 1104 | if (!(pregion->vm_flags & VM_MAYSHARE)) |
1da177e4 LT |
1105 | continue; |
1106 | ||
1107 | /* search for overlapping mappings on the same file */ | |
496ad9aa AV |
1108 | if (file_inode(pregion->vm_file) != |
1109 | file_inode(file)) | |
1da177e4 LT |
1110 | continue; |
1111 | ||
8feae131 | 1112 | if (pregion->vm_pgoff >= pgend) |
1da177e4 LT |
1113 | continue; |
1114 | ||
8feae131 DH |
1115 | rpglen = pregion->vm_end - pregion->vm_start; |
1116 | rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
1117 | rpgend = pregion->vm_pgoff + rpglen; | |
1118 | if (pgoff >= rpgend) | |
1da177e4 LT |
1119 | continue; |
1120 | ||
8feae131 DH |
1121 | /* handle inexactly overlapping matches between |
1122 | * mappings */ | |
1123 | if ((pregion->vm_pgoff != pgoff || rpglen != pglen) && | |
1124 | !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) { | |
1125 | /* new mapping is not a subset of the region */ | |
b4caecd4 | 1126 | if (!(capabilities & NOMMU_MAP_DIRECT)) |
1da177e4 LT |
1127 | goto sharing_violation; |
1128 | continue; | |
1129 | } | |
1130 | ||
8feae131 | 1131 | /* we've found a region we can share */ |
1e2ae599 | 1132 | pregion->vm_usage++; |
8feae131 DH |
1133 | vma->vm_region = pregion; |
1134 | start = pregion->vm_start; | |
1135 | start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT; | |
1136 | vma->vm_start = start; | |
1137 | vma->vm_end = start + len; | |
1138 | ||
22cc877b | 1139 | if (pregion->vm_flags & VM_MAPPED_COPY) |
8feae131 | 1140 | vma->vm_flags |= VM_MAPPED_COPY; |
22cc877b | 1141 | else { |
8feae131 DH |
1142 | ret = do_mmap_shared_file(vma); |
1143 | if (ret < 0) { | |
1144 | vma->vm_region = NULL; | |
1145 | vma->vm_start = 0; | |
1146 | vma->vm_end = 0; | |
1e2ae599 | 1147 | pregion->vm_usage--; |
8feae131 DH |
1148 | pregion = NULL; |
1149 | goto error_just_free; | |
1150 | } | |
1151 | } | |
1152 | fput(region->vm_file); | |
1153 | kmem_cache_free(vm_region_jar, region); | |
1154 | region = pregion; | |
1155 | result = start; | |
1156 | goto share; | |
1da177e4 LT |
1157 | } |
1158 | ||
1da177e4 LT |
1159 | /* obtain the address at which to make a shared mapping |
1160 | * - this is the hook for quasi-memory character devices to | |
1161 | * tell us the location of a shared mapping | |
1162 | */ | |
b4caecd4 | 1163 | if (capabilities & NOMMU_MAP_DIRECT) { |
1da177e4 LT |
1164 | addr = file->f_op->get_unmapped_area(file, addr, len, |
1165 | pgoff, flags); | |
bb005a59 | 1166 | if (IS_ERR_VALUE(addr)) { |
1da177e4 | 1167 | ret = addr; |
bb005a59 | 1168 | if (ret != -ENOSYS) |
8feae131 | 1169 | goto error_just_free; |
1da177e4 LT |
1170 | |
1171 | /* the driver refused to tell us where to site | |
1172 | * the mapping so we'll have to attempt to copy | |
1173 | * it */ | |
bb005a59 | 1174 | ret = -ENODEV; |
b4caecd4 | 1175 | if (!(capabilities & NOMMU_MAP_COPY)) |
8feae131 | 1176 | goto error_just_free; |
1da177e4 | 1177 | |
b4caecd4 | 1178 | capabilities &= ~NOMMU_MAP_DIRECT; |
8feae131 DH |
1179 | } else { |
1180 | vma->vm_start = region->vm_start = addr; | |
1181 | vma->vm_end = region->vm_end = addr + len; | |
1da177e4 LT |
1182 | } |
1183 | } | |
1184 | } | |
1185 | ||
8feae131 | 1186 | vma->vm_region = region; |
1da177e4 | 1187 | |
645d83c5 | 1188 | /* set up the mapping |
b4caecd4 | 1189 | * - the region is filled in if NOMMU_MAP_DIRECT is still set |
645d83c5 | 1190 | */ |
1da177e4 | 1191 | if (file && vma->vm_flags & VM_SHARED) |
