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/moduleparam.h> 11 #include <linux/writeback.h> 12 #include <linux/backing-dev.h> 13 #include <linux/lzo.h> 14 #include <linux/lz4.h> 15 #include <linux/zstd.h> 16 #include <linux/pagevec.h> 17 18 #include "f2fs.h" 19 #include "node.h" 20 #include "segment.h" 21 #include <trace/events/f2fs.h> 22 23 static struct kmem_cache *cic_entry_slab; 24 static struct kmem_cache *dic_entry_slab; 25 26 static void *page_array_alloc(struct inode *inode, int nr) 27 { 28 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 29 unsigned int size = sizeof(struct page *) * nr; 30 31 if (likely(size <= sbi->page_array_slab_size)) 32 return f2fs_kmem_cache_alloc(sbi->page_array_slab, 33 GFP_F2FS_ZERO, false, F2FS_I_SB(inode)); 34 return f2fs_kzalloc(sbi, size, GFP_NOFS); 35 } 36 37 static void page_array_free(struct inode *inode, void *pages, int nr) 38 { 39 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 40 unsigned int size = sizeof(struct page *) * nr; 41 42 if (!pages) 43 return; 44 45 if (likely(size <= sbi->page_array_slab_size)) 46 kmem_cache_free(sbi->page_array_slab, pages); 47 else 48 kfree(pages); 49 } 50 51 struct f2fs_compress_ops { 52 int (*init_compress_ctx)(struct compress_ctx *cc); 53 void (*destroy_compress_ctx)(struct compress_ctx *cc); 54 int (*compress_pages)(struct compress_ctx *cc); 55 int (*init_decompress_ctx)(struct decompress_io_ctx *dic); 56 void (*destroy_decompress_ctx)(struct decompress_io_ctx *dic); 57 int (*decompress_pages)(struct decompress_io_ctx *dic); 58 }; 59 60 static unsigned int offset_in_cluster(struct compress_ctx *cc, pgoff_t index) 61 { 62 return index & (cc->cluster_size - 1); 63 } 64 65 static pgoff_t cluster_idx(struct compress_ctx *cc, pgoff_t index) 66 { 67 return index >> cc->log_cluster_size; 68 } 69 70 static pgoff_t start_idx_of_cluster(struct compress_ctx *cc) 71 { 72 return cc->cluster_idx << cc->log_cluster_size; 73 } 74 75 bool f2fs_is_compressed_page(struct page *page) 76 { 77 if (!PagePrivate(page)) 78 return false; 79 if (!page_private(page)) 80 return false; 81 if (page_private_nonpointer(page)) 82 return false; 83 84 f2fs_bug_on(F2FS_M_SB(page->mapping), 85 *((u32 *)page_private(page)) != F2FS_COMPRESSED_PAGE_MAGIC); 86 return true; 87 } 88 89 static void f2fs_set_compressed_page(struct page *page, 90 struct inode *inode, pgoff_t index, void *data) 91 { 92 attach_page_private(page, (void *)data); 93 94 /* i_crypto_info and iv index */ 95 page->index = index; 96 page->mapping = inode->i_mapping; 97 } 98 99 static void f2fs_drop_rpages(struct compress_ctx *cc, int len, bool unlock) 100 { 101 int i; 102 103 for (i = 0; i < len; i++) { 104 if (!cc->rpages[i]) 105 continue; 106 if (unlock) 107 unlock_page(cc->rpages[i]); 108 else 109 put_page(cc->rpages[i]); 110 } 111 } 112 113 static void f2fs_put_rpages(struct compress_ctx *cc) 114 { 115 f2fs_drop_rpages(cc, cc->cluster_size, false); 116 } 117 118 static void f2fs_unlock_rpages(struct compress_ctx *cc, int len) 119 { 120 f2fs_drop_rpages(cc, len, true); 121 } 122 123 static void f2fs_put_rpages_wbc(struct compress_ctx *cc, 124 struct writeback_control *wbc, bool redirty, int unlock) 125 { 126 unsigned int i; 127 128 for (i = 0; i < cc->cluster_size; i++) { 129 if (!cc->rpages[i]) 130 continue; 131 if (redirty) 132 redirty_page_for_writepage(wbc, cc->rpages[i]); 133 f2fs_put_page(cc->rpages[i], unlock); 134 } 135 } 136 137 struct page *f2fs_compress_control_page(struct page *page) 138 { 139 return ((struct compress_io_ctx *)page_private(page))->rpages[0]; 140 } 141 142 int f2fs_init_compress_ctx(struct compress_ctx *cc) 143 { 144 if (cc->rpages) 145 return 0; 146 147 cc->rpages = page_array_alloc(cc->inode, cc->cluster_size); 148 return cc->rpages ? 0 : -ENOMEM; 149 } 150 151 void f2fs_destroy_compress_ctx(struct compress_ctx *cc, bool reuse) 152 { 153 page_array_free(cc->inode, cc->rpages, cc->cluster_size); 154 cc->rpages = NULL; 155 cc->nr_rpages = 0; 156 cc->nr_cpages = 0; 157 if (!reuse) 158 cc->cluster_idx = NULL_CLUSTER; 159 } 160 161 void f2fs_compress_ctx_add_page(struct compress_ctx *cc, struct page *page) 162 { 163 unsigned int cluster_ofs; 164 165 if (!f2fs_cluster_can_merge_page(cc, page->index)) 166 f2fs_bug_on(F2FS_I_SB(cc->inode), 1); 167 168 cluster_ofs = offset_in_cluster(cc, page->index); 169 cc->rpages[cluster_ofs] = page; 170 cc->nr_rpages++; 171 cc->cluster_idx = cluster_idx(cc, page->index); 172 } 173 174 #ifdef CONFIG_F2FS_FS_LZO 175 static int lzo_init_compress_ctx(struct compress_ctx *cc) 176 { 177 cc->private = f2fs_kvmalloc(F2FS_I_SB(cc->inode), 178 LZO1X_MEM_COMPRESS, GFP_NOFS); 179 if (!cc->private) 180 return -ENOMEM; 181 182 cc->clen = lzo1x_worst_compress(PAGE_SIZE << cc->log_cluster_size); 183 return 0; 184 } 185 186 static void lzo_destroy_compress_ctx(struct compress_ctx *cc) 187 { 188 kvfree(cc->private); 189 cc->private = NULL; 190 } 191 192 static int lzo_compress_pages(struct compress_ctx *cc) 193 { 194 int ret; 195 196 ret = lzo1x_1_compress(cc->rbuf, cc->rlen, cc->cbuf->cdata, 197 &cc->clen, cc->private); 198 if (ret != LZO_E_OK) { 199 printk_ratelimited("%sF2FS-fs (%s): lzo compress failed, ret:%d\n", 200 KERN_ERR, F2FS_I_SB(cc->inode)->sb->s_id, ret); 201 return -EIO; 202 } 203 return 0; 204 } 205 206 static int lzo_decompress_pages(struct decompress_io_ctx *dic) 207 { 208 int ret; 209 210 ret = lzo1x_decompress_safe(dic->cbuf->cdata, dic->clen, 211 dic->rbuf, &dic->rlen); 212 if (ret != LZO_E_OK) { 213 printk_ratelimited("%sF2FS-fs (%s): lzo decompress failed, ret:%d\n", 214 KERN_ERR, F2FS_I_SB(dic->inode)->sb->s_id, ret); 215 return -EIO; 216 } 217 218 if (dic->rlen != PAGE_SIZE << dic->log_cluster_size) { 219 printk_ratelimited("%sF2FS-fs (%s): lzo invalid rlen:%zu, " 220 "expected:%lu\n", KERN_ERR, 221 F2FS_I_SB(dic->inode)->sb->s_id, 222 dic->rlen, 223 PAGE_SIZE << dic->log_cluster_size); 224 return -EIO; 225 } 226 return 0; 227 } 228 229 static const struct f2fs_compress_ops f2fs_lzo_ops = { 230 .init_compress_ctx = lzo_init_compress_ctx, 231 .destroy_compress_ctx = lzo_destroy_compress_ctx, 232 .compress_pages = lzo_compress_pages, 233 .decompress_pages = lzo_decompress_pages, 234 }; 235 #endif 236 237 #ifdef CONFIG_F2FS_FS_LZ4 238 static int lz4_init_compress_ctx(struct compress_ctx *cc) 239 { 240 unsigned int size = LZ4_MEM_COMPRESS; 241 242 #ifdef CONFIG_F2FS_FS_LZ4HC 243 if (F2FS_I(cc->inode)->i_compress_flag >> COMPRESS_LEVEL_OFFSET) 244 size = LZ4HC_MEM_COMPRESS; 245 #endif 246 247 cc->private = f2fs_kvmalloc(F2FS_I_SB(cc->inode), size, GFP_NOFS); 248 if (!cc->private) 249 return -ENOMEM; 250 251 /* 252 * we do not change cc->clen to LZ4_compressBound(inputsize) to 253 * adapt worst compress case, because lz4 compressor can handle 254 * output budget properly. 