1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Copyright (C) 2018 HUAWEI, Inc. 4 * https://www.huawei.com/ 5 * Copyright (C) 2022 Alibaba Cloud 6 */ 7 #include "compress.h" 8 #include <linux/psi.h> 9 #include <linux/cpuhotplug.h> 10 #include <trace/events/erofs.h> 11 12 #define Z_EROFS_PCLUSTER_MAX_PAGES (Z_EROFS_PCLUSTER_MAX_SIZE / PAGE_SIZE) 13 #define Z_EROFS_INLINE_BVECS 2 14 15 /* 16 * let's leave a type here in case of introducing 17 * another tagged pointer later. 18 */ 19 typedef void *z_erofs_next_pcluster_t; 20 21 struct z_erofs_bvec { 22 struct page *page; 23 int offset; 24 unsigned int end; 25 }; 26 27 #define __Z_EROFS_BVSET(name, total) \ 28 struct name { \ 29 /* point to the next page which contains the following bvecs */ \ 30 struct page *nextpage; \ 31 struct z_erofs_bvec bvec[total]; \ 32 } 33 __Z_EROFS_BVSET(z_erofs_bvset,); 34 __Z_EROFS_BVSET(z_erofs_bvset_inline, Z_EROFS_INLINE_BVECS); 35 36 /* 37 * Structure fields follow one of the following exclusion rules. 38 * 39 * I: Modifiable by initialization/destruction paths and read-only 40 * for everyone else; 41 * 42 * L: Field should be protected by the pcluster lock; 43 * 44 * A: Field should be accessed / updated in atomic for parallelized code. 45 */ 46 struct z_erofs_pcluster { 47 struct erofs_workgroup obj; 48 struct mutex lock; 49 50 /* A: point to next chained pcluster or TAILs */ 51 z_erofs_next_pcluster_t next; 52 53 /* L: the maximum decompression size of this round */ 54 unsigned int length; 55 56 /* L: total number of bvecs */ 57 unsigned int vcnt; 58 59 /* I: page offset of start position of decompression */ 60 unsigned short pageofs_out; 61 62 /* I: page offset of inline compressed data */ 63 unsigned short pageofs_in; 64 65 union { 66 /* L: inline a certain number of bvec for bootstrap */ 67 struct z_erofs_bvset_inline bvset; 68 69 /* I: can be used to free the pcluster by RCU. */ 70 struct rcu_head rcu; 71 }; 72 73 union { 74 /* I: physical cluster size in pages */ 75 unsigned short pclusterpages; 76 77 /* I: tailpacking inline compressed size */ 78 unsigned short tailpacking_size; 79 }; 80 81 /* I: compression algorithm format */ 82 unsigned char algorithmformat; 83 84 /* L: whether partial decompression or not */ 85 bool partial; 86 87 /* L: indicate several pageofs_outs or not */ 88 bool multibases; 89 90 /* A: compressed bvecs (can be cached or inplaced pages) */ 91 struct z_erofs_bvec compressed_bvecs[]; 92 }; 93 94 /* let's avoid the valid 32-bit kernel addresses */ 95 96 /* the chained workgroup has't submitted io (still open) */ 97 #define Z_EROFS_PCLUSTER_TAIL ((void *)0x5F0ECAFE) 98 /* the chained workgroup has already submitted io */ 99 #define Z_EROFS_PCLUSTER_TAIL_CLOSED ((void *)0x5F0EDEAD) 100 101 #define Z_EROFS_PCLUSTER_NIL (NULL) 102 103 struct z_erofs_decompressqueue { 104 struct super_block *sb; 105 atomic_t pending_bios; 106 z_erofs_next_pcluster_t head; 107 108 union { 109 struct completion done; 110 struct work_struct work; 111 struct kthread_work kthread_work; 112 } u; 113 bool eio, sync; 114 }; 115 116 static inline bool z_erofs_is_inline_pcluster(struct z_erofs_pcluster *pcl) 117 { 118 return !pcl->obj.index; 119 } 120 121 static inline unsigned int z_erofs_pclusterpages(struct z_erofs_pcluster *pcl) 122 { 123 if (z_erofs_is_inline_pcluster(pcl)) 124 return 1; 125 return pcl->pclusterpages; 126 } 127 128 /* 129 * bit 30: I/O error occurred on this page 130 * bit 0 - 29: remaining parts to complete this page 131 */ 132 #define Z_EROFS_PAGE_EIO (1 << 30) 133 134 static inline void z_erofs_onlinepage_init(struct page *page) 135 { 136 union { 137 atomic_t o; 138 unsigned long v; 139 } u = { .o = ATOMIC_INIT(1) }; 140 141 set_page_private(page, u.v); 142 smp_wmb(); 143 SetPagePrivate(page); 144 } 145 146 static inline void z_erofs_onlinepage_split(struct page *page) 147 { 148 atomic_inc((atomic_t *)&page->private); 149 } 150 151 static inline void z_erofs_page_mark_eio(struct page *page) 152 { 153 int orig; 154 155 do { 156 orig = atomic_read((atomic_t *)&page->private); 157 } while (atomic_cmpxchg((atomic_t *)&page->private, orig, 158 orig | Z_EROFS_PAGE_EIO) != orig); 159 } 160 161 static inline void z_erofs_onlinepage_endio(struct page *page) 162 { 163 unsigned int v; 164 165 DBG_BUGON(!PagePrivate(page)); 166 v = atomic_dec_return((atomic_t *)&page->private); 167 if (!(v & ~Z_EROFS_PAGE_EIO)) { 168 set_page_private(page, 0); 169 ClearPagePrivate(page); 170 if (!(v & Z_EROFS_PAGE_EIO)) 171 SetPageUptodate(page); 172 unlock_page(page); 173 } 174 } 175 176 #define Z_EROFS_ONSTACK_PAGES 32 177 178 /* 179 * since pclustersize is variable for big pcluster feature, introduce slab 180 * pools implementation for different pcluster sizes. 181 */ 182 struct z_erofs_pcluster_slab { 183 struct kmem_cache *slab; 184 unsigned int maxpages; 185 char name[48]; 186 }; 187 188 #define _PCLP(n) { .maxpages = n } 189 190 static struct z_erofs_pcluster_slab pcluster_pool[] __read_mostly = { 191 _PCLP(1), _PCLP(4), _PCLP(16), _PCLP(64), _PCLP(128), 192 _PCLP(Z_EROFS_PCLUSTER_MAX_PAGES) 193 }; 194 195 struct z_erofs_bvec_iter { 196 struct page *bvpage; 197 struct z_erofs_bvset *bvset; 198 unsigned int nr, cur; 199 }; 200 201 static struct page *z_erofs_bvec_iter_end(struct z_erofs_bvec_iter *iter) 202 { 203 if (iter->bvpage) 204 kunmap_local(iter->bvset); 205 return iter->bvpage; 206 } 207 208 static struct page *z_erofs_bvset_flip(struct z_erofs_bvec_iter *iter) 209 { 210 unsigned long base = (unsigned long)((struct z_erofs_bvset *)0)->bvec; 211 /* have to access nextpage in advance, otherwise it will be unmapped */ 212 struct page *nextpage = iter->bvset->nextpage; 213 struct page *oldpage; 214 215 DBG_BUGON(!nextpage); 216 oldpage = z_erofs_bvec_iter_end(iter); 217 iter->bvpage = nextpage; 218 iter->bvset = kmap_local_page(nextpage); 219 iter->nr = (PAGE_SIZE - base) / sizeof(struct z_erofs_bvec); 220 iter->cur = 0; 221 return oldpage; 222 } 223 224 static void z_erofs_bvec_iter_begin(struct z_erofs_bvec_iter *iter, 225 struct z_erofs_bvset_inline *bvset, 226 unsigned int bootstrap_nr, 227 unsigned int cur) 228 { 229 *iter = (struct z_erofs_bvec_iter) { 230 .nr = bootstrap_nr, 231 .bvset = (struct z_erofs_bvset *)bvset, 232 }; 233 234 while (cur > iter->nr) { 235 cur -= iter->nr; 236 z_erofs_bvset_flip(iter); 237 } 238 iter->cur = cur; 239 } 240 241 static int z_erofs_bvec_enqueue(struct z_erofs_bvec_iter *iter, 242 struct z_erofs_bvec *bvec, 243 struct page **candidate_bvpage) 244 { 245 if (iter->cur >= iter->nr) { 246 struct page *nextpage = *candidate_bvpage; 247 248 if (!nextpage) { 249 nextpage = alloc_page(GFP_NOFS); 250 if (!nextpage) 251 return -ENOMEM; 252 set_page_private(nextpage, Z_EROFS_SHORTLIVED_PAGE); 253 } 254 DBG_BUGON(iter->bvset->nextpage); 255 iter->bvset->nextpage = nextpage; 256 z_erofs_bvset_flip(iter); 257 258 iter->bvset->nextpage = NULL; 259 *candidate_bvpage = NULL; 260 } 261 iter->bvset->bvec[iter->cur++] = *bvec; 262 return 0; 263 } 264 265 static void z_erofs_bvec_dequeue(struct z_erofs_bvec_iter *iter, 266 struct z_erofs_bvec *bvec, 267 struct page **old_bvpage) 268 { 269 if (iter->cur == iter->nr) 270 *old_bvpage = z_erofs_bvset_flip(iter); 271 else 272 *old_bvpage = NULL; 273 *bvec = iter->bvset->bvec[iter->cur++]; 274 } 275 276 static void z_erofs_destroy_pcluster_pool(void) 277 { 278 int i; 279 280 for (i = 0; i < ARRAY_SIZE(pcluster_pool); ++i) { 281 if (!pcluster_pool[i].slab) 282 continue; 283 kmem_cache_destroy(pcluster_pool[i].slab); 284 pcluster_pool[i].slab = NULL; 285 } 286 } 287 288 static int z_erofs_create_pcluster_pool(void) 289 { 290 struct z_erofs_pcluster_slab *pcs; 291 struct z_erofs_pcluster *a; 292 unsigned int size; 293 294 for (pcs = pcluster_pool; 295 pcs < pcluster_pool + ARRAY_SIZE(pcluster_pool); ++pcs) { 296 size = struct_size(a, compressed_bvecs, pcs->maxpages); 297 298 sprintf(pcs->name, "erofs_pcluster-%u", pcs->maxpages); 299 pcs->slab = kmem_cache_create(pcs->name, size, 0, 300 SLAB_RECLAIM_ACCOUNT, NULL); 301 if (pcs->slab) 302 continue; 303 304 z_erofs_destroy_pcluster_pool(); 305 return -ENOMEM; 306 } 307 return 0; 308 } 309 310 static struct z_erofs_pcluster *z_erofs_alloc_pcluster(unsigned int nrpages) 311 { 312 int i; 313 314 for (i = 0; i < ARRAY_SIZE(pcluster_pool); ++i) { 315 struct z_erofs_pcluster_slab *pcs = pcluster_pool + i; 316 struct z_erofs_pcluster *pcl; 317 318 if (nrpages > pcs->maxpages) 319 continue; 320 321 pcl = kmem_cache_zalloc(pcs->slab, GFP_NOFS); 322 if (!pcl) 323 return ERR_PTR(-ENOMEM); 324 pcl->pclusterpages = nrpages; 325 return pcl; 326 } 327 return ERR_PTR(-EINVAL); 328 } 329 330 static void z_erofs_free_pcluster(struct z_erofs_pcluster *pcl) 331 { 332 unsigned int pclusterpages = z_erofs_pclusterpages(pcl); 333 int i; 334 335 for (i = 0; i < ARRAY_SIZE(pcluster_pool); ++i) { 336 struct z_erofs_pcluster_slab *pcs = pcluster_pool + i; 337 338 if (pclusterpages > pcs->maxpages) 339 continue; 340 341 kmem_cache_free(pcs->slab, pcl); 342 return; 343 } 344 DBG_BUGON(1); 345 } 346 347 static struct workqueue_struct *z_erofs_workqueue __read_mostly; 348 349 #ifdef CONFIG_EROFS_FS_PCPU_KTHREAD 350 static struct kthread_worker __rcu **z_erofs_pcpu_workers; 351 352 static void erofs_destroy_percpu_workers(void) 353 { 354 struct kthread_worker *worker; 355 unsigned int cpu; 356 357 for_each_possible_cpu(cpu) { 358 worker = rcu_dereference_protected( 359 z_erofs_pcpu_workers[cpu], 1); 360 rcu_assign_pointer(z_erofs_pcpu_workers[cpu], NULL); 361 if (worker) 362 kthread_destroy_worker(worker); 363 } 364 kfree(z_erofs_pcpu_workers); 365 } 366 367 static struct kthread_worker *erofs_init_percpu_worker(int cpu) 368 { 369 struct kthread_worker *worker = 370 kthread_create_worker_on_cpu(cpu, 0, "erofs_worker/%u", cpu); 371 372 if (IS_ERR(worker)) 373 return worker; 374 if (IS_ENABLED(CONFIG_EROFS_FS_PCPU_KTHREAD_HIPRI)) 375 sched_set_fifo_low(worker->task); 376 return worker; 377 } 378 379 static int erofs_init_percpu_workers(void) 380 { 381 struct kthread_worker *worker; 382 unsigned int cpu; 383 384 z_erofs_pcpu_workers = kcalloc(num_possible_cpus(), 385 sizeof(struct kthread_worker *), GFP_ATOMIC); 386 if (!z_erofs_pcpu_workers) 387 return -ENOMEM; 388 389 for_each_online_cpu(cpu) { /* could miss cpu{off,on}line? */ 390 worker = erofs_init_percpu_worker(cpu); 391 if (!IS_ERR(worker)) 392 rcu_assign_pointer(z_erofs_pcpu_workers[cpu], worker); 393 } 394 return 0; 395 } 396 #else 397 static inline void erofs_destroy_percpu_workers(void) {} 398 static inline int erofs_init_percpu_workers(void) { return 0; } 399 #endif 400 401 #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_EROFS_FS_PCPU_KTHREAD) 402 static DEFINE_SPINLOCK(z_erofs_pcpu_worker_lock); 403 static enum cpuhp_state erofs_cpuhp_state; 404 405 static int erofs_cpu_online(unsigned int cpu) 406 { 407 struct kthread_worker *worker, *old; 408 409 worker = erofs_init_percpu_worker(cpu); 410 if (IS_ERR(worker)) 411 return PTR_ERR(worker); 412 413 spin_lock(&z_erofs_pcpu_worker_lock); 414 old = rcu_dereference_protected(z_erofs_pcpu_workers[cpu], 415 lockdep_is_held(&z_erofs_pcpu_worker_lock)); 416 if (!old) 417 rcu_assign_pointer(z_erofs_pcpu_workers[cpu], worker); 418 spin_unlock(&z_erofs_pcpu_worker_lock); 419 if (old) 420 kthread_destroy_worker(worker); 421 return 0; 422 } 423 424 static int erofs_cpu_offline(unsigned int cpu) 425 { 426 struct kthread_worker *worker; 427 428 spin_lock(&z_erofs_pcpu_worker_lock); 429 worker = rcu_dereference_protected(z_erofs_pcpu_workers[cpu], 430 lockdep_is_held(&z_erofs_pcpu_worker_lock)); 431 rcu_assign_pointer(z_erofs_pcpu_workers[cpu], NULL); 432 spin_unlock(&z_erofs_pcpu_worker_lock); 433 434 synchronize_rcu(); 435 if (worker) 436 kthread_destroy_worker(worker); 437 return 0; 438 } 439 440 static int erofs_cpu_hotplug_init(void) 441 { 442 int state; 443 444 state = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, 445 "fs/erofs:online", erofs_cpu_online, erofs_cpu_offline); 446 if (state < 0) 447 return state; 448 449 erofs_cpuhp_state = state; 450 return 0; 451 } 452 453 static void erofs_cpu_hotplug_destroy(void) 454 { 455 if (erofs_cpuhp_state) 456 cpuhp_remove_state_nocalls(erofs_cpuhp_state); 457 } 458 #else /* !CONFIG_HOTPLUG_CPU || !CONFIG_EROFS_FS_PCPU_KTHREAD */ 459 static inline int erofs_cpu_hotplug_init(void) { return 0; } 460 static inline void erofs_cpu_hotplug_destroy(void) {} 461 #endif 462 463 void z_erofs_exit_zip_subsystem(void) 464 { 465 erofs_cpu_hotplug_destroy(); 466 erofs_destroy_percpu_workers(); 467 destroy_workqueue(z_erofs_workqueue); 468 z_erofs_destroy_pcluster_pool(); 469 } 470 471 int __init z_erofs_init_zip_subsystem(void) 472 { 473 int err = z_erofs_create_pcluster_pool(); 474 475 if (err) 476 goto out_error_pcluster_pool; 477 478 z_erofs_workqueue = alloc_workqueue("erofs_worker", 479 WQ_UNBOUND | WQ_HIGHPRI, num_possible_cpus()); 480 if (!z_erofs_workqueue) { 481 err = -ENOMEM; 482 goto out_error_workqueue_init; 483 } 484 485 err = erofs_init_percpu_workers(); 486 if (err) 487 goto out_error_pcpu_worker; 488 489 err = erofs_cpu_hotplug_init(); 490 if (err < 0) 491 goto out_error_cpuhp_init; 492 return err; 493 494 out_error_cpuhp_init: 495 erofs_destroy_percpu_workers(); 496 out_error_pcpu_worker: 497 destroy_workqueue(z_erofs_workqueue); 498 out_error_workqueue_init: 499 z_erofs_destroy_pcluster_pool(); 500 out_error_pcluster_pool: 501 return err; 502 } 503 504 enum z_erofs_pclustermode { 505 Z_EROFS_PCLUSTER_INFLIGHT, 506 /* 507 * The current pclusters was the tail of an exist chain, in addition 508 * that the previous processed chained pclusters are all decided to 509 * be hooked up to it. 510 * A new chain will be created for the remaining pclusters which are 511 * not processed yet, so different from Z_EROFS_PCLUSTER_FOLLOWED, 512 * the next pcluster cannot reuse the whole page safely for inplace I/O 513 * in the following scenario: 514 * ________________________________________________________________ 515 * | tail (partial) page | head (partial) page | 516 * | (belongs to the next pcl) | (belongs to the current pcl) | 517 * |_______PCLUSTER_FOLLOWED______|________PCLUSTER_HOOKED__________| 518 */ 519 Z_EROFS_PCLUSTER_HOOKED, 520 /* 521 * a weak form of Z_EROFS_PCLUSTER_FOLLOWED, the difference is that it 522 * could be dispatched into bypass queue later due to uptodated managed 523 * pages. All related online pages cannot be reused for inplace I/O (or 524 * bvpage) since it can be directly decoded without I/O submission. 