1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Copyright (C) 2018 HUAWEI, Inc. 4 * https://www.huawei.com/ 5 * Created by Gao Xiang <gaoxiang25@huawei.com> 6 */ 7 #include "zdata.h" 8 #include "compress.h" 9 #include <linux/prefetch.h> 10 11 #include <trace/events/erofs.h> 12 13 /* 14 * a compressed_pages[] placeholder in order to avoid 15 * being filled with file pages for in-place decompression. 16 */ 17 #define PAGE_UNALLOCATED ((void *)0x5F0E4B1D) 18 19 /* how to allocate cached pages for a pcluster */ 20 enum z_erofs_cache_alloctype { 21 DONTALLOC, /* don't allocate any cached pages */ 22 DELAYEDALLOC, /* delayed allocation (at the time of submitting io) */ 23 }; 24 25 /* 26 * tagged pointer with 1-bit tag for all compressed pages 27 * tag 0 - the page is just found with an extra page reference 28 */ 29 typedef tagptr1_t compressed_page_t; 30 31 #define tag_compressed_page_justfound(page) \ 32 tagptr_fold(compressed_page_t, page, 1) 33 34 static struct workqueue_struct *z_erofs_workqueue __read_mostly; 35 static struct kmem_cache *pcluster_cachep __read_mostly; 36 37 void z_erofs_exit_zip_subsystem(void) 38 { 39 destroy_workqueue(z_erofs_workqueue); 40 kmem_cache_destroy(pcluster_cachep); 41 } 42 43 static inline int z_erofs_init_workqueue(void) 44 { 45 const unsigned int onlinecpus = num_possible_cpus(); 46 47 /* 48 * no need to spawn too many threads, limiting threads could minimum 49 * scheduling overhead, perhaps per-CPU threads should be better? 50 */ 51 z_erofs_workqueue = alloc_workqueue("erofs_unzipd", 52 WQ_UNBOUND | WQ_HIGHPRI, 53 onlinecpus + onlinecpus / 4); 54 return z_erofs_workqueue ? 0 : -ENOMEM; 55 } 56 57 static void z_erofs_pcluster_init_once(void *ptr) 58 { 59 struct z_erofs_pcluster *pcl = ptr; 60 struct z_erofs_collection *cl = z_erofs_primarycollection(pcl); 61 unsigned int i; 62 63 mutex_init(&cl->lock); 64 cl->nr_pages = 0; 65 cl->vcnt = 0; 66 for (i = 0; i < Z_EROFS_CLUSTER_MAX_PAGES; ++i) 67 pcl->compressed_pages[i] = NULL; 68 } 69 70 int __init z_erofs_init_zip_subsystem(void) 71 { 72 pcluster_cachep = kmem_cache_create("erofs_compress", 73 Z_EROFS_WORKGROUP_SIZE, 0, 74 SLAB_RECLAIM_ACCOUNT, 75 z_erofs_pcluster_init_once); 76 if (pcluster_cachep) { 77 if (!z_erofs_init_workqueue()) 78 return 0; 79 80 kmem_cache_destroy(pcluster_cachep); 81 } 82 return -ENOMEM; 83 } 84 85 enum z_erofs_collectmode { 86 COLLECT_SECONDARY, 87 COLLECT_PRIMARY, 88 /* 89 * The current collection was the tail of an exist chain, in addition 90 * that the previous processed chained collections are all decided to 91 * be hooked up to it. 92 * A new chain will be created for the remaining collections which are 93 * not processed yet, therefore different from COLLECT_PRIMARY_FOLLOWED, 94 * the next collection cannot reuse the whole page safely in 95 * the following scenario: 96 * ________________________________________________________________ 97 * | tail (partial) page | head (partial) page | 98 * | (belongs to the next cl) | (belongs to the current cl) | 99 * |_______PRIMARY_FOLLOWED_______|________PRIMARY_HOOKED___________| 100 */ 101 COLLECT_PRIMARY_HOOKED, 102 COLLECT_PRIMARY_FOLLOWED_NOINPLACE, 103 /* 104 * The current collection has been linked with the owned chain, and 105 * could also be linked with the remaining collections, which means 106 * if the processing page is the tail page of the collection, thus 107 * the current collection can safely use the whole page (since 108 * the previous collection is under control) for in-place I/O, as 109 * illustrated below: 110 * ________________________________________________________________ 111 * | tail (partial) page | head (partial) page | 112 * | (of the current cl) | (of the previous collection) | 113 * | PRIMARY_FOLLOWED or | | 114 * |_____PRIMARY_HOOKED___|____________PRIMARY_FOLLOWED____________| 115 * 116 * [ (*) the above page can be used as inplace I/O. ] 117 */ 118 COLLECT_PRIMARY_FOLLOWED, 119 }; 120 121 struct z_erofs_collector { 122 struct z_erofs_pagevec_ctor vector; 123 124 struct z_erofs_pcluster *pcl, *tailpcl; 125 struct z_erofs_collection *cl; 126 struct page **compressedpages; 127 z_erofs_next_pcluster_t owned_head; 128 129 enum z_erofs_collectmode mode; 130 }; 131 132 struct z_erofs_decompress_frontend { 133 struct inode *const inode; 134 135 struct z_erofs_collector clt; 136 struct erofs_map_blocks map; 137 138 bool readahead; 139 /* used for applying cache strategy on the fly */ 140 bool backmost; 141 erofs_off_t headoffset; 142 }; 143 144 #define COLLECTOR_INIT() { \ 145 .owned_head = Z_EROFS_PCLUSTER_TAIL, \ 146 .mode = COLLECT_PRIMARY_FOLLOWED } 147 148 #define DECOMPRESS_FRONTEND_INIT(__i) { \ 149 .inode = __i, .clt = COLLECTOR_INIT(), \ 150 .backmost = true, } 151 152 static struct page *z_pagemap_global[Z_EROFS_VMAP_GLOBAL_PAGES]; 153 static DEFINE_MUTEX(z_pagemap_global_lock); 154 155 static void preload_compressed_pages(struct