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/prefetch.h> 9 #include <linux/psi.h> 10 #include <linux/cpuhotplug.h> 11 #include <trace/events/erofs.h> 12 13 #define Z_EROFS_PCLUSTER_MAX_PAGES (Z_EROFS_PCLUSTER_MAX_SIZE / PAGE_SIZE) 14 #define Z_EROFS_INLINE_BVECS 2 15 16 /* 17 * let's leave a type here in case of introducing 18 * another tagged pointer later. 19 */ 20 typedef void *z_erofs_next_pcluster_t; 21 22 struct z_erofs_bvec { 23 struct page *page; 24 int offset; 25 unsigned int end; 26 }; 27 28 #define __Z_EROFS_BVSET(name, total) \ 29 struct name { \ 30 /* point to the next page which contains the following bvecs */ \ 31 struct page *nextpage; \ 32 struct z_erofs_bvec bvec[total]; \ 33 } 34 __Z_EROFS_BVSET(z_erofs_bvset,); 35 __Z_EROFS_BVSET(z_erofs_bvset_inline, Z_EROFS_INLINE_BVECS); 36 37 /* 38 * Structure fields follow one of the following exclusion rules. 39 * 40 * I: Modifiable by initialization/destruction paths and read-only 41 * for everyone else; 42 * 43 * L: Field should be protected by the pcluster lock; 44 * 45 * A: Field should be accessed / updated in atomic for parallelized code. 46 */ 47 struct z_erofs_pcluster { 48 struct erofs_workgroup obj; 49 struct mutex lock; 50 51 /* A: point to next chained pcluster or TAILs */ 52 z_erofs_next_pcluster_t next; 53 54 /* L: the maximum decompression size of this round */ 55 unsigned int length; 56 57 /* L: total number of bvecs */ 58 unsigned int vcnt; 59 60 /* I: page offset of start position of decompression */ 61 unsigned short pageofs_out; 62 63 /* I: page offset of inline compressed data */ 64 unsigned short pageofs_in; 65 66 union { 67 /* L: inline a certain number of bvec for bootstrap */ 68 struct z_erofs_bvset_inline bvset; 69 70 /* I: can be used to free the pcluster by RCU. */ 71 struct rcu_head rcu; 72 }; 73 74 union { 75 /* I: physical cluster size in pages */ 76 unsigned short pclusterpages; 77 78 /* I: tailpacking inline compressed size */ 79 unsigned short tailpacking_size; 80 }; 81 82 /* I: compression algorithm format */ 83 unsigned char algorithmformat; 84 85 /* L: whether partial decompression or not */ 86 bool partial; 87 88 /* L: indicate several pageofs_outs or not */ 89 bool multibases; 90 91 /* A: compressed bvecs (can be cached or inplaced pages) */ 92 struct z_erofs_bvec compressed_bvecs[]; 93 }; 94 95 /* let's avoid the valid 32-bit kernel addresses */ 96 97 /* the chained workgroup has't submitted io (still open) */ 98 #define Z_EROFS_PCLUSTER_TAIL ((void *)0x5F0ECAFE) 99 /* the chained workgroup has already submitted io */ 100 #define Z_EROFS_PCLUSTER_TAIL_CLOSED ((void *)0x5F0EDEAD) 101 102 #define Z_EROFS_PCLUSTER_NIL (NULL) 103 104 struct z_erofs_decompressqueue { 105 struct super_block *sb; 106 atomic_t pending_bios; 107 z_erofs_next_pcluster_t head; 108 109 union { 110 struct completion done; 111 struct work_struct work; 112 struct kthread_work kthread_work; 113 } u; 114 bool eio, sync; 115 }; 116 117 static inline bool z_erofs_is_inline_pcluster(struct z_erofs_pcluster *pcl) 118 { 119 return !pcl->obj.index; 120 } 121 122 static inline unsigned int z_erofs_pclusterpages(struct z_erofs_pcluster *pcl) 123 { 124 if (z_erofs_is_inline_pcluster(pcl)) 125 return 1; 126 return pcl->pclusterpages; 127 } 128 129 /* 130 * bit 30: I/O error occurred on this page 131 * bit 0 - 29: remaining parts to complete this page 132 */ 133 #define Z_EROFS_PAGE_EIO (1 << 30) 134 135 static inline void z_erofs_onlinepage_init(struct page *page) 136 { 137 union { 138 atomic_t o; 139 unsigned long v; 140 } u = { .o = ATOMIC_INIT(1) }; 141 142 set_page_private(page, u.v); 143 smp_wmb(); 144 SetPagePrivate(page); 145 } 146 147 static inline void z_erofs_onlinepage_split(struct page *page) 148 { 149 atomic_inc((atomic_t *)&page->private); 150 } 151 152 static inline void z_erofs_page_mark_eio(struct page *page) 153 { 154 int orig; 155 156 do { 157 orig = atomic_read((atomic_t *)&page->private); 158 } while (atomic_cmpxchg((atomic_t *)&page->private, orig, 159 orig | Z_EROFS_PAGE_EIO) != orig); 160 } 161 162 static inline void z_erofs_onlinepage_endio(struct page *page) 163 { 164 unsigned int v; 165 166 DBG_BUGON(!PagePrivate(page)); 167 v = atomic_dec_return((atomic_t *)&page->private); 168 if (!(v & ~Z_EROFS_PAGE_EIO)) { 169 set_page_private(page, 0); 170 ClearPagePrivate(page); 171 if (!(v & Z_EROFS_PAGE_EIO)) 172 SetPageUptodate(page); 173 unlock_page(page); 174 } 175 } 176 177 #define Z_EROFS_ONSTACK_PAGES 32 178 179 /* 180 * since pclustersize is variable for big pcluster feature, introduce slab 181 * pools implementation for different pcluster sizes. 182 */ 183 struct z_erofs_pcluster_slab { 184 struct kmem_cache *slab; 185 unsigned int maxpages; 186 char name[48]; 187 }; 188 189 #define _PCLP(n) { .maxpages = n } 190 191 static struct z_erofs_pcluster_slab pcluster_pool[] __read_mostly = { 192 _PCLP(1), _PCLP(4), _PCLP(16), _PCLP(64), _PCLP(128), 193 _PCLP(Z_EROFS_PCLUSTER_MAX_PAGES) 194 }; 195 196 struct z_erofs_bvec_iter { 197 struct page *bvpage; 198 struct z_erofs_bvset *bvset; 199 unsigned int nr, cur; 200 }; 201 202 static struct page *z_erofs_bvec_iter_end(struct z_erofs_bvec_iter *iter) 203 { 204 if (iter->bvpage) 205 kunmap_local(iter->bvset); 206 return iter->bvpage; 207 } 208 209 static struct page *z_erofs_bvset_flip(struct z_erofs_bvec_iter *iter) 210 { 211 unsigned long base = (unsigned long)((struct z_erofs_bvset *)0)->bvec; 212 /* have to access nextpage in advance, otherwise it will be unmapped */ 213 struct page *nextpage = iter->bvset->nextpage; 214 struct page *oldpage; 215 216 DBG_BUGON(!nextpage); 217 oldpage = z_erofs_bvec_iter_end(iter); 218 iter->bvpage = nextpage; 219 iter->bvset = kmap_local_page(nextpage); 220 iter->nr = (PAGE_SIZE - base) / sizeof(struct z_erofs_bvec); 221 iter->cur = 0; 222 return oldpage; 223 } 224 225 static void z_erofs_bvec_iter_begin(struct z_erofs_bvec_iter *iter, 226 struct z_erofs_bvset_inline *bvset, 227 unsigned int bootstrap_nr, 228 unsigned int cur) 229 { 230 *iter = (struct z_erofs_bvec_iter) { 231 .nr = bootstrap_nr, 232 .bvset = (struct z_erofs_bvset *)bvset, 233 }; 234 235 while (cur > iter->nr) { 236 cur -= iter->nr; 237 z_erofs_bvset_flip(iter); 238 } 239 iter->cur = cur; 240 } 241 242 static int z_erofs_bvec_enqueue(struct z_erofs_bvec_iter *iter, 243 struct z_erofs_bvec *bvec, 244 struct page **candidate_bvpage) 245 { 246 if (iter->cur == iter->nr) { 247 if (!