525 */ 526 Z_EROFS_PCLUSTER_FOLLOWED_NOINPLACE, 527 /* 528 * The current collection has been linked with the owned chain, and 529 * could also be linked with the remaining collections, which means 530 * if the processing page is the tail page of the collection, thus 531 * the current collection can safely use the whole page (since 532 * the previous collection is under control) for in-place I/O, as 533 * illustrated below: 534 * ________________________________________________________________ 535 * | tail (partial) page | head (partial) page | 536 * | (of the current cl) | (of the previous collection) | 537 * | PCLUSTER_FOLLOWED or | | 538 * |_____PCLUSTER_HOOKED__|___________PCLUSTER_FOLLOWED____________| 539 * 540 * [ (*) the above page can be used as inplace I/O. ] 541 */ 542 Z_EROFS_PCLUSTER_FOLLOWED, 543 }; 544 545 struct z_erofs_decompress_frontend { 546 struct inode *const inode; 547 struct erofs_map_blocks map; 548 struct z_erofs_bvec_iter biter; 549 550 struct page *candidate_bvpage; 551 struct z_erofs_pcluster *pcl, *tailpcl; 552 z_erofs_next_pcluster_t owned_head; 553 enum z_erofs_pclustermode mode; 554 555 /* used for applying cache strategy on the fly */ 556 bool backmost; 557 erofs_off_t headoffset; 558 559 /* a pointer used to pick up inplace I/O pages */ 560 unsigned int icur; 561 }; 562 563 #define DECOMPRESS_FRONTEND_INIT(__i) { \ 564 .inode = __i, .owned_head = Z_EROFS_PCLUSTER_TAIL, \ 565 .mode = Z_EROFS_PCLUSTER_FOLLOWED, .backmost = true } 566 567 static bool z_erofs_should_alloc_cache(struct z_erofs_decompress_frontend *fe) 568 { 569 unsigned int cachestrategy = EROFS_I_SB(fe->inode)->opt.cache_strategy; 570 571 if (cachestrategy <= EROFS_ZIP_CACHE_DISABLED) 572 return false; 573 574 if (fe->backmost) 575 return true; 576 577 if (cachestrategy >= EROFS_ZIP_CACHE_READAROUND && 578 fe->map.m_la < fe->headoffset) 579 return true; 580 581 return false; 582 } 583 584 static void z_erofs_bind_cache(struct z_erofs_decompress_frontend *fe, 585 struct page **pagepool) 586 { 587 struct address_space *mc = MNGD_MAPPING(EROFS_I_SB(fe->inode)); 588 struct z_erofs_pcluster *pcl = fe->pcl; 589 bool shouldalloc = z_erofs_should_alloc_cache(fe); 590 bool standalone = true; 591 /* 592 * optimistic allocation without direct reclaim since inplace I/O 593 * can be used if low memory otherwise. 594 */ 595 gfp_t gfp = (mapping_gfp_mask(mc) & ~__GFP_DIRECT_RECLAIM) | 596 __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN; 597 unsigned int i; 598 599 if (fe->mode < Z_EROFS_PCLUSTER_FOLLOWED) 600 return; 601 602 for (i = 0; i < pcl->pclusterpages; ++i) { 603 struct page *page; 604 void *t; /* mark pages just found for debugging */ 605 struct page *newpage = NULL; 606 607 /* the compressed page was loaded before */ 608 if (READ_ONCE(pcl->compressed_bvecs[i].page)) 609 continue; 610 611 page = find_get_page(mc, pcl->obj.index + i); 612 613 if (page) { 614 t = (void *)((unsigned long)page | 1); 615 } else { 616 /* I/O is needed, no possible to decompress directly */ 617 standalone = false; 618 if (!shouldalloc) 619 continue; 620 621 /* 622 * try to use cached I/O if page allocation 623 * succeeds or fallback to in-place I/O instead 624 * to avoid any direct reclaim. 625 */ 626 newpage = erofs_allocpage(pagepool, gfp); 627 if (!newpage) 628 continue; 629 set_page_private(newpage, Z_EROFS_PREALLOCATED_PAGE); 630 t = (void *)((unsigned long)newpage | 1); 631 } 632 633 if (!cmpxchg_relaxed(&pcl->compressed_bvecs[i].page, NULL, t)) 634 continue; 635 636 if (page) 637 put_page(page); 638 else if (newpage) 639 erofs_pagepool_add(pagepool, newpage); 640 } 641 642 /* 643 * don't do inplace I/O if all compressed pages are available in 644 * managed cache since it can be moved to the bypass queue instead. 645 */ 646 if (standalone) 647 fe->mode = Z_EROFS_PCLUSTER_FOLLOWED_NOINPLACE; 648 } 649 650 /* called by erofs_shrinker to get rid of all compressed_pages */ 651 int erofs_try_to_free_all_cached_pages(struct erofs_sb_info *sbi, 652 struct erofs_workgroup *grp) 653 { 654 struct z_erofs_pcluster *const pcl = 655 container_of(grp, struct z_erofs_pcluster, obj); 656 int i; 657 658 DBG_BUGON(z_erofs_is_inline_pcluster(pcl)); 659 /* 660 * refcount of workgroup is now freezed as 1, 661 * therefore no need to worry about available decompression users. 662 */ 663 for (i = 0; i < pcl->pclusterpages; ++i) { 664 struct page *page = pcl->compressed_bvecs[i].page; 665 666 if (!page) 667 continue; 668 669 /* block other users from reclaiming or migrating the page */ 670 if (!trylock_page(page)) 671 return -EBUSY; 672 673 if (!erofs_page_is_managed(sbi, page)) 674 continue; 675 676 /* barrier is implied in the following 'unlock_page' */ 677 WRITE_ONCE(pcl->compressed_bvecs[i].page, NULL); 678 detach_page_private(page); 679 unlock_page(page); 680 } 681 return 0; 682 } 683 684 int erofs_try_to_free_cached_page(struct page *page) 685 { 686 struct z_erofs_pcluster *const pcl = (void *)page_private(page); 687 int ret, i; 688 689 if (!erofs_workgroup_try_to_freeze(&pcl->obj, 1)) 690 return 0; 691 692 ret = 0; 693 DBG_BUGON(z_erofs_is_inline_pcluster(pcl)); 694 for (i = 0; i < pcl->pclusterpages; ++i) { 695 if (pcl->compressed_bvecs[i].page == page) { 696 WRITE_ONCE(pcl->compressed_bvecs[i].page, NULL); 697 ret = 1; 698 break; 699 } 700 } 701 erofs_workgroup_unfreeze(&pcl->obj, 1); 702 if (ret) 703 detach_page_private(page); 704 return ret; 705 } 706 707 static bool z_erofs_try_inplace_io(struct z_erofs_decompress_frontend *fe, 708 struct z_erofs_bvec *bvec) 709 { 710 struct z_erofs_pcluster *const pcl = fe->pcl; 711 712 while (fe->icur > 0) { 713 if (!cmpxchg(&pcl->compressed_bvecs[--fe->icur].page, 714 NULL, bvec->page)) { 715 pcl->compressed_bvecs[fe->icur] = *bvec; 716 return true; 717 } 718 } 719 return false; 720 } 721 722 /* callers must be with pcluster lock held */ 723 static int z_erofs_attach_page(struct z_erofs_decompress_frontend *fe, 724 struct z_erofs_bvec *bvec, bool exclusive) 725 { 726 int ret; 727 728 if (exclusive) { 729 /* give priority for inplaceio to use file pages first */ 730 if (z_erofs_try_inplace_io(fe, bvec)) 731 return 0; 732 /* otherwise, check if it can be used as a bvpage */ 733 if (fe->mode >= Z_EROFS_PCLUSTER_FOLLOWED && 734 !fe->candidate_bvpage) 735 fe->candidate_bvpage = bvec->page; 736 } 737 ret = z_erofs_bvec_enqueue(&fe->biter, bvec, &fe->candidate_bvpage); 738 fe->pcl->vcnt += (ret >= 0); 739 return ret; 740 } 741 742 static void z_erofs_try_to_claim_pcluster(struct z_erofs_decompress_frontend *f) 743 { 744 struct z_erofs_pcluster *pcl = f->pcl; 745 z_erofs_next_pcluster_t *owned_head = &f->owned_head; 746 747 /* type 1, nil pcluster (this pcluster doesn't belong to any chain.) */ 748 if (cmpxchg(&pcl->next, Z_EROFS_PCLUSTER_NIL, 749 *owned_head) == Z_EROFS_PCLUSTER_NIL) { 750 *owned_head = &pcl->next; 751 /* so we can attach this pcluster to our submission chain. */ 752 f->mode = Z_EROFS_PCLUSTER_FOLLOWED; 753 return; 754 } 755 756 /* 757 * type 2, link to the end of an existing open chain, be careful 758 * that its submission is controlled by the original attached chain. 