255 */ 256 cc->clen = cc->rlen - PAGE_SIZE - COMPRESS_HEADER_SIZE; 257 return 0; 258 } 259 260 static void lz4_destroy_compress_ctx(struct compress_ctx *cc) 261 { 262 kvfree(cc->private); 263 cc->private = NULL; 264 } 265 266 #ifdef CONFIG_F2FS_FS_LZ4HC 267 static int lz4hc_compress_pages(struct compress_ctx *cc) 268 { 269 unsigned char level = F2FS_I(cc->inode)->i_compress_flag >> 270 COMPRESS_LEVEL_OFFSET; 271 int len; 272 273 if (level) 274 len = LZ4_compress_HC(cc->rbuf, cc->cbuf->cdata, cc->rlen, 275 cc->clen, level, cc->private); 276 else 277 len = LZ4_compress_default(cc->rbuf, cc->cbuf->cdata, cc->rlen, 278 cc->clen, cc->private); 279 if (!len) 280 return -EAGAIN; 281 282 cc->clen = len; 283 return 0; 284 } 285 #endif 286 287 static int lz4_compress_pages(struct compress_ctx *cc) 288 { 289 int len; 290 291 #ifdef CONFIG_F2FS_FS_LZ4HC 292 return lz4hc_compress_pages(cc); 293 #endif 294 len = LZ4_compress_default(cc->rbuf, cc->cbuf->cdata, cc->rlen, 295 cc->clen, cc->private); 296 if (!len) 297 return -EAGAIN; 298 299 cc->clen = len; 300 return 0; 301 } 302 303 static int lz4_decompress_pages(struct decompress_io_ctx *dic) 304 { 305 int ret; 306 307 ret = LZ4_decompress_safe(dic->cbuf->cdata, dic->rbuf, 308 dic->clen, dic->rlen); 309 if (ret < 0) { 310 printk_ratelimited("%sF2FS-fs (%s): lz4 decompress failed, ret:%d\n", 311 KERN_ERR, F2FS_I_SB(dic->inode)->sb->s_id, ret); 312 return -EIO; 313 } 314 315 if (ret != PAGE_SIZE << dic->log_cluster_size) { 316 printk_ratelimited("%sF2FS-fs (%s): lz4 invalid rlen:%zu, " 317 "expected:%lu\n", KERN_ERR, 318 F2FS_I_SB(dic->inode)->sb->s_id, 319 dic->rlen, 320 PAGE_SIZE << dic->log_cluster_size); 321 return -EIO; 322 } 323 return 0; 324 } 325 326 static const struct f2fs_compress_ops f2fs_lz4_ops = { 327 .init_compress_ctx = lz4_init_compress_ctx, 328 .destroy_compress_ctx = lz4_destroy_compress_ctx, 329 .compress_pages = lz4_compress_pages, 330 .decompress_pages = lz4_decompress_pages, 331 }; 332 #endif 333 334 #ifdef CONFIG_F2FS_FS_ZSTD 335 #define F2FS_ZSTD_DEFAULT_CLEVEL 1 336 337 static int zstd_init_compress_ctx(struct compress_ctx *cc) 338 { 339 ZSTD_parameters params; 340 ZSTD_CStream *stream; 341 void *workspace; 342 unsigned int workspace_size; 343 unsigned char level = F2FS_I(cc->inode)->i_compress_flag >> 344 COMPRESS_LEVEL_OFFSET; 345 346 if (!level) 347 level = F2FS_ZSTD_DEFAULT_CLEVEL; 348 349 params = ZSTD_getParams(level, cc->rlen, 0); 350 workspace_size = ZSTD_CStreamWorkspaceBound(params.cParams); 351 352 workspace = f2fs_kvmalloc(F2FS_I_SB(cc->inode), 353 workspace_size, GFP_NOFS); 354 if (!workspace) 355 return -ENOMEM; 356 357 stream = ZSTD_initCStream(params, 0, workspace, workspace_size); 358 if (!stream) { 359 printk_ratelimited("%sF2FS-fs (%s): %s ZSTD_initCStream failed\n", 360 KERN_ERR, F2FS_I_SB(cc->inode)->sb->s_id, 361 __func__); 362 kvfree(workspace); 363 return -EIO; 364 } 365 366 cc->private = workspace; 367 cc->private2 = stream; 368 369 cc->clen = cc->rlen - PAGE_SIZE - COMPRESS_HEADER_SIZE; 370 return 0; 371 } 372 373 static void zstd_destroy_compress_ctx(struct compress_ctx *cc) 374 { 375 kvfree(cc->private); 376 cc->private = NULL; 377 cc->private2 = NULL; 378 } 379 380 static int zstd_compress_pages(struct compress_ctx *cc) 381 { 382 ZSTD_CStream *stream = cc->private2; 383 ZSTD_inBuffer inbuf; 384 ZSTD_outBuffer outbuf; 385 int src_size = cc->rlen; 386 int dst_size = src_size - PAGE_SIZE - COMPRESS_HEADER_SIZE; 387 int ret; 388 389 inbuf.pos = 0; 390 inbuf.src = cc->rbuf; 391 inbuf.size = src_size; 392 393 outbuf.pos = 0; 394 outbuf.dst = cc->cbuf->cdata; 395 outbuf.size = dst_size; 396 397 ret = ZSTD_compressStream(stream, &outbuf, &inbuf); 398 if (ZSTD_isError(ret)) { 399 printk_ratelimited("%sF2FS-fs (%s): %s ZSTD_compressStream failed, ret: %d\n", 400 KERN_ERR, F2FS_I_SB(cc->inode)->sb->s_id, 401 __func__, ZSTD_getErrorCode(ret)); 402 return -EIO; 403 } 404 405 ret = ZSTD_endStream(stream, &outbuf); 406 if (ZSTD_isError(ret)) { 407 printk_ratelimited("%sF2FS-fs (%s): %s ZSTD_endStream returned %d\n", 408 KERN_ERR, F2FS_I_SB(cc->inode)->sb->s_id, 409 __func__, ZSTD_getErrorCode(ret)); 410 return -EIO; 411 } 412 413 /* 414 * there is compressed data remained in intermediate buffer due to 415 * no more space in cbuf.cdata 416 */ 417 if (ret) 418 return -EAGAIN; 419 420 cc->clen = outbuf.pos; 421 return 0; 422 } 423 424 static int zstd_init_decompress_ctx(struct decompress_io_ctx *dic) 425 { 426 ZSTD_DStream *stream; 427 void *workspace; 428 unsigned int workspace_size; 429 unsigned int max_window_size = 430 MAX_COMPRESS_WINDOW_SIZE(dic->log_cluster_size); 431 432 workspace_size = ZSTD_DStreamWorkspaceBound(max_window_size); 433 434 workspace = f2fs_kvmalloc(F2FS_I_SB(dic->inode), 435 workspace_size, GFP_NOFS); 436 if (!workspace) 437 return -ENOMEM; 438 439 stream = ZSTD_initDStream(max_window_size, workspace, workspace_size); 440 if (!stream) { 441 printk_ratelimited("%sF2FS-fs (%s): %s ZSTD_initDStream failed\n", 442 KERN_ERR, F2FS_I_SB(dic->inode)->sb->s_id, 443 __func__); 444 kvfree(workspace); 445 return -EIO; 446 } 447 448 dic->private = workspace; 449 dic->private2 = stream; 450 451 return 0; 452 } 453 454 static void zstd_destroy_decompress_ctx(struct decompress_io_ctx *dic) 455 { 456 kvfree(dic->private); 457 dic->private = NULL; 458 dic->private2 = NULL; 459 } 460 461 static int zstd_decompress_pages(struct decompress_io_ctx *dic) 462 { 463 ZSTD_DStream *stream = dic->private2; 464 ZSTD_inBuffer inbuf; 465 ZSTD_outBuffer outbuf; 466 int ret; 467 468 inbuf.pos = 0; 469 inbuf.src = dic->cbuf->cdata; 470 inbuf.size = dic->clen; 471 472 outbuf.pos = 0; 473 outbuf.dst = dic->rbuf; 474 outbuf.size = dic->rlen; 475 476 ret = ZSTD_decompressStream(stream, &outbuf, &inbuf); 477 if (ZSTD_isError(ret)) { 478 printk_ratelimited("%sF2FS-fs (%s): %s ZSTD_compressStream failed, ret: %d\n", 479 KERN_ERR, F2FS_I_SB(dic->inode)->sb->s_id, 480 __func__, ZSTD_getErrorCode(ret)); 481 return -EIO; 482 } 483 484 if (dic->rlen != outbuf.pos) { 485 printk_ratelimited("%sF2FS-fs (%s): %s ZSTD invalid rlen:%zu, " 486 "expected:%lu\n", KERN_ERR, 487 F2FS_I_SB(dic->inode)->sb->s_id, 488 __func__, dic->rlen, 489 PAGE_SIZE << dic->log_cluster_size); 490 return -EIO; 491 } 492 493 return 0; 494 } 495 496 static const struct f2fs_compress_ops f2fs_zstd_ops = { 497 .init_compress_ctx = zstd_init_compress_ctx, 498 .destroy_compress_ctx = zstd_destroy_compress_ctx, 499 .compress_pages = zstd_compress_pages, 500 .init_decompress_ctx = zstd_init_decompress_ctx, 501 .destroy_decompress_ctx = zstd_destroy_decompress_ctx, 502 .decompress_pages = zstd_decompress_pages, 503 }; 504 #endif 505 506 #ifdef CONFIG_F2FS_FS_LZO 507 #ifdef CONFIG_F2FS_FS_LZORLE 508 static int lzorle_compress_pages(struct compress_ctx *cc) 509 { 510 int ret; 511 512 ret = lzorle1x_1_compress(cc->rbuf, cc->rlen, cc->cbuf->cdata, 513 &cc->clen, cc->private); 514 if (ret != LZO_E_OK) { 515 printk_ratelimited("%sF2FS-fs (%s): lzo-rle compress failed, ret:%d\n", 516 KERN_ERR, F2FS_I_SB(cc->inode)->sb->s_id, ret); 517 return -EIO; 518 } 519 return 0; 520 } 521 522 static const struct f2fs_compress_ops f2fs_lzorle_ops = { 523 .init_compress_ctx = lzo_init_compress_ctx, 524 .destroy_compress_ctx = lzo_destroy_compress_ctx, 525 .compress_pages = lzorle_compress_pages, 526 .decompress_pages = lzo_decompress_pages, 527 }; 528 #endif 529 #endif 530 531 static const struct f2fs_compress_ops *f2fs_cops[COMPRESS_MAX] = { 532 #ifdef CONFIG_F2FS_FS_LZO 533 &f2fs_lzo_ops, 534 #else 535 NULL, 536 #endif 537 #ifdef CONFIG_F2FS_FS_LZ4 538 &f2fs_lz4_ops, 539 #else 540 NULL, 541 #endif 542 #ifdef CONFIG_F2FS_FS_ZSTD 543 &f2fs_zstd_ops, 544 #else 545 NULL, 546 #endif 547 #if defined(CONFIG_F2FS_FS_LZO) && defined(CONFIG_F2FS_FS_LZORLE) 548 &f2fs_lzorle_ops, 549 #else 550 NULL, 551 #endif 552 }; 553 554 bool f2fs_is_compress_backend_ready(struct inode *inode) 555 { 556 if (!f2fs_compressed_file(inode)) 557 return true; 558 return f2fs_cops[F2FS_I(inode)->i_compress_algorithm]; 559 } 560 561 static mempool_t *compress_page_pool; 562 static int num_compress_pages = 512; 563 module_param(num_compress_pages, uint, 0444); 564 MODULE_PARM_DESC(num_compress_pages, 565 "Number of intermediate compress pages to preallocate"); 566 567 int f2fs_init_compress_mempool(void) 568 { 569 compress_page_pool = mempool_create_page_pool(num_compress_pages, 0); 570 if (!compress_page_pool) 571 return -ENOMEM; 572 573 return 0; 574 } 575 576 void f2fs_destroy_compress_mempool(void) 577 { 578 mempool_destroy(compress_page_pool); 579 } 580 581 static struct page *f2fs_compress_alloc_page(void) 582 { 583 struct page *page; 584 585 page = mempool_alloc(compress_page_pool, GFP_NOFS); 586 lock_page(page); 587 588 return page; 589 } 590 591 static void f2fs_compress_free_page(struct page *page) 592 { 593 if (!page) 594 return; 595 detach_page_private(page); 596 page->mapping = NULL; 597 unlock_page(page); 598 mempool_free(page, compress_page_pool); 599 } 600 601 #define MAX_VMAP_RETRIES 3 602 603 static void *f2fs_vmap(struct page **pages, unsigned int count) 604 { 605 int i; 606 void *buf = NULL; 607 608 for (i = 0; i < MAX_VMAP_RETRIES; i++) { 609 buf = vm_map_ram(pages, count, -1); 610 if (buf) 611 break; 612 vm_unmap_aliases(); 613 } 614 return buf; 615 } 616 617 static int f2fs_compress_pages(struct compress_ctx *cc) 618 { 619 struct f2fs_inode_info *fi = F2FS_I(cc->inode); 620 const struct f2fs_compress_ops *cops = 621 f2fs_cops[fi->i_compress_algorithm]; 622 unsigned int max_len, new_nr_cpages; 623 struct page **new_cpages; 624 u32 chksum = 0; 625 int i, ret; 626 627 trace_f2fs_compress_pages_start(cc->inode, cc->cluster_idx, 628 cc->cluster_size, fi->i_compress_algorithm); 629 630 if (cops->init_compress_ctx) { 631 ret = cops->init_compress_ctx(cc); 632 if (ret) 633 goto out; 634 } 635 636 max_len = COMPRESS_HEADER_SIZE + cc->clen; 637 cc->nr_cpages = DIV_ROUND_UP(max_len, PAGE_SIZE); 638 639 cc->cpages = page_array_alloc(cc->inode, cc->nr_cpages); 640 if (!cc->cpages) { 641 ret = -ENOMEM; 642 goto destroy_compress_ctx; 643 } 644 645 for (i = 0; i < cc->nr_cpages; i++) { 646 cc->cpages[i] = f2fs_compress_alloc_page(); 647 if (!cc->cpages[i]) { 648 ret = -ENOMEM; 649 goto out_free_cpages; 650 } 651 } 652 653 cc->rbuf = f2fs_vmap(cc->rpages, cc->cluster_size); 654 if (!cc->rbuf) { 655 ret = -ENOMEM; 656 goto out_free_cpages; 657 } 658 659 cc->cbuf = f2fs_vmap(cc->cpages, cc->nr_cpages); 660 if (!cc->cbuf) { 661 ret = -ENOMEM; 662 goto out_vunmap_rbuf; 663 } 664 665 ret = cops->compress_pages(cc); 666 if (ret) 667 goto out_vunmap_cbuf; 668 669 max_len = PAGE_SIZE * (cc->cluster_size - 1) - COMPRESS_HEADER_SIZE; 670 671 if (cc->clen > max_len) { 672 ret = -EAGAIN; 673 goto out_vunmap_cbuf; 674 } 675 676 cc->cbuf->clen = cpu_to_le32(cc->clen); 677 678 if (fi->i_compress_flag & 1 << COMPRESS_CHKSUM) 679 chksum = f2fs_crc32(F2FS_I_SB(cc->inode), 680 cc->cbuf->cdata, cc->clen); 681 cc->cbuf->chksum = cpu_to_le32(chksum); 682 683 for (i = 0; i < COMPRESS_DATA_RESERVED_SIZE; i++) 684 cc->cbuf->reserved[i] = cpu_to_le32(0); 685 686 new_nr_cpages = DIV_ROUND_UP(cc->clen + COMPRESS_HEADER_SIZE, PAGE_SIZE); 687 688 /* Now we're going to cut unnecessary tail pages */ 689 new_cpages = page_array_alloc(cc->inode, new_nr_cpages); 690 if (!new_cpages) { 691 ret = -ENOMEM; 692 goto out_vunmap_cbuf; 693 } 694 695 /* zero out any unused part of the last page */ 696 memset(&cc->cbuf->cdata[cc->clen], 0, 697 (new_nr_cpages * PAGE_SIZE) - 698 (cc->clen + COMPRESS_HEADER_SIZE)); 699 700 vm_unmap_ram(cc->cbuf, cc->nr_cpages); 701 vm_unmap_ram(cc->rbuf, cc->cluster_size); 702 703 for (i = 0; i < cc->nr_cpages; i++) { 704 if (i < new_nr_cpages) { 705 new_cpages[i] = cc->cpages[i]; 706 continue; 707 } 708 f2fs_compress_free_page(cc->cpages[i]); 709 cc->cpages[i] = NULL; 710 } 711 712 if (cops->destroy_compress_ctx) 713 cops->destroy_compress_ctx(cc); 714 715 page_array_free(cc->inode, cc->cpages, cc->nr_cpages); 716 cc->cpages = new_cpages; 717 cc->nr_cpages = new_nr_cpages; 718 719 trace_f2fs_compress_pages_end(cc->inode, cc->cluster_idx, 720 cc->clen, ret); 721 return 0; 722 723 out_vunmap_cbuf: 724 vm_unmap_ram(cc->cbuf, cc->nr_cpages); 725 out_vunmap_rbuf: 726 vm_unmap_ram(cc->rbuf, cc->cluster_size); 727 out_free_cpages: 728 for (i = 0; i < cc->nr_cpages; i++) { 729 if (cc->cpages[i]) 730 f2fs_compress_free_page(cc->cpages[i]); 731 } 732 page_array_free(cc->inode, cc->cpages, cc->nr_cpages); 733 cc->cpages = NULL; 734 destroy_compress_ctx: 735 if (cops->destroy_compress_ctx) 736 cops->destroy_compress_ctx(cc); 737 out: 738 trace_f2fs_compress_pages_end(cc->inode, cc->cluster_idx, 739 cc->clen, ret); 740 return ret; 741 } 742 743 void f2fs_decompress_cluster(struct decompress_io_ctx *dic) 744 { 745 struct f2fs_sb_info *sbi = F2FS_I_SB(dic->inode); 746 struct f2fs_inode_info *fi = F2FS_I(dic->inode); 747 const struct f2fs_compress_ops *cops = 748 f2fs_cops[fi->i_compress_algorithm]; 749 int ret; 750 int i; 751 752 trace_f2fs_decompress_pages_start(dic->inode, dic->cluster_idx, 753 dic->cluster_size, fi->i_compress_algorithm); 754 755 if (dic->failed) { 756 ret = -EIO; 757 goto out_end_io; 758 } 759 760 dic->tpages = page_array_alloc(dic->inode, dic->cluster_size); 761 if (!dic->tpages) { 762 ret = -ENOMEM; 763 goto out_end_io; 764 } 765 766 for (i = 0; i < dic->cluster_size; i++) { 767 if (dic->rpages[i]) { 768 dic->tpages[i] = dic->rpages[i]; 769 continue; 770 } 771 772 dic->tpages[i] = f2fs_compress_alloc_page(); 773 if (!dic->tpages[i]) { 774 ret = -ENOMEM; 775 goto out_end_io; 776 } 777 } 778 779 if (cops->init_decompress_ctx) { 780 ret = cops->init_decompress_ctx(dic); 781 if (ret) 782 goto out_end_io; 783 } 784 785 dic->rbuf = f2fs_vmap(dic->tpages, dic->cluster_size); 786 if (!dic->rbuf) { 787 ret = -ENOMEM; 788 goto out_destroy_decompress_ctx; 789 } 790 791 dic->cbuf = f2fs_vmap(dic->cpages, dic->nr_cpages); 792 if (!dic->cbuf) { 793 ret = -ENOMEM; 794 goto out_vunmap_rbuf; 795 } 796 797 dic->clen = le32_to_cpu(dic->cbuf->clen); 798 dic->rlen = PAGE_SIZE << dic->log_cluster_size; 799 800 if (dic->clen > PAGE_SIZE * dic->nr_cpages - COMPRESS_HEADER_SIZE) { 801 ret = -EFSCORRUPTED; 802 goto out_vunmap_cbuf; 803 } 804 805 ret = cops->decompress_pages(dic); 806 807 if (!ret && (fi->i_compress_flag & 1 << COMPRESS_CHKSUM)) { 808 u32 provided = le32_to_cpu(dic->cbuf->chksum); 809 u32 calculated = f2fs_crc32(sbi, dic->cbuf->cdata, dic->clen); 810 811 if (provided != calculated) { 812 if (!is_inode_flag_set(dic->inode, FI_COMPRESS_CORRUPT)) { 813 set_inode_flag(dic->inode, FI_COMPRESS_CORRUPT); 814 printk_ratelimited( 815 "%sF2FS-fs (%s): checksum invalid, nid = %lu, %x vs %x", 816 KERN_INFO, sbi->sb->s_id, dic->inode->i_ino, 817 provided, calculated); 818 } 819 set_sbi_flag(sbi, SBI_NEED_FSCK); 820 } 821 } 822 823 out_vunmap_cbuf: 824 vm_unmap_ram(dic->cbuf, dic->nr_cpages); 825 out_vunmap_rbuf: 826 vm_unmap_ram(dic->rbuf, dic->cluster_size); 827 out_destroy_decompress_ctx: 828 if (cops->destroy_decompress_ctx) 829 cops->destroy_decompress_ctx(dic); 830 out_end_io: 831 trace_f2fs_decompress_pages_end(dic->inode, dic->cluster_idx, 832 dic->clen, ret); 833 f2fs_decompress_end_io(dic, ret); 834 } 835 836 /* 837 * This is called when a page of a compressed cluster has been read from disk 838 * (or failed to be read from disk). It checks whether this page was the last 839 * page being waited on in the cluster, and if so, it decompresses the cluster 840 * (or in the case of a failure, cleans up without actually decompressing). 841 */ 842 void f2fs_end_read_compressed_page(struct page *page, bool failed, 843 block_t blkaddr) 844 { 845 struct decompress_io_ctx *dic = 846 (struct decompress_io_ctx *)page_private(page); 847 struct f2fs_sb_info *sbi = F2FS_I_SB(dic->inode); 848 849 dec_page_count(sbi, F2FS_RD_DATA); 850 851 if (failed) 852 WRITE_ONCE(dic->failed, true); 853 else if (blkaddr) 854 f2fs_cache_compressed_page(sbi, page, 855 dic->inode->i_ino, blkaddr); 856 857 if (atomic_dec_and_test(&dic->remaining_pages)) 858 f2fs_decompress_cluster(dic); 859 } 860 861 static bool is_page_in_cluster(struct compress_ctx *cc, pgoff_t index) 862 { 863 if (cc->cluster_idx == NULL_CLUSTER) 864 return true; 865 return cc->cluster_idx == cluster_idx(cc, index); 866 } 867 868 bool f2fs_cluster_is_empty(struct compress_ctx *cc) 869 { 870 return cc->nr_rpages == 0; 871 } 872 873 static bool f2fs_cluster_is_full(struct compress_ctx *cc) 874 { 875 return cc->cluster_size == cc->nr_rpages; 876 } 877 878 bool f2fs_cluster_can_merge_page(struct compress_ctx *cc, pgoff_t index) 879 { 880 if (f2fs_cluster_is_empty(cc)) 881 return true; 882 return is_page_in_cluster(cc, index); 883 } 884 885 static bool cluster_has_invalid_data(struct compress_ctx *cc) 886 { 887 loff_t i_size = i_size_read(cc->inode); 888 unsigned nr_pages = DIV_ROUND_UP(i_size, PAGE_SIZE); 889 int i; 890 891 for (i = 0; i < cc->cluster_size; i++) { 892 struct page *page = cc->rpages[i]; 893 894 f2fs_bug_on(F2FS_I_SB(cc->inode), !page); 895 896 /* beyond EOF */ 897 if (page->index >= nr_pages) 898 return true; 899 } 900 return false; 901 } 902 903 bool f2fs_sanity_check_cluster(struct dnode_of_data *dn) 904 { 905 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode); 906 unsigned int cluster_size = F2FS_I(dn->inode)->i_cluster_size; 907 bool compressed = dn->data_blkaddr == COMPRESS_ADDR; 908 int cluster_end = 0; 909 int i; 910 char *reason = ""; 911 912 if (!compressed) 913 return false; 914 915 /* [..., COMPR_ADDR, ...] */ 916 if (dn->ofs_in_node % cluster_size) { 917 reason = "[*|C|*|*]"; 918 goto out; 919 } 920 921 for (i = 1; i < cluster_size; i++) { 922 block_t blkaddr = data_blkaddr(dn->inode, dn->node_page, 923 dn->ofs_in_node + i); 924 925 /* [COMPR_ADDR, ..., COMPR_ADDR] */ 926 if (blkaddr == COMPRESS_ADDR) { 927 reason = "[C|*|C|*]"; 928 goto out; 929 } 930 if (compressed) { 931 if (!__is_valid_data_blkaddr(blkaddr)) { 932 if (!cluster_end) 933 cluster_end = i; 934 continue; 935 } 936 /* [COMPR_ADDR, NULL_ADDR or NEW_ADDR, valid_blkaddr] */ 937 if (cluster_end) { 938 reason = "[C|N|N|V]"; 939 goto out; 940 } 941 } 942 } 943 return false; 944 out: 945 f2fs_warn(sbi, "access invalid cluster, ino:%lu, nid:%u, ofs_in_node:%u, reason:%s", 946 dn->inode->i_ino, dn->nid, dn->ofs_in_node, reason); 947 set_sbi_flag(sbi, SBI_NEED_FSCK); 948 return true; 949 } 950 951 static int __f2fs_cluster_blocks(struct inode *inode, 952 unsigned int cluster_idx, bool compr) 953 { 954 struct dnode_of_data dn; 955 unsigned int cluster_size = F2FS_I(inode)->i_cluster_size; 956 unsigned int start_idx = cluster_idx << 957 F2FS_I(inode)->i_log_cluster_size; 958 int ret; 959 960 set_new_dnode(&dn, inode, NULL, NULL, 0); 961 ret = f2fs_get_dnode_of_data(&dn, start_idx, LOOKUP_NODE); 962 if (ret) { 963 if (ret == -ENOENT) 964 ret = 0; 965 goto fail; 966 } 967 968 if (f2fs_sanity_check_cluster(&dn)) { 969 ret = -EFSCORRUPTED; 970 goto fail; 971 } 972 973 if (dn.data_blkaddr == COMPRESS_ADDR) { 974 int i; 975 976 ret = 1; 977 for (i = 1; i < cluster_size; i++) { 978 block_t blkaddr; 979 980 blkaddr = data_blkaddr(dn.inode, 981 dn.node_page, dn.ofs_in_node + i); 982 if (compr) { 983 if (__is_valid_data_blkaddr(blkaddr)) 984 ret++; 985 } else { 986 if (blkaddr != NULL_ADDR) 987 ret++; 988 } 989 } 990 991 f2fs_bug_on(F2FS_I_SB(inode), 992 !compr && ret != cluster_size && 993 !is_inode_flag_set(inode, FI_COMPRESS_RELEASED)); 994 } 995 fail: 996 f2fs_put_dnode(&dn); 997 