*candidate_bvpage) 248 return -EAGAIN; 249 250 DBG_BUGON(iter->bvset->nextpage); 251 iter->bvset->nextpage = *candidate_bvpage; 252 z_erofs_bvset_flip(iter); 253 254 iter->bvset->nextpage = NULL; 255 *candidate_bvpage = NULL; 256 } 257 iter->bvset->bvec[iter->cur++] = *bvec; 258 return 0; 259 } 260 261 static void z_erofs_bvec_dequeue(struct z_erofs_bvec_iter *iter, 262 struct z_erofs_bvec *bvec, 263 struct page **old_bvpage) 264 { 265 if (iter->cur == iter->nr) 266 *old_bvpage = z_erofs_bvset_flip(iter); 267 else 268 *old_bvpage = NULL; 269 *bvec = iter->bvset->bvec[iter->cur++]; 270 } 271 272 static void z_erofs_destroy_pcluster_pool(void) 273 { 274 int i; 275 276 for (i = 0; i < ARRAY_SIZE(pcluster_pool); ++i) { 277 if (!pcluster_pool[i].slab) 278 continue; 279 kmem_cache_destroy(pcluster_pool[i].slab); 280 pcluster_pool[i].slab = NULL; 281 } 282 } 283 284 static int z_erofs_create_pcluster_pool(void) 285 { 286 struct z_erofs_pcluster_slab *pcs; 287 struct z_erofs_pcluster *a; 288 unsigned int size; 289 290 for (pcs = pcluster_pool; 291 pcs < pcluster_pool + ARRAY_SIZE(pcluster_pool); ++pcs) { 292 size = struct_size(a, compressed_bvecs, pcs->maxpages); 293 294 sprintf(pcs->name, "erofs_pcluster-%u", pcs->maxpages); 295 pcs->slab = kmem_cache_create(pcs->name, size, 0, 296 SLAB_RECLAIM_ACCOUNT, NULL); 297 if (pcs->slab) 298 continue; 299 300 z_erofs_destroy_pcluster_pool(); 301 return -ENOMEM; 302 } 303 return 0; 304 } 305 306 static struct z_erofs_pcluster *z_erofs_alloc_pcluster(unsigned int nrpages) 307 { 308 int i; 309 310 for (i = 0; i < ARRAY_SIZE(pcluster_pool); ++i) { 311 struct z_erofs_pcluster_slab *pcs = pcluster_pool + i; 312 struct z_erofs_pcluster *pcl; 313 314 if (nrpages > pcs->maxpages) 315 continue; 316 317 pcl = kmem_cache_zalloc(pcs->slab, GFP_NOFS); 318 if (!pcl) 319 return ERR_PTR(-ENOMEM); 320 pcl->pclusterpages = nrpages; 321 return pcl; 322 } 323 return ERR_PTR(-EINVAL); 324 } 325 326 static void z_erofs_free_pcluster(struct z_erofs_pcluster *pcl) 327 { 328 unsigned int pclusterpages = z_erofs_pclusterpages(pcl); 329 int i; 330 331 for (i = 0; i < ARRAY_SIZE(pcluster_pool); ++i) { 332 struct z_erofs_pcluster_slab *pcs = pcluster_pool + i; 333 334 if (pclusterpages > pcs->maxpages) 335 continue; 336 337 kmem_cache_free(pcs->slab, pcl); 338 return; 339 } 340 DBG_BUGON(1); 341 } 342 343 static struct workqueue_struct *z_erofs_workqueue __read_mostly; 344 345 #ifdef CONFIG_EROFS_FS_PCPU_KTHREAD 346 static struct kthread_worker __rcu **z_erofs_pcpu_workers; 347 348 static void erofs_destroy_percpu_workers(void) 349 { 350 struct kthread_worker *worker; 351 unsigned int cpu; 352 353 for_each_possible_cpu(cpu) { 354 worker = rcu_dereference_protected( 355 z_erofs_pcpu_workers[cpu], 1); 356 rcu_assign_pointer(z_erofs_pcpu_workers[cpu], NULL); 357 if (worker) 358 kthread_destroy_worker(worker); 359 } 360 kfree(z_erofs_pcpu_workers); 361 } 362 363 static struct kthread_worker *erofs_init_percpu_worker(int cpu) 364 { 365 struct kthread_worker *worker = 366 kthread_create_worker_on_cpu(cpu, 0, "erofs_worker/%u", cpu); 367 368 if (IS_ERR(worker)) 369 return worker; 370 if (IS_ENABLED(CONFIG_EROFS_FS_PCPU_KTHREAD_HIPRI)) 371 sched_set_fifo_low(worker->task); 372 return worker; 373 } 374 375 static int erofs_init_percpu_workers(void) 376 { 377 struct kthread_worker *worker; 378 unsigned int cpu; 379 380 z_erofs_pcpu_workers = kcalloc(num_possible_cpus(), 381 sizeof(struct kthread_worker *), GFP_ATOMIC); 382 if (!z_erofs_pcpu_workers) 383 return -ENOMEM; 384 385 for_each_online_cpu(cpu) { /* could miss cpu{off,on}line? */ 386 worker = erofs_init_percpu_worker(cpu); 387 if (!IS_ERR(worker)) 388 rcu_assign_pointer(z_erofs_pcpu_workers[cpu], worker); 389 } 390 return 0; 391 } 392 #else 393 static inline void erofs_destroy_percpu_workers(void) {} 394 static inline int erofs_init_percpu_workers(void) { return 0; } 395 #endif 396 397 #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_EROFS_FS_PCPU_KTHREAD) 398 static DEFINE_SPINLOCK(z_erofs_pcpu_worker_lock); 399 static enum cpuhp_state erofs_cpuhp_state; 400 401 static int erofs_cpu_online(unsigned int cpu) 402 { 403 struct kthread_worker *worker, *old; 404 405 worker = erofs_init_percpu_worker(cpu); 406 if (IS_ERR(worker)) 407 return PTR_ERR(worker); 408 409 spin_lock(&z_erofs_pcpu_worker_lock); 410 old = rcu_dereference_protected(z_erofs_pcpu_workers[cpu], 411 lockdep_is_held(&z_erofs_pcpu_worker_lock)); 412 if (!old) 413 rcu_assign_pointer(z_erofs_pcpu_workers[cpu], worker); 414 spin_unlock(&z_erofs_pcpu_worker_lock); 415 if (old) 416 kthread_destroy_worker(worker); 417 return 0; 418 } 419 420 static int erofs_cpu_offline(unsigned int cpu) 421 { 422 struct kthread_worker *worker; 423 424 spin_lock(&z_erofs_pcpu_worker_lock); 425 worker = rcu_dereference_protected(z_erofs_pcpu_workers[cpu], 426 lockdep_is_held(&z_erofs_pcpu_worker_lock)); 427 rcu_assign_pointer(z_erofs_pcpu_workers[cpu], NULL); 428 spin_unlock(&z_erofs_pcpu_worker_lock); 429 430 synchronize_rcu(); 431 if (worker) 432 kthread_destroy_worker(worker); 433 return 0; 434 } 435 436 static int erofs_cpu_hotplug_init(void) 437 { 438 int state; 439 440 state = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, 441 "fs/erofs:online", erofs_cpu_online, erofs_cpu_offline); 442 if (state < 0) 443 return state; 444 445 erofs_cpuhp_state = state; 446 return 0; 447 } 448 449 static void erofs_cpu_hotplug_destroy(void) 450 { 451 if (erofs_cpuhp_state) 452 cpuhp_remove_state_nocalls(erofs_cpuhp_state); 453 } 454 #else /* !CONFIG_HOTPLUG_CPU || !CONFIG_EROFS_FS_PCPU_KTHREAD */ 455 static inline int erofs_cpu_hotplug_init(void) { return 0; } 456 static inline void erofs_cpu_hotplug_destroy(void) {} 457 #endif 458 459 void z_erofs_exit_zip_subsystem(void) 460 { 461 erofs_cpu_hotplug_destroy(); 462 erofs_destroy_percpu_workers(); 463 destroy_workqueue(z_erofs_workqueue); 464 z_erofs_destroy_pcluster_pool(); 465 } 466 467 int __init z_erofs_init_zip_subsystem(void) 468 { 469 int err = z_erofs_create_pcluster_pool(); 470 471 if (err) 472 goto out_error_pcluster_pool; 473 474 z_erofs_workqueue = alloc_workqueue("erofs_worker", 475 WQ_UNBOUND | WQ_HIGHPRI, num_possible_cpus()); 476 if (!z_erofs_workqueue) { 477 err = -ENOMEM; 478 goto out_error_workqueue_init; 479 } 480 481 err = erofs_init_percpu_workers(); 482 if (err) 483 goto out_error_pcpu_worker; 484 485 err = erofs_cpu_hotplug_init(); 486 if (err < 0) 487 goto out_error_cpuhp_init; 488 return err; 489 490 out_error_cpuhp_init: 491 erofs_destroy_percpu_workers(); 492 out_error_pcpu_worker: 493 destroy_workqueue(z_erofs_workqueue); 494 out_error_workqueue_init: 495 z_erofs_destroy_pcluster_pool(); 496 out_error_pcluster_pool: 497 return err; 498 } 499 500 enum z_erofs_pclustermode { 501 Z_EROFS_PCLUSTER_INFLIGHT, 502 /* 503 * The current pclusters was the tail of an exist chain, in addition 504 * that the previous processed chained pclusters are all decided to 505 * be hooked up to it. 506 * A new chain will be created for the remaining pclusters which are 507 * not processed yet, so different from Z_EROFS_PCLUSTER_FOLLOWED, 508 * the next pcluster cannot reuse the whole page safely for inplace I/O 509 * in the following scenario: 510 * ________________________________________________________________ 511 * | tail (partial) page | head (partial) page | 512 * | (belongs to the next pcl) | (belongs to the current pcl) | 513 * |_______PCLUSTER_FOLLOWED______|________PCLUSTER_HOOKED__________| 514 */ 515 Z_EROFS_PCLUSTER_HOOKED, 516 /* 517 * a weak form of Z_EROFS_PCLUSTER_FOLLOWED, the difference is that it 518 * could be dispatched into bypass queue later due to uptodated managed 519 * pages. All related online pages cannot be reused for inplace I/O (or 520 * bvpage) since it can be directly decoded without I/O submission. 