linux/fs/f2fs/compress.c
<<
>>
Prefs
   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#include <linux/pagevec.h>
  16
  17#include "f2fs.h"
  18#include "node.h"
  19#include "segment.h"
  20#include <trace/events/f2fs.h>
  21
  22static struct kmem_cache *cic_entry_slab;
  23static struct kmem_cache *dic_entry_slab;
  24
  25static void *page_array_alloc(struct inode *inode, int nr)
  26{
  27        struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
  28        unsigned int size = sizeof(struct page *) * nr;
  29
  30        if (likely(size <= sbi->page_array_slab_size))
  31                return f2fs_kmem_cache_alloc(sbi->page_array_slab,
  32                                        GFP_F2FS_ZERO, false, F2FS_I_SB(inode));
  33        return f2fs_kzalloc(sbi, size, GFP_NOFS);
  34}
  35
  36static void page_array_free(struct inode *inode, void *pages, int nr)
  37{
  38        struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
  39        unsigned int size = sizeof(struct page *) * nr;
  40
  41        if (!pages)
  42                return;
  43
  44        if (likely(size <= sbi->page_array_slab_size))
  45                kmem_cache_free(sbi->page_array_slab, pages);
  46        else
  47                kfree(pages);
  48}
  49
  50struct f2fs_compress_ops {
  51        int (*init_compress_ctx)(struct compress_ctx *cc);
  52        void (*destroy_compress_ctx)(struct compress_ctx *cc);
  53        int (*compress_pages)(struct compress_ctx *cc);
  54        int (*init_decompress_ctx)(struct decompress_io_ctx *dic);
  55        void (*destroy_decompress_ctx)(struct decompress_io_ctx *dic);
  56        int (*decompress_pages)(struct decompress_io_ctx *dic);
  57};
  58
  59static unsigned int offset_in_cluster(struct compress_ctx *cc, pgoff_t index)
  60{
  61        return index & (cc->cluster_size - 1);
  62}
  63
  64static pgoff_t cluster_idx(struct compress_ctx *cc, pgoff_t index)
  65{
  66        return index >> cc->log_cluster_size;
  67}
  68
  69static pgoff_t start_idx_of_cluster(struct compress_ctx *cc)
  70{
  71        return cc->cluster_idx << cc->log_cluster_size;
  72}
  73
  74bool f2fs_is_compressed_page(struct page *page)
  75{
  76        if (!PagePrivate(page))
  77                return false;
  78        if (!page_private(page))
  79                return false;
  80        if (page_private_nonpointer(page))
  81                return false;
  82
  83        f2fs_bug_on(F2FS_M_SB(page->mapping),
  84                *((u32 *)page_private(page)) != F2FS_COMPRESSED_PAGE_MAGIC);
  85        return true;
  86}
  87
  88static void f2fs_set_compressed_page(struct page *page,
  89                struct inode *inode, pgoff_t index, void *data)
  90{
  91        attach_page_private(page, (void *)data);
  92
  93        /* i_crypto_info and iv index */
  94        page->index = index;
  95        page->mapping = inode->i_mapping;
  96}
  97
  98static void f2fs_drop_rpages(struct compress_ctx *cc, int len, bool unlock)
  99{
 100        int i;
 101
 102        for (i = 0; i < len; i++) {
 103                if (!cc->rpages[i])
 104                        continue;
 105                if (unlock)
 106                        unlock_page(cc->rpages[i]);
 107                else
 108                        put_page(cc->rpages[i]);
 109        }
 110}
 111
 112static void f2fs_put_rpages(struct compress_ctx *cc)
 113{
 114        f2fs_drop_rpages(cc, cc->cluster_size, false);
 115}
 116
 117static void f2fs_unlock_rpages(struct compress_ctx *cc, int len)
 118{
 119        f2fs_drop_rpages(cc, len, true);
 120}
 121
 122static void f2fs_put_rpages_wbc(struct compress_ctx *cc,
 123                struct writeback_control *wbc, bool redirty, int unlock)
 124{
 125        unsigned int i;
 126
 127        for (i = 0; i < cc->cluster_size; i++) {
 128                if (!cc->rpages[i])
 129                        continue;
 130                if (redirty)
 131                        redirty_page_for_writepage(wbc, cc->rpages[i]);
 132                f2fs_put_page(cc->rpages[i], unlock);
 133        }
 134}
 135
 136struct page *f2fs_compress_control_page(struct page *page)
 137{
 138        return ((struct compress_io_ctx *)page_private(page))->rpages[0];
 139}
 140
 141int f2fs_init_compress_ctx(struct compress_ctx *cc)
 142{
 143        if (cc->rpages)
 144                return 0;
 145
 146        cc->rpages = page_array_alloc(cc->inode, cc->cluster_size);
 147        return cc->rpages ? 0 : -ENOMEM;
 148}
 149
 150void f2fs_destroy_compress_ctx(struct compress_ctx *cc, bool reuse)
 151{
 152        page_array_free(cc->inode, cc->rpages, cc->cluster_size);
 153        cc->rpages = NULL;
 154        cc->nr_rpages = 0;
 155        cc->nr_cpages = 0;
 156        if (!reuse)
 157                cc->cluster_idx = NULL_CLUSTER;
 158}
 159
 160void f2fs_compress_ctx_add_page(struct compress_ctx *cc, struct page *page)
 161{
 162        unsigned int cluster_ofs;
 163
 164        if (!f2fs_cluster_can_merge_page(cc, page->index))
 165                f2fs_bug_on(F2FS_I_SB(cc->inode), 1);
 166
 167        cluster_ofs = offset_in_cluster(cc, page->index);
 168        cc->rpages[cluster_ofs] = page;
 169        cc->nr_rpages++;
 170        cc->cluster_idx = cluster_idx(cc, page->index);
 171}
 172
 173#ifdef CONFIG_F2FS_FS_LZO
 174static int lzo_init_compress_ctx(struct compress_ctx *cc)
 175{
 176        cc->private = f2fs_kvmalloc(F2FS_I_SB(cc->inode),
 177                                LZO1X_MEM_COMPRESS, GFP_NOFS);
 178        if (!cc->private)
 179                return -ENOMEM;
 180
 181        cc->clen = lzo1x_worst_compress(PAGE_SIZE << cc->log_cluster_size);
 182        return 0;
 183}
 184
 185static void lzo_destroy_compress_ctx(struct compress_ctx *cc)
 186{
 187        kvfree(cc->private);
 188        cc->private = NULL;
 189}
 190
 191static int lzo_compress_pages(struct compress_ctx *cc)
 192{
 193        int ret;
 194
 195        ret = lzo1x_1_compress(cc->rbuf, cc->rlen, cc->cbuf->cdata,
 196                                        &cc->clen, cc->private);
 197        if (ret != LZO_E_OK) {
 198                printk_ratelimited("%sF2FS-fs (%s): lzo compress failed, ret:%d\n",
 199                                KERN_ERR, F2FS_I_SB(cc->inode)->sb->s_id, ret);
 200                return -EIO;
 201        }
 202        return 0;
 203}
 204
 205static int lzo_decompress_pages(struct decompress_io_ctx *dic)
 206{
 207        int ret;
 208
 209        ret = lzo1x_decompress_safe(dic->cbuf->cdata, dic->clen,
 210                                                dic->rbuf, &dic->rlen);
 211        if (ret != LZO_E_OK) {
 212                printk_ratelimited("%sF2FS-fs (%s): lzo decompress failed, ret:%d\n",
 213                                KERN_ERR, F2FS_I_SB(dic->inode)->sb->s_id, ret);
 214                return -EIO;
 215        }
 216
 217        if (dic->rlen != PAGE_SIZE << dic->log_cluster_size) {
 218                printk_ratelimited("%sF2FS-fs (%s): lzo invalid rlen:%zu, "
 219                                        "expected:%lu\n", KERN_ERR,
 220                                        F2FS_I_SB(dic->inode)->sb->s_id,
 221                                        dic->rlen,
 222                                        PAGE_SIZE << dic->log_cluster_size);
 223                return -EIO;
 224        }
 225        return 0;
 226}
 227
 228static const struct f2fs_compress_ops f2fs_lzo_ops = {
 229        .init_compress_ctx      = lzo_init_compress_ctx,
 230        .destroy_compress_ctx   = lzo_destroy_compress_ctx,
 231        .compress_pages         = lzo_compress_pages,
 232        .decompress_pages       = lzo_decompress_pages,
 233};
 234#endif
 235
 236#ifdef CONFIG_F2FS_FS_LZ4
 237static int lz4_init_compress_ctx(struct compress_ctx *cc)
 238{
 239        unsigned int size = LZ4_MEM_COMPRESS;
 240
 241#ifdef CONFIG_F2FS_FS_LZ4HC
 242        if (F2FS_I(cc->inode)->i_compress_flag >> COMPRESS_LEVEL_OFFSET)
 243                size = LZ4HC_MEM_COMPRESS;
 244#endif
 245
 246        cc->private = f2fs_kvmalloc(F2FS_I_SB(cc->inode), size, GFP_NOFS);
 247        if (!cc->private)
 248                return -ENOMEM;
 249
 250        /*
 251         * we do not change cc->clen to LZ4_compressBound(inputsize) to
 252         * adapt worst compress case, because lz4 compressor can handle
 253         * output budget properly.
