fs/f2fs/compress.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * f2fs compress support
   4  *
   5  * Copyright (c) 2019 Chao Yu <chao@kernel.org>
   6  */
   7
   8 #include <linux/fs.h>
   9 #include <linux/f2fs_fs.h>
  10 #include <linux/writeback.h>
  11 #include <linux/backing-dev.h>
  12 #include <linux/lzo.h>
  13 #include <linux/lz4.h>
  14 #include <linux/zstd.h>
  15
  16 #include "f2fs.h"
  17 #include "node.h"
  18 #include <trace/events/f2fs.h>
  19
  20 static struct kmem_cache *cic_entry_slab;
  21 static struct kmem_cache *dic_entry_slab;
  22
  23 static void *page_array_alloc(struct inode *inode, int nr)
  24 {
  25         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
  26         unsigned int size = sizeof(struct page *) * nr;
  27
  28         if (likely(size <= sbi->page_array_slab_size))
  29                 return kmem_cache_zalloc(sbi->page_array_slab, GFP_NOFS);
  30         return f2fs_kzalloc(sbi, size, GFP_NOFS);
  31 }
  32
  33 static void page_array_free(struct inode *inode, void *pages, int nr)
  34 {
  35         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
  36         unsigned int size = sizeof(struct page *) * nr;
  37
  38         if (!pages)
  39                 return;
  40
  41         if (likely(size <= sbi->page_array_slab_size))
  42                 kmem_cache_free(sbi->page_array_slab, pages);
  43         else
  44                 kfree(pages);
  45 }
  46
  47 struct f2fs_compress_ops {
  48         int (*init_compress_ctx)(struct compress_ctx *cc);
  49         void (*destroy_compress_ctx)(struct compress_ctx *cc);
  50         int (*compress_pages)(struct compress_ctx *cc);
  51         int (*init_decompress_ctx)(struct decompress_io_ctx *dic);
  52         void (*destroy_decompress_ctx)(struct decompress_io_ctx *dic);
  53         int (*decompress_pages)(struct decompress_io_ctx *dic);
  54 };
  55
  56 static unsigned int offset_in_cluster(struct compress_ctx *cc, pgoff_t index)
  57 {
  58         return index & (cc->cluster_size - 1);
  59 }
  60
  61 static pgoff_t cluster_idx(struct compress_ctx *cc, pgoff_t index)
  62 {
  63         return index >> cc->log_cluster_size;
  64 }
  65
  66 static pgoff_t start_idx_of_cluster(struct compress_ctx *cc)
  67 {
  68         return cc->cluster_idx << cc->log_cluster_size;
  69 }
  70
  71 bool f2fs_is_compressed_page(struct page *page)
  72 {
  73         if (!PagePrivate(page))
  74                 return false;
  75         if (!page_private(page))
  76                 return false;
  77         if (IS_ATOMIC_WRITTEN_PAGE(page) || IS_DUMMY_WRITTEN_PAGE(page))
  78                 return false;
  79         /*
  80          * page->private may be set with pid.
  81          * pid_max is enough to check if it is traced.
  82          */
  83         if (IS_IO_TRACED_PAGE(page))
  84                 return false;
  85
  86         f2fs_bug_on(F2FS_M_SB(page->mapping),
  87                 *((u32 *)page_private(page)) != F2FS_COMPRESSED_PAGE_MAGIC);
  88         return true;
  89 }
  90
  91 static void f2fs_set_compressed_page(struct page *page,
  92                 struct inode *inode, pgoff_t index, void *data)
  93 {
  94         SetPagePrivate(page);
  95         set_page_private(page, (unsigned long)data);
  96
  97         /* i_crypto_info and iv index */
  98         page->index = index;
  99         page->mapping = inode->i_mapping;
 100 }
 101
 102 static void f2fs_drop_rpages(struct compress_ctx *cc, int len, bool unlock)
 103 {
 104         int i;
 105
 106         for (i = 0; i < len; i++) {
 107                 if (!cc->rpages[i])
 108                         continue;
 109                 if (unlock)
 110                         unlock_page(cc->rpages[i]);
 111                 else
 112                         put_page(cc->rpages[i]);
 113         }
 114 }
 115
 116 static void f2fs_put_rpages(struct compress_ctx *cc)
 117 {
 118         f2fs_drop_rpages(cc, cc->cluster_size, false);
 119 }
 120
 121 static void f2fs_unlock_rpages(struct compress_ctx *cc, int len)
 122 {
 123         f2fs_drop_rpages(cc, len, true);
 124 }
 125
 126 static void f2fs_put_rpages_mapping(struct address_space *mapping,
 127                                 pgoff_t start, int len)
 128 {
 129         int i;
 130
 131         for (i = 0; i < len; i++) {
 132                 struct page *page = find_get_page(mapping, start + i);
 133
 134                 put_page(page);
 135                 put_page(page);
 136         }
 137 }
 138
 139 static void f2fs_put_rpages_wbc(struct compress_ctx *cc,
 140                 struct writeback_control *wbc, bool redirty, int unlock)
 141 {
 142         unsigned int i;
 143
 144         for (i = 0; i < cc->cluster_size; i++) {
 145                 if (!cc->rpages[i])
 146                         continue;
 147                 if (redirty)
 148                         redirty_page_for_writepage(wbc, cc->rpages[i]);
 149                 f2fs_put_page(cc->rpages[i], unlock);
 150         }
 151 }
 152
 153 struct page *f2fs_compress_control_page(struct page *page)
 154 {
 155         return ((struct compress_io_ctx *)page_private(page))->rpages[0];
 156 }
 157
 158 int f2fs_init_compress_ctx(struct compress_ctx *cc)
 159 {
 160         if (cc->rpages)
 161                 return 0;
 162
 163         cc->rpages = page_array_alloc(cc->inode, cc->cluster_size);
 164         return cc->rpages ? 0 : -ENOMEM;
 165 }
 166
 167 void f2fs_destroy_compress_ctx(struct compress_ctx *cc)
 168 {
 169         page_array_free(cc->inode, cc->rpages, cc->cluster_size);
 170         cc->rpages = NULL;
 171         cc->nr_rpages = 0;
 172         cc->nr_cpages = 0;
 173         cc->cluster_idx = NULL_CLUSTER;
 174 }
 175
 176 void f2fs_compress_ctx_add_page(struct compress_ctx *cc, struct page *page)
 177 {
 178         unsigned int cluster_ofs;
 179
 180         if (!f2fs_cluster_can_merge_page(cc, page->index))
 181                 f2fs_bug_on(F2FS_I_SB(cc->inode), 1);
 182
 183         cluster_ofs = offset_in_cluster(cc, page->index);
 184         cc->rpages[cluster_ofs] = page;
 185         cc->nr_rpages++;
 186         cc->cluster_idx = cluster_idx(cc, page->index);
 187 }
 188
 189 #ifdef CONFIG_F2FS_FS_LZO
 190 static int lzo_init_compress_ctx(struct compress_ctx *cc)
 191 {
 192         cc->private = f2fs_kvmalloc(F2FS_I_SB(cc->inode),
 193                                 LZO1X_MEM_COMPRESS, GFP_NOFS);
 194         if (!cc->private)
 195                 return -ENOMEM;
 196
 197         cc->clen = lzo1x_worst_compress(PAGE_SIZE << cc->log_cluster_size);
 198         return 0;
 199 }
 200
 201 static void lzo_destroy_compress_ctx(struct compress_ctx *cc)
 202 {
 203         kvfree(cc->private);
 204         cc->private = NULL;
 205 }
 206
 207 static int lzo_compress_pages(struct compress_ctx *cc)
 208 {
 209         int ret;
 210
 211         ret = lzo1x_1_compress(cc->rbuf, cc->rlen, cc->cbuf->cdata,
 212                                         &cc->clen, cc->private);
 213         if (ret != LZO_E_OK) {
 214                 printk_ratelimited("%sF2FS-fs (%s): lzo compress failed, ret:%d\n",
 215                                 KERN_ERR, F2FS_I_SB(cc->inode)->sb->s_id, ret);
 216                 return -EIO;
 217         }
 218         return 0;
 219 }
 220
 221 static int lzo_decompress_pages(struct decompress_io_ctx *dic)
 222 {
 223         int ret;
 224
 225         ret = lzo1x_decompress_safe(dic->cbuf->cdata, dic->clen,
 226                                                 dic->rbuf, &dic->rlen);
 227         if (ret != LZO_E_OK) {
 228                 printk_ratelimited("%sF2FS-fs (%s): lzo decompress failed, ret:%d\n",
 229                                 KERN_ERR, F2FS_I_SB(dic->inode)->sb->s_id, ret);
 230                 return -EIO;
 231         }
 232
 233         if (dic->rlen != PAGE_SIZE << dic->log_cluster_size) {
 234                 printk_ratelimited("%sF2FS-fs (%s): lzo invalid rlen:%zu, "
 235                                         "expected:%lu\n", KERN_ERR,
 236                                         F2FS_I_SB(dic->inode)->sb->s_id,
 237                                         dic->rlen,
 238                                         PAGE_SIZE << dic->log_cluster_size);
 239                 return -EIO;
 240         }
 241         return 0;
 242 }
 243
 244 static const struct f2fs_compress_ops f2fs_lzo_ops = {
 245         .init_compress_ctx      = lzo_init_compress_ctx,
 246         .destroy_compress_ctx   = lzo_destroy_compress_ctx,
 247         .compress_pages         = lzo_compress_pages,
 248         .decompress_pages       = lzo_decompress_pages,
 249 };
 250 #endif
 251
 252 #ifdef CONFIG_F2FS_FS_LZ4
 253 static int lz4_init_compress_ctx(struct compress_ctx *cc)
 254 {
 255         unsigned int size = LZ4_MEM_COMPRESS;
 256
 257 #ifdef CONFIG_F2FS_FS_LZ4HC
 258         if (F2FS_I(cc->inode)->i_compress_flag >> COMPRESS_LEVEL_OFFSET)
 259                 size = LZ4HC_MEM_COMPRESS;
 260 #endif
 261
 262         cc->private = f2fs_kvmalloc(F2FS_I_SB(cc->inode), size, GFP_NOFS);
 263         if (!cc->private)
 264                 return -ENOMEM;
 265
 266         /*
 267          * we do not change cc->clen to LZ4_compressBound(inputsize) to
 268          * adapt worst compress case, because lz4 compressor can handle
 269          * output budget properly.
