Commit | Line | Data |
---|---|---|
9c92ab61 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
0c972a05 TK |
2 | /* binder_alloc.c |
3 | * | |
4 | * Android IPC Subsystem | |
5 | * | |
6 | * Copyright (C) 2007-2017 Google, Inc. | |
0c972a05 TK |
7 | */ |
8 | ||
9 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt | |
10 | ||
0c972a05 TK |
11 | #include <linux/list.h> |
12 | #include <linux/sched/mm.h> | |
13 | #include <linux/module.h> | |
14 | #include <linux/rtmutex.h> | |
15 | #include <linux/rbtree.h> | |
16 | #include <linux/seq_file.h> | |
17 | #include <linux/vmalloc.h> | |
18 | #include <linux/slab.h> | |
19 | #include <linux/sched.h> | |
f2517eb7 | 20 | #include <linux/list_lru.h> |
128f3804 | 21 | #include <linux/ratelimit.h> |
1e81c57b | 22 | #include <asm/cacheflush.h> |
1a7c3d9b TK |
23 | #include <linux/uaccess.h> |
24 | #include <linux/highmem.h> | |
45d02f79 | 25 | #include <linux/sizes.h> |
0c972a05 TK |
26 | #include "binder_alloc.h" |
27 | #include "binder_trace.h" | |
28 | ||
f2517eb7 SY |
29 | struct list_lru binder_alloc_lru; |
30 | ||
0c972a05 TK |
31 | static DEFINE_MUTEX(binder_alloc_mmap_lock); |
32 | ||
33 | enum { | |
128f3804 | 34 | BINDER_DEBUG_USER_ERROR = 1U << 0, |
0c972a05 TK |
35 | BINDER_DEBUG_OPEN_CLOSE = 1U << 1, |
36 | BINDER_DEBUG_BUFFER_ALLOC = 1U << 2, | |
37 | BINDER_DEBUG_BUFFER_ALLOC_ASYNC = 1U << 3, | |
38 | }; | |
128f3804 | 39 | static uint32_t binder_alloc_debug_mask = BINDER_DEBUG_USER_ERROR; |
0c972a05 TK |
40 | |
41 | module_param_named(debug_mask, binder_alloc_debug_mask, | |
42 | uint, 0644); | |
43 | ||
44 | #define binder_alloc_debug(mask, x...) \ | |
45 | do { \ | |
46 | if (binder_alloc_debug_mask & mask) \ | |
128f3804 | 47 | pr_info_ratelimited(x); \ |
0c972a05 TK |
48 | } while (0) |
49 | ||
e2176219 SY |
50 | static struct binder_buffer *binder_buffer_next(struct binder_buffer *buffer) |
51 | { | |
52 | return list_entry(buffer->entry.next, struct binder_buffer, entry); | |
53 | } | |
54 | ||
55 | static struct binder_buffer *binder_buffer_prev(struct binder_buffer *buffer) | |
56 | { | |
57 | return list_entry(buffer->entry.prev, struct binder_buffer, entry); | |
58 | } | |
59 | ||
0c972a05 TK |
60 | static size_t binder_alloc_buffer_size(struct binder_alloc *alloc, |
61 | struct binder_buffer *buffer) | |
62 | { | |
63 | if (list_is_last(&buffer->entry, &alloc->buffers)) | |
bde4a19f TK |
64 | return alloc->buffer + alloc->buffer_size - buffer->user_data; |
65 | return binder_buffer_next(buffer)->user_data - buffer->user_data; | |
0c972a05 TK |
66 | } |
67 | ||
68 | static void binder_insert_free_buffer(struct binder_alloc *alloc, | |
69 | struct binder_buffer *new_buffer) | |
70 | { | |
71 | struct rb_node **p = &alloc->free_buffers.rb_node; | |
72 | struct rb_node *parent = NULL; | |
73 | struct binder_buffer *buffer; | |
74 | size_t buffer_size; | |
75 | size_t new_buffer_size; | |
76 | ||
77 | BUG_ON(!new_buffer->free); | |
78 | ||
79 | new_buffer_size = binder_alloc_buffer_size(alloc, new_buffer); | |
80 | ||
81 | binder_alloc_debug(BINDER_DEBUG_BUFFER_ALLOC, | |
82 | "%d: add free buffer, size %zd, at %pK\n", | |
83 | alloc->pid, new_buffer_size, new_buffer); | |
84 | ||
85 | while (*p) { | |
86 | parent = *p; | |
87 | buffer = rb_entry(parent, struct binder_buffer, rb_node); | |
88 | BUG_ON(!buffer->free); | |
89 | ||
90 | buffer_size = binder_alloc_buffer_size(alloc, buffer); | |
91 | ||
92 | if (new_buffer_size < buffer_size) | |
93 | p = &parent->rb_left; | |
94 | else | |
95 | p = &parent->rb_right; | |
96 | } | |
97 | rb_link_node(&new_buffer->rb_node, parent, p); | |
98 | rb_insert_color(&new_buffer->rb_node, &alloc->free_buffers); | |
99 | } | |
100 | ||
101 | static void binder_insert_allocated_buffer_locked( | |
102 | struct binder_alloc *alloc, struct binder_buffer *new_buffer) | |
103 | { | |
104 | struct rb_node **p = &alloc->allocated_buffers.rb_node; | |
105 | struct rb_node *parent = NULL; | |
106 | struct binder_buffer *buffer; | |
107 | ||
108 | BUG_ON(new_buffer->free); | |
109 | ||
110 | while (*p) { | |
111 | parent = *p; | |
112 | buffer = rb_entry(parent, struct binder_buffer, rb_node); | |
113 | BUG_ON(buffer->free); | |
114 | ||
bde4a19f | 115 | if (new_buffer->user_data < buffer->user_data) |
0c972a05 | 116 | p = &parent->rb_left; |
bde4a19f | 117 | else if (new_buffer->user_data > buffer->user_data) |
0c972a05 TK |
118 | p = &parent->rb_right; |
119 | else | |
120 | BUG(); | |
121 | } | |
122 | rb_link_node(&new_buffer->rb_node, parent, p); | |
123 | rb_insert_color(&new_buffer->rb_node, &alloc->allocated_buffers); | |
124 | } | |
125 | ||
53d311cf | 126 | static struct binder_buffer *binder_alloc_prepare_to_free_locked( |
0c972a05 TK |
127 | struct binder_alloc *alloc, |
128 | uintptr_t user_ptr) | |
129 | { | |
130 | struct rb_node *n = alloc->allocated_buffers.rb_node; | |
131 | struct binder_buffer *buffer; | |
bde4a19f | 132 | void __user *uptr; |
0c972a05 | 133 | |
bde4a19f | 134 | uptr = (void __user *)user_ptr; |
0c972a05 TK |
135 | |
136 | while (n) { | |
137 | buffer = rb_entry(n, struct binder_buffer, rb_node); | |
138 | BUG_ON(buffer->free); | |
139 | ||
bde4a19f | 140 | if (uptr < buffer->user_data) |
0c972a05 | 141 | n = n->rb_left; |
bde4a19f | 142 | else if (uptr > buffer->user_data) |
0c972a05 | 143 | n = n->rb_right; |
53d311cf TK |
144 | else { |
145 | /* | |
146 | * Guard against user threads attempting to | |
7bada55a TK |
147 | * free the buffer when in use by kernel or |
148 | * after it's already been freed. | |
53d311cf | 149 | */ |
7bada55a TK |
150 | if (!buffer->allow_user_free) |
151 | return ERR_PTR(-EPERM); | |
152 | buffer->allow_user_free = 0; | |
0c972a05 | 153 | return buffer; |
53d311cf | 154 | } |
0c972a05 TK |
155 | } |
156 | return NULL; | |
157 | } | |
158 | ||
159 | /** | |
5dc54a06 | 160 | * binder_alloc_prepare_to_free() - get buffer given user ptr |
0c972a05 TK |
161 | * @alloc: binder_alloc for this proc |
162 | * @user_ptr: User pointer to buffer data | |
163 | * | |
164 | * Validate userspace pointer to buffer data and return buffer corresponding to | |
165 | * that user pointer. Search the rb tree for buffer that matches user data | |
166 | * pointer. | |
167 | * | |
168 | * Return: Pointer to buffer or NULL | |
