Merge tag 'for-linus' of git://git.kernel.org/pub/scm/virt/kvm/kvm
[platform/kernel/linux-starfive.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->mm))
212                 mm = alloc->mm;
213
214         if (mm) {
215                 mmap_read_lock(mm);
216                 vma = vma_lookup(mm, alloc->vma_addr);
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 static inline struct vm_area_struct *binder_alloc_get_vma(
313                 struct binder_alloc *alloc)
314 {
315         struct vm_area_struct *vma = NULL;
316
317         if (alloc->vma_addr)
318                 vma = vma_lookup(alloc->mm, alloc->vma_addr);
319
320         return vma;
321 }
322
323 static bool debug_low_async_space_locked(struct binder_alloc *alloc, int pid)
324 {
325         /*
326          * Find the amount and size of buffers allocated by the current caller;
327          * The idea is that once we cross the threshold, whoever is responsible
328          * for the low async space is likely to try to send another async txn,
329          * and at some point we'll catch them in the act. This is more efficient
330          * than keeping a map per pid.
331          */
332         struct rb_node *n;
333         struct binder_buffer *buffer;
334         size_t total_alloc_size = 0;
335         size_t num_buffers = 0;
336
337         for (n = rb_first(&alloc->allocated_buffers); n != NULL;
338                  n = rb_next(n)) {
339                 buffer = rb_entry(n, struct binder_buffer, rb_node);
340                 if (buffer->pid != pid)
341                         continue;
342                 if (!buffer->async_transaction)
343                         continue;
344                 total_alloc_size += binder_alloc_buffer_size(alloc, buffer)
345                         + sizeof(struct binder_buffer);
346                 num_buffers++;
347         }
348
349         /*
350          * Warn if this pid has more than 50 transactions, or more than 50% of
351          * async space (which is 25% of total buffer size). Oneway spam is only
352          * detected when the threshold is exceeded.
353          */
354         if (num_buffers > 50 || total_alloc_size > alloc->buffer_size / 4) {
355                 binder_alloc_debug(BINDER_DEBUG_USER_ERROR,
356                              "%d: pid %d spamming oneway? %zd buffers allocated for a total size of %zd\n",
357                               alloc->pid, pid, num_buffers, total_alloc_size);
358                 if (!alloc->oneway_spam_detected) {
359                         alloc->oneway_spam_detected = true;
360                         return true;
361                 }
362         }
363         return false;
364 }
365
366 static struct binder_buffer *binder_alloc_new_buf_locked(
367                                 struct binder_alloc *alloc,
368                                 size_t data_size,
369                                 size_t offsets_size,
370                                 size_t extra_buffers_size,
371                                 int is_async,
372                                 int pid)
373 {
374         struct rb_node *n = alloc->free_buffers.rb_node;
375         struct binder_buffer *buffer;
376         size_t buffer_size;
377         struct rb_node *best_fit = NULL;
378         void __user *has_page_addr;
379         void __user *end_page_addr;
380         size_t size, data_offsets_size;
381         int ret;
382
383         mmap_read_lock(alloc->mm);
384         if (!binder_alloc_get_vma(alloc)) {
385                 mmap_read_unlock(alloc->mm);
386                 binder_alloc_debug(BINDER_DEBUG_USER_ERROR,
387                                    "%d: binder_alloc_buf, no vma\n",
388                                    alloc->pid);
389                 return ERR_PTR(-ESRCH);
390         }
391         mmap_read_unlock(alloc->mm);
392
393         data_offsets_size = ALIGN(data_size, sizeof(void *)) +
394                 ALIGN(offsets_size, sizeof(void *));
395
396         if (data_offsets_size < data_size || data_offsets_size < offsets_size) {
397                 binder_alloc_debug(BINDER_DEBUG_BUFFER_ALLOC,
398                                 "%d: got transaction with invalid size %zd-%zd\n",
399                                 alloc->pid, data_size, offsets_size);
400                 return ERR_PTR(-EINVAL);
401         }
402         size = data_offsets_size + ALIGN(extra_buffers_size, sizeof(void *));
403         if (size < data_offsets_size || size < extra_buffers_size) {
404                 binder_alloc_debug(BINDER_DEBUG_BUFFER_ALLOC,
405                                 "%d: got transaction with invalid extra_buffers_size %zd\n",
406                                 alloc->pid, extra_buffers_size);
407                 return ERR_PTR(-EINVAL);
408         }
409         if (is_async &&