759 */ 760 if (*owned_head != &pcl->next && pcl != f->tailpcl && 761 cmpxchg(&pcl->next, Z_EROFS_PCLUSTER_TAIL, 762 *owned_head) == Z_EROFS_PCLUSTER_TAIL) { 763 *owned_head = Z_EROFS_PCLUSTER_TAIL; 764 f->mode = Z_EROFS_PCLUSTER_HOOKED; 765 f->tailpcl = NULL; 766 return; 767 } 768 /* type 3, it belongs to a chain, but it isn't the end of the chain */ 769 f->mode = Z_EROFS_PCLUSTER_INFLIGHT; 770 } 771 772 static int z_erofs_register_pcluster(struct z_erofs_decompress_frontend *fe) 773 { 774 struct erofs_map_blocks *map = &fe->map; 775 bool ztailpacking = map->m_flags & EROFS_MAP_META; 776 struct z_erofs_pcluster *pcl; 777 struct erofs_workgroup *grp; 778 int err; 779 780 if (!(map->m_flags & EROFS_MAP_ENCODED) || 781 (!ztailpacking && !(map->m_pa >> PAGE_SHIFT))) { 782 DBG_BUGON(1); 783 return -EFSCORRUPTED; 784 } 785 786 /* no available pcluster, let's allocate one */ 787 pcl = z_erofs_alloc_pcluster(ztailpacking ? 1 : 788 map->m_plen >> PAGE_SHIFT); 789 if (IS_ERR(pcl)) 790 return PTR_ERR(pcl); 791 792 atomic_set(&pcl->obj.refcount, 1); 793 pcl->algorithmformat = map->m_algorithmformat; 794 pcl->length = 0; 795 pcl->partial = true; 796 797 /* new pclusters should be claimed as type 1, primary and followed */ 798 pcl->next = fe->owned_head; 799 pcl->pageofs_out = map->m_la & ~PAGE_MASK; 800 fe->mode = Z_EROFS_PCLUSTER_FOLLOWED; 801 802 /* 803 * lock all primary followed works before visible to others 804 * and mutex_trylock *never* fails for a new pcluster. 805 */ 806 mutex_init(&pcl->lock); 807 DBG_BUGON(!mutex_trylock(&pcl->lock)); 808 809 if (ztailpacking) { 810 pcl->obj.index = 0; /* which indicates ztailpacking */ 811 pcl->pageofs_in = erofs_blkoff(fe->inode->i_sb, map->m_pa); 812 pcl->tailpacking_size = map->m_plen; 813 } else { 814 pcl->obj.index = map->m_pa >> PAGE_SHIFT; 815 816 grp = erofs_insert_workgroup(fe->inode->i_sb, &pcl->obj); 817 if (IS_ERR(grp)) { 818 err = PTR_ERR(grp); 819 goto err_out; 820 } 821 822 if (grp != &pcl->obj) { 823 fe->pcl = container_of(grp, 824 struct z_erofs_pcluster, obj); 825 err = -EEXIST; 826 goto err_out; 827 } 828 } 829 /* used to check tail merging loop due to corrupted images */ 830 if (fe->owned_head == Z_EROFS_PCLUSTER_TAIL) 831 fe->tailpcl = pcl; 832 fe->owned_head = &pcl->next; 833 fe->pcl = pcl; 834 return 0; 835 836 err_out: 837 mutex_unlock(&pcl->lock); 838 z_erofs_free_pcluster(pcl); 839 return err; 840 } 841 842 static int z_erofs_collector_begin(struct z_erofs_decompress_frontend *fe) 843 { 844 struct erofs_map_blocks *map = &fe->map; 845 struct erofs_workgroup *grp = NULL; 846 int ret; 847 848 DBG_BUGON(fe->pcl); 849 850 /* must be Z_EROFS_PCLUSTER_TAIL or pointed to previous pcluster */ 851 DBG_BUGON(fe->owned_head == Z_EROFS_PCLUSTER_NIL); 852 DBG_BUGON(fe->owned_head == Z_EROFS_PCLUSTER_TAIL_CLOSED); 853 854 if (!(map->m_flags & EROFS_MAP_META)) { 855 grp = erofs_find_workgroup(fe->inode->i_sb, 856 map->m_pa >> PAGE_SHIFT); 857 } else if ((map->m_pa & ~PAGE_MASK) + map->m_plen > PAGE_SIZE) { 858 DBG_BUGON(1); 859 return -EFSCORRUPTED; 860 } 861 862 if (grp) { 863 fe->pcl = container_of(grp, struct z_erofs_pcluster, obj); 864 ret = -EEXIST; 865 } else { 866 ret = z_erofs_register_pcluster(fe); 867 } 868 869 if (ret == -EEXIST) { 870 mutex_lock(&fe->pcl->lock); 871 /* used to check tail merging loop due to corrupted images */ 872 if (fe->owned_head == Z_EROFS_PCLUSTER_TAIL) 873 fe->tailpcl = fe->pcl; 874 875 z_erofs_try_to_claim_pcluster(fe); 876 } else if (ret) { 877 return ret; 878 } 879 z_erofs_bvec_iter_begin(&fe->biter, &fe->pcl->bvset, 880 Z_EROFS_INLINE_BVECS, fe->pcl->vcnt); 881 /* since file-backed online pages are traversed in reverse order */ 882 fe->icur = z_erofs_pclusterpages(fe->pcl); 883 return 0; 884 } 885 886 /* 887 * keep in mind that no referenced pclusters will be freed 888 * only after a RCU grace period. 889 */ 890 static void z_erofs_rcu_callback(struct rcu_head *head) 891 { 892 z_erofs_free_pcluster(container_of(head, 893 struct z_erofs_pcluster, rcu)); 894 } 895 896 void erofs_workgroup_free_rcu(struct erofs_workgroup *grp) 897 { 898 struct z_erofs_pcluster *const pcl = 899 container_of(grp, struct z_erofs_pcluster, obj); 900 901 call_rcu(&pcl->rcu, z_erofs_rcu_callback); 902 } 903 904 static bool z_erofs_collector_end(struct z_erofs_decompress_frontend *fe) 905 { 906 struct z_erofs_pcluster *pcl = fe->pcl; 907 908 if (!pcl) 909 return false; 910 911 z_erofs_bvec_iter_end(&fe->biter); 912 mutex_unlock(&pcl->lock); 913 914 if (fe->candidate_bvpage) 915 fe->candidate_bvpage = NULL; 916 917 /* 918 * if all pending pages are added, don't hold its reference 919 * any longer if the pcluster isn't hosted by ourselves. 920 */ 921 if (fe->mode < Z_EROFS_PCLUSTER_FOLLOWED_NOINPLACE) 922 erofs_workgroup_put(&pcl->obj); 923 924 fe->pcl = NULL; 925 return true; 926 } 927 928 static int z_erofs_read_fragment(struct inode *inode, erofs_off_t pos, 929 struct page *page, unsigned int pageofs, 930 unsigned int len) 931 { 932 struct super_block *sb = inode->i_sb; 933 struct inode *packed_inode = EROFS_I_SB(inode)->packed_inode; 934 struct erofs_buf buf = __EROFS_BUF_INITIALIZER; 935 u8 *src, *dst; 936 unsigned int i, cnt; 937 938 if (!packed_inode) 939 return -EFSCORRUPTED; 940 941 buf.inode = packed_inode; 942 pos += EROFS_I(inode)->z_fragmentoff; 943 for (i = 0; i < len; i += cnt) { 944 cnt = min_t(unsigned int, len - i, 945 sb->s_blocksize - erofs_blkoff(sb, pos)); 946 src = erofs_bread(&buf, erofs_blknr(sb, pos), EROFS_KMAP); 947 if (IS_ERR(src)) { 948 erofs_put_metabuf(&buf); 949 return PTR_ERR(src); 950 } 951 952 dst = kmap_local_page(page); 953 memcpy(dst + pageofs + i, src + erofs_blkoff(sb, pos), cnt); 954 kunmap_local(dst); 955 pos += cnt; 956 } 957 erofs_put_metabuf(&buf); 958 return 0; 959 } 960 961 static int z_erofs_do_read_page(struct z_erofs_decompress_frontend *fe, 962 struct page *page, struct page **pagepool) 963 { 964 struct inode *const inode = fe->inode; 965 struct erofs_map_blocks *const map = &fe->map; 966 const loff_t offset = page_offset(page); 967 bool tight = true, exclusive; 968 unsigned int cur, end, spiltted; 969 int err = 0; 970 971 /* register locked file pages as online pages in pack */ 972 z_erofs_onlinepage_init(page); 973 974 spiltted = 0; 975 end = PAGE_SIZE; 976 repeat: 977 cur = end - 1; 978 979 if (offset + cur < map->m_la || 980 offset + cur >= map->m_la + map->m_llen) { 981 if (z_erofs_collector_end(fe)) 982 fe->backmost = false; 983 map->m_la = offset + cur; 984 map->m_llen = 0; 985 err = z_erofs_map_blocks_iter(inode, map, 0); 986 if (err) 987 goto out; 988 } else { 989 if (fe->pcl) 990 goto hitted; 991 /* didn't get a valid pcluster previously (very rare) */ 992 } 993 994 if (!