Merge tag 'sched-urgent-2024-03-24' of git://git.kernel.org/pub/scm/linux/kernel...
[sfrench/cifs-2.6.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_freelist;
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                 unsigned long user_ptr)
129 {
130         struct rb_node *n = alloc->allocated_buffers.rb_node;
131         struct binder_buffer *buffer;
132
133         while (n) {
134                 buffer = rb_entry(n, struct binder_buffer, rb_node);
135                 BUG_ON(buffer->free);
136
137                 if (user_ptr < buffer->user_data) {
138                         n = n->rb_left;
139                 } else if (user_ptr > buffer->user_data) {
140                         n = n->rb_right;
141                 } else {
142                         /*
143                          * Guard against user threads attempting to
144                          * free the buffer when in use by kernel or
145                          * after it's already been freed.
146                          */
147                         if (!buffer->allow_user_free)
148                                 return ERR_PTR(-EPERM);
149                         buffer->allow_user_free = 0;
150                         return buffer;
151                 }
152         }
153         return NULL;
154 }
155
156 /**
157  * binder_alloc_prepare_to_free() - get buffer given user ptr
158  * @alloc:      binder_alloc for this proc
159  * @user_ptr:   User pointer to buffer data
160  *
161  * Validate userspace pointer to buffer data and return buffer corresponding to
162  * that user pointer. Search the rb tree for buffer that matches user data
163  * pointer.
164  *
165  * Return:      Pointer to buffer or NULL
166  */
167 struct binder_buffer *binder_alloc_prepare_to_free(struct binder_alloc *alloc,
168                                                    unsigned long user_ptr)
169 {
170         struct binder_buffer *buffer;
171
172         spin_lock(&alloc->lock);
173         buffer = binder_alloc_prepare_to_free_locked(alloc, user_ptr);
174         spin_unlock(&alloc->lock);
175         return buffer;
176 }
177
178 static inline void
179 binder_set_installed_page(struct binder_lru_page *lru_page,
180                           struct page *page)
181 {
182         /* Pairs with acquire in binder_get_installed_page() */
183         smp_store_release(&lru_page->page_ptr, page);
184 }
185
186 static inline struct page *
187 binder_get_installed_page(struct binder_lru_page *lru_page)
188 {
189         /* Pairs with release in binder_set_installed_page() */
190         return smp_load_acquire(&lru_page->page_ptr);
191 }
192
193 static void binder_lru_freelist_add(struct binder_alloc *alloc,
194                                     unsigned long start, unsigned long end)
195 {
196         struct binder_lru_page *page;
197         unsigned long page_addr;
198
199         trace_binder_update_page_range(alloc, false, start, end);
200
201         for (page_addr = start; page_addr < end; page_addr += PAGE_SIZE) {
202                 size_t index;
203                 int ret;
204
205                 index = (page_addr - alloc->buffer) / PAGE_SIZE;
206                 page = &alloc->pages[index];
207
208                 if (!binder_get_installed_page(page))
209                         continue;
210
211                 trace_binder_free_lru_start(alloc, index);
212
213                 ret = list_lru_add_obj(&binder_freelist, &page->lru);
214                 WARN_ON(!ret);
215
216                 trace_binder_free_lru_end(alloc, index);
217         }
218 }
219
220 static int binder_install_single_page(struct binder_alloc *alloc,
221                                       struct binder_lru_page *lru_page,
222                                       unsigned long addr)
223 {
224         struct page *page;
225         int ret = 0;
226
227         if (!mmget_not_zero(alloc->mm))
228                 return -ESRCH;
229
230         /*
231          * Protected with mmap_sem in write mode as multiple tasks
232          * might race to install the same page.
233          */
234         mmap_write_lock(alloc->mm);
235         if (binder_get_installed_page(lru_page))
236                 goto out;
237
238         if (!alloc->vma) {
239                 pr_err("%d: %s failed, no vma\n", alloc->pid, __func__);
240                 ret = -ESRCH;
241                 goto out;
242         }
243
244         page = alloc_page(GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO);
245         if (!page) {
246                 pr_err("%d: failed to allocate page\n", alloc->pid);
247                 ret = -ENOMEM;
248                 goto out;
249         }
250
251         ret = vm_insert_page(alloc->vma, addr, page);
252         if (ret) {
253                 pr_err("%d: %s failed to insert page at offset %lx with %d\n",
254                        alloc->pid, __func__, addr - alloc->buffer, ret);
255                 __free_page(page);
256                 ret = -ENOMEM;
257                 goto out;
258         }
259
260         /* Mark page installation complete and safe to use */
261         binder_set_installed_page(lru_page, page);
262 out:
263         mmap_write_unlock(alloc->mm);
264         mmput_async(alloc->mm);
265         return ret;
266 }
267
268 static int binder_install_buffer_pages(struct binder_alloc *alloc,
269                                        struct binder_buffer *buffer,
270                                        size_t size)
271 {
272         struct binder_lru_page *page;
273         unsigned long start, final;
274         unsigned long page_addr;
275
276         start = buffer->user_data & PAGE_MASK;
277         final = PAGE_ALIGN(buffer->user_data + size);
278
279         for (page_addr = start; page_addr < final; page_addr += PAGE_SIZE) {
