Merge branches 'acpi-pm' and 'acpi-docs'
[linux-2.6-microblaze.git] / drivers / android / binder_alloc.c
1 // SPDX-License-Identifier: GPL-2.0-only
2 /* binder_alloc.c
3  *
4  * Android IPC Subsystem
5  *
6  * Copyright (C) 2007-2017 Google, Inc.
7  */
8
9 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
10
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>
20 #include <linux/list_lru.h>
21 #include <linux/ratelimit.h>
22 #include <asm/cacheflush.h>
23 #include <linux/uaccess.h>
24 #include <linux/highmem.h>
25 #include <linux/sizes.h>
26 #include "binder_alloc.h"
27 #include "binder_trace.h"
28
29 struct list_lru binder_alloc_lru;
30
31 static DEFINE_MUTEX(binder_alloc_mmap_lock);
32
33 enum {
34         BINDER_DEBUG_USER_ERROR             = 1U << 0,
35         BINDER_DEBUG_OPEN_CLOSE             = 1U << 1,
36         BINDER_DEBUG_BUFFER_ALLOC           = 1U << 2,
37         BINDER_DEBUG_BUFFER_ALLOC_ASYNC     = 1U << 3,
38 };
39 static uint32_t binder_alloc_debug_mask = BINDER_DEBUG_USER_ERROR;
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) \
47                         pr_info_ratelimited(x); \
48         } while (0)
49
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
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))
64                 return alloc->buffer + alloc->buffer_size - buffer->user_data;
65         return binder_buffer_next(buffer)->user_data - buffer->user_data;
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
115                 if (new_buffer->user_data < buffer->user_data)
116                         p = &parent->rb_left;
117                 else if (new_buffer->user_data > buffer->user_data)
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
126 static struct binder_buffer *binder_alloc_prepare_to_free_locked(
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;
132         void __user *uptr;
133
134         uptr = (void __user *)user_ptr;
135
136         while (n) {
137                 buffer = rb_entry(n, struct binder_buffer, rb_node);
138                 BUG_ON(buffer->free);
139
140                 if (uptr < buffer->user_data)
141                         n = n->rb_left;
142                 else if (uptr > buffer->user_data)
143                         n = n->rb_right;
144                 else {
145                         /*
146                          * Guard against user threads attempting to
147                          * free the buffer when in use by kernel or
148                          * after it's already been freed.
149                          */
150                         if (!buffer->allow_user_free)
151                                 return ERR_PTR(-EPERM);
152                         buffer->allow_user_free = 0;
153                         return buffer;
154                 }
155         }
156         return NULL;
157 }
158
159 /**
160  * binder_alloc_prepare_to_free() - get buffer given user ptr
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  */
170 struct binder_buffer *binder_alloc_prepare_to_free(struct binder_alloc *alloc,
171                                                    uintptr_t user_ptr)
172 {
173         struct binder_buffer *buffer;
174
175         mutex_lock(&alloc->mutex);
176         buffer = binder_alloc_prepare_to_free_locked(alloc, user_ptr);
177         mutex_unlock(&alloc->mutex);
178         return buffer;
179 }
180
181 static int binder_update_page_range(struct binder_alloc *alloc, int allocate,
182                                     void __user *start, void __user *end)
183 {
184         void __user *page_addr;
185         unsigned long user_page_addr;
186         struct binder_lru_page *page;
187         struct vm_area_struct *vma = NULL;
188         struct mm_struct *mm = NULL;
189         bool need_mm = false;
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
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
211         if (need_mm && mmget_not_zero(alloc->vma_vm_mm))
212                 mm = alloc->vma_vm_mm;
213
214         if (mm) {
215                 mmap_read_lock(mm);
216                 vma = alloc->vma;
217         }
218
219         if (!vma && need_mm) {
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);
223                 goto err_no_vma;
224         }
225
226         for (page_addr = start; page_addr < end; page_addr += PAGE_SIZE) {
227                 int ret;
228                 bool on_lru;
229                 size_t index;
230
231                 index = (page_addr - alloc->buffer) / PAGE_SIZE;
232                 page = &alloc->pages[index];
233
234                 if (page->page_ptr) {
235                         trace_binder_alloc_lru_start(alloc, index);
236
237                         on_lru = list_lru_del(&binder_alloc_lru, &page->lru);
238                         WARN_ON(!on_lru);
239
240                         trace_binder_alloc_lru_end(alloc, index);
241                         continue;
242                 }
243
244                 if (WARN_ON(!vma))
245                         goto err_page_ptr_cleared;
246
247                 trace_binder_alloc_page_start(alloc, index);
248                 page->page_ptr = alloc_page(GFP_KERNEL |
249                                             __GFP_HIGHMEM |
250                                             __GFP_ZERO);
251                 if (!page->page_ptr) {
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                 }
256                 page->alloc = alloc;
257                 INIT_LIST_HEAD(&page->lru);
258
259                 user_page_addr = (uintptr_t)page_addr;
260                 ret = vm_insert_page(vma, user_page_addr, page[0].page_ptr);
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                 }
266
267                 if (index + 1 > alloc->pages_high)
268                         alloc->pages_high = index + 1;
269
270                 trace_binder_alloc_page_end(alloc, index);
271         }
272         if (mm) {
273                 mmap_read_unlock(mm);
274                 mmput(mm);
275         }
276         return 0;
277
278 free_range:
279         for (page_addr = end - PAGE_SIZE; 1; page_addr -= PAGE_SIZE) {
280                 bool ret;
281                 size_t index;
282
283                 index = (page_addr - alloc->buffer) / PAGE_SIZE;
284                 page = &alloc->pages[index];
285
286                 trace_binder_free_lru_start(alloc, index);
287
288                 ret = list_lru_add(&binder_alloc_lru, &page->lru);
289                 WARN_ON(!ret);
290
291                 trace_binder_free_lru_end(alloc, index);
292                 if (page_addr == start)
293                         break;
294                 continue;
295
296 err_vm_insert_page_failed:
297                 __free_page(page->page_ptr);
298                 page->page_ptr = NULL;
299 err_alloc_page_failed:
300 err_page_ptr_cleared:
301                 if (page_addr == start)
302                         break;
303         }
304 err_no_vma:
305         if (mm) {
306                 mmap_read_unlock(mm);
307                 mmput(mm);
308         }
309         return vma ? -ENOMEM : -ESRCH;
310 }
311
312
313 static inline void binder_alloc_set_vma(struct binder_alloc *alloc,
314                 struct vm_area_struct *vma)
315 {
316         if (vma)
317                 alloc->vma_vm_mm = vma->vm_mm;
318         /*
319          * If we see alloc->vma is not NULL, buffer data structures set up
320          * completely. Look at smp_rmb side binder_alloc_get_vma.
