Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/mm/nommu.c | |
3 | * | |
4 | * Replacement code for mm functions to support CPU's that don't | |
5 | * have any form of memory management unit (thus no virtual memory). | |
6 | * | |
7 | * See Documentation/nommu-mmap.txt | |
8 | * | |
8feae131 | 9 | * Copyright (c) 2004-2008 David Howells <dhowells@redhat.com> |
1da177e4 LT |
10 | * Copyright (c) 2000-2003 David McCullough <davidm@snapgear.com> |
11 | * Copyright (c) 2000-2001 D Jeff Dionne <jeff@uClinux.org> | |
12 | * Copyright (c) 2002 Greg Ungerer <gerg@snapgear.com> | |
29c185e5 | 13 | * Copyright (c) 2007-2010 Paul Mundt <lethal@linux-sh.org> |
1da177e4 LT |
14 | */ |
15 | ||
b1de0d13 MH |
16 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
17 | ||
b95f1b31 | 18 | #include <linux/export.h> |
1da177e4 | 19 | #include <linux/mm.h> |
615d6e87 | 20 | #include <linux/vmacache.h> |
1da177e4 LT |
21 | #include <linux/mman.h> |
22 | #include <linux/swap.h> | |
23 | #include <linux/file.h> | |
24 | #include <linux/highmem.h> | |
25 | #include <linux/pagemap.h> | |
26 | #include <linux/slab.h> | |
27 | #include <linux/vmalloc.h> | |
1da177e4 LT |
28 | #include <linux/blkdev.h> |
29 | #include <linux/backing-dev.h> | |
3b32123d | 30 | #include <linux/compiler.h> |
1da177e4 LT |
31 | #include <linux/mount.h> |
32 | #include <linux/personality.h> | |
33 | #include <linux/security.h> | |
34 | #include <linux/syscalls.h> | |
120a795d | 35 | #include <linux/audit.h> |
cf4aebc2 | 36 | #include <linux/sched/sysctl.h> |
b1de0d13 | 37 | #include <linux/printk.h> |
1da177e4 LT |
38 | |
39 | #include <asm/uaccess.h> | |
40 | #include <asm/tlb.h> | |
41 | #include <asm/tlbflush.h> | |
eb8cdec4 | 42 | #include <asm/mmu_context.h> |
8feae131 DH |
43 | #include "internal.h" |
44 | ||
8feae131 DH |
45 | #if 0 |
46 | #define kenter(FMT, ...) \ | |
47 | printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__) | |
48 | #define kleave(FMT, ...) \ | |
49 | printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__) | |
50 | #define kdebug(FMT, ...) \ | |
51 | printk(KERN_DEBUG "xxx" FMT"yyy\n", ##__VA_ARGS__) | |
52 | #else | |
53 | #define kenter(FMT, ...) \ | |
54 | no_printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__) | |
55 | #define kleave(FMT, ...) \ | |
56 | no_printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__) | |
57 | #define kdebug(FMT, ...) \ | |
58 | no_printk(KERN_DEBUG FMT"\n", ##__VA_ARGS__) | |
59 | #endif | |
1da177e4 LT |
60 | |
61 | void *high_memory; | |
62 | struct page *mem_map; | |
63 | unsigned long max_mapnr; | |
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% */ | |
49f0ce5f | 68 | unsigned long sysctl_overcommit_kbytes __read_mostly; |
1da177e4 | 69 | int sysctl_max_map_count = DEFAULT_MAX_MAP_COUNT; |
fc4d5c29 | 70 | int sysctl_nr_trim_pages = CONFIG_NOMMU_INITIAL_TRIM_EXCESS; |
c9b1d098 | 71 | unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */ |
4eeab4f5 | 72 | unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */ |
1da177e4 LT |
73 | int heap_stack_gap = 0; |
74 | ||
33e5d769 | 75 | atomic_long_t mmap_pages_allocated; |
8feae131 | 76 | |
997071bc S |
77 | /* |
78 | * The global memory commitment made in the system can be a metric | |
79 | * that can be used to drive ballooning decisions when Linux is hosted | |
80 | * as a guest. On Hyper-V, the host implements a policy engine for dynamically | |
81 | * balancing memory across competing virtual machines that are hosted. | |
82 | * Several metrics drive this policy engine including the guest reported | |
83 | * memory commitment. | |
84 | */ | |
85 | unsigned long vm_memory_committed(void) | |
86 | { | |
87 | return percpu_counter_read_positive(&vm_committed_as); | |
88 | } | |
89 | ||
90 | EXPORT_SYMBOL_GPL(vm_memory_committed); | |
91 | ||
1da177e4 | 92 | EXPORT_SYMBOL(mem_map); |
1da177e4 | 93 | |
8feae131 DH |
94 | /* list of mapped, potentially shareable regions */ |
95 | static struct kmem_cache *vm_region_jar; | |
96 | struct rb_root nommu_region_tree = RB_ROOT; | |
97 | DECLARE_RWSEM(nommu_region_sem); | |
1da177e4 | 98 | |
f0f37e2f | 99 | const struct vm_operations_struct generic_file_vm_ops = { |
1da177e4 LT |
100 | }; |
101 | ||
1da177e4 LT |
102 | /* |
103 | * Return the total memory allocated for this pointer, not | |
104 | * just what the caller asked for. | |
105 | * | |
106 | * Doesn't have to be accurate, i.e. may have races. | |
107 | */ | |
108 | unsigned int kobjsize(const void *objp) | |
109 | { | |
110 | struct page *page; | |
111 | ||
4016a139 MH |
112 | /* |
113 | * If the object we have should not have ksize performed on it, | |
114 | * return size of 0 | |
115 | */ | |
5a1603be | 116 | if (!objp || !virt_addr_valid(objp)) |
6cfd53fc PM |
117 | return 0; |
118 | ||
119 | page = virt_to_head_page(objp); | |
6cfd53fc PM |
120 | |
121 | /* | |
122 | * If the allocator sets PageSlab, we know the pointer came from | |
123 | * kmalloc(). | |
124 | */ | |
1da177e4 LT |
125 | if (PageSlab(page)) |
126 | return ksize(objp); | |
127 | ||
ab2e83ea PM |
128 | /* |
129 | * If it's not a compound page, see if we have a matching VMA | |
130 | * region. This test is intentionally done in reverse order, | |
131 | * so if there's no VMA, we still fall through and hand back | |
132 | * PAGE_SIZE for 0-order pages. | |
133 | */ | |
134 | if (!PageCompound(page)) { | |
135 | struct vm_area_struct *vma; | |
136 | ||
137 | vma = find_vma(current->mm, (unsigned long)objp); | |
138 | if (vma) | |
139 | return vma->vm_end - vma->vm_start; | |
140 | } | |
141 | ||
6cfd53fc PM |
142 | /* |
143 | * The ksize() function is only guaranteed to work for pointers | |
5a1603be | 144 | * returned by kmalloc(). So handle arbitrary pointers here. |
6cfd53fc | 145 | */ |
5a1603be | 146 | return PAGE_SIZE << compound_order(page); |
1da177e4 LT |
147 | } |
148 | ||
28a35716 ML |
149 | long __get_user_pages(struct task_struct *tsk, struct mm_struct *mm, |
150 | unsigned long start, unsigned long nr_pages, | |
151 | unsigned int foll_flags, struct page **pages, | |
152 | struct vm_area_struct **vmas, int *nonblocking) | |
1da177e4 | 153 | { |
910e46da | 154 | struct vm_area_struct *vma; |
7b4d5b8b DH |
155 | unsigned long vm_flags; |
156 | int i; | |
157 | ||
158 | /* calculate required read or write permissions. | |
58fa879e | 159 | * If FOLL_FORCE is set, we only require the "MAY" flags. |
7b4d5b8b | 160 | */ |
58fa879e HD |
161 | vm_flags = (foll_flags & FOLL_WRITE) ? |
162 | (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD); | |
163 | vm_flags &= (foll_flags & FOLL_FORCE) ? | |
164 | (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE); | |
1da177e4 | 165 | |
9d73777e | 166 | for (i = 0; i < nr_pages; i++) { |
7561e8ca | 167 | vma = find_vma(mm, start); |
7b4d5b8b DH |
168 | if (!vma) |
169 | goto finish_or_fault; | |
170 | ||
171 | /* protect what we can, including chardevs */ | |
1c3aff1c HD |
172 | if ((vma->vm_flags & (VM_IO | VM_PFNMAP)) || |
173 | !(vm_flags & vma->vm_flags)) | |
7b4d5b8b | 174 | goto finish_or_fault; |
910e46da | 175 | |
1da177e4 LT |
176 | if (pages) { |
177 | pages[i] = virt_to_page(start); | |
178 | if (pages[i]) | |
179 | page_cache_get(pages[i]); | |
180 | } | |
181 | if (vmas) | |
910e46da | 182 | vmas[i] = vma; |
e1ee65d8 | 183 | start = (start + PAGE_SIZE) & PAGE_MASK; |
1da177e4 | 184 | } |
7b4d5b8b DH |
185 | |
186 | return i; | |
187 | ||
188 | finish_or_fault: | |
189 | return i ? : -EFAULT; | |
1da177e4 | 190 | } |
b291f000 | 191 | |
b291f000 NP |
192 | /* |
193 | * get a list of pages in an address range belonging to the specified process | |
194 | * and indicate the VMA that covers each page | |
195 | * - this is potentially dodgy as we may end incrementing the page count of a | |
196 | * slab page or a secondary page from a compound page | |
197 | * - don't permit access to VMAs that don't support it, such as I/O mappings | |
198 | */ | |
28a35716 ML |
199 | long get_user_pages(struct task_struct *tsk, struct mm_struct *mm, |
200 | unsigned long start, unsigned long nr_pages, | |
201 | int write, int force, struct page **pages, | |
202 | struct vm_area_struct **vmas) | |
b291f000 NP |
203 | { |
204 | int flags = 0; | |
205 | ||
206 | if (write) | |
58fa879e | 207 | flags |= FOLL_WRITE; |
b291f000 | 208 | if (force) |
58fa879e | 209 | flags |= FOLL_FORCE; |
b291f000 | 210 | |
53a7706d ML |
211 | return __get_user_pages(tsk, mm, start, nr_pages, flags, pages, vmas, |
212 | NULL); | |
b291f000 | 213 | } |
66aa2b4b GU |
214 | EXPORT_SYMBOL(get_user_pages); |
215 | ||
dfc2f91a PM |
216 | /** |
217 | * follow_pfn - look up PFN at a user virtual address | |
218 | * @vma: memory mapping | |
219 | * @address: user virtual address | |
220 | * @pfn: location to store found PFN | |
221 | * | |
222 | * Only IO mappings and raw PFN mappings are allowed. | |
223 | * | |
224 | * Returns zero and the pfn at @pfn on success, -ve otherwise. | |
225 | */ | |
226 | int follow_pfn(struct vm_area_struct *vma, unsigned long address, | |
227 | unsigned long *pfn) | |
228 | { | |
229 | if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) | |
230 | return -EINVAL; | |
231 | ||
232 | *pfn = address >> PAGE_SHIFT; | |
233 | return 0; | |
234 | } | |
235 | EXPORT_SYMBOL(follow_pfn); | |
236 | ||
f1c4069e | 237 | LIST_HEAD(vmap_area_list); |
1da177e4 | 238 | |
b3bdda02 | 239 | void vfree(const void *addr) |
1da177e4 LT |
240 | { |
241 | kfree(addr); | |
242 | } | |
b5073173 | 243 | EXPORT_SYMBOL(vfree); |
1da177e4 | 244 | |
dd0fc66f | 245 | void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot) |
1da177e4 LT |
246 | { |
247 | /* | |
8518609d RD |
248 | * You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc() |
249 | * returns only a logical address. | |
1da177e4 | 250 | */ |
84097518 | 251 | return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM); |
1da177e4 | 252 | } |
b5073173 | 253 | EXPORT_SYMBOL(__vmalloc); |
1da177e4 | 254 | |
f905bc44 PM |
255 | void *vmalloc_user(unsigned long size) |
256 | { | |
257 | void *ret; | |
258 | ||
259 | ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, | |