z_erofs_collector *clt, 156 struct address_space *mc, 157 enum z_erofs_cache_alloctype type) 158 { 159 const struct z_erofs_pcluster *pcl = clt->pcl; 160 const unsigned int clusterpages = BIT(pcl->clusterbits); 161 struct page **pages = clt->compressedpages; 162 pgoff_t index = pcl->obj.index + (pages - pcl->compressed_pages); 163 bool standalone = true; 164 165 if (clt->mode < COLLECT_PRIMARY_FOLLOWED) 166 return; 167 168 for (; pages < pcl->compressed_pages + clusterpages; ++pages) { 169 struct page *page; 170 compressed_page_t t; 171 172 /* the compressed page was loaded before */ 173 if (READ_ONCE(*pages)) 174 continue; 175 176 page = find_get_page(mc, index); 177 178 if (page) { 179 t = tag_compressed_page_justfound(page); 180 } else if (type == DELAYEDALLOC) { 181 t = tagptr_init(compressed_page_t, PAGE_UNALLOCATED); 182 } else { /* DONTALLOC */ 183 if (standalone) 184 clt->compressedpages = pages; 185 standalone = false; 186 continue; 187 } 188 189 if (!cmpxchg_relaxed(pages, NULL, tagptr_cast_ptr(t))) 190 continue; 191 192 if (page) 193 put_page(page); 194 } 195 196 if (standalone) /* downgrade to PRIMARY_FOLLOWED_NOINPLACE */ 197 clt->mode = COLLECT_PRIMARY_FOLLOWED_NOINPLACE; 198 } 199 200 /* called by erofs_shrinker to get rid of all compressed_pages */ 201 int erofs_try_to_free_all_cached_pages(struct erofs_sb_info *sbi, 202 struct erofs_workgroup *grp) 203 { 204 struct z_erofs_pcluster *const pcl = 205 container_of(grp, struct z_erofs_pcluster, obj); 206 struct address_space *const mapping = MNGD_MAPPING(sbi); 207 const unsigned int clusterpages = BIT(pcl->clusterbits); 208 int i; 209 210 /* 211 * refcount of workgroup is now freezed as 1, 212 * therefore no need to worry about available decompression users. 213 */ 214 for (i = 0; i < clusterpages; ++i) { 215 struct page *page = pcl->compressed_pages[i]; 216 217 if (!page) 218 continue; 219 220 /* block other users from reclaiming or migrating the page */ 221 if (!trylock_page(page)) 222 return -EBUSY; 223 224 if (page->mapping != mapping) 225 continue; 226 227 /* barrier is implied in the following 'unlock_page' */ 228 WRITE_ONCE(pcl->compressed_pages[i], NULL); 229 detach_page_private(page); 230 unlock_page(page); 231 } 232 return 0; 233 } 234 235 int erofs_try_to_free_cached_page(struct address_space *mapping, 236 struct page *page) 237 { 238 struct z_erofs_pcluster *const pcl = (void *)page_private(page); 239 const unsigned int clusterpages = BIT(pcl->clusterbits); 240 int ret = 0; /* 0 - busy */ 241 242 if (erofs_workgroup_try_to_freeze(&pcl->obj, 1)) { 243 unsigned int i; 244 245 for (i = 0; i < clusterpages; ++i) { 246 if (pcl->compressed_pages[i] == page) { 247 WRITE_ONCE(pcl->compressed_pages[i], NULL); 248 ret = 1; 249 break; 250 } 251 } 252 erofs_workgroup_unfreeze(&pcl->obj, 1); 253 254 if (ret) 255 detach_page_private(page); 256 } 257 return ret; 258 } 259 260 /* page_type must be Z_EROFS_PAGE_TYPE_EXCLUSIVE */ 261 static inline bool z_erofs_try_inplace_io(struct z_erofs_collector *clt, 262 struct page *page) 263 { 264 struct z_erofs_pcluster *const pcl = clt->pcl; 265 const unsigned int clusterpages = BIT(pcl->clusterbits); 266 267 while (clt->compressedpages < pcl->compressed_pages + clusterpages) { 268 if (!cmpxchg(clt->compressedpages++, NULL, page)) 269 return true; 270 } 271 return false; 272 } 273 274 /* callers must be with collection lock held */ 275 static int z_erofs_attach_page(struct z_erofs_collector *clt, 276 struct page *page, 277 enum z_erofs_page_type type) 278 { 279 int ret; 280 bool occupied; 281 282 /* give priority for inplaceio */ 283 if (clt->mode >= COLLECT_PRIMARY && 284 type == Z_EROFS_PAGE_TYPE_EXCLUSIVE && 285 z_erofs_try_inplace_io(clt, page)) 286 return 0; 287 288 ret = z_erofs_pagevec_enqueue(&clt->vector, 289 page, type, &occupied); 290 clt->cl->vcnt += (unsigned int)ret; 291 292 return ret ? 0 : -EAGAIN; 293 } 294 295 static enum z_erofs_collectmode 296 try_to_claim_pcluster(struct z_erofs_pcluster *pcl, 297 z_erofs_next_pcluster_t *owned_head) 298 { 299 /* let's claim these following types of pclusters */ 300 retry: 301 if (pcl->next == Z_EROFS_PCLUSTER_NIL) { 302 /* type 1, nil pcluster */ 303 if (cmpxchg(&pcl->next, Z_EROFS_PCLUSTER_NIL, 304 *owned_head) != Z_EROFS_PCLUSTER_NIL) 305 goto retry; 306 307 *owned_head = &pcl->next; 308 /* lucky, I am the followee :) */ 309 return COLLECT_PRIMARY_FOLLOWED; 310 } else if (pcl->next == Z_EROFS_PCLUSTER_TAIL) { 311 /* 312 * type 2, link to the end of a existing open chain, 313 * be careful that its submission itself is governed 314 * by the original owned chain. 