321          * We also want to guarantee new alloc->vma_vm_mm is always visible
322          * if alloc->vma is set.
323          */
324         smp_wmb();
325         alloc->vma = vma;
326 }
327
328 static inline struct vm_area_struct *binder_alloc_get_vma(
329                 struct binder_alloc *alloc)
330 {
331         struct vm_area_struct *vma = NULL;
332
333         if (alloc->vma) {
334                 /* Look at description in binder_alloc_set_vma */
335                 smp_rmb();
336                 vma = alloc->vma;
337         }
338         return vma;
339 }
340
341 static bool debug_low_async_space_locked(struct binder_alloc *alloc, int pid)
342 {
343         /*
344          * Find the amount and size of buffers allocated by the current caller;
345          * The idea is that once we cross the threshold, whoever is responsible
346          * for the low async space is likely to try to send another async txn,
347          * and at some point we'll catch them in the act. This is more efficient
348          * than keeping a map per pid.
349          */
350         struct rb_node *n;
351         struct binder_buffer *buffer;
352         size_t total_alloc_size = 0;
353         size_t num_buffers = 0;
354
355         for (n = rb_first(&alloc->allocated_buffers); n != NULL;
356                  n = rb_next(n)) {
357                 buffer = rb_entry(n, struct binder_buffer, rb_node);
358                 if (buffer->pid != pid)
359                         continue;
360                 if (!buffer->async_transaction)
361                         continue;
362                 total_alloc_size += binder_alloc_buffer_size(alloc, buffer)
363                         + sizeof(struct binder_buffer);
364                 num_buffers++;
365         }
366
367         /*
368          * Warn if this pid has more than 50 transactions, or more than 50% of
369          * async space (which is 25% of total buffer size). Oneway spam is only
370          * detected when the threshold is exceeded.
371          */
372         if (num_buffers > 50 || total_alloc_size > alloc->buffer_size / 4) {
373                 binder_alloc_debug(BINDER_DEBUG_USER_ERROR,
374                              "%d: pid %d spamming oneway? %zd buffers allocated for a total size of %zd\n",
375                               alloc->pid, pid, num_buffers, total_alloc_size);
376                 if (!alloc->oneway_spam_detected) {
377                         alloc->oneway_spam_detected = true;
378                         return true;
379                 }
380         }
381         return false;
382 }
383
384 static struct binder_buffer *binder_alloc_new_buf_locked(
385                                 struct binder_alloc *alloc,
386                                 size_t data_size,
387                                 size_t offsets_size,
388                                 size_t extra_buffers_size,
389                                 int is_async,
390                                 int pid)
391 {
392         struct rb_node *n = alloc->free_buffers.rb_node;
393         struct binder_buffer *buffer;
394         size_t buffer_size;
395         struct rb_node *best_fit = NULL;
396         void __user *has_page_addr;
397         void __user *end_page_addr;
398         size_t size, data_offsets_size;
399         int ret;
400
401         if (!binder_alloc_get_vma(alloc)) {
402                 binder_alloc_debug(BINDER_DEBUG_USER_ERROR,
403                                    "%d: binder_alloc_buf, no vma\n",
404                                    alloc->pid);
405                 return ERR_PTR(-ESRCH);
406         }
407
408         data_offsets_size = ALIGN(data_size, sizeof(void *)) +
409                 ALIGN(offsets_size, sizeof(void *));
410
411         if (data_offsets_size < data_size || data_offsets_size < offsets_size) {
412                 binder_alloc_debug(BINDER_DEBUG_BUFFER_ALLOC,
413                                 "%d: got transaction with invalid size %zd-%zd\n",
414                                 alloc->pid, data_size, offsets_size);
415                 return ERR_PTR(-EINVAL);
416         }
417         size = data_offsets_size + ALIGN(extra_buffers_size, sizeof(void *));
418         if (size < data_offsets_size || size < extra_buffers_size) {
419                 binder_alloc_debug(BINDER_DEBUG_BUFFER_ALLOC,
420                                 "%d: got transaction with invalid extra_buffers_size %zd\n",
421                                 alloc->pid, extra_buffers_size);
422                 return ERR_PTR(-EINVAL);
423         }
424         if (is_async &&
425             alloc->free_async_space < size + sizeof(struct binder_buffer)) {
426                 binder_alloc_debug(BINDER_DEBUG_BUFFER_ALLOC,
427                              "%d: binder_alloc_buf size %zd failed, no async space left\n",
428                               alloc->pid, size);
429                 return ERR_PTR(-ENOSPC);
430         }
431
432         /* Pad 0-size buffers so they get assigned unique addresses */
433         size = max(size, sizeof(void *));
434
435         while (n) {
436                 buffer = rb_entry(n, struct binder_buffer, rb_node);
437                 BUG_ON(!buffer->free);
438                 buffer_size = binder_alloc_buffer_size(alloc, buffer);
439
440                 if (size < buffer_size) {
441                         best_fit = n;