280                 unsigned long index;
281                 int ret;
282
283                 index = (page_addr - alloc->buffer) / PAGE_SIZE;
284                 page = &alloc->pages[index];
285
286                 if (binder_get_installed_page(page))
287                         continue;
288
289                 trace_binder_alloc_page_start(alloc, index);
290
291                 ret = binder_install_single_page(alloc, page, page_addr);
292                 if (ret)
293                         return ret;
294
295                 trace_binder_alloc_page_end(alloc, index);
296         }
297
298         return 0;
299 }
300
301 /* The range of pages should exclude those shared with other buffers */
302 static void binder_lru_freelist_del(struct binder_alloc *alloc,
303                                     unsigned long start, unsigned long end)
304 {
305         struct binder_lru_page *page;
306         unsigned long page_addr;
307
308         trace_binder_update_page_range(alloc, true, start, end);
309
310         for (page_addr = start; page_addr < end; page_addr += PAGE_SIZE) {
311                 unsigned long index;
312                 bool on_lru;
313
314                 index = (page_addr - alloc->buffer) / PAGE_SIZE;
315                 page = &alloc->pages[index];
316
317                 if (page->page_ptr) {
318                         trace_binder_alloc_lru_start(alloc, index);
319
320                         on_lru = list_lru_del_obj(&binder_freelist, &page->lru);
321                         WARN_ON(!on_lru);
322
323                         trace_binder_alloc_lru_end(alloc, index);
324                         continue;
325                 }
326
327                 if (index + 1 > alloc->pages_high)
328                         alloc->pages_high = index + 1;
329         }
330 }
331
332 static inline void binder_alloc_set_vma(struct binder_alloc *alloc,
333                 struct vm_area_struct *vma)
334 {
335         /* pairs with smp_load_acquire in binder_alloc_get_vma() */
336         smp_store_release(&alloc->vma, vma);
337 }
338
339 static inline struct vm_area_struct *binder_alloc_get_vma(
340                 struct binder_alloc *alloc)
341 {
342         /* pairs with smp_store_release in binder_alloc_set_vma() */
343         return smp_load_acquire(&alloc->vma);
344 }
345
346 static void debug_no_space_locked(struct binder_alloc *alloc)
347 {
348         size_t largest_alloc_size = 0;
349         struct binder_buffer *buffer;
350         size_t allocated_buffers = 0;
351         size_t largest_free_size = 0;
352         size_t total_alloc_size = 0;
353         size_t total_free_size = 0;
354         size_t free_buffers = 0;
355         size_t buffer_size;
356         struct rb_node *n;
357
358         for (n = rb_first(&alloc->allocated_buffers); n; n = rb_next(n)) {
359                 buffer = rb_entry(n, struct binder_buffer, rb_node);
360                 buffer_size = binder_alloc_buffer_size(alloc, buffer);
361                 allocated_buffers++;
362                 total_alloc_size += buffer_size;
363                 if (buffer_size > largest_alloc_size)
364                         largest_alloc_size = buffer_size;
365         }
366
367         for (n = rb_first(&alloc->free_buffers); n; n = rb_next(n)) {
368                 buffer = rb_entry(n, struct binder_buffer, rb_node);
369                 buffer_size = binder_alloc_buffer_size(alloc, buffer);
370                 free_buffers++;
371                 total_free_size += buffer_size;
372                 if (buffer_size > largest_free_size)
373                         largest_free_size = buffer_size;
374         }
375
376         binder_alloc_debug(BINDER_DEBUG_USER_ERROR,
377                            "allocated: %zd (num: %zd largest: %zd), free: %zd (num: %zd largest: %zd)\n",
378                            total_alloc_size, allocated_buffers,
379                            largest_alloc_size, total_free_size,
380                            free_buffers, largest_free_size);
381 }
382
383 static bool debug_low_async_space_locked(struct binder_alloc *alloc)
384 {
385         /*
386          * Find the amount and size of buffers allocated by the current caller;
387          * The idea is that once we cross the threshold, whoever is responsible
388          * for the low async space is likely to try to send another async txn,
389          * and at some point we'll catch them in the act. This is more efficient
390          * than keeping a map per pid.
391          */
392         struct binder_buffer *buffer;
393         size_t total_alloc_size = 0;
394         int pid = current->tgid;
395         size_t num_buffers = 0;
396         struct rb_node *n;
397
398         /*
399          * Only start detecting spammers once we have less than 20% of async
400          * space left (which is less than 10% of total buffer size).
401          */
402         if (alloc->free_async_space >= alloc->buffer_size / 10) {
403                 alloc->oneway_spam_detected = false;
404                 return false;
405         }
406
407         for (n = rb_first(&alloc->allocated_buffers); n != NULL;
408                  n = rb_next(n)) {
409                 buffer = rb_entry(n, struct binder_buffer, rb_node);
410                 if (buffer->pid != pid)
411                         continue;
412                 if (!buffer->async_transaction)
413                         continue;
414                 total_alloc_size += binder_alloc_buffer_size(alloc, buffer);
415                 num_buffers++;
416         }
417
418         /*
419          * Warn if this pid has more than 50 transactions, or more than 50% of
420          * async space (which is 25% of total buffer size). Oneway spam is only
421          * detected when the threshold is exceeded.
