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; | |
944b6874 | 62 | EXPORT_SYMBOL(high_memory); |
1da177e4 LT |
63 | struct page *mem_map; |
64 | unsigned long max_mapnr; | |
4266c97a | 65 | unsigned long highest_memmap_pfn; |
00a62ce9 | 66 | struct percpu_counter vm_committed_as; |
1da177e4 LT |
67 | int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */ |
68 | int sysctl_overcommit_ratio = 50; /* default is 50% */ | |
49f0ce5f | 69 | unsigned long sysctl_overcommit_kbytes __read_mostly; |
1da177e4 | 70 | int sysctl_max_map_count = DEFAULT_MAX_MAP_COUNT; |
fc4d5c29 | 71 | int sysctl_nr_trim_pages = CONFIG_NOMMU_INITIAL_TRIM_EXCESS; |
c9b1d098 | 72 | unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */ |
4eeab4f5 | 73 | unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */ |
1da177e4 LT |
74 | int heap_stack_gap = 0; |
75 | ||
33e5d769 | 76 | atomic_long_t mmap_pages_allocated; |
8feae131 | 77 | |
997071bc S |
78 | /* |
79 | * The global memory commitment made in the system can be a metric | |
80 | * that can be used to drive ballooning decisions when Linux is hosted | |
81 | * as a guest. On Hyper-V, the host implements a policy engine for dynamically | |
82 | * balancing memory across competing virtual machines that are hosted. | |
83 | * Several metrics drive this policy engine including the guest reported | |
84 | * memory commitment. | |
85 | */ | |
86 | unsigned long vm_memory_committed(void) | |
87 | { | |
88 | return percpu_counter_read_positive(&vm_committed_as); | |
89 | } | |
90 | ||
91 | EXPORT_SYMBOL_GPL(vm_memory_committed); | |
92 | ||
1da177e4 | 93 | EXPORT_SYMBOL(mem_map); |
1da177e4 | 94 | |
8feae131 DH |
95 | /* list of mapped, potentially shareable regions */ |
96 | static struct kmem_cache *vm_region_jar; | |
97 | struct rb_root nommu_region_tree = RB_ROOT; | |
98 | DECLARE_RWSEM(nommu_region_sem); | |
1da177e4 | 99 | |
f0f37e2f | 100 | const struct vm_operations_struct generic_file_vm_ops = { |
1da177e4 LT |
101 | }; |
102 | ||
1da177e4 LT |
103 | /* |
104 | * Return the total memory allocated for this pointer, not | |
105 | * just what the caller asked for. | |
106 | * | |
107 | * Doesn't have to be accurate, i.e. may have races. | |
108 | */ | |
109 | unsigned int kobjsize(const void *objp) | |
110 | { | |
111 | struct page *page; | |
112 | ||
4016a139 MH |
113 | /* |
114 | * If the object we have should not have ksize performed on it, | |
115 | * return size of 0 | |
116 | */ | |
5a1603be | 117 | if (!objp || !virt_addr_valid(objp)) |
6cfd53fc PM |
118 | return 0; |
119 | ||
120 | page = virt_to_head_page(objp); | |
6cfd53fc PM |
121 | |
122 | /* | |
123 | * If the allocator sets PageSlab, we know the pointer came from | |
124 | * kmalloc(). | |
125 | */ | |
1da177e4 LT |
126 | if (PageSlab(page)) |
127 | return ksize(objp); | |
128 | ||
ab2e83ea PM |
129 | /* |
130 | * If it's not a compound page, see if we have a matching VMA | |
131 | * region. This test is intentionally done in reverse order, | |
132 | * so if there's no VMA, we still fall through and hand back | |
133 | * PAGE_SIZE for 0-order pages. | |
134 | */ | |
135 | if (!PageCompound(page)) { | |
136 | struct vm_area_struct *vma; | |
137 | ||
138 | vma = find_vma(current->mm, (unsigned long)objp); | |
139 | if (vma) | |
140 | return vma->vm_end - vma->vm_start; | |
141 | } | |
142 | ||
6cfd53fc PM |
143 | /* |
144 | * The ksize() function is only guaranteed to work for pointers | |
5a1603be | 145 | * returned by kmalloc(). So handle arbitrary pointers here. |
6cfd53fc | 146 | */ |
5a1603be | 147 | return PAGE_SIZE << compound_order(page); |
1da177e4 LT |
148 | } |
149 | ||
28a35716 ML |
150 | long __get_user_pages(struct task_struct *tsk, struct mm_struct *mm, |
151 | unsigned long start, unsigned long nr_pages, | |
152 | unsigned int foll_flags, struct page **pages, | |
153 | struct vm_area_struct **vmas, int *nonblocking) | |
1da177e4 | 154 | { |
910e46da | 155 | struct vm_area_struct *vma; |
7b4d5b8b DH |
156 | unsigned long vm_flags; |
157 | int i; | |
158 | ||
159 | /* calculate required read or write permissions. | |
58fa879e | 160 | * If FOLL_FORCE is set, we only require the "MAY" flags. |
7b4d5b8b | 161 | */ |
58fa879e HD |
162 | vm_flags = (foll_flags & FOLL_WRITE) ? |
163 | (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD); | |
164 | vm_flags &= (foll_flags & FOLL_FORCE) ? | |
165 | (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE); | |
1da177e4 | 166 | |
9d73777e | 167 | for (i = 0; i < nr_pages; i++) { |
7561e8ca | 168 | vma = find_vma(mm, start); |
7b4d5b8b DH |
169 | if (!vma) |
170 | goto finish_or_fault; | |
171 | ||
172 | /* protect what we can, including chardevs */ | |
1c3aff1c HD |
173 | if ((vma->vm_flags & (VM_IO | VM_PFNMAP)) || |
174 | !(vm_flags & vma->vm_flags)) | |
7b4d5b8b | 175 | goto finish_or_fault; |
910e46da | 176 | |
1da177e4 LT |
177 | if (pages) { |
178 | pages[i] = virt_to_page(start); | |
179 | if (pages[i]) | |
180 | page_cache_get(pages[i]); | |
181 | } | |
182 | if (vmas) | |
910e46da | 183 | vmas[i] = vma; |
e1ee65d8 | 184 | start = (start + PAGE_SIZE) & PAGE_MASK; |
1da177e4 | 185 | } |
7b4d5b8b DH |
186 | |
187 | return i; | |
188 | ||
189 | finish_or_fault: | |
190 | return i ? : -EFAULT; | |
1da177e4 | 191 | } |
b291f000 | 192 | |
b291f000 NP |
193 | /* |
194 | * get a list of pages in an address range belonging to the specified process | |
195 | * and indicate the VMA that covers each page | |
196 | * - this is potentially dodgy as we may end incrementing the page count of a | |
197 | * slab page or a secondary page from a compound page | |
198 | * - don't permit access to VMAs that don't support it, such as I/O mappings | |
199 | */ | |
28a35716 ML |
200 | long get_user_pages(struct task_struct *tsk, struct mm_struct *mm, |
201 | unsigned long start, unsigned long nr_pages, | |
202 | int write, int force, struct page **pages, | |
203 | struct vm_area_struct **vmas) | |
b291f000 NP |
204 | { |
205 | int flags = 0; | |
206 | ||
207 | if (write) | |
58fa879e | 208 | flags |= FOLL_WRITE; |
b291f000 | 209 | if (force) |
58fa879e | 210 | flags |= FOLL_FORCE; |
b291f000 | 211 | |
53a7706d ML |
212 | return __get_user_pages(tsk, mm, start, nr_pages, flags, pages, vmas, |
213 | NULL); | |
b291f000 | 214 | } |
66aa2b4b GU |
215 | EXPORT_SYMBOL(get_user_pages); |
216 | ||
f0818f47 AA |
217 | long get_user_pages_locked(struct task_struct *tsk, struct mm_struct *mm, |
218 | unsigned long start, unsigned long nr_pages, | |
219 | int write, int force, struct page **pages, | |
220 | int *locked) | |
221 | { | |
222 | return get_user_pages(tsk, mm, start, nr_pages, write, force, | |
223 | pages, NULL); | |
224 | } | |
225 | EXPORT_SYMBOL(get_user_pages_locked); | |
226 | ||
227 | long get_user_pages_unlocked(struct task_struct *tsk, struct mm_struct *mm, | |
228 | unsigned long start, unsigned long nr_pages, | |
229 | int write, int force, struct page **pages) | |
230 | { | |
231 | long ret; | |
232 | down_read(&mm->mmap_sem); | |
233 | ret = get_user_pages(tsk, mm, start, nr_pages, write, force, | |
234 | pages, NULL); | |
235 | up_read(&mm->mmap_sem); | |
236 | return ret; | |
237 | } | |
238 | EXPORT_SYMBOL(get_user_pages_unlocked); | |
239 | ||
dfc2f91a PM |
240 | /** |
241 | * follow_pfn - look up PFN at a user virtual address | |
242 | * @vma: memory mapping | |
243 | * @address: user virtual address | |
244 | * @pfn: location to store found PFN | |
245 | * | |
246 | * Only IO mappings and raw PFN mappings are allowed. | |
247 | * | |
248 | * Returns zero and the pfn at @pfn on success, -ve otherwise. | |
249 | */ | |
250 | int follow_pfn(struct vm_area_struct *vma, unsigned long address, | |
251 | unsigned long *pfn) | |
252 | { | |
253 | if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) | |
254 | return -EINVAL; | |
255 | ||
256 | *pfn = address >> PAGE_SHIFT; | |
257 | return 0; | |
258 | } | |
259 | EXPORT_SYMBOL(follow_pfn); | |
260 | ||
f1c4069e | 261 | LIST_HEAD(vmap_area_list); |
1da177e4 | 262 | |
b3bdda02 | 263 | void vfree(const void *addr) |
1da177e4 LT |
264 | { |
265 | kfree(addr); | |
266 | } | |
b5073173 | 267 | EXPORT_SYMBOL(vfree); |
1da177e4 | 268 | |
dd0fc66f | 269 | void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot) |
1da177e4 LT |
270 | { |
271 | /* | |
8518609d RD |
272 | * You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc() |
273 | * returns only a logical address. | |
1da177e4 | 274 | */ |
84097518 | 275 | return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM); |
1da177e4 | 276 | } |
b5073173 | 277 | EXPORT_SYMBOL(__vmalloc); |
1da177e4 | 278 | |
f905bc44 PM |
279 | void *vmalloc_user(unsigned long size) |
280 | { | |
281 | void *ret; | |
282 | ||
283 | ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, | |
284 | PAGE_KERNEL); | |
285 | if (ret) { | |
286 | struct vm_area_struct *vma; | |
287 | ||
288 | down_write(¤t->mm->mmap_sem); | |
289 | vma = find_vma(current->mm, (unsigned long)ret); | |
290 | if (vma) | |
291 | vma->vm_flags |= VM_USERMAP; | |
292 | up_write(¤t->mm->mmap_sem); | |
293 | } | |
294 | ||
295 | return ret; | |
296 | } | |
297 | EXPORT_SYMBOL(vmalloc_user); | |
298 | ||
b3bdda02 | 299 | struct page *vmalloc_to_page(const void *addr) |
1da177e4 LT |
300 | { |
301 | return virt_to_page(addr); | |
302 | } | |
b5073173 | 303 | EXPORT_SYMBOL(vmalloc_to_page); |
1da177e4 | 304 | |
b3bdda02 | 305 | unsigned long vmalloc_to_pfn(const void *addr) |
1da177e4 LT |
306 | { |
307 | return page_to_pfn(virt_to_page(addr)); | |
308 | } | |
b5073173 | 309 | EXPORT_SYMBOL(vmalloc_to_pfn); |
1da177e4 LT |
310 | |
311 | long vread(char *buf, char *addr, unsigned long count) | |
312 | { | |
9bde916b CG |
313 | /* Don't allow overflow */ |
314 | if ((unsigned long) buf + count < count) | |
315 | count = -(unsigned long) buf; | |
316 | ||
1da177e4 LT |
317 | memcpy(buf, addr, count); |
318 | return count; | |
