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 cc->valid_nr_cpages = 0; 158 if (!reuse) 159 cc->cluster_idx = NULL_CLUSTER; 160 } 161 162 void f2fs_compress_ctx_add_page(struct compress_ctx *cc, struct page *page) 163 { 164 unsigned int cluster_ofs; 165 166 if (!f2fs_cluster_can_merge_page(cc, page->index)) 167 f2fs_bug_on(F2FS_I_SB(cc->inode), 1); 168 169 cluster_ofs = offset_in_cluster(cc, page->index); 170 cc->rpages[cluster_ofs] = page; 171 cc->nr_rpages++; 172 cc->cluster_idx = cluster_idx(cc, page->index); 173 } 174 175 #ifdef CONFIG_F2FS_FS_LZO 176 static int lzo_init_compress_ctx(struct compress_ctx *cc) 177 { 178 cc->private = f2fs_kvmalloc(F2FS_I_SB(cc->inode), 179 LZO1X_MEM_COMPRESS, GFP_NOFS); 180 if (!cc->private) 181 return -ENOMEM; 182 183 cc->clen = lzo1x_worst_compress(PAGE_SIZE << cc->log_cluster_size); 184 return 0; 185 } 186 187 static void lzo_destroy_compress_ctx(struct compress_ctx *cc) 188 { 189 kvfree(cc->private); 190 cc->private = NULL; 191 } 192 193 static int lzo_compress_pages(struct compress_ctx *cc) 194 { 195 int ret; 196 197 ret = lzo1x_1_compress(cc->rbuf, cc->rlen, cc->cbuf->cdata, 198 &cc->clen, cc->private); 199 if (ret != LZO_E_OK) { 200 printk_ratelimited("%sF2FS-fs (%s): lzo compress failed, ret:%d\n", 201 KERN_ERR, F2FS_I_SB(cc->inode)->sb->s_id, ret); 202 return -EIO; 203 } 204 return 0; 205 } 206 207 static int lzo_decompress_pages(struct decompress_io_ctx *dic) 208 { 209 int ret; 210 211 ret = lzo1x_decompress_safe(dic->cbuf->cdata, dic->clen, 212 dic->rbuf, &dic->rlen); 213 if (ret != LZO_E_OK) { 214 printk_ratelimited("%sF2FS-fs (%s): lzo decompress failed, ret:%d\n", 215 KERN_ERR, F2FS_I_SB(dic->inode)->sb->s_id, ret); 216 return -EIO; 217 } 218 219 if (dic->rlen != PAGE_SIZE << dic->log_cluster_size) { 220 printk_ratelimited("%sF2FS-fs (%s): lzo invalid rlen:%zu, " 221 "expected:%lu\n", KERN_ERR, 222 F2FS_I_SB(dic->inode)->sb->s_id, 223 dic->rlen, 224 PAGE_SIZE << dic->log_cluster_size); 225 return -EIO; 226 } 227 return 0; 228 } 229 230 static const struct f2fs_compress_ops f2fs_lzo_ops = { 231 .init_compress_ctx = lzo_init_compress_ctx, 232 .destroy_compress_ctx = lzo_destroy_compress_ctx, 233 .compress_pages = lzo_compress_pages, 234 .decompress_pages = lzo_decompress_pages, 235 }; 236 #endif 237 238 #ifdef CONFIG_F2FS_FS_LZ4 239 static int lz4_init_compress_ctx(struct compress_ctx *cc) 240 { 241 unsigned int size = LZ4_MEM_COMPRESS; 242 243 #ifdef CONFIG_F2FS_FS_LZ4HC 244 if (F2FS_I(cc->inode)->i_compress_flag >> COMPRESS_LEVEL_OFFSET) 245 size = LZ4HC_MEM_COMPRESS; 246 #endif 247 248 cc->private = f2fs_kvmalloc(F2FS_I_SB(cc->inode), size, GFP_NOFS); 249 if (!cc->private) 250 return -ENOMEM; 251 252 /* 253 * we do not change cc->clen to LZ4_compressBound(inputsize) to 254 * adapt worst compress case, because lz4 compressor can handle 255 * output budget properly. 256 */ 257 cc->clen = cc->rlen - PAGE_SIZE - COMPRESS_HEADER_SIZE; 258 return 0; 259 } 260 261 static void lz4_destroy_compress_ctx(struct compress_ctx *cc) 262 { 263 kvfree(cc->private); 264 cc->private = NULL; 265 } 266 267 #ifdef CONFIG_F2FS_FS_LZ4HC 268 static int lz4hc_compress_pages(struct compress_ctx *cc) 269 { 270 unsigned char level = F2FS_I(cc->inode)->i_compress_flag >> 271 COMPRESS_LEVEL_OFFSET; 272 int len; 273 274 if (level) 275 len = LZ4_compress_HC(cc->rbuf, cc->cbuf->cdata, cc->rlen, 276 cc->clen, level, cc->private); 277 else 278 len = LZ4_compress_default(cc->rbuf, cc->cbuf->cdata, cc->rlen, 279 cc->clen, cc->private); 280 if (!len) 281 return -EAGAIN; 282 283 cc->clen = len; 284 return 0; 285 } 286 #endif 287 288 static int lz4_compress_pages(struct compress_ctx *cc) 289 { 290 int len; 291 292 #ifdef CONFIG_F2FS_FS_LZ4HC 293 return lz4hc_compress_pages(cc); 294 #endif 295 len = LZ4_compress_default(cc->rbuf, cc->cbuf->cdata, cc->rlen, 296 cc->clen, cc->private); 297 if (!len) 298 return -EAGAIN; 299 300 cc->clen = len; 301 return 0; 302 } 303 304 static int