260 | PAGE_KERNEL); | |
261 | if (ret) { | |
262 | struct vm_area_struct *vma; | |
263 | ||
264 | down_write(¤t->mm->mmap_sem); | |
265 | vma = find_vma(current->mm, (unsigned long)ret); | |
266 | if (vma) | |
267 | vma->vm_flags |= VM_USERMAP; | |
268 | up_write(¤t->mm->mmap_sem); | |
269 | } | |
270 | ||
271 | return ret; | |
272 | } | |
273 | EXPORT_SYMBOL(vmalloc_user); | |
274 | ||
b3bdda02 | 275 | struct page *vmalloc_to_page(const void *addr) |
1da177e4 LT |
276 | { |
277 | return virt_to_page(addr); | |
278 | } | |
b5073173 | 279 | EXPORT_SYMBOL(vmalloc_to_page); |
1da177e4 | 280 | |
b3bdda02 | 281 | unsigned long vmalloc_to_pfn(const void *addr) |
1da177e4 LT |
282 | { |
283 | return page_to_pfn(virt_to_page(addr)); | |
284 | } | |
b5073173 | 285 | EXPORT_SYMBOL(vmalloc_to_pfn); |
1da177e4 LT |
286 | |
287 | long vread(char *buf, char *addr, unsigned long count) | |
288 | { | |
9bde916b CG |
289 | /* Don't allow overflow */ |
290 | if ((unsigned long) buf + count < count) | |
291 | count = -(unsigned long) buf; | |
292 | ||
1da177e4 LT |
293 | memcpy(buf, addr, count); |
294 | return count; | |
295 | } | |
296 | ||
297 | long vwrite(char *buf, char *addr, unsigned long count) | |
298 | { | |
299 | /* Don't allow overflow */ | |
300 | if ((unsigned long) addr + count < count) | |
301 | count = -(unsigned long) addr; | |
302 | ||
303 | memcpy(addr, buf, count); | |
ac714904 | 304 | return count; |
1da177e4 LT |
305 | } |
306 | ||
307 | /* | |
308 | * vmalloc - allocate virtually continguos memory | |
309 | * | |
310 | * @size: allocation size | |
311 | * | |
312 | * Allocate enough pages to cover @size from the page level | |
313 | * allocator and map them into continguos kernel virtual space. | |
314 | * | |
c1c8897f | 315 | * For tight control over page level allocator and protection flags |
1da177e4 LT |
316 | * use __vmalloc() instead. |
317 | */ | |
318 | void *vmalloc(unsigned long size) | |
319 | { | |
320 | return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL); | |
321 | } | |
f6138882 AM |
322 | EXPORT_SYMBOL(vmalloc); |
323 | ||
e1ca7788 DY |
324 | /* |
325 | * vzalloc - allocate virtually continguos memory with zero fill | |
326 | * | |
327 | * @size: allocation size | |
328 | * | |
329 | * Allocate enough pages to cover @size from the page level | |
330 | * allocator and map them into continguos kernel virtual space. | |
331 | * The memory allocated is set to zero. | |
332 | * | |
333 | * For tight control over page level allocator and protection flags | |
334 | * use __vmalloc() instead. | |
335 | */ | |
336 | void *vzalloc(unsigned long size) | |
337 | { | |
338 | return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, | |
339 | PAGE_KERNEL); | |
340 | } | |
341 | EXPORT_SYMBOL(vzalloc); | |
342 | ||
343 | /** | |
344 | * vmalloc_node - allocate memory on a specific node | |
345 | * @size: allocation size | |
346 | * @node: numa node | |
347 | * | |
348 | * Allocate enough pages to cover @size from the page level | |
349 | * allocator and map them into contiguous kernel virtual space. | |
350 | * | |
351 | * For tight control over page level allocator and protection flags | |
352 | * use __vmalloc() instead. | |
353 | */ | |
f6138882 AM |
354 | void *vmalloc_node(unsigned long size, int node) |
355 | { | |
356 | return vmalloc(size); | |
357 | } | |
9a14f653 | 358 | EXPORT_SYMBOL(vmalloc_node); |
e1ca7788 DY |
359 | |
360 | /** | |
361 | * vzalloc_node - allocate memory on a specific node with zero fill | |
362 | * @size: allocation size | |
363 | * @node: numa node | |
364 | * | |
365 | * Allocate enough pages to cover @size from the page level | |
366 | * allocator and map them into contiguous kernel virtual space. | |
367 | * The memory allocated is set to zero. | |
368 | * | |
369 | * For tight control over page level allocator and protection flags | |
370 | * use __vmalloc() instead. | |
371 | */ | |
372 | void *vzalloc_node(unsigned long size, int node) | |
373 | { | |
374 | return vzalloc(size); | |
375 | } | |
376 | EXPORT_SYMBOL(vzalloc_node); | |
1da177e4 | 377 | |
1af446ed PM |
378 | #ifndef PAGE_KERNEL_EXEC |
379 | # define PAGE_KERNEL_EXEC PAGE_KERNEL | |
380 | #endif | |
381 | ||
382 | /** | |
383 | * vmalloc_exec - allocate virtually contiguous, executable memory | |
384 | * @size: allocation size | |
385 | * | |
386 | * Kernel-internal function to allocate enough pages to cover @size | |
387 | * the page level allocator and map them into contiguous and | |
388 | * executable kernel virtual space. | |
389 | * | |
390 | * For tight control over page level allocator and protection flags | |
391 | * use __vmalloc() instead. | |
392 | */ | |
393 | ||
394 | void *vmalloc_exec(unsigned long size) | |
395 | { | |
396 | return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC); | |
397 | } | |
398 | ||
b5073173 PM |
399 | /** |
400 | * vmalloc_32 - allocate virtually contiguous memory (32bit addressable) | |
1da177e4 LT |
401 | * @size: allocation size |
402 | * | |
403 | * Allocate enough 32bit PA addressable pages to cover @size from the | |
404 | * page level allocator and map them into continguos kernel virtual space. | |
405 | */ | |
406 | void *vmalloc_32(unsigned long size) | |
407 | { | |
408 | return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL); | |
409 | } | |
b5073173 PM |
410 | EXPORT_SYMBOL(vmalloc_32); |
411 | ||
412 | /** | |
413 | * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory | |
414 | * @size: allocation size | |
415 | * | |
416 | * The resulting memory area is 32bit addressable and zeroed so it can be | |
417 | * mapped to userspace without leaking data. | |
f905bc44 PM |
418 | * |
419 | * VM_USERMAP is set on the corresponding VMA so that subsequent calls to | |
420 | * remap_vmalloc_range() are permissible. | |
b5073173 PM |
421 | */ |
422 | void *vmalloc_32_user(unsigned long size) | |
423 | { | |
f905bc44 PM |
424 | /* |
425 | * We'll have to sort out the ZONE_DMA bits for 64-bit, | |
426 | * but for now this can simply use vmalloc_user() directly. | |
427 | */ | |
428 | return vmalloc_user(size); | |
b5073173 PM |
429 | } |
430 | EXPORT_SYMBOL(vmalloc_32_user); | |
1da177e4 LT |
431 | |
432 | void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot) | |
433 | { | |
434 | BUG(); | |
435 | return NULL; | |
436 | } | |
b5073173 | 437 | EXPORT_SYMBOL(vmap); |
1da177e4 | 438 | |
b3bdda02 | 439 | void vunmap(const void *addr) |
1da177e4 LT |
440 | { |
441 | BUG(); | |
442 | } | |
b5073173 | 443 | EXPORT_SYMBOL(vunmap); |
1da177e4 | 444 | |
eb6434d9 PM |
445 | void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot) |
446 | { | |
447 | BUG(); | |
448 | return NULL; | |
449 | } | |
450 | EXPORT_SYMBOL(vm_map_ram); | |
451 | ||
452 | void vm_unmap_ram(const void *mem, unsigned int count) | |
453 | { | |
454 | BUG(); | |
455 | } | |
456 | EXPORT_SYMBOL(vm_unmap_ram); | |
457 | ||
458 | void vm_unmap_aliases(void) | |
459 | { | |
460 | } | |
461 | EXPORT_SYMBOL_GPL(vm_unmap_aliases); | |
462 | ||
1eeb66a1 CH |
463 | /* |
464 | * Implement a stub for vmalloc_sync_all() if the architecture chose not to | |
465 | * have one. | |
466 | */ | |
3b32123d | 467 | void __weak vmalloc_sync_all(void) |
1eeb66a1 CH |
468 | { |
469 | } | |
470 | ||
29c185e5 PM |
471 | /** |
472 | * alloc_vm_area - allocate a range of kernel address space | |
473 | * @size: size of the area | |
474 | * | |
475 | * Returns: NULL on failure, vm_struct on success | |
476 | * | |
477 | * This function reserves a range of kernel address space, and | |
478 | * allocates pagetables to map that range. No actual mappings | |
479 | * are created. If the kernel address space is not shared | |
480 | * between processes, it syncs the pagetable across all | |
481 | * processes. | |
482 | */ | |
cd12909c | 483 | struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes) |
29c185e5 PM |
484 | { |
485 | BUG(); | |
486 | return NULL; | |
487 | } | |
488 | EXPORT_SYMBOL_GPL(alloc_vm_area); | |
489 | ||
490 | void free_vm_area(struct vm_struct *area) | |
491 | { | |
492 | BUG(); | |
493 | } | |
494 | EXPORT_SYMBOL_GPL(free_vm_area); | |
495 | ||
b5073173 PM |
496 | int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, |
497 | struct page *page) | |
498 | { | |
499 | return -EINVAL; | |
500 | } | |
501 | EXPORT_SYMBOL(vm_insert_page); | |
502 | ||
1da177e4 LT |
503 | /* |
504 | * sys_brk() for the most part doesn't need the global kernel | |
505 | * lock, except when an application is doing something nasty | |
506 | * like trying to un-brk an area that has already been mapped | |
507 | * to a regular file. in this case, the unmapping will need | |
508 | * to invoke file system routines that need the global lock. | |
509 | */ | |
6a6160a7 | 510 | SYSCALL_DEFINE1(brk, unsigned long, brk) |
1da177e4 LT |
511 | { |
512 | struct mm_struct *mm = current->mm; | |
513 | ||
514 | if (brk < mm->start_brk || brk > mm->context.end_brk) | |
515 | return mm->brk; | |
516 | ||
517 | if (mm->brk == brk) | |
518 | return mm->brk; | |
519 | ||
520 | /* | |
521 | * Always allow shrinking brk | |
522 | */ | |
523 | if (brk <= mm->brk) { | |
524 | mm->brk = brk; | |
525 | return brk; | |
526 | } | |
527 | ||
528 | /* | |
529 | * Ok, looks good - let it rip. | |
530 | */ | |
cfe79c00 | 531 | flush_icache_range(mm->brk, brk); |
1da177e4 LT |
532 | return mm->brk = brk; |
533 | } | |
534 | ||
8feae131 DH |
535 | /* |
536 | * initialise the VMA and region record slabs | |
537 | */ | |
538 | void __init mmap_init(void) | |
1da177e4 | 539 | { |
00a62ce9 KM |
540 | int ret; |
541 | ||
908c7f19 | 542 | ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL); |
00a62ce9 | 543 | VM_BUG_ON(ret); |
33e5d769 | 544 | vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC); |
1da177e4 | 545 | } |
1da177e4 | 546 | |
3034097a | 547 | /* |
8feae131 DH |
548 | * validate the region tree |
549 | * - the caller must hold the region lock | |
3034097a | 550 | */ |
8feae131 DH |
551 | #ifdef CONFIG_DEBUG_NOMMU_REGIONS |
552 | static noinline void validate_nommu_regions(void) | |
3034097a | 553 | { |
8feae131 DH |