 270          */
 271         cc->clen = cc->rlen - PAGE_SIZE - COMPRESS_HEADER_SIZE;
 272         return 0;
 273 }
 274
 275 static void lz4_destroy_compress_ctx(struct compress_ctx *cc)
 276 {
 277         kvfree(cc->private);
 278         cc->private = NULL;
 279 }
 280
 281 #ifdef CONFIG_F2FS_FS_LZ4HC
 282 static int lz4hc_compress_pages(struct compress_ctx *cc)
 283 {
 284         unsigned char level = F2FS_I(cc->inode)->i_compress_flag >>
 285                                                 COMPRESS_LEVEL_OFFSET;
 286         int len;
 287
 288         if (level)
 289                 len = LZ4_compress_HC(cc->rbuf, cc->cbuf->cdata, cc->rlen,
 290                                         cc->clen, level, cc->private);
 291         else
 292                 len = LZ4_compress_default(cc->rbuf, cc->cbuf->cdata, cc->rlen,
 293                                                 cc->clen, cc->private);
 294         if (!len)
 295                 return -EAGAIN;
 296
 297         cc->clen = len;
 298         return 0;
 299 }
 300 #endif
 301
 302 static int lz4_compress_pages(struct compress_ctx *cc)
 303 {
 304         int len;
 305
 306 #ifdef CONFIG_F2FS_FS_LZ4HC
 307         return lz4hc_compress_pages(cc);
 308 #endif
 309         len = LZ4_compress_default(cc->rbuf, cc->cbuf->cdata, cc->rlen,
 310                                                 cc->clen, cc->private);
 311         if (!len)
 312                 return -EAGAIN;
 313
 314         cc->clen = len;
 315         return 0;
 316 }
 317
 318 static int lz4_decompress_pages(struct decompress_io_ctx *dic)
 319 {
 320         int ret;
 321
 322         ret = LZ4_decompress_safe(dic->cbuf->cdata, dic->rbuf,
 323                                                 dic->clen, dic->rlen);
 324         if (ret < 0) {
 325                 printk_ratelimited("%sF2FS-fs (%s): lz4 decompress failed, ret:%d\n",
 326                                 KERN_ERR, F2FS_I_SB(dic->inode)->sb->s_id, ret);
 327                 return -EIO;
 328         }
 329
 330         if (ret != PAGE_SIZE << dic->log_cluster_size) {
 331                 printk_ratelimited("%sF2FS-fs (%s): lz4 invalid rlen:%zu, "
 332                                         "expected:%lu\n", KERN_ERR,
 333                                         F2FS_I_SB(dic->inode)->sb->s_id,
 334                                         dic->rlen,
 335                                         PAGE_SIZE << dic->log_cluster_size);
 336                 return -EIO;
 337         }
 338         return 0;
 339 }
 340
 341 static const struct f2fs_compress_ops f2fs_lz4_ops = {
 342         .init_compress_ctx      = lz4_init_compress_ctx,
 343         .destroy_compress_ctx   = lz4_destroy_compress_ctx,
 344         .compress_pages         = lz4_compress_pages,
 345         .decompress_pages       = lz4_decompress_pages,
 346 };
 347 #endif
 348
 349 #ifdef CONFIG_F2FS_FS_ZSTD
 350 #define F2FS_ZSTD_DEFAULT_CLEVEL        1
 351
 352 static int zstd_init_compress_ctx(struct compress_ctx *cc)
 353 {
 354         ZSTD_parameters params;
 355         ZSTD_CStream *stream;
 356         void *workspace;
 357         unsigned int workspace_size;
 358         unsigned char level = F2FS_I(cc->inode)->i_compress_flag >>
 359                                                 COMPRESS_LEVEL_OFFSET;
 360
 361         if (!level)
 362                 level = F2FS_ZSTD_DEFAULT_CLEVEL;
 363
 364         params = ZSTD_getParams(level, cc->rlen, 0);
 365         workspace_size = ZSTD_CStreamWorkspaceBound(params.cParams);
 366
 367         workspace = f2fs_kvmalloc(F2FS_I_SB(cc->inode),
 368                                         workspace_size, GFP_NOFS);
 369         if (!workspace)
 370                 return -ENOMEM;
 371
 372         stream = ZSTD_initCStream(params, 0, workspace, workspace_size);
 373         if (!stream) {
 374                 printk_ratelimited("%sF2FS-fs (%s): %s ZSTD_initCStream failed\n",
 375                                 KERN_ERR, F2FS_I_SB(cc->inode)->sb->s_id,
 376                                 __func__);
 377                 kvfree(workspace);
 378                 return -EIO;
 379         }
 380
 381         cc->private = workspace;
 382         cc->private2 = stream;
 383
 384         cc->clen = cc->rlen - PAGE_SIZE - COMPRESS_HEADER_SIZE;
 385         return 0;
 386 }
 387
 388 static void zstd_destroy_compress_ctx(struct compress_ctx *cc)
 389 {
 390         kvfree(cc->private);
 391         cc->private = NULL;
 392         cc->private2 = NULL;
 393 }
 394
 395 static int zstd_compress_pages(struct compress_ctx *cc)
 396 {
 397         ZSTD_CStream *stream = cc->private2;
 398         ZSTD_inBuffer inbuf;
 399         ZSTD_outBuffer outbuf;
 400         int src_size = cc->rlen;
 401         int dst_size = src_size - PAGE_SIZE - COMPRESS_HEADER_SIZE;
 402         int ret;
 403
 404         inbuf.pos = 0;
 405         inbuf.src = cc->rbuf;
 406         inbuf.size = src_size;
 407
 408         outbuf.pos = 0;
 409         outbuf.dst = cc->cbuf->cdata;
 410         outbuf.size = dst_size;
 411
 412         ret = ZSTD_compressStream(stream, &outbuf, &inbuf);
 413         if (ZSTD_isError(ret)) {
 414                 printk_ratelimited("%sF2FS-fs (%s): %s ZSTD_compressStream failed, ret: %d\n",
 415                                 KERN_ERR, F2FS_I_SB(cc->inode)->sb->s_id,
 416                                 __func__, ZSTD_getErrorCode(ret));
 417                 return -EIO;
 418         }
 419
 420         ret = ZSTD_endStream(stream, &outbuf);
 421         if (ZSTD_isError(ret)) {
 422                 printk_ratelimited("%sF2FS-fs (%s): %s ZSTD_endStream returned %d\n",
 423                                 KERN_ERR, F2FS_I_SB(cc->inode)->sb->s_id,
 424                                 __func__, ZSTD_getErrorCode(ret));
 425                 return -EIO;
 426         }
 427
 428         /*
 429          * there is compressed data remained in intermediate buffer due to
 430          * no more space in cbuf.cdata
 431          */
 432         if (ret)
 433                 return -EAGAIN;
 434
 435         cc->clen = outbuf.pos;
 436         return 0;
 437 }
 438
 439 static int zstd_init_decompress_ctx(struct decompress_io_ctx *dic)
 440 {
 441         ZSTD_DStream *stream;
 442         void *workspace;
 443         unsigned int workspace_size;
 444         unsigned int max_window_size =
 445                         MAX_COMPRESS_WINDOW_SIZE(dic->log_cluster_size);
 446
 447         workspace_size = ZSTD_DStreamWorkspaceBound(max_window_size);
 448
 449         workspace = f2fs_kvmalloc(F2FS_I_SB(dic->inode),
 450                                         workspace_size, GFP_NOFS);
 451         if (!workspace)
 452                 return -ENOMEM;
 453
 454         stream = ZSTD_initDStream(max_window_size, workspace, workspace_size);
 455         if (!stream) {
 456                 printk_ratelimited("%sF2FS-fs (%s): %s ZSTD_initDStream failed\n",
 457                                 KERN_ERR, F2FS_I_SB(dic->inode)->sb->s_id,
 458                                 __func__);
 459                 kvfree(workspace);
 460                 return -EIO;
 461         }