 254         */
 255        cc->clen = cc->rlen - PAGE_SIZE - COMPRESS_HEADER_SIZE;
 256        return 0;
 257}
 258
 259static void lz4_destroy_compress_ctx(struct compress_ctx *cc)
 260{
 261        kvfree(cc->private);
 262        cc->private = NULL;
 263}
 264
 265#ifdef CONFIG_F2FS_FS_LZ4HC
 266static int lz4hc_compress_pages(struct compress_ctx *cc)
 267{
 268        unsigned char level = F2FS_I(cc->inode)->i_compress_flag >>
 269                                                COMPRESS_LEVEL_OFFSET;
 270        int len;
 271
 272        if (level)
 273                len = LZ4_compress_HC(cc->rbuf, cc->cbuf->cdata, cc->rlen,
 274                                        cc->clen, level, cc->private);
 275        else
 276                len = LZ4_compress_default(cc->rbuf, cc->cbuf->cdata, cc->rlen,
 277                                                cc->clen, cc->private);
 278        if (!len)
 279                return -EAGAIN;
 280
 281        cc->clen = len;
 282        return 0;
 283}
 284#endif
 285
 286static int lz4_compress_pages(struct compress_ctx *cc)
 287{
 288        int len;
 289
 290#ifdef CONFIG_F2FS_FS_LZ4HC
 291        return lz4hc_compress_pages(cc);
 292#endif
 293        len = LZ4_compress_default(cc->rbuf, cc->cbuf->cdata, cc->rlen,
 294                                                cc->clen, cc->private);
 295        if (!len)
 296                return -EAGAIN;
 297
 298        cc->clen = len;
 299        return 0;
 300}
 301
 302static int lz4_decompress_pages(struct decompress_io_ctx *dic)
 303{
 304        int ret;
 305
 306        ret = LZ4_decompress_safe(dic->cbuf->cdata, dic->rbuf,
 307                                                dic->clen, dic->rlen);
 308        if (ret < 0) {
 309                printk_ratelimited("%sF2FS-fs (%s): lz4 decompress failed, ret:%d\n",
 310                                KERN_ERR, F2FS_I_SB(dic->inode)->sb->s_id, ret);
 311                return -EIO;
 312        }
 313
 314        if (ret != PAGE_SIZE << dic->log_cluster_size) {
 315                printk_ratelimited("%sF2FS-fs (%s): lz4 invalid rlen:%zu, "
 316                                        "expected:%lu\n", KERN_ERR,
 317                                        F2FS_I_SB(dic->inode)->sb->s_id,
 318                                        dic->rlen,
 319                                        PAGE_SIZE << dic->log_cluster_size);
 320                return -EIO;
 321        }
 322        return 0;
 323}
 324
 325static const struct f2fs_compress_ops f2fs_lz4_ops = {
 326        .init_compress_ctx      = lz4_init_compress_ctx,
 327        .destroy_compress_ctx   = lz4_destroy_compress_ctx,
 328        .compress_pages         = lz4_compress_pages,
 329        .decompress_pages       = lz4_decompress_pages,
 330};
 331#endif
 332
 333#ifdef CONFIG_F2FS_FS_ZSTD
 334#define F2FS_ZSTD_DEFAULT_CLEVEL        1
 335
 336static int zstd_init_compress_ctx(struct compress_ctx *cc)
 337{
 338        ZSTD_parameters params;
 339        ZSTD_CStream *stream;
 340        void *workspace;
 341        unsigned int workspace_size;
 342        unsigned char level = F2FS_I(cc->inode)->i_compress_flag >>
 343                                                COMPRESS_LEVEL_OFFSET;
 344
 345        if (!level)
 346                level = F2FS_ZSTD_DEFAULT_CLEVEL;
 347
 348        params = ZSTD_getParams(level, cc->rlen, 0);
 349        workspace_size = ZSTD_CStreamWorkspaceBound(params.cParams);
 350
 351        workspace = f2fs_kvmalloc(F2FS_I_SB(cc->inode),
 352                                        workspace_size, GFP_NOFS);
 353        if (!workspace)
 354                return -ENOMEM;
 355
 356        stream = ZSTD_initCStream(params, 0, workspace, workspace_size);
 357        if (!stream) {
 358                printk_ratelimited("%sF2FS-fs (%s): %s ZSTD_initCStream failed\n",
 359                                KERN_ERR, F2FS_I_SB(cc->inode)->sb->s_id,
 360                                __func__);
 361                kvfree(workspace);
 362                return -EIO;
 363        }
 364
 365        cc->private = workspace;
 366        cc->private2 = stream;
 367
 368        cc->clen = cc->rlen - PAGE_SIZE - COMPRESS_HEADER_SIZE;
 369        return 0;
 370}
 371
 372static void zstd_destroy_compress_ctx(struct compress_ctx *cc)
 373{
 374        kvfree(cc->private);
 375        cc->private = NULL;
 376        cc->private2 = NULL;
 377}
 378
 379static int zstd_compress_pages(struct compress_ctx *cc)
 380{
 381        ZSTD_CStream *stream = cc->private2;
 382        ZSTD_inBuffer inbuf;
 383        ZSTD_outBuffer outbuf;
 384        int src_size = cc->rlen;
 385        int dst_size = src_size - PAGE_SIZE - COMPRESS_HEADER_SIZE;
 386        int ret;
 387
 388        inbuf.pos = 0;
 389        inbuf.src = cc->rbuf;
 390        inbuf.size = src_size;
 391
 392        outbuf.pos = 0;
 393        outbuf.dst = cc->cbuf->cdata;
 394        outbuf.size = dst_size;
 395
 396        ret = ZSTD_compressStream(stream, &outbuf, &inbuf);
 397        if (ZSTD_isError(ret)) {
 398                printk_ratelimited("%sF2FS-fs (%s): %s ZSTD_compressStream failed, ret: %d\n",
 399                                KERN_ERR, F2FS_I_SB(cc->inode)->sb->s_id,
 400                                __func__, ZSTD_getErrorCode(ret));
 401                return -EIO;
 402        }
 403
 404        ret = ZSTD_endStream(stream, &outbuf);
 405        if (ZSTD_isError(ret)) {
 406                printk_ratelimited("%sF2FS-fs (%s): %s ZSTD_endStream returned %d\n",
 407                                KERN_ERR, F2FS_I_SB(cc->inode)->sb->s_id,
 408                                __func__, ZSTD_getErrorCode(ret));
 409                return -EIO;
 410        }
 411
 412        /*
 413         * there is compressed data remained in intermediate buffer due to
 414         * no more space in cbuf.cdata
 415         */
 416        if (ret)
 417                return -EAGAIN;
 418
 419        cc->clen = outbuf.pos;
 420        return 0;
 421}
 422
 423static int zstd_init_decompress_ctx(struct decompress_io_ctx *dic)
 424{
 425        ZSTD_DStream *stream;
 426        void *workspace;
 427        unsigned int workspace_size;
 428        unsigned int max_window_size =
 429                        MAX_COMPRESS_WINDOW_SIZE(dic->log_cluster_size);
 430
 431        workspace_size = ZSTD_DStreamWorkspaceBound(max_window_size);
 432
 433        workspace = f2fs_kvmalloc(F2FS_I_SB(dic->inode),
 434                                        workspace_size, GFP_NOFS);
 435        if (!workspace)
 436                return -ENOMEM;
 437
 438        stream = ZSTD_initDStream(max_window_size, workspace, workspace_size);
 439        if (!stream) {
 440                printk_ratelimited("%sF2FS-fs (%s): %s ZSTD_initDStream failed\n",
 441                                KERN_ERR, F2FS_I_SB(dic->inode)->sb->s_id,
 442                                __func__);
 443                kvfree(workspace);
 444                return -EIO;
 445        }
 446
 447        dic->private = workspace;
 448        dic->private2 = stream;
 449
 450        return 0;
 451}
 452
 453static void zstd_destroy_decompress_ctx(struct decompress_io_ctx *dic)
 454{
 455        kvfree(dic->private);
 456        dic->private = NULL;
 457        dic->private2 = NULL;
 458}
 459
 460static int zstd_decompress_pages(struct decompress_io_ctx *dic)
 461{
 462        ZSTD_DStream *stream = dic->private2;
 463        ZSTD_inBuffer inbuf;
 464        ZSTD_outBuffer outbuf;
 465        int ret;
 466
 467        inbuf.pos = 0;
 468        inbuf.src = dic->cbuf->cdata;
 469        inbuf.size = dic->clen;
 470
 471        outbuf.pos = 0;
 472        outbuf.dst = dic->rbuf;
 473        outbuf.size = dic->rlen;
 474
 475        ret = ZSTD_decompressStream(stream, &outbuf, &inbuf);
 476        if (ZSTD_isError(ret)) {
 477                printk_ratelimited("%sF2FS-fs (%s): %s ZSTD_compressStream failed, ret: %d\n",
 478                                KERN_ERR, F2FS_I_SB(dic->inode)->sb->s_id,
 479                                __func__, ZSTD_getErrorCode(ret));
 480                return -EIO;
 481        }
 482
 483        if (dic->rlen != outbuf.pos) {
 484                printk_ratelimited("%sF2FS-fs (%s): %s ZSTD invalid rlen:%zu, "
 485                                "expected:%lu\n", KERN_ERR,
 486                                F2FS_I_SB(dic->inode)->sb->s_id,
 487                                __func__, dic->rlen,
 488                                PAGE_SIZE << dic->log_cluster_size);
 489                return -EIO;
 490        }
 491
 492        return 0;
 493}
 494
 495static const struct f2fs_compress_ops f2fs_zstd_ops = {
 496        .init_compress_ctx      = zstd_init_compress_ctx,
 497        .destroy_compress_ctx   = zstd_destroy_compress_ctx,
 498        .compress_pages         = zstd_compress_pages,
 499        .init_decompress_ctx    = zstd_init_decompress_ctx,
 500        .destroy_decompress_ctx = zstd_destroy_decompress_ctx,
 501        .decompress_pages       = zstd_decompress_pages,
 502};
 503#endif
 504
 505#ifdef CONFIG_F2FS_FS_LZO
 506#ifdef CONFIG_F2FS_FS_LZORLE