lz4_decompress_pages(struct decompress_io_ctx *dic) 305 { 306 int ret; 307 308 ret = LZ4_decompress_safe(dic->cbuf->cdata, dic->rbuf, 309 dic->clen, dic->rlen); 310 if (ret < 0) { 311 printk_ratelimited("%sF2FS-fs (%s): lz4 decompress failed, ret:%d\n", 312 KERN_ERR, F2FS_I_SB(dic->inode)->sb->s_id, ret); 313 return -EIO; 314 } 315 316 if (ret != PAGE_SIZE << dic->log_cluster_size) { 317 printk_ratelimited("%sF2FS-fs (%s): lz4 invalid ret:%d, " 318 "expected:%lu\n", KERN_ERR, 319 F2FS_I_SB(dic->inode)->sb->s_id, ret, 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_get_params(F2FS_ZSTD_DEFAULT_CLEVEL, cc->rlen); 350 workspace_size = zstd_cstream_workspace_bound(¶ms.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_init_cstream(¶ms, 0, workspace, workspace_size); 358 if (!stream) { 359 printk_ratelimited("%sF2FS-fs (%s): %s zstd_init_cstream 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_in_buffer inbuf; 384 zstd_out_buffer 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_compress_stream(stream, &outbuf, &inbuf); 398 if (zstd_is_error(ret)) { 399 printk_ratelimited("%sF2FS-fs (%s): %s zstd_compress_stream failed, ret: %d\n", 400 KERN_ERR, F2FS_I_SB(cc->inode)->sb->s_id, 401 __func__, zstd_get_error_code(ret)); 402 return -EIO; 403 } 404 405 ret = zstd_end_stream(stream, &outbuf); 406 if (zstd_is_error(ret)) { 407 printk_ratelimited("%sF2FS-fs (%s): %s zstd_end_stream returned %d\n", 408 KERN_ERR, F2FS_I_SB(cc->inode)->sb->s_id, 409 __func__, zstd_get_error_code(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_dstream_workspace_bound(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_init_dstream(max_window_size, workspace, workspace_size); 440 if (!stream) { 441 printk_ratelimited("%sF2FS-fs (%s): %s zstd_init_dstream 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_in_buffer inbuf; 465 zstd_out_buffer 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_decompress_stream(stream, &outbuf, &inbuf); 477 if (zstd_is_error(ret)) { 478 printk_ratelimited("%sF2FS-fs (%s): %s zstd_decompress_stream failed, ret: %d\n", 479 KERN_ERR, F2FS_I_SB(dic->inode)->sb->s_id, 480 __func__, zstd_get_error_code(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 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 cc->valid_nr_cpages = cc->nr_cpages; 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 /* zero out any unused part of the last page */ 689 memset(&cc->cbuf->cdata[cc->clen], 0, 690 (new_nr_cpages * PAGE_SIZE) - 691 (cc->clen + COMPRESS_HEADER_SIZE)); 692 693 vm_unmap_ram(cc->cbuf, cc->nr_cpages); 694 vm_unmap_ram(cc->rbuf, cc->cluster_size); 695 696 for (i = 0; i < cc->nr_cpages; i++) { 697 if (i < new_nr_cpages) 698 continue; 699 f2fs_compress_free_page(cc->cpages[i]); 700 cc->cpages[i] = NULL; 701 } 702 703 if (cops->destroy_compress_ctx) 704 cops->destroy_compress_ctx(cc); 705 706 cc->valid_nr_cpages = new_nr_cpages; 707 708 trace_f2fs_compress_pages_end(cc->inode, cc->cluster_idx, 709 cc->clen, ret); 710 return 0; 711 712 out_vunmap_cbuf: 713 vm_unmap_ram(cc->cbuf, cc->nr_cpages); 714 out_vunmap_rbuf: 715 vm_unmap_ram(cc->rbuf, cc->cluster_size); 716 out_free_cpages: 717 for (i = 0; i < cc->nr_cpages; i++) { 718 if (cc->cpages[i]) 719 f2fs_compress_free_page(cc->cpages[i]); 720 } 721 page_array_free(cc->inode, cc->cpages, cc->nr_cpages); 722 cc->cpages = NULL; 723 destroy_compress_ctx: 724 if (cops->destroy_compress_ctx) 725 cops->destroy_compress_ctx(cc); 726 out: 727 trace_f2fs_compress_pages_end(cc->inode, cc->cluster_idx, 728 cc->clen, ret); 729 return ret; 730 } 731 732 void f2fs_decompress_cluster(struct decompress_io_ctx *dic) 733 { 734 struct f2fs_sb_info *sbi = F2FS_I_SB(dic->inode); 735 struct f2fs_inode_info *fi = F2FS_I(dic->inode); 736 const struct f2fs_compress_ops *cops = 737 f2fs_cops[fi->i_compress_algorithm]; 738 int ret; 739 int i; 740 741 trace_f2fs_decompress_pages_start(dic->inode, dic->cluster_idx, 742 dic->cluster_size, fi->i_compress_algorithm); 