319 | } | |
320 | ||
321 | long vwrite(char *buf, char *addr, unsigned long count) | |
322 | { | |
323 | /* Don't allow overflow */ | |
324 | if ((unsigned long) addr + count < count) | |
325 | count = -(unsigned long) addr; | |
326 | ||
327 | memcpy(addr, buf, count); | |
ac714904 | 328 | return count; |
1da177e4 LT |
329 | } |
330 | ||
331 | /* | |
332 | * vmalloc - allocate virtually continguos memory | |
333 | * | |
334 | * @size: allocation size | |
335 | * | |
336 | * Allocate enough pages to cover @size from the page level | |
337 | * allocator and map them into continguos kernel virtual space. | |
338 | * | |
c1c8897f | 339 | * For tight control over page level allocator and protection flags |
1da177e4 LT |
340 | * use __vmalloc() instead. |
341 | */ | |
342 | void *vmalloc(unsigned long size) | |
343 | { | |
344 | return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL); | |
345 | } | |
f6138882 AM |
346 | EXPORT_SYMBOL(vmalloc); |
347 | ||
e1ca7788 DY |
348 | /* |
349 | * vzalloc - allocate virtually continguos memory with zero fill | |
350 | * | |
351 | * @size: allocation size | |
352 | * | |
353 | * Allocate enough pages to cover @size from the page level | |
354 | * allocator and map them into continguos kernel virtual space. | |
355 | * The memory allocated is set to zero. | |
356 | * | |
357 | * For tight control over page level allocator and protection flags | |
358 | * use __vmalloc() instead. | |
359 | */ | |
360 | void *vzalloc(unsigned long size) | |
361 | { | |
362 | return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, | |
363 | PAGE_KERNEL); | |
364 | } | |
365 | EXPORT_SYMBOL(vzalloc); | |
366 | ||
367 | /** | |
368 | * vmalloc_node - allocate memory on a specific node | |
369 | * @size: allocation size | |
370 | * @node: numa node | |
371 | * | |
372 | * Allocate enough pages to cover @size from the page level | |
373 | * allocator and map them into contiguous kernel virtual space. | |
374 | * | |
375 | * For tight control over page level allocator and protection flags | |
376 | * use __vmalloc() instead. | |
377 | */ | |
f6138882 AM |
378 | void *vmalloc_node(unsigned long size, int node) |
379 | { | |
380 | return vmalloc(size); | |
381 | } | |
9a14f653 | 382 | EXPORT_SYMBOL(vmalloc_node); |
e1ca7788 DY |
383 | |
384 | /** | |
385 | * vzalloc_node - allocate memory on a specific node with zero fill | |
386 | * @size: allocation size | |
387 | * @node: numa node | |
388 | * | |
389 | * Allocate enough pages to cover @size from the page level | |
390 | * allocator and map them into contiguous kernel virtual space. | |
391 | * The memory allocated is set to zero. | |
392 | * | |
393 | * For tight control over page level allocator and protection flags | |
394 | * use __vmalloc() instead. | |
395 | */ | |
396 | void *vzalloc_node(unsigned long size, int node) | |
397 | { | |
398 | return vzalloc(size); | |
399 | } | |
400 | EXPORT_SYMBOL(vzalloc_node); | |
1da177e4 | 401 | |
1af446ed PM |
402 | #ifndef PAGE_KERNEL_EXEC |
403 | # define PAGE_KERNEL_EXEC PAGE_KERNEL | |
404 | #endif | |
405 | ||
406 | /** | |
407 | * vmalloc_exec - allocate virtually contiguous, executable memory | |
408 | * @size: allocation size | |
409 | * | |
410 | * Kernel-internal function to allocate enough pages to cover @size | |
411 | * the page level allocator and map them into contiguous and | |
412 | * executable kernel virtual space. | |
413 | * | |
414 | * For tight control over page level allocator and protection flags | |
415 | * use __vmalloc() instead. | |
416 | */ | |
417 | ||
418 | void *vmalloc_exec(unsigned long size) | |
419 | { | |
420 | return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC); | |
421 | } | |
422 | ||
b5073173 PM |
423 | /** |
424 | * vmalloc_32 - allocate virtually contiguous memory (32bit addressable) | |
1da177e4 LT |
425 | * @size: allocation size |
426 | * | |
427 | * Allocate enough 32bit PA addressable pages to cover @size from the | |
428 | * page level allocator and map them into continguos kernel virtual space. | |
429 | */ | |
430 | void *vmalloc_32(unsigned long size) | |
431 | { | |
432 | return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL); | |
433 | } | |
b5073173 PM |
434 | EXPORT_SYMBOL(vmalloc_32); |
435 | ||
436 | /** | |
437 | * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory | |
438 | * @size: allocation size | |
439 | * | |
440 | * The resulting memory area is 32bit addressable and zeroed so it can be | |
441 | * mapped to userspace without leaking data. | |
f905bc44 PM |
442 | * |
443 | * VM_USERMAP is set on the corresponding VMA so that subsequent calls to | |
444 | * remap_vmalloc_range() are permissible. | |
b5073173 PM |
445 | */ |
446 | void *vmalloc_32_user(unsigned long size) | |
447 | { | |
f905bc44 PM |
448 | /* |
449 | * We'll have to sort out the ZONE_DMA bits for 64-bit, | |
450 | * but for now this can simply use vmalloc_user() directly. | |
451 | */ | |
452 | return vmalloc_user(size); | |
b5073173 PM |
453 | } |
454 | EXPORT_SYMBOL(vmalloc_32_user); | |
1da177e4 LT |
455 | |
456 | void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot) | |
457 | { | |
458 | BUG(); | |
459 | return NULL; | |
460 | } | |
b5073173 | 461 | EXPORT_SYMBOL(vmap); |
1da177e4 | 462 | |
b3bdda02 | 463 | void vunmap(const void *addr) |
1da177e4 LT |
464 | { |
465 | BUG(); | |
466 | } | |
b5073173 | 467 | EXPORT_SYMBOL(vunmap); |
1da177e4 | 468 | |
eb6434d9 PM |
469 | void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot) |
470 | { | |
471 | BUG(); | |
472 | return NULL; | |
473 | } | |
474 | EXPORT_SYMBOL(vm_map_ram); | |
475 | ||
476 | void vm_unmap_ram(const void *mem, unsigned int count) | |
477 | { | |
478 | BUG(); | |
479 | } | |
480 | EXPORT_SYMBOL(vm_unmap_ram); | |
481 | ||
482 | void vm_unmap_aliases(void) | |
483 | { | |
484 | } | |
485 | EXPORT_SYMBOL_GPL(vm_unmap_aliases); | |
486 | ||
1eeb66a1 CH |
487 | /* |
488 | * Implement a stub for vmalloc_sync_all() if the architecture chose not to | |
489 | * have one. | |
490 | */ | |
3b32123d | 491 | void __weak vmalloc_sync_all(void) |
1eeb66a1 CH |
492 | { |
493 | } | |
494 | ||
29c185e5 PM |
495 | /** |
496 | * alloc_vm_area - allocate a range of kernel address space | |
497 | * @size: size of the area | |
498 | * | |
499 | * Returns: NULL on failure, vm_struct on success | |
500 | * | |
501 | * This function reserves a range of kernel address space, and | |
502 | * allocates pagetables to map that range. No actual mappings | |
503 | * are created. If the kernel address space is not shared | |
504 | * between processes, it syncs the pagetable across all | |
505 | * processes. | |
506 | */ | |
cd12909c | 507 | struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes) |
29c185e5 PM |
508 | { |
509 | BUG(); | |
510 | return NULL; | |
511 | } | |
512 | EXPORT_SYMBOL_GPL(alloc_vm_area); | |
513 | ||
514 | void free_vm_area(struct vm_struct *area) | |
515 | { | |
516 | BUG(); | |
517 | } | |
518 | EXPORT_SYMBOL_GPL(free_vm_area); | |
519 | ||
b5073173 PM |
520 | int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, |
521 | struct page *page) | |
522 | { | |
523 | return -EINVAL; | |
524 | } | |
525 | EXPORT_SYMBOL(vm_insert_page); | |
526 | ||
1da177e4 LT |
527 | /* |
528 | * sys_brk() for the most part doesn't need the global kernel | |
529 | * lock, except when an application is doing something nasty | |
530 | * like trying to un-brk an area that has already been mapped | |
531 | * to a regular file. in this case, the unmapping will need | |
532 | * to invoke file system routines that need the global lock. | |
533 | */ | |
6a6160a7 | 534 | SYSCALL_DEFINE1(brk, unsigned long, brk) |
1da177e4 LT |
535 | { |
536 | struct mm_struct *mm = current->mm; | |
537 | ||
538 | if (brk < mm->start_brk || brk > mm->context.end_brk) | |
539 | return mm->brk; | |
540 | ||
541 | if (mm->brk == brk) | |
542 | return mm->brk; | |
543 | ||
544 | /* | |
545 | * Always allow shrinking brk | |
546 | */ | |
547 | if (brk <= mm->brk) { | |
548 | mm->brk = brk; | |
549 | return brk; | |
550 | } | |
551 | ||
552 | /* | |
553 | * Ok, looks good - let it rip. | |
554 | */ | |
cfe79c00 | 555 | flush_icache_range(mm->brk, brk); |
1da177e4 LT |
556 | return mm->brk = brk; |
557 | } | |
558 | ||
8feae131 DH |
559 | /* |
560 | * initialise the VMA and region record slabs | |
561 | */ | |
562 | void __init mmap_init(void) | |
1da177e4 | 563 | { |
00a62ce9 KM |
564 | int ret; |
565 | ||
908c7f19 | 566 | ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL); |
00a62ce9 | 567 | VM_BUG_ON(ret); |
33e5d769 | 568 | vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC); |
1da177e4 | 569 | } |
1da177e4 | 570 | |
3034097a | 571 | /* |
8feae131 DH |
572 | * validate the region tree |
573 | * - the caller must hold the region lock | |
3034097a | 574 | */ |
8feae131 DH |
575 | #ifdef CONFIG_DEBUG_NOMMU_REGIONS |
576 | static noinline void validate_nommu_regions(void) | |
3034097a | 577 | { |
8feae131 DH |
578 | struct vm_region *region, *last; |
579 | struct rb_node *p, *lastp; | |
3034097a | 580 | |
8feae131 DH |
581 | lastp = rb_first(&nommu_region_tree); |
582 | if (!lastp) | |
583 | return; | |
584 | ||
585 | last = rb_entry(lastp, struct vm_region, vm_rb); | |
33e5d769 DH |
586 | BUG_ON(unlikely(last->vm_end <= last->vm_start)); |
587 | BUG_ON(unlikely(last->vm_top < last->vm_end)); | |
8feae131 DH |
588 | |
589 | while ((p = rb_next(lastp))) { | |
590 | region = rb_entry(p, struct vm_region, vm_rb); | |
591 | last = rb_entry(lastp, struct vm_region, vm_rb); | |
592 | ||
33e5d769 DH |
593 | BUG_ON(unlikely(region->vm_end <= region->vm_start)); |
594 | BUG_ON(unlikely(region->vm_top < region->vm_end)); | |
595 | BUG_ON(unlikely(region->vm_start < last->vm_top)); | |
3034097a | 596 | |
8feae131 DH |
597 | lastp = p; |
598 | } | |
3034097a | 599 | } |
8feae131 | 600 | #else |
33e5d769 DH |
601 | static void validate_nommu_regions(void) |