(map->m_flags & EROFS_MAP_MAPPED) || 995 map->m_flags & EROFS_MAP_FRAGMENT) 996 goto hitted; 997 998 err = z_erofs_collector_begin(fe); 999 if (err) 1000 goto out; 1001 1002 if (z_erofs_is_inline_pcluster(fe->pcl)) { 1003 void *mp; 1004 1005 mp = erofs_read_metabuf(&fe->map.buf, inode->i_sb, 1006 erofs_blknr(inode->i_sb, map->m_pa), 1007 EROFS_NO_KMAP); 1008 if (IS_ERR(mp)) { 1009 err = PTR_ERR(mp); 1010 erofs_err(inode->i_sb, 1011 "failed to get inline page, err %d", err); 1012 goto out; 1013 } 1014 get_page(fe->map.buf.page); 1015 WRITE_ONCE(fe->pcl->compressed_bvecs[0].page, 1016 fe->map.buf.page); 1017 fe->mode = Z_EROFS_PCLUSTER_FOLLOWED_NOINPLACE; 1018 } else { 1019 /* bind cache first when cached decompression is preferred */ 1020 z_erofs_bind_cache(fe, pagepool); 1021 } 1022 hitted: 1023 /* 1024 * Ensure the current partial page belongs to this submit chain rather 1025 * than other concurrent submit chains or the noio(bypass) chain since 1026 * those chains are handled asynchronously thus the page cannot be used 1027 * for inplace I/O or bvpage (should be processed in a strict order.) 1028 */ 1029 tight &= (fe->mode >= Z_EROFS_PCLUSTER_HOOKED && 1030 fe->mode != Z_EROFS_PCLUSTER_FOLLOWED_NOINPLACE); 1031 1032 cur = end - min_t(unsigned int, offset + end - map->m_la, end); 1033 if (!(map->m_flags & EROFS_MAP_MAPPED)) { 1034 zero_user_segment(page, cur, end); 1035 goto next_part; 1036 } 1037 if (map->m_flags & EROFS_MAP_FRAGMENT) { 1038 unsigned int pageofs, skip, len; 1039 1040 if (offset > map->m_la) { 1041 pageofs = 0; 1042 skip = offset - map->m_la; 1043 } else { 1044 pageofs = map->m_la & ~PAGE_MASK; 1045 skip = 0; 1046 } 1047 len = min_t(unsigned int, map->m_llen - skip, end - cur); 1048 err = z_erofs_read_fragment(inode, skip, page, pageofs, len); 1049 if (err) 1050 goto out; 1051 ++spiltted; 1052 tight = false; 1053 goto next_part; 1054 } 1055 1056 exclusive = (!cur && (!spiltted || tight)); 1057 if (cur) 1058 tight &= (fe->mode >= Z_EROFS_PCLUSTER_FOLLOWED); 1059 1060 err = z_erofs_attach_page(fe, &((struct z_erofs_bvec) { 1061 .page = page, 1062 .offset = offset - map->m_la, 1063 .end = end, 1064 }), exclusive); 1065 if (err) 1066 goto out; 1067 1068 z_erofs_onlinepage_split(page); 1069 /* bump up the number of spiltted parts of a page */ 1070 ++spiltted; 1071 if (fe->pcl->pageofs_out != (map->m_la & ~PAGE_MASK)) 1072 fe->pcl->multibases = true; 1073 if (fe->pcl->length < offset + end - map->m_la) { 1074 fe->pcl->length = offset + end - map->m_la; 1075 fe->pcl->pageofs_out = map->m_la & ~PAGE_MASK; 1076 } 1077 if ((map->m_flags & EROFS_MAP_FULL_MAPPED) && 1078 !(map->m_flags & EROFS_MAP_PARTIAL_REF) && 1079 fe->pcl->length == map->m_llen) 1080 fe->pcl->partial = false; 1081 next_part: 1082 /* shorten the remaining extent to update progress */ 1083 map->m_llen = offset + cur - map->m_la; 1084 map->m_flags &= ~EROFS_MAP_FULL_MAPPED; 1085 1086 end = cur; 1087 if (end > 0) 1088 goto repeat; 1089 1090 out: 1091 if (err) 1092 z_erofs_page_mark_eio(page); 1093 z_erofs_onlinepage_endio(page); 1094 return err; 1095 } 1096 1097 static bool z_erofs_is_sync_decompress(struct erofs_sb_info *sbi, 1098 unsigned int readahead_pages) 1099 { 1100 /* auto: enable for read_folio, disable for readahead */ 1101 if ((sbi->opt.sync_decompress == EROFS_SYNC_DECOMPRESS_AUTO) && 1102 !readahead_pages) 1103 return true; 1104 1105 if ((sbi->opt.sync_decompress == EROFS_SYNC_DECOMPRESS_FORCE_ON) && 1106 (readahead_pages <= sbi->opt.max_sync_decompress_pages)) 1107 return true; 1108 1109 return false; 1110 } 1111 1112 static bool z_erofs_page_is_invalidated(struct page *page) 1113 { 1114 return !page->mapping && !z_erofs_is_shortlived_page(page); 1115 } 1116 1117 struct z_erofs_decompress_backend { 1118 struct page *onstack_pages[Z_EROFS_ONSTACK_PAGES]; 1119 struct super_block *sb; 1120 struct z_erofs_pcluster *pcl; 1121 1122 /* pages with the longest decompressed length for deduplication */ 1123 struct page **decompressed_pages; 1124 /* pages to keep the compressed data */ 1125 struct page **compressed_pages; 1126 1127 struct list_head decompressed_secondary_bvecs; 1128 struct page **pagepool; 1129 unsigned int onstack_used, nr_pages; 1130 }; 1131 1132 struct z_erofs_bvec_item { 1133 struct z_erofs_bvec bvec; 1134 struct list_head list; 1135 }; 1136 1137 static void z_erofs_do_decompressed_bvec(struct z_erofs_decompress_backend *be, 1138 struct z_erofs_bvec *bvec) 1139 { 1140 struct z_erofs_bvec_item *item; 1141 1142 if (!((bvec->offset + be->pcl->pageofs_out) & ~PAGE_MASK)) { 1143 unsigned int pgnr; 1144 1145 pgnr = (bvec->offset + be->pcl->pageofs_out) >> PAGE_SHIFT; 1146 DBG_BUGON(pgnr >= be->nr_pages); 1147 if (!be->decompressed_pages[pgnr]) { 1148 be->decompressed_pages[pgnr] = bvec->page; 1149 return; 1150 } 1151 } 1152 1153 /* (cold path) one pcluster is requested multiple times */ 1154 item = kmalloc(sizeof(*item), GFP_KERNEL | __GFP_NOFAIL); 1155 item->bvec = *bvec; 1156 list_add(&item->list, &be->decompressed_secondary_bvecs); 1157 } 1158 1159 static void z_erofs_fill_other_copies(struct z_erofs_decompress_backend *be, 1160 int err) 1161 { 1162 unsigned int off0 = be->pcl->pageofs_out; 1163 struct list_head *p, *n; 1164 1165 list_for_each_safe(p, n, &be->decompressed_secondary_bvecs) { 1166 struct z_erofs_bvec_item *bvi; 1167 unsigned int end, cur; 1168 void *dst, *src; 1169 1170 bvi = container_of(p, struct z_erofs_bvec_item, list); 1171 cur = bvi->bvec.offset < 0 ? -bvi->bvec.offset : 0; 1172 end = min_t(unsigned int, be->pcl->length - bvi->bvec.offset, 1173 bvi->bvec.end); 1174 dst = kmap_local_page(bvi->bvec.page); 1175 while (cur < end) { 1176 unsigned int pgnr, scur, len; 1177 1178 pgnr = (bvi->bvec.offset + cur + off0) >> PAGE_SHIFT; 1179 DBG_BUGON(pgnr >= be->nr_pages); 1180 1181 scur = bvi->bvec.offset + cur - 1182 ((pgnr << PAGE_SHIFT) - off0); 1183 len = min_t(unsigned int, end - cur, PAGE_SIZE - scur); 1184 if (!be->decompressed_pages[pgnr]) { 1185 err = -EFSCORRUPTED; 1186 cur += len; 1187 continue; 1188 } 1189 src = kmap_local_page(be->decompressed_pages[pgnr]); 1190 memcpy(dst + cur, src + scur, len); 1191 kunmap_local(src); 1192 cur += len; 1193 } 1194 kunmap_local(dst); 1195 if (err) 1196 z_erofs_page_mark_eio(bvi->bvec.page); 1197 z_erofs_onlinepage_endio(bvi->bvec.page); 1198 list_del(p); 1199 kfree(bvi); 1200 } 1201 } 1202 1203 static void z_erofs_parse_out_bvecs(struct z_erofs_decompress_backend *be) 1204 { 1205 struct z_erofs_pcluster *pcl = be->pcl; 1206 struct z_erofs_bvec_iter biter; 1207 struct page *old_bvpage; 1208 int i; 1209 1210 z_erofs_bvec_iter_begin(&biter, &pcl->bvset, Z_EROFS_INLINE_BVECS, 0); 1211 for (i = 0; i < pcl->vcnt; ++i) { 1212 struct z_erofs_bvec bvec; 1213 1214 z_erofs_bvec_dequeue(&biter, &bvec, &old_bvpage); 1215 1216 if (old_bvpage) 1217 z_erofs_put_shortlivedpage(be->pagepool, old_bvpage); 1218 1219 DBG_BUGON(z_erofs_page_is_invalidated(bvec.page)); 1220 z_erofs_do_decompressed_bvec(be, &bvec); 1221 } 1222 1223 old_bvpage = z_erofs_bvec_iter_end(&biter); 1224 if (old_bvpage) 1225 z_erofs_put_shortlivedpage(be->pagepool, old_bvpage); 1226 } 1227 1228 static int z_erofs_parse_in_bvecs(struct z_erofs_decompress_backend *be, 1229 bool *overlapped) 1230 { 1231 struct z_erofs_pcluster *pcl = be->pcl; 1232 unsigned int pclusterpages = z_erofs_pclusterpages(pcl); 