743 744 if (dic->failed) { 745 ret = -EIO; 746 goto out_end_io; 747 } 748 749 dic->tpages = page_array_alloc(dic->inode, dic->cluster_size); 750 if (!dic->tpages) { 751 ret = -ENOMEM; 752 goto out_end_io; 753 } 754 755 for (i = 0; i < dic->cluster_size; i++) { 756 if (dic->rpages[i]) { 757 dic->tpages[i] = dic->rpages[i]; 758 continue; 759 } 760 761 dic->tpages[i] = f2fs_compress_alloc_page(); 762 if (!dic->tpages[i]) { 763 ret = -ENOMEM; 764 goto out_end_io; 765 } 766 } 767 768 if (cops->init_decompress_ctx) { 769 ret = cops->init_decompress_ctx(dic); 770 if (ret) 771 goto out_end_io; 772 } 773 774 dic->rbuf = f2fs_vmap(dic->tpages, dic->cluster_size); 775 if (!dic->rbuf) { 776 ret = -ENOMEM; 777 goto out_destroy_decompress_ctx; 778 } 779 780 dic->cbuf = f2fs_vmap(dic->cpages, dic->nr_cpages); 781 if (!dic->cbuf) { 782 ret = -ENOMEM; 783 goto out_vunmap_rbuf; 784 } 785 786 dic->clen = le32_to_cpu(dic->cbuf->clen); 787 dic->rlen = PAGE_SIZE << dic->log_cluster_size; 788 789 if (dic->clen > PAGE_SIZE * dic->nr_cpages - COMPRESS_HEADER_SIZE) { 790 ret = -EFSCORRUPTED; 791 goto out_vunmap_cbuf; 792 } 793 794 ret = cops->decompress_pages(dic); 795 796 if (!ret && (fi->i_compress_flag & 1 << COMPRESS_CHKSUM)) { 797 u32 provided = le32_to_cpu(dic->cbuf->chksum); 798 u32 calculated = f2fs_crc32(sbi, dic->cbuf->cdata, dic->clen); 799 800 if (provided != calculated) { 801 if (!is_inode_flag_set(dic->inode, FI_COMPRESS_CORRUPT)) { 802 set_inode_flag(dic->inode, FI_COMPRESS_CORRUPT); 803 printk_ratelimited( 804 "%sF2FS-fs (%s): checksum invalid, nid = %lu, %x vs %x", 805 KERN_INFO, sbi->sb->s_id, dic->inode->i_ino, 806 provided, calculated); 807 } 808 set_sbi_flag(sbi, SBI_NEED_FSCK); 809 } 810 } 811 812 out_vunmap_cbuf: 813 vm_unmap_ram(dic->cbuf, dic->nr_cpages); 814 out_vunmap_rbuf: 815 vm_unmap_ram(dic->rbuf, dic->cluster_size); 816 out_destroy_decompress_ctx: 817 if (cops->destroy_decompress_ctx) 818 cops->destroy_decompress_ctx(dic); 819 out_end_io: 820 trace_f2fs_decompress_pages_end(dic->inode, dic->cluster_idx, 821 dic->clen, ret); 822 f2fs_decompress_end_io(dic, ret); 823 } 824 825 /* 826 * This is called when a page of a compressed cluster has been read from disk 827 * (or failed to be read from disk). It checks whether this page was the last 828 * page being waited on in the cluster, and if so, it decompresses the cluster 829 * (or in the case of a failure, cleans up without actually decompressing). 830 */ 831 void f2fs_end_read_compressed_page(struct page *page, bool failed, 832 block_t blkaddr) 833 { 834 struct decompress_io_ctx *dic = 835 (struct decompress_io_ctx *)page_private(page); 836 struct f2fs_sb_info *sbi = F2FS_I_SB(dic->inode); 837 838 dec_page_count(sbi, F2FS_RD_DATA); 839 840 if (failed) 841 WRITE_ONCE(dic->failed, true); 842 else if (blkaddr) 843 f2fs_cache_compressed_page(sbi, page, 844 dic->inode->i_ino, blkaddr); 845 846 if (atomic_dec_and_test(&dic->remaining_pages)) 847 f2fs_decompress_cluster(dic); 848 } 849 850 static bool is_page_in_cluster(struct compress_ctx *cc, pgoff_t index) 851 { 852 if (cc->cluster_idx == NULL_CLUSTER) 853 return true; 854 return cc->cluster_idx == cluster_idx(cc, index); 855 } 856 857 bool f2fs_cluster_is_empty(struct compress_ctx *cc) 858 { 859 return cc->nr_rpages == 0; 860 } 861 862 static bool f2fs_cluster_is_full(struct compress_ctx *cc) 863 { 864 return cc->cluster_size == cc->nr_rpages; 865 } 866 867 bool f2fs_cluster_can_merge_page(struct compress_ctx *cc, pgoff_t index) 868 { 869 if (f2fs_cluster_is_empty(cc)) 870 return true; 871 return is_page_in_cluster(cc, index); 872 } 873 874 bool f2fs_all_cluster_page_loaded(struct compress_ctx *cc, struct pagevec *pvec, 875 int index, int nr_pages) 876 { 877 unsigned long pgidx; 878 int i; 879 880 if (nr_pages - index < cc->cluster_size) 881 return false; 882 883 pgidx = pvec->pages[index]->index; 884 885 for (i = 1; i < cc->cluster_size; i++) { 886 if (pvec->pages[index + i]->index != pgidx + i) 887 return false; 888 } 889 890 return true; 891 } 892 893 static bool cluster_has_invalid_data(struct compress_ctx *cc) 894 { 895 loff_t i_size = i_size_read(cc->inode); 896 unsigned nr_pages = DIV_ROUND_UP(i_size, PAGE_SIZE); 897 int i; 898 899 for (i = 0; i < cc->cluster_size; i++) { 900 struct page *page = cc->rpages[i]; 901 902 f2fs_bug_on(F2FS_I_SB(cc->inode), !page); 903 904 /* beyond EOF */ 905 if (page->index >= nr_pages) 906 return true; 907 } 908 return false; 909 } 910 911 bool f2fs_sanity_check_cluster(struct dnode_of_data *dn) 912 { 913 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode); 914 unsigned int cluster_size = F2FS_I(dn->inode)->i_cluster_size; 915 bool compressed = dn->data_blkaddr == COMPRESS_ADDR; 916 int cluster_end = 0; 917 int i; 918 char *reason = ""; 919 920 if (!compressed) 921 return false; 922 923 /* [..., COMPR_ADDR, ...] */ 924 if (dn->ofs_in_node % cluster_size) { 925 reason = "[*|C|*|*]"; 926 goto out; 927 } 928 929 for (i = 1; i < cluster_size; i++) { 930 block_t blkaddr = data_blkaddr(dn->inode, dn->node_page, 931 dn->ofs_in_node + i); 932 933 /* [COMPR_ADDR, ..., COMPR_ADDR] */ 934 if (blkaddr == COMPRESS_ADDR) { 935 reason = "[C|*|C|*]"; 936 goto out; 937 } 938 if (compressed) { 939 if (!__is_valid_data_blkaddr(blkaddr)) { 940 if (!cluster_end) 941 cluster_end = i; 942 continue; 943 } 944 /* [COMPR_ADDR, NULL_ADDR or NEW_ADDR, valid_blkaddr] */ 945 if (cluster_end) { 946 reason = "[C|N|N|V]"; 947 goto out; 948 } 949 } 950 } 951 return false; 952 out: 953 f2fs_warn(sbi, "access invalid cluster, ino:%lu, nid:%u, ofs_in_node:%u, reason:%s", 954 dn->inode->i_ino, dn->nid, dn->ofs_in_node, reason); 955 set_sbi_flag(sbi, SBI_NEED_FSCK); 956 return true; 957 } 958 959 static int __f2fs_cluster_blocks(struct inode *inode, 960 unsigned int cluster_idx, bool compr) 961 { 962 struct dnode_of_data dn; 963 unsigned int cluster_size = F2FS_I(inode)->i_cluster_size; 964 unsigned int start_idx = cluster_idx << 965 F2FS_I(inode)->i_log_cluster_size; 966 int ret; 967 968 set_new_dnode(&dn, inode, NULL, NULL, 0); 969 ret = f2fs_get_dnode_of_data(&dn, start_idx, LOOKUP_NODE); 970 if (ret) { 971 if (ret == -ENOENT) 972 ret = 0; 973 goto fail; 974 } 975 976 if (f2fs_sanity_check_cluster(&dn)) { 977 ret = -EFSCORRUPTED; 978 goto fail; 979 } 980 981 if (dn.data_blkaddr == COMPRESS_ADDR) { 982 int i; 983 984 ret = 1; 985 for (i = 1; i < cluster_size; i++) { 986 block_t blkaddr; 987 988 blkaddr = data_blkaddr(dn.inode, 989 dn.node_page, dn.ofs_in_node + i); 990 if (compr) { 991 if (__is_valid_data_blkaddr(blkaddr)) 992 ret++; 993 } else { 994 if (blkaddr != NULL_ADDR) 995 ret++; 996 } 997 } 998 999 f2fs_bug_on(F2FS_I_SB(inode), 1000 !compr && ret != cluster_size && 1001 !is_inode_flag_set(inode, FI_COMPRESS_RELEASED)); 1002 } 1003 fail: 1004 f2fs_put_dnode(&dn); 1005 return ret; 1006 } 1007 1008 /* return # of compressed blocks in compressed cluster */ 1009 static int f2fs_compressed_blocks(struct compress_ctx *cc) 1010 { 1011 return __f2fs_cluster_blocks(cc->inode, cc->cluster_idx, true); 1012 } 1013 1014 /* return # of valid blocks in compressed cluster */ 1015 int f2fs_is_compressed_cluster(struct inode *inode, pgoff_t index) 1016 { 1017 return __f2fs_cluster_blocks(inode, 1018 index >> F2FS_I(inode)->i_log_cluster_size, 1019 false); 1020 } 1021 1022 static bool cluster_may_compress(struct compress_ctx *cc) 1023 { 1024 if (!f2fs_need_compress_data(cc->inode)) 1025 return false; 1026 if (f2fs_is_atomic_file(cc->inode)) 1027 return false; 1028 if (!f2fs_cluster_is_full(cc)) 1029 return false; 1030 if (unlikely(f2fs_cp_error(F2FS_I_SB(cc->inode)))) 1031 return false; 1032 return !cluster_has_invalid_data(cc); 1033 } 1034 1035 static void set_cluster_writeback(struct compress_ctx *cc) 1036 { 1037 int i; 1038 1039 for (i = 0; i < cc->cluster_size; i++) { 1040 if (cc->rpages[i]) 1041 set_page_writeback(cc->rpages[i]); 1042 } 1043 } 1044 1045 static void