 462
 463         dic->private = workspace;
 464         dic->private2 = stream;
 465
 466         return 0;
 467 }
 468
 469 static void zstd_destroy_decompress_ctx(struct decompress_io_ctx *dic)
 470 {
 471         kvfree(dic->private);
 472         dic->private = NULL;
 473         dic->private2 = NULL;
 474 }
 475
 476 static int zstd_decompress_pages(struct decompress_io_ctx *dic)
 477 {
 478         ZSTD_DStream *stream = dic->private2;
 479         ZSTD_inBuffer inbuf;
 480         ZSTD_outBuffer outbuf;
 481         int ret;
 482
 483         inbuf.pos = 0;
 484         inbuf.src = dic->cbuf->cdata;
 485         inbuf.size = dic->clen;
 486
 487         outbuf.pos = 0;
 488         outbuf.dst = dic->rbuf;
 489         outbuf.size = dic->rlen;
 490
 491         ret = ZSTD_decompressStream(stream, &outbuf, &inbuf);
 492         if (ZSTD_isError(ret)) {
 493                 printk_ratelimited("%sF2FS-fs (%s): %s ZSTD_compressStream failed, ret: %d\n",
 494                                 KERN_ERR, F2FS_I_SB(dic->inode)->sb->s_id,
 495                                 __func__, ZSTD_getErrorCode(ret));
 496                 return -EIO;
 497         }
 498
 499         if (dic->rlen != outbuf.pos) {
 500                 printk_ratelimited("%sF2FS-fs (%s): %s ZSTD invalid rlen:%zu, "
 501                                 "expected:%lu\n", KERN_ERR,
 502                                 F2FS_I_SB(dic->inode)->sb->s_id,
 503                                 __func__, dic->rlen,
 504                                 PAGE_SIZE << dic->log_cluster_size);
 505                 return -EIO;
 506         }
 507
 508         return 0;
 509 }
 510
 511 static const struct f2fs_compress_ops f2fs_zstd_ops = {
 512         .init_compress_ctx      = zstd_init_compress_ctx,
 513         .destroy_compress_ctx   = zstd_destroy_compress_ctx,
 514         .compress_pages         = zstd_compress_pages,
 515         .init_decompress_ctx    = zstd_init_decompress_ctx,
 516         .destroy_decompress_ctx = zstd_destroy_decompress_ctx,
 517         .decompress_pages       = zstd_decompress_pages,
 518 };
 519 #endif
 520
 521 #ifdef CONFIG_F2FS_FS_LZO
 522 #ifdef CONFIG_F2FS_FS_LZORLE
 523 static int lzorle_compress_pages(struct compress_ctx *cc)
 524 {
 525         int ret;
 526
 527         ret = lzorle1x_1_compress(cc->rbuf, cc->rlen, cc->cbuf->cdata,
 528                                         &cc->clen, cc->private);
 529         if (ret != LZO_E_OK) {
 530                 printk_ratelimited("%sF2FS-fs (%s): lzo-rle compress failed, ret:%d\n",
 531                                 KERN_ERR, F2FS_I_SB(cc->inode)->sb->s_id, ret);
 532                 return -EIO;
 533         }
 534         return 0;
 535 }
 536
 537 static const struct f2fs_compress_ops f2fs_lzorle_ops = {
 538         .init_compress_ctx      = lzo_init_compress_ctx,
 539         .destroy_compress_ctx   = lzo_destroy_compress_ctx,
 540         .compress_pages         = lzorle_compress_pages,
 541         .decompress_pages       = lzo_decompress_pages,
 542 };
 543 #endif
 544 #endif
 545
 546 static const struct f2fs_compress_ops *f2fs_cops[COMPRESS_MAX] = {
 547 #ifdef CONFIG_F2FS_FS_LZO
 548         &f2fs_lzo_ops,
 549 #else
 550         NULL,
 551 #endif
 552 #ifdef CONFIG_F2FS_FS_LZ4
 553         &f2fs_lz4_ops,
 554 #else
 555         NULL,
 556 #endif
 557 #ifdef CONFIG_F2FS_FS_ZSTD
 558         &f2fs_zstd_ops,
 559 #else
 560         NULL,
 561 #endif
 562 #if defined(CONFIG_F2FS_FS_LZO) && defined(CONFIG_F2FS_FS_LZORLE)
 563         &f2fs_lzorle_ops,
 564 #else
 565         NULL,
 566 #endif
 567 };
 568
 569 bool f2fs_is_compress_backend_ready(struct inode *inode)
 570 {
 571         if (!f2fs_compressed_file(inode))
 572                 return true;
 573         return f2fs_cops[F2FS_I(inode)->i_compress_algorithm];
 574 }
 575
 576 static mempool_t *compress_page_pool;
 577 static int num_compress_pages = 512;
 578 module_param(num_compress_pages, uint, 0444);
 579 MODULE_PARM_DESC(num_compress_pages,
 580                 "Number of intermediate compress pages to preallocate");
 581
 582 int f2fs_init_compress_mempool(void)
 583 {
 584         compress_page_pool = mempool_create_page_pool(num_compress_pages, 0);
 585         if (!compress_page_pool)
 586                 return -ENOMEM;
 587
 588         return 0;
 589 }
 590
 591 void f2fs_destroy_compress_mempool(void)
 592 {
 593         mempool_destroy(compress_page_pool);
 594 }
 595
 596 static struct page *f2fs_compress_alloc_page(void)
 597 {
 598         struct page *page;
 599
 600         page = mempool_alloc(compress_page_pool, GFP_NOFS);
 601         lock_page(page);
 602
 603         return page;
 604 }
 605
 606 static void f2fs_compress_free_page(struct page *page)
 607 {
 608         if (!page)
 609                 return;
 610         set_page_private(page, (unsigned long)NULL);
 611         ClearPagePrivate(page);
 612         page->mapping = NULL;
 613         unlock_page(page);
 614         mempool_free(page, compress_page_pool);
 615 }
 616
 617 #define MAX_VMAP_RETRIES        3
 618
 619 static void *f2fs_vmap(struct page **pages, unsigned int count)
 620 {
 621         int i;
 622         void *buf = NULL;
 623
 624         for (i = 0; i < MAX_VMAP_RETRIES; i++) {
 625                 buf = vm_map_ram(pages, count, -1);
 626                 if (buf)
 627                         break;
 628                 vm_unmap_aliases();
 629         }
 630         return buf;
 631 }
 632
 633 static int f2fs_compress_pages(struct compress_ctx *cc)
 634 {
 635         struct f2fs_inode_info *fi = F2FS_I(cc->inode);
 636         const struct f2fs_compress_ops *cops =
 637                                 f2fs_cops[fi->i_compress_algorithm];
 638         unsigned int max_len, new_nr_cpages;
 639         struct page **new_cpages;
 640         u32 chksum = 0;
 641         int i, ret;
 642
 643         trace_f2fs_compress_pages_start(cc->inode, cc->cluster_idx,
 644                                 cc->cluster_size, fi->i_compress_algorithm);
 645
 646         if (cops->init_compress_ctx) {
 647                 ret = cops->init_compress_ctx(cc);
 648                 if (ret)
 649                         goto out;
 650         }
 651
 652         max_len = COMPRESS_HEADER_SIZE + cc->clen;
 653         cc->nr_cpages = DIV_ROUND_UP(max_len, PAGE_SIZE);
 654
 655         cc->cpages = page_array_alloc(cc->inode, cc->nr_cpages);
 656         if (!cc->cpages) {
 657                 ret = -ENOMEM;
 658                 goto destroy_compress_ctx;
 659         }
 660
 661         for (i = 0; i < cc->nr_cpages; i++) {
 662                 cc->cpages[i] = f2fs_compress_alloc_page();
 663                 if (!cc->cpages[i]) {
 664                         ret = -ENOMEM;
 665                         goto out_free_cpages;
 666                 }
 667         }
 668
 669         cc->rbuf = f2fs_vmap(cc->rpages, cc->cluster_size);
 670         if (!cc->rbuf) {
 671                 ret = -ENOMEM;
 672                 goto out_free_cpages;
 673         }
 674
 675         cc->cbuf = f2fs_vmap(cc->cpages, cc->nr_cpages);
 676         if (!cc->cbuf) {