554 | struct vm_region *region, *last; |
555 | struct rb_node *p, *lastp; | |
3034097a | 556 | |
8feae131 DH |
557 | lastp = rb_first(&nommu_region_tree); |
558 | if (!lastp) | |
559 | return; | |
560 | ||
561 | last = rb_entry(lastp, struct vm_region, vm_rb); | |
33e5d769 DH |
562 | BUG_ON(unlikely(last->vm_end <= last->vm_start)); |
563 | BUG_ON(unlikely(last->vm_top < last->vm_end)); | |
8feae131 DH |
564 | |
565 | while ((p = rb_next(lastp))) { | |
566 | region = rb_entry(p, struct vm_region, vm_rb); | |
567 | last = rb_entry(lastp, struct vm_region, vm_rb); | |
568 | ||
33e5d769 DH |
569 | BUG_ON(unlikely(region->vm_end <= region->vm_start)); |
570 | BUG_ON(unlikely(region->vm_top < region->vm_end)); | |
571 | BUG_ON(unlikely(region->vm_start < last->vm_top)); | |
3034097a | 572 | |
8feae131 DH |
573 | lastp = p; |
574 | } | |
3034097a | 575 | } |
8feae131 | 576 | #else |
33e5d769 DH |
577 | static void validate_nommu_regions(void) |
578 | { | |
579 | } | |
8feae131 | 580 | #endif |
3034097a DH |
581 | |
582 | /* | |
8feae131 | 583 | * add a region into the global tree |
3034097a | 584 | */ |
8feae131 | 585 | static void add_nommu_region(struct vm_region *region) |
3034097a | 586 | { |
8feae131 DH |
587 | struct vm_region *pregion; |
588 | struct rb_node **p, *parent; | |
3034097a | 589 | |
8feae131 DH |
590 | validate_nommu_regions(); |
591 | ||
8feae131 DH |
592 | parent = NULL; |
593 | p = &nommu_region_tree.rb_node; | |
594 | while (*p) { | |
595 | parent = *p; | |
596 | pregion = rb_entry(parent, struct vm_region, vm_rb); | |
597 | if (region->vm_start < pregion->vm_start) | |
598 | p = &(*p)->rb_left; | |
599 | else if (region->vm_start > pregion->vm_start) | |
600 | p = &(*p)->rb_right; | |
601 | else if (pregion == region) | |
602 | return; | |
603 | else | |
604 | BUG(); | |
3034097a DH |
605 | } |
606 | ||
8feae131 DH |
607 | rb_link_node(®ion->vm_rb, parent, p); |
608 | rb_insert_color(®ion->vm_rb, &nommu_region_tree); | |
3034097a | 609 | |
8feae131 | 610 | validate_nommu_regions(); |
3034097a | 611 | } |
3034097a | 612 | |
930e652a | 613 | /* |
8feae131 | 614 | * delete a region from the global tree |
930e652a | 615 | */ |
8feae131 | 616 | static void delete_nommu_region(struct vm_region *region) |
930e652a | 617 | { |
8feae131 | 618 | BUG_ON(!nommu_region_tree.rb_node); |
930e652a | 619 | |
8feae131 DH |
620 | validate_nommu_regions(); |
621 | rb_erase(®ion->vm_rb, &nommu_region_tree); | |
622 | validate_nommu_regions(); | |
57c8f63e GU |
623 | } |
624 | ||
6fa5f80b | 625 | /* |
8feae131 | 626 | * free a contiguous series of pages |
6fa5f80b | 627 | */ |
8feae131 | 628 | static void free_page_series(unsigned long from, unsigned long to) |
6fa5f80b | 629 | { |
8feae131 DH |
630 | for (; from < to; from += PAGE_SIZE) { |
631 | struct page *page = virt_to_page(from); | |
632 | ||
633 | kdebug("- free %lx", from); | |
33e5d769 | 634 | atomic_long_dec(&mmap_pages_allocated); |
8feae131 | 635 | if (page_count(page) != 1) |
33e5d769 DH |
636 | kdebug("free page %p: refcount not one: %d", |
637 | page, page_count(page)); | |
8feae131 | 638 | put_page(page); |
6fa5f80b | 639 | } |
6fa5f80b DH |
640 | } |
641 | ||
3034097a | 642 | /* |
8feae131 | 643 | * release a reference to a region |
33e5d769 | 644 | * - the caller must hold the region semaphore for writing, which this releases |
dd8632a1 | 645 | * - the region may not have been added to the tree yet, in which case vm_top |
8feae131 | 646 | * will equal vm_start |
3034097a | 647 | */ |
8feae131 DH |
648 | static void __put_nommu_region(struct vm_region *region) |
649 | __releases(nommu_region_sem) | |
1da177e4 | 650 | { |
1e2ae599 | 651 | kenter("%p{%d}", region, region->vm_usage); |
1da177e4 | 652 | |
8feae131 | 653 | BUG_ON(!nommu_region_tree.rb_node); |
1da177e4 | 654 | |
1e2ae599 | 655 | if (--region->vm_usage == 0) { |
dd8632a1 | 656 | if (region->vm_top > region->vm_start) |
8feae131 DH |
657 | delete_nommu_region(region); |
658 | up_write(&nommu_region_sem); | |
659 | ||
660 | if (region->vm_file) | |
661 | fput(region->vm_file); | |
662 | ||
663 | /* IO memory and memory shared directly out of the pagecache | |
664 | * from ramfs/tmpfs mustn't be released here */ | |
665 | if (region->vm_flags & VM_MAPPED_COPY) { | |
666 | kdebug("free series"); | |
dd8632a1 | 667 | free_page_series(region->vm_start, region->vm_top); |
8feae131 DH |
668 | } |
669 | kmem_cache_free(vm_region_jar, region); | |
670 | } else { | |
671 | up_write(&nommu_region_sem); | |
1da177e4 | 672 | } |
8feae131 | 673 | } |
1da177e4 | 674 | |
8feae131 DH |
675 | /* |
676 | * release a reference to a region | |
677 | */ | |
678 | static void put_nommu_region(struct vm_region *region) | |
679 | { | |
680 | down_write(&nommu_region_sem); | |
681 | __put_nommu_region(region); | |
1da177e4 LT |
682 | } |
683 | ||
eb8cdec4 BS |
684 | /* |
685 | * update protection on a vma | |
686 | */ | |
687 | static void protect_vma(struct vm_area_struct *vma, unsigned long flags) | |
688 | { | |
689 | #ifdef CONFIG_MPU | |
690 | struct mm_struct *mm = vma->vm_mm; | |
691 | long start = vma->vm_start & PAGE_MASK; | |
692 | while (start < vma->vm_end) { | |
693 | protect_page(mm, start, flags); | |
694 | start += PAGE_SIZE; | |
695 | } | |
696 | update_protections(mm); | |
697 | #endif | |
698 | } | |
699 | ||
3034097a | 700 | /* |
8feae131 DH |
701 | * add a VMA into a process's mm_struct in the appropriate place in the list |
702 | * and tree and add to the address space's page tree also if not an anonymous | |
703 | * page | |
704 | * - should be called with mm->mmap_sem held writelocked | |
3034097a | 705 | */ |
8feae131 | 706 | static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma) |
1da177e4 | 707 | { |
6038def0 | 708 | struct vm_area_struct *pvma, *prev; |
1da177e4 | 709 | struct address_space *mapping; |
6038def0 | 710 | struct rb_node **p, *parent, *rb_prev; |
8feae131 DH |
711 | |
712 | kenter(",%p", vma); | |
713 | ||
714 | BUG_ON(!vma->vm_region); | |
715 | ||
716 | mm->map_count++; | |
717 | vma->vm_mm = mm; | |
1da177e4 | 718 | |
eb8cdec4 BS |
719 | protect_vma(vma, vma->vm_flags); |
720 | ||
1da177e4 LT |
721 | /* add the VMA to the mapping */ |
722 | if (vma->vm_file) { | |
723 | mapping = vma->vm_file->f_mapping; | |
724 | ||
83cde9e8 | 725 | i_mmap_lock_write(mapping); |
1da177e4 | 726 | flush_dcache_mmap_lock(mapping); |
6b2dbba8 | 727 | vma_interval_tree_insert(vma, &mapping->i_mmap); |
1da177e4 | 728 | flush_dcache_mmap_unlock(mapping); |
83cde9e8 | 729 | i_mmap_unlock_write(mapping); |
1da177e4 LT |
730 | } |
731 | ||
8feae131 | 732 | /* add the VMA to the tree */ |
6038def0 | 733 | parent = rb_prev = NULL; |
8feae131 | 734 | p = &mm->mm_rb.rb_node; |
1da177e4 LT |
735 | while (*p) { |
736 | parent = *p; | |
737 | pvma = rb_entry(parent, struct vm_area_struct, vm_rb); | |
738 | ||
8feae131 DH |
739 | /* sort by: start addr, end addr, VMA struct addr in that order |
740 | * (the latter is necessary as we may get identical VMAs) */ | |
741 | if (vma->vm_start < pvma->vm_start) | |
1da177e4 | 742 | p = &(*p)->rb_left; |
6038def0 NK |
743 | else if (vma->vm_start > pvma->vm_start) { |
744 | rb_prev = parent; | |
1da177e4 | 745 | p = &(*p)->rb_right; |
6038def0 | 746 | } else if (vma->vm_end < pvma->vm_end) |
8feae131 | 747 | p = &(*p)->rb_left; |
6038def0 NK |
748 | else if (vma->vm_end > pvma->vm_end) { |
749 | rb_prev = parent; | |
8feae131 | 750 | p = &(*p)->rb_right; |
6038def0 | 751 | } else if (vma < pvma) |
8feae131 | 752 | p = &(*p)->rb_left; |
6038def0 NK |
753 | else if (vma > pvma) { |
754 | rb_prev = parent; | |
8feae131 | 755 | p = &(*p)->rb_right; |
6038def0 | 756 | } else |
8feae131 | 757 | BUG(); |
1da177e4 LT |
758 | } |
759 | ||
760 | rb_link_node(&vma->vm_rb, parent, p); | |
8feae131 DH |
761 | rb_insert_color(&vma->vm_rb, &mm->mm_rb); |
762 | ||
763 | /* add VMA to the VMA list also */ | |
6038def0 NK |
764 | prev = NULL; |
765 | if (rb_prev) | |
766 | prev = rb_entry(rb_prev, struct vm_area_struct, vm_rb); | |
8feae131 | 767 | |
6038def0 | 768 | __vma_link_list(mm, vma, prev, parent); |
1da177e4 LT |
769 | } |
770 | ||
3034097a | 771 | /* |
8feae131 | 772 | * delete a VMA from its owning mm_struct and address space |
3034097a | 773 | */ |
8feae131 | 774 | static void delete_vma_from_mm(struct vm_area_struct *vma) |
1da177e4 | 775 | { |
615d6e87 | 776 | int i; |
1da177e4 | 777 | struct address_space *mapping; |
8feae131 | 778 | struct mm_struct *mm = vma->vm_mm; |
615d6e87 | 779 | struct task_struct *curr = current; |
8feae131 DH |
780 | |
781 | kenter("%p", vma); | |
782 | ||
eb8cdec4 BS |
783 | protect_vma(vma, 0); |
784 | ||
8feae131 | 785 | mm->map_count--; |
615d6e87 DB |
786 | for (i = 0; i < VMACACHE_SIZE; i++) { |
787 | /* if the vma is cached, invalidate the entire cache */ | |
788 | if (curr->vmacache[i] == vma) { | |
e020d5bd | 789 | vmacache_invalidate(mm); |
615d6e87 DB |
790 | break; |
791 | } | |
792 | } | |
1da177e4 LT |
793 | |
794 | /* remove the VMA from the mapping */ | |
795 | if (vma->vm_file) { | |
796 | mapping = vma->vm_file->f_mapping; | |
797 | ||
83cde9e8 | 798 | i_mmap_lock_write(mapping); |
1da177e4 | 799 | flush_dcache_mmap_lock(mapping); |
6b2dbba8 | 800 | vma_interval_tree_remove(vma, &mapping->i_mmap); |
1da177e4 | 801 | flush_dcache_mmap_unlock(mapping); |
83cde9e8 | 802 | i_mmap_unlock_write(mapping); |
1da177e4 LT |
803 | } |
804 | ||
8feae131 DH |
805 | /* remove from the MM's tree and list */ |
806 | rb_erase(&vma->vm_rb, &mm->mm_rb); | |
b951bf2c NK |
807 | |
808 | if (vma->vm_prev) | |
809 | vma->vm_prev->vm_next = vma->vm_next; | |
810 | else | |
811 | mm->mmap = vma->vm_next; | |
812 | ||
813 | if (vma->vm_next) | |
814 | vma->vm_next->vm_prev = vma->vm_prev; | |
8feae131 DH |
815 | } |
816 | ||
817 | /* | |
818 | * destroy a VMA record | |
819 | */ | |
820 | static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma) | |
821 | { | |
822 | kenter("%p", vma); | |
823 | if (vma->vm_ops && vma->vm_ops->close) | |
824 | vma->vm_ops->close(vma); | |
e9714acf | 825 | if (vma->vm_file) |
8feae131 | 826 | fput(vma->vm_file); |
