]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/mm/nommu.c | |
3 | * | |
4 | * Replacement code for mm functions to support CPU's that don't | |
5 | * have any form of memory management unit (thus no virtual memory). | |
6 | * | |
7 | * See Documentation/nommu-mmap.txt | |
8 | * | |
8feae131 | 9 | * Copyright (c) 2004-2008 David Howells <dhowells@redhat.com> |
1da177e4 LT |
10 | * Copyright (c) 2000-2003 David McCullough <davidm@snapgear.com> |
11 | * Copyright (c) 2000-2001 D Jeff Dionne <jeff@uClinux.org> | |
12 | * Copyright (c) 2002 Greg Ungerer <gerg@snapgear.com> | |
eb6434d9 | 13 | * Copyright (c) 2007-2009 Paul Mundt <lethal@linux-sh.org> |
1da177e4 LT |
14 | */ |
15 | ||
f2b8544f | 16 | #include <linux/module.h> |
1da177e4 LT |
17 | #include <linux/mm.h> |
18 | #include <linux/mman.h> | |
19 | #include <linux/swap.h> | |
20 | #include <linux/file.h> | |
21 | #include <linux/highmem.h> | |
22 | #include <linux/pagemap.h> | |
23 | #include <linux/slab.h> | |
24 | #include <linux/vmalloc.h> | |
fa8e26cc | 25 | #include <linux/tracehook.h> |
1da177e4 LT |
26 | #include <linux/blkdev.h> |
27 | #include <linux/backing-dev.h> | |
28 | #include <linux/mount.h> | |
29 | #include <linux/personality.h> | |
30 | #include <linux/security.h> | |
31 | #include <linux/syscalls.h> | |
32 | ||
33 | #include <asm/uaccess.h> | |
34 | #include <asm/tlb.h> | |
35 | #include <asm/tlbflush.h> | |
8feae131 DH |
36 | #include "internal.h" |
37 | ||
38 | static inline __attribute__((format(printf, 1, 2))) | |
39 | void no_printk(const char *fmt, ...) | |
40 | { | |
41 | } | |
42 | ||
43 | #if 0 | |
44 | #define kenter(FMT, ...) \ | |
45 | printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__) | |
46 | #define kleave(FMT, ...) \ | |
47 | printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__) | |
48 | #define kdebug(FMT, ...) \ | |
49 | printk(KERN_DEBUG "xxx" FMT"yyy\n", ##__VA_ARGS__) | |
50 | #else | |
51 | #define kenter(FMT, ...) \ | |
52 | no_printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__) | |
53 | #define kleave(FMT, ...) \ | |
54 | no_printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__) | |
55 | #define kdebug(FMT, ...) \ | |
56 | no_printk(KERN_DEBUG FMT"\n", ##__VA_ARGS__) | |
57 | #endif | |
1da177e4 | 58 | |
b291f000 NP |
59 | #include "internal.h" |
60 | ||
1da177e4 LT |
61 | void *high_memory; |
62 | struct page *mem_map; | |
63 | unsigned long max_mapnr; | |
64 | unsigned long num_physpages; | |
00a62ce9 | 65 | struct percpu_counter vm_committed_as; |
1da177e4 LT |
66 | int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */ |
67 | int sysctl_overcommit_ratio = 50; /* default is 50% */ | |
68 | int sysctl_max_map_count = DEFAULT_MAX_MAP_COUNT; | |
fc4d5c29 | 69 | int sysctl_nr_trim_pages = CONFIG_NOMMU_INITIAL_TRIM_EXCESS; |
1da177e4 LT |
70 | int heap_stack_gap = 0; |
71 | ||
33e5d769 | 72 | atomic_long_t mmap_pages_allocated; |
8feae131 | 73 | |
1da177e4 | 74 | EXPORT_SYMBOL(mem_map); |
6a04de6d | 75 | EXPORT_SYMBOL(num_physpages); |
1da177e4 | 76 | |
8feae131 DH |
77 | /* list of mapped, potentially shareable regions */ |
78 | static struct kmem_cache *vm_region_jar; | |
79 | struct rb_root nommu_region_tree = RB_ROOT; | |
80 | DECLARE_RWSEM(nommu_region_sem); | |
1da177e4 LT |
81 | |
82 | struct vm_operations_struct generic_file_vm_ops = { | |
83 | }; | |
84 | ||
85 | /* | |
86 | * Handle all mappings that got truncated by a "truncate()" | |
87 | * system call. | |
88 | * | |
89 | * NOTE! We have to be ready to update the memory sharing | |
90 | * between the file and the memory map for a potential last | |
91 | * incomplete page. Ugly, but necessary. | |
92 | */ | |
93 | int vmtruncate(struct inode *inode, loff_t offset) | |
94 | { | |
95 | struct address_space *mapping = inode->i_mapping; | |
96 | unsigned long limit; | |
97 | ||
98 | if (inode->i_size < offset) | |
99 | goto do_expand; | |
100 | i_size_write(inode, offset); | |
101 | ||
102 | truncate_inode_pages(mapping, offset); | |
103 | goto out_truncate; | |
104 | ||
105 | do_expand: | |
106 | limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur; | |
107 | if (limit != RLIM_INFINITY && offset > limit) | |
108 | goto out_sig; | |
109 | if (offset > inode->i_sb->s_maxbytes) | |
110 | goto out; | |
111 | i_size_write(inode, offset); | |
112 | ||
113 | out_truncate: | |
acfa4380 | 114 | if (inode->i_op->truncate) |
1da177e4 LT |
115 | inode->i_op->truncate(inode); |
116 | return 0; | |
117 | out_sig: | |
118 | send_sig(SIGXFSZ, current, 0); | |
119 | out: | |
120 | return -EFBIG; | |
121 | } | |
122 | ||
123 | EXPORT_SYMBOL(vmtruncate); | |
124 | ||
125 | /* | |
126 | * Return the total memory allocated for this pointer, not | |
127 | * just what the caller asked for. | |
128 | * | |
129 | * Doesn't have to be accurate, i.e. may have races. | |
130 | */ | |
131 | unsigned int kobjsize(const void *objp) | |
132 | { | |
133 | struct page *page; | |
134 | ||
4016a139 MH |
135 | /* |
136 | * If the object we have should not have ksize performed on it, | |
137 | * return size of 0 | |
138 | */ | |
5a1603be | 139 | if (!objp || !virt_addr_valid(objp)) |
6cfd53fc PM |
140 | return 0; |
141 | ||
142 | page = virt_to_head_page(objp); | |
6cfd53fc PM |
143 | |
144 | /* | |
145 | * If the allocator sets PageSlab, we know the pointer came from | |
146 | * kmalloc(). | |
147 | */ | |
1da177e4 LT |
148 | if (PageSlab(page)) |
149 | return ksize(objp); | |
150 | ||
ab2e83ea PM |
151 | /* |
152 | * If it's not a compound page, see if we have a matching VMA | |
153 | * region. This test is intentionally done in reverse order, | |
154 | * so if there's no VMA, we still fall through and hand back | |
155 | * PAGE_SIZE for 0-order pages. | |
156 | */ | |
157 | if (!PageCompound(page)) { | |
158 | struct vm_area_struct *vma; | |
159 | ||
160 | vma = find_vma(current->mm, (unsigned long)objp); | |
161 | if (vma) | |
162 | return vma->vm_end - vma->vm_start; | |
163 | } | |
164 | ||
6cfd53fc PM |
165 | /* |
166 | * The ksize() function is only guaranteed to work for pointers | |
5a1603be | 167 | * returned by kmalloc(). So handle arbitrary pointers here. |
6cfd53fc | 168 | */ |
5a1603be | 169 | return PAGE_SIZE << compound_order(page); |
1da177e4 LT |
170 | } |
171 | ||
b291f000 NP |
172 | int __get_user_pages(struct task_struct *tsk, struct mm_struct *mm, |
173 | unsigned long start, int len, int flags, | |
174 | struct page **pages, struct vm_area_struct **vmas) | |
1da177e4 | 175 | { |
910e46da | 176 | struct vm_area_struct *vma; |
7b4d5b8b DH |
177 | unsigned long vm_flags; |
178 | int i; | |
b291f000 NP |
179 | int write = !!(flags & GUP_FLAGS_WRITE); |
180 | int force = !!(flags & GUP_FLAGS_FORCE); | |
181 | int ignore = !!(flags & GUP_FLAGS_IGNORE_VMA_PERMISSIONS); | |
7b4d5b8b DH |
182 | |
183 | /* calculate required read or write permissions. | |
184 | * - if 'force' is set, we only require the "MAY" flags. | |
185 | */ | |
186 | vm_flags = write ? (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD); | |
187 | vm_flags &= force ? (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE); | |
1da177e4 LT |
188 | |
189 | for (i = 0; i < len; i++) { | |
910e46da | 190 | vma = find_vma(mm, start); |
7b4d5b8b DH |
191 | if (!vma) |
192 | goto finish_or_fault; | |
193 | ||
194 | /* protect what we can, including chardevs */ | |
195 | if (vma->vm_flags & (VM_IO | VM_PFNMAP) || | |
b291f000 | 196 | (!ignore && !(vm_flags & vma->vm_flags))) |
7b4d5b8b | 197 | goto finish_or_fault; |
910e46da | 198 | |
1da177e4 LT |
199 | if (pages) { |
200 | pages[i] = virt_to_page(start); | |
201 | if (pages[i]) | |
202 | page_cache_get(pages[i]); | |
203 | } | |
204 | if (vmas) | |
910e46da | 205 | vmas[i] = vma; |
1da177e4 LT |
206 | start += PAGE_SIZE; |
207 | } | |
7b4d5b8b DH |
208 | |
209 | return i; | |
210 | ||
211 | finish_or_fault: | |
212 | return i ? : -EFAULT; | |
1da177e4 | 213 | } |
b291f000 NP |
214 | |
215 | ||
216 | /* | |
217 | * get a list of pages in an address range belonging to the specified process | |
218 | * and indicate the VMA that covers each page | |
219 | * - this is potentially dodgy as we may end incrementing the page count of a | |
220 | * slab page or a secondary page from a compound page | |
221 | * - don't permit access to VMAs that don't support it, such as I/O mappings | |
222 | */ | |
223 | int get_user_pages(struct task_struct *tsk, struct mm_struct *mm, | |
224 | unsigned long start, int len, int write, int force, | |