 677                 ret = -ENOMEM;
 678                 goto out_vunmap_rbuf;
 679         }
 680
 681         ret = cops->compress_pages(cc);
 682         if (ret)
 683                 goto out_vunmap_cbuf;
 684
 685         max_len = PAGE_SIZE * (cc->cluster_size - 1) - COMPRESS_HEADER_SIZE;
 686
 687         if (cc->clen > max_len) {
 688                 ret = -EAGAIN;
 689                 goto out_vunmap_cbuf;
 690         }
 691
 692         cc->cbuf->clen = cpu_to_le32(cc->clen);
 693
 694         if (fi->i_compress_flag & 1 << COMPRESS_CHKSUM)
 695                 chksum = f2fs_crc32(F2FS_I_SB(cc->inode),
 696                                         cc->cbuf->cdata, cc->clen);
 697         cc->cbuf->chksum = cpu_to_le32(chksum);
 698
 699         for (i = 0; i < COMPRESS_DATA_RESERVED_SIZE; i++)
 700                 cc->cbuf->reserved[i] = cpu_to_le32(0);
 701
 702         new_nr_cpages = DIV_ROUND_UP(cc->clen + COMPRESS_HEADER_SIZE, PAGE_SIZE);
 703
 704         /* Now we're going to cut unnecessary tail pages */
 705         new_cpages = page_array_alloc(cc->inode, new_nr_cpages);
 706         if (!new_cpages) {
 707                 ret = -ENOMEM;
 708                 goto out_vunmap_cbuf;
 709         }
 710
 711         /* zero out any unused part of the last page */
 712         memset(&cc->cbuf->cdata[cc->clen], 0,
 713                         (new_nr_cpages * PAGE_SIZE) -
 714                         (cc->clen + COMPRESS_HEADER_SIZE));
 715
 716         vm_unmap_ram(cc->cbuf, cc->nr_cpages);
 717         vm_unmap_ram(cc->rbuf, cc->cluster_size);
 718
 719         for (i = 0; i < cc->nr_cpages; i++) {
 720                 if (i < new_nr_cpages) {
 721                         new_cpages[i] = cc->cpages[i];
 722                         continue;
 723                 }
 724                 f2fs_compress_free_page(cc->cpages[i]);
 725                 cc->cpages[i] = NULL;
 726         }
 727
 728         if (cops->destroy_compress_ctx)
 729                 cops->destroy_compress_ctx(cc);
 730
 731         page_array_free(cc->inode, cc->cpages, cc->nr_cpages);
 732         cc->cpages = new_cpages;
 733         cc->nr_cpages = new_nr_cpages;
 734
 735         trace_f2fs_compress_pages_end(cc->inode, cc->cluster_idx,
 736                                                         cc->clen, ret);
 737         return 0;
 738
 739 out_vunmap_cbuf:
 740         vm_unmap_ram(cc->cbuf, cc->nr_cpages);
 741 out_vunmap_rbuf:
 742         vm_unmap_ram(cc->rbuf, cc->cluster_size);
 743 out_free_cpages:
 744         for (i = 0; i < cc->nr_cpages; i++) {
 745                 if (cc->cpages[i])
 746                         f2fs_compress_free_page(cc->cpages[i]);
 747         }
 748         page_array_free(cc->inode, cc->cpages, cc->nr_cpages);
 749         cc->cpages = NULL;
 750 destroy_compress_ctx:
 751         if (cops->destroy_compress_ctx)
 752                 cops->destroy_compress_ctx(cc);
 753 out:
 754         trace_f2fs_compress_pages_end(cc->inode, cc->cluster_idx,
 755                                                         cc->clen, ret);
 756         return ret;
 757 }
 758
 759 static void f2fs_decompress_cluster(struct decompress_io_ctx *dic)
 760 {
 761         struct f2fs_sb_info *sbi = F2FS_I_SB(dic->inode);
 762         struct f2fs_inode_info *fi = F2FS_I(dic->inode);
 763         const struct f2fs_compress_ops *cops =
 764                         f2fs_cops[fi->i_compress_algorithm];
 765         int ret;
 766         int i;
 767
 768         trace_f2fs_decompress_pages_start(dic->inode, dic->cluster_idx,
 769                                 dic->cluster_size, fi->i_compress_algorithm);
 770
 771         if (dic->failed) {
 772                 ret = -EIO;
 773                 goto out_end_io;
 774         }
 775
 776         dic->tpages = page_array_alloc(dic->inode, dic->cluster_size);
 777         if (!dic->tpages) {
 778                 ret = -ENOMEM;
 779                 goto out_end_io;
 780         }
 781
 782         for (i = 0; i < dic->cluster_size; i++) {
 783                 if (dic->rpages[i]) {
 784                         dic->tpages[i] = dic->rpages[i];
 785                         continue;
 786                 }
 787
 788                 dic->tpages[i] = f2fs_compress_alloc_page();
 789                 if (!dic->tpages[i]) {
 790                         ret = -ENOMEM;
 791                         goto out_end_io;
 792                 }
 793         }
 794
 795         if (cops->init_decompress_ctx) {
 796                 ret = cops->init_decompress_ctx(dic);
 797                 if (ret)
 798                         goto out_end_io;
 799         }
 800
 801         dic->rbuf = f2fs_vmap(dic->tpages, dic->cluster_size);
 802         if (!dic->rbuf) {
 803                 ret = -ENOMEM;
 804                 goto out_destroy_decompress_ctx;
 805         }
 806
 807         dic->cbuf = f2fs_vmap(dic->cpages, dic->nr_cpages);
 808         if (!dic->cbuf) {
 809                 ret = -ENOMEM;
 810                 goto out_vunmap_rbuf;
 811         }
 812
 813         dic->clen = le32_to_cpu(dic->cbuf->clen);
 814         dic->rlen = PAGE_SIZE << dic->log_cluster_size;
 815
 816         if (dic->clen > PAGE_SIZE * dic->nr_cpages - COMPRESS_HEADER_SIZE) {
 817                 ret = -EFSCORRUPTED;
 818                 goto out_vunmap_cbuf;
 819         }
 820
 821         ret = cops->decompress_pages(dic);
 822
 823         if (!ret && (fi->i_compress_flag & 1 << COMPRESS_CHKSUM)) {
 824                 u32 provided = le32_to_cpu(dic->cbuf->chksum);
 825                 u32 calculated = f2fs_crc32(sbi, dic->cbuf->cdata, dic->clen);
 826
 827                 if (provided != calculated) {
 828                         if (!is_inode_flag_set(dic->inode, FI_COMPRESS_CORRUPT)) {
 829                                 set_inode_flag(dic->inode, FI_COMPRESS_CORRUPT);
 830                                 printk_ratelimited(
 831                                         "%sF2FS-fs (%s): checksum invalid, nid = %lu, %x vs %x",
 832                                         KERN_INFO, sbi->sb->s_id, dic->inode->i_ino,
 833                                         provided, calculated);
 834                         }
 835                         set_sbi_flag(sbi, SBI_NEED_FSCK);
 836                 }
 837         }
 838
 839 out_vunmap_cbuf:
 840         vm_unmap_ram(dic->cbuf, dic->nr_cpages);
 841 out_vunmap_rbuf:
 842         vm_unmap_ram(dic->rbuf, dic->cluster_size);
 843 out_destroy_decompress_ctx:
 844         if (cops->destroy_decompress_ctx)
 845                 cops->destroy_decompress_ctx(dic);
 846 out_end_io:
 847         trace_f2fs_decompress_pages_end(dic->inode, dic->cluster_idx,
 848                                                         dic->clen, ret);
 849         f2fs_decompress_end_io(dic, ret);
 850 }
 851
 852 /*
 853  * This is called when a page of a compressed cluster has been read from disk
 854  * (or failed to be read from disk).  It checks whether this page was the last
 855  * page being waited on in the cluster, and if so, it decompresses the cluster
 856  * (or in the case of a failure, cleans up without actually decompressing).