169 | */ | |
53d311cf TK |
170 | struct binder_buffer *binder_alloc_prepare_to_free(struct binder_alloc *alloc, |
171 | uintptr_t user_ptr) | |
0c972a05 TK |
172 | { |
173 | struct binder_buffer *buffer; | |
174 | ||
175 | mutex_lock(&alloc->mutex); | |
53d311cf | 176 | buffer = binder_alloc_prepare_to_free_locked(alloc, user_ptr); |
0c972a05 TK |
177 | mutex_unlock(&alloc->mutex); |
178 | return buffer; | |
179 | } | |
180 | ||
181 | static int binder_update_page_range(struct binder_alloc *alloc, int allocate, | |
bde4a19f | 182 | void __user *start, void __user *end) |
0c972a05 | 183 | { |
bde4a19f | 184 | void __user *page_addr; |
0c972a05 | 185 | unsigned long user_page_addr; |
f2517eb7 | 186 | struct binder_lru_page *page; |
6ae33b9c | 187 | struct vm_area_struct *vma = NULL; |
f2517eb7 SY |
188 | struct mm_struct *mm = NULL; |
189 | bool need_mm = false; | |
0c972a05 TK |
190 | |
191 | binder_alloc_debug(BINDER_DEBUG_BUFFER_ALLOC, | |
192 | "%d: %s pages %pK-%pK\n", alloc->pid, | |
193 | allocate ? "allocate" : "free", start, end); | |
194 | ||
195 | if (end <= start) | |
196 | return 0; | |
197 | ||
198 | trace_binder_update_page_range(alloc, allocate, start, end); | |
199 | ||
f2517eb7 SY |
200 | if (allocate == 0) |
201 | goto free_range; | |
202 | ||
203 | for (page_addr = start; page_addr < end; page_addr += PAGE_SIZE) { | |
204 | page = &alloc->pages[(page_addr - alloc->buffer) / PAGE_SIZE]; | |
205 | if (!page->page_ptr) { | |
206 | need_mm = true; | |
207 | break; | |
208 | } | |
209 | } | |
210 | ||
e66b77e5 CL |
211 | if (need_mm && mmget_not_zero(alloc->mm)) |
212 | mm = alloc->mm; | |
0c972a05 TK |
213 | |
214 | if (mm) { | |
d1d8875c | 215 | mmap_write_lock(mm); |
c0fd2101 | 216 | vma = alloc->vma; |
0c972a05 TK |
217 | } |
218 | ||
f2517eb7 | 219 | if (!vma && need_mm) { |
128f3804 SY |
220 | binder_alloc_debug(BINDER_DEBUG_USER_ERROR, |
221 | "%d: binder_alloc_buf failed to map pages in userspace, no vma\n", | |
222 | alloc->pid); | |
0c972a05 TK |
223 | goto err_no_vma; |
224 | } | |
225 | ||
226 | for (page_addr = start; page_addr < end; page_addr += PAGE_SIZE) { | |
227 | int ret; | |
f2517eb7 | 228 | bool on_lru; |
e41e164c | 229 | size_t index; |
0c972a05 | 230 | |
e41e164c SY |
231 | index = (page_addr - alloc->buffer) / PAGE_SIZE; |
232 | page = &alloc->pages[index]; | |
0c972a05 | 233 | |
f2517eb7 | 234 | if (page->page_ptr) { |
e41e164c SY |
235 | trace_binder_alloc_lru_start(alloc, index); |
236 | ||
f2517eb7 SY |
237 | on_lru = list_lru_del(&binder_alloc_lru, &page->lru); |
238 | WARN_ON(!on_lru); | |
e41e164c SY |
239 | |
240 | trace_binder_alloc_lru_end(alloc, index); | |
f2517eb7 SY |
241 | continue; |
242 | } | |
243 | ||
244 | if (WARN_ON(!vma)) | |
245 | goto err_page_ptr_cleared; | |
246 | ||
e41e164c | 247 | trace_binder_alloc_page_start(alloc, index); |
f2517eb7 SY |
248 | page->page_ptr = alloc_page(GFP_KERNEL | |
249 | __GFP_HIGHMEM | | |
250 | __GFP_ZERO); | |
251 | if (!page->page_ptr) { | |
0c972a05 TK |
252 | pr_err("%d: binder_alloc_buf failed for page at %pK\n", |
253 | alloc->pid, page_addr); | |
254 | goto err_alloc_page_failed; | |
255 | } | |
f2517eb7 SY |
256 | page->alloc = alloc; |
257 | INIT_LIST_HEAD(&page->lru); | |
258 | ||
c41358a5 | 259 | user_page_addr = (uintptr_t)page_addr; |
f2517eb7 | 260 | ret = vm_insert_page(vma, user_page_addr, page[0].page_ptr); |
0c972a05 TK |
261 | if (ret) { |
262 | pr_err("%d: binder_alloc_buf failed to map page at %lx in userspace\n", | |
263 | alloc->pid, user_page_addr); | |
264 | goto err_vm_insert_page_failed; | |
265 | } | |
e41e164c | 266 | |
8d9a3ab6 MC |
267 | if (index + 1 > alloc->pages_high) |
268 | alloc->pages_high = index + 1; | |
269 | ||
e41e164c | 270 | trace_binder_alloc_page_end(alloc, index); |
0c972a05 TK |
271 | } |
272 | if (mm) { | |
d1d8875c | 273 | mmap_write_unlock(mm); |
0c972a05 TK |
274 | mmput(mm); |
275 | } | |
276 | return 0; | |
277 | ||
278 | free_range: | |
2a9edd05 | 279 | for (page_addr = end - PAGE_SIZE; 1; page_addr -= PAGE_SIZE) { |
f2517eb7 | 280 | bool ret; |
e41e164c | 281 | size_t index; |
f2517eb7 | 282 | |
e41e164c SY |
283 | index = (page_addr - alloc->buffer) / PAGE_SIZE; |
284 | page = &alloc->pages[index]; | |
285 | ||
286 | trace_binder_free_lru_start(alloc, index); | |
f2517eb7 SY |
287 | |
288 | ret = list_lru_add(&binder_alloc_lru, &page->lru); | |
289 | WARN_ON(!ret); | |
e41e164c SY |
290 | |
291 | trace_binder_free_lru_end(alloc, index); | |
2a9edd05 JH |
292 | if (page_addr == start) |
293 | break; | |
f2517eb7 SY |
294 | continue; |
295 | ||
0c972a05 | 296 | err_vm_insert_page_failed: |
f2517eb7 SY |
297 | __free_page(page->page_ptr); |
298 | page->page_ptr = NULL; | |
0c972a05 | 299 | err_alloc_page_failed: |
f2517eb7 | 300 | err_page_ptr_cleared: |
2a9edd05 JH |
301 | if (page_addr == start) |
302 | break; | |
0c972a05 TK |
303 | } |
304 | err_no_vma: | |
305 | if (mm) { | |
d1d8875c | 306 | mmap_write_unlock(mm); |
0c972a05 TK |
307 | mmput(mm); |
308 | } | |
57ada2fb | 309 | return vma ? -ENOMEM : -ESRCH; |
0c972a05 TK |
310 | } |
311 | ||
0fa53349 CL |
312 | static inline void binder_alloc_set_vma(struct binder_alloc *alloc, |
313 | struct vm_area_struct *vma) | |
314 | { | |
315 | /* pairs with smp_load_acquire in binder_alloc_get_vma() */ | |
316 | smp_store_release(&alloc->vma, vma); | |
317 | } | |
318 | ||
da1b9564 MK |
319 | static inline struct vm_area_struct *binder_alloc_get_vma( |
320 | struct binder_alloc *alloc) | |
321 | { | |
0fa53349 CL |
322 | /* pairs with smp_store_release in binder_alloc_set_vma() */ |
323 | return smp_load_acquire(&alloc->vma); | |
da1b9564 MK |
324 | } |
325 | ||
a7dc1e6f | 326 | static bool debug_low_async_space_locked(struct binder_alloc *alloc, int pid) |
261e7818 MC |
327 | { |
328 | /* | |
329 | * Find the amount and size of buffers allocated by the current caller; | |
330 | * The idea is that once we cross the threshold, whoever is responsible | |
331 | * for the low async space is likely to try to send another async txn, | |
332 | * and at some point we'll catch them in the act. This is more efficient | |
333 | * than keeping a map per pid. | |
334 | */ | |
7369fa47 | 335 | struct rb_node *n; |
261e7818 MC |
336 | struct binder_buffer *buffer; |