442                         n = n->rb_left;
443                 } else if (size > buffer_size)
444                         n = n->rb_right;
445                 else {
446                         best_fit = n;
447                         break;
448                 }
449         }
450         if (best_fit == NULL) {
451                 size_t allocated_buffers = 0;
452                 size_t largest_alloc_size = 0;
453                 size_t total_alloc_size = 0;
454                 size_t free_buffers = 0;
455                 size_t largest_free_size = 0;
456                 size_t total_free_size = 0;
457
458                 for (n = rb_first(&alloc->allocated_buffers); n != NULL;
459                      n = rb_next(n)) {
460                         buffer = rb_entry(n, struct binder_buffer, rb_node);
461                         buffer_size = binder_alloc_buffer_size(alloc, buffer);
462                         allocated_buffers++;
463                         total_alloc_size += buffer_size;
464                         if (buffer_size > largest_alloc_size)
465                                 largest_alloc_size = buffer_size;
466                 }
467                 for (n = rb_first(&alloc->free_buffers); n != NULL;
468                      n = rb_next(n)) {
469                         buffer = rb_entry(n, struct binder_buffer, rb_node);
470                         buffer_size = binder_alloc_buffer_size(alloc, buffer);
471                         free_buffers++;
472                         total_free_size += buffer_size;
473                         if (buffer_size > largest_free_size)
474                                 largest_free_size = buffer_size;
475                 }
476                 binder_alloc_debug(BINDER_DEBUG_USER_ERROR,
477                                    "%d: binder_alloc_buf size %zd failed, no address space\n",
478                                    alloc->pid, size);
479                 binder_alloc_debug(BINDER_DEBUG_USER_ERROR,
480                                    "allocated: %zd (num: %zd largest: %zd), free: %zd (num: %zd largest: %zd)\n",
481                                    total_alloc_size, allocated_buffers,
482                                    largest_alloc_size, total_free_size,
483                                    free_buffers, largest_free_size);
484                 return ERR_PTR(-ENOSPC);
485         }
486         if (n == NULL) {
487                 buffer = rb_entry(best_fit, struct binder_buffer, rb_node);
488                 buffer_size = binder_alloc_buffer_size(alloc, buffer);
489         }
490
491         binder_alloc_debug(BINDER_DEBUG_BUFFER_ALLOC,
492                      "%d: binder_alloc_buf size %zd got buffer %pK size %zd\n",
493                       alloc->pid, size, buffer, buffer_size);
494
495         has_page_addr = (void __user *)
496                 (((uintptr_t)buffer->user_data + buffer_size) & PAGE_MASK);
497         WARN_ON(n && buffer_size != size);
498         end_page_addr =
499                 (void __user *)PAGE_ALIGN((uintptr_t)buffer->user_data + size);
500         if (end_page_addr > has_page_addr)
501                 end_page_addr = has_page_addr;
502         ret = binder_update_page_range(alloc, 1, (void __user *)
503                 PAGE_ALIGN((uintptr_t)buffer->user_data), end_page_addr);
504         if (ret)
505                 return ERR_PTR(ret);
506
507         if (buffer_size != size) {
508                 struct binder_buffer *new_buffer;
509
510                 new_buffer = kzalloc(sizeof(*buffer), GFP_KERNEL);
511                 if (!new_buffer) {
512                         pr_err("%s: %d failed to alloc new buffer struct\n",
513                                __func__, alloc->pid);
514                         goto err_alloc_buf_struct_failed;
515                 }
516                 new_buffer->user_data = (u8 __user *)buffer->user_data + size;
517                 list_add(&new_buffer->entry, &buffer->entry);
518                 new_buffer->free = 1;
519                 binder_insert_free_buffer(alloc, new_buffer);
520         }
521
522         rb_erase(best_fit, &alloc->free_buffers);
523         buffer->free = 0;
524         buffer->allow_user_free = 0;
525         binder_insert_allocated_buffer_locked(alloc, buffer);
526         binder_alloc_debug(BINDER_DEBUG_BUFFER_ALLOC,
527                      "%d: binder_alloc_buf size %zd got %pK\n",
528                       alloc->pid, size, buffer);
529         buffer->data_size = data_size;
530         buffer->offsets_size = offsets_size;
531         buffer->async_transaction = is_async;
532         buffer->extra_buffers_size = extra_buffers_size;
533         buffer->pid = pid;
534         buffer->oneway_spam_suspect = false;
535         if (is_async) {
536                 alloc->free_async_space -= size + sizeof(struct binder_buffer);
537                 binder_alloc_debug(BINDER_DEBUG_BUFFER_ALLOC_ASYNC,
538                              "%d: binder_alloc_buf size %zd async free %zd\n",
539                               alloc->pid, size, alloc->free_async_space);
540                 if (alloc->free_async_space < alloc->buffer_size / 10) {
541                         /*
542                          * Start detecting spammers once we have less than 20%
543                          * of async space left (which is less than 10% of total
544                          * buffer size).