315 */ 316 if (cmpxchg(&pcl->next, Z_EROFS_PCLUSTER_TAIL, 317 *owned_head) != Z_EROFS_PCLUSTER_TAIL) 318 goto retry; 319 *owned_head = Z_EROFS_PCLUSTER_TAIL; 320 return COLLECT_PRIMARY_HOOKED; 321 } 322 return COLLECT_PRIMARY; /* :( better luck next time */ 323 } 324 325 static int z_erofs_lookup_collection(struct z_erofs_collector *clt, 326 struct inode *inode, 327 struct erofs_map_blocks *map) 328 { 329 struct z_erofs_pcluster *pcl = clt->pcl; 330 struct z_erofs_collection *cl; 331 unsigned int length; 332 333 /* to avoid unexpected loop formed by corrupted images */ 334 if (clt->owned_head == &pcl->next || pcl == clt->tailpcl) { 335 DBG_BUGON(1); 336 return -EFSCORRUPTED; 337 } 338 339 cl = z_erofs_primarycollection(pcl); 340 if (cl->pageofs != (map->m_la & ~PAGE_MASK)) { 341 DBG_BUGON(1); 342 return -EFSCORRUPTED; 343 } 344 345 length = READ_ONCE(pcl->length); 346 if (length & Z_EROFS_PCLUSTER_FULL_LENGTH) { 347 if ((map->m_llen << Z_EROFS_PCLUSTER_LENGTH_BIT) > length) { 348 DBG_BUGON(1); 349 return -EFSCORRUPTED; 350 } 351 } else { 352 unsigned int llen = map->m_llen << Z_EROFS_PCLUSTER_LENGTH_BIT; 353 354 if (map->m_flags & EROFS_MAP_FULL_MAPPED) 355 llen |= Z_EROFS_PCLUSTER_FULL_LENGTH; 356 357 while (llen > length && 358 length != cmpxchg_relaxed(&pcl->length, length, llen)) { 359 cpu_relax(); 360 length = READ_ONCE(pcl->length); 361 } 362 } 363 mutex_lock(&cl->lock); 364 /* used to check tail merging loop due to corrupted images */ 365 if (clt->owned_head == Z_EROFS_PCLUSTER_TAIL) 366 clt->tailpcl = pcl; 367 clt->mode = try_to_claim_pcluster(pcl, &clt->owned_head); 368 /* clean tailpcl if the current owned_head is Z_EROFS_PCLUSTER_TAIL */ 369 if (clt->owned_head == Z_EROFS_PCLUSTER_TAIL) 370 clt->tailpcl = NULL; 371 clt->cl = cl; 372 return 0; 373 } 374 375 static int z_erofs_register_collection(struct z_erofs_collector *clt, 376 struct inode *inode, 377 struct erofs_map_blocks *map) 378 { 379 struct z_erofs_pcluster *pcl; 380 struct z_erofs_collection *cl; 381 struct erofs_workgroup *grp; 382 int err; 383 384 /* no available workgroup, let's allocate one */ 385 pcl = kmem_cache_alloc(pcluster_cachep, GFP_NOFS); 386 if (!pcl) 387 return -ENOMEM; 388 389 atomic_set(&pcl->obj.refcount, 1); 390 pcl->obj.index = map->m_pa >> PAGE_SHIFT; 391 392 pcl->length = (map->m_llen << Z_EROFS_PCLUSTER_LENGTH_BIT) | 393 (map->m_flags & EROFS_MAP_FULL_MAPPED ? 394 Z_EROFS_PCLUSTER_FULL_LENGTH : 0); 395 396 if (map->m_flags & EROFS_MAP_ZIPPED) 397 pcl->algorithmformat = Z_EROFS_COMPRESSION_LZ4; 398 else 399 pcl->algorithmformat = Z_EROFS_COMPRESSION_SHIFTED; 400 401 pcl->clusterbits = EROFS_I(inode)->z_physical_clusterbits[0]; 402 pcl->clusterbits -= PAGE_SHIFT; 403 404 /* new pclusters should be claimed as type 1, primary and followed */ 405 pcl->next = clt->owned_head; 406 clt->mode = COLLECT_PRIMARY_FOLLOWED; 407 408 cl = z_erofs_primarycollection(pcl); 409 410 /* must be cleaned before freeing to slab */ 411 DBG_BUGON(cl->nr_pages); 412 DBG_BUGON(cl->vcnt); 413 414 cl->pageofs = map->m_la & ~PAGE_MASK; 415 416 /* 417 * lock all primary followed works before visible to others 418 * and mutex_trylock *never* fails for a new pcluster. 419 */ 420 DBG_BUGON(!mutex_trylock(&cl->lock)); 421 422 grp = erofs_insert_workgroup(inode->i_sb, &pcl->obj); 423 if (IS_ERR(grp)) { 424 err = PTR_ERR(grp); 425 goto err_out; 426 } 427 428 if (grp != &pcl->obj) { 429 clt->pcl = container_of(grp, struct z_erofs_pcluster, obj); 430 err = -EEXIST; 431 goto err_out; 432 } 433 /* used to check tail merging loop due to corrupted images */ 434 if (clt->owned_head == Z_EROFS_PCLUSTER_TAIL) 435 clt->tailpcl = pcl; 436 clt->owned_head = &pcl->next; 437 clt->pcl = pcl; 438 clt->cl = cl; 439 return 0; 440 441 err_out: 442 mutex_unlock(&cl->lock); 443 kmem_cache_free(pcluster_cachep, pcl); 444 return err; 445 } 446 447 static int z_erofs_collector_begin(struct z_erofs_collector *clt, 448 struct inode *inode, 449 struct erofs_map_blocks *map) 450 { 451 struct erofs_workgroup *grp; 452 int ret; 453 454 DBG_BUGON(clt->cl); 455 456 /* must be Z_EROFS_PCLUSTER_TAIL or pointed to previous collection */ 457 DBG_BUGON(clt->owned_head == Z_EROFS_PCLUSTER_NIL); 458 DBG_BUGON(clt->owned_head == Z_EROFS_PCLUSTER_TAIL_CLOSED); 459 460 if (!PAGE_ALIGNED(map->m_pa)) { 461 DBG_BUGON(1); 462 return -EINVAL; 463 } 464 465 grp = erofs_find_workgroup(inode->i_sb, map->m_pa >> PAGE_SHIFT); 466 if (grp) { 467 clt->pcl = container_of(grp, struct z_erofs_pcluster, obj); 468 } else { 469 ret = z_erofs_register_collection(clt, inode, map); 470 471 if (!ret) 472 goto out; 473 if (ret != -EEXIST) 474 return ret; 475 } 476 477 ret = z_erofs_lookup_collection(clt, inode, map); 478 if (ret) { 479 erofs_workgroup_put(&clt->pcl->obj); 480 return ret; 481 } 482 483 out: 484 z_erofs_pagevec_ctor_init(&clt->vector, Z_EROFS_NR_INLINE_PAGEVECS, 485 clt->cl->pagevec, clt->cl->vcnt); 486 487 clt->compressedpages = clt->pcl->compressed_pages; 488 if (clt->mode <= COLLECT_PRIMARY) /* cannot do in-place I/O */ 489 clt->compressedpages += Z_EROFS_CLUSTER_MAX_PAGES; 490 return 0; 491 } 492 493 /* 494 * keep in mind that no referenced pclusters will be freed 495 * only after a RCU grace period. 