521 */ 522 Z_EROFS_PCLUSTER_FOLLOWED_NOINPLACE, 523 /* 524 * The current collection has been linked with the owned chain, and 525 * could also be linked with the remaining collections, which means 526 * if the processing page is the tail page of the collection, thus 527 * the current collection can safely use the whole page (since 528 * the previous collection is under control) for in-place I/O, as 529 * illustrated below: 530 * ________________________________________________________________ 531 * | tail (partial) page | head (partial) page | 532 * | (of the current cl) | (of the previous collection) | 533 * | PCLUSTER_FOLLOWED or | | 534 * |_____PCLUSTER_HOOKED__|___________PCLUSTER_FOLLOWED____________| 535 * 536 * [ (*) the above page can be used as inplace I/O. ] 537 */ 538 Z_EROFS_PCLUSTER_FOLLOWED, 539 }; 540 541 struct z_erofs_decompress_frontend { 542 struct inode *const inode; 543 struct erofs_map_blocks map; 544 struct z_erofs_bvec_iter biter; 545 546 struct page *candidate_bvpage; 547 struct z_erofs_pcluster *pcl, *tailpcl; 548 z_erofs_next_pcluster_t owned_head; 549 enum z_erofs_pclustermode mode; 550 551 /* used for applying cache strategy on the fly */ 552 bool backmost; 553 erofs_off_t headoffset; 554 555 /* a pointer used to pick up inplace I/O pages */ 556 unsigned int icur; 557 }; 558 559 #define DECOMPRESS_FRONTEND_INIT(__i) { \ 560 .inode = __i, .owned_head = Z_EROFS_PCLUSTER_TAIL, \ 561 .mode = Z_EROFS_PCLUSTER_FOLLOWED, .backmost = true } 562 563 static bool z_erofs_should_alloc_cache(struct z_erofs_decompress_frontend *fe) 564 { 565 unsigned int cachestrategy = EROFS_I_SB(fe->inode)->opt.cache_strategy; 566 567 if (cachestrategy <= EROFS_ZIP_CACHE_DISABLED) 568 return false; 569 570 if (fe->backmost) 571 return true; 572 573 if (cachestrategy >= EROFS_ZIP_CACHE_READAROUND && 574 fe->map.m_la < fe->headoffset) 575 return true; 576 577 return false; 578 } 579 580 static void z_erofs_bind_cache(struct z_erofs_decompress_frontend *fe, 581 struct page **pagepool) 582 { 583 struct address_space *mc = MNGD_MAPPING(EROFS_I_SB(fe->inode)); 584 struct z_erofs_pcluster *pcl = fe->pcl; 585 bool shouldalloc = z_erofs_should_alloc_cache(fe); 586 bool standalone = true; 587 /* 588 * optimistic allocation without direct reclaim since inplace I/O 589 * can be used if low memory otherwise. 590 */ 591 gfp_t gfp = (mapping_gfp_mask(mc) & ~__GFP_DIRECT_RECLAIM) | 592 __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN; 593 unsigned int i; 594 595 if (fe->mode < Z_EROFS_PCLUSTER_FOLLOWED) 596 return; 597 598 for (i = 0; i < pcl->pclusterpages; ++i) { 599 struct page *page; 600 void *t; /* mark pages just found for debugging */ 601 struct page *newpage = NULL; 602 603 /* the compressed page was loaded before */ 604 if (READ_ONCE(pcl->compressed_bvecs[i].page)) 605 continue; 606 607 page = find_get_page(mc, pcl->obj.index + i); 608 609 if (page) { 610 t = (void *)((unsigned long)page | 1); 611 } else { 612 /* I/O is needed, no possible to decompress directly */ 613 standalone = false; 614 if (!shouldalloc) 615 continue; 616 617 /* 618 * try to use cached I/O if page allocation 619 * succeeds or fallback to in-place I/O instead 620 * to avoid any direct reclaim. 621 */ 622 newpage = erofs_allocpage(pagepool, gfp); 623 if (!newpage) 624 continue; 625 set_page_private(newpage, Z_EROFS_PREALLOCATED_PAGE); 626 t = (void *)((unsigned long)newpage | 1); 627 } 628 629 if (!cmpxchg_relaxed(&pcl->compressed_bvecs[i].page, NULL, t)) 630 continue; 631 632 if (page) 633 put_page(page); 634 else if (newpage) 635 erofs_pagepool_add(pagepool, newpage); 636 } 637 638 /* 639 * don't do inplace I/O if all compressed pages are available in 640 * managed cache since it can be moved to the bypass queue instead. 641 */ 642 if (standalone) 643 fe->mode = Z_EROFS_PCLUSTER_FOLLOWED_NOINPLACE; 644 } 645 646 /* called by erofs_shrinker to get rid of all compressed_pages */ 647 int erofs_try_to_free_all_cached_pages(struct erofs_sb_info *sbi, 648 struct erofs_workgroup *grp) 649 { 650 struct z_erofs_pcluster *const pcl = 651 container_of(grp, struct z_erofs_pcluster, obj); 652 int i; 653 654 DBG_BUGON(z_erofs_is_inline_pcluster(pcl)); 655 /* 656 * refcount of workgroup is now freezed as 1, 657 * therefore no need to worry about available decompression users. 658 */ 659 for (i = 0; i < pcl->pclusterpages; ++i) { 660 struct page *page = pcl->compressed_bvecs[i].page; 661 662 if (!page) 663 continue; 664 665 /* block other users from reclaiming or migrating the page */ 666 if (!trylock_page(page)) 667 return -EBUSY; 668 669 if (!erofs_page_is_managed(sbi, page)) 670 continue; 671 672 /* barrier is implied in the following 'unlock_page' */ 673 WRITE_ONCE(pcl->compressed_bvecs[i].page, NULL); 674 detach_page_private(page); 675 unlock_page(page); 676 } 677 return 0; 678 } 679 680 int erofs_try_to_free_cached_page(struct page *page) 681 { 682 struct z_erofs_pcluster *const pcl = (void *)page_private(page); 683 int ret, i; 684 685 if (!erofs_workgroup_try_to_freeze(&pcl->obj, 1)) 686 return 0; 687 688 ret = 0; 689 DBG_BUGON(z_erofs_is_inline_pcluster(pcl)); 690 for (i = 0; i < pcl->pclusterpages; ++i) { 691 if (pcl->compressed_bvecs[i].page == page) { 692 WRITE_ONCE(pcl->compressed_bvecs[i].page, NULL); 693 ret = 1; 694 break; 695 } 696 } 697 erofs_workgroup_unfreeze(&pcl->obj, 1); 698 if (ret) 699 detach_page_private(page); 700 return ret; 701 } 702 703 static bool z_erofs_try_inplace_io(struct z_erofs_decompress_frontend *fe, 704 struct z_erofs_bvec *bvec) 705 { 706 struct z_erofs_pcluster *const pcl = fe->pcl; 707 708 while (fe->icur > 0) { 709 if (!cmpxchg(&pcl->compressed_bvecs[--fe->icur].page, 710 NULL, bvec->page)) { 711 pcl->compressed_bvecs[fe->icur] = *bvec; 712 return true; 713 } 714 } 715 return false; 716 } 717 718 /* callers must be with pcluster lock held */ 719 static int z_erofs_attach_page(struct z_erofs_decompress_frontend *fe, 720 struct z_erofs_bvec *bvec, bool exclusive) 721 { 722 int ret; 723 724 if (exclusive) { 725 /* give priority for inplaceio to use file pages first */ 726 if (z_erofs_try_inplace_io(fe, bvec)) 727 return 0; 728 /* otherwise, check if it can be used as a bvpage */ 729 if (fe->mode >= Z_EROFS_PCLUSTER_FOLLOWED && 730 !fe->candidate_bvpage) 731 fe->candidate_bvpage = bvec->page; 732 } 733 ret = z_erofs_bvec_enqueue(&fe->biter, bvec, &fe->candidate_bvpage); 734 fe->pcl->vcnt += (ret >= 0); 735 return ret; 736 } 737 738 static void z_erofs_try_to_claim_pcluster(struct z_erofs_decompress_frontend *f) 739 { 740 struct z_erofs_pcluster *pcl = f->pcl; 741 z_erofs_next_pcluster_t *owned_head = &f->owned_head; 742 743 /* type 1, nil pcluster (this pcluster doesn't belong to any chain.) */ 744 if (cmpxchg(&pcl->next, Z_EROFS_PCLUSTER_NIL, 745 *owned_head) == Z_EROFS_PCLUSTER_NIL) { 746 *owned_head = &pcl->next; 747 /* so we can attach this pcluster to our submission chain. */ 748 f->mode = Z_EROFS_PCLUSTER_FOLLOWED; 749 return; 750 } 751 752 /* 753 * type 2, link to the end of an existing open chain, be careful 754 * that its submission is controlled by the original attached chain. 