set_cluster_dirty(struct compress_ctx *cc) 1046 { 1047 int i; 1048 1049 for (i = 0; i < cc->cluster_size; i++) 1050 if (cc->rpages[i]) 1051 set_page_dirty(cc->rpages[i]); 1052 } 1053 1054 static int prepare_compress_overwrite(struct compress_ctx *cc, 1055 struct page **pagep, pgoff_t index, void **fsdata) 1056 { 1057 struct f2fs_sb_info *sbi = F2FS_I_SB(cc->inode); 1058 struct address_space *mapping = cc->inode->i_mapping; 1059 struct page *page; 1060 sector_t last_block_in_bio; 1061 unsigned fgp_flag = FGP_LOCK | FGP_WRITE | FGP_CREAT; 1062 pgoff_t start_idx = start_idx_of_cluster(cc); 1063 int i, ret; 1064 1065 retry: 1066 ret = f2fs_is_compressed_cluster(cc->inode, start_idx); 1067 if (ret <= 0) 1068 return ret; 1069 1070 ret = f2fs_init_compress_ctx(cc); 1071 if (ret) 1072 return ret; 1073 1074 /* keep page reference to avoid page reclaim */ 1075 for (i = 0; i < cc->cluster_size; i++) { 1076 page = f2fs_pagecache_get_page(mapping, start_idx + i, 1077 fgp_flag, GFP_NOFS); 1078 if (!page) { 1079 ret = -ENOMEM; 1080 goto unlock_pages; 1081 } 1082 1083 if (PageUptodate(page)) 1084 f2fs_put_page(page, 1); 1085 else 1086 f2fs_compress_ctx_add_page(cc, page); 1087 } 1088 1089 if (!f2fs_cluster_is_empty(cc)) { 1090 struct bio *bio = NULL; 1091 1092 ret = f2fs_read_multi_pages(cc, &bio, cc->cluster_size, 1093 &last_block_in_bio, false, true); 1094 f2fs_put_rpages(cc); 1095 f2fs_destroy_compress_ctx(cc, true); 1096 if (ret) 1097 goto out; 1098 if (bio) 1099 f2fs_submit_bio(sbi, bio, DATA); 1100 1101 ret = f2fs_init_compress_ctx(cc); 1102 if (ret) 1103 goto out; 1104 } 1105 1106 for (i = 0; i < cc->cluster_size; i++) { 1107 f2fs_bug_on(sbi, cc->rpages[i]); 1108 1109 page = find_lock_page(mapping, start_idx + i); 1110 if (!page) { 1111 /* page can be truncated */ 1112 goto release_and_retry; 1113 } 1114 1115 f2fs_wait_on_page_writeback(page, DATA, true, true); 1116 f2fs_compress_ctx_add_page(cc, page); 1117 1118 if (!PageUptodate(page)) { 1119 release_and_retry: 1120 f2fs_put_rpages(cc); 1121 f2fs_unlock_rpages(cc, i + 1); 1122 f2fs_destroy_compress_ctx(cc, true); 1123 goto retry; 1124 } 1125 } 1126 1127 if (likely(!ret)) { 1128 *fsdata = cc->rpages; 1129 *pagep = cc->rpages[offset_in_cluster(cc, index)]; 1130 return cc->cluster_size; 1131 } 1132 1133 unlock_pages: 1134 f2fs_put_rpages(cc); 1135 f2fs_unlock_rpages(cc, i); 1136 f2fs_destroy_compress_ctx(cc, true); 1137 out: 1138 return ret; 1139 } 1140 1141 int f2fs_prepare_compress_overwrite(struct inode *inode, 1142 struct page **pagep, pgoff_t index, void **fsdata) 1143 { 1144 struct compress_ctx cc = { 1145 .inode = inode, 1146 .log_cluster_size = F2FS_I(inode)->i_log_cluster_size, 1147 .cluster_size = F2FS_I(inode)->i_cluster_size, 1148 .cluster_idx = index >> F2FS_I(inode)->i_log_cluster_size, 1149 .rpages = NULL, 1150 .nr_rpages = 0, 1151 }; 1152 1153 return prepare_compress_overwrite(&cc, pagep, index, fsdata); 1154 } 1155 1156 bool f2fs_compress_write_end(struct inode *inode, void *fsdata, 1157 pgoff_t index, unsigned copied) 1158 1159 { 1160 struct compress_ctx cc = { 1161 .inode = inode, 1162 .log_cluster_size = F2FS_I(inode)->i_log_cluster_size, 1163 .cluster_size = F2FS_I(inode)->i_cluster_size, 1164 .rpages = fsdata, 1165 }; 1166 bool first_index = (index == cc.rpages[0]->index); 1167 1168 if (copied) 1169 set_cluster_dirty(&cc); 1170 1171 f2fs_put_rpages_wbc(&cc, NULL, false, 1); 1172 f2fs_destroy_compress_ctx(&cc, false); 1173 1174 return first_index; 1175 } 1176 1177 int f2fs_truncate_partial_cluster(struct inode *inode, u64 from, bool lock) 1178 { 1179 void *fsdata = NULL; 1180 struct page *pagep; 1181 int log_cluster_size = F2FS_I(inode)->i_log_cluster_size; 1182 pgoff_t start_idx = from >> (PAGE_SHIFT + log_cluster_size) << 1183 log_cluster_size; 1184 int err; 1185 1186 err = f2fs_is_compressed_cluster(inode, start_idx); 1187 if (err < 0) 1188 return err; 1189 1190 /* truncate normal cluster */ 1191 if (!err) 1192 return