8feae131 | 1192 | ret = do_mmap_shared_file(vma); |
1da177e4 | 1193 | else |
645d83c5 | 1194 | ret = do_mmap_private(vma, region, len, capabilities); |
1da177e4 | 1195 | if (ret < 0) |
645d83c5 DH |
1196 | goto error_just_free; |
1197 | add_nommu_region(region); | |
8feae131 | 1198 | |
ea637639 | 1199 | /* clear anonymous mappings that don't ask for uninitialized data */ |
0bf5f949 CH |
1200 | if (!vma->vm_file && |
1201 | (!IS_ENABLED(CONFIG_MMAP_ALLOW_UNINITIALIZED) || | |
1202 | !(flags & MAP_UNINITIALIZED))) | |
ea637639 JZ |
1203 | memset((void *)region->vm_start, 0, |
1204 | region->vm_end - region->vm_start); | |
1205 | ||
1da177e4 | 1206 | /* okay... we have a mapping; now we have to register it */ |
8feae131 | 1207 | result = vma->vm_start; |
1da177e4 | 1208 | |
1da177e4 LT |
1209 | current->mm->total_vm += len >> PAGE_SHIFT; |
1210 | ||
8feae131 DH |
1211 | share: |
1212 | add_vma_to_mm(current->mm, vma); | |
1da177e4 | 1213 | |
cfe79c00 MF |
1214 | /* we flush the region from the icache only when the first executable |
1215 | * mapping of it is made */ | |
1216 | if (vma->vm_flags & VM_EXEC && !region->vm_icache_flushed) { | |
a75a2df6 | 1217 | flush_icache_user_range(region->vm_start, region->vm_end); |
cfe79c00 MF |
1218 | region->vm_icache_flushed = true; |
1219 | } | |
1da177e4 | 1220 | |
cfe79c00 | 1221 | up_write(&nommu_region_sem); |
1da177e4 | 1222 | |
8feae131 | 1223 | return result; |
1da177e4 | 1224 | |
8feae131 DH |
1225 | error_just_free: |
1226 | up_write(&nommu_region_sem); | |
1227 | error: | |
89a86402 DH |
1228 | if (region->vm_file) |
1229 | fput(region->vm_file); | |
8feae131 | 1230 | kmem_cache_free(vm_region_jar, region); |
89a86402 DH |
1231 | if (vma->vm_file) |
1232 | fput(vma->vm_file); | |
3928d4f5 | 1233 | vm_area_free(vma); |
8feae131 DH |
1234 | return ret; |
1235 | ||
1236 | sharing_violation: | |
1237 | up_write(&nommu_region_sem); | |
22cc877b | 1238 | pr_warn("Attempt to share mismatched mappings\n"); |
8feae131 DH |
1239 | ret = -EINVAL; |
1240 | goto error; | |
1da177e4 | 1241 | |
8feae131 DH |
1242 | error_getting_vma: |
1243 | kmem_cache_free(vm_region_jar, region); | |
22cc877b LR |
1244 | pr_warn("Allocation of vma for %lu byte allocation from process %d failed\n", |
1245 | len, current->pid); | |
9af744d7 | 1246 | show_free_areas(0, NULL); |
1da177e4 LT |
1247 | return -ENOMEM; |
1248 | ||
8feae131 | 1249 | error_getting_region: |
22cc877b LR |
1250 | pr_warn("Allocation of vm region for %lu byte allocation from process %d failed\n", |
1251 | len, current->pid); | |
9af744d7 | 1252 | show_free_areas(0, NULL); |
1da177e4 LT |
1253 | return -ENOMEM; |
1254 | } | |
6be5ceb0 | 1255 | |
a90f590a DB |
1256 | unsigned long ksys_mmap_pgoff(unsigned long addr, unsigned long len, |
1257 | unsigned long prot, unsigned long flags, | |
1258 | unsigned long fd, unsigned long pgoff) | |
66f0dc48 HD |
1259 | { |
1260 | struct file *file = NULL; | |
1261 | unsigned long retval = -EBADF; | |
1262 | ||
120a795d | 1263 | audit_mmap_fd(fd, flags); |
66f0dc48 HD |
1264 | if (!(flags & MAP_ANONYMOUS)) { |
1265 | file = fget(fd); | |
1266 | if (!file) | |
1267 | goto out; | |
1268 | } | |
1269 | ||
ad1ed293 | 1270 | retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff); |
66f0dc48 HD |
1271 | |
1272 | if (file) | |
1273 | fput(file); | |
1274 | out: | |
1275 | return retval; | |
1276 | } | |
1277 | ||
a90f590a DB |