337 | size_t total_alloc_size = 0; | |
338 | size_t num_buffers = 0; | |
339 | ||
340 | for (n = rb_first(&alloc->allocated_buffers); n != NULL; | |
341 | n = rb_next(n)) { | |
342 | buffer = rb_entry(n, struct binder_buffer, rb_node); | |
343 | if (buffer->pid != pid) | |
344 | continue; | |
345 | if (!buffer->async_transaction) | |
346 | continue; | |
347 | total_alloc_size += binder_alloc_buffer_size(alloc, buffer) | |
348 | + sizeof(struct binder_buffer); | |
349 | num_buffers++; | |
350 | } | |
351 | ||
352 | /* | |
353 | * Warn if this pid has more than 50 transactions, or more than 50% of | |
a7dc1e6f HL |
354 | * async space (which is 25% of total buffer size). Oneway spam is only |
355 | * detected when the threshold is exceeded. | |
261e7818 MC |
356 | */ |
357 | if (num_buffers > 50 || total_alloc_size > alloc->buffer_size / 4) { | |
358 | binder_alloc_debug(BINDER_DEBUG_USER_ERROR, | |
359 | "%d: pid %d spamming oneway? %zd buffers allocated for a total size of %zd\n", | |
360 | alloc->pid, pid, num_buffers, total_alloc_size); | |
a7dc1e6f HL |
361 | if (!alloc->oneway_spam_detected) { |
362 | alloc->oneway_spam_detected = true; | |
363 | return true; | |
364 | } | |
261e7818 | 365 | } |
a7dc1e6f | 366 | return false; |
261e7818 MC |
367 | } |
368 | ||
3f827245 XS |
369 | static struct binder_buffer *binder_alloc_new_buf_locked( |
370 | struct binder_alloc *alloc, | |
371 | size_t data_size, | |
372 | size_t offsets_size, | |
373 | size_t extra_buffers_size, | |
261e7818 MC |
374 | int is_async, |
375 | int pid) | |
0c972a05 TK |
376 | { |
377 | struct rb_node *n = alloc->free_buffers.rb_node; | |
378 | struct binder_buffer *buffer; | |
379 | size_t buffer_size; | |
380 | struct rb_node *best_fit = NULL; | |
bde4a19f TK |
381 | void __user *has_page_addr; |
382 | void __user *end_page_addr; | |
0c972a05 | 383 | size_t size, data_offsets_size; |
57ada2fb | 384 | int ret; |
0c972a05 | 385 | |
0fa53349 | 386 | /* Check binder_alloc is fully initialized */ |
da1b9564 | 387 | if (!binder_alloc_get_vma(alloc)) { |
128f3804 SY |
388 | binder_alloc_debug(BINDER_DEBUG_USER_ERROR, |
389 | "%d: binder_alloc_buf, no vma\n", | |
390 | alloc->pid); | |
57ada2fb | 391 | return ERR_PTR(-ESRCH); |
0c972a05 TK |
392 | } |
393 | ||
394 | data_offsets_size = ALIGN(data_size, sizeof(void *)) + | |
395 | ALIGN(offsets_size, sizeof(void *)); | |
396 | ||
397 | if (data_offsets_size < data_size || data_offsets_size < offsets_size) { | |
398 | binder_alloc_debug(BINDER_DEBUG_BUFFER_ALLOC, | |
399 | "%d: got transaction with invalid size %zd-%zd\n", | |
400 | alloc->pid, data_size, offsets_size); | |
57ada2fb | 401 | return ERR_PTR(-EINVAL); |
0c972a05 TK |
402 | } |
403 | size = data_offsets_size + ALIGN(extra_buffers_size, sizeof(void *)); | |
404 | if (size < data_offsets_size || size < extra_buffers_size) { | |
405 | binder_alloc_debug(BINDER_DEBUG_BUFFER_ALLOC, | |
406 | "%d: got transaction with invalid extra_buffers_size %zd\n", | |
407 | alloc->pid, extra_buffers_size); | |
57ada2fb | 408 | return ERR_PTR(-EINVAL); |
0c972a05 TK |
409 | } |
410 | if (is_async && | |
411 | alloc->free_async_space < size + sizeof(struct binder_buffer)) { | |
412 | binder_alloc_debug(BINDER_DEBUG_BUFFER_ALLOC, | |
413 | "%d: binder_alloc_buf size %zd failed, no async space left\n", | |
414 | alloc->pid, size); | |
57ada2fb | 415 | return ERR_PTR(-ENOSPC); |
0c972a05 TK |
416 | } |
417 | ||
74310e06 SY |
418 | /* Pad 0-size buffers so they get assigned unique addresses */ |
419 | size = max(size, sizeof(void *)); | |
420 | ||
0c972a05 TK |
421 | while (n) { |
422 | buffer = rb_entry(n, struct binder_buffer, rb_node); | |
423 | BUG_ON(!buffer->free); | |
424 | buffer_size = binder_alloc_buffer_size(alloc, buffer); | |
425 | ||
426 | if (size < buffer_size) { | |
427 | best_fit = n; | |
428 | n = n->rb_left; | |
429 | } else if (size > buffer_size) | |
430 | n = n->rb_right; | |
431 | else { | |
432 | best_fit = n; | |
433 | break; | |
434 | } | |
435 | } | |
436 | if (best_fit == NULL) { | |
b05a68e9 MC |
437 | size_t allocated_buffers = 0; |
438 | size_t largest_alloc_size = 0; | |
439 | size_t total_alloc_size = 0; | |
440 | size_t free_buffers = 0; | |
441 | size_t largest_free_size = 0; | |
442 | size_t total_free_size = 0; | |
443 | ||
444 | for (n = rb_first(&alloc->allocated_buffers); n != NULL; | |
445 | n = rb_next(n)) { | |
446 | buffer = rb_entry(n, struct binder_buffer, rb_node); | |
447 | buffer_size = binder_alloc_buffer_size(alloc, buffer); | |
448 | allocated_buffers++; | |
449 | total_alloc_size += buffer_size; | |
450 | if (buffer_size > largest_alloc_size) | |
451 | largest_alloc_size = buffer_size; | |
452 | } | |
453 | for (n = rb_first(&alloc->free_buffers); n != NULL; | |
454 | n = rb_next(n)) { | |
455 | buffer = rb_entry(n, struct binder_buffer, rb_node); | |
456 | buffer_size = binder_alloc_buffer_size(alloc, buffer); | |
457 | free_buffers++; | |
458 | total_free_size += buffer_size; | |
459 | if (buffer_size > largest_free_size) | |
460 | largest_free_size = buffer_size; | |
461 | } | |
128f3804 SY |
462 | binder_alloc_debug(BINDER_DEBUG_USER_ERROR, |
463 | "%d: binder_alloc_buf size %zd failed, no address space\n", | |
464 | alloc->pid, size); | |
465 | binder_alloc_debug(BINDER_DEBUG_USER_ERROR, | |
466 | "allocated: %zd (num: %zd largest: %zd), free: %zd (num: %zd largest: %zd)\n", | |
467 | total_alloc_size, allocated_buffers, | |
468 | largest_alloc_size, total_free_size, | |
469 | free_buffers, largest_free_size); | |
57ada2fb | 470 | return ERR_PTR(-ENOSPC); |
0c972a05 TK |
471 | } |
472 | if (n == NULL) { | |
473 | buffer = rb_entry(best_fit, struct binder_buffer, rb_node); | |
474 | buffer_size = binder_alloc_buffer_size(alloc, buffer); | |
475 | } | |
476 | ||
477 | binder_alloc_debug(BINDER_DEBUG_BUFFER_ALLOC, | |
478 | "%d: binder_alloc_buf size %zd got buffer %pK size %zd\n", | |
479 | alloc->pid, size, buffer, buffer_size); | |
480 | ||
bde4a19f TK |
481 | has_page_addr = (void __user *) |
482 | (((uintptr_t)buffer->user_data + buffer_size) & PAGE_MASK); | |
74310e06 | 483 | WARN_ON(n && buffer_size != size); |
0c972a05 | 484 | end_page_addr = |
bde4a19f | 485 | (void __user *)PAGE_ALIGN((uintptr_t)buffer->user_data + size); |
0c972a05 TK |