1233 int i, err = 0; 1234 1235 *overlapped = false; 1236 for (i = 0; i < pclusterpages; ++i) { 1237 struct z_erofs_bvec *bvec = &pcl->compressed_bvecs[i]; 1238 struct page *page = bvec->page; 1239 1240 /* compressed pages ought to be present before decompressing */ 1241 if (!page) { 1242 DBG_BUGON(1); 1243 continue; 1244 } 1245 be->compressed_pages[i] = page; 1246 1247 if (z_erofs_is_inline_pcluster(pcl)) { 1248 if (!PageUptodate(page)) 1249 err = -EIO; 1250 continue; 1251 } 1252 1253 DBG_BUGON(z_erofs_page_is_invalidated(page)); 1254 if (!z_erofs_is_shortlived_page(page)) { 1255 if (erofs_page_is_managed(EROFS_SB(be->sb), page)) { 1256 if (!PageUptodate(page)) 1257 err = -EIO; 1258 continue; 1259 } 1260 z_erofs_do_decompressed_bvec(be, bvec); 1261 *overlapped = true; 1262 } 1263 } 1264 1265 if (err) 1266 return err; 1267 return 0; 1268 } 1269 1270 static int z_erofs_decompress_pcluster(struct z_erofs_decompress_backend *be, 1271 int err) 1272 { 1273 struct erofs_sb_info *const sbi = EROFS_SB(be->sb); 1274 struct z_erofs_pcluster *pcl = be->pcl; 1275 unsigned int pclusterpages = z_erofs_pclusterpages(pcl); 1276 const struct z_erofs_decompressor *decompressor = 1277 &erofs_decompressors[pcl->algorithmformat]; 1278 unsigned int i, inputsize; 1279 int err2; 1280 struct page *page; 1281 bool overlapped; 1282 1283 mutex_lock(&pcl->lock); 1284 be->nr_pages = PAGE_ALIGN(pcl->length + pcl->pageofs_out) >> PAGE_SHIFT; 1285 1286 /* allocate (de)compressed page arrays if cannot be kept on stack */ 1287 be->decompressed_pages = NULL; 1288 be->compressed_pages = NULL; 1289 be->onstack_used = 0; 1290 if (be->nr_pages <= Z_EROFS_ONSTACK_PAGES) { 1291 be->decompressed_pages = be->onstack_pages; 1292 be->onstack_used = be->nr_pages; 1293 memset(be->decompressed_pages, 0, 1294 sizeof(struct page *) * be->nr_pages); 1295 } 1296 1297 if (pclusterpages + be->onstack_used <= Z_EROFS_ONSTACK_PAGES) 1298 be->compressed_pages = be->onstack_pages + be->onstack_used; 1299 1300 if (!be->decompressed_pages) 1301 be->decompressed_pages = 1302 kvcalloc(be->nr_pages, sizeof(struct page *), 1303 GFP_KERNEL | __GFP_NOFAIL); 1304 if (!be->compressed_pages) 1305 be->compressed_pages = 1306 kvcalloc(pclusterpages, sizeof(struct page *), 1307 GFP_KERNEL | __GFP_NOFAIL); 1308 1309 z_erofs_parse_out_bvecs(be); 1310 err2 = z_erofs_parse_in_bvecs(be, &overlapped); 1311 if (err2) 1312 err = err2; 1313 if (err) 1314 goto out; 1315 1316 if (z_erofs_is_inline_pcluster(pcl)) 1317 inputsize = pcl->tailpacking_size; 1318 else 1319 inputsize = pclusterpages * PAGE_SIZE; 1320 1321 err = decompressor->decompress(&(struct z_erofs_decompress_req) { 1322 .sb = be->sb, 1323 .in = be->compressed_pages, 1324 .out = be->decompressed_pages, 1325 .pageofs_in = pcl->pageofs_in, 1326 .pageofs_out = pcl->pageofs_out, 1327 .inputsize = inputsize, 1328 .outputsize = pcl->length, 1329 .alg = pcl->algorithmformat, 1330 .inplace_io = overlapped, 1331 .partial_decoding = pcl->partial, 1332 .fillgaps = pcl->multibases, 1333 }, be->pagepool); 1334 1335 out: 1336 /* must handle all compressed pages before actual file pages */ 1337 if (z_erofs_is_inline_pcluster(pcl)) { 1338 page = pcl->compressed_bvecs[0].page; 1339 WRITE_ONCE(pcl->compressed_bvecs[0].page, NULL); 1340 put_page(page); 1341 } else { 1342 for (i = 0; i < pclusterpages; ++i) { 1343 page = pcl->compressed_bvecs[i].page; 1344 1345 if (erofs_page_is_managed(sbi, page)) 1346 continue; 1347 1348 /* recycle all individual short-lived pages */ 1349 (void)z_erofs_put_shortlivedpage(be->pagepool, page); 1350 WRITE_ONCE(pcl->compressed_bvecs[i].page, NULL); 1351 } 1352 } 1353 if (be->compressed_pages < be->onstack_pages || 1354 be->compressed_pages >= be->onstack_pages + Z_EROFS_ONSTACK_PAGES) 1355 kvfree(be->compressed_pages); 1356 z_erofs_fill_other_copies(be, err); 1357 1358 for (i = 0; i < be->nr_pages; ++i) { 1359 page = be->decompressed_pages[i]; 1360 if (!page) 1361 continue; 1362 1363 DBG_BUGON(z_erofs_page_is_invalidated(page)); 1364 1365 /* recycle all individual short-lived pages */ 1366 if (z_erofs_put_shortlivedpage(be->pagepool, page)) 1367 continue; 1368 if (err) 1369 z_erofs_page_mark_eio(page); 1370 z_erofs_onlinepage_endio(page); 1371 } 1372 1373 if (be->decompressed_pages != be->onstack_pages) 1374 kvfree(be->decompressed_pages); 1375 1376 pcl->length = 0; 1377 pcl->partial = true; 1378 pcl->multibases = false; 1379 pcl->bvset.nextpage = NULL; 1380 pcl->vcnt = 0; 1381 1382 /* pcluster lock MUST be taken before the following line */ 1383 WRITE_ONCE(pcl->next, Z_EROFS_PCLUSTER_NIL); 1384 mutex_unlock(&pcl->lock); 1385 return err; 1386 } 1387 1388 static void z_erofs_decompress_queue(const struct z_erofs_decompressqueue *io, 1389 struct page **pagepool) 1390 { 1391 struct z_erofs_decompress_backend be = { 1392 .sb = io->sb, 1393 .pagepool = pagepool, 1394 .decompressed_secondary_bvecs = 1395 LIST_HEAD_INIT(be.decompressed_secondary_bvecs), 1396 }; 1397 z_erofs_next_pcluster_t owned = io->head; 1398 1399 while (owned != Z_EROFS_PCLUSTER_TAIL_CLOSED) { 1400 /* impossible that 'owned' equals Z_EROFS_WORK_TPTR_TAIL */ 1401 DBG_BUGON(owned == Z_EROFS_PCLUSTER_TAIL); 1402 /* impossible that 'owned' equals Z_EROFS_PCLUSTER_NIL */ 1403 DBG_BUGON(owned == Z_EROFS_PCLUSTER_NIL); 1404 1405 be.pcl = container_of(owned, struct z_erofs_pcluster, next); 1406 owned = READ_ONCE(be.pcl->next); 1407 1408 z_erofs_decompress_pcluster(&be, io->eio ? -EIO : 0); 1409 erofs_workgroup_put(&be.pcl->obj); 1410 } 1411 } 1412 1413 static void z_erofs_decompressqueue_work(struct work_struct *work) 1414 { 1415 struct z_erofs_decompressqueue *bgq = 1416 container_of(work, struct z_erofs_decompressqueue, u.work); 1417 struct page *pagepool = NULL; 1418 1419 DBG_BUGON(bgq->head == Z_EROFS_PCLUSTER_TAIL_CLOSED); 1420 z_erofs_decompress_queue(bgq, &pagepool); 1421 erofs_release_pages(&pagepool); 1422 kvfree(bgq); 1423 } 1424 1425 #ifdef CONFIG_EROFS_FS_PCPU_KTHREAD 1426 static void z_erofs_decompressqueue_kthread_work(struct kthread_work *work) 1427 { 1428 z_erofs_decompressqueue_work((struct work_struct *)work); 1429 } 1430 #endif 1431 1432 static void z_erofs_decompress_kickoff(struct z_erofs_decompressqueue *io, 1433 int bios) 1434 { 1435 struct erofs_sb_info *const sbi = EROFS_SB(io->sb); 1436 1437 /* wake up the caller thread for sync decompression */ 1438 if (io->sync) { 1439 if (!atomic_add_return(bios, &io->pending_bios)) 1440 complete(&io->u.done); 1441 return; 1442 } 1443 1444 if (atomic_add_return(bios, &io->pending_bios)) 1445 return; 1446 /* Use (kthread_)work and sync decompression for atomic contexts only */ 1447 if (in_atomic() || irqs_disabled()) { 1448 #ifdef CONFIG_EROFS_FS_PCPU_KTHREAD 1449 struct kthread_worker *worker; 1450 1451 rcu_read_lock(); 1452 worker = rcu_dereference( 1453 z_erofs_pcpu_workers[raw_smp_processor_id()]); 1454 if (!worker) { 1455 INIT_WORK(&io->u.work, z_erofs_decompressqueue_work); 1456 queue_work(z_erofs_workqueue, &io->u.work); 1457 } else { 1458 kthread_queue_work(worker, &io->u.kthread_work); 1459 } 1460 rcu_read_unlock(); 1461 #else 1462 queue_work(z_erofs_workqueue, &io->u.work); 1463 #endif 1464 /* enable sync decompression for readahead */ 1465 if (sbi->opt.sync_decompress == EROFS_SYNC_DECOMPRESS_AUTO) 1466 sbi->opt.sync_decompress = EROFS_SYNC_DECOMPRESS_FORCE_ON; 1467 return; 1468 } 1469 z_erofs_decompressqueue_work(&io->u.work); 1470 } 1471 1472 static struct page *pickup_page_for_submission(struct z_erofs_pcluster *pcl, 1473 unsigned int nr, 1474 struct page **pagepool, 1475 struct address_space *mc) 1476 { 1477 const pgoff_t index = pcl->obj.index; 1478 gfp_t gfp = mapping_gfp_mask(mc); 1479 bool tocache = false; 1480 1481 struct address_space *mapping; 1482 struct page *oldpage, *page; 1483 int justfound; 1484 1485 repeat: 1486 page = READ_ONCE(pcl->compressed_bvecs[nr].page); 1487 oldpage = page; 1488 1489 if (!page) 1490 goto out_allocpage; 1491 1492 justfound = (unsigned long)page & 1UL; 1493 page = (struct page *)((unsigned long)page & ~1UL); 1494 1495 /* 1496 * preallocated cached pages, which is used to avoid direct reclaim 1497 * otherwise, it will go inplace I/O path instead. 