422          */
423         if (num_buffers > 50 || total_alloc_size > alloc->buffer_size / 4) {
424                 binder_alloc_debug(BINDER_DEBUG_USER_ERROR,
425                              "%d: pid %d spamming oneway? %zd buffers allocated for a total size of %zd\n",
426                               alloc->pid, pid, num_buffers, total_alloc_size);
427                 if (!alloc->oneway_spam_detected) {
428                         alloc->oneway_spam_detected = true;
429                         return true;
430                 }
431         }
432         return false;
433 }
434
435 /* Callers preallocate @new_buffer, it is freed by this function if unused */
436 static struct binder_buffer *binder_alloc_new_buf_locked(
437                                 struct binder_alloc *alloc,
438                                 struct binder_buffer *new_buffer,
439                                 size_t size,
440                                 int is_async)
441 {
442         struct rb_node *n = alloc->free_buffers.rb_node;
443         struct rb_node *best_fit = NULL;
444         struct binder_buffer *buffer;
445         unsigned long next_used_page;
446         unsigned long curr_last_page;
447         size_t buffer_size;
448
449         if (is_async && alloc->free_async_space < size) {
450                 binder_alloc_debug(BINDER_DEBUG_BUFFER_ALLOC,
451                              "%d: binder_alloc_buf size %zd failed, no async space left\n",
452                               alloc->pid, size);
453                 buffer = ERR_PTR(-ENOSPC);
454                 goto out;
455         }
456
457         while (n) {
458                 buffer = rb_entry(n, struct binder_buffer, rb_node);
459                 BUG_ON(!buffer->free);
460                 buffer_size = binder_alloc_buffer_size(alloc, buffer);
461
462                 if (size < buffer_size) {
463                         best_fit = n;
464                         n = n->rb_left;
465                 } else if (size > buffer_size) {
466                         n = n->rb_right;
467                 } else {
468                         best_fit = n;
469                         break;
470                 }
471         }
472
473         if (unlikely(!best_fit)) {
474                 binder_alloc_debug(BINDER_DEBUG_USER_ERROR,
475                                    "%d: binder_alloc_buf size %zd failed, no address space\n",
476                                    alloc->pid, size);
477                 debug_no_space_locked(alloc);
478                 buffer = ERR_PTR(-ENOSPC);
479                 goto out;
480         }
481
482         if (buffer_size != size) {
483                 /* Found an oversized buffer and needs to be split */
484                 buffer = rb_entry(best_fit, struct binder_buffer, rb_node);
485                 buffer_size = binder_alloc_buffer_size(alloc, buffer);
486
487                 WARN_ON(n || buffer_size == size);
488                 new_buffer->user_data = buffer->user_data + size;
489                 list_add(&new_buffer->entry, &buffer->entry);
490                 new_buffer->free = 1;
491                 binder_insert_free_buffer(alloc, new_buffer);
492                 new_buffer = NULL;
493         }
494
495         binder_alloc_debug(BINDER_DEBUG_BUFFER_ALLOC,
496                      "%d: binder_alloc_buf size %zd got buffer %pK size %zd\n",
497                       alloc->pid, size, buffer, buffer_size);
498
499         /*
500          * Now we remove the pages from the freelist. A clever calculation
501          * with buffer_size determines if the last page is shared with an
502          * adjacent in-use buffer. In such case, the page has been already
503          * removed from the freelist so we trim our range short.
504          */
505         next_used_page = (buffer->user_data + buffer_size) & PAGE_MASK;
506         curr_last_page = PAGE_ALIGN(buffer->user_data + size);
507         binder_lru_freelist_del(alloc, PAGE_ALIGN(buffer->user_data),
508                                 min(next_used_page, curr_last_page));
509
510         rb_erase(&buffer->rb_node, &alloc->free_buffers);
511         buffer->free = 0;
512         buffer->allow_user_free = 0;
513         binder_insert_allocated_buffer_locked(alloc, buffer);
514         buffer->async_transaction = is_async;
515         buffer->oneway_spam_suspect = false;
516         if (is_async) {
517                 alloc->free_async_space -= size;
518                 binder_alloc_debug(BINDER_DEBUG_BUFFER_ALLOC_ASYNC,
519                              "%d: binder_alloc_buf size %zd async free %zd\n",
520                               alloc->pid, size, alloc->free_async_space);
521                 if (debug_low_async_space_locked(alloc))
522                         buffer->oneway_spam_suspect = true;
523         }
524
525 out:
526         /* Discard possibly unused new_buffer */
527         kfree(new_buffer);
528         return buffer;
529 }
530
531 /* Calculate the sanitized total size, returns 0 for invalid request */
532 static inline size_t sanitized_size(size_t data_size,
533                                     size_t offsets_size,
534                                     size_t extra_buffers_size)
535 {
536         size_t total, tmp;
537
538         /* Align to pointer size and check for overflows */
539         tmp = ALIGN(data_size, sizeof(void *)) +
540                 ALIGN(offsets_size, sizeof(void *));
541         if (tmp < data_size || tmp < offsets_size)
542                 return 0;
543         total = tmp + ALIGN(extra_buffers_size, sizeof(void *));
544         if (total < tmp || total < extra_buffers_size)
545                 return 0;
546
547         /* Pad 0-sized buffers so they get a unique address */
548         total = max(total, sizeof(void *));
549
550         return total;
551 }
552
553 /**
554  * binder_alloc_new_buf() - Allocate a new binder buffer
555  * @alloc:              binder_alloc for this proc
556  * @data_size:          size of user data buffer
557  * @offsets_size:       user specified buffer offset
558  * @extra_buffers_size: size of extra space for meta-data (eg, security context)
559  * @is_async:           buffer for async transaction
560  *
561  * Allocate a new buffer given the requested sizes. Returns
562  * the kernel version of the buffer pointer. The size allocated
563  * is the sum of the three given sizes (each rounded up to
564  * pointer-sized boundary)