1278 | SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len, |
1279 | unsigned long, prot, unsigned long, flags, | |
1280 | unsigned long, fd, unsigned long, pgoff) | |
1281 | { | |
1282 | return ksys_mmap_pgoff(addr, len, prot, flags, fd, pgoff); | |
1283 | } | |
1284 | ||
a4679373 CH |
1285 | #ifdef __ARCH_WANT_SYS_OLD_MMAP |
1286 | struct mmap_arg_struct { | |
1287 | unsigned long addr; | |
1288 | unsigned long len; | |
1289 | unsigned long prot; | |
1290 | unsigned long flags; | |
1291 | unsigned long fd; | |
1292 | unsigned long offset; | |
1293 | }; | |
1294 | ||
1295 | SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg) | |
1296 | { | |
1297 | struct mmap_arg_struct a; | |
1298 | ||
1299 | if (copy_from_user(&a, arg, sizeof(a))) | |
1300 | return -EFAULT; | |
1824cb75 | 1301 | if (offset_in_page(a.offset)) |
a4679373 CH |
1302 | return -EINVAL; |
1303 | ||
a90f590a DB |
1304 | return ksys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd, |
1305 | a.offset >> PAGE_SHIFT); | |
a4679373 CH |
1306 | } |
1307 | #endif /* __ARCH_WANT_SYS_OLD_MMAP */ | |
1308 | ||
1da177e4 | 1309 | /* |
8feae131 DH |
1310 | * split a vma into two pieces at address 'addr', a new vma is allocated either |
1311 | * for the first part or the tail. | |
1da177e4 | 1312 | */ |
8feae131 DH |
1313 | int split_vma(struct mm_struct *mm, struct vm_area_struct *vma, |
1314 | unsigned long addr, int new_below) | |
1da177e4 | 1315 | { |
8feae131 DH |
1316 | struct vm_area_struct *new; |
1317 | struct vm_region *region; | |
1318 | unsigned long npages; | |
1da177e4 | 1319 | |
779c1023 DH |
1320 | /* we're only permitted to split anonymous regions (these should have |
1321 | * only a single usage on the region) */ | |
1322 | if (vma->vm_file) | |
8feae131 | 1323 | return -ENOMEM; |
1da177e4 | 1324 | |
8feae131 DH |
1325 | if (mm->map_count >= sysctl_max_map_count) |
1326 | return -ENOMEM; | |
1da177e4 | 1327 | |
8feae131 DH |
1328 | region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL); |
1329 | if (!region) | |
1330 | return -ENOMEM; | |
1da177e4 | 1331 | |
3928d4f5 | 1332 | new = vm_area_dup(vma); |
8feae131 DH |
1333 | if (!new) { |
1334 | kmem_cache_free(vm_region_jar, region); | |
1335 | return -ENOMEM; | |
1336 | } | |
1337 | ||
1338 | /* most fields are the same, copy all, and then fixup */ | |
8feae131 DH |
1339 | *region = *vma->vm_region; |
1340 | new->vm_region = region; | |
1341 | ||
1342 | npages = (addr - vma->vm_start) >> PAGE_SHIFT; | |
1343 | ||
1344 | if (new_below) { | |
dd8632a1 | 1345 | region->vm_top = region->vm_end = new->vm_end = addr; |
8feae131 DH |
1346 | } else { |
1347 | region->vm_start = new->vm_start = addr; | |
1348 | region->vm_pgoff = new->vm_pgoff += npages; | |
1da177e4 | 1349 | } |
8feae131 DH |
1350 | |
1351 | if (new->vm_ops && new->vm_ops->open) | |
1352 | new->vm_ops->open(new); | |
1353 | ||
1354 | delete_vma_from_mm(vma); | |
1355 | down_write(&nommu_region_sem); | |
1356 | delete_nommu_region(vma->vm_region); | |
1357 | if (new_below) { | |
1358 | vma->vm_region->vm_start = vma->vm_start = addr; | |
1359 | vma->vm_region->vm_pgoff = vma->vm_pgoff += npages; | |
1360 | } else { | |
1361 | vma->vm_region->vm_end = vma->vm_end = addr; | |
dd8632a1 | 1362 | vma->vm_region->vm_top = addr; |
8feae131 DH |
1363 | } |
1364 | add_nommu_region(vma->vm_region); | |
1365 | add_nommu_region(new->vm_region); | |
1366 | up_write(&nommu_region_sem); | |