 507static int lzorle_compress_pages(struct compress_ctx *cc)
 508{
 509        int ret;
 510
 511        ret = lzorle1x_1_compress(cc->rbuf, cc->rlen, cc->cbuf->cdata,
 512                                        &cc->clen, cc->private);
 513        if (ret != LZO_E_OK) {
 514                printk_ratelimited("%sF2FS-fs (%s): lzo-rle compress failed, ret:%d\n",
 515                                KERN_ERR, F2FS_I_SB(cc->inode)->sb->s_id, ret);
 516                return -EIO;
 517        }
 518        return 0;
 519}
 520
 521static const struct f2fs_compress_ops f2fs_lzorle_ops = {
 522        .init_compress_ctx      = lzo_init_compress_ctx,
 523        .destroy_compress_ctx   = lzo_destroy_compress_ctx,
 524        .compress_pages         = lzorle_compress_pages,
 525        .decompress_pages       = lzo_decompress_pages,
 526};
 527#endif
 528#endif
 529
 530static const struct f2fs_compress_ops *f2fs_cops[COMPRESS_MAX] = {
 531#ifdef CONFIG_F2FS_FS_LZO
 532        &f2fs_lzo_ops,
 533#else
 534        NULL,
 535#endif
 536#ifdef CONFIG_F2FS_FS_LZ4
 537        &f2fs_lz4_ops,
 538#else
 539        NULL,
 540#endif
 541#ifdef CONFIG_F2FS_FS_ZSTD
 542        &f2fs_zstd_ops,
 543#else
 544        NULL,
 545#endif
 546#if defined(CONFIG_F2FS_FS_LZO) && defined(CONFIG_F2FS_FS_LZORLE)
 547        &f2fs_lzorle_ops,
 548#else
 549        NULL,
 550#endif
 551};
 552
 553bool f2fs_is_compress_backend_ready(struct inode *inode)
 554{
 555        if (!f2fs_compressed_file(inode))
 556                return true;
 557        return f2fs_cops[F2FS_I(inode)->i_compress_algorithm];
 558}
 559
 560static mempool_t *compress_page_pool;
 561static int num_compress_pages = 512;
 562module_param(num_compress_pages, uint, 0444);
 563MODULE_PARM_DESC(num_compress_pages,
 564                "Number of intermediate compress pages to preallocate");
 565
 566int f2fs_init_compress_mempool(void)
 567{
 568        compress_page_pool = mempool_create_page_pool(num_compress_pages, 0);
 569        if (!compress_page_pool)
 570                return -ENOMEM;
 571
 572        return 0;
 573}
 574
 575void f2fs_destroy_compress_mempool(void)
 576{
 577        mempool_destroy(compress_page_pool);
 578}
 579
 580static struct page *f2fs_compress_alloc_page(void)
 581{
 582        struct page *page;
 583
 584        page = mempool_alloc(compress_page_pool, GFP_NOFS);
 585        lock_page(page);
 586
 587        return page;
 588}
 589
 590static void f2fs_compress_free_page(struct page *page)
 591{
 592        if (!page)
 593                return;
 594        detach_page_private(page);
 595        page->mapping = NULL;
 596        unlock_page(page);
 597        mempool_free(page, compress_page_pool);
 598}
 599
 600#define MAX_VMAP_RETRIES        3
 601
 602static void *f2fs_vmap(struct page **pages, unsigned int count)
 603{
 604        int i;
 605        void *buf = NULL;
 606
 607        for (i = 0; i < MAX_VMAP_RETRIES; i++) {
 608                buf = vm_map_ram(pages, count, -1);
 609                if (buf)
 610                        break;
 611                vm_unmap_aliases();
 612        }
 613        return buf;
 614}
 615
 616static int f2fs_compress_pages(struct compress_ctx *cc)
 617{
 618        struct f2fs_inode_info *fi = F2FS_I(cc->inode);
 619        const struct f2fs_compress_ops *cops =
 620                                f2fs_cops[fi->i_compress_algorithm];
 621        unsigned int max_len, new_nr_cpages;
 622        struct page **new_cpages;
 623        u32 chksum = 0;
 624        int i, ret;
 625
 626        trace_f2fs_compress_pages_start(cc->inode, cc->cluster_idx,
 627                                cc->cluster_size, fi->i_compress_algorithm);
 628
 629        if (cops->init_compress_ctx) {
 630                ret = cops->init_compress_ctx(cc);
 631                if (ret)
 632                        goto out;
 633        }
 634
 635        max_len = COMPRESS_HEADER_SIZE + cc->clen;
 636        cc->nr_cpages = DIV_ROUND_UP(max_len, PAGE_SIZE);
 637
 638        cc->cpages = page_array_alloc(cc->inode, cc->nr_cpages);
 639        if (!cc->cpages) {
 640                ret = -ENOMEM;
 641                goto destroy_compress_ctx;
 642        }
 643
 644        for (i = 0; i < cc->nr_cpages; i++) {
 645                cc->cpages[i] = f2fs_compress_alloc_page();
 646                if (!cc->cpages[i]) {
 647                        ret = -ENOMEM;
 648                        goto out_free_cpages;
 649                }
 650        }
 651
 652        cc->rbuf = f2fs_vmap(cc->rpages, cc->cluster_size);
 653        if (!cc->rbuf) {
 654                ret = -ENOMEM;
 655                goto out_free_cpages;
 656        }
 657
 658        cc->cbuf = f2fs_vmap(cc->cpages, cc->nr_cpages);
 659        if (!cc->cbuf) {
 660                ret = -ENOMEM;
 661                goto out_vunmap_rbuf;
 662        }
 663
 664        ret = cops->compress_pages(cc);
 665        if (ret)
 666                goto out_vunmap_cbuf;
 667
 668        max_len = PAGE_SIZE * (cc->cluster_size - 1) - COMPRESS_HEADER_SIZE;
 669
 670        if (cc->clen > max_len) {
 671                ret = -EAGAIN;
 672                goto out_vunmap_cbuf;
 673        }
 674
 675        cc->cbuf->clen = cpu_to_le32(cc->clen);
 676
 677        if (fi->i_compress_flag & 1 << COMPRESS_CHKSUM)
 678                chksum = f2fs_crc32(F2FS_I_SB(cc->inode),
 679                                        cc->cbuf->cdata, cc->clen);
 680        cc->cbuf->chksum = cpu_to_le32(chksum);
 681
 682        for (i = 0; i < COMPRESS_DATA_RESERVED_SIZE; i++)
 683                cc->cbuf->reserved[i] = cpu_to_le32(0);
 684
 685        new_nr_cpages = DIV_ROUND_UP(cc->clen + COMPRESS_HEADER_SIZE, PAGE_SIZE);
 686
 687        /* Now we're going to cut unnecessary tail pages */
 688        new_cpages = page_array_alloc(cc->inode, new_nr_cpages);
 689        if (!new_cpages) {
 690                ret = -ENOMEM;
 691                goto out_vunmap_cbuf;
 692        }
 693
 694        /* zero out any unused part of the last page */
 695        memset(&cc->cbuf->cdata[cc->clen], 0,
 696                        (new_nr_cpages * PAGE_SIZE) -
 697                        (cc->clen + COMPRESS_HEADER_SIZE));
 698
 699        vm_unmap_ram(cc->cbuf, cc->nr_cpages);
 700        vm_unmap_ram(cc->rbuf, cc->cluster_size);
 701
 702        for (i = 0; i < cc->nr_cpages; i++) {
 703                if (i < new_nr_cpages) {
 704                        new_cpages[i] = cc->cpages[i];
 705                        continue;
 706                }
 707                f2fs_compress_free_page(cc->cpages[i]);
 708                cc->cpages[i] = NULL;
 709        }
 710
 711        if (cops->destroy_compress_ctx)
 712                cops->destroy_compress_ctx(cc);
 713
 714        page_array_free(cc->inode, cc->cpages, cc->nr_cpages);
 715        cc->cpages = new_cpages;
 716        cc->nr_cpages = new_nr_cpages;
 717
 718        trace_f2fs_compress_pages_end(cc->inode, cc->cluster_idx,
 719                                                        cc->clen, ret);
 720        return 0;
 721
 722out_vunmap_cbuf:
 723        vm_unmap_ram(cc->cbuf, cc->nr_cpages);
 724out_vunmap_rbuf:
 725        vm_unmap_ram(cc->rbuf, cc->cluster_size);
 726out_free_cpages:
 727        for (i = 0; i < cc->nr_cpages; i++) {
 728                if (cc->cpages[i])
 729                        f2fs_compress_free_page(cc->cpages[i]);
 730        }
 731        page_array_free(cc->inode, cc->cpages, cc->nr_cpages);
 732        cc->cpages = NULL;
 733destroy_compress_ctx:
 734        if (cops->destroy_compress_ctx)
 735                cops->destroy_compress_ctx(cc);
 736out:
 737        trace_f2fs_compress_pages_end(cc->inode, cc->cluster_idx,
 738                                                        cc->clen, ret);
 739        return ret;
 740}
 741
 742void f2fs_decompress_cluster(struct decompress_io_ctx *dic)
 743{
 744        struct f2fs_sb_info *sbi = F2FS_I_SB(dic->inode);
 745        struct f2fs_inode_info *fi = F2FS_I(dic->inode);
 746        const struct f2fs_compress_ops *cops =
 747                        f2fs_cops[fi->i_compress_algorithm];
 748        int ret;
 749        int i;
 750
 751        trace_f2fs_decompress_pages_start(dic->inode, dic->cluster_idx,
 752                                dic->cluster_size, fi->i_compress_algorithm);
 753
 754        if (dic->failed) {
 755                ret = -EIO;
 756                goto out_end_io;
 757        }
 758
 759        dic->tpages = page_array_alloc(dic->inode, dic->cluster_size);
 760        if (!dic->tpages) {
 761                ret = -ENOMEM;
 762                goto out_end_io;
 763        }
 764
 765        for (i = 0; i < dic->cluster_size; i++) {
 766                if (dic->rpages[i]) {
 767                        dic->tpages[i] = dic->rpages[i];
 768                        continue;
 769                }
 770
 771                dic->tpages[i] = f2fs_compress_alloc_page();
 772                if (!dic->tpages[i]) {
 773                        ret = -ENOMEM;
 774                        goto out_end_io;
 775                }
 776        }