225 | struct page **pages, struct vm_area_struct **vmas) | |
226 | { | |
227 | int flags = 0; | |
228 | ||
229 | if (write) | |
230 | flags |= GUP_FLAGS_WRITE; | |
231 | if (force) | |
232 | flags |= GUP_FLAGS_FORCE; | |
233 | ||
234 | return __get_user_pages(tsk, mm, | |
235 | start, len, flags, | |
236 | pages, vmas); | |
237 | } | |
66aa2b4b GU |
238 | EXPORT_SYMBOL(get_user_pages); |
239 | ||
1da177e4 LT |
240 | DEFINE_RWLOCK(vmlist_lock); |
241 | struct vm_struct *vmlist; | |
242 | ||
b3bdda02 | 243 | void vfree(const void *addr) |
1da177e4 LT |
244 | { |
245 | kfree(addr); | |
246 | } | |
b5073173 | 247 | EXPORT_SYMBOL(vfree); |
1da177e4 | 248 | |
dd0fc66f | 249 | void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot) |
1da177e4 LT |
250 | { |
251 | /* | |
8518609d RD |
252 | * You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc() |
253 | * returns only a logical address. | |
1da177e4 | 254 | */ |
84097518 | 255 | return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM); |
1da177e4 | 256 | } |
b5073173 | 257 | EXPORT_SYMBOL(__vmalloc); |
1da177e4 | 258 | |
f905bc44 PM |
259 | void *vmalloc_user(unsigned long size) |
260 | { | |
261 | void *ret; | |
262 | ||
263 | ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, | |
264 | PAGE_KERNEL); | |
265 | if (ret) { | |
266 | struct vm_area_struct *vma; | |
267 | ||
268 | down_write(¤t->mm->mmap_sem); | |
269 | vma = find_vma(current->mm, (unsigned long)ret); | |
270 | if (vma) | |
271 | vma->vm_flags |= VM_USERMAP; | |
272 | up_write(¤t->mm->mmap_sem); | |
273 | } | |
274 | ||
275 | return ret; | |
276 | } | |
277 | EXPORT_SYMBOL(vmalloc_user); | |
278 | ||
b3bdda02 | 279 | struct page *vmalloc_to_page(const void *addr) |
1da177e4 LT |
280 | { |
281 | return virt_to_page(addr); | |
282 | } | |
b5073173 | 283 | EXPORT_SYMBOL(vmalloc_to_page); |
1da177e4 | 284 | |
b3bdda02 | 285 | unsigned long vmalloc_to_pfn(const void *addr) |
1da177e4 LT |
286 | { |
287 | return page_to_pfn(virt_to_page(addr)); | |
288 | } | |
b5073173 | 289 | EXPORT_SYMBOL(vmalloc_to_pfn); |
1da177e4 LT |
290 | |
291 | long vread(char *buf, char *addr, unsigned long count) | |
292 | { | |
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); | |
304 | return(count); | |
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 | ||
324 | void *vmalloc_node(unsigned long size, int node) | |
325 | { | |
326 | return vmalloc(size); | |
327 | } | |
328 | EXPORT_SYMBOL(vmalloc_node); | |
1da177e4 | 329 | |
1af446ed PM |
330 | #ifndef PAGE_KERNEL_EXEC |
331 | # define PAGE_KERNEL_EXEC PAGE_KERNEL | |
332 | #endif | |
333 | ||
334 | /** | |
335 | * vmalloc_exec - allocate virtually contiguous, executable memory | |
336 | * @size: allocation size | |
337 | * | |
338 | * Kernel-internal function to allocate enough pages to cover @size | |
339 | * the page level allocator and map them into contiguous and | |
340 | * executable kernel virtual space. | |
341 | * | |
342 | * For tight control over page level allocator and protection flags | |
343 | * use __vmalloc() instead. | |
344 | */ | |
345 | ||
346 | void *vmalloc_exec(unsigned long size) | |
347 | { | |
348 | return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC); | |
349 | } | |
350 | ||
b5073173 PM |
351 | /** |
352 | * vmalloc_32 - allocate virtually contiguous memory (32bit addressable) | |
1da177e4 LT |
353 | * @size: allocation size |
354 | * | |
355 | * Allocate enough 32bit PA addressable pages to cover @size from the | |
356 | * page level allocator and map them into continguos kernel virtual space. | |
357 | */ | |
358 | void *vmalloc_32(unsigned long size) | |
359 | { | |
360 | return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL); | |
361 | } | |
b5073173 PM |
362 | EXPORT_SYMBOL(vmalloc_32); |
363 | ||
364 | /** | |
365 | * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory | |
366 | * @size: allocation size | |
367 | * | |
368 | * The resulting memory area is 32bit addressable and zeroed so it can be | |
369 | * mapped to userspace without leaking data. | |
f905bc44 PM |
370 | * |
371 | * VM_USERMAP is set on the corresponding VMA so that subsequent calls to | |
372 | * remap_vmalloc_range() are permissible. | |
b5073173 PM |
373 | */ |
374 | void *vmalloc_32_user(unsigned long size) | |
375 | { | |
f905bc44 PM |
376 | /* |
377 | * We'll have to sort out the ZONE_DMA bits for 64-bit, | |
378 | * but for now this can simply use vmalloc_user() directly. | |
379 | */ | |
380 | return vmalloc_user(size); | |
b5073173 PM |
381 | } |
382 | EXPORT_SYMBOL(vmalloc_32_user); | |
1da177e4 LT |
383 | |
384 | void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot) | |
385 | { | |
386 | BUG(); | |
387 | return NULL; | |
388 | } | |
b5073173 | 389 | EXPORT_SYMBOL(vmap); |
1da177e4 | 390 | |
b3bdda02 | 391 | void vunmap(const void *addr) |
1da177e4 LT |
392 | { |
393 | BUG(); | |
394 | } | |
b5073173 | 395 | EXPORT_SYMBOL(vunmap); |
1da177e4 | 396 | |
eb6434d9 PM |
397 | void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot) |
398 | { | |
399 | BUG(); | |
400 | return NULL; | |
401 | } | |
402 | EXPORT_SYMBOL(vm_map_ram); | |
403 | ||
404 | void vm_unmap_ram(const void *mem, unsigned int count) | |
405 | { | |
406 | BUG(); | |
407 | } | |
408 | EXPORT_SYMBOL(vm_unmap_ram); | |
409 | ||
410 | void vm_unmap_aliases(void) | |
411 | { | |
412 | } | |
413 | EXPORT_SYMBOL_GPL(vm_unmap_aliases); | |
414 | ||
1eeb66a1 CH |
415 | /* |
416 | * Implement a stub for vmalloc_sync_all() if the architecture chose not to | |
417 | * have one. | |
418 | */ | |
419 | void __attribute__((weak)) vmalloc_sync_all(void) | |
420 | { | |
421 | } | |
422 | ||
b5073173 PM |
423 | int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, |
424 | struct page *page) | |
425 | { | |
426 | return -EINVAL; | |
427 | } | |
428 | EXPORT_SYMBOL(vm_insert_page); | |
429 | ||
1da177e4 LT |
430 | /* |
431 | * sys_brk() for the most part doesn't need the global kernel | |
432 | * lock, except when an application is doing something nasty | |
433 | * like trying to un-brk an area that has already been mapped | |
434 | * to a regular file. in this case, the unmapping will need | |
435 | * to invoke file system routines that need the global lock. | |
436 | */ | |
6a6160a7 | 437 | SYSCALL_DEFINE1(brk, unsigned long, brk) |
1da177e4 LT |
438 | { |
439 | struct mm_struct *mm = current->mm; | |
440 | ||
441 | if (brk < mm->start_brk || brk > mm->context.end_brk) | |
442 | return mm->brk; | |
443 | ||
444 | if (mm->brk == brk) | |
445 | return mm->brk; | |
446 | ||
447 | /* | |
448 | * Always allow shrinking brk | |
449 | */ | |
450 | if (brk <= mm->brk) { | |
451 | mm->brk = brk; | |
452 | return brk; | |
453 | } | |
454 | ||
455 | /* | |
456 | * Ok, looks good - let it rip. | |
457 | */ | |
458 | return mm->brk = brk; | |
459 | } | |
460 | ||
8feae131 DH |
461 | /* |
462 | * initialise the VMA and region record slabs | |
463 | */ | |
464 | void __init mmap_init(void) | |
1da177e4 | 465 | { |
00a62ce9 KM |
466 | int ret; |
467 | ||
468 | ret = percpu_counter_init(&vm_committed_as, 0); | |
469 | VM_BUG_ON(ret); | |
33e5d769 | 470 | vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC); |
1da177e4 | 471 | } |
1da177e4 | 472 | |
3034097a | 473 | /* |
8feae131 DH |
474 | * validate the region tree |
475 | * - the caller must hold the region lock | |
3034097a | 476 | */ |
8feae131 DH |
477 | #ifdef CONFIG_DEBUG_NOMMU_REGIONS |
478 | static noinline void validate_nommu_regions(void) | |
3034097a | 479 | { |
8feae131 DH |
480 | struct vm_region *region, *last; |
481 | struct rb_node *p, *lastp; | |
3034097a | 482 | |
8feae131 DH |
483 | lastp = rb_first(&nommu_region_tree); |
484 | if (!lastp) | |
485 | return; | |
486 | ||
487 | last = rb_entry(lastp, struct vm_region, vm_rb); | |
33e5d769 DH |
488 | BUG_ON(unlikely(last->vm_end <= last->vm_start)); |
489 | BUG_ON(unlikely(last->vm_top < last->vm_end)); | |
8feae131 DH |
490 | |
491 | while ((p = rb_next(lastp))) { | |
492 | region = rb_entry(p, struct vm_region, vm_rb); | |
493 | last = rb_entry(lastp, struct vm_region, vm_rb); | |
494 | ||
33e5d769 DH |
495 | BUG_ON(unlikely(region->vm_end <= region->vm_start)); |
496 | BUG_ON(unlikely(region->vm_top < region->vm_end)); | |
497 | BUG_ON(unlikely(region->vm_start < last->vm_top)); | |
3034097a | 498 | |
8feae131 DH |
499 | lastp = p; |
500 | } | |
3034097a | 501 | } |
8feae131 | 502 | #else |
33e5d769 DH |
503 | static void validate_nommu_regions(void) |
504 | { | |
505 | } | |
8feae131 | 506 | #endif |
3034097a DH |
507 | |