486 | if (end_page_addr > has_page_addr) |
487 | end_page_addr = has_page_addr; | |
bde4a19f TK |
488 | ret = binder_update_page_range(alloc, 1, (void __user *) |
489 | PAGE_ALIGN((uintptr_t)buffer->user_data), end_page_addr); | |
57ada2fb TK |
490 | if (ret) |
491 | return ERR_PTR(ret); | |
0c972a05 | 492 | |
0c972a05 | 493 | if (buffer_size != size) { |
74310e06 | 494 | struct binder_buffer *new_buffer; |
0c972a05 | 495 | |
74310e06 SY |
496 | new_buffer = kzalloc(sizeof(*buffer), GFP_KERNEL); |
497 | if (!new_buffer) { | |
498 | pr_err("%s: %d failed to alloc new buffer struct\n", | |
499 | __func__, alloc->pid); | |
500 | goto err_alloc_buf_struct_failed; | |
501 | } | |
bde4a19f | 502 | new_buffer->user_data = (u8 __user *)buffer->user_data + size; |
0c972a05 TK |
503 | list_add(&new_buffer->entry, &buffer->entry); |
504 | new_buffer->free = 1; | |
505 | binder_insert_free_buffer(alloc, new_buffer); | |
506 | } | |
74310e06 SY |
507 | |
508 | rb_erase(best_fit, &alloc->free_buffers); | |
509 | buffer->free = 0; | |
7bada55a | 510 | buffer->allow_user_free = 0; |
74310e06 | 511 | binder_insert_allocated_buffer_locked(alloc, buffer); |
0c972a05 TK |
512 | binder_alloc_debug(BINDER_DEBUG_BUFFER_ALLOC, |
513 | "%d: binder_alloc_buf size %zd got %pK\n", | |
514 | alloc->pid, size, buffer); | |
515 | buffer->data_size = data_size; | |
516 | buffer->offsets_size = offsets_size; | |
517 | buffer->async_transaction = is_async; | |
518 | buffer->extra_buffers_size = extra_buffers_size; | |
261e7818 | 519 | buffer->pid = pid; |
a7dc1e6f | 520 | buffer->oneway_spam_suspect = false; |
0c972a05 TK |
521 | if (is_async) { |
522 | alloc->free_async_space -= size + sizeof(struct binder_buffer); | |
523 | binder_alloc_debug(BINDER_DEBUG_BUFFER_ALLOC_ASYNC, | |
524 | "%d: binder_alloc_buf size %zd async free %zd\n", | |
525 | alloc->pid, size, alloc->free_async_space); | |
261e7818 MC |
526 | if (alloc->free_async_space < alloc->buffer_size / 10) { |
527 | /* | |
528 | * Start detecting spammers once we have less than 20% | |
529 | * of async space left (which is less than 10% of total | |
530 | * buffer size). | |
531 | */ | |
a7dc1e6f HL |
532 | buffer->oneway_spam_suspect = debug_low_async_space_locked(alloc, pid); |
533 | } else { | |
534 | alloc->oneway_spam_detected = false; | |
261e7818 | 535 | } |
0c972a05 TK |
536 | } |
537 | return buffer; | |
74310e06 SY |
538 | |
539 | err_alloc_buf_struct_failed: | |
bde4a19f TK |
540 | binder_update_page_range(alloc, 0, (void __user *) |
541 | PAGE_ALIGN((uintptr_t)buffer->user_data), | |
6ae33b9c | 542 | end_page_addr); |
74310e06 | 543 | return ERR_PTR(-ENOMEM); |
0c972a05 TK |
544 | } |
545 | ||
546 | /** | |
547 | * binder_alloc_new_buf() - Allocate a new binder buffer | |
548 | * @alloc: binder_alloc for this proc | |
549 | * @data_size: size of user data buffer | |
550 | * @offsets_size: user specified buffer offset | |
551 | * @extra_buffers_size: size of extra space for meta-data (eg, security context) | |
552 | * @is_async: buffer for async transaction | |
261e7818 | 553 | * @pid: pid to attribute allocation to (used for debugging) |
0c972a05 TK |
554 | * |
555 | * Allocate a new buffer given the requested sizes. Returns | |
556 | * the kernel version of the buffer pointer. The size allocated | |
557 | * is the sum of the three given sizes (each rounded up to | |
558 | * pointer-sized boundary) | |
559 | * | |
560 | * Return: The allocated buffer or %NULL if error | |
561 | */ | |
562 | struct binder_buffer *binder_alloc_new_buf(struct binder_alloc *alloc, | |
563 | size_t data_size, | |
564 | size_t offsets_size, | |
565 | size_t extra_buffers_size, | |
261e7818 MC |
566 | int is_async, |
567 | int pid) | |
0c972a05 TK |
568 | { |
569 | struct binder_buffer *buffer; | |
570 | ||
571 | mutex_lock(&alloc->mutex); | |
572 | buffer = binder_alloc_new_buf_locked(alloc, data_size, offsets_size, | |
261e7818 | 573 | extra_buffers_size, is_async, pid); |
0c972a05 TK |
574 | mutex_unlock(&alloc->mutex); |
575 | return buffer; | |
576 | } | |
577 | ||
bde4a19f | 578 | static void __user *buffer_start_page(struct binder_buffer *buffer) |
0c972a05 | 579 | { |
bde4a19f | 580 | return (void __user *)((uintptr_t)buffer->user_data & PAGE_MASK); |
0c972a05 TK |
581 | } |
582 | ||
bde4a19f | 583 | static void __user *prev_buffer_end_page(struct binder_buffer *buffer) |
0c972a05 | 584 | { |
bde4a19f TK |
585 | return (void __user *) |
586 | (((uintptr_t)(buffer->user_data) - 1) & PAGE_MASK); | |
0c972a05 TK |
587 | } |
588 | ||
589 | static void binder_delete_free_buffer(struct binder_alloc *alloc, | |
590 | struct binder_buffer *buffer) | |
591 | { | |
592 | struct binder_buffer *prev, *next = NULL; | |
74310e06 | 593 | bool to_free = true; |
4df9772c | 594 | |
0c972a05 | 595 | BUG_ON(alloc->buffers.next == &buffer->entry); |
e2176219 | 596 | prev = binder_buffer_prev(buffer); |
0c972a05 | 597 | BUG_ON(!prev->free); |
74310e06 SY |
598 | if (prev_buffer_end_page(prev) == buffer_start_page(buffer)) { |
599 | to_free = false; | |
0c972a05 | 600 | binder_alloc_debug(BINDER_DEBUG_BUFFER_ALLOC, |
74310e06 | 601 | "%d: merge free, buffer %pK share page with %pK\n", |
bde4a19f TK |
602 | alloc->pid, buffer->user_data, |
603 | prev->user_data); | |
0c972a05 TK |
604 | } |
605 | ||
606 | if (!list_is_last(&buffer->entry, &alloc->buffers)) { | |
e2176219 | 607 | next = binder_buffer_next(buffer); |
74310e06 SY |
608 | if (buffer_start_page(next) == buffer_start_page(buffer)) { |
609 | to_free = false; | |
0c972a05 | 610 | binder_alloc_debug(BINDER_DEBUG_BUFFER_ALLOC, |
74310e06 SY |
611 | "%d: merge free, buffer %pK share page with %pK\n", |
612 | alloc->pid, | |
bde4a19f TK |
613 | buffer->user_data, |
614 | next->user_data); | |
0c972a05 TK |
615 | } |
616 | } | |
74310e06 | 617 | |
bde4a19f | 618 | if (PAGE_ALIGNED(buffer->user_data)) { |
74310e06 SY |
619 | binder_alloc_debug(BINDER_DEBUG_BUFFER_ALLOC, |
620 | "%d: merge free, buffer start %pK is page aligned\n", | |
bde4a19f | 621 | alloc->pid, buffer->user_data); |
74310e06 SY |
622 | to_free = false; |
623 | } | |
624 | ||
625 | if (to_free) { | |
0c972a05 | 626 | binder_alloc_debug(BINDER_DEBUG_BUFFER_ALLOC, |
74310e06 | 627 | "%d: merge free, buffer %pK do not share page with %pK or %pK\n", |