410             alloc->free_async_space < size + sizeof(struct binder_buffer)) {
411                 binder_alloc_debug(BINDER_DEBUG_BUFFER_ALLOC,
412                              "%d: binder_alloc_buf size %zd failed, no async space left\n",
413                               alloc->pid, size);
414                 return ERR_PTR(-ENOSPC);
415         }
416
417         /* Pad 0-size buffers so they get assigned unique addresses */
418         size = max(size, sizeof(void *));
419
420         while (n) {
421                 buffer = rb_entry(n, struct binder_buffer, rb_node);
422                 BUG_ON(!buffer->free);
423                 buffer_size = binder_alloc_buffer_size(alloc, buffer);
424
425                 if (size < buffer_size) {
426                         best_fit = n;
427                         n = n->rb_left;
428                 } else if (size > buffer_size)
429                         n = n->rb_right;
430                 else {
431                         best_fit = n;
432                         break;
433                 }
434         }
435         if (best_fit == NULL) {
436                 size_t allocated_buffers = 0;
437                 size_t largest_alloc_size = 0;
438                 size_t total_alloc_size = 0;
439                 size_t free_buffers = 0;
440                 size_t largest_free_size = 0;
441                 size_t total_free_size = 0;
442
443                 for (n = rb_first(&alloc->allocated_buffers); n != NULL;
444                      n = rb_next(n)) {
445                         buffer = rb_entry(n, struct binder_buffer, rb_node);
446                         buffer_size = binder_alloc_buffer_size(alloc, buffer);
447                         allocated_buffers++;
448                         total_alloc_size += buffer_size;
449                         if (buffer_size > largest_alloc_size)
450                                 largest_alloc_size = buffer_size;
451                 }
452                 for (n = rb_first(&alloc->free_buffers); n != NULL;
453                      n = rb_next(n)) {
454                         buffer = rb_entry(n, struct binder_buffer, rb_node);
455                         buffer_size = binder_alloc_buffer_size(alloc, buffer);
456                         free_buffers++;
457                         total_free_size += buffer_size;
458                         if (buffer_size > largest_free_size)
459                                 largest_free_size = buffer_size;
460                 }
461                 binder_alloc_debug(BINDER_DEBUG_USER_ERROR,
462                                    "%d: binder_alloc_buf size %zd failed, no address space\n",
463                                    alloc->pid, size);
464                 binder_alloc_debug(BINDER_DEBUG_USER_ERROR,
465                                    "allocated: %zd (num: %zd largest: %zd), free: %zd (num: %zd largest: %zd)\n",
466                                    total_alloc_size, allocated_buffers,
467                                    largest_alloc_size, total_free_size,
468                                    free_buffers, largest_free_size);
469                 return ERR_PTR(-ENOSPC);
470         }
471         if (n == NULL) {
472                 buffer = rb_entry(best_fit, struct binder_buffer, rb_node);
473                 buffer_size = binder_alloc_buffer_size(alloc, buffer);
474         }
475
476         binder_alloc_debug(BINDER_DEBUG_BUFFER_ALLOC,
477                      "%d: binder_alloc_buf size %zd got buffer %pK size %zd\n",
478                       alloc->pid, size, buffer, buffer_size);
479
480         has_page_addr = (void __user *)
481                 (((uintptr_t)buffer->user_data + buffer_size) & PAGE_MASK);
482         WARN_ON(n && buffer_size != size);
483         end_page_addr =
484                 (void __user *)PAGE_ALIGN((uintptr_t)buffer->user_data + size);
485         if (end_page_addr > has_page_addr)
486                 end_page_addr = has_page_addr;
487         ret = binder_update_page_range(alloc, 1, (void __user *)
488                 PAGE_ALIGN((uintptr_t)buffer->user_data), end_page_addr);
489         if (ret)
490                 return ERR_PTR(ret);
491
492         if (buffer_size != size) {
493                 struct binder_buffer *new_buffer;
494
495                 new_buffer = kzalloc(sizeof(*buffer), GFP_KERNEL);
496                 if (!new_buffer) {
497                         pr_err("%s: %d failed to alloc new buffer struct\n",
498                                __func__, alloc->pid);
499                         goto err_alloc_buf_struct_failed;
500                 }
501                 new_buffer->user_data = (u8 __user *)buffer->user_data + size;
502                 list_add(&new_buffer->entry, &buffer->entry);
503                 new_buffer->free = 1;