602 | { | |
603 | } | |
8feae131 | 604 | #endif |
3034097a DH |
605 | |
606 | /* | |
8feae131 | 607 | * add a region into the global tree |
3034097a | 608 | */ |
8feae131 | 609 | static void add_nommu_region(struct vm_region *region) |
3034097a | 610 | { |
8feae131 DH |
611 | struct vm_region *pregion; |
612 | struct rb_node **p, *parent; | |
3034097a | 613 | |
8feae131 DH |
614 | validate_nommu_regions(); |
615 | ||
8feae131 DH |
616 | parent = NULL; |
617 | p = &nommu_region_tree.rb_node; | |
618 | while (*p) { | |
619 | parent = *p; | |
620 | pregion = rb_entry(parent, struct vm_region, vm_rb); | |
621 | if (region->vm_start < pregion->vm_start) | |
622 | p = &(*p)->rb_left; | |
623 | else if (region->vm_start > pregion->vm_start) | |
624 | p = &(*p)->rb_right; | |
625 | else if (pregion == region) | |
626 | return; | |
627 | else | |
628 | BUG(); | |
3034097a DH |
629 | } |
630 | ||
8feae131 DH |
631 | rb_link_node(®ion->vm_rb, parent, p); |
632 | rb_insert_color(®ion->vm_rb, &nommu_region_tree); | |
3034097a | 633 | |
8feae131 | 634 | validate_nommu_regions(); |
3034097a | 635 | } |
3034097a | 636 | |
930e652a | 637 | /* |
8feae131 | 638 | * delete a region from the global tree |
930e652a | 639 | */ |
8feae131 | 640 | static void delete_nommu_region(struct vm_region *region) |
930e652a | 641 | { |
8feae131 | 642 | BUG_ON(!nommu_region_tree.rb_node); |
930e652a | 643 | |
8feae131 DH |
644 | validate_nommu_regions(); |
645 | rb_erase(®ion->vm_rb, &nommu_region_tree); | |
646 | validate_nommu_regions(); | |
57c8f63e GU |
647 | } |
648 | ||
6fa5f80b | 649 | /* |
8feae131 | 650 | * free a contiguous series of pages |
6fa5f80b | 651 | */ |
8feae131 | 652 | static void free_page_series(unsigned long from, unsigned long to) |
6fa5f80b | 653 | { |
8feae131 DH |
654 | for (; from < to; from += PAGE_SIZE) { |
655 | struct page *page = virt_to_page(from); | |
656 | ||
657 | kdebug("- free %lx", from); | |
33e5d769 | 658 | atomic_long_dec(&mmap_pages_allocated); |
8feae131 | 659 | if (page_count(page) != 1) |
33e5d769 DH |
660 | kdebug("free page %p: refcount not one: %d", |
661 | page, page_count(page)); | |
8feae131 | 662 | put_page(page); |
6fa5f80b | 663 | } |
6fa5f80b DH |
664 | } |
665 | ||
3034097a | 666 | /* |
8feae131 | 667 | * release a reference to a region |
33e5d769 | 668 | * - the caller must hold the region semaphore for writing, which this releases |
dd8632a1 | 669 | * - the region may not have been added to the tree yet, in which case vm_top |
8feae131 | 670 | * will equal vm_start |
3034097a | 671 | */ |
8feae131 DH |
672 | static void __put_nommu_region(struct vm_region *region) |
673 | __releases(nommu_region_sem) | |
1da177e4 | 674 | { |
1e2ae599 | 675 | kenter("%p{%d}", region, region->vm_usage); |
1da177e4 | 676 | |
8feae131 | 677 | BUG_ON(!nommu_region_tree.rb_node); |
1da177e4 | 678 | |
1e2ae599 | 679 | if (--region->vm_usage == 0) { |
dd8632a1 | 680 | if (region->vm_top > region->vm_start) |
8feae131 DH |
681 | delete_nommu_region(region); |
682 | up_write(&nommu_region_sem); | |
683 | ||
684 | if (region->vm_file) | |
685 | fput(region->vm_file); | |
686 | ||
687 | /* IO memory and memory shared directly out of the pagecache | |
688 | * from ramfs/tmpfs mustn't be released here */ | |
689 | if (region->vm_flags & VM_MAPPED_COPY) { | |
690 | kdebug("free series"); | |
dd8632a1 | 691 | free_page_series(region->vm_start, region->vm_top); |
8feae131 DH |
692 | } |
693 | kmem_cache_free(vm_region_jar, region); | |
694 | } else { | |
695 | up_write(&nommu_region_sem); | |
1da177e4 | 696 | } |
8feae131 | 697 | } |
1da177e4 | 698 | |
8feae131 DH |
699 | /* |
700 | * release a reference to a region | |
701 | */ | |
702 | static void put_nommu_region(struct vm_region *region) | |
703 | { | |
704 | down_write(&nommu_region_sem); | |
705 | __put_nommu_region(region); | |
1da177e4 LT |
706 | } |
707 | ||
eb8cdec4 BS |
708 | /* |
709 | * update protection on a vma | |
710 | */ | |
711 | static void protect_vma(struct vm_area_struct *vma, unsigned long flags) | |
712 | { | |
713 | #ifdef CONFIG_MPU | |
714 | struct mm_struct *mm = vma->vm_mm; | |
715 | long start = vma->vm_start & PAGE_MASK; | |
716 | while (start < vma->vm_end) { | |
717 | protect_page(mm, start, flags); | |
718 | start += PAGE_SIZE; | |
719 | } | |
720 | update_protections(mm); | |
721 | #endif | |
722 | } | |
723 | ||
3034097a | 724 | /* |
8feae131 DH |
725 | * add a VMA into a process's mm_struct in the appropriate place in the list |
726 | * and tree and add to the address space's page tree also if not an anonymous | |
727 | * page | |
728 | * - should be called with mm->mmap_sem held writelocked | |
3034097a | 729 | */ |
8feae131 | 730 | static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma) |
1da177e4 | 731 | { |
6038def0 | 732 | struct vm_area_struct *pvma, *prev; |
1da177e4 | 733 | struct address_space *mapping; |
6038def0 | 734 | struct rb_node **p, *parent, *rb_prev; |
8feae131 DH |
735 | |
736 | kenter(",%p", vma); | |
737 | ||
738 | BUG_ON(!vma->vm_region); | |
739 | ||
740 | mm->map_count++; | |
741 | vma->vm_mm = mm; | |
1da177e4 | 742 | |
eb8cdec4 BS |
743 | protect_vma(vma, vma->vm_flags); |
744 | ||
1da177e4 LT |
745 | /* add the VMA to the mapping */ |
746 | if (vma->vm_file) { | |
747 | mapping = vma->vm_file->f_mapping; | |
748 | ||
83cde9e8 | 749 | i_mmap_lock_write(mapping); |
1da177e4 | 750 | flush_dcache_mmap_lock(mapping); |
6b2dbba8 | 751 | vma_interval_tree_insert(vma, &mapping->i_mmap); |
1da177e4 | 752 | flush_dcache_mmap_unlock(mapping); |
83cde9e8 | 753 | i_mmap_unlock_write(mapping); |
1da177e4 LT |
754 | } |
755 | ||
8feae131 | 756 | /* add the VMA to the tree */ |
6038def0 | 757 | parent = rb_prev = NULL; |
8feae131 | 758 | p = &mm->mm_rb.rb_node; |
1da177e4 LT |
759 | while (*p) { |
760 | parent = *p; | |
761 | pvma = rb_entry(parent, struct vm_area_struct, vm_rb); | |
762 | ||
8feae131 DH |
763 | /* sort by: start addr, end addr, VMA struct addr in that order |
764 | * (the latter is necessary as we may get identical VMAs) */ | |
765 | if (vma->vm_start < pvma->vm_start) | |
1da177e4 | 766 | p = &(*p)->rb_left; |
6038def0 NK |
767 | else if (vma->vm_start > pvma->vm_start) { |
768 | rb_prev = parent; | |
1da177e4 | 769 | p = &(*p)->rb_right; |
6038def0 | 770 | } else if (vma->vm_end < pvma->vm_end) |
8feae131 | 771 | p = &(*p)->rb_left; |
6038def0 NK |
772 | else if (vma->vm_end > pvma->vm_end) { |
773 | rb_prev = parent; | |
8feae131 | 774 | p = &(*p)->rb_right; |
6038def0 | 775 | } else if (vma < pvma) |
8feae131 | 776 | p = &(*p)->rb_left; |
6038def0 NK |
777 | else if (vma > pvma) { |
778 | rb_prev = parent; | |
8feae131 | 779 | p = &(*p)->rb_right; |
6038def0 | 780 | } else |
8feae131 | 781 | BUG(); |
1da177e4 LT |
782 | } |
783 | ||
784 | rb_link_node(&vma->vm_rb, parent, p); | |
8feae131 DH |
785 | rb_insert_color(&vma->vm_rb, &mm->mm_rb); |
786 | ||
787 | /* add VMA to the VMA list also */ | |
6038def0 NK |
788 | prev = NULL; |
789 | if (rb_prev) | |
790 | prev = rb_entry(rb_prev, struct vm_area_struct, vm_rb); | |
8feae131 | 791 | |
6038def0 | 792 | __vma_link_list(mm, vma, prev, parent); |
1da177e4 LT |
793 | } |
794 | ||
3034097a | 795 | /* |
8feae131 | 796 | * delete a VMA from its owning mm_struct and address space |
3034097a | 797 | */ |
8feae131 | 798 | static void delete_vma_from_mm(struct vm_area_struct *vma) |
1da177e4 | 799 | { |
615d6e87 | 800 | int i; |
1da177e4 | 801 | struct address_space *mapping; |
8feae131 | 802 | struct mm_struct *mm = vma->vm_mm; |
615d6e87 | 803 | struct task_struct *curr = current; |
8feae131 DH |
804 | |
805 | kenter("%p", vma); | |
806 | ||
eb8cdec4 BS |
807 | protect_vma(vma, 0); |
808 | ||
8feae131 | 809 | mm->map_count--; |
615d6e87 DB |
810 | for (i = 0; i < VMACACHE_SIZE; i++) { |
811 | /* if the vma is cached, invalidate the entire cache */ | |
812 | if (curr->vmacache[i] == vma) { | |
e020d5bd | 813 | vmacache_invalidate(mm); |
615d6e87 DB |
814 | break; |
815 | } | |
816 | } | |
1da177e4 LT |
817 | |
818 | /* remove the VMA from the mapping */ | |
819 | if (vma->vm_file) { | |
820 | mapping = vma->vm_file->f_mapping; | |
821 | ||
83cde9e8 | 822 | i_mmap_lock_write(mapping); |
1da177e4 | 823 | flush_dcache_mmap_lock(mapping); |
6b2dbba8 | 824 | vma_interval_tree_remove(vma, &mapping->i_mmap); |
1da177e4 | 825 | flush_dcache_mmap_unlock(mapping); |
83cde9e8 | 826 | i_mmap_unlock_write(mapping); |
1da177e4 LT |
827 | } |
828 | ||
8feae131 DH |
829 | /* remove from the MM's tree and list */ |
830 | rb_erase(&vma->vm_rb, &mm->mm_rb); | |
b951bf2c NK |
831 | |
832 | if (vma->vm_prev) | |
833 | vma->vm_prev->vm_next = vma->vm_next; | |
834 | else | |
835 | mm->mmap = vma->vm_next; | |
836 | ||
837 | if (vma->vm_next) | |
838 | vma->vm_next->vm_prev = vma->vm_prev; | |
8feae131 DH |
839 | } |
840 | ||
841 | /* | |
842 | * destroy a VMA record | |
843 | */ | |
844 | static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma) | |
845 | { | |
846 | kenter("%p", vma); | |
847 | if (vma->vm_ops && vma->vm_ops->close) | |
848 | vma->vm_ops->close(vma); | |
e9714acf | 849 | if (vma->vm_file) |
8feae131 | 850 | fput(vma->vm_file); |
8feae131 DH |
851 | put_nommu_region(vma->vm_region); |
852 | kmem_cache_free(vm_area_cachep, vma); | |
853 | } | |
854 | ||
855 | /* | |
856 | * look up the first VMA in which addr resides, NULL if none | |
857 | * - should be called with mm->mmap_sem at least held readlocked | |
858 | */ | |
859 | struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr) | |
860 | { | |
861 | struct vm_area_struct *vma; | |
8feae131 DH |
862 | |
863 | /* check the cache first */ | |
615d6e87 DB |
864 | vma = vmacache_find(mm, addr); |
865 | if (likely(vma)) | |
8feae131 DH |
866 | return vma; |
867 | ||
e922c4c5 | 868 | /* trawl the list (there may be multiple mappings in which addr |
8feae131 | 869 | * resides) */ |