f2fs_do_truncate_blocks(inode, from, lock); 1193 1194 /* truncate compressed cluster */ 1195 err = f2fs_prepare_compress_overwrite(inode, &pagep, 1196 start_idx, &fsdata); 1197 1198 /* should not be a normal cluster */ 1199 f2fs_bug_on(F2FS_I_SB(inode), err == 0); 1200 1201 if (err <= 0) 1202 return err; 1203 1204 if (err > 0) { 1205 struct page **rpages = fsdata; 1206 int cluster_size = F2FS_I(inode)->i_cluster_size; 1207 int i; 1208 1209 for (i = cluster_size - 1; i >= 0; i--) { 1210 loff_t start = rpages[i]->index << PAGE_SHIFT; 1211 1212 if (from <= start) { 1213 zero_user_segment(rpages[i], 0, PAGE_SIZE); 1214 } else { 1215 zero_user_segment(rpages[i], from - start, 1216 PAGE_SIZE); 1217 break; 1218 } 1219 } 1220 1221 f2fs_compress_write_end(inode, fsdata, start_idx, true); 1222 } 1223 return 0; 1224 } 1225 1226 static int f2fs_write_compressed_pages(struct compress_ctx *cc, 1227 int *submitted, 1228 struct writeback_control *wbc, 1229 enum iostat_type io_type) 1230 { 1231 struct inode *inode = cc->inode; 1232 struct f2fs_sb_info *sbi = F2FS_I_SB(inode); 1233 struct f2fs_inode_info *fi = F2FS_I(inode); 1234 struct f2fs_io_info fio = { 1235 .sbi = sbi, 1236 .ino = cc->inode->i_ino, 1237 .type = DATA, 1238 .op = REQ_OP_WRITE, 1239 .op_flags = wbc_to_write_flags(wbc), 1240 .old_blkaddr = NEW_ADDR, 1241 .page = NULL, 1242 .encrypted_page = NULL, 1243 .compressed_page = NULL, 1244 .submitted = false, 1245 .io_type = io_type, 1246 .io_wbc = wbc, 1247 .encrypted = fscrypt_inode_uses_fs_layer_crypto(cc->inode), 1248 }; 1249 struct dnode_of_data dn; 1250 struct node_info ni; 1251 struct compress_io_ctx *cic; 1252 pgoff_t start_idx = start_idx_of_cluster(cc); 1253 unsigned int last_index = cc->cluster_size - 1; 1254 loff_t psize; 1255 int i, err; 1256 1257 /* we should bypass data pages to proceed the kworkder jobs */ 1258 if (unlikely(f2fs_cp_error(sbi))) { 1259 mapping_set_error(cc->rpages[0]->mapping, -EIO); 1260 goto out_free; 1261 } 1262 1263 if (IS_NOQUOTA(inode)) { 1264 /* 1265 * We need to wait for node_write to avoid block allocation during 1266 * checkpoint. This can only happen to quota writes which can cause 1267 * the below discard race condition. 1268 */ 1269 f2fs_down_read(&sbi->node_write); 1270 } else if (!f2fs_trylock_op(sbi)) { 1271 goto out_free; 1272 } 1273 1274 set_new_dnode(&dn, cc->inode, NULL, NULL, 0); 1275 1276 err = f2fs_get_dnode_of_data(&dn, start_idx, LOOKUP_NODE); 1277 if (err) 1278 goto out_unlock_op; 1279 1280 for (i = 0; i < cc->cluster_size; i++) { 1281 if (data_blkaddr(dn.inode, dn.node_page, 1282 dn.ofs_in_node + i) == NULL_ADDR) 1283 goto out_put_dnode; 1284 } 1285 1286 psize = (loff_t)(cc->rpages[last_index]->index + 1) << PAGE_SHIFT; 1287 1288 err = f2fs_get_node_info(fio.sbi, dn.nid, &ni, false); 1289 if (err) 1290 goto out_put_dnode; 1291 1292 fio.version = ni.version; 1293 1294 cic = f2fs_kmem_cache_alloc(cic_entry_slab, GFP_F2FS_ZERO, false, sbi); 1295 if (!cic) 1296 goto out_put_dnode; 1297 1298 cic->magic = F2FS_COMPRESSED_PAGE_MAGIC; 1299 cic->inode = inode; 1300 atomic_set(&cic->pending_pages, cc->valid_nr_cpages); 1301 cic->rpages = page_array_alloc(cc->inode, cc->cluster_size); 1302 if (!cic->rpages) 1303 goto out_put_cic; 1304 1305 cic->nr_rpages = cc->cluster_size; 1306 1307 for (i = 0; i < cc->valid_nr_cpages; i++) { 1308 f2fs_set_compressed_page(cc->cpages[i], inode, 1309 cc->rpages[i + 1]->index, cic); 1310 fio.compressed_page = cc->cpages[i]; 1311 1312 fio.old_blkaddr = data_blkaddr(dn.inode, dn.node_page, 1313 dn.ofs_in_node + i + 1); 1314 1315 /* wait for GCed page writeback via META_MAPPING */ 1316 f2fs_wait_on_block_writeback(inode, fio.old_blkaddr); 1317 1318 if (fio.encrypted) { 1319 fio.page = cc->rpages[i + 1]; 1320 err = f2fs_encrypt_one_page(&fio); 1321 if (err) 1322 goto out_destroy_crypt; 1323 cc->cpages[i] = fio.encrypted_page; 1324 } 1325 } 1326 1327 set_cluster_writeback(cc); 1328 1329 for (i = 0; i < cc->cluster_size; i++) 1330 cic->rpages[i] = cc->rpages[i]; 