 777
 778        if (cops->init_decompress_ctx) {
 779                ret = cops->init_decompress_ctx(dic);
 780                if (ret)
 781                        goto out_end_io;
 782        }
 783
 784        dic->rbuf = f2fs_vmap(dic->tpages, dic->cluster_size);
 785        if (!dic->rbuf) {
 786                ret = -ENOMEM;
 787                goto out_destroy_decompress_ctx;
 788        }
 789
 790        dic->cbuf = f2fs_vmap(dic->cpages, dic->nr_cpages);
 791        if (!dic->cbuf) {
 792                ret = -ENOMEM;
 793                goto out_vunmap_rbuf;
 794        }
 795
 796        dic->clen = le32_to_cpu(dic->cbuf->clen);
 797        dic->rlen = PAGE_SIZE << dic->log_cluster_size;
 798
 799        if (dic->clen > PAGE_SIZE * dic->nr_cpages - COMPRESS_HEADER_SIZE) {
 800                ret = -EFSCORRUPTED;
 801                goto out_vunmap_cbuf;
 802        }
 803
 804        ret = cops->decompress_pages(dic);
 805
 806        if (!ret && (fi->i_compress_flag & 1 << COMPRESS_CHKSUM)) {
 807                u32 provided = le32_to_cpu(dic->cbuf->chksum);
 808                u32 calculated = f2fs_crc32(sbi, dic->cbuf->cdata, dic->clen);
 809
 810                if (provided != calculated) {
 811                        if (!is_inode_flag_set(dic->inode, FI_COMPRESS_CORRUPT)) {
 812                                set_inode_flag(dic->inode, FI_COMPRESS_CORRUPT);
 813                                printk_ratelimited(
 814                                        "%sF2FS-fs (%s): checksum invalid, nid = %lu, %x vs %x",
 815                                        KERN_INFO, sbi->sb->s_id, dic->inode->i_ino,
 816                                        provided, calculated);
 817                        }
 818                        set_sbi_flag(sbi, SBI_NEED_FSCK);
 819                }
 820        }
 821
 822out_vunmap_cbuf:
 823        vm_unmap_ram(dic->cbuf, dic->nr_cpages);
 824out_vunmap_rbuf:
 825        vm_unmap_ram(dic->rbuf, dic->cluster_size);
 826out_destroy_decompress_ctx:
 827        if (cops->destroy_decompress_ctx)
 828                cops->destroy_decompress_ctx(dic);
 829out_end_io:
 830        trace_f2fs_decompress_pages_end(dic->inode, dic->cluster_idx,
 831                                                        dic->clen, ret);
 832        f2fs_decompress_end_io(dic, ret);
 833}
 834
 835/*
 836 * This is called when a page of a compressed cluster has been read from disk
 837 * (or failed to be read from disk).  It checks whether this page was the last
 838 * page being waited on in the cluster, and if so, it decompresses the cluster
 839 * (or in the case of a failure, cleans up without actually decompressing).
 840 */
 841void f2fs_end_read_compressed_page(struct page *page, bool failed,
 842                                                block_t blkaddr)
 843{
 844        struct decompress_io_ctx *dic =
 845                        (struct decompress_io_ctx *)page_private(page);
 846        struct f2fs_sb_info *sbi = F2FS_I_SB(dic->inode);
 847
 848        dec_page_count(sbi, F2FS_RD_DATA);
 849
 850        if (failed)
 851                WRITE_ONCE(dic->failed, true);
 852        else if (blkaddr)
 853                f2fs_cache_compressed_page(sbi, page,
 854                                        dic->inode->i_ino, blkaddr);
 855
 856        if (atomic_dec_and_test(&dic->remaining_pages))
 857                f2fs_decompress_cluster(dic);
 858}
 859
 860static bool is_page_in_cluster(struct compress_ctx *cc, pgoff_t index)
 861{
 862        if (cc->cluster_idx == NULL_CLUSTER)
 863                return true;
 864        return cc->cluster_idx == cluster_idx(cc, index);
 865}
 866
 867bool f2fs_cluster_is_empty(struct compress_ctx *cc)
 868{
 869        return cc->nr_rpages == 0;
 870}
 871
 872static bool f2fs_cluster_is_full(struct compress_ctx *cc)
 873{
 874        return cc->cluster_size == cc->nr_rpages;
 875}
 876
 877bool f2fs_cluster_can_merge_page(struct compress_ctx *cc, pgoff_t index)
 878{
 879        if (f2fs_cluster_is_empty(cc))
 880                return true;
 881        return is_page_in_cluster(cc, index);
 882}
 883
 884static bool cluster_has_invalid_data(struct compress_ctx *cc)
 885{
 886        loff_t i_size = i_size_read(cc->inode);
 887        unsigned nr_pages = DIV_ROUND_UP(i_size, PAGE_SIZE);
 888        int i;
 889
 890        for (i = 0; i < cc->cluster_size; i++) {
 891                struct page *page = cc->rpages[i];
 892
 893                f2fs_bug_on(F2FS_I_SB(cc->inode), !page);
 894
 895                /* beyond EOF */
 896                if (page->index >= nr_pages)
 897                        return true;
 898        }
 899        return false;
 900}
 901
 902bool f2fs_sanity_check_cluster(struct dnode_of_data *dn)
 903{
 904        struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
 905        unsigned int cluster_size = F2FS_I(dn->inode)->i_cluster_size;
 906        bool compressed = dn->data_blkaddr == COMPRESS_ADDR;
 907        int cluster_end = 0;
 908        int i;
 909        char *reason = "";
 910
 911        if (!compressed)
 912                return false;
 913
 914        /* [..., COMPR_ADDR, ...] */
 915        if (dn->ofs_in_node % cluster_size) {
 916                reason = "[*|C|*|*]";
 917                goto out;
 918        }
 919
 920        for (i = 1; i < cluster_size; i++) {
 921                block_t blkaddr = data_blkaddr(dn->inode, dn->node_page,
 922                                                        dn->ofs_in_node + i);
 923
 924                /* [COMPR_ADDR, ..., COMPR_ADDR] */
 925                if (blkaddr == COMPRESS_ADDR) {
 926                        reason = "[C|*|C|*]";
 927                        goto out;
 928                }
 929                if (compressed) {
 930                        if (!__is_valid_data_blkaddr(blkaddr)) {
 931                                if (!cluster_end)
 932                                        cluster_end = i;
 933                                continue;
 934                        }
 935                        /* [COMPR_ADDR, NULL_ADDR or NEW_ADDR, valid_blkaddr] */
 936                        if (cluster_end) {
 937                                reason = "[C|N|N|V]";
 938                                goto out;
 939                        }
 940                }
 941        }
 942        return false;
 943out:
 944        f2fs_warn(sbi, "access invalid cluster, ino:%lu, nid:%u, ofs_in_node:%u, reason:%s",
 945                        dn->inode->i_ino, dn->nid, dn->ofs_in_node, reason);
 946        set_sbi_flag(sbi, SBI_NEED_FSCK);
 947        return true;
 948}
 949
 950static int __f2fs_cluster_blocks(struct inode *inode,
 951                                unsigned int cluster_idx, bool compr)
 952{
 953        struct dnode_of_data dn;
 954        unsigned int cluster_size = F2FS_I(inode)->i_cluster_size;
 955        unsigned int start_idx = cluster_idx <<
 956                                F2FS_I(inode)->i_log_cluster_size;
 957        int ret;
 958
 959        set_new_dnode(&dn, inode, NULL, NULL, 0);
 960        ret = f2fs_get_dnode_of_data(&dn, start_idx, LOOKUP_NODE);
 961        if (ret) {
 962                if (ret == -ENOENT)
 963                        ret = 0;
 964                goto fail;
 965        }
 966
 967        if (f2fs_sanity_check_cluster(&dn)) {
 968                ret = -EFSCORRUPTED;
 969                goto fail;
 970        }
 971
 972        if (dn.data_blkaddr == COMPRESS_ADDR) {
 973                int i;
 974
 975                ret = 1;
 976                for (i = 1; i < cluster_size; i++) {
 977                        block_t blkaddr;
 978
 979                        blkaddr = data_blkaddr(dn.inode,
 980                                        dn.node_page, dn.ofs_in_node + i);
 981                        if (compr) {
 982                                if (__is_valid_data_blkaddr(blkaddr))
 983                                        ret++;
 984                        } else {
 985                                if (blkaddr != NULL_ADDR)
 986                                        ret++;
 987                        }
 988                }
 989
 990                f2fs_bug_on(F2FS_I_SB(inode),
 991                        !compr && ret != cluster_size &&
 992                        !is_inode_flag_set(inode, FI_COMPRESS_RELEASED));
 993        }
 994fail:
 995        f2fs_put_dnode(&dn);
 996        return ret;
 997}
 998
 999/* return # of compressed blocks in compressed cluster */
1000static int f2fs_compressed_blocks(struct compress_ctx *cc)
1001{
1002        return __f2fs_cluster_blocks(cc->inode, cc->cluster_idx, true);
1003}
1004
1005/* return # of valid blocks in compressed cluster */
1006int f2fs_is_compressed_cluster(struct inode *inode, pgoff_t index)
1007{
1008        return __f2fs_cluster_blocks(inode,
1009                index >> F2FS_I(inode)->i_log_cluster_size,
1010                false);
1011}
1012
1013static bool cluster_may_compress(struct compress_ctx *cc)
1014{
1015        if (!f2fs_need_compress_data(cc->inode))
1016                return false;
