drivers/android/binder_alloc.c

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