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