 857  */
 858 void f2fs_end_read_compressed_page(struct page *page, bool failed)
 859 {
 860         struct decompress_io_ctx *dic =
 861                         (struct decompress_io_ctx *)page_private(page);
 862         struct f2fs_sb_info *sbi = F2FS_I_SB(dic->inode);
 863
 864         dec_page_count(sbi, F2FS_RD_DATA);
 865
 866         if (failed)
 867                 WRITE_ONCE(dic->failed, true);
 868
 869         if (atomic_dec_and_test(&dic->remaining_pages))
 870                 f2fs_decompress_cluster(dic);
 871 }
 872
 873 static bool is_page_in_cluster(struct compress_ctx *cc, pgoff_t index)
 874 {
 875         if (cc->cluster_idx == NULL_CLUSTER)
 876                 return true;
 877         return cc->cluster_idx == cluster_idx(cc, index);
 878 }
 879
 880 bool f2fs_cluster_is_empty(struct compress_ctx *cc)
 881 {
 882         return cc->nr_rpages == 0;
 883 }
 884
 885 static bool f2fs_cluster_is_full(struct compress_ctx *cc)
 886 {
 887         return cc->cluster_size == cc->nr_rpages;
 888 }
 889
 890 bool f2fs_cluster_can_merge_page(struct compress_ctx *cc, pgoff_t index)
 891 {
 892         if (f2fs_cluster_is_empty(cc))
 893                 return true;
 894         return is_page_in_cluster(cc, index);
 895 }
 896
 897 static bool __cluster_may_compress(struct compress_ctx *cc)
 898 {
 899         struct f2fs_sb_info *sbi = F2FS_I_SB(cc->inode);
 900         loff_t i_size = i_size_read(cc->inode);
 901         unsigned nr_pages = DIV_ROUND_UP(i_size, PAGE_SIZE);
 902         int i;
 903
 904         for (i = 0; i < cc->cluster_size; i++) {
 905                 struct page *page = cc->rpages[i];
 906
 907                 f2fs_bug_on(sbi, !page);
 908
 909                 if (unlikely(f2fs_cp_error(sbi)))
 910                         return false;
 911                 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 912                         return false;
 913
 914                 /* beyond EOF */
 915                 if (page->index >= nr_pages)
 916                         return false;
 917         }
 918         return true;
 919 }
 920
 921 static int __f2fs_cluster_blocks(struct compress_ctx *cc, bool compr)
 922 {
 923         struct dnode_of_data dn;
 924         int ret;
 925
 926         set_new_dnode(&dn, cc->inode, NULL, NULL, 0);
 927         ret = f2fs_get_dnode_of_data(&dn, start_idx_of_cluster(cc),
 928                                                         LOOKUP_NODE);
 929         if (ret) {
 930                 if (ret == -ENOENT)
 931                         ret = 0;
 932                 goto fail;
 933         }
 934
 935         if (dn.data_blkaddr == COMPRESS_ADDR) {
 936                 int i;
 937
 938                 ret = 1;
 939                 for (i = 1; i < cc->cluster_size; i++) {
 940                         block_t blkaddr;
 941
 942                         blkaddr = data_blkaddr(dn.inode,
 943                                         dn.node_page, dn.ofs_in_node + i);
 944                         if (compr) {
 945                                 if (__is_valid_data_blkaddr(blkaddr))
 946                                         ret++;
 947                         } else {
 948                                 if (blkaddr != NULL_ADDR)
 949                                         ret++;
 950                         }
 951                 }
 952         }
 953 fail:
 954         f2fs_put_dnode(&dn);
 955         return ret;
 956 }
 957
 958 /* return # of compressed blocks in compressed cluster */
 959 static int f2fs_compressed_blocks(struct compress_ctx *cc)
 960 {
 961         return __f2fs_cluster_blocks(cc, true);
 962 }
 963
 964 /* return # of valid blocks in compressed cluster */
 965 static int f2fs_cluster_blocks(struct compress_ctx *cc)
 966 {
 967         return __f2fs_cluster_blocks(cc, false);
 968 }
 969
 970 int f2fs_is_compressed_cluster(struct inode *inode, pgoff_t index)
 971 {
 972         struct compress_ctx cc = {
 973                 .inode = inode,
 974                 .log_cluster_size = F2FS_I(inode)->i_log_cluster_size,
 975                 .cluster_size = F2FS_I(inode)->i_cluster_size,
 976                 .cluster_idx = index >> F2FS_I(inode)->i_log_cluster_size,
 977         };
 978
 979         return f2fs_cluster_blocks(&cc);
 980 }
 981
 982 static bool cluster_may_compress(struct compress_ctx *cc)
 983 {
 984         if (!f2fs_need_compress_data(cc->inode))
 985                 return false;
 986         if (f2fs_is_atomic_file(cc->inode))
 987                 return false;
 988         if (f2fs_is_mmap_file(cc->inode))
 989                 return false;
 990         if (!f2fs_cluster_is_full(cc))
 991                 return false;
 992         if (unlikely(f2fs_cp_error(F2FS_I_SB(cc->inode))))
 993                 return false;
 994         return __cluster_may_compress(cc);
 995 }
 996
 997 static void set_cluster_writeback(struct compress_ctx *cc)
 998 {
 999         int i;
1000
1001         for (i = 0; i < cc->cluster_size; i++) {
1002                 if (cc->rpages[i])
1003                         set_page_writeback(cc->rpages[i]);
1004         }
1005 }
1006
1007 static void set_cluster_dirty(struct compress_ctx *cc)
1008 {
1009         int i;
1010
1011         for (i = 0; i < cc->cluster_size; i++)
1012                 if (cc->rpages[i])
1013                         set_page_dirty(cc->rpages[i]);
1014 }
1015
1016 static int prepare_compress_overwrite(struct compress_ctx *cc,
1017                 struct page **pagep, pgoff_t index, void **fsdata)
1018 {
1019         struct f2fs_sb_info *sbi = F2FS_I_SB(cc->inode);
1020         struct address_space *mapping = cc->inode->i_mapping;
1021         struct page *page;
1022         struct dnode_of_data dn;
1023         sector_t last_block_in_bio;
1024         unsigned fgp_flag = FGP_LOCK | FGP_WRITE | FGP_CREAT;
1025         pgoff_t start_idx = start_idx_of_cluster(cc);
1026         int i, ret;
1027         bool prealloc;
1028
1029 retry:
1030         ret = f2fs_cluster_blocks(cc);
1031         if (ret <= 0)
1032                 return ret;
1033
1034         /* compressed case */
1035         prealloc = (ret < cc->cluster_size);
1036
1037         ret = f2fs_init_compress_ctx(cc);
1038         if (ret)
1039                 return ret;
1040
1041         /* keep page reference to avoid page reclaim */
1042         for (i = 0; i < cc->cluster_size; i++) {
1043                 page = f2fs_pagecache_get_page(mapping, start_idx + i,
1044                                                         fgp_flag, GFP_NOFS);
1045                 if (!page) {
1046                         ret = -ENOMEM;
1047                         goto unlock_pages;
1048                 }
1049
1050                 if (PageUptodate(page))
1051                         unlock_page(page);
1052                 else
1053                         f2fs_compress_ctx_add_page(cc, page);
1054         }
1055
1056         if (!f2fs_cluster_is_empty(cc)) {
1057                 struct bio *bio = NULL;
1058
1059                 ret = f2fs_read_multi_pages(cc, &bio, cc->cluster_size,
1060                                         &last_block_in_bio, false, true);
1061                 f2fs_destroy_compress_ctx(cc);
1062                 if (ret)
1063                         goto release_pages;
1064                 if (bio)
1065                         f2fs_submit_bio(sbi, bio, DATA);
1066
1067                 ret = f2fs_init_compress_ctx(cc);
1068                 if (ret)
1069                         goto release_pages;
1070         }
1071
1072         for (i = 0; i < cc->cluster_size; i++) {
1073                 f2fs_bug_on(sbi, cc->rpages[i]);
1074
1075                 page = find_lock_page(mapping, start_idx + i);
1076                 f2fs_bug_on(sbi, !page);
1077