e922c4c5 | 870 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
8feae131 DH |
871 | if (vma->vm_start > addr) |
872 | return NULL; | |
873 | if (vma->vm_end > addr) { | |
615d6e87 | 874 | vmacache_update(addr, vma); |
8feae131 DH |
875 | return vma; |
876 | } | |
877 | } | |
878 | ||
879 | return NULL; | |
880 | } | |
881 | EXPORT_SYMBOL(find_vma); | |
882 | ||
883 | /* | |
884 | * find a VMA | |
885 | * - we don't extend stack VMAs under NOMMU conditions | |
886 | */ | |
887 | struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr) | |
888 | { | |
7561e8ca | 889 | return find_vma(mm, addr); |
8feae131 DH |
890 | } |
891 | ||
892 | /* | |
893 | * expand a stack to a given address | |
894 | * - not supported under NOMMU conditions | |
895 | */ | |
896 | int expand_stack(struct vm_area_struct *vma, unsigned long address) | |
897 | { | |
898 | return -ENOMEM; | |
899 | } | |
900 | ||
901 | /* | |
902 | * look up the first VMA exactly that exactly matches addr | |
903 | * - should be called with mm->mmap_sem at least held readlocked | |
904 | */ | |
905 | static struct vm_area_struct *find_vma_exact(struct mm_struct *mm, | |
906 | unsigned long addr, | |
907 | unsigned long len) | |
908 | { | |
909 | struct vm_area_struct *vma; | |
8feae131 DH |
910 | unsigned long end = addr + len; |
911 | ||
912 | /* check the cache first */ | |
615d6e87 DB |
913 | vma = vmacache_find_exact(mm, addr, end); |
914 | if (vma) | |
8feae131 DH |
915 | return vma; |
916 | ||
e922c4c5 | 917 | /* trawl the list (there may be multiple mappings in which addr |
8feae131 | 918 | * resides) */ |
e922c4c5 | 919 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
8feae131 DH |
920 | if (vma->vm_start < addr) |
921 | continue; | |
922 | if (vma->vm_start > addr) | |
923 | return NULL; | |
924 | if (vma->vm_end == end) { | |
615d6e87 | 925 | vmacache_update(addr, vma); |
8feae131 DH |
926 | return vma; |
927 | } | |
928 | } | |
929 | ||
930 | return NULL; | |
1da177e4 LT |
931 | } |
932 | ||
933 | /* | |
934 | * determine whether a mapping should be permitted and, if so, what sort of | |
935 | * mapping we're capable of supporting | |
936 | */ | |
937 | static int validate_mmap_request(struct file *file, | |
938 | unsigned long addr, | |
939 | unsigned long len, | |
940 | unsigned long prot, | |
941 | unsigned long flags, | |
942 | unsigned long pgoff, | |
943 | unsigned long *_capabilities) | |
944 | { | |
8feae131 | 945 | unsigned long capabilities, rlen; |
1da177e4 LT |
946 | int ret; |
947 | ||
948 | /* do the simple checks first */ | |
06aab5a3 | 949 | if (flags & MAP_FIXED) { |
1da177e4 LT |
950 | printk(KERN_DEBUG |
951 | "%d: Can't do fixed-address/overlay mmap of RAM\n", | |
952 | current->pid); | |
953 | return -EINVAL; | |
954 | } | |
955 | ||
956 | if ((flags & MAP_TYPE) != MAP_PRIVATE && | |
957 | (flags & MAP_TYPE) != MAP_SHARED) | |
958 | return -EINVAL; | |
959 | ||
f81cff0d | 960 | if (!len) |
1da177e4 LT |
961 | return -EINVAL; |
962 | ||
f81cff0d | 963 | /* Careful about overflows.. */ |
8feae131 DH |
964 | rlen = PAGE_ALIGN(len); |
965 | if (!rlen || rlen > TASK_SIZE) | |
f81cff0d MF |
966 | return -ENOMEM; |
967 | ||
1da177e4 | 968 | /* offset overflow? */ |
8feae131 | 969 | if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff) |
f81cff0d | 970 | return -EOVERFLOW; |
1da177e4 LT |
971 | |
972 | if (file) { | |
973 | /* validate file mapping requests */ | |
974 | struct address_space *mapping; | |
975 | ||
976 | /* files must support mmap */ | |
72c2d531 | 977 | if (!file->f_op->mmap) |
1da177e4 LT |
978 | return -ENODEV; |
979 | ||
980 | /* work out if what we've got could possibly be shared | |
981 | * - we support chardevs that provide their own "memory" | |
982 | * - we support files/blockdevs that are memory backed | |
983 | */ | |
984 | mapping = file->f_mapping; | |
985 | if (!mapping) | |
496ad9aa | 986 | mapping = file_inode(file)->i_mapping; |
1da177e4 LT |
987 | |
988 | capabilities = 0; | |
989 | if (mapping && mapping->backing_dev_info) | |
990 | capabilities = mapping->backing_dev_info->capabilities; | |
991 | ||
992 | if (!capabilities) { | |
993 | /* no explicit capabilities set, so assume some | |
994 | * defaults */ | |
496ad9aa | 995 | switch (file_inode(file)->i_mode & S_IFMT) { |
1da177e4 LT |
996 | case S_IFREG: |
997 | case S_IFBLK: | |
998 | capabilities = BDI_CAP_MAP_COPY; | |
999 | break; | |
1000 | ||
1001 | case S_IFCHR: | |
1002 | capabilities = | |
1003 | BDI_CAP_MAP_DIRECT | | |
1004 | BDI_CAP_READ_MAP | | |
1005 | BDI_CAP_WRITE_MAP; | |
1006 | break; | |
1007 | ||
1008 | default: | |
1009 | return -EINVAL; | |
1010 | } | |
1011 | } | |
1012 | ||
1013 | /* eliminate any capabilities that we can't support on this | |
1014 | * device */ | |
1015 | if (!file->f_op->get_unmapped_area) | |
1016 | capabilities &= ~BDI_CAP_MAP_DIRECT; | |
1017 | if (!file->f_op->read) | |
1018 | capabilities &= ~BDI_CAP_MAP_COPY; | |
1019 | ||
28d7a6ae GY |
1020 | /* The file shall have been opened with read permission. */ |
1021 | if (!(file->f_mode & FMODE_READ)) | |
1022 | return -EACCES; | |
1023 | ||
1da177e4 LT |
1024 | if (flags & MAP_SHARED) { |
1025 | /* do checks for writing, appending and locking */ | |
1026 | if ((prot & PROT_WRITE) && | |
1027 | !(file->f_mode & FMODE_WRITE)) | |
1028 | return -EACCES; | |
1029 | ||
496ad9aa | 1030 | if (IS_APPEND(file_inode(file)) && |
1da177e4 LT |
1031 | (file->f_mode & FMODE_WRITE)) |
1032 | return -EACCES; | |
1033 | ||
d7a06983 | 1034 | if (locks_verify_locked(file)) |
1da177e4 LT |
1035 | return -EAGAIN; |
1036 | ||
1037 | if (!(capabilities & BDI_CAP_MAP_DIRECT)) | |
1038 | return -ENODEV; | |
1039 | ||
1da177e4 LT |
1040 | /* we mustn't privatise shared mappings */ |
1041 | capabilities &= ~BDI_CAP_MAP_COPY; | |
ac714904 | 1042 | } else { |
1da177e4 LT |
1043 | /* we're going to read the file into private memory we |
1044 | * allocate */ | |
1045 | if (!(capabilities & BDI_CAP_MAP_COPY)) | |
1046 | return -ENODEV; | |
1047 | ||
1048 | /* we don't permit a private writable mapping to be | |
1049 | * shared with the backing device */ | |
1050 | if (prot & PROT_WRITE) | |
1051 | capabilities &= ~BDI_CAP_MAP_DIRECT; | |
1052 | } | |
1053 | ||
3c7b2045 BS |
1054 | if (capabilities & BDI_CAP_MAP_DIRECT) { |
1055 | if (((prot & PROT_READ) && !(capabilities & BDI_CAP_READ_MAP)) || | |
1056 | ((prot & PROT_WRITE) && !(capabilities & BDI_CAP_WRITE_MAP)) || | |
1057 | ((prot & PROT_EXEC) && !(capabilities & BDI_CAP_EXEC_MAP)) | |
1058 | ) { | |
1059 | capabilities &= ~BDI_CAP_MAP_DIRECT; | |
1060 | if (flags & MAP_SHARED) { | |
1061 | printk(KERN_WARNING | |
1062 | "MAP_SHARED not completely supported on !MMU\n"); | |
1063 | return -EINVAL; | |
1064 | } | |
1065 | } | |
1066 | } | |
1067 | ||
1da177e4 LT |
1068 | /* handle executable mappings and implied executable |
1069 | * mappings */ | |
e9536ae7 | 1070 | if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) { |
1da177e4 LT |
1071 | if (prot & PROT_EXEC) |
1072 | return -EPERM; | |
ac714904 | 1073 | } else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) { |
1da177e4 LT |
1074 | /* handle implication of PROT_EXEC by PROT_READ */ |
1075 | if (current->personality & READ_IMPLIES_EXEC) { | |
1076 | if (capabilities & BDI_CAP_EXEC_MAP) | |
1077 | prot |= PROT_EXEC; | |
1078 | } | |
ac714904 | 1079 | } else if ((prot & PROT_READ) && |
1da177e4 LT |
1080 | (prot & PROT_EXEC) && |
1081 | !(capabilities & BDI_CAP_EXEC_MAP) | |
1082 | ) { | |
1083 | /* backing file is not executable, try to copy */ | |
1084 | capabilities &= ~BDI_CAP_MAP_DIRECT; | |
1085 | } | |
ac714904 | 1086 | } else { |
1da177e4 LT |
1087 | /* anonymous mappings are always memory backed and can be |
1088 | * privately mapped | |
1089 | */ | |
1090 | capabilities = BDI_CAP_MAP_COPY; | |
1091 | ||
1092 | /* handle PROT_EXEC implication by PROT_READ */ | |
1093 | if ((prot & PROT_READ) && | |
1094 | (current->personality & READ_IMPLIES_EXEC)) | |
1095 | prot |= PROT_EXEC; | |
1096 | } | |
1097 | ||
1098 | /* allow the security API to have its say */ | |
e5467859 | 1099 | ret = security_mmap_addr(addr); |
1da177e4 LT |
1100 | if (ret < 0) |
1101 | return ret; | |
1102 | ||
1103 | /* looks okay */ | |
1104 | *_capabilities = capabilities; | |
1105 | return 0; | |
1106 | } | |
1107 | ||
1108 | /* | |
1109 | * we've determined that we can make the mapping, now translate what we | |
1110 | * now know into VMA flags | |
1111 | */ | |
1112 | static unsigned long determine_vm_flags(struct file *file, | |
1113 | unsigned long prot, | |
1114 | unsigned long flags, | |
1115 | unsigned long capabilities) | |
1116 | { | |
1117 | unsigned long vm_flags; | |
1118 | ||
1119 | vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags); | |
1da177e4 LT |
1120 | /* vm_flags |= mm->def_flags; */ |
1121 | ||
1122 | if (!(capabilities & BDI_CAP_MAP_DIRECT)) { | |
1123 | /* attempt to share read-only copies of mapped file chunks */ | |
3c7b2045 | 1124 | vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC; |
1da177e4 LT |
1125 | if (file && !(prot & PROT_WRITE)) |
1126 | vm_flags |= VM_MAYSHARE; | |
3c7b2045 | 1127 | } else { |
1da177e4 LT |
1128 | /* overlay a shareable mapping on the backing device or inode |
1129 | * if possible - used for chardevs, ramfs/tmpfs/shmfs and | |
1130 | * romfs/cramfs */ | |
3c7b2045 | 1131 | vm_flags |= VM_MAYSHARE | (capabilities & BDI_CAP_VMFLAGS); |
1da177e4 | 1132 | if (flags & MAP_SHARED) |
3c7b2045 | 1133 | vm_flags |= VM_SHARED; |
1da177e4 LT |
1134 | } |
1135 | ||
1136 | /* refuse to let anyone share private mappings with this process if | |
1137 | * it's being traced - otherwise breakpoints set in it may interfere | |
1138 | * with another untraced process | |
1139 | */ | |
a288eecc | 1140 | if ((flags & MAP_PRIVATE) && current->ptrace) |
1da177e4 LT |
1141 | vm_flags &= ~VM_MAYSHARE; |
1142 | ||
1143 | return vm_flags; | |
1144 | } | |
1145 | ||
1146 | /* | |