1017        if (f2fs_is_atomic_file(cc->inode))
1018                return false;
1019        if (!f2fs_cluster_is_full(cc))
1020                return false;
1021        if (unlikely(f2fs_cp_error(F2FS_I_SB(cc->inode))))
1022                return false;
1023        return !cluster_has_invalid_data(cc);
1024}
1025
1026static void set_cluster_writeback(struct compress_ctx *cc)
1027{
1028        int i;
1029
1030        for (i = 0; i < cc->cluster_size; i++) {
1031                if (cc->rpages[i])
1032                        set_page_writeback(cc->rpages[i]);
1033        }
1034}
1035
1036static void set_cluster_dirty(struct compress_ctx *cc)
1037{
1038        int i;
1039
1040        for (i = 0; i < cc->cluster_size; i++)
1041                if (cc->rpages[i])
1042                        set_page_dirty(cc->rpages[i]);
1043}
1044
1045static int prepare_compress_overwrite(struct compress_ctx *cc,
1046                struct page **pagep, pgoff_t index, void **fsdata)
1047{
1048        struct f2fs_sb_info *sbi = F2FS_I_SB(cc->inode);
1049        struct address_space *mapping = cc->inode->i_mapping;
1050        struct page *page;
1051        sector_t last_block_in_bio;
1052        unsigned fgp_flag = FGP_LOCK | FGP_WRITE | FGP_CREAT;
1053        pgoff_t start_idx = start_idx_of_cluster(cc);
1054        int i, ret;
1055
1056retry:
1057        ret = f2fs_is_compressed_cluster(cc->inode, start_idx);
1058        if (ret <= 0)
1059                return ret;
1060
1061        ret = f2fs_init_compress_ctx(cc);
1062        if (ret)
1063                return ret;
1064
1065        /* keep page reference to avoid page reclaim */
1066        for (i = 0; i < cc->cluster_size; i++) {
1067                page = f2fs_pagecache_get_page(mapping, start_idx + i,
1068                                                        fgp_flag, GFP_NOFS);
1069                if (!page) {
1070                        ret = -ENOMEM;
1071                        goto unlock_pages;
1072                }
1073
1074                if (PageUptodate(page))
1075                        f2fs_put_page(page, 1);
1076                else
1077                        f2fs_compress_ctx_add_page(cc, page);
1078        }
1079
1080        if (!f2fs_cluster_is_empty(cc)) {
1081                struct bio *bio = NULL;
1082
1083                ret = f2fs_read_multi_pages(cc, &bio, cc->cluster_size,
1084                                        &last_block_in_bio, false, true);
1085                f2fs_put_rpages(cc);
1086                f2fs_destroy_compress_ctx(cc, true);
1087                if (ret)
1088                        goto out;
1089                if (bio)
1090                        f2fs_submit_bio(sbi, bio, DATA);
1091
1092                ret = f2fs_init_compress_ctx(cc);
1093                if (ret)
1094                        goto out;
1095        }
1096
1097        for (i = 0; i < cc->cluster_size; i++) {
1098                f2fs_bug_on(sbi, cc->rpages[i]);
1099
1100                page = find_lock_page(mapping, start_idx + i);
1101                if (!page) {
1102                        /* page can be truncated */
1103                        goto release_and_retry;
1104                }
1105
1106                f2fs_wait_on_page_writeback(page, DATA, true, true);
1107                f2fs_compress_ctx_add_page(cc, page);
1108
1109                if (!PageUptodate(page)) {
1110release_and_retry:
1111                        f2fs_put_rpages(cc);
1112                        f2fs_unlock_rpages(cc, i + 1);
1113                        f2fs_destroy_compress_ctx(cc, true);
1114                        goto retry;
1115                }
1116        }
1117
1118        if (likely(!ret)) {
1119                *fsdata = cc->rpages;
1120                *pagep = cc->rpages[offset_in_cluster(cc, index)];
1121                return cc->cluster_size;
1122        }
1123
1124unlock_pages:
1125        f2fs_put_rpages(cc);
1126        f2fs_unlock_rpages(cc, i);
1127        f2fs_destroy_compress_ctx(cc, true);
1128out:
1129        return ret;
1130}
1131
1132int f2fs_prepare_compress_overwrite(struct inode *inode,
1133                struct page **pagep, pgoff_t index, void **fsdata)
1134{
1135        struct compress_ctx cc = {
1136                .inode = inode,
1137                .log_cluster_size = F2FS_I(inode)->i_log_cluster_size,
1138                .cluster_size = F2FS_I(inode)->i_cluster_size,
1139                .cluster_idx = index >> F2FS_I(inode)->i_log_cluster_size,
1140                .rpages = NULL,
1141                .nr_rpages = 0,
1142        };
1143
1144        return prepare_compress_overwrite(&cc, pagep, index, fsdata);
1145}
1146
1147bool f2fs_compress_write_end(struct inode *inode, void *fsdata,
1148                                        pgoff_t index, unsigned copied)
1149
1150{
1151        struct compress_ctx cc = {
1152                .inode = inode,
1153                .log_cluster_size = F2FS_I(inode)->i_log_cluster_size,
1154                .cluster_size = F2FS_I(inode)->i_cluster_size,
1155                .rpages = fsdata,
1156        };
1157        bool first_index = (index == cc.rpages[0]->index);
1158
1159        if (copied)
1160                set_cluster_dirty(&cc);
1161
1162        f2fs_put_rpages_wbc(&cc, NULL, false, 1);
1163        f2fs_destroy_compress_ctx(&cc, false);
1164
1165        return first_index;
1166}
1167
1168int f2fs_truncate_partial_cluster(struct inode *inode, u64 from, bool lock)
1169{
1170        void *fsdata = NULL;
1171        struct page *pagep;
1172        int log_cluster_size = F2FS_I(inode)->i_log_cluster_size;
1173        pgoff_t start_idx = from >> (PAGE_SHIFT + log_cluster_size) <<
1174                                                        log_cluster_size;
1175        int err;
1176
1177        err = f2fs_is_compressed_cluster(inode, start_idx);
1178        if (err < 0)
1179                return err;
1180
1181        /* truncate normal cluster */
1182        if (!err)
1183                return f2fs_do_truncate_blocks(inode, from, lock);
1184
1185        /* truncate compressed cluster */
1186        err = f2fs_prepare_compress_overwrite(inode, &pagep,
1187                                                start_idx, &fsdata);
1188
1189        /* should not be a normal cluster */
1190        f2fs_bug_on(F2FS_I_SB(inode), err == 0);
1191
1192        if (err <= 0)
1193                return err;
1194
1195        if (err > 0) {
1196                struct page **rpages = fsdata;
1197                int cluster_size = F2FS_I(inode)->i_cluster_size;
1198                int i;
1199
1200                for (i = cluster_size - 1; i >= 0; i--) {
1201                        loff_t start = rpages[i]->index << PAGE_SHIFT;
1202
1203                        if (from <= start) {
1204                                zero_user_segment(rpages[i], 0, PAGE_SIZE);
1205                        } else {
1206                                zero_user_segment(rpages[i], from - start,
1207                                                                PAGE_SIZE);
1208                                break;
1209                        }
1210                }
1211
1212                f2fs_compress_write_end(inode, fsdata, start_idx, true);
1213        }
1214        return 0;
1215}
1216
1217static int f2fs_write_compressed_pages(struct compress_ctx *cc,
1218                                        int *submitted,
1219                                        struct writeback_control *wbc,
1220                                        enum iostat_type io_type)
1221{
1222        struct inode *inode = cc->inode;
1223        struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1224        struct f2fs_inode_info *fi = F2FS_I(inode);
1225        struct f2fs_io_info fio = {
1226                .sbi = sbi,
1227                .ino = cc->inode->i_ino,
1228                .type = DATA,
1229                .op = REQ_OP_WRITE,
1230                .op_flags = wbc_to_write_flags(wbc),
1231                .old_blkaddr = NEW_ADDR,
1232                .page = NULL,
1233                .encrypted_page = NULL,
1234                .compressed_page = NULL,
1235                .submitted = false,
1236                .io_type = io_type,
1237                .io_wbc = wbc,
1238                .encrypted = fscrypt_inode_uses_fs_layer_crypto(cc->inode),
1239        };
1240        struct dnode_of_data dn;
1241        struct node_info ni;
1242        struct compress_io_ctx *cic;
1243        pgoff_t start_idx = start_idx_of_cluster(cc);
1244        unsigned int last_index = cc->cluster_size - 1;
1245        loff_t psize;
1246        int i, err;
1247
1248        /* we should bypass data pages to proceed the kworkder jobs */
1249        if (unlikely(f2fs_cp_error(sbi))) {
1250                mapping_set_error(cc->rpages[0]->mapping, -EIO);
1251                goto out_free;
1252        }
1253
1254        if (IS_NOQUOTA(inode)) {
1255                /*
1256                 * We need to wait for node_write to avoid block allocation during
1257                 * checkpoint. This can only happen to quota writes which can cause
1258                 * the below discard race condition.