508 | /* | |
8feae131 | 509 | * add a region into the global tree |
3034097a | 510 | */ |
8feae131 | 511 | static void add_nommu_region(struct vm_region *region) |
3034097a | 512 | { |
8feae131 DH |
513 | struct vm_region *pregion; |
514 | struct rb_node **p, *parent; | |
3034097a | 515 | |
8feae131 DH |
516 | validate_nommu_regions(); |
517 | ||
518 | BUG_ON(region->vm_start & ~PAGE_MASK); | |
519 | ||
520 | parent = NULL; | |
521 | p = &nommu_region_tree.rb_node; | |
522 | while (*p) { | |
523 | parent = *p; | |
524 | pregion = rb_entry(parent, struct vm_region, vm_rb); | |
525 | if (region->vm_start < pregion->vm_start) | |
526 | p = &(*p)->rb_left; | |
527 | else if (region->vm_start > pregion->vm_start) | |
528 | p = &(*p)->rb_right; | |
529 | else if (pregion == region) | |
530 | return; | |
531 | else | |
532 | BUG(); | |
3034097a DH |
533 | } |
534 | ||
8feae131 DH |
535 | rb_link_node(®ion->vm_rb, parent, p); |
536 | rb_insert_color(®ion->vm_rb, &nommu_region_tree); | |
3034097a | 537 | |
8feae131 | 538 | validate_nommu_regions(); |
3034097a | 539 | } |
3034097a | 540 | |
930e652a | 541 | /* |
8feae131 | 542 | * delete a region from the global tree |
930e652a | 543 | */ |
8feae131 | 544 | static void delete_nommu_region(struct vm_region *region) |
930e652a | 545 | { |
8feae131 | 546 | BUG_ON(!nommu_region_tree.rb_node); |
930e652a | 547 | |
8feae131 DH |
548 | validate_nommu_regions(); |
549 | rb_erase(®ion->vm_rb, &nommu_region_tree); | |
550 | validate_nommu_regions(); | |
57c8f63e GU |
551 | } |
552 | ||
6fa5f80b | 553 | /* |
8feae131 | 554 | * free a contiguous series of pages |
6fa5f80b | 555 | */ |
8feae131 | 556 | static void free_page_series(unsigned long from, unsigned long to) |
6fa5f80b | 557 | { |
8feae131 DH |
558 | for (; from < to; from += PAGE_SIZE) { |
559 | struct page *page = virt_to_page(from); | |
560 | ||
561 | kdebug("- free %lx", from); | |
33e5d769 | 562 | atomic_long_dec(&mmap_pages_allocated); |
8feae131 | 563 | if (page_count(page) != 1) |
33e5d769 DH |
564 | kdebug("free page %p: refcount not one: %d", |
565 | page, page_count(page)); | |
8feae131 | 566 | put_page(page); |
6fa5f80b | 567 | } |
6fa5f80b DH |
568 | } |
569 | ||
3034097a | 570 | /* |
8feae131 | 571 | * release a reference to a region |
33e5d769 | 572 | * - the caller must hold the region semaphore for writing, which this releases |
dd8632a1 | 573 | * - the region may not have been added to the tree yet, in which case vm_top |
8feae131 | 574 | * will equal vm_start |
3034097a | 575 | */ |
8feae131 DH |
576 | static void __put_nommu_region(struct vm_region *region) |
577 | __releases(nommu_region_sem) | |
1da177e4 | 578 | { |
8feae131 | 579 | kenter("%p{%d}", region, atomic_read(®ion->vm_usage)); |
1da177e4 | 580 | |
8feae131 | 581 | BUG_ON(!nommu_region_tree.rb_node); |
1da177e4 | 582 | |
8feae131 | 583 | if (atomic_dec_and_test(®ion->vm_usage)) { |
dd8632a1 | 584 | if (region->vm_top > region->vm_start) |
8feae131 DH |
585 | delete_nommu_region(region); |
586 | up_write(&nommu_region_sem); | |
587 | ||
588 | if (region->vm_file) | |
589 | fput(region->vm_file); | |
590 | ||
591 | /* IO memory and memory shared directly out of the pagecache | |
592 | * from ramfs/tmpfs mustn't be released here */ | |
593 | if (region->vm_flags & VM_MAPPED_COPY) { | |
594 | kdebug("free series"); | |
dd8632a1 | 595 | free_page_series(region->vm_start, region->vm_top); |
8feae131 DH |
596 | } |
597 | kmem_cache_free(vm_region_jar, region); | |
598 | } else { | |
599 | up_write(&nommu_region_sem); | |
1da177e4 | 600 | } |
8feae131 | 601 | } |
1da177e4 | 602 | |
8feae131 DH |
603 | /* |
604 | * release a reference to a region | |
605 | */ | |
606 | static void put_nommu_region(struct vm_region *region) | |
607 | { | |
608 | down_write(&nommu_region_sem); | |
609 | __put_nommu_region(region); | |
1da177e4 LT |
610 | } |
611 | ||
3034097a | 612 | /* |
8feae131 DH |
613 | * add a VMA into a process's mm_struct in the appropriate place in the list |
614 | * and tree and add to the address space's page tree also if not an anonymous | |
615 | * page | |
616 | * - should be called with mm->mmap_sem held writelocked | |
3034097a | 617 | */ |
8feae131 | 618 | static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma) |
1da177e4 | 619 | { |
8feae131 | 620 | struct vm_area_struct *pvma, **pp; |
1da177e4 | 621 | struct address_space *mapping; |
8feae131 DH |
622 | struct rb_node **p, *parent; |
623 | ||
624 | kenter(",%p", vma); | |
625 | ||
626 | BUG_ON(!vma->vm_region); | |
627 | ||
628 | mm->map_count++; | |
629 | vma->vm_mm = mm; | |
1da177e4 LT |
630 | |
631 | /* add the VMA to the mapping */ | |
632 | if (vma->vm_file) { | |
633 | mapping = vma->vm_file->f_mapping; | |
634 | ||
635 | flush_dcache_mmap_lock(mapping); | |
636 | vma_prio_tree_insert(vma, &mapping->i_mmap); | |
637 | flush_dcache_mmap_unlock(mapping); | |
638 | } | |
639 | ||
8feae131 DH |
640 | /* add the VMA to the tree */ |
641 | parent = NULL; | |
642 | p = &mm->mm_rb.rb_node; | |
1da177e4 LT |
643 | while (*p) { |
644 | parent = *p; | |
645 | pvma = rb_entry(parent, struct vm_area_struct, vm_rb); | |
646 | ||
8feae131 DH |
647 | /* sort by: start addr, end addr, VMA struct addr in that order |
648 | * (the latter is necessary as we may get identical VMAs) */ | |
649 | if (vma->vm_start < pvma->vm_start) | |
1da177e4 | 650 | p = &(*p)->rb_left; |
8feae131 | 651 | else if (vma->vm_start > pvma->vm_start) |
1da177e4 | 652 | p = &(*p)->rb_right; |
8feae131 DH |
653 | else if (vma->vm_end < pvma->vm_end) |
654 | p = &(*p)->rb_left; | |
655 | else if (vma->vm_end > pvma->vm_end) | |
656 | p = &(*p)->rb_right; | |
657 | else if (vma < pvma) | |
658 | p = &(*p)->rb_left; | |
659 | else if (vma > pvma) | |
660 | p = &(*p)->rb_right; | |
661 | else | |
662 | BUG(); | |
1da177e4 LT |
663 | } |
664 | ||
665 | rb_link_node(&vma->vm_rb, parent, p); | |
8feae131 DH |
666 | rb_insert_color(&vma->vm_rb, &mm->mm_rb); |
667 | ||
668 | /* add VMA to the VMA list also */ | |
669 | for (pp = &mm->mmap; (pvma = *pp); pp = &(*pp)->vm_next) { | |
670 | if (pvma->vm_start > vma->vm_start) | |
671 | break; | |
672 | if (pvma->vm_start < vma->vm_start) | |
673 | continue; | |
674 | if (pvma->vm_end < vma->vm_end) | |
675 | break; | |
676 | } | |
677 | ||
678 | vma->vm_next = *pp; | |
679 | *pp = vma; | |
1da177e4 LT |
680 | } |
681 | ||
3034097a | 682 | /* |
8feae131 | 683 | * delete a VMA from its owning mm_struct and address space |
3034097a | 684 | */ |
8feae131 | 685 | static void delete_vma_from_mm(struct vm_area_struct *vma) |
1da177e4 | 686 | { |
8feae131 | 687 | struct vm_area_struct **pp; |
1da177e4 | 688 | struct address_space *mapping; |
8feae131 DH |
689 | struct mm_struct *mm = vma->vm_mm; |
690 | ||
691 | kenter("%p", vma); | |
692 | ||
693 | mm->map_count--; | |
694 | if (mm->mmap_cache == vma) | |
695 | mm->mmap_cache = NULL; | |
1da177e4 LT |
696 | |
697 | /* remove the VMA from the mapping */ | |
698 | if (vma->vm_file) { | |
699 | mapping = vma->vm_file->f_mapping; | |
700 | ||
701 | flush_dcache_mmap_lock(mapping); | |
702 | vma_prio_tree_remove(vma, &mapping->i_mmap); | |
703 | flush_dcache_mmap_unlock(mapping); | |
704 | } | |
705 | ||
8feae131 DH |
706 | /* remove from the MM's tree and list */ |
707 | rb_erase(&vma->vm_rb, &mm->mm_rb); | |
708 | for (pp = &mm->mmap; *pp; pp = &(*pp)->vm_next) { | |
709 | if (*pp == vma) { | |
710 | *pp = vma->vm_next; | |
711 | break; | |
712 | } | |
713 | } | |
714 | ||
715 | vma->vm_mm = NULL; | |
716 | } | |
717 | ||
718 | /* | |
719 | * destroy a VMA record | |
720 | */ | |
721 | static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma) | |
722 | { | |
723 | kenter("%p", vma); | |
724 | if (vma->vm_ops && vma->vm_ops->close) | |
725 | vma->vm_ops->close(vma); | |
726 | if (vma->vm_file) { | |
727 | fput(vma->vm_file); | |
728 | if (vma->vm_flags & VM_EXECUTABLE) | |
729 | removed_exe_file_vma(mm); | |
730 | } | |
731 | put_nommu_region(vma->vm_region); | |
732 | kmem_cache_free(vm_area_cachep, vma); | |
733 | } | |
734 | ||
735 | /* | |
736 | * look up the first VMA in which addr resides, NULL if none | |
737 | * - should be called with mm->mmap_sem at least held readlocked | |
738 | */ | |
739 | struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr) | |
740 | { | |
741 | struct vm_area_struct *vma; | |
742 | struct rb_node *n = mm->mm_rb.rb_node; | |
743 | ||
744 | /* check the cache first */ | |
745 | vma = mm->mmap_cache; | |
746 | if (vma && vma->vm_start <= addr && vma->vm_end > addr) | |