8feae131 DH |
1147 | * set up a shared mapping on a file (the driver or filesystem provides and |
1148 | * pins the storage) | |
1da177e4 | 1149 | */ |
8feae131 | 1150 | static int do_mmap_shared_file(struct vm_area_struct *vma) |
1da177e4 LT |
1151 | { |
1152 | int ret; | |
1153 | ||
1154 | ret = vma->vm_file->f_op->mmap(vma->vm_file, vma); | |
dd8632a1 PM |
1155 | if (ret == 0) { |
1156 | vma->vm_region->vm_top = vma->vm_region->vm_end; | |
645d83c5 | 1157 | return 0; |
dd8632a1 | 1158 | } |
1da177e4 LT |
1159 | if (ret != -ENOSYS) |
1160 | return ret; | |
1161 | ||
3fa30460 DH |
1162 | /* getting -ENOSYS indicates that direct mmap isn't possible (as |
1163 | * opposed to tried but failed) so we can only give a suitable error as | |
1164 | * it's not possible to make a private copy if MAP_SHARED was given */ | |
1da177e4 LT |
1165 | return -ENODEV; |
1166 | } | |
1167 | ||
1168 | /* | |
1169 | * set up a private mapping or an anonymous shared mapping | |
1170 | */ | |
8feae131 DH |
1171 | static int do_mmap_private(struct vm_area_struct *vma, |
1172 | struct vm_region *region, | |
645d83c5 DH |
1173 | unsigned long len, |
1174 | unsigned long capabilities) | |
1da177e4 | 1175 | { |
dbc8358c | 1176 | unsigned long total, point; |
1da177e4 | 1177 | void *base; |
8feae131 | 1178 | int ret, order; |
1da177e4 LT |
1179 | |
1180 | /* invoke the file's mapping function so that it can keep track of | |
1181 | * shared mappings on devices or memory | |
1182 | * - VM_MAYSHARE will be set if it may attempt to share | |
1183 | */ | |
645d83c5 | 1184 | if (capabilities & BDI_CAP_MAP_DIRECT) { |
1da177e4 | 1185 | ret = vma->vm_file->f_op->mmap(vma->vm_file, vma); |
dd8632a1 | 1186 | if (ret == 0) { |
1da177e4 | 1187 | /* shouldn't return success if we're not sharing */ |
dd8632a1 PM |
1188 | BUG_ON(!(vma->vm_flags & VM_MAYSHARE)); |
1189 | vma->vm_region->vm_top = vma->vm_region->vm_end; | |
645d83c5 | 1190 | return 0; |
1da177e4 | 1191 | } |
dd8632a1 PM |
1192 | if (ret != -ENOSYS) |
1193 | return ret; | |
1da177e4 LT |
1194 | |
1195 | /* getting an ENOSYS error indicates that direct mmap isn't | |
1196 | * possible (as opposed to tried but failed) so we'll try to | |
1197 | * make a private copy of the data and map that instead */ | |
1198 | } | |
1199 | ||
8feae131 | 1200 | |
1da177e4 LT |
1201 | /* allocate some memory to hold the mapping |
1202 | * - note that this may not return a page-aligned address if the object | |
1203 | * we're allocating is smaller than a page | |
1204 | */ | |
f67d9b15 | 1205 | order = get_order(len); |
8feae131 DH |
1206 | kdebug("alloc order %d for %lx", order, len); |
1207 | ||
8feae131 | 1208 | total = 1 << order; |
f67d9b15 | 1209 | point = len >> PAGE_SHIFT; |
dd8632a1 | 1210 | |
dbc8358c | 1211 | /* we don't want to allocate a power-of-2 sized page set */ |
dd8632a1 | 1212 | if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages) { |
dbc8358c JK |
1213 | total = point; |
1214 | kdebug("try to alloc exact %lu pages", total); | |
1215 | base = alloc_pages_exact(len, GFP_KERNEL); | |
1216 | } else { | |
1217 | base = (void *)__get_free_pages(GFP_KERNEL, order); | |
8feae131 DH |
1218 | } |
1219 | ||
dbc8358c JK |
1220 | if (!base) |
1221 | goto enomem; | |
1222 | ||
1223 | atomic_long_add(total, &mmap_pages_allocated); | |
1da177e4 | 1224 | |
8feae131 DH |
1225 | region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY; |
1226 | region->vm_start = (unsigned long) base; | |
f67d9b15 | 1227 | region->vm_end = region->vm_start + len; |
dd8632a1 | 1228 | region->vm_top = region->vm_start + (total << PAGE_SHIFT); |
8feae131 DH |
1229 | |
1230 | vma->vm_start = region->vm_start; | |
1231 | vma->vm_end = region->vm_start + len; | |
1da177e4 LT |
1232 | |
1233 | if (vma->vm_file) { | |
1234 | /* read the contents of a file into the copy */ | |
1235 | mm_segment_t old_fs; | |
1236 | loff_t fpos; | |
1237 | ||
1238 | fpos = vma->vm_pgoff; | |
1239 | fpos <<= PAGE_SHIFT; | |
1240 | ||
1241 | old_fs = get_fs(); | |
1242 | set_fs(KERNEL_DS); | |
f67d9b15 | 1243 | ret = vma->vm_file->f_op->read(vma->vm_file, base, len, &fpos); |
1da177e4 LT |
1244 | set_fs(old_fs); |
1245 | ||
1246 | if (ret < 0) | |
1247 | goto error_free; | |
1248 | ||
1249 | /* clear the last little bit */ | |
f67d9b15 BL |
1250 | if (ret < len) |
1251 | memset(base + ret, 0, len - ret); | |
1da177e4 | 1252 | |
1da177e4 LT |
1253 | } |
1254 | ||
1255 | return 0; | |
1256 | ||
1257 | error_free: | |
7223bb4a | 1258 | free_page_series(region->vm_start, region->vm_top); |
8feae131 DH |
1259 | region->vm_start = vma->vm_start = 0; |
1260 | region->vm_end = vma->vm_end = 0; | |
dd8632a1 | 1261 | region->vm_top = 0; |
1da177e4 LT |
1262 | return ret; |
1263 | ||
1264 | enomem: | |
b1de0d13 | 1265 | pr_err("Allocation of length %lu from process %d (%s) failed\n", |
05ae6fa3 | 1266 | len, current->pid, current->comm); |
7bf02ea2 | 1267 | show_free_areas(0); |
1da177e4 LT |
1268 | return -ENOMEM; |
1269 | } | |
1270 | ||
1271 | /* | |
1272 | * handle mapping creation for uClinux | |
1273 | */ | |
e3fc629d | 1274 | unsigned long do_mmap_pgoff(struct file *file, |
1da177e4 LT |
1275 | unsigned long addr, |
1276 | unsigned long len, | |
1277 | unsigned long prot, | |
1278 | unsigned long flags, | |
bebeb3d6 | 1279 | unsigned long pgoff, |
41badc15 | 1280 | unsigned long *populate) |
1da177e4 | 1281 | { |
8feae131 DH |
1282 | struct vm_area_struct *vma; |
1283 | struct vm_region *region; | |
1da177e4 | 1284 | struct rb_node *rb; |
8feae131 | 1285 | unsigned long capabilities, vm_flags, result; |
1da177e4 LT |
1286 | int ret; |
1287 | ||
8feae131 DH |
1288 | kenter(",%lx,%lx,%lx,%lx,%lx", addr, len, prot, flags, pgoff); |
1289 | ||
41badc15 | 1290 | *populate = 0; |
bebeb3d6 | 1291 | |
1da177e4 LT |
1292 | /* decide whether we should attempt the mapping, and if so what sort of |
1293 | * mapping */ | |
1294 | ret = validate_mmap_request(file, addr, len, prot, flags, pgoff, | |
1295 | &capabilities); | |
8feae131 DH |
1296 | if (ret < 0) { |
1297 | kleave(" = %d [val]", ret); | |
1da177e4 | 1298 | return ret; |
8feae131 | 1299 | } |
1da177e4 | 1300 | |
06aab5a3 DH |
1301 | /* we ignore the address hint */ |
1302 | addr = 0; | |
f67d9b15 | 1303 | len = PAGE_ALIGN(len); |
06aab5a3 | 1304 | |
1da177e4 LT |
1305 | /* we've determined that we can make the mapping, now translate what we |
1306 | * now know into VMA flags */ | |
1307 | vm_flags = determine_vm_flags(file, prot, flags, capabilities); | |
1308 | ||
8feae131 DH |
1309 | /* we're going to need to record the mapping */ |
1310 | region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL); | |
1311 | if (!region) | |
1312 | goto error_getting_region; | |
1313 | ||
1314 | vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); | |
1315 | if (!vma) | |
1316 | goto error_getting_vma; | |
1da177e4 | 1317 | |
1e2ae599 | 1318 | region->vm_usage = 1; |
8feae131 DH |
1319 | region->vm_flags = vm_flags; |
1320 | region->vm_pgoff = pgoff; | |
1321 | ||
5beb4930 | 1322 | INIT_LIST_HEAD(&vma->anon_vma_chain); |
8feae131 DH |
1323 | vma->vm_flags = vm_flags; |
1324 | vma->vm_pgoff = pgoff; | |
1da177e4 | 1325 | |
8feae131 | 1326 | if (file) { |
cb0942b8 AV |
1327 | region->vm_file = get_file(file); |
1328 | vma->vm_file = get_file(file); | |
8feae131 DH |
1329 | } |
1330 | ||
1331 | down_write(&nommu_region_sem); | |
1332 | ||
1333 | /* if we want to share, we need to check for regions created by other | |
1da177e4 | 1334 | * mmap() calls that overlap with our proposed mapping |
8feae131 | 1335 | * - we can only share with a superset match on most regular files |
1da177e4 LT |
1336 | * - shared mappings on character devices and memory backed files are |
1337 | * permitted to overlap inexactly as far as we are concerned for in | |
1338 | * these cases, sharing is handled in the driver or filesystem rather | |
1339 | * than here | |
1340 | */ | |
1341 | if (vm_flags & VM_MAYSHARE) { | |
8feae131 DH |
1342 | struct vm_region *pregion; |
1343 | unsigned long pglen, rpglen, pgend, rpgend, start; | |
1da177e4 | 1344 | |
8feae131 DH |
1345 | pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; |
1346 | pgend = pgoff + pglen; | |
165b2392 | 1347 | |
8feae131 DH |
1348 | for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) { |
1349 | pregion = rb_entry(rb, struct vm_region, vm_rb); | |
1da177e4 | 1350 | |
8feae131 | 1351 | if (!(pregion->vm_flags & VM_MAYSHARE)) |
1da177e4 LT |
1352 | continue; |
1353 | ||
1354 | /* search for overlapping mappings on the same file */ | |
496ad9aa AV |
1355 | if (file_inode(pregion->vm_file) != |
1356 | file_inode(file)) | |
1da177e4 LT |
1357 | continue; |
1358 | ||
8feae131 | 1359 | if (pregion->vm_pgoff >= pgend) |
1da177e4 LT |
1360 | continue; |
1361 | ||
8feae131 DH |
1362 | rpglen = pregion->vm_end - pregion->vm_start; |
1363 | rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
1364 | rpgend = pregion->vm_pgoff + rpglen; | |
1365 | if (pgoff >= rpgend) | |
1da177e4 LT |
1366 | continue; |
1367 | ||
8feae131 DH |
1368 | /* handle inexactly overlapping matches between |
1369 | * mappings */ | |
1370 | if ((pregion->vm_pgoff != pgoff || rpglen != pglen) && | |
1371 | !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) { | |
1372 | /* new mapping is not a subset of the region */ | |
1da177e4 LT |
1373 | if (!(capabilities & BDI_CAP_MAP_DIRECT)) |
1374 | goto sharing_violation; | |
1375 | continue; | |
1376 | } | |
1377 | ||
8feae131 | 1378 | /* we've found a region we can share */ |
1e2ae599 | 1379 | pregion->vm_usage++; |
8feae131 DH |
1380 | vma->vm_region = pregion; |
1381 | start = pregion->vm_start; | |
1382 | start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT; | |
1383 | vma->vm_start = start; | |
1384 | vma->vm_end = start + len; | |
1385 | ||
1386 | if (pregion->vm_flags & VM_MAPPED_COPY) { | |
1387 | kdebug("share copy"); | |
1388 | vma->vm_flags |= VM_MAPPED_COPY; | |