1367 | add_vma_to_mm(mm, vma); | |
1368 | add_vma_to_mm(mm, new); | |
1369 | return 0; | |
1da177e4 LT |
1370 | } |
1371 | ||
3034097a | 1372 | /* |
8feae131 DH |
1373 | * shrink a VMA by removing the specified chunk from either the beginning or |
1374 | * the end | |
3034097a | 1375 | */ |
8feae131 DH |
1376 | static int shrink_vma(struct mm_struct *mm, |
1377 | struct vm_area_struct *vma, | |
1378 | unsigned long from, unsigned long to) | |
1da177e4 | 1379 | { |
8feae131 | 1380 | struct vm_region *region; |
1da177e4 | 1381 | |
8feae131 DH |
1382 | /* adjust the VMA's pointers, which may reposition it in the MM's tree |
1383 | * and list */ | |
1384 | delete_vma_from_mm(vma); | |
1385 | if (from > vma->vm_start) | |
1386 | vma->vm_end = from; | |
1387 | else | |
1388 | vma->vm_start = to; | |
1389 | add_vma_to_mm(mm, vma); | |
1da177e4 | 1390 | |
8feae131 DH |
1391 | /* cut the backing region down to size */ |
1392 | region = vma->vm_region; | |
1e2ae599 | 1393 | BUG_ON(region->vm_usage != 1); |
8feae131 DH |
1394 | |
1395 | down_write(&nommu_region_sem); | |
1396 | delete_nommu_region(region); | |
dd8632a1 PM |
1397 | if (from > region->vm_start) { |
1398 | to = region->vm_top; | |
1399 | region->vm_top = region->vm_end = from; | |
1400 | } else { | |
8feae131 | 1401 | region->vm_start = to; |
dd8632a1 | 1402 | } |
8feae131 DH |
1403 | add_nommu_region(region); |
1404 | up_write(&nommu_region_sem); | |
1405 | ||
1406 | free_page_series(from, to); | |
1407 | return 0; | |
1408 | } | |
1da177e4 | 1409 | |
8feae131 DH |
1410 | /* |
1411 | * release a mapping | |
1412 | * - under NOMMU conditions the chunk to be unmapped must be backed by a single | |
1413 | * VMA, though it need not cover the whole VMA | |
1414 | */ | |
897ab3e0 | 1415 | int do_munmap(struct mm_struct *mm, unsigned long start, size_t len, struct list_head *uf) |
8feae131 DH |
1416 | { |
1417 | struct vm_area_struct *vma; | |
f67d9b15 | 1418 | unsigned long end; |
8feae131 | 1419 | int ret; |
1da177e4 | 1420 | |
f67d9b15 | 1421 | len = PAGE_ALIGN(len); |
8feae131 DH |
1422 | if (len == 0) |
1423 | return -EINVAL; | |
365e9c87 | 1424 | |
f67d9b15 BL |
1425 | end = start + len; |
1426 | ||
8feae131 DH |
1427 | /* find the first potentially overlapping VMA */ |
1428 | vma = find_vma(mm, start); | |
1429 | if (!vma) { | |
ac714904 | 1430 | static int limit; |
33e5d769 | 1431 | if (limit < 5) { |
22cc877b LR |
1432 | pr_warn("munmap of memory not mmapped by process %d (%s): 0x%lx-0x%lx\n", |
1433 | current->pid, current->comm, | |
1434 | start, start + len - 1); | |
33e5d769 DH |
1435 | limit++; |
1436 | } | |
8feae131 DH |
1437 | return -EINVAL; |
1438 | } | |
1da177e4 | 1439 | |
8feae131 DH |
1440 | /* we're allowed to split an anonymous VMA but not a file-backed one */ |
1441 | if (vma->vm_file) { | |
1442 | do { | |
22cc877b | 1443 | if (start > vma->vm_start) |
8feae131 | 1444 | return -EINVAL; |
8feae131 DH |
1445 | if (end == vma->vm_end) |
1446 | goto erase_whole_vma; | |
d75a310c NK |
1447 | vma = vma->vm_next; |
1448 | } while (vma); | |
8feae131 DH |
1449 | return -EINVAL; |
1450 | } else { | |
1451 | /* the chunk must be a subset of the VMA found */ | |
1452 | if (start == vma->vm_start && end == vma->vm_end) | |
1453 | goto erase_whole_vma; | |
22cc877b | 1454 | if (start < vma->vm_start || end > vma->vm_end) |
8feae131 | 1455 | return -EINVAL; |
1824cb75 | 1456 | if (offset_in_page(start)) |
