drivers/android/binder_alloc.c

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