1389 | } else { | |
1390 | kdebug("share mmap"); | |
1391 | ret = do_mmap_shared_file(vma); | |
1392 | if (ret < 0) { | |
1393 | vma->vm_region = NULL; | |
1394 | vma->vm_start = 0; | |
1395 | vma->vm_end = 0; | |
1e2ae599 | 1396 | pregion->vm_usage--; |
8feae131 DH |
1397 | pregion = NULL; |
1398 | goto error_just_free; | |
1399 | } | |
1400 | } | |
1401 | fput(region->vm_file); | |
1402 | kmem_cache_free(vm_region_jar, region); | |
1403 | region = pregion; | |
1404 | result = start; | |
1405 | goto share; | |
1da177e4 LT |
1406 | } |
1407 | ||
1da177e4 LT |
1408 | /* obtain the address at which to make a shared mapping |
1409 | * - this is the hook for quasi-memory character devices to | |
1410 | * tell us the location of a shared mapping | |
1411 | */ | |
645d83c5 | 1412 | if (capabilities & BDI_CAP_MAP_DIRECT) { |
1da177e4 LT |
1413 | addr = file->f_op->get_unmapped_area(file, addr, len, |
1414 | pgoff, flags); | |
bb005a59 | 1415 | if (IS_ERR_VALUE(addr)) { |
1da177e4 | 1416 | ret = addr; |
bb005a59 | 1417 | if (ret != -ENOSYS) |
8feae131 | 1418 | goto error_just_free; |
1da177e4 LT |
1419 | |
1420 | /* the driver refused to tell us where to site | |
1421 | * the mapping so we'll have to attempt to copy | |
1422 | * it */ | |
bb005a59 | 1423 | ret = -ENODEV; |
1da177e4 | 1424 | if (!(capabilities & BDI_CAP_MAP_COPY)) |
8feae131 | 1425 | goto error_just_free; |
1da177e4 LT |
1426 | |
1427 | capabilities &= ~BDI_CAP_MAP_DIRECT; | |
8feae131 DH |
1428 | } else { |
1429 | vma->vm_start = region->vm_start = addr; | |
1430 | vma->vm_end = region->vm_end = addr + len; | |
1da177e4 LT |
1431 | } |
1432 | } | |
1433 | } | |
1434 | ||
8feae131 | 1435 | vma->vm_region = region; |
1da177e4 | 1436 | |
645d83c5 DH |
1437 | /* set up the mapping |
1438 | * - the region is filled in if BDI_CAP_MAP_DIRECT is still set | |
1439 | */ | |
1da177e4 | 1440 | if (file && vma->vm_flags & VM_SHARED) |
8feae131 | 1441 | ret = do_mmap_shared_file(vma); |
1da177e4 | 1442 | else |
645d83c5 | 1443 | ret = do_mmap_private(vma, region, len, capabilities); |
1da177e4 | 1444 | if (ret < 0) |
645d83c5 DH |
1445 | goto error_just_free; |
1446 | add_nommu_region(region); | |
8feae131 | 1447 | |
ea637639 JZ |
1448 | /* clear anonymous mappings that don't ask for uninitialized data */ |
1449 | if (!vma->vm_file && !(flags & MAP_UNINITIALIZED)) | |
1450 | memset((void *)region->vm_start, 0, | |
1451 | region->vm_end - region->vm_start); | |
1452 | ||
1da177e4 | 1453 | /* okay... we have a mapping; now we have to register it */ |
8feae131 | 1454 | result = vma->vm_start; |
1da177e4 | 1455 | |
1da177e4 LT |
1456 | current->mm->total_vm += len >> PAGE_SHIFT; |
1457 | ||
8feae131 DH |
1458 | share: |
1459 | add_vma_to_mm(current->mm, vma); | |
1da177e4 | 1460 | |
cfe79c00 MF |
1461 | /* we flush the region from the icache only when the first executable |
1462 | * mapping of it is made */ | |
1463 | if (vma->vm_flags & VM_EXEC && !region->vm_icache_flushed) { | |
1464 | flush_icache_range(region->vm_start, region->vm_end); | |
1465 | region->vm_icache_flushed = true; | |
1466 | } | |
1da177e4 | 1467 | |
cfe79c00 | 1468 | up_write(&nommu_region_sem); |
1da177e4 | 1469 | |
8feae131 DH |
1470 | kleave(" = %lx", result); |
1471 | return result; | |
1da177e4 | 1472 | |
8feae131 DH |
1473 | error_just_free: |
1474 | up_write(&nommu_region_sem); | |
1475 | error: | |
89a86402 DH |
1476 | if (region->vm_file) |
1477 | fput(region->vm_file); | |
8feae131 | 1478 | kmem_cache_free(vm_region_jar, region); |
89a86402 DH |
1479 | if (vma->vm_file) |
1480 | fput(vma->vm_file); | |
8feae131 DH |
1481 | kmem_cache_free(vm_area_cachep, vma); |
1482 | kleave(" = %d", ret); | |
1483 | return ret; | |
1484 | ||
1485 | sharing_violation: | |
1486 | up_write(&nommu_region_sem); | |
1487 | printk(KERN_WARNING "Attempt to share mismatched mappings\n"); | |
1488 | ret = -EINVAL; | |
1489 | goto error; | |
1da177e4 | 1490 | |
8feae131 DH |
1491 | error_getting_vma: |
1492 | kmem_cache_free(vm_region_jar, region); | |
1493 | printk(KERN_WARNING "Allocation of vma for %lu byte allocation" | |
1494 | " from process %d failed\n", | |
1da177e4 | 1495 | len, current->pid); |
7bf02ea2 | 1496 | show_free_areas(0); |
1da177e4 LT |
1497 | return -ENOMEM; |
1498 | ||
8feae131 DH |
1499 | error_getting_region: |
1500 | printk(KERN_WARNING "Allocation of vm region for %lu byte allocation" | |
1501 | " from process %d failed\n", | |
1da177e4 | 1502 | len, current->pid); |
7bf02ea2 | 1503 | show_free_areas(0); |
1da177e4 LT |
1504 | return -ENOMEM; |
1505 | } | |
6be5ceb0 | 1506 | |
66f0dc48 HD |
1507 | SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len, |
1508 | unsigned long, prot, unsigned long, flags, | |
1509 | unsigned long, fd, unsigned long, pgoff) | |
1510 | { | |
1511 | struct file *file = NULL; | |
1512 | unsigned long retval = -EBADF; | |
1513 | ||
120a795d | 1514 | audit_mmap_fd(fd, flags); |
66f0dc48 HD |
1515 | if (!(flags & MAP_ANONYMOUS)) { |
1516 | file = fget(fd); | |
1517 | if (!file) | |
1518 | goto out; | |
1519 | } | |
1520 | ||
1521 | flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE); | |
1522 | ||
ad1ed293 | 1523 | retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff); |
66f0dc48 HD |
1524 | |
1525 | if (file) | |
1526 | fput(file); | |
1527 | out: | |
1528 | return retval; | |
1529 | } | |
1530 | ||
a4679373 CH |
1531 | #ifdef __ARCH_WANT_SYS_OLD_MMAP |
1532 | struct mmap_arg_struct { | |
1533 | unsigned long addr; | |
1534 | unsigned long len; | |
1535 | unsigned long prot; | |
1536 | unsigned long flags; | |
1537 | unsigned long fd; | |
1538 | unsigned long offset; | |
1539 | }; | |
1540 | ||
1541 | SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg) | |
1542 | { | |
1543 | struct mmap_arg_struct a; | |
1544 | ||
1545 | if (copy_from_user(&a, arg, sizeof(a))) | |
1546 | return -EFAULT; | |
1547 | if (a.offset & ~PAGE_MASK) | |
1548 | return -EINVAL; | |
1549 | ||
1550 | return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd, | |
1551 | a.offset >> PAGE_SHIFT); | |
1552 | } | |
1553 | #endif /* __ARCH_WANT_SYS_OLD_MMAP */ | |
1554 | ||
1da177e4 | 1555 | /* |
8feae131 DH |
1556 | * split a vma into two pieces at address 'addr', a new vma is allocated either |
1557 | * for the first part or the tail. | |
1da177e4 | 1558 | */ |
8feae131 DH |
1559 | int split_vma(struct mm_struct *mm, struct vm_area_struct *vma, |
1560 | unsigned long addr, int new_below) | |
1da177e4 | 1561 | { |
8feae131 DH |
1562 | struct vm_area_struct *new; |
1563 | struct vm_region *region; | |
1564 | unsigned long npages; | |
1da177e4 | 1565 | |
8feae131 | 1566 | kenter(""); |
1da177e4 | 1567 | |
779c1023 DH |
1568 | /* we're only permitted to split anonymous regions (these should have |
1569 | * only a single usage on the region) */ | |
1570 | if (vma->vm_file) | |
8feae131 | 1571 | return -ENOMEM; |
1da177e4 | 1572 | |
8feae131 DH |
1573 | if (mm->map_count >= sysctl_max_map_count) |
1574 | return -ENOMEM; | |
1da177e4 | 1575 | |
8feae131 DH |
1576 | region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL); |
1577 | if (!region) | |
1578 | return -ENOMEM; | |
1da177e4 | 1579 | |
8feae131 DH |
1580 | new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); |
1581 | if (!new) { | |
1582 | kmem_cache_free(vm_region_jar, region); | |
1583 | return -ENOMEM; | |
1584 | } | |
1585 | ||
1586 | /* most fields are the same, copy all, and then fixup */ | |
1587 | *new = *vma; | |
1588 | *region = *vma->vm_region; | |
1589 | new->vm_region = region; | |
1590 | ||
1591 | npages = (addr - vma->vm_start) >> PAGE_SHIFT; | |
1592 | ||
1593 | if (new_below) { | |
dd8632a1 | 1594 | region->vm_top = region->vm_end = new->vm_end = addr; |
8feae131 DH |
1595 | } else { |
1596 | region->vm_start = new->vm_start = addr; | |
1597 | region->vm_pgoff = new->vm_pgoff += npages; | |
1da177e4 | 1598 | } |
8feae131 DH |
1599 | |
1600 | if (new->vm_ops && new->vm_ops->open) | |
1601 | new->vm_ops->open(new); | |
1602 | ||
1603 | delete_vma_from_mm(vma); | |
1604 | down_write(&nommu_region_sem); | |
1605 | delete_nommu_region(vma->vm_region); | |
1606 | if (new_below) { | |
1607 | vma->vm_region->vm_start = vma->vm_start = addr; | |
1608 | vma->vm_region->vm_pgoff = vma->vm_pgoff += npages; | |
1609 | } else { | |
1610 | vma->vm_region->vm_end = vma->vm_end = addr; | |
dd8632a1 | 1611 | vma->vm_region->vm_top = addr; |
8feae131 DH |
1612 | } |
1613 | add_nommu_region(vma->vm_region); | |
1614 | add_nommu_region(new->vm_region); | |
1615 | up_write(&nommu_region_sem); | |
1616 | add_vma_to_mm(mm, vma); | |
1617 | add_vma_to_mm(mm, new); | |
1618 | return 0; | |
1da177e4 LT |
1619 | } |
1620 | ||
3034097a | 1621 | /* |
8feae131 DH |
1622 | * shrink a VMA by removing the specified chunk from either the beginning or |
1623 | * the end | |
3034097a | 1624 | */ |
8feae131 DH |
1625 | static int shrink_vma(struct mm_struct *mm, |
1626 | struct vm_area_struct *vma, | |
1627 | unsigned long from, unsigned long to) | |
1da177e4 | 1628 | { |
8feae131 | 1629 | struct vm_region *region; |
1da177e4 | 1630 | |
8feae131 | 1631 | kenter(""); |
1da177e4 | 1632 | |
8feae131 DH |
1633 | /* adjust the VMA's pointers, which may reposition it in the MM's tree |
1634 | * and list */ | |
1635 | delete_vma_from_mm(vma); | |
1636 | if (from > vma->vm_start) | |
1637 | vma->vm_end = from; | |
1638 | else | |
1639 | vma->vm_start = to; | |
1640 | add_vma_to_mm(mm, vma); | |
1da177e4 | 1641 | |
8feae131 DH |
1642 | /* cut the backing region down to size */ |
1643 | region = vma->vm_region; | |
1e2ae599 | 1644 | BUG_ON(region->vm_usage != 1); |
8feae131 DH |
1645 | |
1646 | down_write(&nommu_region_sem); | |
1647 | delete_nommu_region(region); | |
dd8632a1 PM |
1648 | if (from > region->vm_start) { |
1649 | to = region->vm_top; | |
1650 | region->vm_top = region->vm_end = from; | |
1651 | } else { | |
8feae131 | 1652 | region->vm_start = to; |