1259                 */
1260                down_read(&sbi->node_write);
1261        } else if (!f2fs_trylock_op(sbi)) {
1262                goto out_free;
1263        }
1264
1265        set_new_dnode(&dn, cc->inode, NULL, NULL, 0);
1266
1267        err = f2fs_get_dnode_of_data(&dn, start_idx, LOOKUP_NODE);
1268        if (err)
1269                goto out_unlock_op;
1270
1271        for (i = 0; i < cc->cluster_size; i++) {
1272                if (data_blkaddr(dn.inode, dn.node_page,
1273                                        dn.ofs_in_node + i) == NULL_ADDR)
1274                        goto out_put_dnode;
1275        }
1276
1277        psize = (loff_t)(cc->rpages[last_index]->index + 1) << PAGE_SHIFT;
1278
1279        err = f2fs_get_node_info(fio.sbi, dn.nid, &ni);
1280        if (err)
1281                goto out_put_dnode;
1282
1283        fio.version = ni.version;
1284
1285        cic = f2fs_kmem_cache_alloc(cic_entry_slab, GFP_F2FS_ZERO, false, sbi);
1286        if (!cic)
1287                goto out_put_dnode;
1288
1289        cic->magic = F2FS_COMPRESSED_PAGE_MAGIC;
1290        cic->inode = inode;
1291        atomic_set(&cic->pending_pages, cc->nr_cpages);
1292        cic->rpages = page_array_alloc(cc->inode, cc->cluster_size);
1293        if (!cic->rpages)
1294                goto out_put_cic;
1295
1296        cic->nr_rpages = cc->cluster_size;
1297
1298        for (i = 0; i < cc->nr_cpages; i++) {
1299                f2fs_set_compressed_page(cc->cpages[i], inode,
1300                                        cc->rpages[i + 1]->index, cic);
1301                fio.compressed_page = cc->cpages[i];
1302
1303                fio.old_blkaddr = data_blkaddr(dn.inode, dn.node_page,
1304                                                dn.ofs_in_node + i + 1);
1305
1306                /* wait for GCed page writeback via META_MAPPING */
1307                f2fs_wait_on_block_writeback(inode, fio.old_blkaddr);
1308
1309                if (fio.encrypted) {
1310                        fio.page = cc->rpages[i + 1];
1311                        err = f2fs_encrypt_one_page(&fio);
1312                        if (err)
1313                                goto out_destroy_crypt;
1314                        cc->cpages[i] = fio.encrypted_page;
1315                }
1316        }
1317
1318        set_cluster_writeback(cc);
1319
1320        for (i = 0; i < cc->cluster_size; i++)
1321                cic->rpages[i] = cc->rpages[i];
1322
1323        for (i = 0; i < cc->cluster_size; i++, dn.ofs_in_node++) {
1324                block_t blkaddr;
1325
1326                blkaddr = f2fs_data_blkaddr(&dn);
1327                fio.page = cc->rpages[i];
1328                fio.old_blkaddr = blkaddr;
1329
1330                /* cluster header */
1331                if (i == 0) {
1332                        if (blkaddr == COMPRESS_ADDR)
1333                                fio.compr_blocks++;
1334                        if (__is_valid_data_blkaddr(blkaddr))
1335                                f2fs_invalidate_blocks(sbi, blkaddr);
1336                        f2fs_update_data_blkaddr(&dn, COMPRESS_ADDR);
1337                        goto unlock_continue;
1338                }
1339
1340                if (fio.compr_blocks && __is_valid_data_blkaddr(blkaddr))
1341                        fio.compr_blocks++;
1342
1343                if (i > cc->nr_cpages) {
1344                        if (__is_valid_data_blkaddr(blkaddr)) {
1345                                f2fs_invalidate_blocks(sbi, blkaddr);
1346                                f2fs_update_data_blkaddr(&dn, NEW_ADDR);
1347                        }
1348                        goto unlock_continue;
1349                }
1350
1351                f2fs_bug_on(fio.sbi, blkaddr == NULL_ADDR);
1352
1353                if (fio.encrypted)
1354                        fio.encrypted_page = cc->cpages[i - 1];
1355                else
1356                        fio.compressed_page = cc->cpages[i - 1];
1357
1358                cc->cpages[i - 1] = NULL;
1359                f2fs_outplace_write_data(&dn, &fio);
1360                (*submitted)++;
1361unlock_continue:
1362                inode_dec_dirty_pages(cc->inode);
1363                unlock_page(fio.page);
1364        }
1365
1366        if (fio.compr_blocks)
1367                f2fs_i_compr_blocks_update(inode, fio.compr_blocks - 1, false);
1368        f2fs_i_compr_blocks_update(inode, cc->nr_cpages, true);
1369        add_compr_block_stat(inode, cc->nr_cpages);
1370
1371        set_inode_flag(cc->inode, FI_APPEND_WRITE);
1372        if (cc->cluster_idx == 0)
1373                set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
1374
1375        f2fs_put_dnode(&dn);
1376        if (IS_NOQUOTA(inode))
1377                up_read(&sbi->node_write);
1378        else
1379                f2fs_unlock_op(sbi);
1380
1381        spin_lock(&fi->i_size_lock);
1382        if (fi->last_disk_size < psize)
1383                fi->last_disk_size = psize;
1384        spin_unlock(&fi->i_size_lock);
1385
1386        f2fs_put_rpages(cc);
1387        page_array_free(cc->inode, cc->cpages, cc->nr_cpages);
1388        cc->cpages = NULL;
1389        f2fs_destroy_compress_ctx(cc, false);
1390        return 0;
1391
1392out_destroy_crypt:
1393        page_array_free(cc->inode, cic->rpages, cc->cluster_size);
1394
1395        for (--i; i >= 0; i--)
1396                fscrypt_finalize_bounce_page(&cc->cpages[i]);
1397out_put_cic:
1398        kmem_cache_free(cic_entry_slab, cic);
1399out_put_dnode:
1400        f2fs_put_dnode(&dn);
1401out_unlock_op:
1402        if (IS_NOQUOTA(inode))
1403                up_read(&sbi->node_write);
1404        else
1405                f2fs_unlock_op(sbi);
1406out_free:
1407        for (i = 0; i < cc->nr_cpages; i++) {
1408                if (!cc->cpages[i])
1409                        continue;
1410                f2fs_compress_free_page(cc->cpages[i]);
1411                cc->cpages[i] = NULL;
1412        }
1413        page_array_free(cc->inode, cc->cpages, cc->nr_cpages);
1414        cc->cpages = NULL;
1415        return -EAGAIN;
1416}
1417
1418void f2fs_compress_write_end_io(struct bio *bio, struct page *page)
1419{
1420        struct f2fs_sb_info *sbi = bio->bi_private;
1421        struct compress_io_ctx *cic =
1422                        (struct compress_io_ctx *)page_private(page);
1423        int i;
1424
1425        if (unlikely(bio->bi_status))
1426                mapping_set_error(cic->inode->i_mapping, -EIO);
1427
1428        f2fs_compress_free_page(page);
1429
1430        dec_page_count(sbi, F2FS_WB_DATA);
1431
1432        if (atomic_dec_return(&cic->pending_pages))
1433                return;
1434
1435        for (i = 0; i < cic->nr_rpages; i++) {
1436                WARN_ON(!cic->rpages[i]);
1437                clear_page_private_gcing(cic->rpages[i]);
1438                end_page_writeback(cic->rpages[i]);
1439        }
1440
1441        page_array_free(cic->inode, cic->rpages, cic->nr_rpages);
1442        kmem_cache_free(cic_entry_slab, cic);
1443}
1444
1445static int f2fs_write_raw_pages(struct compress_ctx *cc,
1446                                        int *submitted,
1447                                        struct writeback_control *wbc,
1448                                        enum iostat_type io_type)
1449{
1450        struct address_space *mapping = cc->inode->i_mapping;
1451        int _submitted, compr_blocks, ret;
1452        int i = -1, err = 0;
1453
1454        compr_blocks = f2fs_compressed_blocks(cc);
1455        if (compr_blocks < 0) {
1456                err = compr_blocks;
1457                goto out_err;
1458        }
1459
1460        for (i = 0; i < cc->cluster_size; i++) {
1461                if (!cc->rpages[i])
1462                        continue;
1463retry_write:
1464                if (cc->rpages[i]->mapping != mapping) {
1465                        unlock_page(cc->rpages[i]);
1466                        continue;
1467                }
1468
1469                BUG_ON(!PageLocked(cc->rpages[i]));
1470
1471                ret = f2fs_write_single_data_page(cc->rpages[i], &_submitted,
1472                                                NULL, NULL, wbc, io_type,
1473                                                compr_blocks, false);
1474                if (ret) {
1475                        if (ret == AOP_WRITEPAGE_ACTIVATE) {
1476                                unlock_page(cc->rpages[i]);
1477                                ret = 0;
1478                        } else if (ret == -EAGAIN) {
1479                                /*
1480                                 * for quota file, just redirty left pages to
1481                                 * avoid deadlock caused by cluster update race
1482                                 * from foreground operation.