1498 */ 1499 if (page->private == Z_EROFS_PREALLOCATED_PAGE) { 1500 WRITE_ONCE(pcl->compressed_bvecs[nr].page, page); 1501 set_page_private(page, 0); 1502 tocache = true; 1503 goto out_tocache; 1504 } 1505 mapping = READ_ONCE(page->mapping); 1506 1507 /* 1508 * file-backed online pages in plcuster are all locked steady, 1509 * therefore it is impossible for `mapping' to be NULL. 1510 */ 1511 if (mapping && mapping != mc) 1512 /* ought to be unmanaged pages */ 1513 goto out; 1514 1515 /* directly return for shortlived page as well */ 1516 if (z_erofs_is_shortlived_page(page)) 1517 goto out; 1518 1519 lock_page(page); 1520 1521 /* only true if page reclaim goes wrong, should never happen */ 1522 DBG_BUGON(justfound && PagePrivate(page)); 1523 1524 /* the page is still in manage cache */ 1525 if (page->mapping == mc) { 1526 WRITE_ONCE(pcl->compressed_bvecs[nr].page, page); 1527 1528 if (!PagePrivate(page)) { 1529 /* 1530 * impossible to be !PagePrivate(page) for 1531 * the current restriction as well if 1532 * the page is already in compressed_bvecs[]. 1533 */ 1534 DBG_BUGON(!justfound); 1535 1536 justfound = 0; 1537 set_page_private(page, (unsigned long)pcl); 1538 SetPagePrivate(page); 1539 } 1540 1541 /* no need to submit io if it is already up-to-date */ 1542 if (PageUptodate(page)) { 1543 unlock_page(page); 1544 page = NULL; 1545 } 1546 goto out; 1547 } 1548 1549 /* 1550 * the managed page has been truncated, it's unsafe to 1551 * reuse this one, let's allocate a new cache-managed page. 1552 */ 1553 DBG_BUGON(page->mapping); 1554 DBG_BUGON(!justfound); 1555 1556 tocache = true; 1557 unlock_page(page); 1558 put_page(page); 1559 out_allocpage: 1560 page = erofs_allocpage(pagepool, gfp | __GFP_NOFAIL); 1561 if (oldpage != cmpxchg(&pcl->compressed_bvecs[nr].page, 1562 oldpage, page)) { 1563 erofs_pagepool_add(pagepool, page); 1564 cond_resched(); 1565 goto repeat; 1566 } 1567 out_tocache: 1568 if (!tocache || add_to_page_cache_lru(page, mc, index + nr, gfp)) { 1569 /* turn into temporary page if fails (1 ref) */ 1570 set_page_private(page, Z_EROFS_SHORTLIVED_PAGE); 1571 goto out; 1572 } 1573 attach_page_private(page, pcl); 1574 /* drop a refcount added by allocpage (then we have 2 refs here) */ 1575 put_page(page); 1576 1577 out: /* the only exit (for tracing and debugging) */ 1578 return page; 1579 } 1580 1581 static struct z_erofs_decompressqueue *jobqueue_init(struct super_block *sb, 1582 struct z_erofs_decompressqueue *fgq, bool *fg) 1583 { 1584 struct z_erofs_decompressqueue *q; 1585 1586 if (fg && !*fg) { 1587 q = kvzalloc(sizeof(*q), GFP_KERNEL | __GFP_NOWARN); 1588 if (!q) { 1589 *fg = true; 1590 goto fg_out; 1591 } 1592 #ifdef CONFIG_EROFS_FS_PCPU_KTHREAD 1593 kthread_init_work(&q->u.kthread_work, 1594 z_erofs_decompressqueue_kthread_work); 1595 #else 1596 INIT_WORK(&q->u.work, z_erofs_decompressqueue_work); 1597 #endif 1598 } else { 1599 fg_out: 1600 q = fgq; 1601 init_completion(&fgq->u.done); 1602 atomic_set(&fgq->pending_bios, 0); 1603 q->eio = false; 1604 q->sync = true; 1605 } 1606 q->sb = sb; 1607 q->head = Z_EROFS_PCLUSTER_TAIL_CLOSED; 1608 return q; 1609 } 1610 1611 /* define decompression jobqueue types */ 1612 enum { 1613 JQ_BYPASS, 1614 JQ_SUBMIT, 1615 NR_JOBQUEUES, 1616 }; 1617 1618 static void move_to_bypass_jobqueue(struct z_erofs_pcluster *pcl, 1619 z_erofs_next_pcluster_t qtail[], 1620 z_erofs_next_pcluster_t owned_head) 1621 { 1622 z_erofs_next_pcluster_t *const submit_qtail = qtail[JQ_SUBMIT]; 1623 z_erofs_next_pcluster_t *const bypass_qtail = qtail[JQ_BYPASS]; 1624 1625 DBG_BUGON(owned_head == Z_EROFS_PCLUSTER_TAIL_CLOSED); 1626 if (owned_head == Z_EROFS_PCLUSTER_TAIL) 1627 owned_head = Z_EROFS_PCLUSTER_TAIL_CLOSED; 1628 1629 WRITE_ONCE(pcl->next, Z_EROFS_PCLUSTER_TAIL_CLOSED); 1630 1631 WRITE_ONCE(*submit_qtail, owned_head); 1632 WRITE_ONCE(*bypass_qtail, &pcl->next); 1633 1634 qtail[JQ_BYPASS] = &pcl->next; 1635 } 1636 1637 static void z_erofs_decompressqueue_endio(struct bio *bio) 1638 { 1639 struct z_erofs_decompressqueue *q = bio->bi_private; 1640 blk_status_t err = bio->bi_status; 1641 struct bio_vec *bvec; 1642 struct bvec_iter_all iter_all; 1643 1644 bio_for_each_segment_all(bvec, bio, iter_all) { 1645 struct page *page = bvec->bv_page; 1646 1647 DBG_BUGON(PageUptodate(page)); 1648 DBG_BUGON(z_erofs_page_is_invalidated(page)); 1649 1650 if (erofs_page_is_managed(EROFS_SB(q->sb), page)) { 1651 if (!err) 1652 SetPageUptodate(page); 1653 unlock_page(page); 1654 } 1655 } 1656 if (err) 1657 q->eio = true; 1658 z_erofs_decompress_kickoff(q, -1); 1659 bio_put(bio); 1660 } 1661 1662 static void z_erofs_submit_queue(struct z_erofs_decompress_frontend *f, 1663 struct page **pagepool, 1664 struct z_erofs_decompressqueue *fgq, 1665 bool *force_fg, bool readahead) 1666 { 1667 struct super_block *sb = f->inode->i_sb; 1668 struct address_space *mc = MNGD_MAPPING(EROFS_SB(sb)); 1669 z_erofs_next_pcluster_t qtail[NR_JOBQUEUES]; 1670 struct z_erofs_decompressqueue *q[NR_JOBQUEUES]; 1671 z_erofs_next_pcluster_t owned_head = f->owned_head; 1672 /* bio is NULL initially, so no need to initialize last_{index,bdev} */ 1673 pgoff_t last_index; 1674 struct block_device *last_bdev; 1675 unsigned int nr_bios = 0; 1676 struct bio *bio = NULL; 1677 unsigned long pflags; 1678 int memstall = 0; 1679 1680 /* 1681 * if managed cache is enabled, bypass jobqueue is needed, 1682 * no need to read from device for all pclusters in this queue. 