504                 binder_insert_free_buffer(alloc, new_buffer);
505         }
506
507         rb_erase(best_fit, &alloc->free_buffers);
508         buffer->free = 0;
509         buffer->allow_user_free = 0;
510         binder_insert_allocated_buffer_locked(alloc, buffer);
511         binder_alloc_debug(BINDER_DEBUG_BUFFER_ALLOC,
512                      "%d: binder_alloc_buf size %zd got %pK\n",
513                       alloc->pid, size, buffer);
514         buffer->data_size = data_size;
515         buffer->offsets_size = offsets_size;
516         buffer->async_transaction = is_async;
517         buffer->extra_buffers_size = extra_buffers_size;
518         buffer->pid = pid;
519         buffer->oneway_spam_suspect = false;
520         if (is_async) {
521                 alloc->free_async_space -= size + sizeof(struct binder_buffer);
522                 binder_alloc_debug(BINDER_DEBUG_BUFFER_ALLOC_ASYNC,
523                              "%d: binder_alloc_buf size %zd async free %zd\n",
524                               alloc->pid, size, alloc->free_async_space);
525                 if (alloc->free_async_space < alloc->buffer_size / 10) {
526                         /*
527                          * Start detecting spammers once we have less than 20%
528                          * of async space left (which is less than 10% of total
529                          * buffer size).
530                          */
531                         buffer->oneway_spam_suspect = debug_low_async_space_locked(alloc, pid);
532                 } else {
533                         alloc->oneway_spam_detected = false;
534                 }
535         }
536         return buffer;
537
538 err_alloc_buf_struct_failed:
539         binder_update_page_range(alloc, 0, (void __user *)
540                                  PAGE_ALIGN((uintptr_t)buffer->user_data),
541                                  end_page_addr);
542         return ERR_PTR(-ENOMEM);
543 }
544
545 /**
546  * binder_alloc_new_buf() - Allocate a new binder buffer
547  * @alloc:              binder_alloc for this proc
548  * @data_size:          size of user data buffer
549  * @offsets_size:       user specified buffer offset
550  * @extra_buffers_size: size of extra space for meta-data (eg, security context)
551  * @is_async:           buffer for async transaction
552  * @pid:                                pid to attribute allocation to (used for debugging)
553  *
554  * Allocate a new buffer given the requested sizes. Returns
555  * the kernel version of the buffer pointer. The size allocated
556  * is the sum of the three given sizes (each rounded up to
557  * pointer-sized boundary)
558  *
559  * Return:      The allocated buffer or %NULL if error
560  */
561 struct binder_buffer *binder_alloc_new_buf(struct binder_alloc *alloc,
562                                            size_t data_size,
563                                            size_t offsets_size,
564                                            size_t extra_buffers_size,
565                                            int is_async,
566                                            int pid)
567 {
568         struct binder_buffer *buffer;
569
570         mutex_lock(&alloc->mutex);
571         buffer = binder_alloc_new_buf_locked(alloc, data_size, offsets_size,
572                                              extra_buffers_size, is_async, pid);
573         mutex_unlock(&alloc->mutex);
574         return buffer;
575 }
576
577 static void __user *buffer_start_page(struct binder_buffer *buffer)
578 {
579         return (void __user *)((uintptr_t)buffer->user_data & PAGE_MASK);
580 }
581
582 static void __user *prev_buffer_end_page(struct binder_buffer *buffer)
583 {
584         return (void __user *)
585                 (((uintptr_t)(buffer->user_data) - 1) & PAGE_MASK);
586 }
587
588 static void binder_delete_free_buffer(struct binder_alloc *alloc,
589                                       struct binder_buffer *buffer)
590 {
591         struct binder_buffer *prev, *next = NULL;
592         bool to_free = true;
593
594         BUG_ON(alloc->buffers.next == &buffer->entry);
595         prev = binder_buffer_prev(buffer);
596         BUG_ON(!prev->free);
597         if (prev_buffer_end_page(prev) == buffer_start_page(buffer)) {
598                 to_free = false;
599                 binder_alloc_debug(BINDER_DEBUG_BUFFER_ALLOC,
600                                    "%d: merge free, buffer %pK share page with %pK\n",
601                                    alloc->pid, buffer->user_data,
602                                    prev->user_data);
603         }
604
605         if (!list_is_last(&buffer->entry, &alloc->buffers)) {
606                 next = binder_buffer_next(buffer);
607                 if (buffer_start_page(next) == buffer_start_page(buffer)) {