1483                                 */
1484                                if (IS_NOQUOTA(cc->inode)) {
1485                                        err = 0;
1486                                        goto out_err;
1487                                }
1488                                ret = 0;
1489                                cond_resched();
1490                                congestion_wait(BLK_RW_ASYNC,
1491                                                DEFAULT_IO_TIMEOUT);
1492                                lock_page(cc->rpages[i]);
1493
1494                                if (!PageDirty(cc->rpages[i])) {
1495                                        unlock_page(cc->rpages[i]);
1496                                        continue;
1497                                }
1498
1499                                clear_page_dirty_for_io(cc->rpages[i]);
1500                                goto retry_write;
1501                        }
1502                        err = ret;
1503                        goto out_err;
1504                }
1505
1506                *submitted += _submitted;
1507        }
1508
1509        f2fs_balance_fs(F2FS_M_SB(mapping), true);
1510
1511        return 0;
1512out_err:
1513        for (++i; i < cc->cluster_size; i++) {
1514                if (!cc->rpages[i])
1515                        continue;
1516                redirty_page_for_writepage(wbc, cc->rpages[i]);
1517                unlock_page(cc->rpages[i]);
1518        }
1519        return err;
1520}
1521
1522int f2fs_write_multi_pages(struct compress_ctx *cc,
1523                                        int *submitted,
1524                                        struct writeback_control *wbc,
1525                                        enum iostat_type io_type)
1526{
1527        int err;
1528
1529        *submitted = 0;
1530        if (cluster_may_compress(cc)) {
1531                err = f2fs_compress_pages(cc);
1532                if (err == -EAGAIN) {
1533                        goto write;
1534                } else if (err) {
1535                        f2fs_put_rpages_wbc(cc, wbc, true, 1);
1536                        goto destroy_out;
1537                }
1538
1539                err = f2fs_write_compressed_pages(cc, submitted,
1540                                                        wbc, io_type);
1541                if (!err)
1542                        return 0;
1543                f2fs_bug_on(F2FS_I_SB(cc->inode), err != -EAGAIN);
1544        }
1545write:
1546        f2fs_bug_on(F2FS_I_SB(cc->inode), *submitted);
1547
1548        err = f2fs_write_raw_pages(cc, submitted, wbc, io_type);
1549        f2fs_put_rpages_wbc(cc, wbc, false, 0);
1550destroy_out:
1551        f2fs_destroy_compress_ctx(cc, false);
1552        return err;
1553}
1554
1555static void f2fs_free_dic(struct decompress_io_ctx *dic);
1556
1557struct decompress_io_ctx *f2fs_alloc_dic(struct compress_ctx *cc)
1558{
1559        struct decompress_io_ctx *dic;
1560        pgoff_t start_idx = start_idx_of_cluster(cc);
1561        int i;
1562
1563        dic = f2fs_kmem_cache_alloc(dic_entry_slab, GFP_F2FS_ZERO,
1564                                        false, F2FS_I_SB(cc->inode));
1565        if (!dic)
1566                return ERR_PTR(-ENOMEM);
1567
1568        dic->rpages = page_array_alloc(cc->inode, cc->cluster_size);
1569        if (!dic->rpages) {
1570                kmem_cache_free(dic_entry_slab, dic);
1571                return ERR_PTR(-ENOMEM);
1572        }
1573
1574        dic->magic = F2FS_COMPRESSED_PAGE_MAGIC;
1575        dic->inode = cc->inode;
1576        atomic_set(&dic->remaining_pages, cc->nr_cpages);
1577        dic->cluster_idx = cc->cluster_idx;
1578        dic->cluster_size = cc->cluster_size;
1579        dic->log_cluster_size = cc->log_cluster_size;
1580        dic->nr_cpages = cc->nr_cpages;
1581        refcount_set(&dic->refcnt, 1);
1582        dic->failed = false;
1583        dic->need_verity = f2fs_need_verity(cc->inode, start_idx);
1584
1585        for (i = 0; i < dic->cluster_size; i++)
1586                dic->rpages[i] = cc->rpages[i];
1587        dic->nr_rpages = cc->cluster_size;
1588
1589        dic->cpages = page_array_alloc(dic->inode, dic->nr_cpages);
1590        if (!dic->cpages)
1591                goto out_free;
1592
1593        for (i = 0; i < dic->nr_cpages; i++) {
1594                struct page *page;
1595
1596                page = f2fs_compress_alloc_page();
1597                if (!page)
1598                        goto out_free;
1599
1600                f2fs_set_compressed_page(page, cc->inode,
1601                                        start_idx + i + 1, dic);
1602                dic->cpages[i] = page;
1603        }
1604
1605        return dic;
1606
1607out_free:
1608        f2fs_free_dic(dic);
1609        return ERR_PTR(-ENOMEM);
1610}
1611
1612static void f2fs_free_dic(struct decompress_io_ctx *dic)
1613{
1614        int i;
1615
1616        if (dic->tpages) {
1617                for (i = 0; i < dic->cluster_size; i++) {
1618                        if (dic->rpages[i])
1619                                continue;
1620                        if (!dic->tpages[i])
1621                                continue;
1622                        f2fs_compress_free_page(dic->tpages[i]);
1623                }
1624                page_array_free(dic->inode, dic->tpages, dic->cluster_size);
1625        }
1626
1627        if (dic->cpages) {
1628                for (i = 0; i < dic->nr_cpages; i++) {
1629                        if (!dic->cpages[i])
1630                                continue;
1631                        f2fs_compress_free_page(dic->cpages[i]);
1632                }
1633                page_array_free(dic->inode, dic->cpages, dic->nr_cpages);
1634        }
1635
1636        page_array_free(dic->inode, dic->rpages, dic->nr_rpages);
1637        kmem_cache_free(dic_entry_slab, dic);
1638}
1639
1640static void f2fs_put_dic(struct decompress_io_ctx *dic)
1641{
1642        if (refcount_dec_and_test(&dic->refcnt))
1643                f2fs_free_dic(dic);
1644}
1645
1646/*
1647 * Update and unlock the cluster's pagecache pages, and release the reference to
1648 * the decompress_io_ctx that was being held for I/O completion.
1649 */
1650static void __f2fs_decompress_end_io(struct decompress_io_ctx *dic, bool failed)
1651{
1652        int i;
1653
1654        for (i = 0; i < dic->cluster_size; i++) {
1655                struct page *rpage = dic->rpages[i];
1656
1657                if (!rpage)
1658                        continue;
1659
1660                /* PG_error was set if verity failed. */
1661                if (failed || PageError(rpage)) {
1662                        ClearPageUptodate(rpage);
1663                        /* will re-read again later */
1664                        ClearPageError(rpage);
1665                } else {
1666                        SetPageUptodate(rpage);
1667                }
1668                unlock_page(rpage);
1669        }
1670
1671        f2fs_put_dic(dic);
1672}
1673
1674static void f2fs_verify_cluster(struct work_struct *work)
1675{
1676        struct decompress_io_ctx *dic =
1677                container_of(work, struct decompress_io_ctx, verity_work);
1678        int i;
1679
1680        /* Verify the cluster's decompressed pages with fs-verity. */
1681        for (i = 0; i < dic->cluster_size; i++) {
1682                struct page *rpage = dic->rpages[i];
1683
1684                if (rpage && !fsverity_verify_page(rpage))
1685                        SetPageError(rpage);
1686        }
1687
1688        __f2fs_decompress_end_io(dic, false);
1689}
1690
1691/*
1692 * This is called when a compressed cluster has been decompressed
1693 * (or failed to be read and/or decompressed).
1694 */
1695void f2fs_decompress_end_io(struct decompress_io_ctx *dic, bool failed)
1696{
1697        if (!failed && dic->need_verity) {
1698                /*
1699                 * Note that to avoid deadlocks, the verity work can't be done
1700                 * on the decompression workqueue.  This is because verifying
1701                 * the data pages can involve reading metadata pages from the
1702                 * file, and these metadata pages may be compressed.