dd8632a1 | 1653 | } |
8feae131 DH |
1654 | add_nommu_region(region); |
1655 | up_write(&nommu_region_sem); | |
1656 | ||
1657 | free_page_series(from, to); | |
1658 | return 0; | |
1659 | } | |
1da177e4 | 1660 | |
8feae131 DH |
1661 | /* |
1662 | * release a mapping | |
1663 | * - under NOMMU conditions the chunk to be unmapped must be backed by a single | |
1664 | * VMA, though it need not cover the whole VMA | |
1665 | */ | |
1666 | int do_munmap(struct mm_struct *mm, unsigned long start, size_t len) | |
1667 | { | |
1668 | struct vm_area_struct *vma; | |
f67d9b15 | 1669 | unsigned long end; |
8feae131 | 1670 | int ret; |
1da177e4 | 1671 | |
8feae131 | 1672 | kenter(",%lx,%zx", start, len); |
1da177e4 | 1673 | |
f67d9b15 | 1674 | len = PAGE_ALIGN(len); |
8feae131 DH |
1675 | if (len == 0) |
1676 | return -EINVAL; | |
365e9c87 | 1677 | |
f67d9b15 BL |
1678 | end = start + len; |
1679 | ||
8feae131 DH |
1680 | /* find the first potentially overlapping VMA */ |
1681 | vma = find_vma(mm, start); | |
1682 | if (!vma) { | |
ac714904 | 1683 | static int limit; |
33e5d769 DH |
1684 | if (limit < 5) { |
1685 | printk(KERN_WARNING | |
1686 | "munmap of memory not mmapped by process %d" | |
1687 | " (%s): 0x%lx-0x%lx\n", | |
1688 | current->pid, current->comm, | |
1689 | start, start + len - 1); | |
1690 | limit++; | |
1691 | } | |
8feae131 DH |
1692 | return -EINVAL; |
1693 | } | |
1da177e4 | 1694 | |
8feae131 DH |
1695 | /* we're allowed to split an anonymous VMA but not a file-backed one */ |
1696 | if (vma->vm_file) { | |
1697 | do { | |
1698 | if (start > vma->vm_start) { | |
1699 | kleave(" = -EINVAL [miss]"); | |
1700 | return -EINVAL; | |
1701 | } | |
1702 | if (end == vma->vm_end) | |
1703 | goto erase_whole_vma; | |
d75a310c NK |
1704 | vma = vma->vm_next; |
1705 | } while (vma); | |
8feae131 DH |
1706 | kleave(" = -EINVAL [split file]"); |
1707 | return -EINVAL; | |
1708 | } else { | |
1709 | /* the chunk must be a subset of the VMA found */ | |
1710 | if (start == vma->vm_start && end == vma->vm_end) | |
1711 | goto erase_whole_vma; | |
1712 | if (start < vma->vm_start || end > vma->vm_end) { | |
1713 | kleave(" = -EINVAL [superset]"); | |
1714 | return -EINVAL; | |
1715 | } | |
1716 | if (start & ~PAGE_MASK) { | |
1717 | kleave(" = -EINVAL [unaligned start]"); | |
1718 | return -EINVAL; | |
1719 | } | |
1720 | if (end != vma->vm_end && end & ~PAGE_MASK) { | |
1721 | kleave(" = -EINVAL [unaligned split]"); | |
1722 | return -EINVAL; | |
1723 | } | |
1724 | if (start != vma->vm_start && end != vma->vm_end) { | |
1725 | ret = split_vma(mm, vma, start, 1); | |
1726 | if (ret < 0) { | |
1727 | kleave(" = %d [split]", ret); | |
1728 | return ret; | |
1729 | } | |
1730 | } | |
1731 | return shrink_vma(mm, vma, start, end); | |
1732 | } | |
1da177e4 | 1733 | |
8feae131 DH |
1734 | erase_whole_vma: |
1735 | delete_vma_from_mm(vma); | |
1736 | delete_vma(mm, vma); | |
1737 | kleave(" = 0"); | |
1da177e4 LT |
1738 | return 0; |
1739 | } | |
b5073173 | 1740 | EXPORT_SYMBOL(do_munmap); |
1da177e4 | 1741 | |
bfce281c | 1742 | int vm_munmap(unsigned long addr, size_t len) |
3034097a | 1743 | { |
bfce281c | 1744 | struct mm_struct *mm = current->mm; |
3034097a | 1745 | int ret; |
3034097a DH |
1746 | |
1747 | down_write(&mm->mmap_sem); | |
1748 | ret = do_munmap(mm, addr, len); | |
1749 | up_write(&mm->mmap_sem); | |
1750 | return ret; | |
1751 | } | |
a46ef99d LT |
1752 | EXPORT_SYMBOL(vm_munmap); |
1753 | ||
1754 | SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len) | |
1755 | { | |
bfce281c | 1756 | return vm_munmap(addr, len); |
a46ef99d | 1757 | } |
3034097a DH |
1758 | |
1759 | /* | |
8feae131 | 1760 | * release all the mappings made in a process's VM space |
3034097a | 1761 | */ |
8feae131 | 1762 | void exit_mmap(struct mm_struct *mm) |
1da177e4 | 1763 | { |
8feae131 | 1764 | struct vm_area_struct *vma; |
1da177e4 | 1765 | |
8feae131 DH |
1766 | if (!mm) |
1767 | return; | |
1da177e4 | 1768 | |
8feae131 | 1769 | kenter(""); |
1da177e4 | 1770 | |
8feae131 | 1771 | mm->total_vm = 0; |
1da177e4 | 1772 | |
8feae131 DH |
1773 | while ((vma = mm->mmap)) { |
1774 | mm->mmap = vma->vm_next; | |
1775 | delete_vma_from_mm(vma); | |
1776 | delete_vma(mm, vma); | |
04c34961 | 1777 | cond_resched(); |
1da177e4 | 1778 | } |
8feae131 DH |
1779 | |
1780 | kleave(""); | |
1da177e4 LT |
1781 | } |
1782 | ||
e4eb1ff6 | 1783 | unsigned long vm_brk(unsigned long addr, unsigned long len) |
1da177e4 LT |
1784 | { |
1785 | return -ENOMEM; | |
1786 | } | |
1787 | ||
1788 | /* | |
6fa5f80b DH |
1789 | * expand (or shrink) an existing mapping, potentially moving it at the same |
1790 | * time (controlled by the MREMAP_MAYMOVE flag and available VM space) | |
1da177e4 | 1791 | * |
6fa5f80b | 1792 | * under NOMMU conditions, we only permit changing a mapping's size, and only |
8feae131 DH |
1793 | * as long as it stays within the region allocated by do_mmap_private() and the |
1794 | * block is not shareable | |
1da177e4 | 1795 | * |
6fa5f80b | 1796 | * MREMAP_FIXED is not supported under NOMMU conditions |
1da177e4 | 1797 | */ |
4b377bab | 1798 | static unsigned long do_mremap(unsigned long addr, |
1da177e4 LT |
1799 | unsigned long old_len, unsigned long new_len, |
1800 | unsigned long flags, unsigned long new_addr) | |
1801 | { | |
6fa5f80b | 1802 | struct vm_area_struct *vma; |
1da177e4 LT |
1803 | |
1804 | /* insanity checks first */ | |
f67d9b15 BL |
1805 | old_len = PAGE_ALIGN(old_len); |
1806 | new_len = PAGE_ALIGN(new_len); | |
8feae131 | 1807 | if (old_len == 0 || new_len == 0) |
1da177e4 LT |
1808 | return (unsigned long) -EINVAL; |
1809 | ||
8feae131 DH |
1810 | if (addr & ~PAGE_MASK) |
1811 | return -EINVAL; | |
1812 | ||
1da177e4 LT |
1813 | if (flags & MREMAP_FIXED && new_addr != addr) |
1814 | return (unsigned long) -EINVAL; | |
1815 | ||
8feae131 | 1816 | vma = find_vma_exact(current->mm, addr, old_len); |
6fa5f80b DH |
1817 | if (!vma) |
1818 | return (unsigned long) -EINVAL; | |
1da177e4 | 1819 | |
6fa5f80b | 1820 | if (vma->vm_end != vma->vm_start + old_len) |
1da177e4 LT |
1821 | return (unsigned long) -EFAULT; |
1822 | ||
6fa5f80b | 1823 | if (vma->vm_flags & VM_MAYSHARE) |
1da177e4 LT |
1824 | return (unsigned long) -EPERM; |
1825 | ||
8feae131 | 1826 | if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start) |
1da177e4 LT |
1827 | return (unsigned long) -ENOMEM; |
1828 | ||
1829 | /* all checks complete - do it */ | |
6fa5f80b | 1830 | vma->vm_end = vma->vm_start + new_len; |
6fa5f80b DH |
1831 | return vma->vm_start; |
1832 | } | |
1833 | ||
6a6160a7 HC |
1834 | SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len, |
1835 | unsigned long, new_len, unsigned long, flags, | |
1836 | unsigned long, new_addr) | |
6fa5f80b DH |
1837 | { |
1838 | unsigned long ret; | |
1839 | ||
1840 | down_write(¤t->mm->mmap_sem); | |
1841 | ret = do_mremap(addr, old_len, new_len, flags, new_addr); | |
1842 | up_write(¤t->mm->mmap_sem); | |
1843 | return ret; | |
1da177e4 LT |
1844 | } |
1845 | ||
240aadee ML |
1846 | struct page *follow_page_mask(struct vm_area_struct *vma, |
1847 | unsigned long address, unsigned int flags, | |
1848 | unsigned int *page_mask) | |
1da177e4 | 1849 | { |
240aadee | 1850 | *page_mask = 0; |
1da177e4 LT |
1851 | return NULL; |
1852 | } | |
1853 | ||
8f3b1327 BL |
1854 | int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr, |
1855 | unsigned long pfn, unsigned long size, pgprot_t prot) | |
1da177e4 | 1856 | { |
8f3b1327 BL |
1857 | if (addr != (pfn << PAGE_SHIFT)) |
1858 | return -EINVAL; | |
1859 | ||
314e51b9 | 1860 | vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP; |
66aa2b4b | 1861 | return 0; |
1da177e4 | 1862 | } |
22c4af40 | 1863 | EXPORT_SYMBOL(remap_pfn_range); |
1da177e4 | 1864 | |
3c0b9de6 LT |
1865 | int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len) |
1866 | { | |
1867 | unsigned long pfn = start >> PAGE_SHIFT; | |
1868 | unsigned long vm_len = vma->vm_end - vma->vm_start; | |
1869 | ||
1870 | pfn += vma->vm_pgoff; | |
1871 | return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot); | |
1872 | } | |
1873 | EXPORT_SYMBOL(vm_iomap_memory); | |
1874 | ||
f905bc44 PM |
1875 | int remap_vmalloc_range(struct vm_area_struct *vma, void *addr, |
1876 | unsigned long pgoff) | |
1877 | { | |
1878 | unsigned int size = vma->vm_end - vma->vm_start; | |
1879 | ||
1880 | if (!(vma->vm_flags & VM_USERMAP)) | |
1881 | return -EINVAL; | |
1882 | ||
1883 | vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT)); | |
1884 | vma->vm_end = vma->vm_start + size; | |
1885 | ||
1886 | return 0; | |
1887 | } | |
1888 | EXPORT_SYMBOL(remap_vmalloc_range); | |
1889 | ||
1da177e4 LT |
1890 | unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr, |
1891 | unsigned long len, unsigned long pgoff, unsigned long flags) | |
1892 | { | |
1893 | return -ENOMEM; | |
1894 | } | |
1895 | ||
1da177e4 LT |
1896 | void unmap_mapping_range(struct address_space *mapping, |
1897 | loff_t const holebegin, loff_t const holelen, | |
1898 | int even_cows) | |
1899 | { | |
1900 | } | |
22c4af40 | 1901 | EXPORT_SYMBOL(unmap_mapping_range); |
1da177e4 LT |
1902 | |
1903 | /* | |
1904 | * Check that a process has enough memory to allocate a new virtual | |
1905 | * mapping. 0 means there is enough memory for the allocation to | |
1906 | * succeed and -ENOMEM implies there is not. | |
1907 | * | |
1908 | * We currently support three overcommit policies, which are set via the | |
1909 | * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting | |
1910 | * | |
1911 | * Strict overcommit modes added 2002 Feb 26 by Alan Cox. | |
1912 | * Additional code 2002 Jul 20 by Robert Love. | |
1913 | * | |
1914 | * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise. | |
1915 | * | |
1916 | * Note this is a helper function intended to be used by LSMs which | |
1917 | * wish to use this logic. | |
1918 | */ | |