bde4a19f TK |
628 | alloc->pid, buffer->user_data, |
629 | prev->user_data, | |
630 | next ? next->user_data : NULL); | |
74310e06 | 631 | binder_update_page_range(alloc, 0, buffer_start_page(buffer), |
6ae33b9c | 632 | buffer_start_page(buffer) + PAGE_SIZE); |
0c972a05 | 633 | } |
74310e06 SY |
634 | list_del(&buffer->entry); |
635 | kfree(buffer); | |
0c972a05 TK |
636 | } |
637 | ||
638 | static void binder_free_buf_locked(struct binder_alloc *alloc, | |
639 | struct binder_buffer *buffer) | |
640 | { | |
641 | size_t size, buffer_size; | |
642 | ||
643 | buffer_size = binder_alloc_buffer_size(alloc, buffer); | |
644 | ||
645 | size = ALIGN(buffer->data_size, sizeof(void *)) + | |
646 | ALIGN(buffer->offsets_size, sizeof(void *)) + | |
647 | ALIGN(buffer->extra_buffers_size, sizeof(void *)); | |
648 | ||
649 | binder_alloc_debug(BINDER_DEBUG_BUFFER_ALLOC, | |
650 | "%d: binder_free_buf %pK size %zd buffer_size %zd\n", | |
651 | alloc->pid, buffer, size, buffer_size); | |
652 | ||
653 | BUG_ON(buffer->free); | |
654 | BUG_ON(size > buffer_size); | |
655 | BUG_ON(buffer->transaction != NULL); | |
bde4a19f TK |
656 | BUG_ON(buffer->user_data < alloc->buffer); |
657 | BUG_ON(buffer->user_data > alloc->buffer + alloc->buffer_size); | |
0c972a05 TK |
658 | |
659 | if (buffer->async_transaction) { | |
cfd0d84b | 660 | alloc->free_async_space += buffer_size + sizeof(struct binder_buffer); |
0c972a05 TK |
661 | |
662 | binder_alloc_debug(BINDER_DEBUG_BUFFER_ALLOC_ASYNC, | |
663 | "%d: binder_free_buf size %zd async free %zd\n", | |
664 | alloc->pid, size, alloc->free_async_space); | |
665 | } | |
666 | ||
667 | binder_update_page_range(alloc, 0, | |
bde4a19f TK |
668 | (void __user *)PAGE_ALIGN((uintptr_t)buffer->user_data), |
669 | (void __user *)(((uintptr_t) | |
670 | buffer->user_data + buffer_size) & PAGE_MASK)); | |
0c972a05 TK |
671 | |
672 | rb_erase(&buffer->rb_node, &alloc->allocated_buffers); | |
673 | buffer->free = 1; | |
674 | if (!list_is_last(&buffer->entry, &alloc->buffers)) { | |
e2176219 | 675 | struct binder_buffer *next = binder_buffer_next(buffer); |
0c972a05 TK |
676 | |
677 | if (next->free) { | |
678 | rb_erase(&next->rb_node, &alloc->free_buffers); | |
679 | binder_delete_free_buffer(alloc, next); | |
680 | } | |
681 | } | |
682 | if (alloc->buffers.next != &buffer->entry) { | |
e2176219 | 683 | struct binder_buffer *prev = binder_buffer_prev(buffer); |
0c972a05 TK |
684 | |
685 | if (prev->free) { | |
686 | binder_delete_free_buffer(alloc, buffer); | |
687 | rb_erase(&prev->rb_node, &alloc->free_buffers); | |
688 | buffer = prev; | |
689 | } | |
690 | } | |
691 | binder_insert_free_buffer(alloc, buffer); | |
692 | } | |
693 | ||
0f966cba TK |
694 | static void binder_alloc_clear_buf(struct binder_alloc *alloc, |
695 | struct binder_buffer *buffer); | |
0c972a05 TK |
696 | /** |
697 | * binder_alloc_free_buf() - free a binder buffer | |
698 | * @alloc: binder_alloc for this proc | |
699 | * @buffer: kernel pointer to buffer | |
700 | * | |
4b463822 | 701 | * Free the buffer allocated via binder_alloc_new_buf() |
0c972a05 TK |
702 | */ |
703 | void binder_alloc_free_buf(struct binder_alloc *alloc, | |
704 | struct binder_buffer *buffer) | |
705 | { | |
0f966cba TK |
706 | /* |
707 | * We could eliminate the call to binder_alloc_clear_buf() | |
708 | * from binder_alloc_deferred_release() by moving this to | |
709 | * binder_alloc_free_buf_locked(). However, that could | |
710 | * increase contention for the alloc mutex if clear_on_free | |
711 | * is used frequently for large buffers. The mutex is not | |
712 | * needed for correctness here. | |
713 | */ | |
714 | if (buffer->clear_on_free) { | |
715 | binder_alloc_clear_buf(alloc, buffer); | |
716 | buffer->clear_on_free = false; | |
717 | } | |
0c972a05 TK |
718 | mutex_lock(&alloc->mutex); |
719 | binder_free_buf_locked(alloc, buffer); | |
720 | mutex_unlock(&alloc->mutex); | |
721 | } | |
722 | ||
723 | /** | |
724 | * binder_alloc_mmap_handler() - map virtual address space for proc | |
725 | * @alloc: alloc structure for this proc | |
726 | * @vma: vma passed to mmap() | |
727 | * | |
728 | * Called by binder_mmap() to initialize the space specified in | |
729 | * vma for allocating binder buffers | |
730 | * | |
731 | * Return: | |
732 | * 0 = success | |
733 | * -EBUSY = address space already mapped | |
734 | * -ENOMEM = failed to map memory to given address space | |
735 | */ | |
736 | int binder_alloc_mmap_handler(struct binder_alloc *alloc, | |
737 | struct vm_area_struct *vma) | |
738 | { | |
739 | int ret; | |
0c972a05 TK |
740 | const char *failure_string; |
741 | struct binder_buffer *buffer; | |
742 | ||
3ce00bb7 CL |
743 | if (unlikely(vma->vm_mm != alloc->mm)) { |
744 | ret = -EINVAL; | |
745 | failure_string = "invalid vma->vm_mm"; | |
746 | goto err_invalid_mm; | |
747 | } | |
748 | ||
0c972a05 | 749 | mutex_lock(&binder_alloc_mmap_lock); |
a7a74d7f | 750 | if (alloc->buffer_size) { |
0c972a05 TK |
751 | ret = -EBUSY; |
752 | failure_string = "already mapped"; | |
753 | goto err_already_mapped; | |
754 | } | |
a7a74d7f JH |
755 | alloc->buffer_size = min_t(unsigned long, vma->vm_end - vma->vm_start, |
756 | SZ_4M); | |
757 | mutex_unlock(&binder_alloc_mmap_lock); | |
0c972a05 | 758 | |
bde4a19f | 759 | alloc->buffer = (void __user *)vma->vm_start; |
0c972a05 | 760 | |
45d02f79 | 761 | alloc->pages = kcalloc(alloc->buffer_size / PAGE_SIZE, |
6396bb22 | 762 | sizeof(alloc->pages[0]), |
0c972a05 TK |
763 | GFP_KERNEL); |
764 | if (alloc->pages == NULL) { | |
765 | ret = -ENOMEM; | |
766 | failure_string = "alloc page array"; | |
767 | goto err_alloc_pages_failed; | |
768 | } | |
0c972a05 | 769 | |
74310e06 SY |
770 | buffer = kzalloc(sizeof(*buffer), GFP_KERNEL); |
771 | if (!buffer) { | |
0c972a05 | 772 | ret = -ENOMEM; |
74310e06 SY |
773 | failure_string = "alloc buffer struct"; |
774 | goto err_alloc_buf_struct_failed; | |
0c972a05 | 775 | } |
74310e06 | 776 | |
bde4a19f | 777 | buffer->user_data = alloc->buffer; |
0c972a05 TK |
778 | list_add(&buffer->entry, &alloc->buffers); |
779 | buffer->free = 1; | |
780 | binder_insert_free_buffer(alloc, buffer); | |
781 | alloc->free_async_space = alloc->buffer_size / 2; | |
0fa53349 CL |
782 | |
783 | /* Signal binder_alloc is fully initialized */ | |