545                          */
546                         buffer->oneway_spam_suspect = debug_low_async_space_locked(alloc, pid);
547                 } else {
548                         alloc->oneway_spam_detected = false;
549                 }
550         }
551         return buffer;
552
553 err_alloc_buf_struct_failed:
554         binder_update_page_range(alloc, 0, (void __user *)
555                                  PAGE_ALIGN((uintptr_t)buffer->user_data),
556                                  end_page_addr);
557         return ERR_PTR(-ENOMEM);
558 }
559
560 /**
561  * binder_alloc_new_buf() - Allocate a new binder buffer
562  * @alloc:              binder_alloc for this proc
563  * @data_size:          size of user data buffer
564  * @offsets_size:       user specified buffer offset
565  * @extra_buffers_size: size of extra space for meta-data (eg, security context)
566  * @is_async:           buffer for async transaction
567  * @pid:                                pid to attribute allocation to (used for debugging)
568  *
569  * Allocate a new buffer given the requested sizes. Returns
570  * the kernel version of the buffer pointer. The size allocated
571  * is the sum of the three given sizes (each rounded up to
572  * pointer-sized boundary)
573  *
574  * Return:      The allocated buffer or %NULL if error
575  */
576 struct binder_buffer *binder_alloc_new_buf(struct binder_alloc *alloc,
577                                            size_t data_size,
578                                            size_t offsets_size,
579                                            size_t extra_buffers_size,
580                                            int is_async,
581                                            int pid)
582 {
583         struct binder_buffer *buffer;
584
585         mutex_lock(&alloc->mutex);
586         buffer = binder_alloc_new_buf_locked(alloc, data_size, offsets_size,
587                                              extra_buffers_size, is_async, pid);
588         mutex_unlock(&alloc->mutex);
589         return buffer;
590 }
591
592 static void __user *buffer_start_page(struct binder_buffer *buffer)
593 {
594         return (void __user *)((uintptr_t)buffer->user_data & PAGE_MASK);
595 }
596
597 static void __user *prev_buffer_end_page(struct binder_buffer *buffer)
598 {
599         return (void __user *)
600                 (((uintptr_t)(buffer->user_data) - 1) & PAGE_MASK);
601 }
602
603 static void binder_delete_free_buffer(struct binder_alloc *alloc,
604                                       struct binder_buffer *buffer)
605 {
606         struct binder_buffer *prev, *next = NULL;
607         bool to_free = true;
608
609         BUG_ON(alloc->buffers.next == &buffer->entry);
610         prev = binder_buffer_prev(buffer);
611         BUG_ON(!prev->free);
612         if (prev_buffer_end_page(prev) == buffer_start_page(buffer)) {
613                 to_free = false;
614                 binder_alloc_debug(BINDER_DEBUG_BUFFER_ALLOC,
615                                    "%d: merge free, buffer %pK share page with %pK\n",
616                                    alloc->pid, buffer->user_data,
617                                    prev->user_data);
618         }
619
620         if (!list_is_last(&buffer->entry, &alloc->buffers)) {
621                 next = binder_buffer_next(buffer);
622                 if (buffer_start_page(next) == buffer_start_page(buffer)) {
623                         to_free = false;
624                         binder_alloc_debug(BINDER_DEBUG_BUFFER_ALLOC,
625                                            "%d: merge free, buffer %pK share page with %pK\n",
626                                            alloc->pid,
627                                            buffer->user_data,
628                                            next->user_data);
629                 }
630         }
631
632         if (PAGE_ALIGNED(buffer->user_data)) {
633                 binder_alloc_debug(BINDER_DEBUG_BUFFER_ALLOC,
634                                    "%d: merge free, buffer start %pK is page aligned\n",
635                                    alloc->pid, buffer->user_data);
636                 to_free = false;
637         }
638
639         if (to_free) {
640                 binder_alloc_debug(BINDER_DEBUG_BUFFER_ALLOC,
641                                    "%d: merge free, buffer %pK do not share page with %pK or %pK\n",
642                                    alloc->pid, buffer->user_data,
643                                    prev->user_data,
644                                    next ? next->user_data : NULL);
645                 binder_update_page_range(alloc, 0, buffer_start_page(buffer),
646                                          buffer_start_page(buffer) + PAGE_SIZE);
647         }
648         list_del(&buffer->entry);
649         kfree(buffer);
650 }
651
652 static void binder_free_buf_locked(struct binder_alloc *alloc,
653                                    struct binder_buffer *buffer)
654 {
655         size_t size, buffer_size;
656
657         buffer_size = binder_alloc_buffer_size(alloc, buffer);
658
659         size = ALIGN(buffer->data_size, sizeof(void *)) +
660                 ALIGN(buffer->offsets_size, sizeof(void *)) +
661                 ALIGN(buffer->extra_buffers_size, sizeof(void *));
662
663         binder_alloc_debug(BINDER_DEBUG_BUFFER_ALLOC,
664                      "%d: binder_free_buf %pK size %zd buffer_size %zd\n",
665                       alloc->pid, buffer, size, buffer_size);
666
667         BUG_ON(buffer->free);
668         BUG_ON(size > buffer_size);
669         BUG_ON(buffer->transaction != NULL);
670         BUG_ON(buffer->user_data < alloc->buffer);
671         BUG_ON(buffer->user_data > alloc->buffer + alloc->buffer_size);
672
673         if (buffer->async_transaction) {
674                 alloc->free_async_space += size + sizeof(struct binder_buffer);
675
676                 binder_alloc_debug(BINDER_DEBUG_BUFFER_ALLOC_ASYNC,
677                              "%d: binder_free_buf size %zd async free %zd\n",
678                               alloc->pid, size, alloc->free_async_space);
679         }
680
681         binder_update_page_range(alloc, 0,
682                 (void __user *)PAGE_ALIGN((uintptr_t)buffer->user_data),
683                 (void __user *)(((uintptr_t)
684                           buffer->user_data + buffer_size) & PAGE_MASK));
685
686         rb_erase(&buffer->rb_node, &alloc->allocated_buffers);
687         buffer->free = 1;
688         if (!list_is_last(&buffer->entry, &alloc->buffers)) {
689                 struct binder_buffer *next = binder_buffer_next(buffer);
690
691                 if (next->free) {
692                         rb_erase(&next->rb_node, &alloc->free_buffers);
693                         binder_delete_free_buffer(alloc, next);
694                 }
695         }
696         if (alloc->buffers.next != &buffer->entry) {
697                 struct binder_buffer *prev = binder_buffer_prev(buffer);
698
699                 if (prev->free) {
700                         binder_delete_free_buffer(alloc, buffer);
701                         rb_erase(&prev->rb_node, &alloc->free_buffers);
702                         buffer = prev;
703                 }
704         }
705         binder_insert_free_buffer(alloc, buffer);
706 }
707
708 static void binder_alloc_clear_buf(struct binder_alloc *alloc,
709                                    struct binder_buffer *buffer);
710 /**
711  * binder_alloc_free_buf() - free a binder buffer
712  * @alloc:      binder_alloc for this proc
713  * @buffer:     kernel pointer to buffer
714  *
715  * Free the buffer allocated via binder_alloc_new_buf()
716  */
717 void binder_alloc_free_buf(struct binder_alloc *alloc,
718                             struct binder_buffer *buffer)
719 {
720         /*
721          * We could eliminate the call to binder_alloc_clear_buf()
722          * from binder_alloc_deferred_release() by moving this to
723          * binder_alloc_free_buf_locked(). However, that could
724          * increase contention for the alloc mutex if clear_on_free
725          * is used frequently for large buffers. The mutex is not
726          * needed for correctness here.