1331 1332 for (i = 0; i < cc->cluster_size; i++, dn.ofs_in_node++) { 1333 block_t blkaddr; 1334 1335 blkaddr = f2fs_data_blkaddr(&dn); 1336 fio.page = cc->rpages[i]; 1337 fio.old_blkaddr = blkaddr; 1338 1339 /* cluster header */ 1340 if (i == 0) { 1341 if (blkaddr == COMPRESS_ADDR) 1342 fio.compr_blocks++; 1343 if (__is_valid_data_blkaddr(blkaddr)) 1344 f2fs_invalidate_blocks(sbi, blkaddr); 1345 f2fs_update_data_blkaddr(&dn, COMPRESS_ADDR); 1346 goto unlock_continue; 1347 } 1348 1349 if (fio.compr_blocks && __is_valid_data_blkaddr(blkaddr)) 1350 fio.compr_blocks++; 1351 1352 if (i > cc->valid_nr_cpages) { 1353 if (__is_valid_data_blkaddr(blkaddr)) { 1354 f2fs_invalidate_blocks(sbi, blkaddr); 1355 f2fs_update_data_blkaddr(&dn, NEW_ADDR); 1356 } 1357 goto unlock_continue; 1358 } 1359 1360 f2fs_bug_on(fio.sbi, blkaddr == NULL_ADDR); 1361 1362 if (fio.encrypted) 1363 fio.encrypted_page = cc->cpages[i - 1]; 1364 else 1365 fio.compressed_page = cc->cpages[i - 1]; 1366 1367 cc->cpages[i - 1] = NULL; 1368 f2fs_outplace_write_data(&dn, &fio); 1369 (*submitted)++; 1370 unlock_continue: 1371 inode_dec_dirty_pages(cc->inode); 1372 unlock_page(fio.page); 1373 } 1374 1375 if (fio.compr_blocks) 1376 f2fs_i_compr_blocks_update(inode, fio.compr_blocks - 1, false); 1377 f2fs_i_compr_blocks_update(inode, cc->valid_nr_cpages, true); 1378 add_compr_block_stat(inode, cc->valid_nr_cpages); 1379 1380 set_inode_flag(cc->inode, FI_APPEND_WRITE); 1381 if (cc->cluster_idx == 0) 1382 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN); 1383 1384 f2fs_put_dnode(&dn); 1385 if (IS_NOQUOTA(inode)) 1386 f2fs_up_read(&sbi->node_write); 1387 else 1388 f2fs_unlock_op(sbi); 1389 1390 spin_lock(&fi->i_size_lock); 1391 if (fi->last_disk_size < psize) 1392 fi->last_disk_size = psize; 1393 spin_unlock(&fi->i_size_lock); 1394 1395 f2fs_put_rpages(cc); 1396 page_array_free(cc->inode, cc->cpages, cc->nr_cpages); 1397 cc->cpages = NULL; 1398 f2fs_destroy_compress_ctx(cc, false); 1399 return 0; 1400 1401 out_destroy_crypt: 1402 page_array_free(cc->inode, cic->rpages, cc->cluster_size); 1403 1404 for (--i; i >= 0; i--) 1405 fscrypt_finalize_bounce_page(&cc->cpages[i]); 1406 out_put_cic: 1407 kmem_cache_free(cic_entry_slab, cic); 1408 out_put_dnode: 1409 f2fs_put_dnode(&dn); 1410 out_unlock_op: 1411 if (IS_NOQUOTA(inode)) 1412 f2fs_up_read(&sbi->node_write); 1413 else 1414 f2fs_unlock_op(sbi); 1415 out_free: 1416 for (i = 0; i < cc->valid_nr_cpages; i++) { 1417 f2fs_compress_free_page(cc->cpages[i]); 1418 cc->cpages[i] = NULL; 1419 } 1420 page_array_free(cc->inode, cc->cpages, cc->nr_cpages); 1421 cc->cpages = NULL; 1422 return -EAGAIN; 1423 } 1424 1425 void f2fs_compress_write_end_io(struct bio *bio, struct page *page) 1426 { 1427 struct f2fs_sb_info *sbi = bio->bi_private; 1428 struct compress_io_ctx *cic = 1429 (struct compress_io_ctx *)page_private(page); 1430 int i; 1431 1432 if (unlikely(bio->bi_status)) 1433 mapping_set_error(cic->inode->i_mapping, -EIO); 1434 1435 f2fs_compress_free_page(page); 1436 1437 dec_page_count(sbi, F2FS_WB_DATA); 1438 1439 if (atomic_dec_return(&cic->pending_pages)) 1440 return; 1441 1442 for (i = 0; i < cic->nr_rpages; i++) { 1443 WARN_ON(!cic->rpages[i]); 1444 clear_page_private_gcing(cic->rpages[i]); 1445 end_page_writeback(cic->rpages[i]); 1446 } 1447 1448 page_array_free(cic->inode, cic->rpages, cic->nr_rpages); 1449 kmem_cache_free(cic_entry_slab, cic); 1450 } 1451 1452 static int f2fs_write_raw_pages(struct compress_ctx *cc, 1453 int *submitted, 1454 struct writeback_control *wbc, 1455 enum iostat_type io_type) 1456 { 1457 struct address_space *mapping = cc->inode->i_mapping; 1458 int _submitted, compr_blocks, ret, i; 1459 1460 compr_blocks = f2fs_compressed_blocks(cc); 1461 1462 for (i = 0; i < cc->cluster_size; i++) { 1463 if (!cc->rpages[i]) 1464 continue; 1465 1466 redirty_page_for_writepage(wbc, cc->rpages[i]); 1467 unlock_page(cc->rpages[i]); 1468 } 1469 1470 if (compr_blocks < 