8feae131 DH |
827 | put_nommu_region(vma->vm_region); |
828 | kmem_cache_free(vm_area_cachep, vma); | |
829 | } | |
830 | ||
831 | /* | |
832 | * look up the first VMA in which addr resides, NULL if none | |
833 | * - should be called with mm->mmap_sem at least held readlocked | |
834 | */ | |
835 | struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr) | |
836 | { | |
837 | struct vm_area_struct *vma; | |
8feae131 DH |
838 | |
839 | /* check the cache first */ | |
615d6e87 DB |
840 | vma = vmacache_find(mm, addr); |
841 | if (likely(vma)) | |
8feae131 DH |
842 | return vma; |
843 | ||
e922c4c5 | 844 | /* trawl the list (there may be multiple mappings in which addr |
8feae131 | 845 | * resides) */ |
e922c4c5 | 846 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
8feae131 DH |
847 | if (vma->vm_start > addr) |
848 | return NULL; | |
849 | if (vma->vm_end > addr) { | |
615d6e87 | 850 | vmacache_update(addr, vma); |
8feae131 DH |
851 | return vma; |
852 | } | |
853 | } | |
854 | ||
855 | return NULL; | |
856 | } | |
857 | EXPORT_SYMBOL(find_vma); | |
858 | ||
859 | /* | |
860 | * find a VMA | |
861 | * - we don't extend stack VMAs under NOMMU conditions | |
862 | */ | |
863 | struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr) | |
864 | { | |
7561e8ca | 865 | return find_vma(mm, addr); |
8feae131 DH |
866 | } |
867 | ||
868 | /* | |
869 | * expand a stack to a given address | |
870 | * - not supported under NOMMU conditions | |
871 | */ | |
872 | int expand_stack(struct vm_area_struct *vma, unsigned long address) | |
873 | { | |
874 | return -ENOMEM; | |
875 | } | |
876 | ||
877 | /* | |
878 | * look up the first VMA exactly that exactly matches addr | |
879 | * - should be called with mm->mmap_sem at least held readlocked | |
880 | */ | |
881 | static struct vm_area_struct *find_vma_exact(struct mm_struct *mm, | |
882 | unsigned long addr, | |
883 | unsigned long len) | |
884 | { | |
885 | struct vm_area_struct *vma; | |
8feae131 DH |
886 | unsigned long end = addr + len; |
887 | ||
888 | /* check the cache first */ | |
615d6e87 DB |
889 | vma = vmacache_find_exact(mm, addr, end); |
890 | if (vma) | |
8feae131 DH |
891 | return vma; |
892 | ||
e922c4c5 | 893 | /* trawl the list (there may be multiple mappings in which addr |
8feae131 | 894 | * resides) */ |
e922c4c5 | 895 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
8feae131 DH |
896 | if (vma->vm_start < addr) |
897 | continue; | |
898 | if (vma->vm_start > addr) | |
899 | return NULL; | |
900 | if (vma->vm_end == end) { | |
615d6e87 | 901 | vmacache_update(addr, vma); |
8feae131 DH |
902 | return vma; |
903 | } | |
904 | } | |
905 | ||
906 | return NULL; | |
1da177e4 LT |
907 | } |
908 | ||
909 | /* | |
910 | * determine whether a mapping should be permitted and, if so, what sort of | |
911 | * mapping we're capable of supporting | |
912 | */ | |
913 | static int validate_mmap_request(struct file *file, | |
914 | unsigned long addr, | |
915 | unsigned long len, | |
916 | unsigned long prot, | |
917 | unsigned long flags, | |
918 | unsigned long pgoff, | |
919 | unsigned long *_capabilities) | |
920 | { | |
8feae131 | 921 | unsigned long capabilities, rlen; |
1da177e4 LT |
922 | int ret; |
923 | ||
924 | /* do the simple checks first */ | |
06aab5a3 | 925 | if (flags & MAP_FIXED) { |
1da177e4 LT |
926 | printk(KERN_DEBUG |
927 | "%d: Can't do fixed-address/overlay mmap of RAM\n", | |
928 | current->pid); | |
929 | return -EINVAL; | |
930 | } | |
931 | ||
932 | if ((flags & MAP_TYPE) != MAP_PRIVATE && | |
933 | (flags & MAP_TYPE) != MAP_SHARED) | |
934 | return -EINVAL; | |
935 | ||
f81cff0d | 936 | if (!len) |
1da177e4 LT |
937 | return -EINVAL; |
938 | ||
f81cff0d | 939 | /* Careful about overflows.. */ |
8feae131 DH |
940 | rlen = PAGE_ALIGN(len); |
941 | if (!rlen || rlen > TASK_SIZE) | |
f81cff0d MF |
942 | return -ENOMEM; |
943 | ||
1da177e4 | 944 | /* offset overflow? */ |
8feae131 | 945 | if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff) |
f81cff0d | 946 | return -EOVERFLOW; |
1da177e4 LT |
947 | |
948 | if (file) { | |
1da177e4 | 949 | /* files must support mmap */ |
72c2d531 | 950 | if (!file->f_op->mmap) |
1da177e4 LT |
951 | return -ENODEV; |
952 | ||
953 | /* work out if what we've got could possibly be shared | |
954 | * - we support chardevs that provide their own "memory" | |
955 | * - we support files/blockdevs that are memory backed | |
956 | */ | |
b4caecd4 CH |
957 | if (file->f_op->mmap_capabilities) { |
958 | capabilities = file->f_op->mmap_capabilities(file); | |
959 | } else { | |
1da177e4 LT |
960 | /* no explicit capabilities set, so assume some |
961 | * defaults */ | |
496ad9aa | 962 | switch (file_inode(file)->i_mode & S_IFMT) { |
1da177e4 LT |
963 | case S_IFREG: |
964 | case S_IFBLK: | |
b4caecd4 | 965 | capabilities = NOMMU_MAP_COPY; |
1da177e4 LT |
966 | break; |
967 | ||
968 | case S_IFCHR: | |
969 | capabilities = | |
b4caecd4 CH |
970 | NOMMU_MAP_DIRECT | |
971 | NOMMU_MAP_READ | | |
972 | NOMMU_MAP_WRITE; | |
1da177e4 LT |
973 | break; |
974 | ||
975 | default: | |
976 | return -EINVAL; | |
977 | } | |
978 | } | |
979 | ||
980 | /* eliminate any capabilities that we can't support on this | |
981 | * device */ | |
982 | if (!file->f_op->get_unmapped_area) | |
b4caecd4 | 983 | capabilities &= ~NOMMU_MAP_DIRECT; |
1da177e4 | 984 | if (!file->f_op->read) |
b4caecd4 | 985 | capabilities &= ~NOMMU_MAP_COPY; |
1da177e4 | 986 | |
28d7a6ae GY |
987 | /* The file shall have been opened with read permission. */ |
988 | if (!(file->f_mode & FMODE_READ)) | |
989 | return -EACCES; | |
990 | ||
1da177e4 LT |
991 | if (flags & MAP_SHARED) { |
992 | /* do checks for writing, appending and locking */ | |
993 | if ((prot & PROT_WRITE) && | |
994 | !(file->f_mode & FMODE_WRITE)) | |
995 | return -EACCES; | |
996 | ||
496ad9aa | 997 | if (IS_APPEND(file_inode(file)) && |
1da177e4 LT |
998 | (file->f_mode & FMODE_WRITE)) |
999 | return -EACCES; | |
1000 | ||
d7a06983 | 1001 | if (locks_verify_locked(file)) |
1da177e4 LT |
1002 | return -EAGAIN; |
1003 | ||
b4caecd4 | 1004 | if (!(capabilities & NOMMU_MAP_DIRECT)) |
1da177e4 LT |
1005 | return -ENODEV; |
1006 | ||
1da177e4 | 1007 | /* we mustn't privatise shared mappings */ |
b4caecd4 | 1008 | capabilities &= ~NOMMU_MAP_COPY; |
ac714904 | 1009 | } else { |
1da177e4 LT |
1010 | /* we're going to read the file into private memory we |
1011 | * allocate */ | |
b4caecd4 | 1012 | if (!(capabilities & NOMMU_MAP_COPY)) |
1da177e4 LT |
1013 | return -ENODEV; |
1014 | ||
1015 | /* we don't permit a private writable mapping to be | |
1016 | * shared with the backing device */ | |
1017 | if (prot & PROT_WRITE) | |
b4caecd4 | 1018 | capabilities &= ~NOMMU_MAP_DIRECT; |
1da177e4 LT |
1019 | } |
1020 | ||
b4caecd4 CH |
1021 | if (capabilities & NOMMU_MAP_DIRECT) { |
1022 | if (((prot & PROT_READ) && !(capabilities & NOMMU_MAP_READ)) || | |
1023 | ((prot & PROT_WRITE) && !(capabilities & NOMMU_MAP_WRITE)) || | |
1024 | ((prot & PROT_EXEC) && !(capabilities & NOMMU_MAP_EXEC)) | |
3c7b2045 | 1025 | ) { |
b4caecd4 | 1026 | capabilities &= ~NOMMU_MAP_DIRECT; |
3c7b2045 BS |
1027 | if (flags & MAP_SHARED) { |
1028 | printk(KERN_WARNING | |
1029 | "MAP_SHARED not completely supported on !MMU\n"); | |
1030 | return -EINVAL; | |
1031 | } | |
1032 | } | |
1033 | } | |
1034 | ||
1da177e4 LT |
1035 | /* handle executable mappings and implied executable |
1036 | * mappings */ | |
e9536ae7 | 1037 | if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) { |
1da177e4 LT |
1038 | if (prot & PROT_EXEC) |
1039 | return -EPERM; | |
ac714904 | 1040 | } else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) { |
1da177e4 LT |
1041 | /* handle implication of PROT_EXEC by PROT_READ */ |
1042 | if (current->personality & READ_IMPLIES_EXEC) { | |
b4caecd4 | 1043 | if (capabilities & NOMMU_MAP_EXEC) |
1da177e4 LT |
1044 | prot |= PROT_EXEC; |
1045 | } | |
ac714904 | 1046 | } else if ((prot & PROT_READ) && |
1da177e4 | 1047 | (prot & PROT_EXEC) && |
b4caecd4 | 1048 | !(capabilities & NOMMU_MAP_EXEC) |
1da177e4 LT |
1049 | ) { |
1050 | /* backing file is not executable, try to copy */ | |
b4caecd4 | 1051 | capabilities &= ~NOMMU_MAP_DIRECT; |
1da177e4 | 1052 | } |
ac714904 | 1053 | } else { |
1da177e4 LT |
1054 | /* anonymous mappings are always memory backed and can be |
1055 | * privately mapped | |
1056 | */ | |
b4caecd4 | 1057 | capabilities = NOMMU_MAP_COPY; |
1da177e4 LT |
1058 | |
1059 | /* handle PROT_EXEC implication by PROT_READ */ | |
1060 | if ((prot & PROT_READ) && | |
1061 | (current->personality & READ_IMPLIES_EXEC)) | |
1062 | prot |= PROT_EXEC; | |
1063 | } | |
1064 | ||
1065 | /* allow the security API to have its say */ | |
e5467859 | 1066 | ret = security_mmap_addr(addr); |
1da177e4 LT |
1067 | if (ret < 0) |
1068 | return ret; | |
1069 | ||
1070 | /* looks okay */ | |
1071 | *_capabilities = capabilities; | |
1072 | return 0; | |
1073 | } | |
1074 | ||
1075 | /* | |
1076 | * we've determined that we can make the mapping, now translate what we | |
1077 | * now know into VMA flags | |
1078 | */ | |
1079 | static unsigned long determine_vm_flags(struct file *file, | |
1080 | unsigned long prot, | |
1081 | unsigned long flags, | |
1082 | unsigned long capabilities) | |
1083 | { | |
1084 | unsigned long vm_flags; | |
1085 | ||
1086 | vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags); | |
1da177e4 LT |
1087 | /* vm_flags |= mm->def_flags; */ |
1088 | ||
b4caecd4 | 1089 | if (!(capabilities & NOMMU_MAP_DIRECT)) { |
1da177e4 | 1090 | /* attempt to share read-only copies of mapped file chunks */ |
3c7b2045 | 1091 | vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC; |
1da177e4 LT |
1092 | if (file && !(prot & PROT_WRITE)) |
1093 | vm_flags |= VM_MAYSHARE; | |
3c7b2045 | 1094 | } else { |
1da177e4 LT |
1095 | /* overlay a shareable mapping on the backing device or inode |
1096 | * if possible - used for chardevs, ramfs/tmpfs/shmfs and | |
1097 | * romfs/cramfs */ | |
b4caecd4 | 1098 | vm_flags |= VM_MAYSHARE | (capabilities & NOMMU_VMFLAGS); |
1da177e4 | 1099 | if (flags & MAP_SHARED) |