565  *
566  * Return:      The allocated buffer or %ERR_PTR(-errno) if error
567  */
568 struct binder_buffer *binder_alloc_new_buf(struct binder_alloc *alloc,
569                                            size_t data_size,
570                                            size_t offsets_size,
571                                            size_t extra_buffers_size,
572                                            int is_async)
573 {
574         struct binder_buffer *buffer, *next;
575         size_t size;
576         int ret;
577
578         /* Check binder_alloc is fully initialized */
579         if (!binder_alloc_get_vma(alloc)) {
580                 binder_alloc_debug(BINDER_DEBUG_USER_ERROR,
581                                    "%d: binder_alloc_buf, no vma\n",
582                                    alloc->pid);
583                 return ERR_PTR(-ESRCH);
584         }
585
586         size = sanitized_size(data_size, offsets_size, extra_buffers_size);
587         if (unlikely(!size)) {
588                 binder_alloc_debug(BINDER_DEBUG_BUFFER_ALLOC,
589                                    "%d: got transaction with invalid size %zd-%zd-%zd\n",
590                                    alloc->pid, data_size, offsets_size,
591                                    extra_buffers_size);
592                 return ERR_PTR(-EINVAL);
593         }
594
595         /* Preallocate the next buffer */
596         next = kzalloc(sizeof(*next), GFP_KERNEL);
597         if (!next)
598                 return ERR_PTR(-ENOMEM);
599
600         spin_lock(&alloc->lock);
601         buffer = binder_alloc_new_buf_locked(alloc, next, size, is_async);
602         if (IS_ERR(buffer)) {
603                 spin_unlock(&alloc->lock);
604                 goto out;
605         }
606
607         buffer->data_size = data_size;
608         buffer->offsets_size = offsets_size;
609         buffer->extra_buffers_size = extra_buffers_size;
610         buffer->pid = current->tgid;
611         spin_unlock(&alloc->lock);
612
613         ret = binder_install_buffer_pages(alloc, buffer, size);
614         if (ret) {
615                 binder_alloc_free_buf(alloc, buffer);
616                 buffer = ERR_PTR(ret);
617         }
618 out:
619         return buffer;
620 }
621
622 static unsigned long buffer_start_page(struct binder_buffer *buffer)
623 {
624         return buffer->user_data & PAGE_MASK;
625 }
626
627 static unsigned long prev_buffer_end_page(struct binder_buffer *buffer)
628 {
629         return (buffer->user_data - 1) & PAGE_MASK;
630 }
631
632 static void binder_delete_free_buffer(struct binder_alloc *alloc,
633                                       struct binder_buffer *buffer)
634 {
635         struct binder_buffer *prev, *next;
636
637         if (PAGE_ALIGNED(buffer->user_data))
638                 goto skip_freelist;
639
640         BUG_ON(alloc->buffers.next == &buffer->entry);
641         prev = binder_buffer_prev(buffer);
642         BUG_ON(!prev->free);
643         if (prev_buffer_end_page(prev) == buffer_start_page(buffer))
644                 goto skip_freelist;
645
646         if (!list_is_last(&buffer->entry, &alloc->buffers)) {
647                 next = binder_buffer_next(buffer);
648                 if (buffer_start_page(next) == buffer_start_page(buffer))
649                         goto skip_freelist;
650         }
651
652         binder_lru_freelist_add(alloc, buffer_start_page(buffer),
653                                 buffer_start_page(buffer) + PAGE_SIZE);
654 skip_freelist:
655         list_del(&buffer->entry);
656         kfree(buffer);
657 }
658
659 static void binder_free_buf_locked(struct binder_alloc *alloc,
660                                    struct binder_buffer *buffer)
661 {
662         size_t size, buffer_size;
663
664         buffer_size = binder_alloc_buffer_size(alloc, buffer);
665
666         size = ALIGN(buffer->data_size, sizeof(void *)) +
667                 ALIGN(buffer->offsets_size, sizeof(void *)) +
668                 ALIGN(buffer->extra_buffers_size, sizeof(void *));
669
670         binder_alloc_debug(BINDER_DEBUG_BUFFER_ALLOC,
671                      "%d: binder_free_buf %pK size %zd buffer_size %zd\n",
672                       alloc->pid, buffer, size, buffer_size);
673
674         BUG_ON(buffer->free);
675         BUG_ON(size > buffer_size);
676         BUG_ON(buffer->transaction != NULL);
677         BUG_ON(buffer->user_data < alloc->buffer);
678         BUG_ON(buffer->user_data > alloc->buffer + alloc->buffer_size);
679
680         if (buffer->async_transaction) {
681                 alloc->free_async_space += buffer_size;
682                 binder_alloc_debug(BINDER_DEBUG_BUFFER_ALLOC_ASYNC,
683                              "%d: binder_free_buf size %zd async free %zd\n",
684                               alloc->pid, size, alloc->free_async_space);
685         }
686
687         binder_lru_freelist_add(alloc, PAGE_ALIGN(buffer->user_data),
688                                 (buffer->user_data + buffer_size) & PAGE_MASK);
689
690         rb_erase(&buffer->rb_node, &alloc->allocated_buffers);
691         buffer->free = 1;
692         if (!list_is_last(&buffer->entry, &alloc->buffers)) {
693                 struct binder_buffer *next = binder_buffer_next(buffer);
694
695                 if (next->free) {
696                         rb_erase(&next->rb_node, &alloc->free_buffers);
697                         binder_delete_free_buffer(alloc, next);
698                 }
699         }
700         if (alloc->buffers.next != &buffer->entry) {
701                 struct binder_buffer *prev = binder_buffer_prev(buffer);
702
703                 if (prev->free) {
704                         binder_delete_free_buffer(alloc, buffer);
705                         rb_erase(&prev->rb_node, &alloc->free_buffers);
706                         buffer = prev;
707                 }
708         }
709         binder_insert_free_buffer(alloc, buffer);
710 }
711
712 /**
713  * binder_alloc_get_page() - get kernel pointer for given buffer offset
714  * @alloc: binder_alloc for this proc
715  * @buffer: binder buffer to be accessed
716  * @buffer_offset: offset into @buffer data
717  * @pgoffp: address to copy final page offset to
718  *
719  * Lookup the struct page corresponding to the address
720  * at @buffer_offset into @buffer->user_data. If @pgoffp is not
721  * NULL, the byte-offset into the page is written there.