755 */ 756 if (*owned_head != &pcl->next && pcl != f->tailpcl && 757 cmpxchg(&pcl->next, Z_EROFS_PCLUSTER_TAIL, 758 *owned_head) == Z_EROFS_PCLUSTER_TAIL) { 759 *owned_head = Z_EROFS_PCLUSTER_TAIL; 760 f->mode = Z_EROFS_PCLUSTER_HOOKED; 761 f->tailpcl = NULL; 762 return; 763 } 764 /* type 3, it belongs to a chain, but it isn't the end of the chain */ 765 f->mode = Z_EROFS_PCLUSTER_INFLIGHT; 766 } 767 768 static int z_erofs_register_pcluster(struct z_erofs_decompress_frontend *fe) 769 { 770 struct erofs_map_blocks *map = &fe->map; 771 bool ztailpacking = map->m_flags & EROFS_MAP_META; 772 struct z_erofs_pcluster *pcl; 773 struct erofs_workgroup *grp; 774 int err; 775 776 if (!(map->m_flags & EROFS_MAP_ENCODED) || 777 (!ztailpacking && !(map->m_pa >> PAGE_SHIFT))) { 778 DBG_BUGON(1); 779 return -EFSCORRUPTED; 780 } 781 782 /* no available pcluster, let's allocate one */ 783 pcl = z_erofs_alloc_pcluster(ztailpacking ? 1 : 784 map->m_plen >> PAGE_SHIFT); 785 if (IS_ERR(pcl)) 786 return PTR_ERR(pcl); 787 788 atomic_set(&pcl->obj.refcount, 1); 789 pcl->algorithmformat = map->m_algorithmformat; 790 pcl->length = 0; 791 pcl->partial = true; 792 793 /* new pclusters should be claimed as type 1, primary and followed */ 794 pcl->next = fe->owned_head; 795 pcl->pageofs_out = map->m_la & ~PAGE_MASK; 796 fe->mode = Z_EROFS_PCLUSTER_FOLLOWED; 797 798 /* 799 * lock all primary followed works before visible to others 800 * and mutex_trylock *never* fails for a new pcluster. 801 */ 802 mutex_init(&pcl->lock); 803 DBG_BUGON(!mutex_trylock(&pcl->lock)); 804 805 if (ztailpacking) { 806 pcl->obj.index = 0; /* which indicates ztailpacking */ 807 pcl->pageofs_in = erofs_blkoff(fe->inode->i_sb, map->m_pa); 808 pcl->tailpacking_size = map->m_plen; 809 } else { 810 pcl->obj.index = map->m_pa >> PAGE_SHIFT; 811 812 grp = erofs_insert_workgroup(fe->inode->i_sb, &pcl->obj); 813 if (IS_ERR(grp)) { 814 err = PTR_ERR(grp); 815 goto err_out; 816 } 817 818 if (grp != &pcl->obj) { 819 fe->pcl = container_of(grp, 820 struct z_erofs_pcluster, obj); 821 err = -EEXIST; 822 goto err_out; 823 } 824 } 825 /* used to check tail merging loop due to corrupted images */ 826 if (fe->owned_head == Z_EROFS_PCLUSTER_TAIL) 827 fe->tailpcl = pcl; 828 fe->owned_head = &pcl->next; 829 fe->pcl = pcl; 830 return 0; 831 832 err_out: 833 mutex_unlock(&pcl->lock); 834 z_erofs_free_pcluster(pcl); 835 return err; 836 } 837 838 static int z_erofs_collector_begin(struct z_erofs_decompress_frontend *fe) 839 { 840 struct erofs_map_blocks *map = &fe->map; 841 struct erofs_workgroup *grp = NULL; 842 int ret; 843 844 DBG_BUGON(fe->pcl); 845 846 /* must be Z_EROFS_PCLUSTER_TAIL or pointed to previous pcluster */ 847 DBG_BUGON(fe->owned_head == Z_EROFS_PCLUSTER_NIL); 848 DBG_BUGON(fe->owned_head == Z_EROFS_PCLUSTER_TAIL_CLOSED); 849 850 if (!(map->m_flags & EROFS_MAP_META)) { 851 grp = erofs_find_workgroup(fe->inode->i_sb, 852 map->m_pa >> PAGE_SHIFT); 853 } else if ((map->m_pa & ~PAGE_MASK) + map->m_plen > PAGE_SIZE) { 854 DBG_BUGON(1); 855 return -EFSCORRUPTED; 856 } 857 858 if (grp) { 859 fe->pcl = container_of(grp, struct z_erofs_pcluster, obj); 860 ret = -EEXIST; 861 } else { 862 ret = z_erofs_register_pcluster(fe); 863 } 864 865 if (ret == -EEXIST) { 866 mutex_lock(&fe->pcl->lock); 867 /* used to check tail merging loop due to corrupted images */ 868 if (fe->owned_head == Z_EROFS_PCLUSTER_TAIL) 869 fe->tailpcl = fe->pcl; 870 871 z_erofs_try_to_claim_pcluster(fe); 872 } else if (ret) { 873 return ret; 874 } 875 z_erofs_bvec_iter_begin(&fe->biter, &fe->pcl->bvset, 876 Z_EROFS_INLINE_BVECS, fe->pcl->vcnt); 877 /* since file-backed online pages are traversed in reverse order */ 878 fe->icur = z_erofs_pclusterpages(fe->pcl); 879 return 0; 880 } 881 882 /* 883 * keep in mind that no referenced pclusters will be freed 884 * only after a RCU grace period. 885 */ 886 static void z_erofs_rcu_callback(struct rcu_head *head) 887 { 888 z_erofs_free_pcluster(container_of(head, 889 struct z_erofs_pcluster, rcu)); 890 } 891 892 void erofs_workgroup_free_rcu(struct erofs_workgroup *grp) 893 { 894 struct z_erofs_pcluster *const pcl = 895 container_of(grp, struct z_erofs_pcluster, obj); 896 897 call_rcu(&pcl->rcu, z_erofs_rcu_callback); 898 } 899 900 static bool z_erofs_collector_end(struct z_erofs_decompress_frontend *fe) 901 { 902 struct z_erofs_pcluster *pcl = fe->pcl; 903 904 if (!pcl) 905 return false; 906 907 z_erofs_bvec_iter_end(&fe->biter); 908 mutex_unlock(&pcl->lock); 909 910 if (fe->candidate_bvpage) { 911 DBG_BUGON(z_erofs_is_shortlived_page(fe->candidate_bvpage)); 912 fe->candidate_bvpage = NULL; 913 } 914 915 /* 916 * if all pending pages are added, don't hold its reference 917 * any longer if the pcluster isn't hosted by ourselves. 918 */ 919 if (fe->mode < Z_EROFS_PCLUSTER_FOLLOWED_NOINPLACE) 920 erofs_workgroup_put(&pcl->obj); 921 922 fe->pcl = NULL; 923 return true; 924 } 925 926 static int z_erofs_read_fragment(struct inode *inode, erofs_off_t pos, 927 struct page *page, unsigned int pageofs, 928 unsigned int len) 929 { 930 struct super_block *sb = inode->i_sb; 931 struct inode *packed_inode = EROFS_I_SB(inode)->packed_inode; 932 struct erofs_buf buf = __EROFS_BUF_INITIALIZER; 933 u8 *src, *dst; 934 unsigned int i, cnt; 935 936 if (!packed_inode) 937 return -EFSCORRUPTED; 938 939 buf.inode = packed_inode; 940 pos += EROFS_I(inode)->z_fragmentoff; 941 for (i = 0; i < len; i += cnt) { 942 cnt = min_t(unsigned int, len - i, 943 sb->s_blocksize - erofs_blkoff(sb, pos)); 944 src = erofs_bread(&buf, erofs_blknr(sb, pos), EROFS_KMAP); 945 if (IS_ERR(src)) { 946 erofs_put_metabuf(&buf); 947 return PTR_ERR(src); 948 } 949 950 dst = kmap_local_page(page); 951 memcpy(dst + pageofs + i, src + erofs_blkoff(sb, pos), cnt); 952 kunmap_local(dst); 953 pos += cnt; 954 } 955 erofs_put_metabuf(&buf); 956 return 0; 957 } 958 959 static int z_erofs_do_read_page(struct z_erofs_decompress_frontend *fe, 960 struct page *page, struct page **pagepool) 961 { 962 struct inode *const inode = fe->inode; 963 struct erofs_map_blocks *const map = &fe->map; 964 const loff_t offset = page_offset(page); 965 bool tight = true, exclusive; 966 unsigned int cur, end, spiltted; 967 int err = 0; 968 969 /* register locked file pages as online pages in pack */ 970 z_erofs_onlinepage_init(page); 971 972 spiltted = 0; 973 end = PAGE_SIZE; 974 repeat: 975 cur = end - 1; 976 977 if (offset + cur < map->m_la || 978 offset + cur >= map->m_la + map->m_llen) { 979 if (z_erofs_collector_end(fe)) 980 fe->backmost = false; 981 map->m_la = offset + cur; 982 map->m_llen = 0; 983 err = z_erofs_map_blocks_iter(inode, map, 0); 984 if (err) 985 goto out; 986 } else { 987 if (fe->pcl) 988 goto hitted; 989 /* didn't get a valid pcluster previously (very rare) */ 990 } 991 992 if (!