784 | binder_alloc_set_vma(alloc, vma); | |
0c972a05 TK |
785 | |
786 | return 0; | |
787 | ||
74310e06 | 788 | err_alloc_buf_struct_failed: |
0c972a05 TK |
789 | kfree(alloc->pages); |
790 | alloc->pages = NULL; | |
791 | err_alloc_pages_failed: | |
0c972a05 | 792 | alloc->buffer = NULL; |
a7a74d7f JH |
793 | mutex_lock(&binder_alloc_mmap_lock); |
794 | alloc->buffer_size = 0; | |
0c972a05 TK |
795 | err_already_mapped: |
796 | mutex_unlock(&binder_alloc_mmap_lock); | |
3ce00bb7 | 797 | err_invalid_mm: |
128f3804 SY |
798 | binder_alloc_debug(BINDER_DEBUG_USER_ERROR, |
799 | "%s: %d %lx-%lx %s failed %d\n", __func__, | |
800 | alloc->pid, vma->vm_start, vma->vm_end, | |
801 | failure_string, ret); | |
0c972a05 TK |
802 | return ret; |
803 | } | |
804 | ||
805 | ||
806 | void binder_alloc_deferred_release(struct binder_alloc *alloc) | |
807 | { | |
808 | struct rb_node *n; | |
809 | int buffers, page_count; | |
74310e06 | 810 | struct binder_buffer *buffer; |
0c972a05 | 811 | |
0c972a05 TK |
812 | buffers = 0; |
813 | mutex_lock(&alloc->mutex); | |
c0fd2101 | 814 | BUG_ON(alloc->vma); |
da1b9564 | 815 | |
0c972a05 | 816 | while ((n = rb_first(&alloc->allocated_buffers))) { |
0c972a05 TK |
817 | buffer = rb_entry(n, struct binder_buffer, rb_node); |
818 | ||
819 | /* Transaction should already have been freed */ | |
820 | BUG_ON(buffer->transaction); | |
821 | ||
0f966cba TK |
822 | if (buffer->clear_on_free) { |
823 | binder_alloc_clear_buf(alloc, buffer); | |
824 | buffer->clear_on_free = false; | |
825 | } | |
0c972a05 TK |
826 | binder_free_buf_locked(alloc, buffer); |
827 | buffers++; | |
828 | } | |
829 | ||
74310e06 SY |
830 | while (!list_empty(&alloc->buffers)) { |
831 | buffer = list_first_entry(&alloc->buffers, | |
832 | struct binder_buffer, entry); | |
833 | WARN_ON(!buffer->free); | |
834 | ||
835 | list_del(&buffer->entry); | |
836 | WARN_ON_ONCE(!list_empty(&alloc->buffers)); | |
837 | kfree(buffer); | |
838 | } | |
839 | ||
0c972a05 TK |
840 | page_count = 0; |
841 | if (alloc->pages) { | |
842 | int i; | |
843 | ||
844 | for (i = 0; i < alloc->buffer_size / PAGE_SIZE; i++) { | |
bde4a19f | 845 | void __user *page_addr; |
f2517eb7 | 846 | bool on_lru; |
0c972a05 | 847 | |
f2517eb7 | 848 | if (!alloc->pages[i].page_ptr) |
0c972a05 TK |
849 | continue; |
850 | ||
f2517eb7 SY |
851 | on_lru = list_lru_del(&binder_alloc_lru, |
852 | &alloc->pages[i].lru); | |
0c972a05 TK |
853 | page_addr = alloc->buffer + i * PAGE_SIZE; |
854 | binder_alloc_debug(BINDER_DEBUG_BUFFER_ALLOC, | |
f2517eb7 SY |
855 | "%s: %d: page %d at %pK %s\n", |
856 | __func__, alloc->pid, i, page_addr, | |
857 | on_lru ? "on lru" : "active"); | |
f2517eb7 | 858 | __free_page(alloc->pages[i].page_ptr); |
0c972a05 TK |
859 | page_count++; |
860 | } | |
861 | kfree(alloc->pages); | |
0c972a05 TK |
862 | } |
863 | mutex_unlock(&alloc->mutex); | |
e66b77e5 CL |
864 | if (alloc->mm) |
865 | mmdrop(alloc->mm); | |
0c972a05 TK |
866 | |
867 | binder_alloc_debug(BINDER_DEBUG_OPEN_CLOSE, | |
868 | "%s: %d buffers %d, pages %d\n", | |
869 | __func__, alloc->pid, buffers, page_count); | |
870 | } | |
871 | ||
872 | static void print_binder_buffer(struct seq_file *m, const char *prefix, | |
873 | struct binder_buffer *buffer) | |
874 | { | |
b05a68e9 | 875 | seq_printf(m, "%s %d: %pK size %zd:%zd:%zd %s\n", |
bde4a19f | 876 | prefix, buffer->debug_id, buffer->user_data, |
0c972a05 | 877 | buffer->data_size, buffer->offsets_size, |
b05a68e9 | 878 | buffer->extra_buffers_size, |
0c972a05 TK |
879 | buffer->transaction ? "active" : "delivered"); |
880 | } | |
881 | ||
882 | /** | |
883 | * binder_alloc_print_allocated() - print buffer info | |
884 | * @m: seq_file for output via seq_printf() | |
885 | * @alloc: binder_alloc for this proc | |
886 | * | |
887 | * Prints information about every buffer associated with | |
888 | * the binder_alloc state to the given seq_file | |
889 | */ | |
890 | void binder_alloc_print_allocated(struct seq_file *m, | |
891 | struct binder_alloc *alloc) | |
892 | { | |
893 | struct rb_node *n; | |
894 | ||
895 | mutex_lock(&alloc->mutex); | |
896 | for (n = rb_first(&alloc->allocated_buffers); n != NULL; n = rb_next(n)) | |
897 | print_binder_buffer(m, " buffer", | |
898 | rb_entry(n, struct binder_buffer, rb_node)); | |
899 | mutex_unlock(&alloc->mutex); | |
900 | } | |
901 | ||
8ef4665a SY |
902 | /** |
903 | * binder_alloc_print_pages() - print page usage | |
904 | * @m: seq_file for output via seq_printf() | |
905 | * @alloc: binder_alloc for this proc | |
906 | */ | |
907 | void binder_alloc_print_pages(struct seq_file *m, | |
908 | struct binder_alloc *alloc) | |
909 | { | |
910 | struct binder_lru_page *page; | |
911 | int i; | |
912 | int active = 0; | |
913 | int lru = 0; | |
914 | int free = 0; | |
915 | ||
916 | mutex_lock(&alloc->mutex); | |
8eb52a1e JH |
917 | /* |
918 | * Make sure the binder_alloc is fully initialized, otherwise we might | |
919 | * read inconsistent state. | |
920 | */ | |
b15655b1 CL |
921 | if (binder_alloc_get_vma(alloc) != NULL) { |
922 | for (i = 0; i < alloc->buffer_size / PAGE_SIZE; i++) { | |
923 | page = &alloc->pages[i]; | |
924 | if (!page->page_ptr) | |
925 | free++; | |
926 | else if (list_empty(&page->lru)) | |
927 | active++; | |
928 | else | |
929 | lru++; | |
930 | } | |
44e602b4 | 931 | } |
8ef4665a SY |
932 | mutex_unlock(&alloc->mutex); |
933 | seq_printf(m, " pages: %d:%d:%d\n", active, lru, free); | |
8d9a3ab6 | 934 | seq_printf(m, " pages high watermark: %zu\n", alloc->pages_high); |
8ef4665a SY |
935 | } |
936 | ||
0c972a05 TK |
937 | /** |
938 | * binder_alloc_get_allocated_count() - return count of buffers | |
939 | * @alloc: binder_alloc for this proc | |
940 | * | |
941 | * Return: count of allocated buffers | |
942 | */ | |
943 | int binder_alloc_get_allocated_count(struct binder_alloc *alloc) | |
944 | { | |
945 | struct rb_node *n; | |
946 | int count = 0; | |
947 | ||
948 | mutex_lock(&alloc->mutex); | |
949 | for (n = rb_first(&alloc->allocated_buffers); n != NULL; n = rb_next(n)) | |
950 | count++; | |
951 | mutex_unlock(&alloc->mutex); | |
952 | return count; | |
953 | } | |
954 | ||
955 | ||
956 | /** | |
957 | * binder_alloc_vma_close() - invalidate address space | |