727          */
728         if (buffer->clear_on_free) {
729                 binder_alloc_clear_buf(alloc, buffer);
730                 buffer->clear_on_free = false;
731         }
732         mutex_lock(&alloc->mutex);
733         binder_free_buf_locked(alloc, buffer);
734         mutex_unlock(&alloc->mutex);
735 }
736
737 /**
738  * binder_alloc_mmap_handler() - map virtual address space for proc
739  * @alloc:      alloc structure for this proc
740  * @vma:        vma passed to mmap()
741  *
742  * Called by binder_mmap() to initialize the space specified in
743  * vma for allocating binder buffers
744  *
745  * Return:
746  *      0 = success
747  *      -EBUSY = address space already mapped
748  *      -ENOMEM = failed to map memory to given address space
749  */
750 int binder_alloc_mmap_handler(struct binder_alloc *alloc,
751                               struct vm_area_struct *vma)
752 {
753         int ret;
754         const char *failure_string;
755         struct binder_buffer *buffer;
756
757         mutex_lock(&binder_alloc_mmap_lock);
758         if (alloc->buffer_size) {
759                 ret = -EBUSY;
760                 failure_string = "already mapped";
761                 goto err_already_mapped;
762         }
763         alloc->buffer_size = min_t(unsigned long, vma->vm_end - vma->vm_start,
764                                    SZ_4M);
765         mutex_unlock(&binder_alloc_mmap_lock);
766
767         alloc->buffer = (void __user *)vma->vm_start;
768
769         alloc->pages = kcalloc(alloc->buffer_size / PAGE_SIZE,
770                                sizeof(alloc->pages[0]),
771                                GFP_KERNEL);
772         if (alloc->pages == NULL) {
773                 ret = -ENOMEM;
774                 failure_string = "alloc page array";
775                 goto err_alloc_pages_failed;
776         }
777
778         buffer = kzalloc(sizeof(*buffer), GFP_KERNEL);
779         if (!buffer) {
780                 ret = -ENOMEM;
781                 failure_string = "alloc buffer struct";
782                 goto err_alloc_buf_struct_failed;
783         }
784
785         buffer->user_data = alloc->buffer;
786         list_add(&buffer->entry, &alloc->buffers);
787         buffer->free = 1;
788         binder_insert_free_buffer(alloc, buffer);
789         alloc->free_async_space = alloc->buffer_size / 2;
790         binder_alloc_set_vma(alloc, vma);
791         mmgrab(alloc->vma_vm_mm);
792
793         return 0;
794
795 err_alloc_buf_struct_failed:
796         kfree(alloc->pages);
797         alloc->pages = NULL;
798 err_alloc_pages_failed:
799         alloc->buffer = NULL;
800         mutex_lock(&binder_alloc_mmap_lock);
801         alloc->buffer_size = 0;
802 err_already_mapped:
803         mutex_unlock(&binder_alloc_mmap_lock);
804         binder_alloc_debug(BINDER_DEBUG_USER_ERROR,
805                            "%s: %d %lx-%lx %s failed %d\n", __func__,
806                            alloc->pid, vma->vm_start, vma->vm_end,
807                            failure_string, ret);
808         return ret;
809 }
810
811
812 void binder_alloc_deferred_release(struct binder_alloc *alloc)
813 {
814         struct rb_node *n;
815         int buffers, page_count;
816         struct binder_buffer *buffer;
817
818         buffers = 0;
819         mutex_lock(&alloc->mutex);
820         BUG_ON(alloc->vma);
821
822         while ((n = rb_first(&alloc->allocated_buffers))) {
823                 buffer = rb_entry(n, struct binder_buffer, rb_node);
824
825                 /* Transaction should already have been freed */
826                 BUG_ON(buffer->transaction);
827
828                 if (buffer->clear_on_free) {
829                         binder_alloc_clear_buf(alloc, buffer);
830                         buffer->clear_on_free = false;
831                 }
832                 binder_free_buf_locked(alloc, buffer);
833                 buffers++;
834         }
835
836         while (!list_empty(&alloc->buffers)) {
837                 buffer = list_first_entry(&alloc->buffers,
838                                           struct binder_buffer, entry);
839                 WARN_ON(!buffer->free);
840
841                 list_del(&buffer->entry);
842                 WARN_ON_ONCE(!list_empty(&alloc->buffers));
843                 kfree(buffer);
844         }
845
846         page_count = 0;
847         if (alloc->pages) {
848                 int i;
849
850                 for (i = 0; i < alloc->buffer_size / PAGE_SIZE; i++) {
851                         void __user *page_addr;
852                         bool on_lru;
853
854                         if (!alloc->pages[i].page_ptr)
855                                 continue;
856
857                         on_lru = list_lru_del(&binder_alloc_lru,