722  *
723  * The caller is responsible to ensure that the offset points
724  * to a valid address within the @buffer and that @buffer is
725  * not freeable by the user. Since it can't be freed, we are
726  * guaranteed that the corresponding elements of @alloc->pages[]
727  * cannot change.
728  *
729  * Return: struct page
730  */
731 static struct page *binder_alloc_get_page(struct binder_alloc *alloc,
732                                           struct binder_buffer *buffer,
733                                           binder_size_t buffer_offset,
734                                           pgoff_t *pgoffp)
735 {
736         binder_size_t buffer_space_offset = buffer_offset +
737                 (buffer->user_data - alloc->buffer);
738         pgoff_t pgoff = buffer_space_offset & ~PAGE_MASK;
739         size_t index = buffer_space_offset >> PAGE_SHIFT;
740         struct binder_lru_page *lru_page;
741
742         lru_page = &alloc->pages[index];
743         *pgoffp = pgoff;
744         return lru_page->page_ptr;
745 }
746
747 /**
748  * binder_alloc_clear_buf() - zero out buffer
749  * @alloc: binder_alloc for this proc
750  * @buffer: binder buffer to be cleared
751  *
752  * memset the given buffer to 0
753  */
754 static void binder_alloc_clear_buf(struct binder_alloc *alloc,
755                                    struct binder_buffer *buffer)
756 {
757         size_t bytes = binder_alloc_buffer_size(alloc, buffer);
758         binder_size_t buffer_offset = 0;
759
760         while (bytes) {
761                 unsigned long size;
762                 struct page *page;
763                 pgoff_t pgoff;
764
765                 page = binder_alloc_get_page(alloc, buffer,
766                                              buffer_offset, &pgoff);
767                 size = min_t(size_t, bytes, PAGE_SIZE - pgoff);
768                 memset_page(page, pgoff, 0, size);
769                 bytes -= size;
770                 buffer_offset += size;
771         }
772 }
773
774 /**
775  * binder_alloc_free_buf() - free a binder buffer
776  * @alloc:      binder_alloc for this proc
777  * @buffer:     kernel pointer to buffer
778  *
779  * Free the buffer allocated via binder_alloc_new_buf()
780  */
781 void binder_alloc_free_buf(struct binder_alloc *alloc,
782                             struct binder_buffer *buffer)
783 {
784         /*
785          * We could eliminate the call to binder_alloc_clear_buf()
786          * from binder_alloc_deferred_release() by moving this to
787          * binder_free_buf_locked(). However, that could
788          * increase contention for the alloc->lock if clear_on_free
789          * is used frequently for large buffers. This lock is not
790          * needed for correctness here.
791          */
792         if (buffer->clear_on_free) {
793                 binder_alloc_clear_buf(alloc, buffer);
794                 buffer->clear_on_free = false;
795         }
796         spin_lock(&alloc->lock);
797         binder_free_buf_locked(alloc, buffer);
798         spin_unlock(&alloc->lock);
799 }
800
801 /**
802  * binder_alloc_mmap_handler() - map virtual address space for proc
803  * @alloc:      alloc structure for this proc
804  * @vma:        vma passed to mmap()
805  *
806  * Called by binder_mmap() to initialize the space specified in
807  * vma for allocating binder buffers
808  *
809  * Return:
810  *      0 = success
811  *      -EBUSY = address space already mapped
812  *      -ENOMEM = failed to map memory to given address space
813  */
814 int binder_alloc_mmap_handler(struct binder_alloc *alloc,
815                               struct vm_area_struct *vma)
816 {
817         struct binder_buffer *buffer;
818         const char *failure_string;
819         int ret, i;
820
821         if (unlikely(vma->vm_mm != alloc->mm)) {
822                 ret = -EINVAL;
823                 failure_string = "invalid vma->vm_mm";
824                 goto err_invalid_mm;
825         }
826
827         mutex_lock(&binder_alloc_mmap_lock);
828         if (alloc->buffer_size) {
829                 ret = -EBUSY;
830                 failure_string = "already mapped";
831                 goto err_already_mapped;
832         }
833         alloc->buffer_size = min_t(unsigned long, vma->vm_end - vma->vm_start,
834                                    SZ_4M);
835         mutex_unlock(&binder_alloc_mmap_lock);
836
837         alloc->buffer = vma->vm_start;
838
839         alloc->pages = kcalloc(alloc->buffer_size / PAGE_SIZE,
840                                sizeof(alloc->pages[0]),
841                                GFP_KERNEL);
842         if (alloc->pages == NULL) {
843                 ret = -ENOMEM;
844                 failure_string = "alloc page array";
845                 goto err_alloc_pages_failed;
846         }
847
848         for (i = 0; i < alloc->buffer_size / PAGE_SIZE; i++) {
849                 alloc->pages[i].alloc = alloc;
850                 INIT_LIST_HEAD(&alloc->pages[i].lru);
851         }
852
853         buffer = kzalloc(sizeof(*buffer), GFP_KERNEL);
854         if (!buffer) {
855                 ret = -ENOMEM;
856                 failure_string = "alloc buffer struct";
857                 goto err_alloc_buf_struct_failed;