3c7b2045 | 1100 | vm_flags |= VM_SHARED; |
1da177e4 LT |
1101 | } |
1102 | ||
1103 | /* refuse to let anyone share private mappings with this process if | |
1104 | * it's being traced - otherwise breakpoints set in it may interfere | |
1105 | * with another untraced process | |
1106 | */ | |
a288eecc | 1107 | if ((flags & MAP_PRIVATE) && current->ptrace) |
1da177e4 LT |
1108 | vm_flags &= ~VM_MAYSHARE; |
1109 | ||
1110 | return vm_flags; | |
1111 | } | |
1112 | ||
1113 | /* | |
8feae131 DH |
1114 | * set up a shared mapping on a file (the driver or filesystem provides and |
1115 | * pins the storage) | |
1da177e4 | 1116 | */ |
8feae131 | 1117 | static int do_mmap_shared_file(struct vm_area_struct *vma) |
1da177e4 LT |
1118 | { |
1119 | int ret; | |
1120 | ||
1121 | ret = vma->vm_file->f_op->mmap(vma->vm_file, vma); | |
dd8632a1 PM |
1122 | if (ret == 0) { |
1123 | vma->vm_region->vm_top = vma->vm_region->vm_end; | |
645d83c5 | 1124 | return 0; |
dd8632a1 | 1125 | } |
1da177e4 LT |
1126 | if (ret != -ENOSYS) |
1127 | return ret; | |
1128 | ||
3fa30460 DH |
1129 | /* getting -ENOSYS indicates that direct mmap isn't possible (as |
1130 | * opposed to tried but failed) so we can only give a suitable error as | |
1131 | * it's not possible to make a private copy if MAP_SHARED was given */ | |
1da177e4 LT |
1132 | return -ENODEV; |
1133 | } | |
1134 | ||
1135 | /* | |
1136 | * set up a private mapping or an anonymous shared mapping | |
1137 | */ | |
8feae131 DH |
1138 | static int do_mmap_private(struct vm_area_struct *vma, |
1139 | struct vm_region *region, | |
645d83c5 DH |
1140 | unsigned long len, |
1141 | unsigned long capabilities) | |
1da177e4 | 1142 | { |
dbc8358c | 1143 | unsigned long total, point; |
1da177e4 | 1144 | void *base; |
8feae131 | 1145 | int ret, order; |
1da177e4 LT |
1146 | |
1147 | /* invoke the file's mapping function so that it can keep track of | |
1148 | * shared mappings on devices or memory | |
1149 | * - VM_MAYSHARE will be set if it may attempt to share | |
1150 | */ | |
b4caecd4 | 1151 | if (capabilities & NOMMU_MAP_DIRECT) { |
1da177e4 | 1152 | ret = vma->vm_file->f_op->mmap(vma->vm_file, vma); |
dd8632a1 | 1153 | if (ret == 0) { |
1da177e4 | 1154 | /* shouldn't return success if we're not sharing */ |
dd8632a1 PM |
1155 | BUG_ON(!(vma->vm_flags & VM_MAYSHARE)); |
1156 | vma->vm_region->vm_top = vma->vm_region->vm_end; | |
645d83c5 | 1157 | return 0; |
1da177e4 | 1158 | } |
dd8632a1 PM |
1159 | if (ret != -ENOSYS) |
1160 | return ret; | |
1da177e4 LT |
1161 | |
1162 | /* getting an ENOSYS error indicates that direct mmap isn't | |
1163 | * possible (as opposed to tried but failed) so we'll try to | |
1164 | * make a private copy of the data and map that instead */ | |
1165 | } | |
1166 | ||
8feae131 | 1167 | |
1da177e4 LT |
1168 | /* allocate some memory to hold the mapping |
1169 | * - note that this may not return a page-aligned address if the object | |
1170 | * we're allocating is smaller than a page | |
1171 | */ | |
f67d9b15 | 1172 | order = get_order(len); |
8feae131 DH |
1173 | kdebug("alloc order %d for %lx", order, len); |
1174 | ||
8feae131 | 1175 | total = 1 << order; |
f67d9b15 | 1176 | point = len >> PAGE_SHIFT; |
dd8632a1 | 1177 | |
dbc8358c | 1178 | /* we don't want to allocate a power-of-2 sized page set */ |
dd8632a1 | 1179 | if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages) { |
dbc8358c JK |
1180 | total = point; |
1181 | kdebug("try to alloc exact %lu pages", total); | |
1182 | base = alloc_pages_exact(len, GFP_KERNEL); | |
1183 | } else { | |
1184 | base = (void *)__get_free_pages(GFP_KERNEL, order); | |
8feae131 DH |
1185 | } |
1186 | ||
dbc8358c JK |
1187 | if (!base) |
1188 | goto enomem; | |
1189 | ||
1190 | atomic_long_add(total, &mmap_pages_allocated); | |
1da177e4 | 1191 | |
8feae131 DH |
1192 | region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY; |
1193 | region->vm_start = (unsigned long) base; | |
f67d9b15 | 1194 | region->vm_end = region->vm_start + len; |
dd8632a1 | 1195 | region->vm_top = region->vm_start + (total << PAGE_SHIFT); |
8feae131 DH |
1196 | |
1197 | vma->vm_start = region->vm_start; | |
1198 | vma->vm_end = region->vm_start + len; | |
1da177e4 LT |
1199 | |
1200 | if (vma->vm_file) { | |
1201 | /* read the contents of a file into the copy */ | |
1202 | mm_segment_t old_fs; | |
1203 | loff_t fpos; | |
1204 | ||
1205 | fpos = vma->vm_pgoff; | |
1206 | fpos <<= PAGE_SHIFT; | |
1207 | ||
1208 | old_fs = get_fs(); | |
1209 | set_fs(KERNEL_DS); | |
f67d9b15 | 1210 | ret = vma->vm_file->f_op->read(vma->vm_file, base, len, &fpos); |
1da177e4 LT |
1211 | set_fs(old_fs); |
1212 | ||
1213 | if (ret < 0) | |
1214 | goto error_free; | |
1215 | ||
1216 | /* clear the last little bit */ | |
f67d9b15 BL |
1217 | if (ret < len) |
1218 | memset(base + ret, 0, len - ret); | |
1da177e4 | 1219 | |
1da177e4 LT |
1220 | } |
1221 | ||
1222 | return 0; | |
1223 | ||
1224 | error_free: | |
7223bb4a | 1225 | free_page_series(region->vm_start, region->vm_top); |
8feae131 DH |
1226 | region->vm_start = vma->vm_start = 0; |
1227 | region->vm_end = vma->vm_end = 0; | |
dd8632a1 | 1228 | region->vm_top = 0; |
1da177e4 LT |
1229 | return ret; |
1230 | ||
1231 | enomem: | |
b1de0d13 | 1232 | pr_err("Allocation of length %lu from process %d (%s) failed\n", |
05ae6fa3 | 1233 | len, current->pid, current->comm); |
7bf02ea2 | 1234 | show_free_areas(0); |
1da177e4 LT |
1235 | return -ENOMEM; |
1236 | } | |
1237 | ||
1238 | /* | |
1239 | * handle mapping creation for uClinux | |
1240 | */ | |
e3fc629d | 1241 | unsigned long do_mmap_pgoff(struct file *file, |
1da177e4 LT |
1242 | unsigned long addr, |
1243 | unsigned long len, | |
1244 | unsigned long prot, | |
1245 | unsigned long flags, | |
bebeb3d6 | 1246 | unsigned long pgoff, |
41badc15 | 1247 | unsigned long *populate) |
1da177e4 | 1248 | { |
8feae131 DH |
1249 | struct vm_area_struct *vma; |
1250 | struct vm_region *region; | |
1da177e4 | 1251 | struct rb_node *rb; |
8feae131 | 1252 | unsigned long capabilities, vm_flags, result; |
1da177e4 LT |
1253 | int ret; |
1254 | ||
8feae131 DH |
1255 | kenter(",%lx,%lx,%lx,%lx,%lx", addr, len, prot, flags, pgoff); |
1256 | ||
41badc15 | 1257 | *populate = 0; |
bebeb3d6 | 1258 | |
1da177e4 LT |
1259 | /* decide whether we should attempt the mapping, and if so what sort of |
1260 | * mapping */ | |
1261 | ret = validate_mmap_request(file, addr, len, prot, flags, pgoff, | |
1262 | &capabilities); | |
8feae131 DH |
1263 | if (ret < 0) { |
1264 | kleave(" = %d [val]", ret); | |
1da177e4 | 1265 | return ret; |
8feae131 | 1266 | } |
1da177e4 | 1267 | |
06aab5a3 DH |
1268 | /* we ignore the address hint */ |
1269 | addr = 0; | |
f67d9b15 | 1270 | len = PAGE_ALIGN(len); |
06aab5a3 | 1271 | |
1da177e4 LT |
1272 | /* we've determined that we can make the mapping, now translate what we |
1273 | * now know into VMA flags */ | |
1274 | vm_flags = determine_vm_flags(file, prot, flags, capabilities); | |
1275 | ||
8feae131 DH |
1276 | /* we're going to need to record the mapping */ |
1277 | region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL); | |
1278 | if (!region) | |
1279 | goto error_getting_region; | |
1280 | ||
1281 | vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); | |
1282 | if (!vma) | |
1283 | goto error_getting_vma; | |
1da177e4 | 1284 | |
1e2ae599 | 1285 | region->vm_usage = 1; |
8feae131 DH |
1286 | region->vm_flags = vm_flags; |
1287 | region->vm_pgoff = pgoff; | |
1288 | ||
5beb4930 | 1289 | INIT_LIST_HEAD(&vma->anon_vma_chain); |
8feae131 DH |
1290 | vma->vm_flags = vm_flags; |
1291 | vma->vm_pgoff = pgoff; | |
1da177e4 | 1292 | |
8feae131 | 1293 | if (file) { |
cb0942b8 AV |
1294 | region->vm_file = get_file(file); |
1295 | vma->vm_file = get_file(file); | |
8feae131 DH |
1296 | } |
1297 | ||
1298 | down_write(&nommu_region_sem); | |
1299 | ||
1300 | /* if we want to share, we need to check for regions created by other | |
1da177e4 | 1301 | * mmap() calls that overlap with our proposed mapping |
8feae131 | 1302 | * - we can only share with a superset match on most regular files |
1da177e4 LT |
1303 | * - shared mappings on character devices and memory backed files are |
1304 | * permitted to overlap inexactly as far as we are concerned for in | |
1305 | * these cases, sharing is handled in the driver or filesystem rather | |
1306 | * than here | |
1307 | */ | |
1308 | if (vm_flags & VM_MAYSHARE) { | |
8feae131 DH |
1309 | struct vm_region *pregion; |
1310 | unsigned long pglen, rpglen, pgend, rpgend, start; | |
1da177e4 | 1311 | |
8feae131 DH |
1312 | pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; |
1313 | pgend = pgoff + pglen; | |
165b2392 | 1314 | |
8feae131 DH |
1315 | for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) { |
1316 | pregion = rb_entry(rb, struct vm_region, vm_rb); | |
1da177e4 | 1317 | |
8feae131 | 1318 | if (!(pregion->vm_flags & VM_MAYSHARE)) |
1da177e4 LT |
1319 | continue; |
1320 | ||
1321 | /* search for overlapping mappings on the same file */ | |
496ad9aa AV |
1322 | if (file_inode(pregion->vm_file) != |
1323 | file_inode(file)) | |
1da177e4 LT |
1324 | continue; |
1325 | ||
8feae131 | 1326 | if (pregion->vm_pgoff >= pgend) |
1da177e4 LT |
1327 | continue; |
1328 | ||
8feae131 DH |
1329 | rpglen = pregion->vm_end - pregion->vm_start; |
1330 | rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
1331 | rpgend = pregion->vm_pgoff + rpglen; | |
1332 | if (pgoff >= rpgend) | |
1da177e4 LT |
1333 | continue; |
1334 | ||
8feae131 DH |
1335 | /* handle inexactly overlapping matches between |
1336 | * mappings */ | |
1337 | if ((pregion->vm_pgoff != pgoff || rpglen != pglen) && | |
1338 | !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) { | |
1339 | /* new mapping is not a subset of the region */ | |
b4caecd4 | 1340 | if (!(capabilities & NOMMU_MAP_DIRECT)) |
1da177e4 LT |
1341 | goto sharing_violation; |
1342 | continue; | |
1343 | } | |
1344 | ||
8feae131 | 1345 | /* we've found a region we can share */ |
1e2ae599 | 1346 | pregion->vm_usage++; |
8feae131 DH |
1347 | vma->vm_region = pregion; |