747 | return vma; | |
748 | ||
749 | /* trawl the tree (there may be multiple mappings in which addr | |
750 | * resides) */ | |
751 | for (n = rb_first(&mm->mm_rb); n; n = rb_next(n)) { | |
752 | vma = rb_entry(n, struct vm_area_struct, vm_rb); | |
753 | if (vma->vm_start > addr) | |
754 | return NULL; | |
755 | if (vma->vm_end > addr) { | |
756 | mm->mmap_cache = vma; | |
757 | return vma; | |
758 | } | |
759 | } | |
760 | ||
761 | return NULL; | |
762 | } | |
763 | EXPORT_SYMBOL(find_vma); | |
764 | ||
765 | /* | |
766 | * find a VMA | |
767 | * - we don't extend stack VMAs under NOMMU conditions | |
768 | */ | |
769 | struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr) | |
770 | { | |
771 | return find_vma(mm, addr); | |
772 | } | |
773 | ||
774 | /* | |
775 | * expand a stack to a given address | |
776 | * - not supported under NOMMU conditions | |
777 | */ | |
778 | int expand_stack(struct vm_area_struct *vma, unsigned long address) | |
779 | { | |
780 | return -ENOMEM; | |
781 | } | |
782 | ||
783 | /* | |
784 | * look up the first VMA exactly that exactly matches addr | |
785 | * - should be called with mm->mmap_sem at least held readlocked | |
786 | */ | |
787 | static struct vm_area_struct *find_vma_exact(struct mm_struct *mm, | |
788 | unsigned long addr, | |
789 | unsigned long len) | |
790 | { | |
791 | struct vm_area_struct *vma; | |
792 | struct rb_node *n = mm->mm_rb.rb_node; | |
793 | unsigned long end = addr + len; | |
794 | ||
795 | /* check the cache first */ | |
796 | vma = mm->mmap_cache; | |
797 | if (vma && vma->vm_start == addr && vma->vm_end == end) | |
798 | return vma; | |
799 | ||
800 | /* trawl the tree (there may be multiple mappings in which addr | |
801 | * resides) */ | |
802 | for (n = rb_first(&mm->mm_rb); n; n = rb_next(n)) { | |
803 | vma = rb_entry(n, struct vm_area_struct, vm_rb); | |
804 | if (vma->vm_start < addr) | |
805 | continue; | |
806 | if (vma->vm_start > addr) | |
807 | return NULL; | |
808 | if (vma->vm_end == end) { | |
809 | mm->mmap_cache = vma; | |
810 | return vma; | |
811 | } | |
812 | } | |
813 | ||
814 | return NULL; | |
1da177e4 LT |
815 | } |
816 | ||
817 | /* | |
818 | * determine whether a mapping should be permitted and, if so, what sort of | |
819 | * mapping we're capable of supporting | |
820 | */ | |
821 | static int validate_mmap_request(struct file *file, | |
822 | unsigned long addr, | |
823 | unsigned long len, | |
824 | unsigned long prot, | |
825 | unsigned long flags, | |
826 | unsigned long pgoff, | |
827 | unsigned long *_capabilities) | |
828 | { | |
8feae131 | 829 | unsigned long capabilities, rlen; |
1da177e4 LT |
830 | unsigned long reqprot = prot; |
831 | int ret; | |
832 | ||
833 | /* do the simple checks first */ | |
834 | if (flags & MAP_FIXED || addr) { | |
835 | printk(KERN_DEBUG | |
836 | "%d: Can't do fixed-address/overlay mmap of RAM\n", | |
837 | current->pid); | |
838 | return -EINVAL; | |
839 | } | |
840 | ||
841 | if ((flags & MAP_TYPE) != MAP_PRIVATE && | |
842 | (flags & MAP_TYPE) != MAP_SHARED) | |
843 | return -EINVAL; | |
844 | ||
f81cff0d | 845 | if (!len) |
1da177e4 LT |
846 | return -EINVAL; |
847 | ||
f81cff0d | 848 | /* Careful about overflows.. */ |
8feae131 DH |
849 | rlen = PAGE_ALIGN(len); |
850 | if (!rlen || rlen > TASK_SIZE) | |
f81cff0d MF |
851 | return -ENOMEM; |
852 | ||
1da177e4 | 853 | /* offset overflow? */ |
8feae131 | 854 | if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff) |
f81cff0d | 855 | return -EOVERFLOW; |
1da177e4 LT |
856 | |
857 | if (file) { | |
858 | /* validate file mapping requests */ | |
859 | struct address_space *mapping; | |
860 | ||
861 | /* files must support mmap */ | |
862 | if (!file->f_op || !file->f_op->mmap) | |
863 | return -ENODEV; | |
864 | ||
865 | /* work out if what we've got could possibly be shared | |
866 | * - we support chardevs that provide their own "memory" | |
867 | * - we support files/blockdevs that are memory backed | |
868 | */ | |
869 | mapping = file->f_mapping; | |
870 | if (!mapping) | |
e9536ae7 | 871 | mapping = file->f_path.dentry->d_inode->i_mapping; |
1da177e4 LT |
872 | |
873 | capabilities = 0; | |
874 | if (mapping && mapping->backing_dev_info) | |
875 | capabilities = mapping->backing_dev_info->capabilities; | |
876 | ||
877 | if (!capabilities) { | |
878 | /* no explicit capabilities set, so assume some | |
879 | * defaults */ | |
e9536ae7 | 880 | switch (file->f_path.dentry->d_inode->i_mode & S_IFMT) { |
1da177e4 LT |
881 | case S_IFREG: |
882 | case S_IFBLK: | |
883 | capabilities = BDI_CAP_MAP_COPY; | |
884 | break; | |
885 | ||
886 | case S_IFCHR: | |
887 | capabilities = | |
888 | BDI_CAP_MAP_DIRECT | | |
889 | BDI_CAP_READ_MAP | | |
890 | BDI_CAP_WRITE_MAP; | |
891 | break; | |
892 | ||
893 | default: | |
894 | return -EINVAL; | |
895 | } | |
896 | } | |
897 | ||
898 | /* eliminate any capabilities that we can't support on this | |
899 | * device */ | |
900 | if (!file->f_op->get_unmapped_area) | |
901 | capabilities &= ~BDI_CAP_MAP_DIRECT; | |
902 | if (!file->f_op->read) | |
903 | capabilities &= ~BDI_CAP_MAP_COPY; | |
904 | ||
905 | if (flags & MAP_SHARED) { | |
906 | /* do checks for writing, appending and locking */ | |
907 | if ((prot & PROT_WRITE) && | |
908 | !(file->f_mode & FMODE_WRITE)) | |
909 | return -EACCES; | |
910 | ||
e9536ae7 | 911 | if (IS_APPEND(file->f_path.dentry->d_inode) && |
1da177e4 LT |
912 | (file->f_mode & FMODE_WRITE)) |
913 | return -EACCES; | |
914 | ||
e9536ae7 | 915 | if (locks_verify_locked(file->f_path.dentry->d_inode)) |
1da177e4 LT |
916 | return -EAGAIN; |
917 | ||
918 | if (!(capabilities & BDI_CAP_MAP_DIRECT)) | |
919 | return -ENODEV; | |
920 | ||
921 | if (((prot & PROT_READ) && !(capabilities & BDI_CAP_READ_MAP)) || | |
922 | ((prot & PROT_WRITE) && !(capabilities & BDI_CAP_WRITE_MAP)) || | |
923 | ((prot & PROT_EXEC) && !(capabilities & BDI_CAP_EXEC_MAP)) | |
924 | ) { | |
925 | printk("MAP_SHARED not completely supported on !MMU\n"); | |
926 | return -EINVAL; | |
927 | } | |
928 | ||
929 | /* we mustn't privatise shared mappings */ | |
930 | capabilities &= ~BDI_CAP_MAP_COPY; | |
931 | } | |
932 | else { | |
933 | /* we're going to read the file into private memory we | |
934 | * allocate */ | |
935 | if (!(capabilities & BDI_CAP_MAP_COPY)) | |
936 | return -ENODEV; | |
937 | ||
938 | /* we don't permit a private writable mapping to be | |
939 | * shared with the backing device */ | |
940 | if (prot & PROT_WRITE) | |
941 | capabilities &= ~BDI_CAP_MAP_DIRECT; | |
942 | } | |
943 | ||
944 | /* handle executable mappings and implied executable | |
945 | * mappings */ | |
e9536ae7 | 946 | if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) { |
1da177e4 LT |
947 | if (prot & PROT_EXEC) |
948 | return -EPERM; | |
949 | } | |
950 | else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) { | |
951 | /* handle implication of PROT_EXEC by PROT_READ */ | |
952 | if (current->personality & READ_IMPLIES_EXEC) { | |
953 | if (capabilities & BDI_CAP_EXEC_MAP) | |
954 | prot |= PROT_EXEC; | |
955 | } | |
956 | } | |
957 | else if ((prot & PROT_READ) && | |
958 | (prot & PROT_EXEC) && | |
959 | !(capabilities & BDI_CAP_EXEC_MAP) | |
960 | ) { | |
961 | /* backing file is not executable, try to copy */ | |
962 | capabilities &= ~BDI_CAP_MAP_DIRECT; | |
963 | } | |
964 | } | |
965 | else { | |
966 | /* anonymous mappings are always memory backed and can be | |
967 | * privately mapped | |
968 | */ | |
969 | capabilities = BDI_CAP_MAP_COPY; | |
970 | ||
971 | /* handle PROT_EXEC implication by PROT_READ */ | |
972 | if ((prot & PROT_READ) && | |
973 | (current->personality & READ_IMPLIES_EXEC)) | |
974 | prot |= PROT_EXEC; | |
975 | } | |
976 | ||
977 | /* allow the security API to have its say */ | |
ed032189 | 978 | ret = security_file_mmap(file, reqprot, prot, flags, addr, 0); |
1da177e4 LT |
979 | if (ret < 0) |
980 | return ret; | |
981 | ||
982 | /* looks okay */ | |
983 | *_capabilities = capabilities; | |
984 | return 0; | |
985 | } | |
986 | ||
987 | /* | |
988 | * we've determined that we can make the mapping, now translate what we | |
989 | * now know into VMA flags | |
990 | */ | |
991 | static unsigned long determine_vm_flags(struct file *file, | |
992 | unsigned long prot, | |
993 | unsigned long flags, | |
994 | unsigned long capabilities) | |
995 | { | |
996 | unsigned long vm_flags; | |
997 | ||
998 | vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags); | |