858         }
859
860         buffer->user_data = alloc->buffer;
861         list_add(&buffer->entry, &alloc->buffers);
862         buffer->free = 1;
863         binder_insert_free_buffer(alloc, buffer);
864         alloc->free_async_space = alloc->buffer_size / 2;
865
866         /* Signal binder_alloc is fully initialized */
867         binder_alloc_set_vma(alloc, vma);
868
869         return 0;
870
871 err_alloc_buf_struct_failed:
872         kfree(alloc->pages);
873         alloc->pages = NULL;
874 err_alloc_pages_failed:
875         alloc->buffer = 0;
876         mutex_lock(&binder_alloc_mmap_lock);
877         alloc->buffer_size = 0;
878 err_already_mapped:
879         mutex_unlock(&binder_alloc_mmap_lock);
880 err_invalid_mm:
881         binder_alloc_debug(BINDER_DEBUG_USER_ERROR,
882                            "%s: %d %lx-%lx %s failed %d\n", __func__,
883                            alloc->pid, vma->vm_start, vma->vm_end,
884                            failure_string, ret);
885         return ret;
886 }
887
888
889 void binder_alloc_deferred_release(struct binder_alloc *alloc)
890 {
891         struct rb_node *n;
892         int buffers, page_count;
893         struct binder_buffer *buffer;
894
895         buffers = 0;
896         spin_lock(&alloc->lock);
897         BUG_ON(alloc->vma);
898
899         while ((n = rb_first(&alloc->allocated_buffers))) {
900                 buffer = rb_entry(n, struct binder_buffer, rb_node);
901
902                 /* Transaction should already have been freed */
903                 BUG_ON(buffer->transaction);
904
905                 if (buffer->clear_on_free) {
906                         binder_alloc_clear_buf(alloc, buffer);
907                         buffer->clear_on_free = false;
908                 }
909                 binder_free_buf_locked(alloc, buffer);
910                 buffers++;
911         }
912
913         while (!list_empty(&alloc->buffers)) {
914                 buffer = list_first_entry(&alloc->buffers,
915                                           struct binder_buffer, entry);
916                 WARN_ON(!buffer->free);
917
918                 list_del(&buffer->entry);
919                 WARN_ON_ONCE(!list_empty(&alloc->buffers));
920                 kfree(buffer);
921         }
922
923         page_count = 0;
924         if (alloc->pages) {
925                 int i;
926
927                 for (i = 0; i < alloc->buffer_size / PAGE_SIZE; i++) {
928                         bool on_lru;
929
930                         if (!alloc->pages[i].page_ptr)
931                                 continue;
932
933                         on_lru = list_lru_del_obj(&binder_freelist,
934                                                   &alloc->pages[i].lru);
935                         binder_alloc_debug(BINDER_DEBUG_BUFFER_ALLOC,
936                                      "%s: %d: page %d %s\n",
937                                      __func__, alloc->pid, i,
938                                      on_lru ? "on lru" : "active");
939                         __free_page(alloc->pages[i].page_ptr);
940                         page_count++;
941                 }
942                 kfree(alloc->pages);
943         }
944         spin_unlock(&alloc->lock);
945         if (alloc->mm)
946                 mmdrop(alloc->mm);
947
948         binder_alloc_debug(BINDER_DEBUG_OPEN_CLOSE,
949                      "%s: %d buffers %d, pages %d\n",
950                      __func__, alloc->pid, buffers, page_count);
951 }
952
953 /**
954  * binder_alloc_print_allocated() - print buffer info
955  * @m:     seq_file for output via seq_printf()
956  * @alloc: binder_alloc for this proc
957  *
958  * Prints information about every buffer associated with
959  * the binder_alloc state to the given seq_file
960  */
961 void binder_alloc_print_allocated(struct seq_file *m,
962                                   struct binder_alloc *alloc)
963 {
964         struct binder_buffer *buffer;
965         struct rb_node *n;
966
967         spin_lock(&alloc->lock);
968         for (n = rb_first(&alloc->allocated_buffers); n; n = rb_next(n)) {
969                 buffer = rb_entry(n, struct binder_buffer, rb_node);
970                 seq_printf(m, "  buffer %d: %lx size %zd:%zd:%zd %s\n",
971                            buffer->debug_id,
972                            buffer->user_data - alloc->buffer,
973                            buffer->data_size, buffer->offsets_size,
974                            buffer->extra_buffers_size,
975                            buffer->transaction ? "active" : "delivered");
976         }
977         spin_unlock(&alloc->lock);
978 }
979
980 /**
981  * binder_alloc_print_pages() - print page usage
982  * @m:     seq_file for output via seq_printf()
983  * @alloc: binder_alloc for this proc
984  */
985 void binder_alloc_print_pages(struct seq_file *m,
986                               struct binder_alloc *alloc)
987 {
988         struct binder_lru_page *page;
989         int i;
990         int active = 0;
991         int lru = 0;
992         int free = 0;
993
994         spin_lock(&alloc->lock);
995         /*
996          * Make sure the binder_alloc is fully initialized, otherwise we might
997          * read inconsistent state.