return ret; 998 } 999 1000 /* return # of compressed blocks in compressed cluster */ 1001 static int f2fs_compressed_blocks(struct compress_ctx *cc) 1002 { 1003 return __f2fs_cluster_blocks(cc->inode, cc->cluster_idx, true); 1004 } 1005 1006 /* return # of valid blocks in compressed cluster */ 1007 int f2fs_is_compressed_cluster(struct inode *inode, pgoff_t index) 1008 { 1009 return __f2fs_cluster_blocks(inode, 1010 index >> F2FS_I(inode)->i_log_cluster_size, 1011 false); 1012 } 1013 1014 static bool cluster_may_compress(struct compress_ctx *cc) 1015 { 1016 if (!f2fs_need_compress_data(cc->inode)) 1017 return false; 1018 if (f2fs_is_atomic_file(cc->inode)) 1019 return false; 1020 if (!f2fs_cluster_is_full(cc)) 1021 return false; 1022 if (unlikely(f2fs_cp_error(F2FS_I_SB(cc->inode)))) 1023 return false; 1024 return !cluster_has_invalid_data(cc); 1025 } 1026 1027 static void set_cluster_writeback(struct compress_ctx *cc) 1028 { 1029 int i; 1030 1031 for (i = 0; i < cc->cluster_size; i++) { 1032 if (cc->rpages[i]) 1033 set_page_writeback(cc->rpages[i]); 1034 } 1035 } 1036 1037 static void set_cluster_dirty(struct compress_ctx *cc) 1038 { 1039 int i; 1040 1041 for (i = 0; i < cc->cluster_size; i++) 1042 if (cc->rpages[i]) 1043 set_page_dirty(cc->rpages[i]); 1044 } 1045 1046 static int prepare_compress_overwrite(struct compress_ctx *cc, 1047 struct page **pagep, pgoff_t index, void **fsdata) 1048 { 1049 struct f2fs_sb_info *sbi = F2FS_I_SB(cc->inode); 1050 struct address_space *mapping = cc->inode->i_mapping; 1051 struct page *page; 1052 sector_t last_block_in_bio; 1053 unsigned fgp_flag = FGP_LOCK | FGP_WRITE | FGP_CREAT; 1054 pgoff_t start_idx = start_idx_of_cluster(cc); 1055 int i, ret; 1056 1057 retry: 1058 ret = f2fs_is_compressed_cluster(cc->inode, start_idx); 1059 if (ret <= 0) 1060 return ret; 1061 1062 ret = f2fs_init_compress_ctx(cc); 1063 if (ret) 1064 return ret; 1065 1066 /* keep page reference to avoid page reclaim */ 1067 for (i = 0; i < cc->cluster_size; i++) { 1068 page = f2fs_pagecache_get_page(mapping, start_idx + i, 1069 fgp_flag, GFP_NOFS); 1070 if (!page) { 1071 ret = -ENOMEM; 1072 goto unlock_pages; 1073 } 1074 1075 if (PageUptodate(page)) 1076 f2fs_put_page(page, 1); 1077 else 1078 f2fs_compress_ctx_add_page(cc, page); 1079 } 1080 1081 if (!f2fs_cluster_is_empty(cc)) { 1082 struct bio *bio = NULL; 1083 1084 ret = f2fs_read_multi_pages(cc, &bio, cc->cluster_size, 1085 &last_block_in_bio, false, true); 1086 f2fs_put_rpages(cc); 1087 f2fs_destroy_compress_ctx(cc, true); 1088 if (ret) 1089 goto out; 1090 if (bio) 1091 f2fs_submit_bio(sbi, bio, DATA); 1092 1093 ret = f2fs_init_compress_ctx(cc); 1094 if (ret) 1095 goto out; 1096 } 1097 1098 for (i = 0; i < cc->cluster_size; i++) { 1099 f2fs_bug_on(sbi, cc->rpages[i]); 1100 1101 page = find_lock_page(mapping, start_idx + i); 1102 if (!page) { 1103 /* page can be truncated */ 1104 goto release_and_retry; 1105 } 1106 1107 f2fs_wait_on_page_writeback(page, DATA, true, true); 1108 f2fs_compress_ctx_add_page(cc, page); 1109 1110 if (!PageUptodate(page)) { 1111 release_and_retry: 1112 f2fs_put_rpages(cc); 1113 f2fs_unlock_rpages(cc, i + 1); 1114 f2fs_destroy_compress_ctx(cc, true); 1115 goto retry; 1116 } 1117 } 1118 1119 if (likely(!ret)) { 1120 *fsdata = cc->rpages; 1121 *pagep = cc->rpages[offset_in_cluster(cc, index)]; 1122 return cc->cluster_size; 1123 } 1124 1125 unlock_pages: 1126 f2fs_put_rpages(cc); 1127 f2fs_unlock_rpages(cc, i); 1128 f2fs_destroy_compress_ctx(cc, true); 1129 out: 1130 return ret; 1131 } 1132 1133 int f2fs_prepare_compress_overwrite(struct inode *inode, 1134 struct page **pagep, pgoff_t index, void **fsdata) 1135 { 1136 struct compress_ctx cc = { 1137 .inode = inode, 1138 .log_cluster_size = F2FS_I(inode)->i_log_cluster_size, 1139 .cluster_size = F2FS_I(inode)->i_cluster_size, 1140 .cluster_idx = index >> F2FS_I(inode)->i_log_cluster_size, 1141 .rpages = NULL, 1142 .nr_rpages = 0, 1143 }; 1144 1145 return prepare_compress_overwrite(&cc, pagep, index, fsdata); 1146 } 1147 1148 bool f2fs_compress_write_end(struct inode *inode, void *fsdata, 1149 pgoff_t index, unsigned copied) 1150 1151 { 1152 struct compress_ctx cc = { 1153 .inode = inode, 1154 .log_cluster_size = F2FS_I(inode)->i_log_cluster_size, 1155 .cluster_size = F2FS_I(inode)->i_cluster_size, 1156 .rpages = fsdata, 1157 }; 1158 bool first_index = (index == cc.rpages[0]->index); 1159 1160 if (copied) 1161 set_cluster_dirty(&cc); 1162 1163 f2fs_put_rpages_wbc(&cc, NULL, false, 1); 1164 f2fs_destroy_compress_ctx(&cc, false); 1165 1166 return first_index; 1167 } 1168 1169 int f2fs_truncate_partial_cluster(struct inode *inode, u64 from, bool lock) 1170 { 1171 void *fsdata = NULL; 1172 struct page *pagep; 1173 int log_cluster_size = F2FS_I(inode)->i_log_cluster_size; 1174 pgoff_t start_idx = from >> (PAGE_SHIFT + log_cluster_size) << 1175 log_cluster_size; 1176 int err; 1177 1178 err = f2fs_is_compressed_cluster(inode, start_idx); 1179 if (err < 0) 1180 return err; 1181 1182 /* truncate normal cluster */ 1183 if (!err) 1184 return f2fs_do_truncate_blocks(inode, from, lock); 1185 1186 /* truncate compressed cluster */ 1187 err = f2fs_prepare_compress_overwrite(inode, &pagep, 1188 start_idx, &fsdata); 1189 1190 /* should not be a normal cluster */ 1191 f2fs_bug_on(F2FS_I_SB(inode), err == 0); 1192 1193 if (err <= 0) 1194 return err; 1195 1196 if (err > 0) { 1197 struct page **rpages = fsdata; 1198 int cluster_size = F2FS_I(inode)->i_cluster_size; 1199 int i; 1200 1201 for (i = cluster_size - 1; i >= 0; i--) { 1202 loff_t start = rpages[i]->index << PAGE_SHIFT; 1203 1204 if (from <= start) { 1205 zero_user_segment(rpages[i], 0, PAGE_SIZE); 1206 } else { 1207 zero_user_segment(rpages[i], from - start, 1208 PAGE_SIZE); 1209 break; 1210 } 1211 } 1212 1213 f2fs_compress_write_end(inode, fsdata, start_idx, true); 1214 } 1215 return 0; 1216 } 1217 1218 static int f2fs_write_compressed_pages(struct compress_ctx *cc, 1219 int *submitted, 1220 struct writeback_control *wbc, 1221 enum iostat_type io_type) 1222 { 1223 struct inode *inode = cc->inode; 1224 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1225 struct f2fs_inode_info *fi = F2FS_I(inode); 1226 struct f2fs_io_info fio = { 1227 .sbi = sbi, 1228 .ino = cc->inode->i_ino, 1229 .type = DATA, 1230 .op = REQ_OP_WRITE, 1231 .op_flags = wbc_to_write_flags(wbc), 1232 .old_blkaddr = NEW_ADDR, 1233 .page = NULL, 1234 .encrypted_page = NULL, 1235 .compressed_page = NULL, 1236 .submitted = false, 1237 .io_type = io_type, 1238 .io_wbc = wbc, 1239 .encrypted = fscrypt_inode_uses_fs_layer_crypto(cc->inode), 1240 }; 1241 struct dnode_of_data dn; 1242 struct node_info ni; 1243 struct compress_io_ctx *cic; 1244 pgoff_t start_idx = start_idx_of_cluster(cc); 1245 unsigned int last_index = cc->cluster_size - 1; 1246 loff_t psize; 1247 int i, err; 1248 1249 /* we should bypass data pages to proceed the kworkder jobs */ 1250 if (unlikely(f2fs_cp_error(sbi))) { 1251 mapping_set_error(cc->rpages[0]->mapping, -EIO); 1252 goto out_free; 1253 } 1254 1255 if (IS_NOQUOTA(inode)) { 1256 /* 1257 * We need to wait for node_write to avoid block allocation during 1258 * checkpoint. This can only happen to quota writes which can cause 1259 * the below discard race condition. 