999 | vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC; | |
1000 | /* vm_flags |= mm->def_flags; */ | |
1001 | ||
1002 | if (!(capabilities & BDI_CAP_MAP_DIRECT)) { | |
1003 | /* attempt to share read-only copies of mapped file chunks */ | |
1004 | if (file && !(prot & PROT_WRITE)) | |
1005 | vm_flags |= VM_MAYSHARE; | |
1006 | } | |
1007 | else { | |
1008 | /* overlay a shareable mapping on the backing device or inode | |
1009 | * if possible - used for chardevs, ramfs/tmpfs/shmfs and | |
1010 | * romfs/cramfs */ | |
1011 | if (flags & MAP_SHARED) | |
1012 | vm_flags |= VM_MAYSHARE | VM_SHARED; | |
1013 | else if ((((vm_flags & capabilities) ^ vm_flags) & BDI_CAP_VMFLAGS) == 0) | |
1014 | vm_flags |= VM_MAYSHARE; | |
1015 | } | |
1016 | ||
1017 | /* refuse to let anyone share private mappings with this process if | |
1018 | * it's being traced - otherwise breakpoints set in it may interfere | |
1019 | * with another untraced process | |
1020 | */ | |
fa8e26cc | 1021 | if ((flags & MAP_PRIVATE) && tracehook_expect_breakpoints(current)) |
1da177e4 LT |
1022 | vm_flags &= ~VM_MAYSHARE; |
1023 | ||
1024 | return vm_flags; | |
1025 | } | |
1026 | ||
1027 | /* | |
8feae131 DH |
1028 | * set up a shared mapping on a file (the driver or filesystem provides and |
1029 | * pins the storage) | |
1da177e4 | 1030 | */ |
8feae131 | 1031 | static int do_mmap_shared_file(struct vm_area_struct *vma) |
1da177e4 LT |
1032 | { |
1033 | int ret; | |
1034 | ||
1035 | ret = vma->vm_file->f_op->mmap(vma->vm_file, vma); | |
dd8632a1 PM |
1036 | if (ret == 0) { |
1037 | vma->vm_region->vm_top = vma->vm_region->vm_end; | |
1038 | return ret; | |
1039 | } | |
1da177e4 LT |
1040 | if (ret != -ENOSYS) |
1041 | return ret; | |
1042 | ||
1043 | /* getting an ENOSYS error indicates that direct mmap isn't | |
1044 | * possible (as opposed to tried but failed) so we'll fall | |
1045 | * through to making a private copy of the data and mapping | |
1046 | * that if we can */ | |
1047 | return -ENODEV; | |
1048 | } | |
1049 | ||
1050 | /* | |
1051 | * set up a private mapping or an anonymous shared mapping | |
1052 | */ | |
8feae131 DH |
1053 | static int do_mmap_private(struct vm_area_struct *vma, |
1054 | struct vm_region *region, | |
1055 | unsigned long len) | |
1da177e4 | 1056 | { |
8feae131 DH |
1057 | struct page *pages; |
1058 | unsigned long total, point, n, rlen; | |
1da177e4 | 1059 | void *base; |
8feae131 | 1060 | int ret, order; |
1da177e4 LT |
1061 | |
1062 | /* invoke the file's mapping function so that it can keep track of | |
1063 | * shared mappings on devices or memory | |
1064 | * - VM_MAYSHARE will be set if it may attempt to share | |
1065 | */ | |
1066 | if (vma->vm_file) { | |
1067 | ret = vma->vm_file->f_op->mmap(vma->vm_file, vma); | |
dd8632a1 | 1068 | if (ret == 0) { |
1da177e4 | 1069 | /* shouldn't return success if we're not sharing */ |
dd8632a1 PM |
1070 | BUG_ON(!(vma->vm_flags & VM_MAYSHARE)); |
1071 | vma->vm_region->vm_top = vma->vm_region->vm_end; | |
1072 | return ret; | |
1da177e4 | 1073 | } |
dd8632a1 PM |
1074 | if (ret != -ENOSYS) |
1075 | return ret; | |
1da177e4 LT |
1076 | |
1077 | /* getting an ENOSYS error indicates that direct mmap isn't | |
1078 | * possible (as opposed to tried but failed) so we'll try to | |
1079 | * make a private copy of the data and map that instead */ | |
1080 | } | |
1081 | ||
8feae131 DH |
1082 | rlen = PAGE_ALIGN(len); |
1083 | ||
1da177e4 LT |
1084 | /* allocate some memory to hold the mapping |
1085 | * - note that this may not return a page-aligned address if the object | |
1086 | * we're allocating is smaller than a page | |
1087 | */ | |
8feae131 DH |
1088 | order = get_order(rlen); |
1089 | kdebug("alloc order %d for %lx", order, len); | |
1090 | ||
1091 | pages = alloc_pages(GFP_KERNEL, order); | |
1092 | if (!pages) | |
1da177e4 LT |
1093 | goto enomem; |
1094 | ||
8feae131 | 1095 | total = 1 << order; |
33e5d769 | 1096 | atomic_long_add(total, &mmap_pages_allocated); |
8feae131 DH |
1097 | |
1098 | point = rlen >> PAGE_SHIFT; | |
dd8632a1 PM |
1099 | |
1100 | /* we allocated a power-of-2 sized page set, so we may want to trim off | |
1101 | * the excess */ | |
1102 | if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages) { | |
1103 | while (total > point) { | |
1104 | order = ilog2(total - point); | |
1105 | n = 1 << order; | |
1106 | kdebug("shave %lu/%lu @%lu", n, total - point, total); | |
33e5d769 | 1107 | atomic_long_sub(n, &mmap_pages_allocated); |
dd8632a1 PM |
1108 | total -= n; |
1109 | set_page_refcounted(pages + total); | |
1110 | __free_pages(pages + total, order); | |
1111 | } | |
8feae131 DH |
1112 | } |
1113 | ||
8feae131 DH |
1114 | for (point = 1; point < total; point++) |
1115 | set_page_refcounted(&pages[point]); | |
1da177e4 | 1116 | |
8feae131 DH |
1117 | base = page_address(pages); |
1118 | region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY; | |
1119 | region->vm_start = (unsigned long) base; | |
1120 | region->vm_end = region->vm_start + rlen; | |
dd8632a1 | 1121 | region->vm_top = region->vm_start + (total << PAGE_SHIFT); |
8feae131 DH |
1122 | |
1123 | vma->vm_start = region->vm_start; | |
1124 | vma->vm_end = region->vm_start + len; | |
1da177e4 LT |
1125 | |
1126 | if (vma->vm_file) { | |
1127 | /* read the contents of a file into the copy */ | |
1128 | mm_segment_t old_fs; | |
1129 | loff_t fpos; | |
1130 | ||
1131 | fpos = vma->vm_pgoff; | |
1132 | fpos <<= PAGE_SHIFT; | |
1133 | ||
1134 | old_fs = get_fs(); | |
1135 | set_fs(KERNEL_DS); | |
8feae131 | 1136 | ret = vma->vm_file->f_op->read(vma->vm_file, base, rlen, &fpos); |
1da177e4 LT |
1137 | set_fs(old_fs); |
1138 | ||
1139 | if (ret < 0) | |
1140 | goto error_free; | |
1141 | ||
1142 | /* clear the last little bit */ | |
8feae131 DH |
1143 | if (ret < rlen) |
1144 | memset(base + ret, 0, rlen - ret); | |
1da177e4 LT |
1145 | |
1146 | } else { | |
1147 | /* if it's an anonymous mapping, then just clear it */ | |
8feae131 | 1148 | memset(base, 0, rlen); |
1da177e4 LT |
1149 | } |
1150 | ||
1151 | return 0; | |
1152 | ||
1153 | error_free: | |
8feae131 DH |
1154 | free_page_series(region->vm_start, region->vm_end); |
1155 | region->vm_start = vma->vm_start = 0; | |
1156 | region->vm_end = vma->vm_end = 0; | |
dd8632a1 | 1157 | region->vm_top = 0; |
1da177e4 LT |
1158 | return ret; |
1159 | ||
1160 | enomem: | |
05ae6fa3 GU |
1161 | printk("Allocation of length %lu from process %d (%s) failed\n", |
1162 | len, current->pid, current->comm); | |
1da177e4 LT |
1163 | show_free_areas(); |
1164 | return -ENOMEM; | |
1165 | } | |
1166 | ||
1167 | /* | |
1168 | * handle mapping creation for uClinux | |
1169 | */ | |
1170 | unsigned long do_mmap_pgoff(struct file *file, | |
1171 | unsigned long addr, | |
1172 | unsigned long len, | |
1173 | unsigned long prot, | |
1174 | unsigned long flags, | |
1175 | unsigned long pgoff) | |
1176 | { | |
8feae131 DH |
1177 | struct vm_area_struct *vma; |
1178 | struct vm_region *region; | |
1da177e4 | 1179 | struct rb_node *rb; |
8feae131 | 1180 | unsigned long capabilities, vm_flags, result; |
1da177e4 LT |
1181 | int ret; |
1182 | ||
8feae131 DH |
1183 | kenter(",%lx,%lx,%lx,%lx,%lx", addr, len, prot, flags, pgoff); |
1184 | ||
7cd94146 EP |
1185 | if (!(flags & MAP_FIXED)) |
1186 | addr = round_hint_to_min(addr); | |
1187 | ||
1da177e4 LT |
1188 | /* decide whether we should attempt the mapping, and if so what sort of |
1189 | * mapping */ | |
1190 | ret = validate_mmap_request(file, addr, len, prot, flags, pgoff, | |
1191 | &capabilities); | |
8feae131 DH |
1192 | if (ret < 0) { |
1193 | kleave(" = %d [val]", ret); | |
1da177e4 | 1194 | return ret; |
8feae131 | 1195 | } |
1da177e4 LT |
1196 | |
1197 | /* we've determined that we can make the mapping, now translate what we | |
1198 | * now know into VMA flags */ | |
1199 | vm_flags = determine_vm_flags(file, prot, flags, capabilities); | |
1200 | ||
8feae131 DH |
1201 | /* we're going to need to record the mapping */ |
1202 | region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL); | |
1203 | if (!region) | |
1204 | goto error_getting_region; | |
1205 | ||
1206 | vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); | |
1207 | if (!vma) | |
1208 | goto error_getting_vma; | |
1da177e4 | 1209 | |
8feae131 DH |
1210 | atomic_set(®ion->vm_usage, 1); |
1211 | region->vm_flags = vm_flags; | |
1212 | region->vm_pgoff = pgoff; | |
1213 | ||
1214 | INIT_LIST_HEAD(&vma->anon_vma_node); | |
1215 | vma->vm_flags = vm_flags; | |