1348 | start = pregion->vm_start; | |
1349 | start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT; | |
1350 | vma->vm_start = start; | |
1351 | vma->vm_end = start + len; | |
1352 | ||
1353 | if (pregion->vm_flags & VM_MAPPED_COPY) { | |
1354 | kdebug("share copy"); | |
1355 | vma->vm_flags |= VM_MAPPED_COPY; | |
1356 | } else { | |
1357 | kdebug("share mmap"); | |
1358 | ret = do_mmap_shared_file(vma); | |
1359 | if (ret < 0) { | |
1360 | vma->vm_region = NULL; | |
1361 | vma->vm_start = 0; | |
1362 | vma->vm_end = 0; | |
1e2ae599 | 1363 | pregion->vm_usage--; |
8feae131 DH |
1364 | pregion = NULL; |
1365 | goto error_just_free; | |
1366 | } | |
1367 | } | |
1368 | fput(region->vm_file); | |
1369 | kmem_cache_free(vm_region_jar, region); | |
1370 | region = pregion; | |
1371 | result = start; | |
1372 | goto share; | |
1da177e4 LT |
1373 | } |
1374 | ||
1da177e4 LT |
1375 | /* obtain the address at which to make a shared mapping |
1376 | * - this is the hook for quasi-memory character devices to | |
1377 | * tell us the location of a shared mapping | |
1378 | */ | |
b4caecd4 | 1379 | if (capabilities & NOMMU_MAP_DIRECT) { |
1da177e4 LT |
1380 | addr = file->f_op->get_unmapped_area(file, addr, len, |
1381 | pgoff, flags); | |
bb005a59 | 1382 | if (IS_ERR_VALUE(addr)) { |
1da177e4 | 1383 | ret = addr; |
bb005a59 | 1384 | if (ret != -ENOSYS) |
8feae131 | 1385 | goto error_just_free; |
1da177e4 LT |
1386 | |
1387 | /* the driver refused to tell us where to site | |
1388 | * the mapping so we'll have to attempt to copy | |
1389 | * it */ | |
bb005a59 | 1390 | ret = -ENODEV; |
b4caecd4 | 1391 | if (!(capabilities & NOMMU_MAP_COPY)) |
8feae131 | 1392 | goto error_just_free; |
1da177e4 | 1393 | |
b4caecd4 | 1394 | capabilities &= ~NOMMU_MAP_DIRECT; |
8feae131 DH |
1395 | } else { |
1396 | vma->vm_start = region->vm_start = addr; | |
1397 | vma->vm_end = region->vm_end = addr + len; | |
1da177e4 LT |
1398 | } |
1399 | } | |
1400 | } | |
1401 | ||
8feae131 | 1402 | vma->vm_region = region; |
1da177e4 | 1403 | |
645d83c5 | 1404 | /* set up the mapping |
b4caecd4 | 1405 | * - the region is filled in if NOMMU_MAP_DIRECT is still set |
645d83c5 | 1406 | */ |
1da177e4 | 1407 | if (file && vma->vm_flags & VM_SHARED) |
8feae131 | 1408 | ret = do_mmap_shared_file(vma); |
1da177e4 | 1409 | else |
645d83c5 | 1410 | ret = do_mmap_private(vma, region, len, capabilities); |
1da177e4 | 1411 | if (ret < 0) |
645d83c5 DH |
1412 | goto error_just_free; |
1413 | add_nommu_region(region); | |
8feae131 | 1414 | |
ea637639 JZ |
1415 | /* clear anonymous mappings that don't ask for uninitialized data */ |
1416 | if (!vma->vm_file && !(flags & MAP_UNINITIALIZED)) | |
1417 | memset((void *)region->vm_start, 0, | |
1418 | region->vm_end - region->vm_start); | |
1419 | ||
1da177e4 | 1420 | /* okay... we have a mapping; now we have to register it */ |
8feae131 | 1421 | result = vma->vm_start; |
1da177e4 | 1422 | |
1da177e4 LT |
1423 | current->mm->total_vm += len >> PAGE_SHIFT; |
1424 | ||
8feae131 DH |
1425 | share: |
1426 | add_vma_to_mm(current->mm, vma); | |
1da177e4 | 1427 | |
cfe79c00 MF |
1428 | /* we flush the region from the icache only when the first executable |
1429 | * mapping of it is made */ | |
1430 | if (vma->vm_flags & VM_EXEC && !region->vm_icache_flushed) { | |
1431 | flush_icache_range(region->vm_start, region->vm_end); | |
1432 | region->vm_icache_flushed = true; | |
1433 | } | |
1da177e4 | 1434 | |
cfe79c00 | 1435 | up_write(&nommu_region_sem); |
1da177e4 | 1436 | |
8feae131 DH |
1437 | kleave(" = %lx", result); |
1438 | return result; | |
1da177e4 | 1439 | |
8feae131 DH |
1440 | error_just_free: |
1441 | up_write(&nommu_region_sem); | |
1442 | error: | |
89a86402 DH |
1443 | if (region->vm_file) |
1444 | fput(region->vm_file); | |
8feae131 | 1445 | kmem_cache_free(vm_region_jar, region); |
89a86402 DH |
1446 | if (vma->vm_file) |
1447 | fput(vma->vm_file); | |
8feae131 DH |
1448 | kmem_cache_free(vm_area_cachep, vma); |
1449 | kleave(" = %d", ret); | |
1450 | return ret; | |
1451 | ||
1452 | sharing_violation: | |
1453 | up_write(&nommu_region_sem); | |
1454 | printk(KERN_WARNING "Attempt to share mismatched mappings\n"); | |
1455 | ret = -EINVAL; | |
1456 | goto error; | |
1da177e4 | 1457 | |
8feae131 DH |
1458 | error_getting_vma: |
1459 | kmem_cache_free(vm_region_jar, region); | |
1460 | printk(KERN_WARNING "Allocation of vma for %lu byte allocation" | |
1461 | " from process %d failed\n", | |
1da177e4 | 1462 | len, current->pid); |
7bf02ea2 | 1463 | show_free_areas(0); |
1da177e4 LT |
1464 | return -ENOMEM; |
1465 | ||
8feae131 DH |
1466 | error_getting_region: |
1467 | printk(KERN_WARNING "Allocation of vm region for %lu byte allocation" | |
1468 | " from process %d failed\n", | |
1da177e4 | 1469 | len, current->pid); |
7bf02ea2 | 1470 | show_free_areas(0); |
1da177e4 LT |
1471 | return -ENOMEM; |
1472 | } | |
6be5ceb0 | 1473 | |
66f0dc48 HD |
1474 | SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len, |
1475 | unsigned long, prot, unsigned long, flags, | |
1476 | unsigned long, fd, unsigned long, pgoff) | |
1477 | { | |
1478 | struct file *file = NULL; | |
1479 | unsigned long retval = -EBADF; | |
1480 | ||
120a795d | 1481 | audit_mmap_fd(fd, flags); |
66f0dc48 HD |
1482 | if (!(flags & MAP_ANONYMOUS)) { |
1483 | file = fget(fd); | |
1484 | if (!file) | |
1485 | goto out; | |
1486 | } | |
1487 | ||
1488 | flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE); | |
1489 | ||
ad1ed293 | 1490 | retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff); |
66f0dc48 HD |
1491 | |
1492 | if (file) | |
1493 | fput(file); | |
1494 | out: | |
1495 | return retval; | |
1496 | } | |
1497 | ||
a4679373 CH |
1498 | #ifdef __ARCH_WANT_SYS_OLD_MMAP |
1499 | struct mmap_arg_struct { | |
1500 | unsigned long addr; | |
1501 | unsigned long len; | |
1502 | unsigned long prot; | |
1503 | unsigned long flags; | |
1504 | unsigned long fd; | |
1505 | unsigned long offset; | |
1506 | }; | |
1507 | ||
1508 | SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg) | |
1509 | { | |
1510 | struct mmap_arg_struct a; | |
1511 | ||
1512 | if (copy_from_user(&a, arg, sizeof(a))) | |
1513 | return -EFAULT; | |
1514 | if (a.offset & ~PAGE_MASK) | |
1515 | return -EINVAL; | |
1516 | ||
1517 | return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd, | |
1518 | a.offset >> PAGE_SHIFT); | |
1519 | } | |
1520 | #endif /* __ARCH_WANT_SYS_OLD_MMAP */ | |
1521 | ||
1da177e4 | 1522 | /* |
8feae131 DH |
1523 | * split a vma into two pieces at address 'addr', a new vma is allocated either |
1524 | * for the first part or the tail. | |
1da177e4 | 1525 | */ |
8feae131 DH |
1526 | int split_vma(struct mm_struct *mm, struct vm_area_struct *vma, |
1527 | unsigned long addr, int new_below) | |
1da177e4 | 1528 | { |
8feae131 DH |
1529 | struct vm_area_struct *new; |
1530 | struct vm_region *region; | |
1531 | unsigned long npages; | |
1da177e4 | 1532 | |
8feae131 | 1533 | kenter(""); |
1da177e4 | 1534 | |
779c1023 DH |
1535 | /* we're only permitted to split anonymous regions (these should have |
1536 | * only a single usage on the region) */ | |
1537 | if (vma->vm_file) | |
8feae131 | 1538 | return -ENOMEM; |
1da177e4 | 1539 | |
8feae131 DH |
1540 | if (mm->map_count >= sysctl_max_map_count) |
1541 | return -ENOMEM; | |
1da177e4 | 1542 | |
8feae131 DH |
1543 | region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL); |
1544 | if (!region) | |
1545 | return -ENOMEM; | |
1da177e4 | 1546 | |
8feae131 DH |
1547 | new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); |
1548 | if (!new) { | |
1549 | kmem_cache_free(vm_region_jar, region); | |
1550 | return -ENOMEM; | |
1551 | } | |
1552 | ||
1553 | /* most fields are the same, copy all, and then fixup */ | |
1554 | *new = *vma; | |
1555 | *region = *vma->vm_region; | |
1556 | new->vm_region = region; | |
1557 | ||
1558 | npages = (addr - vma->vm_start) >> PAGE_SHIFT; | |
1559 | ||
1560 | if (new_below) { | |
dd8632a1 | 1561 | region->vm_top = region->vm_end = new->vm_end = addr; |
8feae131 DH |
1562 | } else { |
1563 | region->vm_start = new->vm_start = addr; | |
1564 | region->vm_pgoff = new->vm_pgoff += npages; | |
1da177e4 | 1565 | } |
8feae131 DH |
1566 | |
1567 | if (new->vm_ops && new->vm_ops->open) | |
1568 | new->vm_ops->open(new); | |
1569 | ||
1570 | delete_vma_from_mm(vma); | |
1571 | down_write(&nommu_region_sem); | |
1572 | delete_nommu_region(vma->vm_region); | |
1573 | if (new_below) { | |
1574 | vma->vm_region->vm_start = vma->vm_start = addr; | |
1575 | vma->vm_region->vm_pgoff = vma->vm_pgoff += npages; | |
1576 | } else { | |
1577 | vma->vm_region->vm_end = vma->vm_end = addr; | |
dd8632a1 | 1578 | vma->vm_region->vm_top = addr; |
8feae131 DH |
1579 | } |
1580 | add_nommu_region(vma->vm_region); | |
1581 | add_nommu_region(new->vm_region); | |
1582 | up_write(&nommu_region_sem); | |
1583 | add_vma_to_mm(mm, vma); | |
1584 | add_vma_to_mm(mm, new); | |
1585 | return 0; | |
1da177e4 LT |
1586 | } |
1587 | ||
3034097a | 1588 | /* |
8feae131 DH |
1589 | * shrink a VMA by removing the specified chunk from either the beginning or |
1590 | * the end | |
3034097a | 1591 | */ |
8feae131 DH |
1592 | static int shrink_vma(struct mm_struct *mm, |
1593 | struct vm_area_struct *vma, | |
1594 | unsigned long from, unsigned long to) | |
1da177e4 | 1595 | { |
8feae131 | 1596 | struct vm_region *region; |
1da177e4 | 1597 | |
8feae131 | 1598 | kenter(""); |
1da177e4 | 1599 | |
8feae131 DH |
1600 | /* adjust the VMA's pointers, which may reposition it in the MM's tree |
1601 | * and list */ | |
1602 | delete_vma_from_mm(vma); | |
1603 | if (from > vma->vm_start) | |
1604 | vma->vm_end = from; | |
1605 | else | |
1606 | vma->vm_start = to; | |
1607 | add_vma_to_mm(mm, vma); | |
1da177e4 | 1608 | |
8feae131 DH |
1609 | /* cut the backing region down to size */ |
1610 | region = vma->vm_region; | |
1e2ae599 | 1611 | BUG_ON(region->vm_usage != 1); |
8feae131 DH |
1612 | |
1613 | down_write(&nommu_region_sem); | |
1614 | delete_nommu_region(region); | |
dd8632a1 PM |
1615 | if (from > region->vm_start) { |