608                         to_free = false;
609                         binder_alloc_debug(BINDER_DEBUG_BUFFER_ALLOC,
610                                            "%d: merge free, buffer %pK share page with %pK\n",
611                                            alloc->pid,
612                                            buffer->user_data,
613                                            next->user_data);
614                 }
615         }
616
617         if (PAGE_ALIGNED(buffer->user_data)) {
618                 binder_alloc_debug(BINDER_DEBUG_BUFFER_ALLOC,
619                                    "%d: merge free, buffer start %pK is page aligned\n",
620                                    alloc->pid, buffer->user_data);
621                 to_free = false;
622         }
623
624         if (to_free) {
625                 binder_alloc_debug(BINDER_DEBUG_BUFFER_ALLOC,
626                                    "%d: merge free, buffer %pK do not share page with %pK or %pK\n",
627                                    alloc->pid, buffer->user_data,
628                                    prev->user_data,
629                                    next ? next->user_data : NULL);
630                 binder_update_page_range(alloc, 0, buffer_start_page(buffer),
631                                          buffer_start_page(buffer) + PAGE_SIZE);
632         }
633         list_del(&buffer->entry);
634         kfree(buffer);
635 }
636
637 static void binder_free_buf_locked(struct binder_alloc *alloc,
638                                    struct binder_buffer *buffer)
639 {
640         size_t size, buffer_size;
641
642         buffer_size = binder_alloc_buffer_size(alloc, buffer);
643
644         size = ALIGN(buffer->data_size, sizeof(void *)) +
645                 ALIGN(buffer->offsets_size, sizeof(void *)) +
646                 ALIGN(buffer->extra_buffers_size, sizeof(void *));
647
648         binder_alloc_debug(BINDER_DEBUG_BUFFER_ALLOC,
649                      "%d: binder_free_buf %pK size %zd buffer_size %zd\n",
650                       alloc->pid, buffer, size, buffer_size);
651
652         BUG_ON(buffer->free);
653         BUG_ON(size > buffer_size);
654         BUG_ON(buffer->transaction != NULL);
655         BUG_ON(buffer->user_data < alloc->buffer);
656         BUG_ON(buffer->user_data > alloc->buffer + alloc->buffer_size);
657
658         if (buffer->async_transaction) {
659                 alloc->free_async_space += buffer_size + sizeof(struct binder_buffer);
660
661                 binder_alloc_debug(BINDER_DEBUG_BUFFER_ALLOC_ASYNC,
662                              "%d: binder_free_buf size %zd async free %zd\n",
663                               alloc->pid, size, alloc->free_async_space);
664         }
665
666         binder_update_page_range(alloc, 0,
667                 (void __user *)PAGE_ALIGN((uintptr_t)buffer->user_data),
668                 (void __user *)(((uintptr_t)
669                           buffer->user_data + buffer_size) & PAGE_MASK));
670
671         rb_erase(&buffer->rb_node, &alloc->allocated_buffers);
672         buffer->free = 1;
673         if (!list_is_last(&buffer->entry, &alloc->buffers)) {
674                 struct binder_buffer *next = binder_buffer_next(buffer);
675
676                 if (next->free) {
677                         rb_erase(&next->rb_node, &alloc->free_buffers);
678                         binder_delete_free_buffer(alloc, next);
679                 }
680         }
681         if (alloc->buffers.next != &buffer->entry) {
682                 struct binder_buffer *prev = binder_buffer_prev(buffer);
683
684                 if (prev->free) {
685                         binder_delete_free_buffer(alloc, buffer);
686                         rb_erase(&prev->rb_node, &alloc->free_buffers);
687                         buffer = prev;
688                 }
689         }
690         binder_insert_free_buffer(alloc, buffer);
691 }
692
693 static void binder_alloc_clear_buf(struct binder_alloc *alloc,
694                                    struct binder_buffer *buffer);
695 /**
696  * binder_alloc_free_buf() - free a binder buffer
697  * @alloc:      binder_alloc for this proc
698  * @buffer:     kernel pointer to buffer
699  *
700  * Free the buffer allocated via binder_alloc_new_buf()
701  */
702 void binder_alloc_free_buf(struct binder_alloc *alloc,
703                             struct binder_buffer *buffer)
704 {
705         /*
706          * We could eliminate the call to binder_alloc_clear_buf()
707          * from binder_alloc_deferred_release() by moving this to
708          * binder_alloc_free_buf_locked(). However, that could
709          * increase contention for the alloc mutex if clear_on_free
710          * is used frequently for large buffers. The mutex is not
711          * needed for correctness here.