0) 1471 return compr_blocks; 1472 1473 for (i = 0; i < cc->cluster_size; i++) { 1474 if (!cc->rpages[i]) 1475 continue; 1476 retry_write: 1477 lock_page(cc->rpages[i]); 1478 1479 if (cc->rpages[i]->mapping != mapping) { 1480 continue_unlock: 1481 unlock_page(cc->rpages[i]); 1482 continue; 1483 } 1484 1485 if (!PageDirty(cc->rpages[i])) 1486 goto continue_unlock; 1487 1488 if (!clear_page_dirty_for_io(cc->rpages[i])) 1489 goto continue_unlock; 1490 1491 ret = f2fs_write_single_data_page(cc->rpages[i], &_submitted, 1492 NULL, NULL, wbc, io_type, 1493 compr_blocks, false); 1494 if (ret) { 1495 if (ret == AOP_WRITEPAGE_ACTIVATE) { 1496 unlock_page(cc->rpages[i]); 1497 ret = 0; 1498 } else if (ret == -EAGAIN) { 1499 /* 1500 * for quota file, just redirty left pages to 1501 * avoid deadlock caused by cluster update race 1502 * from foreground operation. 1503 */ 1504 if (IS_NOQUOTA(cc->inode)) 1505 return 0; 1506 ret = 0; 1507 f2fs_io_schedule_timeout(DEFAULT_IO_TIMEOUT); 1508 goto retry_write; 1509 } 1510 return ret; 1511 } 1512 1513 *submitted += _submitted; 1514 } 1515 1516 f2fs_balance_fs(F2FS_M_SB(mapping), true); 1517 1518 return 0; 1519 } 1520 1521 int f2fs_write_multi_pages(struct compress_ctx *cc, 1522 int *submitted, 1523 struct writeback_control *wbc, 1524 enum iostat_type io_type) 1525 { 1526 int err; 1527 1528 *submitted = 0; 1529 if (cluster_may_compress(cc)) { 1530 err = f2fs_compress_pages(cc); 1531 if (err == -EAGAIN) { 1532 add_compr_block_stat(cc->inode, cc->cluster_size); 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 } 1545 write: 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); 1550 destroy_out: 1551 f2fs_destroy_compress_ctx(cc, false); 1552 return err; 1553 } 1554 1555 static void f2fs_free_dic(struct decompress_io_ctx *dic); 1556 1557 struct 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 1607 out_free: 1608 f2fs_free_dic(dic); 1609 return ERR_PTR(-ENOMEM); 1610 } 1611 1612 static 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 1640 static 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 */ 1650 static 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 1674 static 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 */ 1695 void 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 */ 1716 void 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 */ 1728 unsigned 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 1748 const struct address_space_operations f2fs_compress_aops = { 1749 .releasepage = f2fs_release_page, 1750 .invalidate_folio = f2fs_invalidate_folio, 1751 }; 1752 1753 struct address_space *COMPRESS_MAPPING(struct f2fs_sb_info *sbi) 1754 { 1755 return sbi->compress_inode->i_mapping; 1756 } 1757 1758 void 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 1765 void 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); 1804 out: 1805 f2fs_put_page(cpage, 1); 1806 } 1807 1808 bool 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 1832 void 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 1878 int 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 1898 void 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 1906 int 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 1923 void f2fs_destroy_page_array_cache(struct f2fs_sb_info *sbi) 1924 { 1925 kmem_cache_destroy(sbi->page_array_slab); 1926 } 1927 1928 static 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 1937 static void f2fs_destroy_cic_cache(void) 1938 { 1939 kmem_cache_destroy(cic_entry_slab); 1940 } 1941 1942 static 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 1951 static void f2fs_destroy_dic_cache(void) 1952 { 1953 kmem_cache_destroy(dic_entry_slab); 1954 } 1955 1956 int __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; 1967 free_cic: 1968 f2fs_destroy_cic_cache(); 1969 out: 1970 return -ENOMEM; 1971 } 1972 1973 void f2fs_destroy_compress_cache(void) 1974 { 1975 f2fs_destroy_dic_cache(); 1976 f2fs_destroy_cic_cache(); 1977 } 1978