858                                               &alloc->pages[i].lru);
859                         page_addr = alloc->buffer + i * PAGE_SIZE;
860                         binder_alloc_debug(BINDER_DEBUG_BUFFER_ALLOC,
861                                      "%s: %d: page %d at %pK %s\n",
862                                      __func__, alloc->pid, i, page_addr,
863                                      on_lru ? "on lru" : "active");
864                         __free_page(alloc->pages[i].page_ptr);
865                         page_count++;
866                 }
867                 kfree(alloc->pages);
868         }
869         mutex_unlock(&alloc->mutex);
870         if (alloc->vma_vm_mm)
871                 mmdrop(alloc->vma_vm_mm);
872
873         binder_alloc_debug(BINDER_DEBUG_OPEN_CLOSE,
874                      "%s: %d buffers %d, pages %d\n",
875                      __func__, alloc->pid, buffers, page_count);
876 }
877
878 static void print_binder_buffer(struct seq_file *m, const char *prefix,
879                                 struct binder_buffer *buffer)
880 {
881         seq_printf(m, "%s %d: %pK size %zd:%zd:%zd %s\n",
882                    prefix, buffer->debug_id, buffer->user_data,
883                    buffer->data_size, buffer->offsets_size,
884                    buffer->extra_buffers_size,
885                    buffer->transaction ? "active" : "delivered");
886 }
887
888 /**
889  * binder_alloc_print_allocated() - print buffer info
890  * @m:     seq_file for output via seq_printf()
891  * @alloc: binder_alloc for this proc
892  *
893  * Prints information about every buffer associated with
894  * the binder_alloc state to the given seq_file
895  */
896 void binder_alloc_print_allocated(struct seq_file *m,
897                                   struct binder_alloc *alloc)
898 {
899         struct rb_node *n;
900
901         mutex_lock(&alloc->mutex);
902         for (n = rb_first(&alloc->allocated_buffers); n != NULL; n = rb_next(n))
903                 print_binder_buffer(m, "  buffer",
904                                     rb_entry(n, struct binder_buffer, rb_node));
905         mutex_unlock(&alloc->mutex);
906 }
907
908 /**
909  * binder_alloc_print_pages() - print page usage
910  * @m:     seq_file for output via seq_printf()
911  * @alloc: binder_alloc for this proc
912  */
913 void binder_alloc_print_pages(struct seq_file *m,
914                               struct binder_alloc *alloc)
915 {
916         struct binder_lru_page *page;
917         int i;
918         int active = 0;
919         int lru = 0;
920         int free = 0;
921
922         mutex_lock(&alloc->mutex);
923         /*
924          * Make sure the binder_alloc is fully initialized, otherwise we might
925          * read inconsistent state.
926          */
927         if (binder_alloc_get_vma(alloc) != NULL) {
928                 for (i = 0; i < alloc->buffer_size / PAGE_SIZE; i++) {
929                         page = &alloc->pages[i];
930                         if (!page->page_ptr)
931                                 free++;
932                         else if (list_empty(&page->lru))
933                                 active++;
934                         else
935                                 lru++;
936                 }
937         }
938         mutex_unlock(&alloc->mutex);
939         seq_printf(m, "  pages: %d:%d:%d\n", active, lru, free);
940         seq_printf(m, "  pages high watermark: %zu\n", alloc->pages_high);
941 }
942
943 /**
944  * binder_alloc_get_allocated_count() - return count of buffers
945  * @alloc: binder_alloc for this proc
946  *
947  * Return: count of allocated buffers
948  */
949 int binder_alloc_get_allocated_count(struct binder_alloc *alloc)
950 {
951         struct rb_node *n;
952         int count = 0;
953
954         mutex_lock(&alloc->mutex);
955         for (n = rb_first(&alloc->allocated_buffers); n != NULL; n = rb_next(n))
956                 count++;
957         mutex_unlock(&alloc->mutex);
958         return count;
959 }
960
961
962 /**
963  * binder_alloc_vma_close() - invalidate address space
964  * @alloc: binder_alloc for this proc
965  *
966  * Called from binder_vma_close() when releasing address space.
967  * Clears alloc->vma to prevent new incoming transactions from
968  * allocating more buffers.
969  */
970 void binder_alloc_vma_close(struct binder_alloc *alloc)
971 {
972         binder_alloc_set_vma(alloc, NULL);
973 }
974
975 /**
976  * binder_alloc_free_page() - shrinker callback to free pages
977  * @item:   item to free
978  * @lock:   lock protecting the item
979  * @cb_arg: callback argument
980  *
981  * Called from list_lru_walk() in binder_shrink_scan() to free
982  * up pages when the system is under memory pressure.