1078                 f2fs_wait_on_page_writeback(page, DATA, true, true);
1079
1080                 f2fs_compress_ctx_add_page(cc, page);
1081                 f2fs_put_page(page, 0);
1082
1083                 if (!PageUptodate(page)) {
1084                         f2fs_unlock_rpages(cc, i + 1);
1085                         f2fs_put_rpages_mapping(mapping, start_idx,
1086                                         cc->cluster_size);
1087                         f2fs_destroy_compress_ctx(cc);
1088                         goto retry;
1089                 }
1090         }
1091
1092         if (prealloc) {
1093                 f2fs_do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO, true);
1094
1095                 set_new_dnode(&dn, cc->inode, NULL, NULL, 0);
1096
1097                 for (i = cc->cluster_size - 1; i > 0; i--) {
1098                         ret = f2fs_get_block(&dn, start_idx + i);
1099                         if (ret) {
1100                                 i = cc->cluster_size;
1101                                 break;
1102                         }
1103
1104                         if (dn.data_blkaddr != NEW_ADDR)
1105                                 break;
1106                 }
1107
1108                 f2fs_do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO, false);
1109         }
1110
1111         if (likely(!ret)) {
1112                 *fsdata = cc->rpages;
1113                 *pagep = cc->rpages[offset_in_cluster(cc, index)];
1114                 return cc->cluster_size;
1115         }
1116
1117 unlock_pages:
1118         f2fs_unlock_rpages(cc, i);
1119 release_pages:
1120         f2fs_put_rpages_mapping(mapping, start_idx, i);
1121         f2fs_destroy_compress_ctx(cc);
1122         return ret;
1123 }
1124
1125 int f2fs_prepare_compress_overwrite(struct inode *inode,
1126                 struct page **pagep, pgoff_t index, void **fsdata)
1127 {
1128         struct compress_ctx cc = {
1129                 .inode = inode,
1130                 .log_cluster_size = F2FS_I(inode)->i_log_cluster_size,
1131                 .cluster_size = F2FS_I(inode)->i_cluster_size,
1132                 .cluster_idx = index >> F2FS_I(inode)->i_log_cluster_size,
1133                 .rpages = NULL,
1134                 .nr_rpages = 0,
1135         };
1136
1137         return prepare_compress_overwrite(&cc, pagep, index, fsdata);
1138 }
1139
1140 bool f2fs_compress_write_end(struct inode *inode, void *fsdata,
1141                                         pgoff_t index, unsigned copied)
1142
1143 {
1144         struct compress_ctx cc = {
1145                 .inode = inode,
1146                 .log_cluster_size = F2FS_I(inode)->i_log_cluster_size,
1147                 .cluster_size = F2FS_I(inode)->i_cluster_size,
1148                 .rpages = fsdata,
1149         };
1150         bool first_index = (index == cc.rpages[0]->index);
1151
1152         if (copied)
1153                 set_cluster_dirty(&cc);
1154
1155         f2fs_put_rpages_wbc(&cc, NULL, false, 1);
1156         f2fs_destroy_compress_ctx(&cc);
1157
1158         return first_index;
1159 }
1160
1161 int f2fs_truncate_partial_cluster(struct inode *inode, u64 from, bool lock)
1162 {
1163         void *fsdata = NULL;
1164         struct page *pagep;
1165         int log_cluster_size = F2FS_I(inode)->i_log_cluster_size;
1166         pgoff_t start_idx = from >> (PAGE_SHIFT + log_cluster_size) <<
1167                                                         log_cluster_size;
1168         int err;
1169
1170         err = f2fs_is_compressed_cluster(inode, start_idx);
1171         if (err < 0)
1172                 return err;
1173
1174         /* truncate normal cluster */
1175         if (!err)
1176                 return f2fs_do_truncate_blocks(inode, from, lock);
1177
1178         /* truncate compressed cluster */
1179         err = f2fs_prepare_compress_overwrite(inode, &pagep,
1180                                                 start_idx, &fsdata);
1181
1182         /* should not be a normal cluster */
1183         f2fs_bug_on(F2FS_I_SB(inode), err == 0);
1184
1185         if (err <= 0)
1186                 return err;
1187
1188         if (err > 0) {
1189                 struct page **rpages = fsdata;
1190                 int cluster_size = F2FS_I(inode)->i_cluster_size;
1191                 int i;
1192
1193                 for (i = cluster_size - 1; i >= 0; i--) {
1194                         loff_t start = rpages[i]->index << PAGE_SHIFT;
1195
1196                         if (from <= start) {
1197                                 zero_user_segment(rpages[i], 0, PAGE_SIZE);
1198                         } else {
1199                                 zero_user_segment(rpages[i], from - start,
1200                                                                 PAGE_SIZE);
1201                                 break;
1202                         }
1203                 }
1204
1205                 f2fs_compress_write_end(inode, fsdata, start_idx, true);
1206         }
1207         return 0;
1208 }
1209
1210 static int f2fs_write_compressed_pages(struct compress_ctx *cc,
1211                                         int *submitted,
1212                                         struct writeback_control *wbc,
1213                                         enum iostat_type io_type)
1214 {
1215         struct inode *inode = cc->inode;
1216         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1217         struct f2fs_inode_info *fi = F2FS_I(inode);
1218         struct f2fs_io_info fio = {
1219                 .sbi = sbi,
1220                 .ino = cc->inode->i_ino,
1221                 .type = DATA,
1222                 .op = REQ_OP_WRITE,
1223                 .op_flags = wbc_to_write_flags(wbc),
1224                 .old_blkaddr = NEW_ADDR,
1225                 .page = NULL,
1226                 .encrypted_page = NULL,
1227                 .compressed_page = NULL,
1228                 .submitted = false,
1229                 .io_type = io_type,
1230                 .io_wbc = wbc,
1231                 .encrypted = fscrypt_inode_uses_fs_layer_crypto(cc->inode),
1232         };
1233         struct dnode_of_data dn;
1234         struct node_info ni;
1235         struct compress_io_ctx *cic;
1236         pgoff_t start_idx = start_idx_of_cluster(cc);
1237         unsigned int last_index = cc->cluster_size - 1;
1238         loff_t psize;
1239         int i, err;
1240
1241         if (IS_NOQUOTA(inode)) {
1242                 /*
1243                  * We need to wait for node_write to avoid block allocation during
1244                  * checkpoint. This can only happen to quota writes which can cause
1245                  * the below discard race condition.
1246                  */
1247                 down_read(&sbi->node_write);
1248         } else if (!f2fs_trylock_op(sbi)) {
1249                 goto out_free;
1250         }
1251
1252         set_new_dnode(&dn, cc->inode, NULL, NULL, 0);
1253
1254         err = f2fs_get_dnode_of_data(&dn, start_idx, LOOKUP_NODE);
1255         if (err)
1256                 goto out_unlock_op;
1257
1258         for (i = 0; i < cc->cluster_size; i++) {
1259                 if (data_blkaddr(dn.inode, dn.node_page,
1260                                         dn.ofs_in_node + i) == NULL_ADDR)
1261                         goto out_put_dnode;
1262         }
1263
1264         psize = (loff_t)(cc->rpages[last_index]->index + 1) << PAGE_SHIFT;
1265
1266         err = f2fs_get_node_info(fio.sbi, dn.nid, &ni);
1267         if (err)
1268                 goto out_put_dnode;
1269
1270         fio.version = ni.version;
1271
1272         cic = kmem_cache_zalloc(cic_entry_slab, GFP_NOFS);
1273         if (!cic)
1274                 goto out_put_dnode;
1275