1216 | vma->vm_pgoff = pgoff; | |
1da177e4 | 1217 | |
8feae131 DH |
1218 | if (file) { |
1219 | region->vm_file = file; | |
1220 | get_file(file); | |
1221 | vma->vm_file = file; | |
1222 | get_file(file); | |
1223 | if (vm_flags & VM_EXECUTABLE) { | |
1224 | added_exe_file_vma(current->mm); | |
1225 | vma->vm_mm = current->mm; | |
1226 | } | |
1227 | } | |
1228 | ||
1229 | down_write(&nommu_region_sem); | |
1230 | ||
1231 | /* if we want to share, we need to check for regions created by other | |
1da177e4 | 1232 | * mmap() calls that overlap with our proposed mapping |
8feae131 | 1233 | * - we can only share with a superset match on most regular files |
1da177e4 LT |
1234 | * - shared mappings on character devices and memory backed files are |
1235 | * permitted to overlap inexactly as far as we are concerned for in | |
1236 | * these cases, sharing is handled in the driver or filesystem rather | |
1237 | * than here | |
1238 | */ | |
1239 | if (vm_flags & VM_MAYSHARE) { | |
8feae131 DH |
1240 | struct vm_region *pregion; |
1241 | unsigned long pglen, rpglen, pgend, rpgend, start; | |
1da177e4 | 1242 | |
8feae131 DH |
1243 | pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; |
1244 | pgend = pgoff + pglen; | |
165b2392 | 1245 | |
8feae131 DH |
1246 | for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) { |
1247 | pregion = rb_entry(rb, struct vm_region, vm_rb); | |
1da177e4 | 1248 | |
8feae131 | 1249 | if (!(pregion->vm_flags & VM_MAYSHARE)) |
1da177e4 LT |
1250 | continue; |
1251 | ||
1252 | /* search for overlapping mappings on the same file */ | |
8feae131 DH |
1253 | if (pregion->vm_file->f_path.dentry->d_inode != |
1254 | file->f_path.dentry->d_inode) | |
1da177e4 LT |
1255 | continue; |
1256 | ||
8feae131 | 1257 | if (pregion->vm_pgoff >= pgend) |
1da177e4 LT |
1258 | continue; |
1259 | ||
8feae131 DH |
1260 | rpglen = pregion->vm_end - pregion->vm_start; |
1261 | rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
1262 | rpgend = pregion->vm_pgoff + rpglen; | |
1263 | if (pgoff >= rpgend) | |
1da177e4 LT |
1264 | continue; |
1265 | ||
8feae131 DH |
1266 | /* handle inexactly overlapping matches between |
1267 | * mappings */ | |
1268 | if ((pregion->vm_pgoff != pgoff || rpglen != pglen) && | |
1269 | !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) { | |
1270 | /* new mapping is not a subset of the region */ | |
1da177e4 LT |
1271 | if (!(capabilities & BDI_CAP_MAP_DIRECT)) |
1272 | goto sharing_violation; | |
1273 | continue; | |
1274 | } | |
1275 | ||
8feae131 DH |
1276 | /* we've found a region we can share */ |
1277 | atomic_inc(&pregion->vm_usage); | |
1278 | vma->vm_region = pregion; | |
1279 | start = pregion->vm_start; | |
1280 | start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT; | |
1281 | vma->vm_start = start; | |
1282 | vma->vm_end = start + len; | |
1283 | ||
1284 | if (pregion->vm_flags & VM_MAPPED_COPY) { | |
1285 | kdebug("share copy"); | |
1286 | vma->vm_flags |= VM_MAPPED_COPY; | |
1287 | } else { | |
1288 | kdebug("share mmap"); | |
1289 | ret = do_mmap_shared_file(vma); | |
1290 | if (ret < 0) { | |
1291 | vma->vm_region = NULL; | |
1292 | vma->vm_start = 0; | |
1293 | vma->vm_end = 0; | |
1294 | atomic_dec(&pregion->vm_usage); | |
1295 | pregion = NULL; | |
1296 | goto error_just_free; | |
1297 | } | |
1298 | } | |
1299 | fput(region->vm_file); | |
1300 | kmem_cache_free(vm_region_jar, region); | |
1301 | region = pregion; | |
1302 | result = start; | |
1303 | goto share; | |
1da177e4 LT |
1304 | } |
1305 | ||
1da177e4 LT |
1306 | /* obtain the address at which to make a shared mapping |
1307 | * - this is the hook for quasi-memory character devices to | |
1308 | * tell us the location of a shared mapping | |
1309 | */ | |
1310 | if (file && file->f_op->get_unmapped_area) { | |
1311 | addr = file->f_op->get_unmapped_area(file, addr, len, | |
1312 | pgoff, flags); | |
1313 | if (IS_ERR((void *) addr)) { | |
1314 | ret = addr; | |
1315 | if (ret != (unsigned long) -ENOSYS) | |
8feae131 | 1316 | goto error_just_free; |
1da177e4 LT |
1317 | |
1318 | /* the driver refused to tell us where to site | |
1319 | * the mapping so we'll have to attempt to copy | |
1320 | * it */ | |
1321 | ret = (unsigned long) -ENODEV; | |
1322 | if (!(capabilities & BDI_CAP_MAP_COPY)) | |
8feae131 | 1323 | goto error_just_free; |
1da177e4 LT |
1324 | |
1325 | capabilities &= ~BDI_CAP_MAP_DIRECT; | |
8feae131 DH |
1326 | } else { |
1327 | vma->vm_start = region->vm_start = addr; | |
1328 | vma->vm_end = region->vm_end = addr + len; | |
1da177e4 LT |
1329 | } |
1330 | } | |
1331 | } | |
1332 | ||
8feae131 | 1333 | vma->vm_region = region; |
1da177e4 LT |
1334 | |
1335 | /* set up the mapping */ | |
1336 | if (file && vma->vm_flags & VM_SHARED) | |
8feae131 | 1337 | ret = do_mmap_shared_file(vma); |
1da177e4 | 1338 | else |
8feae131 | 1339 | ret = do_mmap_private(vma, region, len); |
1da177e4 | 1340 | if (ret < 0) |
8feae131 DH |
1341 | goto error_put_region; |
1342 | ||
1343 | add_nommu_region(region); | |
1da177e4 LT |
1344 | |
1345 | /* okay... we have a mapping; now we have to register it */ | |
8feae131 | 1346 | result = vma->vm_start; |
1da177e4 | 1347 | |
1da177e4 LT |
1348 | current->mm->total_vm += len >> PAGE_SHIFT; |
1349 | ||
8feae131 DH |
1350 | share: |
1351 | add_vma_to_mm(current->mm, vma); | |
1da177e4 | 1352 | |
8feae131 | 1353 | up_write(&nommu_region_sem); |
1da177e4 LT |
1354 | |
1355 | if (prot & PROT_EXEC) | |
8feae131 | 1356 | flush_icache_range(result, result + len); |
1da177e4 | 1357 | |
8feae131 DH |
1358 | kleave(" = %lx", result); |
1359 | return result; | |
1da177e4 | 1360 | |
8feae131 DH |
1361 | error_put_region: |
1362 | __put_nommu_region(region); | |
1da177e4 | 1363 | if (vma) { |
925d1c40 | 1364 | if (vma->vm_file) { |
3fcd03e0 | 1365 | fput(vma->vm_file); |
925d1c40 MH |
1366 | if (vma->vm_flags & VM_EXECUTABLE) |
1367 | removed_exe_file_vma(vma->vm_mm); | |
1368 | } | |
8feae131 | 1369 | kmem_cache_free(vm_area_cachep, vma); |
1da177e4 | 1370 | } |
8feae131 | 1371 | kleave(" = %d [pr]", ret); |
1da177e4 LT |
1372 | return ret; |
1373 | ||
8feae131 DH |
1374 | error_just_free: |
1375 | up_write(&nommu_region_sem); | |
1376 | error: | |
1377 | fput(region->vm_file); | |
1378 | kmem_cache_free(vm_region_jar, region); | |
1379 | fput(vma->vm_file); | |
1380 | if (vma->vm_flags & VM_EXECUTABLE) | |
1381 | removed_exe_file_vma(vma->vm_mm); | |
1382 | kmem_cache_free(vm_area_cachep, vma); | |
1383 | kleave(" = %d", ret); | |
1384 | return ret; | |
1385 | ||
1386 | sharing_violation: | |
1387 | up_write(&nommu_region_sem); | |
1388 | printk(KERN_WARNING "Attempt to share mismatched mappings\n"); | |
1389 | ret = -EINVAL; | |
1390 | goto error; | |
1da177e4 | 1391 | |
8feae131 DH |
1392 | error_getting_vma: |
1393 | kmem_cache_free(vm_region_jar, region); | |
1394 | printk(KERN_WARNING "Allocation of vma for %lu byte allocation" | |
1395 | " from process %d failed\n", | |
1da177e4 LT |
1396 | len, current->pid); |
1397 | show_free_areas(); | |
1398 | return -ENOMEM; | |
1399 | ||
8feae131 DH |
1400 | error_getting_region: |
1401 | printk(KERN_WARNING "Allocation of vm region for %lu byte allocation" | |
1402 | " from process %d failed\n", | |
1da177e4 LT |
1403 | len, current->pid); |
1404 | show_free_areas(); | |
1405 | return -ENOMEM; | |
1406 | } | |
b5073173 | 1407 | EXPORT_SYMBOL(do_mmap_pgoff); |
1da177e4 LT |
1408 | |
1409 | /* | |
8feae131 DH |
1410 | * split a vma into two pieces at address 'addr', a new vma is allocated either |
1411 | * for the first part or the tail. | |
1da177e4 | 1412 | */ |
8feae131 DH |
1413 | int split_vma(struct mm_struct *mm, struct vm_area_struct *vma, |
1414 | unsigned long addr, int new_below) | |
1da177e4 | 1415 | { |
8feae131 DH |
1416 | struct vm_area_struct *new; |
1417 | struct vm_region *region; | |
1418 | unsigned long npages; | |
1da177e4 | 1419 | |
8feae131 | 1420 | kenter(""); |
1da177e4 | 1421 | |
8feae131 DH |
1422 | /* we're only permitted to split anonymous regions that have a single |
1423 | * owner */ | |
1424 | if (vma->vm_file || | |
1425 | atomic_read(&vma->vm_region->vm_usage) != 1) | |
1426 | return -ENOMEM; | |
1da177e4 | 1427 | |
8feae131 DH |
1428 | if (mm->map_count >= sysctl_max_map_count) |
1429 | return -ENOMEM; | |
1da177e4 | 1430 | |
8feae131 DH |
1431 | region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL); |
1432 | if (!region) | |
1433 | return -ENOMEM; | |
1da177e4 | 1434 | |
8feae131 DH |
1435 | new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); |