8feae131 | 1457 | return -EINVAL; |
1824cb75 | 1458 | if (end != vma->vm_end && offset_in_page(end)) |
8feae131 | 1459 | return -EINVAL; |
8feae131 DH |
1460 | if (start != vma->vm_start && end != vma->vm_end) { |
1461 | ret = split_vma(mm, vma, start, 1); | |
22cc877b | 1462 | if (ret < 0) |
8feae131 | 1463 | return ret; |
8feae131 DH |
1464 | } |
1465 | return shrink_vma(mm, vma, start, end); | |
1466 | } | |
1da177e4 | 1467 | |
8feae131 DH |
1468 | erase_whole_vma: |
1469 | delete_vma_from_mm(vma); | |
1470 | delete_vma(mm, vma); | |
1da177e4 LT |
1471 | return 0; |
1472 | } | |
1473 | ||
bfce281c | 1474 | int vm_munmap(unsigned long addr, size_t len) |
3034097a | 1475 | { |
bfce281c | 1476 | struct mm_struct *mm = current->mm; |
3034097a | 1477 | int ret; |
3034097a | 1478 | |
d8ed45c5 | 1479 | mmap_write_lock(mm); |
897ab3e0 | 1480 | ret = do_munmap(mm, addr, len, NULL); |
d8ed45c5 | 1481 | mmap_write_unlock(mm); |
3034097a DH |
1482 | return ret; |
1483 | } | |
a46ef99d LT |
1484 | EXPORT_SYMBOL(vm_munmap); |
1485 | ||
1486 | SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len) | |
1487 | { | |
bfce281c | 1488 | return vm_munmap(addr, len); |
a46ef99d | 1489 | } |
3034097a DH |
1490 | |
1491 | /* | |
8feae131 | 1492 | * release all the mappings made in a process's VM space |
3034097a | 1493 | */ |
8feae131 | 1494 | void exit_mmap(struct mm_struct *mm) |
1da177e4 | 1495 | { |
8feae131 | 1496 | struct vm_area_struct *vma; |
1da177e4 | 1497 | |
8feae131 DH |
1498 | if (!mm) |
1499 | return; | |
1da177e4 | 1500 | |
8feae131 | 1501 | mm->total_vm = 0; |
1da177e4 | 1502 | |
8feae131 DH |
1503 | while ((vma = mm->mmap)) { |
1504 | mm->mmap = vma->vm_next; | |
1505 | delete_vma_from_mm(vma); | |
1506 | delete_vma(mm, vma); | |
04c34961 | 1507 | cond_resched(); |
1da177e4 | 1508 | } |
524e00b3 | 1509 | __mt_destroy(&mm->mm_mt); |
1da177e4 LT |
1510 | } |
1511 | ||
5d22fc25 | 1512 | int vm_brk(unsigned long addr, unsigned long len) |
1da177e4 LT |
1513 | { |
1514 | return -ENOMEM; | |
1515 | } | |
1516 | ||
1517 | /* | |
6fa5f80b DH |
1518 | * expand (or shrink) an existing mapping, potentially moving it at the same |
1519 | * time (controlled by the MREMAP_MAYMOVE flag and available VM space) | |
1da177e4 | 1520 | * |
6fa5f80b | 1521 | * under NOMMU conditions, we only permit changing a mapping's size, and only |
8feae131 DH |
1522 | * as long as it stays within the region allocated by do_mmap_private() and the |
1523 | * block is not shareable | |
1da177e4 | 1524 | * |
6fa5f80b | 1525 | * MREMAP_FIXED is not supported under NOMMU conditions |
1da177e4 | 1526 | */ |
4b377bab | 1527 | static unsigned long do_mremap(unsigned long addr, |
1da177e4 LT |
1528 | unsigned long old_len, unsigned long new_len, |
1529 | unsigned long flags, unsigned long new_addr) | |
1530 | { | |
6fa5f80b | 1531 | struct vm_area_struct *vma; |
1da177e4 LT |
1532 | |
1533 | /* insanity checks first */ | |
f67d9b15 BL |
1534 | old_len = PAGE_ALIGN(old_len); |
1535 | new_len = PAGE_ALIGN(new_len); | |
8feae131 | 1536 | if (old_len == 0 || new_len == 0) |
1da177e4 LT |
1537 | return (unsigned long) -EINVAL; |
1538 | ||
1824cb75 | 1539 | if (offset_in_page(addr)) |
8feae131 DH |
1540 | return -EINVAL; |
1541 | ||
1da177e4 LT |
1542 | if (flags & MREMAP_FIXED && new_addr != addr) |
1543 | return (unsigned long) -EINVAL; | |
1544 | ||
8feae131 | 1545 | vma = find_vma_exact(current->mm, addr, old_len); |