1276         cic->magic = F2FS_COMPRESSED_PAGE_MAGIC;
1277         cic->inode = inode;
1278         atomic_set(&cic->pending_pages, cc->nr_cpages);
1279         cic->rpages = page_array_alloc(cc->inode, cc->cluster_size);
1280         if (!cic->rpages)
1281                 goto out_put_cic;
1282
1283         cic->nr_rpages = cc->cluster_size;
1284
1285         for (i = 0; i < cc->nr_cpages; i++) {
1286                 f2fs_set_compressed_page(cc->cpages[i], inode,
1287                                         cc->rpages[i + 1]->index, cic);
1288                 fio.compressed_page = cc->cpages[i];
1289
1290                 fio.old_blkaddr = data_blkaddr(dn.inode, dn.node_page,
1291                                                 dn.ofs_in_node + i + 1);
1292
1293                 /* wait for GCed page writeback via META_MAPPING */
1294                 f2fs_wait_on_block_writeback(inode, fio.old_blkaddr);
1295
1296                 if (fio.encrypted) {
1297                         fio.page = cc->rpages[i + 1];
1298                         err = f2fs_encrypt_one_page(&fio);
1299                         if (err)
1300                                 goto out_destroy_crypt;
1301                         cc->cpages[i] = fio.encrypted_page;
1302                 }
1303         }
1304
1305         set_cluster_writeback(cc);
1306
1307         for (i = 0; i < cc->cluster_size; i++)
1308                 cic->rpages[i] = cc->rpages[i];
1309
1310         for (i = 0; i < cc->cluster_size; i++, dn.ofs_in_node++) {
1311                 block_t blkaddr;
1312
1313                 blkaddr = f2fs_data_blkaddr(&dn);
1314                 fio.page = cc->rpages[i];
1315                 fio.old_blkaddr = blkaddr;
1316
1317                 /* cluster header */
1318                 if (i == 0) {
1319                         if (blkaddr == COMPRESS_ADDR)
1320                                 fio.compr_blocks++;
1321                         if (__is_valid_data_blkaddr(blkaddr))
1322                                 f2fs_invalidate_blocks(sbi, blkaddr);
1323                         f2fs_update_data_blkaddr(&dn, COMPRESS_ADDR);
1324                         goto unlock_continue;
1325                 }
1326
1327                 if (fio.compr_blocks && __is_valid_data_blkaddr(blkaddr))
1328                         fio.compr_blocks++;
1329
1330                 if (i > cc->nr_cpages) {
1331                         if (__is_valid_data_blkaddr(blkaddr)) {
1332                                 f2fs_invalidate_blocks(sbi, blkaddr);
1333                                 f2fs_update_data_blkaddr(&dn, NEW_ADDR);
1334                         }
1335                         goto unlock_continue;
1336                 }
1337
1338                 f2fs_bug_on(fio.sbi, blkaddr == NULL_ADDR);
1339
1340                 if (fio.encrypted)
1341                         fio.encrypted_page = cc->cpages[i - 1];
1342                 else
1343                         fio.compressed_page = cc->cpages[i - 1];
1344
1345                 cc->cpages[i - 1] = NULL;
1346                 f2fs_outplace_write_data(&dn, &fio);
1347                 (*submitted)++;
1348 unlock_continue:
1349                 inode_dec_dirty_pages(cc->inode);
1350                 unlock_page(fio.page);
1351         }
1352
1353         if (fio.compr_blocks)
1354                 f2fs_i_compr_blocks_update(inode, fio.compr_blocks - 1, false);
1355         f2fs_i_compr_blocks_update(inode, cc->nr_cpages, true);
1356
1357         set_inode_flag(cc->inode, FI_APPEND_WRITE);
1358         if (cc->cluster_idx == 0)
1359                 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
1360
1361         f2fs_put_dnode(&dn);
1362         if (IS_NOQUOTA(inode))
1363                 up_read(&sbi->node_write);
1364         else
1365                 f2fs_unlock_op(sbi);
1366
1367         spin_lock(&fi->i_size_lock);
1368         if (fi->last_disk_size < psize)
1369                 fi->last_disk_size = psize;
1370         spin_unlock(&fi->i_size_lock);
1371
1372         f2fs_put_rpages(cc);
1373         page_array_free(cc->inode, cc->cpages, cc->nr_cpages);
1374         cc->cpages = NULL;
1375         f2fs_destroy_compress_ctx(cc);
1376         return 0;
1377
1378 out_destroy_crypt:
1379         page_array_free(cc->inode, cic->rpages, cc->cluster_size);
1380
1381         for (--i; i >= 0; i--)
1382                 fscrypt_finalize_bounce_page(&cc->cpages[i]);
1383         for (i = 0; i < cc->nr_cpages; i++) {
1384                 if (!cc->cpages[i])
1385                         continue;
1386                 f2fs_put_page(cc->cpages[i], 1);
1387         }
1388 out_put_cic:
1389         kmem_cache_free(cic_entry_slab, cic);
1390 out_put_dnode:
1391         f2fs_put_dnode(&dn);
1392 out_unlock_op:
1393         if (IS_NOQUOTA(inode))
1394                 up_read(&sbi->node_write);
1395         else
1396                 f2fs_unlock_op(sbi);
1397 out_free:
1398         page_array_free(cc->inode, cc->cpages, cc->nr_cpages);
1399         cc->cpages = NULL;
1400         return -EAGAIN;
1401 }
1402
1403 void f2fs_compress_write_end_io(struct bio *bio, struct page *page)
1404 {
1405         struct f2fs_sb_info *sbi = bio->bi_private;
1406         struct compress_io_ctx *cic =
1407                         (struct compress_io_ctx *)page_private(page);
1408         int i;
1409
1410         if (unlikely(bio->bi_status))
1411                 mapping_set_error(cic->inode->i_mapping, -EIO);
1412
1413         f2fs_compress_free_page(page);
1414
1415         dec_page_count(sbi, F2FS_WB_DATA);
1416
1417         if (atomic_dec_return(&cic->pending_pages))
1418                 return;
1419
1420         for (i = 0; i < cic->nr_rpages; i++) {
1421                 WARN_ON(!cic->rpages[i]);
1422                 clear_cold_data(cic->rpages[i]);
1423                 end_page_writeback(cic->rpages[i]);
1424         }
1425
1426         page_array_free(cic->inode, cic->rpages, cic->nr_rpages);
1427         kmem_cache_free(cic_entry_slab, cic);
1428 }
1429
1430 static int f2fs_write_raw_pages(struct compress_ctx *cc,
1431                                         int *submitted,
1432                                         struct writeback_control *wbc,
1433                                         enum iostat_type io_type)
1434 {
1435         struct address_space *mapping = cc->inode->i_mapping;
1436         int _submitted, compr_blocks, ret;
1437         int i = -1, err = 0;
1438
1439         compr_blocks = f2fs_compressed_blocks(cc);
1440         if (compr_blocks < 0) {
1441                 err = compr_blocks;
1442                 goto out_err;
1443         }
1444
1445         for (i = 0; i < cc->cluster_size; i++) {
1446                 if (!cc->rpages[i])
1447                         continue;
1448 retry_write:
1449                 if (cc->rpages[i]->mapping != mapping) {
1450                         unlock_page(cc->rpages[i]);
1451                         continue;
1452                 }
1453
1454                 BUG_ON(!PageLocked(cc->rpages[i]));
1455
1456                 ret = f2fs_write_single_data_page(cc->rpages[i], &_submitted,
1457                                                 NULL, NULL, wbc, io_type,
1458                                                 compr_blocks, false);
1459                 if (ret) {
1460                         if (ret == AOP_WRITEPAGE_ACTIVATE) {
1461                                 unlock_page(cc->rpages[i]);
1462                                 ret = 0;
1463                         } else if (ret == -EAGAIN) {
1464                                 /*
1465                                  * for quota file, just redirty left pages to
1466                                  * avoid deadlock caused by cluster update race
1467                                  * from foreground operation.