1436 | if (!new) { | |
1437 | kmem_cache_free(vm_region_jar, region); | |
1438 | return -ENOMEM; | |
1439 | } | |
1440 | ||
1441 | /* most fields are the same, copy all, and then fixup */ | |
1442 | *new = *vma; | |
1443 | *region = *vma->vm_region; | |
1444 | new->vm_region = region; | |
1445 | ||
1446 | npages = (addr - vma->vm_start) >> PAGE_SHIFT; | |
1447 | ||
1448 | if (new_below) { | |
dd8632a1 | 1449 | region->vm_top = region->vm_end = new->vm_end = addr; |
8feae131 DH |
1450 | } else { |
1451 | region->vm_start = new->vm_start = addr; | |
1452 | region->vm_pgoff = new->vm_pgoff += npages; | |
1da177e4 | 1453 | } |
8feae131 DH |
1454 | |
1455 | if (new->vm_ops && new->vm_ops->open) | |
1456 | new->vm_ops->open(new); | |
1457 | ||
1458 | delete_vma_from_mm(vma); | |
1459 | down_write(&nommu_region_sem); | |
1460 | delete_nommu_region(vma->vm_region); | |
1461 | if (new_below) { | |
1462 | vma->vm_region->vm_start = vma->vm_start = addr; | |
1463 | vma->vm_region->vm_pgoff = vma->vm_pgoff += npages; | |
1464 | } else { | |
1465 | vma->vm_region->vm_end = vma->vm_end = addr; | |
dd8632a1 | 1466 | vma->vm_region->vm_top = addr; |
8feae131 DH |
1467 | } |
1468 | add_nommu_region(vma->vm_region); | |
1469 | add_nommu_region(new->vm_region); | |
1470 | up_write(&nommu_region_sem); | |
1471 | add_vma_to_mm(mm, vma); | |
1472 | add_vma_to_mm(mm, new); | |
1473 | return 0; | |
1da177e4 LT |
1474 | } |
1475 | ||
3034097a | 1476 | /* |
8feae131 DH |
1477 | * shrink a VMA by removing the specified chunk from either the beginning or |
1478 | * the end | |
3034097a | 1479 | */ |
8feae131 DH |
1480 | static int shrink_vma(struct mm_struct *mm, |
1481 | struct vm_area_struct *vma, | |
1482 | unsigned long from, unsigned long to) | |
1da177e4 | 1483 | { |
8feae131 | 1484 | struct vm_region *region; |
1da177e4 | 1485 | |
8feae131 | 1486 | kenter(""); |
1da177e4 | 1487 | |
8feae131 DH |
1488 | /* adjust the VMA's pointers, which may reposition it in the MM's tree |
1489 | * and list */ | |
1490 | delete_vma_from_mm(vma); | |
1491 | if (from > vma->vm_start) | |
1492 | vma->vm_end = from; | |
1493 | else | |
1494 | vma->vm_start = to; | |
1495 | add_vma_to_mm(mm, vma); | |
1da177e4 | 1496 | |
8feae131 DH |
1497 | /* cut the backing region down to size */ |
1498 | region = vma->vm_region; | |
1499 | BUG_ON(atomic_read(®ion->vm_usage) != 1); | |
1500 | ||
1501 | down_write(&nommu_region_sem); | |
1502 | delete_nommu_region(region); | |
dd8632a1 PM |
1503 | if (from > region->vm_start) { |
1504 | to = region->vm_top; | |
1505 | region->vm_top = region->vm_end = from; | |
1506 | } else { | |
8feae131 | 1507 | region->vm_start = to; |
dd8632a1 | 1508 | } |
8feae131 DH |
1509 | add_nommu_region(region); |
1510 | up_write(&nommu_region_sem); | |
1511 | ||
1512 | free_page_series(from, to); | |
1513 | return 0; | |
1514 | } | |
1da177e4 | 1515 | |
8feae131 DH |
1516 | /* |
1517 | * release a mapping | |
1518 | * - under NOMMU conditions the chunk to be unmapped must be backed by a single | |
1519 | * VMA, though it need not cover the whole VMA | |
1520 | */ | |
1521 | int do_munmap(struct mm_struct *mm, unsigned long start, size_t len) | |
1522 | { | |
1523 | struct vm_area_struct *vma; | |
1524 | struct rb_node *rb; | |
1525 | unsigned long end = start + len; | |
1526 | int ret; | |
1da177e4 | 1527 | |
8feae131 | 1528 | kenter(",%lx,%zx", start, len); |
1da177e4 | 1529 | |
8feae131 DH |
1530 | if (len == 0) |
1531 | return -EINVAL; | |
365e9c87 | 1532 | |
8feae131 DH |
1533 | /* find the first potentially overlapping VMA */ |
1534 | vma = find_vma(mm, start); | |
1535 | if (!vma) { | |
33e5d769 DH |
1536 | static int limit = 0; |
1537 | if (limit < 5) { | |
1538 | printk(KERN_WARNING | |
1539 | "munmap of memory not mmapped by process %d" | |
1540 | " (%s): 0x%lx-0x%lx\n", | |
1541 | current->pid, current->comm, | |
1542 | start, start + len - 1); | |
1543 | limit++; | |
1544 | } | |
8feae131 DH |
1545 | return -EINVAL; |
1546 | } | |
1da177e4 | 1547 | |
8feae131 DH |
1548 | /* we're allowed to split an anonymous VMA but not a file-backed one */ |
1549 | if (vma->vm_file) { | |
1550 | do { | |
1551 | if (start > vma->vm_start) { | |
1552 | kleave(" = -EINVAL [miss]"); | |
1553 | return -EINVAL; | |
1554 | } | |
1555 | if (end == vma->vm_end) | |
1556 | goto erase_whole_vma; | |
1557 | rb = rb_next(&vma->vm_rb); | |
1558 | vma = rb_entry(rb, struct vm_area_struct, vm_rb); | |
1559 | } while (rb); | |
1560 | kleave(" = -EINVAL [split file]"); | |
1561 | return -EINVAL; | |
1562 | } else { | |
1563 | /* the chunk must be a subset of the VMA found */ | |
1564 | if (start == vma->vm_start && end == vma->vm_end) | |
1565 | goto erase_whole_vma; | |
1566 | if (start < vma->vm_start || end > vma->vm_end) { | |
1567 | kleave(" = -EINVAL [superset]"); | |
1568 | return -EINVAL; | |
1569 | } | |
1570 | if (start & ~PAGE_MASK) { | |
1571 | kleave(" = -EINVAL [unaligned start]"); | |
1572 | return -EINVAL; | |
1573 | } | |
1574 | if (end != vma->vm_end && end & ~PAGE_MASK) { | |
1575 | kleave(" = -EINVAL [unaligned split]"); | |
1576 | return -EINVAL; | |
1577 | } | |
1578 | if (start != vma->vm_start && end != vma->vm_end) { | |
1579 | ret = split_vma(mm, vma, start, 1); | |
1580 | if (ret < 0) { | |
1581 | kleave(" = %d [split]", ret); | |
1582 | return ret; | |
1583 | } | |
1584 | } | |
1585 | return shrink_vma(mm, vma, start, end); | |
1586 | } | |
1da177e4 | 1587 | |
8feae131 DH |
1588 | erase_whole_vma: |
1589 | delete_vma_from_mm(vma); | |
1590 | delete_vma(mm, vma); | |
1591 | kleave(" = 0"); | |
1da177e4 LT |
1592 | return 0; |
1593 | } | |
b5073173 | 1594 | EXPORT_SYMBOL(do_munmap); |
1da177e4 | 1595 | |
6a6160a7 | 1596 | SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len) |
3034097a DH |
1597 | { |
1598 | int ret; | |
1599 | struct mm_struct *mm = current->mm; | |
1600 | ||
1601 | down_write(&mm->mmap_sem); | |
1602 | ret = do_munmap(mm, addr, len); | |
1603 | up_write(&mm->mmap_sem); | |
1604 | return ret; | |
1605 | } | |
1606 | ||
1607 | /* | |
8feae131 | 1608 | * release all the mappings made in a process's VM space |
3034097a | 1609 | */ |
8feae131 | 1610 | void exit_mmap(struct mm_struct *mm) |
1da177e4 | 1611 | { |
8feae131 | 1612 | struct vm_area_struct *vma; |
1da177e4 | 1613 | |
8feae131 DH |
1614 | if (!mm) |
1615 | return; | |
1da177e4 | 1616 | |
8feae131 | 1617 | kenter(""); |
1da177e4 | 1618 | |
8feae131 | 1619 | mm->total_vm = 0; |
1da177e4 | 1620 | |
8feae131 DH |
1621 | while ((vma = mm->mmap)) { |
1622 | mm->mmap = vma->vm_next; | |
1623 | delete_vma_from_mm(vma); | |
1624 | delete_vma(mm, vma); | |
1da177e4 | 1625 | } |
8feae131 DH |
1626 | |
1627 | kleave(""); | |
1da177e4 LT |
1628 | } |
1629 | ||
1da177e4 LT |
1630 | unsigned long do_brk(unsigned long addr, unsigned long len) |
1631 | { | |
1632 | return -ENOMEM; | |
1633 | } | |
1634 | ||
1635 | /* | |
6fa5f80b DH |
1636 | * expand (or shrink) an existing mapping, potentially moving it at the same |
1637 | * time (controlled by the MREMAP_MAYMOVE flag and available VM space) | |
1da177e4 | 1638 | * |
6fa5f80b | 1639 | * under NOMMU conditions, we only permit changing a mapping's size, and only |
8feae131 DH |
1640 | * as long as it stays within the region allocated by do_mmap_private() and the |
1641 | * block is not shareable | |
1da177e4 | 1642 | * |
6fa5f80b | 1643 | * MREMAP_FIXED is not supported under NOMMU conditions |
1da177e4 LT |
1644 | */ |
1645 | unsigned long do_mremap(unsigned long addr, | |
1646 | unsigned long old_len, unsigned long new_len, | |
1647 | unsigned long flags, unsigned long new_addr) | |
1648 | { | |
6fa5f80b | 1649 | struct vm_area_struct *vma; |
1da177e4 LT |
1650 | |
1651 | /* insanity checks first */ | |
8feae131 | 1652 | if (old_len == 0 || new_len == 0) |
1da177e4 LT |
1653 | return (unsigned long) -EINVAL; |
1654 | ||
8feae131 DH |
1655 | if (addr & ~PAGE_MASK) |
1656 | return -EINVAL; | |
1657 | ||
1da177e4 LT |
1658 | if (flags & MREMAP_FIXED && new_addr != addr) |
1659 | return (unsigned long) -EINVAL; | |
1660 | ||
8feae131 | 1661 | vma = find_vma_exact(current->mm, addr, old_len); |
6fa5f80b DH |
1662 | if (!vma) |
1663 | return (unsigned long) -EINVAL; | |
1da177e4 | 1664 | |
6fa5f80b | 1665 | if (vma->vm_end != vma->vm_start + old_len) |
1da177e4 LT |
1666 | return (unsigned long) -EFAULT; |
1667 | ||
6fa5f80b | 1668 | if (vma->vm_flags & VM_MAYSHARE) |
1da177e4 LT |
1669 | return (unsigned long) -EPERM; |
1670 | ||