6fa5f80b DH |
1546 | if (!vma) |
1547 | return (unsigned long) -EINVAL; | |
1da177e4 | 1548 | |
6fa5f80b | 1549 | if (vma->vm_end != vma->vm_start + old_len) |
1da177e4 LT |
1550 | return (unsigned long) -EFAULT; |
1551 | ||
6fa5f80b | 1552 | if (vma->vm_flags & VM_MAYSHARE) |
1da177e4 LT |
1553 | return (unsigned long) -EPERM; |
1554 | ||
8feae131 | 1555 | if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start) |
1da177e4 LT |
1556 | return (unsigned long) -ENOMEM; |
1557 | ||
1558 | /* all checks complete - do it */ | |
6fa5f80b | 1559 | vma->vm_end = vma->vm_start + new_len; |
6fa5f80b DH |
1560 | return vma->vm_start; |
1561 | } | |
1562 | ||
6a6160a7 HC |
1563 | SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len, |
1564 | unsigned long, new_len, unsigned long, flags, | |
1565 | unsigned long, new_addr) | |
6fa5f80b DH |
1566 | { |
1567 | unsigned long ret; | |
1568 | ||
d8ed45c5 | 1569 | mmap_write_lock(current->mm); |
6fa5f80b | 1570 | ret = do_mremap(addr, old_len, new_len, flags, new_addr); |
d8ed45c5 | 1571 | mmap_write_unlock(current->mm); |
6fa5f80b | 1572 | return ret; |
1da177e4 LT |
1573 | } |
1574 | ||
df06b37f KB |
1575 | struct page *follow_page(struct vm_area_struct *vma, unsigned long address, |
1576 | unsigned int foll_flags) | |
1da177e4 LT |
1577 | { |
1578 | return NULL; | |
1579 | } | |
1580 | ||
8f3b1327 BL |
1581 | int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr, |
1582 | unsigned long pfn, unsigned long size, pgprot_t prot) | |
1da177e4 | 1583 | { |
8f3b1327 BL |
1584 | if (addr != (pfn << PAGE_SHIFT)) |
1585 | return -EINVAL; | |
1586 | ||
314e51b9 | 1587 | vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP; |
66aa2b4b | 1588 | return 0; |
1da177e4 | 1589 | } |
22c4af40 | 1590 | EXPORT_SYMBOL(remap_pfn_range); |
1da177e4 | 1591 | |
3c0b9de6 LT |
1592 | int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len) |
1593 | { | |
1594 | unsigned long pfn = start >> PAGE_SHIFT; | |
1595 | unsigned long vm_len = vma->vm_end - vma->vm_start; | |
1596 | ||
1597 | pfn += vma->vm_pgoff; | |
1598 | return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot); | |
1599 | } | |
1600 | EXPORT_SYMBOL(vm_iomap_memory); | |
1601 | ||
f905bc44 PM |
1602 | int remap_vmalloc_range(struct vm_area_struct *vma, void *addr, |
1603 | unsigned long pgoff) | |
1604 | { | |
1605 | unsigned int size = vma->vm_end - vma->vm_start; | |
1606 | ||
1607 | if (!(vma->vm_flags & VM_USERMAP)) | |
1608 | return -EINVAL; | |
1609 | ||
1610 | vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT)); | |
1611 | vma->vm_end = vma->vm_start + size; | |
1612 | ||
1613 | return 0; | |
1614 | } | |
1615 | EXPORT_SYMBOL(remap_vmalloc_range); | |
1616 | ||
2bcd6454 | 1617 | vm_fault_t filemap_fault(struct vm_fault *vmf) |
b0e15190 DH |
1618 | { |
1619 | BUG(); | |
d0217ac0 | 1620 | return 0; |
b0e15190 | 1621 | } |
b5073173 | 1622 | EXPORT_SYMBOL(filemap_fault); |
0ec76a11 | 1623 | |
3f98a28c | 1624 | vm_fault_t filemap_map_pages(struct vm_fault *vmf, |
bae473a4 | 1625 | pgoff_t start_pgoff, pgoff_t end_pgoff) |
f1820361 KS |
1626 | { |
1627 | BUG(); | |
3f98a28c | 1628 | return 0; |
f1820361 KS |
1629 | } |
1630 | EXPORT_SYMBOL(filemap_map_pages); | |
1631 | ||
d3f5ffca JH |
1632 | int __access_remote_vm(struct mm_struct *mm, unsigned long addr, void *buf, |
1633 | int len, unsigned int gup_flags) | |
0ec76a11 | 1634 | { |
0ec76a11 | 1635 | struct vm_area_struct *vma; |