(map->m_flags & EROFS_MAP_MAPPED) || 993 map->m_flags & EROFS_MAP_FRAGMENT) 994 goto hitted; 995 996 err = z_erofs_collector_begin(fe); 997 if (err) 998 goto out; 999 1000 if (z_erofs_is_inline_pcluster(fe->pcl)) { 1001 void *mp; 1002 1003 mp = erofs_read_metabuf(&fe->map.buf, inode->i_sb, 1004 erofs_blknr(inode->i_sb, map->m_pa), 1005 EROFS_NO_KMAP); 1006 if (IS_ERR(mp)) { 1007 err = PTR_ERR(mp); 1008 erofs_err(inode->i_sb, 1009 "failed to get inline page, err %d", err); 1010 goto out; 1011 } 1012 get_page(fe->map.buf.page); 1013 WRITE_ONCE(fe->pcl->compressed_bvecs[0].page, 1014 fe->map.buf.page); 1015 fe->mode = Z_EROFS_PCLUSTER_FOLLOWED_NOINPLACE; 1016 } else { 1017 /* bind cache first when cached decompression is preferred */ 1018 z_erofs_bind_cache(fe, pagepool); 1019 } 1020 hitted: 1021 /* 1022 * Ensure the current partial page belongs to this submit chain rather 1023 * than other concurrent submit chains or the noio(bypass) chain since 1024 * those chains are handled asynchronously thus the page cannot be used 1025 * for inplace I/O or bvpage (should be processed in a strict order.) 1026 */ 1027 tight &= (fe->mode >= Z_EROFS_PCLUSTER_HOOKED && 1028 fe->mode != Z_EROFS_PCLUSTER_FOLLOWED_NOINPLACE); 1029 1030 cur = end - min_t(unsigned int, offset + end - map->m_la, end); 1031 if (!(map->m_flags & EROFS_MAP_MAPPED)) { 1032 zero_user_segment(page, cur, end); 1033 goto next_part; 1034 } 1035 if (map->m_flags & EROFS_MAP_FRAGMENT) { 1036 unsigned int pageofs, skip, len; 1037 1038 if (offset > map->m_la) { 1039 pageofs = 0; 1040 skip = offset - map->m_la; 1041 } else { 1042 pageofs = map->m_la & ~PAGE_MASK; 1043 skip = 0; 1044 } 1045 len = min_t(unsigned int, map->m_llen - skip, end - cur); 1046 err = z_erofs_read_fragment(inode, skip, page, pageofs, len); 1047 if (err) 1048 goto out; 1049 ++spiltted; 1050 tight = false; 1051 goto next_part; 1052 } 1053 1054 exclusive = (!cur && (!spiltted || tight)); 1055 if (cur) 1056 tight &= (fe->mode >= Z_EROFS_PCLUSTER_FOLLOWED); 1057 1058 retry: 1059 err = z_erofs_attach_page(fe, &((struct z_erofs_bvec) { 1060 .page = page, 1061 .offset = offset - map->m_la, 1062 .end = end, 1063 }), exclusive); 1064 /* should allocate an additional short-lived page for bvset */ 1065 if (err == -EAGAIN && !fe->candidate_bvpage) { 1066 fe->candidate_bvpage = alloc_page(GFP_NOFS | __GFP_NOFAIL); 1067 set_page_private(fe->candidate_bvpage, 1068 Z_EROFS_SHORTLIVED_PAGE); 1069 goto retry; 1070 } 1071 1072 if (err) { 1073 DBG_BUGON(err == -EAGAIN && fe->candidate_bvpage); 1074 goto out; 1075 } 1076 1077 z_erofs_onlinepage_split(page); 1078 /* bump up the number of spiltted parts of a page */ 1079 ++spiltted; 1080 if (fe->pcl->pageofs_out != (map->m_la & ~PAGE_MASK)) 1081 fe->pcl->multibases = true; 1082 if (fe->pcl->length < offset + end - map->m_la) { 1083 fe->pcl->length = offset + end - map->m_la; 1084 fe->pcl->pageofs_out = map->m_la & ~PAGE_MASK; 1085 } 1086 if ((map->m_flags & EROFS_MAP_FULL_MAPPED) && 1087 !(map->m_flags & EROFS_MAP_PARTIAL_REF) && 1088 fe->pcl->length == map->m_llen) 1089 fe->pcl->partial = false; 1090 next_part: 1091 /* shorten the remaining extent to update progress */ 1092 map->m_llen = offset + cur - map->m_la; 1093 map->m_flags &= ~EROFS_MAP_FULL_MAPPED; 1094 1095 end = cur; 1096 if (end > 0) 1097 goto repeat; 1098 1099 out: 1100 if (err) 1101 z_erofs_page_mark_eio(page); 1102 z_erofs_onlinepage_endio(page); 1103 return err; 1104 } 1105 1106 static bool z_erofs_is_sync_decompress(struct erofs_sb_info *sbi, 1107 unsigned int readahead_pages) 1108 { 1109 /* auto: enable for read_folio, disable for readahead */ 1110 if ((sbi->opt.sync_decompress == EROFS_SYNC_DECOMPRESS_AUTO) && 1111 !readahead_pages) 1112 return true; 1113 1114 if ((sbi->opt.sync_decompress == EROFS_SYNC_DECOMPRESS_FORCE_ON) && 1115 (readahead_pages <= sbi->opt.max_sync_decompress_pages)) 1116 return true; 1117 1118 return false; 1119 } 1120 1121 static bool z_erofs_page_is_invalidated(struct page *page) 1122 { 1123 return !page->mapping && !z_erofs_is_shortlived_page(page); 1124 } 1125 1126 struct z_erofs_decompress_backend { 1127 struct page *onstack_pages[Z_EROFS_ONSTACK_PAGES]; 1128 struct super_block *sb; 1129 struct z_erofs_pcluster *pcl; 1130 1131 /* pages with the longest decompressed length for deduplication */ 1132 struct page **decompressed_pages; 1133 /* pages to keep the compressed data */ 1134 struct page **compressed_pages; 1135 1136 struct list_head decompressed_secondary_bvecs; 1137 struct page **pagepool; 1138 unsigned int onstack_used, nr_pages; 1139 }; 1140 1141 struct z_erofs_bvec_item { 1142 struct z_erofs_bvec bvec; 1143 struct list_head list; 1144 }; 1145 1146 static void z_erofs_do_decompressed_bvec(struct z_erofs_decompress_backend *be, 1147 struct z_erofs_bvec *bvec) 1148 { 1149 struct z_erofs_bvec_item *item; 1150 1151 if (!((bvec->offset + be->pcl->pageofs_out) & ~PAGE_MASK)) { 1152 unsigned int pgnr; 1153 1154 pgnr = (bvec->offset + be->pcl->pageofs_out) >> PAGE_SHIFT; 1155 DBG_BUGON(pgnr >= be->nr_pages); 1156 if (!be->decompressed_pages[pgnr]) { 1157 be->decompressed_pages[pgnr] = bvec->page; 1158 return; 1159 } 1160 } 1161 1162 /* (cold path) one pcluster is requested multiple times */ 1163 item = kmalloc(sizeof(*item), GFP_KERNEL | __GFP_NOFAIL); 1164 item->bvec = *bvec; 1165 list_add(&item->list, &be->decompressed_secondary_bvecs); 1166 } 1167 1168 static void z_erofs_fill_other_copies(struct z_erofs_decompress_backend *be, 1169 int err) 1170 { 1171 unsigned int off0 = be->pcl->pageofs_out; 1172 struct list_head *p, *n; 1173 1174 list_for_each_safe(p, n, &be->decompressed_secondary_bvecs) { 1175 struct z_erofs_bvec_item *bvi; 1176 unsigned int end, cur; 1177 void *dst, *src; 1178 1179 bvi = container_of(p, struct z_erofs_bvec_item, list); 1180 cur = bvi->bvec.offset < 0 ? -bvi->bvec.offset : 0; 1181 end = min_t(unsigned int, be->pcl->length - bvi->bvec.offset, 1182 bvi->bvec.end); 1183 dst = kmap_local_page(bvi->bvec.page); 1184 while (cur < end) { 1185 unsigned int pgnr, scur, len; 1186 1187 pgnr = (bvi->bvec.offset + cur + off0) >> PAGE_SHIFT; 1188 DBG_BUGON(pgnr >= be->nr_pages); 1189 1190 scur = bvi->bvec.offset + cur - 1191 ((pgnr << PAGE_SHIFT) - off0); 1192 len = min_t(unsigned int, end - cur, PAGE_SIZE - scur); 1193 if (!be->decompressed_pages[pgnr]) { 1194 err = -EFSCORRUPTED; 1195 cur += len; 1196 continue; 1197 } 1198 src = kmap_local_page(be->decompressed_pages[pgnr]); 1199 memcpy(dst + cur, src + scur, len); 1200 kunmap_local(src); 1201 cur += len; 1202 } 1203 kunmap_local(dst); 1204 if (err) 1205 z_erofs_page_mark_eio(bvi->bvec.page); 1206 z_erofs_onlinepage_endio(bvi->bvec.page); 1207 list_del(p); 1208 kfree(bvi); 1209 } 1210 } 1211 1212 static void z_erofs_parse_out_bvecs(struct z_erofs_decompress_backend *be) 1213 { 1214 struct z_erofs_pcluster *pcl = be->pcl; 1215 struct z_erofs_bvec_iter biter; 1216 struct page *old_bvpage; 1217 int i; 1218 1219 z_erofs_bvec_iter_begin(&biter, &pcl->bvset, Z_EROFS_INLINE_BVECS, 0); 1220 for (i = 0; i < pcl->vcnt; ++i) { 1221 struct z_erofs_bvec bvec; 1222 1223 z_erofs_bvec_dequeue(&biter, &bvec, &old_bvpage); 1224 1225 if (old_bvpage) 1226 z_erofs_put_shortlivedpage(be->pagepool, old_bvpage); 1227 1228 DBG_BUGON(z_erofs_page_is_invalidated(bvec.page)); 1229 z_erofs_do_decompressed_bvec(be, &bvec); 1230 } 1231 1232 old_bvpage = z_erofs_bvec_iter_end(&biter); 1233 if (old_bvpage) 1234 z_erofs_put_shortlivedpage(be->pagepool, old_bvpage); 1235 } 1236 1237 static int z_erofs_parse_in_bvecs(struct z_erofs_decompress_backend *be, 1238 bool *overlapped) 1239 { 1240 struct z_erofs_pcluster *pcl = be->pcl; 1241 