712          */
713         if (buffer->clear_on_free) {
714                 binder_alloc_clear_buf(alloc, buffer);
715                 buffer->clear_on_free = false;
716         }
717         mutex_lock(&alloc->mutex);
718         binder_free_buf_locked(alloc, buffer);
719         mutex_unlock(&alloc->mutex);
720 }
721
722 /**
723  * binder_alloc_mmap_handler() - map virtual address space for proc
724  * @alloc:      alloc structure for this proc
725  * @vma:        vma passed to mmap()
726  *
727  * Called by binder_mmap() to initialize the space specified in
728  * vma for allocating binder buffers
729  *
730  * Return:
731  *      0 = success
732  *      -EBUSY = address space already mapped
733  *      -ENOMEM = failed to map memory to given address space
734  */
735 int binder_alloc_mmap_handler(struct binder_alloc *alloc,
736                               struct vm_area_struct *vma)
737 {
738         int ret;
739         const char *failure_string;
740         struct binder_buffer *buffer;
741
742         if (unlikely(vma->vm_mm != alloc->mm)) {
743                 ret = -EINVAL;
744                 failure_string = "invalid vma->vm_mm";
745                 goto err_invalid_mm;
746         }
747
748         mutex_lock(&binder_alloc_mmap_lock);
749         if (alloc->buffer_size) {
750                 ret = -EBUSY;
751                 failure_string = "already mapped";
752                 goto err_already_mapped;
753         }
754         alloc->buffer_size = min_t(unsigned long, vma->vm_end - vma->vm_start,
755                                    SZ_4M);
756         mutex_unlock(&binder_alloc_mmap_lock);
757
758         alloc->buffer = (void __user *)vma->vm_start;
759
760         alloc->pages = kcalloc(alloc->buffer_size / PAGE_SIZE,
761                                sizeof(alloc->pages[0]),
762                                GFP_KERNEL);
763         if (alloc->pages == NULL) {
764                 ret = -ENOMEM;
765                 failure_string = "alloc page array";
766                 goto err_alloc_pages_failed;
767         }
768
769         buffer = kzalloc(sizeof(*buffer), GFP_KERNEL);
770         if (!buffer) {
771                 ret = -ENOMEM;
772                 failure_string = "alloc buffer struct";
773                 goto err_alloc_buf_struct_failed;
774         }
775
776         buffer->user_data = alloc->buffer;
777         list_add(&buffer->entry, &alloc->buffers);
778         buffer->free = 1;
779         binder_insert_free_buffer(alloc, buffer);
780         alloc->free_async_space = alloc->buffer_size / 2;
781         alloc->vma_addr = vma->vm_start;
782
783         return 0;
784
785 err_alloc_buf_struct_failed:
786         kfree(alloc->pages);
787         alloc->pages = NULL;
788 err_alloc_pages_failed:
789         alloc->buffer = NULL;
790         mutex_lock(&binder_alloc_mmap_lock);
791         alloc->buffer_size = 0;
792 err_already_mapped:
793         mutex_unlock(&binder_alloc_mmap_lock);
794 err_invalid_mm:
795         binder_alloc_debug(BINDER_DEBUG_USER_ERROR,
796                            "%s: %d %lx-%lx %s failed %d\n", __func__,
797                            alloc->pid, vma->vm_start, vma->vm_end,
798                            failure_string, ret);
799         return ret;
800 }
801
802
803 void binder_alloc_deferred_release(struct binder_alloc *alloc)
804 {
805         struct rb_node *n;
806         int buffers, page_count;
807         struct binder_buffer *buffer;
808
809         buffers = 0;
810         mutex_lock(&alloc->mutex);
811         BUG_ON(alloc->vma_addr &&
812                vma_lookup(alloc->mm, alloc->vma_addr));
813
814         while ((n = rb_first(&alloc->allocated_buffers))) {
815                 buffer = rb_entry(n, struct binder_buffer, rb_node);
816
817                 /* Transaction should already have been freed */
818                 BUG_ON(buffer->transaction);
819