496 */ 497 static void z_erofs_rcu_callback(struct rcu_head *head) 498 { 499 struct z_erofs_collection *const cl = 500 container_of(head, struct z_erofs_collection, rcu); 501 502 kmem_cache_free(pcluster_cachep, 503 container_of(cl, struct z_erofs_pcluster, 504 primary_collection)); 505 } 506 507 void erofs_workgroup_free_rcu(struct erofs_workgroup *grp) 508 { 509 struct z_erofs_pcluster *const pcl = 510 container_of(grp, struct z_erofs_pcluster, obj); 511 struct z_erofs_collection *const cl = z_erofs_primarycollection(pcl); 512 513 call_rcu(&cl->rcu, z_erofs_rcu_callback); 514 } 515 516 static void z_erofs_collection_put(struct z_erofs_collection *cl) 517 { 518 struct z_erofs_pcluster *const pcl = 519 container_of(cl, struct z_erofs_pcluster, primary_collection); 520 521 erofs_workgroup_put(&pcl->obj); 522 } 523 524 static bool z_erofs_collector_end(struct z_erofs_collector *clt) 525 { 526 struct z_erofs_collection *cl = clt->cl; 527 528 if (!cl) 529 return false; 530 531 z_erofs_pagevec_ctor_exit(&clt->vector, false); 532 mutex_unlock(&cl->lock); 533 534 /* 535 * if all pending pages are added, don't hold its reference 536 * any longer if the pcluster isn't hosted by ourselves. 537 */ 538 if (clt->mode < COLLECT_PRIMARY_FOLLOWED_NOINPLACE) 539 z_erofs_collection_put(cl); 540 541 clt->cl = NULL; 542 return true; 543 } 544 545 static bool should_alloc_managed_pages(struct z_erofs_decompress_frontend *fe, 546 unsigned int cachestrategy, 547 erofs_off_t la) 548 { 549 if (cachestrategy <= EROFS_ZIP_CACHE_DISABLED) 550 return false; 551 552 if (fe->backmost) 553 return true; 554 555 return cachestrategy >= EROFS_ZIP_CACHE_READAROUND && 556 la < fe->headoffset; 557 } 558 559 static int z_erofs_do_read_page(struct z_erofs_decompress_frontend *fe, 560 struct page *page) 561 { 562 struct inode *const inode = fe->inode; 563 struct erofs_sb_info *const sbi = EROFS_I_SB(inode); 564 struct erofs_map_blocks *const map = &fe->map; 565 struct z_erofs_collector *const clt = &fe->clt; 566 const loff_t offset = page_offset(page); 567 bool tight = true; 568 569 enum z_erofs_cache_alloctype cache_strategy; 570 enum z_erofs_page_type page_type; 571 unsigned int cur, end, spiltted, index; 572 int err = 0; 573 574 /* register locked file pages as online pages in pack */ 575 z_erofs_onlinepage_init(page); 576 577 spiltted = 0; 578 end = PAGE_SIZE; 579 repeat: 580 cur = end - 1; 581 582 /* lucky, within the range of the current map_blocks */ 583 if (offset + cur >= map->m_la && 584 offset + cur < map->m_la + map->m_llen) { 585 /* didn't get a valid collection previously (very rare) */ 586 if (!clt->cl) 587 goto restart_now; 588 goto hitted; 589 } 590 591 /* go ahead the next map_blocks */ 592 erofs_dbg("%s: [out-of-range] pos %llu", __func__, offset + cur); 593 594 if (z_erofs_collector_end(clt)) 595 fe->backmost = false; 596 597 map->m_la = offset + cur; 598 map->m_llen = 0; 599 err = z_erofs_map_blocks_iter(inode, map, 0); 600 if (err) 601 goto err_out; 602 603 restart_now: 604 if (!(map->m_flags & EROFS_MAP_MAPPED)) 605 goto hitted; 606 607 err = z_erofs_collector_begin(clt, inode, map); 608 if (err) 609 goto err_out; 610 611 /* preload all compressed pages (maybe downgrade role if necessary) */ 612 if (should_alloc_managed_pages(fe, sbi->ctx.cache_strategy, map->m_la)) 613 cache_strategy = DELAYEDALLOC; 614 else 615 cache_strategy = DONTALLOC; 616 617 preload_compressed_pages(clt, MNGD_MAPPING(sbi), cache_strategy); 618 619 hitted: 620 /* 621 * Ensure the current partial page belongs to this submit chain rather 622 * than other concurrent submit chains or the noio(bypass) chain since 623 * those chains are handled asynchronously thus the page cannot be used 624 * for inplace I/O or pagevec (should be processed in strict order.) 625 */ 626 tight &= (clt->mode >= COLLECT_PRIMARY_HOOKED && 627 clt->mode != COLLECT_PRIMARY_FOLLOWED_NOINPLACE); 628 629 cur = end - min_t(unsigned int, offset + end - map->m_la, end); 630 if (!(map->m_flags & EROFS_MAP_MAPPED)) { 631 zero_user_segment(page, cur, end); 632 goto next_part; 633 } 634 635 /* let's derive page type */ 636 page_type = cur ? Z_EROFS_VLE_PAGE_TYPE_HEAD : 637 (!spiltted ? Z_EROFS_PAGE_TYPE_EXCLUSIVE : 638 (tight ? Z_EROFS_PAGE_TYPE_EXCLUSIVE : 639 Z_EROFS_VLE_PAGE_TYPE_TAIL_SHARED)); 640 641 if (cur) 642 tight &= (clt->mode >= COLLECT_PRIMARY_FOLLOWED); 643 644 retry: 645 err = z_erofs_attach_page(clt, page, page_type); 646 /* should allocate an additional short-lived page for pagevec */ 647 if (err == -EAGAIN) { 648 struct page *const newpage = 649 alloc_page(GFP_NOFS | __GFP_NOFAIL); 650 651 set_page_private(newpage, Z_EROFS_SHORTLIVED_PAGE); 652 err = z_erofs_attach_page(clt, newpage, 653 Z_EROFS_PAGE_TYPE_EXCLUSIVE); 654 if (!err) 655 goto retry; 656 } 657 658 if (err) 659 goto err_out; 660 661 index = page->index - (map->m_la >> PAGE_SHIFT); 662 663 z_erofs_onlinepage_fixup(page, index, true); 664 665 /* bump up the number of spiltted parts of a page */ 666 ++spiltted; 667 /* also update nr_pages */ 668 clt->cl->nr_pages = max_t(pgoff_t, clt->cl->nr_pages, index + 1); 669 next_part: 670 /* can be used for verification */ 671 map->m_llen = offset + cur - map->m_la; 672 673 end = cur; 674 if (end > 0) 675 goto repeat; 676 677 out: 678 z_erofs_onlinepage_endio(page); 679 680 erofs_dbg("%s, finish page: %pK spiltted: %u map->m_llen %llu", 681 __func__, page, spiltted, map->m_llen); 682 return err; 683 684 /* if some error occurred while processing this page */ 685 err_out: 686 SetPageError(page); 687 goto out; 688 } 689 690 static void z_erofs_decompress_kickoff(struct z_erofs_decompressqueue *io, 691 bool sync, int bios) 692 { 693 /* wake up the caller thread for sync decompression */ 694 if (sync) { 695 unsigned long flags; 696 697 spin_lock_irqsave(&io->u.wait.lock, flags); 698 if (!atomic_add_return(bios, &io->pending_bios)) 699 wake_up_locked(&io->u.wait); 700 spin_unlock_irqrestore(&io->u.wait.lock, flags); 701 return; 702 } 703 704 if (!atomic_add_return(bios, &io->pending_bios)) 705 queue_work(z_erofs_workqueue, &io->u.work); 706 } 707 708 static bool z_erofs_page_is_invalidated(struct page *page) 709 { 710 return !page->mapping && !z_erofs_is_shortlived_page(page); 711 } 712 713 static void z_erofs_decompressqueue_endio(struct bio *bio) 714 { 715 tagptr1_t t = tagptr_init(tagptr1_t, bio->bi_private); 716 struct z_erofs_decompressqueue *q = tagptr_unfold_ptr(t); 717 blk_status_t err = bio->bi_status; 718 struct bio_vec *bvec; 719 struct bvec_iter_all iter_all; 720 721 bio_for_each_segment_all(bvec, bio, iter_all) { 722 struct page *page = bvec->bv_page; 723 724 DBG_BUGON(PageUptodate(page)); 725 DBG_BUGON(z_erofs_page_is_invalidated(page)); 726 727 if (err) 728 SetPageError(page); 729 730 if (erofs_page_is_managed(EROFS_SB(q->sb), page)) { 731 if (!err) 732 SetPageUptodate(page); 733 unlock_page(page); 734 } 735 } 736 z_erofs_decompress_kickoff(q, tagptr_unfold_tags(t), -1); 737 bio_put(bio); 738 } 739 740 static int z_erofs_decompress_pcluster(struct super_block *sb, 741 struct z_erofs_pcluster *pcl, 742 struct list_head *pagepool) 743 { 744 struct erofs_sb_info *const sbi = EROFS_SB(sb); 745 const unsigned int clusterpages = BIT(pcl->clusterbits); 746 struct z_erofs_pagevec_ctor ctor; 747 unsigned int i, outputsize, llen, nr_pages; 748 struct page *pages_onstack[Z_EROFS_VMAP_ONSTACK_PAGES]; 749 struct page **pages, **compressed_pages, *page; 750 751 enum z_erofs_page_type page_type; 752 bool overlapped, partial; 753 struct z_erofs_collection *cl; 754 int err; 755 756 might_sleep(); 757 cl = z_erofs_primarycollection(pcl); 758 DBG_BUGON(!READ_ONCE(cl->nr_pages)); 759 760 mutex_lock(&cl->lock); 761 nr_pages = cl->nr_pages; 762 763 if (nr_pages <= Z_EROFS_VMAP_ONSTACK_PAGES) { 764 pages = pages_onstack; 765 } else if (nr_pages <= Z_EROFS_VMAP_GLOBAL_PAGES && 766 mutex_trylock(&z_pagemap_global_lock)) { 767 pages = z_pagemap_global; 768 } else { 769 gfp_t gfp_flags = GFP_KERNEL; 770 771 if (nr_pages > Z_EROFS_VMAP_GLOBAL_PAGES) 772 gfp_flags |= __GFP_NOFAIL; 773 774 pages = kvmalloc_array(nr_pages, sizeof(struct page *), 775 gfp_flags); 776 777 /* fallback to global pagemap for the lowmem scenario */ 778 if (!pages) { 779 mutex_lock(&z_pagemap_global_lock); 780 pages = z_pagemap_global; 781 } 782 } 783 784 for (i = 0; i < nr_pages; ++i) 785 pages[i] = NULL; 786 787 err = 0; 788 z_erofs_pagevec_ctor_init(&ctor, Z_EROFS_NR_INLINE_PAGEVECS, 789 cl->pagevec, 0); 790 791 for (i = 0; i < cl->vcnt; ++i) { 792 unsigned int pagenr; 793 794 page = z_erofs_pagevec_dequeue(&ctor, &page_type); 795 796 /* all pages in pagevec ought to be valid */ 797 DBG_BUGON(!page); 798 DBG_BUGON(z_erofs_page_is_invalidated(page)); 799 800 if (z_erofs_put_shortlivedpage(pagepool, page)) 801 continue; 802 803 if (page_type == Z_EROFS_VLE_PAGE_TYPE_HEAD) 804 pagenr = 0; 805 else 806 pagenr = z_erofs_onlinepage_index(page); 807 808 DBG_BUGON(pagenr >= nr_pages); 809 810 /* 811 * currently EROFS doesn't support multiref(dedup), 812 * so here erroring out one multiref page. 813 */ 814 if (pages[pagenr]) { 815 DBG_BUGON(1); 816 SetPageError(pages[pagenr]); 817 z_erofs_onlinepage_endio(pages[pagenr]); 818 err = -EFSCORRUPTED; 819 } 820 pages[pagenr] = page; 821 } 822 z_erofs_pagevec_ctor_exit(&ctor, true); 823 824 overlapped = false; 825 compressed_pages = pcl->compressed_pages; 826 827 for (i = 0; i < clusterpages; ++i) { 828 unsigned int pagenr; 829 830 page = compressed_pages[i]; 831 832 /* all compressed pages ought to be valid */ 833 DBG_BUGON(!page); 834 DBG_BUGON(z_erofs_page_is_invalidated(page)); 835 836 if (!z_erofs_is_shortlived_page(page)) { 837 if (erofs_page_is_managed(sbi, page)) { 838 if (!PageUptodate(page)) 839 err = -EIO; 840 continue; 841 } 842 843 /* 844 * only if non-head page can be selected 845 * for inplace decompression 846 */ 847 pagenr = z_erofs_onlinepage_index(page); 848 849 DBG_BUGON(pagenr >= nr_pages); 850 if (pages[pagenr]) { 851 DBG_BUGON(1); 852 SetPageError(pages[pagenr]); 853 z_erofs_onlinepage_endio(pages[pagenr]); 854 err = -EFSCORRUPTED; 855 } 856 pages[pagenr] = page; 857 858 overlapped = true; 859 } 860 861 /* PG_error needs checking for all non-managed pages */ 862 if (PageError(page)) { 863 DBG_BUGON(PageUptodate(page)); 864 err = -EIO; 865 } 866 } 867 868 if (err) 869 goto out; 870 871 llen = pcl->length >> Z_EROFS_PCLUSTER_LENGTH_BIT; 872 if (nr_pages << PAGE_SHIFT >= cl->pageofs + llen) { 873 outputsize = llen; 874 partial = !