unsigned int pclusterpages = z_erofs_pclusterpages(pcl); 1242 int i, err = 0; 1243 1244 *overlapped = false; 1245 for (i = 0; i < pclusterpages; ++i) { 1246 struct z_erofs_bvec *bvec = &pcl->compressed_bvecs[i]; 1247 struct page *page = bvec->page; 1248 1249 /* compressed pages ought to be present before decompressing */ 1250 if (!page) { 1251 DBG_BUGON(1); 1252 continue; 1253 } 1254 be->compressed_pages[i] = page; 1255 1256 if (z_erofs_is_inline_pcluster(pcl)) { 1257 if (!PageUptodate(page)) 1258 err = -EIO; 1259 continue; 1260 } 1261 1262 DBG_BUGON(z_erofs_page_is_invalidated(page)); 1263 if (!z_erofs_is_shortlived_page(page)) { 1264 if (erofs_page_is_managed(EROFS_SB(be->sb), page)) { 1265 if (!PageUptodate(page)) 1266 err = -EIO; 1267 continue; 1268 } 1269 z_erofs_do_decompressed_bvec(be, bvec); 1270 *overlapped = true; 1271 } 1272 } 1273 1274 if (err) 1275 return err; 1276 return 0; 1277 } 1278 1279 static int z_erofs_decompress_pcluster(struct z_erofs_decompress_backend *be, 1280 int err) 1281 { 1282 struct erofs_sb_info *const sbi = EROFS_SB(be->sb); 1283 struct z_erofs_pcluster *pcl = be->pcl; 1284 unsigned int pclusterpages = z_erofs_pclusterpages(pcl); 1285 const struct z_erofs_decompressor *decompressor = 1286 &erofs_decompressors[pcl->algorithmformat]; 1287 unsigned int i, inputsize; 1288 int err2; 1289 struct page *page; 1290 bool overlapped; 1291 1292 mutex_lock(&pcl->lock); 1293 be->nr_pages = PAGE_ALIGN(pcl->length + pcl->pageofs_out) >> PAGE_SHIFT; 1294 1295 /* allocate (de)compressed page arrays if cannot be kept on stack */ 1296 be->decompressed_pages = NULL; 1297 be->compressed_pages = NULL; 1298 be->onstack_used = 0; 1299 if (be->nr_pages <= Z_EROFS_ONSTACK_PAGES) { 1300 be->decompressed_pages = be->onstack_pages; 1301 be->onstack_used = be->nr_pages; 1302 memset(be->decompressed_pages, 0, 1303 sizeof(struct page *) * be->nr_pages); 1304 } 1305 1306 if (pclusterpages + be->onstack_used <= Z_EROFS_ONSTACK_PAGES) 1307 be->compressed_pages = be->onstack_pages + be->onstack_used; 1308 1309 if (!be->decompressed_pages) 1310 be->decompressed_pages = 1311 kvcalloc(be->nr_pages, sizeof(struct page *), 1312 GFP_KERNEL | __GFP_NOFAIL); 1313 if (!be->compressed_pages) 1314 be->compressed_pages = 1315 kvcalloc(pclusterpages, sizeof(struct page *), 1316 GFP_KERNEL | __GFP_NOFAIL); 1317 1318 z_erofs_parse_out_bvecs(be); 1319 err2 = z_erofs_parse_in_bvecs(be, &overlapped); 1320 if (err2) 1321 err = err2; 1322 if (err) 1323 goto out; 1324 1325 if (z_erofs_is_inline_pcluster(pcl)) 1326 inputsize = pcl->tailpacking_size; 1327 else 1328 inputsize = pclusterpages * PAGE_SIZE; 1329 1330 err = decompressor->decompress(&(struct z_erofs_decompress_req) { 1331 .sb = be->sb, 1332 .in = be->compressed_pages, 1333 .out = be->decompressed_pages, 1334 .pageofs_in = pcl->pageofs_in, 1335 .pageofs_out = pcl->pageofs_out, 1336 .inputsize = inputsize, 1337 .outputsize = pcl->length, 1338 .alg = pcl->algorithmformat, 1339 .inplace_io = overlapped, 1340 .partial_decoding = pcl->partial, 1341 .fillgaps = pcl->multibases, 1342 }, be->pagepool); 1343 1344 out: 1345 /* must handle all compressed pages before actual file pages */ 1346 if (z_erofs_is_inline_pcluster(pcl)) { 1347 page = pcl->compressed_bvecs[0].page; 1348 WRITE_ONCE(pcl->compressed_bvecs[0].page, NULL); 1349 put_page(page); 1350 } else { 1351 for (i = 0; i < pclusterpages; ++i) { 1352 page = pcl->compressed_bvecs[i].page; 1353 1354 if (erofs_page_is_managed(sbi, page)) 1355 continue; 1356 1357 /* recycle all individual short-lived pages */ 1358 (void)z_erofs_put_shortlivedpage(be->pagepool, page); 1359 WRITE_ONCE(pcl->compressed_bvecs[i].page, NULL); 1360 } 1361 } 1362 if (be->compressed_pages < be->onstack_pages || 1363 be->compressed_pages >= be->onstack_pages + Z_EROFS_ONSTACK_PAGES) 1364 kvfree(be->compressed_pages); 1365 z_erofs_fill_other_copies(be, err); 1366 1367 for (i = 0; i < be->nr_pages; ++i) { 1368 page = be->decompressed_pages[i]; 1369 if (!page) 1370 continue; 1371 1372 DBG_BUGON(z_erofs_page_is_invalidated(page)); 1373 1374 /* recycle all individual short-lived pages */ 1375 if (z_erofs_put_shortlivedpage(be->pagepool, page)) 1376 continue; 1377 if (err) 1378 z_erofs_page_mark_eio(page); 1379 z_erofs_onlinepage_endio(page); 1380 } 1381 1382 if (be->decompressed_pages != be->onstack_pages) 1383 kvfree(be->decompressed_pages); 1384 1385 pcl->length = 0; 1386 pcl->partial = true; 1387 pcl->multibases = false; 1388 pcl->bvset.nextpage = NULL; 1389 pcl->vcnt = 0; 1390 1391 /* pcluster lock MUST be taken before the following line */ 1392 WRITE_ONCE(pcl->next, Z_EROFS_PCLUSTER_NIL); 1393 mutex_unlock(&pcl->lock); 1394 return err; 1395 } 1396 1397 static void z_erofs_decompress_queue(const struct z_erofs_decompressqueue *io, 1398 struct page **pagepool) 1399 { 1400 struct z_erofs_decompress_backend be = { 1401 .sb = io->sb, 1402 .pagepool = pagepool, 1403 .decompressed_secondary_bvecs = 1404 LIST_HEAD_INIT(be.decompressed_secondary_bvecs), 1405 }; 1406 z_erofs_next_pcluster_t owned = io->head; 1407 1408 while (owned != Z_EROFS_PCLUSTER_TAIL_CLOSED) { 1409 /* impossible that 'owned' equals Z_EROFS_WORK_TPTR_TAIL */ 1410 DBG_BUGON(owned == Z_EROFS_PCLUSTER_TAIL); 1411 /* impossible that 'owned' equals Z_EROFS_PCLUSTER_NIL */ 1412 DBG_BUGON(owned == Z_EROFS_PCLUSTER_NIL); 1413 1414 be.pcl = container_of(owned, struct z_erofs_pcluster, next); 1415 owned = READ_ONCE(be.pcl->next); 1416 1417 z_erofs_decompress_pcluster(&be, io->eio ? -EIO : 0); 1418 erofs_workgroup_put(&be.pcl->obj); 1419 } 1420 } 1421 1422 static void z_erofs_decompressqueue_work(struct work_struct *work) 1423 { 1424 struct z_erofs_decompressqueue *bgq = 1425 container_of(work, struct z_erofs_decompressqueue, u.work); 1426 struct page *pagepool = NULL; 1427 1428 DBG_BUGON(bgq->head == Z_EROFS_PCLUSTER_TAIL_CLOSED); 1429 z_erofs_decompress_queue(bgq, &pagepool); 1430 erofs_release_pages(&pagepool); 1431 kvfree(bgq); 1432 } 1433 1434 #ifdef CONFIG_EROFS_FS_PCPU_KTHREAD 1435 static void z_erofs_decompressqueue_kthread_work(struct kthread_work *work) 1436 { 1437 z_erofs_decompressqueue_work((struct work_struct *)work); 1438 } 1439 #endif 1440 1441 static void z_erofs_decompress_kickoff(struct z_erofs_decompressqueue *io, 1442 int bios) 1443 { 1444 struct erofs_sb_info *const sbi = EROFS_SB(io->sb); 1445 1446 /* wake up the caller thread for sync decompression */ 1447 if (io->sync) { 1448 if (!atomic_add_return(bios, &io->pending_bios)) 1449 complete(&io->u.done); 1450 return; 1451 } 1452 1453 if (atomic_add_return(bios, &io->pending_bios)) 1454 return; 1455 /* Use (kthread_)work and sync decompression for atomic contexts only */ 1456 if (in_atomic() || irqs_disabled()) { 1457 #ifdef CONFIG_EROFS_FS_PCPU_KTHREAD 1458 struct kthread_worker *worker; 1459 1460 rcu_read_lock(); 1461 worker = rcu_dereference( 1462 z_erofs_pcpu_workers[raw_smp_processor_id()]); 1463 if (!worker) { 1464 INIT_WORK(&io->u.work, z_erofs_decompressqueue_work); 1465 queue_work(z_erofs_workqueue, &io->u.work); 1466 } else { 1467 kthread_queue_work(worker, &io->u.kthread_work); 1468 } 1469 rcu_read_unlock(); 1470 #else 1471 queue_work(z_erofs_workqueue, &io->u.work); 1472 #endif 1473 /* enable sync decompression for readahead */ 1474 if (sbi->opt.sync_decompress == EROFS_SYNC_DECOMPRESS_AUTO) 1475 sbi->opt.sync_decompress = EROFS_SYNC_DECOMPRESS_FORCE_ON; 1476 return; 1477 } 1478 z_erofs_decompressqueue_work(&io->u.work); 1479 } 1480 1481 static struct page *pickup_page_for_submission(struct z_erofs_pcluster *pcl, 1482 unsigned int nr, 1483 struct page **pagepool, 1484 struct address_space *mc) 1485 { 1486 const pgoff_t index = pcl->obj.index; 1487 gfp_t gfp = mapping_gfp_mask(mc); 1488 bool tocache = false; 1489 1490 struct address_space *mapping; 1491 struct page *oldpage, *page; 1492 int justfound; 1493 1494 repeat: 1495 page = READ_ONCE(pcl->compressed_bvecs[nr].page); 1496 oldpage = page; 1497 1498 if (!page) 1499 goto out_allocpage; 1500 1501 justfound = (unsigned long)page & 1UL; 1502 page = (struct page *)((unsigned long)page & ~1UL); 1503 1504 /* 1505 * preallocated cached pages, which is used to avoid direct reclaim 1506 * otherwise, it will go inplace I/O path instead. 