958 | * @alloc: binder_alloc for this proc | |
959 | * | |
960 | * Called from binder_vma_close() when releasing address space. | |
961 | * Clears alloc->vma to prevent new incoming transactions from | |
962 | * allocating more buffers. | |
963 | */ | |
964 | void binder_alloc_vma_close(struct binder_alloc *alloc) | |
965 | { | |
0fa53349 | 966 | binder_alloc_set_vma(alloc, NULL); |
0c972a05 TK |
967 | } |
968 | ||
f2517eb7 SY |
969 | /** |
970 | * binder_alloc_free_page() - shrinker callback to free pages | |
971 | * @item: item to free | |
972 | * @lock: lock protecting the item | |
973 | * @cb_arg: callback argument | |
974 | * | |
975 | * Called from list_lru_walk() in binder_shrink_scan() to free | |
976 | * up pages when the system is under memory pressure. | |
977 | */ | |
978 | enum lru_status binder_alloc_free_page(struct list_head *item, | |
979 | struct list_lru_one *lru, | |
980 | spinlock_t *lock, | |
981 | void *cb_arg) | |
324fa64c | 982 | __must_hold(lock) |
f2517eb7 SY |
983 | { |
984 | struct mm_struct *mm = NULL; | |
985 | struct binder_lru_page *page = container_of(item, | |
986 | struct binder_lru_page, | |
987 | lru); | |
988 | struct binder_alloc *alloc; | |
989 | uintptr_t page_addr; | |
990 | size_t index; | |
a1b2289c | 991 | struct vm_area_struct *vma; |
f2517eb7 SY |
992 | |
993 | alloc = page->alloc; | |
994 | if (!mutex_trylock(&alloc->mutex)) | |
995 | goto err_get_alloc_mutex_failed; | |
996 | ||
997 | if (!page->page_ptr) | |
998 | goto err_page_already_freed; | |
999 | ||
1000 | index = page - alloc->pages; | |
1001 | page_addr = (uintptr_t)alloc->buffer + index * PAGE_SIZE; | |
5cec2d2e | 1002 | |
e66b77e5 | 1003 | mm = alloc->mm; |
5cec2d2e TK |
1004 | if (!mmget_not_zero(mm)) |
1005 | goto err_mmget; | |
d8ed45c5 | 1006 | if (!mmap_read_trylock(mm)) |
3e4e28c5 | 1007 | goto err_mmap_read_lock_failed; |
da1b9564 | 1008 | vma = binder_alloc_get_vma(alloc); |
a1b2289c SY |
1009 | |
1010 | list_lru_isolate(lru, item); | |
1011 | spin_unlock(lock); | |
f2517eb7 | 1012 | |
a1b2289c | 1013 | if (vma) { |
e41e164c SY |
1014 | trace_binder_unmap_user_start(alloc, index); |
1015 | ||
e9adcfec | 1016 | zap_page_range_single(vma, page_addr, PAGE_SIZE, NULL); |
f2517eb7 | 1017 | |
e41e164c | 1018 | trace_binder_unmap_user_end(alloc, index); |
f2517eb7 | 1019 | } |
d8ed45c5 | 1020 | mmap_read_unlock(mm); |
f867c771 | 1021 | mmput_async(mm); |
f2517eb7 | 1022 | |
e41e164c SY |
1023 | trace_binder_unmap_kernel_start(alloc, index); |
1024 | ||
f2517eb7 SY |
1025 | __free_page(page->page_ptr); |
1026 | page->page_ptr = NULL; | |
1027 | ||
e41e164c SY |
1028 | trace_binder_unmap_kernel_end(alloc, index); |
1029 | ||
a1b2289c | 1030 | spin_lock(lock); |
f2517eb7 | 1031 | mutex_unlock(&alloc->mutex); |
a1b2289c | 1032 | return LRU_REMOVED_RETRY; |
f2517eb7 | 1033 | |
3e4e28c5 | 1034 | err_mmap_read_lock_failed: |
a1b2289c | 1035 | mmput_async(mm); |
a0c2baaf | 1036 | err_mmget: |
f2517eb7 SY |
1037 | err_page_already_freed: |
1038 | mutex_unlock(&alloc->mutex); | |
1039 | err_get_alloc_mutex_failed: | |
1040 | return LRU_SKIP; | |
1041 | } | |
1042 | ||
1043 | static unsigned long | |
1044 | binder_shrink_count(struct shrinker *shrink, struct shrink_control *sc) | |
1045 | { | |
b2fb28de | 1046 | return list_lru_count(&binder_alloc_lru); |
f2517eb7 SY |
1047 | } |
1048 | ||
1049 | static unsigned long | |
1050 | binder_shrink_scan(struct shrinker *shrink, struct shrink_control *sc) | |
1051 | { | |
b2fb28de | 1052 | return list_lru_walk(&binder_alloc_lru, binder_alloc_free_page, |
f2517eb7 | 1053 | NULL, sc->nr_to_scan); |
f2517eb7 SY |
1054 | } |
1055 | ||
de7bbe3d | 1056 | static struct shrinker binder_shrinker = { |
f2517eb7 SY |
1057 | .count_objects = binder_shrink_count, |
1058 | .scan_objects = binder_shrink_scan, | |
1059 | .seeks = DEFAULT_SEEKS, | |
1060 | }; | |
1061 | ||
0c972a05 TK |
1062 | /** |
1063 | * binder_alloc_init() - called by binder_open() for per-proc initialization | |
1064 | * @alloc: binder_alloc for this proc | |
1065 | * | |
1066 | * Called from binder_open() to initialize binder_alloc fields for | |
1067 | * new binder proc | |
1068 | */ | |
1069 | void binder_alloc_init(struct binder_alloc *alloc) | |
1070 | { | |
0c972a05 | 1071 | alloc->pid = current->group_leader->pid; |
e66b77e5 CL |
1072 | alloc->mm = current->mm; |
1073 | mmgrab(alloc->mm); | |
0c972a05 | 1074 | mutex_init(&alloc->mutex); |
957ccc2b | 1075 | INIT_LIST_HEAD(&alloc->buffers); |
0c972a05 TK |
1076 | } |
1077 | ||
533dfb25 | 1078 | int binder_alloc_shrinker_init(void) |
f2517eb7 | 1079 | { |
533dfb25 TH |
1080 | int ret = list_lru_init(&binder_alloc_lru); |
1081 | ||
1082 | if (ret == 0) { | |
e33c267a | 1083 | ret = register_shrinker(&binder_shrinker, "android-binder"); |
533dfb25 TH |
1084 | if (ret) |
1085 | list_lru_destroy(&binder_alloc_lru); | |
1086 | } | |
1087 | return ret; | |
f2517eb7 | 1088 | } |
1a7c3d9b | 1089 | |
adb9743d QZ |
1090 | void binder_alloc_shrinker_exit(void) |
1091 | { | |
1092 | unregister_shrinker(&binder_shrinker); | |
1093 | list_lru_destroy(&binder_alloc_lru); | |
1094 | } | |
1095 | ||
1a7c3d9b TK |
1096 | /** |
1097 | * check_buffer() - verify that buffer/offset is safe to access | |
1098 | * @alloc: binder_alloc for this proc | |
1099 | * @buffer: binder buffer to be accessed | |
1100 | * @offset: offset into @buffer data | |
1101 | * @bytes: bytes to access from offset | |
1102 | * | |
1103 | * Check that the @offset/@bytes are within the size of the given | |
1104 | * @buffer and that the buffer is currently active and not freeable. | |
1105 | * Offsets must also be multiples of sizeof(u32). The kernel is | |
1106 | * allowed to touch the buffer in two cases: | |
1107 | * | |
1108 | * 1) when the buffer is being created: | |
1109 | * (buffer->free == 0 && buffer->allow_user_free == 0) | |
1110 | * 2) when the buffer is being torn down: | |
1111 | * (buffer->free == 0 && buffer->transaction == NULL). | |
1112 | * | |
1113 | * Return: true if the buffer is safe to access | |
1114 | */ | |
1115 | static inline bool check_buffer(struct binder_alloc *alloc, | |
1116 | struct binder_buffer *buffer, | |
1117 | binder_size_t offset, size_t bytes) | |
1118 | { | |
1119 | size_t buffer_size = binder_alloc_buffer_size(alloc, buffer); | |