1703                 */
1704                INIT_WORK(&dic->verity_work, f2fs_verify_cluster);
1705                fsverity_enqueue_verify_work(&dic->verity_work);
1706        } else {
1707                __f2fs_decompress_end_io(dic, failed);
1708        }
1709}
1710
1711/*
1712 * Put a reference to a compressed page's decompress_io_ctx.
1713 *
1714 * This is called when the page is no longer needed and can be freed.
1715 */
1716void f2fs_put_page_dic(struct page *page)
1717{
1718        struct decompress_io_ctx *dic =
1719                        (struct decompress_io_ctx *)page_private(page);
1720
1721        f2fs_put_dic(dic);
1722}
1723
1724/*
1725 * check whether cluster blocks are contiguous, and add extent cache entry
1726 * only if cluster blocks are logically and physically contiguous.
1727 */
1728unsigned int f2fs_cluster_blocks_are_contiguous(struct dnode_of_data *dn)
1729{
1730        bool compressed = f2fs_data_blkaddr(dn) == COMPRESS_ADDR;
1731        int i = compressed ? 1 : 0;
1732        block_t first_blkaddr = data_blkaddr(dn->inode, dn->node_page,
1733                                                dn->ofs_in_node + i);
1734
1735        for (i += 1; i < F2FS_I(dn->inode)->i_cluster_size; i++) {
1736                block_t blkaddr = data_blkaddr(dn->inode, dn->node_page,
1737                                                dn->ofs_in_node + i);
1738
1739                if (!__is_valid_data_blkaddr(blkaddr))
1740                        break;
1741                if (first_blkaddr + i - (compressed ? 1 : 0) != blkaddr)
1742                        return 0;
1743        }
1744
1745        return compressed ? i - 1 : i;
1746}
1747
1748const struct address_space_operations f2fs_compress_aops = {
1749        .releasepage = f2fs_release_page,
1750        .invalidatepage = f2fs_invalidate_page,
1751};
1752
1753struct address_space *COMPRESS_MAPPING(struct f2fs_sb_info *sbi)
1754{
1755        return sbi->compress_inode->i_mapping;
1756}
1757
1758void f2fs_invalidate_compress_page(struct f2fs_sb_info *sbi, block_t blkaddr)
1759{
1760        if (!sbi->compress_inode)
1761                return;
1762        invalidate_mapping_pages(COMPRESS_MAPPING(sbi), blkaddr, blkaddr);
1763}
1764
1765void f2fs_cache_compressed_page(struct f2fs_sb_info *sbi, struct page *page,
1766                                                nid_t ino, block_t blkaddr)
1767{
1768        struct page *cpage;
1769        int ret;
1770
1771        if (!test_opt(sbi, COMPRESS_CACHE))
1772                return;
1773
1774        if (!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE_READ))
1775                return;
1776
1777        if (!f2fs_available_free_memory(sbi, COMPRESS_PAGE))
1778                return;
1779
1780        cpage = find_get_page(COMPRESS_MAPPING(sbi), blkaddr);
1781        if (cpage) {
1782                f2fs_put_page(cpage, 0);
1783                return;
1784        }
1785
1786        cpage = alloc_page(__GFP_NOWARN | __GFP_IO);
1787        if (!cpage)
1788                return;
1789
1790        ret = add_to_page_cache_lru(cpage, COMPRESS_MAPPING(sbi),
1791                                                blkaddr, GFP_NOFS);
1792        if (ret) {
1793                f2fs_put_page(cpage, 0);
1794                return;
1795        }
1796
1797        set_page_private_data(cpage, ino);
1798
1799        if (!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE_READ))
1800                goto out;
1801
1802        memcpy(page_address(cpage), page_address(page), PAGE_SIZE);
1803        SetPageUptodate(cpage);
1804out:
1805        f2fs_put_page(cpage, 1);
1806}
1807
1808bool f2fs_load_compressed_page(struct f2fs_sb_info *sbi, struct page *page,
1809                                                                block_t blkaddr)
1810{
1811        struct page *cpage;
1812        bool hitted = false;
1813
1814        if (!test_opt(sbi, COMPRESS_CACHE))
1815                return false;
1816
1817        cpage = f2fs_pagecache_get_page(COMPRESS_MAPPING(sbi),
1818                                blkaddr, FGP_LOCK | FGP_NOWAIT, GFP_NOFS);
1819        if (cpage) {
1820                if (PageUptodate(cpage)) {
1821                        atomic_inc(&sbi->compress_page_hit);
1822                        memcpy(page_address(page),
1823                                page_address(cpage), PAGE_SIZE);
1824                        hitted = true;
1825                }
1826                f2fs_put_page(cpage, 1);
1827        }
1828
1829        return hitted;
1830}
1831
1832void f2fs_invalidate_compress_pages(struct f2fs_sb_info *sbi, nid_t ino)
1833{
1834        struct address_space *mapping = sbi->compress_inode->i_mapping;
1835        struct pagevec pvec;
1836        pgoff_t index = 0;
1837        pgoff_t end = MAX_BLKADDR(sbi);
1838
1839        if (!mapping->nrpages)
1840                return;
1841
1842        pagevec_init(&pvec);
1843
1844        do {
1845                unsigned int nr_pages;
1846                int i;
1847
1848                nr_pages = pagevec_lookup_range(&pvec, mapping,
1849                                                &index, end - 1);
1850                if (!nr_pages)
1851                        break;
1852
1853                for (i = 0; i < nr_pages; i++) {
1854                        struct page *page = pvec.pages[i];
1855
1856                        if (page->index > end)
1857                                break;
1858
1859                        lock_page(page);
1860                        if (page->mapping != mapping) {
1861                                unlock_page(page);
1862                                continue;
1863                        }
1864
1865                        if (ino != get_page_private_data(page)) {
1866                                unlock_page(page);
1867                                continue;
1868                        }
1869
1870                        generic_error_remove_page(mapping, page);
1871                        unlock_page(page);
1872                }
1873                pagevec_release(&pvec);
1874                cond_resched();
1875        } while (index < end);
1876}
1877
1878int f2fs_init_compress_inode(struct f2fs_sb_info *sbi)
1879{
1880        struct inode *inode;
1881
1882        if (!test_opt(sbi, COMPRESS_CACHE))
1883                return 0;
1884
1885        inode = f2fs_iget(sbi->sb, F2FS_COMPRESS_INO(sbi));
1886        if (IS_ERR(inode))
1887                return PTR_ERR(inode);
1888        sbi->compress_inode = inode;
1889
1890        sbi->compress_percent = COMPRESS_PERCENT;
1891        sbi->compress_watermark = COMPRESS_WATERMARK;
1892
1893        atomic_set(&sbi->compress_page_hit, 0);
1894
1895        return 0;
1896}
1897
1898void f2fs_destroy_compress_inode(struct f2fs_sb_info *sbi)
1899{
1900        if (!sbi->compress_inode)
1901                return;
1902        iput(sbi->compress_inode);
1903        sbi->compress_inode = NULL;
1904}
1905
1906int f2fs_init_page_array_cache(struct f2fs_sb_info *sbi)
1907{
1908        dev_t dev = sbi->sb->s_bdev->bd_dev;
1909        char slab_name[32];
1910
1911        sprintf(slab_name, "f2fs_page_array_entry-%u:%u", MAJOR(dev), MINOR(dev));
1912
1913        sbi->page_array_slab_size = sizeof(struct page *) <<
1914                                        F2FS_OPTION(sbi).compress_log_size;
1915
1916        sbi->page_array_slab = f2fs_kmem_cache_create(slab_name,
1917                                        sbi->page_array_slab_size);
1918        if (!sbi->page_array_slab)
1919                return -ENOMEM;
1920        return 0;
1921}
1922
1923void f2fs_destroy_page_array_cache(struct f2fs_sb_info *sbi)
1924{
1925        kmem_cache_destroy(sbi->page_array_slab);
1926}
1927
1928static int __init f2fs_init_cic_cache(void)
1929{
1930        cic_entry_slab = f2fs_kmem_cache_create("f2fs_cic_entry",
1931                                        sizeof(struct compress_io_ctx));
1932        if (!cic_entry_slab)
1933                return -ENOMEM;
1934        return 0;
1935}
1936
1937static void f2fs_destroy_cic_cache(void)
1938{
1939        kmem_cache_destroy(cic_entry_slab);
1940}
1941
1942static int __init f2fs_init_dic_cache(void)
1943{
1944        dic_entry_slab = f2fs_kmem_cache_create("f2fs_dic_entry",
1945                                        sizeof(struct decompress_io_ctx));
1946        if (!dic_entry_slab)
1947                return -ENOMEM;
1948        return 0;
1949}
1950
1951static void f2fs_destroy_dic_cache(void)
1952{
1953        kmem_cache_destroy(dic_entry_slab);
1954}
1955
1956int __init f2fs_init_compress_cache(void)
1957{
1958        int err;
1959
1960        err = f2fs_init_cic_cache();
1961        if (err)
1962                goto out;
1963        err = f2fs_init_dic_cache();
1964        if (err)
1965                goto free_cic;
1966        return 0;
1967free_cic:
1968        f2fs_destroy_cic_cache();
1969out:
1970        return -ENOMEM;
1971}
1972
1973void f2fs_destroy_compress_cache(void)
1974{
1975        f2fs_destroy_dic_cache();
1976        f2fs_destroy_cic_cache();
1977}
1978