998          */
999         if (binder_alloc_get_vma(alloc) != NULL) {
1000                 for (i = 0; i < alloc->buffer_size / PAGE_SIZE; i++) {
1001                         page = &alloc->pages[i];
1002                         if (!page->page_ptr)
1003                                 free++;
1004                         else if (list_empty(&page->lru))
1005                                 active++;
1006                         else
1007                                 lru++;
1008                 }
1009         }
1010         spin_unlock(&alloc->lock);
1011         seq_printf(m, "  pages: %d:%d:%d\n", active, lru, free);
1012         seq_printf(m, "  pages high watermark: %zu\n", alloc->pages_high);
1013 }
1014
1015 /**
1016  * binder_alloc_get_allocated_count() - return count of buffers
1017  * @alloc: binder_alloc for this proc
1018  *
1019  * Return: count of allocated buffers
1020  */
1021 int binder_alloc_get_allocated_count(struct binder_alloc *alloc)
1022 {
1023         struct rb_node *n;
1024         int count = 0;
1025
1026         spin_lock(&alloc->lock);
1027         for (n = rb_first(&alloc->allocated_buffers); n != NULL; n = rb_next(n))
1028                 count++;
1029         spin_unlock(&alloc->lock);
1030         return count;
1031 }
1032
1033
1034 /**
1035  * binder_alloc_vma_close() - invalidate address space
1036  * @alloc: binder_alloc for this proc
1037  *
1038  * Called from binder_vma_close() when releasing address space.
1039  * Clears alloc->vma to prevent new incoming transactions from
1040  * allocating more buffers.
1041  */
1042 void binder_alloc_vma_close(struct binder_alloc *alloc)
1043 {
1044         binder_alloc_set_vma(alloc, NULL);
1045 }
1046
1047 /**
1048  * binder_alloc_free_page() - shrinker callback to free pages
1049  * @item:   item to free
1050  * @lock:   lock protecting the item
1051  * @cb_arg: callback argument
1052  *
1053  * Called from list_lru_walk() in binder_shrink_scan() to free
1054  * up pages when the system is under memory pressure.
1055  */
1056 enum lru_status binder_alloc_free_page(struct list_head *item,
1057                                        struct list_lru_one *lru,
1058                                        spinlock_t *lock,
1059                                        void *cb_arg)
1060         __must_hold(lock)
1061 {
1062         struct binder_lru_page *page = container_of(item, typeof(*page), lru);
1063         struct binder_alloc *alloc = page->alloc;
1064         struct mm_struct *mm = alloc->mm;
1065         struct vm_area_struct *vma;
1066         struct page *page_to_free;
1067         unsigned long page_addr;
1068         size_t index;
1069
1070         if (!mmget_not_zero(mm))
1071                 goto err_mmget;
1072         if (!mmap_read_trylock(mm))
1073                 goto err_mmap_read_lock_failed;
1074         if (!spin_trylock(&alloc->lock))
1075                 goto err_get_alloc_lock_failed;
1076         if (!page->page_ptr)
1077                 goto err_page_already_freed;
1078
1079         index = page - alloc->pages;
1080         page_addr = alloc->buffer + index * PAGE_SIZE;
1081
1082         vma = vma_lookup(mm, page_addr);
1083         if (vma && vma != binder_alloc_get_vma(alloc))
1084                 goto err_invalid_vma;
1085
1086         trace_binder_unmap_kernel_start(alloc, index);
1087
1088         page_to_free = page->page_ptr;
1089         page->page_ptr = NULL;
1090
1091         trace_binder_unmap_kernel_end(alloc, index);
1092
1093         list_lru_isolate(lru, item);
1094         spin_unlock(&alloc->lock);
1095         spin_unlock(lock);
1096
1097         if (vma) {
1098                 trace_binder_unmap_user_start(alloc, index);
1099
1100                 zap_page_range_single(vma, page_addr, PAGE_SIZE, NULL);
1101
1102                 trace_binder_unmap_user_end(alloc, index);
1103         }
1104
1105         mmap_read_unlock(mm);
1106         mmput_async(mm);
1107         __free_page(page_to_free);
1108
1109         spin_lock(lock);
1110         return LRU_REMOVED_RETRY;
1111
1112 err_invalid_vma:
1113 err_page_already_freed:
1114         spin_unlock(&alloc->lock);
1115 err_get_alloc_lock_failed:
1116         mmap_read_unlock(mm);
1117 err_mmap_read_lock_failed:
1118         mmput_async(mm);
1119 err_mmget:
1120         return LRU_SKIP;
1121 }
1122
1123 static unsigned long
1124 binder_shrink_count(struct shrinker *shrink, struct shrink_control *sc)
1125 {
1126         return list_lru_count(&binder_freelist);
1127 }
1128
1129 static unsigned long
1130 binder_shrink_scan(struct shrinker *shrink, struct shrink_control *sc)
1131 {
1132         return list_lru_walk(&binder_freelist, binder_alloc_free_page,
1133                             NULL, sc->nr_to_scan);
1134 }
1135
1136 static struct shrinker *binder_shrinker;
1137
1138 /**
1139  * binder_alloc_init() - called by binder_open() for per-proc initialization
1140  * @alloc: binder_alloc for this proc
1141  *
1142  * Called from binder_open() to initialize binder_alloc fields for
1143  * new binder proc
1144  */
1145 void binder_alloc_init(struct binder_alloc *alloc)
1146 {
1147         alloc->pid = current->group_leader->pid;
1148         alloc->mm = current->mm;
1149         mmgrab(alloc->mm);
1150         spin_lock_init(&alloc->lock);
1151         INIT_LIST_HEAD(&alloc->buffers);
1152 }
1153
1154 int binder_alloc_shrinker_init(void)
1155 {
1156         int ret;
1157
1158         ret = list_lru_init(&binder_freelist);
1159         if (ret)
1160                 return ret;
1161
1162         binder_shrinker = shrinker_alloc(0, "android-binder");
1163         if (!binder_shrinker) {
1164                 list_lru_destroy(&binder_freelist);
1165                 return -ENOMEM;
1166         }
1167
1168         binder_shrinker->count_objects = binder_shrink_count;
1169         binder_shrinker->scan_objects = binder_shrink_scan;
1170
1171         shrinker_register(binder_shrinker);
1172
1173         return 0;
1174 }
1175
1176 void binder_alloc_shrinker_exit(void)
1177 {
1178         shrinker_free(binder_shrinker);
1179         list_lru_destroy(&binder_freelist);
1180 }
1181
1182 /**
1183  * check_buffer() - verify that buffer/offset is safe to access
1184  * @alloc: binder_alloc for this proc
1185  * @buffer: binder buffer to be accessed
1186  * @offset: offset into @buffer data
1187  * @bytes: bytes to access from offset
1188  *
1189  * Check that the @offset/@bytes are within the size of the given
1190  * @buffer and that the buffer is currently active and not freeable.