1260 */ 1261 down_read(&sbi->node_write); 1262 } else if (!f2fs_trylock_op(sbi)) { 1263 goto out_free; 1264 } 1265 1266 set_new_dnode(&dn, cc->inode, NULL, NULL, 0); 1267 1268 err = f2fs_get_dnode_of_data(&dn, start_idx, LOOKUP_NODE); 1269 if (err) 1270 goto out_unlock_op; 1271 1272 for (i = 0; i < cc->cluster_size; i++) { 1273 if (data_blkaddr(dn.inode, dn.node_page, 1274 dn.ofs_in_node + i) == NULL_ADDR) 1275 goto out_put_dnode; 1276 } 1277 1278 psize = (loff_t)(cc->rpages[last_index]->index + 1) << PAGE_SHIFT; 1279 1280 err = f2fs_get_node_info(fio.sbi, dn.nid, &ni); 1281 if (err) 1282 goto out_put_dnode; 1283 1284 fio.version = ni.version; 1285 1286 cic = f2fs_kmem_cache_alloc(cic_entry_slab, GFP_F2FS_ZERO, false, sbi); 1287 if (!cic) 1288 goto out_put_dnode; 1289 1290 cic->magic = F2FS_COMPRESSED_PAGE_MAGIC; 1291 cic->inode = inode; 1292 atomic_set(&cic->pending_pages, cc->nr_cpages); 1293 cic->rpages = page_array_alloc(cc->inode, cc->cluster_size); 1294 if (!cic->rpages) 1295 goto out_put_cic; 1296 1297 cic->nr_rpages = cc->cluster_size; 1298 1299 for (i = 0; i < cc->nr_cpages; i++) { 1300 f2fs_set_compressed_page(cc->cpages[i], inode, 1301 cc->rpages[i + 1]->index, cic); 1302 fio.compressed_page = cc->cpages[i]; 1303 1304 fio.old_blkaddr = data_blkaddr(dn.inode, dn.node_page, 1305 dn.ofs_in_node + i + 1); 1306 1307 /* wait for GCed page writeback via META_MAPPING */ 1308 f2fs_wait_on_block_writeback(inode, fio.old_blkaddr); 1309 1310 if (fio.encrypted) { 1311 fio.page = cc->rpages[i + 1]; 1312 err = f2fs_encrypt_one_page(&fio); 1313 if (err) 1314 goto out_destroy_crypt; 1315 cc->cpages[i] = fio.encrypted_page; 1316 } 1317 } 1318 1319 set_cluster_writeback(cc); 1320 1321 for (i = 0; i < cc->cluster_size; i++) 1322 cic->rpages[i] = cc->rpages[i]; 1323 1324 for (i = 0; i < cc->cluster_size; i++, dn.ofs_in_node++) { 1325 block_t blkaddr; 1326 1327 blkaddr = f2fs_data_blkaddr(&dn); 1328 fio.page = cc->rpages[i]; 1329 fio.old_blkaddr = blkaddr; 1330 1331 /* cluster header */ 1332 if (i == 0) { 1333 if (blkaddr == COMPRESS_ADDR) 1334 fio.compr_blocks++; 1335 if (__is_valid_data_blkaddr(blkaddr)) 1336 f2fs_invalidate_blocks(sbi, blkaddr); 1337 f2fs_update_data_blkaddr(&dn, COMPRESS_ADDR); 1338 goto unlock_continue; 1339 } 1340 1341 if (fio.compr_blocks && __is_valid_data_blkaddr(blkaddr)) 1342 fio.compr_blocks++; 1343 1344 if (i > cc->nr_cpages) { 1345 if (__is_valid_data_blkaddr(blkaddr)) { 1346 f2fs_invalidate_blocks(sbi, blkaddr); 1347 f2fs_update_data_blkaddr(&dn, NEW_ADDR); 1348 } 1349 goto unlock_continue; 1350 } 1351 1352 f2fs_bug_on(fio.sbi, blkaddr == NULL_ADDR); 1353 1354 if (fio.encrypted) 1355 fio.encrypted_page = cc->cpages[i - 1]; 1356 else 1357 fio.compressed_page = cc->cpages[i - 1]; 1358 1359 cc->cpages[i - 1] = NULL; 1360 f2fs_outplace_write_data(&dn, &fio); 1361 (*submitted)++; 1362 unlock_continue: 1363 inode_dec_dirty_pages(cc->inode); 1364 unlock_page(fio.page); 1365 } 1366 1367 if (fio.compr_blocks) 1368 f2fs_i_compr_blocks_update(inode, fio.compr_blocks - 1, false); 1369 f2fs_i_compr_blocks_update(inode, cc->nr_cpages, true); 1370 add_compr_block_stat(inode, cc->nr_cpages); 1371 1372 set_inode_flag(cc->inode, FI_APPEND_WRITE); 1373 if (cc->cluster_idx == 0) 1374 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN); 1375 1376 f2fs_put_dnode(&dn); 1377 if (IS_NOQUOTA(inode)) 1378 up_read(&sbi->node_write); 1379 else 1380 f2fs_unlock_op(sbi); 1381 1382 spin_lock(&fi->i_size_lock); 1383 if (fi->last_disk_size < psize) 1384 fi->last_disk_size = psize; 1385 spin_unlock(&fi->i_size_lock); 1386 1387 f2fs_put_rpages(cc); 1388 page_array_free(cc->inode, cc->cpages, cc->nr_cpages); 1389 cc->cpages = NULL; 1390 f2fs_destroy_compress_ctx(cc, false); 1391 return 0; 1392 1393 out_destroy_crypt: 1394 page_array_free(cc->inode, cic->rpages, cc->cluster_size); 1395 1396 for (--i; i >= 0; i--) 1397 fscrypt_finalize_bounce_page(&cc->cpages[i]); 1398 out_put_cic: 1399 kmem_cache_free(cic_entry_slab, cic); 1400 out_put_dnode: 1401 f2fs_put_dnode(&dn); 1402 out_unlock_op: 1403 if (IS_NOQUOTA(inode)) 1404 up_read(&sbi->node_write); 1405 else 1406 f2fs_unlock_op(sbi); 1407 out_free: 1408 for (i = 0; i < cc->nr_cpages; i++) { 1409 if (!cc->cpages[i]) 1410 continue; 1411 f2fs_compress_free_page(cc->cpages[i]); 1412 cc->cpages[i] = NULL; 1413 } 1414 page_array_free(cc->inode, cc->cpages, cc->nr_cpages); 1415 cc->cpages = NULL; 1416 return -EAGAIN; 1417 } 1418 1419 void f2fs_compress_write_end_io(struct bio *bio, struct page *page) 1420 { 1421 struct f2fs_sb_info *sbi = bio->bi_private; 1422 struct compress_io_ctx *cic = 1423 (struct compress_io_ctx *)page_private(page); 1424 int i; 1425 1426 if (unlikely(bio->bi_status)) 1427 mapping_set_error(cic->inode->i_mapping, -EIO); 1428 1429 f2fs_compress_free_page(page); 1430 1431 dec_page_count(sbi, F2FS_WB_DATA); 1432 1433 if (atomic_dec_return(&cic->pending_pages)) 1434 return; 1435 1436 for (i = 0; i < cic->nr_rpages; i++) { 1437 WARN_ON(!cic->rpages[i]); 1438 clear_page_private_gcing(cic->rpages[i]); 1439 end_page_writeback(cic->rpages[i]); 1440 } 1441 1442 page_array_free(cic->inode, cic->rpages, cic->nr_rpages); 1443 kmem_cache_free(cic_entry_slab, cic); 1444 } 1445 1446 static int f2fs_write_raw_pages(struct compress_ctx *cc, 1447 int *submitted, 1448 struct writeback_control *wbc, 1449 enum iostat_type io_type) 1450 { 1451 struct address_space *mapping = cc->inode->i_mapping; 1452 int _submitted, compr_blocks, ret; 1453 int i = -1, err = 0; 1454 1455 compr_blocks = f2fs_compressed_blocks(cc); 1456 if (compr_blocks < 0) { 1457 err = compr_blocks; 1458 goto out_err; 1459 } 1460 1461 for (i = 0; i < cc->cluster_size; i++) { 1462 if (!cc->rpages[i]) 1463 continue; 1464 retry_write: 1465 if (cc->rpages[i]->mapping != mapping) { 1466 unlock_page(cc->rpages[i]); 1467 continue; 1468 } 1469 1470 BUG_ON(!PageLocked(cc->rpages[i])); 1471 1472 ret = f2fs_write_single_data_page(cc->rpages[i], &_submitted, 1473 NULL, NULL, wbc, io_type, 1474 compr_blocks, false); 1475 if (ret) { 1476 if (ret == AOP_WRITEPAGE_ACTIVATE) { 1477 unlock_page(cc->rpages[i]); 1478 ret = 0; 1479 } else if (ret == -EAGAIN) { 1480 /* 1481 * for quota file, just redirty left pages to 1482 * avoid deadlock caused by cluster update race 1483 * from foreground operation. 