1468                                  */
1469                                 if (IS_NOQUOTA(cc->inode)) {
1470                                         err = 0;
1471                                         goto out_err;
1472                                 }
1473                                 ret = 0;
1474                                 cond_resched();
1475                                 congestion_wait(BLK_RW_ASYNC,
1476                                                 DEFAULT_IO_TIMEOUT);
1477                                 lock_page(cc->rpages[i]);
1478
1479                                 if (!PageDirty(cc->rpages[i])) {
1480                                         unlock_page(cc->rpages[i]);
1481                                         continue;
1482                                 }
1483
1484                                 clear_page_dirty_for_io(cc->rpages[i]);
1485                                 goto retry_write;
1486                         }
1487                         err = ret;
1488                         goto out_err;
1489                 }
1490
1491                 *submitted += _submitted;
1492         }
1493
1494         f2fs_balance_fs(F2FS_M_SB(mapping), true);
1495
1496         return 0;
1497 out_err:
1498         for (++i; i < cc->cluster_size; i++) {
1499                 if (!cc->rpages[i])
1500                         continue;
1501                 redirty_page_for_writepage(wbc, cc->rpages[i]);
1502                 unlock_page(cc->rpages[i]);
1503         }
1504         return err;
1505 }
1506
1507 int f2fs_write_multi_pages(struct compress_ctx *cc,
1508                                         int *submitted,
1509                                         struct writeback_control *wbc,
1510                                         enum iostat_type io_type)
1511 {
1512         int err;
1513
1514         *submitted = 0;
1515         if (cluster_may_compress(cc)) {
1516                 err = f2fs_compress_pages(cc);
1517                 if (err == -EAGAIN) {
1518                         goto write;
1519                 } else if (err) {
1520                         f2fs_put_rpages_wbc(cc, wbc, true, 1);
1521                         goto destroy_out;
1522                 }
1523
1524                 err = f2fs_write_compressed_pages(cc, submitted,
1525                                                         wbc, io_type);
1526                 if (!err)
1527                         return 0;
1528                 f2fs_bug_on(F2FS_I_SB(cc->inode), err != -EAGAIN);
1529         }
1530 write:
1531         f2fs_bug_on(F2FS_I_SB(cc->inode), *submitted);
1532
1533         err = f2fs_write_raw_pages(cc, submitted, wbc, io_type);
1534         f2fs_put_rpages_wbc(cc, wbc, false, 0);
1535 destroy_out:
1536         f2fs_destroy_compress_ctx(cc);
1537         return err;
1538 }
1539
1540 static void f2fs_free_dic(struct decompress_io_ctx *dic);
1541
1542 struct decompress_io_ctx *f2fs_alloc_dic(struct compress_ctx *cc)
1543 {
1544         struct decompress_io_ctx *dic;
1545         pgoff_t start_idx = start_idx_of_cluster(cc);
1546         int i;
1547
1548         dic = kmem_cache_zalloc(dic_entry_slab, GFP_NOFS);
1549         if (!dic)
1550                 return ERR_PTR(-ENOMEM);
1551
1552         dic->rpages = page_array_alloc(cc->inode, cc->cluster_size);
1553         if (!dic->rpages) {
1554                 kmem_cache_free(dic_entry_slab, dic);
1555                 return ERR_PTR(-ENOMEM);
1556         }
1557
1558         dic->magic = F2FS_COMPRESSED_PAGE_MAGIC;
1559         dic->inode = cc->inode;
1560         atomic_set(&dic->remaining_pages, cc->nr_cpages);
1561         dic->cluster_idx = cc->cluster_idx;
1562         dic->cluster_size = cc->cluster_size;
1563         dic->log_cluster_size = cc->log_cluster_size;
1564         dic->nr_cpages = cc->nr_cpages;
1565         refcount_set(&dic->refcnt, 1);
1566         dic->failed = false;
1567         dic->need_verity = f2fs_need_verity(cc->inode, start_idx);
1568
1569         for (i = 0; i < dic->cluster_size; i++)
1570                 dic->rpages[i] = cc->rpages[i];
1571         dic->nr_rpages = cc->cluster_size;
1572
1573         dic->cpages = page_array_alloc(dic->inode, dic->nr_cpages);
1574         if (!dic->cpages)
1575                 goto out_free;
1576
1577         for (i = 0; i < dic->nr_cpages; i++) {
1578                 struct page *page;
1579
1580                 page = f2fs_compress_alloc_page();
1581                 if (!page)
1582                         goto out_free;
1583
1584                 f2fs_set_compressed_page(page, cc->inode,
1585                                         start_idx + i + 1, dic);
1586                 dic->cpages[i] = page;
1587         }
1588
1589         return dic;
1590
1591 out_free:
1592         f2fs_free_dic(dic);
1593         return ERR_PTR(-ENOMEM);
1594 }
1595
1596 static void f2fs_free_dic(struct decompress_io_ctx *dic)
1597 {
1598         int i;
1599
1600         if (dic->tpages) {
1601                 for (i = 0; i < dic->cluster_size; i++) {
1602                         if (dic->rpages[i])
1603                                 continue;
1604                         if (!dic->tpages[i])
1605                                 continue;
1606                         f2fs_compress_free_page(dic->tpages[i]);
1607                 }
1608                 page_array_free(dic->inode, dic->tpages, dic->cluster_size);
1609         }
1610
1611         if (dic->cpages) {
1612                 for (i = 0; i < dic->nr_cpages; i++) {
1613                         if (!dic->cpages[i])
1614                                 continue;
1615                         f2fs_compress_free_page(dic->cpages[i]);
1616                 }
1617                 page_array_free(dic->inode, dic->cpages, dic->nr_cpages);
1618         }
1619
1620         page_array_free(dic->inode, dic->rpages, dic->nr_rpages);
1621         kmem_cache_free(dic_entry_slab, dic);
1622 }
1623
1624 static void f2fs_put_dic(struct decompress_io_ctx *dic)
1625 {
1626         if (refcount_dec_and_test(&dic->refcnt))
1627                 f2fs_free_dic(dic);
1628 }
1629
1630 /*
1631  * Update and unlock the cluster's pagecache pages, and release the reference to
1632  * the decompress_io_ctx that was being held for I/O completion.
1633  */
1634 static void __f2fs_decompress_end_io(struct decompress_io_ctx *dic, bool failed)
1635 {
1636         int i;
1637
1638         for (i = 0; i < dic->cluster_size; i++) {
1639                 struct page *rpage = dic->rpages[i];
1640
1641                 if (!rpage)
1642                         continue;
1643
1644                 /* PG_error was set if verity failed. */
1645                 if (failed || PageError(rpage)) {
1646                         ClearPageUptodate(rpage);
1647                         /* will re-read again later */
1648                         ClearPageError(rpage);
1649                 } else {
1650                         SetPageUptodate(rpage);
1651                 }
1652                 unlock_page(rpage);
1653         }
1654
1655         f2fs_put_dic(dic);
1656 }
1657
1658 static void f2fs_verify_cluster(struct work_struct *work)
1659 {
1660         struct decompress_io_ctx *dic =
1661                 container_of(work, struct decompress_io_ctx, verity_work);
1662         int i;
1663
1664         /* Verify the cluster's decompressed pages with fs-verity. */
1665         for (i = 0; i < dic->cluster_size; i++) {
1666                 struct page *rpage = dic->rpages[i];
1667
1668                 if (rpage && !fsverity_verify_page(rpage))
1669                         SetPageError(rpage);
1670         }
1671
1672         __f2fs_decompress_end_io(dic, false);
1673 }
1674
1675 /*
1676  * This is called when a compressed cluster has been decompressed
1677  * (or failed to be read and/or decompressed).
1678  */
1679 void f2fs_decompress_end_io(struct decompress_io_ctx *dic, bool failed)
1680 {
1681         if (!failed && dic->need_verity) {
1682                 /*
1683                  * Note that to avoid deadlocks, the verity work can't be done
1684                  * on the decompression workqueue.  This is because verifying
1685                  * the data pages can involve reading metadata pages from the
1686                  * file, and these metadata pages may be compressed.
1687                  */
1688                 INIT_WORK(&dic->verity_work, f2fs_verify_cluster);
1689                 fsverity_enqueue_verify_work(&dic->verity_work);
1690         } else {
1691                 __f2fs_decompress_end_io(dic, failed);
1692         }
1693 }
1694
1695 /*
1696  * Put a reference to a compressed page's decompress_io_ctx.
1697  *
1698  * This is called when the page is no longer needed and can be freed.
1699  */
1700 void f2fs_put_page_dic(struct page *page)
1701 {
1702         struct decompress_io_ctx *dic =
1703                         (struct decompress_io_ctx *)page_private(page);
1704
1705         f2fs_put_dic(dic);
1706 }
1707
1708 int f2fs_init_page_array_cache(struct f2fs_sb_info *sbi)
1709 {
1710         dev_t dev = sbi->sb->s_bdev->bd_dev;
1711         char slab_name[32];
1712
1713         sprintf(slab_name, "f2fs_page_array_entry-%u:%u", MAJOR(dev), MINOR(dev));
1714
1715         sbi->page_array_slab_size = sizeof(struct page *) <<
1716                                         F2FS_OPTION(sbi).compress_log_size;
1717
1718         sbi->page_array_slab = f2fs_kmem_cache_create(slab_name,
1719                                         sbi->page_array_slab_size);
1720         if (!sbi->page_array_slab)
1721                 return -ENOMEM;
1722         return 0;
1723 }
1724
1725 void f2fs_destroy_page_array_cache(struct f2fs_sb_info *sbi)
1726 {
1727         kmem_cache_destroy(sbi->page_array_slab);
1728 }
1729
1730 static int __init f2fs_init_cic_cache(void)
1731 {
1732         cic_entry_slab = f2fs_kmem_cache_create("f2fs_cic_entry",
1733                                         sizeof(struct compress_io_ctx));
1734         if (!cic_entry_slab)
1735                 return -ENOMEM;
1736         return 0;
1737 }
1738
1739 static void f2fs_destroy_cic_cache(void)
1740 {
1741         kmem_cache_destroy(cic_entry_slab);
1742 }
1743
1744 static int __init f2fs_init_dic_cache(void)
1745 {
1746         dic_entry_slab = f2fs_kmem_cache_create("f2fs_dic_entry",
1747                                         sizeof(struct decompress_io_ctx));
1748         if (!dic_entry_slab)
1749                 return -ENOMEM;
1750         return 0;
1751 }
1752
1753 static void f2fs_destroy_dic_cache(void)
1754 {
1755         kmem_cache_destroy(dic_entry_slab);
1756 }
1757
1758 int __init f2fs_init_compress_cache(void)
1759 {
1760         int err;
1761
1762         err = f2fs_init_cic_cache();
1763         if (err)
1764                 goto out;
1765         err = f2fs_init_dic_cache();
1766         if (err)
1767                 goto free_cic;
1768         return 0;
1769 free_cic:
1770         f2fs_destroy_cic_cache();
1771 out:
1772         return -ENOMEM;
1773 }
1774
1775 void f2fs_destroy_compress_cache(void)
1776 {
1777         f2fs_destroy_dic_cache();
1778         f2fs_destroy_cic_cache();
1779 }