1191  * Offsets must also be multiples of sizeof(u32). The kernel is
1192  * allowed to touch the buffer in two cases:
1193  *
1194  * 1) when the buffer is being created:
1195  *     (buffer->free == 0 && buffer->allow_user_free == 0)
1196  * 2) when the buffer is being torn down:
1197  *     (buffer->free == 0 && buffer->transaction == NULL).
1198  *
1199  * Return: true if the buffer is safe to access
1200  */
1201 static inline bool check_buffer(struct binder_alloc *alloc,
1202                                 struct binder_buffer *buffer,
1203                                 binder_size_t offset, size_t bytes)
1204 {
1205         size_t buffer_size = binder_alloc_buffer_size(alloc, buffer);
1206
1207         return buffer_size >= bytes &&
1208                 offset <= buffer_size - bytes &&
1209                 IS_ALIGNED(offset, sizeof(u32)) &&
1210                 !buffer->free &&
1211                 (!buffer->allow_user_free || !buffer->transaction);
1212 }
1213
1214 /**
1215  * binder_alloc_copy_user_to_buffer() - copy src user to tgt user
1216  * @alloc: binder_alloc for this proc
1217  * @buffer: binder buffer to be accessed
1218  * @buffer_offset: offset into @buffer data
1219  * @from: userspace pointer to source buffer
1220  * @bytes: bytes to copy
1221  *
1222  * Copy bytes from source userspace to target buffer.
1223  *
1224  * Return: bytes remaining to be copied
1225  */
1226 unsigned long
1227 binder_alloc_copy_user_to_buffer(struct binder_alloc *alloc,
1228                                  struct binder_buffer *buffer,
1229                                  binder_size_t buffer_offset,
1230                                  const void __user *from,
1231                                  size_t bytes)
1232 {
1233         if (!check_buffer(alloc, buffer, buffer_offset, bytes))
1234                 return bytes;
1235
1236         while (bytes) {
1237                 unsigned long size;
1238                 unsigned long ret;
1239                 struct page *page;
1240                 pgoff_t pgoff;
1241                 void *kptr;
1242
1243                 page = binder_alloc_get_page(alloc, buffer,
1244                                              buffer_offset, &pgoff);
1245                 size = min_t(size_t, bytes, PAGE_SIZE - pgoff);
1246                 kptr = kmap_local_page(page) + pgoff;
1247                 ret = copy_from_user(kptr, from, size);
1248                 kunmap_local(kptr);
1249                 if (ret)
1250                         return bytes - size + ret;
1251                 bytes -= size;
1252                 from += size;
1253                 buffer_offset += size;
1254         }
1255         return 0;
1256 }
1257
1258 static int binder_alloc_do_buffer_copy(struct binder_alloc *alloc,
1259                                        bool to_buffer,
1260                                        struct binder_buffer *buffer,
1261                                        binder_size_t buffer_offset,
1262                                        void *ptr,
1263                                        size_t bytes)
1264 {
1265         /* All copies must be 32-bit aligned and 32-bit size */
1266         if (!check_buffer(alloc, buffer, buffer_offset, bytes))
1267                 return -EINVAL;
1268
1269         while (bytes) {
1270                 unsigned long size;
1271                 struct page *page;
1272                 pgoff_t pgoff;
1273
1274                 page = binder_alloc_get_page(alloc, buffer,
1275                                              buffer_offset, &pgoff);
1276                 size = min_t(size_t, bytes, PAGE_SIZE - pgoff);
1277                 if (to_buffer)
1278                         memcpy_to_page(page, pgoff, ptr, size);
1279                 else
1280                         memcpy_from_page(ptr, page, pgoff, size);
1281                 bytes -= size;
1282                 pgoff = 0;
1283                 ptr = ptr + size;
1284                 buffer_offset += size;
1285         }
1286         return 0;
1287 }
1288
1289 int binder_alloc_copy_to_buffer(struct binder_alloc *alloc,
1290                                 struct binder_buffer *buffer,
1291                                 binder_size_t buffer_offset,
1292                                 void *src,
1293                                 size_t bytes)
1294 {
1295         return binder_alloc_do_buffer_copy(alloc, true, buffer, buffer_offset,
1296                                            src, bytes);
1297 }
1298
1299 int binder_alloc_copy_from_buffer(struct binder_alloc *alloc,
1300                                   void *dest,
1301                                   struct binder_buffer *buffer,
1302                                   binder_size_t buffer_offset,
1303                                   size_t bytes)
1304 {
1305         return binder_alloc_do_buffer_copy(alloc, false, buffer, buffer_offset,
1306                                            dest, bytes);
1307 }