fs/f2fs/compress.c

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