34b4e4aa | 1919 | int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin) |
1da177e4 | 1920 | { |
c9b1d098 | 1921 | unsigned long free, allowed, reserve; |
1da177e4 LT |
1922 | |
1923 | vm_acct_memory(pages); | |
1924 | ||
1925 | /* | |
1926 | * Sometimes we want to use more memory than we have | |
1927 | */ | |
1928 | if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS) | |
1929 | return 0; | |
1930 | ||
1931 | if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) { | |
c15bef30 DF |
1932 | free = global_page_state(NR_FREE_PAGES); |
1933 | free += global_page_state(NR_FILE_PAGES); | |
1934 | ||
1935 | /* | |
1936 | * shmem pages shouldn't be counted as free in this | |
1937 | * case, they can't be purged, only swapped out, and | |
1938 | * that won't affect the overall amount of available | |
1939 | * memory in the system. | |
1940 | */ | |
1941 | free -= global_page_state(NR_SHMEM); | |
1da177e4 | 1942 | |
ec8acf20 | 1943 | free += get_nr_swap_pages(); |
1da177e4 LT |
1944 | |
1945 | /* | |
1946 | * Any slabs which are created with the | |
1947 | * SLAB_RECLAIM_ACCOUNT flag claim to have contents | |
1948 | * which are reclaimable, under pressure. The dentry | |
1949 | * cache and most inode caches should fall into this | |
1950 | */ | |
972d1a7b | 1951 | free += global_page_state(NR_SLAB_RECLAIMABLE); |
1da177e4 | 1952 | |
d5ddc79b HA |
1953 | /* |
1954 | * Leave reserved pages. The pages are not for anonymous pages. | |
1955 | */ | |
c15bef30 | 1956 | if (free <= totalreserve_pages) |
d5ddc79b HA |
1957 | goto error; |
1958 | else | |
c15bef30 | 1959 | free -= totalreserve_pages; |
d5ddc79b HA |
1960 | |
1961 | /* | |
4eeab4f5 | 1962 | * Reserve some for root |
d5ddc79b | 1963 | */ |
1da177e4 | 1964 | if (!cap_sys_admin) |
4eeab4f5 | 1965 | free -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10); |
1da177e4 LT |
1966 | |
1967 | if (free > pages) | |
1968 | return 0; | |
d5ddc79b HA |
1969 | |
1970 | goto error; | |
1da177e4 LT |
1971 | } |
1972 | ||
00619bcc | 1973 | allowed = vm_commit_limit(); |
1da177e4 | 1974 | /* |
4eeab4f5 | 1975 | * Reserve some 3% for root |
1da177e4 LT |
1976 | */ |
1977 | if (!cap_sys_admin) | |
4eeab4f5 | 1978 | allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10); |
1da177e4 | 1979 | |
c9b1d098 AS |
1980 | /* |
1981 | * Don't let a single process grow so big a user can't recover | |
1982 | */ | |
1983 | if (mm) { | |
1984 | reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10); | |
1985 | allowed -= min(mm->total_vm / 32, reserve); | |
1986 | } | |
1da177e4 | 1987 | |
00a62ce9 | 1988 | if (percpu_counter_read_positive(&vm_committed_as) < allowed) |
1da177e4 | 1989 | return 0; |
00a62ce9 | 1990 | |
d5ddc79b | 1991 | error: |
1da177e4 LT |
1992 | vm_unacct_memory(pages); |
1993 | ||
1994 | return -ENOMEM; | |
1995 | } | |
1996 | ||
d0217ac0 | 1997 | int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf) |
b0e15190 DH |
1998 | { |
1999 | BUG(); | |
d0217ac0 | 2000 | return 0; |
b0e15190 | 2001 | } |
b5073173 | 2002 | EXPORT_SYMBOL(filemap_fault); |
0ec76a11 | 2003 | |
f1820361 KS |
2004 | void filemap_map_pages(struct vm_area_struct *vma, struct vm_fault *vmf) |
2005 | { | |
2006 | BUG(); | |
2007 | } | |
2008 | EXPORT_SYMBOL(filemap_map_pages); | |
2009 | ||
f55f199b MF |
2010 | static int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm, |
2011 | unsigned long addr, void *buf, int len, int write) | |
0ec76a11 | 2012 | { |
0ec76a11 | 2013 | struct vm_area_struct *vma; |
0ec76a11 DH |
2014 | |
2015 | down_read(&mm->mmap_sem); | |
2016 | ||
2017 | /* the access must start within one of the target process's mappings */ | |
0159b141 DH |
2018 | vma = find_vma(mm, addr); |
2019 | if (vma) { | |
0ec76a11 DH |
2020 | /* don't overrun this mapping */ |
2021 | if (addr + len >= vma->vm_end) | |
2022 | len = vma->vm_end - addr; | |
2023 | ||
2024 | /* only read or write mappings where it is permitted */ | |
d00c7b99 | 2025 | if (write && vma->vm_flags & VM_MAYWRITE) |
7959722b JZ |
2026 | copy_to_user_page(vma, NULL, addr, |
2027 | (void *) addr, buf, len); | |
d00c7b99 | 2028 | else if (!write && vma->vm_flags & VM_MAYREAD) |
7959722b JZ |
2029 | copy_from_user_page(vma, NULL, addr, |
2030 | buf, (void *) addr, len); | |
0ec76a11 DH |
2031 | else |
2032 | len = 0; | |
2033 | } else { | |
2034 | len = 0; | |
2035 | } | |
2036 | ||
2037 | up_read(&mm->mmap_sem); | |
f55f199b MF |
2038 | |
2039 | return len; | |
2040 | } | |
2041 | ||
2042 | /** | |
2043 | * @access_remote_vm - access another process' address space | |
2044 | * @mm: the mm_struct of the target address space | |
2045 | * @addr: start address to access | |
2046 | * @buf: source or destination buffer | |
2047 | * @len: number of bytes to transfer | |
2048 | * @write: whether the access is a write | |
2049 | * | |
2050 | * The caller must hold a reference on @mm. | |
2051 | */ | |
2052 | int access_remote_vm(struct mm_struct *mm, unsigned long addr, | |
2053 | void *buf, int len, int write) | |
2054 | { | |
2055 | return __access_remote_vm(NULL, mm, addr, buf, len, write); | |
2056 | } | |
2057 | ||
2058 | /* | |
2059 | * Access another process' address space. | |
2060 | * - source/target buffer must be kernel space | |
2061 | */ | |
2062 | int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write) | |
2063 | { | |
2064 | struct mm_struct *mm; | |
2065 | ||
2066 | if (addr + len < addr) | |
2067 | return 0; | |
2068 | ||
2069 | mm = get_task_mm(tsk); | |
2070 | if (!mm) | |
2071 | return 0; | |
2072 | ||
2073 | len = __access_remote_vm(tsk, mm, addr, buf, len, write); | |
2074 | ||
0ec76a11 DH |
2075 | mmput(mm); |
2076 | return len; | |
2077 | } | |
7e660872 DH |
2078 | |
2079 | /** | |
2080 | * nommu_shrink_inode_mappings - Shrink the shared mappings on an inode | |
2081 | * @inode: The inode to check | |
2082 | * @size: The current filesize of the inode | |
2083 | * @newsize: The proposed filesize of the inode | |
2084 | * | |
2085 | * Check the shared mappings on an inode on behalf of a shrinking truncate to | |
2086 | * make sure that that any outstanding VMAs aren't broken and then shrink the | |
2087 | * vm_regions that extend that beyond so that do_mmap_pgoff() doesn't | |
2088 | * automatically grant mappings that are too large. | |
2089 | */ | |
2090 | int nommu_shrink_inode_mappings(struct inode *inode, size_t size, | |
2091 | size_t newsize) | |
2092 | { | |
2093 | struct vm_area_struct *vma; | |
7e660872 DH |
2094 | struct vm_region *region; |
2095 | pgoff_t low, high; | |
2096 | size_t r_size, r_top; | |
2097 | ||
2098 | low = newsize >> PAGE_SHIFT; | |
2099 | high = (size + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
2100 | ||
2101 | down_write(&nommu_region_sem); | |
1acf2e04 | 2102 | i_mmap_lock_read(inode->i_mapping); |
7e660872 DH |
2103 | |
2104 | /* search for VMAs that fall within the dead zone */ | |
6b2dbba8 | 2105 | vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, low, high) { |
7e660872 DH |
2106 | /* found one - only interested if it's shared out of the page |
2107 | * cache */ | |
2108 | if (vma->vm_flags & VM_SHARED) { | |
1acf2e04 | 2109 | i_mmap_unlock_read(inode->i_mapping); |
7e660872 DH |
2110 | up_write(&nommu_region_sem); |
2111 | return -ETXTBSY; /* not quite true, but near enough */ | |
2112 | } | |
2113 | } | |
2114 | ||
2115 | /* reduce any regions that overlap the dead zone - if in existence, | |
2116 | * these will be pointed to by VMAs that don't overlap the dead zone | |
2117 | * | |
2118 | * we don't check for any regions that start beyond the EOF as there | |
2119 | * shouldn't be any | |
2120 | */ | |
1acf2e04 | 2121 | vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, 0, ULONG_MAX) { |
7e660872 DH |
2122 | if (!(vma->vm_flags & VM_SHARED)) |
2123 | continue; | |
2124 | ||
2125 | region = vma->vm_region; | |
2126 | r_size = region->vm_top - region->vm_start; | |
2127 | r_top = (region->vm_pgoff << PAGE_SHIFT) + r_size; | |
2128 | ||
2129 | if (r_top > newsize) { | |
2130 | region->vm_top -= r_top - newsize; | |
2131 | if (region->vm_end > region->vm_top) | |
2132 | region->vm_end = region->vm_top; | |
2133 | } | |
2134 | } | |
2135 | ||
1acf2e04 | 2136 | i_mmap_unlock_read(inode->i_mapping); |
7e660872 DH |
2137 | up_write(&nommu_region_sem); |
2138 | return 0; | |
2139 | } | |
c9b1d098 AS |
2140 | |
2141 | /* | |
2142 | * Initialise sysctl_user_reserve_kbytes. | |
2143 | * | |
2144 | * This is intended to prevent a user from starting a single memory hogging | |
2145 | * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER | |
2146 | * mode. | |
2147 | * | |
2148 | * The default value is min(3% of free memory, 128MB) | |
2149 | * 128MB is enough to recover with sshd/login, bash, and top/kill. | |
2150 | */ | |
2151 | static int __meminit init_user_reserve(void) | |
2152 | { | |
2153 | unsigned long free_kbytes; | |
2154 | ||
2155 | free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); | |
2156 | ||
2157 | sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17); | |
2158 | return 0; | |
2159 | } | |
2160 | module_init(init_user_reserve) | |
4eeab4f5 AS |
2161 | |
2162 | /* | |
2163 | * Initialise sysctl_admin_reserve_kbytes. | |
2164 | * | |
2165 | * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin | |
2166 | * to log in and kill a memory hogging process. | |
2167 | * | |
2168 | * Systems with more than 256MB will reserve 8MB, enough to recover | |
2169 | * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will | |
2170 | * only reserve 3% of free pages by default. | |
2171 | */ | |
2172 | static int __meminit init_admin_reserve(void) | |
2173 | { | |
2174 | unsigned long free_kbytes; | |
2175 | ||
2176 | free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); | |
2177 | ||
2178 | sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13); | |
2179 | return 0; | |
2180 | } | |
2181 | module_init(init_admin_reserve) |