(pcl->length & Z_EROFS_PCLUSTER_FULL_LENGTH); 875 } else { 876 outputsize = (nr_pages << PAGE_SHIFT) - cl->pageofs; 877 partial = true; 878 } 879 880 err = z_erofs_decompress(&(struct z_erofs_decompress_req) { 881 .sb = sb, 882 .in = compressed_pages, 883 .out = pages, 884 .pageofs_out = cl->pageofs, 885 .inputsize = PAGE_SIZE, 886 .outputsize = outputsize, 887 .alg = pcl->algorithmformat, 888 .inplace_io = overlapped, 889 .partial_decoding = partial 890 }, pagepool); 891 892 out: 893 /* must handle all compressed pages before endding pages */ 894 for (i = 0; i < clusterpages; ++i) { 895 page = compressed_pages[i]; 896 897 if (erofs_page_is_managed(sbi, page)) 898 continue; 899 900 /* recycle all individual short-lived pages */ 901 (void)z_erofs_put_shortlivedpage(pagepool, page); 902 903 WRITE_ONCE(compressed_pages[i], NULL); 904 } 905 906 for (i = 0; i < nr_pages; ++i) { 907 page = pages[i]; 908 if (!page) 909 continue; 910 911 DBG_BUGON(z_erofs_page_is_invalidated(page)); 912 913 /* recycle all individual short-lived pages */ 914 if (z_erofs_put_shortlivedpage(pagepool, page)) 915 continue; 916 917 if (err < 0) 918 SetPageError(page); 919 920 z_erofs_onlinepage_endio(page); 921 } 922 923 if (pages == z_pagemap_global) 924 mutex_unlock(&z_pagemap_global_lock); 925 else if (pages != pages_onstack) 926 kvfree(pages); 927 928 cl->nr_pages = 0; 929 cl->vcnt = 0; 930 931 /* all cl locks MUST be taken before the following line */ 932 WRITE_ONCE(pcl->next, Z_EROFS_PCLUSTER_NIL); 933 934 /* all cl locks SHOULD be released right now */ 935 mutex_unlock(&cl->lock); 936 937 z_erofs_collection_put(cl); 938 return err; 939 } 940 941 static void z_erofs_decompress_queue(const struct z_erofs_decompressqueue *io, 942 struct list_head *pagepool) 943 { 944 z_erofs_next_pcluster_t owned = io->head; 945 946 while (owned != Z_EROFS_PCLUSTER_TAIL_CLOSED) { 947 struct z_erofs_pcluster *pcl; 948 949 /* no possible that 'owned' equals Z_EROFS_WORK_TPTR_TAIL */ 950 DBG_BUGON(owned == Z_EROFS_PCLUSTER_TAIL); 951 952 /* no possible that 'owned' equals NULL */ 953 DBG_BUGON(owned == Z_EROFS_PCLUSTER_NIL); 954 955 pcl = container_of(owned, struct z_erofs_pcluster, next); 956 owned = READ_ONCE(pcl->next); 957 958 z_erofs_decompress_pcluster(io->sb, pcl, pagepool); 959 } 960 } 961 962 static void z_erofs_decompressqueue_work(struct work_struct *work) 963 { 964 struct z_erofs_decompressqueue *bgq = 965 container_of(work, struct z_erofs_decompressqueue, u.work); 966 LIST_HEAD(pagepool); 967 968 DBG_BUGON(bgq->head == Z_EROFS_PCLUSTER_TAIL_CLOSED); 969 z_erofs_decompress_queue(bgq, &pagepool); 970 971 put_pages_list(&pagepool); 972 kvfree(bgq); 973 } 974 975 static struct page *pickup_page_for_submission(struct z_erofs_pcluster *pcl, 976 unsigned int nr, 977 struct list_head *pagepool, 978 struct address_space *mc, 979 gfp_t gfp) 980 { 981 const pgoff_t index = pcl->obj.index; 982 bool tocache = false; 983 984 struct address_space *mapping; 985 struct page *oldpage, *page; 986 987 compressed_page_t t; 988 int justfound; 989 990 repeat: 991 page = READ_ONCE(pcl->compressed_pages[nr]); 992 oldpage = page; 993 994 if (!page) 995 goto out_allocpage; 996 997 /* 998 * the cached page has not been allocated and 999 * an placeholder is out there, prepare it now. 1000 */ 1001 if (page == PAGE_UNALLOCATED) { 1002 tocache = true; 1003 goto out_allocpage; 1004 } 1005 1006 /* process the target tagged pointer */ 1007 t = tagptr_init(compressed_page_t, page); 1008 justfound = tagptr_unfold_tags(t); 1009 page = tagptr_unfold_ptr(t); 1010 1011 mapping = READ_ONCE(page->mapping); 1012 1013 /* 1014 * file-backed online pages in plcuster are all locked steady, 1015 * therefore it is impossible for `mapping' to be NULL. 1016 */ 1017 if (mapping && mapping != mc) 1018 /* ought to be unmanaged pages */ 1019 goto out; 1020 1021 /* directly return for shortlived page as well */ 1022 if (z_erofs_is_shortlived_page(page)) 1023 goto out; 1024 1025 lock_page(page); 1026 1027 /* only true if page reclaim goes wrong, should never happen */ 1028 DBG_BUGON(justfound && PagePrivate(page)); 1029 1030 /* the page is still in manage cache */ 1031 if (page->mapping == mc) { 1032 WRITE_ONCE(pcl->compressed_pages[nr], page); 1033 1034 ClearPageError(page); 1035 if (!PagePrivate(page)) { 1036 /* 1037 * impossible to be !PagePrivate(page) for 1038 * the current restriction as well if 1039 * the page is already in compressed_pages[]. 