983  */
984 enum lru_status binder_alloc_free_page(struct list_head *item,
985                                        struct list_lru_one *lru,
986                                        spinlock_t *lock,
987                                        void *cb_arg)
988         __must_hold(lock)
989 {
990         struct mm_struct *mm = NULL;
991         struct binder_lru_page *page = container_of(item,
992                                                     struct binder_lru_page,
993                                                     lru);
994         struct binder_alloc *alloc;
995         uintptr_t page_addr;
996         size_t index;
997         struct vm_area_struct *vma;
998
999         alloc = page->alloc;
1000         if (!mutex_trylock(&alloc->mutex))
1001                 goto err_get_alloc_mutex_failed;
1002
1003         if (!page->page_ptr)
1004                 goto err_page_already_freed;
1005
1006         index = page - alloc->pages;
1007         page_addr = (uintptr_t)alloc->buffer + index * PAGE_SIZE;
1008
1009         mm = alloc->vma_vm_mm;
1010         if (!mmget_not_zero(mm))
1011                 goto err_mmget;
1012         if (!mmap_read_trylock(mm))
1013                 goto err_mmap_read_lock_failed;
1014         vma = binder_alloc_get_vma(alloc);
1015
1016         list_lru_isolate(lru, item);
1017         spin_unlock(lock);
1018
1019         if (vma) {
1020                 trace_binder_unmap_user_start(alloc, index);
1021
1022                 zap_page_range(vma, page_addr, PAGE_SIZE);
1023
1024                 trace_binder_unmap_user_end(alloc, index);
1025         }
1026         mmap_read_unlock(mm);
1027         mmput_async(mm);
1028
1029         trace_binder_unmap_kernel_start(alloc, index);
1030
1031         __free_page(page->page_ptr);
1032         page->page_ptr = NULL;
1033
1034         trace_binder_unmap_kernel_end(alloc, index);
1035
1036         spin_lock(lock);
1037         mutex_unlock(&alloc->mutex);
1038         return LRU_REMOVED_RETRY;
1039
1040 err_mmap_read_lock_failed:
1041         mmput_async(mm);
1042 err_mmget:
1043 err_page_already_freed:
1044         mutex_unlock(&alloc->mutex);
1045 err_get_alloc_mutex_failed:
1046         return LRU_SKIP;
1047 }
1048
1049 static unsigned long
1050 binder_shrink_count(struct shrinker *shrink, struct shrink_control *sc)
1051 {
1052         unsigned long ret = list_lru_count(&binder_alloc_lru);
1053         return ret;
1054 }
1055
1056 static unsigned long
1057 binder_shrink_scan(struct shrinker *shrink, struct shrink_control *sc)
1058 {
1059         unsigned long ret;
1060
1061         ret = list_lru_walk(&binder_alloc_lru, binder_alloc_free_page,
1062                             NULL, sc->nr_to_scan);
1063         return ret;
1064 }
1065
1066 static struct shrinker binder_shrinker = {
1067         .count_objects = binder_shrink_count,
1068         .scan_objects = binder_shrink_scan,
1069         .seeks = DEFAULT_SEEKS,
1070 };
1071
1072 /**
1073  * binder_alloc_init() - called by binder_open() for per-proc initialization
1074  * @alloc: binder_alloc for this proc
1075  *
1076  * Called from binder_open() to initialize binder_alloc fields for
1077  * new binder proc
1078  */
1079 void binder_alloc_init(struct binder_alloc *alloc)
1080 {
1081         alloc->pid = current->group_leader->pid;
1082         mutex_init(&alloc->mutex);
1083         INIT_LIST_HEAD(&alloc->buffers);
1084 }
1085
1086 int binder_alloc_shrinker_init(void)
1087 {
1088         int ret = list_lru_init(&binder_alloc_lru);
1089
1090         if (ret == 0) {
1091                 ret = register_shrinker(&binder_shrinker);
1092                 if (ret)
1093                         list_lru_destroy(&binder_alloc_lru);
1094         }
1095         return ret;
1096 }
1097
1098 /**
1099  * check_buffer() - verify that buffer/offset is safe to access
1100  * @alloc: binder_alloc for this proc
1101  * @buffer: binder buffer to be accessed
1102  * @offset: offset into @buffer data
1103  * @bytes: bytes to access from offset
1104  *
1105  * Check that the @offset/@bytes are within the size of the given
1106  * @buffer and that the buffer is currently active and not freeable.
1107  * Offsets must also be multiples of sizeof(u32). The kernel is
1108  * allowed to touch the buffer in two cases:
1109  *
1110  * 1) when the buffer is being created:
1111  *     (buffer->free == 0 && buffer->allow_user_free == 0)
1112  * 2) when the buffer is being torn down:
1113  *     (buffer->free == 0 && buffer->transaction == NULL).
1114  *
1115  * Return: true if the buffer is safe to access
1116  */
1117 static inline bool check_buffer(struct binder_alloc *alloc,
1118                                 struct binder_buffer *buffer,
1119                                 binder_size_t offset, size_t bytes)
1120 {
1121         size_t buffer_size = binder_alloc_buffer_size(alloc, buffer);
1122
1123         return buffer_size >= bytes &&
1124                 offset <= buffer_size - bytes &&
1125                 IS_ALIGNED(offset, sizeof(u32)) &&
1126                 !buffer->free &&
1127                 (!buffer->allow_user_free || !buffer->transaction);
1128 }
1129
1130 /**
1131  * binder_alloc_get_page() - get kernel pointer for given buffer offset
1132  * @alloc: binder_alloc for this proc
1133  * @buffer: binder buffer to be accessed
1134  * @buffer_offset: offset into @buffer data
1135  * @pgoffp: address to copy final page offset to
1136  *
1137  * Lookup the struct page corresponding to the address
1138  * at @buffer_offset into @buffer->user_data. If @pgoffp is not
1139  * NULL, the byte-offset into the page is written there.
1140  *
1141  * The caller is responsible to ensure that the offset points
1142  * to a valid address within the @buffer and that @buffer is
1143  * not freeable by the user. Since it can't be freed, we are
1144  * guaranteed that the corresponding elements of @alloc->pages[]
1145  * cannot change.