1616 | to = region->vm_top; | |
1617 | region->vm_top = region->vm_end = from; | |
1618 | } else { | |
8feae131 | 1619 | region->vm_start = to; |
dd8632a1 | 1620 | } |
8feae131 DH |
1621 | add_nommu_region(region); |
1622 | up_write(&nommu_region_sem); | |
1623 | ||
1624 | free_page_series(from, to); | |
1625 | return 0; | |
1626 | } | |
1da177e4 | 1627 | |
8feae131 DH |
1628 | /* |
1629 | * release a mapping | |
1630 | * - under NOMMU conditions the chunk to be unmapped must be backed by a single | |
1631 | * VMA, though it need not cover the whole VMA | |
1632 | */ | |
1633 | int do_munmap(struct mm_struct *mm, unsigned long start, size_t len) | |
1634 | { | |
1635 | struct vm_area_struct *vma; | |
f67d9b15 | 1636 | unsigned long end; |
8feae131 | 1637 | int ret; |
1da177e4 | 1638 | |
8feae131 | 1639 | kenter(",%lx,%zx", start, len); |
1da177e4 | 1640 | |
f67d9b15 | 1641 | len = PAGE_ALIGN(len); |
8feae131 DH |
1642 | if (len == 0) |
1643 | return -EINVAL; | |
365e9c87 | 1644 | |
f67d9b15 BL |
1645 | end = start + len; |
1646 | ||
8feae131 DH |
1647 | /* find the first potentially overlapping VMA */ |
1648 | vma = find_vma(mm, start); | |
1649 | if (!vma) { | |
ac714904 | 1650 | static int limit; |
33e5d769 DH |
1651 | if (limit < 5) { |
1652 | printk(KERN_WARNING | |
1653 | "munmap of memory not mmapped by process %d" | |
1654 | " (%s): 0x%lx-0x%lx\n", | |
1655 | current->pid, current->comm, | |
1656 | start, start + len - 1); | |
1657 | limit++; | |
1658 | } | |
8feae131 DH |
1659 | return -EINVAL; |
1660 | } | |
1da177e4 | 1661 | |
8feae131 DH |
1662 | /* we're allowed to split an anonymous VMA but not a file-backed one */ |
1663 | if (vma->vm_file) { | |
1664 | do { | |
1665 | if (start > vma->vm_start) { | |
1666 | kleave(" = -EINVAL [miss]"); | |
1667 | return -EINVAL; | |
1668 | } | |
1669 | if (end == vma->vm_end) | |
1670 | goto erase_whole_vma; | |
d75a310c NK |
1671 | vma = vma->vm_next; |
1672 | } while (vma); | |
8feae131 DH |
1673 | kleave(" = -EINVAL [split file]"); |
1674 | return -EINVAL; | |
1675 | } else { | |
1676 | /* the chunk must be a subset of the VMA found */ | |
1677 | if (start == vma->vm_start && end == vma->vm_end) | |
1678 | goto erase_whole_vma; | |
1679 | if (start < vma->vm_start || end > vma->vm_end) { | |
1680 | kleave(" = -EINVAL [superset]"); | |
1681 | return -EINVAL; | |
1682 | } | |
1683 | if (start & ~PAGE_MASK) { | |
1684 | kleave(" = -EINVAL [unaligned start]"); | |
1685 | return -EINVAL; | |
1686 | } | |
1687 | if (end != vma->vm_end && end & ~PAGE_MASK) { | |
1688 | kleave(" = -EINVAL [unaligned split]"); | |
1689 | return -EINVAL; | |
1690 | } | |
1691 | if (start != vma->vm_start && end != vma->vm_end) { | |
1692 | ret = split_vma(mm, vma, start, 1); | |
1693 | if (ret < 0) { | |
1694 | kleave(" = %d [split]", ret); | |
1695 | return ret; | |
1696 | } | |
1697 | } | |
1698 | return shrink_vma(mm, vma, start, end); | |
1699 | } | |
1da177e4 | 1700 | |
8feae131 DH |
1701 | erase_whole_vma: |
1702 | delete_vma_from_mm(vma); | |
1703 | delete_vma(mm, vma); | |
1704 | kleave(" = 0"); | |
1da177e4 LT |
1705 | return 0; |
1706 | } | |
b5073173 | 1707 | EXPORT_SYMBOL(do_munmap); |
1da177e4 | 1708 | |
bfce281c | 1709 | int vm_munmap(unsigned long addr, size_t len) |
3034097a | 1710 | { |
bfce281c | 1711 | struct mm_struct *mm = current->mm; |
3034097a | 1712 | int ret; |
3034097a DH |
1713 | |
1714 | down_write(&mm->mmap_sem); | |
1715 | ret = do_munmap(mm, addr, len); | |
1716 | up_write(&mm->mmap_sem); | |
1717 | return ret; | |
1718 | } | |
a46ef99d LT |
1719 | EXPORT_SYMBOL(vm_munmap); |
1720 | ||
1721 | SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len) | |
1722 | { | |
bfce281c | 1723 | return vm_munmap(addr, len); |
a46ef99d | 1724 | } |
3034097a DH |
1725 | |
1726 | /* | |
8feae131 | 1727 | * release all the mappings made in a process's VM space |
3034097a | 1728 | */ |
8feae131 | 1729 | void exit_mmap(struct mm_struct *mm) |
1da177e4 | 1730 | { |
8feae131 | 1731 | struct vm_area_struct *vma; |
1da177e4 | 1732 | |
8feae131 DH |
1733 | if (!mm) |
1734 | return; | |
1da177e4 | 1735 | |
8feae131 | 1736 | kenter(""); |
1da177e4 | 1737 | |
8feae131 | 1738 | mm->total_vm = 0; |
1da177e4 | 1739 | |
8feae131 DH |
1740 | while ((vma = mm->mmap)) { |
1741 | mm->mmap = vma->vm_next; | |
1742 | delete_vma_from_mm(vma); | |
1743 | delete_vma(mm, vma); | |
04c34961 | 1744 | cond_resched(); |
1da177e4 | 1745 | } |
8feae131 DH |
1746 | |
1747 | kleave(""); | |
1da177e4 LT |
1748 | } |
1749 | ||
e4eb1ff6 | 1750 | unsigned long vm_brk(unsigned long addr, unsigned long len) |
1da177e4 LT |
1751 | { |
1752 | return -ENOMEM; | |
1753 | } | |
1754 | ||
1755 | /* | |
6fa5f80b DH |
1756 | * expand (or shrink) an existing mapping, potentially moving it at the same |
1757 | * time (controlled by the MREMAP_MAYMOVE flag and available VM space) | |
1da177e4 | 1758 | * |
6fa5f80b | 1759 | * under NOMMU conditions, we only permit changing a mapping's size, and only |
8feae131 DH |
1760 | * as long as it stays within the region allocated by do_mmap_private() and the |
1761 | * block is not shareable | |
1da177e4 | 1762 | * |
6fa5f80b | 1763 | * MREMAP_FIXED is not supported under NOMMU conditions |
1da177e4 | 1764 | */ |
4b377bab | 1765 | static unsigned long do_mremap(unsigned long addr, |
1da177e4 LT |
1766 | unsigned long old_len, unsigned long new_len, |
1767 | unsigned long flags, unsigned long new_addr) | |
1768 | { | |
6fa5f80b | 1769 | struct vm_area_struct *vma; |
1da177e4 LT |
1770 | |
1771 | /* insanity checks first */ | |
f67d9b15 BL |
1772 | old_len = PAGE_ALIGN(old_len); |
1773 | new_len = PAGE_ALIGN(new_len); | |
8feae131 | 1774 | if (old_len == 0 || new_len == 0) |
1da177e4 LT |
1775 | return (unsigned long) -EINVAL; |
1776 | ||
8feae131 DH |
1777 | if (addr & ~PAGE_MASK) |
1778 | return -EINVAL; | |
1779 | ||
1da177e4 LT |
1780 | if (flags & MREMAP_FIXED && new_addr != addr) |
1781 | return (unsigned long) -EINVAL; | |
1782 | ||
8feae131 | 1783 | vma = find_vma_exact(current->mm, addr, old_len); |
6fa5f80b DH |
1784 | if (!vma) |
1785 | return (unsigned long) -EINVAL; | |
1da177e4 | 1786 | |
6fa5f80b | 1787 | if (vma->vm_end != vma->vm_start + old_len) |
1da177e4 LT |
1788 | return (unsigned long) -EFAULT; |
1789 | ||
6fa5f80b | 1790 | if (vma->vm_flags & VM_MAYSHARE) |
1da177e4 LT |
1791 | return (unsigned long) -EPERM; |
1792 | ||
8feae131 | 1793 | if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start) |
1da177e4 LT |
1794 | return (unsigned long) -ENOMEM; |
1795 | ||
1796 | /* all checks complete - do it */ | |
6fa5f80b | 1797 | vma->vm_end = vma->vm_start + new_len; |
6fa5f80b DH |
1798 | return vma->vm_start; |
1799 | } | |
1800 | ||
6a6160a7 HC |
1801 | SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len, |
1802 | unsigned long, new_len, unsigned long, flags, | |
1803 | unsigned long, new_addr) | |
6fa5f80b DH |
1804 | { |
1805 | unsigned long ret; | |
1806 | ||
1807 | down_write(¤t->mm->mmap_sem); | |
1808 | ret = do_mremap(addr, old_len, new_len, flags, new_addr); | |
1809 | up_write(¤t->mm->mmap_sem); | |
1810 | return ret; | |
1da177e4 LT |
1811 | } |
1812 | ||
240aadee ML |
1813 | struct page *follow_page_mask(struct vm_area_struct *vma, |
1814 | unsigned long address, unsigned int flags, | |
1815 | unsigned int *page_mask) | |
1da177e4 | 1816 | { |
240aadee | 1817 | *page_mask = 0; |
1da177e4 LT |
1818 | return NULL; |
1819 | } | |
1820 | ||
8f3b1327 BL |
1821 | int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr, |
1822 | unsigned long pfn, unsigned long size, pgprot_t prot) | |
1da177e4 | 1823 | { |
8f3b1327 BL |
1824 | if (addr != (pfn << PAGE_SHIFT)) |
1825 | return -EINVAL; | |
1826 | ||
314e51b9 | 1827 | vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP; |
66aa2b4b | 1828 | return 0; |
1da177e4 | 1829 | } |
22c4af40 | 1830 | EXPORT_SYMBOL(remap_pfn_range); |
1da177e4 | 1831 | |
3c0b9de6 LT |
1832 | int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len) |
1833 | { | |
1834 | unsigned long pfn = start >> PAGE_SHIFT; | |
1835 | unsigned long vm_len = vma->vm_end - vma->vm_start; | |
1836 | ||
1837 | pfn += vma->vm_pgoff; | |
1838 | return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot); | |
1839 | } | |
1840 | EXPORT_SYMBOL(vm_iomap_memory); | |
1841 | ||
f905bc44 PM |
1842 | int remap_vmalloc_range(struct vm_area_struct *vma, void *addr, |
1843 | unsigned long pgoff) | |
1844 | { | |
1845 | unsigned int size = vma->vm_end - vma->vm_start; | |
1846 | ||
1847 | if (!(vma->vm_flags & VM_USERMAP)) | |
1848 | return -EINVAL; | |
1849 | ||
1850 | vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT)); | |
1851 | vma->vm_end = vma->vm_start + size; | |
1852 | ||
1853 | return 0; | |
1854 | } | |
1855 | EXPORT_SYMBOL(remap_vmalloc_range); | |
1856 | ||
1da177e4 LT |
1857 | unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr, |
1858 | unsigned long len, unsigned long pgoff, unsigned long flags) | |
1859 | { | |
1860 | return -ENOMEM; | |
1861 | } | |
1862 | ||
1da177e4 LT |
1863 | void unmap_mapping_range(struct address_space *mapping, |
1864 | loff_t const holebegin, loff_t const holelen, | |
1865 | int even_cows) | |
1866 | { | |
1867 | } | |
22c4af40 | 1868 | EXPORT_SYMBOL(unmap_mapping_range); |
1da177e4 LT |
1869 | |
1870 | /* | |
1871 | * Check that a process has enough memory to allocate a new virtual | |
1872 | * mapping. 0 means there is enough memory for the allocation to | |
1873 | * succeed and -ENOMEM implies there is not. | |
1874 | * | |
1875 | * We currently support three overcommit policies, which are set via the | |
1876 | * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting | |
1877 | * | |
1878 | * Strict overcommit modes added 2002 Feb 26 by Alan Cox. | |
1879 | * Additional code 2002 Jul 20 by Robert Love. | |
1880 | * | |
1881 | * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise. | |
1882 | * | |
1883 | * Note this is a helper function intended to be used by LSMs which | |
1884 | * wish to use this logic. | |
1885 | */ | |
34b4e4aa | 1886 | int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin) |