1683 */ 1684 q[JQ_BYPASS] = jobqueue_init(sb, fgq + JQ_BYPASS, NULL); 1685 q[JQ_SUBMIT] = jobqueue_init(sb, fgq + JQ_SUBMIT, force_fg); 1686 1687 qtail[JQ_BYPASS] = &q[JQ_BYPASS]->head; 1688 qtail[JQ_SUBMIT] = &q[JQ_SUBMIT]->head; 1689 1690 /* by default, all need io submission */ 1691 q[JQ_SUBMIT]->head = owned_head; 1692 1693 do { 1694 struct erofs_map_dev mdev; 1695 struct z_erofs_pcluster *pcl; 1696 pgoff_t cur, end; 1697 unsigned int i = 0; 1698 bool bypass = true; 1699 1700 /* no possible 'owned_head' equals the following */ 1701 DBG_BUGON(owned_head == Z_EROFS_PCLUSTER_TAIL_CLOSED); 1702 DBG_BUGON(owned_head == Z_EROFS_PCLUSTER_NIL); 1703 1704 pcl = container_of(owned_head, struct z_erofs_pcluster, next); 1705 1706 /* close the main owned chain at first */ 1707 owned_head = cmpxchg(&pcl->next, Z_EROFS_PCLUSTER_TAIL, 1708 Z_EROFS_PCLUSTER_TAIL_CLOSED); 1709 if (z_erofs_is_inline_pcluster(pcl)) { 1710 move_to_bypass_jobqueue(pcl, qtail, owned_head); 1711 continue; 1712 } 1713 1714 /* no device id here, thus it will always succeed */ 1715 mdev = (struct erofs_map_dev) { 1716 .m_pa = erofs_pos(sb, pcl->obj.index), 1717 }; 1718 (void)erofs_map_dev(sb, &mdev); 1719 1720 cur = erofs_blknr(sb, mdev.m_pa); 1721 end = cur + pcl->pclusterpages; 1722 1723 do { 1724 struct page *page; 1725 1726 page = pickup_page_for_submission(pcl, i++, pagepool, 1727 mc); 1728 if (!page) 1729 continue; 1730 1731 if (bio && (cur != last_index + 1 || 1732 last_bdev != mdev.m_bdev)) { 1733 submit_bio_retry: 1734 submit_bio(bio); 1735 if (memstall) { 1736 psi_memstall_leave(&pflags); 1737 memstall = 0; 1738 } 1739 bio = NULL; 1740 } 1741 1742 if (unlikely(PageWorkingset(page)) && !memstall) { 1743 psi_memstall_enter(&pflags); 1744 memstall = 1; 1745 } 1746 1747 if (!bio) { 1748 bio = bio_alloc(mdev.m_bdev, BIO_MAX_VECS, 1749 REQ_OP_READ, GFP_NOIO); 1750 bio->bi_end_io = z_erofs_decompressqueue_endio; 1751 1752 last_bdev = mdev.m_bdev; 1753 bio->bi_iter.bi_sector = (sector_t)cur << 1754 (sb->s_blocksize_bits - 9); 1755 bio->bi_private = q[JQ_SUBMIT]; 1756 if (readahead) 1757 bio->bi_opf |= REQ_RAHEAD; 1758 ++nr_bios; 1759 } 1760 1761 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) 1762 goto submit_bio_retry; 1763 1764 last_index = cur; 1765 bypass = false; 1766 } while (++cur < end); 1767 1768 if (!bypass) 1769 qtail[JQ_SUBMIT] = &pcl->next; 1770 else 1771 move_to_bypass_jobqueue(pcl, qtail, owned_head); 1772 } while (owned_head != Z_EROFS_PCLUSTER_TAIL); 1773 1774 if (bio) { 1775 submit_bio(bio); 1776 if (memstall) 1777 psi_memstall_leave(&pflags); 1778 } 1779 1780 /* 1781 * although background is preferred, no one is pending for submission. 1782 * don't issue decompression but drop it directly instead. 1783 */ 1784 if (!*force_fg && !nr_bios) { 1785 kvfree(q[JQ_SUBMIT]); 1786 return; 1787 } 1788 z_erofs_decompress_kickoff(q[JQ_SUBMIT], nr_bios); 1789 } 1790 1791 static void z_erofs_runqueue(struct z_erofs_decompress_frontend *f, 1792 struct page **pagepool, bool force_fg, bool ra) 1793 { 1794 struct z_erofs_decompressqueue io[NR_JOBQUEUES]; 1795 1796 if (f->owned_head == Z_EROFS_PCLUSTER_TAIL) 1797 return; 1798 z_erofs_submit_queue(f, pagepool, io, &force_fg, ra); 1799 1800 /* handle bypass queue (no i/o pclusters) immediately */ 1801 z_erofs_decompress_queue(&io[JQ_BYPASS], pagepool); 1802 1803 if (!force_fg) 1804 return; 1805 1806 /* wait until all bios are completed */ 1807 wait_for_completion_io(&io[JQ_SUBMIT].u.done); 1808 1809 /* handle synchronous decompress queue in the caller context */ 1810 z_erofs_decompress_queue(&io[JQ_SUBMIT], pagepool); 1811 } 1812 1813 /* 1814 * Since partial uptodate is still unimplemented for now, we have to use 1815 * approximate readmore strategies as a start. 1816 */ 1817 static void z_erofs_pcluster_readmore(struct z_erofs_decompress_frontend *f, 1818 struct readahead_control *rac, 1819 struct page **pagepool, bool backmost) 1820 { 1821 struct inode *inode = f->inode; 1822 struct erofs_map_blocks *map = &f->map; 1823 erofs_off_t cur, end, headoffset = f->headoffset; 1824 int err; 1825 1826 if (backmost) { 1827 if (rac) 1828 end = headoffset + readahead_length(rac) - 1; 1829 else 1830 end = headoffset + PAGE_SIZE - 1; 1831 map->m_la = end; 1832 err = z_erofs_map_blocks_iter(inode, map, 1833 EROFS_GET_BLOCKS_READMORE); 1834 if (err) 1835 return; 1836 1837 /* expand ra for the trailing edge if readahead */ 1838 if (rac) { 1839 cur = round_up(map->m_la + map->m_llen, PAGE_SIZE); 1840 readahead_expand(rac, headoffset, cur - headoffset); 1841 return; 1842 } 1843 end = round_up(end, PAGE_SIZE); 1844 } else { 1845 end = round_up(map->m_la, PAGE_SIZE); 1846 1847 if (!map->m_llen) 1848 return; 1849 } 1850 1851 cur = map->m_la + map->m_llen - 1; 1852 while (cur >= end) { 1853 pgoff_t index = cur >> PAGE_SHIFT; 1854 struct page *page; 1855 1856 page = erofs_grab_cache_page_nowait(inode->i_mapping, index); 1857 if (page) { 1858 if (PageUptodate(page)) { 1859 unlock_page(page); 1860 } else { 1861 err = z_erofs_do_read_page(f, page, pagepool); 1862 if (err) 1863 erofs_err(inode->i_sb, 1864 "readmore error at page %lu @ nid %llu", 1865 index, EROFS_I(inode)->nid); 1866 } 1867 put_page(page); 1868 } 1869 1870 if (cur < PAGE_SIZE) 1871 break; 1872 cur = (index << PAGE_SHIFT) - 1; 1873 } 1874 } 1875 1876 static int z_erofs_read_folio(struct file *file, struct folio *folio) 1877 { 1878 struct page *page = &folio->page; 1879 struct inode *const inode = page->mapping->host; 1880 struct erofs_sb_info *const sbi = EROFS_I_SB(inode); 1881 struct z_erofs_decompress_frontend f = DECOMPRESS_FRONTEND_INIT(inode); 1882 struct page *pagepool = NULL; 1883 int err; 1884 1885 trace_erofs_readpage(page, false); 1886 f.headoffset = (erofs_off_t)page->index << PAGE_SHIFT; 1887 1888 z_erofs_pcluster_readmore(&f, NULL, &pagepool, true); 1889 err = z_erofs_do_read_page(&f, page, &pagepool); 1890 z_erofs_pcluster_readmore(&f, NULL, &pagepool, false); 1891 1892 (void)z_erofs_collector_end(&f); 1893 1894 /* if some compressed cluster ready, need submit them anyway */ 1895 z_erofs_runqueue(&f, &pagepool, z_erofs_is_sync_decompress(sbi, 0), 1896 false); 1897 1898 if (err) 1899 erofs_err(inode->i_sb, "failed to read, err [%d]", err); 1900 1901 erofs_put_metabuf(&f.map.buf); 1902 erofs_release_pages(&pagepool); 1903 return err; 1904 } 1905 1906 static void z_erofs_readahead(struct readahead_control *rac) 1907 { 1908 struct inode *const inode = rac->mapping->host; 1909 struct erofs_sb_info *const sbi = EROFS_I_SB(inode); 1910 struct z_erofs_decompress_frontend f = DECOMPRESS_FRONTEND_INIT(inode); 1911 struct page *pagepool = NULL, *head = NULL, *page; 1912 unsigned int nr_pages; 1913 1914 f.headoffset = readahead_pos(rac); 1915 1916 z_erofs_pcluster_readmore(&f, rac, &pagepool, true); 1917 nr_pages = readahead_count(rac); 1918 trace_erofs_readpages(inode, readahead_index(rac), nr_pages, false); 1919 1920 while ((page = readahead_page(rac))) { 1921 set_page_private(page, (unsigned long)head); 1922 head = page; 1923 } 1924 1925 while (head) { 1926 struct page *page = head; 1927 int err; 1928 1929 /* traversal in reverse order */ 1930 head = (void *)page_private(page); 1931 1932 err = z_erofs_do_read_page(&f, page, &pagepool); 1933 if (err) 1934 erofs_err(inode->i_sb, 1935 "readahead error at page %lu @ nid %llu", 1936 page->index, EROFS_I(inode)->nid); 1937 put_page(page); 1938 } 1939 z_erofs_pcluster_readmore(&f, rac, &pagepool, false); 1940 (void)z_erofs_collector_end(&f); 1941 1942 z_erofs_runqueue(&f, &pagepool, 1943 z_erofs_is_sync_decompress(sbi, nr_pages), true); 1944 erofs_put_metabuf(&f.map.buf); 1945 erofs_release_pages(&pagepool); 1946 } 1947 1948 const struct address_space_operations z_erofs_aops = { 1949 .read_folio = z_erofs_read_folio, 1950 .readahead = z_erofs_readahead, 1951 }; 1952