1146  *
1147  * Return: struct page
1148  */
1149 static struct page *binder_alloc_get_page(struct binder_alloc *alloc,
1150                                           struct binder_buffer *buffer,
1151                                           binder_size_t buffer_offset,
1152                                           pgoff_t *pgoffp)
1153 {
1154         binder_size_t buffer_space_offset = buffer_offset +
1155                 (buffer->user_data - alloc->buffer);
1156         pgoff_t pgoff = buffer_space_offset & ~PAGE_MASK;
1157         size_t index = buffer_space_offset >> PAGE_SHIFT;
1158         struct binder_lru_page *lru_page;
1159
1160         lru_page = &alloc->pages[index];
1161         *pgoffp = pgoff;
1162         return lru_page->page_ptr;
1163 }
1164
1165 /**
1166  * binder_alloc_clear_buf() - zero out buffer
1167  * @alloc: binder_alloc for this proc
1168  * @buffer: binder buffer to be cleared
1169  *
1170  * memset the given buffer to 0
1171  */
1172 static void binder_alloc_clear_buf(struct binder_alloc *alloc,
1173                                    struct binder_buffer *buffer)
1174 {
1175         size_t bytes = binder_alloc_buffer_size(alloc, buffer);
1176         binder_size_t buffer_offset = 0;
1177
1178         while (bytes) {
1179                 unsigned long size;
1180                 struct page *page;
1181                 pgoff_t pgoff;
1182                 void *kptr;
1183
1184                 page = binder_alloc_get_page(alloc, buffer,
1185                                              buffer_offset, &pgoff);
1186                 size = min_t(size_t, bytes, PAGE_SIZE - pgoff);
1187                 kptr = kmap(page) + pgoff;
1188                 memset(kptr, 0, size);
1189                 kunmap(page);
1190                 bytes -= size;
1191                 buffer_offset += size;
1192         }
1193 }
1194
1195 /**
1196  * binder_alloc_copy_user_to_buffer() - copy src user to tgt user
1197  * @alloc: binder_alloc for this proc
1198  * @buffer: binder buffer to be accessed
1199  * @buffer_offset: offset into @buffer data
1200  * @from: userspace pointer to source buffer
1201  * @bytes: bytes to copy
1202  *
1203  * Copy bytes from source userspace to target buffer.
1204  *
1205  * Return: bytes remaining to be copied
1206  */
1207 unsigned long
1208 binder_alloc_copy_user_to_buffer(struct binder_alloc *alloc,
1209                                  struct binder_buffer *buffer,
1210                                  binder_size_t buffer_offset,
1211                                  const void __user *from,
1212                                  size_t bytes)
1213 {
1214         if (!check_buffer(alloc, buffer, buffer_offset, bytes))
1215                 return bytes;
1216
1217         while (bytes) {
1218                 unsigned long size;
1219                 unsigned long ret;
1220                 struct page *page;
1221                 pgoff_t pgoff;
1222                 void *kptr;
1223
1224                 page = binder_alloc_get_page(alloc, buffer,
1225                                              buffer_offset, &pgoff);
1226                 size = min_t(size_t, bytes, PAGE_SIZE - pgoff);
1227                 kptr = kmap(page) + pgoff;
1228                 ret = copy_from_user(kptr, from, size);
1229                 kunmap(page);
1230                 if (ret)
1231                         return bytes - size + ret;
1232                 bytes -= size;
1233                 from += size;
1234                 buffer_offset += size;
1235         }
1236         return 0;
1237 }
1238
1239 static int binder_alloc_do_buffer_copy(struct binder_alloc *alloc,
1240                                        bool to_buffer,
1241                                        struct binder_buffer *buffer,
1242                                        binder_size_t buffer_offset,
1243                                        void *ptr,
1244                                        size_t bytes)
1245 {
1246         /* All copies must be 32-bit aligned and 32-bit size */
1247         if (!check_buffer(alloc, buffer, buffer_offset, bytes))
1248                 return -EINVAL;
1249
1250         while (bytes) {
1251                 unsigned long size;
1252                 struct page *page;
1253                 pgoff_t pgoff;
1254                 void *tmpptr;
1255                 void *base_ptr;
1256
1257                 page = binder_alloc_get_page(alloc, buffer,
1258                                              buffer_offset, &pgoff);
1259                 size = min_t(size_t, bytes, PAGE_SIZE - pgoff);
1260                 base_ptr = kmap_atomic(page);
1261                 tmpptr = base_ptr + pgoff;
1262                 if (to_buffer)
1263                         memcpy(tmpptr, ptr, size);
1264                 else
1265                         memcpy(ptr, tmpptr, size);
1266                 /*
1267                  * kunmap_atomic() takes care of flushing the cache
1268                  * if this device has VIVT cache arch
1269                  */
1270                 kunmap_atomic(base_ptr);
1271                 bytes -= size;
1272                 pgoff = 0;
1273                 ptr = ptr + size;
1274                 buffer_offset += size;
1275         }
1276         return 0;
1277 }
1278
1279 int binder_alloc_copy_to_buffer(struct binder_alloc *alloc,
1280                                 struct binder_buffer *buffer,
1281                                 binder_size_t buffer_offset,
1282                                 void *src,
1283                                 size_t bytes)
1284 {
1285         return binder_alloc_do_buffer_copy(alloc, true, buffer, buffer_offset,
1286                                            src, bytes);
1287 }
1288
1289 int binder_alloc_copy_from_buffer(struct binder_alloc *alloc,
1290                                   void *dest,
1291                                   struct binder_buffer *buffer,
1292                                   binder_size_t buffer_offset,
1293                                   size_t bytes)
1294 {
1295         return binder_alloc_do_buffer_copy(alloc, false, buffer, buffer_offset,
1296                                            dest, bytes);
1297 }
1298