442486ec | 1636 | int write = gup_flags & FOLL_WRITE; |
0ec76a11 | 1637 | |
d8ed45c5 | 1638 | if (mmap_read_lock_killable(mm)) |
1e426fe2 | 1639 | return 0; |
0ec76a11 DH |
1640 | |
1641 | /* the access must start within one of the target process's mappings */ | |
0159b141 DH |
1642 | vma = find_vma(mm, addr); |
1643 | if (vma) { | |
0ec76a11 DH |
1644 | /* don't overrun this mapping */ |
1645 | if (addr + len >= vma->vm_end) | |
1646 | len = vma->vm_end - addr; | |
1647 | ||
1648 | /* only read or write mappings where it is permitted */ | |
d00c7b99 | 1649 | if (write && vma->vm_flags & VM_MAYWRITE) |
7959722b JZ |
1650 | copy_to_user_page(vma, NULL, addr, |
1651 | (void *) addr, buf, len); | |
d00c7b99 | 1652 | else if (!write && vma->vm_flags & VM_MAYREAD) |
7959722b JZ |
1653 | copy_from_user_page(vma, NULL, addr, |
1654 | buf, (void *) addr, len); | |
0ec76a11 DH |
1655 | else |
1656 | len = 0; | |
1657 | } else { | |
1658 | len = 0; | |
1659 | } | |
1660 | ||
d8ed45c5 | 1661 | mmap_read_unlock(mm); |
f55f199b MF |
1662 | |
1663 | return len; | |
1664 | } | |
1665 | ||
1666 | /** | |
b7701a5f | 1667 | * access_remote_vm - access another process' address space |
f55f199b MF |
1668 | * @mm: the mm_struct of the target address space |
1669 | * @addr: start address to access | |
1670 | * @buf: source or destination buffer | |
1671 | * @len: number of bytes to transfer | |
6347e8d5 | 1672 | * @gup_flags: flags modifying lookup behaviour |
f55f199b MF |
1673 | * |
1674 | * The caller must hold a reference on @mm. | |
1675 | */ | |
1676 | int access_remote_vm(struct mm_struct *mm, unsigned long addr, | |
6347e8d5 | 1677 | void *buf, int len, unsigned int gup_flags) |
f55f199b | 1678 | { |
d3f5ffca | 1679 | return __access_remote_vm(mm, addr, buf, len, gup_flags); |
f55f199b MF |
1680 | } |
1681 | ||
1682 | /* | |
1683 | * Access another process' address space. | |
1684 | * - source/target buffer must be kernel space | |
1685 | */ | |
f307ab6d LS |
1686 | int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, |
1687 | unsigned int gup_flags) | |
f55f199b MF |
1688 | { |
1689 | struct mm_struct *mm; | |
1690 | ||
1691 | if (addr + len < addr) | |
1692 | return 0; | |
1693 | ||
1694 | mm = get_task_mm(tsk); | |
1695 | if (!mm) | |
1696 | return 0; | |
1697 | ||
d3f5ffca | 1698 | len = __access_remote_vm(mm, addr, buf, len, gup_flags); |
f55f199b | 1699 | |
0ec76a11 DH |
1700 | mmput(mm); |
1701 | return len; | |
1702 | } | |
fcd35857 | 1703 | EXPORT_SYMBOL_GPL(access_process_vm); |
7e660872 DH |
1704 | |
1705 | /** | |
1706 | * nommu_shrink_inode_mappings - Shrink the shared mappings on an inode | |
1707 | * @inode: The inode to check | |
1708 | * @size: The current filesize of the inode | |
1709 | * @newsize: The proposed filesize of the inode | |
1710 | * | |
1711 | * Check the shared mappings on an inode on behalf of a shrinking truncate to | |
c08b342c RD |
1712 | * make sure that any outstanding VMAs aren't broken and then shrink the |
1713 | * vm_regions that extend beyond so that do_mmap() doesn't | |
7e660872 DH |
1714 | * automatically grant mappings that are too large. |
1715 | */ | |
1716 | int nommu_shrink_inode_mappings(struct inode *inode, size_t size, | |
1717 | size_t newsize) | |
1718 | { | |
1719 | struct vm_area_struct *vma; | |
7e660872 DH |
1720 | struct vm_region *region; |
1721 | pgoff_t low, high; | |
1722 | size_t r_size, r_top; | |