1507 */ 1508 if (page->private == Z_EROFS_PREALLOCATED_PAGE) { 1509 WRITE_ONCE(pcl->compressed_bvecs[nr].page, page); 1510 set_page_private(page, 0); 1511 tocache = true; 1512 goto out_tocache; 1513 } 1514 mapping = READ_ONCE(page->mapping); 1515 1516 /* 1517 * file-backed online pages in plcuster are all locked steady, 1518 * therefore it is impossible for `mapping' to be NULL. 1519 */ 1520 if (mapping && mapping != mc) 1521 /* ought to be unmanaged pages */ 1522 goto out; 1523 1524 /* directly return for shortlived page as well */ 1525 if (z_erofs_is_shortlived_page(page)) 1526 goto out; 1527 1528 lock_page(page); 1529 1530 /* only true if page reclaim goes wrong, should never happen */ 1531 DBG_BUGON(justfound && PagePrivate(page)); 1532 1533 /* the page is still in manage cache */ 1534 if (page->mapping == mc) { 1535 WRITE_ONCE(pcl->compressed_bvecs[nr].page, page); 1536 1537 if (!PagePrivate(page)) { 1538 /* 1539 * impossible to be !PagePrivate(page) for 1540 * the current restriction as well if 1541 * the page is already in compressed_bvecs[]. 1542 */ 1543 DBG_BUGON(!justfound); 1544 1545 justfound = 0; 1546 set_page_private(page, (unsigned long)pcl); 1547 SetPagePrivate(page); 1548 } 1549 1550 /* no need to submit io if it is already up-to-date */ 1551 if (PageUptodate(page)) { 1552 unlock_page(page); 1553 page = NULL; 1554 } 1555 goto out; 1556 } 1557 1558 /* 1559 * the managed page has been truncated, it's unsafe to 1560 * reuse this one, let's allocate a new cache-managed page. 1561 */ 1562 DBG_BUGON(page->mapping); 1563 DBG_BUGON(!justfound); 1564 1565 tocache = true; 1566 unlock_page(page); 1567 put_page(page); 1568 out_allocpage: 1569 page = erofs_allocpage(pagepool, gfp | __GFP_NOFAIL); 1570 if (oldpage != cmpxchg(&pcl->compressed_bvecs[nr].page, 1571 oldpage, page)) { 1572 erofs_pagepool_add(pagepool, page); 1573 cond_resched(); 1574 goto repeat; 1575 } 1576 out_tocache: 1577 if (!tocache || add_to_page_cache_lru(page, mc, index + nr, gfp)) { 1578 /* turn into temporary page if fails (1 ref) */ 1579 set_page_private(page, Z_EROFS_SHORTLIVED_PAGE); 1580 goto out; 1581 } 1582 attach_page_private(page, pcl); 1583 /* drop a refcount added by allocpage (then we have 2 refs here) */ 1584 put_page(page); 1585 1586 out: /* the only exit (for tracing and debugging) */ 1587 return page; 1588 } 1589 1590 static struct z_erofs_decompressqueue *jobqueue_init(struct super_block *sb, 1591 struct z_erofs_decompressqueue *fgq, bool *fg) 1592 { 1593 struct z_erofs_decompressqueue *q; 1594 1595 if (fg && !*fg) { 1596 q = kvzalloc(sizeof(*q), GFP_KERNEL | __GFP_NOWARN); 1597 if (!q) { 1598 *fg = true; 1599 goto fg_out; 1600 } 1601 #ifdef CONFIG_EROFS_FS_PCPU_KTHREAD 1602 kthread_init_work(&q->u.kthread_work, 1603 z_erofs_decompressqueue_kthread_work); 1604 #else 1605 INIT_WORK(&q->u.work, z_erofs_decompressqueue_work); 1606 #endif 1607 } else { 1608 fg_out: 1609 q = fgq; 1610 init_completion(&fgq->u.done); 1611 atomic_set(&fgq->pending_bios, 0); 1612 q->eio = false; 1613 q->sync = true; 1614 } 1615 q->sb = sb; 1616 q->head = Z_EROFS_PCLUSTER_TAIL_CLOSED; 1617 return q; 1618 } 1619 1620 /* define decompression jobqueue types */ 1621 enum { 1622 JQ_BYPASS, 1623 JQ_SUBMIT, 1624 NR_JOBQUEUES, 1625 }; 1626 1627 static void move_to_bypass_jobqueue(struct z_erofs_pcluster *pcl, 1628 z_erofs_next_pcluster_t qtail[], 1629 z_erofs_next_pcluster_t owned_head) 1630 { 1631 z_erofs_next_pcluster_t *const submit_qtail = qtail[JQ_SUBMIT]; 1632 z_erofs_next_pcluster_t *const bypass_qtail = qtail[JQ_BYPASS]; 1633 1634 DBG_BUGON(owned_head == Z_EROFS_PCLUSTER_TAIL_CLOSED); 1635 if (owned_head == Z_EROFS_PCLUSTER_TAIL) 1636 owned_head = Z_EROFS_PCLUSTER_TAIL_CLOSED; 1637 1638 WRITE_ONCE(pcl->next, Z_EROFS_PCLUSTER_TAIL_CLOSED); 1639 1640 WRITE_ONCE(*submit_qtail, owned_head); 1641 WRITE_ONCE(*bypass_qtail, &pcl->next); 1642 1643 qtail[JQ_BYPASS] = &pcl->next; 1644 } 1645 1646 static void z_erofs_decompressqueue_endio(struct bio *bio) 1647 { 1648 struct z_erofs_decompressqueue *q = bio->bi_private; 1649 blk_status_t err = bio->bi_status; 1650 struct bio_vec *bvec; 1651 struct bvec_iter_all iter_all; 1652 1653 bio_for_each_segment_all(bvec, bio, iter_all) { 1654 struct page *page = bvec->bv_page; 1655 1656 DBG_BUGON(PageUptodate(page)); 1657 DBG_BUGON(z_erofs_page_is_invalidated(page)); 1658 1659 if (erofs_page_is_managed(EROFS_SB(q->sb), page)) { 1660 if (!err) 1661 SetPageUptodate(page); 1662 unlock_page(page); 1663 } 1664 } 1665 if (err) 1666 q->eio = true; 1667 z_erofs_decompress_kickoff(q, -1); 1668 bio_put(bio); 1669 } 1670 1671 static void z_erofs_submit_queue(struct z_erofs_decompress_frontend *f, 1672 struct page **pagepool, 1673 struct z_erofs_decompressqueue *fgq, 1674 bool *force_fg, bool readahead) 1675 { 1676 struct super_block *sb = f->inode->i_sb; 1677 struct address_space *mc = MNGD_MAPPING(EROFS_SB(sb)); 1678 z_erofs_next_pcluster_t qtail[NR_JOBQUEUES]; 1679 struct z_erofs_decompressqueue *q[NR_JOBQUEUES]; 1680 z_erofs_next_pcluster_t owned_head = f->owned_head; 1681 /* bio is NULL initially, so no need to initialize last_{index,bdev} */ 1682 pgoff_t last_index; 1683 struct block_device *last_bdev; 1684 unsigned int nr_bios = 0; 1685 struct bio *bio = NULL; 1686 unsigned long pflags; 1687 int memstall = 0; 1688 1689 /* 1690 * if managed cache is enabled, bypass jobqueue is needed, 1691 * no need to read from device for all pclusters in this queue. 