1120 | ||
1121 | return buffer_size >= bytes && | |
1122 | offset <= buffer_size - bytes && | |
1123 | IS_ALIGNED(offset, sizeof(u32)) && | |
1124 | !buffer->free && | |
1125 | (!buffer->allow_user_free || !buffer->transaction); | |
1126 | } | |
1127 | ||
1128 | /** | |
1129 | * binder_alloc_get_page() - get kernel pointer for given buffer offset | |
1130 | * @alloc: binder_alloc for this proc | |
1131 | * @buffer: binder buffer to be accessed | |
1132 | * @buffer_offset: offset into @buffer data | |
1133 | * @pgoffp: address to copy final page offset to | |
1134 | * | |
1135 | * Lookup the struct page corresponding to the address | |
bde4a19f | 1136 | * at @buffer_offset into @buffer->user_data. If @pgoffp is not |
1a7c3d9b TK |
1137 | * NULL, the byte-offset into the page is written there. |
1138 | * | |
1139 | * The caller is responsible to ensure that the offset points | |
1140 | * to a valid address within the @buffer and that @buffer is | |
1141 | * not freeable by the user. Since it can't be freed, we are | |
1142 | * guaranteed that the corresponding elements of @alloc->pages[] | |
1143 | * cannot change. | |
1144 | * | |
1145 | * Return: struct page | |
1146 | */ | |
1147 | static struct page *binder_alloc_get_page(struct binder_alloc *alloc, | |
1148 | struct binder_buffer *buffer, | |
1149 | binder_size_t buffer_offset, | |
1150 | pgoff_t *pgoffp) | |
1151 | { | |
1152 | binder_size_t buffer_space_offset = buffer_offset + | |
bde4a19f | 1153 | (buffer->user_data - alloc->buffer); |
1a7c3d9b TK |
1154 | pgoff_t pgoff = buffer_space_offset & ~PAGE_MASK; |
1155 | size_t index = buffer_space_offset >> PAGE_SHIFT; | |
1156 | struct binder_lru_page *lru_page; | |
1157 | ||
1158 | lru_page = &alloc->pages[index]; | |
1159 | *pgoffp = pgoff; | |
1160 | return lru_page->page_ptr; | |
1161 | } | |
1162 | ||
0f966cba TK |
1163 | /** |
1164 | * binder_alloc_clear_buf() - zero out buffer | |
1165 | * @alloc: binder_alloc for this proc | |
1166 | * @buffer: binder buffer to be cleared | |
1167 | * | |
1168 | * memset the given buffer to 0 | |
1169 | */ | |
1170 | static void binder_alloc_clear_buf(struct binder_alloc *alloc, | |
1171 | struct binder_buffer *buffer) | |
1172 | { | |
1173 | size_t bytes = binder_alloc_buffer_size(alloc, buffer); | |
1174 | binder_size_t buffer_offset = 0; | |
1175 | ||
1176 | while (bytes) { | |
1177 | unsigned long size; | |
1178 | struct page *page; | |
1179 | pgoff_t pgoff; | |
0f966cba TK |
1180 | |
1181 | page = binder_alloc_get_page(alloc, buffer, | |
1182 | buffer_offset, &pgoff); | |
1183 | size = min_t(size_t, bytes, PAGE_SIZE - pgoff); | |
26eff2d6 | 1184 | memset_page(page, pgoff, 0, size); |
0f966cba TK |
1185 | bytes -= size; |
1186 | buffer_offset += size; | |
1187 | } | |
1188 | } | |
1189 | ||
1a7c3d9b TK |
1190 | /** |
1191 | * binder_alloc_copy_user_to_buffer() - copy src user to tgt user | |
1192 | * @alloc: binder_alloc for this proc | |
1193 | * @buffer: binder buffer to be accessed | |
1194 | * @buffer_offset: offset into @buffer data | |
1195 | * @from: userspace pointer to source buffer | |
1196 | * @bytes: bytes to copy | |
1197 | * | |
1198 | * Copy bytes from source userspace to target buffer. | |
1199 | * | |
1200 | * Return: bytes remaining to be copied | |
1201 | */ | |
1202 | unsigned long | |
1203 | binder_alloc_copy_user_to_buffer(struct binder_alloc *alloc, | |
1204 | struct binder_buffer *buffer, | |
1205 | binder_size_t buffer_offset, | |
1206 | const void __user *from, | |
1207 | size_t bytes) | |
1208 | { | |
1209 | if (!check_buffer(alloc, buffer, buffer_offset, bytes)) | |
1210 | return bytes; | |
1211 | ||
1212 | while (bytes) { | |
1213 | unsigned long size; | |
1214 | unsigned long ret; | |
1215 | struct page *page; | |
1216 | pgoff_t pgoff; | |
1217 | void *kptr; | |
1218 | ||
1219 | page = binder_alloc_get_page(alloc, buffer, | |
1220 | buffer_offset, &pgoff); | |
1221 | size = min_t(size_t, bytes, PAGE_SIZE - pgoff); | |
1d625960 | 1222 | kptr = kmap_local_page(page) + pgoff; |
1a7c3d9b | 1223 | ret = copy_from_user(kptr, from, size); |
1d625960 | 1224 | kunmap_local(kptr); |
1a7c3d9b TK |
1225 | if (ret) |
1226 | return bytes - size + ret; | |
1227 | bytes -= size; | |
1228 | from += size; | |
1229 | buffer_offset += size; | |
1230 | } | |
1231 | return 0; | |
1232 | } | |
8ced0c62 | 1233 | |
bb4a2e48 TK |
1234 | static int binder_alloc_do_buffer_copy(struct binder_alloc *alloc, |
1235 | bool to_buffer, | |
1236 | struct binder_buffer *buffer, | |
1237 | binder_size_t buffer_offset, | |
1238 | void *ptr, | |
1239 | size_t bytes) | |
8ced0c62 TK |
1240 | { |
1241 | /* All copies must be 32-bit aligned and 32-bit size */ | |
bb4a2e48 TK |
1242 | if (!check_buffer(alloc, buffer, buffer_offset, bytes)) |
1243 | return -EINVAL; | |
8ced0c62 TK |
1244 | |
1245 | while (bytes) { | |
1246 | unsigned long size; | |
1247 | struct page *page; | |
1248 | pgoff_t pgoff; | |
8ced0c62 TK |
1249 | |
1250 | page = binder_alloc_get_page(alloc, buffer, | |
1251 | buffer_offset, &pgoff); | |
1252 | size = min_t(size_t, bytes, PAGE_SIZE - pgoff); | |
8ced0c62 | 1253 | if (to_buffer) |
e88a6a8f | 1254 | memcpy_to_page(page, pgoff, ptr, size); |
8ced0c62 | 1255 | else |
e88a6a8f | 1256 | memcpy_from_page(ptr, page, pgoff, size); |
8ced0c62 TK |
1257 | bytes -= size; |
1258 | pgoff = 0; | |
1259 | ptr = ptr + size; | |
1260 | buffer_offset += size; | |
1261 | } | |
bb4a2e48 | 1262 | return 0; |
8ced0c62 TK |
1263 | } |
1264 | ||
bb4a2e48 TK |
1265 | int binder_alloc_copy_to_buffer(struct binder_alloc *alloc, |
1266 | struct binder_buffer *buffer, | |
1267 | binder_size_t buffer_offset, | |
1268 | void *src, | |
1269 | size_t bytes) | |
8ced0c62 | 1270 | { |
bb4a2e48 TK |
1271 | return binder_alloc_do_buffer_copy(alloc, true, buffer, buffer_offset, |
1272 | src, bytes); | |
8ced0c62 TK |
1273 | } |
1274 | ||
bb4a2e48 TK |
1275 | int binder_alloc_copy_from_buffer(struct binder_alloc *alloc, |
1276 | void *dest, | |
1277 | struct binder_buffer *buffer, | |
1278 | binder_size_t buffer_offset, | |
1279 | size_t bytes) | |
8ced0c62 | 1280 | { |
bb4a2e48 TK |
1281 | return binder_alloc_do_buffer_copy(alloc, false, buffer, buffer_offset, |
1282 | dest, bytes); | |
8ced0c62 TK |
1283 | } |
1284 |