1484 */ 1485 if (IS_NOQUOTA(cc->inode)) { 1486 err = 0; 1487 goto out_err; 1488 } 1489 ret = 0; 1490 cond_resched(); 1491 congestion_wait(BLK_RW_ASYNC, 1492 DEFAULT_IO_TIMEOUT); 1493 lock_page(cc->rpages[i]); 1494 1495 if (!PageDirty(cc->rpages[i])) { 1496 unlock_page(cc->rpages[i]); 1497 continue; 1498 } 1499 1500 clear_page_dirty_for_io(cc->rpages[i]); 1501 goto retry_write; 1502 } 1503 err = ret; 1504 goto out_err; 1505 } 1506 1507 *submitted += _submitted; 1508 } 1509 1510 f2fs_balance_fs(F2FS_M_SB(mapping), true); 1511 1512 return 0; 1513 out_err: 1514 for (++i; i < cc->cluster_size; i++) { 1515 if (!cc->rpages[i]) 1516 continue; 1517 redirty_page_for_writepage(wbc, cc->rpages[i]); 1518 unlock_page(cc->rpages[i]); 1519 } 1520 return err; 1521 } 1522 1523 int f2fs_write_multi_pages(struct compress_ctx *cc, 1524 int *submitted, 1525 struct writeback_control *wbc, 1526 enum iostat_type io_type) 1527 { 1528 int err; 1529 1530 *submitted = 0; 1531 if (cluster_may_compress(cc)) { 1532 err = f2fs_compress_pages(cc); 1533 if (err == -EAGAIN) { 1534 goto write; 1535 } else if (err) { 1536 f2fs_put_rpages_wbc(cc, wbc, true, 1); 1537 goto destroy_out; 1538 } 1539 1540 err = f2fs_write_compressed_pages(cc, submitted, 1541 wbc, io_type); 1542 if (!err) 1543 return 0; 1544 f2fs_bug_on(F2FS_I_SB(cc->inode), err != -EAGAIN); 1545 } 1546 write: 1547 f2fs_bug_on(F2FS_I_SB(cc->inode), *submitted); 1548 1549 err = f2fs_write_raw_pages(cc, submitted, wbc, io_type); 1550 f2fs_put_rpages_wbc(cc, wbc, false, 0); 1551 destroy_out: 1552 f2fs_destroy_compress_ctx(cc, false); 1553 return err; 1554 } 1555 1556 static void f2fs_free_dic(struct decompress_io_ctx *dic); 1557 1558 struct decompress_io_ctx *f2fs_alloc_dic(struct compress_ctx *cc) 1559 { 1560 struct decompress_io_ctx *dic; 1561 pgoff_t start_idx = start_idx_of_cluster(cc); 1562 int i; 1563 1564 dic = f2fs_kmem_cache_alloc(dic_entry_slab, GFP_F2FS_ZERO, 1565 false, F2FS_I_SB(cc->inode)); 1566 if (!dic) 1567 return ERR_PTR(-ENOMEM); 1568 1569 dic->rpages = page_array_alloc(cc->inode, cc->cluster_size); 1570 if (!dic->rpages) { 1571 kmem_cache_free(dic_entry_slab, dic); 1572 return ERR_PTR(-ENOMEM); 1573 } 1574 1575 dic->magic = F2FS_COMPRESSED_PAGE_MAGIC; 1576 dic->inode = cc->inode; 1577 atomic_set(&dic->remaining_pages, cc->nr_cpages); 1578 dic->cluster_idx = cc->cluster_idx; 1579 dic->cluster_size = cc->cluster_size; 1580 dic->log_cluster_size = cc->log_cluster_size; 1581 dic->nr_cpages = cc->nr_cpages; 1582 refcount_set(&dic->refcnt, 1); 1583 dic->failed = false; 1584 dic->need_verity = f2fs_need_verity(cc->inode, start_idx); 1585 1586 for (i = 0; i < dic->cluster_size; i++) 1587 dic->rpages[i] = cc->rpages[i]; 1588 dic->nr_rpages = cc->cluster_size; 1589 1590 dic->cpages = page_array_alloc(dic->inode, dic->nr_cpages); 1591 if (!dic->cpages) 1592 goto out_free; 1593 1594 for (i = 0; i < dic->nr_cpages; i++) { 1595 struct page *page; 1596 1597 page = f2fs_compress_alloc_page(); 1598 if (!page) 1599 goto out_free; 1600 1601 f2fs_set_compressed_page(page, cc->inode, 1602 start_idx + i + 1, dic); 1603 dic->cpages[i] = page; 1604 } 1605 1606 return dic; 1607 1608 out_free: 1609 f2fs_free_dic(dic); 1610 return ERR_PTR(-ENOMEM); 1611 } 1612 1613 static void f2fs_free_dic(struct decompress_io_ctx *dic) 1614 { 1615 int i; 1616 1617 if (dic->tpages) { 1618 for (i = 0; i < dic->cluster_size; i++) { 1619 if (dic->rpages[i]) 1620 continue; 1621 if (!dic->tpages[i]) 1622 continue; 1623 f2fs_compress_free_page(dic->tpages[i]); 1624 } 1625 page_array_free(dic->inode, dic->tpages, dic->cluster_size); 1626 } 1627 1628 if (dic->cpages) { 1629 for (i = 0; i < dic->nr_cpages; i++) { 1630 if (!dic->cpages[i]) 1631 continue; 1632 f2fs_compress_free_page(dic->cpages[i]); 1633 } 1634 page_array_free(dic->inode, dic->cpages, dic->nr_cpages); 1635 } 1636 1637 page_array_free(dic->inode, dic->rpages, dic->nr_rpages); 1638 kmem_cache_free(dic_entry_slab, dic); 1639 } 1640 1641 static void f2fs_put_dic(struct decompress_io_ctx *dic) 1642 { 1643 if (refcount_dec_and_test(&dic->refcnt)) 1644 f2fs_free_dic(dic); 1645 } 1646 1647 /* 1648 * Update and unlock the cluster's pagecache pages, and release the reference to 1649 * the decompress_io_ctx that was being held for I/O completion. 1650 */ 1651 static void __f2fs_decompress_end_io(struct decompress_io_ctx *dic, bool failed) 1652 { 1653 int i; 1654 1655 for (i = 0; i < dic->cluster_size; i++) { 1656 struct page *rpage = dic->rpages[i]; 1657 1658 if (!rpage) 1659 continue; 1660 1661 /* PG_error was set if verity failed. */ 1662 if (failed || PageError(rpage)) { 1663 ClearPageUptodate(rpage); 1664 /* will re-read again later */ 1665 ClearPageError(rpage); 1666 } else { 1667 SetPageUptodate(rpage); 1668 } 1669 unlock_page(rpage); 1670 } 1671 1672 f2fs_put_dic(dic); 1673 } 1674 1675 static void f2fs_verify_cluster(struct work_struct *work) 1676 { 1677 struct decompress_io_ctx *dic = 1678 container_of(work, struct decompress_io_ctx, verity_work); 1679 int i; 1680 1681 /* Verify the cluster's decompressed pages with fs-verity. */ 1682 for (i = 0; i < dic->cluster_size; i++) { 1683 struct page *rpage = dic->rpages[i]; 1684 1685 if (rpage && !fsverity_verify_page(rpage)) 1686 SetPageError(rpage); 1687 } 1688 1689 __f2fs_decompress_end_io(dic, false); 1690 } 1691 1692 /* 1693 * This is called when a compressed cluster has been decompressed 1694 * (or failed to be read and/or decompressed). 1695 */ 1696 void f2fs_decompress_end_io(struct decompress_io_ctx *dic, bool failed) 1697 { 1698 if (!failed && dic->need_verity) { 1699 /* 1700 * Note that to avoid deadlocks, the verity work can't be done 1701 * on the decompression workqueue. This is because verifying 1702 * the data pages can involve reading metadata pages from the 1703 * file, and these metadata pages may be compressed. 1704 */ 1705 INIT_WORK(&dic->verity_work, f2fs_verify_cluster); 1706 fsverity_enqueue_verify_work(&dic->verity_work); 1707 } else { 1708 __f2fs_decompress_end_io(dic, failed); 1709 } 1710 } 1711 1712 /* 1713 * Put a reference to a compressed page's decompress_io_ctx. 1714 * 1715 * This is called when the page is no longer needed and can be freed. 