1da177e4 | 1887 | { |
c9b1d098 | 1888 | unsigned long free, allowed, reserve; |
1da177e4 LT |
1889 | |
1890 | vm_acct_memory(pages); | |
1891 | ||
1892 | /* | |
1893 | * Sometimes we want to use more memory than we have | |
1894 | */ | |
1895 | if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS) | |
1896 | return 0; | |
1897 | ||
1898 | if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) { | |
c15bef30 DF |
1899 | free = global_page_state(NR_FREE_PAGES); |
1900 | free += global_page_state(NR_FILE_PAGES); | |
1901 | ||
1902 | /* | |
1903 | * shmem pages shouldn't be counted as free in this | |
1904 | * case, they can't be purged, only swapped out, and | |
1905 | * that won't affect the overall amount of available | |
1906 | * memory in the system. | |
1907 | */ | |
1908 | free -= global_page_state(NR_SHMEM); | |
1da177e4 | 1909 | |
ec8acf20 | 1910 | free += get_nr_swap_pages(); |
1da177e4 LT |
1911 | |
1912 | /* | |
1913 | * Any slabs which are created with the | |
1914 | * SLAB_RECLAIM_ACCOUNT flag claim to have contents | |
1915 | * which are reclaimable, under pressure. The dentry | |
1916 | * cache and most inode caches should fall into this | |
1917 | */ | |
972d1a7b | 1918 | free += global_page_state(NR_SLAB_RECLAIMABLE); |
1da177e4 | 1919 | |
d5ddc79b HA |
1920 | /* |
1921 | * Leave reserved pages. The pages are not for anonymous pages. | |
1922 | */ | |
c15bef30 | 1923 | if (free <= totalreserve_pages) |
d5ddc79b HA |
1924 | goto error; |
1925 | else | |
c15bef30 | 1926 | free -= totalreserve_pages; |
d5ddc79b HA |
1927 | |
1928 | /* | |
4eeab4f5 | 1929 | * Reserve some for root |
d5ddc79b | 1930 | */ |
1da177e4 | 1931 | if (!cap_sys_admin) |
4eeab4f5 | 1932 | free -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10); |
1da177e4 LT |
1933 | |
1934 | if (free > pages) | |
1935 | return 0; | |
d5ddc79b HA |
1936 | |
1937 | goto error; | |
1da177e4 LT |
1938 | } |
1939 | ||
00619bcc | 1940 | allowed = vm_commit_limit(); |
1da177e4 | 1941 | /* |
4eeab4f5 | 1942 | * Reserve some 3% for root |
1da177e4 LT |
1943 | */ |
1944 | if (!cap_sys_admin) | |
4eeab4f5 | 1945 | allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10); |
1da177e4 | 1946 | |
c9b1d098 AS |
1947 | /* |
1948 | * Don't let a single process grow so big a user can't recover | |
1949 | */ | |
1950 | if (mm) { | |
1951 | reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10); | |
1952 | allowed -= min(mm->total_vm / 32, reserve); | |
1953 | } | |
1da177e4 | 1954 | |
00a62ce9 | 1955 | if (percpu_counter_read_positive(&vm_committed_as) < allowed) |
1da177e4 | 1956 | return 0; |
00a62ce9 | 1957 | |
d5ddc79b | 1958 | error: |
1da177e4 LT |
1959 | vm_unacct_memory(pages); |
1960 | ||
1961 | return -ENOMEM; | |
1962 | } | |
1963 | ||
d0217ac0 | 1964 | int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf) |
b0e15190 DH |
1965 | { |
1966 | BUG(); | |
d0217ac0 | 1967 | return 0; |
b0e15190 | 1968 | } |
b5073173 | 1969 | EXPORT_SYMBOL(filemap_fault); |
0ec76a11 | 1970 | |
f1820361 KS |
1971 | void filemap_map_pages(struct vm_area_struct *vma, struct vm_fault *vmf) |
1972 | { | |
1973 | BUG(); | |
1974 | } | |
1975 | EXPORT_SYMBOL(filemap_map_pages); | |
1976 | ||
0b173bc4 KK |
1977 | int generic_file_remap_pages(struct vm_area_struct *vma, unsigned long addr, |
1978 | unsigned long size, pgoff_t pgoff) | |
1979 | { | |
1980 | BUG(); | |
1981 | return 0; | |
1982 | } | |
1983 | EXPORT_SYMBOL(generic_file_remap_pages); | |
1984 | ||
f55f199b MF |
1985 | static int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm, |
1986 | unsigned long addr, void *buf, int len, int write) | |
0ec76a11 | 1987 | { |
0ec76a11 | 1988 | struct vm_area_struct *vma; |
0ec76a11 DH |
1989 | |
1990 | down_read(&mm->mmap_sem); | |
1991 | ||
1992 | /* the access must start within one of the target process's mappings */ | |
0159b141 DH |
1993 | vma = find_vma(mm, addr); |
1994 | if (vma) { | |
0ec76a11 DH |
1995 | /* don't overrun this mapping */ |
1996 | if (addr + len >= vma->vm_end) | |
1997 | len = vma->vm_end - addr; | |
1998 | ||
1999 | /* only read or write mappings where it is permitted */ | |
d00c7b99 | 2000 | if (write && vma->vm_flags & VM_MAYWRITE) |
7959722b JZ |
2001 | copy_to_user_page(vma, NULL, addr, |
2002 | (void *) addr, buf, len); | |
d00c7b99 | 2003 | else if (!write && vma->vm_flags & VM_MAYREAD) |
7959722b JZ |
2004 | copy_from_user_page(vma, NULL, addr, |
2005 | buf, (void *) addr, len); | |
0ec76a11 DH |
2006 | else |
2007 | len = 0; | |
2008 | } else { | |
2009 | len = 0; | |
2010 | } | |
2011 | ||
2012 | up_read(&mm->mmap_sem); | |
f55f199b MF |
2013 | |
2014 | return len; | |
2015 | } | |
2016 | ||
2017 | /** | |
2018 | * @access_remote_vm - access another process' address space | |
2019 | * @mm: the mm_struct of the target address space | |
2020 | * @addr: start address to access | |
2021 | * @buf: source or destination buffer | |
2022 | * @len: number of bytes to transfer | |
2023 | * @write: whether the access is a write | |
2024 | * | |
2025 | * The caller must hold a reference on @mm. | |
2026 | */ | |
2027 | int access_remote_vm(struct mm_struct *mm, unsigned long addr, | |
2028 | void *buf, int len, int write) | |
2029 | { | |
2030 | return __access_remote_vm(NULL, mm, addr, buf, len, write); | |
2031 | } | |
2032 | ||
2033 | /* | |
2034 | * Access another process' address space. | |
2035 | * - source/target buffer must be kernel space | |
2036 | */ | |
2037 | int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write) | |
2038 | { | |
2039 | struct mm_struct *mm; | |
2040 | ||
2041 | if (addr + len < addr) | |
2042 | return 0; | |
2043 | ||
2044 | mm = get_task_mm(tsk); | |
2045 | if (!mm) | |
2046 | return 0; | |
2047 | ||
2048 | len = __access_remote_vm(tsk, mm, addr, buf, len, write); | |
2049 | ||
0ec76a11 DH |
2050 | mmput(mm); |
2051 | return len; | |
2052 | } | |
7e660872 DH |
2053 | |
2054 | /** | |
2055 | * nommu_shrink_inode_mappings - Shrink the shared mappings on an inode | |
2056 | * @inode: The inode to check | |
2057 | * @size: The current filesize of the inode | |
2058 | * @newsize: The proposed filesize of the inode | |
2059 | * | |
2060 | * Check the shared mappings on an inode on behalf of a shrinking truncate to | |
2061 | * make sure that that any outstanding VMAs aren't broken and then shrink the | |
2062 | * vm_regions that extend that beyond so that do_mmap_pgoff() doesn't | |
2063 | * automatically grant mappings that are too large. | |
2064 | */ | |
2065 | int nommu_shrink_inode_mappings(struct inode *inode, size_t size, | |
2066 | size_t newsize) | |
2067 | { | |
2068 | struct vm_area_struct *vma; | |
7e660872 DH |
2069 | struct vm_region *region; |
2070 | pgoff_t low, high; | |
2071 | size_t r_size, r_top; | |
2072 | ||
2073 | low = newsize >> PAGE_SHIFT; | |
2074 | high = (size + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
2075 | ||
2076 | down_write(&nommu_region_sem); | |
1acf2e04 | 2077 | i_mmap_lock_read(inode->i_mapping); |
7e660872 DH |
2078 | |
2079 | /* search for VMAs that fall within the dead zone */ | |
6b2dbba8 | 2080 | vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, low, high) { |
7e660872 DH |
2081 | /* found one - only interested if it's shared out of the page |
2082 | * cache */ | |
2083 | if (vma->vm_flags & VM_SHARED) { | |
1acf2e04 | 2084 | i_mmap_unlock_read(inode->i_mapping); |
7e660872 DH |
2085 | up_write(&nommu_region_sem); |
2086 | return -ETXTBSY; /* not quite true, but near enough */ | |
2087 | } | |
2088 | } | |
2089 | ||
2090 | /* reduce any regions that overlap the dead zone - if in existence, | |
2091 | * these will be pointed to by VMAs that don't overlap the dead zone | |
2092 | * | |
2093 | * we don't check for any regions that start beyond the EOF as there | |
2094 | * shouldn't be any | |
2095 | */ | |
1acf2e04 | 2096 | vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, 0, ULONG_MAX) { |
7e660872 DH |
2097 | if (!(vma->vm_flags & VM_SHARED)) |
2098 | continue; | |
2099 | ||
2100 | region = vma->vm_region; | |
2101 | r_size = region->vm_top - region->vm_start; | |
2102 | r_top = (region->vm_pgoff << PAGE_SHIFT) + r_size; | |
2103 | ||
2104 | if (r_top > newsize) { | |
2105 | region->vm_top -= r_top - newsize; | |
2106 | if (region->vm_end > region->vm_top) | |
2107 | region->vm_end = region->vm_top; | |
2108 | } | |
2109 | } | |
2110 | ||
1acf2e04 | 2111 | i_mmap_unlock_read(inode->i_mapping); |
7e660872 DH |
2112 | up_write(&nommu_region_sem); |
2113 | return 0; | |
2114 | } | |
c9b1d098 AS |
2115 | |
2116 | /* | |
2117 | * Initialise sysctl_user_reserve_kbytes. | |
2118 | * | |
2119 | * This is intended to prevent a user from starting a single memory hogging | |
2120 | * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER | |
2121 | * mode. | |
2122 | * | |
2123 | * The default value is min(3% of free memory, 128MB) | |
2124 | * 128MB is enough to recover with sshd/login, bash, and top/kill. | |
2125 | */ | |
2126 | static int __meminit init_user_reserve(void) | |
2127 | { | |
2128 | unsigned long free_kbytes; | |
2129 | ||
2130 | free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); | |
2131 | ||
2132 | sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17); | |
2133 | return 0; | |
2134 | } | |
2135 | module_init(init_user_reserve) | |
4eeab4f5 AS |
2136 | |
2137 | /* | |
2138 | * Initialise sysctl_admin_reserve_kbytes. | |
2139 | * | |
2140 | * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin | |
2141 | * to log in and kill a memory hogging process. | |
2142 | * | |
2143 | * Systems with more than 256MB will reserve 8MB, enough to recover | |
2144 | * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will | |
2145 | * only reserve 3% of free pages by default. | |
2146 | */ | |
2147 | static int __meminit init_admin_reserve(void) | |
2148 | { | |
2149 | unsigned long free_kbytes; | |
2150 | ||
2151 | free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); | |
2152 | ||
2153 | sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13); | |
2154 | return 0; | |
2155 | } | |
2156 | module_init(init_admin_reserve) |