1723 | ||
1724 | low = newsize >> PAGE_SHIFT; | |
1725 | high = (size + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
1726 | ||
1727 | down_write(&nommu_region_sem); | |
1acf2e04 | 1728 | i_mmap_lock_read(inode->i_mapping); |
7e660872 DH |
1729 | |
1730 | /* search for VMAs that fall within the dead zone */ | |
6b2dbba8 | 1731 | vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, low, high) { |
7e660872 DH |
1732 | /* found one - only interested if it's shared out of the page |
1733 | * cache */ | |
1734 | if (vma->vm_flags & VM_SHARED) { | |
1acf2e04 | 1735 | i_mmap_unlock_read(inode->i_mapping); |
7e660872 DH |
1736 | up_write(&nommu_region_sem); |
1737 | return -ETXTBSY; /* not quite true, but near enough */ | |
1738 | } | |
1739 | } | |
1740 | ||
1741 | /* reduce any regions that overlap the dead zone - if in existence, | |
1742 | * these will be pointed to by VMAs that don't overlap the dead zone | |
1743 | * | |
1744 | * we don't check for any regions that start beyond the EOF as there | |
1745 | * shouldn't be any | |
1746 | */ | |
1acf2e04 | 1747 | vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, 0, ULONG_MAX) { |
7e660872 DH |
1748 | if (!(vma->vm_flags & VM_SHARED)) |
1749 | continue; | |
1750 | ||
1751 | region = vma->vm_region; | |
1752 | r_size = region->vm_top - region->vm_start; | |
1753 | r_top = (region->vm_pgoff << PAGE_SHIFT) + r_size; | |
1754 | ||
1755 | if (r_top > newsize) { | |
1756 | region->vm_top -= r_top - newsize; | |
1757 | if (region->vm_end > region->vm_top) | |
1758 | region->vm_end = region->vm_top; | |
1759 | } | |
1760 | } | |
1761 | ||
1acf2e04 | 1762 | i_mmap_unlock_read(inode->i_mapping); |
7e660872 DH |
1763 | up_write(&nommu_region_sem); |
1764 | return 0; | |
1765 | } | |
c9b1d098 AS |
1766 | |
1767 | /* | |
1768 | * Initialise sysctl_user_reserve_kbytes. | |
1769 | * | |
1770 | * This is intended to prevent a user from starting a single memory hogging | |
1771 | * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER | |
1772 | * mode. | |
1773 | * | |
1774 | * The default value is min(3% of free memory, 128MB) | |
1775 | * 128MB is enough to recover with sshd/login, bash, and top/kill. | |
1776 | */ | |
1777 | static int __meminit init_user_reserve(void) | |
1778 | { | |
1779 | unsigned long free_kbytes; | |
1780 | ||
c41f012a | 1781 | free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); |
c9b1d098 AS |
1782 | |
1783 | sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17); | |
1784 | return 0; | |
1785 | } | |
a4bc6fc7 | 1786 | subsys_initcall(init_user_reserve); |
4eeab4f5 AS |
1787 | |
1788 | /* | |
1789 | * Initialise sysctl_admin_reserve_kbytes. | |
1790 | * | |
1791 | * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin | |
1792 | * to log in and kill a memory hogging process. | |
1793 | * | |
1794 | * Systems with more than 256MB will reserve 8MB, enough to recover | |
1795 | * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will | |
1796 | * only reserve 3% of free pages by default. | |
1797 | */ | |
1798 | static int __meminit init_admin_reserve(void) | |
1799 | { | |
1800 | unsigned long free_kbytes; | |
1801 | ||
c41f012a | 1802 | free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); |
4eeab4f5 AS |
1803 | |
1804 | sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13); | |
1805 | return 0; | |
1806 | } | |
a4bc6fc7 | 1807 | subsys_initcall(init_admin_reserve); |