1040 */ 1041 DBG_BUGON(!justfound); 1042 1043 justfound = 0; 1044 set_page_private(page, (unsigned long)pcl); 1045 SetPagePrivate(page); 1046 } 1047 1048 /* no need to submit io if it is already up-to-date */ 1049 if (PageUptodate(page)) { 1050 unlock_page(page); 1051 page = NULL; 1052 } 1053 goto out; 1054 } 1055 1056 /* 1057 * the managed page has been truncated, it's unsafe to 1058 * reuse this one, let's allocate a new cache-managed page. 1059 */ 1060 DBG_BUGON(page->mapping); 1061 DBG_BUGON(!justfound); 1062 1063 tocache = true; 1064 unlock_page(page); 1065 put_page(page); 1066 out_allocpage: 1067 page = erofs_allocpage(pagepool, gfp | __GFP_NOFAIL); 1068 if (!tocache || add_to_page_cache_lru(page, mc, index + nr, gfp)) { 1069 /* turn into temporary page if fails */ 1070 set_page_private(page, Z_EROFS_SHORTLIVED_PAGE); 1071 tocache = false; 1072 } 1073 1074 if (oldpage != cmpxchg(&pcl->compressed_pages[nr], oldpage, page)) { 1075 if (tocache) { 1076 /* since it added to managed cache successfully */ 1077 unlock_page(page); 1078 put_page(page); 1079 } else { 1080 list_add(&page->lru, pagepool); 1081 } 1082 cond_resched(); 1083 goto repeat; 1084 } 1085 1086 if (tocache) { 1087 attach_page_private(page, pcl); 1088 /* drop a ref added by allocpage (then we have 2 refs here) */ 1089 put_page(page); 1090 } 1091 out: /* the only exit (for tracing and debugging) */ 1092 return page; 1093 } 1094 1095 static struct z_erofs_decompressqueue * 1096 jobqueue_init(struct super_block *sb, 1097 struct z_erofs_decompressqueue *fgq, bool *fg) 1098 { 1099 struct z_erofs_decompressqueue *q; 1100 1101 if (fg && !*fg) { 1102 q = kvzalloc(sizeof(*q), GFP_KERNEL | __GFP_NOWARN); 1103 if (!q) { 1104 *fg = true; 1105 goto fg_out; 1106 } 1107 INIT_WORK(&q->u.work, z_erofs_decompressqueue_work); 1108 } else { 1109 fg_out: 1110 q = fgq; 1111 init_waitqueue_head(&fgq->u.wait); 1112 atomic_set(&fgq->pending_bios, 0); 1113 } 1114 q->sb = sb; 1115 q->head = Z_EROFS_PCLUSTER_TAIL_CLOSED; 1116 return q; 1117 } 1118 1119 /* define decompression jobqueue types */ 1120 enum { 1121 JQ_BYPASS, 1122 JQ_SUBMIT, 1123 NR_JOBQUEUES, 1124 }; 1125 1126 static void *jobqueueset_init(struct super_block *sb, 1127 struct z_erofs_decompressqueue *q[], 1128 struct z_erofs_decompressqueue *fgq, bool *fg) 1129 { 1130 /* 1131 * if managed cache is enabled, bypass jobqueue is needed, 1132 * no need to read from device for all pclusters in this queue. 1133 */ 1134 q[JQ_BYPASS] = jobqueue_init(sb, fgq + JQ_BYPASS, NULL); 1135 q[JQ_SUBMIT] = jobqueue_init(sb, fgq + JQ_SUBMIT, fg); 1136 1137 return tagptr_cast_ptr(tagptr_fold(tagptr1_t, q[JQ_SUBMIT], *fg)); 1138 } 1139 1140 static void move_to_bypass_jobqueue(struct z_erofs_pcluster *pcl, 1141 z_erofs_next_pcluster_t qtail[], 1142 z_erofs_next_pcluster_t owned_head) 1143 { 1144 z_erofs_next_pcluster_t *const submit_qtail = qtail[JQ_SUBMIT]; 1145 z_erofs_next_pcluster_t *const bypass_qtail = qtail[JQ_BYPASS]; 1146 1147 DBG_BUGON(owned_head == Z_EROFS_PCLUSTER_TAIL_CLOSED); 1148 if (owned_head == Z_EROFS_PCLUSTER_TAIL) 1149 owned_head = Z_EROFS_PCLUSTER_TAIL_CLOSED; 1150 1151 WRITE_ONCE(pcl->next, Z_EROFS_PCLUSTER_TAIL_CLOSED); 1152 1153 WRITE_ONCE(*submit_qtail, owned_head); 1154 WRITE_ONCE(*bypass_qtail, &pcl->next); 1155 1156 qtail[JQ_BYPASS] = &pcl->next; 1157 } 1158 1159 static void z_erofs_submit_queue(struct super_block *sb, 1160 struct z_erofs_decompress_frontend *f, 1161 struct list_head *pagepool, 1162 struct z_erofs_decompressqueue *fgq, 1163 bool *force_fg) 1164 { 1165 struct erofs_sb_info *const sbi = EROFS_SB(sb); 1166 z_erofs_next_pcluster_t qtail[NR_JOBQUEUES]; 1167 struct z_erofs_decompressqueue *q[NR_JOBQUEUES]; 1168 void *bi_private; 1169 z_erofs_next_pcluster_t owned_head = f->clt.owned_head; 1170 /* since bio will be NULL, no need to initialize last_index */ 1171 pgoff_t last_index; 1172 unsigned int nr_bios = 0; 1173 struct bio *bio = NULL; 1174 1175 bi_private = jobqueueset_init(sb, q, fgq, force_fg); 1176 qtail[JQ_BYPASS] = &q[JQ_BYPASS]->head; 1177 qtail[JQ_SUBMIT] = &q[JQ_SUBMIT]->head; 1178 1179 /* by default, all need io submission */ 1180 q[JQ_SUBMIT]->head = owned_head; 1181 1182 do { 1183 struct z_erofs_pcluster *pcl; 1184 pgoff_t cur, end; 1185 unsigned int i = 0; 1186 bool bypass = true; 1187 1188 /* no possible 'owned_head' equals the following */ 1189 DBG_BUGON(owned_head == Z_EROFS_PCLUSTER_TAIL_CLOSED); 1190 DBG_BUGON(owned_head == Z_EROFS_PCLUSTER_NIL); 1191 1192 pcl = container_of(owned_head, struct z_erofs_pcluster, next); 1193 1194 cur = pcl->obj.index; 1195 end = cur + BIT(pcl->clusterbits); 1196 1197 /* close