8feae131 | 1671 | if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start) |
1da177e4 LT |
1672 | return (unsigned long) -ENOMEM; |
1673 | ||
1674 | /* all checks complete - do it */ | |
6fa5f80b | 1675 | vma->vm_end = vma->vm_start + new_len; |
6fa5f80b DH |
1676 | return vma->vm_start; |
1677 | } | |
b5073173 | 1678 | EXPORT_SYMBOL(do_mremap); |
6fa5f80b | 1679 | |
6a6160a7 HC |
1680 | SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len, |
1681 | unsigned long, new_len, unsigned long, flags, | |
1682 | unsigned long, new_addr) | |
6fa5f80b DH |
1683 | { |
1684 | unsigned long ret; | |
1685 | ||
1686 | down_write(¤t->mm->mmap_sem); | |
1687 | ret = do_mremap(addr, old_len, new_len, flags, new_addr); | |
1688 | up_write(¤t->mm->mmap_sem); | |
1689 | return ret; | |
1da177e4 LT |
1690 | } |
1691 | ||
6aab341e | 1692 | struct page *follow_page(struct vm_area_struct *vma, unsigned long address, |
deceb6cd | 1693 | unsigned int foll_flags) |
1da177e4 LT |
1694 | { |
1695 | return NULL; | |
1696 | } | |
1697 | ||
1da177e4 LT |
1698 | int remap_pfn_range(struct vm_area_struct *vma, unsigned long from, |
1699 | unsigned long to, unsigned long size, pgprot_t prot) | |
1700 | { | |
66aa2b4b GU |
1701 | vma->vm_start = vma->vm_pgoff << PAGE_SHIFT; |
1702 | return 0; | |
1da177e4 | 1703 | } |
22c4af40 | 1704 | EXPORT_SYMBOL(remap_pfn_range); |
1da177e4 | 1705 | |
f905bc44 PM |
1706 | int remap_vmalloc_range(struct vm_area_struct *vma, void *addr, |
1707 | unsigned long pgoff) | |
1708 | { | |
1709 | unsigned int size = vma->vm_end - vma->vm_start; | |
1710 | ||
1711 | if (!(vma->vm_flags & VM_USERMAP)) | |
1712 | return -EINVAL; | |
1713 | ||
1714 | vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT)); | |
1715 | vma->vm_end = vma->vm_start + size; | |
1716 | ||
1717 | return 0; | |
1718 | } | |
1719 | EXPORT_SYMBOL(remap_vmalloc_range); | |
1720 | ||
1da177e4 LT |
1721 | void swap_unplug_io_fn(struct backing_dev_info *bdi, struct page *page) |
1722 | { | |
1723 | } | |
1724 | ||
1725 | unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr, | |
1726 | unsigned long len, unsigned long pgoff, unsigned long flags) | |
1727 | { | |
1728 | return -ENOMEM; | |
1729 | } | |
1730 | ||
1363c3cd | 1731 | void arch_unmap_area(struct mm_struct *mm, unsigned long addr) |
1da177e4 LT |
1732 | { |
1733 | } | |
1734 | ||
1da177e4 LT |
1735 | void unmap_mapping_range(struct address_space *mapping, |
1736 | loff_t const holebegin, loff_t const holelen, | |
1737 | int even_cows) | |
1738 | { | |
1739 | } | |
22c4af40 | 1740 | EXPORT_SYMBOL(unmap_mapping_range); |
1da177e4 | 1741 | |
d56e03cd DH |
1742 | /* |
1743 | * ask for an unmapped area at which to create a mapping on a file | |
1744 | */ | |
1745 | unsigned long get_unmapped_area(struct file *file, unsigned long addr, | |
1746 | unsigned long len, unsigned long pgoff, | |
1747 | unsigned long flags) | |
1748 | { | |
1749 | unsigned long (*get_area)(struct file *, unsigned long, unsigned long, | |
1750 | unsigned long, unsigned long); | |
1751 | ||
1752 | get_area = current->mm->get_unmapped_area; | |
1753 | if (file && file->f_op && file->f_op->get_unmapped_area) | |
1754 | get_area = file->f_op->get_unmapped_area; | |
1755 | ||
1756 | if (!get_area) | |
1757 | return -ENOSYS; | |
1758 | ||
1759 | return get_area(file, addr, len, pgoff, flags); | |
1760 | } | |
d56e03cd DH |
1761 | EXPORT_SYMBOL(get_unmapped_area); |
1762 | ||
1da177e4 LT |
1763 | /* |
1764 | * Check that a process has enough memory to allocate a new virtual | |
1765 | * mapping. 0 means there is enough memory for the allocation to | |
1766 | * succeed and -ENOMEM implies there is not. | |
1767 | * | |
1768 | * We currently support three overcommit policies, which are set via the | |
1769 | * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting | |
1770 | * | |
1771 | * Strict overcommit modes added 2002 Feb 26 by Alan Cox. | |
1772 | * Additional code 2002 Jul 20 by Robert Love. | |
1773 | * | |
1774 | * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise. | |
1775 | * | |
1776 | * Note this is a helper function intended to be used by LSMs which | |
1777 | * wish to use this logic. | |
1778 | */ | |
34b4e4aa | 1779 | int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin) |
1da177e4 LT |
1780 | { |
1781 | unsigned long free, allowed; | |
1782 | ||
1783 | vm_acct_memory(pages); | |
1784 | ||
1785 | /* | |
1786 | * Sometimes we want to use more memory than we have | |
1787 | */ | |
1788 | if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS) | |
1789 | return 0; | |
1790 | ||
1791 | if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) { | |
1792 | unsigned long n; | |
1793 | ||
347ce434 | 1794 | free = global_page_state(NR_FILE_PAGES); |
1da177e4 LT |
1795 | free += nr_swap_pages; |
1796 | ||
1797 | /* | |
1798 | * Any slabs which are created with the | |
1799 | * SLAB_RECLAIM_ACCOUNT flag claim to have contents | |
1800 | * which are reclaimable, under pressure. The dentry | |
1801 | * cache and most inode caches should fall into this | |
1802 | */ | |
972d1a7b | 1803 | free += global_page_state(NR_SLAB_RECLAIMABLE); |
1da177e4 LT |
1804 | |
1805 | /* | |
1806 | * Leave the last 3% for root | |
1807 | */ | |
1808 | if (!cap_sys_admin) | |
1809 | free -= free / 32; | |
1810 | ||
1811 | if (free > pages) | |
1812 | return 0; | |
1813 | ||
1814 | /* | |
1815 | * nr_free_pages() is very expensive on large systems, | |
1816 | * only call if we're about to fail. | |
1817 | */ | |
1818 | n = nr_free_pages(); | |
d5ddc79b HA |
1819 | |
1820 | /* | |
1821 | * Leave reserved pages. The pages are not for anonymous pages. | |
1822 | */ | |
1823 | if (n <= totalreserve_pages) | |
1824 | goto error; | |
1825 | else | |
1826 | n -= totalreserve_pages; | |
1827 | ||
1828 | /* | |
1829 | * Leave the last 3% for root | |
1830 | */ | |
1da177e4 LT |
1831 | if (!cap_sys_admin) |
1832 | n -= n / 32; | |
1833 | free += n; | |
1834 | ||
1835 | if (free > pages) | |
1836 | return 0; | |
d5ddc79b HA |
1837 | |
1838 | goto error; | |
1da177e4 LT |
1839 | } |
1840 | ||
1841 | allowed = totalram_pages * sysctl_overcommit_ratio / 100; | |
1842 | /* | |
1843 | * Leave the last 3% for root | |
1844 | */ | |
1845 | if (!cap_sys_admin) | |
1846 | allowed -= allowed / 32; | |
1847 | allowed += total_swap_pages; | |
1848 | ||
1849 | /* Don't let a single process grow too big: | |
1850 | leave 3% of the size of this process for other processes */ | |
731572d3 AC |
1851 | if (mm) |
1852 | allowed -= mm->total_vm / 32; | |
1da177e4 | 1853 | |
00a62ce9 | 1854 | if (percpu_counter_read_positive(&vm_committed_as) < allowed) |
1da177e4 | 1855 | return 0; |
00a62ce9 | 1856 | |
d5ddc79b | 1857 | error: |
1da177e4 LT |
1858 | vm_unacct_memory(pages); |
1859 | ||
1860 | return -ENOMEM; | |
1861 | } | |
1862 | ||
1863 | int in_gate_area_no_task(unsigned long addr) | |
1864 | { | |
1865 | return 0; | |
1866 | } | |
b0e15190 | 1867 | |
d0217ac0 | 1868 | int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf) |
b0e15190 DH |
1869 | { |
1870 | BUG(); | |
d0217ac0 | 1871 | return 0; |
b0e15190 | 1872 | } |
b5073173 | 1873 | EXPORT_SYMBOL(filemap_fault); |
0ec76a11 DH |
1874 | |
1875 | /* | |
1876 | * Access another process' address space. | |
1877 | * - source/target buffer must be kernel space | |
1878 | */ | |
1879 | int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write) | |
1880 | { | |
0ec76a11 DH |
1881 | struct vm_area_struct *vma; |
1882 | struct mm_struct *mm; | |
1883 | ||
1884 | if (addr + len < addr) | |
1885 | return 0; | |
1886 | ||
1887 | mm = get_task_mm(tsk); | |
1888 | if (!mm) | |
1889 | return 0; | |
1890 | ||
1891 | down_read(&mm->mmap_sem); | |
1892 | ||
1893 | /* the access must start within one of the target process's mappings */ | |
0159b141 DH |
1894 | vma = find_vma(mm, addr); |
1895 | if (vma) { | |
0ec76a11 DH |
1896 | /* don't overrun this mapping */ |
1897 | if (addr + len >= vma->vm_end) | |
1898 | len = vma->vm_end - addr; | |
1899 | ||
1900 | /* only read or write mappings where it is permitted */ | |
d00c7b99 | 1901 | if (write && vma->vm_flags & VM_MAYWRITE) |
0ec76a11 | 1902 | len -= copy_to_user((void *) addr, buf, len); |
d00c7b99 | 1903 | else if (!write && vma->vm_flags & VM_MAYREAD) |
0ec76a11 DH |
1904 | len -= copy_from_user(buf, (void *) addr, len); |
1905 | else | |
1906 | len = 0; | |
1907 | } else { | |
1908 | len = 0; | |
1909 | } | |
1910 | ||
1911 | up_read(&mm->mmap_sem); | |
1912 | mmput(mm); | |
1913 | return len; | |
1914 | } |