1692 */ 1693 q[JQ_BYPASS] = jobqueue_init(sb, fgq + JQ_BYPASS, NULL); 1694 q[JQ_SUBMIT] = jobqueue_init(sb, fgq + JQ_SUBMIT, force_fg); 1695 1696 qtail[JQ_BYPASS] = &q[JQ_BYPASS]->head; 1697 qtail[JQ_SUBMIT] = &q[JQ_SUBMIT]->head; 1698 1699 /* by default, all need io submission */ 1700 q[JQ_SUBMIT]->head = owned_head; 1701 1702 do { 1703 struct erofs_map_dev mdev; 1704 struct z_erofs_pcluster *pcl; 1705 pgoff_t cur, end; 1706 unsigned int i = 0; 1707 bool bypass = true; 1708 1709 /* no possible 'owned_head' equals the following */ 1710 DBG_BUGON(owned_head == Z_EROFS_PCLUSTER_TAIL_CLOSED); 1711 DBG_BUGON(owned_head == Z_EROFS_PCLUSTER_NIL); 1712 1713 pcl = container_of(owned_head, struct z_erofs_pcluster, next); 1714 1715 /* close the main owned chain at first */ 1716 owned_head = cmpxchg(&pcl->next, Z_EROFS_PCLUSTER_TAIL, 1717 Z_EROFS_PCLUSTER_TAIL_CLOSED); 1718 if (z_erofs_is_inline_pcluster(pcl)) { 1719 move_to_bypass_jobqueue(pcl, qtail, owned_head); 1720 continue; 1721 } 1722 1723 /* no device id here, thus it will always succeed */ 1724 mdev = (struct erofs_map_dev) { 1725 .m_pa = erofs_pos(sb, pcl->obj.index), 1726 }; 1727 (void)erofs_map_dev(sb, &mdev); 1728 1729 cur = erofs_blknr(sb, mdev.m_pa); 1730 end = cur + pcl->pclusterpages; 1731 1732 do { 1733 struct page *page; 1734 1735 page = pickup_page_for_submission(pcl, i++, pagepool, 1736 mc); 1737 if (!page) 1738 continue; 1739 1740 if (bio && (cur != last_index + 1 || 1741 last_bdev != mdev.m_bdev)) { 1742 submit_bio_retry: 1743 submit_bio(bio); 1744 if (memstall) { 1745 psi_memstall_leave(&pflags); 1746 memstall = 0; 1747 } 1748 bio = NULL; 1749 } 1750 1751 if (unlikely(PageWorkingset(page)) && !memstall) { 1752 psi_memstall_enter(&pflags); 1753 memstall = 1; 1754 } 1755 1756 if (!bio) { 1757 bio = bio_alloc(mdev.m_bdev, BIO_MAX_VECS, 1758 REQ_OP_READ, GFP_NOIO); 1759 bio->bi_end_io = z_erofs_decompressqueue_endio; 1760 1761 last_bdev = mdev.m_bdev; 1762 bio->bi_iter.bi_sector = (sector_t)cur << 1763 (sb->s_blocksize_bits - 9); 1764 bio->bi_private = q[JQ_SUBMIT]; 1765 if (readahead) 1766 bio->bi_opf |= REQ_RAHEAD; 1767 ++nr_bios; 1768 } 1769 1770 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) 1771 goto submit_bio_retry; 1772 1773 last_index = cur; 1774 bypass = false; 1775 } while (++cur < end); 1776 1777 if (!bypass) 1778 qtail[JQ_SUBMIT] = &pcl->next; 1779 else 1780 move_to_bypass_jobqueue(pcl, qtail, owned_head); 1781 } while (owned_head != Z_EROFS_PCLUSTER_TAIL); 1782 1783 if (bio) { 1784 submit_bio(bio); 1785 if (memstall) 1786 psi_memstall_leave(&pflags); 1787 } 1788 1789 /* 1790 * although background is preferred, no one is pending for submission. 1791 * don't issue decompression but drop it directly instead. 1792 */ 1793 if (!*force_fg && !nr_bios) { 1794 kvfree(q[JQ_SUBMIT]); 1795 return; 1796 } 1797 z_erofs_decompress_kickoff(q[JQ_SUBMIT], nr_bios); 1798 } 1799 1800 static void z_erofs_runqueue(struct z_erofs_decompress_frontend *f, 1801 struct page **pagepool, bool force_fg, bool ra) 1802 { 1803 struct z_erofs_decompressqueue io[NR_JOBQUEUES]; 1804 1805 if (f->owned_head == Z_EROFS_PCLUSTER_TAIL) 1806 return; 1807 z_erofs_submit_queue(f, pagepool, io, &force_fg, ra); 1808 1809 /* handle bypass queue (no i/o pclusters) immediately */ 1810 z_erofs_decompress_queue(&io[JQ_BYPASS], pagepool); 1811 1812 if (!force_fg) 1813 return; 1814 1815 /* wait until all bios are completed */ 1816 wait_for_completion_io(&io[JQ_SUBMIT].u.done); 1817 1818 /* handle synchronous decompress queue in the caller context */ 1819 z_erofs_decompress_queue(&io[JQ_SUBMIT], pagepool); 1820 } 1821 1822 /* 1823 * Since partial uptodate is still unimplemented for now, we have to use 1824 * approximate readmore strategies as a start. 1825 */ 1826 static void z_erofs_pcluster_readmore(struct z_erofs_decompress_frontend *f, 1827 struct readahead_control *rac, 1828 erofs_off_t end, 1829 struct page **pagepool, 1830 bool backmost) 1831 { 1832 struct inode *inode = f->inode; 1833 struct erofs_map_blocks *map = &f->map; 1834 erofs_off_t cur; 1835 int err; 1836 1837 if (backmost) { 1838 map->m_la = end; 1839 err = z_erofs_map_blocks_iter(inode, map, 1840 EROFS_GET_BLOCKS_READMORE); 1841 if (err) 1842 return; 1843 1844 /* expend ra for the trailing edge if readahead */ 1845 if (rac) { 1846 loff_t newstart = readahead_pos(rac); 1847 1848 cur = round_up(map->m_la + map->m_llen, PAGE_SIZE); 1849 readahead_expand(rac, newstart, cur - newstart); 1850 return; 1851 } 1852 end = round_up(end, PAGE_SIZE); 1853 } else { 1854 end = round_up(map->m_la, PAGE_SIZE); 1855 1856 if (!map->m_llen) 1857 return; 1858 } 1859 1860 cur = map->m_la + map->m_llen - 1; 1861 while (cur >= end) { 1862 pgoff_t index = cur >> PAGE_SHIFT; 1863 struct page *page; 1864 1865 page = erofs_grab_cache_page_nowait(inode->i_mapping, index); 1866 if (page) { 1867 if (PageUptodate(page)) { 1868 unlock_page(page); 1869 } else { 1870 err = z_erofs_do_read_page(f, page, pagepool); 1871 if (err) 1872 erofs_err(inode->i_sb, 1873 "readmore error at page %lu @ nid %llu", 1874 index, EROFS_I(inode)->nid); 1875 } 1876 put_page(page); 1877 } 1878 1879 if (cur < PAGE_SIZE) 1880 break; 1881 cur = (index << PAGE_SHIFT) - 1; 1882 } 1883 } 1884 1885 static int z_erofs_read_folio(struct file *file, struct folio *folio) 1886 { 1887 struct page *page = &folio->page; 1888 struct inode *const inode = page->mapping->host; 1889 struct erofs_sb_info *const sbi = EROFS_I_SB(inode); 1890 struct z_erofs_decompress_frontend f = DECOMPRESS_FRONTEND_INIT(inode); 1891 struct page *pagepool = NULL; 1892 int err; 1893 1894 trace_erofs_readpage(page, false); 1895 f.headoffset = (erofs_off_t)page->index << PAGE_SHIFT; 1896 1897 z_erofs_pcluster_readmore(&f, NULL, f.headoffset + PAGE_SIZE - 1, 1898 &pagepool, true); 1899 err = z_erofs_do_read_page(&f, page, &pagepool); 1900 z_erofs_pcluster_readmore(&f, NULL, 0, &pagepool, false); 1901 1902 (void)z_erofs_collector_end(&f); 1903 1904 /* if some compressed cluster ready, need submit them anyway */ 1905 z_erofs_runqueue(&f, &pagepool, z_erofs_is_sync_decompress(sbi, 0), 1906 false); 1907 1908 if (err) 1909 erofs_err(inode->i_sb, "failed to read, err [%d]", err); 1910 1911 erofs_put_metabuf(&f.map.buf); 1912 erofs_release_pages(&pagepool); 1913 return err; 1914 } 1915 1916 static void z_erofs_readahead(struct readahead_control *rac) 1917 { 1918 struct inode *const inode = rac->mapping->host; 1919 struct erofs_sb_info *const sbi = EROFS_I_SB(inode); 1920 struct z_erofs_decompress_frontend f = DECOMPRESS_FRONTEND_INIT(inode); 1921 struct page *pagepool = NULL, *head = NULL, *page; 1922 unsigned int nr_pages; 1923 1924 f.headoffset = readahead_pos(rac); 1925 1926 z_erofs_pcluster_readmore(&f, rac, f.headoffset + 1927 readahead_length(rac) - 1, &pagepool, true); 1928 nr_pages = readahead_count(rac); 1929 trace_erofs_readpages(inode, readahead_index(rac), nr_pages, false); 1930 1931 while ((page = readahead_page(rac))) { 1932 set_page_private(page, (unsigned long)head); 1933 head = page; 1934 } 1935 1936 while (head) { 1937 struct page *page = head; 1938 int err; 1939 1940 /* traversal in reverse order */ 1941 head = (void *)page_private(page); 1942 1943 err = z_erofs_do_read_page(&f, page, &pagepool); 1944 if (err) 1945 erofs_err(inode->i_sb, 1946 "readahead error at page %lu @ nid %llu", 1947 page->index, EROFS_I(inode)->nid); 1948 put_page(page); 1949 } 1950 z_erofs_pcluster_readmore(&f, rac, 0, &pagepool, false); 1951 (void)z_erofs_collector_end(&f); 1952 1953 z_erofs_runqueue(&f, &pagepool, 1954 z_erofs_is_sync_decompress(sbi, nr_pages), true); 1955 erofs_put_metabuf(&f.map.buf); 1956 erofs_release_pages(&pagepool); 1957 } 1958 1959 const struct address_space_operations z_erofs_aops = { 1960 .read_folio = z_erofs_read_folio, 1961 .readahead = z_erofs_readahead, 1962 }; 1963