1716 */ 1717 void f2fs_put_page_dic(struct page *page) 1718 { 1719 struct decompress_io_ctx *dic = 1720 (struct decompress_io_ctx *)page_private(page); 1721 1722 f2fs_put_dic(dic); 1723 } 1724 1725 /* 1726 * check whether cluster blocks are contiguous, and add extent cache entry 1727 * only if cluster blocks are logically and physically contiguous. 1728 */ 1729 unsigned int f2fs_cluster_blocks_are_contiguous(struct dnode_of_data *dn) 1730 { 1731 bool compressed = f2fs_data_blkaddr(dn) == COMPRESS_ADDR; 1732 int i = compressed ? 1 : 0; 1733 block_t first_blkaddr = data_blkaddr(dn->inode, dn->node_page, 1734 dn->ofs_in_node + i); 1735 1736 for (i += 1; i < F2FS_I(dn->inode)->i_cluster_size; i++) { 1737 block_t blkaddr = data_blkaddr(dn->inode, dn->node_page, 1738 dn->ofs_in_node + i); 1739 1740 if (!__is_valid_data_blkaddr(blkaddr)) 1741 break; 1742 if (first_blkaddr + i - (compressed ? 1 : 0) != blkaddr) 1743 return 0; 1744 } 1745 1746 return compressed ? i - 1 : i; 1747 } 1748 1749 const struct address_space_operations f2fs_compress_aops = { 1750 .releasepage = f2fs_release_page, 1751 .invalidatepage = f2fs_invalidate_page, 1752 }; 1753 1754 struct address_space *COMPRESS_MAPPING(struct f2fs_sb_info *sbi) 1755 { 1756 return sbi->compress_inode->i_mapping; 1757 } 1758 1759 void f2fs_invalidate_compress_page(struct f2fs_sb_info *sbi, block_t blkaddr) 1760 { 1761 if (!sbi->compress_inode) 1762 return; 1763 invalidate_mapping_pages(COMPRESS_MAPPING(sbi), blkaddr, blkaddr); 1764 } 1765 1766 void f2fs_cache_compressed_page(struct f2fs_sb_info *sbi, struct page *page, 1767 nid_t ino, block_t blkaddr) 1768 { 1769 struct page *cpage; 1770 int ret; 1771 1772 if (!test_opt(sbi, COMPRESS_CACHE)) 1773 return; 1774 1775 if (!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE_READ)) 1776 return; 1777 1778 if (!f2fs_available_free_memory(sbi, COMPRESS_PAGE)) 1779 return; 1780 1781 cpage = find_get_page(COMPRESS_MAPPING(sbi), blkaddr); 1782 if (cpage) { 1783 f2fs_put_page(cpage, 0); 1784 return; 1785 } 1786 1787 cpage = alloc_page(__GFP_NOWARN | __GFP_IO); 1788 if (!cpage) 1789 return; 1790 1791 ret = add_to_page_cache_lru(cpage, COMPRESS_MAPPING(sbi), 1792 blkaddr, GFP_NOFS); 1793 if (ret) { 1794 f2fs_put_page(cpage, 0); 1795 return; 1796 } 1797 1798 set_page_private_data(cpage, ino); 1799 1800 if (!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE_READ)) 1801 goto out; 1802 1803 memcpy(page_address(cpage), page_address(page), PAGE_SIZE); 1804 SetPageUptodate(cpage); 1805 out: 1806 f2fs_put_page(cpage, 1); 1807 } 1808 1809 bool f2fs_load_compressed_page(struct f2fs_sb_info *sbi, struct page *page, 1810 block_t blkaddr) 1811 { 1812 struct page *cpage; 1813 bool hitted = false; 1814 1815 if (!test_opt(sbi, COMPRESS_CACHE)) 1816 return false; 1817 1818 cpage = f2fs_pagecache_get_page(COMPRESS_MAPPING(sbi), 1819 blkaddr, FGP_LOCK | FGP_NOWAIT, GFP_NOFS); 1820 if (cpage) { 1821 if (PageUptodate(cpage)) { 1822 atomic_inc(&sbi->compress_page_hit); 1823 memcpy(page_address(page), 1824 page_address(cpage), PAGE_SIZE); 1825 hitted = true; 1826 } 1827 f2fs_put_page(cpage, 1); 1828 } 1829 1830 return hitted; 1831 } 1832 1833 void f2fs_invalidate_compress_pages(struct f2fs_sb_info *sbi, nid_t ino) 1834 { 1835 struct address_space *mapping = sbi->compress_inode->i_mapping; 1836 struct pagevec pvec; 1837 pgoff_t index = 0; 1838 pgoff_t end = MAX_BLKADDR(sbi); 1839 1840 if (!mapping->nrpages) 1841 return; 1842 1843 pagevec_init(&pvec); 1844 1845 do { 1846 unsigned int nr_pages; 1847 int i; 1848 1849 nr_pages = pagevec_lookup_range(&pvec, mapping, 1850 &index, end - 1); 1851 if (!nr_pages) 1852 break; 1853 1854 for (i = 0; i < nr_pages; i++) { 1855 struct page *page = pvec.pages[i]; 1856 1857 if (page->index > end) 1858 break; 1859 1860 lock_page(page); 1861 if (page->mapping != mapping) { 1862 unlock_page(page); 1863 continue; 1864 } 1865 1866 if (ino != get_page_private_data(page)) { 1867 unlock_page(page); 1868 continue; 1869 } 1870 1871 generic_error_remove_page(mapping, page); 1872 unlock_page(page); 1873 } 1874 pagevec_release(&pvec); 1875 cond_resched(); 1876 } while (index < end); 1877 } 1878 1879 int f2fs_init_compress_inode(struct f2fs_sb_info *sbi) 1880 { 1881 struct inode *inode; 1882 1883 if (!test_opt(sbi, COMPRESS_CACHE)) 1884 return 0; 1885 1886 inode = f2fs_iget(sbi->sb, F2FS_COMPRESS_INO(sbi)); 1887 if (IS_ERR(inode)) 1888 return PTR_ERR(inode); 1889 sbi->compress_inode = inode; 1890 1891 sbi->compress_percent = COMPRESS_PERCENT; 1892 sbi->compress_watermark = COMPRESS_WATERMARK; 1893 1894 atomic_set(&sbi->compress_page_hit, 0); 1895 1896 return 0; 1897 } 1898 1899 void f2fs_destroy_compress_inode(struct f2fs_sb_info *sbi) 1900 { 1901 if (!sbi->compress_inode) 1902 return; 1903 iput(sbi->compress_inode); 1904 sbi->compress_inode = NULL; 1905 } 1906 1907 int f2fs_init_page_array_cache(struct f2fs_sb_info *sbi) 1908 { 1909 dev_t dev = sbi->sb->s_bdev->bd_dev; 1910 char slab_name[32]; 1911 1912 sprintf(slab_name, "f2fs_page_array_entry-%u:%u", MAJOR(dev), MINOR(dev)); 1913 1914 sbi->page_array_slab_size = sizeof(struct page *) << 1915 F2FS_OPTION(sbi).compress_log_size; 1916 1917 sbi->page_array_slab = f2fs_kmem_cache_create(slab_name, 1918 sbi->page_array_slab_size); 1919 if (!sbi->page_array_slab) 1920 return -ENOMEM; 1921 return 0; 1922 } 1923 1924 void f2fs_destroy_page_array_cache(struct f2fs_sb_info *sbi) 1925 { 1926 kmem_cache_destroy(sbi->page_array_slab); 1927 } 1928 1929 static int __init f2fs_init_cic_cache(void) 1930 { 1931 cic_entry_slab = f2fs_kmem_cache_create("f2fs_cic_entry", 1932 sizeof(struct compress_io_ctx)); 1933 if (!cic_entry_slab) 1934 return -ENOMEM; 1935 return 0; 1936 } 1937 1938 static void f2fs_destroy_cic_cache(void) 1939 { 1940 kmem_cache_destroy(cic_entry_slab); 1941 } 1942 1943 static int __init f2fs_init_dic_cache(void) 1944 { 1945 dic_entry_slab = f2fs_kmem_cache_create("f2fs_dic_entry", 1946 sizeof(struct decompress_io_ctx)); 1947 if (!dic_entry_slab) 1948 return -ENOMEM; 1949 return 0; 1950 } 1951 1952 static void f2fs_destroy_dic_cache(void) 1953 { 1954 kmem_cache_destroy(dic_entry_slab); 1955 } 1956 1957 int __init f2fs_init_compress_cache(void) 1958 { 1959 int err; 1960 1961 err = f2fs_init_cic_cache(); 1962 if (err) 1963 goto out; 1964 err = f2fs_init_dic_cache(); 1965 if (err) 1966 goto free_cic; 1967 return 0; 1968 free_cic: 1969 f2fs_destroy_cic_cache(); 1970 out: 1971 return -ENOMEM; 1972 } 1973 1974 void f2fs_destroy_compress_cache(void) 1975 { 1976 f2fs_destroy_dic_cache(); 1977 f2fs_destroy_cic_cache(); 1978 } 1979