820                 if (buffer->clear_on_free) {
821                         binder_alloc_clear_buf(alloc, buffer);
822                         buffer->clear_on_free = false;
823                 }
824                 binder_free_buf_locked(alloc, buffer);
825                 buffers++;
826         }
827
828         while (!list_empty(&alloc->buffers)) {
829                 buffer = list_first_entry(&alloc->buffers,
830                                           struct binder_buffer, entry);
831                 WARN_ON(!buffer->free);
832
833                 list_del(&buffer->entry);
834                 WARN_ON_ONCE(!list_empty(&alloc->buffers));
835                 kfree(buffer);
836         }
837
838         page_count = 0;
839         if (alloc->pages) {
840                 int i;
841
842                 for (i = 0; i < alloc->buffer_size / PAGE_SIZE; i++) {
843                         void __user *page_addr;
844                         bool on_lru;
845
846                         if (!alloc->pages[i].page_ptr)
847                                 continue;
848
849                         on_lru = list_lru_del(&binder_alloc_lru,
850                                               &alloc->pages[i].lru);
851                         page_addr = alloc->buffer + i * PAGE_SIZE;
852                         binder_alloc_debug(BINDER_DEBUG_BUFFER_ALLOC,
853                                      "%s: %d: page %d at %pK %s\n",
854                                      __func__, alloc->pid, i, page_addr,
855                                      on_lru ? "on lru" : "active");
856                         __free_page(alloc->pages[i].page_ptr);
857                         page_count++;
858                 }
859                 kfree(alloc->pages);
860         }
861         mutex_unlock(&alloc->mutex);
862         if (alloc->mm)
863                 mmdrop(alloc->mm);
864
865         binder_alloc_debug(BINDER_DEBUG_OPEN_CLOSE,
866                      "%s: %d buffers %d, pages %d\n",
867                      __func__, alloc->pid, buffers, page_count);
868 }
869
870 static void print_binder_buffer(struct seq_file *m, const char *prefix,
871                                 struct binder_buffer *buffer)
872 {
873         seq_printf(m, "%s %d: %pK size %zd:%zd:%zd %s\n",
874                    prefix, buffer->debug_id, buffer->user_data,
875                    buffer->data_size, buffer->offsets_size,
876                    buffer->extra_buffers_size,
877                    buffer->transaction ? "active" : "delivered");
878 }
879
880 /**
881  * binder_alloc_print_allocated() - print buffer info
882  * @m:     seq_file for output via seq_printf()
883  * @alloc: binder_alloc for this proc
884  *
885  * Prints information about every buffer associated with
886  * the binder_alloc state to the given seq_file
887  */
888 void binder_alloc_print_allocated(struct seq_file *m,
889                                   struct binder_alloc *alloc)
890 {
891         struct rb_node *n;
892
893         mutex_lock(&alloc->mutex);
894         for (n = rb_first(&alloc->allocated_buffers); n != NULL; n = rb_next(n))
895                 print_binder_buffer(m, "  buffer",
896                                     rb_entry(n, struct binder_buffer, rb_node));
897         mutex_unlock(&alloc->mutex);
898 }
899
900 /**
901  * binder_alloc_print_pages() - print page usage
902  * @m:     seq_file for output via seq_printf()
903  * @alloc: binder_alloc for this proc
904  */
905 void binder_alloc_print_pages(struct seq_file *m,
906                               struct binder_alloc *alloc)
907 {
908         struct binder_lru_page *page;
909         int i;
910         int active = 0;
911         int lru = 0;
912         int free = 0;
913
914         mutex_lock(&alloc->mutex);
915         /*
916          * Make sure the binder_alloc is fully initialized, otherwise we might
917          * read inconsistent state.