the main owned chain at first */ 1198 owned_head = cmpxchg(&pcl->next, Z_EROFS_PCLUSTER_TAIL, 1199 Z_EROFS_PCLUSTER_TAIL_CLOSED); 1200 1201 do { 1202 struct page *page; 1203 1204 page = pickup_page_for_submission(pcl, i++, pagepool, 1205 MNGD_MAPPING(sbi), 1206 GFP_NOFS); 1207 if (!page) 1208 continue; 1209 1210 if (bio && cur != last_index + 1) { 1211 submit_bio_retry: 1212 submit_bio(bio); 1213 bio = NULL; 1214 } 1215 1216 if (!bio) { 1217 bio = bio_alloc(GFP_NOIO, BIO_MAX_PAGES); 1218 1219 bio->bi_end_io = z_erofs_decompressqueue_endio; 1220 bio_set_dev(bio, sb->s_bdev); 1221 bio->bi_iter.bi_sector = (sector_t)cur << 1222 LOG_SECTORS_PER_BLOCK; 1223 bio->bi_private = bi_private; 1224 bio->bi_opf = REQ_OP_READ; 1225 if (f->readahead) 1226 bio->bi_opf |= REQ_RAHEAD; 1227 ++nr_bios; 1228 } 1229 1230 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) 1231 goto submit_bio_retry; 1232 1233 last_index = cur; 1234 bypass = false; 1235 } while (++cur < end); 1236 1237 if (!bypass) 1238 qtail[JQ_SUBMIT] = &pcl->next; 1239 else 1240 move_to_bypass_jobqueue(pcl, qtail, owned_head); 1241 } while (owned_head != Z_EROFS_PCLUSTER_TAIL); 1242 1243 if (bio) 1244 submit_bio(bio); 1245 1246 /* 1247 * although background is preferred, no one is pending for submission. 1248 * don't issue workqueue for decompression but drop it directly instead. 1249 */ 1250 if (!*force_fg && !nr_bios) { 1251 kvfree(q[JQ_SUBMIT]); 1252 return; 1253 } 1254 z_erofs_decompress_kickoff(q[JQ_SUBMIT], *force_fg, nr_bios); 1255 } 1256 1257 static void z_erofs_runqueue(struct super_block *sb, 1258 struct z_erofs_decompress_frontend *f, 1259 struct list_head *pagepool, bool force_fg) 1260 { 1261 struct z_erofs_decompressqueue io[NR_JOBQUEUES]; 1262 1263 if (f->clt.owned_head == Z_EROFS_PCLUSTER_TAIL) 1264 return; 1265 z_erofs_submit_queue(sb, f, pagepool, io, &force_fg); 1266 1267 /* handle bypass queue (no i/o pclusters) immediately */ 1268 z_erofs_decompress_queue(&io[JQ_BYPASS], pagepool); 1269 1270 if (!force_fg) 1271 return; 1272 1273 /* wait until all bios are completed */ 1274 io_wait_event(io[JQ_SUBMIT].u.wait, 1275 !atomic_read(&io[JQ_SUBMIT].pending_bios)); 1276 1277 /* handle synchronous decompress queue in the caller context */ 1278 z_erofs_decompress_queue(&io[JQ_SUBMIT], pagepool); 1279 } 1280 1281 static int z_erofs_readpage(struct file *file, struct page *page) 1282 { 1283 struct inode *const inode = page->mapping->host; 1284 struct z_erofs_decompress_frontend f = DECOMPRESS_FRONTEND_INIT(inode); 1285 int err; 1286 LIST_HEAD(pagepool); 1287 1288 trace_erofs_readpage(page, false); 1289 1290 f.headoffset = (erofs_off_t)page->index << PAGE_SHIFT; 1291 1292 err = z_erofs_do_read_page(&f, page); 1293 (void)z_erofs_collector_end(&f.clt); 1294 1295 /* if some compressed cluster ready, need submit them anyway */ 1296 z_erofs_runqueue(inode->i_sb, &f, &pagepool, true); 1297 1298 if (err) 1299 erofs_err(inode->i_sb, "failed to read, err [%d]", err); 1300 1301 if (f.map.mpage) 1302 put_page(f.map.mpage); 1303 1304 /* clean up the remaining free pages */ 1305 put_pages_list(&pagepool); 1306 return err; 1307 } 1308 1309 static void z_erofs_readahead(struct readahead_control *rac) 1310 { 1311 struct inode *const inode = rac->mapping->host; 1312 struct erofs_sb_info *const sbi = EROFS_I_SB(inode); 1313 1314 unsigned int nr_pages = readahead_count(rac); 1315 bool sync = (nr_pages <= sbi->ctx.max_sync_decompress_pages); 1316 struct z_erofs_decompress_frontend f = DECOMPRESS_FRONTEND_INIT(inode); 1317 struct page *page, *head = NULL; 1318 LIST_HEAD(pagepool); 1319 1320 trace_erofs_readpages(inode, readahead_index(rac), nr_pages, false); 1321 1322 f.readahead = true; 1323 f.headoffset = readahead_pos(rac); 1324 1325 while ((page = readahead_page(rac))) { 1326 prefetchw(&page->flags); 1327 1328 /* 1329 * A pure asynchronous readahead is indicated if 1330 * a PG_readahead marked page is hitted at first. 1331 * Let's also do asynchronous decompression for this case. 1332 */ 1333 sync &= !(PageReadahead(page) && !head); 1334 1335 set_page_private(page, (unsigned long)head); 1336 head = page; 1337 } 1338 1339 while (head) { 1340 struct page *page = head; 1341 int err; 1342 1343 /* traversal in reverse order */ 1344 head = (void *)page_private(page); 1345 1346 err = z_erofs_do_read_page(&f, page); 1347 if (err) 1348 erofs_err(inode->i_sb, 1349 "readahead error at page %lu @ nid %llu", 1350 page->index, EROFS_I(inode)->nid); 1351 put_page(page); 1352 } 1353 1354 (void)z_erofs_collector_end(&f.clt); 1355 1356 z_erofs_runqueue(inode->i_sb, &f, &pagepool, sync); 1357 1358 if (f.map.mpage) 1359 put_page(f.map.mpage); 1360 1361 /* clean up the remaining free pages */ 1362 put_pages_list(&pagepool); 1363 } 1364 1365 const struct address_space_operations z_erofs_aops = { 1366 .readpage = z_erofs_readpage, 1367 .readahead = z_erofs_readahead, 1368 }; 1369 1370