918          */
919
920         mmap_read_lock(alloc->mm);
921         if (binder_alloc_get_vma(alloc) == NULL) {
922                 mmap_read_unlock(alloc->mm);
923                 goto uninitialized;
924         }
925
926         mmap_read_unlock(alloc->mm);
927         for (i = 0; i < alloc->buffer_size / PAGE_SIZE; i++) {
928                 page = &alloc->pages[i];
929                 if (!page->page_ptr)
930                         free++;
931                 else if (list_empty(&page->lru))
932                         active++;
933                 else
934                         lru++;
935         }
936
937 uninitialized:
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         alloc->vma_addr = 0;
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->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         return list_lru_count(&binder_alloc_lru);
1053 }
1054
1055 static unsigned long
1056 binder_shrink_scan(struct shrinker *shrink, struct shrink_control *sc)
1057 {
1058         return list_lru_walk(&binder_alloc_lru, binder_alloc_free_page,
1059                             NULL, sc->nr_to_scan);
1060 }
1061
1062 static struct shrinker binder_shrinker = {
1063         .count_objects = binder_shrink_count,
1064         .scan_objects = binder_shrink_scan,
1065         .seeks = DEFAULT_SEEKS,
1066 };
1067
1068 /**
1069  * binder_alloc_init() - called by binder_open() for per-proc initialization
1070  * @alloc: binder_alloc for this proc
1071  *
1072  * Called from binder_open() to initialize binder_alloc fields for
1073  * new binder proc
1074  */
1075 void binder_alloc_init(struct binder_alloc *alloc)
1076 {
1077         alloc->pid = current->group_leader->pid;
1078         alloc->mm = current->mm;
1079         mmgrab(alloc->mm);
1080         mutex_init(&alloc->mutex);
1081         INIT_LIST_HEAD(&alloc->buffers);
1082 }
1083
1084 int binder_alloc_shrinker_init(void)
1085 {
1086         int ret = list_lru_init(&binder_alloc_lru);
1087
1088         if (ret == 0) {
1089                 ret = register_shrinker(&binder_shrinker, "android-binder");
1090                 if (ret)
1091                         list_lru_destroy(&binder_alloc_lru);
1092         }
1093         return ret;
1094 }
1095
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
1136  * at @buffer_offset into @buffer->user_data. If @pgoffp is not
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 +
1153                 (buffer->user_data - alloc->buffer);
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
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;
1180
1181                 page = binder_alloc_get_page(alloc, buffer,
1182                                              buffer_offset, &pgoff);
1183                 size = min_t(size_t, bytes, PAGE_SIZE - pgoff);
1184                 memset_page(page, pgoff, 0, size);
1185                 bytes -= size;
1186                 buffer_offset += size;
1187         }
1188 }
1189
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);
1222                 kptr = kmap_local_page(page) + pgoff;
1223                 ret = copy_from_user(kptr, from, size);
1224                 kunmap_local(kptr);
1225                 if (ret)
1226                         return bytes - size + ret;
1227                 bytes -= size;
1228                 from += size;
1229                 buffer_offset += size;
1230         }
1231         return 0;
1232 }
1233
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)
1240 {
1241         /* All copies must be 32-bit aligned and 32-bit size */
1242         if (!check_buffer(alloc, buffer, buffer_offset, bytes))
1243                 return -EINVAL;
1244
1245         while (bytes) {
1246                 unsigned long size;
1247                 struct page *page;
1248                 pgoff_t pgoff;
1249
1250                 page = binder_alloc_get_page(alloc, buffer,
1251                                              buffer_offset, &pgoff);
1252                 size = min_t(size_t, bytes, PAGE_SIZE - pgoff);
1253                 if (to_buffer)
1254                         memcpy_to_page(page, pgoff, ptr, size);
1255                 else
1256                         memcpy_from_page(ptr, page, pgoff, size);
1257                 bytes -= size;
1258                 pgoff = 0;
1259                 ptr = ptr + size;
1260                 buffer_offset += size;
1261         }
1262         return 0;
1263 }
1264
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)
1270 {
1271         return binder_alloc_do_buffer_copy(alloc, true, buffer, buffer_offset,
1272                                            src, bytes);
1273 }
1274
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)
1280 {
1281         return binder_alloc_do_buffer_copy(alloc, false, buffer, buffer_offset,
1282                                            dest, bytes);
1283 }
1284